xref: /linux/fs/bcachefs/recovery.c (revision d8310914848223de7ec04d55bd15f013f0dad803)
1 // SPDX-License-Identifier: GPL-2.0
2 
3 #include "bcachefs.h"
4 #include "backpointers.h"
5 #include "bkey_buf.h"
6 #include "alloc_background.h"
7 #include "btree_gc.h"
8 #include "btree_journal_iter.h"
9 #include "btree_update.h"
10 #include "btree_update_interior.h"
11 #include "btree_io.h"
12 #include "buckets.h"
13 #include "dirent.h"
14 #include "ec.h"
15 #include "errcode.h"
16 #include "error.h"
17 #include "fs-common.h"
18 #include "fsck.h"
19 #include "journal_io.h"
20 #include "journal_reclaim.h"
21 #include "journal_seq_blacklist.h"
22 #include "lru.h"
23 #include "logged_ops.h"
24 #include "move.h"
25 #include "quota.h"
26 #include "rebalance.h"
27 #include "recovery.h"
28 #include "replicas.h"
29 #include "sb-clean.h"
30 #include "sb-downgrade.h"
31 #include "snapshot.h"
32 #include "subvolume.h"
33 #include "super-io.h"
34 
35 #include <linux/sort.h>
36 #include <linux/stat.h>
37 
38 #define QSTR(n) { { { .len = strlen(n) } }, .name = n }
39 
40 static bool btree_id_is_alloc(enum btree_id id)
41 {
42 	switch (id) {
43 	case BTREE_ID_alloc:
44 	case BTREE_ID_backpointers:
45 	case BTREE_ID_need_discard:
46 	case BTREE_ID_freespace:
47 	case BTREE_ID_bucket_gens:
48 		return true;
49 	default:
50 		return false;
51 	}
52 }
53 
54 /* for -o reconstruct_alloc: */
55 static void drop_alloc_keys(struct journal_keys *keys)
56 {
57 	size_t src, dst;
58 
59 	for (src = 0, dst = 0; src < keys->nr; src++)
60 		if (!btree_id_is_alloc(keys->d[src].btree_id))
61 			keys->d[dst++] = keys->d[src];
62 
63 	keys->nr = dst;
64 }
65 
66 /*
67  * Btree node pointers have a field to stack a pointer to the in memory btree
68  * node; we need to zero out this field when reading in btree nodes, or when
69  * reading in keys from the journal:
70  */
71 static void zero_out_btree_mem_ptr(struct journal_keys *keys)
72 {
73 	struct journal_key *i;
74 
75 	for (i = keys->d; i < keys->d + keys->nr; i++)
76 		if (i->k->k.type == KEY_TYPE_btree_ptr_v2)
77 			bkey_i_to_btree_ptr_v2(i->k)->v.mem_ptr = 0;
78 }
79 
80 /* journal replay: */
81 
82 static void replay_now_at(struct journal *j, u64 seq)
83 {
84 	BUG_ON(seq < j->replay_journal_seq);
85 
86 	seq = min(seq, j->replay_journal_seq_end);
87 
88 	while (j->replay_journal_seq < seq)
89 		bch2_journal_pin_put(j, j->replay_journal_seq++);
90 }
91 
92 static int bch2_journal_replay_key(struct btree_trans *trans,
93 				   struct journal_key *k)
94 {
95 	struct btree_iter iter;
96 	unsigned iter_flags =
97 		BTREE_ITER_INTENT|
98 		BTREE_ITER_NOT_EXTENTS;
99 	unsigned update_flags = BTREE_TRIGGER_NORUN;
100 	int ret;
101 
102 	if (k->overwritten)
103 		return 0;
104 
105 	trans->journal_res.seq = k->journal_seq;
106 
107 	/*
108 	 * BTREE_UPDATE_KEY_CACHE_RECLAIM disables key cache lookup/update to
109 	 * keep the key cache coherent with the underlying btree. Nothing
110 	 * besides the allocator is doing updates yet so we don't need key cache
111 	 * coherency for non-alloc btrees, and key cache fills for snapshots
112 	 * btrees use BTREE_ITER_FILTER_SNAPSHOTS, which isn't available until
113 	 * the snapshots recovery pass runs.
114 	 */
115 	if (!k->level && k->btree_id == BTREE_ID_alloc)
116 		iter_flags |= BTREE_ITER_CACHED;
117 	else
118 		update_flags |= BTREE_UPDATE_KEY_CACHE_RECLAIM;
119 
120 	bch2_trans_node_iter_init(trans, &iter, k->btree_id, k->k->k.p,
121 				  BTREE_MAX_DEPTH, k->level,
122 				  iter_flags);
123 	ret = bch2_btree_iter_traverse(&iter);
124 	if (ret)
125 		goto out;
126 
127 	/* Must be checked with btree locked: */
128 	if (k->overwritten)
129 		goto out;
130 
131 	ret = bch2_trans_update(trans, &iter, k->k, update_flags);
132 out:
133 	bch2_trans_iter_exit(trans, &iter);
134 	return ret;
135 }
136 
137 static int journal_sort_seq_cmp(const void *_l, const void *_r)
138 {
139 	const struct journal_key *l = *((const struct journal_key **)_l);
140 	const struct journal_key *r = *((const struct journal_key **)_r);
141 
142 	return cmp_int(l->journal_seq, r->journal_seq);
143 }
144 
145 static int bch2_journal_replay(struct bch_fs *c)
146 {
147 	struct journal_keys *keys = &c->journal_keys;
148 	DARRAY(struct journal_key *) keys_sorted = { 0 };
149 	struct journal *j = &c->journal;
150 	u64 start_seq	= c->journal_replay_seq_start;
151 	u64 end_seq	= c->journal_replay_seq_start;
152 	struct btree_trans *trans = bch2_trans_get(c);
153 	int ret = 0;
154 
155 	if (keys->nr) {
156 		ret = bch2_journal_log_msg(c, "Starting journal replay (%zu keys in entries %llu-%llu)",
157 					   keys->nr, start_seq, end_seq);
158 		if (ret)
159 			goto err;
160 	}
161 
162 	BUG_ON(!atomic_read(&keys->ref));
163 
164 	/*
165 	 * First, attempt to replay keys in sorted order. This is more
166 	 * efficient - better locality of btree access -  but some might fail if
167 	 * that would cause a journal deadlock.
168 	 */
169 	for (size_t i = 0; i < keys->nr; i++) {
170 		cond_resched();
171 
172 		struct journal_key *k = keys->d + i;
173 
174 		/* Skip fastpath if we're low on space in the journal */
175 		ret = c->journal.watermark ? -1 :
176 			commit_do(trans, NULL, NULL,
177 				  BCH_TRANS_COMMIT_no_enospc|
178 				  BCH_TRANS_COMMIT_journal_reclaim|
179 				  (!k->allocated ? BCH_TRANS_COMMIT_no_journal_res : 0),
180 			     bch2_journal_replay_key(trans, k));
181 		BUG_ON(!ret && !k->overwritten);
182 		if (ret) {
183 			ret = darray_push(&keys_sorted, k);
184 			if (ret)
185 				goto err;
186 		}
187 	}
188 
189 	/*
190 	 * Now, replay any remaining keys in the order in which they appear in
191 	 * the journal, unpinning those journal entries as we go:
192 	 */
193 	sort(keys_sorted.data, keys_sorted.nr,
194 	     sizeof(keys_sorted.data[0]),
195 	     journal_sort_seq_cmp, NULL);
196 
197 	darray_for_each(keys_sorted, kp) {
198 		cond_resched();
199 
200 		struct journal_key *k = *kp;
201 
202 		replay_now_at(j, k->journal_seq);
203 
204 		ret = commit_do(trans, NULL, NULL,
205 				BCH_TRANS_COMMIT_no_enospc|
206 				(!k->allocated
207 				 ? BCH_TRANS_COMMIT_no_journal_res|BCH_WATERMARK_reclaim
208 				 : 0),
209 			     bch2_journal_replay_key(trans, k));
210 		bch_err_msg(c, ret, "while replaying key at btree %s level %u:",
211 			    bch2_btree_id_str(k->btree_id), k->level);
212 		if (ret)
213 			goto err;
214 
215 		BUG_ON(!k->overwritten);
216 	}
217 
218 	/*
219 	 * We need to put our btree_trans before calling flush_all_pins(), since
220 	 * that will use a btree_trans internally
221 	 */
222 	bch2_trans_put(trans);
223 	trans = NULL;
224 
225 	if (!c->opts.keep_journal)
226 		bch2_journal_keys_put_initial(c);
227 
228 	replay_now_at(j, j->replay_journal_seq_end);
229 	j->replay_journal_seq = 0;
230 
231 	bch2_journal_set_replay_done(j);
232 
233 	if (keys->nr)
234 		bch2_journal_log_msg(c, "journal replay finished");
235 err:
236 	if (trans)
237 		bch2_trans_put(trans);
238 	darray_exit(&keys_sorted);
239 	bch_err_fn(c, ret);
240 	return ret;
241 }
242 
243 /* journal replay early: */
244 
245 static int journal_replay_entry_early(struct bch_fs *c,
246 				      struct jset_entry *entry)
247 {
248 	int ret = 0;
249 
250 	switch (entry->type) {
251 	case BCH_JSET_ENTRY_btree_root: {
252 		struct btree_root *r;
253 
254 		while (entry->btree_id >= c->btree_roots_extra.nr + BTREE_ID_NR) {
255 			ret = darray_push(&c->btree_roots_extra, (struct btree_root) { NULL });
256 			if (ret)
257 				return ret;
258 		}
259 
260 		r = bch2_btree_id_root(c, entry->btree_id);
261 
262 		if (entry->u64s) {
263 			r->level = entry->level;
264 			bkey_copy(&r->key, (struct bkey_i *) entry->start);
265 			r->error = 0;
266 		} else {
267 			r->error = -EIO;
268 		}
269 		r->alive = true;
270 		break;
271 	}
272 	case BCH_JSET_ENTRY_usage: {
273 		struct jset_entry_usage *u =
274 			container_of(entry, struct jset_entry_usage, entry);
275 
276 		switch (entry->btree_id) {
277 		case BCH_FS_USAGE_reserved:
278 			if (entry->level < BCH_REPLICAS_MAX)
279 				c->usage_base->persistent_reserved[entry->level] =
280 					le64_to_cpu(u->v);
281 			break;
282 		case BCH_FS_USAGE_inodes:
283 			c->usage_base->b.nr_inodes = le64_to_cpu(u->v);
284 			break;
285 		case BCH_FS_USAGE_key_version:
286 			atomic64_set(&c->key_version,
287 				     le64_to_cpu(u->v));
288 			break;
289 		}
290 
291 		break;
292 	}
293 	case BCH_JSET_ENTRY_data_usage: {
294 		struct jset_entry_data_usage *u =
295 			container_of(entry, struct jset_entry_data_usage, entry);
296 
297 		ret = bch2_replicas_set_usage(c, &u->r,
298 					      le64_to_cpu(u->v));
299 		break;
300 	}
301 	case BCH_JSET_ENTRY_dev_usage: {
302 		struct jset_entry_dev_usage *u =
303 			container_of(entry, struct jset_entry_dev_usage, entry);
304 		struct bch_dev *ca = bch_dev_bkey_exists(c, le32_to_cpu(u->dev));
305 		unsigned i, nr_types = jset_entry_dev_usage_nr_types(u);
306 
307 		for (i = 0; i < min_t(unsigned, nr_types, BCH_DATA_NR); i++) {
308 			ca->usage_base->d[i].buckets	= le64_to_cpu(u->d[i].buckets);
309 			ca->usage_base->d[i].sectors	= le64_to_cpu(u->d[i].sectors);
310 			ca->usage_base->d[i].fragmented	= le64_to_cpu(u->d[i].fragmented);
311 		}
312 
313 		break;
314 	}
315 	case BCH_JSET_ENTRY_blacklist: {
316 		struct jset_entry_blacklist *bl_entry =
317 			container_of(entry, struct jset_entry_blacklist, entry);
318 
319 		ret = bch2_journal_seq_blacklist_add(c,
320 				le64_to_cpu(bl_entry->seq),
321 				le64_to_cpu(bl_entry->seq) + 1);
322 		break;
323 	}
324 	case BCH_JSET_ENTRY_blacklist_v2: {
325 		struct jset_entry_blacklist_v2 *bl_entry =
326 			container_of(entry, struct jset_entry_blacklist_v2, entry);
327 
328 		ret = bch2_journal_seq_blacklist_add(c,
329 				le64_to_cpu(bl_entry->start),
330 				le64_to_cpu(bl_entry->end) + 1);
331 		break;
332 	}
333 	case BCH_JSET_ENTRY_clock: {
334 		struct jset_entry_clock *clock =
335 			container_of(entry, struct jset_entry_clock, entry);
336 
337 		atomic64_set(&c->io_clock[clock->rw].now, le64_to_cpu(clock->time));
338 	}
339 	}
340 
341 	return ret;
342 }
343 
344 static int journal_replay_early(struct bch_fs *c,
345 				struct bch_sb_field_clean *clean)
346 {
347 	if (clean) {
348 		for (struct jset_entry *entry = clean->start;
349 		     entry != vstruct_end(&clean->field);
350 		     entry = vstruct_next(entry)) {
351 			int ret = journal_replay_entry_early(c, entry);
352 			if (ret)
353 				return ret;
354 		}
355 	} else {
356 		struct genradix_iter iter;
357 		struct journal_replay *i, **_i;
358 
359 		genradix_for_each(&c->journal_entries, iter, _i) {
360 			i = *_i;
361 
362 			if (!i || i->ignore)
363 				continue;
364 
365 			vstruct_for_each(&i->j, entry) {
366 				int ret = journal_replay_entry_early(c, entry);
367 				if (ret)
368 					return ret;
369 			}
370 		}
371 	}
372 
373 	bch2_fs_usage_initialize(c);
374 
375 	return 0;
376 }
377 
378 /* sb clean section: */
379 
380 static int read_btree_roots(struct bch_fs *c)
381 {
382 	unsigned i;
383 	int ret = 0;
384 
385 	for (i = 0; i < btree_id_nr_alive(c); i++) {
386 		struct btree_root *r = bch2_btree_id_root(c, i);
387 
388 		if (!r->alive)
389 			continue;
390 
391 		if (btree_id_is_alloc(i) &&
392 		    c->opts.reconstruct_alloc) {
393 			c->sb.compat &= ~(1ULL << BCH_COMPAT_alloc_info);
394 			continue;
395 		}
396 
397 		if (r->error) {
398 			__fsck_err(c,
399 				   btree_id_is_alloc(i)
400 				   ? FSCK_CAN_IGNORE : 0,
401 				   btree_root_bkey_invalid,
402 				   "invalid btree root %s",
403 				   bch2_btree_id_str(i));
404 			if (i == BTREE_ID_alloc)
405 				c->sb.compat &= ~(1ULL << BCH_COMPAT_alloc_info);
406 		}
407 
408 		ret = bch2_btree_root_read(c, i, &r->key, r->level);
409 		if (ret) {
410 			fsck_err(c,
411 				 btree_root_read_error,
412 				 "error reading btree root %s",
413 				 bch2_btree_id_str(i));
414 			if (btree_id_is_alloc(i))
415 				c->sb.compat &= ~(1ULL << BCH_COMPAT_alloc_info);
416 			ret = 0;
417 		}
418 	}
419 
420 	for (i = 0; i < BTREE_ID_NR; i++) {
421 		struct btree_root *r = bch2_btree_id_root(c, i);
422 
423 		if (!r->b) {
424 			r->alive = false;
425 			r->level = 0;
426 			bch2_btree_root_alloc(c, i);
427 		}
428 	}
429 fsck_err:
430 	return ret;
431 }
432 
433 static int bch2_initialize_subvolumes(struct bch_fs *c)
434 {
435 	struct bkey_i_snapshot_tree	root_tree;
436 	struct bkey_i_snapshot		root_snapshot;
437 	struct bkey_i_subvolume		root_volume;
438 	int ret;
439 
440 	bkey_snapshot_tree_init(&root_tree.k_i);
441 	root_tree.k.p.offset		= 1;
442 	root_tree.v.master_subvol	= cpu_to_le32(1);
443 	root_tree.v.root_snapshot	= cpu_to_le32(U32_MAX);
444 
445 	bkey_snapshot_init(&root_snapshot.k_i);
446 	root_snapshot.k.p.offset = U32_MAX;
447 	root_snapshot.v.flags	= 0;
448 	root_snapshot.v.parent	= 0;
449 	root_snapshot.v.subvol	= cpu_to_le32(BCACHEFS_ROOT_SUBVOL);
450 	root_snapshot.v.tree	= cpu_to_le32(1);
451 	SET_BCH_SNAPSHOT_SUBVOL(&root_snapshot.v, true);
452 
453 	bkey_subvolume_init(&root_volume.k_i);
454 	root_volume.k.p.offset = BCACHEFS_ROOT_SUBVOL;
455 	root_volume.v.flags	= 0;
456 	root_volume.v.snapshot	= cpu_to_le32(U32_MAX);
457 	root_volume.v.inode	= cpu_to_le64(BCACHEFS_ROOT_INO);
458 
459 	ret =   bch2_btree_insert(c, BTREE_ID_snapshot_trees,	&root_tree.k_i, NULL, 0) ?:
460 		bch2_btree_insert(c, BTREE_ID_snapshots,	&root_snapshot.k_i, NULL, 0) ?:
461 		bch2_btree_insert(c, BTREE_ID_subvolumes,	&root_volume.k_i, NULL, 0);
462 	bch_err_fn(c, ret);
463 	return ret;
464 }
465 
466 static int __bch2_fs_upgrade_for_subvolumes(struct btree_trans *trans)
467 {
468 	struct btree_iter iter;
469 	struct bkey_s_c k;
470 	struct bch_inode_unpacked inode;
471 	int ret;
472 
473 	k = bch2_bkey_get_iter(trans, &iter, BTREE_ID_inodes,
474 			       SPOS(0, BCACHEFS_ROOT_INO, U32_MAX), 0);
475 	ret = bkey_err(k);
476 	if (ret)
477 		return ret;
478 
479 	if (!bkey_is_inode(k.k)) {
480 		bch_err(trans->c, "root inode not found");
481 		ret = -BCH_ERR_ENOENT_inode;
482 		goto err;
483 	}
484 
485 	ret = bch2_inode_unpack(k, &inode);
486 	BUG_ON(ret);
487 
488 	inode.bi_subvol = BCACHEFS_ROOT_SUBVOL;
489 
490 	ret = bch2_inode_write(trans, &iter, &inode);
491 err:
492 	bch2_trans_iter_exit(trans, &iter);
493 	return ret;
494 }
495 
496 /* set bi_subvol on root inode */
497 noinline_for_stack
498 static int bch2_fs_upgrade_for_subvolumes(struct bch_fs *c)
499 {
500 	int ret = bch2_trans_do(c, NULL, NULL, BCH_TRANS_COMMIT_lazy_rw,
501 				__bch2_fs_upgrade_for_subvolumes(trans));
502 	bch_err_fn(c, ret);
503 	return ret;
504 }
505 
506 const char * const bch2_recovery_passes[] = {
507 #define x(_fn, ...)	#_fn,
508 	BCH_RECOVERY_PASSES()
509 #undef x
510 	NULL
511 };
512 
513 static int bch2_check_allocations(struct bch_fs *c)
514 {
515 	return bch2_gc(c, true, c->opts.norecovery);
516 }
517 
518 static int bch2_set_may_go_rw(struct bch_fs *c)
519 {
520 	struct journal_keys *keys = &c->journal_keys;
521 
522 	/*
523 	 * After we go RW, the journal keys buffer can't be modified (except for
524 	 * setting journal_key->overwritten: it will be accessed by multiple
525 	 * threads
526 	 */
527 	move_gap(keys->d, keys->nr, keys->size, keys->gap, keys->nr);
528 	keys->gap = keys->nr;
529 
530 	set_bit(BCH_FS_may_go_rw, &c->flags);
531 
532 	if (keys->nr || c->opts.fsck || !c->sb.clean)
533 		return bch2_fs_read_write_early(c);
534 	return 0;
535 }
536 
537 struct recovery_pass_fn {
538 	int		(*fn)(struct bch_fs *);
539 	unsigned	when;
540 };
541 
542 static struct recovery_pass_fn recovery_pass_fns[] = {
543 #define x(_fn, _id, _when)	{ .fn = bch2_##_fn, .when = _when },
544 	BCH_RECOVERY_PASSES()
545 #undef x
546 };
547 
548 u64 bch2_recovery_passes_to_stable(u64 v)
549 {
550 	static const u8 map[] = {
551 #define x(n, id, ...)	[BCH_RECOVERY_PASS_##n] = BCH_RECOVERY_PASS_STABLE_##n,
552 	BCH_RECOVERY_PASSES()
553 #undef x
554 	};
555 
556 	u64 ret = 0;
557 	for (unsigned i = 0; i < ARRAY_SIZE(map); i++)
558 		if (v & BIT_ULL(i))
559 			ret |= BIT_ULL(map[i]);
560 	return ret;
561 }
562 
563 u64 bch2_recovery_passes_from_stable(u64 v)
564 {
565 	static const u8 map[] = {
566 #define x(n, id, ...)	[BCH_RECOVERY_PASS_STABLE_##n] = BCH_RECOVERY_PASS_##n,
567 	BCH_RECOVERY_PASSES()
568 #undef x
569 	};
570 
571 	u64 ret = 0;
572 	for (unsigned i = 0; i < ARRAY_SIZE(map); i++)
573 		if (v & BIT_ULL(i))
574 			ret |= BIT_ULL(map[i]);
575 	return ret;
576 }
577 
578 static bool check_version_upgrade(struct bch_fs *c)
579 {
580 	unsigned latest_version	= bcachefs_metadata_version_current;
581 	unsigned latest_compatible = min(latest_version,
582 					 bch2_latest_compatible_version(c->sb.version));
583 	unsigned old_version = c->sb.version_upgrade_complete ?: c->sb.version;
584 	unsigned new_version = 0;
585 
586 	if (old_version < bcachefs_metadata_required_upgrade_below) {
587 		if (c->opts.version_upgrade == BCH_VERSION_UPGRADE_incompatible ||
588 		    latest_compatible < bcachefs_metadata_required_upgrade_below)
589 			new_version = latest_version;
590 		else
591 			new_version = latest_compatible;
592 	} else {
593 		switch (c->opts.version_upgrade) {
594 		case BCH_VERSION_UPGRADE_compatible:
595 			new_version = latest_compatible;
596 			break;
597 		case BCH_VERSION_UPGRADE_incompatible:
598 			new_version = latest_version;
599 			break;
600 		case BCH_VERSION_UPGRADE_none:
601 			new_version = min(old_version, latest_version);
602 			break;
603 		}
604 	}
605 
606 	if (new_version > old_version) {
607 		struct printbuf buf = PRINTBUF;
608 
609 		if (old_version < bcachefs_metadata_required_upgrade_below)
610 			prt_str(&buf, "Version upgrade required:\n");
611 
612 		if (old_version != c->sb.version) {
613 			prt_str(&buf, "Version upgrade from ");
614 			bch2_version_to_text(&buf, c->sb.version_upgrade_complete);
615 			prt_str(&buf, " to ");
616 			bch2_version_to_text(&buf, c->sb.version);
617 			prt_str(&buf, " incomplete\n");
618 		}
619 
620 		prt_printf(&buf, "Doing %s version upgrade from ",
621 			   BCH_VERSION_MAJOR(old_version) != BCH_VERSION_MAJOR(new_version)
622 			   ? "incompatible" : "compatible");
623 		bch2_version_to_text(&buf, old_version);
624 		prt_str(&buf, " to ");
625 		bch2_version_to_text(&buf, new_version);
626 		prt_newline(&buf);
627 
628 		struct bch_sb_field_ext *ext = bch2_sb_field_get(c->disk_sb.sb, ext);
629 		__le64 passes = ext->recovery_passes_required[0];
630 		bch2_sb_set_upgrade(c, old_version, new_version);
631 		passes = ext->recovery_passes_required[0] & ~passes;
632 
633 		if (passes) {
634 			prt_str(&buf, "  running recovery passes: ");
635 			prt_bitflags(&buf, bch2_recovery_passes,
636 				     bch2_recovery_passes_from_stable(le64_to_cpu(passes)));
637 		}
638 
639 		bch_info(c, "%s", buf.buf);
640 
641 		bch2_sb_upgrade(c, new_version);
642 
643 		printbuf_exit(&buf);
644 		return true;
645 	}
646 
647 	return false;
648 }
649 
650 u64 bch2_fsck_recovery_passes(void)
651 {
652 	u64 ret = 0;
653 
654 	for (unsigned i = 0; i < ARRAY_SIZE(recovery_pass_fns); i++)
655 		if (recovery_pass_fns[i].when & PASS_FSCK)
656 			ret |= BIT_ULL(i);
657 	return ret;
658 }
659 
660 static bool should_run_recovery_pass(struct bch_fs *c, enum bch_recovery_pass pass)
661 {
662 	struct recovery_pass_fn *p = recovery_pass_fns + pass;
663 
664 	if (c->opts.norecovery && pass > BCH_RECOVERY_PASS_snapshots_read)
665 		return false;
666 	if (c->recovery_passes_explicit & BIT_ULL(pass))
667 		return true;
668 	if ((p->when & PASS_FSCK) && c->opts.fsck)
669 		return true;
670 	if ((p->when & PASS_UNCLEAN) && !c->sb.clean)
671 		return true;
672 	if (p->when & PASS_ALWAYS)
673 		return true;
674 	return false;
675 }
676 
677 static int bch2_run_recovery_pass(struct bch_fs *c, enum bch_recovery_pass pass)
678 {
679 	struct recovery_pass_fn *p = recovery_pass_fns + pass;
680 	int ret;
681 
682 	if (!(p->when & PASS_SILENT))
683 		bch2_print(c, KERN_INFO bch2_log_msg(c, "%s..."),
684 			   bch2_recovery_passes[pass]);
685 	ret = p->fn(c);
686 	if (ret)
687 		return ret;
688 	if (!(p->when & PASS_SILENT))
689 		bch2_print(c, KERN_CONT " done\n");
690 
691 	return 0;
692 }
693 
694 static int bch2_run_recovery_passes(struct bch_fs *c)
695 {
696 	int ret = 0;
697 
698 	while (c->curr_recovery_pass < ARRAY_SIZE(recovery_pass_fns)) {
699 		if (should_run_recovery_pass(c, c->curr_recovery_pass)) {
700 			unsigned pass = c->curr_recovery_pass;
701 
702 			ret = bch2_run_recovery_pass(c, c->curr_recovery_pass);
703 			if (bch2_err_matches(ret, BCH_ERR_restart_recovery) ||
704 			    (ret && c->curr_recovery_pass < pass))
705 				continue;
706 			if (ret)
707 				break;
708 
709 			c->recovery_passes_complete |= BIT_ULL(c->curr_recovery_pass);
710 		}
711 		c->curr_recovery_pass++;
712 		c->recovery_pass_done = max(c->recovery_pass_done, c->curr_recovery_pass);
713 	}
714 
715 	return ret;
716 }
717 
718 int bch2_run_online_recovery_passes(struct bch_fs *c)
719 {
720 	int ret = 0;
721 
722 	for (unsigned i = 0; i < ARRAY_SIZE(recovery_pass_fns); i++) {
723 		struct recovery_pass_fn *p = recovery_pass_fns + i;
724 
725 		if (!(p->when & PASS_ONLINE))
726 			continue;
727 
728 		ret = bch2_run_recovery_pass(c, i);
729 		if (bch2_err_matches(ret, BCH_ERR_restart_recovery)) {
730 			i = c->curr_recovery_pass;
731 			continue;
732 		}
733 		if (ret)
734 			break;
735 	}
736 
737 	return ret;
738 }
739 
740 int bch2_fs_recovery(struct bch_fs *c)
741 {
742 	struct bch_sb_field_clean *clean = NULL;
743 	struct jset *last_journal_entry = NULL;
744 	u64 last_seq = 0, blacklist_seq, journal_seq;
745 	int ret = 0;
746 
747 	if (c->sb.clean) {
748 		clean = bch2_read_superblock_clean(c);
749 		ret = PTR_ERR_OR_ZERO(clean);
750 		if (ret)
751 			goto err;
752 
753 		bch_info(c, "recovering from clean shutdown, journal seq %llu",
754 			 le64_to_cpu(clean->journal_seq));
755 	} else {
756 		bch_info(c, "recovering from unclean shutdown");
757 	}
758 
759 	if (!(c->sb.features & (1ULL << BCH_FEATURE_new_extent_overwrite))) {
760 		bch_err(c, "feature new_extent_overwrite not set, filesystem no longer supported");
761 		ret = -EINVAL;
762 		goto err;
763 	}
764 
765 	if (!c->sb.clean &&
766 	    !(c->sb.features & (1ULL << BCH_FEATURE_extents_above_btree_updates))) {
767 		bch_err(c, "filesystem needs recovery from older version; run fsck from older bcachefs-tools to fix");
768 		ret = -EINVAL;
769 		goto err;
770 	}
771 
772 	if (c->opts.fsck && c->opts.norecovery) {
773 		bch_err(c, "cannot select both norecovery and fsck");
774 		ret = -EINVAL;
775 		goto err;
776 	}
777 
778 	if (!c->opts.nochanges) {
779 		mutex_lock(&c->sb_lock);
780 		bool write_sb = false;
781 
782 		struct bch_sb_field_ext *ext =
783 			bch2_sb_field_get_minsize(&c->disk_sb, ext, sizeof(*ext) / sizeof(u64));
784 		if (!ext) {
785 			ret = -BCH_ERR_ENOSPC_sb;
786 			mutex_unlock(&c->sb_lock);
787 			goto err;
788 		}
789 
790 		if (BCH_SB_HAS_TOPOLOGY_ERRORS(c->disk_sb.sb)) {
791 			ext->recovery_passes_required[0] |=
792 				cpu_to_le64(bch2_recovery_passes_to_stable(BIT_ULL(BCH_RECOVERY_PASS_check_topology)));
793 			write_sb = true;
794 		}
795 
796 		u64 sb_passes = bch2_recovery_passes_from_stable(le64_to_cpu(ext->recovery_passes_required[0]));
797 		if (sb_passes) {
798 			struct printbuf buf = PRINTBUF;
799 			prt_str(&buf, "superblock requires following recovery passes to be run:\n  ");
800 			prt_bitflags(&buf, bch2_recovery_passes, sb_passes);
801 			bch_info(c, "%s", buf.buf);
802 			printbuf_exit(&buf);
803 		}
804 
805 		if (bch2_check_version_downgrade(c)) {
806 			struct printbuf buf = PRINTBUF;
807 
808 			prt_str(&buf, "Version downgrade required:");
809 
810 			__le64 passes = ext->recovery_passes_required[0];
811 			bch2_sb_set_downgrade(c,
812 					BCH_VERSION_MINOR(bcachefs_metadata_version_current),
813 					BCH_VERSION_MINOR(c->sb.version));
814 			passes = ext->recovery_passes_required[0] & ~passes;
815 			if (passes) {
816 				prt_str(&buf, "\n  running recovery passes: ");
817 				prt_bitflags(&buf, bch2_recovery_passes,
818 					     bch2_recovery_passes_from_stable(le64_to_cpu(passes)));
819 			}
820 
821 			bch_info(c, "%s", buf.buf);
822 			printbuf_exit(&buf);
823 			write_sb = true;
824 		}
825 
826 		if (check_version_upgrade(c))
827 			write_sb = true;
828 
829 		if (write_sb)
830 			bch2_write_super(c);
831 
832 		c->recovery_passes_explicit |= bch2_recovery_passes_from_stable(le64_to_cpu(ext->recovery_passes_required[0]));
833 		mutex_unlock(&c->sb_lock);
834 	}
835 
836 	if (c->opts.fsck && IS_ENABLED(CONFIG_BCACHEFS_DEBUG))
837 		c->recovery_passes_explicit |= BIT_ULL(BCH_RECOVERY_PASS_check_topology);
838 
839 	if (c->opts.fsck)
840 		set_bit(BCH_FS_fsck_running, &c->flags);
841 
842 	ret = bch2_blacklist_table_initialize(c);
843 	if (ret) {
844 		bch_err(c, "error initializing blacklist table");
845 		goto err;
846 	}
847 
848 	if (!c->sb.clean || c->opts.fsck || c->opts.keep_journal) {
849 		struct genradix_iter iter;
850 		struct journal_replay **i;
851 
852 		bch_verbose(c, "starting journal read");
853 		ret = bch2_journal_read(c, &last_seq, &blacklist_seq, &journal_seq);
854 		if (ret)
855 			goto err;
856 
857 		/*
858 		 * note: cmd_list_journal needs the blacklist table fully up to date so
859 		 * it can asterisk ignored journal entries:
860 		 */
861 		if (c->opts.read_journal_only)
862 			goto out;
863 
864 		genradix_for_each_reverse(&c->journal_entries, iter, i)
865 			if (*i && !(*i)->ignore) {
866 				last_journal_entry = &(*i)->j;
867 				break;
868 			}
869 
870 		if (mustfix_fsck_err_on(c->sb.clean &&
871 					last_journal_entry &&
872 					!journal_entry_empty(last_journal_entry), c,
873 				clean_but_journal_not_empty,
874 				"filesystem marked clean but journal not empty")) {
875 			c->sb.compat &= ~(1ULL << BCH_COMPAT_alloc_info);
876 			SET_BCH_SB_CLEAN(c->disk_sb.sb, false);
877 			c->sb.clean = false;
878 		}
879 
880 		if (!last_journal_entry) {
881 			fsck_err_on(!c->sb.clean, c,
882 				    dirty_but_no_journal_entries,
883 				    "no journal entries found");
884 			if (clean)
885 				goto use_clean;
886 
887 			genradix_for_each_reverse(&c->journal_entries, iter, i)
888 				if (*i) {
889 					last_journal_entry = &(*i)->j;
890 					(*i)->ignore = false;
891 					/*
892 					 * This was probably a NO_FLUSH entry,
893 					 * so last_seq was garbage - but we know
894 					 * we're only using a single journal
895 					 * entry, set it here:
896 					 */
897 					(*i)->j.last_seq = (*i)->j.seq;
898 					break;
899 				}
900 		}
901 
902 		ret = bch2_journal_keys_sort(c);
903 		if (ret)
904 			goto err;
905 
906 		if (c->sb.clean && last_journal_entry) {
907 			ret = bch2_verify_superblock_clean(c, &clean,
908 						      last_journal_entry);
909 			if (ret)
910 				goto err;
911 		}
912 	} else {
913 use_clean:
914 		if (!clean) {
915 			bch_err(c, "no superblock clean section found");
916 			ret = -BCH_ERR_fsck_repair_impossible;
917 			goto err;
918 
919 		}
920 		blacklist_seq = journal_seq = le64_to_cpu(clean->journal_seq) + 1;
921 	}
922 
923 	c->journal_replay_seq_start	= last_seq;
924 	c->journal_replay_seq_end	= blacklist_seq - 1;
925 
926 	if (c->opts.reconstruct_alloc) {
927 		c->sb.compat &= ~(1ULL << BCH_COMPAT_alloc_info);
928 		drop_alloc_keys(&c->journal_keys);
929 	}
930 
931 	zero_out_btree_mem_ptr(&c->journal_keys);
932 
933 	ret = journal_replay_early(c, clean);
934 	if (ret)
935 		goto err;
936 
937 	/*
938 	 * After an unclean shutdown, skip then next few journal sequence
939 	 * numbers as they may have been referenced by btree writes that
940 	 * happened before their corresponding journal writes - those btree
941 	 * writes need to be ignored, by skipping and blacklisting the next few
942 	 * journal sequence numbers:
943 	 */
944 	if (!c->sb.clean)
945 		journal_seq += 8;
946 
947 	if (blacklist_seq != journal_seq) {
948 		ret =   bch2_journal_log_msg(c, "blacklisting entries %llu-%llu",
949 					     blacklist_seq, journal_seq) ?:
950 			bch2_journal_seq_blacklist_add(c,
951 					blacklist_seq, journal_seq);
952 		if (ret) {
953 			bch_err(c, "error creating new journal seq blacklist entry");
954 			goto err;
955 		}
956 	}
957 
958 	ret =   bch2_journal_log_msg(c, "starting journal at entry %llu, replaying %llu-%llu",
959 				     journal_seq, last_seq, blacklist_seq - 1) ?:
960 		bch2_fs_journal_start(&c->journal, journal_seq);
961 	if (ret)
962 		goto err;
963 
964 	if (c->opts.reconstruct_alloc)
965 		bch2_journal_log_msg(c, "dropping alloc info");
966 
967 	/*
968 	 * Skip past versions that might have possibly been used (as nonces),
969 	 * but hadn't had their pointers written:
970 	 */
971 	if (c->sb.encryption_type && !c->sb.clean)
972 		atomic64_add(1 << 16, &c->key_version);
973 
974 	ret = read_btree_roots(c);
975 	if (ret)
976 		goto err;
977 
978 	ret = bch2_run_recovery_passes(c);
979 	if (ret)
980 		goto err;
981 
982 	clear_bit(BCH_FS_fsck_running, &c->flags);
983 
984 	/* If we fixed errors, verify that fs is actually clean now: */
985 	if (IS_ENABLED(CONFIG_BCACHEFS_DEBUG) &&
986 	    test_bit(BCH_FS_errors_fixed, &c->flags) &&
987 	    !test_bit(BCH_FS_errors_not_fixed, &c->flags) &&
988 	    !test_bit(BCH_FS_error, &c->flags)) {
989 		bch2_flush_fsck_errs(c);
990 
991 		bch_info(c, "Fixed errors, running fsck a second time to verify fs is clean");
992 		clear_bit(BCH_FS_errors_fixed, &c->flags);
993 
994 		c->curr_recovery_pass = BCH_RECOVERY_PASS_check_alloc_info;
995 
996 		ret = bch2_run_recovery_passes(c);
997 		if (ret)
998 			goto err;
999 
1000 		if (test_bit(BCH_FS_errors_fixed, &c->flags) ||
1001 		    test_bit(BCH_FS_errors_not_fixed, &c->flags)) {
1002 			bch_err(c, "Second fsck run was not clean");
1003 			set_bit(BCH_FS_errors_not_fixed, &c->flags);
1004 		}
1005 
1006 		set_bit(BCH_FS_errors_fixed, &c->flags);
1007 	}
1008 
1009 	if (enabled_qtypes(c)) {
1010 		bch_verbose(c, "reading quotas");
1011 		ret = bch2_fs_quota_read(c);
1012 		if (ret)
1013 			goto err;
1014 		bch_verbose(c, "quotas done");
1015 	}
1016 
1017 	mutex_lock(&c->sb_lock);
1018 	bool write_sb = false;
1019 
1020 	if (BCH_SB_VERSION_UPGRADE_COMPLETE(c->disk_sb.sb) != le16_to_cpu(c->disk_sb.sb->version)) {
1021 		SET_BCH_SB_VERSION_UPGRADE_COMPLETE(c->disk_sb.sb, le16_to_cpu(c->disk_sb.sb->version));
1022 		write_sb = true;
1023 	}
1024 
1025 	if (!test_bit(BCH_FS_error, &c->flags) &&
1026 	    !(c->disk_sb.sb->compat[0] & cpu_to_le64(1ULL << BCH_COMPAT_alloc_info))) {
1027 		c->disk_sb.sb->compat[0] |= cpu_to_le64(1ULL << BCH_COMPAT_alloc_info);
1028 		write_sb = true;
1029 	}
1030 
1031 	if (!test_bit(BCH_FS_error, &c->flags)) {
1032 		struct bch_sb_field_ext *ext = bch2_sb_field_get(c->disk_sb.sb, ext);
1033 		if (ext &&
1034 		    (!bch2_is_zero(ext->recovery_passes_required, sizeof(ext->recovery_passes_required)) ||
1035 		     !bch2_is_zero(ext->errors_silent, sizeof(ext->errors_silent)))) {
1036 			memset(ext->recovery_passes_required, 0, sizeof(ext->recovery_passes_required));
1037 			memset(ext->errors_silent, 0, sizeof(ext->errors_silent));
1038 			write_sb = true;
1039 		}
1040 	}
1041 
1042 	if (c->opts.fsck &&
1043 	    !test_bit(BCH_FS_error, &c->flags) &&
1044 	    !test_bit(BCH_FS_errors_not_fixed, &c->flags)) {
1045 		SET_BCH_SB_HAS_ERRORS(c->disk_sb.sb, 0);
1046 		SET_BCH_SB_HAS_TOPOLOGY_ERRORS(c->disk_sb.sb, 0);
1047 		write_sb = true;
1048 	}
1049 
1050 	if (write_sb)
1051 		bch2_write_super(c);
1052 	mutex_unlock(&c->sb_lock);
1053 
1054 	if (!(c->sb.compat & (1ULL << BCH_COMPAT_extents_above_btree_updates_done)) ||
1055 	    c->sb.version_min < bcachefs_metadata_version_btree_ptr_sectors_written) {
1056 		struct bch_move_stats stats;
1057 
1058 		bch2_move_stats_init(&stats, "recovery");
1059 
1060 		struct printbuf buf = PRINTBUF;
1061 		bch2_version_to_text(&buf, c->sb.version_min);
1062 		bch_info(c, "scanning for old btree nodes: min_version %s", buf.buf);
1063 		printbuf_exit(&buf);
1064 
1065 		ret =   bch2_fs_read_write_early(c) ?:
1066 			bch2_scan_old_btree_nodes(c, &stats);
1067 		if (ret)
1068 			goto err;
1069 		bch_info(c, "scanning for old btree nodes done");
1070 	}
1071 
1072 	if (c->journal_seq_blacklist_table &&
1073 	    c->journal_seq_blacklist_table->nr > 128)
1074 		queue_work(system_long_wq, &c->journal_seq_blacklist_gc_work);
1075 
1076 	ret = 0;
1077 out:
1078 	bch2_flush_fsck_errs(c);
1079 
1080 	if (!c->opts.keep_journal &&
1081 	    test_bit(JOURNAL_REPLAY_DONE, &c->journal.flags))
1082 		bch2_journal_keys_put_initial(c);
1083 	kfree(clean);
1084 
1085 	if (!ret &&
1086 	    test_bit(BCH_FS_need_delete_dead_snapshots, &c->flags) &&
1087 	    !c->opts.nochanges) {
1088 		bch2_fs_read_write_early(c);
1089 		bch2_delete_dead_snapshots_async(c);
1090 	}
1091 
1092 	bch_err_fn(c, ret);
1093 	return ret;
1094 err:
1095 fsck_err:
1096 	bch2_fs_emergency_read_only(c);
1097 	goto out;
1098 }
1099 
1100 int bch2_fs_initialize(struct bch_fs *c)
1101 {
1102 	struct bch_inode_unpacked root_inode, lostfound_inode;
1103 	struct bkey_inode_buf packed_inode;
1104 	struct qstr lostfound = QSTR("lost+found");
1105 	int ret;
1106 
1107 	bch_notice(c, "initializing new filesystem");
1108 
1109 	mutex_lock(&c->sb_lock);
1110 	c->disk_sb.sb->compat[0] |= cpu_to_le64(1ULL << BCH_COMPAT_extents_above_btree_updates_done);
1111 	c->disk_sb.sb->compat[0] |= cpu_to_le64(1ULL << BCH_COMPAT_bformat_overflow_done);
1112 
1113 	bch2_check_version_downgrade(c);
1114 
1115 	if (c->opts.version_upgrade != BCH_VERSION_UPGRADE_none) {
1116 		bch2_sb_upgrade(c, bcachefs_metadata_version_current);
1117 		SET_BCH_SB_VERSION_UPGRADE_COMPLETE(c->disk_sb.sb, bcachefs_metadata_version_current);
1118 		bch2_write_super(c);
1119 	}
1120 	mutex_unlock(&c->sb_lock);
1121 
1122 	c->curr_recovery_pass = ARRAY_SIZE(recovery_pass_fns);
1123 	set_bit(BCH_FS_may_go_rw, &c->flags);
1124 
1125 	for (unsigned i = 0; i < BTREE_ID_NR; i++)
1126 		bch2_btree_root_alloc(c, i);
1127 
1128 	for_each_member_device(c, ca)
1129 		bch2_dev_usage_init(ca);
1130 
1131 	ret = bch2_fs_journal_alloc(c);
1132 	if (ret)
1133 		goto err;
1134 
1135 	/*
1136 	 * journal_res_get() will crash if called before this has
1137 	 * set up the journal.pin FIFO and journal.cur pointer:
1138 	 */
1139 	bch2_fs_journal_start(&c->journal, 1);
1140 	bch2_journal_set_replay_done(&c->journal);
1141 
1142 	ret = bch2_fs_read_write_early(c);
1143 	if (ret)
1144 		goto err;
1145 
1146 	/*
1147 	 * Write out the superblock and journal buckets, now that we can do
1148 	 * btree updates
1149 	 */
1150 	bch_verbose(c, "marking superblocks");
1151 	ret = bch2_trans_mark_dev_sbs(c);
1152 	bch_err_msg(c, ret, "marking superblocks");
1153 	if (ret)
1154 		goto err;
1155 
1156 	for_each_online_member(c, ca)
1157 		ca->new_fs_bucket_idx = 0;
1158 
1159 	ret = bch2_fs_freespace_init(c);
1160 	if (ret)
1161 		goto err;
1162 
1163 	ret = bch2_initialize_subvolumes(c);
1164 	if (ret)
1165 		goto err;
1166 
1167 	bch_verbose(c, "reading snapshots table");
1168 	ret = bch2_snapshots_read(c);
1169 	if (ret)
1170 		goto err;
1171 	bch_verbose(c, "reading snapshots done");
1172 
1173 	bch2_inode_init(c, &root_inode, 0, 0, S_IFDIR|0755, 0, NULL);
1174 	root_inode.bi_inum	= BCACHEFS_ROOT_INO;
1175 	root_inode.bi_subvol	= BCACHEFS_ROOT_SUBVOL;
1176 	bch2_inode_pack(&packed_inode, &root_inode);
1177 	packed_inode.inode.k.p.snapshot = U32_MAX;
1178 
1179 	ret = bch2_btree_insert(c, BTREE_ID_inodes, &packed_inode.inode.k_i, NULL, 0);
1180 	bch_err_msg(c, ret, "creating root directory");
1181 	if (ret)
1182 		goto err;
1183 
1184 	bch2_inode_init_early(c, &lostfound_inode);
1185 
1186 	ret = bch2_trans_do(c, NULL, NULL, 0,
1187 		bch2_create_trans(trans,
1188 				  BCACHEFS_ROOT_SUBVOL_INUM,
1189 				  &root_inode, &lostfound_inode,
1190 				  &lostfound,
1191 				  0, 0, S_IFDIR|0700, 0,
1192 				  NULL, NULL, (subvol_inum) { 0 }, 0));
1193 	bch_err_msg(c, ret, "creating lost+found");
1194 	if (ret)
1195 		goto err;
1196 
1197 	c->recovery_pass_done = ARRAY_SIZE(recovery_pass_fns) - 1;
1198 
1199 	if (enabled_qtypes(c)) {
1200 		ret = bch2_fs_quota_read(c);
1201 		if (ret)
1202 			goto err;
1203 	}
1204 
1205 	ret = bch2_journal_flush(&c->journal);
1206 	bch_err_msg(c, ret, "writing first journal entry");
1207 	if (ret)
1208 		goto err;
1209 
1210 	mutex_lock(&c->sb_lock);
1211 	SET_BCH_SB_INITIALIZED(c->disk_sb.sb, true);
1212 	SET_BCH_SB_CLEAN(c->disk_sb.sb, false);
1213 
1214 	bch2_write_super(c);
1215 	mutex_unlock(&c->sb_lock);
1216 
1217 	return 0;
1218 err:
1219 	bch_err_fn(c, ret);
1220 	return ret;
1221 }
1222