]> git.proxmox.com Git - mirror_ubuntu-artful-kernel.git/blob - fs/btrfs/relocation.c
projects / mirror_ubuntu-artful-kernel.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
btrfs: root->fs_info cleanup, add fs_info convenience variables
[mirror_ubuntu-artful-kernel.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 #include "inode-map.h"
34 #include "qgroup.h"
35
36 /*
37 * backref_node, mapping_node and tree_block start with this
38 */
39 struct tree_entry {
40 struct rb_node rb_node;
41 u64 bytenr;
42 };
43
44 /*
45 * present a tree block in the backref cache
46 */
47 struct backref_node {
48 struct rb_node rb_node;
49 u64 bytenr;
50
51 u64 new_bytenr;
52 /* objectid of tree block owner, can be not uptodate */
53 u64 owner;
54 /* link to pending, changed or detached list */
55 struct list_head list;
56 /* list of upper level blocks reference this block */
57 struct list_head upper;
58 /* list of child blocks in the cache */
59 struct list_head lower;
60 /* NULL if this node is not tree root */
61 struct btrfs_root *root;
62 /* extent buffer got by COW the block */
63 struct extent_buffer *eb;
64 /* level of tree block */
65 unsigned int level:8;
66 /* is the block in non-reference counted tree */
67 unsigned int cowonly:1;
68 /* 1 if no child node in the cache */
69 unsigned int lowest:1;
70 /* is the extent buffer locked */
71 unsigned int locked:1;
72 /* has the block been processed */
73 unsigned int processed:1;
74 /* have backrefs of this block been checked */
75 unsigned int checked:1;
76 /*
77 * 1 if corresponding block has been cowed but some upper
78 * level block pointers may not point to the new location
79 */
80 unsigned int pending:1;
81 /*
82 * 1 if the backref node isn't connected to any other
83 * backref node.
84 */
85 unsigned int detached:1;
86 };
87
88 /*
89 * present a block pointer in the backref cache
90 */
91 struct backref_edge {
92 struct list_head list[2];
93 struct backref_node *node[2];
94 };
95
96 #define LOWER 0
97 #define UPPER 1
98 #define RELOCATION_RESERVED_NODES 256
99
100 struct backref_cache {
101 /* red black tree of all backref nodes in the cache */
102 struct rb_root rb_root;
103 /* for passing backref nodes to btrfs_reloc_cow_block */
104 struct backref_node *path[BTRFS_MAX_LEVEL];
105 /*
106 * list of blocks that have been cowed but some block
107 * pointers in upper level blocks may not reflect the
108 * new location
109 */
110 struct list_head pending[BTRFS_MAX_LEVEL];
111 /* list of backref nodes with no child node */
112 struct list_head leaves;
113 /* list of blocks that have been cowed in current transaction */
114 struct list_head changed;
115 /* list of detached backref node. */
116 struct list_head detached;
117
118 u64 last_trans;
119
120 int nr_nodes;
121 int nr_edges;
122 };
123
124 /*
125 * map address of tree root to tree
126 */
127 struct mapping_node {
128 struct rb_node rb_node;
129 u64 bytenr;
130 void *data;
131 };
132
133 struct mapping_tree {
134 struct rb_root rb_root;
135 spinlock_t lock;
136 };
137
138 /*
139 * present a tree block to process
140 */
141 struct tree_block {
142 struct rb_node rb_node;
143 u64 bytenr;
144 struct btrfs_key key;
145 unsigned int level:8;
146 unsigned int key_ready:1;
147 };
148
149 #define MAX_EXTENTS 128
150
151 struct file_extent_cluster {
152 u64 start;
153 u64 end;
154 u64 boundary[MAX_EXTENTS];
155 unsigned int nr;
156 };
157
158 struct reloc_control {
159 /* block group to relocate */
160 struct btrfs_block_group_cache *block_group;
161 /* extent tree */
162 struct btrfs_root *extent_root;
163 /* inode for moving data */
164 struct inode *data_inode;
165
166 struct btrfs_block_rsv *block_rsv;
167
168 struct backref_cache backref_cache;
169
170 struct file_extent_cluster cluster;
171 /* tree blocks have been processed */
172 struct extent_io_tree processed_blocks;
173 /* map start of tree root to corresponding reloc tree */
174 struct mapping_tree reloc_root_tree;
175 /* list of reloc trees */
176 struct list_head reloc_roots;
177 /* size of metadata reservation for merging reloc trees */
178 u64 merging_rsv_size;
179 /* size of relocated tree nodes */
180 u64 nodes_relocated;
181 /* reserved size for block group relocation*/
182 u64 reserved_bytes;
183
184 u64 search_start;
185 u64 extents_found;
186
187 unsigned int stage:8;
188 unsigned int create_reloc_tree:1;
189 unsigned int merge_reloc_tree:1;
190 unsigned int found_file_extent:1;
191 };
192
193 /* stages of data relocation */
194 #define MOVE_DATA_EXTENTS 0
195 #define UPDATE_DATA_PTRS 1
196
197 static void remove_backref_node(struct backref_cache *cache,
198 struct backref_node *node);
199 static void __mark_block_processed(struct reloc_control *rc,
200 struct backref_node *node);
201
202 static void mapping_tree_init(struct mapping_tree *tree)
203 {
204 tree->rb_root = RB_ROOT;
205 spin_lock_init(&tree->lock);
206 }
207
208 static void backref_cache_init(struct backref_cache *cache)
209 {
210 int i;
211 cache->rb_root = RB_ROOT;
212 for (i = 0; i < BTRFS_MAX_LEVEL; i++)
213 INIT_LIST_HEAD(&cache->pending[i]);
214 INIT_LIST_HEAD(&cache->changed);
215 INIT_LIST_HEAD(&cache->detached);
216 INIT_LIST_HEAD(&cache->leaves);
217 }
218
219 static void backref_cache_cleanup(struct backref_cache *cache)
220 {
221 struct backref_node *node;
222 int i;
223
224 while (!list_empty(&cache->detached)) {
225 node = list_entry(cache->detached.next,
226 struct backref_node, list);
227 remove_backref_node(cache, node);
228 }
229
230 while (!list_empty(&cache->leaves)) {
231 node = list_entry(cache->leaves.next,
232 struct backref_node, lower);
233 remove_backref_node(cache, node);
234 }
235
236 cache->last_trans = 0;
237
238 for (i = 0; i < BTRFS_MAX_LEVEL; i++)
239 ASSERT(list_empty(&cache->pending[i]));
240 ASSERT(list_empty(&cache->changed));
241 ASSERT(list_empty(&cache->detached));
242 ASSERT(RB_EMPTY_ROOT(&cache->rb_root));
243 ASSERT(!cache->nr_nodes);
244 ASSERT(!cache->nr_edges);
245 }
246
247 static struct backref_node *alloc_backref_node(struct backref_cache *cache)
248 {
249 struct backref_node *node;
250
251 node = kzalloc(sizeof(*node), GFP_NOFS);
252 if (node) {
253 INIT_LIST_HEAD(&node->list);
254 INIT_LIST_HEAD(&node->upper);
255 INIT_LIST_HEAD(&node->lower);
256 RB_CLEAR_NODE(&node->rb_node);
257 cache->nr_nodes++;
258 }
259 return node;
260 }
261
262 static void free_backref_node(struct backref_cache *cache,
263 struct backref_node *node)
264 {
265 if (node) {
266 cache->nr_nodes--;
267 kfree(node);
268 }
269 }
270
271 static struct backref_edge *alloc_backref_edge(struct backref_cache *cache)
272 {
273 struct backref_edge *edge;
274
275 edge = kzalloc(sizeof(*edge), GFP_NOFS);
276 if (edge)
277 cache->nr_edges++;
278 return edge;
279 }
280
281 static void free_backref_edge(struct backref_cache *cache,
282 struct backref_edge *edge)
283 {
284 if (edge) {
285 cache->nr_edges--;
286 kfree(edge);
287 }
288 }
289
290 static struct rb_node *tree_insert(struct rb_root *root, u64 bytenr,
291 struct rb_node *node)
292 {
293 struct rb_node **p = &root->rb_node;
294 struct rb_node *parent = NULL;
295 struct tree_entry *entry;
296
297 while (*p) {
298 parent = *p;
299 entry = rb_entry(parent, struct tree_entry, rb_node);
300
301 if (bytenr < entry->bytenr)
302 p = &(*p)->rb_left;
303 else if (bytenr > entry->bytenr)
304 p = &(*p)->rb_right;
305 else
306 return parent;
307 }
308
309 rb_link_node(node, parent, p);
310 rb_insert_color(node, root);
311 return NULL;
312 }
313
314 static struct rb_node *tree_search(struct rb_root *root, u64 bytenr)
315 {
316 struct rb_node *n = root->rb_node;
317 struct tree_entry *entry;
318
319 while (n) {
320 entry = rb_entry(n, struct tree_entry, rb_node);
321
322 if (bytenr < entry->bytenr)
323 n = n->rb_left;
324 else if (bytenr > entry->bytenr)
325 n = n->rb_right;
326 else
327 return n;
328 }
329 return NULL;
330 }
331
332 static void backref_tree_panic(struct rb_node *rb_node, int errno, u64 bytenr)
333 {
334
335 struct btrfs_fs_info *fs_info = NULL;
336 struct backref_node *bnode = rb_entry(rb_node, struct backref_node,
337 rb_node);
338 if (bnode->root)
339 fs_info = bnode->root->fs_info;
340 btrfs_panic(fs_info, errno,
341 "Inconsistency in backref cache found at offset %llu",
342 bytenr);
343 }
344
345 /*
346 * walk up backref nodes until reach node presents tree root
347 */
348 static struct backref_node *walk_up_backref(struct backref_node *node,
349 struct backref_edge *edges[],
350 int *index)
351 {
352 struct backref_edge *edge;
353 int idx = *index;
354
355 while (!list_empty(&node->upper)) {
356 edge = list_entry(node->upper.next,
357 struct backref_edge, list[LOWER]);
358 edges[idx++] = edge;
359 node = edge->node[UPPER];
360 }
361 BUG_ON(node->detached);
362 *index = idx;
363 return node;
364 }
365
366 /*
367 * walk down backref nodes to find start of next reference path
368 */
369 static struct backref_node *walk_down_backref(struct backref_edge *edges[],
370 int *index)
371 {
372 struct backref_edge *edge;
373 struct backref_node *lower;
374 int idx = *index;
375
376 while (idx > 0) {
377 edge = edges[idx - 1];
378 lower = edge->node[LOWER];
379 if (list_is_last(&edge->list[LOWER], &lower->upper)) {
380 idx--;
381 continue;
382 }
383 edge = list_entry(edge->list[LOWER].next,
384 struct backref_edge, list[LOWER]);
385 edges[idx - 1] = edge;
386 *index = idx;
387 return edge->node[UPPER];
388 }
389 *index = 0;
390 return NULL;
391 }
392
393 static void unlock_node_buffer(struct backref_node *node)
394 {
395 if (node->locked) {
396 btrfs_tree_unlock(node->eb);
397 node->locked = 0;
398 }
399 }
400
401 static void drop_node_buffer(struct backref_node *node)
402 {
403 if (node->eb) {
404 unlock_node_buffer(node);
405 free_extent_buffer(node->eb);
406 node->eb = NULL;
407 }
408 }
409
410 static void drop_backref_node(struct backref_cache *tree,
411 struct backref_node *node)
412 {
413 BUG_ON(!list_empty(&node->upper));
414
415 drop_node_buffer(node);
416 list_del(&node->list);
417 list_del(&node->lower);
418 if (!RB_EMPTY_NODE(&node->rb_node))
419 rb_erase(&node->rb_node, &tree->rb_root);
420 free_backref_node(tree, node);
421 }
422
423 /*
424 * remove a backref node from the backref cache
425 */
426 static void remove_backref_node(struct backref_cache *cache,
427 struct backref_node *node)
428 {
429 struct backref_node *upper;
430 struct backref_edge *edge;
431
432 if (!node)
433 return;
434
435 BUG_ON(!node->lowest && !node->detached);
436 while (!list_empty(&node->upper)) {
437 edge = list_entry(node->upper.next, struct backref_edge,
438 list[LOWER]);
439 upper = edge->node[UPPER];
440 list_del(&edge->list[LOWER]);
441 list_del(&edge->list[UPPER]);
442 free_backref_edge(cache, edge);
443
444 if (RB_EMPTY_NODE(&upper->rb_node)) {
445 BUG_ON(!list_empty(&node->upper));
446 drop_backref_node(cache, node);
447 node = upper;
448 node->lowest = 1;
449 continue;
450 }
451 /*
452 * add the node to leaf node list if no other
453 * child block cached.
454 */
455 if (list_empty(&upper->lower)) {
456 list_add_tail(&upper->lower, &cache->leaves);
457 upper->lowest = 1;
458 }
459 }
460
461 drop_backref_node(cache, node);
462 }
463
464 static void update_backref_node(struct backref_cache *cache,
465 struct backref_node *node, u64 bytenr)
466 {
467 struct rb_node *rb_node;
468 rb_erase(&node->rb_node, &cache->rb_root);
469 node->bytenr = bytenr;
470 rb_node = tree_insert(&cache->rb_root, node->bytenr, &node->rb_node);
471 if (rb_node)
472 backref_tree_panic(rb_node, -EEXIST, bytenr);
473 }
474
475 /*
476 * update backref cache after a transaction commit
477 */
478 static int update_backref_cache(struct btrfs_trans_handle *trans,
479 struct backref_cache *cache)
480 {
481 struct backref_node *node;
482 int level = 0;
483
484 if (cache->last_trans == 0) {
485 cache->last_trans = trans->transid;
486 return 0;
487 }
488
489 if (cache->last_trans == trans->transid)
490 return 0;
491
492 /*
493 * detached nodes are used to avoid unnecessary backref
494 * lookup. transaction commit changes the extent tree.
495 * so the detached nodes are no longer useful.
496 */
497 while (!list_empty(&cache->detached)) {
498 node = list_entry(cache->detached.next,
499 struct backref_node, list);
500 remove_backref_node(cache, node);
501 }
502
503 while (!list_empty(&cache->changed)) {
504 node = list_entry(cache->changed.next,
505 struct backref_node, list);
506 list_del_init(&node->list);
507 BUG_ON(node->pending);
508 update_backref_node(cache, node, node->new_bytenr);
509 }
510
511 /*
512 * some nodes can be left in the pending list if there were
513 * errors during processing the pending nodes.
514 */
515 for (level = 0; level < BTRFS_MAX_LEVEL; level++) {
516 list_for_each_entry(node, &cache->pending[level], list) {
517 BUG_ON(!node->pending);
518 if (node->bytenr == node->new_bytenr)
519 continue;
520 update_backref_node(cache, node, node->new_bytenr);
521 }
522 }
523
524 cache->last_trans = 0;
525 return 1;
526 }
527
528
529 static int should_ignore_root(struct btrfs_root *root)
530 {
531 struct btrfs_root *reloc_root;
532
533 if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state))
534 return 0;
535
536 reloc_root = root->reloc_root;
537 if (!reloc_root)
538 return 0;
539
540 if (btrfs_root_last_snapshot(&reloc_root->root_item) ==
541 root->fs_info->running_transaction->transid - 1)
542 return 0;
543 /*
544 * if there is reloc tree and it was created in previous
545 * transaction backref lookup can find the reloc tree,
546 * so backref node for the fs tree root is useless for
547 * relocation.
548 */
549 return 1;
550 }
551 /*
552 * find reloc tree by address of tree root
553 */
554 static struct btrfs_root *find_reloc_root(struct reloc_control *rc,
555 u64 bytenr)
556 {
557 struct rb_node *rb_node;
558 struct mapping_node *node;
559 struct btrfs_root *root = NULL;
560
561 spin_lock(&rc->reloc_root_tree.lock);
562 rb_node = tree_search(&rc->reloc_root_tree.rb_root, bytenr);
563 if (rb_node) {
564 node = rb_entry(rb_node, struct mapping_node, rb_node);
565 root = (struct btrfs_root *)node->data;
566 }
567 spin_unlock(&rc->reloc_root_tree.lock);
568 return root;
569 }
570
571 static int is_cowonly_root(u64 root_objectid)
572 {
573 if (root_objectid == BTRFS_ROOT_TREE_OBJECTID ||
574 root_objectid == BTRFS_EXTENT_TREE_OBJECTID ||
575 root_objectid == BTRFS_CHUNK_TREE_OBJECTID ||
576 root_objectid == BTRFS_DEV_TREE_OBJECTID ||
577 root_objectid == BTRFS_TREE_LOG_OBJECTID ||
578 root_objectid == BTRFS_CSUM_TREE_OBJECTID ||
579 root_objectid == BTRFS_UUID_TREE_OBJECTID ||
580 root_objectid == BTRFS_QUOTA_TREE_OBJECTID ||
581 root_objectid == BTRFS_FREE_SPACE_TREE_OBJECTID)
582 return 1;
583 return 0;
584 }
585
586 static struct btrfs_root *read_fs_root(struct btrfs_fs_info *fs_info,
587 u64 root_objectid)
588 {
589 struct btrfs_key key;
590
591 key.objectid = root_objectid;
592 key.type = BTRFS_ROOT_ITEM_KEY;
593 if (is_cowonly_root(root_objectid))
594 key.offset = 0;
595 else
596 key.offset = (u64)-1;
597
598 return btrfs_get_fs_root(fs_info, &key, false);
599 }
600
601 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
602 static noinline_for_stack
603 struct btrfs_root *find_tree_root(struct reloc_control *rc,
604 struct extent_buffer *leaf,
605 struct btrfs_extent_ref_v0 *ref0)
606 {
607 struct btrfs_root *root;
608 u64 root_objectid = btrfs_ref_root_v0(leaf, ref0);
609 u64 generation = btrfs_ref_generation_v0(leaf, ref0);
610
611 BUG_ON(root_objectid == BTRFS_TREE_RELOC_OBJECTID);
612
613 root = read_fs_root(rc->extent_root->fs_info, root_objectid);
614 BUG_ON(IS_ERR(root));
615
616 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) &&
617 generation != btrfs_root_generation(&root->root_item))
618 return NULL;
619
620 return root;
621 }
622 #endif
623
624 static noinline_for_stack
625 int find_inline_backref(struct extent_buffer *leaf, int slot,
626 unsigned long *ptr, unsigned long *end)
627 {
628 struct btrfs_key key;
629 struct btrfs_extent_item *ei;
630 struct btrfs_tree_block_info *bi;
631 u32 item_size;
632
633 btrfs_item_key_to_cpu(leaf, &key, slot);
634
635 item_size = btrfs_item_size_nr(leaf, slot);
636 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
637 if (item_size < sizeof(*ei)) {
638 WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0));
639 return 1;
640 }
641 #endif
642 ei = btrfs_item_ptr(leaf, slot, struct btrfs_extent_item);
643 WARN_ON(!(btrfs_extent_flags(leaf, ei) &
644 BTRFS_EXTENT_FLAG_TREE_BLOCK));
645
646 if (key.type == BTRFS_EXTENT_ITEM_KEY &&
647 item_size <= sizeof(*ei) + sizeof(*bi)) {
648 WARN_ON(item_size < sizeof(*ei) + sizeof(*bi));
649 return 1;
650 }
651 if (key.type == BTRFS_METADATA_ITEM_KEY &&
652 item_size <= sizeof(*ei)) {
653 WARN_ON(item_size < sizeof(*ei));
654 return 1;
655 }
656
657 if (key.type == BTRFS_EXTENT_ITEM_KEY) {
658 bi = (struct btrfs_tree_block_info *)(ei + 1);
659 *ptr = (unsigned long)(bi + 1);
660 } else {
661 *ptr = (unsigned long)(ei + 1);
662 }
663 *end = (unsigned long)ei + item_size;
664 return 0;
665 }
666
667 /*
668 * build backref tree for a given tree block. root of the backref tree
669 * corresponds the tree block, leaves of the backref tree correspond
670 * roots of b-trees that reference the tree block.
671 *
672 * the basic idea of this function is check backrefs of a given block
673 * to find upper level blocks that reference the block, and then check
674 * backrefs of these upper level blocks recursively. the recursion stop
675 * when tree root is reached or backrefs for the block is cached.
676 *
677 * NOTE: if we find backrefs for a block are cached, we know backrefs
678 * for all upper level blocks that directly/indirectly reference the
679 * block are also cached.
680 */
681 static noinline_for_stack
682 struct backref_node *build_backref_tree(struct reloc_control *rc,
683 struct btrfs_key *node_key,
684 int level, u64 bytenr)
685 {
686 struct backref_cache *cache = &rc->backref_cache;
687 struct btrfs_path *path1;
688 struct btrfs_path *path2;
689 struct extent_buffer *eb;
690 struct btrfs_root *root;
691 struct backref_node *cur;
692 struct backref_node *upper;
693 struct backref_node *lower;
694 struct backref_node *node = NULL;
695 struct backref_node *exist = NULL;
696 struct backref_edge *edge;
697 struct rb_node *rb_node;
698 struct btrfs_key key;
699 unsigned long end;
700 unsigned long ptr;
701 LIST_HEAD(list);
702 LIST_HEAD(useless);
703 int cowonly;
704 int ret;
705 int err = 0;
706 bool need_check = true;
707
708 path1 = btrfs_alloc_path();
709 path2 = btrfs_alloc_path();
710 if (!path1 || !path2) {
711 err = -ENOMEM;
712 goto out;
713 }
714 path1->reada = READA_FORWARD;
715 path2->reada = READA_FORWARD;
716
717 node = alloc_backref_node(cache);
718 if (!node) {
719 err = -ENOMEM;
720 goto out;
721 }
722
723 node->bytenr = bytenr;
724 node->level = level;
725 node->lowest = 1;
726 cur = node;
727 again:
728 end = 0;
729 ptr = 0;
730 key.objectid = cur->bytenr;
731 key.type = BTRFS_METADATA_ITEM_KEY;
732 key.offset = (u64)-1;
733
734 path1->search_commit_root = 1;
735 path1->skip_locking = 1;
736 ret = btrfs_search_slot(NULL, rc->extent_root, &key, path1,
737 0, 0);
738 if (ret < 0) {
739 err = ret;
740 goto out;
741 }
742 ASSERT(ret);
743 ASSERT(path1->slots[0]);
744
745 path1->slots[0]--;
746
747 WARN_ON(cur->checked);
748 if (!list_empty(&cur->upper)) {
749 /*
750 * the backref was added previously when processing
751 * backref of type BTRFS_TREE_BLOCK_REF_KEY
752 */
753 ASSERT(list_is_singular(&cur->upper));
754 edge = list_entry(cur->upper.next, struct backref_edge,
755 list[LOWER]);
756 ASSERT(list_empty(&edge->list[UPPER]));
757 exist = edge->node[UPPER];
758 /*
759 * add the upper level block to pending list if we need
760 * check its backrefs
761 */
762 if (!exist->checked)
763 list_add_tail(&edge->list[UPPER], &list);
764 } else {
765 exist = NULL;
766 }
767
768 while (1) {
769 cond_resched();
770 eb = path1->nodes[0];
771
772 if (ptr >= end) {
773 if (path1->slots[0] >= btrfs_header_nritems(eb)) {
774 ret = btrfs_next_leaf(rc->extent_root, path1);
775 if (ret < 0) {
776 err = ret;
777 goto out;
778 }
779 if (ret > 0)
780 break;
781 eb = path1->nodes[0];
782 }
783
784 btrfs_item_key_to_cpu(eb, &key, path1->slots[0]);
785 if (key.objectid != cur->bytenr) {
786 WARN_ON(exist);
787 break;
788 }
789
790 if (key.type == BTRFS_EXTENT_ITEM_KEY ||
791 key.type == BTRFS_METADATA_ITEM_KEY) {
792 ret = find_inline_backref(eb, path1->slots[0],
793 &ptr, &end);
794 if (ret)
795 goto next;
796 }
797 }
798
799 if (ptr < end) {
800 /* update key for inline back ref */
801 struct btrfs_extent_inline_ref *iref;
802 iref = (struct btrfs_extent_inline_ref *)ptr;
803 key.type = btrfs_extent_inline_ref_type(eb, iref);
804 key.offset = btrfs_extent_inline_ref_offset(eb, iref);
805 WARN_ON(key.type != BTRFS_TREE_BLOCK_REF_KEY &&
806 key.type != BTRFS_SHARED_BLOCK_REF_KEY);
807 }
808
809 if (exist &&
810 ((key.type == BTRFS_TREE_BLOCK_REF_KEY &&
811 exist->owner == key.offset) ||
812 (key.type == BTRFS_SHARED_BLOCK_REF_KEY &&
813 exist->bytenr == key.offset))) {
814 exist = NULL;
815 goto next;
816 }
817
818 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
819 if (key.type == BTRFS_SHARED_BLOCK_REF_KEY ||
820 key.type == BTRFS_EXTENT_REF_V0_KEY) {
821 if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
822 struct btrfs_extent_ref_v0 *ref0;
823 ref0 = btrfs_item_ptr(eb, path1->slots[0],
824 struct btrfs_extent_ref_v0);
825 if (key.objectid == key.offset) {
826 root = find_tree_root(rc, eb, ref0);
827 if (root && !should_ignore_root(root))
828 cur->root = root;
829 else
830 list_add(&cur->list, &useless);
831 break;
832 }
833 if (is_cowonly_root(btrfs_ref_root_v0(eb,
834 ref0)))
835 cur->cowonly = 1;
836 }
837 #else
838 ASSERT(key.type != BTRFS_EXTENT_REF_V0_KEY);
839 if (key.type == BTRFS_SHARED_BLOCK_REF_KEY) {
840 #endif
841 if (key.objectid == key.offset) {
842 /*
843 * only root blocks of reloc trees use
844 * backref of this type.
845 */
846 root = find_reloc_root(rc, cur->bytenr);
847 ASSERT(root);
848 cur->root = root;
849 break;
850 }
851
852 edge = alloc_backref_edge(cache);
853 if (!edge) {
854 err = -ENOMEM;
855 goto out;
856 }
857 rb_node = tree_search(&cache->rb_root, key.offset);
858 if (!rb_node) {
859 upper = alloc_backref_node(cache);
860 if (!upper) {
861 free_backref_edge(cache, edge);
862 err = -ENOMEM;
863 goto out;
864 }
865 upper->bytenr = key.offset;
866 upper->level = cur->level + 1;
867 /*
868 * backrefs for the upper level block isn't
869 * cached, add the block to pending list
870 */
871 list_add_tail(&edge->list[UPPER], &list);
872 } else {
873 upper = rb_entry(rb_node, struct backref_node,
874 rb_node);
875 ASSERT(upper->checked);
876 INIT_LIST_HEAD(&edge->list[UPPER]);
877 }
878 list_add_tail(&edge->list[LOWER], &cur->upper);
879 edge->node[LOWER] = cur;
880 edge->node[UPPER] = upper;
881
882 goto next;
883 } else if (key.type != BTRFS_TREE_BLOCK_REF_KEY) {
884 goto next;
885 }
886
887 /* key.type == BTRFS_TREE_BLOCK_REF_KEY */
888 root = read_fs_root(rc->extent_root->fs_info, key.offset);
889 if (IS_ERR(root)) {
890 err = PTR_ERR(root);
891 goto out;
892 }
893
894 if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state))
895 cur->cowonly = 1;
896
897 if (btrfs_root_level(&root->root_item) == cur->level) {
898 /* tree root */
899 ASSERT(btrfs_root_bytenr(&root->root_item) ==
900 cur->bytenr);
901 if (should_ignore_root(root))
902 list_add(&cur->list, &useless);
903 else
904 cur->root = root;
905 break;
906 }
907
908 level = cur->level + 1;
909
910 /*
911 * searching the tree to find upper level blocks
912 * reference the block.
913 */
914 path2->search_commit_root = 1;
915 path2->skip_locking = 1;
916 path2->lowest_level = level;
917 ret = btrfs_search_slot(NULL, root, node_key, path2, 0, 0);
918 path2->lowest_level = 0;
919 if (ret < 0) {
920 err = ret;
921 goto out;
922 }
923 if (ret > 0 && path2->slots[level] > 0)
924 path2->slots[level]--;
925
926 eb = path2->nodes[level];
927 if (btrfs_node_blockptr(eb, path2->slots[level]) !=
928 cur->bytenr) {
929 btrfs_err(root->fs_info,
930 "couldn't find block (%llu) (level %d) in tree (%llu) with key (%llu %u %llu)",
931 cur->bytenr, level - 1, root->objectid,
932 node_key->objectid, node_key->type,
933 node_key->offset);
934 err = -ENOENT;
935 goto out;
936 }
937 lower = cur;
938 need_check = true;
939 for (; level < BTRFS_MAX_LEVEL; level++) {
940 if (!path2->nodes[level]) {
941 ASSERT(btrfs_root_bytenr(&root->root_item) ==
942 lower->bytenr);
943 if (should_ignore_root(root))
944 list_add(&lower->list, &useless);
945 else
946 lower->root = root;
947 break;
948 }
949
950 edge = alloc_backref_edge(cache);
951 if (!edge) {
952 err = -ENOMEM;
953 goto out;
954 }
955
956 eb = path2->nodes[level];
957 rb_node = tree_search(&cache->rb_root, eb->start);
958 if (!rb_node) {
959 upper = alloc_backref_node(cache);
960 if (!upper) {
961 free_backref_edge(cache, edge);
962 err = -ENOMEM;
963 goto out;
964 }
965 upper->bytenr = eb->start;
966 upper->owner = btrfs_header_owner(eb);
967 upper->level = lower->level + 1;
968 if (!test_bit(BTRFS_ROOT_REF_COWS,
969 &root->state))
970 upper->cowonly = 1;
971
972 /*
973 * if we know the block isn't shared
974 * we can void checking its backrefs.
975 */
976 if (btrfs_block_can_be_shared(root, eb))
977 upper->checked = 0;
978 else
979 upper->checked = 1;
980
981 /*
982 * add the block to pending list if we
983 * need check its backrefs, we only do this once
984 * while walking up a tree as we will catch
985 * anything else later on.
986 */
987 if (!upper->checked && need_check) {
988 need_check = false;
989 list_add_tail(&edge->list[UPPER],
990 &list);
991 } else {
992 if (upper->checked)
993 need_check = true;
994 INIT_LIST_HEAD(&edge->list[UPPER]);
995 }
996 } else {
997 upper = rb_entry(rb_node, struct backref_node,
998 rb_node);
999 ASSERT(upper->checked);
1000 INIT_LIST_HEAD(&edge->list[UPPER]);
1001 if (!upper->owner)
1002 upper->owner = btrfs_header_owner(eb);
1003 }
1004 list_add_tail(&edge->list[LOWER], &lower->upper);
1005 edge->node[LOWER] = lower;
1006 edge->node[UPPER] = upper;
1007
1008 if (rb_node)
1009 break;
1010 lower = upper;
1011 upper = NULL;
1012 }
1013 btrfs_release_path(path2);
1014 next:
1015 if (ptr < end) {
1016 ptr += btrfs_extent_inline_ref_size(key.type);
1017 if (ptr >= end) {
1018 WARN_ON(ptr > end);
1019 ptr = 0;
1020 end = 0;
1021 }
1022 }
1023 if (ptr >= end)
1024 path1->slots[0]++;
1025 }
1026 btrfs_release_path(path1);
1027
1028 cur->checked = 1;
1029 WARN_ON(exist);
1030
1031 /* the pending list isn't empty, take the first block to process */
1032 if (!list_empty(&list)) {
1033 edge = list_entry(list.next, struct backref_edge, list[UPPER]);
1034 list_del_init(&edge->list[UPPER]);
1035 cur = edge->node[UPPER];
1036 goto again;
1037 }
1038
1039 /*
1040 * everything goes well, connect backref nodes and insert backref nodes
1041 * into the cache.
1042 */
1043 ASSERT(node->checked);
1044 cowonly = node->cowonly;
1045 if (!cowonly) {
1046 rb_node = tree_insert(&cache->rb_root, node->bytenr,
1047 &node->rb_node);
1048 if (rb_node)
1049 backref_tree_panic(rb_node, -EEXIST, node->bytenr);
1050 list_add_tail(&node->lower, &cache->leaves);
1051 }
1052
1053 list_for_each_entry(edge, &node->upper, list[LOWER])
1054 list_add_tail(&edge->list[UPPER], &list);
1055
1056 while (!list_empty(&list)) {
1057 edge = list_entry(list.next, struct backref_edge, list[UPPER]);
1058 list_del_init(&edge->list[UPPER]);
1059 upper = edge->node[UPPER];
1060 if (upper->detached) {
1061 list_del(&edge->list[LOWER]);
1062 lower = edge->node[LOWER];
1063 free_backref_edge(cache, edge);
1064 if (list_empty(&lower->upper))
1065 list_add(&lower->list, &useless);
1066 continue;
1067 }
1068
1069 if (!RB_EMPTY_NODE(&upper->rb_node)) {
1070 if (upper->lowest) {
1071 list_del_init(&upper->lower);
1072 upper->lowest = 0;
1073 }
1074
1075 list_add_tail(&edge->list[UPPER], &upper->lower);
1076 continue;
1077 }
1078
1079 if (!upper->checked) {
1080 /*
1081 * Still want to blow up for developers since this is a
1082 * logic bug.
1083 */
1084 ASSERT(0);
1085 err = -EINVAL;
1086 goto out;
1087 }
1088 if (cowonly != upper->cowonly) {
1089 ASSERT(0);
1090 err = -EINVAL;
1091 goto out;
1092 }
1093
1094 if (!cowonly) {
1095 rb_node = tree_insert(&cache->rb_root, upper->bytenr,
1096 &upper->rb_node);
1097 if (rb_node)
1098 backref_tree_panic(rb_node, -EEXIST,
1099 upper->bytenr);
1100 }
1101
1102 list_add_tail(&edge->list[UPPER], &upper->lower);
1103
1104 list_for_each_entry(edge, &upper->upper, list[LOWER])
1105 list_add_tail(&edge->list[UPPER], &list);
1106 }
1107 /*
1108 * process useless backref nodes. backref nodes for tree leaves
1109 * are deleted from the cache. backref nodes for upper level
1110 * tree blocks are left in the cache to avoid unnecessary backref
1111 * lookup.
1112 */
1113 while (!list_empty(&useless)) {
1114 upper = list_entry(useless.next, struct backref_node, list);
1115 list_del_init(&upper->list);
1116 ASSERT(list_empty(&upper->upper));
1117 if (upper == node)
1118 node = NULL;
1119 if (upper->lowest) {
1120 list_del_init(&upper->lower);
1121 upper->lowest = 0;
1122 }
1123 while (!list_empty(&upper->lower)) {
1124 edge = list_entry(upper->lower.next,
1125 struct backref_edge, list[UPPER]);
1126 list_del(&edge->list[UPPER]);
1127 list_del(&edge->list[LOWER]);
1128 lower = edge->node[LOWER];
1129 free_backref_edge(cache, edge);
1130
1131 if (list_empty(&lower->upper))
1132 list_add(&lower->list, &useless);
1133 }
1134 __mark_block_processed(rc, upper);
1135 if (upper->level > 0) {
1136 list_add(&upper->list, &cache->detached);
1137 upper->detached = 1;
1138 } else {
1139 rb_erase(&upper->rb_node, &cache->rb_root);
1140 free_backref_node(cache, upper);
1141 }
1142 }
1143 out:
1144 btrfs_free_path(path1);
1145 btrfs_free_path(path2);
1146 if (err) {
1147 while (!list_empty(&useless)) {
1148 lower = list_entry(useless.next,
1149 struct backref_node, list);
1150 list_del_init(&lower->list);
1151 }
1152 while (!list_empty(&list)) {
1153 edge = list_first_entry(&list, struct backref_edge,
1154 list[UPPER]);
1155 list_del(&edge->list[UPPER]);
1156 list_del(&edge->list[LOWER]);
1157 lower = edge->node[LOWER];
1158 upper = edge->node[UPPER];
1159 free_backref_edge(cache, edge);
1160
1161 /*
1162 * Lower is no longer linked to any upper backref nodes
1163 * and isn't in the cache, we can free it ourselves.
1164 */
1165 if (list_empty(&lower->upper) &&
1166 RB_EMPTY_NODE(&lower->rb_node))
1167 list_add(&lower->list, &useless);
1168
1169 if (!RB_EMPTY_NODE(&upper->rb_node))
1170 continue;
1171
1172 /* Add this guy's upper edges to the list to process */
1173 list_for_each_entry(edge, &upper->upper, list[LOWER])
1174 list_add_tail(&edge->list[UPPER], &list);
1175 if (list_empty(&upper->upper))
1176 list_add(&upper->list, &useless);
1177 }
1178
1179 while (!list_empty(&useless)) {
1180 lower = list_entry(useless.next,
1181 struct backref_node, list);
1182 list_del_init(&lower->list);
1183 if (lower == node)
1184 node = NULL;
1185 free_backref_node(cache, lower);
1186 }
1187
1188 free_backref_node(cache, node);
1189 return ERR_PTR(err);
1190 }
1191 ASSERT(!node || !node->detached);
1192 return node;
1193 }
1194
1195 /*
1196 * helper to add backref node for the newly created snapshot.
1197 * the backref node is created by cloning backref node that
1198 * corresponds to root of source tree
1199 */
1200 static int clone_backref_node(struct btrfs_trans_handle *trans,
1201 struct reloc_control *rc,
1202 struct btrfs_root *src,
1203 struct btrfs_root *dest)
1204 {
1205 struct btrfs_root *reloc_root = src->reloc_root;
1206 struct backref_cache *cache = &rc->backref_cache;
1207 struct backref_node *node = NULL;
1208 struct backref_node *new_node;
1209 struct backref_edge *edge;
1210 struct backref_edge *new_edge;
1211 struct rb_node *rb_node;
1212
1213 if (cache->last_trans > 0)
1214 update_backref_cache(trans, cache);
1215
1216 rb_node = tree_search(&cache->rb_root, src->commit_root->start);
1217 if (rb_node) {
1218 node = rb_entry(rb_node, struct backref_node, rb_node);
1219 if (node->detached)
1220 node = NULL;
1221 else
1222 BUG_ON(node->new_bytenr != reloc_root->node->start);
1223 }
1224
1225 if (!node) {
1226 rb_node = tree_search(&cache->rb_root,
1227 reloc_root->commit_root->start);
1228 if (rb_node) {
1229 node = rb_entry(rb_node, struct backref_node,
1230 rb_node);
1231 BUG_ON(node->detached);
1232 }
1233 }
1234
1235 if (!node)
1236 return 0;
1237
1238 new_node = alloc_backref_node(cache);
1239 if (!new_node)
1240 return -ENOMEM;
1241
1242 new_node->bytenr = dest->node->start;
1243 new_node->level = node->level;
1244 new_node->lowest = node->lowest;
1245 new_node->checked = 1;
1246 new_node->root = dest;
1247
1248 if (!node->lowest) {
1249 list_for_each_entry(edge, &node->lower, list[UPPER]) {
1250 new_edge = alloc_backref_edge(cache);
1251 if (!new_edge)
1252 goto fail;
1253
1254 new_edge->node[UPPER] = new_node;
1255 new_edge->node[LOWER] = edge->node[LOWER];
1256 list_add_tail(&new_edge->list[UPPER],
1257 &new_node->lower);
1258 }
1259 } else {
1260 list_add_tail(&new_node->lower, &cache->leaves);
1261 }
1262
1263 rb_node = tree_insert(&cache->rb_root, new_node->bytenr,
1264 &new_node->rb_node);
1265 if (rb_node)
1266 backref_tree_panic(rb_node, -EEXIST, new_node->bytenr);
1267
1268 if (!new_node->lowest) {
1269 list_for_each_entry(new_edge, &new_node->lower, list[UPPER]) {
1270 list_add_tail(&new_edge->list[LOWER],
1271 &new_edge->node[LOWER]->upper);
1272 }
1273 }
1274 return 0;
1275 fail:
1276 while (!list_empty(&new_node->lower)) {
1277 new_edge = list_entry(new_node->lower.next,
1278 struct backref_edge, list[UPPER]);
1279 list_del(&new_edge->list[UPPER]);
1280 free_backref_edge(cache, new_edge);
1281 }
1282 free_backref_node(cache, new_node);
1283 return -ENOMEM;
1284 }
1285
1286 /*
1287 * helper to add 'address of tree root -> reloc tree' mapping
1288 */
1289 static int __must_check __add_reloc_root(struct btrfs_root *root)
1290 {
1291 struct btrfs_fs_info *fs_info = root->fs_info;
1292 struct rb_node *rb_node;
1293 struct mapping_node *node;
1294 struct reloc_control *rc = fs_info->reloc_ctl;
1295
1296 node = kmalloc(sizeof(*node), GFP_NOFS);
1297 if (!node)
1298 return -ENOMEM;
1299
1300 node->bytenr = root->node->start;
1301 node->data = root;
1302
1303 spin_lock(&rc->reloc_root_tree.lock);
1304 rb_node = tree_insert(&rc->reloc_root_tree.rb_root,
1305 node->bytenr, &node->rb_node);
1306 spin_unlock(&rc->reloc_root_tree.lock);
1307 if (rb_node) {
1308 btrfs_panic(fs_info, -EEXIST,
1309 "Duplicate root found for start=%llu while inserting into relocation tree",
1310 node->bytenr);
1311 kfree(node);
1312 return -EEXIST;
1313 }
1314
1315 list_add_tail(&root->root_list, &rc->reloc_roots);
1316 return 0;
1317 }
1318
1319 /*
1320 * helper to delete the 'address of tree root -> reloc tree'
1321 * mapping
1322 */
1323 static void __del_reloc_root(struct btrfs_root *root)
1324 {
1325 struct btrfs_fs_info *fs_info = root->fs_info;
1326 struct rb_node *rb_node;
1327 struct mapping_node *node = NULL;
1328 struct reloc_control *rc = fs_info->reloc_ctl;
1329
1330 spin_lock(&rc->reloc_root_tree.lock);
1331 rb_node = tree_search(&rc->reloc_root_tree.rb_root,
1332 root->node->start);
1333 if (rb_node) {
1334 node = rb_entry(rb_node, struct mapping_node, rb_node);
1335 rb_erase(&node->rb_node, &rc->reloc_root_tree.rb_root);
1336 }
1337 spin_unlock(&rc->reloc_root_tree.lock);
1338
1339 if (!node)
1340 return;
1341 BUG_ON((struct btrfs_root *)node->data != root);
1342
1343 spin_lock(&fs_info->trans_lock);
1344 list_del_init(&root->root_list);
1345 spin_unlock(&fs_info->trans_lock);
1346 kfree(node);
1347 }
1348
1349 /*
1350 * helper to update the 'address of tree root -> reloc tree'
1351 * mapping
1352 */
1353 static int __update_reloc_root(struct btrfs_root *root, u64 new_bytenr)
1354 {
1355 struct btrfs_fs_info *fs_info = root->fs_info;
1356 struct rb_node *rb_node;
1357 struct mapping_node *node = NULL;
1358 struct reloc_control *rc = fs_info->reloc_ctl;
1359
1360 spin_lock(&rc->reloc_root_tree.lock);
1361 rb_node = tree_search(&rc->reloc_root_tree.rb_root,
1362 root->node->start);
1363 if (rb_node) {
1364 node = rb_entry(rb_node, struct mapping_node, rb_node);
1365 rb_erase(&node->rb_node, &rc->reloc_root_tree.rb_root);
1366 }
1367 spin_unlock(&rc->reloc_root_tree.lock);
1368
1369 if (!node)
1370 return 0;
1371 BUG_ON((struct btrfs_root *)node->data != root);
1372
1373 spin_lock(&rc->reloc_root_tree.lock);
1374 node->bytenr = new_bytenr;
1375 rb_node = tree_insert(&rc->reloc_root_tree.rb_root,
1376 node->bytenr, &node->rb_node);
1377 spin_unlock(&rc->reloc_root_tree.lock);
1378 if (rb_node)
1379 backref_tree_panic(rb_node, -EEXIST, node->bytenr);
1380 return 0;
1381 }
1382
1383 static struct btrfs_root *create_reloc_root(struct btrfs_trans_handle *trans,
1384 struct btrfs_root *root, u64 objectid)
1385 {
1386 struct btrfs_fs_info *fs_info = root->fs_info;
1387 struct btrfs_root *reloc_root;
1388 struct extent_buffer *eb;
1389 struct btrfs_root_item *root_item;
1390 struct btrfs_key root_key;
1391 u64 last_snap = 0;
1392 int ret;
1393
1394 root_item = kmalloc(sizeof(*root_item), GFP_NOFS);
1395 BUG_ON(!root_item);
1396
1397 root_key.objectid = BTRFS_TREE_RELOC_OBJECTID;
1398 root_key.type = BTRFS_ROOT_ITEM_KEY;
1399 root_key.offset = objectid;
1400
1401 if (root->root_key.objectid == objectid) {
1402 /* called by btrfs_init_reloc_root */
1403 ret = btrfs_copy_root(trans, root, root->commit_root, &eb,
1404 BTRFS_TREE_RELOC_OBJECTID);
1405 BUG_ON(ret);
1406
1407 last_snap = btrfs_root_last_snapshot(&root->root_item);
1408 btrfs_set_root_last_snapshot(&root->root_item,
1409 trans->transid - 1);
1410 } else {
1411 /*
1412 * called by btrfs_reloc_post_snapshot_hook.
1413 * the source tree is a reloc tree, all tree blocks
1414 * modified after it was created have RELOC flag
1415 * set in their headers. so it's OK to not update
1416 * the 'last_snapshot'.
1417 */
1418 ret = btrfs_copy_root(trans, root, root->node, &eb,
1419 BTRFS_TREE_RELOC_OBJECTID);
1420 BUG_ON(ret);
1421 }
1422
1423 memcpy(root_item, &root->root_item, sizeof(*root_item));
1424 btrfs_set_root_bytenr(root_item, eb->start);
1425 btrfs_set_root_level(root_item, btrfs_header_level(eb));
1426 btrfs_set_root_generation(root_item, trans->transid);
1427
1428 if (root->root_key.objectid == objectid) {
1429 btrfs_set_root_refs(root_item, 0);
1430 memset(&root_item->drop_progress, 0,
1431 sizeof(struct btrfs_disk_key));
1432 root_item->drop_level = 0;
1433 /*
1434 * abuse rtransid, it is safe because it is impossible to
1435 * receive data into a relocation tree.
1436 */
1437 btrfs_set_root_rtransid(root_item, last_snap);
1438 btrfs_set_root_otransid(root_item, trans->transid);
1439 }
1440
1441 btrfs_tree_unlock(eb);
1442 free_extent_buffer(eb);
1443
1444 ret = btrfs_insert_root(trans, fs_info->tree_root,
1445 &root_key, root_item);
1446 BUG_ON(ret);
1447 kfree(root_item);
1448
1449 reloc_root = btrfs_read_fs_root(fs_info->tree_root, &root_key);
1450 BUG_ON(IS_ERR(reloc_root));
1451 reloc_root->last_trans = trans->transid;
1452 return reloc_root;
1453 }
1454
1455 /*
1456 * create reloc tree for a given fs tree. reloc tree is just a
1457 * snapshot of the fs tree with special root objectid.
1458 */
1459 int btrfs_init_reloc_root(struct btrfs_trans_handle *trans,
1460 struct btrfs_root *root)
1461 {
1462 struct btrfs_fs_info *fs_info = root->fs_info;
1463 struct btrfs_root *reloc_root;
1464 struct reloc_control *rc = fs_info->reloc_ctl;
1465 struct btrfs_block_rsv *rsv;
1466 int clear_rsv = 0;
1467 int ret;
1468
1469 if (root->reloc_root) {
1470 reloc_root = root->reloc_root;
1471 reloc_root->last_trans = trans->transid;
1472 return 0;
1473 }
1474
1475 if (!rc || !rc->create_reloc_tree ||
1476 root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID)
1477 return 0;
1478
1479 if (!trans->reloc_reserved) {
1480 rsv = trans->block_rsv;
1481 trans->block_rsv = rc->block_rsv;
1482 clear_rsv = 1;
1483 }
1484 reloc_root = create_reloc_root(trans, root, root->root_key.objectid);
1485 if (clear_rsv)
1486 trans->block_rsv = rsv;
1487
1488 ret = __add_reloc_root(reloc_root);
1489 BUG_ON(ret < 0);
1490 root->reloc_root = reloc_root;
1491 return 0;
1492 }
1493
1494 /*
1495 * update root item of reloc tree
1496 */
1497 int btrfs_update_reloc_root(struct btrfs_trans_handle *trans,
1498 struct btrfs_root *root)
1499 {
1500 struct btrfs_fs_info *fs_info = root->fs_info;
1501 struct btrfs_root *reloc_root;
1502 struct btrfs_root_item *root_item;
1503 int ret;
1504
1505 if (!root->reloc_root)
1506 goto out;
1507
1508 reloc_root = root->reloc_root;
1509 root_item = &reloc_root->root_item;
1510
1511 if (fs_info->reloc_ctl->merge_reloc_tree &&
1512 btrfs_root_refs(root_item) == 0) {
1513 root->reloc_root = NULL;
1514 __del_reloc_root(reloc_root);
1515 }
1516
1517 if (reloc_root->commit_root != reloc_root->node) {
1518 btrfs_set_root_node(root_item, reloc_root->node);
1519 free_extent_buffer(reloc_root->commit_root);
1520 reloc_root->commit_root = btrfs_root_node(reloc_root);
1521 }
1522
1523 ret = btrfs_update_root(trans, fs_info->tree_root,
1524 &reloc_root->root_key, root_item);
1525 BUG_ON(ret);
1526
1527 out:
1528 return 0;
1529 }
1530
1531 /*
1532 * helper to find first cached inode with inode number >= objectid
1533 * in a subvolume
1534 */
1535 static struct inode *find_next_inode(struct btrfs_root *root, u64 objectid)
1536 {
1537 struct rb_node *node;
1538 struct rb_node *prev;
1539 struct btrfs_inode *entry;
1540 struct inode *inode;
1541
1542 spin_lock(&root->inode_lock);
1543 again:
1544 node = root->inode_tree.rb_node;
1545 prev = NULL;
1546 while (node) {
1547 prev = node;
1548 entry = rb_entry(node, struct btrfs_inode, rb_node);
1549
1550 if (objectid < btrfs_ino(&entry->vfs_inode))
1551 node = node->rb_left;
1552 else if (objectid > btrfs_ino(&entry->vfs_inode))
1553 node = node->rb_right;
1554 else
1555 break;
1556 }
1557 if (!node) {
1558 while (prev) {
1559 entry = rb_entry(prev, struct btrfs_inode, rb_node);
1560 if (objectid <= btrfs_ino(&entry->vfs_inode)) {
1561 node = prev;
1562 break;
1563 }
1564 prev = rb_next(prev);
1565 }
1566 }
1567 while (node) {
1568 entry = rb_entry(node, struct btrfs_inode, rb_node);
1569 inode = igrab(&entry->vfs_inode);
1570 if (inode) {
1571 spin_unlock(&root->inode_lock);
1572 return inode;
1573 }
1574
1575 objectid = btrfs_ino(&entry->vfs_inode) + 1;
1576 if (cond_resched_lock(&root->inode_lock))
1577 goto again;
1578
1579 node = rb_next(node);
1580 }
1581 spin_unlock(&root->inode_lock);
1582 return NULL;
1583 }
1584
1585 static int in_block_group(u64 bytenr,
1586 struct btrfs_block_group_cache *block_group)
1587 {
1588 if (bytenr >= block_group->key.objectid &&
1589 bytenr < block_group->key.objectid + block_group->key.offset)
1590 return 1;
1591 return 0;
1592 }
1593
1594 /*
1595 * get new location of data
1596 */
1597 static int get_new_location(struct inode *reloc_inode, u64 *new_bytenr,
1598 u64 bytenr, u64 num_bytes)
1599 {
1600 struct btrfs_root *root = BTRFS_I(reloc_inode)->root;
1601 struct btrfs_path *path;
1602 struct btrfs_file_extent_item *fi;
1603 struct extent_buffer *leaf;
1604 int ret;
1605
1606 path = btrfs_alloc_path();
1607 if (!path)
1608 return -ENOMEM;
1609
1610 bytenr -= BTRFS_I(reloc_inode)->index_cnt;
1611 ret = btrfs_lookup_file_extent(NULL, root, path, btrfs_ino(reloc_inode),
1612 bytenr, 0);
1613 if (ret < 0)
1614 goto out;
1615 if (ret > 0) {
1616 ret = -ENOENT;
1617 goto out;
1618 }
1619
1620 leaf = path->nodes[0];
1621 fi = btrfs_item_ptr(leaf, path->slots[0],
1622 struct btrfs_file_extent_item);
1623
1624 BUG_ON(btrfs_file_extent_offset(leaf, fi) ||
1625 btrfs_file_extent_compression(leaf, fi) ||
1626 btrfs_file_extent_encryption(leaf, fi) ||
1627 btrfs_file_extent_other_encoding(leaf, fi));
1628
1629 if (num_bytes != btrfs_file_extent_disk_num_bytes(leaf, fi)) {
1630 ret = -EINVAL;
1631 goto out;
1632 }
1633
1634 *new_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1635 ret = 0;
1636 out:
1637 btrfs_free_path(path);
1638 return ret;
1639 }
1640
1641 /*
1642 * update file extent items in the tree leaf to point to
1643 * the new locations.
1644 */
1645 static noinline_for_stack
1646 int replace_file_extents(struct btrfs_trans_handle *trans,
1647 struct reloc_control *rc,
1648 struct btrfs_root *root,
1649 struct extent_buffer *leaf)
1650 {
1651 struct btrfs_fs_info *fs_info = root->fs_info;
1652 struct btrfs_key key;
1653 struct btrfs_file_extent_item *fi;
1654 struct inode *inode = NULL;
1655 u64 parent;
1656 u64 bytenr;
1657 u64 new_bytenr = 0;
1658 u64 num_bytes;
1659 u64 end;
1660 u32 nritems;
1661 u32 i;
1662 int ret = 0;
1663 int first = 1;
1664 int dirty = 0;
1665
1666 if (rc->stage != UPDATE_DATA_PTRS)
1667 return 0;
1668
1669 /* reloc trees always use full backref */
1670 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID)
1671 parent = leaf->start;
1672 else
1673 parent = 0;
1674
1675 nritems = btrfs_header_nritems(leaf);
1676 for (i = 0; i < nritems; i++) {
1677 cond_resched();
1678 btrfs_item_key_to_cpu(leaf, &key, i);
1679 if (key.type != BTRFS_EXTENT_DATA_KEY)
1680 continue;
1681 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
1682 if (btrfs_file_extent_type(leaf, fi) ==
1683 BTRFS_FILE_EXTENT_INLINE)
1684 continue;
1685 bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1686 num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
1687 if (bytenr == 0)
1688 continue;
1689 if (!in_block_group(bytenr, rc->block_group))
1690 continue;
1691
1692 /*
1693 * if we are modifying block in fs tree, wait for readpage
1694 * to complete and drop the extent cache
1695 */
1696 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
1697 if (first) {
1698 inode = find_next_inode(root, key.objectid);
1699 first = 0;
1700 } else if (inode && btrfs_ino(inode) < key.objectid) {
1701 btrfs_add_delayed_iput(inode);
1702 inode = find_next_inode(root, key.objectid);
1703 }
1704 if (inode && btrfs_ino(inode) == key.objectid) {
1705 end = key.offset +
1706 btrfs_file_extent_num_bytes(leaf, fi);
1707 WARN_ON(!IS_ALIGNED(key.offset,
1708 fs_info->sectorsize));
1709 WARN_ON(!IS_ALIGNED(end, fs_info->sectorsize));
1710 end--;
1711 ret = try_lock_extent(&BTRFS_I(inode)->io_tree,
1712 key.offset, end);
1713 if (!ret)
1714 continue;
1715
1716 btrfs_drop_extent_cache(inode, key.offset, end,
1717 1);
1718 unlock_extent(&BTRFS_I(inode)->io_tree,
1719 key.offset, end);
1720 }
1721 }
1722
1723 ret = get_new_location(rc->data_inode, &new_bytenr,
1724 bytenr, num_bytes);
1725 if (ret) {
1726 /*
1727 * Don't have to abort since we've not changed anything
1728 * in the file extent yet.
1729 */
1730 break;
1731 }
1732
1733 btrfs_set_file_extent_disk_bytenr(leaf, fi, new_bytenr);
1734 dirty = 1;
1735
1736 key.offset -= btrfs_file_extent_offset(leaf, fi);
1737 ret = btrfs_inc_extent_ref(trans, root, new_bytenr,
1738 num_bytes, parent,
1739 btrfs_header_owner(leaf),
1740 key.objectid, key.offset);
1741 if (ret) {
1742 btrfs_abort_transaction(trans, ret);
1743 break;
1744 }
1745
1746 ret = btrfs_free_extent(trans, root, bytenr, num_bytes,
1747 parent, btrfs_header_owner(leaf),
1748 key.objectid, key.offset);
1749 if (ret) {
1750 btrfs_abort_transaction(trans, ret);
1751 break;
1752 }
1753 }
1754 if (dirty)
1755 btrfs_mark_buffer_dirty(leaf);
1756 if (inode)
1757 btrfs_add_delayed_iput(inode);
1758 return ret;
1759 }
1760
1761 static noinline_for_stack
1762 int memcmp_node_keys(struct extent_buffer *eb, int slot,
1763 struct btrfs_path *path, int level)
1764 {
1765 struct btrfs_disk_key key1;
1766 struct btrfs_disk_key key2;
1767 btrfs_node_key(eb, &key1, slot);
1768 btrfs_node_key(path->nodes[level], &key2, path->slots[level]);
1769 return memcmp(&key1, &key2, sizeof(key1));
1770 }
1771
1772 /*
1773 * try to replace tree blocks in fs tree with the new blocks
1774 * in reloc tree. tree blocks haven't been modified since the
1775 * reloc tree was create can be replaced.
1776 *
1777 * if a block was replaced, level of the block + 1 is returned.
1778 * if no block got replaced, 0 is returned. if there are other
1779 * errors, a negative error number is returned.
1780 */
1781 static noinline_for_stack
1782 int replace_path(struct btrfs_trans_handle *trans,
1783 struct btrfs_root *dest, struct btrfs_root *src,
1784 struct btrfs_path *path, struct btrfs_key *next_key,
1785 int lowest_level, int max_level)
1786 {
1787 struct btrfs_fs_info *fs_info = dest->fs_info;
1788 struct extent_buffer *eb;
1789 struct extent_buffer *parent;
1790 struct btrfs_key key;
1791 u64 old_bytenr;
1792 u64 new_bytenr;
1793 u64 old_ptr_gen;
1794 u64 new_ptr_gen;
1795 u64 last_snapshot;
1796 u32 blocksize;
1797 int cow = 0;
1798 int level;
1799 int ret;
1800 int slot;
1801
1802 BUG_ON(src->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
1803 BUG_ON(dest->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID);
1804
1805 last_snapshot = btrfs_root_last_snapshot(&src->root_item);
1806 again:
1807 slot = path->slots[lowest_level];
1808 btrfs_node_key_to_cpu(path->nodes[lowest_level], &key, slot);
1809
1810 eb = btrfs_lock_root_node(dest);
1811 btrfs_set_lock_blocking(eb);
1812 level = btrfs_header_level(eb);
1813
1814 if (level < lowest_level) {
1815 btrfs_tree_unlock(eb);
1816 free_extent_buffer(eb);
1817 return 0;
1818 }
1819
1820 if (cow) {
1821 ret = btrfs_cow_block(trans, dest, eb, NULL, 0, &eb);
1822 BUG_ON(ret);
1823 }
1824 btrfs_set_lock_blocking(eb);
1825
1826 if (next_key) {
1827 next_key->objectid = (u64)-1;
1828 next_key->type = (u8)-1;
1829 next_key->offset = (u64)-1;
1830 }
1831
1832 parent = eb;
1833 while (1) {
1834 level = btrfs_header_level(parent);
1835 BUG_ON(level < lowest_level);
1836
1837 ret = btrfs_bin_search(parent, &key, level, &slot);
1838 if (ret && slot > 0)
1839 slot--;
1840
1841 if (next_key && slot + 1 < btrfs_header_nritems(parent))
1842 btrfs_node_key_to_cpu(parent, next_key, slot + 1);
1843
1844 old_bytenr = btrfs_node_blockptr(parent, slot);
1845 blocksize = fs_info->nodesize;
1846 old_ptr_gen = btrfs_node_ptr_generation(parent, slot);
1847
1848 if (level <= max_level) {
1849 eb = path->nodes[level];
1850 new_bytenr = btrfs_node_blockptr(eb,
1851 path->slots[level]);
1852 new_ptr_gen = btrfs_node_ptr_generation(eb,
1853 path->slots[level]);
1854 } else {
1855 new_bytenr = 0;
1856 new_ptr_gen = 0;
1857 }
1858
1859 if (WARN_ON(new_bytenr > 0 && new_bytenr == old_bytenr)) {
1860 ret = level;
1861 break;
1862 }
1863
1864 if (new_bytenr == 0 || old_ptr_gen > last_snapshot ||
1865 memcmp_node_keys(parent, slot, path, level)) {
1866 if (level <= lowest_level) {
1867 ret = 0;
1868 break;
1869 }
1870
1871 eb = read_tree_block(dest, old_bytenr, old_ptr_gen);
1872 if (IS_ERR(eb)) {
1873 ret = PTR_ERR(eb);
1874 break;
1875 } else if (!extent_buffer_uptodate(eb)) {
1876 ret = -EIO;
1877 free_extent_buffer(eb);
1878 break;
1879 }
1880 btrfs_tree_lock(eb);
1881 if (cow) {
1882 ret = btrfs_cow_block(trans, dest, eb, parent,
1883 slot, &eb);
1884 BUG_ON(ret);
1885 }
1886 btrfs_set_lock_blocking(eb);
1887
1888 btrfs_tree_unlock(parent);
1889 free_extent_buffer(parent);
1890
1891 parent = eb;
1892 continue;
1893 }
1894
1895 if (!cow) {
1896 btrfs_tree_unlock(parent);
1897 free_extent_buffer(parent);
1898 cow = 1;
1899 goto again;
1900 }
1901
1902 btrfs_node_key_to_cpu(path->nodes[level], &key,
1903 path->slots[level]);
1904 btrfs_release_path(path);
1905
1906 path->lowest_level = level;
1907 ret = btrfs_search_slot(trans, src, &key, path, 0, 1);
1908 path->lowest_level = 0;
1909 BUG_ON(ret);
1910
1911 /*
1912 * Info qgroup to trace both subtrees.
1913 *
1914 * We must trace both trees.
1915 * 1) Tree reloc subtree
1916 * If not traced, we will leak data numbers
1917 * 2) Fs subtree
1918 * If not traced, we will double count old data
1919 * and tree block numbers, if current trans doesn't free
1920 * data reloc tree inode.
1921 */
1922 ret = btrfs_qgroup_trace_subtree(trans, src, parent,
1923 btrfs_header_generation(parent),
1924 btrfs_header_level(parent));
1925 if (ret < 0)
1926 break;
1927 ret = btrfs_qgroup_trace_subtree(trans, dest,
1928 path->nodes[level],
1929 btrfs_header_generation(path->nodes[level]),
1930 btrfs_header_level(path->nodes[level]));
1931 if (ret < 0)
1932 break;
1933
1934 /*
1935 * swap blocks in fs tree and reloc tree.
1936 */
1937 btrfs_set_node_blockptr(parent, slot, new_bytenr);
1938 btrfs_set_node_ptr_generation(parent, slot, new_ptr_gen);
1939 btrfs_mark_buffer_dirty(parent);
1940
1941 btrfs_set_node_blockptr(path->nodes[level],
1942 path->slots[level], old_bytenr);
1943 btrfs_set_node_ptr_generation(path->nodes[level],
1944 path->slots[level], old_ptr_gen);
1945 btrfs_mark_buffer_dirty(path->nodes[level]);
1946
1947 ret = btrfs_inc_extent_ref(trans, src, old_bytenr, blocksize,
1948 path->nodes[level]->start,
1949 src->root_key.objectid, level - 1, 0);
1950 BUG_ON(ret);
1951 ret = btrfs_inc_extent_ref(trans, dest, new_bytenr, blocksize,
1952 0, dest->root_key.objectid, level - 1,
1953 0);
1954 BUG_ON(ret);
1955
1956 ret = btrfs_free_extent(trans, src, new_bytenr, blocksize,
1957 path->nodes[level]->start,
1958 src->root_key.objectid, level - 1, 0);
1959 BUG_ON(ret);
1960
1961 ret = btrfs_free_extent(trans, dest, old_bytenr, blocksize,
1962 0, dest->root_key.objectid, level - 1,
1963 0);
1964 BUG_ON(ret);
1965
1966 btrfs_unlock_up_safe(path, 0);
1967
1968 ret = level;
1969 break;
1970 }
1971 btrfs_tree_unlock(parent);
1972 free_extent_buffer(parent);
1973 return ret;
1974 }
1975
1976 /*
1977 * helper to find next relocated block in reloc tree
1978 */
1979 static noinline_for_stack
1980 int walk_up_reloc_tree(struct btrfs_root *root, struct btrfs_path *path,
1981 int *level)
1982 {
1983 struct extent_buffer *eb;
1984 int i;
1985 u64 last_snapshot;
1986 u32 nritems;
1987
1988 last_snapshot = btrfs_root_last_snapshot(&root->root_item);
1989
1990 for (i = 0; i < *level; i++) {
1991 free_extent_buffer(path->nodes[i]);
1992 path->nodes[i] = NULL;
1993 }
1994
1995 for (i = *level; i < BTRFS_MAX_LEVEL && path->nodes[i]; i++) {
1996 eb = path->nodes[i];
1997 nritems = btrfs_header_nritems(eb);
1998 while (path->slots[i] + 1 < nritems) {
1999 path->slots[i]++;
2000 if (btrfs_node_ptr_generation(eb, path->slots[i]) <=
2001 last_snapshot)
2002 continue;
2003
2004 *level = i;
2005 return 0;
2006 }
2007 free_extent_buffer(path->nodes[i]);
2008 path->nodes[i] = NULL;
2009 }
2010 return 1;
2011 }
2012
2013 /*
2014 * walk down reloc tree to find relocated block of lowest level
2015 */
2016 static noinline_for_stack
2017 int walk_down_reloc_tree(struct btrfs_root *root, struct btrfs_path *path,
2018 int *level)
2019 {
2020 struct extent_buffer *eb = NULL;
2021 int i;
2022 u64 bytenr;
2023 u64 ptr_gen = 0;
2024 u64 last_snapshot;
2025 u32 nritems;
2026
2027 last_snapshot = btrfs_root_last_snapshot(&root->root_item);
2028
2029 for (i = *level; i > 0; i--) {
2030 eb = path->nodes[i];
2031 nritems = btrfs_header_nritems(eb);
2032 while (path->slots[i] < nritems) {
2033 ptr_gen = btrfs_node_ptr_generation(eb, path->slots[i]);
2034 if (ptr_gen > last_snapshot)
2035 break;
2036 path->slots[i]++;
2037 }
2038 if (path->slots[i] >= nritems) {
2039 if (i == *level)
2040 break;
2041 *level = i + 1;
2042 return 0;
2043 }
2044 if (i == 1) {
2045 *level = i;
2046 return 0;
2047 }
2048
2049 bytenr = btrfs_node_blockptr(eb, path->slots[i]);
2050 eb = read_tree_block(root, bytenr, ptr_gen);
2051 if (IS_ERR(eb)) {
2052 return PTR_ERR(eb);
2053 } else if (!extent_buffer_uptodate(eb)) {
2054 free_extent_buffer(eb);
2055 return -EIO;
2056 }
2057 BUG_ON(btrfs_header_level(eb) != i - 1);
2058 path->nodes[i - 1] = eb;
2059 path->slots[i - 1] = 0;
2060 }
2061 return 1;
2062 }
2063
2064 /*
2065 * invalidate extent cache for file extents whose key in range of
2066 * [min_key, max_key)
2067 */
2068 static int invalidate_extent_cache(struct btrfs_root *root,
2069 struct btrfs_key *min_key,
2070 struct btrfs_key *max_key)
2071 {
2072 struct btrfs_fs_info *fs_info = root->fs_info;
2073 struct inode *inode = NULL;
2074 u64 objectid;
2075 u64 start, end;
2076 u64 ino;
2077
2078 objectid = min_key->objectid;
2079 while (1) {
2080 cond_resched();
2081 iput(inode);
2082
2083 if (objectid > max_key->objectid)
2084 break;
2085
2086 inode = find_next_inode(root, objectid);
2087 if (!inode)
2088 break;
2089 ino = btrfs_ino(inode);
2090
2091 if (ino > max_key->objectid) {
2092 iput(inode);
2093 break;
2094 }
2095
2096 objectid = ino + 1;
2097 if (!S_ISREG(inode->i_mode))
2098 continue;
2099
2100 if (unlikely(min_key->objectid == ino)) {
2101 if (min_key->type > BTRFS_EXTENT_DATA_KEY)
2102 continue;
2103 if (min_key->type < BTRFS_EXTENT_DATA_KEY)
2104 start = 0;
2105 else {
2106 start = min_key->offset;
2107 WARN_ON(!IS_ALIGNED(start, fs_info->sectorsize));
2108 }
2109 } else {
2110 start = 0;
2111 }
2112
2113 if (unlikely(max_key->objectid == ino)) {
2114 if (max_key->type < BTRFS_EXTENT_DATA_KEY)
2115 continue;
2116 if (max_key->type > BTRFS_EXTENT_DATA_KEY) {
2117 end = (u64)-1;
2118 } else {
2119 if (max_key->offset == 0)
2120 continue;
2121 end = max_key->offset;
2122 WARN_ON(!IS_ALIGNED(end, fs_info->sectorsize));
2123 end--;
2124 }
2125 } else {
2126 end = (u64)-1;
2127 }
2128
2129 /* the lock_extent waits for readpage to complete */
2130 lock_extent(&BTRFS_I(inode)->io_tree, start, end);
2131 btrfs_drop_extent_cache(inode, start, end, 1);
2132 unlock_extent(&BTRFS_I(inode)->io_tree, start, end);
2133 }
2134 return 0;
2135 }
2136
2137 static int find_next_key(struct btrfs_path *path, int level,
2138 struct btrfs_key *key)
2139
2140 {
2141 while (level < BTRFS_MAX_LEVEL) {
2142 if (!path->nodes[level])
2143 break;
2144 if (path->slots[level] + 1 <
2145 btrfs_header_nritems(path->nodes[level])) {
2146 btrfs_node_key_to_cpu(path->nodes[level], key,
2147 path->slots[level] + 1);
2148 return 0;
2149 }
2150 level++;
2151 }
2152 return 1;
2153 }
2154
2155 /*
2156 * merge the relocated tree blocks in reloc tree with corresponding
2157 * fs tree.
2158 */
2159 static noinline_for_stack int merge_reloc_root(struct reloc_control *rc,
2160 struct btrfs_root *root)
2161 {
2162 struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
2163 LIST_HEAD(inode_list);
2164 struct btrfs_key key;
2165 struct btrfs_key next_key;
2166 struct btrfs_trans_handle *trans = NULL;
2167 struct btrfs_root *reloc_root;
2168 struct btrfs_root_item *root_item;
2169 struct btrfs_path *path;
2170 struct extent_buffer *leaf;
2171 int level;
2172 int max_level;
2173 int replaced = 0;
2174 int ret;
2175 int err = 0;
2176 u32 min_reserved;
2177
2178 path = btrfs_alloc_path();
2179 if (!path)
2180 return -ENOMEM;
2181 path->reada = READA_FORWARD;
2182
2183 reloc_root = root->reloc_root;
2184 root_item = &reloc_root->root_item;
2185
2186 if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
2187 level = btrfs_root_level(root_item);
2188 extent_buffer_get(reloc_root->node);
2189 path->nodes[level] = reloc_root->node;
2190 path->slots[level] = 0;
2191 } else {
2192 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
2193
2194 level = root_item->drop_level;
2195 BUG_ON(level == 0);
2196 path->lowest_level = level;
2197 ret = btrfs_search_slot(NULL, reloc_root, &key, path, 0, 0);
2198 path->lowest_level = 0;
2199 if (ret < 0) {
2200 btrfs_free_path(path);
2201 return ret;
2202 }
2203
2204 btrfs_node_key_to_cpu(path->nodes[level], &next_key,
2205 path->slots[level]);
2206 WARN_ON(memcmp(&key, &next_key, sizeof(key)));
2207
2208 btrfs_unlock_up_safe(path, 0);
2209 }
2210
2211 min_reserved = fs_info->nodesize * (BTRFS_MAX_LEVEL - 1) * 2;
2212 memset(&next_key, 0, sizeof(next_key));
2213
2214 while (1) {
2215 ret = btrfs_block_rsv_refill(root, rc->block_rsv, min_reserved,
2216 BTRFS_RESERVE_FLUSH_ALL);
2217 if (ret) {
2218 err = ret;
2219 goto out;
2220 }
2221 trans = btrfs_start_transaction(root, 0);
2222 if (IS_ERR(trans)) {
2223 err = PTR_ERR(trans);
2224 trans = NULL;
2225 goto out;
2226 }
2227 trans->block_rsv = rc->block_rsv;
2228
2229 replaced = 0;
2230 max_level = level;
2231
2232 ret = walk_down_reloc_tree(reloc_root, path, &level);
2233 if (ret < 0) {
2234 err = ret;
2235 goto out;
2236 }
2237 if (ret > 0)
2238 break;
2239
2240 if (!find_next_key(path, level, &key) &&
2241 btrfs_comp_cpu_keys(&next_key, &key) >= 0) {
2242 ret = 0;
2243 } else {
2244 ret = replace_path(trans, root, reloc_root, path,
2245 &next_key, level, max_level);
2246 }
2247 if (ret < 0) {
2248 err = ret;
2249 goto out;
2250 }
2251
2252 if (ret > 0) {
2253 level = ret;
2254 btrfs_node_key_to_cpu(path->nodes[level], &key,
2255 path->slots[level]);
2256 replaced = 1;
2257 }
2258
2259 ret = walk_up_reloc_tree(reloc_root, path, &level);
2260 if (ret > 0)
2261 break;
2262
2263 BUG_ON(level == 0);
2264 /*
2265 * save the merging progress in the drop_progress.
2266 * this is OK since root refs == 1 in this case.
2267 */
2268 btrfs_node_key(path->nodes[level], &root_item->drop_progress,
2269 path->slots[level]);
2270 root_item->drop_level = level;
2271
2272 btrfs_end_transaction_throttle(trans, root);
2273 trans = NULL;
2274
2275 btrfs_btree_balance_dirty(root);
2276
2277 if (replaced && rc->stage == UPDATE_DATA_PTRS)
2278 invalidate_extent_cache(root, &key, &next_key);
2279 }
2280
2281 /*
2282 * handle the case only one block in the fs tree need to be
2283 * relocated and the block is tree root.
2284 */
2285 leaf = btrfs_lock_root_node(root);
2286 ret = btrfs_cow_block(trans, root, leaf, NULL, 0, &leaf);
2287 btrfs_tree_unlock(leaf);
2288 free_extent_buffer(leaf);
2289 if (ret < 0)
2290 err = ret;
2291 out:
2292 btrfs_free_path(path);
2293
2294 if (err == 0) {
2295 memset(&root_item->drop_progress, 0,
2296 sizeof(root_item->drop_progress));
2297 root_item->drop_level = 0;
2298 btrfs_set_root_refs(root_item, 0);
2299 btrfs_update_reloc_root(trans, root);
2300 }
2301
2302 if (trans)
2303 btrfs_end_transaction_throttle(trans, root);
2304
2305 btrfs_btree_balance_dirty(root);
2306
2307 if (replaced && rc->stage == UPDATE_DATA_PTRS)
2308 invalidate_extent_cache(root, &key, &next_key);
2309
2310 return err;
2311 }
2312
2313 static noinline_for_stack
2314 int prepare_to_merge(struct reloc_control *rc, int err)
2315 {
2316 struct btrfs_root *root = rc->extent_root;
2317 struct btrfs_fs_info *fs_info = root->fs_info;
2318 struct btrfs_root *reloc_root;
2319 struct btrfs_trans_handle *trans;
2320 LIST_HEAD(reloc_roots);
2321 u64 num_bytes = 0;
2322 int ret;
2323
2324 mutex_lock(&fs_info->reloc_mutex);
2325 rc->merging_rsv_size += fs_info->nodesize * (BTRFS_MAX_LEVEL - 1) * 2;
2326 rc->merging_rsv_size += rc->nodes_relocated * 2;
2327 mutex_unlock(&fs_info->reloc_mutex);
2328
2329 again:
2330 if (!err) {
2331 num_bytes = rc->merging_rsv_size;
2332 ret = btrfs_block_rsv_add(root, rc->block_rsv, num_bytes,
2333 BTRFS_RESERVE_FLUSH_ALL);
2334 if (ret)
2335 err = ret;
2336 }
2337
2338 trans = btrfs_join_transaction(rc->extent_root);
2339 if (IS_ERR(trans)) {
2340 if (!err)
2341 btrfs_block_rsv_release(rc->extent_root,
2342 rc->block_rsv, num_bytes);
2343 return PTR_ERR(trans);
2344 }
2345
2346 if (!err) {
2347 if (num_bytes != rc->merging_rsv_size) {
2348 btrfs_end_transaction(trans, rc->extent_root);
2349 btrfs_block_rsv_release(rc->extent_root,
2350 rc->block_rsv, num_bytes);
2351 goto again;
2352 }
2353 }
2354
2355 rc->merge_reloc_tree = 1;
2356
2357 while (!list_empty(&rc->reloc_roots)) {
2358 reloc_root = list_entry(rc->reloc_roots.next,
2359 struct btrfs_root, root_list);
2360 list_del_init(&reloc_root->root_list);
2361
2362 root = read_fs_root(fs_info, reloc_root->root_key.offset);
2363 BUG_ON(IS_ERR(root));
2364 BUG_ON(root->reloc_root != reloc_root);
2365
2366 /*
2367 * set reference count to 1, so btrfs_recover_relocation
2368 * knows it should resumes merging
2369 */
2370 if (!err)
2371 btrfs_set_root_refs(&reloc_root->root_item, 1);
2372 btrfs_update_reloc_root(trans, root);
2373
2374 list_add(&reloc_root->root_list, &reloc_roots);
2375 }
2376
2377 list_splice(&reloc_roots, &rc->reloc_roots);
2378
2379 if (!err)
2380 btrfs_commit_transaction(trans, rc->extent_root);
2381 else
2382 btrfs_end_transaction(trans, rc->extent_root);
2383 return err;
2384 }
2385
2386 static noinline_for_stack
2387 void free_reloc_roots(struct list_head *list)
2388 {
2389 struct btrfs_root *reloc_root;
2390
2391 while (!list_empty(list)) {
2392 reloc_root = list_entry(list->next, struct btrfs_root,
2393 root_list);
2394 free_extent_buffer(reloc_root->node);
2395 free_extent_buffer(reloc_root->commit_root);
2396 reloc_root->node = NULL;
2397 reloc_root->commit_root = NULL;
2398 __del_reloc_root(reloc_root);
2399 }
2400 }
2401
2402 static noinline_for_stack
2403 void merge_reloc_roots(struct reloc_control *rc)
2404 {
2405 struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
2406 struct btrfs_root *root;
2407 struct btrfs_root *reloc_root;
2408 u64 last_snap;
2409 u64 otransid;
2410 u64 objectid;
2411 LIST_HEAD(reloc_roots);
2412 int found = 0;
2413 int ret = 0;
2414 again:
2415 root = rc->extent_root;
2416
2417 /*
2418 * this serializes us with btrfs_record_root_in_transaction,
2419 * we have to make sure nobody is in the middle of
2420 * adding their roots to the list while we are
2421 * doing this splice
2422 */
2423 mutex_lock(&fs_info->reloc_mutex);
2424 list_splice_init(&rc->reloc_roots, &reloc_roots);
2425 mutex_unlock(&fs_info->reloc_mutex);
2426
2427 while (!list_empty(&reloc_roots)) {
2428 found = 1;
2429 reloc_root = list_entry(reloc_roots.next,
2430 struct btrfs_root, root_list);
2431
2432 if (btrfs_root_refs(&reloc_root->root_item) > 0) {
2433 root = read_fs_root(fs_info,
2434 reloc_root->root_key.offset);
2435 BUG_ON(IS_ERR(root));
2436 BUG_ON(root->reloc_root != reloc_root);
2437
2438 ret = merge_reloc_root(rc, root);
2439 if (ret) {
2440 if (list_empty(&reloc_root->root_list))
2441 list_add_tail(&reloc_root->root_list,
2442 &reloc_roots);
2443 goto out;
2444 }
2445 } else {
2446 list_del_init(&reloc_root->root_list);
2447 }
2448
2449 /*
2450 * we keep the old last snapshot transid in rtranid when we
2451 * created the relocation tree.
2452 */
2453 last_snap = btrfs_root_rtransid(&reloc_root->root_item);
2454 otransid = btrfs_root_otransid(&reloc_root->root_item);
2455 objectid = reloc_root->root_key.offset;
2456
2457 ret = btrfs_drop_snapshot(reloc_root, rc->block_rsv, 0, 1);
2458 if (ret < 0) {
2459 if (list_empty(&reloc_root->root_list))
2460 list_add_tail(&reloc_root->root_list,
2461 &reloc_roots);
2462 goto out;
2463 }
2464 }
2465
2466 if (found) {
2467 found = 0;
2468 goto again;
2469 }
2470 out:
2471 if (ret) {
2472 btrfs_handle_fs_error(fs_info, ret, NULL);
2473 if (!list_empty(&reloc_roots))
2474 free_reloc_roots(&reloc_roots);
2475
2476 /* new reloc root may be added */
2477 mutex_lock(&fs_info->reloc_mutex);
2478 list_splice_init(&rc->reloc_roots, &reloc_roots);
2479 mutex_unlock(&fs_info->reloc_mutex);
2480 if (!list_empty(&reloc_roots))
2481 free_reloc_roots(&reloc_roots);
2482 }
2483
2484 BUG_ON(!RB_EMPTY_ROOT(&rc->reloc_root_tree.rb_root));
2485 }
2486
2487 static void free_block_list(struct rb_root *blocks)
2488 {
2489 struct tree_block *block;
2490 struct rb_node *rb_node;
2491 while ((rb_node = rb_first(blocks))) {
2492 block = rb_entry(rb_node, struct tree_block, rb_node);
2493 rb_erase(rb_node, blocks);
2494 kfree(block);
2495 }
2496 }
2497
2498 static int record_reloc_root_in_trans(struct btrfs_trans_handle *trans,
2499 struct btrfs_root *reloc_root)
2500 {
2501 struct btrfs_fs_info *fs_info = reloc_root->fs_info;
2502 struct btrfs_root *root;
2503
2504 if (reloc_root->last_trans == trans->transid)
2505 return 0;
2506
2507 root = read_fs_root(fs_info, reloc_root->root_key.offset);
2508 BUG_ON(IS_ERR(root));
2509 BUG_ON(root->reloc_root != reloc_root);
2510
2511 return btrfs_record_root_in_trans(trans, root);
2512 }
2513
2514 static noinline_for_stack
2515 struct btrfs_root *select_reloc_root(struct btrfs_trans_handle *trans,
2516 struct reloc_control *rc,
2517 struct backref_node *node,
2518 struct backref_edge *edges[])
2519 {
2520 struct backref_node *next;
2521 struct btrfs_root *root;
2522 int index = 0;
2523
2524 next = node;
2525 while (1) {
2526 cond_resched();
2527 next = walk_up_backref(next, edges, &index);
2528 root = next->root;
2529 BUG_ON(!root);
2530 BUG_ON(!test_bit(BTRFS_ROOT_REF_COWS, &root->state));
2531
2532 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) {
2533 record_reloc_root_in_trans(trans, root);
2534 break;
2535 }
2536
2537 btrfs_record_root_in_trans(trans, root);
2538 root = root->reloc_root;
2539
2540 if (next->new_bytenr != root->node->start) {
2541 BUG_ON(next->new_bytenr);
2542 BUG_ON(!list_empty(&next->list));
2543 next->new_bytenr = root->node->start;
2544 next->root = root;
2545 list_add_tail(&next->list,
2546 &rc->backref_cache.changed);
2547 __mark_block_processed(rc, next);
2548 break;
2549 }
2550
2551 WARN_ON(1);
2552 root = NULL;
2553 next = walk_down_backref(edges, &index);
2554 if (!next || next->level <= node->level)
2555 break;
2556 }
2557 if (!root)
2558 return NULL;
2559
2560 next = node;
2561 /* setup backref node path for btrfs_reloc_cow_block */
2562 while (1) {
2563 rc->backref_cache.path[next->level] = next;
2564 if (--index < 0)
2565 break;
2566 next = edges[index]->node[UPPER];
2567 }
2568 return root;
2569 }
2570
2571 /*
2572 * select a tree root for relocation. return NULL if the block
2573 * is reference counted. we should use do_relocation() in this
2574 * case. return a tree root pointer if the block isn't reference
2575 * counted. return -ENOENT if the block is root of reloc tree.
2576 */
2577 static noinline_for_stack
2578 struct btrfs_root *select_one_root(struct backref_node *node)
2579 {
2580 struct backref_node *next;
2581 struct btrfs_root *root;
2582 struct btrfs_root *fs_root = NULL;
2583 struct backref_edge *edges[BTRFS_MAX_LEVEL - 1];
2584 int index = 0;
2585
2586 next = node;
2587 while (1) {
2588 cond_resched();
2589 next = walk_up_backref(next, edges, &index);
2590 root = next->root;
2591 BUG_ON(!root);
2592
2593 /* no other choice for non-references counted tree */
2594 if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state))
2595 return root;
2596
2597 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID)
2598 fs_root = root;
2599
2600 if (next != node)
2601 return NULL;
2602
2603 next = walk_down_backref(edges, &index);
2604 if (!next || next->level <= node->level)
2605 break;
2606 }
2607
2608 if (!fs_root)
2609 return ERR_PTR(-ENOENT);
2610 return fs_root;
2611 }
2612
2613 static noinline_for_stack
2614 u64 calcu_metadata_size(struct reloc_control *rc,
2615 struct backref_node *node, int reserve)
2616 {
2617 struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
2618 struct backref_node *next = node;
2619 struct backref_edge *edge;
2620 struct backref_edge *edges[BTRFS_MAX_LEVEL - 1];
2621 u64 num_bytes = 0;
2622 int index = 0;
2623
2624 BUG_ON(reserve && node->processed);
2625
2626 while (next) {
2627 cond_resched();
2628 while (1) {
2629 if (next->processed && (reserve || next != node))
2630 break;
2631
2632 num_bytes += fs_info->nodesize;
2633
2634 if (list_empty(&next->upper))
2635 break;
2636
2637 edge = list_entry(next->upper.next,
2638 struct backref_edge, list[LOWER]);
2639 edges[index++] = edge;
2640 next = edge->node[UPPER];
2641 }
2642 next = walk_down_backref(edges, &index);
2643 }
2644 return num_bytes;
2645 }
2646
2647 static int reserve_metadata_space(struct btrfs_trans_handle *trans,
2648 struct reloc_control *rc,
2649 struct backref_node *node)
2650 {
2651 struct btrfs_root *root = rc->extent_root;
2652 struct btrfs_fs_info *fs_info = root->fs_info;
2653 u64 num_bytes;
2654 int ret;
2655 u64 tmp;
2656
2657 num_bytes = calcu_metadata_size(rc, node, 1) * 2;
2658
2659 trans->block_rsv = rc->block_rsv;
2660 rc->reserved_bytes += num_bytes;
2661
2662 /*
2663 * We are under a transaction here so we can only do limited flushing.
2664 * If we get an enospc just kick back -EAGAIN so we know to drop the
2665 * transaction and try to refill when we can flush all the things.
2666 */
2667 ret = btrfs_block_rsv_refill(root, rc->block_rsv, num_bytes,
2668 BTRFS_RESERVE_FLUSH_LIMIT);
2669 if (ret) {
2670 tmp = fs_info->nodesize * RELOCATION_RESERVED_NODES;
2671 while (tmp <= rc->reserved_bytes)
2672 tmp <<= 1;
2673 /*
2674 * only one thread can access block_rsv at this point,
2675 * so we don't need hold lock to protect block_rsv.
2676 * we expand more reservation size here to allow enough
2677 * space for relocation and we will return eailer in
2678 * enospc case.
2679 */
2680 rc->block_rsv->size = tmp + fs_info->nodesize *
2681 RELOCATION_RESERVED_NODES;
2682 return -EAGAIN;
2683 }
2684
2685 return 0;
2686 }
2687
2688 /*
2689 * relocate a block tree, and then update pointers in upper level
2690 * blocks that reference the block to point to the new location.
2691 *
2692 * if called by link_to_upper, the block has already been relocated.
2693 * in that case this function just updates pointers.
2694 */
2695 static int do_relocation(struct btrfs_trans_handle *trans,
2696 struct reloc_control *rc,
2697 struct backref_node *node,
2698 struct btrfs_key *key,
2699 struct btrfs_path *path, int lowest)
2700 {
2701 struct backref_node *upper;
2702 struct backref_edge *edge;
2703 struct backref_edge *edges[BTRFS_MAX_LEVEL - 1];
2704 struct btrfs_root *root;
2705 struct extent_buffer *eb;
2706 u32 blocksize;
2707 u64 bytenr;
2708 u64 generation;
2709 int slot;
2710 int ret;
2711 int err = 0;
2712
2713 BUG_ON(lowest && node->eb);
2714
2715 path->lowest_level = node->level + 1;
2716 rc->backref_cache.path[node->level] = node;
2717 list_for_each_entry(edge, &node->upper, list[LOWER]) {
2718 cond_resched();
2719
2720 upper = edge->node[UPPER];
2721 root = select_reloc_root(trans, rc, upper, edges);
2722 BUG_ON(!root);
2723
2724 if (upper->eb && !upper->locked) {
2725 if (!lowest) {
2726 ret = btrfs_bin_search(upper->eb, key,
2727 upper->level, &slot);
2728 BUG_ON(ret);
2729 bytenr = btrfs_node_blockptr(upper->eb, slot);
2730 if (node->eb->start == bytenr)
2731 goto next;
2732 }
2733 drop_node_buffer(upper);
2734 }
2735
2736 if (!upper->eb) {
2737 ret = btrfs_search_slot(trans, root, key, path, 0, 1);
2738 if (ret) {
2739 if (ret < 0)
2740 err = ret;
2741 else
2742 err = -ENOENT;
2743
2744 btrfs_release_path(path);
2745 break;
2746 }
2747
2748 if (!upper->eb) {
2749 upper->eb = path->nodes[upper->level];
2750 path->nodes[upper->level] = NULL;
2751 } else {
2752 BUG_ON(upper->eb != path->nodes[upper->level]);
2753 }
2754
2755 upper->locked = 1;
2756 path->locks[upper->level] = 0;
2757
2758 slot = path->slots[upper->level];
2759 btrfs_release_path(path);
2760 } else {
2761 ret = btrfs_bin_search(upper->eb, key, upper->level,
2762 &slot);
2763 BUG_ON(ret);
2764 }
2765
2766 bytenr = btrfs_node_blockptr(upper->eb, slot);
2767 if (lowest) {
2768 if (bytenr != node->bytenr) {
2769 btrfs_err(root->fs_info,
2770 "lowest leaf/node mismatch: bytenr %llu node->bytenr %llu slot %d upper %llu",
2771 bytenr, node->bytenr, slot,
2772 upper->eb->start);
2773 err = -EIO;
2774 goto next;
2775 }
2776 } else {
2777 if (node->eb->start == bytenr)
2778 goto next;
2779 }
2780
2781 blocksize = root->fs_info->nodesize;
2782 generation = btrfs_node_ptr_generation(upper->eb, slot);
2783 eb = read_tree_block(root, bytenr, generation);
2784 if (IS_ERR(eb)) {
2785 err = PTR_ERR(eb);
2786 goto next;
2787 } else if (!extent_buffer_uptodate(eb)) {
2788 free_extent_buffer(eb);
2789 err = -EIO;
2790 goto next;
2791 }
2792 btrfs_tree_lock(eb);
2793 btrfs_set_lock_blocking(eb);
2794
2795 if (!node->eb) {
2796 ret = btrfs_cow_block(trans, root, eb, upper->eb,
2797 slot, &eb);
2798 btrfs_tree_unlock(eb);
2799 free_extent_buffer(eb);
2800 if (ret < 0) {
2801 err = ret;
2802 goto next;
2803 }
2804 BUG_ON(node->eb != eb);
2805 } else {
2806 btrfs_set_node_blockptr(upper->eb, slot,
2807 node->eb->start);
2808 btrfs_set_node_ptr_generation(upper->eb, slot,
2809 trans->transid);
2810 btrfs_mark_buffer_dirty(upper->eb);
2811
2812 ret = btrfs_inc_extent_ref(trans, root,
2813 node->eb->start, blocksize,
2814 upper->eb->start,
2815 btrfs_header_owner(upper->eb),
2816 node->level, 0);
2817 BUG_ON(ret);
2818
2819 ret = btrfs_drop_subtree(trans, root, eb, upper->eb);
2820 BUG_ON(ret);
2821 }
2822 next:
2823 if (!upper->pending)
2824 drop_node_buffer(upper);
2825 else
2826 unlock_node_buffer(upper);
2827 if (err)
2828 break;
2829 }
2830
2831 if (!err && node->pending) {
2832 drop_node_buffer(node);
2833 list_move_tail(&node->list, &rc->backref_cache.changed);
2834 node->pending = 0;
2835 }
2836
2837 path->lowest_level = 0;
2838 BUG_ON(err == -ENOSPC);
2839 return err;
2840 }
2841
2842 static int link_to_upper(struct btrfs_trans_handle *trans,
2843 struct reloc_control *rc,
2844 struct backref_node *node,
2845 struct btrfs_path *path)
2846 {
2847 struct btrfs_key key;
2848
2849 btrfs_node_key_to_cpu(node->eb, &key, 0);
2850 return do_relocation(trans, rc, node, &key, path, 0);
2851 }
2852
2853 static int finish_pending_nodes(struct btrfs_trans_handle *trans,
2854 struct reloc_control *rc,
2855 struct btrfs_path *path, int err)
2856 {
2857 LIST_HEAD(list);
2858 struct backref_cache *cache = &rc->backref_cache;
2859 struct backref_node *node;
2860 int level;
2861 int ret;
2862
2863 for (level = 0; level < BTRFS_MAX_LEVEL; level++) {
2864 while (!list_empty(&cache->pending[level])) {
2865 node = list_entry(cache->pending[level].next,
2866 struct backref_node, list);
2867 list_move_tail(&node->list, &list);
2868 BUG_ON(!node->pending);
2869
2870 if (!err) {
2871 ret = link_to_upper(trans, rc, node, path);
2872 if (ret < 0)
2873 err = ret;
2874 }
2875 }
2876 list_splice_init(&list, &cache->pending[level]);
2877 }
2878 return err;
2879 }
2880
2881 static void mark_block_processed(struct reloc_control *rc,
2882 u64 bytenr, u32 blocksize)
2883 {
2884 set_extent_bits(&rc->processed_blocks, bytenr, bytenr + blocksize - 1,
2885 EXTENT_DIRTY);
2886 }
2887
2888 static void __mark_block_processed(struct reloc_control *rc,
2889 struct backref_node *node)
2890 {
2891 u32 blocksize;
2892 if (node->level == 0 ||
2893 in_block_group(node->bytenr, rc->block_group)) {
2894 blocksize = rc->extent_root->fs_info->nodesize;
2895 mark_block_processed(rc, node->bytenr, blocksize);
2896 }
2897 node->processed = 1;
2898 }
2899
2900 /*
2901 * mark a block and all blocks directly/indirectly reference the block
2902 * as processed.
2903 */
2904 static void update_processed_blocks(struct reloc_control *rc,
2905 struct backref_node *node)
2906 {
2907 struct backref_node *next = node;
2908 struct backref_edge *edge;
2909 struct backref_edge *edges[BTRFS_MAX_LEVEL - 1];
2910 int index = 0;
2911
2912 while (next) {
2913 cond_resched();
2914 while (1) {
2915 if (next->processed)
2916 break;
2917
2918 __mark_block_processed(rc, next);
2919
2920 if (list_empty(&next->upper))
2921 break;
2922
2923 edge = list_entry(next->upper.next,
2924 struct backref_edge, list[LOWER]);
2925 edges[index++] = edge;
2926 next = edge->node[UPPER];
2927 }
2928 next = walk_down_backref(edges, &index);
2929 }
2930 }
2931
2932 static int tree_block_processed(u64 bytenr, struct reloc_control *rc)
2933 {
2934 u32 blocksize = rc->extent_root->fs_info->nodesize;
2935
2936 if (test_range_bit(&rc->processed_blocks, bytenr,
2937 bytenr + blocksize - 1, EXTENT_DIRTY, 1, NULL))
2938 return 1;
2939 return 0;
2940 }
2941
2942 static int get_tree_block_key(struct reloc_control *rc,
2943 struct tree_block *block)
2944 {
2945 struct extent_buffer *eb;
2946
2947 BUG_ON(block->key_ready);
2948 eb = read_tree_block(rc->extent_root, block->bytenr,
2949 block->key.offset);
2950 if (IS_ERR(eb)) {
2951 return PTR_ERR(eb);
2952 } else if (!extent_buffer_uptodate(eb)) {
2953 free_extent_buffer(eb);
2954 return -EIO;
2955 }
2956 WARN_ON(btrfs_header_level(eb) != block->level);
2957 if (block->level == 0)
2958 btrfs_item_key_to_cpu(eb, &block->key, 0);
2959 else
2960 btrfs_node_key_to_cpu(eb, &block->key, 0);
2961 free_extent_buffer(eb);
2962 block->key_ready = 1;
2963 return 0;
2964 }
2965
2966 /*
2967 * helper function to relocate a tree block
2968 */
2969 static int relocate_tree_block(struct btrfs_trans_handle *trans,
2970 struct reloc_control *rc,
2971 struct backref_node *node,
2972 struct btrfs_key *key,
2973 struct btrfs_path *path)
2974 {
2975 struct btrfs_root *root;
2976 int ret = 0;
2977
2978 if (!node)
2979 return 0;
2980
2981 BUG_ON(node->processed);
2982 root = select_one_root(node);
2983 if (root == ERR_PTR(-ENOENT)) {
2984 update_processed_blocks(rc, node);
2985 goto out;
2986 }
2987
2988 if (!root || test_bit(BTRFS_ROOT_REF_COWS, &root->state)) {
2989 ret = reserve_metadata_space(trans, rc, node);
2990 if (ret)
2991 goto out;
2992 }
2993
2994 if (root) {
2995 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state)) {
2996 BUG_ON(node->new_bytenr);
2997 BUG_ON(!list_empty(&node->list));
2998 btrfs_record_root_in_trans(trans, root);
2999 root = root->reloc_root;
3000 node->new_bytenr = root->node->start;
3001 node->root = root;
3002 list_add_tail(&node->list, &rc->backref_cache.changed);
3003 } else {
3004 path->lowest_level = node->level;
3005 ret = btrfs_search_slot(trans, root, key, path, 0, 1);
3006 btrfs_release_path(path);
3007 if (ret > 0)
3008 ret = 0;
3009 }
3010 if (!ret)
3011 update_processed_blocks(rc, node);
3012 } else {
3013 ret = do_relocation(trans, rc, node, key, path, 1);
3014 }
3015 out:
3016 if (ret || node->level == 0 || node->cowonly)
3017 remove_backref_node(&rc->backref_cache, node);
3018 return ret;
3019 }
3020
3021 /*
3022 * relocate a list of blocks
3023 */
3024 static noinline_for_stack
3025 int relocate_tree_blocks(struct btrfs_trans_handle *trans,
3026 struct reloc_control *rc, struct rb_root *blocks)
3027 {
3028 struct backref_node *node;
3029 struct btrfs_path *path;
3030 struct tree_block *block;
3031 struct rb_node *rb_node;
3032 int ret;
3033 int err = 0;
3034
3035 path = btrfs_alloc_path();
3036 if (!path) {
3037 err = -ENOMEM;
3038 goto out_free_blocks;
3039 }
3040
3041 rb_node = rb_first(blocks);
3042 while (rb_node) {
3043 block = rb_entry(rb_node, struct tree_block, rb_node);
3044 if (!block->key_ready)
3045 readahead_tree_block(rc->extent_root, block->bytenr);
3046 rb_node = rb_next(rb_node);
3047 }
3048
3049 rb_node = rb_first(blocks);
3050 while (rb_node) {
3051 block = rb_entry(rb_node, struct tree_block, rb_node);
3052 if (!block->key_ready) {
3053 err = get_tree_block_key(rc, block);
3054 if (err)
3055 goto out_free_path;
3056 }
3057 rb_node = rb_next(rb_node);
3058 }
3059
3060 rb_node = rb_first(blocks);
3061 while (rb_node) {
3062 block = rb_entry(rb_node, struct tree_block, rb_node);
3063
3064 node = build_backref_tree(rc, &block->key,
3065 block->level, block->bytenr);
3066 if (IS_ERR(node)) {
3067 err = PTR_ERR(node);
3068 goto out;
3069 }
3070
3071 ret = relocate_tree_block(trans, rc, node, &block->key,
3072 path);
3073 if (ret < 0) {
3074 if (ret != -EAGAIN || rb_node == rb_first(blocks))
3075 err = ret;
3076 goto out;
3077 }
3078 rb_node = rb_next(rb_node);
3079 }
3080 out:
3081 err = finish_pending_nodes(trans, rc, path, err);
3082
3083 out_free_path:
3084 btrfs_free_path(path);
3085 out_free_blocks:
3086 free_block_list(blocks);
3087 return err;
3088 }
3089
3090 static noinline_for_stack
3091 int prealloc_file_extent_cluster(struct inode *inode,
3092 struct file_extent_cluster *cluster)
3093 {
3094 u64 alloc_hint = 0;
3095 u64 start;
3096 u64 end;
3097 u64 offset = BTRFS_I(inode)->index_cnt;
3098 u64 num_bytes;
3099 int nr = 0;
3100 int ret = 0;
3101 u64 prealloc_start = cluster->start - offset;
3102 u64 prealloc_end = cluster->end - offset;
3103 u64 cur_offset;
3104
3105 BUG_ON(cluster->start != cluster->boundary[0]);
3106 inode_lock(inode);
3107
3108 ret = btrfs_check_data_free_space(inode, prealloc_start,
3109 prealloc_end + 1 - prealloc_start);
3110 if (ret)
3111 goto out;
3112
3113 cur_offset = prealloc_start;
3114 while (nr < cluster->nr) {
3115 start = cluster->boundary[nr] - offset;
3116 if (nr + 1 < cluster->nr)
3117 end = cluster->boundary[nr + 1] - 1 - offset;
3118 else
3119 end = cluster->end - offset;
3120
3121 lock_extent(&BTRFS_I(inode)->io_tree, start, end);
3122 num_bytes = end + 1 - start;
3123 if (cur_offset < start)
3124 btrfs_free_reserved_data_space(inode, cur_offset,
3125 start - cur_offset);
3126 ret = btrfs_prealloc_file_range(inode, 0, start,
3127 num_bytes, num_bytes,
3128 end + 1, &alloc_hint);
3129 cur_offset = end + 1;
3130 unlock_extent(&BTRFS_I(inode)->io_tree, start, end);
3131 if (ret)
3132 break;
3133 nr++;
3134 }
3135 if (cur_offset < prealloc_end)
3136 btrfs_free_reserved_data_space(inode, cur_offset,
3137 prealloc_end + 1 - cur_offset);
3138 out:
3139 inode_unlock(inode);
3140 return ret;
3141 }
3142
3143 static noinline_for_stack
3144 int setup_extent_mapping(struct inode *inode, u64 start, u64 end,
3145 u64 block_start)
3146 {
3147 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3148 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
3149 struct extent_map *em;
3150 int ret = 0;
3151
3152 em = alloc_extent_map();
3153 if (!em)
3154 return -ENOMEM;
3155
3156 em->start = start;
3157 em->len = end + 1 - start;
3158 em->block_len = em->len;
3159 em->block_start = block_start;
3160 em->bdev = fs_info->fs_devices->latest_bdev;
3161 set_bit(EXTENT_FLAG_PINNED, &em->flags);
3162
3163 lock_extent(&BTRFS_I(inode)->io_tree, start, end);
3164 while (1) {
3165 write_lock(&em_tree->lock);
3166 ret = add_extent_mapping(em_tree, em, 0);
3167 write_unlock(&em_tree->lock);
3168 if (ret != -EEXIST) {
3169 free_extent_map(em);
3170 break;
3171 }
3172 btrfs_drop_extent_cache(inode, start, end, 0);
3173 }
3174 unlock_extent(&BTRFS_I(inode)->io_tree, start, end);
3175 return ret;
3176 }
3177
3178 static int relocate_file_extent_cluster(struct inode *inode,
3179 struct file_extent_cluster *cluster)
3180 {
3181 u64 page_start;
3182 u64 page_end;
3183 u64 offset = BTRFS_I(inode)->index_cnt;
3184 unsigned long index;
3185 unsigned long last_index;
3186 struct page *page;
3187 struct file_ra_state *ra;
3188 gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping);
3189 int nr = 0;
3190 int ret = 0;
3191
3192 if (!cluster->nr)
3193 return 0;
3194
3195 ra = kzalloc(sizeof(*ra), GFP_NOFS);
3196 if (!ra)
3197 return -ENOMEM;
3198
3199 ret = prealloc_file_extent_cluster(inode, cluster);
3200 if (ret)
3201 goto out;
3202
3203 file_ra_state_init(ra, inode->i_mapping);
3204
3205 ret = setup_extent_mapping(inode, cluster->start - offset,
3206 cluster->end - offset, cluster->start);
3207 if (ret)
3208 goto out;
3209
3210 index = (cluster->start - offset) >> PAGE_SHIFT;
3211 last_index = (cluster->end - offset) >> PAGE_SHIFT;
3212 while (index <= last_index) {
3213 ret = btrfs_delalloc_reserve_metadata(inode, PAGE_SIZE);
3214 if (ret)
3215 goto out;
3216
3217 page = find_lock_page(inode->i_mapping, index);
3218 if (!page) {
3219 page_cache_sync_readahead(inode->i_mapping,
3220 ra, NULL, index,
3221 last_index + 1 - index);
3222 page = find_or_create_page(inode->i_mapping, index,
3223 mask);
3224 if (!page) {
3225 btrfs_delalloc_release_metadata(inode,
3226 PAGE_SIZE);
3227 ret = -ENOMEM;
3228 goto out;
3229 }
3230 }
3231
3232 if (PageReadahead(page)) {
3233 page_cache_async_readahead(inode->i_mapping,
3234 ra, NULL, page, index,
3235 last_index + 1 - index);
3236 }
3237
3238 if (!PageUptodate(page)) {
3239 btrfs_readpage(NULL, page);
3240 lock_page(page);
3241 if (!PageUptodate(page)) {
3242 unlock_page(page);
3243 put_page(page);
3244 btrfs_delalloc_release_metadata(inode,
3245 PAGE_SIZE);
3246 ret = -EIO;
3247 goto out;
3248 }
3249 }
3250
3251 page_start = page_offset(page);
3252 page_end = page_start + PAGE_SIZE - 1;
3253
3254 lock_extent(&BTRFS_I(inode)->io_tree, page_start, page_end);
3255
3256 set_page_extent_mapped(page);
3257
3258 if (nr < cluster->nr &&
3259 page_start + offset == cluster->boundary[nr]) {
3260 set_extent_bits(&BTRFS_I(inode)->io_tree,
3261 page_start, page_end,
3262 EXTENT_BOUNDARY);
3263 nr++;
3264 }
3265
3266 btrfs_set_extent_delalloc(inode, page_start, page_end, NULL, 0);
3267 set_page_dirty(page);
3268
3269 unlock_extent(&BTRFS_I(inode)->io_tree,
3270 page_start, page_end);
3271 unlock_page(page);
3272 put_page(page);
3273
3274 index++;
3275 balance_dirty_pages_ratelimited(inode->i_mapping);
3276 btrfs_throttle(BTRFS_I(inode)->root);
3277 }
3278 WARN_ON(nr != cluster->nr);
3279 out:
3280 kfree(ra);
3281 return ret;
3282 }
3283
3284 static noinline_for_stack
3285 int relocate_data_extent(struct inode *inode, struct btrfs_key *extent_key,
3286 struct file_extent_cluster *cluster)
3287 {
3288 int ret;
3289
3290 if (cluster->nr > 0 && extent_key->objectid != cluster->end + 1) {
3291 ret = relocate_file_extent_cluster(inode, cluster);
3292 if (ret)
3293 return ret;
3294 cluster->nr = 0;
3295 }
3296
3297 if (!cluster->nr)
3298 cluster->start = extent_key->objectid;
3299 else
3300 BUG_ON(cluster->nr >= MAX_EXTENTS);
3301 cluster->end = extent_key->objectid + extent_key->offset - 1;
3302 cluster->boundary[cluster->nr] = extent_key->objectid;
3303 cluster->nr++;
3304
3305 if (cluster->nr >= MAX_EXTENTS) {
3306 ret = relocate_file_extent_cluster(inode, cluster);
3307 if (ret)
3308 return ret;
3309 cluster->nr = 0;
3310 }
3311 return 0;
3312 }
3313
3314 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
3315 static int get_ref_objectid_v0(struct reloc_control *rc,
3316 struct btrfs_path *path,
3317 struct btrfs_key *extent_key,
3318 u64 *ref_objectid, int *path_change)
3319 {
3320 struct btrfs_key key;
3321 struct extent_buffer *leaf;
3322 struct btrfs_extent_ref_v0 *ref0;
3323 int ret;
3324 int slot;
3325
3326 leaf = path->nodes[0];
3327 slot = path->slots[0];
3328 while (1) {
3329 if (slot >= btrfs_header_nritems(leaf)) {
3330 ret = btrfs_next_leaf(rc->extent_root, path);
3331 if (ret < 0)
3332 return ret;
3333 BUG_ON(ret > 0);
3334 leaf = path->nodes[0];
3335 slot = path->slots[0];
3336 if (path_change)
3337 *path_change = 1;
3338 }
3339 btrfs_item_key_to_cpu(leaf, &key, slot);
3340 if (key.objectid != extent_key->objectid)
3341 return -ENOENT;
3342
3343 if (key.type != BTRFS_EXTENT_REF_V0_KEY) {
3344 slot++;
3345 continue;
3346 }
3347 ref0 = btrfs_item_ptr(leaf, slot,
3348 struct btrfs_extent_ref_v0);
3349 *ref_objectid = btrfs_ref_objectid_v0(leaf, ref0);
3350 break;
3351 }
3352 return 0;
3353 }
3354 #endif
3355
3356 /*
3357 * helper to add a tree block to the list.
3358 * the major work is getting the generation and level of the block
3359 */
3360 static int add_tree_block(struct reloc_control *rc,
3361 struct btrfs_key *extent_key,
3362 struct btrfs_path *path,
3363 struct rb_root *blocks)
3364 {
3365 struct extent_buffer *eb;
3366 struct btrfs_extent_item *ei;
3367 struct btrfs_tree_block_info *bi;
3368 struct tree_block *block;
3369 struct rb_node *rb_node;
3370 u32 item_size;
3371 int level = -1;
3372 u64 generation;
3373
3374 eb = path->nodes[0];
3375 item_size = btrfs_item_size_nr(eb, path->slots[0]);
3376
3377 if (extent_key->type == BTRFS_METADATA_ITEM_KEY ||
3378 item_size >= sizeof(*ei) + sizeof(*bi)) {
3379 ei = btrfs_item_ptr(eb, path->slots[0],
3380 struct btrfs_extent_item);
3381 if (extent_key->type == BTRFS_EXTENT_ITEM_KEY) {
3382 bi = (struct btrfs_tree_block_info *)(ei + 1);
3383 level = btrfs_tree_block_level(eb, bi);
3384 } else {
3385 level = (int)extent_key->offset;
3386 }
3387 generation = btrfs_extent_generation(eb, ei);
3388 } else {
3389 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
3390 u64 ref_owner;
3391 int ret;
3392
3393 BUG_ON(item_size != sizeof(struct btrfs_extent_item_v0));
3394 ret = get_ref_objectid_v0(rc, path, extent_key,
3395 &ref_owner, NULL);
3396 if (ret < 0)
3397 return ret;
3398 BUG_ON(ref_owner >= BTRFS_MAX_LEVEL);
3399 level = (int)ref_owner;
3400 /* FIXME: get real generation */
3401 generation = 0;
3402 #else
3403 BUG();
3404 #endif
3405 }
3406
3407 btrfs_release_path(path);
3408
3409 BUG_ON(level == -1);
3410
3411 block = kmalloc(sizeof(*block), GFP_NOFS);
3412 if (!block)
3413 return -ENOMEM;
3414
3415 block->bytenr = extent_key->objectid;
3416 block->key.objectid = rc->extent_root->fs_info->nodesize;
3417 block->key.offset = generation;
3418 block->level = level;
3419 block->key_ready = 0;
3420
3421 rb_node = tree_insert(blocks, block->bytenr, &block->rb_node);
3422 if (rb_node)
3423 backref_tree_panic(rb_node, -EEXIST, block->bytenr);
3424
3425 return 0;
3426 }
3427
3428 /*
3429 * helper to add tree blocks for backref of type BTRFS_SHARED_DATA_REF_KEY
3430 */
3431 static int __add_tree_block(struct reloc_control *rc,
3432 u64 bytenr, u32 blocksize,
3433 struct rb_root *blocks)
3434 {
3435 struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
3436 struct btrfs_path *path;
3437 struct btrfs_key key;
3438 int ret;
3439 bool skinny = btrfs_fs_incompat(fs_info, SKINNY_METADATA);
3440
3441 if (tree_block_processed(bytenr, rc))
3442 return 0;
3443
3444 if (tree_search(blocks, bytenr))
3445 return 0;
3446
3447 path = btrfs_alloc_path();
3448 if (!path)
3449 return -ENOMEM;
3450 again:
3451 key.objectid = bytenr;
3452 if (skinny) {
3453 key.type = BTRFS_METADATA_ITEM_KEY;
3454 key.offset = (u64)-1;
3455 } else {
3456 key.type = BTRFS_EXTENT_ITEM_KEY;
3457 key.offset = blocksize;
3458 }
3459
3460 path->search_commit_root = 1;
3461 path->skip_locking = 1;
3462 ret = btrfs_search_slot(NULL, rc->extent_root, &key, path, 0, 0);
3463 if (ret < 0)
3464 goto out;
3465
3466 if (ret > 0 && skinny) {
3467 if (path->slots[0]) {
3468 path->slots[0]--;
3469 btrfs_item_key_to_cpu(path->nodes[0], &key,
3470 path->slots[0]);
3471 if (key.objectid == bytenr &&
3472 (key.type == BTRFS_METADATA_ITEM_KEY ||
3473 (key.type == BTRFS_EXTENT_ITEM_KEY &&
3474 key.offset == blocksize)))
3475 ret = 0;
3476 }
3477
3478 if (ret) {
3479 skinny = false;
3480 btrfs_release_path(path);
3481 goto again;
3482 }
3483 }
3484 BUG_ON(ret);
3485
3486 ret = add_tree_block(rc, &key, path, blocks);
3487 out:
3488 btrfs_free_path(path);
3489 return ret;
3490 }
3491
3492 /*
3493 * helper to check if the block use full backrefs for pointers in it
3494 */
3495 static int block_use_full_backref(struct reloc_control *rc,
3496 struct extent_buffer *eb)
3497 {
3498 u64 flags;
3499 int ret;
3500
3501 if (btrfs_header_flag(eb, BTRFS_HEADER_FLAG_RELOC) ||
3502 btrfs_header_backref_rev(eb) < BTRFS_MIXED_BACKREF_REV)
3503 return 1;
3504
3505 ret = btrfs_lookup_extent_info(NULL, rc->extent_root,
3506 eb->start, btrfs_header_level(eb), 1,
3507 NULL, &flags);
3508 BUG_ON(ret);
3509
3510 if (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)
3511 ret = 1;
3512 else
3513 ret = 0;
3514 return ret;
3515 }
3516
3517 static int delete_block_group_cache(struct btrfs_fs_info *fs_info,
3518 struct btrfs_block_group_cache *block_group,
3519 struct inode *inode,
3520 u64 ino)
3521 {
3522 struct btrfs_key key;
3523 struct btrfs_root *root = fs_info->tree_root;
3524 struct btrfs_trans_handle *trans;
3525 int ret = 0;
3526
3527 if (inode)
3528 goto truncate;
3529
3530 key.objectid = ino;
3531 key.type = BTRFS_INODE_ITEM_KEY;
3532 key.offset = 0;
3533
3534 inode = btrfs_iget(fs_info->sb, &key, root, NULL);
3535 if (IS_ERR(inode) || is_bad_inode(inode)) {
3536 if (!IS_ERR(inode))
3537 iput(inode);
3538 return -ENOENT;
3539 }
3540
3541 truncate:
3542 ret = btrfs_check_trunc_cache_free_space(root,
3543 &fs_info->global_block_rsv);
3544 if (ret)
3545 goto out;
3546
3547 trans = btrfs_join_transaction(root);
3548 if (IS_ERR(trans)) {
3549 ret = PTR_ERR(trans);
3550 goto out;
3551 }
3552
3553 ret = btrfs_truncate_free_space_cache(root, trans, block_group, inode);
3554
3555 btrfs_end_transaction(trans, root);
3556 btrfs_btree_balance_dirty(root);
3557 out:
3558 iput(inode);
3559 return ret;
3560 }
3561
3562 /*
3563 * helper to add tree blocks for backref of type BTRFS_EXTENT_DATA_REF_KEY
3564 * this function scans fs tree to find blocks reference the data extent
3565 */
3566 static int find_data_references(struct reloc_control *rc,
3567 struct btrfs_key *extent_key,
3568 struct extent_buffer *leaf,
3569 struct btrfs_extent_data_ref *ref,
3570 struct rb_root *blocks)
3571 {
3572 struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
3573 struct btrfs_path *path;
3574 struct tree_block *block;
3575 struct btrfs_root *root;
3576 struct btrfs_file_extent_item *fi;
3577 struct rb_node *rb_node;
3578 struct btrfs_key key;
3579 u64 ref_root;
3580 u64 ref_objectid;
3581 u64 ref_offset;
3582 u32 ref_count;
3583 u32 nritems;
3584 int err = 0;
3585 int added = 0;
3586 int counted;
3587 int ret;
3588
3589 ref_root = btrfs_extent_data_ref_root(leaf, ref);
3590 ref_objectid = btrfs_extent_data_ref_objectid(leaf, ref);
3591 ref_offset = btrfs_extent_data_ref_offset(leaf, ref);
3592 ref_count = btrfs_extent_data_ref_count(leaf, ref);
3593
3594 /*
3595 * This is an extent belonging to the free space cache, lets just delete
3596 * it and redo the search.
3597 */
3598 if (ref_root == BTRFS_ROOT_TREE_OBJECTID) {
3599 ret = delete_block_group_cache(fs_info, rc->block_group,
3600 NULL, ref_objectid);
3601 if (ret != -ENOENT)
3602 return ret;
3603 ret = 0;
3604 }
3605
3606 path = btrfs_alloc_path();
3607 if (!path)
3608 return -ENOMEM;
3609 path->reada = READA_FORWARD;
3610
3611 root = read_fs_root(fs_info, ref_root);
3612 if (IS_ERR(root)) {
3613 err = PTR_ERR(root);
3614 goto out;
3615 }
3616
3617 key.objectid = ref_objectid;
3618 key.type = BTRFS_EXTENT_DATA_KEY;
3619 if (ref_offset > ((u64)-1 << 32))
3620 key.offset = 0;
3621 else
3622 key.offset = ref_offset;
3623
3624 path->search_commit_root = 1;
3625 path->skip_locking = 1;
3626 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3627 if (ret < 0) {
3628 err = ret;
3629 goto out;
3630 }
3631
3632 leaf = path->nodes[0];
3633 nritems = btrfs_header_nritems(leaf);
3634 /*
3635 * the references in tree blocks that use full backrefs
3636 * are not counted in
3637 */
3638 if (block_use_full_backref(rc, leaf))
3639 counted = 0;
3640 else
3641 counted = 1;
3642 rb_node = tree_search(blocks, leaf->start);
3643 if (rb_node) {
3644 if (counted)
3645 added = 1;
3646 else
3647 path->slots[0] = nritems;
3648 }
3649
3650 while (ref_count > 0) {
3651 while (path->slots[0] >= nritems) {
3652 ret = btrfs_next_leaf(root, path);
3653 if (ret < 0) {
3654 err = ret;
3655 goto out;
3656 }
3657 if (WARN_ON(ret > 0))
3658 goto out;
3659
3660 leaf = path->nodes[0];
3661 nritems = btrfs_header_nritems(leaf);
3662 added = 0;
3663
3664 if (block_use_full_backref(rc, leaf))
3665 counted = 0;
3666 else
3667 counted = 1;
3668 rb_node = tree_search(blocks, leaf->start);
3669 if (rb_node) {
3670 if (counted)
3671 added = 1;
3672 else
3673 path->slots[0] = nritems;
3674 }
3675 }
3676
3677 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3678 if (WARN_ON(key.objectid != ref_objectid ||
3679 key.type != BTRFS_EXTENT_DATA_KEY))
3680 break;
3681
3682 fi = btrfs_item_ptr(leaf, path->slots[0],
3683 struct btrfs_file_extent_item);
3684
3685 if (btrfs_file_extent_type(leaf, fi) ==
3686 BTRFS_FILE_EXTENT_INLINE)
3687 goto next;
3688
3689 if (btrfs_file_extent_disk_bytenr(leaf, fi) !=
3690 extent_key->objectid)
3691 goto next;
3692
3693 key.offset -= btrfs_file_extent_offset(leaf, fi);
3694 if (key.offset != ref_offset)
3695 goto next;
3696
3697 if (counted)
3698 ref_count--;
3699 if (added)
3700 goto next;
3701
3702 if (!tree_block_processed(leaf->start, rc)) {
3703 block = kmalloc(sizeof(*block), GFP_NOFS);
3704 if (!block) {
3705 err = -ENOMEM;
3706 break;
3707 }
3708 block->bytenr = leaf->start;
3709 btrfs_item_key_to_cpu(leaf, &block->key, 0);
3710 block->level = 0;
3711 block->key_ready = 1;
3712 rb_node = tree_insert(blocks, block->bytenr,
3713 &block->rb_node);
3714 if (rb_node)
3715 backref_tree_panic(rb_node, -EEXIST,
3716 block->bytenr);
3717 }
3718 if (counted)
3719 added = 1;
3720 else
3721 path->slots[0] = nritems;
3722 next:
3723 path->slots[0]++;
3724
3725 }
3726 out:
3727 btrfs_free_path(path);
3728 return err;
3729 }
3730
3731 /*
3732 * helper to find all tree blocks that reference a given data extent
3733 */
3734 static noinline_for_stack
3735 int add_data_references(struct reloc_control *rc,
3736 struct btrfs_key *extent_key,
3737 struct btrfs_path *path,
3738 struct rb_root *blocks)
3739 {
3740 struct btrfs_key key;
3741 struct extent_buffer *eb;
3742 struct btrfs_extent_data_ref *dref;
3743 struct btrfs_extent_inline_ref *iref;
3744 unsigned long ptr;
3745 unsigned long end;
3746 u32 blocksize = rc->extent_root->fs_info->nodesize;
3747 int ret = 0;
3748 int err = 0;
3749
3750 eb = path->nodes[0];
3751 ptr = btrfs_item_ptr_offset(eb, path->slots[0]);
3752 end = ptr + btrfs_item_size_nr(eb, path->slots[0]);
3753 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
3754 if (ptr + sizeof(struct btrfs_extent_item_v0) == end)
3755 ptr = end;
3756 else
3757 #endif
3758 ptr += sizeof(struct btrfs_extent_item);
3759
3760 while (ptr < end) {
3761 iref = (struct btrfs_extent_inline_ref *)ptr;
3762 key.type = btrfs_extent_inline_ref_type(eb, iref);
3763 if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
3764 key.offset = btrfs_extent_inline_ref_offset(eb, iref);
3765 ret = __add_tree_block(rc, key.offset, blocksize,
3766 blocks);
3767 } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
3768 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
3769 ret = find_data_references(rc, extent_key,
3770 eb, dref, blocks);
3771 } else {
3772 BUG();
3773 }
3774 if (ret) {
3775 err = ret;
3776 goto out;
3777 }
3778 ptr += btrfs_extent_inline_ref_size(key.type);
3779 }
3780 WARN_ON(ptr > end);
3781
3782 while (1) {
3783 cond_resched();
3784 eb = path->nodes[0];
3785 if (path->slots[0] >= btrfs_header_nritems(eb)) {
3786 ret = btrfs_next_leaf(rc->extent_root, path);
3787 if (ret < 0) {
3788 err = ret;
3789 break;
3790 }
3791 if (ret > 0)
3792 break;
3793 eb = path->nodes[0];
3794 }
3795
3796 btrfs_item_key_to_cpu(eb, &key, path->slots[0]);
3797 if (key.objectid != extent_key->objectid)
3798 break;
3799
3800 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
3801 if (key.type == BTRFS_SHARED_DATA_REF_KEY ||
3802 key.type == BTRFS_EXTENT_REF_V0_KEY) {
3803 #else
3804 BUG_ON(key.type == BTRFS_EXTENT_REF_V0_KEY);
3805 if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
3806 #endif
3807 ret = __add_tree_block(rc, key.offset, blocksize,
3808 blocks);
3809 } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
3810 dref = btrfs_item_ptr(eb, path->slots[0],
3811 struct btrfs_extent_data_ref);
3812 ret = find_data_references(rc, extent_key,
3813 eb, dref, blocks);
3814 } else {
3815 ret = 0;
3816 }
3817 if (ret) {
3818 err = ret;
3819 break;
3820 }
3821 path->slots[0]++;
3822 }
3823 out:
3824 btrfs_release_path(path);
3825 if (err)
3826 free_block_list(blocks);
3827 return err;
3828 }
3829
3830 /*
3831 * helper to find next unprocessed extent
3832 */
3833 static noinline_for_stack
3834 int find_next_extent(struct reloc_control *rc, struct btrfs_path *path,
3835 struct btrfs_key *extent_key)
3836 {
3837 struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
3838 struct btrfs_key key;
3839 struct extent_buffer *leaf;
3840 u64 start, end, last;
3841 int ret;
3842
3843 last = rc->block_group->key.objectid + rc->block_group->key.offset;
3844 while (1) {
3845 cond_resched();
3846 if (rc->search_start >= last) {
3847 ret = 1;
3848 break;
3849 }
3850
3851 key.objectid = rc->search_start;
3852 key.type = BTRFS_EXTENT_ITEM_KEY;
3853 key.offset = 0;
3854
3855 path->search_commit_root = 1;
3856 path->skip_locking = 1;
3857 ret = btrfs_search_slot(NULL, rc->extent_root, &key, path,
3858 0, 0);
3859 if (ret < 0)
3860 break;
3861 next:
3862 leaf = path->nodes[0];
3863 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
3864 ret = btrfs_next_leaf(rc->extent_root, path);
3865 if (ret != 0)
3866 break;
3867 leaf = path->nodes[0];
3868 }
3869
3870 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3871 if (key.objectid >= last) {
3872 ret = 1;
3873 break;
3874 }
3875
3876 if (key.type != BTRFS_EXTENT_ITEM_KEY &&
3877 key.type != BTRFS_METADATA_ITEM_KEY) {
3878 path->slots[0]++;
3879 goto next;
3880 }
3881
3882 if (key.type == BTRFS_EXTENT_ITEM_KEY &&
3883 key.objectid + key.offset <= rc->search_start) {
3884 path->slots[0]++;
3885 goto next;
3886 }
3887
3888 if (key.type == BTRFS_METADATA_ITEM_KEY &&
3889 key.objectid + fs_info->nodesize <=
3890 rc->search_start) {
3891 path->slots[0]++;
3892 goto next;
3893 }
3894
3895 ret = find_first_extent_bit(&rc->processed_blocks,
3896 key.objectid, &start, &end,
3897 EXTENT_DIRTY, NULL);
3898
3899 if (ret == 0 && start <= key.objectid) {
3900 btrfs_release_path(path);
3901 rc->search_start = end + 1;
3902 } else {
3903 if (key.type == BTRFS_EXTENT_ITEM_KEY)
3904 rc->search_start = key.objectid + key.offset;
3905 else
3906 rc->search_start = key.objectid +
3907 fs_info->nodesize;
3908 memcpy(extent_key, &key, sizeof(key));
3909 return 0;
3910 }
3911 }
3912 btrfs_release_path(path);
3913 return ret;
3914 }
3915
3916 static void set_reloc_control(struct reloc_control *rc)
3917 {
3918 struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
3919
3920 mutex_lock(&fs_info->reloc_mutex);
3921 fs_info->reloc_ctl = rc;
3922 mutex_unlock(&fs_info->reloc_mutex);
3923 }
3924
3925 static void unset_reloc_control(struct reloc_control *rc)
3926 {
3927 struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
3928
3929 mutex_lock(&fs_info->reloc_mutex);
3930 fs_info->reloc_ctl = NULL;
3931 mutex_unlock(&fs_info->reloc_mutex);
3932 }
3933
3934 static int check_extent_flags(u64 flags)
3935 {
3936 if ((flags & BTRFS_EXTENT_FLAG_DATA) &&
3937 (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK))
3938 return 1;
3939 if (!(flags & BTRFS_EXTENT_FLAG_DATA) &&
3940 !(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK))
3941 return 1;
3942 if ((flags & BTRFS_EXTENT_FLAG_DATA) &&
3943 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
3944 return 1;
3945 return 0;
3946 }
3947
3948 static noinline_for_stack
3949 int prepare_to_relocate(struct reloc_control *rc)
3950 {
3951 struct btrfs_trans_handle *trans;
3952 int ret;
3953
3954 rc->block_rsv = btrfs_alloc_block_rsv(rc->extent_root,
3955 BTRFS_BLOCK_RSV_TEMP);
3956 if (!rc->block_rsv)
3957 return -ENOMEM;
3958
3959 memset(&rc->cluster, 0, sizeof(rc->cluster));
3960 rc->search_start = rc->block_group->key.objectid;
3961 rc->extents_found = 0;
3962 rc->nodes_relocated = 0;
3963 rc->merging_rsv_size = 0;
3964 rc->reserved_bytes = 0;
3965 rc->block_rsv->size = rc->extent_root->fs_info->nodesize *
3966 RELOCATION_RESERVED_NODES;
3967 ret = btrfs_block_rsv_refill(rc->extent_root,
3968 rc->block_rsv, rc->block_rsv->size,
3969 BTRFS_RESERVE_FLUSH_ALL);
3970 if (ret)
3971 return ret;
3972
3973 rc->create_reloc_tree = 1;
3974 set_reloc_control(rc);
3975
3976 trans = btrfs_join_transaction(rc->extent_root);
3977 if (IS_ERR(trans)) {
3978 unset_reloc_control(rc);
3979 /*
3980 * extent tree is not a ref_cow tree and has no reloc_root to
3981 * cleanup. And callers are responsible to free the above
3982 * block rsv.
3983 */
3984 return PTR_ERR(trans);
3985 }
3986 btrfs_commit_transaction(trans, rc->extent_root);
3987 return 0;
3988 }
3989
3990 static noinline_for_stack int relocate_block_group(struct reloc_control *rc)
3991 {
3992 struct rb_root blocks = RB_ROOT;
3993 struct btrfs_key key;
3994 struct btrfs_trans_handle *trans = NULL;
3995 struct btrfs_path *path;
3996 struct btrfs_extent_item *ei;
3997 u64 flags;
3998 u32 item_size;
3999 int ret;
4000 int err = 0;
4001 int progress = 0;
4002
4003 path = btrfs_alloc_path();
4004 if (!path)
4005 return -ENOMEM;
4006 path->reada = READA_FORWARD;
4007
4008 ret = prepare_to_relocate(rc);
4009 if (ret) {
4010 err = ret;
4011 goto out_free;
4012 }
4013
4014 while (1) {
4015 rc->reserved_bytes = 0;
4016 ret = btrfs_block_rsv_refill(rc->extent_root,
4017 rc->block_rsv, rc->block_rsv->size,
4018 BTRFS_RESERVE_FLUSH_ALL);
4019 if (ret) {
4020 err = ret;
4021 break;
4022 }
4023 progress++;
4024 trans = btrfs_start_transaction(rc->extent_root, 0);
4025 if (IS_ERR(trans)) {
4026 err = PTR_ERR(trans);
4027 trans = NULL;
4028 break;
4029 }
4030 restart:
4031 if (update_backref_cache(trans, &rc->backref_cache)) {
4032 btrfs_end_transaction(trans, rc->extent_root);
4033 continue;
4034 }
4035
4036 ret = find_next_extent(rc, path, &key);
4037 if (ret < 0)
4038 err = ret;
4039 if (ret != 0)
4040 break;
4041
4042 rc->extents_found++;
4043
4044 ei = btrfs_item_ptr(path->nodes[0], path->slots[0],
4045 struct btrfs_extent_item);
4046 item_size = btrfs_item_size_nr(path->nodes[0], path->slots[0]);
4047 if (item_size >= sizeof(*ei)) {
4048 flags = btrfs_extent_flags(path->nodes[0], ei);
4049 ret = check_extent_flags(flags);
4050 BUG_ON(ret);
4051
4052 } else {
4053 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
4054 u64 ref_owner;
4055 int path_change = 0;
4056
4057 BUG_ON(item_size !=
4058 sizeof(struct btrfs_extent_item_v0));
4059 ret = get_ref_objectid_v0(rc, path, &key, &ref_owner,
4060 &path_change);
4061 if (ret < 0) {
4062 err = ret;
4063 break;
4064 }
4065 if (ref_owner < BTRFS_FIRST_FREE_OBJECTID)
4066 flags = BTRFS_EXTENT_FLAG_TREE_BLOCK;
4067 else
4068 flags = BTRFS_EXTENT_FLAG_DATA;
4069
4070 if (path_change) {
4071 btrfs_release_path(path);
4072
4073 path->search_commit_root = 1;
4074 path->skip_locking = 1;
4075 ret = btrfs_search_slot(NULL, rc->extent_root,
4076 &key, path, 0, 0);
4077 if (ret < 0) {
4078 err = ret;
4079 break;
4080 }
4081 BUG_ON(ret > 0);
4082 }
4083 #else
4084 BUG();
4085 #endif
4086 }
4087
4088 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
4089 ret = add_tree_block(rc, &key, path, &blocks);
4090 } else if (rc->stage == UPDATE_DATA_PTRS &&
4091 (flags & BTRFS_EXTENT_FLAG_DATA)) {
4092 ret = add_data_references(rc, &key, path, &blocks);
4093 } else {
4094 btrfs_release_path(path);
4095 ret = 0;
4096 }
4097 if (ret < 0) {
4098 err = ret;
4099 break;
4100 }
4101
4102 if (!RB_EMPTY_ROOT(&blocks)) {
4103 ret = relocate_tree_blocks(trans, rc, &blocks);
4104 if (ret < 0) {
4105 /*
4106 * if we fail to relocate tree blocks, force to update
4107 * backref cache when committing transaction.
4108 */
4109 rc->backref_cache.last_trans = trans->transid - 1;
4110
4111 if (ret != -EAGAIN) {
4112 err = ret;
4113 break;
4114 }
4115 rc->extents_found--;
4116 rc->search_start = key.objectid;
4117 }
4118 }
4119
4120 btrfs_end_transaction_throttle(trans, rc->extent_root);
4121 btrfs_btree_balance_dirty(rc->extent_root);
4122 trans = NULL;
4123
4124 if (rc->stage == MOVE_DATA_EXTENTS &&
4125 (flags & BTRFS_EXTENT_FLAG_DATA)) {
4126 rc->found_file_extent = 1;
4127 ret = relocate_data_extent(rc->data_inode,
4128 &key, &rc->cluster);
4129 if (ret < 0) {
4130 err = ret;
4131 break;
4132 }
4133 }
4134 }
4135 if (trans && progress && err == -ENOSPC) {
4136 ret = btrfs_force_chunk_alloc(trans, rc->extent_root,
4137 rc->block_group->flags);
4138 if (ret == 1) {
4139 err = 0;
4140 progress = 0;
4141 goto restart;
4142 }
4143 }
4144
4145 btrfs_release_path(path);
4146 clear_extent_bits(&rc->processed_blocks, 0, (u64)-1, EXTENT_DIRTY);
4147
4148 if (trans) {
4149 btrfs_end_transaction_throttle(trans, rc->extent_root);
4150 btrfs_btree_balance_dirty(rc->extent_root);
4151 }
4152
4153 if (!err) {
4154 ret = relocate_file_extent_cluster(rc->data_inode,
4155 &rc->cluster);
4156 if (ret < 0)
4157 err = ret;
4158 }
4159
4160 rc->create_reloc_tree = 0;
4161 set_reloc_control(rc);
4162
4163 backref_cache_cleanup(&rc->backref_cache);
4164 btrfs_block_rsv_release(rc->extent_root, rc->block_rsv, (u64)-1);
4165
4166 err = prepare_to_merge(rc, err);
4167
4168 merge_reloc_roots(rc);
4169
4170 rc->merge_reloc_tree = 0;
4171 unset_reloc_control(rc);
4172 btrfs_block_rsv_release(rc->extent_root, rc->block_rsv, (u64)-1);
4173
4174 /* get rid of pinned extents */
4175 trans = btrfs_join_transaction(rc->extent_root);
4176 if (IS_ERR(trans)) {
4177 err = PTR_ERR(trans);
4178 goto out_free;
4179 }
4180 btrfs_commit_transaction(trans, rc->extent_root);
4181 out_free:
4182 btrfs_free_block_rsv(rc->extent_root, rc->block_rsv);
4183 btrfs_free_path(path);
4184 return err;
4185 }
4186
4187 static int __insert_orphan_inode(struct btrfs_trans_handle *trans,
4188 struct btrfs_root *root, u64 objectid)
4189 {
4190 struct btrfs_path *path;
4191 struct btrfs_inode_item *item;
4192 struct extent_buffer *leaf;
4193 int ret;
4194
4195 path = btrfs_alloc_path();
4196 if (!path)
4197 return -ENOMEM;
4198
4199 ret = btrfs_insert_empty_inode(trans, root, path, objectid);
4200 if (ret)
4201 goto out;
4202
4203 leaf = path->nodes[0];
4204 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_inode_item);
4205 memzero_extent_buffer(leaf, (unsigned long)item, sizeof(*item));
4206 btrfs_set_inode_generation(leaf, item, 1);
4207 btrfs_set_inode_size(leaf, item, 0);
4208 btrfs_set_inode_mode(leaf, item, S_IFREG | 0600);
4209 btrfs_set_inode_flags(leaf, item, BTRFS_INODE_NOCOMPRESS |
4210 BTRFS_INODE_PREALLOC);
4211 btrfs_mark_buffer_dirty(leaf);
4212 out:
4213 btrfs_free_path(path);
4214 return ret;
4215 }
4216
4217 /*
4218 * helper to create inode for data relocation.
4219 * the inode is in data relocation tree and its link count is 0
4220 */
4221 static noinline_for_stack
4222 struct inode *create_reloc_inode(struct btrfs_fs_info *fs_info,
4223 struct btrfs_block_group_cache *group)
4224 {
4225 struct inode *inode = NULL;
4226 struct btrfs_trans_handle *trans;
4227 struct btrfs_root *root;
4228 struct btrfs_key key;
4229 u64 objectid;
4230 int err = 0;
4231
4232 root = read_fs_root(fs_info, BTRFS_DATA_RELOC_TREE_OBJECTID);
4233 if (IS_ERR(root))
4234 return ERR_CAST(root);
4235
4236 trans = btrfs_start_transaction(root, 6);
4237 if (IS_ERR(trans))
4238 return ERR_CAST(trans);
4239
4240 err = btrfs_find_free_objectid(root, &objectid);
4241 if (err)
4242 goto out;
4243
4244 err = __insert_orphan_inode(trans, root, objectid);
4245 BUG_ON(err);
4246
4247 key.objectid = objectid;
4248 key.type = BTRFS_INODE_ITEM_KEY;
4249 key.offset = 0;
4250 inode = btrfs_iget(fs_info->sb, &key, root, NULL);
4251 BUG_ON(IS_ERR(inode) || is_bad_inode(inode));
4252 BTRFS_I(inode)->index_cnt = group->key.objectid;
4253
4254 err = btrfs_orphan_add(trans, inode);
4255 out:
4256 btrfs_end_transaction(trans, root);
4257 btrfs_btree_balance_dirty(root);
4258 if (err) {
4259 if (inode)
4260 iput(inode);
4261 inode = ERR_PTR(err);
4262 }
4263 return inode;
4264 }
4265
4266 static struct reloc_control *alloc_reloc_control(struct btrfs_fs_info *fs_info)
4267 {
4268 struct reloc_control *rc;
4269
4270 rc = kzalloc(sizeof(*rc), GFP_NOFS);
4271 if (!rc)
4272 return NULL;
4273
4274 INIT_LIST_HEAD(&rc->reloc_roots);
4275 backref_cache_init(&rc->backref_cache);
4276 mapping_tree_init(&rc->reloc_root_tree);
4277 extent_io_tree_init(&rc->processed_blocks,
4278 fs_info->btree_inode->i_mapping);
4279 return rc;
4280 }
4281
4282 /*
4283 * Print the block group being relocated
4284 */
4285 static void describe_relocation(struct btrfs_fs_info *fs_info,
4286 struct btrfs_block_group_cache *block_group)
4287 {
4288 char buf[128]; /* prefixed by a '|' that'll be dropped */
4289 u64 flags = block_group->flags;
4290
4291 /* Shouldn't happen */
4292 if (!flags) {
4293 strcpy(buf, "|NONE");
4294 } else {
4295 char *bp = buf;
4296
4297 #define DESCRIBE_FLAG(f, d) \
4298 if (flags & BTRFS_BLOCK_GROUP_##f) { \
4299 bp += snprintf(bp, buf - bp + sizeof(buf), "|%s", d); \
4300 flags &= ~BTRFS_BLOCK_GROUP_##f; \
4301 }
4302 DESCRIBE_FLAG(DATA, "data");
4303 DESCRIBE_FLAG(SYSTEM, "system");
4304 DESCRIBE_FLAG(METADATA, "metadata");
4305 DESCRIBE_FLAG(RAID0, "raid0");
4306 DESCRIBE_FLAG(RAID1, "raid1");
4307 DESCRIBE_FLAG(DUP, "dup");
4308 DESCRIBE_FLAG(RAID10, "raid10");
4309 DESCRIBE_FLAG(RAID5, "raid5");
4310 DESCRIBE_FLAG(RAID6, "raid6");
4311 if (flags)
4312 snprintf(buf, buf - bp + sizeof(buf), "|0x%llx", flags);
4313 #undef DESCRIBE_FLAG
4314 }
4315
4316 btrfs_info(fs_info,
4317 "relocating block group %llu flags %s",
4318 block_group->key.objectid, buf + 1);
4319 }
4320
4321 /*
4322 * function to relocate all extents in a block group.
4323 */
4324 int btrfs_relocate_block_group(struct btrfs_fs_info *fs_info, u64 group_start)
4325 {
4326 struct btrfs_root *extent_root = fs_info->extent_root;
4327 struct reloc_control *rc;
4328 struct inode *inode;
4329 struct btrfs_path *path;
4330 int ret;
4331 int rw = 0;
4332 int err = 0;
4333
4334 rc = alloc_reloc_control(fs_info);
4335 if (!rc)
4336 return -ENOMEM;
4337
4338 rc->extent_root = extent_root;
4339
4340 rc->block_group = btrfs_lookup_block_group(fs_info, group_start);
4341 BUG_ON(!rc->block_group);
4342
4343 ret = btrfs_inc_block_group_ro(extent_root, rc->block_group);
4344 if (ret) {
4345 err = ret;
4346 goto out;
4347 }
4348 rw = 1;
4349
4350 path = btrfs_alloc_path();
4351 if (!path) {
4352 err = -ENOMEM;
4353 goto out;
4354 }
4355
4356 inode = lookup_free_space_inode(fs_info->tree_root, rc->block_group,
4357 path);
4358 btrfs_free_path(path);
4359
4360 if (!IS_ERR(inode))
4361 ret = delete_block_group_cache(fs_info, rc->block_group, inode, 0);
4362 else
4363 ret = PTR_ERR(inode);
4364
4365 if (ret && ret != -ENOENT) {
4366 err = ret;
4367 goto out;
4368 }
4369
4370 rc->data_inode = create_reloc_inode(fs_info, rc->block_group);
4371 if (IS_ERR(rc->data_inode)) {
4372 err = PTR_ERR(rc->data_inode);
4373 rc->data_inode = NULL;
4374 goto out;
4375 }
4376
4377 describe_relocation(fs_info, rc->block_group);
4378
4379 btrfs_wait_block_group_reservations(rc->block_group);
4380 btrfs_wait_nocow_writers(rc->block_group);
4381 btrfs_wait_ordered_roots(fs_info, -1,
4382 rc->block_group->key.objectid,
4383 rc->block_group->key.offset);
4384
4385 while (1) {
4386 mutex_lock(&fs_info->cleaner_mutex);
4387 ret = relocate_block_group(rc);
4388 mutex_unlock(&fs_info->cleaner_mutex);
4389 if (ret < 0) {
4390 err = ret;
4391 goto out;
4392 }
4393
4394 if (rc->extents_found == 0)
4395 break;
4396
4397 btrfs_info(fs_info, "found %llu extents", rc->extents_found);
4398
4399 if (rc->stage == MOVE_DATA_EXTENTS && rc->found_file_extent) {
4400 ret = btrfs_wait_ordered_range(rc->data_inode, 0,
4401 (u64)-1);
4402 if (ret) {
4403 err = ret;
4404 goto out;
4405 }
4406 invalidate_mapping_pages(rc->data_inode->i_mapping,
4407 0, -1);
4408 rc->stage = UPDATE_DATA_PTRS;
4409 }
4410 }
4411
4412 WARN_ON(rc->block_group->pinned > 0);
4413 WARN_ON(rc->block_group->reserved > 0);
4414 WARN_ON(btrfs_block_group_used(&rc->block_group->item) > 0);
4415 out:
4416 if (err && rw)
4417 btrfs_dec_block_group_ro(extent_root, rc->block_group);
4418 iput(rc->data_inode);
4419 btrfs_put_block_group(rc->block_group);
4420 kfree(rc);
4421 return err;
4422 }
4423
4424 static noinline_for_stack int mark_garbage_root(struct btrfs_root *root)
4425 {
4426 struct btrfs_fs_info *fs_info = root->fs_info;
4427 struct btrfs_trans_handle *trans;
4428 int ret, err;
4429
4430 trans = btrfs_start_transaction(fs_info->tree_root, 0);
4431 if (IS_ERR(trans))
4432 return PTR_ERR(trans);
4433
4434 memset(&root->root_item.drop_progress, 0,
4435 sizeof(root->root_item.drop_progress));
4436 root->root_item.drop_level = 0;
4437 btrfs_set_root_refs(&root->root_item, 0);
4438 ret = btrfs_update_root(trans, fs_info->tree_root,
4439 &root->root_key, &root->root_item);
4440
4441 err = btrfs_end_transaction(trans, fs_info->tree_root);
4442 if (err)
4443 return err;
4444 return ret;
4445 }
4446
4447 /*
4448 * recover relocation interrupted by system crash.
4449 *
4450 * this function resumes merging reloc trees with corresponding fs trees.
4451 * this is important for keeping the sharing of tree blocks
4452 */
4453 int btrfs_recover_relocation(struct btrfs_root *root)
4454 {
4455 struct btrfs_fs_info *fs_info = root->fs_info;
4456 LIST_HEAD(reloc_roots);
4457 struct btrfs_key key;
4458 struct btrfs_root *fs_root;
4459 struct btrfs_root *reloc_root;
4460 struct btrfs_path *path;
4461 struct extent_buffer *leaf;
4462 struct reloc_control *rc = NULL;
4463 struct btrfs_trans_handle *trans;
4464 int ret;
4465 int err = 0;
4466
4467 path = btrfs_alloc_path();
4468 if (!path)
4469 return -ENOMEM;
4470 path->reada = READA_BACK;
4471
4472 key.objectid = BTRFS_TREE_RELOC_OBJECTID;
4473 key.type = BTRFS_ROOT_ITEM_KEY;
4474 key.offset = (u64)-1;
4475
4476 while (1) {
4477 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key,
4478 path, 0, 0);
4479 if (ret < 0) {
4480 err = ret;
4481 goto out;
4482 }
4483 if (ret > 0) {
4484 if (path->slots[0] == 0)
4485 break;
4486 path->slots[0]--;
4487 }
4488 leaf = path->nodes[0];
4489 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4490 btrfs_release_path(path);
4491
4492 if (key.objectid != BTRFS_TREE_RELOC_OBJECTID ||
4493 key.type != BTRFS_ROOT_ITEM_KEY)
4494 break;
4495
4496 reloc_root = btrfs_read_fs_root(root, &key);
4497 if (IS_ERR(reloc_root)) {
4498 err = PTR_ERR(reloc_root);
4499 goto out;
4500 }
4501
4502 list_add(&reloc_root->root_list, &reloc_roots);
4503
4504 if (btrfs_root_refs(&reloc_root->root_item) > 0) {
4505 fs_root = read_fs_root(fs_info,
4506 reloc_root->root_key.offset);
4507 if (IS_ERR(fs_root)) {
4508 ret = PTR_ERR(fs_root);
4509 if (ret != -ENOENT) {
4510 err = ret;
4511 goto out;
4512 }
4513 ret = mark_garbage_root(reloc_root);
4514 if (ret < 0) {
4515 err = ret;
4516 goto out;
4517 }
4518 }
4519 }
4520
4521 if (key.offset == 0)
4522 break;
4523
4524 key.offset--;
4525 }
4526 btrfs_release_path(path);
4527
4528 if (list_empty(&reloc_roots))
4529 goto out;
4530
4531 rc = alloc_reloc_control(fs_info);
4532 if (!rc) {
4533 err = -ENOMEM;
4534 goto out;
4535 }
4536
4537 rc->extent_root = fs_info->extent_root;
4538
4539 set_reloc_control(rc);
4540
4541 trans = btrfs_join_transaction(rc->extent_root);
4542 if (IS_ERR(trans)) {
4543 unset_reloc_control(rc);
4544 err = PTR_ERR(trans);
4545 goto out_free;
4546 }
4547
4548 rc->merge_reloc_tree = 1;
4549
4550 while (!list_empty(&reloc_roots)) {
4551 reloc_root = list_entry(reloc_roots.next,
4552 struct btrfs_root, root_list);
4553 list_del(&reloc_root->root_list);
4554
4555 if (btrfs_root_refs(&reloc_root->root_item) == 0) {
4556 list_add_tail(&reloc_root->root_list,
4557 &rc->reloc_roots);
4558 continue;
4559 }
4560
4561 fs_root = read_fs_root(fs_info, reloc_root->root_key.offset);
4562 if (IS_ERR(fs_root)) {
4563 err = PTR_ERR(fs_root);
4564 goto out_free;
4565 }
4566
4567 err = __add_reloc_root(reloc_root);
4568 BUG_ON(err < 0); /* -ENOMEM or logic error */
4569 fs_root->reloc_root = reloc_root;
4570 }
4571
4572 err = btrfs_commit_transaction(trans, rc->extent_root);
4573 if (err)
4574 goto out_free;
4575
4576 merge_reloc_roots(rc);
4577
4578 unset_reloc_control(rc);
4579
4580 trans = btrfs_join_transaction(rc->extent_root);
4581 if (IS_ERR(trans)) {
4582 err = PTR_ERR(trans);
4583 goto out_free;
4584 }
4585 err = btrfs_commit_transaction(trans, rc->extent_root);
4586 out_free:
4587 kfree(rc);
4588 out:
4589 if (!list_empty(&reloc_roots))
4590 free_reloc_roots(&reloc_roots);
4591
4592 btrfs_free_path(path);
4593
4594 if (err == 0) {
4595 /* cleanup orphan inode in data relocation tree */
4596 fs_root = read_fs_root(fs_info, BTRFS_DATA_RELOC_TREE_OBJECTID);
4597 if (IS_ERR(fs_root))
4598 err = PTR_ERR(fs_root);
4599 else
4600 err = btrfs_orphan_cleanup(fs_root);
4601 }
4602 return err;
4603 }
4604
4605 /*
4606 * helper to add ordered checksum for data relocation.
4607 *
4608 * cloning checksum properly handles the nodatasum extents.
4609 * it also saves CPU time to re-calculate the checksum.
4610 */
4611 int btrfs_reloc_clone_csums(struct inode *inode, u64 file_pos, u64 len)
4612 {
4613 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4614 struct btrfs_ordered_sum *sums;
4615 struct btrfs_ordered_extent *ordered;
4616 int ret;
4617 u64 disk_bytenr;
4618 u64 new_bytenr;
4619 LIST_HEAD(list);
4620
4621 ordered = btrfs_lookup_ordered_extent(inode, file_pos);
4622 BUG_ON(ordered->file_offset != file_pos || ordered->len != len);
4623
4624 disk_bytenr = file_pos + BTRFS_I(inode)->index_cnt;
4625 ret = btrfs_lookup_csums_range(fs_info->csum_root, disk_bytenr,
4626 disk_bytenr + len - 1, &list, 0);
4627 if (ret)
4628 goto out;
4629
4630 while (!list_empty(&list)) {
4631 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
4632 list_del_init(&sums->list);
4633
4634 /*
4635 * We need to offset the new_bytenr based on where the csum is.
4636 * We need to do this because we will read in entire prealloc
4637 * extents but we may have written to say the middle of the
4638 * prealloc extent, so we need to make sure the csum goes with
4639 * the right disk offset.
4640 *
4641 * We can do this because the data reloc inode refers strictly
4642 * to the on disk bytes, so we don't have to worry about
4643 * disk_len vs real len like with real inodes since it's all
4644 * disk length.
4645 */
4646 new_bytenr = ordered->start + (sums->bytenr - disk_bytenr);
4647 sums->bytenr = new_bytenr;
4648
4649 btrfs_add_ordered_sum(inode, ordered, sums);
4650 }
4651 out:
4652 btrfs_put_ordered_extent(ordered);
4653 return ret;
4654 }
4655
4656 int btrfs_reloc_cow_block(struct btrfs_trans_handle *trans,
4657 struct btrfs_root *root, struct extent_buffer *buf,
4658 struct extent_buffer *cow)
4659 {
4660 struct btrfs_fs_info *fs_info = root->fs_info;
4661 struct reloc_control *rc;
4662 struct backref_node *node;
4663 int first_cow = 0;
4664 int level;
4665 int ret = 0;
4666
4667 rc = fs_info->reloc_ctl;
4668 if (!rc)
4669 return 0;
4670
4671 BUG_ON(rc->stage == UPDATE_DATA_PTRS &&
4672 root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID);
4673
4674 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) {
4675 if (buf == root->node)
4676 __update_reloc_root(root, cow->start);
4677 }
4678
4679 level = btrfs_header_level(buf);
4680 if (btrfs_header_generation(buf) <=
4681 btrfs_root_last_snapshot(&root->root_item))
4682 first_cow = 1;
4683
4684 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID &&
4685 rc->create_reloc_tree) {
4686 WARN_ON(!first_cow && level == 0);
4687
4688 node = rc->backref_cache.path[level];
4689 BUG_ON(node->bytenr != buf->start &&
4690 node->new_bytenr != buf->start);
4691
4692 drop_node_buffer(node);
4693 extent_buffer_get(cow);
4694 node->eb = cow;
4695 node->new_bytenr = cow->start;
4696
4697 if (!node->pending) {
4698 list_move_tail(&node->list,
4699 &rc->backref_cache.pending[level]);
4700 node->pending = 1;
4701 }
4702
4703 if (first_cow)
4704 __mark_block_processed(rc, node);
4705
4706 if (first_cow && level > 0)
4707 rc->nodes_relocated += buf->len;
4708 }
4709
4710 if (level == 0 && first_cow && rc->stage == UPDATE_DATA_PTRS)
4711 ret = replace_file_extents(trans, rc, root, cow);
4712 return ret;
4713 }
4714
4715 /*
4716 * called before creating snapshot. it calculates metadata reservation
4717 * required for relocating tree blocks in the snapshot
4718 */
4719 void btrfs_reloc_pre_snapshot(struct btrfs_pending_snapshot *pending,
4720 u64 *bytes_to_reserve)
4721 {
4722 struct btrfs_root *root;
4723 struct reloc_control *rc;
4724
4725 root = pending->root;
4726 if (!root->reloc_root)
4727 return;
4728
4729 rc = root->fs_info->reloc_ctl;
4730 if (!rc->merge_reloc_tree)
4731 return;
4732
4733 root = root->reloc_root;
4734 BUG_ON(btrfs_root_refs(&root->root_item) == 0);
4735 /*
4736 * relocation is in the stage of merging trees. the space
4737 * used by merging a reloc tree is twice the size of
4738 * relocated tree nodes in the worst case. half for cowing
4739 * the reloc tree, half for cowing the fs tree. the space
4740 * used by cowing the reloc tree will be freed after the
4741 * tree is dropped. if we create snapshot, cowing the fs
4742 * tree may use more space than it frees. so we need
4743 * reserve extra space.
4744 */
4745 *bytes_to_reserve += rc->nodes_relocated;
4746 }
4747
4748 /*
4749 * called after snapshot is created. migrate block reservation
4750 * and create reloc root for the newly created snapshot
4751 */
4752 int btrfs_reloc_post_snapshot(struct btrfs_trans_handle *trans,
4753 struct btrfs_pending_snapshot *pending)
4754 {
4755 struct btrfs_root *root = pending->root;
4756 struct btrfs_root *reloc_root;
4757 struct btrfs_root *new_root;
4758 struct reloc_control *rc;
4759 int ret;
4760
4761 if (!root->reloc_root)
4762 return 0;
4763
4764 rc = root->fs_info->reloc_ctl;
4765 rc->merging_rsv_size += rc->nodes_relocated;
4766
4767 if (rc->merge_reloc_tree) {
4768 ret = btrfs_block_rsv_migrate(&pending->block_rsv,
4769 rc->block_rsv,
4770 rc->nodes_relocated, 1);
4771 if (ret)
4772 return ret;
4773 }
4774
4775 new_root = pending->snap;
4776 reloc_root = create_reloc_root(trans, root->reloc_root,
4777 new_root->root_key.objectid);
4778 if (IS_ERR(reloc_root))
4779 return PTR_ERR(reloc_root);
4780
4781 ret = __add_reloc_root(reloc_root);
4782 BUG_ON(ret < 0);
4783 new_root->reloc_root = reloc_root;
4784
4785 if (rc->create_reloc_tree)
4786 ret = clone_backref_node(trans, rc, root, reloc_root);
4787 return ret;
4788 }
Ubuntu Artful kernel mirror