xref: /linux/fs/btrfs/tree-mod-log.c (revision c532de5a67a70f8533d495f8f2aaa9a0491c3ad0)
1 // SPDX-License-Identifier: GPL-2.0
2 
3 #include "messages.h"
4 #include "tree-mod-log.h"
5 #include "disk-io.h"
6 #include "fs.h"
7 #include "accessors.h"
8 #include "tree-checker.h"
9 
10 struct tree_mod_root {
11 	u64 logical;
12 	u8 level;
13 };
14 
15 struct tree_mod_elem {
16 	struct rb_node node;
17 	u64 logical;
18 	u64 seq;
19 	enum btrfs_mod_log_op op;
20 
21 	/*
22 	 * This is used for BTRFS_MOD_LOG_KEY_* and BTRFS_MOD_LOG_MOVE_KEYS
23 	 * operations.
24 	 */
25 	int slot;
26 
27 	/* This is used for BTRFS_MOD_LOG_KEY* and BTRFS_MOD_LOG_ROOT_REPLACE. */
28 	u64 generation;
29 
30 	/* Those are used for op == BTRFS_MOD_LOG_KEY_{REPLACE,REMOVE}. */
31 	struct btrfs_disk_key key;
32 	u64 blockptr;
33 
34 	/* This is used for op == BTRFS_MOD_LOG_MOVE_KEYS. */
35 	struct {
36 		int dst_slot;
37 		int nr_items;
38 	} move;
39 
40 	/* This is used for op == BTRFS_MOD_LOG_ROOT_REPLACE. */
41 	struct tree_mod_root old_root;
42 };
43 
44 /*
45  * Pull a new tree mod seq number for our operation.
46  */
47 static u64 btrfs_inc_tree_mod_seq(struct btrfs_fs_info *fs_info)
48 {
49 	return atomic64_inc_return(&fs_info->tree_mod_seq);
50 }
51 
52 /*
53  * This adds a new blocker to the tree mod log's blocker list if the @elem
54  * passed does not already have a sequence number set. So when a caller expects
55  * to record tree modifications, it should ensure to set elem->seq to zero
56  * before calling btrfs_get_tree_mod_seq.
57  * Returns a fresh, unused tree log modification sequence number, even if no new
58  * blocker was added.
59  */
60 u64 btrfs_get_tree_mod_seq(struct btrfs_fs_info *fs_info,
61 			   struct btrfs_seq_list *elem)
62 {
63 	write_lock(&fs_info->tree_mod_log_lock);
64 	if (!elem->seq) {
65 		elem->seq = btrfs_inc_tree_mod_seq(fs_info);
66 		list_add_tail(&elem->list, &fs_info->tree_mod_seq_list);
67 		set_bit(BTRFS_FS_TREE_MOD_LOG_USERS, &fs_info->flags);
68 	}
69 	write_unlock(&fs_info->tree_mod_log_lock);
70 
71 	return elem->seq;
72 }
73 
74 void btrfs_put_tree_mod_seq(struct btrfs_fs_info *fs_info,
75 			    struct btrfs_seq_list *elem)
76 {
77 	struct rb_root *tm_root;
78 	struct rb_node *node;
79 	struct rb_node *next;
80 	struct tree_mod_elem *tm;
81 	u64 min_seq = BTRFS_SEQ_LAST;
82 	u64 seq_putting = elem->seq;
83 
84 	if (!seq_putting)
85 		return;
86 
87 	write_lock(&fs_info->tree_mod_log_lock);
88 	list_del(&elem->list);
89 	elem->seq = 0;
90 
91 	if (list_empty(&fs_info->tree_mod_seq_list)) {
92 		clear_bit(BTRFS_FS_TREE_MOD_LOG_USERS, &fs_info->flags);
93 	} else {
94 		struct btrfs_seq_list *first;
95 
96 		first = list_first_entry(&fs_info->tree_mod_seq_list,
97 					 struct btrfs_seq_list, list);
98 		if (seq_putting > first->seq) {
99 			/*
100 			 * Blocker with lower sequence number exists, we cannot
101 			 * remove anything from the log.
102 			 */
103 			write_unlock(&fs_info->tree_mod_log_lock);
104 			return;
105 		}
106 		min_seq = first->seq;
107 	}
108 
109 	/*
110 	 * Anything that's lower than the lowest existing (read: blocked)
111 	 * sequence number can be removed from the tree.
112 	 */
113 	tm_root = &fs_info->tree_mod_log;
114 	for (node = rb_first(tm_root); node; node = next) {
115 		next = rb_next(node);
116 		tm = rb_entry(node, struct tree_mod_elem, node);
117 		if (tm->seq >= min_seq)
118 			continue;
119 		rb_erase(node, tm_root);
120 		kfree(tm);
121 	}
122 	write_unlock(&fs_info->tree_mod_log_lock);
123 }
124 
125 /*
126  * Key order of the log:
127  *       node/leaf start address -> sequence
128  *
129  * The 'start address' is the logical address of the *new* root node for root
130  * replace operations, or the logical address of the affected block for all
131  * other operations.
132  */
133 static noinline int tree_mod_log_insert(struct btrfs_fs_info *fs_info,
134 					struct tree_mod_elem *tm)
135 {
136 	struct rb_root *tm_root;
137 	struct rb_node **new;
138 	struct rb_node *parent = NULL;
139 	struct tree_mod_elem *cur;
140 
141 	lockdep_assert_held_write(&fs_info->tree_mod_log_lock);
142 
143 	tm->seq = btrfs_inc_tree_mod_seq(fs_info);
144 
145 	tm_root = &fs_info->tree_mod_log;
146 	new = &tm_root->rb_node;
147 	while (*new) {
148 		cur = rb_entry(*new, struct tree_mod_elem, node);
149 		parent = *new;
150 		if (cur->logical < tm->logical)
151 			new = &((*new)->rb_left);
152 		else if (cur->logical > tm->logical)
153 			new = &((*new)->rb_right);
154 		else if (cur->seq < tm->seq)
155 			new = &((*new)->rb_left);
156 		else if (cur->seq > tm->seq)
157 			new = &((*new)->rb_right);
158 		else
159 			return -EEXIST;
160 	}
161 
162 	rb_link_node(&tm->node, parent, new);
163 	rb_insert_color(&tm->node, tm_root);
164 	return 0;
165 }
166 
167 /*
168  * Determines if logging can be omitted. Returns true if it can. Otherwise, it
169  * returns false with the tree_mod_log_lock acquired. The caller must hold
170  * this until all tree mod log insertions are recorded in the rb tree and then
171  * write unlock fs_info::tree_mod_log_lock.
172  */
173 static bool tree_mod_dont_log(struct btrfs_fs_info *fs_info, const struct extent_buffer *eb)
174 {
175 	if (!test_bit(BTRFS_FS_TREE_MOD_LOG_USERS, &fs_info->flags))
176 		return true;
177 	if (eb && btrfs_header_level(eb) == 0)
178 		return true;
179 
180 	write_lock(&fs_info->tree_mod_log_lock);
181 	if (list_empty(&(fs_info)->tree_mod_seq_list)) {
182 		write_unlock(&fs_info->tree_mod_log_lock);
183 		return true;
184 	}
185 
186 	return false;
187 }
188 
189 /* Similar to tree_mod_dont_log, but doesn't acquire any locks. */
190 static bool tree_mod_need_log(const struct btrfs_fs_info *fs_info,
191 			      const struct extent_buffer *eb)
192 {
193 	if (!test_bit(BTRFS_FS_TREE_MOD_LOG_USERS, &fs_info->flags))
194 		return false;
195 	if (eb && btrfs_header_level(eb) == 0)
196 		return false;
197 
198 	return true;
199 }
200 
201 static struct tree_mod_elem *alloc_tree_mod_elem(const struct extent_buffer *eb,
202 						 int slot,
203 						 enum btrfs_mod_log_op op)
204 {
205 	struct tree_mod_elem *tm;
206 
207 	tm = kzalloc(sizeof(*tm), GFP_NOFS);
208 	if (!tm)
209 		return NULL;
210 
211 	tm->logical = eb->start;
212 	if (op != BTRFS_MOD_LOG_KEY_ADD) {
213 		btrfs_node_key(eb, &tm->key, slot);
214 		tm->blockptr = btrfs_node_blockptr(eb, slot);
215 	}
216 	tm->op = op;
217 	tm->slot = slot;
218 	tm->generation = btrfs_node_ptr_generation(eb, slot);
219 	RB_CLEAR_NODE(&tm->node);
220 
221 	return tm;
222 }
223 
224 int btrfs_tree_mod_log_insert_key(const struct extent_buffer *eb, int slot,
225 				  enum btrfs_mod_log_op op)
226 {
227 	struct tree_mod_elem *tm;
228 	int ret = 0;
229 
230 	if (!tree_mod_need_log(eb->fs_info, eb))
231 		return 0;
232 
233 	tm = alloc_tree_mod_elem(eb, slot, op);
234 	if (!tm)
235 		ret = -ENOMEM;
236 
237 	if (tree_mod_dont_log(eb->fs_info, eb)) {
238 		kfree(tm);
239 		/*
240 		 * Don't error if we failed to allocate memory because we don't
241 		 * need to log.
242 		 */
243 		return 0;
244 	} else if (ret != 0) {
245 		/*
246 		 * We previously failed to allocate memory and we need to log,
247 		 * so we have to fail.
248 		 */
249 		goto out_unlock;
250 	}
251 
252 	ret = tree_mod_log_insert(eb->fs_info, tm);
253 out_unlock:
254 	write_unlock(&eb->fs_info->tree_mod_log_lock);
255 	if (ret)
256 		kfree(tm);
257 
258 	return ret;
259 }
260 
261 static struct tree_mod_elem *tree_mod_log_alloc_move(const struct extent_buffer *eb,
262 						     int dst_slot, int src_slot,
263 						     int nr_items)
264 {
265 	struct tree_mod_elem *tm;
266 
267 	tm = kzalloc(sizeof(*tm), GFP_NOFS);
268 	if (!tm)
269 		return ERR_PTR(-ENOMEM);
270 
271 	tm->logical = eb->start;
272 	tm->slot = src_slot;
273 	tm->move.dst_slot = dst_slot;
274 	tm->move.nr_items = nr_items;
275 	tm->op = BTRFS_MOD_LOG_MOVE_KEYS;
276 	RB_CLEAR_NODE(&tm->node);
277 
278 	return tm;
279 }
280 
281 int btrfs_tree_mod_log_insert_move(const struct extent_buffer *eb,
282 				   int dst_slot, int src_slot,
283 				   int nr_items)
284 {
285 	struct tree_mod_elem *tm = NULL;
286 	struct tree_mod_elem **tm_list = NULL;
287 	int ret = 0;
288 	int i;
289 	bool locked = false;
290 
291 	if (!tree_mod_need_log(eb->fs_info, eb))
292 		return 0;
293 
294 	tm_list = kcalloc(nr_items, sizeof(struct tree_mod_elem *), GFP_NOFS);
295 	if (!tm_list) {
296 		ret = -ENOMEM;
297 		goto lock;
298 	}
299 
300 	tm = tree_mod_log_alloc_move(eb, dst_slot, src_slot, nr_items);
301 	if (IS_ERR(tm)) {
302 		ret = PTR_ERR(tm);
303 		tm = NULL;
304 		goto lock;
305 	}
306 
307 	for (i = 0; i + dst_slot < src_slot && i < nr_items; i++) {
308 		tm_list[i] = alloc_tree_mod_elem(eb, i + dst_slot,
309 				BTRFS_MOD_LOG_KEY_REMOVE_WHILE_MOVING);
310 		if (!tm_list[i]) {
311 			ret = -ENOMEM;
312 			goto lock;
313 		}
314 	}
315 
316 lock:
317 	if (tree_mod_dont_log(eb->fs_info, eb)) {
318 		/*
319 		 * Don't error if we failed to allocate memory because we don't
320 		 * need to log.
321 		 */
322 		ret = 0;
323 		goto free_tms;
324 	}
325 	locked = true;
326 
327 	/*
328 	 * We previously failed to allocate memory and we need to log, so we
329 	 * have to fail.
330 	 */
331 	if (ret != 0)
332 		goto free_tms;
333 
334 	/*
335 	 * When we override something during the move, we log these removals.
336 	 * This can only happen when we move towards the beginning of the
337 	 * buffer, i.e. dst_slot < src_slot.
338 	 */
339 	for (i = 0; i + dst_slot < src_slot && i < nr_items; i++) {
340 		ret = tree_mod_log_insert(eb->fs_info, tm_list[i]);
341 		if (ret)
342 			goto free_tms;
343 	}
344 
345 	ret = tree_mod_log_insert(eb->fs_info, tm);
346 	if (ret)
347 		goto free_tms;
348 	write_unlock(&eb->fs_info->tree_mod_log_lock);
349 	kfree(tm_list);
350 
351 	return 0;
352 
353 free_tms:
354 	if (tm_list) {
355 		for (i = 0; i < nr_items; i++) {
356 			if (tm_list[i] && !RB_EMPTY_NODE(&tm_list[i]->node))
357 				rb_erase(&tm_list[i]->node, &eb->fs_info->tree_mod_log);
358 			kfree(tm_list[i]);
359 		}
360 	}
361 	if (locked)
362 		write_unlock(&eb->fs_info->tree_mod_log_lock);
363 	kfree(tm_list);
364 	kfree(tm);
365 
366 	return ret;
367 }
368 
369 static int tree_mod_log_free_eb(struct btrfs_fs_info *fs_info,
370 				struct tree_mod_elem **tm_list,
371 				int nritems)
372 {
373 	int i, j;
374 	int ret;
375 
376 	for (i = nritems - 1; i >= 0; i--) {
377 		ret = tree_mod_log_insert(fs_info, tm_list[i]);
378 		if (ret) {
379 			for (j = nritems - 1; j > i; j--)
380 				rb_erase(&tm_list[j]->node,
381 					 &fs_info->tree_mod_log);
382 			return ret;
383 		}
384 	}
385 
386 	return 0;
387 }
388 
389 int btrfs_tree_mod_log_insert_root(struct extent_buffer *old_root,
390 				   struct extent_buffer *new_root,
391 				   bool log_removal)
392 {
393 	struct btrfs_fs_info *fs_info = old_root->fs_info;
394 	struct tree_mod_elem *tm = NULL;
395 	struct tree_mod_elem **tm_list = NULL;
396 	int nritems = 0;
397 	int ret = 0;
398 	int i;
399 
400 	if (!tree_mod_need_log(fs_info, NULL))
401 		return 0;
402 
403 	if (log_removal && btrfs_header_level(old_root) > 0) {
404 		nritems = btrfs_header_nritems(old_root);
405 		tm_list = kcalloc(nritems, sizeof(struct tree_mod_elem *),
406 				  GFP_NOFS);
407 		if (!tm_list) {
408 			ret = -ENOMEM;
409 			goto lock;
410 		}
411 		for (i = 0; i < nritems; i++) {
412 			tm_list[i] = alloc_tree_mod_elem(old_root, i,
413 			    BTRFS_MOD_LOG_KEY_REMOVE_WHILE_FREEING);
414 			if (!tm_list[i]) {
415 				ret = -ENOMEM;
416 				goto lock;
417 			}
418 		}
419 	}
420 
421 	tm = kzalloc(sizeof(*tm), GFP_NOFS);
422 	if (!tm) {
423 		ret = -ENOMEM;
424 		goto lock;
425 	}
426 
427 	tm->logical = new_root->start;
428 	tm->old_root.logical = old_root->start;
429 	tm->old_root.level = btrfs_header_level(old_root);
430 	tm->generation = btrfs_header_generation(old_root);
431 	tm->op = BTRFS_MOD_LOG_ROOT_REPLACE;
432 
433 lock:
434 	if (tree_mod_dont_log(fs_info, NULL)) {
435 		/*
436 		 * Don't error if we failed to allocate memory because we don't
437 		 * need to log.
438 		 */
439 		ret = 0;
440 		goto free_tms;
441 	} else if (ret != 0) {
442 		/*
443 		 * We previously failed to allocate memory and we need to log,
444 		 * so we have to fail.
445 		 */
446 		goto out_unlock;
447 	}
448 
449 	if (tm_list)
450 		ret = tree_mod_log_free_eb(fs_info, tm_list, nritems);
451 	if (!ret)
452 		ret = tree_mod_log_insert(fs_info, tm);
453 
454 out_unlock:
455 	write_unlock(&fs_info->tree_mod_log_lock);
456 	if (ret)
457 		goto free_tms;
458 	kfree(tm_list);
459 
460 	return ret;
461 
462 free_tms:
463 	if (tm_list) {
464 		for (i = 0; i < nritems; i++)
465 			kfree(tm_list[i]);
466 		kfree(tm_list);
467 	}
468 	kfree(tm);
469 
470 	return ret;
471 }
472 
473 static struct tree_mod_elem *__tree_mod_log_search(struct btrfs_fs_info *fs_info,
474 						   u64 start, u64 min_seq,
475 						   bool smallest)
476 {
477 	struct rb_root *tm_root;
478 	struct rb_node *node;
479 	struct tree_mod_elem *cur = NULL;
480 	struct tree_mod_elem *found = NULL;
481 
482 	read_lock(&fs_info->tree_mod_log_lock);
483 	tm_root = &fs_info->tree_mod_log;
484 	node = tm_root->rb_node;
485 	while (node) {
486 		cur = rb_entry(node, struct tree_mod_elem, node);
487 		if (cur->logical < start) {
488 			node = node->rb_left;
489 		} else if (cur->logical > start) {
490 			node = node->rb_right;
491 		} else if (cur->seq < min_seq) {
492 			node = node->rb_left;
493 		} else if (!smallest) {
494 			/* We want the node with the highest seq */
495 			if (found)
496 				BUG_ON(found->seq > cur->seq);
497 			found = cur;
498 			node = node->rb_left;
499 		} else if (cur->seq > min_seq) {
500 			/* We want the node with the smallest seq */
501 			if (found)
502 				BUG_ON(found->seq < cur->seq);
503 			found = cur;
504 			node = node->rb_right;
505 		} else {
506 			found = cur;
507 			break;
508 		}
509 	}
510 	read_unlock(&fs_info->tree_mod_log_lock);
511 
512 	return found;
513 }
514 
515 /*
516  * This returns the element from the log with the smallest time sequence
517  * value that's in the log (the oldest log item). Any element with a time
518  * sequence lower than min_seq will be ignored.
519  */
520 static struct tree_mod_elem *tree_mod_log_search_oldest(struct btrfs_fs_info *fs_info,
521 							u64 start, u64 min_seq)
522 {
523 	return __tree_mod_log_search(fs_info, start, min_seq, true);
524 }
525 
526 /*
527  * This returns the element from the log with the largest time sequence
528  * value that's in the log (the most recent log item). Any element with
529  * a time sequence lower than min_seq will be ignored.
530  */
531 static struct tree_mod_elem *tree_mod_log_search(struct btrfs_fs_info *fs_info,
532 						 u64 start, u64 min_seq)
533 {
534 	return __tree_mod_log_search(fs_info, start, min_seq, false);
535 }
536 
537 int btrfs_tree_mod_log_eb_copy(struct extent_buffer *dst,
538 			       const struct extent_buffer *src,
539 			       unsigned long dst_offset,
540 			       unsigned long src_offset,
541 			       int nr_items)
542 {
543 	struct btrfs_fs_info *fs_info = dst->fs_info;
544 	int ret = 0;
545 	struct tree_mod_elem **tm_list = NULL;
546 	struct tree_mod_elem **tm_list_add = NULL;
547 	struct tree_mod_elem **tm_list_rem = NULL;
548 	int i;
549 	bool locked = false;
550 	struct tree_mod_elem *dst_move_tm = NULL;
551 	struct tree_mod_elem *src_move_tm = NULL;
552 	u32 dst_move_nr_items = btrfs_header_nritems(dst) - dst_offset;
553 	u32 src_move_nr_items = btrfs_header_nritems(src) - (src_offset + nr_items);
554 
555 	if (!tree_mod_need_log(fs_info, NULL))
556 		return 0;
557 
558 	if (btrfs_header_level(dst) == 0 && btrfs_header_level(src) == 0)
559 		return 0;
560 
561 	tm_list = kcalloc(nr_items * 2, sizeof(struct tree_mod_elem *),
562 			  GFP_NOFS);
563 	if (!tm_list) {
564 		ret = -ENOMEM;
565 		goto lock;
566 	}
567 
568 	if (dst_move_nr_items) {
569 		dst_move_tm = tree_mod_log_alloc_move(dst, dst_offset + nr_items,
570 						      dst_offset, dst_move_nr_items);
571 		if (IS_ERR(dst_move_tm)) {
572 			ret = PTR_ERR(dst_move_tm);
573 			dst_move_tm = NULL;
574 			goto lock;
575 		}
576 	}
577 	if (src_move_nr_items) {
578 		src_move_tm = tree_mod_log_alloc_move(src, src_offset,
579 						      src_offset + nr_items,
580 						      src_move_nr_items);
581 		if (IS_ERR(src_move_tm)) {
582 			ret = PTR_ERR(src_move_tm);
583 			src_move_tm = NULL;
584 			goto lock;
585 		}
586 	}
587 
588 	tm_list_add = tm_list;
589 	tm_list_rem = tm_list + nr_items;
590 	for (i = 0; i < nr_items; i++) {
591 		tm_list_rem[i] = alloc_tree_mod_elem(src, i + src_offset,
592 						     BTRFS_MOD_LOG_KEY_REMOVE);
593 		if (!tm_list_rem[i]) {
594 			ret = -ENOMEM;
595 			goto lock;
596 		}
597 
598 		tm_list_add[i] = alloc_tree_mod_elem(dst, i + dst_offset,
599 						     BTRFS_MOD_LOG_KEY_ADD);
600 		if (!tm_list_add[i]) {
601 			ret = -ENOMEM;
602 			goto lock;
603 		}
604 	}
605 
606 lock:
607 	if (tree_mod_dont_log(fs_info, NULL)) {
608 		/*
609 		 * Don't error if we failed to allocate memory because we don't
610 		 * need to log.
611 		 */
612 		ret = 0;
613 		goto free_tms;
614 	}
615 	locked = true;
616 
617 	/*
618 	 * We previously failed to allocate memory and we need to log, so we
619 	 * have to fail.
620 	 */
621 	if (ret != 0)
622 		goto free_tms;
623 
624 	if (dst_move_tm) {
625 		ret = tree_mod_log_insert(fs_info, dst_move_tm);
626 		if (ret)
627 			goto free_tms;
628 	}
629 	for (i = 0; i < nr_items; i++) {
630 		ret = tree_mod_log_insert(fs_info, tm_list_rem[i]);
631 		if (ret)
632 			goto free_tms;
633 		ret = tree_mod_log_insert(fs_info, tm_list_add[i]);
634 		if (ret)
635 			goto free_tms;
636 	}
637 	if (src_move_tm) {
638 		ret = tree_mod_log_insert(fs_info, src_move_tm);
639 		if (ret)
640 			goto free_tms;
641 	}
642 
643 	write_unlock(&fs_info->tree_mod_log_lock);
644 	kfree(tm_list);
645 
646 	return 0;
647 
648 free_tms:
649 	if (dst_move_tm && !RB_EMPTY_NODE(&dst_move_tm->node))
650 		rb_erase(&dst_move_tm->node, &fs_info->tree_mod_log);
651 	kfree(dst_move_tm);
652 	if (src_move_tm && !RB_EMPTY_NODE(&src_move_tm->node))
653 		rb_erase(&src_move_tm->node, &fs_info->tree_mod_log);
654 	kfree(src_move_tm);
655 	if (tm_list) {
656 		for (i = 0; i < nr_items * 2; i++) {
657 			if (tm_list[i] && !RB_EMPTY_NODE(&tm_list[i]->node))
658 				rb_erase(&tm_list[i]->node, &fs_info->tree_mod_log);
659 			kfree(tm_list[i]);
660 		}
661 	}
662 	if (locked)
663 		write_unlock(&fs_info->tree_mod_log_lock);
664 	kfree(tm_list);
665 
666 	return ret;
667 }
668 
669 int btrfs_tree_mod_log_free_eb(struct extent_buffer *eb)
670 {
671 	struct tree_mod_elem **tm_list = NULL;
672 	int nritems = 0;
673 	int i;
674 	int ret = 0;
675 
676 	if (!tree_mod_need_log(eb->fs_info, eb))
677 		return 0;
678 
679 	nritems = btrfs_header_nritems(eb);
680 	tm_list = kcalloc(nritems, sizeof(struct tree_mod_elem *), GFP_NOFS);
681 	if (!tm_list) {
682 		ret = -ENOMEM;
683 		goto lock;
684 	}
685 
686 	for (i = 0; i < nritems; i++) {
687 		tm_list[i] = alloc_tree_mod_elem(eb, i,
688 				    BTRFS_MOD_LOG_KEY_REMOVE_WHILE_FREEING);
689 		if (!tm_list[i]) {
690 			ret = -ENOMEM;
691 			goto lock;
692 		}
693 	}
694 
695 lock:
696 	if (tree_mod_dont_log(eb->fs_info, eb)) {
697 		/*
698 		 * Don't error if we failed to allocate memory because we don't
699 		 * need to log.
700 		 */
701 		ret = 0;
702 		goto free_tms;
703 	} else if (ret != 0) {
704 		/*
705 		 * We previously failed to allocate memory and we need to log,
706 		 * so we have to fail.
707 		 */
708 		goto out_unlock;
709 	}
710 
711 	ret = tree_mod_log_free_eb(eb->fs_info, tm_list, nritems);
712 out_unlock:
713 	write_unlock(&eb->fs_info->tree_mod_log_lock);
714 	if (ret)
715 		goto free_tms;
716 	kfree(tm_list);
717 
718 	return 0;
719 
720 free_tms:
721 	if (tm_list) {
722 		for (i = 0; i < nritems; i++)
723 			kfree(tm_list[i]);
724 		kfree(tm_list);
725 	}
726 
727 	return ret;
728 }
729 
730 /*
731  * Returns the logical address of the oldest predecessor of the given root.
732  * Entries older than time_seq are ignored.
733  */
734 static struct tree_mod_elem *tree_mod_log_oldest_root(struct extent_buffer *eb_root,
735 						      u64 time_seq)
736 {
737 	struct tree_mod_elem *tm;
738 	struct tree_mod_elem *found = NULL;
739 	u64 root_logical = eb_root->start;
740 	bool looped = false;
741 
742 	if (!time_seq)
743 		return NULL;
744 
745 	/*
746 	 * The very last operation that's logged for a root is the replacement
747 	 * operation (if it is replaced at all). This has the logical address
748 	 * of the *new* root, making it the very first operation that's logged
749 	 * for this root.
750 	 */
751 	while (1) {
752 		tm = tree_mod_log_search_oldest(eb_root->fs_info, root_logical,
753 						time_seq);
754 		if (!looped && !tm)
755 			return NULL;
756 		/*
757 		 * If there are no tree operation for the oldest root, we simply
758 		 * return it. This should only happen if that (old) root is at
759 		 * level 0.
760 		 */
761 		if (!tm)
762 			break;
763 
764 		/*
765 		 * If there's an operation that's not a root replacement, we
766 		 * found the oldest version of our root. Normally, we'll find a
767 		 * BTRFS_MOD_LOG_KEY_REMOVE_WHILE_FREEING operation here.
768 		 */
769 		if (tm->op != BTRFS_MOD_LOG_ROOT_REPLACE)
770 			break;
771 
772 		found = tm;
773 		root_logical = tm->old_root.logical;
774 		looped = true;
775 	}
776 
777 	/* If there's no old root to return, return what we found instead */
778 	if (!found)
779 		found = tm;
780 
781 	return found;
782 }
783 
784 
785 /*
786  * tm is a pointer to the first operation to rewind within eb. Then, all
787  * previous operations will be rewound (until we reach something older than
788  * time_seq).
789  */
790 static void tree_mod_log_rewind(struct btrfs_fs_info *fs_info,
791 				struct extent_buffer *eb,
792 				u64 time_seq,
793 				struct tree_mod_elem *first_tm)
794 {
795 	u32 n;
796 	struct rb_node *next;
797 	struct tree_mod_elem *tm = first_tm;
798 	unsigned long o_dst;
799 	unsigned long o_src;
800 	unsigned long p_size = sizeof(struct btrfs_key_ptr);
801 	/*
802 	 * max_slot tracks the maximum valid slot of the rewind eb at every
803 	 * step of the rewind. This is in contrast with 'n' which eventually
804 	 * matches the number of items, but can be wrong during moves or if
805 	 * removes overlap on already valid slots (which is probably separately
806 	 * a bug). We do this to validate the offsets of memmoves for rewinding
807 	 * moves and detect invalid memmoves.
808 	 *
809 	 * Since a rewind eb can start empty, max_slot is a signed integer with
810 	 * a special meaning for -1, which is that no slot is valid to move out
811 	 * of. Any other negative value is invalid.
812 	 */
813 	int max_slot;
814 	int move_src_end_slot;
815 	int move_dst_end_slot;
816 
817 	n = btrfs_header_nritems(eb);
818 	max_slot = n - 1;
819 	read_lock(&fs_info->tree_mod_log_lock);
820 	while (tm && tm->seq >= time_seq) {
821 		ASSERT(max_slot >= -1);
822 		/*
823 		 * All the operations are recorded with the operator used for
824 		 * the modification. As we're going backwards, we do the
825 		 * opposite of each operation here.
826 		 */
827 		switch (tm->op) {
828 		case BTRFS_MOD_LOG_KEY_REMOVE_WHILE_FREEING:
829 			BUG_ON(tm->slot < n);
830 			fallthrough;
831 		case BTRFS_MOD_LOG_KEY_REMOVE_WHILE_MOVING:
832 		case BTRFS_MOD_LOG_KEY_REMOVE:
833 			btrfs_set_node_key(eb, &tm->key, tm->slot);
834 			btrfs_set_node_blockptr(eb, tm->slot, tm->blockptr);
835 			btrfs_set_node_ptr_generation(eb, tm->slot,
836 						      tm->generation);
837 			n++;
838 			if (tm->slot > max_slot)
839 				max_slot = tm->slot;
840 			break;
841 		case BTRFS_MOD_LOG_KEY_REPLACE:
842 			BUG_ON(tm->slot >= n);
843 			btrfs_set_node_key(eb, &tm->key, tm->slot);
844 			btrfs_set_node_blockptr(eb, tm->slot, tm->blockptr);
845 			btrfs_set_node_ptr_generation(eb, tm->slot,
846 						      tm->generation);
847 			break;
848 		case BTRFS_MOD_LOG_KEY_ADD:
849 			/*
850 			 * It is possible we could have already removed keys
851 			 * behind the known max slot, so this will be an
852 			 * overestimate. In practice, the copy operation
853 			 * inserts them in increasing order, and overestimating
854 			 * just means we miss some warnings, so it's OK. It
855 			 * isn't worth carefully tracking the full array of
856 			 * valid slots to check against when moving.
857 			 */
858 			if (tm->slot == max_slot)
859 				max_slot--;
860 			/* if a move operation is needed it's in the log */
861 			n--;
862 			break;
863 		case BTRFS_MOD_LOG_MOVE_KEYS:
864 			ASSERT(tm->move.nr_items > 0);
865 			move_src_end_slot = tm->move.dst_slot + tm->move.nr_items - 1;
866 			move_dst_end_slot = tm->slot + tm->move.nr_items - 1;
867 			o_dst = btrfs_node_key_ptr_offset(eb, tm->slot);
868 			o_src = btrfs_node_key_ptr_offset(eb, tm->move.dst_slot);
869 			if (WARN_ON(move_src_end_slot > max_slot ||
870 				    tm->move.nr_items <= 0)) {
871 				btrfs_warn(fs_info,
872 "move from invalid tree mod log slot eb %llu slot %d dst_slot %d nr_items %d seq %llu n %u max_slot %d",
873 					   eb->start, tm->slot,
874 					   tm->move.dst_slot, tm->move.nr_items,
875 					   tm->seq, n, max_slot);
876 			}
877 			memmove_extent_buffer(eb, o_dst, o_src,
878 					      tm->move.nr_items * p_size);
879 			max_slot = move_dst_end_slot;
880 			break;
881 		case BTRFS_MOD_LOG_ROOT_REPLACE:
882 			/*
883 			 * This operation is special. For roots, this must be
884 			 * handled explicitly before rewinding.
885 			 * For non-roots, this operation may exist if the node
886 			 * was a root: root A -> child B; then A gets empty and
887 			 * B is promoted to the new root. In the mod log, we'll
888 			 * have a root-replace operation for B, a tree block
889 			 * that is no root. We simply ignore that operation.
890 			 */
891 			break;
892 		}
893 		next = rb_next(&tm->node);
894 		if (!next)
895 			break;
896 		tm = rb_entry(next, struct tree_mod_elem, node);
897 		if (tm->logical != first_tm->logical)
898 			break;
899 	}
900 	read_unlock(&fs_info->tree_mod_log_lock);
901 	btrfs_set_header_nritems(eb, n);
902 }
903 
904 /*
905  * Called with eb read locked. If the buffer cannot be rewound, the same buffer
906  * is returned. If rewind operations happen, a fresh buffer is returned. The
907  * returned buffer is always read-locked. If the returned buffer is not the
908  * input buffer, the lock on the input buffer is released and the input buffer
909  * is freed (its refcount is decremented).
910  */
911 struct extent_buffer *btrfs_tree_mod_log_rewind(struct btrfs_fs_info *fs_info,
912 						struct btrfs_path *path,
913 						struct extent_buffer *eb,
914 						u64 time_seq)
915 {
916 	struct extent_buffer *eb_rewin;
917 	struct tree_mod_elem *tm;
918 
919 	if (!time_seq)
920 		return eb;
921 
922 	if (btrfs_header_level(eb) == 0)
923 		return eb;
924 
925 	tm = tree_mod_log_search(fs_info, eb->start, time_seq);
926 	if (!tm)
927 		return eb;
928 
929 	if (tm->op == BTRFS_MOD_LOG_KEY_REMOVE_WHILE_FREEING) {
930 		BUG_ON(tm->slot != 0);
931 		eb_rewin = alloc_dummy_extent_buffer(fs_info, eb->start);
932 		if (!eb_rewin) {
933 			btrfs_tree_read_unlock(eb);
934 			free_extent_buffer(eb);
935 			return NULL;
936 		}
937 		btrfs_set_header_bytenr(eb_rewin, eb->start);
938 		btrfs_set_header_backref_rev(eb_rewin,
939 					     btrfs_header_backref_rev(eb));
940 		btrfs_set_header_owner(eb_rewin, btrfs_header_owner(eb));
941 		btrfs_set_header_level(eb_rewin, btrfs_header_level(eb));
942 	} else {
943 		eb_rewin = btrfs_clone_extent_buffer(eb);
944 		if (!eb_rewin) {
945 			btrfs_tree_read_unlock(eb);
946 			free_extent_buffer(eb);
947 			return NULL;
948 		}
949 	}
950 
951 	btrfs_tree_read_unlock(eb);
952 	free_extent_buffer(eb);
953 
954 	btrfs_set_buffer_lockdep_class(btrfs_header_owner(eb_rewin),
955 				       eb_rewin, btrfs_header_level(eb_rewin));
956 	btrfs_tree_read_lock(eb_rewin);
957 	tree_mod_log_rewind(fs_info, eb_rewin, time_seq, tm);
958 	WARN_ON(btrfs_header_nritems(eb_rewin) >
959 		BTRFS_NODEPTRS_PER_BLOCK(fs_info));
960 
961 	return eb_rewin;
962 }
963 
964 /*
965  * Rewind the state of @root's root node to the given @time_seq value.
966  * If there are no changes, the current root->root_node is returned. If anything
967  * changed in between, there's a fresh buffer allocated on which the rewind
968  * operations are done. In any case, the returned buffer is read locked.
969  * Returns NULL on error (with no locks held).
970  */
971 struct extent_buffer *btrfs_get_old_root(struct btrfs_root *root, u64 time_seq)
972 {
973 	struct btrfs_fs_info *fs_info = root->fs_info;
974 	struct tree_mod_elem *tm;
975 	struct extent_buffer *eb = NULL;
976 	struct extent_buffer *eb_root;
977 	u64 eb_root_owner = 0;
978 	struct extent_buffer *old;
979 	struct tree_mod_root *old_root = NULL;
980 	u64 old_generation = 0;
981 	u64 logical;
982 	int level;
983 
984 	eb_root = btrfs_read_lock_root_node(root);
985 	tm = tree_mod_log_oldest_root(eb_root, time_seq);
986 	if (!tm)
987 		return eb_root;
988 
989 	if (tm->op == BTRFS_MOD_LOG_ROOT_REPLACE) {
990 		old_root = &tm->old_root;
991 		old_generation = tm->generation;
992 		logical = old_root->logical;
993 		level = old_root->level;
994 	} else {
995 		logical = eb_root->start;
996 		level = btrfs_header_level(eb_root);
997 	}
998 
999 	tm = tree_mod_log_search(fs_info, logical, time_seq);
1000 	if (old_root && tm && tm->op != BTRFS_MOD_LOG_KEY_REMOVE_WHILE_FREEING) {
1001 		struct btrfs_tree_parent_check check = { 0 };
1002 
1003 		btrfs_tree_read_unlock(eb_root);
1004 		free_extent_buffer(eb_root);
1005 
1006 		check.level = level;
1007 		check.owner_root = btrfs_root_id(root);
1008 
1009 		old = read_tree_block(fs_info, logical, &check);
1010 		if (WARN_ON(IS_ERR(old) || !extent_buffer_uptodate(old))) {
1011 			if (!IS_ERR(old))
1012 				free_extent_buffer(old);
1013 			btrfs_warn(fs_info,
1014 				   "failed to read tree block %llu from get_old_root",
1015 				   logical);
1016 		} else {
1017 			struct tree_mod_elem *tm2;
1018 
1019 			btrfs_tree_read_lock(old);
1020 			eb = btrfs_clone_extent_buffer(old);
1021 			/*
1022 			 * After the lookup for the most recent tree mod operation
1023 			 * above and before we locked and cloned the extent buffer
1024 			 * 'old', a new tree mod log operation may have been added.
1025 			 * So lookup for a more recent one to make sure the number
1026 			 * of mod log operations we replay is consistent with the
1027 			 * number of items we have in the cloned extent buffer,
1028 			 * otherwise we can hit a BUG_ON when rewinding the extent
1029 			 * buffer.
1030 			 */
1031 			tm2 = tree_mod_log_search(fs_info, logical, time_seq);
1032 			btrfs_tree_read_unlock(old);
1033 			free_extent_buffer(old);
1034 			ASSERT(tm2);
1035 			ASSERT(tm2 == tm || tm2->seq > tm->seq);
1036 			if (!tm2 || tm2->seq < tm->seq) {
1037 				free_extent_buffer(eb);
1038 				return NULL;
1039 			}
1040 			tm = tm2;
1041 		}
1042 	} else if (old_root) {
1043 		eb_root_owner = btrfs_header_owner(eb_root);
1044 		btrfs_tree_read_unlock(eb_root);
1045 		free_extent_buffer(eb_root);
1046 		eb = alloc_dummy_extent_buffer(fs_info, logical);
1047 	} else {
1048 		eb = btrfs_clone_extent_buffer(eb_root);
1049 		btrfs_tree_read_unlock(eb_root);
1050 		free_extent_buffer(eb_root);
1051 	}
1052 
1053 	if (!eb)
1054 		return NULL;
1055 	if (old_root) {
1056 		btrfs_set_header_bytenr(eb, eb->start);
1057 		btrfs_set_header_backref_rev(eb, BTRFS_MIXED_BACKREF_REV);
1058 		btrfs_set_header_owner(eb, eb_root_owner);
1059 		btrfs_set_header_level(eb, old_root->level);
1060 		btrfs_set_header_generation(eb, old_generation);
1061 	}
1062 	btrfs_set_buffer_lockdep_class(btrfs_header_owner(eb), eb,
1063 				       btrfs_header_level(eb));
1064 	btrfs_tree_read_lock(eb);
1065 	if (tm)
1066 		tree_mod_log_rewind(fs_info, eb, time_seq, tm);
1067 	else
1068 		WARN_ON(btrfs_header_level(eb) != 0);
1069 	WARN_ON(btrfs_header_nritems(eb) > BTRFS_NODEPTRS_PER_BLOCK(fs_info));
1070 
1071 	return eb;
1072 }
1073 
1074 int btrfs_old_root_level(struct btrfs_root *root, u64 time_seq)
1075 {
1076 	struct tree_mod_elem *tm;
1077 	int level;
1078 	struct extent_buffer *eb_root = btrfs_root_node(root);
1079 
1080 	tm = tree_mod_log_oldest_root(eb_root, time_seq);
1081 	if (tm && tm->op == BTRFS_MOD_LOG_ROOT_REPLACE)
1082 		level = tm->old_root.level;
1083 	else
1084 		level = btrfs_header_level(eb_root);
1085 
1086 	free_extent_buffer(eb_root);
1087 
1088 	return level;
1089 }
1090 
1091 /*
1092  * Return the lowest sequence number in the tree modification log.
1093  *
1094  * Return the sequence number of the oldest tree modification log user, which
1095  * corresponds to the lowest sequence number of all existing users. If there are
1096  * no users it returns 0.
1097  */
1098 u64 btrfs_tree_mod_log_lowest_seq(struct btrfs_fs_info *fs_info)
1099 {
1100 	u64 ret = 0;
1101 
1102 	read_lock(&fs_info->tree_mod_log_lock);
1103 	if (!list_empty(&fs_info->tree_mod_seq_list)) {
1104 		struct btrfs_seq_list *elem;
1105 
1106 		elem = list_first_entry(&fs_info->tree_mod_seq_list,
1107 					struct btrfs_seq_list, list);
1108 		ret = elem->seq;
1109 	}
1110 	read_unlock(&fs_info->tree_mod_log_lock);
1111 
1112 	return ret;
1113 }
1114