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>
29 #include "print-tree.h"
30 #include "transaction.h"
33 #include "free-space-cache.h"
35 static int update_block_group(struct btrfs_trans_handle *trans,
36 struct btrfs_root *root,
37 u64 bytenr, u64 num_bytes, int alloc,
39 static int update_reserved_extents(struct btrfs_block_group_cache *cache,
40 u64 num_bytes, int reserve);
41 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
42 struct btrfs_root *root,
43 u64 bytenr, u64 num_bytes, u64 parent,
44 u64 root_objectid, u64 owner_objectid,
45 u64 owner_offset, int refs_to_drop,
46 struct btrfs_delayed_extent_op *extra_op);
47 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
48 struct extent_buffer *leaf,
49 struct btrfs_extent_item *ei);
50 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
51 struct btrfs_root *root,
52 u64 parent, u64 root_objectid,
53 u64 flags, u64 owner, u64 offset,
54 struct btrfs_key *ins, int ref_mod);
55 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
56 struct btrfs_root *root,
57 u64 parent, u64 root_objectid,
58 u64 flags, struct btrfs_disk_key *key,
59 int level, struct btrfs_key *ins);
60 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
61 struct btrfs_root *extent_root, u64 alloc_bytes,
62 u64 flags, int force);
63 static int pin_down_bytes(struct btrfs_trans_handle *trans,
64 struct btrfs_root *root,
65 struct btrfs_path *path,
66 u64 bytenr, u64 num_bytes,
67 int is_data, int reserved,
68 struct extent_buffer **must_clean);
69 static int find_next_key(struct btrfs_path *path, int level,
70 struct btrfs_key *key);
71 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
72 int dump_block_groups);
75 block_group_cache_done(struct btrfs_block_group_cache *cache)
78 return cache->cached == BTRFS_CACHE_FINISHED;
81 static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits)
83 return (cache->flags & bits) == bits;
87 * this adds the block group to the fs_info rb tree for the block group
90 static int btrfs_add_block_group_cache(struct btrfs_fs_info *info,
91 struct btrfs_block_group_cache *block_group)
94 struct rb_node *parent = NULL;
95 struct btrfs_block_group_cache *cache;
97 spin_lock(&info->block_group_cache_lock);
98 p = &info->block_group_cache_tree.rb_node;
102 cache = rb_entry(parent, struct btrfs_block_group_cache,
104 if (block_group->key.objectid < cache->key.objectid) {
106 } else if (block_group->key.objectid > cache->key.objectid) {
109 spin_unlock(&info->block_group_cache_lock);
114 rb_link_node(&block_group->cache_node, parent, p);
115 rb_insert_color(&block_group->cache_node,
116 &info->block_group_cache_tree);
117 spin_unlock(&info->block_group_cache_lock);
123 * This will return the block group at or after bytenr if contains is 0, else
124 * it will return the block group that contains the bytenr
126 static struct btrfs_block_group_cache *
127 block_group_cache_tree_search(struct btrfs_fs_info *info, u64 bytenr,
130 struct btrfs_block_group_cache *cache, *ret = NULL;
134 spin_lock(&info->block_group_cache_lock);
135 n = info->block_group_cache_tree.rb_node;
138 cache = rb_entry(n, struct btrfs_block_group_cache,
140 end = cache->key.objectid + cache->key.offset - 1;
141 start = cache->key.objectid;
143 if (bytenr < start) {
144 if (!contains && (!ret || start < ret->key.objectid))
147 } else if (bytenr > start) {
148 if (contains && bytenr <= end) {
159 atomic_inc(&ret->count);
160 spin_unlock(&info->block_group_cache_lock);
165 static int add_excluded_extent(struct btrfs_root *root,
166 u64 start, u64 num_bytes)
168 u64 end = start + num_bytes - 1;
169 set_extent_bits(&root->fs_info->freed_extents[0],
170 start, end, EXTENT_UPTODATE, GFP_NOFS);
171 set_extent_bits(&root->fs_info->freed_extents[1],
172 start, end, EXTENT_UPTODATE, GFP_NOFS);
176 static void free_excluded_extents(struct btrfs_root *root,
177 struct btrfs_block_group_cache *cache)
181 start = cache->key.objectid;
182 end = start + cache->key.offset - 1;
184 clear_extent_bits(&root->fs_info->freed_extents[0],
185 start, end, EXTENT_UPTODATE, GFP_NOFS);
186 clear_extent_bits(&root->fs_info->freed_extents[1],
187 start, end, EXTENT_UPTODATE, GFP_NOFS);
190 static int exclude_super_stripes(struct btrfs_root *root,
191 struct btrfs_block_group_cache *cache)
198 for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
199 bytenr = btrfs_sb_offset(i);
200 ret = btrfs_rmap_block(&root->fs_info->mapping_tree,
201 cache->key.objectid, bytenr,
202 0, &logical, &nr, &stripe_len);
206 cache->bytes_super += stripe_len;
207 ret = add_excluded_extent(root, logical[nr],
217 static struct btrfs_caching_control *
218 get_caching_control(struct btrfs_block_group_cache *cache)
220 struct btrfs_caching_control *ctl;
222 spin_lock(&cache->lock);
223 if (cache->cached != BTRFS_CACHE_STARTED) {
224 spin_unlock(&cache->lock);
228 ctl = cache->caching_ctl;
229 atomic_inc(&ctl->count);
230 spin_unlock(&cache->lock);
234 static void put_caching_control(struct btrfs_caching_control *ctl)
236 if (atomic_dec_and_test(&ctl->count))
241 * this is only called by cache_block_group, since we could have freed extents
242 * we need to check the pinned_extents for any extents that can't be used yet
243 * since their free space will be released as soon as the transaction commits.
245 static u64 add_new_free_space(struct btrfs_block_group_cache *block_group,
246 struct btrfs_fs_info *info, u64 start, u64 end)
248 u64 extent_start, extent_end, size, total_added = 0;
251 while (start < end) {
252 ret = find_first_extent_bit(info->pinned_extents, start,
253 &extent_start, &extent_end,
254 EXTENT_DIRTY | EXTENT_UPTODATE);
258 if (extent_start == start) {
259 start = extent_end + 1;
260 } else if (extent_start > start && extent_start < end) {
261 size = extent_start - start;
263 ret = btrfs_add_free_space(block_group, start,
266 start = extent_end + 1;
275 ret = btrfs_add_free_space(block_group, start, size);
282 static int caching_kthread(void *data)
284 struct btrfs_block_group_cache *block_group = data;
285 struct btrfs_fs_info *fs_info = block_group->fs_info;
286 struct btrfs_caching_control *caching_ctl = block_group->caching_ctl;
287 struct btrfs_root *extent_root = fs_info->extent_root;
288 struct btrfs_path *path;
289 struct extent_buffer *leaf;
290 struct btrfs_key key;
296 path = btrfs_alloc_path();
300 exclude_super_stripes(extent_root, block_group);
301 spin_lock(&block_group->space_info->lock);
302 block_group->space_info->bytes_super += block_group->bytes_super;
303 spin_unlock(&block_group->space_info->lock);
305 last = max_t(u64, block_group->key.objectid, BTRFS_SUPER_INFO_OFFSET);
308 * We don't want to deadlock with somebody trying to allocate a new
309 * extent for the extent root while also trying to search the extent
310 * root to add free space. So we skip locking and search the commit
311 * root, since its read-only
313 path->skip_locking = 1;
314 path->search_commit_root = 1;
319 key.type = BTRFS_EXTENT_ITEM_KEY;
321 mutex_lock(&caching_ctl->mutex);
322 /* need to make sure the commit_root doesn't disappear */
323 down_read(&fs_info->extent_commit_sem);
325 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
329 leaf = path->nodes[0];
330 nritems = btrfs_header_nritems(leaf);
334 if (fs_info->closing > 1) {
339 if (path->slots[0] < nritems) {
340 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
342 ret = find_next_key(path, 0, &key);
346 caching_ctl->progress = last;
347 btrfs_release_path(extent_root, path);
348 up_read(&fs_info->extent_commit_sem);
349 mutex_unlock(&caching_ctl->mutex);
350 if (btrfs_transaction_in_commit(fs_info))
357 if (key.objectid < block_group->key.objectid) {
362 if (key.objectid >= block_group->key.objectid +
363 block_group->key.offset)
366 if (key.type == BTRFS_EXTENT_ITEM_KEY) {
367 total_found += add_new_free_space(block_group,
370 last = key.objectid + key.offset;
372 if (total_found > (1024 * 1024 * 2)) {
374 wake_up(&caching_ctl->wait);
381 total_found += add_new_free_space(block_group, fs_info, last,
382 block_group->key.objectid +
383 block_group->key.offset);
384 caching_ctl->progress = (u64)-1;
386 spin_lock(&block_group->lock);
387 block_group->caching_ctl = NULL;
388 block_group->cached = BTRFS_CACHE_FINISHED;
389 spin_unlock(&block_group->lock);
392 btrfs_free_path(path);
393 up_read(&fs_info->extent_commit_sem);
395 free_excluded_extents(extent_root, block_group);
397 mutex_unlock(&caching_ctl->mutex);
398 wake_up(&caching_ctl->wait);
400 put_caching_control(caching_ctl);
401 atomic_dec(&block_group->space_info->caching_threads);
405 static int cache_block_group(struct btrfs_block_group_cache *cache)
407 struct btrfs_fs_info *fs_info = cache->fs_info;
408 struct btrfs_caching_control *caching_ctl;
409 struct task_struct *tsk;
413 if (cache->cached != BTRFS_CACHE_NO)
416 caching_ctl = kzalloc(sizeof(*caching_ctl), GFP_KERNEL);
417 BUG_ON(!caching_ctl);
419 INIT_LIST_HEAD(&caching_ctl->list);
420 mutex_init(&caching_ctl->mutex);
421 init_waitqueue_head(&caching_ctl->wait);
422 caching_ctl->block_group = cache;
423 caching_ctl->progress = cache->key.objectid;
424 /* one for caching kthread, one for caching block group list */
425 atomic_set(&caching_ctl->count, 2);
427 spin_lock(&cache->lock);
428 if (cache->cached != BTRFS_CACHE_NO) {
429 spin_unlock(&cache->lock);
433 cache->caching_ctl = caching_ctl;
434 cache->cached = BTRFS_CACHE_STARTED;
435 spin_unlock(&cache->lock);
437 down_write(&fs_info->extent_commit_sem);
438 list_add_tail(&caching_ctl->list, &fs_info->caching_block_groups);
439 up_write(&fs_info->extent_commit_sem);
441 atomic_inc(&cache->space_info->caching_threads);
443 tsk = kthread_run(caching_kthread, cache, "btrfs-cache-%llu\n",
444 cache->key.objectid);
447 printk(KERN_ERR "error running thread %d\n", ret);
455 * return the block group that starts at or after bytenr
457 static struct btrfs_block_group_cache *
458 btrfs_lookup_first_block_group(struct btrfs_fs_info *info, u64 bytenr)
460 struct btrfs_block_group_cache *cache;
462 cache = block_group_cache_tree_search(info, bytenr, 0);
468 * return the block group that contains the given bytenr
470 struct btrfs_block_group_cache *btrfs_lookup_block_group(
471 struct btrfs_fs_info *info,
474 struct btrfs_block_group_cache *cache;
476 cache = block_group_cache_tree_search(info, bytenr, 1);
481 void btrfs_put_block_group(struct btrfs_block_group_cache *cache)
483 if (atomic_dec_and_test(&cache->count))
487 static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info,
490 struct list_head *head = &info->space_info;
491 struct btrfs_space_info *found;
494 list_for_each_entry_rcu(found, head, list) {
495 if (found->flags == flags) {
505 * after adding space to the filesystem, we need to clear the full flags
506 * on all the space infos.
508 void btrfs_clear_space_info_full(struct btrfs_fs_info *info)
510 struct list_head *head = &info->space_info;
511 struct btrfs_space_info *found;
514 list_for_each_entry_rcu(found, head, list)
519 static u64 div_factor(u64 num, int factor)
528 u64 btrfs_find_block_group(struct btrfs_root *root,
529 u64 search_start, u64 search_hint, int owner)
531 struct btrfs_block_group_cache *cache;
533 u64 last = max(search_hint, search_start);
540 cache = btrfs_lookup_first_block_group(root->fs_info, last);
544 spin_lock(&cache->lock);
545 last = cache->key.objectid + cache->key.offset;
546 used = btrfs_block_group_used(&cache->item);
548 if ((full_search || !cache->ro) &&
549 block_group_bits(cache, BTRFS_BLOCK_GROUP_METADATA)) {
550 if (used + cache->pinned + cache->reserved <
551 div_factor(cache->key.offset, factor)) {
552 group_start = cache->key.objectid;
553 spin_unlock(&cache->lock);
554 btrfs_put_block_group(cache);
558 spin_unlock(&cache->lock);
559 btrfs_put_block_group(cache);
567 if (!full_search && factor < 10) {
577 /* simple helper to search for an existing extent at a given offset */
578 int btrfs_lookup_extent(struct btrfs_root *root, u64 start, u64 len)
581 struct btrfs_key key;
582 struct btrfs_path *path;
584 path = btrfs_alloc_path();
586 key.objectid = start;
588 btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
589 ret = btrfs_search_slot(NULL, root->fs_info->extent_root, &key, path,
591 btrfs_free_path(path);
596 * Back reference rules. Back refs have three main goals:
598 * 1) differentiate between all holders of references to an extent so that
599 * when a reference is dropped we can make sure it was a valid reference
600 * before freeing the extent.
602 * 2) Provide enough information to quickly find the holders of an extent
603 * if we notice a given block is corrupted or bad.
605 * 3) Make it easy to migrate blocks for FS shrinking or storage pool
606 * maintenance. This is actually the same as #2, but with a slightly
607 * different use case.
609 * There are two kinds of back refs. The implicit back refs is optimized
610 * for pointers in non-shared tree blocks. For a given pointer in a block,
611 * back refs of this kind provide information about the block's owner tree
612 * and the pointer's key. These information allow us to find the block by
613 * b-tree searching. The full back refs is for pointers in tree blocks not
614 * referenced by their owner trees. The location of tree block is recorded
615 * in the back refs. Actually the full back refs is generic, and can be
616 * used in all cases the implicit back refs is used. The major shortcoming
617 * of the full back refs is its overhead. Every time a tree block gets
618 * COWed, we have to update back refs entry for all pointers in it.
620 * For a newly allocated tree block, we use implicit back refs for
621 * pointers in it. This means most tree related operations only involve
622 * implicit back refs. For a tree block created in old transaction, the
623 * only way to drop a reference to it is COW it. So we can detect the
624 * event that tree block loses its owner tree's reference and do the
625 * back refs conversion.
627 * When a tree block is COW'd through a tree, there are four cases:
629 * The reference count of the block is one and the tree is the block's
630 * owner tree. Nothing to do in this case.
632 * The reference count of the block is one and the tree is not the
633 * block's owner tree. In this case, full back refs is used for pointers
634 * in the block. Remove these full back refs, add implicit back refs for
635 * every pointers in the new block.
637 * The reference count of the block is greater than one and the tree is
638 * the block's owner tree. In this case, implicit back refs is used for
639 * pointers in the block. Add full back refs for every pointers in the
640 * block, increase lower level extents' reference counts. The original
641 * implicit back refs are entailed to the new block.
643 * The reference count of the block is greater than one and the tree is
644 * not the block's owner tree. Add implicit back refs for every pointer in
645 * the new block, increase lower level extents' reference count.
647 * Back Reference Key composing:
649 * The key objectid corresponds to the first byte in the extent,
650 * The key type is used to differentiate between types of back refs.
651 * There are different meanings of the key offset for different types
654 * File extents can be referenced by:
656 * - multiple snapshots, subvolumes, or different generations in one subvol
657 * - different files inside a single subvolume
658 * - different offsets inside a file (bookend extents in file.c)
660 * The extent ref structure for the implicit back refs has fields for:
662 * - Objectid of the subvolume root
663 * - objectid of the file holding the reference
664 * - original offset in the file
665 * - how many bookend extents
667 * The key offset for the implicit back refs is hash of the first
670 * The extent ref structure for the full back refs has field for:
672 * - number of pointers in the tree leaf
674 * The key offset for the implicit back refs is the first byte of
677 * When a file extent is allocated, The implicit back refs is used.
678 * the fields are filled in:
680 * (root_key.objectid, inode objectid, offset in file, 1)
682 * When a file extent is removed file truncation, we find the
683 * corresponding implicit back refs and check the following fields:
685 * (btrfs_header_owner(leaf), inode objectid, offset in file)
687 * Btree extents can be referenced by:
689 * - Different subvolumes
691 * Both the implicit back refs and the full back refs for tree blocks
692 * only consist of key. The key offset for the implicit back refs is
693 * objectid of block's owner tree. The key offset for the full back refs
694 * is the first byte of parent block.
696 * When implicit back refs is used, information about the lowest key and
697 * level of the tree block are required. These information are stored in
698 * tree block info structure.
701 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
702 static int convert_extent_item_v0(struct btrfs_trans_handle *trans,
703 struct btrfs_root *root,
704 struct btrfs_path *path,
705 u64 owner, u32 extra_size)
707 struct btrfs_extent_item *item;
708 struct btrfs_extent_item_v0 *ei0;
709 struct btrfs_extent_ref_v0 *ref0;
710 struct btrfs_tree_block_info *bi;
711 struct extent_buffer *leaf;
712 struct btrfs_key key;
713 struct btrfs_key found_key;
714 u32 new_size = sizeof(*item);
718 leaf = path->nodes[0];
719 BUG_ON(btrfs_item_size_nr(leaf, path->slots[0]) != sizeof(*ei0));
721 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
722 ei0 = btrfs_item_ptr(leaf, path->slots[0],
723 struct btrfs_extent_item_v0);
724 refs = btrfs_extent_refs_v0(leaf, ei0);
726 if (owner == (u64)-1) {
728 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
729 ret = btrfs_next_leaf(root, path);
733 leaf = path->nodes[0];
735 btrfs_item_key_to_cpu(leaf, &found_key,
737 BUG_ON(key.objectid != found_key.objectid);
738 if (found_key.type != BTRFS_EXTENT_REF_V0_KEY) {
742 ref0 = btrfs_item_ptr(leaf, path->slots[0],
743 struct btrfs_extent_ref_v0);
744 owner = btrfs_ref_objectid_v0(leaf, ref0);
748 btrfs_release_path(root, path);
750 if (owner < BTRFS_FIRST_FREE_OBJECTID)
751 new_size += sizeof(*bi);
753 new_size -= sizeof(*ei0);
754 ret = btrfs_search_slot(trans, root, &key, path,
755 new_size + extra_size, 1);
760 ret = btrfs_extend_item(trans, root, path, new_size);
763 leaf = path->nodes[0];
764 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
765 btrfs_set_extent_refs(leaf, item, refs);
766 /* FIXME: get real generation */
767 btrfs_set_extent_generation(leaf, item, 0);
768 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
769 btrfs_set_extent_flags(leaf, item,
770 BTRFS_EXTENT_FLAG_TREE_BLOCK |
771 BTRFS_BLOCK_FLAG_FULL_BACKREF);
772 bi = (struct btrfs_tree_block_info *)(item + 1);
773 /* FIXME: get first key of the block */
774 memset_extent_buffer(leaf, 0, (unsigned long)bi, sizeof(*bi));
775 btrfs_set_tree_block_level(leaf, bi, (int)owner);
777 btrfs_set_extent_flags(leaf, item, BTRFS_EXTENT_FLAG_DATA);
779 btrfs_mark_buffer_dirty(leaf);
784 static u64 hash_extent_data_ref(u64 root_objectid, u64 owner, u64 offset)
786 u32 high_crc = ~(u32)0;
787 u32 low_crc = ~(u32)0;
790 lenum = cpu_to_le64(root_objectid);
791 high_crc = crc32c(high_crc, &lenum, sizeof(lenum));
792 lenum = cpu_to_le64(owner);
793 low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
794 lenum = cpu_to_le64(offset);
795 low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
797 return ((u64)high_crc << 31) ^ (u64)low_crc;
800 static u64 hash_extent_data_ref_item(struct extent_buffer *leaf,
801 struct btrfs_extent_data_ref *ref)
803 return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf, ref),
804 btrfs_extent_data_ref_objectid(leaf, ref),
805 btrfs_extent_data_ref_offset(leaf, ref));
808 static int match_extent_data_ref(struct extent_buffer *leaf,
809 struct btrfs_extent_data_ref *ref,
810 u64 root_objectid, u64 owner, u64 offset)
812 if (btrfs_extent_data_ref_root(leaf, ref) != root_objectid ||
813 btrfs_extent_data_ref_objectid(leaf, ref) != owner ||
814 btrfs_extent_data_ref_offset(leaf, ref) != offset)
819 static noinline int lookup_extent_data_ref(struct btrfs_trans_handle *trans,
820 struct btrfs_root *root,
821 struct btrfs_path *path,
822 u64 bytenr, u64 parent,
824 u64 owner, u64 offset)
826 struct btrfs_key key;
827 struct btrfs_extent_data_ref *ref;
828 struct extent_buffer *leaf;
834 key.objectid = bytenr;
836 key.type = BTRFS_SHARED_DATA_REF_KEY;
839 key.type = BTRFS_EXTENT_DATA_REF_KEY;
840 key.offset = hash_extent_data_ref(root_objectid,
845 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
854 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
855 key.type = BTRFS_EXTENT_REF_V0_KEY;
856 btrfs_release_path(root, path);
857 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
868 leaf = path->nodes[0];
869 nritems = btrfs_header_nritems(leaf);
871 if (path->slots[0] >= nritems) {
872 ret = btrfs_next_leaf(root, path);
878 leaf = path->nodes[0];
879 nritems = btrfs_header_nritems(leaf);
883 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
884 if (key.objectid != bytenr ||
885 key.type != BTRFS_EXTENT_DATA_REF_KEY)
888 ref = btrfs_item_ptr(leaf, path->slots[0],
889 struct btrfs_extent_data_ref);
891 if (match_extent_data_ref(leaf, ref, root_objectid,
894 btrfs_release_path(root, path);
906 static noinline int insert_extent_data_ref(struct btrfs_trans_handle *trans,
907 struct btrfs_root *root,
908 struct btrfs_path *path,
909 u64 bytenr, u64 parent,
910 u64 root_objectid, u64 owner,
911 u64 offset, int refs_to_add)
913 struct btrfs_key key;
914 struct extent_buffer *leaf;
919 key.objectid = bytenr;
921 key.type = BTRFS_SHARED_DATA_REF_KEY;
923 size = sizeof(struct btrfs_shared_data_ref);
925 key.type = BTRFS_EXTENT_DATA_REF_KEY;
926 key.offset = hash_extent_data_ref(root_objectid,
928 size = sizeof(struct btrfs_extent_data_ref);
931 ret = btrfs_insert_empty_item(trans, root, path, &key, size);
932 if (ret && ret != -EEXIST)
935 leaf = path->nodes[0];
937 struct btrfs_shared_data_ref *ref;
938 ref = btrfs_item_ptr(leaf, path->slots[0],
939 struct btrfs_shared_data_ref);
941 btrfs_set_shared_data_ref_count(leaf, ref, refs_to_add);
943 num_refs = btrfs_shared_data_ref_count(leaf, ref);
944 num_refs += refs_to_add;
945 btrfs_set_shared_data_ref_count(leaf, ref, num_refs);
948 struct btrfs_extent_data_ref *ref;
949 while (ret == -EEXIST) {
950 ref = btrfs_item_ptr(leaf, path->slots[0],
951 struct btrfs_extent_data_ref);
952 if (match_extent_data_ref(leaf, ref, root_objectid,
955 btrfs_release_path(root, path);
957 ret = btrfs_insert_empty_item(trans, root, path, &key,
959 if (ret && ret != -EEXIST)
962 leaf = path->nodes[0];
964 ref = btrfs_item_ptr(leaf, path->slots[0],
965 struct btrfs_extent_data_ref);
967 btrfs_set_extent_data_ref_root(leaf, ref,
969 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
970 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
971 btrfs_set_extent_data_ref_count(leaf, ref, refs_to_add);
973 num_refs = btrfs_extent_data_ref_count(leaf, ref);
974 num_refs += refs_to_add;
975 btrfs_set_extent_data_ref_count(leaf, ref, num_refs);
978 btrfs_mark_buffer_dirty(leaf);
981 btrfs_release_path(root, path);
985 static noinline int remove_extent_data_ref(struct btrfs_trans_handle *trans,
986 struct btrfs_root *root,
987 struct btrfs_path *path,
990 struct btrfs_key key;
991 struct btrfs_extent_data_ref *ref1 = NULL;
992 struct btrfs_shared_data_ref *ref2 = NULL;
993 struct extent_buffer *leaf;
997 leaf = path->nodes[0];
998 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1000 if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1001 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1002 struct btrfs_extent_data_ref);
1003 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1004 } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1005 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1006 struct btrfs_shared_data_ref);
1007 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1008 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1009 } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1010 struct btrfs_extent_ref_v0 *ref0;
1011 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1012 struct btrfs_extent_ref_v0);
1013 num_refs = btrfs_ref_count_v0(leaf, ref0);
1019 BUG_ON(num_refs < refs_to_drop);
1020 num_refs -= refs_to_drop;
1022 if (num_refs == 0) {
1023 ret = btrfs_del_item(trans, root, path);
1025 if (key.type == BTRFS_EXTENT_DATA_REF_KEY)
1026 btrfs_set_extent_data_ref_count(leaf, ref1, num_refs);
1027 else if (key.type == BTRFS_SHARED_DATA_REF_KEY)
1028 btrfs_set_shared_data_ref_count(leaf, ref2, num_refs);
1029 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1031 struct btrfs_extent_ref_v0 *ref0;
1032 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1033 struct btrfs_extent_ref_v0);
1034 btrfs_set_ref_count_v0(leaf, ref0, num_refs);
1037 btrfs_mark_buffer_dirty(leaf);
1042 static noinline u32 extent_data_ref_count(struct btrfs_root *root,
1043 struct btrfs_path *path,
1044 struct btrfs_extent_inline_ref *iref)
1046 struct btrfs_key key;
1047 struct extent_buffer *leaf;
1048 struct btrfs_extent_data_ref *ref1;
1049 struct btrfs_shared_data_ref *ref2;
1052 leaf = path->nodes[0];
1053 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1055 if (btrfs_extent_inline_ref_type(leaf, iref) ==
1056 BTRFS_EXTENT_DATA_REF_KEY) {
1057 ref1 = (struct btrfs_extent_data_ref *)(&iref->offset);
1058 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1060 ref2 = (struct btrfs_shared_data_ref *)(iref + 1);
1061 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1063 } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1064 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1065 struct btrfs_extent_data_ref);
1066 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1067 } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1068 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1069 struct btrfs_shared_data_ref);
1070 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1071 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1072 } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1073 struct btrfs_extent_ref_v0 *ref0;
1074 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1075 struct btrfs_extent_ref_v0);
1076 num_refs = btrfs_ref_count_v0(leaf, ref0);
1084 static noinline int lookup_tree_block_ref(struct btrfs_trans_handle *trans,
1085 struct btrfs_root *root,
1086 struct btrfs_path *path,
1087 u64 bytenr, u64 parent,
1090 struct btrfs_key key;
1093 key.objectid = bytenr;
1095 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1096 key.offset = parent;
1098 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1099 key.offset = root_objectid;
1102 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1105 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1106 if (ret == -ENOENT && parent) {
1107 btrfs_release_path(root, path);
1108 key.type = BTRFS_EXTENT_REF_V0_KEY;
1109 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1117 static noinline int insert_tree_block_ref(struct btrfs_trans_handle *trans,
1118 struct btrfs_root *root,
1119 struct btrfs_path *path,
1120 u64 bytenr, u64 parent,
1123 struct btrfs_key key;
1126 key.objectid = bytenr;
1128 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1129 key.offset = parent;
1131 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1132 key.offset = root_objectid;
1135 ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
1136 btrfs_release_path(root, path);
1140 static inline int extent_ref_type(u64 parent, u64 owner)
1143 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1145 type = BTRFS_SHARED_BLOCK_REF_KEY;
1147 type = BTRFS_TREE_BLOCK_REF_KEY;
1150 type = BTRFS_SHARED_DATA_REF_KEY;
1152 type = BTRFS_EXTENT_DATA_REF_KEY;
1157 static int find_next_key(struct btrfs_path *path, int level,
1158 struct btrfs_key *key)
1161 for (; level < BTRFS_MAX_LEVEL; level++) {
1162 if (!path->nodes[level])
1164 if (path->slots[level] + 1 >=
1165 btrfs_header_nritems(path->nodes[level]))
1168 btrfs_item_key_to_cpu(path->nodes[level], key,
1169 path->slots[level] + 1);
1171 btrfs_node_key_to_cpu(path->nodes[level], key,
1172 path->slots[level] + 1);
1179 * look for inline back ref. if back ref is found, *ref_ret is set
1180 * to the address of inline back ref, and 0 is returned.
1182 * if back ref isn't found, *ref_ret is set to the address where it
1183 * should be inserted, and -ENOENT is returned.
1185 * if insert is true and there are too many inline back refs, the path
1186 * points to the extent item, and -EAGAIN is returned.
1188 * NOTE: inline back refs are ordered in the same way that back ref
1189 * items in the tree are ordered.
1191 static noinline_for_stack
1192 int lookup_inline_extent_backref(struct btrfs_trans_handle *trans,
1193 struct btrfs_root *root,
1194 struct btrfs_path *path,
1195 struct btrfs_extent_inline_ref **ref_ret,
1196 u64 bytenr, u64 num_bytes,
1197 u64 parent, u64 root_objectid,
1198 u64 owner, u64 offset, int insert)
1200 struct btrfs_key key;
1201 struct extent_buffer *leaf;
1202 struct btrfs_extent_item *ei;
1203 struct btrfs_extent_inline_ref *iref;
1214 key.objectid = bytenr;
1215 key.type = BTRFS_EXTENT_ITEM_KEY;
1216 key.offset = num_bytes;
1218 want = extent_ref_type(parent, owner);
1220 extra_size = btrfs_extent_inline_ref_size(want);
1221 path->keep_locks = 1;
1224 ret = btrfs_search_slot(trans, root, &key, path, extra_size, 1);
1231 leaf = path->nodes[0];
1232 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1233 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1234 if (item_size < sizeof(*ei)) {
1239 ret = convert_extent_item_v0(trans, root, path, owner,
1245 leaf = path->nodes[0];
1246 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1249 BUG_ON(item_size < sizeof(*ei));
1251 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1252 flags = btrfs_extent_flags(leaf, ei);
1254 ptr = (unsigned long)(ei + 1);
1255 end = (unsigned long)ei + item_size;
1257 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
1258 ptr += sizeof(struct btrfs_tree_block_info);
1261 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_DATA));
1270 iref = (struct btrfs_extent_inline_ref *)ptr;
1271 type = btrfs_extent_inline_ref_type(leaf, iref);
1275 ptr += btrfs_extent_inline_ref_size(type);
1279 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1280 struct btrfs_extent_data_ref *dref;
1281 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1282 if (match_extent_data_ref(leaf, dref, root_objectid,
1287 if (hash_extent_data_ref_item(leaf, dref) <
1288 hash_extent_data_ref(root_objectid, owner, offset))
1292 ref_offset = btrfs_extent_inline_ref_offset(leaf, iref);
1294 if (parent == ref_offset) {
1298 if (ref_offset < parent)
1301 if (root_objectid == ref_offset) {
1305 if (ref_offset < root_objectid)
1309 ptr += btrfs_extent_inline_ref_size(type);
1311 if (err == -ENOENT && insert) {
1312 if (item_size + extra_size >=
1313 BTRFS_MAX_EXTENT_ITEM_SIZE(root)) {
1318 * To add new inline back ref, we have to make sure
1319 * there is no corresponding back ref item.
1320 * For simplicity, we just do not add new inline back
1321 * ref if there is any kind of item for this block
1323 if (find_next_key(path, 0, &key) == 0 &&
1324 key.objectid == bytenr &&
1325 key.type < BTRFS_BLOCK_GROUP_ITEM_KEY) {
1330 *ref_ret = (struct btrfs_extent_inline_ref *)ptr;
1333 path->keep_locks = 0;
1334 btrfs_unlock_up_safe(path, 1);
1340 * helper to add new inline back ref
1342 static noinline_for_stack
1343 int setup_inline_extent_backref(struct btrfs_trans_handle *trans,
1344 struct btrfs_root *root,
1345 struct btrfs_path *path,
1346 struct btrfs_extent_inline_ref *iref,
1347 u64 parent, u64 root_objectid,
1348 u64 owner, u64 offset, int refs_to_add,
1349 struct btrfs_delayed_extent_op *extent_op)
1351 struct extent_buffer *leaf;
1352 struct btrfs_extent_item *ei;
1355 unsigned long item_offset;
1361 leaf = path->nodes[0];
1362 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1363 item_offset = (unsigned long)iref - (unsigned long)ei;
1365 type = extent_ref_type(parent, owner);
1366 size = btrfs_extent_inline_ref_size(type);
1368 ret = btrfs_extend_item(trans, root, path, size);
1371 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1372 refs = btrfs_extent_refs(leaf, ei);
1373 refs += refs_to_add;
1374 btrfs_set_extent_refs(leaf, ei, refs);
1376 __run_delayed_extent_op(extent_op, leaf, ei);
1378 ptr = (unsigned long)ei + item_offset;
1379 end = (unsigned long)ei + btrfs_item_size_nr(leaf, path->slots[0]);
1380 if (ptr < end - size)
1381 memmove_extent_buffer(leaf, ptr + size, ptr,
1384 iref = (struct btrfs_extent_inline_ref *)ptr;
1385 btrfs_set_extent_inline_ref_type(leaf, iref, type);
1386 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1387 struct btrfs_extent_data_ref *dref;
1388 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1389 btrfs_set_extent_data_ref_root(leaf, dref, root_objectid);
1390 btrfs_set_extent_data_ref_objectid(leaf, dref, owner);
1391 btrfs_set_extent_data_ref_offset(leaf, dref, offset);
1392 btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add);
1393 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1394 struct btrfs_shared_data_ref *sref;
1395 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1396 btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add);
1397 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1398 } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) {
1399 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1401 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
1403 btrfs_mark_buffer_dirty(leaf);
1407 static int lookup_extent_backref(struct btrfs_trans_handle *trans,
1408 struct btrfs_root *root,
1409 struct btrfs_path *path,
1410 struct btrfs_extent_inline_ref **ref_ret,
1411 u64 bytenr, u64 num_bytes, u64 parent,
1412 u64 root_objectid, u64 owner, u64 offset)
1416 ret = lookup_inline_extent_backref(trans, root, path, ref_ret,
1417 bytenr, num_bytes, parent,
1418 root_objectid, owner, offset, 0);
1422 btrfs_release_path(root, path);
1425 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1426 ret = lookup_tree_block_ref(trans, root, path, bytenr, parent,
1429 ret = lookup_extent_data_ref(trans, root, path, bytenr, parent,
1430 root_objectid, owner, offset);
1436 * helper to update/remove inline back ref
1438 static noinline_for_stack
1439 int update_inline_extent_backref(struct btrfs_trans_handle *trans,
1440 struct btrfs_root *root,
1441 struct btrfs_path *path,
1442 struct btrfs_extent_inline_ref *iref,
1444 struct btrfs_delayed_extent_op *extent_op)
1446 struct extent_buffer *leaf;
1447 struct btrfs_extent_item *ei;
1448 struct btrfs_extent_data_ref *dref = NULL;
1449 struct btrfs_shared_data_ref *sref = NULL;
1458 leaf = path->nodes[0];
1459 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1460 refs = btrfs_extent_refs(leaf, ei);
1461 WARN_ON(refs_to_mod < 0 && refs + refs_to_mod <= 0);
1462 refs += refs_to_mod;
1463 btrfs_set_extent_refs(leaf, ei, refs);
1465 __run_delayed_extent_op(extent_op, leaf, ei);
1467 type = btrfs_extent_inline_ref_type(leaf, iref);
1469 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1470 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1471 refs = btrfs_extent_data_ref_count(leaf, dref);
1472 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1473 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1474 refs = btrfs_shared_data_ref_count(leaf, sref);
1477 BUG_ON(refs_to_mod != -1);
1480 BUG_ON(refs_to_mod < 0 && refs < -refs_to_mod);
1481 refs += refs_to_mod;
1484 if (type == BTRFS_EXTENT_DATA_REF_KEY)
1485 btrfs_set_extent_data_ref_count(leaf, dref, refs);
1487 btrfs_set_shared_data_ref_count(leaf, sref, refs);
1489 size = btrfs_extent_inline_ref_size(type);
1490 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1491 ptr = (unsigned long)iref;
1492 end = (unsigned long)ei + item_size;
1493 if (ptr + size < end)
1494 memmove_extent_buffer(leaf, ptr, ptr + size,
1497 ret = btrfs_truncate_item(trans, root, path, item_size, 1);
1500 btrfs_mark_buffer_dirty(leaf);
1504 static noinline_for_stack
1505 int insert_inline_extent_backref(struct btrfs_trans_handle *trans,
1506 struct btrfs_root *root,
1507 struct btrfs_path *path,
1508 u64 bytenr, u64 num_bytes, u64 parent,
1509 u64 root_objectid, u64 owner,
1510 u64 offset, int refs_to_add,
1511 struct btrfs_delayed_extent_op *extent_op)
1513 struct btrfs_extent_inline_ref *iref;
1516 ret = lookup_inline_extent_backref(trans, root, path, &iref,
1517 bytenr, num_bytes, parent,
1518 root_objectid, owner, offset, 1);
1520 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID);
1521 ret = update_inline_extent_backref(trans, root, path, iref,
1522 refs_to_add, extent_op);
1523 } else if (ret == -ENOENT) {
1524 ret = setup_inline_extent_backref(trans, root, path, iref,
1525 parent, root_objectid,
1526 owner, offset, refs_to_add,
1532 static int insert_extent_backref(struct btrfs_trans_handle *trans,
1533 struct btrfs_root *root,
1534 struct btrfs_path *path,
1535 u64 bytenr, u64 parent, u64 root_objectid,
1536 u64 owner, u64 offset, int refs_to_add)
1539 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1540 BUG_ON(refs_to_add != 1);
1541 ret = insert_tree_block_ref(trans, root, path, bytenr,
1542 parent, root_objectid);
1544 ret = insert_extent_data_ref(trans, root, path, bytenr,
1545 parent, root_objectid,
1546 owner, offset, refs_to_add);
1551 static int remove_extent_backref(struct btrfs_trans_handle *trans,
1552 struct btrfs_root *root,
1553 struct btrfs_path *path,
1554 struct btrfs_extent_inline_ref *iref,
1555 int refs_to_drop, int is_data)
1559 BUG_ON(!is_data && refs_to_drop != 1);
1561 ret = update_inline_extent_backref(trans, root, path, iref,
1562 -refs_to_drop, NULL);
1563 } else if (is_data) {
1564 ret = remove_extent_data_ref(trans, root, path, refs_to_drop);
1566 ret = btrfs_del_item(trans, root, path);
1571 static void btrfs_issue_discard(struct block_device *bdev,
1574 blkdev_issue_discard(bdev, start >> 9, len >> 9, GFP_KERNEL);
1577 static int btrfs_discard_extent(struct btrfs_root *root, u64 bytenr,
1581 u64 map_length = num_bytes;
1582 struct btrfs_multi_bio *multi = NULL;
1584 if (!btrfs_test_opt(root, DISCARD))
1587 /* Tell the block device(s) that the sectors can be discarded */
1588 ret = btrfs_map_block(&root->fs_info->mapping_tree, READ,
1589 bytenr, &map_length, &multi, 0);
1591 struct btrfs_bio_stripe *stripe = multi->stripes;
1594 if (map_length > num_bytes)
1595 map_length = num_bytes;
1597 for (i = 0; i < multi->num_stripes; i++, stripe++) {
1598 btrfs_issue_discard(stripe->dev->bdev,
1608 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1609 struct btrfs_root *root,
1610 u64 bytenr, u64 num_bytes, u64 parent,
1611 u64 root_objectid, u64 owner, u64 offset)
1614 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID &&
1615 root_objectid == BTRFS_TREE_LOG_OBJECTID);
1617 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1618 ret = btrfs_add_delayed_tree_ref(trans, bytenr, num_bytes,
1619 parent, root_objectid, (int)owner,
1620 BTRFS_ADD_DELAYED_REF, NULL);
1622 ret = btrfs_add_delayed_data_ref(trans, bytenr, num_bytes,
1623 parent, root_objectid, owner, offset,
1624 BTRFS_ADD_DELAYED_REF, NULL);
1629 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1630 struct btrfs_root *root,
1631 u64 bytenr, u64 num_bytes,
1632 u64 parent, u64 root_objectid,
1633 u64 owner, u64 offset, int refs_to_add,
1634 struct btrfs_delayed_extent_op *extent_op)
1636 struct btrfs_path *path;
1637 struct extent_buffer *leaf;
1638 struct btrfs_extent_item *item;
1643 path = btrfs_alloc_path();
1648 path->leave_spinning = 1;
1649 /* this will setup the path even if it fails to insert the back ref */
1650 ret = insert_inline_extent_backref(trans, root->fs_info->extent_root,
1651 path, bytenr, num_bytes, parent,
1652 root_objectid, owner, offset,
1653 refs_to_add, extent_op);
1657 if (ret != -EAGAIN) {
1662 leaf = path->nodes[0];
1663 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1664 refs = btrfs_extent_refs(leaf, item);
1665 btrfs_set_extent_refs(leaf, item, refs + refs_to_add);
1667 __run_delayed_extent_op(extent_op, leaf, item);
1669 btrfs_mark_buffer_dirty(leaf);
1670 btrfs_release_path(root->fs_info->extent_root, path);
1673 path->leave_spinning = 1;
1675 /* now insert the actual backref */
1676 ret = insert_extent_backref(trans, root->fs_info->extent_root,
1677 path, bytenr, parent, root_objectid,
1678 owner, offset, refs_to_add);
1681 btrfs_free_path(path);
1685 static int run_delayed_data_ref(struct btrfs_trans_handle *trans,
1686 struct btrfs_root *root,
1687 struct btrfs_delayed_ref_node *node,
1688 struct btrfs_delayed_extent_op *extent_op,
1689 int insert_reserved)
1692 struct btrfs_delayed_data_ref *ref;
1693 struct btrfs_key ins;
1698 ins.objectid = node->bytenr;
1699 ins.offset = node->num_bytes;
1700 ins.type = BTRFS_EXTENT_ITEM_KEY;
1702 ref = btrfs_delayed_node_to_data_ref(node);
1703 if (node->type == BTRFS_SHARED_DATA_REF_KEY)
1704 parent = ref->parent;
1706 ref_root = ref->root;
1708 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
1710 BUG_ON(extent_op->update_key);
1711 flags |= extent_op->flags_to_set;
1713 ret = alloc_reserved_file_extent(trans, root,
1714 parent, ref_root, flags,
1715 ref->objectid, ref->offset,
1716 &ins, node->ref_mod);
1717 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
1718 ret = __btrfs_inc_extent_ref(trans, root, node->bytenr,
1719 node->num_bytes, parent,
1720 ref_root, ref->objectid,
1721 ref->offset, node->ref_mod,
1723 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
1724 ret = __btrfs_free_extent(trans, root, node->bytenr,
1725 node->num_bytes, parent,
1726 ref_root, ref->objectid,
1727 ref->offset, node->ref_mod,
1735 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
1736 struct extent_buffer *leaf,
1737 struct btrfs_extent_item *ei)
1739 u64 flags = btrfs_extent_flags(leaf, ei);
1740 if (extent_op->update_flags) {
1741 flags |= extent_op->flags_to_set;
1742 btrfs_set_extent_flags(leaf, ei, flags);
1745 if (extent_op->update_key) {
1746 struct btrfs_tree_block_info *bi;
1747 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK));
1748 bi = (struct btrfs_tree_block_info *)(ei + 1);
1749 btrfs_set_tree_block_key(leaf, bi, &extent_op->key);
1753 static int run_delayed_extent_op(struct btrfs_trans_handle *trans,
1754 struct btrfs_root *root,
1755 struct btrfs_delayed_ref_node *node,
1756 struct btrfs_delayed_extent_op *extent_op)
1758 struct btrfs_key key;
1759 struct btrfs_path *path;
1760 struct btrfs_extent_item *ei;
1761 struct extent_buffer *leaf;
1766 path = btrfs_alloc_path();
1770 key.objectid = node->bytenr;
1771 key.type = BTRFS_EXTENT_ITEM_KEY;
1772 key.offset = node->num_bytes;
1775 path->leave_spinning = 1;
1776 ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key,
1787 leaf = path->nodes[0];
1788 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1789 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1790 if (item_size < sizeof(*ei)) {
1791 ret = convert_extent_item_v0(trans, root->fs_info->extent_root,
1797 leaf = path->nodes[0];
1798 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1801 BUG_ON(item_size < sizeof(*ei));
1802 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1803 __run_delayed_extent_op(extent_op, leaf, ei);
1805 btrfs_mark_buffer_dirty(leaf);
1807 btrfs_free_path(path);
1811 static int run_delayed_tree_ref(struct btrfs_trans_handle *trans,
1812 struct btrfs_root *root,
1813 struct btrfs_delayed_ref_node *node,
1814 struct btrfs_delayed_extent_op *extent_op,
1815 int insert_reserved)
1818 struct btrfs_delayed_tree_ref *ref;
1819 struct btrfs_key ins;
1823 ins.objectid = node->bytenr;
1824 ins.offset = node->num_bytes;
1825 ins.type = BTRFS_EXTENT_ITEM_KEY;
1827 ref = btrfs_delayed_node_to_tree_ref(node);
1828 if (node->type == BTRFS_SHARED_BLOCK_REF_KEY)
1829 parent = ref->parent;
1831 ref_root = ref->root;
1833 BUG_ON(node->ref_mod != 1);
1834 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
1835 BUG_ON(!extent_op || !extent_op->update_flags ||
1836 !extent_op->update_key);
1837 ret = alloc_reserved_tree_block(trans, root,
1839 extent_op->flags_to_set,
1842 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
1843 ret = __btrfs_inc_extent_ref(trans, root, node->bytenr,
1844 node->num_bytes, parent, ref_root,
1845 ref->level, 0, 1, extent_op);
1846 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
1847 ret = __btrfs_free_extent(trans, root, node->bytenr,
1848 node->num_bytes, parent, ref_root,
1849 ref->level, 0, 1, extent_op);
1857 /* helper function to actually process a single delayed ref entry */
1858 static int run_one_delayed_ref(struct btrfs_trans_handle *trans,
1859 struct btrfs_root *root,
1860 struct btrfs_delayed_ref_node *node,
1861 struct btrfs_delayed_extent_op *extent_op,
1862 int insert_reserved)
1865 if (btrfs_delayed_ref_is_head(node)) {
1866 struct btrfs_delayed_ref_head *head;
1868 * we've hit the end of the chain and we were supposed
1869 * to insert this extent into the tree. But, it got
1870 * deleted before we ever needed to insert it, so all
1871 * we have to do is clean up the accounting
1874 head = btrfs_delayed_node_to_head(node);
1875 if (insert_reserved) {
1877 struct extent_buffer *must_clean = NULL;
1879 ret = pin_down_bytes(trans, root, NULL,
1880 node->bytenr, node->num_bytes,
1881 head->is_data, 1, &must_clean);
1886 clean_tree_block(NULL, root, must_clean);
1887 btrfs_tree_unlock(must_clean);
1888 free_extent_buffer(must_clean);
1890 if (head->is_data) {
1891 ret = btrfs_del_csums(trans, root,
1897 ret = btrfs_free_reserved_extent(root,
1903 mutex_unlock(&head->mutex);
1907 if (node->type == BTRFS_TREE_BLOCK_REF_KEY ||
1908 node->type == BTRFS_SHARED_BLOCK_REF_KEY)
1909 ret = run_delayed_tree_ref(trans, root, node, extent_op,
1911 else if (node->type == BTRFS_EXTENT_DATA_REF_KEY ||
1912 node->type == BTRFS_SHARED_DATA_REF_KEY)
1913 ret = run_delayed_data_ref(trans, root, node, extent_op,
1920 static noinline struct btrfs_delayed_ref_node *
1921 select_delayed_ref(struct btrfs_delayed_ref_head *head)
1923 struct rb_node *node;
1924 struct btrfs_delayed_ref_node *ref;
1925 int action = BTRFS_ADD_DELAYED_REF;
1928 * select delayed ref of type BTRFS_ADD_DELAYED_REF first.
1929 * this prevents ref count from going down to zero when
1930 * there still are pending delayed ref.
1932 node = rb_prev(&head->node.rb_node);
1936 ref = rb_entry(node, struct btrfs_delayed_ref_node,
1938 if (ref->bytenr != head->node.bytenr)
1940 if (ref->action == action)
1942 node = rb_prev(node);
1944 if (action == BTRFS_ADD_DELAYED_REF) {
1945 action = BTRFS_DROP_DELAYED_REF;
1951 static noinline int run_clustered_refs(struct btrfs_trans_handle *trans,
1952 struct btrfs_root *root,
1953 struct list_head *cluster)
1955 struct btrfs_delayed_ref_root *delayed_refs;
1956 struct btrfs_delayed_ref_node *ref;
1957 struct btrfs_delayed_ref_head *locked_ref = NULL;
1958 struct btrfs_delayed_extent_op *extent_op;
1961 int must_insert_reserved = 0;
1963 delayed_refs = &trans->transaction->delayed_refs;
1966 /* pick a new head ref from the cluster list */
1967 if (list_empty(cluster))
1970 locked_ref = list_entry(cluster->next,
1971 struct btrfs_delayed_ref_head, cluster);
1973 /* grab the lock that says we are going to process
1974 * all the refs for this head */
1975 ret = btrfs_delayed_ref_lock(trans, locked_ref);
1978 * we may have dropped the spin lock to get the head
1979 * mutex lock, and that might have given someone else
1980 * time to free the head. If that's true, it has been
1981 * removed from our list and we can move on.
1983 if (ret == -EAGAIN) {
1991 * record the must insert reserved flag before we
1992 * drop the spin lock.
1994 must_insert_reserved = locked_ref->must_insert_reserved;
1995 locked_ref->must_insert_reserved = 0;
1997 extent_op = locked_ref->extent_op;
1998 locked_ref->extent_op = NULL;
2001 * locked_ref is the head node, so we have to go one
2002 * node back for any delayed ref updates
2004 ref = select_delayed_ref(locked_ref);
2006 /* All delayed refs have been processed, Go ahead
2007 * and send the head node to run_one_delayed_ref,
2008 * so that any accounting fixes can happen
2010 ref = &locked_ref->node;
2012 if (extent_op && must_insert_reserved) {
2018 spin_unlock(&delayed_refs->lock);
2020 ret = run_delayed_extent_op(trans, root,
2026 spin_lock(&delayed_refs->lock);
2030 list_del_init(&locked_ref->cluster);
2035 rb_erase(&ref->rb_node, &delayed_refs->root);
2036 delayed_refs->num_entries--;
2038 spin_unlock(&delayed_refs->lock);
2040 ret = run_one_delayed_ref(trans, root, ref, extent_op,
2041 must_insert_reserved);
2044 btrfs_put_delayed_ref(ref);
2049 spin_lock(&delayed_refs->lock);
2055 * this starts processing the delayed reference count updates and
2056 * extent insertions we have queued up so far. count can be
2057 * 0, which means to process everything in the tree at the start
2058 * of the run (but not newly added entries), or it can be some target
2059 * number you'd like to process.
2061 int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2062 struct btrfs_root *root, unsigned long count)
2064 struct rb_node *node;
2065 struct btrfs_delayed_ref_root *delayed_refs;
2066 struct btrfs_delayed_ref_node *ref;
2067 struct list_head cluster;
2069 int run_all = count == (unsigned long)-1;
2072 if (root == root->fs_info->extent_root)
2073 root = root->fs_info->tree_root;
2075 delayed_refs = &trans->transaction->delayed_refs;
2076 INIT_LIST_HEAD(&cluster);
2078 spin_lock(&delayed_refs->lock);
2080 count = delayed_refs->num_entries * 2;
2084 if (!(run_all || run_most) &&
2085 delayed_refs->num_heads_ready < 64)
2089 * go find something we can process in the rbtree. We start at
2090 * the beginning of the tree, and then build a cluster
2091 * of refs to process starting at the first one we are able to
2094 ret = btrfs_find_ref_cluster(trans, &cluster,
2095 delayed_refs->run_delayed_start);
2099 ret = run_clustered_refs(trans, root, &cluster);
2102 count -= min_t(unsigned long, ret, count);
2109 node = rb_first(&delayed_refs->root);
2112 count = (unsigned long)-1;
2115 ref = rb_entry(node, struct btrfs_delayed_ref_node,
2117 if (btrfs_delayed_ref_is_head(ref)) {
2118 struct btrfs_delayed_ref_head *head;
2120 head = btrfs_delayed_node_to_head(ref);
2121 atomic_inc(&ref->refs);
2123 spin_unlock(&delayed_refs->lock);
2124 mutex_lock(&head->mutex);
2125 mutex_unlock(&head->mutex);
2127 btrfs_put_delayed_ref(ref);
2131 node = rb_next(node);
2133 spin_unlock(&delayed_refs->lock);
2134 schedule_timeout(1);
2138 spin_unlock(&delayed_refs->lock);
2142 int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans,
2143 struct btrfs_root *root,
2144 u64 bytenr, u64 num_bytes, u64 flags,
2147 struct btrfs_delayed_extent_op *extent_op;
2150 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
2154 extent_op->flags_to_set = flags;
2155 extent_op->update_flags = 1;
2156 extent_op->update_key = 0;
2157 extent_op->is_data = is_data ? 1 : 0;
2159 ret = btrfs_add_delayed_extent_op(trans, bytenr, num_bytes, extent_op);
2165 static noinline int check_delayed_ref(struct btrfs_trans_handle *trans,
2166 struct btrfs_root *root,
2167 struct btrfs_path *path,
2168 u64 objectid, u64 offset, u64 bytenr)
2170 struct btrfs_delayed_ref_head *head;
2171 struct btrfs_delayed_ref_node *ref;
2172 struct btrfs_delayed_data_ref *data_ref;
2173 struct btrfs_delayed_ref_root *delayed_refs;
2174 struct rb_node *node;
2178 delayed_refs = &trans->transaction->delayed_refs;
2179 spin_lock(&delayed_refs->lock);
2180 head = btrfs_find_delayed_ref_head(trans, bytenr);
2184 if (!mutex_trylock(&head->mutex)) {
2185 atomic_inc(&head->node.refs);
2186 spin_unlock(&delayed_refs->lock);
2188 btrfs_release_path(root->fs_info->extent_root, path);
2190 mutex_lock(&head->mutex);
2191 mutex_unlock(&head->mutex);
2192 btrfs_put_delayed_ref(&head->node);
2196 node = rb_prev(&head->node.rb_node);
2200 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2202 if (ref->bytenr != bytenr)
2206 if (ref->type != BTRFS_EXTENT_DATA_REF_KEY)
2209 data_ref = btrfs_delayed_node_to_data_ref(ref);
2211 node = rb_prev(node);
2213 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2214 if (ref->bytenr == bytenr)
2218 if (data_ref->root != root->root_key.objectid ||
2219 data_ref->objectid != objectid || data_ref->offset != offset)
2224 mutex_unlock(&head->mutex);
2226 spin_unlock(&delayed_refs->lock);
2230 static noinline int check_committed_ref(struct btrfs_trans_handle *trans,
2231 struct btrfs_root *root,
2232 struct btrfs_path *path,
2233 u64 objectid, u64 offset, u64 bytenr)
2235 struct btrfs_root *extent_root = root->fs_info->extent_root;
2236 struct extent_buffer *leaf;
2237 struct btrfs_extent_data_ref *ref;
2238 struct btrfs_extent_inline_ref *iref;
2239 struct btrfs_extent_item *ei;
2240 struct btrfs_key key;
2244 key.objectid = bytenr;
2245 key.offset = (u64)-1;
2246 key.type = BTRFS_EXTENT_ITEM_KEY;
2248 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
2254 if (path->slots[0] == 0)
2258 leaf = path->nodes[0];
2259 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2261 if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY)
2265 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2266 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2267 if (item_size < sizeof(*ei)) {
2268 WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0));
2272 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2274 if (item_size != sizeof(*ei) +
2275 btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY))
2278 if (btrfs_extent_generation(leaf, ei) <=
2279 btrfs_root_last_snapshot(&root->root_item))
2282 iref = (struct btrfs_extent_inline_ref *)(ei + 1);
2283 if (btrfs_extent_inline_ref_type(leaf, iref) !=
2284 BTRFS_EXTENT_DATA_REF_KEY)
2287 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
2288 if (btrfs_extent_refs(leaf, ei) !=
2289 btrfs_extent_data_ref_count(leaf, ref) ||
2290 btrfs_extent_data_ref_root(leaf, ref) !=
2291 root->root_key.objectid ||
2292 btrfs_extent_data_ref_objectid(leaf, ref) != objectid ||
2293 btrfs_extent_data_ref_offset(leaf, ref) != offset)
2301 int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans,
2302 struct btrfs_root *root,
2303 u64 objectid, u64 offset, u64 bytenr)
2305 struct btrfs_path *path;
2309 path = btrfs_alloc_path();
2314 ret = check_committed_ref(trans, root, path, objectid,
2316 if (ret && ret != -ENOENT)
2319 ret2 = check_delayed_ref(trans, root, path, objectid,
2321 } while (ret2 == -EAGAIN);
2323 if (ret2 && ret2 != -ENOENT) {
2328 if (ret != -ENOENT || ret2 != -ENOENT)
2331 btrfs_free_path(path);
2336 int btrfs_cache_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2337 struct extent_buffer *buf, u32 nr_extents)
2339 struct btrfs_key key;
2340 struct btrfs_file_extent_item *fi;
2348 if (!root->ref_cows)
2351 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
2353 root_gen = root->root_key.offset;
2356 root_gen = trans->transid - 1;
2359 level = btrfs_header_level(buf);
2360 nritems = btrfs_header_nritems(buf);
2363 struct btrfs_leaf_ref *ref;
2364 struct btrfs_extent_info *info;
2366 ref = btrfs_alloc_leaf_ref(root, nr_extents);
2372 ref->root_gen = root_gen;
2373 ref->bytenr = buf->start;
2374 ref->owner = btrfs_header_owner(buf);
2375 ref->generation = btrfs_header_generation(buf);
2376 ref->nritems = nr_extents;
2377 info = ref->extents;
2379 for (i = 0; nr_extents > 0 && i < nritems; i++) {
2381 btrfs_item_key_to_cpu(buf, &key, i);
2382 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
2384 fi = btrfs_item_ptr(buf, i,
2385 struct btrfs_file_extent_item);
2386 if (btrfs_file_extent_type(buf, fi) ==
2387 BTRFS_FILE_EXTENT_INLINE)
2389 disk_bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
2390 if (disk_bytenr == 0)
2393 info->bytenr = disk_bytenr;
2395 btrfs_file_extent_disk_num_bytes(buf, fi);
2396 info->objectid = key.objectid;
2397 info->offset = key.offset;
2401 ret = btrfs_add_leaf_ref(root, ref, shared);
2402 if (ret == -EEXIST && shared) {
2403 struct btrfs_leaf_ref *old;
2404 old = btrfs_lookup_leaf_ref(root, ref->bytenr);
2406 btrfs_remove_leaf_ref(root, old);
2407 btrfs_free_leaf_ref(root, old);
2408 ret = btrfs_add_leaf_ref(root, ref, shared);
2411 btrfs_free_leaf_ref(root, ref);
2417 /* when a block goes through cow, we update the reference counts of
2418 * everything that block points to. The internal pointers of the block
2419 * can be in just about any order, and it is likely to have clusters of
2420 * things that are close together and clusters of things that are not.
2422 * To help reduce the seeks that come with updating all of these reference
2423 * counts, sort them by byte number before actual updates are done.
2425 * struct refsort is used to match byte number to slot in the btree block.
2426 * we sort based on the byte number and then use the slot to actually
2429 * struct refsort is smaller than strcut btrfs_item and smaller than
2430 * struct btrfs_key_ptr. Since we're currently limited to the page size
2431 * for a btree block, there's no way for a kmalloc of refsorts for a
2432 * single node to be bigger than a page.
2440 * for passing into sort()
2442 static int refsort_cmp(const void *a_void, const void *b_void)
2444 const struct refsort *a = a_void;
2445 const struct refsort *b = b_void;
2447 if (a->bytenr < b->bytenr)
2449 if (a->bytenr > b->bytenr)
2455 static int __btrfs_mod_ref(struct btrfs_trans_handle *trans,
2456 struct btrfs_root *root,
2457 struct extent_buffer *buf,
2458 int full_backref, int inc)
2465 struct btrfs_key key;
2466 struct btrfs_file_extent_item *fi;
2470 int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *,
2471 u64, u64, u64, u64, u64, u64);
2473 ref_root = btrfs_header_owner(buf);
2474 nritems = btrfs_header_nritems(buf);
2475 level = btrfs_header_level(buf);
2477 if (!root->ref_cows && level == 0)
2481 process_func = btrfs_inc_extent_ref;
2483 process_func = btrfs_free_extent;
2486 parent = buf->start;
2490 for (i = 0; i < nritems; i++) {
2492 btrfs_item_key_to_cpu(buf, &key, i);
2493 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
2495 fi = btrfs_item_ptr(buf, i,
2496 struct btrfs_file_extent_item);
2497 if (btrfs_file_extent_type(buf, fi) ==
2498 BTRFS_FILE_EXTENT_INLINE)
2500 bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
2504 num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi);
2505 key.offset -= btrfs_file_extent_offset(buf, fi);
2506 ret = process_func(trans, root, bytenr, num_bytes,
2507 parent, ref_root, key.objectid,
2512 bytenr = btrfs_node_blockptr(buf, i);
2513 num_bytes = btrfs_level_size(root, level - 1);
2514 ret = process_func(trans, root, bytenr, num_bytes,
2515 parent, ref_root, level - 1, 0);
2526 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2527 struct extent_buffer *buf, int full_backref)
2529 return __btrfs_mod_ref(trans, root, buf, full_backref, 1);
2532 int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2533 struct extent_buffer *buf, int full_backref)
2535 return __btrfs_mod_ref(trans, root, buf, full_backref, 0);
2538 static int write_one_cache_group(struct btrfs_trans_handle *trans,
2539 struct btrfs_root *root,
2540 struct btrfs_path *path,
2541 struct btrfs_block_group_cache *cache)
2544 struct btrfs_root *extent_root = root->fs_info->extent_root;
2546 struct extent_buffer *leaf;
2548 ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
2553 leaf = path->nodes[0];
2554 bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
2555 write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
2556 btrfs_mark_buffer_dirty(leaf);
2557 btrfs_release_path(extent_root, path);
2565 static struct btrfs_block_group_cache *
2566 next_block_group(struct btrfs_root *root,
2567 struct btrfs_block_group_cache *cache)
2569 struct rb_node *node;
2570 spin_lock(&root->fs_info->block_group_cache_lock);
2571 node = rb_next(&cache->cache_node);
2572 btrfs_put_block_group(cache);
2574 cache = rb_entry(node, struct btrfs_block_group_cache,
2576 atomic_inc(&cache->count);
2579 spin_unlock(&root->fs_info->block_group_cache_lock);
2583 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
2584 struct btrfs_root *root)
2586 struct btrfs_block_group_cache *cache;
2588 struct btrfs_path *path;
2591 path = btrfs_alloc_path();
2597 err = btrfs_run_delayed_refs(trans, root,
2602 cache = btrfs_lookup_first_block_group(root->fs_info, last);
2606 cache = next_block_group(root, cache);
2616 last = cache->key.objectid + cache->key.offset;
2618 err = write_one_cache_group(trans, root, path, cache);
2620 btrfs_put_block_group(cache);
2623 btrfs_free_path(path);
2627 int btrfs_extent_readonly(struct btrfs_root *root, u64 bytenr)
2629 struct btrfs_block_group_cache *block_group;
2632 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
2633 if (!block_group || block_group->ro)
2636 btrfs_put_block_group(block_group);
2640 static int update_space_info(struct btrfs_fs_info *info, u64 flags,
2641 u64 total_bytes, u64 bytes_used,
2642 struct btrfs_space_info **space_info)
2644 struct btrfs_space_info *found;
2646 found = __find_space_info(info, flags);
2648 spin_lock(&found->lock);
2649 found->total_bytes += total_bytes;
2650 found->bytes_used += bytes_used;
2652 spin_unlock(&found->lock);
2653 *space_info = found;
2656 found = kzalloc(sizeof(*found), GFP_NOFS);
2660 INIT_LIST_HEAD(&found->block_groups);
2661 init_rwsem(&found->groups_sem);
2662 spin_lock_init(&found->lock);
2663 found->flags = flags;
2664 found->total_bytes = total_bytes;
2665 found->bytes_used = bytes_used;
2666 found->bytes_pinned = 0;
2667 found->bytes_reserved = 0;
2668 found->bytes_readonly = 0;
2669 found->bytes_delalloc = 0;
2671 found->force_alloc = 0;
2672 *space_info = found;
2673 list_add_rcu(&found->list, &info->space_info);
2674 atomic_set(&found->caching_threads, 0);
2678 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
2680 u64 extra_flags = flags & (BTRFS_BLOCK_GROUP_RAID0 |
2681 BTRFS_BLOCK_GROUP_RAID1 |
2682 BTRFS_BLOCK_GROUP_RAID10 |
2683 BTRFS_BLOCK_GROUP_DUP);
2685 if (flags & BTRFS_BLOCK_GROUP_DATA)
2686 fs_info->avail_data_alloc_bits |= extra_flags;
2687 if (flags & BTRFS_BLOCK_GROUP_METADATA)
2688 fs_info->avail_metadata_alloc_bits |= extra_flags;
2689 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
2690 fs_info->avail_system_alloc_bits |= extra_flags;
2694 static void set_block_group_readonly(struct btrfs_block_group_cache *cache)
2696 spin_lock(&cache->space_info->lock);
2697 spin_lock(&cache->lock);
2699 cache->space_info->bytes_readonly += cache->key.offset -
2700 btrfs_block_group_used(&cache->item);
2703 spin_unlock(&cache->lock);
2704 spin_unlock(&cache->space_info->lock);
2707 u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags)
2709 u64 num_devices = root->fs_info->fs_devices->rw_devices;
2711 if (num_devices == 1)
2712 flags &= ~(BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID0);
2713 if (num_devices < 4)
2714 flags &= ~BTRFS_BLOCK_GROUP_RAID10;
2716 if ((flags & BTRFS_BLOCK_GROUP_DUP) &&
2717 (flags & (BTRFS_BLOCK_GROUP_RAID1 |
2718 BTRFS_BLOCK_GROUP_RAID10))) {
2719 flags &= ~BTRFS_BLOCK_GROUP_DUP;
2722 if ((flags & BTRFS_BLOCK_GROUP_RAID1) &&
2723 (flags & BTRFS_BLOCK_GROUP_RAID10)) {
2724 flags &= ~BTRFS_BLOCK_GROUP_RAID1;
2727 if ((flags & BTRFS_BLOCK_GROUP_RAID0) &&
2728 ((flags & BTRFS_BLOCK_GROUP_RAID1) |
2729 (flags & BTRFS_BLOCK_GROUP_RAID10) |
2730 (flags & BTRFS_BLOCK_GROUP_DUP)))
2731 flags &= ~BTRFS_BLOCK_GROUP_RAID0;
2735 static u64 btrfs_get_alloc_profile(struct btrfs_root *root, u64 data)
2737 struct btrfs_fs_info *info = root->fs_info;
2741 alloc_profile = info->avail_data_alloc_bits &
2742 info->data_alloc_profile;
2743 data = BTRFS_BLOCK_GROUP_DATA | alloc_profile;
2744 } else if (root == root->fs_info->chunk_root) {
2745 alloc_profile = info->avail_system_alloc_bits &
2746 info->system_alloc_profile;
2747 data = BTRFS_BLOCK_GROUP_SYSTEM | alloc_profile;
2749 alloc_profile = info->avail_metadata_alloc_bits &
2750 info->metadata_alloc_profile;
2751 data = BTRFS_BLOCK_GROUP_METADATA | alloc_profile;
2754 return btrfs_reduce_alloc_profile(root, data);
2757 void btrfs_set_inode_space_info(struct btrfs_root *root, struct inode *inode)
2761 alloc_target = btrfs_get_alloc_profile(root, 1);
2762 BTRFS_I(inode)->space_info = __find_space_info(root->fs_info,
2766 static u64 calculate_bytes_needed(struct btrfs_root *root, int num_items)
2771 level = BTRFS_MAX_LEVEL - 2;
2773 * NOTE: these calculations are absolutely the worst possible case.
2774 * This assumes that _every_ item we insert will require a new leaf, and
2775 * that the tree has grown to its maximum level size.
2779 * for every item we insert we could insert both an extent item and a
2780 * extent ref item. Then for ever item we insert, we will need to cow
2781 * both the original leaf, plus the leaf to the left and right of it.
2783 * Unless we are talking about the extent root, then we just want the
2784 * number of items * 2, since we just need the extent item plus its ref.
2786 if (root == root->fs_info->extent_root)
2787 num_bytes = num_items * 2;
2789 num_bytes = (num_items + (2 * num_items)) * 3;
2792 * num_bytes is total number of leaves we could need times the leaf
2793 * size, and then for every leaf we could end up cow'ing 2 nodes per
2794 * level, down to the leaf level.
2796 num_bytes = (num_bytes * root->leafsize) +
2797 (num_bytes * (level * 2)) * root->nodesize;
2803 * Unreserve metadata space for delalloc. If we have less reserved credits than
2804 * we have extents, this function does nothing.
2806 int btrfs_unreserve_metadata_for_delalloc(struct btrfs_root *root,
2807 struct inode *inode, int num_items)
2809 struct btrfs_fs_info *info = root->fs_info;
2810 struct btrfs_space_info *meta_sinfo;
2815 /* get the space info for where the metadata will live */
2816 alloc_target = btrfs_get_alloc_profile(root, 0);
2817 meta_sinfo = __find_space_info(info, alloc_target);
2819 num_bytes = calculate_bytes_needed(root->fs_info->extent_root,
2822 spin_lock(&meta_sinfo->lock);
2823 spin_lock(&BTRFS_I(inode)->accounting_lock);
2824 if (BTRFS_I(inode)->reserved_extents <=
2825 BTRFS_I(inode)->outstanding_extents) {
2826 spin_unlock(&BTRFS_I(inode)->accounting_lock);
2827 spin_unlock(&meta_sinfo->lock);
2830 spin_unlock(&BTRFS_I(inode)->accounting_lock);
2832 BTRFS_I(inode)->reserved_extents--;
2833 BUG_ON(BTRFS_I(inode)->reserved_extents < 0);
2835 if (meta_sinfo->bytes_delalloc < num_bytes) {
2837 meta_sinfo->bytes_delalloc = 0;
2839 meta_sinfo->bytes_delalloc -= num_bytes;
2841 spin_unlock(&meta_sinfo->lock);
2848 static void check_force_delalloc(struct btrfs_space_info *meta_sinfo)
2852 thresh = meta_sinfo->bytes_used + meta_sinfo->bytes_reserved +
2853 meta_sinfo->bytes_pinned + meta_sinfo->bytes_readonly +
2854 meta_sinfo->bytes_super + meta_sinfo->bytes_root +
2855 meta_sinfo->bytes_may_use;
2857 thresh = meta_sinfo->total_bytes - thresh;
2859 do_div(thresh, 100);
2860 if (thresh <= meta_sinfo->bytes_delalloc)
2861 meta_sinfo->force_delalloc = 1;
2863 meta_sinfo->force_delalloc = 0;
2866 struct async_flush {
2867 struct btrfs_root *root;
2868 struct btrfs_space_info *info;
2869 struct btrfs_work work;
2872 static noinline void flush_delalloc_async(struct btrfs_work *work)
2874 struct async_flush *async;
2875 struct btrfs_root *root;
2876 struct btrfs_space_info *info;
2878 async = container_of(work, struct async_flush, work);
2882 btrfs_start_delalloc_inodes(root);
2883 wake_up(&info->flush_wait);
2884 btrfs_wait_ordered_extents(root, 0);
2886 spin_lock(&info->lock);
2888 spin_unlock(&info->lock);
2889 wake_up(&info->flush_wait);
2894 static void wait_on_flush(struct btrfs_space_info *info)
2900 prepare_to_wait(&info->flush_wait, &wait,
2901 TASK_UNINTERRUPTIBLE);
2902 spin_lock(&info->lock);
2903 if (!info->flushing) {
2904 spin_unlock(&info->lock);
2908 used = info->bytes_used + info->bytes_reserved +
2909 info->bytes_pinned + info->bytes_readonly +
2910 info->bytes_super + info->bytes_root +
2911 info->bytes_may_use + info->bytes_delalloc;
2912 if (used < info->total_bytes) {
2913 spin_unlock(&info->lock);
2916 spin_unlock(&info->lock);
2919 finish_wait(&info->flush_wait, &wait);
2922 static void flush_delalloc(struct btrfs_root *root,
2923 struct btrfs_space_info *info)
2925 struct async_flush *async;
2928 spin_lock(&info->lock);
2930 if (!info->flushing) {
2932 init_waitqueue_head(&info->flush_wait);
2937 spin_unlock(&info->lock);
2940 wait_on_flush(info);
2944 async = kzalloc(sizeof(*async), GFP_NOFS);
2950 async->work.func = flush_delalloc_async;
2952 btrfs_queue_worker(&root->fs_info->enospc_workers,
2954 wait_on_flush(info);
2958 btrfs_start_delalloc_inodes(root);
2959 btrfs_wait_ordered_extents(root, 0);
2961 spin_lock(&info->lock);
2963 spin_unlock(&info->lock);
2964 wake_up(&info->flush_wait);
2967 static int maybe_allocate_chunk(struct btrfs_root *root,
2968 struct btrfs_space_info *info)
2970 struct btrfs_super_block *disk_super = &root->fs_info->super_copy;
2971 struct btrfs_trans_handle *trans;
2977 free_space = btrfs_super_total_bytes(disk_super);
2979 * we allow the metadata to grow to a max of either 5gb or 5% of the
2980 * space in the volume.
2982 min_metadata = min((u64)5 * 1024 * 1024 * 1024,
2983 div64_u64(free_space * 5, 100));
2984 if (info->total_bytes >= min_metadata) {
2985 spin_unlock(&info->lock);
2990 spin_unlock(&info->lock);
2994 if (!info->allocating_chunk) {
2995 info->force_alloc = 1;
2996 info->allocating_chunk = 1;
2997 init_waitqueue_head(&info->allocate_wait);
3002 spin_unlock(&info->lock);
3005 wait_event(info->allocate_wait,
3006 !info->allocating_chunk);
3010 trans = btrfs_start_transaction(root, 1);
3016 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
3017 4096 + 2 * 1024 * 1024,
3019 btrfs_end_transaction(trans, root);
3023 spin_lock(&info->lock);
3024 info->allocating_chunk = 0;
3025 spin_unlock(&info->lock);
3026 wake_up(&info->allocate_wait);
3034 * Reserve metadata space for delalloc.
3036 int btrfs_reserve_metadata_for_delalloc(struct btrfs_root *root,
3037 struct inode *inode, int num_items)
3039 struct btrfs_fs_info *info = root->fs_info;
3040 struct btrfs_space_info *meta_sinfo;
3047 /* get the space info for where the metadata will live */
3048 alloc_target = btrfs_get_alloc_profile(root, 0);
3049 meta_sinfo = __find_space_info(info, alloc_target);
3051 num_bytes = calculate_bytes_needed(root->fs_info->extent_root,
3054 spin_lock(&meta_sinfo->lock);
3056 force_delalloc = meta_sinfo->force_delalloc;
3058 if (unlikely(!meta_sinfo->bytes_root))
3059 meta_sinfo->bytes_root = calculate_bytes_needed(root, 6);
3062 meta_sinfo->bytes_delalloc += num_bytes;
3064 used = meta_sinfo->bytes_used + meta_sinfo->bytes_reserved +
3065 meta_sinfo->bytes_pinned + meta_sinfo->bytes_readonly +
3066 meta_sinfo->bytes_super + meta_sinfo->bytes_root +
3067 meta_sinfo->bytes_may_use + meta_sinfo->bytes_delalloc;
3069 if (used > meta_sinfo->total_bytes) {
3073 if (maybe_allocate_chunk(root, meta_sinfo))
3077 spin_unlock(&meta_sinfo->lock);
3081 filemap_flush(inode->i_mapping);
3083 } else if (flushed == 3) {
3084 flush_delalloc(root, meta_sinfo);
3087 spin_lock(&meta_sinfo->lock);
3088 meta_sinfo->bytes_delalloc -= num_bytes;
3089 spin_unlock(&meta_sinfo->lock);
3090 printk(KERN_ERR "enospc, has %d, reserved %d\n",
3091 BTRFS_I(inode)->outstanding_extents,
3092 BTRFS_I(inode)->reserved_extents);
3093 dump_space_info(meta_sinfo, 0, 0);
3097 BTRFS_I(inode)->reserved_extents++;
3098 check_force_delalloc(meta_sinfo);
3099 spin_unlock(&meta_sinfo->lock);
3101 if (!flushed && force_delalloc)
3102 filemap_flush(inode->i_mapping);
3108 * unreserve num_items number of items worth of metadata space. This needs to
3109 * be paired with btrfs_reserve_metadata_space.
3111 * NOTE: if you have the option, run this _AFTER_ you do a
3112 * btrfs_end_transaction, since btrfs_end_transaction will run delayed ref
3113 * oprations which will result in more used metadata, so we want to make sure we
3114 * can do that without issue.
3116 int btrfs_unreserve_metadata_space(struct btrfs_root *root, int num_items)
3118 struct btrfs_fs_info *info = root->fs_info;
3119 struct btrfs_space_info *meta_sinfo;
3124 /* get the space info for where the metadata will live */
3125 alloc_target = btrfs_get_alloc_profile(root, 0);
3126 meta_sinfo = __find_space_info(info, alloc_target);
3128 num_bytes = calculate_bytes_needed(root, num_items);
3130 spin_lock(&meta_sinfo->lock);
3131 if (meta_sinfo->bytes_may_use < num_bytes) {
3133 meta_sinfo->bytes_may_use = 0;
3135 meta_sinfo->bytes_may_use -= num_bytes;
3137 spin_unlock(&meta_sinfo->lock);
3145 * Reserve some metadata space for use. We'll calculate the worste case number
3146 * of bytes that would be needed to modify num_items number of items. If we
3147 * have space, fantastic, if not, you get -ENOSPC. Please call
3148 * btrfs_unreserve_metadata_space when you are done for the _SAME_ number of
3149 * items you reserved, since whatever metadata you needed should have already
3152 * This will commit the transaction to make more space if we don't have enough
3153 * metadata space. THe only time we don't do this is if we're reserving space
3154 * inside of a transaction, then we will just return -ENOSPC and it is the
3155 * callers responsibility to handle it properly.
3157 int btrfs_reserve_metadata_space(struct btrfs_root *root, int num_items)
3159 struct btrfs_fs_info *info = root->fs_info;
3160 struct btrfs_space_info *meta_sinfo;
3166 /* get the space info for where the metadata will live */
3167 alloc_target = btrfs_get_alloc_profile(root, 0);
3168 meta_sinfo = __find_space_info(info, alloc_target);
3170 num_bytes = calculate_bytes_needed(root, num_items);
3172 spin_lock(&meta_sinfo->lock);
3174 if (unlikely(!meta_sinfo->bytes_root))
3175 meta_sinfo->bytes_root = calculate_bytes_needed(root, 6);
3178 meta_sinfo->bytes_may_use += num_bytes;
3180 used = meta_sinfo->bytes_used + meta_sinfo->bytes_reserved +
3181 meta_sinfo->bytes_pinned + meta_sinfo->bytes_readonly +
3182 meta_sinfo->bytes_super + meta_sinfo->bytes_root +
3183 meta_sinfo->bytes_may_use + meta_sinfo->bytes_delalloc;
3185 if (used > meta_sinfo->total_bytes) {
3188 if (maybe_allocate_chunk(root, meta_sinfo))
3192 spin_unlock(&meta_sinfo->lock);
3196 flush_delalloc(root, meta_sinfo);
3199 spin_lock(&meta_sinfo->lock);
3200 meta_sinfo->bytes_may_use -= num_bytes;
3201 spin_unlock(&meta_sinfo->lock);
3203 dump_space_info(meta_sinfo, 0, 0);
3207 check_force_delalloc(meta_sinfo);
3208 spin_unlock(&meta_sinfo->lock);
3214 * This will check the space that the inode allocates from to make sure we have
3215 * enough space for bytes.
3217 int btrfs_check_data_free_space(struct btrfs_root *root, struct inode *inode,
3220 struct btrfs_space_info *data_sinfo;
3221 int ret = 0, committed = 0;
3223 /* make sure bytes are sectorsize aligned */
3224 bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
3226 data_sinfo = BTRFS_I(inode)->space_info;
3231 /* make sure we have enough space to handle the data first */
3232 spin_lock(&data_sinfo->lock);
3233 if (data_sinfo->total_bytes - data_sinfo->bytes_used -
3234 data_sinfo->bytes_delalloc - data_sinfo->bytes_reserved -
3235 data_sinfo->bytes_pinned - data_sinfo->bytes_readonly -
3236 data_sinfo->bytes_may_use - data_sinfo->bytes_super < bytes) {
3237 struct btrfs_trans_handle *trans;
3240 * if we don't have enough free bytes in this space then we need
3241 * to alloc a new chunk.
3243 if (!data_sinfo->full) {
3246 data_sinfo->force_alloc = 1;
3247 spin_unlock(&data_sinfo->lock);
3249 alloc_target = btrfs_get_alloc_profile(root, 1);
3250 trans = btrfs_start_transaction(root, 1);
3254 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
3255 bytes + 2 * 1024 * 1024,
3257 btrfs_end_transaction(trans, root);
3262 btrfs_set_inode_space_info(root, inode);
3263 data_sinfo = BTRFS_I(inode)->space_info;
3267 spin_unlock(&data_sinfo->lock);
3269 /* commit the current transaction and try again */
3270 if (!committed && !root->fs_info->open_ioctl_trans) {
3272 trans = btrfs_join_transaction(root, 1);
3275 ret = btrfs_commit_transaction(trans, root);
3281 printk(KERN_ERR "no space left, need %llu, %llu delalloc bytes"
3282 ", %llu bytes_used, %llu bytes_reserved, "
3283 "%llu bytes_pinned, %llu bytes_readonly, %llu may use "
3284 "%llu total\n", (unsigned long long)bytes,
3285 (unsigned long long)data_sinfo->bytes_delalloc,
3286 (unsigned long long)data_sinfo->bytes_used,
3287 (unsigned long long)data_sinfo->bytes_reserved,
3288 (unsigned long long)data_sinfo->bytes_pinned,
3289 (unsigned long long)data_sinfo->bytes_readonly,
3290 (unsigned long long)data_sinfo->bytes_may_use,
3291 (unsigned long long)data_sinfo->total_bytes);
3294 data_sinfo->bytes_may_use += bytes;
3295 BTRFS_I(inode)->reserved_bytes += bytes;
3296 spin_unlock(&data_sinfo->lock);
3302 * if there was an error for whatever reason after calling
3303 * btrfs_check_data_free_space, call this so we can cleanup the counters.
3305 void btrfs_free_reserved_data_space(struct btrfs_root *root,
3306 struct inode *inode, u64 bytes)
3308 struct btrfs_space_info *data_sinfo;
3310 /* make sure bytes are sectorsize aligned */
3311 bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
3313 data_sinfo = BTRFS_I(inode)->space_info;
3314 spin_lock(&data_sinfo->lock);
3315 data_sinfo->bytes_may_use -= bytes;
3316 BTRFS_I(inode)->reserved_bytes -= bytes;
3317 spin_unlock(&data_sinfo->lock);
3320 /* called when we are adding a delalloc extent to the inode's io_tree */
3321 void btrfs_delalloc_reserve_space(struct btrfs_root *root, struct inode *inode,
3324 struct btrfs_space_info *data_sinfo;
3326 /* get the space info for where this inode will be storing its data */
3327 data_sinfo = BTRFS_I(inode)->space_info;
3329 /* make sure we have enough space to handle the data first */
3330 spin_lock(&data_sinfo->lock);
3331 data_sinfo->bytes_delalloc += bytes;
3334 * we are adding a delalloc extent without calling
3335 * btrfs_check_data_free_space first. This happens on a weird
3336 * writepage condition, but shouldn't hurt our accounting
3338 if (unlikely(bytes > BTRFS_I(inode)->reserved_bytes)) {
3339 data_sinfo->bytes_may_use -= BTRFS_I(inode)->reserved_bytes;
3340 BTRFS_I(inode)->reserved_bytes = 0;
3342 data_sinfo->bytes_may_use -= bytes;
3343 BTRFS_I(inode)->reserved_bytes -= bytes;
3346 spin_unlock(&data_sinfo->lock);
3349 /* called when we are clearing an delalloc extent from the inode's io_tree */
3350 void btrfs_delalloc_free_space(struct btrfs_root *root, struct inode *inode,
3353 struct btrfs_space_info *info;
3355 info = BTRFS_I(inode)->space_info;
3357 spin_lock(&info->lock);
3358 info->bytes_delalloc -= bytes;
3359 spin_unlock(&info->lock);
3362 static void force_metadata_allocation(struct btrfs_fs_info *info)
3364 struct list_head *head = &info->space_info;
3365 struct btrfs_space_info *found;
3368 list_for_each_entry_rcu(found, head, list) {
3369 if (found->flags & BTRFS_BLOCK_GROUP_METADATA)
3370 found->force_alloc = 1;
3375 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
3376 struct btrfs_root *extent_root, u64 alloc_bytes,
3377 u64 flags, int force)
3379 struct btrfs_space_info *space_info;
3380 struct btrfs_fs_info *fs_info = extent_root->fs_info;
3384 mutex_lock(&fs_info->chunk_mutex);
3386 flags = btrfs_reduce_alloc_profile(extent_root, flags);
3388 space_info = __find_space_info(extent_root->fs_info, flags);
3390 ret = update_space_info(extent_root->fs_info, flags,
3394 BUG_ON(!space_info);
3396 spin_lock(&space_info->lock);
3397 if (space_info->force_alloc)
3399 if (space_info->full) {
3400 spin_unlock(&space_info->lock);
3404 thresh = space_info->total_bytes - space_info->bytes_readonly;
3405 thresh = div_factor(thresh, 8);
3407 (space_info->bytes_used + space_info->bytes_pinned +
3408 space_info->bytes_reserved + alloc_bytes) < thresh) {
3409 spin_unlock(&space_info->lock);
3412 spin_unlock(&space_info->lock);
3415 * if we're doing a data chunk, go ahead and make sure that
3416 * we keep a reasonable number of metadata chunks allocated in the
3419 if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) {
3420 fs_info->data_chunk_allocations++;
3421 if (!(fs_info->data_chunk_allocations %
3422 fs_info->metadata_ratio))
3423 force_metadata_allocation(fs_info);
3426 ret = btrfs_alloc_chunk(trans, extent_root, flags);
3427 spin_lock(&space_info->lock);
3429 space_info->full = 1;
3430 space_info->force_alloc = 0;
3431 spin_unlock(&space_info->lock);
3433 mutex_unlock(&extent_root->fs_info->chunk_mutex);
3437 static int update_block_group(struct btrfs_trans_handle *trans,
3438 struct btrfs_root *root,
3439 u64 bytenr, u64 num_bytes, int alloc,
3442 struct btrfs_block_group_cache *cache;
3443 struct btrfs_fs_info *info = root->fs_info;
3444 u64 total = num_bytes;
3448 /* block accounting for super block */
3449 spin_lock(&info->delalloc_lock);
3450 old_val = btrfs_super_bytes_used(&info->super_copy);
3452 old_val += num_bytes;
3454 old_val -= num_bytes;
3455 btrfs_set_super_bytes_used(&info->super_copy, old_val);
3457 /* block accounting for root item */
3458 old_val = btrfs_root_used(&root->root_item);
3460 old_val += num_bytes;
3462 old_val -= num_bytes;
3463 btrfs_set_root_used(&root->root_item, old_val);
3464 spin_unlock(&info->delalloc_lock);
3467 cache = btrfs_lookup_block_group(info, bytenr);
3470 byte_in_group = bytenr - cache->key.objectid;
3471 WARN_ON(byte_in_group > cache->key.offset);
3473 spin_lock(&cache->space_info->lock);
3474 spin_lock(&cache->lock);
3476 old_val = btrfs_block_group_used(&cache->item);
3477 num_bytes = min(total, cache->key.offset - byte_in_group);
3479 old_val += num_bytes;
3480 btrfs_set_block_group_used(&cache->item, old_val);
3481 cache->reserved -= num_bytes;
3482 cache->space_info->bytes_used += num_bytes;
3483 cache->space_info->bytes_reserved -= num_bytes;
3485 cache->space_info->bytes_readonly -= num_bytes;
3486 spin_unlock(&cache->lock);
3487 spin_unlock(&cache->space_info->lock);
3489 old_val -= num_bytes;
3490 cache->space_info->bytes_used -= num_bytes;
3492 cache->space_info->bytes_readonly += num_bytes;
3493 btrfs_set_block_group_used(&cache->item, old_val);
3494 spin_unlock(&cache->lock);
3495 spin_unlock(&cache->space_info->lock);
3499 ret = btrfs_discard_extent(root, bytenr,
3503 ret = btrfs_add_free_space(cache, bytenr,
3508 btrfs_put_block_group(cache);
3510 bytenr += num_bytes;
3515 static u64 first_logical_byte(struct btrfs_root *root, u64 search_start)
3517 struct btrfs_block_group_cache *cache;
3520 cache = btrfs_lookup_first_block_group(root->fs_info, search_start);
3524 bytenr = cache->key.objectid;
3525 btrfs_put_block_group(cache);
3531 * this function must be called within transaction
3533 int btrfs_pin_extent(struct btrfs_root *root,
3534 u64 bytenr, u64 num_bytes, int reserved)
3536 struct btrfs_fs_info *fs_info = root->fs_info;
3537 struct btrfs_block_group_cache *cache;
3539 cache = btrfs_lookup_block_group(fs_info, bytenr);
3542 spin_lock(&cache->space_info->lock);
3543 spin_lock(&cache->lock);
3544 cache->pinned += num_bytes;
3545 cache->space_info->bytes_pinned += num_bytes;
3547 cache->reserved -= num_bytes;
3548 cache->space_info->bytes_reserved -= num_bytes;
3550 spin_unlock(&cache->lock);
3551 spin_unlock(&cache->space_info->lock);
3553 btrfs_put_block_group(cache);
3555 set_extent_dirty(fs_info->pinned_extents,
3556 bytenr, bytenr + num_bytes - 1, GFP_NOFS);
3560 static int update_reserved_extents(struct btrfs_block_group_cache *cache,
3561 u64 num_bytes, int reserve)
3563 spin_lock(&cache->space_info->lock);
3564 spin_lock(&cache->lock);
3566 cache->reserved += num_bytes;
3567 cache->space_info->bytes_reserved += num_bytes;
3569 cache->reserved -= num_bytes;
3570 cache->space_info->bytes_reserved -= num_bytes;
3572 spin_unlock(&cache->lock);
3573 spin_unlock(&cache->space_info->lock);
3577 int btrfs_prepare_extent_commit(struct btrfs_trans_handle *trans,
3578 struct btrfs_root *root)
3580 struct btrfs_fs_info *fs_info = root->fs_info;
3581 struct btrfs_caching_control *next;
3582 struct btrfs_caching_control *caching_ctl;
3583 struct btrfs_block_group_cache *cache;
3585 down_write(&fs_info->extent_commit_sem);
3587 list_for_each_entry_safe(caching_ctl, next,
3588 &fs_info->caching_block_groups, list) {
3589 cache = caching_ctl->block_group;
3590 if (block_group_cache_done(cache)) {
3591 cache->last_byte_to_unpin = (u64)-1;
3592 list_del_init(&caching_ctl->list);
3593 put_caching_control(caching_ctl);
3595 cache->last_byte_to_unpin = caching_ctl->progress;
3599 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
3600 fs_info->pinned_extents = &fs_info->freed_extents[1];
3602 fs_info->pinned_extents = &fs_info->freed_extents[0];
3604 up_write(&fs_info->extent_commit_sem);
3608 static int unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
3610 struct btrfs_fs_info *fs_info = root->fs_info;
3611 struct btrfs_block_group_cache *cache = NULL;
3614 while (start <= end) {
3616 start >= cache->key.objectid + cache->key.offset) {
3618 btrfs_put_block_group(cache);
3619 cache = btrfs_lookup_block_group(fs_info, start);
3623 len = cache->key.objectid + cache->key.offset - start;
3624 len = min(len, end + 1 - start);
3626 if (start < cache->last_byte_to_unpin) {
3627 len = min(len, cache->last_byte_to_unpin - start);
3628 btrfs_add_free_space(cache, start, len);
3631 spin_lock(&cache->space_info->lock);
3632 spin_lock(&cache->lock);
3633 cache->pinned -= len;
3634 cache->space_info->bytes_pinned -= len;
3635 spin_unlock(&cache->lock);
3636 spin_unlock(&cache->space_info->lock);
3642 btrfs_put_block_group(cache);
3646 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
3647 struct btrfs_root *root)
3649 struct btrfs_fs_info *fs_info = root->fs_info;
3650 struct extent_io_tree *unpin;
3655 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
3656 unpin = &fs_info->freed_extents[1];
3658 unpin = &fs_info->freed_extents[0];
3661 ret = find_first_extent_bit(unpin, 0, &start, &end,
3666 ret = btrfs_discard_extent(root, start, end + 1 - start);
3668 clear_extent_dirty(unpin, start, end, GFP_NOFS);
3669 unpin_extent_range(root, start, end);
3676 static int pin_down_bytes(struct btrfs_trans_handle *trans,
3677 struct btrfs_root *root,
3678 struct btrfs_path *path,
3679 u64 bytenr, u64 num_bytes,
3680 int is_data, int reserved,
3681 struct extent_buffer **must_clean)
3684 struct extent_buffer *buf;
3690 * discard is sloooow, and so triggering discards on
3691 * individual btree blocks isn't a good plan. Just
3692 * pin everything in discard mode.
3694 if (btrfs_test_opt(root, DISCARD))
3697 buf = btrfs_find_tree_block(root, bytenr, num_bytes);
3701 /* we can reuse a block if it hasn't been written
3702 * and it is from this transaction. We can't
3703 * reuse anything from the tree log root because
3704 * it has tiny sub-transactions.
3706 if (btrfs_buffer_uptodate(buf, 0) &&
3707 btrfs_try_tree_lock(buf)) {
3708 u64 header_owner = btrfs_header_owner(buf);
3709 u64 header_transid = btrfs_header_generation(buf);
3710 if (header_owner != BTRFS_TREE_LOG_OBJECTID &&
3711 header_transid == trans->transid &&
3712 !btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
3716 btrfs_tree_unlock(buf);
3718 free_extent_buffer(buf);
3721 btrfs_set_path_blocking(path);
3722 /* unlocks the pinned mutex */
3723 btrfs_pin_extent(root, bytenr, num_bytes, reserved);
3729 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
3730 struct btrfs_root *root,
3731 u64 bytenr, u64 num_bytes, u64 parent,
3732 u64 root_objectid, u64 owner_objectid,
3733 u64 owner_offset, int refs_to_drop,
3734 struct btrfs_delayed_extent_op *extent_op)
3736 struct btrfs_key key;
3737 struct btrfs_path *path;
3738 struct btrfs_fs_info *info = root->fs_info;
3739 struct btrfs_root *extent_root = info->extent_root;
3740 struct extent_buffer *leaf;
3741 struct btrfs_extent_item *ei;
3742 struct btrfs_extent_inline_ref *iref;
3745 int extent_slot = 0;
3746 int found_extent = 0;
3751 path = btrfs_alloc_path();
3756 path->leave_spinning = 1;
3758 is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID;
3759 BUG_ON(!is_data && refs_to_drop != 1);
3761 ret = lookup_extent_backref(trans, extent_root, path, &iref,
3762 bytenr, num_bytes, parent,
3763 root_objectid, owner_objectid,
3766 extent_slot = path->slots[0];
3767 while (extent_slot >= 0) {
3768 btrfs_item_key_to_cpu(path->nodes[0], &key,
3770 if (key.objectid != bytenr)
3772 if (key.type == BTRFS_EXTENT_ITEM_KEY &&
3773 key.offset == num_bytes) {
3777 if (path->slots[0] - extent_slot > 5)
3781 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
3782 item_size = btrfs_item_size_nr(path->nodes[0], extent_slot);
3783 if (found_extent && item_size < sizeof(*ei))
3786 if (!found_extent) {
3788 ret = remove_extent_backref(trans, extent_root, path,
3792 btrfs_release_path(extent_root, path);
3793 path->leave_spinning = 1;
3795 key.objectid = bytenr;
3796 key.type = BTRFS_EXTENT_ITEM_KEY;
3797 key.offset = num_bytes;
3799 ret = btrfs_search_slot(trans, extent_root,
3802 printk(KERN_ERR "umm, got %d back from search"
3803 ", was looking for %llu\n", ret,
3804 (unsigned long long)bytenr);
3805 btrfs_print_leaf(extent_root, path->nodes[0]);
3808 extent_slot = path->slots[0];
3811 btrfs_print_leaf(extent_root, path->nodes[0]);
3813 printk(KERN_ERR "btrfs unable to find ref byte nr %llu "
3814 "parent %llu root %llu owner %llu offset %llu\n",
3815 (unsigned long long)bytenr,
3816 (unsigned long long)parent,
3817 (unsigned long long)root_objectid,
3818 (unsigned long long)owner_objectid,
3819 (unsigned long long)owner_offset);
3822 leaf = path->nodes[0];
3823 item_size = btrfs_item_size_nr(leaf, extent_slot);
3824 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
3825 if (item_size < sizeof(*ei)) {
3826 BUG_ON(found_extent || extent_slot != path->slots[0]);
3827 ret = convert_extent_item_v0(trans, extent_root, path,
3831 btrfs_release_path(extent_root, path);
3832 path->leave_spinning = 1;
3834 key.objectid = bytenr;
3835 key.type = BTRFS_EXTENT_ITEM_KEY;
3836 key.offset = num_bytes;
3838 ret = btrfs_search_slot(trans, extent_root, &key, path,
3841 printk(KERN_ERR "umm, got %d back from search"
3842 ", was looking for %llu\n", ret,
3843 (unsigned long long)bytenr);
3844 btrfs_print_leaf(extent_root, path->nodes[0]);
3847 extent_slot = path->slots[0];
3848 leaf = path->nodes[0];
3849 item_size = btrfs_item_size_nr(leaf, extent_slot);
3852 BUG_ON(item_size < sizeof(*ei));
3853 ei = btrfs_item_ptr(leaf, extent_slot,
3854 struct btrfs_extent_item);
3855 if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID) {
3856 struct btrfs_tree_block_info *bi;
3857 BUG_ON(item_size < sizeof(*ei) + sizeof(*bi));
3858 bi = (struct btrfs_tree_block_info *)(ei + 1);
3859 WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi));
3862 refs = btrfs_extent_refs(leaf, ei);
3863 BUG_ON(refs < refs_to_drop);
3864 refs -= refs_to_drop;
3868 __run_delayed_extent_op(extent_op, leaf, ei);
3870 * In the case of inline back ref, reference count will
3871 * be updated by remove_extent_backref
3874 BUG_ON(!found_extent);
3876 btrfs_set_extent_refs(leaf, ei, refs);
3877 btrfs_mark_buffer_dirty(leaf);
3880 ret = remove_extent_backref(trans, extent_root, path,
3887 struct extent_buffer *must_clean = NULL;
3890 BUG_ON(is_data && refs_to_drop !=
3891 extent_data_ref_count(root, path, iref));
3893 BUG_ON(path->slots[0] != extent_slot);
3895 BUG_ON(path->slots[0] != extent_slot + 1);
3896 path->slots[0] = extent_slot;
3901 ret = pin_down_bytes(trans, root, path, bytenr,
3902 num_bytes, is_data, 0, &must_clean);
3907 * it is going to be very rare for someone to be waiting
3908 * on the block we're freeing. del_items might need to
3909 * schedule, so rather than get fancy, just force it
3913 btrfs_set_lock_blocking(must_clean);
3915 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
3918 btrfs_release_path(extent_root, path);
3921 clean_tree_block(NULL, root, must_clean);
3922 btrfs_tree_unlock(must_clean);
3923 free_extent_buffer(must_clean);
3927 ret = btrfs_del_csums(trans, root, bytenr, num_bytes);
3930 invalidate_mapping_pages(info->btree_inode->i_mapping,
3931 bytenr >> PAGE_CACHE_SHIFT,
3932 (bytenr + num_bytes - 1) >> PAGE_CACHE_SHIFT);
3935 ret = update_block_group(trans, root, bytenr, num_bytes, 0,
3939 btrfs_free_path(path);
3944 * when we free an extent, it is possible (and likely) that we free the last
3945 * delayed ref for that extent as well. This searches the delayed ref tree for
3946 * a given extent, and if there are no other delayed refs to be processed, it
3947 * removes it from the tree.
3949 static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans,
3950 struct btrfs_root *root, u64 bytenr)
3952 struct btrfs_delayed_ref_head *head;
3953 struct btrfs_delayed_ref_root *delayed_refs;
3954 struct btrfs_delayed_ref_node *ref;
3955 struct rb_node *node;
3958 delayed_refs = &trans->transaction->delayed_refs;
3959 spin_lock(&delayed_refs->lock);
3960 head = btrfs_find_delayed_ref_head(trans, bytenr);
3964 node = rb_prev(&head->node.rb_node);
3968 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
3970 /* there are still entries for this ref, we can't drop it */
3971 if (ref->bytenr == bytenr)
3974 if (head->extent_op) {
3975 if (!head->must_insert_reserved)
3977 kfree(head->extent_op);
3978 head->extent_op = NULL;
3982 * waiting for the lock here would deadlock. If someone else has it
3983 * locked they are already in the process of dropping it anyway
3985 if (!mutex_trylock(&head->mutex))
3989 * at this point we have a head with no other entries. Go
3990 * ahead and process it.
3992 head->node.in_tree = 0;
3993 rb_erase(&head->node.rb_node, &delayed_refs->root);
3995 delayed_refs->num_entries--;
3998 * we don't take a ref on the node because we're removing it from the
3999 * tree, so we just steal the ref the tree was holding.
4001 delayed_refs->num_heads--;
4002 if (list_empty(&head->cluster))
4003 delayed_refs->num_heads_ready--;
4005 list_del_init(&head->cluster);
4006 spin_unlock(&delayed_refs->lock);
4008 ret = run_one_delayed_ref(trans, root->fs_info->tree_root,
4009 &head->node, head->extent_op,
4010 head->must_insert_reserved);
4012 btrfs_put_delayed_ref(&head->node);
4015 spin_unlock(&delayed_refs->lock);
4019 int btrfs_free_extent(struct btrfs_trans_handle *trans,
4020 struct btrfs_root *root,
4021 u64 bytenr, u64 num_bytes, u64 parent,
4022 u64 root_objectid, u64 owner, u64 offset)
4027 * tree log blocks never actually go into the extent allocation
4028 * tree, just update pinning info and exit early.
4030 if (root_objectid == BTRFS_TREE_LOG_OBJECTID) {
4031 WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID);
4032 /* unlocks the pinned mutex */
4033 btrfs_pin_extent(root, bytenr, num_bytes, 1);
4035 } else if (owner < BTRFS_FIRST_FREE_OBJECTID) {
4036 ret = btrfs_add_delayed_tree_ref(trans, bytenr, num_bytes,
4037 parent, root_objectid, (int)owner,
4038 BTRFS_DROP_DELAYED_REF, NULL);
4040 ret = check_ref_cleanup(trans, root, bytenr);
4043 ret = btrfs_add_delayed_data_ref(trans, bytenr, num_bytes,
4044 parent, root_objectid, owner,
4045 offset, BTRFS_DROP_DELAYED_REF, NULL);
4051 static u64 stripe_align(struct btrfs_root *root, u64 val)
4053 u64 mask = ((u64)root->stripesize - 1);
4054 u64 ret = (val + mask) & ~mask;
4059 * when we wait for progress in the block group caching, its because
4060 * our allocation attempt failed at least once. So, we must sleep
4061 * and let some progress happen before we try again.
4063 * This function will sleep at least once waiting for new free space to
4064 * show up, and then it will check the block group free space numbers
4065 * for our min num_bytes. Another option is to have it go ahead
4066 * and look in the rbtree for a free extent of a given size, but this
4070 wait_block_group_cache_progress(struct btrfs_block_group_cache *cache,
4073 struct btrfs_caching_control *caching_ctl;
4076 caching_ctl = get_caching_control(cache);
4080 wait_event(caching_ctl->wait, block_group_cache_done(cache) ||
4081 (cache->free_space >= num_bytes));
4083 put_caching_control(caching_ctl);
4088 wait_block_group_cache_done(struct btrfs_block_group_cache *cache)
4090 struct btrfs_caching_control *caching_ctl;
4093 caching_ctl = get_caching_control(cache);
4097 wait_event(caching_ctl->wait, block_group_cache_done(cache));
4099 put_caching_control(caching_ctl);
4103 enum btrfs_loop_type {
4104 LOOP_CACHED_ONLY = 0,
4105 LOOP_CACHING_NOWAIT = 1,
4106 LOOP_CACHING_WAIT = 2,
4107 LOOP_ALLOC_CHUNK = 3,
4108 LOOP_NO_EMPTY_SIZE = 4,
4112 * walks the btree of allocated extents and find a hole of a given size.
4113 * The key ins is changed to record the hole:
4114 * ins->objectid == block start
4115 * ins->flags = BTRFS_EXTENT_ITEM_KEY
4116 * ins->offset == number of blocks
4117 * Any available blocks before search_start are skipped.
4119 static noinline int find_free_extent(struct btrfs_trans_handle *trans,
4120 struct btrfs_root *orig_root,
4121 u64 num_bytes, u64 empty_size,
4122 u64 search_start, u64 search_end,
4123 u64 hint_byte, struct btrfs_key *ins,
4124 u64 exclude_start, u64 exclude_nr,
4128 struct btrfs_root *root = orig_root->fs_info->extent_root;
4129 struct btrfs_free_cluster *last_ptr = NULL;
4130 struct btrfs_block_group_cache *block_group = NULL;
4131 int empty_cluster = 2 * 1024 * 1024;
4132 int allowed_chunk_alloc = 0;
4133 struct btrfs_space_info *space_info;
4134 int last_ptr_loop = 0;
4136 bool found_uncached_bg = false;
4137 bool failed_cluster_refill = false;
4138 bool failed_alloc = false;
4140 WARN_ON(num_bytes < root->sectorsize);
4141 btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY);
4145 space_info = __find_space_info(root->fs_info, data);
4147 if (orig_root->ref_cows || empty_size)
4148 allowed_chunk_alloc = 1;
4150 if (data & BTRFS_BLOCK_GROUP_METADATA) {
4151 last_ptr = &root->fs_info->meta_alloc_cluster;
4152 if (!btrfs_test_opt(root, SSD))
4153 empty_cluster = 64 * 1024;
4156 if ((data & BTRFS_BLOCK_GROUP_DATA) && btrfs_test_opt(root, SSD)) {
4157 last_ptr = &root->fs_info->data_alloc_cluster;
4161 spin_lock(&last_ptr->lock);
4162 if (last_ptr->block_group)
4163 hint_byte = last_ptr->window_start;
4164 spin_unlock(&last_ptr->lock);
4167 search_start = max(search_start, first_logical_byte(root, 0));
4168 search_start = max(search_start, hint_byte);
4173 if (search_start == hint_byte) {
4174 block_group = btrfs_lookup_block_group(root->fs_info,
4177 * we don't want to use the block group if it doesn't match our
4178 * allocation bits, or if its not cached.
4180 if (block_group && block_group_bits(block_group, data) &&
4181 block_group_cache_done(block_group)) {
4182 down_read(&space_info->groups_sem);
4183 if (list_empty(&block_group->list) ||
4186 * someone is removing this block group,
4187 * we can't jump into the have_block_group
4188 * target because our list pointers are not
4191 btrfs_put_block_group(block_group);
4192 up_read(&space_info->groups_sem);
4194 goto have_block_group;
4195 } else if (block_group) {
4196 btrfs_put_block_group(block_group);
4201 down_read(&space_info->groups_sem);
4202 list_for_each_entry(block_group, &space_info->block_groups, list) {
4206 atomic_inc(&block_group->count);
4207 search_start = block_group->key.objectid;
4210 if (unlikely(block_group->cached == BTRFS_CACHE_NO)) {
4212 * we want to start caching kthreads, but not too many
4213 * right off the bat so we don't overwhelm the system,
4214 * so only start them if there are less than 2 and we're
4215 * in the initial allocation phase.
4217 if (loop > LOOP_CACHING_NOWAIT ||
4218 atomic_read(&space_info->caching_threads) < 2) {
4219 ret = cache_block_group(block_group);
4224 cached = block_group_cache_done(block_group);
4225 if (unlikely(!cached)) {
4226 found_uncached_bg = true;
4228 /* if we only want cached bgs, loop */
4229 if (loop == LOOP_CACHED_ONLY)
4233 if (unlikely(block_group->ro))
4237 * Ok we want to try and use the cluster allocator, so lets look
4238 * there, unless we are on LOOP_NO_EMPTY_SIZE, since we will
4239 * have tried the cluster allocator plenty of times at this
4240 * point and not have found anything, so we are likely way too
4241 * fragmented for the clustering stuff to find anything, so lets
4242 * just skip it and let the allocator find whatever block it can
4245 if (last_ptr && loop < LOOP_NO_EMPTY_SIZE) {
4247 * the refill lock keeps out other
4248 * people trying to start a new cluster
4250 spin_lock(&last_ptr->refill_lock);
4251 if (last_ptr->block_group &&
4252 (last_ptr->block_group->ro ||
4253 !block_group_bits(last_ptr->block_group, data))) {
4255 goto refill_cluster;
4258 offset = btrfs_alloc_from_cluster(block_group, last_ptr,
4259 num_bytes, search_start);
4261 /* we have a block, we're done */
4262 spin_unlock(&last_ptr->refill_lock);
4266 spin_lock(&last_ptr->lock);
4268 * whoops, this cluster doesn't actually point to
4269 * this block group. Get a ref on the block
4270 * group is does point to and try again
4272 if (!last_ptr_loop && last_ptr->block_group &&
4273 last_ptr->block_group != block_group) {
4275 btrfs_put_block_group(block_group);
4276 block_group = last_ptr->block_group;
4277 atomic_inc(&block_group->count);
4278 spin_unlock(&last_ptr->lock);
4279 spin_unlock(&last_ptr->refill_lock);
4282 search_start = block_group->key.objectid;
4284 * we know this block group is properly
4285 * in the list because
4286 * btrfs_remove_block_group, drops the
4287 * cluster before it removes the block
4288 * group from the list
4290 goto have_block_group;
4292 spin_unlock(&last_ptr->lock);
4295 * this cluster didn't work out, free it and
4298 btrfs_return_cluster_to_free_space(NULL, last_ptr);
4302 /* allocate a cluster in this block group */
4303 ret = btrfs_find_space_cluster(trans, root,
4304 block_group, last_ptr,
4306 empty_cluster + empty_size);
4309 * now pull our allocation out of this
4312 offset = btrfs_alloc_from_cluster(block_group,
4313 last_ptr, num_bytes,
4316 /* we found one, proceed */
4317 spin_unlock(&last_ptr->refill_lock);
4320 } else if (!cached && loop > LOOP_CACHING_NOWAIT
4321 && !failed_cluster_refill) {
4322 spin_unlock(&last_ptr->refill_lock);
4324 failed_cluster_refill = true;
4325 wait_block_group_cache_progress(block_group,
4326 num_bytes + empty_cluster + empty_size);
4327 goto have_block_group;
4331 * at this point we either didn't find a cluster
4332 * or we weren't able to allocate a block from our
4333 * cluster. Free the cluster we've been trying
4334 * to use, and go to the next block group
4336 btrfs_return_cluster_to_free_space(NULL, last_ptr);
4337 spin_unlock(&last_ptr->refill_lock);
4341 offset = btrfs_find_space_for_alloc(block_group, search_start,
4342 num_bytes, empty_size);
4344 * If we didn't find a chunk, and we haven't failed on this
4345 * block group before, and this block group is in the middle of
4346 * caching and we are ok with waiting, then go ahead and wait
4347 * for progress to be made, and set failed_alloc to true.
4349 * If failed_alloc is true then we've already waited on this
4350 * block group once and should move on to the next block group.
4352 if (!offset && !failed_alloc && !cached &&
4353 loop > LOOP_CACHING_NOWAIT) {
4354 wait_block_group_cache_progress(block_group,
4355 num_bytes + empty_size);
4356 failed_alloc = true;
4357 goto have_block_group;
4358 } else if (!offset) {
4362 search_start = stripe_align(root, offset);
4363 /* move on to the next group */
4364 if (search_start + num_bytes >= search_end) {
4365 btrfs_add_free_space(block_group, offset, num_bytes);
4369 /* move on to the next group */
4370 if (search_start + num_bytes >
4371 block_group->key.objectid + block_group->key.offset) {
4372 btrfs_add_free_space(block_group, offset, num_bytes);
4376 if (exclude_nr > 0 &&
4377 (search_start + num_bytes > exclude_start &&
4378 search_start < exclude_start + exclude_nr)) {
4379 search_start = exclude_start + exclude_nr;
4381 btrfs_add_free_space(block_group, offset, num_bytes);
4383 * if search_start is still in this block group
4384 * then we just re-search this block group
4386 if (search_start >= block_group->key.objectid &&
4387 search_start < (block_group->key.objectid +
4388 block_group->key.offset))
4389 goto have_block_group;
4393 ins->objectid = search_start;
4394 ins->offset = num_bytes;
4396 if (offset < search_start)
4397 btrfs_add_free_space(block_group, offset,
4398 search_start - offset);
4399 BUG_ON(offset > search_start);
4401 update_reserved_extents(block_group, num_bytes, 1);
4403 /* we are all good, lets return */
4406 failed_cluster_refill = false;
4407 failed_alloc = false;
4408 btrfs_put_block_group(block_group);
4410 up_read(&space_info->groups_sem);
4412 /* LOOP_CACHED_ONLY, only search fully cached block groups
4413 * LOOP_CACHING_NOWAIT, search partially cached block groups, but
4414 * dont wait foR them to finish caching
4415 * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
4416 * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
4417 * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
4420 if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE &&
4421 (found_uncached_bg || empty_size || empty_cluster ||
4422 allowed_chunk_alloc)) {
4423 if (found_uncached_bg) {
4424 found_uncached_bg = false;
4425 if (loop < LOOP_CACHING_WAIT) {
4431 if (loop == LOOP_ALLOC_CHUNK) {
4436 if (allowed_chunk_alloc) {
4437 ret = do_chunk_alloc(trans, root, num_bytes +
4438 2 * 1024 * 1024, data, 1);
4439 allowed_chunk_alloc = 0;
4441 space_info->force_alloc = 1;
4444 if (loop < LOOP_NO_EMPTY_SIZE) {
4449 } else if (!ins->objectid) {
4453 /* we found what we needed */
4454 if (ins->objectid) {
4455 if (!(data & BTRFS_BLOCK_GROUP_DATA))
4456 trans->block_group = block_group->key.objectid;
4458 btrfs_put_block_group(block_group);
4465 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
4466 int dump_block_groups)
4468 struct btrfs_block_group_cache *cache;
4470 spin_lock(&info->lock);
4471 printk(KERN_INFO "space_info has %llu free, is %sfull\n",
4472 (unsigned long long)(info->total_bytes - info->bytes_used -
4473 info->bytes_pinned - info->bytes_reserved -
4475 (info->full) ? "" : "not ");
4476 printk(KERN_INFO "space_info total=%llu, pinned=%llu, delalloc=%llu,"
4477 " may_use=%llu, used=%llu, root=%llu, super=%llu, reserved=%llu"
4479 (unsigned long long)info->total_bytes,
4480 (unsigned long long)info->bytes_pinned,
4481 (unsigned long long)info->bytes_delalloc,
4482 (unsigned long long)info->bytes_may_use,
4483 (unsigned long long)info->bytes_used,
4484 (unsigned long long)info->bytes_root,
4485 (unsigned long long)info->bytes_super,
4486 (unsigned long long)info->bytes_reserved);
4487 spin_unlock(&info->lock);
4489 if (!dump_block_groups)
4492 down_read(&info->groups_sem);
4493 list_for_each_entry(cache, &info->block_groups, list) {
4494 spin_lock(&cache->lock);
4495 printk(KERN_INFO "block group %llu has %llu bytes, %llu used "
4496 "%llu pinned %llu reserved\n",
4497 (unsigned long long)cache->key.objectid,
4498 (unsigned long long)cache->key.offset,
4499 (unsigned long long)btrfs_block_group_used(&cache->item),
4500 (unsigned long long)cache->pinned,
4501 (unsigned long long)cache->reserved);
4502 btrfs_dump_free_space(cache, bytes);
4503 spin_unlock(&cache->lock);
4505 up_read(&info->groups_sem);
4508 int btrfs_reserve_extent(struct btrfs_trans_handle *trans,
4509 struct btrfs_root *root,
4510 u64 num_bytes, u64 min_alloc_size,
4511 u64 empty_size, u64 hint_byte,
4512 u64 search_end, struct btrfs_key *ins,
4516 u64 search_start = 0;
4517 struct btrfs_fs_info *info = root->fs_info;
4519 data = btrfs_get_alloc_profile(root, data);
4522 * the only place that sets empty_size is btrfs_realloc_node, which
4523 * is not called recursively on allocations
4525 if (empty_size || root->ref_cows) {
4526 if (!(data & BTRFS_BLOCK_GROUP_METADATA)) {
4527 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
4529 BTRFS_BLOCK_GROUP_METADATA |
4530 (info->metadata_alloc_profile &
4531 info->avail_metadata_alloc_bits), 0);
4533 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
4534 num_bytes + 2 * 1024 * 1024, data, 0);
4537 WARN_ON(num_bytes < root->sectorsize);
4538 ret = find_free_extent(trans, root, num_bytes, empty_size,
4539 search_start, search_end, hint_byte, ins,
4540 trans->alloc_exclude_start,
4541 trans->alloc_exclude_nr, data);
4543 if (ret == -ENOSPC && num_bytes > min_alloc_size) {
4544 num_bytes = num_bytes >> 1;
4545 num_bytes = num_bytes & ~(root->sectorsize - 1);
4546 num_bytes = max(num_bytes, min_alloc_size);
4547 do_chunk_alloc(trans, root->fs_info->extent_root,
4548 num_bytes, data, 1);
4551 if (ret == -ENOSPC) {
4552 struct btrfs_space_info *sinfo;
4554 sinfo = __find_space_info(root->fs_info, data);
4555 printk(KERN_ERR "btrfs allocation failed flags %llu, "
4556 "wanted %llu\n", (unsigned long long)data,
4557 (unsigned long long)num_bytes);
4558 dump_space_info(sinfo, num_bytes, 1);
4564 int btrfs_free_reserved_extent(struct btrfs_root *root, u64 start, u64 len)
4566 struct btrfs_block_group_cache *cache;
4569 cache = btrfs_lookup_block_group(root->fs_info, start);
4571 printk(KERN_ERR "Unable to find block group for %llu\n",
4572 (unsigned long long)start);
4576 ret = btrfs_discard_extent(root, start, len);
4578 btrfs_add_free_space(cache, start, len);
4579 update_reserved_extents(cache, len, 0);
4580 btrfs_put_block_group(cache);
4585 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
4586 struct btrfs_root *root,
4587 u64 parent, u64 root_objectid,
4588 u64 flags, u64 owner, u64 offset,
4589 struct btrfs_key *ins, int ref_mod)
4592 struct btrfs_fs_info *fs_info = root->fs_info;
4593 struct btrfs_extent_item *extent_item;
4594 struct btrfs_extent_inline_ref *iref;
4595 struct btrfs_path *path;
4596 struct extent_buffer *leaf;
4601 type = BTRFS_SHARED_DATA_REF_KEY;
4603 type = BTRFS_EXTENT_DATA_REF_KEY;
4605 size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type);
4607 path = btrfs_alloc_path();
4610 path->leave_spinning = 1;
4611 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
4615 leaf = path->nodes[0];
4616 extent_item = btrfs_item_ptr(leaf, path->slots[0],
4617 struct btrfs_extent_item);
4618 btrfs_set_extent_refs(leaf, extent_item, ref_mod);
4619 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
4620 btrfs_set_extent_flags(leaf, extent_item,
4621 flags | BTRFS_EXTENT_FLAG_DATA);
4623 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
4624 btrfs_set_extent_inline_ref_type(leaf, iref, type);
4626 struct btrfs_shared_data_ref *ref;
4627 ref = (struct btrfs_shared_data_ref *)(iref + 1);
4628 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
4629 btrfs_set_shared_data_ref_count(leaf, ref, ref_mod);
4631 struct btrfs_extent_data_ref *ref;
4632 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
4633 btrfs_set_extent_data_ref_root(leaf, ref, root_objectid);
4634 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
4635 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
4636 btrfs_set_extent_data_ref_count(leaf, ref, ref_mod);
4639 btrfs_mark_buffer_dirty(path->nodes[0]);
4640 btrfs_free_path(path);
4642 ret = update_block_group(trans, root, ins->objectid, ins->offset,
4645 printk(KERN_ERR "btrfs update block group failed for %llu "
4646 "%llu\n", (unsigned long long)ins->objectid,
4647 (unsigned long long)ins->offset);
4653 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
4654 struct btrfs_root *root,
4655 u64 parent, u64 root_objectid,
4656 u64 flags, struct btrfs_disk_key *key,
4657 int level, struct btrfs_key *ins)
4660 struct btrfs_fs_info *fs_info = root->fs_info;
4661 struct btrfs_extent_item *extent_item;
4662 struct btrfs_tree_block_info *block_info;
4663 struct btrfs_extent_inline_ref *iref;
4664 struct btrfs_path *path;
4665 struct extent_buffer *leaf;
4666 u32 size = sizeof(*extent_item) + sizeof(*block_info) + sizeof(*iref);
4668 path = btrfs_alloc_path();
4671 path->leave_spinning = 1;
4672 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
4676 leaf = path->nodes[0];
4677 extent_item = btrfs_item_ptr(leaf, path->slots[0],
4678 struct btrfs_extent_item);
4679 btrfs_set_extent_refs(leaf, extent_item, 1);
4680 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
4681 btrfs_set_extent_flags(leaf, extent_item,
4682 flags | BTRFS_EXTENT_FLAG_TREE_BLOCK);
4683 block_info = (struct btrfs_tree_block_info *)(extent_item + 1);
4685 btrfs_set_tree_block_key(leaf, block_info, key);
4686 btrfs_set_tree_block_level(leaf, block_info, level);
4688 iref = (struct btrfs_extent_inline_ref *)(block_info + 1);
4690 BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
4691 btrfs_set_extent_inline_ref_type(leaf, iref,
4692 BTRFS_SHARED_BLOCK_REF_KEY);
4693 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
4695 btrfs_set_extent_inline_ref_type(leaf, iref,
4696 BTRFS_TREE_BLOCK_REF_KEY);
4697 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
4700 btrfs_mark_buffer_dirty(leaf);
4701 btrfs_free_path(path);
4703 ret = update_block_group(trans, root, ins->objectid, ins->offset,
4706 printk(KERN_ERR "btrfs update block group failed for %llu "
4707 "%llu\n", (unsigned long long)ins->objectid,
4708 (unsigned long long)ins->offset);
4714 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
4715 struct btrfs_root *root,
4716 u64 root_objectid, u64 owner,
4717 u64 offset, struct btrfs_key *ins)
4721 BUG_ON(root_objectid == BTRFS_TREE_LOG_OBJECTID);
4723 ret = btrfs_add_delayed_data_ref(trans, ins->objectid, ins->offset,
4724 0, root_objectid, owner, offset,
4725 BTRFS_ADD_DELAYED_EXTENT, NULL);
4730 * this is used by the tree logging recovery code. It records that
4731 * an extent has been allocated and makes sure to clear the free
4732 * space cache bits as well
4734 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans,
4735 struct btrfs_root *root,
4736 u64 root_objectid, u64 owner, u64 offset,
4737 struct btrfs_key *ins)
4740 struct btrfs_block_group_cache *block_group;
4741 struct btrfs_caching_control *caching_ctl;
4742 u64 start = ins->objectid;
4743 u64 num_bytes = ins->offset;
4745 block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid);
4746 cache_block_group(block_group);
4747 caching_ctl = get_caching_control(block_group);
4750 BUG_ON(!block_group_cache_done(block_group));
4751 ret = btrfs_remove_free_space(block_group, start, num_bytes);
4754 mutex_lock(&caching_ctl->mutex);
4756 if (start >= caching_ctl->progress) {
4757 ret = add_excluded_extent(root, start, num_bytes);
4759 } else if (start + num_bytes <= caching_ctl->progress) {
4760 ret = btrfs_remove_free_space(block_group,
4764 num_bytes = caching_ctl->progress - start;
4765 ret = btrfs_remove_free_space(block_group,
4769 start = caching_ctl->progress;
4770 num_bytes = ins->objectid + ins->offset -
4771 caching_ctl->progress;
4772 ret = add_excluded_extent(root, start, num_bytes);
4776 mutex_unlock(&caching_ctl->mutex);
4777 put_caching_control(caching_ctl);
4780 update_reserved_extents(block_group, ins->offset, 1);
4781 btrfs_put_block_group(block_group);
4782 ret = alloc_reserved_file_extent(trans, root, 0, root_objectid,
4783 0, owner, offset, ins, 1);
4788 * finds a free extent and does all the dirty work required for allocation
4789 * returns the key for the extent through ins, and a tree buffer for
4790 * the first block of the extent through buf.
4792 * returns 0 if everything worked, non-zero otherwise.
4794 static int alloc_tree_block(struct btrfs_trans_handle *trans,
4795 struct btrfs_root *root,
4796 u64 num_bytes, u64 parent, u64 root_objectid,
4797 struct btrfs_disk_key *key, int level,
4798 u64 empty_size, u64 hint_byte, u64 search_end,
4799 struct btrfs_key *ins)
4804 ret = btrfs_reserve_extent(trans, root, num_bytes, num_bytes,
4805 empty_size, hint_byte, search_end,
4810 if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) {
4812 parent = ins->objectid;
4813 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
4817 if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
4818 struct btrfs_delayed_extent_op *extent_op;
4819 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
4822 memcpy(&extent_op->key, key, sizeof(extent_op->key));
4824 memset(&extent_op->key, 0, sizeof(extent_op->key));
4825 extent_op->flags_to_set = flags;
4826 extent_op->update_key = 1;
4827 extent_op->update_flags = 1;
4828 extent_op->is_data = 0;
4830 ret = btrfs_add_delayed_tree_ref(trans, ins->objectid,
4831 ins->offset, parent, root_objectid,
4832 level, BTRFS_ADD_DELAYED_EXTENT,
4839 struct extent_buffer *btrfs_init_new_buffer(struct btrfs_trans_handle *trans,
4840 struct btrfs_root *root,
4841 u64 bytenr, u32 blocksize,
4844 struct extent_buffer *buf;
4846 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
4848 return ERR_PTR(-ENOMEM);
4849 btrfs_set_header_generation(buf, trans->transid);
4850 btrfs_set_buffer_lockdep_class(buf, level);
4851 btrfs_tree_lock(buf);
4852 clean_tree_block(trans, root, buf);
4854 btrfs_set_lock_blocking(buf);
4855 btrfs_set_buffer_uptodate(buf);
4857 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
4858 set_extent_dirty(&root->dirty_log_pages, buf->start,
4859 buf->start + buf->len - 1, GFP_NOFS);
4861 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
4862 buf->start + buf->len - 1, GFP_NOFS);
4864 trans->blocks_used++;
4865 /* this returns a buffer locked for blocking */
4870 * helper function to allocate a block for a given tree
4871 * returns the tree buffer or NULL.
4873 struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
4874 struct btrfs_root *root, u32 blocksize,
4875 u64 parent, u64 root_objectid,
4876 struct btrfs_disk_key *key, int level,
4877 u64 hint, u64 empty_size)
4879 struct btrfs_key ins;
4881 struct extent_buffer *buf;
4883 ret = alloc_tree_block(trans, root, blocksize, parent, root_objectid,
4884 key, level, empty_size, hint, (u64)-1, &ins);
4887 return ERR_PTR(ret);
4890 buf = btrfs_init_new_buffer(trans, root, ins.objectid,
4895 struct walk_control {
4896 u64 refs[BTRFS_MAX_LEVEL];
4897 u64 flags[BTRFS_MAX_LEVEL];
4898 struct btrfs_key update_progress;
4908 #define DROP_REFERENCE 1
4909 #define UPDATE_BACKREF 2
4911 static noinline void reada_walk_down(struct btrfs_trans_handle *trans,
4912 struct btrfs_root *root,
4913 struct walk_control *wc,
4914 struct btrfs_path *path)
4923 struct btrfs_key key;
4924 struct extent_buffer *eb;
4929 if (path->slots[wc->level] < wc->reada_slot) {
4930 wc->reada_count = wc->reada_count * 2 / 3;
4931 wc->reada_count = max(wc->reada_count, 2);
4933 wc->reada_count = wc->reada_count * 3 / 2;
4934 wc->reada_count = min_t(int, wc->reada_count,
4935 BTRFS_NODEPTRS_PER_BLOCK(root));
4938 eb = path->nodes[wc->level];
4939 nritems = btrfs_header_nritems(eb);
4940 blocksize = btrfs_level_size(root, wc->level - 1);
4942 for (slot = path->slots[wc->level]; slot < nritems; slot++) {
4943 if (nread >= wc->reada_count)
4947 bytenr = btrfs_node_blockptr(eb, slot);
4948 generation = btrfs_node_ptr_generation(eb, slot);
4950 if (slot == path->slots[wc->level])
4953 if (wc->stage == UPDATE_BACKREF &&
4954 generation <= root->root_key.offset)
4957 /* We don't lock the tree block, it's OK to be racy here */
4958 ret = btrfs_lookup_extent_info(trans, root, bytenr, blocksize,
4963 if (wc->stage == DROP_REFERENCE) {
4967 if (wc->level == 1 &&
4968 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
4970 if (!wc->update_ref ||
4971 generation <= root->root_key.offset)
4973 btrfs_node_key_to_cpu(eb, &key, slot);
4974 ret = btrfs_comp_cpu_keys(&key,
4975 &wc->update_progress);
4979 if (wc->level == 1 &&
4980 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
4984 ret = readahead_tree_block(root, bytenr, blocksize,
4988 last = bytenr + blocksize;
4991 wc->reada_slot = slot;
4995 * hepler to process tree block while walking down the tree.
4997 * when wc->stage == UPDATE_BACKREF, this function updates
4998 * back refs for pointers in the block.
5000 * NOTE: return value 1 means we should stop walking down.
5002 static noinline int walk_down_proc(struct btrfs_trans_handle *trans,
5003 struct btrfs_root *root,
5004 struct btrfs_path *path,
5005 struct walk_control *wc, int lookup_info)
5007 int level = wc->level;
5008 struct extent_buffer *eb = path->nodes[level];
5009 u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF;
5012 if (wc->stage == UPDATE_BACKREF &&
5013 btrfs_header_owner(eb) != root->root_key.objectid)
5017 * when reference count of tree block is 1, it won't increase
5018 * again. once full backref flag is set, we never clear it.
5021 ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) ||
5022 (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) {
5023 BUG_ON(!path->locks[level]);
5024 ret = btrfs_lookup_extent_info(trans, root,
5029 BUG_ON(wc->refs[level] == 0);
5032 if (wc->stage == DROP_REFERENCE) {
5033 if (wc->refs[level] > 1)
5036 if (path->locks[level] && !wc->keep_locks) {
5037 btrfs_tree_unlock(eb);
5038 path->locks[level] = 0;
5043 /* wc->stage == UPDATE_BACKREF */
5044 if (!(wc->flags[level] & flag)) {
5045 BUG_ON(!path->locks[level]);
5046 ret = btrfs_inc_ref(trans, root, eb, 1);
5048 ret = btrfs_dec_ref(trans, root, eb, 0);
5050 ret = btrfs_set_disk_extent_flags(trans, root, eb->start,
5053 wc->flags[level] |= flag;
5057 * the block is shared by multiple trees, so it's not good to
5058 * keep the tree lock
5060 if (path->locks[level] && level > 0) {
5061 btrfs_tree_unlock(eb);
5062 path->locks[level] = 0;
5068 * hepler to process tree block pointer.
5070 * when wc->stage == DROP_REFERENCE, this function checks
5071 * reference count of the block pointed to. if the block
5072 * is shared and we need update back refs for the subtree
5073 * rooted at the block, this function changes wc->stage to
5074 * UPDATE_BACKREF. if the block is shared and there is no
5075 * need to update back, this function drops the reference
5078 * NOTE: return value 1 means we should stop walking down.
5080 static noinline int do_walk_down(struct btrfs_trans_handle *trans,
5081 struct btrfs_root *root,
5082 struct btrfs_path *path,
5083 struct walk_control *wc, int *lookup_info)
5089 struct btrfs_key key;
5090 struct extent_buffer *next;
5091 int level = wc->level;
5095 generation = btrfs_node_ptr_generation(path->nodes[level],
5096 path->slots[level]);
5098 * if the lower level block was created before the snapshot
5099 * was created, we know there is no need to update back refs
5102 if (wc->stage == UPDATE_BACKREF &&
5103 generation <= root->root_key.offset) {
5108 bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]);
5109 blocksize = btrfs_level_size(root, level - 1);
5111 next = btrfs_find_tree_block(root, bytenr, blocksize);
5113 next = btrfs_find_create_tree_block(root, bytenr, blocksize);
5116 btrfs_tree_lock(next);
5117 btrfs_set_lock_blocking(next);
5119 ret = btrfs_lookup_extent_info(trans, root, bytenr, blocksize,
5120 &wc->refs[level - 1],
5121 &wc->flags[level - 1]);
5123 BUG_ON(wc->refs[level - 1] == 0);
5126 if (wc->stage == DROP_REFERENCE) {
5127 if (wc->refs[level - 1] > 1) {
5129 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
5132 if (!wc->update_ref ||
5133 generation <= root->root_key.offset)
5136 btrfs_node_key_to_cpu(path->nodes[level], &key,
5137 path->slots[level]);
5138 ret = btrfs_comp_cpu_keys(&key, &wc->update_progress);
5142 wc->stage = UPDATE_BACKREF;
5143 wc->shared_level = level - 1;
5147 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
5151 if (!btrfs_buffer_uptodate(next, generation)) {
5152 btrfs_tree_unlock(next);
5153 free_extent_buffer(next);
5159 if (reada && level == 1)
5160 reada_walk_down(trans, root, wc, path);
5161 next = read_tree_block(root, bytenr, blocksize, generation);
5162 btrfs_tree_lock(next);
5163 btrfs_set_lock_blocking(next);
5167 BUG_ON(level != btrfs_header_level(next));
5168 path->nodes[level] = next;
5169 path->slots[level] = 0;
5170 path->locks[level] = 1;
5176 wc->refs[level - 1] = 0;
5177 wc->flags[level - 1] = 0;
5178 if (wc->stage == DROP_REFERENCE) {
5179 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
5180 parent = path->nodes[level]->start;
5182 BUG_ON(root->root_key.objectid !=
5183 btrfs_header_owner(path->nodes[level]));
5187 ret = btrfs_free_extent(trans, root, bytenr, blocksize, parent,
5188 root->root_key.objectid, level - 1, 0);
5191 btrfs_tree_unlock(next);
5192 free_extent_buffer(next);
5198 * hepler to process tree block while walking up the tree.
5200 * when wc->stage == DROP_REFERENCE, this function drops
5201 * reference count on the block.
5203 * when wc->stage == UPDATE_BACKREF, this function changes
5204 * wc->stage back to DROP_REFERENCE if we changed wc->stage
5205 * to UPDATE_BACKREF previously while processing the block.
5207 * NOTE: return value 1 means we should stop walking up.
5209 static noinline int walk_up_proc(struct btrfs_trans_handle *trans,
5210 struct btrfs_root *root,
5211 struct btrfs_path *path,
5212 struct walk_control *wc)
5215 int level = wc->level;
5216 struct extent_buffer *eb = path->nodes[level];
5219 if (wc->stage == UPDATE_BACKREF) {
5220 BUG_ON(wc->shared_level < level);
5221 if (level < wc->shared_level)
5224 ret = find_next_key(path, level + 1, &wc->update_progress);
5228 wc->stage = DROP_REFERENCE;
5229 wc->shared_level = -1;
5230 path->slots[level] = 0;
5233 * check reference count again if the block isn't locked.
5234 * we should start walking down the tree again if reference
5237 if (!path->locks[level]) {
5239 btrfs_tree_lock(eb);
5240 btrfs_set_lock_blocking(eb);
5241 path->locks[level] = 1;
5243 ret = btrfs_lookup_extent_info(trans, root,
5248 BUG_ON(wc->refs[level] == 0);
5249 if (wc->refs[level] == 1) {
5250 btrfs_tree_unlock(eb);
5251 path->locks[level] = 0;
5257 /* wc->stage == DROP_REFERENCE */
5258 BUG_ON(wc->refs[level] > 1 && !path->locks[level]);
5260 if (wc->refs[level] == 1) {
5262 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
5263 ret = btrfs_dec_ref(trans, root, eb, 1);
5265 ret = btrfs_dec_ref(trans, root, eb, 0);
5268 /* make block locked assertion in clean_tree_block happy */
5269 if (!path->locks[level] &&
5270 btrfs_header_generation(eb) == trans->transid) {
5271 btrfs_tree_lock(eb);
5272 btrfs_set_lock_blocking(eb);
5273 path->locks[level] = 1;
5275 clean_tree_block(trans, root, eb);
5278 if (eb == root->node) {
5279 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
5282 BUG_ON(root->root_key.objectid !=
5283 btrfs_header_owner(eb));
5285 if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
5286 parent = path->nodes[level + 1]->start;
5288 BUG_ON(root->root_key.objectid !=
5289 btrfs_header_owner(path->nodes[level + 1]));
5292 ret = btrfs_free_extent(trans, root, eb->start, eb->len, parent,
5293 root->root_key.objectid, level, 0);
5296 wc->refs[level] = 0;
5297 wc->flags[level] = 0;
5301 static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
5302 struct btrfs_root *root,
5303 struct btrfs_path *path,
5304 struct walk_control *wc)
5306 int level = wc->level;
5307 int lookup_info = 1;
5310 while (level >= 0) {
5311 if (path->slots[level] >=
5312 btrfs_header_nritems(path->nodes[level]))
5315 ret = walk_down_proc(trans, root, path, wc, lookup_info);
5322 ret = do_walk_down(trans, root, path, wc, &lookup_info);
5324 path->slots[level]++;
5332 static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
5333 struct btrfs_root *root,
5334 struct btrfs_path *path,
5335 struct walk_control *wc, int max_level)
5337 int level = wc->level;
5340 path->slots[level] = btrfs_header_nritems(path->nodes[level]);
5341 while (level < max_level && path->nodes[level]) {
5343 if (path->slots[level] + 1 <
5344 btrfs_header_nritems(path->nodes[level])) {
5345 path->slots[level]++;
5348 ret = walk_up_proc(trans, root, path, wc);
5352 if (path->locks[level]) {
5353 btrfs_tree_unlock(path->nodes[level]);
5354 path->locks[level] = 0;
5356 free_extent_buffer(path->nodes[level]);
5357 path->nodes[level] = NULL;
5365 * drop a subvolume tree.
5367 * this function traverses the tree freeing any blocks that only
5368 * referenced by the tree.
5370 * when a shared tree block is found. this function decreases its
5371 * reference count by one. if update_ref is true, this function
5372 * also make sure backrefs for the shared block and all lower level
5373 * blocks are properly updated.
5375 int btrfs_drop_snapshot(struct btrfs_root *root, int update_ref)
5377 struct btrfs_path *path;
5378 struct btrfs_trans_handle *trans;
5379 struct btrfs_root *tree_root = root->fs_info->tree_root;
5380 struct btrfs_root_item *root_item = &root->root_item;
5381 struct walk_control *wc;
5382 struct btrfs_key key;
5387 path = btrfs_alloc_path();
5390 wc = kzalloc(sizeof(*wc), GFP_NOFS);
5393 trans = btrfs_start_transaction(tree_root, 1);
5395 if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
5396 level = btrfs_header_level(root->node);
5397 path->nodes[level] = btrfs_lock_root_node(root);
5398 btrfs_set_lock_blocking(path->nodes[level]);
5399 path->slots[level] = 0;
5400 path->locks[level] = 1;
5401 memset(&wc->update_progress, 0,
5402 sizeof(wc->update_progress));
5404 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
5405 memcpy(&wc->update_progress, &key,
5406 sizeof(wc->update_progress));
5408 level = root_item->drop_level;
5410 path->lowest_level = level;
5411 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5412 path->lowest_level = 0;
5420 * unlock our path, this is safe because only this
5421 * function is allowed to delete this snapshot
5423 btrfs_unlock_up_safe(path, 0);
5425 level = btrfs_header_level(root->node);
5427 btrfs_tree_lock(path->nodes[level]);
5428 btrfs_set_lock_blocking(path->nodes[level]);
5430 ret = btrfs_lookup_extent_info(trans, root,
5431 path->nodes[level]->start,
5432 path->nodes[level]->len,
5436 BUG_ON(wc->refs[level] == 0);
5438 if (level == root_item->drop_level)
5441 btrfs_tree_unlock(path->nodes[level]);
5442 WARN_ON(wc->refs[level] != 1);
5448 wc->shared_level = -1;
5449 wc->stage = DROP_REFERENCE;
5450 wc->update_ref = update_ref;
5452 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
5455 ret = walk_down_tree(trans, root, path, wc);
5461 ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL);
5468 BUG_ON(wc->stage != DROP_REFERENCE);
5472 if (wc->stage == DROP_REFERENCE) {
5474 btrfs_node_key(path->nodes[level],
5475 &root_item->drop_progress,
5476 path->slots[level]);
5477 root_item->drop_level = level;
5480 BUG_ON(wc->level == 0);
5481 if (trans->transaction->in_commit ||
5482 trans->transaction->delayed_refs.flushing) {
5483 ret = btrfs_update_root(trans, tree_root,
5488 btrfs_end_transaction(trans, tree_root);
5489 trans = btrfs_start_transaction(tree_root, 1);
5491 unsigned long update;
5492 update = trans->delayed_ref_updates;
5493 trans->delayed_ref_updates = 0;
5495 btrfs_run_delayed_refs(trans, tree_root,
5499 btrfs_release_path(root, path);
5502 ret = btrfs_del_root(trans, tree_root, &root->root_key);
5505 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
5506 ret = btrfs_find_last_root(tree_root, root->root_key.objectid,
5510 ret = btrfs_del_orphan_item(trans, tree_root,
5511 root->root_key.objectid);
5516 if (root->in_radix) {
5517 btrfs_free_fs_root(tree_root->fs_info, root);
5519 free_extent_buffer(root->node);
5520 free_extent_buffer(root->commit_root);
5524 btrfs_end_transaction(trans, tree_root);
5526 btrfs_free_path(path);
5531 * drop subtree rooted at tree block 'node'.
5533 * NOTE: this function will unlock and release tree block 'node'
5535 int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
5536 struct btrfs_root *root,
5537 struct extent_buffer *node,
5538 struct extent_buffer *parent)
5540 struct btrfs_path *path;
5541 struct walk_control *wc;
5547 BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
5549 path = btrfs_alloc_path();
5552 wc = kzalloc(sizeof(*wc), GFP_NOFS);
5555 btrfs_assert_tree_locked(parent);
5556 parent_level = btrfs_header_level(parent);
5557 extent_buffer_get(parent);
5558 path->nodes[parent_level] = parent;
5559 path->slots[parent_level] = btrfs_header_nritems(parent);
5561 btrfs_assert_tree_locked(node);
5562 level = btrfs_header_level(node);
5563 path->nodes[level] = node;
5564 path->slots[level] = 0;
5565 path->locks[level] = 1;
5567 wc->refs[parent_level] = 1;
5568 wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF;
5570 wc->shared_level = -1;
5571 wc->stage = DROP_REFERENCE;
5574 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
5577 wret = walk_down_tree(trans, root, path, wc);
5583 wret = walk_up_tree(trans, root, path, wc, parent_level);
5591 btrfs_free_path(path);
5596 static unsigned long calc_ra(unsigned long start, unsigned long last,
5599 return min(last, start + nr - 1);
5602 static noinline int relocate_inode_pages(struct inode *inode, u64 start,
5607 unsigned long first_index;
5608 unsigned long last_index;
5611 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
5612 struct file_ra_state *ra;
5613 struct btrfs_ordered_extent *ordered;
5614 unsigned int total_read = 0;
5615 unsigned int total_dirty = 0;
5618 ra = kzalloc(sizeof(*ra), GFP_NOFS);
5620 mutex_lock(&inode->i_mutex);
5621 first_index = start >> PAGE_CACHE_SHIFT;
5622 last_index = (start + len - 1) >> PAGE_CACHE_SHIFT;
5624 /* make sure the dirty trick played by the caller work */
5625 ret = invalidate_inode_pages2_range(inode->i_mapping,
5626 first_index, last_index);
5630 file_ra_state_init(ra, inode->i_mapping);
5632 for (i = first_index ; i <= last_index; i++) {
5633 if (total_read % ra->ra_pages == 0) {
5634 btrfs_force_ra(inode->i_mapping, ra, NULL, i,
5635 calc_ra(i, last_index, ra->ra_pages));
5639 if (((u64)i << PAGE_CACHE_SHIFT) > i_size_read(inode))
5641 page = grab_cache_page(inode->i_mapping, i);
5646 if (!PageUptodate(page)) {
5647 btrfs_readpage(NULL, page);
5649 if (!PageUptodate(page)) {
5651 page_cache_release(page);
5656 wait_on_page_writeback(page);
5658 page_start = (u64)page->index << PAGE_CACHE_SHIFT;
5659 page_end = page_start + PAGE_CACHE_SIZE - 1;
5660 lock_extent(io_tree, page_start, page_end, GFP_NOFS);
5662 ordered = btrfs_lookup_ordered_extent(inode, page_start);
5664 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
5666 page_cache_release(page);
5667 btrfs_start_ordered_extent(inode, ordered, 1);
5668 btrfs_put_ordered_extent(ordered);
5671 set_page_extent_mapped(page);
5673 if (i == first_index)
5674 set_extent_bits(io_tree, page_start, page_end,
5675 EXTENT_BOUNDARY, GFP_NOFS);
5676 btrfs_set_extent_delalloc(inode, page_start, page_end);
5678 set_page_dirty(page);
5681 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
5683 page_cache_release(page);
5688 mutex_unlock(&inode->i_mutex);
5689 balance_dirty_pages_ratelimited_nr(inode->i_mapping, total_dirty);
5693 static noinline int relocate_data_extent(struct inode *reloc_inode,
5694 struct btrfs_key *extent_key,
5697 struct btrfs_root *root = BTRFS_I(reloc_inode)->root;
5698 struct extent_map_tree *em_tree = &BTRFS_I(reloc_inode)->extent_tree;
5699 struct extent_map *em;
5700 u64 start = extent_key->objectid - offset;
5701 u64 end = start + extent_key->offset - 1;
5703 em = alloc_extent_map(GFP_NOFS);
5704 BUG_ON(!em || IS_ERR(em));
5707 em->len = extent_key->offset;
5708 em->block_len = extent_key->offset;
5709 em->block_start = extent_key->objectid;
5710 em->bdev = root->fs_info->fs_devices->latest_bdev;
5711 set_bit(EXTENT_FLAG_PINNED, &em->flags);
5713 /* setup extent map to cheat btrfs_readpage */
5714 lock_extent(&BTRFS_I(reloc_inode)->io_tree, start, end, GFP_NOFS);
5717 write_lock(&em_tree->lock);
5718 ret = add_extent_mapping(em_tree, em);
5719 write_unlock(&em_tree->lock);
5720 if (ret != -EEXIST) {
5721 free_extent_map(em);
5724 btrfs_drop_extent_cache(reloc_inode, start, end, 0);
5726 unlock_extent(&BTRFS_I(reloc_inode)->io_tree, start, end, GFP_NOFS);
5728 return relocate_inode_pages(reloc_inode, start, extent_key->offset);
5731 struct btrfs_ref_path {
5733 u64 nodes[BTRFS_MAX_LEVEL];
5735 u64 root_generation;
5742 struct btrfs_key node_keys[BTRFS_MAX_LEVEL];
5743 u64 new_nodes[BTRFS_MAX_LEVEL];
5746 struct disk_extent {
5757 static int is_cowonly_root(u64 root_objectid)
5759 if (root_objectid == BTRFS_ROOT_TREE_OBJECTID ||
5760 root_objectid == BTRFS_EXTENT_TREE_OBJECTID ||
5761 root_objectid == BTRFS_CHUNK_TREE_OBJECTID ||
5762 root_objectid == BTRFS_DEV_TREE_OBJECTID ||
5763 root_objectid == BTRFS_TREE_LOG_OBJECTID ||
5764 root_objectid == BTRFS_CSUM_TREE_OBJECTID)
5769 static noinline int __next_ref_path(struct btrfs_trans_handle *trans,
5770 struct btrfs_root *extent_root,
5771 struct btrfs_ref_path *ref_path,
5774 struct extent_buffer *leaf;
5775 struct btrfs_path *path;
5776 struct btrfs_extent_ref *ref;
5777 struct btrfs_key key;
5778 struct btrfs_key found_key;
5784 path = btrfs_alloc_path();
5789 ref_path->lowest_level = -1;
5790 ref_path->current_level = -1;
5791 ref_path->shared_level = -1;
5795 level = ref_path->current_level - 1;
5796 while (level >= -1) {
5798 if (level < ref_path->lowest_level)
5802 bytenr = ref_path->nodes[level];
5804 bytenr = ref_path->extent_start;
5805 BUG_ON(bytenr == 0);
5807 parent = ref_path->nodes[level + 1];
5808 ref_path->nodes[level + 1] = 0;
5809 ref_path->current_level = level;
5810 BUG_ON(parent == 0);
5812 key.objectid = bytenr;
5813 key.offset = parent + 1;
5814 key.type = BTRFS_EXTENT_REF_KEY;
5816 ret = btrfs_search_slot(trans, extent_root, &key, path, 0, 0);
5821 leaf = path->nodes[0];
5822 nritems = btrfs_header_nritems(leaf);
5823 if (path->slots[0] >= nritems) {
5824 ret = btrfs_next_leaf(extent_root, path);
5829 leaf = path->nodes[0];
5832 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5833 if (found_key.objectid == bytenr &&
5834 found_key.type == BTRFS_EXTENT_REF_KEY) {
5835 if (level < ref_path->shared_level)
5836 ref_path->shared_level = level;
5841 btrfs_release_path(extent_root, path);
5844 /* reached lowest level */
5848 level = ref_path->current_level;
5849 while (level < BTRFS_MAX_LEVEL - 1) {
5853 bytenr = ref_path->nodes[level];
5855 bytenr = ref_path->extent_start;
5857 BUG_ON(bytenr == 0);
5859 key.objectid = bytenr;
5861 key.type = BTRFS_EXTENT_REF_KEY;
5863 ret = btrfs_search_slot(trans, extent_root, &key, path, 0, 0);
5867 leaf = path->nodes[0];
5868 nritems = btrfs_header_nritems(leaf);
5869 if (path->slots[0] >= nritems) {
5870 ret = btrfs_next_leaf(extent_root, path);
5874 /* the extent was freed by someone */
5875 if (ref_path->lowest_level == level)
5877 btrfs_release_path(extent_root, path);
5880 leaf = path->nodes[0];
5883 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5884 if (found_key.objectid != bytenr ||
5885 found_key.type != BTRFS_EXTENT_REF_KEY) {
5886 /* the extent was freed by someone */
5887 if (ref_path->lowest_level == level) {
5891 btrfs_release_path(extent_root, path);
5895 ref = btrfs_item_ptr(leaf, path->slots[0],
5896 struct btrfs_extent_ref);
5897 ref_objectid = btrfs_ref_objectid(leaf, ref);
5898 if (ref_objectid < BTRFS_FIRST_FREE_OBJECTID) {
5900 level = (int)ref_objectid;
5901 BUG_ON(level >= BTRFS_MAX_LEVEL);
5902 ref_path->lowest_level = level;
5903 ref_path->current_level = level;
5904 ref_path->nodes[level] = bytenr;
5906 WARN_ON(ref_objectid != level);
5909 WARN_ON(level != -1);
5913 if (ref_path->lowest_level == level) {
5914 ref_path->owner_objectid = ref_objectid;
5915 ref_path->num_refs = btrfs_ref_num_refs(leaf, ref);
5919 * the block is tree root or the block isn't in reference
5922 if (found_key.objectid == found_key.offset ||
5923 is_cowonly_root(btrfs_ref_root(leaf, ref))) {
5924 ref_path->root_objectid = btrfs_ref_root(leaf, ref);
5925 ref_path->root_generation =
5926 btrfs_ref_generation(leaf, ref);
5928 /* special reference from the tree log */
5929 ref_path->nodes[0] = found_key.offset;
5930 ref_path->current_level = 0;
5937 BUG_ON(ref_path->nodes[level] != 0);
5938 ref_path->nodes[level] = found_key.offset;
5939 ref_path->current_level = level;
5942 * the reference was created in the running transaction,
5943 * no need to continue walking up.
5945 if (btrfs_ref_generation(leaf, ref) == trans->transid) {
5946 ref_path->root_objectid = btrfs_ref_root(leaf, ref);
5947 ref_path->root_generation =
5948 btrfs_ref_generation(leaf, ref);
5953 btrfs_release_path(extent_root, path);
5956 /* reached max tree level, but no tree root found. */
5959 btrfs_free_path(path);
5963 static int btrfs_first_ref_path(struct btrfs_trans_handle *trans,
5964 struct btrfs_root *extent_root,
5965 struct btrfs_ref_path *ref_path,
5968 memset(ref_path, 0, sizeof(*ref_path));
5969 ref_path->extent_start = extent_start;
5971 return __next_ref_path(trans, extent_root, ref_path, 1);
5974 static int btrfs_next_ref_path(struct btrfs_trans_handle *trans,
5975 struct btrfs_root *extent_root,
5976 struct btrfs_ref_path *ref_path)
5978 return __next_ref_path(trans, extent_root, ref_path, 0);
5981 static noinline int get_new_locations(struct inode *reloc_inode,
5982 struct btrfs_key *extent_key,
5983 u64 offset, int no_fragment,
5984 struct disk_extent **extents,
5987 struct btrfs_root *root = BTRFS_I(reloc_inode)->root;
5988 struct btrfs_path *path;
5989 struct btrfs_file_extent_item *fi;
5990 struct extent_buffer *leaf;
5991 struct disk_extent *exts = *extents;
5992 struct btrfs_key found_key;
5997 int max = *nr_extents;
6000 WARN_ON(!no_fragment && *extents);
6003 exts = kmalloc(sizeof(*exts) * max, GFP_NOFS);
6008 path = btrfs_alloc_path();
6011 cur_pos = extent_key->objectid - offset;
6012 last_byte = extent_key->objectid + extent_key->offset;
6013 ret = btrfs_lookup_file_extent(NULL, root, path, reloc_inode->i_ino,
6023 leaf = path->nodes[0];
6024 nritems = btrfs_header_nritems(leaf);
6025 if (path->slots[0] >= nritems) {
6026 ret = btrfs_next_leaf(root, path);
6031 leaf = path->nodes[0];
6034 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6035 if (found_key.offset != cur_pos ||
6036 found_key.type != BTRFS_EXTENT_DATA_KEY ||
6037 found_key.objectid != reloc_inode->i_ino)
6040 fi = btrfs_item_ptr(leaf, path->slots[0],
6041 struct btrfs_file_extent_item);
6042 if (btrfs_file_extent_type(leaf, fi) !=
6043 BTRFS_FILE_EXTENT_REG ||
6044 btrfs_file_extent_disk_bytenr(leaf, fi) == 0)
6048 struct disk_extent *old = exts;
6050 exts = kzalloc(sizeof(*exts) * max, GFP_NOFS);
6051 memcpy(exts, old, sizeof(*exts) * nr);
6052 if (old != *extents)
6056 exts[nr].disk_bytenr =
6057 btrfs_file_extent_disk_bytenr(leaf, fi);
6058 exts[nr].disk_num_bytes =
6059 btrfs_file_extent_disk_num_bytes(leaf, fi);
6060 exts[nr].offset = btrfs_file_extent_offset(leaf, fi);
6061 exts[nr].num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
6062 exts[nr].ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
6063 exts[nr].compression = btrfs_file_extent_compression(leaf, fi);
6064 exts[nr].encryption = btrfs_file_extent_encryption(leaf, fi);
6065 exts[nr].other_encoding = btrfs_file_extent_other_encoding(leaf,
6067 BUG_ON(exts[nr].offset > 0);
6068 BUG_ON(exts[nr].compression || exts[nr].encryption);
6069 BUG_ON(exts[nr].num_bytes != exts[nr].disk_num_bytes);
6071 cur_pos += exts[nr].num_bytes;
6074 if (cur_pos + offset >= last_byte)
6084 BUG_ON(cur_pos + offset > last_byte);
6085 if (cur_pos + offset < last_byte) {
6091 btrfs_free_path(path);
6093 if (exts != *extents)
6102 static noinline int replace_one_extent(struct btrfs_trans_handle *trans,
6103 struct btrfs_root *root,
6104 struct btrfs_path *path,
6105 struct btrfs_key *extent_key,
6106 struct btrfs_key *leaf_key,
6107 struct btrfs_ref_path *ref_path,
6108 struct disk_extent *new_extents,
6111 struct extent_buffer *leaf;
6112 struct btrfs_file_extent_item *fi;
6113 struct inode *inode = NULL;
6114 struct btrfs_key key;
6119 u64 search_end = (u64)-1;
6122 int extent_locked = 0;
6126 memcpy(&key, leaf_key, sizeof(key));
6127 if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS) {
6128 if (key.objectid < ref_path->owner_objectid ||
6129 (key.objectid == ref_path->owner_objectid &&
6130 key.type < BTRFS_EXTENT_DATA_KEY)) {
6131 key.objectid = ref_path->owner_objectid;
6132 key.type = BTRFS_EXTENT_DATA_KEY;
6138 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
6142 leaf = path->nodes[0];
6143 nritems = btrfs_header_nritems(leaf);
6145 if (extent_locked && ret > 0) {
6147 * the file extent item was modified by someone
6148 * before the extent got locked.
6150 unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
6151 lock_end, GFP_NOFS);
6155 if (path->slots[0] >= nritems) {
6156 if (++nr_scaned > 2)
6159 BUG_ON(extent_locked);
6160 ret = btrfs_next_leaf(root, path);
6165 leaf = path->nodes[0];
6166 nritems = btrfs_header_nritems(leaf);
6169 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
6171 if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS) {
6172 if ((key.objectid > ref_path->owner_objectid) ||
6173 (key.objectid == ref_path->owner_objectid &&
6174 key.type > BTRFS_EXTENT_DATA_KEY) ||
6175 key.offset >= search_end)
6179 if (inode && key.objectid != inode->i_ino) {
6180 BUG_ON(extent_locked);
6181 btrfs_release_path(root, path);
6182 mutex_unlock(&inode->i_mutex);
6188 if (key.type != BTRFS_EXTENT_DATA_KEY) {
6193 fi = btrfs_item_ptr(leaf, path->slots[0],
6194 struct btrfs_file_extent_item);
6195 extent_type = btrfs_file_extent_type(leaf, fi);
6196 if ((extent_type != BTRFS_FILE_EXTENT_REG &&
6197 extent_type != BTRFS_FILE_EXTENT_PREALLOC) ||
6198 (btrfs_file_extent_disk_bytenr(leaf, fi) !=
6199 extent_key->objectid)) {
6205 num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
6206 ext_offset = btrfs_file_extent_offset(leaf, fi);
6208 if (search_end == (u64)-1) {
6209 search_end = key.offset - ext_offset +
6210 btrfs_file_extent_ram_bytes(leaf, fi);
6213 if (!extent_locked) {
6214 lock_start = key.offset;
6215 lock_end = lock_start + num_bytes - 1;
6217 if (lock_start > key.offset ||
6218 lock_end + 1 < key.offset + num_bytes) {
6219 unlock_extent(&BTRFS_I(inode)->io_tree,
6220 lock_start, lock_end, GFP_NOFS);
6226 btrfs_release_path(root, path);
6228 inode = btrfs_iget_locked(root->fs_info->sb,
6229 key.objectid, root);
6230 if (inode->i_state & I_NEW) {
6231 BTRFS_I(inode)->root = root;
6232 BTRFS_I(inode)->location.objectid =
6234 BTRFS_I(inode)->location.type =
6235 BTRFS_INODE_ITEM_KEY;
6236 BTRFS_I(inode)->location.offset = 0;
6237 btrfs_read_locked_inode(inode);
6238 unlock_new_inode(inode);
6241 * some code call btrfs_commit_transaction while
6242 * holding the i_mutex, so we can't use mutex_lock
6245 if (is_bad_inode(inode) ||
6246 !mutex_trylock(&inode->i_mutex)) {
6249 key.offset = (u64)-1;
6254 if (!extent_locked) {
6255 struct btrfs_ordered_extent *ordered;
6257 btrfs_release_path(root, path);
6259 lock_extent(&BTRFS_I(inode)->io_tree, lock_start,
6260 lock_end, GFP_NOFS);
6261 ordered = btrfs_lookup_first_ordered_extent(inode,
6264 ordered->file_offset <= lock_end &&
6265 ordered->file_offset + ordered->len > lock_start) {
6266 unlock_extent(&BTRFS_I(inode)->io_tree,
6267 lock_start, lock_end, GFP_NOFS);
6268 btrfs_start_ordered_extent(inode, ordered, 1);
6269 btrfs_put_ordered_extent(ordered);
6270 key.offset += num_bytes;
6274 btrfs_put_ordered_extent(ordered);
6280 if (nr_extents == 1) {
6281 /* update extent pointer in place */
6282 btrfs_set_file_extent_disk_bytenr(leaf, fi,
6283 new_extents[0].disk_bytenr);
6284 btrfs_set_file_extent_disk_num_bytes(leaf, fi,
6285 new_extents[0].disk_num_bytes);
6286 btrfs_mark_buffer_dirty(leaf);
6288 btrfs_drop_extent_cache(inode, key.offset,
6289 key.offset + num_bytes - 1, 0);
6291 ret = btrfs_inc_extent_ref(trans, root,
6292 new_extents[0].disk_bytenr,
6293 new_extents[0].disk_num_bytes,
6295 root->root_key.objectid,
6300 ret = btrfs_free_extent(trans, root,
6301 extent_key->objectid,
6304 btrfs_header_owner(leaf),
6305 btrfs_header_generation(leaf),
6309 btrfs_release_path(root, path);
6310 key.offset += num_bytes;
6318 * drop old extent pointer at first, then insert the
6319 * new pointers one bye one
6321 btrfs_release_path(root, path);
6322 ret = btrfs_drop_extents(trans, root, inode, key.offset,
6323 key.offset + num_bytes,
6324 key.offset, &alloc_hint);
6327 for (i = 0; i < nr_extents; i++) {
6328 if (ext_offset >= new_extents[i].num_bytes) {
6329 ext_offset -= new_extents[i].num_bytes;
6332 extent_len = min(new_extents[i].num_bytes -
6333 ext_offset, num_bytes);
6335 ret = btrfs_insert_empty_item(trans, root,
6340 leaf = path->nodes[0];
6341 fi = btrfs_item_ptr(leaf, path->slots[0],
6342 struct btrfs_file_extent_item);
6343 btrfs_set_file_extent_generation(leaf, fi,
6345 btrfs_set_file_extent_type(leaf, fi,
6346 BTRFS_FILE_EXTENT_REG);
6347 btrfs_set_file_extent_disk_bytenr(leaf, fi,
6348 new_extents[i].disk_bytenr);
6349 btrfs_set_file_extent_disk_num_bytes(leaf, fi,
6350 new_extents[i].disk_num_bytes);
6351 btrfs_set_file_extent_ram_bytes(leaf, fi,
6352 new_extents[i].ram_bytes);
6354 btrfs_set_file_extent_compression(leaf, fi,
6355 new_extents[i].compression);
6356 btrfs_set_file_extent_encryption(leaf, fi,
6357 new_extents[i].encryption);
6358 btrfs_set_file_extent_other_encoding(leaf, fi,
6359 new_extents[i].other_encoding);
6361 btrfs_set_file_extent_num_bytes(leaf, fi,
6363 ext_offset += new_extents[i].offset;
6364 btrfs_set_file_extent_offset(leaf, fi,
6366 btrfs_mark_buffer_dirty(leaf);
6368 btrfs_drop_extent_cache(inode, key.offset,
6369 key.offset + extent_len - 1, 0);
6371 ret = btrfs_inc_extent_ref(trans, root,
6372 new_extents[i].disk_bytenr,
6373 new_extents[i].disk_num_bytes,
6375 root->root_key.objectid,
6376 trans->transid, key.objectid);
6378 btrfs_release_path(root, path);
6380 inode_add_bytes(inode, extent_len);
6383 num_bytes -= extent_len;
6384 key.offset += extent_len;
6389 BUG_ON(i >= nr_extents);
6393 if (extent_locked) {
6394 unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
6395 lock_end, GFP_NOFS);
6399 if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS &&
6400 key.offset >= search_end)
6407 btrfs_release_path(root, path);
6409 mutex_unlock(&inode->i_mutex);
6410 if (extent_locked) {
6411 unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
6412 lock_end, GFP_NOFS);
6419 int btrfs_reloc_tree_cache_ref(struct btrfs_trans_handle *trans,
6420 struct btrfs_root *root,
6421 struct extent_buffer *buf, u64 orig_start)
6426 BUG_ON(btrfs_header_generation(buf) != trans->transid);
6427 BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
6429 level = btrfs_header_level(buf);
6431 struct btrfs_leaf_ref *ref;
6432 struct btrfs_leaf_ref *orig_ref;
6434 orig_ref = btrfs_lookup_leaf_ref(root, orig_start);
6438 ref = btrfs_alloc_leaf_ref(root, orig_ref->nritems);
6440 btrfs_free_leaf_ref(root, orig_ref);
6444 ref->nritems = orig_ref->nritems;
6445 memcpy(ref->extents, orig_ref->extents,
6446 sizeof(ref->extents[0]) * ref->nritems);
6448 btrfs_free_leaf_ref(root, orig_ref);
6450 ref->root_gen = trans->transid;
6451 ref->bytenr = buf->start;
6452 ref->owner = btrfs_header_owner(buf);
6453 ref->generation = btrfs_header_generation(buf);
6455 ret = btrfs_add_leaf_ref(root, ref, 0);
6457 btrfs_free_leaf_ref(root, ref);
6462 static noinline int invalidate_extent_cache(struct btrfs_root *root,
6463 struct extent_buffer *leaf,
6464 struct btrfs_block_group_cache *group,
6465 struct btrfs_root *target_root)
6467 struct btrfs_key key;
6468 struct inode *inode = NULL;
6469 struct btrfs_file_extent_item *fi;
6471 u64 skip_objectid = 0;
6475 nritems = btrfs_header_nritems(leaf);
6476 for (i = 0; i < nritems; i++) {
6477 btrfs_item_key_to_cpu(leaf, &key, i);
6478 if (key.objectid == skip_objectid ||
6479 key.type != BTRFS_EXTENT_DATA_KEY)
6481 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
6482 if (btrfs_file_extent_type(leaf, fi) ==
6483 BTRFS_FILE_EXTENT_INLINE)
6485 if (btrfs_file_extent_disk_bytenr(leaf, fi) == 0)
6487 if (!inode || inode->i_ino != key.objectid) {
6489 inode = btrfs_ilookup(target_root->fs_info->sb,
6490 key.objectid, target_root, 1);
6493 skip_objectid = key.objectid;
6496 num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
6498 lock_extent(&BTRFS_I(inode)->io_tree, key.offset,
6499 key.offset + num_bytes - 1, GFP_NOFS);
6500 btrfs_drop_extent_cache(inode, key.offset,
6501 key.offset + num_bytes - 1, 1);
6502 unlock_extent(&BTRFS_I(inode)->io_tree, key.offset,
6503 key.offset + num_bytes - 1, GFP_NOFS);
6510 static noinline int replace_extents_in_leaf(struct btrfs_trans_handle *trans,
6511 struct btrfs_root *root,
6512 struct extent_buffer *leaf,
6513 struct btrfs_block_group_cache *group,
6514 struct inode *reloc_inode)
6516 struct btrfs_key key;
6517 struct btrfs_key extent_key;
6518 struct btrfs_file_extent_item *fi;
6519 struct btrfs_leaf_ref *ref;
6520 struct disk_extent *new_extent;
6529 new_extent = kmalloc(sizeof(*new_extent), GFP_NOFS);
6530 BUG_ON(!new_extent);
6532 ref = btrfs_lookup_leaf_ref(root, leaf->start);
6536 nritems = btrfs_header_nritems(leaf);
6537 for (i = 0; i < nritems; i++) {
6538 btrfs_item_key_to_cpu(leaf, &key, i);
6539 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
6541 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
6542 if (btrfs_file_extent_type(leaf, fi) ==
6543 BTRFS_FILE_EXTENT_INLINE)
6545 bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
6546 num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
6551 if (bytenr >= group->key.objectid + group->key.offset ||
6552 bytenr + num_bytes <= group->key.objectid)
6555 extent_key.objectid = bytenr;
6556 extent_key.offset = num_bytes;
6557 extent_key.type = BTRFS_EXTENT_ITEM_KEY;
6559 ret = get_new_locations(reloc_inode, &extent_key,
6560 group->key.objectid, 1,
6561 &new_extent, &nr_extent);
6566 BUG_ON(ref->extents[ext_index].bytenr != bytenr);
6567 BUG_ON(ref->extents[ext_index].num_bytes != num_bytes);
6568 ref->extents[ext_index].bytenr = new_extent->disk_bytenr;
6569 ref->extents[ext_index].num_bytes = new_extent->disk_num_bytes;
6571 btrfs_set_file_extent_disk_bytenr(leaf, fi,
6572 new_extent->disk_bytenr);
6573 btrfs_set_file_extent_disk_num_bytes(leaf, fi,
6574 new_extent->disk_num_bytes);
6575 btrfs_mark_buffer_dirty(leaf);
6577 ret = btrfs_inc_extent_ref(trans, root,
6578 new_extent->disk_bytenr,
6579 new_extent->disk_num_bytes,
6581 root->root_key.objectid,
6582 trans->transid, key.objectid);
6585 ret = btrfs_free_extent(trans, root,
6586 bytenr, num_bytes, leaf->start,
6587 btrfs_header_owner(leaf),
6588 btrfs_header_generation(leaf),
6594 BUG_ON(ext_index + 1 != ref->nritems);
6595 btrfs_free_leaf_ref(root, ref);
6599 int btrfs_free_reloc_root(struct btrfs_trans_handle *trans,
6600 struct btrfs_root *root)
6602 struct btrfs_root *reloc_root;
6605 if (root->reloc_root) {
6606 reloc_root = root->reloc_root;
6607 root->reloc_root = NULL;
6608 list_add(&reloc_root->dead_list,
6609 &root->fs_info->dead_reloc_roots);
6611 btrfs_set_root_bytenr(&reloc_root->root_item,
6612 reloc_root->node->start);
6613 btrfs_set_root_level(&root->root_item,
6614 btrfs_header_level(reloc_root->node));
6615 memset(&reloc_root->root_item.drop_progress, 0,
6616 sizeof(struct btrfs_disk_key));
6617 reloc_root->root_item.drop_level = 0;
6619 ret = btrfs_update_root(trans, root->fs_info->tree_root,
6620 &reloc_root->root_key,
6621 &reloc_root->root_item);
6627 int btrfs_drop_dead_reloc_roots(struct btrfs_root *root)
6629 struct btrfs_trans_handle *trans;
6630 struct btrfs_root *reloc_root;
6631 struct btrfs_root *prev_root = NULL;
6632 struct list_head dead_roots;
6636 INIT_LIST_HEAD(&dead_roots);
6637 list_splice_init(&root->fs_info->dead_reloc_roots, &dead_roots);
6639 while (!list_empty(&dead_roots)) {
6640 reloc_root = list_entry(dead_roots.prev,
6641 struct btrfs_root, dead_list);
6642 list_del_init(&reloc_root->dead_list);
6644 BUG_ON(reloc_root->commit_root != NULL);
6646 trans = btrfs_join_transaction(root, 1);
6649 mutex_lock(&root->fs_info->drop_mutex);
6650 ret = btrfs_drop_snapshot(trans, reloc_root);
6653 mutex_unlock(&root->fs_info->drop_mutex);
6655 nr = trans->blocks_used;
6656 ret = btrfs_end_transaction(trans, root);
6658 btrfs_btree_balance_dirty(root, nr);
6661 free_extent_buffer(reloc_root->node);
6663 ret = btrfs_del_root(trans, root->fs_info->tree_root,
6664 &reloc_root->root_key);
6666 mutex_unlock(&root->fs_info->drop_mutex);
6668 nr = trans->blocks_used;
6669 ret = btrfs_end_transaction(trans, root);
6671 btrfs_btree_balance_dirty(root, nr);
6674 prev_root = reloc_root;
6677 btrfs_remove_leaf_refs(prev_root, (u64)-1, 0);
6683 int btrfs_add_dead_reloc_root(struct btrfs_root *root)
6685 list_add(&root->dead_list, &root->fs_info->dead_reloc_roots);
6689 int btrfs_cleanup_reloc_trees(struct btrfs_root *root)
6691 struct btrfs_root *reloc_root;
6692 struct btrfs_trans_handle *trans;
6693 struct btrfs_key location;
6697 mutex_lock(&root->fs_info->tree_reloc_mutex);
6698 ret = btrfs_find_dead_roots(root, BTRFS_TREE_RELOC_OBJECTID, NULL);
6700 found = !list_empty(&root->fs_info->dead_reloc_roots);
6701 mutex_unlock(&root->fs_info->tree_reloc_mutex);
6704 trans = btrfs_start_transaction(root, 1);
6706 ret = btrfs_commit_transaction(trans, root);
6710 location.objectid = BTRFS_DATA_RELOC_TREE_OBJECTID;
6711 location.offset = (u64)-1;
6712 location.type = BTRFS_ROOT_ITEM_KEY;
6714 reloc_root = btrfs_read_fs_root_no_name(root->fs_info, &location);
6715 BUG_ON(!reloc_root);
6716 btrfs_orphan_cleanup(reloc_root);
6720 static noinline int init_reloc_tree(struct btrfs_trans_handle *trans,
6721 struct btrfs_root *root)
6723 struct btrfs_root *reloc_root;
6724 struct extent_buffer *eb;
6725 struct btrfs_root_item *root_item;
6726 struct btrfs_key root_key;
6729 BUG_ON(!root->ref_cows);
6730 if (root->reloc_root)
6733 root_item = kmalloc(sizeof(*root_item), GFP_NOFS);
6736 ret = btrfs_copy_root(trans, root, root->commit_root,
6737 &eb, BTRFS_TREE_RELOC_OBJECTID);
6740 root_key.objectid = BTRFS_TREE_RELOC_OBJECTID;
6741 root_key.offset = root->root_key.objectid;
6742 root_key.type = BTRFS_ROOT_ITEM_KEY;
6744 memcpy(root_item, &root->root_item, sizeof(root_item));
6745 btrfs_set_root_refs(root_item, 0);
6746 btrfs_set_root_bytenr(root_item, eb->start);
6747 btrfs_set_root_level(root_item, btrfs_header_level(eb));
6748 btrfs_set_root_generation(root_item, trans->transid);
6750 btrfs_tree_unlock(eb);
6751 free_extent_buffer(eb);
6753 ret = btrfs_insert_root(trans, root->fs_info->tree_root,
6754 &root_key, root_item);
6758 reloc_root = btrfs_read_fs_root_no_radix(root->fs_info->tree_root,
6760 BUG_ON(!reloc_root);
6761 reloc_root->last_trans = trans->transid;
6762 reloc_root->commit_root = NULL;
6763 reloc_root->ref_tree = &root->fs_info->reloc_ref_tree;
6765 root->reloc_root = reloc_root;
6770 * Core function of space balance.
6772 * The idea is using reloc trees to relocate tree blocks in reference
6773 * counted roots. There is one reloc tree for each subvol, and all
6774 * reloc trees share same root key objectid. Reloc trees are snapshots
6775 * of the latest committed roots of subvols (root->commit_root).
6777 * To relocate a tree block referenced by a subvol, there are two steps.
6778 * COW the block through subvol's reloc tree, then update block pointer
6779 * in the subvol to point to the new block. Since all reloc trees share
6780 * same root key objectid, doing special handing for tree blocks owned
6781 * by them is easy. Once a tree block has been COWed in one reloc tree,
6782 * we can use the resulting new block directly when the same block is
6783 * required to COW again through other reloc trees. By this way, relocated
6784 * tree blocks are shared between reloc trees, so they are also shared
6787 static noinline int relocate_one_path(struct btrfs_trans_handle *trans,
6788 struct btrfs_root *root,
6789 struct btrfs_path *path,
6790 struct btrfs_key *first_key,
6791 struct btrfs_ref_path *ref_path,
6792 struct btrfs_block_group_cache *group,
6793 struct inode *reloc_inode)
6795 struct btrfs_root *reloc_root;
6796 struct extent_buffer *eb = NULL;
6797 struct btrfs_key *keys;
6801 int lowest_level = 0;
6804 if (ref_path->owner_objectid < BTRFS_FIRST_FREE_OBJECTID)
6805 lowest_level = ref_path->owner_objectid;
6807 if (!root->ref_cows) {
6808 path->lowest_level = lowest_level;
6809 ret = btrfs_search_slot(trans, root, first_key, path, 0, 1);
6811 path->lowest_level = 0;
6812 btrfs_release_path(root, path);
6816 mutex_lock(&root->fs_info->tree_reloc_mutex);
6817 ret = init_reloc_tree(trans, root);
6819 reloc_root = root->reloc_root;
6821 shared_level = ref_path->shared_level;
6822 ref_path->shared_level = BTRFS_MAX_LEVEL - 1;
6824 keys = ref_path->node_keys;
6825 nodes = ref_path->new_nodes;
6826 memset(&keys[shared_level + 1], 0,
6827 sizeof(*keys) * (BTRFS_MAX_LEVEL - shared_level - 1));
6828 memset(&nodes[shared_level + 1], 0,
6829 sizeof(*nodes) * (BTRFS_MAX_LEVEL - shared_level - 1));
6831 if (nodes[lowest_level] == 0) {
6832 path->lowest_level = lowest_level;
6833 ret = btrfs_search_slot(trans, reloc_root, first_key, path,
6836 for (level = lowest_level; level < BTRFS_MAX_LEVEL; level++) {
6837 eb = path->nodes[level];
6838 if (!eb || eb == reloc_root->node)
6840 nodes[level] = eb->start;
6842 btrfs_item_key_to_cpu(eb, &keys[level], 0);
6844 btrfs_node_key_to_cpu(eb, &keys[level], 0);
6847 ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
6848 eb = path->nodes[0];
6849 ret = replace_extents_in_leaf(trans, reloc_root, eb,
6850 group, reloc_inode);
6853 btrfs_release_path(reloc_root, path);
6855 ret = btrfs_merge_path(trans, reloc_root, keys, nodes,
6861 * replace tree blocks in the fs tree with tree blocks in
6864 ret = btrfs_merge_path(trans, root, keys, nodes, lowest_level);
6867 if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
6868 ret = btrfs_search_slot(trans, reloc_root, first_key, path,
6871 extent_buffer_get(path->nodes[0]);
6872 eb = path->nodes[0];
6873 btrfs_release_path(reloc_root, path);
6874 ret = invalidate_extent_cache(reloc_root, eb, group, root);
6876 free_extent_buffer(eb);
6879 mutex_unlock(&root->fs_info->tree_reloc_mutex);
6880 path->lowest_level = 0;
6884 static noinline int relocate_tree_block(struct btrfs_trans_handle *trans,
6885 struct btrfs_root *root,
6886 struct btrfs_path *path,
6887 struct btrfs_key *first_key,
6888 struct btrfs_ref_path *ref_path)
6892 ret = relocate_one_path(trans, root, path, first_key,
6893 ref_path, NULL, NULL);
6899 static noinline int del_extent_zero(struct btrfs_trans_handle *trans,
6900 struct btrfs_root *extent_root,
6901 struct btrfs_path *path,
6902 struct btrfs_key *extent_key)
6906 ret = btrfs_search_slot(trans, extent_root, extent_key, path, -1, 1);
6909 ret = btrfs_del_item(trans, extent_root, path);
6911 btrfs_release_path(extent_root, path);
6915 static noinline struct btrfs_root *read_ref_root(struct btrfs_fs_info *fs_info,
6916 struct btrfs_ref_path *ref_path)
6918 struct btrfs_key root_key;
6920 root_key.objectid = ref_path->root_objectid;
6921 root_key.type = BTRFS_ROOT_ITEM_KEY;
6922 if (is_cowonly_root(ref_path->root_objectid))
6923 root_key.offset = 0;
6925 root_key.offset = (u64)-1;
6927 return btrfs_read_fs_root_no_name(fs_info, &root_key);
6930 static noinline int relocate_one_extent(struct btrfs_root *extent_root,
6931 struct btrfs_path *path,
6932 struct btrfs_key *extent_key,
6933 struct btrfs_block_group_cache *group,
6934 struct inode *reloc_inode, int pass)
6936 struct btrfs_trans_handle *trans;
6937 struct btrfs_root *found_root;
6938 struct btrfs_ref_path *ref_path = NULL;
6939 struct disk_extent *new_extents = NULL;
6944 struct btrfs_key first_key;
6948 trans = btrfs_start_transaction(extent_root, 1);
6951 if (extent_key->objectid == 0) {
6952 ret = del_extent_zero(trans, extent_root, path, extent_key);
6956 ref_path = kmalloc(sizeof(*ref_path), GFP_NOFS);
6962 for (loops = 0; ; loops++) {
6964 ret = btrfs_first_ref_path(trans, extent_root, ref_path,
6965 extent_key->objectid);
6967 ret = btrfs_next_ref_path(trans, extent_root, ref_path);
6974 if (ref_path->root_objectid == BTRFS_TREE_LOG_OBJECTID ||
6975 ref_path->root_objectid == BTRFS_TREE_RELOC_OBJECTID)
6978 found_root = read_ref_root(extent_root->fs_info, ref_path);
6979 BUG_ON(!found_root);
6981 * for reference counted tree, only process reference paths
6982 * rooted at the latest committed root.
6984 if (found_root->ref_cows &&
6985 ref_path->root_generation != found_root->root_key.offset)
6988 if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
6991 * copy data extents to new locations
6993 u64 group_start = group->key.objectid;
6994 ret = relocate_data_extent(reloc_inode,
7003 level = ref_path->owner_objectid;
7006 if (prev_block != ref_path->nodes[level]) {
7007 struct extent_buffer *eb;
7008 u64 block_start = ref_path->nodes[level];
7009 u64 block_size = btrfs_level_size(found_root, level);
7011 eb = read_tree_block(found_root, block_start,
7013 btrfs_tree_lock(eb);
7014 BUG_ON(level != btrfs_header_level(eb));
7017 btrfs_item_key_to_cpu(eb, &first_key, 0);
7019 btrfs_node_key_to_cpu(eb, &first_key, 0);
7021 btrfs_tree_unlock(eb);
7022 free_extent_buffer(eb);
7023 prev_block = block_start;
7026 mutex_lock(&extent_root->fs_info->trans_mutex);
7027 btrfs_record_root_in_trans(found_root);
7028 mutex_unlock(&extent_root->fs_info->trans_mutex);
7029 if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
7031 * try to update data extent references while
7032 * keeping metadata shared between snapshots.
7035 ret = relocate_one_path(trans, found_root,
7036 path, &first_key, ref_path,
7037 group, reloc_inode);
7043 * use fallback method to process the remaining
7047 u64 group_start = group->key.objectid;
7048 new_extents = kmalloc(sizeof(*new_extents),
7051 ret = get_new_locations(reloc_inode,
7059 ret = replace_one_extent(trans, found_root,
7061 &first_key, ref_path,
7062 new_extents, nr_extents);
7064 ret = relocate_tree_block(trans, found_root, path,
7065 &first_key, ref_path);
7072 btrfs_end_transaction(trans, extent_root);
7079 static u64 update_block_group_flags(struct btrfs_root *root, u64 flags)
7082 u64 stripped = BTRFS_BLOCK_GROUP_RAID0 |
7083 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
7085 num_devices = root->fs_info->fs_devices->rw_devices;
7086 if (num_devices == 1) {
7087 stripped |= BTRFS_BLOCK_GROUP_DUP;
7088 stripped = flags & ~stripped;
7090 /* turn raid0 into single device chunks */
7091 if (flags & BTRFS_BLOCK_GROUP_RAID0)
7094 /* turn mirroring into duplication */
7095 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
7096 BTRFS_BLOCK_GROUP_RAID10))
7097 return stripped | BTRFS_BLOCK_GROUP_DUP;
7100 /* they already had raid on here, just return */
7101 if (flags & stripped)
7104 stripped |= BTRFS_BLOCK_GROUP_DUP;
7105 stripped = flags & ~stripped;
7107 /* switch duplicated blocks with raid1 */
7108 if (flags & BTRFS_BLOCK_GROUP_DUP)
7109 return stripped | BTRFS_BLOCK_GROUP_RAID1;
7111 /* turn single device chunks into raid0 */
7112 return stripped | BTRFS_BLOCK_GROUP_RAID0;
7117 static int __alloc_chunk_for_shrink(struct btrfs_root *root,
7118 struct btrfs_block_group_cache *shrink_block_group,
7121 struct btrfs_trans_handle *trans;
7122 u64 new_alloc_flags;
7125 spin_lock(&shrink_block_group->lock);
7126 if (btrfs_block_group_used(&shrink_block_group->item) +
7127 shrink_block_group->reserved > 0) {
7128 spin_unlock(&shrink_block_group->lock);
7130 trans = btrfs_start_transaction(root, 1);
7131 spin_lock(&shrink_block_group->lock);
7133 new_alloc_flags = update_block_group_flags(root,
7134 shrink_block_group->flags);
7135 if (new_alloc_flags != shrink_block_group->flags) {
7137 btrfs_block_group_used(&shrink_block_group->item);
7139 calc = shrink_block_group->key.offset;
7141 spin_unlock(&shrink_block_group->lock);
7143 do_chunk_alloc(trans, root->fs_info->extent_root,
7144 calc + 2 * 1024 * 1024, new_alloc_flags, force);
7146 btrfs_end_transaction(trans, root);
7148 spin_unlock(&shrink_block_group->lock);
7153 int btrfs_prepare_block_group_relocation(struct btrfs_root *root,
7154 struct btrfs_block_group_cache *group)
7157 __alloc_chunk_for_shrink(root, group, 1);
7158 set_block_group_readonly(group);
7163 * checks to see if its even possible to relocate this block group.
7165 * @return - -1 if it's not a good idea to relocate this block group, 0 if its
7166 * ok to go ahead and try.
7168 int btrfs_can_relocate(struct btrfs_root *root, u64 bytenr)
7170 struct btrfs_block_group_cache *block_group;
7171 struct btrfs_space_info *space_info;
7172 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
7173 struct btrfs_device *device;
7177 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
7179 /* odd, couldn't find the block group, leave it alone */
7183 /* no bytes used, we're good */
7184 if (!btrfs_block_group_used(&block_group->item))
7187 space_info = block_group->space_info;
7188 spin_lock(&space_info->lock);
7190 full = space_info->full;
7193 * if this is the last block group we have in this space, we can't
7194 * relocate it unless we're able to allocate a new chunk below.
7196 * Otherwise, we need to make sure we have room in the space to handle
7197 * all of the extents from this block group. If we can, we're good
7199 if ((space_info->total_bytes != block_group->key.offset) &&
7200 (space_info->bytes_used + space_info->bytes_reserved +
7201 space_info->bytes_pinned + space_info->bytes_readonly +
7202 btrfs_block_group_used(&block_group->item) <
7203 space_info->total_bytes)) {
7204 spin_unlock(&space_info->lock);
7207 spin_unlock(&space_info->lock);
7210 * ok we don't have enough space, but maybe we have free space on our
7211 * devices to allocate new chunks for relocation, so loop through our
7212 * alloc devices and guess if we have enough space. However, if we
7213 * were marked as full, then we know there aren't enough chunks, and we
7220 mutex_lock(&root->fs_info->chunk_mutex);
7221 list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
7222 u64 min_free = btrfs_block_group_used(&block_group->item);
7223 u64 dev_offset, max_avail;
7226 * check to make sure we can actually find a chunk with enough
7227 * space to fit our block group in.
7229 if (device->total_bytes > device->bytes_used + min_free) {
7230 ret = find_free_dev_extent(NULL, device, min_free,
7231 &dev_offset, &max_avail);
7237 mutex_unlock(&root->fs_info->chunk_mutex);
7239 btrfs_put_block_group(block_group);
7243 static int find_first_block_group(struct btrfs_root *root,
7244 struct btrfs_path *path, struct btrfs_key *key)
7247 struct btrfs_key found_key;
7248 struct extent_buffer *leaf;
7251 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
7256 slot = path->slots[0];
7257 leaf = path->nodes[0];
7258 if (slot >= btrfs_header_nritems(leaf)) {
7259 ret = btrfs_next_leaf(root, path);
7266 btrfs_item_key_to_cpu(leaf, &found_key, slot);
7268 if (found_key.objectid >= key->objectid &&
7269 found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
7280 int btrfs_free_block_groups(struct btrfs_fs_info *info)
7282 struct btrfs_block_group_cache *block_group;
7283 struct btrfs_space_info *space_info;
7284 struct btrfs_caching_control *caching_ctl;
7287 down_write(&info->extent_commit_sem);
7288 while (!list_empty(&info->caching_block_groups)) {
7289 caching_ctl = list_entry(info->caching_block_groups.next,
7290 struct btrfs_caching_control, list);
7291 list_del(&caching_ctl->list);
7292 put_caching_control(caching_ctl);
7294 up_write(&info->extent_commit_sem);
7296 spin_lock(&info->block_group_cache_lock);
7297 while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
7298 block_group = rb_entry(n, struct btrfs_block_group_cache,
7300 rb_erase(&block_group->cache_node,
7301 &info->block_group_cache_tree);
7302 spin_unlock(&info->block_group_cache_lock);
7304 down_write(&block_group->space_info->groups_sem);
7305 list_del(&block_group->list);
7306 up_write(&block_group->space_info->groups_sem);
7308 if (block_group->cached == BTRFS_CACHE_STARTED)
7309 wait_block_group_cache_done(block_group);
7311 btrfs_remove_free_space_cache(block_group);
7313 WARN_ON(atomic_read(&block_group->count) != 1);
7316 spin_lock(&info->block_group_cache_lock);
7318 spin_unlock(&info->block_group_cache_lock);
7320 /* now that all the block groups are freed, go through and
7321 * free all the space_info structs. This is only called during
7322 * the final stages of unmount, and so we know nobody is
7323 * using them. We call synchronize_rcu() once before we start,
7324 * just to be on the safe side.
7328 while(!list_empty(&info->space_info)) {
7329 space_info = list_entry(info->space_info.next,
7330 struct btrfs_space_info,
7333 list_del(&space_info->list);
7339 int btrfs_read_block_groups(struct btrfs_root *root)
7341 struct btrfs_path *path;
7343 struct btrfs_block_group_cache *cache;
7344 struct btrfs_fs_info *info = root->fs_info;
7345 struct btrfs_space_info *space_info;
7346 struct btrfs_key key;
7347 struct btrfs_key found_key;
7348 struct extent_buffer *leaf;
7350 root = info->extent_root;
7353 btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY);
7354 path = btrfs_alloc_path();
7359 ret = find_first_block_group(root, path, &key);
7367 leaf = path->nodes[0];
7368 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
7369 cache = kzalloc(sizeof(*cache), GFP_NOFS);
7375 atomic_set(&cache->count, 1);
7376 spin_lock_init(&cache->lock);
7377 spin_lock_init(&cache->tree_lock);
7378 cache->fs_info = info;
7379 INIT_LIST_HEAD(&cache->list);
7380 INIT_LIST_HEAD(&cache->cluster_list);
7383 * we only want to have 32k of ram per block group for keeping
7384 * track of free space, and if we pass 1/2 of that we want to
7385 * start converting things over to using bitmaps
7387 cache->extents_thresh = ((1024 * 32) / 2) /
7388 sizeof(struct btrfs_free_space);
7390 read_extent_buffer(leaf, &cache->item,
7391 btrfs_item_ptr_offset(leaf, path->slots[0]),
7392 sizeof(cache->item));
7393 memcpy(&cache->key, &found_key, sizeof(found_key));
7395 key.objectid = found_key.objectid + found_key.offset;
7396 btrfs_release_path(root, path);
7397 cache->flags = btrfs_block_group_flags(&cache->item);
7398 cache->sectorsize = root->sectorsize;
7401 * check for two cases, either we are full, and therefore
7402 * don't need to bother with the caching work since we won't
7403 * find any space, or we are empty, and we can just add all
7404 * the space in and be done with it. This saves us _alot_ of
7405 * time, particularly in the full case.
7407 if (found_key.offset == btrfs_block_group_used(&cache->item)) {
7408 exclude_super_stripes(root, cache);
7409 cache->last_byte_to_unpin = (u64)-1;
7410 cache->cached = BTRFS_CACHE_FINISHED;
7411 free_excluded_extents(root, cache);
7412 } else if (btrfs_block_group_used(&cache->item) == 0) {
7413 exclude_super_stripes(root, cache);
7414 cache->last_byte_to_unpin = (u64)-1;
7415 cache->cached = BTRFS_CACHE_FINISHED;
7416 add_new_free_space(cache, root->fs_info,
7418 found_key.objectid +
7420 free_excluded_extents(root, cache);
7423 ret = update_space_info(info, cache->flags, found_key.offset,
7424 btrfs_block_group_used(&cache->item),
7427 cache->space_info = space_info;
7428 spin_lock(&cache->space_info->lock);
7429 cache->space_info->bytes_super += cache->bytes_super;
7430 spin_unlock(&cache->space_info->lock);
7432 down_write(&space_info->groups_sem);
7433 list_add_tail(&cache->list, &space_info->block_groups);
7434 up_write(&space_info->groups_sem);
7436 ret = btrfs_add_block_group_cache(root->fs_info, cache);
7439 set_avail_alloc_bits(root->fs_info, cache->flags);
7440 if (btrfs_chunk_readonly(root, cache->key.objectid))
7441 set_block_group_readonly(cache);
7445 btrfs_free_path(path);
7449 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
7450 struct btrfs_root *root, u64 bytes_used,
7451 u64 type, u64 chunk_objectid, u64 chunk_offset,
7455 struct btrfs_root *extent_root;
7456 struct btrfs_block_group_cache *cache;
7458 extent_root = root->fs_info->extent_root;
7460 root->fs_info->last_trans_log_full_commit = trans->transid;
7462 cache = kzalloc(sizeof(*cache), GFP_NOFS);
7466 cache->key.objectid = chunk_offset;
7467 cache->key.offset = size;
7468 cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
7469 cache->sectorsize = root->sectorsize;
7472 * we only want to have 32k of ram per block group for keeping track
7473 * of free space, and if we pass 1/2 of that we want to start
7474 * converting things over to using bitmaps
7476 cache->extents_thresh = ((1024 * 32) / 2) /
7477 sizeof(struct btrfs_free_space);
7478 atomic_set(&cache->count, 1);
7479 spin_lock_init(&cache->lock);
7480 spin_lock_init(&cache->tree_lock);
7481 INIT_LIST_HEAD(&cache->list);
7482 INIT_LIST_HEAD(&cache->cluster_list);
7484 btrfs_set_block_group_used(&cache->item, bytes_used);
7485 btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
7486 cache->flags = type;
7487 btrfs_set_block_group_flags(&cache->item, type);
7489 cache->last_byte_to_unpin = (u64)-1;
7490 cache->cached = BTRFS_CACHE_FINISHED;
7491 exclude_super_stripes(root, cache);
7493 add_new_free_space(cache, root->fs_info, chunk_offset,
7494 chunk_offset + size);
7496 free_excluded_extents(root, cache);
7498 ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
7499 &cache->space_info);
7502 spin_lock(&cache->space_info->lock);
7503 cache->space_info->bytes_super += cache->bytes_super;
7504 spin_unlock(&cache->space_info->lock);
7506 down_write(&cache->space_info->groups_sem);
7507 list_add_tail(&cache->list, &cache->space_info->block_groups);
7508 up_write(&cache->space_info->groups_sem);
7510 ret = btrfs_add_block_group_cache(root->fs_info, cache);
7513 ret = btrfs_insert_item(trans, extent_root, &cache->key, &cache->item,
7514 sizeof(cache->item));
7517 set_avail_alloc_bits(extent_root->fs_info, type);
7522 int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
7523 struct btrfs_root *root, u64 group_start)
7525 struct btrfs_path *path;
7526 struct btrfs_block_group_cache *block_group;
7527 struct btrfs_free_cluster *cluster;
7528 struct btrfs_key key;
7531 root = root->fs_info->extent_root;
7533 block_group = btrfs_lookup_block_group(root->fs_info, group_start);
7534 BUG_ON(!block_group);
7535 BUG_ON(!block_group->ro);
7537 memcpy(&key, &block_group->key, sizeof(key));
7539 /* make sure this block group isn't part of an allocation cluster */
7540 cluster = &root->fs_info->data_alloc_cluster;
7541 spin_lock(&cluster->refill_lock);
7542 btrfs_return_cluster_to_free_space(block_group, cluster);
7543 spin_unlock(&cluster->refill_lock);
7546 * make sure this block group isn't part of a metadata
7547 * allocation cluster
7549 cluster = &root->fs_info->meta_alloc_cluster;
7550 spin_lock(&cluster->refill_lock);
7551 btrfs_return_cluster_to_free_space(block_group, cluster);
7552 spin_unlock(&cluster->refill_lock);
7554 path = btrfs_alloc_path();
7557 spin_lock(&root->fs_info->block_group_cache_lock);
7558 rb_erase(&block_group->cache_node,
7559 &root->fs_info->block_group_cache_tree);
7560 spin_unlock(&root->fs_info->block_group_cache_lock);
7562 down_write(&block_group->space_info->groups_sem);
7564 * we must use list_del_init so people can check to see if they
7565 * are still on the list after taking the semaphore
7567 list_del_init(&block_group->list);
7568 up_write(&block_group->space_info->groups_sem);
7570 if (block_group->cached == BTRFS_CACHE_STARTED)
7571 wait_block_group_cache_done(block_group);
7573 btrfs_remove_free_space_cache(block_group);
7575 spin_lock(&block_group->space_info->lock);
7576 block_group->space_info->total_bytes -= block_group->key.offset;
7577 block_group->space_info->bytes_readonly -= block_group->key.offset;
7578 spin_unlock(&block_group->space_info->lock);
7580 btrfs_clear_space_info_full(root->fs_info);
7582 btrfs_put_block_group(block_group);
7583 btrfs_put_block_group(block_group);
7585 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
7591 ret = btrfs_del_item(trans, root, path);
7593 btrfs_free_path(path);