xref: /linux/fs/bcachefs/snapshot.c (revision c532de5a67a70f8533d495f8f2aaa9a0491c3ad0)
1 // SPDX-License-Identifier: GPL-2.0
2 
3 #include "bcachefs.h"
4 #include "bkey_buf.h"
5 #include "btree_key_cache.h"
6 #include "btree_update.h"
7 #include "buckets.h"
8 #include "errcode.h"
9 #include "error.h"
10 #include "fs.h"
11 #include "recovery_passes.h"
12 #include "snapshot.h"
13 
14 #include <linux/random.h>
15 
16 /*
17  * Snapshot trees:
18  *
19  * Keys in BTREE_ID_snapshot_trees identify a whole tree of snapshot nodes; they
20  * exist to provide a stable identifier for the whole lifetime of a snapshot
21  * tree.
22  */
23 
24 void bch2_snapshot_tree_to_text(struct printbuf *out, struct bch_fs *c,
25 				struct bkey_s_c k)
26 {
27 	struct bkey_s_c_snapshot_tree t = bkey_s_c_to_snapshot_tree(k);
28 
29 	prt_printf(out, "subvol %u root snapshot %u",
30 		   le32_to_cpu(t.v->master_subvol),
31 		   le32_to_cpu(t.v->root_snapshot));
32 }
33 
34 int bch2_snapshot_tree_validate(struct bch_fs *c, struct bkey_s_c k,
35 			       enum bch_validate_flags flags)
36 {
37 	int ret = 0;
38 
39 	bkey_fsck_err_on(bkey_gt(k.k->p, POS(0, U32_MAX)) ||
40 			 bkey_lt(k.k->p, POS(0, 1)),
41 			 c, snapshot_tree_pos_bad,
42 			 "bad pos");
43 fsck_err:
44 	return ret;
45 }
46 
47 int bch2_snapshot_tree_lookup(struct btree_trans *trans, u32 id,
48 			      struct bch_snapshot_tree *s)
49 {
50 	int ret = bch2_bkey_get_val_typed(trans, BTREE_ID_snapshot_trees, POS(0, id),
51 					  BTREE_ITER_with_updates, snapshot_tree, s);
52 
53 	if (bch2_err_matches(ret, ENOENT))
54 		ret = -BCH_ERR_ENOENT_snapshot_tree;
55 	return ret;
56 }
57 
58 struct bkey_i_snapshot_tree *
59 __bch2_snapshot_tree_create(struct btree_trans *trans)
60 {
61 	struct btree_iter iter;
62 	int ret = bch2_bkey_get_empty_slot(trans, &iter,
63 			BTREE_ID_snapshot_trees, POS(0, U32_MAX));
64 	struct bkey_i_snapshot_tree *s_t;
65 
66 	if (ret == -BCH_ERR_ENOSPC_btree_slot)
67 		ret = -BCH_ERR_ENOSPC_snapshot_tree;
68 	if (ret)
69 		return ERR_PTR(ret);
70 
71 	s_t = bch2_bkey_alloc(trans, &iter, 0, snapshot_tree);
72 	ret = PTR_ERR_OR_ZERO(s_t);
73 	bch2_trans_iter_exit(trans, &iter);
74 	return ret ? ERR_PTR(ret) : s_t;
75 }
76 
77 static int bch2_snapshot_tree_create(struct btree_trans *trans,
78 				u32 root_id, u32 subvol_id, u32 *tree_id)
79 {
80 	struct bkey_i_snapshot_tree *n_tree =
81 		__bch2_snapshot_tree_create(trans);
82 
83 	if (IS_ERR(n_tree))
84 		return PTR_ERR(n_tree);
85 
86 	n_tree->v.master_subvol	= cpu_to_le32(subvol_id);
87 	n_tree->v.root_snapshot	= cpu_to_le32(root_id);
88 	*tree_id = n_tree->k.p.offset;
89 	return 0;
90 }
91 
92 /* Snapshot nodes: */
93 
94 static bool __bch2_snapshot_is_ancestor_early(struct snapshot_table *t, u32 id, u32 ancestor)
95 {
96 	while (id && id < ancestor) {
97 		const struct snapshot_t *s = __snapshot_t(t, id);
98 		id = s ? s->parent : 0;
99 	}
100 	return id == ancestor;
101 }
102 
103 static bool bch2_snapshot_is_ancestor_early(struct bch_fs *c, u32 id, u32 ancestor)
104 {
105 	rcu_read_lock();
106 	bool ret = __bch2_snapshot_is_ancestor_early(rcu_dereference(c->snapshots), id, ancestor);
107 	rcu_read_unlock();
108 
109 	return ret;
110 }
111 
112 static inline u32 get_ancestor_below(struct snapshot_table *t, u32 id, u32 ancestor)
113 {
114 	const struct snapshot_t *s = __snapshot_t(t, id);
115 	if (!s)
116 		return 0;
117 
118 	if (s->skip[2] <= ancestor)
119 		return s->skip[2];
120 	if (s->skip[1] <= ancestor)
121 		return s->skip[1];
122 	if (s->skip[0] <= ancestor)
123 		return s->skip[0];
124 	return s->parent;
125 }
126 
127 static bool test_ancestor_bitmap(struct snapshot_table *t, u32 id, u32 ancestor)
128 {
129 	const struct snapshot_t *s = __snapshot_t(t, id);
130 	if (!s)
131 		return false;
132 
133 	return test_bit(ancestor - id - 1, s->is_ancestor);
134 }
135 
136 bool __bch2_snapshot_is_ancestor(struct bch_fs *c, u32 id, u32 ancestor)
137 {
138 	bool ret;
139 
140 	rcu_read_lock();
141 	struct snapshot_table *t = rcu_dereference(c->snapshots);
142 
143 	if (unlikely(c->recovery_pass_done < BCH_RECOVERY_PASS_check_snapshots)) {
144 		ret = __bch2_snapshot_is_ancestor_early(t, id, ancestor);
145 		goto out;
146 	}
147 
148 	while (id && id < ancestor - IS_ANCESTOR_BITMAP)
149 		id = get_ancestor_below(t, id, ancestor);
150 
151 	ret = id && id < ancestor
152 		? test_ancestor_bitmap(t, id, ancestor)
153 		: id == ancestor;
154 
155 	EBUG_ON(ret != __bch2_snapshot_is_ancestor_early(t, id, ancestor));
156 out:
157 	rcu_read_unlock();
158 
159 	return ret;
160 }
161 
162 static noinline struct snapshot_t *__snapshot_t_mut(struct bch_fs *c, u32 id)
163 {
164 	size_t idx = U32_MAX - id;
165 	struct snapshot_table *new, *old;
166 
167 	size_t new_bytes = kmalloc_size_roundup(struct_size(new, s, idx + 1));
168 	size_t new_size = (new_bytes - sizeof(*new)) / sizeof(new->s[0]);
169 
170 	if (unlikely(new_bytes > INT_MAX))
171 		return NULL;
172 
173 	new = kvzalloc(new_bytes, GFP_KERNEL);
174 	if (!new)
175 		return NULL;
176 
177 	new->nr = new_size;
178 
179 	old = rcu_dereference_protected(c->snapshots, true);
180 	if (old)
181 		memcpy(new->s, old->s, sizeof(old->s[0]) * old->nr);
182 
183 	rcu_assign_pointer(c->snapshots, new);
184 	kvfree_rcu(old, rcu);
185 
186 	return &rcu_dereference_protected(c->snapshots,
187 				lockdep_is_held(&c->snapshot_table_lock))->s[idx];
188 }
189 
190 static inline struct snapshot_t *snapshot_t_mut(struct bch_fs *c, u32 id)
191 {
192 	size_t idx = U32_MAX - id;
193 	struct snapshot_table *table =
194 		rcu_dereference_protected(c->snapshots,
195 				lockdep_is_held(&c->snapshot_table_lock));
196 
197 	lockdep_assert_held(&c->snapshot_table_lock);
198 
199 	if (likely(table && idx < table->nr))
200 		return &table->s[idx];
201 
202 	return __snapshot_t_mut(c, id);
203 }
204 
205 void bch2_snapshot_to_text(struct printbuf *out, struct bch_fs *c,
206 			   struct bkey_s_c k)
207 {
208 	struct bkey_s_c_snapshot s = bkey_s_c_to_snapshot(k);
209 
210 	prt_printf(out, "is_subvol %llu deleted %llu parent %10u children %10u %10u subvol %u tree %u",
211 	       BCH_SNAPSHOT_SUBVOL(s.v),
212 	       BCH_SNAPSHOT_DELETED(s.v),
213 	       le32_to_cpu(s.v->parent),
214 	       le32_to_cpu(s.v->children[0]),
215 	       le32_to_cpu(s.v->children[1]),
216 	       le32_to_cpu(s.v->subvol),
217 	       le32_to_cpu(s.v->tree));
218 
219 	if (bkey_val_bytes(k.k) > offsetof(struct bch_snapshot, depth))
220 		prt_printf(out, " depth %u skiplist %u %u %u",
221 			   le32_to_cpu(s.v->depth),
222 			   le32_to_cpu(s.v->skip[0]),
223 			   le32_to_cpu(s.v->skip[1]),
224 			   le32_to_cpu(s.v->skip[2]));
225 }
226 
227 int bch2_snapshot_validate(struct bch_fs *c, struct bkey_s_c k,
228 			  enum bch_validate_flags flags)
229 {
230 	struct bkey_s_c_snapshot s;
231 	u32 i, id;
232 	int ret = 0;
233 
234 	bkey_fsck_err_on(bkey_gt(k.k->p, POS(0, U32_MAX)) ||
235 			 bkey_lt(k.k->p, POS(0, 1)),
236 			 c, snapshot_pos_bad,
237 			 "bad pos");
238 
239 	s = bkey_s_c_to_snapshot(k);
240 
241 	id = le32_to_cpu(s.v->parent);
242 	bkey_fsck_err_on(id && id <= k.k->p.offset,
243 			 c, snapshot_parent_bad,
244 			 "bad parent node (%u <= %llu)",
245 			 id, k.k->p.offset);
246 
247 	bkey_fsck_err_on(le32_to_cpu(s.v->children[0]) < le32_to_cpu(s.v->children[1]),
248 			 c, snapshot_children_not_normalized,
249 			 "children not normalized");
250 
251 	bkey_fsck_err_on(s.v->children[0] && s.v->children[0] == s.v->children[1],
252 			 c, snapshot_child_duplicate,
253 			 "duplicate child nodes");
254 
255 	for (i = 0; i < 2; i++) {
256 		id = le32_to_cpu(s.v->children[i]);
257 
258 		bkey_fsck_err_on(id >= k.k->p.offset,
259 				 c, snapshot_child_bad,
260 				 "bad child node (%u >= %llu)",
261 				 id, k.k->p.offset);
262 	}
263 
264 	if (bkey_val_bytes(k.k) > offsetof(struct bch_snapshot, skip)) {
265 		bkey_fsck_err_on(le32_to_cpu(s.v->skip[0]) > le32_to_cpu(s.v->skip[1]) ||
266 				 le32_to_cpu(s.v->skip[1]) > le32_to_cpu(s.v->skip[2]),
267 				 c, snapshot_skiplist_not_normalized,
268 				 "skiplist not normalized");
269 
270 		for (i = 0; i < ARRAY_SIZE(s.v->skip); i++) {
271 			id = le32_to_cpu(s.v->skip[i]);
272 
273 			bkey_fsck_err_on(id && id < le32_to_cpu(s.v->parent),
274 					 c, snapshot_skiplist_bad,
275 					 "bad skiplist node %u", id);
276 		}
277 	}
278 fsck_err:
279 	return ret;
280 }
281 
282 static void __set_is_ancestor_bitmap(struct bch_fs *c, u32 id)
283 {
284 	struct snapshot_t *t = snapshot_t_mut(c, id);
285 	u32 parent = id;
286 
287 	while ((parent = bch2_snapshot_parent_early(c, parent)) &&
288 	       parent - id - 1 < IS_ANCESTOR_BITMAP)
289 		__set_bit(parent - id - 1, t->is_ancestor);
290 }
291 
292 static void set_is_ancestor_bitmap(struct bch_fs *c, u32 id)
293 {
294 	mutex_lock(&c->snapshot_table_lock);
295 	__set_is_ancestor_bitmap(c, id);
296 	mutex_unlock(&c->snapshot_table_lock);
297 }
298 
299 static int __bch2_mark_snapshot(struct btree_trans *trans,
300 		       enum btree_id btree, unsigned level,
301 		       struct bkey_s_c old, struct bkey_s_c new,
302 		       enum btree_iter_update_trigger_flags flags)
303 {
304 	struct bch_fs *c = trans->c;
305 	struct snapshot_t *t;
306 	u32 id = new.k->p.offset;
307 	int ret = 0;
308 
309 	mutex_lock(&c->snapshot_table_lock);
310 
311 	t = snapshot_t_mut(c, id);
312 	if (!t) {
313 		ret = -BCH_ERR_ENOMEM_mark_snapshot;
314 		goto err;
315 	}
316 
317 	if (new.k->type == KEY_TYPE_snapshot) {
318 		struct bkey_s_c_snapshot s = bkey_s_c_to_snapshot(new);
319 
320 		t->parent	= le32_to_cpu(s.v->parent);
321 		t->children[0]	= le32_to_cpu(s.v->children[0]);
322 		t->children[1]	= le32_to_cpu(s.v->children[1]);
323 		t->subvol	= BCH_SNAPSHOT_SUBVOL(s.v) ? le32_to_cpu(s.v->subvol) : 0;
324 		t->tree		= le32_to_cpu(s.v->tree);
325 
326 		if (bkey_val_bytes(s.k) > offsetof(struct bch_snapshot, depth)) {
327 			t->depth	= le32_to_cpu(s.v->depth);
328 			t->skip[0]	= le32_to_cpu(s.v->skip[0]);
329 			t->skip[1]	= le32_to_cpu(s.v->skip[1]);
330 			t->skip[2]	= le32_to_cpu(s.v->skip[2]);
331 		} else {
332 			t->depth	= 0;
333 			t->skip[0]	= 0;
334 			t->skip[1]	= 0;
335 			t->skip[2]	= 0;
336 		}
337 
338 		__set_is_ancestor_bitmap(c, id);
339 
340 		if (BCH_SNAPSHOT_DELETED(s.v)) {
341 			set_bit(BCH_FS_need_delete_dead_snapshots, &c->flags);
342 			if (c->curr_recovery_pass > BCH_RECOVERY_PASS_delete_dead_snapshots)
343 				bch2_delete_dead_snapshots_async(c);
344 		}
345 	} else {
346 		memset(t, 0, sizeof(*t));
347 	}
348 err:
349 	mutex_unlock(&c->snapshot_table_lock);
350 	return ret;
351 }
352 
353 int bch2_mark_snapshot(struct btree_trans *trans,
354 		       enum btree_id btree, unsigned level,
355 		       struct bkey_s_c old, struct bkey_s new,
356 		       enum btree_iter_update_trigger_flags flags)
357 {
358 	return __bch2_mark_snapshot(trans, btree, level, old, new.s_c, flags);
359 }
360 
361 int bch2_snapshot_lookup(struct btree_trans *trans, u32 id,
362 			 struct bch_snapshot *s)
363 {
364 	return bch2_bkey_get_val_typed(trans, BTREE_ID_snapshots, POS(0, id),
365 				       BTREE_ITER_with_updates, snapshot, s);
366 }
367 
368 static int bch2_snapshot_live(struct btree_trans *trans, u32 id)
369 {
370 	struct bch_snapshot v;
371 	int ret;
372 
373 	if (!id)
374 		return 0;
375 
376 	ret = bch2_snapshot_lookup(trans, id, &v);
377 	if (bch2_err_matches(ret, ENOENT))
378 		bch_err(trans->c, "snapshot node %u not found", id);
379 	if (ret)
380 		return ret;
381 
382 	return !BCH_SNAPSHOT_DELETED(&v);
383 }
384 
385 /*
386  * If @k is a snapshot with just one live child, it's part of a linear chain,
387  * which we consider to be an equivalence class: and then after snapshot
388  * deletion cleanup, there should only be a single key at a given position in
389  * this equivalence class.
390  *
391  * This sets the equivalence class of @k to be the child's equivalence class, if
392  * it's part of such a linear chain: this correctly sets equivalence classes on
393  * startup if we run leaf to root (i.e. in natural key order).
394  */
395 static int bch2_snapshot_set_equiv(struct btree_trans *trans, struct bkey_s_c k)
396 {
397 	struct bch_fs *c = trans->c;
398 	unsigned i, nr_live = 0, live_idx = 0;
399 	struct bkey_s_c_snapshot snap;
400 	u32 id = k.k->p.offset, child[2];
401 
402 	if (k.k->type != KEY_TYPE_snapshot)
403 		return 0;
404 
405 	snap = bkey_s_c_to_snapshot(k);
406 
407 	child[0] = le32_to_cpu(snap.v->children[0]);
408 	child[1] = le32_to_cpu(snap.v->children[1]);
409 
410 	for (i = 0; i < 2; i++) {
411 		int ret = bch2_snapshot_live(trans, child[i]);
412 
413 		if (ret < 0)
414 			return ret;
415 
416 		if (ret)
417 			live_idx = i;
418 		nr_live += ret;
419 	}
420 
421 	mutex_lock(&c->snapshot_table_lock);
422 
423 	snapshot_t_mut(c, id)->equiv = nr_live == 1
424 		? snapshot_t_mut(c, child[live_idx])->equiv
425 		: id;
426 
427 	mutex_unlock(&c->snapshot_table_lock);
428 
429 	return 0;
430 }
431 
432 /* fsck: */
433 
434 static u32 bch2_snapshot_child(struct bch_fs *c, u32 id, unsigned child)
435 {
436 	return snapshot_t(c, id)->children[child];
437 }
438 
439 static u32 bch2_snapshot_left_child(struct bch_fs *c, u32 id)
440 {
441 	return bch2_snapshot_child(c, id, 0);
442 }
443 
444 static u32 bch2_snapshot_right_child(struct bch_fs *c, u32 id)
445 {
446 	return bch2_snapshot_child(c, id, 1);
447 }
448 
449 static u32 bch2_snapshot_tree_next(struct bch_fs *c, u32 id)
450 {
451 	u32 n, parent;
452 
453 	n = bch2_snapshot_left_child(c, id);
454 	if (n)
455 		return n;
456 
457 	while ((parent = bch2_snapshot_parent(c, id))) {
458 		n = bch2_snapshot_right_child(c, parent);
459 		if (n && n != id)
460 			return n;
461 		id = parent;
462 	}
463 
464 	return 0;
465 }
466 
467 static u32 bch2_snapshot_tree_oldest_subvol(struct bch_fs *c, u32 snapshot_root)
468 {
469 	u32 id = snapshot_root;
470 	u32 subvol = 0, s;
471 
472 	rcu_read_lock();
473 	while (id) {
474 		s = snapshot_t(c, id)->subvol;
475 
476 		if (s && (!subvol || s < subvol))
477 			subvol = s;
478 
479 		id = bch2_snapshot_tree_next(c, id);
480 	}
481 	rcu_read_unlock();
482 
483 	return subvol;
484 }
485 
486 static int bch2_snapshot_tree_master_subvol(struct btree_trans *trans,
487 					    u32 snapshot_root, u32 *subvol_id)
488 {
489 	struct bch_fs *c = trans->c;
490 	struct btree_iter iter;
491 	struct bkey_s_c k;
492 	bool found = false;
493 	int ret;
494 
495 	for_each_btree_key_norestart(trans, iter, BTREE_ID_subvolumes, POS_MIN,
496 				     0, k, ret) {
497 		if (k.k->type != KEY_TYPE_subvolume)
498 			continue;
499 
500 		struct bkey_s_c_subvolume s = bkey_s_c_to_subvolume(k);
501 		if (!bch2_snapshot_is_ancestor(c, le32_to_cpu(s.v->snapshot), snapshot_root))
502 			continue;
503 		if (!BCH_SUBVOLUME_SNAP(s.v)) {
504 			*subvol_id = s.k->p.offset;
505 			found = true;
506 			break;
507 		}
508 	}
509 
510 	bch2_trans_iter_exit(trans, &iter);
511 
512 	if (!ret && !found) {
513 		struct bkey_i_subvolume *u;
514 
515 		*subvol_id = bch2_snapshot_tree_oldest_subvol(c, snapshot_root);
516 
517 		u = bch2_bkey_get_mut_typed(trans, &iter,
518 					    BTREE_ID_subvolumes, POS(0, *subvol_id),
519 					    0, subvolume);
520 		ret = PTR_ERR_OR_ZERO(u);
521 		if (ret)
522 			return ret;
523 
524 		SET_BCH_SUBVOLUME_SNAP(&u->v, false);
525 	}
526 
527 	return ret;
528 }
529 
530 static int check_snapshot_tree(struct btree_trans *trans,
531 			       struct btree_iter *iter,
532 			       struct bkey_s_c k)
533 {
534 	struct bch_fs *c = trans->c;
535 	struct bkey_s_c_snapshot_tree st;
536 	struct bch_snapshot s;
537 	struct bch_subvolume subvol;
538 	struct printbuf buf = PRINTBUF;
539 	u32 root_id;
540 	int ret;
541 
542 	if (k.k->type != KEY_TYPE_snapshot_tree)
543 		return 0;
544 
545 	st = bkey_s_c_to_snapshot_tree(k);
546 	root_id = le32_to_cpu(st.v->root_snapshot);
547 
548 	ret = bch2_snapshot_lookup(trans, root_id, &s);
549 	if (ret && !bch2_err_matches(ret, ENOENT))
550 		goto err;
551 
552 	if (fsck_err_on(ret ||
553 			root_id != bch2_snapshot_root(c, root_id) ||
554 			st.k->p.offset != le32_to_cpu(s.tree),
555 			trans, snapshot_tree_to_missing_snapshot,
556 			"snapshot tree points to missing/incorrect snapshot:\n  %s",
557 			(bch2_bkey_val_to_text(&buf, c, st.s_c), buf.buf))) {
558 		ret = bch2_btree_delete_at(trans, iter, 0);
559 		goto err;
560 	}
561 
562 	ret = bch2_subvolume_get(trans, le32_to_cpu(st.v->master_subvol),
563 				 false, 0, &subvol);
564 	if (ret && !bch2_err_matches(ret, ENOENT))
565 		goto err;
566 
567 	if (fsck_err_on(ret,
568 			trans, snapshot_tree_to_missing_subvol,
569 			"snapshot tree points to missing subvolume:\n  %s",
570 			(printbuf_reset(&buf),
571 			 bch2_bkey_val_to_text(&buf, c, st.s_c), buf.buf)) ||
572 	    fsck_err_on(!bch2_snapshot_is_ancestor(c,
573 						le32_to_cpu(subvol.snapshot),
574 						root_id),
575 			trans, snapshot_tree_to_wrong_subvol,
576 			"snapshot tree points to subvolume that does not point to snapshot in this tree:\n  %s",
577 			(printbuf_reset(&buf),
578 			 bch2_bkey_val_to_text(&buf, c, st.s_c), buf.buf)) ||
579 	    fsck_err_on(BCH_SUBVOLUME_SNAP(&subvol),
580 			trans, snapshot_tree_to_snapshot_subvol,
581 			"snapshot tree points to snapshot subvolume:\n  %s",
582 			(printbuf_reset(&buf),
583 			 bch2_bkey_val_to_text(&buf, c, st.s_c), buf.buf))) {
584 		struct bkey_i_snapshot_tree *u;
585 		u32 subvol_id;
586 
587 		ret = bch2_snapshot_tree_master_subvol(trans, root_id, &subvol_id);
588 		bch_err_fn(c, ret);
589 
590 		if (bch2_err_matches(ret, ENOENT)) { /* nothing to be done here */
591 			ret = 0;
592 			goto err;
593 		}
594 
595 		if (ret)
596 			goto err;
597 
598 		u = bch2_bkey_make_mut_typed(trans, iter, &k, 0, snapshot_tree);
599 		ret = PTR_ERR_OR_ZERO(u);
600 		if (ret)
601 			goto err;
602 
603 		u->v.master_subvol = cpu_to_le32(subvol_id);
604 		st = snapshot_tree_i_to_s_c(u);
605 	}
606 err:
607 fsck_err:
608 	printbuf_exit(&buf);
609 	return ret;
610 }
611 
612 /*
613  * For each snapshot_tree, make sure it points to the root of a snapshot tree
614  * and that snapshot entry points back to it, or delete it.
615  *
616  * And, make sure it points to a subvolume within that snapshot tree, or correct
617  * it to point to the oldest subvolume within that snapshot tree.
618  */
619 int bch2_check_snapshot_trees(struct bch_fs *c)
620 {
621 	int ret = bch2_trans_run(c,
622 		for_each_btree_key_commit(trans, iter,
623 			BTREE_ID_snapshot_trees, POS_MIN,
624 			BTREE_ITER_prefetch, k,
625 			NULL, NULL, BCH_TRANS_COMMIT_no_enospc,
626 		check_snapshot_tree(trans, &iter, k)));
627 	bch_err_fn(c, ret);
628 	return ret;
629 }
630 
631 /*
632  * Look up snapshot tree for @tree_id and find root,
633  * make sure @snap_id is a descendent:
634  */
635 static int snapshot_tree_ptr_good(struct btree_trans *trans,
636 				  u32 snap_id, u32 tree_id)
637 {
638 	struct bch_snapshot_tree s_t;
639 	int ret = bch2_snapshot_tree_lookup(trans, tree_id, &s_t);
640 
641 	if (bch2_err_matches(ret, ENOENT))
642 		return 0;
643 	if (ret)
644 		return ret;
645 
646 	return bch2_snapshot_is_ancestor_early(trans->c, snap_id, le32_to_cpu(s_t.root_snapshot));
647 }
648 
649 u32 bch2_snapshot_skiplist_get(struct bch_fs *c, u32 id)
650 {
651 	const struct snapshot_t *s;
652 
653 	if (!id)
654 		return 0;
655 
656 	rcu_read_lock();
657 	s = snapshot_t(c, id);
658 	if (s->parent)
659 		id = bch2_snapshot_nth_parent(c, id, get_random_u32_below(s->depth));
660 	rcu_read_unlock();
661 
662 	return id;
663 }
664 
665 static int snapshot_skiplist_good(struct btree_trans *trans, u32 id, struct bch_snapshot s)
666 {
667 	unsigned i;
668 
669 	for (i = 0; i < 3; i++)
670 		if (!s.parent) {
671 			if (s.skip[i])
672 				return false;
673 		} else {
674 			if (!bch2_snapshot_is_ancestor_early(trans->c, id, le32_to_cpu(s.skip[i])))
675 				return false;
676 		}
677 
678 	return true;
679 }
680 
681 /*
682  * snapshot_tree pointer was incorrect: look up root snapshot node, make sure
683  * its snapshot_tree pointer is correct (allocate new one if necessary), then
684  * update this node's pointer to root node's pointer:
685  */
686 static int snapshot_tree_ptr_repair(struct btree_trans *trans,
687 				    struct btree_iter *iter,
688 				    struct bkey_s_c k,
689 				    struct bch_snapshot *s)
690 {
691 	struct bch_fs *c = trans->c;
692 	struct btree_iter root_iter;
693 	struct bch_snapshot_tree s_t;
694 	struct bkey_s_c_snapshot root;
695 	struct bkey_i_snapshot *u;
696 	u32 root_id = bch2_snapshot_root(c, k.k->p.offset), tree_id;
697 	int ret;
698 
699 	root = bch2_bkey_get_iter_typed(trans, &root_iter,
700 			       BTREE_ID_snapshots, POS(0, root_id),
701 			       BTREE_ITER_with_updates, snapshot);
702 	ret = bkey_err(root);
703 	if (ret)
704 		goto err;
705 
706 	tree_id = le32_to_cpu(root.v->tree);
707 
708 	ret = bch2_snapshot_tree_lookup(trans, tree_id, &s_t);
709 	if (ret && !bch2_err_matches(ret, ENOENT))
710 		return ret;
711 
712 	if (ret || le32_to_cpu(s_t.root_snapshot) != root_id) {
713 		u = bch2_bkey_make_mut_typed(trans, &root_iter, &root.s_c, 0, snapshot);
714 		ret =   PTR_ERR_OR_ZERO(u) ?:
715 			bch2_snapshot_tree_create(trans, root_id,
716 				bch2_snapshot_tree_oldest_subvol(c, root_id),
717 				&tree_id);
718 		if (ret)
719 			goto err;
720 
721 		u->v.tree = cpu_to_le32(tree_id);
722 		if (k.k->p.offset == root_id)
723 			*s = u->v;
724 	}
725 
726 	if (k.k->p.offset != root_id) {
727 		u = bch2_bkey_make_mut_typed(trans, iter, &k, 0, snapshot);
728 		ret = PTR_ERR_OR_ZERO(u);
729 		if (ret)
730 			goto err;
731 
732 		u->v.tree = cpu_to_le32(tree_id);
733 		*s = u->v;
734 	}
735 err:
736 	bch2_trans_iter_exit(trans, &root_iter);
737 	return ret;
738 }
739 
740 static int check_snapshot(struct btree_trans *trans,
741 			  struct btree_iter *iter,
742 			  struct bkey_s_c k)
743 {
744 	struct bch_fs *c = trans->c;
745 	struct bch_snapshot s;
746 	struct bch_subvolume subvol;
747 	struct bch_snapshot v;
748 	struct bkey_i_snapshot *u;
749 	u32 parent_id = bch2_snapshot_parent_early(c, k.k->p.offset);
750 	u32 real_depth;
751 	struct printbuf buf = PRINTBUF;
752 	u32 i, id;
753 	int ret = 0;
754 
755 	if (k.k->type != KEY_TYPE_snapshot)
756 		return 0;
757 
758 	memset(&s, 0, sizeof(s));
759 	memcpy(&s, k.v, min(sizeof(s), bkey_val_bytes(k.k)));
760 
761 	id = le32_to_cpu(s.parent);
762 	if (id) {
763 		ret = bch2_snapshot_lookup(trans, id, &v);
764 		if (bch2_err_matches(ret, ENOENT))
765 			bch_err(c, "snapshot with nonexistent parent:\n  %s",
766 				(bch2_bkey_val_to_text(&buf, c, k), buf.buf));
767 		if (ret)
768 			goto err;
769 
770 		if (le32_to_cpu(v.children[0]) != k.k->p.offset &&
771 		    le32_to_cpu(v.children[1]) != k.k->p.offset) {
772 			bch_err(c, "snapshot parent %u missing pointer to child %llu",
773 				id, k.k->p.offset);
774 			ret = -EINVAL;
775 			goto err;
776 		}
777 	}
778 
779 	for (i = 0; i < 2 && s.children[i]; i++) {
780 		id = le32_to_cpu(s.children[i]);
781 
782 		ret = bch2_snapshot_lookup(trans, id, &v);
783 		if (bch2_err_matches(ret, ENOENT))
784 			bch_err(c, "snapshot node %llu has nonexistent child %u",
785 				k.k->p.offset, id);
786 		if (ret)
787 			goto err;
788 
789 		if (le32_to_cpu(v.parent) != k.k->p.offset) {
790 			bch_err(c, "snapshot child %u has wrong parent (got %u should be %llu)",
791 				id, le32_to_cpu(v.parent), k.k->p.offset);
792 			ret = -EINVAL;
793 			goto err;
794 		}
795 	}
796 
797 	bool should_have_subvol = BCH_SNAPSHOT_SUBVOL(&s) &&
798 		!BCH_SNAPSHOT_DELETED(&s);
799 
800 	if (should_have_subvol) {
801 		id = le32_to_cpu(s.subvol);
802 		ret = bch2_subvolume_get(trans, id, 0, false, &subvol);
803 		if (bch2_err_matches(ret, ENOENT))
804 			bch_err(c, "snapshot points to nonexistent subvolume:\n  %s",
805 				(bch2_bkey_val_to_text(&buf, c, k), buf.buf));
806 		if (ret)
807 			goto err;
808 
809 		if (BCH_SNAPSHOT_SUBVOL(&s) != (le32_to_cpu(subvol.snapshot) == k.k->p.offset)) {
810 			bch_err(c, "snapshot node %llu has wrong BCH_SNAPSHOT_SUBVOL",
811 				k.k->p.offset);
812 			ret = -EINVAL;
813 			goto err;
814 		}
815 	} else {
816 		if (fsck_err_on(s.subvol,
817 				trans, snapshot_should_not_have_subvol,
818 				"snapshot should not point to subvol:\n  %s",
819 				(bch2_bkey_val_to_text(&buf, c, k), buf.buf))) {
820 			u = bch2_bkey_make_mut_typed(trans, iter, &k, 0, snapshot);
821 			ret = PTR_ERR_OR_ZERO(u);
822 			if (ret)
823 				goto err;
824 
825 			u->v.subvol = 0;
826 			s = u->v;
827 		}
828 	}
829 
830 	ret = snapshot_tree_ptr_good(trans, k.k->p.offset, le32_to_cpu(s.tree));
831 	if (ret < 0)
832 		goto err;
833 
834 	if (fsck_err_on(!ret,
835 			trans, snapshot_to_bad_snapshot_tree,
836 			"snapshot points to missing/incorrect tree:\n  %s",
837 			(bch2_bkey_val_to_text(&buf, c, k), buf.buf))) {
838 		ret = snapshot_tree_ptr_repair(trans, iter, k, &s);
839 		if (ret)
840 			goto err;
841 	}
842 	ret = 0;
843 
844 	real_depth = bch2_snapshot_depth(c, parent_id);
845 
846 	if (fsck_err_on(le32_to_cpu(s.depth) != real_depth,
847 			trans, snapshot_bad_depth,
848 			"snapshot with incorrect depth field, should be %u:\n  %s",
849 			real_depth, (bch2_bkey_val_to_text(&buf, c, k), buf.buf))) {
850 		u = bch2_bkey_make_mut_typed(trans, iter, &k, 0, snapshot);
851 		ret = PTR_ERR_OR_ZERO(u);
852 		if (ret)
853 			goto err;
854 
855 		u->v.depth = cpu_to_le32(real_depth);
856 		s = u->v;
857 	}
858 
859 	ret = snapshot_skiplist_good(trans, k.k->p.offset, s);
860 	if (ret < 0)
861 		goto err;
862 
863 	if (fsck_err_on(!ret,
864 			trans, snapshot_bad_skiplist,
865 			"snapshot with bad skiplist field:\n  %s",
866 			(bch2_bkey_val_to_text(&buf, c, k), buf.buf))) {
867 		u = bch2_bkey_make_mut_typed(trans, iter, &k, 0, snapshot);
868 		ret = PTR_ERR_OR_ZERO(u);
869 		if (ret)
870 			goto err;
871 
872 		for (i = 0; i < ARRAY_SIZE(u->v.skip); i++)
873 			u->v.skip[i] = cpu_to_le32(bch2_snapshot_skiplist_get(c, parent_id));
874 
875 		bubble_sort(u->v.skip, ARRAY_SIZE(u->v.skip), cmp_le32);
876 		s = u->v;
877 	}
878 	ret = 0;
879 err:
880 fsck_err:
881 	printbuf_exit(&buf);
882 	return ret;
883 }
884 
885 int bch2_check_snapshots(struct bch_fs *c)
886 {
887 	/*
888 	 * We iterate backwards as checking/fixing the depth field requires that
889 	 * the parent's depth already be correct:
890 	 */
891 	int ret = bch2_trans_run(c,
892 		for_each_btree_key_reverse_commit(trans, iter,
893 				BTREE_ID_snapshots, POS_MAX,
894 				BTREE_ITER_prefetch, k,
895 				NULL, NULL, BCH_TRANS_COMMIT_no_enospc,
896 			check_snapshot(trans, &iter, k)));
897 	bch_err_fn(c, ret);
898 	return ret;
899 }
900 
901 static int check_snapshot_exists(struct btree_trans *trans, u32 id)
902 {
903 	struct bch_fs *c = trans->c;
904 
905 	if (bch2_snapshot_equiv(c, id))
906 		return 0;
907 
908 	/* Do we need to reconstruct the snapshot_tree entry as well? */
909 	struct btree_iter iter;
910 	struct bkey_s_c k;
911 	int ret = 0;
912 	u32 tree_id = 0;
913 
914 	for_each_btree_key_norestart(trans, iter, BTREE_ID_snapshot_trees, POS_MIN,
915 				     0, k, ret) {
916 		if (le32_to_cpu(bkey_s_c_to_snapshot_tree(k).v->root_snapshot) == id) {
917 			tree_id = k.k->p.offset;
918 			break;
919 		}
920 	}
921 	bch2_trans_iter_exit(trans, &iter);
922 
923 	if (ret)
924 		return ret;
925 
926 	if (!tree_id) {
927 		ret = bch2_snapshot_tree_create(trans, id, 0, &tree_id);
928 		if (ret)
929 			return ret;
930 	}
931 
932 	struct bkey_i_snapshot *snapshot = bch2_trans_kmalloc(trans, sizeof(*snapshot));
933 	ret = PTR_ERR_OR_ZERO(snapshot);
934 	if (ret)
935 		return ret;
936 
937 	bkey_snapshot_init(&snapshot->k_i);
938 	snapshot->k.p		= POS(0, id);
939 	snapshot->v.tree	= cpu_to_le32(tree_id);
940 	snapshot->v.btime.lo	= cpu_to_le64(bch2_current_time(c));
941 
942 	for_each_btree_key_norestart(trans, iter, BTREE_ID_subvolumes, POS_MIN,
943 				     0, k, ret) {
944 		if (le32_to_cpu(bkey_s_c_to_subvolume(k).v->snapshot) == id) {
945 			snapshot->v.subvol = cpu_to_le32(k.k->p.offset);
946 			SET_BCH_SNAPSHOT_SUBVOL(&snapshot->v, true);
947 			break;
948 		}
949 	}
950 	bch2_trans_iter_exit(trans, &iter);
951 
952 	return  bch2_btree_insert_trans(trans, BTREE_ID_snapshots, &snapshot->k_i, 0) ?:
953 		bch2_mark_snapshot(trans, BTREE_ID_snapshots, 0,
954 				   bkey_s_c_null, bkey_i_to_s(&snapshot->k_i), 0) ?:
955 		bch2_snapshot_set_equiv(trans, bkey_i_to_s_c(&snapshot->k_i));
956 }
957 
958 /* Figure out which snapshot nodes belong in the same tree: */
959 struct snapshot_tree_reconstruct {
960 	enum btree_id			btree;
961 	struct bpos			cur_pos;
962 	snapshot_id_list		cur_ids;
963 	DARRAY(snapshot_id_list)	trees;
964 };
965 
966 static void snapshot_tree_reconstruct_exit(struct snapshot_tree_reconstruct *r)
967 {
968 	darray_for_each(r->trees, i)
969 		darray_exit(i);
970 	darray_exit(&r->trees);
971 	darray_exit(&r->cur_ids);
972 }
973 
974 static inline bool same_snapshot(struct snapshot_tree_reconstruct *r, struct bpos pos)
975 {
976 	return r->btree == BTREE_ID_inodes
977 		? r->cur_pos.offset == pos.offset
978 		: r->cur_pos.inode == pos.inode;
979 }
980 
981 static inline bool snapshot_id_lists_have_common(snapshot_id_list *l, snapshot_id_list *r)
982 {
983 	darray_for_each(*l, i)
984 		if (snapshot_list_has_id(r, *i))
985 			return true;
986 	return false;
987 }
988 
989 static void snapshot_id_list_to_text(struct printbuf *out, snapshot_id_list *s)
990 {
991 	bool first = true;
992 	darray_for_each(*s, i) {
993 		if (!first)
994 			prt_char(out, ' ');
995 		first = false;
996 		prt_printf(out, "%u", *i);
997 	}
998 }
999 
1000 static int snapshot_tree_reconstruct_next(struct bch_fs *c, struct snapshot_tree_reconstruct *r)
1001 {
1002 	if (r->cur_ids.nr) {
1003 		darray_for_each(r->trees, i)
1004 			if (snapshot_id_lists_have_common(i, &r->cur_ids)) {
1005 				int ret = snapshot_list_merge(c, i, &r->cur_ids);
1006 				if (ret)
1007 					return ret;
1008 				goto out;
1009 			}
1010 		darray_push(&r->trees, r->cur_ids);
1011 		darray_init(&r->cur_ids);
1012 	}
1013 out:
1014 	r->cur_ids.nr = 0;
1015 	return 0;
1016 }
1017 
1018 static int get_snapshot_trees(struct bch_fs *c, struct snapshot_tree_reconstruct *r, struct bpos pos)
1019 {
1020 	if (!same_snapshot(r, pos))
1021 		snapshot_tree_reconstruct_next(c, r);
1022 	r->cur_pos = pos;
1023 	return snapshot_list_add_nodup(c, &r->cur_ids, pos.snapshot);
1024 }
1025 
1026 int bch2_reconstruct_snapshots(struct bch_fs *c)
1027 {
1028 	struct btree_trans *trans = bch2_trans_get(c);
1029 	struct printbuf buf = PRINTBUF;
1030 	struct snapshot_tree_reconstruct r = {};
1031 	int ret = 0;
1032 
1033 	for (unsigned btree = 0; btree < BTREE_ID_NR; btree++) {
1034 		if (btree_type_has_snapshots(btree)) {
1035 			r.btree = btree;
1036 
1037 			ret = for_each_btree_key(trans, iter, btree, POS_MIN,
1038 					BTREE_ITER_all_snapshots|BTREE_ITER_prefetch, k, ({
1039 				get_snapshot_trees(c, &r, k.k->p);
1040 			}));
1041 			if (ret)
1042 				goto err;
1043 
1044 			snapshot_tree_reconstruct_next(c, &r);
1045 		}
1046 	}
1047 
1048 	darray_for_each(r.trees, t) {
1049 		printbuf_reset(&buf);
1050 		snapshot_id_list_to_text(&buf, t);
1051 
1052 		darray_for_each(*t, id) {
1053 			if (fsck_err_on(!bch2_snapshot_equiv(c, *id),
1054 					trans, snapshot_node_missing,
1055 					"snapshot node %u from tree %s missing, recreate?", *id, buf.buf)) {
1056 				if (t->nr > 1) {
1057 					bch_err(c, "cannot reconstruct snapshot trees with multiple nodes");
1058 					ret = -BCH_ERR_fsck_repair_unimplemented;
1059 					goto err;
1060 				}
1061 
1062 				ret = commit_do(trans, NULL, NULL, BCH_TRANS_COMMIT_no_enospc,
1063 						check_snapshot_exists(trans, *id));
1064 				if (ret)
1065 					goto err;
1066 			}
1067 		}
1068 	}
1069 fsck_err:
1070 err:
1071 	bch2_trans_put(trans);
1072 	snapshot_tree_reconstruct_exit(&r);
1073 	printbuf_exit(&buf);
1074 	bch_err_fn(c, ret);
1075 	return ret;
1076 }
1077 
1078 int bch2_check_key_has_snapshot(struct btree_trans *trans,
1079 				struct btree_iter *iter,
1080 				struct bkey_s_c k)
1081 {
1082 	struct bch_fs *c = trans->c;
1083 	struct printbuf buf = PRINTBUF;
1084 	int ret = 0;
1085 
1086 	if (fsck_err_on(!bch2_snapshot_equiv(c, k.k->p.snapshot),
1087 			trans, bkey_in_missing_snapshot,
1088 			"key in missing snapshot %s, delete?",
1089 			(bch2_bkey_val_to_text(&buf, c, k), buf.buf)))
1090 		ret = bch2_btree_delete_at(trans, iter,
1091 					    BTREE_UPDATE_internal_snapshot_node) ?: 1;
1092 fsck_err:
1093 	printbuf_exit(&buf);
1094 	return ret;
1095 }
1096 
1097 /*
1098  * Mark a snapshot as deleted, for future cleanup:
1099  */
1100 int bch2_snapshot_node_set_deleted(struct btree_trans *trans, u32 id)
1101 {
1102 	struct btree_iter iter;
1103 	struct bkey_i_snapshot *s;
1104 	int ret = 0;
1105 
1106 	s = bch2_bkey_get_mut_typed(trans, &iter,
1107 				    BTREE_ID_snapshots, POS(0, id),
1108 				    0, snapshot);
1109 	ret = PTR_ERR_OR_ZERO(s);
1110 	if (unlikely(ret)) {
1111 		bch2_fs_inconsistent_on(bch2_err_matches(ret, ENOENT),
1112 					trans->c, "missing snapshot %u", id);
1113 		return ret;
1114 	}
1115 
1116 	/* already deleted? */
1117 	if (BCH_SNAPSHOT_DELETED(&s->v))
1118 		goto err;
1119 
1120 	SET_BCH_SNAPSHOT_DELETED(&s->v, true);
1121 	SET_BCH_SNAPSHOT_SUBVOL(&s->v, false);
1122 	s->v.subvol = 0;
1123 err:
1124 	bch2_trans_iter_exit(trans, &iter);
1125 	return ret;
1126 }
1127 
1128 static inline void normalize_snapshot_child_pointers(struct bch_snapshot *s)
1129 {
1130 	if (le32_to_cpu(s->children[0]) < le32_to_cpu(s->children[1]))
1131 		swap(s->children[0], s->children[1]);
1132 }
1133 
1134 static int bch2_snapshot_node_delete(struct btree_trans *trans, u32 id)
1135 {
1136 	struct bch_fs *c = trans->c;
1137 	struct btree_iter iter, p_iter = (struct btree_iter) { NULL };
1138 	struct btree_iter c_iter = (struct btree_iter) { NULL };
1139 	struct btree_iter tree_iter = (struct btree_iter) { NULL };
1140 	struct bkey_s_c_snapshot s;
1141 	u32 parent_id, child_id;
1142 	unsigned i;
1143 	int ret = 0;
1144 
1145 	s = bch2_bkey_get_iter_typed(trans, &iter, BTREE_ID_snapshots, POS(0, id),
1146 				     BTREE_ITER_intent, snapshot);
1147 	ret = bkey_err(s);
1148 	bch2_fs_inconsistent_on(bch2_err_matches(ret, ENOENT), c,
1149 				"missing snapshot %u", id);
1150 
1151 	if (ret)
1152 		goto err;
1153 
1154 	BUG_ON(s.v->children[1]);
1155 
1156 	parent_id = le32_to_cpu(s.v->parent);
1157 	child_id = le32_to_cpu(s.v->children[0]);
1158 
1159 	if (parent_id) {
1160 		struct bkey_i_snapshot *parent;
1161 
1162 		parent = bch2_bkey_get_mut_typed(trans, &p_iter,
1163 				     BTREE_ID_snapshots, POS(0, parent_id),
1164 				     0, snapshot);
1165 		ret = PTR_ERR_OR_ZERO(parent);
1166 		bch2_fs_inconsistent_on(bch2_err_matches(ret, ENOENT), c,
1167 					"missing snapshot %u", parent_id);
1168 		if (unlikely(ret))
1169 			goto err;
1170 
1171 		/* find entry in parent->children for node being deleted */
1172 		for (i = 0; i < 2; i++)
1173 			if (le32_to_cpu(parent->v.children[i]) == id)
1174 				break;
1175 
1176 		if (bch2_fs_inconsistent_on(i == 2, c,
1177 					"snapshot %u missing child pointer to %u",
1178 					parent_id, id))
1179 			goto err;
1180 
1181 		parent->v.children[i] = cpu_to_le32(child_id);
1182 
1183 		normalize_snapshot_child_pointers(&parent->v);
1184 	}
1185 
1186 	if (child_id) {
1187 		struct bkey_i_snapshot *child;
1188 
1189 		child = bch2_bkey_get_mut_typed(trans, &c_iter,
1190 				     BTREE_ID_snapshots, POS(0, child_id),
1191 				     0, snapshot);
1192 		ret = PTR_ERR_OR_ZERO(child);
1193 		bch2_fs_inconsistent_on(bch2_err_matches(ret, ENOENT), c,
1194 					"missing snapshot %u", child_id);
1195 		if (unlikely(ret))
1196 			goto err;
1197 
1198 		child->v.parent = cpu_to_le32(parent_id);
1199 
1200 		if (!child->v.parent) {
1201 			child->v.skip[0] = 0;
1202 			child->v.skip[1] = 0;
1203 			child->v.skip[2] = 0;
1204 		}
1205 	}
1206 
1207 	if (!parent_id) {
1208 		/*
1209 		 * We're deleting the root of a snapshot tree: update the
1210 		 * snapshot_tree entry to point to the new root, or delete it if
1211 		 * this is the last snapshot ID in this tree:
1212 		 */
1213 		struct bkey_i_snapshot_tree *s_t;
1214 
1215 		BUG_ON(s.v->children[1]);
1216 
1217 		s_t = bch2_bkey_get_mut_typed(trans, &tree_iter,
1218 				BTREE_ID_snapshot_trees, POS(0, le32_to_cpu(s.v->tree)),
1219 				0, snapshot_tree);
1220 		ret = PTR_ERR_OR_ZERO(s_t);
1221 		if (ret)
1222 			goto err;
1223 
1224 		if (s.v->children[0]) {
1225 			s_t->v.root_snapshot = s.v->children[0];
1226 		} else {
1227 			s_t->k.type = KEY_TYPE_deleted;
1228 			set_bkey_val_u64s(&s_t->k, 0);
1229 		}
1230 	}
1231 
1232 	ret = bch2_btree_delete_at(trans, &iter, 0);
1233 err:
1234 	bch2_trans_iter_exit(trans, &tree_iter);
1235 	bch2_trans_iter_exit(trans, &p_iter);
1236 	bch2_trans_iter_exit(trans, &c_iter);
1237 	bch2_trans_iter_exit(trans, &iter);
1238 	return ret;
1239 }
1240 
1241 static int create_snapids(struct btree_trans *trans, u32 parent, u32 tree,
1242 			  u32 *new_snapids,
1243 			  u32 *snapshot_subvols,
1244 			  unsigned nr_snapids)
1245 {
1246 	struct bch_fs *c = trans->c;
1247 	struct btree_iter iter;
1248 	struct bkey_i_snapshot *n;
1249 	struct bkey_s_c k;
1250 	unsigned i, j;
1251 	u32 depth = bch2_snapshot_depth(c, parent);
1252 	int ret;
1253 
1254 	bch2_trans_iter_init(trans, &iter, BTREE_ID_snapshots,
1255 			     POS_MIN, BTREE_ITER_intent);
1256 	k = bch2_btree_iter_peek(&iter);
1257 	ret = bkey_err(k);
1258 	if (ret)
1259 		goto err;
1260 
1261 	for (i = 0; i < nr_snapids; i++) {
1262 		k = bch2_btree_iter_prev_slot(&iter);
1263 		ret = bkey_err(k);
1264 		if (ret)
1265 			goto err;
1266 
1267 		if (!k.k || !k.k->p.offset) {
1268 			ret = -BCH_ERR_ENOSPC_snapshot_create;
1269 			goto err;
1270 		}
1271 
1272 		n = bch2_bkey_alloc(trans, &iter, 0, snapshot);
1273 		ret = PTR_ERR_OR_ZERO(n);
1274 		if (ret)
1275 			goto err;
1276 
1277 		n->v.flags	= 0;
1278 		n->v.parent	= cpu_to_le32(parent);
1279 		n->v.subvol	= cpu_to_le32(snapshot_subvols[i]);
1280 		n->v.tree	= cpu_to_le32(tree);
1281 		n->v.depth	= cpu_to_le32(depth);
1282 		n->v.btime.lo	= cpu_to_le64(bch2_current_time(c));
1283 		n->v.btime.hi	= 0;
1284 
1285 		for (j = 0; j < ARRAY_SIZE(n->v.skip); j++)
1286 			n->v.skip[j] = cpu_to_le32(bch2_snapshot_skiplist_get(c, parent));
1287 
1288 		bubble_sort(n->v.skip, ARRAY_SIZE(n->v.skip), cmp_le32);
1289 		SET_BCH_SNAPSHOT_SUBVOL(&n->v, true);
1290 
1291 		ret = __bch2_mark_snapshot(trans, BTREE_ID_snapshots, 0,
1292 					 bkey_s_c_null, bkey_i_to_s_c(&n->k_i), 0);
1293 		if (ret)
1294 			goto err;
1295 
1296 		new_snapids[i]	= iter.pos.offset;
1297 
1298 		mutex_lock(&c->snapshot_table_lock);
1299 		snapshot_t_mut(c, new_snapids[i])->equiv = new_snapids[i];
1300 		mutex_unlock(&c->snapshot_table_lock);
1301 	}
1302 err:
1303 	bch2_trans_iter_exit(trans, &iter);
1304 	return ret;
1305 }
1306 
1307 /*
1308  * Create new snapshot IDs as children of an existing snapshot ID:
1309  */
1310 static int bch2_snapshot_node_create_children(struct btree_trans *trans, u32 parent,
1311 			      u32 *new_snapids,
1312 			      u32 *snapshot_subvols,
1313 			      unsigned nr_snapids)
1314 {
1315 	struct btree_iter iter;
1316 	struct bkey_i_snapshot *n_parent;
1317 	int ret = 0;
1318 
1319 	n_parent = bch2_bkey_get_mut_typed(trans, &iter,
1320 			BTREE_ID_snapshots, POS(0, parent),
1321 			0, snapshot);
1322 	ret = PTR_ERR_OR_ZERO(n_parent);
1323 	if (unlikely(ret)) {
1324 		if (bch2_err_matches(ret, ENOENT))
1325 			bch_err(trans->c, "snapshot %u not found", parent);
1326 		return ret;
1327 	}
1328 
1329 	if (n_parent->v.children[0] || n_parent->v.children[1]) {
1330 		bch_err(trans->c, "Trying to add child snapshot nodes to parent that already has children");
1331 		ret = -EINVAL;
1332 		goto err;
1333 	}
1334 
1335 	ret = create_snapids(trans, parent, le32_to_cpu(n_parent->v.tree),
1336 			     new_snapids, snapshot_subvols, nr_snapids);
1337 	if (ret)
1338 		goto err;
1339 
1340 	n_parent->v.children[0] = cpu_to_le32(new_snapids[0]);
1341 	n_parent->v.children[1] = cpu_to_le32(new_snapids[1]);
1342 	n_parent->v.subvol = 0;
1343 	SET_BCH_SNAPSHOT_SUBVOL(&n_parent->v, false);
1344 err:
1345 	bch2_trans_iter_exit(trans, &iter);
1346 	return ret;
1347 }
1348 
1349 /*
1350  * Create a snapshot node that is the root of a new tree:
1351  */
1352 static int bch2_snapshot_node_create_tree(struct btree_trans *trans,
1353 			      u32 *new_snapids,
1354 			      u32 *snapshot_subvols,
1355 			      unsigned nr_snapids)
1356 {
1357 	struct bkey_i_snapshot_tree *n_tree;
1358 	int ret;
1359 
1360 	n_tree = __bch2_snapshot_tree_create(trans);
1361 	ret =   PTR_ERR_OR_ZERO(n_tree) ?:
1362 		create_snapids(trans, 0, n_tree->k.p.offset,
1363 			     new_snapids, snapshot_subvols, nr_snapids);
1364 	if (ret)
1365 		return ret;
1366 
1367 	n_tree->v.master_subvol	= cpu_to_le32(snapshot_subvols[0]);
1368 	n_tree->v.root_snapshot	= cpu_to_le32(new_snapids[0]);
1369 	return 0;
1370 }
1371 
1372 int bch2_snapshot_node_create(struct btree_trans *trans, u32 parent,
1373 			      u32 *new_snapids,
1374 			      u32 *snapshot_subvols,
1375 			      unsigned nr_snapids)
1376 {
1377 	BUG_ON((parent == 0) != (nr_snapids == 1));
1378 	BUG_ON((parent != 0) != (nr_snapids == 2));
1379 
1380 	return parent
1381 		? bch2_snapshot_node_create_children(trans, parent,
1382 				new_snapids, snapshot_subvols, nr_snapids)
1383 		: bch2_snapshot_node_create_tree(trans,
1384 				new_snapids, snapshot_subvols, nr_snapids);
1385 
1386 }
1387 
1388 /*
1389  * If we have an unlinked inode in an internal snapshot node, and the inode
1390  * really has been deleted in all child snapshots, how does this get cleaned up?
1391  *
1392  * first there is the problem of how keys that have been overwritten in all
1393  * child snapshots get deleted (unimplemented?), but inodes may perhaps be
1394  * special?
1395  *
1396  * also: unlinked inode in internal snapshot appears to not be getting deleted
1397  * correctly if inode doesn't exist in leaf snapshots
1398  *
1399  * solution:
1400  *
1401  * for a key in an interior snapshot node that needs work to be done that
1402  * requires it to be mutated: iterate over all descendent leaf nodes and copy
1403  * that key to snapshot leaf nodes, where we can mutate it
1404  */
1405 
1406 static int delete_dead_snapshots_process_key(struct btree_trans *trans,
1407 			       struct btree_iter *iter,
1408 			       struct bkey_s_c k,
1409 			       snapshot_id_list *deleted,
1410 			       snapshot_id_list *equiv_seen,
1411 			       struct bpos *last_pos)
1412 {
1413 	int ret = bch2_check_key_has_snapshot(trans, iter, k);
1414 	if (ret)
1415 		return ret < 0 ? ret : 0;
1416 
1417 	struct bch_fs *c = trans->c;
1418 	u32 equiv = bch2_snapshot_equiv(c, k.k->p.snapshot);
1419 	if (!equiv) /* key for invalid snapshot node, but we chose not to delete */
1420 		return 0;
1421 
1422 	if (!bkey_eq(k.k->p, *last_pos))
1423 		equiv_seen->nr = 0;
1424 
1425 	if (snapshot_list_has_id(deleted, k.k->p.snapshot))
1426 		return bch2_btree_delete_at(trans, iter,
1427 					    BTREE_UPDATE_internal_snapshot_node);
1428 
1429 	if (!bpos_eq(*last_pos, k.k->p) &&
1430 	    snapshot_list_has_id(equiv_seen, equiv))
1431 		return bch2_btree_delete_at(trans, iter,
1432 					    BTREE_UPDATE_internal_snapshot_node);
1433 
1434 	*last_pos = k.k->p;
1435 
1436 	ret = snapshot_list_add_nodup(c, equiv_seen, equiv);
1437 	if (ret)
1438 		return ret;
1439 
1440 	/*
1441 	 * When we have a linear chain of snapshot nodes, we consider
1442 	 * those to form an equivalence class: we're going to collapse
1443 	 * them all down to a single node, and keep the leaf-most node -
1444 	 * which has the same id as the equivalence class id.
1445 	 *
1446 	 * If there are multiple keys in different snapshots at the same
1447 	 * position, we're only going to keep the one in the newest
1448 	 * snapshot (we delete the others above) - the rest have been
1449 	 * overwritten and are redundant, and for the key we're going to keep we
1450 	 * need to move it to the equivalance class ID if it's not there
1451 	 * already.
1452 	 */
1453 	if (equiv != k.k->p.snapshot) {
1454 		struct bkey_i *new = bch2_bkey_make_mut_noupdate(trans, k);
1455 		int ret = PTR_ERR_OR_ZERO(new);
1456 		if (ret)
1457 			return ret;
1458 
1459 		new->k.p.snapshot = equiv;
1460 
1461 		struct btree_iter new_iter;
1462 		bch2_trans_iter_init(trans, &new_iter, iter->btree_id, new->k.p,
1463 				     BTREE_ITER_all_snapshots|
1464 				     BTREE_ITER_cached|
1465 				     BTREE_ITER_intent);
1466 
1467 		ret =   bch2_btree_iter_traverse(&new_iter) ?:
1468 			bch2_trans_update(trans, &new_iter, new,
1469 					BTREE_UPDATE_internal_snapshot_node) ?:
1470 			bch2_btree_delete_at(trans, iter,
1471 					BTREE_UPDATE_internal_snapshot_node);
1472 		bch2_trans_iter_exit(trans, &new_iter);
1473 		if (ret)
1474 			return ret;
1475 	}
1476 
1477 	return 0;
1478 }
1479 
1480 static int bch2_snapshot_needs_delete(struct btree_trans *trans, struct bkey_s_c k)
1481 {
1482 	struct bkey_s_c_snapshot snap;
1483 	u32 children[2];
1484 	int ret;
1485 
1486 	if (k.k->type != KEY_TYPE_snapshot)
1487 		return 0;
1488 
1489 	snap = bkey_s_c_to_snapshot(k);
1490 	if (BCH_SNAPSHOT_DELETED(snap.v) ||
1491 	    BCH_SNAPSHOT_SUBVOL(snap.v))
1492 		return 0;
1493 
1494 	children[0] = le32_to_cpu(snap.v->children[0]);
1495 	children[1] = le32_to_cpu(snap.v->children[1]);
1496 
1497 	ret   = bch2_snapshot_live(trans, children[0]) ?:
1498 		bch2_snapshot_live(trans, children[1]);
1499 	if (ret < 0)
1500 		return ret;
1501 	return !ret;
1502 }
1503 
1504 /*
1505  * For a given snapshot, if it doesn't have a subvolume that points to it, and
1506  * it doesn't have child snapshot nodes - it's now redundant and we can mark it
1507  * as deleted.
1508  */
1509 static int bch2_delete_redundant_snapshot(struct btree_trans *trans, struct bkey_s_c k)
1510 {
1511 	int ret = bch2_snapshot_needs_delete(trans, k);
1512 
1513 	return ret <= 0
1514 		? ret
1515 		: bch2_snapshot_node_set_deleted(trans, k.k->p.offset);
1516 }
1517 
1518 static inline u32 bch2_snapshot_nth_parent_skip(struct bch_fs *c, u32 id, u32 n,
1519 						snapshot_id_list *skip)
1520 {
1521 	rcu_read_lock();
1522 	while (snapshot_list_has_id(skip, id))
1523 		id = __bch2_snapshot_parent(c, id);
1524 
1525 	while (n--) {
1526 		do {
1527 			id = __bch2_snapshot_parent(c, id);
1528 		} while (snapshot_list_has_id(skip, id));
1529 	}
1530 	rcu_read_unlock();
1531 
1532 	return id;
1533 }
1534 
1535 static int bch2_fix_child_of_deleted_snapshot(struct btree_trans *trans,
1536 					      struct btree_iter *iter, struct bkey_s_c k,
1537 					      snapshot_id_list *deleted)
1538 {
1539 	struct bch_fs *c = trans->c;
1540 	u32 nr_deleted_ancestors = 0;
1541 	struct bkey_i_snapshot *s;
1542 	int ret;
1543 
1544 	if (k.k->type != KEY_TYPE_snapshot)
1545 		return 0;
1546 
1547 	if (snapshot_list_has_id(deleted, k.k->p.offset))
1548 		return 0;
1549 
1550 	s = bch2_bkey_make_mut_noupdate_typed(trans, k, snapshot);
1551 	ret = PTR_ERR_OR_ZERO(s);
1552 	if (ret)
1553 		return ret;
1554 
1555 	darray_for_each(*deleted, i)
1556 		nr_deleted_ancestors += bch2_snapshot_is_ancestor(c, s->k.p.offset, *i);
1557 
1558 	if (!nr_deleted_ancestors)
1559 		return 0;
1560 
1561 	le32_add_cpu(&s->v.depth, -nr_deleted_ancestors);
1562 
1563 	if (!s->v.depth) {
1564 		s->v.skip[0] = 0;
1565 		s->v.skip[1] = 0;
1566 		s->v.skip[2] = 0;
1567 	} else {
1568 		u32 depth = le32_to_cpu(s->v.depth);
1569 		u32 parent = bch2_snapshot_parent(c, s->k.p.offset);
1570 
1571 		for (unsigned j = 0; j < ARRAY_SIZE(s->v.skip); j++) {
1572 			u32 id = le32_to_cpu(s->v.skip[j]);
1573 
1574 			if (snapshot_list_has_id(deleted, id)) {
1575 				id = bch2_snapshot_nth_parent_skip(c,
1576 							parent,
1577 							depth > 1
1578 							? get_random_u32_below(depth - 1)
1579 							: 0,
1580 							deleted);
1581 				s->v.skip[j] = cpu_to_le32(id);
1582 			}
1583 		}
1584 
1585 		bubble_sort(s->v.skip, ARRAY_SIZE(s->v.skip), cmp_le32);
1586 	}
1587 
1588 	return bch2_trans_update(trans, iter, &s->k_i, 0);
1589 }
1590 
1591 int bch2_delete_dead_snapshots(struct bch_fs *c)
1592 {
1593 	struct btree_trans *trans;
1594 	snapshot_id_list deleted = { 0 };
1595 	snapshot_id_list deleted_interior = { 0 };
1596 	int ret = 0;
1597 
1598 	if (!test_and_clear_bit(BCH_FS_need_delete_dead_snapshots, &c->flags))
1599 		return 0;
1600 
1601 	trans = bch2_trans_get(c);
1602 
1603 	/*
1604 	 * For every snapshot node: If we have no live children and it's not
1605 	 * pointed to by a subvolume, delete it:
1606 	 */
1607 	ret = for_each_btree_key_commit(trans, iter, BTREE_ID_snapshots,
1608 			POS_MIN, 0, k,
1609 			NULL, NULL, 0,
1610 		bch2_delete_redundant_snapshot(trans, k));
1611 	bch_err_msg(c, ret, "deleting redundant snapshots");
1612 	if (ret)
1613 		goto err;
1614 
1615 	ret = for_each_btree_key(trans, iter, BTREE_ID_snapshots,
1616 				 POS_MIN, 0, k,
1617 		bch2_snapshot_set_equiv(trans, k));
1618 	bch_err_msg(c, ret, "in bch2_snapshots_set_equiv");
1619 	if (ret)
1620 		goto err;
1621 
1622 	ret = for_each_btree_key(trans, iter, BTREE_ID_snapshots,
1623 				 POS_MIN, 0, k, ({
1624 		if (k.k->type != KEY_TYPE_snapshot)
1625 			continue;
1626 
1627 		BCH_SNAPSHOT_DELETED(bkey_s_c_to_snapshot(k).v)
1628 			? snapshot_list_add(c, &deleted, k.k->p.offset)
1629 			: 0;
1630 	}));
1631 	bch_err_msg(c, ret, "walking snapshots");
1632 	if (ret)
1633 		goto err;
1634 
1635 	for (unsigned btree = 0; btree < BTREE_ID_NR; btree++) {
1636 		struct bpos last_pos = POS_MIN;
1637 		snapshot_id_list equiv_seen = { 0 };
1638 		struct disk_reservation res = { 0 };
1639 
1640 		if (!btree_type_has_snapshots(btree))
1641 			continue;
1642 
1643 		ret = for_each_btree_key_commit(trans, iter,
1644 				btree, POS_MIN,
1645 				BTREE_ITER_prefetch|BTREE_ITER_all_snapshots, k,
1646 				&res, NULL, BCH_TRANS_COMMIT_no_enospc,
1647 			delete_dead_snapshots_process_key(trans, &iter, k, &deleted,
1648 							  &equiv_seen, &last_pos));
1649 
1650 		bch2_disk_reservation_put(c, &res);
1651 		darray_exit(&equiv_seen);
1652 
1653 		bch_err_msg(c, ret, "deleting keys from dying snapshots");
1654 		if (ret)
1655 			goto err;
1656 	}
1657 
1658 	bch2_trans_unlock(trans);
1659 	down_write(&c->snapshot_create_lock);
1660 
1661 	ret = for_each_btree_key(trans, iter, BTREE_ID_snapshots,
1662 				 POS_MIN, 0, k, ({
1663 		u32 snapshot = k.k->p.offset;
1664 		u32 equiv = bch2_snapshot_equiv(c, snapshot);
1665 
1666 		equiv != snapshot
1667 			? snapshot_list_add(c, &deleted_interior, snapshot)
1668 			: 0;
1669 	}));
1670 
1671 	bch_err_msg(c, ret, "walking snapshots");
1672 	if (ret)
1673 		goto err_create_lock;
1674 
1675 	/*
1676 	 * Fixing children of deleted snapshots can't be done completely
1677 	 * atomically, if we crash between here and when we delete the interior
1678 	 * nodes some depth fields will be off:
1679 	 */
1680 	ret = for_each_btree_key_commit(trans, iter, BTREE_ID_snapshots, POS_MIN,
1681 				  BTREE_ITER_intent, k,
1682 				  NULL, NULL, BCH_TRANS_COMMIT_no_enospc,
1683 		bch2_fix_child_of_deleted_snapshot(trans, &iter, k, &deleted_interior));
1684 	if (ret)
1685 		goto err_create_lock;
1686 
1687 	darray_for_each(deleted, i) {
1688 		ret = commit_do(trans, NULL, NULL, 0,
1689 			bch2_snapshot_node_delete(trans, *i));
1690 		bch_err_msg(c, ret, "deleting snapshot %u", *i);
1691 		if (ret)
1692 			goto err_create_lock;
1693 	}
1694 
1695 	darray_for_each(deleted_interior, i) {
1696 		ret = commit_do(trans, NULL, NULL, 0,
1697 			bch2_snapshot_node_delete(trans, *i));
1698 		bch_err_msg(c, ret, "deleting snapshot %u", *i);
1699 		if (ret)
1700 			goto err_create_lock;
1701 	}
1702 err_create_lock:
1703 	up_write(&c->snapshot_create_lock);
1704 err:
1705 	darray_exit(&deleted_interior);
1706 	darray_exit(&deleted);
1707 	bch2_trans_put(trans);
1708 	bch_err_fn(c, ret);
1709 	return ret;
1710 }
1711 
1712 void bch2_delete_dead_snapshots_work(struct work_struct *work)
1713 {
1714 	struct bch_fs *c = container_of(work, struct bch_fs, snapshot_delete_work);
1715 
1716 	set_worker_desc("bcachefs-delete-dead-snapshots/%s", c->name);
1717 
1718 	bch2_delete_dead_snapshots(c);
1719 	bch2_write_ref_put(c, BCH_WRITE_REF_delete_dead_snapshots);
1720 }
1721 
1722 void bch2_delete_dead_snapshots_async(struct bch_fs *c)
1723 {
1724 	if (bch2_write_ref_tryget(c, BCH_WRITE_REF_delete_dead_snapshots) &&
1725 	    !queue_work(c->write_ref_wq, &c->snapshot_delete_work))
1726 		bch2_write_ref_put(c, BCH_WRITE_REF_delete_dead_snapshots);
1727 }
1728 
1729 int __bch2_key_has_snapshot_overwrites(struct btree_trans *trans,
1730 				       enum btree_id id,
1731 				       struct bpos pos)
1732 {
1733 	struct bch_fs *c = trans->c;
1734 	struct btree_iter iter;
1735 	struct bkey_s_c k;
1736 	int ret;
1737 
1738 	bch2_trans_iter_init(trans, &iter, id, pos,
1739 			     BTREE_ITER_not_extents|
1740 			     BTREE_ITER_all_snapshots);
1741 	while (1) {
1742 		k = bch2_btree_iter_prev(&iter);
1743 		ret = bkey_err(k);
1744 		if (ret)
1745 			break;
1746 
1747 		if (!k.k)
1748 			break;
1749 
1750 		if (!bkey_eq(pos, k.k->p))
1751 			break;
1752 
1753 		if (bch2_snapshot_is_ancestor(c, k.k->p.snapshot, pos.snapshot)) {
1754 			ret = 1;
1755 			break;
1756 		}
1757 	}
1758 	bch2_trans_iter_exit(trans, &iter);
1759 
1760 	return ret;
1761 }
1762 
1763 static int bch2_check_snapshot_needs_deletion(struct btree_trans *trans, struct bkey_s_c k)
1764 {
1765 	struct bch_fs *c = trans->c;
1766 	struct bkey_s_c_snapshot snap;
1767 	int ret = 0;
1768 
1769 	if (k.k->type != KEY_TYPE_snapshot)
1770 		return 0;
1771 
1772 	snap = bkey_s_c_to_snapshot(k);
1773 	if (BCH_SNAPSHOT_DELETED(snap.v) ||
1774 	    bch2_snapshot_equiv(c, k.k->p.offset) != k.k->p.offset ||
1775 	    (ret = bch2_snapshot_needs_delete(trans, k)) > 0) {
1776 		set_bit(BCH_FS_need_delete_dead_snapshots, &c->flags);
1777 		return 0;
1778 	}
1779 
1780 	return ret;
1781 }
1782 
1783 int bch2_snapshots_read(struct bch_fs *c)
1784 {
1785 	int ret = bch2_trans_run(c,
1786 		for_each_btree_key(trans, iter, BTREE_ID_snapshots,
1787 				   POS_MIN, 0, k,
1788 			__bch2_mark_snapshot(trans, BTREE_ID_snapshots, 0, bkey_s_c_null, k, 0) ?:
1789 			bch2_snapshot_set_equiv(trans, k) ?:
1790 			bch2_check_snapshot_needs_deletion(trans, k)) ?:
1791 		for_each_btree_key(trans, iter, BTREE_ID_snapshots,
1792 				   POS_MIN, 0, k,
1793 			   (set_is_ancestor_bitmap(c, k.k->p.offset), 0)));
1794 	bch_err_fn(c, ret);
1795 
1796 	/*
1797 	 * It's important that we check if we need to reconstruct snapshots
1798 	 * before going RW, so we mark that pass as required in the superblock -
1799 	 * otherwise, we could end up deleting keys with missing snapshot nodes
1800 	 * instead
1801 	 */
1802 	BUG_ON(!test_bit(BCH_FS_new_fs, &c->flags) &&
1803 	       test_bit(BCH_FS_may_go_rw, &c->flags));
1804 
1805 	if (bch2_err_matches(ret, EIO) ||
1806 	    (c->sb.btrees_lost_data & BIT_ULL(BTREE_ID_snapshots)))
1807 		ret = bch2_run_explicit_recovery_pass_persistent(c, BCH_RECOVERY_PASS_reconstruct_snapshots);
1808 
1809 	return ret;
1810 }
1811 
1812 void bch2_fs_snapshots_exit(struct bch_fs *c)
1813 {
1814 	kvfree(rcu_dereference_protected(c->snapshots, true));
1815 }
1816