]> git.sesse.net Git - bcachefs-tools-debian/blob - libbcachefs/subvolume.c
5b145d850b24083a0b1a8af201d7f1d4099da05e
[bcachefs-tools-debian] / libbcachefs / subvolume.c
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 "subvolume.h"
10
11 static int bch2_subvolume_delete(struct btree_trans *, u32);
12
13 /* Snapshot tree: */
14
15 void bch2_snapshot_tree_to_text(struct printbuf *out, struct bch_fs *c,
16                                 struct bkey_s_c k)
17 {
18         struct bkey_s_c_snapshot_tree t = bkey_s_c_to_snapshot_tree(k);
19
20         prt_printf(out, "subvol %u root snapshot %u",
21                    le32_to_cpu(t.v->master_subvol),
22                    le32_to_cpu(t.v->root_snapshot));
23 }
24
25 int bch2_snapshot_tree_invalid(const struct bch_fs *c, struct bkey_s_c k,
26                                unsigned flags, struct printbuf *err)
27 {
28         if (bkey_gt(k.k->p, POS(0, U32_MAX)) ||
29             bkey_lt(k.k->p, POS(0, 1))) {
30                 prt_printf(err, "bad pos");
31                 return -BCH_ERR_invalid_bkey;
32         }
33
34         return 0;
35 }
36
37 int bch2_snapshot_tree_lookup(struct btree_trans *trans, u32 id,
38                               struct bch_snapshot_tree *s)
39 {
40         int ret;
41
42         ret = bch2_bkey_get_val_typed(trans, BTREE_ID_snapshot_trees, POS(0, id),
43                                       BTREE_ITER_WITH_UPDATES, snapshot_tree, s);
44
45         if (bch2_err_matches(ret, ENOENT))
46                 bch_err(trans->c, "snapshot tree %u not found", id);
47         return ret;
48 }
49
50 static struct bkey_i_snapshot_tree *
51 __snapshot_tree_create(struct btree_trans *trans)
52 {
53         struct btree_iter iter;
54         int ret = bch2_bkey_get_empty_slot(trans, &iter,
55                         BTREE_ID_snapshot_trees, POS(0, U32_MAX));
56         struct bkey_i_snapshot_tree *s_t;
57
58         if (ret == -BCH_ERR_ENOSPC_btree_slot)
59                 ret = -BCH_ERR_ENOSPC_snapshot_tree;
60         if (ret)
61                 return ERR_PTR(ret);
62
63         s_t = bch2_bkey_alloc(trans, &iter, 0, snapshot_tree);
64         ret = PTR_ERR_OR_ZERO(s_t);
65         bch2_trans_iter_exit(trans, &iter);
66         return ret ? ERR_PTR(ret) : s_t;
67 }
68
69 static int snapshot_tree_create(struct btree_trans *trans,
70                                 u32 root_id, u32 subvol_id, u32 *tree_id)
71 {
72         struct bkey_i_snapshot_tree *n_tree =
73                 __snapshot_tree_create(trans);
74
75         if (IS_ERR(n_tree))
76                 return PTR_ERR(n_tree);
77
78         n_tree->v.master_subvol = cpu_to_le32(subvol_id);
79         n_tree->v.root_snapshot = cpu_to_le32(root_id);
80         *tree_id = n_tree->k.p.offset;
81         return 0;
82 }
83
84 /* Snapshot nodes: */
85
86 void bch2_snapshot_to_text(struct printbuf *out, struct bch_fs *c,
87                            struct bkey_s_c k)
88 {
89         struct bkey_s_c_snapshot s = bkey_s_c_to_snapshot(k);
90
91         prt_printf(out, "is_subvol %llu deleted %llu parent %10u children %10u %10u subvol %u",
92                BCH_SNAPSHOT_SUBVOL(s.v),
93                BCH_SNAPSHOT_DELETED(s.v),
94                le32_to_cpu(s.v->parent),
95                le32_to_cpu(s.v->children[0]),
96                le32_to_cpu(s.v->children[1]),
97                le32_to_cpu(s.v->subvol));
98 }
99
100 int bch2_snapshot_invalid(const struct bch_fs *c, struct bkey_s_c k,
101                           unsigned flags, struct printbuf *err)
102 {
103         struct bkey_s_c_snapshot s;
104         u32 i, id;
105
106         if (bkey_gt(k.k->p, POS(0, U32_MAX)) ||
107             bkey_lt(k.k->p, POS(0, 1))) {
108                 prt_printf(err, "bad pos");
109                 return -BCH_ERR_invalid_bkey;
110         }
111
112         s = bkey_s_c_to_snapshot(k);
113
114         id = le32_to_cpu(s.v->parent);
115         if (id && id <= k.k->p.offset) {
116                 prt_printf(err, "bad parent node (%u <= %llu)",
117                        id, k.k->p.offset);
118                 return -BCH_ERR_invalid_bkey;
119         }
120
121         if (le32_to_cpu(s.v->children[0]) < le32_to_cpu(s.v->children[1])) {
122                 prt_printf(err, "children not normalized");
123                 return -BCH_ERR_invalid_bkey;
124         }
125
126         if (s.v->children[0] &&
127             s.v->children[0] == s.v->children[1]) {
128                 prt_printf(err, "duplicate child nodes");
129                 return -BCH_ERR_invalid_bkey;
130         }
131
132         for (i = 0; i < 2; i++) {
133                 id = le32_to_cpu(s.v->children[i]);
134
135                 if (id >= k.k->p.offset) {
136                         prt_printf(err, "bad child node (%u >= %llu)",
137                                id, k.k->p.offset);
138                         return -BCH_ERR_invalid_bkey;
139                 }
140         }
141
142         return 0;
143 }
144
145 int bch2_mark_snapshot(struct btree_trans *trans,
146                        enum btree_id btree, unsigned level,
147                        struct bkey_s_c old, struct bkey_s_c new,
148                        unsigned flags)
149 {
150         struct bch_fs *c = trans->c;
151         struct snapshot_t *t;
152
153         t = genradix_ptr_alloc(&c->snapshots,
154                                U32_MAX - new.k->p.offset,
155                                GFP_KERNEL);
156         if (!t)
157                 return -BCH_ERR_ENOMEM_mark_snapshot;
158
159         if (new.k->type == KEY_TYPE_snapshot) {
160                 struct bkey_s_c_snapshot s = bkey_s_c_to_snapshot(new);
161
162                 t->parent       = le32_to_cpu(s.v->parent);
163                 t->children[0]  = le32_to_cpu(s.v->children[0]);
164                 t->children[1]  = le32_to_cpu(s.v->children[1]);
165                 t->subvol       = BCH_SNAPSHOT_SUBVOL(s.v) ? le32_to_cpu(s.v->subvol) : 0;
166                 t->tree         = le32_to_cpu(s.v->tree);
167         } else {
168                 t->parent       = 0;
169                 t->children[0]  = 0;
170                 t->children[1]  = 0;
171                 t->subvol       = 0;
172                 t->tree         = 0;
173         }
174
175         return 0;
176 }
177
178 static int snapshot_lookup(struct btree_trans *trans, u32 id,
179                            struct bch_snapshot *s)
180 {
181         return bch2_bkey_get_val_typed(trans, BTREE_ID_snapshots, POS(0, id),
182                                        BTREE_ITER_WITH_UPDATES, snapshot, s);
183 }
184
185 static int snapshot_live(struct btree_trans *trans, u32 id)
186 {
187         struct bch_snapshot v;
188         int ret;
189
190         if (!id)
191                 return 0;
192
193         ret = snapshot_lookup(trans, id, &v);
194         if (bch2_err_matches(ret, ENOENT))
195                 bch_err(trans->c, "snapshot node %u not found", id);
196         if (ret)
197                 return ret;
198
199         return !BCH_SNAPSHOT_DELETED(&v);
200 }
201
202 static int bch2_snapshot_set_equiv(struct btree_trans *trans, struct bkey_s_c k)
203 {
204         struct bch_fs *c = trans->c;
205         unsigned i, nr_live = 0, live_idx = 0;
206         struct bkey_s_c_snapshot snap;
207         u32 id = k.k->p.offset, child[2];
208
209         if (k.k->type != KEY_TYPE_snapshot)
210                 return 0;
211
212         snap = bkey_s_c_to_snapshot(k);
213
214         child[0] = le32_to_cpu(snap.v->children[0]);
215         child[1] = le32_to_cpu(snap.v->children[1]);
216
217         for (i = 0; i < 2; i++) {
218                 int ret = snapshot_live(trans, child[i]);
219
220                 if (ret < 0)
221                         return ret;
222
223                 if (ret)
224                         live_idx = i;
225                 nr_live += ret;
226         }
227
228         snapshot_t(c, id)->equiv = nr_live == 1
229                 ? snapshot_t(c, child[live_idx])->equiv
230                 : id;
231         return 0;
232 }
233
234 /* fsck: */
235
236 static u32 bch2_snapshot_child(struct bch_fs *c, u32 id, unsigned child)
237 {
238         return snapshot_t(c, id)->children[child];
239 }
240
241 static u32 bch2_snapshot_left_child(struct bch_fs *c, u32 id)
242 {
243         return bch2_snapshot_child(c, id, 0);
244 }
245
246 static u32 bch2_snapshot_right_child(struct bch_fs *c, u32 id)
247 {
248         return bch2_snapshot_child(c, id, 1);
249 }
250
251 static u32 bch2_snapshot_tree_next(struct bch_fs *c, u32 id)
252 {
253         u32 n, parent;
254
255         n = bch2_snapshot_left_child(c, id);
256         if (n)
257                 return n;
258
259         while ((parent = bch2_snapshot_parent(c, id))) {
260                 n = bch2_snapshot_right_child(c, parent);
261                 if (n && n != id)
262                         return n;
263                 id = parent;
264         }
265
266         return 0;
267 }
268
269 static u32 bch2_snapshot_tree_oldest_subvol(struct bch_fs *c, u32 snapshot_root)
270 {
271         u32 id = snapshot_root;
272         u32 subvol = 0, s;
273
274         while (id) {
275                 s = snapshot_t(c, id)->subvol;
276
277                 if (s && (!subvol || s < subvol))
278                         subvol = s;
279
280                 id = bch2_snapshot_tree_next(c, id);
281         }
282
283         return subvol;
284 }
285
286 static int bch2_snapshot_tree_master_subvol(struct btree_trans *trans,
287                                             u32 snapshot_root, u32 *subvol_id)
288 {
289         struct bch_fs *c = trans->c;
290         struct btree_iter iter;
291         struct bkey_s_c k;
292         struct bkey_s_c_subvolume s;
293         int ret;
294
295         for_each_btree_key_norestart(trans, iter, BTREE_ID_subvolumes, POS_MIN,
296                                      0, k, ret) {
297                 if (k.k->type != KEY_TYPE_subvolume)
298                         continue;
299
300                 s = bkey_s_c_to_subvolume(k);
301                 if (!bch2_snapshot_is_ancestor(c, le32_to_cpu(s.v->snapshot), snapshot_root))
302                         continue;
303                 if (!BCH_SUBVOLUME_SNAP(s.v)) {
304                         *subvol_id = s.k->p.offset;
305                         goto found;
306                 }
307         }
308         ret = ret ?: -ENOENT;
309 found:
310         bch2_trans_iter_exit(trans, &iter);
311
312         if (bch2_err_matches(ret, ENOENT)) {
313                 struct bkey_i_subvolume *s;
314
315                 *subvol_id = bch2_snapshot_tree_oldest_subvol(c, snapshot_root);
316
317                 s = bch2_bkey_get_mut_typed(trans, &iter,
318                                             BTREE_ID_subvolumes, POS(0, *subvol_id),
319                                             0, subvolume);
320                 ret = PTR_ERR_OR_ZERO(s);
321                 if (ret)
322                         return ret;
323
324                 SET_BCH_SUBVOLUME_SNAP(&s->v, false);
325         }
326
327         return ret;
328 }
329
330 static int check_snapshot_tree(struct btree_trans *trans,
331                                struct btree_iter *iter,
332                                struct bkey_s_c k)
333 {
334         struct bch_fs *c = trans->c;
335         struct bkey_s_c_snapshot_tree st;
336         struct bch_snapshot s;
337         struct bch_subvolume subvol;
338         struct printbuf buf = PRINTBUF;
339         u32 root_id;
340         int ret;
341
342         if (k.k->type != KEY_TYPE_snapshot_tree)
343                 return 0;
344
345         st = bkey_s_c_to_snapshot_tree(k);
346         root_id = le32_to_cpu(st.v->root_snapshot);
347
348         ret = snapshot_lookup(trans, root_id, &s);
349         if (ret && !bch2_err_matches(ret, ENOENT))
350                 goto err;
351
352         if (fsck_err_on(ret ||
353                         root_id != bch2_snapshot_root(c, root_id) ||
354                         st.k->p.offset != le32_to_cpu(s.tree),
355                         c,
356                         "snapshot tree points to missing/incorrect snapshot:\n  %s",
357                         (bch2_bkey_val_to_text(&buf, c, st.s_c), buf.buf))) {
358                 ret = bch2_btree_delete_at(trans, iter, 0);
359                 goto err;
360         }
361
362         ret = bch2_subvolume_get(trans, le32_to_cpu(st.v->master_subvol),
363                                  false, 0, &subvol);
364         if (ret && !bch2_err_matches(ret, ENOENT))
365                 goto err;
366
367         if (fsck_err_on(ret, c,
368                         "snapshot tree points to missing subvolume:\n  %s",
369                         (printbuf_reset(&buf),
370                          bch2_bkey_val_to_text(&buf, c, st.s_c), buf.buf)) ||
371             fsck_err_on(!bch2_snapshot_is_ancestor(c,
372                                                    le32_to_cpu(subvol.snapshot),
373                                                    root_id), c,
374                         "snapshot tree points to subvolume that does not point to snapshot in this tree:\n  %s",
375                         (printbuf_reset(&buf),
376                          bch2_bkey_val_to_text(&buf, c, st.s_c), buf.buf)) ||
377             fsck_err_on(BCH_SUBVOLUME_SNAP(&subvol), c,
378                         "snapshot tree points to snapshot subvolume:\n  %s",
379                         (printbuf_reset(&buf),
380                          bch2_bkey_val_to_text(&buf, c, st.s_c), buf.buf))) {
381                 struct bkey_i_snapshot_tree *u;
382                 u32 subvol_id;
383
384                 ret = bch2_snapshot_tree_master_subvol(trans, root_id, &subvol_id);
385                 if (ret)
386                         goto err;
387
388                 u = bch2_bkey_make_mut_typed(trans, iter, k, 0, snapshot_tree);
389                 ret = PTR_ERR_OR_ZERO(u);
390                 if (ret)
391                         goto err;
392
393                 u->v.master_subvol = cpu_to_le32(subvol_id);
394                 st = snapshot_tree_i_to_s_c(u);
395         }
396 err:
397 fsck_err:
398         printbuf_exit(&buf);
399         return ret;
400 }
401
402 /*
403  * For each snapshot_tree, make sure it points to the root of a snapshot tree
404  * and that snapshot entry points back to it, or delete it.
405  *
406  * And, make sure it points to a subvolume within that snapshot tree, or correct
407  * it to point to the oldest subvolume within that snapshot tree.
408  */
409 int bch2_fs_check_snapshot_trees(struct bch_fs *c)
410 {
411         struct btree_iter iter;
412         struct bkey_s_c k;
413         int ret;
414
415         ret = bch2_trans_run(c,
416                 for_each_btree_key_commit(&trans, iter,
417                         BTREE_ID_snapshot_trees, POS_MIN,
418                         BTREE_ITER_PREFETCH, k,
419                         NULL, NULL, BTREE_INSERT_LAZY_RW|BTREE_INSERT_NOFAIL,
420                 check_snapshot_tree(&trans, &iter, k)));
421
422         if (ret)
423                 bch_err(c, "error %i checking snapshot trees", ret);
424         return ret;
425 }
426
427 /*
428  * Look up snapshot tree for @tree_id and find root,
429  * make sure @snap_id is a descendent:
430  */
431 static int snapshot_tree_ptr_good(struct btree_trans *trans,
432                                   u32 snap_id, u32 tree_id)
433 {
434         struct bch_snapshot_tree s_t;
435         int ret = bch2_snapshot_tree_lookup(trans, tree_id, &s_t);
436
437         if (ret)
438                 return ret;
439
440         return bch2_snapshot_is_ancestor(trans->c, snap_id, le32_to_cpu(s_t.root_snapshot));
441 }
442
443 /*
444  * snapshot_tree pointer was incorrect: look up root snapshot node, make sure
445  * its snapshot_tree pointer is correct (allocate new one if necessary), then
446  * update this node's pointer to root node's pointer:
447  */
448 static int snapshot_tree_ptr_repair(struct btree_trans *trans,
449                                     struct btree_iter *iter,
450                                     struct bkey_s_c_snapshot *s)
451 {
452         struct bch_fs *c = trans->c;
453         struct btree_iter root_iter;
454         struct bch_snapshot_tree s_t;
455         struct bkey_s_c_snapshot root;
456         struct bkey_i_snapshot *u;
457         u32 root_id = bch2_snapshot_root(c, s->k->p.offset), tree_id;
458         int ret;
459
460         root = bch2_bkey_get_iter_typed(trans, &root_iter,
461                                BTREE_ID_snapshots, POS(0, root_id),
462                                BTREE_ITER_WITH_UPDATES, snapshot);
463         ret = bkey_err(root);
464         if (ret)
465                 goto err;
466
467         tree_id = le32_to_cpu(root.v->tree);
468
469         ret = bch2_snapshot_tree_lookup(trans, tree_id, &s_t);
470         if (ret)
471                 return ret;
472
473         if (le32_to_cpu(s_t.root_snapshot) != root_id) {
474                 u = bch2_bkey_make_mut_typed(trans, &root_iter, root.s_c, 0, snapshot);
475                 ret =   PTR_ERR_OR_ZERO(u) ?:
476                         snapshot_tree_create(trans, root_id,
477                                 bch2_snapshot_tree_oldest_subvol(c, root_id),
478                                 &tree_id);
479                 if (ret)
480                         goto err;
481
482                 u->v.tree = cpu_to_le32(tree_id);
483                 if (s->k->p.snapshot == root_id)
484                         *s = snapshot_i_to_s_c(u);
485         }
486
487         if (s->k->p.snapshot != root_id) {
488                 u = bch2_bkey_make_mut_typed(trans, iter, s->s_c, 0, snapshot);
489                 ret = PTR_ERR_OR_ZERO(u);
490                 if (ret)
491                         goto err;
492
493                 u->v.tree = cpu_to_le32(tree_id);
494                 *s = snapshot_i_to_s_c(u);
495         }
496 err:
497         bch2_trans_iter_exit(trans, &root_iter);
498         return ret;
499 }
500
501 static int check_snapshot(struct btree_trans *trans,
502                           struct btree_iter *iter,
503                           struct bkey_s_c k)
504 {
505         struct bch_fs *c = trans->c;
506         struct bkey_s_c_snapshot s;
507         struct bch_subvolume subvol;
508         struct bch_snapshot v;
509         struct printbuf buf = PRINTBUF;
510         bool should_have_subvol;
511         u32 i, id;
512         int ret = 0;
513
514         if (k.k->type != KEY_TYPE_snapshot)
515                 return 0;
516
517         s = bkey_s_c_to_snapshot(k);
518         id = le32_to_cpu(s.v->parent);
519         if (id) {
520                 ret = snapshot_lookup(trans, id, &v);
521                 if (bch2_err_matches(ret, ENOENT))
522                         bch_err(c, "snapshot with nonexistent parent:\n  %s",
523                                 (bch2_bkey_val_to_text(&buf, c, s.s_c), buf.buf));
524                 if (ret)
525                         goto err;
526
527                 if (le32_to_cpu(v.children[0]) != s.k->p.offset &&
528                     le32_to_cpu(v.children[1]) != s.k->p.offset) {
529                         bch_err(c, "snapshot parent %u missing pointer to child %llu",
530                                 id, s.k->p.offset);
531                         ret = -EINVAL;
532                         goto err;
533                 }
534         }
535
536         for (i = 0; i < 2 && s.v->children[i]; i++) {
537                 id = le32_to_cpu(s.v->children[i]);
538
539                 ret = snapshot_lookup(trans, id, &v);
540                 if (bch2_err_matches(ret, ENOENT))
541                         bch_err(c, "snapshot node %llu has nonexistent child %u",
542                                 s.k->p.offset, id);
543                 if (ret)
544                         goto err;
545
546                 if (le32_to_cpu(v.parent) != s.k->p.offset) {
547                         bch_err(c, "snapshot child %u has wrong parent (got %u should be %llu)",
548                                 id, le32_to_cpu(v.parent), s.k->p.offset);
549                         ret = -EINVAL;
550                         goto err;
551                 }
552         }
553
554         should_have_subvol = BCH_SNAPSHOT_SUBVOL(s.v) &&
555                 !BCH_SNAPSHOT_DELETED(s.v);
556
557         if (should_have_subvol) {
558                 id = le32_to_cpu(s.v->subvol);
559                 ret = bch2_subvolume_get(trans, id, 0, false, &subvol);
560                 if (bch2_err_matches(ret, ENOENT))
561                         bch_err(c, "snapshot points to nonexistent subvolume:\n  %s",
562                                 (bch2_bkey_val_to_text(&buf, c, s.s_c), buf.buf));
563                 if (ret)
564                         goto err;
565
566                 if (BCH_SNAPSHOT_SUBVOL(s.v) != (le32_to_cpu(subvol.snapshot) == s.k->p.offset)) {
567                         bch_err(c, "snapshot node %llu has wrong BCH_SNAPSHOT_SUBVOL",
568                                 s.k->p.offset);
569                         ret = -EINVAL;
570                         goto err;
571                 }
572         } else {
573                 if (fsck_err_on(s.v->subvol, c, "snapshot should not point to subvol:\n  %s",
574                                 (bch2_bkey_val_to_text(&buf, c, s.s_c), buf.buf))) {
575                         struct bkey_i_snapshot *u = bch2_trans_kmalloc(trans, sizeof(*u));
576
577                         ret = PTR_ERR_OR_ZERO(u);
578                         if (ret)
579                                 goto err;
580
581                         bkey_reassemble(&u->k_i, s.s_c);
582                         u->v.subvol = 0;
583                         ret = bch2_trans_update(trans, iter, &u->k_i, 0);
584                         if (ret)
585                                 goto err;
586
587                         s = snapshot_i_to_s_c(u);
588                 }
589         }
590
591         ret = snapshot_tree_ptr_good(trans, s.k->p.offset, le32_to_cpu(s.v->tree));
592         if (ret < 0)
593                 goto err;
594
595         if (fsck_err_on(!ret, c, "snapshot points to missing/incorrect tree:\n  %s",
596                         (bch2_bkey_val_to_text(&buf, c, s.s_c), buf.buf))) {
597                 ret = snapshot_tree_ptr_repair(trans, iter, &s);
598                 if (ret)
599                         goto err;
600         }
601         ret = 0;
602
603         if (BCH_SNAPSHOT_DELETED(s.v))
604                 set_bit(BCH_FS_HAVE_DELETED_SNAPSHOTS, &c->flags);
605 err:
606 fsck_err:
607         printbuf_exit(&buf);
608         return ret;
609 }
610
611 int bch2_fs_check_snapshots(struct bch_fs *c)
612 {
613         struct btree_iter iter;
614         struct bkey_s_c k;
615         int ret;
616
617         ret = bch2_trans_run(c,
618                 for_each_btree_key_commit(&trans, iter,
619                         BTREE_ID_snapshots, POS_MIN,
620                         BTREE_ITER_PREFETCH, k,
621                         NULL, NULL, BTREE_INSERT_LAZY_RW|BTREE_INSERT_NOFAIL,
622                 check_snapshot(&trans, &iter, k)));
623         if (ret)
624                 bch_err(c, "%s: error %s", __func__, bch2_err_str(ret));
625         return ret;
626 }
627
628 static int check_subvol(struct btree_trans *trans,
629                         struct btree_iter *iter,
630                         struct bkey_s_c k)
631 {
632         struct bch_fs *c = trans->c;
633         struct bkey_s_c_subvolume subvol;
634         struct bch_snapshot snapshot;
635         unsigned snapid;
636         int ret = 0;
637
638         if (k.k->type != KEY_TYPE_subvolume)
639                 return 0;
640
641         subvol = bkey_s_c_to_subvolume(k);
642         snapid = le32_to_cpu(subvol.v->snapshot);
643         ret = snapshot_lookup(trans, snapid, &snapshot);
644
645         if (bch2_err_matches(ret, ENOENT))
646                 bch_err(c, "subvolume %llu points to nonexistent snapshot %u",
647                         k.k->p.offset, snapid);
648         if (ret)
649                 return ret;
650
651         if (BCH_SUBVOLUME_UNLINKED(subvol.v)) {
652                 bch2_fs_lazy_rw(c);
653
654                 ret = bch2_subvolume_delete(trans, iter->pos.offset);
655                 if (ret)
656                         bch_err(c, "error deleting subvolume %llu: %s",
657                                 iter->pos.offset, bch2_err_str(ret));
658                 return ret ?: -BCH_ERR_transaction_restart_nested;
659         }
660
661         if (!BCH_SUBVOLUME_SNAP(subvol.v)) {
662                 u32 snapshot_root = bch2_snapshot_root(c, le32_to_cpu(subvol.v->snapshot));
663                 u32 snapshot_tree = snapshot_t(c, snapshot_root)->tree;
664                 struct bch_snapshot_tree st;
665
666                 ret = bch2_snapshot_tree_lookup(trans, snapshot_tree, &st);
667                 if (ret)
668                         return ret;
669
670                 if (fsck_err_on(le32_to_cpu(st.master_subvol) != subvol.k->p.offset, c,
671                                 "subvolume %llu is not set as snapshot but is not master subvolume",
672                                 k.k->p.offset)) {
673                         struct bkey_i_subvolume *s =
674                                 bch2_bkey_make_mut_typed(trans, iter, subvol.s_c, 0, subvolume);
675                         ret = PTR_ERR_OR_ZERO(s);
676                         if (ret)
677                                 return ret;
678
679                         SET_BCH_SUBVOLUME_SNAP(&s->v, true);
680                 }
681         }
682
683 fsck_err:
684         return ret;
685 }
686
687 int bch2_fs_check_subvols(struct bch_fs *c)
688 {
689         struct btree_iter iter;
690         struct bkey_s_c k;
691         int ret;
692
693         ret = bch2_trans_run(c,
694                 for_each_btree_key_commit(&trans, iter,
695                         BTREE_ID_subvolumes, POS_MIN, BTREE_ITER_PREFETCH, k,
696                         NULL, NULL, BTREE_INSERT_LAZY_RW|BTREE_INSERT_NOFAIL,
697                 check_subvol(&trans, &iter, k)));
698         if (ret)
699                 bch_err(c, "%s: error %s", __func__, bch2_err_str(ret));
700
701         return ret;
702 }
703
704 void bch2_fs_snapshots_exit(struct bch_fs *c)
705 {
706         genradix_free(&c->snapshots);
707 }
708
709 int bch2_fs_snapshots_start(struct bch_fs *c)
710 {
711         struct btree_iter iter;
712         struct bkey_s_c k;
713         int ret = 0;
714
715         ret = bch2_trans_run(c,
716                 for_each_btree_key2(&trans, iter, BTREE_ID_snapshots,
717                            POS_MIN, 0, k,
718                         bch2_mark_snapshot(&trans, BTREE_ID_snapshots, 0, bkey_s_c_null, k, 0) ?:
719                         bch2_snapshot_set_equiv(&trans, k)));
720         if (ret)
721                 bch_err(c, "error starting snapshots: %s", bch2_err_str(ret));
722         return ret;
723 }
724
725 /*
726  * Mark a snapshot as deleted, for future cleanup:
727  */
728 static int bch2_snapshot_node_set_deleted(struct btree_trans *trans, u32 id)
729 {
730         struct btree_iter iter;
731         struct bkey_i_snapshot *s;
732         int ret = 0;
733
734         s = bch2_bkey_get_mut_typed(trans, &iter,
735                                     BTREE_ID_snapshots, POS(0, id),
736                                     0, snapshot);
737         ret = PTR_ERR_OR_ZERO(s);
738         if (unlikely(ret)) {
739                 bch2_fs_inconsistent_on(bch2_err_matches(ret, ENOENT),
740                                         trans->c, "missing snapshot %u", id);
741                 return ret;
742         }
743
744         /* already deleted? */
745         if (BCH_SNAPSHOT_DELETED(&s->v))
746                 goto err;
747
748         SET_BCH_SNAPSHOT_DELETED(&s->v, true);
749         SET_BCH_SNAPSHOT_SUBVOL(&s->v, false);
750         s->v.subvol = 0;
751 err:
752         bch2_trans_iter_exit(trans, &iter);
753         return ret;
754 }
755
756 static int bch2_snapshot_node_delete(struct btree_trans *trans, u32 id)
757 {
758         struct bch_fs *c = trans->c;
759         struct btree_iter iter, p_iter = (struct btree_iter) { NULL };
760         struct btree_iter tree_iter = (struct btree_iter) { NULL };
761         struct bkey_s_c_snapshot s;
762         u32 parent_id;
763         unsigned i;
764         int ret = 0;
765
766         s = bch2_bkey_get_iter_typed(trans, &iter, BTREE_ID_snapshots, POS(0, id),
767                                      BTREE_ITER_INTENT, snapshot);
768         ret = bkey_err(s);
769         bch2_fs_inconsistent_on(bch2_err_matches(ret, ENOENT), c,
770                                 "missing snapshot %u", id);
771
772         if (ret)
773                 goto err;
774
775         BUG_ON(!BCH_SNAPSHOT_DELETED(s.v));
776         parent_id = le32_to_cpu(s.v->parent);
777
778         if (parent_id) {
779                 struct bkey_i_snapshot *parent;
780
781                 parent = bch2_bkey_get_mut_typed(trans, &p_iter,
782                                      BTREE_ID_snapshots, POS(0, parent_id),
783                                      0, snapshot);
784                 ret = PTR_ERR_OR_ZERO(parent);
785                 if (unlikely(ret)) {
786                         bch2_fs_inconsistent_on(bch2_err_matches(ret, ENOENT), c,
787                                                 "missing snapshot %u", parent_id);
788                         goto err;
789                 }
790
791                 for (i = 0; i < 2; i++)
792                         if (le32_to_cpu(parent->v.children[i]) == id)
793                                 break;
794
795                 if (i == 2)
796                         bch_err(c, "snapshot %u missing child pointer to %u",
797                                 parent_id, id);
798                 else
799                         parent->v.children[i] = 0;
800
801                 if (le32_to_cpu(parent->v.children[0]) <
802                     le32_to_cpu(parent->v.children[1]))
803                         swap(parent->v.children[0],
804                              parent->v.children[1]);
805         } else {
806                 /*
807                  * We're deleting the root of a snapshot tree: update the
808                  * snapshot_tree entry to point to the new root, or delete it if
809                  * this is the last snapshot ID in this tree:
810                  */
811                 struct bkey_i_snapshot_tree *s_t;
812
813                 BUG_ON(s.v->children[1]);
814
815                 s_t = bch2_bkey_get_mut_typed(trans, &tree_iter,
816                                 BTREE_ID_snapshot_trees, POS(0, le32_to_cpu(s.v->tree)),
817                                 0, snapshot_tree);
818                 ret = PTR_ERR_OR_ZERO(s_t);
819                 if (ret)
820                         goto err;
821
822                 if (s.v->children[0]) {
823                         s_t->v.root_snapshot = cpu_to_le32(s.v->children[0]);
824                 } else {
825                         s_t->k.type = KEY_TYPE_deleted;
826                         set_bkey_val_u64s(&s_t->k, 0);
827                 }
828         }
829
830         ret = bch2_btree_delete_at(trans, &iter, 0);
831 err:
832         bch2_trans_iter_exit(trans, &tree_iter);
833         bch2_trans_iter_exit(trans, &p_iter);
834         bch2_trans_iter_exit(trans, &iter);
835         return ret;
836 }
837
838 static int create_snapids(struct btree_trans *trans, u32 parent, u32 tree,
839                           u32 *new_snapids,
840                           u32 *snapshot_subvols,
841                           unsigned nr_snapids)
842 {
843         struct btree_iter iter;
844         struct bkey_i_snapshot *n;
845         struct bkey_s_c k;
846         unsigned i;
847         int ret;
848
849         bch2_trans_iter_init(trans, &iter, BTREE_ID_snapshots,
850                              POS_MIN, BTREE_ITER_INTENT);
851         k = bch2_btree_iter_peek(&iter);
852         ret = bkey_err(k);
853         if (ret)
854                 goto err;
855
856         for (i = 0; i < nr_snapids; i++) {
857                 k = bch2_btree_iter_prev_slot(&iter);
858                 ret = bkey_err(k);
859                 if (ret)
860                         goto err;
861
862                 if (!k.k || !k.k->p.offset) {
863                         ret = -BCH_ERR_ENOSPC_snapshot_create;
864                         goto err;
865                 }
866
867                 n = bch2_bkey_alloc(trans, &iter, 0, snapshot);
868                 ret = PTR_ERR_OR_ZERO(n);
869                 if (ret)
870                         goto err;
871
872                 n->v.flags      = 0;
873                 n->v.parent     = cpu_to_le32(parent);
874                 n->v.subvol     = cpu_to_le32(snapshot_subvols[i]);
875                 n->v.tree       = cpu_to_le32(tree);
876                 SET_BCH_SNAPSHOT_SUBVOL(&n->v, true);
877
878                 ret = bch2_mark_snapshot(trans, BTREE_ID_snapshots, 0,
879                                          bkey_s_c_null, bkey_i_to_s_c(&n->k_i), 0);
880                 if (ret)
881                         goto err;
882
883                 new_snapids[i]  = iter.pos.offset;
884         }
885 err:
886         bch2_trans_iter_exit(trans, &iter);
887         return ret;
888 }
889
890 /*
891  * Create new snapshot IDs as children of an existing snapshot ID:
892  */
893 static int bch2_snapshot_node_create_children(struct btree_trans *trans, u32 parent,
894                               u32 *new_snapids,
895                               u32 *snapshot_subvols,
896                               unsigned nr_snapids)
897 {
898         struct btree_iter iter;
899         struct bkey_i_snapshot *n_parent;
900         int ret = 0;
901
902         n_parent = bch2_bkey_get_mut_typed(trans, &iter,
903                         BTREE_ID_snapshots, POS(0, parent),
904                         0, snapshot);
905         ret = PTR_ERR_OR_ZERO(n_parent);
906         if (unlikely(ret)) {
907                 if (bch2_err_matches(ret, ENOENT))
908                         bch_err(trans->c, "snapshot %u not found", parent);
909                 return ret;
910         }
911
912         if (n_parent->v.children[0] || n_parent->v.children[1]) {
913                 bch_err(trans->c, "Trying to add child snapshot nodes to parent that already has children");
914                 ret = -EINVAL;
915                 goto err;
916         }
917
918         ret = create_snapids(trans, parent, le32_to_cpu(n_parent->v.tree),
919                              new_snapids, snapshot_subvols, nr_snapids);
920         if (ret)
921                 goto err;
922
923         n_parent->v.children[0] = cpu_to_le32(new_snapids[0]);
924         n_parent->v.children[1] = cpu_to_le32(new_snapids[1]);
925         n_parent->v.subvol = 0;
926         SET_BCH_SNAPSHOT_SUBVOL(&n_parent->v, false);
927 err:
928         bch2_trans_iter_exit(trans, &iter);
929         return ret;
930 }
931
932 /*
933  * Create a snapshot node that is the root of a new tree:
934  */
935 static int bch2_snapshot_node_create_tree(struct btree_trans *trans,
936                               u32 *new_snapids,
937                               u32 *snapshot_subvols,
938                               unsigned nr_snapids)
939 {
940         struct bkey_i_snapshot_tree *n_tree;
941         int ret;
942
943         n_tree = __snapshot_tree_create(trans);
944         ret =   PTR_ERR_OR_ZERO(n_tree) ?:
945                 create_snapids(trans, 0, n_tree->k.p.offset,
946                              new_snapids, snapshot_subvols, nr_snapids);
947         if (ret)
948                 return ret;
949
950         n_tree->v.master_subvol = cpu_to_le32(snapshot_subvols[0]);
951         n_tree->v.root_snapshot = cpu_to_le32(new_snapids[0]);
952         return 0;
953 }
954
955 int bch2_snapshot_node_create(struct btree_trans *trans, u32 parent,
956                               u32 *new_snapids,
957                               u32 *snapshot_subvols,
958                               unsigned nr_snapids)
959 {
960         BUG_ON((parent == 0) != (nr_snapids == 1));
961         BUG_ON((parent != 0) != (nr_snapids == 2));
962
963         return parent
964                 ? bch2_snapshot_node_create_children(trans, parent,
965                                 new_snapids, snapshot_subvols, nr_snapids)
966                 : bch2_snapshot_node_create_tree(trans,
967                                 new_snapids, snapshot_subvols, nr_snapids);
968
969 }
970
971 static int snapshot_delete_key(struct btree_trans *trans,
972                                struct btree_iter *iter,
973                                struct bkey_s_c k,
974                                snapshot_id_list *deleted,
975                                snapshot_id_list *equiv_seen,
976                                struct bpos *last_pos)
977 {
978         struct bch_fs *c = trans->c;
979         u32 equiv = snapshot_t(c, k.k->p.snapshot)->equiv;
980
981         if (!bkey_eq(k.k->p, *last_pos))
982                 equiv_seen->nr = 0;
983         *last_pos = k.k->p;
984
985         if (snapshot_list_has_id(deleted, k.k->p.snapshot) ||
986             snapshot_list_has_id(equiv_seen, equiv)) {
987                 return bch2_btree_delete_at(trans, iter,
988                                             BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE);
989         } else {
990                 return snapshot_list_add(c, equiv_seen, equiv);
991         }
992 }
993
994 static int bch2_delete_redundant_snapshot(struct btree_trans *trans, struct btree_iter *iter,
995                                           struct bkey_s_c k)
996 {
997         struct bkey_s_c_snapshot snap;
998         u32 children[2];
999         int ret;
1000
1001         if (k.k->type != KEY_TYPE_snapshot)
1002                 return 0;
1003
1004         snap = bkey_s_c_to_snapshot(k);
1005         if (BCH_SNAPSHOT_DELETED(snap.v) ||
1006             BCH_SNAPSHOT_SUBVOL(snap.v))
1007                 return 0;
1008
1009         children[0] = le32_to_cpu(snap.v->children[0]);
1010         children[1] = le32_to_cpu(snap.v->children[1]);
1011
1012         ret   = snapshot_live(trans, children[0]) ?:
1013                 snapshot_live(trans, children[1]);
1014         if (ret < 0)
1015                 return ret;
1016
1017         if (!ret)
1018                 return bch2_snapshot_node_set_deleted(trans, k.k->p.offset);
1019         return 0;
1020 }
1021
1022 int bch2_delete_dead_snapshots(struct bch_fs *c)
1023 {
1024         struct btree_trans trans;
1025         struct btree_iter iter;
1026         struct bkey_s_c k;
1027         struct bkey_s_c_snapshot snap;
1028         snapshot_id_list deleted = { 0 };
1029         u32 i, id;
1030         int ret = 0;
1031
1032         if (!test_bit(BCH_FS_HAVE_DELETED_SNAPSHOTS, &c->flags))
1033                 return 0;
1034
1035         if (!test_bit(BCH_FS_STARTED, &c->flags)) {
1036                 ret = bch2_fs_read_write_early(c);
1037                 if (ret) {
1038                         bch_err(c, "error deleleting dead snapshots: error going rw: %s", bch2_err_str(ret));
1039                         return ret;
1040                 }
1041         }
1042
1043         bch2_trans_init(&trans, c, 0, 0);
1044
1045         /*
1046          * For every snapshot node: If we have no live children and it's not
1047          * pointed to by a subvolume, delete it:
1048          */
1049         ret = for_each_btree_key_commit(&trans, iter, BTREE_ID_snapshots,
1050                         POS_MIN, 0, k,
1051                         NULL, NULL, 0,
1052                 bch2_delete_redundant_snapshot(&trans, &iter, k));
1053         if (ret) {
1054                 bch_err(c, "error deleting redundant snapshots: %s", bch2_err_str(ret));
1055                 goto err;
1056         }
1057
1058         for_each_btree_key2(&trans, iter, BTREE_ID_snapshots,
1059                            POS_MIN, 0, k,
1060                 bch2_snapshot_set_equiv(&trans, k));
1061         if (ret) {
1062                 bch_err(c, "error in bch2_snapshots_set_equiv: %s", bch2_err_str(ret));
1063                 goto err;
1064         }
1065
1066         for_each_btree_key(&trans, iter, BTREE_ID_snapshots,
1067                            POS_MIN, 0, k, ret) {
1068                 if (k.k->type != KEY_TYPE_snapshot)
1069                         continue;
1070
1071                 snap = bkey_s_c_to_snapshot(k);
1072                 if (BCH_SNAPSHOT_DELETED(snap.v)) {
1073                         ret = snapshot_list_add(c, &deleted, k.k->p.offset);
1074                         if (ret)
1075                                 break;
1076                 }
1077         }
1078         bch2_trans_iter_exit(&trans, &iter);
1079
1080         if (ret) {
1081                 bch_err(c, "error walking snapshots: %s", bch2_err_str(ret));
1082                 goto err;
1083         }
1084
1085         for (id = 0; id < BTREE_ID_NR; id++) {
1086                 struct bpos last_pos = POS_MIN;
1087                 snapshot_id_list equiv_seen = { 0 };
1088
1089                 if (!btree_type_has_snapshots(id))
1090                         continue;
1091
1092                 ret = for_each_btree_key_commit(&trans, iter,
1093                                 id, POS_MIN,
1094                                 BTREE_ITER_PREFETCH|BTREE_ITER_ALL_SNAPSHOTS, k,
1095                                 NULL, NULL, BTREE_INSERT_NOFAIL,
1096                         snapshot_delete_key(&trans, &iter, k, &deleted, &equiv_seen, &last_pos));
1097
1098                 darray_exit(&equiv_seen);
1099
1100                 if (ret) {
1101                         bch_err(c, "error deleting snapshot keys: %s", bch2_err_str(ret));
1102                         goto err;
1103                 }
1104         }
1105
1106         for (i = 0; i < deleted.nr; i++) {
1107                 ret = commit_do(&trans, NULL, NULL, 0,
1108                         bch2_snapshot_node_delete(&trans, deleted.data[i]));
1109                 if (ret) {
1110                         bch_err(c, "error deleting snapshot %u: %s",
1111                                 deleted.data[i], bch2_err_str(ret));
1112                         goto err;
1113                 }
1114         }
1115
1116         clear_bit(BCH_FS_HAVE_DELETED_SNAPSHOTS, &c->flags);
1117 err:
1118         darray_exit(&deleted);
1119         bch2_trans_exit(&trans);
1120         return ret;
1121 }
1122
1123 static void bch2_delete_dead_snapshots_work(struct work_struct *work)
1124 {
1125         struct bch_fs *c = container_of(work, struct bch_fs, snapshot_delete_work);
1126
1127         bch2_delete_dead_snapshots(c);
1128         bch2_write_ref_put(c, BCH_WRITE_REF_delete_dead_snapshots);
1129 }
1130
1131 void bch2_delete_dead_snapshots_async(struct bch_fs *c)
1132 {
1133         if (bch2_write_ref_tryget(c, BCH_WRITE_REF_delete_dead_snapshots) &&
1134             !queue_work(c->write_ref_wq, &c->snapshot_delete_work))
1135                 bch2_write_ref_put(c, BCH_WRITE_REF_delete_dead_snapshots);
1136 }
1137
1138 static int bch2_delete_dead_snapshots_hook(struct btree_trans *trans,
1139                                            struct btree_trans_commit_hook *h)
1140 {
1141         struct bch_fs *c = trans->c;
1142
1143         set_bit(BCH_FS_HAVE_DELETED_SNAPSHOTS, &c->flags);
1144
1145         if (!test_bit(BCH_FS_FSCK_DONE, &c->flags))
1146                 return 0;
1147
1148         bch2_delete_dead_snapshots_async(c);
1149         return 0;
1150 }
1151
1152 /* Subvolumes: */
1153
1154 int bch2_subvolume_invalid(const struct bch_fs *c, struct bkey_s_c k,
1155                            unsigned flags, struct printbuf *err)
1156 {
1157         if (bkey_lt(k.k->p, SUBVOL_POS_MIN) ||
1158             bkey_gt(k.k->p, SUBVOL_POS_MAX)) {
1159                 prt_printf(err, "invalid pos");
1160                 return -BCH_ERR_invalid_bkey;
1161         }
1162
1163         return 0;
1164 }
1165
1166 void bch2_subvolume_to_text(struct printbuf *out, struct bch_fs *c,
1167                             struct bkey_s_c k)
1168 {
1169         struct bkey_s_c_subvolume s = bkey_s_c_to_subvolume(k);
1170
1171         prt_printf(out, "root %llu snapshot id %u",
1172                    le64_to_cpu(s.v->inode),
1173                    le32_to_cpu(s.v->snapshot));
1174
1175         if (bkey_val_bytes(s.k) > offsetof(struct bch_subvolume, parent))
1176                 prt_printf(out, " parent %u", le32_to_cpu(s.v->parent));
1177 }
1178
1179 static __always_inline int
1180 bch2_subvolume_get_inlined(struct btree_trans *trans, unsigned subvol,
1181                            bool inconsistent_if_not_found,
1182                            int iter_flags,
1183                            struct bch_subvolume *s)
1184 {
1185         int ret = bch2_bkey_get_val_typed(trans, BTREE_ID_subvolumes, POS(0, subvol),
1186                                           iter_flags, subvolume, s);
1187         bch2_fs_inconsistent_on(bch2_err_matches(ret, ENOENT) &&
1188                                 inconsistent_if_not_found,
1189                                 trans->c, "missing subvolume %u", subvol);
1190         return ret;
1191 }
1192
1193 int bch2_subvolume_get(struct btree_trans *trans, unsigned subvol,
1194                        bool inconsistent_if_not_found,
1195                        int iter_flags,
1196                        struct bch_subvolume *s)
1197 {
1198         return bch2_subvolume_get_inlined(trans, subvol, inconsistent_if_not_found, iter_flags, s);
1199 }
1200
1201 int bch2_snapshot_get_subvol(struct btree_trans *trans, u32 snapshot,
1202                              struct bch_subvolume *subvol)
1203 {
1204         struct bch_snapshot snap;
1205
1206         return  snapshot_lookup(trans, snapshot, &snap) ?:
1207                 bch2_subvolume_get(trans, le32_to_cpu(snap.subvol), true, 0, subvol);
1208 }
1209
1210 int bch2_subvolume_get_snapshot(struct btree_trans *trans, u32 subvol,
1211                                 u32 *snapid)
1212 {
1213         struct btree_iter iter;
1214         struct bkey_s_c k;
1215         int ret;
1216
1217         k = bch2_bkey_get_iter(trans, &iter, BTREE_ID_subvolumes, POS(0, subvol),
1218                                BTREE_ITER_CACHED|
1219                                BTREE_ITER_WITH_UPDATES);
1220         ret = bkey_err(k) ?: k.k->type == KEY_TYPE_subvolume ? 0 : -ENOENT;
1221
1222         if (likely(!ret))
1223                 *snapid = le32_to_cpu(bkey_s_c_to_subvolume(k).v->snapshot);
1224         else if (bch2_err_matches(ret, ENOENT))
1225                 bch2_fs_inconsistent(trans->c, "missing subvolume %u", subvol);
1226         bch2_trans_iter_exit(trans, &iter);
1227         return ret;
1228 }
1229
1230 static int bch2_subvolume_reparent(struct btree_trans *trans,
1231                                    struct btree_iter *iter,
1232                                    struct bkey_s_c k,
1233                                    u32 old_parent, u32 new_parent)
1234 {
1235         struct bkey_i_subvolume *s;
1236         int ret;
1237
1238         if (k.k->type != KEY_TYPE_subvolume)
1239                 return 0;
1240
1241         if (bkey_val_bytes(k.k) > offsetof(struct bch_subvolume, parent) &&
1242             le32_to_cpu(bkey_s_c_to_subvolume(k).v->parent) != old_parent)
1243                 return 0;
1244
1245         s = bch2_bkey_make_mut_typed(trans, iter, k, 0, subvolume);
1246         ret = PTR_ERR_OR_ZERO(s);
1247         if (ret)
1248                 return ret;
1249
1250         s->v.parent = cpu_to_le32(new_parent);
1251         return 0;
1252 }
1253
1254 /*
1255  * Scan for subvolumes with parent @subvolid_to_delete, reparent:
1256  */
1257 static int bch2_subvolumes_reparent(struct btree_trans *trans, u32 subvolid_to_delete)
1258 {
1259         struct btree_iter iter;
1260         struct bkey_s_c k;
1261         struct bch_subvolume s;
1262
1263         return lockrestart_do(trans,
1264                         bch2_subvolume_get(trans, subvolid_to_delete, true,
1265                                    BTREE_ITER_CACHED, &s)) ?:
1266                 for_each_btree_key_commit(trans, iter,
1267                                 BTREE_ID_subvolumes, POS_MIN, BTREE_ITER_PREFETCH, k,
1268                                 NULL, NULL, BTREE_INSERT_NOFAIL,
1269                         bch2_subvolume_reparent(trans, &iter, k,
1270                                         subvolid_to_delete, le32_to_cpu(s.parent)));
1271 }
1272
1273 /*
1274  * Delete subvolume, mark snapshot ID as deleted, queue up snapshot
1275  * deletion/cleanup:
1276  */
1277 static int __bch2_subvolume_delete(struct btree_trans *trans, u32 subvolid)
1278 {
1279         struct btree_iter iter;
1280         struct bkey_s_c_subvolume subvol;
1281         struct btree_trans_commit_hook *h;
1282         u32 snapid;
1283         int ret = 0;
1284
1285         subvol = bch2_bkey_get_iter_typed(trans, &iter,
1286                                 BTREE_ID_subvolumes, POS(0, subvolid),
1287                                 BTREE_ITER_CACHED|BTREE_ITER_INTENT,
1288                                 subvolume);
1289         ret = bkey_err(subvol);
1290         bch2_fs_inconsistent_on(bch2_err_matches(ret, ENOENT), trans->c,
1291                                 "missing subvolume %u", subvolid);
1292         if (ret)
1293                 return ret;
1294
1295         snapid = le32_to_cpu(subvol.v->snapshot);
1296
1297         ret = bch2_btree_delete_at(trans, &iter, 0);
1298         if (ret)
1299                 goto err;
1300
1301         ret = bch2_snapshot_node_set_deleted(trans, snapid);
1302         if (ret)
1303                 goto err;
1304
1305         h = bch2_trans_kmalloc(trans, sizeof(*h));
1306         ret = PTR_ERR_OR_ZERO(h);
1307         if (ret)
1308                 goto err;
1309
1310         h->fn = bch2_delete_dead_snapshots_hook;
1311         bch2_trans_commit_hook(trans, h);
1312 err:
1313         bch2_trans_iter_exit(trans, &iter);
1314         return ret;
1315 }
1316
1317 static int bch2_subvolume_delete(struct btree_trans *trans, u32 subvolid)
1318 {
1319         return bch2_subvolumes_reparent(trans, subvolid) ?:
1320                 commit_do(trans, NULL, NULL, BTREE_INSERT_NOFAIL,
1321                           __bch2_subvolume_delete(trans, subvolid));
1322 }
1323
1324 void bch2_subvolume_wait_for_pagecache_and_delete(struct work_struct *work)
1325 {
1326         struct bch_fs *c = container_of(work, struct bch_fs,
1327                                 snapshot_wait_for_pagecache_and_delete_work);
1328         snapshot_id_list s;
1329         u32 *id;
1330         int ret = 0;
1331
1332         while (!ret) {
1333                 mutex_lock(&c->snapshots_unlinked_lock);
1334                 s = c->snapshots_unlinked;
1335                 darray_init(&c->snapshots_unlinked);
1336                 mutex_unlock(&c->snapshots_unlinked_lock);
1337
1338                 if (!s.nr)
1339                         break;
1340
1341                 bch2_evict_subvolume_inodes(c, &s);
1342
1343                 for (id = s.data; id < s.data + s.nr; id++) {
1344                         ret = bch2_trans_run(c, bch2_subvolume_delete(&trans, *id));
1345                         if (ret) {
1346                                 bch_err(c, "error deleting subvolume %u: %s", *id, bch2_err_str(ret));
1347                                 break;
1348                         }
1349                 }
1350
1351                 darray_exit(&s);
1352         }
1353
1354         bch2_write_ref_put(c, BCH_WRITE_REF_snapshot_delete_pagecache);
1355 }
1356
1357 struct subvolume_unlink_hook {
1358         struct btree_trans_commit_hook  h;
1359         u32                             subvol;
1360 };
1361
1362 int bch2_subvolume_wait_for_pagecache_and_delete_hook(struct btree_trans *trans,
1363                                                       struct btree_trans_commit_hook *_h)
1364 {
1365         struct subvolume_unlink_hook *h = container_of(_h, struct subvolume_unlink_hook, h);
1366         struct bch_fs *c = trans->c;
1367         int ret = 0;
1368
1369         mutex_lock(&c->snapshots_unlinked_lock);
1370         if (!snapshot_list_has_id(&c->snapshots_unlinked, h->subvol))
1371                 ret = snapshot_list_add(c, &c->snapshots_unlinked, h->subvol);
1372         mutex_unlock(&c->snapshots_unlinked_lock);
1373
1374         if (ret)
1375                 return ret;
1376
1377         if (!bch2_write_ref_tryget(c, BCH_WRITE_REF_snapshot_delete_pagecache))
1378                 return -EROFS;
1379
1380         if (!queue_work(c->write_ref_wq, &c->snapshot_wait_for_pagecache_and_delete_work))
1381                 bch2_write_ref_put(c, BCH_WRITE_REF_snapshot_delete_pagecache);
1382         return 0;
1383 }
1384
1385 int bch2_subvolume_unlink(struct btree_trans *trans, u32 subvolid)
1386 {
1387         struct btree_iter iter;
1388         struct bkey_i_subvolume *n;
1389         struct subvolume_unlink_hook *h;
1390         int ret = 0;
1391
1392         h = bch2_trans_kmalloc(trans, sizeof(*h));
1393         ret = PTR_ERR_OR_ZERO(h);
1394         if (ret)
1395                 return ret;
1396
1397         h->h.fn         = bch2_subvolume_wait_for_pagecache_and_delete_hook;
1398         h->subvol       = subvolid;
1399         bch2_trans_commit_hook(trans, &h->h);
1400
1401         n = bch2_bkey_get_mut_typed(trans, &iter,
1402                         BTREE_ID_subvolumes, POS(0, subvolid),
1403                         BTREE_ITER_CACHED, subvolume);
1404         ret = PTR_ERR_OR_ZERO(n);
1405         if (unlikely(ret)) {
1406                 bch2_fs_inconsistent_on(bch2_err_matches(ret, ENOENT), trans->c,
1407                                         "missing subvolume %u", subvolid);
1408                 return ret;
1409         }
1410
1411         SET_BCH_SUBVOLUME_UNLINKED(&n->v, true);
1412         bch2_trans_iter_exit(trans, &iter);
1413         return ret;
1414 }
1415
1416 int bch2_subvolume_create(struct btree_trans *trans, u64 inode,
1417                           u32 src_subvolid,
1418                           u32 *new_subvolid,
1419                           u32 *new_snapshotid,
1420                           bool ro)
1421 {
1422         struct bch_fs *c = trans->c;
1423         struct btree_iter dst_iter, src_iter = (struct btree_iter) { NULL };
1424         struct bkey_i_subvolume *new_subvol = NULL;
1425         struct bkey_i_subvolume *src_subvol = NULL;
1426         u32 parent = 0, new_nodes[2], snapshot_subvols[2];
1427         int ret = 0;
1428
1429         ret = bch2_bkey_get_empty_slot(trans, &dst_iter,
1430                                 BTREE_ID_subvolumes, POS(0, U32_MAX));
1431         if (ret == -BCH_ERR_ENOSPC_btree_slot)
1432                 ret = -BCH_ERR_ENOSPC_subvolume_create;
1433         if (ret)
1434                 return ret;
1435
1436         snapshot_subvols[0] = dst_iter.pos.offset;
1437         snapshot_subvols[1] = src_subvolid;
1438
1439         if (src_subvolid) {
1440                 /* Creating a snapshot: */
1441
1442                 src_subvol = bch2_bkey_get_mut_typed(trans, &src_iter,
1443                                 BTREE_ID_subvolumes, POS(0, src_subvolid),
1444                                 BTREE_ITER_CACHED, subvolume);
1445                 ret = PTR_ERR_OR_ZERO(src_subvol);
1446                 if (unlikely(ret)) {
1447                         bch2_fs_inconsistent_on(ret == -ENOENT, c,
1448                                                 "subvolume %u not found", src_subvolid);
1449                         goto err;
1450                 }
1451
1452                 parent = le32_to_cpu(src_subvol->v.snapshot);
1453         }
1454
1455         ret = bch2_snapshot_node_create(trans, parent, new_nodes,
1456                                         snapshot_subvols,
1457                                         src_subvolid ? 2 : 1);
1458         if (ret)
1459                 goto err;
1460
1461         if (src_subvolid) {
1462                 src_subvol->v.snapshot = cpu_to_le32(new_nodes[1]);
1463                 ret = bch2_trans_update(trans, &src_iter, &src_subvol->k_i, 0);
1464                 if (ret)
1465                         goto err;
1466         }
1467
1468         new_subvol = bch2_bkey_alloc(trans, &dst_iter, 0, subvolume);
1469         ret = PTR_ERR_OR_ZERO(new_subvol);
1470         if (ret)
1471                 goto err;
1472
1473         new_subvol->v.flags     = 0;
1474         new_subvol->v.snapshot  = cpu_to_le32(new_nodes[0]);
1475         new_subvol->v.inode     = cpu_to_le64(inode);
1476         new_subvol->v.parent    = cpu_to_le32(src_subvolid);
1477         new_subvol->v.otime.lo  = cpu_to_le64(bch2_current_time(c));
1478         new_subvol->v.otime.hi  = 0;
1479
1480         SET_BCH_SUBVOLUME_RO(&new_subvol->v, ro);
1481         SET_BCH_SUBVOLUME_SNAP(&new_subvol->v, src_subvolid != 0);
1482
1483         *new_subvolid   = new_subvol->k.p.offset;
1484         *new_snapshotid = new_nodes[0];
1485 err:
1486         bch2_trans_iter_exit(trans, &src_iter);
1487         bch2_trans_iter_exit(trans, &dst_iter);
1488         return ret;
1489 }
1490
1491 int bch2_fs_subvolumes_init(struct bch_fs *c)
1492 {
1493         INIT_WORK(&c->snapshot_delete_work, bch2_delete_dead_snapshots_work);
1494         INIT_WORK(&c->snapshot_wait_for_pagecache_and_delete_work,
1495                   bch2_subvolume_wait_for_pagecache_and_delete);
1496         mutex_init(&c->snapshots_unlinked_lock);
1497         return 0;
1498 }