]> git.proxmox.com Git - mirror_ubuntu-jammy-kernel.git/blob - fs/btrfs/relocation.c
projects / mirror_ubuntu-jammy-kernel.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
Merge tag 'asoc-fix-v5.6-rc2' of https://git.kernel.org/pub/scm/linux/kernel/git...
[mirror_ubuntu-jammy-kernel.git] / fs / btrfs / relocation.c
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3 * Copyright (C) 2009 Oracle. All rights reserved.
4 */
5
6 #include <linux/sched.h>
7 #include <linux/pagemap.h>
8 #include <linux/writeback.h>
9 #include <linux/blkdev.h>
10 #include <linux/rbtree.h>
11 #include <linux/slab.h>
12 #include "ctree.h"
13 #include "disk-io.h"
14 #include "transaction.h"
15 #include "volumes.h"
16 #include "locking.h"
17 #include "btrfs_inode.h"
18 #include "async-thread.h"
19 #include "free-space-cache.h"
20 #include "inode-map.h"
21 #include "qgroup.h"
22 #include "print-tree.h"
23 #include "delalloc-space.h"
24 #include "block-group.h"
25
26 /*
27 * backref_node, mapping_node and tree_block start with this
28 */
29 struct tree_entry {
30 struct rb_node rb_node;
31 u64 bytenr;
32 };
33
34 /*
35 * present a tree block in the backref cache
36 */
37 struct backref_node {
38 struct rb_node rb_node;
39 u64 bytenr;
40
41 u64 new_bytenr;
42 /* objectid of tree block owner, can be not uptodate */
43 u64 owner;
44 /* link to pending, changed or detached list */
45 struct list_head list;
46 /* list of upper level blocks reference this block */
47 struct list_head upper;
48 /* list of child blocks in the cache */
49 struct list_head lower;
50 /* NULL if this node is not tree root */
51 struct btrfs_root *root;
52 /* extent buffer got by COW the block */
53 struct extent_buffer *eb;
54 /* level of tree block */
55 unsigned int level:8;
56 /* is the block in non-reference counted tree */
57 unsigned int cowonly:1;
58 /* 1 if no child node in the cache */
59 unsigned int lowest:1;
60 /* is the extent buffer locked */
61 unsigned int locked:1;
62 /* has the block been processed */
63 unsigned int processed:1;
64 /* have backrefs of this block been checked */
65 unsigned int checked:1;
66 /*
67 * 1 if corresponding block has been cowed but some upper
68 * level block pointers may not point to the new location
69 */
70 unsigned int pending:1;
71 /*
72 * 1 if the backref node isn't connected to any other
73 * backref node.
74 */
75 unsigned int detached:1;
76 };
77
78 /*
79 * present a block pointer in the backref cache
80 */
81 struct backref_edge {
82 struct list_head list[2];
83 struct backref_node *node[2];
84 };
85
86 #define LOWER 0
87 #define UPPER 1
88 #define RELOCATION_RESERVED_NODES 256
89
90 struct backref_cache {
91 /* red black tree of all backref nodes in the cache */
92 struct rb_root rb_root;
93 /* for passing backref nodes to btrfs_reloc_cow_block */
94 struct backref_node *path[BTRFS_MAX_LEVEL];
95 /*
96 * list of blocks that have been cowed but some block
97 * pointers in upper level blocks may not reflect the
98 * new location
99 */
100 struct list_head pending[BTRFS_MAX_LEVEL];
101 /* list of backref nodes with no child node */
102 struct list_head leaves;
103 /* list of blocks that have been cowed in current transaction */
104 struct list_head changed;
105 /* list of detached backref node. */
106 struct list_head detached;
107
108 u64 last_trans;
109
110 int nr_nodes;
111 int nr_edges;
112 };
113
114 /*
115 * map address of tree root to tree
116 */
117 struct mapping_node {
118 struct rb_node rb_node;
119 u64 bytenr;
120 void *data;
121 };
122
123 struct mapping_tree {
124 struct rb_root rb_root;
125 spinlock_t lock;
126 };
127
128 /*
129 * present a tree block to process
130 */
131 struct tree_block {
132 struct rb_node rb_node;
133 u64 bytenr;
134 struct btrfs_key key;
135 unsigned int level:8;
136 unsigned int key_ready:1;
137 };
138
139 #define MAX_EXTENTS 128
140
141 struct file_extent_cluster {
142 u64 start;
143 u64 end;
144 u64 boundary[MAX_EXTENTS];
145 unsigned int nr;
146 };
147
148 struct reloc_control {
149 /* block group to relocate */
150 struct btrfs_block_group *block_group;
151 /* extent tree */
152 struct btrfs_root *extent_root;
153 /* inode for moving data */
154 struct inode *data_inode;
155
156 struct btrfs_block_rsv *block_rsv;
157
158 struct backref_cache backref_cache;
159
160 struct file_extent_cluster cluster;
161 /* tree blocks have been processed */
162 struct extent_io_tree processed_blocks;
163 /* map start of tree root to corresponding reloc tree */
164 struct mapping_tree reloc_root_tree;
165 /* list of reloc trees */
166 struct list_head reloc_roots;
167 /* list of subvolume trees that get relocated */
168 struct list_head dirty_subvol_roots;
169 /* size of metadata reservation for merging reloc trees */
170 u64 merging_rsv_size;
171 /* size of relocated tree nodes */
172 u64 nodes_relocated;
173 /* reserved size for block group relocation*/
174 u64 reserved_bytes;
175
176 u64 search_start;
177 u64 extents_found;
178
179 unsigned int stage:8;
180 unsigned int create_reloc_tree:1;
181 unsigned int merge_reloc_tree:1;
182 unsigned int found_file_extent:1;
183 };
184
185 /* stages of data relocation */
186 #define MOVE_DATA_EXTENTS 0
187 #define UPDATE_DATA_PTRS 1
188
189 static void remove_backref_node(struct backref_cache *cache,
190 struct backref_node *node);
191 static void __mark_block_processed(struct reloc_control *rc,
192 struct backref_node *node);
193
194 static void mapping_tree_init(struct mapping_tree *tree)
195 {
196 tree->rb_root = RB_ROOT;
197 spin_lock_init(&tree->lock);
198 }
199
200 static void backref_cache_init(struct backref_cache *cache)
201 {
202 int i;
203 cache->rb_root = RB_ROOT;
204 for (i = 0; i < BTRFS_MAX_LEVEL; i++)
205 INIT_LIST_HEAD(&cache->pending[i]);
206 INIT_LIST_HEAD(&cache->changed);
207 INIT_LIST_HEAD(&cache->detached);
208 INIT_LIST_HEAD(&cache->leaves);
209 }
210
211 static void backref_cache_cleanup(struct backref_cache *cache)
212 {
213 struct backref_node *node;
214 int i;
215
216 while (!list_empty(&cache->detached)) {
217 node = list_entry(cache->detached.next,
218 struct backref_node, list);
219 remove_backref_node(cache, node);
220 }
221
222 while (!list_empty(&cache->leaves)) {
223 node = list_entry(cache->leaves.next,
224 struct backref_node, lower);
225 remove_backref_node(cache, node);
226 }
227
228 cache->last_trans = 0;
229
230 for (i = 0; i < BTRFS_MAX_LEVEL; i++)
231 ASSERT(list_empty(&cache->pending[i]));
232 ASSERT(list_empty(&cache->changed));
233 ASSERT(list_empty(&cache->detached));
234 ASSERT(RB_EMPTY_ROOT(&cache->rb_root));
235 ASSERT(!cache->nr_nodes);
236 ASSERT(!cache->nr_edges);
237 }
238
239 static struct backref_node *alloc_backref_node(struct backref_cache *cache)
240 {
241 struct backref_node *node;
242
243 node = kzalloc(sizeof(*node), GFP_NOFS);
244 if (node) {
245 INIT_LIST_HEAD(&node->list);
246 INIT_LIST_HEAD(&node->upper);
247 INIT_LIST_HEAD(&node->lower);
248 RB_CLEAR_NODE(&node->rb_node);
249 cache->nr_nodes++;
250 }
251 return node;
252 }
253
254 static void free_backref_node(struct backref_cache *cache,
255 struct backref_node *node)
256 {
257 if (node) {
258 cache->nr_nodes--;
259 kfree(node);
260 }
261 }
262
263 static struct backref_edge *alloc_backref_edge(struct backref_cache *cache)
264 {
265 struct backref_edge *edge;
266
267 edge = kzalloc(sizeof(*edge), GFP_NOFS);
268 if (edge)
269 cache->nr_edges++;
270 return edge;
271 }
272
273 static void free_backref_edge(struct backref_cache *cache,
274 struct backref_edge *edge)
275 {
276 if (edge) {
277 cache->nr_edges--;
278 kfree(edge);
279 }
280 }
281
282 static struct rb_node *tree_insert(struct rb_root *root, u64 bytenr,
283 struct rb_node *node)
284 {
285 struct rb_node **p = &root->rb_node;
286 struct rb_node *parent = NULL;
287 struct tree_entry *entry;
288
289 while (*p) {
290 parent = *p;
291 entry = rb_entry(parent, struct tree_entry, rb_node);
292
293 if (bytenr < entry->bytenr)
294 p = &(*p)->rb_left;
295 else if (bytenr > entry->bytenr)
296 p = &(*p)->rb_right;
297 else
298 return parent;
299 }
300
301 rb_link_node(node, parent, p);
302 rb_insert_color(node, root);
303 return NULL;
304 }
305
306 static struct rb_node *tree_search(struct rb_root *root, u64 bytenr)
307 {
308 struct rb_node *n = root->rb_node;
309 struct tree_entry *entry;
310
311 while (n) {
312 entry = rb_entry(n, struct tree_entry, rb_node);
313
314 if (bytenr < entry->bytenr)
315 n = n->rb_left;
316 else if (bytenr > entry->bytenr)
317 n = n->rb_right;
318 else
319 return n;
320 }
321 return NULL;
322 }
323
324 static void backref_tree_panic(struct rb_node *rb_node, int errno, u64 bytenr)
325 {
326
327 struct btrfs_fs_info *fs_info = NULL;
328 struct backref_node *bnode = rb_entry(rb_node, struct backref_node,
329 rb_node);
330 if (bnode->root)
331 fs_info = bnode->root->fs_info;
332 btrfs_panic(fs_info, errno,
333 "Inconsistency in backref cache found at offset %llu",
334 bytenr);
335 }
336
337 /*
338 * walk up backref nodes until reach node presents tree root
339 */
340 static struct backref_node *walk_up_backref(struct backref_node *node,
341 struct backref_edge *edges[],
342 int *index)
343 {
344 struct backref_edge *edge;
345 int idx = *index;
346
347 while (!list_empty(&node->upper)) {
348 edge = list_entry(node->upper.next,
349 struct backref_edge, list[LOWER]);
350 edges[idx++] = edge;
351 node = edge->node[UPPER];
352 }
353 BUG_ON(node->detached);
354 *index = idx;
355 return node;
356 }
357
358 /*
359 * walk down backref nodes to find start of next reference path
360 */
361 static struct backref_node *walk_down_backref(struct backref_edge *edges[],
362 int *index)
363 {
364 struct backref_edge *edge;
365 struct backref_node *lower;
366 int idx = *index;
367
368 while (idx > 0) {
369 edge = edges[idx - 1];
370 lower = edge->node[LOWER];
371 if (list_is_last(&edge->list[LOWER], &lower->upper)) {
372 idx--;
373 continue;
374 }
375 edge = list_entry(edge->list[LOWER].next,
376 struct backref_edge, list[LOWER]);
377 edges[idx - 1] = edge;
378 *index = idx;
379 return edge->node[UPPER];
380 }
381 *index = 0;
382 return NULL;
383 }
384
385 static void unlock_node_buffer(struct backref_node *node)
386 {
387 if (node->locked) {
388 btrfs_tree_unlock(node->eb);
389 node->locked = 0;
390 }
391 }
392
393 static void drop_node_buffer(struct backref_node *node)
394 {
395 if (node->eb) {
396 unlock_node_buffer(node);
397 free_extent_buffer(node->eb);
398 node->eb = NULL;
399 }
400 }
401
402 static void drop_backref_node(struct backref_cache *tree,
403 struct backref_node *node)
404 {
405 BUG_ON(!list_empty(&node->upper));
406
407 drop_node_buffer(node);
408 list_del(&node->list);
409 list_del(&node->lower);
410 if (!RB_EMPTY_NODE(&node->rb_node))
411 rb_erase(&node->rb_node, &tree->rb_root);
412 free_backref_node(tree, node);
413 }
414
415 /*
416 * remove a backref node from the backref cache
417 */
418 static void remove_backref_node(struct backref_cache *cache,
419 struct backref_node *node)
420 {
421 struct backref_node *upper;
422 struct backref_edge *edge;
423
424 if (!node)
425 return;
426
427 BUG_ON(!node->lowest && !node->detached);
428 while (!list_empty(&node->upper)) {
429 edge = list_entry(node->upper.next, struct backref_edge,
430 list[LOWER]);
431 upper = edge->node[UPPER];
432 list_del(&edge->list[LOWER]);
433 list_del(&edge->list[UPPER]);
434 free_backref_edge(cache, edge);
435
436 if (RB_EMPTY_NODE(&upper->rb_node)) {
437 BUG_ON(!list_empty(&node->upper));
438 drop_backref_node(cache, node);
439 node = upper;
440 node->lowest = 1;
441 continue;
442 }
443 /*
444 * add the node to leaf node list if no other
445 * child block cached.
446 */
447 if (list_empty(&upper->lower)) {
448 list_add_tail(&upper->lower, &cache->leaves);
449 upper->lowest = 1;
450 }
451 }
452
453 drop_backref_node(cache, node);
454 }
455
456 static void update_backref_node(struct backref_cache *cache,
457 struct backref_node *node, u64 bytenr)
458 {
459 struct rb_node *rb_node;
460 rb_erase(&node->rb_node, &cache->rb_root);
461 node->bytenr = bytenr;
462 rb_node = tree_insert(&cache->rb_root, node->bytenr, &node->rb_node);
463 if (rb_node)
464 backref_tree_panic(rb_node, -EEXIST, bytenr);
465 }
466
467 /*
468 * update backref cache after a transaction commit
469 */
470 static int update_backref_cache(struct btrfs_trans_handle *trans,
471 struct backref_cache *cache)
472 {
473 struct backref_node *node;
474 int level = 0;
475
476 if (cache->last_trans == 0) {
477 cache->last_trans = trans->transid;
478 return 0;
479 }
480
481 if (cache->last_trans == trans->transid)
482 return 0;
483
484 /*
485 * detached nodes are used to avoid unnecessary backref
486 * lookup. transaction commit changes the extent tree.
487 * so the detached nodes are no longer useful.
488 */
489 while (!list_empty(&cache->detached)) {
490 node = list_entry(cache->detached.next,
491 struct backref_node, list);
492 remove_backref_node(cache, node);
493 }
494
495 while (!list_empty(&cache->changed)) {
496 node = list_entry(cache->changed.next,
497 struct backref_node, list);
498 list_del_init(&node->list);
499 BUG_ON(node->pending);
500 update_backref_node(cache, node, node->new_bytenr);
501 }
502
503 /*
504 * some nodes can be left in the pending list if there were
505 * errors during processing the pending nodes.
506 */
507 for (level = 0; level < BTRFS_MAX_LEVEL; level++) {
508 list_for_each_entry(node, &cache->pending[level], list) {
509 BUG_ON(!node->pending);
510 if (node->bytenr == node->new_bytenr)
511 continue;
512 update_backref_node(cache, node, node->new_bytenr);
513 }
514 }
515
516 cache->last_trans = 0;
517 return 1;
518 }
519
520 static bool reloc_root_is_dead(struct btrfs_root *root)
521 {
522 /*
523 * Pair with set_bit/clear_bit in clean_dirty_subvols and
524 * btrfs_update_reloc_root. We need to see the updated bit before
525 * trying to access reloc_root
526 */
527 smp_rmb();
528 if (test_bit(BTRFS_ROOT_DEAD_RELOC_TREE, &root->state))
529 return true;
530 return false;
531 }
532
533 /*
534 * Check if this subvolume tree has valid reloc tree.
535 *
536 * Reloc tree after swap is considered dead, thus not considered as valid.
537 * This is enough for most callers, as they don't distinguish dead reloc root
538 * from no reloc root. But should_ignore_root() below is a special case.
539 */
540 static bool have_reloc_root(struct btrfs_root *root)
541 {
542 if (reloc_root_is_dead(root))
543 return false;
544 if (!root->reloc_root)
545 return false;
546 return true;
547 }
548
549 static int should_ignore_root(struct btrfs_root *root)
550 {
551 struct btrfs_root *reloc_root;
552
553 if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state))
554 return 0;
555
556 /* This root has been merged with its reloc tree, we can ignore it */
557 if (reloc_root_is_dead(root))
558 return 1;
559
560 reloc_root = root->reloc_root;
561 if (!reloc_root)
562 return 0;
563
564 if (btrfs_root_last_snapshot(&reloc_root->root_item) ==
565 root->fs_info->running_transaction->transid - 1)
566 return 0;
567 /*
568 * if there is reloc tree and it was created in previous
569 * transaction backref lookup can find the reloc tree,
570 * so backref node for the fs tree root is useless for
571 * relocation.
572 */
573 return 1;
574 }
575 /*
576 * find reloc tree by address of tree root
577 */
578 static struct btrfs_root *find_reloc_root(struct reloc_control *rc,
579 u64 bytenr)
580 {
581 struct rb_node *rb_node;
582 struct mapping_node *node;
583 struct btrfs_root *root = NULL;
584
585 spin_lock(&rc->reloc_root_tree.lock);
586 rb_node = tree_search(&rc->reloc_root_tree.rb_root, bytenr);
587 if (rb_node) {
588 node = rb_entry(rb_node, struct mapping_node, rb_node);
589 root = (struct btrfs_root *)node->data;
590 }
591 spin_unlock(&rc->reloc_root_tree.lock);
592 return root;
593 }
594
595 static int is_cowonly_root(u64 root_objectid)
596 {
597 if (root_objectid == BTRFS_ROOT_TREE_OBJECTID ||
598 root_objectid == BTRFS_EXTENT_TREE_OBJECTID ||
599 root_objectid == BTRFS_CHUNK_TREE_OBJECTID ||
600 root_objectid == BTRFS_DEV_TREE_OBJECTID ||
601 root_objectid == BTRFS_TREE_LOG_OBJECTID ||
602 root_objectid == BTRFS_CSUM_TREE_OBJECTID ||
603 root_objectid == BTRFS_UUID_TREE_OBJECTID ||
604 root_objectid == BTRFS_QUOTA_TREE_OBJECTID ||
605 root_objectid == BTRFS_FREE_SPACE_TREE_OBJECTID)
606 return 1;
607 return 0;
608 }
609
610 static struct btrfs_root *read_fs_root(struct btrfs_fs_info *fs_info,
611 u64 root_objectid)
612 {
613 struct btrfs_key key;
614
615 key.objectid = root_objectid;
616 key.type = BTRFS_ROOT_ITEM_KEY;
617 if (is_cowonly_root(root_objectid))
618 key.offset = 0;
619 else
620 key.offset = (u64)-1;
621
622 return btrfs_get_fs_root(fs_info, &key, false);
623 }
624
625 static noinline_for_stack
626 int find_inline_backref(struct extent_buffer *leaf, int slot,
627 unsigned long *ptr, unsigned long *end)
628 {
629 struct btrfs_key key;
630 struct btrfs_extent_item *ei;
631 struct btrfs_tree_block_info *bi;
632 u32 item_size;
633
634 btrfs_item_key_to_cpu(leaf, &key, slot);
635
636 item_size = btrfs_item_size_nr(leaf, slot);
637 if (item_size < sizeof(*ei)) {
638 btrfs_print_v0_err(leaf->fs_info);
639 btrfs_handle_fs_error(leaf->fs_info, -EINVAL, NULL);
640 return 1;
641 }
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; /* For searching extent root */
688 struct btrfs_path *path2; /* For searching parent of TREE_BLOCK_REF */
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); /* Pending edge list, upper node needs to be checked */
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 int type;
803 iref = (struct btrfs_extent_inline_ref *)ptr;
804 type = btrfs_get_extent_inline_ref_type(eb, iref,
805 BTRFS_REF_TYPE_BLOCK);
806 if (type == BTRFS_REF_TYPE_INVALID) {
807 err = -EUCLEAN;
808 goto out;
809 }
810 key.type = type;
811 key.offset = btrfs_extent_inline_ref_offset(eb, iref);
812
813 WARN_ON(key.type != BTRFS_TREE_BLOCK_REF_KEY &&
814 key.type != BTRFS_SHARED_BLOCK_REF_KEY);
815 }
816
817 /*
818 * Parent node found and matches current inline ref, no need to
819 * rebuild this node for this inline ref.
820 */
821 if (exist &&
822 ((key.type == BTRFS_TREE_BLOCK_REF_KEY &&
823 exist->owner == key.offset) ||
824 (key.type == BTRFS_SHARED_BLOCK_REF_KEY &&
825 exist->bytenr == key.offset))) {
826 exist = NULL;
827 goto next;
828 }
829
830 /* SHARED_BLOCK_REF means key.offset is the parent bytenr */
831 if (key.type == BTRFS_SHARED_BLOCK_REF_KEY) {
832 if (key.objectid == key.offset) {
833 /*
834 * Only root blocks of reloc trees use backref
835 * pointing to itself.
836 */
837 root = find_reloc_root(rc, cur->bytenr);
838 ASSERT(root);
839 cur->root = root;
840 break;
841 }
842
843 edge = alloc_backref_edge(cache);
844 if (!edge) {
845 err = -ENOMEM;
846 goto out;
847 }
848 rb_node = tree_search(&cache->rb_root, key.offset);
849 if (!rb_node) {
850 upper = alloc_backref_node(cache);
851 if (!upper) {
852 free_backref_edge(cache, edge);
853 err = -ENOMEM;
854 goto out;
855 }
856 upper->bytenr = key.offset;
857 upper->level = cur->level + 1;
858 /*
859 * backrefs for the upper level block isn't
860 * cached, add the block to pending list
861 */
862 list_add_tail(&edge->list[UPPER], &list);
863 } else {
864 upper = rb_entry(rb_node, struct backref_node,
865 rb_node);
866 ASSERT(upper->checked);
867 INIT_LIST_HEAD(&edge->list[UPPER]);
868 }
869 list_add_tail(&edge->list[LOWER], &cur->upper);
870 edge->node[LOWER] = cur;
871 edge->node[UPPER] = upper;
872
873 goto next;
874 } else if (unlikely(key.type == BTRFS_EXTENT_REF_V0_KEY)) {
875 err = -EINVAL;
876 btrfs_print_v0_err(rc->extent_root->fs_info);
877 btrfs_handle_fs_error(rc->extent_root->fs_info, err,
878 NULL);
879 goto out;
880 } else if (key.type != BTRFS_TREE_BLOCK_REF_KEY) {
881 goto next;
882 }
883
884 /*
885 * key.type == BTRFS_TREE_BLOCK_REF_KEY, inline ref offset
886 * means the root objectid. We need to search the tree to get
887 * its parent bytenr.
888 */
889 root = read_fs_root(rc->extent_root->fs_info, key.offset);
890 if (IS_ERR(root)) {
891 err = PTR_ERR(root);
892 goto out;
893 }
894
895 if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state))
896 cur->cowonly = 1;
897
898 if (btrfs_root_level(&root->root_item) == cur->level) {
899 /* tree root */
900 ASSERT(btrfs_root_bytenr(&root->root_item) ==
901 cur->bytenr);
902 if (should_ignore_root(root))
903 list_add(&cur->list, &useless);
904 else
905 cur->root = root;
906 break;
907 }
908
909 level = cur->level + 1;
910
911 /* Search the tree to find parent blocks referring the block. */
912 path2->search_commit_root = 1;
913 path2->skip_locking = 1;
914 path2->lowest_level = level;
915 ret = btrfs_search_slot(NULL, root, node_key, path2, 0, 0);
916 path2->lowest_level = 0;
917 if (ret < 0) {
918 err = ret;
919 goto out;
920 }
921 if (ret > 0 && path2->slots[level] > 0)
922 path2->slots[level]--;
923
924 eb = path2->nodes[level];
925 if (btrfs_node_blockptr(eb, path2->slots[level]) !=
926 cur->bytenr) {
927 btrfs_err(root->fs_info,
928 "couldn't find block (%llu) (level %d) in tree (%llu) with key (%llu %u %llu)",
929 cur->bytenr, level - 1,
930 root->root_key.objectid,
931 node_key->objectid, node_key->type,
932 node_key->offset);
933 err = -ENOENT;
934 goto out;
935 }
936 lower = cur;
937 need_check = true;
938
939 /* Add all nodes and edges in the path */
940 for (; level < BTRFS_MAX_LEVEL; level++) {
941 if (!path2->nodes[level]) {
942 ASSERT(btrfs_root_bytenr(&root->root_item) ==
943 lower->bytenr);
944 if (should_ignore_root(root))
945 list_add(&lower->list, &useless);
946 else
947 lower->root = root;
948 break;
949 }
950
951 edge = alloc_backref_edge(cache);
952 if (!edge) {
953 err = -ENOMEM;
954 goto out;
955 }
956
957 eb = path2->nodes[level];
958 rb_node = tree_search(&cache->rb_root, eb->start);
959 if (!rb_node) {
960 upper = alloc_backref_node(cache);
961 if (!upper) {
962 free_backref_edge(cache, edge);
963 err = -ENOMEM;
964 goto out;
965 }
966 upper->bytenr = eb->start;
967 upper->owner = btrfs_header_owner(eb);
968 upper->level = lower->level + 1;
969 if (!test_bit(BTRFS_ROOT_REF_COWS,
970 &root->state))
971 upper->cowonly = 1;
972
973 /*
974 * if we know the block isn't shared
975 * we can void checking its backrefs.
976 */
977 if (btrfs_block_can_be_shared(root, eb))
978 upper->checked = 0;
979 else
980 upper->checked = 1;
981
982 /*
983 * add the block to pending list if we
984 * need check its backrefs, we only do this once
985 * while walking up a tree as we will catch
986 * anything else later on.
987 */
988 if (!upper->checked && need_check) {
989 need_check = false;
990 list_add_tail(&edge->list[UPPER],
991 &list);
992 } else {
993 if (upper->checked)
994 need_check = true;
995 INIT_LIST_HEAD(&edge->list[UPPER]);
996 }
997 } else {
998 upper = rb_entry(rb_node, struct backref_node,
999 rb_node);
1000 ASSERT(upper->checked);
1001 INIT_LIST_HEAD(&edge->list[UPPER]);
1002 if (!upper->owner)
1003 upper->owner = btrfs_header_owner(eb);
1004 }
1005 list_add_tail(&edge->list[LOWER], &lower->upper);
1006 edge->node[LOWER] = lower;
1007 edge->node[UPPER] = upper;
1008
1009 if (rb_node)
1010 break;
1011 lower = upper;
1012 upper = NULL;
1013 }
1014 btrfs_release_path(path2);
1015 next:
1016 if (ptr < end) {
1017 ptr += btrfs_extent_inline_ref_size(key.type);
1018 if (ptr >= end) {
1019 WARN_ON(ptr > end);
1020 ptr = 0;
1021 end = 0;
1022 }
1023 }
1024 if (ptr >= end)
1025 path1->slots[0]++;
1026 }
1027 btrfs_release_path(path1);
1028
1029 cur->checked = 1;
1030 WARN_ON(exist);
1031
1032 /* the pending list isn't empty, take the first block to process */
1033 if (!list_empty(&list)) {
1034 edge = list_entry(list.next, struct backref_edge, list[UPPER]);
1035 list_del_init(&edge->list[UPPER]);
1036 cur = edge->node[UPPER];
1037 goto again;
1038 }
1039
1040 /*
1041 * everything goes well, connect backref nodes and insert backref nodes
1042 * into the cache.
1043 */
1044 ASSERT(node->checked);
1045 cowonly = node->cowonly;
1046 if (!cowonly) {
1047 rb_node = tree_insert(&cache->rb_root, node->bytenr,
1048 &node->rb_node);
1049 if (rb_node)
1050 backref_tree_panic(rb_node, -EEXIST, node->bytenr);
1051 list_add_tail(&node->lower, &cache->leaves);
1052 }
1053
1054 list_for_each_entry(edge, &node->upper, list[LOWER])
1055 list_add_tail(&edge->list[UPPER], &list);
1056
1057 while (!list_empty(&list)) {
1058 edge = list_entry(list.next, struct backref_edge, list[UPPER]);
1059 list_del_init(&edge->list[UPPER]);
1060 upper = edge->node[UPPER];
1061 if (upper->detached) {
1062 list_del(&edge->list[LOWER]);
1063 lower = edge->node[LOWER];
1064 free_backref_edge(cache, edge);
1065 if (list_empty(&lower->upper))
1066 list_add(&lower->list, &useless);
1067 continue;
1068 }
1069
1070 if (!RB_EMPTY_NODE(&upper->rb_node)) {
1071 if (upper->lowest) {
1072 list_del_init(&upper->lower);
1073 upper->lowest = 0;
1074 }
1075
1076 list_add_tail(&edge->list[UPPER], &upper->lower);
1077 continue;
1078 }
1079
1080 if (!upper->checked) {
1081 /*
1082 * Still want to blow up for developers since this is a
1083 * logic bug.
1084 */
1085 ASSERT(0);
1086 err = -EINVAL;
1087 goto out;
1088 }
1089 if (cowonly != upper->cowonly) {
1090 ASSERT(0);
1091 err = -EINVAL;
1092 goto out;
1093 }
1094
1095 if (!cowonly) {
1096 rb_node = tree_insert(&cache->rb_root, upper->bytenr,
1097 &upper->rb_node);
1098 if (rb_node)
1099 backref_tree_panic(rb_node, -EEXIST,
1100 upper->bytenr);
1101 }
1102
1103 list_add_tail(&edge->list[UPPER], &upper->lower);
1104
1105 list_for_each_entry(edge, &upper->upper, list[LOWER])
1106 list_add_tail(&edge->list[UPPER], &list);
1107 }
1108 /*
1109 * process useless backref nodes. backref nodes for tree leaves
1110 * are deleted from the cache. backref nodes for upper level
1111 * tree blocks are left in the cache to avoid unnecessary backref
1112 * lookup.
1113 */
1114 while (!list_empty(&useless)) {
1115 upper = list_entry(useless.next, struct backref_node, list);
1116 list_del_init(&upper->list);
1117 ASSERT(list_empty(&upper->upper));
1118 if (upper == node)
1119 node = NULL;
1120 if (upper->lowest) {
1121 list_del_init(&upper->lower);
1122 upper->lowest = 0;
1123 }
1124 while (!list_empty(&upper->lower)) {
1125 edge = list_entry(upper->lower.next,
1126 struct backref_edge, list[UPPER]);
1127 list_del(&edge->list[UPPER]);
1128 list_del(&edge->list[LOWER]);
1129 lower = edge->node[LOWER];
1130 free_backref_edge(cache, edge);
1131
1132 if (list_empty(&lower->upper))
1133 list_add(&lower->list, &useless);
1134 }
1135 __mark_block_processed(rc, upper);
1136 if (upper->level > 0) {
1137 list_add(&upper->list, &cache->detached);
1138 upper->detached = 1;
1139 } else {
1140 rb_erase(&upper->rb_node, &cache->rb_root);
1141 free_backref_node(cache, upper);
1142 }
1143 }
1144 out:
1145 btrfs_free_path(path1);
1146 btrfs_free_path(path2);
1147 if (err) {
1148 while (!list_empty(&useless)) {
1149 lower = list_entry(useless.next,
1150 struct backref_node, list);
1151 list_del_init(&lower->list);
1152 }
1153 while (!list_empty(&list)) {
1154 edge = list_first_entry(&list, struct backref_edge,
1155 list[UPPER]);
1156 list_del(&edge->list[UPPER]);
1157 list_del(&edge->list[LOWER]);
1158 lower = edge->node[LOWER];
1159 upper = edge->node[UPPER];
1160 free_backref_edge(cache, edge);
1161
1162 /*
1163 * Lower is no longer linked to any upper backref nodes
1164 * and isn't in the cache, we can free it ourselves.
1165 */
1166 if (list_empty(&lower->upper) &&
1167 RB_EMPTY_NODE(&lower->rb_node))
1168 list_add(&lower->list, &useless);
1169
1170 if (!RB_EMPTY_NODE(&upper->rb_node))
1171 continue;
1172
1173 /* Add this guy's upper edges to the list to process */
1174 list_for_each_entry(edge, &upper->upper, list[LOWER])
1175 list_add_tail(&edge->list[UPPER], &list);
1176 if (list_empty(&upper->upper))
1177 list_add(&upper->list, &useless);
1178 }
1179
1180 while (!list_empty(&useless)) {
1181 lower = list_entry(useless.next,
1182 struct backref_node, list);
1183 list_del_init(&lower->list);
1184 if (lower == node)
1185 node = NULL;
1186 free_backref_node(cache, lower);
1187 }
1188
1189 free_backref_node(cache, node);
1190 return ERR_PTR(err);
1191 }
1192 ASSERT(!node || !node->detached);
1193 return node;
1194 }
1195
1196 /*
1197 * helper to add backref node for the newly created snapshot.
1198 * the backref node is created by cloning backref node that
1199 * corresponds to root of source tree
1200 */
1201 static int clone_backref_node(struct btrfs_trans_handle *trans,
1202 struct reloc_control *rc,
1203 struct btrfs_root *src,
1204 struct btrfs_root *dest)
1205 {
1206 struct btrfs_root *reloc_root = src->reloc_root;
1207 struct backref_cache *cache = &rc->backref_cache;
1208 struct backref_node *node = NULL;
1209 struct backref_node *new_node;
1210 struct backref_edge *edge;
1211 struct backref_edge *new_edge;
1212 struct rb_node *rb_node;
1213
1214 if (cache->last_trans > 0)
1215 update_backref_cache(trans, cache);
1216
1217 rb_node = tree_search(&cache->rb_root, src->commit_root->start);
1218 if (rb_node) {
1219 node = rb_entry(rb_node, struct backref_node, rb_node);
1220 if (node->detached)
1221 node = NULL;
1222 else
1223 BUG_ON(node->new_bytenr != reloc_root->node->start);
1224 }
1225
1226 if (!node) {
1227 rb_node = tree_search(&cache->rb_root,
1228 reloc_root->commit_root->start);
1229 if (rb_node) {
1230 node = rb_entry(rb_node, struct backref_node,
1231 rb_node);
1232 BUG_ON(node->detached);
1233 }
1234 }
1235
1236 if (!node)
1237 return 0;
1238
1239 new_node = alloc_backref_node(cache);
1240 if (!new_node)
1241 return -ENOMEM;
1242
1243 new_node->bytenr = dest->node->start;
1244 new_node->level = node->level;
1245 new_node->lowest = node->lowest;
1246 new_node->checked = 1;
1247 new_node->root = dest;
1248
1249 if (!node->lowest) {
1250 list_for_each_entry(edge, &node->lower, list[UPPER]) {
1251 new_edge = alloc_backref_edge(cache);
1252 if (!new_edge)
1253 goto fail;
1254
1255 new_edge->node[UPPER] = new_node;
1256 new_edge->node[LOWER] = edge->node[LOWER];
1257 list_add_tail(&new_edge->list[UPPER],
1258 &new_node->lower);
1259 }
1260 } else {
1261 list_add_tail(&new_node->lower, &cache->leaves);
1262 }
1263
1264 rb_node = tree_insert(&cache->rb_root, new_node->bytenr,
1265 &new_node->rb_node);
1266 if (rb_node)
1267 backref_tree_panic(rb_node, -EEXIST, new_node->bytenr);
1268
1269 if (!new_node->lowest) {
1270 list_for_each_entry(new_edge, &new_node->lower, list[UPPER]) {
1271 list_add_tail(&new_edge->list[LOWER],
1272 &new_edge->node[LOWER]->upper);
1273 }
1274 }
1275 return 0;
1276 fail:
1277 while (!list_empty(&new_node->lower)) {
1278 new_edge = list_entry(new_node->lower.next,
1279 struct backref_edge, list[UPPER]);
1280 list_del(&new_edge->list[UPPER]);
1281 free_backref_edge(cache, new_edge);
1282 }
1283 free_backref_node(cache, new_node);
1284 return -ENOMEM;
1285 }
1286
1287 /*
1288 * helper to add 'address of tree root -> reloc tree' mapping
1289 */
1290 static int __must_check __add_reloc_root(struct btrfs_root *root)
1291 {
1292 struct btrfs_fs_info *fs_info = root->fs_info;
1293 struct rb_node *rb_node;
1294 struct mapping_node *node;
1295 struct reloc_control *rc = fs_info->reloc_ctl;
1296
1297 node = kmalloc(sizeof(*node), GFP_NOFS);
1298 if (!node)
1299 return -ENOMEM;
1300
1301 node->bytenr = root->node->start;
1302 node->data = root;
1303
1304 spin_lock(&rc->reloc_root_tree.lock);
1305 rb_node = tree_insert(&rc->reloc_root_tree.rb_root,
1306 node->bytenr, &node->rb_node);
1307 spin_unlock(&rc->reloc_root_tree.lock);
1308 if (rb_node) {
1309 btrfs_panic(fs_info, -EEXIST,
1310 "Duplicate root found for start=%llu while inserting into relocation tree",
1311 node->bytenr);
1312 }
1313
1314 list_add_tail(&root->root_list, &rc->reloc_roots);
1315 return 0;
1316 }
1317
1318 /*
1319 * helper to delete the 'address of tree root -> reloc tree'
1320 * mapping
1321 */
1322 static void __del_reloc_root(struct btrfs_root *root)
1323 {
1324 struct btrfs_fs_info *fs_info = root->fs_info;
1325 struct rb_node *rb_node;
1326 struct mapping_node *node = NULL;
1327 struct reloc_control *rc = fs_info->reloc_ctl;
1328
1329 if (rc && root->node) {
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 if (!node)
1339 return;
1340 BUG_ON((struct btrfs_root *)node->data != root);
1341 }
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 int ret;
1392
1393 root_item = kmalloc(sizeof(*root_item), GFP_NOFS);
1394 BUG_ON(!root_item);
1395
1396 root_key.objectid = BTRFS_TREE_RELOC_OBJECTID;
1397 root_key.type = BTRFS_ROOT_ITEM_KEY;
1398 root_key.offset = objectid;
1399
1400 if (root->root_key.objectid == objectid) {
1401 u64 commit_root_gen;
1402
1403 /* called by btrfs_init_reloc_root */
1404 ret = btrfs_copy_root(trans, root, root->commit_root, &eb,
1405 BTRFS_TREE_RELOC_OBJECTID);
1406 BUG_ON(ret);
1407 /*
1408 * Set the last_snapshot field to the generation of the commit
1409 * root - like this ctree.c:btrfs_block_can_be_shared() behaves
1410 * correctly (returns true) when the relocation root is created
1411 * either inside the critical section of a transaction commit
1412 * (through transaction.c:qgroup_account_snapshot()) and when
1413 * it's created before the transaction commit is started.
1414 */
1415 commit_root_gen = btrfs_header_generation(root->commit_root);
1416 btrfs_set_root_last_snapshot(&root->root_item, commit_root_gen);
1417 } else {
1418 /*
1419 * called by btrfs_reloc_post_snapshot_hook.
1420 * the source tree is a reloc tree, all tree blocks
1421 * modified after it was created have RELOC flag
1422 * set in their headers. so it's OK to not update
1423 * the 'last_snapshot'.
1424 */
1425 ret = btrfs_copy_root(trans, root, root->node, &eb,
1426 BTRFS_TREE_RELOC_OBJECTID);
1427 BUG_ON(ret);
1428 }
1429
1430 memcpy(root_item, &root->root_item, sizeof(*root_item));
1431 btrfs_set_root_bytenr(root_item, eb->start);
1432 btrfs_set_root_level(root_item, btrfs_header_level(eb));
1433 btrfs_set_root_generation(root_item, trans->transid);
1434
1435 if (root->root_key.objectid == objectid) {
1436 btrfs_set_root_refs(root_item, 0);
1437 memset(&root_item->drop_progress, 0,
1438 sizeof(struct btrfs_disk_key));
1439 root_item->drop_level = 0;
1440 }
1441
1442 btrfs_tree_unlock(eb);
1443 free_extent_buffer(eb);
1444
1445 ret = btrfs_insert_root(trans, fs_info->tree_root,
1446 &root_key, root_item);
1447 BUG_ON(ret);
1448 kfree(root_item);
1449
1450 reloc_root = btrfs_read_fs_root(fs_info->tree_root, &root_key);
1451 BUG_ON(IS_ERR(reloc_root));
1452 reloc_root->last_trans = trans->transid;
1453 return reloc_root;
1454 }
1455
1456 /*
1457 * create reloc tree for a given fs tree. reloc tree is just a
1458 * snapshot of the fs tree with special root objectid.
1459 */
1460 int btrfs_init_reloc_root(struct btrfs_trans_handle *trans,
1461 struct btrfs_root *root)
1462 {
1463 struct btrfs_fs_info *fs_info = root->fs_info;
1464 struct btrfs_root *reloc_root;
1465 struct reloc_control *rc = fs_info->reloc_ctl;
1466 struct btrfs_block_rsv *rsv;
1467 int clear_rsv = 0;
1468 int ret;
1469
1470 /*
1471 * The subvolume has reloc tree but the swap is finished, no need to
1472 * create/update the dead reloc tree
1473 */
1474 if (reloc_root_is_dead(root))
1475 return 0;
1476
1477 if (root->reloc_root) {
1478 reloc_root = root->reloc_root;
1479 reloc_root->last_trans = trans->transid;
1480 return 0;
1481 }
1482
1483 if (!rc || !rc->create_reloc_tree ||
1484 root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID)
1485 return 0;
1486
1487 if (!trans->reloc_reserved) {
1488 rsv = trans->block_rsv;
1489 trans->block_rsv = rc->block_rsv;
1490 clear_rsv = 1;
1491 }
1492 reloc_root = create_reloc_root(trans, root, root->root_key.objectid);
1493 if (clear_rsv)
1494 trans->block_rsv = rsv;
1495
1496 ret = __add_reloc_root(reloc_root);
1497 BUG_ON(ret < 0);
1498 root->reloc_root = reloc_root;
1499 return 0;
1500 }
1501
1502 /*
1503 * update root item of reloc tree
1504 */
1505 int btrfs_update_reloc_root(struct btrfs_trans_handle *trans,
1506 struct btrfs_root *root)
1507 {
1508 struct btrfs_fs_info *fs_info = root->fs_info;
1509 struct btrfs_root *reloc_root;
1510 struct btrfs_root_item *root_item;
1511 int ret;
1512
1513 if (!have_reloc_root(root))
1514 goto out;
1515
1516 reloc_root = root->reloc_root;
1517 root_item = &reloc_root->root_item;
1518
1519 /* root->reloc_root will stay until current relocation finished */
1520 if (fs_info->reloc_ctl->merge_reloc_tree &&
1521 btrfs_root_refs(root_item) == 0) {
1522 set_bit(BTRFS_ROOT_DEAD_RELOC_TREE, &root->state);
1523 /*
1524 * Mark the tree as dead before we change reloc_root so
1525 * have_reloc_root will not touch it from now on.
1526 */
1527 smp_wmb();
1528 __del_reloc_root(reloc_root);
1529 }
1530
1531 if (reloc_root->commit_root != reloc_root->node) {
1532 btrfs_set_root_node(root_item, reloc_root->node);
1533 free_extent_buffer(reloc_root->commit_root);
1534 reloc_root->commit_root = btrfs_root_node(reloc_root);
1535 }
1536
1537 ret = btrfs_update_root(trans, fs_info->tree_root,
1538 &reloc_root->root_key, root_item);
1539 BUG_ON(ret);
1540
1541 out:
1542 return 0;
1543 }
1544
1545 /*
1546 * helper to find first cached inode with inode number >= objectid
1547 * in a subvolume
1548 */
1549 static struct inode *find_next_inode(struct btrfs_root *root, u64 objectid)
1550 {
1551 struct rb_node *node;
1552 struct rb_node *prev;
1553 struct btrfs_inode *entry;
1554 struct inode *inode;
1555
1556 spin_lock(&root->inode_lock);
1557 again:
1558 node = root->inode_tree.rb_node;
1559 prev = NULL;
1560 while (node) {
1561 prev = node;
1562 entry = rb_entry(node, struct btrfs_inode, rb_node);
1563
1564 if (objectid < btrfs_ino(entry))
1565 node = node->rb_left;
1566 else if (objectid > btrfs_ino(entry))
1567 node = node->rb_right;
1568 else
1569 break;
1570 }
1571 if (!node) {
1572 while (prev) {
1573 entry = rb_entry(prev, struct btrfs_inode, rb_node);
1574 if (objectid <= btrfs_ino(entry)) {
1575 node = prev;
1576 break;
1577 }
1578 prev = rb_next(prev);
1579 }
1580 }
1581 while (node) {
1582 entry = rb_entry(node, struct btrfs_inode, rb_node);
1583 inode = igrab(&entry->vfs_inode);
1584 if (inode) {
1585 spin_unlock(&root->inode_lock);
1586 return inode;
1587 }
1588
1589 objectid = btrfs_ino(entry) + 1;
1590 if (cond_resched_lock(&root->inode_lock))
1591 goto again;
1592
1593 node = rb_next(node);
1594 }
1595 spin_unlock(&root->inode_lock);
1596 return NULL;
1597 }
1598
1599 static int in_block_group(u64 bytenr, struct btrfs_block_group *block_group)
1600 {
1601 if (bytenr >= block_group->start &&
1602 bytenr < block_group->start + block_group->length)
1603 return 1;
1604 return 0;
1605 }
1606
1607 /*
1608 * get new location of data
1609 */
1610 static int get_new_location(struct inode *reloc_inode, u64 *new_bytenr,
1611 u64 bytenr, u64 num_bytes)
1612 {
1613 struct btrfs_root *root = BTRFS_I(reloc_inode)->root;
1614 struct btrfs_path *path;
1615 struct btrfs_file_extent_item *fi;
1616 struct extent_buffer *leaf;
1617 int ret;
1618
1619 path = btrfs_alloc_path();
1620 if (!path)
1621 return -ENOMEM;
1622
1623 bytenr -= BTRFS_I(reloc_inode)->index_cnt;
1624 ret = btrfs_lookup_file_extent(NULL, root, path,
1625 btrfs_ino(BTRFS_I(reloc_inode)), bytenr, 0);
1626 if (ret < 0)
1627 goto out;
1628 if (ret > 0) {
1629 ret = -ENOENT;
1630 goto out;
1631 }
1632
1633 leaf = path->nodes[0];
1634 fi = btrfs_item_ptr(leaf, path->slots[0],
1635 struct btrfs_file_extent_item);
1636
1637 BUG_ON(btrfs_file_extent_offset(leaf, fi) ||
1638 btrfs_file_extent_compression(leaf, fi) ||
1639 btrfs_file_extent_encryption(leaf, fi) ||
1640 btrfs_file_extent_other_encoding(leaf, fi));
1641
1642 if (num_bytes != btrfs_file_extent_disk_num_bytes(leaf, fi)) {
1643 ret = -EINVAL;
1644 goto out;
1645 }
1646
1647 *new_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1648 ret = 0;
1649 out:
1650 btrfs_free_path(path);
1651 return ret;
1652 }
1653
1654 /*
1655 * update file extent items in the tree leaf to point to
1656 * the new locations.
1657 */
1658 static noinline_for_stack
1659 int replace_file_extents(struct btrfs_trans_handle *trans,
1660 struct reloc_control *rc,
1661 struct btrfs_root *root,
1662 struct extent_buffer *leaf)
1663 {
1664 struct btrfs_fs_info *fs_info = root->fs_info;
1665 struct btrfs_key key;
1666 struct btrfs_file_extent_item *fi;
1667 struct inode *inode = NULL;
1668 u64 parent;
1669 u64 bytenr;
1670 u64 new_bytenr = 0;
1671 u64 num_bytes;
1672 u64 end;
1673 u32 nritems;
1674 u32 i;
1675 int ret = 0;
1676 int first = 1;
1677 int dirty = 0;
1678
1679 if (rc->stage != UPDATE_DATA_PTRS)
1680 return 0;
1681
1682 /* reloc trees always use full backref */
1683 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID)
1684 parent = leaf->start;
1685 else
1686 parent = 0;
1687
1688 nritems = btrfs_header_nritems(leaf);
1689 for (i = 0; i < nritems; i++) {
1690 struct btrfs_ref ref = { 0 };
1691
1692 cond_resched();
1693 btrfs_item_key_to_cpu(leaf, &key, i);
1694 if (key.type != BTRFS_EXTENT_DATA_KEY)
1695 continue;
1696 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
1697 if (btrfs_file_extent_type(leaf, fi) ==
1698 BTRFS_FILE_EXTENT_INLINE)
1699 continue;
1700 bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1701 num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
1702 if (bytenr == 0)
1703 continue;
1704 if (!in_block_group(bytenr, rc->block_group))
1705 continue;
1706
1707 /*
1708 * if we are modifying block in fs tree, wait for readpage
1709 * to complete and drop the extent cache
1710 */
1711 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
1712 if (first) {
1713 inode = find_next_inode(root, key.objectid);
1714 first = 0;
1715 } else if (inode && btrfs_ino(BTRFS_I(inode)) < key.objectid) {
1716 btrfs_add_delayed_iput(inode);
1717 inode = find_next_inode(root, key.objectid);
1718 }
1719 if (inode && btrfs_ino(BTRFS_I(inode)) == key.objectid) {
1720 end = key.offset +
1721 btrfs_file_extent_num_bytes(leaf, fi);
1722 WARN_ON(!IS_ALIGNED(key.offset,
1723 fs_info->sectorsize));
1724 WARN_ON(!IS_ALIGNED(end, fs_info->sectorsize));
1725 end--;
1726 ret = try_lock_extent(&BTRFS_I(inode)->io_tree,
1727 key.offset, end);
1728 if (!ret)
1729 continue;
1730
1731 btrfs_drop_extent_cache(BTRFS_I(inode),
1732 key.offset, end, 1);
1733 unlock_extent(&BTRFS_I(inode)->io_tree,
1734 key.offset, end);
1735 }
1736 }
1737
1738 ret = get_new_location(rc->data_inode, &new_bytenr,
1739 bytenr, num_bytes);
1740 if (ret) {
1741 /*
1742 * Don't have to abort since we've not changed anything
1743 * in the file extent yet.
1744 */
1745 break;
1746 }
1747
1748 btrfs_set_file_extent_disk_bytenr(leaf, fi, new_bytenr);
1749 dirty = 1;
1750
1751 key.offset -= btrfs_file_extent_offset(leaf, fi);
1752 btrfs_init_generic_ref(&ref, BTRFS_ADD_DELAYED_REF, new_bytenr,
1753 num_bytes, parent);
1754 ref.real_root = root->root_key.objectid;
1755 btrfs_init_data_ref(&ref, btrfs_header_owner(leaf),
1756 key.objectid, key.offset);
1757 ret = btrfs_inc_extent_ref(trans, &ref);
1758 if (ret) {
1759 btrfs_abort_transaction(trans, ret);
1760 break;
1761 }
1762
1763 btrfs_init_generic_ref(&ref, BTRFS_DROP_DELAYED_REF, bytenr,
1764 num_bytes, parent);
1765 ref.real_root = root->root_key.objectid;
1766 btrfs_init_data_ref(&ref, btrfs_header_owner(leaf),
1767 key.objectid, key.offset);
1768 ret = btrfs_free_extent(trans, &ref);
1769 if (ret) {
1770 btrfs_abort_transaction(trans, ret);
1771 break;
1772 }
1773 }
1774 if (dirty)
1775 btrfs_mark_buffer_dirty(leaf);
1776 if (inode)
1777 btrfs_add_delayed_iput(inode);
1778 return ret;
1779 }
1780
1781 static noinline_for_stack
1782 int memcmp_node_keys(struct extent_buffer *eb, int slot,
1783 struct btrfs_path *path, int level)
1784 {
1785 struct btrfs_disk_key key1;
1786 struct btrfs_disk_key key2;
1787 btrfs_node_key(eb, &key1, slot);
1788 btrfs_node_key(path->nodes[level], &key2, path->slots[level]);
1789 return memcmp(&key1, &key2, sizeof(key1));
1790 }
1791
1792 /*
1793 * try to replace tree blocks in fs tree with the new blocks
1794 * in reloc tree. tree blocks haven't been modified since the
1795 * reloc tree was create can be replaced.
1796 *
1797 * if a block was replaced, level of the block + 1 is returned.
1798 * if no block got replaced, 0 is returned. if there are other
1799 * errors, a negative error number is returned.
1800 */
1801 static noinline_for_stack
1802 int replace_path(struct btrfs_trans_handle *trans, struct reloc_control *rc,
1803 struct btrfs_root *dest, struct btrfs_root *src,
1804 struct btrfs_path *path, struct btrfs_key *next_key,
1805 int lowest_level, int max_level)
1806 {
1807 struct btrfs_fs_info *fs_info = dest->fs_info;
1808 struct extent_buffer *eb;
1809 struct extent_buffer *parent;
1810 struct btrfs_ref ref = { 0 };
1811 struct btrfs_key key;
1812 u64 old_bytenr;
1813 u64 new_bytenr;
1814 u64 old_ptr_gen;
1815 u64 new_ptr_gen;
1816 u64 last_snapshot;
1817 u32 blocksize;
1818 int cow = 0;
1819 int level;
1820 int ret;
1821 int slot;
1822
1823 BUG_ON(src->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
1824 BUG_ON(dest->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID);
1825
1826 last_snapshot = btrfs_root_last_snapshot(&src->root_item);
1827 again:
1828 slot = path->slots[lowest_level];
1829 btrfs_node_key_to_cpu(path->nodes[lowest_level], &key, slot);
1830
1831 eb = btrfs_lock_root_node(dest);
1832 btrfs_set_lock_blocking_write(eb);
1833 level = btrfs_header_level(eb);
1834
1835 if (level < lowest_level) {
1836 btrfs_tree_unlock(eb);
1837 free_extent_buffer(eb);
1838 return 0;
1839 }
1840
1841 if (cow) {
1842 ret = btrfs_cow_block(trans, dest, eb, NULL, 0, &eb);
1843 BUG_ON(ret);
1844 }
1845 btrfs_set_lock_blocking_write(eb);
1846
1847 if (next_key) {
1848 next_key->objectid = (u64)-1;
1849 next_key->type = (u8)-1;
1850 next_key->offset = (u64)-1;
1851 }
1852
1853 parent = eb;
1854 while (1) {
1855 struct btrfs_key first_key;
1856
1857 level = btrfs_header_level(parent);
1858 BUG_ON(level < lowest_level);
1859
1860 ret = btrfs_bin_search(parent, &key, level, &slot);
1861 if (ret < 0)
1862 break;
1863 if (ret && slot > 0)
1864 slot--;
1865
1866 if (next_key && slot + 1 < btrfs_header_nritems(parent))
1867 btrfs_node_key_to_cpu(parent, next_key, slot + 1);
1868
1869 old_bytenr = btrfs_node_blockptr(parent, slot);
1870 blocksize = fs_info->nodesize;
1871 old_ptr_gen = btrfs_node_ptr_generation(parent, slot);
1872 btrfs_node_key_to_cpu(parent, &first_key, slot);
1873
1874 if (level <= max_level) {
1875 eb = path->nodes[level];
1876 new_bytenr = btrfs_node_blockptr(eb,
1877 path->slots[level]);
1878 new_ptr_gen = btrfs_node_ptr_generation(eb,
1879 path->slots[level]);
1880 } else {
1881 new_bytenr = 0;
1882 new_ptr_gen = 0;
1883 }
1884
1885 if (WARN_ON(new_bytenr > 0 && new_bytenr == old_bytenr)) {
1886 ret = level;
1887 break;
1888 }
1889
1890 if (new_bytenr == 0 || old_ptr_gen > last_snapshot ||
1891 memcmp_node_keys(parent, slot, path, level)) {
1892 if (level <= lowest_level) {
1893 ret = 0;
1894 break;
1895 }
1896
1897 eb = read_tree_block(fs_info, old_bytenr, old_ptr_gen,
1898 level - 1, &first_key);
1899 if (IS_ERR(eb)) {
1900 ret = PTR_ERR(eb);
1901 break;
1902 } else if (!extent_buffer_uptodate(eb)) {
1903 ret = -EIO;
1904 free_extent_buffer(eb);
1905 break;
1906 }
1907 btrfs_tree_lock(eb);
1908 if (cow) {
1909 ret = btrfs_cow_block(trans, dest, eb, parent,
1910 slot, &eb);
1911 BUG_ON(ret);
1912 }
1913 btrfs_set_lock_blocking_write(eb);
1914
1915 btrfs_tree_unlock(parent);
1916 free_extent_buffer(parent);
1917
1918 parent = eb;
1919 continue;
1920 }
1921
1922 if (!cow) {
1923 btrfs_tree_unlock(parent);
1924 free_extent_buffer(parent);
1925 cow = 1;
1926 goto again;
1927 }
1928
1929 btrfs_node_key_to_cpu(path->nodes[level], &key,
1930 path->slots[level]);
1931 btrfs_release_path(path);
1932
1933 path->lowest_level = level;
1934 ret = btrfs_search_slot(trans, src, &key, path, 0, 1);
1935 path->lowest_level = 0;
1936 BUG_ON(ret);
1937
1938 /*
1939 * Info qgroup to trace both subtrees.
1940 *
1941 * We must trace both trees.
1942 * 1) Tree reloc subtree
1943 * If not traced, we will leak data numbers
1944 * 2) Fs subtree
1945 * If not traced, we will double count old data
1946 *
1947 * We don't scan the subtree right now, but only record
1948 * the swapped tree blocks.
1949 * The real subtree rescan is delayed until we have new
1950 * CoW on the subtree root node before transaction commit.
1951 */
1952 ret = btrfs_qgroup_add_swapped_blocks(trans, dest,
1953 rc->block_group, parent, slot,
1954 path->nodes[level], path->slots[level],
1955 last_snapshot);
1956 if (ret < 0)
1957 break;
1958 /*
1959 * swap blocks in fs tree and reloc tree.
1960 */
1961 btrfs_set_node_blockptr(parent, slot, new_bytenr);
1962 btrfs_set_node_ptr_generation(parent, slot, new_ptr_gen);
1963 btrfs_mark_buffer_dirty(parent);
1964
1965 btrfs_set_node_blockptr(path->nodes[level],
1966 path->slots[level], old_bytenr);
1967 btrfs_set_node_ptr_generation(path->nodes[level],
1968 path->slots[level], old_ptr_gen);
1969 btrfs_mark_buffer_dirty(path->nodes[level]);
1970
1971 btrfs_init_generic_ref(&ref, BTRFS_ADD_DELAYED_REF, old_bytenr,
1972 blocksize, path->nodes[level]->start);
1973 ref.skip_qgroup = true;
1974 btrfs_init_tree_ref(&ref, level - 1, src->root_key.objectid);
1975 ret = btrfs_inc_extent_ref(trans, &ref);
1976 BUG_ON(ret);
1977 btrfs_init_generic_ref(&ref, BTRFS_ADD_DELAYED_REF, new_bytenr,
1978 blocksize, 0);
1979 ref.skip_qgroup = true;
1980 btrfs_init_tree_ref(&ref, level - 1, dest->root_key.objectid);
1981 ret = btrfs_inc_extent_ref(trans, &ref);
1982 BUG_ON(ret);
1983
1984 btrfs_init_generic_ref(&ref, BTRFS_DROP_DELAYED_REF, new_bytenr,
1985 blocksize, path->nodes[level]->start);
1986 btrfs_init_tree_ref(&ref, level - 1, src->root_key.objectid);
1987 ref.skip_qgroup = true;
1988 ret = btrfs_free_extent(trans, &ref);
1989 BUG_ON(ret);
1990
1991 btrfs_init_generic_ref(&ref, BTRFS_DROP_DELAYED_REF, old_bytenr,
1992 blocksize, 0);
1993 btrfs_init_tree_ref(&ref, level - 1, dest->root_key.objectid);
1994 ref.skip_qgroup = true;
1995 ret = btrfs_free_extent(trans, &ref);
1996 BUG_ON(ret);
1997
1998 btrfs_unlock_up_safe(path, 0);
1999
2000 ret = level;
2001 break;
2002 }
2003 btrfs_tree_unlock(parent);
2004 free_extent_buffer(parent);
2005 return ret;
2006 }
2007
2008 /*
2009 * helper to find next relocated block in reloc tree
2010 */
2011 static noinline_for_stack
2012 int walk_up_reloc_tree(struct btrfs_root *root, struct btrfs_path *path,
2013 int *level)
2014 {
2015 struct extent_buffer *eb;
2016 int i;
2017 u64 last_snapshot;
2018 u32 nritems;
2019
2020 last_snapshot = btrfs_root_last_snapshot(&root->root_item);
2021
2022 for (i = 0; i < *level; i++) {
2023 free_extent_buffer(path->nodes[i]);
2024 path->nodes[i] = NULL;
2025 }
2026
2027 for (i = *level; i < BTRFS_MAX_LEVEL && path->nodes[i]; i++) {
2028 eb = path->nodes[i];
2029 nritems = btrfs_header_nritems(eb);
2030 while (path->slots[i] + 1 < nritems) {
2031 path->slots[i]++;
2032 if (btrfs_node_ptr_generation(eb, path->slots[i]) <=
2033 last_snapshot)
2034 continue;
2035
2036 *level = i;
2037 return 0;
2038 }
2039 free_extent_buffer(path->nodes[i]);
2040 path->nodes[i] = NULL;
2041 }
2042 return 1;
2043 }
2044
2045 /*
2046 * walk down reloc tree to find relocated block of lowest level
2047 */
2048 static noinline_for_stack
2049 int walk_down_reloc_tree(struct btrfs_root *root, struct btrfs_path *path,
2050 int *level)
2051 {
2052 struct btrfs_fs_info *fs_info = root->fs_info;
2053 struct extent_buffer *eb = NULL;
2054 int i;
2055 u64 bytenr;
2056 u64 ptr_gen = 0;
2057 u64 last_snapshot;
2058 u32 nritems;
2059
2060 last_snapshot = btrfs_root_last_snapshot(&root->root_item);
2061
2062 for (i = *level; i > 0; i--) {
2063 struct btrfs_key first_key;
2064
2065 eb = path->nodes[i];
2066 nritems = btrfs_header_nritems(eb);
2067 while (path->slots[i] < nritems) {
2068 ptr_gen = btrfs_node_ptr_generation(eb, path->slots[i]);
2069 if (ptr_gen > last_snapshot)
2070 break;
2071 path->slots[i]++;
2072 }
2073 if (path->slots[i] >= nritems) {
2074 if (i == *level)
2075 break;
2076 *level = i + 1;
2077 return 0;
2078 }
2079 if (i == 1) {
2080 *level = i;
2081 return 0;
2082 }
2083
2084 bytenr = btrfs_node_blockptr(eb, path->slots[i]);
2085 btrfs_node_key_to_cpu(eb, &first_key, path->slots[i]);
2086 eb = read_tree_block(fs_info, bytenr, ptr_gen, i - 1,
2087 &first_key);
2088 if (IS_ERR(eb)) {
2089 return PTR_ERR(eb);
2090 } else if (!extent_buffer_uptodate(eb)) {
2091 free_extent_buffer(eb);
2092 return -EIO;
2093 }
2094 BUG_ON(btrfs_header_level(eb) != i - 1);
2095 path->nodes[i - 1] = eb;
2096 path->slots[i - 1] = 0;
2097 }
2098 return 1;
2099 }
2100
2101 /*
2102 * invalidate extent cache for file extents whose key in range of
2103 * [min_key, max_key)
2104 */
2105 static int invalidate_extent_cache(struct btrfs_root *root,
2106 struct btrfs_key *min_key,
2107 struct btrfs_key *max_key)
2108 {
2109 struct btrfs_fs_info *fs_info = root->fs_info;
2110 struct inode *inode = NULL;
2111 u64 objectid;
2112 u64 start, end;
2113 u64 ino;
2114
2115 objectid = min_key->objectid;
2116 while (1) {
2117 cond_resched();
2118 iput(inode);
2119
2120 if (objectid > max_key->objectid)
2121 break;
2122
2123 inode = find_next_inode(root, objectid);
2124 if (!inode)
2125 break;
2126 ino = btrfs_ino(BTRFS_I(inode));
2127
2128 if (ino > max_key->objectid) {
2129 iput(inode);
2130 break;
2131 }
2132
2133 objectid = ino + 1;
2134 if (!S_ISREG(inode->i_mode))
2135 continue;
2136
2137 if (unlikely(min_key->objectid == ino)) {
2138 if (min_key->type > BTRFS_EXTENT_DATA_KEY)
2139 continue;
2140 if (min_key->type < BTRFS_EXTENT_DATA_KEY)
2141 start = 0;
2142 else {
2143 start = min_key->offset;
2144 WARN_ON(!IS_ALIGNED(start, fs_info->sectorsize));
2145 }
2146 } else {
2147 start = 0;
2148 }
2149
2150 if (unlikely(max_key->objectid == ino)) {
2151 if (max_key->type < BTRFS_EXTENT_DATA_KEY)
2152 continue;
2153 if (max_key->type > BTRFS_EXTENT_DATA_KEY) {
2154 end = (u64)-1;
2155 } else {
2156 if (max_key->offset == 0)
2157 continue;
2158 end = max_key->offset;
2159 WARN_ON(!IS_ALIGNED(end, fs_info->sectorsize));
2160 end--;
2161 }
2162 } else {
2163 end = (u64)-1;
2164 }
2165
2166 /* the lock_extent waits for readpage to complete */
2167 lock_extent(&BTRFS_I(inode)->io_tree, start, end);
2168 btrfs_drop_extent_cache(BTRFS_I(inode), start, end, 1);
2169 unlock_extent(&BTRFS_I(inode)->io_tree, start, end);
2170 }
2171 return 0;
2172 }
2173
2174 static int find_next_key(struct btrfs_path *path, int level,
2175 struct btrfs_key *key)
2176
2177 {
2178 while (level < BTRFS_MAX_LEVEL) {
2179 if (!path->nodes[level])
2180 break;
2181 if (path->slots[level] + 1 <
2182 btrfs_header_nritems(path->nodes[level])) {
2183 btrfs_node_key_to_cpu(path->nodes[level], key,
2184 path->slots[level] + 1);
2185 return 0;
2186 }
2187 level++;
2188 }
2189 return 1;
2190 }
2191
2192 /*
2193 * Insert current subvolume into reloc_control::dirty_subvol_roots
2194 */
2195 static void insert_dirty_subvol(struct btrfs_trans_handle *trans,
2196 struct reloc_control *rc,
2197 struct btrfs_root *root)
2198 {
2199 struct btrfs_root *reloc_root = root->reloc_root;
2200 struct btrfs_root_item *reloc_root_item;
2201
2202 /* @root must be a subvolume tree root with a valid reloc tree */
2203 ASSERT(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
2204 ASSERT(reloc_root);
2205
2206 reloc_root_item = &reloc_root->root_item;
2207 memset(&reloc_root_item->drop_progress, 0,
2208 sizeof(reloc_root_item->drop_progress));
2209 reloc_root_item->drop_level = 0;
2210 btrfs_set_root_refs(reloc_root_item, 0);
2211 btrfs_update_reloc_root(trans, root);
2212
2213 if (list_empty(&root->reloc_dirty_list)) {
2214 btrfs_grab_fs_root(root);
2215 list_add_tail(&root->reloc_dirty_list, &rc->dirty_subvol_roots);
2216 }
2217 }
2218
2219 static int clean_dirty_subvols(struct reloc_control *rc)
2220 {
2221 struct btrfs_root *root;
2222 struct btrfs_root *next;
2223 int ret = 0;
2224 int ret2;
2225
2226 list_for_each_entry_safe(root, next, &rc->dirty_subvol_roots,
2227 reloc_dirty_list) {
2228 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
2229 /* Merged subvolume, cleanup its reloc root */
2230 struct btrfs_root *reloc_root = root->reloc_root;
2231
2232 list_del_init(&root->reloc_dirty_list);
2233 root->reloc_root = NULL;
2234 if (reloc_root) {
2235
2236 ret2 = btrfs_drop_snapshot(reloc_root, NULL, 0, 1);
2237 if (ret2 < 0 && !ret)
2238 ret = ret2;
2239 }
2240 /*
2241 * Need barrier to ensure clear_bit() only happens after
2242 * root->reloc_root = NULL. Pairs with have_reloc_root.
2243 */
2244 smp_wmb();
2245 clear_bit(BTRFS_ROOT_DEAD_RELOC_TREE, &root->state);
2246 btrfs_put_fs_root(root);
2247 } else {
2248 /* Orphan reloc tree, just clean it up */
2249 ret2 = btrfs_drop_snapshot(root, NULL, 0, 1);
2250 if (ret2 < 0 && !ret)
2251 ret = ret2;
2252 }
2253 }
2254 return ret;
2255 }
2256
2257 /*
2258 * merge the relocated tree blocks in reloc tree with corresponding
2259 * fs tree.
2260 */
2261 static noinline_for_stack int merge_reloc_root(struct reloc_control *rc,
2262 struct btrfs_root *root)
2263 {
2264 struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
2265 struct btrfs_key key;
2266 struct btrfs_key next_key;
2267 struct btrfs_trans_handle *trans = NULL;
2268 struct btrfs_root *reloc_root;
2269 struct btrfs_root_item *root_item;
2270 struct btrfs_path *path;
2271 struct extent_buffer *leaf;
2272 int level;
2273 int max_level;
2274 int replaced = 0;
2275 int ret;
2276 int err = 0;
2277 u32 min_reserved;
2278
2279 path = btrfs_alloc_path();
2280 if (!path)
2281 return -ENOMEM;
2282 path->reada = READA_FORWARD;
2283
2284 reloc_root = root->reloc_root;
2285 root_item = &reloc_root->root_item;
2286
2287 if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
2288 level = btrfs_root_level(root_item);
2289 atomic_inc(&reloc_root->node->refs);
2290 path->nodes[level] = reloc_root->node;
2291 path->slots[level] = 0;
2292 } else {
2293 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
2294
2295 level = root_item->drop_level;
2296 BUG_ON(level == 0);
2297 path->lowest_level = level;
2298 ret = btrfs_search_slot(NULL, reloc_root, &key, path, 0, 0);
2299 path->lowest_level = 0;
2300 if (ret < 0) {
2301 btrfs_free_path(path);
2302 return ret;
2303 }
2304
2305 btrfs_node_key_to_cpu(path->nodes[level], &next_key,
2306 path->slots[level]);
2307 WARN_ON(memcmp(&key, &next_key, sizeof(key)));
2308
2309 btrfs_unlock_up_safe(path, 0);
2310 }
2311
2312 min_reserved = fs_info->nodesize * (BTRFS_MAX_LEVEL - 1) * 2;
2313 memset(&next_key, 0, sizeof(next_key));
2314
2315 while (1) {
2316 ret = btrfs_block_rsv_refill(root, rc->block_rsv, min_reserved,
2317 BTRFS_RESERVE_FLUSH_ALL);
2318 if (ret) {
2319 err = ret;
2320 goto out;
2321 }
2322 trans = btrfs_start_transaction(root, 0);
2323 if (IS_ERR(trans)) {
2324 err = PTR_ERR(trans);
2325 trans = NULL;
2326 goto out;
2327 }
2328 trans->block_rsv = rc->block_rsv;
2329
2330 replaced = 0;
2331 max_level = level;
2332
2333 ret = walk_down_reloc_tree(reloc_root, path, &level);
2334 if (ret < 0) {
2335 err = ret;
2336 goto out;
2337 }
2338 if (ret > 0)
2339 break;
2340
2341 if (!find_next_key(path, level, &key) &&
2342 btrfs_comp_cpu_keys(&next_key, &key) >= 0) {
2343 ret = 0;
2344 } else {
2345 ret = replace_path(trans, rc, root, reloc_root, path,
2346 &next_key, level, max_level);
2347 }
2348 if (ret < 0) {
2349 err = ret;
2350 goto out;
2351 }
2352
2353 if (ret > 0) {
2354 level = ret;
2355 btrfs_node_key_to_cpu(path->nodes[level], &key,
2356 path->slots[level]);
2357 replaced = 1;
2358 }
2359
2360 ret = walk_up_reloc_tree(reloc_root, path, &level);
2361 if (ret > 0)
2362 break;
2363
2364 BUG_ON(level == 0);
2365 /*
2366 * save the merging progress in the drop_progress.
2367 * this is OK since root refs == 1 in this case.
2368 */
2369 btrfs_node_key(path->nodes[level], &root_item->drop_progress,
2370 path->slots[level]);
2371 root_item->drop_level = level;
2372
2373 btrfs_end_transaction_throttle(trans);
2374 trans = NULL;
2375
2376 btrfs_btree_balance_dirty(fs_info);
2377
2378 if (replaced && rc->stage == UPDATE_DATA_PTRS)
2379 invalidate_extent_cache(root, &key, &next_key);
2380 }
2381
2382 /*
2383 * handle the case only one block in the fs tree need to be
2384 * relocated and the block is tree root.
2385 */
2386 leaf = btrfs_lock_root_node(root);
2387 ret = btrfs_cow_block(trans, root, leaf, NULL, 0, &leaf);
2388 btrfs_tree_unlock(leaf);
2389 free_extent_buffer(leaf);
2390 if (ret < 0)
2391 err = ret;
2392 out:
2393 btrfs_free_path(path);
2394
2395 if (err == 0)
2396 insert_dirty_subvol(trans, rc, root);
2397
2398 if (trans)
2399 btrfs_end_transaction_throttle(trans);
2400
2401 btrfs_btree_balance_dirty(fs_info);
2402
2403 if (replaced && rc->stage == UPDATE_DATA_PTRS)
2404 invalidate_extent_cache(root, &key, &next_key);
2405
2406 return err;
2407 }
2408
2409 static noinline_for_stack
2410 int prepare_to_merge(struct reloc_control *rc, int err)
2411 {
2412 struct btrfs_root *root = rc->extent_root;
2413 struct btrfs_fs_info *fs_info = root->fs_info;
2414 struct btrfs_root *reloc_root;
2415 struct btrfs_trans_handle *trans;
2416 LIST_HEAD(reloc_roots);
2417 u64 num_bytes = 0;
2418 int ret;
2419
2420 mutex_lock(&fs_info->reloc_mutex);
2421 rc->merging_rsv_size += fs_info->nodesize * (BTRFS_MAX_LEVEL - 1) * 2;
2422 rc->merging_rsv_size += rc->nodes_relocated * 2;
2423 mutex_unlock(&fs_info->reloc_mutex);
2424
2425 again:
2426 if (!err) {
2427 num_bytes = rc->merging_rsv_size;
2428 ret = btrfs_block_rsv_add(root, rc->block_rsv, num_bytes,
2429 BTRFS_RESERVE_FLUSH_ALL);
2430 if (ret)
2431 err = ret;
2432 }
2433
2434 trans = btrfs_join_transaction(rc->extent_root);
2435 if (IS_ERR(trans)) {
2436 if (!err)
2437 btrfs_block_rsv_release(fs_info, rc->block_rsv,
2438 num_bytes);
2439 return PTR_ERR(trans);
2440 }
2441
2442 if (!err) {
2443 if (num_bytes != rc->merging_rsv_size) {
2444 btrfs_end_transaction(trans);
2445 btrfs_block_rsv_release(fs_info, rc->block_rsv,
2446 num_bytes);
2447 goto again;
2448 }
2449 }
2450
2451 rc->merge_reloc_tree = 1;
2452
2453 while (!list_empty(&rc->reloc_roots)) {
2454 reloc_root = list_entry(rc->reloc_roots.next,
2455 struct btrfs_root, root_list);
2456 list_del_init(&reloc_root->root_list);
2457
2458 root = read_fs_root(fs_info, reloc_root->root_key.offset);
2459 BUG_ON(IS_ERR(root));
2460 BUG_ON(root->reloc_root != reloc_root);
2461
2462 /*
2463 * set reference count to 1, so btrfs_recover_relocation
2464 * knows it should resumes merging
2465 */
2466 if (!err)
2467 btrfs_set_root_refs(&reloc_root->root_item, 1);
2468 btrfs_update_reloc_root(trans, root);
2469
2470 list_add(&reloc_root->root_list, &reloc_roots);
2471 }
2472
2473 list_splice(&reloc_roots, &rc->reloc_roots);
2474
2475 if (!err)
2476 btrfs_commit_transaction(trans);
2477 else
2478 btrfs_end_transaction(trans);
2479 return err;
2480 }
2481
2482 static noinline_for_stack
2483 void free_reloc_roots(struct list_head *list)
2484 {
2485 struct btrfs_root *reloc_root;
2486
2487 while (!list_empty(list)) {
2488 reloc_root = list_entry(list->next, struct btrfs_root,
2489 root_list);
2490 __del_reloc_root(reloc_root);
2491 free_extent_buffer(reloc_root->node);
2492 free_extent_buffer(reloc_root->commit_root);
2493 reloc_root->node = NULL;
2494 reloc_root->commit_root = NULL;
2495 }
2496 }
2497
2498 static noinline_for_stack
2499 void merge_reloc_roots(struct reloc_control *rc)
2500 {
2501 struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
2502 struct btrfs_root *root;
2503 struct btrfs_root *reloc_root;
2504 LIST_HEAD(reloc_roots);
2505 int found = 0;
2506 int ret = 0;
2507 again:
2508 root = rc->extent_root;
2509
2510 /*
2511 * this serializes us with btrfs_record_root_in_transaction,
2512 * we have to make sure nobody is in the middle of
2513 * adding their roots to the list while we are
2514 * doing this splice
2515 */
2516 mutex_lock(&fs_info->reloc_mutex);
2517 list_splice_init(&rc->reloc_roots, &reloc_roots);
2518 mutex_unlock(&fs_info->reloc_mutex);
2519
2520 while (!list_empty(&reloc_roots)) {
2521 found = 1;
2522 reloc_root = list_entry(reloc_roots.next,
2523 struct btrfs_root, root_list);
2524
2525 if (btrfs_root_refs(&reloc_root->root_item) > 0) {
2526 root = read_fs_root(fs_info,
2527 reloc_root->root_key.offset);
2528 BUG_ON(IS_ERR(root));
2529 BUG_ON(root->reloc_root != reloc_root);
2530
2531 ret = merge_reloc_root(rc, root);
2532 if (ret) {
2533 if (list_empty(&reloc_root->root_list))
2534 list_add_tail(&reloc_root->root_list,
2535 &reloc_roots);
2536 goto out;
2537 }
2538 } else {
2539 list_del_init(&reloc_root->root_list);
2540 /* Don't forget to queue this reloc root for cleanup */
2541 list_add_tail(&reloc_root->reloc_dirty_list,
2542 &rc->dirty_subvol_roots);
2543 }
2544 }
2545
2546 if (found) {
2547 found = 0;
2548 goto again;
2549 }
2550 out:
2551 if (ret) {
2552 btrfs_handle_fs_error(fs_info, ret, NULL);
2553 if (!list_empty(&reloc_roots))
2554 free_reloc_roots(&reloc_roots);
2555
2556 /* new reloc root may be added */
2557 mutex_lock(&fs_info->reloc_mutex);
2558 list_splice_init(&rc->reloc_roots, &reloc_roots);
2559 mutex_unlock(&fs_info->reloc_mutex);
2560 if (!list_empty(&reloc_roots))
2561 free_reloc_roots(&reloc_roots);
2562 }
2563
2564 BUG_ON(!RB_EMPTY_ROOT(&rc->reloc_root_tree.rb_root));
2565 }
2566
2567 static void free_block_list(struct rb_root *blocks)
2568 {
2569 struct tree_block *block;
2570 struct rb_node *rb_node;
2571 while ((rb_node = rb_first(blocks))) {
2572 block = rb_entry(rb_node, struct tree_block, rb_node);
2573 rb_erase(rb_node, blocks);
2574 kfree(block);
2575 }
2576 }
2577
2578 static int record_reloc_root_in_trans(struct btrfs_trans_handle *trans,
2579 struct btrfs_root *reloc_root)
2580 {
2581 struct btrfs_fs_info *fs_info = reloc_root->fs_info;
2582 struct btrfs_root *root;
2583
2584 if (reloc_root->last_trans == trans->transid)
2585 return 0;
2586
2587 root = read_fs_root(fs_info, reloc_root->root_key.offset);
2588 BUG_ON(IS_ERR(root));
2589 BUG_ON(root->reloc_root != reloc_root);
2590
2591 return btrfs_record_root_in_trans(trans, root);
2592 }
2593
2594 static noinline_for_stack
2595 struct btrfs_root *select_reloc_root(struct btrfs_trans_handle *trans,
2596 struct reloc_control *rc,
2597 struct backref_node *node,
2598 struct backref_edge *edges[])
2599 {
2600 struct backref_node *next;
2601 struct btrfs_root *root;
2602 int index = 0;
2603
2604 next = node;
2605 while (1) {
2606 cond_resched();
2607 next = walk_up_backref(next, edges, &index);
2608 root = next->root;
2609 BUG_ON(!root);
2610 BUG_ON(!test_bit(BTRFS_ROOT_REF_COWS, &root->state));
2611
2612 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) {
2613 record_reloc_root_in_trans(trans, root);
2614 break;
2615 }
2616
2617 btrfs_record_root_in_trans(trans, root);
2618 root = root->reloc_root;
2619
2620 if (next->new_bytenr != root->node->start) {
2621 BUG_ON(next->new_bytenr);
2622 BUG_ON(!list_empty(&next->list));
2623 next->new_bytenr = root->node->start;
2624 next->root = root;
2625 list_add_tail(&next->list,
2626 &rc->backref_cache.changed);
2627 __mark_block_processed(rc, next);
2628 break;
2629 }
2630
2631 WARN_ON(1);
2632 root = NULL;
2633 next = walk_down_backref(edges, &index);
2634 if (!next || next->level <= node->level)
2635 break;
2636 }
2637 if (!root)
2638 return NULL;
2639
2640 next = node;
2641 /* setup backref node path for btrfs_reloc_cow_block */
2642 while (1) {
2643 rc->backref_cache.path[next->level] = next;
2644 if (--index < 0)
2645 break;
2646 next = edges[index]->node[UPPER];
2647 }
2648 return root;
2649 }
2650
2651 /*
2652 * select a tree root for relocation. return NULL if the block
2653 * is reference counted. we should use do_relocation() in this
2654 * case. return a tree root pointer if the block isn't reference
2655 * counted. return -ENOENT if the block is root of reloc tree.
2656 */
2657 static noinline_for_stack
2658 struct btrfs_root *select_one_root(struct backref_node *node)
2659 {
2660 struct backref_node *next;
2661 struct btrfs_root *root;
2662 struct btrfs_root *fs_root = NULL;
2663 struct backref_edge *edges[BTRFS_MAX_LEVEL - 1];
2664 int index = 0;
2665
2666 next = node;
2667 while (1) {
2668 cond_resched();
2669 next = walk_up_backref(next, edges, &index);
2670 root = next->root;
2671 BUG_ON(!root);
2672
2673 /* no other choice for non-references counted tree */
2674 if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state))
2675 return root;
2676
2677 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID)
2678 fs_root = root;
2679
2680 if (next != node)
2681 return NULL;
2682
2683 next = walk_down_backref(edges, &index);
2684 if (!next || next->level <= node->level)
2685 break;
2686 }
2687
2688 if (!fs_root)
2689 return ERR_PTR(-ENOENT);
2690 return fs_root;
2691 }
2692
2693 static noinline_for_stack
2694 u64 calcu_metadata_size(struct reloc_control *rc,
2695 struct backref_node *node, int reserve)
2696 {
2697 struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
2698 struct backref_node *next = node;
2699 struct backref_edge *edge;
2700 struct backref_edge *edges[BTRFS_MAX_LEVEL - 1];
2701 u64 num_bytes = 0;
2702 int index = 0;
2703
2704 BUG_ON(reserve && node->processed);
2705
2706 while (next) {
2707 cond_resched();
2708 while (1) {
2709 if (next->processed && (reserve || next != node))
2710 break;
2711
2712 num_bytes += fs_info->nodesize;
2713
2714 if (list_empty(&next->upper))
2715 break;
2716
2717 edge = list_entry(next->upper.next,
2718 struct backref_edge, list[LOWER]);
2719 edges[index++] = edge;
2720 next = edge->node[UPPER];
2721 }
2722 next = walk_down_backref(edges, &index);
2723 }
2724 return num_bytes;
2725 }
2726
2727 static int reserve_metadata_space(struct btrfs_trans_handle *trans,
2728 struct reloc_control *rc,
2729 struct backref_node *node)
2730 {
2731 struct btrfs_root *root = rc->extent_root;
2732 struct btrfs_fs_info *fs_info = root->fs_info;
2733 u64 num_bytes;
2734 int ret;
2735 u64 tmp;
2736
2737 num_bytes = calcu_metadata_size(rc, node, 1) * 2;
2738
2739 trans->block_rsv = rc->block_rsv;
2740 rc->reserved_bytes += num_bytes;
2741
2742 /*
2743 * We are under a transaction here so we can only do limited flushing.
2744 * If we get an enospc just kick back -EAGAIN so we know to drop the
2745 * transaction and try to refill when we can flush all the things.
2746 */
2747 ret = btrfs_block_rsv_refill(root, rc->block_rsv, num_bytes,
2748 BTRFS_RESERVE_FLUSH_LIMIT);
2749 if (ret) {
2750 tmp = fs_info->nodesize * RELOCATION_RESERVED_NODES;
2751 while (tmp <= rc->reserved_bytes)
2752 tmp <<= 1;
2753 /*
2754 * only one thread can access block_rsv at this point,
2755 * so we don't need hold lock to protect block_rsv.
2756 * we expand more reservation size here to allow enough
2757 * space for relocation and we will return earlier in
2758 * enospc case.
2759 */
2760 rc->block_rsv->size = tmp + fs_info->nodesize *
2761 RELOCATION_RESERVED_NODES;
2762 return -EAGAIN;
2763 }
2764
2765 return 0;
2766 }
2767
2768 /*
2769 * relocate a block tree, and then update pointers in upper level
2770 * blocks that reference the block to point to the new location.
2771 *
2772 * if called by link_to_upper, the block has already been relocated.
2773 * in that case this function just updates pointers.
2774 */
2775 static int do_relocation(struct btrfs_trans_handle *trans,
2776 struct reloc_control *rc,
2777 struct backref_node *node,
2778 struct btrfs_key *key,
2779 struct btrfs_path *path, int lowest)
2780 {
2781 struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
2782 struct backref_node *upper;
2783 struct backref_edge *edge;
2784 struct backref_edge *edges[BTRFS_MAX_LEVEL - 1];
2785 struct btrfs_root *root;
2786 struct extent_buffer *eb;
2787 u32 blocksize;
2788 u64 bytenr;
2789 u64 generation;
2790 int slot;
2791 int ret;
2792 int err = 0;
2793
2794 BUG_ON(lowest && node->eb);
2795
2796 path->lowest_level = node->level + 1;
2797 rc->backref_cache.path[node->level] = node;
2798 list_for_each_entry(edge, &node->upper, list[LOWER]) {
2799 struct btrfs_key first_key;
2800 struct btrfs_ref ref = { 0 };
2801
2802 cond_resched();
2803
2804 upper = edge->node[UPPER];
2805 root = select_reloc_root(trans, rc, upper, edges);
2806 BUG_ON(!root);
2807
2808 if (upper->eb && !upper->locked) {
2809 if (!lowest) {
2810 ret = btrfs_bin_search(upper->eb, key,
2811 upper->level, &slot);
2812 if (ret < 0) {
2813 err = ret;
2814 goto next;
2815 }
2816 BUG_ON(ret);
2817 bytenr = btrfs_node_blockptr(upper->eb, slot);
2818 if (node->eb->start == bytenr)
2819 goto next;
2820 }
2821 drop_node_buffer(upper);
2822 }
2823
2824 if (!upper->eb) {
2825 ret = btrfs_search_slot(trans, root, key, path, 0, 1);
2826 if (ret) {
2827 if (ret < 0)
2828 err = ret;
2829 else
2830 err = -ENOENT;
2831
2832 btrfs_release_path(path);
2833 break;
2834 }
2835
2836 if (!upper->eb) {
2837 upper->eb = path->nodes[upper->level];
2838 path->nodes[upper->level] = NULL;
2839 } else {
2840 BUG_ON(upper->eb != path->nodes[upper->level]);
2841 }
2842
2843 upper->locked = 1;
2844 path->locks[upper->level] = 0;
2845
2846 slot = path->slots[upper->level];
2847 btrfs_release_path(path);
2848 } else {
2849 ret = btrfs_bin_search(upper->eb, key, upper->level,
2850 &slot);
2851 if (ret < 0) {
2852 err = ret;
2853 goto next;
2854 }
2855 BUG_ON(ret);
2856 }
2857
2858 bytenr = btrfs_node_blockptr(upper->eb, slot);
2859 if (lowest) {
2860 if (bytenr != node->bytenr) {
2861 btrfs_err(root->fs_info,
2862 "lowest leaf/node mismatch: bytenr %llu node->bytenr %llu slot %d upper %llu",
2863 bytenr, node->bytenr, slot,
2864 upper->eb->start);
2865 err = -EIO;
2866 goto next;
2867 }
2868 } else {
2869 if (node->eb->start == bytenr)
2870 goto next;
2871 }
2872
2873 blocksize = root->fs_info->nodesize;
2874 generation = btrfs_node_ptr_generation(upper->eb, slot);
2875 btrfs_node_key_to_cpu(upper->eb, &first_key, slot);
2876 eb = read_tree_block(fs_info, bytenr, generation,
2877 upper->level - 1, &first_key);
2878 if (IS_ERR(eb)) {
2879 err = PTR_ERR(eb);
2880 goto next;
2881 } else if (!extent_buffer_uptodate(eb)) {
2882 free_extent_buffer(eb);
2883 err = -EIO;
2884 goto next;
2885 }
2886 btrfs_tree_lock(eb);
2887 btrfs_set_lock_blocking_write(eb);
2888
2889 if (!node->eb) {
2890 ret = btrfs_cow_block(trans, root, eb, upper->eb,
2891 slot, &eb);
2892 btrfs_tree_unlock(eb);
2893 free_extent_buffer(eb);
2894 if (ret < 0) {
2895 err = ret;
2896 goto next;
2897 }
2898 BUG_ON(node->eb != eb);
2899 } else {
2900 btrfs_set_node_blockptr(upper->eb, slot,
2901 node->eb->start);
2902 btrfs_set_node_ptr_generation(upper->eb, slot,
2903 trans->transid);
2904 btrfs_mark_buffer_dirty(upper->eb);
2905
2906 btrfs_init_generic_ref(&ref, BTRFS_ADD_DELAYED_REF,
2907 node->eb->start, blocksize,
2908 upper->eb->start);
2909 ref.real_root = root->root_key.objectid;
2910 btrfs_init_tree_ref(&ref, node->level,
2911 btrfs_header_owner(upper->eb));
2912 ret = btrfs_inc_extent_ref(trans, &ref);
2913 BUG_ON(ret);
2914
2915 ret = btrfs_drop_subtree(trans, root, eb, upper->eb);
2916 BUG_ON(ret);
2917 }
2918 next:
2919 if (!upper->pending)
2920 drop_node_buffer(upper);
2921 else
2922 unlock_node_buffer(upper);
2923 if (err)
2924 break;
2925 }
2926
2927 if (!err && node->pending) {
2928 drop_node_buffer(node);
2929 list_move_tail(&node->list, &rc->backref_cache.changed);
2930 node->pending = 0;
2931 }
2932
2933 path->lowest_level = 0;
2934 BUG_ON(err == -ENOSPC);
2935 return err;
2936 }
2937
2938 static int link_to_upper(struct btrfs_trans_handle *trans,
2939 struct reloc_control *rc,
2940 struct backref_node *node,
2941 struct btrfs_path *path)
2942 {
2943 struct btrfs_key key;
2944
2945 btrfs_node_key_to_cpu(node->eb, &key, 0);
2946 return do_relocation(trans, rc, node, &key, path, 0);
2947 }
2948
2949 static int finish_pending_nodes(struct btrfs_trans_handle *trans,
2950 struct reloc_control *rc,
2951 struct btrfs_path *path, int err)
2952 {
2953 LIST_HEAD(list);
2954 struct backref_cache *cache = &rc->backref_cache;
2955 struct backref_node *node;
2956 int level;
2957 int ret;
2958
2959 for (level = 0; level < BTRFS_MAX_LEVEL; level++) {
2960 while (!list_empty(&cache->pending[level])) {
2961 node = list_entry(cache->pending[level].next,
2962 struct backref_node, list);
2963 list_move_tail(&node->list, &list);
2964 BUG_ON(!node->pending);
2965
2966 if (!err) {
2967 ret = link_to_upper(trans, rc, node, path);
2968 if (ret < 0)
2969 err = ret;
2970 }
2971 }
2972 list_splice_init(&list, &cache->pending[level]);
2973 }
2974 return err;
2975 }
2976
2977 static void mark_block_processed(struct reloc_control *rc,
2978 u64 bytenr, u32 blocksize)
2979 {
2980 set_extent_bits(&rc->processed_blocks, bytenr, bytenr + blocksize - 1,
2981 EXTENT_DIRTY);
2982 }
2983
2984 static void __mark_block_processed(struct reloc_control *rc,
2985 struct backref_node *node)
2986 {
2987 u32 blocksize;
2988 if (node->level == 0 ||
2989 in_block_group(node->bytenr, rc->block_group)) {
2990 blocksize = rc->extent_root->fs_info->nodesize;
2991 mark_block_processed(rc, node->bytenr, blocksize);
2992 }
2993 node->processed = 1;
2994 }
2995
2996 /*
2997 * mark a block and all blocks directly/indirectly reference the block
2998 * as processed.
2999 */
3000 static void update_processed_blocks(struct reloc_control *rc,
3001 struct backref_node *node)
3002 {
3003 struct backref_node *next = node;
3004 struct backref_edge *edge;
3005 struct backref_edge *edges[BTRFS_MAX_LEVEL - 1];
3006 int index = 0;
3007
3008 while (next) {
3009 cond_resched();
3010 while (1) {
3011 if (next->processed)
3012 break;
3013
3014 __mark_block_processed(rc, next);
3015
3016 if (list_empty(&next->upper))
3017 break;
3018
3019 edge = list_entry(next->upper.next,
3020 struct backref_edge, list[LOWER]);
3021 edges[index++] = edge;
3022 next = edge->node[UPPER];
3023 }
3024 next = walk_down_backref(edges, &index);
3025 }
3026 }
3027
3028 static int tree_block_processed(u64 bytenr, struct reloc_control *rc)
3029 {
3030 u32 blocksize = rc->extent_root->fs_info->nodesize;
3031
3032 if (test_range_bit(&rc->processed_blocks, bytenr,
3033 bytenr + blocksize - 1, EXTENT_DIRTY, 1, NULL))
3034 return 1;
3035 return 0;
3036 }
3037
3038 static int get_tree_block_key(struct btrfs_fs_info *fs_info,
3039 struct tree_block *block)
3040 {
3041 struct extent_buffer *eb;
3042
3043 BUG_ON(block->key_ready);
3044 eb = read_tree_block(fs_info, block->bytenr, block->key.offset,
3045 block->level, NULL);
3046 if (IS_ERR(eb)) {
3047 return PTR_ERR(eb);
3048 } else if (!extent_buffer_uptodate(eb)) {
3049 free_extent_buffer(eb);
3050 return -EIO;
3051 }
3052 if (block->level == 0)
3053 btrfs_item_key_to_cpu(eb, &block->key, 0);
3054 else
3055 btrfs_node_key_to_cpu(eb, &block->key, 0);
3056 free_extent_buffer(eb);
3057 block->key_ready = 1;
3058 return 0;
3059 }
3060
3061 /*
3062 * helper function to relocate a tree block
3063 */
3064 static int relocate_tree_block(struct btrfs_trans_handle *trans,
3065 struct reloc_control *rc,
3066 struct backref_node *node,
3067 struct btrfs_key *key,
3068 struct btrfs_path *path)
3069 {
3070 struct btrfs_root *root;
3071 int ret = 0;
3072
3073 if (!node)
3074 return 0;
3075
3076 BUG_ON(node->processed);
3077 root = select_one_root(node);
3078 if (root == ERR_PTR(-ENOENT)) {
3079 update_processed_blocks(rc, node);
3080 goto out;
3081 }
3082
3083 if (!root || test_bit(BTRFS_ROOT_REF_COWS, &root->state)) {
3084 ret = reserve_metadata_space(trans, rc, node);
3085 if (ret)
3086 goto out;
3087 }
3088
3089 if (root) {
3090 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state)) {
3091 BUG_ON(node->new_bytenr);
3092 BUG_ON(!list_empty(&node->list));
3093 btrfs_record_root_in_trans(trans, root);
3094 root = root->reloc_root;
3095 node->new_bytenr = root->node->start;
3096 node->root = root;
3097 list_add_tail(&node->list, &rc->backref_cache.changed);
3098 } else {
3099 path->lowest_level = node->level;
3100 ret = btrfs_search_slot(trans, root, key, path, 0, 1);
3101 btrfs_release_path(path);
3102 if (ret > 0)
3103 ret = 0;
3104 }
3105 if (!ret)
3106 update_processed_blocks(rc, node);
3107 } else {
3108 ret = do_relocation(trans, rc, node, key, path, 1);
3109 }
3110 out:
3111 if (ret || node->level == 0 || node->cowonly)
3112 remove_backref_node(&rc->backref_cache, node);
3113 return ret;
3114 }
3115
3116 /*
3117 * relocate a list of blocks
3118 */
3119 static noinline_for_stack
3120 int relocate_tree_blocks(struct btrfs_trans_handle *trans,
3121 struct reloc_control *rc, struct rb_root *blocks)
3122 {
3123 struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
3124 struct backref_node *node;
3125 struct btrfs_path *path;
3126 struct tree_block *block;
3127 struct tree_block *next;
3128 int ret;
3129 int err = 0;
3130
3131 path = btrfs_alloc_path();
3132 if (!path) {
3133 err = -ENOMEM;
3134 goto out_free_blocks;
3135 }
3136
3137 /* Kick in readahead for tree blocks with missing keys */
3138 rbtree_postorder_for_each_entry_safe(block, next, blocks, rb_node) {
3139 if (!block->key_ready)
3140 readahead_tree_block(fs_info, block->bytenr);
3141 }
3142
3143 /* Get first keys */
3144 rbtree_postorder_for_each_entry_safe(block, next, blocks, rb_node) {
3145 if (!block->key_ready) {
3146 err = get_tree_block_key(fs_info, block);
3147 if (err)
3148 goto out_free_path;
3149 }
3150 }
3151
3152 /* Do tree relocation */
3153 rbtree_postorder_for_each_entry_safe(block, next, blocks, rb_node) {
3154 node = build_backref_tree(rc, &block->key,
3155 block->level, block->bytenr);
3156 if (IS_ERR(node)) {
3157 err = PTR_ERR(node);
3158 goto out;
3159 }
3160
3161 ret = relocate_tree_block(trans, rc, node, &block->key,
3162 path);
3163 if (ret < 0) {
3164 if (ret != -EAGAIN || &block->rb_node == rb_first(blocks))
3165 err = ret;
3166 goto out;
3167 }
3168 }
3169 out:
3170 err = finish_pending_nodes(trans, rc, path, err);
3171
3172 out_free_path:
3173 btrfs_free_path(path);
3174 out_free_blocks:
3175 free_block_list(blocks);
3176 return err;
3177 }
3178
3179 static noinline_for_stack
3180 int prealloc_file_extent_cluster(struct inode *inode,
3181 struct file_extent_cluster *cluster)
3182 {
3183 u64 alloc_hint = 0;
3184 u64 start;
3185 u64 end;
3186 u64 offset = BTRFS_I(inode)->index_cnt;
3187 u64 num_bytes;
3188 int nr = 0;
3189 int ret = 0;
3190 u64 prealloc_start = cluster->start - offset;
3191 u64 prealloc_end = cluster->end - offset;
3192 u64 cur_offset;
3193 struct extent_changeset *data_reserved = NULL;
3194
3195 BUG_ON(cluster->start != cluster->boundary[0]);
3196 inode_lock(inode);
3197
3198 ret = btrfs_check_data_free_space(inode, &data_reserved, prealloc_start,
3199 prealloc_end + 1 - prealloc_start);
3200 if (ret)
3201 goto out;
3202
3203 cur_offset = prealloc_start;
3204 while (nr < cluster->nr) {
3205 start = cluster->boundary[nr] - offset;
3206 if (nr + 1 < cluster->nr)
3207 end = cluster->boundary[nr + 1] - 1 - offset;
3208 else
3209 end = cluster->end - offset;
3210
3211 lock_extent(&BTRFS_I(inode)->io_tree, start, end);
3212 num_bytes = end + 1 - start;
3213 if (cur_offset < start)
3214 btrfs_free_reserved_data_space(inode, data_reserved,
3215 cur_offset, start - cur_offset);
3216 ret = btrfs_prealloc_file_range(inode, 0, start,
3217 num_bytes, num_bytes,
3218 end + 1, &alloc_hint);
3219 cur_offset = end + 1;
3220 unlock_extent(&BTRFS_I(inode)->io_tree, start, end);
3221 if (ret)
3222 break;
3223 nr++;
3224 }
3225 if (cur_offset < prealloc_end)
3226 btrfs_free_reserved_data_space(inode, data_reserved,
3227 cur_offset, prealloc_end + 1 - cur_offset);
3228 out:
3229 inode_unlock(inode);
3230 extent_changeset_free(data_reserved);
3231 return ret;
3232 }
3233
3234 static noinline_for_stack
3235 int setup_extent_mapping(struct inode *inode, u64 start, u64 end,
3236 u64 block_start)
3237 {
3238 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
3239 struct extent_map *em;
3240 int ret = 0;
3241
3242 em = alloc_extent_map();
3243 if (!em)
3244 return -ENOMEM;
3245
3246 em->start = start;
3247 em->len = end + 1 - start;
3248 em->block_len = em->len;
3249 em->block_start = block_start;
3250 set_bit(EXTENT_FLAG_PINNED, &em->flags);
3251
3252 lock_extent(&BTRFS_I(inode)->io_tree, start, end);
3253 while (1) {
3254 write_lock(&em_tree->lock);
3255 ret = add_extent_mapping(em_tree, em, 0);
3256 write_unlock(&em_tree->lock);
3257 if (ret != -EEXIST) {
3258 free_extent_map(em);
3259 break;
3260 }
3261 btrfs_drop_extent_cache(BTRFS_I(inode), start, end, 0);
3262 }
3263 unlock_extent(&BTRFS_I(inode)->io_tree, start, end);
3264 return ret;
3265 }
3266
3267 static int relocate_file_extent_cluster(struct inode *inode,
3268 struct file_extent_cluster *cluster)
3269 {
3270 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3271 u64 page_start;
3272 u64 page_end;
3273 u64 offset = BTRFS_I(inode)->index_cnt;
3274 unsigned long index;
3275 unsigned long last_index;
3276 struct page *page;
3277 struct file_ra_state *ra;
3278 gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping);
3279 int nr = 0;
3280 int ret = 0;
3281
3282 if (!cluster->nr)
3283 return 0;
3284
3285 ra = kzalloc(sizeof(*ra), GFP_NOFS);
3286 if (!ra)
3287 return -ENOMEM;
3288
3289 ret = prealloc_file_extent_cluster(inode, cluster);
3290 if (ret)
3291 goto out;
3292
3293 file_ra_state_init(ra, inode->i_mapping);
3294
3295 ret = setup_extent_mapping(inode, cluster->start - offset,
3296 cluster->end - offset, cluster->start);
3297 if (ret)
3298 goto out;
3299
3300 index = (cluster->start - offset) >> PAGE_SHIFT;
3301 last_index = (cluster->end - offset) >> PAGE_SHIFT;
3302 while (index <= last_index) {
3303 ret = btrfs_delalloc_reserve_metadata(BTRFS_I(inode),
3304 PAGE_SIZE);
3305 if (ret)
3306 goto out;
3307
3308 page = find_lock_page(inode->i_mapping, index);
3309 if (!page) {
3310 page_cache_sync_readahead(inode->i_mapping,
3311 ra, NULL, index,
3312 last_index + 1 - index);
3313 page = find_or_create_page(inode->i_mapping, index,
3314 mask);
3315 if (!page) {
3316 btrfs_delalloc_release_metadata(BTRFS_I(inode),
3317 PAGE_SIZE, true);
3318 btrfs_delalloc_release_extents(BTRFS_I(inode),
3319 PAGE_SIZE);
3320 ret = -ENOMEM;
3321 goto out;
3322 }
3323 }
3324
3325 if (PageReadahead(page)) {
3326 page_cache_async_readahead(inode->i_mapping,
3327 ra, NULL, page, index,
3328 last_index + 1 - index);
3329 }
3330
3331 if (!PageUptodate(page)) {
3332 btrfs_readpage(NULL, page);
3333 lock_page(page);
3334 if (!PageUptodate(page)) {
3335 unlock_page(page);
3336 put_page(page);
3337 btrfs_delalloc_release_metadata(BTRFS_I(inode),
3338 PAGE_SIZE, true);
3339 btrfs_delalloc_release_extents(BTRFS_I(inode),
3340 PAGE_SIZE);
3341 ret = -EIO;
3342 goto out;
3343 }
3344 }
3345
3346 page_start = page_offset(page);
3347 page_end = page_start + PAGE_SIZE - 1;
3348
3349 lock_extent(&BTRFS_I(inode)->io_tree, page_start, page_end);
3350
3351 set_page_extent_mapped(page);
3352
3353 if (nr < cluster->nr &&
3354 page_start + offset == cluster->boundary[nr]) {
3355 set_extent_bits(&BTRFS_I(inode)->io_tree,
3356 page_start, page_end,
3357 EXTENT_BOUNDARY);
3358 nr++;
3359 }
3360
3361 ret = btrfs_set_extent_delalloc(inode, page_start, page_end, 0,
3362 NULL);
3363 if (ret) {
3364 unlock_page(page);
3365 put_page(page);
3366 btrfs_delalloc_release_metadata(BTRFS_I(inode),
3367 PAGE_SIZE, true);
3368 btrfs_delalloc_release_extents(BTRFS_I(inode),
3369 PAGE_SIZE);
3370
3371 clear_extent_bits(&BTRFS_I(inode)->io_tree,
3372 page_start, page_end,
3373 EXTENT_LOCKED | EXTENT_BOUNDARY);
3374 goto out;
3375
3376 }
3377 set_page_dirty(page);
3378
3379 unlock_extent(&BTRFS_I(inode)->io_tree,
3380 page_start, page_end);
3381 unlock_page(page);
3382 put_page(page);
3383
3384 index++;
3385 btrfs_delalloc_release_extents(BTRFS_I(inode), PAGE_SIZE);
3386 balance_dirty_pages_ratelimited(inode->i_mapping);
3387 btrfs_throttle(fs_info);
3388 }
3389 WARN_ON(nr != cluster->nr);
3390 out:
3391 kfree(ra);
3392 return ret;
3393 }
3394
3395 static noinline_for_stack
3396 int relocate_data_extent(struct inode *inode, struct btrfs_key *extent_key,
3397 struct file_extent_cluster *cluster)
3398 {
3399 int ret;
3400
3401 if (cluster->nr > 0 && extent_key->objectid != cluster->end + 1) {
3402 ret = relocate_file_extent_cluster(inode, cluster);
3403 if (ret)
3404 return ret;
3405 cluster->nr = 0;
3406 }
3407
3408 if (!cluster->nr)
3409 cluster->start = extent_key->objectid;
3410 else
3411 BUG_ON(cluster->nr >= MAX_EXTENTS);
3412 cluster->end = extent_key->objectid + extent_key->offset - 1;
3413 cluster->boundary[cluster->nr] = extent_key->objectid;
3414 cluster->nr++;
3415
3416 if (cluster->nr >= MAX_EXTENTS) {
3417 ret = relocate_file_extent_cluster(inode, cluster);
3418 if (ret)
3419 return ret;
3420 cluster->nr = 0;
3421 }
3422 return 0;
3423 }
3424
3425 /*
3426 * helper to add a tree block to the list.
3427 * the major work is getting the generation and level of the block
3428 */
3429 static int add_tree_block(struct reloc_control *rc,
3430 struct btrfs_key *extent_key,
3431 struct btrfs_path *path,
3432 struct rb_root *blocks)
3433 {
3434 struct extent_buffer *eb;
3435 struct btrfs_extent_item *ei;
3436 struct btrfs_tree_block_info *bi;
3437 struct tree_block *block;
3438 struct rb_node *rb_node;
3439 u32 item_size;
3440 int level = -1;
3441 u64 generation;
3442
3443 eb = path->nodes[0];
3444 item_size = btrfs_item_size_nr(eb, path->slots[0]);
3445
3446 if (extent_key->type == BTRFS_METADATA_ITEM_KEY ||
3447 item_size >= sizeof(*ei) + sizeof(*bi)) {
3448 ei = btrfs_item_ptr(eb, path->slots[0],
3449 struct btrfs_extent_item);
3450 if (extent_key->type == BTRFS_EXTENT_ITEM_KEY) {
3451 bi = (struct btrfs_tree_block_info *)(ei + 1);
3452 level = btrfs_tree_block_level(eb, bi);
3453 } else {
3454 level = (int)extent_key->offset;
3455 }
3456 generation = btrfs_extent_generation(eb, ei);
3457 } else if (unlikely(item_size == sizeof(struct btrfs_extent_item_v0))) {
3458 btrfs_print_v0_err(eb->fs_info);
3459 btrfs_handle_fs_error(eb->fs_info, -EINVAL, NULL);
3460 return -EINVAL;
3461 } else {
3462 BUG();
3463 }
3464
3465 btrfs_release_path(path);
3466
3467 BUG_ON(level == -1);
3468
3469 block = kmalloc(sizeof(*block), GFP_NOFS);
3470 if (!block)
3471 return -ENOMEM;
3472
3473 block->bytenr = extent_key->objectid;
3474 block->key.objectid = rc->extent_root->fs_info->nodesize;
3475 block->key.offset = generation;
3476 block->level = level;
3477 block->key_ready = 0;
3478
3479 rb_node = tree_insert(blocks, block->bytenr, &block->rb_node);
3480 if (rb_node)
3481 backref_tree_panic(rb_node, -EEXIST, block->bytenr);
3482
3483 return 0;
3484 }
3485
3486 /*
3487 * helper to add tree blocks for backref of type BTRFS_SHARED_DATA_REF_KEY
3488 */
3489 static int __add_tree_block(struct reloc_control *rc,
3490 u64 bytenr, u32 blocksize,
3491 struct rb_root *blocks)
3492 {
3493 struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
3494 struct btrfs_path *path;
3495 struct btrfs_key key;
3496 int ret;
3497 bool skinny = btrfs_fs_incompat(fs_info, SKINNY_METADATA);
3498
3499 if (tree_block_processed(bytenr, rc))
3500 return 0;
3501
3502 if (tree_search(blocks, bytenr))
3503 return 0;
3504
3505 path = btrfs_alloc_path();
3506 if (!path)
3507 return -ENOMEM;
3508 again:
3509 key.objectid = bytenr;
3510 if (skinny) {
3511 key.type = BTRFS_METADATA_ITEM_KEY;
3512 key.offset = (u64)-1;
3513 } else {
3514 key.type = BTRFS_EXTENT_ITEM_KEY;
3515 key.offset = blocksize;
3516 }
3517
3518 path->search_commit_root = 1;
3519 path->skip_locking = 1;
3520 ret = btrfs_search_slot(NULL, rc->extent_root, &key, path, 0, 0);
3521 if (ret < 0)
3522 goto out;
3523
3524 if (ret > 0 && skinny) {
3525 if (path->slots[0]) {
3526 path->slots[0]--;
3527 btrfs_item_key_to_cpu(path->nodes[0], &key,
3528 path->slots[0]);
3529 if (key.objectid == bytenr &&
3530 (key.type == BTRFS_METADATA_ITEM_KEY ||
3531 (key.type == BTRFS_EXTENT_ITEM_KEY &&
3532 key.offset == blocksize)))
3533 ret = 0;
3534 }
3535
3536 if (ret) {
3537 skinny = false;
3538 btrfs_release_path(path);
3539 goto again;
3540 }
3541 }
3542 if (ret) {
3543 ASSERT(ret == 1);
3544 btrfs_print_leaf(path->nodes[0]);
3545 btrfs_err(fs_info,
3546 "tree block extent item (%llu) is not found in extent tree",
3547 bytenr);
3548 WARN_ON(1);
3549 ret = -EINVAL;
3550 goto out;
3551 }
3552
3553 ret = add_tree_block(rc, &key, path, blocks);
3554 out:
3555 btrfs_free_path(path);
3556 return ret;
3557 }
3558
3559 /*
3560 * helper to check if the block use full backrefs for pointers in it
3561 */
3562 static int block_use_full_backref(struct reloc_control *rc,
3563 struct extent_buffer *eb)
3564 {
3565 u64 flags;
3566 int ret;
3567
3568 if (btrfs_header_flag(eb, BTRFS_HEADER_FLAG_RELOC) ||
3569 btrfs_header_backref_rev(eb) < BTRFS_MIXED_BACKREF_REV)
3570 return 1;
3571
3572 ret = btrfs_lookup_extent_info(NULL, rc->extent_root->fs_info,
3573 eb->start, btrfs_header_level(eb), 1,
3574 NULL, &flags);
3575 BUG_ON(ret);
3576
3577 if (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)
3578 ret = 1;
3579 else
3580 ret = 0;
3581 return ret;
3582 }
3583
3584 static int delete_block_group_cache(struct btrfs_fs_info *fs_info,
3585 struct btrfs_block_group *block_group,
3586 struct inode *inode,
3587 u64 ino)
3588 {
3589 struct btrfs_key key;
3590 struct btrfs_root *root = fs_info->tree_root;
3591 struct btrfs_trans_handle *trans;
3592 int ret = 0;
3593
3594 if (inode)
3595 goto truncate;
3596
3597 key.objectid = ino;
3598 key.type = BTRFS_INODE_ITEM_KEY;
3599 key.offset = 0;
3600
3601 inode = btrfs_iget(fs_info->sb, &key, root);
3602 if (IS_ERR(inode))
3603 return -ENOENT;
3604
3605 truncate:
3606 ret = btrfs_check_trunc_cache_free_space(fs_info,
3607 &fs_info->global_block_rsv);
3608 if (ret)
3609 goto out;
3610
3611 trans = btrfs_join_transaction(root);
3612 if (IS_ERR(trans)) {
3613 ret = PTR_ERR(trans);
3614 goto out;
3615 }
3616
3617 ret = btrfs_truncate_free_space_cache(trans, block_group, inode);
3618
3619 btrfs_end_transaction(trans);
3620 btrfs_btree_balance_dirty(fs_info);
3621 out:
3622 iput(inode);
3623 return ret;
3624 }
3625
3626 /*
3627 * helper to add tree blocks for backref of type BTRFS_EXTENT_DATA_REF_KEY
3628 * this function scans fs tree to find blocks reference the data extent
3629 */
3630 static int find_data_references(struct reloc_control *rc,
3631 struct btrfs_key *extent_key,
3632 struct extent_buffer *leaf,
3633 struct btrfs_extent_data_ref *ref,
3634 struct rb_root *blocks)
3635 {
3636 struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
3637 struct btrfs_path *path;
3638 struct tree_block *block;
3639 struct btrfs_root *root;
3640 struct btrfs_file_extent_item *fi;
3641 struct rb_node *rb_node;
3642 struct btrfs_key key;
3643 u64 ref_root;
3644 u64 ref_objectid;
3645 u64 ref_offset;
3646 u32 ref_count;
3647 u32 nritems;
3648 int err = 0;
3649 int added = 0;
3650 int counted;
3651 int ret;
3652
3653 ref_root = btrfs_extent_data_ref_root(leaf, ref);
3654 ref_objectid = btrfs_extent_data_ref_objectid(leaf, ref);
3655 ref_offset = btrfs_extent_data_ref_offset(leaf, ref);
3656 ref_count = btrfs_extent_data_ref_count(leaf, ref);
3657
3658 /*
3659 * This is an extent belonging to the free space cache, lets just delete
3660 * it and redo the search.
3661 */
3662 if (ref_root == BTRFS_ROOT_TREE_OBJECTID) {
3663 ret = delete_block_group_cache(fs_info, rc->block_group,
3664 NULL, ref_objectid);
3665 if (ret != -ENOENT)
3666 return ret;
3667 ret = 0;
3668 }
3669
3670 path = btrfs_alloc_path();
3671 if (!path)
3672 return -ENOMEM;
3673 path->reada = READA_FORWARD;
3674
3675 root = read_fs_root(fs_info, ref_root);
3676 if (IS_ERR(root)) {
3677 err = PTR_ERR(root);
3678 goto out;
3679 }
3680
3681 key.objectid = ref_objectid;
3682 key.type = BTRFS_EXTENT_DATA_KEY;
3683 if (ref_offset > ((u64)-1 << 32))
3684 key.offset = 0;
3685 else
3686 key.offset = ref_offset;
3687
3688 path->search_commit_root = 1;
3689 path->skip_locking = 1;
3690 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3691 if (ret < 0) {
3692 err = ret;
3693 goto out;
3694 }
3695
3696 leaf = path->nodes[0];
3697 nritems = btrfs_header_nritems(leaf);
3698 /*
3699 * the references in tree blocks that use full backrefs
3700 * are not counted in
3701 */
3702 if (block_use_full_backref(rc, leaf))
3703 counted = 0;
3704 else
3705 counted = 1;
3706 rb_node = tree_search(blocks, leaf->start);
3707 if (rb_node) {
3708 if (counted)
3709 added = 1;
3710 else
3711 path->slots[0] = nritems;
3712 }
3713
3714 while (ref_count > 0) {
3715 while (path->slots[0] >= nritems) {
3716 ret = btrfs_next_leaf(root, path);
3717 if (ret < 0) {
3718 err = ret;
3719 goto out;
3720 }
3721 if (WARN_ON(ret > 0))
3722 goto out;
3723
3724 leaf = path->nodes[0];
3725 nritems = btrfs_header_nritems(leaf);
3726 added = 0;
3727
3728 if (block_use_full_backref(rc, leaf))
3729 counted = 0;
3730 else
3731 counted = 1;
3732 rb_node = tree_search(blocks, leaf->start);
3733 if (rb_node) {
3734 if (counted)
3735 added = 1;
3736 else
3737 path->slots[0] = nritems;
3738 }
3739 }
3740
3741 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3742 if (WARN_ON(key.objectid != ref_objectid ||
3743 key.type != BTRFS_EXTENT_DATA_KEY))
3744 break;
3745
3746 fi = btrfs_item_ptr(leaf, path->slots[0],
3747 struct btrfs_file_extent_item);
3748
3749 if (btrfs_file_extent_type(leaf, fi) ==
3750 BTRFS_FILE_EXTENT_INLINE)
3751 goto next;
3752
3753 if (btrfs_file_extent_disk_bytenr(leaf, fi) !=
3754 extent_key->objectid)
3755 goto next;
3756
3757 key.offset -= btrfs_file_extent_offset(leaf, fi);
3758 if (key.offset != ref_offset)
3759 goto next;
3760
3761 if (counted)
3762 ref_count--;
3763 if (added)
3764 goto next;
3765
3766 if (!tree_block_processed(leaf->start, rc)) {
3767 block = kmalloc(sizeof(*block), GFP_NOFS);
3768 if (!block) {
3769 err = -ENOMEM;
3770 break;
3771 }
3772 block->bytenr = leaf->start;
3773 btrfs_item_key_to_cpu(leaf, &block->key, 0);
3774 block->level = 0;
3775 block->key_ready = 1;
3776 rb_node = tree_insert(blocks, block->bytenr,
3777 &block->rb_node);
3778 if (rb_node)
3779 backref_tree_panic(rb_node, -EEXIST,
3780 block->bytenr);
3781 }
3782 if (counted)
3783 added = 1;
3784 else
3785 path->slots[0] = nritems;
3786 next:
3787 path->slots[0]++;
3788
3789 }
3790 out:
3791 btrfs_free_path(path);
3792 return err;
3793 }
3794
3795 /*
3796 * helper to find all tree blocks that reference a given data extent
3797 */
3798 static noinline_for_stack
3799 int add_data_references(struct reloc_control *rc,
3800 struct btrfs_key *extent_key,
3801 struct btrfs_path *path,
3802 struct rb_root *blocks)
3803 {
3804 struct btrfs_key key;
3805 struct extent_buffer *eb;
3806 struct btrfs_extent_data_ref *dref;
3807 struct btrfs_extent_inline_ref *iref;
3808 unsigned long ptr;
3809 unsigned long end;
3810 u32 blocksize = rc->extent_root->fs_info->nodesize;
3811 int ret = 0;
3812 int err = 0;
3813
3814 eb = path->nodes[0];
3815 ptr = btrfs_item_ptr_offset(eb, path->slots[0]);
3816 end = ptr + btrfs_item_size_nr(eb, path->slots[0]);
3817 ptr += sizeof(struct btrfs_extent_item);
3818
3819 while (ptr < end) {
3820 iref = (struct btrfs_extent_inline_ref *)ptr;
3821 key.type = btrfs_get_extent_inline_ref_type(eb, iref,
3822 BTRFS_REF_TYPE_DATA);
3823 if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
3824 key.offset = btrfs_extent_inline_ref_offset(eb, iref);
3825 ret = __add_tree_block(rc, key.offset, blocksize,
3826 blocks);
3827 } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
3828 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
3829 ret = find_data_references(rc, extent_key,
3830 eb, dref, blocks);
3831 } else {
3832 ret = -EUCLEAN;
3833 btrfs_err(rc->extent_root->fs_info,
3834 "extent %llu slot %d has an invalid inline ref type",
3835 eb->start, path->slots[0]);
3836 }
3837 if (ret) {
3838 err = ret;
3839 goto out;
3840 }
3841 ptr += btrfs_extent_inline_ref_size(key.type);
3842 }
3843 WARN_ON(ptr > end);
3844
3845 while (1) {
3846 cond_resched();
3847 eb = path->nodes[0];
3848 if (path->slots[0] >= btrfs_header_nritems(eb)) {
3849 ret = btrfs_next_leaf(rc->extent_root, path);
3850 if (ret < 0) {
3851 err = ret;
3852 break;
3853 }
3854 if (ret > 0)
3855 break;
3856 eb = path->nodes[0];
3857 }
3858
3859 btrfs_item_key_to_cpu(eb, &key, path->slots[0]);
3860 if (key.objectid != extent_key->objectid)
3861 break;
3862
3863 if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
3864 ret = __add_tree_block(rc, key.offset, blocksize,
3865 blocks);
3866 } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
3867 dref = btrfs_item_ptr(eb, path->slots[0],
3868 struct btrfs_extent_data_ref);
3869 ret = find_data_references(rc, extent_key,
3870 eb, dref, blocks);
3871 } else if (unlikely(key.type == BTRFS_EXTENT_REF_V0_KEY)) {
3872 btrfs_print_v0_err(eb->fs_info);
3873 btrfs_handle_fs_error(eb->fs_info, -EINVAL, NULL);
3874 ret = -EINVAL;
3875 } else {
3876 ret = 0;
3877 }
3878 if (ret) {
3879 err = ret;
3880 break;
3881 }
3882 path->slots[0]++;
3883 }
3884 out:
3885 btrfs_release_path(path);
3886 if (err)
3887 free_block_list(blocks);
3888 return err;
3889 }
3890
3891 /*
3892 * helper to find next unprocessed extent
3893 */
3894 static noinline_for_stack
3895 int find_next_extent(struct reloc_control *rc, struct btrfs_path *path,
3896 struct btrfs_key *extent_key)
3897 {
3898 struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
3899 struct btrfs_key key;
3900 struct extent_buffer *leaf;
3901 u64 start, end, last;
3902 int ret;
3903
3904 last = rc->block_group->start + rc->block_group->length;
3905 while (1) {
3906 cond_resched();
3907 if (rc->search_start >= last) {
3908 ret = 1;
3909 break;
3910 }
3911
3912 key.objectid = rc->search_start;
3913 key.type = BTRFS_EXTENT_ITEM_KEY;
3914 key.offset = 0;
3915
3916 path->search_commit_root = 1;
3917 path->skip_locking = 1;
3918 ret = btrfs_search_slot(NULL, rc->extent_root, &key, path,
3919 0, 0);
3920 if (ret < 0)
3921 break;
3922 next:
3923 leaf = path->nodes[0];
3924 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
3925 ret = btrfs_next_leaf(rc->extent_root, path);
3926 if (ret != 0)
3927 break;
3928 leaf = path->nodes[0];
3929 }
3930
3931 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3932 if (key.objectid >= last) {
3933 ret = 1;
3934 break;
3935 }
3936
3937 if (key.type != BTRFS_EXTENT_ITEM_KEY &&
3938 key.type != BTRFS_METADATA_ITEM_KEY) {
3939 path->slots[0]++;
3940 goto next;
3941 }
3942
3943 if (key.type == BTRFS_EXTENT_ITEM_KEY &&
3944 key.objectid + key.offset <= rc->search_start) {
3945 path->slots[0]++;
3946 goto next;
3947 }
3948
3949 if (key.type == BTRFS_METADATA_ITEM_KEY &&
3950 key.objectid + fs_info->nodesize <=
3951 rc->search_start) {
3952 path->slots[0]++;
3953 goto next;
3954 }
3955
3956 ret = find_first_extent_bit(&rc->processed_blocks,
3957 key.objectid, &start, &end,
3958 EXTENT_DIRTY, NULL);
3959
3960 if (ret == 0 && start <= key.objectid) {
3961 btrfs_release_path(path);
3962 rc->search_start = end + 1;
3963 } else {
3964 if (key.type == BTRFS_EXTENT_ITEM_KEY)
3965 rc->search_start = key.objectid + key.offset;
3966 else
3967 rc->search_start = key.objectid +
3968 fs_info->nodesize;
3969 memcpy(extent_key, &key, sizeof(key));
3970 return 0;
3971 }
3972 }
3973 btrfs_release_path(path);
3974 return ret;
3975 }
3976
3977 static void set_reloc_control(struct reloc_control *rc)
3978 {
3979 struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
3980
3981 mutex_lock(&fs_info->reloc_mutex);
3982 fs_info->reloc_ctl = rc;
3983 mutex_unlock(&fs_info->reloc_mutex);
3984 }
3985
3986 static void unset_reloc_control(struct reloc_control *rc)
3987 {
3988 struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
3989
3990 mutex_lock(&fs_info->reloc_mutex);
3991 fs_info->reloc_ctl = NULL;
3992 mutex_unlock(&fs_info->reloc_mutex);
3993 }
3994
3995 static int check_extent_flags(u64 flags)
3996 {
3997 if ((flags & BTRFS_EXTENT_FLAG_DATA) &&
3998 (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK))
3999 return 1;
4000 if (!(flags & BTRFS_EXTENT_FLAG_DATA) &&
4001 !(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK))
4002 return 1;
4003 if ((flags & BTRFS_EXTENT_FLAG_DATA) &&
4004 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
4005 return 1;
4006 return 0;
4007 }
4008
4009 static noinline_for_stack
4010 int prepare_to_relocate(struct reloc_control *rc)
4011 {
4012 struct btrfs_trans_handle *trans;
4013 int ret;
4014
4015 rc->block_rsv = btrfs_alloc_block_rsv(rc->extent_root->fs_info,
4016 BTRFS_BLOCK_RSV_TEMP);
4017 if (!rc->block_rsv)
4018 return -ENOMEM;
4019
4020 memset(&rc->cluster, 0, sizeof(rc->cluster));
4021 rc->search_start = rc->block_group->start;
4022 rc->extents_found = 0;
4023 rc->nodes_relocated = 0;
4024 rc->merging_rsv_size = 0;
4025 rc->reserved_bytes = 0;
4026 rc->block_rsv->size = rc->extent_root->fs_info->nodesize *
4027 RELOCATION_RESERVED_NODES;
4028 ret = btrfs_block_rsv_refill(rc->extent_root,
4029 rc->block_rsv, rc->block_rsv->size,
4030 BTRFS_RESERVE_FLUSH_ALL);
4031 if (ret)
4032 return ret;
4033
4034 rc->create_reloc_tree = 1;
4035 set_reloc_control(rc);
4036
4037 trans = btrfs_join_transaction(rc->extent_root);
4038 if (IS_ERR(trans)) {
4039 unset_reloc_control(rc);
4040 /*
4041 * extent tree is not a ref_cow tree and has no reloc_root to
4042 * cleanup. And callers are responsible to free the above
4043 * block rsv.
4044 */
4045 return PTR_ERR(trans);
4046 }
4047 btrfs_commit_transaction(trans);
4048 return 0;
4049 }
4050
4051 static noinline_for_stack int relocate_block_group(struct reloc_control *rc)
4052 {
4053 struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
4054 struct rb_root blocks = RB_ROOT;
4055 struct btrfs_key key;
4056 struct btrfs_trans_handle *trans = NULL;
4057 struct btrfs_path *path;
4058 struct btrfs_extent_item *ei;
4059 u64 flags;
4060 u32 item_size;
4061 int ret;
4062 int err = 0;
4063 int progress = 0;
4064
4065 path = btrfs_alloc_path();
4066 if (!path)
4067 return -ENOMEM;
4068 path->reada = READA_FORWARD;
4069
4070 ret = prepare_to_relocate(rc);
4071 if (ret) {
4072 err = ret;
4073 goto out_free;
4074 }
4075
4076 while (1) {
4077 rc->reserved_bytes = 0;
4078 ret = btrfs_block_rsv_refill(rc->extent_root,
4079 rc->block_rsv, rc->block_rsv->size,
4080 BTRFS_RESERVE_FLUSH_ALL);
4081 if (ret) {
4082 err = ret;
4083 break;
4084 }
4085 progress++;
4086 trans = btrfs_start_transaction(rc->extent_root, 0);
4087 if (IS_ERR(trans)) {
4088 err = PTR_ERR(trans);
4089 trans = NULL;
4090 break;
4091 }
4092 restart:
4093 if (update_backref_cache(trans, &rc->backref_cache)) {
4094 btrfs_end_transaction(trans);
4095 trans = NULL;
4096 continue;
4097 }
4098
4099 ret = find_next_extent(rc, path, &key);
4100 if (ret < 0)
4101 err = ret;
4102 if (ret != 0)
4103 break;
4104
4105 rc->extents_found++;
4106
4107 ei = btrfs_item_ptr(path->nodes[0], path->slots[0],
4108 struct btrfs_extent_item);
4109 item_size = btrfs_item_size_nr(path->nodes[0], path->slots[0]);
4110 if (item_size >= sizeof(*ei)) {
4111 flags = btrfs_extent_flags(path->nodes[0], ei);
4112 ret = check_extent_flags(flags);
4113 BUG_ON(ret);
4114 } else if (unlikely(item_size == sizeof(struct btrfs_extent_item_v0))) {
4115 err = -EINVAL;
4116 btrfs_print_v0_err(trans->fs_info);
4117 btrfs_abort_transaction(trans, err);
4118 break;
4119 } else {
4120 BUG();
4121 }
4122
4123 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
4124 ret = add_tree_block(rc, &key, path, &blocks);
4125 } else if (rc->stage == UPDATE_DATA_PTRS &&
4126 (flags & BTRFS_EXTENT_FLAG_DATA)) {
4127 ret = add_data_references(rc, &key, path, &blocks);
4128 } else {
4129 btrfs_release_path(path);
4130 ret = 0;
4131 }
4132 if (ret < 0) {
4133 err = ret;
4134 break;
4135 }
4136
4137 if (!RB_EMPTY_ROOT(&blocks)) {
4138 ret = relocate_tree_blocks(trans, rc, &blocks);
4139 if (ret < 0) {
4140 /*
4141 * if we fail to relocate tree blocks, force to update
4142 * backref cache when committing transaction.
4143 */
4144 rc->backref_cache.last_trans = trans->transid - 1;
4145
4146 if (ret != -EAGAIN) {
4147 err = ret;
4148 break;
4149 }
4150 rc->extents_found--;
4151 rc->search_start = key.objectid;
4152 }
4153 }
4154
4155 btrfs_end_transaction_throttle(trans);
4156 btrfs_btree_balance_dirty(fs_info);
4157 trans = NULL;
4158
4159 if (rc->stage == MOVE_DATA_EXTENTS &&
4160 (flags & BTRFS_EXTENT_FLAG_DATA)) {
4161 rc->found_file_extent = 1;
4162 ret = relocate_data_extent(rc->data_inode,
4163 &key, &rc->cluster);
4164 if (ret < 0) {
4165 err = ret;
4166 break;
4167 }
4168 }
4169 }
4170 if (trans && progress && err == -ENOSPC) {
4171 ret = btrfs_force_chunk_alloc(trans, rc->block_group->flags);
4172 if (ret == 1) {
4173 err = 0;
4174 progress = 0;
4175 goto restart;
4176 }
4177 }
4178
4179 btrfs_release_path(path);
4180 clear_extent_bits(&rc->processed_blocks, 0, (u64)-1, EXTENT_DIRTY);
4181
4182 if (trans) {
4183 btrfs_end_transaction_throttle(trans);
4184 btrfs_btree_balance_dirty(fs_info);
4185 }
4186
4187 if (!err) {
4188 ret = relocate_file_extent_cluster(rc->data_inode,
4189 &rc->cluster);
4190 if (ret < 0)
4191 err = ret;
4192 }
4193
4194 rc->create_reloc_tree = 0;
4195 set_reloc_control(rc);
4196
4197 backref_cache_cleanup(&rc->backref_cache);
4198 btrfs_block_rsv_release(fs_info, rc->block_rsv, (u64)-1);
4199
4200 err = prepare_to_merge(rc, err);
4201
4202 merge_reloc_roots(rc);
4203
4204 rc->merge_reloc_tree = 0;
4205 unset_reloc_control(rc);
4206 btrfs_block_rsv_release(fs_info, rc->block_rsv, (u64)-1);
4207
4208 /* get rid of pinned extents */
4209 trans = btrfs_join_transaction(rc->extent_root);
4210 if (IS_ERR(trans)) {
4211 err = PTR_ERR(trans);
4212 goto out_free;
4213 }
4214 btrfs_commit_transaction(trans);
4215 ret = clean_dirty_subvols(rc);
4216 if (ret < 0 && !err)
4217 err = ret;
4218 out_free:
4219 btrfs_free_block_rsv(fs_info, rc->block_rsv);
4220 btrfs_free_path(path);
4221 return err;
4222 }
4223
4224 static int __insert_orphan_inode(struct btrfs_trans_handle *trans,
4225 struct btrfs_root *root, u64 objectid)
4226 {
4227 struct btrfs_path *path;
4228 struct btrfs_inode_item *item;
4229 struct extent_buffer *leaf;
4230 int ret;
4231
4232 path = btrfs_alloc_path();
4233 if (!path)
4234 return -ENOMEM;
4235
4236 ret = btrfs_insert_empty_inode(trans, root, path, objectid);
4237 if (ret)
4238 goto out;
4239
4240 leaf = path->nodes[0];
4241 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_inode_item);
4242 memzero_extent_buffer(leaf, (unsigned long)item, sizeof(*item));
4243 btrfs_set_inode_generation(leaf, item, 1);
4244 btrfs_set_inode_size(leaf, item, 0);
4245 btrfs_set_inode_mode(leaf, item, S_IFREG | 0600);
4246 btrfs_set_inode_flags(leaf, item, BTRFS_INODE_NOCOMPRESS |
4247 BTRFS_INODE_PREALLOC);
4248 btrfs_mark_buffer_dirty(leaf);
4249 out:
4250 btrfs_free_path(path);
4251 return ret;
4252 }
4253
4254 /*
4255 * helper to create inode for data relocation.
4256 * the inode is in data relocation tree and its link count is 0
4257 */
4258 static noinline_for_stack
4259 struct inode *create_reloc_inode(struct btrfs_fs_info *fs_info,
4260 struct btrfs_block_group *group)
4261 {
4262 struct inode *inode = NULL;
4263 struct btrfs_trans_handle *trans;
4264 struct btrfs_root *root;
4265 struct btrfs_key key;
4266 u64 objectid;
4267 int err = 0;
4268
4269 root = read_fs_root(fs_info, BTRFS_DATA_RELOC_TREE_OBJECTID);
4270 if (IS_ERR(root))
4271 return ERR_CAST(root);
4272
4273 trans = btrfs_start_transaction(root, 6);
4274 if (IS_ERR(trans))
4275 return ERR_CAST(trans);
4276
4277 err = btrfs_find_free_objectid(root, &objectid);
4278 if (err)
4279 goto out;
4280
4281 err = __insert_orphan_inode(trans, root, objectid);
4282 BUG_ON(err);
4283
4284 key.objectid = objectid;
4285 key.type = BTRFS_INODE_ITEM_KEY;
4286 key.offset = 0;
4287 inode = btrfs_iget(fs_info->sb, &key, root);
4288 BUG_ON(IS_ERR(inode));
4289 BTRFS_I(inode)->index_cnt = group->start;
4290
4291 err = btrfs_orphan_add(trans, BTRFS_I(inode));
4292 out:
4293 btrfs_end_transaction(trans);
4294 btrfs_btree_balance_dirty(fs_info);
4295 if (err) {
4296 if (inode)
4297 iput(inode);
4298 inode = ERR_PTR(err);
4299 }
4300 return inode;
4301 }
4302
4303 static struct reloc_control *alloc_reloc_control(struct btrfs_fs_info *fs_info)
4304 {
4305 struct reloc_control *rc;
4306
4307 rc = kzalloc(sizeof(*rc), GFP_NOFS);
4308 if (!rc)
4309 return NULL;
4310
4311 INIT_LIST_HEAD(&rc->reloc_roots);
4312 INIT_LIST_HEAD(&rc->dirty_subvol_roots);
4313 backref_cache_init(&rc->backref_cache);
4314 mapping_tree_init(&rc->reloc_root_tree);
4315 extent_io_tree_init(fs_info, &rc->processed_blocks,
4316 IO_TREE_RELOC_BLOCKS, NULL);
4317 return rc;
4318 }
4319
4320 /*
4321 * Print the block group being relocated
4322 */
4323 static void describe_relocation(struct btrfs_fs_info *fs_info,
4324 struct btrfs_block_group *block_group)
4325 {
4326 char buf[128] = {'\0'};
4327
4328 btrfs_describe_block_groups(block_group->flags, buf, sizeof(buf));
4329
4330 btrfs_info(fs_info,
4331 "relocating block group %llu flags %s",
4332 block_group->start, buf);
4333 }
4334
4335 static const char *stage_to_string(int stage)
4336 {
4337 if (stage == MOVE_DATA_EXTENTS)
4338 return "move data extents";
4339 if (stage == UPDATE_DATA_PTRS)
4340 return "update data pointers";
4341 return "unknown";
4342 }
4343
4344 /*
4345 * function to relocate all extents in a block group.
4346 */
4347 int btrfs_relocate_block_group(struct btrfs_fs_info *fs_info, u64 group_start)
4348 {
4349 struct btrfs_block_group *bg;
4350 struct btrfs_root *extent_root = fs_info->extent_root;
4351 struct reloc_control *rc;
4352 struct inode *inode;
4353 struct btrfs_path *path;
4354 int ret;
4355 int rw = 0;
4356 int err = 0;
4357
4358 bg = btrfs_lookup_block_group(fs_info, group_start);
4359 if (!bg)
4360 return -ENOENT;
4361
4362 if (btrfs_pinned_by_swapfile(fs_info, bg)) {
4363 btrfs_put_block_group(bg);
4364 return -ETXTBSY;
4365 }
4366
4367 rc = alloc_reloc_control(fs_info);
4368 if (!rc) {
4369 btrfs_put_block_group(bg);
4370 return -ENOMEM;
4371 }
4372
4373 rc->extent_root = extent_root;
4374 rc->block_group = bg;
4375
4376 ret = btrfs_inc_block_group_ro(rc->block_group, true);
4377 if (ret) {
4378 err = ret;
4379 goto out;
4380 }
4381 rw = 1;
4382
4383 path = btrfs_alloc_path();
4384 if (!path) {
4385 err = -ENOMEM;
4386 goto out;
4387 }
4388
4389 inode = lookup_free_space_inode(rc->block_group, path);
4390 btrfs_free_path(path);
4391
4392 if (!IS_ERR(inode))
4393 ret = delete_block_group_cache(fs_info, rc->block_group, inode, 0);
4394 else
4395 ret = PTR_ERR(inode);
4396
4397 if (ret && ret != -ENOENT) {
4398 err = ret;
4399 goto out;
4400 }
4401
4402 rc->data_inode = create_reloc_inode(fs_info, rc->block_group);
4403 if (IS_ERR(rc->data_inode)) {
4404 err = PTR_ERR(rc->data_inode);
4405 rc->data_inode = NULL;
4406 goto out;
4407 }
4408
4409 describe_relocation(fs_info, rc->block_group);
4410
4411 btrfs_wait_block_group_reservations(rc->block_group);
4412 btrfs_wait_nocow_writers(rc->block_group);
4413 btrfs_wait_ordered_roots(fs_info, U64_MAX,
4414 rc->block_group->start,
4415 rc->block_group->length);
4416
4417 while (1) {
4418 int finishes_stage;
4419
4420 mutex_lock(&fs_info->cleaner_mutex);
4421 ret = relocate_block_group(rc);
4422 mutex_unlock(&fs_info->cleaner_mutex);
4423 if (ret < 0)
4424 err = ret;
4425
4426 finishes_stage = rc->stage;
4427 /*
4428 * We may have gotten ENOSPC after we already dirtied some
4429 * extents. If writeout happens while we're relocating a
4430 * different block group we could end up hitting the
4431 * BUG_ON(rc->stage == UPDATE_DATA_PTRS) in
4432 * btrfs_reloc_cow_block. Make sure we write everything out
4433 * properly so we don't trip over this problem, and then break
4434 * out of the loop if we hit an error.
4435 */
4436 if (rc->stage == MOVE_DATA_EXTENTS && rc->found_file_extent) {
4437 ret = btrfs_wait_ordered_range(rc->data_inode, 0,
4438 (u64)-1);
4439 if (ret)
4440 err = ret;
4441 invalidate_mapping_pages(rc->data_inode->i_mapping,
4442 0, -1);
4443 rc->stage = UPDATE_DATA_PTRS;
4444 }
4445
4446 if (err < 0)
4447 goto out;
4448
4449 if (rc->extents_found == 0)
4450 break;
4451
4452 btrfs_info(fs_info, "found %llu extents, stage: %s",
4453 rc->extents_found, stage_to_string(finishes_stage));
4454 }
4455
4456 WARN_ON(rc->block_group->pinned > 0);
4457 WARN_ON(rc->block_group->reserved > 0);
4458 WARN_ON(rc->block_group->used > 0);
4459 out:
4460 if (err && rw)
4461 btrfs_dec_block_group_ro(rc->block_group);
4462 iput(rc->data_inode);
4463 btrfs_put_block_group(rc->block_group);
4464 kfree(rc);
4465 return err;
4466 }
4467
4468 static noinline_for_stack int mark_garbage_root(struct btrfs_root *root)
4469 {
4470 struct btrfs_fs_info *fs_info = root->fs_info;
4471 struct btrfs_trans_handle *trans;
4472 int ret, err;
4473
4474 trans = btrfs_start_transaction(fs_info->tree_root, 0);
4475 if (IS_ERR(trans))
4476 return PTR_ERR(trans);
4477
4478 memset(&root->root_item.drop_progress, 0,
4479 sizeof(root->root_item.drop_progress));
4480 root->root_item.drop_level = 0;
4481 btrfs_set_root_refs(&root->root_item, 0);
4482 ret = btrfs_update_root(trans, fs_info->tree_root,
4483 &root->root_key, &root->root_item);
4484
4485 err = btrfs_end_transaction(trans);
4486 if (err)
4487 return err;
4488 return ret;
4489 }
4490
4491 /*
4492 * recover relocation interrupted by system crash.
4493 *
4494 * this function resumes merging reloc trees with corresponding fs trees.
4495 * this is important for keeping the sharing of tree blocks
4496 */
4497 int btrfs_recover_relocation(struct btrfs_root *root)
4498 {
4499 struct btrfs_fs_info *fs_info = root->fs_info;
4500 LIST_HEAD(reloc_roots);
4501 struct btrfs_key key;
4502 struct btrfs_root *fs_root;
4503 struct btrfs_root *reloc_root;
4504 struct btrfs_path *path;
4505 struct extent_buffer *leaf;
4506 struct reloc_control *rc = NULL;
4507 struct btrfs_trans_handle *trans;
4508 int ret;
4509 int err = 0;
4510
4511 path = btrfs_alloc_path();
4512 if (!path)
4513 return -ENOMEM;
4514 path->reada = READA_BACK;
4515
4516 key.objectid = BTRFS_TREE_RELOC_OBJECTID;
4517 key.type = BTRFS_ROOT_ITEM_KEY;
4518 key.offset = (u64)-1;
4519
4520 while (1) {
4521 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key,
4522 path, 0, 0);
4523 if (ret < 0) {
4524 err = ret;
4525 goto out;
4526 }
4527 if (ret > 0) {
4528 if (path->slots[0] == 0)
4529 break;
4530 path->slots[0]--;
4531 }
4532 leaf = path->nodes[0];
4533 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4534 btrfs_release_path(path);
4535
4536 if (key.objectid != BTRFS_TREE_RELOC_OBJECTID ||
4537 key.type != BTRFS_ROOT_ITEM_KEY)
4538 break;
4539
4540 reloc_root = btrfs_read_fs_root(root, &key);
4541 if (IS_ERR(reloc_root)) {
4542 err = PTR_ERR(reloc_root);
4543 goto out;
4544 }
4545
4546 list_add(&reloc_root->root_list, &reloc_roots);
4547
4548 if (btrfs_root_refs(&reloc_root->root_item) > 0) {
4549 fs_root = read_fs_root(fs_info,
4550 reloc_root->root_key.offset);
4551 if (IS_ERR(fs_root)) {
4552 ret = PTR_ERR(fs_root);
4553 if (ret != -ENOENT) {
4554 err = ret;
4555 goto out;
4556 }
4557 ret = mark_garbage_root(reloc_root);
4558 if (ret < 0) {
4559 err = ret;
4560 goto out;
4561 }
4562 }
4563 }
4564
4565 if (key.offset == 0)
4566 break;
4567
4568 key.offset--;
4569 }
4570 btrfs_release_path(path);
4571
4572 if (list_empty(&reloc_roots))
4573 goto out;
4574
4575 rc = alloc_reloc_control(fs_info);
4576 if (!rc) {
4577 err = -ENOMEM;
4578 goto out;
4579 }
4580
4581 rc->extent_root = fs_info->extent_root;
4582
4583 set_reloc_control(rc);
4584
4585 trans = btrfs_join_transaction(rc->extent_root);
4586 if (IS_ERR(trans)) {
4587 unset_reloc_control(rc);
4588 err = PTR_ERR(trans);
4589 goto out_free;
4590 }
4591
4592 rc->merge_reloc_tree = 1;
4593
4594 while (!list_empty(&reloc_roots)) {
4595 reloc_root = list_entry(reloc_roots.next,
4596 struct btrfs_root, root_list);
4597 list_del(&reloc_root->root_list);
4598
4599 if (btrfs_root_refs(&reloc_root->root_item) == 0) {
4600 list_add_tail(&reloc_root->root_list,
4601 &rc->reloc_roots);
4602 continue;
4603 }
4604
4605 fs_root = read_fs_root(fs_info, reloc_root->root_key.offset);
4606 if (IS_ERR(fs_root)) {
4607 err = PTR_ERR(fs_root);
4608 list_add_tail(&reloc_root->root_list, &reloc_roots);
4609 goto out_free;
4610 }
4611
4612 err = __add_reloc_root(reloc_root);
4613 BUG_ON(err < 0); /* -ENOMEM or logic error */
4614 fs_root->reloc_root = reloc_root;
4615 }
4616
4617 err = btrfs_commit_transaction(trans);
4618 if (err)
4619 goto out_free;
4620
4621 merge_reloc_roots(rc);
4622
4623 unset_reloc_control(rc);
4624
4625 trans = btrfs_join_transaction(rc->extent_root);
4626 if (IS_ERR(trans)) {
4627 err = PTR_ERR(trans);
4628 goto out_free;
4629 }
4630 err = btrfs_commit_transaction(trans);
4631
4632 ret = clean_dirty_subvols(rc);
4633 if (ret < 0 && !err)
4634 err = ret;
4635 out_free:
4636 kfree(rc);
4637 out:
4638 if (!list_empty(&reloc_roots))
4639 free_reloc_roots(&reloc_roots);
4640
4641 btrfs_free_path(path);
4642
4643 if (err == 0) {
4644 /* cleanup orphan inode in data relocation tree */
4645 fs_root = read_fs_root(fs_info, BTRFS_DATA_RELOC_TREE_OBJECTID);
4646 if (IS_ERR(fs_root))
4647 err = PTR_ERR(fs_root);
4648 else
4649 err = btrfs_orphan_cleanup(fs_root);
4650 }
4651 return err;
4652 }
4653
4654 /*
4655 * helper to add ordered checksum for data relocation.
4656 *
4657 * cloning checksum properly handles the nodatasum extents.
4658 * it also saves CPU time to re-calculate the checksum.
4659 */
4660 int btrfs_reloc_clone_csums(struct inode *inode, u64 file_pos, u64 len)
4661 {
4662 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4663 struct btrfs_ordered_sum *sums;
4664 struct btrfs_ordered_extent *ordered;
4665 int ret;
4666 u64 disk_bytenr;
4667 u64 new_bytenr;
4668 LIST_HEAD(list);
4669
4670 ordered = btrfs_lookup_ordered_extent(inode, file_pos);
4671 BUG_ON(ordered->file_offset != file_pos || ordered->num_bytes != len);
4672
4673 disk_bytenr = file_pos + BTRFS_I(inode)->index_cnt;
4674 ret = btrfs_lookup_csums_range(fs_info->csum_root, disk_bytenr,
4675 disk_bytenr + len - 1, &list, 0);
4676 if (ret)
4677 goto out;
4678
4679 while (!list_empty(&list)) {
4680 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
4681 list_del_init(&sums->list);
4682
4683 /*
4684 * We need to offset the new_bytenr based on where the csum is.
4685 * We need to do this because we will read in entire prealloc
4686 * extents but we may have written to say the middle of the
4687 * prealloc extent, so we need to make sure the csum goes with
4688 * the right disk offset.
4689 *
4690 * We can do this because the data reloc inode refers strictly
4691 * to the on disk bytes, so we don't have to worry about
4692 * disk_len vs real len like with real inodes since it's all
4693 * disk length.
4694 */
4695 new_bytenr = ordered->disk_bytenr + sums->bytenr - disk_bytenr;
4696 sums->bytenr = new_bytenr;
4697
4698 btrfs_add_ordered_sum(ordered, sums);
4699 }
4700 out:
4701 btrfs_put_ordered_extent(ordered);
4702 return ret;
4703 }
4704
4705 int btrfs_reloc_cow_block(struct btrfs_trans_handle *trans,
4706 struct btrfs_root *root, struct extent_buffer *buf,
4707 struct extent_buffer *cow)
4708 {
4709 struct btrfs_fs_info *fs_info = root->fs_info;
4710 struct reloc_control *rc;
4711 struct backref_node *node;
4712 int first_cow = 0;
4713 int level;
4714 int ret = 0;
4715
4716 rc = fs_info->reloc_ctl;
4717 if (!rc)
4718 return 0;
4719
4720 BUG_ON(rc->stage == UPDATE_DATA_PTRS &&
4721 root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID);
4722
4723 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) {
4724 if (buf == root->node)
4725 __update_reloc_root(root, cow->start);
4726 }
4727
4728 level = btrfs_header_level(buf);
4729 if (btrfs_header_generation(buf) <=
4730 btrfs_root_last_snapshot(&root->root_item))
4731 first_cow = 1;
4732
4733 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID &&
4734 rc->create_reloc_tree) {
4735 WARN_ON(!first_cow && level == 0);
4736
4737 node = rc->backref_cache.path[level];
4738 BUG_ON(node->bytenr != buf->start &&
4739 node->new_bytenr != buf->start);
4740
4741 drop_node_buffer(node);
4742 atomic_inc(&cow->refs);
4743 node->eb = cow;
4744 node->new_bytenr = cow->start;
4745
4746 if (!node->pending) {
4747 list_move_tail(&node->list,
4748 &rc->backref_cache.pending[level]);
4749 node->pending = 1;
4750 }
4751
4752 if (first_cow)
4753 __mark_block_processed(rc, node);
4754
4755 if (first_cow && level > 0)
4756 rc->nodes_relocated += buf->len;
4757 }
4758
4759 if (level == 0 && first_cow && rc->stage == UPDATE_DATA_PTRS)
4760 ret = replace_file_extents(trans, rc, root, cow);
4761 return ret;
4762 }
4763
4764 /*
4765 * called before creating snapshot. it calculates metadata reservation
4766 * required for relocating tree blocks in the snapshot
4767 */
4768 void btrfs_reloc_pre_snapshot(struct btrfs_pending_snapshot *pending,
4769 u64 *bytes_to_reserve)
4770 {
4771 struct btrfs_root *root = pending->root;
4772 struct reloc_control *rc = root->fs_info->reloc_ctl;
4773
4774 if (!rc || !have_reloc_root(root))
4775 return;
4776
4777 if (!rc->merge_reloc_tree)
4778 return;
4779
4780 root = root->reloc_root;
4781 BUG_ON(btrfs_root_refs(&root->root_item) == 0);
4782 /*
4783 * relocation is in the stage of merging trees. the space
4784 * used by merging a reloc tree is twice the size of
4785 * relocated tree nodes in the worst case. half for cowing
4786 * the reloc tree, half for cowing the fs tree. the space
4787 * used by cowing the reloc tree will be freed after the
4788 * tree is dropped. if we create snapshot, cowing the fs
4789 * tree may use more space than it frees. so we need
4790 * reserve extra space.
4791 */
4792 *bytes_to_reserve += rc->nodes_relocated;
4793 }
4794
4795 /*
4796 * called after snapshot is created. migrate block reservation
4797 * and create reloc root for the newly created snapshot
4798 */
4799 int btrfs_reloc_post_snapshot(struct btrfs_trans_handle *trans,
4800 struct btrfs_pending_snapshot *pending)
4801 {
4802 struct btrfs_root *root = pending->root;
4803 struct btrfs_root *reloc_root;
4804 struct btrfs_root *new_root;
4805 struct reloc_control *rc = root->fs_info->reloc_ctl;
4806 int ret;
4807
4808 if (!rc || !have_reloc_root(root))
4809 return 0;
4810
4811 rc = root->fs_info->reloc_ctl;
4812 rc->merging_rsv_size += rc->nodes_relocated;
4813
4814 if (rc->merge_reloc_tree) {
4815 ret = btrfs_block_rsv_migrate(&pending->block_rsv,
4816 rc->block_rsv,
4817 rc->nodes_relocated, true);
4818 if (ret)
4819 return ret;
4820 }
4821
4822 new_root = pending->snap;
4823 reloc_root = create_reloc_root(trans, root->reloc_root,
4824 new_root->root_key.objectid);
4825 if (IS_ERR(reloc_root))
4826 return PTR_ERR(reloc_root);
4827
4828 ret = __add_reloc_root(reloc_root);
4829 BUG_ON(ret < 0);
4830 new_root->reloc_root = reloc_root;
4831
4832 if (rc->create_reloc_tree)
4833 ret = clone_backref_node(trans, rc, root, reloc_root);
4834 return ret;
4835 }
Ubuntu Jammy Linux kernel mirror