ocfs2: fix locking for res->tracking and dlm->tracking_list
[linux/fpc-iii.git] / fs / btrfs / relocation.c
blobcfe913d2d3df98b315d03e141c08b4d626f38fb9
1 /*
2 * Copyright (C) 2009 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/pagemap.h>
21 #include <linux/writeback.h>
22 #include <linux/blkdev.h>
23 #include <linux/rbtree.h>
24 #include <linux/slab.h>
25 #include "ctree.h"
26 #include "disk-io.h"
27 #include "transaction.h"
28 #include "volumes.h"
29 #include "locking.h"
30 #include "btrfs_inode.h"
31 #include "async-thread.h"
32 #include "free-space-cache.h"
33 #include "inode-map.h"
36 * backref_node, mapping_node and tree_block start with this
38 struct tree_entry {
39 struct rb_node rb_node;
40 u64 bytenr;
44 * present a tree block in the backref cache
46 struct backref_node {
47 struct rb_node rb_node;
48 u64 bytenr;
50 u64 new_bytenr;
51 /* objectid of tree block owner, can be not uptodate */
52 u64 owner;
53 /* link to pending, changed or detached list */
54 struct list_head list;
55 /* list of upper level blocks reference this block */
56 struct list_head upper;
57 /* list of child blocks in the cache */
58 struct list_head lower;
59 /* NULL if this node is not tree root */
60 struct btrfs_root *root;
61 /* extent buffer got by COW the block */
62 struct extent_buffer *eb;
63 /* level of tree block */
64 unsigned int level:8;
65 /* is the block in non-reference counted tree */
66 unsigned int cowonly:1;
67 /* 1 if no child node in the cache */
68 unsigned int lowest:1;
69 /* is the extent buffer locked */
70 unsigned int locked:1;
71 /* has the block been processed */
72 unsigned int processed:1;
73 /* have backrefs of this block been checked */
74 unsigned int checked:1;
76 * 1 if corresponding block has been cowed but some upper
77 * level block pointers may not point to the new location
79 unsigned int pending:1;
81 * 1 if the backref node isn't connected to any other
82 * backref node.
84 unsigned int detached:1;
88 * present a block pointer in the backref cache
90 struct backref_edge {
91 struct list_head list[2];
92 struct backref_node *node[2];
95 #define LOWER 0
96 #define UPPER 1
97 #define RELOCATION_RESERVED_NODES 256
99 struct backref_cache {
100 /* red black tree of all backref nodes in the cache */
101 struct rb_root rb_root;
102 /* for passing backref nodes to btrfs_reloc_cow_block */
103 struct backref_node *path[BTRFS_MAX_LEVEL];
105 * list of blocks that have been cowed but some block
106 * pointers in upper level blocks may not reflect the
107 * new location
109 struct list_head pending[BTRFS_MAX_LEVEL];
110 /* list of backref nodes with no child node */
111 struct list_head leaves;
112 /* list of blocks that have been cowed in current transaction */
113 struct list_head changed;
114 /* list of detached backref node. */
115 struct list_head detached;
117 u64 last_trans;
119 int nr_nodes;
120 int nr_edges;
124 * map address of tree root to tree
126 struct mapping_node {
127 struct rb_node rb_node;
128 u64 bytenr;
129 void *data;
132 struct mapping_tree {
133 struct rb_root rb_root;
134 spinlock_t lock;
138 * present a tree block to process
140 struct tree_block {
141 struct rb_node rb_node;
142 u64 bytenr;
143 struct btrfs_key key;
144 unsigned int level:8;
145 unsigned int key_ready:1;
148 #define MAX_EXTENTS 128
150 struct file_extent_cluster {
151 u64 start;
152 u64 end;
153 u64 boundary[MAX_EXTENTS];
154 unsigned int nr;
157 struct reloc_control {
158 /* block group to relocate */
159 struct btrfs_block_group_cache *block_group;
160 /* extent tree */
161 struct btrfs_root *extent_root;
162 /* inode for moving data */
163 struct inode *data_inode;
165 struct btrfs_block_rsv *block_rsv;
167 struct backref_cache backref_cache;
169 struct file_extent_cluster cluster;
170 /* tree blocks have been processed */
171 struct extent_io_tree processed_blocks;
172 /* map start of tree root to corresponding reloc tree */
173 struct mapping_tree reloc_root_tree;
174 /* list of reloc trees */
175 struct list_head reloc_roots;
176 /* size of metadata reservation for merging reloc trees */
177 u64 merging_rsv_size;
178 /* size of relocated tree nodes */
179 u64 nodes_relocated;
180 /* reserved size for block group relocation*/
181 u64 reserved_bytes;
183 u64 search_start;
184 u64 extents_found;
186 unsigned int stage:8;
187 unsigned int create_reloc_tree:1;
188 unsigned int merge_reloc_tree:1;
189 unsigned int found_file_extent:1;
192 /* stages of data relocation */
193 #define MOVE_DATA_EXTENTS 0
194 #define UPDATE_DATA_PTRS 1
196 static void remove_backref_node(struct backref_cache *cache,
197 struct backref_node *node);
198 static void __mark_block_processed(struct reloc_control *rc,
199 struct backref_node *node);
201 static void mapping_tree_init(struct mapping_tree *tree)
203 tree->rb_root = RB_ROOT;
204 spin_lock_init(&tree->lock);
207 static void backref_cache_init(struct backref_cache *cache)
209 int i;
210 cache->rb_root = RB_ROOT;
211 for (i = 0; i < BTRFS_MAX_LEVEL; i++)
212 INIT_LIST_HEAD(&cache->pending[i]);
213 INIT_LIST_HEAD(&cache->changed);
214 INIT_LIST_HEAD(&cache->detached);
215 INIT_LIST_HEAD(&cache->leaves);
218 static void backref_cache_cleanup(struct backref_cache *cache)
220 struct backref_node *node;
221 int i;
223 while (!list_empty(&cache->detached)) {
224 node = list_entry(cache->detached.next,
225 struct backref_node, list);
226 remove_backref_node(cache, node);
229 while (!list_empty(&cache->leaves)) {
230 node = list_entry(cache->leaves.next,
231 struct backref_node, lower);
232 remove_backref_node(cache, node);
235 cache->last_trans = 0;
237 for (i = 0; i < BTRFS_MAX_LEVEL; i++)
238 BUG_ON(!list_empty(&cache->pending[i]));
239 BUG_ON(!list_empty(&cache->changed));
240 BUG_ON(!list_empty(&cache->detached));
241 BUG_ON(!RB_EMPTY_ROOT(&cache->rb_root));
242 BUG_ON(cache->nr_nodes);
243 BUG_ON(cache->nr_edges);
246 static struct backref_node *alloc_backref_node(struct backref_cache *cache)
248 struct backref_node *node;
250 node = kzalloc(sizeof(*node), GFP_NOFS);
251 if (node) {
252 INIT_LIST_HEAD(&node->list);
253 INIT_LIST_HEAD(&node->upper);
254 INIT_LIST_HEAD(&node->lower);
255 RB_CLEAR_NODE(&node->rb_node);
256 cache->nr_nodes++;
258 return node;
261 static void free_backref_node(struct backref_cache *cache,
262 struct backref_node *node)
264 if (node) {
265 cache->nr_nodes--;
266 kfree(node);
270 static struct backref_edge *alloc_backref_edge(struct backref_cache *cache)
272 struct backref_edge *edge;
274 edge = kzalloc(sizeof(*edge), GFP_NOFS);
275 if (edge)
276 cache->nr_edges++;
277 return edge;
280 static void free_backref_edge(struct backref_cache *cache,
281 struct backref_edge *edge)
283 if (edge) {
284 cache->nr_edges--;
285 kfree(edge);
289 static struct rb_node *tree_insert(struct rb_root *root, u64 bytenr,
290 struct rb_node *node)
292 struct rb_node **p = &root->rb_node;
293 struct rb_node *parent = NULL;
294 struct tree_entry *entry;
296 while (*p) {
297 parent = *p;
298 entry = rb_entry(parent, struct tree_entry, rb_node);
300 if (bytenr < entry->bytenr)
301 p = &(*p)->rb_left;
302 else if (bytenr > entry->bytenr)
303 p = &(*p)->rb_right;
304 else
305 return parent;
308 rb_link_node(node, parent, p);
309 rb_insert_color(node, root);
310 return NULL;
313 static struct rb_node *tree_search(struct rb_root *root, u64 bytenr)
315 struct rb_node *n = root->rb_node;
316 struct tree_entry *entry;
318 while (n) {
319 entry = rb_entry(n, struct tree_entry, rb_node);
321 if (bytenr < entry->bytenr)
322 n = n->rb_left;
323 else if (bytenr > entry->bytenr)
324 n = n->rb_right;
325 else
326 return n;
328 return NULL;
331 static void backref_tree_panic(struct rb_node *rb_node, int errno, u64 bytenr)
334 struct btrfs_fs_info *fs_info = NULL;
335 struct backref_node *bnode = rb_entry(rb_node, struct backref_node,
336 rb_node);
337 if (bnode->root)
338 fs_info = bnode->root->fs_info;
339 btrfs_panic(fs_info, errno, "Inconsistency in backref cache "
340 "found at offset %llu", bytenr);
344 * walk up backref nodes until reach node presents tree root
346 static struct backref_node *walk_up_backref(struct backref_node *node,
347 struct backref_edge *edges[],
348 int *index)
350 struct backref_edge *edge;
351 int idx = *index;
353 while (!list_empty(&node->upper)) {
354 edge = list_entry(node->upper.next,
355 struct backref_edge, list[LOWER]);
356 edges[idx++] = edge;
357 node = edge->node[UPPER];
359 BUG_ON(node->detached);
360 *index = idx;
361 return node;
365 * walk down backref nodes to find start of next reference path
367 static struct backref_node *walk_down_backref(struct backref_edge *edges[],
368 int *index)
370 struct backref_edge *edge;
371 struct backref_node *lower;
372 int idx = *index;
374 while (idx > 0) {
375 edge = edges[idx - 1];
376 lower = edge->node[LOWER];
377 if (list_is_last(&edge->list[LOWER], &lower->upper)) {
378 idx--;
379 continue;
381 edge = list_entry(edge->list[LOWER].next,
382 struct backref_edge, list[LOWER]);
383 edges[idx - 1] = edge;
384 *index = idx;
385 return edge->node[UPPER];
387 *index = 0;
388 return NULL;
391 static void unlock_node_buffer(struct backref_node *node)
393 if (node->locked) {
394 btrfs_tree_unlock(node->eb);
395 node->locked = 0;
399 static void drop_node_buffer(struct backref_node *node)
401 if (node->eb) {
402 unlock_node_buffer(node);
403 free_extent_buffer(node->eb);
404 node->eb = NULL;
408 static void drop_backref_node(struct backref_cache *tree,
409 struct backref_node *node)
411 BUG_ON(!list_empty(&node->upper));
413 drop_node_buffer(node);
414 list_del(&node->list);
415 list_del(&node->lower);
416 if (!RB_EMPTY_NODE(&node->rb_node))
417 rb_erase(&node->rb_node, &tree->rb_root);
418 free_backref_node(tree, node);
422 * remove a backref node from the backref cache
424 static void remove_backref_node(struct backref_cache *cache,
425 struct backref_node *node)
427 struct backref_node *upper;
428 struct backref_edge *edge;
430 if (!node)
431 return;
433 BUG_ON(!node->lowest && !node->detached);
434 while (!list_empty(&node->upper)) {
435 edge = list_entry(node->upper.next, struct backref_edge,
436 list[LOWER]);
437 upper = edge->node[UPPER];
438 list_del(&edge->list[LOWER]);
439 list_del(&edge->list[UPPER]);
440 free_backref_edge(cache, edge);
442 if (RB_EMPTY_NODE(&upper->rb_node)) {
443 BUG_ON(!list_empty(&node->upper));
444 drop_backref_node(cache, node);
445 node = upper;
446 node->lowest = 1;
447 continue;
450 * add the node to leaf node list if no other
451 * child block cached.
453 if (list_empty(&upper->lower)) {
454 list_add_tail(&upper->lower, &cache->leaves);
455 upper->lowest = 1;
459 drop_backref_node(cache, node);
462 static void update_backref_node(struct backref_cache *cache,
463 struct backref_node *node, u64 bytenr)
465 struct rb_node *rb_node;
466 rb_erase(&node->rb_node, &cache->rb_root);
467 node->bytenr = bytenr;
468 rb_node = tree_insert(&cache->rb_root, node->bytenr, &node->rb_node);
469 if (rb_node)
470 backref_tree_panic(rb_node, -EEXIST, bytenr);
474 * update backref cache after a transaction commit
476 static int update_backref_cache(struct btrfs_trans_handle *trans,
477 struct backref_cache *cache)
479 struct backref_node *node;
480 int level = 0;
482 if (cache->last_trans == 0) {
483 cache->last_trans = trans->transid;
484 return 0;
487 if (cache->last_trans == trans->transid)
488 return 0;
491 * detached nodes are used to avoid unnecessary backref
492 * lookup. transaction commit changes the extent tree.
493 * so the detached nodes are no longer useful.
495 while (!list_empty(&cache->detached)) {
496 node = list_entry(cache->detached.next,
497 struct backref_node, list);
498 remove_backref_node(cache, node);
501 while (!list_empty(&cache->changed)) {
502 node = list_entry(cache->changed.next,
503 struct backref_node, list);
504 list_del_init(&node->list);
505 BUG_ON(node->pending);
506 update_backref_node(cache, node, node->new_bytenr);
510 * some nodes can be left in the pending list if there were
511 * errors during processing the pending nodes.
513 for (level = 0; level < BTRFS_MAX_LEVEL; level++) {
514 list_for_each_entry(node, &cache->pending[level], list) {
515 BUG_ON(!node->pending);
516 if (node->bytenr == node->new_bytenr)
517 continue;
518 update_backref_node(cache, node, node->new_bytenr);
522 cache->last_trans = 0;
523 return 1;
527 static int should_ignore_root(struct btrfs_root *root)
529 struct btrfs_root *reloc_root;
531 if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state))
532 return 0;
534 reloc_root = root->reloc_root;
535 if (!reloc_root)
536 return 0;
538 if (btrfs_root_last_snapshot(&reloc_root->root_item) ==
539 root->fs_info->running_transaction->transid - 1)
540 return 0;
542 * if there is reloc tree and it was created in previous
543 * transaction backref lookup can find the reloc tree,
544 * so backref node for the fs tree root is useless for
545 * relocation.
547 return 1;
550 * find reloc tree by address of tree root
552 static struct btrfs_root *find_reloc_root(struct reloc_control *rc,
553 u64 bytenr)
555 struct rb_node *rb_node;
556 struct mapping_node *node;
557 struct btrfs_root *root = NULL;
559 spin_lock(&rc->reloc_root_tree.lock);
560 rb_node = tree_search(&rc->reloc_root_tree.rb_root, bytenr);
561 if (rb_node) {
562 node = rb_entry(rb_node, struct mapping_node, rb_node);
563 root = (struct btrfs_root *)node->data;
565 spin_unlock(&rc->reloc_root_tree.lock);
566 return root;
569 static int is_cowonly_root(u64 root_objectid)
571 if (root_objectid == BTRFS_ROOT_TREE_OBJECTID ||
572 root_objectid == BTRFS_EXTENT_TREE_OBJECTID ||
573 root_objectid == BTRFS_CHUNK_TREE_OBJECTID ||
574 root_objectid == BTRFS_DEV_TREE_OBJECTID ||
575 root_objectid == BTRFS_TREE_LOG_OBJECTID ||
576 root_objectid == BTRFS_CSUM_TREE_OBJECTID ||
577 root_objectid == BTRFS_UUID_TREE_OBJECTID ||
578 root_objectid == BTRFS_QUOTA_TREE_OBJECTID)
579 return 1;
580 return 0;
583 static struct btrfs_root *read_fs_root(struct btrfs_fs_info *fs_info,
584 u64 root_objectid)
586 struct btrfs_key key;
588 key.objectid = root_objectid;
589 key.type = BTRFS_ROOT_ITEM_KEY;
590 if (is_cowonly_root(root_objectid))
591 key.offset = 0;
592 else
593 key.offset = (u64)-1;
595 return btrfs_get_fs_root(fs_info, &key, false);
598 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
599 static noinline_for_stack
600 struct btrfs_root *find_tree_root(struct reloc_control *rc,
601 struct extent_buffer *leaf,
602 struct btrfs_extent_ref_v0 *ref0)
604 struct btrfs_root *root;
605 u64 root_objectid = btrfs_ref_root_v0(leaf, ref0);
606 u64 generation = btrfs_ref_generation_v0(leaf, ref0);
608 BUG_ON(root_objectid == BTRFS_TREE_RELOC_OBJECTID);
610 root = read_fs_root(rc->extent_root->fs_info, root_objectid);
611 BUG_ON(IS_ERR(root));
613 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) &&
614 generation != btrfs_root_generation(&root->root_item))
615 return NULL;
617 return root;
619 #endif
621 static noinline_for_stack
622 int find_inline_backref(struct extent_buffer *leaf, int slot,
623 unsigned long *ptr, unsigned long *end)
625 struct btrfs_key key;
626 struct btrfs_extent_item *ei;
627 struct btrfs_tree_block_info *bi;
628 u32 item_size;
630 btrfs_item_key_to_cpu(leaf, &key, slot);
632 item_size = btrfs_item_size_nr(leaf, slot);
633 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
634 if (item_size < sizeof(*ei)) {
635 WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0));
636 return 1;
638 #endif
639 ei = btrfs_item_ptr(leaf, slot, struct btrfs_extent_item);
640 WARN_ON(!(btrfs_extent_flags(leaf, ei) &
641 BTRFS_EXTENT_FLAG_TREE_BLOCK));
643 if (key.type == BTRFS_EXTENT_ITEM_KEY &&
644 item_size <= sizeof(*ei) + sizeof(*bi)) {
645 WARN_ON(item_size < sizeof(*ei) + sizeof(*bi));
646 return 1;
648 if (key.type == BTRFS_METADATA_ITEM_KEY &&
649 item_size <= sizeof(*ei)) {
650 WARN_ON(item_size < sizeof(*ei));
651 return 1;
654 if (key.type == BTRFS_EXTENT_ITEM_KEY) {
655 bi = (struct btrfs_tree_block_info *)(ei + 1);
656 *ptr = (unsigned long)(bi + 1);
657 } else {
658 *ptr = (unsigned long)(ei + 1);
660 *end = (unsigned long)ei + item_size;
661 return 0;
665 * build backref tree for a given tree block. root of the backref tree
666 * corresponds the tree block, leaves of the backref tree correspond
667 * roots of b-trees that reference the tree block.
669 * the basic idea of this function is check backrefs of a given block
670 * to find upper level blocks that refernece the block, and then check
671 * bakcrefs of these upper level blocks recursively. the recursion stop
672 * when tree root is reached or backrefs for the block is cached.
674 * NOTE: if we find backrefs for a block are cached, we know backrefs
675 * for all upper level blocks that directly/indirectly reference the
676 * block are also cached.
678 static noinline_for_stack
679 struct backref_node *build_backref_tree(struct reloc_control *rc,
680 struct btrfs_key *node_key,
681 int level, u64 bytenr)
683 struct backref_cache *cache = &rc->backref_cache;
684 struct btrfs_path *path1;
685 struct btrfs_path *path2;
686 struct extent_buffer *eb;
687 struct btrfs_root *root;
688 struct backref_node *cur;
689 struct backref_node *upper;
690 struct backref_node *lower;
691 struct backref_node *node = NULL;
692 struct backref_node *exist = NULL;
693 struct backref_edge *edge;
694 struct rb_node *rb_node;
695 struct btrfs_key key;
696 unsigned long end;
697 unsigned long ptr;
698 LIST_HEAD(list);
699 LIST_HEAD(useless);
700 int cowonly;
701 int ret;
702 int err = 0;
703 bool need_check = true;
705 path1 = btrfs_alloc_path();
706 path2 = btrfs_alloc_path();
707 if (!path1 || !path2) {
708 err = -ENOMEM;
709 goto out;
711 path1->reada = 1;
712 path2->reada = 2;
714 node = alloc_backref_node(cache);
715 if (!node) {
716 err = -ENOMEM;
717 goto out;
720 node->bytenr = bytenr;
721 node->level = level;
722 node->lowest = 1;
723 cur = node;
724 again:
725 end = 0;
726 ptr = 0;
727 key.objectid = cur->bytenr;
728 key.type = BTRFS_METADATA_ITEM_KEY;
729 key.offset = (u64)-1;
731 path1->search_commit_root = 1;
732 path1->skip_locking = 1;
733 ret = btrfs_search_slot(NULL, rc->extent_root, &key, path1,
734 0, 0);
735 if (ret < 0) {
736 err = ret;
737 goto out;
739 ASSERT(ret);
740 ASSERT(path1->slots[0]);
742 path1->slots[0]--;
744 WARN_ON(cur->checked);
745 if (!list_empty(&cur->upper)) {
747 * the backref was added previously when processing
748 * backref of type BTRFS_TREE_BLOCK_REF_KEY
750 ASSERT(list_is_singular(&cur->upper));
751 edge = list_entry(cur->upper.next, struct backref_edge,
752 list[LOWER]);
753 ASSERT(list_empty(&edge->list[UPPER]));
754 exist = edge->node[UPPER];
756 * add the upper level block to pending list if we need
757 * check its backrefs
759 if (!exist->checked)
760 list_add_tail(&edge->list[UPPER], &list);
761 } else {
762 exist = NULL;
765 while (1) {
766 cond_resched();
767 eb = path1->nodes[0];
769 if (ptr >= end) {
770 if (path1->slots[0] >= btrfs_header_nritems(eb)) {
771 ret = btrfs_next_leaf(rc->extent_root, path1);
772 if (ret < 0) {
773 err = ret;
774 goto out;
776 if (ret > 0)
777 break;
778 eb = path1->nodes[0];
781 btrfs_item_key_to_cpu(eb, &key, path1->slots[0]);
782 if (key.objectid != cur->bytenr) {
783 WARN_ON(exist);
784 break;
787 if (key.type == BTRFS_EXTENT_ITEM_KEY ||
788 key.type == BTRFS_METADATA_ITEM_KEY) {
789 ret = find_inline_backref(eb, path1->slots[0],
790 &ptr, &end);
791 if (ret)
792 goto next;
796 if (ptr < end) {
797 /* update key for inline back ref */
798 struct btrfs_extent_inline_ref *iref;
799 iref = (struct btrfs_extent_inline_ref *)ptr;
800 key.type = btrfs_extent_inline_ref_type(eb, iref);
801 key.offset = btrfs_extent_inline_ref_offset(eb, iref);
802 WARN_ON(key.type != BTRFS_TREE_BLOCK_REF_KEY &&
803 key.type != BTRFS_SHARED_BLOCK_REF_KEY);
806 if (exist &&
807 ((key.type == BTRFS_TREE_BLOCK_REF_KEY &&
808 exist->owner == key.offset) ||
809 (key.type == BTRFS_SHARED_BLOCK_REF_KEY &&
810 exist->bytenr == key.offset))) {
811 exist = NULL;
812 goto next;
815 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
816 if (key.type == BTRFS_SHARED_BLOCK_REF_KEY ||
817 key.type == BTRFS_EXTENT_REF_V0_KEY) {
818 if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
819 struct btrfs_extent_ref_v0 *ref0;
820 ref0 = btrfs_item_ptr(eb, path1->slots[0],
821 struct btrfs_extent_ref_v0);
822 if (key.objectid == key.offset) {
823 root = find_tree_root(rc, eb, ref0);
824 if (root && !should_ignore_root(root))
825 cur->root = root;
826 else
827 list_add(&cur->list, &useless);
828 break;
830 if (is_cowonly_root(btrfs_ref_root_v0(eb,
831 ref0)))
832 cur->cowonly = 1;
834 #else
835 ASSERT(key.type != BTRFS_EXTENT_REF_V0_KEY);
836 if (key.type == BTRFS_SHARED_BLOCK_REF_KEY) {
837 #endif
838 if (key.objectid == key.offset) {
840 * only root blocks of reloc trees use
841 * backref of this type.
843 root = find_reloc_root(rc, cur->bytenr);
844 ASSERT(root);
845 cur->root = root;
846 break;
849 edge = alloc_backref_edge(cache);
850 if (!edge) {
851 err = -ENOMEM;
852 goto out;
854 rb_node = tree_search(&cache->rb_root, key.offset);
855 if (!rb_node) {
856 upper = alloc_backref_node(cache);
857 if (!upper) {
858 free_backref_edge(cache, edge);
859 err = -ENOMEM;
860 goto out;
862 upper->bytenr = key.offset;
863 upper->level = cur->level + 1;
865 * backrefs for the upper level block isn't
866 * cached, add the block to pending list
868 list_add_tail(&edge->list[UPPER], &list);
869 } else {
870 upper = rb_entry(rb_node, struct backref_node,
871 rb_node);
872 ASSERT(upper->checked);
873 INIT_LIST_HEAD(&edge->list[UPPER]);
875 list_add_tail(&edge->list[LOWER], &cur->upper);
876 edge->node[LOWER] = cur;
877 edge->node[UPPER] = upper;
879 goto next;
880 } else if (key.type != BTRFS_TREE_BLOCK_REF_KEY) {
881 goto next;
884 /* key.type == BTRFS_TREE_BLOCK_REF_KEY */
885 root = read_fs_root(rc->extent_root->fs_info, key.offset);
886 if (IS_ERR(root)) {
887 err = PTR_ERR(root);
888 goto out;
891 if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state))
892 cur->cowonly = 1;
894 if (btrfs_root_level(&root->root_item) == cur->level) {
895 /* tree root */
896 ASSERT(btrfs_root_bytenr(&root->root_item) ==
897 cur->bytenr);
898 if (should_ignore_root(root))
899 list_add(&cur->list, &useless);
900 else
901 cur->root = root;
902 break;
905 level = cur->level + 1;
908 * searching the tree to find upper level blocks
909 * reference the block.
911 path2->search_commit_root = 1;
912 path2->skip_locking = 1;
913 path2->lowest_level = level;
914 ret = btrfs_search_slot(NULL, root, node_key, path2, 0, 0);
915 path2->lowest_level = 0;
916 if (ret < 0) {
917 err = ret;
918 goto out;
920 if (ret > 0 && path2->slots[level] > 0)
921 path2->slots[level]--;
923 eb = path2->nodes[level];
924 if (btrfs_node_blockptr(eb, path2->slots[level]) !=
925 cur->bytenr) {
926 btrfs_err(root->fs_info,
927 "couldn't find block (%llu) (level %d) in tree (%llu) with key (%llu %u %llu)",
928 cur->bytenr, level - 1, root->objectid,
929 node_key->objectid, node_key->type,
930 node_key->offset);
931 err = -ENOENT;
932 goto out;
934 lower = cur;
935 need_check = true;
936 for (; level < BTRFS_MAX_LEVEL; level++) {
937 if (!path2->nodes[level]) {
938 ASSERT(btrfs_root_bytenr(&root->root_item) ==
939 lower->bytenr);
940 if (should_ignore_root(root))
941 list_add(&lower->list, &useless);
942 else
943 lower->root = root;
944 break;
947 edge = alloc_backref_edge(cache);
948 if (!edge) {
949 err = -ENOMEM;
950 goto out;
953 eb = path2->nodes[level];
954 rb_node = tree_search(&cache->rb_root, eb->start);
955 if (!rb_node) {
956 upper = alloc_backref_node(cache);
957 if (!upper) {
958 free_backref_edge(cache, edge);
959 err = -ENOMEM;
960 goto out;
962 upper->bytenr = eb->start;
963 upper->owner = btrfs_header_owner(eb);
964 upper->level = lower->level + 1;
965 if (!test_bit(BTRFS_ROOT_REF_COWS,
966 &root->state))
967 upper->cowonly = 1;
970 * if we know the block isn't shared
971 * we can void checking its backrefs.
973 if (btrfs_block_can_be_shared(root, eb))
974 upper->checked = 0;
975 else
976 upper->checked = 1;
979 * add the block to pending list if we
980 * need check its backrefs, we only do this once
981 * while walking up a tree as we will catch
982 * anything else later on.
984 if (!upper->checked && need_check) {
985 need_check = false;
986 list_add_tail(&edge->list[UPPER],
987 &list);
988 } else {
989 if (upper->checked)
990 need_check = true;
991 INIT_LIST_HEAD(&edge->list[UPPER]);
993 } else {
994 upper = rb_entry(rb_node, struct backref_node,
995 rb_node);
996 ASSERT(upper->checked);
997 INIT_LIST_HEAD(&edge->list[UPPER]);
998 if (!upper->owner)
999 upper->owner = btrfs_header_owner(eb);
1001 list_add_tail(&edge->list[LOWER], &lower->upper);
1002 edge->node[LOWER] = lower;
1003 edge->node[UPPER] = upper;
1005 if (rb_node)
1006 break;
1007 lower = upper;
1008 upper = NULL;
1010 btrfs_release_path(path2);
1011 next:
1012 if (ptr < end) {
1013 ptr += btrfs_extent_inline_ref_size(key.type);
1014 if (ptr >= end) {
1015 WARN_ON(ptr > end);
1016 ptr = 0;
1017 end = 0;
1020 if (ptr >= end)
1021 path1->slots[0]++;
1023 btrfs_release_path(path1);
1025 cur->checked = 1;
1026 WARN_ON(exist);
1028 /* the pending list isn't empty, take the first block to process */
1029 if (!list_empty(&list)) {
1030 edge = list_entry(list.next, struct backref_edge, list[UPPER]);
1031 list_del_init(&edge->list[UPPER]);
1032 cur = edge->node[UPPER];
1033 goto again;
1037 * everything goes well, connect backref nodes and insert backref nodes
1038 * into the cache.
1040 ASSERT(node->checked);
1041 cowonly = node->cowonly;
1042 if (!cowonly) {
1043 rb_node = tree_insert(&cache->rb_root, node->bytenr,
1044 &node->rb_node);
1045 if (rb_node)
1046 backref_tree_panic(rb_node, -EEXIST, node->bytenr);
1047 list_add_tail(&node->lower, &cache->leaves);
1050 list_for_each_entry(edge, &node->upper, list[LOWER])
1051 list_add_tail(&edge->list[UPPER], &list);
1053 while (!list_empty(&list)) {
1054 edge = list_entry(list.next, struct backref_edge, list[UPPER]);
1055 list_del_init(&edge->list[UPPER]);
1056 upper = edge->node[UPPER];
1057 if (upper->detached) {
1058 list_del(&edge->list[LOWER]);
1059 lower = edge->node[LOWER];
1060 free_backref_edge(cache, edge);
1061 if (list_empty(&lower->upper))
1062 list_add(&lower->list, &useless);
1063 continue;
1066 if (!RB_EMPTY_NODE(&upper->rb_node)) {
1067 if (upper->lowest) {
1068 list_del_init(&upper->lower);
1069 upper->lowest = 0;
1072 list_add_tail(&edge->list[UPPER], &upper->lower);
1073 continue;
1076 if (!upper->checked) {
1078 * Still want to blow up for developers since this is a
1079 * logic bug.
1081 ASSERT(0);
1082 err = -EINVAL;
1083 goto out;
1085 if (cowonly != upper->cowonly) {
1086 ASSERT(0);
1087 err = -EINVAL;
1088 goto out;
1091 if (!cowonly) {
1092 rb_node = tree_insert(&cache->rb_root, upper->bytenr,
1093 &upper->rb_node);
1094 if (rb_node)
1095 backref_tree_panic(rb_node, -EEXIST,
1096 upper->bytenr);
1099 list_add_tail(&edge->list[UPPER], &upper->lower);
1101 list_for_each_entry(edge, &upper->upper, list[LOWER])
1102 list_add_tail(&edge->list[UPPER], &list);
1105 * process useless backref nodes. backref nodes for tree leaves
1106 * are deleted from the cache. backref nodes for upper level
1107 * tree blocks are left in the cache to avoid unnecessary backref
1108 * lookup.
1110 while (!list_empty(&useless)) {
1111 upper = list_entry(useless.next, struct backref_node, list);
1112 list_del_init(&upper->list);
1113 ASSERT(list_empty(&upper->upper));
1114 if (upper == node)
1115 node = NULL;
1116 if (upper->lowest) {
1117 list_del_init(&upper->lower);
1118 upper->lowest = 0;
1120 while (!list_empty(&upper->lower)) {
1121 edge = list_entry(upper->lower.next,
1122 struct backref_edge, list[UPPER]);
1123 list_del(&edge->list[UPPER]);
1124 list_del(&edge->list[LOWER]);
1125 lower = edge->node[LOWER];
1126 free_backref_edge(cache, edge);
1128 if (list_empty(&lower->upper))
1129 list_add(&lower->list, &useless);
1131 __mark_block_processed(rc, upper);
1132 if (upper->level > 0) {
1133 list_add(&upper->list, &cache->detached);
1134 upper->detached = 1;
1135 } else {
1136 rb_erase(&upper->rb_node, &cache->rb_root);
1137 free_backref_node(cache, upper);
1140 out:
1141 btrfs_free_path(path1);
1142 btrfs_free_path(path2);
1143 if (err) {
1144 while (!list_empty(&useless)) {
1145 lower = list_entry(useless.next,
1146 struct backref_node, list);
1147 list_del_init(&lower->list);
1149 while (!list_empty(&list)) {
1150 edge = list_first_entry(&list, struct backref_edge,
1151 list[UPPER]);
1152 list_del(&edge->list[UPPER]);
1153 list_del(&edge->list[LOWER]);
1154 lower = edge->node[LOWER];
1155 upper = edge->node[UPPER];
1156 free_backref_edge(cache, edge);
1159 * Lower is no longer linked to any upper backref nodes
1160 * and isn't in the cache, we can free it ourselves.
1162 if (list_empty(&lower->upper) &&
1163 RB_EMPTY_NODE(&lower->rb_node))
1164 list_add(&lower->list, &useless);
1166 if (!RB_EMPTY_NODE(&upper->rb_node))
1167 continue;
1169 /* Add this guy's upper edges to the list to proces */
1170 list_for_each_entry(edge, &upper->upper, list[LOWER])
1171 list_add_tail(&edge->list[UPPER], &list);
1172 if (list_empty(&upper->upper))
1173 list_add(&upper->list, &useless);
1176 while (!list_empty(&useless)) {
1177 lower = list_entry(useless.next,
1178 struct backref_node, list);
1179 list_del_init(&lower->list);
1180 free_backref_node(cache, lower);
1182 return ERR_PTR(err);
1184 ASSERT(!node || !node->detached);
1185 return node;
1189 * helper to add backref node for the newly created snapshot.
1190 * the backref node is created by cloning backref node that
1191 * corresponds to root of source tree
1193 static int clone_backref_node(struct btrfs_trans_handle *trans,
1194 struct reloc_control *rc,
1195 struct btrfs_root *src,
1196 struct btrfs_root *dest)
1198 struct btrfs_root *reloc_root = src->reloc_root;
1199 struct backref_cache *cache = &rc->backref_cache;
1200 struct backref_node *node = NULL;
1201 struct backref_node *new_node;
1202 struct backref_edge *edge;
1203 struct backref_edge *new_edge;
1204 struct rb_node *rb_node;
1206 if (cache->last_trans > 0)
1207 update_backref_cache(trans, cache);
1209 rb_node = tree_search(&cache->rb_root, src->commit_root->start);
1210 if (rb_node) {
1211 node = rb_entry(rb_node, struct backref_node, rb_node);
1212 if (node->detached)
1213 node = NULL;
1214 else
1215 BUG_ON(node->new_bytenr != reloc_root->node->start);
1218 if (!node) {
1219 rb_node = tree_search(&cache->rb_root,
1220 reloc_root->commit_root->start);
1221 if (rb_node) {
1222 node = rb_entry(rb_node, struct backref_node,
1223 rb_node);
1224 BUG_ON(node->detached);
1228 if (!node)
1229 return 0;
1231 new_node = alloc_backref_node(cache);
1232 if (!new_node)
1233 return -ENOMEM;
1235 new_node->bytenr = dest->node->start;
1236 new_node->level = node->level;
1237 new_node->lowest = node->lowest;
1238 new_node->checked = 1;
1239 new_node->root = dest;
1241 if (!node->lowest) {
1242 list_for_each_entry(edge, &node->lower, list[UPPER]) {
1243 new_edge = alloc_backref_edge(cache);
1244 if (!new_edge)
1245 goto fail;
1247 new_edge->node[UPPER] = new_node;
1248 new_edge->node[LOWER] = edge->node[LOWER];
1249 list_add_tail(&new_edge->list[UPPER],
1250 &new_node->lower);
1252 } else {
1253 list_add_tail(&new_node->lower, &cache->leaves);
1256 rb_node = tree_insert(&cache->rb_root, new_node->bytenr,
1257 &new_node->rb_node);
1258 if (rb_node)
1259 backref_tree_panic(rb_node, -EEXIST, new_node->bytenr);
1261 if (!new_node->lowest) {
1262 list_for_each_entry(new_edge, &new_node->lower, list[UPPER]) {
1263 list_add_tail(&new_edge->list[LOWER],
1264 &new_edge->node[LOWER]->upper);
1267 return 0;
1268 fail:
1269 while (!list_empty(&new_node->lower)) {
1270 new_edge = list_entry(new_node->lower.next,
1271 struct backref_edge, list[UPPER]);
1272 list_del(&new_edge->list[UPPER]);
1273 free_backref_edge(cache, new_edge);
1275 free_backref_node(cache, new_node);
1276 return -ENOMEM;
1280 * helper to add 'address of tree root -> reloc tree' mapping
1282 static int __must_check __add_reloc_root(struct btrfs_root *root)
1284 struct rb_node *rb_node;
1285 struct mapping_node *node;
1286 struct reloc_control *rc = root->fs_info->reloc_ctl;
1288 node = kmalloc(sizeof(*node), GFP_NOFS);
1289 if (!node)
1290 return -ENOMEM;
1292 node->bytenr = root->node->start;
1293 node->data = root;
1295 spin_lock(&rc->reloc_root_tree.lock);
1296 rb_node = tree_insert(&rc->reloc_root_tree.rb_root,
1297 node->bytenr, &node->rb_node);
1298 spin_unlock(&rc->reloc_root_tree.lock);
1299 if (rb_node) {
1300 btrfs_panic(root->fs_info, -EEXIST, "Duplicate root found "
1301 "for start=%llu while inserting into relocation "
1302 "tree", node->bytenr);
1303 kfree(node);
1304 return -EEXIST;
1307 list_add_tail(&root->root_list, &rc->reloc_roots);
1308 return 0;
1312 * helper to delete the 'address of tree root -> reloc tree'
1313 * mapping
1315 static void __del_reloc_root(struct btrfs_root *root)
1317 struct rb_node *rb_node;
1318 struct mapping_node *node = NULL;
1319 struct reloc_control *rc = root->fs_info->reloc_ctl;
1321 if (rc) {
1322 spin_lock(&rc->reloc_root_tree.lock);
1323 rb_node = tree_search(&rc->reloc_root_tree.rb_root,
1324 root->node->start);
1325 if (rb_node) {
1326 node = rb_entry(rb_node, struct mapping_node, rb_node);
1327 rb_erase(&node->rb_node, &rc->reloc_root_tree.rb_root);
1329 spin_unlock(&rc->reloc_root_tree.lock);
1330 if (!node)
1331 return;
1332 BUG_ON((struct btrfs_root *)node->data != root);
1335 spin_lock(&root->fs_info->trans_lock);
1336 list_del_init(&root->root_list);
1337 spin_unlock(&root->fs_info->trans_lock);
1338 kfree(node);
1342 * helper to update the 'address of tree root -> reloc tree'
1343 * mapping
1345 static int __update_reloc_root(struct btrfs_root *root, u64 new_bytenr)
1347 struct rb_node *rb_node;
1348 struct mapping_node *node = NULL;
1349 struct reloc_control *rc = root->fs_info->reloc_ctl;
1351 spin_lock(&rc->reloc_root_tree.lock);
1352 rb_node = tree_search(&rc->reloc_root_tree.rb_root,
1353 root->node->start);
1354 if (rb_node) {
1355 node = rb_entry(rb_node, struct mapping_node, rb_node);
1356 rb_erase(&node->rb_node, &rc->reloc_root_tree.rb_root);
1358 spin_unlock(&rc->reloc_root_tree.lock);
1360 if (!node)
1361 return 0;
1362 BUG_ON((struct btrfs_root *)node->data != root);
1364 spin_lock(&rc->reloc_root_tree.lock);
1365 node->bytenr = new_bytenr;
1366 rb_node = tree_insert(&rc->reloc_root_tree.rb_root,
1367 node->bytenr, &node->rb_node);
1368 spin_unlock(&rc->reloc_root_tree.lock);
1369 if (rb_node)
1370 backref_tree_panic(rb_node, -EEXIST, node->bytenr);
1371 return 0;
1374 static struct btrfs_root *create_reloc_root(struct btrfs_trans_handle *trans,
1375 struct btrfs_root *root, u64 objectid)
1377 struct btrfs_root *reloc_root;
1378 struct extent_buffer *eb;
1379 struct btrfs_root_item *root_item;
1380 struct btrfs_key root_key;
1381 u64 last_snap = 0;
1382 int ret;
1384 root_item = kmalloc(sizeof(*root_item), GFP_NOFS);
1385 BUG_ON(!root_item);
1387 root_key.objectid = BTRFS_TREE_RELOC_OBJECTID;
1388 root_key.type = BTRFS_ROOT_ITEM_KEY;
1389 root_key.offset = objectid;
1391 if (root->root_key.objectid == objectid) {
1392 /* called by btrfs_init_reloc_root */
1393 ret = btrfs_copy_root(trans, root, root->commit_root, &eb,
1394 BTRFS_TREE_RELOC_OBJECTID);
1395 BUG_ON(ret);
1397 last_snap = btrfs_root_last_snapshot(&root->root_item);
1398 btrfs_set_root_last_snapshot(&root->root_item,
1399 trans->transid - 1);
1400 } else {
1402 * called by btrfs_reloc_post_snapshot_hook.
1403 * the source tree is a reloc tree, all tree blocks
1404 * modified after it was created have RELOC flag
1405 * set in their headers. so it's OK to not update
1406 * the 'last_snapshot'.
1408 ret = btrfs_copy_root(trans, root, root->node, &eb,
1409 BTRFS_TREE_RELOC_OBJECTID);
1410 BUG_ON(ret);
1413 memcpy(root_item, &root->root_item, sizeof(*root_item));
1414 btrfs_set_root_bytenr(root_item, eb->start);
1415 btrfs_set_root_level(root_item, btrfs_header_level(eb));
1416 btrfs_set_root_generation(root_item, trans->transid);
1418 if (root->root_key.objectid == objectid) {
1419 btrfs_set_root_refs(root_item, 0);
1420 memset(&root_item->drop_progress, 0,
1421 sizeof(struct btrfs_disk_key));
1422 root_item->drop_level = 0;
1424 * abuse rtransid, it is safe because it is impossible to
1425 * receive data into a relocation tree.
1427 btrfs_set_root_rtransid(root_item, last_snap);
1428 btrfs_set_root_otransid(root_item, trans->transid);
1431 btrfs_tree_unlock(eb);
1432 free_extent_buffer(eb);
1434 ret = btrfs_insert_root(trans, root->fs_info->tree_root,
1435 &root_key, root_item);
1436 BUG_ON(ret);
1437 kfree(root_item);
1439 reloc_root = btrfs_read_fs_root(root->fs_info->tree_root, &root_key);
1440 BUG_ON(IS_ERR(reloc_root));
1441 reloc_root->last_trans = trans->transid;
1442 return reloc_root;
1446 * create reloc tree for a given fs tree. reloc tree is just a
1447 * snapshot of the fs tree with special root objectid.
1449 int btrfs_init_reloc_root(struct btrfs_trans_handle *trans,
1450 struct btrfs_root *root)
1452 struct btrfs_root *reloc_root;
1453 struct reloc_control *rc = root->fs_info->reloc_ctl;
1454 struct btrfs_block_rsv *rsv;
1455 int clear_rsv = 0;
1456 int ret;
1458 if (root->reloc_root) {
1459 reloc_root = root->reloc_root;
1460 reloc_root->last_trans = trans->transid;
1461 return 0;
1464 if (!rc || !rc->create_reloc_tree ||
1465 root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID)
1466 return 0;
1468 if (!trans->reloc_reserved) {
1469 rsv = trans->block_rsv;
1470 trans->block_rsv = rc->block_rsv;
1471 clear_rsv = 1;
1473 reloc_root = create_reloc_root(trans, root, root->root_key.objectid);
1474 if (clear_rsv)
1475 trans->block_rsv = rsv;
1477 ret = __add_reloc_root(reloc_root);
1478 BUG_ON(ret < 0);
1479 root->reloc_root = reloc_root;
1480 return 0;
1484 * update root item of reloc tree
1486 int btrfs_update_reloc_root(struct btrfs_trans_handle *trans,
1487 struct btrfs_root *root)
1489 struct btrfs_root *reloc_root;
1490 struct btrfs_root_item *root_item;
1491 int ret;
1493 if (!root->reloc_root)
1494 goto out;
1496 reloc_root = root->reloc_root;
1497 root_item = &reloc_root->root_item;
1499 if (root->fs_info->reloc_ctl->merge_reloc_tree &&
1500 btrfs_root_refs(root_item) == 0) {
1501 root->reloc_root = NULL;
1502 __del_reloc_root(reloc_root);
1505 if (reloc_root->commit_root != reloc_root->node) {
1506 btrfs_set_root_node(root_item, reloc_root->node);
1507 free_extent_buffer(reloc_root->commit_root);
1508 reloc_root->commit_root = btrfs_root_node(reloc_root);
1511 ret = btrfs_update_root(trans, root->fs_info->tree_root,
1512 &reloc_root->root_key, root_item);
1513 BUG_ON(ret);
1515 out:
1516 return 0;
1520 * helper to find first cached inode with inode number >= objectid
1521 * in a subvolume
1523 static struct inode *find_next_inode(struct btrfs_root *root, u64 objectid)
1525 struct rb_node *node;
1526 struct rb_node *prev;
1527 struct btrfs_inode *entry;
1528 struct inode *inode;
1530 spin_lock(&root->inode_lock);
1531 again:
1532 node = root->inode_tree.rb_node;
1533 prev = NULL;
1534 while (node) {
1535 prev = node;
1536 entry = rb_entry(node, struct btrfs_inode, rb_node);
1538 if (objectid < btrfs_ino(&entry->vfs_inode))
1539 node = node->rb_left;
1540 else if (objectid > btrfs_ino(&entry->vfs_inode))
1541 node = node->rb_right;
1542 else
1543 break;
1545 if (!node) {
1546 while (prev) {
1547 entry = rb_entry(prev, struct btrfs_inode, rb_node);
1548 if (objectid <= btrfs_ino(&entry->vfs_inode)) {
1549 node = prev;
1550 break;
1552 prev = rb_next(prev);
1555 while (node) {
1556 entry = rb_entry(node, struct btrfs_inode, rb_node);
1557 inode = igrab(&entry->vfs_inode);
1558 if (inode) {
1559 spin_unlock(&root->inode_lock);
1560 return inode;
1563 objectid = btrfs_ino(&entry->vfs_inode) + 1;
1564 if (cond_resched_lock(&root->inode_lock))
1565 goto again;
1567 node = rb_next(node);
1569 spin_unlock(&root->inode_lock);
1570 return NULL;
1573 static int in_block_group(u64 bytenr,
1574 struct btrfs_block_group_cache *block_group)
1576 if (bytenr >= block_group->key.objectid &&
1577 bytenr < block_group->key.objectid + block_group->key.offset)
1578 return 1;
1579 return 0;
1583 * get new location of data
1585 static int get_new_location(struct inode *reloc_inode, u64 *new_bytenr,
1586 u64 bytenr, u64 num_bytes)
1588 struct btrfs_root *root = BTRFS_I(reloc_inode)->root;
1589 struct btrfs_path *path;
1590 struct btrfs_file_extent_item *fi;
1591 struct extent_buffer *leaf;
1592 int ret;
1594 path = btrfs_alloc_path();
1595 if (!path)
1596 return -ENOMEM;
1598 bytenr -= BTRFS_I(reloc_inode)->index_cnt;
1599 ret = btrfs_lookup_file_extent(NULL, root, path, btrfs_ino(reloc_inode),
1600 bytenr, 0);
1601 if (ret < 0)
1602 goto out;
1603 if (ret > 0) {
1604 ret = -ENOENT;
1605 goto out;
1608 leaf = path->nodes[0];
1609 fi = btrfs_item_ptr(leaf, path->slots[0],
1610 struct btrfs_file_extent_item);
1612 BUG_ON(btrfs_file_extent_offset(leaf, fi) ||
1613 btrfs_file_extent_compression(leaf, fi) ||
1614 btrfs_file_extent_encryption(leaf, fi) ||
1615 btrfs_file_extent_other_encoding(leaf, fi));
1617 if (num_bytes != btrfs_file_extent_disk_num_bytes(leaf, fi)) {
1618 ret = -EINVAL;
1619 goto out;
1622 *new_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1623 ret = 0;
1624 out:
1625 btrfs_free_path(path);
1626 return ret;
1630 * update file extent items in the tree leaf to point to
1631 * the new locations.
1633 static noinline_for_stack
1634 int replace_file_extents(struct btrfs_trans_handle *trans,
1635 struct reloc_control *rc,
1636 struct btrfs_root *root,
1637 struct extent_buffer *leaf)
1639 struct btrfs_key key;
1640 struct btrfs_file_extent_item *fi;
1641 struct inode *inode = NULL;
1642 u64 parent;
1643 u64 bytenr;
1644 u64 new_bytenr = 0;
1645 u64 num_bytes;
1646 u64 end;
1647 u32 nritems;
1648 u32 i;
1649 int ret = 0;
1650 int first = 1;
1651 int dirty = 0;
1653 if (rc->stage != UPDATE_DATA_PTRS)
1654 return 0;
1656 /* reloc trees always use full backref */
1657 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID)
1658 parent = leaf->start;
1659 else
1660 parent = 0;
1662 nritems = btrfs_header_nritems(leaf);
1663 for (i = 0; i < nritems; i++) {
1664 cond_resched();
1665 btrfs_item_key_to_cpu(leaf, &key, i);
1666 if (key.type != BTRFS_EXTENT_DATA_KEY)
1667 continue;
1668 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
1669 if (btrfs_file_extent_type(leaf, fi) ==
1670 BTRFS_FILE_EXTENT_INLINE)
1671 continue;
1672 bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1673 num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
1674 if (bytenr == 0)
1675 continue;
1676 if (!in_block_group(bytenr, rc->block_group))
1677 continue;
1680 * if we are modifying block in fs tree, wait for readpage
1681 * to complete and drop the extent cache
1683 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
1684 if (first) {
1685 inode = find_next_inode(root, key.objectid);
1686 first = 0;
1687 } else if (inode && btrfs_ino(inode) < key.objectid) {
1688 btrfs_add_delayed_iput(inode);
1689 inode = find_next_inode(root, key.objectid);
1691 if (inode && btrfs_ino(inode) == key.objectid) {
1692 end = key.offset +
1693 btrfs_file_extent_num_bytes(leaf, fi);
1694 WARN_ON(!IS_ALIGNED(key.offset,
1695 root->sectorsize));
1696 WARN_ON(!IS_ALIGNED(end, root->sectorsize));
1697 end--;
1698 ret = try_lock_extent(&BTRFS_I(inode)->io_tree,
1699 key.offset, end);
1700 if (!ret)
1701 continue;
1703 btrfs_drop_extent_cache(inode, key.offset, end,
1705 unlock_extent(&BTRFS_I(inode)->io_tree,
1706 key.offset, end);
1710 ret = get_new_location(rc->data_inode, &new_bytenr,
1711 bytenr, num_bytes);
1712 if (ret) {
1714 * Don't have to abort since we've not changed anything
1715 * in the file extent yet.
1717 break;
1720 btrfs_set_file_extent_disk_bytenr(leaf, fi, new_bytenr);
1721 dirty = 1;
1723 key.offset -= btrfs_file_extent_offset(leaf, fi);
1724 ret = btrfs_inc_extent_ref(trans, root, new_bytenr,
1725 num_bytes, parent,
1726 btrfs_header_owner(leaf),
1727 key.objectid, key.offset);
1728 if (ret) {
1729 btrfs_abort_transaction(trans, root, ret);
1730 break;
1733 ret = btrfs_free_extent(trans, root, bytenr, num_bytes,
1734 parent, btrfs_header_owner(leaf),
1735 key.objectid, key.offset);
1736 if (ret) {
1737 btrfs_abort_transaction(trans, root, ret);
1738 break;
1741 if (dirty)
1742 btrfs_mark_buffer_dirty(leaf);
1743 if (inode)
1744 btrfs_add_delayed_iput(inode);
1745 return ret;
1748 static noinline_for_stack
1749 int memcmp_node_keys(struct extent_buffer *eb, int slot,
1750 struct btrfs_path *path, int level)
1752 struct btrfs_disk_key key1;
1753 struct btrfs_disk_key key2;
1754 btrfs_node_key(eb, &key1, slot);
1755 btrfs_node_key(path->nodes[level], &key2, path->slots[level]);
1756 return memcmp(&key1, &key2, sizeof(key1));
1760 * try to replace tree blocks in fs tree with the new blocks
1761 * in reloc tree. tree blocks haven't been modified since the
1762 * reloc tree was create can be replaced.
1764 * if a block was replaced, level of the block + 1 is returned.
1765 * if no block got replaced, 0 is returned. if there are other
1766 * errors, a negative error number is returned.
1768 static noinline_for_stack
1769 int replace_path(struct btrfs_trans_handle *trans,
1770 struct btrfs_root *dest, struct btrfs_root *src,
1771 struct btrfs_path *path, struct btrfs_key *next_key,
1772 int lowest_level, int max_level)
1774 struct extent_buffer *eb;
1775 struct extent_buffer *parent;
1776 struct btrfs_key key;
1777 u64 old_bytenr;
1778 u64 new_bytenr;
1779 u64 old_ptr_gen;
1780 u64 new_ptr_gen;
1781 u64 last_snapshot;
1782 u32 blocksize;
1783 int cow = 0;
1784 int level;
1785 int ret;
1786 int slot;
1788 BUG_ON(src->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
1789 BUG_ON(dest->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID);
1791 last_snapshot = btrfs_root_last_snapshot(&src->root_item);
1792 again:
1793 slot = path->slots[lowest_level];
1794 btrfs_node_key_to_cpu(path->nodes[lowest_level], &key, slot);
1796 eb = btrfs_lock_root_node(dest);
1797 btrfs_set_lock_blocking(eb);
1798 level = btrfs_header_level(eb);
1800 if (level < lowest_level) {
1801 btrfs_tree_unlock(eb);
1802 free_extent_buffer(eb);
1803 return 0;
1806 if (cow) {
1807 ret = btrfs_cow_block(trans, dest, eb, NULL, 0, &eb);
1808 BUG_ON(ret);
1810 btrfs_set_lock_blocking(eb);
1812 if (next_key) {
1813 next_key->objectid = (u64)-1;
1814 next_key->type = (u8)-1;
1815 next_key->offset = (u64)-1;
1818 parent = eb;
1819 while (1) {
1820 level = btrfs_header_level(parent);
1821 BUG_ON(level < lowest_level);
1823 ret = btrfs_bin_search(parent, &key, level, &slot);
1824 if (ret && slot > 0)
1825 slot--;
1827 if (next_key && slot + 1 < btrfs_header_nritems(parent))
1828 btrfs_node_key_to_cpu(parent, next_key, slot + 1);
1830 old_bytenr = btrfs_node_blockptr(parent, slot);
1831 blocksize = dest->nodesize;
1832 old_ptr_gen = btrfs_node_ptr_generation(parent, slot);
1834 if (level <= max_level) {
1835 eb = path->nodes[level];
1836 new_bytenr = btrfs_node_blockptr(eb,
1837 path->slots[level]);
1838 new_ptr_gen = btrfs_node_ptr_generation(eb,
1839 path->slots[level]);
1840 } else {
1841 new_bytenr = 0;
1842 new_ptr_gen = 0;
1845 if (WARN_ON(new_bytenr > 0 && new_bytenr == old_bytenr)) {
1846 ret = level;
1847 break;
1850 if (new_bytenr == 0 || old_ptr_gen > last_snapshot ||
1851 memcmp_node_keys(parent, slot, path, level)) {
1852 if (level <= lowest_level) {
1853 ret = 0;
1854 break;
1857 eb = read_tree_block(dest, old_bytenr, old_ptr_gen);
1858 if (IS_ERR(eb)) {
1859 ret = PTR_ERR(eb);
1860 } else if (!extent_buffer_uptodate(eb)) {
1861 ret = -EIO;
1862 free_extent_buffer(eb);
1863 break;
1865 btrfs_tree_lock(eb);
1866 if (cow) {
1867 ret = btrfs_cow_block(trans, dest, eb, parent,
1868 slot, &eb);
1869 BUG_ON(ret);
1871 btrfs_set_lock_blocking(eb);
1873 btrfs_tree_unlock(parent);
1874 free_extent_buffer(parent);
1876 parent = eb;
1877 continue;
1880 if (!cow) {
1881 btrfs_tree_unlock(parent);
1882 free_extent_buffer(parent);
1883 cow = 1;
1884 goto again;
1887 btrfs_node_key_to_cpu(path->nodes[level], &key,
1888 path->slots[level]);
1889 btrfs_release_path(path);
1891 path->lowest_level = level;
1892 ret = btrfs_search_slot(trans, src, &key, path, 0, 1);
1893 path->lowest_level = 0;
1894 BUG_ON(ret);
1897 * swap blocks in fs tree and reloc tree.
1899 btrfs_set_node_blockptr(parent, slot, new_bytenr);
1900 btrfs_set_node_ptr_generation(parent, slot, new_ptr_gen);
1901 btrfs_mark_buffer_dirty(parent);
1903 btrfs_set_node_blockptr(path->nodes[level],
1904 path->slots[level], old_bytenr);
1905 btrfs_set_node_ptr_generation(path->nodes[level],
1906 path->slots[level], old_ptr_gen);
1907 btrfs_mark_buffer_dirty(path->nodes[level]);
1909 ret = btrfs_inc_extent_ref(trans, src, old_bytenr, blocksize,
1910 path->nodes[level]->start,
1911 src->root_key.objectid, level - 1, 0);
1912 BUG_ON(ret);
1913 ret = btrfs_inc_extent_ref(trans, dest, new_bytenr, blocksize,
1914 0, dest->root_key.objectid, level - 1,
1916 BUG_ON(ret);
1918 ret = btrfs_free_extent(trans, src, new_bytenr, blocksize,
1919 path->nodes[level]->start,
1920 src->root_key.objectid, level - 1, 0);
1921 BUG_ON(ret);
1923 ret = btrfs_free_extent(trans, dest, old_bytenr, blocksize,
1924 0, dest->root_key.objectid, level - 1,
1926 BUG_ON(ret);
1928 btrfs_unlock_up_safe(path, 0);
1930 ret = level;
1931 break;
1933 btrfs_tree_unlock(parent);
1934 free_extent_buffer(parent);
1935 return ret;
1939 * helper to find next relocated block in reloc tree
1941 static noinline_for_stack
1942 int walk_up_reloc_tree(struct btrfs_root *root, struct btrfs_path *path,
1943 int *level)
1945 struct extent_buffer *eb;
1946 int i;
1947 u64 last_snapshot;
1948 u32 nritems;
1950 last_snapshot = btrfs_root_last_snapshot(&root->root_item);
1952 for (i = 0; i < *level; i++) {
1953 free_extent_buffer(path->nodes[i]);
1954 path->nodes[i] = NULL;
1957 for (i = *level; i < BTRFS_MAX_LEVEL && path->nodes[i]; i++) {
1958 eb = path->nodes[i];
1959 nritems = btrfs_header_nritems(eb);
1960 while (path->slots[i] + 1 < nritems) {
1961 path->slots[i]++;
1962 if (btrfs_node_ptr_generation(eb, path->slots[i]) <=
1963 last_snapshot)
1964 continue;
1966 *level = i;
1967 return 0;
1969 free_extent_buffer(path->nodes[i]);
1970 path->nodes[i] = NULL;
1972 return 1;
1976 * walk down reloc tree to find relocated block of lowest level
1978 static noinline_for_stack
1979 int walk_down_reloc_tree(struct btrfs_root *root, struct btrfs_path *path,
1980 int *level)
1982 struct extent_buffer *eb = NULL;
1983 int i;
1984 u64 bytenr;
1985 u64 ptr_gen = 0;
1986 u64 last_snapshot;
1987 u32 nritems;
1989 last_snapshot = btrfs_root_last_snapshot(&root->root_item);
1991 for (i = *level; i > 0; i--) {
1992 eb = path->nodes[i];
1993 nritems = btrfs_header_nritems(eb);
1994 while (path->slots[i] < nritems) {
1995 ptr_gen = btrfs_node_ptr_generation(eb, path->slots[i]);
1996 if (ptr_gen > last_snapshot)
1997 break;
1998 path->slots[i]++;
2000 if (path->slots[i] >= nritems) {
2001 if (i == *level)
2002 break;
2003 *level = i + 1;
2004 return 0;
2006 if (i == 1) {
2007 *level = i;
2008 return 0;
2011 bytenr = btrfs_node_blockptr(eb, path->slots[i]);
2012 eb = read_tree_block(root, bytenr, ptr_gen);
2013 if (IS_ERR(eb)) {
2014 return PTR_ERR(eb);
2015 } else if (!extent_buffer_uptodate(eb)) {
2016 free_extent_buffer(eb);
2017 return -EIO;
2019 BUG_ON(btrfs_header_level(eb) != i - 1);
2020 path->nodes[i - 1] = eb;
2021 path->slots[i - 1] = 0;
2023 return 1;
2027 * invalidate extent cache for file extents whose key in range of
2028 * [min_key, max_key)
2030 static int invalidate_extent_cache(struct btrfs_root *root,
2031 struct btrfs_key *min_key,
2032 struct btrfs_key *max_key)
2034 struct inode *inode = NULL;
2035 u64 objectid;
2036 u64 start, end;
2037 u64 ino;
2039 objectid = min_key->objectid;
2040 while (1) {
2041 cond_resched();
2042 iput(inode);
2044 if (objectid > max_key->objectid)
2045 break;
2047 inode = find_next_inode(root, objectid);
2048 if (!inode)
2049 break;
2050 ino = btrfs_ino(inode);
2052 if (ino > max_key->objectid) {
2053 iput(inode);
2054 break;
2057 objectid = ino + 1;
2058 if (!S_ISREG(inode->i_mode))
2059 continue;
2061 if (unlikely(min_key->objectid == ino)) {
2062 if (min_key->type > BTRFS_EXTENT_DATA_KEY)
2063 continue;
2064 if (min_key->type < BTRFS_EXTENT_DATA_KEY)
2065 start = 0;
2066 else {
2067 start = min_key->offset;
2068 WARN_ON(!IS_ALIGNED(start, root->sectorsize));
2070 } else {
2071 start = 0;
2074 if (unlikely(max_key->objectid == ino)) {
2075 if (max_key->type < BTRFS_EXTENT_DATA_KEY)
2076 continue;
2077 if (max_key->type > BTRFS_EXTENT_DATA_KEY) {
2078 end = (u64)-1;
2079 } else {
2080 if (max_key->offset == 0)
2081 continue;
2082 end = max_key->offset;
2083 WARN_ON(!IS_ALIGNED(end, root->sectorsize));
2084 end--;
2086 } else {
2087 end = (u64)-1;
2090 /* the lock_extent waits for readpage to complete */
2091 lock_extent(&BTRFS_I(inode)->io_tree, start, end);
2092 btrfs_drop_extent_cache(inode, start, end, 1);
2093 unlock_extent(&BTRFS_I(inode)->io_tree, start, end);
2095 return 0;
2098 static int find_next_key(struct btrfs_path *path, int level,
2099 struct btrfs_key *key)
2102 while (level < BTRFS_MAX_LEVEL) {
2103 if (!path->nodes[level])
2104 break;
2105 if (path->slots[level] + 1 <
2106 btrfs_header_nritems(path->nodes[level])) {
2107 btrfs_node_key_to_cpu(path->nodes[level], key,
2108 path->slots[level] + 1);
2109 return 0;
2111 level++;
2113 return 1;
2117 * merge the relocated tree blocks in reloc tree with corresponding
2118 * fs tree.
2120 static noinline_for_stack int merge_reloc_root(struct reloc_control *rc,
2121 struct btrfs_root *root)
2123 LIST_HEAD(inode_list);
2124 struct btrfs_key key;
2125 struct btrfs_key next_key;
2126 struct btrfs_trans_handle *trans = NULL;
2127 struct btrfs_root *reloc_root;
2128 struct btrfs_root_item *root_item;
2129 struct btrfs_path *path;
2130 struct extent_buffer *leaf;
2131 int level;
2132 int max_level;
2133 int replaced = 0;
2134 int ret;
2135 int err = 0;
2136 u32 min_reserved;
2138 path = btrfs_alloc_path();
2139 if (!path)
2140 return -ENOMEM;
2141 path->reada = 1;
2143 reloc_root = root->reloc_root;
2144 root_item = &reloc_root->root_item;
2146 if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
2147 level = btrfs_root_level(root_item);
2148 extent_buffer_get(reloc_root->node);
2149 path->nodes[level] = reloc_root->node;
2150 path->slots[level] = 0;
2151 } else {
2152 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
2154 level = root_item->drop_level;
2155 BUG_ON(level == 0);
2156 path->lowest_level = level;
2157 ret = btrfs_search_slot(NULL, reloc_root, &key, path, 0, 0);
2158 path->lowest_level = 0;
2159 if (ret < 0) {
2160 btrfs_free_path(path);
2161 return ret;
2164 btrfs_node_key_to_cpu(path->nodes[level], &next_key,
2165 path->slots[level]);
2166 WARN_ON(memcmp(&key, &next_key, sizeof(key)));
2168 btrfs_unlock_up_safe(path, 0);
2171 min_reserved = root->nodesize * (BTRFS_MAX_LEVEL - 1) * 2;
2172 memset(&next_key, 0, sizeof(next_key));
2174 while (1) {
2175 ret = btrfs_block_rsv_refill(root, rc->block_rsv, min_reserved,
2176 BTRFS_RESERVE_FLUSH_ALL);
2177 if (ret) {
2178 err = ret;
2179 goto out;
2181 trans = btrfs_start_transaction(root, 0);
2182 if (IS_ERR(trans)) {
2183 err = PTR_ERR(trans);
2184 trans = NULL;
2185 goto out;
2187 trans->block_rsv = rc->block_rsv;
2189 replaced = 0;
2190 max_level = level;
2192 ret = walk_down_reloc_tree(reloc_root, path, &level);
2193 if (ret < 0) {
2194 err = ret;
2195 goto out;
2197 if (ret > 0)
2198 break;
2200 if (!find_next_key(path, level, &key) &&
2201 btrfs_comp_cpu_keys(&next_key, &key) >= 0) {
2202 ret = 0;
2203 } else {
2204 ret = replace_path(trans, root, reloc_root, path,
2205 &next_key, level, max_level);
2207 if (ret < 0) {
2208 err = ret;
2209 goto out;
2212 if (ret > 0) {
2213 level = ret;
2214 btrfs_node_key_to_cpu(path->nodes[level], &key,
2215 path->slots[level]);
2216 replaced = 1;
2219 ret = walk_up_reloc_tree(reloc_root, path, &level);
2220 if (ret > 0)
2221 break;
2223 BUG_ON(level == 0);
2225 * save the merging progress in the drop_progress.
2226 * this is OK since root refs == 1 in this case.
2228 btrfs_node_key(path->nodes[level], &root_item->drop_progress,
2229 path->slots[level]);
2230 root_item->drop_level = level;
2232 btrfs_end_transaction_throttle(trans, root);
2233 trans = NULL;
2235 btrfs_btree_balance_dirty(root);
2237 if (replaced && rc->stage == UPDATE_DATA_PTRS)
2238 invalidate_extent_cache(root, &key, &next_key);
2242 * handle the case only one block in the fs tree need to be
2243 * relocated and the block is tree root.
2245 leaf = btrfs_lock_root_node(root);
2246 ret = btrfs_cow_block(trans, root, leaf, NULL, 0, &leaf);
2247 btrfs_tree_unlock(leaf);
2248 free_extent_buffer(leaf);
2249 if (ret < 0)
2250 err = ret;
2251 out:
2252 btrfs_free_path(path);
2254 if (err == 0) {
2255 memset(&root_item->drop_progress, 0,
2256 sizeof(root_item->drop_progress));
2257 root_item->drop_level = 0;
2258 btrfs_set_root_refs(root_item, 0);
2259 btrfs_update_reloc_root(trans, root);
2262 if (trans)
2263 btrfs_end_transaction_throttle(trans, root);
2265 btrfs_btree_balance_dirty(root);
2267 if (replaced && rc->stage == UPDATE_DATA_PTRS)
2268 invalidate_extent_cache(root, &key, &next_key);
2270 return err;
2273 static noinline_for_stack
2274 int prepare_to_merge(struct reloc_control *rc, int err)
2276 struct btrfs_root *root = rc->extent_root;
2277 struct btrfs_root *reloc_root;
2278 struct btrfs_trans_handle *trans;
2279 LIST_HEAD(reloc_roots);
2280 u64 num_bytes = 0;
2281 int ret;
2283 mutex_lock(&root->fs_info->reloc_mutex);
2284 rc->merging_rsv_size += root->nodesize * (BTRFS_MAX_LEVEL - 1) * 2;
2285 rc->merging_rsv_size += rc->nodes_relocated * 2;
2286 mutex_unlock(&root->fs_info->reloc_mutex);
2288 again:
2289 if (!err) {
2290 num_bytes = rc->merging_rsv_size;
2291 ret = btrfs_block_rsv_add(root, rc->block_rsv, num_bytes,
2292 BTRFS_RESERVE_FLUSH_ALL);
2293 if (ret)
2294 err = ret;
2297 trans = btrfs_join_transaction(rc->extent_root);
2298 if (IS_ERR(trans)) {
2299 if (!err)
2300 btrfs_block_rsv_release(rc->extent_root,
2301 rc->block_rsv, num_bytes);
2302 return PTR_ERR(trans);
2305 if (!err) {
2306 if (num_bytes != rc->merging_rsv_size) {
2307 btrfs_end_transaction(trans, rc->extent_root);
2308 btrfs_block_rsv_release(rc->extent_root,
2309 rc->block_rsv, num_bytes);
2310 goto again;
2314 rc->merge_reloc_tree = 1;
2316 while (!list_empty(&rc->reloc_roots)) {
2317 reloc_root = list_entry(rc->reloc_roots.next,
2318 struct btrfs_root, root_list);
2319 list_del_init(&reloc_root->root_list);
2321 root = read_fs_root(reloc_root->fs_info,
2322 reloc_root->root_key.offset);
2323 BUG_ON(IS_ERR(root));
2324 BUG_ON(root->reloc_root != reloc_root);
2327 * set reference count to 1, so btrfs_recover_relocation
2328 * knows it should resumes merging
2330 if (!err)
2331 btrfs_set_root_refs(&reloc_root->root_item, 1);
2332 btrfs_update_reloc_root(trans, root);
2334 list_add(&reloc_root->root_list, &reloc_roots);
2337 list_splice(&reloc_roots, &rc->reloc_roots);
2339 if (!err)
2340 btrfs_commit_transaction(trans, rc->extent_root);
2341 else
2342 btrfs_end_transaction(trans, rc->extent_root);
2343 return err;
2346 static noinline_for_stack
2347 void free_reloc_roots(struct list_head *list)
2349 struct btrfs_root *reloc_root;
2351 while (!list_empty(list)) {
2352 reloc_root = list_entry(list->next, struct btrfs_root,
2353 root_list);
2354 __del_reloc_root(reloc_root);
2355 free_extent_buffer(reloc_root->node);
2356 free_extent_buffer(reloc_root->commit_root);
2357 reloc_root->node = NULL;
2358 reloc_root->commit_root = NULL;
2362 static noinline_for_stack
2363 void merge_reloc_roots(struct reloc_control *rc)
2365 struct btrfs_root *root;
2366 struct btrfs_root *reloc_root;
2367 u64 last_snap;
2368 u64 otransid;
2369 u64 objectid;
2370 LIST_HEAD(reloc_roots);
2371 int found = 0;
2372 int ret = 0;
2373 again:
2374 root = rc->extent_root;
2377 * this serializes us with btrfs_record_root_in_transaction,
2378 * we have to make sure nobody is in the middle of
2379 * adding their roots to the list while we are
2380 * doing this splice
2382 mutex_lock(&root->fs_info->reloc_mutex);
2383 list_splice_init(&rc->reloc_roots, &reloc_roots);
2384 mutex_unlock(&root->fs_info->reloc_mutex);
2386 while (!list_empty(&reloc_roots)) {
2387 found = 1;
2388 reloc_root = list_entry(reloc_roots.next,
2389 struct btrfs_root, root_list);
2391 if (btrfs_root_refs(&reloc_root->root_item) > 0) {
2392 root = read_fs_root(reloc_root->fs_info,
2393 reloc_root->root_key.offset);
2394 BUG_ON(IS_ERR(root));
2395 BUG_ON(root->reloc_root != reloc_root);
2397 ret = merge_reloc_root(rc, root);
2398 if (ret) {
2399 if (list_empty(&reloc_root->root_list))
2400 list_add_tail(&reloc_root->root_list,
2401 &reloc_roots);
2402 goto out;
2404 } else {
2405 list_del_init(&reloc_root->root_list);
2409 * we keep the old last snapshod transid in rtranid when we
2410 * created the relocation tree.
2412 last_snap = btrfs_root_rtransid(&reloc_root->root_item);
2413 otransid = btrfs_root_otransid(&reloc_root->root_item);
2414 objectid = reloc_root->root_key.offset;
2416 ret = btrfs_drop_snapshot(reloc_root, rc->block_rsv, 0, 1);
2417 if (ret < 0) {
2418 if (list_empty(&reloc_root->root_list))
2419 list_add_tail(&reloc_root->root_list,
2420 &reloc_roots);
2421 goto out;
2425 if (found) {
2426 found = 0;
2427 goto again;
2429 out:
2430 if (ret) {
2431 btrfs_std_error(root->fs_info, ret, NULL);
2432 if (!list_empty(&reloc_roots))
2433 free_reloc_roots(&reloc_roots);
2435 /* new reloc root may be added */
2436 mutex_lock(&root->fs_info->reloc_mutex);
2437 list_splice_init(&rc->reloc_roots, &reloc_roots);
2438 mutex_unlock(&root->fs_info->reloc_mutex);
2439 if (!list_empty(&reloc_roots))
2440 free_reloc_roots(&reloc_roots);
2443 BUG_ON(!RB_EMPTY_ROOT(&rc->reloc_root_tree.rb_root));
2446 static void free_block_list(struct rb_root *blocks)
2448 struct tree_block *block;
2449 struct rb_node *rb_node;
2450 while ((rb_node = rb_first(blocks))) {
2451 block = rb_entry(rb_node, struct tree_block, rb_node);
2452 rb_erase(rb_node, blocks);
2453 kfree(block);
2457 static int record_reloc_root_in_trans(struct btrfs_trans_handle *trans,
2458 struct btrfs_root *reloc_root)
2460 struct btrfs_root *root;
2462 if (reloc_root->last_trans == trans->transid)
2463 return 0;
2465 root = read_fs_root(reloc_root->fs_info, reloc_root->root_key.offset);
2466 BUG_ON(IS_ERR(root));
2467 BUG_ON(root->reloc_root != reloc_root);
2469 return btrfs_record_root_in_trans(trans, root);
2472 static noinline_for_stack
2473 struct btrfs_root *select_reloc_root(struct btrfs_trans_handle *trans,
2474 struct reloc_control *rc,
2475 struct backref_node *node,
2476 struct backref_edge *edges[])
2478 struct backref_node *next;
2479 struct btrfs_root *root;
2480 int index = 0;
2482 next = node;
2483 while (1) {
2484 cond_resched();
2485 next = walk_up_backref(next, edges, &index);
2486 root = next->root;
2487 BUG_ON(!root);
2488 BUG_ON(!test_bit(BTRFS_ROOT_REF_COWS, &root->state));
2490 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) {
2491 record_reloc_root_in_trans(trans, root);
2492 break;
2495 btrfs_record_root_in_trans(trans, root);
2496 root = root->reloc_root;
2498 if (next->new_bytenr != root->node->start) {
2499 BUG_ON(next->new_bytenr);
2500 BUG_ON(!list_empty(&next->list));
2501 next->new_bytenr = root->node->start;
2502 next->root = root;
2503 list_add_tail(&next->list,
2504 &rc->backref_cache.changed);
2505 __mark_block_processed(rc, next);
2506 break;
2509 WARN_ON(1);
2510 root = NULL;
2511 next = walk_down_backref(edges, &index);
2512 if (!next || next->level <= node->level)
2513 break;
2515 if (!root)
2516 return NULL;
2518 next = node;
2519 /* setup backref node path for btrfs_reloc_cow_block */
2520 while (1) {
2521 rc->backref_cache.path[next->level] = next;
2522 if (--index < 0)
2523 break;
2524 next = edges[index]->node[UPPER];
2526 return root;
2530 * select a tree root for relocation. return NULL if the block
2531 * is reference counted. we should use do_relocation() in this
2532 * case. return a tree root pointer if the block isn't reference
2533 * counted. return -ENOENT if the block is root of reloc tree.
2535 static noinline_for_stack
2536 struct btrfs_root *select_one_root(struct backref_node *node)
2538 struct backref_node *next;
2539 struct btrfs_root *root;
2540 struct btrfs_root *fs_root = NULL;
2541 struct backref_edge *edges[BTRFS_MAX_LEVEL - 1];
2542 int index = 0;
2544 next = node;
2545 while (1) {
2546 cond_resched();
2547 next = walk_up_backref(next, edges, &index);
2548 root = next->root;
2549 BUG_ON(!root);
2551 /* no other choice for non-references counted tree */
2552 if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state))
2553 return root;
2555 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID)
2556 fs_root = root;
2558 if (next != node)
2559 return NULL;
2561 next = walk_down_backref(edges, &index);
2562 if (!next || next->level <= node->level)
2563 break;
2566 if (!fs_root)
2567 return ERR_PTR(-ENOENT);
2568 return fs_root;
2571 static noinline_for_stack
2572 u64 calcu_metadata_size(struct reloc_control *rc,
2573 struct backref_node *node, int reserve)
2575 struct backref_node *next = node;
2576 struct backref_edge *edge;
2577 struct backref_edge *edges[BTRFS_MAX_LEVEL - 1];
2578 u64 num_bytes = 0;
2579 int index = 0;
2581 BUG_ON(reserve && node->processed);
2583 while (next) {
2584 cond_resched();
2585 while (1) {
2586 if (next->processed && (reserve || next != node))
2587 break;
2589 num_bytes += rc->extent_root->nodesize;
2591 if (list_empty(&next->upper))
2592 break;
2594 edge = list_entry(next->upper.next,
2595 struct backref_edge, list[LOWER]);
2596 edges[index++] = edge;
2597 next = edge->node[UPPER];
2599 next = walk_down_backref(edges, &index);
2601 return num_bytes;
2604 static int reserve_metadata_space(struct btrfs_trans_handle *trans,
2605 struct reloc_control *rc,
2606 struct backref_node *node)
2608 struct btrfs_root *root = rc->extent_root;
2609 u64 num_bytes;
2610 int ret;
2611 u64 tmp;
2613 num_bytes = calcu_metadata_size(rc, node, 1) * 2;
2615 trans->block_rsv = rc->block_rsv;
2616 rc->reserved_bytes += num_bytes;
2617 ret = btrfs_block_rsv_refill(root, rc->block_rsv, num_bytes,
2618 BTRFS_RESERVE_FLUSH_ALL);
2619 if (ret) {
2620 if (ret == -EAGAIN) {
2621 tmp = rc->extent_root->nodesize *
2622 RELOCATION_RESERVED_NODES;
2623 while (tmp <= rc->reserved_bytes)
2624 tmp <<= 1;
2626 * only one thread can access block_rsv at this point,
2627 * so we don't need hold lock to protect block_rsv.
2628 * we expand more reservation size here to allow enough
2629 * space for relocation and we will return eailer in
2630 * enospc case.
2632 rc->block_rsv->size = tmp + rc->extent_root->nodesize *
2633 RELOCATION_RESERVED_NODES;
2635 return ret;
2638 return 0;
2642 * relocate a block tree, and then update pointers in upper level
2643 * blocks that reference the block to point to the new location.
2645 * if called by link_to_upper, the block has already been relocated.
2646 * in that case this function just updates pointers.
2648 static int do_relocation(struct btrfs_trans_handle *trans,
2649 struct reloc_control *rc,
2650 struct backref_node *node,
2651 struct btrfs_key *key,
2652 struct btrfs_path *path, int lowest)
2654 struct backref_node *upper;
2655 struct backref_edge *edge;
2656 struct backref_edge *edges[BTRFS_MAX_LEVEL - 1];
2657 struct btrfs_root *root;
2658 struct extent_buffer *eb;
2659 u32 blocksize;
2660 u64 bytenr;
2661 u64 generation;
2662 int slot;
2663 int ret;
2664 int err = 0;
2666 BUG_ON(lowest && node->eb);
2668 path->lowest_level = node->level + 1;
2669 rc->backref_cache.path[node->level] = node;
2670 list_for_each_entry(edge, &node->upper, list[LOWER]) {
2671 cond_resched();
2673 upper = edge->node[UPPER];
2674 root = select_reloc_root(trans, rc, upper, edges);
2675 BUG_ON(!root);
2677 if (upper->eb && !upper->locked) {
2678 if (!lowest) {
2679 ret = btrfs_bin_search(upper->eb, key,
2680 upper->level, &slot);
2681 BUG_ON(ret);
2682 bytenr = btrfs_node_blockptr(upper->eb, slot);
2683 if (node->eb->start == bytenr)
2684 goto next;
2686 drop_node_buffer(upper);
2689 if (!upper->eb) {
2690 ret = btrfs_search_slot(trans, root, key, path, 0, 1);
2691 if (ret) {
2692 if (ret < 0)
2693 err = ret;
2694 else
2695 err = -ENOENT;
2697 btrfs_release_path(path);
2698 break;
2701 if (!upper->eb) {
2702 upper->eb = path->nodes[upper->level];
2703 path->nodes[upper->level] = NULL;
2704 } else {
2705 BUG_ON(upper->eb != path->nodes[upper->level]);
2708 upper->locked = 1;
2709 path->locks[upper->level] = 0;
2711 slot = path->slots[upper->level];
2712 btrfs_release_path(path);
2713 } else {
2714 ret = btrfs_bin_search(upper->eb, key, upper->level,
2715 &slot);
2716 BUG_ON(ret);
2719 bytenr = btrfs_node_blockptr(upper->eb, slot);
2720 if (lowest) {
2721 BUG_ON(bytenr != node->bytenr);
2722 } else {
2723 if (node->eb->start == bytenr)
2724 goto next;
2727 blocksize = root->nodesize;
2728 generation = btrfs_node_ptr_generation(upper->eb, slot);
2729 eb = read_tree_block(root, bytenr, generation);
2730 if (IS_ERR(eb)) {
2731 err = PTR_ERR(eb);
2732 goto next;
2733 } else if (!extent_buffer_uptodate(eb)) {
2734 free_extent_buffer(eb);
2735 err = -EIO;
2736 goto next;
2738 btrfs_tree_lock(eb);
2739 btrfs_set_lock_blocking(eb);
2741 if (!node->eb) {
2742 ret = btrfs_cow_block(trans, root, eb, upper->eb,
2743 slot, &eb);
2744 btrfs_tree_unlock(eb);
2745 free_extent_buffer(eb);
2746 if (ret < 0) {
2747 err = ret;
2748 goto next;
2750 BUG_ON(node->eb != eb);
2751 } else {
2752 btrfs_set_node_blockptr(upper->eb, slot,
2753 node->eb->start);
2754 btrfs_set_node_ptr_generation(upper->eb, slot,
2755 trans->transid);
2756 btrfs_mark_buffer_dirty(upper->eb);
2758 ret = btrfs_inc_extent_ref(trans, root,
2759 node->eb->start, blocksize,
2760 upper->eb->start,
2761 btrfs_header_owner(upper->eb),
2762 node->level, 0);
2763 BUG_ON(ret);
2765 ret = btrfs_drop_subtree(trans, root, eb, upper->eb);
2766 BUG_ON(ret);
2768 next:
2769 if (!upper->pending)
2770 drop_node_buffer(upper);
2771 else
2772 unlock_node_buffer(upper);
2773 if (err)
2774 break;
2777 if (!err && node->pending) {
2778 drop_node_buffer(node);
2779 list_move_tail(&node->list, &rc->backref_cache.changed);
2780 node->pending = 0;
2783 path->lowest_level = 0;
2784 BUG_ON(err == -ENOSPC);
2785 return err;
2788 static int link_to_upper(struct btrfs_trans_handle *trans,
2789 struct reloc_control *rc,
2790 struct backref_node *node,
2791 struct btrfs_path *path)
2793 struct btrfs_key key;
2795 btrfs_node_key_to_cpu(node->eb, &key, 0);
2796 return do_relocation(trans, rc, node, &key, path, 0);
2799 static int finish_pending_nodes(struct btrfs_trans_handle *trans,
2800 struct reloc_control *rc,
2801 struct btrfs_path *path, int err)
2803 LIST_HEAD(list);
2804 struct backref_cache *cache = &rc->backref_cache;
2805 struct backref_node *node;
2806 int level;
2807 int ret;
2809 for (level = 0; level < BTRFS_MAX_LEVEL; level++) {
2810 while (!list_empty(&cache->pending[level])) {
2811 node = list_entry(cache->pending[level].next,
2812 struct backref_node, list);
2813 list_move_tail(&node->list, &list);
2814 BUG_ON(!node->pending);
2816 if (!err) {
2817 ret = link_to_upper(trans, rc, node, path);
2818 if (ret < 0)
2819 err = ret;
2822 list_splice_init(&list, &cache->pending[level]);
2824 return err;
2827 static void mark_block_processed(struct reloc_control *rc,
2828 u64 bytenr, u32 blocksize)
2830 set_extent_bits(&rc->processed_blocks, bytenr, bytenr + blocksize - 1,
2831 EXTENT_DIRTY, GFP_NOFS);
2834 static void __mark_block_processed(struct reloc_control *rc,
2835 struct backref_node *node)
2837 u32 blocksize;
2838 if (node->level == 0 ||
2839 in_block_group(node->bytenr, rc->block_group)) {
2840 blocksize = rc->extent_root->nodesize;
2841 mark_block_processed(rc, node->bytenr, blocksize);
2843 node->processed = 1;
2847 * mark a block and all blocks directly/indirectly reference the block
2848 * as processed.
2850 static void update_processed_blocks(struct reloc_control *rc,
2851 struct backref_node *node)
2853 struct backref_node *next = node;
2854 struct backref_edge *edge;
2855 struct backref_edge *edges[BTRFS_MAX_LEVEL - 1];
2856 int index = 0;
2858 while (next) {
2859 cond_resched();
2860 while (1) {
2861 if (next->processed)
2862 break;
2864 __mark_block_processed(rc, next);
2866 if (list_empty(&next->upper))
2867 break;
2869 edge = list_entry(next->upper.next,
2870 struct backref_edge, list[LOWER]);
2871 edges[index++] = edge;
2872 next = edge->node[UPPER];
2874 next = walk_down_backref(edges, &index);
2878 static int tree_block_processed(u64 bytenr, struct reloc_control *rc)
2880 u32 blocksize = rc->extent_root->nodesize;
2882 if (test_range_bit(&rc->processed_blocks, bytenr,
2883 bytenr + blocksize - 1, EXTENT_DIRTY, 1, NULL))
2884 return 1;
2885 return 0;
2888 static int get_tree_block_key(struct reloc_control *rc,
2889 struct tree_block *block)
2891 struct extent_buffer *eb;
2893 BUG_ON(block->key_ready);
2894 eb = read_tree_block(rc->extent_root, block->bytenr,
2895 block->key.offset);
2896 if (IS_ERR(eb)) {
2897 return PTR_ERR(eb);
2898 } else if (!extent_buffer_uptodate(eb)) {
2899 free_extent_buffer(eb);
2900 return -EIO;
2902 WARN_ON(btrfs_header_level(eb) != block->level);
2903 if (block->level == 0)
2904 btrfs_item_key_to_cpu(eb, &block->key, 0);
2905 else
2906 btrfs_node_key_to_cpu(eb, &block->key, 0);
2907 free_extent_buffer(eb);
2908 block->key_ready = 1;
2909 return 0;
2913 * helper function to relocate a tree block
2915 static int relocate_tree_block(struct btrfs_trans_handle *trans,
2916 struct reloc_control *rc,
2917 struct backref_node *node,
2918 struct btrfs_key *key,
2919 struct btrfs_path *path)
2921 struct btrfs_root *root;
2922 int ret = 0;
2924 if (!node)
2925 return 0;
2927 BUG_ON(node->processed);
2928 root = select_one_root(node);
2929 if (root == ERR_PTR(-ENOENT)) {
2930 update_processed_blocks(rc, node);
2931 goto out;
2934 if (!root || test_bit(BTRFS_ROOT_REF_COWS, &root->state)) {
2935 ret = reserve_metadata_space(trans, rc, node);
2936 if (ret)
2937 goto out;
2940 if (root) {
2941 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state)) {
2942 BUG_ON(node->new_bytenr);
2943 BUG_ON(!list_empty(&node->list));
2944 btrfs_record_root_in_trans(trans, root);
2945 root = root->reloc_root;
2946 node->new_bytenr = root->node->start;
2947 node->root = root;
2948 list_add_tail(&node->list, &rc->backref_cache.changed);
2949 } else {
2950 path->lowest_level = node->level;
2951 ret = btrfs_search_slot(trans, root, key, path, 0, 1);
2952 btrfs_release_path(path);
2953 if (ret > 0)
2954 ret = 0;
2956 if (!ret)
2957 update_processed_blocks(rc, node);
2958 } else {
2959 ret = do_relocation(trans, rc, node, key, path, 1);
2961 out:
2962 if (ret || node->level == 0 || node->cowonly)
2963 remove_backref_node(&rc->backref_cache, node);
2964 return ret;
2968 * relocate a list of blocks
2970 static noinline_for_stack
2971 int relocate_tree_blocks(struct btrfs_trans_handle *trans,
2972 struct reloc_control *rc, struct rb_root *blocks)
2974 struct backref_node *node;
2975 struct btrfs_path *path;
2976 struct tree_block *block;
2977 struct rb_node *rb_node;
2978 int ret;
2979 int err = 0;
2981 path = btrfs_alloc_path();
2982 if (!path) {
2983 err = -ENOMEM;
2984 goto out_free_blocks;
2987 rb_node = rb_first(blocks);
2988 while (rb_node) {
2989 block = rb_entry(rb_node, struct tree_block, rb_node);
2990 if (!block->key_ready)
2991 readahead_tree_block(rc->extent_root, block->bytenr);
2992 rb_node = rb_next(rb_node);
2995 rb_node = rb_first(blocks);
2996 while (rb_node) {
2997 block = rb_entry(rb_node, struct tree_block, rb_node);
2998 if (!block->key_ready) {
2999 err = get_tree_block_key(rc, block);
3000 if (err)
3001 goto out_free_path;
3003 rb_node = rb_next(rb_node);
3006 rb_node = rb_first(blocks);
3007 while (rb_node) {
3008 block = rb_entry(rb_node, struct tree_block, rb_node);
3010 node = build_backref_tree(rc, &block->key,
3011 block->level, block->bytenr);
3012 if (IS_ERR(node)) {
3013 err = PTR_ERR(node);
3014 goto out;
3017 ret = relocate_tree_block(trans, rc, node, &block->key,
3018 path);
3019 if (ret < 0) {
3020 if (ret != -EAGAIN || rb_node == rb_first(blocks))
3021 err = ret;
3022 goto out;
3024 rb_node = rb_next(rb_node);
3026 out:
3027 err = finish_pending_nodes(trans, rc, path, err);
3029 out_free_path:
3030 btrfs_free_path(path);
3031 out_free_blocks:
3032 free_block_list(blocks);
3033 return err;
3036 static noinline_for_stack
3037 int prealloc_file_extent_cluster(struct inode *inode,
3038 struct file_extent_cluster *cluster)
3040 u64 alloc_hint = 0;
3041 u64 start;
3042 u64 end;
3043 u64 offset = BTRFS_I(inode)->index_cnt;
3044 u64 num_bytes;
3045 int nr = 0;
3046 int ret = 0;
3048 BUG_ON(cluster->start != cluster->boundary[0]);
3049 mutex_lock(&inode->i_mutex);
3051 ret = btrfs_check_data_free_space(inode, cluster->start,
3052 cluster->end + 1 - cluster->start);
3053 if (ret)
3054 goto out;
3056 while (nr < cluster->nr) {
3057 start = cluster->boundary[nr] - offset;
3058 if (nr + 1 < cluster->nr)
3059 end = cluster->boundary[nr + 1] - 1 - offset;
3060 else
3061 end = cluster->end - offset;
3063 lock_extent(&BTRFS_I(inode)->io_tree, start, end);
3064 num_bytes = end + 1 - start;
3065 ret = btrfs_prealloc_file_range(inode, 0, start,
3066 num_bytes, num_bytes,
3067 end + 1, &alloc_hint);
3068 unlock_extent(&BTRFS_I(inode)->io_tree, start, end);
3069 if (ret)
3070 break;
3071 nr++;
3073 btrfs_free_reserved_data_space(inode, cluster->start,
3074 cluster->end + 1 - cluster->start);
3075 out:
3076 mutex_unlock(&inode->i_mutex);
3077 return ret;
3080 static noinline_for_stack
3081 int setup_extent_mapping(struct inode *inode, u64 start, u64 end,
3082 u64 block_start)
3084 struct btrfs_root *root = BTRFS_I(inode)->root;
3085 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
3086 struct extent_map *em;
3087 int ret = 0;
3089 em = alloc_extent_map();
3090 if (!em)
3091 return -ENOMEM;
3093 em->start = start;
3094 em->len = end + 1 - start;
3095 em->block_len = em->len;
3096 em->block_start = block_start;
3097 em->bdev = root->fs_info->fs_devices->latest_bdev;
3098 set_bit(EXTENT_FLAG_PINNED, &em->flags);
3100 lock_extent(&BTRFS_I(inode)->io_tree, start, end);
3101 while (1) {
3102 write_lock(&em_tree->lock);
3103 ret = add_extent_mapping(em_tree, em, 0);
3104 write_unlock(&em_tree->lock);
3105 if (ret != -EEXIST) {
3106 free_extent_map(em);
3107 break;
3109 btrfs_drop_extent_cache(inode, start, end, 0);
3111 unlock_extent(&BTRFS_I(inode)->io_tree, start, end);
3112 return ret;
3115 static int relocate_file_extent_cluster(struct inode *inode,
3116 struct file_extent_cluster *cluster)
3118 u64 page_start;
3119 u64 page_end;
3120 u64 offset = BTRFS_I(inode)->index_cnt;
3121 unsigned long index;
3122 unsigned long last_index;
3123 struct page *page;
3124 struct file_ra_state *ra;
3125 gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping);
3126 int nr = 0;
3127 int ret = 0;
3129 if (!cluster->nr)
3130 return 0;
3132 ra = kzalloc(sizeof(*ra), GFP_NOFS);
3133 if (!ra)
3134 return -ENOMEM;
3136 ret = prealloc_file_extent_cluster(inode, cluster);
3137 if (ret)
3138 goto out;
3140 file_ra_state_init(ra, inode->i_mapping);
3142 ret = setup_extent_mapping(inode, cluster->start - offset,
3143 cluster->end - offset, cluster->start);
3144 if (ret)
3145 goto out;
3147 index = (cluster->start - offset) >> PAGE_CACHE_SHIFT;
3148 last_index = (cluster->end - offset) >> PAGE_CACHE_SHIFT;
3149 while (index <= last_index) {
3150 ret = btrfs_delalloc_reserve_metadata(inode, PAGE_CACHE_SIZE);
3151 if (ret)
3152 goto out;
3154 page = find_lock_page(inode->i_mapping, index);
3155 if (!page) {
3156 page_cache_sync_readahead(inode->i_mapping,
3157 ra, NULL, index,
3158 last_index + 1 - index);
3159 page = find_or_create_page(inode->i_mapping, index,
3160 mask);
3161 if (!page) {
3162 btrfs_delalloc_release_metadata(inode,
3163 PAGE_CACHE_SIZE);
3164 ret = -ENOMEM;
3165 goto out;
3169 if (PageReadahead(page)) {
3170 page_cache_async_readahead(inode->i_mapping,
3171 ra, NULL, page, index,
3172 last_index + 1 - index);
3175 if (!PageUptodate(page)) {
3176 btrfs_readpage(NULL, page);
3177 lock_page(page);
3178 if (!PageUptodate(page)) {
3179 unlock_page(page);
3180 page_cache_release(page);
3181 btrfs_delalloc_release_metadata(inode,
3182 PAGE_CACHE_SIZE);
3183 ret = -EIO;
3184 goto out;
3188 page_start = page_offset(page);
3189 page_end = page_start + PAGE_CACHE_SIZE - 1;
3191 lock_extent(&BTRFS_I(inode)->io_tree, page_start, page_end);
3193 set_page_extent_mapped(page);
3195 if (nr < cluster->nr &&
3196 page_start + offset == cluster->boundary[nr]) {
3197 set_extent_bits(&BTRFS_I(inode)->io_tree,
3198 page_start, page_end,
3199 EXTENT_BOUNDARY, GFP_NOFS);
3200 nr++;
3203 btrfs_set_extent_delalloc(inode, page_start, page_end, NULL);
3204 set_page_dirty(page);
3206 unlock_extent(&BTRFS_I(inode)->io_tree,
3207 page_start, page_end);
3208 unlock_page(page);
3209 page_cache_release(page);
3211 index++;
3212 balance_dirty_pages_ratelimited(inode->i_mapping);
3213 btrfs_throttle(BTRFS_I(inode)->root);
3215 WARN_ON(nr != cluster->nr);
3216 out:
3217 kfree(ra);
3218 return ret;
3221 static noinline_for_stack
3222 int relocate_data_extent(struct inode *inode, struct btrfs_key *extent_key,
3223 struct file_extent_cluster *cluster)
3225 int ret;
3227 if (cluster->nr > 0 && extent_key->objectid != cluster->end + 1) {
3228 ret = relocate_file_extent_cluster(inode, cluster);
3229 if (ret)
3230 return ret;
3231 cluster->nr = 0;
3234 if (!cluster->nr)
3235 cluster->start = extent_key->objectid;
3236 else
3237 BUG_ON(cluster->nr >= MAX_EXTENTS);
3238 cluster->end = extent_key->objectid + extent_key->offset - 1;
3239 cluster->boundary[cluster->nr] = extent_key->objectid;
3240 cluster->nr++;
3242 if (cluster->nr >= MAX_EXTENTS) {
3243 ret = relocate_file_extent_cluster(inode, cluster);
3244 if (ret)
3245 return ret;
3246 cluster->nr = 0;
3248 return 0;
3251 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
3252 static int get_ref_objectid_v0(struct reloc_control *rc,
3253 struct btrfs_path *path,
3254 struct btrfs_key *extent_key,
3255 u64 *ref_objectid, int *path_change)
3257 struct btrfs_key key;
3258 struct extent_buffer *leaf;
3259 struct btrfs_extent_ref_v0 *ref0;
3260 int ret;
3261 int slot;
3263 leaf = path->nodes[0];
3264 slot = path->slots[0];
3265 while (1) {
3266 if (slot >= btrfs_header_nritems(leaf)) {
3267 ret = btrfs_next_leaf(rc->extent_root, path);
3268 if (ret < 0)
3269 return ret;
3270 BUG_ON(ret > 0);
3271 leaf = path->nodes[0];
3272 slot = path->slots[0];
3273 if (path_change)
3274 *path_change = 1;
3276 btrfs_item_key_to_cpu(leaf, &key, slot);
3277 if (key.objectid != extent_key->objectid)
3278 return -ENOENT;
3280 if (key.type != BTRFS_EXTENT_REF_V0_KEY) {
3281 slot++;
3282 continue;
3284 ref0 = btrfs_item_ptr(leaf, slot,
3285 struct btrfs_extent_ref_v0);
3286 *ref_objectid = btrfs_ref_objectid_v0(leaf, ref0);
3287 break;
3289 return 0;
3291 #endif
3294 * helper to add a tree block to the list.
3295 * the major work is getting the generation and level of the block
3297 static int add_tree_block(struct reloc_control *rc,
3298 struct btrfs_key *extent_key,
3299 struct btrfs_path *path,
3300 struct rb_root *blocks)
3302 struct extent_buffer *eb;
3303 struct btrfs_extent_item *ei;
3304 struct btrfs_tree_block_info *bi;
3305 struct tree_block *block;
3306 struct rb_node *rb_node;
3307 u32 item_size;
3308 int level = -1;
3309 u64 generation;
3311 eb = path->nodes[0];
3312 item_size = btrfs_item_size_nr(eb, path->slots[0]);
3314 if (extent_key->type == BTRFS_METADATA_ITEM_KEY ||
3315 item_size >= sizeof(*ei) + sizeof(*bi)) {
3316 ei = btrfs_item_ptr(eb, path->slots[0],
3317 struct btrfs_extent_item);
3318 if (extent_key->type == BTRFS_EXTENT_ITEM_KEY) {
3319 bi = (struct btrfs_tree_block_info *)(ei + 1);
3320 level = btrfs_tree_block_level(eb, bi);
3321 } else {
3322 level = (int)extent_key->offset;
3324 generation = btrfs_extent_generation(eb, ei);
3325 } else {
3326 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
3327 u64 ref_owner;
3328 int ret;
3330 BUG_ON(item_size != sizeof(struct btrfs_extent_item_v0));
3331 ret = get_ref_objectid_v0(rc, path, extent_key,
3332 &ref_owner, NULL);
3333 if (ret < 0)
3334 return ret;
3335 BUG_ON(ref_owner >= BTRFS_MAX_LEVEL);
3336 level = (int)ref_owner;
3337 /* FIXME: get real generation */
3338 generation = 0;
3339 #else
3340 BUG();
3341 #endif
3344 btrfs_release_path(path);
3346 BUG_ON(level == -1);
3348 block = kmalloc(sizeof(*block), GFP_NOFS);
3349 if (!block)
3350 return -ENOMEM;
3352 block->bytenr = extent_key->objectid;
3353 block->key.objectid = rc->extent_root->nodesize;
3354 block->key.offset = generation;
3355 block->level = level;
3356 block->key_ready = 0;
3358 rb_node = tree_insert(blocks, block->bytenr, &block->rb_node);
3359 if (rb_node)
3360 backref_tree_panic(rb_node, -EEXIST, block->bytenr);
3362 return 0;
3366 * helper to add tree blocks for backref of type BTRFS_SHARED_DATA_REF_KEY
3368 static int __add_tree_block(struct reloc_control *rc,
3369 u64 bytenr, u32 blocksize,
3370 struct rb_root *blocks)
3372 struct btrfs_path *path;
3373 struct btrfs_key key;
3374 int ret;
3375 bool skinny = btrfs_fs_incompat(rc->extent_root->fs_info,
3376 SKINNY_METADATA);
3378 if (tree_block_processed(bytenr, rc))
3379 return 0;
3381 if (tree_search(blocks, bytenr))
3382 return 0;
3384 path = btrfs_alloc_path();
3385 if (!path)
3386 return -ENOMEM;
3387 again:
3388 key.objectid = bytenr;
3389 if (skinny) {
3390 key.type = BTRFS_METADATA_ITEM_KEY;
3391 key.offset = (u64)-1;
3392 } else {
3393 key.type = BTRFS_EXTENT_ITEM_KEY;
3394 key.offset = blocksize;
3397 path->search_commit_root = 1;
3398 path->skip_locking = 1;
3399 ret = btrfs_search_slot(NULL, rc->extent_root, &key, path, 0, 0);
3400 if (ret < 0)
3401 goto out;
3403 if (ret > 0 && skinny) {
3404 if (path->slots[0]) {
3405 path->slots[0]--;
3406 btrfs_item_key_to_cpu(path->nodes[0], &key,
3407 path->slots[0]);
3408 if (key.objectid == bytenr &&
3409 (key.type == BTRFS_METADATA_ITEM_KEY ||
3410 (key.type == BTRFS_EXTENT_ITEM_KEY &&
3411 key.offset == blocksize)))
3412 ret = 0;
3415 if (ret) {
3416 skinny = false;
3417 btrfs_release_path(path);
3418 goto again;
3421 BUG_ON(ret);
3423 ret = add_tree_block(rc, &key, path, blocks);
3424 out:
3425 btrfs_free_path(path);
3426 return ret;
3430 * helper to check if the block use full backrefs for pointers in it
3432 static int block_use_full_backref(struct reloc_control *rc,
3433 struct extent_buffer *eb)
3435 u64 flags;
3436 int ret;
3438 if (btrfs_header_flag(eb, BTRFS_HEADER_FLAG_RELOC) ||
3439 btrfs_header_backref_rev(eb) < BTRFS_MIXED_BACKREF_REV)
3440 return 1;
3442 ret = btrfs_lookup_extent_info(NULL, rc->extent_root,
3443 eb->start, btrfs_header_level(eb), 1,
3444 NULL, &flags);
3445 BUG_ON(ret);
3447 if (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)
3448 ret = 1;
3449 else
3450 ret = 0;
3451 return ret;
3454 static int delete_block_group_cache(struct btrfs_fs_info *fs_info,
3455 struct btrfs_block_group_cache *block_group,
3456 struct inode *inode,
3457 u64 ino)
3459 struct btrfs_key key;
3460 struct btrfs_root *root = fs_info->tree_root;
3461 struct btrfs_trans_handle *trans;
3462 int ret = 0;
3464 if (inode)
3465 goto truncate;
3467 key.objectid = ino;
3468 key.type = BTRFS_INODE_ITEM_KEY;
3469 key.offset = 0;
3471 inode = btrfs_iget(fs_info->sb, &key, root, NULL);
3472 if (IS_ERR(inode) || is_bad_inode(inode)) {
3473 if (!IS_ERR(inode))
3474 iput(inode);
3475 return -ENOENT;
3478 truncate:
3479 ret = btrfs_check_trunc_cache_free_space(root,
3480 &fs_info->global_block_rsv);
3481 if (ret)
3482 goto out;
3484 trans = btrfs_join_transaction(root);
3485 if (IS_ERR(trans)) {
3486 ret = PTR_ERR(trans);
3487 goto out;
3490 ret = btrfs_truncate_free_space_cache(root, trans, block_group, inode);
3492 btrfs_end_transaction(trans, root);
3493 btrfs_btree_balance_dirty(root);
3494 out:
3495 iput(inode);
3496 return ret;
3500 * helper to add tree blocks for backref of type BTRFS_EXTENT_DATA_REF_KEY
3501 * this function scans fs tree to find blocks reference the data extent
3503 static int find_data_references(struct reloc_control *rc,
3504 struct btrfs_key *extent_key,
3505 struct extent_buffer *leaf,
3506 struct btrfs_extent_data_ref *ref,
3507 struct rb_root *blocks)
3509 struct btrfs_path *path;
3510 struct tree_block *block;
3511 struct btrfs_root *root;
3512 struct btrfs_file_extent_item *fi;
3513 struct rb_node *rb_node;
3514 struct btrfs_key key;
3515 u64 ref_root;
3516 u64 ref_objectid;
3517 u64 ref_offset;
3518 u32 ref_count;
3519 u32 nritems;
3520 int err = 0;
3521 int added = 0;
3522 int counted;
3523 int ret;
3525 ref_root = btrfs_extent_data_ref_root(leaf, ref);
3526 ref_objectid = btrfs_extent_data_ref_objectid(leaf, ref);
3527 ref_offset = btrfs_extent_data_ref_offset(leaf, ref);
3528 ref_count = btrfs_extent_data_ref_count(leaf, ref);
3531 * This is an extent belonging to the free space cache, lets just delete
3532 * it and redo the search.
3534 if (ref_root == BTRFS_ROOT_TREE_OBJECTID) {
3535 ret = delete_block_group_cache(rc->extent_root->fs_info,
3536 rc->block_group,
3537 NULL, ref_objectid);
3538 if (ret != -ENOENT)
3539 return ret;
3540 ret = 0;
3543 path = btrfs_alloc_path();
3544 if (!path)
3545 return -ENOMEM;
3546 path->reada = 1;
3548 root = read_fs_root(rc->extent_root->fs_info, ref_root);
3549 if (IS_ERR(root)) {
3550 err = PTR_ERR(root);
3551 goto out;
3554 key.objectid = ref_objectid;
3555 key.type = BTRFS_EXTENT_DATA_KEY;
3556 if (ref_offset > ((u64)-1 << 32))
3557 key.offset = 0;
3558 else
3559 key.offset = ref_offset;
3561 path->search_commit_root = 1;
3562 path->skip_locking = 1;
3563 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3564 if (ret < 0) {
3565 err = ret;
3566 goto out;
3569 leaf = path->nodes[0];
3570 nritems = btrfs_header_nritems(leaf);
3572 * the references in tree blocks that use full backrefs
3573 * are not counted in
3575 if (block_use_full_backref(rc, leaf))
3576 counted = 0;
3577 else
3578 counted = 1;
3579 rb_node = tree_search(blocks, leaf->start);
3580 if (rb_node) {
3581 if (counted)
3582 added = 1;
3583 else
3584 path->slots[0] = nritems;
3587 while (ref_count > 0) {
3588 while (path->slots[0] >= nritems) {
3589 ret = btrfs_next_leaf(root, path);
3590 if (ret < 0) {
3591 err = ret;
3592 goto out;
3594 if (WARN_ON(ret > 0))
3595 goto out;
3597 leaf = path->nodes[0];
3598 nritems = btrfs_header_nritems(leaf);
3599 added = 0;
3601 if (block_use_full_backref(rc, leaf))
3602 counted = 0;
3603 else
3604 counted = 1;
3605 rb_node = tree_search(blocks, leaf->start);
3606 if (rb_node) {
3607 if (counted)
3608 added = 1;
3609 else
3610 path->slots[0] = nritems;
3614 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3615 if (WARN_ON(key.objectid != ref_objectid ||
3616 key.type != BTRFS_EXTENT_DATA_KEY))
3617 break;
3619 fi = btrfs_item_ptr(leaf, path->slots[0],
3620 struct btrfs_file_extent_item);
3622 if (btrfs_file_extent_type(leaf, fi) ==
3623 BTRFS_FILE_EXTENT_INLINE)
3624 goto next;
3626 if (btrfs_file_extent_disk_bytenr(leaf, fi) !=
3627 extent_key->objectid)
3628 goto next;
3630 key.offset -= btrfs_file_extent_offset(leaf, fi);
3631 if (key.offset != ref_offset)
3632 goto next;
3634 if (counted)
3635 ref_count--;
3636 if (added)
3637 goto next;
3639 if (!tree_block_processed(leaf->start, rc)) {
3640 block = kmalloc(sizeof(*block), GFP_NOFS);
3641 if (!block) {
3642 err = -ENOMEM;
3643 break;
3645 block->bytenr = leaf->start;
3646 btrfs_item_key_to_cpu(leaf, &block->key, 0);
3647 block->level = 0;
3648 block->key_ready = 1;
3649 rb_node = tree_insert(blocks, block->bytenr,
3650 &block->rb_node);
3651 if (rb_node)
3652 backref_tree_panic(rb_node, -EEXIST,
3653 block->bytenr);
3655 if (counted)
3656 added = 1;
3657 else
3658 path->slots[0] = nritems;
3659 next:
3660 path->slots[0]++;
3663 out:
3664 btrfs_free_path(path);
3665 return err;
3669 * helper to find all tree blocks that reference a given data extent
3671 static noinline_for_stack
3672 int add_data_references(struct reloc_control *rc,
3673 struct btrfs_key *extent_key,
3674 struct btrfs_path *path,
3675 struct rb_root *blocks)
3677 struct btrfs_key key;
3678 struct extent_buffer *eb;
3679 struct btrfs_extent_data_ref *dref;
3680 struct btrfs_extent_inline_ref *iref;
3681 unsigned long ptr;
3682 unsigned long end;
3683 u32 blocksize = rc->extent_root->nodesize;
3684 int ret = 0;
3685 int err = 0;
3687 eb = path->nodes[0];
3688 ptr = btrfs_item_ptr_offset(eb, path->slots[0]);
3689 end = ptr + btrfs_item_size_nr(eb, path->slots[0]);
3690 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
3691 if (ptr + sizeof(struct btrfs_extent_item_v0) == end)
3692 ptr = end;
3693 else
3694 #endif
3695 ptr += sizeof(struct btrfs_extent_item);
3697 while (ptr < end) {
3698 iref = (struct btrfs_extent_inline_ref *)ptr;
3699 key.type = btrfs_extent_inline_ref_type(eb, iref);
3700 if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
3701 key.offset = btrfs_extent_inline_ref_offset(eb, iref);
3702 ret = __add_tree_block(rc, key.offset, blocksize,
3703 blocks);
3704 } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
3705 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
3706 ret = find_data_references(rc, extent_key,
3707 eb, dref, blocks);
3708 } else {
3709 BUG();
3711 if (ret) {
3712 err = ret;
3713 goto out;
3715 ptr += btrfs_extent_inline_ref_size(key.type);
3717 WARN_ON(ptr > end);
3719 while (1) {
3720 cond_resched();
3721 eb = path->nodes[0];
3722 if (path->slots[0] >= btrfs_header_nritems(eb)) {
3723 ret = btrfs_next_leaf(rc->extent_root, path);
3724 if (ret < 0) {
3725 err = ret;
3726 break;
3728 if (ret > 0)
3729 break;
3730 eb = path->nodes[0];
3733 btrfs_item_key_to_cpu(eb, &key, path->slots[0]);
3734 if (key.objectid != extent_key->objectid)
3735 break;
3737 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
3738 if (key.type == BTRFS_SHARED_DATA_REF_KEY ||
3739 key.type == BTRFS_EXTENT_REF_V0_KEY) {
3740 #else
3741 BUG_ON(key.type == BTRFS_EXTENT_REF_V0_KEY);
3742 if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
3743 #endif
3744 ret = __add_tree_block(rc, key.offset, blocksize,
3745 blocks);
3746 } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
3747 dref = btrfs_item_ptr(eb, path->slots[0],
3748 struct btrfs_extent_data_ref);
3749 ret = find_data_references(rc, extent_key,
3750 eb, dref, blocks);
3751 } else {
3752 ret = 0;
3754 if (ret) {
3755 err = ret;
3756 break;
3758 path->slots[0]++;
3760 out:
3761 btrfs_release_path(path);
3762 if (err)
3763 free_block_list(blocks);
3764 return err;
3768 * helper to find next unprocessed extent
3770 static noinline_for_stack
3771 int find_next_extent(struct reloc_control *rc, struct btrfs_path *path,
3772 struct btrfs_key *extent_key)
3774 struct btrfs_key key;
3775 struct extent_buffer *leaf;
3776 u64 start, end, last;
3777 int ret;
3779 last = rc->block_group->key.objectid + rc->block_group->key.offset;
3780 while (1) {
3781 cond_resched();
3782 if (rc->search_start >= last) {
3783 ret = 1;
3784 break;
3787 key.objectid = rc->search_start;
3788 key.type = BTRFS_EXTENT_ITEM_KEY;
3789 key.offset = 0;
3791 path->search_commit_root = 1;
3792 path->skip_locking = 1;
3793 ret = btrfs_search_slot(NULL, rc->extent_root, &key, path,
3794 0, 0);
3795 if (ret < 0)
3796 break;
3797 next:
3798 leaf = path->nodes[0];
3799 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
3800 ret = btrfs_next_leaf(rc->extent_root, path);
3801 if (ret != 0)
3802 break;
3803 leaf = path->nodes[0];
3806 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3807 if (key.objectid >= last) {
3808 ret = 1;
3809 break;
3812 if (key.type != BTRFS_EXTENT_ITEM_KEY &&
3813 key.type != BTRFS_METADATA_ITEM_KEY) {
3814 path->slots[0]++;
3815 goto next;
3818 if (key.type == BTRFS_EXTENT_ITEM_KEY &&
3819 key.objectid + key.offset <= rc->search_start) {
3820 path->slots[0]++;
3821 goto next;
3824 if (key.type == BTRFS_METADATA_ITEM_KEY &&
3825 key.objectid + rc->extent_root->nodesize <=
3826 rc->search_start) {
3827 path->slots[0]++;
3828 goto next;
3831 ret = find_first_extent_bit(&rc->processed_blocks,
3832 key.objectid, &start, &end,
3833 EXTENT_DIRTY, NULL);
3835 if (ret == 0 && start <= key.objectid) {
3836 btrfs_release_path(path);
3837 rc->search_start = end + 1;
3838 } else {
3839 if (key.type == BTRFS_EXTENT_ITEM_KEY)
3840 rc->search_start = key.objectid + key.offset;
3841 else
3842 rc->search_start = key.objectid +
3843 rc->extent_root->nodesize;
3844 memcpy(extent_key, &key, sizeof(key));
3845 return 0;
3848 btrfs_release_path(path);
3849 return ret;
3852 static void set_reloc_control(struct reloc_control *rc)
3854 struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
3856 mutex_lock(&fs_info->reloc_mutex);
3857 fs_info->reloc_ctl = rc;
3858 mutex_unlock(&fs_info->reloc_mutex);
3861 static void unset_reloc_control(struct reloc_control *rc)
3863 struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
3865 mutex_lock(&fs_info->reloc_mutex);
3866 fs_info->reloc_ctl = NULL;
3867 mutex_unlock(&fs_info->reloc_mutex);
3870 static int check_extent_flags(u64 flags)
3872 if ((flags & BTRFS_EXTENT_FLAG_DATA) &&
3873 (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK))
3874 return 1;
3875 if (!(flags & BTRFS_EXTENT_FLAG_DATA) &&
3876 !(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK))
3877 return 1;
3878 if ((flags & BTRFS_EXTENT_FLAG_DATA) &&
3879 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
3880 return 1;
3881 return 0;
3884 static noinline_for_stack
3885 int prepare_to_relocate(struct reloc_control *rc)
3887 struct btrfs_trans_handle *trans;
3889 rc->block_rsv = btrfs_alloc_block_rsv(rc->extent_root,
3890 BTRFS_BLOCK_RSV_TEMP);
3891 if (!rc->block_rsv)
3892 return -ENOMEM;
3894 memset(&rc->cluster, 0, sizeof(rc->cluster));
3895 rc->search_start = rc->block_group->key.objectid;
3896 rc->extents_found = 0;
3897 rc->nodes_relocated = 0;
3898 rc->merging_rsv_size = 0;
3899 rc->reserved_bytes = 0;
3900 rc->block_rsv->size = rc->extent_root->nodesize *
3901 RELOCATION_RESERVED_NODES;
3903 rc->create_reloc_tree = 1;
3904 set_reloc_control(rc);
3906 trans = btrfs_join_transaction(rc->extent_root);
3907 if (IS_ERR(trans)) {
3908 unset_reloc_control(rc);
3910 * extent tree is not a ref_cow tree and has no reloc_root to
3911 * cleanup. And callers are responsible to free the above
3912 * block rsv.
3914 return PTR_ERR(trans);
3916 btrfs_commit_transaction(trans, rc->extent_root);
3917 return 0;
3920 static noinline_for_stack int relocate_block_group(struct reloc_control *rc)
3922 struct rb_root blocks = RB_ROOT;
3923 struct btrfs_key key;
3924 struct btrfs_trans_handle *trans = NULL;
3925 struct btrfs_path *path;
3926 struct btrfs_extent_item *ei;
3927 u64 flags;
3928 u32 item_size;
3929 int ret;
3930 int err = 0;
3931 int progress = 0;
3933 path = btrfs_alloc_path();
3934 if (!path)
3935 return -ENOMEM;
3936 path->reada = 1;
3938 ret = prepare_to_relocate(rc);
3939 if (ret) {
3940 err = ret;
3941 goto out_free;
3944 while (1) {
3945 rc->reserved_bytes = 0;
3946 ret = btrfs_block_rsv_refill(rc->extent_root,
3947 rc->block_rsv, rc->block_rsv->size,
3948 BTRFS_RESERVE_FLUSH_ALL);
3949 if (ret) {
3950 err = ret;
3951 break;
3953 progress++;
3954 trans = btrfs_start_transaction(rc->extent_root, 0);
3955 if (IS_ERR(trans)) {
3956 err = PTR_ERR(trans);
3957 trans = NULL;
3958 break;
3960 restart:
3961 if (update_backref_cache(trans, &rc->backref_cache)) {
3962 btrfs_end_transaction(trans, rc->extent_root);
3963 continue;
3966 ret = find_next_extent(rc, path, &key);
3967 if (ret < 0)
3968 err = ret;
3969 if (ret != 0)
3970 break;
3972 rc->extents_found++;
3974 ei = btrfs_item_ptr(path->nodes[0], path->slots[0],
3975 struct btrfs_extent_item);
3976 item_size = btrfs_item_size_nr(path->nodes[0], path->slots[0]);
3977 if (item_size >= sizeof(*ei)) {
3978 flags = btrfs_extent_flags(path->nodes[0], ei);
3979 ret = check_extent_flags(flags);
3980 BUG_ON(ret);
3982 } else {
3983 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
3984 u64 ref_owner;
3985 int path_change = 0;
3987 BUG_ON(item_size !=
3988 sizeof(struct btrfs_extent_item_v0));
3989 ret = get_ref_objectid_v0(rc, path, &key, &ref_owner,
3990 &path_change);
3991 if (ret < 0) {
3992 err = ret;
3993 break;
3995 if (ref_owner < BTRFS_FIRST_FREE_OBJECTID)
3996 flags = BTRFS_EXTENT_FLAG_TREE_BLOCK;
3997 else
3998 flags = BTRFS_EXTENT_FLAG_DATA;
4000 if (path_change) {
4001 btrfs_release_path(path);
4003 path->search_commit_root = 1;
4004 path->skip_locking = 1;
4005 ret = btrfs_search_slot(NULL, rc->extent_root,
4006 &key, path, 0, 0);
4007 if (ret < 0) {
4008 err = ret;
4009 break;
4011 BUG_ON(ret > 0);
4013 #else
4014 BUG();
4015 #endif
4018 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
4019 ret = add_tree_block(rc, &key, path, &blocks);
4020 } else if (rc->stage == UPDATE_DATA_PTRS &&
4021 (flags & BTRFS_EXTENT_FLAG_DATA)) {
4022 ret = add_data_references(rc, &key, path, &blocks);
4023 } else {
4024 btrfs_release_path(path);
4025 ret = 0;
4027 if (ret < 0) {
4028 err = ret;
4029 break;
4032 if (!RB_EMPTY_ROOT(&blocks)) {
4033 ret = relocate_tree_blocks(trans, rc, &blocks);
4034 if (ret < 0) {
4036 * if we fail to relocate tree blocks, force to update
4037 * backref cache when committing transaction.
4039 rc->backref_cache.last_trans = trans->transid - 1;
4041 if (ret != -EAGAIN) {
4042 err = ret;
4043 break;
4045 rc->extents_found--;
4046 rc->search_start = key.objectid;
4050 btrfs_end_transaction_throttle(trans, rc->extent_root);
4051 btrfs_btree_balance_dirty(rc->extent_root);
4052 trans = NULL;
4054 if (rc->stage == MOVE_DATA_EXTENTS &&
4055 (flags & BTRFS_EXTENT_FLAG_DATA)) {
4056 rc->found_file_extent = 1;
4057 ret = relocate_data_extent(rc->data_inode,
4058 &key, &rc->cluster);
4059 if (ret < 0) {
4060 err = ret;
4061 break;
4065 if (trans && progress && err == -ENOSPC) {
4066 ret = btrfs_force_chunk_alloc(trans, rc->extent_root,
4067 rc->block_group->flags);
4068 if (ret == 1) {
4069 err = 0;
4070 progress = 0;
4071 goto restart;
4075 btrfs_release_path(path);
4076 clear_extent_bits(&rc->processed_blocks, 0, (u64)-1, EXTENT_DIRTY,
4077 GFP_NOFS);
4079 if (trans) {
4080 btrfs_end_transaction_throttle(trans, rc->extent_root);
4081 btrfs_btree_balance_dirty(rc->extent_root);
4084 if (!err) {
4085 ret = relocate_file_extent_cluster(rc->data_inode,
4086 &rc->cluster);
4087 if (ret < 0)
4088 err = ret;
4091 rc->create_reloc_tree = 0;
4092 set_reloc_control(rc);
4094 backref_cache_cleanup(&rc->backref_cache);
4095 btrfs_block_rsv_release(rc->extent_root, rc->block_rsv, (u64)-1);
4097 err = prepare_to_merge(rc, err);
4099 merge_reloc_roots(rc);
4101 rc->merge_reloc_tree = 0;
4102 unset_reloc_control(rc);
4103 btrfs_block_rsv_release(rc->extent_root, rc->block_rsv, (u64)-1);
4105 /* get rid of pinned extents */
4106 trans = btrfs_join_transaction(rc->extent_root);
4107 if (IS_ERR(trans))
4108 err = PTR_ERR(trans);
4109 else
4110 btrfs_commit_transaction(trans, rc->extent_root);
4111 out_free:
4112 btrfs_free_block_rsv(rc->extent_root, rc->block_rsv);
4113 btrfs_free_path(path);
4114 return err;
4117 static int __insert_orphan_inode(struct btrfs_trans_handle *trans,
4118 struct btrfs_root *root, u64 objectid)
4120 struct btrfs_path *path;
4121 struct btrfs_inode_item *item;
4122 struct extent_buffer *leaf;
4123 int ret;
4125 path = btrfs_alloc_path();
4126 if (!path)
4127 return -ENOMEM;
4129 ret = btrfs_insert_empty_inode(trans, root, path, objectid);
4130 if (ret)
4131 goto out;
4133 leaf = path->nodes[0];
4134 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_inode_item);
4135 memset_extent_buffer(leaf, 0, (unsigned long)item, sizeof(*item));
4136 btrfs_set_inode_generation(leaf, item, 1);
4137 btrfs_set_inode_size(leaf, item, 0);
4138 btrfs_set_inode_mode(leaf, item, S_IFREG | 0600);
4139 btrfs_set_inode_flags(leaf, item, BTRFS_INODE_NOCOMPRESS |
4140 BTRFS_INODE_PREALLOC);
4141 btrfs_mark_buffer_dirty(leaf);
4142 out:
4143 btrfs_free_path(path);
4144 return ret;
4148 * helper to create inode for data relocation.
4149 * the inode is in data relocation tree and its link count is 0
4151 static noinline_for_stack
4152 struct inode *create_reloc_inode(struct btrfs_fs_info *fs_info,
4153 struct btrfs_block_group_cache *group)
4155 struct inode *inode = NULL;
4156 struct btrfs_trans_handle *trans;
4157 struct btrfs_root *root;
4158 struct btrfs_key key;
4159 u64 objectid;
4160 int err = 0;
4162 root = read_fs_root(fs_info, BTRFS_DATA_RELOC_TREE_OBJECTID);
4163 if (IS_ERR(root))
4164 return ERR_CAST(root);
4166 trans = btrfs_start_transaction(root, 6);
4167 if (IS_ERR(trans))
4168 return ERR_CAST(trans);
4170 err = btrfs_find_free_objectid(root, &objectid);
4171 if (err)
4172 goto out;
4174 err = __insert_orphan_inode(trans, root, objectid);
4175 BUG_ON(err);
4177 key.objectid = objectid;
4178 key.type = BTRFS_INODE_ITEM_KEY;
4179 key.offset = 0;
4180 inode = btrfs_iget(root->fs_info->sb, &key, root, NULL);
4181 BUG_ON(IS_ERR(inode) || is_bad_inode(inode));
4182 BTRFS_I(inode)->index_cnt = group->key.objectid;
4184 err = btrfs_orphan_add(trans, inode);
4185 out:
4186 btrfs_end_transaction(trans, root);
4187 btrfs_btree_balance_dirty(root);
4188 if (err) {
4189 if (inode)
4190 iput(inode);
4191 inode = ERR_PTR(err);
4193 return inode;
4196 static struct reloc_control *alloc_reloc_control(struct btrfs_fs_info *fs_info)
4198 struct reloc_control *rc;
4200 rc = kzalloc(sizeof(*rc), GFP_NOFS);
4201 if (!rc)
4202 return NULL;
4204 INIT_LIST_HEAD(&rc->reloc_roots);
4205 backref_cache_init(&rc->backref_cache);
4206 mapping_tree_init(&rc->reloc_root_tree);
4207 extent_io_tree_init(&rc->processed_blocks,
4208 fs_info->btree_inode->i_mapping);
4209 return rc;
4213 * function to relocate all extents in a block group.
4215 int btrfs_relocate_block_group(struct btrfs_root *extent_root, u64 group_start)
4217 struct btrfs_fs_info *fs_info = extent_root->fs_info;
4218 struct reloc_control *rc;
4219 struct inode *inode;
4220 struct btrfs_path *path;
4221 int ret;
4222 int rw = 0;
4223 int err = 0;
4225 rc = alloc_reloc_control(fs_info);
4226 if (!rc)
4227 return -ENOMEM;
4229 rc->extent_root = extent_root;
4231 rc->block_group = btrfs_lookup_block_group(fs_info, group_start);
4232 BUG_ON(!rc->block_group);
4234 ret = btrfs_inc_block_group_ro(extent_root, rc->block_group);
4235 if (ret) {
4236 err = ret;
4237 goto out;
4239 rw = 1;
4241 path = btrfs_alloc_path();
4242 if (!path) {
4243 err = -ENOMEM;
4244 goto out;
4247 inode = lookup_free_space_inode(fs_info->tree_root, rc->block_group,
4248 path);
4249 btrfs_free_path(path);
4251 if (!IS_ERR(inode))
4252 ret = delete_block_group_cache(fs_info, rc->block_group, inode, 0);
4253 else
4254 ret = PTR_ERR(inode);
4256 if (ret && ret != -ENOENT) {
4257 err = ret;
4258 goto out;
4261 rc->data_inode = create_reloc_inode(fs_info, rc->block_group);
4262 if (IS_ERR(rc->data_inode)) {
4263 err = PTR_ERR(rc->data_inode);
4264 rc->data_inode = NULL;
4265 goto out;
4268 btrfs_info(extent_root->fs_info, "relocating block group %llu flags %llu",
4269 rc->block_group->key.objectid, rc->block_group->flags);
4271 ret = btrfs_start_delalloc_roots(fs_info, 0, -1);
4272 if (ret < 0) {
4273 err = ret;
4274 goto out;
4276 btrfs_wait_ordered_roots(fs_info, -1);
4278 while (1) {
4279 mutex_lock(&fs_info->cleaner_mutex);
4280 ret = relocate_block_group(rc);
4281 mutex_unlock(&fs_info->cleaner_mutex);
4282 if (ret < 0) {
4283 err = ret;
4284 goto out;
4287 if (rc->extents_found == 0)
4288 break;
4290 btrfs_info(extent_root->fs_info, "found %llu extents",
4291 rc->extents_found);
4293 if (rc->stage == MOVE_DATA_EXTENTS && rc->found_file_extent) {
4294 ret = btrfs_wait_ordered_range(rc->data_inode, 0,
4295 (u64)-1);
4296 if (ret) {
4297 err = ret;
4298 goto out;
4300 invalidate_mapping_pages(rc->data_inode->i_mapping,
4301 0, -1);
4302 rc->stage = UPDATE_DATA_PTRS;
4306 WARN_ON(rc->block_group->pinned > 0);
4307 WARN_ON(rc->block_group->reserved > 0);
4308 WARN_ON(btrfs_block_group_used(&rc->block_group->item) > 0);
4309 out:
4310 if (err && rw)
4311 btrfs_dec_block_group_ro(extent_root, rc->block_group);
4312 iput(rc->data_inode);
4313 btrfs_put_block_group(rc->block_group);
4314 kfree(rc);
4315 return err;
4318 static noinline_for_stack int mark_garbage_root(struct btrfs_root *root)
4320 struct btrfs_trans_handle *trans;
4321 int ret, err;
4323 trans = btrfs_start_transaction(root->fs_info->tree_root, 0);
4324 if (IS_ERR(trans))
4325 return PTR_ERR(trans);
4327 memset(&root->root_item.drop_progress, 0,
4328 sizeof(root->root_item.drop_progress));
4329 root->root_item.drop_level = 0;
4330 btrfs_set_root_refs(&root->root_item, 0);
4331 ret = btrfs_update_root(trans, root->fs_info->tree_root,
4332 &root->root_key, &root->root_item);
4334 err = btrfs_end_transaction(trans, root->fs_info->tree_root);
4335 if (err)
4336 return err;
4337 return ret;
4341 * recover relocation interrupted by system crash.
4343 * this function resumes merging reloc trees with corresponding fs trees.
4344 * this is important for keeping the sharing of tree blocks
4346 int btrfs_recover_relocation(struct btrfs_root *root)
4348 LIST_HEAD(reloc_roots);
4349 struct btrfs_key key;
4350 struct btrfs_root *fs_root;
4351 struct btrfs_root *reloc_root;
4352 struct btrfs_path *path;
4353 struct extent_buffer *leaf;
4354 struct reloc_control *rc = NULL;
4355 struct btrfs_trans_handle *trans;
4356 int ret;
4357 int err = 0;
4359 path = btrfs_alloc_path();
4360 if (!path)
4361 return -ENOMEM;
4362 path->reada = -1;
4364 key.objectid = BTRFS_TREE_RELOC_OBJECTID;
4365 key.type = BTRFS_ROOT_ITEM_KEY;
4366 key.offset = (u64)-1;
4368 while (1) {
4369 ret = btrfs_search_slot(NULL, root->fs_info->tree_root, &key,
4370 path, 0, 0);
4371 if (ret < 0) {
4372 err = ret;
4373 goto out;
4375 if (ret > 0) {
4376 if (path->slots[0] == 0)
4377 break;
4378 path->slots[0]--;
4380 leaf = path->nodes[0];
4381 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4382 btrfs_release_path(path);
4384 if (key.objectid != BTRFS_TREE_RELOC_OBJECTID ||
4385 key.type != BTRFS_ROOT_ITEM_KEY)
4386 break;
4388 reloc_root = btrfs_read_fs_root(root, &key);
4389 if (IS_ERR(reloc_root)) {
4390 err = PTR_ERR(reloc_root);
4391 goto out;
4394 list_add(&reloc_root->root_list, &reloc_roots);
4396 if (btrfs_root_refs(&reloc_root->root_item) > 0) {
4397 fs_root = read_fs_root(root->fs_info,
4398 reloc_root->root_key.offset);
4399 if (IS_ERR(fs_root)) {
4400 ret = PTR_ERR(fs_root);
4401 if (ret != -ENOENT) {
4402 err = ret;
4403 goto out;
4405 ret = mark_garbage_root(reloc_root);
4406 if (ret < 0) {
4407 err = ret;
4408 goto out;
4413 if (key.offset == 0)
4414 break;
4416 key.offset--;
4418 btrfs_release_path(path);
4420 if (list_empty(&reloc_roots))
4421 goto out;
4423 rc = alloc_reloc_control(root->fs_info);
4424 if (!rc) {
4425 err = -ENOMEM;
4426 goto out;
4429 rc->extent_root = root->fs_info->extent_root;
4431 set_reloc_control(rc);
4433 trans = btrfs_join_transaction(rc->extent_root);
4434 if (IS_ERR(trans)) {
4435 unset_reloc_control(rc);
4436 err = PTR_ERR(trans);
4437 goto out_free;
4440 rc->merge_reloc_tree = 1;
4442 while (!list_empty(&reloc_roots)) {
4443 reloc_root = list_entry(reloc_roots.next,
4444 struct btrfs_root, root_list);
4445 list_del(&reloc_root->root_list);
4447 if (btrfs_root_refs(&reloc_root->root_item) == 0) {
4448 list_add_tail(&reloc_root->root_list,
4449 &rc->reloc_roots);
4450 continue;
4453 fs_root = read_fs_root(root->fs_info,
4454 reloc_root->root_key.offset);
4455 if (IS_ERR(fs_root)) {
4456 err = PTR_ERR(fs_root);
4457 goto out_free;
4460 err = __add_reloc_root(reloc_root);
4461 BUG_ON(err < 0); /* -ENOMEM or logic error */
4462 fs_root->reloc_root = reloc_root;
4465 err = btrfs_commit_transaction(trans, rc->extent_root);
4466 if (err)
4467 goto out_free;
4469 merge_reloc_roots(rc);
4471 unset_reloc_control(rc);
4473 trans = btrfs_join_transaction(rc->extent_root);
4474 if (IS_ERR(trans))
4475 err = PTR_ERR(trans);
4476 else
4477 err = btrfs_commit_transaction(trans, rc->extent_root);
4478 out_free:
4479 kfree(rc);
4480 out:
4481 if (!list_empty(&reloc_roots))
4482 free_reloc_roots(&reloc_roots);
4484 btrfs_free_path(path);
4486 if (err == 0) {
4487 /* cleanup orphan inode in data relocation tree */
4488 fs_root = read_fs_root(root->fs_info,
4489 BTRFS_DATA_RELOC_TREE_OBJECTID);
4490 if (IS_ERR(fs_root))
4491 err = PTR_ERR(fs_root);
4492 else
4493 err = btrfs_orphan_cleanup(fs_root);
4495 return err;
4499 * helper to add ordered checksum for data relocation.
4501 * cloning checksum properly handles the nodatasum extents.
4502 * it also saves CPU time to re-calculate the checksum.
4504 int btrfs_reloc_clone_csums(struct inode *inode, u64 file_pos, u64 len)
4506 struct btrfs_ordered_sum *sums;
4507 struct btrfs_ordered_extent *ordered;
4508 struct btrfs_root *root = BTRFS_I(inode)->root;
4509 int ret;
4510 u64 disk_bytenr;
4511 u64 new_bytenr;
4512 LIST_HEAD(list);
4514 ordered = btrfs_lookup_ordered_extent(inode, file_pos);
4515 BUG_ON(ordered->file_offset != file_pos || ordered->len != len);
4517 disk_bytenr = file_pos + BTRFS_I(inode)->index_cnt;
4518 ret = btrfs_lookup_csums_range(root->fs_info->csum_root, disk_bytenr,
4519 disk_bytenr + len - 1, &list, 0);
4520 if (ret)
4521 goto out;
4523 while (!list_empty(&list)) {
4524 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
4525 list_del_init(&sums->list);
4528 * We need to offset the new_bytenr based on where the csum is.
4529 * We need to do this because we will read in entire prealloc
4530 * extents but we may have written to say the middle of the
4531 * prealloc extent, so we need to make sure the csum goes with
4532 * the right disk offset.
4534 * We can do this because the data reloc inode refers strictly
4535 * to the on disk bytes, so we don't have to worry about
4536 * disk_len vs real len like with real inodes since it's all
4537 * disk length.
4539 new_bytenr = ordered->start + (sums->bytenr - disk_bytenr);
4540 sums->bytenr = new_bytenr;
4542 btrfs_add_ordered_sum(inode, ordered, sums);
4544 out:
4545 btrfs_put_ordered_extent(ordered);
4546 return ret;
4549 int btrfs_reloc_cow_block(struct btrfs_trans_handle *trans,
4550 struct btrfs_root *root, struct extent_buffer *buf,
4551 struct extent_buffer *cow)
4553 struct reloc_control *rc;
4554 struct backref_node *node;
4555 int first_cow = 0;
4556 int level;
4557 int ret = 0;
4559 rc = root->fs_info->reloc_ctl;
4560 if (!rc)
4561 return 0;
4563 BUG_ON(rc->stage == UPDATE_DATA_PTRS &&
4564 root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID);
4566 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) {
4567 if (buf == root->node)
4568 __update_reloc_root(root, cow->start);
4571 level = btrfs_header_level(buf);
4572 if (btrfs_header_generation(buf) <=
4573 btrfs_root_last_snapshot(&root->root_item))
4574 first_cow = 1;
4576 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID &&
4577 rc->create_reloc_tree) {
4578 WARN_ON(!first_cow && level == 0);
4580 node = rc->backref_cache.path[level];
4581 BUG_ON(node->bytenr != buf->start &&
4582 node->new_bytenr != buf->start);
4584 drop_node_buffer(node);
4585 extent_buffer_get(cow);
4586 node->eb = cow;
4587 node->new_bytenr = cow->start;
4589 if (!node->pending) {
4590 list_move_tail(&node->list,
4591 &rc->backref_cache.pending[level]);
4592 node->pending = 1;
4595 if (first_cow)
4596 __mark_block_processed(rc, node);
4598 if (first_cow && level > 0)
4599 rc->nodes_relocated += buf->len;
4602 if (level == 0 && first_cow && rc->stage == UPDATE_DATA_PTRS)
4603 ret = replace_file_extents(trans, rc, root, cow);
4604 return ret;
4608 * called before creating snapshot. it calculates metadata reservation
4609 * requried for relocating tree blocks in the snapshot
4611 void btrfs_reloc_pre_snapshot(struct btrfs_pending_snapshot *pending,
4612 u64 *bytes_to_reserve)
4614 struct btrfs_root *root;
4615 struct reloc_control *rc;
4617 root = pending->root;
4618 if (!root->reloc_root)
4619 return;
4621 rc = root->fs_info->reloc_ctl;
4622 if (!rc->merge_reloc_tree)
4623 return;
4625 root = root->reloc_root;
4626 BUG_ON(btrfs_root_refs(&root->root_item) == 0);
4628 * relocation is in the stage of merging trees. the space
4629 * used by merging a reloc tree is twice the size of
4630 * relocated tree nodes in the worst case. half for cowing
4631 * the reloc tree, half for cowing the fs tree. the space
4632 * used by cowing the reloc tree will be freed after the
4633 * tree is dropped. if we create snapshot, cowing the fs
4634 * tree may use more space than it frees. so we need
4635 * reserve extra space.
4637 *bytes_to_reserve += rc->nodes_relocated;
4641 * called after snapshot is created. migrate block reservation
4642 * and create reloc root for the newly created snapshot
4644 int btrfs_reloc_post_snapshot(struct btrfs_trans_handle *trans,
4645 struct btrfs_pending_snapshot *pending)
4647 struct btrfs_root *root = pending->root;
4648 struct btrfs_root *reloc_root;
4649 struct btrfs_root *new_root;
4650 struct reloc_control *rc;
4651 int ret;
4653 if (!root->reloc_root)
4654 return 0;
4656 rc = root->fs_info->reloc_ctl;
4657 rc->merging_rsv_size += rc->nodes_relocated;
4659 if (rc->merge_reloc_tree) {
4660 ret = btrfs_block_rsv_migrate(&pending->block_rsv,
4661 rc->block_rsv,
4662 rc->nodes_relocated);
4663 if (ret)
4664 return ret;
4667 new_root = pending->snap;
4668 reloc_root = create_reloc_root(trans, root->reloc_root,
4669 new_root->root_key.objectid);
4670 if (IS_ERR(reloc_root))
4671 return PTR_ERR(reloc_root);
4673 ret = __add_reloc_root(reloc_root);
4674 BUG_ON(ret < 0);
4675 new_root->reloc_root = reloc_root;
4677 if (rc->create_reloc_tree)
4678 ret = clone_backref_node(trans, rc, root, reloc_root);
4679 return ret;