projects / deliverable / linux.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
btrfs: fix return value check of btrfs_start_transaction()
[deliverable/linux.git] / fs / btrfs / relocation.c
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
34 /*
35 * backref_node, mapping_node and tree_block start with this
36 */
37 struct tree_entry {
38 struct rb_node rb_node;
39 u64 bytenr;
40 };
41
42 /*
43 * present a tree block in the backref cache
44 */
45 struct backref_node {
46 struct rb_node rb_node;
47 u64 bytenr;
48
49 u64 new_bytenr;
50 /* objectid of tree block owner, can be not uptodate */
51 u64 owner;
52 /* link to pending, changed or detached list */
53 struct list_head list;
54 /* list of upper level blocks reference this block */
55 struct list_head upper;
56 /* list of child blocks in the cache */
57 struct list_head lower;
58 /* NULL if this node is not tree root */
59 struct btrfs_root *root;
60 /* extent buffer got by COW the block */
61 struct extent_buffer *eb;
62 /* level of tree block */
63 unsigned int level:8;
64 /* is the block in non-reference counted tree */
65 unsigned int cowonly:1;
66 /* 1 if no child node in the cache */
67 unsigned int lowest:1;
68 /* is the extent buffer locked */
69 unsigned int locked:1;
70 /* has the block been processed */
71 unsigned int processed:1;
72 /* have backrefs of this block been checked */
73 unsigned int checked:1;
74 /*
75 * 1 if corresponding block has been cowed but some upper
76 * level block pointers may not point to the new location
77 */
78 unsigned int pending:1;
79 /*
80 * 1 if the backref node isn't connected to any other
81 * backref node.
82 */
83 unsigned int detached:1;
84 };
85
86 /*
87 * present a block pointer in the backref cache
88 */
89 struct backref_edge {
90 struct list_head list[2];
91 struct backref_node *node[2];
92 };
93
94 #define LOWER 0
95 #define UPPER 1
96
97 struct backref_cache {
98 /* red black tree of all backref nodes in the cache */
99 struct rb_root rb_root;
100 /* for passing backref nodes to btrfs_reloc_cow_block */
101 struct backref_node *path[BTRFS_MAX_LEVEL];
102 /*
103 * list of blocks that have been cowed but some block
104 * pointers in upper level blocks may not reflect the
105 * new location
106 */
107 struct list_head pending[BTRFS_MAX_LEVEL];
108 /* list of backref nodes with no child node */
109 struct list_head leaves;
110 /* list of blocks that have been cowed in current transaction */
111 struct list_head changed;
112 /* list of detached backref node. */
113 struct list_head detached;
114
115 u64 last_trans;
116
117 int nr_nodes;
118 int nr_edges;
119 };
120
121 /*
122 * map address of tree root to tree
123 */
124 struct mapping_node {
125 struct rb_node rb_node;
126 u64 bytenr;
127 void *data;
128 };
129
130 struct mapping_tree {
131 struct rb_root rb_root;
132 spinlock_t lock;
133 };
134
135 /*
136 * present a tree block to process
137 */
138 struct tree_block {
139 struct rb_node rb_node;
140 u64 bytenr;
141 struct btrfs_key key;
142 unsigned int level:8;
143 unsigned int key_ready:1;
144 };
145
146 #define MAX_EXTENTS 128
147
148 struct file_extent_cluster {
149 u64 start;
150 u64 end;
151 u64 boundary[MAX_EXTENTS];
152 unsigned int nr;
153 };
154
155 struct reloc_control {
156 /* block group to relocate */
157 struct btrfs_block_group_cache *block_group;
158 /* extent tree */
159 struct btrfs_root *extent_root;
160 /* inode for moving data */
161 struct inode *data_inode;
162
163 struct btrfs_block_rsv *block_rsv;
164
165 struct backref_cache backref_cache;
166
167 struct file_extent_cluster cluster;
168 /* tree blocks have been processed */
169 struct extent_io_tree processed_blocks;
170 /* map start of tree root to corresponding reloc tree */
171 struct mapping_tree reloc_root_tree;
172 /* list of reloc trees */
173 struct list_head reloc_roots;
174 /* size of metadata reservation for merging reloc trees */
175 u64 merging_rsv_size;
176 /* size of relocated tree nodes */
177 u64 nodes_relocated;
178
179 u64 search_start;
180 u64 extents_found;
181
182 unsigned int stage:8;
183 unsigned int create_reloc_tree:1;
184 unsigned int merge_reloc_tree:1;
185 unsigned int found_file_extent:1;
186 unsigned int commit_transaction:1;
187 };
188
189 /* stages of data relocation */
190 #define MOVE_DATA_EXTENTS 0
191 #define UPDATE_DATA_PTRS 1
192
193 static void remove_backref_node(struct backref_cache *cache,
194 struct backref_node *node);
195 static void __mark_block_processed(struct reloc_control *rc,
196 struct backref_node *node);
197
198 static void mapping_tree_init(struct mapping_tree *tree)
199 {
200 tree->rb_root = RB_ROOT;
201 spin_lock_init(&tree->lock);
202 }
203
204 static void backref_cache_init(struct backref_cache *cache)
205 {
206 int i;
207 cache->rb_root = RB_ROOT;
208 for (i = 0; i < BTRFS_MAX_LEVEL; i++)
209 INIT_LIST_HEAD(&cache->pending[i]);
210 INIT_LIST_HEAD(&cache->changed);
211 INIT_LIST_HEAD(&cache->detached);
212 INIT_LIST_HEAD(&cache->leaves);
213 }
214
215 static void backref_cache_cleanup(struct backref_cache *cache)
216 {
217 struct backref_node *node;
218 int i;
219
220 while (!list_empty(&cache->detached)) {
221 node = list_entry(cache->detached.next,
222 struct backref_node, list);
223 remove_backref_node(cache, node);
224 }
225
226 while (!list_empty(&cache->leaves)) {
227 node = list_entry(cache->leaves.next,
228 struct backref_node, lower);
229 remove_backref_node(cache, node);
230 }
231
232 cache->last_trans = 0;
233
234 for (i = 0; i < BTRFS_MAX_LEVEL; i++)
235 BUG_ON(!list_empty(&cache->pending[i]));
236 BUG_ON(!list_empty(&cache->changed));
237 BUG_ON(!list_empty(&cache->detached));
238 BUG_ON(!RB_EMPTY_ROOT(&cache->rb_root));
239 BUG_ON(cache->nr_nodes);
240 BUG_ON(cache->nr_edges);
241 }
242
243 static struct backref_node *alloc_backref_node(struct backref_cache *cache)
244 {
245 struct backref_node *node;
246
247 node = kzalloc(sizeof(*node), GFP_NOFS);
248 if (node) {
249 INIT_LIST_HEAD(&node->list);
250 INIT_LIST_HEAD(&node->upper);
251 INIT_LIST_HEAD(&node->lower);
252 RB_CLEAR_NODE(&node->rb_node);
253 cache->nr_nodes++;
254 }
255 return node;
256 }
257
258 static void free_backref_node(struct backref_cache *cache,
259 struct backref_node *node)
260 {
261 if (node) {
262 cache->nr_nodes--;
263 kfree(node);
264 }
265 }
266
267 static struct backref_edge *alloc_backref_edge(struct backref_cache *cache)
268 {
269 struct backref_edge *edge;
270
271 edge = kzalloc(sizeof(*edge), GFP_NOFS);
272 if (edge)
273 cache->nr_edges++;
274 return edge;
275 }
276
277 static void free_backref_edge(struct backref_cache *cache,
278 struct backref_edge *edge)
279 {
280 if (edge) {
281 cache->nr_edges--;
282 kfree(edge);
283 }
284 }
285
286 static struct rb_node *tree_insert(struct rb_root *root, u64 bytenr,
287 struct rb_node *node)
288 {
289 struct rb_node **p = &root->rb_node;
290 struct rb_node *parent = NULL;
291 struct tree_entry *entry;
292
293 while (*p) {
294 parent = *p;
295 entry = rb_entry(parent, struct tree_entry, rb_node);
296
297 if (bytenr < entry->bytenr)
298 p = &(*p)->rb_left;
299 else if (bytenr > entry->bytenr)
300 p = &(*p)->rb_right;
301 else
302 return parent;
303 }
304
305 rb_link_node(node, parent, p);
306 rb_insert_color(node, root);
307 return NULL;
308 }
309
310 static struct rb_node *tree_search(struct rb_root *root, u64 bytenr)
311 {
312 struct rb_node *n = root->rb_node;
313 struct tree_entry *entry;
314
315 while (n) {
316 entry = rb_entry(n, struct tree_entry, rb_node);
317
318 if (bytenr < entry->bytenr)
319 n = n->rb_left;
320 else if (bytenr > entry->bytenr)
321 n = n->rb_right;
322 else
323 return n;
324 }
325 return NULL;
326 }
327
328 /*
329 * walk up backref nodes until reach node presents tree root
330 */
331 static struct backref_node *walk_up_backref(struct backref_node *node,
332 struct backref_edge *edges[],
333 int *index)
334 {
335 struct backref_edge *edge;
336 int idx = *index;
337
338 while (!list_empty(&node->upper)) {
339 edge = list_entry(node->upper.next,
340 struct backref_edge, list[LOWER]);
341 edges[idx++] = edge;
342 node = edge->node[UPPER];
343 }
344 BUG_ON(node->detached);
345 *index = idx;
346 return node;
347 }
348
349 /*
350 * walk down backref nodes to find start of next reference path
351 */
352 static struct backref_node *walk_down_backref(struct backref_edge *edges[],
353 int *index)
354 {
355 struct backref_edge *edge;
356 struct backref_node *lower;
357 int idx = *index;
358
359 while (idx > 0) {
360 edge = edges[idx - 1];
361 lower = edge->node[LOWER];
362 if (list_is_last(&edge->list[LOWER], &lower->upper)) {
363 idx--;
364 continue;
365 }
366 edge = list_entry(edge->list[LOWER].next,
367 struct backref_edge, list[LOWER]);
368 edges[idx - 1] = edge;
369 *index = idx;
370 return edge->node[UPPER];
371 }
372 *index = 0;
373 return NULL;
374 }
375
376 static void unlock_node_buffer(struct backref_node *node)
377 {
378 if (node->locked) {
379 btrfs_tree_unlock(node->eb);
380 node->locked = 0;
381 }
382 }
383
384 static void drop_node_buffer(struct backref_node *node)
385 {
386 if (node->eb) {
387 unlock_node_buffer(node);
388 free_extent_buffer(node->eb);
389 node->eb = NULL;
390 }
391 }
392
393 static void drop_backref_node(struct backref_cache *tree,
394 struct backref_node *node)
395 {
396 BUG_ON(!list_empty(&node->upper));
397
398 drop_node_buffer(node);
399 list_del(&node->list);
400 list_del(&node->lower);
401 if (!RB_EMPTY_NODE(&node->rb_node))
402 rb_erase(&node->rb_node, &tree->rb_root);
403 free_backref_node(tree, node);
404 }
405
406 /*
407 * remove a backref node from the backref cache
408 */
409 static void remove_backref_node(struct backref_cache *cache,
410 struct backref_node *node)
411 {
412 struct backref_node *upper;
413 struct backref_edge *edge;
414
415 if (!node)
416 return;
417
418 BUG_ON(!node->lowest && !node->detached);
419 while (!list_empty(&node->upper)) {
420 edge = list_entry(node->upper.next, struct backref_edge,
421 list[LOWER]);
422 upper = edge->node[UPPER];
423 list_del(&edge->list[LOWER]);
424 list_del(&edge->list[UPPER]);
425 free_backref_edge(cache, edge);
426
427 if (RB_EMPTY_NODE(&upper->rb_node)) {
428 BUG_ON(!list_empty(&node->upper));
429 drop_backref_node(cache, node);
430 node = upper;
431 node->lowest = 1;
432 continue;
433 }
434 /*
435 * add the node to leaf node list if no other
436 * child block cached.
437 */
438 if (list_empty(&upper->lower)) {
439 list_add_tail(&upper->lower, &cache->leaves);
440 upper->lowest = 1;
441 }
442 }
443
444 drop_backref_node(cache, node);
445 }
446
447 static void update_backref_node(struct backref_cache *cache,
448 struct backref_node *node, u64 bytenr)
449 {
450 struct rb_node *rb_node;
451 rb_erase(&node->rb_node, &cache->rb_root);
452 node->bytenr = bytenr;
453 rb_node = tree_insert(&cache->rb_root, node->bytenr, &node->rb_node);
454 BUG_ON(rb_node);
455 }
456
457 /*
458 * update backref cache after a transaction commit
459 */
460 static int update_backref_cache(struct btrfs_trans_handle *trans,
461 struct backref_cache *cache)
462 {
463 struct backref_node *node;
464 int level = 0;
465
466 if (cache->last_trans == 0) {
467 cache->last_trans = trans->transid;
468 return 0;
469 }
470
471 if (cache->last_trans == trans->transid)
472 return 0;
473
474 /*
475 * detached nodes are used to avoid unnecessary backref
476 * lookup. transaction commit changes the extent tree.
477 * so the detached nodes are no longer useful.
478 */
479 while (!list_empty(&cache->detached)) {
480 node = list_entry(cache->detached.next,
481 struct backref_node, list);
482 remove_backref_node(cache, node);
483 }
484
485 while (!list_empty(&cache->changed)) {
486 node = list_entry(cache->changed.next,
487 struct backref_node, list);
488 list_del_init(&node->list);
489 BUG_ON(node->pending);
490 update_backref_node(cache, node, node->new_bytenr);
491 }
492
493 /*
494 * some nodes can be left in the pending list if there were
495 * errors during processing the pending nodes.
496 */
497 for (level = 0; level < BTRFS_MAX_LEVEL; level++) {
498 list_for_each_entry(node, &cache->pending[level], list) {
499 BUG_ON(!node->pending);
500 if (node->bytenr == node->new_bytenr)
501 continue;
502 update_backref_node(cache, node, node->new_bytenr);
503 }
504 }
505
506 cache->last_trans = 0;
507 return 1;
508 }
509
510 static int should_ignore_root(struct btrfs_root *root)
511 {
512 struct btrfs_root *reloc_root;
513
514 if (!root->ref_cows)
515 return 0;
516
517 reloc_root = root->reloc_root;
518 if (!reloc_root)
519 return 0;
520
521 if (btrfs_root_last_snapshot(&reloc_root->root_item) ==
522 root->fs_info->running_transaction->transid - 1)
523 return 0;
524 /*
525 * if there is reloc tree and it was created in previous
526 * transaction backref lookup can find the reloc tree,
527 * so backref node for the fs tree root is useless for
528 * relocation.
529 */
530 return 1;
531 }
532
533 /*
534 * find reloc tree by address of tree root
535 */
536 static struct btrfs_root *find_reloc_root(struct reloc_control *rc,
537 u64 bytenr)
538 {
539 struct rb_node *rb_node;
540 struct mapping_node *node;
541 struct btrfs_root *root = NULL;
542
543 spin_lock(&rc->reloc_root_tree.lock);
544 rb_node = tree_search(&rc->reloc_root_tree.rb_root, bytenr);
545 if (rb_node) {
546 node = rb_entry(rb_node, struct mapping_node, rb_node);
547 root = (struct btrfs_root *)node->data;
548 }
549 spin_unlock(&rc->reloc_root_tree.lock);
550 return root;
551 }
552
553 static int is_cowonly_root(u64 root_objectid)
554 {
555 if (root_objectid == BTRFS_ROOT_TREE_OBJECTID ||
556 root_objectid == BTRFS_EXTENT_TREE_OBJECTID ||
557 root_objectid == BTRFS_CHUNK_TREE_OBJECTID ||
558 root_objectid == BTRFS_DEV_TREE_OBJECTID ||
559 root_objectid == BTRFS_TREE_LOG_OBJECTID ||
560 root_objectid == BTRFS_CSUM_TREE_OBJECTID)
561 return 1;
562 return 0;
563 }
564
565 static struct btrfs_root *read_fs_root(struct btrfs_fs_info *fs_info,
566 u64 root_objectid)
567 {
568 struct btrfs_key key;
569
570 key.objectid = root_objectid;
571 key.type = BTRFS_ROOT_ITEM_KEY;
572 if (is_cowonly_root(root_objectid))
573 key.offset = 0;
574 else
575 key.offset = (u64)-1;
576
577 return btrfs_read_fs_root_no_name(fs_info, &key);
578 }
579
580 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
581 static noinline_for_stack
582 struct btrfs_root *find_tree_root(struct reloc_control *rc,
583 struct extent_buffer *leaf,
584 struct btrfs_extent_ref_v0 *ref0)
585 {
586 struct btrfs_root *root;
587 u64 root_objectid = btrfs_ref_root_v0(leaf, ref0);
588 u64 generation = btrfs_ref_generation_v0(leaf, ref0);
589
590 BUG_ON(root_objectid == BTRFS_TREE_RELOC_OBJECTID);
591
592 root = read_fs_root(rc->extent_root->fs_info, root_objectid);
593 BUG_ON(IS_ERR(root));
594
595 if (root->ref_cows &&
596 generation != btrfs_root_generation(&root->root_item))
597 return NULL;
598
599 return root;
600 }
601 #endif
602
603 static noinline_for_stack
604 int find_inline_backref(struct extent_buffer *leaf, int slot,
605 unsigned long *ptr, unsigned long *end)
606 {
607 struct btrfs_extent_item *ei;
608 struct btrfs_tree_block_info *bi;
609 u32 item_size;
610
611 item_size = btrfs_item_size_nr(leaf, slot);
612 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
613 if (item_size < sizeof(*ei)) {
614 WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0));
615 return 1;
616 }
617 #endif
618 ei = btrfs_item_ptr(leaf, slot, struct btrfs_extent_item);
619 WARN_ON(!(btrfs_extent_flags(leaf, ei) &
620 BTRFS_EXTENT_FLAG_TREE_BLOCK));
621
622 if (item_size <= sizeof(*ei) + sizeof(*bi)) {
623 WARN_ON(item_size < sizeof(*ei) + sizeof(*bi));
624 return 1;
625 }
626
627 bi = (struct btrfs_tree_block_info *)(ei + 1);
628 *ptr = (unsigned long)(bi + 1);
629 *end = (unsigned long)ei + item_size;
630 return 0;
631 }
632
633 /*
634 * build backref tree for a given tree block. root of the backref tree
635 * corresponds the tree block, leaves of the backref tree correspond
636 * roots of b-trees that reference the tree block.
637 *
638 * the basic idea of this function is check backrefs of a given block
639 * to find upper level blocks that refernece the block, and then check
640 * bakcrefs of these upper level blocks recursively. the recursion stop
641 * when tree root is reached or backrefs for the block is cached.
642 *
643 * NOTE: if we find backrefs for a block are cached, we know backrefs
644 * for all upper level blocks that directly/indirectly reference the
645 * block are also cached.
646 */
647 static noinline_for_stack
648 struct backref_node *build_backref_tree(struct reloc_control *rc,
649 struct btrfs_key *node_key,
650 int level, u64 bytenr)
651 {
652 struct backref_cache *cache = &rc->backref_cache;
653 struct btrfs_path *path1;
654 struct btrfs_path *path2;
655 struct extent_buffer *eb;
656 struct btrfs_root *root;
657 struct backref_node *cur;
658 struct backref_node *upper;
659 struct backref_node *lower;
660 struct backref_node *node = NULL;
661 struct backref_node *exist = NULL;
662 struct backref_edge *edge;
663 struct rb_node *rb_node;
664 struct btrfs_key key;
665 unsigned long end;
666 unsigned long ptr;
667 LIST_HEAD(list);
668 LIST_HEAD(useless);
669 int cowonly;
670 int ret;
671 int err = 0;
672
673 path1 = btrfs_alloc_path();
674 path2 = btrfs_alloc_path();
675 if (!path1 || !path2) {
676 err = -ENOMEM;
677 goto out;
678 }
679
680 node = alloc_backref_node(cache);
681 if (!node) {
682 err = -ENOMEM;
683 goto out;
684 }
685
686 node->bytenr = bytenr;
687 node->level = level;
688 node->lowest = 1;
689 cur = node;
690 again:
691 end = 0;
692 ptr = 0;
693 key.objectid = cur->bytenr;
694 key.type = BTRFS_EXTENT_ITEM_KEY;
695 key.offset = (u64)-1;
696
697 path1->search_commit_root = 1;
698 path1->skip_locking = 1;
699 ret = btrfs_search_slot(NULL, rc->extent_root, &key, path1,
700 0, 0);
701 if (ret < 0) {
702 err = ret;
703 goto out;
704 }
705 BUG_ON(!ret || !path1->slots[0]);
706
707 path1->slots[0]--;
708
709 WARN_ON(cur->checked);
710 if (!list_empty(&cur->upper)) {
711 /*
712 * the backref was added previously when processsing
713 * backref of type BTRFS_TREE_BLOCK_REF_KEY
714 */
715 BUG_ON(!list_is_singular(&cur->upper));
716 edge = list_entry(cur->upper.next, struct backref_edge,
717 list[LOWER]);
718 BUG_ON(!list_empty(&edge->list[UPPER]));
719 exist = edge->node[UPPER];
720 /*
721 * add the upper level block to pending list if we need
722 * check its backrefs
723 */
724 if (!exist->checked)
725 list_add_tail(&edge->list[UPPER], &list);
726 } else {
727 exist = NULL;
728 }
729
730 while (1) {
731 cond_resched();
732 eb = path1->nodes[0];
733
734 if (ptr >= end) {
735 if (path1->slots[0] >= btrfs_header_nritems(eb)) {
736 ret = btrfs_next_leaf(rc->extent_root, path1);
737 if (ret < 0) {
738 err = ret;
739 goto out;
740 }
741 if (ret > 0)
742 break;
743 eb = path1->nodes[0];
744 }
745
746 btrfs_item_key_to_cpu(eb, &key, path1->slots[0]);
747 if (key.objectid != cur->bytenr) {
748 WARN_ON(exist);
749 break;
750 }
751
752 if (key.type == BTRFS_EXTENT_ITEM_KEY) {
753 ret = find_inline_backref(eb, path1->slots[0],
754 &ptr, &end);
755 if (ret)
756 goto next;
757 }
758 }
759
760 if (ptr < end) {
761 /* update key for inline back ref */
762 struct btrfs_extent_inline_ref *iref;
763 iref = (struct btrfs_extent_inline_ref *)ptr;
764 key.type = btrfs_extent_inline_ref_type(eb, iref);
765 key.offset = btrfs_extent_inline_ref_offset(eb, iref);
766 WARN_ON(key.type != BTRFS_TREE_BLOCK_REF_KEY &&
767 key.type != BTRFS_SHARED_BLOCK_REF_KEY);
768 }
769
770 if (exist &&
771 ((key.type == BTRFS_TREE_BLOCK_REF_KEY &&
772 exist->owner == key.offset) ||
773 (key.type == BTRFS_SHARED_BLOCK_REF_KEY &&
774 exist->bytenr == key.offset))) {
775 exist = NULL;
776 goto next;
777 }
778
779 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
780 if (key.type == BTRFS_SHARED_BLOCK_REF_KEY ||
781 key.type == BTRFS_EXTENT_REF_V0_KEY) {
782 if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
783 struct btrfs_extent_ref_v0 *ref0;
784 ref0 = btrfs_item_ptr(eb, path1->slots[0],
785 struct btrfs_extent_ref_v0);
786 if (key.objectid == key.offset) {
787 root = find_tree_root(rc, eb, ref0);
788 if (root && !should_ignore_root(root))
789 cur->root = root;
790 else
791 list_add(&cur->list, &useless);
792 break;
793 }
794 if (is_cowonly_root(btrfs_ref_root_v0(eb,
795 ref0)))
796 cur->cowonly = 1;
797 }
798 #else
799 BUG_ON(key.type == BTRFS_EXTENT_REF_V0_KEY);
800 if (key.type == BTRFS_SHARED_BLOCK_REF_KEY) {
801 #endif
802 if (key.objectid == key.offset) {
803 /*
804 * only root blocks of reloc trees use
805 * backref of this type.
806 */
807 root = find_reloc_root(rc, cur->bytenr);
808 BUG_ON(!root);
809 cur->root = root;
810 break;
811 }
812
813 edge = alloc_backref_edge(cache);
814 if (!edge) {
815 err = -ENOMEM;
816 goto out;
817 }
818 rb_node = tree_search(&cache->rb_root, key.offset);
819 if (!rb_node) {
820 upper = alloc_backref_node(cache);
821 if (!upper) {
822 free_backref_edge(cache, edge);
823 err = -ENOMEM;
824 goto out;
825 }
826 upper->bytenr = key.offset;
827 upper->level = cur->level + 1;
828 /*
829 * backrefs for the upper level block isn't
830 * cached, add the block to pending list
831 */
832 list_add_tail(&edge->list[UPPER], &list);
833 } else {
834 upper = rb_entry(rb_node, struct backref_node,
835 rb_node);
836 BUG_ON(!upper->checked);
837 INIT_LIST_HEAD(&edge->list[UPPER]);
838 }
839 list_add_tail(&edge->list[LOWER], &cur->upper);
840 edge->node[LOWER] = cur;
841 edge->node[UPPER] = upper;
842
843 goto next;
844 } else if (key.type != BTRFS_TREE_BLOCK_REF_KEY) {
845 goto next;
846 }
847
848 /* key.type == BTRFS_TREE_BLOCK_REF_KEY */
849 root = read_fs_root(rc->extent_root->fs_info, key.offset);
850 if (IS_ERR(root)) {
851 err = PTR_ERR(root);
852 goto out;
853 }
854
855 if (!root->ref_cows)
856 cur->cowonly = 1;
857
858 if (btrfs_root_level(&root->root_item) == cur->level) {
859 /* tree root */
860 BUG_ON(btrfs_root_bytenr(&root->root_item) !=
861 cur->bytenr);
862 if (should_ignore_root(root))
863 list_add(&cur->list, &useless);
864 else
865 cur->root = root;
866 break;
867 }
868
869 level = cur->level + 1;
870
871 /*
872 * searching the tree to find upper level blocks
873 * reference the block.
874 */
875 path2->search_commit_root = 1;
876 path2->skip_locking = 1;
877 path2->lowest_level = level;
878 ret = btrfs_search_slot(NULL, root, node_key, path2, 0, 0);
879 path2->lowest_level = 0;
880 if (ret < 0) {
881 err = ret;
882 goto out;
883 }
884 if (ret > 0 && path2->slots[level] > 0)
885 path2->slots[level]--;
886
887 eb = path2->nodes[level];
888 WARN_ON(btrfs_node_blockptr(eb, path2->slots[level]) !=
889 cur->bytenr);
890
891 lower = cur;
892 for (; level < BTRFS_MAX_LEVEL; level++) {
893 if (!path2->nodes[level]) {
894 BUG_ON(btrfs_root_bytenr(&root->root_item) !=
895 lower->bytenr);
896 if (should_ignore_root(root))
897 list_add(&lower->list, &useless);
898 else
899 lower->root = root;
900 break;
901 }
902
903 edge = alloc_backref_edge(cache);
904 if (!edge) {
905 err = -ENOMEM;
906 goto out;
907 }
908
909 eb = path2->nodes[level];
910 rb_node = tree_search(&cache->rb_root, eb->start);
911 if (!rb_node) {
912 upper = alloc_backref_node(cache);
913 if (!upper) {
914 free_backref_edge(cache, edge);
915 err = -ENOMEM;
916 goto out;
917 }
918 upper->bytenr = eb->start;
919 upper->owner = btrfs_header_owner(eb);
920 upper->level = lower->level + 1;
921 if (!root->ref_cows)
922 upper->cowonly = 1;
923
924 /*
925 * if we know the block isn't shared
926 * we can void checking its backrefs.
927 */
928 if (btrfs_block_can_be_shared(root, eb))
929 upper->checked = 0;
930 else
931 upper->checked = 1;
932
933 /*
934 * add the block to pending list if we
935 * need check its backrefs. only block
936 * at 'cur->level + 1' is added to the
937 * tail of pending list. this guarantees
938 * we check backrefs from lower level
939 * blocks to upper level blocks.
940 */
941 if (!upper->checked &&
942 level == cur->level + 1) {
943 list_add_tail(&edge->list[UPPER],
944 &list);
945 } else
946 INIT_LIST_HEAD(&edge->list[UPPER]);
947 } else {
948 upper = rb_entry(rb_node, struct backref_node,
949 rb_node);
950 BUG_ON(!upper->checked);
951 INIT_LIST_HEAD(&edge->list[UPPER]);
952 if (!upper->owner)
953 upper->owner = btrfs_header_owner(eb);
954 }
955 list_add_tail(&edge->list[LOWER], &lower->upper);
956 edge->node[LOWER] = lower;
957 edge->node[UPPER] = upper;
958
959 if (rb_node)
960 break;
961 lower = upper;
962 upper = NULL;
963 }
964 btrfs_release_path(root, path2);
965 next:
966 if (ptr < end) {
967 ptr += btrfs_extent_inline_ref_size(key.type);
968 if (ptr >= end) {
969 WARN_ON(ptr > end);
970 ptr = 0;
971 end = 0;
972 }
973 }
974 if (ptr >= end)
975 path1->slots[0]++;
976 }
977 btrfs_release_path(rc->extent_root, path1);
978
979 cur->checked = 1;
980 WARN_ON(exist);
981
982 /* the pending list isn't empty, take the first block to process */
983 if (!list_empty(&list)) {
984 edge = list_entry(list.next, struct backref_edge, list[UPPER]);
985 list_del_init(&edge->list[UPPER]);
986 cur = edge->node[UPPER];
987 goto again;
988 }
989
990 /*
991 * everything goes well, connect backref nodes and insert backref nodes
992 * into the cache.
993 */
994 BUG_ON(!node->checked);
995 cowonly = node->cowonly;
996 if (!cowonly) {
997 rb_node = tree_insert(&cache->rb_root, node->bytenr,
998 &node->rb_node);
999 BUG_ON(rb_node);
1000 list_add_tail(&node->lower, &cache->leaves);
1001 }
1002
1003 list_for_each_entry(edge, &node->upper, list[LOWER])
1004 list_add_tail(&edge->list[UPPER], &list);
1005
1006 while (!list_empty(&list)) {
1007 edge = list_entry(list.next, struct backref_edge, list[UPPER]);
1008 list_del_init(&edge->list[UPPER]);
1009 upper = edge->node[UPPER];
1010 if (upper->detached) {
1011 list_del(&edge->list[LOWER]);
1012 lower = edge->node[LOWER];
1013 free_backref_edge(cache, edge);
1014 if (list_empty(&lower->upper))
1015 list_add(&lower->list, &useless);
1016 continue;
1017 }
1018
1019 if (!RB_EMPTY_NODE(&upper->rb_node)) {
1020 if (upper->lowest) {
1021 list_del_init(&upper->lower);
1022 upper->lowest = 0;
1023 }
1024
1025 list_add_tail(&edge->list[UPPER], &upper->lower);
1026 continue;
1027 }
1028
1029 BUG_ON(!upper->checked);
1030 BUG_ON(cowonly != upper->cowonly);
1031 if (!cowonly) {
1032 rb_node = tree_insert(&cache->rb_root, upper->bytenr,
1033 &upper->rb_node);
1034 BUG_ON(rb_node);
1035 }
1036
1037 list_add_tail(&edge->list[UPPER], &upper->lower);
1038
1039 list_for_each_entry(edge, &upper->upper, list[LOWER])
1040 list_add_tail(&edge->list[UPPER], &list);
1041 }
1042 /*
1043 * process useless backref nodes. backref nodes for tree leaves
1044 * are deleted from the cache. backref nodes for upper level
1045 * tree blocks are left in the cache to avoid unnecessary backref
1046 * lookup.
1047 */
1048 while (!list_empty(&useless)) {
1049 upper = list_entry(useless.next, struct backref_node, list);
1050 list_del_init(&upper->list);
1051 BUG_ON(!list_empty(&upper->upper));
1052 if (upper == node)
1053 node = NULL;
1054 if (upper->lowest) {
1055 list_del_init(&upper->lower);
1056 upper->lowest = 0;
1057 }
1058 while (!list_empty(&upper->lower)) {
1059 edge = list_entry(upper->lower.next,
1060 struct backref_edge, list[UPPER]);
1061 list_del(&edge->list[UPPER]);
1062 list_del(&edge->list[LOWER]);
1063 lower = edge->node[LOWER];
1064 free_backref_edge(cache, edge);
1065
1066 if (list_empty(&lower->upper))
1067 list_add(&lower->list, &useless);
1068 }
1069 __mark_block_processed(rc, upper);
1070 if (upper->level > 0) {
1071 list_add(&upper->list, &cache->detached);
1072 upper->detached = 1;
1073 } else {
1074 rb_erase(&upper->rb_node, &cache->rb_root);
1075 free_backref_node(cache, upper);
1076 }
1077 }
1078 out:
1079 btrfs_free_path(path1);
1080 btrfs_free_path(path2);
1081 if (err) {
1082 while (!list_empty(&useless)) {
1083 lower = list_entry(useless.next,
1084 struct backref_node, upper);
1085 list_del_init(&lower->upper);
1086 }
1087 upper = node;
1088 INIT_LIST_HEAD(&list);
1089 while (upper) {
1090 if (RB_EMPTY_NODE(&upper->rb_node)) {
1091 list_splice_tail(&upper->upper, &list);
1092 free_backref_node(cache, upper);
1093 }
1094
1095 if (list_empty(&list))
1096 break;
1097
1098 edge = list_entry(list.next, struct backref_edge,
1099 list[LOWER]);
1100 list_del(&edge->list[LOWER]);
1101 upper = edge->node[UPPER];
1102 free_backref_edge(cache, edge);
1103 }
1104 return ERR_PTR(err);
1105 }
1106 BUG_ON(node && node->detached);
1107 return node;
1108 }
1109
1110 /*
1111 * helper to add backref node for the newly created snapshot.
1112 * the backref node is created by cloning backref node that
1113 * corresponds to root of source tree
1114 */
1115 static int clone_backref_node(struct btrfs_trans_handle *trans,
1116 struct reloc_control *rc,
1117 struct btrfs_root *src,
1118 struct btrfs_root *dest)
1119 {
1120 struct btrfs_root *reloc_root = src->reloc_root;
1121 struct backref_cache *cache = &rc->backref_cache;
1122 struct backref_node *node = NULL;
1123 struct backref_node *new_node;
1124 struct backref_edge *edge;
1125 struct backref_edge *new_edge;
1126 struct rb_node *rb_node;
1127
1128 if (cache->last_trans > 0)
1129 update_backref_cache(trans, cache);
1130
1131 rb_node = tree_search(&cache->rb_root, src->commit_root->start);
1132 if (rb_node) {
1133 node = rb_entry(rb_node, struct backref_node, rb_node);
1134 if (node->detached)
1135 node = NULL;
1136 else
1137 BUG_ON(node->new_bytenr != reloc_root->node->start);
1138 }
1139
1140 if (!node) {
1141 rb_node = tree_search(&cache->rb_root,
1142 reloc_root->commit_root->start);
1143 if (rb_node) {
1144 node = rb_entry(rb_node, struct backref_node,
1145 rb_node);
1146 BUG_ON(node->detached);
1147 }
1148 }
1149
1150 if (!node)
1151 return 0;
1152
1153 new_node = alloc_backref_node(cache);
1154 if (!new_node)
1155 return -ENOMEM;
1156
1157 new_node->bytenr = dest->node->start;
1158 new_node->level = node->level;
1159 new_node->lowest = node->lowest;
1160 new_node->root = dest;
1161
1162 if (!node->lowest) {
1163 list_for_each_entry(edge, &node->lower, list[UPPER]) {
1164 new_edge = alloc_backref_edge(cache);
1165 if (!new_edge)
1166 goto fail;
1167
1168 new_edge->node[UPPER] = new_node;
1169 new_edge->node[LOWER] = edge->node[LOWER];
1170 list_add_tail(&new_edge->list[UPPER],
1171 &new_node->lower);
1172 }
1173 }
1174
1175 rb_node = tree_insert(&cache->rb_root, new_node->bytenr,
1176 &new_node->rb_node);
1177 BUG_ON(rb_node);
1178
1179 if (!new_node->lowest) {
1180 list_for_each_entry(new_edge, &new_node->lower, list[UPPER]) {
1181 list_add_tail(&new_edge->list[LOWER],
1182 &new_edge->node[LOWER]->upper);
1183 }
1184 }
1185 return 0;
1186 fail:
1187 while (!list_empty(&new_node->lower)) {
1188 new_edge = list_entry(new_node->lower.next,
1189 struct backref_edge, list[UPPER]);
1190 list_del(&new_edge->list[UPPER]);
1191 free_backref_edge(cache, new_edge);
1192 }
1193 free_backref_node(cache, new_node);
1194 return -ENOMEM;
1195 }
1196
1197 /*
1198 * helper to add 'address of tree root -> reloc tree' mapping
1199 */
1200 static int __add_reloc_root(struct btrfs_root *root)
1201 {
1202 struct rb_node *rb_node;
1203 struct mapping_node *node;
1204 struct reloc_control *rc = root->fs_info->reloc_ctl;
1205
1206 node = kmalloc(sizeof(*node), GFP_NOFS);
1207 BUG_ON(!node);
1208
1209 node->bytenr = root->node->start;
1210 node->data = root;
1211
1212 spin_lock(&rc->reloc_root_tree.lock);
1213 rb_node = tree_insert(&rc->reloc_root_tree.rb_root,
1214 node->bytenr, &node->rb_node);
1215 spin_unlock(&rc->reloc_root_tree.lock);
1216 BUG_ON(rb_node);
1217
1218 list_add_tail(&root->root_list, &rc->reloc_roots);
1219 return 0;
1220 }
1221
1222 /*
1223 * helper to update/delete the 'address of tree root -> reloc tree'
1224 * mapping
1225 */
1226 static int __update_reloc_root(struct btrfs_root *root, int del)
1227 {
1228 struct rb_node *rb_node;
1229 struct mapping_node *node = NULL;
1230 struct reloc_control *rc = root->fs_info->reloc_ctl;
1231
1232 spin_lock(&rc->reloc_root_tree.lock);
1233 rb_node = tree_search(&rc->reloc_root_tree.rb_root,
1234 root->commit_root->start);
1235 if (rb_node) {
1236 node = rb_entry(rb_node, struct mapping_node, rb_node);
1237 rb_erase(&node->rb_node, &rc->reloc_root_tree.rb_root);
1238 }
1239 spin_unlock(&rc->reloc_root_tree.lock);
1240
1241 BUG_ON((struct btrfs_root *)node->data != root);
1242
1243 if (!del) {
1244 spin_lock(&rc->reloc_root_tree.lock);
1245 node->bytenr = root->node->start;
1246 rb_node = tree_insert(&rc->reloc_root_tree.rb_root,
1247 node->bytenr, &node->rb_node);
1248 spin_unlock(&rc->reloc_root_tree.lock);
1249 BUG_ON(rb_node);
1250 } else {
1251 list_del_init(&root->root_list);
1252 kfree(node);
1253 }
1254 return 0;
1255 }
1256
1257 static struct btrfs_root *create_reloc_root(struct btrfs_trans_handle *trans,
1258 struct btrfs_root *root, u64 objectid)
1259 {
1260 struct btrfs_root *reloc_root;
1261 struct extent_buffer *eb;
1262 struct btrfs_root_item *root_item;
1263 struct btrfs_key root_key;
1264 int ret;
1265
1266 root_item = kmalloc(sizeof(*root_item), GFP_NOFS);
1267 BUG_ON(!root_item);
1268
1269 root_key.objectid = BTRFS_TREE_RELOC_OBJECTID;
1270 root_key.type = BTRFS_ROOT_ITEM_KEY;
1271 root_key.offset = objectid;
1272
1273 if (root->root_key.objectid == objectid) {
1274 /* called by btrfs_init_reloc_root */
1275 ret = btrfs_copy_root(trans, root, root->commit_root, &eb,
1276 BTRFS_TREE_RELOC_OBJECTID);
1277 BUG_ON(ret);
1278
1279 btrfs_set_root_last_snapshot(&root->root_item,
1280 trans->transid - 1);
1281 } else {
1282 /*
1283 * called by btrfs_reloc_post_snapshot_hook.
1284 * the source tree is a reloc tree, all tree blocks
1285 * modified after it was created have RELOC flag
1286 * set in their headers. so it's OK to not update
1287 * the 'last_snapshot'.
1288 */
1289 ret = btrfs_copy_root(trans, root, root->node, &eb,
1290 BTRFS_TREE_RELOC_OBJECTID);
1291 BUG_ON(ret);
1292 }
1293
1294 memcpy(root_item, &root->root_item, sizeof(*root_item));
1295 btrfs_set_root_bytenr(root_item, eb->start);
1296 btrfs_set_root_level(root_item, btrfs_header_level(eb));
1297 btrfs_set_root_generation(root_item, trans->transid);
1298
1299 if (root->root_key.objectid == objectid) {
1300 btrfs_set_root_refs(root_item, 0);
1301 memset(&root_item->drop_progress, 0,
1302 sizeof(struct btrfs_disk_key));
1303 root_item->drop_level = 0;
1304 }
1305
1306 btrfs_tree_unlock(eb);
1307 free_extent_buffer(eb);
1308
1309 ret = btrfs_insert_root(trans, root->fs_info->tree_root,
1310 &root_key, root_item);
1311 BUG_ON(ret);
1312 kfree(root_item);
1313
1314 reloc_root = btrfs_read_fs_root_no_radix(root->fs_info->tree_root,
1315 &root_key);
1316 BUG_ON(IS_ERR(reloc_root));
1317 reloc_root->last_trans = trans->transid;
1318 return reloc_root;
1319 }
1320
1321 /*
1322 * create reloc tree for a given fs tree. reloc tree is just a
1323 * snapshot of the fs tree with special root objectid.
1324 */
1325 int btrfs_init_reloc_root(struct btrfs_trans_handle *trans,
1326 struct btrfs_root *root)
1327 {
1328 struct btrfs_root *reloc_root;
1329 struct reloc_control *rc = root->fs_info->reloc_ctl;
1330 int clear_rsv = 0;
1331
1332 if (root->reloc_root) {
1333 reloc_root = root->reloc_root;
1334 reloc_root->last_trans = trans->transid;
1335 return 0;
1336 }
1337
1338 if (!rc || !rc->create_reloc_tree ||
1339 root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID)
1340 return 0;
1341
1342 if (!trans->block_rsv) {
1343 trans->block_rsv = rc->block_rsv;
1344 clear_rsv = 1;
1345 }
1346 reloc_root = create_reloc_root(trans, root, root->root_key.objectid);
1347 if (clear_rsv)
1348 trans->block_rsv = NULL;
1349
1350 __add_reloc_root(reloc_root);
1351 root->reloc_root = reloc_root;
1352 return 0;
1353 }
1354
1355 /*
1356 * update root item of reloc tree
1357 */
1358 int btrfs_update_reloc_root(struct btrfs_trans_handle *trans,
1359 struct btrfs_root *root)
1360 {
1361 struct btrfs_root *reloc_root;
1362 struct btrfs_root_item *root_item;
1363 int del = 0;
1364 int ret;
1365
1366 if (!root->reloc_root)
1367 return 0;
1368
1369 reloc_root = root->reloc_root;
1370 root_item = &reloc_root->root_item;
1371
1372 if (root->fs_info->reloc_ctl->merge_reloc_tree &&
1373 btrfs_root_refs(root_item) == 0) {
1374 root->reloc_root = NULL;
1375 del = 1;
1376 }
1377
1378 __update_reloc_root(reloc_root, del);
1379
1380 if (reloc_root->commit_root != reloc_root->node) {
1381 btrfs_set_root_node(root_item, reloc_root->node);
1382 free_extent_buffer(reloc_root->commit_root);
1383 reloc_root->commit_root = btrfs_root_node(reloc_root);
1384 }
1385
1386 ret = btrfs_update_root(trans, root->fs_info->tree_root,
1387 &reloc_root->root_key, root_item);
1388 BUG_ON(ret);
1389 return 0;
1390 }
1391
1392 /*
1393 * helper to find first cached inode with inode number >= objectid
1394 * in a subvolume
1395 */
1396 static struct inode *find_next_inode(struct btrfs_root *root, u64 objectid)
1397 {
1398 struct rb_node *node;
1399 struct rb_node *prev;
1400 struct btrfs_inode *entry;
1401 struct inode *inode;
1402
1403 spin_lock(&root->inode_lock);
1404 again:
1405 node = root->inode_tree.rb_node;
1406 prev = NULL;
1407 while (node) {
1408 prev = node;
1409 entry = rb_entry(node, struct btrfs_inode, rb_node);
1410
1411 if (objectid < entry->vfs_inode.i_ino)
1412 node = node->rb_left;
1413 else if (objectid > entry->vfs_inode.i_ino)
1414 node = node->rb_right;
1415 else
1416 break;
1417 }
1418 if (!node) {
1419 while (prev) {
1420 entry = rb_entry(prev, struct btrfs_inode, rb_node);
1421 if (objectid <= entry->vfs_inode.i_ino) {
1422 node = prev;
1423 break;
1424 }
1425 prev = rb_next(prev);
1426 }
1427 }
1428 while (node) {
1429 entry = rb_entry(node, struct btrfs_inode, rb_node);
1430 inode = igrab(&entry->vfs_inode);
1431 if (inode) {
1432 spin_unlock(&root->inode_lock);
1433 return inode;
1434 }
1435
1436 objectid = entry->vfs_inode.i_ino + 1;
1437 if (cond_resched_lock(&root->inode_lock))
1438 goto again;
1439
1440 node = rb_next(node);
1441 }
1442 spin_unlock(&root->inode_lock);
1443 return NULL;
1444 }
1445
1446 static int in_block_group(u64 bytenr,
1447 struct btrfs_block_group_cache *block_group)
1448 {
1449 if (bytenr >= block_group->key.objectid &&
1450 bytenr < block_group->key.objectid + block_group->key.offset)
1451 return 1;
1452 return 0;
1453 }
1454
1455 /*
1456 * get new location of data
1457 */
1458 static int get_new_location(struct inode *reloc_inode, u64 *new_bytenr,
1459 u64 bytenr, u64 num_bytes)
1460 {
1461 struct btrfs_root *root = BTRFS_I(reloc_inode)->root;
1462 struct btrfs_path *path;
1463 struct btrfs_file_extent_item *fi;
1464 struct extent_buffer *leaf;
1465 int ret;
1466
1467 path = btrfs_alloc_path();
1468 if (!path)
1469 return -ENOMEM;
1470
1471 bytenr -= BTRFS_I(reloc_inode)->index_cnt;
1472 ret = btrfs_lookup_file_extent(NULL, root, path, reloc_inode->i_ino,
1473 bytenr, 0);
1474 if (ret < 0)
1475 goto out;
1476 if (ret > 0) {
1477 ret = -ENOENT;
1478 goto out;
1479 }
1480
1481 leaf = path->nodes[0];
1482 fi = btrfs_item_ptr(leaf, path->slots[0],
1483 struct btrfs_file_extent_item);
1484
1485 BUG_ON(btrfs_file_extent_offset(leaf, fi) ||
1486 btrfs_file_extent_compression(leaf, fi) ||
1487 btrfs_file_extent_encryption(leaf, fi) ||
1488 btrfs_file_extent_other_encoding(leaf, fi));
1489
1490 if (num_bytes != btrfs_file_extent_disk_num_bytes(leaf, fi)) {
1491 ret = 1;
1492 goto out;
1493 }
1494
1495 *new_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1496 ret = 0;
1497 out:
1498 btrfs_free_path(path);
1499 return ret;
1500 }
1501
1502 /*
1503 * update file extent items in the tree leaf to point to
1504 * the new locations.
1505 */
1506 static noinline_for_stack
1507 int replace_file_extents(struct btrfs_trans_handle *trans,
1508 struct reloc_control *rc,
1509 struct btrfs_root *root,
1510 struct extent_buffer *leaf)
1511 {
1512 struct btrfs_key key;
1513 struct btrfs_file_extent_item *fi;
1514 struct inode *inode = NULL;
1515 u64 parent;
1516 u64 bytenr;
1517 u64 new_bytenr = 0;
1518 u64 num_bytes;
1519 u64 end;
1520 u32 nritems;
1521 u32 i;
1522 int ret;
1523 int first = 1;
1524 int dirty = 0;
1525
1526 if (rc->stage != UPDATE_DATA_PTRS)
1527 return 0;
1528
1529 /* reloc trees always use full backref */
1530 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID)
1531 parent = leaf->start;
1532 else
1533 parent = 0;
1534
1535 nritems = btrfs_header_nritems(leaf);
1536 for (i = 0; i < nritems; i++) {
1537 cond_resched();
1538 btrfs_item_key_to_cpu(leaf, &key, i);
1539 if (key.type != BTRFS_EXTENT_DATA_KEY)
1540 continue;
1541 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
1542 if (btrfs_file_extent_type(leaf, fi) ==
1543 BTRFS_FILE_EXTENT_INLINE)
1544 continue;
1545 bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1546 num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
1547 if (bytenr == 0)
1548 continue;
1549 if (!in_block_group(bytenr, rc->block_group))
1550 continue;
1551
1552 /*
1553 * if we are modifying block in fs tree, wait for readpage
1554 * to complete and drop the extent cache
1555 */
1556 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
1557 if (first) {
1558 inode = find_next_inode(root, key.objectid);
1559 first = 0;
1560 } else if (inode && inode->i_ino < key.objectid) {
1561 btrfs_add_delayed_iput(inode);
1562 inode = find_next_inode(root, key.objectid);
1563 }
1564 if (inode && inode->i_ino == key.objectid) {
1565 end = key.offset +
1566 btrfs_file_extent_num_bytes(leaf, fi);
1567 WARN_ON(!IS_ALIGNED(key.offset,
1568 root->sectorsize));
1569 WARN_ON(!IS_ALIGNED(end, root->sectorsize));
1570 end--;
1571 ret = try_lock_extent(&BTRFS_I(inode)->io_tree,
1572 key.offset, end,
1573 GFP_NOFS);
1574 if (!ret)
1575 continue;
1576
1577 btrfs_drop_extent_cache(inode, key.offset, end,
1578 1);
1579 unlock_extent(&BTRFS_I(inode)->io_tree,
1580 key.offset, end, GFP_NOFS);
1581 }
1582 }
1583
1584 ret = get_new_location(rc->data_inode, &new_bytenr,
1585 bytenr, num_bytes);
1586 if (ret > 0) {
1587 WARN_ON(1);
1588 continue;
1589 }
1590 BUG_ON(ret < 0);
1591
1592 btrfs_set_file_extent_disk_bytenr(leaf, fi, new_bytenr);
1593 dirty = 1;
1594
1595 key.offset -= btrfs_file_extent_offset(leaf, fi);
1596 ret = btrfs_inc_extent_ref(trans, root, new_bytenr,
1597 num_bytes, parent,
1598 btrfs_header_owner(leaf),
1599 key.objectid, key.offset);
1600 BUG_ON(ret);
1601
1602 ret = btrfs_free_extent(trans, root, bytenr, num_bytes,
1603 parent, btrfs_header_owner(leaf),
1604 key.objectid, key.offset);
1605 BUG_ON(ret);
1606 }
1607 if (dirty)
1608 btrfs_mark_buffer_dirty(leaf);
1609 if (inode)
1610 btrfs_add_delayed_iput(inode);
1611 return 0;
1612 }
1613
1614 static noinline_for_stack
1615 int memcmp_node_keys(struct extent_buffer *eb, int slot,
1616 struct btrfs_path *path, int level)
1617 {
1618 struct btrfs_disk_key key1;
1619 struct btrfs_disk_key key2;
1620 btrfs_node_key(eb, &key1, slot);
1621 btrfs_node_key(path->nodes[level], &key2, path->slots[level]);
1622 return memcmp(&key1, &key2, sizeof(key1));
1623 }
1624
1625 /*
1626 * try to replace tree blocks in fs tree with the new blocks
1627 * in reloc tree. tree blocks haven't been modified since the
1628 * reloc tree was create can be replaced.
1629 *
1630 * if a block was replaced, level of the block + 1 is returned.
1631 * if no block got replaced, 0 is returned. if there are other
1632 * errors, a negative error number is returned.
1633 */
1634 static noinline_for_stack
1635 int replace_path(struct btrfs_trans_handle *trans,
1636 struct btrfs_root *dest, struct btrfs_root *src,
1637 struct btrfs_path *path, struct btrfs_key *next_key,
1638 int lowest_level, int max_level)
1639 {
1640 struct extent_buffer *eb;
1641 struct extent_buffer *parent;
1642 struct btrfs_key key;
1643 u64 old_bytenr;
1644 u64 new_bytenr;
1645 u64 old_ptr_gen;
1646 u64 new_ptr_gen;
1647 u64 last_snapshot;
1648 u32 blocksize;
1649 int cow = 0;
1650 int level;
1651 int ret;
1652 int slot;
1653
1654 BUG_ON(src->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
1655 BUG_ON(dest->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID);
1656
1657 last_snapshot = btrfs_root_last_snapshot(&src->root_item);
1658 again:
1659 slot = path->slots[lowest_level];
1660 btrfs_node_key_to_cpu(path->nodes[lowest_level], &key, slot);
1661
1662 eb = btrfs_lock_root_node(dest);
1663 btrfs_set_lock_blocking(eb);
1664 level = btrfs_header_level(eb);
1665
1666 if (level < lowest_level) {
1667 btrfs_tree_unlock(eb);
1668 free_extent_buffer(eb);
1669 return 0;
1670 }
1671
1672 if (cow) {
1673 ret = btrfs_cow_block(trans, dest, eb, NULL, 0, &eb);
1674 BUG_ON(ret);
1675 }
1676 btrfs_set_lock_blocking(eb);
1677
1678 if (next_key) {
1679 next_key->objectid = (u64)-1;
1680 next_key->type = (u8)-1;
1681 next_key->offset = (u64)-1;
1682 }
1683
1684 parent = eb;
1685 while (1) {
1686 level = btrfs_header_level(parent);
1687 BUG_ON(level < lowest_level);
1688
1689 ret = btrfs_bin_search(parent, &key, level, &slot);
1690 if (ret && slot > 0)
1691 slot--;
1692
1693 if (next_key && slot + 1 < btrfs_header_nritems(parent))
1694 btrfs_node_key_to_cpu(parent, next_key, slot + 1);
1695
1696 old_bytenr = btrfs_node_blockptr(parent, slot);
1697 blocksize = btrfs_level_size(dest, level - 1);
1698 old_ptr_gen = btrfs_node_ptr_generation(parent, slot);
1699
1700 if (level <= max_level) {
1701 eb = path->nodes[level];
1702 new_bytenr = btrfs_node_blockptr(eb,
1703 path->slots[level]);
1704 new_ptr_gen = btrfs_node_ptr_generation(eb,
1705 path->slots[level]);
1706 } else {
1707 new_bytenr = 0;
1708 new_ptr_gen = 0;
1709 }
1710
1711 if (new_bytenr > 0 && new_bytenr == old_bytenr) {
1712 WARN_ON(1);
1713 ret = level;
1714 break;
1715 }
1716
1717 if (new_bytenr == 0 || old_ptr_gen > last_snapshot ||
1718 memcmp_node_keys(parent, slot, path, level)) {
1719 if (level <= lowest_level) {
1720 ret = 0;
1721 break;
1722 }
1723
1724 eb = read_tree_block(dest, old_bytenr, blocksize,
1725 old_ptr_gen);
1726 btrfs_tree_lock(eb);
1727 if (cow) {
1728 ret = btrfs_cow_block(trans, dest, eb, parent,
1729 slot, &eb);
1730 BUG_ON(ret);
1731 }
1732 btrfs_set_lock_blocking(eb);
1733
1734 btrfs_tree_unlock(parent);
1735 free_extent_buffer(parent);
1736
1737 parent = eb;
1738 continue;
1739 }
1740
1741 if (!cow) {
1742 btrfs_tree_unlock(parent);
1743 free_extent_buffer(parent);
1744 cow = 1;
1745 goto again;
1746 }
1747
1748 btrfs_node_key_to_cpu(path->nodes[level], &key,
1749 path->slots[level]);
1750 btrfs_release_path(src, path);
1751
1752 path->lowest_level = level;
1753 ret = btrfs_search_slot(trans, src, &key, path, 0, 1);
1754 path->lowest_level = 0;
1755 BUG_ON(ret);
1756
1757 /*
1758 * swap blocks in fs tree and reloc tree.
1759 */
1760 btrfs_set_node_blockptr(parent, slot, new_bytenr);
1761 btrfs_set_node_ptr_generation(parent, slot, new_ptr_gen);
1762 btrfs_mark_buffer_dirty(parent);
1763
1764 btrfs_set_node_blockptr(path->nodes[level],
1765 path->slots[level], old_bytenr);
1766 btrfs_set_node_ptr_generation(path->nodes[level],
1767 path->slots[level], old_ptr_gen);
1768 btrfs_mark_buffer_dirty(path->nodes[level]);
1769
1770 ret = btrfs_inc_extent_ref(trans, src, old_bytenr, blocksize,
1771 path->nodes[level]->start,
1772 src->root_key.objectid, level - 1, 0);
1773 BUG_ON(ret);
1774 ret = btrfs_inc_extent_ref(trans, dest, new_bytenr, blocksize,
1775 0, dest->root_key.objectid, level - 1,
1776 0);
1777 BUG_ON(ret);
1778
1779 ret = btrfs_free_extent(trans, src, new_bytenr, blocksize,
1780 path->nodes[level]->start,
1781 src->root_key.objectid, level - 1, 0);
1782 BUG_ON(ret);
1783
1784 ret = btrfs_free_extent(trans, dest, old_bytenr, blocksize,
1785 0, dest->root_key.objectid, level - 1,
1786 0);
1787 BUG_ON(ret);
1788
1789 btrfs_unlock_up_safe(path, 0);
1790
1791 ret = level;
1792 break;
1793 }
1794 btrfs_tree_unlock(parent);
1795 free_extent_buffer(parent);
1796 return ret;
1797 }
1798
1799 /*
1800 * helper to find next relocated block in reloc tree
1801 */
1802 static noinline_for_stack
1803 int walk_up_reloc_tree(struct btrfs_root *root, struct btrfs_path *path,
1804 int *level)
1805 {
1806 struct extent_buffer *eb;
1807 int i;
1808 u64 last_snapshot;
1809 u32 nritems;
1810
1811 last_snapshot = btrfs_root_last_snapshot(&root->root_item);
1812
1813 for (i = 0; i < *level; i++) {
1814 free_extent_buffer(path->nodes[i]);
1815 path->nodes[i] = NULL;
1816 }
1817
1818 for (i = *level; i < BTRFS_MAX_LEVEL && path->nodes[i]; i++) {
1819 eb = path->nodes[i];
1820 nritems = btrfs_header_nritems(eb);
1821 while (path->slots[i] + 1 < nritems) {
1822 path->slots[i]++;
1823 if (btrfs_node_ptr_generation(eb, path->slots[i]) <=
1824 last_snapshot)
1825 continue;
1826
1827 *level = i;
1828 return 0;
1829 }
1830 free_extent_buffer(path->nodes[i]);
1831 path->nodes[i] = NULL;
1832 }
1833 return 1;
1834 }
1835
1836 /*
1837 * walk down reloc tree to find relocated block of lowest level
1838 */
1839 static noinline_for_stack
1840 int walk_down_reloc_tree(struct btrfs_root *root, struct btrfs_path *path,
1841 int *level)
1842 {
1843 struct extent_buffer *eb = NULL;
1844 int i;
1845 u64 bytenr;
1846 u64 ptr_gen = 0;
1847 u64 last_snapshot;
1848 u32 blocksize;
1849 u32 nritems;
1850
1851 last_snapshot = btrfs_root_last_snapshot(&root->root_item);
1852
1853 for (i = *level; i > 0; i--) {
1854 eb = path->nodes[i];
1855 nritems = btrfs_header_nritems(eb);
1856 while (path->slots[i] < nritems) {
1857 ptr_gen = btrfs_node_ptr_generation(eb, path->slots[i]);
1858 if (ptr_gen > last_snapshot)
1859 break;
1860 path->slots[i]++;
1861 }
1862 if (path->slots[i] >= nritems) {
1863 if (i == *level)
1864 break;
1865 *level = i + 1;
1866 return 0;
1867 }
1868 if (i == 1) {
1869 *level = i;
1870 return 0;
1871 }
1872
1873 bytenr = btrfs_node_blockptr(eb, path->slots[i]);
1874 blocksize = btrfs_level_size(root, i - 1);
1875 eb = read_tree_block(root, bytenr, blocksize, ptr_gen);
1876 BUG_ON(btrfs_header_level(eb) != i - 1);
1877 path->nodes[i - 1] = eb;
1878 path->slots[i - 1] = 0;
1879 }
1880 return 1;
1881 }
1882
1883 /*
1884 * invalidate extent cache for file extents whose key in range of
1885 * [min_key, max_key)
1886 */
1887 static int invalidate_extent_cache(struct btrfs_root *root,
1888 struct btrfs_key *min_key,
1889 struct btrfs_key *max_key)
1890 {
1891 struct inode *inode = NULL;
1892 u64 objectid;
1893 u64 start, end;
1894
1895 objectid = min_key->objectid;
1896 while (1) {
1897 cond_resched();
1898 iput(inode);
1899
1900 if (objectid > max_key->objectid)
1901 break;
1902
1903 inode = find_next_inode(root, objectid);
1904 if (!inode)
1905 break;
1906
1907 if (inode->i_ino > max_key->objectid) {
1908 iput(inode);
1909 break;
1910 }
1911
1912 objectid = inode->i_ino + 1;
1913 if (!S_ISREG(inode->i_mode))
1914 continue;
1915
1916 if (unlikely(min_key->objectid == inode->i_ino)) {
1917 if (min_key->type > BTRFS_EXTENT_DATA_KEY)
1918 continue;
1919 if (min_key->type < BTRFS_EXTENT_DATA_KEY)
1920 start = 0;
1921 else {
1922 start = min_key->offset;
1923 WARN_ON(!IS_ALIGNED(start, root->sectorsize));
1924 }
1925 } else {
1926 start = 0;
1927 }
1928
1929 if (unlikely(max_key->objectid == inode->i_ino)) {
1930 if (max_key->type < BTRFS_EXTENT_DATA_KEY)
1931 continue;
1932 if (max_key->type > BTRFS_EXTENT_DATA_KEY) {
1933 end = (u64)-1;
1934 } else {
1935 if (max_key->offset == 0)
1936 continue;
1937 end = max_key->offset;
1938 WARN_ON(!IS_ALIGNED(end, root->sectorsize));
1939 end--;
1940 }
1941 } else {
1942 end = (u64)-1;
1943 }
1944
1945 /* the lock_extent waits for readpage to complete */
1946 lock_extent(&BTRFS_I(inode)->io_tree, start, end, GFP_NOFS);
1947 btrfs_drop_extent_cache(inode, start, end, 1);
1948 unlock_extent(&BTRFS_I(inode)->io_tree, start, end, GFP_NOFS);
1949 }
1950 return 0;
1951 }
1952
1953 static int find_next_key(struct btrfs_path *path, int level,
1954 struct btrfs_key *key)
1955
1956 {
1957 while (level < BTRFS_MAX_LEVEL) {
1958 if (!path->nodes[level])
1959 break;
1960 if (path->slots[level] + 1 <
1961 btrfs_header_nritems(path->nodes[level])) {
1962 btrfs_node_key_to_cpu(path->nodes[level], key,
1963 path->slots[level] + 1);
1964 return 0;
1965 }
1966 level++;
1967 }
1968 return 1;
1969 }
1970
1971 /*
1972 * merge the relocated tree blocks in reloc tree with corresponding
1973 * fs tree.
1974 */
1975 static noinline_for_stack int merge_reloc_root(struct reloc_control *rc,
1976 struct btrfs_root *root)
1977 {
1978 LIST_HEAD(inode_list);
1979 struct btrfs_key key;
1980 struct btrfs_key next_key;
1981 struct btrfs_trans_handle *trans;
1982 struct btrfs_root *reloc_root;
1983 struct btrfs_root_item *root_item;
1984 struct btrfs_path *path;
1985 struct extent_buffer *leaf;
1986 unsigned long nr;
1987 int level;
1988 int max_level;
1989 int replaced = 0;
1990 int ret;
1991 int err = 0;
1992 u32 min_reserved;
1993
1994 path = btrfs_alloc_path();
1995 if (!path)
1996 return -ENOMEM;
1997
1998 reloc_root = root->reloc_root;
1999 root_item = &reloc_root->root_item;
2000
2001 if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
2002 level = btrfs_root_level(root_item);
2003 extent_buffer_get(reloc_root->node);
2004 path->nodes[level] = reloc_root->node;
2005 path->slots[level] = 0;
2006 } else {
2007 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
2008
2009 level = root_item->drop_level;
2010 BUG_ON(level == 0);
2011 path->lowest_level = level;
2012 ret = btrfs_search_slot(NULL, reloc_root, &key, path, 0, 0);
2013 path->lowest_level = 0;
2014 if (ret < 0) {
2015 btrfs_free_path(path);
2016 return ret;
2017 }
2018
2019 btrfs_node_key_to_cpu(path->nodes[level], &next_key,
2020 path->slots[level]);
2021 WARN_ON(memcmp(&key, &next_key, sizeof(key)));
2022
2023 btrfs_unlock_up_safe(path, 0);
2024 }
2025
2026 min_reserved = root->nodesize * (BTRFS_MAX_LEVEL - 1) * 2;
2027 memset(&next_key, 0, sizeof(next_key));
2028
2029 while (1) {
2030 trans = btrfs_start_transaction(root, 0);
2031 BUG_ON(IS_ERR(trans));
2032 trans->block_rsv = rc->block_rsv;
2033
2034 ret = btrfs_block_rsv_check(trans, root, rc->block_rsv,
2035 min_reserved, 0);
2036 if (ret) {
2037 BUG_ON(ret != -EAGAIN);
2038 ret = btrfs_commit_transaction(trans, root);
2039 BUG_ON(ret);
2040 continue;
2041 }
2042
2043 replaced = 0;
2044 max_level = level;
2045
2046 ret = walk_down_reloc_tree(reloc_root, path, &level);
2047 if (ret < 0) {
2048 err = ret;
2049 goto out;
2050 }
2051 if (ret > 0)
2052 break;
2053
2054 if (!find_next_key(path, level, &key) &&
2055 btrfs_comp_cpu_keys(&next_key, &key) >= 0) {
2056 ret = 0;
2057 } else {
2058 ret = replace_path(trans, root, reloc_root, path,
2059 &next_key, level, max_level);
2060 }
2061 if (ret < 0) {
2062 err = ret;
2063 goto out;
2064 }
2065
2066 if (ret > 0) {
2067 level = ret;
2068 btrfs_node_key_to_cpu(path->nodes[level], &key,
2069 path->slots[level]);
2070 replaced = 1;
2071 }
2072
2073 ret = walk_up_reloc_tree(reloc_root, path, &level);
2074 if (ret > 0)
2075 break;
2076
2077 BUG_ON(level == 0);
2078 /*
2079 * save the merging progress in the drop_progress.
2080 * this is OK since root refs == 1 in this case.
2081 */
2082 btrfs_node_key(path->nodes[level], &root_item->drop_progress,
2083 path->slots[level]);
2084 root_item->drop_level = level;
2085
2086 nr = trans->blocks_used;
2087 btrfs_end_transaction_throttle(trans, root);
2088
2089 btrfs_btree_balance_dirty(root, nr);
2090
2091 if (replaced && rc->stage == UPDATE_DATA_PTRS)
2092 invalidate_extent_cache(root, &key, &next_key);
2093 }
2094
2095 /*
2096 * handle the case only one block in the fs tree need to be
2097 * relocated and the block is tree root.
2098 */
2099 leaf = btrfs_lock_root_node(root);
2100 ret = btrfs_cow_block(trans, root, leaf, NULL, 0, &leaf);
2101 btrfs_tree_unlock(leaf);
2102 free_extent_buffer(leaf);
2103 if (ret < 0)
2104 err = ret;
2105 out:
2106 btrfs_free_path(path);
2107
2108 if (err == 0) {
2109 memset(&root_item->drop_progress, 0,
2110 sizeof(root_item->drop_progress));
2111 root_item->drop_level = 0;
2112 btrfs_set_root_refs(root_item, 0);
2113 btrfs_update_reloc_root(trans, root);
2114 }
2115
2116 nr = trans->blocks_used;
2117 btrfs_end_transaction_throttle(trans, root);
2118
2119 btrfs_btree_balance_dirty(root, nr);
2120
2121 if (replaced && rc->stage == UPDATE_DATA_PTRS)
2122 invalidate_extent_cache(root, &key, &next_key);
2123
2124 return err;
2125 }
2126
2127 static noinline_for_stack
2128 int prepare_to_merge(struct reloc_control *rc, int err)
2129 {
2130 struct btrfs_root *root = rc->extent_root;
2131 struct btrfs_root *reloc_root;
2132 struct btrfs_trans_handle *trans;
2133 LIST_HEAD(reloc_roots);
2134 u64 num_bytes = 0;
2135 int ret;
2136
2137 mutex_lock(&root->fs_info->trans_mutex);
2138 rc->merging_rsv_size += root->nodesize * (BTRFS_MAX_LEVEL - 1) * 2;
2139 rc->merging_rsv_size += rc->nodes_relocated * 2;
2140 mutex_unlock(&root->fs_info->trans_mutex);
2141 again:
2142 if (!err) {
2143 num_bytes = rc->merging_rsv_size;
2144 ret = btrfs_block_rsv_add(NULL, root, rc->block_rsv,
2145 num_bytes);
2146 if (ret)
2147 err = ret;
2148 }
2149
2150 trans = btrfs_join_transaction(rc->extent_root, 1);
2151 if (IS_ERR(trans)) {
2152 if (!err)
2153 btrfs_block_rsv_release(rc->extent_root,
2154 rc->block_rsv, num_bytes);
2155 return PTR_ERR(trans);
2156 }
2157
2158 if (!err) {
2159 if (num_bytes != rc->merging_rsv_size) {
2160 btrfs_end_transaction(trans, rc->extent_root);
2161 btrfs_block_rsv_release(rc->extent_root,
2162 rc->block_rsv, num_bytes);
2163 goto again;
2164 }
2165 }
2166
2167 rc->merge_reloc_tree = 1;
2168
2169 while (!list_empty(&rc->reloc_roots)) {
2170 reloc_root = list_entry(rc->reloc_roots.next,
2171 struct btrfs_root, root_list);
2172 list_del_init(&reloc_root->root_list);
2173
2174 root = read_fs_root(reloc_root->fs_info,
2175 reloc_root->root_key.offset);
2176 BUG_ON(IS_ERR(root));
2177 BUG_ON(root->reloc_root != reloc_root);
2178
2179 /*
2180 * set reference count to 1, so btrfs_recover_relocation
2181 * knows it should resumes merging
2182 */
2183 if (!err)
2184 btrfs_set_root_refs(&reloc_root->root_item, 1);
2185 btrfs_update_reloc_root(trans, root);
2186
2187 list_add(&reloc_root->root_list, &reloc_roots);
2188 }
2189
2190 list_splice(&reloc_roots, &rc->reloc_roots);
2191
2192 if (!err)
2193 btrfs_commit_transaction(trans, rc->extent_root);
2194 else
2195 btrfs_end_transaction(trans, rc->extent_root);
2196 return err;
2197 }
2198
2199 static noinline_for_stack
2200 int merge_reloc_roots(struct reloc_control *rc)
2201 {
2202 struct btrfs_root *root;
2203 struct btrfs_root *reloc_root;
2204 LIST_HEAD(reloc_roots);
2205 int found = 0;
2206 int ret;
2207 again:
2208 root = rc->extent_root;
2209 mutex_lock(&root->fs_info->trans_mutex);
2210 list_splice_init(&rc->reloc_roots, &reloc_roots);
2211 mutex_unlock(&root->fs_info->trans_mutex);
2212
2213 while (!list_empty(&reloc_roots)) {
2214 found = 1;
2215 reloc_root = list_entry(reloc_roots.next,
2216 struct btrfs_root, root_list);
2217
2218 if (btrfs_root_refs(&reloc_root->root_item) > 0) {
2219 root = read_fs_root(reloc_root->fs_info,
2220 reloc_root->root_key.offset);
2221 BUG_ON(IS_ERR(root));
2222 BUG_ON(root->reloc_root != reloc_root);
2223
2224 ret = merge_reloc_root(rc, root);
2225 BUG_ON(ret);
2226 } else {
2227 list_del_init(&reloc_root->root_list);
2228 }
2229 btrfs_drop_snapshot(reloc_root, rc->block_rsv, 0);
2230 }
2231
2232 if (found) {
2233 found = 0;
2234 goto again;
2235 }
2236 BUG_ON(!RB_EMPTY_ROOT(&rc->reloc_root_tree.rb_root));
2237 return 0;
2238 }
2239
2240 static void free_block_list(struct rb_root *blocks)
2241 {
2242 struct tree_block *block;
2243 struct rb_node *rb_node;
2244 while ((rb_node = rb_first(blocks))) {
2245 block = rb_entry(rb_node, struct tree_block, rb_node);
2246 rb_erase(rb_node, blocks);
2247 kfree(block);
2248 }
2249 }
2250
2251 static int record_reloc_root_in_trans(struct btrfs_trans_handle *trans,
2252 struct btrfs_root *reloc_root)
2253 {
2254 struct btrfs_root *root;
2255
2256 if (reloc_root->last_trans == trans->transid)
2257 return 0;
2258
2259 root = read_fs_root(reloc_root->fs_info, reloc_root->root_key.offset);
2260 BUG_ON(IS_ERR(root));
2261 BUG_ON(root->reloc_root != reloc_root);
2262
2263 return btrfs_record_root_in_trans(trans, root);
2264 }
2265
2266 static noinline_for_stack
2267 struct btrfs_root *select_reloc_root(struct btrfs_trans_handle *trans,
2268 struct reloc_control *rc,
2269 struct backref_node *node,
2270 struct backref_edge *edges[], int *nr)
2271 {
2272 struct backref_node *next;
2273 struct btrfs_root *root;
2274 int index = 0;
2275
2276 next = node;
2277 while (1) {
2278 cond_resched();
2279 next = walk_up_backref(next, edges, &index);
2280 root = next->root;
2281 BUG_ON(!root);
2282 BUG_ON(!root->ref_cows);
2283
2284 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) {
2285 record_reloc_root_in_trans(trans, root);
2286 break;
2287 }
2288
2289 btrfs_record_root_in_trans(trans, root);
2290 root = root->reloc_root;
2291
2292 if (next->new_bytenr != root->node->start) {
2293 BUG_ON(next->new_bytenr);
2294 BUG_ON(!list_empty(&next->list));
2295 next->new_bytenr = root->node->start;
2296 next->root = root;
2297 list_add_tail(&next->list,
2298 &rc->backref_cache.changed);
2299 __mark_block_processed(rc, next);
2300 break;
2301 }
2302
2303 WARN_ON(1);
2304 root = NULL;
2305 next = walk_down_backref(edges, &index);
2306 if (!next || next->level <= node->level)
2307 break;
2308 }
2309 if (!root)
2310 return NULL;
2311
2312 *nr = index;
2313 next = node;
2314 /* setup backref node path for btrfs_reloc_cow_block */
2315 while (1) {
2316 rc->backref_cache.path[next->level] = next;
2317 if (--index < 0)
2318 break;
2319 next = edges[index]->node[UPPER];
2320 }
2321 return root;
2322 }
2323
2324 /*
2325 * select a tree root for relocation. return NULL if the block
2326 * is reference counted. we should use do_relocation() in this
2327 * case. return a tree root pointer if the block isn't reference
2328 * counted. return -ENOENT if the block is root of reloc tree.
2329 */
2330 static noinline_for_stack
2331 struct btrfs_root *select_one_root(struct btrfs_trans_handle *trans,
2332 struct backref_node *node)
2333 {
2334 struct backref_node *next;
2335 struct btrfs_root *root;
2336 struct btrfs_root *fs_root = NULL;
2337 struct backref_edge *edges[BTRFS_MAX_LEVEL - 1];
2338 int index = 0;
2339
2340 next = node;
2341 while (1) {
2342 cond_resched();
2343 next = walk_up_backref(next, edges, &index);
2344 root = next->root;
2345 BUG_ON(!root);
2346
2347 /* no other choice for non-refernce counted tree */
2348 if (!root->ref_cows)
2349 return root;
2350
2351 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID)
2352 fs_root = root;
2353
2354 if (next != node)
2355 return NULL;
2356
2357 next = walk_down_backref(edges, &index);
2358 if (!next || next->level <= node->level)
2359 break;
2360 }
2361
2362 if (!fs_root)
2363 return ERR_PTR(-ENOENT);
2364 return fs_root;
2365 }
2366
2367 static noinline_for_stack
2368 u64 calcu_metadata_size(struct reloc_control *rc,
2369 struct backref_node *node, int reserve)
2370 {
2371 struct backref_node *next = node;
2372 struct backref_edge *edge;
2373 struct backref_edge *edges[BTRFS_MAX_LEVEL - 1];
2374 u64 num_bytes = 0;
2375 int index = 0;
2376
2377 BUG_ON(reserve && node->processed);
2378
2379 while (next) {
2380 cond_resched();
2381 while (1) {
2382 if (next->processed && (reserve || next != node))
2383 break;
2384
2385 num_bytes += btrfs_level_size(rc->extent_root,
2386 next->level);
2387
2388 if (list_empty(&next->upper))
2389 break;
2390
2391 edge = list_entry(next->upper.next,
2392 struct backref_edge, list[LOWER]);
2393 edges[index++] = edge;
2394 next = edge->node[UPPER];
2395 }
2396 next = walk_down_backref(edges, &index);
2397 }
2398 return num_bytes;
2399 }
2400
2401 static int reserve_metadata_space(struct btrfs_trans_handle *trans,
2402 struct reloc_control *rc,
2403 struct backref_node *node)
2404 {
2405 struct btrfs_root *root = rc->extent_root;
2406 u64 num_bytes;
2407 int ret;
2408
2409 num_bytes = calcu_metadata_size(rc, node, 1) * 2;
2410
2411 trans->block_rsv = rc->block_rsv;
2412 ret = btrfs_block_rsv_add(trans, root, rc->block_rsv, num_bytes);
2413 if (ret) {
2414 if (ret == -EAGAIN)
2415 rc->commit_transaction = 1;
2416 return ret;
2417 }
2418
2419 return 0;
2420 }
2421
2422 static void release_metadata_space(struct reloc_control *rc,
2423 struct backref_node *node)
2424 {
2425 u64 num_bytes = calcu_metadata_size(rc, node, 0) * 2;
2426 btrfs_block_rsv_release(rc->extent_root, rc->block_rsv, num_bytes);
2427 }
2428
2429 /*
2430 * relocate a block tree, and then update pointers in upper level
2431 * blocks that reference the block to point to the new location.
2432 *
2433 * if called by link_to_upper, the block has already been relocated.
2434 * in that case this function just updates pointers.
2435 */
2436 static int do_relocation(struct btrfs_trans_handle *trans,
2437 struct reloc_control *rc,
2438 struct backref_node *node,
2439 struct btrfs_key *key,
2440 struct btrfs_path *path, int lowest)
2441 {
2442 struct backref_node *upper;
2443 struct backref_edge *edge;
2444 struct backref_edge *edges[BTRFS_MAX_LEVEL - 1];
2445 struct btrfs_root *root;
2446 struct extent_buffer *eb;
2447 u32 blocksize;
2448 u64 bytenr;
2449 u64 generation;
2450 int nr;
2451 int slot;
2452 int ret;
2453 int err = 0;
2454
2455 BUG_ON(lowest && node->eb);
2456
2457 path->lowest_level = node->level + 1;
2458 rc->backref_cache.path[node->level] = node;
2459 list_for_each_entry(edge, &node->upper, list[LOWER]) {
2460 cond_resched();
2461
2462 upper = edge->node[UPPER];
2463 root = select_reloc_root(trans, rc, upper, edges, &nr);
2464 BUG_ON(!root);
2465
2466 if (upper->eb && !upper->locked) {
2467 if (!lowest) {
2468 ret = btrfs_bin_search(upper->eb, key,
2469 upper->level, &slot);
2470 BUG_ON(ret);
2471 bytenr = btrfs_node_blockptr(upper->eb, slot);
2472 if (node->eb->start == bytenr)
2473 goto next;
2474 }
2475 drop_node_buffer(upper);
2476 }
2477
2478 if (!upper->eb) {
2479 ret = btrfs_search_slot(trans, root, key, path, 0, 1);
2480 if (ret < 0) {
2481 err = ret;
2482 break;
2483 }
2484 BUG_ON(ret > 0);
2485
2486 if (!upper->eb) {
2487 upper->eb = path->nodes[upper->level];
2488 path->nodes[upper->level] = NULL;
2489 } else {
2490 BUG_ON(upper->eb != path->nodes[upper->level]);
2491 }
2492
2493 upper->locked = 1;
2494 path->locks[upper->level] = 0;
2495
2496 slot = path->slots[upper->level];
2497 btrfs_release_path(NULL, path);
2498 } else {
2499 ret = btrfs_bin_search(upper->eb, key, upper->level,
2500 &slot);
2501 BUG_ON(ret);
2502 }
2503
2504 bytenr = btrfs_node_blockptr(upper->eb, slot);
2505 if (lowest) {
2506 BUG_ON(bytenr != node->bytenr);
2507 } else {
2508 if (node->eb->start == bytenr)
2509 goto next;
2510 }
2511
2512 blocksize = btrfs_level_size(root, node->level);
2513 generation = btrfs_node_ptr_generation(upper->eb, slot);
2514 eb = read_tree_block(root, bytenr, blocksize, generation);
2515 btrfs_tree_lock(eb);
2516 btrfs_set_lock_blocking(eb);
2517
2518 if (!node->eb) {
2519 ret = btrfs_cow_block(trans, root, eb, upper->eb,
2520 slot, &eb);
2521 btrfs_tree_unlock(eb);
2522 free_extent_buffer(eb);
2523 if (ret < 0) {
2524 err = ret;
2525 goto next;
2526 }
2527 BUG_ON(node->eb != eb);
2528 } else {
2529 btrfs_set_node_blockptr(upper->eb, slot,
2530 node->eb->start);
2531 btrfs_set_node_ptr_generation(upper->eb, slot,
2532 trans->transid);
2533 btrfs_mark_buffer_dirty(upper->eb);
2534
2535 ret = btrfs_inc_extent_ref(trans, root,
2536 node->eb->start, blocksize,
2537 upper->eb->start,
2538 btrfs_header_owner(upper->eb),
2539 node->level, 0);
2540 BUG_ON(ret);
2541
2542 ret = btrfs_drop_subtree(trans, root, eb, upper->eb);
2543 BUG_ON(ret);
2544 }
2545 next:
2546 if (!upper->pending)
2547 drop_node_buffer(upper);
2548 else
2549 unlock_node_buffer(upper);
2550 if (err)
2551 break;
2552 }
2553
2554 if (!err && node->pending) {
2555 drop_node_buffer(node);
2556 list_move_tail(&node->list, &rc->backref_cache.changed);
2557 node->pending = 0;
2558 }
2559
2560 path->lowest_level = 0;
2561 BUG_ON(err == -ENOSPC);
2562 return err;
2563 }
2564
2565 static int link_to_upper(struct btrfs_trans_handle *trans,
2566 struct reloc_control *rc,
2567 struct backref_node *node,
2568 struct btrfs_path *path)
2569 {
2570 struct btrfs_key key;
2571
2572 btrfs_node_key_to_cpu(node->eb, &key, 0);
2573 return do_relocation(trans, rc, node, &key, path, 0);
2574 }
2575
2576 static int finish_pending_nodes(struct btrfs_trans_handle *trans,
2577 struct reloc_control *rc,
2578 struct btrfs_path *path, int err)
2579 {
2580 LIST_HEAD(list);
2581 struct backref_cache *cache = &rc->backref_cache;
2582 struct backref_node *node;
2583 int level;
2584 int ret;
2585
2586 for (level = 0; level < BTRFS_MAX_LEVEL; level++) {
2587 while (!list_empty(&cache->pending[level])) {
2588 node = list_entry(cache->pending[level].next,
2589 struct backref_node, list);
2590 list_move_tail(&node->list, &list);
2591 BUG_ON(!node->pending);
2592
2593 if (!err) {
2594 ret = link_to_upper(trans, rc, node, path);
2595 if (ret < 0)
2596 err = ret;
2597 }
2598 }
2599 list_splice_init(&list, &cache->pending[level]);
2600 }
2601 return err;
2602 }
2603
2604 static void mark_block_processed(struct reloc_control *rc,
2605 u64 bytenr, u32 blocksize)
2606 {
2607 set_extent_bits(&rc->processed_blocks, bytenr, bytenr + blocksize - 1,
2608 EXTENT_DIRTY, GFP_NOFS);
2609 }
2610
2611 static void __mark_block_processed(struct reloc_control *rc,
2612 struct backref_node *node)
2613 {
2614 u32 blocksize;
2615 if (node->level == 0 ||
2616 in_block_group(node->bytenr, rc->block_group)) {
2617 blocksize = btrfs_level_size(rc->extent_root, node->level);
2618 mark_block_processed(rc, node->bytenr, blocksize);
2619 }
2620 node->processed = 1;
2621 }
2622
2623 /*
2624 * mark a block and all blocks directly/indirectly reference the block
2625 * as processed.
2626 */
2627 static void update_processed_blocks(struct reloc_control *rc,
2628 struct backref_node *node)
2629 {
2630 struct backref_node *next = node;
2631 struct backref_edge *edge;
2632 struct backref_edge *edges[BTRFS_MAX_LEVEL - 1];
2633 int index = 0;
2634
2635 while (next) {
2636 cond_resched();
2637 while (1) {
2638 if (next->processed)
2639 break;
2640
2641 __mark_block_processed(rc, next);
2642
2643 if (list_empty(&next->upper))
2644 break;
2645
2646 edge = list_entry(next->upper.next,
2647 struct backref_edge, list[LOWER]);
2648 edges[index++] = edge;
2649 next = edge->node[UPPER];
2650 }
2651 next = walk_down_backref(edges, &index);
2652 }
2653 }
2654
2655 static int tree_block_processed(u64 bytenr, u32 blocksize,
2656 struct reloc_control *rc)
2657 {
2658 if (test_range_bit(&rc->processed_blocks, bytenr,
2659 bytenr + blocksize - 1, EXTENT_DIRTY, 1, NULL))
2660 return 1;
2661 return 0;
2662 }
2663
2664 static int get_tree_block_key(struct reloc_control *rc,
2665 struct tree_block *block)
2666 {
2667 struct extent_buffer *eb;
2668
2669 BUG_ON(block->key_ready);
2670 eb = read_tree_block(rc->extent_root, block->bytenr,
2671 block->key.objectid, block->key.offset);
2672 WARN_ON(btrfs_header_level(eb) != block->level);
2673 if (block->level == 0)
2674 btrfs_item_key_to_cpu(eb, &block->key, 0);
2675 else
2676 btrfs_node_key_to_cpu(eb, &block->key, 0);
2677 free_extent_buffer(eb);
2678 block->key_ready = 1;
2679 return 0;
2680 }
2681
2682 static int reada_tree_block(struct reloc_control *rc,
2683 struct tree_block *block)
2684 {
2685 BUG_ON(block->key_ready);
2686 readahead_tree_block(rc->extent_root, block->bytenr,
2687 block->key.objectid, block->key.offset);
2688 return 0;
2689 }
2690
2691 /*
2692 * helper function to relocate a tree block
2693 */
2694 static int relocate_tree_block(struct btrfs_trans_handle *trans,
2695 struct reloc_control *rc,
2696 struct backref_node *node,
2697 struct btrfs_key *key,
2698 struct btrfs_path *path)
2699 {
2700 struct btrfs_root *root;
2701 int release = 0;
2702 int ret = 0;
2703
2704 if (!node)
2705 return 0;
2706
2707 BUG_ON(node->processed);
2708 root = select_one_root(trans, node);
2709 if (root == ERR_PTR(-ENOENT)) {
2710 update_processed_blocks(rc, node);
2711 goto out;
2712 }
2713
2714 if (!root || root->ref_cows) {
2715 ret = reserve_metadata_space(trans, rc, node);
2716 if (ret)
2717 goto out;
2718 release = 1;
2719 }
2720
2721 if (root) {
2722 if (root->ref_cows) {
2723 BUG_ON(node->new_bytenr);
2724 BUG_ON(!list_empty(&node->list));
2725 btrfs_record_root_in_trans(trans, root);
2726 root = root->reloc_root;
2727 node->new_bytenr = root->node->start;
2728 node->root = root;
2729 list_add_tail(&node->list, &rc->backref_cache.changed);
2730 } else {
2731 path->lowest_level = node->level;
2732 ret = btrfs_search_slot(trans, root, key, path, 0, 1);
2733 btrfs_release_path(root, path);
2734 if (ret > 0)
2735 ret = 0;
2736 }
2737 if (!ret)
2738 update_processed_blocks(rc, node);
2739 } else {
2740 ret = do_relocation(trans, rc, node, key, path, 1);
2741 }
2742 out:
2743 if (ret || node->level == 0 || node->cowonly) {
2744 if (release)
2745 release_metadata_space(rc, node);
2746 remove_backref_node(&rc->backref_cache, node);
2747 }
2748 return ret;
2749 }
2750
2751 /*
2752 * relocate a list of blocks
2753 */
2754 static noinline_for_stack
2755 int relocate_tree_blocks(struct btrfs_trans_handle *trans,
2756 struct reloc_control *rc, struct rb_root *blocks)
2757 {
2758 struct backref_node *node;
2759 struct btrfs_path *path;
2760 struct tree_block *block;
2761 struct rb_node *rb_node;
2762 int ret;
2763 int err = 0;
2764
2765 path = btrfs_alloc_path();
2766 if (!path)
2767 return -ENOMEM;
2768
2769 rb_node = rb_first(blocks);
2770 while (rb_node) {
2771 block = rb_entry(rb_node, struct tree_block, rb_node);
2772 if (!block->key_ready)
2773 reada_tree_block(rc, block);
2774 rb_node = rb_next(rb_node);
2775 }
2776
2777 rb_node = rb_first(blocks);
2778 while (rb_node) {
2779 block = rb_entry(rb_node, struct tree_block, rb_node);
2780 if (!block->key_ready)
2781 get_tree_block_key(rc, block);
2782 rb_node = rb_next(rb_node);
2783 }
2784
2785 rb_node = rb_first(blocks);
2786 while (rb_node) {
2787 block = rb_entry(rb_node, struct tree_block, rb_node);
2788
2789 node = build_backref_tree(rc, &block->key,
2790 block->level, block->bytenr);
2791 if (IS_ERR(node)) {
2792 err = PTR_ERR(node);
2793 goto out;
2794 }
2795
2796 ret = relocate_tree_block(trans, rc, node, &block->key,
2797 path);
2798 if (ret < 0) {
2799 if (ret != -EAGAIN || rb_node == rb_first(blocks))
2800 err = ret;
2801 goto out;
2802 }
2803 rb_node = rb_next(rb_node);
2804 }
2805 out:
2806 free_block_list(blocks);
2807 err = finish_pending_nodes(trans, rc, path, err);
2808
2809 btrfs_free_path(path);
2810 return err;
2811 }
2812
2813 static noinline_for_stack
2814 int prealloc_file_extent_cluster(struct inode *inode,
2815 struct file_extent_cluster *cluster)
2816 {
2817 u64 alloc_hint = 0;
2818 u64 start;
2819 u64 end;
2820 u64 offset = BTRFS_I(inode)->index_cnt;
2821 u64 num_bytes;
2822 int nr = 0;
2823 int ret = 0;
2824
2825 BUG_ON(cluster->start != cluster->boundary[0]);
2826 mutex_lock(&inode->i_mutex);
2827
2828 ret = btrfs_check_data_free_space(inode, cluster->end +
2829 1 - cluster->start);
2830 if (ret)
2831 goto out;
2832
2833 while (nr < cluster->nr) {
2834 start = cluster->boundary[nr] - offset;
2835 if (nr + 1 < cluster->nr)
2836 end = cluster->boundary[nr + 1] - 1 - offset;
2837 else
2838 end = cluster->end - offset;
2839
2840 lock_extent(&BTRFS_I(inode)->io_tree, start, end, GFP_NOFS);
2841 num_bytes = end + 1 - start;
2842 ret = btrfs_prealloc_file_range(inode, 0, start,
2843 num_bytes, num_bytes,
2844 end + 1, &alloc_hint);
2845 unlock_extent(&BTRFS_I(inode)->io_tree, start, end, GFP_NOFS);
2846 if (ret)
2847 break;
2848 nr++;
2849 }
2850 btrfs_free_reserved_data_space(inode, cluster->end +
2851 1 - cluster->start);
2852 out:
2853 mutex_unlock(&inode->i_mutex);
2854 return ret;
2855 }
2856
2857 static noinline_for_stack
2858 int setup_extent_mapping(struct inode *inode, u64 start, u64 end,
2859 u64 block_start)
2860 {
2861 struct btrfs_root *root = BTRFS_I(inode)->root;
2862 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
2863 struct extent_map *em;
2864 int ret = 0;
2865
2866 em = alloc_extent_map(GFP_NOFS);
2867 if (!em)
2868 return -ENOMEM;
2869
2870 em->start = start;
2871 em->len = end + 1 - start;
2872 em->block_len = em->len;
2873 em->block_start = block_start;
2874 em->bdev = root->fs_info->fs_devices->latest_bdev;
2875 set_bit(EXTENT_FLAG_PINNED, &em->flags);
2876
2877 lock_extent(&BTRFS_I(inode)->io_tree, start, end, GFP_NOFS);
2878 while (1) {
2879 write_lock(&em_tree->lock);
2880 ret = add_extent_mapping(em_tree, em);
2881 write_unlock(&em_tree->lock);
2882 if (ret != -EEXIST) {
2883 free_extent_map(em);
2884 break;
2885 }
2886 btrfs_drop_extent_cache(inode, start, end, 0);
2887 }
2888 unlock_extent(&BTRFS_I(inode)->io_tree, start, end, GFP_NOFS);
2889 return ret;
2890 }
2891
2892 static int relocate_file_extent_cluster(struct inode *inode,
2893 struct file_extent_cluster *cluster)
2894 {
2895 u64 page_start;
2896 u64 page_end;
2897 u64 offset = BTRFS_I(inode)->index_cnt;
2898 unsigned long index;
2899 unsigned long last_index;
2900 struct page *page;
2901 struct file_ra_state *ra;
2902 int nr = 0;
2903 int ret = 0;
2904
2905 if (!cluster->nr)
2906 return 0;
2907
2908 ra = kzalloc(sizeof(*ra), GFP_NOFS);
2909 if (!ra)
2910 return -ENOMEM;
2911
2912 ret = prealloc_file_extent_cluster(inode, cluster);
2913 if (ret)
2914 goto out;
2915
2916 file_ra_state_init(ra, inode->i_mapping);
2917
2918 ret = setup_extent_mapping(inode, cluster->start - offset,
2919 cluster->end - offset, cluster->start);
2920 if (ret)
2921 goto out;
2922
2923 index = (cluster->start - offset) >> PAGE_CACHE_SHIFT;
2924 last_index = (cluster->end - offset) >> PAGE_CACHE_SHIFT;
2925 while (index <= last_index) {
2926 ret = btrfs_delalloc_reserve_metadata(inode, PAGE_CACHE_SIZE);
2927 if (ret)
2928 goto out;
2929
2930 page = find_lock_page(inode->i_mapping, index);
2931 if (!page) {
2932 page_cache_sync_readahead(inode->i_mapping,
2933 ra, NULL, index,
2934 last_index + 1 - index);
2935 page = grab_cache_page(inode->i_mapping, index);
2936 if (!page) {
2937 btrfs_delalloc_release_metadata(inode,
2938 PAGE_CACHE_SIZE);
2939 ret = -ENOMEM;
2940 goto out;
2941 }
2942 }
2943
2944 if (PageReadahead(page)) {
2945 page_cache_async_readahead(inode->i_mapping,
2946 ra, NULL, page, index,
2947 last_index + 1 - index);
2948 }
2949
2950 if (!PageUptodate(page)) {
2951 btrfs_readpage(NULL, page);
2952 lock_page(page);
2953 if (!PageUptodate(page)) {
2954 unlock_page(page);
2955 page_cache_release(page);
2956 btrfs_delalloc_release_metadata(inode,
2957 PAGE_CACHE_SIZE);
2958 ret = -EIO;
2959 goto out;
2960 }
2961 }
2962
2963 page_start = (u64)page->index << PAGE_CACHE_SHIFT;
2964 page_end = page_start + PAGE_CACHE_SIZE - 1;
2965
2966 lock_extent(&BTRFS_I(inode)->io_tree,
2967 page_start, page_end, GFP_NOFS);
2968
2969 set_page_extent_mapped(page);
2970
2971 if (nr < cluster->nr &&
2972 page_start + offset == cluster->boundary[nr]) {
2973 set_extent_bits(&BTRFS_I(inode)->io_tree,
2974 page_start, page_end,
2975 EXTENT_BOUNDARY, GFP_NOFS);
2976 nr++;
2977 }
2978
2979 btrfs_set_extent_delalloc(inode, page_start, page_end, NULL);
2980 set_page_dirty(page);
2981
2982 unlock_extent(&BTRFS_I(inode)->io_tree,
2983 page_start, page_end, GFP_NOFS);
2984 unlock_page(page);
2985 page_cache_release(page);
2986
2987 index++;
2988 balance_dirty_pages_ratelimited(inode->i_mapping);
2989 btrfs_throttle(BTRFS_I(inode)->root);
2990 }
2991 WARN_ON(nr != cluster->nr);
2992 out:
2993 kfree(ra);
2994 return ret;
2995 }
2996
2997 static noinline_for_stack
2998 int relocate_data_extent(struct inode *inode, struct btrfs_key *extent_key,
2999 struct file_extent_cluster *cluster)
3000 {
3001 int ret;
3002
3003 if (cluster->nr > 0 && extent_key->objectid != cluster->end + 1) {
3004 ret = relocate_file_extent_cluster(inode, cluster);
3005 if (ret)
3006 return ret;
3007 cluster->nr = 0;
3008 }
3009
3010 if (!cluster->nr)
3011 cluster->start = extent_key->objectid;
3012 else
3013 BUG_ON(cluster->nr >= MAX_EXTENTS);
3014 cluster->end = extent_key->objectid + extent_key->offset - 1;
3015 cluster->boundary[cluster->nr] = extent_key->objectid;
3016 cluster->nr++;
3017
3018 if (cluster->nr >= MAX_EXTENTS) {
3019 ret = relocate_file_extent_cluster(inode, cluster);
3020 if (ret)
3021 return ret;
3022 cluster->nr = 0;
3023 }
3024 return 0;
3025 }
3026
3027 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
3028 static int get_ref_objectid_v0(struct reloc_control *rc,
3029 struct btrfs_path *path,
3030 struct btrfs_key *extent_key,
3031 u64 *ref_objectid, int *path_change)
3032 {
3033 struct btrfs_key key;
3034 struct extent_buffer *leaf;
3035 struct btrfs_extent_ref_v0 *ref0;
3036 int ret;
3037 int slot;
3038
3039 leaf = path->nodes[0];
3040 slot = path->slots[0];
3041 while (1) {
3042 if (slot >= btrfs_header_nritems(leaf)) {
3043 ret = btrfs_next_leaf(rc->extent_root, path);
3044 if (ret < 0)
3045 return ret;
3046 BUG_ON(ret > 0);
3047 leaf = path->nodes[0];
3048 slot = path->slots[0];
3049 if (path_change)
3050 *path_change = 1;
3051 }
3052 btrfs_item_key_to_cpu(leaf, &key, slot);
3053 if (key.objectid != extent_key->objectid)
3054 return -ENOENT;
3055
3056 if (key.type != BTRFS_EXTENT_REF_V0_KEY) {
3057 slot++;
3058 continue;
3059 }
3060 ref0 = btrfs_item_ptr(leaf, slot,
3061 struct btrfs_extent_ref_v0);
3062 *ref_objectid = btrfs_ref_objectid_v0(leaf, ref0);
3063 break;
3064 }
3065 return 0;
3066 }
3067 #endif
3068
3069 /*
3070 * helper to add a tree block to the list.
3071 * the major work is getting the generation and level of the block
3072 */
3073 static int add_tree_block(struct reloc_control *rc,
3074 struct btrfs_key *extent_key,
3075 struct btrfs_path *path,
3076 struct rb_root *blocks)
3077 {
3078 struct extent_buffer *eb;
3079 struct btrfs_extent_item *ei;
3080 struct btrfs_tree_block_info *bi;
3081 struct tree_block *block;
3082 struct rb_node *rb_node;
3083 u32 item_size;
3084 int level = -1;
3085 int generation;
3086
3087 eb = path->nodes[0];
3088 item_size = btrfs_item_size_nr(eb, path->slots[0]);
3089
3090 if (item_size >= sizeof(*ei) + sizeof(*bi)) {
3091 ei = btrfs_item_ptr(eb, path->slots[0],
3092 struct btrfs_extent_item);
3093 bi = (struct btrfs_tree_block_info *)(ei + 1);
3094 generation = btrfs_extent_generation(eb, ei);
3095 level = btrfs_tree_block_level(eb, bi);
3096 } else {
3097 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
3098 u64 ref_owner;
3099 int ret;
3100
3101 BUG_ON(item_size != sizeof(struct btrfs_extent_item_v0));
3102 ret = get_ref_objectid_v0(rc, path, extent_key,
3103 &ref_owner, NULL);
3104 if (ret < 0)
3105 return ret;
3106 BUG_ON(ref_owner >= BTRFS_MAX_LEVEL);
3107 level = (int)ref_owner;
3108 /* FIXME: get real generation */
3109 generation = 0;
3110 #else
3111 BUG();
3112 #endif
3113 }
3114
3115 btrfs_release_path(rc->extent_root, path);
3116
3117 BUG_ON(level == -1);
3118
3119 block = kmalloc(sizeof(*block), GFP_NOFS);
3120 if (!block)
3121 return -ENOMEM;
3122
3123 block->bytenr = extent_key->objectid;
3124 block->key.objectid = extent_key->offset;
3125 block->key.offset = generation;
3126 block->level = level;
3127 block->key_ready = 0;
3128
3129 rb_node = tree_insert(blocks, block->bytenr, &block->rb_node);
3130 BUG_ON(rb_node);
3131
3132 return 0;
3133 }
3134
3135 /*
3136 * helper to add tree blocks for backref of type BTRFS_SHARED_DATA_REF_KEY
3137 */
3138 static int __add_tree_block(struct reloc_control *rc,
3139 u64 bytenr, u32 blocksize,
3140 struct rb_root *blocks)
3141 {
3142 struct btrfs_path *path;
3143 struct btrfs_key key;
3144 int ret;
3145
3146 if (tree_block_processed(bytenr, blocksize, rc))
3147 return 0;
3148
3149 if (tree_search(blocks, bytenr))
3150 return 0;
3151
3152 path = btrfs_alloc_path();
3153 if (!path)
3154 return -ENOMEM;
3155
3156 key.objectid = bytenr;
3157 key.type = BTRFS_EXTENT_ITEM_KEY;
3158 key.offset = blocksize;
3159
3160 path->search_commit_root = 1;
3161 path->skip_locking = 1;
3162 ret = btrfs_search_slot(NULL, rc->extent_root, &key, path, 0, 0);
3163 if (ret < 0)
3164 goto out;
3165 BUG_ON(ret);
3166
3167 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
3168 ret = add_tree_block(rc, &key, path, blocks);
3169 out:
3170 btrfs_free_path(path);
3171 return ret;
3172 }
3173
3174 /*
3175 * helper to check if the block use full backrefs for pointers in it
3176 */
3177 static int block_use_full_backref(struct reloc_control *rc,
3178 struct extent_buffer *eb)
3179 {
3180 u64 flags;
3181 int ret;
3182
3183 if (btrfs_header_flag(eb, BTRFS_HEADER_FLAG_RELOC) ||
3184 btrfs_header_backref_rev(eb) < BTRFS_MIXED_BACKREF_REV)
3185 return 1;
3186
3187 ret = btrfs_lookup_extent_info(NULL, rc->extent_root,
3188 eb->start, eb->len, NULL, &flags);
3189 BUG_ON(ret);
3190
3191 if (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)
3192 ret = 1;
3193 else
3194 ret = 0;
3195 return ret;
3196 }
3197
3198 static int delete_block_group_cache(struct btrfs_fs_info *fs_info,
3199 struct inode *inode, u64 ino)
3200 {
3201 struct btrfs_key key;
3202 struct btrfs_path *path;
3203 struct btrfs_root *root = fs_info->tree_root;
3204 struct btrfs_trans_handle *trans;
3205 unsigned long nr;
3206 int ret = 0;
3207
3208 if (inode)
3209 goto truncate;
3210
3211 key.objectid = ino;
3212 key.type = BTRFS_INODE_ITEM_KEY;
3213 key.offset = 0;
3214
3215 inode = btrfs_iget(fs_info->sb, &key, root, NULL);
3216 if (!inode || IS_ERR(inode) || is_bad_inode(inode)) {
3217 if (inode && !IS_ERR(inode))
3218 iput(inode);
3219 return -ENOENT;
3220 }
3221
3222 truncate:
3223 path = btrfs_alloc_path();
3224 if (!path) {
3225 ret = -ENOMEM;
3226 goto out;
3227 }
3228
3229 trans = btrfs_join_transaction(root, 0);
3230 if (IS_ERR(trans)) {
3231 btrfs_free_path(path);
3232 ret = PTR_ERR(trans);
3233 goto out;
3234 }
3235
3236 ret = btrfs_truncate_free_space_cache(root, trans, path, inode);
3237
3238 btrfs_free_path(path);
3239 nr = trans->blocks_used;
3240 btrfs_end_transaction(trans, root);
3241 btrfs_btree_balance_dirty(root, nr);
3242 out:
3243 iput(inode);
3244 return ret;
3245 }
3246
3247 /*
3248 * helper to add tree blocks for backref of type BTRFS_EXTENT_DATA_REF_KEY
3249 * this function scans fs tree to find blocks reference the data extent
3250 */
3251 static int find_data_references(struct reloc_control *rc,
3252 struct btrfs_key *extent_key,
3253 struct extent_buffer *leaf,
3254 struct btrfs_extent_data_ref *ref,
3255 struct rb_root *blocks)
3256 {
3257 struct btrfs_path *path;
3258 struct tree_block *block;
3259 struct btrfs_root *root;
3260 struct btrfs_file_extent_item *fi;
3261 struct rb_node *rb_node;
3262 struct btrfs_key key;
3263 u64 ref_root;
3264 u64 ref_objectid;
3265 u64 ref_offset;
3266 u32 ref_count;
3267 u32 nritems;
3268 int err = 0;
3269 int added = 0;
3270 int counted;
3271 int ret;
3272
3273 ref_root = btrfs_extent_data_ref_root(leaf, ref);
3274 ref_objectid = btrfs_extent_data_ref_objectid(leaf, ref);
3275 ref_offset = btrfs_extent_data_ref_offset(leaf, ref);
3276 ref_count = btrfs_extent_data_ref_count(leaf, ref);
3277
3278 /*
3279 * This is an extent belonging to the free space cache, lets just delete
3280 * it and redo the search.
3281 */
3282 if (ref_root == BTRFS_ROOT_TREE_OBJECTID) {
3283 ret = delete_block_group_cache(rc->extent_root->fs_info,
3284 NULL, ref_objectid);
3285 if (ret != -ENOENT)
3286 return ret;
3287 ret = 0;
3288 }
3289
3290 path = btrfs_alloc_path();
3291 if (!path)
3292 return -ENOMEM;
3293
3294 root = read_fs_root(rc->extent_root->fs_info, ref_root);
3295 if (IS_ERR(root)) {
3296 err = PTR_ERR(root);
3297 goto out;
3298 }
3299
3300 key.objectid = ref_objectid;
3301 key.offset = ref_offset;
3302 key.type = BTRFS_EXTENT_DATA_KEY;
3303
3304 path->search_commit_root = 1;
3305 path->skip_locking = 1;
3306 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3307 if (ret < 0) {
3308 err = ret;
3309 goto out;
3310 }
3311
3312 leaf = path->nodes[0];
3313 nritems = btrfs_header_nritems(leaf);
3314 /*
3315 * the references in tree blocks that use full backrefs
3316 * are not counted in
3317 */
3318 if (block_use_full_backref(rc, leaf))
3319 counted = 0;
3320 else
3321 counted = 1;
3322 rb_node = tree_search(blocks, leaf->start);
3323 if (rb_node) {
3324 if (counted)
3325 added = 1;
3326 else
3327 path->slots[0] = nritems;
3328 }
3329
3330 while (ref_count > 0) {
3331 while (path->slots[0] >= nritems) {
3332 ret = btrfs_next_leaf(root, path);
3333 if (ret < 0) {
3334 err = ret;
3335 goto out;
3336 }
3337 if (ret > 0) {
3338 WARN_ON(1);
3339 goto out;
3340 }
3341
3342 leaf = path->nodes[0];
3343 nritems = btrfs_header_nritems(leaf);
3344 added = 0;
3345
3346 if (block_use_full_backref(rc, leaf))
3347 counted = 0;
3348 else
3349 counted = 1;
3350 rb_node = tree_search(blocks, leaf->start);
3351 if (rb_node) {
3352 if (counted)
3353 added = 1;
3354 else
3355 path->slots[0] = nritems;
3356 }
3357 }
3358
3359 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3360 if (key.objectid != ref_objectid ||
3361 key.type != BTRFS_EXTENT_DATA_KEY) {
3362 WARN_ON(1);
3363 break;
3364 }
3365
3366 fi = btrfs_item_ptr(leaf, path->slots[0],
3367 struct btrfs_file_extent_item);
3368
3369 if (btrfs_file_extent_type(leaf, fi) ==
3370 BTRFS_FILE_EXTENT_INLINE)
3371 goto next;
3372
3373 if (btrfs_file_extent_disk_bytenr(leaf, fi) !=
3374 extent_key->objectid)
3375 goto next;
3376
3377 key.offset -= btrfs_file_extent_offset(leaf, fi);
3378 if (key.offset != ref_offset)
3379 goto next;
3380
3381 if (counted)
3382 ref_count--;
3383 if (added)
3384 goto next;
3385
3386 if (!tree_block_processed(leaf->start, leaf->len, rc)) {
3387 block = kmalloc(sizeof(*block), GFP_NOFS);
3388 if (!block) {
3389 err = -ENOMEM;
3390 break;
3391 }
3392 block->bytenr = leaf->start;
3393 btrfs_item_key_to_cpu(leaf, &block->key, 0);
3394 block->level = 0;
3395 block->key_ready = 1;
3396 rb_node = tree_insert(blocks, block->bytenr,
3397 &block->rb_node);
3398 BUG_ON(rb_node);
3399 }
3400 if (counted)
3401 added = 1;
3402 else
3403 path->slots[0] = nritems;
3404 next:
3405 path->slots[0]++;
3406
3407 }
3408 out:
3409 btrfs_free_path(path);
3410 return err;
3411 }
3412
3413 /*
3414 * hepler to find all tree blocks that reference a given data extent
3415 */
3416 static noinline_for_stack
3417 int add_data_references(struct reloc_control *rc,
3418 struct btrfs_key *extent_key,
3419 struct btrfs_path *path,
3420 struct rb_root *blocks)
3421 {
3422 struct btrfs_key key;
3423 struct extent_buffer *eb;
3424 struct btrfs_extent_data_ref *dref;
3425 struct btrfs_extent_inline_ref *iref;
3426 unsigned long ptr;
3427 unsigned long end;
3428 u32 blocksize = btrfs_level_size(rc->extent_root, 0);
3429 int ret;
3430 int err = 0;
3431
3432 eb = path->nodes[0];
3433 ptr = btrfs_item_ptr_offset(eb, path->slots[0]);
3434 end = ptr + btrfs_item_size_nr(eb, path->slots[0]);
3435 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
3436 if (ptr + sizeof(struct btrfs_extent_item_v0) == end)
3437 ptr = end;
3438 else
3439 #endif
3440 ptr += sizeof(struct btrfs_extent_item);
3441
3442 while (ptr < end) {
3443 iref = (struct btrfs_extent_inline_ref *)ptr;
3444 key.type = btrfs_extent_inline_ref_type(eb, iref);
3445 if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
3446 key.offset = btrfs_extent_inline_ref_offset(eb, iref);
3447 ret = __add_tree_block(rc, key.offset, blocksize,
3448 blocks);
3449 } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
3450 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
3451 ret = find_data_references(rc, extent_key,
3452 eb, dref, blocks);
3453 } else {
3454 BUG();
3455 }
3456 ptr += btrfs_extent_inline_ref_size(key.type);
3457 }
3458 WARN_ON(ptr > end);
3459
3460 while (1) {
3461 cond_resched();
3462 eb = path->nodes[0];
3463 if (path->slots[0] >= btrfs_header_nritems(eb)) {
3464 ret = btrfs_next_leaf(rc->extent_root, path);
3465 if (ret < 0) {
3466 err = ret;
3467 break;
3468 }
3469 if (ret > 0)
3470 break;
3471 eb = path->nodes[0];
3472 }
3473
3474 btrfs_item_key_to_cpu(eb, &key, path->slots[0]);
3475 if (key.objectid != extent_key->objectid)
3476 break;
3477
3478 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
3479 if (key.type == BTRFS_SHARED_DATA_REF_KEY ||
3480 key.type == BTRFS_EXTENT_REF_V0_KEY) {
3481 #else
3482 BUG_ON(key.type == BTRFS_EXTENT_REF_V0_KEY);
3483 if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
3484 #endif
3485 ret = __add_tree_block(rc, key.offset, blocksize,
3486 blocks);
3487 } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
3488 dref = btrfs_item_ptr(eb, path->slots[0],
3489 struct btrfs_extent_data_ref);
3490 ret = find_data_references(rc, extent_key,
3491 eb, dref, blocks);
3492 } else {
3493 ret = 0;
3494 }
3495 if (ret) {
3496 err = ret;
3497 break;
3498 }
3499 path->slots[0]++;
3500 }
3501 btrfs_release_path(rc->extent_root, path);
3502 if (err)
3503 free_block_list(blocks);
3504 return err;
3505 }
3506
3507 /*
3508 * hepler to find next unprocessed extent
3509 */
3510 static noinline_for_stack
3511 int find_next_extent(struct btrfs_trans_handle *trans,
3512 struct reloc_control *rc, struct btrfs_path *path,
3513 struct btrfs_key *extent_key)
3514 {
3515 struct btrfs_key key;
3516 struct extent_buffer *leaf;
3517 u64 start, end, last;
3518 int ret;
3519
3520 last = rc->block_group->key.objectid + rc->block_group->key.offset;
3521 while (1) {
3522 cond_resched();
3523 if (rc->search_start >= last) {
3524 ret = 1;
3525 break;
3526 }
3527
3528 key.objectid = rc->search_start;
3529 key.type = BTRFS_EXTENT_ITEM_KEY;
3530 key.offset = 0;
3531
3532 path->search_commit_root = 1;
3533 path->skip_locking = 1;
3534 ret = btrfs_search_slot(NULL, rc->extent_root, &key, path,
3535 0, 0);
3536 if (ret < 0)
3537 break;
3538 next:
3539 leaf = path->nodes[0];
3540 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
3541 ret = btrfs_next_leaf(rc->extent_root, path);
3542 if (ret != 0)
3543 break;
3544 leaf = path->nodes[0];
3545 }
3546
3547 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3548 if (key.objectid >= last) {
3549 ret = 1;
3550 break;
3551 }
3552
3553 if (key.type != BTRFS_EXTENT_ITEM_KEY ||
3554 key.objectid + key.offset <= rc->search_start) {
3555 path->slots[0]++;
3556 goto next;
3557 }
3558
3559 ret = find_first_extent_bit(&rc->processed_blocks,
3560 key.objectid, &start, &end,
3561 EXTENT_DIRTY);
3562
3563 if (ret == 0 && start <= key.objectid) {
3564 btrfs_release_path(rc->extent_root, path);
3565 rc->search_start = end + 1;
3566 } else {
3567 rc->search_start = key.objectid + key.offset;
3568 memcpy(extent_key, &key, sizeof(key));
3569 return 0;
3570 }
3571 }
3572 btrfs_release_path(rc->extent_root, path);
3573 return ret;
3574 }
3575
3576 static void set_reloc_control(struct reloc_control *rc)
3577 {
3578 struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
3579 mutex_lock(&fs_info->trans_mutex);
3580 fs_info->reloc_ctl = rc;
3581 mutex_unlock(&fs_info->trans_mutex);
3582 }
3583
3584 static void unset_reloc_control(struct reloc_control *rc)
3585 {
3586 struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
3587 mutex_lock(&fs_info->trans_mutex);
3588 fs_info->reloc_ctl = NULL;
3589 mutex_unlock(&fs_info->trans_mutex);
3590 }
3591
3592 static int check_extent_flags(u64 flags)
3593 {
3594 if ((flags & BTRFS_EXTENT_FLAG_DATA) &&
3595 (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK))
3596 return 1;
3597 if (!(flags & BTRFS_EXTENT_FLAG_DATA) &&
3598 !(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK))
3599 return 1;
3600 if ((flags & BTRFS_EXTENT_FLAG_DATA) &&
3601 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
3602 return 1;
3603 return 0;
3604 }
3605
3606 static noinline_for_stack
3607 int prepare_to_relocate(struct reloc_control *rc)
3608 {
3609 struct btrfs_trans_handle *trans;
3610 int ret;
3611
3612 rc->block_rsv = btrfs_alloc_block_rsv(rc->extent_root);
3613 if (!rc->block_rsv)
3614 return -ENOMEM;
3615
3616 /*
3617 * reserve some space for creating reloc trees.
3618 * btrfs_init_reloc_root will use them when there
3619 * is no reservation in transaction handle.
3620 */
3621 ret = btrfs_block_rsv_add(NULL, rc->extent_root, rc->block_rsv,
3622 rc->extent_root->nodesize * 256);
3623 if (ret)
3624 return ret;
3625
3626 rc->block_rsv->refill_used = 1;
3627 btrfs_add_durable_block_rsv(rc->extent_root->fs_info, rc->block_rsv);
3628
3629 memset(&rc->cluster, 0, sizeof(rc->cluster));
3630 rc->search_start = rc->block_group->key.objectid;
3631 rc->extents_found = 0;
3632 rc->nodes_relocated = 0;
3633 rc->merging_rsv_size = 0;
3634
3635 rc->create_reloc_tree = 1;
3636 set_reloc_control(rc);
3637
3638 trans = btrfs_join_transaction(rc->extent_root, 1);
3639 BUG_ON(IS_ERR(trans));
3640 btrfs_commit_transaction(trans, rc->extent_root);
3641 return 0;
3642 }
3643
3644 static noinline_for_stack int relocate_block_group(struct reloc_control *rc)
3645 {
3646 struct rb_root blocks = RB_ROOT;
3647 struct btrfs_key key;
3648 struct btrfs_trans_handle *trans = NULL;
3649 struct btrfs_path *path;
3650 struct btrfs_extent_item *ei;
3651 unsigned long nr;
3652 u64 flags;
3653 u32 item_size;
3654 int ret;
3655 int err = 0;
3656
3657 path = btrfs_alloc_path();
3658 if (!path)
3659 return -ENOMEM;
3660
3661 ret = prepare_to_relocate(rc);
3662 if (ret) {
3663 err = ret;
3664 goto out_free;
3665 }
3666
3667 while (1) {
3668 trans = btrfs_start_transaction(rc->extent_root, 0);
3669 BUG_ON(IS_ERR(trans));
3670
3671 if (update_backref_cache(trans, &rc->backref_cache)) {
3672 btrfs_end_transaction(trans, rc->extent_root);
3673 continue;
3674 }
3675
3676 ret = find_next_extent(trans, rc, path, &key);
3677 if (ret < 0)
3678 err = ret;
3679 if (ret != 0)
3680 break;
3681
3682 rc->extents_found++;
3683
3684 ei = btrfs_item_ptr(path->nodes[0], path->slots[0],
3685 struct btrfs_extent_item);
3686 item_size = btrfs_item_size_nr(path->nodes[0], path->slots[0]);
3687 if (item_size >= sizeof(*ei)) {
3688 flags = btrfs_extent_flags(path->nodes[0], ei);
3689 ret = check_extent_flags(flags);
3690 BUG_ON(ret);
3691
3692 } else {
3693 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
3694 u64 ref_owner;
3695 int path_change = 0;
3696
3697 BUG_ON(item_size !=
3698 sizeof(struct btrfs_extent_item_v0));
3699 ret = get_ref_objectid_v0(rc, path, &key, &ref_owner,
3700 &path_change);
3701 if (ref_owner < BTRFS_FIRST_FREE_OBJECTID)
3702 flags = BTRFS_EXTENT_FLAG_TREE_BLOCK;
3703 else
3704 flags = BTRFS_EXTENT_FLAG_DATA;
3705
3706 if (path_change) {
3707 btrfs_release_path(rc->extent_root, path);
3708
3709 path->search_commit_root = 1;
3710 path->skip_locking = 1;
3711 ret = btrfs_search_slot(NULL, rc->extent_root,
3712 &key, path, 0, 0);
3713 if (ret < 0) {
3714 err = ret;
3715 break;
3716 }
3717 BUG_ON(ret > 0);
3718 }
3719 #else
3720 BUG();
3721 #endif
3722 }
3723
3724 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
3725 ret = add_tree_block(rc, &key, path, &blocks);
3726 } else if (rc->stage == UPDATE_DATA_PTRS &&
3727 (flags & BTRFS_EXTENT_FLAG_DATA)) {
3728 ret = add_data_references(rc, &key, path, &blocks);
3729 } else {
3730 btrfs_release_path(rc->extent_root, path);
3731 ret = 0;
3732 }
3733 if (ret < 0) {
3734 err = ret;
3735 break;
3736 }
3737
3738 if (!RB_EMPTY_ROOT(&blocks)) {
3739 ret = relocate_tree_blocks(trans, rc, &blocks);
3740 if (ret < 0) {
3741 if (ret != -EAGAIN) {
3742 err = ret;
3743 break;
3744 }
3745 rc->extents_found--;
3746 rc->search_start = key.objectid;
3747 }
3748 }
3749
3750 ret = btrfs_block_rsv_check(trans, rc->extent_root,
3751 rc->block_rsv, 0, 5);
3752 if (ret < 0) {
3753 if (ret != -EAGAIN) {
3754 err = ret;
3755 WARN_ON(1);
3756 break;
3757 }
3758 rc->commit_transaction = 1;
3759 }
3760
3761 if (rc->commit_transaction) {
3762 rc->commit_transaction = 0;
3763 ret = btrfs_commit_transaction(trans, rc->extent_root);
3764 BUG_ON(ret);
3765 } else {
3766 nr = trans->blocks_used;
3767 btrfs_end_transaction_throttle(trans, rc->extent_root);
3768 btrfs_btree_balance_dirty(rc->extent_root, nr);
3769 }
3770 trans = NULL;
3771
3772 if (rc->stage == MOVE_DATA_EXTENTS &&
3773 (flags & BTRFS_EXTENT_FLAG_DATA)) {
3774 rc->found_file_extent = 1;
3775 ret = relocate_data_extent(rc->data_inode,
3776 &key, &rc->cluster);
3777 if (ret < 0) {
3778 err = ret;
3779 break;
3780 }
3781 }
3782 }
3783
3784 btrfs_release_path(rc->extent_root, path);
3785 clear_extent_bits(&rc->processed_blocks, 0, (u64)-1, EXTENT_DIRTY,
3786 GFP_NOFS);
3787
3788 if (trans) {
3789 nr = trans->blocks_used;
3790 btrfs_end_transaction_throttle(trans, rc->extent_root);
3791 btrfs_btree_balance_dirty(rc->extent_root, nr);
3792 }
3793
3794 if (!err) {
3795 ret = relocate_file_extent_cluster(rc->data_inode,
3796 &rc->cluster);
3797 if (ret < 0)
3798 err = ret;
3799 }
3800
3801 rc->create_reloc_tree = 0;
3802 set_reloc_control(rc);
3803
3804 backref_cache_cleanup(&rc->backref_cache);
3805 btrfs_block_rsv_release(rc->extent_root, rc->block_rsv, (u64)-1);
3806
3807 err = prepare_to_merge(rc, err);
3808
3809 merge_reloc_roots(rc);
3810
3811 rc->merge_reloc_tree = 0;
3812 unset_reloc_control(rc);
3813 btrfs_block_rsv_release(rc->extent_root, rc->block_rsv, (u64)-1);
3814
3815 /* get rid of pinned extents */
3816 trans = btrfs_join_transaction(rc->extent_root, 1);
3817 if (IS_ERR(trans))
3818 err = PTR_ERR(trans);
3819 else
3820 btrfs_commit_transaction(trans, rc->extent_root);
3821 out_free:
3822 btrfs_free_block_rsv(rc->extent_root, rc->block_rsv);
3823 btrfs_free_path(path);
3824 return err;
3825 }
3826
3827 static int __insert_orphan_inode(struct btrfs_trans_handle *trans,
3828 struct btrfs_root *root, u64 objectid)
3829 {
3830 struct btrfs_path *path;
3831 struct btrfs_inode_item *item;
3832 struct extent_buffer *leaf;
3833 int ret;
3834
3835 path = btrfs_alloc_path();
3836 if (!path)
3837 return -ENOMEM;
3838
3839 ret = btrfs_insert_empty_inode(trans, root, path, objectid);
3840 if (ret)
3841 goto out;
3842
3843 leaf = path->nodes[0];
3844 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_inode_item);
3845 memset_extent_buffer(leaf, 0, (unsigned long)item, sizeof(*item));
3846 btrfs_set_inode_generation(leaf, item, 1);
3847 btrfs_set_inode_size(leaf, item, 0);
3848 btrfs_set_inode_mode(leaf, item, S_IFREG | 0600);
3849 btrfs_set_inode_flags(leaf, item, BTRFS_INODE_NOCOMPRESS |
3850 BTRFS_INODE_PREALLOC);
3851 btrfs_mark_buffer_dirty(leaf);
3852 btrfs_release_path(root, path);
3853 out:
3854 btrfs_free_path(path);
3855 return ret;
3856 }
3857
3858 /*
3859 * helper to create inode for data relocation.
3860 * the inode is in data relocation tree and its link count is 0
3861 */
3862 static noinline_for_stack
3863 struct inode *create_reloc_inode(struct btrfs_fs_info *fs_info,
3864 struct btrfs_block_group_cache *group)
3865 {
3866 struct inode *inode = NULL;
3867 struct btrfs_trans_handle *trans;
3868 struct btrfs_root *root;
3869 struct btrfs_key key;
3870 unsigned long nr;
3871 u64 objectid = BTRFS_FIRST_FREE_OBJECTID;
3872 int err = 0;
3873
3874 root = read_fs_root(fs_info, BTRFS_DATA_RELOC_TREE_OBJECTID);
3875 if (IS_ERR(root))
3876 return ERR_CAST(root);
3877
3878 trans = btrfs_start_transaction(root, 6);
3879 if (IS_ERR(trans))
3880 return ERR_CAST(trans);
3881
3882 err = btrfs_find_free_objectid(trans, root, objectid, &objectid);
3883 if (err)
3884 goto out;
3885
3886 err = __insert_orphan_inode(trans, root, objectid);
3887 BUG_ON(err);
3888
3889 key.objectid = objectid;
3890 key.type = BTRFS_INODE_ITEM_KEY;
3891 key.offset = 0;
3892 inode = btrfs_iget(root->fs_info->sb, &key, root, NULL);
3893 BUG_ON(IS_ERR(inode) || is_bad_inode(inode));
3894 BTRFS_I(inode)->index_cnt = group->key.objectid;
3895
3896 err = btrfs_orphan_add(trans, inode);
3897 out:
3898 nr = trans->blocks_used;
3899 btrfs_end_transaction(trans, root);
3900 btrfs_btree_balance_dirty(root, nr);
3901 if (err) {
3902 if (inode)
3903 iput(inode);
3904 inode = ERR_PTR(err);
3905 }
3906 return inode;
3907 }
3908
3909 static struct reloc_control *alloc_reloc_control(void)
3910 {
3911 struct reloc_control *rc;
3912
3913 rc = kzalloc(sizeof(*rc), GFP_NOFS);
3914 if (!rc)
3915 return NULL;
3916
3917 INIT_LIST_HEAD(&rc->reloc_roots);
3918 backref_cache_init(&rc->backref_cache);
3919 mapping_tree_init(&rc->reloc_root_tree);
3920 extent_io_tree_init(&rc->processed_blocks, NULL, GFP_NOFS);
3921 return rc;
3922 }
3923
3924 /*
3925 * function to relocate all extents in a block group.
3926 */
3927 int btrfs_relocate_block_group(struct btrfs_root *extent_root, u64 group_start)
3928 {
3929 struct btrfs_fs_info *fs_info = extent_root->fs_info;
3930 struct reloc_control *rc;
3931 struct inode *inode;
3932 struct btrfs_path *path;
3933 int ret;
3934 int rw = 0;
3935 int err = 0;
3936
3937 rc = alloc_reloc_control();
3938 if (!rc)
3939 return -ENOMEM;
3940
3941 rc->extent_root = extent_root;
3942
3943 rc->block_group = btrfs_lookup_block_group(fs_info, group_start);
3944 BUG_ON(!rc->block_group);
3945
3946 if (!rc->block_group->ro) {
3947 ret = btrfs_set_block_group_ro(extent_root, rc->block_group);
3948 if (ret) {
3949 err = ret;
3950 goto out;
3951 }
3952 rw = 1;
3953 }
3954
3955 path = btrfs_alloc_path();
3956 if (!path) {
3957 err = -ENOMEM;
3958 goto out;
3959 }
3960
3961 inode = lookup_free_space_inode(fs_info->tree_root, rc->block_group,
3962 path);
3963 btrfs_free_path(path);
3964
3965 if (!IS_ERR(inode))
3966 ret = delete_block_group_cache(fs_info, inode, 0);
3967 else
3968 ret = PTR_ERR(inode);
3969
3970 if (ret && ret != -ENOENT) {
3971 err = ret;
3972 goto out;
3973 }
3974
3975 rc->data_inode = create_reloc_inode(fs_info, rc->block_group);
3976 if (IS_ERR(rc->data_inode)) {
3977 err = PTR_ERR(rc->data_inode);
3978 rc->data_inode = NULL;
3979 goto out;
3980 }
3981
3982 printk(KERN_INFO "btrfs: relocating block group %llu flags %llu\n",
3983 (unsigned long long)rc->block_group->key.objectid,
3984 (unsigned long long)rc->block_group->flags);
3985
3986 btrfs_start_delalloc_inodes(fs_info->tree_root, 0);
3987 btrfs_wait_ordered_extents(fs_info->tree_root, 0, 0);
3988
3989 while (1) {
3990 mutex_lock(&fs_info->cleaner_mutex);
3991
3992 btrfs_clean_old_snapshots(fs_info->tree_root);
3993 ret = relocate_block_group(rc);
3994
3995 mutex_unlock(&fs_info->cleaner_mutex);
3996 if (ret < 0) {
3997 err = ret;
3998 goto out;
3999 }
4000
4001 if (rc->extents_found == 0)
4002 break;
4003
4004 printk(KERN_INFO "btrfs: found %llu extents\n",
4005 (unsigned long long)rc->extents_found);
4006
4007 if (rc->stage == MOVE_DATA_EXTENTS && rc->found_file_extent) {
4008 btrfs_wait_ordered_range(rc->data_inode, 0, (u64)-1);
4009 invalidate_mapping_pages(rc->data_inode->i_mapping,
4010 0, -1);
4011 rc->stage = UPDATE_DATA_PTRS;
4012 }
4013 }
4014
4015 filemap_write_and_wait_range(fs_info->btree_inode->i_mapping,
4016 rc->block_group->key.objectid,
4017 rc->block_group->key.objectid +
4018 rc->block_group->key.offset - 1);
4019
4020 WARN_ON(rc->block_group->pinned > 0);
4021 WARN_ON(rc->block_group->reserved > 0);
4022 WARN_ON(btrfs_block_group_used(&rc->block_group->item) > 0);
4023 out:
4024 if (err && rw)
4025 btrfs_set_block_group_rw(extent_root, rc->block_group);
4026 iput(rc->data_inode);
4027 btrfs_put_block_group(rc->block_group);
4028 kfree(rc);
4029 return err;
4030 }
4031
4032 static noinline_for_stack int mark_garbage_root(struct btrfs_root *root)
4033 {
4034 struct btrfs_trans_handle *trans;
4035 int ret;
4036
4037 trans = btrfs_start_transaction(root->fs_info->tree_root, 0);
4038 BUG_ON(IS_ERR(trans));
4039
4040 memset(&root->root_item.drop_progress, 0,
4041 sizeof(root->root_item.drop_progress));
4042 root->root_item.drop_level = 0;
4043 btrfs_set_root_refs(&root->root_item, 0);
4044 ret = btrfs_update_root(trans, root->fs_info->tree_root,
4045 &root->root_key, &root->root_item);
4046 BUG_ON(ret);
4047
4048 ret = btrfs_end_transaction(trans, root->fs_info->tree_root);
4049 BUG_ON(ret);
4050 return 0;
4051 }
4052
4053 /*
4054 * recover relocation interrupted by system crash.
4055 *
4056 * this function resumes merging reloc trees with corresponding fs trees.
4057 * this is important for keeping the sharing of tree blocks
4058 */
4059 int btrfs_recover_relocation(struct btrfs_root *root)
4060 {
4061 LIST_HEAD(reloc_roots);
4062 struct btrfs_key key;
4063 struct btrfs_root *fs_root;
4064 struct btrfs_root *reloc_root;
4065 struct btrfs_path *path;
4066 struct extent_buffer *leaf;
4067 struct reloc_control *rc = NULL;
4068 struct btrfs_trans_handle *trans;
4069 int ret;
4070 int err = 0;
4071
4072 path = btrfs_alloc_path();
4073 if (!path)
4074 return -ENOMEM;
4075
4076 key.objectid = BTRFS_TREE_RELOC_OBJECTID;
4077 key.type = BTRFS_ROOT_ITEM_KEY;
4078 key.offset = (u64)-1;
4079
4080 while (1) {
4081 ret = btrfs_search_slot(NULL, root->fs_info->tree_root, &key,
4082 path, 0, 0);
4083 if (ret < 0) {
4084 err = ret;
4085 goto out;
4086 }
4087 if (ret > 0) {
4088 if (path->slots[0] == 0)
4089 break;
4090 path->slots[0]--;
4091 }
4092 leaf = path->nodes[0];
4093 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4094 btrfs_release_path(root->fs_info->tree_root, path);
4095
4096 if (key.objectid != BTRFS_TREE_RELOC_OBJECTID ||
4097 key.type != BTRFS_ROOT_ITEM_KEY)
4098 break;
4099
4100 reloc_root = btrfs_read_fs_root_no_radix(root, &key);
4101 if (IS_ERR(reloc_root)) {
4102 err = PTR_ERR(reloc_root);
4103 goto out;
4104 }
4105
4106 list_add(&reloc_root->root_list, &reloc_roots);
4107
4108 if (btrfs_root_refs(&reloc_root->root_item) > 0) {
4109 fs_root = read_fs_root(root->fs_info,
4110 reloc_root->root_key.offset);
4111 if (IS_ERR(fs_root)) {
4112 ret = PTR_ERR(fs_root);
4113 if (ret != -ENOENT) {
4114 err = ret;
4115 goto out;
4116 }
4117 mark_garbage_root(reloc_root);
4118 }
4119 }
4120
4121 if (key.offset == 0)
4122 break;
4123
4124 key.offset--;
4125 }
4126 btrfs_release_path(root->fs_info->tree_root, path);
4127
4128 if (list_empty(&reloc_roots))
4129 goto out;
4130
4131 rc = alloc_reloc_control();
4132 if (!rc) {
4133 err = -ENOMEM;
4134 goto out;
4135 }
4136
4137 rc->extent_root = root->fs_info->extent_root;
4138
4139 set_reloc_control(rc);
4140
4141 trans = btrfs_join_transaction(rc->extent_root, 1);
4142 if (IS_ERR(trans)) {
4143 unset_reloc_control(rc);
4144 err = PTR_ERR(trans);
4145 goto out_free;
4146 }
4147
4148 rc->merge_reloc_tree = 1;
4149
4150 while (!list_empty(&reloc_roots)) {
4151 reloc_root = list_entry(reloc_roots.next,
4152 struct btrfs_root, root_list);
4153 list_del(&reloc_root->root_list);
4154
4155 if (btrfs_root_refs(&reloc_root->root_item) == 0) {
4156 list_add_tail(&reloc_root->root_list,
4157 &rc->reloc_roots);
4158 continue;
4159 }
4160
4161 fs_root = read_fs_root(root->fs_info,
4162 reloc_root->root_key.offset);
4163 BUG_ON(IS_ERR(fs_root));
4164
4165 __add_reloc_root(reloc_root);
4166 fs_root->reloc_root = reloc_root;
4167 }
4168
4169 btrfs_commit_transaction(trans, rc->extent_root);
4170
4171 merge_reloc_roots(rc);
4172
4173 unset_reloc_control(rc);
4174
4175 trans = btrfs_join_transaction(rc->extent_root, 1);
4176 if (IS_ERR(trans))
4177 err = PTR_ERR(trans);
4178 else
4179 btrfs_commit_transaction(trans, rc->extent_root);
4180 out_free:
4181 kfree(rc);
4182 out:
4183 while (!list_empty(&reloc_roots)) {
4184 reloc_root = list_entry(reloc_roots.next,
4185 struct btrfs_root, root_list);
4186 list_del(&reloc_root->root_list);
4187 free_extent_buffer(reloc_root->node);
4188 free_extent_buffer(reloc_root->commit_root);
4189 kfree(reloc_root);
4190 }
4191 btrfs_free_path(path);
4192
4193 if (err == 0) {
4194 /* cleanup orphan inode in data relocation tree */
4195 fs_root = read_fs_root(root->fs_info,
4196 BTRFS_DATA_RELOC_TREE_OBJECTID);
4197 if (IS_ERR(fs_root))
4198 err = PTR_ERR(fs_root);
4199 else
4200 btrfs_orphan_cleanup(fs_root);
4201 }
4202 return err;
4203 }
4204
4205 /*
4206 * helper to add ordered checksum for data relocation.
4207 *
4208 * cloning checksum properly handles the nodatasum extents.
4209 * it also saves CPU time to re-calculate the checksum.
4210 */
4211 int btrfs_reloc_clone_csums(struct inode *inode, u64 file_pos, u64 len)
4212 {
4213 struct btrfs_ordered_sum *sums;
4214 struct btrfs_sector_sum *sector_sum;
4215 struct btrfs_ordered_extent *ordered;
4216 struct btrfs_root *root = BTRFS_I(inode)->root;
4217 size_t offset;
4218 int ret;
4219 u64 disk_bytenr;
4220 LIST_HEAD(list);
4221
4222 ordered = btrfs_lookup_ordered_extent(inode, file_pos);
4223 BUG_ON(ordered->file_offset != file_pos || ordered->len != len);
4224
4225 disk_bytenr = file_pos + BTRFS_I(inode)->index_cnt;
4226 ret = btrfs_lookup_csums_range(root->fs_info->csum_root, disk_bytenr,
4227 disk_bytenr + len - 1, &list);
4228
4229 while (!list_empty(&list)) {
4230 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
4231 list_del_init(&sums->list);
4232
4233 sector_sum = sums->sums;
4234 sums->bytenr = ordered->start;
4235
4236 offset = 0;
4237 while (offset < sums->len) {
4238 sector_sum->bytenr += ordered->start - disk_bytenr;
4239 sector_sum++;
4240 offset += root->sectorsize;
4241 }
4242
4243 btrfs_add_ordered_sum(inode, ordered, sums);
4244 }
4245 btrfs_put_ordered_extent(ordered);
4246 return ret;
4247 }
4248
4249 void btrfs_reloc_cow_block(struct btrfs_trans_handle *trans,
4250 struct btrfs_root *root, struct extent_buffer *buf,
4251 struct extent_buffer *cow)
4252 {
4253 struct reloc_control *rc;
4254 struct backref_node *node;
4255 int first_cow = 0;
4256 int level;
4257 int ret;
4258
4259 rc = root->fs_info->reloc_ctl;
4260 if (!rc)
4261 return;
4262
4263 BUG_ON(rc->stage == UPDATE_DATA_PTRS &&
4264 root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID);
4265
4266 level = btrfs_header_level(buf);
4267 if (btrfs_header_generation(buf) <=
4268 btrfs_root_last_snapshot(&root->root_item))
4269 first_cow = 1;
4270
4271 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID &&
4272 rc->create_reloc_tree) {
4273 WARN_ON(!first_cow && level == 0);
4274
4275 node = rc->backref_cache.path[level];
4276 BUG_ON(node->bytenr != buf->start &&
4277 node->new_bytenr != buf->start);
4278
4279 drop_node_buffer(node);
4280 extent_buffer_get(cow);
4281 node->eb = cow;
4282 node->new_bytenr = cow->start;
4283
4284 if (!node->pending) {
4285 list_move_tail(&node->list,
4286 &rc->backref_cache.pending[level]);
4287 node->pending = 1;
4288 }
4289
4290 if (first_cow)
4291 __mark_block_processed(rc, node);
4292
4293 if (first_cow && level > 0)
4294 rc->nodes_relocated += buf->len;
4295 }
4296
4297 if (level == 0 && first_cow && rc->stage == UPDATE_DATA_PTRS) {
4298 ret = replace_file_extents(trans, rc, root, cow);
4299 BUG_ON(ret);
4300 }
4301 }
4302
4303 /*
4304 * called before creating snapshot. it calculates metadata reservation
4305 * requried for relocating tree blocks in the snapshot
4306 */
4307 void btrfs_reloc_pre_snapshot(struct btrfs_trans_handle *trans,
4308 struct btrfs_pending_snapshot *pending,
4309 u64 *bytes_to_reserve)
4310 {
4311 struct btrfs_root *root;
4312 struct reloc_control *rc;
4313
4314 root = pending->root;
4315 if (!root->reloc_root)
4316 return;
4317
4318 rc = root->fs_info->reloc_ctl;
4319 if (!rc->merge_reloc_tree)
4320 return;
4321
4322 root = root->reloc_root;
4323 BUG_ON(btrfs_root_refs(&root->root_item) == 0);
4324 /*
4325 * relocation is in the stage of merging trees. the space
4326 * used by merging a reloc tree is twice the size of
4327 * relocated tree nodes in the worst case. half for cowing
4328 * the reloc tree, half for cowing the fs tree. the space
4329 * used by cowing the reloc tree will be freed after the
4330 * tree is dropped. if we create snapshot, cowing the fs
4331 * tree may use more space than it frees. so we need
4332 * reserve extra space.
4333 */
4334 *bytes_to_reserve += rc->nodes_relocated;
4335 }
4336
4337 /*
4338 * called after snapshot is created. migrate block reservation
4339 * and create reloc root for the newly created snapshot
4340 */
4341 void btrfs_reloc_post_snapshot(struct btrfs_trans_handle *trans,
4342 struct btrfs_pending_snapshot *pending)
4343 {
4344 struct btrfs_root *root = pending->root;
4345 struct btrfs_root *reloc_root;
4346 struct btrfs_root *new_root;
4347 struct reloc_control *rc;
4348 int ret;
4349
4350 if (!root->reloc_root)
4351 return;
4352
4353 rc = root->fs_info->reloc_ctl;
4354 rc->merging_rsv_size += rc->nodes_relocated;
4355
4356 if (rc->merge_reloc_tree) {
4357 ret = btrfs_block_rsv_migrate(&pending->block_rsv,
4358 rc->block_rsv,
4359 rc->nodes_relocated);
4360 BUG_ON(ret);
4361 }
4362
4363 new_root = pending->snap;
4364 reloc_root = create_reloc_root(trans, root->reloc_root,
4365 new_root->root_key.objectid);
4366
4367 __add_reloc_root(reloc_root);
4368 new_root->reloc_root = reloc_root;
4369
4370 if (rc->create_reloc_tree) {
4371 ret = clone_backref_node(trans, rc, root, reloc_root);
4372 BUG_ON(ret);
4373 }
4374 }
Linux kernel - Deliverables
This page took 0.227279 seconds and 5 git commands to generate.