xref: /linux/fs/bcachefs/subvolume.c (revision 06a130e42a5bfc84795464bff023bff4c16f58c5)
1 // SPDX-License-Identifier: GPL-2.0
2 
3 #include "bcachefs.h"
4 #include "btree_key_cache.h"
5 #include "btree_update.h"
6 #include "errcode.h"
7 #include "error.h"
8 #include "fs.h"
9 #include "snapshot.h"
10 #include "subvolume.h"
11 
12 #include <linux/random.h>
13 
14 static int bch2_subvolume_delete(struct btree_trans *, u32);
15 
16 static struct bpos subvolume_children_pos(struct bkey_s_c k)
17 {
18 	if (k.k->type != KEY_TYPE_subvolume)
19 		return POS_MIN;
20 
21 	struct bkey_s_c_subvolume s = bkey_s_c_to_subvolume(k);
22 	if (!s.v->fs_path_parent)
23 		return POS_MIN;
24 	return POS(le32_to_cpu(s.v->fs_path_parent), s.k->p.offset);
25 }
26 
27 static int check_subvol(struct btree_trans *trans,
28 			struct btree_iter *iter,
29 			struct bkey_s_c k)
30 {
31 	struct bch_fs *c = trans->c;
32 	struct bkey_s_c_subvolume subvol;
33 	struct btree_iter subvol_children_iter = {};
34 	struct bch_snapshot snapshot;
35 	struct printbuf buf = PRINTBUF;
36 	unsigned snapid;
37 	int ret = 0;
38 
39 	if (k.k->type != KEY_TYPE_subvolume)
40 		return 0;
41 
42 	subvol = bkey_s_c_to_subvolume(k);
43 	snapid = le32_to_cpu(subvol.v->snapshot);
44 	ret = bch2_snapshot_lookup(trans, snapid, &snapshot);
45 
46 	if (bch2_err_matches(ret, ENOENT))
47 		bch_err(c, "subvolume %llu points to nonexistent snapshot %u",
48 			k.k->p.offset, snapid);
49 	if (ret)
50 		return ret;
51 
52 	if (BCH_SUBVOLUME_UNLINKED(subvol.v)) {
53 		ret = bch2_subvolume_delete(trans, iter->pos.offset);
54 		bch_err_msg(c, ret, "deleting subvolume %llu", iter->pos.offset);
55 		return ret ?: -BCH_ERR_transaction_restart_nested;
56 	}
57 
58 	if (fsck_err_on(subvol.k->p.offset == BCACHEFS_ROOT_SUBVOL &&
59 			subvol.v->fs_path_parent,
60 			trans, subvol_root_fs_path_parent_nonzero,
61 			"root subvolume has nonzero fs_path_parent\n%s",
62 			(bch2_bkey_val_to_text(&buf, c, k), buf.buf))) {
63 		struct bkey_i_subvolume *n =
64 			bch2_bkey_make_mut_typed(trans, iter, &subvol.s_c, 0, subvolume);
65 		ret = PTR_ERR_OR_ZERO(n);
66 		if (ret)
67 			goto err;
68 
69 		n->v.fs_path_parent = 0;
70 	}
71 
72 	if (subvol.v->fs_path_parent) {
73 		struct bpos pos = subvolume_children_pos(k);
74 
75 		struct bkey_s_c subvol_children_k =
76 			bch2_bkey_get_iter(trans, &subvol_children_iter,
77 					   BTREE_ID_subvolume_children, pos, 0);
78 		ret = bkey_err(subvol_children_k);
79 		if (ret)
80 			goto err;
81 
82 		if (fsck_err_on(subvol_children_k.k->type != KEY_TYPE_set,
83 				trans, subvol_children_not_set,
84 				"subvolume not set in subvolume_children btree at %llu:%llu\n%s",
85 				pos.inode, pos.offset,
86 				(printbuf_reset(&buf),
87 				 bch2_bkey_val_to_text(&buf, c, k), buf.buf))) {
88 			ret = bch2_btree_bit_mod(trans, BTREE_ID_subvolume_children, pos, true);
89 			if (ret)
90 				goto err;
91 		}
92 	}
93 
94 	struct bch_inode_unpacked inode;
95 	ret = bch2_inode_find_by_inum_nowarn_trans(trans,
96 				    (subvol_inum) { k.k->p.offset, le64_to_cpu(subvol.v->inode) },
97 				    &inode);
98 	if (!ret) {
99 		if (fsck_err_on(inode.bi_subvol != subvol.k->p.offset,
100 				trans, subvol_root_wrong_bi_subvol,
101 				"subvol root %llu:%u has wrong bi_subvol field: got %u, should be %llu",
102 				inode.bi_inum, inode.bi_snapshot,
103 				inode.bi_subvol, subvol.k->p.offset)) {
104 			inode.bi_subvol = subvol.k->p.offset;
105 			inode.bi_snapshot = le32_to_cpu(subvol.v->snapshot);
106 			ret = __bch2_fsck_write_inode(trans, &inode);
107 			if (ret)
108 				goto err;
109 		}
110 	} else if (bch2_err_matches(ret, ENOENT)) {
111 		if (fsck_err(trans, subvol_to_missing_root,
112 			     "subvolume %llu points to missing subvolume root %llu:%u",
113 			     k.k->p.offset, le64_to_cpu(subvol.v->inode),
114 			     le32_to_cpu(subvol.v->snapshot))) {
115 			ret = bch2_subvolume_delete(trans, iter->pos.offset);
116 			bch_err_msg(c, ret, "deleting subvolume %llu", iter->pos.offset);
117 			ret = ret ?: -BCH_ERR_transaction_restart_nested;
118 			goto err;
119 		}
120 	} else {
121 		goto err;
122 	}
123 
124 	if (!BCH_SUBVOLUME_SNAP(subvol.v)) {
125 		u32 snapshot_root = bch2_snapshot_root(c, le32_to_cpu(subvol.v->snapshot));
126 		u32 snapshot_tree;
127 		struct bch_snapshot_tree st;
128 
129 		rcu_read_lock();
130 		snapshot_tree = snapshot_t(c, snapshot_root)->tree;
131 		rcu_read_unlock();
132 
133 		ret = bch2_snapshot_tree_lookup(trans, snapshot_tree, &st);
134 
135 		bch2_fs_inconsistent_on(bch2_err_matches(ret, ENOENT), c,
136 				"%s: snapshot tree %u not found", __func__, snapshot_tree);
137 
138 		if (ret)
139 			goto err;
140 
141 		if (fsck_err_on(le32_to_cpu(st.master_subvol) != subvol.k->p.offset,
142 				trans, subvol_not_master_and_not_snapshot,
143 				"subvolume %llu is not set as snapshot but is not master subvolume",
144 				k.k->p.offset)) {
145 			struct bkey_i_subvolume *s =
146 				bch2_bkey_make_mut_typed(trans, iter, &subvol.s_c, 0, subvolume);
147 			ret = PTR_ERR_OR_ZERO(s);
148 			if (ret)
149 				goto err;
150 
151 			SET_BCH_SUBVOLUME_SNAP(&s->v, true);
152 		}
153 	}
154 err:
155 fsck_err:
156 	bch2_trans_iter_exit(trans, &subvol_children_iter);
157 	printbuf_exit(&buf);
158 	return ret;
159 }
160 
161 int bch2_check_subvols(struct bch_fs *c)
162 {
163 	int ret = bch2_trans_run(c,
164 		for_each_btree_key_commit(trans, iter,
165 				BTREE_ID_subvolumes, POS_MIN, BTREE_ITER_prefetch, k,
166 				NULL, NULL, BCH_TRANS_COMMIT_no_enospc,
167 			check_subvol(trans, &iter, k)));
168 	bch_err_fn(c, ret);
169 	return ret;
170 }
171 
172 static int check_subvol_child(struct btree_trans *trans,
173 			      struct btree_iter *child_iter,
174 			      struct bkey_s_c child_k)
175 {
176 	struct bch_subvolume s;
177 	int ret = bch2_bkey_get_val_typed(trans, BTREE_ID_subvolumes, POS(0, child_k.k->p.offset),
178 					  0, subvolume, &s);
179 	if (ret && !bch2_err_matches(ret, ENOENT))
180 		return ret;
181 
182 	if (fsck_err_on(ret ||
183 			le32_to_cpu(s.fs_path_parent) != child_k.k->p.inode,
184 			trans, subvol_children_bad,
185 			"incorrect entry in subvolume_children btree %llu:%llu",
186 			child_k.k->p.inode, child_k.k->p.offset)) {
187 		ret = bch2_btree_delete_at(trans, child_iter, 0);
188 		if (ret)
189 			goto err;
190 	}
191 err:
192 fsck_err:
193 	return ret;
194 }
195 
196 int bch2_check_subvol_children(struct bch_fs *c)
197 {
198 	int ret = bch2_trans_run(c,
199 		for_each_btree_key_commit(trans, iter,
200 				BTREE_ID_subvolume_children, POS_MIN, BTREE_ITER_prefetch, k,
201 				NULL, NULL, BCH_TRANS_COMMIT_no_enospc,
202 			check_subvol_child(trans, &iter, k)));
203 	bch_err_fn(c, ret);
204 	return 0;
205 }
206 
207 /* Subvolumes: */
208 
209 int bch2_subvolume_validate(struct bch_fs *c, struct bkey_s_c k,
210 			   enum bch_validate_flags flags)
211 {
212 	struct bkey_s_c_subvolume subvol = bkey_s_c_to_subvolume(k);
213 	int ret = 0;
214 
215 	bkey_fsck_err_on(bkey_lt(k.k->p, SUBVOL_POS_MIN) ||
216 			 bkey_gt(k.k->p, SUBVOL_POS_MAX),
217 			 c, subvol_pos_bad,
218 			 "invalid pos");
219 
220 	bkey_fsck_err_on(!subvol.v->snapshot,
221 			 c, subvol_snapshot_bad,
222 			 "invalid snapshot");
223 
224 	bkey_fsck_err_on(!subvol.v->inode,
225 			 c, subvol_inode_bad,
226 			 "invalid inode");
227 fsck_err:
228 	return ret;
229 }
230 
231 void bch2_subvolume_to_text(struct printbuf *out, struct bch_fs *c,
232 			    struct bkey_s_c k)
233 {
234 	struct bkey_s_c_subvolume s = bkey_s_c_to_subvolume(k);
235 
236 	prt_printf(out, "root %llu snapshot id %u",
237 		   le64_to_cpu(s.v->inode),
238 		   le32_to_cpu(s.v->snapshot));
239 
240 	if (bkey_val_bytes(s.k) > offsetof(struct bch_subvolume, creation_parent)) {
241 		prt_printf(out, " creation_parent %u", le32_to_cpu(s.v->creation_parent));
242 		prt_printf(out, " fs_parent %u", le32_to_cpu(s.v->fs_path_parent));
243 	}
244 }
245 
246 static int subvolume_children_mod(struct btree_trans *trans, struct bpos pos, bool set)
247 {
248 	return !bpos_eq(pos, POS_MIN)
249 		? bch2_btree_bit_mod(trans, BTREE_ID_subvolume_children, pos, set)
250 		: 0;
251 }
252 
253 int bch2_subvolume_trigger(struct btree_trans *trans,
254 			   enum btree_id btree_id, unsigned level,
255 			   struct bkey_s_c old, struct bkey_s new,
256 			   enum btree_iter_update_trigger_flags flags)
257 {
258 	if (flags & BTREE_TRIGGER_transactional) {
259 		struct bpos children_pos_old = subvolume_children_pos(old);
260 		struct bpos children_pos_new = subvolume_children_pos(new.s_c);
261 
262 		if (!bpos_eq(children_pos_old, children_pos_new)) {
263 			int ret = subvolume_children_mod(trans, children_pos_old, false) ?:
264 				  subvolume_children_mod(trans, children_pos_new, true);
265 			if (ret)
266 				return ret;
267 		}
268 	}
269 
270 	return 0;
271 }
272 
273 int bch2_subvol_has_children(struct btree_trans *trans, u32 subvol)
274 {
275 	struct btree_iter iter;
276 
277 	bch2_trans_iter_init(trans, &iter, BTREE_ID_subvolume_children, POS(subvol, 0), 0);
278 	struct bkey_s_c k = bch2_btree_iter_peek(&iter);
279 	bch2_trans_iter_exit(trans, &iter);
280 
281 	return bkey_err(k) ?: k.k && k.k->p.inode == subvol
282 		? -BCH_ERR_ENOTEMPTY_subvol_not_empty
283 		: 0;
284 }
285 
286 static __always_inline int
287 bch2_subvolume_get_inlined(struct btree_trans *trans, unsigned subvol,
288 			   bool inconsistent_if_not_found,
289 			   int iter_flags,
290 			   struct bch_subvolume *s)
291 {
292 	int ret = bch2_bkey_get_val_typed(trans, BTREE_ID_subvolumes, POS(0, subvol),
293 					  iter_flags, subvolume, s);
294 	bch2_fs_inconsistent_on(bch2_err_matches(ret, ENOENT) &&
295 				inconsistent_if_not_found,
296 				trans->c, "missing subvolume %u", subvol);
297 	return ret;
298 }
299 
300 int bch2_subvolume_get(struct btree_trans *trans, unsigned subvol,
301 		       bool inconsistent_if_not_found,
302 		       int iter_flags,
303 		       struct bch_subvolume *s)
304 {
305 	return bch2_subvolume_get_inlined(trans, subvol, inconsistent_if_not_found, iter_flags, s);
306 }
307 
308 int bch2_subvol_is_ro_trans(struct btree_trans *trans, u32 subvol)
309 {
310 	struct bch_subvolume s;
311 	int ret = bch2_subvolume_get_inlined(trans, subvol, true, 0, &s);
312 	if (ret)
313 		return ret;
314 
315 	if (BCH_SUBVOLUME_RO(&s))
316 		return -EROFS;
317 	return 0;
318 }
319 
320 int bch2_subvol_is_ro(struct bch_fs *c, u32 subvol)
321 {
322 	return bch2_trans_do(c, NULL, NULL, 0,
323 		bch2_subvol_is_ro_trans(trans, subvol));
324 }
325 
326 int bch2_snapshot_get_subvol(struct btree_trans *trans, u32 snapshot,
327 			     struct bch_subvolume *subvol)
328 {
329 	struct bch_snapshot snap;
330 
331 	return  bch2_snapshot_lookup(trans, snapshot, &snap) ?:
332 		bch2_subvolume_get(trans, le32_to_cpu(snap.subvol), true, 0, subvol);
333 }
334 
335 int __bch2_subvolume_get_snapshot(struct btree_trans *trans, u32 subvolid,
336 				  u32 *snapid, bool warn)
337 {
338 	struct btree_iter iter;
339 	struct bkey_s_c_subvolume subvol;
340 	int ret;
341 
342 	subvol = bch2_bkey_get_iter_typed(trans, &iter,
343 					  BTREE_ID_subvolumes, POS(0, subvolid),
344 					  BTREE_ITER_cached|BTREE_ITER_with_updates,
345 					  subvolume);
346 	ret = bkey_err(subvol);
347 
348 	bch2_fs_inconsistent_on(warn && bch2_err_matches(ret, ENOENT), trans->c,
349 				"missing subvolume %u", subvolid);
350 
351 	if (likely(!ret))
352 		*snapid = le32_to_cpu(subvol.v->snapshot);
353 	bch2_trans_iter_exit(trans, &iter);
354 	return ret;
355 }
356 
357 int bch2_subvolume_get_snapshot(struct btree_trans *trans, u32 subvolid,
358 				u32 *snapid)
359 {
360 	return __bch2_subvolume_get_snapshot(trans, subvolid, snapid, true);
361 }
362 
363 static int bch2_subvolume_reparent(struct btree_trans *trans,
364 				   struct btree_iter *iter,
365 				   struct bkey_s_c k,
366 				   u32 old_parent, u32 new_parent)
367 {
368 	struct bkey_i_subvolume *s;
369 	int ret;
370 
371 	if (k.k->type != KEY_TYPE_subvolume)
372 		return 0;
373 
374 	if (bkey_val_bytes(k.k) > offsetof(struct bch_subvolume, creation_parent) &&
375 	    le32_to_cpu(bkey_s_c_to_subvolume(k).v->creation_parent) != old_parent)
376 		return 0;
377 
378 	s = bch2_bkey_make_mut_typed(trans, iter, &k, 0, subvolume);
379 	ret = PTR_ERR_OR_ZERO(s);
380 	if (ret)
381 		return ret;
382 
383 	s->v.creation_parent = cpu_to_le32(new_parent);
384 	return 0;
385 }
386 
387 /*
388  * Separate from the snapshot tree in the snapshots btree, we record the tree
389  * structure of how snapshot subvolumes were created - the parent subvolume of
390  * each snapshot subvolume.
391  *
392  * When a subvolume is deleted, we scan for child subvolumes and reparant them,
393  * to avoid dangling references:
394  */
395 static int bch2_subvolumes_reparent(struct btree_trans *trans, u32 subvolid_to_delete)
396 {
397 	struct bch_subvolume s;
398 
399 	return lockrestart_do(trans,
400 			bch2_subvolume_get(trans, subvolid_to_delete, true,
401 				   BTREE_ITER_cached, &s)) ?:
402 		for_each_btree_key_commit(trans, iter,
403 				BTREE_ID_subvolumes, POS_MIN, BTREE_ITER_prefetch, k,
404 				NULL, NULL, BCH_TRANS_COMMIT_no_enospc,
405 			bch2_subvolume_reparent(trans, &iter, k,
406 					subvolid_to_delete, le32_to_cpu(s.creation_parent)));
407 }
408 
409 /*
410  * Delete subvolume, mark snapshot ID as deleted, queue up snapshot
411  * deletion/cleanup:
412  */
413 static int __bch2_subvolume_delete(struct btree_trans *trans, u32 subvolid)
414 {
415 	struct btree_iter iter;
416 	struct bkey_s_c_subvolume subvol;
417 	u32 snapid;
418 	int ret = 0;
419 
420 	subvol = bch2_bkey_get_iter_typed(trans, &iter,
421 				BTREE_ID_subvolumes, POS(0, subvolid),
422 				BTREE_ITER_cached|BTREE_ITER_intent,
423 				subvolume);
424 	ret = bkey_err(subvol);
425 	bch2_fs_inconsistent_on(bch2_err_matches(ret, ENOENT), trans->c,
426 				"missing subvolume %u", subvolid);
427 	if (ret)
428 		return ret;
429 
430 	snapid = le32_to_cpu(subvol.v->snapshot);
431 
432 	ret =   bch2_btree_delete_at(trans, &iter, 0) ?:
433 		bch2_snapshot_node_set_deleted(trans, snapid);
434 	bch2_trans_iter_exit(trans, &iter);
435 	return ret;
436 }
437 
438 static int bch2_subvolume_delete(struct btree_trans *trans, u32 subvolid)
439 {
440 	return bch2_subvolumes_reparent(trans, subvolid) ?:
441 		commit_do(trans, NULL, NULL, BCH_TRANS_COMMIT_no_enospc,
442 			  __bch2_subvolume_delete(trans, subvolid));
443 }
444 
445 static void bch2_subvolume_wait_for_pagecache_and_delete(struct work_struct *work)
446 {
447 	struct bch_fs *c = container_of(work, struct bch_fs,
448 				snapshot_wait_for_pagecache_and_delete_work);
449 	snapshot_id_list s;
450 	u32 *id;
451 	int ret = 0;
452 
453 	while (!ret) {
454 		mutex_lock(&c->snapshots_unlinked_lock);
455 		s = c->snapshots_unlinked;
456 		darray_init(&c->snapshots_unlinked);
457 		mutex_unlock(&c->snapshots_unlinked_lock);
458 
459 		if (!s.nr)
460 			break;
461 
462 		bch2_evict_subvolume_inodes(c, &s);
463 
464 		for (id = s.data; id < s.data + s.nr; id++) {
465 			ret = bch2_trans_run(c, bch2_subvolume_delete(trans, *id));
466 			bch_err_msg(c, ret, "deleting subvolume %u", *id);
467 			if (ret)
468 				break;
469 		}
470 
471 		darray_exit(&s);
472 	}
473 
474 	bch2_write_ref_put(c, BCH_WRITE_REF_snapshot_delete_pagecache);
475 }
476 
477 struct subvolume_unlink_hook {
478 	struct btree_trans_commit_hook	h;
479 	u32				subvol;
480 };
481 
482 static int bch2_subvolume_wait_for_pagecache_and_delete_hook(struct btree_trans *trans,
483 						      struct btree_trans_commit_hook *_h)
484 {
485 	struct subvolume_unlink_hook *h = container_of(_h, struct subvolume_unlink_hook, h);
486 	struct bch_fs *c = trans->c;
487 	int ret = 0;
488 
489 	mutex_lock(&c->snapshots_unlinked_lock);
490 	if (!snapshot_list_has_id(&c->snapshots_unlinked, h->subvol))
491 		ret = snapshot_list_add(c, &c->snapshots_unlinked, h->subvol);
492 	mutex_unlock(&c->snapshots_unlinked_lock);
493 
494 	if (ret)
495 		return ret;
496 
497 	if (!bch2_write_ref_tryget(c, BCH_WRITE_REF_snapshot_delete_pagecache))
498 		return -EROFS;
499 
500 	if (!queue_work(c->write_ref_wq, &c->snapshot_wait_for_pagecache_and_delete_work))
501 		bch2_write_ref_put(c, BCH_WRITE_REF_snapshot_delete_pagecache);
502 	return 0;
503 }
504 
505 int bch2_subvolume_unlink(struct btree_trans *trans, u32 subvolid)
506 {
507 	struct btree_iter iter;
508 	struct bkey_i_subvolume *n;
509 	struct subvolume_unlink_hook *h;
510 	int ret = 0;
511 
512 	h = bch2_trans_kmalloc(trans, sizeof(*h));
513 	ret = PTR_ERR_OR_ZERO(h);
514 	if (ret)
515 		return ret;
516 
517 	h->h.fn		= bch2_subvolume_wait_for_pagecache_and_delete_hook;
518 	h->subvol	= subvolid;
519 	bch2_trans_commit_hook(trans, &h->h);
520 
521 	n = bch2_bkey_get_mut_typed(trans, &iter,
522 			BTREE_ID_subvolumes, POS(0, subvolid),
523 			BTREE_ITER_cached, subvolume);
524 	ret = PTR_ERR_OR_ZERO(n);
525 	if (unlikely(ret)) {
526 		bch2_fs_inconsistent_on(bch2_err_matches(ret, ENOENT), trans->c,
527 					"missing subvolume %u", subvolid);
528 		return ret;
529 	}
530 
531 	SET_BCH_SUBVOLUME_UNLINKED(&n->v, true);
532 	bch2_trans_iter_exit(trans, &iter);
533 	return ret;
534 }
535 
536 int bch2_subvolume_create(struct btree_trans *trans, u64 inode,
537 			  u32 parent_subvolid,
538 			  u32 src_subvolid,
539 			  u32 *new_subvolid,
540 			  u32 *new_snapshotid,
541 			  bool ro)
542 {
543 	struct bch_fs *c = trans->c;
544 	struct btree_iter dst_iter, src_iter = (struct btree_iter) { NULL };
545 	struct bkey_i_subvolume *new_subvol = NULL;
546 	struct bkey_i_subvolume *src_subvol = NULL;
547 	u32 parent = 0, new_nodes[2], snapshot_subvols[2];
548 	int ret = 0;
549 
550 	ret = bch2_bkey_get_empty_slot(trans, &dst_iter,
551 				BTREE_ID_subvolumes, POS(0, U32_MAX));
552 	if (ret == -BCH_ERR_ENOSPC_btree_slot)
553 		ret = -BCH_ERR_ENOSPC_subvolume_create;
554 	if (ret)
555 		return ret;
556 
557 	snapshot_subvols[0] = dst_iter.pos.offset;
558 	snapshot_subvols[1] = src_subvolid;
559 
560 	if (src_subvolid) {
561 		/* Creating a snapshot: */
562 
563 		src_subvol = bch2_bkey_get_mut_typed(trans, &src_iter,
564 				BTREE_ID_subvolumes, POS(0, src_subvolid),
565 				BTREE_ITER_cached, subvolume);
566 		ret = PTR_ERR_OR_ZERO(src_subvol);
567 		if (unlikely(ret)) {
568 			bch2_fs_inconsistent_on(bch2_err_matches(ret, ENOENT), c,
569 						"subvolume %u not found", src_subvolid);
570 			goto err;
571 		}
572 
573 		parent = le32_to_cpu(src_subvol->v.snapshot);
574 	}
575 
576 	ret = bch2_snapshot_node_create(trans, parent, new_nodes,
577 					snapshot_subvols,
578 					src_subvolid ? 2 : 1);
579 	if (ret)
580 		goto err;
581 
582 	if (src_subvolid) {
583 		src_subvol->v.snapshot = cpu_to_le32(new_nodes[1]);
584 		ret = bch2_trans_update(trans, &src_iter, &src_subvol->k_i, 0);
585 		if (ret)
586 			goto err;
587 	}
588 
589 	new_subvol = bch2_bkey_alloc(trans, &dst_iter, 0, subvolume);
590 	ret = PTR_ERR_OR_ZERO(new_subvol);
591 	if (ret)
592 		goto err;
593 
594 	new_subvol->v.flags		= 0;
595 	new_subvol->v.snapshot		= cpu_to_le32(new_nodes[0]);
596 	new_subvol->v.inode		= cpu_to_le64(inode);
597 	new_subvol->v.creation_parent	= cpu_to_le32(src_subvolid);
598 	new_subvol->v.fs_path_parent	= cpu_to_le32(parent_subvolid);
599 	new_subvol->v.otime.lo		= cpu_to_le64(bch2_current_time(c));
600 	new_subvol->v.otime.hi		= 0;
601 
602 	SET_BCH_SUBVOLUME_RO(&new_subvol->v, ro);
603 	SET_BCH_SUBVOLUME_SNAP(&new_subvol->v, src_subvolid != 0);
604 
605 	*new_subvolid	= new_subvol->k.p.offset;
606 	*new_snapshotid	= new_nodes[0];
607 err:
608 	bch2_trans_iter_exit(trans, &src_iter);
609 	bch2_trans_iter_exit(trans, &dst_iter);
610 	return ret;
611 }
612 
613 int bch2_initialize_subvolumes(struct bch_fs *c)
614 {
615 	struct bkey_i_snapshot_tree	root_tree;
616 	struct bkey_i_snapshot		root_snapshot;
617 	struct bkey_i_subvolume		root_volume;
618 	int ret;
619 
620 	bkey_snapshot_tree_init(&root_tree.k_i);
621 	root_tree.k.p.offset		= 1;
622 	root_tree.v.master_subvol	= cpu_to_le32(1);
623 	root_tree.v.root_snapshot	= cpu_to_le32(U32_MAX);
624 
625 	bkey_snapshot_init(&root_snapshot.k_i);
626 	root_snapshot.k.p.offset = U32_MAX;
627 	root_snapshot.v.flags	= 0;
628 	root_snapshot.v.parent	= 0;
629 	root_snapshot.v.subvol	= cpu_to_le32(BCACHEFS_ROOT_SUBVOL);
630 	root_snapshot.v.tree	= cpu_to_le32(1);
631 	SET_BCH_SNAPSHOT_SUBVOL(&root_snapshot.v, true);
632 
633 	bkey_subvolume_init(&root_volume.k_i);
634 	root_volume.k.p.offset = BCACHEFS_ROOT_SUBVOL;
635 	root_volume.v.flags	= 0;
636 	root_volume.v.snapshot	= cpu_to_le32(U32_MAX);
637 	root_volume.v.inode	= cpu_to_le64(BCACHEFS_ROOT_INO);
638 
639 	ret =   bch2_btree_insert(c, BTREE_ID_snapshot_trees,	&root_tree.k_i, NULL, 0, 0) ?:
640 		bch2_btree_insert(c, BTREE_ID_snapshots,	&root_snapshot.k_i, NULL, 0, 0) ?:
641 		bch2_btree_insert(c, BTREE_ID_subvolumes,	&root_volume.k_i, NULL, 0, 0);
642 	bch_err_fn(c, ret);
643 	return ret;
644 }
645 
646 static int __bch2_fs_upgrade_for_subvolumes(struct btree_trans *trans)
647 {
648 	struct btree_iter iter;
649 	struct bkey_s_c k;
650 	struct bch_inode_unpacked inode;
651 	int ret;
652 
653 	k = bch2_bkey_get_iter(trans, &iter, BTREE_ID_inodes,
654 			       SPOS(0, BCACHEFS_ROOT_INO, U32_MAX), 0);
655 	ret = bkey_err(k);
656 	if (ret)
657 		return ret;
658 
659 	if (!bkey_is_inode(k.k)) {
660 		bch_err(trans->c, "root inode not found");
661 		ret = -BCH_ERR_ENOENT_inode;
662 		goto err;
663 	}
664 
665 	ret = bch2_inode_unpack(k, &inode);
666 	BUG_ON(ret);
667 
668 	inode.bi_subvol = BCACHEFS_ROOT_SUBVOL;
669 
670 	ret = bch2_inode_write(trans, &iter, &inode);
671 err:
672 	bch2_trans_iter_exit(trans, &iter);
673 	return ret;
674 }
675 
676 /* set bi_subvol on root inode */
677 int bch2_fs_upgrade_for_subvolumes(struct bch_fs *c)
678 {
679 	int ret = bch2_trans_do(c, NULL, NULL, BCH_TRANS_COMMIT_lazy_rw,
680 				__bch2_fs_upgrade_for_subvolumes(trans));
681 	bch_err_fn(c, ret);
682 	return ret;
683 }
684 
685 int bch2_fs_subvolumes_init(struct bch_fs *c)
686 {
687 	INIT_WORK(&c->snapshot_delete_work, bch2_delete_dead_snapshots_work);
688 	INIT_WORK(&c->snapshot_wait_for_pagecache_and_delete_work,
689 		  bch2_subvolume_wait_for_pagecache_and_delete);
690 	mutex_init(&c->snapshots_unlinked_lock);
691 	return 0;
692 }
693