xref: /linux/fs/bcachefs/subvolume.c (revision 41f7adb676f6c4aef439d78b15c7e1119216bc2b)
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 			c, 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 				c, 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 	struct btree_iter inode_iter = {};
96 	ret = bch2_inode_peek_nowarn(trans, &inode_iter, &inode,
97 				    (subvol_inum) { k.k->p.offset, le64_to_cpu(subvol.v->inode) },
98 				    0);
99 	bch2_trans_iter_exit(trans, &inode_iter);
100 
101 	if (ret && !bch2_err_matches(ret, ENOENT))
102 		return ret;
103 
104 	if (fsck_err_on(ret, c, subvol_to_missing_root,
105 			"subvolume %llu points to missing subvolume root %llu:%u",
106 			k.k->p.offset, le64_to_cpu(subvol.v->inode),
107 			le32_to_cpu(subvol.v->snapshot))) {
108 		ret = bch2_subvolume_delete(trans, iter->pos.offset);
109 		bch_err_msg(c, ret, "deleting subvolume %llu", iter->pos.offset);
110 		return ret ?: -BCH_ERR_transaction_restart_nested;
111 	}
112 
113 	if (fsck_err_on(inode.bi_subvol != subvol.k->p.offset,
114 			c, subvol_root_wrong_bi_subvol,
115 			"subvol root %llu:%u has wrong bi_subvol field: got %u, should be %llu",
116 			inode.bi_inum, inode_iter.k.p.snapshot,
117 			inode.bi_subvol, subvol.k->p.offset)) {
118 		inode.bi_subvol = subvol.k->p.offset;
119 		ret = __bch2_fsck_write_inode(trans, &inode, le32_to_cpu(subvol.v->snapshot));
120 		if (ret)
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 			return ret;
140 
141 		if (fsck_err_on(le32_to_cpu(st.master_subvol) != subvol.k->p.offset,
142 				c, 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 				return ret;
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_fs *c = trans->c;
177 	struct bch_subvolume s;
178 	int ret = bch2_bkey_get_val_typed(trans, BTREE_ID_subvolumes, POS(0, child_k.k->p.offset),
179 					  0, subvolume, &s);
180 	if (ret && !bch2_err_matches(ret, ENOENT))
181 		return ret;
182 
183 	if (fsck_err_on(ret ||
184 			le32_to_cpu(s.fs_path_parent) != child_k.k->p.inode,
185 			c, subvol_children_bad,
186 			"incorrect entry in subvolume_children btree %llu:%llu",
187 			child_k.k->p.inode, child_k.k->p.offset)) {
188 		ret = bch2_btree_delete_at(trans, child_iter, 0);
189 		if (ret)
190 			goto err;
191 	}
192 err:
193 fsck_err:
194 	return ret;
195 }
196 
197 int bch2_check_subvol_children(struct bch_fs *c)
198 {
199 	int ret = bch2_trans_run(c,
200 		for_each_btree_key_commit(trans, iter,
201 				BTREE_ID_subvolume_children, POS_MIN, BTREE_ITER_prefetch, k,
202 				NULL, NULL, BCH_TRANS_COMMIT_no_enospc,
203 			check_subvol_child(trans, &iter, k)));
204 	bch_err_fn(c, ret);
205 	return 0;
206 }
207 
208 /* Subvolumes: */
209 
210 int bch2_subvolume_invalid(struct bch_fs *c, struct bkey_s_c k,
211 			   enum bch_validate_flags flags, struct printbuf *err)
212 {
213 	struct bkey_s_c_subvolume subvol = bkey_s_c_to_subvolume(k);
214 	int ret = 0;
215 
216 	bkey_fsck_err_on(bkey_lt(k.k->p, SUBVOL_POS_MIN) ||
217 			 bkey_gt(k.k->p, SUBVOL_POS_MAX), c, err,
218 			 subvol_pos_bad,
219 			 "invalid pos");
220 
221 	bkey_fsck_err_on(!subvol.v->snapshot, c, err,
222 			 subvol_snapshot_bad,
223 			 "invalid snapshot");
224 
225 	bkey_fsck_err_on(!subvol.v->inode, c, err,
226 			 subvol_inode_bad,
227 			 "invalid inode");
228 fsck_err:
229 	return ret;
230 }
231 
232 void bch2_subvolume_to_text(struct printbuf *out, struct bch_fs *c,
233 			    struct bkey_s_c k)
234 {
235 	struct bkey_s_c_subvolume s = bkey_s_c_to_subvolume(k);
236 
237 	prt_printf(out, "root %llu snapshot id %u",
238 		   le64_to_cpu(s.v->inode),
239 		   le32_to_cpu(s.v->snapshot));
240 
241 	if (bkey_val_bytes(s.k) > offsetof(struct bch_subvolume, creation_parent)) {
242 		prt_printf(out, " creation_parent %u", le32_to_cpu(s.v->creation_parent));
243 		prt_printf(out, " fs_parent %u", le32_to_cpu(s.v->fs_path_parent));
244 	}
245 }
246 
247 static int subvolume_children_mod(struct btree_trans *trans, struct bpos pos, bool set)
248 {
249 	return !bpos_eq(pos, POS_MIN)
250 		? bch2_btree_bit_mod(trans, BTREE_ID_subvolume_children, pos, set)
251 		: 0;
252 }
253 
254 int bch2_subvolume_trigger(struct btree_trans *trans,
255 			   enum btree_id btree_id, unsigned level,
256 			   struct bkey_s_c old, struct bkey_s new,
257 			   enum btree_iter_update_trigger_flags flags)
258 {
259 	if (flags & BTREE_TRIGGER_transactional) {
260 		struct bpos children_pos_old = subvolume_children_pos(old);
261 		struct bpos children_pos_new = subvolume_children_pos(new.s_c);
262 
263 		if (!bpos_eq(children_pos_old, children_pos_new)) {
264 			int ret = subvolume_children_mod(trans, children_pos_old, false) ?:
265 				  subvolume_children_mod(trans, children_pos_new, true);
266 			if (ret)
267 				return ret;
268 		}
269 	}
270 
271 	return 0;
272 }
273 
274 int bch2_subvol_has_children(struct btree_trans *trans, u32 subvol)
275 {
276 	struct btree_iter iter;
277 
278 	bch2_trans_iter_init(trans, &iter, BTREE_ID_subvolume_children, POS(subvol, 0), 0);
279 	struct bkey_s_c k = bch2_btree_iter_peek(&iter);
280 	bch2_trans_iter_exit(trans, &iter);
281 
282 	return bkey_err(k) ?: k.k && k.k->p.inode == subvol
283 		? -BCH_ERR_ENOTEMPTY_subvol_not_empty
284 		: 0;
285 }
286 
287 static __always_inline int
288 bch2_subvolume_get_inlined(struct btree_trans *trans, unsigned subvol,
289 			   bool inconsistent_if_not_found,
290 			   int iter_flags,
291 			   struct bch_subvolume *s)
292 {
293 	int ret = bch2_bkey_get_val_typed(trans, BTREE_ID_subvolumes, POS(0, subvol),
294 					  iter_flags, subvolume, s);
295 	bch2_fs_inconsistent_on(bch2_err_matches(ret, ENOENT) &&
296 				inconsistent_if_not_found,
297 				trans->c, "missing subvolume %u", subvol);
298 	return ret;
299 }
300 
301 int bch2_subvolume_get(struct btree_trans *trans, unsigned subvol,
302 		       bool inconsistent_if_not_found,
303 		       int iter_flags,
304 		       struct bch_subvolume *s)
305 {
306 	return bch2_subvolume_get_inlined(trans, subvol, inconsistent_if_not_found, iter_flags, s);
307 }
308 
309 int bch2_subvol_is_ro_trans(struct btree_trans *trans, u32 subvol)
310 {
311 	struct bch_subvolume s;
312 	int ret = bch2_subvolume_get_inlined(trans, subvol, true, 0, &s);
313 	if (ret)
314 		return ret;
315 
316 	if (BCH_SUBVOLUME_RO(&s))
317 		return -EROFS;
318 	return 0;
319 }
320 
321 int bch2_subvol_is_ro(struct bch_fs *c, u32 subvol)
322 {
323 	return bch2_trans_do(c, NULL, NULL, 0,
324 		bch2_subvol_is_ro_trans(trans, subvol));
325 }
326 
327 int bch2_snapshot_get_subvol(struct btree_trans *trans, u32 snapshot,
328 			     struct bch_subvolume *subvol)
329 {
330 	struct bch_snapshot snap;
331 
332 	return  bch2_snapshot_lookup(trans, snapshot, &snap) ?:
333 		bch2_subvolume_get(trans, le32_to_cpu(snap.subvol), true, 0, subvol);
334 }
335 
336 int bch2_subvolume_get_snapshot(struct btree_trans *trans, u32 subvolid,
337 				u32 *snapid)
338 {
339 	struct btree_iter iter;
340 	struct bkey_s_c_subvolume subvol;
341 	int ret;
342 
343 	subvol = bch2_bkey_get_iter_typed(trans, &iter,
344 					  BTREE_ID_subvolumes, POS(0, subvolid),
345 					  BTREE_ITER_cached|BTREE_ITER_with_updates,
346 					  subvolume);
347 	ret = bkey_err(subvol);
348 	bch2_fs_inconsistent_on(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 static int bch2_subvolume_reparent(struct btree_trans *trans,
358 				   struct btree_iter *iter,
359 				   struct bkey_s_c k,
360 				   u32 old_parent, u32 new_parent)
361 {
362 	struct bkey_i_subvolume *s;
363 	int ret;
364 
365 	if (k.k->type != KEY_TYPE_subvolume)
366 		return 0;
367 
368 	if (bkey_val_bytes(k.k) > offsetof(struct bch_subvolume, creation_parent) &&
369 	    le32_to_cpu(bkey_s_c_to_subvolume(k).v->creation_parent) != old_parent)
370 		return 0;
371 
372 	s = bch2_bkey_make_mut_typed(trans, iter, &k, 0, subvolume);
373 	ret = PTR_ERR_OR_ZERO(s);
374 	if (ret)
375 		return ret;
376 
377 	s->v.creation_parent = cpu_to_le32(new_parent);
378 	return 0;
379 }
380 
381 /*
382  * Separate from the snapshot tree in the snapshots btree, we record the tree
383  * structure of how snapshot subvolumes were created - the parent subvolume of
384  * each snapshot subvolume.
385  *
386  * When a subvolume is deleted, we scan for child subvolumes and reparant them,
387  * to avoid dangling references:
388  */
389 static int bch2_subvolumes_reparent(struct btree_trans *trans, u32 subvolid_to_delete)
390 {
391 	struct bch_subvolume s;
392 
393 	return lockrestart_do(trans,
394 			bch2_subvolume_get(trans, subvolid_to_delete, true,
395 				   BTREE_ITER_cached, &s)) ?:
396 		for_each_btree_key_commit(trans, iter,
397 				BTREE_ID_subvolumes, POS_MIN, BTREE_ITER_prefetch, k,
398 				NULL, NULL, BCH_TRANS_COMMIT_no_enospc,
399 			bch2_subvolume_reparent(trans, &iter, k,
400 					subvolid_to_delete, le32_to_cpu(s.creation_parent)));
401 }
402 
403 /*
404  * Delete subvolume, mark snapshot ID as deleted, queue up snapshot
405  * deletion/cleanup:
406  */
407 static int __bch2_subvolume_delete(struct btree_trans *trans, u32 subvolid)
408 {
409 	struct btree_iter iter;
410 	struct bkey_s_c_subvolume subvol;
411 	u32 snapid;
412 	int ret = 0;
413 
414 	subvol = bch2_bkey_get_iter_typed(trans, &iter,
415 				BTREE_ID_subvolumes, POS(0, subvolid),
416 				BTREE_ITER_cached|BTREE_ITER_intent,
417 				subvolume);
418 	ret = bkey_err(subvol);
419 	bch2_fs_inconsistent_on(bch2_err_matches(ret, ENOENT), trans->c,
420 				"missing subvolume %u", subvolid);
421 	if (ret)
422 		return ret;
423 
424 	snapid = le32_to_cpu(subvol.v->snapshot);
425 
426 	ret =   bch2_btree_delete_at(trans, &iter, 0) ?:
427 		bch2_snapshot_node_set_deleted(trans, snapid);
428 	bch2_trans_iter_exit(trans, &iter);
429 	return ret;
430 }
431 
432 static int bch2_subvolume_delete(struct btree_trans *trans, u32 subvolid)
433 {
434 	return bch2_subvolumes_reparent(trans, subvolid) ?:
435 		commit_do(trans, NULL, NULL, BCH_TRANS_COMMIT_no_enospc,
436 			  __bch2_subvolume_delete(trans, subvolid));
437 }
438 
439 static void bch2_subvolume_wait_for_pagecache_and_delete(struct work_struct *work)
440 {
441 	struct bch_fs *c = container_of(work, struct bch_fs,
442 				snapshot_wait_for_pagecache_and_delete_work);
443 	snapshot_id_list s;
444 	u32 *id;
445 	int ret = 0;
446 
447 	while (!ret) {
448 		mutex_lock(&c->snapshots_unlinked_lock);
449 		s = c->snapshots_unlinked;
450 		darray_init(&c->snapshots_unlinked);
451 		mutex_unlock(&c->snapshots_unlinked_lock);
452 
453 		if (!s.nr)
454 			break;
455 
456 		bch2_evict_subvolume_inodes(c, &s);
457 
458 		for (id = s.data; id < s.data + s.nr; id++) {
459 			ret = bch2_trans_run(c, bch2_subvolume_delete(trans, *id));
460 			bch_err_msg(c, ret, "deleting subvolume %u", *id);
461 			if (ret)
462 				break;
463 		}
464 
465 		darray_exit(&s);
466 	}
467 
468 	bch2_write_ref_put(c, BCH_WRITE_REF_snapshot_delete_pagecache);
469 }
470 
471 struct subvolume_unlink_hook {
472 	struct btree_trans_commit_hook	h;
473 	u32				subvol;
474 };
475 
476 static int bch2_subvolume_wait_for_pagecache_and_delete_hook(struct btree_trans *trans,
477 						      struct btree_trans_commit_hook *_h)
478 {
479 	struct subvolume_unlink_hook *h = container_of(_h, struct subvolume_unlink_hook, h);
480 	struct bch_fs *c = trans->c;
481 	int ret = 0;
482 
483 	mutex_lock(&c->snapshots_unlinked_lock);
484 	if (!snapshot_list_has_id(&c->snapshots_unlinked, h->subvol))
485 		ret = snapshot_list_add(c, &c->snapshots_unlinked, h->subvol);
486 	mutex_unlock(&c->snapshots_unlinked_lock);
487 
488 	if (ret)
489 		return ret;
490 
491 	if (!bch2_write_ref_tryget(c, BCH_WRITE_REF_snapshot_delete_pagecache))
492 		return -EROFS;
493 
494 	if (!queue_work(c->write_ref_wq, &c->snapshot_wait_for_pagecache_and_delete_work))
495 		bch2_write_ref_put(c, BCH_WRITE_REF_snapshot_delete_pagecache);
496 	return 0;
497 }
498 
499 int bch2_subvolume_unlink(struct btree_trans *trans, u32 subvolid)
500 {
501 	struct btree_iter iter;
502 	struct bkey_i_subvolume *n;
503 	struct subvolume_unlink_hook *h;
504 	int ret = 0;
505 
506 	h = bch2_trans_kmalloc(trans, sizeof(*h));
507 	ret = PTR_ERR_OR_ZERO(h);
508 	if (ret)
509 		return ret;
510 
511 	h->h.fn		= bch2_subvolume_wait_for_pagecache_and_delete_hook;
512 	h->subvol	= subvolid;
513 	bch2_trans_commit_hook(trans, &h->h);
514 
515 	n = bch2_bkey_get_mut_typed(trans, &iter,
516 			BTREE_ID_subvolumes, POS(0, subvolid),
517 			BTREE_ITER_cached, subvolume);
518 	ret = PTR_ERR_OR_ZERO(n);
519 	if (unlikely(ret)) {
520 		bch2_fs_inconsistent_on(bch2_err_matches(ret, ENOENT), trans->c,
521 					"missing subvolume %u", subvolid);
522 		return ret;
523 	}
524 
525 	SET_BCH_SUBVOLUME_UNLINKED(&n->v, true);
526 	bch2_trans_iter_exit(trans, &iter);
527 	return ret;
528 }
529 
530 int bch2_subvolume_create(struct btree_trans *trans, u64 inode,
531 			  u32 parent_subvolid,
532 			  u32 src_subvolid,
533 			  u32 *new_subvolid,
534 			  u32 *new_snapshotid,
535 			  bool ro)
536 {
537 	struct bch_fs *c = trans->c;
538 	struct btree_iter dst_iter, src_iter = (struct btree_iter) { NULL };
539 	struct bkey_i_subvolume *new_subvol = NULL;
540 	struct bkey_i_subvolume *src_subvol = NULL;
541 	u32 parent = 0, new_nodes[2], snapshot_subvols[2];
542 	int ret = 0;
543 
544 	ret = bch2_bkey_get_empty_slot(trans, &dst_iter,
545 				BTREE_ID_subvolumes, POS(0, U32_MAX));
546 	if (ret == -BCH_ERR_ENOSPC_btree_slot)
547 		ret = -BCH_ERR_ENOSPC_subvolume_create;
548 	if (ret)
549 		return ret;
550 
551 	snapshot_subvols[0] = dst_iter.pos.offset;
552 	snapshot_subvols[1] = src_subvolid;
553 
554 	if (src_subvolid) {
555 		/* Creating a snapshot: */
556 
557 		src_subvol = bch2_bkey_get_mut_typed(trans, &src_iter,
558 				BTREE_ID_subvolumes, POS(0, src_subvolid),
559 				BTREE_ITER_cached, subvolume);
560 		ret = PTR_ERR_OR_ZERO(src_subvol);
561 		if (unlikely(ret)) {
562 			bch2_fs_inconsistent_on(bch2_err_matches(ret, ENOENT), c,
563 						"subvolume %u not found", src_subvolid);
564 			goto err;
565 		}
566 
567 		parent = le32_to_cpu(src_subvol->v.snapshot);
568 	}
569 
570 	ret = bch2_snapshot_node_create(trans, parent, new_nodes,
571 					snapshot_subvols,
572 					src_subvolid ? 2 : 1);
573 	if (ret)
574 		goto err;
575 
576 	if (src_subvolid) {
577 		src_subvol->v.snapshot = cpu_to_le32(new_nodes[1]);
578 		ret = bch2_trans_update(trans, &src_iter, &src_subvol->k_i, 0);
579 		if (ret)
580 			goto err;
581 	}
582 
583 	new_subvol = bch2_bkey_alloc(trans, &dst_iter, 0, subvolume);
584 	ret = PTR_ERR_OR_ZERO(new_subvol);
585 	if (ret)
586 		goto err;
587 
588 	new_subvol->v.flags		= 0;
589 	new_subvol->v.snapshot		= cpu_to_le32(new_nodes[0]);
590 	new_subvol->v.inode		= cpu_to_le64(inode);
591 	new_subvol->v.creation_parent	= cpu_to_le32(src_subvolid);
592 	new_subvol->v.fs_path_parent	= cpu_to_le32(parent_subvolid);
593 	new_subvol->v.otime.lo		= cpu_to_le64(bch2_current_time(c));
594 	new_subvol->v.otime.hi		= 0;
595 
596 	SET_BCH_SUBVOLUME_RO(&new_subvol->v, ro);
597 	SET_BCH_SUBVOLUME_SNAP(&new_subvol->v, src_subvolid != 0);
598 
599 	*new_subvolid	= new_subvol->k.p.offset;
600 	*new_snapshotid	= new_nodes[0];
601 err:
602 	bch2_trans_iter_exit(trans, &src_iter);
603 	bch2_trans_iter_exit(trans, &dst_iter);
604 	return ret;
605 }
606 
607 int bch2_initialize_subvolumes(struct bch_fs *c)
608 {
609 	struct bkey_i_snapshot_tree	root_tree;
610 	struct bkey_i_snapshot		root_snapshot;
611 	struct bkey_i_subvolume		root_volume;
612 	int ret;
613 
614 	bkey_snapshot_tree_init(&root_tree.k_i);
615 	root_tree.k.p.offset		= 1;
616 	root_tree.v.master_subvol	= cpu_to_le32(1);
617 	root_tree.v.root_snapshot	= cpu_to_le32(U32_MAX);
618 
619 	bkey_snapshot_init(&root_snapshot.k_i);
620 	root_snapshot.k.p.offset = U32_MAX;
621 	root_snapshot.v.flags	= 0;
622 	root_snapshot.v.parent	= 0;
623 	root_snapshot.v.subvol	= cpu_to_le32(BCACHEFS_ROOT_SUBVOL);
624 	root_snapshot.v.tree	= cpu_to_le32(1);
625 	SET_BCH_SNAPSHOT_SUBVOL(&root_snapshot.v, true);
626 
627 	bkey_subvolume_init(&root_volume.k_i);
628 	root_volume.k.p.offset = BCACHEFS_ROOT_SUBVOL;
629 	root_volume.v.flags	= 0;
630 	root_volume.v.snapshot	= cpu_to_le32(U32_MAX);
631 	root_volume.v.inode	= cpu_to_le64(BCACHEFS_ROOT_INO);
632 
633 	ret =   bch2_btree_insert(c, BTREE_ID_snapshot_trees,	&root_tree.k_i, NULL, 0) ?:
634 		bch2_btree_insert(c, BTREE_ID_snapshots,	&root_snapshot.k_i, NULL, 0) ?:
635 		bch2_btree_insert(c, BTREE_ID_subvolumes,	&root_volume.k_i, NULL, 0);
636 	bch_err_fn(c, ret);
637 	return ret;
638 }
639 
640 static int __bch2_fs_upgrade_for_subvolumes(struct btree_trans *trans)
641 {
642 	struct btree_iter iter;
643 	struct bkey_s_c k;
644 	struct bch_inode_unpacked inode;
645 	int ret;
646 
647 	k = bch2_bkey_get_iter(trans, &iter, BTREE_ID_inodes,
648 			       SPOS(0, BCACHEFS_ROOT_INO, U32_MAX), 0);
649 	ret = bkey_err(k);
650 	if (ret)
651 		return ret;
652 
653 	if (!bkey_is_inode(k.k)) {
654 		bch_err(trans->c, "root inode not found");
655 		ret = -BCH_ERR_ENOENT_inode;
656 		goto err;
657 	}
658 
659 	ret = bch2_inode_unpack(k, &inode);
660 	BUG_ON(ret);
661 
662 	inode.bi_subvol = BCACHEFS_ROOT_SUBVOL;
663 
664 	ret = bch2_inode_write(trans, &iter, &inode);
665 err:
666 	bch2_trans_iter_exit(trans, &iter);
667 	return ret;
668 }
669 
670 /* set bi_subvol on root inode */
671 int bch2_fs_upgrade_for_subvolumes(struct bch_fs *c)
672 {
673 	int ret = bch2_trans_do(c, NULL, NULL, BCH_TRANS_COMMIT_lazy_rw,
674 				__bch2_fs_upgrade_for_subvolumes(trans));
675 	bch_err_fn(c, ret);
676 	return ret;
677 }
678 
679 int bch2_fs_subvolumes_init(struct bch_fs *c)
680 {
681 	INIT_WORK(&c->snapshot_delete_work, bch2_delete_dead_snapshots_work);
682 	INIT_WORK(&c->snapshot_wait_for_pagecache_and_delete_work,
683 		  bch2_subvolume_wait_for_pagecache_and_delete);
684 	mutex_init(&c->snapshots_unlinked_lock);
685 	return 0;
686 }
687