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);
1226 mutex_unlock(&extent_root->fs_info->chunk_mutex);
1231 static int update_block_group(struct btrfs_trans_handle *trans,
1232 struct btrfs_root *root,
1233 u64 bytenr, u64 num_bytes, int alloc,
1236 struct btrfs_block_group_cache *cache;
1237 struct btrfs_fs_info *info = root->fs_info;
1238 u64 total = num_bytes;
1245 cache = btrfs_lookup_block_group(info, bytenr);
1249 byte_in_group = bytenr - cache->key.objectid;
1250 WARN_ON(byte_in_group > cache->key.offset);
1251 start = cache->key.objectid;
1252 end = start + cache->key.offset - 1;
1253 set_extent_bits(&info->block_group_cache, start, end,
1254 BLOCK_GROUP_DIRTY, GFP_NOFS);
1256 old_val = btrfs_block_group_used(&cache->item);
1257 num_bytes = min(total, cache->key.offset - byte_in_group);
1259 old_val += num_bytes;
1260 cache->space_info->bytes_used += num_bytes;
1262 old_val -= num_bytes;
1263 cache->space_info->bytes_used -= num_bytes;
1265 set_extent_dirty(&info->free_space_cache,
1266 bytenr, bytenr + num_bytes - 1,
1270 btrfs_set_block_group_used(&cache->item, old_val);
1272 bytenr += num_bytes;
1277 static u64 first_logical_byte(struct btrfs_root *root, u64 search_start)
1282 ret = find_first_extent_bit(&root->fs_info->block_group_cache,
1283 search_start, &start, &end,
1284 BLOCK_GROUP_DATA | BLOCK_GROUP_METADATA |
1285 BLOCK_GROUP_SYSTEM);
1292 static int update_pinned_extents(struct btrfs_root *root,
1293 u64 bytenr, u64 num, int pin)
1296 struct btrfs_block_group_cache *cache;
1297 struct btrfs_fs_info *fs_info = root->fs_info;
1300 set_extent_dirty(&fs_info->pinned_extents,
1301 bytenr, bytenr + num - 1, GFP_NOFS);
1303 clear_extent_dirty(&fs_info->pinned_extents,
1304 bytenr, bytenr + num - 1, GFP_NOFS);
1307 cache = btrfs_lookup_block_group(fs_info, bytenr);
1309 u64 first = first_logical_byte(root, bytenr);
1310 WARN_ON(first < bytenr);
1311 len = min(first - bytenr, num);
1313 len = min(num, cache->key.offset -
1314 (bytenr - cache->key.objectid));
1318 cache->pinned += len;
1319 cache->space_info->bytes_pinned += len;
1321 fs_info->total_pinned += len;
1324 cache->pinned -= len;
1325 cache->space_info->bytes_pinned -= len;
1327 fs_info->total_pinned -= len;
1335 int btrfs_copy_pinned(struct btrfs_root *root, struct extent_io_tree *copy)
1340 struct extent_io_tree *pinned_extents = &root->fs_info->pinned_extents;
1344 ret = find_first_extent_bit(pinned_extents, last,
1345 &start, &end, EXTENT_DIRTY);
1348 set_extent_dirty(copy, start, end, GFP_NOFS);
1354 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
1355 struct btrfs_root *root,
1356 struct extent_io_tree *unpin)
1361 struct extent_io_tree *free_space_cache;
1362 free_space_cache = &root->fs_info->free_space_cache;
1364 mutex_lock(&root->fs_info->alloc_mutex);
1366 ret = find_first_extent_bit(unpin, 0, &start, &end,
1370 update_pinned_extents(root, start, end + 1 - start, 0);
1371 clear_extent_dirty(unpin, start, end, GFP_NOFS);
1372 set_extent_dirty(free_space_cache, start, end, GFP_NOFS);
1374 mutex_unlock(&root->fs_info->alloc_mutex);
1378 static int finish_current_insert(struct btrfs_trans_handle *trans,
1379 struct btrfs_root *extent_root)
1383 struct btrfs_fs_info *info = extent_root->fs_info;
1384 struct extent_buffer *eb;
1385 struct btrfs_path *path;
1386 struct btrfs_key ins;
1387 struct btrfs_disk_key first;
1388 struct btrfs_extent_item extent_item;
1393 btrfs_set_stack_extent_refs(&extent_item, 1);
1394 btrfs_set_key_type(&ins, BTRFS_EXTENT_ITEM_KEY);
1395 path = btrfs_alloc_path();
1398 ret = find_first_extent_bit(&info->extent_ins, 0, &start,
1399 &end, EXTENT_LOCKED);
1403 ins.objectid = start;
1404 ins.offset = end + 1 - start;
1405 err = btrfs_insert_item(trans, extent_root, &ins,
1406 &extent_item, sizeof(extent_item));
1407 clear_extent_bits(&info->extent_ins, start, end, EXTENT_LOCKED,
1409 eb = read_tree_block(extent_root, ins.objectid, ins.offset,
1411 btrfs_tree_lock(eb);
1412 level = btrfs_header_level(eb);
1414 btrfs_item_key(eb, &first, 0);
1416 btrfs_node_key(eb, &first, 0);
1418 btrfs_tree_unlock(eb);
1419 free_extent_buffer(eb);
1421 * the first key is just a hint, so the race we've created
1422 * against reading it is fine
1424 err = btrfs_insert_extent_backref(trans, extent_root, path,
1425 start, extent_root->root_key.objectid,
1427 btrfs_disk_key_objectid(&first));
1430 btrfs_free_path(path);
1434 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);
1463 update_pinned_extents(root, bytenr, num_bytes, 1);
1465 set_extent_bits(&root->fs_info->pending_del,
1466 bytenr, bytenr + num_bytes - 1,
1467 EXTENT_LOCKED, GFP_NOFS);
1474 * remove an extent from the root, returns 0 on success
1476 static int __free_extent(struct btrfs_trans_handle *trans, struct btrfs_root
1477 *root, u64 bytenr, u64 num_bytes,
1478 u64 root_objectid, u64 ref_generation,
1479 u64 owner_objectid, u64 owner_offset, int pin,
1482 struct btrfs_path *path;
1483 struct btrfs_key key;
1484 struct btrfs_fs_info *info = root->fs_info;
1485 struct btrfs_root *extent_root = info->extent_root;
1486 struct extent_buffer *leaf;
1488 int extent_slot = 0;
1489 int found_extent = 0;
1491 struct btrfs_extent_item *ei;
1494 key.objectid = bytenr;
1495 btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
1496 key.offset = num_bytes;
1497 path = btrfs_alloc_path();
1502 ret = lookup_extent_backref(trans, extent_root, path,
1503 bytenr, root_objectid,
1505 owner_objectid, owner_offset, 1);
1507 struct btrfs_key found_key;
1508 extent_slot = path->slots[0];
1509 while(extent_slot > 0) {
1511 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1513 if (found_key.objectid != bytenr)
1515 if (found_key.type == BTRFS_EXTENT_ITEM_KEY &&
1516 found_key.offset == num_bytes) {
1520 if (path->slots[0] - extent_slot > 5)
1524 ret = btrfs_del_item(trans, extent_root, path);
1526 btrfs_print_leaf(extent_root, path->nodes[0]);
1528 printk("Unable to find ref byte nr %Lu root %Lu "
1529 " gen %Lu owner %Lu offset %Lu\n", bytenr,
1530 root_objectid, ref_generation, owner_objectid,
1533 if (!found_extent) {
1534 btrfs_release_path(extent_root, path);
1535 ret = btrfs_search_slot(trans, extent_root, &key, path, -1, 1);
1539 extent_slot = path->slots[0];
1542 leaf = path->nodes[0];
1543 ei = btrfs_item_ptr(leaf, extent_slot,
1544 struct btrfs_extent_item);
1545 refs = btrfs_extent_refs(leaf, ei);
1548 btrfs_set_extent_refs(leaf, ei, refs);
1550 btrfs_mark_buffer_dirty(leaf);
1552 if (refs == 0 && found_extent && path->slots[0] == extent_slot + 1) {
1553 /* if the back ref and the extent are next to each other
1554 * they get deleted below in one shot
1556 path->slots[0] = extent_slot;
1558 } else if (found_extent) {
1559 /* otherwise delete the extent back ref */
1560 ret = btrfs_del_item(trans, extent_root, path);
1562 /* if refs are 0, we need to setup the path for deletion */
1564 btrfs_release_path(extent_root, path);
1565 ret = btrfs_search_slot(trans, extent_root, &key, path,
1578 ret = pin_down_bytes(root, bytenr, num_bytes, 0);
1584 /* block accounting for super block */
1585 spin_lock_irq(&info->delalloc_lock);
1586 super_used = btrfs_super_bytes_used(&info->super_copy);
1587 btrfs_set_super_bytes_used(&info->super_copy,
1588 super_used - num_bytes);
1589 spin_unlock_irq(&info->delalloc_lock);
1591 /* block accounting for root item */
1592 root_used = btrfs_root_used(&root->root_item);
1593 btrfs_set_root_used(&root->root_item,
1594 root_used - num_bytes);
1595 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
1600 ret = update_block_group(trans, root, bytenr, num_bytes, 0,
1604 btrfs_free_path(path);
1605 finish_current_insert(trans, extent_root);
1610 * find all the blocks marked as pending in the radix tree and remove
1611 * them from the extent map
1613 static int del_pending_extents(struct btrfs_trans_handle *trans, struct
1614 btrfs_root *extent_root)
1620 struct extent_io_tree *pending_del;
1621 struct extent_io_tree *pinned_extents;
1623 pending_del = &extent_root->fs_info->pending_del;
1624 pinned_extents = &extent_root->fs_info->pinned_extents;
1627 ret = find_first_extent_bit(pending_del, 0, &start, &end,
1631 update_pinned_extents(extent_root, start, end + 1 - start, 1);
1632 clear_extent_bits(pending_del, start, end, EXTENT_LOCKED,
1634 ret = __free_extent(trans, extent_root,
1635 start, end + 1 - start,
1636 extent_root->root_key.objectid,
1645 * remove an extent from the root, returns 0 on success
1647 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
1648 struct btrfs_root *root, u64 bytenr,
1649 u64 num_bytes, u64 root_objectid,
1650 u64 ref_generation, u64 owner_objectid,
1651 u64 owner_offset, int pin)
1653 struct btrfs_root *extent_root = root->fs_info->extent_root;
1657 WARN_ON(num_bytes < root->sectorsize);
1658 if (!root->ref_cows)
1661 if (root == extent_root) {
1662 pin_down_bytes(root, bytenr, num_bytes, 1);
1665 ret = __free_extent(trans, root, bytenr, num_bytes, root_objectid,
1666 ref_generation, owner_objectid, owner_offset,
1668 pending_ret = del_pending_extents(trans, root->fs_info->extent_root);
1669 return ret ? ret : pending_ret;
1672 int btrfs_free_extent(struct btrfs_trans_handle *trans,
1673 struct btrfs_root *root, u64 bytenr,
1674 u64 num_bytes, u64 root_objectid,
1675 u64 ref_generation, u64 owner_objectid,
1676 u64 owner_offset, int pin)
1680 maybe_lock_mutex(root);
1681 ret = __btrfs_free_extent(trans, root, bytenr, num_bytes,
1682 root_objectid, ref_generation,
1683 owner_objectid, owner_offset, pin);
1684 maybe_unlock_mutex(root);
1688 static u64 stripe_align(struct btrfs_root *root, u64 val)
1690 u64 mask = ((u64)root->stripesize - 1);
1691 u64 ret = (val + mask) & ~mask;
1696 * walks the btree of allocated extents and find a hole of a given size.
1697 * The key ins is changed to record the hole:
1698 * ins->objectid == block start
1699 * ins->flags = BTRFS_EXTENT_ITEM_KEY
1700 * ins->offset == number of blocks
1701 * Any available blocks before search_start are skipped.
1703 static int noinline find_free_extent(struct btrfs_trans_handle *trans,
1704 struct btrfs_root *orig_root,
1705 u64 num_bytes, u64 empty_size,
1706 u64 search_start, u64 search_end,
1707 u64 hint_byte, struct btrfs_key *ins,
1708 u64 exclude_start, u64 exclude_nr,
1712 u64 orig_search_start;
1713 struct btrfs_root * root = orig_root->fs_info->extent_root;
1714 struct btrfs_fs_info *info = root->fs_info;
1715 u64 total_needed = num_bytes;
1716 u64 *last_ptr = NULL;
1717 struct btrfs_block_group_cache *block_group;
1720 int chunk_alloc_done = 0;
1721 int empty_cluster = 2 * 1024 * 1024;
1722 int allowed_chunk_alloc = 0;
1724 WARN_ON(num_bytes < root->sectorsize);
1725 btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY);
1727 if (orig_root->ref_cows || empty_size)
1728 allowed_chunk_alloc = 1;
1730 if (data & BTRFS_BLOCK_GROUP_METADATA) {
1731 last_ptr = &root->fs_info->last_alloc;
1732 empty_cluster = 256 * 1024;
1735 if ((data & BTRFS_BLOCK_GROUP_DATA) && btrfs_test_opt(root, SSD)) {
1736 last_ptr = &root->fs_info->last_data_alloc;
1741 hint_byte = *last_ptr;
1743 empty_size += empty_cluster;
1747 search_start = max(search_start, first_logical_byte(root, 0));
1748 orig_search_start = search_start;
1750 if (search_end == (u64)-1)
1751 search_end = btrfs_super_total_bytes(&info->super_copy);
1754 block_group = btrfs_lookup_first_block_group(info, hint_byte);
1756 hint_byte = search_start;
1757 block_group = __btrfs_find_block_group(root, block_group,
1758 hint_byte, data, 1);
1759 if (last_ptr && *last_ptr == 0 && block_group)
1760 hint_byte = block_group->key.objectid;
1762 block_group = __btrfs_find_block_group(root,
1764 search_start, data, 1);
1766 search_start = max(search_start, hint_byte);
1768 total_needed += empty_size;
1772 block_group = btrfs_lookup_first_block_group(info,
1775 block_group = btrfs_lookup_first_block_group(info,
1778 if (full_scan && !chunk_alloc_done) {
1779 if (allowed_chunk_alloc) {
1780 do_chunk_alloc(trans, root,
1781 num_bytes + 2 * 1024 * 1024, data, 1);
1782 allowed_chunk_alloc = 0;
1783 } else if (block_group && block_group_bits(block_group, data)) {
1784 block_group->space_info->force_alloc = 1;
1786 chunk_alloc_done = 1;
1788 ret = find_search_start(root, &block_group, &search_start,
1789 total_needed, data);
1790 if (ret == -ENOSPC && last_ptr && *last_ptr) {
1792 block_group = btrfs_lookup_first_block_group(info,
1794 search_start = orig_search_start;
1795 ret = find_search_start(root, &block_group, &search_start,
1796 total_needed, data);
1803 if (last_ptr && *last_ptr && search_start != *last_ptr) {
1806 empty_size += empty_cluster;
1807 total_needed += empty_size;
1809 block_group = btrfs_lookup_first_block_group(info,
1811 search_start = orig_search_start;
1812 ret = find_search_start(root, &block_group,
1813 &search_start, total_needed, data);
1820 search_start = stripe_align(root, search_start);
1821 ins->objectid = search_start;
1822 ins->offset = num_bytes;
1824 if (ins->objectid + num_bytes >= search_end)
1827 if (ins->objectid + num_bytes >
1828 block_group->key.objectid + block_group->key.offset) {
1829 search_start = block_group->key.objectid +
1830 block_group->key.offset;
1834 if (test_range_bit(&info->extent_ins, ins->objectid,
1835 ins->objectid + num_bytes -1, EXTENT_LOCKED, 0)) {
1836 search_start = ins->objectid + num_bytes;
1840 if (test_range_bit(&info->pinned_extents, ins->objectid,
1841 ins->objectid + num_bytes -1, EXTENT_DIRTY, 0)) {
1842 search_start = ins->objectid + num_bytes;
1846 if (exclude_nr > 0 && (ins->objectid + num_bytes > exclude_start &&
1847 ins->objectid < exclude_start + exclude_nr)) {
1848 search_start = exclude_start + exclude_nr;
1852 if (!(data & BTRFS_BLOCK_GROUP_DATA)) {
1853 block_group = btrfs_lookup_block_group(info, ins->objectid);
1855 trans->block_group = block_group;
1857 ins->offset = num_bytes;
1859 *last_ptr = ins->objectid + ins->offset;
1861 btrfs_super_total_bytes(&root->fs_info->super_copy)) {
1868 if (search_start + num_bytes >= search_end) {
1870 search_start = orig_search_start;
1877 total_needed -= empty_size;
1882 block_group = btrfs_lookup_first_block_group(info, search_start);
1884 block_group = __btrfs_find_block_group(root, block_group,
1885 search_start, data, 0);
1893 * finds a free extent and does all the dirty work required for allocation
1894 * returns the key for the extent through ins, and a tree buffer for
1895 * the first block of the extent through buf.
1897 * returns 0 if everything worked, non-zero otherwise.
1899 int btrfs_alloc_extent(struct btrfs_trans_handle *trans,
1900 struct btrfs_root *root,
1901 u64 num_bytes, u64 min_alloc_size,
1902 u64 root_objectid, u64 ref_generation,
1903 u64 owner, u64 owner_offset,
1904 u64 empty_size, u64 hint_byte,
1905 u64 search_end, struct btrfs_key *ins, u64 data)
1911 u64 search_start = 0;
1914 struct btrfs_fs_info *info = root->fs_info;
1915 struct btrfs_root *extent_root = info->extent_root;
1916 struct btrfs_extent_item *extent_item;
1917 struct btrfs_extent_ref *ref;
1918 struct btrfs_path *path;
1919 struct btrfs_key keys[2];
1921 maybe_lock_mutex(root);
1924 alloc_profile = info->avail_data_alloc_bits &
1925 info->data_alloc_profile;
1926 data = BTRFS_BLOCK_GROUP_DATA | alloc_profile;
1927 } else if (root == root->fs_info->chunk_root) {
1928 alloc_profile = info->avail_system_alloc_bits &
1929 info->system_alloc_profile;
1930 data = BTRFS_BLOCK_GROUP_SYSTEM | alloc_profile;
1932 alloc_profile = info->avail_metadata_alloc_bits &
1933 info->metadata_alloc_profile;
1934 data = BTRFS_BLOCK_GROUP_METADATA | alloc_profile;
1937 data = reduce_alloc_profile(root, data);
1939 * the only place that sets empty_size is btrfs_realloc_node, which
1940 * is not called recursively on allocations
1942 if (empty_size || root->ref_cows) {
1943 if (!(data & BTRFS_BLOCK_GROUP_METADATA)) {
1944 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
1946 BTRFS_BLOCK_GROUP_METADATA |
1947 (info->metadata_alloc_profile &
1948 info->avail_metadata_alloc_bits), 0);
1951 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
1952 num_bytes + 2 * 1024 * 1024, data, 0);
1956 WARN_ON(num_bytes < root->sectorsize);
1957 ret = find_free_extent(trans, root, num_bytes, empty_size,
1958 search_start, search_end, hint_byte, ins,
1959 trans->alloc_exclude_start,
1960 trans->alloc_exclude_nr, data);
1962 if (ret == -ENOSPC && num_bytes > min_alloc_size) {
1963 num_bytes = num_bytes >> 1;
1964 num_bytes = max(num_bytes, min_alloc_size);
1965 do_chunk_alloc(trans, root->fs_info->extent_root,
1966 num_bytes, data, 1);
1970 printk("allocation failed flags %Lu\n", data);
1977 /* block accounting for super block */
1978 spin_lock_irq(&info->delalloc_lock);
1979 super_used = btrfs_super_bytes_used(&info->super_copy);
1980 btrfs_set_super_bytes_used(&info->super_copy, super_used + num_bytes);
1981 spin_unlock_irq(&info->delalloc_lock);
1983 /* block accounting for root item */
1984 root_used = btrfs_root_used(&root->root_item);
1985 btrfs_set_root_used(&root->root_item, root_used + num_bytes);
1987 clear_extent_dirty(&root->fs_info->free_space_cache,
1988 ins->objectid, ins->objectid + ins->offset - 1,
1991 if (root == extent_root) {
1992 set_extent_bits(&root->fs_info->extent_ins, ins->objectid,
1993 ins->objectid + ins->offset - 1,
1994 EXTENT_LOCKED, GFP_NOFS);
1998 WARN_ON(trans->alloc_exclude_nr);
1999 trans->alloc_exclude_start = ins->objectid;
2000 trans->alloc_exclude_nr = ins->offset;
2002 memcpy(&keys[0], ins, sizeof(*ins));
2003 keys[1].offset = hash_extent_ref(root_objectid, ref_generation,
2004 owner, owner_offset);
2005 keys[1].objectid = ins->objectid;
2006 keys[1].type = BTRFS_EXTENT_REF_KEY;
2007 sizes[0] = sizeof(*extent_item);
2008 sizes[1] = sizeof(*ref);
2010 path = btrfs_alloc_path();
2013 ret = btrfs_insert_empty_items(trans, extent_root, path, keys,
2017 extent_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
2018 struct btrfs_extent_item);
2019 btrfs_set_extent_refs(path->nodes[0], extent_item, 1);
2020 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
2021 struct btrfs_extent_ref);
2023 btrfs_set_ref_root(path->nodes[0], ref, root_objectid);
2024 btrfs_set_ref_generation(path->nodes[0], ref, ref_generation);
2025 btrfs_set_ref_objectid(path->nodes[0], ref, owner);
2026 btrfs_set_ref_offset(path->nodes[0], ref, owner_offset);
2028 btrfs_mark_buffer_dirty(path->nodes[0]);
2030 trans->alloc_exclude_start = 0;
2031 trans->alloc_exclude_nr = 0;
2032 btrfs_free_path(path);
2033 finish_current_insert(trans, extent_root);
2034 pending_ret = del_pending_extents(trans, extent_root);
2044 ret = update_block_group(trans, root, ins->objectid, ins->offset, 1, 0);
2046 printk("update block group failed for %Lu %Lu\n",
2047 ins->objectid, ins->offset);
2051 maybe_unlock_mutex(root);
2055 * helper function to allocate a block for a given tree
2056 * returns the tree buffer or NULL.
2058 struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
2059 struct btrfs_root *root,
2068 struct btrfs_key ins;
2070 struct extent_buffer *buf;
2072 ret = btrfs_alloc_extent(trans, root, blocksize, blocksize,
2073 root_objectid, ref_generation,
2074 level, first_objectid, empty_size, hint,
2078 return ERR_PTR(ret);
2080 buf = btrfs_find_create_tree_block(root, ins.objectid, blocksize);
2082 btrfs_free_extent(trans, root, ins.objectid, blocksize,
2083 root->root_key.objectid, ref_generation,
2085 return ERR_PTR(-ENOMEM);
2087 btrfs_set_header_generation(buf, trans->transid);
2088 btrfs_tree_lock(buf);
2089 clean_tree_block(trans, root, buf);
2090 btrfs_set_buffer_uptodate(buf);
2092 if (PageDirty(buf->first_page)) {
2093 printk("page %lu dirty\n", buf->first_page->index);
2097 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
2098 buf->start + buf->len - 1, GFP_NOFS);
2099 if (!btrfs_test_opt(root, SSD))
2100 btrfs_set_buffer_defrag(buf);
2101 trans->blocks_used++;
2105 static int noinline drop_leaf_ref(struct btrfs_trans_handle *trans,
2106 struct btrfs_root *root,
2107 struct extent_buffer *leaf)
2110 u64 leaf_generation;
2111 struct btrfs_key key;
2112 struct btrfs_file_extent_item *fi;
2117 BUG_ON(!btrfs_is_leaf(leaf));
2118 nritems = btrfs_header_nritems(leaf);
2119 leaf_owner = btrfs_header_owner(leaf);
2120 leaf_generation = btrfs_header_generation(leaf);
2122 for (i = 0; i < nritems; i++) {
2125 btrfs_item_key_to_cpu(leaf, &key, i);
2126 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
2128 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
2129 if (btrfs_file_extent_type(leaf, fi) ==
2130 BTRFS_FILE_EXTENT_INLINE)
2133 * FIXME make sure to insert a trans record that
2134 * repeats the snapshot del on crash
2136 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
2137 if (disk_bytenr == 0)
2139 ret = __btrfs_free_extent(trans, root, disk_bytenr,
2140 btrfs_file_extent_disk_num_bytes(leaf, fi),
2141 leaf_owner, leaf_generation,
2142 key.objectid, key.offset, 0);
2148 static void noinline reada_walk_down(struct btrfs_root *root,
2149 struct extent_buffer *node,
2162 nritems = btrfs_header_nritems(node);
2163 level = btrfs_header_level(node);
2167 for (i = slot; i < nritems && skipped < 32; i++) {
2168 bytenr = btrfs_node_blockptr(node, i);
2169 if (last && ((bytenr > last && bytenr - last > 32 * 1024) ||
2170 (last > bytenr && last - bytenr > 32 * 1024))) {
2174 blocksize = btrfs_level_size(root, level - 1);
2176 ret = lookup_extent_ref(NULL, root, bytenr,
2184 mutex_unlock(&root->fs_info->alloc_mutex);
2185 ret = readahead_tree_block(root, bytenr, blocksize,
2186 btrfs_node_ptr_generation(node, i));
2187 last = bytenr + blocksize;
2189 mutex_lock(&root->fs_info->alloc_mutex);
2196 * helper function for drop_snapshot, this walks down the tree dropping ref
2197 * counts as it goes.
2199 static int noinline walk_down_tree(struct btrfs_trans_handle *trans,
2200 struct btrfs_root *root,
2201 struct btrfs_path *path, int *level)
2207 struct extent_buffer *next;
2208 struct extent_buffer *cur;
2209 struct extent_buffer *parent;
2214 mutex_lock(&root->fs_info->alloc_mutex);
2216 WARN_ON(*level < 0);
2217 WARN_ON(*level >= BTRFS_MAX_LEVEL);
2218 ret = lookup_extent_ref(trans, root,
2219 path->nodes[*level]->start,
2220 path->nodes[*level]->len, &refs);
2226 * walk down to the last node level and free all the leaves
2228 while(*level >= 0) {
2229 WARN_ON(*level < 0);
2230 WARN_ON(*level >= BTRFS_MAX_LEVEL);
2231 cur = path->nodes[*level];
2233 if (btrfs_header_level(cur) != *level)
2236 if (path->slots[*level] >=
2237 btrfs_header_nritems(cur))
2240 ret = drop_leaf_ref(trans, root, cur);
2244 bytenr = btrfs_node_blockptr(cur, path->slots[*level]);
2245 ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]);
2246 blocksize = btrfs_level_size(root, *level - 1);
2248 ret = lookup_extent_ref(trans, root, bytenr, blocksize, &refs);
2251 parent = path->nodes[*level];
2252 root_owner = btrfs_header_owner(parent);
2253 root_gen = btrfs_header_generation(parent);
2254 path->slots[*level]++;
2255 ret = __btrfs_free_extent(trans, root, bytenr,
2256 blocksize, root_owner,
2261 next = btrfs_find_tree_block(root, bytenr, blocksize);
2262 if (!next || !btrfs_buffer_uptodate(next, ptr_gen)) {
2263 free_extent_buffer(next);
2264 reada_walk_down(root, cur, path->slots[*level]);
2266 mutex_unlock(&root->fs_info->alloc_mutex);
2267 next = read_tree_block(root, bytenr, blocksize,
2269 mutex_lock(&root->fs_info->alloc_mutex);
2271 /* we've dropped the lock, double check */
2272 ret = lookup_extent_ref(trans, root, bytenr,
2276 parent = path->nodes[*level];
2277 root_owner = btrfs_header_owner(parent);
2278 root_gen = btrfs_header_generation(parent);
2280 path->slots[*level]++;
2281 free_extent_buffer(next);
2282 ret = __btrfs_free_extent(trans, root, bytenr,
2290 WARN_ON(*level <= 0);
2291 if (path->nodes[*level-1])
2292 free_extent_buffer(path->nodes[*level-1]);
2293 path->nodes[*level-1] = next;
2294 *level = btrfs_header_level(next);
2295 path->slots[*level] = 0;
2298 WARN_ON(*level < 0);
2299 WARN_ON(*level >= BTRFS_MAX_LEVEL);
2301 if (path->nodes[*level] == root->node) {
2302 root_owner = root->root_key.objectid;
2303 parent = path->nodes[*level];
2305 parent = path->nodes[*level + 1];
2306 root_owner = btrfs_header_owner(parent);
2309 root_gen = btrfs_header_generation(parent);
2310 ret = __btrfs_free_extent(trans, root, path->nodes[*level]->start,
2311 path->nodes[*level]->len,
2312 root_owner, root_gen, 0, 0, 1);
2313 free_extent_buffer(path->nodes[*level]);
2314 path->nodes[*level] = NULL;
2317 mutex_unlock(&root->fs_info->alloc_mutex);
2322 * helper for dropping snapshots. This walks back up the tree in the path
2323 * to find the first node higher up where we haven't yet gone through
2326 static int noinline walk_up_tree(struct btrfs_trans_handle *trans,
2327 struct btrfs_root *root,
2328 struct btrfs_path *path, int *level)
2332 struct btrfs_root_item *root_item = &root->root_item;
2337 for(i = *level; i < BTRFS_MAX_LEVEL - 1 && path->nodes[i]; i++) {
2338 slot = path->slots[i];
2339 if (slot < btrfs_header_nritems(path->nodes[i]) - 1) {
2340 struct extent_buffer *node;
2341 struct btrfs_disk_key disk_key;
2342 node = path->nodes[i];
2345 WARN_ON(*level == 0);
2346 btrfs_node_key(node, &disk_key, path->slots[i]);
2347 memcpy(&root_item->drop_progress,
2348 &disk_key, sizeof(disk_key));
2349 root_item->drop_level = i;
2352 if (path->nodes[*level] == root->node) {
2353 root_owner = root->root_key.objectid;
2355 btrfs_header_generation(path->nodes[*level]);
2357 struct extent_buffer *node;
2358 node = path->nodes[*level + 1];
2359 root_owner = btrfs_header_owner(node);
2360 root_gen = btrfs_header_generation(node);
2362 ret = btrfs_free_extent(trans, root,
2363 path->nodes[*level]->start,
2364 path->nodes[*level]->len,
2365 root_owner, root_gen, 0, 0, 1);
2367 free_extent_buffer(path->nodes[*level]);
2368 path->nodes[*level] = NULL;
2376 * drop the reference count on the tree rooted at 'snap'. This traverses
2377 * the tree freeing any blocks that have a ref count of zero after being
2380 int btrfs_drop_snapshot(struct btrfs_trans_handle *trans, struct btrfs_root
2386 struct btrfs_path *path;
2389 struct btrfs_root_item *root_item = &root->root_item;
2391 WARN_ON(!mutex_is_locked(&root->fs_info->drop_mutex));
2392 path = btrfs_alloc_path();
2395 level = btrfs_header_level(root->node);
2397 if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
2398 path->nodes[level] = root->node;
2399 extent_buffer_get(root->node);
2400 path->slots[level] = 0;
2402 struct btrfs_key key;
2403 struct btrfs_disk_key found_key;
2404 struct extent_buffer *node;
2406 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
2407 level = root_item->drop_level;
2408 path->lowest_level = level;
2409 wret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2414 node = path->nodes[level];
2415 btrfs_node_key(node, &found_key, path->slots[level]);
2416 WARN_ON(memcmp(&found_key, &root_item->drop_progress,
2417 sizeof(found_key)));
2418 for (i = 0; i < BTRFS_MAX_LEVEL; i++) {
2419 if (path->nodes[i] && path->locks[i]) {
2421 btrfs_tree_unlock(path->nodes[i]);
2426 wret = walk_down_tree(trans, root, path, &level);
2432 wret = walk_up_tree(trans, root, path, &level);
2440 for (i = 0; i <= orig_level; i++) {
2441 if (path->nodes[i]) {
2442 free_extent_buffer(path->nodes[i]);
2443 path->nodes[i] = NULL;
2447 btrfs_free_path(path);
2451 int btrfs_free_block_groups(struct btrfs_fs_info *info)
2458 mutex_lock(&info->alloc_mutex);
2460 ret = find_first_extent_bit(&info->block_group_cache, 0,
2461 &start, &end, (unsigned int)-1);
2464 ret = get_state_private(&info->block_group_cache, start, &ptr);
2466 kfree((void *)(unsigned long)ptr);
2467 clear_extent_bits(&info->block_group_cache, start,
2468 end, (unsigned int)-1, GFP_NOFS);
2471 ret = find_first_extent_bit(&info->free_space_cache, 0,
2472 &start, &end, EXTENT_DIRTY);
2475 clear_extent_dirty(&info->free_space_cache, start,
2478 mutex_unlock(&info->alloc_mutex);
2482 static unsigned long calc_ra(unsigned long start, unsigned long last,
2485 return min(last, start + nr - 1);
2488 static int noinline relocate_inode_pages(struct inode *inode, u64 start,
2493 unsigned long last_index;
2496 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2497 struct file_ra_state *ra;
2498 unsigned long total_read = 0;
2499 unsigned long ra_pages;
2500 struct btrfs_trans_handle *trans;
2502 ra = kzalloc(sizeof(*ra), GFP_NOFS);
2504 mutex_lock(&inode->i_mutex);
2505 i = start >> PAGE_CACHE_SHIFT;
2506 last_index = (start + len - 1) >> PAGE_CACHE_SHIFT;
2508 ra_pages = BTRFS_I(inode)->root->fs_info->bdi.ra_pages;
2510 file_ra_state_init(ra, inode->i_mapping);
2512 for (; i <= last_index; i++) {
2513 if (total_read % ra_pages == 0) {
2514 btrfs_force_ra(inode->i_mapping, ra, NULL, i,
2515 calc_ra(i, last_index, ra_pages));
2518 if (((u64)i << PAGE_CACHE_SHIFT) > inode->i_size)
2519 goto truncate_racing;
2521 page = grab_cache_page(inode->i_mapping, i);
2525 if (!PageUptodate(page)) {
2526 btrfs_readpage(NULL, page);
2528 if (!PageUptodate(page)) {
2530 page_cache_release(page);
2534 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,18)
2535 ClearPageDirty(page);
2537 cancel_dirty_page(page, PAGE_CACHE_SIZE);
2539 wait_on_page_writeback(page);
2540 set_page_extent_mapped(page);
2541 page_start = (u64)page->index << PAGE_CACHE_SHIFT;
2542 page_end = page_start + PAGE_CACHE_SIZE - 1;
2544 lock_extent(io_tree, page_start, page_end, GFP_NOFS);
2546 set_extent_delalloc(io_tree, page_start,
2547 page_end, GFP_NOFS);
2548 set_page_dirty(page);
2550 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
2552 page_cache_release(page);
2554 balance_dirty_pages_ratelimited_nr(inode->i_mapping,
2559 trans = btrfs_start_transaction(BTRFS_I(inode)->root, 1);
2561 btrfs_add_ordered_inode(inode);
2562 btrfs_end_transaction(trans, BTRFS_I(inode)->root);
2563 mark_inode_dirty(inode);
2565 mutex_unlock(&inode->i_mutex);
2569 vmtruncate(inode, inode->i_size);
2570 balance_dirty_pages_ratelimited_nr(inode->i_mapping,
2576 * The back references tell us which tree holds a ref on a block,
2577 * but it is possible for the tree root field in the reference to
2578 * reflect the original root before a snapshot was made. In this
2579 * case we should search through all the children of a given root
2580 * to find potential holders of references on a block.
2582 * Instead, we do something a little less fancy and just search
2583 * all the roots for a given key/block combination.
2585 static int find_root_for_ref(struct btrfs_root *root,
2586 struct btrfs_path *path,
2587 struct btrfs_key *key0,
2590 struct btrfs_root **found_root,
2593 struct btrfs_key root_location;
2594 struct btrfs_root *cur_root = *found_root;
2595 struct btrfs_file_extent_item *file_extent;
2596 u64 root_search_start = BTRFS_FS_TREE_OBJECTID;
2601 root_location.offset = (u64)-1;
2602 root_location.type = BTRFS_ROOT_ITEM_KEY;
2603 path->lowest_level = level;
2606 ret = btrfs_search_slot(NULL, cur_root, key0, path, 0, 0);
2608 if (ret == 0 && file_key) {
2609 struct extent_buffer *leaf = path->nodes[0];
2610 file_extent = btrfs_item_ptr(leaf, path->slots[0],
2611 struct btrfs_file_extent_item);
2612 if (btrfs_file_extent_type(leaf, file_extent) ==
2613 BTRFS_FILE_EXTENT_REG) {
2615 btrfs_file_extent_disk_bytenr(leaf,
2618 } else if (!file_key) {
2619 if (path->nodes[level])
2620 found_bytenr = path->nodes[level]->start;
2623 for (i = level; i < BTRFS_MAX_LEVEL; i++) {
2624 if (!path->nodes[i])
2626 free_extent_buffer(path->nodes[i]);
2627 path->nodes[i] = NULL;
2629 btrfs_release_path(cur_root, path);
2631 if (found_bytenr == bytenr) {
2632 *found_root = cur_root;
2636 ret = btrfs_search_root(root->fs_info->tree_root,
2637 root_search_start, &root_search_start);
2641 root_location.objectid = root_search_start;
2642 cur_root = btrfs_read_fs_root_no_name(root->fs_info,
2650 path->lowest_level = 0;
2655 * note, this releases the path
2657 static int noinline relocate_one_reference(struct btrfs_root *extent_root,
2658 struct btrfs_path *path,
2659 struct btrfs_key *extent_key,
2660 u64 *last_file_objectid,
2661 u64 *last_file_offset,
2662 u64 *last_file_root,
2665 struct inode *inode;
2666 struct btrfs_root *found_root;
2667 struct btrfs_key root_location;
2668 struct btrfs_key found_key;
2669 struct btrfs_extent_ref *ref;
2677 ref = btrfs_item_ptr(path->nodes[0], path->slots[0],
2678 struct btrfs_extent_ref);
2679 ref_root = btrfs_ref_root(path->nodes[0], ref);
2680 ref_gen = btrfs_ref_generation(path->nodes[0], ref);
2681 ref_objectid = btrfs_ref_objectid(path->nodes[0], ref);
2682 ref_offset = btrfs_ref_offset(path->nodes[0], ref);
2683 btrfs_release_path(extent_root, path);
2685 root_location.objectid = ref_root;
2687 root_location.offset = 0;
2689 root_location.offset = (u64)-1;
2690 root_location.type = BTRFS_ROOT_ITEM_KEY;
2692 found_root = btrfs_read_fs_root_no_name(extent_root->fs_info,
2694 BUG_ON(!found_root);
2696 if (ref_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
2697 found_key.objectid = ref_objectid;
2698 found_key.type = BTRFS_EXTENT_DATA_KEY;
2699 found_key.offset = ref_offset;
2702 if (last_extent == extent_key->objectid &&
2703 *last_file_objectid == ref_objectid &&
2704 *last_file_offset == ref_offset &&
2705 *last_file_root == ref_root)
2708 ret = find_root_for_ref(extent_root, path, &found_key,
2709 level, 1, &found_root,
2710 extent_key->objectid);
2715 if (last_extent == extent_key->objectid &&
2716 *last_file_objectid == ref_objectid &&
2717 *last_file_offset == ref_offset &&
2718 *last_file_root == ref_root)
2721 inode = btrfs_iget_locked(extent_root->fs_info->sb,
2722 ref_objectid, found_root);
2723 if (inode->i_state & I_NEW) {
2724 /* the inode and parent dir are two different roots */
2725 BTRFS_I(inode)->root = found_root;
2726 BTRFS_I(inode)->location.objectid = ref_objectid;
2727 BTRFS_I(inode)->location.type = BTRFS_INODE_ITEM_KEY;
2728 BTRFS_I(inode)->location.offset = 0;
2729 btrfs_read_locked_inode(inode);
2730 unlock_new_inode(inode);
2733 /* this can happen if the reference is not against
2734 * the latest version of the tree root
2736 if (is_bad_inode(inode)) {
2739 *last_file_objectid = inode->i_ino;
2740 *last_file_root = found_root->root_key.objectid;
2741 *last_file_offset = ref_offset;
2743 relocate_inode_pages(inode, ref_offset, extent_key->offset);
2746 struct btrfs_trans_handle *trans;
2747 struct extent_buffer *eb;
2750 eb = read_tree_block(found_root, extent_key->objectid,
2751 extent_key->offset, 0);
2752 btrfs_tree_lock(eb);
2753 level = btrfs_header_level(eb);
2756 btrfs_item_key_to_cpu(eb, &found_key, 0);
2758 btrfs_node_key_to_cpu(eb, &found_key, 0);
2760 btrfs_tree_unlock(eb);
2761 free_extent_buffer(eb);
2763 ret = find_root_for_ref(extent_root, path, &found_key,
2764 level, 0, &found_root,
2765 extent_key->objectid);
2770 trans = btrfs_start_transaction(found_root, 1);
2772 path->lowest_level = level;
2774 ret = btrfs_search_slot(trans, found_root, &found_key, path,
2776 path->lowest_level = 0;
2777 for (i = level; i < BTRFS_MAX_LEVEL; i++) {
2778 if (!path->nodes[i])
2780 free_extent_buffer(path->nodes[i]);
2781 path->nodes[i] = NULL;
2783 btrfs_release_path(found_root, path);
2784 if (found_root == found_root->fs_info->extent_root)
2785 btrfs_extent_post_op(trans, found_root);
2786 btrfs_end_transaction(trans, found_root);
2793 static int noinline del_extent_zero(struct btrfs_root *extent_root,
2794 struct btrfs_path *path,
2795 struct btrfs_key *extent_key)
2798 struct btrfs_trans_handle *trans;
2800 trans = btrfs_start_transaction(extent_root, 1);
2801 ret = btrfs_search_slot(trans, extent_root, extent_key, path, -1, 1);
2808 ret = btrfs_del_item(trans, extent_root, path);
2810 btrfs_end_transaction(trans, extent_root);
2814 static int noinline relocate_one_extent(struct btrfs_root *extent_root,
2815 struct btrfs_path *path,
2816 struct btrfs_key *extent_key)
2818 struct btrfs_key key;
2819 struct btrfs_key found_key;
2820 struct extent_buffer *leaf;
2821 u64 last_file_objectid = 0;
2822 u64 last_file_root = 0;
2823 u64 last_file_offset = (u64)-1;
2824 u64 last_extent = 0;
2829 if (extent_key->objectid == 0) {
2830 ret = del_extent_zero(extent_root, path, extent_key);
2833 key.objectid = extent_key->objectid;
2834 key.type = BTRFS_EXTENT_REF_KEY;
2838 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
2844 leaf = path->nodes[0];
2845 nritems = btrfs_header_nritems(leaf);
2846 if (path->slots[0] == nritems) {
2847 ret = btrfs_next_leaf(extent_root, path);
2854 leaf = path->nodes[0];
2857 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2858 if (found_key.objectid != extent_key->objectid) {
2862 if (found_key.type != BTRFS_EXTENT_REF_KEY) {
2866 key.offset = found_key.offset + 1;
2867 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2869 ret = relocate_one_reference(extent_root, path, extent_key,
2870 &last_file_objectid,
2872 &last_file_root, last_extent);
2875 last_extent = extent_key->objectid;
2879 btrfs_release_path(extent_root, path);
2883 static u64 update_block_group_flags(struct btrfs_root *root, u64 flags)
2886 u64 stripped = BTRFS_BLOCK_GROUP_RAID0 |
2887 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
2889 num_devices = root->fs_info->fs_devices->num_devices;
2890 if (num_devices == 1) {
2891 stripped |= BTRFS_BLOCK_GROUP_DUP;
2892 stripped = flags & ~stripped;
2894 /* turn raid0 into single device chunks */
2895 if (flags & BTRFS_BLOCK_GROUP_RAID0)
2898 /* turn mirroring into duplication */
2899 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
2900 BTRFS_BLOCK_GROUP_RAID10))
2901 return stripped | BTRFS_BLOCK_GROUP_DUP;
2904 /* they already had raid on here, just return */
2905 if (flags & stripped)
2908 stripped |= BTRFS_BLOCK_GROUP_DUP;
2909 stripped = flags & ~stripped;
2911 /* switch duplicated blocks with raid1 */
2912 if (flags & BTRFS_BLOCK_GROUP_DUP)
2913 return stripped | BTRFS_BLOCK_GROUP_RAID1;
2915 /* turn single device chunks into raid0 */
2916 return stripped | BTRFS_BLOCK_GROUP_RAID0;
2921 int __alloc_chunk_for_shrink(struct btrfs_root *root,
2922 struct btrfs_block_group_cache *shrink_block_group,
2925 struct btrfs_trans_handle *trans;
2926 u64 new_alloc_flags;
2929 if (btrfs_block_group_used(&shrink_block_group->item) > 0) {
2931 trans = btrfs_start_transaction(root, 1);
2932 new_alloc_flags = update_block_group_flags(root,
2933 shrink_block_group->flags);
2934 if (new_alloc_flags != shrink_block_group->flags) {
2936 btrfs_block_group_used(&shrink_block_group->item);
2938 calc = shrink_block_group->key.offset;
2940 do_chunk_alloc(trans, root->fs_info->extent_root,
2941 calc + 2 * 1024 * 1024, new_alloc_flags, force);
2942 btrfs_end_transaction(trans, root);
2947 int btrfs_shrink_extent_tree(struct btrfs_root *root, u64 shrink_start)
2949 struct btrfs_trans_handle *trans;
2950 struct btrfs_root *tree_root = root->fs_info->tree_root;
2951 struct btrfs_path *path;
2954 u64 shrink_last_byte;
2955 struct btrfs_block_group_cache *shrink_block_group;
2956 struct btrfs_fs_info *info = root->fs_info;
2957 struct btrfs_key key;
2958 struct btrfs_key found_key;
2959 struct extent_buffer *leaf;
2964 mutex_lock(&root->fs_info->alloc_mutex);
2965 shrink_block_group = btrfs_lookup_block_group(root->fs_info,
2967 BUG_ON(!shrink_block_group);
2969 shrink_last_byte = shrink_block_group->key.objectid +
2970 shrink_block_group->key.offset;
2972 shrink_block_group->space_info->total_bytes -=
2973 shrink_block_group->key.offset;
2974 path = btrfs_alloc_path();
2975 root = root->fs_info->extent_root;
2978 printk("btrfs relocating block group %llu flags %llu\n",
2979 (unsigned long long)shrink_start,
2980 (unsigned long long)shrink_block_group->flags);
2982 __alloc_chunk_for_shrink(root, shrink_block_group, 1);
2986 shrink_block_group->ro = 1;
2990 key.objectid = shrink_start;
2993 cur_byte = key.objectid;
2995 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2999 ret = btrfs_previous_item(root, path, 0, BTRFS_EXTENT_ITEM_KEY);
3004 leaf = path->nodes[0];
3005 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3006 if (found_key.objectid + found_key.offset > shrink_start &&
3007 found_key.objectid < shrink_last_byte) {
3008 cur_byte = found_key.objectid;
3009 key.objectid = cur_byte;
3012 btrfs_release_path(root, path);
3015 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3019 leaf = path->nodes[0];
3020 nritems = btrfs_header_nritems(leaf);
3022 if (path->slots[0] >= nritems) {
3023 ret = btrfs_next_leaf(root, path);
3030 leaf = path->nodes[0];
3031 nritems = btrfs_header_nritems(leaf);
3034 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3036 if (found_key.objectid >= shrink_last_byte)
3039 if (progress && need_resched()) {
3040 memcpy(&key, &found_key, sizeof(key));
3042 btrfs_release_path(root, path);
3043 btrfs_search_slot(NULL, root, &key, path, 0, 0);
3049 if (btrfs_key_type(&found_key) != BTRFS_EXTENT_ITEM_KEY ||
3050 found_key.objectid + found_key.offset <= cur_byte) {
3051 memcpy(&key, &found_key, sizeof(key));
3058 cur_byte = found_key.objectid + found_key.offset;
3059 key.objectid = cur_byte;
3060 btrfs_release_path(root, path);
3061 ret = relocate_one_extent(root, path, &found_key);
3062 __alloc_chunk_for_shrink(root, shrink_block_group, 0);
3065 btrfs_release_path(root, path);
3067 if (total_found > 0) {
3068 printk("btrfs relocate found %llu last extent was %llu\n",
3069 (unsigned long long)total_found,
3070 (unsigned long long)found_key.objectid);
3071 trans = btrfs_start_transaction(tree_root, 1);
3072 btrfs_commit_transaction(trans, tree_root);
3074 btrfs_clean_old_snapshots(tree_root);
3076 trans = btrfs_start_transaction(tree_root, 1);
3077 btrfs_commit_transaction(trans, tree_root);
3082 * we've freed all the extents, now remove the block
3083 * group item from the tree
3085 trans = btrfs_start_transaction(root, 1);
3086 memcpy(&key, &shrink_block_group->key, sizeof(key));
3088 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3094 clear_extent_bits(&info->block_group_cache, key.objectid,
3095 key.objectid + key.offset - 1,
3096 (unsigned int)-1, GFP_NOFS);
3099 clear_extent_bits(&info->free_space_cache,
3100 key.objectid, key.objectid + key.offset - 1,
3101 (unsigned int)-1, GFP_NOFS);
3103 memset(shrink_block_group, 0, sizeof(*shrink_block_group));
3104 kfree(shrink_block_group);
3106 btrfs_del_item(trans, root, path);
3107 btrfs_commit_transaction(trans, root);
3109 /* the code to unpin extents might set a few bits in the free
3110 * space cache for this range again
3112 clear_extent_bits(&info->free_space_cache,
3113 key.objectid, key.objectid + key.offset - 1,
3114 (unsigned int)-1, GFP_NOFS);
3116 btrfs_free_path(path);
3117 mutex_unlock(&root->fs_info->alloc_mutex);
3121 int find_first_block_group(struct btrfs_root *root, struct btrfs_path *path,
3122 struct btrfs_key *key)
3125 struct btrfs_key found_key;
3126 struct extent_buffer *leaf;
3129 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
3134 slot = path->slots[0];
3135 leaf = path->nodes[0];
3136 if (slot >= btrfs_header_nritems(leaf)) {
3137 ret = btrfs_next_leaf(root, path);
3144 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3146 if (found_key.objectid >= key->objectid &&
3147 found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
3158 int btrfs_read_block_groups(struct btrfs_root *root)
3160 struct btrfs_path *path;
3163 struct btrfs_block_group_cache *cache;
3164 struct btrfs_fs_info *info = root->fs_info;
3165 struct btrfs_space_info *space_info;
3166 struct extent_io_tree *block_group_cache;
3167 struct btrfs_key key;
3168 struct btrfs_key found_key;
3169 struct extent_buffer *leaf;
3171 block_group_cache = &info->block_group_cache;
3172 root = info->extent_root;
3175 btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY);
3176 path = btrfs_alloc_path();
3180 mutex_lock(&root->fs_info->alloc_mutex);
3182 ret = find_first_block_group(root, path, &key);
3190 leaf = path->nodes[0];
3191 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3192 cache = kzalloc(sizeof(*cache), GFP_NOFS);
3198 read_extent_buffer(leaf, &cache->item,
3199 btrfs_item_ptr_offset(leaf, path->slots[0]),
3200 sizeof(cache->item));
3201 memcpy(&cache->key, &found_key, sizeof(found_key));
3203 key.objectid = found_key.objectid + found_key.offset;
3204 btrfs_release_path(root, path);
3205 cache->flags = btrfs_block_group_flags(&cache->item);
3207 if (cache->flags & BTRFS_BLOCK_GROUP_DATA) {
3208 bit = BLOCK_GROUP_DATA;
3209 } else if (cache->flags & BTRFS_BLOCK_GROUP_SYSTEM) {
3210 bit = BLOCK_GROUP_SYSTEM;
3211 } else if (cache->flags & BTRFS_BLOCK_GROUP_METADATA) {
3212 bit = BLOCK_GROUP_METADATA;
3214 set_avail_alloc_bits(info, cache->flags);
3216 ret = update_space_info(info, cache->flags, found_key.offset,
3217 btrfs_block_group_used(&cache->item),
3220 cache->space_info = space_info;
3222 /* use EXTENT_LOCKED to prevent merging */
3223 set_extent_bits(block_group_cache, found_key.objectid,
3224 found_key.objectid + found_key.offset - 1,
3225 bit | EXTENT_LOCKED, GFP_NOFS);
3226 set_state_private(block_group_cache, found_key.objectid,
3227 (unsigned long)cache);
3230 btrfs_super_total_bytes(&info->super_copy))
3235 btrfs_free_path(path);
3236 mutex_unlock(&root->fs_info->alloc_mutex);
3240 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
3241 struct btrfs_root *root, u64 bytes_used,
3242 u64 type, u64 chunk_objectid, u64 chunk_offset,
3247 struct btrfs_root *extent_root;
3248 struct btrfs_block_group_cache *cache;
3249 struct extent_io_tree *block_group_cache;
3251 extent_root = root->fs_info->extent_root;
3252 block_group_cache = &root->fs_info->block_group_cache;
3254 cache = kzalloc(sizeof(*cache), GFP_NOFS);
3256 cache->key.objectid = chunk_offset;
3257 cache->key.offset = size;
3258 btrfs_set_key_type(&cache->key, BTRFS_BLOCK_GROUP_ITEM_KEY);
3260 btrfs_set_block_group_used(&cache->item, bytes_used);
3261 btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
3262 cache->flags = type;
3263 btrfs_set_block_group_flags(&cache->item, type);
3265 ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
3266 &cache->space_info);
3269 bit = block_group_state_bits(type);
3270 set_extent_bits(block_group_cache, chunk_offset,
3271 chunk_offset + size - 1,
3272 bit | EXTENT_LOCKED, GFP_NOFS);
3274 set_state_private(block_group_cache, chunk_offset,
3275 (unsigned long)cache);
3276 ret = btrfs_insert_item(trans, extent_root, &cache->key, &cache->item,
3277 sizeof(cache->item));
3280 finish_current_insert(trans, extent_root);
3281 ret = del_pending_extents(trans, extent_root);
3283 set_avail_alloc_bits(extent_root->fs_info, type);