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>
25 #include "print-tree.h"
26 #include "transaction.h"
30 #define BLOCK_GROUP_DATA EXTENT_WRITEBACK
31 #define BLOCK_GROUP_METADATA EXTENT_UPTODATE
32 #define BLOCK_GROUP_SYSTEM EXTENT_NEW
34 #define BLOCK_GROUP_DIRTY EXTENT_DIRTY
36 static int finish_current_insert(struct btrfs_trans_handle *trans, struct
37 btrfs_root *extent_root);
38 static int del_pending_extents(struct btrfs_trans_handle *trans, struct
39 btrfs_root *extent_root);
40 static struct btrfs_block_group_cache *
41 __btrfs_find_block_group(struct btrfs_root *root,
42 struct btrfs_block_group_cache *hint,
43 u64 search_start, int data, int owner);
45 void maybe_lock_mutex(struct btrfs_root *root)
47 if (root != root->fs_info->extent_root &&
48 root != root->fs_info->chunk_root &&
49 root != root->fs_info->dev_root) {
50 mutex_lock(&root->fs_info->alloc_mutex);
54 void maybe_unlock_mutex(struct btrfs_root *root)
56 if (root != root->fs_info->extent_root &&
57 root != root->fs_info->chunk_root &&
58 root != root->fs_info->dev_root) {
59 mutex_unlock(&root->fs_info->alloc_mutex);
63 static int cache_block_group(struct btrfs_root *root,
64 struct btrfs_block_group_cache *block_group)
66 struct btrfs_path *path;
69 struct extent_buffer *leaf;
70 struct extent_io_tree *free_space_cache;
80 root = root->fs_info->extent_root;
81 free_space_cache = &root->fs_info->free_space_cache;
83 if (block_group->cached)
86 path = btrfs_alloc_path();
92 * we get into deadlocks with paths held by callers of this function.
93 * since the alloc_mutex is protecting things right now, just
94 * skip the locking here
96 path->skip_locking = 1;
97 first_free = block_group->key.objectid;
98 key.objectid = block_group->key.objectid;
100 btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
101 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
104 ret = btrfs_previous_item(root, path, 0, BTRFS_EXTENT_ITEM_KEY);
108 leaf = path->nodes[0];
109 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
110 if (key.objectid + key.offset > first_free)
111 first_free = key.objectid + key.offset;
114 leaf = path->nodes[0];
115 slot = path->slots[0];
116 if (slot >= btrfs_header_nritems(leaf)) {
117 ret = btrfs_next_leaf(root, path);
126 btrfs_item_key_to_cpu(leaf, &key, slot);
127 if (key.objectid < block_group->key.objectid) {
130 if (key.objectid >= block_group->key.objectid +
131 block_group->key.offset) {
135 if (btrfs_key_type(&key) == BTRFS_EXTENT_ITEM_KEY) {
140 if (key.objectid > last) {
141 hole_size = key.objectid - last;
142 set_extent_dirty(free_space_cache, last,
143 last + hole_size - 1,
146 last = key.objectid + key.offset;
154 if (block_group->key.objectid +
155 block_group->key.offset > last) {
156 hole_size = block_group->key.objectid +
157 block_group->key.offset - last;
158 set_extent_dirty(free_space_cache, last,
159 last + hole_size - 1, GFP_NOFS);
161 block_group->cached = 1;
163 btrfs_free_path(path);
167 struct btrfs_block_group_cache *btrfs_lookup_first_block_group(struct
171 struct extent_io_tree *block_group_cache;
172 struct btrfs_block_group_cache *block_group = NULL;
178 bytenr = max_t(u64, bytenr,
179 BTRFS_SUPER_INFO_OFFSET + BTRFS_SUPER_INFO_SIZE);
180 block_group_cache = &info->block_group_cache;
181 ret = find_first_extent_bit(block_group_cache,
182 bytenr, &start, &end,
183 BLOCK_GROUP_DATA | BLOCK_GROUP_METADATA |
188 ret = get_state_private(block_group_cache, start, &ptr);
192 block_group = (struct btrfs_block_group_cache *)(unsigned long)ptr;
196 struct btrfs_block_group_cache *btrfs_lookup_block_group(struct
200 struct extent_io_tree *block_group_cache;
201 struct btrfs_block_group_cache *block_group = NULL;
207 bytenr = max_t(u64, bytenr,
208 BTRFS_SUPER_INFO_OFFSET + BTRFS_SUPER_INFO_SIZE);
209 block_group_cache = &info->block_group_cache;
210 ret = find_first_extent_bit(block_group_cache,
211 bytenr, &start, &end,
212 BLOCK_GROUP_DATA | BLOCK_GROUP_METADATA |
217 ret = get_state_private(block_group_cache, start, &ptr);
221 block_group = (struct btrfs_block_group_cache *)(unsigned long)ptr;
222 if (block_group->key.objectid <= bytenr && bytenr <
223 block_group->key.objectid + block_group->key.offset)
228 static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits)
230 return (cache->flags & bits) == bits;
233 static int noinline find_search_start(struct btrfs_root *root,
234 struct btrfs_block_group_cache **cache_ret,
235 u64 *start_ret, u64 num, int data)
238 struct btrfs_block_group_cache *cache = *cache_ret;
239 struct extent_io_tree *free_space_cache;
240 struct extent_state *state;
245 u64 search_start = *start_ret;
248 total_fs_bytes = btrfs_super_total_bytes(&root->fs_info->super_copy);
249 free_space_cache = &root->fs_info->free_space_cache;
255 ret = cache_block_group(root, cache);
260 last = max(search_start, cache->key.objectid);
261 if (!block_group_bits(cache, data) || cache->ro)
264 spin_lock_irq(&free_space_cache->lock);
265 state = find_first_extent_bit_state(free_space_cache, last, EXTENT_DIRTY);
270 spin_unlock_irq(&free_space_cache->lock);
274 start = max(last, state->start);
275 last = state->end + 1;
276 if (last - start < num) {
278 state = extent_state_next(state);
279 } while(state && !(state->state & EXTENT_DIRTY));
282 spin_unlock_irq(&free_space_cache->lock);
286 if (start + num > cache->key.objectid + cache->key.offset)
288 if (!block_group_bits(cache, data)) {
289 printk("block group bits don't match %Lu %d\n", cache->flags, data);
295 cache = btrfs_lookup_block_group(root->fs_info, search_start);
297 printk("Unable to find block group for %Lu\n", search_start);
303 last = cache->key.objectid + cache->key.offset;
305 cache = btrfs_lookup_first_block_group(root->fs_info, last);
306 if (!cache || cache->key.objectid >= total_fs_bytes) {
315 if (cache_miss && !cache->cached) {
316 cache_block_group(root, cache);
318 cache = btrfs_lookup_first_block_group(root->fs_info, last);
321 cache = __btrfs_find_block_group(root, cache, last, data, 0);
328 static u64 div_factor(u64 num, int factor)
337 static int block_group_state_bits(u64 flags)
340 if (flags & BTRFS_BLOCK_GROUP_DATA)
341 bits |= BLOCK_GROUP_DATA;
342 if (flags & BTRFS_BLOCK_GROUP_METADATA)
343 bits |= BLOCK_GROUP_METADATA;
344 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
345 bits |= BLOCK_GROUP_SYSTEM;
349 static struct btrfs_block_group_cache *
350 __btrfs_find_block_group(struct btrfs_root *root,
351 struct btrfs_block_group_cache *hint,
352 u64 search_start, int data, int owner)
354 struct btrfs_block_group_cache *cache;
355 struct extent_io_tree *block_group_cache;
356 struct btrfs_block_group_cache *found_group = NULL;
357 struct btrfs_fs_info *info = root->fs_info;
370 block_group_cache = &info->block_group_cache;
372 if (data & BTRFS_BLOCK_GROUP_METADATA)
375 bit = block_group_state_bits(data);
378 struct btrfs_block_group_cache *shint;
379 shint = btrfs_lookup_first_block_group(info, search_start);
380 if (shint && block_group_bits(shint, data) && !shint->ro) {
381 used = btrfs_block_group_used(&shint->item);
382 if (used + shint->pinned <
383 div_factor(shint->key.offset, factor)) {
388 if (hint && !hint->ro && block_group_bits(hint, data)) {
389 used = btrfs_block_group_used(&hint->item);
390 if (used + hint->pinned <
391 div_factor(hint->key.offset, factor)) {
394 last = hint->key.objectid + hint->key.offset;
397 last = max(hint->key.objectid, search_start);
403 ret = find_first_extent_bit(block_group_cache, last,
408 ret = get_state_private(block_group_cache, start, &ptr);
414 cache = (struct btrfs_block_group_cache *)(unsigned long)ptr;
415 last = cache->key.objectid + cache->key.offset;
416 used = btrfs_block_group_used(&cache->item);
418 if (!cache->ro && block_group_bits(cache, data)) {
419 free_check = div_factor(cache->key.offset, factor);
420 if (used + cache->pinned < free_check) {
432 if (!full_search && factor < 10) {
442 struct btrfs_block_group_cache *btrfs_find_block_group(struct btrfs_root *root,
443 struct btrfs_block_group_cache
444 *hint, u64 search_start,
448 struct btrfs_block_group_cache *ret;
449 mutex_lock(&root->fs_info->alloc_mutex);
450 ret = __btrfs_find_block_group(root, hint, search_start, data, owner);
451 mutex_unlock(&root->fs_info->alloc_mutex);
454 static u64 hash_extent_ref(u64 root_objectid, u64 ref_generation,
455 u64 owner, u64 owner_offset)
457 u32 high_crc = ~(u32)0;
458 u32 low_crc = ~(u32)0;
460 lenum = cpu_to_le64(root_objectid);
461 high_crc = btrfs_crc32c(high_crc, &lenum, sizeof(lenum));
462 lenum = cpu_to_le64(ref_generation);
463 low_crc = btrfs_crc32c(low_crc, &lenum, sizeof(lenum));
464 if (owner >= BTRFS_FIRST_FREE_OBJECTID) {
465 lenum = cpu_to_le64(owner);
466 low_crc = btrfs_crc32c(low_crc, &lenum, sizeof(lenum));
467 lenum = cpu_to_le64(owner_offset);
468 low_crc = btrfs_crc32c(low_crc, &lenum, sizeof(lenum));
470 return ((u64)high_crc << 32) | (u64)low_crc;
473 static int match_extent_ref(struct extent_buffer *leaf,
474 struct btrfs_extent_ref *disk_ref,
475 struct btrfs_extent_ref *cpu_ref)
480 if (cpu_ref->objectid)
481 len = sizeof(*cpu_ref);
483 len = 2 * sizeof(u64);
484 ret = memcmp_extent_buffer(leaf, cpu_ref, (unsigned long)disk_ref,
489 static int noinline lookup_extent_backref(struct btrfs_trans_handle *trans,
490 struct btrfs_root *root,
491 struct btrfs_path *path, u64 bytenr,
493 u64 ref_generation, u64 owner,
494 u64 owner_offset, int del)
497 struct btrfs_key key;
498 struct btrfs_key found_key;
499 struct btrfs_extent_ref ref;
500 struct extent_buffer *leaf;
501 struct btrfs_extent_ref *disk_ref;
505 btrfs_set_stack_ref_root(&ref, root_objectid);
506 btrfs_set_stack_ref_generation(&ref, ref_generation);
507 btrfs_set_stack_ref_objectid(&ref, owner);
508 btrfs_set_stack_ref_offset(&ref, owner_offset);
510 hash = hash_extent_ref(root_objectid, ref_generation, owner,
513 key.objectid = bytenr;
514 key.type = BTRFS_EXTENT_REF_KEY;
517 ret = btrfs_search_slot(trans, root, &key, path,
521 leaf = path->nodes[0];
523 u32 nritems = btrfs_header_nritems(leaf);
524 if (path->slots[0] >= nritems) {
525 ret2 = btrfs_next_leaf(root, path);
528 leaf = path->nodes[0];
530 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
531 if (found_key.objectid != bytenr ||
532 found_key.type != BTRFS_EXTENT_REF_KEY)
534 key.offset = found_key.offset;
536 btrfs_release_path(root, path);
540 disk_ref = btrfs_item_ptr(path->nodes[0],
542 struct btrfs_extent_ref);
543 if (match_extent_ref(path->nodes[0], disk_ref, &ref)) {
547 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
548 key.offset = found_key.offset + 1;
549 btrfs_release_path(root, path);
556 * Back reference rules. Back refs have three main goals:
558 * 1) differentiate between all holders of references to an extent so that
559 * when a reference is dropped we can make sure it was a valid reference
560 * before freeing the extent.
562 * 2) Provide enough information to quickly find the holders of an extent
563 * if we notice a given block is corrupted or bad.
565 * 3) Make it easy to migrate blocks for FS shrinking or storage pool
566 * maintenance. This is actually the same as #2, but with a slightly
567 * different use case.
569 * File extents can be referenced by:
571 * - multiple snapshots, subvolumes, or different generations in one subvol
572 * - different files inside a single subvolume (in theory, not implemented yet)
573 * - different offsets inside a file (bookend extents in file.c)
575 * The extent ref structure has fields for:
577 * - Objectid of the subvolume root
578 * - Generation number of the tree holding the reference
579 * - objectid of the file holding the reference
580 * - offset in the file corresponding to the key holding the reference
582 * When a file extent is allocated the fields are filled in:
583 * (root_key.objectid, trans->transid, inode objectid, offset in file)
585 * When a leaf is cow'd new references are added for every file extent found
586 * in the leaf. It looks the same as the create case, but trans->transid
587 * will be different when the block is cow'd.
589 * (root_key.objectid, trans->transid, inode objectid, offset in file)
591 * When a file extent is removed either during snapshot deletion or file
592 * truncation, the corresponding back reference is found
595 * (btrfs_header_owner(leaf), btrfs_header_generation(leaf),
596 * inode objectid, offset in file)
598 * Btree extents can be referenced by:
600 * - Different subvolumes
601 * - Different generations of the same subvolume
603 * Storing sufficient information for a full reverse mapping of a btree
604 * block would require storing the lowest key of the block in the backref,
605 * and it would require updating that lowest key either before write out or
606 * every time it changed. Instead, the objectid of the lowest key is stored
607 * along with the level of the tree block. This provides a hint
608 * about where in the btree the block can be found. Searches through the
609 * btree only need to look for a pointer to that block, so they stop one
610 * level higher than the level recorded in the backref.
612 * Some btrees do not do reference counting on their extents. These
613 * include the extent tree and the tree of tree roots. Backrefs for these
614 * trees always have a generation of zero.
616 * When a tree block is created, back references are inserted:
618 * (root->root_key.objectid, trans->transid or zero, level, lowest_key_objectid)
620 * When a tree block is cow'd in a reference counted root,
621 * new back references are added for all the blocks it points to.
622 * These are of the form (trans->transid will have increased since creation):
624 * (root->root_key.objectid, trans->transid, level, lowest_key_objectid)
626 * Because the lowest_key_objectid and the level are just hints
627 * they are not used when backrefs are deleted. When a backref is deleted:
629 * if backref was for a tree root:
630 * root_objectid = root->root_key.objectid
632 * root_objectid = btrfs_header_owner(parent)
634 * (root_objectid, btrfs_header_generation(parent) or zero, 0, 0)
636 * Back Reference Key hashing:
638 * Back references have four fields, each 64 bits long. Unfortunately,
639 * This is hashed into a single 64 bit number and placed into the key offset.
640 * The key objectid corresponds to the first byte in the extent, and the
641 * key type is set to BTRFS_EXTENT_REF_KEY
643 int btrfs_insert_extent_backref(struct btrfs_trans_handle *trans,
644 struct btrfs_root *root,
645 struct btrfs_path *path, u64 bytenr,
646 u64 root_objectid, u64 ref_generation,
647 u64 owner, u64 owner_offset)
650 struct btrfs_key key;
651 struct btrfs_extent_ref ref;
652 struct btrfs_extent_ref *disk_ref;
655 btrfs_set_stack_ref_root(&ref, root_objectid);
656 btrfs_set_stack_ref_generation(&ref, ref_generation);
657 btrfs_set_stack_ref_objectid(&ref, owner);
658 btrfs_set_stack_ref_offset(&ref, owner_offset);
660 hash = hash_extent_ref(root_objectid, ref_generation, owner,
663 key.objectid = bytenr;
664 key.type = BTRFS_EXTENT_REF_KEY;
666 ret = btrfs_insert_empty_item(trans, root, path, &key, sizeof(ref));
667 while (ret == -EEXIST) {
668 disk_ref = btrfs_item_ptr(path->nodes[0], path->slots[0],
669 struct btrfs_extent_ref);
670 if (match_extent_ref(path->nodes[0], disk_ref, &ref))
673 btrfs_release_path(root, path);
674 ret = btrfs_insert_empty_item(trans, root, path, &key,
679 disk_ref = btrfs_item_ptr(path->nodes[0], path->slots[0],
680 struct btrfs_extent_ref);
681 write_extent_buffer(path->nodes[0], &ref, (unsigned long)disk_ref,
683 btrfs_mark_buffer_dirty(path->nodes[0]);
685 btrfs_release_path(root, path);
689 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
690 struct btrfs_root *root,
691 u64 bytenr, u64 num_bytes,
692 u64 root_objectid, u64 ref_generation,
693 u64 owner, u64 owner_offset)
695 struct btrfs_path *path;
697 struct btrfs_key key;
698 struct extent_buffer *l;
699 struct btrfs_extent_item *item;
702 WARN_ON(num_bytes < root->sectorsize);
703 path = btrfs_alloc_path();
708 key.objectid = bytenr;
709 btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
710 key.offset = num_bytes;
711 ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key, path,
720 item = btrfs_item_ptr(l, path->slots[0], struct btrfs_extent_item);
721 refs = btrfs_extent_refs(l, item);
722 btrfs_set_extent_refs(l, item, refs + 1);
723 btrfs_mark_buffer_dirty(path->nodes[0]);
725 btrfs_release_path(root->fs_info->extent_root, path);
728 ret = btrfs_insert_extent_backref(trans, root->fs_info->extent_root,
729 path, bytenr, root_objectid,
730 ref_generation, owner, owner_offset);
732 finish_current_insert(trans, root->fs_info->extent_root);
733 del_pending_extents(trans, root->fs_info->extent_root);
735 btrfs_free_path(path);
739 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
740 struct btrfs_root *root,
741 u64 bytenr, u64 num_bytes,
742 u64 root_objectid, u64 ref_generation,
743 u64 owner, u64 owner_offset)
747 mutex_lock(&root->fs_info->alloc_mutex);
748 ret = __btrfs_inc_extent_ref(trans, root, bytenr, num_bytes,
749 root_objectid, ref_generation,
750 owner, owner_offset);
751 mutex_unlock(&root->fs_info->alloc_mutex);
755 int btrfs_extent_post_op(struct btrfs_trans_handle *trans,
756 struct btrfs_root *root)
758 finish_current_insert(trans, root->fs_info->extent_root);
759 del_pending_extents(trans, root->fs_info->extent_root);
763 static int lookup_extent_ref(struct btrfs_trans_handle *trans,
764 struct btrfs_root *root, u64 bytenr,
765 u64 num_bytes, u32 *refs)
767 struct btrfs_path *path;
769 struct btrfs_key key;
770 struct extent_buffer *l;
771 struct btrfs_extent_item *item;
773 WARN_ON(num_bytes < root->sectorsize);
774 path = btrfs_alloc_path();
776 key.objectid = bytenr;
777 key.offset = num_bytes;
778 btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
779 ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key, path,
784 btrfs_print_leaf(root, path->nodes[0]);
785 printk("failed to find block number %Lu\n", bytenr);
789 item = btrfs_item_ptr(l, path->slots[0], struct btrfs_extent_item);
790 *refs = btrfs_extent_refs(l, item);
792 btrfs_free_path(path);
796 u32 btrfs_count_snapshots_in_path(struct btrfs_root *root,
797 struct btrfs_path *count_path,
801 struct btrfs_root *extent_root = root->fs_info->extent_root;
802 struct btrfs_path *path;
806 u64 root_objectid = root->root_key.objectid;
812 struct btrfs_key key;
813 struct btrfs_key found_key;
814 struct extent_buffer *l;
815 struct btrfs_extent_item *item;
816 struct btrfs_extent_ref *ref_item;
819 /* FIXME, needs locking */
822 mutex_lock(&root->fs_info->alloc_mutex);
823 path = btrfs_alloc_path();
826 bytenr = first_extent;
828 bytenr = count_path->nodes[level]->start;
831 key.objectid = bytenr;
834 btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
835 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
841 btrfs_item_key_to_cpu(l, &found_key, path->slots[0]);
843 if (found_key.objectid != bytenr ||
844 found_key.type != BTRFS_EXTENT_ITEM_KEY) {
848 item = btrfs_item_ptr(l, path->slots[0], struct btrfs_extent_item);
849 extent_refs = btrfs_extent_refs(l, item);
852 nritems = btrfs_header_nritems(l);
853 if (path->slots[0] >= nritems) {
854 ret = btrfs_next_leaf(extent_root, path);
859 btrfs_item_key_to_cpu(l, &found_key, path->slots[0]);
860 if (found_key.objectid != bytenr)
863 if (found_key.type != BTRFS_EXTENT_REF_KEY) {
869 ref_item = btrfs_item_ptr(l, path->slots[0],
870 struct btrfs_extent_ref);
871 found_objectid = btrfs_ref_root(l, ref_item);
873 if (found_objectid != root_objectid) {
878 found_owner = btrfs_ref_objectid(l, ref_item);
879 if (found_owner != expected_owner) {
884 * nasty. we don't count a reference held by
885 * the running transaction. This allows nodatacow
886 * to avoid cow most of the time
888 if (found_owner >= BTRFS_FIRST_FREE_OBJECTID &&
889 btrfs_ref_generation(l, ref_item) ==
890 root->fs_info->generation) {
898 * if there is more than one reference against a data extent,
899 * we have to assume the other ref is another snapshot
901 if (level == -1 && extent_refs > 1) {
905 if (cur_count == 0) {
909 if (level >= 0 && root->node == count_path->nodes[level])
912 btrfs_release_path(root, path);
916 btrfs_free_path(path);
917 mutex_unlock(&root->fs_info->alloc_mutex);
921 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
922 struct extent_buffer *buf)
926 struct btrfs_key key;
927 struct btrfs_file_extent_item *fi;
936 mutex_lock(&root->fs_info->alloc_mutex);
937 level = btrfs_header_level(buf);
938 nritems = btrfs_header_nritems(buf);
939 for (i = 0; i < nritems; i++) {
942 btrfs_item_key_to_cpu(buf, &key, i);
943 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
945 fi = btrfs_item_ptr(buf, i,
946 struct btrfs_file_extent_item);
947 if (btrfs_file_extent_type(buf, fi) ==
948 BTRFS_FILE_EXTENT_INLINE)
950 disk_bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
951 if (disk_bytenr == 0)
953 ret = __btrfs_inc_extent_ref(trans, root, disk_bytenr,
954 btrfs_file_extent_disk_num_bytes(buf, fi),
955 root->root_key.objectid, trans->transid,
956 key.objectid, key.offset);
962 bytenr = btrfs_node_blockptr(buf, i);
963 btrfs_node_key_to_cpu(buf, &key, i);
964 ret = __btrfs_inc_extent_ref(trans, root, bytenr,
965 btrfs_level_size(root, level - 1),
966 root->root_key.objectid,
968 level - 1, key.objectid);
975 mutex_unlock(&root->fs_info->alloc_mutex);
980 for (i =0; i < faili; i++) {
983 btrfs_item_key_to_cpu(buf, &key, i);
984 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
986 fi = btrfs_item_ptr(buf, i,
987 struct btrfs_file_extent_item);
988 if (btrfs_file_extent_type(buf, fi) ==
989 BTRFS_FILE_EXTENT_INLINE)
991 disk_bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
992 if (disk_bytenr == 0)
994 err = btrfs_free_extent(trans, root, disk_bytenr,
995 btrfs_file_extent_disk_num_bytes(buf,
999 bytenr = btrfs_node_blockptr(buf, i);
1000 err = btrfs_free_extent(trans, root, bytenr,
1001 btrfs_level_size(root, level - 1), 0);
1006 mutex_unlock(&root->fs_info->alloc_mutex);
1010 static int write_one_cache_group(struct btrfs_trans_handle *trans,
1011 struct btrfs_root *root,
1012 struct btrfs_path *path,
1013 struct btrfs_block_group_cache *cache)
1017 struct btrfs_root *extent_root = root->fs_info->extent_root;
1019 struct extent_buffer *leaf;
1021 ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
1026 leaf = path->nodes[0];
1027 bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
1028 write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
1029 btrfs_mark_buffer_dirty(leaf);
1030 btrfs_release_path(extent_root, path);
1032 finish_current_insert(trans, extent_root);
1033 pending_ret = del_pending_extents(trans, extent_root);
1042 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
1043 struct btrfs_root *root)
1045 struct extent_io_tree *block_group_cache;
1046 struct btrfs_block_group_cache *cache;
1050 struct btrfs_path *path;
1056 block_group_cache = &root->fs_info->block_group_cache;
1057 path = btrfs_alloc_path();
1061 mutex_lock(&root->fs_info->alloc_mutex);
1063 ret = find_first_extent_bit(block_group_cache, last,
1064 &start, &end, BLOCK_GROUP_DIRTY);
1069 ret = get_state_private(block_group_cache, start, &ptr);
1072 cache = (struct btrfs_block_group_cache *)(unsigned long)ptr;
1073 err = write_one_cache_group(trans, root,
1076 * if we fail to write the cache group, we want
1077 * to keep it marked dirty in hopes that a later
1084 clear_extent_bits(block_group_cache, start, end,
1085 BLOCK_GROUP_DIRTY, GFP_NOFS);
1087 btrfs_free_path(path);
1088 mutex_unlock(&root->fs_info->alloc_mutex);
1092 static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info,
1095 struct list_head *head = &info->space_info;
1096 struct list_head *cur;
1097 struct btrfs_space_info *found;
1098 list_for_each(cur, head) {
1099 found = list_entry(cur, struct btrfs_space_info, list);
1100 if (found->flags == flags)
1107 static int update_space_info(struct btrfs_fs_info *info, u64 flags,
1108 u64 total_bytes, u64 bytes_used,
1109 struct btrfs_space_info **space_info)
1111 struct btrfs_space_info *found;
1113 found = __find_space_info(info, flags);
1115 found->total_bytes += total_bytes;
1116 found->bytes_used += bytes_used;
1118 WARN_ON(found->total_bytes < found->bytes_used);
1119 *space_info = found;
1122 found = kmalloc(sizeof(*found), GFP_NOFS);
1126 list_add(&found->list, &info->space_info);
1127 found->flags = flags;
1128 found->total_bytes = total_bytes;
1129 found->bytes_used = bytes_used;
1130 found->bytes_pinned = 0;
1132 found->force_alloc = 0;
1133 *space_info = found;
1137 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
1139 u64 extra_flags = flags & (BTRFS_BLOCK_GROUP_RAID0 |
1140 BTRFS_BLOCK_GROUP_RAID1 |
1141 BTRFS_BLOCK_GROUP_RAID10 |
1142 BTRFS_BLOCK_GROUP_DUP);
1144 if (flags & BTRFS_BLOCK_GROUP_DATA)
1145 fs_info->avail_data_alloc_bits |= extra_flags;
1146 if (flags & BTRFS_BLOCK_GROUP_METADATA)
1147 fs_info->avail_metadata_alloc_bits |= extra_flags;
1148 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
1149 fs_info->avail_system_alloc_bits |= extra_flags;
1153 static u64 reduce_alloc_profile(struct btrfs_root *root, u64 flags)
1155 u64 num_devices = root->fs_info->fs_devices->num_devices;
1157 if (num_devices == 1)
1158 flags &= ~(BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID0);
1159 if (num_devices < 4)
1160 flags &= ~BTRFS_BLOCK_GROUP_RAID10;
1162 if ((flags & BTRFS_BLOCK_GROUP_DUP) &&
1163 (flags & (BTRFS_BLOCK_GROUP_RAID1 |
1164 BTRFS_BLOCK_GROUP_RAID10))) {
1165 flags &= ~BTRFS_BLOCK_GROUP_DUP;
1168 if ((flags & BTRFS_BLOCK_GROUP_RAID1) &&
1169 (flags & BTRFS_BLOCK_GROUP_RAID10)) {
1170 flags &= ~BTRFS_BLOCK_GROUP_RAID1;
1173 if ((flags & BTRFS_BLOCK_GROUP_RAID0) &&
1174 ((flags & BTRFS_BLOCK_GROUP_RAID1) |
1175 (flags & BTRFS_BLOCK_GROUP_RAID10) |
1176 (flags & BTRFS_BLOCK_GROUP_DUP)))
1177 flags &= ~BTRFS_BLOCK_GROUP_RAID0;
1181 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
1182 struct btrfs_root *extent_root, u64 alloc_bytes,
1183 u64 flags, int force)
1185 struct btrfs_space_info *space_info;
1191 flags = reduce_alloc_profile(extent_root, flags);
1193 space_info = __find_space_info(extent_root->fs_info, flags);
1195 ret = update_space_info(extent_root->fs_info, flags,
1199 BUG_ON(!space_info);
1201 if (space_info->force_alloc) {
1203 space_info->force_alloc = 0;
1205 if (space_info->full)
1208 thresh = div_factor(space_info->total_bytes, 6);
1210 (space_info->bytes_used + space_info->bytes_pinned + alloc_bytes) <
1214 mutex_lock(&extent_root->fs_info->chunk_mutex);
1215 ret = btrfs_alloc_chunk(trans, extent_root, &start, &num_bytes, flags);
1216 if (ret == -ENOSPC) {
1217 printk("space info full %Lu\n", flags);
1218 space_info->full = 1;
1223 ret = btrfs_make_block_group(trans, extent_root, 0, flags,
1224 BTRFS_FIRST_CHUNK_TREE_OBJECTID, start, num_bytes);
1227 mutex_unlock(&extent_root->fs_info->chunk_mutex);
1232 static int update_block_group(struct btrfs_trans_handle *trans,
1233 struct btrfs_root *root,
1234 u64 bytenr, u64 num_bytes, int alloc,
1237 struct btrfs_block_group_cache *cache;
1238 struct btrfs_fs_info *info = root->fs_info;
1239 u64 total = num_bytes;
1246 cache = btrfs_lookup_block_group(info, bytenr);
1250 byte_in_group = bytenr - cache->key.objectid;
1251 WARN_ON(byte_in_group > cache->key.offset);
1252 start = cache->key.objectid;
1253 end = start + cache->key.offset - 1;
1254 set_extent_bits(&info->block_group_cache, start, end,
1255 BLOCK_GROUP_DIRTY, GFP_NOFS);
1257 old_val = btrfs_block_group_used(&cache->item);
1258 num_bytes = min(total, cache->key.offset - byte_in_group);
1260 old_val += num_bytes;
1261 cache->space_info->bytes_used += num_bytes;
1263 old_val -= num_bytes;
1264 cache->space_info->bytes_used -= num_bytes;
1266 set_extent_dirty(&info->free_space_cache,
1267 bytenr, bytenr + num_bytes - 1,
1271 btrfs_set_block_group_used(&cache->item, old_val);
1273 bytenr += num_bytes;
1278 static u64 first_logical_byte(struct btrfs_root *root, u64 search_start)
1283 ret = find_first_extent_bit(&root->fs_info->block_group_cache,
1284 search_start, &start, &end,
1285 BLOCK_GROUP_DATA | BLOCK_GROUP_METADATA |
1286 BLOCK_GROUP_SYSTEM);
1293 static int update_pinned_extents(struct btrfs_root *root,
1294 u64 bytenr, u64 num, int pin)
1297 struct btrfs_block_group_cache *cache;
1298 struct btrfs_fs_info *fs_info = root->fs_info;
1301 set_extent_dirty(&fs_info->pinned_extents,
1302 bytenr, bytenr + num - 1, GFP_NOFS);
1304 clear_extent_dirty(&fs_info->pinned_extents,
1305 bytenr, bytenr + num - 1, GFP_NOFS);
1308 cache = btrfs_lookup_block_group(fs_info, bytenr);
1310 u64 first = first_logical_byte(root, bytenr);
1311 WARN_ON(first < bytenr);
1312 len = min(first - bytenr, num);
1314 len = min(num, cache->key.offset -
1315 (bytenr - cache->key.objectid));
1319 cache->pinned += len;
1320 cache->space_info->bytes_pinned += len;
1322 fs_info->total_pinned += len;
1325 cache->pinned -= len;
1326 cache->space_info->bytes_pinned -= len;
1328 fs_info->total_pinned -= len;
1336 int btrfs_copy_pinned(struct btrfs_root *root, struct extent_io_tree *copy)
1341 struct extent_io_tree *pinned_extents = &root->fs_info->pinned_extents;
1345 ret = find_first_extent_bit(pinned_extents, last,
1346 &start, &end, EXTENT_DIRTY);
1349 set_extent_dirty(copy, start, end, GFP_NOFS);
1355 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
1356 struct btrfs_root *root,
1357 struct extent_io_tree *unpin)
1362 struct extent_io_tree *free_space_cache;
1363 free_space_cache = &root->fs_info->free_space_cache;
1365 mutex_lock(&root->fs_info->alloc_mutex);
1367 ret = find_first_extent_bit(unpin, 0, &start, &end,
1371 update_pinned_extents(root, start, end + 1 - start, 0);
1372 clear_extent_dirty(unpin, start, end, GFP_NOFS);
1373 set_extent_dirty(free_space_cache, start, end, GFP_NOFS);
1375 mutex_unlock(&root->fs_info->alloc_mutex);
1379 static int finish_current_insert(struct btrfs_trans_handle *trans,
1380 struct btrfs_root *extent_root)
1384 struct btrfs_fs_info *info = extent_root->fs_info;
1385 struct extent_buffer *eb;
1386 struct btrfs_path *path;
1387 struct btrfs_key ins;
1388 struct btrfs_disk_key first;
1389 struct btrfs_extent_item extent_item;
1394 btrfs_set_stack_extent_refs(&extent_item, 1);
1395 btrfs_set_key_type(&ins, BTRFS_EXTENT_ITEM_KEY);
1396 path = btrfs_alloc_path();
1399 ret = find_first_extent_bit(&info->extent_ins, 0, &start,
1400 &end, EXTENT_LOCKED);
1404 ins.objectid = start;
1405 ins.offset = end + 1 - start;
1406 err = btrfs_insert_item(trans, extent_root, &ins,
1407 &extent_item, sizeof(extent_item));
1408 clear_extent_bits(&info->extent_ins, start, end, EXTENT_LOCKED,
1410 eb = read_tree_block(extent_root, ins.objectid, ins.offset,
1412 btrfs_tree_lock(eb);
1413 level = btrfs_header_level(eb);
1415 btrfs_item_key(eb, &first, 0);
1417 btrfs_node_key(eb, &first, 0);
1419 btrfs_tree_unlock(eb);
1420 free_extent_buffer(eb);
1422 * the first key is just a hint, so the race we've created
1423 * against reading it is fine
1425 err = btrfs_insert_extent_backref(trans, extent_root, path,
1426 start, extent_root->root_key.objectid,
1428 btrfs_disk_key_objectid(&first));
1431 btrfs_free_path(path);
1435 static int pin_down_bytes(struct btrfs_root *root, u64 bytenr, u32 num_bytes,
1441 struct extent_buffer *buf;
1442 buf = btrfs_find_tree_block(root, bytenr, num_bytes);
1444 if (!btrfs_try_tree_lock(buf) &&
1445 btrfs_buffer_uptodate(buf, 0)) {
1447 root->fs_info->running_transaction->transid;
1448 u64 header_transid =
1449 btrfs_header_generation(buf);
1450 if (header_transid == transid &&
1451 !btrfs_header_flag(buf,
1452 BTRFS_HEADER_FLAG_WRITTEN)) {
1453 clean_tree_block(NULL, root, buf);
1454 btrfs_tree_unlock(buf);
1455 free_extent_buffer(buf);
1458 btrfs_tree_unlock(buf);
1460 free_extent_buffer(buf);
1462 update_pinned_extents(root, bytenr, num_bytes, 1);
1464 set_extent_bits(&root->fs_info->pending_del,
1465 bytenr, bytenr + num_bytes - 1,
1466 EXTENT_LOCKED, GFP_NOFS);
1473 * remove an extent from the root, returns 0 on success
1475 static int __free_extent(struct btrfs_trans_handle *trans, struct btrfs_root
1476 *root, u64 bytenr, u64 num_bytes,
1477 u64 root_objectid, u64 ref_generation,
1478 u64 owner_objectid, u64 owner_offset, int pin,
1481 struct btrfs_path *path;
1482 struct btrfs_key key;
1483 struct btrfs_fs_info *info = root->fs_info;
1484 struct btrfs_root *extent_root = info->extent_root;
1485 struct extent_buffer *leaf;
1487 int extent_slot = 0;
1488 int found_extent = 0;
1490 struct btrfs_extent_item *ei;
1493 key.objectid = bytenr;
1494 btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
1495 key.offset = num_bytes;
1496 path = btrfs_alloc_path();
1501 ret = lookup_extent_backref(trans, extent_root, path,
1502 bytenr, root_objectid,
1504 owner_objectid, owner_offset, 1);
1506 struct btrfs_key found_key;
1507 extent_slot = path->slots[0];
1508 while(extent_slot > 0) {
1510 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1512 if (found_key.objectid != bytenr)
1514 if (found_key.type == BTRFS_EXTENT_ITEM_KEY &&
1515 found_key.offset == num_bytes) {
1519 if (path->slots[0] - extent_slot > 5)
1523 ret = btrfs_del_item(trans, extent_root, path);
1525 btrfs_print_leaf(extent_root, path->nodes[0]);
1527 printk("Unable to find ref byte nr %Lu root %Lu "
1528 " gen %Lu owner %Lu offset %Lu\n", bytenr,
1529 root_objectid, ref_generation, owner_objectid,
1532 if (!found_extent) {
1533 btrfs_release_path(extent_root, path);
1534 ret = btrfs_search_slot(trans, extent_root, &key, path, -1, 1);
1538 extent_slot = path->slots[0];
1541 leaf = path->nodes[0];
1542 ei = btrfs_item_ptr(leaf, extent_slot,
1543 struct btrfs_extent_item);
1544 refs = btrfs_extent_refs(leaf, ei);
1547 btrfs_set_extent_refs(leaf, ei, refs);
1549 btrfs_mark_buffer_dirty(leaf);
1551 if (refs == 0 && found_extent && path->slots[0] == extent_slot + 1) {
1552 /* if the back ref and the extent are next to each other
1553 * they get deleted below in one shot
1555 path->slots[0] = extent_slot;
1557 } else if (found_extent) {
1558 /* otherwise delete the extent back ref */
1559 ret = btrfs_del_item(trans, extent_root, path);
1561 /* if refs are 0, we need to setup the path for deletion */
1563 btrfs_release_path(extent_root, path);
1564 ret = btrfs_search_slot(trans, extent_root, &key, path,
1577 ret = pin_down_bytes(root, bytenr, num_bytes, 0);
1583 /* block accounting for super block */
1584 spin_lock_irq(&info->delalloc_lock);
1585 super_used = btrfs_super_bytes_used(&info->super_copy);
1586 btrfs_set_super_bytes_used(&info->super_copy,
1587 super_used - num_bytes);
1588 spin_unlock_irq(&info->delalloc_lock);
1590 /* block accounting for root item */
1591 root_used = btrfs_root_used(&root->root_item);
1592 btrfs_set_root_used(&root->root_item,
1593 root_used - num_bytes);
1594 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
1599 ret = update_block_group(trans, root, bytenr, num_bytes, 0,
1603 btrfs_free_path(path);
1604 finish_current_insert(trans, extent_root);
1609 * find all the blocks marked as pending in the radix tree and remove
1610 * them from the extent map
1612 static int del_pending_extents(struct btrfs_trans_handle *trans, struct
1613 btrfs_root *extent_root)
1619 struct extent_io_tree *pending_del;
1620 struct extent_io_tree *pinned_extents;
1622 pending_del = &extent_root->fs_info->pending_del;
1623 pinned_extents = &extent_root->fs_info->pinned_extents;
1626 ret = find_first_extent_bit(pending_del, 0, &start, &end,
1630 update_pinned_extents(extent_root, start, end + 1 - start, 1);
1631 clear_extent_bits(pending_del, start, end, EXTENT_LOCKED,
1633 ret = __free_extent(trans, extent_root,
1634 start, end + 1 - start,
1635 extent_root->root_key.objectid,
1644 * remove an extent from the root, returns 0 on success
1646 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
1647 struct btrfs_root *root, u64 bytenr,
1648 u64 num_bytes, u64 root_objectid,
1649 u64 ref_generation, u64 owner_objectid,
1650 u64 owner_offset, int pin)
1652 struct btrfs_root *extent_root = root->fs_info->extent_root;
1656 WARN_ON(num_bytes < root->sectorsize);
1657 if (!root->ref_cows)
1660 if (root == extent_root) {
1661 pin_down_bytes(root, bytenr, num_bytes, 1);
1664 ret = __free_extent(trans, root, bytenr, num_bytes, root_objectid,
1665 ref_generation, owner_objectid, owner_offset,
1667 pending_ret = del_pending_extents(trans, root->fs_info->extent_root);
1668 return ret ? ret : pending_ret;
1671 int btrfs_free_extent(struct btrfs_trans_handle *trans,
1672 struct btrfs_root *root, u64 bytenr,
1673 u64 num_bytes, u64 root_objectid,
1674 u64 ref_generation, u64 owner_objectid,
1675 u64 owner_offset, int pin)
1679 maybe_lock_mutex(root);
1680 ret = __btrfs_free_extent(trans, root, bytenr, num_bytes,
1681 root_objectid, ref_generation,
1682 owner_objectid, owner_offset, pin);
1683 maybe_unlock_mutex(root);
1687 static u64 stripe_align(struct btrfs_root *root, u64 val)
1689 u64 mask = ((u64)root->stripesize - 1);
1690 u64 ret = (val + mask) & ~mask;
1695 * walks the btree of allocated extents and find a hole of a given size.
1696 * The key ins is changed to record the hole:
1697 * ins->objectid == block start
1698 * ins->flags = BTRFS_EXTENT_ITEM_KEY
1699 * ins->offset == number of blocks
1700 * Any available blocks before search_start are skipped.
1702 static int noinline find_free_extent(struct btrfs_trans_handle *trans,
1703 struct btrfs_root *orig_root,
1704 u64 num_bytes, u64 empty_size,
1705 u64 search_start, u64 search_end,
1706 u64 hint_byte, struct btrfs_key *ins,
1707 u64 exclude_start, u64 exclude_nr,
1711 u64 orig_search_start;
1712 struct btrfs_root * root = orig_root->fs_info->extent_root;
1713 struct btrfs_fs_info *info = root->fs_info;
1714 u64 total_needed = num_bytes;
1715 u64 *last_ptr = NULL;
1716 struct btrfs_block_group_cache *block_group;
1719 int chunk_alloc_done = 0;
1720 int empty_cluster = 2 * 1024 * 1024;
1721 int allowed_chunk_alloc = 0;
1723 WARN_ON(num_bytes < root->sectorsize);
1724 btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY);
1726 if (orig_root->ref_cows || empty_size)
1727 allowed_chunk_alloc = 1;
1729 if (data & BTRFS_BLOCK_GROUP_METADATA) {
1730 last_ptr = &root->fs_info->last_alloc;
1731 empty_cluster = 256 * 1024;
1734 if ((data & BTRFS_BLOCK_GROUP_DATA) && btrfs_test_opt(root, SSD)) {
1735 last_ptr = &root->fs_info->last_data_alloc;
1740 hint_byte = *last_ptr;
1742 empty_size += empty_cluster;
1746 search_start = max(search_start, first_logical_byte(root, 0));
1747 orig_search_start = search_start;
1749 if (search_end == (u64)-1)
1750 search_end = btrfs_super_total_bytes(&info->super_copy);
1753 block_group = btrfs_lookup_first_block_group(info, hint_byte);
1755 hint_byte = search_start;
1756 block_group = __btrfs_find_block_group(root, block_group,
1757 hint_byte, data, 1);
1758 if (last_ptr && *last_ptr == 0 && block_group)
1759 hint_byte = block_group->key.objectid;
1761 block_group = __btrfs_find_block_group(root,
1763 search_start, data, 1);
1765 search_start = max(search_start, hint_byte);
1767 total_needed += empty_size;
1771 block_group = btrfs_lookup_first_block_group(info,
1774 block_group = btrfs_lookup_first_block_group(info,
1777 if (full_scan && !chunk_alloc_done) {
1778 if (allowed_chunk_alloc) {
1779 do_chunk_alloc(trans, root,
1780 num_bytes + 2 * 1024 * 1024, data, 1);
1781 allowed_chunk_alloc = 0;
1782 } else if (block_group && block_group_bits(block_group, data)) {
1783 block_group->space_info->force_alloc = 1;
1785 chunk_alloc_done = 1;
1787 ret = find_search_start(root, &block_group, &search_start,
1788 total_needed, data);
1789 if (ret == -ENOSPC && last_ptr && *last_ptr) {
1791 block_group = btrfs_lookup_first_block_group(info,
1793 search_start = orig_search_start;
1794 ret = find_search_start(root, &block_group, &search_start,
1795 total_needed, data);
1802 if (last_ptr && *last_ptr && search_start != *last_ptr) {
1805 empty_size += empty_cluster;
1806 total_needed += empty_size;
1808 block_group = btrfs_lookup_first_block_group(info,
1810 search_start = orig_search_start;
1811 ret = find_search_start(root, &block_group,
1812 &search_start, total_needed, data);
1819 search_start = stripe_align(root, search_start);
1820 ins->objectid = search_start;
1821 ins->offset = num_bytes;
1823 if (ins->objectid + num_bytes >= search_end)
1826 if (ins->objectid + num_bytes >
1827 block_group->key.objectid + block_group->key.offset) {
1828 search_start = block_group->key.objectid +
1829 block_group->key.offset;
1833 if (test_range_bit(&info->extent_ins, ins->objectid,
1834 ins->objectid + num_bytes -1, EXTENT_LOCKED, 0)) {
1835 search_start = ins->objectid + num_bytes;
1839 if (test_range_bit(&info->pinned_extents, ins->objectid,
1840 ins->objectid + num_bytes -1, EXTENT_DIRTY, 0)) {
1841 search_start = ins->objectid + num_bytes;
1845 if (exclude_nr > 0 && (ins->objectid + num_bytes > exclude_start &&
1846 ins->objectid < exclude_start + exclude_nr)) {
1847 search_start = exclude_start + exclude_nr;
1851 if (!(data & BTRFS_BLOCK_GROUP_DATA)) {
1852 block_group = btrfs_lookup_block_group(info, ins->objectid);
1854 trans->block_group = block_group;
1856 ins->offset = num_bytes;
1858 *last_ptr = ins->objectid + ins->offset;
1860 btrfs_super_total_bytes(&root->fs_info->super_copy)) {
1867 if (search_start + num_bytes >= search_end) {
1869 search_start = orig_search_start;
1876 total_needed -= empty_size;
1881 block_group = btrfs_lookup_first_block_group(info, search_start);
1883 block_group = __btrfs_find_block_group(root, block_group,
1884 search_start, data, 0);
1892 * finds a free extent and does all the dirty work required for allocation
1893 * returns the key for the extent through ins, and a tree buffer for
1894 * the first block of the extent through buf.
1896 * returns 0 if everything worked, non-zero otherwise.
1898 int btrfs_alloc_extent(struct btrfs_trans_handle *trans,
1899 struct btrfs_root *root,
1900 u64 num_bytes, u64 min_alloc_size,
1901 u64 root_objectid, u64 ref_generation,
1902 u64 owner, u64 owner_offset,
1903 u64 empty_size, u64 hint_byte,
1904 u64 search_end, struct btrfs_key *ins, u64 data)
1910 u64 search_start = 0;
1913 struct btrfs_fs_info *info = root->fs_info;
1914 struct btrfs_root *extent_root = info->extent_root;
1915 struct btrfs_extent_item *extent_item;
1916 struct btrfs_extent_ref *ref;
1917 struct btrfs_path *path;
1918 struct btrfs_key keys[2];
1920 maybe_lock_mutex(root);
1923 alloc_profile = info->avail_data_alloc_bits &
1924 info->data_alloc_profile;
1925 data = BTRFS_BLOCK_GROUP_DATA | alloc_profile;
1926 } else if (root == root->fs_info->chunk_root) {
1927 alloc_profile = info->avail_system_alloc_bits &
1928 info->system_alloc_profile;
1929 data = BTRFS_BLOCK_GROUP_SYSTEM | alloc_profile;
1931 alloc_profile = info->avail_metadata_alloc_bits &
1932 info->metadata_alloc_profile;
1933 data = BTRFS_BLOCK_GROUP_METADATA | alloc_profile;
1936 data = reduce_alloc_profile(root, data);
1938 * the only place that sets empty_size is btrfs_realloc_node, which
1939 * is not called recursively on allocations
1941 if (empty_size || root->ref_cows) {
1942 if (!(data & BTRFS_BLOCK_GROUP_METADATA)) {
1943 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
1945 BTRFS_BLOCK_GROUP_METADATA |
1946 (info->metadata_alloc_profile &
1947 info->avail_metadata_alloc_bits), 0);
1950 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
1951 num_bytes + 2 * 1024 * 1024, data, 0);
1955 WARN_ON(num_bytes < root->sectorsize);
1956 ret = find_free_extent(trans, root, num_bytes, empty_size,
1957 search_start, search_end, hint_byte, ins,
1958 trans->alloc_exclude_start,
1959 trans->alloc_exclude_nr, data);
1961 if (ret == -ENOSPC && num_bytes > min_alloc_size) {
1962 num_bytes = num_bytes >> 1;
1963 num_bytes = max(num_bytes, min_alloc_size);
1964 do_chunk_alloc(trans, root->fs_info->extent_root,
1965 num_bytes, data, 1);
1969 printk("allocation failed flags %Lu\n", data);
1976 /* block accounting for super block */
1977 spin_lock_irq(&info->delalloc_lock);
1978 super_used = btrfs_super_bytes_used(&info->super_copy);
1979 btrfs_set_super_bytes_used(&info->super_copy, super_used + num_bytes);
1980 spin_unlock_irq(&info->delalloc_lock);
1982 /* block accounting for root item */
1983 root_used = btrfs_root_used(&root->root_item);
1984 btrfs_set_root_used(&root->root_item, root_used + num_bytes);
1986 clear_extent_dirty(&root->fs_info->free_space_cache,
1987 ins->objectid, ins->objectid + ins->offset - 1,
1990 if (root == extent_root) {
1991 set_extent_bits(&root->fs_info->extent_ins, ins->objectid,
1992 ins->objectid + ins->offset - 1,
1993 EXTENT_LOCKED, GFP_NOFS);
1997 WARN_ON(trans->alloc_exclude_nr);
1998 trans->alloc_exclude_start = ins->objectid;
1999 trans->alloc_exclude_nr = ins->offset;
2001 memcpy(&keys[0], ins, sizeof(*ins));
2002 keys[1].offset = hash_extent_ref(root_objectid, ref_generation,
2003 owner, owner_offset);
2004 keys[1].objectid = ins->objectid;
2005 keys[1].type = BTRFS_EXTENT_REF_KEY;
2006 sizes[0] = sizeof(*extent_item);
2007 sizes[1] = sizeof(*ref);
2009 path = btrfs_alloc_path();
2012 ret = btrfs_insert_empty_items(trans, extent_root, path, keys,
2016 extent_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
2017 struct btrfs_extent_item);
2018 btrfs_set_extent_refs(path->nodes[0], extent_item, 1);
2019 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
2020 struct btrfs_extent_ref);
2022 btrfs_set_ref_root(path->nodes[0], ref, root_objectid);
2023 btrfs_set_ref_generation(path->nodes[0], ref, ref_generation);
2024 btrfs_set_ref_objectid(path->nodes[0], ref, owner);
2025 btrfs_set_ref_offset(path->nodes[0], ref, owner_offset);
2027 btrfs_mark_buffer_dirty(path->nodes[0]);
2029 trans->alloc_exclude_start = 0;
2030 trans->alloc_exclude_nr = 0;
2031 btrfs_free_path(path);
2032 finish_current_insert(trans, extent_root);
2033 pending_ret = del_pending_extents(trans, extent_root);
2043 ret = update_block_group(trans, root, ins->objectid, ins->offset, 1, 0);
2045 printk("update block group failed for %Lu %Lu\n",
2046 ins->objectid, ins->offset);
2050 maybe_unlock_mutex(root);
2054 * helper function to allocate a block for a given tree
2055 * returns the tree buffer or NULL.
2057 struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
2058 struct btrfs_root *root,
2067 struct btrfs_key ins;
2069 struct extent_buffer *buf;
2071 ret = btrfs_alloc_extent(trans, root, blocksize, blocksize,
2072 root_objectid, ref_generation,
2073 level, first_objectid, empty_size, hint,
2077 return ERR_PTR(ret);
2079 buf = btrfs_find_create_tree_block(root, ins.objectid, blocksize);
2081 btrfs_free_extent(trans, root, ins.objectid, blocksize,
2082 root->root_key.objectid, ref_generation,
2084 return ERR_PTR(-ENOMEM);
2086 btrfs_set_header_generation(buf, trans->transid);
2087 btrfs_tree_lock(buf);
2088 clean_tree_block(trans, root, buf);
2089 btrfs_set_buffer_uptodate(buf);
2091 if (PageDirty(buf->first_page)) {
2092 printk("page %lu dirty\n", buf->first_page->index);
2096 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
2097 buf->start + buf->len - 1, GFP_NOFS);
2098 if (!btrfs_test_opt(root, SSD))
2099 btrfs_set_buffer_defrag(buf);
2100 trans->blocks_used++;
2104 static int noinline drop_leaf_ref(struct btrfs_trans_handle *trans,
2105 struct btrfs_root *root,
2106 struct extent_buffer *leaf)
2109 u64 leaf_generation;
2110 struct btrfs_key key;
2111 struct btrfs_file_extent_item *fi;
2116 BUG_ON(!btrfs_is_leaf(leaf));
2117 nritems = btrfs_header_nritems(leaf);
2118 leaf_owner = btrfs_header_owner(leaf);
2119 leaf_generation = btrfs_header_generation(leaf);
2121 for (i = 0; i < nritems; i++) {
2124 btrfs_item_key_to_cpu(leaf, &key, i);
2125 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
2127 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
2128 if (btrfs_file_extent_type(leaf, fi) ==
2129 BTRFS_FILE_EXTENT_INLINE)
2132 * FIXME make sure to insert a trans record that
2133 * repeats the snapshot del on crash
2135 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
2136 if (disk_bytenr == 0)
2138 ret = __btrfs_free_extent(trans, root, disk_bytenr,
2139 btrfs_file_extent_disk_num_bytes(leaf, fi),
2140 leaf_owner, leaf_generation,
2141 key.objectid, key.offset, 0);
2147 static void noinline reada_walk_down(struct btrfs_root *root,
2148 struct extent_buffer *node,
2161 nritems = btrfs_header_nritems(node);
2162 level = btrfs_header_level(node);
2166 for (i = slot; i < nritems && skipped < 32; i++) {
2167 bytenr = btrfs_node_blockptr(node, i);
2168 if (last && ((bytenr > last && bytenr - last > 32 * 1024) ||
2169 (last > bytenr && last - bytenr > 32 * 1024))) {
2173 blocksize = btrfs_level_size(root, level - 1);
2175 ret = lookup_extent_ref(NULL, root, bytenr,
2183 ret = readahead_tree_block(root, bytenr, blocksize,
2184 btrfs_node_ptr_generation(node, i));
2185 last = bytenr + blocksize;
2193 * we want to avoid as much random IO as we can with the alloc mutex
2194 * held, so drop the lock and do the lookup, then do it again with the
2197 int drop_snap_lookup_refcount(struct btrfs_root *root, u64 start, u64 len,
2200 mutex_unlock(&root->fs_info->alloc_mutex);
2201 lookup_extent_ref(NULL, root, start, len, refs);
2203 mutex_lock(&root->fs_info->alloc_mutex);
2204 return lookup_extent_ref(NULL, root, start, len, refs);
2208 * helper function for drop_snapshot, this walks down the tree dropping ref
2209 * counts as it goes.
2211 static int noinline walk_down_tree(struct btrfs_trans_handle *trans,
2212 struct btrfs_root *root,
2213 struct btrfs_path *path, int *level)
2219 struct extent_buffer *next;
2220 struct extent_buffer *cur;
2221 struct extent_buffer *parent;
2226 mutex_lock(&root->fs_info->alloc_mutex);
2228 WARN_ON(*level < 0);
2229 WARN_ON(*level >= BTRFS_MAX_LEVEL);
2230 ret = drop_snap_lookup_refcount(root, path->nodes[*level]->start,
2231 path->nodes[*level]->len, &refs);
2237 * walk down to the last node level and free all the leaves
2239 while(*level >= 0) {
2240 WARN_ON(*level < 0);
2241 WARN_ON(*level >= BTRFS_MAX_LEVEL);
2242 cur = path->nodes[*level];
2244 if (btrfs_header_level(cur) != *level)
2247 if (path->slots[*level] >=
2248 btrfs_header_nritems(cur))
2251 ret = drop_leaf_ref(trans, root, cur);
2255 bytenr = btrfs_node_blockptr(cur, path->slots[*level]);
2256 ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]);
2257 blocksize = btrfs_level_size(root, *level - 1);
2259 ret = drop_snap_lookup_refcount(root, bytenr, blocksize, &refs);
2262 parent = path->nodes[*level];
2263 root_owner = btrfs_header_owner(parent);
2264 root_gen = btrfs_header_generation(parent);
2265 path->slots[*level]++;
2266 ret = __btrfs_free_extent(trans, root, bytenr,
2267 blocksize, root_owner,
2272 next = btrfs_find_tree_block(root, bytenr, blocksize);
2273 if (!next || !btrfs_buffer_uptodate(next, ptr_gen)) {
2274 free_extent_buffer(next);
2275 mutex_unlock(&root->fs_info->alloc_mutex);
2277 if (path->slots[*level] == 0)
2278 reada_walk_down(root, cur, path->slots[*level]);
2280 next = read_tree_block(root, bytenr, blocksize,
2283 mutex_lock(&root->fs_info->alloc_mutex);
2285 /* we've dropped the lock, double check */
2286 ret = drop_snap_lookup_refcount(root, bytenr,
2290 parent = path->nodes[*level];
2291 root_owner = btrfs_header_owner(parent);
2292 root_gen = btrfs_header_generation(parent);
2294 path->slots[*level]++;
2295 free_extent_buffer(next);
2296 ret = __btrfs_free_extent(trans, root, bytenr,
2304 WARN_ON(*level <= 0);
2305 if (path->nodes[*level-1])
2306 free_extent_buffer(path->nodes[*level-1]);
2307 path->nodes[*level-1] = next;
2308 *level = btrfs_header_level(next);
2309 path->slots[*level] = 0;
2312 WARN_ON(*level < 0);
2313 WARN_ON(*level >= BTRFS_MAX_LEVEL);
2315 if (path->nodes[*level] == root->node) {
2316 root_owner = root->root_key.objectid;
2317 parent = path->nodes[*level];
2319 parent = path->nodes[*level + 1];
2320 root_owner = btrfs_header_owner(parent);
2323 root_gen = btrfs_header_generation(parent);
2324 ret = __btrfs_free_extent(trans, root, path->nodes[*level]->start,
2325 path->nodes[*level]->len,
2326 root_owner, root_gen, 0, 0, 1);
2327 free_extent_buffer(path->nodes[*level]);
2328 path->nodes[*level] = NULL;
2331 mutex_unlock(&root->fs_info->alloc_mutex);
2337 * helper for dropping snapshots. This walks back up the tree in the path
2338 * to find the first node higher up where we haven't yet gone through
2341 static int noinline walk_up_tree(struct btrfs_trans_handle *trans,
2342 struct btrfs_root *root,
2343 struct btrfs_path *path, int *level)
2347 struct btrfs_root_item *root_item = &root->root_item;
2352 for(i = *level; i < BTRFS_MAX_LEVEL - 1 && path->nodes[i]; i++) {
2353 slot = path->slots[i];
2354 if (slot < btrfs_header_nritems(path->nodes[i]) - 1) {
2355 struct extent_buffer *node;
2356 struct btrfs_disk_key disk_key;
2357 node = path->nodes[i];
2360 WARN_ON(*level == 0);
2361 btrfs_node_key(node, &disk_key, path->slots[i]);
2362 memcpy(&root_item->drop_progress,
2363 &disk_key, sizeof(disk_key));
2364 root_item->drop_level = i;
2367 if (path->nodes[*level] == root->node) {
2368 root_owner = root->root_key.objectid;
2370 btrfs_header_generation(path->nodes[*level]);
2372 struct extent_buffer *node;
2373 node = path->nodes[*level + 1];
2374 root_owner = btrfs_header_owner(node);
2375 root_gen = btrfs_header_generation(node);
2377 ret = btrfs_free_extent(trans, root,
2378 path->nodes[*level]->start,
2379 path->nodes[*level]->len,
2380 root_owner, root_gen, 0, 0, 1);
2382 free_extent_buffer(path->nodes[*level]);
2383 path->nodes[*level] = NULL;
2391 * drop the reference count on the tree rooted at 'snap'. This traverses
2392 * the tree freeing any blocks that have a ref count of zero after being
2395 int btrfs_drop_snapshot(struct btrfs_trans_handle *trans, struct btrfs_root
2401 struct btrfs_path *path;
2404 struct btrfs_root_item *root_item = &root->root_item;
2406 WARN_ON(!mutex_is_locked(&root->fs_info->drop_mutex));
2407 path = btrfs_alloc_path();
2410 level = btrfs_header_level(root->node);
2412 if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
2413 path->nodes[level] = root->node;
2414 extent_buffer_get(root->node);
2415 path->slots[level] = 0;
2417 struct btrfs_key key;
2418 struct btrfs_disk_key found_key;
2419 struct extent_buffer *node;
2421 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
2422 level = root_item->drop_level;
2423 path->lowest_level = level;
2424 wret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2429 node = path->nodes[level];
2430 btrfs_node_key(node, &found_key, path->slots[level]);
2431 WARN_ON(memcmp(&found_key, &root_item->drop_progress,
2432 sizeof(found_key)));
2433 for (i = 0; i < BTRFS_MAX_LEVEL; i++) {
2434 if (path->nodes[i] && path->locks[i]) {
2436 btrfs_tree_unlock(path->nodes[i]);
2441 wret = walk_down_tree(trans, root, path, &level);
2447 wret = walk_up_tree(trans, root, path, &level);
2452 if (trans->transaction->in_commit) {
2457 for (i = 0; i <= orig_level; i++) {
2458 if (path->nodes[i]) {
2459 free_extent_buffer(path->nodes[i]);
2460 path->nodes[i] = NULL;
2464 btrfs_free_path(path);
2468 int btrfs_free_block_groups(struct btrfs_fs_info *info)
2475 mutex_lock(&info->alloc_mutex);
2477 ret = find_first_extent_bit(&info->block_group_cache, 0,
2478 &start, &end, (unsigned int)-1);
2481 ret = get_state_private(&info->block_group_cache, start, &ptr);
2483 kfree((void *)(unsigned long)ptr);
2484 clear_extent_bits(&info->block_group_cache, start,
2485 end, (unsigned int)-1, GFP_NOFS);
2488 ret = find_first_extent_bit(&info->free_space_cache, 0,
2489 &start, &end, EXTENT_DIRTY);
2492 clear_extent_dirty(&info->free_space_cache, start,
2495 mutex_unlock(&info->alloc_mutex);
2499 static unsigned long calc_ra(unsigned long start, unsigned long last,
2502 return min(last, start + nr - 1);
2505 static int noinline relocate_inode_pages(struct inode *inode, u64 start,
2510 unsigned long last_index;
2513 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2514 struct file_ra_state *ra;
2515 unsigned long total_read = 0;
2516 unsigned long ra_pages;
2517 struct btrfs_trans_handle *trans;
2519 ra = kzalloc(sizeof(*ra), GFP_NOFS);
2521 mutex_lock(&inode->i_mutex);
2522 i = start >> PAGE_CACHE_SHIFT;
2523 last_index = (start + len - 1) >> PAGE_CACHE_SHIFT;
2525 ra_pages = BTRFS_I(inode)->root->fs_info->bdi.ra_pages;
2527 file_ra_state_init(ra, inode->i_mapping);
2529 for (; i <= last_index; i++) {
2530 if (total_read % ra_pages == 0) {
2531 btrfs_force_ra(inode->i_mapping, ra, NULL, i,
2532 calc_ra(i, last_index, ra_pages));
2535 if (((u64)i << PAGE_CACHE_SHIFT) > inode->i_size)
2536 goto truncate_racing;
2538 page = grab_cache_page(inode->i_mapping, i);
2542 if (!PageUptodate(page)) {
2543 btrfs_readpage(NULL, page);
2545 if (!PageUptodate(page)) {
2547 page_cache_release(page);
2551 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,18)
2552 ClearPageDirty(page);
2554 cancel_dirty_page(page, PAGE_CACHE_SIZE);
2556 wait_on_page_writeback(page);
2557 set_page_extent_mapped(page);
2558 page_start = (u64)page->index << PAGE_CACHE_SHIFT;
2559 page_end = page_start + PAGE_CACHE_SIZE - 1;
2561 lock_extent(io_tree, page_start, page_end, GFP_NOFS);
2563 set_extent_delalloc(io_tree, page_start,
2564 page_end, GFP_NOFS);
2565 set_page_dirty(page);
2567 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
2569 page_cache_release(page);
2571 balance_dirty_pages_ratelimited_nr(inode->i_mapping,
2576 trans = btrfs_start_transaction(BTRFS_I(inode)->root, 1);
2578 btrfs_add_ordered_inode(inode);
2579 btrfs_end_transaction(trans, BTRFS_I(inode)->root);
2580 mark_inode_dirty(inode);
2582 mutex_unlock(&inode->i_mutex);
2586 vmtruncate(inode, inode->i_size);
2587 balance_dirty_pages_ratelimited_nr(inode->i_mapping,
2593 * The back references tell us which tree holds a ref on a block,
2594 * but it is possible for the tree root field in the reference to
2595 * reflect the original root before a snapshot was made. In this
2596 * case we should search through all the children of a given root
2597 * to find potential holders of references on a block.
2599 * Instead, we do something a little less fancy and just search
2600 * all the roots for a given key/block combination.
2602 static int find_root_for_ref(struct btrfs_root *root,
2603 struct btrfs_path *path,
2604 struct btrfs_key *key0,
2607 struct btrfs_root **found_root,
2610 struct btrfs_key root_location;
2611 struct btrfs_root *cur_root = *found_root;
2612 struct btrfs_file_extent_item *file_extent;
2613 u64 root_search_start = BTRFS_FS_TREE_OBJECTID;
2618 root_location.offset = (u64)-1;
2619 root_location.type = BTRFS_ROOT_ITEM_KEY;
2620 path->lowest_level = level;
2623 ret = btrfs_search_slot(NULL, cur_root, key0, path, 0, 0);
2625 if (ret == 0 && file_key) {
2626 struct extent_buffer *leaf = path->nodes[0];
2627 file_extent = btrfs_item_ptr(leaf, path->slots[0],
2628 struct btrfs_file_extent_item);
2629 if (btrfs_file_extent_type(leaf, file_extent) ==
2630 BTRFS_FILE_EXTENT_REG) {
2632 btrfs_file_extent_disk_bytenr(leaf,
2635 } else if (!file_key) {
2636 if (path->nodes[level])
2637 found_bytenr = path->nodes[level]->start;
2640 for (i = level; i < BTRFS_MAX_LEVEL; i++) {
2641 if (!path->nodes[i])
2643 free_extent_buffer(path->nodes[i]);
2644 path->nodes[i] = NULL;
2646 btrfs_release_path(cur_root, path);
2648 if (found_bytenr == bytenr) {
2649 *found_root = cur_root;
2653 ret = btrfs_search_root(root->fs_info->tree_root,
2654 root_search_start, &root_search_start);
2658 root_location.objectid = root_search_start;
2659 cur_root = btrfs_read_fs_root_no_name(root->fs_info,
2667 path->lowest_level = 0;
2672 * note, this releases the path
2674 static int noinline relocate_one_reference(struct btrfs_root *extent_root,
2675 struct btrfs_path *path,
2676 struct btrfs_key *extent_key,
2677 u64 *last_file_objectid,
2678 u64 *last_file_offset,
2679 u64 *last_file_root,
2682 struct inode *inode;
2683 struct btrfs_root *found_root;
2684 struct btrfs_key root_location;
2685 struct btrfs_key found_key;
2686 struct btrfs_extent_ref *ref;
2694 ref = btrfs_item_ptr(path->nodes[0], path->slots[0],
2695 struct btrfs_extent_ref);
2696 ref_root = btrfs_ref_root(path->nodes[0], ref);
2697 ref_gen = btrfs_ref_generation(path->nodes[0], ref);
2698 ref_objectid = btrfs_ref_objectid(path->nodes[0], ref);
2699 ref_offset = btrfs_ref_offset(path->nodes[0], ref);
2700 btrfs_release_path(extent_root, path);
2702 root_location.objectid = ref_root;
2704 root_location.offset = 0;
2706 root_location.offset = (u64)-1;
2707 root_location.type = BTRFS_ROOT_ITEM_KEY;
2709 found_root = btrfs_read_fs_root_no_name(extent_root->fs_info,
2711 BUG_ON(!found_root);
2713 if (ref_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
2714 found_key.objectid = ref_objectid;
2715 found_key.type = BTRFS_EXTENT_DATA_KEY;
2716 found_key.offset = ref_offset;
2719 if (last_extent == extent_key->objectid &&
2720 *last_file_objectid == ref_objectid &&
2721 *last_file_offset == ref_offset &&
2722 *last_file_root == ref_root)
2725 ret = find_root_for_ref(extent_root, path, &found_key,
2726 level, 1, &found_root,
2727 extent_key->objectid);
2732 if (last_extent == extent_key->objectid &&
2733 *last_file_objectid == ref_objectid &&
2734 *last_file_offset == ref_offset &&
2735 *last_file_root == ref_root)
2738 inode = btrfs_iget_locked(extent_root->fs_info->sb,
2739 ref_objectid, found_root);
2740 if (inode->i_state & I_NEW) {
2741 /* the inode and parent dir are two different roots */
2742 BTRFS_I(inode)->root = found_root;
2743 BTRFS_I(inode)->location.objectid = ref_objectid;
2744 BTRFS_I(inode)->location.type = BTRFS_INODE_ITEM_KEY;
2745 BTRFS_I(inode)->location.offset = 0;
2746 btrfs_read_locked_inode(inode);
2747 unlock_new_inode(inode);
2750 /* this can happen if the reference is not against
2751 * the latest version of the tree root
2753 if (is_bad_inode(inode)) {
2756 *last_file_objectid = inode->i_ino;
2757 *last_file_root = found_root->root_key.objectid;
2758 *last_file_offset = ref_offset;
2760 relocate_inode_pages(inode, ref_offset, extent_key->offset);
2763 struct btrfs_trans_handle *trans;
2764 struct extent_buffer *eb;
2767 eb = read_tree_block(found_root, extent_key->objectid,
2768 extent_key->offset, 0);
2769 btrfs_tree_lock(eb);
2770 level = btrfs_header_level(eb);
2773 btrfs_item_key_to_cpu(eb, &found_key, 0);
2775 btrfs_node_key_to_cpu(eb, &found_key, 0);
2777 btrfs_tree_unlock(eb);
2778 free_extent_buffer(eb);
2780 ret = find_root_for_ref(extent_root, path, &found_key,
2781 level, 0, &found_root,
2782 extent_key->objectid);
2787 trans = btrfs_start_transaction(found_root, 1);
2789 path->lowest_level = level;
2791 ret = btrfs_search_slot(trans, found_root, &found_key, path,
2793 path->lowest_level = 0;
2794 for (i = level; i < BTRFS_MAX_LEVEL; i++) {
2795 if (!path->nodes[i])
2797 free_extent_buffer(path->nodes[i]);
2798 path->nodes[i] = NULL;
2800 btrfs_release_path(found_root, path);
2801 if (found_root == found_root->fs_info->extent_root)
2802 btrfs_extent_post_op(trans, found_root);
2803 btrfs_end_transaction(trans, found_root);
2810 static int noinline del_extent_zero(struct btrfs_root *extent_root,
2811 struct btrfs_path *path,
2812 struct btrfs_key *extent_key)
2815 struct btrfs_trans_handle *trans;
2817 trans = btrfs_start_transaction(extent_root, 1);
2818 ret = btrfs_search_slot(trans, extent_root, extent_key, path, -1, 1);
2825 ret = btrfs_del_item(trans, extent_root, path);
2827 btrfs_end_transaction(trans, extent_root);
2831 static int noinline relocate_one_extent(struct btrfs_root *extent_root,
2832 struct btrfs_path *path,
2833 struct btrfs_key *extent_key)
2835 struct btrfs_key key;
2836 struct btrfs_key found_key;
2837 struct extent_buffer *leaf;
2838 u64 last_file_objectid = 0;
2839 u64 last_file_root = 0;
2840 u64 last_file_offset = (u64)-1;
2841 u64 last_extent = 0;
2846 if (extent_key->objectid == 0) {
2847 ret = del_extent_zero(extent_root, path, extent_key);
2850 key.objectid = extent_key->objectid;
2851 key.type = BTRFS_EXTENT_REF_KEY;
2855 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
2861 leaf = path->nodes[0];
2862 nritems = btrfs_header_nritems(leaf);
2863 if (path->slots[0] == nritems) {
2864 ret = btrfs_next_leaf(extent_root, path);
2871 leaf = path->nodes[0];
2874 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2875 if (found_key.objectid != extent_key->objectid) {
2879 if (found_key.type != BTRFS_EXTENT_REF_KEY) {
2883 key.offset = found_key.offset + 1;
2884 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2886 ret = relocate_one_reference(extent_root, path, extent_key,
2887 &last_file_objectid,
2889 &last_file_root, last_extent);
2892 last_extent = extent_key->objectid;
2896 btrfs_release_path(extent_root, path);
2900 static u64 update_block_group_flags(struct btrfs_root *root, u64 flags)
2903 u64 stripped = BTRFS_BLOCK_GROUP_RAID0 |
2904 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
2906 num_devices = root->fs_info->fs_devices->num_devices;
2907 if (num_devices == 1) {
2908 stripped |= BTRFS_BLOCK_GROUP_DUP;
2909 stripped = flags & ~stripped;
2911 /* turn raid0 into single device chunks */
2912 if (flags & BTRFS_BLOCK_GROUP_RAID0)
2915 /* turn mirroring into duplication */
2916 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
2917 BTRFS_BLOCK_GROUP_RAID10))
2918 return stripped | BTRFS_BLOCK_GROUP_DUP;
2921 /* they already had raid on here, just return */
2922 if (flags & stripped)
2925 stripped |= BTRFS_BLOCK_GROUP_DUP;
2926 stripped = flags & ~stripped;
2928 /* switch duplicated blocks with raid1 */
2929 if (flags & BTRFS_BLOCK_GROUP_DUP)
2930 return stripped | BTRFS_BLOCK_GROUP_RAID1;
2932 /* turn single device chunks into raid0 */
2933 return stripped | BTRFS_BLOCK_GROUP_RAID0;
2938 int __alloc_chunk_for_shrink(struct btrfs_root *root,
2939 struct btrfs_block_group_cache *shrink_block_group,
2942 struct btrfs_trans_handle *trans;
2943 u64 new_alloc_flags;
2946 if (btrfs_block_group_used(&shrink_block_group->item) > 0) {
2948 trans = btrfs_start_transaction(root, 1);
2949 new_alloc_flags = update_block_group_flags(root,
2950 shrink_block_group->flags);
2951 if (new_alloc_flags != shrink_block_group->flags) {
2953 btrfs_block_group_used(&shrink_block_group->item);
2955 calc = shrink_block_group->key.offset;
2957 do_chunk_alloc(trans, root->fs_info->extent_root,
2958 calc + 2 * 1024 * 1024, new_alloc_flags, force);
2959 btrfs_end_transaction(trans, root);
2964 int btrfs_shrink_extent_tree(struct btrfs_root *root, u64 shrink_start)
2966 struct btrfs_trans_handle *trans;
2967 struct btrfs_root *tree_root = root->fs_info->tree_root;
2968 struct btrfs_path *path;
2971 u64 shrink_last_byte;
2972 struct btrfs_block_group_cache *shrink_block_group;
2973 struct btrfs_fs_info *info = root->fs_info;
2974 struct btrfs_key key;
2975 struct btrfs_key found_key;
2976 struct extent_buffer *leaf;
2981 mutex_lock(&root->fs_info->alloc_mutex);
2982 shrink_block_group = btrfs_lookup_block_group(root->fs_info,
2984 BUG_ON(!shrink_block_group);
2986 shrink_last_byte = shrink_block_group->key.objectid +
2987 shrink_block_group->key.offset;
2989 shrink_block_group->space_info->total_bytes -=
2990 shrink_block_group->key.offset;
2991 path = btrfs_alloc_path();
2992 root = root->fs_info->extent_root;
2995 printk("btrfs relocating block group %llu flags %llu\n",
2996 (unsigned long long)shrink_start,
2997 (unsigned long long)shrink_block_group->flags);
2999 __alloc_chunk_for_shrink(root, shrink_block_group, 1);
3003 shrink_block_group->ro = 1;
3007 key.objectid = shrink_start;
3010 cur_byte = key.objectid;
3012 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3016 ret = btrfs_previous_item(root, path, 0, BTRFS_EXTENT_ITEM_KEY);
3021 leaf = path->nodes[0];
3022 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3023 if (found_key.objectid + found_key.offset > shrink_start &&
3024 found_key.objectid < shrink_last_byte) {
3025 cur_byte = found_key.objectid;
3026 key.objectid = cur_byte;
3029 btrfs_release_path(root, path);
3032 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3036 leaf = path->nodes[0];
3037 nritems = btrfs_header_nritems(leaf);
3039 if (path->slots[0] >= nritems) {
3040 ret = btrfs_next_leaf(root, path);
3047 leaf = path->nodes[0];
3048 nritems = btrfs_header_nritems(leaf);
3051 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3053 if (found_key.objectid >= shrink_last_byte)
3056 if (progress && need_resched()) {
3057 memcpy(&key, &found_key, sizeof(key));
3059 btrfs_release_path(root, path);
3060 btrfs_search_slot(NULL, root, &key, path, 0, 0);
3066 if (btrfs_key_type(&found_key) != BTRFS_EXTENT_ITEM_KEY ||
3067 found_key.objectid + found_key.offset <= cur_byte) {
3068 memcpy(&key, &found_key, sizeof(key));
3075 cur_byte = found_key.objectid + found_key.offset;
3076 key.objectid = cur_byte;
3077 btrfs_release_path(root, path);
3078 ret = relocate_one_extent(root, path, &found_key);
3079 __alloc_chunk_for_shrink(root, shrink_block_group, 0);
3082 btrfs_release_path(root, path);
3084 if (total_found > 0) {
3085 printk("btrfs relocate found %llu last extent was %llu\n",
3086 (unsigned long long)total_found,
3087 (unsigned long long)found_key.objectid);
3088 trans = btrfs_start_transaction(tree_root, 1);
3089 btrfs_commit_transaction(trans, tree_root);
3091 btrfs_clean_old_snapshots(tree_root);
3093 trans = btrfs_start_transaction(tree_root, 1);
3094 btrfs_commit_transaction(trans, tree_root);
3099 * we've freed all the extents, now remove the block
3100 * group item from the tree
3102 trans = btrfs_start_transaction(root, 1);
3103 memcpy(&key, &shrink_block_group->key, sizeof(key));
3105 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3111 clear_extent_bits(&info->block_group_cache, key.objectid,
3112 key.objectid + key.offset - 1,
3113 (unsigned int)-1, GFP_NOFS);
3116 clear_extent_bits(&info->free_space_cache,
3117 key.objectid, key.objectid + key.offset - 1,
3118 (unsigned int)-1, GFP_NOFS);
3120 memset(shrink_block_group, 0, sizeof(*shrink_block_group));
3121 kfree(shrink_block_group);
3123 btrfs_del_item(trans, root, path);
3124 btrfs_commit_transaction(trans, root);
3126 /* the code to unpin extents might set a few bits in the free
3127 * space cache for this range again
3129 clear_extent_bits(&info->free_space_cache,
3130 key.objectid, key.objectid + key.offset - 1,
3131 (unsigned int)-1, GFP_NOFS);
3133 btrfs_free_path(path);
3134 mutex_unlock(&root->fs_info->alloc_mutex);
3138 int find_first_block_group(struct btrfs_root *root, struct btrfs_path *path,
3139 struct btrfs_key *key)
3142 struct btrfs_key found_key;
3143 struct extent_buffer *leaf;
3146 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
3151 slot = path->slots[0];
3152 leaf = path->nodes[0];
3153 if (slot >= btrfs_header_nritems(leaf)) {
3154 ret = btrfs_next_leaf(root, path);
3161 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3163 if (found_key.objectid >= key->objectid &&
3164 found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
3175 int btrfs_read_block_groups(struct btrfs_root *root)
3177 struct btrfs_path *path;
3180 struct btrfs_block_group_cache *cache;
3181 struct btrfs_fs_info *info = root->fs_info;
3182 struct btrfs_space_info *space_info;
3183 struct extent_io_tree *block_group_cache;
3184 struct btrfs_key key;
3185 struct btrfs_key found_key;
3186 struct extent_buffer *leaf;
3188 block_group_cache = &info->block_group_cache;
3189 root = info->extent_root;
3192 btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY);
3193 path = btrfs_alloc_path();
3197 mutex_lock(&root->fs_info->alloc_mutex);
3199 ret = find_first_block_group(root, path, &key);
3207 leaf = path->nodes[0];
3208 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3209 cache = kzalloc(sizeof(*cache), GFP_NOFS);
3215 read_extent_buffer(leaf, &cache->item,
3216 btrfs_item_ptr_offset(leaf, path->slots[0]),
3217 sizeof(cache->item));
3218 memcpy(&cache->key, &found_key, sizeof(found_key));
3220 key.objectid = found_key.objectid + found_key.offset;
3221 btrfs_release_path(root, path);
3222 cache->flags = btrfs_block_group_flags(&cache->item);
3224 if (cache->flags & BTRFS_BLOCK_GROUP_DATA) {
3225 bit = BLOCK_GROUP_DATA;
3226 } else if (cache->flags & BTRFS_BLOCK_GROUP_SYSTEM) {
3227 bit = BLOCK_GROUP_SYSTEM;
3228 } else if (cache->flags & BTRFS_BLOCK_GROUP_METADATA) {
3229 bit = BLOCK_GROUP_METADATA;
3231 set_avail_alloc_bits(info, cache->flags);
3233 ret = update_space_info(info, cache->flags, found_key.offset,
3234 btrfs_block_group_used(&cache->item),
3237 cache->space_info = space_info;
3239 /* use EXTENT_LOCKED to prevent merging */
3240 set_extent_bits(block_group_cache, found_key.objectid,
3241 found_key.objectid + found_key.offset - 1,
3242 bit | EXTENT_LOCKED, GFP_NOFS);
3243 set_state_private(block_group_cache, found_key.objectid,
3244 (unsigned long)cache);
3247 btrfs_super_total_bytes(&info->super_copy))
3252 btrfs_free_path(path);
3253 mutex_unlock(&root->fs_info->alloc_mutex);
3257 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
3258 struct btrfs_root *root, u64 bytes_used,
3259 u64 type, u64 chunk_objectid, u64 chunk_offset,
3264 struct btrfs_root *extent_root;
3265 struct btrfs_block_group_cache *cache;
3266 struct extent_io_tree *block_group_cache;
3268 extent_root = root->fs_info->extent_root;
3269 block_group_cache = &root->fs_info->block_group_cache;
3271 cache = kzalloc(sizeof(*cache), GFP_NOFS);
3273 cache->key.objectid = chunk_offset;
3274 cache->key.offset = size;
3275 btrfs_set_key_type(&cache->key, BTRFS_BLOCK_GROUP_ITEM_KEY);
3277 btrfs_set_block_group_used(&cache->item, bytes_used);
3278 btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
3279 cache->flags = type;
3280 btrfs_set_block_group_flags(&cache->item, type);
3282 ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
3283 &cache->space_info);
3286 bit = block_group_state_bits(type);
3287 set_extent_bits(block_group_cache, chunk_offset,
3288 chunk_offset + size - 1,
3289 bit | EXTENT_LOCKED, GFP_NOFS);
3291 set_state_private(block_group_cache, chunk_offset,
3292 (unsigned long)cache);
3293 ret = btrfs_insert_item(trans, extent_root, &cache->key, &cache->item,
3294 sizeof(cache->item));
3297 finish_current_insert(trans, extent_root);
3298 ret = del_pending_extents(trans, extent_root);
3300 set_avail_alloc_bits(extent_root->fs_info, type);