xref: /openbmc/linux/fs/btrfs/relocation.c (revision 907717ee)
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