Loading...
1/*
2 * Copyright (C) 2009 Oracle. All rights reserved.
3 *
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.
7 *
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.
12 *
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.
17 */
18
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"
34
35/*
36 * backref_node, mapping_node and tree_block start with this
37 */
38struct tree_entry {
39 struct rb_node rb_node;
40 u64 bytenr;
41};
42
43/*
44 * present a tree block in the backref cache
45 */
46struct backref_node {
47 struct rb_node rb_node;
48 u64 bytenr;
49
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;
75 /*
76 * 1 if corresponding block has been cowed but some upper
77 * level block pointers may not point to the new location
78 */
79 unsigned int pending:1;
80 /*
81 * 1 if the backref node isn't connected to any other
82 * backref node.
83 */
84 unsigned int detached:1;
85};
86
87/*
88 * present a block pointer in the backref cache
89 */
90struct backref_edge {
91 struct list_head list[2];
92 struct backref_node *node[2];
93};
94
95#define LOWER 0
96#define UPPER 1
97
98struct backref_cache {
99 /* red black tree of all backref nodes in the cache */
100 struct rb_root rb_root;
101 /* for passing backref nodes to btrfs_reloc_cow_block */
102 struct backref_node *path[BTRFS_MAX_LEVEL];
103 /*
104 * list of blocks that have been cowed but some block
105 * pointers in upper level blocks may not reflect the
106 * new location
107 */
108 struct list_head pending[BTRFS_MAX_LEVEL];
109 /* list of backref nodes with no child node */
110 struct list_head leaves;
111 /* list of blocks that have been cowed in current transaction */
112 struct list_head changed;
113 /* list of detached backref node. */
114 struct list_head detached;
115
116 u64 last_trans;
117
118 int nr_nodes;
119 int nr_edges;
120};
121
122/*
123 * map address of tree root to tree
124 */
125struct mapping_node {
126 struct rb_node rb_node;
127 u64 bytenr;
128 void *data;
129};
130
131struct mapping_tree {
132 struct rb_root rb_root;
133 spinlock_t lock;
134};
135
136/*
137 * present a tree block to process
138 */
139struct tree_block {
140 struct rb_node rb_node;
141 u64 bytenr;
142 struct btrfs_key key;
143 unsigned int level:8;
144 unsigned int key_ready:1;
145};
146
147#define MAX_EXTENTS 128
148
149struct file_extent_cluster {
150 u64 start;
151 u64 end;
152 u64 boundary[MAX_EXTENTS];
153 unsigned int nr;
154};
155
156struct reloc_control {
157 /* block group to relocate */
158 struct btrfs_block_group_cache *block_group;
159 /* extent tree */
160 struct btrfs_root *extent_root;
161 /* inode for moving data */
162 struct inode *data_inode;
163
164 struct btrfs_block_rsv *block_rsv;
165
166 struct backref_cache backref_cache;
167
168 struct file_extent_cluster cluster;
169 /* tree blocks have been processed */
170 struct extent_io_tree processed_blocks;
171 /* map start of tree root to corresponding reloc tree */
172 struct mapping_tree reloc_root_tree;
173 /* list of reloc trees */
174 struct list_head reloc_roots;
175 /* size of metadata reservation for merging reloc trees */
176 u64 merging_rsv_size;
177 /* size of relocated tree nodes */
178 u64 nodes_relocated;
179
180 u64 search_start;
181 u64 extents_found;
182
183 unsigned int stage:8;
184 unsigned int create_reloc_tree:1;
185 unsigned int merge_reloc_tree:1;
186 unsigned int found_file_extent:1;
187 unsigned int commit_transaction:1;
188};
189
190/* stages of data relocation */
191#define MOVE_DATA_EXTENTS 0
192#define UPDATE_DATA_PTRS 1
193
194static void remove_backref_node(struct backref_cache *cache,
195 struct backref_node *node);
196static void __mark_block_processed(struct reloc_control *rc,
197 struct backref_node *node);
198
199static void mapping_tree_init(struct mapping_tree *tree)
200{
201 tree->rb_root = RB_ROOT;
202 spin_lock_init(&tree->lock);
203}
204
205static void backref_cache_init(struct backref_cache *cache)
206{
207 int i;
208 cache->rb_root = RB_ROOT;
209 for (i = 0; i < BTRFS_MAX_LEVEL; i++)
210 INIT_LIST_HEAD(&cache->pending[i]);
211 INIT_LIST_HEAD(&cache->changed);
212 INIT_LIST_HEAD(&cache->detached);
213 INIT_LIST_HEAD(&cache->leaves);
214}
215
216static void backref_cache_cleanup(struct backref_cache *cache)
217{
218 struct backref_node *node;
219 int i;
220
221 while (!list_empty(&cache->detached)) {
222 node = list_entry(cache->detached.next,
223 struct backref_node, list);
224 remove_backref_node(cache, node);
225 }
226
227 while (!list_empty(&cache->leaves)) {
228 node = list_entry(cache->leaves.next,
229 struct backref_node, lower);
230 remove_backref_node(cache, node);
231 }
232
233 cache->last_trans = 0;
234
235 for (i = 0; i < BTRFS_MAX_LEVEL; i++)
236 BUG_ON(!list_empty(&cache->pending[i]));
237 BUG_ON(!list_empty(&cache->changed));
238 BUG_ON(!list_empty(&cache->detached));
239 BUG_ON(!RB_EMPTY_ROOT(&cache->rb_root));
240 BUG_ON(cache->nr_nodes);
241 BUG_ON(cache->nr_edges);
242}
243
244static struct backref_node *alloc_backref_node(struct backref_cache *cache)
245{
246 struct backref_node *node;
247
248 node = kzalloc(sizeof(*node), GFP_NOFS);
249 if (node) {
250 INIT_LIST_HEAD(&node->list);
251 INIT_LIST_HEAD(&node->upper);
252 INIT_LIST_HEAD(&node->lower);
253 RB_CLEAR_NODE(&node->rb_node);
254 cache->nr_nodes++;
255 }
256 return node;
257}
258
259static void free_backref_node(struct backref_cache *cache,
260 struct backref_node *node)
261{
262 if (node) {
263 cache->nr_nodes--;
264 kfree(node);
265 }
266}
267
268static struct backref_edge *alloc_backref_edge(struct backref_cache *cache)
269{
270 struct backref_edge *edge;
271
272 edge = kzalloc(sizeof(*edge), GFP_NOFS);
273 if (edge)
274 cache->nr_edges++;
275 return edge;
276}
277
278static void free_backref_edge(struct backref_cache *cache,
279 struct backref_edge *edge)
280{
281 if (edge) {
282 cache->nr_edges--;
283 kfree(edge);
284 }
285}
286
287static struct rb_node *tree_insert(struct rb_root *root, u64 bytenr,
288 struct rb_node *node)
289{
290 struct rb_node **p = &root->rb_node;
291 struct rb_node *parent = NULL;
292 struct tree_entry *entry;
293
294 while (*p) {
295 parent = *p;
296 entry = rb_entry(parent, struct tree_entry, rb_node);
297
298 if (bytenr < entry->bytenr)
299 p = &(*p)->rb_left;
300 else if (bytenr > entry->bytenr)
301 p = &(*p)->rb_right;
302 else
303 return parent;
304 }
305
306 rb_link_node(node, parent, p);
307 rb_insert_color(node, root);
308 return NULL;
309}
310
311static struct rb_node *tree_search(struct rb_root *root, u64 bytenr)
312{
313 struct rb_node *n = root->rb_node;
314 struct tree_entry *entry;
315
316 while (n) {
317 entry = rb_entry(n, struct tree_entry, rb_node);
318
319 if (bytenr < entry->bytenr)
320 n = n->rb_left;
321 else if (bytenr > entry->bytenr)
322 n = n->rb_right;
323 else
324 return n;
325 }
326 return NULL;
327}
328
329void backref_tree_panic(struct rb_node *rb_node, int errno,
330 u64 bytenr)
331{
332
333 struct btrfs_fs_info *fs_info = NULL;
334 struct backref_node *bnode = rb_entry(rb_node, struct backref_node,
335 rb_node);
336 if (bnode->root)
337 fs_info = bnode->root->fs_info;
338 btrfs_panic(fs_info, errno, "Inconsistency in backref cache "
339 "found at offset %llu\n", (unsigned long long)bytenr);
340}
341
342/*
343 * walk up backref nodes until reach node presents tree root
344 */
345static struct backref_node *walk_up_backref(struct backref_node *node,
346 struct backref_edge *edges[],
347 int *index)
348{
349 struct backref_edge *edge;
350 int idx = *index;
351
352 while (!list_empty(&node->upper)) {
353 edge = list_entry(node->upper.next,
354 struct backref_edge, list[LOWER]);
355 edges[idx++] = edge;
356 node = edge->node[UPPER];
357 }
358 BUG_ON(node->detached);
359 *index = idx;
360 return node;
361}
362
363/*
364 * walk down backref nodes to find start of next reference path
365 */
366static struct backref_node *walk_down_backref(struct backref_edge *edges[],
367 int *index)
368{
369 struct backref_edge *edge;
370 struct backref_node *lower;
371 int idx = *index;
372
373 while (idx > 0) {
374 edge = edges[idx - 1];
375 lower = edge->node[LOWER];
376 if (list_is_last(&edge->list[LOWER], &lower->upper)) {
377 idx--;
378 continue;
379 }
380 edge = list_entry(edge->list[LOWER].next,
381 struct backref_edge, list[LOWER]);
382 edges[idx - 1] = edge;
383 *index = idx;
384 return edge->node[UPPER];
385 }
386 *index = 0;
387 return NULL;
388}
389
390static void unlock_node_buffer(struct backref_node *node)
391{
392 if (node->locked) {
393 btrfs_tree_unlock(node->eb);
394 node->locked = 0;
395 }
396}
397
398static void drop_node_buffer(struct backref_node *node)
399{
400 if (node->eb) {
401 unlock_node_buffer(node);
402 free_extent_buffer(node->eb);
403 node->eb = NULL;
404 }
405}
406
407static void drop_backref_node(struct backref_cache *tree,
408 struct backref_node *node)
409{
410 BUG_ON(!list_empty(&node->upper));
411
412 drop_node_buffer(node);
413 list_del(&node->list);
414 list_del(&node->lower);
415 if (!RB_EMPTY_NODE(&node->rb_node))
416 rb_erase(&node->rb_node, &tree->rb_root);
417 free_backref_node(tree, node);
418}
419
420/*
421 * remove a backref node from the backref cache
422 */
423static void remove_backref_node(struct backref_cache *cache,
424 struct backref_node *node)
425{
426 struct backref_node *upper;
427 struct backref_edge *edge;
428
429 if (!node)
430 return;
431
432 BUG_ON(!node->lowest && !node->detached);
433 while (!list_empty(&node->upper)) {
434 edge = list_entry(node->upper.next, struct backref_edge,
435 list[LOWER]);
436 upper = edge->node[UPPER];
437 list_del(&edge->list[LOWER]);
438 list_del(&edge->list[UPPER]);
439 free_backref_edge(cache, edge);
440
441 if (RB_EMPTY_NODE(&upper->rb_node)) {
442 BUG_ON(!list_empty(&node->upper));
443 drop_backref_node(cache, node);
444 node = upper;
445 node->lowest = 1;
446 continue;
447 }
448 /*
449 * add the node to leaf node list if no other
450 * child block cached.
451 */
452 if (list_empty(&upper->lower)) {
453 list_add_tail(&upper->lower, &cache->leaves);
454 upper->lowest = 1;
455 }
456 }
457
458 drop_backref_node(cache, node);
459}
460
461static void update_backref_node(struct backref_cache *cache,
462 struct backref_node *node, u64 bytenr)
463{
464 struct rb_node *rb_node;
465 rb_erase(&node->rb_node, &cache->rb_root);
466 node->bytenr = bytenr;
467 rb_node = tree_insert(&cache->rb_root, node->bytenr, &node->rb_node);
468 if (rb_node)
469 backref_tree_panic(rb_node, -EEXIST, bytenr);
470}
471
472/*
473 * update backref cache after a transaction commit
474 */
475static int update_backref_cache(struct btrfs_trans_handle *trans,
476 struct backref_cache *cache)
477{
478 struct backref_node *node;
479 int level = 0;
480
481 if (cache->last_trans == 0) {
482 cache->last_trans = trans->transid;
483 return 0;
484 }
485
486 if (cache->last_trans == trans->transid)
487 return 0;
488
489 /*
490 * detached nodes are used to avoid unnecessary backref
491 * lookup. transaction commit changes the extent tree.
492 * so the detached nodes are no longer useful.
493 */
494 while (!list_empty(&cache->detached)) {
495 node = list_entry(cache->detached.next,
496 struct backref_node, list);
497 remove_backref_node(cache, node);
498 }
499
500 while (!list_empty(&cache->changed)) {
501 node = list_entry(cache->changed.next,
502 struct backref_node, list);
503 list_del_init(&node->list);
504 BUG_ON(node->pending);
505 update_backref_node(cache, node, node->new_bytenr);
506 }
507
508 /*
509 * some nodes can be left in the pending list if there were
510 * errors during processing the pending nodes.
511 */
512 for (level = 0; level < BTRFS_MAX_LEVEL; level++) {
513 list_for_each_entry(node, &cache->pending[level], list) {
514 BUG_ON(!node->pending);
515 if (node->bytenr == node->new_bytenr)
516 continue;
517 update_backref_node(cache, node, node->new_bytenr);
518 }
519 }
520
521 cache->last_trans = 0;
522 return 1;
523}
524
525
526static int should_ignore_root(struct btrfs_root *root)
527{
528 struct btrfs_root *reloc_root;
529
530 if (!root->ref_cows)
531 return 0;
532
533 reloc_root = root->reloc_root;
534 if (!reloc_root)
535 return 0;
536
537 if (btrfs_root_last_snapshot(&reloc_root->root_item) ==
538 root->fs_info->running_transaction->transid - 1)
539 return 0;
540 /*
541 * if there is reloc tree and it was created in previous
542 * transaction backref lookup can find the reloc tree,
543 * so backref node for the fs tree root is useless for
544 * relocation.
545 */
546 return 1;
547}
548/*
549 * find reloc tree by address of tree root
550 */
551static struct btrfs_root *find_reloc_root(struct reloc_control *rc,
552 u64 bytenr)
553{
554 struct rb_node *rb_node;
555 struct mapping_node *node;
556 struct btrfs_root *root = NULL;
557
558 spin_lock(&rc->reloc_root_tree.lock);
559 rb_node = tree_search(&rc->reloc_root_tree.rb_root, bytenr);
560 if (rb_node) {
561 node = rb_entry(rb_node, struct mapping_node, rb_node);
562 root = (struct btrfs_root *)node->data;
563 }
564 spin_unlock(&rc->reloc_root_tree.lock);
565 return root;
566}
567
568static int is_cowonly_root(u64 root_objectid)
569{
570 if (root_objectid == BTRFS_ROOT_TREE_OBJECTID ||
571 root_objectid == BTRFS_EXTENT_TREE_OBJECTID ||
572 root_objectid == BTRFS_CHUNK_TREE_OBJECTID ||
573 root_objectid == BTRFS_DEV_TREE_OBJECTID ||
574 root_objectid == BTRFS_TREE_LOG_OBJECTID ||
575 root_objectid == BTRFS_CSUM_TREE_OBJECTID)
576 return 1;
577 return 0;
578}
579
580static struct btrfs_root *read_fs_root(struct btrfs_fs_info *fs_info,
581 u64 root_objectid)
582{
583 struct btrfs_key key;
584
585 key.objectid = root_objectid;
586 key.type = BTRFS_ROOT_ITEM_KEY;
587 if (is_cowonly_root(root_objectid))
588 key.offset = 0;
589 else
590 key.offset = (u64)-1;
591
592 return btrfs_read_fs_root_no_name(fs_info, &key);
593}
594
595#ifdef BTRFS_COMPAT_EXTENT_TREE_V0
596static noinline_for_stack
597struct btrfs_root *find_tree_root(struct reloc_control *rc,
598 struct extent_buffer *leaf,
599 struct btrfs_extent_ref_v0 *ref0)
600{
601 struct btrfs_root *root;
602 u64 root_objectid = btrfs_ref_root_v0(leaf, ref0);
603 u64 generation = btrfs_ref_generation_v0(leaf, ref0);
604
605 BUG_ON(root_objectid == BTRFS_TREE_RELOC_OBJECTID);
606
607 root = read_fs_root(rc->extent_root->fs_info, root_objectid);
608 BUG_ON(IS_ERR(root));
609
610 if (root->ref_cows &&
611 generation != btrfs_root_generation(&root->root_item))
612 return NULL;
613
614 return root;
615}
616#endif
617
618static noinline_for_stack
619int find_inline_backref(struct extent_buffer *leaf, int slot,
620 unsigned long *ptr, unsigned long *end)
621{
622 struct btrfs_extent_item *ei;
623 struct btrfs_tree_block_info *bi;
624 u32 item_size;
625
626 item_size = btrfs_item_size_nr(leaf, slot);
627#ifdef BTRFS_COMPAT_EXTENT_TREE_V0
628 if (item_size < sizeof(*ei)) {
629 WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0));
630 return 1;
631 }
632#endif
633 ei = btrfs_item_ptr(leaf, slot, struct btrfs_extent_item);
634 WARN_ON(!(btrfs_extent_flags(leaf, ei) &
635 BTRFS_EXTENT_FLAG_TREE_BLOCK));
636
637 if (item_size <= sizeof(*ei) + sizeof(*bi)) {
638 WARN_ON(item_size < sizeof(*ei) + sizeof(*bi));
639 return 1;
640 }
641
642 bi = (struct btrfs_tree_block_info *)(ei + 1);
643 *ptr = (unsigned long)(bi + 1);
644 *end = (unsigned long)ei + item_size;
645 return 0;
646}
647
648/*
649 * build backref tree for a given tree block. root of the backref tree
650 * corresponds the tree block, leaves of the backref tree correspond
651 * roots of b-trees that reference the tree block.
652 *
653 * the basic idea of this function is check backrefs of a given block
654 * to find upper level blocks that refernece the block, and then check
655 * bakcrefs of these upper level blocks recursively. the recursion stop
656 * when tree root is reached or backrefs for the block is cached.
657 *
658 * NOTE: if we find backrefs for a block are cached, we know backrefs
659 * for all upper level blocks that directly/indirectly reference the
660 * block are also cached.
661 */
662static noinline_for_stack
663struct backref_node *build_backref_tree(struct reloc_control *rc,
664 struct btrfs_key *node_key,
665 int level, u64 bytenr)
666{
667 struct backref_cache *cache = &rc->backref_cache;
668 struct btrfs_path *path1;
669 struct btrfs_path *path2;
670 struct extent_buffer *eb;
671 struct btrfs_root *root;
672 struct backref_node *cur;
673 struct backref_node *upper;
674 struct backref_node *lower;
675 struct backref_node *node = NULL;
676 struct backref_node *exist = NULL;
677 struct backref_edge *edge;
678 struct rb_node *rb_node;
679 struct btrfs_key key;
680 unsigned long end;
681 unsigned long ptr;
682 LIST_HEAD(list);
683 LIST_HEAD(useless);
684 int cowonly;
685 int ret;
686 int err = 0;
687
688 path1 = btrfs_alloc_path();
689 path2 = btrfs_alloc_path();
690 if (!path1 || !path2) {
691 err = -ENOMEM;
692 goto out;
693 }
694 path1->reada = 1;
695 path2->reada = 2;
696
697 node = alloc_backref_node(cache);
698 if (!node) {
699 err = -ENOMEM;
700 goto out;
701 }
702
703 node->bytenr = bytenr;
704 node->level = level;
705 node->lowest = 1;
706 cur = node;
707again:
708 end = 0;
709 ptr = 0;
710 key.objectid = cur->bytenr;
711 key.type = BTRFS_EXTENT_ITEM_KEY;
712 key.offset = (u64)-1;
713
714 path1->search_commit_root = 1;
715 path1->skip_locking = 1;
716 ret = btrfs_search_slot(NULL, rc->extent_root, &key, path1,
717 0, 0);
718 if (ret < 0) {
719 err = ret;
720 goto out;
721 }
722 BUG_ON(!ret || !path1->slots[0]);
723
724 path1->slots[0]--;
725
726 WARN_ON(cur->checked);
727 if (!list_empty(&cur->upper)) {
728 /*
729 * the backref was added previously when processing
730 * backref of type BTRFS_TREE_BLOCK_REF_KEY
731 */
732 BUG_ON(!list_is_singular(&cur->upper));
733 edge = list_entry(cur->upper.next, struct backref_edge,
734 list[LOWER]);
735 BUG_ON(!list_empty(&edge->list[UPPER]));
736 exist = edge->node[UPPER];
737 /*
738 * add the upper level block to pending list if we need
739 * check its backrefs
740 */
741 if (!exist->checked)
742 list_add_tail(&edge->list[UPPER], &list);
743 } else {
744 exist = NULL;
745 }
746
747 while (1) {
748 cond_resched();
749 eb = path1->nodes[0];
750
751 if (ptr >= end) {
752 if (path1->slots[0] >= btrfs_header_nritems(eb)) {
753 ret = btrfs_next_leaf(rc->extent_root, path1);
754 if (ret < 0) {
755 err = ret;
756 goto out;
757 }
758 if (ret > 0)
759 break;
760 eb = path1->nodes[0];
761 }
762
763 btrfs_item_key_to_cpu(eb, &key, path1->slots[0]);
764 if (key.objectid != cur->bytenr) {
765 WARN_ON(exist);
766 break;
767 }
768
769 if (key.type == BTRFS_EXTENT_ITEM_KEY) {
770 ret = find_inline_backref(eb, path1->slots[0],
771 &ptr, &end);
772 if (ret)
773 goto next;
774 }
775 }
776
777 if (ptr < end) {
778 /* update key for inline back ref */
779 struct btrfs_extent_inline_ref *iref;
780 iref = (struct btrfs_extent_inline_ref *)ptr;
781 key.type = btrfs_extent_inline_ref_type(eb, iref);
782 key.offset = btrfs_extent_inline_ref_offset(eb, iref);
783 WARN_ON(key.type != BTRFS_TREE_BLOCK_REF_KEY &&
784 key.type != BTRFS_SHARED_BLOCK_REF_KEY);
785 }
786
787 if (exist &&
788 ((key.type == BTRFS_TREE_BLOCK_REF_KEY &&
789 exist->owner == key.offset) ||
790 (key.type == BTRFS_SHARED_BLOCK_REF_KEY &&
791 exist->bytenr == key.offset))) {
792 exist = NULL;
793 goto next;
794 }
795
796#ifdef BTRFS_COMPAT_EXTENT_TREE_V0
797 if (key.type == BTRFS_SHARED_BLOCK_REF_KEY ||
798 key.type == BTRFS_EXTENT_REF_V0_KEY) {
799 if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
800 struct btrfs_extent_ref_v0 *ref0;
801 ref0 = btrfs_item_ptr(eb, path1->slots[0],
802 struct btrfs_extent_ref_v0);
803 if (key.objectid == key.offset) {
804 root = find_tree_root(rc, eb, ref0);
805 if (root && !should_ignore_root(root))
806 cur->root = root;
807 else
808 list_add(&cur->list, &useless);
809 break;
810 }
811 if (is_cowonly_root(btrfs_ref_root_v0(eb,
812 ref0)))
813 cur->cowonly = 1;
814 }
815#else
816 BUG_ON(key.type == BTRFS_EXTENT_REF_V0_KEY);
817 if (key.type == BTRFS_SHARED_BLOCK_REF_KEY) {
818#endif
819 if (key.objectid == key.offset) {
820 /*
821 * only root blocks of reloc trees use
822 * backref of this type.
823 */
824 root = find_reloc_root(rc, cur->bytenr);
825 BUG_ON(!root);
826 cur->root = root;
827 break;
828 }
829
830 edge = alloc_backref_edge(cache);
831 if (!edge) {
832 err = -ENOMEM;
833 goto out;
834 }
835 rb_node = tree_search(&cache->rb_root, key.offset);
836 if (!rb_node) {
837 upper = alloc_backref_node(cache);
838 if (!upper) {
839 free_backref_edge(cache, edge);
840 err = -ENOMEM;
841 goto out;
842 }
843 upper->bytenr = key.offset;
844 upper->level = cur->level + 1;
845 /*
846 * backrefs for the upper level block isn't
847 * cached, add the block to pending list
848 */
849 list_add_tail(&edge->list[UPPER], &list);
850 } else {
851 upper = rb_entry(rb_node, struct backref_node,
852 rb_node);
853 BUG_ON(!upper->checked);
854 INIT_LIST_HEAD(&edge->list[UPPER]);
855 }
856 list_add_tail(&edge->list[LOWER], &cur->upper);
857 edge->node[LOWER] = cur;
858 edge->node[UPPER] = upper;
859
860 goto next;
861 } else if (key.type != BTRFS_TREE_BLOCK_REF_KEY) {
862 goto next;
863 }
864
865 /* key.type == BTRFS_TREE_BLOCK_REF_KEY */
866 root = read_fs_root(rc->extent_root->fs_info, key.offset);
867 if (IS_ERR(root)) {
868 err = PTR_ERR(root);
869 goto out;
870 }
871
872 if (!root->ref_cows)
873 cur->cowonly = 1;
874
875 if (btrfs_root_level(&root->root_item) == cur->level) {
876 /* tree root */
877 BUG_ON(btrfs_root_bytenr(&root->root_item) !=
878 cur->bytenr);
879 if (should_ignore_root(root))
880 list_add(&cur->list, &useless);
881 else
882 cur->root = root;
883 break;
884 }
885
886 level = cur->level + 1;
887
888 /*
889 * searching the tree to find upper level blocks
890 * reference the block.
891 */
892 path2->search_commit_root = 1;
893 path2->skip_locking = 1;
894 path2->lowest_level = level;
895 ret = btrfs_search_slot(NULL, root, node_key, path2, 0, 0);
896 path2->lowest_level = 0;
897 if (ret < 0) {
898 err = ret;
899 goto out;
900 }
901 if (ret > 0 && path2->slots[level] > 0)
902 path2->slots[level]--;
903
904 eb = path2->nodes[level];
905 WARN_ON(btrfs_node_blockptr(eb, path2->slots[level]) !=
906 cur->bytenr);
907
908 lower = cur;
909 for (; level < BTRFS_MAX_LEVEL; level++) {
910 if (!path2->nodes[level]) {
911 BUG_ON(btrfs_root_bytenr(&root->root_item) !=
912 lower->bytenr);
913 if (should_ignore_root(root))
914 list_add(&lower->list, &useless);
915 else
916 lower->root = root;
917 break;
918 }
919
920 edge = alloc_backref_edge(cache);
921 if (!edge) {
922 err = -ENOMEM;
923 goto out;
924 }
925
926 eb = path2->nodes[level];
927 rb_node = tree_search(&cache->rb_root, eb->start);
928 if (!rb_node) {
929 upper = alloc_backref_node(cache);
930 if (!upper) {
931 free_backref_edge(cache, edge);
932 err = -ENOMEM;
933 goto out;
934 }
935 upper->bytenr = eb->start;
936 upper->owner = btrfs_header_owner(eb);
937 upper->level = lower->level + 1;
938 if (!root->ref_cows)
939 upper->cowonly = 1;
940
941 /*
942 * if we know the block isn't shared
943 * we can void checking its backrefs.
944 */
945 if (btrfs_block_can_be_shared(root, eb))
946 upper->checked = 0;
947 else
948 upper->checked = 1;
949
950 /*
951 * add the block to pending list if we
952 * need check its backrefs. only block
953 * at 'cur->level + 1' is added to the
954 * tail of pending list. this guarantees
955 * we check backrefs from lower level
956 * blocks to upper level blocks.
957 */
958 if (!upper->checked &&
959 level == cur->level + 1) {
960 list_add_tail(&edge->list[UPPER],
961 &list);
962 } else
963 INIT_LIST_HEAD(&edge->list[UPPER]);
964 } else {
965 upper = rb_entry(rb_node, struct backref_node,
966 rb_node);
967 BUG_ON(!upper->checked);
968 INIT_LIST_HEAD(&edge->list[UPPER]);
969 if (!upper->owner)
970 upper->owner = btrfs_header_owner(eb);
971 }
972 list_add_tail(&edge->list[LOWER], &lower->upper);
973 edge->node[LOWER] = lower;
974 edge->node[UPPER] = upper;
975
976 if (rb_node)
977 break;
978 lower = upper;
979 upper = NULL;
980 }
981 btrfs_release_path(path2);
982next:
983 if (ptr < end) {
984 ptr += btrfs_extent_inline_ref_size(key.type);
985 if (ptr >= end) {
986 WARN_ON(ptr > end);
987 ptr = 0;
988 end = 0;
989 }
990 }
991 if (ptr >= end)
992 path1->slots[0]++;
993 }
994 btrfs_release_path(path1);
995
996 cur->checked = 1;
997 WARN_ON(exist);
998
999 /* the pending list isn't empty, take the first block to process */
1000 if (!list_empty(&list)) {
1001 edge = list_entry(list.next, struct backref_edge, list[UPPER]);
1002 list_del_init(&edge->list[UPPER]);
1003 cur = edge->node[UPPER];
1004 goto again;
1005 }
1006
1007 /*
1008 * everything goes well, connect backref nodes and insert backref nodes
1009 * into the cache.
1010 */
1011 BUG_ON(!node->checked);
1012 cowonly = node->cowonly;
1013 if (!cowonly) {
1014 rb_node = tree_insert(&cache->rb_root, node->bytenr,
1015 &node->rb_node);
1016 if (rb_node)
1017 backref_tree_panic(rb_node, -EEXIST, node->bytenr);
1018 list_add_tail(&node->lower, &cache->leaves);
1019 }
1020
1021 list_for_each_entry(edge, &node->upper, list[LOWER])
1022 list_add_tail(&edge->list[UPPER], &list);
1023
1024 while (!list_empty(&list)) {
1025 edge = list_entry(list.next, struct backref_edge, list[UPPER]);
1026 list_del_init(&edge->list[UPPER]);
1027 upper = edge->node[UPPER];
1028 if (upper->detached) {
1029 list_del(&edge->list[LOWER]);
1030 lower = edge->node[LOWER];
1031 free_backref_edge(cache, edge);
1032 if (list_empty(&lower->upper))
1033 list_add(&lower->list, &useless);
1034 continue;
1035 }
1036
1037 if (!RB_EMPTY_NODE(&upper->rb_node)) {
1038 if (upper->lowest) {
1039 list_del_init(&upper->lower);
1040 upper->lowest = 0;
1041 }
1042
1043 list_add_tail(&edge->list[UPPER], &upper->lower);
1044 continue;
1045 }
1046
1047 BUG_ON(!upper->checked);
1048 BUG_ON(cowonly != upper->cowonly);
1049 if (!cowonly) {
1050 rb_node = tree_insert(&cache->rb_root, upper->bytenr,
1051 &upper->rb_node);
1052 if (rb_node)
1053 backref_tree_panic(rb_node, -EEXIST,
1054 upper->bytenr);
1055 }
1056
1057 list_add_tail(&edge->list[UPPER], &upper->lower);
1058
1059 list_for_each_entry(edge, &upper->upper, list[LOWER])
1060 list_add_tail(&edge->list[UPPER], &list);
1061 }
1062 /*
1063 * process useless backref nodes. backref nodes for tree leaves
1064 * are deleted from the cache. backref nodes for upper level
1065 * tree blocks are left in the cache to avoid unnecessary backref
1066 * lookup.
1067 */
1068 while (!list_empty(&useless)) {
1069 upper = list_entry(useless.next, struct backref_node, list);
1070 list_del_init(&upper->list);
1071 BUG_ON(!list_empty(&upper->upper));
1072 if (upper == node)
1073 node = NULL;
1074 if (upper->lowest) {
1075 list_del_init(&upper->lower);
1076 upper->lowest = 0;
1077 }
1078 while (!list_empty(&upper->lower)) {
1079 edge = list_entry(upper->lower.next,
1080 struct backref_edge, list[UPPER]);
1081 list_del(&edge->list[UPPER]);
1082 list_del(&edge->list[LOWER]);
1083 lower = edge->node[LOWER];
1084 free_backref_edge(cache, edge);
1085
1086 if (list_empty(&lower->upper))
1087 list_add(&lower->list, &useless);
1088 }
1089 __mark_block_processed(rc, upper);
1090 if (upper->level > 0) {
1091 list_add(&upper->list, &cache->detached);
1092 upper->detached = 1;
1093 } else {
1094 rb_erase(&upper->rb_node, &cache->rb_root);
1095 free_backref_node(cache, upper);
1096 }
1097 }
1098out:
1099 btrfs_free_path(path1);
1100 btrfs_free_path(path2);
1101 if (err) {
1102 while (!list_empty(&useless)) {
1103 lower = list_entry(useless.next,
1104 struct backref_node, upper);
1105 list_del_init(&lower->upper);
1106 }
1107 upper = node;
1108 INIT_LIST_HEAD(&list);
1109 while (upper) {
1110 if (RB_EMPTY_NODE(&upper->rb_node)) {
1111 list_splice_tail(&upper->upper, &list);
1112 free_backref_node(cache, upper);
1113 }
1114
1115 if (list_empty(&list))
1116 break;
1117
1118 edge = list_entry(list.next, struct backref_edge,
1119 list[LOWER]);
1120 list_del(&edge->list[LOWER]);
1121 upper = edge->node[UPPER];
1122 free_backref_edge(cache, edge);
1123 }
1124 return ERR_PTR(err);
1125 }
1126 BUG_ON(node && node->detached);
1127 return node;
1128}
1129
1130/*
1131 * helper to add backref node for the newly created snapshot.
1132 * the backref node is created by cloning backref node that
1133 * corresponds to root of source tree
1134 */
1135static int clone_backref_node(struct btrfs_trans_handle *trans,
1136 struct reloc_control *rc,
1137 struct btrfs_root *src,
1138 struct btrfs_root *dest)
1139{
1140 struct btrfs_root *reloc_root = src->reloc_root;
1141 struct backref_cache *cache = &rc->backref_cache;
1142 struct backref_node *node = NULL;
1143 struct backref_node *new_node;
1144 struct backref_edge *edge;
1145 struct backref_edge *new_edge;
1146 struct rb_node *rb_node;
1147
1148 if (cache->last_trans > 0)
1149 update_backref_cache(trans, cache);
1150
1151 rb_node = tree_search(&cache->rb_root, src->commit_root->start);
1152 if (rb_node) {
1153 node = rb_entry(rb_node, struct backref_node, rb_node);
1154 if (node->detached)
1155 node = NULL;
1156 else
1157 BUG_ON(node->new_bytenr != reloc_root->node->start);
1158 }
1159
1160 if (!node) {
1161 rb_node = tree_search(&cache->rb_root,
1162 reloc_root->commit_root->start);
1163 if (rb_node) {
1164 node = rb_entry(rb_node, struct backref_node,
1165 rb_node);
1166 BUG_ON(node->detached);
1167 }
1168 }
1169
1170 if (!node)
1171 return 0;
1172
1173 new_node = alloc_backref_node(cache);
1174 if (!new_node)
1175 return -ENOMEM;
1176
1177 new_node->bytenr = dest->node->start;
1178 new_node->level = node->level;
1179 new_node->lowest = node->lowest;
1180 new_node->checked = 1;
1181 new_node->root = dest;
1182
1183 if (!node->lowest) {
1184 list_for_each_entry(edge, &node->lower, list[UPPER]) {
1185 new_edge = alloc_backref_edge(cache);
1186 if (!new_edge)
1187 goto fail;
1188
1189 new_edge->node[UPPER] = new_node;
1190 new_edge->node[LOWER] = edge->node[LOWER];
1191 list_add_tail(&new_edge->list[UPPER],
1192 &new_node->lower);
1193 }
1194 } else {
1195 list_add_tail(&new_node->lower, &cache->leaves);
1196 }
1197
1198 rb_node = tree_insert(&cache->rb_root, new_node->bytenr,
1199 &new_node->rb_node);
1200 if (rb_node)
1201 backref_tree_panic(rb_node, -EEXIST, new_node->bytenr);
1202
1203 if (!new_node->lowest) {
1204 list_for_each_entry(new_edge, &new_node->lower, list[UPPER]) {
1205 list_add_tail(&new_edge->list[LOWER],
1206 &new_edge->node[LOWER]->upper);
1207 }
1208 }
1209 return 0;
1210fail:
1211 while (!list_empty(&new_node->lower)) {
1212 new_edge = list_entry(new_node->lower.next,
1213 struct backref_edge, list[UPPER]);
1214 list_del(&new_edge->list[UPPER]);
1215 free_backref_edge(cache, new_edge);
1216 }
1217 free_backref_node(cache, new_node);
1218 return -ENOMEM;
1219}
1220
1221/*
1222 * helper to add 'address of tree root -> reloc tree' mapping
1223 */
1224static int __must_check __add_reloc_root(struct btrfs_root *root)
1225{
1226 struct rb_node *rb_node;
1227 struct mapping_node *node;
1228 struct reloc_control *rc = root->fs_info->reloc_ctl;
1229
1230 node = kmalloc(sizeof(*node), GFP_NOFS);
1231 if (!node)
1232 return -ENOMEM;
1233
1234 node->bytenr = root->node->start;
1235 node->data = root;
1236
1237 spin_lock(&rc->reloc_root_tree.lock);
1238 rb_node = tree_insert(&rc->reloc_root_tree.rb_root,
1239 node->bytenr, &node->rb_node);
1240 spin_unlock(&rc->reloc_root_tree.lock);
1241 if (rb_node) {
1242 kfree(node);
1243 btrfs_panic(root->fs_info, -EEXIST, "Duplicate root found "
1244 "for start=%llu while inserting into relocation "
1245 "tree\n");
1246 }
1247
1248 list_add_tail(&root->root_list, &rc->reloc_roots);
1249 return 0;
1250}
1251
1252/*
1253 * helper to update/delete the 'address of tree root -> reloc tree'
1254 * mapping
1255 */
1256static int __update_reloc_root(struct btrfs_root *root, int del)
1257{
1258 struct rb_node *rb_node;
1259 struct mapping_node *node = NULL;
1260 struct reloc_control *rc = root->fs_info->reloc_ctl;
1261
1262 spin_lock(&rc->reloc_root_tree.lock);
1263 rb_node = tree_search(&rc->reloc_root_tree.rb_root,
1264 root->commit_root->start);
1265 if (rb_node) {
1266 node = rb_entry(rb_node, struct mapping_node, rb_node);
1267 rb_erase(&node->rb_node, &rc->reloc_root_tree.rb_root);
1268 }
1269 spin_unlock(&rc->reloc_root_tree.lock);
1270
1271 BUG_ON((struct btrfs_root *)node->data != root);
1272
1273 if (!del) {
1274 spin_lock(&rc->reloc_root_tree.lock);
1275 node->bytenr = root->node->start;
1276 rb_node = tree_insert(&rc->reloc_root_tree.rb_root,
1277 node->bytenr, &node->rb_node);
1278 spin_unlock(&rc->reloc_root_tree.lock);
1279 if (rb_node)
1280 backref_tree_panic(rb_node, -EEXIST, node->bytenr);
1281 } else {
1282 spin_lock(&root->fs_info->trans_lock);
1283 list_del_init(&root->root_list);
1284 spin_unlock(&root->fs_info->trans_lock);
1285 kfree(node);
1286 }
1287 return 0;
1288}
1289
1290static struct btrfs_root *create_reloc_root(struct btrfs_trans_handle *trans,
1291 struct btrfs_root *root, u64 objectid)
1292{
1293 struct btrfs_root *reloc_root;
1294 struct extent_buffer *eb;
1295 struct btrfs_root_item *root_item;
1296 struct btrfs_key root_key;
1297 int ret;
1298
1299 root_item = kmalloc(sizeof(*root_item), GFP_NOFS);
1300 BUG_ON(!root_item);
1301
1302 root_key.objectid = BTRFS_TREE_RELOC_OBJECTID;
1303 root_key.type = BTRFS_ROOT_ITEM_KEY;
1304 root_key.offset = objectid;
1305
1306 if (root->root_key.objectid == objectid) {
1307 /* called by btrfs_init_reloc_root */
1308 ret = btrfs_copy_root(trans, root, root->commit_root, &eb,
1309 BTRFS_TREE_RELOC_OBJECTID);
1310 BUG_ON(ret);
1311
1312 btrfs_set_root_last_snapshot(&root->root_item,
1313 trans->transid - 1);
1314 } else {
1315 /*
1316 * called by btrfs_reloc_post_snapshot_hook.
1317 * the source tree is a reloc tree, all tree blocks
1318 * modified after it was created have RELOC flag
1319 * set in their headers. so it's OK to not update
1320 * the 'last_snapshot'.
1321 */
1322 ret = btrfs_copy_root(trans, root, root->node, &eb,
1323 BTRFS_TREE_RELOC_OBJECTID);
1324 BUG_ON(ret);
1325 }
1326
1327 memcpy(root_item, &root->root_item, sizeof(*root_item));
1328 btrfs_set_root_bytenr(root_item, eb->start);
1329 btrfs_set_root_level(root_item, btrfs_header_level(eb));
1330 btrfs_set_root_generation(root_item, trans->transid);
1331
1332 if (root->root_key.objectid == objectid) {
1333 btrfs_set_root_refs(root_item, 0);
1334 memset(&root_item->drop_progress, 0,
1335 sizeof(struct btrfs_disk_key));
1336 root_item->drop_level = 0;
1337 }
1338
1339 btrfs_tree_unlock(eb);
1340 free_extent_buffer(eb);
1341
1342 ret = btrfs_insert_root(trans, root->fs_info->tree_root,
1343 &root_key, root_item);
1344 BUG_ON(ret);
1345 kfree(root_item);
1346
1347 reloc_root = btrfs_read_fs_root_no_radix(root->fs_info->tree_root,
1348 &root_key);
1349 BUG_ON(IS_ERR(reloc_root));
1350 reloc_root->last_trans = trans->transid;
1351 return reloc_root;
1352}
1353
1354/*
1355 * create reloc tree for a given fs tree. reloc tree is just a
1356 * snapshot of the fs tree with special root objectid.
1357 */
1358int btrfs_init_reloc_root(struct btrfs_trans_handle *trans,
1359 struct btrfs_root *root)
1360{
1361 struct btrfs_root *reloc_root;
1362 struct reloc_control *rc = root->fs_info->reloc_ctl;
1363 int clear_rsv = 0;
1364 int ret;
1365
1366 if (root->reloc_root) {
1367 reloc_root = root->reloc_root;
1368 reloc_root->last_trans = trans->transid;
1369 return 0;
1370 }
1371
1372 if (!rc || !rc->create_reloc_tree ||
1373 root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID)
1374 return 0;
1375
1376 if (!trans->block_rsv) {
1377 trans->block_rsv = rc->block_rsv;
1378 clear_rsv = 1;
1379 }
1380 reloc_root = create_reloc_root(trans, root, root->root_key.objectid);
1381 if (clear_rsv)
1382 trans->block_rsv = NULL;
1383
1384 ret = __add_reloc_root(reloc_root);
1385 BUG_ON(ret < 0);
1386 root->reloc_root = reloc_root;
1387 return 0;
1388}
1389
1390/*
1391 * update root item of reloc tree
1392 */
1393int btrfs_update_reloc_root(struct btrfs_trans_handle *trans,
1394 struct btrfs_root *root)
1395{
1396 struct btrfs_root *reloc_root;
1397 struct btrfs_root_item *root_item;
1398 int del = 0;
1399 int ret;
1400
1401 if (!root->reloc_root)
1402 goto out;
1403
1404 reloc_root = root->reloc_root;
1405 root_item = &reloc_root->root_item;
1406
1407 if (root->fs_info->reloc_ctl->merge_reloc_tree &&
1408 btrfs_root_refs(root_item) == 0) {
1409 root->reloc_root = NULL;
1410 del = 1;
1411 }
1412
1413 __update_reloc_root(reloc_root, del);
1414
1415 if (reloc_root->commit_root != reloc_root->node) {
1416 btrfs_set_root_node(root_item, reloc_root->node);
1417 free_extent_buffer(reloc_root->commit_root);
1418 reloc_root->commit_root = btrfs_root_node(reloc_root);
1419 }
1420
1421 ret = btrfs_update_root(trans, root->fs_info->tree_root,
1422 &reloc_root->root_key, root_item);
1423 BUG_ON(ret);
1424
1425out:
1426 return 0;
1427}
1428
1429/*
1430 * helper to find first cached inode with inode number >= objectid
1431 * in a subvolume
1432 */
1433static struct inode *find_next_inode(struct btrfs_root *root, u64 objectid)
1434{
1435 struct rb_node *node;
1436 struct rb_node *prev;
1437 struct btrfs_inode *entry;
1438 struct inode *inode;
1439
1440 spin_lock(&root->inode_lock);
1441again:
1442 node = root->inode_tree.rb_node;
1443 prev = NULL;
1444 while (node) {
1445 prev = node;
1446 entry = rb_entry(node, struct btrfs_inode, rb_node);
1447
1448 if (objectid < btrfs_ino(&entry->vfs_inode))
1449 node = node->rb_left;
1450 else if (objectid > btrfs_ino(&entry->vfs_inode))
1451 node = node->rb_right;
1452 else
1453 break;
1454 }
1455 if (!node) {
1456 while (prev) {
1457 entry = rb_entry(prev, struct btrfs_inode, rb_node);
1458 if (objectid <= btrfs_ino(&entry->vfs_inode)) {
1459 node = prev;
1460 break;
1461 }
1462 prev = rb_next(prev);
1463 }
1464 }
1465 while (node) {
1466 entry = rb_entry(node, struct btrfs_inode, rb_node);
1467 inode = igrab(&entry->vfs_inode);
1468 if (inode) {
1469 spin_unlock(&root->inode_lock);
1470 return inode;
1471 }
1472
1473 objectid = btrfs_ino(&entry->vfs_inode) + 1;
1474 if (cond_resched_lock(&root->inode_lock))
1475 goto again;
1476
1477 node = rb_next(node);
1478 }
1479 spin_unlock(&root->inode_lock);
1480 return NULL;
1481}
1482
1483static int in_block_group(u64 bytenr,
1484 struct btrfs_block_group_cache *block_group)
1485{
1486 if (bytenr >= block_group->key.objectid &&
1487 bytenr < block_group->key.objectid + block_group->key.offset)
1488 return 1;
1489 return 0;
1490}
1491
1492/*
1493 * get new location of data
1494 */
1495static int get_new_location(struct inode *reloc_inode, u64 *new_bytenr,
1496 u64 bytenr, u64 num_bytes)
1497{
1498 struct btrfs_root *root = BTRFS_I(reloc_inode)->root;
1499 struct btrfs_path *path;
1500 struct btrfs_file_extent_item *fi;
1501 struct extent_buffer *leaf;
1502 int ret;
1503
1504 path = btrfs_alloc_path();
1505 if (!path)
1506 return -ENOMEM;
1507
1508 bytenr -= BTRFS_I(reloc_inode)->index_cnt;
1509 ret = btrfs_lookup_file_extent(NULL, root, path, btrfs_ino(reloc_inode),
1510 bytenr, 0);
1511 if (ret < 0)
1512 goto out;
1513 if (ret > 0) {
1514 ret = -ENOENT;
1515 goto out;
1516 }
1517
1518 leaf = path->nodes[0];
1519 fi = btrfs_item_ptr(leaf, path->slots[0],
1520 struct btrfs_file_extent_item);
1521
1522 BUG_ON(btrfs_file_extent_offset(leaf, fi) ||
1523 btrfs_file_extent_compression(leaf, fi) ||
1524 btrfs_file_extent_encryption(leaf, fi) ||
1525 btrfs_file_extent_other_encoding(leaf, fi));
1526
1527 if (num_bytes != btrfs_file_extent_disk_num_bytes(leaf, fi)) {
1528 ret = 1;
1529 goto out;
1530 }
1531
1532 *new_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1533 ret = 0;
1534out:
1535 btrfs_free_path(path);
1536 return ret;
1537}
1538
1539/*
1540 * update file extent items in the tree leaf to point to
1541 * the new locations.
1542 */
1543static noinline_for_stack
1544int replace_file_extents(struct btrfs_trans_handle *trans,
1545 struct reloc_control *rc,
1546 struct btrfs_root *root,
1547 struct extent_buffer *leaf)
1548{
1549 struct btrfs_key key;
1550 struct btrfs_file_extent_item *fi;
1551 struct inode *inode = NULL;
1552 u64 parent;
1553 u64 bytenr;
1554 u64 new_bytenr = 0;
1555 u64 num_bytes;
1556 u64 end;
1557 u32 nritems;
1558 u32 i;
1559 int ret;
1560 int first = 1;
1561 int dirty = 0;
1562
1563 if (rc->stage != UPDATE_DATA_PTRS)
1564 return 0;
1565
1566 /* reloc trees always use full backref */
1567 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID)
1568 parent = leaf->start;
1569 else
1570 parent = 0;
1571
1572 nritems = btrfs_header_nritems(leaf);
1573 for (i = 0; i < nritems; i++) {
1574 cond_resched();
1575 btrfs_item_key_to_cpu(leaf, &key, i);
1576 if (key.type != BTRFS_EXTENT_DATA_KEY)
1577 continue;
1578 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
1579 if (btrfs_file_extent_type(leaf, fi) ==
1580 BTRFS_FILE_EXTENT_INLINE)
1581 continue;
1582 bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1583 num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
1584 if (bytenr == 0)
1585 continue;
1586 if (!in_block_group(bytenr, rc->block_group))
1587 continue;
1588
1589 /*
1590 * if we are modifying block in fs tree, wait for readpage
1591 * to complete and drop the extent cache
1592 */
1593 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
1594 if (first) {
1595 inode = find_next_inode(root, key.objectid);
1596 first = 0;
1597 } else if (inode && btrfs_ino(inode) < key.objectid) {
1598 btrfs_add_delayed_iput(inode);
1599 inode = find_next_inode(root, key.objectid);
1600 }
1601 if (inode && btrfs_ino(inode) == key.objectid) {
1602 end = key.offset +
1603 btrfs_file_extent_num_bytes(leaf, fi);
1604 WARN_ON(!IS_ALIGNED(key.offset,
1605 root->sectorsize));
1606 WARN_ON(!IS_ALIGNED(end, root->sectorsize));
1607 end--;
1608 ret = try_lock_extent(&BTRFS_I(inode)->io_tree,
1609 key.offset, end);
1610 if (!ret)
1611 continue;
1612
1613 btrfs_drop_extent_cache(inode, key.offset, end,
1614 1);
1615 unlock_extent(&BTRFS_I(inode)->io_tree,
1616 key.offset, end);
1617 }
1618 }
1619
1620 ret = get_new_location(rc->data_inode, &new_bytenr,
1621 bytenr, num_bytes);
1622 if (ret > 0) {
1623 WARN_ON(1);
1624 continue;
1625 }
1626 BUG_ON(ret < 0);
1627
1628 btrfs_set_file_extent_disk_bytenr(leaf, fi, new_bytenr);
1629 dirty = 1;
1630
1631 key.offset -= btrfs_file_extent_offset(leaf, fi);
1632 ret = btrfs_inc_extent_ref(trans, root, new_bytenr,
1633 num_bytes, parent,
1634 btrfs_header_owner(leaf),
1635 key.objectid, key.offset, 1);
1636 BUG_ON(ret);
1637
1638 ret = btrfs_free_extent(trans, root, bytenr, num_bytes,
1639 parent, btrfs_header_owner(leaf),
1640 key.objectid, key.offset, 1);
1641 BUG_ON(ret);
1642 }
1643 if (dirty)
1644 btrfs_mark_buffer_dirty(leaf);
1645 if (inode)
1646 btrfs_add_delayed_iput(inode);
1647 return 0;
1648}
1649
1650static noinline_for_stack
1651int memcmp_node_keys(struct extent_buffer *eb, int slot,
1652 struct btrfs_path *path, int level)
1653{
1654 struct btrfs_disk_key key1;
1655 struct btrfs_disk_key key2;
1656 btrfs_node_key(eb, &key1, slot);
1657 btrfs_node_key(path->nodes[level], &key2, path->slots[level]);
1658 return memcmp(&key1, &key2, sizeof(key1));
1659}
1660
1661/*
1662 * try to replace tree blocks in fs tree with the new blocks
1663 * in reloc tree. tree blocks haven't been modified since the
1664 * reloc tree was create can be replaced.
1665 *
1666 * if a block was replaced, level of the block + 1 is returned.
1667 * if no block got replaced, 0 is returned. if there are other
1668 * errors, a negative error number is returned.
1669 */
1670static noinline_for_stack
1671int replace_path(struct btrfs_trans_handle *trans,
1672 struct btrfs_root *dest, struct btrfs_root *src,
1673 struct btrfs_path *path, struct btrfs_key *next_key,
1674 int lowest_level, int max_level)
1675{
1676 struct extent_buffer *eb;
1677 struct extent_buffer *parent;
1678 struct btrfs_key key;
1679 u64 old_bytenr;
1680 u64 new_bytenr;
1681 u64 old_ptr_gen;
1682 u64 new_ptr_gen;
1683 u64 last_snapshot;
1684 u32 blocksize;
1685 int cow = 0;
1686 int level;
1687 int ret;
1688 int slot;
1689
1690 BUG_ON(src->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
1691 BUG_ON(dest->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID);
1692
1693 last_snapshot = btrfs_root_last_snapshot(&src->root_item);
1694again:
1695 slot = path->slots[lowest_level];
1696 btrfs_node_key_to_cpu(path->nodes[lowest_level], &key, slot);
1697
1698 eb = btrfs_lock_root_node(dest);
1699 btrfs_set_lock_blocking(eb);
1700 level = btrfs_header_level(eb);
1701
1702 if (level < lowest_level) {
1703 btrfs_tree_unlock(eb);
1704 free_extent_buffer(eb);
1705 return 0;
1706 }
1707
1708 if (cow) {
1709 ret = btrfs_cow_block(trans, dest, eb, NULL, 0, &eb);
1710 BUG_ON(ret);
1711 }
1712 btrfs_set_lock_blocking(eb);
1713
1714 if (next_key) {
1715 next_key->objectid = (u64)-1;
1716 next_key->type = (u8)-1;
1717 next_key->offset = (u64)-1;
1718 }
1719
1720 parent = eb;
1721 while (1) {
1722 level = btrfs_header_level(parent);
1723 BUG_ON(level < lowest_level);
1724
1725 ret = btrfs_bin_search(parent, &key, level, &slot);
1726 if (ret && slot > 0)
1727 slot--;
1728
1729 if (next_key && slot + 1 < btrfs_header_nritems(parent))
1730 btrfs_node_key_to_cpu(parent, next_key, slot + 1);
1731
1732 old_bytenr = btrfs_node_blockptr(parent, slot);
1733 blocksize = btrfs_level_size(dest, level - 1);
1734 old_ptr_gen = btrfs_node_ptr_generation(parent, slot);
1735
1736 if (level <= max_level) {
1737 eb = path->nodes[level];
1738 new_bytenr = btrfs_node_blockptr(eb,
1739 path->slots[level]);
1740 new_ptr_gen = btrfs_node_ptr_generation(eb,
1741 path->slots[level]);
1742 } else {
1743 new_bytenr = 0;
1744 new_ptr_gen = 0;
1745 }
1746
1747 if (new_bytenr > 0 && new_bytenr == old_bytenr) {
1748 WARN_ON(1);
1749 ret = level;
1750 break;
1751 }
1752
1753 if (new_bytenr == 0 || old_ptr_gen > last_snapshot ||
1754 memcmp_node_keys(parent, slot, path, level)) {
1755 if (level <= lowest_level) {
1756 ret = 0;
1757 break;
1758 }
1759
1760 eb = read_tree_block(dest, old_bytenr, blocksize,
1761 old_ptr_gen);
1762 BUG_ON(!eb);
1763 btrfs_tree_lock(eb);
1764 if (cow) {
1765 ret = btrfs_cow_block(trans, dest, eb, parent,
1766 slot, &eb);
1767 BUG_ON(ret);
1768 }
1769 btrfs_set_lock_blocking(eb);
1770
1771 btrfs_tree_unlock(parent);
1772 free_extent_buffer(parent);
1773
1774 parent = eb;
1775 continue;
1776 }
1777
1778 if (!cow) {
1779 btrfs_tree_unlock(parent);
1780 free_extent_buffer(parent);
1781 cow = 1;
1782 goto again;
1783 }
1784
1785 btrfs_node_key_to_cpu(path->nodes[level], &key,
1786 path->slots[level]);
1787 btrfs_release_path(path);
1788
1789 path->lowest_level = level;
1790 ret = btrfs_search_slot(trans, src, &key, path, 0, 1);
1791 path->lowest_level = 0;
1792 BUG_ON(ret);
1793
1794 /*
1795 * swap blocks in fs tree and reloc tree.
1796 */
1797 btrfs_set_node_blockptr(parent, slot, new_bytenr);
1798 btrfs_set_node_ptr_generation(parent, slot, new_ptr_gen);
1799 btrfs_mark_buffer_dirty(parent);
1800
1801 btrfs_set_node_blockptr(path->nodes[level],
1802 path->slots[level], old_bytenr);
1803 btrfs_set_node_ptr_generation(path->nodes[level],
1804 path->slots[level], old_ptr_gen);
1805 btrfs_mark_buffer_dirty(path->nodes[level]);
1806
1807 ret = btrfs_inc_extent_ref(trans, src, old_bytenr, blocksize,
1808 path->nodes[level]->start,
1809 src->root_key.objectid, level - 1, 0,
1810 1);
1811 BUG_ON(ret);
1812 ret = btrfs_inc_extent_ref(trans, dest, new_bytenr, blocksize,
1813 0, dest->root_key.objectid, level - 1,
1814 0, 1);
1815 BUG_ON(ret);
1816
1817 ret = btrfs_free_extent(trans, src, new_bytenr, blocksize,
1818 path->nodes[level]->start,
1819 src->root_key.objectid, level - 1, 0,
1820 1);
1821 BUG_ON(ret);
1822
1823 ret = btrfs_free_extent(trans, dest, old_bytenr, blocksize,
1824 0, dest->root_key.objectid, level - 1,
1825 0, 1);
1826 BUG_ON(ret);
1827
1828 btrfs_unlock_up_safe(path, 0);
1829
1830 ret = level;
1831 break;
1832 }
1833 btrfs_tree_unlock(parent);
1834 free_extent_buffer(parent);
1835 return ret;
1836}
1837
1838/*
1839 * helper to find next relocated block in reloc tree
1840 */
1841static noinline_for_stack
1842int walk_up_reloc_tree(struct btrfs_root *root, struct btrfs_path *path,
1843 int *level)
1844{
1845 struct extent_buffer *eb;
1846 int i;
1847 u64 last_snapshot;
1848 u32 nritems;
1849
1850 last_snapshot = btrfs_root_last_snapshot(&root->root_item);
1851
1852 for (i = 0; i < *level; i++) {
1853 free_extent_buffer(path->nodes[i]);
1854 path->nodes[i] = NULL;
1855 }
1856
1857 for (i = *level; i < BTRFS_MAX_LEVEL && path->nodes[i]; i++) {
1858 eb = path->nodes[i];
1859 nritems = btrfs_header_nritems(eb);
1860 while (path->slots[i] + 1 < nritems) {
1861 path->slots[i]++;
1862 if (btrfs_node_ptr_generation(eb, path->slots[i]) <=
1863 last_snapshot)
1864 continue;
1865
1866 *level = i;
1867 return 0;
1868 }
1869 free_extent_buffer(path->nodes[i]);
1870 path->nodes[i] = NULL;
1871 }
1872 return 1;
1873}
1874
1875/*
1876 * walk down reloc tree to find relocated block of lowest level
1877 */
1878static noinline_for_stack
1879int walk_down_reloc_tree(struct btrfs_root *root, struct btrfs_path *path,
1880 int *level)
1881{
1882 struct extent_buffer *eb = NULL;
1883 int i;
1884 u64 bytenr;
1885 u64 ptr_gen = 0;
1886 u64 last_snapshot;
1887 u32 blocksize;
1888 u32 nritems;
1889
1890 last_snapshot = btrfs_root_last_snapshot(&root->root_item);
1891
1892 for (i = *level; i > 0; i--) {
1893 eb = path->nodes[i];
1894 nritems = btrfs_header_nritems(eb);
1895 while (path->slots[i] < nritems) {
1896 ptr_gen = btrfs_node_ptr_generation(eb, path->slots[i]);
1897 if (ptr_gen > last_snapshot)
1898 break;
1899 path->slots[i]++;
1900 }
1901 if (path->slots[i] >= nritems) {
1902 if (i == *level)
1903 break;
1904 *level = i + 1;
1905 return 0;
1906 }
1907 if (i == 1) {
1908 *level = i;
1909 return 0;
1910 }
1911
1912 bytenr = btrfs_node_blockptr(eb, path->slots[i]);
1913 blocksize = btrfs_level_size(root, i - 1);
1914 eb = read_tree_block(root, bytenr, blocksize, ptr_gen);
1915 BUG_ON(btrfs_header_level(eb) != i - 1);
1916 path->nodes[i - 1] = eb;
1917 path->slots[i - 1] = 0;
1918 }
1919 return 1;
1920}
1921
1922/*
1923 * invalidate extent cache for file extents whose key in range of
1924 * [min_key, max_key)
1925 */
1926static int invalidate_extent_cache(struct btrfs_root *root,
1927 struct btrfs_key *min_key,
1928 struct btrfs_key *max_key)
1929{
1930 struct inode *inode = NULL;
1931 u64 objectid;
1932 u64 start, end;
1933 u64 ino;
1934
1935 objectid = min_key->objectid;
1936 while (1) {
1937 cond_resched();
1938 iput(inode);
1939
1940 if (objectid > max_key->objectid)
1941 break;
1942
1943 inode = find_next_inode(root, objectid);
1944 if (!inode)
1945 break;
1946 ino = btrfs_ino(inode);
1947
1948 if (ino > max_key->objectid) {
1949 iput(inode);
1950 break;
1951 }
1952
1953 objectid = ino + 1;
1954 if (!S_ISREG(inode->i_mode))
1955 continue;
1956
1957 if (unlikely(min_key->objectid == ino)) {
1958 if (min_key->type > BTRFS_EXTENT_DATA_KEY)
1959 continue;
1960 if (min_key->type < BTRFS_EXTENT_DATA_KEY)
1961 start = 0;
1962 else {
1963 start = min_key->offset;
1964 WARN_ON(!IS_ALIGNED(start, root->sectorsize));
1965 }
1966 } else {
1967 start = 0;
1968 }
1969
1970 if (unlikely(max_key->objectid == ino)) {
1971 if (max_key->type < BTRFS_EXTENT_DATA_KEY)
1972 continue;
1973 if (max_key->type > BTRFS_EXTENT_DATA_KEY) {
1974 end = (u64)-1;
1975 } else {
1976 if (max_key->offset == 0)
1977 continue;
1978 end = max_key->offset;
1979 WARN_ON(!IS_ALIGNED(end, root->sectorsize));
1980 end--;
1981 }
1982 } else {
1983 end = (u64)-1;
1984 }
1985
1986 /* the lock_extent waits for readpage to complete */
1987 lock_extent(&BTRFS_I(inode)->io_tree, start, end);
1988 btrfs_drop_extent_cache(inode, start, end, 1);
1989 unlock_extent(&BTRFS_I(inode)->io_tree, start, end);
1990 }
1991 return 0;
1992}
1993
1994static int find_next_key(struct btrfs_path *path, int level,
1995 struct btrfs_key *key)
1996
1997{
1998 while (level < BTRFS_MAX_LEVEL) {
1999 if (!path->nodes[level])
2000 break;
2001 if (path->slots[level] + 1 <
2002 btrfs_header_nritems(path->nodes[level])) {
2003 btrfs_node_key_to_cpu(path->nodes[level], key,
2004 path->slots[level] + 1);
2005 return 0;
2006 }
2007 level++;
2008 }
2009 return 1;
2010}
2011
2012/*
2013 * merge the relocated tree blocks in reloc tree with corresponding
2014 * fs tree.
2015 */
2016static noinline_for_stack int merge_reloc_root(struct reloc_control *rc,
2017 struct btrfs_root *root)
2018{
2019 LIST_HEAD(inode_list);
2020 struct btrfs_key key;
2021 struct btrfs_key next_key;
2022 struct btrfs_trans_handle *trans;
2023 struct btrfs_root *reloc_root;
2024 struct btrfs_root_item *root_item;
2025 struct btrfs_path *path;
2026 struct extent_buffer *leaf;
2027 unsigned long nr;
2028 int level;
2029 int max_level;
2030 int replaced = 0;
2031 int ret;
2032 int err = 0;
2033 u32 min_reserved;
2034
2035 path = btrfs_alloc_path();
2036 if (!path)
2037 return -ENOMEM;
2038 path->reada = 1;
2039
2040 reloc_root = root->reloc_root;
2041 root_item = &reloc_root->root_item;
2042
2043 if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
2044 level = btrfs_root_level(root_item);
2045 extent_buffer_get(reloc_root->node);
2046 path->nodes[level] = reloc_root->node;
2047 path->slots[level] = 0;
2048 } else {
2049 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
2050
2051 level = root_item->drop_level;
2052 BUG_ON(level == 0);
2053 path->lowest_level = level;
2054 ret = btrfs_search_slot(NULL, reloc_root, &key, path, 0, 0);
2055 path->lowest_level = 0;
2056 if (ret < 0) {
2057 btrfs_free_path(path);
2058 return ret;
2059 }
2060
2061 btrfs_node_key_to_cpu(path->nodes[level], &next_key,
2062 path->slots[level]);
2063 WARN_ON(memcmp(&key, &next_key, sizeof(key)));
2064
2065 btrfs_unlock_up_safe(path, 0);
2066 }
2067
2068 min_reserved = root->nodesize * (BTRFS_MAX_LEVEL - 1) * 2;
2069 memset(&next_key, 0, sizeof(next_key));
2070
2071 while (1) {
2072 trans = btrfs_start_transaction(root, 0);
2073 BUG_ON(IS_ERR(trans));
2074 trans->block_rsv = rc->block_rsv;
2075
2076 ret = btrfs_block_rsv_refill(root, rc->block_rsv, min_reserved);
2077 if (ret) {
2078 BUG_ON(ret != -EAGAIN);
2079 ret = btrfs_commit_transaction(trans, root);
2080 BUG_ON(ret);
2081 continue;
2082 }
2083
2084 replaced = 0;
2085 max_level = level;
2086
2087 ret = walk_down_reloc_tree(reloc_root, path, &level);
2088 if (ret < 0) {
2089 err = ret;
2090 goto out;
2091 }
2092 if (ret > 0)
2093 break;
2094
2095 if (!find_next_key(path, level, &key) &&
2096 btrfs_comp_cpu_keys(&next_key, &key) >= 0) {
2097 ret = 0;
2098 } else {
2099 ret = replace_path(trans, root, reloc_root, path,
2100 &next_key, level, max_level);
2101 }
2102 if (ret < 0) {
2103 err = ret;
2104 goto out;
2105 }
2106
2107 if (ret > 0) {
2108 level = ret;
2109 btrfs_node_key_to_cpu(path->nodes[level], &key,
2110 path->slots[level]);
2111 replaced = 1;
2112 }
2113
2114 ret = walk_up_reloc_tree(reloc_root, path, &level);
2115 if (ret > 0)
2116 break;
2117
2118 BUG_ON(level == 0);
2119 /*
2120 * save the merging progress in the drop_progress.
2121 * this is OK since root refs == 1 in this case.
2122 */
2123 btrfs_node_key(path->nodes[level], &root_item->drop_progress,
2124 path->slots[level]);
2125 root_item->drop_level = level;
2126
2127 nr = trans->blocks_used;
2128 btrfs_end_transaction_throttle(trans, root);
2129
2130 btrfs_btree_balance_dirty(root, nr);
2131
2132 if (replaced && rc->stage == UPDATE_DATA_PTRS)
2133 invalidate_extent_cache(root, &key, &next_key);
2134 }
2135
2136 /*
2137 * handle the case only one block in the fs tree need to be
2138 * relocated and the block is tree root.
2139 */
2140 leaf = btrfs_lock_root_node(root);
2141 ret = btrfs_cow_block(trans, root, leaf, NULL, 0, &leaf);
2142 btrfs_tree_unlock(leaf);
2143 free_extent_buffer(leaf);
2144 if (ret < 0)
2145 err = ret;
2146out:
2147 btrfs_free_path(path);
2148
2149 if (err == 0) {
2150 memset(&root_item->drop_progress, 0,
2151 sizeof(root_item->drop_progress));
2152 root_item->drop_level = 0;
2153 btrfs_set_root_refs(root_item, 0);
2154 btrfs_update_reloc_root(trans, root);
2155 }
2156
2157 nr = trans->blocks_used;
2158 btrfs_end_transaction_throttle(trans, root);
2159
2160 btrfs_btree_balance_dirty(root, nr);
2161
2162 if (replaced && rc->stage == UPDATE_DATA_PTRS)
2163 invalidate_extent_cache(root, &key, &next_key);
2164
2165 return err;
2166}
2167
2168static noinline_for_stack
2169int prepare_to_merge(struct reloc_control *rc, int err)
2170{
2171 struct btrfs_root *root = rc->extent_root;
2172 struct btrfs_root *reloc_root;
2173 struct btrfs_trans_handle *trans;
2174 LIST_HEAD(reloc_roots);
2175 u64 num_bytes = 0;
2176 int ret;
2177
2178 mutex_lock(&root->fs_info->reloc_mutex);
2179 rc->merging_rsv_size += root->nodesize * (BTRFS_MAX_LEVEL - 1) * 2;
2180 rc->merging_rsv_size += rc->nodes_relocated * 2;
2181 mutex_unlock(&root->fs_info->reloc_mutex);
2182
2183again:
2184 if (!err) {
2185 num_bytes = rc->merging_rsv_size;
2186 ret = btrfs_block_rsv_add(root, rc->block_rsv, num_bytes);
2187 if (ret)
2188 err = ret;
2189 }
2190
2191 trans = btrfs_join_transaction(rc->extent_root);
2192 if (IS_ERR(trans)) {
2193 if (!err)
2194 btrfs_block_rsv_release(rc->extent_root,
2195 rc->block_rsv, num_bytes);
2196 return PTR_ERR(trans);
2197 }
2198
2199 if (!err) {
2200 if (num_bytes != rc->merging_rsv_size) {
2201 btrfs_end_transaction(trans, rc->extent_root);
2202 btrfs_block_rsv_release(rc->extent_root,
2203 rc->block_rsv, num_bytes);
2204 goto again;
2205 }
2206 }
2207
2208 rc->merge_reloc_tree = 1;
2209
2210 while (!list_empty(&rc->reloc_roots)) {
2211 reloc_root = list_entry(rc->reloc_roots.next,
2212 struct btrfs_root, root_list);
2213 list_del_init(&reloc_root->root_list);
2214
2215 root = read_fs_root(reloc_root->fs_info,
2216 reloc_root->root_key.offset);
2217 BUG_ON(IS_ERR(root));
2218 BUG_ON(root->reloc_root != reloc_root);
2219
2220 /*
2221 * set reference count to 1, so btrfs_recover_relocation
2222 * knows it should resumes merging
2223 */
2224 if (!err)
2225 btrfs_set_root_refs(&reloc_root->root_item, 1);
2226 btrfs_update_reloc_root(trans, root);
2227
2228 list_add(&reloc_root->root_list, &reloc_roots);
2229 }
2230
2231 list_splice(&reloc_roots, &rc->reloc_roots);
2232
2233 if (!err)
2234 btrfs_commit_transaction(trans, rc->extent_root);
2235 else
2236 btrfs_end_transaction(trans, rc->extent_root);
2237 return err;
2238}
2239
2240static noinline_for_stack
2241int merge_reloc_roots(struct reloc_control *rc)
2242{
2243 struct btrfs_root *root;
2244 struct btrfs_root *reloc_root;
2245 LIST_HEAD(reloc_roots);
2246 int found = 0;
2247 int ret;
2248again:
2249 root = rc->extent_root;
2250
2251 /*
2252 * this serializes us with btrfs_record_root_in_transaction,
2253 * we have to make sure nobody is in the middle of
2254 * adding their roots to the list while we are
2255 * doing this splice
2256 */
2257 mutex_lock(&root->fs_info->reloc_mutex);
2258 list_splice_init(&rc->reloc_roots, &reloc_roots);
2259 mutex_unlock(&root->fs_info->reloc_mutex);
2260
2261 while (!list_empty(&reloc_roots)) {
2262 found = 1;
2263 reloc_root = list_entry(reloc_roots.next,
2264 struct btrfs_root, root_list);
2265
2266 if (btrfs_root_refs(&reloc_root->root_item) > 0) {
2267 root = read_fs_root(reloc_root->fs_info,
2268 reloc_root->root_key.offset);
2269 BUG_ON(IS_ERR(root));
2270 BUG_ON(root->reloc_root != reloc_root);
2271
2272 ret = merge_reloc_root(rc, root);
2273 BUG_ON(ret);
2274 } else {
2275 list_del_init(&reloc_root->root_list);
2276 }
2277 ret = btrfs_drop_snapshot(reloc_root, rc->block_rsv, 0, 1);
2278 BUG_ON(ret < 0);
2279 }
2280
2281 if (found) {
2282 found = 0;
2283 goto again;
2284 }
2285 BUG_ON(!RB_EMPTY_ROOT(&rc->reloc_root_tree.rb_root));
2286 return 0;
2287}
2288
2289static void free_block_list(struct rb_root *blocks)
2290{
2291 struct tree_block *block;
2292 struct rb_node *rb_node;
2293 while ((rb_node = rb_first(blocks))) {
2294 block = rb_entry(rb_node, struct tree_block, rb_node);
2295 rb_erase(rb_node, blocks);
2296 kfree(block);
2297 }
2298}
2299
2300static int record_reloc_root_in_trans(struct btrfs_trans_handle *trans,
2301 struct btrfs_root *reloc_root)
2302{
2303 struct btrfs_root *root;
2304
2305 if (reloc_root->last_trans == trans->transid)
2306 return 0;
2307
2308 root = read_fs_root(reloc_root->fs_info, reloc_root->root_key.offset);
2309 BUG_ON(IS_ERR(root));
2310 BUG_ON(root->reloc_root != reloc_root);
2311
2312 return btrfs_record_root_in_trans(trans, root);
2313}
2314
2315static noinline_for_stack
2316struct btrfs_root *select_reloc_root(struct btrfs_trans_handle *trans,
2317 struct reloc_control *rc,
2318 struct backref_node *node,
2319 struct backref_edge *edges[], int *nr)
2320{
2321 struct backref_node *next;
2322 struct btrfs_root *root;
2323 int index = 0;
2324
2325 next = node;
2326 while (1) {
2327 cond_resched();
2328 next = walk_up_backref(next, edges, &index);
2329 root = next->root;
2330 BUG_ON(!root);
2331 BUG_ON(!root->ref_cows);
2332
2333 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) {
2334 record_reloc_root_in_trans(trans, root);
2335 break;
2336 }
2337
2338 btrfs_record_root_in_trans(trans, root);
2339 root = root->reloc_root;
2340
2341 if (next->new_bytenr != root->node->start) {
2342 BUG_ON(next->new_bytenr);
2343 BUG_ON(!list_empty(&next->list));
2344 next->new_bytenr = root->node->start;
2345 next->root = root;
2346 list_add_tail(&next->list,
2347 &rc->backref_cache.changed);
2348 __mark_block_processed(rc, next);
2349 break;
2350 }
2351
2352 WARN_ON(1);
2353 root = NULL;
2354 next = walk_down_backref(edges, &index);
2355 if (!next || next->level <= node->level)
2356 break;
2357 }
2358 if (!root)
2359 return NULL;
2360
2361 *nr = index;
2362 next = node;
2363 /* setup backref node path for btrfs_reloc_cow_block */
2364 while (1) {
2365 rc->backref_cache.path[next->level] = next;
2366 if (--index < 0)
2367 break;
2368 next = edges[index]->node[UPPER];
2369 }
2370 return root;
2371}
2372
2373/*
2374 * select a tree root for relocation. return NULL if the block
2375 * is reference counted. we should use do_relocation() in this
2376 * case. return a tree root pointer if the block isn't reference
2377 * counted. return -ENOENT if the block is root of reloc tree.
2378 */
2379static noinline_for_stack
2380struct btrfs_root *select_one_root(struct btrfs_trans_handle *trans,
2381 struct backref_node *node)
2382{
2383 struct backref_node *next;
2384 struct btrfs_root *root;
2385 struct btrfs_root *fs_root = NULL;
2386 struct backref_edge *edges[BTRFS_MAX_LEVEL - 1];
2387 int index = 0;
2388
2389 next = node;
2390 while (1) {
2391 cond_resched();
2392 next = walk_up_backref(next, edges, &index);
2393 root = next->root;
2394 BUG_ON(!root);
2395
2396 /* no other choice for non-references counted tree */
2397 if (!root->ref_cows)
2398 return root;
2399
2400 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID)
2401 fs_root = root;
2402
2403 if (next != node)
2404 return NULL;
2405
2406 next = walk_down_backref(edges, &index);
2407 if (!next || next->level <= node->level)
2408 break;
2409 }
2410
2411 if (!fs_root)
2412 return ERR_PTR(-ENOENT);
2413 return fs_root;
2414}
2415
2416static noinline_for_stack
2417u64 calcu_metadata_size(struct reloc_control *rc,
2418 struct backref_node *node, int reserve)
2419{
2420 struct backref_node *next = node;
2421 struct backref_edge *edge;
2422 struct backref_edge *edges[BTRFS_MAX_LEVEL - 1];
2423 u64 num_bytes = 0;
2424 int index = 0;
2425
2426 BUG_ON(reserve && node->processed);
2427
2428 while (next) {
2429 cond_resched();
2430 while (1) {
2431 if (next->processed && (reserve || next != node))
2432 break;
2433
2434 num_bytes += btrfs_level_size(rc->extent_root,
2435 next->level);
2436
2437 if (list_empty(&next->upper))
2438 break;
2439
2440 edge = list_entry(next->upper.next,
2441 struct backref_edge, list[LOWER]);
2442 edges[index++] = edge;
2443 next = edge->node[UPPER];
2444 }
2445 next = walk_down_backref(edges, &index);
2446 }
2447 return num_bytes;
2448}
2449
2450static int reserve_metadata_space(struct btrfs_trans_handle *trans,
2451 struct reloc_control *rc,
2452 struct backref_node *node)
2453{
2454 struct btrfs_root *root = rc->extent_root;
2455 u64 num_bytes;
2456 int ret;
2457
2458 num_bytes = calcu_metadata_size(rc, node, 1) * 2;
2459
2460 trans->block_rsv = rc->block_rsv;
2461 ret = btrfs_block_rsv_add(root, rc->block_rsv, num_bytes);
2462 if (ret) {
2463 if (ret == -EAGAIN)
2464 rc->commit_transaction = 1;
2465 return ret;
2466 }
2467
2468 return 0;
2469}
2470
2471static void release_metadata_space(struct reloc_control *rc,
2472 struct backref_node *node)
2473{
2474 u64 num_bytes = calcu_metadata_size(rc, node, 0) * 2;
2475 btrfs_block_rsv_release(rc->extent_root, rc->block_rsv, num_bytes);
2476}
2477
2478/*
2479 * relocate a block tree, and then update pointers in upper level
2480 * blocks that reference the block to point to the new location.
2481 *
2482 * if called by link_to_upper, the block has already been relocated.
2483 * in that case this function just updates pointers.
2484 */
2485static int do_relocation(struct btrfs_trans_handle *trans,
2486 struct reloc_control *rc,
2487 struct backref_node *node,
2488 struct btrfs_key *key,
2489 struct btrfs_path *path, int lowest)
2490{
2491 struct backref_node *upper;
2492 struct backref_edge *edge;
2493 struct backref_edge *edges[BTRFS_MAX_LEVEL - 1];
2494 struct btrfs_root *root;
2495 struct extent_buffer *eb;
2496 u32 blocksize;
2497 u64 bytenr;
2498 u64 generation;
2499 int nr;
2500 int slot;
2501 int ret;
2502 int err = 0;
2503
2504 BUG_ON(lowest && node->eb);
2505
2506 path->lowest_level = node->level + 1;
2507 rc->backref_cache.path[node->level] = node;
2508 list_for_each_entry(edge, &node->upper, list[LOWER]) {
2509 cond_resched();
2510
2511 upper = edge->node[UPPER];
2512 root = select_reloc_root(trans, rc, upper, edges, &nr);
2513 BUG_ON(!root);
2514
2515 if (upper->eb && !upper->locked) {
2516 if (!lowest) {
2517 ret = btrfs_bin_search(upper->eb, key,
2518 upper->level, &slot);
2519 BUG_ON(ret);
2520 bytenr = btrfs_node_blockptr(upper->eb, slot);
2521 if (node->eb->start == bytenr)
2522 goto next;
2523 }
2524 drop_node_buffer(upper);
2525 }
2526
2527 if (!upper->eb) {
2528 ret = btrfs_search_slot(trans, root, key, path, 0, 1);
2529 if (ret < 0) {
2530 err = ret;
2531 break;
2532 }
2533 BUG_ON(ret > 0);
2534
2535 if (!upper->eb) {
2536 upper->eb = path->nodes[upper->level];
2537 path->nodes[upper->level] = NULL;
2538 } else {
2539 BUG_ON(upper->eb != path->nodes[upper->level]);
2540 }
2541
2542 upper->locked = 1;
2543 path->locks[upper->level] = 0;
2544
2545 slot = path->slots[upper->level];
2546 btrfs_release_path(path);
2547 } else {
2548 ret = btrfs_bin_search(upper->eb, key, upper->level,
2549 &slot);
2550 BUG_ON(ret);
2551 }
2552
2553 bytenr = btrfs_node_blockptr(upper->eb, slot);
2554 if (lowest) {
2555 BUG_ON(bytenr != node->bytenr);
2556 } else {
2557 if (node->eb->start == bytenr)
2558 goto next;
2559 }
2560
2561 blocksize = btrfs_level_size(root, node->level);
2562 generation = btrfs_node_ptr_generation(upper->eb, slot);
2563 eb = read_tree_block(root, bytenr, blocksize, generation);
2564 if (!eb) {
2565 err = -EIO;
2566 goto next;
2567 }
2568 btrfs_tree_lock(eb);
2569 btrfs_set_lock_blocking(eb);
2570
2571 if (!node->eb) {
2572 ret = btrfs_cow_block(trans, root, eb, upper->eb,
2573 slot, &eb);
2574 btrfs_tree_unlock(eb);
2575 free_extent_buffer(eb);
2576 if (ret < 0) {
2577 err = ret;
2578 goto next;
2579 }
2580 BUG_ON(node->eb != eb);
2581 } else {
2582 btrfs_set_node_blockptr(upper->eb, slot,
2583 node->eb->start);
2584 btrfs_set_node_ptr_generation(upper->eb, slot,
2585 trans->transid);
2586 btrfs_mark_buffer_dirty(upper->eb);
2587
2588 ret = btrfs_inc_extent_ref(trans, root,
2589 node->eb->start, blocksize,
2590 upper->eb->start,
2591 btrfs_header_owner(upper->eb),
2592 node->level, 0, 1);
2593 BUG_ON(ret);
2594
2595 ret = btrfs_drop_subtree(trans, root, eb, upper->eb);
2596 BUG_ON(ret);
2597 }
2598next:
2599 if (!upper->pending)
2600 drop_node_buffer(upper);
2601 else
2602 unlock_node_buffer(upper);
2603 if (err)
2604 break;
2605 }
2606
2607 if (!err && node->pending) {
2608 drop_node_buffer(node);
2609 list_move_tail(&node->list, &rc->backref_cache.changed);
2610 node->pending = 0;
2611 }
2612
2613 path->lowest_level = 0;
2614 BUG_ON(err == -ENOSPC);
2615 return err;
2616}
2617
2618static int link_to_upper(struct btrfs_trans_handle *trans,
2619 struct reloc_control *rc,
2620 struct backref_node *node,
2621 struct btrfs_path *path)
2622{
2623 struct btrfs_key key;
2624
2625 btrfs_node_key_to_cpu(node->eb, &key, 0);
2626 return do_relocation(trans, rc, node, &key, path, 0);
2627}
2628
2629static int finish_pending_nodes(struct btrfs_trans_handle *trans,
2630 struct reloc_control *rc,
2631 struct btrfs_path *path, int err)
2632{
2633 LIST_HEAD(list);
2634 struct backref_cache *cache = &rc->backref_cache;
2635 struct backref_node *node;
2636 int level;
2637 int ret;
2638
2639 for (level = 0; level < BTRFS_MAX_LEVEL; level++) {
2640 while (!list_empty(&cache->pending[level])) {
2641 node = list_entry(cache->pending[level].next,
2642 struct backref_node, list);
2643 list_move_tail(&node->list, &list);
2644 BUG_ON(!node->pending);
2645
2646 if (!err) {
2647 ret = link_to_upper(trans, rc, node, path);
2648 if (ret < 0)
2649 err = ret;
2650 }
2651 }
2652 list_splice_init(&list, &cache->pending[level]);
2653 }
2654 return err;
2655}
2656
2657static void mark_block_processed(struct reloc_control *rc,
2658 u64 bytenr, u32 blocksize)
2659{
2660 set_extent_bits(&rc->processed_blocks, bytenr, bytenr + blocksize - 1,
2661 EXTENT_DIRTY, GFP_NOFS);
2662}
2663
2664static void __mark_block_processed(struct reloc_control *rc,
2665 struct backref_node *node)
2666{
2667 u32 blocksize;
2668 if (node->level == 0 ||
2669 in_block_group(node->bytenr, rc->block_group)) {
2670 blocksize = btrfs_level_size(rc->extent_root, node->level);
2671 mark_block_processed(rc, node->bytenr, blocksize);
2672 }
2673 node->processed = 1;
2674}
2675
2676/*
2677 * mark a block and all blocks directly/indirectly reference the block
2678 * as processed.
2679 */
2680static void update_processed_blocks(struct reloc_control *rc,
2681 struct backref_node *node)
2682{
2683 struct backref_node *next = node;
2684 struct backref_edge *edge;
2685 struct backref_edge *edges[BTRFS_MAX_LEVEL - 1];
2686 int index = 0;
2687
2688 while (next) {
2689 cond_resched();
2690 while (1) {
2691 if (next->processed)
2692 break;
2693
2694 __mark_block_processed(rc, next);
2695
2696 if (list_empty(&next->upper))
2697 break;
2698
2699 edge = list_entry(next->upper.next,
2700 struct backref_edge, list[LOWER]);
2701 edges[index++] = edge;
2702 next = edge->node[UPPER];
2703 }
2704 next = walk_down_backref(edges, &index);
2705 }
2706}
2707
2708static int tree_block_processed(u64 bytenr, u32 blocksize,
2709 struct reloc_control *rc)
2710{
2711 if (test_range_bit(&rc->processed_blocks, bytenr,
2712 bytenr + blocksize - 1, EXTENT_DIRTY, 1, NULL))
2713 return 1;
2714 return 0;
2715}
2716
2717static int get_tree_block_key(struct reloc_control *rc,
2718 struct tree_block *block)
2719{
2720 struct extent_buffer *eb;
2721
2722 BUG_ON(block->key_ready);
2723 eb = read_tree_block(rc->extent_root, block->bytenr,
2724 block->key.objectid, block->key.offset);
2725 BUG_ON(!eb);
2726 WARN_ON(btrfs_header_level(eb) != block->level);
2727 if (block->level == 0)
2728 btrfs_item_key_to_cpu(eb, &block->key, 0);
2729 else
2730 btrfs_node_key_to_cpu(eb, &block->key, 0);
2731 free_extent_buffer(eb);
2732 block->key_ready = 1;
2733 return 0;
2734}
2735
2736static int reada_tree_block(struct reloc_control *rc,
2737 struct tree_block *block)
2738{
2739 BUG_ON(block->key_ready);
2740 readahead_tree_block(rc->extent_root, block->bytenr,
2741 block->key.objectid, block->key.offset);
2742 return 0;
2743}
2744
2745/*
2746 * helper function to relocate a tree block
2747 */
2748static int relocate_tree_block(struct btrfs_trans_handle *trans,
2749 struct reloc_control *rc,
2750 struct backref_node *node,
2751 struct btrfs_key *key,
2752 struct btrfs_path *path)
2753{
2754 struct btrfs_root *root;
2755 int release = 0;
2756 int ret = 0;
2757
2758 if (!node)
2759 return 0;
2760
2761 BUG_ON(node->processed);
2762 root = select_one_root(trans, node);
2763 if (root == ERR_PTR(-ENOENT)) {
2764 update_processed_blocks(rc, node);
2765 goto out;
2766 }
2767
2768 if (!root || root->ref_cows) {
2769 ret = reserve_metadata_space(trans, rc, node);
2770 if (ret)
2771 goto out;
2772 release = 1;
2773 }
2774
2775 if (root) {
2776 if (root->ref_cows) {
2777 BUG_ON(node->new_bytenr);
2778 BUG_ON(!list_empty(&node->list));
2779 btrfs_record_root_in_trans(trans, root);
2780 root = root->reloc_root;
2781 node->new_bytenr = root->node->start;
2782 node->root = root;
2783 list_add_tail(&node->list, &rc->backref_cache.changed);
2784 } else {
2785 path->lowest_level = node->level;
2786 ret = btrfs_search_slot(trans, root, key, path, 0, 1);
2787 btrfs_release_path(path);
2788 if (ret > 0)
2789 ret = 0;
2790 }
2791 if (!ret)
2792 update_processed_blocks(rc, node);
2793 } else {
2794 ret = do_relocation(trans, rc, node, key, path, 1);
2795 }
2796out:
2797 if (ret || node->level == 0 || node->cowonly) {
2798 if (release)
2799 release_metadata_space(rc, node);
2800 remove_backref_node(&rc->backref_cache, node);
2801 }
2802 return ret;
2803}
2804
2805/*
2806 * relocate a list of blocks
2807 */
2808static noinline_for_stack
2809int relocate_tree_blocks(struct btrfs_trans_handle *trans,
2810 struct reloc_control *rc, struct rb_root *blocks)
2811{
2812 struct backref_node *node;
2813 struct btrfs_path *path;
2814 struct tree_block *block;
2815 struct rb_node *rb_node;
2816 int ret;
2817 int err = 0;
2818
2819 path = btrfs_alloc_path();
2820 if (!path)
2821 return -ENOMEM;
2822
2823 rb_node = rb_first(blocks);
2824 while (rb_node) {
2825 block = rb_entry(rb_node, struct tree_block, rb_node);
2826 if (!block->key_ready)
2827 reada_tree_block(rc, block);
2828 rb_node = rb_next(rb_node);
2829 }
2830
2831 rb_node = rb_first(blocks);
2832 while (rb_node) {
2833 block = rb_entry(rb_node, struct tree_block, rb_node);
2834 if (!block->key_ready)
2835 get_tree_block_key(rc, block);
2836 rb_node = rb_next(rb_node);
2837 }
2838
2839 rb_node = rb_first(blocks);
2840 while (rb_node) {
2841 block = rb_entry(rb_node, struct tree_block, rb_node);
2842
2843 node = build_backref_tree(rc, &block->key,
2844 block->level, block->bytenr);
2845 if (IS_ERR(node)) {
2846 err = PTR_ERR(node);
2847 goto out;
2848 }
2849
2850 ret = relocate_tree_block(trans, rc, node, &block->key,
2851 path);
2852 if (ret < 0) {
2853 if (ret != -EAGAIN || rb_node == rb_first(blocks))
2854 err = ret;
2855 goto out;
2856 }
2857 rb_node = rb_next(rb_node);
2858 }
2859out:
2860 free_block_list(blocks);
2861 err = finish_pending_nodes(trans, rc, path, err);
2862
2863 btrfs_free_path(path);
2864 return err;
2865}
2866
2867static noinline_for_stack
2868int prealloc_file_extent_cluster(struct inode *inode,
2869 struct file_extent_cluster *cluster)
2870{
2871 u64 alloc_hint = 0;
2872 u64 start;
2873 u64 end;
2874 u64 offset = BTRFS_I(inode)->index_cnt;
2875 u64 num_bytes;
2876 int nr = 0;
2877 int ret = 0;
2878
2879 BUG_ON(cluster->start != cluster->boundary[0]);
2880 mutex_lock(&inode->i_mutex);
2881
2882 ret = btrfs_check_data_free_space(inode, cluster->end +
2883 1 - cluster->start);
2884 if (ret)
2885 goto out;
2886
2887 while (nr < cluster->nr) {
2888 start = cluster->boundary[nr] - offset;
2889 if (nr + 1 < cluster->nr)
2890 end = cluster->boundary[nr + 1] - 1 - offset;
2891 else
2892 end = cluster->end - offset;
2893
2894 lock_extent(&BTRFS_I(inode)->io_tree, start, end);
2895 num_bytes = end + 1 - start;
2896 ret = btrfs_prealloc_file_range(inode, 0, start,
2897 num_bytes, num_bytes,
2898 end + 1, &alloc_hint);
2899 unlock_extent(&BTRFS_I(inode)->io_tree, start, end);
2900 if (ret)
2901 break;
2902 nr++;
2903 }
2904 btrfs_free_reserved_data_space(inode, cluster->end +
2905 1 - cluster->start);
2906out:
2907 mutex_unlock(&inode->i_mutex);
2908 return ret;
2909}
2910
2911static noinline_for_stack
2912int setup_extent_mapping(struct inode *inode, u64 start, u64 end,
2913 u64 block_start)
2914{
2915 struct btrfs_root *root = BTRFS_I(inode)->root;
2916 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
2917 struct extent_map *em;
2918 int ret = 0;
2919
2920 em = alloc_extent_map();
2921 if (!em)
2922 return -ENOMEM;
2923
2924 em->start = start;
2925 em->len = end + 1 - start;
2926 em->block_len = em->len;
2927 em->block_start = block_start;
2928 em->bdev = root->fs_info->fs_devices->latest_bdev;
2929 set_bit(EXTENT_FLAG_PINNED, &em->flags);
2930
2931 lock_extent(&BTRFS_I(inode)->io_tree, start, end);
2932 while (1) {
2933 write_lock(&em_tree->lock);
2934 ret = add_extent_mapping(em_tree, em);
2935 write_unlock(&em_tree->lock);
2936 if (ret != -EEXIST) {
2937 free_extent_map(em);
2938 break;
2939 }
2940 btrfs_drop_extent_cache(inode, start, end, 0);
2941 }
2942 unlock_extent(&BTRFS_I(inode)->io_tree, start, end);
2943 return ret;
2944}
2945
2946static int relocate_file_extent_cluster(struct inode *inode,
2947 struct file_extent_cluster *cluster)
2948{
2949 u64 page_start;
2950 u64 page_end;
2951 u64 offset = BTRFS_I(inode)->index_cnt;
2952 unsigned long index;
2953 unsigned long last_index;
2954 struct page *page;
2955 struct file_ra_state *ra;
2956 gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping);
2957 int nr = 0;
2958 int ret = 0;
2959
2960 if (!cluster->nr)
2961 return 0;
2962
2963 ra = kzalloc(sizeof(*ra), GFP_NOFS);
2964 if (!ra)
2965 return -ENOMEM;
2966
2967 ret = prealloc_file_extent_cluster(inode, cluster);
2968 if (ret)
2969 goto out;
2970
2971 file_ra_state_init(ra, inode->i_mapping);
2972
2973 ret = setup_extent_mapping(inode, cluster->start - offset,
2974 cluster->end - offset, cluster->start);
2975 if (ret)
2976 goto out;
2977
2978 index = (cluster->start - offset) >> PAGE_CACHE_SHIFT;
2979 last_index = (cluster->end - offset) >> PAGE_CACHE_SHIFT;
2980 while (index <= last_index) {
2981 ret = btrfs_delalloc_reserve_metadata(inode, PAGE_CACHE_SIZE);
2982 if (ret)
2983 goto out;
2984
2985 page = find_lock_page(inode->i_mapping, index);
2986 if (!page) {
2987 page_cache_sync_readahead(inode->i_mapping,
2988 ra, NULL, index,
2989 last_index + 1 - index);
2990 page = find_or_create_page(inode->i_mapping, index,
2991 mask);
2992 if (!page) {
2993 btrfs_delalloc_release_metadata(inode,
2994 PAGE_CACHE_SIZE);
2995 ret = -ENOMEM;
2996 goto out;
2997 }
2998 }
2999
3000 if (PageReadahead(page)) {
3001 page_cache_async_readahead(inode->i_mapping,
3002 ra, NULL, page, index,
3003 last_index + 1 - index);
3004 }
3005
3006 if (!PageUptodate(page)) {
3007 btrfs_readpage(NULL, page);
3008 lock_page(page);
3009 if (!PageUptodate(page)) {
3010 unlock_page(page);
3011 page_cache_release(page);
3012 btrfs_delalloc_release_metadata(inode,
3013 PAGE_CACHE_SIZE);
3014 ret = -EIO;
3015 goto out;
3016 }
3017 }
3018
3019 page_start = (u64)page->index << PAGE_CACHE_SHIFT;
3020 page_end = page_start + PAGE_CACHE_SIZE - 1;
3021
3022 lock_extent(&BTRFS_I(inode)->io_tree, page_start, page_end);
3023
3024 set_page_extent_mapped(page);
3025
3026 if (nr < cluster->nr &&
3027 page_start + offset == cluster->boundary[nr]) {
3028 set_extent_bits(&BTRFS_I(inode)->io_tree,
3029 page_start, page_end,
3030 EXTENT_BOUNDARY, GFP_NOFS);
3031 nr++;
3032 }
3033
3034 btrfs_set_extent_delalloc(inode, page_start, page_end, NULL);
3035 set_page_dirty(page);
3036
3037 unlock_extent(&BTRFS_I(inode)->io_tree,
3038 page_start, page_end);
3039 unlock_page(page);
3040 page_cache_release(page);
3041
3042 index++;
3043 balance_dirty_pages_ratelimited(inode->i_mapping);
3044 btrfs_throttle(BTRFS_I(inode)->root);
3045 }
3046 WARN_ON(nr != cluster->nr);
3047out:
3048 kfree(ra);
3049 return ret;
3050}
3051
3052static noinline_for_stack
3053int relocate_data_extent(struct inode *inode, struct btrfs_key *extent_key,
3054 struct file_extent_cluster *cluster)
3055{
3056 int ret;
3057
3058 if (cluster->nr > 0 && extent_key->objectid != cluster->end + 1) {
3059 ret = relocate_file_extent_cluster(inode, cluster);
3060 if (ret)
3061 return ret;
3062 cluster->nr = 0;
3063 }
3064
3065 if (!cluster->nr)
3066 cluster->start = extent_key->objectid;
3067 else
3068 BUG_ON(cluster->nr >= MAX_EXTENTS);
3069 cluster->end = extent_key->objectid + extent_key->offset - 1;
3070 cluster->boundary[cluster->nr] = extent_key->objectid;
3071 cluster->nr++;
3072
3073 if (cluster->nr >= MAX_EXTENTS) {
3074 ret = relocate_file_extent_cluster(inode, cluster);
3075 if (ret)
3076 return ret;
3077 cluster->nr = 0;
3078 }
3079 return 0;
3080}
3081
3082#ifdef BTRFS_COMPAT_EXTENT_TREE_V0
3083static int get_ref_objectid_v0(struct reloc_control *rc,
3084 struct btrfs_path *path,
3085 struct btrfs_key *extent_key,
3086 u64 *ref_objectid, int *path_change)
3087{
3088 struct btrfs_key key;
3089 struct extent_buffer *leaf;
3090 struct btrfs_extent_ref_v0 *ref0;
3091 int ret;
3092 int slot;
3093
3094 leaf = path->nodes[0];
3095 slot = path->slots[0];
3096 while (1) {
3097 if (slot >= btrfs_header_nritems(leaf)) {
3098 ret = btrfs_next_leaf(rc->extent_root, path);
3099 if (ret < 0)
3100 return ret;
3101 BUG_ON(ret > 0);
3102 leaf = path->nodes[0];
3103 slot = path->slots[0];
3104 if (path_change)
3105 *path_change = 1;
3106 }
3107 btrfs_item_key_to_cpu(leaf, &key, slot);
3108 if (key.objectid != extent_key->objectid)
3109 return -ENOENT;
3110
3111 if (key.type != BTRFS_EXTENT_REF_V0_KEY) {
3112 slot++;
3113 continue;
3114 }
3115 ref0 = btrfs_item_ptr(leaf, slot,
3116 struct btrfs_extent_ref_v0);
3117 *ref_objectid = btrfs_ref_objectid_v0(leaf, ref0);
3118 break;
3119 }
3120 return 0;
3121}
3122#endif
3123
3124/*
3125 * helper to add a tree block to the list.
3126 * the major work is getting the generation and level of the block
3127 */
3128static int add_tree_block(struct reloc_control *rc,
3129 struct btrfs_key *extent_key,
3130 struct btrfs_path *path,
3131 struct rb_root *blocks)
3132{
3133 struct extent_buffer *eb;
3134 struct btrfs_extent_item *ei;
3135 struct btrfs_tree_block_info *bi;
3136 struct tree_block *block;
3137 struct rb_node *rb_node;
3138 u32 item_size;
3139 int level = -1;
3140 int generation;
3141
3142 eb = path->nodes[0];
3143 item_size = btrfs_item_size_nr(eb, path->slots[0]);
3144
3145 if (item_size >= sizeof(*ei) + sizeof(*bi)) {
3146 ei = btrfs_item_ptr(eb, path->slots[0],
3147 struct btrfs_extent_item);
3148 bi = (struct btrfs_tree_block_info *)(ei + 1);
3149 generation = btrfs_extent_generation(eb, ei);
3150 level = btrfs_tree_block_level(eb, bi);
3151 } else {
3152#ifdef BTRFS_COMPAT_EXTENT_TREE_V0
3153 u64 ref_owner;
3154 int ret;
3155
3156 BUG_ON(item_size != sizeof(struct btrfs_extent_item_v0));
3157 ret = get_ref_objectid_v0(rc, path, extent_key,
3158 &ref_owner, NULL);
3159 if (ret < 0)
3160 return ret;
3161 BUG_ON(ref_owner >= BTRFS_MAX_LEVEL);
3162 level = (int)ref_owner;
3163 /* FIXME: get real generation */
3164 generation = 0;
3165#else
3166 BUG();
3167#endif
3168 }
3169
3170 btrfs_release_path(path);
3171
3172 BUG_ON(level == -1);
3173
3174 block = kmalloc(sizeof(*block), GFP_NOFS);
3175 if (!block)
3176 return -ENOMEM;
3177
3178 block->bytenr = extent_key->objectid;
3179 block->key.objectid = extent_key->offset;
3180 block->key.offset = generation;
3181 block->level = level;
3182 block->key_ready = 0;
3183
3184 rb_node = tree_insert(blocks, block->bytenr, &block->rb_node);
3185 if (rb_node)
3186 backref_tree_panic(rb_node, -EEXIST, block->bytenr);
3187
3188 return 0;
3189}
3190
3191/*
3192 * helper to add tree blocks for backref of type BTRFS_SHARED_DATA_REF_KEY
3193 */
3194static int __add_tree_block(struct reloc_control *rc,
3195 u64 bytenr, u32 blocksize,
3196 struct rb_root *blocks)
3197{
3198 struct btrfs_path *path;
3199 struct btrfs_key key;
3200 int ret;
3201
3202 if (tree_block_processed(bytenr, blocksize, rc))
3203 return 0;
3204
3205 if (tree_search(blocks, bytenr))
3206 return 0;
3207
3208 path = btrfs_alloc_path();
3209 if (!path)
3210 return -ENOMEM;
3211
3212 key.objectid = bytenr;
3213 key.type = BTRFS_EXTENT_ITEM_KEY;
3214 key.offset = blocksize;
3215
3216 path->search_commit_root = 1;
3217 path->skip_locking = 1;
3218 ret = btrfs_search_slot(NULL, rc->extent_root, &key, path, 0, 0);
3219 if (ret < 0)
3220 goto out;
3221 BUG_ON(ret);
3222
3223 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
3224 ret = add_tree_block(rc, &key, path, blocks);
3225out:
3226 btrfs_free_path(path);
3227 return ret;
3228}
3229
3230/*
3231 * helper to check if the block use full backrefs for pointers in it
3232 */
3233static int block_use_full_backref(struct reloc_control *rc,
3234 struct extent_buffer *eb)
3235{
3236 u64 flags;
3237 int ret;
3238
3239 if (btrfs_header_flag(eb, BTRFS_HEADER_FLAG_RELOC) ||
3240 btrfs_header_backref_rev(eb) < BTRFS_MIXED_BACKREF_REV)
3241 return 1;
3242
3243 ret = btrfs_lookup_extent_info(NULL, rc->extent_root,
3244 eb->start, eb->len, NULL, &flags);
3245 BUG_ON(ret);
3246
3247 if (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)
3248 ret = 1;
3249 else
3250 ret = 0;
3251 return ret;
3252}
3253
3254static int delete_block_group_cache(struct btrfs_fs_info *fs_info,
3255 struct inode *inode, u64 ino)
3256{
3257 struct btrfs_key key;
3258 struct btrfs_path *path;
3259 struct btrfs_root *root = fs_info->tree_root;
3260 struct btrfs_trans_handle *trans;
3261 unsigned long nr;
3262 int ret = 0;
3263
3264 if (inode)
3265 goto truncate;
3266
3267 key.objectid = ino;
3268 key.type = BTRFS_INODE_ITEM_KEY;
3269 key.offset = 0;
3270
3271 inode = btrfs_iget(fs_info->sb, &key, root, NULL);
3272 if (IS_ERR_OR_NULL(inode) || is_bad_inode(inode)) {
3273 if (inode && !IS_ERR(inode))
3274 iput(inode);
3275 return -ENOENT;
3276 }
3277
3278truncate:
3279 path = btrfs_alloc_path();
3280 if (!path) {
3281 ret = -ENOMEM;
3282 goto out;
3283 }
3284
3285 trans = btrfs_join_transaction(root);
3286 if (IS_ERR(trans)) {
3287 btrfs_free_path(path);
3288 ret = PTR_ERR(trans);
3289 goto out;
3290 }
3291
3292 ret = btrfs_truncate_free_space_cache(root, trans, path, inode);
3293
3294 btrfs_free_path(path);
3295 nr = trans->blocks_used;
3296 btrfs_end_transaction(trans, root);
3297 btrfs_btree_balance_dirty(root, nr);
3298out:
3299 iput(inode);
3300 return ret;
3301}
3302
3303/*
3304 * helper to add tree blocks for backref of type BTRFS_EXTENT_DATA_REF_KEY
3305 * this function scans fs tree to find blocks reference the data extent
3306 */
3307static int find_data_references(struct reloc_control *rc,
3308 struct btrfs_key *extent_key,
3309 struct extent_buffer *leaf,
3310 struct btrfs_extent_data_ref *ref,
3311 struct rb_root *blocks)
3312{
3313 struct btrfs_path *path;
3314 struct tree_block *block;
3315 struct btrfs_root *root;
3316 struct btrfs_file_extent_item *fi;
3317 struct rb_node *rb_node;
3318 struct btrfs_key key;
3319 u64 ref_root;
3320 u64 ref_objectid;
3321 u64 ref_offset;
3322 u32 ref_count;
3323 u32 nritems;
3324 int err = 0;
3325 int added = 0;
3326 int counted;
3327 int ret;
3328
3329 ref_root = btrfs_extent_data_ref_root(leaf, ref);
3330 ref_objectid = btrfs_extent_data_ref_objectid(leaf, ref);
3331 ref_offset = btrfs_extent_data_ref_offset(leaf, ref);
3332 ref_count = btrfs_extent_data_ref_count(leaf, ref);
3333
3334 /*
3335 * This is an extent belonging to the free space cache, lets just delete
3336 * it and redo the search.
3337 */
3338 if (ref_root == BTRFS_ROOT_TREE_OBJECTID) {
3339 ret = delete_block_group_cache(rc->extent_root->fs_info,
3340 NULL, ref_objectid);
3341 if (ret != -ENOENT)
3342 return ret;
3343 ret = 0;
3344 }
3345
3346 path = btrfs_alloc_path();
3347 if (!path)
3348 return -ENOMEM;
3349 path->reada = 1;
3350
3351 root = read_fs_root(rc->extent_root->fs_info, ref_root);
3352 if (IS_ERR(root)) {
3353 err = PTR_ERR(root);
3354 goto out;
3355 }
3356
3357 key.objectid = ref_objectid;
3358 key.type = BTRFS_EXTENT_DATA_KEY;
3359 if (ref_offset > ((u64)-1 << 32))
3360 key.offset = 0;
3361 else
3362 key.offset = ref_offset;
3363
3364 path->search_commit_root = 1;
3365 path->skip_locking = 1;
3366 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3367 if (ret < 0) {
3368 err = ret;
3369 goto out;
3370 }
3371
3372 leaf = path->nodes[0];
3373 nritems = btrfs_header_nritems(leaf);
3374 /*
3375 * the references in tree blocks that use full backrefs
3376 * are not counted in
3377 */
3378 if (block_use_full_backref(rc, leaf))
3379 counted = 0;
3380 else
3381 counted = 1;
3382 rb_node = tree_search(blocks, leaf->start);
3383 if (rb_node) {
3384 if (counted)
3385 added = 1;
3386 else
3387 path->slots[0] = nritems;
3388 }
3389
3390 while (ref_count > 0) {
3391 while (path->slots[0] >= nritems) {
3392 ret = btrfs_next_leaf(root, path);
3393 if (ret < 0) {
3394 err = ret;
3395 goto out;
3396 }
3397 if (ret > 0) {
3398 WARN_ON(1);
3399 goto out;
3400 }
3401
3402 leaf = path->nodes[0];
3403 nritems = btrfs_header_nritems(leaf);
3404 added = 0;
3405
3406 if (block_use_full_backref(rc, leaf))
3407 counted = 0;
3408 else
3409 counted = 1;
3410 rb_node = tree_search(blocks, leaf->start);
3411 if (rb_node) {
3412 if (counted)
3413 added = 1;
3414 else
3415 path->slots[0] = nritems;
3416 }
3417 }
3418
3419 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3420 if (key.objectid != ref_objectid ||
3421 key.type != BTRFS_EXTENT_DATA_KEY) {
3422 WARN_ON(1);
3423 break;
3424 }
3425
3426 fi = btrfs_item_ptr(leaf, path->slots[0],
3427 struct btrfs_file_extent_item);
3428
3429 if (btrfs_file_extent_type(leaf, fi) ==
3430 BTRFS_FILE_EXTENT_INLINE)
3431 goto next;
3432
3433 if (btrfs_file_extent_disk_bytenr(leaf, fi) !=
3434 extent_key->objectid)
3435 goto next;
3436
3437 key.offset -= btrfs_file_extent_offset(leaf, fi);
3438 if (key.offset != ref_offset)
3439 goto next;
3440
3441 if (counted)
3442 ref_count--;
3443 if (added)
3444 goto next;
3445
3446 if (!tree_block_processed(leaf->start, leaf->len, rc)) {
3447 block = kmalloc(sizeof(*block), GFP_NOFS);
3448 if (!block) {
3449 err = -ENOMEM;
3450 break;
3451 }
3452 block->bytenr = leaf->start;
3453 btrfs_item_key_to_cpu(leaf, &block->key, 0);
3454 block->level = 0;
3455 block->key_ready = 1;
3456 rb_node = tree_insert(blocks, block->bytenr,
3457 &block->rb_node);
3458 if (rb_node)
3459 backref_tree_panic(rb_node, -EEXIST,
3460 block->bytenr);
3461 }
3462 if (counted)
3463 added = 1;
3464 else
3465 path->slots[0] = nritems;
3466next:
3467 path->slots[0]++;
3468
3469 }
3470out:
3471 btrfs_free_path(path);
3472 return err;
3473}
3474
3475/*
3476 * hepler to find all tree blocks that reference a given data extent
3477 */
3478static noinline_for_stack
3479int add_data_references(struct reloc_control *rc,
3480 struct btrfs_key *extent_key,
3481 struct btrfs_path *path,
3482 struct rb_root *blocks)
3483{
3484 struct btrfs_key key;
3485 struct extent_buffer *eb;
3486 struct btrfs_extent_data_ref *dref;
3487 struct btrfs_extent_inline_ref *iref;
3488 unsigned long ptr;
3489 unsigned long end;
3490 u32 blocksize = btrfs_level_size(rc->extent_root, 0);
3491 int ret;
3492 int err = 0;
3493
3494 eb = path->nodes[0];
3495 ptr = btrfs_item_ptr_offset(eb, path->slots[0]);
3496 end = ptr + btrfs_item_size_nr(eb, path->slots[0]);
3497#ifdef BTRFS_COMPAT_EXTENT_TREE_V0
3498 if (ptr + sizeof(struct btrfs_extent_item_v0) == end)
3499 ptr = end;
3500 else
3501#endif
3502 ptr += sizeof(struct btrfs_extent_item);
3503
3504 while (ptr < end) {
3505 iref = (struct btrfs_extent_inline_ref *)ptr;
3506 key.type = btrfs_extent_inline_ref_type(eb, iref);
3507 if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
3508 key.offset = btrfs_extent_inline_ref_offset(eb, iref);
3509 ret = __add_tree_block(rc, key.offset, blocksize,
3510 blocks);
3511 } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
3512 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
3513 ret = find_data_references(rc, extent_key,
3514 eb, dref, blocks);
3515 } else {
3516 BUG();
3517 }
3518 ptr += btrfs_extent_inline_ref_size(key.type);
3519 }
3520 WARN_ON(ptr > end);
3521
3522 while (1) {
3523 cond_resched();
3524 eb = path->nodes[0];
3525 if (path->slots[0] >= btrfs_header_nritems(eb)) {
3526 ret = btrfs_next_leaf(rc->extent_root, path);
3527 if (ret < 0) {
3528 err = ret;
3529 break;
3530 }
3531 if (ret > 0)
3532 break;
3533 eb = path->nodes[0];
3534 }
3535
3536 btrfs_item_key_to_cpu(eb, &key, path->slots[0]);
3537 if (key.objectid != extent_key->objectid)
3538 break;
3539
3540#ifdef BTRFS_COMPAT_EXTENT_TREE_V0
3541 if (key.type == BTRFS_SHARED_DATA_REF_KEY ||
3542 key.type == BTRFS_EXTENT_REF_V0_KEY) {
3543#else
3544 BUG_ON(key.type == BTRFS_EXTENT_REF_V0_KEY);
3545 if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
3546#endif
3547 ret = __add_tree_block(rc, key.offset, blocksize,
3548 blocks);
3549 } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
3550 dref = btrfs_item_ptr(eb, path->slots[0],
3551 struct btrfs_extent_data_ref);
3552 ret = find_data_references(rc, extent_key,
3553 eb, dref, blocks);
3554 } else {
3555 ret = 0;
3556 }
3557 if (ret) {
3558 err = ret;
3559 break;
3560 }
3561 path->slots[0]++;
3562 }
3563 btrfs_release_path(path);
3564 if (err)
3565 free_block_list(blocks);
3566 return err;
3567}
3568
3569/*
3570 * hepler to find next unprocessed extent
3571 */
3572static noinline_for_stack
3573int find_next_extent(struct btrfs_trans_handle *trans,
3574 struct reloc_control *rc, struct btrfs_path *path,
3575 struct btrfs_key *extent_key)
3576{
3577 struct btrfs_key key;
3578 struct extent_buffer *leaf;
3579 u64 start, end, last;
3580 int ret;
3581
3582 last = rc->block_group->key.objectid + rc->block_group->key.offset;
3583 while (1) {
3584 cond_resched();
3585 if (rc->search_start >= last) {
3586 ret = 1;
3587 break;
3588 }
3589
3590 key.objectid = rc->search_start;
3591 key.type = BTRFS_EXTENT_ITEM_KEY;
3592 key.offset = 0;
3593
3594 path->search_commit_root = 1;
3595 path->skip_locking = 1;
3596 ret = btrfs_search_slot(NULL, rc->extent_root, &key, path,
3597 0, 0);
3598 if (ret < 0)
3599 break;
3600next:
3601 leaf = path->nodes[0];
3602 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
3603 ret = btrfs_next_leaf(rc->extent_root, path);
3604 if (ret != 0)
3605 break;
3606 leaf = path->nodes[0];
3607 }
3608
3609 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3610 if (key.objectid >= last) {
3611 ret = 1;
3612 break;
3613 }
3614
3615 if (key.type != BTRFS_EXTENT_ITEM_KEY ||
3616 key.objectid + key.offset <= rc->search_start) {
3617 path->slots[0]++;
3618 goto next;
3619 }
3620
3621 ret = find_first_extent_bit(&rc->processed_blocks,
3622 key.objectid, &start, &end,
3623 EXTENT_DIRTY);
3624
3625 if (ret == 0 && start <= key.objectid) {
3626 btrfs_release_path(path);
3627 rc->search_start = end + 1;
3628 } else {
3629 rc->search_start = key.objectid + key.offset;
3630 memcpy(extent_key, &key, sizeof(key));
3631 return 0;
3632 }
3633 }
3634 btrfs_release_path(path);
3635 return ret;
3636}
3637
3638static void set_reloc_control(struct reloc_control *rc)
3639{
3640 struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
3641
3642 mutex_lock(&fs_info->reloc_mutex);
3643 fs_info->reloc_ctl = rc;
3644 mutex_unlock(&fs_info->reloc_mutex);
3645}
3646
3647static void unset_reloc_control(struct reloc_control *rc)
3648{
3649 struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
3650
3651 mutex_lock(&fs_info->reloc_mutex);
3652 fs_info->reloc_ctl = NULL;
3653 mutex_unlock(&fs_info->reloc_mutex);
3654}
3655
3656static int check_extent_flags(u64 flags)
3657{
3658 if ((flags & BTRFS_EXTENT_FLAG_DATA) &&
3659 (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK))
3660 return 1;
3661 if (!(flags & BTRFS_EXTENT_FLAG_DATA) &&
3662 !(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK))
3663 return 1;
3664 if ((flags & BTRFS_EXTENT_FLAG_DATA) &&
3665 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
3666 return 1;
3667 return 0;
3668}
3669
3670static noinline_for_stack
3671int prepare_to_relocate(struct reloc_control *rc)
3672{
3673 struct btrfs_trans_handle *trans;
3674 int ret;
3675
3676 rc->block_rsv = btrfs_alloc_block_rsv(rc->extent_root);
3677 if (!rc->block_rsv)
3678 return -ENOMEM;
3679
3680 /*
3681 * reserve some space for creating reloc trees.
3682 * btrfs_init_reloc_root will use them when there
3683 * is no reservation in transaction handle.
3684 */
3685 ret = btrfs_block_rsv_add(rc->extent_root, rc->block_rsv,
3686 rc->extent_root->nodesize * 256);
3687 if (ret)
3688 return ret;
3689
3690 memset(&rc->cluster, 0, sizeof(rc->cluster));
3691 rc->search_start = rc->block_group->key.objectid;
3692 rc->extents_found = 0;
3693 rc->nodes_relocated = 0;
3694 rc->merging_rsv_size = 0;
3695
3696 rc->create_reloc_tree = 1;
3697 set_reloc_control(rc);
3698
3699 trans = btrfs_join_transaction(rc->extent_root);
3700 BUG_ON(IS_ERR(trans));
3701 btrfs_commit_transaction(trans, rc->extent_root);
3702 return 0;
3703}
3704
3705static noinline_for_stack int relocate_block_group(struct reloc_control *rc)
3706{
3707 struct rb_root blocks = RB_ROOT;
3708 struct btrfs_key key;
3709 struct btrfs_trans_handle *trans = NULL;
3710 struct btrfs_path *path;
3711 struct btrfs_extent_item *ei;
3712 unsigned long nr;
3713 u64 flags;
3714 u32 item_size;
3715 int ret;
3716 int err = 0;
3717 int progress = 0;
3718
3719 path = btrfs_alloc_path();
3720 if (!path)
3721 return -ENOMEM;
3722 path->reada = 1;
3723
3724 ret = prepare_to_relocate(rc);
3725 if (ret) {
3726 err = ret;
3727 goto out_free;
3728 }
3729
3730 while (1) {
3731 progress++;
3732 trans = btrfs_start_transaction(rc->extent_root, 0);
3733 BUG_ON(IS_ERR(trans));
3734restart:
3735 if (update_backref_cache(trans, &rc->backref_cache)) {
3736 btrfs_end_transaction(trans, rc->extent_root);
3737 continue;
3738 }
3739
3740 ret = find_next_extent(trans, rc, path, &key);
3741 if (ret < 0)
3742 err = ret;
3743 if (ret != 0)
3744 break;
3745
3746 rc->extents_found++;
3747
3748 ei = btrfs_item_ptr(path->nodes[0], path->slots[0],
3749 struct btrfs_extent_item);
3750 item_size = btrfs_item_size_nr(path->nodes[0], path->slots[0]);
3751 if (item_size >= sizeof(*ei)) {
3752 flags = btrfs_extent_flags(path->nodes[0], ei);
3753 ret = check_extent_flags(flags);
3754 BUG_ON(ret);
3755
3756 } else {
3757#ifdef BTRFS_COMPAT_EXTENT_TREE_V0
3758 u64 ref_owner;
3759 int path_change = 0;
3760
3761 BUG_ON(item_size !=
3762 sizeof(struct btrfs_extent_item_v0));
3763 ret = get_ref_objectid_v0(rc, path, &key, &ref_owner,
3764 &path_change);
3765 if (ref_owner < BTRFS_FIRST_FREE_OBJECTID)
3766 flags = BTRFS_EXTENT_FLAG_TREE_BLOCK;
3767 else
3768 flags = BTRFS_EXTENT_FLAG_DATA;
3769
3770 if (path_change) {
3771 btrfs_release_path(path);
3772
3773 path->search_commit_root = 1;
3774 path->skip_locking = 1;
3775 ret = btrfs_search_slot(NULL, rc->extent_root,
3776 &key, path, 0, 0);
3777 if (ret < 0) {
3778 err = ret;
3779 break;
3780 }
3781 BUG_ON(ret > 0);
3782 }
3783#else
3784 BUG();
3785#endif
3786 }
3787
3788 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
3789 ret = add_tree_block(rc, &key, path, &blocks);
3790 } else if (rc->stage == UPDATE_DATA_PTRS &&
3791 (flags & BTRFS_EXTENT_FLAG_DATA)) {
3792 ret = add_data_references(rc, &key, path, &blocks);
3793 } else {
3794 btrfs_release_path(path);
3795 ret = 0;
3796 }
3797 if (ret < 0) {
3798 err = ret;
3799 break;
3800 }
3801
3802 if (!RB_EMPTY_ROOT(&blocks)) {
3803 ret = relocate_tree_blocks(trans, rc, &blocks);
3804 if (ret < 0) {
3805 if (ret != -EAGAIN) {
3806 err = ret;
3807 break;
3808 }
3809 rc->extents_found--;
3810 rc->search_start = key.objectid;
3811 }
3812 }
3813
3814 ret = btrfs_block_rsv_check(rc->extent_root, rc->block_rsv, 5);
3815 if (ret < 0) {
3816 if (ret != -ENOSPC) {
3817 err = ret;
3818 WARN_ON(1);
3819 break;
3820 }
3821 rc->commit_transaction = 1;
3822 }
3823
3824 if (rc->commit_transaction) {
3825 rc->commit_transaction = 0;
3826 ret = btrfs_commit_transaction(trans, rc->extent_root);
3827 BUG_ON(ret);
3828 } else {
3829 nr = trans->blocks_used;
3830 btrfs_end_transaction_throttle(trans, rc->extent_root);
3831 btrfs_btree_balance_dirty(rc->extent_root, nr);
3832 }
3833 trans = NULL;
3834
3835 if (rc->stage == MOVE_DATA_EXTENTS &&
3836 (flags & BTRFS_EXTENT_FLAG_DATA)) {
3837 rc->found_file_extent = 1;
3838 ret = relocate_data_extent(rc->data_inode,
3839 &key, &rc->cluster);
3840 if (ret < 0) {
3841 err = ret;
3842 break;
3843 }
3844 }
3845 }
3846 if (trans && progress && err == -ENOSPC) {
3847 ret = btrfs_force_chunk_alloc(trans, rc->extent_root,
3848 rc->block_group->flags);
3849 if (ret == 0) {
3850 err = 0;
3851 progress = 0;
3852 goto restart;
3853 }
3854 }
3855
3856 btrfs_release_path(path);
3857 clear_extent_bits(&rc->processed_blocks, 0, (u64)-1, EXTENT_DIRTY,
3858 GFP_NOFS);
3859
3860 if (trans) {
3861 nr = trans->blocks_used;
3862 btrfs_end_transaction_throttle(trans, rc->extent_root);
3863 btrfs_btree_balance_dirty(rc->extent_root, nr);
3864 }
3865
3866 if (!err) {
3867 ret = relocate_file_extent_cluster(rc->data_inode,
3868 &rc->cluster);
3869 if (ret < 0)
3870 err = ret;
3871 }
3872
3873 rc->create_reloc_tree = 0;
3874 set_reloc_control(rc);
3875
3876 backref_cache_cleanup(&rc->backref_cache);
3877 btrfs_block_rsv_release(rc->extent_root, rc->block_rsv, (u64)-1);
3878
3879 err = prepare_to_merge(rc, err);
3880
3881 merge_reloc_roots(rc);
3882
3883 rc->merge_reloc_tree = 0;
3884 unset_reloc_control(rc);
3885 btrfs_block_rsv_release(rc->extent_root, rc->block_rsv, (u64)-1);
3886
3887 /* get rid of pinned extents */
3888 trans = btrfs_join_transaction(rc->extent_root);
3889 if (IS_ERR(trans))
3890 err = PTR_ERR(trans);
3891 else
3892 btrfs_commit_transaction(trans, rc->extent_root);
3893out_free:
3894 btrfs_free_block_rsv(rc->extent_root, rc->block_rsv);
3895 btrfs_free_path(path);
3896 return err;
3897}
3898
3899static int __insert_orphan_inode(struct btrfs_trans_handle *trans,
3900 struct btrfs_root *root, u64 objectid)
3901{
3902 struct btrfs_path *path;
3903 struct btrfs_inode_item *item;
3904 struct extent_buffer *leaf;
3905 int ret;
3906
3907 path = btrfs_alloc_path();
3908 if (!path)
3909 return -ENOMEM;
3910
3911 ret = btrfs_insert_empty_inode(trans, root, path, objectid);
3912 if (ret)
3913 goto out;
3914
3915 leaf = path->nodes[0];
3916 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_inode_item);
3917 memset_extent_buffer(leaf, 0, (unsigned long)item, sizeof(*item));
3918 btrfs_set_inode_generation(leaf, item, 1);
3919 btrfs_set_inode_size(leaf, item, 0);
3920 btrfs_set_inode_mode(leaf, item, S_IFREG | 0600);
3921 btrfs_set_inode_flags(leaf, item, BTRFS_INODE_NOCOMPRESS |
3922 BTRFS_INODE_PREALLOC);
3923 btrfs_mark_buffer_dirty(leaf);
3924 btrfs_release_path(path);
3925out:
3926 btrfs_free_path(path);
3927 return ret;
3928}
3929
3930/*
3931 * helper to create inode for data relocation.
3932 * the inode is in data relocation tree and its link count is 0
3933 */
3934static noinline_for_stack
3935struct inode *create_reloc_inode(struct btrfs_fs_info *fs_info,
3936 struct btrfs_block_group_cache *group)
3937{
3938 struct inode *inode = NULL;
3939 struct btrfs_trans_handle *trans;
3940 struct btrfs_root *root;
3941 struct btrfs_key key;
3942 unsigned long nr;
3943 u64 objectid = BTRFS_FIRST_FREE_OBJECTID;
3944 int err = 0;
3945
3946 root = read_fs_root(fs_info, BTRFS_DATA_RELOC_TREE_OBJECTID);
3947 if (IS_ERR(root))
3948 return ERR_CAST(root);
3949
3950 trans = btrfs_start_transaction(root, 6);
3951 if (IS_ERR(trans))
3952 return ERR_CAST(trans);
3953
3954 err = btrfs_find_free_objectid(root, &objectid);
3955 if (err)
3956 goto out;
3957
3958 err = __insert_orphan_inode(trans, root, objectid);
3959 BUG_ON(err);
3960
3961 key.objectid = objectid;
3962 key.type = BTRFS_INODE_ITEM_KEY;
3963 key.offset = 0;
3964 inode = btrfs_iget(root->fs_info->sb, &key, root, NULL);
3965 BUG_ON(IS_ERR(inode) || is_bad_inode(inode));
3966 BTRFS_I(inode)->index_cnt = group->key.objectid;
3967
3968 err = btrfs_orphan_add(trans, inode);
3969out:
3970 nr = trans->blocks_used;
3971 btrfs_end_transaction(trans, root);
3972 btrfs_btree_balance_dirty(root, nr);
3973 if (err) {
3974 if (inode)
3975 iput(inode);
3976 inode = ERR_PTR(err);
3977 }
3978 return inode;
3979}
3980
3981static struct reloc_control *alloc_reloc_control(void)
3982{
3983 struct reloc_control *rc;
3984
3985 rc = kzalloc(sizeof(*rc), GFP_NOFS);
3986 if (!rc)
3987 return NULL;
3988
3989 INIT_LIST_HEAD(&rc->reloc_roots);
3990 backref_cache_init(&rc->backref_cache);
3991 mapping_tree_init(&rc->reloc_root_tree);
3992 extent_io_tree_init(&rc->processed_blocks, NULL);
3993 return rc;
3994}
3995
3996/*
3997 * function to relocate all extents in a block group.
3998 */
3999int btrfs_relocate_block_group(struct btrfs_root *extent_root, u64 group_start)
4000{
4001 struct btrfs_fs_info *fs_info = extent_root->fs_info;
4002 struct reloc_control *rc;
4003 struct inode *inode;
4004 struct btrfs_path *path;
4005 int ret;
4006 int rw = 0;
4007 int err = 0;
4008
4009 rc = alloc_reloc_control();
4010 if (!rc)
4011 return -ENOMEM;
4012
4013 rc->extent_root = extent_root;
4014
4015 rc->block_group = btrfs_lookup_block_group(fs_info, group_start);
4016 BUG_ON(!rc->block_group);
4017
4018 if (!rc->block_group->ro) {
4019 ret = btrfs_set_block_group_ro(extent_root, rc->block_group);
4020 if (ret) {
4021 err = ret;
4022 goto out;
4023 }
4024 rw = 1;
4025 }
4026
4027 path = btrfs_alloc_path();
4028 if (!path) {
4029 err = -ENOMEM;
4030 goto out;
4031 }
4032
4033 inode = lookup_free_space_inode(fs_info->tree_root, rc->block_group,
4034 path);
4035 btrfs_free_path(path);
4036
4037 if (!IS_ERR(inode))
4038 ret = delete_block_group_cache(fs_info, inode, 0);
4039 else
4040 ret = PTR_ERR(inode);
4041
4042 if (ret && ret != -ENOENT) {
4043 err = ret;
4044 goto out;
4045 }
4046
4047 rc->data_inode = create_reloc_inode(fs_info, rc->block_group);
4048 if (IS_ERR(rc->data_inode)) {
4049 err = PTR_ERR(rc->data_inode);
4050 rc->data_inode = NULL;
4051 goto out;
4052 }
4053
4054 printk(KERN_INFO "btrfs: relocating block group %llu flags %llu\n",
4055 (unsigned long long)rc->block_group->key.objectid,
4056 (unsigned long long)rc->block_group->flags);
4057
4058 btrfs_start_delalloc_inodes(fs_info->tree_root, 0);
4059 btrfs_wait_ordered_extents(fs_info->tree_root, 0, 0);
4060
4061 while (1) {
4062 mutex_lock(&fs_info->cleaner_mutex);
4063
4064 btrfs_clean_old_snapshots(fs_info->tree_root);
4065 ret = relocate_block_group(rc);
4066
4067 mutex_unlock(&fs_info->cleaner_mutex);
4068 if (ret < 0) {
4069 err = ret;
4070 goto out;
4071 }
4072
4073 if (rc->extents_found == 0)
4074 break;
4075
4076 printk(KERN_INFO "btrfs: found %llu extents\n",
4077 (unsigned long long)rc->extents_found);
4078
4079 if (rc->stage == MOVE_DATA_EXTENTS && rc->found_file_extent) {
4080 btrfs_wait_ordered_range(rc->data_inode, 0, (u64)-1);
4081 invalidate_mapping_pages(rc->data_inode->i_mapping,
4082 0, -1);
4083 rc->stage = UPDATE_DATA_PTRS;
4084 }
4085 }
4086
4087 filemap_write_and_wait_range(fs_info->btree_inode->i_mapping,
4088 rc->block_group->key.objectid,
4089 rc->block_group->key.objectid +
4090 rc->block_group->key.offset - 1);
4091
4092 WARN_ON(rc->block_group->pinned > 0);
4093 WARN_ON(rc->block_group->reserved > 0);
4094 WARN_ON(btrfs_block_group_used(&rc->block_group->item) > 0);
4095out:
4096 if (err && rw)
4097 btrfs_set_block_group_rw(extent_root, rc->block_group);
4098 iput(rc->data_inode);
4099 btrfs_put_block_group(rc->block_group);
4100 kfree(rc);
4101 return err;
4102}
4103
4104static noinline_for_stack int mark_garbage_root(struct btrfs_root *root)
4105{
4106 struct btrfs_trans_handle *trans;
4107 int ret, err;
4108
4109 trans = btrfs_start_transaction(root->fs_info->tree_root, 0);
4110 if (IS_ERR(trans))
4111 return PTR_ERR(trans);
4112
4113 memset(&root->root_item.drop_progress, 0,
4114 sizeof(root->root_item.drop_progress));
4115 root->root_item.drop_level = 0;
4116 btrfs_set_root_refs(&root->root_item, 0);
4117 ret = btrfs_update_root(trans, root->fs_info->tree_root,
4118 &root->root_key, &root->root_item);
4119
4120 err = btrfs_end_transaction(trans, root->fs_info->tree_root);
4121 if (err)
4122 return err;
4123 return ret;
4124}
4125
4126/*
4127 * recover relocation interrupted by system crash.
4128 *
4129 * this function resumes merging reloc trees with corresponding fs trees.
4130 * this is important for keeping the sharing of tree blocks
4131 */
4132int btrfs_recover_relocation(struct btrfs_root *root)
4133{
4134 LIST_HEAD(reloc_roots);
4135 struct btrfs_key key;
4136 struct btrfs_root *fs_root;
4137 struct btrfs_root *reloc_root;
4138 struct btrfs_path *path;
4139 struct extent_buffer *leaf;
4140 struct reloc_control *rc = NULL;
4141 struct btrfs_trans_handle *trans;
4142 int ret;
4143 int err = 0;
4144
4145 path = btrfs_alloc_path();
4146 if (!path)
4147 return -ENOMEM;
4148 path->reada = -1;
4149
4150 key.objectid = BTRFS_TREE_RELOC_OBJECTID;
4151 key.type = BTRFS_ROOT_ITEM_KEY;
4152 key.offset = (u64)-1;
4153
4154 while (1) {
4155 ret = btrfs_search_slot(NULL, root->fs_info->tree_root, &key,
4156 path, 0, 0);
4157 if (ret < 0) {
4158 err = ret;
4159 goto out;
4160 }
4161 if (ret > 0) {
4162 if (path->slots[0] == 0)
4163 break;
4164 path->slots[0]--;
4165 }
4166 leaf = path->nodes[0];
4167 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4168 btrfs_release_path(path);
4169
4170 if (key.objectid != BTRFS_TREE_RELOC_OBJECTID ||
4171 key.type != BTRFS_ROOT_ITEM_KEY)
4172 break;
4173
4174 reloc_root = btrfs_read_fs_root_no_radix(root, &key);
4175 if (IS_ERR(reloc_root)) {
4176 err = PTR_ERR(reloc_root);
4177 goto out;
4178 }
4179
4180 list_add(&reloc_root->root_list, &reloc_roots);
4181
4182 if (btrfs_root_refs(&reloc_root->root_item) > 0) {
4183 fs_root = read_fs_root(root->fs_info,
4184 reloc_root->root_key.offset);
4185 if (IS_ERR(fs_root)) {
4186 ret = PTR_ERR(fs_root);
4187 if (ret != -ENOENT) {
4188 err = ret;
4189 goto out;
4190 }
4191 ret = mark_garbage_root(reloc_root);
4192 if (ret < 0) {
4193 err = ret;
4194 goto out;
4195 }
4196 }
4197 }
4198
4199 if (key.offset == 0)
4200 break;
4201
4202 key.offset--;
4203 }
4204 btrfs_release_path(path);
4205
4206 if (list_empty(&reloc_roots))
4207 goto out;
4208
4209 rc = alloc_reloc_control();
4210 if (!rc) {
4211 err = -ENOMEM;
4212 goto out;
4213 }
4214
4215 rc->extent_root = root->fs_info->extent_root;
4216
4217 set_reloc_control(rc);
4218
4219 trans = btrfs_join_transaction(rc->extent_root);
4220 if (IS_ERR(trans)) {
4221 unset_reloc_control(rc);
4222 err = PTR_ERR(trans);
4223 goto out_free;
4224 }
4225
4226 rc->merge_reloc_tree = 1;
4227
4228 while (!list_empty(&reloc_roots)) {
4229 reloc_root = list_entry(reloc_roots.next,
4230 struct btrfs_root, root_list);
4231 list_del(&reloc_root->root_list);
4232
4233 if (btrfs_root_refs(&reloc_root->root_item) == 0) {
4234 list_add_tail(&reloc_root->root_list,
4235 &rc->reloc_roots);
4236 continue;
4237 }
4238
4239 fs_root = read_fs_root(root->fs_info,
4240 reloc_root->root_key.offset);
4241 if (IS_ERR(fs_root)) {
4242 err = PTR_ERR(fs_root);
4243 goto out_free;
4244 }
4245
4246 err = __add_reloc_root(reloc_root);
4247 BUG_ON(err < 0); /* -ENOMEM or logic error */
4248 fs_root->reloc_root = reloc_root;
4249 }
4250
4251 err = btrfs_commit_transaction(trans, rc->extent_root);
4252 if (err)
4253 goto out_free;
4254
4255 merge_reloc_roots(rc);
4256
4257 unset_reloc_control(rc);
4258
4259 trans = btrfs_join_transaction(rc->extent_root);
4260 if (IS_ERR(trans))
4261 err = PTR_ERR(trans);
4262 else
4263 err = btrfs_commit_transaction(trans, rc->extent_root);
4264out_free:
4265 kfree(rc);
4266out:
4267 while (!list_empty(&reloc_roots)) {
4268 reloc_root = list_entry(reloc_roots.next,
4269 struct btrfs_root, root_list);
4270 list_del(&reloc_root->root_list);
4271 free_extent_buffer(reloc_root->node);
4272 free_extent_buffer(reloc_root->commit_root);
4273 kfree(reloc_root);
4274 }
4275 btrfs_free_path(path);
4276
4277 if (err == 0) {
4278 /* cleanup orphan inode in data relocation tree */
4279 fs_root = read_fs_root(root->fs_info,
4280 BTRFS_DATA_RELOC_TREE_OBJECTID);
4281 if (IS_ERR(fs_root))
4282 err = PTR_ERR(fs_root);
4283 else
4284 err = btrfs_orphan_cleanup(fs_root);
4285 }
4286 return err;
4287}
4288
4289/*
4290 * helper to add ordered checksum for data relocation.
4291 *
4292 * cloning checksum properly handles the nodatasum extents.
4293 * it also saves CPU time to re-calculate the checksum.
4294 */
4295int btrfs_reloc_clone_csums(struct inode *inode, u64 file_pos, u64 len)
4296{
4297 struct btrfs_ordered_sum *sums;
4298 struct btrfs_sector_sum *sector_sum;
4299 struct btrfs_ordered_extent *ordered;
4300 struct btrfs_root *root = BTRFS_I(inode)->root;
4301 size_t offset;
4302 int ret;
4303 u64 disk_bytenr;
4304 LIST_HEAD(list);
4305
4306 ordered = btrfs_lookup_ordered_extent(inode, file_pos);
4307 BUG_ON(ordered->file_offset != file_pos || ordered->len != len);
4308
4309 disk_bytenr = file_pos + BTRFS_I(inode)->index_cnt;
4310 ret = btrfs_lookup_csums_range(root->fs_info->csum_root, disk_bytenr,
4311 disk_bytenr + len - 1, &list, 0);
4312 if (ret)
4313 goto out;
4314
4315 while (!list_empty(&list)) {
4316 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
4317 list_del_init(&sums->list);
4318
4319 sector_sum = sums->sums;
4320 sums->bytenr = ordered->start;
4321
4322 offset = 0;
4323 while (offset < sums->len) {
4324 sector_sum->bytenr += ordered->start - disk_bytenr;
4325 sector_sum++;
4326 offset += root->sectorsize;
4327 }
4328
4329 btrfs_add_ordered_sum(inode, ordered, sums);
4330 }
4331out:
4332 btrfs_put_ordered_extent(ordered);
4333 return ret;
4334}
4335
4336void btrfs_reloc_cow_block(struct btrfs_trans_handle *trans,
4337 struct btrfs_root *root, struct extent_buffer *buf,
4338 struct extent_buffer *cow)
4339{
4340 struct reloc_control *rc;
4341 struct backref_node *node;
4342 int first_cow = 0;
4343 int level;
4344 int ret;
4345
4346 rc = root->fs_info->reloc_ctl;
4347 if (!rc)
4348 return;
4349
4350 BUG_ON(rc->stage == UPDATE_DATA_PTRS &&
4351 root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID);
4352
4353 level = btrfs_header_level(buf);
4354 if (btrfs_header_generation(buf) <=
4355 btrfs_root_last_snapshot(&root->root_item))
4356 first_cow = 1;
4357
4358 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID &&
4359 rc->create_reloc_tree) {
4360 WARN_ON(!first_cow && level == 0);
4361
4362 node = rc->backref_cache.path[level];
4363 BUG_ON(node->bytenr != buf->start &&
4364 node->new_bytenr != buf->start);
4365
4366 drop_node_buffer(node);
4367 extent_buffer_get(cow);
4368 node->eb = cow;
4369 node->new_bytenr = cow->start;
4370
4371 if (!node->pending) {
4372 list_move_tail(&node->list,
4373 &rc->backref_cache.pending[level]);
4374 node->pending = 1;
4375 }
4376
4377 if (first_cow)
4378 __mark_block_processed(rc, node);
4379
4380 if (first_cow && level > 0)
4381 rc->nodes_relocated += buf->len;
4382 }
4383
4384 if (level == 0 && first_cow && rc->stage == UPDATE_DATA_PTRS) {
4385 ret = replace_file_extents(trans, rc, root, cow);
4386 BUG_ON(ret);
4387 }
4388}
4389
4390/*
4391 * called before creating snapshot. it calculates metadata reservation
4392 * requried for relocating tree blocks in the snapshot
4393 */
4394void btrfs_reloc_pre_snapshot(struct btrfs_trans_handle *trans,
4395 struct btrfs_pending_snapshot *pending,
4396 u64 *bytes_to_reserve)
4397{
4398 struct btrfs_root *root;
4399 struct reloc_control *rc;
4400
4401 root = pending->root;
4402 if (!root->reloc_root)
4403 return;
4404
4405 rc = root->fs_info->reloc_ctl;
4406 if (!rc->merge_reloc_tree)
4407 return;
4408
4409 root = root->reloc_root;
4410 BUG_ON(btrfs_root_refs(&root->root_item) == 0);
4411 /*
4412 * relocation is in the stage of merging trees. the space
4413 * used by merging a reloc tree is twice the size of
4414 * relocated tree nodes in the worst case. half for cowing
4415 * the reloc tree, half for cowing the fs tree. the space
4416 * used by cowing the reloc tree will be freed after the
4417 * tree is dropped. if we create snapshot, cowing the fs
4418 * tree may use more space than it frees. so we need
4419 * reserve extra space.
4420 */
4421 *bytes_to_reserve += rc->nodes_relocated;
4422}
4423
4424/*
4425 * called after snapshot is created. migrate block reservation
4426 * and create reloc root for the newly created snapshot
4427 */
4428int btrfs_reloc_post_snapshot(struct btrfs_trans_handle *trans,
4429 struct btrfs_pending_snapshot *pending)
4430{
4431 struct btrfs_root *root = pending->root;
4432 struct btrfs_root *reloc_root;
4433 struct btrfs_root *new_root;
4434 struct reloc_control *rc;
4435 int ret;
4436
4437 if (!root->reloc_root)
4438 return 0;
4439
4440 rc = root->fs_info->reloc_ctl;
4441 rc->merging_rsv_size += rc->nodes_relocated;
4442
4443 if (rc->merge_reloc_tree) {
4444 ret = btrfs_block_rsv_migrate(&pending->block_rsv,
4445 rc->block_rsv,
4446 rc->nodes_relocated);
4447 if (ret)
4448 return ret;
4449 }
4450
4451 new_root = pending->snap;
4452 reloc_root = create_reloc_root(trans, root->reloc_root,
4453 new_root->root_key.objectid);
4454 if (IS_ERR(reloc_root))
4455 return PTR_ERR(reloc_root);
4456
4457 ret = __add_reloc_root(reloc_root);
4458 BUG_ON(ret < 0);
4459 new_root->reloc_root = reloc_root;
4460
4461 if (rc->create_reloc_tree)
4462 ret = clone_backref_node(trans, rc, root, reloc_root);
4463 return ret;
4464}
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * Copyright (C) 2009 Oracle. All rights reserved.
4 */
5
6#include <linux/sched.h>
7#include <linux/pagemap.h>
8#include <linux/writeback.h>
9#include <linux/blkdev.h>
10#include <linux/rbtree.h>
11#include <linux/slab.h>
12#include <linux/error-injection.h>
13#include "ctree.h"
14#include "disk-io.h"
15#include "transaction.h"
16#include "volumes.h"
17#include "locking.h"
18#include "btrfs_inode.h"
19#include "async-thread.h"
20#include "free-space-cache.h"
21#include "inode-map.h"
22#include "qgroup.h"
23#include "print-tree.h"
24#include "delalloc-space.h"
25#include "block-group.h"
26#include "backref.h"
27#include "misc.h"
28
29/*
30 * Relocation overview
31 *
32 * [What does relocation do]
33 *
34 * The objective of relocation is to relocate all extents of the target block
35 * group to other block groups.
36 * This is utilized by resize (shrink only), profile converting, compacting
37 * space, or balance routine to spread chunks over devices.
38 *
39 * Before | After
40 * ------------------------------------------------------------------
41 * BG A: 10 data extents | BG A: deleted
42 * BG B: 2 data extents | BG B: 10 data extents (2 old + 8 relocated)
43 * BG C: 1 extents | BG C: 3 data extents (1 old + 2 relocated)
44 *
45 * [How does relocation work]
46 *
47 * 1. Mark the target block group read-only
48 * New extents won't be allocated from the target block group.
49 *
50 * 2.1 Record each extent in the target block group
51 * To build a proper map of extents to be relocated.
52 *
53 * 2.2 Build data reloc tree and reloc trees
54 * Data reloc tree will contain an inode, recording all newly relocated
55 * data extents.
56 * There will be only one data reloc tree for one data block group.
57 *
58 * Reloc tree will be a special snapshot of its source tree, containing
59 * relocated tree blocks.
60 * Each tree referring to a tree block in target block group will get its
61 * reloc tree built.
62 *
63 * 2.3 Swap source tree with its corresponding reloc tree
64 * Each involved tree only refers to new extents after swap.
65 *
66 * 3. Cleanup reloc trees and data reloc tree.
67 * As old extents in the target block group are still referenced by reloc
68 * trees, we need to clean them up before really freeing the target block
69 * group.
70 *
71 * The main complexity is in steps 2.2 and 2.3.
72 *
73 * The entry point of relocation is relocate_block_group() function.
74 */
75
76#define RELOCATION_RESERVED_NODES 256
77/*
78 * map address of tree root to tree
79 */
80struct mapping_node {
81 struct {
82 struct rb_node rb_node;
83 u64 bytenr;
84 }; /* Use rb_simle_node for search/insert */
85 void *data;
86};
87
88struct mapping_tree {
89 struct rb_root rb_root;
90 spinlock_t lock;
91};
92
93/*
94 * present a tree block to process
95 */
96struct tree_block {
97 struct {
98 struct rb_node rb_node;
99 u64 bytenr;
100 }; /* Use rb_simple_node for search/insert */
101 struct btrfs_key key;
102 unsigned int level:8;
103 unsigned int key_ready:1;
104};
105
106#define MAX_EXTENTS 128
107
108struct file_extent_cluster {
109 u64 start;
110 u64 end;
111 u64 boundary[MAX_EXTENTS];
112 unsigned int nr;
113};
114
115struct reloc_control {
116 /* block group to relocate */
117 struct btrfs_block_group *block_group;
118 /* extent tree */
119 struct btrfs_root *extent_root;
120 /* inode for moving data */
121 struct inode *data_inode;
122
123 struct btrfs_block_rsv *block_rsv;
124
125 struct btrfs_backref_cache backref_cache;
126
127 struct file_extent_cluster cluster;
128 /* tree blocks have been processed */
129 struct extent_io_tree processed_blocks;
130 /* map start of tree root to corresponding reloc tree */
131 struct mapping_tree reloc_root_tree;
132 /* list of reloc trees */
133 struct list_head reloc_roots;
134 /* list of subvolume trees that get relocated */
135 struct list_head dirty_subvol_roots;
136 /* size of metadata reservation for merging reloc trees */
137 u64 merging_rsv_size;
138 /* size of relocated tree nodes */
139 u64 nodes_relocated;
140 /* reserved size for block group relocation*/
141 u64 reserved_bytes;
142
143 u64 search_start;
144 u64 extents_found;
145
146 unsigned int stage:8;
147 unsigned int create_reloc_tree:1;
148 unsigned int merge_reloc_tree:1;
149 unsigned int found_file_extent:1;
150};
151
152/* stages of data relocation */
153#define MOVE_DATA_EXTENTS 0
154#define UPDATE_DATA_PTRS 1
155
156static void mark_block_processed(struct reloc_control *rc,
157 struct btrfs_backref_node *node)
158{
159 u32 blocksize;
160
161 if (node->level == 0 ||
162 in_range(node->bytenr, rc->block_group->start,
163 rc->block_group->length)) {
164 blocksize = rc->extent_root->fs_info->nodesize;
165 set_extent_bits(&rc->processed_blocks, node->bytenr,
166 node->bytenr + blocksize - 1, EXTENT_DIRTY);
167 }
168 node->processed = 1;
169}
170
171
172static void mapping_tree_init(struct mapping_tree *tree)
173{
174 tree->rb_root = RB_ROOT;
175 spin_lock_init(&tree->lock);
176}
177
178/*
179 * walk up backref nodes until reach node presents tree root
180 */
181static struct btrfs_backref_node *walk_up_backref(
182 struct btrfs_backref_node *node,
183 struct btrfs_backref_edge *edges[], int *index)
184{
185 struct btrfs_backref_edge *edge;
186 int idx = *index;
187
188 while (!list_empty(&node->upper)) {
189 edge = list_entry(node->upper.next,
190 struct btrfs_backref_edge, list[LOWER]);
191 edges[idx++] = edge;
192 node = edge->node[UPPER];
193 }
194 BUG_ON(node->detached);
195 *index = idx;
196 return node;
197}
198
199/*
200 * walk down backref nodes to find start of next reference path
201 */
202static struct btrfs_backref_node *walk_down_backref(
203 struct btrfs_backref_edge *edges[], int *index)
204{
205 struct btrfs_backref_edge *edge;
206 struct btrfs_backref_node *lower;
207 int idx = *index;
208
209 while (idx > 0) {
210 edge = edges[idx - 1];
211 lower = edge->node[LOWER];
212 if (list_is_last(&edge->list[LOWER], &lower->upper)) {
213 idx--;
214 continue;
215 }
216 edge = list_entry(edge->list[LOWER].next,
217 struct btrfs_backref_edge, list[LOWER]);
218 edges[idx - 1] = edge;
219 *index = idx;
220 return edge->node[UPPER];
221 }
222 *index = 0;
223 return NULL;
224}
225
226static void update_backref_node(struct btrfs_backref_cache *cache,
227 struct btrfs_backref_node *node, u64 bytenr)
228{
229 struct rb_node *rb_node;
230 rb_erase(&node->rb_node, &cache->rb_root);
231 node->bytenr = bytenr;
232 rb_node = rb_simple_insert(&cache->rb_root, node->bytenr, &node->rb_node);
233 if (rb_node)
234 btrfs_backref_panic(cache->fs_info, bytenr, -EEXIST);
235}
236
237/*
238 * update backref cache after a transaction commit
239 */
240static int update_backref_cache(struct btrfs_trans_handle *trans,
241 struct btrfs_backref_cache *cache)
242{
243 struct btrfs_backref_node *node;
244 int level = 0;
245
246 if (cache->last_trans == 0) {
247 cache->last_trans = trans->transid;
248 return 0;
249 }
250
251 if (cache->last_trans == trans->transid)
252 return 0;
253
254 /*
255 * detached nodes are used to avoid unnecessary backref
256 * lookup. transaction commit changes the extent tree.
257 * so the detached nodes are no longer useful.
258 */
259 while (!list_empty(&cache->detached)) {
260 node = list_entry(cache->detached.next,
261 struct btrfs_backref_node, list);
262 btrfs_backref_cleanup_node(cache, node);
263 }
264
265 while (!list_empty(&cache->changed)) {
266 node = list_entry(cache->changed.next,
267 struct btrfs_backref_node, list);
268 list_del_init(&node->list);
269 BUG_ON(node->pending);
270 update_backref_node(cache, node, node->new_bytenr);
271 }
272
273 /*
274 * some nodes can be left in the pending list if there were
275 * errors during processing the pending nodes.
276 */
277 for (level = 0; level < BTRFS_MAX_LEVEL; level++) {
278 list_for_each_entry(node, &cache->pending[level], list) {
279 BUG_ON(!node->pending);
280 if (node->bytenr == node->new_bytenr)
281 continue;
282 update_backref_node(cache, node, node->new_bytenr);
283 }
284 }
285
286 cache->last_trans = 0;
287 return 1;
288}
289
290static bool reloc_root_is_dead(struct btrfs_root *root)
291{
292 /*
293 * Pair with set_bit/clear_bit in clean_dirty_subvols and
294 * btrfs_update_reloc_root. We need to see the updated bit before
295 * trying to access reloc_root
296 */
297 smp_rmb();
298 if (test_bit(BTRFS_ROOT_DEAD_RELOC_TREE, &root->state))
299 return true;
300 return false;
301}
302
303/*
304 * Check if this subvolume tree has valid reloc tree.
305 *
306 * Reloc tree after swap is considered dead, thus not considered as valid.
307 * This is enough for most callers, as they don't distinguish dead reloc root
308 * from no reloc root. But btrfs_should_ignore_reloc_root() below is a
309 * special case.
310 */
311static bool have_reloc_root(struct btrfs_root *root)
312{
313 if (reloc_root_is_dead(root))
314 return false;
315 if (!root->reloc_root)
316 return false;
317 return true;
318}
319
320int btrfs_should_ignore_reloc_root(struct btrfs_root *root)
321{
322 struct btrfs_root *reloc_root;
323
324 if (!test_bit(BTRFS_ROOT_SHAREABLE, &root->state))
325 return 0;
326
327 /* This root has been merged with its reloc tree, we can ignore it */
328 if (reloc_root_is_dead(root))
329 return 1;
330
331 reloc_root = root->reloc_root;
332 if (!reloc_root)
333 return 0;
334
335 if (btrfs_header_generation(reloc_root->commit_root) ==
336 root->fs_info->running_transaction->transid)
337 return 0;
338 /*
339 * if there is reloc tree and it was created in previous
340 * transaction backref lookup can find the reloc tree,
341 * so backref node for the fs tree root is useless for
342 * relocation.
343 */
344 return 1;
345}
346
347/*
348 * find reloc tree by address of tree root
349 */
350struct btrfs_root *find_reloc_root(struct btrfs_fs_info *fs_info, u64 bytenr)
351{
352 struct reloc_control *rc = fs_info->reloc_ctl;
353 struct rb_node *rb_node;
354 struct mapping_node *node;
355 struct btrfs_root *root = NULL;
356
357 ASSERT(rc);
358 spin_lock(&rc->reloc_root_tree.lock);
359 rb_node = rb_simple_search(&rc->reloc_root_tree.rb_root, bytenr);
360 if (rb_node) {
361 node = rb_entry(rb_node, struct mapping_node, rb_node);
362 root = (struct btrfs_root *)node->data;
363 }
364 spin_unlock(&rc->reloc_root_tree.lock);
365 return btrfs_grab_root(root);
366}
367
368/*
369 * For useless nodes, do two major clean ups:
370 *
371 * - Cleanup the children edges and nodes
372 * If child node is also orphan (no parent) during cleanup, then the child
373 * node will also be cleaned up.
374 *
375 * - Freeing up leaves (level 0), keeps nodes detached
376 * For nodes, the node is still cached as "detached"
377 *
378 * Return false if @node is not in the @useless_nodes list.
379 * Return true if @node is in the @useless_nodes list.
380 */
381static bool handle_useless_nodes(struct reloc_control *rc,
382 struct btrfs_backref_node *node)
383{
384 struct btrfs_backref_cache *cache = &rc->backref_cache;
385 struct list_head *useless_node = &cache->useless_node;
386 bool ret = false;
387
388 while (!list_empty(useless_node)) {
389 struct btrfs_backref_node *cur;
390
391 cur = list_first_entry(useless_node, struct btrfs_backref_node,
392 list);
393 list_del_init(&cur->list);
394
395 /* Only tree root nodes can be added to @useless_nodes */
396 ASSERT(list_empty(&cur->upper));
397
398 if (cur == node)
399 ret = true;
400
401 /* The node is the lowest node */
402 if (cur->lowest) {
403 list_del_init(&cur->lower);
404 cur->lowest = 0;
405 }
406
407 /* Cleanup the lower edges */
408 while (!list_empty(&cur->lower)) {
409 struct btrfs_backref_edge *edge;
410 struct btrfs_backref_node *lower;
411
412 edge = list_entry(cur->lower.next,
413 struct btrfs_backref_edge, list[UPPER]);
414 list_del(&edge->list[UPPER]);
415 list_del(&edge->list[LOWER]);
416 lower = edge->node[LOWER];
417 btrfs_backref_free_edge(cache, edge);
418
419 /* Child node is also orphan, queue for cleanup */
420 if (list_empty(&lower->upper))
421 list_add(&lower->list, useless_node);
422 }
423 /* Mark this block processed for relocation */
424 mark_block_processed(rc, cur);
425
426 /*
427 * Backref nodes for tree leaves are deleted from the cache.
428 * Backref nodes for upper level tree blocks are left in the
429 * cache to avoid unnecessary backref lookup.
430 */
431 if (cur->level > 0) {
432 list_add(&cur->list, &cache->detached);
433 cur->detached = 1;
434 } else {
435 rb_erase(&cur->rb_node, &cache->rb_root);
436 btrfs_backref_free_node(cache, cur);
437 }
438 }
439 return ret;
440}
441
442/*
443 * Build backref tree for a given tree block. Root of the backref tree
444 * corresponds the tree block, leaves of the backref tree correspond roots of
445 * b-trees that reference the tree block.
446 *
447 * The basic idea of this function is check backrefs of a given block to find
448 * upper level blocks that reference the block, and then check backrefs of
449 * these upper level blocks recursively. The recursion stops when tree root is
450 * reached or backrefs for the block is cached.
451 *
452 * NOTE: if we find that backrefs for a block are cached, we know backrefs for
453 * all upper level blocks that directly/indirectly reference the block are also
454 * cached.
455 */
456static noinline_for_stack struct btrfs_backref_node *build_backref_tree(
457 struct reloc_control *rc, struct btrfs_key *node_key,
458 int level, u64 bytenr)
459{
460 struct btrfs_backref_iter *iter;
461 struct btrfs_backref_cache *cache = &rc->backref_cache;
462 /* For searching parent of TREE_BLOCK_REF */
463 struct btrfs_path *path;
464 struct btrfs_backref_node *cur;
465 struct btrfs_backref_node *node = NULL;
466 struct btrfs_backref_edge *edge;
467 int ret;
468 int err = 0;
469
470 iter = btrfs_backref_iter_alloc(rc->extent_root->fs_info, GFP_NOFS);
471 if (!iter)
472 return ERR_PTR(-ENOMEM);
473 path = btrfs_alloc_path();
474 if (!path) {
475 err = -ENOMEM;
476 goto out;
477 }
478
479 node = btrfs_backref_alloc_node(cache, bytenr, level);
480 if (!node) {
481 err = -ENOMEM;
482 goto out;
483 }
484
485 node->lowest = 1;
486 cur = node;
487
488 /* Breadth-first search to build backref cache */
489 do {
490 ret = btrfs_backref_add_tree_node(cache, path, iter, node_key,
491 cur);
492 if (ret < 0) {
493 err = ret;
494 goto out;
495 }
496 edge = list_first_entry_or_null(&cache->pending_edge,
497 struct btrfs_backref_edge, list[UPPER]);
498 /*
499 * The pending list isn't empty, take the first block to
500 * process
501 */
502 if (edge) {
503 list_del_init(&edge->list[UPPER]);
504 cur = edge->node[UPPER];
505 }
506 } while (edge);
507
508 /* Finish the upper linkage of newly added edges/nodes */
509 ret = btrfs_backref_finish_upper_links(cache, node);
510 if (ret < 0) {
511 err = ret;
512 goto out;
513 }
514
515 if (handle_useless_nodes(rc, node))
516 node = NULL;
517out:
518 btrfs_backref_iter_free(iter);
519 btrfs_free_path(path);
520 if (err) {
521 btrfs_backref_error_cleanup(cache, node);
522 return ERR_PTR(err);
523 }
524 ASSERT(!node || !node->detached);
525 ASSERT(list_empty(&cache->useless_node) &&
526 list_empty(&cache->pending_edge));
527 return node;
528}
529
530/*
531 * helper to add backref node for the newly created snapshot.
532 * the backref node is created by cloning backref node that
533 * corresponds to root of source tree
534 */
535static int clone_backref_node(struct btrfs_trans_handle *trans,
536 struct reloc_control *rc,
537 struct btrfs_root *src,
538 struct btrfs_root *dest)
539{
540 struct btrfs_root *reloc_root = src->reloc_root;
541 struct btrfs_backref_cache *cache = &rc->backref_cache;
542 struct btrfs_backref_node *node = NULL;
543 struct btrfs_backref_node *new_node;
544 struct btrfs_backref_edge *edge;
545 struct btrfs_backref_edge *new_edge;
546 struct rb_node *rb_node;
547
548 if (cache->last_trans > 0)
549 update_backref_cache(trans, cache);
550
551 rb_node = rb_simple_search(&cache->rb_root, src->commit_root->start);
552 if (rb_node) {
553 node = rb_entry(rb_node, struct btrfs_backref_node, rb_node);
554 if (node->detached)
555 node = NULL;
556 else
557 BUG_ON(node->new_bytenr != reloc_root->node->start);
558 }
559
560 if (!node) {
561 rb_node = rb_simple_search(&cache->rb_root,
562 reloc_root->commit_root->start);
563 if (rb_node) {
564 node = rb_entry(rb_node, struct btrfs_backref_node,
565 rb_node);
566 BUG_ON(node->detached);
567 }
568 }
569
570 if (!node)
571 return 0;
572
573 new_node = btrfs_backref_alloc_node(cache, dest->node->start,
574 node->level);
575 if (!new_node)
576 return -ENOMEM;
577
578 new_node->lowest = node->lowest;
579 new_node->checked = 1;
580 new_node->root = btrfs_grab_root(dest);
581 ASSERT(new_node->root);
582
583 if (!node->lowest) {
584 list_for_each_entry(edge, &node->lower, list[UPPER]) {
585 new_edge = btrfs_backref_alloc_edge(cache);
586 if (!new_edge)
587 goto fail;
588
589 btrfs_backref_link_edge(new_edge, edge->node[LOWER],
590 new_node, LINK_UPPER);
591 }
592 } else {
593 list_add_tail(&new_node->lower, &cache->leaves);
594 }
595
596 rb_node = rb_simple_insert(&cache->rb_root, new_node->bytenr,
597 &new_node->rb_node);
598 if (rb_node)
599 btrfs_backref_panic(trans->fs_info, new_node->bytenr, -EEXIST);
600
601 if (!new_node->lowest) {
602 list_for_each_entry(new_edge, &new_node->lower, list[UPPER]) {
603 list_add_tail(&new_edge->list[LOWER],
604 &new_edge->node[LOWER]->upper);
605 }
606 }
607 return 0;
608fail:
609 while (!list_empty(&new_node->lower)) {
610 new_edge = list_entry(new_node->lower.next,
611 struct btrfs_backref_edge, list[UPPER]);
612 list_del(&new_edge->list[UPPER]);
613 btrfs_backref_free_edge(cache, new_edge);
614 }
615 btrfs_backref_free_node(cache, new_node);
616 return -ENOMEM;
617}
618
619/*
620 * helper to add 'address of tree root -> reloc tree' mapping
621 */
622static int __must_check __add_reloc_root(struct btrfs_root *root)
623{
624 struct btrfs_fs_info *fs_info = root->fs_info;
625 struct rb_node *rb_node;
626 struct mapping_node *node;
627 struct reloc_control *rc = fs_info->reloc_ctl;
628
629 node = kmalloc(sizeof(*node), GFP_NOFS);
630 if (!node)
631 return -ENOMEM;
632
633 node->bytenr = root->commit_root->start;
634 node->data = root;
635
636 spin_lock(&rc->reloc_root_tree.lock);
637 rb_node = rb_simple_insert(&rc->reloc_root_tree.rb_root,
638 node->bytenr, &node->rb_node);
639 spin_unlock(&rc->reloc_root_tree.lock);
640 if (rb_node) {
641 btrfs_panic(fs_info, -EEXIST,
642 "Duplicate root found for start=%llu while inserting into relocation tree",
643 node->bytenr);
644 }
645
646 list_add_tail(&root->root_list, &rc->reloc_roots);
647 return 0;
648}
649
650/*
651 * helper to delete the 'address of tree root -> reloc tree'
652 * mapping
653 */
654static void __del_reloc_root(struct btrfs_root *root)
655{
656 struct btrfs_fs_info *fs_info = root->fs_info;
657 struct rb_node *rb_node;
658 struct mapping_node *node = NULL;
659 struct reloc_control *rc = fs_info->reloc_ctl;
660 bool put_ref = false;
661
662 if (rc && root->node) {
663 spin_lock(&rc->reloc_root_tree.lock);
664 rb_node = rb_simple_search(&rc->reloc_root_tree.rb_root,
665 root->commit_root->start);
666 if (rb_node) {
667 node = rb_entry(rb_node, struct mapping_node, rb_node);
668 rb_erase(&node->rb_node, &rc->reloc_root_tree.rb_root);
669 RB_CLEAR_NODE(&node->rb_node);
670 }
671 spin_unlock(&rc->reloc_root_tree.lock);
672 if (!node)
673 return;
674 BUG_ON((struct btrfs_root *)node->data != root);
675 }
676
677 /*
678 * We only put the reloc root here if it's on the list. There's a lot
679 * of places where the pattern is to splice the rc->reloc_roots, process
680 * the reloc roots, and then add the reloc root back onto
681 * rc->reloc_roots. If we call __del_reloc_root while it's off of the
682 * list we don't want the reference being dropped, because the guy
683 * messing with the list is in charge of the reference.
684 */
685 spin_lock(&fs_info->trans_lock);
686 if (!list_empty(&root->root_list)) {
687 put_ref = true;
688 list_del_init(&root->root_list);
689 }
690 spin_unlock(&fs_info->trans_lock);
691 if (put_ref)
692 btrfs_put_root(root);
693 kfree(node);
694}
695
696/*
697 * helper to update the 'address of tree root -> reloc tree'
698 * mapping
699 */
700static int __update_reloc_root(struct btrfs_root *root)
701{
702 struct btrfs_fs_info *fs_info = root->fs_info;
703 struct rb_node *rb_node;
704 struct mapping_node *node = NULL;
705 struct reloc_control *rc = fs_info->reloc_ctl;
706
707 spin_lock(&rc->reloc_root_tree.lock);
708 rb_node = rb_simple_search(&rc->reloc_root_tree.rb_root,
709 root->commit_root->start);
710 if (rb_node) {
711 node = rb_entry(rb_node, struct mapping_node, rb_node);
712 rb_erase(&node->rb_node, &rc->reloc_root_tree.rb_root);
713 }
714 spin_unlock(&rc->reloc_root_tree.lock);
715
716 if (!node)
717 return 0;
718 BUG_ON((struct btrfs_root *)node->data != root);
719
720 spin_lock(&rc->reloc_root_tree.lock);
721 node->bytenr = root->node->start;
722 rb_node = rb_simple_insert(&rc->reloc_root_tree.rb_root,
723 node->bytenr, &node->rb_node);
724 spin_unlock(&rc->reloc_root_tree.lock);
725 if (rb_node)
726 btrfs_backref_panic(fs_info, node->bytenr, -EEXIST);
727 return 0;
728}
729
730static struct btrfs_root *create_reloc_root(struct btrfs_trans_handle *trans,
731 struct btrfs_root *root, u64 objectid)
732{
733 struct btrfs_fs_info *fs_info = root->fs_info;
734 struct btrfs_root *reloc_root;
735 struct extent_buffer *eb;
736 struct btrfs_root_item *root_item;
737 struct btrfs_key root_key;
738 int ret;
739
740 root_item = kmalloc(sizeof(*root_item), GFP_NOFS);
741 BUG_ON(!root_item);
742
743 root_key.objectid = BTRFS_TREE_RELOC_OBJECTID;
744 root_key.type = BTRFS_ROOT_ITEM_KEY;
745 root_key.offset = objectid;
746
747 if (root->root_key.objectid == objectid) {
748 u64 commit_root_gen;
749
750 /* called by btrfs_init_reloc_root */
751 ret = btrfs_copy_root(trans, root, root->commit_root, &eb,
752 BTRFS_TREE_RELOC_OBJECTID);
753 BUG_ON(ret);
754 /*
755 * Set the last_snapshot field to the generation of the commit
756 * root - like this ctree.c:btrfs_block_can_be_shared() behaves
757 * correctly (returns true) when the relocation root is created
758 * either inside the critical section of a transaction commit
759 * (through transaction.c:qgroup_account_snapshot()) and when
760 * it's created before the transaction commit is started.
761 */
762 commit_root_gen = btrfs_header_generation(root->commit_root);
763 btrfs_set_root_last_snapshot(&root->root_item, commit_root_gen);
764 } else {
765 /*
766 * called by btrfs_reloc_post_snapshot_hook.
767 * the source tree is a reloc tree, all tree blocks
768 * modified after it was created have RELOC flag
769 * set in their headers. so it's OK to not update
770 * the 'last_snapshot'.
771 */
772 ret = btrfs_copy_root(trans, root, root->node, &eb,
773 BTRFS_TREE_RELOC_OBJECTID);
774 BUG_ON(ret);
775 }
776
777 memcpy(root_item, &root->root_item, sizeof(*root_item));
778 btrfs_set_root_bytenr(root_item, eb->start);
779 btrfs_set_root_level(root_item, btrfs_header_level(eb));
780 btrfs_set_root_generation(root_item, trans->transid);
781
782 if (root->root_key.objectid == objectid) {
783 btrfs_set_root_refs(root_item, 0);
784 memset(&root_item->drop_progress, 0,
785 sizeof(struct btrfs_disk_key));
786 root_item->drop_level = 0;
787 }
788
789 btrfs_tree_unlock(eb);
790 free_extent_buffer(eb);
791
792 ret = btrfs_insert_root(trans, fs_info->tree_root,
793 &root_key, root_item);
794 BUG_ON(ret);
795 kfree(root_item);
796
797 reloc_root = btrfs_read_tree_root(fs_info->tree_root, &root_key);
798 BUG_ON(IS_ERR(reloc_root));
799 set_bit(BTRFS_ROOT_SHAREABLE, &reloc_root->state);
800 reloc_root->last_trans = trans->transid;
801 return reloc_root;
802}
803
804/*
805 * create reloc tree for a given fs tree. reloc tree is just a
806 * snapshot of the fs tree with special root objectid.
807 *
808 * The reloc_root comes out of here with two references, one for
809 * root->reloc_root, and another for being on the rc->reloc_roots list.
810 */
811int btrfs_init_reloc_root(struct btrfs_trans_handle *trans,
812 struct btrfs_root *root)
813{
814 struct btrfs_fs_info *fs_info = root->fs_info;
815 struct btrfs_root *reloc_root;
816 struct reloc_control *rc = fs_info->reloc_ctl;
817 struct btrfs_block_rsv *rsv;
818 int clear_rsv = 0;
819 int ret;
820
821 if (!rc)
822 return 0;
823
824 /*
825 * The subvolume has reloc tree but the swap is finished, no need to
826 * create/update the dead reloc tree
827 */
828 if (reloc_root_is_dead(root))
829 return 0;
830
831 /*
832 * This is subtle but important. We do not do
833 * record_root_in_transaction for reloc roots, instead we record their
834 * corresponding fs root, and then here we update the last trans for the
835 * reloc root. This means that we have to do this for the entire life
836 * of the reloc root, regardless of which stage of the relocation we are
837 * in.
838 */
839 if (root->reloc_root) {
840 reloc_root = root->reloc_root;
841 reloc_root->last_trans = trans->transid;
842 return 0;
843 }
844
845 /*
846 * We are merging reloc roots, we do not need new reloc trees. Also
847 * reloc trees never need their own reloc tree.
848 */
849 if (!rc->create_reloc_tree ||
850 root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID)
851 return 0;
852
853 if (!trans->reloc_reserved) {
854 rsv = trans->block_rsv;
855 trans->block_rsv = rc->block_rsv;
856 clear_rsv = 1;
857 }
858 reloc_root = create_reloc_root(trans, root, root->root_key.objectid);
859 if (clear_rsv)
860 trans->block_rsv = rsv;
861
862 ret = __add_reloc_root(reloc_root);
863 BUG_ON(ret < 0);
864 root->reloc_root = btrfs_grab_root(reloc_root);
865 return 0;
866}
867
868/*
869 * update root item of reloc tree
870 */
871int btrfs_update_reloc_root(struct btrfs_trans_handle *trans,
872 struct btrfs_root *root)
873{
874 struct btrfs_fs_info *fs_info = root->fs_info;
875 struct btrfs_root *reloc_root;
876 struct btrfs_root_item *root_item;
877 int ret;
878
879 if (!have_reloc_root(root))
880 goto out;
881
882 reloc_root = root->reloc_root;
883 root_item = &reloc_root->root_item;
884
885 /*
886 * We are probably ok here, but __del_reloc_root() will drop its ref of
887 * the root. We have the ref for root->reloc_root, but just in case
888 * hold it while we update the reloc root.
889 */
890 btrfs_grab_root(reloc_root);
891
892 /* root->reloc_root will stay until current relocation finished */
893 if (fs_info->reloc_ctl->merge_reloc_tree &&
894 btrfs_root_refs(root_item) == 0) {
895 set_bit(BTRFS_ROOT_DEAD_RELOC_TREE, &root->state);
896 /*
897 * Mark the tree as dead before we change reloc_root so
898 * have_reloc_root will not touch it from now on.
899 */
900 smp_wmb();
901 __del_reloc_root(reloc_root);
902 }
903
904 if (reloc_root->commit_root != reloc_root->node) {
905 __update_reloc_root(reloc_root);
906 btrfs_set_root_node(root_item, reloc_root->node);
907 free_extent_buffer(reloc_root->commit_root);
908 reloc_root->commit_root = btrfs_root_node(reloc_root);
909 }
910
911 ret = btrfs_update_root(trans, fs_info->tree_root,
912 &reloc_root->root_key, root_item);
913 BUG_ON(ret);
914 btrfs_put_root(reloc_root);
915out:
916 return 0;
917}
918
919/*
920 * helper to find first cached inode with inode number >= objectid
921 * in a subvolume
922 */
923static struct inode *find_next_inode(struct btrfs_root *root, u64 objectid)
924{
925 struct rb_node *node;
926 struct rb_node *prev;
927 struct btrfs_inode *entry;
928 struct inode *inode;
929
930 spin_lock(&root->inode_lock);
931again:
932 node = root->inode_tree.rb_node;
933 prev = NULL;
934 while (node) {
935 prev = node;
936 entry = rb_entry(node, struct btrfs_inode, rb_node);
937
938 if (objectid < btrfs_ino(entry))
939 node = node->rb_left;
940 else if (objectid > btrfs_ino(entry))
941 node = node->rb_right;
942 else
943 break;
944 }
945 if (!node) {
946 while (prev) {
947 entry = rb_entry(prev, struct btrfs_inode, rb_node);
948 if (objectid <= btrfs_ino(entry)) {
949 node = prev;
950 break;
951 }
952 prev = rb_next(prev);
953 }
954 }
955 while (node) {
956 entry = rb_entry(node, struct btrfs_inode, rb_node);
957 inode = igrab(&entry->vfs_inode);
958 if (inode) {
959 spin_unlock(&root->inode_lock);
960 return inode;
961 }
962
963 objectid = btrfs_ino(entry) + 1;
964 if (cond_resched_lock(&root->inode_lock))
965 goto again;
966
967 node = rb_next(node);
968 }
969 spin_unlock(&root->inode_lock);
970 return NULL;
971}
972
973/*
974 * get new location of data
975 */
976static int get_new_location(struct inode *reloc_inode, u64 *new_bytenr,
977 u64 bytenr, u64 num_bytes)
978{
979 struct btrfs_root *root = BTRFS_I(reloc_inode)->root;
980 struct btrfs_path *path;
981 struct btrfs_file_extent_item *fi;
982 struct extent_buffer *leaf;
983 int ret;
984
985 path = btrfs_alloc_path();
986 if (!path)
987 return -ENOMEM;
988
989 bytenr -= BTRFS_I(reloc_inode)->index_cnt;
990 ret = btrfs_lookup_file_extent(NULL, root, path,
991 btrfs_ino(BTRFS_I(reloc_inode)), bytenr, 0);
992 if (ret < 0)
993 goto out;
994 if (ret > 0) {
995 ret = -ENOENT;
996 goto out;
997 }
998
999 leaf = path->nodes[0];
1000 fi = btrfs_item_ptr(leaf, path->slots[0],
1001 struct btrfs_file_extent_item);
1002
1003 BUG_ON(btrfs_file_extent_offset(leaf, fi) ||
1004 btrfs_file_extent_compression(leaf, fi) ||
1005 btrfs_file_extent_encryption(leaf, fi) ||
1006 btrfs_file_extent_other_encoding(leaf, fi));
1007
1008 if (num_bytes != btrfs_file_extent_disk_num_bytes(leaf, fi)) {
1009 ret = -EINVAL;
1010 goto out;
1011 }
1012
1013 *new_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1014 ret = 0;
1015out:
1016 btrfs_free_path(path);
1017 return ret;
1018}
1019
1020/*
1021 * update file extent items in the tree leaf to point to
1022 * the new locations.
1023 */
1024static noinline_for_stack
1025int replace_file_extents(struct btrfs_trans_handle *trans,
1026 struct reloc_control *rc,
1027 struct btrfs_root *root,
1028 struct extent_buffer *leaf)
1029{
1030 struct btrfs_fs_info *fs_info = root->fs_info;
1031 struct btrfs_key key;
1032 struct btrfs_file_extent_item *fi;
1033 struct inode *inode = NULL;
1034 u64 parent;
1035 u64 bytenr;
1036 u64 new_bytenr = 0;
1037 u64 num_bytes;
1038 u64 end;
1039 u32 nritems;
1040 u32 i;
1041 int ret = 0;
1042 int first = 1;
1043 int dirty = 0;
1044
1045 if (rc->stage != UPDATE_DATA_PTRS)
1046 return 0;
1047
1048 /* reloc trees always use full backref */
1049 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID)
1050 parent = leaf->start;
1051 else
1052 parent = 0;
1053
1054 nritems = btrfs_header_nritems(leaf);
1055 for (i = 0; i < nritems; i++) {
1056 struct btrfs_ref ref = { 0 };
1057
1058 cond_resched();
1059 btrfs_item_key_to_cpu(leaf, &key, i);
1060 if (key.type != BTRFS_EXTENT_DATA_KEY)
1061 continue;
1062 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
1063 if (btrfs_file_extent_type(leaf, fi) ==
1064 BTRFS_FILE_EXTENT_INLINE)
1065 continue;
1066 bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1067 num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
1068 if (bytenr == 0)
1069 continue;
1070 if (!in_range(bytenr, rc->block_group->start,
1071 rc->block_group->length))
1072 continue;
1073
1074 /*
1075 * if we are modifying block in fs tree, wait for readpage
1076 * to complete and drop the extent cache
1077 */
1078 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
1079 if (first) {
1080 inode = find_next_inode(root, key.objectid);
1081 first = 0;
1082 } else if (inode && btrfs_ino(BTRFS_I(inode)) < key.objectid) {
1083 btrfs_add_delayed_iput(inode);
1084 inode = find_next_inode(root, key.objectid);
1085 }
1086 if (inode && btrfs_ino(BTRFS_I(inode)) == key.objectid) {
1087 end = key.offset +
1088 btrfs_file_extent_num_bytes(leaf, fi);
1089 WARN_ON(!IS_ALIGNED(key.offset,
1090 fs_info->sectorsize));
1091 WARN_ON(!IS_ALIGNED(end, fs_info->sectorsize));
1092 end--;
1093 ret = try_lock_extent(&BTRFS_I(inode)->io_tree,
1094 key.offset, end);
1095 if (!ret)
1096 continue;
1097
1098 btrfs_drop_extent_cache(BTRFS_I(inode),
1099 key.offset, end, 1);
1100 unlock_extent(&BTRFS_I(inode)->io_tree,
1101 key.offset, end);
1102 }
1103 }
1104
1105 ret = get_new_location(rc->data_inode, &new_bytenr,
1106 bytenr, num_bytes);
1107 if (ret) {
1108 /*
1109 * Don't have to abort since we've not changed anything
1110 * in the file extent yet.
1111 */
1112 break;
1113 }
1114
1115 btrfs_set_file_extent_disk_bytenr(leaf, fi, new_bytenr);
1116 dirty = 1;
1117
1118 key.offset -= btrfs_file_extent_offset(leaf, fi);
1119 btrfs_init_generic_ref(&ref, BTRFS_ADD_DELAYED_REF, new_bytenr,
1120 num_bytes, parent);
1121 ref.real_root = root->root_key.objectid;
1122 btrfs_init_data_ref(&ref, btrfs_header_owner(leaf),
1123 key.objectid, key.offset);
1124 ret = btrfs_inc_extent_ref(trans, &ref);
1125 if (ret) {
1126 btrfs_abort_transaction(trans, ret);
1127 break;
1128 }
1129
1130 btrfs_init_generic_ref(&ref, BTRFS_DROP_DELAYED_REF, bytenr,
1131 num_bytes, parent);
1132 ref.real_root = root->root_key.objectid;
1133 btrfs_init_data_ref(&ref, btrfs_header_owner(leaf),
1134 key.objectid, key.offset);
1135 ret = btrfs_free_extent(trans, &ref);
1136 if (ret) {
1137 btrfs_abort_transaction(trans, ret);
1138 break;
1139 }
1140 }
1141 if (dirty)
1142 btrfs_mark_buffer_dirty(leaf);
1143 if (inode)
1144 btrfs_add_delayed_iput(inode);
1145 return ret;
1146}
1147
1148static noinline_for_stack
1149int memcmp_node_keys(struct extent_buffer *eb, int slot,
1150 struct btrfs_path *path, int level)
1151{
1152 struct btrfs_disk_key key1;
1153 struct btrfs_disk_key key2;
1154 btrfs_node_key(eb, &key1, slot);
1155 btrfs_node_key(path->nodes[level], &key2, path->slots[level]);
1156 return memcmp(&key1, &key2, sizeof(key1));
1157}
1158
1159/*
1160 * try to replace tree blocks in fs tree with the new blocks
1161 * in reloc tree. tree blocks haven't been modified since the
1162 * reloc tree was create can be replaced.
1163 *
1164 * if a block was replaced, level of the block + 1 is returned.
1165 * if no block got replaced, 0 is returned. if there are other
1166 * errors, a negative error number is returned.
1167 */
1168static noinline_for_stack
1169int replace_path(struct btrfs_trans_handle *trans, struct reloc_control *rc,
1170 struct btrfs_root *dest, struct btrfs_root *src,
1171 struct btrfs_path *path, struct btrfs_key *next_key,
1172 int lowest_level, int max_level)
1173{
1174 struct btrfs_fs_info *fs_info = dest->fs_info;
1175 struct extent_buffer *eb;
1176 struct extent_buffer *parent;
1177 struct btrfs_ref ref = { 0 };
1178 struct btrfs_key key;
1179 u64 old_bytenr;
1180 u64 new_bytenr;
1181 u64 old_ptr_gen;
1182 u64 new_ptr_gen;
1183 u64 last_snapshot;
1184 u32 blocksize;
1185 int cow = 0;
1186 int level;
1187 int ret;
1188 int slot;
1189
1190 BUG_ON(src->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
1191 BUG_ON(dest->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID);
1192
1193 last_snapshot = btrfs_root_last_snapshot(&src->root_item);
1194again:
1195 slot = path->slots[lowest_level];
1196 btrfs_node_key_to_cpu(path->nodes[lowest_level], &key, slot);
1197
1198 eb = btrfs_lock_root_node(dest);
1199 btrfs_set_lock_blocking_write(eb);
1200 level = btrfs_header_level(eb);
1201
1202 if (level < lowest_level) {
1203 btrfs_tree_unlock(eb);
1204 free_extent_buffer(eb);
1205 return 0;
1206 }
1207
1208 if (cow) {
1209 ret = btrfs_cow_block(trans, dest, eb, NULL, 0, &eb);
1210 BUG_ON(ret);
1211 }
1212 btrfs_set_lock_blocking_write(eb);
1213
1214 if (next_key) {
1215 next_key->objectid = (u64)-1;
1216 next_key->type = (u8)-1;
1217 next_key->offset = (u64)-1;
1218 }
1219
1220 parent = eb;
1221 while (1) {
1222 struct btrfs_key first_key;
1223
1224 level = btrfs_header_level(parent);
1225 BUG_ON(level < lowest_level);
1226
1227 ret = btrfs_bin_search(parent, &key, &slot);
1228 if (ret < 0)
1229 break;
1230 if (ret && slot > 0)
1231 slot--;
1232
1233 if (next_key && slot + 1 < btrfs_header_nritems(parent))
1234 btrfs_node_key_to_cpu(parent, next_key, slot + 1);
1235
1236 old_bytenr = btrfs_node_blockptr(parent, slot);
1237 blocksize = fs_info->nodesize;
1238 old_ptr_gen = btrfs_node_ptr_generation(parent, slot);
1239 btrfs_node_key_to_cpu(parent, &first_key, slot);
1240
1241 if (level <= max_level) {
1242 eb = path->nodes[level];
1243 new_bytenr = btrfs_node_blockptr(eb,
1244 path->slots[level]);
1245 new_ptr_gen = btrfs_node_ptr_generation(eb,
1246 path->slots[level]);
1247 } else {
1248 new_bytenr = 0;
1249 new_ptr_gen = 0;
1250 }
1251
1252 if (WARN_ON(new_bytenr > 0 && new_bytenr == old_bytenr)) {
1253 ret = level;
1254 break;
1255 }
1256
1257 if (new_bytenr == 0 || old_ptr_gen > last_snapshot ||
1258 memcmp_node_keys(parent, slot, path, level)) {
1259 if (level <= lowest_level) {
1260 ret = 0;
1261 break;
1262 }
1263
1264 eb = read_tree_block(fs_info, old_bytenr, old_ptr_gen,
1265 level - 1, &first_key);
1266 if (IS_ERR(eb)) {
1267 ret = PTR_ERR(eb);
1268 break;
1269 } else if (!extent_buffer_uptodate(eb)) {
1270 ret = -EIO;
1271 free_extent_buffer(eb);
1272 break;
1273 }
1274 btrfs_tree_lock(eb);
1275 if (cow) {
1276 ret = btrfs_cow_block(trans, dest, eb, parent,
1277 slot, &eb);
1278 BUG_ON(ret);
1279 }
1280 btrfs_set_lock_blocking_write(eb);
1281
1282 btrfs_tree_unlock(parent);
1283 free_extent_buffer(parent);
1284
1285 parent = eb;
1286 continue;
1287 }
1288
1289 if (!cow) {
1290 btrfs_tree_unlock(parent);
1291 free_extent_buffer(parent);
1292 cow = 1;
1293 goto again;
1294 }
1295
1296 btrfs_node_key_to_cpu(path->nodes[level], &key,
1297 path->slots[level]);
1298 btrfs_release_path(path);
1299
1300 path->lowest_level = level;
1301 ret = btrfs_search_slot(trans, src, &key, path, 0, 1);
1302 path->lowest_level = 0;
1303 BUG_ON(ret);
1304
1305 /*
1306 * Info qgroup to trace both subtrees.
1307 *
1308 * We must trace both trees.
1309 * 1) Tree reloc subtree
1310 * If not traced, we will leak data numbers
1311 * 2) Fs subtree
1312 * If not traced, we will double count old data
1313 *
1314 * We don't scan the subtree right now, but only record
1315 * the swapped tree blocks.
1316 * The real subtree rescan is delayed until we have new
1317 * CoW on the subtree root node before transaction commit.
1318 */
1319 ret = btrfs_qgroup_add_swapped_blocks(trans, dest,
1320 rc->block_group, parent, slot,
1321 path->nodes[level], path->slots[level],
1322 last_snapshot);
1323 if (ret < 0)
1324 break;
1325 /*
1326 * swap blocks in fs tree and reloc tree.
1327 */
1328 btrfs_set_node_blockptr(parent, slot, new_bytenr);
1329 btrfs_set_node_ptr_generation(parent, slot, new_ptr_gen);
1330 btrfs_mark_buffer_dirty(parent);
1331
1332 btrfs_set_node_blockptr(path->nodes[level],
1333 path->slots[level], old_bytenr);
1334 btrfs_set_node_ptr_generation(path->nodes[level],
1335 path->slots[level], old_ptr_gen);
1336 btrfs_mark_buffer_dirty(path->nodes[level]);
1337
1338 btrfs_init_generic_ref(&ref, BTRFS_ADD_DELAYED_REF, old_bytenr,
1339 blocksize, path->nodes[level]->start);
1340 ref.skip_qgroup = true;
1341 btrfs_init_tree_ref(&ref, level - 1, src->root_key.objectid);
1342 ret = btrfs_inc_extent_ref(trans, &ref);
1343 BUG_ON(ret);
1344 btrfs_init_generic_ref(&ref, BTRFS_ADD_DELAYED_REF, new_bytenr,
1345 blocksize, 0);
1346 ref.skip_qgroup = true;
1347 btrfs_init_tree_ref(&ref, level - 1, dest->root_key.objectid);
1348 ret = btrfs_inc_extent_ref(trans, &ref);
1349 BUG_ON(ret);
1350
1351 btrfs_init_generic_ref(&ref, BTRFS_DROP_DELAYED_REF, new_bytenr,
1352 blocksize, path->nodes[level]->start);
1353 btrfs_init_tree_ref(&ref, level - 1, src->root_key.objectid);
1354 ref.skip_qgroup = true;
1355 ret = btrfs_free_extent(trans, &ref);
1356 BUG_ON(ret);
1357
1358 btrfs_init_generic_ref(&ref, BTRFS_DROP_DELAYED_REF, old_bytenr,
1359 blocksize, 0);
1360 btrfs_init_tree_ref(&ref, level - 1, dest->root_key.objectid);
1361 ref.skip_qgroup = true;
1362 ret = btrfs_free_extent(trans, &ref);
1363 BUG_ON(ret);
1364
1365 btrfs_unlock_up_safe(path, 0);
1366
1367 ret = level;
1368 break;
1369 }
1370 btrfs_tree_unlock(parent);
1371 free_extent_buffer(parent);
1372 return ret;
1373}
1374
1375/*
1376 * helper to find next relocated block in reloc tree
1377 */
1378static noinline_for_stack
1379int walk_up_reloc_tree(struct btrfs_root *root, struct btrfs_path *path,
1380 int *level)
1381{
1382 struct extent_buffer *eb;
1383 int i;
1384 u64 last_snapshot;
1385 u32 nritems;
1386
1387 last_snapshot = btrfs_root_last_snapshot(&root->root_item);
1388
1389 for (i = 0; i < *level; i++) {
1390 free_extent_buffer(path->nodes[i]);
1391 path->nodes[i] = NULL;
1392 }
1393
1394 for (i = *level; i < BTRFS_MAX_LEVEL && path->nodes[i]; i++) {
1395 eb = path->nodes[i];
1396 nritems = btrfs_header_nritems(eb);
1397 while (path->slots[i] + 1 < nritems) {
1398 path->slots[i]++;
1399 if (btrfs_node_ptr_generation(eb, path->slots[i]) <=
1400 last_snapshot)
1401 continue;
1402
1403 *level = i;
1404 return 0;
1405 }
1406 free_extent_buffer(path->nodes[i]);
1407 path->nodes[i] = NULL;
1408 }
1409 return 1;
1410}
1411
1412/*
1413 * walk down reloc tree to find relocated block of lowest level
1414 */
1415static noinline_for_stack
1416int walk_down_reloc_tree(struct btrfs_root *root, struct btrfs_path *path,
1417 int *level)
1418{
1419 struct btrfs_fs_info *fs_info = root->fs_info;
1420 struct extent_buffer *eb = NULL;
1421 int i;
1422 u64 bytenr;
1423 u64 ptr_gen = 0;
1424 u64 last_snapshot;
1425 u32 nritems;
1426
1427 last_snapshot = btrfs_root_last_snapshot(&root->root_item);
1428
1429 for (i = *level; i > 0; i--) {
1430 struct btrfs_key first_key;
1431
1432 eb = path->nodes[i];
1433 nritems = btrfs_header_nritems(eb);
1434 while (path->slots[i] < nritems) {
1435 ptr_gen = btrfs_node_ptr_generation(eb, path->slots[i]);
1436 if (ptr_gen > last_snapshot)
1437 break;
1438 path->slots[i]++;
1439 }
1440 if (path->slots[i] >= nritems) {
1441 if (i == *level)
1442 break;
1443 *level = i + 1;
1444 return 0;
1445 }
1446 if (i == 1) {
1447 *level = i;
1448 return 0;
1449 }
1450
1451 bytenr = btrfs_node_blockptr(eb, path->slots[i]);
1452 btrfs_node_key_to_cpu(eb, &first_key, path->slots[i]);
1453 eb = read_tree_block(fs_info, bytenr, ptr_gen, i - 1,
1454 &first_key);
1455 if (IS_ERR(eb)) {
1456 return PTR_ERR(eb);
1457 } else if (!extent_buffer_uptodate(eb)) {
1458 free_extent_buffer(eb);
1459 return -EIO;
1460 }
1461 BUG_ON(btrfs_header_level(eb) != i - 1);
1462 path->nodes[i - 1] = eb;
1463 path->slots[i - 1] = 0;
1464 }
1465 return 1;
1466}
1467
1468/*
1469 * invalidate extent cache for file extents whose key in range of
1470 * [min_key, max_key)
1471 */
1472static int invalidate_extent_cache(struct btrfs_root *root,
1473 struct btrfs_key *min_key,
1474 struct btrfs_key *max_key)
1475{
1476 struct btrfs_fs_info *fs_info = root->fs_info;
1477 struct inode *inode = NULL;
1478 u64 objectid;
1479 u64 start, end;
1480 u64 ino;
1481
1482 objectid = min_key->objectid;
1483 while (1) {
1484 cond_resched();
1485 iput(inode);
1486
1487 if (objectid > max_key->objectid)
1488 break;
1489
1490 inode = find_next_inode(root, objectid);
1491 if (!inode)
1492 break;
1493 ino = btrfs_ino(BTRFS_I(inode));
1494
1495 if (ino > max_key->objectid) {
1496 iput(inode);
1497 break;
1498 }
1499
1500 objectid = ino + 1;
1501 if (!S_ISREG(inode->i_mode))
1502 continue;
1503
1504 if (unlikely(min_key->objectid == ino)) {
1505 if (min_key->type > BTRFS_EXTENT_DATA_KEY)
1506 continue;
1507 if (min_key->type < BTRFS_EXTENT_DATA_KEY)
1508 start = 0;
1509 else {
1510 start = min_key->offset;
1511 WARN_ON(!IS_ALIGNED(start, fs_info->sectorsize));
1512 }
1513 } else {
1514 start = 0;
1515 }
1516
1517 if (unlikely(max_key->objectid == ino)) {
1518 if (max_key->type < BTRFS_EXTENT_DATA_KEY)
1519 continue;
1520 if (max_key->type > BTRFS_EXTENT_DATA_KEY) {
1521 end = (u64)-1;
1522 } else {
1523 if (max_key->offset == 0)
1524 continue;
1525 end = max_key->offset;
1526 WARN_ON(!IS_ALIGNED(end, fs_info->sectorsize));
1527 end--;
1528 }
1529 } else {
1530 end = (u64)-1;
1531 }
1532
1533 /* the lock_extent waits for readpage to complete */
1534 lock_extent(&BTRFS_I(inode)->io_tree, start, end);
1535 btrfs_drop_extent_cache(BTRFS_I(inode), start, end, 1);
1536 unlock_extent(&BTRFS_I(inode)->io_tree, start, end);
1537 }
1538 return 0;
1539}
1540
1541static int find_next_key(struct btrfs_path *path, int level,
1542 struct btrfs_key *key)
1543
1544{
1545 while (level < BTRFS_MAX_LEVEL) {
1546 if (!path->nodes[level])
1547 break;
1548 if (path->slots[level] + 1 <
1549 btrfs_header_nritems(path->nodes[level])) {
1550 btrfs_node_key_to_cpu(path->nodes[level], key,
1551 path->slots[level] + 1);
1552 return 0;
1553 }
1554 level++;
1555 }
1556 return 1;
1557}
1558
1559/*
1560 * Insert current subvolume into reloc_control::dirty_subvol_roots
1561 */
1562static void insert_dirty_subvol(struct btrfs_trans_handle *trans,
1563 struct reloc_control *rc,
1564 struct btrfs_root *root)
1565{
1566 struct btrfs_root *reloc_root = root->reloc_root;
1567 struct btrfs_root_item *reloc_root_item;
1568
1569 /* @root must be a subvolume tree root with a valid reloc tree */
1570 ASSERT(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
1571 ASSERT(reloc_root);
1572
1573 reloc_root_item = &reloc_root->root_item;
1574 memset(&reloc_root_item->drop_progress, 0,
1575 sizeof(reloc_root_item->drop_progress));
1576 reloc_root_item->drop_level = 0;
1577 btrfs_set_root_refs(reloc_root_item, 0);
1578 btrfs_update_reloc_root(trans, root);
1579
1580 if (list_empty(&root->reloc_dirty_list)) {
1581 btrfs_grab_root(root);
1582 list_add_tail(&root->reloc_dirty_list, &rc->dirty_subvol_roots);
1583 }
1584}
1585
1586static int clean_dirty_subvols(struct reloc_control *rc)
1587{
1588 struct btrfs_root *root;
1589 struct btrfs_root *next;
1590 int ret = 0;
1591 int ret2;
1592
1593 list_for_each_entry_safe(root, next, &rc->dirty_subvol_roots,
1594 reloc_dirty_list) {
1595 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
1596 /* Merged subvolume, cleanup its reloc root */
1597 struct btrfs_root *reloc_root = root->reloc_root;
1598
1599 list_del_init(&root->reloc_dirty_list);
1600 root->reloc_root = NULL;
1601 /*
1602 * Need barrier to ensure clear_bit() only happens after
1603 * root->reloc_root = NULL. Pairs with have_reloc_root.
1604 */
1605 smp_wmb();
1606 clear_bit(BTRFS_ROOT_DEAD_RELOC_TREE, &root->state);
1607 if (reloc_root) {
1608 /*
1609 * btrfs_drop_snapshot drops our ref we hold for
1610 * ->reloc_root. If it fails however we must
1611 * drop the ref ourselves.
1612 */
1613 ret2 = btrfs_drop_snapshot(reloc_root, 0, 1);
1614 if (ret2 < 0) {
1615 btrfs_put_root(reloc_root);
1616 if (!ret)
1617 ret = ret2;
1618 }
1619 }
1620 btrfs_put_root(root);
1621 } else {
1622 /* Orphan reloc tree, just clean it up */
1623 ret2 = btrfs_drop_snapshot(root, 0, 1);
1624 if (ret2 < 0) {
1625 btrfs_put_root(root);
1626 if (!ret)
1627 ret = ret2;
1628 }
1629 }
1630 }
1631 return ret;
1632}
1633
1634/*
1635 * merge the relocated tree blocks in reloc tree with corresponding
1636 * fs tree.
1637 */
1638static noinline_for_stack int merge_reloc_root(struct reloc_control *rc,
1639 struct btrfs_root *root)
1640{
1641 struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
1642 struct btrfs_key key;
1643 struct btrfs_key next_key;
1644 struct btrfs_trans_handle *trans = NULL;
1645 struct btrfs_root *reloc_root;
1646 struct btrfs_root_item *root_item;
1647 struct btrfs_path *path;
1648 struct extent_buffer *leaf;
1649 int level;
1650 int max_level;
1651 int replaced = 0;
1652 int ret;
1653 int err = 0;
1654 u32 min_reserved;
1655
1656 path = btrfs_alloc_path();
1657 if (!path)
1658 return -ENOMEM;
1659 path->reada = READA_FORWARD;
1660
1661 reloc_root = root->reloc_root;
1662 root_item = &reloc_root->root_item;
1663
1664 if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
1665 level = btrfs_root_level(root_item);
1666 atomic_inc(&reloc_root->node->refs);
1667 path->nodes[level] = reloc_root->node;
1668 path->slots[level] = 0;
1669 } else {
1670 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
1671
1672 level = root_item->drop_level;
1673 BUG_ON(level == 0);
1674 path->lowest_level = level;
1675 ret = btrfs_search_slot(NULL, reloc_root, &key, path, 0, 0);
1676 path->lowest_level = 0;
1677 if (ret < 0) {
1678 btrfs_free_path(path);
1679 return ret;
1680 }
1681
1682 btrfs_node_key_to_cpu(path->nodes[level], &next_key,
1683 path->slots[level]);
1684 WARN_ON(memcmp(&key, &next_key, sizeof(key)));
1685
1686 btrfs_unlock_up_safe(path, 0);
1687 }
1688
1689 /*
1690 * In merge_reloc_root(), we modify the upper level pointer to swap the
1691 * tree blocks between reloc tree and subvolume tree. Thus for tree
1692 * block COW, we COW at most from level 1 to root level for each tree.
1693 *
1694 * Thus the needed metadata size is at most root_level * nodesize,
1695 * and * 2 since we have two trees to COW.
1696 */
1697 min_reserved = fs_info->nodesize * btrfs_root_level(root_item) * 2;
1698 memset(&next_key, 0, sizeof(next_key));
1699
1700 while (1) {
1701 ret = btrfs_block_rsv_refill(root, rc->block_rsv, min_reserved,
1702 BTRFS_RESERVE_FLUSH_LIMIT);
1703 if (ret) {
1704 err = ret;
1705 goto out;
1706 }
1707 trans = btrfs_start_transaction(root, 0);
1708 if (IS_ERR(trans)) {
1709 err = PTR_ERR(trans);
1710 trans = NULL;
1711 goto out;
1712 }
1713
1714 /*
1715 * At this point we no longer have a reloc_control, so we can't
1716 * depend on btrfs_init_reloc_root to update our last_trans.
1717 *
1718 * But that's ok, we started the trans handle on our
1719 * corresponding fs_root, which means it's been added to the
1720 * dirty list. At commit time we'll still call
1721 * btrfs_update_reloc_root() and update our root item
1722 * appropriately.
1723 */
1724 reloc_root->last_trans = trans->transid;
1725 trans->block_rsv = rc->block_rsv;
1726
1727 replaced = 0;
1728 max_level = level;
1729
1730 ret = walk_down_reloc_tree(reloc_root, path, &level);
1731 if (ret < 0) {
1732 err = ret;
1733 goto out;
1734 }
1735 if (ret > 0)
1736 break;
1737
1738 if (!find_next_key(path, level, &key) &&
1739 btrfs_comp_cpu_keys(&next_key, &key) >= 0) {
1740 ret = 0;
1741 } else {
1742 ret = replace_path(trans, rc, root, reloc_root, path,
1743 &next_key, level, max_level);
1744 }
1745 if (ret < 0) {
1746 err = ret;
1747 goto out;
1748 }
1749
1750 if (ret > 0) {
1751 level = ret;
1752 btrfs_node_key_to_cpu(path->nodes[level], &key,
1753 path->slots[level]);
1754 replaced = 1;
1755 }
1756
1757 ret = walk_up_reloc_tree(reloc_root, path, &level);
1758 if (ret > 0)
1759 break;
1760
1761 BUG_ON(level == 0);
1762 /*
1763 * save the merging progress in the drop_progress.
1764 * this is OK since root refs == 1 in this case.
1765 */
1766 btrfs_node_key(path->nodes[level], &root_item->drop_progress,
1767 path->slots[level]);
1768 root_item->drop_level = level;
1769
1770 btrfs_end_transaction_throttle(trans);
1771 trans = NULL;
1772
1773 btrfs_btree_balance_dirty(fs_info);
1774
1775 if (replaced && rc->stage == UPDATE_DATA_PTRS)
1776 invalidate_extent_cache(root, &key, &next_key);
1777 }
1778
1779 /*
1780 * handle the case only one block in the fs tree need to be
1781 * relocated and the block is tree root.
1782 */
1783 leaf = btrfs_lock_root_node(root);
1784 ret = btrfs_cow_block(trans, root, leaf, NULL, 0, &leaf);
1785 btrfs_tree_unlock(leaf);
1786 free_extent_buffer(leaf);
1787 if (ret < 0)
1788 err = ret;
1789out:
1790 btrfs_free_path(path);
1791
1792 if (err == 0)
1793 insert_dirty_subvol(trans, rc, root);
1794
1795 if (trans)
1796 btrfs_end_transaction_throttle(trans);
1797
1798 btrfs_btree_balance_dirty(fs_info);
1799
1800 if (replaced && rc->stage == UPDATE_DATA_PTRS)
1801 invalidate_extent_cache(root, &key, &next_key);
1802
1803 return err;
1804}
1805
1806static noinline_for_stack
1807int prepare_to_merge(struct reloc_control *rc, int err)
1808{
1809 struct btrfs_root *root = rc->extent_root;
1810 struct btrfs_fs_info *fs_info = root->fs_info;
1811 struct btrfs_root *reloc_root;
1812 struct btrfs_trans_handle *trans;
1813 LIST_HEAD(reloc_roots);
1814 u64 num_bytes = 0;
1815 int ret;
1816
1817 mutex_lock(&fs_info->reloc_mutex);
1818 rc->merging_rsv_size += fs_info->nodesize * (BTRFS_MAX_LEVEL - 1) * 2;
1819 rc->merging_rsv_size += rc->nodes_relocated * 2;
1820 mutex_unlock(&fs_info->reloc_mutex);
1821
1822again:
1823 if (!err) {
1824 num_bytes = rc->merging_rsv_size;
1825 ret = btrfs_block_rsv_add(root, rc->block_rsv, num_bytes,
1826 BTRFS_RESERVE_FLUSH_ALL);
1827 if (ret)
1828 err = ret;
1829 }
1830
1831 trans = btrfs_join_transaction(rc->extent_root);
1832 if (IS_ERR(trans)) {
1833 if (!err)
1834 btrfs_block_rsv_release(fs_info, rc->block_rsv,
1835 num_bytes, NULL);
1836 return PTR_ERR(trans);
1837 }
1838
1839 if (!err) {
1840 if (num_bytes != rc->merging_rsv_size) {
1841 btrfs_end_transaction(trans);
1842 btrfs_block_rsv_release(fs_info, rc->block_rsv,
1843 num_bytes, NULL);
1844 goto again;
1845 }
1846 }
1847
1848 rc->merge_reloc_tree = 1;
1849
1850 while (!list_empty(&rc->reloc_roots)) {
1851 reloc_root = list_entry(rc->reloc_roots.next,
1852 struct btrfs_root, root_list);
1853 list_del_init(&reloc_root->root_list);
1854
1855 root = btrfs_get_fs_root(fs_info, reloc_root->root_key.offset,
1856 false);
1857 BUG_ON(IS_ERR(root));
1858 BUG_ON(root->reloc_root != reloc_root);
1859
1860 /*
1861 * set reference count to 1, so btrfs_recover_relocation
1862 * knows it should resumes merging
1863 */
1864 if (!err)
1865 btrfs_set_root_refs(&reloc_root->root_item, 1);
1866 btrfs_update_reloc_root(trans, root);
1867
1868 list_add(&reloc_root->root_list, &reloc_roots);
1869 btrfs_put_root(root);
1870 }
1871
1872 list_splice(&reloc_roots, &rc->reloc_roots);
1873
1874 if (!err)
1875 btrfs_commit_transaction(trans);
1876 else
1877 btrfs_end_transaction(trans);
1878 return err;
1879}
1880
1881static noinline_for_stack
1882void free_reloc_roots(struct list_head *list)
1883{
1884 struct btrfs_root *reloc_root, *tmp;
1885
1886 list_for_each_entry_safe(reloc_root, tmp, list, root_list)
1887 __del_reloc_root(reloc_root);
1888}
1889
1890static noinline_for_stack
1891void merge_reloc_roots(struct reloc_control *rc)
1892{
1893 struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
1894 struct btrfs_root *root;
1895 struct btrfs_root *reloc_root;
1896 LIST_HEAD(reloc_roots);
1897 int found = 0;
1898 int ret = 0;
1899again:
1900 root = rc->extent_root;
1901
1902 /*
1903 * this serializes us with btrfs_record_root_in_transaction,
1904 * we have to make sure nobody is in the middle of
1905 * adding their roots to the list while we are
1906 * doing this splice
1907 */
1908 mutex_lock(&fs_info->reloc_mutex);
1909 list_splice_init(&rc->reloc_roots, &reloc_roots);
1910 mutex_unlock(&fs_info->reloc_mutex);
1911
1912 while (!list_empty(&reloc_roots)) {
1913 found = 1;
1914 reloc_root = list_entry(reloc_roots.next,
1915 struct btrfs_root, root_list);
1916
1917 root = btrfs_get_fs_root(fs_info, reloc_root->root_key.offset,
1918 false);
1919 if (btrfs_root_refs(&reloc_root->root_item) > 0) {
1920 BUG_ON(IS_ERR(root));
1921 BUG_ON(root->reloc_root != reloc_root);
1922 ret = merge_reloc_root(rc, root);
1923 btrfs_put_root(root);
1924 if (ret) {
1925 if (list_empty(&reloc_root->root_list))
1926 list_add_tail(&reloc_root->root_list,
1927 &reloc_roots);
1928 goto out;
1929 }
1930 } else {
1931 if (!IS_ERR(root)) {
1932 if (root->reloc_root == reloc_root) {
1933 root->reloc_root = NULL;
1934 btrfs_put_root(reloc_root);
1935 }
1936 clear_bit(BTRFS_ROOT_DEAD_RELOC_TREE,
1937 &root->state);
1938 btrfs_put_root(root);
1939 }
1940
1941 list_del_init(&reloc_root->root_list);
1942 /* Don't forget to queue this reloc root for cleanup */
1943 list_add_tail(&reloc_root->reloc_dirty_list,
1944 &rc->dirty_subvol_roots);
1945 }
1946 }
1947
1948 if (found) {
1949 found = 0;
1950 goto again;
1951 }
1952out:
1953 if (ret) {
1954 btrfs_handle_fs_error(fs_info, ret, NULL);
1955 free_reloc_roots(&reloc_roots);
1956
1957 /* new reloc root may be added */
1958 mutex_lock(&fs_info->reloc_mutex);
1959 list_splice_init(&rc->reloc_roots, &reloc_roots);
1960 mutex_unlock(&fs_info->reloc_mutex);
1961 free_reloc_roots(&reloc_roots);
1962 }
1963
1964 /*
1965 * We used to have
1966 *
1967 * BUG_ON(!RB_EMPTY_ROOT(&rc->reloc_root_tree.rb_root));
1968 *
1969 * here, but it's wrong. If we fail to start the transaction in
1970 * prepare_to_merge() we will have only 0 ref reloc roots, none of which
1971 * have actually been removed from the reloc_root_tree rb tree. This is
1972 * fine because we're bailing here, and we hold a reference on the root
1973 * for the list that holds it, so these roots will be cleaned up when we
1974 * do the reloc_dirty_list afterwards. Meanwhile the root->reloc_root
1975 * will be cleaned up on unmount.
1976 *
1977 * The remaining nodes will be cleaned up by free_reloc_control.
1978 */
1979}
1980
1981static void free_block_list(struct rb_root *blocks)
1982{
1983 struct tree_block *block;
1984 struct rb_node *rb_node;
1985 while ((rb_node = rb_first(blocks))) {
1986 block = rb_entry(rb_node, struct tree_block, rb_node);
1987 rb_erase(rb_node, blocks);
1988 kfree(block);
1989 }
1990}
1991
1992static int record_reloc_root_in_trans(struct btrfs_trans_handle *trans,
1993 struct btrfs_root *reloc_root)
1994{
1995 struct btrfs_fs_info *fs_info = reloc_root->fs_info;
1996 struct btrfs_root *root;
1997 int ret;
1998
1999 if (reloc_root->last_trans == trans->transid)
2000 return 0;
2001
2002 root = btrfs_get_fs_root(fs_info, reloc_root->root_key.offset, false);
2003 BUG_ON(IS_ERR(root));
2004 BUG_ON(root->reloc_root != reloc_root);
2005 ret = btrfs_record_root_in_trans(trans, root);
2006 btrfs_put_root(root);
2007
2008 return ret;
2009}
2010
2011static noinline_for_stack
2012struct btrfs_root *select_reloc_root(struct btrfs_trans_handle *trans,
2013 struct reloc_control *rc,
2014 struct btrfs_backref_node *node,
2015 struct btrfs_backref_edge *edges[])
2016{
2017 struct btrfs_backref_node *next;
2018 struct btrfs_root *root;
2019 int index = 0;
2020
2021 next = node;
2022 while (1) {
2023 cond_resched();
2024 next = walk_up_backref(next, edges, &index);
2025 root = next->root;
2026 BUG_ON(!root);
2027 BUG_ON(!test_bit(BTRFS_ROOT_SHAREABLE, &root->state));
2028
2029 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) {
2030 record_reloc_root_in_trans(trans, root);
2031 break;
2032 }
2033
2034 btrfs_record_root_in_trans(trans, root);
2035 root = root->reloc_root;
2036
2037 if (next->new_bytenr != root->node->start) {
2038 BUG_ON(next->new_bytenr);
2039 BUG_ON(!list_empty(&next->list));
2040 next->new_bytenr = root->node->start;
2041 btrfs_put_root(next->root);
2042 next->root = btrfs_grab_root(root);
2043 ASSERT(next->root);
2044 list_add_tail(&next->list,
2045 &rc->backref_cache.changed);
2046 mark_block_processed(rc, next);
2047 break;
2048 }
2049
2050 WARN_ON(1);
2051 root = NULL;
2052 next = walk_down_backref(edges, &index);
2053 if (!next || next->level <= node->level)
2054 break;
2055 }
2056 if (!root)
2057 return NULL;
2058
2059 next = node;
2060 /* setup backref node path for btrfs_reloc_cow_block */
2061 while (1) {
2062 rc->backref_cache.path[next->level] = next;
2063 if (--index < 0)
2064 break;
2065 next = edges[index]->node[UPPER];
2066 }
2067 return root;
2068}
2069
2070/*
2071 * Select a tree root for relocation.
2072 *
2073 * Return NULL if the block is not shareable. We should use do_relocation() in
2074 * this case.
2075 *
2076 * Return a tree root pointer if the block is shareable.
2077 * Return -ENOENT if the block is root of reloc tree.
2078 */
2079static noinline_for_stack
2080struct btrfs_root *select_one_root(struct btrfs_backref_node *node)
2081{
2082 struct btrfs_backref_node *next;
2083 struct btrfs_root *root;
2084 struct btrfs_root *fs_root = NULL;
2085 struct btrfs_backref_edge *edges[BTRFS_MAX_LEVEL - 1];
2086 int index = 0;
2087
2088 next = node;
2089 while (1) {
2090 cond_resched();
2091 next = walk_up_backref(next, edges, &index);
2092 root = next->root;
2093 BUG_ON(!root);
2094
2095 /* No other choice for non-shareable tree */
2096 if (!test_bit(BTRFS_ROOT_SHAREABLE, &root->state))
2097 return root;
2098
2099 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID)
2100 fs_root = root;
2101
2102 if (next != node)
2103 return NULL;
2104
2105 next = walk_down_backref(edges, &index);
2106 if (!next || next->level <= node->level)
2107 break;
2108 }
2109
2110 if (!fs_root)
2111 return ERR_PTR(-ENOENT);
2112 return fs_root;
2113}
2114
2115static noinline_for_stack
2116u64 calcu_metadata_size(struct reloc_control *rc,
2117 struct btrfs_backref_node *node, int reserve)
2118{
2119 struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
2120 struct btrfs_backref_node *next = node;
2121 struct btrfs_backref_edge *edge;
2122 struct btrfs_backref_edge *edges[BTRFS_MAX_LEVEL - 1];
2123 u64 num_bytes = 0;
2124 int index = 0;
2125
2126 BUG_ON(reserve && node->processed);
2127
2128 while (next) {
2129 cond_resched();
2130 while (1) {
2131 if (next->processed && (reserve || next != node))
2132 break;
2133
2134 num_bytes += fs_info->nodesize;
2135
2136 if (list_empty(&next->upper))
2137 break;
2138
2139 edge = list_entry(next->upper.next,
2140 struct btrfs_backref_edge, list[LOWER]);
2141 edges[index++] = edge;
2142 next = edge->node[UPPER];
2143 }
2144 next = walk_down_backref(edges, &index);
2145 }
2146 return num_bytes;
2147}
2148
2149static int reserve_metadata_space(struct btrfs_trans_handle *trans,
2150 struct reloc_control *rc,
2151 struct btrfs_backref_node *node)
2152{
2153 struct btrfs_root *root = rc->extent_root;
2154 struct btrfs_fs_info *fs_info = root->fs_info;
2155 u64 num_bytes;
2156 int ret;
2157 u64 tmp;
2158
2159 num_bytes = calcu_metadata_size(rc, node, 1) * 2;
2160
2161 trans->block_rsv = rc->block_rsv;
2162 rc->reserved_bytes += num_bytes;
2163
2164 /*
2165 * We are under a transaction here so we can only do limited flushing.
2166 * If we get an enospc just kick back -EAGAIN so we know to drop the
2167 * transaction and try to refill when we can flush all the things.
2168 */
2169 ret = btrfs_block_rsv_refill(root, rc->block_rsv, num_bytes,
2170 BTRFS_RESERVE_FLUSH_LIMIT);
2171 if (ret) {
2172 tmp = fs_info->nodesize * RELOCATION_RESERVED_NODES;
2173 while (tmp <= rc->reserved_bytes)
2174 tmp <<= 1;
2175 /*
2176 * only one thread can access block_rsv at this point,
2177 * so we don't need hold lock to protect block_rsv.
2178 * we expand more reservation size here to allow enough
2179 * space for relocation and we will return earlier in
2180 * enospc case.
2181 */
2182 rc->block_rsv->size = tmp + fs_info->nodesize *
2183 RELOCATION_RESERVED_NODES;
2184 return -EAGAIN;
2185 }
2186
2187 return 0;
2188}
2189
2190/*
2191 * relocate a block tree, and then update pointers in upper level
2192 * blocks that reference the block to point to the new location.
2193 *
2194 * if called by link_to_upper, the block has already been relocated.
2195 * in that case this function just updates pointers.
2196 */
2197static int do_relocation(struct btrfs_trans_handle *trans,
2198 struct reloc_control *rc,
2199 struct btrfs_backref_node *node,
2200 struct btrfs_key *key,
2201 struct btrfs_path *path, int lowest)
2202{
2203 struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
2204 struct btrfs_backref_node *upper;
2205 struct btrfs_backref_edge *edge;
2206 struct btrfs_backref_edge *edges[BTRFS_MAX_LEVEL - 1];
2207 struct btrfs_root *root;
2208 struct extent_buffer *eb;
2209 u32 blocksize;
2210 u64 bytenr;
2211 u64 generation;
2212 int slot;
2213 int ret;
2214 int err = 0;
2215
2216 BUG_ON(lowest && node->eb);
2217
2218 path->lowest_level = node->level + 1;
2219 rc->backref_cache.path[node->level] = node;
2220 list_for_each_entry(edge, &node->upper, list[LOWER]) {
2221 struct btrfs_key first_key;
2222 struct btrfs_ref ref = { 0 };
2223
2224 cond_resched();
2225
2226 upper = edge->node[UPPER];
2227 root = select_reloc_root(trans, rc, upper, edges);
2228 BUG_ON(!root);
2229
2230 if (upper->eb && !upper->locked) {
2231 if (!lowest) {
2232 ret = btrfs_bin_search(upper->eb, key, &slot);
2233 if (ret < 0) {
2234 err = ret;
2235 goto next;
2236 }
2237 BUG_ON(ret);
2238 bytenr = btrfs_node_blockptr(upper->eb, slot);
2239 if (node->eb->start == bytenr)
2240 goto next;
2241 }
2242 btrfs_backref_drop_node_buffer(upper);
2243 }
2244
2245 if (!upper->eb) {
2246 ret = btrfs_search_slot(trans, root, key, path, 0, 1);
2247 if (ret) {
2248 if (ret < 0)
2249 err = ret;
2250 else
2251 err = -ENOENT;
2252
2253 btrfs_release_path(path);
2254 break;
2255 }
2256
2257 if (!upper->eb) {
2258 upper->eb = path->nodes[upper->level];
2259 path->nodes[upper->level] = NULL;
2260 } else {
2261 BUG_ON(upper->eb != path->nodes[upper->level]);
2262 }
2263
2264 upper->locked = 1;
2265 path->locks[upper->level] = 0;
2266
2267 slot = path->slots[upper->level];
2268 btrfs_release_path(path);
2269 } else {
2270 ret = btrfs_bin_search(upper->eb, key, &slot);
2271 if (ret < 0) {
2272 err = ret;
2273 goto next;
2274 }
2275 BUG_ON(ret);
2276 }
2277
2278 bytenr = btrfs_node_blockptr(upper->eb, slot);
2279 if (lowest) {
2280 if (bytenr != node->bytenr) {
2281 btrfs_err(root->fs_info,
2282 "lowest leaf/node mismatch: bytenr %llu node->bytenr %llu slot %d upper %llu",
2283 bytenr, node->bytenr, slot,
2284 upper->eb->start);
2285 err = -EIO;
2286 goto next;
2287 }
2288 } else {
2289 if (node->eb->start == bytenr)
2290 goto next;
2291 }
2292
2293 blocksize = root->fs_info->nodesize;
2294 generation = btrfs_node_ptr_generation(upper->eb, slot);
2295 btrfs_node_key_to_cpu(upper->eb, &first_key, slot);
2296 eb = read_tree_block(fs_info, bytenr, generation,
2297 upper->level - 1, &first_key);
2298 if (IS_ERR(eb)) {
2299 err = PTR_ERR(eb);
2300 goto next;
2301 } else if (!extent_buffer_uptodate(eb)) {
2302 free_extent_buffer(eb);
2303 err = -EIO;
2304 goto next;
2305 }
2306 btrfs_tree_lock(eb);
2307 btrfs_set_lock_blocking_write(eb);
2308
2309 if (!node->eb) {
2310 ret = btrfs_cow_block(trans, root, eb, upper->eb,
2311 slot, &eb);
2312 btrfs_tree_unlock(eb);
2313 free_extent_buffer(eb);
2314 if (ret < 0) {
2315 err = ret;
2316 goto next;
2317 }
2318 BUG_ON(node->eb != eb);
2319 } else {
2320 btrfs_set_node_blockptr(upper->eb, slot,
2321 node->eb->start);
2322 btrfs_set_node_ptr_generation(upper->eb, slot,
2323 trans->transid);
2324 btrfs_mark_buffer_dirty(upper->eb);
2325
2326 btrfs_init_generic_ref(&ref, BTRFS_ADD_DELAYED_REF,
2327 node->eb->start, blocksize,
2328 upper->eb->start);
2329 ref.real_root = root->root_key.objectid;
2330 btrfs_init_tree_ref(&ref, node->level,
2331 btrfs_header_owner(upper->eb));
2332 ret = btrfs_inc_extent_ref(trans, &ref);
2333 BUG_ON(ret);
2334
2335 ret = btrfs_drop_subtree(trans, root, eb, upper->eb);
2336 BUG_ON(ret);
2337 }
2338next:
2339 if (!upper->pending)
2340 btrfs_backref_drop_node_buffer(upper);
2341 else
2342 btrfs_backref_unlock_node_buffer(upper);
2343 if (err)
2344 break;
2345 }
2346
2347 if (!err && node->pending) {
2348 btrfs_backref_drop_node_buffer(node);
2349 list_move_tail(&node->list, &rc->backref_cache.changed);
2350 node->pending = 0;
2351 }
2352
2353 path->lowest_level = 0;
2354 BUG_ON(err == -ENOSPC);
2355 return err;
2356}
2357
2358static int link_to_upper(struct btrfs_trans_handle *trans,
2359 struct reloc_control *rc,
2360 struct btrfs_backref_node *node,
2361 struct btrfs_path *path)
2362{
2363 struct btrfs_key key;
2364
2365 btrfs_node_key_to_cpu(node->eb, &key, 0);
2366 return do_relocation(trans, rc, node, &key, path, 0);
2367}
2368
2369static int finish_pending_nodes(struct btrfs_trans_handle *trans,
2370 struct reloc_control *rc,
2371 struct btrfs_path *path, int err)
2372{
2373 LIST_HEAD(list);
2374 struct btrfs_backref_cache *cache = &rc->backref_cache;
2375 struct btrfs_backref_node *node;
2376 int level;
2377 int ret;
2378
2379 for (level = 0; level < BTRFS_MAX_LEVEL; level++) {
2380 while (!list_empty(&cache->pending[level])) {
2381 node = list_entry(cache->pending[level].next,
2382 struct btrfs_backref_node, list);
2383 list_move_tail(&node->list, &list);
2384 BUG_ON(!node->pending);
2385
2386 if (!err) {
2387 ret = link_to_upper(trans, rc, node, path);
2388 if (ret < 0)
2389 err = ret;
2390 }
2391 }
2392 list_splice_init(&list, &cache->pending[level]);
2393 }
2394 return err;
2395}
2396
2397/*
2398 * mark a block and all blocks directly/indirectly reference the block
2399 * as processed.
2400 */
2401static void update_processed_blocks(struct reloc_control *rc,
2402 struct btrfs_backref_node *node)
2403{
2404 struct btrfs_backref_node *next = node;
2405 struct btrfs_backref_edge *edge;
2406 struct btrfs_backref_edge *edges[BTRFS_MAX_LEVEL - 1];
2407 int index = 0;
2408
2409 while (next) {
2410 cond_resched();
2411 while (1) {
2412 if (next->processed)
2413 break;
2414
2415 mark_block_processed(rc, next);
2416
2417 if (list_empty(&next->upper))
2418 break;
2419
2420 edge = list_entry(next->upper.next,
2421 struct btrfs_backref_edge, list[LOWER]);
2422 edges[index++] = edge;
2423 next = edge->node[UPPER];
2424 }
2425 next = walk_down_backref(edges, &index);
2426 }
2427}
2428
2429static int tree_block_processed(u64 bytenr, struct reloc_control *rc)
2430{
2431 u32 blocksize = rc->extent_root->fs_info->nodesize;
2432
2433 if (test_range_bit(&rc->processed_blocks, bytenr,
2434 bytenr + blocksize - 1, EXTENT_DIRTY, 1, NULL))
2435 return 1;
2436 return 0;
2437}
2438
2439static int get_tree_block_key(struct btrfs_fs_info *fs_info,
2440 struct tree_block *block)
2441{
2442 struct extent_buffer *eb;
2443
2444 eb = read_tree_block(fs_info, block->bytenr, block->key.offset,
2445 block->level, NULL);
2446 if (IS_ERR(eb)) {
2447 return PTR_ERR(eb);
2448 } else if (!extent_buffer_uptodate(eb)) {
2449 free_extent_buffer(eb);
2450 return -EIO;
2451 }
2452 if (block->level == 0)
2453 btrfs_item_key_to_cpu(eb, &block->key, 0);
2454 else
2455 btrfs_node_key_to_cpu(eb, &block->key, 0);
2456 free_extent_buffer(eb);
2457 block->key_ready = 1;
2458 return 0;
2459}
2460
2461/*
2462 * helper function to relocate a tree block
2463 */
2464static int relocate_tree_block(struct btrfs_trans_handle *trans,
2465 struct reloc_control *rc,
2466 struct btrfs_backref_node *node,
2467 struct btrfs_key *key,
2468 struct btrfs_path *path)
2469{
2470 struct btrfs_root *root;
2471 int ret = 0;
2472
2473 if (!node)
2474 return 0;
2475
2476 /*
2477 * If we fail here we want to drop our backref_node because we are going
2478 * to start over and regenerate the tree for it.
2479 */
2480 ret = reserve_metadata_space(trans, rc, node);
2481 if (ret)
2482 goto out;
2483
2484 BUG_ON(node->processed);
2485 root = select_one_root(node);
2486 if (root == ERR_PTR(-ENOENT)) {
2487 update_processed_blocks(rc, node);
2488 goto out;
2489 }
2490
2491 if (root) {
2492 if (test_bit(BTRFS_ROOT_SHAREABLE, &root->state)) {
2493 BUG_ON(node->new_bytenr);
2494 BUG_ON(!list_empty(&node->list));
2495 btrfs_record_root_in_trans(trans, root);
2496 root = root->reloc_root;
2497 node->new_bytenr = root->node->start;
2498 btrfs_put_root(node->root);
2499 node->root = btrfs_grab_root(root);
2500 ASSERT(node->root);
2501 list_add_tail(&node->list, &rc->backref_cache.changed);
2502 } else {
2503 path->lowest_level = node->level;
2504 ret = btrfs_search_slot(trans, root, key, path, 0, 1);
2505 btrfs_release_path(path);
2506 if (ret > 0)
2507 ret = 0;
2508 }
2509 if (!ret)
2510 update_processed_blocks(rc, node);
2511 } else {
2512 ret = do_relocation(trans, rc, node, key, path, 1);
2513 }
2514out:
2515 if (ret || node->level == 0 || node->cowonly)
2516 btrfs_backref_cleanup_node(&rc->backref_cache, node);
2517 return ret;
2518}
2519
2520/*
2521 * relocate a list of blocks
2522 */
2523static noinline_for_stack
2524int relocate_tree_blocks(struct btrfs_trans_handle *trans,
2525 struct reloc_control *rc, struct rb_root *blocks)
2526{
2527 struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
2528 struct btrfs_backref_node *node;
2529 struct btrfs_path *path;
2530 struct tree_block *block;
2531 struct tree_block *next;
2532 int ret;
2533 int err = 0;
2534
2535 path = btrfs_alloc_path();
2536 if (!path) {
2537 err = -ENOMEM;
2538 goto out_free_blocks;
2539 }
2540
2541 /* Kick in readahead for tree blocks with missing keys */
2542 rbtree_postorder_for_each_entry_safe(block, next, blocks, rb_node) {
2543 if (!block->key_ready)
2544 readahead_tree_block(fs_info, block->bytenr);
2545 }
2546
2547 /* Get first keys */
2548 rbtree_postorder_for_each_entry_safe(block, next, blocks, rb_node) {
2549 if (!block->key_ready) {
2550 err = get_tree_block_key(fs_info, block);
2551 if (err)
2552 goto out_free_path;
2553 }
2554 }
2555
2556 /* Do tree relocation */
2557 rbtree_postorder_for_each_entry_safe(block, next, blocks, rb_node) {
2558 node = build_backref_tree(rc, &block->key,
2559 block->level, block->bytenr);
2560 if (IS_ERR(node)) {
2561 err = PTR_ERR(node);
2562 goto out;
2563 }
2564
2565 ret = relocate_tree_block(trans, rc, node, &block->key,
2566 path);
2567 if (ret < 0) {
2568 err = ret;
2569 break;
2570 }
2571 }
2572out:
2573 err = finish_pending_nodes(trans, rc, path, err);
2574
2575out_free_path:
2576 btrfs_free_path(path);
2577out_free_blocks:
2578 free_block_list(blocks);
2579 return err;
2580}
2581
2582static noinline_for_stack int prealloc_file_extent_cluster(
2583 struct btrfs_inode *inode,
2584 struct file_extent_cluster *cluster)
2585{
2586 u64 alloc_hint = 0;
2587 u64 start;
2588 u64 end;
2589 u64 offset = inode->index_cnt;
2590 u64 num_bytes;
2591 int nr;
2592 int ret = 0;
2593 u64 prealloc_start = cluster->start - offset;
2594 u64 prealloc_end = cluster->end - offset;
2595 u64 cur_offset = prealloc_start;
2596
2597 BUG_ON(cluster->start != cluster->boundary[0]);
2598 ret = btrfs_alloc_data_chunk_ondemand(inode,
2599 prealloc_end + 1 - prealloc_start);
2600 if (ret)
2601 return ret;
2602
2603 inode_lock(&inode->vfs_inode);
2604 for (nr = 0; nr < cluster->nr; nr++) {
2605 start = cluster->boundary[nr] - offset;
2606 if (nr + 1 < cluster->nr)
2607 end = cluster->boundary[nr + 1] - 1 - offset;
2608 else
2609 end = cluster->end - offset;
2610
2611 lock_extent(&inode->io_tree, start, end);
2612 num_bytes = end + 1 - start;
2613 ret = btrfs_prealloc_file_range(&inode->vfs_inode, 0, start,
2614 num_bytes, num_bytes,
2615 end + 1, &alloc_hint);
2616 cur_offset = end + 1;
2617 unlock_extent(&inode->io_tree, start, end);
2618 if (ret)
2619 break;
2620 }
2621 inode_unlock(&inode->vfs_inode);
2622
2623 if (cur_offset < prealloc_end)
2624 btrfs_free_reserved_data_space_noquota(inode->root->fs_info,
2625 prealloc_end + 1 - cur_offset);
2626 return ret;
2627}
2628
2629static noinline_for_stack
2630int setup_extent_mapping(struct inode *inode, u64 start, u64 end,
2631 u64 block_start)
2632{
2633 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
2634 struct extent_map *em;
2635 int ret = 0;
2636
2637 em = alloc_extent_map();
2638 if (!em)
2639 return -ENOMEM;
2640
2641 em->start = start;
2642 em->len = end + 1 - start;
2643 em->block_len = em->len;
2644 em->block_start = block_start;
2645 set_bit(EXTENT_FLAG_PINNED, &em->flags);
2646
2647 lock_extent(&BTRFS_I(inode)->io_tree, start, end);
2648 while (1) {
2649 write_lock(&em_tree->lock);
2650 ret = add_extent_mapping(em_tree, em, 0);
2651 write_unlock(&em_tree->lock);
2652 if (ret != -EEXIST) {
2653 free_extent_map(em);
2654 break;
2655 }
2656 btrfs_drop_extent_cache(BTRFS_I(inode), start, end, 0);
2657 }
2658 unlock_extent(&BTRFS_I(inode)->io_tree, start, end);
2659 return ret;
2660}
2661
2662/*
2663 * Allow error injection to test balance cancellation
2664 */
2665int btrfs_should_cancel_balance(struct btrfs_fs_info *fs_info)
2666{
2667 return atomic_read(&fs_info->balance_cancel_req) ||
2668 fatal_signal_pending(current);
2669}
2670ALLOW_ERROR_INJECTION(btrfs_should_cancel_balance, TRUE);
2671
2672static int relocate_file_extent_cluster(struct inode *inode,
2673 struct file_extent_cluster *cluster)
2674{
2675 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2676 u64 page_start;
2677 u64 page_end;
2678 u64 offset = BTRFS_I(inode)->index_cnt;
2679 unsigned long index;
2680 unsigned long last_index;
2681 struct page *page;
2682 struct file_ra_state *ra;
2683 gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping);
2684 int nr = 0;
2685 int ret = 0;
2686
2687 if (!cluster->nr)
2688 return 0;
2689
2690 ra = kzalloc(sizeof(*ra), GFP_NOFS);
2691 if (!ra)
2692 return -ENOMEM;
2693
2694 ret = prealloc_file_extent_cluster(BTRFS_I(inode), cluster);
2695 if (ret)
2696 goto out;
2697
2698 file_ra_state_init(ra, inode->i_mapping);
2699
2700 ret = setup_extent_mapping(inode, cluster->start - offset,
2701 cluster->end - offset, cluster->start);
2702 if (ret)
2703 goto out;
2704
2705 index = (cluster->start - offset) >> PAGE_SHIFT;
2706 last_index = (cluster->end - offset) >> PAGE_SHIFT;
2707 while (index <= last_index) {
2708 ret = btrfs_delalloc_reserve_metadata(BTRFS_I(inode),
2709 PAGE_SIZE);
2710 if (ret)
2711 goto out;
2712
2713 page = find_lock_page(inode->i_mapping, index);
2714 if (!page) {
2715 page_cache_sync_readahead(inode->i_mapping,
2716 ra, NULL, index,
2717 last_index + 1 - index);
2718 page = find_or_create_page(inode->i_mapping, index,
2719 mask);
2720 if (!page) {
2721 btrfs_delalloc_release_metadata(BTRFS_I(inode),
2722 PAGE_SIZE, true);
2723 btrfs_delalloc_release_extents(BTRFS_I(inode),
2724 PAGE_SIZE);
2725 ret = -ENOMEM;
2726 goto out;
2727 }
2728 }
2729
2730 if (PageReadahead(page)) {
2731 page_cache_async_readahead(inode->i_mapping,
2732 ra, NULL, page, index,
2733 last_index + 1 - index);
2734 }
2735
2736 if (!PageUptodate(page)) {
2737 btrfs_readpage(NULL, page);
2738 lock_page(page);
2739 if (!PageUptodate(page)) {
2740 unlock_page(page);
2741 put_page(page);
2742 btrfs_delalloc_release_metadata(BTRFS_I(inode),
2743 PAGE_SIZE, true);
2744 btrfs_delalloc_release_extents(BTRFS_I(inode),
2745 PAGE_SIZE);
2746 ret = -EIO;
2747 goto out;
2748 }
2749 }
2750
2751 page_start = page_offset(page);
2752 page_end = page_start + PAGE_SIZE - 1;
2753
2754 lock_extent(&BTRFS_I(inode)->io_tree, page_start, page_end);
2755
2756 set_page_extent_mapped(page);
2757
2758 if (nr < cluster->nr &&
2759 page_start + offset == cluster->boundary[nr]) {
2760 set_extent_bits(&BTRFS_I(inode)->io_tree,
2761 page_start, page_end,
2762 EXTENT_BOUNDARY);
2763 nr++;
2764 }
2765
2766 ret = btrfs_set_extent_delalloc(BTRFS_I(inode), page_start,
2767 page_end, 0, NULL);
2768 if (ret) {
2769 unlock_page(page);
2770 put_page(page);
2771 btrfs_delalloc_release_metadata(BTRFS_I(inode),
2772 PAGE_SIZE, true);
2773 btrfs_delalloc_release_extents(BTRFS_I(inode),
2774 PAGE_SIZE);
2775
2776 clear_extent_bits(&BTRFS_I(inode)->io_tree,
2777 page_start, page_end,
2778 EXTENT_LOCKED | EXTENT_BOUNDARY);
2779 goto out;
2780
2781 }
2782 set_page_dirty(page);
2783
2784 unlock_extent(&BTRFS_I(inode)->io_tree,
2785 page_start, page_end);
2786 unlock_page(page);
2787 put_page(page);
2788
2789 index++;
2790 btrfs_delalloc_release_extents(BTRFS_I(inode), PAGE_SIZE);
2791 balance_dirty_pages_ratelimited(inode->i_mapping);
2792 btrfs_throttle(fs_info);
2793 if (btrfs_should_cancel_balance(fs_info)) {
2794 ret = -ECANCELED;
2795 goto out;
2796 }
2797 }
2798 WARN_ON(nr != cluster->nr);
2799out:
2800 kfree(ra);
2801 return ret;
2802}
2803
2804static noinline_for_stack
2805int relocate_data_extent(struct inode *inode, struct btrfs_key *extent_key,
2806 struct file_extent_cluster *cluster)
2807{
2808 int ret;
2809
2810 if (cluster->nr > 0 && extent_key->objectid != cluster->end + 1) {
2811 ret = relocate_file_extent_cluster(inode, cluster);
2812 if (ret)
2813 return ret;
2814 cluster->nr = 0;
2815 }
2816
2817 if (!cluster->nr)
2818 cluster->start = extent_key->objectid;
2819 else
2820 BUG_ON(cluster->nr >= MAX_EXTENTS);
2821 cluster->end = extent_key->objectid + extent_key->offset - 1;
2822 cluster->boundary[cluster->nr] = extent_key->objectid;
2823 cluster->nr++;
2824
2825 if (cluster->nr >= MAX_EXTENTS) {
2826 ret = relocate_file_extent_cluster(inode, cluster);
2827 if (ret)
2828 return ret;
2829 cluster->nr = 0;
2830 }
2831 return 0;
2832}
2833
2834/*
2835 * helper to add a tree block to the list.
2836 * the major work is getting the generation and level of the block
2837 */
2838static int add_tree_block(struct reloc_control *rc,
2839 struct btrfs_key *extent_key,
2840 struct btrfs_path *path,
2841 struct rb_root *blocks)
2842{
2843 struct extent_buffer *eb;
2844 struct btrfs_extent_item *ei;
2845 struct btrfs_tree_block_info *bi;
2846 struct tree_block *block;
2847 struct rb_node *rb_node;
2848 u32 item_size;
2849 int level = -1;
2850 u64 generation;
2851
2852 eb = path->nodes[0];
2853 item_size = btrfs_item_size_nr(eb, path->slots[0]);
2854
2855 if (extent_key->type == BTRFS_METADATA_ITEM_KEY ||
2856 item_size >= sizeof(*ei) + sizeof(*bi)) {
2857 ei = btrfs_item_ptr(eb, path->slots[0],
2858 struct btrfs_extent_item);
2859 if (extent_key->type == BTRFS_EXTENT_ITEM_KEY) {
2860 bi = (struct btrfs_tree_block_info *)(ei + 1);
2861 level = btrfs_tree_block_level(eb, bi);
2862 } else {
2863 level = (int)extent_key->offset;
2864 }
2865 generation = btrfs_extent_generation(eb, ei);
2866 } else if (unlikely(item_size == sizeof(struct btrfs_extent_item_v0))) {
2867 btrfs_print_v0_err(eb->fs_info);
2868 btrfs_handle_fs_error(eb->fs_info, -EINVAL, NULL);
2869 return -EINVAL;
2870 } else {
2871 BUG();
2872 }
2873
2874 btrfs_release_path(path);
2875
2876 BUG_ON(level == -1);
2877
2878 block = kmalloc(sizeof(*block), GFP_NOFS);
2879 if (!block)
2880 return -ENOMEM;
2881
2882 block->bytenr = extent_key->objectid;
2883 block->key.objectid = rc->extent_root->fs_info->nodesize;
2884 block->key.offset = generation;
2885 block->level = level;
2886 block->key_ready = 0;
2887
2888 rb_node = rb_simple_insert(blocks, block->bytenr, &block->rb_node);
2889 if (rb_node)
2890 btrfs_backref_panic(rc->extent_root->fs_info, block->bytenr,
2891 -EEXIST);
2892
2893 return 0;
2894}
2895
2896/*
2897 * helper to add tree blocks for backref of type BTRFS_SHARED_DATA_REF_KEY
2898 */
2899static int __add_tree_block(struct reloc_control *rc,
2900 u64 bytenr, u32 blocksize,
2901 struct rb_root *blocks)
2902{
2903 struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
2904 struct btrfs_path *path;
2905 struct btrfs_key key;
2906 int ret;
2907 bool skinny = btrfs_fs_incompat(fs_info, SKINNY_METADATA);
2908
2909 if (tree_block_processed(bytenr, rc))
2910 return 0;
2911
2912 if (rb_simple_search(blocks, bytenr))
2913 return 0;
2914
2915 path = btrfs_alloc_path();
2916 if (!path)
2917 return -ENOMEM;
2918again:
2919 key.objectid = bytenr;
2920 if (skinny) {
2921 key.type = BTRFS_METADATA_ITEM_KEY;
2922 key.offset = (u64)-1;
2923 } else {
2924 key.type = BTRFS_EXTENT_ITEM_KEY;
2925 key.offset = blocksize;
2926 }
2927
2928 path->search_commit_root = 1;
2929 path->skip_locking = 1;
2930 ret = btrfs_search_slot(NULL, rc->extent_root, &key, path, 0, 0);
2931 if (ret < 0)
2932 goto out;
2933
2934 if (ret > 0 && skinny) {
2935 if (path->slots[0]) {
2936 path->slots[0]--;
2937 btrfs_item_key_to_cpu(path->nodes[0], &key,
2938 path->slots[0]);
2939 if (key.objectid == bytenr &&
2940 (key.type == BTRFS_METADATA_ITEM_KEY ||
2941 (key.type == BTRFS_EXTENT_ITEM_KEY &&
2942 key.offset == blocksize)))
2943 ret = 0;
2944 }
2945
2946 if (ret) {
2947 skinny = false;
2948 btrfs_release_path(path);
2949 goto again;
2950 }
2951 }
2952 if (ret) {
2953 ASSERT(ret == 1);
2954 btrfs_print_leaf(path->nodes[0]);
2955 btrfs_err(fs_info,
2956 "tree block extent item (%llu) is not found in extent tree",
2957 bytenr);
2958 WARN_ON(1);
2959 ret = -EINVAL;
2960 goto out;
2961 }
2962
2963 ret = add_tree_block(rc, &key, path, blocks);
2964out:
2965 btrfs_free_path(path);
2966 return ret;
2967}
2968
2969static int delete_block_group_cache(struct btrfs_fs_info *fs_info,
2970 struct btrfs_block_group *block_group,
2971 struct inode *inode,
2972 u64 ino)
2973{
2974 struct btrfs_root *root = fs_info->tree_root;
2975 struct btrfs_trans_handle *trans;
2976 int ret = 0;
2977
2978 if (inode)
2979 goto truncate;
2980
2981 inode = btrfs_iget(fs_info->sb, ino, root);
2982 if (IS_ERR(inode))
2983 return -ENOENT;
2984
2985truncate:
2986 ret = btrfs_check_trunc_cache_free_space(fs_info,
2987 &fs_info->global_block_rsv);
2988 if (ret)
2989 goto out;
2990
2991 trans = btrfs_join_transaction(root);
2992 if (IS_ERR(trans)) {
2993 ret = PTR_ERR(trans);
2994 goto out;
2995 }
2996
2997 ret = btrfs_truncate_free_space_cache(trans, block_group, inode);
2998
2999 btrfs_end_transaction(trans);
3000 btrfs_btree_balance_dirty(fs_info);
3001out:
3002 iput(inode);
3003 return ret;
3004}
3005
3006/*
3007 * Locate the free space cache EXTENT_DATA in root tree leaf and delete the
3008 * cache inode, to avoid free space cache data extent blocking data relocation.
3009 */
3010static int delete_v1_space_cache(struct extent_buffer *leaf,
3011 struct btrfs_block_group *block_group,
3012 u64 data_bytenr)
3013{
3014 u64 space_cache_ino;
3015 struct btrfs_file_extent_item *ei;
3016 struct btrfs_key key;
3017 bool found = false;
3018 int i;
3019 int ret;
3020
3021 if (btrfs_header_owner(leaf) != BTRFS_ROOT_TREE_OBJECTID)
3022 return 0;
3023
3024 for (i = 0; i < btrfs_header_nritems(leaf); i++) {
3025 btrfs_item_key_to_cpu(leaf, &key, i);
3026 if (key.type != BTRFS_EXTENT_DATA_KEY)
3027 continue;
3028 ei = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
3029 if (btrfs_file_extent_type(leaf, ei) == BTRFS_FILE_EXTENT_REG &&
3030 btrfs_file_extent_disk_bytenr(leaf, ei) == data_bytenr) {
3031 found = true;
3032 space_cache_ino = key.objectid;
3033 break;
3034 }
3035 }
3036 if (!found)
3037 return -ENOENT;
3038 ret = delete_block_group_cache(leaf->fs_info, block_group, NULL,
3039 space_cache_ino);
3040 return ret;
3041}
3042
3043/*
3044 * helper to find all tree blocks that reference a given data extent
3045 */
3046static noinline_for_stack
3047int add_data_references(struct reloc_control *rc,
3048 struct btrfs_key *extent_key,
3049 struct btrfs_path *path,
3050 struct rb_root *blocks)
3051{
3052 struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
3053 struct ulist *leaves = NULL;
3054 struct ulist_iterator leaf_uiter;
3055 struct ulist_node *ref_node = NULL;
3056 const u32 blocksize = fs_info->nodesize;
3057 int ret = 0;
3058
3059 btrfs_release_path(path);
3060 ret = btrfs_find_all_leafs(NULL, fs_info, extent_key->objectid,
3061 0, &leaves, NULL, true);
3062 if (ret < 0)
3063 return ret;
3064
3065 ULIST_ITER_INIT(&leaf_uiter);
3066 while ((ref_node = ulist_next(leaves, &leaf_uiter))) {
3067 struct extent_buffer *eb;
3068
3069 eb = read_tree_block(fs_info, ref_node->val, 0, 0, NULL);
3070 if (IS_ERR(eb)) {
3071 ret = PTR_ERR(eb);
3072 break;
3073 }
3074 ret = delete_v1_space_cache(eb, rc->block_group,
3075 extent_key->objectid);
3076 free_extent_buffer(eb);
3077 if (ret < 0)
3078 break;
3079 ret = __add_tree_block(rc, ref_node->val, blocksize, blocks);
3080 if (ret < 0)
3081 break;
3082 }
3083 if (ret < 0)
3084 free_block_list(blocks);
3085 ulist_free(leaves);
3086 return ret;
3087}
3088
3089/*
3090 * helper to find next unprocessed extent
3091 */
3092static noinline_for_stack
3093int find_next_extent(struct reloc_control *rc, struct btrfs_path *path,
3094 struct btrfs_key *extent_key)
3095{
3096 struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
3097 struct btrfs_key key;
3098 struct extent_buffer *leaf;
3099 u64 start, end, last;
3100 int ret;
3101
3102 last = rc->block_group->start + rc->block_group->length;
3103 while (1) {
3104 cond_resched();
3105 if (rc->search_start >= last) {
3106 ret = 1;
3107 break;
3108 }
3109
3110 key.objectid = rc->search_start;
3111 key.type = BTRFS_EXTENT_ITEM_KEY;
3112 key.offset = 0;
3113
3114 path->search_commit_root = 1;
3115 path->skip_locking = 1;
3116 ret = btrfs_search_slot(NULL, rc->extent_root, &key, path,
3117 0, 0);
3118 if (ret < 0)
3119 break;
3120next:
3121 leaf = path->nodes[0];
3122 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
3123 ret = btrfs_next_leaf(rc->extent_root, path);
3124 if (ret != 0)
3125 break;
3126 leaf = path->nodes[0];
3127 }
3128
3129 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3130 if (key.objectid >= last) {
3131 ret = 1;
3132 break;
3133 }
3134
3135 if (key.type != BTRFS_EXTENT_ITEM_KEY &&
3136 key.type != BTRFS_METADATA_ITEM_KEY) {
3137 path->slots[0]++;
3138 goto next;
3139 }
3140
3141 if (key.type == BTRFS_EXTENT_ITEM_KEY &&
3142 key.objectid + key.offset <= rc->search_start) {
3143 path->slots[0]++;
3144 goto next;
3145 }
3146
3147 if (key.type == BTRFS_METADATA_ITEM_KEY &&
3148 key.objectid + fs_info->nodesize <=
3149 rc->search_start) {
3150 path->slots[0]++;
3151 goto next;
3152 }
3153
3154 ret = find_first_extent_bit(&rc->processed_blocks,
3155 key.objectid, &start, &end,
3156 EXTENT_DIRTY, NULL);
3157
3158 if (ret == 0 && start <= key.objectid) {
3159 btrfs_release_path(path);
3160 rc->search_start = end + 1;
3161 } else {
3162 if (key.type == BTRFS_EXTENT_ITEM_KEY)
3163 rc->search_start = key.objectid + key.offset;
3164 else
3165 rc->search_start = key.objectid +
3166 fs_info->nodesize;
3167 memcpy(extent_key, &key, sizeof(key));
3168 return 0;
3169 }
3170 }
3171 btrfs_release_path(path);
3172 return ret;
3173}
3174
3175static void set_reloc_control(struct reloc_control *rc)
3176{
3177 struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
3178
3179 mutex_lock(&fs_info->reloc_mutex);
3180 fs_info->reloc_ctl = rc;
3181 mutex_unlock(&fs_info->reloc_mutex);
3182}
3183
3184static void unset_reloc_control(struct reloc_control *rc)
3185{
3186 struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
3187
3188 mutex_lock(&fs_info->reloc_mutex);
3189 fs_info->reloc_ctl = NULL;
3190 mutex_unlock(&fs_info->reloc_mutex);
3191}
3192
3193static int check_extent_flags(u64 flags)
3194{
3195 if ((flags & BTRFS_EXTENT_FLAG_DATA) &&
3196 (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK))
3197 return 1;
3198 if (!(flags & BTRFS_EXTENT_FLAG_DATA) &&
3199 !(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK))
3200 return 1;
3201 if ((flags & BTRFS_EXTENT_FLAG_DATA) &&
3202 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
3203 return 1;
3204 return 0;
3205}
3206
3207static noinline_for_stack
3208int prepare_to_relocate(struct reloc_control *rc)
3209{
3210 struct btrfs_trans_handle *trans;
3211 int ret;
3212
3213 rc->block_rsv = btrfs_alloc_block_rsv(rc->extent_root->fs_info,
3214 BTRFS_BLOCK_RSV_TEMP);
3215 if (!rc->block_rsv)
3216 return -ENOMEM;
3217
3218 memset(&rc->cluster, 0, sizeof(rc->cluster));
3219 rc->search_start = rc->block_group->start;
3220 rc->extents_found = 0;
3221 rc->nodes_relocated = 0;
3222 rc->merging_rsv_size = 0;
3223 rc->reserved_bytes = 0;
3224 rc->block_rsv->size = rc->extent_root->fs_info->nodesize *
3225 RELOCATION_RESERVED_NODES;
3226 ret = btrfs_block_rsv_refill(rc->extent_root,
3227 rc->block_rsv, rc->block_rsv->size,
3228 BTRFS_RESERVE_FLUSH_ALL);
3229 if (ret)
3230 return ret;
3231
3232 rc->create_reloc_tree = 1;
3233 set_reloc_control(rc);
3234
3235 trans = btrfs_join_transaction(rc->extent_root);
3236 if (IS_ERR(trans)) {
3237 unset_reloc_control(rc);
3238 /*
3239 * extent tree is not a ref_cow tree and has no reloc_root to
3240 * cleanup. And callers are responsible to free the above
3241 * block rsv.
3242 */
3243 return PTR_ERR(trans);
3244 }
3245 btrfs_commit_transaction(trans);
3246 return 0;
3247}
3248
3249static noinline_for_stack int relocate_block_group(struct reloc_control *rc)
3250{
3251 struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
3252 struct rb_root blocks = RB_ROOT;
3253 struct btrfs_key key;
3254 struct btrfs_trans_handle *trans = NULL;
3255 struct btrfs_path *path;
3256 struct btrfs_extent_item *ei;
3257 u64 flags;
3258 u32 item_size;
3259 int ret;
3260 int err = 0;
3261 int progress = 0;
3262
3263 path = btrfs_alloc_path();
3264 if (!path)
3265 return -ENOMEM;
3266 path->reada = READA_FORWARD;
3267
3268 ret = prepare_to_relocate(rc);
3269 if (ret) {
3270 err = ret;
3271 goto out_free;
3272 }
3273
3274 while (1) {
3275 rc->reserved_bytes = 0;
3276 ret = btrfs_block_rsv_refill(rc->extent_root,
3277 rc->block_rsv, rc->block_rsv->size,
3278 BTRFS_RESERVE_FLUSH_ALL);
3279 if (ret) {
3280 err = ret;
3281 break;
3282 }
3283 progress++;
3284 trans = btrfs_start_transaction(rc->extent_root, 0);
3285 if (IS_ERR(trans)) {
3286 err = PTR_ERR(trans);
3287 trans = NULL;
3288 break;
3289 }
3290restart:
3291 if (update_backref_cache(trans, &rc->backref_cache)) {
3292 btrfs_end_transaction(trans);
3293 trans = NULL;
3294 continue;
3295 }
3296
3297 ret = find_next_extent(rc, path, &key);
3298 if (ret < 0)
3299 err = ret;
3300 if (ret != 0)
3301 break;
3302
3303 rc->extents_found++;
3304
3305 ei = btrfs_item_ptr(path->nodes[0], path->slots[0],
3306 struct btrfs_extent_item);
3307 item_size = btrfs_item_size_nr(path->nodes[0], path->slots[0]);
3308 if (item_size >= sizeof(*ei)) {
3309 flags = btrfs_extent_flags(path->nodes[0], ei);
3310 ret = check_extent_flags(flags);
3311 BUG_ON(ret);
3312 } else if (unlikely(item_size == sizeof(struct btrfs_extent_item_v0))) {
3313 err = -EINVAL;
3314 btrfs_print_v0_err(trans->fs_info);
3315 btrfs_abort_transaction(trans, err);
3316 break;
3317 } else {
3318 BUG();
3319 }
3320
3321 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
3322 ret = add_tree_block(rc, &key, path, &blocks);
3323 } else if (rc->stage == UPDATE_DATA_PTRS &&
3324 (flags & BTRFS_EXTENT_FLAG_DATA)) {
3325 ret = add_data_references(rc, &key, path, &blocks);
3326 } else {
3327 btrfs_release_path(path);
3328 ret = 0;
3329 }
3330 if (ret < 0) {
3331 err = ret;
3332 break;
3333 }
3334
3335 if (!RB_EMPTY_ROOT(&blocks)) {
3336 ret = relocate_tree_blocks(trans, rc, &blocks);
3337 if (ret < 0) {
3338 if (ret != -EAGAIN) {
3339 err = ret;
3340 break;
3341 }
3342 rc->extents_found--;
3343 rc->search_start = key.objectid;
3344 }
3345 }
3346
3347 btrfs_end_transaction_throttle(trans);
3348 btrfs_btree_balance_dirty(fs_info);
3349 trans = NULL;
3350
3351 if (rc->stage == MOVE_DATA_EXTENTS &&
3352 (flags & BTRFS_EXTENT_FLAG_DATA)) {
3353 rc->found_file_extent = 1;
3354 ret = relocate_data_extent(rc->data_inode,
3355 &key, &rc->cluster);
3356 if (ret < 0) {
3357 err = ret;
3358 break;
3359 }
3360 }
3361 if (btrfs_should_cancel_balance(fs_info)) {
3362 err = -ECANCELED;
3363 break;
3364 }
3365 }
3366 if (trans && progress && err == -ENOSPC) {
3367 ret = btrfs_force_chunk_alloc(trans, rc->block_group->flags);
3368 if (ret == 1) {
3369 err = 0;
3370 progress = 0;
3371 goto restart;
3372 }
3373 }
3374
3375 btrfs_release_path(path);
3376 clear_extent_bits(&rc->processed_blocks, 0, (u64)-1, EXTENT_DIRTY);
3377
3378 if (trans) {
3379 btrfs_end_transaction_throttle(trans);
3380 btrfs_btree_balance_dirty(fs_info);
3381 }
3382
3383 if (!err) {
3384 ret = relocate_file_extent_cluster(rc->data_inode,
3385 &rc->cluster);
3386 if (ret < 0)
3387 err = ret;
3388 }
3389
3390 rc->create_reloc_tree = 0;
3391 set_reloc_control(rc);
3392
3393 btrfs_backref_release_cache(&rc->backref_cache);
3394 btrfs_block_rsv_release(fs_info, rc->block_rsv, (u64)-1, NULL);
3395
3396 /*
3397 * Even in the case when the relocation is cancelled, we should all go
3398 * through prepare_to_merge() and merge_reloc_roots().
3399 *
3400 * For error (including cancelled balance), prepare_to_merge() will
3401 * mark all reloc trees orphan, then queue them for cleanup in
3402 * merge_reloc_roots()
3403 */
3404 err = prepare_to_merge(rc, err);
3405
3406 merge_reloc_roots(rc);
3407
3408 rc->merge_reloc_tree = 0;
3409 unset_reloc_control(rc);
3410 btrfs_block_rsv_release(fs_info, rc->block_rsv, (u64)-1, NULL);
3411
3412 /* get rid of pinned extents */
3413 trans = btrfs_join_transaction(rc->extent_root);
3414 if (IS_ERR(trans)) {
3415 err = PTR_ERR(trans);
3416 goto out_free;
3417 }
3418 btrfs_commit_transaction(trans);
3419out_free:
3420 ret = clean_dirty_subvols(rc);
3421 if (ret < 0 && !err)
3422 err = ret;
3423 btrfs_free_block_rsv(fs_info, rc->block_rsv);
3424 btrfs_free_path(path);
3425 return err;
3426}
3427
3428static int __insert_orphan_inode(struct btrfs_trans_handle *trans,
3429 struct btrfs_root *root, u64 objectid)
3430{
3431 struct btrfs_path *path;
3432 struct btrfs_inode_item *item;
3433 struct extent_buffer *leaf;
3434 int ret;
3435
3436 path = btrfs_alloc_path();
3437 if (!path)
3438 return -ENOMEM;
3439
3440 ret = btrfs_insert_empty_inode(trans, root, path, objectid);
3441 if (ret)
3442 goto out;
3443
3444 leaf = path->nodes[0];
3445 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_inode_item);
3446 memzero_extent_buffer(leaf, (unsigned long)item, sizeof(*item));
3447 btrfs_set_inode_generation(leaf, item, 1);
3448 btrfs_set_inode_size(leaf, item, 0);
3449 btrfs_set_inode_mode(leaf, item, S_IFREG | 0600);
3450 btrfs_set_inode_flags(leaf, item, BTRFS_INODE_NOCOMPRESS |
3451 BTRFS_INODE_PREALLOC);
3452 btrfs_mark_buffer_dirty(leaf);
3453out:
3454 btrfs_free_path(path);
3455 return ret;
3456}
3457
3458/*
3459 * helper to create inode for data relocation.
3460 * the inode is in data relocation tree and its link count is 0
3461 */
3462static noinline_for_stack
3463struct inode *create_reloc_inode(struct btrfs_fs_info *fs_info,
3464 struct btrfs_block_group *group)
3465{
3466 struct inode *inode = NULL;
3467 struct btrfs_trans_handle *trans;
3468 struct btrfs_root *root;
3469 u64 objectid;
3470 int err = 0;
3471
3472 root = btrfs_grab_root(fs_info->data_reloc_root);
3473 trans = btrfs_start_transaction(root, 6);
3474 if (IS_ERR(trans)) {
3475 btrfs_put_root(root);
3476 return ERR_CAST(trans);
3477 }
3478
3479 err = btrfs_find_free_objectid(root, &objectid);
3480 if (err)
3481 goto out;
3482
3483 err = __insert_orphan_inode(trans, root, objectid);
3484 BUG_ON(err);
3485
3486 inode = btrfs_iget(fs_info->sb, objectid, root);
3487 BUG_ON(IS_ERR(inode));
3488 BTRFS_I(inode)->index_cnt = group->start;
3489
3490 err = btrfs_orphan_add(trans, BTRFS_I(inode));
3491out:
3492 btrfs_put_root(root);
3493 btrfs_end_transaction(trans);
3494 btrfs_btree_balance_dirty(fs_info);
3495 if (err) {
3496 if (inode)
3497 iput(inode);
3498 inode = ERR_PTR(err);
3499 }
3500 return inode;
3501}
3502
3503static struct reloc_control *alloc_reloc_control(struct btrfs_fs_info *fs_info)
3504{
3505 struct reloc_control *rc;
3506
3507 rc = kzalloc(sizeof(*rc), GFP_NOFS);
3508 if (!rc)
3509 return NULL;
3510
3511 INIT_LIST_HEAD(&rc->reloc_roots);
3512 INIT_LIST_HEAD(&rc->dirty_subvol_roots);
3513 btrfs_backref_init_cache(fs_info, &rc->backref_cache, 1);
3514 mapping_tree_init(&rc->reloc_root_tree);
3515 extent_io_tree_init(fs_info, &rc->processed_blocks,
3516 IO_TREE_RELOC_BLOCKS, NULL);
3517 return rc;
3518}
3519
3520static void free_reloc_control(struct reloc_control *rc)
3521{
3522 struct mapping_node *node, *tmp;
3523
3524 free_reloc_roots(&rc->reloc_roots);
3525 rbtree_postorder_for_each_entry_safe(node, tmp,
3526 &rc->reloc_root_tree.rb_root, rb_node)
3527 kfree(node);
3528
3529 kfree(rc);
3530}
3531
3532/*
3533 * Print the block group being relocated
3534 */
3535static void describe_relocation(struct btrfs_fs_info *fs_info,
3536 struct btrfs_block_group *block_group)
3537{
3538 char buf[128] = {'\0'};
3539
3540 btrfs_describe_block_groups(block_group->flags, buf, sizeof(buf));
3541
3542 btrfs_info(fs_info,
3543 "relocating block group %llu flags %s",
3544 block_group->start, buf);
3545}
3546
3547static const char *stage_to_string(int stage)
3548{
3549 if (stage == MOVE_DATA_EXTENTS)
3550 return "move data extents";
3551 if (stage == UPDATE_DATA_PTRS)
3552 return "update data pointers";
3553 return "unknown";
3554}
3555
3556/*
3557 * function to relocate all extents in a block group.
3558 */
3559int btrfs_relocate_block_group(struct btrfs_fs_info *fs_info, u64 group_start)
3560{
3561 struct btrfs_block_group *bg;
3562 struct btrfs_root *extent_root = fs_info->extent_root;
3563 struct reloc_control *rc;
3564 struct inode *inode;
3565 struct btrfs_path *path;
3566 int ret;
3567 int rw = 0;
3568 int err = 0;
3569
3570 bg = btrfs_lookup_block_group(fs_info, group_start);
3571 if (!bg)
3572 return -ENOENT;
3573
3574 if (btrfs_pinned_by_swapfile(fs_info, bg)) {
3575 btrfs_put_block_group(bg);
3576 return -ETXTBSY;
3577 }
3578
3579 rc = alloc_reloc_control(fs_info);
3580 if (!rc) {
3581 btrfs_put_block_group(bg);
3582 return -ENOMEM;
3583 }
3584
3585 rc->extent_root = extent_root;
3586 rc->block_group = bg;
3587
3588 ret = btrfs_inc_block_group_ro(rc->block_group, true);
3589 if (ret) {
3590 err = ret;
3591 goto out;
3592 }
3593 rw = 1;
3594
3595 path = btrfs_alloc_path();
3596 if (!path) {
3597 err = -ENOMEM;
3598 goto out;
3599 }
3600
3601 inode = lookup_free_space_inode(rc->block_group, path);
3602 btrfs_free_path(path);
3603
3604 if (!IS_ERR(inode))
3605 ret = delete_block_group_cache(fs_info, rc->block_group, inode, 0);
3606 else
3607 ret = PTR_ERR(inode);
3608
3609 if (ret && ret != -ENOENT) {
3610 err = ret;
3611 goto out;
3612 }
3613
3614 rc->data_inode = create_reloc_inode(fs_info, rc->block_group);
3615 if (IS_ERR(rc->data_inode)) {
3616 err = PTR_ERR(rc->data_inode);
3617 rc->data_inode = NULL;
3618 goto out;
3619 }
3620
3621 describe_relocation(fs_info, rc->block_group);
3622
3623 btrfs_wait_block_group_reservations(rc->block_group);
3624 btrfs_wait_nocow_writers(rc->block_group);
3625 btrfs_wait_ordered_roots(fs_info, U64_MAX,
3626 rc->block_group->start,
3627 rc->block_group->length);
3628
3629 while (1) {
3630 int finishes_stage;
3631
3632 mutex_lock(&fs_info->cleaner_mutex);
3633 ret = relocate_block_group(rc);
3634 mutex_unlock(&fs_info->cleaner_mutex);
3635 if (ret < 0)
3636 err = ret;
3637
3638 finishes_stage = rc->stage;
3639 /*
3640 * We may have gotten ENOSPC after we already dirtied some
3641 * extents. If writeout happens while we're relocating a
3642 * different block group we could end up hitting the
3643 * BUG_ON(rc->stage == UPDATE_DATA_PTRS) in
3644 * btrfs_reloc_cow_block. Make sure we write everything out
3645 * properly so we don't trip over this problem, and then break
3646 * out of the loop if we hit an error.
3647 */
3648 if (rc->stage == MOVE_DATA_EXTENTS && rc->found_file_extent) {
3649 ret = btrfs_wait_ordered_range(rc->data_inode, 0,
3650 (u64)-1);
3651 if (ret)
3652 err = ret;
3653 invalidate_mapping_pages(rc->data_inode->i_mapping,
3654 0, -1);
3655 rc->stage = UPDATE_DATA_PTRS;
3656 }
3657
3658 if (err < 0)
3659 goto out;
3660
3661 if (rc->extents_found == 0)
3662 break;
3663
3664 btrfs_info(fs_info, "found %llu extents, stage: %s",
3665 rc->extents_found, stage_to_string(finishes_stage));
3666 }
3667
3668 WARN_ON(rc->block_group->pinned > 0);
3669 WARN_ON(rc->block_group->reserved > 0);
3670 WARN_ON(rc->block_group->used > 0);
3671out:
3672 if (err && rw)
3673 btrfs_dec_block_group_ro(rc->block_group);
3674 iput(rc->data_inode);
3675 btrfs_put_block_group(rc->block_group);
3676 free_reloc_control(rc);
3677 return err;
3678}
3679
3680static noinline_for_stack int mark_garbage_root(struct btrfs_root *root)
3681{
3682 struct btrfs_fs_info *fs_info = root->fs_info;
3683 struct btrfs_trans_handle *trans;
3684 int ret, err;
3685
3686 trans = btrfs_start_transaction(fs_info->tree_root, 0);
3687 if (IS_ERR(trans))
3688 return PTR_ERR(trans);
3689
3690 memset(&root->root_item.drop_progress, 0,
3691 sizeof(root->root_item.drop_progress));
3692 root->root_item.drop_level = 0;
3693 btrfs_set_root_refs(&root->root_item, 0);
3694 ret = btrfs_update_root(trans, fs_info->tree_root,
3695 &root->root_key, &root->root_item);
3696
3697 err = btrfs_end_transaction(trans);
3698 if (err)
3699 return err;
3700 return ret;
3701}
3702
3703/*
3704 * recover relocation interrupted by system crash.
3705 *
3706 * this function resumes merging reloc trees with corresponding fs trees.
3707 * this is important for keeping the sharing of tree blocks
3708 */
3709int btrfs_recover_relocation(struct btrfs_root *root)
3710{
3711 struct btrfs_fs_info *fs_info = root->fs_info;
3712 LIST_HEAD(reloc_roots);
3713 struct btrfs_key key;
3714 struct btrfs_root *fs_root;
3715 struct btrfs_root *reloc_root;
3716 struct btrfs_path *path;
3717 struct extent_buffer *leaf;
3718 struct reloc_control *rc = NULL;
3719 struct btrfs_trans_handle *trans;
3720 int ret;
3721 int err = 0;
3722
3723 path = btrfs_alloc_path();
3724 if (!path)
3725 return -ENOMEM;
3726 path->reada = READA_BACK;
3727
3728 key.objectid = BTRFS_TREE_RELOC_OBJECTID;
3729 key.type = BTRFS_ROOT_ITEM_KEY;
3730 key.offset = (u64)-1;
3731
3732 while (1) {
3733 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key,
3734 path, 0, 0);
3735 if (ret < 0) {
3736 err = ret;
3737 goto out;
3738 }
3739 if (ret > 0) {
3740 if (path->slots[0] == 0)
3741 break;
3742 path->slots[0]--;
3743 }
3744 leaf = path->nodes[0];
3745 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3746 btrfs_release_path(path);
3747
3748 if (key.objectid != BTRFS_TREE_RELOC_OBJECTID ||
3749 key.type != BTRFS_ROOT_ITEM_KEY)
3750 break;
3751
3752 reloc_root = btrfs_read_tree_root(root, &key);
3753 if (IS_ERR(reloc_root)) {
3754 err = PTR_ERR(reloc_root);
3755 goto out;
3756 }
3757
3758 set_bit(BTRFS_ROOT_SHAREABLE, &reloc_root->state);
3759 list_add(&reloc_root->root_list, &reloc_roots);
3760
3761 if (btrfs_root_refs(&reloc_root->root_item) > 0) {
3762 fs_root = btrfs_get_fs_root(fs_info,
3763 reloc_root->root_key.offset, false);
3764 if (IS_ERR(fs_root)) {
3765 ret = PTR_ERR(fs_root);
3766 if (ret != -ENOENT) {
3767 err = ret;
3768 goto out;
3769 }
3770 ret = mark_garbage_root(reloc_root);
3771 if (ret < 0) {
3772 err = ret;
3773 goto out;
3774 }
3775 } else {
3776 btrfs_put_root(fs_root);
3777 }
3778 }
3779
3780 if (key.offset == 0)
3781 break;
3782
3783 key.offset--;
3784 }
3785 btrfs_release_path(path);
3786
3787 if (list_empty(&reloc_roots))
3788 goto out;
3789
3790 rc = alloc_reloc_control(fs_info);
3791 if (!rc) {
3792 err = -ENOMEM;
3793 goto out;
3794 }
3795
3796 rc->extent_root = fs_info->extent_root;
3797
3798 set_reloc_control(rc);
3799
3800 trans = btrfs_join_transaction(rc->extent_root);
3801 if (IS_ERR(trans)) {
3802 err = PTR_ERR(trans);
3803 goto out_unset;
3804 }
3805
3806 rc->merge_reloc_tree = 1;
3807
3808 while (!list_empty(&reloc_roots)) {
3809 reloc_root = list_entry(reloc_roots.next,
3810 struct btrfs_root, root_list);
3811 list_del(&reloc_root->root_list);
3812
3813 if (btrfs_root_refs(&reloc_root->root_item) == 0) {
3814 list_add_tail(&reloc_root->root_list,
3815 &rc->reloc_roots);
3816 continue;
3817 }
3818
3819 fs_root = btrfs_get_fs_root(fs_info, reloc_root->root_key.offset,
3820 false);
3821 if (IS_ERR(fs_root)) {
3822 err = PTR_ERR(fs_root);
3823 list_add_tail(&reloc_root->root_list, &reloc_roots);
3824 btrfs_end_transaction(trans);
3825 goto out_unset;
3826 }
3827
3828 err = __add_reloc_root(reloc_root);
3829 BUG_ON(err < 0); /* -ENOMEM or logic error */
3830 fs_root->reloc_root = btrfs_grab_root(reloc_root);
3831 btrfs_put_root(fs_root);
3832 }
3833
3834 err = btrfs_commit_transaction(trans);
3835 if (err)
3836 goto out_unset;
3837
3838 merge_reloc_roots(rc);
3839
3840 unset_reloc_control(rc);
3841
3842 trans = btrfs_join_transaction(rc->extent_root);
3843 if (IS_ERR(trans)) {
3844 err = PTR_ERR(trans);
3845 goto out_clean;
3846 }
3847 err = btrfs_commit_transaction(trans);
3848out_clean:
3849 ret = clean_dirty_subvols(rc);
3850 if (ret < 0 && !err)
3851 err = ret;
3852out_unset:
3853 unset_reloc_control(rc);
3854 free_reloc_control(rc);
3855out:
3856 free_reloc_roots(&reloc_roots);
3857
3858 btrfs_free_path(path);
3859
3860 if (err == 0) {
3861 /* cleanup orphan inode in data relocation tree */
3862 fs_root = btrfs_grab_root(fs_info->data_reloc_root);
3863 ASSERT(fs_root);
3864 err = btrfs_orphan_cleanup(fs_root);
3865 btrfs_put_root(fs_root);
3866 }
3867 return err;
3868}
3869
3870/*
3871 * helper to add ordered checksum for data relocation.
3872 *
3873 * cloning checksum properly handles the nodatasum extents.
3874 * it also saves CPU time to re-calculate the checksum.
3875 */
3876int btrfs_reloc_clone_csums(struct btrfs_inode *inode, u64 file_pos, u64 len)
3877{
3878 struct btrfs_fs_info *fs_info = inode->root->fs_info;
3879 struct btrfs_ordered_sum *sums;
3880 struct btrfs_ordered_extent *ordered;
3881 int ret;
3882 u64 disk_bytenr;
3883 u64 new_bytenr;
3884 LIST_HEAD(list);
3885
3886 ordered = btrfs_lookup_ordered_extent(inode, file_pos);
3887 BUG_ON(ordered->file_offset != file_pos || ordered->num_bytes != len);
3888
3889 disk_bytenr = file_pos + inode->index_cnt;
3890 ret = btrfs_lookup_csums_range(fs_info->csum_root, disk_bytenr,
3891 disk_bytenr + len - 1, &list, 0);
3892 if (ret)
3893 goto out;
3894
3895 while (!list_empty(&list)) {
3896 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
3897 list_del_init(&sums->list);
3898
3899 /*
3900 * We need to offset the new_bytenr based on where the csum is.
3901 * We need to do this because we will read in entire prealloc
3902 * extents but we may have written to say the middle of the
3903 * prealloc extent, so we need to make sure the csum goes with
3904 * the right disk offset.
3905 *
3906 * We can do this because the data reloc inode refers strictly
3907 * to the on disk bytes, so we don't have to worry about
3908 * disk_len vs real len like with real inodes since it's all
3909 * disk length.
3910 */
3911 new_bytenr = ordered->disk_bytenr + sums->bytenr - disk_bytenr;
3912 sums->bytenr = new_bytenr;
3913
3914 btrfs_add_ordered_sum(ordered, sums);
3915 }
3916out:
3917 btrfs_put_ordered_extent(ordered);
3918 return ret;
3919}
3920
3921int btrfs_reloc_cow_block(struct btrfs_trans_handle *trans,
3922 struct btrfs_root *root, struct extent_buffer *buf,
3923 struct extent_buffer *cow)
3924{
3925 struct btrfs_fs_info *fs_info = root->fs_info;
3926 struct reloc_control *rc;
3927 struct btrfs_backref_node *node;
3928 int first_cow = 0;
3929 int level;
3930 int ret = 0;
3931
3932 rc = fs_info->reloc_ctl;
3933 if (!rc)
3934 return 0;
3935
3936 BUG_ON(rc->stage == UPDATE_DATA_PTRS &&
3937 root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID);
3938
3939 level = btrfs_header_level(buf);
3940 if (btrfs_header_generation(buf) <=
3941 btrfs_root_last_snapshot(&root->root_item))
3942 first_cow = 1;
3943
3944 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID &&
3945 rc->create_reloc_tree) {
3946 WARN_ON(!first_cow && level == 0);
3947
3948 node = rc->backref_cache.path[level];
3949 BUG_ON(node->bytenr != buf->start &&
3950 node->new_bytenr != buf->start);
3951
3952 btrfs_backref_drop_node_buffer(node);
3953 atomic_inc(&cow->refs);
3954 node->eb = cow;
3955 node->new_bytenr = cow->start;
3956
3957 if (!node->pending) {
3958 list_move_tail(&node->list,
3959 &rc->backref_cache.pending[level]);
3960 node->pending = 1;
3961 }
3962
3963 if (first_cow)
3964 mark_block_processed(rc, node);
3965
3966 if (first_cow && level > 0)
3967 rc->nodes_relocated += buf->len;
3968 }
3969
3970 if (level == 0 && first_cow && rc->stage == UPDATE_DATA_PTRS)
3971 ret = replace_file_extents(trans, rc, root, cow);
3972 return ret;
3973}
3974
3975/*
3976 * called before creating snapshot. it calculates metadata reservation
3977 * required for relocating tree blocks in the snapshot
3978 */
3979void btrfs_reloc_pre_snapshot(struct btrfs_pending_snapshot *pending,
3980 u64 *bytes_to_reserve)
3981{
3982 struct btrfs_root *root = pending->root;
3983 struct reloc_control *rc = root->fs_info->reloc_ctl;
3984
3985 if (!rc || !have_reloc_root(root))
3986 return;
3987
3988 if (!rc->merge_reloc_tree)
3989 return;
3990
3991 root = root->reloc_root;
3992 BUG_ON(btrfs_root_refs(&root->root_item) == 0);
3993 /*
3994 * relocation is in the stage of merging trees. the space
3995 * used by merging a reloc tree is twice the size of
3996 * relocated tree nodes in the worst case. half for cowing
3997 * the reloc tree, half for cowing the fs tree. the space
3998 * used by cowing the reloc tree will be freed after the
3999 * tree is dropped. if we create snapshot, cowing the fs
4000 * tree may use more space than it frees. so we need
4001 * reserve extra space.
4002 */
4003 *bytes_to_reserve += rc->nodes_relocated;
4004}
4005
4006/*
4007 * called after snapshot is created. migrate block reservation
4008 * and create reloc root for the newly created snapshot
4009 *
4010 * This is similar to btrfs_init_reloc_root(), we come out of here with two
4011 * references held on the reloc_root, one for root->reloc_root and one for
4012 * rc->reloc_roots.
4013 */
4014int btrfs_reloc_post_snapshot(struct btrfs_trans_handle *trans,
4015 struct btrfs_pending_snapshot *pending)
4016{
4017 struct btrfs_root *root = pending->root;
4018 struct btrfs_root *reloc_root;
4019 struct btrfs_root *new_root;
4020 struct reloc_control *rc = root->fs_info->reloc_ctl;
4021 int ret;
4022
4023 if (!rc || !have_reloc_root(root))
4024 return 0;
4025
4026 rc = root->fs_info->reloc_ctl;
4027 rc->merging_rsv_size += rc->nodes_relocated;
4028
4029 if (rc->merge_reloc_tree) {
4030 ret = btrfs_block_rsv_migrate(&pending->block_rsv,
4031 rc->block_rsv,
4032 rc->nodes_relocated, true);
4033 if (ret)
4034 return ret;
4035 }
4036
4037 new_root = pending->snap;
4038 reloc_root = create_reloc_root(trans, root->reloc_root,
4039 new_root->root_key.objectid);
4040 if (IS_ERR(reloc_root))
4041 return PTR_ERR(reloc_root);
4042
4043 ret = __add_reloc_root(reloc_root);
4044 BUG_ON(ret < 0);
4045 new_root->reloc_root = btrfs_grab_root(reloc_root);
4046
4047 if (rc->create_reloc_tree)
4048 ret = clone_backref_node(trans, rc, root, reloc_root);
4049 return ret;
4050}