xref: /linux/fs/bcachefs/subvolume.c (revision ff0905bbf991f4337b5ebc19c0d43525ebb0d96b)
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 "enumerated_ref.h"
7 #include "errcode.h"
8 #include "error.h"
9 #include "fs.h"
10 #include "recovery_passes.h"
11 #include "snapshot.h"
12 #include "subvolume.h"
13 
14 #include <linux/random.h>
15 
16 static int bch2_subvolume_delete(struct btree_trans *, u32);
17 
bch2_subvolume_missing(struct bch_fs * c,u32 subvolid)18 static int bch2_subvolume_missing(struct bch_fs *c, u32 subvolid)
19 {
20 	struct printbuf buf = PRINTBUF;
21 	bch2_log_msg_start(c, &buf);
22 
23 	prt_printf(&buf, "missing subvolume %u", subvolid);
24 	bool print = bch2_count_fsck_err(c, subvol_missing, &buf);
25 
26 	int ret = bch2_run_explicit_recovery_pass(c, &buf,
27 					BCH_RECOVERY_PASS_check_inodes, 0);
28 	if (print)
29 		bch2_print_str(c, KERN_ERR, buf.buf);
30 	printbuf_exit(&buf);
31 	return ret;
32 }
33 
subvolume_children_pos(struct bkey_s_c k)34 static struct bpos subvolume_children_pos(struct bkey_s_c k)
35 {
36 	if (k.k->type != KEY_TYPE_subvolume)
37 		return POS_MIN;
38 
39 	struct bkey_s_c_subvolume s = bkey_s_c_to_subvolume(k);
40 	if (!s.v->fs_path_parent)
41 		return POS_MIN;
42 	return POS(le32_to_cpu(s.v->fs_path_parent), s.k->p.offset);
43 }
44 
check_subvol(struct btree_trans * trans,struct btree_iter * iter,struct bkey_s_c k)45 static int check_subvol(struct btree_trans *trans,
46 			struct btree_iter *iter,
47 			struct bkey_s_c k)
48 {
49 	struct bch_fs *c = trans->c;
50 	struct bkey_s_c_subvolume subvol;
51 	struct btree_iter subvol_children_iter = {};
52 	struct bch_snapshot snapshot;
53 	struct printbuf buf = PRINTBUF;
54 	unsigned snapid;
55 	int ret = 0;
56 
57 	if (k.k->type != KEY_TYPE_subvolume)
58 		return 0;
59 
60 	subvol = bkey_s_c_to_subvolume(k);
61 	snapid = le32_to_cpu(subvol.v->snapshot);
62 	ret = bch2_snapshot_lookup(trans, snapid, &snapshot);
63 
64 	if (bch2_err_matches(ret, ENOENT))
65 		return bch2_run_print_explicit_recovery_pass(c,
66 					BCH_RECOVERY_PASS_reconstruct_snapshots) ?: ret;
67 	if (ret)
68 		return ret;
69 
70 	if (BCH_SUBVOLUME_UNLINKED(subvol.v)) {
71 		ret = bch2_subvolume_delete(trans, iter->pos.offset);
72 		bch_err_msg(c, ret, "deleting subvolume %llu", iter->pos.offset);
73 		return ret ?: -BCH_ERR_transaction_restart_nested;
74 	}
75 
76 	if (fsck_err_on(subvol.k->p.offset == BCACHEFS_ROOT_SUBVOL &&
77 			subvol.v->fs_path_parent,
78 			trans, subvol_root_fs_path_parent_nonzero,
79 			"root subvolume has nonzero fs_path_parent\n%s",
80 			(bch2_bkey_val_to_text(&buf, c, k), buf.buf))) {
81 		struct bkey_i_subvolume *n =
82 			bch2_bkey_make_mut_typed(trans, iter, &subvol.s_c, 0, subvolume);
83 		ret = PTR_ERR_OR_ZERO(n);
84 		if (ret)
85 			goto err;
86 
87 		n->v.fs_path_parent = 0;
88 	}
89 
90 	if (subvol.v->fs_path_parent) {
91 		struct bpos pos = subvolume_children_pos(k);
92 
93 		struct bkey_s_c subvol_children_k =
94 			bch2_bkey_get_iter(trans, &subvol_children_iter,
95 					   BTREE_ID_subvolume_children, pos, 0);
96 		ret = bkey_err(subvol_children_k);
97 		if (ret)
98 			goto err;
99 
100 		if (fsck_err_on(subvol_children_k.k->type != KEY_TYPE_set,
101 				trans, subvol_children_not_set,
102 				"subvolume not set in subvolume_children btree at %llu:%llu\n%s",
103 				pos.inode, pos.offset,
104 				(printbuf_reset(&buf),
105 				 bch2_bkey_val_to_text(&buf, c, k), buf.buf))) {
106 			ret = bch2_btree_bit_mod(trans, BTREE_ID_subvolume_children, pos, true);
107 			if (ret)
108 				goto err;
109 		}
110 	}
111 
112 	struct bch_inode_unpacked inode;
113 	ret = bch2_inode_find_by_inum_nowarn_trans(trans,
114 				    (subvol_inum) { k.k->p.offset, le64_to_cpu(subvol.v->inode) },
115 				    &inode);
116 	if (!ret) {
117 		if (fsck_err_on(inode.bi_subvol != subvol.k->p.offset,
118 				trans, subvol_root_wrong_bi_subvol,
119 				"subvol root %llu:%u has wrong bi_subvol field: got %u, should be %llu",
120 				inode.bi_inum, inode.bi_snapshot,
121 				inode.bi_subvol, subvol.k->p.offset)) {
122 			inode.bi_subvol = subvol.k->p.offset;
123 			inode.bi_snapshot = le32_to_cpu(subvol.v->snapshot);
124 			ret = __bch2_fsck_write_inode(trans, &inode);
125 			if (ret)
126 				goto err;
127 		}
128 	} else if (bch2_err_matches(ret, ENOENT)) {
129 		if (fsck_err(trans, subvol_to_missing_root,
130 			     "subvolume %llu points to missing subvolume root %llu:%u",
131 			     k.k->p.offset, le64_to_cpu(subvol.v->inode),
132 			     le32_to_cpu(subvol.v->snapshot))) {
133 			/*
134 			 * Recreate - any contents that are still disconnected
135 			 * will then get reattached under lost+found
136 			 */
137 			bch2_inode_init_early(c, &inode);
138 			bch2_inode_init_late(c, &inode, bch2_current_time(c),
139 					     0, 0, S_IFDIR|0700, 0, NULL);
140 			inode.bi_inum			= le64_to_cpu(subvol.v->inode);
141 			inode.bi_snapshot		= le32_to_cpu(subvol.v->snapshot);
142 			inode.bi_subvol			= k.k->p.offset;
143 			inode.bi_parent_subvol		= le32_to_cpu(subvol.v->fs_path_parent);
144 			ret = __bch2_fsck_write_inode(trans, &inode);
145 			if (ret)
146 				goto err;
147 		}
148 	} else {
149 		goto err;
150 	}
151 
152 	if (!BCH_SUBVOLUME_SNAP(subvol.v)) {
153 		u32 snapshot_root = bch2_snapshot_root(c, le32_to_cpu(subvol.v->snapshot));
154 		u32 snapshot_tree = bch2_snapshot_tree(c, snapshot_root);
155 
156 		struct bch_snapshot_tree st;
157 		ret = bch2_snapshot_tree_lookup(trans, snapshot_tree, &st);
158 
159 		bch2_fs_inconsistent_on(bch2_err_matches(ret, ENOENT), c,
160 				"%s: snapshot tree %u not found", __func__, snapshot_tree);
161 
162 		if (ret)
163 			goto err;
164 
165 		if (fsck_err_on(le32_to_cpu(st.master_subvol) != subvol.k->p.offset,
166 				trans, subvol_not_master_and_not_snapshot,
167 				"subvolume %llu is not set as snapshot but is not master subvolume",
168 				k.k->p.offset)) {
169 			struct bkey_i_subvolume *s =
170 				bch2_bkey_make_mut_typed(trans, iter, &subvol.s_c, 0, subvolume);
171 			ret = PTR_ERR_OR_ZERO(s);
172 			if (ret)
173 				goto err;
174 
175 			SET_BCH_SUBVOLUME_SNAP(&s->v, true);
176 		}
177 	}
178 err:
179 fsck_err:
180 	bch2_trans_iter_exit(trans, &subvol_children_iter);
181 	printbuf_exit(&buf);
182 	return ret;
183 }
184 
bch2_check_subvols(struct bch_fs * c)185 int bch2_check_subvols(struct bch_fs *c)
186 {
187 	int ret = bch2_trans_run(c,
188 		for_each_btree_key_commit(trans, iter,
189 				BTREE_ID_subvolumes, POS_MIN, BTREE_ITER_prefetch, k,
190 				NULL, NULL, BCH_TRANS_COMMIT_no_enospc,
191 			check_subvol(trans, &iter, k)));
192 	bch_err_fn(c, ret);
193 	return ret;
194 }
195 
check_subvol_child(struct btree_trans * trans,struct btree_iter * child_iter,struct bkey_s_c child_k)196 static int check_subvol_child(struct btree_trans *trans,
197 			      struct btree_iter *child_iter,
198 			      struct bkey_s_c child_k)
199 {
200 	struct bch_subvolume s;
201 	int ret = bch2_bkey_get_val_typed(trans, BTREE_ID_subvolumes, POS(0, child_k.k->p.offset),
202 					  0, subvolume, &s);
203 	if (ret && !bch2_err_matches(ret, ENOENT))
204 		return ret;
205 
206 	if (fsck_err_on(ret ||
207 			le32_to_cpu(s.fs_path_parent) != child_k.k->p.inode,
208 			trans, subvol_children_bad,
209 			"incorrect entry in subvolume_children btree %llu:%llu",
210 			child_k.k->p.inode, child_k.k->p.offset)) {
211 		ret = bch2_btree_delete_at(trans, child_iter, 0);
212 		if (ret)
213 			goto err;
214 	}
215 err:
216 fsck_err:
217 	return ret;
218 }
219 
bch2_check_subvol_children(struct bch_fs * c)220 int bch2_check_subvol_children(struct bch_fs *c)
221 {
222 	int ret = bch2_trans_run(c,
223 		for_each_btree_key_commit(trans, iter,
224 				BTREE_ID_subvolume_children, POS_MIN, BTREE_ITER_prefetch, k,
225 				NULL, NULL, BCH_TRANS_COMMIT_no_enospc,
226 			check_subvol_child(trans, &iter, k)));
227 	bch_err_fn(c, ret);
228 	return 0;
229 }
230 
231 /* Subvolumes: */
232 
bch2_subvolume_validate(struct bch_fs * c,struct bkey_s_c k,struct bkey_validate_context from)233 int bch2_subvolume_validate(struct bch_fs *c, struct bkey_s_c k,
234 			    struct bkey_validate_context from)
235 {
236 	struct bkey_s_c_subvolume subvol = bkey_s_c_to_subvolume(k);
237 	int ret = 0;
238 
239 	bkey_fsck_err_on(bkey_lt(k.k->p, SUBVOL_POS_MIN) ||
240 			 bkey_gt(k.k->p, SUBVOL_POS_MAX),
241 			 c, subvol_pos_bad,
242 			 "invalid pos");
243 
244 	bkey_fsck_err_on(!subvol.v->snapshot,
245 			 c, subvol_snapshot_bad,
246 			 "invalid snapshot");
247 
248 	bkey_fsck_err_on(!subvol.v->inode,
249 			 c, subvol_inode_bad,
250 			 "invalid inode");
251 fsck_err:
252 	return ret;
253 }
254 
bch2_subvolume_to_text(struct printbuf * out,struct bch_fs * c,struct bkey_s_c k)255 void bch2_subvolume_to_text(struct printbuf *out, struct bch_fs *c,
256 			    struct bkey_s_c k)
257 {
258 	struct bkey_s_c_subvolume s = bkey_s_c_to_subvolume(k);
259 
260 	prt_printf(out, "root %llu snapshot id %u",
261 		   le64_to_cpu(s.v->inode),
262 		   le32_to_cpu(s.v->snapshot));
263 
264 	if (bkey_val_bytes(s.k) > offsetof(struct bch_subvolume, creation_parent)) {
265 		prt_printf(out, " creation_parent %u", le32_to_cpu(s.v->creation_parent));
266 		prt_printf(out, " fs_parent %u", le32_to_cpu(s.v->fs_path_parent));
267 	}
268 
269 	if (BCH_SUBVOLUME_RO(s.v))
270 		prt_printf(out, " ro");
271 	if (BCH_SUBVOLUME_SNAP(s.v))
272 		prt_printf(out, " snapshot");
273 	if (BCH_SUBVOLUME_UNLINKED(s.v))
274 		prt_printf(out, " unlinked");
275 }
276 
subvolume_children_mod(struct btree_trans * trans,struct bpos pos,bool set)277 static int subvolume_children_mod(struct btree_trans *trans, struct bpos pos, bool set)
278 {
279 	return !bpos_eq(pos, POS_MIN)
280 		? bch2_btree_bit_mod(trans, BTREE_ID_subvolume_children, pos, set)
281 		: 0;
282 }
283 
bch2_subvolume_trigger(struct btree_trans * trans,enum btree_id btree_id,unsigned level,struct bkey_s_c old,struct bkey_s new,enum btree_iter_update_trigger_flags flags)284 int bch2_subvolume_trigger(struct btree_trans *trans,
285 			   enum btree_id btree_id, unsigned level,
286 			   struct bkey_s_c old, struct bkey_s new,
287 			   enum btree_iter_update_trigger_flags flags)
288 {
289 	if (flags & BTREE_TRIGGER_transactional) {
290 		struct bpos children_pos_old = subvolume_children_pos(old);
291 		struct bpos children_pos_new = subvolume_children_pos(new.s_c);
292 
293 		if (!bpos_eq(children_pos_old, children_pos_new)) {
294 			int ret = subvolume_children_mod(trans, children_pos_old, false) ?:
295 				  subvolume_children_mod(trans, children_pos_new, true);
296 			if (ret)
297 				return ret;
298 		}
299 	}
300 
301 	return 0;
302 }
303 
bch2_subvol_has_children(struct btree_trans * trans,u32 subvol)304 int bch2_subvol_has_children(struct btree_trans *trans, u32 subvol)
305 {
306 	struct btree_iter iter;
307 
308 	bch2_trans_iter_init(trans, &iter, BTREE_ID_subvolume_children, POS(subvol, 0), 0);
309 	struct bkey_s_c k = bch2_btree_iter_peek(trans, &iter);
310 	bch2_trans_iter_exit(trans, &iter);
311 
312 	return bkey_err(k) ?: k.k && k.k->p.inode == subvol
313 		? -BCH_ERR_ENOTEMPTY_subvol_not_empty
314 		: 0;
315 }
316 
317 static __always_inline int
bch2_subvolume_get_inlined(struct btree_trans * trans,unsigned subvol,bool inconsistent_if_not_found,struct bch_subvolume * s)318 bch2_subvolume_get_inlined(struct btree_trans *trans, unsigned subvol,
319 			   bool inconsistent_if_not_found,
320 			   struct bch_subvolume *s)
321 {
322 	int ret = bch2_bkey_get_val_typed(trans, BTREE_ID_subvolumes, POS(0, subvol),
323 					  BTREE_ITER_cached|
324 					  BTREE_ITER_with_updates, subvolume, s);
325 	if (bch2_err_matches(ret, ENOENT) && inconsistent_if_not_found)
326 		ret = bch2_subvolume_missing(trans->c, subvol) ?: ret;
327 	return ret;
328 }
329 
bch2_subvolume_get(struct btree_trans * trans,unsigned subvol,bool inconsistent_if_not_found,struct bch_subvolume * s)330 int bch2_subvolume_get(struct btree_trans *trans, unsigned subvol,
331 		       bool inconsistent_if_not_found,
332 		       struct bch_subvolume *s)
333 {
334 	return bch2_subvolume_get_inlined(trans, subvol, inconsistent_if_not_found, s);
335 }
336 
bch2_subvol_is_ro_trans(struct btree_trans * trans,u32 subvol)337 int bch2_subvol_is_ro_trans(struct btree_trans *trans, u32 subvol)
338 {
339 	struct bch_subvolume s;
340 	int ret = bch2_subvolume_get_inlined(trans, subvol, true, &s);
341 	if (ret)
342 		return ret;
343 
344 	if (BCH_SUBVOLUME_RO(&s))
345 		return -EROFS;
346 	return 0;
347 }
348 
bch2_subvol_is_ro(struct bch_fs * c,u32 subvol)349 int bch2_subvol_is_ro(struct bch_fs *c, u32 subvol)
350 {
351 	return bch2_trans_do(c, bch2_subvol_is_ro_trans(trans, subvol));
352 }
353 
bch2_snapshot_get_subvol(struct btree_trans * trans,u32 snapshot,struct bch_subvolume * subvol)354 int bch2_snapshot_get_subvol(struct btree_trans *trans, u32 snapshot,
355 			     struct bch_subvolume *subvol)
356 {
357 	struct bch_snapshot snap;
358 
359 	return  bch2_snapshot_lookup(trans, snapshot, &snap) ?:
360 		bch2_subvolume_get(trans, le32_to_cpu(snap.subvol), true, subvol);
361 }
362 
__bch2_subvolume_get_snapshot(struct btree_trans * trans,u32 subvolid,u32 * snapid,bool warn)363 int __bch2_subvolume_get_snapshot(struct btree_trans *trans, u32 subvolid,
364 				  u32 *snapid, bool warn)
365 {
366 	struct btree_iter iter;
367 	struct bkey_s_c_subvolume subvol;
368 	int ret;
369 
370 	subvol = bch2_bkey_get_iter_typed(trans, &iter,
371 					  BTREE_ID_subvolumes, POS(0, subvolid),
372 					  BTREE_ITER_cached|BTREE_ITER_with_updates,
373 					  subvolume);
374 	ret = bkey_err(subvol);
375 
376 	if (bch2_err_matches(ret, ENOENT))
377 		ret = bch2_subvolume_missing(trans->c, subvolid) ?: ret;
378 
379 	if (likely(!ret))
380 		*snapid = le32_to_cpu(subvol.v->snapshot);
381 	bch2_trans_iter_exit(trans, &iter);
382 	return ret;
383 }
384 
bch2_subvolume_get_snapshot(struct btree_trans * trans,u32 subvolid,u32 * snapid)385 int bch2_subvolume_get_snapshot(struct btree_trans *trans, u32 subvolid,
386 				u32 *snapid)
387 {
388 	return __bch2_subvolume_get_snapshot(trans, subvolid, snapid, true);
389 }
390 
bch2_subvolume_reparent(struct btree_trans * trans,struct btree_iter * iter,struct bkey_s_c k,u32 old_parent,u32 new_parent)391 static int bch2_subvolume_reparent(struct btree_trans *trans,
392 				   struct btree_iter *iter,
393 				   struct bkey_s_c k,
394 				   u32 old_parent, u32 new_parent)
395 {
396 	struct bkey_i_subvolume *s;
397 	int ret;
398 
399 	if (k.k->type != KEY_TYPE_subvolume)
400 		return 0;
401 
402 	if (bkey_val_bytes(k.k) > offsetof(struct bch_subvolume, creation_parent) &&
403 	    le32_to_cpu(bkey_s_c_to_subvolume(k).v->creation_parent) != old_parent)
404 		return 0;
405 
406 	s = bch2_bkey_make_mut_typed(trans, iter, &k, 0, subvolume);
407 	ret = PTR_ERR_OR_ZERO(s);
408 	if (ret)
409 		return ret;
410 
411 	s->v.creation_parent = cpu_to_le32(new_parent);
412 	return 0;
413 }
414 
415 /*
416  * Separate from the snapshot tree in the snapshots btree, we record the tree
417  * structure of how snapshot subvolumes were created - the parent subvolume of
418  * each snapshot subvolume.
419  *
420  * When a subvolume is deleted, we scan for child subvolumes and reparant them,
421  * to avoid dangling references:
422  */
bch2_subvolumes_reparent(struct btree_trans * trans,u32 subvolid_to_delete)423 static int bch2_subvolumes_reparent(struct btree_trans *trans, u32 subvolid_to_delete)
424 {
425 	struct bch_subvolume s;
426 
427 	return lockrestart_do(trans,
428 			bch2_subvolume_get(trans, subvolid_to_delete, true, &s)) ?:
429 		for_each_btree_key_commit(trans, iter,
430 				BTREE_ID_subvolumes, POS_MIN, BTREE_ITER_prefetch, k,
431 				NULL, NULL, BCH_TRANS_COMMIT_no_enospc,
432 			bch2_subvolume_reparent(trans, &iter, k,
433 					subvolid_to_delete, le32_to_cpu(s.creation_parent)));
434 }
435 
436 /*
437  * Delete subvolume, mark snapshot ID as deleted, queue up snapshot
438  * deletion/cleanup:
439  */
__bch2_subvolume_delete(struct btree_trans * trans,u32 subvolid)440 static int __bch2_subvolume_delete(struct btree_trans *trans, u32 subvolid)
441 {
442 	struct btree_iter subvol_iter = {}, snapshot_iter = {}, snapshot_tree_iter = {};
443 
444 	struct bkey_s_c_subvolume subvol =
445 		bch2_bkey_get_iter_typed(trans, &subvol_iter,
446 				BTREE_ID_subvolumes, POS(0, subvolid),
447 				BTREE_ITER_cached|BTREE_ITER_intent,
448 				subvolume);
449 	int ret = bkey_err(subvol);
450 	if (bch2_err_matches(ret, ENOENT))
451 		ret = bch2_subvolume_missing(trans->c, subvolid) ?: ret;
452 	if (ret)
453 		goto err;
454 
455 	u32 snapid = le32_to_cpu(subvol.v->snapshot);
456 
457 	struct bkey_s_c_snapshot snapshot =
458 		bch2_bkey_get_iter_typed(trans, &snapshot_iter,
459 				BTREE_ID_snapshots, POS(0, snapid),
460 				0, snapshot);
461 	ret = bkey_err(snapshot);
462 	bch2_fs_inconsistent_on(bch2_err_matches(ret, ENOENT), trans->c,
463 				"missing snapshot %u", snapid);
464 	if (ret)
465 		goto err;
466 
467 	u32 treeid = le32_to_cpu(snapshot.v->tree);
468 
469 	struct bkey_s_c_snapshot_tree snapshot_tree =
470 		bch2_bkey_get_iter_typed(trans, &snapshot_tree_iter,
471 				BTREE_ID_snapshot_trees, POS(0, treeid),
472 				0, snapshot_tree);
473 	ret = bkey_err(snapshot_tree);
474 	bch2_fs_inconsistent_on(bch2_err_matches(ret, ENOENT), trans->c,
475 				"missing snapshot tree %u", treeid);
476 	if (ret)
477 		goto err;
478 
479 	if (le32_to_cpu(snapshot_tree.v->master_subvol) == subvolid) {
480 		struct bkey_i_snapshot_tree *snapshot_tree_mut =
481 			bch2_bkey_make_mut_typed(trans, &snapshot_tree_iter,
482 						 &snapshot_tree.s_c,
483 						 0, snapshot_tree);
484 		ret = PTR_ERR_OR_ZERO(snapshot_tree_mut);
485 		if (ret)
486 			goto err;
487 
488 		snapshot_tree_mut->v.master_subvol = 0;
489 	}
490 
491 	ret =   bch2_btree_delete_at(trans, &subvol_iter, 0) ?:
492 		bch2_snapshot_node_set_deleted(trans, snapid);
493 err:
494 	bch2_trans_iter_exit(trans, &snapshot_tree_iter);
495 	bch2_trans_iter_exit(trans, &snapshot_iter);
496 	bch2_trans_iter_exit(trans, &subvol_iter);
497 	return ret;
498 }
499 
bch2_subvolume_delete(struct btree_trans * trans,u32 subvolid)500 static int bch2_subvolume_delete(struct btree_trans *trans, u32 subvolid)
501 {
502 	int ret = bch2_subvolumes_reparent(trans, subvolid) ?:
503 		commit_do(trans, NULL, NULL, BCH_TRANS_COMMIT_no_enospc,
504 			  __bch2_subvolume_delete(trans, subvolid));
505 
506 	bch2_recovery_pass_set_no_ratelimit(trans->c, BCH_RECOVERY_PASS_check_subvols);
507 	return ret;
508 }
509 
bch2_subvolume_wait_for_pagecache_and_delete(struct work_struct * work)510 static void bch2_subvolume_wait_for_pagecache_and_delete(struct work_struct *work)
511 {
512 	struct bch_fs *c = container_of(work, struct bch_fs,
513 				snapshot_wait_for_pagecache_and_delete_work);
514 	int ret = 0;
515 
516 	while (!ret) {
517 		mutex_lock(&c->snapshots_unlinked_lock);
518 		snapshot_id_list s = c->snapshots_unlinked;
519 		darray_init(&c->snapshots_unlinked);
520 		mutex_unlock(&c->snapshots_unlinked_lock);
521 
522 		if (!s.nr)
523 			break;
524 
525 		bch2_evict_subvolume_inodes(c, &s);
526 
527 		darray_for_each(s, id) {
528 			ret = bch2_trans_run(c, bch2_subvolume_delete(trans, *id));
529 			bch_err_msg(c, ret, "deleting subvolume %u", *id);
530 			if (ret)
531 				break;
532 		}
533 
534 		darray_exit(&s);
535 	}
536 
537 	enumerated_ref_put(&c->writes, BCH_WRITE_REF_snapshot_delete_pagecache);
538 }
539 
540 struct subvolume_unlink_hook {
541 	struct btree_trans_commit_hook	h;
542 	u32				subvol;
543 };
544 
bch2_subvolume_wait_for_pagecache_and_delete_hook(struct btree_trans * trans,struct btree_trans_commit_hook * _h)545 static int bch2_subvolume_wait_for_pagecache_and_delete_hook(struct btree_trans *trans,
546 						      struct btree_trans_commit_hook *_h)
547 {
548 	struct subvolume_unlink_hook *h = container_of(_h, struct subvolume_unlink_hook, h);
549 	struct bch_fs *c = trans->c;
550 	int ret = 0;
551 
552 	mutex_lock(&c->snapshots_unlinked_lock);
553 	if (!snapshot_list_has_id(&c->snapshots_unlinked, h->subvol))
554 		ret = snapshot_list_add(c, &c->snapshots_unlinked, h->subvol);
555 	mutex_unlock(&c->snapshots_unlinked_lock);
556 
557 	if (ret)
558 		return ret;
559 
560 	if (!enumerated_ref_tryget(&c->writes, BCH_WRITE_REF_snapshot_delete_pagecache))
561 		return -EROFS;
562 
563 	if (!queue_work(c->write_ref_wq, &c->snapshot_wait_for_pagecache_and_delete_work))
564 		enumerated_ref_put(&c->writes, BCH_WRITE_REF_snapshot_delete_pagecache);
565 	return 0;
566 }
567 
bch2_subvolume_unlink(struct btree_trans * trans,u32 subvolid)568 int bch2_subvolume_unlink(struct btree_trans *trans, u32 subvolid)
569 {
570 	struct btree_iter iter;
571 	struct bkey_i_subvolume *n;
572 	struct subvolume_unlink_hook *h;
573 	int ret = 0;
574 
575 	h = bch2_trans_kmalloc(trans, sizeof(*h));
576 	ret = PTR_ERR_OR_ZERO(h);
577 	if (ret)
578 		return ret;
579 
580 	h->h.fn		= bch2_subvolume_wait_for_pagecache_and_delete_hook;
581 	h->subvol	= subvolid;
582 	bch2_trans_commit_hook(trans, &h->h);
583 
584 	n = bch2_bkey_get_mut_typed(trans, &iter,
585 			BTREE_ID_subvolumes, POS(0, subvolid),
586 			BTREE_ITER_cached, subvolume);
587 	ret = PTR_ERR_OR_ZERO(n);
588 	if (bch2_err_matches(ret, ENOENT))
589 		ret = bch2_subvolume_missing(trans->c, subvolid) ?: ret;
590 	if (unlikely(ret))
591 		return ret;
592 
593 	SET_BCH_SUBVOLUME_UNLINKED(&n->v, true);
594 	n->v.fs_path_parent = 0;
595 	bch2_trans_iter_exit(trans, &iter);
596 	return ret;
597 }
598 
bch2_subvolume_create(struct btree_trans * trans,u64 inode,u32 parent_subvolid,u32 src_subvolid,u32 * new_subvolid,u32 * new_snapshotid,bool ro)599 int bch2_subvolume_create(struct btree_trans *trans, u64 inode,
600 			  u32 parent_subvolid,
601 			  u32 src_subvolid,
602 			  u32 *new_subvolid,
603 			  u32 *new_snapshotid,
604 			  bool ro)
605 {
606 	struct bch_fs *c = trans->c;
607 	struct btree_iter dst_iter, src_iter = {};
608 	struct bkey_i_subvolume *new_subvol = NULL;
609 	struct bkey_i_subvolume *src_subvol = NULL;
610 	u32 parent = 0, new_nodes[2], snapshot_subvols[2];
611 	int ret = 0;
612 
613 	ret = bch2_bkey_get_empty_slot(trans, &dst_iter,
614 				BTREE_ID_subvolumes, POS(0, U32_MAX));
615 	if (ret == -BCH_ERR_ENOSPC_btree_slot)
616 		ret = bch_err_throw(c, ENOSPC_subvolume_create);
617 	if (ret)
618 		return ret;
619 
620 	snapshot_subvols[0] = dst_iter.pos.offset;
621 	snapshot_subvols[1] = src_subvolid;
622 
623 	if (src_subvolid) {
624 		/* Creating a snapshot: */
625 
626 		src_subvol = bch2_bkey_get_mut_typed(trans, &src_iter,
627 				BTREE_ID_subvolumes, POS(0, src_subvolid),
628 				BTREE_ITER_cached, subvolume);
629 		ret = PTR_ERR_OR_ZERO(src_subvol);
630 		if (bch2_err_matches(ret, ENOENT))
631 			ret = bch2_subvolume_missing(trans->c, src_subvolid) ?: ret;
632 		if (unlikely(ret))
633 			goto err;
634 
635 		parent = le32_to_cpu(src_subvol->v.snapshot);
636 	}
637 
638 	ret = bch2_snapshot_node_create(trans, parent, new_nodes,
639 					snapshot_subvols,
640 					src_subvolid ? 2 : 1);
641 	if (ret)
642 		goto err;
643 
644 	if (src_subvolid) {
645 		src_subvol->v.snapshot = cpu_to_le32(new_nodes[1]);
646 		ret = bch2_trans_update(trans, &src_iter, &src_subvol->k_i, 0);
647 		if (ret)
648 			goto err;
649 	}
650 
651 	new_subvol = bch2_bkey_alloc(trans, &dst_iter, 0, subvolume);
652 	ret = PTR_ERR_OR_ZERO(new_subvol);
653 	if (ret)
654 		goto err;
655 
656 	new_subvol->v.flags		= 0;
657 	new_subvol->v.snapshot		= cpu_to_le32(new_nodes[0]);
658 	new_subvol->v.inode		= cpu_to_le64(inode);
659 	new_subvol->v.creation_parent	= cpu_to_le32(src_subvolid);
660 	new_subvol->v.fs_path_parent	= cpu_to_le32(parent_subvolid);
661 	new_subvol->v.otime.lo		= cpu_to_le64(bch2_current_time(c));
662 	new_subvol->v.otime.hi		= 0;
663 
664 	SET_BCH_SUBVOLUME_RO(&new_subvol->v, ro);
665 	SET_BCH_SUBVOLUME_SNAP(&new_subvol->v, src_subvolid != 0);
666 
667 	*new_subvolid	= new_subvol->k.p.offset;
668 	*new_snapshotid	= new_nodes[0];
669 err:
670 	bch2_trans_iter_exit(trans, &src_iter);
671 	bch2_trans_iter_exit(trans, &dst_iter);
672 	return ret;
673 }
674 
bch2_initialize_subvolumes(struct bch_fs * c)675 int bch2_initialize_subvolumes(struct bch_fs *c)
676 {
677 	struct bkey_i_snapshot_tree	root_tree;
678 	struct bkey_i_snapshot		root_snapshot;
679 	struct bkey_i_subvolume		root_volume;
680 	int ret;
681 
682 	bkey_snapshot_tree_init(&root_tree.k_i);
683 	root_tree.k.p.offset		= 1;
684 	root_tree.v.master_subvol	= cpu_to_le32(1);
685 	root_tree.v.root_snapshot	= cpu_to_le32(U32_MAX);
686 
687 	bkey_snapshot_init(&root_snapshot.k_i);
688 	root_snapshot.k.p.offset = U32_MAX;
689 	root_snapshot.v.flags	= 0;
690 	root_snapshot.v.parent	= 0;
691 	root_snapshot.v.subvol	= cpu_to_le32(BCACHEFS_ROOT_SUBVOL);
692 	root_snapshot.v.tree	= cpu_to_le32(1);
693 	SET_BCH_SNAPSHOT_SUBVOL(&root_snapshot.v, true);
694 
695 	bkey_subvolume_init(&root_volume.k_i);
696 	root_volume.k.p.offset = BCACHEFS_ROOT_SUBVOL;
697 	root_volume.v.flags	= 0;
698 	root_volume.v.snapshot	= cpu_to_le32(U32_MAX);
699 	root_volume.v.inode	= cpu_to_le64(BCACHEFS_ROOT_INO);
700 
701 	ret =   bch2_btree_insert(c, BTREE_ID_snapshot_trees,	&root_tree.k_i, NULL, 0, 0) ?:
702 		bch2_btree_insert(c, BTREE_ID_snapshots,	&root_snapshot.k_i, NULL, 0, 0) ?:
703 		bch2_btree_insert(c, BTREE_ID_subvolumes,	&root_volume.k_i, NULL, 0, 0);
704 	bch_err_fn(c, ret);
705 	return ret;
706 }
707 
__bch2_fs_upgrade_for_subvolumes(struct btree_trans * trans)708 static int __bch2_fs_upgrade_for_subvolumes(struct btree_trans *trans)
709 {
710 	struct btree_iter iter;
711 	struct bkey_s_c k;
712 	struct bch_inode_unpacked inode;
713 	int ret;
714 
715 	k = bch2_bkey_get_iter(trans, &iter, BTREE_ID_inodes,
716 			       SPOS(0, BCACHEFS_ROOT_INO, U32_MAX), 0);
717 	ret = bkey_err(k);
718 	if (ret)
719 		return ret;
720 
721 	if (!bkey_is_inode(k.k)) {
722 		struct bch_fs *c = trans->c;
723 		bch_err(c, "root inode not found");
724 		ret = bch_err_throw(c, ENOENT_inode);
725 		goto err;
726 	}
727 
728 	ret = bch2_inode_unpack(k, &inode);
729 	BUG_ON(ret);
730 
731 	inode.bi_subvol = BCACHEFS_ROOT_SUBVOL;
732 
733 	ret = bch2_inode_write(trans, &iter, &inode);
734 err:
735 	bch2_trans_iter_exit(trans, &iter);
736 	return ret;
737 }
738 
739 /* set bi_subvol on root inode */
bch2_fs_upgrade_for_subvolumes(struct bch_fs * c)740 int bch2_fs_upgrade_for_subvolumes(struct bch_fs *c)
741 {
742 	int ret = bch2_trans_commit_do(c, NULL, NULL, BCH_TRANS_COMMIT_no_enospc,
743 				       __bch2_fs_upgrade_for_subvolumes(trans));
744 	bch_err_fn(c, ret);
745 	return ret;
746 }
747 
bch2_fs_subvolumes_init_early(struct bch_fs * c)748 void bch2_fs_subvolumes_init_early(struct bch_fs *c)
749 {
750 	INIT_WORK(&c->snapshot_wait_for_pagecache_and_delete_work,
751 		  bch2_subvolume_wait_for_pagecache_and_delete);
752 }
753