xref: /linux/fs/bcachefs/fsck.c (revision 55d0969c451159cff86949b38c39171cab962069)
1 // SPDX-License-Identifier: GPL-2.0
2 
3 #include "bcachefs.h"
4 #include "bkey_buf.h"
5 #include "btree_cache.h"
6 #include "btree_update.h"
7 #include "buckets.h"
8 #include "darray.h"
9 #include "dirent.h"
10 #include "error.h"
11 #include "fs.h"
12 #include "fs-common.h"
13 #include "fsck.h"
14 #include "inode.h"
15 #include "keylist.h"
16 #include "recovery_passes.h"
17 #include "snapshot.h"
18 #include "super.h"
19 #include "xattr.h"
20 
21 #include <linux/bsearch.h>
22 #include <linux/dcache.h> /* struct qstr */
23 
24 static bool inode_points_to_dirent(struct bch_inode_unpacked *inode,
25 				   struct bkey_s_c_dirent d)
26 {
27 	return  inode->bi_dir		== d.k->p.inode &&
28 		inode->bi_dir_offset	== d.k->p.offset;
29 }
30 
31 static int dirent_points_to_inode_nowarn(struct bkey_s_c_dirent d,
32 					 struct bch_inode_unpacked *inode)
33 {
34 	if (d.v->d_type == DT_SUBVOL
35 	    ? le32_to_cpu(d.v->d_child_subvol)	== inode->bi_subvol
36 	    : le64_to_cpu(d.v->d_inum)		== inode->bi_inum)
37 		return 0;
38 	return -BCH_ERR_ENOENT_dirent_doesnt_match_inode;
39 }
40 
41 static void dirent_inode_mismatch_msg(struct printbuf *out,
42 				      struct bch_fs *c,
43 				      struct bkey_s_c_dirent dirent,
44 				      struct bch_inode_unpacked *inode)
45 {
46 	prt_str(out, "inode points to dirent that does not point back:");
47 	prt_newline(out);
48 	bch2_bkey_val_to_text(out, c, dirent.s_c);
49 	prt_newline(out);
50 	bch2_inode_unpacked_to_text(out, inode);
51 }
52 
53 static int dirent_points_to_inode(struct bch_fs *c,
54 				  struct bkey_s_c_dirent dirent,
55 				  struct bch_inode_unpacked *inode)
56 {
57 	int ret = dirent_points_to_inode_nowarn(dirent, inode);
58 	if (ret) {
59 		struct printbuf buf = PRINTBUF;
60 		dirent_inode_mismatch_msg(&buf, c, dirent, inode);
61 		bch_warn(c, "%s", buf.buf);
62 		printbuf_exit(&buf);
63 	}
64 	return ret;
65 }
66 
67 /*
68  * XXX: this is handling transaction restarts without returning
69  * -BCH_ERR_transaction_restart_nested, this is not how we do things anymore:
70  */
71 static s64 bch2_count_inode_sectors(struct btree_trans *trans, u64 inum,
72 				    u32 snapshot)
73 {
74 	u64 sectors = 0;
75 
76 	int ret = for_each_btree_key_upto(trans, iter, BTREE_ID_extents,
77 				SPOS(inum, 0, snapshot),
78 				POS(inum, U64_MAX),
79 				0, k, ({
80 		if (bkey_extent_is_allocation(k.k))
81 			sectors += k.k->size;
82 		0;
83 	}));
84 
85 	return ret ?: sectors;
86 }
87 
88 static s64 bch2_count_subdirs(struct btree_trans *trans, u64 inum,
89 				    u32 snapshot)
90 {
91 	u64 subdirs = 0;
92 
93 	int ret = for_each_btree_key_upto(trans, iter, BTREE_ID_dirents,
94 				    SPOS(inum, 0, snapshot),
95 				    POS(inum, U64_MAX),
96 				    0, k, ({
97 		if (k.k->type == KEY_TYPE_dirent &&
98 		    bkey_s_c_to_dirent(k).v->d_type == DT_DIR)
99 			subdirs++;
100 		0;
101 	}));
102 
103 	return ret ?: subdirs;
104 }
105 
106 static int subvol_lookup(struct btree_trans *trans, u32 subvol,
107 			 u32 *snapshot, u64 *inum)
108 {
109 	struct bch_subvolume s;
110 	int ret = bch2_subvolume_get(trans, subvol, false, 0, &s);
111 
112 	*snapshot = le32_to_cpu(s.snapshot);
113 	*inum = le64_to_cpu(s.inode);
114 	return ret;
115 }
116 
117 static int lookup_first_inode(struct btree_trans *trans, u64 inode_nr,
118 			      struct bch_inode_unpacked *inode)
119 {
120 	struct btree_iter iter;
121 	struct bkey_s_c k;
122 	int ret;
123 
124 	for_each_btree_key_norestart(trans, iter, BTREE_ID_inodes, POS(0, inode_nr),
125 				     BTREE_ITER_all_snapshots, k, ret) {
126 		if (k.k->p.offset != inode_nr)
127 			break;
128 		if (!bkey_is_inode(k.k))
129 			continue;
130 		ret = bch2_inode_unpack(k, inode);
131 		goto found;
132 	}
133 	ret = -BCH_ERR_ENOENT_inode;
134 found:
135 	bch_err_msg(trans->c, ret, "fetching inode %llu", inode_nr);
136 	bch2_trans_iter_exit(trans, &iter);
137 	return ret;
138 }
139 
140 static int lookup_inode(struct btree_trans *trans, u64 inode_nr, u32 snapshot,
141 			struct bch_inode_unpacked *inode)
142 {
143 	struct btree_iter iter;
144 	struct bkey_s_c k;
145 	int ret;
146 
147 	k = bch2_bkey_get_iter(trans, &iter, BTREE_ID_inodes,
148 			       SPOS(0, inode_nr, snapshot), 0);
149 	ret = bkey_err(k);
150 	if (ret)
151 		goto err;
152 
153 	ret = bkey_is_inode(k.k)
154 		? bch2_inode_unpack(k, inode)
155 		: -BCH_ERR_ENOENT_inode;
156 err:
157 	bch2_trans_iter_exit(trans, &iter);
158 	return ret;
159 }
160 
161 static int lookup_dirent_in_snapshot(struct btree_trans *trans,
162 			   struct bch_hash_info hash_info,
163 			   subvol_inum dir, struct qstr *name,
164 			   u64 *target, unsigned *type, u32 snapshot)
165 {
166 	struct btree_iter iter;
167 	struct bkey_s_c k = bch2_hash_lookup_in_snapshot(trans, &iter, bch2_dirent_hash_desc,
168 							 &hash_info, dir, name, 0, snapshot);
169 	int ret = bkey_err(k);
170 	if (ret)
171 		return ret;
172 
173 	struct bkey_s_c_dirent d = bkey_s_c_to_dirent(bch2_btree_iter_peek_slot(&iter));
174 	*target = le64_to_cpu(d.v->d_inum);
175 	*type = d.v->d_type;
176 	bch2_trans_iter_exit(trans, &iter);
177 	return 0;
178 }
179 
180 static int __remove_dirent(struct btree_trans *trans, struct bpos pos)
181 {
182 	struct bch_fs *c = trans->c;
183 	struct btree_iter iter;
184 	struct bch_inode_unpacked dir_inode;
185 	struct bch_hash_info dir_hash_info;
186 	int ret;
187 
188 	ret = lookup_first_inode(trans, pos.inode, &dir_inode);
189 	if (ret)
190 		goto err;
191 
192 	dir_hash_info = bch2_hash_info_init(c, &dir_inode);
193 
194 	bch2_trans_iter_init(trans, &iter, BTREE_ID_dirents, pos, BTREE_ITER_intent);
195 
196 	ret =   bch2_btree_iter_traverse(&iter) ?:
197 		bch2_hash_delete_at(trans, bch2_dirent_hash_desc,
198 				    &dir_hash_info, &iter,
199 				    BTREE_UPDATE_internal_snapshot_node);
200 	bch2_trans_iter_exit(trans, &iter);
201 err:
202 	bch_err_fn(c, ret);
203 	return ret;
204 }
205 
206 /* Get lost+found, create if it doesn't exist: */
207 static int lookup_lostfound(struct btree_trans *trans, u32 snapshot,
208 			    struct bch_inode_unpacked *lostfound,
209 			    u64 reattaching_inum)
210 {
211 	struct bch_fs *c = trans->c;
212 	struct qstr lostfound_str = QSTR("lost+found");
213 	u64 inum = 0;
214 	unsigned d_type = 0;
215 	int ret;
216 
217 	struct bch_snapshot_tree st;
218 	ret = bch2_snapshot_tree_lookup(trans,
219 			bch2_snapshot_tree(c, snapshot), &st);
220 	if (ret)
221 		return ret;
222 
223 	subvol_inum root_inum = { .subvol = le32_to_cpu(st.master_subvol) };
224 
225 	struct bch_subvolume subvol;
226 	ret = bch2_subvolume_get(trans, le32_to_cpu(st.master_subvol),
227 				 false, 0, &subvol);
228 	bch_err_msg(c, ret, "looking up root subvol %u for snapshot %u",
229 		    le32_to_cpu(st.master_subvol), snapshot);
230 	if (ret)
231 		return ret;
232 
233 	if (!subvol.inode) {
234 		struct btree_iter iter;
235 		struct bkey_i_subvolume *subvol = bch2_bkey_get_mut_typed(trans, &iter,
236 				BTREE_ID_subvolumes, POS(0, le32_to_cpu(st.master_subvol)),
237 				0, subvolume);
238 		ret = PTR_ERR_OR_ZERO(subvol);
239 		if (ret)
240 			return ret;
241 
242 		subvol->v.inode = cpu_to_le64(reattaching_inum);
243 		bch2_trans_iter_exit(trans, &iter);
244 	}
245 
246 	root_inum.inum = le64_to_cpu(subvol.inode);
247 
248 	struct bch_inode_unpacked root_inode;
249 	struct bch_hash_info root_hash_info;
250 	ret = lookup_inode(trans, root_inum.inum, snapshot, &root_inode);
251 	bch_err_msg(c, ret, "looking up root inode %llu for subvol %u",
252 		    root_inum.inum, le32_to_cpu(st.master_subvol));
253 	if (ret)
254 		return ret;
255 
256 	root_hash_info = bch2_hash_info_init(c, &root_inode);
257 
258 	ret = lookup_dirent_in_snapshot(trans, root_hash_info, root_inum,
259 			      &lostfound_str, &inum, &d_type, snapshot);
260 	if (bch2_err_matches(ret, ENOENT))
261 		goto create_lostfound;
262 
263 	bch_err_fn(c, ret);
264 	if (ret)
265 		return ret;
266 
267 	if (d_type != DT_DIR) {
268 		bch_err(c, "error looking up lost+found: not a directory");
269 		return -BCH_ERR_ENOENT_not_directory;
270 	}
271 
272 	/*
273 	 * The bch2_check_dirents pass has already run, dangling dirents
274 	 * shouldn't exist here:
275 	 */
276 	ret = lookup_inode(trans, inum, snapshot, lostfound);
277 	bch_err_msg(c, ret, "looking up lost+found %llu:%u in (root inode %llu, snapshot root %u)",
278 		    inum, snapshot, root_inum.inum, bch2_snapshot_root(c, snapshot));
279 	return ret;
280 
281 create_lostfound:
282 	/*
283 	 * we always create lost+found in the root snapshot; we don't want
284 	 * different branches of the snapshot tree to have different lost+found
285 	 */
286 	snapshot = le32_to_cpu(st.root_snapshot);
287 	/*
288 	 * XXX: we could have a nicer log message here  if we had a nice way to
289 	 * walk backpointers to print a path
290 	 */
291 	bch_notice(c, "creating lost+found in subvol %llu snapshot %u",
292 		   root_inum.subvol, le32_to_cpu(st.root_snapshot));
293 
294 	u64 now = bch2_current_time(c);
295 	struct btree_iter lostfound_iter = { NULL };
296 	u64 cpu = raw_smp_processor_id();
297 
298 	bch2_inode_init_early(c, lostfound);
299 	bch2_inode_init_late(lostfound, now, 0, 0, S_IFDIR|0700, 0, &root_inode);
300 	lostfound->bi_dir = root_inode.bi_inum;
301 	lostfound->bi_snapshot = le32_to_cpu(st.root_snapshot);
302 
303 	root_inode.bi_nlink++;
304 
305 	ret = bch2_inode_create(trans, &lostfound_iter, lostfound, snapshot, cpu);
306 	if (ret)
307 		goto err;
308 
309 	bch2_btree_iter_set_snapshot(&lostfound_iter, snapshot);
310 	ret = bch2_btree_iter_traverse(&lostfound_iter);
311 	if (ret)
312 		goto err;
313 
314 	ret =   bch2_dirent_create_snapshot(trans,
315 				0, root_inode.bi_inum, snapshot, &root_hash_info,
316 				mode_to_type(lostfound->bi_mode),
317 				&lostfound_str,
318 				lostfound->bi_inum,
319 				&lostfound->bi_dir_offset,
320 				STR_HASH_must_create) ?:
321 		bch2_inode_write_flags(trans, &lostfound_iter, lostfound,
322 				       BTREE_UPDATE_internal_snapshot_node);
323 err:
324 	bch_err_msg(c, ret, "creating lost+found");
325 	bch2_trans_iter_exit(trans, &lostfound_iter);
326 	return ret;
327 }
328 
329 static inline bool inode_should_reattach(struct bch_inode_unpacked *inode)
330 {
331 	if (inode->bi_inum == BCACHEFS_ROOT_INO &&
332 	    inode->bi_subvol == BCACHEFS_ROOT_SUBVOL)
333 		return false;
334 
335 	return !inode->bi_dir && !(inode->bi_flags & BCH_INODE_unlinked);
336 }
337 
338 static int maybe_delete_dirent(struct btree_trans *trans, struct bpos d_pos, u32 snapshot)
339 {
340 	struct btree_iter iter;
341 	struct bkey_s_c k = bch2_bkey_get_iter(trans, &iter, BTREE_ID_dirents,
342 					SPOS(d_pos.inode, d_pos.offset, snapshot),
343 					BTREE_ITER_intent|
344 					BTREE_ITER_with_updates);
345 	int ret = bkey_err(k);
346 	if (ret)
347 		return ret;
348 
349 	if (bpos_eq(k.k->p, d_pos)) {
350 		/*
351 		 * delet_at() doesn't work because the update path doesn't
352 		 * internally use BTREE_ITER_with_updates yet
353 		 */
354 		struct bkey_i *k = bch2_trans_kmalloc(trans, sizeof(*k));
355 		ret = PTR_ERR_OR_ZERO(k);
356 		if (ret)
357 			goto err;
358 
359 		bkey_init(&k->k);
360 		k->k.type = KEY_TYPE_whiteout;
361 		k->k.p = iter.pos;
362 		ret = bch2_trans_update(trans, &iter, k, BTREE_UPDATE_internal_snapshot_node);
363 	}
364 err:
365 	bch2_trans_iter_exit(trans, &iter);
366 	return ret;
367 }
368 
369 static int reattach_inode(struct btree_trans *trans, struct bch_inode_unpacked *inode)
370 {
371 	struct bch_fs *c = trans->c;
372 	struct bch_inode_unpacked lostfound;
373 	char name_buf[20];
374 	int ret;
375 
376 	u32 dirent_snapshot = inode->bi_snapshot;
377 	if (inode->bi_subvol) {
378 		inode->bi_parent_subvol = BCACHEFS_ROOT_SUBVOL;
379 
380 		u64 root_inum;
381 		ret = subvol_lookup(trans, inode->bi_parent_subvol,
382 				    &dirent_snapshot, &root_inum);
383 		if (ret)
384 			return ret;
385 
386 		snprintf(name_buf, sizeof(name_buf), "subvol-%u", inode->bi_subvol);
387 	} else {
388 		snprintf(name_buf, sizeof(name_buf), "%llu", inode->bi_inum);
389 	}
390 
391 	ret = lookup_lostfound(trans, dirent_snapshot, &lostfound, inode->bi_inum);
392 	if (ret)
393 		return ret;
394 
395 	lostfound.bi_nlink += S_ISDIR(inode->bi_mode);
396 
397 	/* ensure lost+found inode is also present in inode snapshot */
398 	if (!inode->bi_subvol) {
399 		BUG_ON(!bch2_snapshot_is_ancestor(c, inode->bi_snapshot, lostfound.bi_snapshot));
400 		lostfound.bi_snapshot = inode->bi_snapshot;
401 	}
402 
403 	ret = __bch2_fsck_write_inode(trans, &lostfound);
404 	if (ret)
405 		return ret;
406 
407 	struct bch_hash_info dir_hash = bch2_hash_info_init(c, &lostfound);
408 	struct qstr name = (struct qstr) QSTR(name_buf);
409 
410 	inode->bi_dir = lostfound.bi_inum;
411 
412 	ret = bch2_dirent_create_snapshot(trans,
413 				inode->bi_parent_subvol, lostfound.bi_inum,
414 				dirent_snapshot,
415 				&dir_hash,
416 				inode_d_type(inode),
417 				&name,
418 				inode->bi_subvol ?: inode->bi_inum,
419 				&inode->bi_dir_offset,
420 				STR_HASH_must_create);
421 	if (ret) {
422 		bch_err_msg(c, ret, "error creating dirent");
423 		return ret;
424 	}
425 
426 	ret = __bch2_fsck_write_inode(trans, inode);
427 	if (ret)
428 		return ret;
429 
430 	/*
431 	 * Fix up inodes in child snapshots: if they should also be reattached
432 	 * update the backpointer field, if they should not be we need to emit
433 	 * whiteouts for the dirent we just created.
434 	 */
435 	if (!inode->bi_subvol && bch2_snapshot_is_leaf(c, inode->bi_snapshot) <= 0) {
436 		snapshot_id_list whiteouts_done;
437 		struct btree_iter iter;
438 		struct bkey_s_c k;
439 
440 		darray_init(&whiteouts_done);
441 
442 		for_each_btree_key_reverse_norestart(trans, iter,
443 				BTREE_ID_inodes, SPOS(0, inode->bi_inum, inode->bi_snapshot - 1),
444 				BTREE_ITER_all_snapshots|BTREE_ITER_intent, k, ret) {
445 			if (k.k->p.offset != inode->bi_inum)
446 				break;
447 
448 			if (!bkey_is_inode(k.k) ||
449 			    !bch2_snapshot_is_ancestor(c, k.k->p.snapshot, inode->bi_snapshot) ||
450 			    snapshot_list_has_ancestor(c, &whiteouts_done, k.k->p.snapshot))
451 				continue;
452 
453 			struct bch_inode_unpacked child_inode;
454 			bch2_inode_unpack(k, &child_inode);
455 
456 			if (!inode_should_reattach(&child_inode)) {
457 				ret = maybe_delete_dirent(trans,
458 							  SPOS(lostfound.bi_inum, inode->bi_dir_offset,
459 							       dirent_snapshot),
460 							  k.k->p.snapshot);
461 				if (ret)
462 					break;
463 
464 				ret = snapshot_list_add(c, &whiteouts_done, k.k->p.snapshot);
465 				if (ret)
466 					break;
467 			} else {
468 				iter.snapshot = k.k->p.snapshot;
469 				child_inode.bi_dir = inode->bi_dir;
470 				child_inode.bi_dir_offset = inode->bi_dir_offset;
471 
472 				ret = bch2_inode_write_flags(trans, &iter, &child_inode,
473 							     BTREE_UPDATE_internal_snapshot_node);
474 				if (ret)
475 					break;
476 			}
477 		}
478 		darray_exit(&whiteouts_done);
479 		bch2_trans_iter_exit(trans, &iter);
480 	}
481 
482 	return ret;
483 }
484 
485 static int remove_backpointer(struct btree_trans *trans,
486 			      struct bch_inode_unpacked *inode)
487 {
488 	if (!inode->bi_dir)
489 		return 0;
490 
491 	struct bch_fs *c = trans->c;
492 	struct btree_iter iter;
493 	struct bkey_s_c_dirent d =
494 		bch2_bkey_get_iter_typed(trans, &iter, BTREE_ID_dirents,
495 				     SPOS(inode->bi_dir, inode->bi_dir_offset, inode->bi_snapshot), 0,
496 				     dirent);
497 	int ret =   bkey_err(d) ?:
498 		dirent_points_to_inode(c, d, inode) ?:
499 		__remove_dirent(trans, d.k->p);
500 	bch2_trans_iter_exit(trans, &iter);
501 	return ret;
502 }
503 
504 static int reattach_subvol(struct btree_trans *trans, struct bkey_s_c_subvolume s)
505 {
506 	struct bch_fs *c = trans->c;
507 
508 	struct bch_inode_unpacked inode;
509 	int ret = bch2_inode_find_by_inum_trans(trans,
510 				(subvol_inum) { s.k->p.offset, le64_to_cpu(s.v->inode) },
511 				&inode);
512 	if (ret)
513 		return ret;
514 
515 	ret = remove_backpointer(trans, &inode);
516 	if (!bch2_err_matches(ret, ENOENT))
517 		bch_err_msg(c, ret, "removing dirent");
518 	if (ret)
519 		return ret;
520 
521 	ret = reattach_inode(trans, &inode);
522 	bch_err_msg(c, ret, "reattaching inode %llu", inode.bi_inum);
523 	return ret;
524 }
525 
526 static int reconstruct_subvol(struct btree_trans *trans, u32 snapshotid, u32 subvolid, u64 inum)
527 {
528 	struct bch_fs *c = trans->c;
529 
530 	if (!bch2_snapshot_is_leaf(c, snapshotid)) {
531 		bch_err(c, "need to reconstruct subvol, but have interior node snapshot");
532 		return -BCH_ERR_fsck_repair_unimplemented;
533 	}
534 
535 	/*
536 	 * If inum isn't set, that means we're being called from check_dirents,
537 	 * not check_inodes - the root of this subvolume doesn't exist or we
538 	 * would have found it there:
539 	 */
540 	if (!inum) {
541 		struct btree_iter inode_iter = {};
542 		struct bch_inode_unpacked new_inode;
543 		u64 cpu = raw_smp_processor_id();
544 
545 		bch2_inode_init_early(c, &new_inode);
546 		bch2_inode_init_late(&new_inode, bch2_current_time(c), 0, 0, S_IFDIR|0755, 0, NULL);
547 
548 		new_inode.bi_subvol = subvolid;
549 
550 		int ret = bch2_inode_create(trans, &inode_iter, &new_inode, snapshotid, cpu) ?:
551 			  bch2_btree_iter_traverse(&inode_iter) ?:
552 			  bch2_inode_write(trans, &inode_iter, &new_inode);
553 		bch2_trans_iter_exit(trans, &inode_iter);
554 		if (ret)
555 			return ret;
556 
557 		inum = new_inode.bi_inum;
558 	}
559 
560 	bch_info(c, "reconstructing subvol %u with root inode %llu", subvolid, inum);
561 
562 	struct bkey_i_subvolume *new_subvol = bch2_trans_kmalloc(trans, sizeof(*new_subvol));
563 	int ret = PTR_ERR_OR_ZERO(new_subvol);
564 	if (ret)
565 		return ret;
566 
567 	bkey_subvolume_init(&new_subvol->k_i);
568 	new_subvol->k.p.offset	= subvolid;
569 	new_subvol->v.snapshot	= cpu_to_le32(snapshotid);
570 	new_subvol->v.inode	= cpu_to_le64(inum);
571 	ret = bch2_btree_insert_trans(trans, BTREE_ID_subvolumes, &new_subvol->k_i, 0);
572 	if (ret)
573 		return ret;
574 
575 	struct btree_iter iter;
576 	struct bkey_i_snapshot *s = bch2_bkey_get_mut_typed(trans, &iter,
577 			BTREE_ID_snapshots, POS(0, snapshotid),
578 			0, snapshot);
579 	ret = PTR_ERR_OR_ZERO(s);
580 	bch_err_msg(c, ret, "getting snapshot %u", snapshotid);
581 	if (ret)
582 		return ret;
583 
584 	u32 snapshot_tree = le32_to_cpu(s->v.tree);
585 
586 	s->v.subvol = cpu_to_le32(subvolid);
587 	SET_BCH_SNAPSHOT_SUBVOL(&s->v, true);
588 	bch2_trans_iter_exit(trans, &iter);
589 
590 	struct bkey_i_snapshot_tree *st = bch2_bkey_get_mut_typed(trans, &iter,
591 			BTREE_ID_snapshot_trees, POS(0, snapshot_tree),
592 			0, snapshot_tree);
593 	ret = PTR_ERR_OR_ZERO(st);
594 	bch_err_msg(c, ret, "getting snapshot tree %u", snapshot_tree);
595 	if (ret)
596 		return ret;
597 
598 	if (!st->v.master_subvol)
599 		st->v.master_subvol = cpu_to_le32(subvolid);
600 
601 	bch2_trans_iter_exit(trans, &iter);
602 	return 0;
603 }
604 
605 static int reconstruct_inode(struct btree_trans *trans, enum btree_id btree, u32 snapshot, u64 inum)
606 {
607 	struct bch_fs *c = trans->c;
608 	unsigned i_mode = S_IFREG;
609 	u64 i_size = 0;
610 
611 	switch (btree) {
612 	case BTREE_ID_extents: {
613 		struct btree_iter iter = {};
614 
615 		bch2_trans_iter_init(trans, &iter, BTREE_ID_extents, SPOS(inum, U64_MAX, snapshot), 0);
616 		struct bkey_s_c k = bch2_btree_iter_peek_prev(&iter);
617 		bch2_trans_iter_exit(trans, &iter);
618 		int ret = bkey_err(k);
619 		if (ret)
620 			return ret;
621 
622 		i_size = k.k->p.offset << 9;
623 		break;
624 	}
625 	case BTREE_ID_dirents:
626 		i_mode = S_IFDIR;
627 		break;
628 	case BTREE_ID_xattrs:
629 		break;
630 	default:
631 		BUG();
632 	}
633 
634 	struct bch_inode_unpacked new_inode;
635 	bch2_inode_init_early(c, &new_inode);
636 	bch2_inode_init_late(&new_inode, bch2_current_time(c), 0, 0, i_mode|0600, 0, NULL);
637 	new_inode.bi_size = i_size;
638 	new_inode.bi_inum = inum;
639 	new_inode.bi_snapshot = snapshot;
640 
641 	return __bch2_fsck_write_inode(trans, &new_inode);
642 }
643 
644 struct snapshots_seen {
645 	struct bpos			pos;
646 	snapshot_id_list		ids;
647 };
648 
649 static inline void snapshots_seen_exit(struct snapshots_seen *s)
650 {
651 	darray_exit(&s->ids);
652 }
653 
654 static inline void snapshots_seen_init(struct snapshots_seen *s)
655 {
656 	memset(s, 0, sizeof(*s));
657 }
658 
659 static int snapshots_seen_add_inorder(struct bch_fs *c, struct snapshots_seen *s, u32 id)
660 {
661 	u32 *i;
662 	__darray_for_each(s->ids, i) {
663 		if (*i == id)
664 			return 0;
665 		if (*i > id)
666 			break;
667 	}
668 
669 	int ret = darray_insert_item(&s->ids, i - s->ids.data, id);
670 	if (ret)
671 		bch_err(c, "error reallocating snapshots_seen table (size %zu)",
672 			s->ids.size);
673 	return ret;
674 }
675 
676 static int snapshots_seen_update(struct bch_fs *c, struct snapshots_seen *s,
677 				 enum btree_id btree_id, struct bpos pos)
678 {
679 	if (!bkey_eq(s->pos, pos))
680 		s->ids.nr = 0;
681 	s->pos = pos;
682 
683 	return snapshot_list_add_nodup(c, &s->ids, pos.snapshot);
684 }
685 
686 /**
687  * key_visible_in_snapshot - returns true if @id is a descendent of @ancestor,
688  * and @ancestor hasn't been overwritten in @seen
689  *
690  * @c:		filesystem handle
691  * @seen:	list of snapshot ids already seen at current position
692  * @id:		descendent snapshot id
693  * @ancestor:	ancestor snapshot id
694  *
695  * Returns:	whether key in @ancestor snapshot is visible in @id snapshot
696  */
697 static bool key_visible_in_snapshot(struct bch_fs *c, struct snapshots_seen *seen,
698 				    u32 id, u32 ancestor)
699 {
700 	ssize_t i;
701 
702 	EBUG_ON(id > ancestor);
703 
704 	/* @ancestor should be the snapshot most recently added to @seen */
705 	EBUG_ON(ancestor != seen->pos.snapshot);
706 	EBUG_ON(ancestor != darray_last(seen->ids));
707 
708 	if (id == ancestor)
709 		return true;
710 
711 	if (!bch2_snapshot_is_ancestor(c, id, ancestor))
712 		return false;
713 
714 	/*
715 	 * We know that @id is a descendant of @ancestor, we're checking if
716 	 * we've seen a key that overwrote @ancestor - i.e. also a descendent of
717 	 * @ascestor and with @id as a descendent.
718 	 *
719 	 * But we already know that we're scanning IDs between @id and @ancestor
720 	 * numerically, since snapshot ID lists are kept sorted, so if we find
721 	 * an id that's an ancestor of @id we're done:
722 	 */
723 
724 	for (i = seen->ids.nr - 2;
725 	     i >= 0 && seen->ids.data[i] >= id;
726 	     --i)
727 		if (bch2_snapshot_is_ancestor(c, id, seen->ids.data[i]))
728 			return false;
729 
730 	return true;
731 }
732 
733 /**
734  * ref_visible - given a key with snapshot id @src that points to a key with
735  * snapshot id @dst, test whether there is some snapshot in which @dst is
736  * visible.
737  *
738  * @c:		filesystem handle
739  * @s:		list of snapshot IDs already seen at @src
740  * @src:	snapshot ID of src key
741  * @dst:	snapshot ID of dst key
742  * Returns:	true if there is some snapshot in which @dst is visible
743  *
744  * Assumes we're visiting @src keys in natural key order
745  */
746 static bool ref_visible(struct bch_fs *c, struct snapshots_seen *s,
747 			u32 src, u32 dst)
748 {
749 	return dst <= src
750 		? key_visible_in_snapshot(c, s, dst, src)
751 		: bch2_snapshot_is_ancestor(c, src, dst);
752 }
753 
754 static int ref_visible2(struct bch_fs *c,
755 			u32 src, struct snapshots_seen *src_seen,
756 			u32 dst, struct snapshots_seen *dst_seen)
757 {
758 	if (dst > src) {
759 		swap(dst, src);
760 		swap(dst_seen, src_seen);
761 	}
762 	return key_visible_in_snapshot(c, src_seen, dst, src);
763 }
764 
765 #define for_each_visible_inode(_c, _s, _w, _snapshot, _i)				\
766 	for (_i = (_w)->inodes.data; _i < (_w)->inodes.data + (_w)->inodes.nr &&	\
767 	     (_i)->snapshot <= (_snapshot); _i++)					\
768 		if (key_visible_in_snapshot(_c, _s, _i->snapshot, _snapshot))
769 
770 struct inode_walker_entry {
771 	struct bch_inode_unpacked inode;
772 	u32			snapshot;
773 	u64			count;
774 };
775 
776 struct inode_walker {
777 	bool				first_this_inode;
778 	bool				have_inodes;
779 	bool				recalculate_sums;
780 	struct bpos			last_pos;
781 
782 	DARRAY(struct inode_walker_entry) inodes;
783 };
784 
785 static void inode_walker_exit(struct inode_walker *w)
786 {
787 	darray_exit(&w->inodes);
788 }
789 
790 static struct inode_walker inode_walker_init(void)
791 {
792 	return (struct inode_walker) { 0, };
793 }
794 
795 static int add_inode(struct bch_fs *c, struct inode_walker *w,
796 		     struct bkey_s_c inode)
797 {
798 	struct bch_inode_unpacked u;
799 
800 	BUG_ON(bch2_inode_unpack(inode, &u));
801 
802 	return darray_push(&w->inodes, ((struct inode_walker_entry) {
803 		.inode		= u,
804 		.snapshot	= inode.k->p.snapshot,
805 	}));
806 }
807 
808 static int get_inodes_all_snapshots(struct btree_trans *trans,
809 				    struct inode_walker *w, u64 inum)
810 {
811 	struct bch_fs *c = trans->c;
812 	struct btree_iter iter;
813 	struct bkey_s_c k;
814 	int ret;
815 
816 	/*
817 	 * We no longer have inodes for w->last_pos; clear this to avoid
818 	 * screwing up check_i_sectors/check_subdir_count if we take a
819 	 * transaction restart here:
820 	 */
821 	w->have_inodes = false;
822 	w->recalculate_sums = false;
823 	w->inodes.nr = 0;
824 
825 	for_each_btree_key_norestart(trans, iter, BTREE_ID_inodes, POS(0, inum),
826 				     BTREE_ITER_all_snapshots, k, ret) {
827 		if (k.k->p.offset != inum)
828 			break;
829 
830 		if (bkey_is_inode(k.k))
831 			add_inode(c, w, k);
832 	}
833 	bch2_trans_iter_exit(trans, &iter);
834 
835 	if (ret)
836 		return ret;
837 
838 	w->first_this_inode = true;
839 	w->have_inodes = true;
840 	return 0;
841 }
842 
843 static struct inode_walker_entry *
844 lookup_inode_for_snapshot(struct bch_fs *c, struct inode_walker *w, struct bkey_s_c k)
845 {
846 	bool is_whiteout = k.k->type == KEY_TYPE_whiteout;
847 
848 	struct inode_walker_entry *i;
849 	__darray_for_each(w->inodes, i)
850 		if (bch2_snapshot_is_ancestor(c, k.k->p.snapshot, i->snapshot))
851 			goto found;
852 
853 	return NULL;
854 found:
855 	BUG_ON(k.k->p.snapshot > i->snapshot);
856 
857 	if (k.k->p.snapshot != i->snapshot && !is_whiteout) {
858 		struct inode_walker_entry new = *i;
859 
860 		new.snapshot = k.k->p.snapshot;
861 		new.count = 0;
862 
863 		struct printbuf buf = PRINTBUF;
864 		bch2_bkey_val_to_text(&buf, c, k);
865 
866 		bch_info(c, "have key for inode %llu:%u but have inode in ancestor snapshot %u\n"
867 			 "unexpected because we should always update the inode when we update a key in that inode\n"
868 			 "%s",
869 			 w->last_pos.inode, k.k->p.snapshot, i->snapshot, buf.buf);
870 		printbuf_exit(&buf);
871 
872 		while (i > w->inodes.data && i[-1].snapshot > k.k->p.snapshot)
873 			--i;
874 
875 		size_t pos = i - w->inodes.data;
876 		int ret = darray_insert_item(&w->inodes, pos, new);
877 		if (ret)
878 			return ERR_PTR(ret);
879 
880 		i = w->inodes.data + pos;
881 	}
882 
883 	return i;
884 }
885 
886 static struct inode_walker_entry *walk_inode(struct btree_trans *trans,
887 					     struct inode_walker *w,
888 					     struct bkey_s_c k)
889 {
890 	if (w->last_pos.inode != k.k->p.inode) {
891 		int ret = get_inodes_all_snapshots(trans, w, k.k->p.inode);
892 		if (ret)
893 			return ERR_PTR(ret);
894 	}
895 
896 	w->last_pos = k.k->p;
897 
898 	return lookup_inode_for_snapshot(trans->c, w, k);
899 }
900 
901 static int get_visible_inodes(struct btree_trans *trans,
902 			      struct inode_walker *w,
903 			      struct snapshots_seen *s,
904 			      u64 inum)
905 {
906 	struct bch_fs *c = trans->c;
907 	struct btree_iter iter;
908 	struct bkey_s_c k;
909 	int ret;
910 
911 	w->inodes.nr = 0;
912 
913 	for_each_btree_key_norestart(trans, iter, BTREE_ID_inodes, POS(0, inum),
914 			   BTREE_ITER_all_snapshots, k, ret) {
915 		if (k.k->p.offset != inum)
916 			break;
917 
918 		if (!ref_visible(c, s, s->pos.snapshot, k.k->p.snapshot))
919 			continue;
920 
921 		if (bkey_is_inode(k.k))
922 			add_inode(c, w, k);
923 
924 		if (k.k->p.snapshot >= s->pos.snapshot)
925 			break;
926 	}
927 	bch2_trans_iter_exit(trans, &iter);
928 
929 	return ret;
930 }
931 
932 static int hash_redo_key(struct btree_trans *trans,
933 			 const struct bch_hash_desc desc,
934 			 struct bch_hash_info *hash_info,
935 			 struct btree_iter *k_iter, struct bkey_s_c k)
936 {
937 	struct bkey_i *delete;
938 	struct bkey_i *tmp;
939 
940 	delete = bch2_trans_kmalloc(trans, sizeof(*delete));
941 	if (IS_ERR(delete))
942 		return PTR_ERR(delete);
943 
944 	tmp = bch2_bkey_make_mut_noupdate(trans, k);
945 	if (IS_ERR(tmp))
946 		return PTR_ERR(tmp);
947 
948 	bkey_init(&delete->k);
949 	delete->k.p = k_iter->pos;
950 	return  bch2_btree_iter_traverse(k_iter) ?:
951 		bch2_trans_update(trans, k_iter, delete, 0) ?:
952 		bch2_hash_set_in_snapshot(trans, desc, hash_info,
953 				       (subvol_inum) { 0, k.k->p.inode },
954 				       k.k->p.snapshot, tmp,
955 				       STR_HASH_must_create|
956 				       BTREE_UPDATE_internal_snapshot_node) ?:
957 		bch2_trans_commit(trans, NULL, NULL, BCH_TRANS_COMMIT_no_enospc);
958 }
959 
960 static int hash_check_key(struct btree_trans *trans,
961 			  const struct bch_hash_desc desc,
962 			  struct bch_hash_info *hash_info,
963 			  struct btree_iter *k_iter, struct bkey_s_c hash_k)
964 {
965 	struct bch_fs *c = trans->c;
966 	struct btree_iter iter = { NULL };
967 	struct printbuf buf = PRINTBUF;
968 	struct bkey_s_c k;
969 	u64 hash;
970 	int ret = 0;
971 
972 	if (hash_k.k->type != desc.key_type)
973 		return 0;
974 
975 	hash = desc.hash_bkey(hash_info, hash_k);
976 
977 	if (likely(hash == hash_k.k->p.offset))
978 		return 0;
979 
980 	if (hash_k.k->p.offset < hash)
981 		goto bad_hash;
982 
983 	for_each_btree_key_norestart(trans, iter, desc.btree_id,
984 				     SPOS(hash_k.k->p.inode, hash, hash_k.k->p.snapshot),
985 				     BTREE_ITER_slots, k, ret) {
986 		if (bkey_eq(k.k->p, hash_k.k->p))
987 			break;
988 
989 		if (fsck_err_on(k.k->type == desc.key_type &&
990 				!desc.cmp_bkey(k, hash_k),
991 				trans, hash_table_key_duplicate,
992 				"duplicate hash table keys:\n%s",
993 				(printbuf_reset(&buf),
994 				 bch2_bkey_val_to_text(&buf, c, hash_k),
995 				 buf.buf))) {
996 			ret = bch2_hash_delete_at(trans, desc, hash_info, k_iter, 0) ?: 1;
997 			break;
998 		}
999 
1000 		if (bkey_deleted(k.k)) {
1001 			bch2_trans_iter_exit(trans, &iter);
1002 			goto bad_hash;
1003 		}
1004 	}
1005 out:
1006 	bch2_trans_iter_exit(trans, &iter);
1007 	printbuf_exit(&buf);
1008 	return ret;
1009 bad_hash:
1010 	if (fsck_err(trans, hash_table_key_wrong_offset,
1011 		     "hash table key at wrong offset: btree %s inode %llu offset %llu, hashed to %llu\n%s",
1012 		     bch2_btree_id_str(desc.btree_id), hash_k.k->p.inode, hash_k.k->p.offset, hash,
1013 		     (printbuf_reset(&buf),
1014 		      bch2_bkey_val_to_text(&buf, c, hash_k), buf.buf))) {
1015 		ret = hash_redo_key(trans, desc, hash_info, k_iter, hash_k);
1016 		bch_err_fn(c, ret);
1017 		if (ret)
1018 			return ret;
1019 		ret = -BCH_ERR_transaction_restart_nested;
1020 	}
1021 fsck_err:
1022 	goto out;
1023 }
1024 
1025 static struct bkey_s_c_dirent dirent_get_by_pos(struct btree_trans *trans,
1026 						struct btree_iter *iter,
1027 						struct bpos pos)
1028 {
1029 	return bch2_bkey_get_iter_typed(trans, iter, BTREE_ID_dirents, pos, 0, dirent);
1030 }
1031 
1032 static struct bkey_s_c_dirent inode_get_dirent(struct btree_trans *trans,
1033 					       struct btree_iter *iter,
1034 					       struct bch_inode_unpacked *inode,
1035 					       u32 *snapshot)
1036 {
1037 	if (inode->bi_subvol) {
1038 		u64 inum;
1039 		int ret = subvol_lookup(trans, inode->bi_parent_subvol, snapshot, &inum);
1040 		if (ret)
1041 			return ((struct bkey_s_c_dirent) { .k = ERR_PTR(ret) });
1042 	}
1043 
1044 	return dirent_get_by_pos(trans, iter, SPOS(inode->bi_dir, inode->bi_dir_offset, *snapshot));
1045 }
1046 
1047 static int check_inode_deleted_list(struct btree_trans *trans, struct bpos p)
1048 {
1049 	struct btree_iter iter;
1050 	struct bkey_s_c k = bch2_bkey_get_iter(trans, &iter, BTREE_ID_deleted_inodes, p, 0);
1051 	int ret = bkey_err(k) ?: k.k->type == KEY_TYPE_set;
1052 	bch2_trans_iter_exit(trans, &iter);
1053 	return ret;
1054 }
1055 
1056 static int check_inode_dirent_inode(struct btree_trans *trans,
1057 				    struct bch_inode_unpacked *inode,
1058 				    bool *write_inode)
1059 {
1060 	struct bch_fs *c = trans->c;
1061 	struct printbuf buf = PRINTBUF;
1062 
1063 	u32 inode_snapshot = inode->bi_snapshot;
1064 	struct btree_iter dirent_iter = {};
1065 	struct bkey_s_c_dirent d = inode_get_dirent(trans, &dirent_iter, inode, &inode_snapshot);
1066 	int ret = bkey_err(d);
1067 	if (ret && !bch2_err_matches(ret, ENOENT))
1068 		return ret;
1069 
1070 	if (fsck_err_on(ret,
1071 			trans, inode_points_to_missing_dirent,
1072 			"inode points to missing dirent\n%s",
1073 			(bch2_inode_unpacked_to_text(&buf, inode), buf.buf)) ||
1074 	    fsck_err_on(!ret && dirent_points_to_inode_nowarn(d, inode),
1075 			trans, inode_points_to_wrong_dirent,
1076 			"%s",
1077 			(printbuf_reset(&buf),
1078 			 dirent_inode_mismatch_msg(&buf, c, d, inode),
1079 			 buf.buf))) {
1080 		/*
1081 		 * We just clear the backpointer fields for now. If we find a
1082 		 * dirent that points to this inode in check_dirents(), we'll
1083 		 * update it then; then when we get to check_path() if the
1084 		 * backpointer is still 0 we'll reattach it.
1085 		 */
1086 		inode->bi_dir = 0;
1087 		inode->bi_dir_offset = 0;
1088 		*write_inode = true;
1089 	}
1090 
1091 	ret = 0;
1092 fsck_err:
1093 	bch2_trans_iter_exit(trans, &dirent_iter);
1094 	printbuf_exit(&buf);
1095 	bch_err_fn(c, ret);
1096 	return ret;
1097 }
1098 
1099 static int check_inode(struct btree_trans *trans,
1100 		       struct btree_iter *iter,
1101 		       struct bkey_s_c k,
1102 		       struct bch_inode_unpacked *prev,
1103 		       struct snapshots_seen *s)
1104 {
1105 	struct bch_fs *c = trans->c;
1106 	struct printbuf buf = PRINTBUF;
1107 	struct bch_inode_unpacked u;
1108 	bool do_update = false;
1109 	int ret;
1110 
1111 	ret = bch2_check_key_has_snapshot(trans, iter, k);
1112 	if (ret < 0)
1113 		goto err;
1114 	if (ret)
1115 		return 0;
1116 
1117 	ret = snapshots_seen_update(c, s, iter->btree_id, k.k->p);
1118 	if (ret)
1119 		goto err;
1120 
1121 	if (!bkey_is_inode(k.k))
1122 		return 0;
1123 
1124 	BUG_ON(bch2_inode_unpack(k, &u));
1125 
1126 	if (prev->bi_inum != u.bi_inum)
1127 		*prev = u;
1128 
1129 	if (fsck_err_on(prev->bi_hash_seed	!= u.bi_hash_seed ||
1130 			inode_d_type(prev)	!= inode_d_type(&u),
1131 			trans, inode_snapshot_mismatch,
1132 			"inodes in different snapshots don't match")) {
1133 		bch_err(c, "repair not implemented yet");
1134 		ret = -BCH_ERR_fsck_repair_unimplemented;
1135 		goto err_noprint;
1136 	}
1137 
1138 	if (u.bi_dir || u.bi_dir_offset) {
1139 		ret = check_inode_dirent_inode(trans, &u, &do_update);
1140 		if (ret)
1141 			goto err;
1142 	}
1143 
1144 	if (fsck_err_on(u.bi_dir && (u.bi_flags & BCH_INODE_unlinked),
1145 			trans, inode_unlinked_but_has_dirent,
1146 			"inode unlinked but has dirent\n%s",
1147 			(printbuf_reset(&buf),
1148 			 bch2_inode_unpacked_to_text(&buf, &u),
1149 			 buf.buf))) {
1150 		u.bi_flags &= ~BCH_INODE_unlinked;
1151 		do_update = true;
1152 	}
1153 
1154 	if (S_ISDIR(u.bi_mode) && (u.bi_flags & BCH_INODE_unlinked)) {
1155 		/* Check for this early so that check_unreachable_inode() will reattach it */
1156 
1157 		ret = bch2_empty_dir_snapshot(trans, k.k->p.offset, 0, k.k->p.snapshot);
1158 		if (ret && ret != -BCH_ERR_ENOTEMPTY_dir_not_empty)
1159 			goto err;
1160 
1161 		fsck_err_on(ret, trans, inode_dir_unlinked_but_not_empty,
1162 			    "dir unlinked but not empty\n%s",
1163 			    (printbuf_reset(&buf),
1164 			     bch2_inode_unpacked_to_text(&buf, &u),
1165 			     buf.buf));
1166 		u.bi_flags &= ~BCH_INODE_unlinked;
1167 		do_update = true;
1168 		ret = 0;
1169 	}
1170 
1171 	ret = bch2_inode_has_child_snapshots(trans, k.k->p);
1172 	if (ret < 0)
1173 		goto err;
1174 
1175 	if (fsck_err_on(ret != !!(u.bi_flags & BCH_INODE_has_child_snapshot),
1176 			trans, inode_has_child_snapshots_wrong,
1177 			"inode has_child_snapshots flag wrong (should be %u)\n%s",
1178 			ret,
1179 			(printbuf_reset(&buf),
1180 			 bch2_inode_unpacked_to_text(&buf, &u),
1181 			 buf.buf))) {
1182 		if (ret)
1183 			u.bi_flags |= BCH_INODE_has_child_snapshot;
1184 		else
1185 			u.bi_flags &= ~BCH_INODE_has_child_snapshot;
1186 		do_update = true;
1187 	}
1188 	ret = 0;
1189 
1190 	if ((u.bi_flags & BCH_INODE_unlinked) &&
1191 	    !(u.bi_flags & BCH_INODE_has_child_snapshot)) {
1192 		if (!test_bit(BCH_FS_started, &c->flags)) {
1193 			/*
1194 			 * If we're not in online fsck, don't delete unlinked
1195 			 * inodes, just make sure they're on the deleted list.
1196 			 *
1197 			 * They might be referred to by a logged operation -
1198 			 * i.e. we might have crashed in the middle of a
1199 			 * truncate on an unlinked but open file - so we want to
1200 			 * let the delete_dead_inodes kill it after resuming
1201 			 * logged ops.
1202 			 */
1203 			ret = check_inode_deleted_list(trans, k.k->p);
1204 			if (ret < 0)
1205 				goto err_noprint;
1206 
1207 			fsck_err_on(!ret,
1208 				    trans, unlinked_inode_not_on_deleted_list,
1209 				    "inode %llu:%u unlinked, but not on deleted list",
1210 				    u.bi_inum, k.k->p.snapshot);
1211 
1212 			ret = bch2_btree_bit_mod_buffered(trans, BTREE_ID_deleted_inodes, k.k->p, 1);
1213 			if (ret)
1214 				goto err;
1215 		} else {
1216 			ret = bch2_inode_or_descendents_is_open(trans, k.k->p);
1217 			if (ret < 0)
1218 				goto err;
1219 
1220 			if (fsck_err_on(!ret,
1221 					trans, inode_unlinked_and_not_open,
1222 				      "inode %llu%u unlinked and not open",
1223 				      u.bi_inum, u.bi_snapshot)) {
1224 				ret = bch2_inode_rm_snapshot(trans, u.bi_inum, iter->pos.snapshot);
1225 				bch_err_msg(c, ret, "in fsck deleting inode");
1226 				goto err_noprint;
1227 			}
1228 			ret = 0;
1229 		}
1230 	}
1231 
1232 	if (fsck_err_on(u.bi_parent_subvol &&
1233 			(u.bi_subvol == 0 ||
1234 			 u.bi_subvol == BCACHEFS_ROOT_SUBVOL),
1235 			trans, inode_bi_parent_nonzero,
1236 			"inode %llu:%u has subvol %u but nonzero parent subvol %u",
1237 			u.bi_inum, k.k->p.snapshot, u.bi_subvol, u.bi_parent_subvol)) {
1238 		u.bi_parent_subvol = 0;
1239 		do_update = true;
1240 	}
1241 
1242 	if (u.bi_subvol) {
1243 		struct bch_subvolume s;
1244 
1245 		ret = bch2_subvolume_get(trans, u.bi_subvol, false, 0, &s);
1246 		if (ret && !bch2_err_matches(ret, ENOENT))
1247 			goto err;
1248 
1249 		if (ret && (c->sb.btrees_lost_data & BIT_ULL(BTREE_ID_subvolumes))) {
1250 			ret = reconstruct_subvol(trans, k.k->p.snapshot, u.bi_subvol, u.bi_inum);
1251 			goto do_update;
1252 		}
1253 
1254 		if (fsck_err_on(ret,
1255 				trans, inode_bi_subvol_missing,
1256 				"inode %llu:%u bi_subvol points to missing subvolume %u",
1257 				u.bi_inum, k.k->p.snapshot, u.bi_subvol) ||
1258 		    fsck_err_on(le64_to_cpu(s.inode) != u.bi_inum ||
1259 				!bch2_snapshot_is_ancestor(c, le32_to_cpu(s.snapshot),
1260 							   k.k->p.snapshot),
1261 				trans, inode_bi_subvol_wrong,
1262 				"inode %llu:%u points to subvol %u, but subvol points to %llu:%u",
1263 				u.bi_inum, k.k->p.snapshot, u.bi_subvol,
1264 				le64_to_cpu(s.inode),
1265 				le32_to_cpu(s.snapshot))) {
1266 			u.bi_subvol = 0;
1267 			u.bi_parent_subvol = 0;
1268 			do_update = true;
1269 		}
1270 	}
1271 do_update:
1272 	if (do_update) {
1273 		ret = __bch2_fsck_write_inode(trans, &u);
1274 		bch_err_msg(c, ret, "in fsck updating inode");
1275 		if (ret)
1276 			goto err_noprint;
1277 	}
1278 err:
1279 fsck_err:
1280 	bch_err_fn(c, ret);
1281 err_noprint:
1282 	printbuf_exit(&buf);
1283 	return ret;
1284 }
1285 
1286 int bch2_check_inodes(struct bch_fs *c)
1287 {
1288 	struct bch_inode_unpacked prev = { 0 };
1289 	struct snapshots_seen s;
1290 
1291 	snapshots_seen_init(&s);
1292 
1293 	int ret = bch2_trans_run(c,
1294 		for_each_btree_key_commit(trans, iter, BTREE_ID_inodes,
1295 				POS_MIN,
1296 				BTREE_ITER_prefetch|BTREE_ITER_all_snapshots, k,
1297 				NULL, NULL, BCH_TRANS_COMMIT_no_enospc,
1298 			check_inode(trans, &iter, k, &prev, &s)));
1299 
1300 	snapshots_seen_exit(&s);
1301 	bch_err_fn(c, ret);
1302 	return ret;
1303 }
1304 
1305 static int find_oldest_inode_needs_reattach(struct btree_trans *trans,
1306 					    struct bch_inode_unpacked *inode)
1307 {
1308 	struct bch_fs *c = trans->c;
1309 	struct btree_iter iter;
1310 	struct bkey_s_c k;
1311 	int ret = 0;
1312 
1313 	/*
1314 	 * We look for inodes to reattach in natural key order, leaves first,
1315 	 * but we should do the reattach at the oldest version that needs to be
1316 	 * reattached:
1317 	 */
1318 	for_each_btree_key_norestart(trans, iter,
1319 				     BTREE_ID_inodes,
1320 				     SPOS(0, inode->bi_inum, inode->bi_snapshot + 1),
1321 				     BTREE_ITER_all_snapshots, k, ret) {
1322 		if (k.k->p.offset != inode->bi_inum)
1323 			break;
1324 
1325 		if (!bch2_snapshot_is_ancestor(c, inode->bi_snapshot, k.k->p.snapshot))
1326 			continue;
1327 
1328 		if (!bkey_is_inode(k.k))
1329 			break;
1330 
1331 		struct bch_inode_unpacked parent_inode;
1332 		bch2_inode_unpack(k, &parent_inode);
1333 
1334 		if (!inode_should_reattach(&parent_inode))
1335 			break;
1336 
1337 		*inode = parent_inode;
1338 	}
1339 	bch2_trans_iter_exit(trans, &iter);
1340 
1341 	return ret;
1342 }
1343 
1344 static int check_unreachable_inode(struct btree_trans *trans,
1345 				   struct btree_iter *iter,
1346 				   struct bkey_s_c k)
1347 {
1348 	struct printbuf buf = PRINTBUF;
1349 	int ret = 0;
1350 
1351 	if (!bkey_is_inode(k.k))
1352 		return 0;
1353 
1354 	struct bch_inode_unpacked inode;
1355 	BUG_ON(bch2_inode_unpack(k, &inode));
1356 
1357 	if (!inode_should_reattach(&inode))
1358 		return 0;
1359 
1360 	ret = find_oldest_inode_needs_reattach(trans, &inode);
1361 	if (ret)
1362 		return ret;
1363 
1364 	if (fsck_err(trans, inode_unreachable,
1365 		     "unreachable inode:\n%s",
1366 		     (bch2_inode_unpacked_to_text(&buf, &inode),
1367 		      buf.buf)))
1368 		ret = reattach_inode(trans, &inode);
1369 fsck_err:
1370 	printbuf_exit(&buf);
1371 	return ret;
1372 }
1373 
1374 /*
1375  * Reattach unreachable (but not unlinked) inodes
1376  *
1377  * Run after check_inodes() and check_dirents(), so we node that inode
1378  * backpointer fields point to valid dirents, and every inode that has a dirent
1379  * that points to it has its backpointer field set - so we're just looking for
1380  * non-unlinked inodes without backpointers:
1381  *
1382  * XXX: this is racy w.r.t. hardlink removal in online fsck
1383  */
1384 int bch2_check_unreachable_inodes(struct bch_fs *c)
1385 {
1386 	int ret = bch2_trans_run(c,
1387 		for_each_btree_key_commit(trans, iter, BTREE_ID_inodes,
1388 				POS_MIN,
1389 				BTREE_ITER_prefetch|BTREE_ITER_all_snapshots, k,
1390 				NULL, NULL, BCH_TRANS_COMMIT_no_enospc,
1391 			check_unreachable_inode(trans, &iter, k)));
1392 	bch_err_fn(c, ret);
1393 	return ret;
1394 }
1395 
1396 static inline bool btree_matches_i_mode(enum btree_id btree, unsigned mode)
1397 {
1398 	switch (btree) {
1399 	case BTREE_ID_extents:
1400 		return S_ISREG(mode) || S_ISLNK(mode);
1401 	case BTREE_ID_dirents:
1402 		return S_ISDIR(mode);
1403 	case BTREE_ID_xattrs:
1404 		return true;
1405 	default:
1406 		BUG();
1407 	}
1408 }
1409 
1410 static int check_key_has_inode(struct btree_trans *trans,
1411 			       struct btree_iter *iter,
1412 			       struct inode_walker *inode,
1413 			       struct inode_walker_entry *i,
1414 			       struct bkey_s_c k)
1415 {
1416 	struct bch_fs *c = trans->c;
1417 	struct printbuf buf = PRINTBUF;
1418 	int ret = PTR_ERR_OR_ZERO(i);
1419 	if (ret)
1420 		return ret;
1421 
1422 	if (k.k->type == KEY_TYPE_whiteout)
1423 		goto out;
1424 
1425 	if (!i && (c->sb.btrees_lost_data & BIT_ULL(BTREE_ID_inodes))) {
1426 		ret =   reconstruct_inode(trans, iter->btree_id, k.k->p.snapshot, k.k->p.inode) ?:
1427 			bch2_trans_commit(trans, NULL, NULL, BCH_TRANS_COMMIT_no_enospc);
1428 		if (ret)
1429 			goto err;
1430 
1431 		inode->last_pos.inode--;
1432 		ret = -BCH_ERR_transaction_restart_nested;
1433 		goto err;
1434 	}
1435 
1436 	if (fsck_err_on(!i,
1437 			trans, key_in_missing_inode,
1438 			"key in missing inode:\n  %s",
1439 			(printbuf_reset(&buf),
1440 			 bch2_bkey_val_to_text(&buf, c, k), buf.buf)))
1441 		goto delete;
1442 
1443 	if (fsck_err_on(i && !btree_matches_i_mode(iter->btree_id, i->inode.bi_mode),
1444 			trans, key_in_wrong_inode_type,
1445 			"key for wrong inode mode %o:\n  %s",
1446 			i->inode.bi_mode,
1447 			(printbuf_reset(&buf),
1448 			 bch2_bkey_val_to_text(&buf, c, k), buf.buf)))
1449 		goto delete;
1450 out:
1451 err:
1452 fsck_err:
1453 	printbuf_exit(&buf);
1454 	bch_err_fn(c, ret);
1455 	return ret;
1456 delete:
1457 	ret = bch2_btree_delete_at(trans, iter, BTREE_UPDATE_internal_snapshot_node);
1458 	goto out;
1459 }
1460 
1461 static int check_i_sectors_notnested(struct btree_trans *trans, struct inode_walker *w)
1462 {
1463 	struct bch_fs *c = trans->c;
1464 	int ret = 0;
1465 	s64 count2;
1466 
1467 	darray_for_each(w->inodes, i) {
1468 		if (i->inode.bi_sectors == i->count)
1469 			continue;
1470 
1471 		count2 = bch2_count_inode_sectors(trans, w->last_pos.inode, i->snapshot);
1472 
1473 		if (w->recalculate_sums)
1474 			i->count = count2;
1475 
1476 		if (i->count != count2) {
1477 			bch_err_ratelimited(c, "fsck counted i_sectors wrong for inode %llu:%u: got %llu should be %llu",
1478 					    w->last_pos.inode, i->snapshot, i->count, count2);
1479 			return -BCH_ERR_internal_fsck_err;
1480 		}
1481 
1482 		if (fsck_err_on(!(i->inode.bi_flags & BCH_INODE_i_sectors_dirty),
1483 				trans, inode_i_sectors_wrong,
1484 				"inode %llu:%u has incorrect i_sectors: got %llu, should be %llu",
1485 				w->last_pos.inode, i->snapshot,
1486 				i->inode.bi_sectors, i->count)) {
1487 			i->inode.bi_sectors = i->count;
1488 			ret = bch2_fsck_write_inode(trans, &i->inode);
1489 			if (ret)
1490 				break;
1491 		}
1492 	}
1493 fsck_err:
1494 	bch_err_fn(c, ret);
1495 	return ret;
1496 }
1497 
1498 static int check_i_sectors(struct btree_trans *trans, struct inode_walker *w)
1499 {
1500 	u32 restart_count = trans->restart_count;
1501 	return check_i_sectors_notnested(trans, w) ?:
1502 		trans_was_restarted(trans, restart_count);
1503 }
1504 
1505 struct extent_end {
1506 	u32			snapshot;
1507 	u64			offset;
1508 	struct snapshots_seen	seen;
1509 };
1510 
1511 struct extent_ends {
1512 	struct bpos			last_pos;
1513 	DARRAY(struct extent_end)	e;
1514 };
1515 
1516 static void extent_ends_reset(struct extent_ends *extent_ends)
1517 {
1518 	darray_for_each(extent_ends->e, i)
1519 		snapshots_seen_exit(&i->seen);
1520 	extent_ends->e.nr = 0;
1521 }
1522 
1523 static void extent_ends_exit(struct extent_ends *extent_ends)
1524 {
1525 	extent_ends_reset(extent_ends);
1526 	darray_exit(&extent_ends->e);
1527 }
1528 
1529 static void extent_ends_init(struct extent_ends *extent_ends)
1530 {
1531 	memset(extent_ends, 0, sizeof(*extent_ends));
1532 }
1533 
1534 static int extent_ends_at(struct bch_fs *c,
1535 			  struct extent_ends *extent_ends,
1536 			  struct snapshots_seen *seen,
1537 			  struct bkey_s_c k)
1538 {
1539 	struct extent_end *i, n = (struct extent_end) {
1540 		.offset		= k.k->p.offset,
1541 		.snapshot	= k.k->p.snapshot,
1542 		.seen		= *seen,
1543 	};
1544 
1545 	n.seen.ids.data = kmemdup(seen->ids.data,
1546 			      sizeof(seen->ids.data[0]) * seen->ids.size,
1547 			      GFP_KERNEL);
1548 	if (!n.seen.ids.data)
1549 		return -BCH_ERR_ENOMEM_fsck_extent_ends_at;
1550 
1551 	__darray_for_each(extent_ends->e, i) {
1552 		if (i->snapshot == k.k->p.snapshot) {
1553 			snapshots_seen_exit(&i->seen);
1554 			*i = n;
1555 			return 0;
1556 		}
1557 
1558 		if (i->snapshot >= k.k->p.snapshot)
1559 			break;
1560 	}
1561 
1562 	return darray_insert_item(&extent_ends->e, i - extent_ends->e.data, n);
1563 }
1564 
1565 static int overlapping_extents_found(struct btree_trans *trans,
1566 				     enum btree_id btree,
1567 				     struct bpos pos1, struct snapshots_seen *pos1_seen,
1568 				     struct bkey pos2,
1569 				     bool *fixed,
1570 				     struct extent_end *extent_end)
1571 {
1572 	struct bch_fs *c = trans->c;
1573 	struct printbuf buf = PRINTBUF;
1574 	struct btree_iter iter1, iter2 = { NULL };
1575 	struct bkey_s_c k1, k2;
1576 	int ret;
1577 
1578 	BUG_ON(bkey_le(pos1, bkey_start_pos(&pos2)));
1579 
1580 	bch2_trans_iter_init(trans, &iter1, btree, pos1,
1581 			     BTREE_ITER_all_snapshots|
1582 			     BTREE_ITER_not_extents);
1583 	k1 = bch2_btree_iter_peek_upto(&iter1, POS(pos1.inode, U64_MAX));
1584 	ret = bkey_err(k1);
1585 	if (ret)
1586 		goto err;
1587 
1588 	prt_str(&buf, "\n  ");
1589 	bch2_bkey_val_to_text(&buf, c, k1);
1590 
1591 	if (!bpos_eq(pos1, k1.k->p)) {
1592 		prt_str(&buf, "\n  wanted\n  ");
1593 		bch2_bpos_to_text(&buf, pos1);
1594 		prt_str(&buf, "\n  ");
1595 		bch2_bkey_to_text(&buf, &pos2);
1596 
1597 		bch_err(c, "%s: error finding first overlapping extent when repairing, got%s",
1598 			__func__, buf.buf);
1599 		ret = -BCH_ERR_internal_fsck_err;
1600 		goto err;
1601 	}
1602 
1603 	bch2_trans_copy_iter(&iter2, &iter1);
1604 
1605 	while (1) {
1606 		bch2_btree_iter_advance(&iter2);
1607 
1608 		k2 = bch2_btree_iter_peek_upto(&iter2, POS(pos1.inode, U64_MAX));
1609 		ret = bkey_err(k2);
1610 		if (ret)
1611 			goto err;
1612 
1613 		if (bpos_ge(k2.k->p, pos2.p))
1614 			break;
1615 	}
1616 
1617 	prt_str(&buf, "\n  ");
1618 	bch2_bkey_val_to_text(&buf, c, k2);
1619 
1620 	if (bpos_gt(k2.k->p, pos2.p) ||
1621 	    pos2.size != k2.k->size) {
1622 		bch_err(c, "%s: error finding seconding overlapping extent when repairing%s",
1623 			__func__, buf.buf);
1624 		ret = -BCH_ERR_internal_fsck_err;
1625 		goto err;
1626 	}
1627 
1628 	prt_printf(&buf, "\n  overwriting %s extent",
1629 		   pos1.snapshot >= pos2.p.snapshot ? "first" : "second");
1630 
1631 	if (fsck_err(trans, extent_overlapping,
1632 		     "overlapping extents%s", buf.buf)) {
1633 		struct btree_iter *old_iter = &iter1;
1634 		struct disk_reservation res = { 0 };
1635 
1636 		if (pos1.snapshot < pos2.p.snapshot) {
1637 			old_iter = &iter2;
1638 			swap(k1, k2);
1639 		}
1640 
1641 		trans->extra_disk_res += bch2_bkey_sectors_compressed(k2);
1642 
1643 		ret =   bch2_trans_update_extent_overwrite(trans, old_iter,
1644 				BTREE_UPDATE_internal_snapshot_node,
1645 				k1, k2) ?:
1646 			bch2_trans_commit(trans, &res, NULL, BCH_TRANS_COMMIT_no_enospc);
1647 		bch2_disk_reservation_put(c, &res);
1648 
1649 		if (ret)
1650 			goto err;
1651 
1652 		*fixed = true;
1653 
1654 		if (pos1.snapshot == pos2.p.snapshot) {
1655 			/*
1656 			 * We overwrote the first extent, and did the overwrite
1657 			 * in the same snapshot:
1658 			 */
1659 			extent_end->offset = bkey_start_offset(&pos2);
1660 		} else if (pos1.snapshot > pos2.p.snapshot) {
1661 			/*
1662 			 * We overwrote the first extent in pos2's snapshot:
1663 			 */
1664 			ret = snapshots_seen_add_inorder(c, pos1_seen, pos2.p.snapshot);
1665 		} else {
1666 			/*
1667 			 * We overwrote the second extent - restart
1668 			 * check_extent() from the top:
1669 			 */
1670 			ret = -BCH_ERR_transaction_restart_nested;
1671 		}
1672 	}
1673 fsck_err:
1674 err:
1675 	bch2_trans_iter_exit(trans, &iter2);
1676 	bch2_trans_iter_exit(trans, &iter1);
1677 	printbuf_exit(&buf);
1678 	return ret;
1679 }
1680 
1681 static int check_overlapping_extents(struct btree_trans *trans,
1682 			      struct snapshots_seen *seen,
1683 			      struct extent_ends *extent_ends,
1684 			      struct bkey_s_c k,
1685 			      struct btree_iter *iter,
1686 			      bool *fixed)
1687 {
1688 	struct bch_fs *c = trans->c;
1689 	int ret = 0;
1690 
1691 	/* transaction restart, running again */
1692 	if (bpos_eq(extent_ends->last_pos, k.k->p))
1693 		return 0;
1694 
1695 	if (extent_ends->last_pos.inode != k.k->p.inode)
1696 		extent_ends_reset(extent_ends);
1697 
1698 	darray_for_each(extent_ends->e, i) {
1699 		if (i->offset <= bkey_start_offset(k.k))
1700 			continue;
1701 
1702 		if (!ref_visible2(c,
1703 				  k.k->p.snapshot, seen,
1704 				  i->snapshot, &i->seen))
1705 			continue;
1706 
1707 		ret = overlapping_extents_found(trans, iter->btree_id,
1708 						SPOS(iter->pos.inode,
1709 						     i->offset,
1710 						     i->snapshot),
1711 						&i->seen,
1712 						*k.k, fixed, i);
1713 		if (ret)
1714 			goto err;
1715 	}
1716 
1717 	extent_ends->last_pos = k.k->p;
1718 err:
1719 	return ret;
1720 }
1721 
1722 static int check_extent_overbig(struct btree_trans *trans, struct btree_iter *iter,
1723 				struct bkey_s_c k)
1724 {
1725 	struct bch_fs *c = trans->c;
1726 	struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
1727 	struct bch_extent_crc_unpacked crc;
1728 	const union bch_extent_entry *i;
1729 	unsigned encoded_extent_max_sectors = c->opts.encoded_extent_max >> 9;
1730 
1731 	bkey_for_each_crc(k.k, ptrs, crc, i)
1732 		if (crc_is_encoded(crc) &&
1733 		    crc.uncompressed_size > encoded_extent_max_sectors) {
1734 			struct printbuf buf = PRINTBUF;
1735 
1736 			bch2_bkey_val_to_text(&buf, c, k);
1737 			bch_err(c, "overbig encoded extent, please report this:\n  %s", buf.buf);
1738 			printbuf_exit(&buf);
1739 		}
1740 
1741 	return 0;
1742 }
1743 
1744 static int check_extent(struct btree_trans *trans, struct btree_iter *iter,
1745 			struct bkey_s_c k,
1746 			struct inode_walker *inode,
1747 			struct snapshots_seen *s,
1748 			struct extent_ends *extent_ends,
1749 			struct disk_reservation *res)
1750 {
1751 	struct bch_fs *c = trans->c;
1752 	struct printbuf buf = PRINTBUF;
1753 	int ret = 0;
1754 
1755 	ret = bch2_check_key_has_snapshot(trans, iter, k);
1756 	if (ret) {
1757 		ret = ret < 0 ? ret : 0;
1758 		goto out;
1759 	}
1760 
1761 	if (inode->last_pos.inode != k.k->p.inode && inode->have_inodes) {
1762 		ret = check_i_sectors(trans, inode);
1763 		if (ret)
1764 			goto err;
1765 	}
1766 
1767 	ret = snapshots_seen_update(c, s, iter->btree_id, k.k->p);
1768 	if (ret)
1769 		goto err;
1770 
1771 	struct inode_walker_entry *extent_i = walk_inode(trans, inode, k);
1772 	ret = PTR_ERR_OR_ZERO(extent_i);
1773 	if (ret)
1774 		goto err;
1775 
1776 	ret = check_key_has_inode(trans, iter, inode, extent_i, k);
1777 	if (ret)
1778 		goto err;
1779 
1780 	if (k.k->type != KEY_TYPE_whiteout) {
1781 		ret = check_overlapping_extents(trans, s, extent_ends, k, iter,
1782 						&inode->recalculate_sums);
1783 		if (ret)
1784 			goto err;
1785 
1786 		/*
1787 		 * Check inodes in reverse order, from oldest snapshots to
1788 		 * newest, starting from the inode that matches this extent's
1789 		 * snapshot. If we didn't have one, iterate over all inodes:
1790 		 */
1791 		for (struct inode_walker_entry *i = extent_i ?: &darray_last(inode->inodes);
1792 		     inode->inodes.data && i >= inode->inodes.data;
1793 		     --i) {
1794 			if (i->snapshot > k.k->p.snapshot ||
1795 			    !key_visible_in_snapshot(c, s, i->snapshot, k.k->p.snapshot))
1796 				continue;
1797 
1798 			if (fsck_err_on(k.k->p.offset > round_up(i->inode.bi_size, block_bytes(c)) >> 9 &&
1799 					!bkey_extent_is_reservation(k),
1800 					trans, extent_past_end_of_inode,
1801 					"extent type past end of inode %llu:%u, i_size %llu\n  %s",
1802 					i->inode.bi_inum, i->snapshot, i->inode.bi_size,
1803 					(bch2_bkey_val_to_text(&buf, c, k), buf.buf))) {
1804 				struct btree_iter iter2;
1805 
1806 				bch2_trans_copy_iter(&iter2, iter);
1807 				bch2_btree_iter_set_snapshot(&iter2, i->snapshot);
1808 				ret =   bch2_btree_iter_traverse(&iter2) ?:
1809 					bch2_btree_delete_at(trans, &iter2,
1810 						BTREE_UPDATE_internal_snapshot_node);
1811 				bch2_trans_iter_exit(trans, &iter2);
1812 				if (ret)
1813 					goto err;
1814 
1815 				iter->k.type = KEY_TYPE_whiteout;
1816 				break;
1817 			}
1818 		}
1819 	}
1820 
1821 	ret = bch2_trans_commit(trans, res, NULL, BCH_TRANS_COMMIT_no_enospc);
1822 	if (ret)
1823 		goto err;
1824 
1825 	if (bkey_extent_is_allocation(k.k)) {
1826 		for (struct inode_walker_entry *i = extent_i ?: &darray_last(inode->inodes);
1827 		     inode->inodes.data && i >= inode->inodes.data;
1828 		     --i) {
1829 			if (i->snapshot > k.k->p.snapshot ||
1830 			    !key_visible_in_snapshot(c, s, i->snapshot, k.k->p.snapshot))
1831 				continue;
1832 
1833 			i->count += k.k->size;
1834 		}
1835 	}
1836 
1837 	if (k.k->type != KEY_TYPE_whiteout) {
1838 		ret = extent_ends_at(c, extent_ends, s, k);
1839 		if (ret)
1840 			goto err;
1841 	}
1842 out:
1843 err:
1844 fsck_err:
1845 	printbuf_exit(&buf);
1846 	bch_err_fn(c, ret);
1847 	return ret;
1848 }
1849 
1850 /*
1851  * Walk extents: verify that extents have a corresponding S_ISREG inode, and
1852  * that i_size an i_sectors are consistent
1853  */
1854 int bch2_check_extents(struct bch_fs *c)
1855 {
1856 	struct inode_walker w = inode_walker_init();
1857 	struct snapshots_seen s;
1858 	struct extent_ends extent_ends;
1859 	struct disk_reservation res = { 0 };
1860 
1861 	snapshots_seen_init(&s);
1862 	extent_ends_init(&extent_ends);
1863 
1864 	int ret = bch2_trans_run(c,
1865 		for_each_btree_key(trans, iter, BTREE_ID_extents,
1866 				POS(BCACHEFS_ROOT_INO, 0),
1867 				BTREE_ITER_prefetch|BTREE_ITER_all_snapshots, k, ({
1868 			bch2_disk_reservation_put(c, &res);
1869 			check_extent(trans, &iter, k, &w, &s, &extent_ends, &res) ?:
1870 			check_extent_overbig(trans, &iter, k);
1871 		})) ?:
1872 		check_i_sectors_notnested(trans, &w));
1873 
1874 	bch2_disk_reservation_put(c, &res);
1875 	extent_ends_exit(&extent_ends);
1876 	inode_walker_exit(&w);
1877 	snapshots_seen_exit(&s);
1878 
1879 	bch_err_fn(c, ret);
1880 	return ret;
1881 }
1882 
1883 int bch2_check_indirect_extents(struct bch_fs *c)
1884 {
1885 	struct disk_reservation res = { 0 };
1886 
1887 	int ret = bch2_trans_run(c,
1888 		for_each_btree_key_commit(trans, iter, BTREE_ID_reflink,
1889 				POS_MIN,
1890 				BTREE_ITER_prefetch, k,
1891 				&res, NULL,
1892 				BCH_TRANS_COMMIT_no_enospc, ({
1893 			bch2_disk_reservation_put(c, &res);
1894 			check_extent_overbig(trans, &iter, k);
1895 		})));
1896 
1897 	bch2_disk_reservation_put(c, &res);
1898 	bch_err_fn(c, ret);
1899 	return ret;
1900 }
1901 
1902 static int check_subdir_count_notnested(struct btree_trans *trans, struct inode_walker *w)
1903 {
1904 	struct bch_fs *c = trans->c;
1905 	int ret = 0;
1906 	s64 count2;
1907 
1908 	darray_for_each(w->inodes, i) {
1909 		if (i->inode.bi_nlink == i->count)
1910 			continue;
1911 
1912 		count2 = bch2_count_subdirs(trans, w->last_pos.inode, i->snapshot);
1913 		if (count2 < 0)
1914 			return count2;
1915 
1916 		if (i->count != count2) {
1917 			bch_err_ratelimited(c, "fsck counted subdirectories wrong for inum %llu:%u: got %llu should be %llu",
1918 					    w->last_pos.inode, i->snapshot, i->count, count2);
1919 			i->count = count2;
1920 			if (i->inode.bi_nlink == i->count)
1921 				continue;
1922 		}
1923 
1924 		if (fsck_err_on(i->inode.bi_nlink != i->count,
1925 				trans, inode_dir_wrong_nlink,
1926 				"directory %llu:%u with wrong i_nlink: got %u, should be %llu",
1927 				w->last_pos.inode, i->snapshot, i->inode.bi_nlink, i->count)) {
1928 			i->inode.bi_nlink = i->count;
1929 			ret = bch2_fsck_write_inode(trans, &i->inode);
1930 			if (ret)
1931 				break;
1932 		}
1933 	}
1934 fsck_err:
1935 	bch_err_fn(c, ret);
1936 	return ret;
1937 }
1938 
1939 static int check_subdir_count(struct btree_trans *trans, struct inode_walker *w)
1940 {
1941 	u32 restart_count = trans->restart_count;
1942 	return check_subdir_count_notnested(trans, w) ?:
1943 		trans_was_restarted(trans, restart_count);
1944 }
1945 
1946 noinline_for_stack
1947 static int check_dirent_inode_dirent(struct btree_trans *trans,
1948 				   struct btree_iter *iter,
1949 				   struct bkey_s_c_dirent d,
1950 				   struct bch_inode_unpacked *target)
1951 {
1952 	struct bch_fs *c = trans->c;
1953 	struct printbuf buf = PRINTBUF;
1954 	struct btree_iter bp_iter = { NULL };
1955 	int ret = 0;
1956 
1957 	if (inode_points_to_dirent(target, d))
1958 		return 0;
1959 
1960 	if (!target->bi_dir &&
1961 	    !target->bi_dir_offset) {
1962 		fsck_err_on(S_ISDIR(target->bi_mode),
1963 			    trans, inode_dir_missing_backpointer,
1964 			    "directory with missing backpointer\n%s",
1965 			    (printbuf_reset(&buf),
1966 			     bch2_bkey_val_to_text(&buf, c, d.s_c),
1967 			     prt_printf(&buf, "\n"),
1968 			     bch2_inode_unpacked_to_text(&buf, target),
1969 			     buf.buf));
1970 
1971 		fsck_err_on(target->bi_flags & BCH_INODE_unlinked,
1972 			    trans, inode_unlinked_but_has_dirent,
1973 			    "inode unlinked but has dirent\n%s",
1974 			    (printbuf_reset(&buf),
1975 			     bch2_bkey_val_to_text(&buf, c, d.s_c),
1976 			     prt_printf(&buf, "\n"),
1977 			     bch2_inode_unpacked_to_text(&buf, target),
1978 			     buf.buf));
1979 
1980 		target->bi_flags &= ~BCH_INODE_unlinked;
1981 		target->bi_dir		= d.k->p.inode;
1982 		target->bi_dir_offset	= d.k->p.offset;
1983 		return __bch2_fsck_write_inode(trans, target);
1984 	}
1985 
1986 	if (bch2_inode_should_have_bp(target) &&
1987 	    !fsck_err(trans, inode_wrong_backpointer,
1988 		      "dirent points to inode that does not point back:\n  %s",
1989 		      (bch2_bkey_val_to_text(&buf, c, d.s_c),
1990 		       prt_printf(&buf, "\n  "),
1991 		       bch2_inode_unpacked_to_text(&buf, target),
1992 		       buf.buf)))
1993 		goto err;
1994 
1995 	struct bkey_s_c_dirent bp_dirent = dirent_get_by_pos(trans, &bp_iter,
1996 			      SPOS(target->bi_dir, target->bi_dir_offset, target->bi_snapshot));
1997 	ret = bkey_err(bp_dirent);
1998 	if (ret && !bch2_err_matches(ret, ENOENT))
1999 		goto err;
2000 
2001 	bool backpointer_exists = !ret;
2002 	ret = 0;
2003 
2004 	if (fsck_err_on(!backpointer_exists,
2005 			trans, inode_wrong_backpointer,
2006 			"inode %llu:%u has wrong backpointer:\n"
2007 			"got       %llu:%llu\n"
2008 			"should be %llu:%llu",
2009 			target->bi_inum, target->bi_snapshot,
2010 			target->bi_dir,
2011 			target->bi_dir_offset,
2012 			d.k->p.inode,
2013 			d.k->p.offset)) {
2014 		target->bi_dir		= d.k->p.inode;
2015 		target->bi_dir_offset	= d.k->p.offset;
2016 		ret = __bch2_fsck_write_inode(trans, target);
2017 		goto out;
2018 	}
2019 
2020 	bch2_bkey_val_to_text(&buf, c, d.s_c);
2021 	prt_newline(&buf);
2022 	if (backpointer_exists)
2023 		bch2_bkey_val_to_text(&buf, c, bp_dirent.s_c);
2024 
2025 	if (fsck_err_on(backpointer_exists &&
2026 			(S_ISDIR(target->bi_mode) ||
2027 			 target->bi_subvol),
2028 			trans, inode_dir_multiple_links,
2029 			"%s %llu:%u with multiple links\n%s",
2030 			S_ISDIR(target->bi_mode) ? "directory" : "subvolume",
2031 			target->bi_inum, target->bi_snapshot, buf.buf)) {
2032 		ret = __remove_dirent(trans, d.k->p);
2033 		goto out;
2034 	}
2035 
2036 	/*
2037 	 * hardlinked file with nlink 0:
2038 	 * We're just adjusting nlink here so check_nlinks() will pick
2039 	 * it up, it ignores inodes with nlink 0
2040 	 */
2041 	if (fsck_err_on(backpointer_exists && !target->bi_nlink,
2042 			trans, inode_multiple_links_but_nlink_0,
2043 			"inode %llu:%u type %s has multiple links but i_nlink 0\n%s",
2044 			target->bi_inum, target->bi_snapshot, bch2_d_types[d.v->d_type], buf.buf)) {
2045 		target->bi_nlink++;
2046 		target->bi_flags &= ~BCH_INODE_unlinked;
2047 		ret = __bch2_fsck_write_inode(trans, target);
2048 		if (ret)
2049 			goto err;
2050 	}
2051 out:
2052 err:
2053 fsck_err:
2054 	bch2_trans_iter_exit(trans, &bp_iter);
2055 	printbuf_exit(&buf);
2056 	bch_err_fn(c, ret);
2057 	return ret;
2058 }
2059 
2060 noinline_for_stack
2061 static int check_dirent_target(struct btree_trans *trans,
2062 			       struct btree_iter *iter,
2063 			       struct bkey_s_c_dirent d,
2064 			       struct bch_inode_unpacked *target)
2065 {
2066 	struct bch_fs *c = trans->c;
2067 	struct bkey_i_dirent *n;
2068 	struct printbuf buf = PRINTBUF;
2069 	int ret = 0;
2070 
2071 	ret = check_dirent_inode_dirent(trans, iter, d, target);
2072 	if (ret)
2073 		goto err;
2074 
2075 	if (fsck_err_on(d.v->d_type != inode_d_type(target),
2076 			trans, dirent_d_type_wrong,
2077 			"incorrect d_type: got %s, should be %s:\n%s",
2078 			bch2_d_type_str(d.v->d_type),
2079 			bch2_d_type_str(inode_d_type(target)),
2080 			(printbuf_reset(&buf),
2081 			 bch2_bkey_val_to_text(&buf, c, d.s_c), buf.buf))) {
2082 		n = bch2_trans_kmalloc(trans, bkey_bytes(d.k));
2083 		ret = PTR_ERR_OR_ZERO(n);
2084 		if (ret)
2085 			goto err;
2086 
2087 		bkey_reassemble(&n->k_i, d.s_c);
2088 		n->v.d_type = inode_d_type(target);
2089 		if (n->v.d_type == DT_SUBVOL) {
2090 			n->v.d_parent_subvol = cpu_to_le32(target->bi_parent_subvol);
2091 			n->v.d_child_subvol = cpu_to_le32(target->bi_subvol);
2092 		} else {
2093 			n->v.d_inum = cpu_to_le64(target->bi_inum);
2094 		}
2095 
2096 		ret = bch2_trans_update(trans, iter, &n->k_i, 0);
2097 		if (ret)
2098 			goto err;
2099 
2100 		d = dirent_i_to_s_c(n);
2101 	}
2102 err:
2103 fsck_err:
2104 	printbuf_exit(&buf);
2105 	bch_err_fn(c, ret);
2106 	return ret;
2107 }
2108 
2109 /* find a subvolume that's a descendent of @snapshot: */
2110 static int find_snapshot_subvol(struct btree_trans *trans, u32 snapshot, u32 *subvolid)
2111 {
2112 	struct btree_iter iter;
2113 	struct bkey_s_c k;
2114 	int ret;
2115 
2116 	for_each_btree_key_norestart(trans, iter, BTREE_ID_subvolumes, POS_MIN, 0, k, ret) {
2117 		if (k.k->type != KEY_TYPE_subvolume)
2118 			continue;
2119 
2120 		struct bkey_s_c_subvolume s = bkey_s_c_to_subvolume(k);
2121 		if (bch2_snapshot_is_ancestor(trans->c, le32_to_cpu(s.v->snapshot), snapshot)) {
2122 			bch2_trans_iter_exit(trans, &iter);
2123 			*subvolid = k.k->p.offset;
2124 			goto found;
2125 		}
2126 	}
2127 	if (!ret)
2128 		ret = -ENOENT;
2129 found:
2130 	bch2_trans_iter_exit(trans, &iter);
2131 	return ret;
2132 }
2133 
2134 noinline_for_stack
2135 static int check_dirent_to_subvol(struct btree_trans *trans, struct btree_iter *iter,
2136 				  struct bkey_s_c_dirent d)
2137 {
2138 	struct bch_fs *c = trans->c;
2139 	struct btree_iter subvol_iter = {};
2140 	struct bch_inode_unpacked subvol_root;
2141 	u32 parent_subvol = le32_to_cpu(d.v->d_parent_subvol);
2142 	u32 target_subvol = le32_to_cpu(d.v->d_child_subvol);
2143 	u32 parent_snapshot;
2144 	u32 new_parent_subvol = 0;
2145 	u64 parent_inum;
2146 	struct printbuf buf = PRINTBUF;
2147 	int ret = 0;
2148 
2149 	ret = subvol_lookup(trans, parent_subvol, &parent_snapshot, &parent_inum);
2150 	if (ret && !bch2_err_matches(ret, ENOENT))
2151 		return ret;
2152 
2153 	if (ret ||
2154 	    (!ret && !bch2_snapshot_is_ancestor(c, parent_snapshot, d.k->p.snapshot))) {
2155 		int ret2 = find_snapshot_subvol(trans, d.k->p.snapshot, &new_parent_subvol);
2156 		if (ret2 && !bch2_err_matches(ret, ENOENT))
2157 			return ret2;
2158 	}
2159 
2160 	if (ret &&
2161 	    !new_parent_subvol &&
2162 	    (c->sb.btrees_lost_data & BIT_ULL(BTREE_ID_subvolumes))) {
2163 		/*
2164 		 * Couldn't find a subvol for dirent's snapshot - but we lost
2165 		 * subvols, so we need to reconstruct:
2166 		 */
2167 		ret = reconstruct_subvol(trans, d.k->p.snapshot, parent_subvol, 0);
2168 		if (ret)
2169 			return ret;
2170 
2171 		parent_snapshot = d.k->p.snapshot;
2172 	}
2173 
2174 	if (fsck_err_on(ret,
2175 			trans, dirent_to_missing_parent_subvol,
2176 			"dirent parent_subvol points to missing subvolume\n%s",
2177 			(bch2_bkey_val_to_text(&buf, c, d.s_c), buf.buf)) ||
2178 	    fsck_err_on(!ret && !bch2_snapshot_is_ancestor(c, parent_snapshot, d.k->p.snapshot),
2179 			trans, dirent_not_visible_in_parent_subvol,
2180 			"dirent not visible in parent_subvol (not an ancestor of subvol snap %u)\n%s",
2181 			parent_snapshot,
2182 			(bch2_bkey_val_to_text(&buf, c, d.s_c), buf.buf))) {
2183 		if (!new_parent_subvol) {
2184 			bch_err(c, "could not find a subvol for snapshot %u", d.k->p.snapshot);
2185 			return -BCH_ERR_fsck_repair_unimplemented;
2186 		}
2187 
2188 		struct bkey_i_dirent *new_dirent = bch2_bkey_make_mut_typed(trans, iter, &d.s_c, 0, dirent);
2189 		ret = PTR_ERR_OR_ZERO(new_dirent);
2190 		if (ret)
2191 			goto err;
2192 
2193 		new_dirent->v.d_parent_subvol = cpu_to_le32(new_parent_subvol);
2194 	}
2195 
2196 	struct bkey_s_c_subvolume s =
2197 		bch2_bkey_get_iter_typed(trans, &subvol_iter,
2198 					 BTREE_ID_subvolumes, POS(0, target_subvol),
2199 					 0, subvolume);
2200 	ret = bkey_err(s.s_c);
2201 	if (ret && !bch2_err_matches(ret, ENOENT))
2202 		return ret;
2203 
2204 	if (ret) {
2205 		if (fsck_err(trans, dirent_to_missing_subvol,
2206 			     "dirent points to missing subvolume\n%s",
2207 			     (bch2_bkey_val_to_text(&buf, c, d.s_c), buf.buf)))
2208 			return __remove_dirent(trans, d.k->p);
2209 		ret = 0;
2210 		goto out;
2211 	}
2212 
2213 	if (fsck_err_on(le32_to_cpu(s.v->fs_path_parent) != parent_subvol,
2214 			trans, subvol_fs_path_parent_wrong,
2215 			"subvol with wrong fs_path_parent, should be be %u\n%s",
2216 			parent_subvol,
2217 			(bch2_bkey_val_to_text(&buf, c, s.s_c), buf.buf))) {
2218 		struct bkey_i_subvolume *n =
2219 			bch2_bkey_make_mut_typed(trans, &subvol_iter, &s.s_c, 0, subvolume);
2220 		ret = PTR_ERR_OR_ZERO(n);
2221 		if (ret)
2222 			goto err;
2223 
2224 		n->v.fs_path_parent = cpu_to_le32(parent_subvol);
2225 	}
2226 
2227 	u64 target_inum = le64_to_cpu(s.v->inode);
2228 	u32 target_snapshot = le32_to_cpu(s.v->snapshot);
2229 
2230 	ret = lookup_inode(trans, target_inum, target_snapshot, &subvol_root);
2231 	if (ret && !bch2_err_matches(ret, ENOENT))
2232 		goto err;
2233 
2234 	if (ret) {
2235 		bch_err(c, "subvol %u points to missing inode root %llu", target_subvol, target_inum);
2236 		ret = -BCH_ERR_fsck_repair_unimplemented;
2237 		goto err;
2238 	}
2239 
2240 	if (fsck_err_on(!ret && parent_subvol != subvol_root.bi_parent_subvol,
2241 			trans, inode_bi_parent_wrong,
2242 			"subvol root %llu has wrong bi_parent_subvol: got %u, should be %u",
2243 			target_inum,
2244 			subvol_root.bi_parent_subvol, parent_subvol)) {
2245 		subvol_root.bi_parent_subvol = parent_subvol;
2246 		subvol_root.bi_snapshot = le32_to_cpu(s.v->snapshot);
2247 		ret = __bch2_fsck_write_inode(trans, &subvol_root);
2248 		if (ret)
2249 			goto err;
2250 	}
2251 
2252 	ret = check_dirent_target(trans, iter, d, &subvol_root);
2253 	if (ret)
2254 		goto err;
2255 out:
2256 err:
2257 fsck_err:
2258 	bch2_trans_iter_exit(trans, &subvol_iter);
2259 	printbuf_exit(&buf);
2260 	return ret;
2261 }
2262 
2263 static int check_dirent(struct btree_trans *trans, struct btree_iter *iter,
2264 			struct bkey_s_c k,
2265 			struct bch_hash_info *hash_info,
2266 			struct inode_walker *dir,
2267 			struct inode_walker *target,
2268 			struct snapshots_seen *s)
2269 {
2270 	struct bch_fs *c = trans->c;
2271 	struct inode_walker_entry *i;
2272 	struct printbuf buf = PRINTBUF;
2273 	int ret = 0;
2274 
2275 	ret = bch2_check_key_has_snapshot(trans, iter, k);
2276 	if (ret) {
2277 		ret = ret < 0 ? ret : 0;
2278 		goto out;
2279 	}
2280 
2281 	ret = snapshots_seen_update(c, s, iter->btree_id, k.k->p);
2282 	if (ret)
2283 		goto err;
2284 
2285 	if (k.k->type == KEY_TYPE_whiteout)
2286 		goto out;
2287 
2288 	if (dir->last_pos.inode != k.k->p.inode && dir->have_inodes) {
2289 		ret = check_subdir_count(trans, dir);
2290 		if (ret)
2291 			goto err;
2292 	}
2293 
2294 	i = walk_inode(trans, dir, k);
2295 	ret = PTR_ERR_OR_ZERO(i);
2296 	if (ret < 0)
2297 		goto err;
2298 
2299 	ret = check_key_has_inode(trans, iter, dir, i, k);
2300 	if (ret)
2301 		goto err;
2302 
2303 	if (!i)
2304 		goto out;
2305 
2306 	if (dir->first_this_inode)
2307 		*hash_info = bch2_hash_info_init(c, &i->inode);
2308 	dir->first_this_inode = false;
2309 
2310 	ret = hash_check_key(trans, bch2_dirent_hash_desc, hash_info, iter, k);
2311 	if (ret < 0)
2312 		goto err;
2313 	if (ret) {
2314 		/* dirent has been deleted */
2315 		ret = 0;
2316 		goto out;
2317 	}
2318 
2319 	if (k.k->type != KEY_TYPE_dirent)
2320 		goto out;
2321 
2322 	struct bkey_s_c_dirent d = bkey_s_c_to_dirent(k);
2323 
2324 	if (d.v->d_type == DT_SUBVOL) {
2325 		ret = check_dirent_to_subvol(trans, iter, d);
2326 		if (ret)
2327 			goto err;
2328 	} else {
2329 		ret = get_visible_inodes(trans, target, s, le64_to_cpu(d.v->d_inum));
2330 		if (ret)
2331 			goto err;
2332 
2333 		if (fsck_err_on(!target->inodes.nr,
2334 				trans, dirent_to_missing_inode,
2335 				"dirent points to missing inode:\n%s",
2336 				(printbuf_reset(&buf),
2337 				 bch2_bkey_val_to_text(&buf, c, k),
2338 				 buf.buf))) {
2339 			ret = __remove_dirent(trans, d.k->p);
2340 			if (ret)
2341 				goto err;
2342 		}
2343 
2344 		darray_for_each(target->inodes, i) {
2345 			ret = check_dirent_target(trans, iter, d, &i->inode);
2346 			if (ret)
2347 				goto err;
2348 		}
2349 	}
2350 
2351 	ret = bch2_trans_commit(trans, NULL, NULL, BCH_TRANS_COMMIT_no_enospc);
2352 	if (ret)
2353 		goto err;
2354 
2355 	if (d.v->d_type == DT_DIR)
2356 		for_each_visible_inode(c, s, dir, d.k->p.snapshot, i)
2357 			i->count++;
2358 out:
2359 err:
2360 fsck_err:
2361 	printbuf_exit(&buf);
2362 	bch_err_fn(c, ret);
2363 	return ret;
2364 }
2365 
2366 /*
2367  * Walk dirents: verify that they all have a corresponding S_ISDIR inode,
2368  * validate d_type
2369  */
2370 int bch2_check_dirents(struct bch_fs *c)
2371 {
2372 	struct inode_walker dir = inode_walker_init();
2373 	struct inode_walker target = inode_walker_init();
2374 	struct snapshots_seen s;
2375 	struct bch_hash_info hash_info;
2376 
2377 	snapshots_seen_init(&s);
2378 
2379 	int ret = bch2_trans_run(c,
2380 		for_each_btree_key(trans, iter, BTREE_ID_dirents,
2381 				POS(BCACHEFS_ROOT_INO, 0),
2382 				BTREE_ITER_prefetch|BTREE_ITER_all_snapshots, k,
2383 			check_dirent(trans, &iter, k, &hash_info, &dir, &target, &s)) ?:
2384 		check_subdir_count_notnested(trans, &dir));
2385 
2386 	snapshots_seen_exit(&s);
2387 	inode_walker_exit(&dir);
2388 	inode_walker_exit(&target);
2389 	bch_err_fn(c, ret);
2390 	return ret;
2391 }
2392 
2393 static int check_xattr(struct btree_trans *trans, struct btree_iter *iter,
2394 		       struct bkey_s_c k,
2395 		       struct bch_hash_info *hash_info,
2396 		       struct inode_walker *inode)
2397 {
2398 	struct bch_fs *c = trans->c;
2399 	struct inode_walker_entry *i;
2400 	int ret;
2401 
2402 	ret = bch2_check_key_has_snapshot(trans, iter, k);
2403 	if (ret < 0)
2404 		return ret;
2405 	if (ret)
2406 		return 0;
2407 
2408 	i = walk_inode(trans, inode, k);
2409 	ret = PTR_ERR_OR_ZERO(i);
2410 	if (ret)
2411 		return ret;
2412 
2413 	ret = check_key_has_inode(trans, iter, inode, i, k);
2414 	if (ret)
2415 		return ret;
2416 
2417 	if (!i)
2418 		return 0;
2419 
2420 	if (inode->first_this_inode)
2421 		*hash_info = bch2_hash_info_init(c, &i->inode);
2422 	inode->first_this_inode = false;
2423 
2424 	ret = hash_check_key(trans, bch2_xattr_hash_desc, hash_info, iter, k);
2425 	bch_err_fn(c, ret);
2426 	return ret;
2427 }
2428 
2429 /*
2430  * Walk xattrs: verify that they all have a corresponding inode
2431  */
2432 int bch2_check_xattrs(struct bch_fs *c)
2433 {
2434 	struct inode_walker inode = inode_walker_init();
2435 	struct bch_hash_info hash_info;
2436 	int ret = 0;
2437 
2438 	ret = bch2_trans_run(c,
2439 		for_each_btree_key_commit(trans, iter, BTREE_ID_xattrs,
2440 			POS(BCACHEFS_ROOT_INO, 0),
2441 			BTREE_ITER_prefetch|BTREE_ITER_all_snapshots,
2442 			k,
2443 			NULL, NULL,
2444 			BCH_TRANS_COMMIT_no_enospc,
2445 		check_xattr(trans, &iter, k, &hash_info, &inode)));
2446 
2447 	inode_walker_exit(&inode);
2448 	bch_err_fn(c, ret);
2449 	return ret;
2450 }
2451 
2452 static int check_root_trans(struct btree_trans *trans)
2453 {
2454 	struct bch_fs *c = trans->c;
2455 	struct bch_inode_unpacked root_inode;
2456 	u32 snapshot;
2457 	u64 inum;
2458 	int ret;
2459 
2460 	ret = subvol_lookup(trans, BCACHEFS_ROOT_SUBVOL, &snapshot, &inum);
2461 	if (ret && !bch2_err_matches(ret, ENOENT))
2462 		return ret;
2463 
2464 	if (mustfix_fsck_err_on(ret, trans, root_subvol_missing,
2465 				"root subvol missing")) {
2466 		struct bkey_i_subvolume *root_subvol =
2467 			bch2_trans_kmalloc(trans, sizeof(*root_subvol));
2468 		ret = PTR_ERR_OR_ZERO(root_subvol);
2469 		if (ret)
2470 			goto err;
2471 
2472 		snapshot	= U32_MAX;
2473 		inum		= BCACHEFS_ROOT_INO;
2474 
2475 		bkey_subvolume_init(&root_subvol->k_i);
2476 		root_subvol->k.p.offset = BCACHEFS_ROOT_SUBVOL;
2477 		root_subvol->v.flags	= 0;
2478 		root_subvol->v.snapshot	= cpu_to_le32(snapshot);
2479 		root_subvol->v.inode	= cpu_to_le64(inum);
2480 		ret = bch2_btree_insert_trans(trans, BTREE_ID_subvolumes, &root_subvol->k_i, 0);
2481 		bch_err_msg(c, ret, "writing root subvol");
2482 		if (ret)
2483 			goto err;
2484 	}
2485 
2486 	ret = lookup_inode(trans, BCACHEFS_ROOT_INO, snapshot, &root_inode);
2487 	if (ret && !bch2_err_matches(ret, ENOENT))
2488 		return ret;
2489 
2490 	if (mustfix_fsck_err_on(ret,
2491 				trans, root_dir_missing,
2492 				"root directory missing") ||
2493 	    mustfix_fsck_err_on(!S_ISDIR(root_inode.bi_mode),
2494 				trans, root_inode_not_dir,
2495 				"root inode not a directory")) {
2496 		bch2_inode_init(c, &root_inode, 0, 0, S_IFDIR|0755,
2497 				0, NULL);
2498 		root_inode.bi_inum = inum;
2499 		root_inode.bi_snapshot = snapshot;
2500 
2501 		ret = __bch2_fsck_write_inode(trans, &root_inode);
2502 		bch_err_msg(c, ret, "writing root inode");
2503 	}
2504 err:
2505 fsck_err:
2506 	return ret;
2507 }
2508 
2509 /* Get root directory, create if it doesn't exist: */
2510 int bch2_check_root(struct bch_fs *c)
2511 {
2512 	int ret = bch2_trans_do(c, NULL, NULL, BCH_TRANS_COMMIT_no_enospc,
2513 		check_root_trans(trans));
2514 	bch_err_fn(c, ret);
2515 	return ret;
2516 }
2517 
2518 typedef DARRAY(u32) darray_u32;
2519 
2520 static bool darray_u32_has(darray_u32 *d, u32 v)
2521 {
2522 	darray_for_each(*d, i)
2523 		if (*i == v)
2524 			return true;
2525 	return false;
2526 }
2527 
2528 static int check_subvol_path(struct btree_trans *trans, struct btree_iter *iter, struct bkey_s_c k)
2529 {
2530 	struct bch_fs *c = trans->c;
2531 	struct btree_iter parent_iter = {};
2532 	darray_u32 subvol_path = {};
2533 	struct printbuf buf = PRINTBUF;
2534 	int ret = 0;
2535 
2536 	if (k.k->type != KEY_TYPE_subvolume)
2537 		return 0;
2538 
2539 	while (k.k->p.offset != BCACHEFS_ROOT_SUBVOL) {
2540 		ret = darray_push(&subvol_path, k.k->p.offset);
2541 		if (ret)
2542 			goto err;
2543 
2544 		struct bkey_s_c_subvolume s = bkey_s_c_to_subvolume(k);
2545 
2546 		struct bch_inode_unpacked subvol_root;
2547 		ret = bch2_inode_find_by_inum_trans(trans,
2548 					(subvol_inum) { s.k->p.offset, le64_to_cpu(s.v->inode) },
2549 					&subvol_root);
2550 		if (ret)
2551 			break;
2552 
2553 		u32 parent = le32_to_cpu(s.v->fs_path_parent);
2554 
2555 		if (darray_u32_has(&subvol_path, parent)) {
2556 			if (fsck_err(c, subvol_loop, "subvolume loop"))
2557 				ret = reattach_subvol(trans, s);
2558 			break;
2559 		}
2560 
2561 		bch2_trans_iter_exit(trans, &parent_iter);
2562 		bch2_trans_iter_init(trans, &parent_iter,
2563 				     BTREE_ID_subvolumes, POS(0, parent), 0);
2564 		k = bch2_btree_iter_peek_slot(&parent_iter);
2565 		ret = bkey_err(k);
2566 		if (ret)
2567 			goto err;
2568 
2569 		if (fsck_err_on(k.k->type != KEY_TYPE_subvolume,
2570 				trans, subvol_unreachable,
2571 				"unreachable subvolume %s",
2572 				(bch2_bkey_val_to_text(&buf, c, s.s_c),
2573 				 buf.buf))) {
2574 			ret = reattach_subvol(trans, s);
2575 			break;
2576 		}
2577 	}
2578 fsck_err:
2579 err:
2580 	printbuf_exit(&buf);
2581 	darray_exit(&subvol_path);
2582 	bch2_trans_iter_exit(trans, &parent_iter);
2583 	return ret;
2584 }
2585 
2586 int bch2_check_subvolume_structure(struct bch_fs *c)
2587 {
2588 	int ret = bch2_trans_run(c,
2589 		for_each_btree_key_commit(trans, iter,
2590 				BTREE_ID_subvolumes, POS_MIN, BTREE_ITER_prefetch, k,
2591 				NULL, NULL, BCH_TRANS_COMMIT_no_enospc,
2592 			check_subvol_path(trans, &iter, k)));
2593 	bch_err_fn(c, ret);
2594 	return ret;
2595 }
2596 
2597 struct pathbuf_entry {
2598 	u64	inum;
2599 	u32	snapshot;
2600 };
2601 
2602 typedef DARRAY(struct pathbuf_entry) pathbuf;
2603 
2604 static bool path_is_dup(pathbuf *p, u64 inum, u32 snapshot)
2605 {
2606 	darray_for_each(*p, i)
2607 		if (i->inum	== inum &&
2608 		    i->snapshot	== snapshot)
2609 			return true;
2610 	return false;
2611 }
2612 
2613 static int check_path(struct btree_trans *trans, pathbuf *p, struct bkey_s_c inode_k)
2614 {
2615 	struct bch_fs *c = trans->c;
2616 	struct btree_iter inode_iter = {};
2617 	struct bch_inode_unpacked inode;
2618 	struct printbuf buf = PRINTBUF;
2619 	u32 snapshot = inode_k.k->p.snapshot;
2620 	int ret = 0;
2621 
2622 	p->nr = 0;
2623 
2624 	BUG_ON(bch2_inode_unpack(inode_k, &inode));
2625 
2626 	if (!S_ISDIR(inode.bi_mode))
2627 		return 0;
2628 
2629 	while (!inode.bi_subvol) {
2630 		struct btree_iter dirent_iter;
2631 		struct bkey_s_c_dirent d;
2632 		u32 parent_snapshot = snapshot;
2633 
2634 		d = inode_get_dirent(trans, &dirent_iter, &inode, &parent_snapshot);
2635 		ret = bkey_err(d.s_c);
2636 		if (ret && !bch2_err_matches(ret, ENOENT))
2637 			break;
2638 
2639 		if (!ret && (ret = dirent_points_to_inode(c, d, &inode)))
2640 			bch2_trans_iter_exit(trans, &dirent_iter);
2641 
2642 		if (bch2_err_matches(ret, ENOENT)) {
2643 			printbuf_reset(&buf);
2644 			bch2_bkey_val_to_text(&buf, c, inode_k);
2645 			bch_err(c, "unreachable inode in check_directory_structure: %s\n%s",
2646 				bch2_err_str(ret), buf.buf);
2647 			goto out;
2648 		}
2649 
2650 		bch2_trans_iter_exit(trans, &dirent_iter);
2651 
2652 		ret = darray_push(p, ((struct pathbuf_entry) {
2653 			.inum		= inode.bi_inum,
2654 			.snapshot	= snapshot,
2655 		}));
2656 		if (ret)
2657 			return ret;
2658 
2659 		snapshot = parent_snapshot;
2660 
2661 		bch2_trans_iter_exit(trans, &inode_iter);
2662 		inode_k = bch2_bkey_get_iter(trans, &inode_iter, BTREE_ID_inodes,
2663 					     SPOS(0, inode.bi_dir, snapshot), 0);
2664 		ret = bkey_err(inode_k) ?:
2665 			!bkey_is_inode(inode_k.k) ? -BCH_ERR_ENOENT_inode
2666 			: bch2_inode_unpack(inode_k, &inode);
2667 		if (ret) {
2668 			/* Should have been caught in dirents pass */
2669 			bch_err_msg(c, ret, "error looking up parent directory");
2670 			break;
2671 		}
2672 
2673 		snapshot = inode_k.k->p.snapshot;
2674 
2675 		if (path_is_dup(p, inode.bi_inum, snapshot)) {
2676 			/* XXX print path */
2677 			bch_err(c, "directory structure loop");
2678 
2679 			darray_for_each(*p, i)
2680 				pr_err("%llu:%u", i->inum, i->snapshot);
2681 			pr_err("%llu:%u", inode.bi_inum, snapshot);
2682 
2683 			if (fsck_err(trans, dir_loop, "directory structure loop")) {
2684 				ret = remove_backpointer(trans, &inode);
2685 				bch_err_msg(c, ret, "removing dirent");
2686 				if (ret)
2687 					break;
2688 
2689 				ret = reattach_inode(trans, &inode);
2690 				bch_err_msg(c, ret, "reattaching inode %llu", inode.bi_inum);
2691 			}
2692 			break;
2693 		}
2694 	}
2695 out:
2696 fsck_err:
2697 	bch2_trans_iter_exit(trans, &inode_iter);
2698 	printbuf_exit(&buf);
2699 	bch_err_fn(c, ret);
2700 	return ret;
2701 }
2702 
2703 /*
2704  * Check for loops in the directory structure: all other connectivity issues
2705  * have been fixed by prior passes
2706  */
2707 int bch2_check_directory_structure(struct bch_fs *c)
2708 {
2709 	pathbuf path = { 0, };
2710 	int ret;
2711 
2712 	ret = bch2_trans_run(c,
2713 		for_each_btree_key_commit(trans, iter, BTREE_ID_inodes, POS_MIN,
2714 					  BTREE_ITER_intent|
2715 					  BTREE_ITER_prefetch|
2716 					  BTREE_ITER_all_snapshots, k,
2717 					  NULL, NULL, BCH_TRANS_COMMIT_no_enospc, ({
2718 			if (!bkey_is_inode(k.k))
2719 				continue;
2720 
2721 			if (bch2_inode_flags(k) & BCH_INODE_unlinked)
2722 				continue;
2723 
2724 			check_path(trans, &path, k);
2725 		})));
2726 	darray_exit(&path);
2727 
2728 	bch_err_fn(c, ret);
2729 	return ret;
2730 }
2731 
2732 struct nlink_table {
2733 	size_t		nr;
2734 	size_t		size;
2735 
2736 	struct nlink {
2737 		u64	inum;
2738 		u32	snapshot;
2739 		u32	count;
2740 	}		*d;
2741 };
2742 
2743 static int add_nlink(struct bch_fs *c, struct nlink_table *t,
2744 		     u64 inum, u32 snapshot)
2745 {
2746 	if (t->nr == t->size) {
2747 		size_t new_size = max_t(size_t, 128UL, t->size * 2);
2748 		void *d = kvmalloc_array(new_size, sizeof(t->d[0]), GFP_KERNEL);
2749 
2750 		if (!d) {
2751 			bch_err(c, "fsck: error allocating memory for nlink_table, size %zu",
2752 				new_size);
2753 			return -BCH_ERR_ENOMEM_fsck_add_nlink;
2754 		}
2755 
2756 		if (t->d)
2757 			memcpy(d, t->d, t->size * sizeof(t->d[0]));
2758 		kvfree(t->d);
2759 
2760 		t->d = d;
2761 		t->size = new_size;
2762 	}
2763 
2764 
2765 	t->d[t->nr++] = (struct nlink) {
2766 		.inum		= inum,
2767 		.snapshot	= snapshot,
2768 	};
2769 
2770 	return 0;
2771 }
2772 
2773 static int nlink_cmp(const void *_l, const void *_r)
2774 {
2775 	const struct nlink *l = _l;
2776 	const struct nlink *r = _r;
2777 
2778 	return cmp_int(l->inum, r->inum);
2779 }
2780 
2781 static void inc_link(struct bch_fs *c, struct snapshots_seen *s,
2782 		     struct nlink_table *links,
2783 		     u64 range_start, u64 range_end, u64 inum, u32 snapshot)
2784 {
2785 	struct nlink *link, key = {
2786 		.inum = inum, .snapshot = U32_MAX,
2787 	};
2788 
2789 	if (inum < range_start || inum >= range_end)
2790 		return;
2791 
2792 	link = __inline_bsearch(&key, links->d, links->nr,
2793 				sizeof(links->d[0]), nlink_cmp);
2794 	if (!link)
2795 		return;
2796 
2797 	while (link > links->d && link[0].inum == link[-1].inum)
2798 		--link;
2799 
2800 	for (; link < links->d + links->nr && link->inum == inum; link++)
2801 		if (ref_visible(c, s, snapshot, link->snapshot)) {
2802 			link->count++;
2803 			if (link->snapshot >= snapshot)
2804 				break;
2805 		}
2806 }
2807 
2808 noinline_for_stack
2809 static int check_nlinks_find_hardlinks(struct bch_fs *c,
2810 				       struct nlink_table *t,
2811 				       u64 start, u64 *end)
2812 {
2813 	int ret = bch2_trans_run(c,
2814 		for_each_btree_key(trans, iter, BTREE_ID_inodes,
2815 				   POS(0, start),
2816 				   BTREE_ITER_intent|
2817 				   BTREE_ITER_prefetch|
2818 				   BTREE_ITER_all_snapshots, k, ({
2819 			if (!bkey_is_inode(k.k))
2820 				continue;
2821 
2822 			/* Should never fail, checked by bch2_inode_invalid: */
2823 			struct bch_inode_unpacked u;
2824 			BUG_ON(bch2_inode_unpack(k, &u));
2825 
2826 			/*
2827 			 * Backpointer and directory structure checks are sufficient for
2828 			 * directories, since they can't have hardlinks:
2829 			 */
2830 			if (S_ISDIR(u.bi_mode))
2831 				continue;
2832 
2833 			/*
2834 			 * Previous passes ensured that bi_nlink is nonzero if
2835 			 * it had multiple hardlinks:
2836 			 */
2837 			if (!u.bi_nlink)
2838 				continue;
2839 
2840 			ret = add_nlink(c, t, k.k->p.offset, k.k->p.snapshot);
2841 			if (ret) {
2842 				*end = k.k->p.offset;
2843 				ret = 0;
2844 				break;
2845 			}
2846 			0;
2847 		})));
2848 
2849 	bch_err_fn(c, ret);
2850 	return ret;
2851 }
2852 
2853 noinline_for_stack
2854 static int check_nlinks_walk_dirents(struct bch_fs *c, struct nlink_table *links,
2855 				     u64 range_start, u64 range_end)
2856 {
2857 	struct snapshots_seen s;
2858 
2859 	snapshots_seen_init(&s);
2860 
2861 	int ret = bch2_trans_run(c,
2862 		for_each_btree_key(trans, iter, BTREE_ID_dirents, POS_MIN,
2863 				   BTREE_ITER_intent|
2864 				   BTREE_ITER_prefetch|
2865 				   BTREE_ITER_all_snapshots, k, ({
2866 			ret = snapshots_seen_update(c, &s, iter.btree_id, k.k->p);
2867 			if (ret)
2868 				break;
2869 
2870 			if (k.k->type == KEY_TYPE_dirent) {
2871 				struct bkey_s_c_dirent d = bkey_s_c_to_dirent(k);
2872 
2873 				if (d.v->d_type != DT_DIR &&
2874 				    d.v->d_type != DT_SUBVOL)
2875 					inc_link(c, &s, links, range_start, range_end,
2876 						 le64_to_cpu(d.v->d_inum), d.k->p.snapshot);
2877 			}
2878 			0;
2879 		})));
2880 
2881 	snapshots_seen_exit(&s);
2882 
2883 	bch_err_fn(c, ret);
2884 	return ret;
2885 }
2886 
2887 static int check_nlinks_update_inode(struct btree_trans *trans, struct btree_iter *iter,
2888 				     struct bkey_s_c k,
2889 				     struct nlink_table *links,
2890 				     size_t *idx, u64 range_end)
2891 {
2892 	struct bch_inode_unpacked u;
2893 	struct nlink *link = &links->d[*idx];
2894 	int ret = 0;
2895 
2896 	if (k.k->p.offset >= range_end)
2897 		return 1;
2898 
2899 	if (!bkey_is_inode(k.k))
2900 		return 0;
2901 
2902 	BUG_ON(bch2_inode_unpack(k, &u));
2903 
2904 	if (S_ISDIR(u.bi_mode))
2905 		return 0;
2906 
2907 	if (!u.bi_nlink)
2908 		return 0;
2909 
2910 	while ((cmp_int(link->inum, k.k->p.offset) ?:
2911 		cmp_int(link->snapshot, k.k->p.snapshot)) < 0) {
2912 		BUG_ON(*idx == links->nr);
2913 		link = &links->d[++*idx];
2914 	}
2915 
2916 	if (fsck_err_on(bch2_inode_nlink_get(&u) != link->count,
2917 			trans, inode_wrong_nlink,
2918 			"inode %llu type %s has wrong i_nlink (%u, should be %u)",
2919 			u.bi_inum, bch2_d_types[mode_to_type(u.bi_mode)],
2920 			bch2_inode_nlink_get(&u), link->count)) {
2921 		bch2_inode_nlink_set(&u, link->count);
2922 		ret = __bch2_fsck_write_inode(trans, &u);
2923 	}
2924 fsck_err:
2925 	return ret;
2926 }
2927 
2928 noinline_for_stack
2929 static int check_nlinks_update_hardlinks(struct bch_fs *c,
2930 			       struct nlink_table *links,
2931 			       u64 range_start, u64 range_end)
2932 {
2933 	size_t idx = 0;
2934 
2935 	int ret = bch2_trans_run(c,
2936 		for_each_btree_key_commit(trans, iter, BTREE_ID_inodes,
2937 				POS(0, range_start),
2938 				BTREE_ITER_intent|BTREE_ITER_prefetch|BTREE_ITER_all_snapshots, k,
2939 				NULL, NULL, BCH_TRANS_COMMIT_no_enospc,
2940 			check_nlinks_update_inode(trans, &iter, k, links, &idx, range_end)));
2941 	if (ret < 0) {
2942 		bch_err(c, "error in fsck walking inodes: %s", bch2_err_str(ret));
2943 		return ret;
2944 	}
2945 
2946 	return 0;
2947 }
2948 
2949 int bch2_check_nlinks(struct bch_fs *c)
2950 {
2951 	struct nlink_table links = { 0 };
2952 	u64 this_iter_range_start, next_iter_range_start = 0;
2953 	int ret = 0;
2954 
2955 	do {
2956 		this_iter_range_start = next_iter_range_start;
2957 		next_iter_range_start = U64_MAX;
2958 
2959 		ret = check_nlinks_find_hardlinks(c, &links,
2960 						  this_iter_range_start,
2961 						  &next_iter_range_start);
2962 
2963 		ret = check_nlinks_walk_dirents(c, &links,
2964 					  this_iter_range_start,
2965 					  next_iter_range_start);
2966 		if (ret)
2967 			break;
2968 
2969 		ret = check_nlinks_update_hardlinks(c, &links,
2970 					 this_iter_range_start,
2971 					 next_iter_range_start);
2972 		if (ret)
2973 			break;
2974 
2975 		links.nr = 0;
2976 	} while (next_iter_range_start != U64_MAX);
2977 
2978 	kvfree(links.d);
2979 	bch_err_fn(c, ret);
2980 	return ret;
2981 }
2982 
2983 static int fix_reflink_p_key(struct btree_trans *trans, struct btree_iter *iter,
2984 			     struct bkey_s_c k)
2985 {
2986 	struct bkey_s_c_reflink_p p;
2987 	struct bkey_i_reflink_p *u;
2988 
2989 	if (k.k->type != KEY_TYPE_reflink_p)
2990 		return 0;
2991 
2992 	p = bkey_s_c_to_reflink_p(k);
2993 
2994 	if (!p.v->front_pad && !p.v->back_pad)
2995 		return 0;
2996 
2997 	u = bch2_trans_kmalloc(trans, sizeof(*u));
2998 	int ret = PTR_ERR_OR_ZERO(u);
2999 	if (ret)
3000 		return ret;
3001 
3002 	bkey_reassemble(&u->k_i, k);
3003 	u->v.front_pad	= 0;
3004 	u->v.back_pad	= 0;
3005 
3006 	return bch2_trans_update(trans, iter, &u->k_i, BTREE_TRIGGER_norun);
3007 }
3008 
3009 int bch2_fix_reflink_p(struct bch_fs *c)
3010 {
3011 	if (c->sb.version >= bcachefs_metadata_version_reflink_p_fix)
3012 		return 0;
3013 
3014 	int ret = bch2_trans_run(c,
3015 		for_each_btree_key_commit(trans, iter,
3016 				BTREE_ID_extents, POS_MIN,
3017 				BTREE_ITER_intent|BTREE_ITER_prefetch|
3018 				BTREE_ITER_all_snapshots, k,
3019 				NULL, NULL, BCH_TRANS_COMMIT_no_enospc,
3020 			fix_reflink_p_key(trans, &iter, k)));
3021 	bch_err_fn(c, ret);
3022 	return ret;
3023 }
3024