git.sesse.net Git - bcachefs-tools-debian/blob - 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 = bch2_bkey_get_val_typed(trans, BTREE_ID_snapshot_trees, POS(0, id),
  41                                           BTREE_ITER_WITH_UPDATES, snapshot_tree, s);
  42
  43         if (bch2_err_matches(ret, ENOENT))
  44                 ret = -BCH_ERR_ENOENT_snapshot_tree;
  45         return ret;
  46 }
  47
  48 static struct bkey_i_snapshot_tree *
  49 __snapshot_tree_create(struct btree_trans *trans)
  50 {
  51         struct btree_iter iter;
  52         int ret = bch2_bkey_get_empty_slot(trans, &iter,
  53                         BTREE_ID_snapshot_trees, POS(0, U32_MAX));
  54         struct bkey_i_snapshot_tree *s_t;
  55
  56         if (ret == -BCH_ERR_ENOSPC_btree_slot)
  57                 ret = -BCH_ERR_ENOSPC_snapshot_tree;
  58         if (ret)
  59                 return ERR_PTR(ret);
  60
  61         s_t = bch2_bkey_alloc(trans, &iter, 0, snapshot_tree);
  62         ret = PTR_ERR_OR_ZERO(s_t);
  63         bch2_trans_iter_exit(trans, &iter);
  64         return ret ? ERR_PTR(ret) : s_t;
  65 }
  66
  67 static int snapshot_tree_create(struct btree_trans *trans,
  68                                 u32 root_id, u32 subvol_id, u32 *tree_id)
  69 {
  70         struct bkey_i_snapshot_tree *n_tree =
  71                 __snapshot_tree_create(trans);
  72
  73         if (IS_ERR(n_tree))
  74                 return PTR_ERR(n_tree);
  75
  76         n_tree->v.master_subvol = cpu_to_le32(subvol_id);
  77         n_tree->v.root_snapshot = cpu_to_le32(root_id);
  78         *tree_id = n_tree->k.p.offset;
  79         return 0;
  80 }
  81
  82 /* Snapshot nodes: */
  83
  84 void bch2_snapshot_to_text(struct printbuf *out, struct bch_fs *c,
  85                            struct bkey_s_c k)
  86 {
  87         struct bkey_s_c_snapshot s = bkey_s_c_to_snapshot(k);
  88
  89         prt_printf(out, "is_subvol %llu deleted %llu parent %10u children %10u %10u subvol %u tree %u",
  90                BCH_SNAPSHOT_SUBVOL(s.v),
  91                BCH_SNAPSHOT_DELETED(s.v),
  92                le32_to_cpu(s.v->parent),
  93                le32_to_cpu(s.v->children[0]),
  94                le32_to_cpu(s.v->children[1]),
  95                le32_to_cpu(s.v->subvol),
  96                le32_to_cpu(s.v->tree));
  97 }
  98
  99 int bch2_snapshot_invalid(const struct bch_fs *c, struct bkey_s_c k,
 100                           unsigned flags, struct printbuf *err)
 101 {
 102         struct bkey_s_c_snapshot s;
 103         u32 i, id;
 104
 105         if (bkey_gt(k.k->p, POS(0, U32_MAX)) ||
 106             bkey_lt(k.k->p, POS(0, 1))) {
 107                 prt_printf(err, "bad pos");
 108                 return -BCH_ERR_invalid_bkey;
 109         }
 110
 111         s = bkey_s_c_to_snapshot(k);
 112
 113         id = le32_to_cpu(s.v->parent);
 114         if (id && id <= k.k->p.offset) {
 115                 prt_printf(err, "bad parent node (%u <= %llu)",
 116                        id, k.k->p.offset);
 117                 return -BCH_ERR_invalid_bkey;
 118         }
 119
 120         if (le32_to_cpu(s.v->children[0]) < le32_to_cpu(s.v->children[1])) {
 121                 prt_printf(err, "children not normalized");
 122                 return -BCH_ERR_invalid_bkey;
 123         }
 124
 125         if (s.v->children[0] &&
 126             s.v->children[0] == s.v->children[1]) {
 127                 prt_printf(err, "duplicate child nodes");
 128                 return -BCH_ERR_invalid_bkey;
 129         }
 130
 131         for (i = 0; i < 2; i++) {
 132                 id = le32_to_cpu(s.v->children[i]);
 133
 134                 if (id >= k.k->p.offset) {
 135                         prt_printf(err, "bad child node (%u >= %llu)",
 136                                id, k.k->p.offset);
 137                         return -BCH_ERR_invalid_bkey;
 138                 }
 139         }
 140
 141         return 0;
 142 }
 143
 144 int bch2_mark_snapshot(struct btree_trans *trans,
 145                        enum btree_id btree, unsigned level,
 146                        struct bkey_s_c old, struct bkey_s_c new,
 147                        unsigned flags)
 148 {
 149         struct bch_fs *c = trans->c;
 150         struct snapshot_t *t;
 151
 152         t = genradix_ptr_alloc(&c->snapshots,
 153                                U32_MAX - new.k->p.offset,
 154                                GFP_KERNEL);
 155         if (!t)
 156                 return -BCH_ERR_ENOMEM_mark_snapshot;
 157
 158         if (new.k->type == KEY_TYPE_snapshot) {
 159                 struct bkey_s_c_snapshot s = bkey_s_c_to_snapshot(new);
 160
 161                 t->parent       = le32_to_cpu(s.v->parent);
 162                 t->children[0]  = le32_to_cpu(s.v->children[0]);
 163                 t->children[1]  = le32_to_cpu(s.v->children[1]);
 164                 t->subvol       = BCH_SNAPSHOT_SUBVOL(s.v) ? le32_to_cpu(s.v->subvol) : 0;
 165                 t->tree         = le32_to_cpu(s.v->tree);
 166         } else {
 167                 t->parent       = 0;
 168                 t->children[0]  = 0;
 169                 t->children[1]  = 0;
 170                 t->subvol       = 0;
 171                 t->tree         = 0;
 172         }
 173
 174         return 0;
 175 }
 176
 177 static int snapshot_lookup(struct btree_trans *trans, u32 id,
 178                            struct bch_snapshot *s)
 179 {
 180         return bch2_bkey_get_val_typed(trans, BTREE_ID_snapshots, POS(0, id),
 181                                        BTREE_ITER_WITH_UPDATES, snapshot, s);
 182 }
 183
 184 static int snapshot_live(struct btree_trans *trans, u32 id)
 185 {
 186         struct bch_snapshot v;
 187         int ret;
 188
 189         if (!id)
 190                 return 0;
 191
 192         ret = snapshot_lookup(trans, id, &v);
 193         if (bch2_err_matches(ret, ENOENT))
 194                 bch_err(trans->c, "snapshot node %u not found", id);
 195         if (ret)
 196                 return ret;
 197
 198         return !BCH_SNAPSHOT_DELETED(&v);
 199 }
 200
 201 static int bch2_snapshot_set_equiv(struct btree_trans *trans, struct bkey_s_c k)
 202 {
 203         struct bch_fs *c = trans->c;
 204         unsigned i, nr_live = 0, live_idx = 0;
 205         struct bkey_s_c_snapshot snap;
 206         u32 id = k.k->p.offset, child[2];
 207
 208         if (k.k->type != KEY_TYPE_snapshot)
 209                 return 0;
 210
 211         snap = bkey_s_c_to_snapshot(k);
 212
 213         child[0] = le32_to_cpu(snap.v->children[0]);
 214         child[1] = le32_to_cpu(snap.v->children[1]);
 215
 216         for (i = 0; i < 2; i++) {
 217                 int ret = snapshot_live(trans, child[i]);
 218
 219                 if (ret < 0)
 220                         return ret;
 221
 222                 if (ret)
 223                         live_idx = i;
 224                 nr_live += ret;
 225         }
 226
 227         snapshot_t(c, id)->equiv = nr_live == 1
 228                 ? snapshot_t(c, child[live_idx])->equiv
 229                 : id;
 230         return 0;
 231 }
 232
 233 /* fsck: */
 234
 235 static u32 bch2_snapshot_child(struct bch_fs *c, u32 id, unsigned child)
 236 {
 237         return snapshot_t(c, id)->children[child];
 238 }
 239
 240 static u32 bch2_snapshot_left_child(struct bch_fs *c, u32 id)
 241 {
 242         return bch2_snapshot_child(c, id, 0);
 243 }
 244
 245 static u32 bch2_snapshot_right_child(struct bch_fs *c, u32 id)
 246 {
 247         return bch2_snapshot_child(c, id, 1);
 248 }
 249
 250 static u32 bch2_snapshot_tree_next(struct bch_fs *c, u32 id)
 251 {
 252         u32 n, parent;
 253
 254         n = bch2_snapshot_left_child(c, id);
 255         if (n)
 256                 return n;
 257
 258         while ((parent = bch2_snapshot_parent(c, id))) {
 259                 n = bch2_snapshot_right_child(c, parent);
 260                 if (n && n != id)
 261                         return n;
 262                 id = parent;
 263         }
 264
 265         return 0;
 266 }
 267
 268 static u32 bch2_snapshot_tree_oldest_subvol(struct bch_fs *c, u32 snapshot_root)
 269 {
 270         u32 id = snapshot_root;
 271         u32 subvol = 0, s;
 272
 273         while (id) {
 274                 s = snapshot_t(c, id)->subvol;
 275
 276                 if (s && (!subvol || s < subvol))
 277                         subvol = s;
 278
 279                 id = bch2_snapshot_tree_next(c, id);
 280         }
 281
 282         return subvol;
 283 }
 284
 285 static int bch2_snapshot_tree_master_subvol(struct btree_trans *trans,
 286                                             u32 snapshot_root, u32 *subvol_id)
 287 {
 288         struct bch_fs *c = trans->c;
 289         struct btree_iter iter;
 290         struct bkey_s_c k;
 291         struct bkey_s_c_subvolume s;
 292         bool found = false;
 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                         found = true;
 306                         break;
 307                 }
 308         }
 309
 310         bch2_trans_iter_exit(trans, &iter);
 311
 312         if (!ret && !found) {
 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 (bch2_err_matches(ret, ENOENT))
 438                 return 0;
 439         if (ret)
 440                 return ret;
 441
 442         return bch2_snapshot_is_ancestor(trans->c, snap_id, le32_to_cpu(s_t.root_snapshot));
 443 }
 444
 445 /*
 446  * snapshot_tree pointer was incorrect: look up root snapshot node, make sure
 447  * its snapshot_tree pointer is correct (allocate new one if necessary), then
 448  * update this node's pointer to root node's pointer:
 449  */
 450 static int snapshot_tree_ptr_repair(struct btree_trans *trans,
 451                                     struct btree_iter *iter,
 452                                     struct bkey_s_c_snapshot *s)
 453 {
 454         struct bch_fs *c = trans->c;
 455         struct btree_iter root_iter;
 456         struct bch_snapshot_tree s_t;
 457         struct bkey_s_c_snapshot root;
 458         struct bkey_i_snapshot *u;
 459         u32 root_id = bch2_snapshot_root(c, s->k->p.offset), tree_id;
 460         int ret;
 461
 462         root = bch2_bkey_get_iter_typed(trans, &root_iter,
 463                                BTREE_ID_snapshots, POS(0, root_id),
 464                                BTREE_ITER_WITH_UPDATES, snapshot);
 465         ret = bkey_err(root);
 466         if (ret)
 467                 goto err;
 468
 469         tree_id = le32_to_cpu(root.v->tree);
 470
 471         ret = bch2_snapshot_tree_lookup(trans, tree_id, &s_t);
 472         if (ret && !bch2_err_matches(ret, ENOENT))
 473                 return ret;
 474
 475         if (ret || le32_to_cpu(s_t.root_snapshot) != root_id) {
 476                 u = bch2_bkey_make_mut_typed(trans, &root_iter, root.s_c, 0, snapshot);
 477                 ret =   PTR_ERR_OR_ZERO(u) ?:
 478                         snapshot_tree_create(trans, root_id,
 479                                 bch2_snapshot_tree_oldest_subvol(c, root_id),
 480                                 &tree_id);
 481                 if (ret)
 482                         goto err;
 483
 484                 u->v.tree = cpu_to_le32(tree_id);
 485                 if (s->k->p.snapshot == root_id)
 486                         *s = snapshot_i_to_s_c(u);
 487         }
 488
 489         if (s->k->p.snapshot != root_id) {
 490                 u = bch2_bkey_make_mut_typed(trans, iter, s->s_c, 0, snapshot);
 491                 ret = PTR_ERR_OR_ZERO(u);
 492                 if (ret)
 493                         goto err;
 494
 495                 u->v.tree = cpu_to_le32(tree_id);
 496                 *s = snapshot_i_to_s_c(u);
 497         }
 498 err:
 499         bch2_trans_iter_exit(trans, &root_iter);
 500         return ret;
 501 }
 502
 503 static int check_snapshot(struct btree_trans *trans,
 504                           struct btree_iter *iter,
 505                           struct bkey_s_c k)
 506 {
 507         struct bch_fs *c = trans->c;
 508         struct bkey_s_c_snapshot s;
 509         struct bch_subvolume subvol;
 510         struct bch_snapshot v;
 511         struct printbuf buf = PRINTBUF;
 512         bool should_have_subvol;
 513         u32 i, id;
 514         int ret = 0;
 515
 516         if (k.k->type != KEY_TYPE_snapshot)
 517                 return 0;
 518
 519         s = bkey_s_c_to_snapshot(k);
 520         id = le32_to_cpu(s.v->parent);
 521         if (id) {
 522                 ret = snapshot_lookup(trans, id, &v);
 523                 if (bch2_err_matches(ret, ENOENT))
 524                         bch_err(c, "snapshot with nonexistent parent:\n  %s",
 525                                 (bch2_bkey_val_to_text(&buf, c, s.s_c), buf.buf));
 526                 if (ret)
 527                         goto err;
 528
 529                 if (le32_to_cpu(v.children[0]) != s.k->p.offset &&
 530                     le32_to_cpu(v.children[1]) != s.k->p.offset) {
 531                         bch_err(c, "snapshot parent %u missing pointer to child %llu",
 532                                 id, s.k->p.offset);
 533                         ret = -EINVAL;
 534                         goto err;
 535                 }
 536         }
 537
 538         for (i = 0; i < 2 && s.v->children[i]; i++) {
 539                 id = le32_to_cpu(s.v->children[i]);
 540
 541                 ret = snapshot_lookup(trans, id, &v);
 542                 if (bch2_err_matches(ret, ENOENT))
 543                         bch_err(c, "snapshot node %llu has nonexistent child %u",
 544                                 s.k->p.offset, id);
 545                 if (ret)
 546                         goto err;
 547
 548                 if (le32_to_cpu(v.parent) != s.k->p.offset) {
 549                         bch_err(c, "snapshot child %u has wrong parent (got %u should be %llu)",
 550                                 id, le32_to_cpu(v.parent), s.k->p.offset);
 551                         ret = -EINVAL;
 552                         goto err;
 553                 }
 554         }
 555
 556         should_have_subvol = BCH_SNAPSHOT_SUBVOL(s.v) &&
 557                 !BCH_SNAPSHOT_DELETED(s.v);
 558
 559         if (should_have_subvol) {
 560                 id = le32_to_cpu(s.v->subvol);
 561                 ret = bch2_subvolume_get(trans, id, 0, false, &subvol);
 562                 if (bch2_err_matches(ret, ENOENT))
 563                         bch_err(c, "snapshot points to nonexistent subvolume:\n  %s",
 564                                 (bch2_bkey_val_to_text(&buf, c, s.s_c), buf.buf));
 565                 if (ret)
 566                         goto err;
 567
 568                 if (BCH_SNAPSHOT_SUBVOL(s.v) != (le32_to_cpu(subvol.snapshot) == s.k->p.offset)) {
 569                         bch_err(c, "snapshot node %llu has wrong BCH_SNAPSHOT_SUBVOL",
 570                                 s.k->p.offset);
 571                         ret = -EINVAL;
 572                         goto err;
 573                 }
 574         } else {
 575                 if (fsck_err_on(s.v->subvol, c, "snapshot should not point to subvol:\n  %s",
 576                                 (bch2_bkey_val_to_text(&buf, c, s.s_c), buf.buf))) {
 577                         struct bkey_i_snapshot *u = bch2_trans_kmalloc(trans, sizeof(*u));
 578
 579                         ret = PTR_ERR_OR_ZERO(u);
 580                         if (ret)
 581                                 goto err;
 582
 583                         bkey_reassemble(&u->k_i, s.s_c);
 584                         u->v.subvol = 0;
 585                         ret = bch2_trans_update(trans, iter, &u->k_i, 0);
 586                         if (ret)
 587                                 goto err;
 588
 589                         s = snapshot_i_to_s_c(u);
 590                 }
 591         }
 592
 593         ret = snapshot_tree_ptr_good(trans, s.k->p.offset, le32_to_cpu(s.v->tree));
 594         if (ret < 0)
 595                 goto err;
 596
 597         if (fsck_err_on(!ret, c, "snapshot points to missing/incorrect tree:\n  %s",
 598                         (bch2_bkey_val_to_text(&buf, c, s.s_c), buf.buf))) {
 599                 ret = snapshot_tree_ptr_repair(trans, iter, &s);
 600                 if (ret)
 601                         goto err;
 602         }
 603         ret = 0;
 604
 605         if (BCH_SNAPSHOT_DELETED(s.v))
 606                 set_bit(BCH_FS_HAVE_DELETED_SNAPSHOTS, &c->flags);
 607 err:
 608 fsck_err:
 609         printbuf_exit(&buf);
 610         return ret;
 611 }
 612
 613 int bch2_fs_check_snapshots(struct bch_fs *c)
 614 {
 615         struct btree_iter iter;
 616         struct bkey_s_c k;
 617         int ret;
 618
 619         ret = bch2_trans_run(c,
 620                 for_each_btree_key_commit(&trans, iter,
 621                         BTREE_ID_snapshots, POS_MIN,
 622                         BTREE_ITER_PREFETCH, k,
 623                         NULL, NULL, BTREE_INSERT_LAZY_RW|BTREE_INSERT_NOFAIL,
 624                 check_snapshot(&trans, &iter, k)));
 625         if (ret)
 626                 bch_err(c, "%s: error %s", __func__, bch2_err_str(ret));
 627         return ret;
 628 }
 629
 630 static int check_subvol(struct btree_trans *trans,
 631                         struct btree_iter *iter,
 632                         struct bkey_s_c k)
 633 {
 634         struct bch_fs *c = trans->c;
 635         struct bkey_s_c_subvolume subvol;
 636         struct bch_snapshot snapshot;
 637         unsigned snapid;
 638         int ret = 0;
 639
 640         if (k.k->type != KEY_TYPE_subvolume)
 641                 return 0;
 642
 643         subvol = bkey_s_c_to_subvolume(k);
 644         snapid = le32_to_cpu(subvol.v->snapshot);
 645         ret = snapshot_lookup(trans, snapid, &snapshot);
 646
 647         if (bch2_err_matches(ret, ENOENT))
 648                 bch_err(c, "subvolume %llu points to nonexistent snapshot %u",
 649                         k.k->p.offset, snapid);
 650         if (ret)
 651                 return ret;
 652
 653         if (BCH_SUBVOLUME_UNLINKED(subvol.v)) {
 654                 bch2_fs_lazy_rw(c);
 655
 656                 ret = bch2_subvolume_delete(trans, iter->pos.offset);
 657                 if (ret)
 658                         bch_err(c, "error deleting subvolume %llu: %s",
 659                                 iter->pos.offset, bch2_err_str(ret));
 660                 return ret ?: -BCH_ERR_transaction_restart_nested;
 661         }
 662
 663         if (!BCH_SUBVOLUME_SNAP(subvol.v)) {
 664                 u32 snapshot_root = bch2_snapshot_root(c, le32_to_cpu(subvol.v->snapshot));
 665                 u32 snapshot_tree = snapshot_t(c, snapshot_root)->tree;
 666                 struct bch_snapshot_tree st;
 667
 668                 ret = bch2_snapshot_tree_lookup(trans, snapshot_tree, &st);
 669
 670                 bch2_fs_inconsistent_on(bch2_err_matches(ret, ENOENT), c,
 671                                 "%s: snapshot tree %u not found", __func__, snapshot_tree);
 672
 673                 if (ret)
 674                         return ret;
 675
 676                 if (fsck_err_on(le32_to_cpu(st.master_subvol) != subvol.k->p.offset, c,
 677                                 "subvolume %llu is not set as snapshot but is not master subvolume",
 678                                 k.k->p.offset)) {
 679                         struct bkey_i_subvolume *s =
 680                                 bch2_bkey_make_mut_typed(trans, iter, subvol.s_c, 0, subvolume);
 681                         ret = PTR_ERR_OR_ZERO(s);
 682                         if (ret)
 683                                 return ret;
 684
 685                         SET_BCH_SUBVOLUME_SNAP(&s->v, true);
 686                 }
 687         }
 688
 689 fsck_err:
 690         return ret;
 691 }
 692
 693 int bch2_fs_check_subvols(struct bch_fs *c)
 694 {
 695         struct btree_iter iter;
 696         struct bkey_s_c k;
 697         int ret;
 698
 699         ret = bch2_trans_run(c,
 700                 for_each_btree_key_commit(&trans, iter,
 701                         BTREE_ID_subvolumes, POS_MIN, BTREE_ITER_PREFETCH, k,
 702                         NULL, NULL, BTREE_INSERT_LAZY_RW|BTREE_INSERT_NOFAIL,
 703                 check_subvol(&trans, &iter, k)));
 704         if (ret)
 705                 bch_err(c, "%s: error %s", __func__, bch2_err_str(ret));
 706
 707         return ret;
 708 }
 709
 710 void bch2_fs_snapshots_exit(struct bch_fs *c)
 711 {
 712         genradix_free(&c->snapshots);
 713 }
 714
 715 int bch2_fs_snapshots_start(struct bch_fs *c)
 716 {
 717         struct btree_iter iter;
 718         struct bkey_s_c k;
 719         int ret = 0;
 720
 721         ret = bch2_trans_run(c,
 722                 for_each_btree_key2(&trans, iter, BTREE_ID_snapshots,
 723                            POS_MIN, 0, k,
 724                         bch2_mark_snapshot(&trans, BTREE_ID_snapshots, 0, bkey_s_c_null, k, 0) ?:
 725                         bch2_snapshot_set_equiv(&trans, k)));
 726         if (ret)
 727                 bch_err(c, "error starting snapshots: %s", bch2_err_str(ret));
 728         return ret;
 729 }
 730
 731 /*
 732  * Mark a snapshot as deleted, for future cleanup:
 733  */
 734 static int bch2_snapshot_node_set_deleted(struct btree_trans *trans, u32 id)
 735 {
 736         struct btree_iter iter;
 737         struct bkey_i_snapshot *s;
 738         int ret = 0;
 739
 740         s = bch2_bkey_get_mut_typed(trans, &iter,
 741                                     BTREE_ID_snapshots, POS(0, id),
 742                                     0, snapshot);
 743         ret = PTR_ERR_OR_ZERO(s);
 744         if (unlikely(ret)) {
 745                 bch2_fs_inconsistent_on(bch2_err_matches(ret, ENOENT),
 746                                         trans->c, "missing snapshot %u", id);
 747                 return ret;
 748         }
 749
 750         /* already deleted? */
 751         if (BCH_SNAPSHOT_DELETED(&s->v))
 752                 goto err;
 753
 754         SET_BCH_SNAPSHOT_DELETED(&s->v, true);
 755         SET_BCH_SNAPSHOT_SUBVOL(&s->v, false);
 756         s->v.subvol = 0;
 757 err:
 758         bch2_trans_iter_exit(trans, &iter);
 759         return ret;
 760 }
 761
 762 static int bch2_snapshot_node_delete(struct btree_trans *trans, u32 id)
 763 {
 764         struct bch_fs *c = trans->c;
 765         struct btree_iter iter, p_iter = (struct btree_iter) { NULL };
 766         struct btree_iter tree_iter = (struct btree_iter) { NULL };
 767         struct bkey_s_c_snapshot s;
 768         u32 parent_id;
 769         unsigned i;
 770         int ret = 0;
 771
 772         s = bch2_bkey_get_iter_typed(trans, &iter, BTREE_ID_snapshots, POS(0, id),
 773                                      BTREE_ITER_INTENT, snapshot);
 774         ret = bkey_err(s);
 775         bch2_fs_inconsistent_on(bch2_err_matches(ret, ENOENT), c,
 776                                 "missing snapshot %u", id);
 777
 778         if (ret)
 779                 goto err;
 780
 781         BUG_ON(!BCH_SNAPSHOT_DELETED(s.v));
 782         parent_id = le32_to_cpu(s.v->parent);
 783
 784         if (parent_id) {
 785                 struct bkey_i_snapshot *parent;
 786
 787                 parent = bch2_bkey_get_mut_typed(trans, &p_iter,
 788                                      BTREE_ID_snapshots, POS(0, parent_id),
 789                                      0, snapshot);
 790                 ret = PTR_ERR_OR_ZERO(parent);
 791                 if (unlikely(ret)) {
 792                         bch2_fs_inconsistent_on(bch2_err_matches(ret, ENOENT), c,
 793                                                 "missing snapshot %u", parent_id);
 794                         goto err;
 795                 }
 796
 797                 for (i = 0; i < 2; i++)
 798                         if (le32_to_cpu(parent->v.children[i]) == id)
 799                                 break;
 800
 801                 if (i == 2)
 802                         bch_err(c, "snapshot %u missing child pointer to %u",
 803                                 parent_id, id);
 804                 else
 805                         parent->v.children[i] = 0;
 806
 807                 if (le32_to_cpu(parent->v.children[0]) <
 808                     le32_to_cpu(parent->v.children[1]))
 809                         swap(parent->v.children[0],
 810                              parent->v.children[1]);
 811         } else {
 812                 /*
 813                  * We're deleting the root of a snapshot tree: update the
 814                  * snapshot_tree entry to point to the new root, or delete it if
 815                  * this is the last snapshot ID in this tree:
 816                  */
 817                 struct bkey_i_snapshot_tree *s_t;
 818
 819                 BUG_ON(s.v->children[1]);
 820
 821                 s_t = bch2_bkey_get_mut_typed(trans, &tree_iter,
 822                                 BTREE_ID_snapshot_trees, POS(0, le32_to_cpu(s.v->tree)),
 823                                 0, snapshot_tree);
 824                 ret = PTR_ERR_OR_ZERO(s_t);
 825                 if (ret)
 826                         goto err;
 827
 828                 if (s.v->children[0]) {
 829                         s_t->v.root_snapshot = cpu_to_le32(s.v->children[0]);
 830                 } else {
 831                         s_t->k.type = KEY_TYPE_deleted;
 832                         set_bkey_val_u64s(&s_t->k, 0);
 833                 }
 834         }
 835
 836         ret = bch2_btree_delete_at(trans, &iter, 0);
 837 err:
 838         bch2_trans_iter_exit(trans, &tree_iter);
 839         bch2_trans_iter_exit(trans, &p_iter);
 840         bch2_trans_iter_exit(trans, &iter);
 841         return ret;
 842 }
 843
 844 static int create_snapids(struct btree_trans *trans, u32 parent, u32 tree,
 845                           u32 *new_snapids,
 846                           u32 *snapshot_subvols,
 847                           unsigned nr_snapids)
 848 {
 849         struct btree_iter iter;
 850         struct bkey_i_snapshot *n;
 851         struct bkey_s_c k;
 852         unsigned i;
 853         int ret;
 854
 855         bch2_trans_iter_init(trans, &iter, BTREE_ID_snapshots,
 856                              POS_MIN, BTREE_ITER_INTENT);
 857         k = bch2_btree_iter_peek(&iter);
 858         ret = bkey_err(k);
 859         if (ret)
 860                 goto err;
 861
 862         for (i = 0; i < nr_snapids; i++) {
 863                 k = bch2_btree_iter_prev_slot(&iter);
 864                 ret = bkey_err(k);
 865                 if (ret)
 866                         goto err;
 867
 868                 if (!k.k || !k.k->p.offset) {
 869                         ret = -BCH_ERR_ENOSPC_snapshot_create;
 870                         goto err;
 871                 }
 872
 873                 n = bch2_bkey_alloc(trans, &iter, 0, snapshot);
 874                 ret = PTR_ERR_OR_ZERO(n);
 875                 if (ret)
 876                         goto err;
 877
 878                 n->v.flags      = 0;
 879                 n->v.parent     = cpu_to_le32(parent);
 880                 n->v.subvol     = cpu_to_le32(snapshot_subvols[i]);
 881                 n->v.tree       = cpu_to_le32(tree);
 882                 SET_BCH_SNAPSHOT_SUBVOL(&n->v, true);
 883
 884                 ret = bch2_mark_snapshot(trans, BTREE_ID_snapshots, 0,
 885                                          bkey_s_c_null, bkey_i_to_s_c(&n->k_i), 0);
 886                 if (ret)
 887                         goto err;
 888
 889                 new_snapids[i]  = iter.pos.offset;
 890         }
 891 err:
 892         bch2_trans_iter_exit(trans, &iter);
 893         return ret;
 894 }
 895
 896 /*
 897  * Create new snapshot IDs as children of an existing snapshot ID:
 898  */
 899 static int bch2_snapshot_node_create_children(struct btree_trans *trans, u32 parent,
 900                               u32 *new_snapids,
 901                               u32 *snapshot_subvols,
 902                               unsigned nr_snapids)
 903 {
 904         struct btree_iter iter;
 905         struct bkey_i_snapshot *n_parent;
 906         int ret = 0;
 907
 908         n_parent = bch2_bkey_get_mut_typed(trans, &iter,
 909                         BTREE_ID_snapshots, POS(0, parent),
 910                         0, snapshot);
 911         ret = PTR_ERR_OR_ZERO(n_parent);
 912         if (unlikely(ret)) {
 913                 if (bch2_err_matches(ret, ENOENT))
 914                         bch_err(trans->c, "snapshot %u not found", parent);
 915                 return ret;
 916         }
 917
 918         if (n_parent->v.children[0] || n_parent->v.children[1]) {
 919                 bch_err(trans->c, "Trying to add child snapshot nodes to parent that already has children");
 920                 ret = -EINVAL;
 921                 goto err;
 922         }
 923
 924         ret = create_snapids(trans, parent, le32_to_cpu(n_parent->v.tree),
 925                              new_snapids, snapshot_subvols, nr_snapids);
 926         if (ret)
 927                 goto err;
 928
 929         n_parent->v.children[0] = cpu_to_le32(new_snapids[0]);
 930         n_parent->v.children[1] = cpu_to_le32(new_snapids[1]);
 931         n_parent->v.subvol = 0;
 932         SET_BCH_SNAPSHOT_SUBVOL(&n_parent->v, false);
 933 err:
 934         bch2_trans_iter_exit(trans, &iter);
 935         return ret;
 936 }
 937
 938 /*
 939  * Create a snapshot node that is the root of a new tree:
 940  */
 941 static int bch2_snapshot_node_create_tree(struct btree_trans *trans,
 942                               u32 *new_snapids,
 943                               u32 *snapshot_subvols,
 944                               unsigned nr_snapids)
 945 {
 946         struct bkey_i_snapshot_tree *n_tree;
 947         int ret;
 948
 949         n_tree = __snapshot_tree_create(trans);
 950         ret =   PTR_ERR_OR_ZERO(n_tree) ?:
 951                 create_snapids(trans, 0, n_tree->k.p.offset,
 952                              new_snapids, snapshot_subvols, nr_snapids);
 953         if (ret)
 954                 return ret;
 955
 956         n_tree->v.master_subvol = cpu_to_le32(snapshot_subvols[0]);
 957         n_tree->v.root_snapshot = cpu_to_le32(new_snapids[0]);
 958         return 0;
 959 }
 960
 961 int bch2_snapshot_node_create(struct btree_trans *trans, u32 parent,
 962                               u32 *new_snapids,
 963                               u32 *snapshot_subvols,
 964                               unsigned nr_snapids)
 965 {
 966         BUG_ON((parent == 0) != (nr_snapids == 1));
 967         BUG_ON((parent != 0) != (nr_snapids == 2));
 968
 969         return parent
 970                 ? bch2_snapshot_node_create_children(trans, parent,
 971                                 new_snapids, snapshot_subvols, nr_snapids)
 972                 : bch2_snapshot_node_create_tree(trans,
 973                                 new_snapids, snapshot_subvols, nr_snapids);
 974
 975 }
 976
 977 static int snapshot_delete_key(struct btree_trans *trans,
 978                                struct btree_iter *iter,
 979                                struct bkey_s_c k,
 980                                snapshot_id_list *deleted,
 981                                snapshot_id_list *equiv_seen,
 982                                struct bpos *last_pos)
 983 {
 984         struct bch_fs *c = trans->c;
 985         u32 equiv = snapshot_t(c, k.k->p.snapshot)->equiv;
 986
 987         if (!bkey_eq(k.k->p, *last_pos))
 988                 equiv_seen->nr = 0;
 989         *last_pos = k.k->p;
 990
 991         if (snapshot_list_has_id(deleted, k.k->p.snapshot) ||
 992             snapshot_list_has_id(equiv_seen, equiv)) {
 993                 return bch2_btree_delete_at(trans, iter,
 994                                             BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE);
 995         } else {
 996                 return snapshot_list_add(c, equiv_seen, equiv);
 997         }
 998 }
 999
1000 static int bch2_delete_redundant_snapshot(struct btree_trans *trans, struct btree_iter *iter,
1001                                           struct bkey_s_c k)
1002 {
1003         struct bkey_s_c_snapshot snap;
1004         u32 children[2];
1005         int ret;
1006
1007         if (k.k->type != KEY_TYPE_snapshot)
1008                 return 0;
1009
1010         snap = bkey_s_c_to_snapshot(k);
1011         if (BCH_SNAPSHOT_DELETED(snap.v) ||
1012             BCH_SNAPSHOT_SUBVOL(snap.v))
1013                 return 0;
1014
1015         children[0] = le32_to_cpu(snap.v->children[0]);
1016         children[1] = le32_to_cpu(snap.v->children[1]);
1017
1018         ret   = snapshot_live(trans, children[0]) ?:
1019                 snapshot_live(trans, children[1]);
1020         if (ret < 0)
1021                 return ret;
1022
1023         if (!ret)
1024                 return bch2_snapshot_node_set_deleted(trans, k.k->p.offset);
1025         return 0;
1026 }
1027
1028 int bch2_delete_dead_snapshots(struct bch_fs *c)
1029 {
1030         struct btree_trans trans;
1031         struct btree_iter iter;
1032         struct bkey_s_c k;
1033         struct bkey_s_c_snapshot snap;
1034         snapshot_id_list deleted = { 0 };
1035         u32 i, id;
1036         int ret = 0;
1037
1038         if (!test_bit(BCH_FS_HAVE_DELETED_SNAPSHOTS, &c->flags))
1039                 return 0;
1040
1041         if (!test_bit(BCH_FS_STARTED, &c->flags)) {
1042                 ret = bch2_fs_read_write_early(c);
1043                 if (ret) {
1044                         bch_err(c, "error deleleting dead snapshots: error going rw: %s", bch2_err_str(ret));
1045                         return ret;
1046                 }
1047         }
1048
1049         bch2_trans_init(&trans, c, 0, 0);
1050
1051         /*
1052          * For every snapshot node: If we have no live children and it's not
1053          * pointed to by a subvolume, delete it:
1054          */
1055         ret = for_each_btree_key_commit(&trans, iter, BTREE_ID_snapshots,
1056                         POS_MIN, 0, k,
1057                         NULL, NULL, 0,
1058                 bch2_delete_redundant_snapshot(&trans, &iter, k));
1059         if (ret) {
1060                 bch_err(c, "error deleting redundant snapshots: %s", bch2_err_str(ret));
1061                 goto err;
1062         }
1063
1064         for_each_btree_key2(&trans, iter, BTREE_ID_snapshots,
1065                            POS_MIN, 0, k,
1066                 bch2_snapshot_set_equiv(&trans, k));
1067         if (ret) {
1068                 bch_err(c, "error in bch2_snapshots_set_equiv: %s", bch2_err_str(ret));
1069                 goto err;
1070         }
1071
1072         for_each_btree_key(&trans, iter, BTREE_ID_snapshots,
1073                            POS_MIN, 0, k, ret) {
1074                 if (k.k->type != KEY_TYPE_snapshot)
1075                         continue;
1076
1077                 snap = bkey_s_c_to_snapshot(k);
1078                 if (BCH_SNAPSHOT_DELETED(snap.v)) {
1079                         ret = snapshot_list_add(c, &deleted, k.k->p.offset);
1080                         if (ret)
1081                                 break;
1082                 }
1083         }
1084         bch2_trans_iter_exit(&trans, &iter);
1085
1086         if (ret) {
1087                 bch_err(c, "error walking snapshots: %s", bch2_err_str(ret));
1088                 goto err;
1089         }
1090
1091         for (id = 0; id < BTREE_ID_NR; id++) {
1092                 struct bpos last_pos = POS_MIN;
1093                 snapshot_id_list equiv_seen = { 0 };
1094
1095                 if (!btree_type_has_snapshots(id))
1096                         continue;
1097
1098                 ret = for_each_btree_key_commit(&trans, iter,
1099                                 id, POS_MIN,
1100                                 BTREE_ITER_PREFETCH|BTREE_ITER_ALL_SNAPSHOTS, k,
1101                                 NULL, NULL, BTREE_INSERT_NOFAIL,
1102                         snapshot_delete_key(&trans, &iter, k, &deleted, &equiv_seen, &last_pos));
1103
1104                 darray_exit(&equiv_seen);
1105
1106                 if (ret) {
1107                         bch_err(c, "error deleting snapshot keys: %s", bch2_err_str(ret));
1108                         goto err;
1109                 }
1110         }
1111
1112         for (i = 0; i < deleted.nr; i++) {
1113                 ret = commit_do(&trans, NULL, NULL, 0,
1114                         bch2_snapshot_node_delete(&trans, deleted.data[i]));
1115                 if (ret) {
1116                         bch_err(c, "error deleting snapshot %u: %s",
1117                                 deleted.data[i], bch2_err_str(ret));
1118                         goto err;
1119                 }
1120         }
1121
1122         clear_bit(BCH_FS_HAVE_DELETED_SNAPSHOTS, &c->flags);
1123 err:
1124         darray_exit(&deleted);
1125         bch2_trans_exit(&trans);
1126         return ret;
1127 }
1128
1129 static void bch2_delete_dead_snapshots_work(struct work_struct *work)
1130 {
1131         struct bch_fs *c = container_of(work, struct bch_fs, snapshot_delete_work);
1132
1133         bch2_delete_dead_snapshots(c);
1134         bch2_write_ref_put(c, BCH_WRITE_REF_delete_dead_snapshots);
1135 }
1136
1137 void bch2_delete_dead_snapshots_async(struct bch_fs *c)
1138 {
1139         if (bch2_write_ref_tryget(c, BCH_WRITE_REF_delete_dead_snapshots) &&
1140             !queue_work(c->write_ref_wq, &c->snapshot_delete_work))
1141                 bch2_write_ref_put(c, BCH_WRITE_REF_delete_dead_snapshots);
1142 }
1143
1144 static int bch2_delete_dead_snapshots_hook(struct btree_trans *trans,
1145                                            struct btree_trans_commit_hook *h)
1146 {
1147         struct bch_fs *c = trans->c;
1148
1149         set_bit(BCH_FS_HAVE_DELETED_SNAPSHOTS, &c->flags);
1150
1151         if (!test_bit(BCH_FS_FSCK_DONE, &c->flags))
1152                 return 0;
1153
1154         bch2_delete_dead_snapshots_async(c);
1155         return 0;
1156 }
1157
1158 /* Subvolumes: */
1159
1160 int bch2_subvolume_invalid(const struct bch_fs *c, struct bkey_s_c k,
1161                            unsigned flags, struct printbuf *err)
1162 {
1163         if (bkey_lt(k.k->p, SUBVOL_POS_MIN) ||
1164             bkey_gt(k.k->p, SUBVOL_POS_MAX)) {
1165                 prt_printf(err, "invalid pos");
1166                 return -BCH_ERR_invalid_bkey;
1167         }
1168
1169         return 0;
1170 }
1171
1172 void bch2_subvolume_to_text(struct printbuf *out, struct bch_fs *c,
1173                             struct bkey_s_c k)
1174 {
1175         struct bkey_s_c_subvolume s = bkey_s_c_to_subvolume(k);
1176
1177         prt_printf(out, "root %llu snapshot id %u",
1178                    le64_to_cpu(s.v->inode),
1179                    le32_to_cpu(s.v->snapshot));
1180
1181         if (bkey_val_bytes(s.k) > offsetof(struct bch_subvolume, parent))
1182                 prt_printf(out, " parent %u", le32_to_cpu(s.v->parent));
1183 }
1184
1185 static __always_inline int
1186 bch2_subvolume_get_inlined(struct btree_trans *trans, unsigned subvol,
1187                            bool inconsistent_if_not_found,
1188                            int iter_flags,
1189                            struct bch_subvolume *s)
1190 {
1191         int ret = bch2_bkey_get_val_typed(trans, BTREE_ID_subvolumes, POS(0, subvol),
1192                                           iter_flags, subvolume, s);
1193         bch2_fs_inconsistent_on(bch2_err_matches(ret, ENOENT) &&
1194                                 inconsistent_if_not_found,
1195                                 trans->c, "missing subvolume %u", subvol);
1196         return ret;
1197 }
1198
1199 int bch2_subvolume_get(struct btree_trans *trans, unsigned subvol,
1200                        bool inconsistent_if_not_found,
1201                        int iter_flags,
1202                        struct bch_subvolume *s)
1203 {
1204         return bch2_subvolume_get_inlined(trans, subvol, inconsistent_if_not_found, iter_flags, s);
1205 }
1206
1207 int bch2_snapshot_get_subvol(struct btree_trans *trans, u32 snapshot,
1208                              struct bch_subvolume *subvol)
1209 {
1210         struct bch_snapshot snap;
1211
1212         return  snapshot_lookup(trans, snapshot, &snap) ?:
1213                 bch2_subvolume_get(trans, le32_to_cpu(snap.subvol), true, 0, subvol);
1214 }
1215
1216 int bch2_subvolume_get_snapshot(struct btree_trans *trans, u32 subvol,
1217                                 u32 *snapid)
1218 {
1219         struct btree_iter iter;
1220         struct bkey_s_c k;
1221         int ret;
1222
1223         k = bch2_bkey_get_iter(trans, &iter, BTREE_ID_subvolumes, POS(0, subvol),
1224                                BTREE_ITER_CACHED|
1225                                BTREE_ITER_WITH_UPDATES);
1226         ret = bkey_err(k) ?: k.k->type == KEY_TYPE_subvolume ? 0 : -BCH_ERR_ENOENT_subvolume;
1227
1228         if (likely(!ret))
1229                 *snapid = le32_to_cpu(bkey_s_c_to_subvolume(k).v->snapshot);
1230         else if (bch2_err_matches(ret, ENOENT))
1231                 bch2_fs_inconsistent(trans->c, "missing subvolume %u", subvol);
1232         bch2_trans_iter_exit(trans, &iter);
1233         return ret;
1234 }
1235
1236 static int bch2_subvolume_reparent(struct btree_trans *trans,
1237                                    struct btree_iter *iter,
1238                                    struct bkey_s_c k,
1239                                    u32 old_parent, u32 new_parent)
1240 {
1241         struct bkey_i_subvolume *s;
1242         int ret;
1243
1244         if (k.k->type != KEY_TYPE_subvolume)
1245                 return 0;
1246
1247         if (bkey_val_bytes(k.k) > offsetof(struct bch_subvolume, parent) &&
1248             le32_to_cpu(bkey_s_c_to_subvolume(k).v->parent) != old_parent)
1249                 return 0;
1250
1251         s = bch2_bkey_make_mut_typed(trans, iter, k, 0, subvolume);
1252         ret = PTR_ERR_OR_ZERO(s);
1253         if (ret)
1254                 return ret;
1255
1256         s->v.parent = cpu_to_le32(new_parent);
1257         return 0;
1258 }
1259
1260 /*
1261  * Scan for subvolumes with parent @subvolid_to_delete, reparent:
1262  */
1263 static int bch2_subvolumes_reparent(struct btree_trans *trans, u32 subvolid_to_delete)
1264 {
1265         struct btree_iter iter;
1266         struct bkey_s_c k;
1267         struct bch_subvolume s;
1268
1269         return lockrestart_do(trans,
1270                         bch2_subvolume_get(trans, subvolid_to_delete, true,
1271                                    BTREE_ITER_CACHED, &s)) ?:
1272                 for_each_btree_key_commit(trans, iter,
1273                                 BTREE_ID_subvolumes, POS_MIN, BTREE_ITER_PREFETCH, k,
1274                                 NULL, NULL, BTREE_INSERT_NOFAIL,
1275                         bch2_subvolume_reparent(trans, &iter, k,
1276                                         subvolid_to_delete, le32_to_cpu(s.parent)));
1277 }
1278
1279 /*
1280  * Delete subvolume, mark snapshot ID as deleted, queue up snapshot
1281  * deletion/cleanup:
1282  */
1283 static int __bch2_subvolume_delete(struct btree_trans *trans, u32 subvolid)
1284 {
1285         struct btree_iter iter;
1286         struct bkey_s_c_subvolume subvol;
1287         struct btree_trans_commit_hook *h;
1288         u32 snapid;
1289         int ret = 0;
1290
1291         subvol = bch2_bkey_get_iter_typed(trans, &iter,
1292                                 BTREE_ID_subvolumes, POS(0, subvolid),
1293                                 BTREE_ITER_CACHED|BTREE_ITER_INTENT,
1294                                 subvolume);
1295         ret = bkey_err(subvol);
1296         bch2_fs_inconsistent_on(bch2_err_matches(ret, ENOENT), trans->c,
1297                                 "missing subvolume %u", subvolid);
1298         if (ret)
1299                 return ret;
1300
1301         snapid = le32_to_cpu(subvol.v->snapshot);
1302
1303         ret = bch2_btree_delete_at(trans, &iter, 0);
1304         if (ret)
1305                 goto err;
1306
1307         ret = bch2_snapshot_node_set_deleted(trans, snapid);
1308         if (ret)
1309                 goto err;
1310
1311         h = bch2_trans_kmalloc(trans, sizeof(*h));
1312         ret = PTR_ERR_OR_ZERO(h);
1313         if (ret)
1314                 goto err;
1315
1316         h->fn = bch2_delete_dead_snapshots_hook;
1317         bch2_trans_commit_hook(trans, h);
1318 err:
1319         bch2_trans_iter_exit(trans, &iter);
1320         return ret;
1321 }
1322
1323 static int bch2_subvolume_delete(struct btree_trans *trans, u32 subvolid)
1324 {
1325         return bch2_subvolumes_reparent(trans, subvolid) ?:
1326                 commit_do(trans, NULL, NULL, BTREE_INSERT_NOFAIL,
1327                           __bch2_subvolume_delete(trans, subvolid));
1328 }
1329
1330 void bch2_subvolume_wait_for_pagecache_and_delete(struct work_struct *work)
1331 {
1332         struct bch_fs *c = container_of(work, struct bch_fs,
1333                                 snapshot_wait_for_pagecache_and_delete_work);
1334         snapshot_id_list s;
1335         u32 *id;
1336         int ret = 0;
1337
1338         while (!ret) {
1339                 mutex_lock(&c->snapshots_unlinked_lock);
1340                 s = c->snapshots_unlinked;
1341                 darray_init(&c->snapshots_unlinked);
1342                 mutex_unlock(&c->snapshots_unlinked_lock);
1343
1344                 if (!s.nr)
1345                         break;
1346
1347                 bch2_evict_subvolume_inodes(c, &s);
1348
1349                 for (id = s.data; id < s.data + s.nr; id++) {
1350                         ret = bch2_trans_run(c, bch2_subvolume_delete(&trans, *id));
1351                         if (ret) {
1352                                 bch_err(c, "error deleting subvolume %u: %s", *id, bch2_err_str(ret));
1353                                 break;
1354                         }
1355                 }
1356
1357                 darray_exit(&s);
1358         }
1359
1360         bch2_write_ref_put(c, BCH_WRITE_REF_snapshot_delete_pagecache);
1361 }
1362
1363 struct subvolume_unlink_hook {
1364         struct btree_trans_commit_hook  h;
1365         u32                             subvol;
1366 };
1367
1368 int bch2_subvolume_wait_for_pagecache_and_delete_hook(struct btree_trans *trans,
1369                                                       struct btree_trans_commit_hook *_h)
1370 {
1371         struct subvolume_unlink_hook *h = container_of(_h, struct subvolume_unlink_hook, h);
1372         struct bch_fs *c = trans->c;
1373         int ret = 0;
1374
1375         mutex_lock(&c->snapshots_unlinked_lock);
1376         if (!snapshot_list_has_id(&c->snapshots_unlinked, h->subvol))
1377                 ret = snapshot_list_add(c, &c->snapshots_unlinked, h->subvol);
1378         mutex_unlock(&c->snapshots_unlinked_lock);
1379
1380         if (ret)
1381                 return ret;
1382
1383         if (!bch2_write_ref_tryget(c, BCH_WRITE_REF_snapshot_delete_pagecache))
1384                 return -EROFS;
1385
1386         if (!queue_work(c->write_ref_wq, &c->snapshot_wait_for_pagecache_and_delete_work))
1387                 bch2_write_ref_put(c, BCH_WRITE_REF_snapshot_delete_pagecache);
1388         return 0;
1389 }
1390
1391 int bch2_subvolume_unlink(struct btree_trans *trans, u32 subvolid)
1392 {
1393         struct btree_iter iter;
1394         struct bkey_i_subvolume *n;
1395         struct subvolume_unlink_hook *h;
1396         int ret = 0;
1397
1398         h = bch2_trans_kmalloc(trans, sizeof(*h));
1399         ret = PTR_ERR_OR_ZERO(h);
1400         if (ret)
1401                 return ret;
1402
1403         h->h.fn         = bch2_subvolume_wait_for_pagecache_and_delete_hook;
1404         h->subvol       = subvolid;
1405         bch2_trans_commit_hook(trans, &h->h);
1406
1407         n = bch2_bkey_get_mut_typed(trans, &iter,
1408                         BTREE_ID_subvolumes, POS(0, subvolid),
1409                         BTREE_ITER_CACHED, subvolume);
1410         ret = PTR_ERR_OR_ZERO(n);
1411         if (unlikely(ret)) {
1412                 bch2_fs_inconsistent_on(bch2_err_matches(ret, ENOENT), trans->c,
1413                                         "missing subvolume %u", subvolid);
1414                 return ret;
1415         }
1416
1417         SET_BCH_SUBVOLUME_UNLINKED(&n->v, true);
1418         bch2_trans_iter_exit(trans, &iter);
1419         return ret;
1420 }
1421
1422 int bch2_subvolume_create(struct btree_trans *trans, u64 inode,
1423                           u32 src_subvolid,
1424                           u32 *new_subvolid,
1425                           u32 *new_snapshotid,
1426                           bool ro)
1427 {
1428         struct bch_fs *c = trans->c;
1429         struct btree_iter dst_iter, src_iter = (struct btree_iter) { NULL };
1430         struct bkey_i_subvolume *new_subvol = NULL;
1431         struct bkey_i_subvolume *src_subvol = NULL;
1432         u32 parent = 0, new_nodes[2], snapshot_subvols[2];
1433         int ret = 0;
1434
1435         ret = bch2_bkey_get_empty_slot(trans, &dst_iter,
1436                                 BTREE_ID_subvolumes, POS(0, U32_MAX));
1437         if (ret == -BCH_ERR_ENOSPC_btree_slot)
1438                 ret = -BCH_ERR_ENOSPC_subvolume_create;
1439         if (ret)
1440                 return ret;
1441
1442         snapshot_subvols[0] = dst_iter.pos.offset;
1443         snapshot_subvols[1] = src_subvolid;
1444
1445         if (src_subvolid) {
1446                 /* Creating a snapshot: */
1447
1448                 src_subvol = bch2_bkey_get_mut_typed(trans, &src_iter,
1449                                 BTREE_ID_subvolumes, POS(0, src_subvolid),
1450                                 BTREE_ITER_CACHED, subvolume);
1451                 ret = PTR_ERR_OR_ZERO(src_subvol);
1452                 if (unlikely(ret)) {
1453                         bch2_fs_inconsistent_on(bch2_err_matches(ret, ENOENT), c,
1454                                                 "subvolume %u not found", src_subvolid);
1455                         goto err;
1456                 }
1457
1458                 parent = le32_to_cpu(src_subvol->v.snapshot);
1459         }
1460
1461         ret = bch2_snapshot_node_create(trans, parent, new_nodes,
1462                                         snapshot_subvols,
1463                                         src_subvolid ? 2 : 1);
1464         if (ret)
1465                 goto err;
1466
1467         if (src_subvolid) {
1468                 src_subvol->v.snapshot = cpu_to_le32(new_nodes[1]);
1469                 ret = bch2_trans_update(trans, &src_iter, &src_subvol->k_i, 0);
1470                 if (ret)
1471                         goto err;
1472         }
1473
1474         new_subvol = bch2_bkey_alloc(trans, &dst_iter, 0, subvolume);
1475         ret = PTR_ERR_OR_ZERO(new_subvol);
1476         if (ret)
1477                 goto err;
1478
1479         new_subvol->v.flags     = 0;
1480         new_subvol->v.snapshot  = cpu_to_le32(new_nodes[0]);
1481         new_subvol->v.inode     = cpu_to_le64(inode);
1482         new_subvol->v.parent    = cpu_to_le32(src_subvolid);
1483         new_subvol->v.otime.lo  = cpu_to_le64(bch2_current_time(c));
1484         new_subvol->v.otime.hi  = 0;
1485
1486         SET_BCH_SUBVOLUME_RO(&new_subvol->v, ro);
1487         SET_BCH_SUBVOLUME_SNAP(&new_subvol->v, src_subvolid != 0);
1488
1489         *new_subvolid   = new_subvol->k.p.offset;
1490         *new_snapshotid = new_nodes[0];
1491 err:
1492         bch2_trans_iter_exit(trans, &src_iter);
1493         bch2_trans_iter_exit(trans, &dst_iter);
1494         return ret;
1495 }
1496
1497 int bch2_fs_subvolumes_init(struct bch_fs *c)
1498 {
1499         INIT_WORK(&c->snapshot_delete_work, bch2_delete_dead_snapshots_work);
1500         INIT_WORK(&c->snapshot_wait_for_pagecache_and_delete_work,
1501                   bch2_subvolume_wait_for_pagecache_and_delete);
1502         mutex_init(&c->snapshots_unlinked_lock);
1503         return 0;
1504 }