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 /* Snapshot tree: */
  12
  13 void bch2_snapshot_to_text(struct printbuf *out, struct bch_fs *c,
  14                            struct bkey_s_c k)
  15 {
  16         struct bkey_s_c_snapshot s = bkey_s_c_to_snapshot(k);
  17
  18         prt_printf(out, "is_subvol %llu deleted %llu parent %10u children %10u %10u subvol %u",
  19                BCH_SNAPSHOT_SUBVOL(s.v),
  20                BCH_SNAPSHOT_DELETED(s.v),
  21                le32_to_cpu(s.v->parent),
  22                le32_to_cpu(s.v->children[0]),
  23                le32_to_cpu(s.v->children[1]),
  24                le32_to_cpu(s.v->subvol));
  25 }
  26
  27 int bch2_snapshot_invalid(const struct bch_fs *c, struct bkey_s_c k,
  28                           unsigned flags, struct printbuf *err)
  29 {
  30         struct bkey_s_c_snapshot s;
  31         u32 i, id;
  32
  33         if (bkey_gt(k.k->p, POS(0, U32_MAX)) ||
  34             bkey_lt(k.k->p, POS(0, 1))) {
  35                 prt_printf(err, "bad pos");
  36                 return -BCH_ERR_invalid_bkey;
  37         }
  38
  39         if (bkey_val_bytes(k.k) != sizeof(struct bch_snapshot)) {
  40                 prt_printf(err, "bad val size (%zu != %zu)",
  41                        bkey_val_bytes(k.k), sizeof(struct bch_snapshot));
  42                 return -BCH_ERR_invalid_bkey;
  43         }
  44
  45         s = bkey_s_c_to_snapshot(k);
  46
  47         id = le32_to_cpu(s.v->parent);
  48         if (id && id <= k.k->p.offset) {
  49                 prt_printf(err, "bad parent node (%u <= %llu)",
  50                        id, k.k->p.offset);
  51                 return -BCH_ERR_invalid_bkey;
  52         }
  53
  54         if (le32_to_cpu(s.v->children[0]) < le32_to_cpu(s.v->children[1])) {
  55                 prt_printf(err, "children not normalized");
  56                 return -BCH_ERR_invalid_bkey;
  57         }
  58
  59         if (s.v->children[0] &&
  60             s.v->children[0] == s.v->children[1]) {
  61                 prt_printf(err, "duplicate child nodes");
  62                 return -BCH_ERR_invalid_bkey;
  63         }
  64
  65         for (i = 0; i < 2; i++) {
  66                 id = le32_to_cpu(s.v->children[i]);
  67
  68                 if (id >= k.k->p.offset) {
  69                         prt_printf(err, "bad child node (%u >= %llu)",
  70                                id, k.k->p.offset);
  71                         return -BCH_ERR_invalid_bkey;
  72                 }
  73         }
  74
  75         return 0;
  76 }
  77
  78 int bch2_mark_snapshot(struct btree_trans *trans,
  79                        struct bkey_s_c old, struct bkey_s_c new,
  80                        unsigned flags)
  81 {
  82         struct bch_fs *c = trans->c;
  83         struct snapshot_t *t;
  84
  85         t = genradix_ptr_alloc(&c->snapshots,
  86                                U32_MAX - new.k->p.offset,
  87                                GFP_KERNEL);
  88         if (!t)
  89                 return -ENOMEM;
  90
  91         if (new.k->type == KEY_TYPE_snapshot) {
  92                 struct bkey_s_c_snapshot s = bkey_s_c_to_snapshot(new);
  93
  94                 t->parent       = le32_to_cpu(s.v->parent);
  95                 t->children[0]  = le32_to_cpu(s.v->children[0]);
  96                 t->children[1]  = le32_to_cpu(s.v->children[1]);
  97                 t->subvol       = BCH_SNAPSHOT_SUBVOL(s.v) ? le32_to_cpu(s.v->subvol) : 0;
  98         } else {
  99                 t->parent       = 0;
 100                 t->children[0]  = 0;
 101                 t->children[1]  = 0;
 102                 t->subvol       = 0;
 103         }
 104
 105         return 0;
 106 }
 107
 108 static int snapshot_lookup(struct btree_trans *trans, u32 id,
 109                            struct bch_snapshot *s)
 110 {
 111         struct btree_iter iter;
 112         struct bkey_s_c k;
 113         int ret;
 114
 115         bch2_trans_iter_init(trans, &iter, BTREE_ID_snapshots, POS(0, id),
 116                              BTREE_ITER_WITH_UPDATES);
 117         k = bch2_btree_iter_peek_slot(&iter);
 118         ret = bkey_err(k) ?: k.k->type == KEY_TYPE_snapshot ? 0 : -ENOENT;
 119
 120         if (!ret)
 121                 *s = *bkey_s_c_to_snapshot(k).v;
 122
 123         bch2_trans_iter_exit(trans, &iter);
 124         return ret;
 125 }
 126
 127 static int snapshot_live(struct btree_trans *trans, u32 id)
 128 {
 129         struct bch_snapshot v;
 130         int ret;
 131
 132         if (!id)
 133                 return 0;
 134
 135         ret = snapshot_lookup(trans, id, &v);
 136         if (ret == -ENOENT)
 137                 bch_err(trans->c, "snapshot node %u not found", id);
 138         if (ret)
 139                 return ret;
 140
 141         return !BCH_SNAPSHOT_DELETED(&v);
 142 }
 143
 144 static int bch2_snapshot_set_equiv(struct btree_trans *trans, struct bkey_s_c k)
 145 {
 146         struct bch_fs *c = trans->c;
 147         unsigned i, nr_live = 0, live_idx = 0;
 148         struct bkey_s_c_snapshot snap;
 149         u32 id = k.k->p.offset, child[2];
 150
 151         if (k.k->type != KEY_TYPE_snapshot)
 152                 return 0;
 153
 154         snap = bkey_s_c_to_snapshot(k);
 155
 156         child[0] = le32_to_cpu(snap.v->children[0]);
 157         child[1] = le32_to_cpu(snap.v->children[1]);
 158
 159         for (i = 0; i < 2; i++) {
 160                 int ret = snapshot_live(trans, child[i]);
 161
 162                 if (ret < 0)
 163                         return ret;
 164
 165                 if (ret)
 166                         live_idx = i;
 167                 nr_live += ret;
 168         }
 169
 170         snapshot_t(c, id)->equiv = nr_live == 1
 171                 ? snapshot_t(c, child[live_idx])->equiv
 172                 : id;
 173         return 0;
 174 }
 175
 176 /* fsck: */
 177 static int check_snapshot(struct btree_trans *trans,
 178                           struct btree_iter *iter,
 179                           struct bkey_s_c k)
 180 {
 181         struct bch_fs *c = trans->c;
 182         struct bkey_s_c_snapshot s;
 183         struct bch_subvolume subvol;
 184         struct bch_snapshot v;
 185         struct printbuf buf = PRINTBUF;
 186         bool should_have_subvol;
 187         u32 i, id;
 188         int ret = 0;
 189
 190         if (k.k->type != KEY_TYPE_snapshot)
 191                 return 0;
 192
 193         s = bkey_s_c_to_snapshot(k);
 194         id = le32_to_cpu(s.v->parent);
 195         if (id) {
 196                 ret = snapshot_lookup(trans, id, &v);
 197                 if (ret == -ENOENT)
 198                         bch_err(c, "snapshot with nonexistent parent:\n  %s",
 199                                 (bch2_bkey_val_to_text(&buf, c, s.s_c), buf.buf));
 200                 if (ret)
 201                         goto err;
 202
 203                 if (le32_to_cpu(v.children[0]) != s.k->p.offset &&
 204                     le32_to_cpu(v.children[1]) != s.k->p.offset) {
 205                         bch_err(c, "snapshot parent %u missing pointer to child %llu",
 206                                 id, s.k->p.offset);
 207                         ret = -EINVAL;
 208                         goto err;
 209                 }
 210         }
 211
 212         for (i = 0; i < 2 && s.v->children[i]; i++) {
 213                 id = le32_to_cpu(s.v->children[i]);
 214
 215                 ret = snapshot_lookup(trans, id, &v);
 216                 if (ret == -ENOENT)
 217                         bch_err(c, "snapshot node %llu has nonexistent child %u",
 218                                 s.k->p.offset, id);
 219                 if (ret)
 220                         goto err;
 221
 222                 if (le32_to_cpu(v.parent) != s.k->p.offset) {
 223                         bch_err(c, "snapshot child %u has wrong parent (got %u should be %llu)",
 224                                 id, le32_to_cpu(v.parent), s.k->p.offset);
 225                         ret = -EINVAL;
 226                         goto err;
 227                 }
 228         }
 229
 230         should_have_subvol = BCH_SNAPSHOT_SUBVOL(s.v) &&
 231                 !BCH_SNAPSHOT_DELETED(s.v);
 232
 233         if (should_have_subvol) {
 234                 id = le32_to_cpu(s.v->subvol);
 235                 ret = bch2_subvolume_get(trans, id, 0, false, &subvol);
 236                 if (ret == -ENOENT)
 237                         bch_err(c, "snapshot points to nonexistent subvolume:\n  %s",
 238                                 (bch2_bkey_val_to_text(&buf, c, s.s_c), buf.buf));
 239                 if (ret)
 240                         goto err;
 241
 242                 if (BCH_SNAPSHOT_SUBVOL(s.v) != (le32_to_cpu(subvol.snapshot) == s.k->p.offset)) {
 243                         bch_err(c, "snapshot node %llu has wrong BCH_SNAPSHOT_SUBVOL",
 244                                 s.k->p.offset);
 245                         ret = -EINVAL;
 246                         goto err;
 247                 }
 248         } else {
 249                 if (fsck_err_on(s.v->subvol, c, "snapshot should not point to subvol:\n  %s",
 250                                 (bch2_bkey_val_to_text(&buf, c, s.s_c), buf.buf))) {
 251                         struct bkey_i_snapshot *u = bch2_trans_kmalloc(trans, sizeof(*u));
 252
 253                         ret = PTR_ERR_OR_ZERO(u);
 254                         if (ret)
 255                                 goto err;
 256
 257                         bkey_reassemble(&u->k_i, s.s_c);
 258                         u->v.subvol = 0;
 259                         ret = bch2_trans_update(trans, iter, &u->k_i, 0);
 260                         if (ret)
 261                                 goto err;
 262                 }
 263         }
 264
 265         if (BCH_SNAPSHOT_DELETED(s.v))
 266                 set_bit(BCH_FS_HAVE_DELETED_SNAPSHOTS, &c->flags);
 267 err:
 268 fsck_err:
 269         printbuf_exit(&buf);
 270         return ret;
 271 }
 272
 273 int bch2_fs_check_snapshots(struct bch_fs *c)
 274 {
 275         struct btree_trans trans;
 276         struct btree_iter iter;
 277         struct bkey_s_c k;
 278         int ret;
 279
 280         bch2_trans_init(&trans, c, 0, 0);
 281
 282         ret = for_each_btree_key_commit(&trans, iter,
 283                         BTREE_ID_snapshots, POS_MIN,
 284                         BTREE_ITER_PREFETCH, k,
 285                         NULL, NULL, BTREE_INSERT_LAZY_RW|BTREE_INSERT_NOFAIL,
 286                 check_snapshot(&trans, &iter, k));
 287
 288         if (ret)
 289                 bch_err(c, "error %i checking snapshots", ret);
 290
 291         bch2_trans_exit(&trans);
 292         return ret;
 293 }
 294
 295 static int check_subvol(struct btree_trans *trans,
 296                         struct btree_iter *iter,
 297                         struct bkey_s_c k)
 298 {
 299         struct bkey_s_c_subvolume subvol;
 300         struct bch_snapshot snapshot;
 301         unsigned snapid;
 302         int ret;
 303
 304         if (k.k->type != KEY_TYPE_subvolume)
 305                 return 0;
 306
 307         subvol = bkey_s_c_to_subvolume(k);
 308         snapid = le32_to_cpu(subvol.v->snapshot);
 309         ret = snapshot_lookup(trans, snapid, &snapshot);
 310
 311         if (ret == -ENOENT)
 312                 bch_err(trans->c, "subvolume %llu points to nonexistent snapshot %u",
 313                         k.k->p.offset, snapid);
 314         if (ret)
 315                 return ret;
 316
 317         if (BCH_SUBVOLUME_UNLINKED(subvol.v)) {
 318                 ret = bch2_subvolume_delete(trans, iter->pos.offset);
 319                 if (ret && !bch2_err_matches(ret, BCH_ERR_transaction_restart))
 320                         bch_err(trans->c, "error deleting subvolume %llu: %s",
 321                                 iter->pos.offset, bch2_err_str(ret));
 322                 if (ret)
 323                         return ret;
 324         }
 325
 326         return 0;
 327 }
 328
 329 int bch2_fs_check_subvols(struct bch_fs *c)
 330 {
 331         struct btree_trans trans;
 332         struct btree_iter iter;
 333         struct bkey_s_c k;
 334         int ret;
 335
 336         bch2_trans_init(&trans, c, 0, 0);
 337
 338         ret = for_each_btree_key_commit(&trans, iter,
 339                         BTREE_ID_subvolumes, POS_MIN, BTREE_ITER_PREFETCH, k,
 340                         NULL, NULL, BTREE_INSERT_LAZY_RW|BTREE_INSERT_NOFAIL,
 341                 check_subvol(&trans, &iter, k));
 342
 343         bch2_trans_exit(&trans);
 344
 345         return ret;
 346 }
 347
 348 void bch2_fs_snapshots_exit(struct bch_fs *c)
 349 {
 350         genradix_free(&c->snapshots);
 351 }
 352
 353 int bch2_fs_snapshots_start(struct bch_fs *c)
 354 {
 355         struct btree_trans trans;
 356         struct btree_iter iter;
 357         struct bkey_s_c k;
 358         int ret = 0;
 359
 360         bch2_trans_init(&trans, c, 0, 0);
 361
 362         for_each_btree_key2(&trans, iter, BTREE_ID_snapshots,
 363                            POS_MIN, 0, k,
 364                 bch2_mark_snapshot(&trans, bkey_s_c_null, k, 0) ?:
 365                 bch2_snapshot_set_equiv(&trans, k));
 366
 367         bch2_trans_exit(&trans);
 368
 369         if (ret)
 370                 bch_err(c, "error starting snapshots: %s", bch2_err_str(ret));
 371         return ret;
 372 }
 373
 374 /*
 375  * Mark a snapshot as deleted, for future cleanup:
 376  */
 377 static int bch2_snapshot_node_set_deleted(struct btree_trans *trans, u32 id)
 378 {
 379         struct btree_iter iter;
 380         struct bkey_i_snapshot *s;
 381         int ret = 0;
 382
 383         bch2_trans_iter_init(trans, &iter, BTREE_ID_snapshots, POS(0, id),
 384                              BTREE_ITER_INTENT);
 385         s = bch2_bkey_get_mut_typed(trans, &iter, snapshot);
 386         ret = PTR_ERR_OR_ZERO(s);
 387         if (unlikely(ret)) {
 388                 bch2_fs_inconsistent_on(ret == -ENOENT, trans->c, "missing snapshot %u", id);
 389                 goto err;
 390         }
 391
 392         /* already deleted? */
 393         if (BCH_SNAPSHOT_DELETED(&s->v))
 394                 goto err;
 395
 396         SET_BCH_SNAPSHOT_DELETED(&s->v, true);
 397         SET_BCH_SNAPSHOT_SUBVOL(&s->v, false);
 398         s->v.subvol = 0;
 399
 400         ret = bch2_trans_update(trans, &iter, &s->k_i, 0);
 401         if (ret)
 402                 goto err;
 403 err:
 404         bch2_trans_iter_exit(trans, &iter);
 405         return ret;
 406 }
 407
 408 static int bch2_snapshot_node_delete(struct btree_trans *trans, u32 id)
 409 {
 410         struct btree_iter iter, p_iter = (struct btree_iter) { NULL };
 411         struct bkey_s_c k;
 412         struct bkey_s_c_snapshot s;
 413         u32 parent_id;
 414         unsigned i;
 415         int ret = 0;
 416
 417         bch2_trans_iter_init(trans, &iter, BTREE_ID_snapshots, POS(0, id),
 418                              BTREE_ITER_INTENT);
 419         k = bch2_btree_iter_peek_slot(&iter);
 420         ret = bkey_err(k);
 421         if (ret)
 422                 goto err;
 423
 424         if (k.k->type != KEY_TYPE_snapshot) {
 425                 bch2_fs_inconsistent(trans->c, "missing snapshot %u", id);
 426                 ret = -ENOENT;
 427                 goto err;
 428         }
 429
 430         s = bkey_s_c_to_snapshot(k);
 431
 432         BUG_ON(!BCH_SNAPSHOT_DELETED(s.v));
 433         parent_id = le32_to_cpu(s.v->parent);
 434
 435         if (parent_id) {
 436                 struct bkey_i_snapshot *parent;
 437
 438                 bch2_trans_iter_init(trans, &p_iter, BTREE_ID_snapshots,
 439                                      POS(0, parent_id),
 440                                      BTREE_ITER_INTENT);
 441                 parent = bch2_bkey_get_mut_typed(trans, &p_iter, snapshot);
 442                 ret = PTR_ERR_OR_ZERO(parent);
 443                 if (unlikely(ret)) {
 444                         bch2_fs_inconsistent_on(ret == -ENOENT, trans->c, "missing snapshot %u", parent_id);
 445                         goto err;
 446                 }
 447
 448                 for (i = 0; i < 2; i++)
 449                         if (le32_to_cpu(parent->v.children[i]) == id)
 450                                 break;
 451
 452                 if (i == 2)
 453                         bch_err(trans->c, "snapshot %u missing child pointer to %u",
 454                                 parent_id, id);
 455                 else
 456                         parent->v.children[i] = 0;
 457
 458                 if (le32_to_cpu(parent->v.children[0]) <
 459                     le32_to_cpu(parent->v.children[1]))
 460                         swap(parent->v.children[0],
 461                              parent->v.children[1]);
 462
 463                 ret = bch2_trans_update(trans, &p_iter, &parent->k_i, 0);
 464                 if (ret)
 465                         goto err;
 466         }
 467
 468         ret = bch2_btree_delete_at(trans, &iter, 0);
 469 err:
 470         bch2_trans_iter_exit(trans, &p_iter);
 471         bch2_trans_iter_exit(trans, &iter);
 472         return ret;
 473 }
 474
 475 int bch2_snapshot_node_create(struct btree_trans *trans, u32 parent,
 476                               u32 *new_snapids,
 477                               u32 *snapshot_subvols,
 478                               unsigned nr_snapids)
 479 {
 480         struct btree_iter iter;
 481         struct bkey_i_snapshot *n;
 482         struct bkey_s_c k;
 483         unsigned i;
 484         int ret = 0;
 485
 486         bch2_trans_iter_init(trans, &iter, BTREE_ID_snapshots,
 487                              POS_MIN, BTREE_ITER_INTENT);
 488         k = bch2_btree_iter_peek(&iter);
 489         ret = bkey_err(k);
 490         if (ret)
 491                 goto err;
 492
 493         for (i = 0; i < nr_snapids; i++) {
 494                 k = bch2_btree_iter_prev_slot(&iter);
 495                 ret = bkey_err(k);
 496                 if (ret)
 497                         goto err;
 498
 499                 if (!k.k || !k.k->p.offset) {
 500                         ret = -BCH_ERR_ENOSPC_snapshot_create;
 501                         goto err;
 502                 }
 503
 504                 n = bch2_bkey_alloc(trans, &iter, snapshot);
 505                 ret = PTR_ERR_OR_ZERO(n);
 506                 if (ret)
 507                         goto err;
 508
 509                 n->v.flags      = 0;
 510                 n->v.parent     = cpu_to_le32(parent);
 511                 n->v.subvol     = cpu_to_le32(snapshot_subvols[i]);
 512                 n->v.pad        = 0;
 513                 SET_BCH_SNAPSHOT_SUBVOL(&n->v, true);
 514
 515                 ret   = bch2_trans_update(trans, &iter, &n->k_i, 0);
 516                 if (ret)
 517                         goto err;
 518
 519                 new_snapids[i]  = iter.pos.offset;
 520         }
 521
 522         if (parent) {
 523                 bch2_btree_iter_set_pos(&iter, POS(0, parent));
 524                 n = bch2_bkey_get_mut_typed(trans, &iter, snapshot);
 525                 ret = PTR_ERR_OR_ZERO(n);
 526                 if (unlikely(ret)) {
 527                         if (ret == -ENOENT)
 528                                 bch_err(trans->c, "snapshot %u not found", parent);
 529                         goto err;
 530                 }
 531
 532                 if (n->v.children[0] || n->v.children[1]) {
 533                         bch_err(trans->c, "Trying to add child snapshot nodes to parent that already has children");
 534                         ret = -EINVAL;
 535                         goto err;
 536                 }
 537
 538                 n->v.children[0] = cpu_to_le32(new_snapids[0]);
 539                 n->v.children[1] = cpu_to_le32(new_snapids[1]);
 540                 n->v.subvol = 0;
 541                 SET_BCH_SNAPSHOT_SUBVOL(&n->v, false);
 542                 ret = bch2_trans_update(trans, &iter, &n->k_i, 0);
 543                 if (ret)
 544                         goto err;
 545         }
 546 err:
 547         bch2_trans_iter_exit(trans, &iter);
 548         return ret;
 549 }
 550
 551 static int snapshot_delete_key(struct btree_trans *trans,
 552                                struct btree_iter *iter,
 553                                struct bkey_s_c k,
 554                                snapshot_id_list *deleted,
 555                                snapshot_id_list *equiv_seen,
 556                                struct bpos *last_pos)
 557 {
 558         struct bch_fs *c = trans->c;
 559         u32 equiv = snapshot_t(c, k.k->p.snapshot)->equiv;
 560
 561         if (!bkey_eq(k.k->p, *last_pos))
 562                 equiv_seen->nr = 0;
 563         *last_pos = k.k->p;
 564
 565         if (snapshot_list_has_id(deleted, k.k->p.snapshot) ||
 566             snapshot_list_has_id(equiv_seen, equiv)) {
 567                 return bch2_btree_delete_at(trans, iter,
 568                                             BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE);
 569         } else {
 570                 return snapshot_list_add(c, equiv_seen, equiv);
 571         }
 572 }
 573
 574 static int bch2_delete_redundant_snapshot(struct btree_trans *trans, struct btree_iter *iter,
 575                                           struct bkey_s_c k)
 576 {
 577         struct bkey_s_c_snapshot snap;
 578         u32 children[2];
 579         int ret;
 580
 581         if (k.k->type != KEY_TYPE_snapshot)
 582                 return 0;
 583
 584         snap = bkey_s_c_to_snapshot(k);
 585         if (BCH_SNAPSHOT_DELETED(snap.v) ||
 586             BCH_SNAPSHOT_SUBVOL(snap.v))
 587                 return 0;
 588
 589         children[0] = le32_to_cpu(snap.v->children[0]);
 590         children[1] = le32_to_cpu(snap.v->children[1]);
 591
 592         ret   = snapshot_live(trans, children[0]) ?:
 593                 snapshot_live(trans, children[1]);
 594         if (ret < 0)
 595                 return ret;
 596
 597         if (!ret)
 598                 return bch2_snapshot_node_set_deleted(trans, k.k->p.offset);
 599         return 0;
 600 }
 601
 602 int bch2_delete_dead_snapshots(struct bch_fs *c)
 603 {
 604         struct btree_trans trans;
 605         struct btree_iter iter;
 606         struct bkey_s_c k;
 607         struct bkey_s_c_snapshot snap;
 608         snapshot_id_list deleted = { 0 };
 609         u32 i, id;
 610         int ret = 0;
 611
 612         if (!test_bit(BCH_FS_HAVE_DELETED_SNAPSHOTS, &c->flags))
 613                 return 0;
 614
 615         if (!test_bit(BCH_FS_STARTED, &c->flags)) {
 616                 ret = bch2_fs_read_write_early(c);
 617                 if (ret) {
 618                         bch_err(c, "error deleleting dead snapshots: error going rw: %s", bch2_err_str(ret));
 619                         return ret;
 620                 }
 621         }
 622
 623         bch2_trans_init(&trans, c, 0, 0);
 624
 625         /*
 626          * For every snapshot node: If we have no live children and it's not
 627          * pointed to by a subvolume, delete it:
 628          */
 629         ret = for_each_btree_key_commit(&trans, iter, BTREE_ID_snapshots,
 630                         POS_MIN, 0, k,
 631                         NULL, NULL, 0,
 632                 bch2_delete_redundant_snapshot(&trans, &iter, k));
 633         if (ret) {
 634                 bch_err(c, "error deleting redundant snapshots: %s", bch2_err_str(ret));
 635                 goto err;
 636         }
 637
 638         for_each_btree_key2(&trans, iter, BTREE_ID_snapshots,
 639                            POS_MIN, 0, k,
 640                 bch2_snapshot_set_equiv(&trans, k));
 641         if (ret) {
 642                 bch_err(c, "error in bch2_snapshots_set_equiv: %s", bch2_err_str(ret));
 643                 goto err;
 644         }
 645
 646         for_each_btree_key(&trans, iter, BTREE_ID_snapshots,
 647                            POS_MIN, 0, k, ret) {
 648                 if (k.k->type != KEY_TYPE_snapshot)
 649                         continue;
 650
 651                 snap = bkey_s_c_to_snapshot(k);
 652                 if (BCH_SNAPSHOT_DELETED(snap.v)) {
 653                         ret = snapshot_list_add(c, &deleted, k.k->p.offset);
 654                         if (ret)
 655                                 break;
 656                 }
 657         }
 658         bch2_trans_iter_exit(&trans, &iter);
 659
 660         if (ret) {
 661                 bch_err(c, "error walking snapshots: %s", bch2_err_str(ret));
 662                 goto err;
 663         }
 664
 665         for (id = 0; id < BTREE_ID_NR; id++) {
 666                 struct bpos last_pos = POS_MIN;
 667                 snapshot_id_list equiv_seen = { 0 };
 668
 669                 if (!btree_type_has_snapshots(id))
 670                         continue;
 671
 672                 ret = for_each_btree_key_commit(&trans, iter,
 673                                 id, POS_MIN,
 674                                 BTREE_ITER_PREFETCH|BTREE_ITER_ALL_SNAPSHOTS, k,
 675                                 NULL, NULL, BTREE_INSERT_NOFAIL,
 676                         snapshot_delete_key(&trans, &iter, k, &deleted, &equiv_seen, &last_pos));
 677
 678                 darray_exit(&equiv_seen);
 679
 680                 if (ret) {
 681                         bch_err(c, "error deleting snapshot keys: %s", bch2_err_str(ret));
 682                         goto err;
 683                 }
 684         }
 685
 686         for (i = 0; i < deleted.nr; i++) {
 687                 ret = commit_do(&trans, NULL, NULL, 0,
 688                         bch2_snapshot_node_delete(&trans, deleted.data[i]));
 689                 if (ret) {
 690                         bch_err(c, "error deleting snapshot %u: %s",
 691                                 deleted.data[i], bch2_err_str(ret));
 692                         goto err;
 693                 }
 694         }
 695
 696         clear_bit(BCH_FS_HAVE_DELETED_SNAPSHOTS, &c->flags);
 697 err:
 698         darray_exit(&deleted);
 699         bch2_trans_exit(&trans);
 700         return ret;
 701 }
 702
 703 static void bch2_delete_dead_snapshots_work(struct work_struct *work)
 704 {
 705         struct bch_fs *c = container_of(work, struct bch_fs, snapshot_delete_work);
 706
 707         bch2_delete_dead_snapshots(c);
 708         bch2_write_ref_put(c, BCH_WRITE_REF_delete_dead_snapshots);
 709 }
 710
 711 void bch2_delete_dead_snapshots_async(struct bch_fs *c)
 712 {
 713         if (bch2_write_ref_tryget(c, BCH_WRITE_REF_delete_dead_snapshots) &&
 714             !queue_work(system_long_wq, &c->snapshot_delete_work))
 715                 bch2_write_ref_put(c, BCH_WRITE_REF_delete_dead_snapshots);
 716 }
 717
 718 static int bch2_delete_dead_snapshots_hook(struct btree_trans *trans,
 719                                            struct btree_trans_commit_hook *h)
 720 {
 721         struct bch_fs *c = trans->c;
 722
 723         set_bit(BCH_FS_HAVE_DELETED_SNAPSHOTS, &c->flags);
 724
 725         if (!test_bit(BCH_FS_FSCK_DONE, &c->flags))
 726                 return 0;
 727
 728         bch2_delete_dead_snapshots_async(c);
 729         return 0;
 730 }
 731
 732 /* Subvolumes: */
 733
 734 int bch2_subvolume_invalid(const struct bch_fs *c, struct bkey_s_c k,
 735                            unsigned flags, struct printbuf *err)
 736 {
 737         if (bkey_lt(k.k->p, SUBVOL_POS_MIN) ||
 738             bkey_gt(k.k->p, SUBVOL_POS_MAX)) {
 739                 prt_printf(err, "invalid pos");
 740                 return -BCH_ERR_invalid_bkey;
 741         }
 742
 743         if (bkey_val_bytes(k.k) != sizeof(struct bch_subvolume)) {
 744                 prt_printf(err, "incorrect value size (%zu != %zu)",
 745                        bkey_val_bytes(k.k), sizeof(struct bch_subvolume));
 746                 return -BCH_ERR_invalid_bkey;
 747         }
 748
 749         return 0;
 750 }
 751
 752 void bch2_subvolume_to_text(struct printbuf *out, struct bch_fs *c,
 753                             struct bkey_s_c k)
 754 {
 755         struct bkey_s_c_subvolume s = bkey_s_c_to_subvolume(k);
 756
 757         prt_printf(out, "root %llu snapshot id %u",
 758                le64_to_cpu(s.v->inode),
 759                le32_to_cpu(s.v->snapshot));
 760 }
 761
 762 static __always_inline int
 763 bch2_subvolume_get_inlined(struct btree_trans *trans, unsigned subvol,
 764                            bool inconsistent_if_not_found,
 765                            int iter_flags,
 766                            struct bch_subvolume *s)
 767 {
 768         struct btree_iter iter;
 769         struct bkey_s_c k;
 770         int ret;
 771
 772         bch2_trans_iter_init(trans, &iter, BTREE_ID_subvolumes, POS(0, subvol),
 773                              iter_flags);
 774         k = bch2_btree_iter_peek_slot(&iter);
 775         ret = bkey_err(k) ?: k.k->type == KEY_TYPE_subvolume ? 0 : -ENOENT;
 776
 777         if (ret == -ENOENT && inconsistent_if_not_found)
 778                 bch2_fs_inconsistent(trans->c, "missing subvolume %u", subvol);
 779         if (!ret)
 780                 *s = *bkey_s_c_to_subvolume(k).v;
 781
 782         bch2_trans_iter_exit(trans, &iter);
 783         return ret;
 784 }
 785
 786 int bch2_subvolume_get(struct btree_trans *trans, unsigned subvol,
 787                        bool inconsistent_if_not_found,
 788                        int iter_flags,
 789                        struct bch_subvolume *s)
 790 {
 791         return bch2_subvolume_get_inlined(trans, subvol, inconsistent_if_not_found, iter_flags, s);
 792 }
 793
 794 int bch2_snapshot_get_subvol(struct btree_trans *trans, u32 snapshot,
 795                              struct bch_subvolume *subvol)
 796 {
 797         struct bch_snapshot snap;
 798
 799         return  snapshot_lookup(trans, snapshot, &snap) ?:
 800                 bch2_subvolume_get(trans, le32_to_cpu(snap.subvol), true, 0, subvol);
 801 }
 802
 803 int bch2_subvolume_get_snapshot(struct btree_trans *trans, u32 subvol,
 804                                 u32 *snapid)
 805 {
 806         struct bch_subvolume s;
 807         int ret;
 808
 809         ret = bch2_subvolume_get_inlined(trans, subvol, true,
 810                                          BTREE_ITER_CACHED|
 811                                          BTREE_ITER_WITH_UPDATES,
 812                                          &s);
 813         if (!ret)
 814                 *snapid = le32_to_cpu(s.snapshot);
 815         return ret;
 816 }
 817
 818 /*
 819  * Delete subvolume, mark snapshot ID as deleted, queue up snapshot
 820  * deletion/cleanup:
 821  */
 822 int bch2_subvolume_delete(struct btree_trans *trans, u32 subvolid)
 823 {
 824         struct btree_iter iter;
 825         struct bkey_s_c k;
 826         struct bkey_s_c_subvolume subvol;
 827         struct btree_trans_commit_hook *h;
 828         u32 snapid;
 829         int ret = 0;
 830
 831         bch2_trans_iter_init(trans, &iter, BTREE_ID_subvolumes,
 832                              POS(0, subvolid),
 833                              BTREE_ITER_CACHED|
 834                              BTREE_ITER_INTENT);
 835         k = bch2_btree_iter_peek_slot(&iter);
 836         ret = bkey_err(k);
 837         if (ret)
 838                 goto err;
 839
 840         if (k.k->type != KEY_TYPE_subvolume) {
 841                 bch2_fs_inconsistent(trans->c, "missing subvolume %u", subvolid);
 842                 ret = -EIO;
 843                 goto err;
 844         }
 845
 846         subvol = bkey_s_c_to_subvolume(k);
 847         snapid = le32_to_cpu(subvol.v->snapshot);
 848
 849         ret = bch2_btree_delete_at(trans, &iter, 0);
 850         if (ret)
 851                 goto err;
 852
 853         ret = bch2_snapshot_node_set_deleted(trans, snapid);
 854         if (ret)
 855                 goto err;
 856
 857         h = bch2_trans_kmalloc(trans, sizeof(*h));
 858         ret = PTR_ERR_OR_ZERO(h);
 859         if (ret)
 860                 goto err;
 861
 862         h->fn = bch2_delete_dead_snapshots_hook;
 863         bch2_trans_commit_hook(trans, h);
 864 err:
 865         bch2_trans_iter_exit(trans, &iter);
 866         return ret;
 867 }
 868
 869 void bch2_subvolume_wait_for_pagecache_and_delete(struct work_struct *work)
 870 {
 871         struct bch_fs *c = container_of(work, struct bch_fs,
 872                                 snapshot_wait_for_pagecache_and_delete_work);
 873         snapshot_id_list s;
 874         u32 *id;
 875         int ret = 0;
 876
 877         while (!ret) {
 878                 mutex_lock(&c->snapshots_unlinked_lock);
 879                 s = c->snapshots_unlinked;
 880                 darray_init(&c->snapshots_unlinked);
 881                 mutex_unlock(&c->snapshots_unlinked_lock);
 882
 883                 if (!s.nr)
 884                         break;
 885
 886                 bch2_evict_subvolume_inodes(c, &s);
 887
 888                 for (id = s.data; id < s.data + s.nr; id++) {
 889                         ret = bch2_trans_do(c, NULL, NULL, BTREE_INSERT_NOFAIL,
 890                                       bch2_subvolume_delete(&trans, *id));
 891                         if (ret) {
 892                                 bch_err(c, "error deleting subvolume %u: %s", *id, bch2_err_str(ret));
 893                                 break;
 894                         }
 895                 }
 896
 897                 darray_exit(&s);
 898         }
 899
 900         bch2_write_ref_put(c, BCH_WRITE_REF_snapshot_delete_pagecache);
 901 }
 902
 903 struct subvolume_unlink_hook {
 904         struct btree_trans_commit_hook  h;
 905         u32                             subvol;
 906 };
 907
 908 int bch2_subvolume_wait_for_pagecache_and_delete_hook(struct btree_trans *trans,
 909                                                       struct btree_trans_commit_hook *_h)
 910 {
 911         struct subvolume_unlink_hook *h = container_of(_h, struct subvolume_unlink_hook, h);
 912         struct bch_fs *c = trans->c;
 913         int ret = 0;
 914
 915         mutex_lock(&c->snapshots_unlinked_lock);
 916         if (!snapshot_list_has_id(&c->snapshots_unlinked, h->subvol))
 917                 ret = snapshot_list_add(c, &c->snapshots_unlinked, h->subvol);
 918         mutex_unlock(&c->snapshots_unlinked_lock);
 919
 920         if (ret)
 921                 return ret;
 922
 923         if (!bch2_write_ref_tryget(c, BCH_WRITE_REF_snapshot_delete_pagecache))
 924                 return -EROFS;
 925
 926         if (!queue_work(system_long_wq, &c->snapshot_wait_for_pagecache_and_delete_work))
 927                 bch2_write_ref_put(c, BCH_WRITE_REF_snapshot_delete_pagecache);
 928         return 0;
 929 }
 930
 931 int bch2_subvolume_unlink(struct btree_trans *trans, u32 subvolid)
 932 {
 933         struct btree_iter iter;
 934         struct bkey_i_subvolume *n;
 935         struct subvolume_unlink_hook *h;
 936         int ret = 0;
 937
 938         bch2_trans_iter_init(trans, &iter, BTREE_ID_subvolumes,
 939                              POS(0, subvolid),
 940                              BTREE_ITER_CACHED|
 941                              BTREE_ITER_INTENT);
 942         n = bch2_bkey_get_mut_typed(trans, &iter, subvolume);
 943         ret = PTR_ERR_OR_ZERO(n);
 944         if (unlikely(ret)) {
 945                 bch2_fs_inconsistent_on(ret == -ENOENT, trans->c, "missing subvolume %u", subvolid);
 946                 goto err;
 947         }
 948
 949         SET_BCH_SUBVOLUME_UNLINKED(&n->v, true);
 950
 951         ret = bch2_trans_update(trans, &iter, &n->k_i, 0);
 952         if (ret)
 953                 goto err;
 954
 955         h = bch2_trans_kmalloc(trans, sizeof(*h));
 956         ret = PTR_ERR_OR_ZERO(h);
 957         if (ret)
 958                 goto err;
 959
 960         h->h.fn         = bch2_subvolume_wait_for_pagecache_and_delete_hook;
 961         h->subvol       = subvolid;
 962         bch2_trans_commit_hook(trans, &h->h);
 963 err:
 964         bch2_trans_iter_exit(trans, &iter);
 965         return ret;
 966 }
 967
 968 int bch2_subvolume_create(struct btree_trans *trans, u64 inode,
 969                           u32 src_subvolid,
 970                           u32 *new_subvolid,
 971                           u32 *new_snapshotid,
 972                           bool ro)
 973 {
 974         struct bch_fs *c = trans->c;
 975         struct btree_iter dst_iter, src_iter = (struct btree_iter) { NULL };
 976         struct bkey_i_subvolume *new_subvol = NULL;
 977         struct bkey_i_subvolume *src_subvol = NULL;
 978         struct bkey_s_c k;
 979         u32 parent = 0, new_nodes[2], snapshot_subvols[2];
 980         int ret = 0;
 981
 982         for_each_btree_key(trans, dst_iter, BTREE_ID_subvolumes, SUBVOL_POS_MIN,
 983                            BTREE_ITER_SLOTS|BTREE_ITER_INTENT, k, ret) {
 984                 if (bkey_gt(k.k->p, SUBVOL_POS_MAX))
 985                         break;
 986
 987                 /*
 988                  * bch2_subvolume_delete() doesn't flush the btree key cache -
 989                  * ideally it would but that's tricky
 990                  */
 991                 if (bkey_deleted(k.k) &&
 992                     !bch2_btree_key_cache_find(c, BTREE_ID_subvolumes, dst_iter.pos))
 993                         goto found_slot;
 994         }
 995
 996         if (!ret)
 997                 ret = -BCH_ERR_ENOSPC_subvolume_create;
 998         goto err;
 999 found_slot:
1000         snapshot_subvols[0] = dst_iter.pos.offset;
1001         snapshot_subvols[1] = src_subvolid;
1002
1003         if (src_subvolid) {
1004                 /* Creating a snapshot: */
1005
1006                 bch2_trans_iter_init(trans, &src_iter, BTREE_ID_subvolumes,
1007                                      POS(0, src_subvolid),
1008                                      BTREE_ITER_CACHED|
1009                                      BTREE_ITER_INTENT);
1010                 src_subvol = bch2_bkey_get_mut_typed(trans, &src_iter, subvolume);
1011                 ret = PTR_ERR_OR_ZERO(src_subvol);
1012                 if (unlikely(ret)) {
1013                         bch2_fs_inconsistent_on(ret == -ENOENT, trans->c,
1014                                                 "subvolume %u not found", src_subvolid);
1015                         goto err;
1016                 }
1017
1018                 parent = le32_to_cpu(src_subvol->v.snapshot);
1019         }
1020
1021         ret = bch2_snapshot_node_create(trans, parent, new_nodes,
1022                                         snapshot_subvols,
1023                                         src_subvolid ? 2 : 1);
1024         if (ret)
1025                 goto err;
1026
1027         if (src_subvolid) {
1028                 src_subvol->v.snapshot = cpu_to_le32(new_nodes[1]);
1029                 ret = bch2_trans_update(trans, &src_iter, &src_subvol->k_i, 0);
1030                 if (ret)
1031                         goto err;
1032         }
1033
1034         new_subvol = bch2_bkey_alloc(trans, &dst_iter, subvolume);
1035         ret = PTR_ERR_OR_ZERO(new_subvol);
1036         if (ret)
1037                 goto err;
1038
1039         new_subvol->v.flags     = 0;
1040         new_subvol->v.snapshot  = cpu_to_le32(new_nodes[0]);
1041         new_subvol->v.inode     = cpu_to_le64(inode);
1042         SET_BCH_SUBVOLUME_RO(&new_subvol->v, ro);
1043         SET_BCH_SUBVOLUME_SNAP(&new_subvol->v, src_subvolid != 0);
1044         ret = bch2_trans_update(trans, &dst_iter, &new_subvol->k_i, 0);
1045         if (ret)
1046                 goto err;
1047
1048         *new_subvolid   = new_subvol->k.p.offset;
1049         *new_snapshotid = new_nodes[0];
1050 err:
1051         bch2_trans_iter_exit(trans, &src_iter);
1052         bch2_trans_iter_exit(trans, &dst_iter);
1053         return ret;
1054 }
1055
1056 int bch2_fs_subvolumes_init(struct bch_fs *c)
1057 {
1058         INIT_WORK(&c->snapshot_delete_work, bch2_delete_dead_snapshots_work);
1059         INIT_WORK(&c->snapshot_wait_for_pagecache_and_delete_work,
1060                   bch2_subvolume_wait_for_pagecache_and_delete);
1061         mutex_init(&c->snapshots_unlinked_lock);
1062         return 0;
1063 }