1 // SPDX-License-Identifier: GPL-2.0
2 /*
3 * Copyright (C) 2009 Oracle. All rights reserved.
4 */
5
6 #include <linux/sched.h>
7 #include <linux/pagemap.h>
8 #include <linux/writeback.h>
9 #include <linux/blkdev.h>
10 #include <linux/rbtree.h>
11 #include <linux/slab.h>
12 #include <linux/error-injection.h>
13 #include "ctree.h"
14 #include "disk-io.h"
15 #include "transaction.h"
16 #include "volumes.h"
17 #include "locking.h"
18 #include "btrfs_inode.h"
19 #include "async-thread.h"
20 #include "free-space-cache.h"
21 #include "qgroup.h"
22 #include "print-tree.h"
23 #include "delalloc-space.h"
24 #include "block-group.h"
25 #include "backref.h"
26 #include "misc.h"
27 #include "subpage.h"
28 #include "zoned.h"
29 #include "inode-item.h"
30 #include "space-info.h"
31 #include "fs.h"
32 #include "accessors.h"
33 #include "extent-tree.h"
34 #include "root-tree.h"
35 #include "file-item.h"
36 #include "relocation.h"
37 #include "super.h"
38 #include "tree-checker.h"
39
40 /*
41 * Relocation overview
42 *
43 * [What does relocation do]
44 *
45 * The objective of relocation is to relocate all extents of the target block
46 * group to other block groups.
47 * This is utilized by resize (shrink only), profile converting, compacting
48 * space, or balance routine to spread chunks over devices.
49 *
50 * Before | After
51 * ------------------------------------------------------------------
52 * BG A: 10 data extents | BG A: deleted
53 * BG B: 2 data extents | BG B: 10 data extents (2 old + 8 relocated)
54 * BG C: 1 extents | BG C: 3 data extents (1 old + 2 relocated)
55 *
56 * [How does relocation work]
57 *
58 * 1. Mark the target block group read-only
59 * New extents won't be allocated from the target block group.
60 *
61 * 2.1 Record each extent in the target block group
62 * To build a proper map of extents to be relocated.
63 *
64 * 2.2 Build data reloc tree and reloc trees
65 * Data reloc tree will contain an inode, recording all newly relocated
66 * data extents.
67 * There will be only one data reloc tree for one data block group.
68 *
69 * Reloc tree will be a special snapshot of its source tree, containing
70 * relocated tree blocks.
71 * Each tree referring to a tree block in target block group will get its
72 * reloc tree built.
73 *
74 * 2.3 Swap source tree with its corresponding reloc tree
75 * Each involved tree only refers to new extents after swap.
76 *
77 * 3. Cleanup reloc trees and data reloc tree.
78 * As old extents in the target block group are still referenced by reloc
79 * trees, we need to clean them up before really freeing the target block
80 * group.
81 *
82 * The main complexity is in steps 2.2 and 2.3.
83 *
84 * The entry point of relocation is relocate_block_group() function.
85 */
86
87 #define RELOCATION_RESERVED_NODES 256
88 /*
89 * map address of tree root to tree
90 */
91 struct mapping_node {
92 struct {
93 struct rb_node rb_node;
94 u64 bytenr;
95 }; /* Use rb_simle_node for search/insert */
96 void *data;
97 };
98
99 struct mapping_tree {
100 struct rb_root rb_root;
101 spinlock_t lock;
102 };
103
104 /*
105 * present a tree block to process
106 */
107 struct tree_block {
108 struct {
109 struct rb_node rb_node;
110 u64 bytenr;
111 }; /* Use rb_simple_node for search/insert */
112 u64 owner;
113 struct btrfs_key key;
114 unsigned int level:8;
115 unsigned int key_ready:1;
116 };
117
118 #define MAX_EXTENTS 128
119
120 struct file_extent_cluster {
121 u64 start;
122 u64 end;
123 u64 boundary[MAX_EXTENTS];
124 unsigned int nr;
125 };
126
127 struct reloc_control {
128 /* block group to relocate */
129 struct btrfs_block_group *block_group;
130 /* extent tree */
131 struct btrfs_root *extent_root;
132 /* inode for moving data */
133 struct inode *data_inode;
134
135 struct btrfs_block_rsv *block_rsv;
136
137 struct btrfs_backref_cache backref_cache;
138
139 struct file_extent_cluster cluster;
140 /* tree blocks have been processed */
141 struct extent_io_tree processed_blocks;
142 /* map start of tree root to corresponding reloc tree */
143 struct mapping_tree reloc_root_tree;
144 /* list of reloc trees */
145 struct list_head reloc_roots;
146 /* list of subvolume trees that get relocated */
147 struct list_head dirty_subvol_roots;
148 /* size of metadata reservation for merging reloc trees */
149 u64 merging_rsv_size;
150 /* size of relocated tree nodes */
151 u64 nodes_relocated;
152 /* reserved size for block group relocation*/
153 u64 reserved_bytes;
154
155 u64 search_start;
156 u64 extents_found;
157
158 unsigned int stage:8;
159 unsigned int create_reloc_tree:1;
160 unsigned int merge_reloc_tree:1;
161 unsigned int found_file_extent:1;
162 };
163
164 /* stages of data relocation */
165 #define MOVE_DATA_EXTENTS 0
166 #define UPDATE_DATA_PTRS 1
167
mark_block_processed(struct reloc_control * rc,struct btrfs_backref_node * node)168 static void mark_block_processed(struct reloc_control *rc,
169 struct btrfs_backref_node *node)
170 {
171 u32 blocksize;
172
173 if (node->level == 0 ||
174 in_range(node->bytenr, rc->block_group->start,
175 rc->block_group->length)) {
176 blocksize = rc->extent_root->fs_info->nodesize;
177 set_extent_bit(&rc->processed_blocks, node->bytenr,
178 node->bytenr + blocksize - 1, EXTENT_DIRTY, NULL);
179 }
180 node->processed = 1;
181 }
182
183
mapping_tree_init(struct mapping_tree * tree)184 static void mapping_tree_init(struct mapping_tree *tree)
185 {
186 tree->rb_root = RB_ROOT;
187 spin_lock_init(&tree->lock);
188 }
189
190 /*
191 * walk up backref nodes until reach node presents tree root
192 */
walk_up_backref(struct btrfs_backref_node * node,struct btrfs_backref_edge * edges[],int * index)193 static struct btrfs_backref_node *walk_up_backref(
194 struct btrfs_backref_node *node,
195 struct btrfs_backref_edge *edges[], int *index)
196 {
197 struct btrfs_backref_edge *edge;
198 int idx = *index;
199
200 while (!list_empty(&node->upper)) {
201 edge = list_entry(node->upper.next,
202 struct btrfs_backref_edge, list[LOWER]);
203 edges[idx++] = edge;
204 node = edge->node[UPPER];
205 }
206 BUG_ON(node->detached);
207 *index = idx;
208 return node;
209 }
210
211 /*
212 * walk down backref nodes to find start of next reference path
213 */
walk_down_backref(struct btrfs_backref_edge * edges[],int * index)214 static struct btrfs_backref_node *walk_down_backref(
215 struct btrfs_backref_edge *edges[], int *index)
216 {
217 struct btrfs_backref_edge *edge;
218 struct btrfs_backref_node *lower;
219 int idx = *index;
220
221 while (idx > 0) {
222 edge = edges[idx - 1];
223 lower = edge->node[LOWER];
224 if (list_is_last(&edge->list[LOWER], &lower->upper)) {
225 idx--;
226 continue;
227 }
228 edge = list_entry(edge->list[LOWER].next,
229 struct btrfs_backref_edge, list[LOWER]);
230 edges[idx - 1] = edge;
231 *index = idx;
232 return edge->node[UPPER];
233 }
234 *index = 0;
235 return NULL;
236 }
237
reloc_root_is_dead(const struct btrfs_root * root)238 static bool reloc_root_is_dead(const struct btrfs_root *root)
239 {
240 /*
241 * Pair with set_bit/clear_bit in clean_dirty_subvols and
242 * btrfs_update_reloc_root. We need to see the updated bit before
243 * trying to access reloc_root
244 */
245 smp_rmb();
246 if (test_bit(BTRFS_ROOT_DEAD_RELOC_TREE, &root->state))
247 return true;
248 return false;
249 }
250
251 /*
252 * Check if this subvolume tree has valid reloc tree.
253 *
254 * Reloc tree after swap is considered dead, thus not considered as valid.
255 * This is enough for most callers, as they don't distinguish dead reloc root
256 * from no reloc root. But btrfs_should_ignore_reloc_root() below is a
257 * special case.
258 */
have_reloc_root(const struct btrfs_root * root)259 static bool have_reloc_root(const struct btrfs_root *root)
260 {
261 if (reloc_root_is_dead(root))
262 return false;
263 if (!root->reloc_root)
264 return false;
265 return true;
266 }
267
btrfs_should_ignore_reloc_root(const struct btrfs_root * root)268 bool btrfs_should_ignore_reloc_root(const struct btrfs_root *root)
269 {
270 struct btrfs_root *reloc_root;
271
272 if (!test_bit(BTRFS_ROOT_SHAREABLE, &root->state))
273 return false;
274
275 /* This root has been merged with its reloc tree, we can ignore it */
276 if (reloc_root_is_dead(root))
277 return true;
278
279 reloc_root = root->reloc_root;
280 if (!reloc_root)
281 return false;
282
283 if (btrfs_header_generation(reloc_root->commit_root) ==
284 root->fs_info->running_transaction->transid)
285 return false;
286 /*
287 * If there is reloc tree and it was created in previous transaction
288 * backref lookup can find the reloc tree, so backref node for the fs
289 * tree root is useless for relocation.
290 */
291 return true;
292 }
293
294 /*
295 * find reloc tree by address of tree root
296 */
find_reloc_root(struct btrfs_fs_info * fs_info,u64 bytenr)297 struct btrfs_root *find_reloc_root(struct btrfs_fs_info *fs_info, u64 bytenr)
298 {
299 struct reloc_control *rc = fs_info->reloc_ctl;
300 struct rb_node *rb_node;
301 struct mapping_node *node;
302 struct btrfs_root *root = NULL;
303
304 ASSERT(rc);
305 spin_lock(&rc->reloc_root_tree.lock);
306 rb_node = rb_simple_search(&rc->reloc_root_tree.rb_root, bytenr);
307 if (rb_node) {
308 node = rb_entry(rb_node, struct mapping_node, rb_node);
309 root = node->data;
310 }
311 spin_unlock(&rc->reloc_root_tree.lock);
312 return btrfs_grab_root(root);
313 }
314
315 /*
316 * For useless nodes, do two major clean ups:
317 *
318 * - Cleanup the children edges and nodes
319 * If child node is also orphan (no parent) during cleanup, then the child
320 * node will also be cleaned up.
321 *
322 * - Freeing up leaves (level 0), keeps nodes detached
323 * For nodes, the node is still cached as "detached"
324 *
325 * Return false if @node is not in the @useless_nodes list.
326 * Return true if @node is in the @useless_nodes list.
327 */
handle_useless_nodes(struct reloc_control * rc,struct btrfs_backref_node * node)328 static bool handle_useless_nodes(struct reloc_control *rc,
329 struct btrfs_backref_node *node)
330 {
331 struct btrfs_backref_cache *cache = &rc->backref_cache;
332 struct list_head *useless_node = &cache->useless_node;
333 bool ret = false;
334
335 while (!list_empty(useless_node)) {
336 struct btrfs_backref_node *cur;
337
338 cur = list_first_entry(useless_node, struct btrfs_backref_node,
339 list);
340 list_del_init(&cur->list);
341
342 /* Only tree root nodes can be added to @useless_nodes */
343 ASSERT(list_empty(&cur->upper));
344
345 if (cur == node)
346 ret = true;
347
348 /* The node is the lowest node */
349 if (cur->lowest) {
350 list_del_init(&cur->lower);
351 cur->lowest = 0;
352 }
353
354 /* Cleanup the lower edges */
355 while (!list_empty(&cur->lower)) {
356 struct btrfs_backref_edge *edge;
357 struct btrfs_backref_node *lower;
358
359 edge = list_entry(cur->lower.next,
360 struct btrfs_backref_edge, list[UPPER]);
361 list_del(&edge->list[UPPER]);
362 list_del(&edge->list[LOWER]);
363 lower = edge->node[LOWER];
364 btrfs_backref_free_edge(cache, edge);
365
366 /* Child node is also orphan, queue for cleanup */
367 if (list_empty(&lower->upper))
368 list_add(&lower->list, useless_node);
369 }
370 /* Mark this block processed for relocation */
371 mark_block_processed(rc, cur);
372
373 /*
374 * Backref nodes for tree leaves are deleted from the cache.
375 * Backref nodes for upper level tree blocks are left in the
376 * cache to avoid unnecessary backref lookup.
377 */
378 if (cur->level > 0) {
379 list_add(&cur->list, &cache->detached);
380 cur->detached = 1;
381 } else {
382 rb_erase(&cur->rb_node, &cache->rb_root);
383 btrfs_backref_free_node(cache, cur);
384 }
385 }
386 return ret;
387 }
388
389 /*
390 * Build backref tree for a given tree block. Root of the backref tree
391 * corresponds the tree block, leaves of the backref tree correspond roots of
392 * b-trees that reference the tree block.
393 *
394 * The basic idea of this function is check backrefs of a given block to find
395 * upper level blocks that reference the block, and then check backrefs of
396 * these upper level blocks recursively. The recursion stops when tree root is
397 * reached or backrefs for the block is cached.
398 *
399 * NOTE: if we find that backrefs for a block are cached, we know backrefs for
400 * all upper level blocks that directly/indirectly reference the block are also
401 * cached.
402 */
build_backref_tree(struct btrfs_trans_handle * trans,struct reloc_control * rc,struct btrfs_key * node_key,int level,u64 bytenr)403 static noinline_for_stack struct btrfs_backref_node *build_backref_tree(
404 struct btrfs_trans_handle *trans,
405 struct reloc_control *rc, struct btrfs_key *node_key,
406 int level, u64 bytenr)
407 {
408 struct btrfs_backref_iter *iter;
409 struct btrfs_backref_cache *cache = &rc->backref_cache;
410 /* For searching parent of TREE_BLOCK_REF */
411 struct btrfs_path *path;
412 struct btrfs_backref_node *cur;
413 struct btrfs_backref_node *node = NULL;
414 struct btrfs_backref_edge *edge;
415 int ret;
416 int err = 0;
417
418 iter = btrfs_backref_iter_alloc(rc->extent_root->fs_info);
419 if (!iter)
420 return ERR_PTR(-ENOMEM);
421 path = btrfs_alloc_path();
422 if (!path) {
423 err = -ENOMEM;
424 goto out;
425 }
426
427 node = btrfs_backref_alloc_node(cache, bytenr, level);
428 if (!node) {
429 err = -ENOMEM;
430 goto out;
431 }
432
433 node->lowest = 1;
434 cur = node;
435
436 /* Breadth-first search to build backref cache */
437 do {
438 ret = btrfs_backref_add_tree_node(trans, cache, path, iter,
439 node_key, cur);
440 if (ret < 0) {
441 err = ret;
442 goto out;
443 }
444 edge = list_first_entry_or_null(&cache->pending_edge,
445 struct btrfs_backref_edge, list[UPPER]);
446 /*
447 * The pending list isn't empty, take the first block to
448 * process
449 */
450 if (edge) {
451 list_del_init(&edge->list[UPPER]);
452 cur = edge->node[UPPER];
453 }
454 } while (edge);
455
456 /* Finish the upper linkage of newly added edges/nodes */
457 ret = btrfs_backref_finish_upper_links(cache, node);
458 if (ret < 0) {
459 err = ret;
460 goto out;
461 }
462
463 if (handle_useless_nodes(rc, node))
464 node = NULL;
465 out:
466 btrfs_backref_iter_free(iter);
467 btrfs_free_path(path);
468 if (err) {
469 btrfs_backref_error_cleanup(cache, node);
470 return ERR_PTR(err);
471 }
472 ASSERT(!node || !node->detached);
473 ASSERT(list_empty(&cache->useless_node) &&
474 list_empty(&cache->pending_edge));
475 return node;
476 }
477
478 /*
479 * helper to add backref node for the newly created snapshot.
480 * the backref node is created by cloning backref node that
481 * corresponds to root of source tree
482 */
clone_backref_node(struct btrfs_trans_handle * trans,struct reloc_control * rc,const struct btrfs_root * src,struct btrfs_root * dest)483 static int clone_backref_node(struct btrfs_trans_handle *trans,
484 struct reloc_control *rc,
485 const struct btrfs_root *src,
486 struct btrfs_root *dest)
487 {
488 struct btrfs_root *reloc_root = src->reloc_root;
489 struct btrfs_backref_cache *cache = &rc->backref_cache;
490 struct btrfs_backref_node *node = NULL;
491 struct btrfs_backref_node *new_node;
492 struct btrfs_backref_edge *edge;
493 struct btrfs_backref_edge *new_edge;
494 struct rb_node *rb_node;
495
496 rb_node = rb_simple_search(&cache->rb_root, src->commit_root->start);
497 if (rb_node) {
498 node = rb_entry(rb_node, struct btrfs_backref_node, rb_node);
499 if (node->detached)
500 node = NULL;
501 else
502 BUG_ON(node->new_bytenr != reloc_root->node->start);
503 }
504
505 if (!node) {
506 rb_node = rb_simple_search(&cache->rb_root,
507 reloc_root->commit_root->start);
508 if (rb_node) {
509 node = rb_entry(rb_node, struct btrfs_backref_node,
510 rb_node);
511 BUG_ON(node->detached);
512 }
513 }
514
515 if (!node)
516 return 0;
517
518 new_node = btrfs_backref_alloc_node(cache, dest->node->start,
519 node->level);
520 if (!new_node)
521 return -ENOMEM;
522
523 new_node->lowest = node->lowest;
524 new_node->checked = 1;
525 new_node->root = btrfs_grab_root(dest);
526 ASSERT(new_node->root);
527
528 if (!node->lowest) {
529 list_for_each_entry(edge, &node->lower, list[UPPER]) {
530 new_edge = btrfs_backref_alloc_edge(cache);
531 if (!new_edge)
532 goto fail;
533
534 btrfs_backref_link_edge(new_edge, edge->node[LOWER],
535 new_node, LINK_UPPER);
536 }
537 } else {
538 list_add_tail(&new_node->lower, &cache->leaves);
539 }
540
541 rb_node = rb_simple_insert(&cache->rb_root, new_node->bytenr,
542 &new_node->rb_node);
543 if (rb_node)
544 btrfs_backref_panic(trans->fs_info, new_node->bytenr, -EEXIST);
545
546 if (!new_node->lowest) {
547 list_for_each_entry(new_edge, &new_node->lower, list[UPPER]) {
548 list_add_tail(&new_edge->list[LOWER],
549 &new_edge->node[LOWER]->upper);
550 }
551 }
552 return 0;
553 fail:
554 while (!list_empty(&new_node->lower)) {
555 new_edge = list_entry(new_node->lower.next,
556 struct btrfs_backref_edge, list[UPPER]);
557 list_del(&new_edge->list[UPPER]);
558 btrfs_backref_free_edge(cache, new_edge);
559 }
560 btrfs_backref_free_node(cache, new_node);
561 return -ENOMEM;
562 }
563
564 /*
565 * helper to add 'address of tree root -> reloc tree' mapping
566 */
__add_reloc_root(struct btrfs_root * root)567 static int __must_check __add_reloc_root(struct btrfs_root *root)
568 {
569 struct btrfs_fs_info *fs_info = root->fs_info;
570 struct rb_node *rb_node;
571 struct mapping_node *node;
572 struct reloc_control *rc = fs_info->reloc_ctl;
573
574 node = kmalloc(sizeof(*node), GFP_NOFS);
575 if (!node)
576 return -ENOMEM;
577
578 node->bytenr = root->commit_root->start;
579 node->data = root;
580
581 spin_lock(&rc->reloc_root_tree.lock);
582 rb_node = rb_simple_insert(&rc->reloc_root_tree.rb_root,
583 node->bytenr, &node->rb_node);
584 spin_unlock(&rc->reloc_root_tree.lock);
585 if (rb_node) {
586 btrfs_err(fs_info,
587 "Duplicate root found for start=%llu while inserting into relocation tree",
588 node->bytenr);
589 return -EEXIST;
590 }
591
592 list_add_tail(&root->root_list, &rc->reloc_roots);
593 return 0;
594 }
595
596 /*
597 * helper to delete the 'address of tree root -> reloc tree'
598 * mapping
599 */
__del_reloc_root(struct btrfs_root * root)600 static void __del_reloc_root(struct btrfs_root *root)
601 {
602 struct btrfs_fs_info *fs_info = root->fs_info;
603 struct rb_node *rb_node;
604 struct mapping_node *node = NULL;
605 struct reloc_control *rc = fs_info->reloc_ctl;
606 bool put_ref = false;
607
608 if (rc && root->node) {
609 spin_lock(&rc->reloc_root_tree.lock);
610 rb_node = rb_simple_search(&rc->reloc_root_tree.rb_root,
611 root->commit_root->start);
612 if (rb_node) {
613 node = rb_entry(rb_node, struct mapping_node, rb_node);
614 rb_erase(&node->rb_node, &rc->reloc_root_tree.rb_root);
615 RB_CLEAR_NODE(&node->rb_node);
616 }
617 spin_unlock(&rc->reloc_root_tree.lock);
618 ASSERT(!node || (struct btrfs_root *)node->data == root);
619 }
620
621 /*
622 * We only put the reloc root here if it's on the list. There's a lot
623 * of places where the pattern is to splice the rc->reloc_roots, process
624 * the reloc roots, and then add the reloc root back onto
625 * rc->reloc_roots. If we call __del_reloc_root while it's off of the
626 * list we don't want the reference being dropped, because the guy
627 * messing with the list is in charge of the reference.
628 */
629 spin_lock(&fs_info->trans_lock);
630 if (!list_empty(&root->root_list)) {
631 put_ref = true;
632 list_del_init(&root->root_list);
633 }
634 spin_unlock(&fs_info->trans_lock);
635 if (put_ref)
636 btrfs_put_root(root);
637 kfree(node);
638 }
639
640 /*
641 * helper to update the 'address of tree root -> reloc tree'
642 * mapping
643 */
__update_reloc_root(struct btrfs_root * root)644 static int __update_reloc_root(struct btrfs_root *root)
645 {
646 struct btrfs_fs_info *fs_info = root->fs_info;
647 struct rb_node *rb_node;
648 struct mapping_node *node = NULL;
649 struct reloc_control *rc = fs_info->reloc_ctl;
650
651 spin_lock(&rc->reloc_root_tree.lock);
652 rb_node = rb_simple_search(&rc->reloc_root_tree.rb_root,
653 root->commit_root->start);
654 if (rb_node) {
655 node = rb_entry(rb_node, struct mapping_node, rb_node);
656 rb_erase(&node->rb_node, &rc->reloc_root_tree.rb_root);
657 }
658 spin_unlock(&rc->reloc_root_tree.lock);
659
660 if (!node)
661 return 0;
662 BUG_ON((struct btrfs_root *)node->data != root);
663
664 spin_lock(&rc->reloc_root_tree.lock);
665 node->bytenr = root->node->start;
666 rb_node = rb_simple_insert(&rc->reloc_root_tree.rb_root,
667 node->bytenr, &node->rb_node);
668 spin_unlock(&rc->reloc_root_tree.lock);
669 if (rb_node)
670 btrfs_backref_panic(fs_info, node->bytenr, -EEXIST);
671 return 0;
672 }
673
create_reloc_root(struct btrfs_trans_handle * trans,struct btrfs_root * root,u64 objectid)674 static struct btrfs_root *create_reloc_root(struct btrfs_trans_handle *trans,
675 struct btrfs_root *root, u64 objectid)
676 {
677 struct btrfs_fs_info *fs_info = root->fs_info;
678 struct btrfs_root *reloc_root;
679 struct extent_buffer *eb;
680 struct btrfs_root_item *root_item;
681 struct btrfs_key root_key;
682 int ret = 0;
683 bool must_abort = false;
684
685 root_item = kmalloc(sizeof(*root_item), GFP_NOFS);
686 if (!root_item)
687 return ERR_PTR(-ENOMEM);
688
689 root_key.objectid = BTRFS_TREE_RELOC_OBJECTID;
690 root_key.type = BTRFS_ROOT_ITEM_KEY;
691 root_key.offset = objectid;
692
693 if (root->root_key.objectid == objectid) {
694 u64 commit_root_gen;
695
696 /* called by btrfs_init_reloc_root */
697 ret = btrfs_copy_root(trans, root, root->commit_root, &eb,
698 BTRFS_TREE_RELOC_OBJECTID);
699 if (ret)
700 goto fail;
701
702 /*
703 * Set the last_snapshot field to the generation of the commit
704 * root - like this ctree.c:btrfs_block_can_be_shared() behaves
705 * correctly (returns true) when the relocation root is created
706 * either inside the critical section of a transaction commit
707 * (through transaction.c:qgroup_account_snapshot()) and when
708 * it's created before the transaction commit is started.
709 */
710 commit_root_gen = btrfs_header_generation(root->commit_root);
711 btrfs_set_root_last_snapshot(&root->root_item, commit_root_gen);
712 } else {
713 /*
714 * called by btrfs_reloc_post_snapshot_hook.
715 * the source tree is a reloc tree, all tree blocks
716 * modified after it was created have RELOC flag
717 * set in their headers. so it's OK to not update
718 * the 'last_snapshot'.
719 */
720 ret = btrfs_copy_root(trans, root, root->node, &eb,
721 BTRFS_TREE_RELOC_OBJECTID);
722 if (ret)
723 goto fail;
724 }
725
726 /*
727 * We have changed references at this point, we must abort the
728 * transaction if anything fails.
729 */
730 must_abort = true;
731
732 memcpy(root_item, &root->root_item, sizeof(*root_item));
733 btrfs_set_root_bytenr(root_item, eb->start);
734 btrfs_set_root_level(root_item, btrfs_header_level(eb));
735 btrfs_set_root_generation(root_item, trans->transid);
736
737 if (root->root_key.objectid == objectid) {
738 btrfs_set_root_refs(root_item, 0);
739 memset(&root_item->drop_progress, 0,
740 sizeof(struct btrfs_disk_key));
741 btrfs_set_root_drop_level(root_item, 0);
742 }
743
744 btrfs_tree_unlock(eb);
745 free_extent_buffer(eb);
746
747 ret = btrfs_insert_root(trans, fs_info->tree_root,
748 &root_key, root_item);
749 if (ret)
750 goto fail;
751
752 kfree(root_item);
753
754 reloc_root = btrfs_read_tree_root(fs_info->tree_root, &root_key);
755 if (IS_ERR(reloc_root)) {
756 ret = PTR_ERR(reloc_root);
757 goto abort;
758 }
759 set_bit(BTRFS_ROOT_SHAREABLE, &reloc_root->state);
760 reloc_root->last_trans = trans->transid;
761 return reloc_root;
762 fail:
763 kfree(root_item);
764 abort:
765 if (must_abort)
766 btrfs_abort_transaction(trans, ret);
767 return ERR_PTR(ret);
768 }
769
770 /*
771 * create reloc tree for a given fs tree. reloc tree is just a
772 * snapshot of the fs tree with special root objectid.
773 *
774 * The reloc_root comes out of here with two references, one for
775 * root->reloc_root, and another for being on the rc->reloc_roots list.
776 */
btrfs_init_reloc_root(struct btrfs_trans_handle * trans,struct btrfs_root * root)777 int btrfs_init_reloc_root(struct btrfs_trans_handle *trans,
778 struct btrfs_root *root)
779 {
780 struct btrfs_fs_info *fs_info = root->fs_info;
781 struct btrfs_root *reloc_root;
782 struct reloc_control *rc = fs_info->reloc_ctl;
783 struct btrfs_block_rsv *rsv;
784 int clear_rsv = 0;
785 int ret;
786
787 if (!rc)
788 return 0;
789
790 /*
791 * The subvolume has reloc tree but the swap is finished, no need to
792 * create/update the dead reloc tree
793 */
794 if (reloc_root_is_dead(root))
795 return 0;
796
797 /*
798 * This is subtle but important. We do not do
799 * record_root_in_transaction for reloc roots, instead we record their
800 * corresponding fs root, and then here we update the last trans for the
801 * reloc root. This means that we have to do this for the entire life
802 * of the reloc root, regardless of which stage of the relocation we are
803 * in.
804 */
805 if (root->reloc_root) {
806 reloc_root = root->reloc_root;
807 reloc_root->last_trans = trans->transid;
808 return 0;
809 }
810
811 /*
812 * We are merging reloc roots, we do not need new reloc trees. Also
813 * reloc trees never need their own reloc tree.
814 */
815 if (!rc->create_reloc_tree ||
816 root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID)
817 return 0;
818
819 if (!trans->reloc_reserved) {
820 rsv = trans->block_rsv;
821 trans->block_rsv = rc->block_rsv;
822 clear_rsv = 1;
823 }
824 reloc_root = create_reloc_root(trans, root, root->root_key.objectid);
825 if (clear_rsv)
826 trans->block_rsv = rsv;
827 if (IS_ERR(reloc_root))
828 return PTR_ERR(reloc_root);
829
830 ret = __add_reloc_root(reloc_root);
831 ASSERT(ret != -EEXIST);
832 if (ret) {
833 /* Pairs with create_reloc_root */
834 btrfs_put_root(reloc_root);
835 return ret;
836 }
837 root->reloc_root = btrfs_grab_root(reloc_root);
838 return 0;
839 }
840
841 /*
842 * update root item of reloc tree
843 */
btrfs_update_reloc_root(struct btrfs_trans_handle * trans,struct btrfs_root * root)844 int btrfs_update_reloc_root(struct btrfs_trans_handle *trans,
845 struct btrfs_root *root)
846 {
847 struct btrfs_fs_info *fs_info = root->fs_info;
848 struct btrfs_root *reloc_root;
849 struct btrfs_root_item *root_item;
850 int ret;
851
852 if (!have_reloc_root(root))
853 return 0;
854
855 reloc_root = root->reloc_root;
856 root_item = &reloc_root->root_item;
857
858 /*
859 * We are probably ok here, but __del_reloc_root() will drop its ref of
860 * the root. We have the ref for root->reloc_root, but just in case
861 * hold it while we update the reloc root.
862 */
863 btrfs_grab_root(reloc_root);
864
865 /* root->reloc_root will stay until current relocation finished */
866 if (fs_info->reloc_ctl && fs_info->reloc_ctl->merge_reloc_tree &&
867 btrfs_root_refs(root_item) == 0) {
868 set_bit(BTRFS_ROOT_DEAD_RELOC_TREE, &root->state);
869 /*
870 * Mark the tree as dead before we change reloc_root so
871 * have_reloc_root will not touch it from now on.
872 */
873 smp_wmb();
874 __del_reloc_root(reloc_root);
875 }
876
877 if (reloc_root->commit_root != reloc_root->node) {
878 __update_reloc_root(reloc_root);
879 btrfs_set_root_node(root_item, reloc_root->node);
880 free_extent_buffer(reloc_root->commit_root);
881 reloc_root->commit_root = btrfs_root_node(reloc_root);
882 }
883
884 ret = btrfs_update_root(trans, fs_info->tree_root,
885 &reloc_root->root_key, root_item);
886 btrfs_put_root(reloc_root);
887 return ret;
888 }
889
890 /*
891 * helper to find first cached inode with inode number >= objectid
892 * in a subvolume
893 */
find_next_inode(struct btrfs_root * root,u64 objectid)894 static struct inode *find_next_inode(struct btrfs_root *root, u64 objectid)
895 {
896 struct rb_node *node;
897 struct rb_node *prev;
898 struct btrfs_inode *entry;
899 struct inode *inode;
900
901 spin_lock(&root->inode_lock);
902 again:
903 node = root->inode_tree.rb_node;
904 prev = NULL;
905 while (node) {
906 prev = node;
907 entry = rb_entry(node, struct btrfs_inode, rb_node);
908
909 if (objectid < btrfs_ino(entry))
910 node = node->rb_left;
911 else if (objectid > btrfs_ino(entry))
912 node = node->rb_right;
913 else
914 break;
915 }
916 if (!node) {
917 while (prev) {
918 entry = rb_entry(prev, struct btrfs_inode, rb_node);
919 if (objectid <= btrfs_ino(entry)) {
920 node = prev;
921 break;
922 }
923 prev = rb_next(prev);
924 }
925 }
926 while (node) {
927 entry = rb_entry(node, struct btrfs_inode, rb_node);
928 inode = igrab(&entry->vfs_inode);
929 if (inode) {
930 spin_unlock(&root->inode_lock);
931 return inode;
932 }
933
934 objectid = btrfs_ino(entry) + 1;
935 if (cond_resched_lock(&root->inode_lock))
936 goto again;
937
938 node = rb_next(node);
939 }
940 spin_unlock(&root->inode_lock);
941 return NULL;
942 }
943
944 /*
945 * get new location of data
946 */
get_new_location(struct inode * reloc_inode,u64 * new_bytenr,u64 bytenr,u64 num_bytes)947 static int get_new_location(struct inode *reloc_inode, u64 *new_bytenr,
948 u64 bytenr, u64 num_bytes)
949 {
950 struct btrfs_root *root = BTRFS_I(reloc_inode)->root;
951 struct btrfs_path *path;
952 struct btrfs_file_extent_item *fi;
953 struct extent_buffer *leaf;
954 int ret;
955
956 path = btrfs_alloc_path();
957 if (!path)
958 return -ENOMEM;
959
960 bytenr -= BTRFS_I(reloc_inode)->index_cnt;
961 ret = btrfs_lookup_file_extent(NULL, root, path,
962 btrfs_ino(BTRFS_I(reloc_inode)), bytenr, 0);
963 if (ret < 0)
964 goto out;
965 if (ret > 0) {
966 ret = -ENOENT;
967 goto out;
968 }
969
970 leaf = path->nodes[0];
971 fi = btrfs_item_ptr(leaf, path->slots[0],
972 struct btrfs_file_extent_item);
973
974 BUG_ON(btrfs_file_extent_offset(leaf, fi) ||
975 btrfs_file_extent_compression(leaf, fi) ||
976 btrfs_file_extent_encryption(leaf, fi) ||
977 btrfs_file_extent_other_encoding(leaf, fi));
978
979 if (num_bytes != btrfs_file_extent_disk_num_bytes(leaf, fi)) {
980 ret = -EINVAL;
981 goto out;
982 }
983
984 *new_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
985 ret = 0;
986 out:
987 btrfs_free_path(path);
988 return ret;
989 }
990
991 /*
992 * update file extent items in the tree leaf to point to
993 * the new locations.
994 */
995 static noinline_for_stack
replace_file_extents(struct btrfs_trans_handle * trans,struct reloc_control * rc,struct btrfs_root * root,struct extent_buffer * leaf)996 int replace_file_extents(struct btrfs_trans_handle *trans,
997 struct reloc_control *rc,
998 struct btrfs_root *root,
999 struct extent_buffer *leaf)
1000 {
1001 struct btrfs_fs_info *fs_info = root->fs_info;
1002 struct btrfs_key key;
1003 struct btrfs_file_extent_item *fi;
1004 struct inode *inode = NULL;
1005 u64 parent;
1006 u64 bytenr;
1007 u64 new_bytenr = 0;
1008 u64 num_bytes;
1009 u64 end;
1010 u32 nritems;
1011 u32 i;
1012 int ret = 0;
1013 int first = 1;
1014 int dirty = 0;
1015
1016 if (rc->stage != UPDATE_DATA_PTRS)
1017 return 0;
1018
1019 /* reloc trees always use full backref */
1020 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID)
1021 parent = leaf->start;
1022 else
1023 parent = 0;
1024
1025 nritems = btrfs_header_nritems(leaf);
1026 for (i = 0; i < nritems; i++) {
1027 struct btrfs_ref ref = { 0 };
1028
1029 cond_resched();
1030 btrfs_item_key_to_cpu(leaf, &key, i);
1031 if (key.type != BTRFS_EXTENT_DATA_KEY)
1032 continue;
1033 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
1034 if (btrfs_file_extent_type(leaf, fi) ==
1035 BTRFS_FILE_EXTENT_INLINE)
1036 continue;
1037 bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1038 num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
1039 if (bytenr == 0)
1040 continue;
1041 if (!in_range(bytenr, rc->block_group->start,
1042 rc->block_group->length))
1043 continue;
1044
1045 /*
1046 * if we are modifying block in fs tree, wait for read_folio
1047 * to complete and drop the extent cache
1048 */
1049 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
1050 if (first) {
1051 inode = find_next_inode(root, key.objectid);
1052 first = 0;
1053 } else if (inode && btrfs_ino(BTRFS_I(inode)) < key.objectid) {
1054 btrfs_add_delayed_iput(BTRFS_I(inode));
1055 inode = find_next_inode(root, key.objectid);
1056 }
1057 if (inode && btrfs_ino(BTRFS_I(inode)) == key.objectid) {
1058 struct extent_state *cached_state = NULL;
1059
1060 end = key.offset +
1061 btrfs_file_extent_num_bytes(leaf, fi);
1062 WARN_ON(!IS_ALIGNED(key.offset,
1063 fs_info->sectorsize));
1064 WARN_ON(!IS_ALIGNED(end, fs_info->sectorsize));
1065 end--;
1066 ret = try_lock_extent(&BTRFS_I(inode)->io_tree,
1067 key.offset, end,
1068 &cached_state);
1069 if (!ret)
1070 continue;
1071
1072 btrfs_drop_extent_map_range(BTRFS_I(inode),
1073 key.offset, end, true);
1074 unlock_extent(&BTRFS_I(inode)->io_tree,
1075 key.offset, end, &cached_state);
1076 }
1077 }
1078
1079 ret = get_new_location(rc->data_inode, &new_bytenr,
1080 bytenr, num_bytes);
1081 if (ret) {
1082 /*
1083 * Don't have to abort since we've not changed anything
1084 * in the file extent yet.
1085 */
1086 break;
1087 }
1088
1089 btrfs_set_file_extent_disk_bytenr(leaf, fi, new_bytenr);
1090 dirty = 1;
1091
1092 key.offset -= btrfs_file_extent_offset(leaf, fi);
1093 btrfs_init_generic_ref(&ref, BTRFS_ADD_DELAYED_REF, new_bytenr,
1094 num_bytes, parent);
1095 btrfs_init_data_ref(&ref, btrfs_header_owner(leaf),
1096 key.objectid, key.offset,
1097 root->root_key.objectid, false);
1098 ret = btrfs_inc_extent_ref(trans, &ref);
1099 if (ret) {
1100 btrfs_abort_transaction(trans, ret);
1101 break;
1102 }
1103
1104 btrfs_init_generic_ref(&ref, BTRFS_DROP_DELAYED_REF, bytenr,
1105 num_bytes, parent);
1106 btrfs_init_data_ref(&ref, btrfs_header_owner(leaf),
1107 key.objectid, key.offset,
1108 root->root_key.objectid, false);
1109 ret = btrfs_free_extent(trans, &ref);
1110 if (ret) {
1111 btrfs_abort_transaction(trans, ret);
1112 break;
1113 }
1114 }
1115 if (dirty)
1116 btrfs_mark_buffer_dirty(trans, leaf);
1117 if (inode)
1118 btrfs_add_delayed_iput(BTRFS_I(inode));
1119 return ret;
1120 }
1121
memcmp_node_keys(const struct extent_buffer * eb,int slot,const struct btrfs_path * path,int level)1122 static noinline_for_stack int memcmp_node_keys(const struct extent_buffer *eb,
1123 int slot, const struct btrfs_path *path,
1124 int level)
1125 {
1126 struct btrfs_disk_key key1;
1127 struct btrfs_disk_key key2;
1128 btrfs_node_key(eb, &key1, slot);
1129 btrfs_node_key(path->nodes[level], &key2, path->slots[level]);
1130 return memcmp(&key1, &key2, sizeof(key1));
1131 }
1132
1133 /*
1134 * try to replace tree blocks in fs tree with the new blocks
1135 * in reloc tree. tree blocks haven't been modified since the
1136 * reloc tree was create can be replaced.
1137 *
1138 * if a block was replaced, level of the block + 1 is returned.
1139 * if no block got replaced, 0 is returned. if there are other
1140 * errors, a negative error number is returned.
1141 */
1142 static noinline_for_stack
replace_path(struct btrfs_trans_handle * trans,struct reloc_control * rc,struct btrfs_root * dest,struct btrfs_root * src,struct btrfs_path * path,struct btrfs_key * next_key,int lowest_level,int max_level)1143 int replace_path(struct btrfs_trans_handle *trans, struct reloc_control *rc,
1144 struct btrfs_root *dest, struct btrfs_root *src,
1145 struct btrfs_path *path, struct btrfs_key *next_key,
1146 int lowest_level, int max_level)
1147 {
1148 struct btrfs_fs_info *fs_info = dest->fs_info;
1149 struct extent_buffer *eb;
1150 struct extent_buffer *parent;
1151 struct btrfs_ref ref = { 0 };
1152 struct btrfs_key key;
1153 u64 old_bytenr;
1154 u64 new_bytenr;
1155 u64 old_ptr_gen;
1156 u64 new_ptr_gen;
1157 u64 last_snapshot;
1158 u32 blocksize;
1159 int cow = 0;
1160 int level;
1161 int ret;
1162 int slot;
1163
1164 ASSERT(src->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID);
1165 ASSERT(dest->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
1166
1167 last_snapshot = btrfs_root_last_snapshot(&src->root_item);
1168 again:
1169 slot = path->slots[lowest_level];
1170 btrfs_node_key_to_cpu(path->nodes[lowest_level], &key, slot);
1171
1172 eb = btrfs_lock_root_node(dest);
1173 level = btrfs_header_level(eb);
1174
1175 if (level < lowest_level) {
1176 btrfs_tree_unlock(eb);
1177 free_extent_buffer(eb);
1178 return 0;
1179 }
1180
1181 if (cow) {
1182 ret = btrfs_cow_block(trans, dest, eb, NULL, 0, &eb,
1183 BTRFS_NESTING_COW);
1184 if (ret) {
1185 btrfs_tree_unlock(eb);
1186 free_extent_buffer(eb);
1187 return ret;
1188 }
1189 }
1190
1191 if (next_key) {
1192 next_key->objectid = (u64)-1;
1193 next_key->type = (u8)-1;
1194 next_key->offset = (u64)-1;
1195 }
1196
1197 parent = eb;
1198 while (1) {
1199 level = btrfs_header_level(parent);
1200 ASSERT(level >= lowest_level);
1201
1202 ret = btrfs_bin_search(parent, 0, &key, &slot);
1203 if (ret < 0)
1204 break;
1205 if (ret && slot > 0)
1206 slot--;
1207
1208 if (next_key && slot + 1 < btrfs_header_nritems(parent))
1209 btrfs_node_key_to_cpu(parent, next_key, slot + 1);
1210
1211 old_bytenr = btrfs_node_blockptr(parent, slot);
1212 blocksize = fs_info->nodesize;
1213 old_ptr_gen = btrfs_node_ptr_generation(parent, slot);
1214
1215 if (level <= max_level) {
1216 eb = path->nodes[level];
1217 new_bytenr = btrfs_node_blockptr(eb,
1218 path->slots[level]);
1219 new_ptr_gen = btrfs_node_ptr_generation(eb,
1220 path->slots[level]);
1221 } else {
1222 new_bytenr = 0;
1223 new_ptr_gen = 0;
1224 }
1225
1226 if (WARN_ON(new_bytenr > 0 && new_bytenr == old_bytenr)) {
1227 ret = level;
1228 break;
1229 }
1230
1231 if (new_bytenr == 0 || old_ptr_gen > last_snapshot ||
1232 memcmp_node_keys(parent, slot, path, level)) {
1233 if (level <= lowest_level) {
1234 ret = 0;
1235 break;
1236 }
1237
1238 eb = btrfs_read_node_slot(parent, slot);
1239 if (IS_ERR(eb)) {
1240 ret = PTR_ERR(eb);
1241 break;
1242 }
1243 btrfs_tree_lock(eb);
1244 if (cow) {
1245 ret = btrfs_cow_block(trans, dest, eb, parent,
1246 slot, &eb,
1247 BTRFS_NESTING_COW);
1248 if (ret) {
1249 btrfs_tree_unlock(eb);
1250 free_extent_buffer(eb);
1251 break;
1252 }
1253 }
1254
1255 btrfs_tree_unlock(parent);
1256 free_extent_buffer(parent);
1257
1258 parent = eb;
1259 continue;
1260 }
1261
1262 if (!cow) {
1263 btrfs_tree_unlock(parent);
1264 free_extent_buffer(parent);
1265 cow = 1;
1266 goto again;
1267 }
1268
1269 btrfs_node_key_to_cpu(path->nodes[level], &key,
1270 path->slots[level]);
1271 btrfs_release_path(path);
1272
1273 path->lowest_level = level;
1274 set_bit(BTRFS_ROOT_RESET_LOCKDEP_CLASS, &src->state);
1275 ret = btrfs_search_slot(trans, src, &key, path, 0, 1);
1276 clear_bit(BTRFS_ROOT_RESET_LOCKDEP_CLASS, &src->state);
1277 path->lowest_level = 0;
1278 if (ret) {
1279 if (ret > 0)
1280 ret = -ENOENT;
1281 break;
1282 }
1283
1284 /*
1285 * Info qgroup to trace both subtrees.
1286 *
1287 * We must trace both trees.
1288 * 1) Tree reloc subtree
1289 * If not traced, we will leak data numbers
1290 * 2) Fs subtree
1291 * If not traced, we will double count old data
1292 *
1293 * We don't scan the subtree right now, but only record
1294 * the swapped tree blocks.
1295 * The real subtree rescan is delayed until we have new
1296 * CoW on the subtree root node before transaction commit.
1297 */
1298 ret = btrfs_qgroup_add_swapped_blocks(trans, dest,
1299 rc->block_group, parent, slot,
1300 path->nodes[level], path->slots[level],
1301 last_snapshot);
1302 if (ret < 0)
1303 break;
1304 /*
1305 * swap blocks in fs tree and reloc tree.
1306 */
1307 btrfs_set_node_blockptr(parent, slot, new_bytenr);
1308 btrfs_set_node_ptr_generation(parent, slot, new_ptr_gen);
1309 btrfs_mark_buffer_dirty(trans, parent);
1310
1311 btrfs_set_node_blockptr(path->nodes[level],
1312 path->slots[level], old_bytenr);
1313 btrfs_set_node_ptr_generation(path->nodes[level],
1314 path->slots[level], old_ptr_gen);
1315 btrfs_mark_buffer_dirty(trans, path->nodes[level]);
1316
1317 btrfs_init_generic_ref(&ref, BTRFS_ADD_DELAYED_REF, old_bytenr,
1318 blocksize, path->nodes[level]->start);
1319 btrfs_init_tree_ref(&ref, level - 1, src->root_key.objectid,
1320 0, true);
1321 ret = btrfs_inc_extent_ref(trans, &ref);
1322 if (ret) {
1323 btrfs_abort_transaction(trans, ret);
1324 break;
1325 }
1326 btrfs_init_generic_ref(&ref, BTRFS_ADD_DELAYED_REF, new_bytenr,
1327 blocksize, 0);
1328 btrfs_init_tree_ref(&ref, level - 1, dest->root_key.objectid, 0,
1329 true);
1330 ret = btrfs_inc_extent_ref(trans, &ref);
1331 if (ret) {
1332 btrfs_abort_transaction(trans, ret);
1333 break;
1334 }
1335
1336 btrfs_init_generic_ref(&ref, BTRFS_DROP_DELAYED_REF, new_bytenr,
1337 blocksize, path->nodes[level]->start);
1338 btrfs_init_tree_ref(&ref, level - 1, src->root_key.objectid,
1339 0, true);
1340 ret = btrfs_free_extent(trans, &ref);
1341 if (ret) {
1342 btrfs_abort_transaction(trans, ret);
1343 break;
1344 }
1345
1346 btrfs_init_generic_ref(&ref, BTRFS_DROP_DELAYED_REF, old_bytenr,
1347 blocksize, 0);
1348 btrfs_init_tree_ref(&ref, level - 1, dest->root_key.objectid,
1349 0, true);
1350 ret = btrfs_free_extent(trans, &ref);
1351 if (ret) {
1352 btrfs_abort_transaction(trans, ret);
1353 break;
1354 }
1355
1356 btrfs_unlock_up_safe(path, 0);
1357
1358 ret = level;
1359 break;
1360 }
1361 btrfs_tree_unlock(parent);
1362 free_extent_buffer(parent);
1363 return ret;
1364 }
1365
1366 /*
1367 * helper to find next relocated block in reloc tree
1368 */
1369 static noinline_for_stack
walk_up_reloc_tree(struct btrfs_root * root,struct btrfs_path * path,int * level)1370 int walk_up_reloc_tree(struct btrfs_root *root, struct btrfs_path *path,
1371 int *level)
1372 {
1373 struct extent_buffer *eb;
1374 int i;
1375 u64 last_snapshot;
1376 u32 nritems;
1377
1378 last_snapshot = btrfs_root_last_snapshot(&root->root_item);
1379
1380 for (i = 0; i < *level; i++) {
1381 free_extent_buffer(path->nodes[i]);
1382 path->nodes[i] = NULL;
1383 }
1384
1385 for (i = *level; i < BTRFS_MAX_LEVEL && path->nodes[i]; i++) {
1386 eb = path->nodes[i];
1387 nritems = btrfs_header_nritems(eb);
1388 while (path->slots[i] + 1 < nritems) {
1389 path->slots[i]++;
1390 if (btrfs_node_ptr_generation(eb, path->slots[i]) <=
1391 last_snapshot)
1392 continue;
1393
1394 *level = i;
1395 return 0;
1396 }
1397 free_extent_buffer(path->nodes[i]);
1398 path->nodes[i] = NULL;
1399 }
1400 return 1;
1401 }
1402
1403 /*
1404 * walk down reloc tree to find relocated block of lowest level
1405 */
1406 static noinline_for_stack
walk_down_reloc_tree(struct btrfs_root * root,struct btrfs_path * path,int * level)1407 int walk_down_reloc_tree(struct btrfs_root *root, struct btrfs_path *path,
1408 int *level)
1409 {
1410 struct extent_buffer *eb = NULL;
1411 int i;
1412 u64 ptr_gen = 0;
1413 u64 last_snapshot;
1414 u32 nritems;
1415
1416 last_snapshot = btrfs_root_last_snapshot(&root->root_item);
1417
1418 for (i = *level; i > 0; i--) {
1419 eb = path->nodes[i];
1420 nritems = btrfs_header_nritems(eb);
1421 while (path->slots[i] < nritems) {
1422 ptr_gen = btrfs_node_ptr_generation(eb, path->slots[i]);
1423 if (ptr_gen > last_snapshot)
1424 break;
1425 path->slots[i]++;
1426 }
1427 if (path->slots[i] >= nritems) {
1428 if (i == *level)
1429 break;
1430 *level = i + 1;
1431 return 0;
1432 }
1433 if (i == 1) {
1434 *level = i;
1435 return 0;
1436 }
1437
1438 eb = btrfs_read_node_slot(eb, path->slots[i]);
1439 if (IS_ERR(eb))
1440 return PTR_ERR(eb);
1441 BUG_ON(btrfs_header_level(eb) != i - 1);
1442 path->nodes[i - 1] = eb;
1443 path->slots[i - 1] = 0;
1444 }
1445 return 1;
1446 }
1447
1448 /*
1449 * invalidate extent cache for file extents whose key in range of
1450 * [min_key, max_key)
1451 */
invalidate_extent_cache(struct btrfs_root * root,const struct btrfs_key * min_key,const struct btrfs_key * max_key)1452 static int invalidate_extent_cache(struct btrfs_root *root,
1453 const struct btrfs_key *min_key,
1454 const struct btrfs_key *max_key)
1455 {
1456 struct btrfs_fs_info *fs_info = root->fs_info;
1457 struct inode *inode = NULL;
1458 u64 objectid;
1459 u64 start, end;
1460 u64 ino;
1461
1462 objectid = min_key->objectid;
1463 while (1) {
1464 struct extent_state *cached_state = NULL;
1465
1466 cond_resched();
1467 iput(inode);
1468
1469 if (objectid > max_key->objectid)
1470 break;
1471
1472 inode = find_next_inode(root, objectid);
1473 if (!inode)
1474 break;
1475 ino = btrfs_ino(BTRFS_I(inode));
1476
1477 if (ino > max_key->objectid) {
1478 iput(inode);
1479 break;
1480 }
1481
1482 objectid = ino + 1;
1483 if (!S_ISREG(inode->i_mode))
1484 continue;
1485
1486 if (unlikely(min_key->objectid == ino)) {
1487 if (min_key->type > BTRFS_EXTENT_DATA_KEY)
1488 continue;
1489 if (min_key->type < BTRFS_EXTENT_DATA_KEY)
1490 start = 0;
1491 else {
1492 start = min_key->offset;
1493 WARN_ON(!IS_ALIGNED(start, fs_info->sectorsize));
1494 }
1495 } else {
1496 start = 0;
1497 }
1498
1499 if (unlikely(max_key->objectid == ino)) {
1500 if (max_key->type < BTRFS_EXTENT_DATA_KEY)
1501 continue;
1502 if (max_key->type > BTRFS_EXTENT_DATA_KEY) {
1503 end = (u64)-1;
1504 } else {
1505 if (max_key->offset == 0)
1506 continue;
1507 end = max_key->offset;
1508 WARN_ON(!IS_ALIGNED(end, fs_info->sectorsize));
1509 end--;
1510 }
1511 } else {
1512 end = (u64)-1;
1513 }
1514
1515 /* the lock_extent waits for read_folio to complete */
1516 lock_extent(&BTRFS_I(inode)->io_tree, start, end, &cached_state);
1517 btrfs_drop_extent_map_range(BTRFS_I(inode), start, end, true);
1518 unlock_extent(&BTRFS_I(inode)->io_tree, start, end, &cached_state);
1519 }
1520 return 0;
1521 }
1522
find_next_key(struct btrfs_path * path,int level,struct btrfs_key * key)1523 static int find_next_key(struct btrfs_path *path, int level,
1524 struct btrfs_key *key)
1525
1526 {
1527 while (level < BTRFS_MAX_LEVEL) {
1528 if (!path->nodes[level])
1529 break;
1530 if (path->slots[level] + 1 <
1531 btrfs_header_nritems(path->nodes[level])) {
1532 btrfs_node_key_to_cpu(path->nodes[level], key,
1533 path->slots[level] + 1);
1534 return 0;
1535 }
1536 level++;
1537 }
1538 return 1;
1539 }
1540
1541 /*
1542 * Insert current subvolume into reloc_control::dirty_subvol_roots
1543 */
insert_dirty_subvol(struct btrfs_trans_handle * trans,struct reloc_control * rc,struct btrfs_root * root)1544 static int insert_dirty_subvol(struct btrfs_trans_handle *trans,
1545 struct reloc_control *rc,
1546 struct btrfs_root *root)
1547 {
1548 struct btrfs_root *reloc_root = root->reloc_root;
1549 struct btrfs_root_item *reloc_root_item;
1550 int ret;
1551
1552 /* @root must be a subvolume tree root with a valid reloc tree */
1553 ASSERT(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
1554 ASSERT(reloc_root);
1555
1556 reloc_root_item = &reloc_root->root_item;
1557 memset(&reloc_root_item->drop_progress, 0,
1558 sizeof(reloc_root_item->drop_progress));
1559 btrfs_set_root_drop_level(reloc_root_item, 0);
1560 btrfs_set_root_refs(reloc_root_item, 0);
1561 ret = btrfs_update_reloc_root(trans, root);
1562 if (ret)
1563 return ret;
1564
1565 if (list_empty(&root->reloc_dirty_list)) {
1566 btrfs_grab_root(root);
1567 list_add_tail(&root->reloc_dirty_list, &rc->dirty_subvol_roots);
1568 }
1569
1570 return 0;
1571 }
1572
clean_dirty_subvols(struct reloc_control * rc)1573 static int clean_dirty_subvols(struct reloc_control *rc)
1574 {
1575 struct btrfs_root *root;
1576 struct btrfs_root *next;
1577 int ret = 0;
1578 int ret2;
1579
1580 list_for_each_entry_safe(root, next, &rc->dirty_subvol_roots,
1581 reloc_dirty_list) {
1582 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
1583 /* Merged subvolume, cleanup its reloc root */
1584 struct btrfs_root *reloc_root = root->reloc_root;
1585
1586 list_del_init(&root->reloc_dirty_list);
1587 root->reloc_root = NULL;
1588 /*
1589 * Need barrier to ensure clear_bit() only happens after
1590 * root->reloc_root = NULL. Pairs with have_reloc_root.
1591 */
1592 smp_wmb();
1593 clear_bit(BTRFS_ROOT_DEAD_RELOC_TREE, &root->state);
1594 if (reloc_root) {
1595 /*
1596 * btrfs_drop_snapshot drops our ref we hold for
1597 * ->reloc_root. If it fails however we must
1598 * drop the ref ourselves.
1599 */
1600 ret2 = btrfs_drop_snapshot(reloc_root, 0, 1);
1601 if (ret2 < 0) {
1602 btrfs_put_root(reloc_root);
1603 if (!ret)
1604 ret = ret2;
1605 }
1606 }
1607 btrfs_put_root(root);
1608 } else {
1609 /* Orphan reloc tree, just clean it up */
1610 ret2 = btrfs_drop_snapshot(root, 0, 1);
1611 if (ret2 < 0) {
1612 btrfs_put_root(root);
1613 if (!ret)
1614 ret = ret2;
1615 }
1616 }
1617 }
1618 return ret;
1619 }
1620
1621 /*
1622 * merge the relocated tree blocks in reloc tree with corresponding
1623 * fs tree.
1624 */
merge_reloc_root(struct reloc_control * rc,struct btrfs_root * root)1625 static noinline_for_stack int merge_reloc_root(struct reloc_control *rc,
1626 struct btrfs_root *root)
1627 {
1628 struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
1629 struct btrfs_key key;
1630 struct btrfs_key next_key;
1631 struct btrfs_trans_handle *trans = NULL;
1632 struct btrfs_root *reloc_root;
1633 struct btrfs_root_item *root_item;
1634 struct btrfs_path *path;
1635 struct extent_buffer *leaf;
1636 int reserve_level;
1637 int level;
1638 int max_level;
1639 int replaced = 0;
1640 int ret = 0;
1641 u32 min_reserved;
1642
1643 path = btrfs_alloc_path();
1644 if (!path)
1645 return -ENOMEM;
1646 path->reada = READA_FORWARD;
1647
1648 reloc_root = root->reloc_root;
1649 root_item = &reloc_root->root_item;
1650
1651 if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
1652 level = btrfs_root_level(root_item);
1653 atomic_inc(&reloc_root->node->refs);
1654 path->nodes[level] = reloc_root->node;
1655 path->slots[level] = 0;
1656 } else {
1657 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
1658
1659 level = btrfs_root_drop_level(root_item);
1660 BUG_ON(level == 0);
1661 path->lowest_level = level;
1662 ret = btrfs_search_slot(NULL, reloc_root, &key, path, 0, 0);
1663 path->lowest_level = 0;
1664 if (ret < 0) {
1665 btrfs_free_path(path);
1666 return ret;
1667 }
1668
1669 btrfs_node_key_to_cpu(path->nodes[level], &next_key,
1670 path->slots[level]);
1671 WARN_ON(memcmp(&key, &next_key, sizeof(key)));
1672
1673 btrfs_unlock_up_safe(path, 0);
1674 }
1675
1676 /*
1677 * In merge_reloc_root(), we modify the upper level pointer to swap the
1678 * tree blocks between reloc tree and subvolume tree. Thus for tree
1679 * block COW, we COW at most from level 1 to root level for each tree.
1680 *
1681 * Thus the needed metadata size is at most root_level * nodesize,
1682 * and * 2 since we have two trees to COW.
1683 */
1684 reserve_level = max_t(int, 1, btrfs_root_level(root_item));
1685 min_reserved = fs_info->nodesize * reserve_level * 2;
1686 memset(&next_key, 0, sizeof(next_key));
1687
1688 while (1) {
1689 ret = btrfs_block_rsv_refill(fs_info, rc->block_rsv,
1690 min_reserved,
1691 BTRFS_RESERVE_FLUSH_LIMIT);
1692 if (ret)
1693 goto out;
1694 trans = btrfs_start_transaction(root, 0);
1695 if (IS_ERR(trans)) {
1696 ret = PTR_ERR(trans);
1697 trans = NULL;
1698 goto out;
1699 }
1700
1701 /*
1702 * At this point we no longer have a reloc_control, so we can't
1703 * depend on btrfs_init_reloc_root to update our last_trans.
1704 *
1705 * But that's ok, we started the trans handle on our
1706 * corresponding fs_root, which means it's been added to the
1707 * dirty list. At commit time we'll still call
1708 * btrfs_update_reloc_root() and update our root item
1709 * appropriately.
1710 */
1711 reloc_root->last_trans = trans->transid;
1712 trans->block_rsv = rc->block_rsv;
1713
1714 replaced = 0;
1715 max_level = level;
1716
1717 ret = walk_down_reloc_tree(reloc_root, path, &level);
1718 if (ret < 0)
1719 goto out;
1720 if (ret > 0)
1721 break;
1722
1723 if (!find_next_key(path, level, &key) &&
1724 btrfs_comp_cpu_keys(&next_key, &key) >= 0) {
1725 ret = 0;
1726 } else {
1727 ret = replace_path(trans, rc, root, reloc_root, path,
1728 &next_key, level, max_level);
1729 }
1730 if (ret < 0)
1731 goto out;
1732 if (ret > 0) {
1733 level = ret;
1734 btrfs_node_key_to_cpu(path->nodes[level], &key,
1735 path->slots[level]);
1736 replaced = 1;
1737 }
1738
1739 ret = walk_up_reloc_tree(reloc_root, path, &level);
1740 if (ret > 0)
1741 break;
1742
1743 BUG_ON(level == 0);
1744 /*
1745 * save the merging progress in the drop_progress.
1746 * this is OK since root refs == 1 in this case.
1747 */
1748 btrfs_node_key(path->nodes[level], &root_item->drop_progress,
1749 path->slots[level]);
1750 btrfs_set_root_drop_level(root_item, level);
1751
1752 btrfs_end_transaction_throttle(trans);
1753 trans = NULL;
1754
1755 btrfs_btree_balance_dirty(fs_info);
1756
1757 if (replaced && rc->stage == UPDATE_DATA_PTRS)
1758 invalidate_extent_cache(root, &key, &next_key);
1759 }
1760
1761 /*
1762 * handle the case only one block in the fs tree need to be
1763 * relocated and the block is tree root.
1764 */
1765 leaf = btrfs_lock_root_node(root);
1766 ret = btrfs_cow_block(trans, root, leaf, NULL, 0, &leaf,
1767 BTRFS_NESTING_COW);
1768 btrfs_tree_unlock(leaf);
1769 free_extent_buffer(leaf);
1770 out:
1771 btrfs_free_path(path);
1772
1773 if (ret == 0) {
1774 ret = insert_dirty_subvol(trans, rc, root);
1775 if (ret)
1776 btrfs_abort_transaction(trans, ret);
1777 }
1778
1779 if (trans)
1780 btrfs_end_transaction_throttle(trans);
1781
1782 btrfs_btree_balance_dirty(fs_info);
1783
1784 if (replaced && rc->stage == UPDATE_DATA_PTRS)
1785 invalidate_extent_cache(root, &key, &next_key);
1786
1787 return ret;
1788 }
1789
1790 static noinline_for_stack
prepare_to_merge(struct reloc_control * rc,int err)1791 int prepare_to_merge(struct reloc_control *rc, int err)
1792 {
1793 struct btrfs_root *root = rc->extent_root;
1794 struct btrfs_fs_info *fs_info = root->fs_info;
1795 struct btrfs_root *reloc_root;
1796 struct btrfs_trans_handle *trans;
1797 LIST_HEAD(reloc_roots);
1798 u64 num_bytes = 0;
1799 int ret;
1800
1801 mutex_lock(&fs_info->reloc_mutex);
1802 rc->merging_rsv_size += fs_info->nodesize * (BTRFS_MAX_LEVEL - 1) * 2;
1803 rc->merging_rsv_size += rc->nodes_relocated * 2;
1804 mutex_unlock(&fs_info->reloc_mutex);
1805
1806 again:
1807 if (!err) {
1808 num_bytes = rc->merging_rsv_size;
1809 ret = btrfs_block_rsv_add(fs_info, rc->block_rsv, num_bytes,
1810 BTRFS_RESERVE_FLUSH_ALL);
1811 if (ret)
1812 err = ret;
1813 }
1814
1815 trans = btrfs_join_transaction(rc->extent_root);
1816 if (IS_ERR(trans)) {
1817 if (!err)
1818 btrfs_block_rsv_release(fs_info, rc->block_rsv,
1819 num_bytes, NULL);
1820 return PTR_ERR(trans);
1821 }
1822
1823 if (!err) {
1824 if (num_bytes != rc->merging_rsv_size) {
1825 btrfs_end_transaction(trans);
1826 btrfs_block_rsv_release(fs_info, rc->block_rsv,
1827 num_bytes, NULL);
1828 goto again;
1829 }
1830 }
1831
1832 rc->merge_reloc_tree = 1;
1833
1834 while (!list_empty(&rc->reloc_roots)) {
1835 reloc_root = list_entry(rc->reloc_roots.next,
1836 struct btrfs_root, root_list);
1837 list_del_init(&reloc_root->root_list);
1838
1839 root = btrfs_get_fs_root(fs_info, reloc_root->root_key.offset,
1840 false);
1841 if (IS_ERR(root)) {
1842 /*
1843 * Even if we have an error we need this reloc root
1844 * back on our list so we can clean up properly.
1845 */
1846 list_add(&reloc_root->root_list, &reloc_roots);
1847 btrfs_abort_transaction(trans, (int)PTR_ERR(root));
1848 if (!err)
1849 err = PTR_ERR(root);
1850 break;
1851 }
1852
1853 if (unlikely(root->reloc_root != reloc_root)) {
1854 if (root->reloc_root) {
1855 btrfs_err(fs_info,
1856 "reloc tree mismatch, root %lld has reloc root key (%lld %u %llu) gen %llu, expect reloc root key (%lld %u %llu) gen %llu",
1857 root->root_key.objectid,
1858 root->reloc_root->root_key.objectid,
1859 root->reloc_root->root_key.type,
1860 root->reloc_root->root_key.offset,
1861 btrfs_root_generation(
1862 &root->reloc_root->root_item),
1863 reloc_root->root_key.objectid,
1864 reloc_root->root_key.type,
1865 reloc_root->root_key.offset,
1866 btrfs_root_generation(
1867 &reloc_root->root_item));
1868 } else {
1869 btrfs_err(fs_info,
1870 "reloc tree mismatch, root %lld has no reloc root, expect reloc root key (%lld %u %llu) gen %llu",
1871 root->root_key.objectid,
1872 reloc_root->root_key.objectid,
1873 reloc_root->root_key.type,
1874 reloc_root->root_key.offset,
1875 btrfs_root_generation(
1876 &reloc_root->root_item));
1877 }
1878 list_add(&reloc_root->root_list, &reloc_roots);
1879 btrfs_put_root(root);
1880 btrfs_abort_transaction(trans, -EUCLEAN);
1881 if (!err)
1882 err = -EUCLEAN;
1883 break;
1884 }
1885
1886 /*
1887 * set reference count to 1, so btrfs_recover_relocation
1888 * knows it should resumes merging
1889 */
1890 if (!err)
1891 btrfs_set_root_refs(&reloc_root->root_item, 1);
1892 ret = btrfs_update_reloc_root(trans, root);
1893
1894 /*
1895 * Even if we have an error we need this reloc root back on our
1896 * list so we can clean up properly.
1897 */
1898 list_add(&reloc_root->root_list, &reloc_roots);
1899 btrfs_put_root(root);
1900
1901 if (ret) {
1902 btrfs_abort_transaction(trans, ret);
1903 if (!err)
1904 err = ret;
1905 break;
1906 }
1907 }
1908
1909 list_splice(&reloc_roots, &rc->reloc_roots);
1910
1911 if (!err)
1912 err = btrfs_commit_transaction(trans);
1913 else
1914 btrfs_end_transaction(trans);
1915 return err;
1916 }
1917
1918 static noinline_for_stack
free_reloc_roots(struct list_head * list)1919 void free_reloc_roots(struct list_head *list)
1920 {
1921 struct btrfs_root *reloc_root, *tmp;
1922
1923 list_for_each_entry_safe(reloc_root, tmp, list, root_list)
1924 __del_reloc_root(reloc_root);
1925 }
1926
1927 static noinline_for_stack
merge_reloc_roots(struct reloc_control * rc)1928 void merge_reloc_roots(struct reloc_control *rc)
1929 {
1930 struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
1931 struct btrfs_root *root;
1932 struct btrfs_root *reloc_root;
1933 LIST_HEAD(reloc_roots);
1934 int found = 0;
1935 int ret = 0;
1936 again:
1937 root = rc->extent_root;
1938
1939 /*
1940 * this serializes us with btrfs_record_root_in_transaction,
1941 * we have to make sure nobody is in the middle of
1942 * adding their roots to the list while we are
1943 * doing this splice
1944 */
1945 mutex_lock(&fs_info->reloc_mutex);
1946 list_splice_init(&rc->reloc_roots, &reloc_roots);
1947 mutex_unlock(&fs_info->reloc_mutex);
1948
1949 while (!list_empty(&reloc_roots)) {
1950 found = 1;
1951 reloc_root = list_entry(reloc_roots.next,
1952 struct btrfs_root, root_list);
1953
1954 root = btrfs_get_fs_root(fs_info, reloc_root->root_key.offset,
1955 false);
1956 if (btrfs_root_refs(&reloc_root->root_item) > 0) {
1957 if (WARN_ON(IS_ERR(root))) {
1958 /*
1959 * For recovery we read the fs roots on mount,
1960 * and if we didn't find the root then we marked
1961 * the reloc root as a garbage root. For normal
1962 * relocation obviously the root should exist in
1963 * memory. However there's no reason we can't
1964 * handle the error properly here just in case.
1965 */
1966 ret = PTR_ERR(root);
1967 goto out;
1968 }
1969 if (WARN_ON(root->reloc_root != reloc_root)) {
1970 /*
1971 * This can happen if on-disk metadata has some
1972 * corruption, e.g. bad reloc tree key offset.
1973 */
1974 ret = -EINVAL;
1975 goto out;
1976 }
1977 ret = merge_reloc_root(rc, root);
1978 btrfs_put_root(root);
1979 if (ret) {
1980 if (list_empty(&reloc_root->root_list))
1981 list_add_tail(&reloc_root->root_list,
1982 &reloc_roots);
1983 goto out;
1984 }
1985 } else {
1986 if (!IS_ERR(root)) {
1987 if (root->reloc_root == reloc_root) {
1988 root->reloc_root = NULL;
1989 btrfs_put_root(reloc_root);
1990 }
1991 clear_bit(BTRFS_ROOT_DEAD_RELOC_TREE,
1992 &root->state);
1993 btrfs_put_root(root);
1994 }
1995
1996 list_del_init(&reloc_root->root_list);
1997 /* Don't forget to queue this reloc root for cleanup */
1998 list_add_tail(&reloc_root->reloc_dirty_list,
1999 &rc->dirty_subvol_roots);
2000 }
2001 }
2002
2003 if (found) {
2004 found = 0;
2005 goto again;
2006 }
2007 out:
2008 if (ret) {
2009 btrfs_handle_fs_error(fs_info, ret, NULL);
2010 free_reloc_roots(&reloc_roots);
2011
2012 /* new reloc root may be added */
2013 mutex_lock(&fs_info->reloc_mutex);
2014 list_splice_init(&rc->reloc_roots, &reloc_roots);
2015 mutex_unlock(&fs_info->reloc_mutex);
2016 free_reloc_roots(&reloc_roots);
2017 }
2018
2019 /*
2020 * We used to have
2021 *
2022 * BUG_ON(!RB_EMPTY_ROOT(&rc->reloc_root_tree.rb_root));
2023 *
2024 * here, but it's wrong. If we fail to start the transaction in
2025 * prepare_to_merge() we will have only 0 ref reloc roots, none of which
2026 * have actually been removed from the reloc_root_tree rb tree. This is
2027 * fine because we're bailing here, and we hold a reference on the root
2028 * for the list that holds it, so these roots will be cleaned up when we
2029 * do the reloc_dirty_list afterwards. Meanwhile the root->reloc_root
2030 * will be cleaned up on unmount.
2031 *
2032 * The remaining nodes will be cleaned up by free_reloc_control.
2033 */
2034 }
2035
free_block_list(struct rb_root * blocks)2036 static void free_block_list(struct rb_root *blocks)
2037 {
2038 struct tree_block *block;
2039 struct rb_node *rb_node;
2040 while ((rb_node = rb_first(blocks))) {
2041 block = rb_entry(rb_node, struct tree_block, rb_node);
2042 rb_erase(rb_node, blocks);
2043 kfree(block);
2044 }
2045 }
2046
record_reloc_root_in_trans(struct btrfs_trans_handle * trans,struct btrfs_root * reloc_root)2047 static int record_reloc_root_in_trans(struct btrfs_trans_handle *trans,
2048 struct btrfs_root *reloc_root)
2049 {
2050 struct btrfs_fs_info *fs_info = reloc_root->fs_info;
2051 struct btrfs_root *root;
2052 int ret;
2053
2054 if (reloc_root->last_trans == trans->transid)
2055 return 0;
2056
2057 root = btrfs_get_fs_root(fs_info, reloc_root->root_key.offset, false);
2058
2059 /*
2060 * This should succeed, since we can't have a reloc root without having
2061 * already looked up the actual root and created the reloc root for this
2062 * root.
2063 *
2064 * However if there's some sort of corruption where we have a ref to a
2065 * reloc root without a corresponding root this could return ENOENT.
2066 */
2067 if (IS_ERR(root)) {
2068 ASSERT(0);
2069 return PTR_ERR(root);
2070 }
2071 if (root->reloc_root != reloc_root) {
2072 ASSERT(0);
2073 btrfs_err(fs_info,
2074 "root %llu has two reloc roots associated with it",
2075 reloc_root->root_key.offset);
2076 btrfs_put_root(root);
2077 return -EUCLEAN;
2078 }
2079 ret = btrfs_record_root_in_trans(trans, root);
2080 btrfs_put_root(root);
2081
2082 return ret;
2083 }
2084
2085 static noinline_for_stack
select_reloc_root(struct btrfs_trans_handle * trans,struct reloc_control * rc,struct btrfs_backref_node * node,struct btrfs_backref_edge * edges[])2086 struct btrfs_root *select_reloc_root(struct btrfs_trans_handle *trans,
2087 struct reloc_control *rc,
2088 struct btrfs_backref_node *node,
2089 struct btrfs_backref_edge *edges[])
2090 {
2091 struct btrfs_backref_node *next;
2092 struct btrfs_root *root;
2093 int index = 0;
2094 int ret;
2095
2096 next = node;
2097 while (1) {
2098 cond_resched();
2099 next = walk_up_backref(next, edges, &index);
2100 root = next->root;
2101
2102 /*
2103 * If there is no root, then our references for this block are
2104 * incomplete, as we should be able to walk all the way up to a
2105 * block that is owned by a root.
2106 *
2107 * This path is only for SHAREABLE roots, so if we come upon a
2108 * non-SHAREABLE root then we have backrefs that resolve
2109 * improperly.
2110 *
2111 * Both of these cases indicate file system corruption, or a bug
2112 * in the backref walking code.
2113 */
2114 if (!root) {
2115 ASSERT(0);
2116 btrfs_err(trans->fs_info,
2117 "bytenr %llu doesn't have a backref path ending in a root",
2118 node->bytenr);
2119 return ERR_PTR(-EUCLEAN);
2120 }
2121 if (!test_bit(BTRFS_ROOT_SHAREABLE, &root->state)) {
2122 ASSERT(0);
2123 btrfs_err(trans->fs_info,
2124 "bytenr %llu has multiple refs with one ending in a non-shareable root",
2125 node->bytenr);
2126 return ERR_PTR(-EUCLEAN);
2127 }
2128
2129 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) {
2130 ret = record_reloc_root_in_trans(trans, root);
2131 if (ret)
2132 return ERR_PTR(ret);
2133 break;
2134 }
2135
2136 ret = btrfs_record_root_in_trans(trans, root);
2137 if (ret)
2138 return ERR_PTR(ret);
2139 root = root->reloc_root;
2140
2141 /*
2142 * We could have raced with another thread which failed, so
2143 * root->reloc_root may not be set, return ENOENT in this case.
2144 */
2145 if (!root)
2146 return ERR_PTR(-ENOENT);
2147
2148 if (next->new_bytenr != root->node->start) {
2149 /*
2150 * We just created the reloc root, so we shouldn't have
2151 * ->new_bytenr set and this shouldn't be in the changed
2152 * list. If it is then we have multiple roots pointing
2153 * at the same bytenr which indicates corruption, or
2154 * we've made a mistake in the backref walking code.
2155 */
2156 ASSERT(next->new_bytenr == 0);
2157 ASSERT(list_empty(&next->list));
2158 if (next->new_bytenr || !list_empty(&next->list)) {
2159 btrfs_err(trans->fs_info,
2160 "bytenr %llu possibly has multiple roots pointing at the same bytenr %llu",
2161 node->bytenr, next->bytenr);
2162 return ERR_PTR(-EUCLEAN);
2163 }
2164
2165 next->new_bytenr = root->node->start;
2166 btrfs_put_root(next->root);
2167 next->root = btrfs_grab_root(root);
2168 ASSERT(next->root);
2169 list_add_tail(&next->list,
2170 &rc->backref_cache.changed);
2171 mark_block_processed(rc, next);
2172 break;
2173 }
2174
2175 WARN_ON(1);
2176 root = NULL;
2177 next = walk_down_backref(edges, &index);
2178 if (!next || next->level <= node->level)
2179 break;
2180 }
2181 if (!root) {
2182 /*
2183 * This can happen if there's fs corruption or if there's a bug
2184 * in the backref lookup code.
2185 */
2186 ASSERT(0);
2187 return ERR_PTR(-ENOENT);
2188 }
2189
2190 next = node;
2191 /* setup backref node path for btrfs_reloc_cow_block */
2192 while (1) {
2193 rc->backref_cache.path[next->level] = next;
2194 if (--index < 0)
2195 break;
2196 next = edges[index]->node[UPPER];
2197 }
2198 return root;
2199 }
2200
2201 /*
2202 * Select a tree root for relocation.
2203 *
2204 * Return NULL if the block is not shareable. We should use do_relocation() in
2205 * this case.
2206 *
2207 * Return a tree root pointer if the block is shareable.
2208 * Return -ENOENT if the block is root of reloc tree.
2209 */
2210 static noinline_for_stack
select_one_root(struct btrfs_backref_node * node)2211 struct btrfs_root *select_one_root(struct btrfs_backref_node *node)
2212 {
2213 struct btrfs_backref_node *next;
2214 struct btrfs_root *root;
2215 struct btrfs_root *fs_root = NULL;
2216 struct btrfs_backref_edge *edges[BTRFS_MAX_LEVEL - 1];
2217 int index = 0;
2218
2219 next = node;
2220 while (1) {
2221 cond_resched();
2222 next = walk_up_backref(next, edges, &index);
2223 root = next->root;
2224
2225 /*
2226 * This can occur if we have incomplete extent refs leading all
2227 * the way up a particular path, in this case return -EUCLEAN.
2228 */
2229 if (!root)
2230 return ERR_PTR(-EUCLEAN);
2231
2232 /* No other choice for non-shareable tree */
2233 if (!test_bit(BTRFS_ROOT_SHAREABLE, &root->state))
2234 return root;
2235
2236 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID)
2237 fs_root = root;
2238
2239 if (next != node)
2240 return NULL;
2241
2242 next = walk_down_backref(edges, &index);
2243 if (!next || next->level <= node->level)
2244 break;
2245 }
2246
2247 if (!fs_root)
2248 return ERR_PTR(-ENOENT);
2249 return fs_root;
2250 }
2251
2252 static noinline_for_stack
calcu_metadata_size(struct reloc_control * rc,struct btrfs_backref_node * node,int reserve)2253 u64 calcu_metadata_size(struct reloc_control *rc,
2254 struct btrfs_backref_node *node, int reserve)
2255 {
2256 struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
2257 struct btrfs_backref_node *next = node;
2258 struct btrfs_backref_edge *edge;
2259 struct btrfs_backref_edge *edges[BTRFS_MAX_LEVEL - 1];
2260 u64 num_bytes = 0;
2261 int index = 0;
2262
2263 BUG_ON(reserve && node->processed);
2264
2265 while (next) {
2266 cond_resched();
2267 while (1) {
2268 if (next->processed && (reserve || next != node))
2269 break;
2270
2271 num_bytes += fs_info->nodesize;
2272
2273 if (list_empty(&next->upper))
2274 break;
2275
2276 edge = list_entry(next->upper.next,
2277 struct btrfs_backref_edge, list[LOWER]);
2278 edges[index++] = edge;
2279 next = edge->node[UPPER];
2280 }
2281 next = walk_down_backref(edges, &index);
2282 }
2283 return num_bytes;
2284 }
2285
reserve_metadata_space(struct btrfs_trans_handle * trans,struct reloc_control * rc,struct btrfs_backref_node * node)2286 static int reserve_metadata_space(struct btrfs_trans_handle *trans,
2287 struct reloc_control *rc,
2288 struct btrfs_backref_node *node)
2289 {
2290 struct btrfs_root *root = rc->extent_root;
2291 struct btrfs_fs_info *fs_info = root->fs_info;
2292 u64 num_bytes;
2293 int ret;
2294 u64 tmp;
2295
2296 num_bytes = calcu_metadata_size(rc, node, 1) * 2;
2297
2298 trans->block_rsv = rc->block_rsv;
2299 rc->reserved_bytes += num_bytes;
2300
2301 /*
2302 * We are under a transaction here so we can only do limited flushing.
2303 * If we get an enospc just kick back -EAGAIN so we know to drop the
2304 * transaction and try to refill when we can flush all the things.
2305 */
2306 ret = btrfs_block_rsv_refill(fs_info, rc->block_rsv, num_bytes,
2307 BTRFS_RESERVE_FLUSH_LIMIT);
2308 if (ret) {
2309 tmp = fs_info->nodesize * RELOCATION_RESERVED_NODES;
2310 while (tmp <= rc->reserved_bytes)
2311 tmp <<= 1;
2312 /*
2313 * only one thread can access block_rsv at this point,
2314 * so we don't need hold lock to protect block_rsv.
2315 * we expand more reservation size here to allow enough
2316 * space for relocation and we will return earlier in
2317 * enospc case.
2318 */
2319 rc->block_rsv->size = tmp + fs_info->nodesize *
2320 RELOCATION_RESERVED_NODES;
2321 return -EAGAIN;
2322 }
2323
2324 return 0;
2325 }
2326
2327 /*
2328 * relocate a block tree, and then update pointers in upper level
2329 * blocks that reference the block to point to the new location.
2330 *
2331 * if called by link_to_upper, the block has already been relocated.
2332 * in that case this function just updates pointers.
2333 */
do_relocation(struct btrfs_trans_handle * trans,struct reloc_control * rc,struct btrfs_backref_node * node,struct btrfs_key * key,struct btrfs_path * path,int lowest)2334 static int do_relocation(struct btrfs_trans_handle *trans,
2335 struct reloc_control *rc,
2336 struct btrfs_backref_node *node,
2337 struct btrfs_key *key,
2338 struct btrfs_path *path, int lowest)
2339 {
2340 struct btrfs_backref_node *upper;
2341 struct btrfs_backref_edge *edge;
2342 struct btrfs_backref_edge *edges[BTRFS_MAX_LEVEL - 1];
2343 struct btrfs_root *root;
2344 struct extent_buffer *eb;
2345 u32 blocksize;
2346 u64 bytenr;
2347 int slot;
2348 int ret = 0;
2349
2350 /*
2351 * If we are lowest then this is the first time we're processing this
2352 * block, and thus shouldn't have an eb associated with it yet.
2353 */
2354 ASSERT(!lowest || !node->eb);
2355
2356 path->lowest_level = node->level + 1;
2357 rc->backref_cache.path[node->level] = node;
2358 list_for_each_entry(edge, &node->upper, list[LOWER]) {
2359 struct btrfs_ref ref = { 0 };
2360
2361 cond_resched();
2362
2363 upper = edge->node[UPPER];
2364 root = select_reloc_root(trans, rc, upper, edges);
2365 if (IS_ERR(root)) {
2366 ret = PTR_ERR(root);
2367 goto next;
2368 }
2369
2370 if (upper->eb && !upper->locked) {
2371 if (!lowest) {
2372 ret = btrfs_bin_search(upper->eb, 0, key, &slot);
2373 if (ret < 0)
2374 goto next;
2375 BUG_ON(ret);
2376 bytenr = btrfs_node_blockptr(upper->eb, slot);
2377 if (node->eb->start == bytenr)
2378 goto next;
2379 }
2380 btrfs_backref_drop_node_buffer(upper);
2381 }
2382
2383 if (!upper->eb) {
2384 ret = btrfs_search_slot(trans, root, key, path, 0, 1);
2385 if (ret) {
2386 if (ret > 0)
2387 ret = -ENOENT;
2388
2389 btrfs_release_path(path);
2390 break;
2391 }
2392
2393 if (!upper->eb) {
2394 upper->eb = path->nodes[upper->level];
2395 path->nodes[upper->level] = NULL;
2396 } else {
2397 BUG_ON(upper->eb != path->nodes[upper->level]);
2398 }
2399
2400 upper->locked = 1;
2401 path->locks[upper->level] = 0;
2402
2403 slot = path->slots[upper->level];
2404 btrfs_release_path(path);
2405 } else {
2406 ret = btrfs_bin_search(upper->eb, 0, key, &slot);
2407 if (ret < 0)
2408 goto next;
2409 BUG_ON(ret);
2410 }
2411
2412 bytenr = btrfs_node_blockptr(upper->eb, slot);
2413 if (lowest) {
2414 if (bytenr != node->bytenr) {
2415 btrfs_err(root->fs_info,
2416 "lowest leaf/node mismatch: bytenr %llu node->bytenr %llu slot %d upper %llu",
2417 bytenr, node->bytenr, slot,
2418 upper->eb->start);
2419 ret = -EIO;
2420 goto next;
2421 }
2422 } else {
2423 if (node->eb->start == bytenr)
2424 goto next;
2425 }
2426
2427 blocksize = root->fs_info->nodesize;
2428 eb = btrfs_read_node_slot(upper->eb, slot);
2429 if (IS_ERR(eb)) {
2430 ret = PTR_ERR(eb);
2431 goto next;
2432 }
2433 btrfs_tree_lock(eb);
2434
2435 if (!node->eb) {
2436 ret = btrfs_cow_block(trans, root, eb, upper->eb,
2437 slot, &eb, BTRFS_NESTING_COW);
2438 btrfs_tree_unlock(eb);
2439 free_extent_buffer(eb);
2440 if (ret < 0)
2441 goto next;
2442 /*
2443 * We've just COWed this block, it should have updated
2444 * the correct backref node entry.
2445 */
2446 ASSERT(node->eb == eb);
2447 } else {
2448 btrfs_set_node_blockptr(upper->eb, slot,
2449 node->eb->start);
2450 btrfs_set_node_ptr_generation(upper->eb, slot,
2451 trans->transid);
2452 btrfs_mark_buffer_dirty(trans, upper->eb);
2453
2454 btrfs_init_generic_ref(&ref, BTRFS_ADD_DELAYED_REF,
2455 node->eb->start, blocksize,
2456 upper->eb->start);
2457 btrfs_init_tree_ref(&ref, node->level,
2458 btrfs_header_owner(upper->eb),
2459 root->root_key.objectid, false);
2460 ret = btrfs_inc_extent_ref(trans, &ref);
2461 if (!ret)
2462 ret = btrfs_drop_subtree(trans, root, eb,
2463 upper->eb);
2464 if (ret)
2465 btrfs_abort_transaction(trans, ret);
2466 }
2467 next:
2468 if (!upper->pending)
2469 btrfs_backref_drop_node_buffer(upper);
2470 else
2471 btrfs_backref_unlock_node_buffer(upper);
2472 if (ret)
2473 break;
2474 }
2475
2476 if (!ret && node->pending) {
2477 btrfs_backref_drop_node_buffer(node);
2478 list_move_tail(&node->list, &rc->backref_cache.changed);
2479 node->pending = 0;
2480 }
2481
2482 path->lowest_level = 0;
2483
2484 /*
2485 * We should have allocated all of our space in the block rsv and thus
2486 * shouldn't ENOSPC.
2487 */
2488 ASSERT(ret != -ENOSPC);
2489 return ret;
2490 }
2491
link_to_upper(struct btrfs_trans_handle * trans,struct reloc_control * rc,struct btrfs_backref_node * node,struct btrfs_path * path)2492 static int link_to_upper(struct btrfs_trans_handle *trans,
2493 struct reloc_control *rc,
2494 struct btrfs_backref_node *node,
2495 struct btrfs_path *path)
2496 {
2497 struct btrfs_key key;
2498
2499 btrfs_node_key_to_cpu(node->eb, &key, 0);
2500 return do_relocation(trans, rc, node, &key, path, 0);
2501 }
2502
finish_pending_nodes(struct btrfs_trans_handle * trans,struct reloc_control * rc,struct btrfs_path * path,int err)2503 static int finish_pending_nodes(struct btrfs_trans_handle *trans,
2504 struct reloc_control *rc,
2505 struct btrfs_path *path, int err)
2506 {
2507 LIST_HEAD(list);
2508 struct btrfs_backref_cache *cache = &rc->backref_cache;
2509 struct btrfs_backref_node *node;
2510 int level;
2511 int ret;
2512
2513 for (level = 0; level < BTRFS_MAX_LEVEL; level++) {
2514 while (!list_empty(&cache->pending[level])) {
2515 node = list_entry(cache->pending[level].next,
2516 struct btrfs_backref_node, list);
2517 list_move_tail(&node->list, &list);
2518 BUG_ON(!node->pending);
2519
2520 if (!err) {
2521 ret = link_to_upper(trans, rc, node, path);
2522 if (ret < 0)
2523 err = ret;
2524 }
2525 }
2526 list_splice_init(&list, &cache->pending[level]);
2527 }
2528 return err;
2529 }
2530
2531 /*
2532 * mark a block and all blocks directly/indirectly reference the block
2533 * as processed.
2534 */
update_processed_blocks(struct reloc_control * rc,struct btrfs_backref_node * node)2535 static void update_processed_blocks(struct reloc_control *rc,
2536 struct btrfs_backref_node *node)
2537 {
2538 struct btrfs_backref_node *next = node;
2539 struct btrfs_backref_edge *edge;
2540 struct btrfs_backref_edge *edges[BTRFS_MAX_LEVEL - 1];
2541 int index = 0;
2542
2543 while (next) {
2544 cond_resched();
2545 while (1) {
2546 if (next->processed)
2547 break;
2548
2549 mark_block_processed(rc, next);
2550
2551 if (list_empty(&next->upper))
2552 break;
2553
2554 edge = list_entry(next->upper.next,
2555 struct btrfs_backref_edge, list[LOWER]);
2556 edges[index++] = edge;
2557 next = edge->node[UPPER];
2558 }
2559 next = walk_down_backref(edges, &index);
2560 }
2561 }
2562
tree_block_processed(u64 bytenr,struct reloc_control * rc)2563 static int tree_block_processed(u64 bytenr, struct reloc_control *rc)
2564 {
2565 u32 blocksize = rc->extent_root->fs_info->nodesize;
2566
2567 if (test_range_bit(&rc->processed_blocks, bytenr,
2568 bytenr + blocksize - 1, EXTENT_DIRTY, 1, NULL))
2569 return 1;
2570 return 0;
2571 }
2572
get_tree_block_key(struct btrfs_fs_info * fs_info,struct tree_block * block)2573 static int get_tree_block_key(struct btrfs_fs_info *fs_info,
2574 struct tree_block *block)
2575 {
2576 struct btrfs_tree_parent_check check = {
2577 .level = block->level,
2578 .owner_root = block->owner,
2579 .transid = block->key.offset
2580 };
2581 struct extent_buffer *eb;
2582
2583 eb = read_tree_block(fs_info, block->bytenr, &check);
2584 if (IS_ERR(eb))
2585 return PTR_ERR(eb);
2586 if (!extent_buffer_uptodate(eb)) {
2587 free_extent_buffer(eb);
2588 return -EIO;
2589 }
2590 if (block->level == 0)
2591 btrfs_item_key_to_cpu(eb, &block->key, 0);
2592 else
2593 btrfs_node_key_to_cpu(eb, &block->key, 0);
2594 free_extent_buffer(eb);
2595 block->key_ready = 1;
2596 return 0;
2597 }
2598
2599 /*
2600 * helper function to relocate a tree block
2601 */
relocate_tree_block(struct btrfs_trans_handle * trans,struct reloc_control * rc,struct btrfs_backref_node * node,struct btrfs_key * key,struct btrfs_path * path)2602 static int relocate_tree_block(struct btrfs_trans_handle *trans,
2603 struct reloc_control *rc,
2604 struct btrfs_backref_node *node,
2605 struct btrfs_key *key,
2606 struct btrfs_path *path)
2607 {
2608 struct btrfs_root *root;
2609 int ret = 0;
2610
2611 if (!node)
2612 return 0;
2613
2614 /*
2615 * If we fail here we want to drop our backref_node because we are going
2616 * to start over and regenerate the tree for it.
2617 */
2618 ret = reserve_metadata_space(trans, rc, node);
2619 if (ret)
2620 goto out;
2621
2622 BUG_ON(node->processed);
2623 root = select_one_root(node);
2624 if (IS_ERR(root)) {
2625 ret = PTR_ERR(root);
2626
2627 /* See explanation in select_one_root for the -EUCLEAN case. */
2628 ASSERT(ret == -ENOENT);
2629 if (ret == -ENOENT) {
2630 ret = 0;
2631 update_processed_blocks(rc, node);
2632 }
2633 goto out;
2634 }
2635
2636 if (root) {
2637 if (test_bit(BTRFS_ROOT_SHAREABLE, &root->state)) {
2638 /*
2639 * This block was the root block of a root, and this is
2640 * the first time we're processing the block and thus it
2641 * should not have had the ->new_bytenr modified and
2642 * should have not been included on the changed list.
2643 *
2644 * However in the case of corruption we could have
2645 * multiple refs pointing to the same block improperly,
2646 * and thus we would trip over these checks. ASSERT()
2647 * for the developer case, because it could indicate a
2648 * bug in the backref code, however error out for a
2649 * normal user in the case of corruption.
2650 */
2651 ASSERT(node->new_bytenr == 0);
2652 ASSERT(list_empty(&node->list));
2653 if (node->new_bytenr || !list_empty(&node->list)) {
2654 btrfs_err(root->fs_info,
2655 "bytenr %llu has improper references to it",
2656 node->bytenr);
2657 ret = -EUCLEAN;
2658 goto out;
2659 }
2660 ret = btrfs_record_root_in_trans(trans, root);
2661 if (ret)
2662 goto out;
2663 /*
2664 * Another thread could have failed, need to check if we
2665 * have reloc_root actually set.
2666 */
2667 if (!root->reloc_root) {
2668 ret = -ENOENT;
2669 goto out;
2670 }
2671 root = root->reloc_root;
2672 node->new_bytenr = root->node->start;
2673 btrfs_put_root(node->root);
2674 node->root = btrfs_grab_root(root);
2675 ASSERT(node->root);
2676 list_add_tail(&node->list, &rc->backref_cache.changed);
2677 } else {
2678 path->lowest_level = node->level;
2679 if (root == root->fs_info->chunk_root)
2680 btrfs_reserve_chunk_metadata(trans, false);
2681 ret = btrfs_search_slot(trans, root, key, path, 0, 1);
2682 btrfs_release_path(path);
2683 if (root == root->fs_info->chunk_root)
2684 btrfs_trans_release_chunk_metadata(trans);
2685 if (ret > 0)
2686 ret = 0;
2687 }
2688 if (!ret)
2689 update_processed_blocks(rc, node);
2690 } else {
2691 ret = do_relocation(trans, rc, node, key, path, 1);
2692 }
2693 out:
2694 if (ret || node->level == 0 || node->cowonly)
2695 btrfs_backref_cleanup_node(&rc->backref_cache, node);
2696 return ret;
2697 }
2698
2699 /*
2700 * relocate a list of blocks
2701 */
2702 static noinline_for_stack
relocate_tree_blocks(struct btrfs_trans_handle * trans,struct reloc_control * rc,struct rb_root * blocks)2703 int relocate_tree_blocks(struct btrfs_trans_handle *trans,
2704 struct reloc_control *rc, struct rb_root *blocks)
2705 {
2706 struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
2707 struct btrfs_backref_node *node;
2708 struct btrfs_path *path;
2709 struct tree_block *block;
2710 struct tree_block *next;
2711 int ret;
2712 int err = 0;
2713
2714 path = btrfs_alloc_path();
2715 if (!path) {
2716 err = -ENOMEM;
2717 goto out_free_blocks;
2718 }
2719
2720 /* Kick in readahead for tree blocks with missing keys */
2721 rbtree_postorder_for_each_entry_safe(block, next, blocks, rb_node) {
2722 if (!block->key_ready)
2723 btrfs_readahead_tree_block(fs_info, block->bytenr,
2724 block->owner, 0,
2725 block->level);
2726 }
2727
2728 /* Get first keys */
2729 rbtree_postorder_for_each_entry_safe(block, next, blocks, rb_node) {
2730 if (!block->key_ready) {
2731 err = get_tree_block_key(fs_info, block);
2732 if (err)
2733 goto out_free_path;
2734 }
2735 }
2736
2737 /* Do tree relocation */
2738 rbtree_postorder_for_each_entry_safe(block, next, blocks, rb_node) {
2739 node = build_backref_tree(trans, rc, &block->key,
2740 block->level, block->bytenr);
2741 if (IS_ERR(node)) {
2742 err = PTR_ERR(node);
2743 goto out;
2744 }
2745
2746 ret = relocate_tree_block(trans, rc, node, &block->key,
2747 path);
2748 if (ret < 0) {
2749 err = ret;
2750 break;
2751 }
2752 }
2753 out:
2754 err = finish_pending_nodes(trans, rc, path, err);
2755
2756 out_free_path:
2757 btrfs_free_path(path);
2758 out_free_blocks:
2759 free_block_list(blocks);
2760 return err;
2761 }
2762
prealloc_file_extent_cluster(struct btrfs_inode * inode,const struct file_extent_cluster * cluster)2763 static noinline_for_stack int prealloc_file_extent_cluster(
2764 struct btrfs_inode *inode,
2765 const struct file_extent_cluster *cluster)
2766 {
2767 u64 alloc_hint = 0;
2768 u64 start;
2769 u64 end;
2770 u64 offset = inode->index_cnt;
2771 u64 num_bytes;
2772 int nr;
2773 int ret = 0;
2774 u64 i_size = i_size_read(&inode->vfs_inode);
2775 u64 prealloc_start = cluster->start - offset;
2776 u64 prealloc_end = cluster->end - offset;
2777 u64 cur_offset = prealloc_start;
2778
2779 /*
2780 * For subpage case, previous i_size may not be aligned to PAGE_SIZE.
2781 * This means the range [i_size, PAGE_END + 1) is filled with zeros by
2782 * btrfs_do_readpage() call of previously relocated file cluster.
2783 *
2784 * If the current cluster starts in the above range, btrfs_do_readpage()
2785 * will skip the read, and relocate_one_page() will later writeback
2786 * the padding zeros as new data, causing data corruption.
2787 *
2788 * Here we have to manually invalidate the range (i_size, PAGE_END + 1).
2789 */
2790 if (!PAGE_ALIGNED(i_size)) {
2791 struct address_space *mapping = inode->vfs_inode.i_mapping;
2792 struct btrfs_fs_info *fs_info = inode->root->fs_info;
2793 const u32 sectorsize = fs_info->sectorsize;
2794 struct page *page;
2795
2796 ASSERT(sectorsize < PAGE_SIZE);
2797 ASSERT(IS_ALIGNED(i_size, sectorsize));
2798
2799 /*
2800 * Subpage can't handle page with DIRTY but without UPTODATE
2801 * bit as it can lead to the following deadlock:
2802 *
2803 * btrfs_read_folio()
2804 * | Page already *locked*
2805 * |- btrfs_lock_and_flush_ordered_range()
2806 * |- btrfs_start_ordered_extent()
2807 * |- extent_write_cache_pages()
2808 * |- lock_page()
2809 * We try to lock the page we already hold.
2810 *
2811 * Here we just writeback the whole data reloc inode, so that
2812 * we will be ensured to have no dirty range in the page, and
2813 * are safe to clear the uptodate bits.
2814 *
2815 * This shouldn't cause too much overhead, as we need to write
2816 * the data back anyway.
2817 */
2818 ret = filemap_write_and_wait(mapping);
2819 if (ret < 0)
2820 return ret;
2821
2822 clear_extent_bits(&inode->io_tree, i_size,
2823 round_up(i_size, PAGE_SIZE) - 1,
2824 EXTENT_UPTODATE);
2825 page = find_lock_page(mapping, i_size >> PAGE_SHIFT);
2826 /*
2827 * If page is freed we don't need to do anything then, as we
2828 * will re-read the whole page anyway.
2829 */
2830 if (page) {
2831 btrfs_subpage_clear_uptodate(fs_info, page, i_size,
2832 round_up(i_size, PAGE_SIZE) - i_size);
2833 unlock_page(page);
2834 put_page(page);
2835 }
2836 }
2837
2838 BUG_ON(cluster->start != cluster->boundary[0]);
2839 ret = btrfs_alloc_data_chunk_ondemand(inode,
2840 prealloc_end + 1 - prealloc_start);
2841 if (ret)
2842 return ret;
2843
2844 btrfs_inode_lock(inode, 0);
2845 for (nr = 0; nr < cluster->nr; nr++) {
2846 struct extent_state *cached_state = NULL;
2847
2848 start = cluster->boundary[nr] - offset;
2849 if (nr + 1 < cluster->nr)
2850 end = cluster->boundary[nr + 1] - 1 - offset;
2851 else
2852 end = cluster->end - offset;
2853
2854 lock_extent(&inode->io_tree, start, end, &cached_state);
2855 num_bytes = end + 1 - start;
2856 ret = btrfs_prealloc_file_range(&inode->vfs_inode, 0, start,
2857 num_bytes, num_bytes,
2858 end + 1, &alloc_hint);
2859 cur_offset = end + 1;
2860 unlock_extent(&inode->io_tree, start, end, &cached_state);
2861 if (ret)
2862 break;
2863 }
2864 btrfs_inode_unlock(inode, 0);
2865
2866 if (cur_offset < prealloc_end)
2867 btrfs_free_reserved_data_space_noquota(inode->root->fs_info,
2868 prealloc_end + 1 - cur_offset);
2869 return ret;
2870 }
2871
setup_relocation_extent_mapping(struct inode * inode,u64 start,u64 end,u64 block_start)2872 static noinline_for_stack int setup_relocation_extent_mapping(struct inode *inode,
2873 u64 start, u64 end, u64 block_start)
2874 {
2875 struct extent_map *em;
2876 struct extent_state *cached_state = NULL;
2877 int ret = 0;
2878
2879 em = alloc_extent_map();
2880 if (!em)
2881 return -ENOMEM;
2882
2883 em->start = start;
2884 em->len = end + 1 - start;
2885 em->block_len = em->len;
2886 em->block_start = block_start;
2887 set_bit(EXTENT_FLAG_PINNED, &em->flags);
2888
2889 lock_extent(&BTRFS_I(inode)->io_tree, start, end, &cached_state);
2890 ret = btrfs_replace_extent_map_range(BTRFS_I(inode), em, false);
2891 unlock_extent(&BTRFS_I(inode)->io_tree, start, end, &cached_state);
2892 free_extent_map(em);
2893
2894 return ret;
2895 }
2896
2897 /*
2898 * Allow error injection to test balance/relocation cancellation
2899 */
btrfs_should_cancel_balance(const struct btrfs_fs_info * fs_info)2900 noinline int btrfs_should_cancel_balance(const struct btrfs_fs_info *fs_info)
2901 {
2902 return atomic_read(&fs_info->balance_cancel_req) ||
2903 atomic_read(&fs_info->reloc_cancel_req) ||
2904 fatal_signal_pending(current);
2905 }
2906 ALLOW_ERROR_INJECTION(btrfs_should_cancel_balance, TRUE);
2907
get_cluster_boundary_end(const struct file_extent_cluster * cluster,int cluster_nr)2908 static u64 get_cluster_boundary_end(const struct file_extent_cluster *cluster,
2909 int cluster_nr)
2910 {
2911 /* Last extent, use cluster end directly */
2912 if (cluster_nr >= cluster->nr - 1)
2913 return cluster->end;
2914
2915 /* Use next boundary start*/
2916 return cluster->boundary[cluster_nr + 1] - 1;
2917 }
2918
relocate_one_page(struct inode * inode,struct file_ra_state * ra,const struct file_extent_cluster * cluster,int * cluster_nr,unsigned long page_index)2919 static int relocate_one_page(struct inode *inode, struct file_ra_state *ra,
2920 const struct file_extent_cluster *cluster,
2921 int *cluster_nr, unsigned long page_index)
2922 {
2923 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2924 u64 offset = BTRFS_I(inode)->index_cnt;
2925 const unsigned long last_index = (cluster->end - offset) >> PAGE_SHIFT;
2926 gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping);
2927 struct page *page;
2928 u64 page_start;
2929 u64 page_end;
2930 u64 cur;
2931 int ret;
2932
2933 ASSERT(page_index <= last_index);
2934 page = find_lock_page(inode->i_mapping, page_index);
2935 if (!page) {
2936 page_cache_sync_readahead(inode->i_mapping, ra, NULL,
2937 page_index, last_index + 1 - page_index);
2938 page = find_or_create_page(inode->i_mapping, page_index, mask);
2939 if (!page)
2940 return -ENOMEM;
2941 }
2942
2943 if (PageReadahead(page))
2944 page_cache_async_readahead(inode->i_mapping, ra, NULL,
2945 page_folio(page), page_index,
2946 last_index + 1 - page_index);
2947
2948 if (!PageUptodate(page)) {
2949 btrfs_read_folio(NULL, page_folio(page));
2950 lock_page(page);
2951 if (!PageUptodate(page)) {
2952 ret = -EIO;
2953 goto release_page;
2954 }
2955 }
2956
2957 /*
2958 * We could have lost page private when we dropped the lock to read the
2959 * page above, make sure we set_page_extent_mapped here so we have any
2960 * of the subpage blocksize stuff we need in place.
2961 */
2962 ret = set_page_extent_mapped(page);
2963 if (ret < 0)
2964 goto release_page;
2965
2966 page_start = page_offset(page);
2967 page_end = page_start + PAGE_SIZE - 1;
2968
2969 /*
2970 * Start from the cluster, as for subpage case, the cluster can start
2971 * inside the page.
2972 */
2973 cur = max(page_start, cluster->boundary[*cluster_nr] - offset);
2974 while (cur <= page_end) {
2975 struct extent_state *cached_state = NULL;
2976 u64 extent_start = cluster->boundary[*cluster_nr] - offset;
2977 u64 extent_end = get_cluster_boundary_end(cluster,
2978 *cluster_nr) - offset;
2979 u64 clamped_start = max(page_start, extent_start);
2980 u64 clamped_end = min(page_end, extent_end);
2981 u32 clamped_len = clamped_end + 1 - clamped_start;
2982
2983 /* Reserve metadata for this range */
2984 ret = btrfs_delalloc_reserve_metadata(BTRFS_I(inode),
2985 clamped_len, clamped_len,
2986 false);
2987 if (ret)
2988 goto release_page;
2989
2990 /* Mark the range delalloc and dirty for later writeback */
2991 lock_extent(&BTRFS_I(inode)->io_tree, clamped_start, clamped_end,
2992 &cached_state);
2993 ret = btrfs_set_extent_delalloc(BTRFS_I(inode), clamped_start,
2994 clamped_end, 0, &cached_state);
2995 if (ret) {
2996 clear_extent_bit(&BTRFS_I(inode)->io_tree,
2997 clamped_start, clamped_end,
2998 EXTENT_LOCKED | EXTENT_BOUNDARY,
2999 &cached_state);
3000 btrfs_delalloc_release_metadata(BTRFS_I(inode),
3001 clamped_len, true);
3002 btrfs_delalloc_release_extents(BTRFS_I(inode),
3003 clamped_len);
3004 goto release_page;
3005 }
3006 btrfs_page_set_dirty(fs_info, page, clamped_start, clamped_len);
3007
3008 /*
3009 * Set the boundary if it's inside the page.
3010 * Data relocation requires the destination extents to have the
3011 * same size as the source.
3012 * EXTENT_BOUNDARY bit prevents current extent from being merged
3013 * with previous extent.
3014 */
3015 if (in_range(cluster->boundary[*cluster_nr] - offset,
3016 page_start, PAGE_SIZE)) {
3017 u64 boundary_start = cluster->boundary[*cluster_nr] -
3018 offset;
3019 u64 boundary_end = boundary_start +
3020 fs_info->sectorsize - 1;
3021
3022 set_extent_bit(&BTRFS_I(inode)->io_tree,
3023 boundary_start, boundary_end,
3024 EXTENT_BOUNDARY, NULL);
3025 }
3026 unlock_extent(&BTRFS_I(inode)->io_tree, clamped_start, clamped_end,
3027 &cached_state);
3028 btrfs_delalloc_release_extents(BTRFS_I(inode), clamped_len);
3029 cur += clamped_len;
3030
3031 /* Crossed extent end, go to next extent */
3032 if (cur >= extent_end) {
3033 (*cluster_nr)++;
3034 /* Just finished the last extent of the cluster, exit. */
3035 if (*cluster_nr >= cluster->nr)
3036 break;
3037 }
3038 }
3039 unlock_page(page);
3040 put_page(page);
3041
3042 balance_dirty_pages_ratelimited(inode->i_mapping);
3043 btrfs_throttle(fs_info);
3044 if (btrfs_should_cancel_balance(fs_info))
3045 ret = -ECANCELED;
3046 return ret;
3047
3048 release_page:
3049 unlock_page(page);
3050 put_page(page);
3051 return ret;
3052 }
3053
relocate_file_extent_cluster(struct inode * inode,const struct file_extent_cluster * cluster)3054 static int relocate_file_extent_cluster(struct inode *inode,
3055 const struct file_extent_cluster *cluster)
3056 {
3057 u64 offset = BTRFS_I(inode)->index_cnt;
3058 unsigned long index;
3059 unsigned long last_index;
3060 struct file_ra_state *ra;
3061 int cluster_nr = 0;
3062 int ret = 0;
3063
3064 if (!cluster->nr)
3065 return 0;
3066
3067 ra = kzalloc(sizeof(*ra), GFP_NOFS);
3068 if (!ra)
3069 return -ENOMEM;
3070
3071 ret = prealloc_file_extent_cluster(BTRFS_I(inode), cluster);
3072 if (ret)
3073 goto out;
3074
3075 file_ra_state_init(ra, inode->i_mapping);
3076
3077 ret = setup_relocation_extent_mapping(inode, cluster->start - offset,
3078 cluster->end - offset, cluster->start);
3079 if (ret)
3080 goto out;
3081
3082 last_index = (cluster->end - offset) >> PAGE_SHIFT;
3083 for (index = (cluster->start - offset) >> PAGE_SHIFT;
3084 index <= last_index && !ret; index++)
3085 ret = relocate_one_page(inode, ra, cluster, &cluster_nr, index);
3086 if (ret == 0)
3087 WARN_ON(cluster_nr != cluster->nr);
3088 out:
3089 kfree(ra);
3090 return ret;
3091 }
3092
relocate_data_extent(struct inode * inode,const struct btrfs_key * extent_key,struct file_extent_cluster * cluster)3093 static noinline_for_stack int relocate_data_extent(struct inode *inode,
3094 const struct btrfs_key *extent_key,
3095 struct file_extent_cluster *cluster)
3096 {
3097 int ret;
3098
3099 if (cluster->nr > 0 && extent_key->objectid != cluster->end + 1) {
3100 ret = relocate_file_extent_cluster(inode, cluster);
3101 if (ret)
3102 return ret;
3103 cluster->nr = 0;
3104 }
3105
3106 if (!cluster->nr)
3107 cluster->start = extent_key->objectid;
3108 else
3109 BUG_ON(cluster->nr >= MAX_EXTENTS);
3110 cluster->end = extent_key->objectid + extent_key->offset - 1;
3111 cluster->boundary[cluster->nr] = extent_key->objectid;
3112 cluster->nr++;
3113
3114 if (cluster->nr >= MAX_EXTENTS) {
3115 ret = relocate_file_extent_cluster(inode, cluster);
3116 if (ret)
3117 return ret;
3118 cluster->nr = 0;
3119 }
3120 return 0;
3121 }
3122
3123 /*
3124 * helper to add a tree block to the list.
3125 * the major work is getting the generation and level of the block
3126 */
add_tree_block(struct reloc_control * rc,const struct btrfs_key * extent_key,struct btrfs_path * path,struct rb_root * blocks)3127 static int add_tree_block(struct reloc_control *rc,
3128 const struct btrfs_key *extent_key,
3129 struct btrfs_path *path,
3130 struct rb_root *blocks)
3131 {
3132 struct extent_buffer *eb;
3133 struct btrfs_extent_item *ei;
3134 struct btrfs_tree_block_info *bi;
3135 struct tree_block *block;
3136 struct rb_node *rb_node;
3137 u32 item_size;
3138 int level = -1;
3139 u64 generation;
3140 u64 owner = 0;
3141
3142 eb = path->nodes[0];
3143 item_size = btrfs_item_size(eb, path->slots[0]);
3144
3145 if (extent_key->type == BTRFS_METADATA_ITEM_KEY ||
3146 item_size >= sizeof(*ei) + sizeof(*bi)) {
3147 unsigned long ptr = 0, end;
3148
3149 ei = btrfs_item_ptr(eb, path->slots[0],
3150 struct btrfs_extent_item);
3151 end = (unsigned long)ei + item_size;
3152 if (extent_key->type == BTRFS_EXTENT_ITEM_KEY) {
3153 bi = (struct btrfs_tree_block_info *)(ei + 1);
3154 level = btrfs_tree_block_level(eb, bi);
3155 ptr = (unsigned long)(bi + 1);
3156 } else {
3157 level = (int)extent_key->offset;
3158 ptr = (unsigned long)(ei + 1);
3159 }
3160 generation = btrfs_extent_generation(eb, ei);
3161
3162 /*
3163 * We're reading random blocks without knowing their owner ahead
3164 * of time. This is ok most of the time, as all reloc roots and
3165 * fs roots have the same lock type. However normal trees do
3166 * not, and the only way to know ahead of time is to read the
3167 * inline ref offset. We know it's an fs root if
3168 *
3169 * 1. There's more than one ref.
3170 * 2. There's a SHARED_DATA_REF_KEY set.
3171 * 3. FULL_BACKREF is set on the flags.
3172 *
3173 * Otherwise it's safe to assume that the ref offset == the
3174 * owner of this block, so we can use that when calling
3175 * read_tree_block.
3176 */
3177 if (btrfs_extent_refs(eb, ei) == 1 &&
3178 !(btrfs_extent_flags(eb, ei) &
3179 BTRFS_BLOCK_FLAG_FULL_BACKREF) &&
3180 ptr < end) {
3181 struct btrfs_extent_inline_ref *iref;
3182 int type;
3183
3184 iref = (struct btrfs_extent_inline_ref *)ptr;
3185 type = btrfs_get_extent_inline_ref_type(eb, iref,
3186 BTRFS_REF_TYPE_BLOCK);
3187 if (type == BTRFS_REF_TYPE_INVALID)
3188 return -EINVAL;
3189 if (type == BTRFS_TREE_BLOCK_REF_KEY)
3190 owner = btrfs_extent_inline_ref_offset(eb, iref);
3191 }
3192 } else {
3193 btrfs_print_leaf(eb);
3194 btrfs_err(rc->block_group->fs_info,
3195 "unrecognized tree backref at tree block %llu slot %u",
3196 eb->start, path->slots[0]);
3197 btrfs_release_path(path);
3198 return -EUCLEAN;
3199 }
3200
3201 btrfs_release_path(path);
3202
3203 BUG_ON(level == -1);
3204
3205 block = kmalloc(sizeof(*block), GFP_NOFS);
3206 if (!block)
3207 return -ENOMEM;
3208
3209 block->bytenr = extent_key->objectid;
3210 block->key.objectid = rc->extent_root->fs_info->nodesize;
3211 block->key.offset = generation;
3212 block->level = level;
3213 block->key_ready = 0;
3214 block->owner = owner;
3215
3216 rb_node = rb_simple_insert(blocks, block->bytenr, &block->rb_node);
3217 if (rb_node)
3218 btrfs_backref_panic(rc->extent_root->fs_info, block->bytenr,
3219 -EEXIST);
3220
3221 return 0;
3222 }
3223
3224 /*
3225 * helper to add tree blocks for backref of type BTRFS_SHARED_DATA_REF_KEY
3226 */
__add_tree_block(struct reloc_control * rc,u64 bytenr,u32 blocksize,struct rb_root * blocks)3227 static int __add_tree_block(struct reloc_control *rc,
3228 u64 bytenr, u32 blocksize,
3229 struct rb_root *blocks)
3230 {
3231 struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
3232 struct btrfs_path *path;
3233 struct btrfs_key key;
3234 int ret;
3235 bool skinny = btrfs_fs_incompat(fs_info, SKINNY_METADATA);
3236
3237 if (tree_block_processed(bytenr, rc))
3238 return 0;
3239
3240 if (rb_simple_search(blocks, bytenr))
3241 return 0;
3242
3243 path = btrfs_alloc_path();
3244 if (!path)
3245 return -ENOMEM;
3246 again:
3247 key.objectid = bytenr;
3248 if (skinny) {
3249 key.type = BTRFS_METADATA_ITEM_KEY;
3250 key.offset = (u64)-1;
3251 } else {
3252 key.type = BTRFS_EXTENT_ITEM_KEY;
3253 key.offset = blocksize;
3254 }
3255
3256 path->search_commit_root = 1;
3257 path->skip_locking = 1;
3258 ret = btrfs_search_slot(NULL, rc->extent_root, &key, path, 0, 0);
3259 if (ret < 0)
3260 goto out;
3261
3262 if (ret > 0 && skinny) {
3263 if (path->slots[0]) {
3264 path->slots[0]--;
3265 btrfs_item_key_to_cpu(path->nodes[0], &key,
3266 path->slots[0]);
3267 if (key.objectid == bytenr &&
3268 (key.type == BTRFS_METADATA_ITEM_KEY ||
3269 (key.type == BTRFS_EXTENT_ITEM_KEY &&
3270 key.offset == blocksize)))
3271 ret = 0;
3272 }
3273
3274 if (ret) {
3275 skinny = false;
3276 btrfs_release_path(path);
3277 goto again;
3278 }
3279 }
3280 if (ret) {
3281 ASSERT(ret == 1);
3282 btrfs_print_leaf(path->nodes[0]);
3283 btrfs_err(fs_info,
3284 "tree block extent item (%llu) is not found in extent tree",
3285 bytenr);
3286 WARN_ON(1);
3287 ret = -EINVAL;
3288 goto out;
3289 }
3290
3291 ret = add_tree_block(rc, &key, path, blocks);
3292 out:
3293 btrfs_free_path(path);
3294 return ret;
3295 }
3296
delete_block_group_cache(struct btrfs_fs_info * fs_info,struct btrfs_block_group * block_group,struct inode * inode,u64 ino)3297 static int delete_block_group_cache(struct btrfs_fs_info *fs_info,
3298 struct btrfs_block_group *block_group,
3299 struct inode *inode,
3300 u64 ino)
3301 {
3302 struct btrfs_root *root = fs_info->tree_root;
3303 struct btrfs_trans_handle *trans;
3304 int ret = 0;
3305
3306 if (inode)
3307 goto truncate;
3308
3309 inode = btrfs_iget(fs_info->sb, ino, root);
3310 if (IS_ERR(inode))
3311 return -ENOENT;
3312
3313 truncate:
3314 ret = btrfs_check_trunc_cache_free_space(fs_info,
3315 &fs_info->global_block_rsv);
3316 if (ret)
3317 goto out;
3318
3319 trans = btrfs_join_transaction(root);
3320 if (IS_ERR(trans)) {
3321 ret = PTR_ERR(trans);
3322 goto out;
3323 }
3324
3325 ret = btrfs_truncate_free_space_cache(trans, block_group, inode);
3326
3327 btrfs_end_transaction(trans);
3328 btrfs_btree_balance_dirty(fs_info);
3329 out:
3330 iput(inode);
3331 return ret;
3332 }
3333
3334 /*
3335 * Locate the free space cache EXTENT_DATA in root tree leaf and delete the
3336 * cache inode, to avoid free space cache data extent blocking data relocation.
3337 */
delete_v1_space_cache(struct extent_buffer * leaf,struct btrfs_block_group * block_group,u64 data_bytenr)3338 static int delete_v1_space_cache(struct extent_buffer *leaf,
3339 struct btrfs_block_group *block_group,
3340 u64 data_bytenr)
3341 {
3342 u64 space_cache_ino;
3343 struct btrfs_file_extent_item *ei;
3344 struct btrfs_key key;
3345 bool found = false;
3346 int i;
3347 int ret;
3348
3349 if (btrfs_header_owner(leaf) != BTRFS_ROOT_TREE_OBJECTID)
3350 return 0;
3351
3352 for (i = 0; i < btrfs_header_nritems(leaf); i++) {
3353 u8 type;
3354
3355 btrfs_item_key_to_cpu(leaf, &key, i);
3356 if (key.type != BTRFS_EXTENT_DATA_KEY)
3357 continue;
3358 ei = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
3359 type = btrfs_file_extent_type(leaf, ei);
3360
3361 if ((type == BTRFS_FILE_EXTENT_REG ||
3362 type == BTRFS_FILE_EXTENT_PREALLOC) &&
3363 btrfs_file_extent_disk_bytenr(leaf, ei) == data_bytenr) {
3364 found = true;
3365 space_cache_ino = key.objectid;
3366 break;
3367 }
3368 }
3369 if (!found)
3370 return -ENOENT;
3371 ret = delete_block_group_cache(leaf->fs_info, block_group, NULL,
3372 space_cache_ino);
3373 return ret;
3374 }
3375
3376 /*
3377 * helper to find all tree blocks that reference a given data extent
3378 */
add_data_references(struct reloc_control * rc,const struct btrfs_key * extent_key,struct btrfs_path * path,struct rb_root * blocks)3379 static noinline_for_stack int add_data_references(struct reloc_control *rc,
3380 const struct btrfs_key *extent_key,
3381 struct btrfs_path *path,
3382 struct rb_root *blocks)
3383 {
3384 struct btrfs_backref_walk_ctx ctx = { 0 };
3385 struct ulist_iterator leaf_uiter;
3386 struct ulist_node *ref_node = NULL;
3387 const u32 blocksize = rc->extent_root->fs_info->nodesize;
3388 int ret = 0;
3389
3390 btrfs_release_path(path);
3391
3392 ctx.bytenr = extent_key->objectid;
3393 ctx.skip_inode_ref_list = true;
3394 ctx.fs_info = rc->extent_root->fs_info;
3395
3396 ret = btrfs_find_all_leafs(&ctx);
3397 if (ret < 0)
3398 return ret;
3399
3400 ULIST_ITER_INIT(&leaf_uiter);
3401 while ((ref_node = ulist_next(ctx.refs, &leaf_uiter))) {
3402 struct btrfs_tree_parent_check check = { 0 };
3403 struct extent_buffer *eb;
3404
3405 eb = read_tree_block(ctx.fs_info, ref_node->val, &check);
3406 if (IS_ERR(eb)) {
3407 ret = PTR_ERR(eb);
3408 break;
3409 }
3410 ret = delete_v1_space_cache(eb, rc->block_group,
3411 extent_key->objectid);
3412 free_extent_buffer(eb);
3413 if (ret < 0)
3414 break;
3415 ret = __add_tree_block(rc, ref_node->val, blocksize, blocks);
3416 if (ret < 0)
3417 break;
3418 }
3419 if (ret < 0)
3420 free_block_list(blocks);
3421 ulist_free(ctx.refs);
3422 return ret;
3423 }
3424
3425 /*
3426 * helper to find next unprocessed extent
3427 */
3428 static noinline_for_stack
find_next_extent(struct reloc_control * rc,struct btrfs_path * path,struct btrfs_key * extent_key)3429 int find_next_extent(struct reloc_control *rc, struct btrfs_path *path,
3430 struct btrfs_key *extent_key)
3431 {
3432 struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
3433 struct btrfs_key key;
3434 struct extent_buffer *leaf;
3435 u64 start, end, last;
3436 int ret;
3437
3438 last = rc->block_group->start + rc->block_group->length;
3439 while (1) {
3440 bool block_found;
3441
3442 cond_resched();
3443 if (rc->search_start >= last) {
3444 ret = 1;
3445 break;
3446 }
3447
3448 key.objectid = rc->search_start;
3449 key.type = BTRFS_EXTENT_ITEM_KEY;
3450 key.offset = 0;
3451
3452 path->search_commit_root = 1;
3453 path->skip_locking = 1;
3454 ret = btrfs_search_slot(NULL, rc->extent_root, &key, path,
3455 0, 0);
3456 if (ret < 0)
3457 break;
3458 next:
3459 leaf = path->nodes[0];
3460 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
3461 ret = btrfs_next_leaf(rc->extent_root, path);
3462 if (ret != 0)
3463 break;
3464 leaf = path->nodes[0];
3465 }
3466
3467 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3468 if (key.objectid >= last) {
3469 ret = 1;
3470 break;
3471 }
3472
3473 if (key.type != BTRFS_EXTENT_ITEM_KEY &&
3474 key.type != BTRFS_METADATA_ITEM_KEY) {
3475 path->slots[0]++;
3476 goto next;
3477 }
3478
3479 if (key.type == BTRFS_EXTENT_ITEM_KEY &&
3480 key.objectid + key.offset <= rc->search_start) {
3481 path->slots[0]++;
3482 goto next;
3483 }
3484
3485 if (key.type == BTRFS_METADATA_ITEM_KEY &&
3486 key.objectid + fs_info->nodesize <=
3487 rc->search_start) {
3488 path->slots[0]++;
3489 goto next;
3490 }
3491
3492 block_found = find_first_extent_bit(&rc->processed_blocks,
3493 key.objectid, &start, &end,
3494 EXTENT_DIRTY, NULL);
3495
3496 if (block_found && start <= key.objectid) {
3497 btrfs_release_path(path);
3498 rc->search_start = end + 1;
3499 } else {
3500 if (key.type == BTRFS_EXTENT_ITEM_KEY)
3501 rc->search_start = key.objectid + key.offset;
3502 else
3503 rc->search_start = key.objectid +
3504 fs_info->nodesize;
3505 memcpy(extent_key, &key, sizeof(key));
3506 return 0;
3507 }
3508 }
3509 btrfs_release_path(path);
3510 return ret;
3511 }
3512
set_reloc_control(struct reloc_control * rc)3513 static void set_reloc_control(struct reloc_control *rc)
3514 {
3515 struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
3516
3517 mutex_lock(&fs_info->reloc_mutex);
3518 fs_info->reloc_ctl = rc;
3519 mutex_unlock(&fs_info->reloc_mutex);
3520 }
3521
unset_reloc_control(struct reloc_control * rc)3522 static void unset_reloc_control(struct reloc_control *rc)
3523 {
3524 struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
3525
3526 mutex_lock(&fs_info->reloc_mutex);
3527 fs_info->reloc_ctl = NULL;
3528 mutex_unlock(&fs_info->reloc_mutex);
3529 }
3530
3531 static noinline_for_stack
prepare_to_relocate(struct reloc_control * rc)3532 int prepare_to_relocate(struct reloc_control *rc)
3533 {
3534 struct btrfs_trans_handle *trans;
3535 int ret;
3536
3537 rc->block_rsv = btrfs_alloc_block_rsv(rc->extent_root->fs_info,
3538 BTRFS_BLOCK_RSV_TEMP);
3539 if (!rc->block_rsv)
3540 return -ENOMEM;
3541
3542 memset(&rc->cluster, 0, sizeof(rc->cluster));
3543 rc->search_start = rc->block_group->start;
3544 rc->extents_found = 0;
3545 rc->nodes_relocated = 0;
3546 rc->merging_rsv_size = 0;
3547 rc->reserved_bytes = 0;
3548 rc->block_rsv->size = rc->extent_root->fs_info->nodesize *
3549 RELOCATION_RESERVED_NODES;
3550 ret = btrfs_block_rsv_refill(rc->extent_root->fs_info,
3551 rc->block_rsv, rc->block_rsv->size,
3552 BTRFS_RESERVE_FLUSH_ALL);
3553 if (ret)
3554 return ret;
3555
3556 rc->create_reloc_tree = 1;
3557 set_reloc_control(rc);
3558
3559 trans = btrfs_join_transaction(rc->extent_root);
3560 if (IS_ERR(trans)) {
3561 unset_reloc_control(rc);
3562 /*
3563 * extent tree is not a ref_cow tree and has no reloc_root to
3564 * cleanup. And callers are responsible to free the above
3565 * block rsv.
3566 */
3567 return PTR_ERR(trans);
3568 }
3569
3570 ret = btrfs_commit_transaction(trans);
3571 if (ret)
3572 unset_reloc_control(rc);
3573
3574 return ret;
3575 }
3576
relocate_block_group(struct reloc_control * rc)3577 static noinline_for_stack int relocate_block_group(struct reloc_control *rc)
3578 {
3579 struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
3580 struct rb_root blocks = RB_ROOT;
3581 struct btrfs_key key;
3582 struct btrfs_trans_handle *trans = NULL;
3583 struct btrfs_path *path;
3584 struct btrfs_extent_item *ei;
3585 u64 flags;
3586 int ret;
3587 int err = 0;
3588 int progress = 0;
3589
3590 path = btrfs_alloc_path();
3591 if (!path)
3592 return -ENOMEM;
3593 path->reada = READA_FORWARD;
3594
3595 ret = prepare_to_relocate(rc);
3596 if (ret) {
3597 err = ret;
3598 goto out_free;
3599 }
3600
3601 while (1) {
3602 rc->reserved_bytes = 0;
3603 ret = btrfs_block_rsv_refill(fs_info, rc->block_rsv,
3604 rc->block_rsv->size,
3605 BTRFS_RESERVE_FLUSH_ALL);
3606 if (ret) {
3607 err = ret;
3608 break;
3609 }
3610 progress++;
3611 trans = btrfs_start_transaction(rc->extent_root, 0);
3612 if (IS_ERR(trans)) {
3613 err = PTR_ERR(trans);
3614 trans = NULL;
3615 break;
3616 }
3617 restart:
3618 if (rc->backref_cache.last_trans != trans->transid)
3619 btrfs_backref_release_cache(&rc->backref_cache);
3620 rc->backref_cache.last_trans = trans->transid;
3621
3622 ret = find_next_extent(rc, path, &key);
3623 if (ret < 0)
3624 err = ret;
3625 if (ret != 0)
3626 break;
3627
3628 rc->extents_found++;
3629
3630 ei = btrfs_item_ptr(path->nodes[0], path->slots[0],
3631 struct btrfs_extent_item);
3632 flags = btrfs_extent_flags(path->nodes[0], ei);
3633
3634 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
3635 ret = add_tree_block(rc, &key, path, &blocks);
3636 } else if (rc->stage == UPDATE_DATA_PTRS &&
3637 (flags & BTRFS_EXTENT_FLAG_DATA)) {
3638 ret = add_data_references(rc, &key, path, &blocks);
3639 } else {
3640 btrfs_release_path(path);
3641 ret = 0;
3642 }
3643 if (ret < 0) {
3644 err = ret;
3645 break;
3646 }
3647
3648 if (!RB_EMPTY_ROOT(&blocks)) {
3649 ret = relocate_tree_blocks(trans, rc, &blocks);
3650 if (ret < 0) {
3651 if (ret != -EAGAIN) {
3652 err = ret;
3653 break;
3654 }
3655 rc->extents_found--;
3656 rc->search_start = key.objectid;
3657 }
3658 }
3659
3660 btrfs_end_transaction_throttle(trans);
3661 btrfs_btree_balance_dirty(fs_info);
3662 trans = NULL;
3663
3664 if (rc->stage == MOVE_DATA_EXTENTS &&
3665 (flags & BTRFS_EXTENT_FLAG_DATA)) {
3666 rc->found_file_extent = 1;
3667 ret = relocate_data_extent(rc->data_inode,
3668 &key, &rc->cluster);
3669 if (ret < 0) {
3670 err = ret;
3671 break;
3672 }
3673 }
3674 if (btrfs_should_cancel_balance(fs_info)) {
3675 err = -ECANCELED;
3676 break;
3677 }
3678 }
3679 if (trans && progress && err == -ENOSPC) {
3680 ret = btrfs_force_chunk_alloc(trans, rc->block_group->flags);
3681 if (ret == 1) {
3682 err = 0;
3683 progress = 0;
3684 goto restart;
3685 }
3686 }
3687
3688 btrfs_release_path(path);
3689 clear_extent_bits(&rc->processed_blocks, 0, (u64)-1, EXTENT_DIRTY);
3690
3691 if (trans) {
3692 btrfs_end_transaction_throttle(trans);
3693 btrfs_btree_balance_dirty(fs_info);
3694 }
3695
3696 if (!err) {
3697 ret = relocate_file_extent_cluster(rc->data_inode,
3698 &rc->cluster);
3699 if (ret < 0)
3700 err = ret;
3701 }
3702
3703 rc->create_reloc_tree = 0;
3704 set_reloc_control(rc);
3705
3706 btrfs_backref_release_cache(&rc->backref_cache);
3707 btrfs_block_rsv_release(fs_info, rc->block_rsv, (u64)-1, NULL);
3708
3709 /*
3710 * Even in the case when the relocation is cancelled, we should all go
3711 * through prepare_to_merge() and merge_reloc_roots().
3712 *
3713 * For error (including cancelled balance), prepare_to_merge() will
3714 * mark all reloc trees orphan, then queue them for cleanup in
3715 * merge_reloc_roots()
3716 */
3717 err = prepare_to_merge(rc, err);
3718
3719 merge_reloc_roots(rc);
3720
3721 rc->merge_reloc_tree = 0;
3722 unset_reloc_control(rc);
3723 btrfs_block_rsv_release(fs_info, rc->block_rsv, (u64)-1, NULL);
3724
3725 /* get rid of pinned extents */
3726 trans = btrfs_join_transaction(rc->extent_root);
3727 if (IS_ERR(trans)) {
3728 err = PTR_ERR(trans);
3729 goto out_free;
3730 }
3731 ret = btrfs_commit_transaction(trans);
3732 if (ret && !err)
3733 err = ret;
3734 out_free:
3735 ret = clean_dirty_subvols(rc);
3736 if (ret < 0 && !err)
3737 err = ret;
3738 btrfs_free_block_rsv(fs_info, rc->block_rsv);
3739 btrfs_free_path(path);
3740 return err;
3741 }
3742
__insert_orphan_inode(struct btrfs_trans_handle * trans,struct btrfs_root * root,u64 objectid)3743 static int __insert_orphan_inode(struct btrfs_trans_handle *trans,
3744 struct btrfs_root *root, u64 objectid)
3745 {
3746 struct btrfs_path *path;
3747 struct btrfs_inode_item *item;
3748 struct extent_buffer *leaf;
3749 int ret;
3750
3751 path = btrfs_alloc_path();
3752 if (!path)
3753 return -ENOMEM;
3754
3755 ret = btrfs_insert_empty_inode(trans, root, path, objectid);
3756 if (ret)
3757 goto out;
3758
3759 leaf = path->nodes[0];
3760 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_inode_item);
3761 memzero_extent_buffer(leaf, (unsigned long)item, sizeof(*item));
3762 btrfs_set_inode_generation(leaf, item, 1);
3763 btrfs_set_inode_size(leaf, item, 0);
3764 btrfs_set_inode_mode(leaf, item, S_IFREG | 0600);
3765 btrfs_set_inode_flags(leaf, item, BTRFS_INODE_NOCOMPRESS |
3766 BTRFS_INODE_PREALLOC);
3767 btrfs_mark_buffer_dirty(trans, leaf);
3768 out:
3769 btrfs_free_path(path);
3770 return ret;
3771 }
3772
delete_orphan_inode(struct btrfs_trans_handle * trans,struct btrfs_root * root,u64 objectid)3773 static void delete_orphan_inode(struct btrfs_trans_handle *trans,
3774 struct btrfs_root *root, u64 objectid)
3775 {
3776 struct btrfs_path *path;
3777 struct btrfs_key key;
3778 int ret = 0;
3779
3780 path = btrfs_alloc_path();
3781 if (!path) {
3782 ret = -ENOMEM;
3783 goto out;
3784 }
3785
3786 key.objectid = objectid;
3787 key.type = BTRFS_INODE_ITEM_KEY;
3788 key.offset = 0;
3789 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3790 if (ret) {
3791 if (ret > 0)
3792 ret = -ENOENT;
3793 goto out;
3794 }
3795 ret = btrfs_del_item(trans, root, path);
3796 out:
3797 if (ret)
3798 btrfs_abort_transaction(trans, ret);
3799 btrfs_free_path(path);
3800 }
3801
3802 /*
3803 * helper to create inode for data relocation.
3804 * the inode is in data relocation tree and its link count is 0
3805 */
create_reloc_inode(struct btrfs_fs_info * fs_info,const struct btrfs_block_group * group)3806 static noinline_for_stack struct inode *create_reloc_inode(
3807 struct btrfs_fs_info *fs_info,
3808 const struct btrfs_block_group *group)
3809 {
3810 struct inode *inode = NULL;
3811 struct btrfs_trans_handle *trans;
3812 struct btrfs_root *root;
3813 u64 objectid;
3814 int err = 0;
3815
3816 root = btrfs_grab_root(fs_info->data_reloc_root);
3817 trans = btrfs_start_transaction(root, 6);
3818 if (IS_ERR(trans)) {
3819 btrfs_put_root(root);
3820 return ERR_CAST(trans);
3821 }
3822
3823 err = btrfs_get_free_objectid(root, &objectid);
3824 if (err)
3825 goto out;
3826
3827 err = __insert_orphan_inode(trans, root, objectid);
3828 if (err)
3829 goto out;
3830
3831 inode = btrfs_iget(fs_info->sb, objectid, root);
3832 if (IS_ERR(inode)) {
3833 delete_orphan_inode(trans, root, objectid);
3834 err = PTR_ERR(inode);
3835 inode = NULL;
3836 goto out;
3837 }
3838 BTRFS_I(inode)->index_cnt = group->start;
3839
3840 err = btrfs_orphan_add(trans, BTRFS_I(inode));
3841 out:
3842 btrfs_put_root(root);
3843 btrfs_end_transaction(trans);
3844 btrfs_btree_balance_dirty(fs_info);
3845 if (err) {
3846 iput(inode);
3847 inode = ERR_PTR(err);
3848 }
3849 return inode;
3850 }
3851
3852 /*
3853 * Mark start of chunk relocation that is cancellable. Check if the cancellation
3854 * has been requested meanwhile and don't start in that case.
3855 *
3856 * Return:
3857 * 0 success
3858 * -EINPROGRESS operation is already in progress, that's probably a bug
3859 * -ECANCELED cancellation request was set before the operation started
3860 */
reloc_chunk_start(struct btrfs_fs_info * fs_info)3861 static int reloc_chunk_start(struct btrfs_fs_info *fs_info)
3862 {
3863 if (test_and_set_bit(BTRFS_FS_RELOC_RUNNING, &fs_info->flags)) {
3864 /* This should not happen */
3865 btrfs_err(fs_info, "reloc already running, cannot start");
3866 return -EINPROGRESS;
3867 }
3868
3869 if (atomic_read(&fs_info->reloc_cancel_req) > 0) {
3870 btrfs_info(fs_info, "chunk relocation canceled on start");
3871 /*
3872 * On cancel, clear all requests but let the caller mark
3873 * the end after cleanup operations.
3874 */
3875 atomic_set(&fs_info->reloc_cancel_req, 0);
3876 return -ECANCELED;
3877 }
3878 return 0;
3879 }
3880
3881 /*
3882 * Mark end of chunk relocation that is cancellable and wake any waiters.
3883 */
reloc_chunk_end(struct btrfs_fs_info * fs_info)3884 static void reloc_chunk_end(struct btrfs_fs_info *fs_info)
3885 {
3886 /* Requested after start, clear bit first so any waiters can continue */
3887 if (atomic_read(&fs_info->reloc_cancel_req) > 0)
3888 btrfs_info(fs_info, "chunk relocation canceled during operation");
3889 clear_and_wake_up_bit(BTRFS_FS_RELOC_RUNNING, &fs_info->flags);
3890 atomic_set(&fs_info->reloc_cancel_req, 0);
3891 }
3892
alloc_reloc_control(struct btrfs_fs_info * fs_info)3893 static struct reloc_control *alloc_reloc_control(struct btrfs_fs_info *fs_info)
3894 {
3895 struct reloc_control *rc;
3896
3897 rc = kzalloc(sizeof(*rc), GFP_NOFS);
3898 if (!rc)
3899 return NULL;
3900
3901 INIT_LIST_HEAD(&rc->reloc_roots);
3902 INIT_LIST_HEAD(&rc->dirty_subvol_roots);
3903 btrfs_backref_init_cache(fs_info, &rc->backref_cache, 1);
3904 mapping_tree_init(&rc->reloc_root_tree);
3905 extent_io_tree_init(fs_info, &rc->processed_blocks, IO_TREE_RELOC_BLOCKS);
3906 return rc;
3907 }
3908
free_reloc_control(struct reloc_control * rc)3909 static void free_reloc_control(struct reloc_control *rc)
3910 {
3911 struct mapping_node *node, *tmp;
3912
3913 free_reloc_roots(&rc->reloc_roots);
3914 rbtree_postorder_for_each_entry_safe(node, tmp,
3915 &rc->reloc_root_tree.rb_root, rb_node)
3916 kfree(node);
3917
3918 kfree(rc);
3919 }
3920
3921 /*
3922 * Print the block group being relocated
3923 */
describe_relocation(struct btrfs_fs_info * fs_info,struct btrfs_block_group * block_group)3924 static void describe_relocation(struct btrfs_fs_info *fs_info,
3925 struct btrfs_block_group *block_group)
3926 {
3927 char buf[128] = {'\0'};
3928
3929 btrfs_describe_block_groups(block_group->flags, buf, sizeof(buf));
3930
3931 btrfs_info(fs_info,
3932 "relocating block group %llu flags %s",
3933 block_group->start, buf);
3934 }
3935
stage_to_string(int stage)3936 static const char *stage_to_string(int stage)
3937 {
3938 if (stage == MOVE_DATA_EXTENTS)
3939 return "move data extents";
3940 if (stage == UPDATE_DATA_PTRS)
3941 return "update data pointers";
3942 return "unknown";
3943 }
3944
3945 /*
3946 * function to relocate all extents in a block group.
3947 */
btrfs_relocate_block_group(struct btrfs_fs_info * fs_info,u64 group_start)3948 int btrfs_relocate_block_group(struct btrfs_fs_info *fs_info, u64 group_start)
3949 {
3950 struct btrfs_block_group *bg;
3951 struct btrfs_root *extent_root = btrfs_extent_root(fs_info, group_start);
3952 struct reloc_control *rc;
3953 struct inode *inode;
3954 struct btrfs_path *path;
3955 int ret;
3956 int rw = 0;
3957 int err = 0;
3958
3959 /*
3960 * This only gets set if we had a half-deleted snapshot on mount. We
3961 * cannot allow relocation to start while we're still trying to clean up
3962 * these pending deletions.
3963 */
3964 ret = wait_on_bit(&fs_info->flags, BTRFS_FS_UNFINISHED_DROPS, TASK_INTERRUPTIBLE);
3965 if (ret)
3966 return ret;
3967
3968 /* We may have been woken up by close_ctree, so bail if we're closing. */
3969 if (btrfs_fs_closing(fs_info))
3970 return -EINTR;
3971
3972 bg = btrfs_lookup_block_group(fs_info, group_start);
3973 if (!bg)
3974 return -ENOENT;
3975
3976 /*
3977 * Relocation of a data block group creates ordered extents. Without
3978 * sb_start_write(), we can freeze the filesystem while unfinished
3979 * ordered extents are left. Such ordered extents can cause a deadlock
3980 * e.g. when syncfs() is waiting for their completion but they can't
3981 * finish because they block when joining a transaction, due to the
3982 * fact that the freeze locks are being held in write mode.
3983 */
3984 if (bg->flags & BTRFS_BLOCK_GROUP_DATA)
3985 ASSERT(sb_write_started(fs_info->sb));
3986
3987 if (btrfs_pinned_by_swapfile(fs_info, bg)) {
3988 btrfs_put_block_group(bg);
3989 return -ETXTBSY;
3990 }
3991
3992 rc = alloc_reloc_control(fs_info);
3993 if (!rc) {
3994 btrfs_put_block_group(bg);
3995 return -ENOMEM;
3996 }
3997
3998 ret = reloc_chunk_start(fs_info);
3999 if (ret < 0) {
4000 err = ret;
4001 goto out_put_bg;
4002 }
4003
4004 rc->extent_root = extent_root;
4005 rc->block_group = bg;
4006
4007 ret = btrfs_inc_block_group_ro(rc->block_group, true);
4008 if (ret) {
4009 err = ret;
4010 goto out;
4011 }
4012 rw = 1;
4013
4014 path = btrfs_alloc_path();
4015 if (!path) {
4016 err = -ENOMEM;
4017 goto out;
4018 }
4019
4020 inode = lookup_free_space_inode(rc->block_group, path);
4021 btrfs_free_path(path);
4022
4023 if (!IS_ERR(inode))
4024 ret = delete_block_group_cache(fs_info, rc->block_group, inode, 0);
4025 else
4026 ret = PTR_ERR(inode);
4027
4028 if (ret && ret != -ENOENT) {
4029 err = ret;
4030 goto out;
4031 }
4032
4033 rc->data_inode = create_reloc_inode(fs_info, rc->block_group);
4034 if (IS_ERR(rc->data_inode)) {
4035 err = PTR_ERR(rc->data_inode);
4036 rc->data_inode = NULL;
4037 goto out;
4038 }
4039
4040 describe_relocation(fs_info, rc->block_group);
4041
4042 btrfs_wait_block_group_reservations(rc->block_group);
4043 btrfs_wait_nocow_writers(rc->block_group);
4044 btrfs_wait_ordered_roots(fs_info, U64_MAX,
4045 rc->block_group->start,
4046 rc->block_group->length);
4047
4048 ret = btrfs_zone_finish(rc->block_group);
4049 WARN_ON(ret && ret != -EAGAIN);
4050
4051 while (1) {
4052 int finishes_stage;
4053
4054 mutex_lock(&fs_info->cleaner_mutex);
4055 ret = relocate_block_group(rc);
4056 mutex_unlock(&fs_info->cleaner_mutex);
4057 if (ret < 0)
4058 err = ret;
4059
4060 finishes_stage = rc->stage;
4061 /*
4062 * We may have gotten ENOSPC after we already dirtied some
4063 * extents. If writeout happens while we're relocating a
4064 * different block group we could end up hitting the
4065 * BUG_ON(rc->stage == UPDATE_DATA_PTRS) in
4066 * btrfs_reloc_cow_block. Make sure we write everything out
4067 * properly so we don't trip over this problem, and then break
4068 * out of the loop if we hit an error.
4069 */
4070 if (rc->stage == MOVE_DATA_EXTENTS && rc->found_file_extent) {
4071 ret = btrfs_wait_ordered_range(rc->data_inode, 0,
4072 (u64)-1);
4073 if (ret)
4074 err = ret;
4075 invalidate_mapping_pages(rc->data_inode->i_mapping,
4076 0, -1);
4077 rc->stage = UPDATE_DATA_PTRS;
4078 }
4079
4080 if (err < 0)
4081 goto out;
4082
4083 if (rc->extents_found == 0)
4084 break;
4085
4086 btrfs_info(fs_info, "found %llu extents, stage: %s",
4087 rc->extents_found, stage_to_string(finishes_stage));
4088 }
4089
4090 WARN_ON(rc->block_group->pinned > 0);
4091 WARN_ON(rc->block_group->reserved > 0);
4092 WARN_ON(rc->block_group->used > 0);
4093 out:
4094 if (err && rw)
4095 btrfs_dec_block_group_ro(rc->block_group);
4096 iput(rc->data_inode);
4097 out_put_bg:
4098 btrfs_put_block_group(bg);
4099 reloc_chunk_end(fs_info);
4100 free_reloc_control(rc);
4101 return err;
4102 }
4103
mark_garbage_root(struct btrfs_root * root)4104 static noinline_for_stack int mark_garbage_root(struct btrfs_root *root)
4105 {
4106 struct btrfs_fs_info *fs_info = root->fs_info;
4107 struct btrfs_trans_handle *trans;
4108 int ret, err;
4109
4110 trans = btrfs_start_transaction(fs_info->tree_root, 0);
4111 if (IS_ERR(trans))
4112 return PTR_ERR(trans);
4113
4114 memset(&root->root_item.drop_progress, 0,
4115 sizeof(root->root_item.drop_progress));
4116 btrfs_set_root_drop_level(&root->root_item, 0);
4117 btrfs_set_root_refs(&root->root_item, 0);
4118 ret = btrfs_update_root(trans, fs_info->tree_root,
4119 &root->root_key, &root->root_item);
4120
4121 err = btrfs_end_transaction(trans);
4122 if (err)
4123 return err;
4124 return ret;
4125 }
4126
4127 /*
4128 * recover relocation interrupted by system crash.
4129 *
4130 * this function resumes merging reloc trees with corresponding fs trees.
4131 * this is important for keeping the sharing of tree blocks
4132 */
btrfs_recover_relocation(struct btrfs_fs_info * fs_info)4133 int btrfs_recover_relocation(struct btrfs_fs_info *fs_info)
4134 {
4135 LIST_HEAD(reloc_roots);
4136 struct btrfs_key key;
4137 struct btrfs_root *fs_root;
4138 struct btrfs_root *reloc_root;
4139 struct btrfs_path *path;
4140 struct extent_buffer *leaf;
4141 struct reloc_control *rc = NULL;
4142 struct btrfs_trans_handle *trans;
4143 int ret;
4144 int err = 0;
4145
4146 path = btrfs_alloc_path();
4147 if (!path)
4148 return -ENOMEM;
4149 path->reada = READA_BACK;
4150
4151 key.objectid = BTRFS_TREE_RELOC_OBJECTID;
4152 key.type = BTRFS_ROOT_ITEM_KEY;
4153 key.offset = (u64)-1;
4154
4155 while (1) {
4156 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key,
4157 path, 0, 0);
4158 if (ret < 0) {
4159 err = ret;
4160 goto out;
4161 }
4162 if (ret > 0) {
4163 if (path->slots[0] == 0)
4164 break;
4165 path->slots[0]--;
4166 }
4167 leaf = path->nodes[0];
4168 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4169 btrfs_release_path(path);
4170
4171 if (key.objectid != BTRFS_TREE_RELOC_OBJECTID ||
4172 key.type != BTRFS_ROOT_ITEM_KEY)
4173 break;
4174
4175 reloc_root = btrfs_read_tree_root(fs_info->tree_root, &key);
4176 if (IS_ERR(reloc_root)) {
4177 err = PTR_ERR(reloc_root);
4178 goto out;
4179 }
4180
4181 set_bit(BTRFS_ROOT_SHAREABLE, &reloc_root->state);
4182 list_add(&reloc_root->root_list, &reloc_roots);
4183
4184 if (btrfs_root_refs(&reloc_root->root_item) > 0) {
4185 fs_root = btrfs_get_fs_root(fs_info,
4186 reloc_root->root_key.offset, false);
4187 if (IS_ERR(fs_root)) {
4188 ret = PTR_ERR(fs_root);
4189 if (ret != -ENOENT) {
4190 err = ret;
4191 goto out;
4192 }
4193 ret = mark_garbage_root(reloc_root);
4194 if (ret < 0) {
4195 err = ret;
4196 goto out;
4197 }
4198 } else {
4199 btrfs_put_root(fs_root);
4200 }
4201 }
4202
4203 if (key.offset == 0)
4204 break;
4205
4206 key.offset--;
4207 }
4208 btrfs_release_path(path);
4209
4210 if (list_empty(&reloc_roots))
4211 goto out;
4212
4213 rc = alloc_reloc_control(fs_info);
4214 if (!rc) {
4215 err = -ENOMEM;
4216 goto out;
4217 }
4218
4219 ret = reloc_chunk_start(fs_info);
4220 if (ret < 0) {
4221 err = ret;
4222 goto out_end;
4223 }
4224
4225 rc->extent_root = btrfs_extent_root(fs_info, 0);
4226
4227 set_reloc_control(rc);
4228
4229 trans = btrfs_join_transaction(rc->extent_root);
4230 if (IS_ERR(trans)) {
4231 err = PTR_ERR(trans);
4232 goto out_unset;
4233 }
4234
4235 rc->merge_reloc_tree = 1;
4236
4237 while (!list_empty(&reloc_roots)) {
4238 reloc_root = list_entry(reloc_roots.next,
4239 struct btrfs_root, root_list);
4240 list_del(&reloc_root->root_list);
4241
4242 if (btrfs_root_refs(&reloc_root->root_item) == 0) {
4243 list_add_tail(&reloc_root->root_list,
4244 &rc->reloc_roots);
4245 continue;
4246 }
4247
4248 fs_root = btrfs_get_fs_root(fs_info, reloc_root->root_key.offset,
4249 false);
4250 if (IS_ERR(fs_root)) {
4251 err = PTR_ERR(fs_root);
4252 list_add_tail(&reloc_root->root_list, &reloc_roots);
4253 btrfs_end_transaction(trans);
4254 goto out_unset;
4255 }
4256
4257 err = __add_reloc_root(reloc_root);
4258 ASSERT(err != -EEXIST);
4259 if (err) {
4260 list_add_tail(&reloc_root->root_list, &reloc_roots);
4261 btrfs_put_root(fs_root);
4262 btrfs_end_transaction(trans);
4263 goto out_unset;
4264 }
4265 fs_root->reloc_root = btrfs_grab_root(reloc_root);
4266 btrfs_put_root(fs_root);
4267 }
4268
4269 err = btrfs_commit_transaction(trans);
4270 if (err)
4271 goto out_unset;
4272
4273 merge_reloc_roots(rc);
4274
4275 unset_reloc_control(rc);
4276
4277 trans = btrfs_join_transaction(rc->extent_root);
4278 if (IS_ERR(trans)) {
4279 err = PTR_ERR(trans);
4280 goto out_clean;
4281 }
4282 err = btrfs_commit_transaction(trans);
4283 out_clean:
4284 ret = clean_dirty_subvols(rc);
4285 if (ret < 0 && !err)
4286 err = ret;
4287 out_unset:
4288 unset_reloc_control(rc);
4289 out_end:
4290 reloc_chunk_end(fs_info);
4291 free_reloc_control(rc);
4292 out:
4293 free_reloc_roots(&reloc_roots);
4294
4295 btrfs_free_path(path);
4296
4297 if (err == 0) {
4298 /* cleanup orphan inode in data relocation tree */
4299 fs_root = btrfs_grab_root(fs_info->data_reloc_root);
4300 ASSERT(fs_root);
4301 err = btrfs_orphan_cleanup(fs_root);
4302 btrfs_put_root(fs_root);
4303 }
4304 return err;
4305 }
4306
4307 /*
4308 * helper to add ordered checksum for data relocation.
4309 *
4310 * cloning checksum properly handles the nodatasum extents.
4311 * it also saves CPU time to re-calculate the checksum.
4312 */
btrfs_reloc_clone_csums(struct btrfs_ordered_extent * ordered)4313 int btrfs_reloc_clone_csums(struct btrfs_ordered_extent *ordered)
4314 {
4315 struct btrfs_inode *inode = BTRFS_I(ordered->inode);
4316 struct btrfs_fs_info *fs_info = inode->root->fs_info;
4317 u64 disk_bytenr = ordered->file_offset + inode->index_cnt;
4318 struct btrfs_root *csum_root = btrfs_csum_root(fs_info, disk_bytenr);
4319 LIST_HEAD(list);
4320 int ret;
4321
4322 ret = btrfs_lookup_csums_list(csum_root, disk_bytenr,
4323 disk_bytenr + ordered->num_bytes - 1,
4324 &list, 0, false);
4325 if (ret)
4326 return ret;
4327
4328 while (!list_empty(&list)) {
4329 struct btrfs_ordered_sum *sums =
4330 list_entry(list.next, struct btrfs_ordered_sum, list);
4331
4332 list_del_init(&sums->list);
4333
4334 /*
4335 * We need to offset the new_bytenr based on where the csum is.
4336 * We need to do this because we will read in entire prealloc
4337 * extents but we may have written to say the middle of the
4338 * prealloc extent, so we need to make sure the csum goes with
4339 * the right disk offset.
4340 *
4341 * We can do this because the data reloc inode refers strictly
4342 * to the on disk bytes, so we don't have to worry about
4343 * disk_len vs real len like with real inodes since it's all
4344 * disk length.
4345 */
4346 sums->logical = ordered->disk_bytenr + sums->logical - disk_bytenr;
4347 btrfs_add_ordered_sum(ordered, sums);
4348 }
4349
4350 return 0;
4351 }
4352
btrfs_reloc_cow_block(struct btrfs_trans_handle * trans,struct btrfs_root * root,const struct extent_buffer * buf,struct extent_buffer * cow)4353 int btrfs_reloc_cow_block(struct btrfs_trans_handle *trans,
4354 struct btrfs_root *root,
4355 const struct extent_buffer *buf,
4356 struct extent_buffer *cow)
4357 {
4358 struct btrfs_fs_info *fs_info = root->fs_info;
4359 struct reloc_control *rc;
4360 struct btrfs_backref_node *node;
4361 int first_cow = 0;
4362 int level;
4363 int ret = 0;
4364
4365 rc = fs_info->reloc_ctl;
4366 if (!rc)
4367 return 0;
4368
4369 BUG_ON(rc->stage == UPDATE_DATA_PTRS && btrfs_is_data_reloc_root(root));
4370
4371 level = btrfs_header_level(buf);
4372 if (btrfs_header_generation(buf) <=
4373 btrfs_root_last_snapshot(&root->root_item))
4374 first_cow = 1;
4375
4376 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID &&
4377 rc->create_reloc_tree) {
4378 WARN_ON(!first_cow && level == 0);
4379
4380 node = rc->backref_cache.path[level];
4381 BUG_ON(node->bytenr != buf->start &&
4382 node->new_bytenr != buf->start);
4383
4384 btrfs_backref_drop_node_buffer(node);
4385 atomic_inc(&cow->refs);
4386 node->eb = cow;
4387 node->new_bytenr = cow->start;
4388
4389 if (!node->pending) {
4390 list_move_tail(&node->list,
4391 &rc->backref_cache.pending[level]);
4392 node->pending = 1;
4393 }
4394
4395 if (first_cow)
4396 mark_block_processed(rc, node);
4397
4398 if (first_cow && level > 0)
4399 rc->nodes_relocated += buf->len;
4400 }
4401
4402 if (level == 0 && first_cow && rc->stage == UPDATE_DATA_PTRS)
4403 ret = replace_file_extents(trans, rc, root, cow);
4404 return ret;
4405 }
4406
4407 /*
4408 * called before creating snapshot. it calculates metadata reservation
4409 * required for relocating tree blocks in the snapshot
4410 */
btrfs_reloc_pre_snapshot(struct btrfs_pending_snapshot * pending,u64 * bytes_to_reserve)4411 void btrfs_reloc_pre_snapshot(struct btrfs_pending_snapshot *pending,
4412 u64 *bytes_to_reserve)
4413 {
4414 struct btrfs_root *root = pending->root;
4415 struct reloc_control *rc = root->fs_info->reloc_ctl;
4416
4417 if (!rc || !have_reloc_root(root))
4418 return;
4419
4420 if (!rc->merge_reloc_tree)
4421 return;
4422
4423 root = root->reloc_root;
4424 BUG_ON(btrfs_root_refs(&root->root_item) == 0);
4425 /*
4426 * relocation is in the stage of merging trees. the space
4427 * used by merging a reloc tree is twice the size of
4428 * relocated tree nodes in the worst case. half for cowing
4429 * the reloc tree, half for cowing the fs tree. the space
4430 * used by cowing the reloc tree will be freed after the
4431 * tree is dropped. if we create snapshot, cowing the fs
4432 * tree may use more space than it frees. so we need
4433 * reserve extra space.
4434 */
4435 *bytes_to_reserve += rc->nodes_relocated;
4436 }
4437
4438 /*
4439 * called after snapshot is created. migrate block reservation
4440 * and create reloc root for the newly created snapshot
4441 *
4442 * This is similar to btrfs_init_reloc_root(), we come out of here with two
4443 * references held on the reloc_root, one for root->reloc_root and one for
4444 * rc->reloc_roots.
4445 */
btrfs_reloc_post_snapshot(struct btrfs_trans_handle * trans,struct btrfs_pending_snapshot * pending)4446 int btrfs_reloc_post_snapshot(struct btrfs_trans_handle *trans,
4447 struct btrfs_pending_snapshot *pending)
4448 {
4449 struct btrfs_root *root = pending->root;
4450 struct btrfs_root *reloc_root;
4451 struct btrfs_root *new_root;
4452 struct reloc_control *rc = root->fs_info->reloc_ctl;
4453 int ret;
4454
4455 if (!rc || !have_reloc_root(root))
4456 return 0;
4457
4458 rc = root->fs_info->reloc_ctl;
4459 rc->merging_rsv_size += rc->nodes_relocated;
4460
4461 if (rc->merge_reloc_tree) {
4462 ret = btrfs_block_rsv_migrate(&pending->block_rsv,
4463 rc->block_rsv,
4464 rc->nodes_relocated, true);
4465 if (ret)
4466 return ret;
4467 }
4468
4469 new_root = pending->snap;
4470 reloc_root = create_reloc_root(trans, root->reloc_root,
4471 new_root->root_key.objectid);
4472 if (IS_ERR(reloc_root))
4473 return PTR_ERR(reloc_root);
4474
4475 ret = __add_reloc_root(reloc_root);
4476 ASSERT(ret != -EEXIST);
4477 if (ret) {
4478 /* Pairs with create_reloc_root */
4479 btrfs_put_root(reloc_root);
4480 return ret;
4481 }
4482 new_root->reloc_root = btrfs_grab_root(reloc_root);
4483
4484 if (rc->create_reloc_tree)
4485 ret = clone_backref_node(trans, rc, root, reloc_root);
4486 return ret;
4487 }
4488
4489 /*
4490 * Get the current bytenr for the block group which is being relocated.
4491 *
4492 * Return U64_MAX if no running relocation.
4493 */
btrfs_get_reloc_bg_bytenr(const struct btrfs_fs_info * fs_info)4494 u64 btrfs_get_reloc_bg_bytenr(const struct btrfs_fs_info *fs_info)
4495 {
4496 u64 logical = U64_MAX;
4497
4498 lockdep_assert_held(&fs_info->reloc_mutex);
4499
4500 if (fs_info->reloc_ctl && fs_info->reloc_ctl->block_group)
4501 logical = fs_info->reloc_ctl->block_group->start;
4502 return logical;
4503 }
4504