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.
19 #include <linux/sched.h>
20 #include <linux/crc32c.h>
21 #include <linux/pagemap.h>
25 #include "print-tree.h"
26 #include "transaction.h"
29 #define BLOCK_GROUP_DATA EXTENT_WRITEBACK
30 #define BLOCK_GROUP_METADATA EXTENT_UPTODATE
31 #define BLOCK_GROUP_SYSTEM EXTENT_NEW
33 #define BLOCK_GROUP_DIRTY EXTENT_DIRTY
35 static int finish_current_insert(struct btrfs_trans_handle *trans, struct
36 btrfs_root *extent_root);
37 static int del_pending_extents(struct btrfs_trans_handle *trans, struct
38 btrfs_root *extent_root);
39 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
40 struct btrfs_root *root, u64 bytes_used,
41 u64 type, u64 chunk_tree, u64 chunk_objectid,
45 static int cache_block_group(struct btrfs_root *root,
46 struct btrfs_block_group_cache *block_group)
48 struct btrfs_path *path;
51 struct extent_buffer *leaf;
52 struct extent_io_tree *free_space_cache;
62 root = root->fs_info->extent_root;
63 free_space_cache = &root->fs_info->free_space_cache;
65 if (block_group->cached)
68 path = btrfs_alloc_path();
73 first_free = block_group->key.objectid;
74 key.objectid = block_group->key.objectid;
76 btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
77 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
80 ret = btrfs_previous_item(root, path, 0, BTRFS_EXTENT_ITEM_KEY);
84 leaf = path->nodes[0];
85 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
86 if (key.objectid + key.offset > first_free)
87 first_free = key.objectid + key.offset;
90 leaf = path->nodes[0];
91 slot = path->slots[0];
92 if (slot >= btrfs_header_nritems(leaf)) {
93 ret = btrfs_next_leaf(root, path);
102 btrfs_item_key_to_cpu(leaf, &key, slot);
103 if (key.objectid < block_group->key.objectid) {
106 if (key.objectid >= block_group->key.objectid +
107 block_group->key.offset) {
111 if (btrfs_key_type(&key) == BTRFS_EXTENT_ITEM_KEY) {
116 if (key.objectid > last) {
117 hole_size = key.objectid - last;
118 set_extent_dirty(free_space_cache, last,
119 last + hole_size - 1,
122 last = key.objectid + key.offset;
130 if (block_group->key.objectid +
131 block_group->key.offset > last) {
132 hole_size = block_group->key.objectid +
133 block_group->key.offset - last;
134 set_extent_dirty(free_space_cache, last,
135 last + hole_size - 1, GFP_NOFS);
137 block_group->cached = 1;
139 btrfs_free_path(path);
143 struct btrfs_block_group_cache *btrfs_lookup_block_group(struct
147 struct extent_io_tree *block_group_cache;
148 struct btrfs_block_group_cache *block_group = NULL;
154 block_group_cache = &info->block_group_cache;
155 ret = find_first_extent_bit(block_group_cache,
156 bytenr, &start, &end,
157 BLOCK_GROUP_DATA | BLOCK_GROUP_METADATA |
162 ret = get_state_private(block_group_cache, start, &ptr);
166 block_group = (struct btrfs_block_group_cache *)(unsigned long)ptr;
167 if (block_group->key.objectid <= bytenr && bytenr <
168 block_group->key.objectid + block_group->key.offset)
173 static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits)
175 return (cache->flags & bits);
178 static int noinline find_search_start(struct btrfs_root *root,
179 struct btrfs_block_group_cache **cache_ret,
180 u64 *start_ret, int num, int data)
183 struct btrfs_block_group_cache *cache = *cache_ret;
184 struct extent_io_tree *free_space_cache;
185 struct extent_state *state;
190 u64 search_start = *start_ret;
195 total_fs_bytes = btrfs_super_total_bytes(&root->fs_info->super_copy);
196 free_space_cache = &root->fs_info->free_space_cache;
199 ret = cache_block_group(root, cache);
203 last = max(search_start, cache->key.objectid);
204 if (!block_group_bits(cache, data)) {
208 spin_lock_irq(&free_space_cache->lock);
209 state = find_first_extent_bit_state(free_space_cache, last, EXTENT_DIRTY);
214 spin_unlock_irq(&free_space_cache->lock);
218 start = max(last, state->start);
219 last = state->end + 1;
220 if (last - start < num) {
221 if (last == cache->key.objectid + cache->key.offset)
224 state = extent_state_next(state);
225 } while(state && !(state->state & EXTENT_DIRTY));
228 spin_unlock_irq(&free_space_cache->lock);
229 if (start + num > cache->key.objectid + cache->key.offset)
231 if (start + num > total_fs_bytes)
236 cache = btrfs_lookup_block_group(root->fs_info, search_start);
238 printk("Unable to find block group for %Lu\n", search_start);
244 last = cache->key.objectid + cache->key.offset;
246 cache = btrfs_lookup_block_group(root->fs_info, last);
247 if (!cache || cache->key.objectid >= total_fs_bytes) {
256 if (cache_miss && !cache->cached) {
257 cache_block_group(root, cache);
259 cache = btrfs_lookup_block_group(root->fs_info, last);
261 cache = btrfs_find_block_group(root, cache, last, data, 0);
269 static u64 div_factor(u64 num, int factor)
278 static int block_group_state_bits(u64 flags)
281 if (flags & BTRFS_BLOCK_GROUP_DATA)
282 bits |= BLOCK_GROUP_DATA;
283 if (flags & BTRFS_BLOCK_GROUP_METADATA)
284 bits |= BLOCK_GROUP_METADATA;
285 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
286 bits |= BLOCK_GROUP_SYSTEM;
290 struct btrfs_block_group_cache *btrfs_find_block_group(struct btrfs_root *root,
291 struct btrfs_block_group_cache
292 *hint, u64 search_start,
295 struct btrfs_block_group_cache *cache;
296 struct extent_io_tree *block_group_cache;
297 struct btrfs_block_group_cache *found_group = NULL;
298 struct btrfs_fs_info *info = root->fs_info;
312 block_group_cache = &info->block_group_cache;
313 total_fs_bytes = btrfs_super_total_bytes(&root->fs_info->super_copy);
318 bit = block_group_state_bits(data);
320 if (search_start && search_start < total_fs_bytes) {
321 struct btrfs_block_group_cache *shint;
322 shint = btrfs_lookup_block_group(info, search_start);
323 if (shint && block_group_bits(shint, data)) {
324 used = btrfs_block_group_used(&shint->item);
325 if (used + shint->pinned <
326 div_factor(shint->key.offset, factor)) {
331 if (hint && block_group_bits(hint, data) &&
332 hint->key.objectid < total_fs_bytes) {
333 used = btrfs_block_group_used(&hint->item);
334 if (used + hint->pinned <
335 div_factor(hint->key.offset, factor)) {
338 last = hint->key.objectid + hint->key.offset;
342 hint_last = max(hint->key.objectid, search_start);
344 hint_last = search_start;
346 if (hint_last >= total_fs_bytes)
347 hint_last = search_start;
352 ret = find_first_extent_bit(block_group_cache, last,
357 ret = get_state_private(block_group_cache, start, &ptr);
361 cache = (struct btrfs_block_group_cache *)(unsigned long)ptr;
362 last = cache->key.objectid + cache->key.offset;
363 used = btrfs_block_group_used(&cache->item);
365 if (cache->key.objectid > total_fs_bytes)
369 free_check = cache->key.offset;
371 free_check = div_factor(cache->key.offset, factor);
373 if (used + cache->pinned < free_check) {
378 printk("failed on cache %Lu used %Lu total %Lu\n",
379 cache->key.objectid, used, cache->key.offset);
392 static u64 hash_extent_ref(u64 root_objectid, u64 ref_generation,
393 u64 owner, u64 owner_offset)
395 u32 high_crc = ~(u32)0;
396 u32 low_crc = ~(u32)0;
399 lenum = cpu_to_le64(root_objectid);
400 high_crc = crc32c(high_crc, &lenum, sizeof(lenum));
401 lenum = cpu_to_le64(ref_generation);
402 low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
403 if (owner >= BTRFS_FIRST_FREE_OBJECTID) {
404 lenum = cpu_to_le64(owner);
405 low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
406 lenum = cpu_to_le64(owner_offset);
407 low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
409 return ((u64)high_crc << 32) | (u64)low_crc;
412 static int match_extent_ref(struct extent_buffer *leaf,
413 struct btrfs_extent_ref *disk_ref,
414 struct btrfs_extent_ref *cpu_ref)
419 if (cpu_ref->objectid)
420 len = sizeof(*cpu_ref);
422 len = 2 * sizeof(u64);
423 ret = memcmp_extent_buffer(leaf, cpu_ref, (unsigned long)disk_ref,
428 static int noinline lookup_extent_backref(struct btrfs_trans_handle *trans,
429 struct btrfs_root *root,
430 struct btrfs_path *path, u64 bytenr,
432 u64 ref_generation, u64 owner,
433 u64 owner_offset, int del)
436 struct btrfs_key key;
437 struct btrfs_key found_key;
438 struct btrfs_extent_ref ref;
439 struct extent_buffer *leaf;
440 struct btrfs_extent_ref *disk_ref;
444 btrfs_set_stack_ref_root(&ref, root_objectid);
445 btrfs_set_stack_ref_generation(&ref, ref_generation);
446 btrfs_set_stack_ref_objectid(&ref, owner);
447 btrfs_set_stack_ref_offset(&ref, owner_offset);
449 hash = hash_extent_ref(root_objectid, ref_generation, owner,
452 key.objectid = bytenr;
453 key.type = BTRFS_EXTENT_REF_KEY;
456 ret = btrfs_search_slot(trans, root, &key, path,
460 leaf = path->nodes[0];
462 u32 nritems = btrfs_header_nritems(leaf);
463 if (path->slots[0] >= nritems) {
464 ret2 = btrfs_next_leaf(root, path);
467 leaf = path->nodes[0];
469 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
470 if (found_key.objectid != bytenr ||
471 found_key.type != BTRFS_EXTENT_REF_KEY)
473 key.offset = found_key.offset;
475 btrfs_release_path(root, path);
479 disk_ref = btrfs_item_ptr(path->nodes[0],
481 struct btrfs_extent_ref);
482 if (match_extent_ref(path->nodes[0], disk_ref, &ref)) {
486 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
487 key.offset = found_key.offset + 1;
488 btrfs_release_path(root, path);
495 * Back reference rules. Back refs have three main goals:
497 * 1) differentiate between all holders of references to an extent so that
498 * when a reference is dropped we can make sure it was a valid reference
499 * before freeing the extent.
501 * 2) Provide enough information to quickly find the holders of an extent
502 * if we notice a given block is corrupted or bad.
504 * 3) Make it easy to migrate blocks for FS shrinking or storage pool
505 * maintenance. This is actually the same as #2, but with a slightly
506 * different use case.
508 * File extents can be referenced by:
510 * - multiple snapshots, subvolumes, or different generations in one subvol
511 * - different files inside a single subvolume (in theory, not implemented yet)
512 * - different offsets inside a file (bookend extents in file.c)
514 * The extent ref structure has fields for:
516 * - Objectid of the subvolume root
517 * - Generation number of the tree holding the reference
518 * - objectid of the file holding the reference
519 * - offset in the file corresponding to the key holding the reference
521 * When a file extent is allocated the fields are filled in:
522 * (root_key.objectid, trans->transid, inode objectid, offset in file)
524 * When a leaf is cow'd new references are added for every file extent found
525 * in the leaf. It looks the same as the create case, but trans->transid
526 * will be different when the block is cow'd.
528 * (root_key.objectid, trans->transid, inode objectid, offset in file)
530 * When a file extent is removed either during snapshot deletion or file
531 * truncation, the corresponding back reference is found
534 * (btrfs_header_owner(leaf), btrfs_header_generation(leaf),
535 * inode objectid, offset in file)
537 * Btree extents can be referenced by:
539 * - Different subvolumes
540 * - Different generations of the same subvolume
542 * Storing sufficient information for a full reverse mapping of a btree
543 * block would require storing the lowest key of the block in the backref,
544 * and it would require updating that lowest key either before write out or
545 * every time it changed. Instead, the objectid of the lowest key is stored
546 * along with the level of the tree block. This provides a hint
547 * about where in the btree the block can be found. Searches through the
548 * btree only need to look for a pointer to that block, so they stop one
549 * level higher than the level recorded in the backref.
551 * Some btrees do not do reference counting on their extents. These
552 * include the extent tree and the tree of tree roots. Backrefs for these
553 * trees always have a generation of zero.
555 * When a tree block is created, back references are inserted:
557 * (root->root_key.objectid, trans->transid or zero, level, lowest_key_objectid)
559 * When a tree block is cow'd in a reference counted root,
560 * new back references are added for all the blocks it points to.
561 * These are of the form (trans->transid will have increased since creation):
563 * (root->root_key.objectid, trans->transid, level, lowest_key_objectid)
565 * Because the lowest_key_objectid and the level are just hints
566 * they are not used when backrefs are deleted. When a backref is deleted:
568 * if backref was for a tree root:
569 * root_objectid = root->root_key.objectid
571 * root_objectid = btrfs_header_owner(parent)
573 * (root_objectid, btrfs_header_generation(parent) or zero, 0, 0)
575 * Back Reference Key hashing:
577 * Back references have four fields, each 64 bits long. Unfortunately,
578 * This is hashed into a single 64 bit number and placed into the key offset.
579 * The key objectid corresponds to the first byte in the extent, and the
580 * key type is set to BTRFS_EXTENT_REF_KEY
582 int btrfs_insert_extent_backref(struct btrfs_trans_handle *trans,
583 struct btrfs_root *root,
584 struct btrfs_path *path, u64 bytenr,
585 u64 root_objectid, u64 ref_generation,
586 u64 owner, u64 owner_offset)
589 struct btrfs_key key;
590 struct btrfs_extent_ref ref;
591 struct btrfs_extent_ref *disk_ref;
594 btrfs_set_stack_ref_root(&ref, root_objectid);
595 btrfs_set_stack_ref_generation(&ref, ref_generation);
596 btrfs_set_stack_ref_objectid(&ref, owner);
597 btrfs_set_stack_ref_offset(&ref, owner_offset);
599 hash = hash_extent_ref(root_objectid, ref_generation, owner,
602 key.objectid = bytenr;
603 key.type = BTRFS_EXTENT_REF_KEY;
605 ret = btrfs_insert_empty_item(trans, root, path, &key, sizeof(ref));
606 while (ret == -EEXIST) {
607 disk_ref = btrfs_item_ptr(path->nodes[0], path->slots[0],
608 struct btrfs_extent_ref);
609 if (match_extent_ref(path->nodes[0], disk_ref, &ref))
612 btrfs_release_path(root, path);
613 ret = btrfs_insert_empty_item(trans, root, path, &key,
618 disk_ref = btrfs_item_ptr(path->nodes[0], path->slots[0],
619 struct btrfs_extent_ref);
620 write_extent_buffer(path->nodes[0], &ref, (unsigned long)disk_ref,
622 btrfs_mark_buffer_dirty(path->nodes[0]);
624 btrfs_release_path(root, path);
628 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
629 struct btrfs_root *root,
630 u64 bytenr, u64 num_bytes,
631 u64 root_objectid, u64 ref_generation,
632 u64 owner, u64 owner_offset)
634 struct btrfs_path *path;
636 struct btrfs_key key;
637 struct extent_buffer *l;
638 struct btrfs_extent_item *item;
641 WARN_ON(num_bytes < root->sectorsize);
642 path = btrfs_alloc_path();
647 key.objectid = bytenr;
648 btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
649 key.offset = num_bytes;
650 ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key, path,
659 item = btrfs_item_ptr(l, path->slots[0], struct btrfs_extent_item);
660 refs = btrfs_extent_refs(l, item);
661 btrfs_set_extent_refs(l, item, refs + 1);
662 btrfs_mark_buffer_dirty(path->nodes[0]);
664 btrfs_release_path(root->fs_info->extent_root, path);
667 ret = btrfs_insert_extent_backref(trans, root->fs_info->extent_root,
668 path, bytenr, root_objectid,
669 ref_generation, owner, owner_offset);
671 finish_current_insert(trans, root->fs_info->extent_root);
672 del_pending_extents(trans, root->fs_info->extent_root);
674 btrfs_free_path(path);
678 int btrfs_extent_post_op(struct btrfs_trans_handle *trans,
679 struct btrfs_root *root)
681 finish_current_insert(trans, root->fs_info->extent_root);
682 del_pending_extents(trans, root->fs_info->extent_root);
686 static int lookup_extent_ref(struct btrfs_trans_handle *trans,
687 struct btrfs_root *root, u64 bytenr,
688 u64 num_bytes, u32 *refs)
690 struct btrfs_path *path;
692 struct btrfs_key key;
693 struct extent_buffer *l;
694 struct btrfs_extent_item *item;
696 WARN_ON(num_bytes < root->sectorsize);
697 path = btrfs_alloc_path();
699 key.objectid = bytenr;
700 key.offset = num_bytes;
701 btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
702 ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key, path,
707 btrfs_print_leaf(root, path->nodes[0]);
708 printk("failed to find block number %Lu\n", bytenr);
712 item = btrfs_item_ptr(l, path->slots[0], struct btrfs_extent_item);
713 *refs = btrfs_extent_refs(l, item);
715 btrfs_free_path(path);
719 u32 btrfs_count_snapshots_in_path(struct btrfs_root *root,
720 struct btrfs_path *count_path,
723 struct btrfs_root *extent_root = root->fs_info->extent_root;
724 struct btrfs_path *path;
727 u64 root_objectid = root->root_key.objectid;
732 struct btrfs_key key;
733 struct btrfs_key found_key;
734 struct extent_buffer *l;
735 struct btrfs_extent_item *item;
736 struct btrfs_extent_ref *ref_item;
739 path = btrfs_alloc_path();
742 bytenr = first_extent;
744 bytenr = count_path->nodes[level]->start;
747 key.objectid = bytenr;
750 btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
751 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
757 btrfs_item_key_to_cpu(l, &found_key, path->slots[0]);
759 if (found_key.objectid != bytenr ||
760 found_key.type != BTRFS_EXTENT_ITEM_KEY) {
764 item = btrfs_item_ptr(l, path->slots[0], struct btrfs_extent_item);
767 nritems = btrfs_header_nritems(l);
768 if (path->slots[0] >= nritems) {
769 ret = btrfs_next_leaf(extent_root, path);
774 btrfs_item_key_to_cpu(l, &found_key, path->slots[0]);
775 if (found_key.objectid != bytenr)
778 if (found_key.type != BTRFS_EXTENT_REF_KEY) {
784 ref_item = btrfs_item_ptr(l, path->slots[0],
785 struct btrfs_extent_ref);
786 found_objectid = btrfs_ref_root(l, ref_item);
788 if (found_objectid != root_objectid) {
795 if (cur_count == 0) {
799 if (level >= 0 && root->node == count_path->nodes[level])
802 btrfs_release_path(root, path);
806 btrfs_free_path(path);
809 int btrfs_inc_root_ref(struct btrfs_trans_handle *trans,
810 struct btrfs_root *root, u64 owner_objectid)
816 struct btrfs_disk_key disk_key;
818 level = btrfs_header_level(root->node);
819 generation = trans->transid;
820 nritems = btrfs_header_nritems(root->node);
823 btrfs_item_key(root->node, &disk_key, 0);
825 btrfs_node_key(root->node, &disk_key, 0);
826 key_objectid = btrfs_disk_key_objectid(&disk_key);
830 return btrfs_inc_extent_ref(trans, root, root->node->start,
831 root->node->len, owner_objectid,
832 generation, level, key_objectid);
835 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
836 struct extent_buffer *buf)
840 struct btrfs_key key;
841 struct btrfs_file_extent_item *fi;
850 level = btrfs_header_level(buf);
851 nritems = btrfs_header_nritems(buf);
852 for (i = 0; i < nritems; i++) {
855 btrfs_item_key_to_cpu(buf, &key, i);
856 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
858 fi = btrfs_item_ptr(buf, i,
859 struct btrfs_file_extent_item);
860 if (btrfs_file_extent_type(buf, fi) ==
861 BTRFS_FILE_EXTENT_INLINE)
863 disk_bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
864 if (disk_bytenr == 0)
866 ret = btrfs_inc_extent_ref(trans, root, disk_bytenr,
867 btrfs_file_extent_disk_num_bytes(buf, fi),
868 root->root_key.objectid, trans->transid,
869 key.objectid, key.offset);
875 bytenr = btrfs_node_blockptr(buf, i);
876 btrfs_node_key_to_cpu(buf, &key, i);
877 ret = btrfs_inc_extent_ref(trans, root, bytenr,
878 btrfs_level_size(root, level - 1),
879 root->root_key.objectid,
881 level - 1, key.objectid);
892 for (i =0; i < faili; i++) {
895 btrfs_item_key_to_cpu(buf, &key, i);
896 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
898 fi = btrfs_item_ptr(buf, i,
899 struct btrfs_file_extent_item);
900 if (btrfs_file_extent_type(buf, fi) ==
901 BTRFS_FILE_EXTENT_INLINE)
903 disk_bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
904 if (disk_bytenr == 0)
906 err = btrfs_free_extent(trans, root, disk_bytenr,
907 btrfs_file_extent_disk_num_bytes(buf,
911 bytenr = btrfs_node_blockptr(buf, i);
912 err = btrfs_free_extent(trans, root, bytenr,
913 btrfs_level_size(root, level - 1), 0);
921 static int write_one_cache_group(struct btrfs_trans_handle *trans,
922 struct btrfs_root *root,
923 struct btrfs_path *path,
924 struct btrfs_block_group_cache *cache)
928 struct btrfs_root *extent_root = root->fs_info->extent_root;
930 struct extent_buffer *leaf;
932 ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
937 leaf = path->nodes[0];
938 bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
939 write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
940 btrfs_mark_buffer_dirty(leaf);
941 btrfs_release_path(extent_root, path);
943 finish_current_insert(trans, extent_root);
944 pending_ret = del_pending_extents(trans, extent_root);
953 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
954 struct btrfs_root *root)
956 struct extent_io_tree *block_group_cache;
957 struct btrfs_block_group_cache *cache;
961 struct btrfs_path *path;
967 block_group_cache = &root->fs_info->block_group_cache;
968 path = btrfs_alloc_path();
973 ret = find_first_extent_bit(block_group_cache, last,
974 &start, &end, BLOCK_GROUP_DIRTY);
979 ret = get_state_private(block_group_cache, start, &ptr);
983 cache = (struct btrfs_block_group_cache *)(unsigned long)ptr;
984 err = write_one_cache_group(trans, root,
987 * if we fail to write the cache group, we want
988 * to keep it marked dirty in hopes that a later
995 clear_extent_bits(block_group_cache, start, end,
996 BLOCK_GROUP_DIRTY, GFP_NOFS);
998 btrfs_free_path(path);
1002 static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info,
1005 struct list_head *head = &info->space_info;
1006 struct list_head *cur;
1007 struct btrfs_space_info *found;
1008 list_for_each(cur, head) {
1009 found = list_entry(cur, struct btrfs_space_info, list);
1010 if (found->flags == flags)
1017 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
1018 struct btrfs_root *extent_root, u64 alloc_bytes,
1021 struct btrfs_space_info *space_info;
1027 space_info = __find_space_info(extent_root->fs_info, flags);
1028 BUG_ON(!space_info);
1030 if (space_info->full)
1033 thresh = div_factor(space_info->total_bytes, 7);
1034 if ((space_info->bytes_used + space_info->bytes_pinned + alloc_bytes) <
1038 ret = btrfs_alloc_chunk(trans, extent_root, &start, &num_bytes, flags);
1039 if (ret == -ENOSPC) {
1040 printk("space info full %Lu\n", flags);
1041 space_info->full = 1;
1047 ret = btrfs_make_block_group(trans, extent_root, 0, flags,
1048 extent_root->fs_info->chunk_root->root_key.objectid,
1054 static int update_block_group(struct btrfs_trans_handle *trans,
1055 struct btrfs_root *root,
1056 u64 bytenr, u64 num_bytes, int alloc,
1059 struct btrfs_block_group_cache *cache;
1060 struct btrfs_fs_info *info = root->fs_info;
1061 u64 total = num_bytes;
1068 cache = btrfs_lookup_block_group(info, bytenr);
1072 byte_in_group = bytenr - cache->key.objectid;
1073 WARN_ON(byte_in_group > cache->key.offset);
1074 start = cache->key.objectid;
1075 end = start + cache->key.offset - 1;
1076 set_extent_bits(&info->block_group_cache, start, end,
1077 BLOCK_GROUP_DIRTY, GFP_NOFS);
1079 old_val = btrfs_block_group_used(&cache->item);
1080 num_bytes = min(total, cache->key.offset - byte_in_group);
1082 old_val += num_bytes;
1083 cache->space_info->bytes_used += num_bytes;
1085 old_val -= num_bytes;
1086 cache->space_info->bytes_used -= num_bytes;
1088 set_extent_dirty(&info->free_space_cache,
1089 bytenr, bytenr + num_bytes - 1,
1093 btrfs_set_block_group_used(&cache->item, old_val);
1095 bytenr += num_bytes;
1100 static int update_pinned_extents(struct btrfs_root *root,
1101 u64 bytenr, u64 num, int pin)
1104 struct btrfs_block_group_cache *cache;
1105 struct btrfs_fs_info *fs_info = root->fs_info;
1108 set_extent_dirty(&fs_info->pinned_extents,
1109 bytenr, bytenr + num - 1, GFP_NOFS);
1111 clear_extent_dirty(&fs_info->pinned_extents,
1112 bytenr, bytenr + num - 1, GFP_NOFS);
1115 cache = btrfs_lookup_block_group(fs_info, bytenr);
1117 len = min(num, cache->key.offset -
1118 (bytenr - cache->key.objectid));
1120 cache->pinned += len;
1121 cache->space_info->bytes_pinned += len;
1122 fs_info->total_pinned += len;
1124 cache->pinned -= len;
1125 cache->space_info->bytes_pinned -= len;
1126 fs_info->total_pinned -= len;
1134 int btrfs_copy_pinned(struct btrfs_root *root, struct extent_io_tree *copy)
1139 struct extent_io_tree *pinned_extents = &root->fs_info->pinned_extents;
1143 ret = find_first_extent_bit(pinned_extents, last,
1144 &start, &end, EXTENT_DIRTY);
1147 set_extent_dirty(copy, start, end, GFP_NOFS);
1153 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
1154 struct btrfs_root *root,
1155 struct extent_io_tree *unpin)
1160 struct extent_io_tree *free_space_cache;
1161 free_space_cache = &root->fs_info->free_space_cache;
1164 ret = find_first_extent_bit(unpin, 0, &start, &end,
1168 update_pinned_extents(root, start, end + 1 - start, 0);
1169 clear_extent_dirty(unpin, start, end, GFP_NOFS);
1170 set_extent_dirty(free_space_cache, start, end, GFP_NOFS);
1175 static int finish_current_insert(struct btrfs_trans_handle *trans,
1176 struct btrfs_root *extent_root)
1180 struct btrfs_fs_info *info = extent_root->fs_info;
1181 struct extent_buffer *eb;
1182 struct btrfs_path *path;
1183 struct btrfs_key ins;
1184 struct btrfs_disk_key first;
1185 struct btrfs_extent_item extent_item;
1190 btrfs_set_stack_extent_refs(&extent_item, 1);
1191 btrfs_set_key_type(&ins, BTRFS_EXTENT_ITEM_KEY);
1192 path = btrfs_alloc_path();
1195 ret = find_first_extent_bit(&info->extent_ins, 0, &start,
1196 &end, EXTENT_LOCKED);
1200 ins.objectid = start;
1201 ins.offset = end + 1 - start;
1202 err = btrfs_insert_item(trans, extent_root, &ins,
1203 &extent_item, sizeof(extent_item));
1204 clear_extent_bits(&info->extent_ins, start, end, EXTENT_LOCKED,
1206 eb = read_tree_block(extent_root, ins.objectid, ins.offset);
1207 level = btrfs_header_level(eb);
1209 btrfs_item_key(eb, &first, 0);
1211 btrfs_node_key(eb, &first, 0);
1213 err = btrfs_insert_extent_backref(trans, extent_root, path,
1214 start, extent_root->root_key.objectid,
1216 btrfs_disk_key_objectid(&first));
1218 free_extent_buffer(eb);
1220 btrfs_free_path(path);
1224 static int pin_down_bytes(struct btrfs_root *root, u64 bytenr, u32 num_bytes,
1228 struct extent_buffer *buf;
1231 buf = btrfs_find_tree_block(root, bytenr, num_bytes);
1233 if (btrfs_buffer_uptodate(buf)) {
1235 root->fs_info->running_transaction->transid;
1236 u64 header_transid =
1237 btrfs_header_generation(buf);
1238 if (header_transid == transid) {
1239 clean_tree_block(NULL, root, buf);
1240 free_extent_buffer(buf);
1244 free_extent_buffer(buf);
1246 update_pinned_extents(root, bytenr, num_bytes, 1);
1248 set_extent_bits(&root->fs_info->pending_del,
1249 bytenr, bytenr + num_bytes - 1,
1250 EXTENT_LOCKED, GFP_NOFS);
1257 * remove an extent from the root, returns 0 on success
1259 static int __free_extent(struct btrfs_trans_handle *trans, struct btrfs_root
1260 *root, u64 bytenr, u64 num_bytes,
1261 u64 root_objectid, u64 ref_generation,
1262 u64 owner_objectid, u64 owner_offset, int pin,
1265 struct btrfs_path *path;
1266 struct btrfs_key key;
1267 struct btrfs_fs_info *info = root->fs_info;
1268 struct btrfs_root *extent_root = info->extent_root;
1269 struct extent_buffer *leaf;
1271 int extent_slot = 0;
1272 int found_extent = 0;
1274 struct btrfs_extent_item *ei;
1277 key.objectid = bytenr;
1278 btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
1279 key.offset = num_bytes;
1280 path = btrfs_alloc_path();
1285 ret = lookup_extent_backref(trans, extent_root, path,
1286 bytenr, root_objectid,
1288 owner_objectid, owner_offset, 1);
1290 struct btrfs_key found_key;
1291 extent_slot = path->slots[0];
1292 while(extent_slot > 0) {
1294 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1296 if (found_key.objectid != bytenr)
1298 if (found_key.type == BTRFS_EXTENT_ITEM_KEY &&
1299 found_key.offset == num_bytes) {
1303 if (path->slots[0] - extent_slot > 5)
1307 ret = btrfs_del_item(trans, extent_root, path);
1309 btrfs_print_leaf(extent_root, path->nodes[0]);
1311 printk("Unable to find ref byte nr %Lu root %Lu "
1312 " gen %Lu owner %Lu offset %Lu\n", bytenr,
1313 root_objectid, ref_generation, owner_objectid,
1316 if (!found_extent) {
1317 btrfs_release_path(extent_root, path);
1318 ret = btrfs_search_slot(trans, extent_root, &key, path, -1, 1);
1322 extent_slot = path->slots[0];
1325 leaf = path->nodes[0];
1326 ei = btrfs_item_ptr(leaf, extent_slot,
1327 struct btrfs_extent_item);
1328 refs = btrfs_extent_refs(leaf, ei);
1331 btrfs_set_extent_refs(leaf, ei, refs);
1333 btrfs_mark_buffer_dirty(leaf);
1335 if (refs == 0 && found_extent && path->slots[0] == extent_slot + 1) {
1336 /* if the back ref and the extent are next to each other
1337 * they get deleted below in one shot
1339 path->slots[0] = extent_slot;
1341 } else if (found_extent) {
1342 /* otherwise delete the extent back ref */
1343 ret = btrfs_del_item(trans, extent_root, path);
1345 /* if refs are 0, we need to setup the path for deletion */
1347 btrfs_release_path(extent_root, path);
1348 ret = btrfs_search_slot(trans, extent_root, &key, path,
1361 ret = pin_down_bytes(root, bytenr, num_bytes, 0);
1367 /* block accounting for super block */
1368 super_used = btrfs_super_bytes_used(&info->super_copy);
1369 btrfs_set_super_bytes_used(&info->super_copy,
1370 super_used - num_bytes);
1372 /* block accounting for root item */
1373 root_used = btrfs_root_used(&root->root_item);
1374 btrfs_set_root_used(&root->root_item,
1375 root_used - num_bytes);
1376 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
1381 ret = update_block_group(trans, root, bytenr, num_bytes, 0,
1385 btrfs_free_path(path);
1386 finish_current_insert(trans, extent_root);
1391 * find all the blocks marked as pending in the radix tree and remove
1392 * them from the extent map
1394 static int del_pending_extents(struct btrfs_trans_handle *trans, struct
1395 btrfs_root *extent_root)
1401 struct extent_io_tree *pending_del;
1402 struct extent_io_tree *pinned_extents;
1404 pending_del = &extent_root->fs_info->pending_del;
1405 pinned_extents = &extent_root->fs_info->pinned_extents;
1408 ret = find_first_extent_bit(pending_del, 0, &start, &end,
1412 update_pinned_extents(extent_root, start, end + 1 - start, 1);
1413 clear_extent_bits(pending_del, start, end, EXTENT_LOCKED,
1415 ret = __free_extent(trans, extent_root,
1416 start, end + 1 - start,
1417 extent_root->root_key.objectid,
1426 * remove an extent from the root, returns 0 on success
1428 int btrfs_free_extent(struct btrfs_trans_handle *trans, struct btrfs_root
1429 *root, u64 bytenr, u64 num_bytes,
1430 u64 root_objectid, u64 ref_generation,
1431 u64 owner_objectid, u64 owner_offset, int pin)
1433 struct btrfs_root *extent_root = root->fs_info->extent_root;
1437 WARN_ON(num_bytes < root->sectorsize);
1438 if (!root->ref_cows)
1441 if (root == extent_root) {
1442 pin_down_bytes(root, bytenr, num_bytes, 1);
1445 ret = __free_extent(trans, root, bytenr, num_bytes, root_objectid,
1446 ref_generation, owner_objectid, owner_offset,
1448 pending_ret = del_pending_extents(trans, root->fs_info->extent_root);
1449 return ret ? ret : pending_ret;
1452 static u64 stripe_align(struct btrfs_root *root, u64 val)
1454 u64 mask = ((u64)root->stripesize - 1);
1455 u64 ret = (val + mask) & ~mask;
1460 * walks the btree of allocated extents and find a hole of a given size.
1461 * The key ins is changed to record the hole:
1462 * ins->objectid == block start
1463 * ins->flags = BTRFS_EXTENT_ITEM_KEY
1464 * ins->offset == number of blocks
1465 * Any available blocks before search_start are skipped.
1467 static int noinline find_free_extent(struct btrfs_trans_handle *trans,
1468 struct btrfs_root *orig_root,
1469 u64 num_bytes, u64 empty_size,
1470 u64 search_start, u64 search_end,
1471 u64 hint_byte, struct btrfs_key *ins,
1472 u64 exclude_start, u64 exclude_nr,
1476 u64 orig_search_start = search_start;
1477 struct btrfs_root * root = orig_root->fs_info->extent_root;
1478 struct btrfs_fs_info *info = root->fs_info;
1479 u64 total_needed = num_bytes;
1480 struct btrfs_block_group_cache *block_group;
1484 WARN_ON(num_bytes < root->sectorsize);
1485 btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY);
1487 if (search_end == (u64)-1)
1488 search_end = btrfs_super_total_bytes(&info->super_copy);
1491 block_group = btrfs_lookup_block_group(info, hint_byte);
1493 hint_byte = search_start;
1494 block_group = btrfs_find_block_group(root, block_group,
1495 hint_byte, data, 1);
1497 block_group = btrfs_find_block_group(root,
1499 search_start, data, 1);
1502 total_needed += empty_size;
1506 block_group = btrfs_lookup_block_group(info, search_start);
1508 block_group = btrfs_lookup_block_group(info,
1511 ret = find_search_start(root, &block_group, &search_start,
1512 total_needed, data);
1516 search_start = stripe_align(root, search_start);
1517 ins->objectid = search_start;
1518 ins->offset = num_bytes;
1520 if (ins->objectid + num_bytes >= search_end)
1523 if (ins->objectid + num_bytes >
1524 block_group->key.objectid + block_group->key.offset) {
1525 search_start = block_group->key.objectid +
1526 block_group->key.offset;
1530 if (test_range_bit(&info->extent_ins, ins->objectid,
1531 ins->objectid + num_bytes -1, EXTENT_LOCKED, 0)) {
1532 search_start = ins->objectid + num_bytes;
1536 if (test_range_bit(&info->pinned_extents, ins->objectid,
1537 ins->objectid + num_bytes -1, EXTENT_DIRTY, 0)) {
1538 search_start = ins->objectid + num_bytes;
1542 if (exclude_nr > 0 && (ins->objectid + num_bytes > exclude_start &&
1543 ins->objectid < exclude_start + exclude_nr)) {
1544 search_start = exclude_start + exclude_nr;
1548 if (!(data & BTRFS_BLOCK_GROUP_DATA)) {
1549 block_group = btrfs_lookup_block_group(info, ins->objectid);
1551 trans->block_group = block_group;
1553 ins->offset = num_bytes;
1557 if (search_start + num_bytes >= search_end) {
1559 search_start = orig_search_start;
1566 total_needed -= empty_size;
1571 block_group = btrfs_lookup_block_group(info, search_start);
1573 block_group = btrfs_find_block_group(root, block_group,
1574 search_start, data, 0);
1581 * finds a free extent and does all the dirty work required for allocation
1582 * returns the key for the extent through ins, and a tree buffer for
1583 * the first block of the extent through buf.
1585 * returns 0 if everything worked, non-zero otherwise.
1587 int btrfs_alloc_extent(struct btrfs_trans_handle *trans,
1588 struct btrfs_root *root,
1589 u64 num_bytes, u64 root_objectid, u64 ref_generation,
1590 u64 owner, u64 owner_offset,
1591 u64 empty_size, u64 hint_byte,
1592 u64 search_end, struct btrfs_key *ins, int data)
1598 u64 search_start = 0;
1601 struct btrfs_fs_info *info = root->fs_info;
1602 struct btrfs_root *extent_root = info->extent_root;
1603 struct btrfs_extent_item *extent_item;
1604 struct btrfs_extent_ref *ref;
1605 struct btrfs_path *path;
1606 struct btrfs_key keys[2];
1609 data = BTRFS_BLOCK_GROUP_DATA;
1610 } else if (root == root->fs_info->chunk_root) {
1611 data = BTRFS_BLOCK_GROUP_SYSTEM;
1613 data = BTRFS_BLOCK_GROUP_METADATA;
1616 if (root->ref_cows) {
1617 if (data != BTRFS_BLOCK_GROUP_METADATA) {
1618 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
1620 BTRFS_BLOCK_GROUP_METADATA);
1623 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
1628 new_hint = max(hint_byte, root->fs_info->alloc_start);
1629 if (new_hint < btrfs_super_total_bytes(&info->super_copy))
1630 hint_byte = new_hint;
1632 WARN_ON(num_bytes < root->sectorsize);
1633 ret = find_free_extent(trans, root, num_bytes, empty_size,
1634 search_start, search_end, hint_byte, ins,
1635 trans->alloc_exclude_start,
1636 trans->alloc_exclude_nr, data);
1641 /* block accounting for super block */
1642 super_used = btrfs_super_bytes_used(&info->super_copy);
1643 btrfs_set_super_bytes_used(&info->super_copy, super_used + num_bytes);
1645 /* block accounting for root item */
1646 root_used = btrfs_root_used(&root->root_item);
1647 btrfs_set_root_used(&root->root_item, root_used + num_bytes);
1649 clear_extent_dirty(&root->fs_info->free_space_cache,
1650 ins->objectid, ins->objectid + ins->offset - 1,
1653 if (root == extent_root) {
1654 set_extent_bits(&root->fs_info->extent_ins, ins->objectid,
1655 ins->objectid + ins->offset - 1,
1656 EXTENT_LOCKED, GFP_NOFS);
1660 WARN_ON(trans->alloc_exclude_nr);
1661 trans->alloc_exclude_start = ins->objectid;
1662 trans->alloc_exclude_nr = ins->offset;
1664 memcpy(&keys[0], ins, sizeof(*ins));
1665 keys[1].offset = hash_extent_ref(root_objectid, ref_generation,
1666 owner, owner_offset);
1667 keys[1].objectid = ins->objectid;
1668 keys[1].type = BTRFS_EXTENT_REF_KEY;
1669 sizes[0] = sizeof(*extent_item);
1670 sizes[1] = sizeof(*ref);
1672 path = btrfs_alloc_path();
1675 ret = btrfs_insert_empty_items(trans, extent_root, path, keys,
1679 extent_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
1680 struct btrfs_extent_item);
1681 btrfs_set_extent_refs(path->nodes[0], extent_item, 1);
1682 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
1683 struct btrfs_extent_ref);
1685 btrfs_set_ref_root(path->nodes[0], ref, root_objectid);
1686 btrfs_set_ref_generation(path->nodes[0], ref, ref_generation);
1687 btrfs_set_ref_objectid(path->nodes[0], ref, owner);
1688 btrfs_set_ref_offset(path->nodes[0], ref, owner_offset);
1690 btrfs_mark_buffer_dirty(path->nodes[0]);
1692 trans->alloc_exclude_start = 0;
1693 trans->alloc_exclude_nr = 0;
1694 btrfs_free_path(path);
1695 finish_current_insert(trans, extent_root);
1696 pending_ret = del_pending_extents(trans, extent_root);
1706 ret = update_block_group(trans, root, ins->objectid, ins->offset, 1, 0);
1708 printk("update block group failed for %Lu %Lu\n",
1709 ins->objectid, ins->offset);
1716 * helper function to allocate a block for a given tree
1717 * returns the tree buffer or NULL.
1719 struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
1720 struct btrfs_root *root,
1722 u64 root_objectid, u64 hint,
1728 ref_generation = trans->transid;
1733 return __btrfs_alloc_free_block(trans, root, blocksize, root_objectid,
1734 ref_generation, 0, 0, hint, empty_size);
1738 * helper function to allocate a block for a given tree
1739 * returns the tree buffer or NULL.
1741 struct extent_buffer *__btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
1742 struct btrfs_root *root,
1751 struct btrfs_key ins;
1753 struct extent_buffer *buf;
1755 ret = btrfs_alloc_extent(trans, root, blocksize,
1756 root_objectid, ref_generation,
1757 level, first_objectid, empty_size, hint,
1761 return ERR_PTR(ret);
1763 buf = btrfs_find_create_tree_block(root, ins.objectid, blocksize);
1765 btrfs_free_extent(trans, root, ins.objectid, blocksize,
1766 root->root_key.objectid, ref_generation,
1768 return ERR_PTR(-ENOMEM);
1770 btrfs_set_header_generation(buf, trans->transid);
1771 clean_tree_block(trans, root, buf);
1772 wait_on_tree_block_writeback(root, buf);
1773 btrfs_set_buffer_uptodate(buf);
1775 if (PageDirty(buf->first_page)) {
1776 printk("page %lu dirty\n", buf->first_page->index);
1780 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
1781 buf->start + buf->len - 1, GFP_NOFS);
1782 set_extent_bits(&BTRFS_I(root->fs_info->btree_inode)->io_tree,
1783 buf->start, buf->start + buf->len - 1,
1784 EXTENT_CSUM, GFP_NOFS);
1785 buf->flags |= EXTENT_CSUM;
1786 if (!btrfs_test_opt(root, SSD))
1787 btrfs_set_buffer_defrag(buf);
1788 trans->blocks_used++;
1792 static int noinline drop_leaf_ref(struct btrfs_trans_handle *trans,
1793 struct btrfs_root *root,
1794 struct extent_buffer *leaf)
1797 u64 leaf_generation;
1798 struct btrfs_key key;
1799 struct btrfs_file_extent_item *fi;
1804 BUG_ON(!btrfs_is_leaf(leaf));
1805 nritems = btrfs_header_nritems(leaf);
1806 leaf_owner = btrfs_header_owner(leaf);
1807 leaf_generation = btrfs_header_generation(leaf);
1809 for (i = 0; i < nritems; i++) {
1812 btrfs_item_key_to_cpu(leaf, &key, i);
1813 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
1815 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
1816 if (btrfs_file_extent_type(leaf, fi) ==
1817 BTRFS_FILE_EXTENT_INLINE)
1820 * FIXME make sure to insert a trans record that
1821 * repeats the snapshot del on crash
1823 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1824 if (disk_bytenr == 0)
1826 ret = btrfs_free_extent(trans, root, disk_bytenr,
1827 btrfs_file_extent_disk_num_bytes(leaf, fi),
1828 leaf_owner, leaf_generation,
1829 key.objectid, key.offset, 0);
1835 static void noinline reada_walk_down(struct btrfs_root *root,
1836 struct extent_buffer *node,
1849 nritems = btrfs_header_nritems(node);
1850 level = btrfs_header_level(node);
1854 for (i = slot; i < nritems && skipped < 32; i++) {
1855 bytenr = btrfs_node_blockptr(node, i);
1856 if (last && ((bytenr > last && bytenr - last > 32 * 1024) ||
1857 (last > bytenr && last - bytenr > 32 * 1024))) {
1861 blocksize = btrfs_level_size(root, level - 1);
1863 ret = lookup_extent_ref(NULL, root, bytenr,
1871 mutex_unlock(&root->fs_info->fs_mutex);
1872 ret = readahead_tree_block(root, bytenr, blocksize);
1873 last = bytenr + blocksize;
1875 mutex_lock(&root->fs_info->fs_mutex);
1882 * helper function for drop_snapshot, this walks down the tree dropping ref
1883 * counts as it goes.
1885 static int noinline walk_down_tree(struct btrfs_trans_handle *trans,
1886 struct btrfs_root *root,
1887 struct btrfs_path *path, int *level)
1892 struct extent_buffer *next;
1893 struct extent_buffer *cur;
1894 struct extent_buffer *parent;
1899 WARN_ON(*level < 0);
1900 WARN_ON(*level >= BTRFS_MAX_LEVEL);
1901 ret = lookup_extent_ref(trans, root,
1902 path->nodes[*level]->start,
1903 path->nodes[*level]->len, &refs);
1909 * walk down to the last node level and free all the leaves
1911 while(*level >= 0) {
1912 WARN_ON(*level < 0);
1913 WARN_ON(*level >= BTRFS_MAX_LEVEL);
1914 cur = path->nodes[*level];
1916 if (btrfs_header_level(cur) != *level)
1919 if (path->slots[*level] >=
1920 btrfs_header_nritems(cur))
1923 ret = drop_leaf_ref(trans, root, cur);
1927 bytenr = btrfs_node_blockptr(cur, path->slots[*level]);
1928 blocksize = btrfs_level_size(root, *level - 1);
1929 ret = lookup_extent_ref(trans, root, bytenr, blocksize, &refs);
1932 parent = path->nodes[*level];
1933 root_owner = btrfs_header_owner(parent);
1934 root_gen = btrfs_header_generation(parent);
1935 path->slots[*level]++;
1936 ret = btrfs_free_extent(trans, root, bytenr,
1937 blocksize, root_owner,
1942 next = btrfs_find_tree_block(root, bytenr, blocksize);
1943 if (!next || !btrfs_buffer_uptodate(next)) {
1944 free_extent_buffer(next);
1945 reada_walk_down(root, cur, path->slots[*level]);
1946 mutex_unlock(&root->fs_info->fs_mutex);
1947 next = read_tree_block(root, bytenr, blocksize);
1948 mutex_lock(&root->fs_info->fs_mutex);
1950 /* we dropped the lock, check one more time */
1951 ret = lookup_extent_ref(trans, root, bytenr,
1955 parent = path->nodes[*level];
1956 root_owner = btrfs_header_owner(parent);
1957 root_gen = btrfs_header_generation(parent);
1959 path->slots[*level]++;
1960 free_extent_buffer(next);
1961 ret = btrfs_free_extent(trans, root, bytenr,
1969 WARN_ON(*level <= 0);
1970 if (path->nodes[*level-1])
1971 free_extent_buffer(path->nodes[*level-1]);
1972 path->nodes[*level-1] = next;
1973 *level = btrfs_header_level(next);
1974 path->slots[*level] = 0;
1977 WARN_ON(*level < 0);
1978 WARN_ON(*level >= BTRFS_MAX_LEVEL);
1980 if (path->nodes[*level] == root->node) {
1981 root_owner = root->root_key.objectid;
1982 parent = path->nodes[*level];
1984 parent = path->nodes[*level + 1];
1985 root_owner = btrfs_header_owner(parent);
1988 root_gen = btrfs_header_generation(parent);
1989 ret = btrfs_free_extent(trans, root, path->nodes[*level]->start,
1990 path->nodes[*level]->len,
1991 root_owner, root_gen, 0, 0, 1);
1992 free_extent_buffer(path->nodes[*level]);
1993 path->nodes[*level] = NULL;
2000 * helper for dropping snapshots. This walks back up the tree in the path
2001 * to find the first node higher up where we haven't yet gone through
2004 static int noinline walk_up_tree(struct btrfs_trans_handle *trans,
2005 struct btrfs_root *root,
2006 struct btrfs_path *path, int *level)
2010 struct btrfs_root_item *root_item = &root->root_item;
2015 for(i = *level; i < BTRFS_MAX_LEVEL - 1 && path->nodes[i]; i++) {
2016 slot = path->slots[i];
2017 if (slot < btrfs_header_nritems(path->nodes[i]) - 1) {
2018 struct extent_buffer *node;
2019 struct btrfs_disk_key disk_key;
2020 node = path->nodes[i];
2023 WARN_ON(*level == 0);
2024 btrfs_node_key(node, &disk_key, path->slots[i]);
2025 memcpy(&root_item->drop_progress,
2026 &disk_key, sizeof(disk_key));
2027 root_item->drop_level = i;
2030 if (path->nodes[*level] == root->node) {
2031 root_owner = root->root_key.objectid;
2033 btrfs_header_generation(path->nodes[*level]);
2035 struct extent_buffer *node;
2036 node = path->nodes[*level + 1];
2037 root_owner = btrfs_header_owner(node);
2038 root_gen = btrfs_header_generation(node);
2040 ret = btrfs_free_extent(trans, root,
2041 path->nodes[*level]->start,
2042 path->nodes[*level]->len,
2043 root_owner, root_gen, 0, 0, 1);
2045 free_extent_buffer(path->nodes[*level]);
2046 path->nodes[*level] = NULL;
2054 * drop the reference count on the tree rooted at 'snap'. This traverses
2055 * the tree freeing any blocks that have a ref count of zero after being
2058 int btrfs_drop_snapshot(struct btrfs_trans_handle *trans, struct btrfs_root
2064 struct btrfs_path *path;
2067 struct btrfs_root_item *root_item = &root->root_item;
2069 path = btrfs_alloc_path();
2072 level = btrfs_header_level(root->node);
2074 if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
2075 path->nodes[level] = root->node;
2076 extent_buffer_get(root->node);
2077 path->slots[level] = 0;
2079 struct btrfs_key key;
2080 struct btrfs_disk_key found_key;
2081 struct extent_buffer *node;
2083 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
2084 level = root_item->drop_level;
2085 path->lowest_level = level;
2086 wret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2091 node = path->nodes[level];
2092 btrfs_node_key(node, &found_key, path->slots[level]);
2093 WARN_ON(memcmp(&found_key, &root_item->drop_progress,
2094 sizeof(found_key)));
2097 wret = walk_down_tree(trans, root, path, &level);
2103 wret = walk_up_tree(trans, root, path, &level);
2111 for (i = 0; i <= orig_level; i++) {
2112 if (path->nodes[i]) {
2113 free_extent_buffer(path->nodes[i]);
2114 path->nodes[i] = NULL;
2118 btrfs_free_path(path);
2122 int btrfs_free_block_groups(struct btrfs_fs_info *info)
2129 ret = find_first_extent_bit(&info->block_group_cache, 0,
2130 &start, &end, (unsigned int)-1);
2133 ret = get_state_private(&info->block_group_cache, start, &ptr);
2135 kfree((void *)(unsigned long)ptr);
2136 clear_extent_bits(&info->block_group_cache, start,
2137 end, (unsigned int)-1, GFP_NOFS);
2140 ret = find_first_extent_bit(&info->free_space_cache, 0,
2141 &start, &end, EXTENT_DIRTY);
2144 clear_extent_dirty(&info->free_space_cache, start,
2150 static int noinline relocate_inode_pages(struct inode *inode, u64 start,
2156 u64 existing_delalloc;
2157 unsigned long last_index;
2160 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2161 struct file_ra_state *ra;
2163 ra = kzalloc(sizeof(*ra), GFP_NOFS);
2165 mutex_lock(&inode->i_mutex);
2166 i = start >> PAGE_CACHE_SHIFT;
2167 last_index = (start + len - 1) >> PAGE_CACHE_SHIFT;
2169 file_ra_state_init(ra, inode->i_mapping);
2170 btrfs_force_ra(inode->i_mapping, ra, NULL, i, last_index);
2173 for (; i <= last_index; i++) {
2174 page = grab_cache_page(inode->i_mapping, i);
2177 if (!PageUptodate(page)) {
2178 btrfs_readpage(NULL, page);
2180 if (!PageUptodate(page)) {
2182 page_cache_release(page);
2186 page_start = (u64)page->index << PAGE_CACHE_SHIFT;
2187 page_end = page_start + PAGE_CACHE_SIZE - 1;
2189 lock_extent(io_tree, page_start, page_end, GFP_NOFS);
2191 delalloc_start = page_start;
2192 existing_delalloc = count_range_bits(io_tree,
2193 &delalloc_start, page_end,
2194 PAGE_CACHE_SIZE, EXTENT_DELALLOC);
2196 set_extent_delalloc(io_tree, page_start,
2197 page_end, GFP_NOFS);
2199 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
2200 set_page_dirty(page);
2202 page_cache_release(page);
2206 mutex_unlock(&inode->i_mutex);
2211 * note, this releases the path
2213 static int noinline relocate_one_reference(struct btrfs_root *extent_root,
2214 struct btrfs_path *path,
2215 struct btrfs_key *extent_key)
2217 struct inode *inode;
2218 struct btrfs_root *found_root;
2219 struct btrfs_key *root_location;
2220 struct btrfs_extent_ref *ref;
2227 ref = btrfs_item_ptr(path->nodes[0], path->slots[0],
2228 struct btrfs_extent_ref);
2229 ref_root = btrfs_ref_root(path->nodes[0], ref);
2230 ref_gen = btrfs_ref_generation(path->nodes[0], ref);
2231 ref_objectid = btrfs_ref_objectid(path->nodes[0], ref);
2232 ref_offset = btrfs_ref_offset(path->nodes[0], ref);
2233 btrfs_release_path(extent_root, path);
2235 root_location = kmalloc(sizeof(*root_location), GFP_NOFS);
2236 root_location->objectid = ref_root;
2238 root_location->offset = 0;
2240 root_location->offset = (u64)-1;
2241 root_location->type = BTRFS_ROOT_ITEM_KEY;
2243 found_root = btrfs_read_fs_root_no_name(extent_root->fs_info,
2245 BUG_ON(!found_root);
2246 kfree(root_location);
2248 if (ref_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
2249 mutex_unlock(&extent_root->fs_info->fs_mutex);
2250 inode = btrfs_iget_locked(extent_root->fs_info->sb,
2251 ref_objectid, found_root);
2252 if (inode->i_state & I_NEW) {
2253 /* the inode and parent dir are two different roots */
2254 BTRFS_I(inode)->root = found_root;
2255 BTRFS_I(inode)->location.objectid = ref_objectid;
2256 BTRFS_I(inode)->location.type = BTRFS_INODE_ITEM_KEY;
2257 BTRFS_I(inode)->location.offset = 0;
2258 btrfs_read_locked_inode(inode);
2259 unlock_new_inode(inode);
2262 /* this can happen if the reference is not against
2263 * the latest version of the tree root
2265 if (is_bad_inode(inode)) {
2266 mutex_lock(&extent_root->fs_info->fs_mutex);
2269 relocate_inode_pages(inode, ref_offset, extent_key->offset);
2270 /* FIXME, data=ordered will help get rid of this */
2271 filemap_fdatawrite(inode->i_mapping);
2273 mutex_lock(&extent_root->fs_info->fs_mutex);
2275 struct btrfs_trans_handle *trans;
2276 struct btrfs_key found_key;
2277 struct extent_buffer *eb;
2281 trans = btrfs_start_transaction(found_root, 1);
2282 eb = read_tree_block(found_root, extent_key->objectid,
2283 extent_key->offset);
2284 level = btrfs_header_level(eb);
2287 btrfs_item_key_to_cpu(eb, &found_key, 0);
2289 btrfs_node_key_to_cpu(eb, &found_key, 0);
2291 free_extent_buffer(eb);
2293 path->lowest_level = level;
2295 ret = btrfs_search_slot(trans, found_root, &found_key, path,
2297 path->lowest_level = 0;
2298 for (i = level; i < BTRFS_MAX_LEVEL; i++) {
2299 if (!path->nodes[i])
2301 free_extent_buffer(path->nodes[i]);
2302 path->nodes[i] = NULL;
2304 btrfs_release_path(found_root, path);
2305 btrfs_end_transaction(trans, found_root);
2312 static int noinline relocate_one_extent(struct btrfs_root *extent_root,
2313 struct btrfs_path *path,
2314 struct btrfs_key *extent_key)
2316 struct btrfs_key key;
2317 struct btrfs_key found_key;
2318 struct extent_buffer *leaf;
2323 key.objectid = extent_key->objectid;
2324 key.type = BTRFS_EXTENT_REF_KEY;
2328 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
2334 leaf = path->nodes[0];
2335 nritems = btrfs_header_nritems(leaf);
2336 if (path->slots[0] == nritems)
2339 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2340 if (found_key.objectid != extent_key->objectid)
2343 if (found_key.type != BTRFS_EXTENT_REF_KEY)
2346 key.offset = found_key.offset + 1;
2347 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2349 ret = relocate_one_reference(extent_root, path, extent_key);
2355 btrfs_release_path(extent_root, path);
2359 int btrfs_shrink_extent_tree(struct btrfs_root *root, u64 new_size)
2361 struct btrfs_trans_handle *trans;
2362 struct btrfs_root *tree_root = root->fs_info->tree_root;
2363 struct btrfs_path *path;
2366 struct btrfs_fs_info *info = root->fs_info;
2367 struct extent_io_tree *block_group_cache;
2368 struct btrfs_key key;
2369 struct btrfs_key found_key;
2370 struct extent_buffer *leaf;
2375 btrfs_set_super_total_bytes(&info->super_copy, new_size);
2376 clear_extent_dirty(&info->free_space_cache, new_size, (u64)-1,
2378 block_group_cache = &info->block_group_cache;
2379 path = btrfs_alloc_path();
2380 root = root->fs_info->extent_root;
2385 key.objectid = new_size;
2388 cur_byte = key.objectid;
2390 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2394 ret = btrfs_previous_item(root, path, 0, BTRFS_EXTENT_ITEM_KEY);
2398 leaf = path->nodes[0];
2399 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2400 if (found_key.objectid + found_key.offset > new_size) {
2401 cur_byte = found_key.objectid;
2402 key.objectid = cur_byte;
2405 btrfs_release_path(root, path);
2408 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2412 leaf = path->nodes[0];
2413 nritems = btrfs_header_nritems(leaf);
2415 if (path->slots[0] >= nritems) {
2416 ret = btrfs_next_leaf(root, path);
2423 leaf = path->nodes[0];
2424 nritems = btrfs_header_nritems(leaf);
2427 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2429 if (progress && need_resched()) {
2430 memcpy(&key, &found_key, sizeof(key));
2431 mutex_unlock(&root->fs_info->fs_mutex);
2433 mutex_lock(&root->fs_info->fs_mutex);
2434 btrfs_release_path(root, path);
2435 btrfs_search_slot(NULL, root, &key, path, 0, 0);
2441 if (btrfs_key_type(&found_key) != BTRFS_EXTENT_ITEM_KEY ||
2442 found_key.objectid + found_key.offset <= cur_byte) {
2448 cur_byte = found_key.objectid + found_key.offset;
2449 key.objectid = cur_byte;
2450 btrfs_release_path(root, path);
2451 ret = relocate_one_extent(root, path, &found_key);
2454 btrfs_release_path(root, path);
2456 if (total_found > 0) {
2457 trans = btrfs_start_transaction(tree_root, 1);
2458 btrfs_commit_transaction(trans, tree_root);
2460 mutex_unlock(&root->fs_info->fs_mutex);
2461 btrfs_clean_old_snapshots(tree_root);
2462 mutex_lock(&root->fs_info->fs_mutex);
2464 trans = btrfs_start_transaction(tree_root, 1);
2465 btrfs_commit_transaction(trans, tree_root);
2469 trans = btrfs_start_transaction(root, 1);
2470 key.objectid = new_size;
2476 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2480 leaf = path->nodes[0];
2481 nritems = btrfs_header_nritems(leaf);
2483 if (path->slots[0] >= nritems) {
2484 ret = btrfs_next_leaf(root, path);
2491 leaf = path->nodes[0];
2492 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2495 * btrfs_next_leaf doesn't cow buffers, we have to
2496 * do the search again
2498 memcpy(&key, &found_key, sizeof(key));
2499 btrfs_release_path(root, path);
2503 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2504 if (btrfs_key_type(&found_key) != BTRFS_BLOCK_GROUP_ITEM_KEY) {
2505 printk("shrinker found key %Lu %u %Lu\n",
2506 found_key.objectid, found_key.type,
2511 ret = get_state_private(&info->block_group_cache,
2512 found_key.objectid, &ptr);
2514 kfree((void *)(unsigned long)ptr);
2516 clear_extent_bits(&info->block_group_cache, found_key.objectid,
2517 found_key.objectid + found_key.offset - 1,
2518 (unsigned int)-1, GFP_NOFS);
2520 key.objectid = found_key.objectid + 1;
2521 btrfs_del_item(trans, root, path);
2522 btrfs_release_path(root, path);
2524 if (need_resched()) {
2525 mutex_unlock(&root->fs_info->fs_mutex);
2527 mutex_lock(&root->fs_info->fs_mutex);
2530 clear_extent_dirty(&info->free_space_cache, new_size, (u64)-1,
2532 btrfs_commit_transaction(trans, root);
2534 btrfs_free_path(path);
2538 int btrfs_grow_extent_tree(struct btrfs_trans_handle *trans,
2539 struct btrfs_root *root, u64 new_size)
2541 btrfs_set_super_total_bytes(&root->fs_info->super_copy, new_size);
2545 int find_first_block_group(struct btrfs_root *root, struct btrfs_path *path,
2546 struct btrfs_key *key)
2549 struct btrfs_key found_key;
2550 struct extent_buffer *leaf;
2553 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
2557 slot = path->slots[0];
2558 leaf = path->nodes[0];
2559 if (slot >= btrfs_header_nritems(leaf)) {
2560 ret = btrfs_next_leaf(root, path);
2567 btrfs_item_key_to_cpu(leaf, &found_key, slot);
2569 if (found_key.objectid >= key->objectid &&
2570 found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY)
2579 static int update_space_info(struct btrfs_fs_info *info, u64 flags,
2580 u64 total_bytes, u64 bytes_used,
2581 struct btrfs_space_info **space_info)
2583 struct btrfs_space_info *found;
2585 found = __find_space_info(info, flags);
2587 found->total_bytes += total_bytes;
2588 found->bytes_used += bytes_used;
2589 WARN_ON(found->total_bytes < found->bytes_used);
2590 *space_info = found;
2593 found = kmalloc(sizeof(*found), GFP_NOFS);
2597 list_add(&found->list, &info->space_info);
2598 found->flags = flags;
2599 found->total_bytes = total_bytes;
2600 found->bytes_used = bytes_used;
2601 found->bytes_pinned = 0;
2603 *space_info = found;
2607 int btrfs_read_block_groups(struct btrfs_root *root)
2609 struct btrfs_path *path;
2612 struct btrfs_block_group_cache *cache;
2613 struct btrfs_fs_info *info = root->fs_info;
2614 struct btrfs_space_info *space_info;
2615 struct extent_io_tree *block_group_cache;
2616 struct btrfs_key key;
2617 struct btrfs_key found_key;
2618 struct extent_buffer *leaf;
2620 block_group_cache = &info->block_group_cache;
2621 root = info->extent_root;
2624 btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY);
2625 path = btrfs_alloc_path();
2630 ret = find_first_block_group(root, path, &key);
2638 leaf = path->nodes[0];
2639 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2640 cache = kmalloc(sizeof(*cache), GFP_NOFS);
2646 read_extent_buffer(leaf, &cache->item,
2647 btrfs_item_ptr_offset(leaf, path->slots[0]),
2648 sizeof(cache->item));
2649 memcpy(&cache->key, &found_key, sizeof(found_key));
2653 key.objectid = found_key.objectid + found_key.offset;
2654 btrfs_release_path(root, path);
2655 cache->flags = btrfs_block_group_flags(&cache->item);
2657 if (cache->flags & BTRFS_BLOCK_GROUP_DATA) {
2658 bit = BLOCK_GROUP_DATA;
2659 } else if (cache->flags & BTRFS_BLOCK_GROUP_SYSTEM) {
2660 bit = BLOCK_GROUP_SYSTEM;
2661 } else if (cache->flags & BTRFS_BLOCK_GROUP_METADATA) {
2662 bit = BLOCK_GROUP_METADATA;
2665 ret = update_space_info(info, cache->flags, found_key.offset,
2666 btrfs_block_group_used(&cache->item),
2669 cache->space_info = space_info;
2671 /* use EXTENT_LOCKED to prevent merging */
2672 set_extent_bits(block_group_cache, found_key.objectid,
2673 found_key.objectid + found_key.offset - 1,
2674 bit | EXTENT_LOCKED, GFP_NOFS);
2675 set_state_private(block_group_cache, found_key.objectid,
2676 (unsigned long)cache);
2679 btrfs_super_total_bytes(&info->super_copy))
2684 btrfs_free_path(path);
2688 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
2689 struct btrfs_root *root, u64 bytes_used,
2690 u64 type, u64 chunk_tree, u64 chunk_objectid,
2695 struct btrfs_root *extent_root;
2696 struct btrfs_block_group_cache *cache;
2697 struct extent_io_tree *block_group_cache;
2699 extent_root = root->fs_info->extent_root;
2700 block_group_cache = &root->fs_info->block_group_cache;
2702 cache = kmalloc(sizeof(*cache), GFP_NOFS);
2704 cache->key.objectid = chunk_objectid;
2705 cache->key.offset = size;
2708 btrfs_set_key_type(&cache->key, BTRFS_BLOCK_GROUP_ITEM_KEY);
2709 memset(&cache->item, 0, sizeof(cache->item));
2710 btrfs_set_block_group_used(&cache->item, bytes_used);
2711 btrfs_set_block_group_chunk_tree(&cache->item, chunk_tree);
2712 btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
2713 cache->flags = type;
2714 btrfs_set_block_group_flags(&cache->item, type);
2716 ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
2717 &cache->space_info);
2720 if (type & BTRFS_BLOCK_GROUP_DATA) {
2721 bit = BLOCK_GROUP_DATA;
2722 } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
2723 bit = BLOCK_GROUP_SYSTEM;
2724 } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
2725 bit = BLOCK_GROUP_METADATA;
2727 set_extent_bits(block_group_cache, chunk_objectid,
2728 chunk_objectid + size - 1,
2729 bit | EXTENT_LOCKED, GFP_NOFS);
2730 set_state_private(block_group_cache, chunk_objectid,
2731 (unsigned long)cache);
2733 ret = btrfs_insert_item(trans, extent_root, &cache->key, &cache->item,
2734 sizeof(cache->item));
2737 finish_current_insert(trans, extent_root);
2738 ret = del_pending_extents(trans, extent_root);