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                        enum btree_id btree, unsigned level,
  80                        struct bkey_s_c old, struct bkey_s_c new,
  81                        unsigned flags)
  82 {
  83         struct bch_fs *c = trans->c;
  84         struct snapshot_t *t;
  85
  86         t = genradix_ptr_alloc(&c->snapshots,
  87                                U32_MAX - new.k->p.offset,
  88                                GFP_KERNEL);
  89         if (!t)
  90                 return -ENOMEM;
  91
  92         if (new.k->type == KEY_TYPE_snapshot) {
  93                 struct bkey_s_c_snapshot s = bkey_s_c_to_snapshot(new);
  94
  95                 t->parent       = le32_to_cpu(s.v->parent);
  96                 t->children[0]  = le32_to_cpu(s.v->children[0]);
  97                 t->children[1]  = le32_to_cpu(s.v->children[1]);
  98                 t->subvol       = BCH_SNAPSHOT_SUBVOL(s.v) ? le32_to_cpu(s.v->subvol) : 0;
  99         } else {
 100                 t->parent       = 0;
 101                 t->children[0]  = 0;
 102                 t->children[1]  = 0;
 103                 t->subvol       = 0;
 104         }
 105
 106         return 0;
 107 }
 108
 109 static int snapshot_lookup(struct btree_trans *trans, u32 id,
 110                            struct bch_snapshot *s)
 111 {
 112         struct btree_iter iter;
 113         struct bkey_s_c k;
 114         int ret;
 115
 116         bch2_trans_iter_init(trans, &iter, BTREE_ID_snapshots, POS(0, id),
 117                              BTREE_ITER_WITH_UPDATES);
 118         k = bch2_btree_iter_peek_slot(&iter);
 119         ret = bkey_err(k) ?: k.k->type == KEY_TYPE_snapshot ? 0 : -ENOENT;
 120
 121         if (!ret)
 122                 *s = *bkey_s_c_to_snapshot(k).v;
 123
 124         bch2_trans_iter_exit(trans, &iter);
 125         return ret;
 126 }
 127
 128 static int snapshot_live(struct btree_trans *trans, u32 id)
 129 {
 130         struct bch_snapshot v;
 131         int ret;
 132
 133         if (!id)
 134                 return 0;
 135
 136         ret = snapshot_lookup(trans, id, &v);
 137         if (ret == -ENOENT)
 138                 bch_err(trans->c, "snapshot node %u not found", id);
 139         if (ret)
 140                 return ret;
 141
 142         return !BCH_SNAPSHOT_DELETED(&v);
 143 }
 144
 145 static int bch2_snapshot_set_equiv(struct btree_trans *trans, struct bkey_s_c k)
 146 {
 147         struct bch_fs *c = trans->c;
 148         unsigned i, nr_live = 0, live_idx = 0;
 149         struct bkey_s_c_snapshot snap;
 150         u32 id = k.k->p.offset, child[2];
 151
 152         if (k.k->type != KEY_TYPE_snapshot)
 153                 return 0;
 154
 155         snap = bkey_s_c_to_snapshot(k);
 156
 157         child[0] = le32_to_cpu(snap.v->children[0]);
 158         child[1] = le32_to_cpu(snap.v->children[1]);
 159
 160         for (i = 0; i < 2; i++) {
 161                 int ret = snapshot_live(trans, child[i]);
 162
 163                 if (ret < 0)
 164                         return ret;
 165
 166                 if (ret)
 167                         live_idx = i;
 168                 nr_live += ret;
 169         }
 170
 171         snapshot_t(c, id)->equiv = nr_live == 1
 172                 ? snapshot_t(c, child[live_idx])->equiv
 173                 : id;
 174         return 0;
 175 }
 176
 177 /* fsck: */
 178 static int check_snapshot(struct btree_trans *trans,
 179                           struct btree_iter *iter,
 180                           struct bkey_s_c k)
 181 {
 182         struct bch_fs *c = trans->c;
 183         struct bkey_s_c_snapshot s;
 184         struct bch_subvolume subvol;
 185         struct bch_snapshot v;
 186         struct printbuf buf = PRINTBUF;
 187         bool should_have_subvol;
 188         u32 i, id;
 189         int ret = 0;
 190
 191         if (k.k->type != KEY_TYPE_snapshot)
 192                 return 0;
 193
 194         s = bkey_s_c_to_snapshot(k);
 195         id = le32_to_cpu(s.v->parent);
 196         if (id) {
 197                 ret = snapshot_lookup(trans, id, &v);
 198                 if (ret == -ENOENT)
 199                         bch_err(c, "snapshot with nonexistent parent:\n  %s",
 200                                 (bch2_bkey_val_to_text(&buf, c, s.s_c), buf.buf));
 201                 if (ret)
 202                         goto err;
 203
 204                 if (le32_to_cpu(v.children[0]) != s.k->p.offset &&
 205                     le32_to_cpu(v.children[1]) != s.k->p.offset) {
 206                         bch_err(c, "snapshot parent %u missing pointer to child %llu",
 207                                 id, s.k->p.offset);
 208                         ret = -EINVAL;
 209                         goto err;
 210                 }
 211         }
 212
 213         for (i = 0; i < 2 && s.v->children[i]; i++) {
 214                 id = le32_to_cpu(s.v->children[i]);
 215
 216                 ret = snapshot_lookup(trans, id, &v);
 217                 if (ret == -ENOENT)
 218                         bch_err(c, "snapshot node %llu has nonexistent child %u",
 219                                 s.k->p.offset, id);
 220                 if (ret)
 221                         goto err;
 222
 223                 if (le32_to_cpu(v.parent) != s.k->p.offset) {
 224                         bch_err(c, "snapshot child %u has wrong parent (got %u should be %llu)",
 225                                 id, le32_to_cpu(v.parent), s.k->p.offset);
 226                         ret = -EINVAL;
 227                         goto err;
 228                 }
 229         }
 230
 231         should_have_subvol = BCH_SNAPSHOT_SUBVOL(s.v) &&
 232                 !BCH_SNAPSHOT_DELETED(s.v);
 233
 234         if (should_have_subvol) {
 235                 id = le32_to_cpu(s.v->subvol);
 236                 ret = bch2_subvolume_get(trans, id, 0, false, &subvol);
 237                 if (ret == -ENOENT)
 238                         bch_err(c, "snapshot points to nonexistent subvolume:\n  %s",
 239                                 (bch2_bkey_val_to_text(&buf, c, s.s_c), buf.buf));
 240                 if (ret)
 241                         goto err;
 242
 243                 if (BCH_SNAPSHOT_SUBVOL(s.v) != (le32_to_cpu(subvol.snapshot) == s.k->p.offset)) {
 244                         bch_err(c, "snapshot node %llu has wrong BCH_SNAPSHOT_SUBVOL",
 245                                 s.k->p.offset);
 246                         ret = -EINVAL;
 247                         goto err;
 248                 }
 249         } else {
 250                 if (fsck_err_on(s.v->subvol, c, "snapshot should not point to subvol:\n  %s",
 251                                 (bch2_bkey_val_to_text(&buf, c, s.s_c), buf.buf))) {
 252                         struct bkey_i_snapshot *u = bch2_trans_kmalloc(trans, sizeof(*u));
 253
 254                         ret = PTR_ERR_OR_ZERO(u);
 255                         if (ret)
 256                                 goto err;
 257
 258                         bkey_reassemble(&u->k_i, s.s_c);
 259                         u->v.subvol = 0;
 260                         ret = bch2_trans_update(trans, iter, &u->k_i, 0);
 261                         if (ret)
 262                                 goto err;
 263                 }
 264         }
 265
 266         if (BCH_SNAPSHOT_DELETED(s.v))
 267                 set_bit(BCH_FS_HAVE_DELETED_SNAPSHOTS, &c->flags);
 268 err:
 269 fsck_err:
 270         printbuf_exit(&buf);
 271         return ret;
 272 }
 273
 274 int bch2_fs_check_snapshots(struct bch_fs *c)
 275 {
 276         struct btree_trans trans;
 277         struct btree_iter iter;
 278         struct bkey_s_c k;
 279         int ret;
 280
 281         bch2_trans_init(&trans, c, 0, 0);
 282
 283         ret = for_each_btree_key_commit(&trans, iter,
 284                         BTREE_ID_snapshots, POS_MIN,
 285                         BTREE_ITER_PREFETCH, k,
 286                         NULL, NULL, BTREE_INSERT_LAZY_RW|BTREE_INSERT_NOFAIL,
 287                 check_snapshot(&trans, &iter, k));
 288
 289         if (ret)
 290                 bch_err(c, "error %i checking snapshots", ret);
 291
 292         bch2_trans_exit(&trans);
 293         return ret;
 294 }
 295
 296 static int check_subvol(struct btree_trans *trans,
 297                         struct btree_iter *iter,
 298                         struct bkey_s_c k)
 299 {
 300         struct bkey_s_c_subvolume subvol;
 301         struct bch_snapshot snapshot;
 302         unsigned snapid;
 303         int ret;
 304
 305         if (k.k->type != KEY_TYPE_subvolume)
 306                 return 0;
 307
 308         subvol = bkey_s_c_to_subvolume(k);
 309         snapid = le32_to_cpu(subvol.v->snapshot);
 310         ret = snapshot_lookup(trans, snapid, &snapshot);
 311
 312         if (ret == -ENOENT)
 313                 bch_err(trans->c, "subvolume %llu points to nonexistent snapshot %u",
 314                         k.k->p.offset, snapid);
 315         if (ret)
 316                 return ret;
 317
 318         if (BCH_SUBVOLUME_UNLINKED(subvol.v)) {
 319                 ret = bch2_subvolume_delete(trans, iter->pos.offset);
 320                 if (ret && !bch2_err_matches(ret, BCH_ERR_transaction_restart))
 321                         bch_err(trans->c, "error deleting subvolume %llu: %s",
 322                                 iter->pos.offset, bch2_err_str(ret));
 323                 if (ret)
 324                         return ret;
 325         }
 326
 327         return 0;
 328 }
 329
 330 int bch2_fs_check_subvols(struct bch_fs *c)
 331 {
 332         struct btree_trans trans;
 333         struct btree_iter iter;
 334         struct bkey_s_c k;
 335         int ret;
 336
 337         bch2_trans_init(&trans, c, 0, 0);
 338
 339         ret = for_each_btree_key_commit(&trans, iter,
 340                         BTREE_ID_subvolumes, POS_MIN, BTREE_ITER_PREFETCH, k,
 341                         NULL, NULL, BTREE_INSERT_LAZY_RW|BTREE_INSERT_NOFAIL,
 342                 check_subvol(&trans, &iter, k));
 343
 344         bch2_trans_exit(&trans);
 345
 346         return ret;
 347 }
 348
 349 void bch2_fs_snapshots_exit(struct bch_fs *c)
 350 {
 351         genradix_free(&c->snapshots);
 352 }
 353
 354 int bch2_fs_snapshots_start(struct bch_fs *c)
 355 {
 356         struct btree_trans trans;
 357         struct btree_iter iter;
 358         struct bkey_s_c k;
 359         int ret = 0;
 360
 361         bch2_trans_init(&trans, c, 0, 0);
 362
 363         for_each_btree_key2(&trans, iter, BTREE_ID_snapshots,
 364                            POS_MIN, 0, k,
 365                 bch2_mark_snapshot(&trans, BTREE_ID_snapshots, 0, bkey_s_c_null, k, 0) ?:
 366                 bch2_snapshot_set_equiv(&trans, k));
 367
 368         bch2_trans_exit(&trans);
 369
 370         if (ret)
 371                 bch_err(c, "error starting snapshots: %s", bch2_err_str(ret));
 372         return ret;
 373 }
 374
 375 /*
 376  * Mark a snapshot as deleted, for future cleanup:
 377  */
 378 static int bch2_snapshot_node_set_deleted(struct btree_trans *trans, u32 id)
 379 {
 380         struct btree_iter iter;
 381         struct bkey_i_snapshot *s;
 382         int ret = 0;
 383
 384         bch2_trans_iter_init(trans, &iter, BTREE_ID_snapshots, POS(0, id),
 385                              BTREE_ITER_INTENT);
 386         s = bch2_bkey_get_mut_typed(trans, &iter, snapshot);
 387         ret = PTR_ERR_OR_ZERO(s);
 388         if (unlikely(ret)) {
 389                 bch2_fs_inconsistent_on(ret == -ENOENT, trans->c, "missing snapshot %u", id);
 390                 goto err;
 391         }
 392
 393         /* already deleted? */
 394         if (BCH_SNAPSHOT_DELETED(&s->v))
 395                 goto err;
 396
 397         SET_BCH_SNAPSHOT_DELETED(&s->v, true);
 398         SET_BCH_SNAPSHOT_SUBVOL(&s->v, false);
 399         s->v.subvol = 0;
 400
 401         ret = bch2_trans_update(trans, &iter, &s->k_i, 0);
 402         if (ret)
 403                 goto err;
 404 err:
 405         bch2_trans_iter_exit(trans, &iter);
 406         return ret;
 407 }
 408
 409 static int bch2_snapshot_node_delete(struct btree_trans *trans, u32 id)
 410 {
 411         struct btree_iter iter, p_iter = (struct btree_iter) { NULL };
 412         struct bkey_s_c k;
 413         struct bkey_s_c_snapshot s;
 414         u32 parent_id;
 415         unsigned i;
 416         int ret = 0;
 417
 418         bch2_trans_iter_init(trans, &iter, BTREE_ID_snapshots, POS(0, id),
 419                              BTREE_ITER_INTENT);
 420         k = bch2_btree_iter_peek_slot(&iter);
 421         ret = bkey_err(k);
 422         if (ret)
 423                 goto err;
 424
 425         if (k.k->type != KEY_TYPE_snapshot) {
 426                 bch2_fs_inconsistent(trans->c, "missing snapshot %u", id);
 427                 ret = -ENOENT;
 428                 goto err;
 429         }
 430
 431         s = bkey_s_c_to_snapshot(k);
 432
 433         BUG_ON(!BCH_SNAPSHOT_DELETED(s.v));
 434         parent_id = le32_to_cpu(s.v->parent);
 435
 436         if (parent_id) {
 437                 struct bkey_i_snapshot *parent;
 438
 439                 bch2_trans_iter_init(trans, &p_iter, BTREE_ID_snapshots,
 440                                      POS(0, parent_id),
 441                                      BTREE_ITER_INTENT);
 442                 parent = bch2_bkey_get_mut_typed(trans, &p_iter, snapshot);
 443                 ret = PTR_ERR_OR_ZERO(parent);
 444                 if (unlikely(ret)) {
 445                         bch2_fs_inconsistent_on(ret == -ENOENT, trans->c, "missing snapshot %u", parent_id);
 446                         goto err;
 447                 }
 448
 449                 for (i = 0; i < 2; i++)
 450                         if (le32_to_cpu(parent->v.children[i]) == id)
 451                                 break;
 452
 453                 if (i == 2)
 454                         bch_err(trans->c, "snapshot %u missing child pointer to %u",
 455                                 parent_id, id);
 456                 else
 457                         parent->v.children[i] = 0;
 458
 459                 if (le32_to_cpu(parent->v.children[0]) <
 460                     le32_to_cpu(parent->v.children[1]))
 461                         swap(parent->v.children[0],
 462                              parent->v.children[1]);
 463
 464                 ret = bch2_trans_update(trans, &p_iter, &parent->k_i, 0);
 465                 if (ret)
 466                         goto err;
 467         }
 468
 469         ret = bch2_btree_delete_at(trans, &iter, 0);
 470 err:
 471         bch2_trans_iter_exit(trans, &p_iter);
 472         bch2_trans_iter_exit(trans, &iter);
 473         return ret;
 474 }
 475
 476 int bch2_snapshot_node_create(struct btree_trans *trans, u32 parent,
 477                               u32 *new_snapids,
 478                               u32 *snapshot_subvols,
 479                               unsigned nr_snapids)
 480 {
 481         struct btree_iter iter;
 482         struct bkey_i_snapshot *n;
 483         struct bkey_s_c k;
 484         unsigned i;
 485         int ret = 0;
 486
 487         bch2_trans_iter_init(trans, &iter, BTREE_ID_snapshots,
 488                              POS_MIN, BTREE_ITER_INTENT);
 489         k = bch2_btree_iter_peek(&iter);
 490         ret = bkey_err(k);
 491         if (ret)
 492                 goto err;
 493
 494         for (i = 0; i < nr_snapids; i++) {
 495                 k = bch2_btree_iter_prev_slot(&iter);
 496                 ret = bkey_err(k);
 497                 if (ret)
 498                         goto err;
 499
 500                 if (!k.k || !k.k->p.offset) {
 501                         ret = -BCH_ERR_ENOSPC_snapshot_create;
 502                         goto err;
 503                 }
 504
 505                 n = bch2_bkey_alloc(trans, &iter, snapshot);
 506                 ret = PTR_ERR_OR_ZERO(n);
 507                 if (ret)
 508                         goto err;
 509
 510                 n->v.flags      = 0;
 511                 n->v.parent     = cpu_to_le32(parent);
 512                 n->v.subvol     = cpu_to_le32(snapshot_subvols[i]);
 513                 n->v.pad        = 0;
 514                 SET_BCH_SNAPSHOT_SUBVOL(&n->v, true);
 515
 516                 ret   = bch2_trans_update(trans, &iter, &n->k_i, 0);
 517                 if (ret)
 518                         goto err;
 519
 520                 new_snapids[i]  = iter.pos.offset;
 521         }
 522
 523         if (parent) {
 524                 bch2_btree_iter_set_pos(&iter, POS(0, parent));
 525                 n = bch2_bkey_get_mut_typed(trans, &iter, snapshot);
 526                 ret = PTR_ERR_OR_ZERO(n);
 527                 if (unlikely(ret)) {
 528                         if (ret == -ENOENT)
 529                                 bch_err(trans->c, "snapshot %u not found", parent);
 530                         goto err;
 531                 }
 532
 533                 if (n->v.children[0] || n->v.children[1]) {
 534                         bch_err(trans->c, "Trying to add child snapshot nodes to parent that already has children");
 535                         ret = -EINVAL;
 536                         goto err;
 537                 }
 538
 539                 n->v.children[0] = cpu_to_le32(new_snapids[0]);
 540                 n->v.children[1] = cpu_to_le32(new_snapids[1]);
 541                 n->v.subvol = 0;
 542                 SET_BCH_SNAPSHOT_SUBVOL(&n->v, false);
 543                 ret = bch2_trans_update(trans, &iter, &n->k_i, 0);
 544                 if (ret)
 545                         goto err;
 546         }
 547 err:
 548         bch2_trans_iter_exit(trans, &iter);
 549         return ret;
 550 }
 551
 552 static int snapshot_delete_key(struct btree_trans *trans,
 553                                struct btree_iter *iter,
 554                                struct bkey_s_c k,
 555                                snapshot_id_list *deleted,
 556                                snapshot_id_list *equiv_seen,
 557                                struct bpos *last_pos)
 558 {
 559         struct bch_fs *c = trans->c;
 560         u32 equiv = snapshot_t(c, k.k->p.snapshot)->equiv;
 561
 562         if (!bkey_eq(k.k->p, *last_pos))
 563                 equiv_seen->nr = 0;
 564         *last_pos = k.k->p;
 565
 566         if (snapshot_list_has_id(deleted, k.k->p.snapshot) ||
 567             snapshot_list_has_id(equiv_seen, equiv)) {
 568                 return bch2_btree_delete_at(trans, iter,
 569                                             BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE);
 570         } else {
 571                 return snapshot_list_add(c, equiv_seen, equiv);
 572         }
 573 }
 574
 575 static int bch2_delete_redundant_snapshot(struct btree_trans *trans, struct btree_iter *iter,
 576                                           struct bkey_s_c k)
 577 {
 578         struct bkey_s_c_snapshot snap;
 579         u32 children[2];
 580         int ret;
 581
 582         if (k.k->type != KEY_TYPE_snapshot)
 583                 return 0;
 584
 585         snap = bkey_s_c_to_snapshot(k);
 586         if (BCH_SNAPSHOT_DELETED(snap.v) ||
 587             BCH_SNAPSHOT_SUBVOL(snap.v))
 588                 return 0;
 589
 590         children[0] = le32_to_cpu(snap.v->children[0]);
 591         children[1] = le32_to_cpu(snap.v->children[1]);
 592
 593         ret   = snapshot_live(trans, children[0]) ?:
 594                 snapshot_live(trans, children[1]);
 595         if (ret < 0)
 596                 return ret;
 597
 598         if (!ret)
 599                 return bch2_snapshot_node_set_deleted(trans, k.k->p.offset);
 600         return 0;
 601 }
 602
 603 int bch2_delete_dead_snapshots(struct bch_fs *c)
 604 {
 605         struct btree_trans trans;
 606         struct btree_iter iter;
 607         struct bkey_s_c k;
 608         struct bkey_s_c_snapshot snap;
 609         snapshot_id_list deleted = { 0 };
 610         u32 i, id;
 611         int ret = 0;
 612
 613         if (!test_bit(BCH_FS_HAVE_DELETED_SNAPSHOTS, &c->flags))
 614                 return 0;
 615
 616         if (!test_bit(BCH_FS_STARTED, &c->flags)) {
 617                 ret = bch2_fs_read_write_early(c);
 618                 if (ret) {
 619                         bch_err(c, "error deleleting dead snapshots: error going rw: %s", bch2_err_str(ret));
 620                         return ret;
 621                 }
 622         }
 623
 624         bch2_trans_init(&trans, c, 0, 0);
 625
 626         /*
 627          * For every snapshot node: If we have no live children and it's not
 628          * pointed to by a subvolume, delete it:
 629          */
 630         ret = for_each_btree_key_commit(&trans, iter, BTREE_ID_snapshots,
 631                         POS_MIN, 0, k,
 632                         NULL, NULL, 0,
 633                 bch2_delete_redundant_snapshot(&trans, &iter, k));
 634         if (ret) {
 635                 bch_err(c, "error deleting redundant snapshots: %s", bch2_err_str(ret));
 636                 goto err;
 637         }
 638
 639         for_each_btree_key2(&trans, iter, BTREE_ID_snapshots,
 640                            POS_MIN, 0, k,
 641                 bch2_snapshot_set_equiv(&trans, k));
 642         if (ret) {
 643                 bch_err(c, "error in bch2_snapshots_set_equiv: %s", bch2_err_str(ret));
 644                 goto err;
 645         }
 646
 647         for_each_btree_key(&trans, iter, BTREE_ID_snapshots,
 648                            POS_MIN, 0, k, ret) {
 649                 if (k.k->type != KEY_TYPE_snapshot)
 650                         continue;
 651
 652                 snap = bkey_s_c_to_snapshot(k);
 653                 if (BCH_SNAPSHOT_DELETED(snap.v)) {
 654                         ret = snapshot_list_add(c, &deleted, k.k->p.offset);
 655                         if (ret)
 656                                 break;
 657                 }
 658         }
 659         bch2_trans_iter_exit(&trans, &iter);
 660
 661         if (ret) {
 662                 bch_err(c, "error walking snapshots: %s", bch2_err_str(ret));
 663                 goto err;
 664         }
 665
 666         for (id = 0; id < BTREE_ID_NR; id++) {
 667                 struct bpos last_pos = POS_MIN;
 668                 snapshot_id_list equiv_seen = { 0 };
 669
 670                 if (!btree_type_has_snapshots(id))
 671                         continue;
 672
 673                 ret = for_each_btree_key_commit(&trans, iter,
 674                                 id, POS_MIN,
 675                                 BTREE_ITER_PREFETCH|BTREE_ITER_ALL_SNAPSHOTS, k,
 676                                 NULL, NULL, BTREE_INSERT_NOFAIL,
 677                         snapshot_delete_key(&trans, &iter, k, &deleted, &equiv_seen, &last_pos));
 678
 679                 darray_exit(&equiv_seen);
 680
 681                 if (ret) {
 682                         bch_err(c, "error deleting snapshot keys: %s", bch2_err_str(ret));
 683                         goto err;
 684                 }
 685         }
 686
 687         for (i = 0; i < deleted.nr; i++) {
 688                 ret = commit_do(&trans, NULL, NULL, 0,
 689                         bch2_snapshot_node_delete(&trans, deleted.data[i]));
 690                 if (ret) {
 691                         bch_err(c, "error deleting snapshot %u: %s",
 692                                 deleted.data[i], bch2_err_str(ret));
 693                         goto err;
 694                 }
 695         }
 696
 697         clear_bit(BCH_FS_HAVE_DELETED_SNAPSHOTS, &c->flags);
 698 err:
 699         darray_exit(&deleted);
 700         bch2_trans_exit(&trans);
 701         return ret;
 702 }
 703
 704 static void bch2_delete_dead_snapshots_work(struct work_struct *work)
 705 {
 706         struct bch_fs *c = container_of(work, struct bch_fs, snapshot_delete_work);
 707
 708         bch2_delete_dead_snapshots(c);
 709         bch2_write_ref_put(c, BCH_WRITE_REF_delete_dead_snapshots);
 710 }
 711
 712 void bch2_delete_dead_snapshots_async(struct bch_fs *c)
 713 {
 714         if (bch2_write_ref_tryget(c, BCH_WRITE_REF_delete_dead_snapshots) &&
 715             !queue_work(system_long_wq, &c->snapshot_delete_work))
 716                 bch2_write_ref_put(c, BCH_WRITE_REF_delete_dead_snapshots);
 717 }
 718
 719 static int bch2_delete_dead_snapshots_hook(struct btree_trans *trans,
 720                                            struct btree_trans_commit_hook *h)
 721 {
 722         struct bch_fs *c = trans->c;
 723
 724         set_bit(BCH_FS_HAVE_DELETED_SNAPSHOTS, &c->flags);
 725
 726         if (!test_bit(BCH_FS_FSCK_DONE, &c->flags))
 727                 return 0;
 728
 729         bch2_delete_dead_snapshots_async(c);
 730         return 0;
 731 }
 732
 733 /* Subvolumes: */
 734
 735 int bch2_subvolume_invalid(const struct bch_fs *c, struct bkey_s_c k,
 736                            unsigned flags, struct printbuf *err)
 737 {
 738         if (bkey_lt(k.k->p, SUBVOL_POS_MIN) ||
 739             bkey_gt(k.k->p, SUBVOL_POS_MAX)) {
 740                 prt_printf(err, "invalid pos");
 741                 return -BCH_ERR_invalid_bkey;
 742         }
 743
 744         if (bkey_val_bytes(k.k) != sizeof(struct bch_subvolume)) {
 745                 prt_printf(err, "incorrect value size (%zu != %zu)",
 746                        bkey_val_bytes(k.k), sizeof(struct bch_subvolume));
 747                 return -BCH_ERR_invalid_bkey;
 748         }
 749
 750         return 0;
 751 }
 752
 753 void bch2_subvolume_to_text(struct printbuf *out, struct bch_fs *c,
 754                             struct bkey_s_c k)
 755 {
 756         struct bkey_s_c_subvolume s = bkey_s_c_to_subvolume(k);
 757
 758         prt_printf(out, "root %llu snapshot id %u",
 759                le64_to_cpu(s.v->inode),
 760                le32_to_cpu(s.v->snapshot));
 761 }
 762
 763 static __always_inline int
 764 bch2_subvolume_get_inlined(struct btree_trans *trans, unsigned subvol,
 765                            bool inconsistent_if_not_found,
 766                            int iter_flags,
 767                            struct bch_subvolume *s)
 768 {
 769         struct btree_iter iter;
 770         struct bkey_s_c k;
 771         int ret;
 772
 773         bch2_trans_iter_init(trans, &iter, BTREE_ID_subvolumes, POS(0, subvol),
 774                              iter_flags);
 775         k = bch2_btree_iter_peek_slot(&iter);
 776         ret = bkey_err(k) ?: k.k->type == KEY_TYPE_subvolume ? 0 : -ENOENT;
 777
 778         if (ret == -ENOENT && inconsistent_if_not_found)
 779                 bch2_fs_inconsistent(trans->c, "missing subvolume %u", subvol);
 780         if (!ret)
 781                 *s = *bkey_s_c_to_subvolume(k).v;
 782
 783         bch2_trans_iter_exit(trans, &iter);
 784         return ret;
 785 }
 786
 787 int bch2_subvolume_get(struct btree_trans *trans, unsigned subvol,
 788                        bool inconsistent_if_not_found,
 789                        int iter_flags,
 790                        struct bch_subvolume *s)
 791 {
 792         return bch2_subvolume_get_inlined(trans, subvol, inconsistent_if_not_found, iter_flags, s);
 793 }
 794
 795 int bch2_snapshot_get_subvol(struct btree_trans *trans, u32 snapshot,
 796                              struct bch_subvolume *subvol)
 797 {
 798         struct bch_snapshot snap;
 799
 800         return  snapshot_lookup(trans, snapshot, &snap) ?:
 801                 bch2_subvolume_get(trans, le32_to_cpu(snap.subvol), true, 0, subvol);
 802 }
 803
 804 int bch2_subvolume_get_snapshot(struct btree_trans *trans, u32 subvol,
 805                                 u32 *snapid)
 806 {
 807         struct bch_subvolume s;
 808         int ret;
 809
 810         ret = bch2_subvolume_get_inlined(trans, subvol, true,
 811                                          BTREE_ITER_CACHED|
 812                                          BTREE_ITER_WITH_UPDATES,
 813                                          &s);
 814         if (!ret)
 815                 *snapid = le32_to_cpu(s.snapshot);
 816         return ret;
 817 }
 818
 819 /*
 820  * Delete subvolume, mark snapshot ID as deleted, queue up snapshot
 821  * deletion/cleanup:
 822  */
 823 int bch2_subvolume_delete(struct btree_trans *trans, u32 subvolid)
 824 {
 825         struct btree_iter iter;
 826         struct bkey_s_c k;
 827         struct bkey_s_c_subvolume subvol;
 828         struct btree_trans_commit_hook *h;
 829         u32 snapid;
 830         int ret = 0;
 831
 832         bch2_trans_iter_init(trans, &iter, BTREE_ID_subvolumes,
 833                              POS(0, subvolid),
 834                              BTREE_ITER_CACHED|
 835                              BTREE_ITER_INTENT);
 836         k = bch2_btree_iter_peek_slot(&iter);
 837         ret = bkey_err(k);
 838         if (ret)
 839                 goto err;
 840
 841         if (k.k->type != KEY_TYPE_subvolume) {
 842                 bch2_fs_inconsistent(trans->c, "missing subvolume %u", subvolid);
 843                 ret = -EIO;
 844                 goto err;
 845         }
 846
 847         subvol = bkey_s_c_to_subvolume(k);
 848         snapid = le32_to_cpu(subvol.v->snapshot);
 849
 850         ret = bch2_btree_delete_at(trans, &iter, 0);
 851         if (ret)
 852                 goto err;
 853
 854         ret = bch2_snapshot_node_set_deleted(trans, snapid);
 855         if (ret)
 856                 goto err;
 857
 858         h = bch2_trans_kmalloc(trans, sizeof(*h));
 859         ret = PTR_ERR_OR_ZERO(h);
 860         if (ret)
 861                 goto err;
 862
 863         h->fn = bch2_delete_dead_snapshots_hook;
 864         bch2_trans_commit_hook(trans, h);
 865 err:
 866         bch2_trans_iter_exit(trans, &iter);
 867         return ret;
 868 }
 869
 870 void bch2_subvolume_wait_for_pagecache_and_delete(struct work_struct *work)
 871 {
 872         struct bch_fs *c = container_of(work, struct bch_fs,
 873                                 snapshot_wait_for_pagecache_and_delete_work);
 874         snapshot_id_list s;
 875         u32 *id;
 876         int ret = 0;
 877
 878         while (!ret) {
 879                 mutex_lock(&c->snapshots_unlinked_lock);
 880                 s = c->snapshots_unlinked;
 881                 darray_init(&c->snapshots_unlinked);
 882                 mutex_unlock(&c->snapshots_unlinked_lock);
 883
 884                 if (!s.nr)
 885                         break;
 886
 887                 bch2_evict_subvolume_inodes(c, &s);
 888
 889                 for (id = s.data; id < s.data + s.nr; id++) {
 890                         ret = bch2_trans_do(c, NULL, NULL, BTREE_INSERT_NOFAIL,
 891                                       bch2_subvolume_delete(&trans, *id));
 892                         if (ret) {
 893                                 bch_err(c, "error deleting subvolume %u: %s", *id, bch2_err_str(ret));
 894                                 break;
 895                         }
 896                 }
 897
 898                 darray_exit(&s);
 899         }
 900
 901         bch2_write_ref_put(c, BCH_WRITE_REF_snapshot_delete_pagecache);
 902 }
 903
 904 struct subvolume_unlink_hook {
 905         struct btree_trans_commit_hook  h;
 906         u32                             subvol;
 907 };
 908
 909 int bch2_subvolume_wait_for_pagecache_and_delete_hook(struct btree_trans *trans,
 910                                                       struct btree_trans_commit_hook *_h)
 911 {
 912         struct subvolume_unlink_hook *h = container_of(_h, struct subvolume_unlink_hook, h);
 913         struct bch_fs *c = trans->c;
 914         int ret = 0;
 915
 916         mutex_lock(&c->snapshots_unlinked_lock);
 917         if (!snapshot_list_has_id(&c->snapshots_unlinked, h->subvol))
 918                 ret = snapshot_list_add(c, &c->snapshots_unlinked, h->subvol);
 919         mutex_unlock(&c->snapshots_unlinked_lock);
 920
 921         if (ret)
 922                 return ret;
 923
 924         if (!bch2_write_ref_tryget(c, BCH_WRITE_REF_snapshot_delete_pagecache))
 925                 return -EROFS;
 926
 927         if (!queue_work(system_long_wq, &c->snapshot_wait_for_pagecache_and_delete_work))
 928                 bch2_write_ref_put(c, BCH_WRITE_REF_snapshot_delete_pagecache);
 929         return 0;
 930 }
 931
 932 int bch2_subvolume_unlink(struct btree_trans *trans, u32 subvolid)
 933 {
 934         struct btree_iter iter;
 935         struct bkey_i_subvolume *n;
 936         struct subvolume_unlink_hook *h;
 937         int ret = 0;
 938
 939         bch2_trans_iter_init(trans, &iter, BTREE_ID_subvolumes,
 940                              POS(0, subvolid),
 941                              BTREE_ITER_CACHED|
 942                              BTREE_ITER_INTENT);
 943         n = bch2_bkey_get_mut_typed(trans, &iter, subvolume);
 944         ret = PTR_ERR_OR_ZERO(n);
 945         if (unlikely(ret)) {
 946                 bch2_fs_inconsistent_on(ret == -ENOENT, trans->c, "missing subvolume %u", subvolid);
 947                 goto err;
 948         }
 949
 950         SET_BCH_SUBVOLUME_UNLINKED(&n->v, true);
 951
 952         ret = bch2_trans_update(trans, &iter, &n->k_i, 0);
 953         if (ret)
 954                 goto err;
 955
 956         h = bch2_trans_kmalloc(trans, sizeof(*h));
 957         ret = PTR_ERR_OR_ZERO(h);
 958         if (ret)
 959                 goto err;
 960
 961         h->h.fn         = bch2_subvolume_wait_for_pagecache_and_delete_hook;
 962         h->subvol       = subvolid;
 963         bch2_trans_commit_hook(trans, &h->h);
 964 err:
 965         bch2_trans_iter_exit(trans, &iter);
 966         return ret;
 967 }
 968
 969 int bch2_subvolume_create(struct btree_trans *trans, u64 inode,
 970                           u32 src_subvolid,
 971                           u32 *new_subvolid,
 972                           u32 *new_snapshotid,
 973                           bool ro)
 974 {
 975         struct bch_fs *c = trans->c;
 976         struct btree_iter dst_iter, src_iter = (struct btree_iter) { NULL };
 977         struct bkey_i_subvolume *new_subvol = NULL;
 978         struct bkey_i_subvolume *src_subvol = NULL;
 979         struct bkey_s_c k;
 980         u32 parent = 0, new_nodes[2], snapshot_subvols[2];
 981         int ret = 0;
 982
 983         for_each_btree_key(trans, dst_iter, BTREE_ID_subvolumes, SUBVOL_POS_MIN,
 984                            BTREE_ITER_SLOTS|BTREE_ITER_INTENT, k, ret) {
 985                 if (bkey_gt(k.k->p, SUBVOL_POS_MAX))
 986                         break;
 987
 988                 /*
 989                  * bch2_subvolume_delete() doesn't flush the btree key cache -
 990                  * ideally it would but that's tricky
 991                  */
 992                 if (bkey_deleted(k.k) &&
 993                     !bch2_btree_key_cache_find(c, BTREE_ID_subvolumes, dst_iter.pos))
 994                         goto found_slot;
 995         }
 996
 997         if (!ret)
 998                 ret = -BCH_ERR_ENOSPC_subvolume_create;
 999         goto err;
1000 found_slot:
1001         snapshot_subvols[0] = dst_iter.pos.offset;
1002         snapshot_subvols[1] = src_subvolid;
1003
1004         if (src_subvolid) {
1005                 /* Creating a snapshot: */
1006
1007                 bch2_trans_iter_init(trans, &src_iter, BTREE_ID_subvolumes,
1008                                      POS(0, src_subvolid),
1009                                      BTREE_ITER_CACHED|
1010                                      BTREE_ITER_INTENT);
1011                 src_subvol = bch2_bkey_get_mut_typed(trans, &src_iter, subvolume);
1012                 ret = PTR_ERR_OR_ZERO(src_subvol);
1013                 if (unlikely(ret)) {
1014                         bch2_fs_inconsistent_on(ret == -ENOENT, trans->c,
1015                                                 "subvolume %u not found", src_subvolid);
1016                         goto err;
1017                 }
1018
1019                 parent = le32_to_cpu(src_subvol->v.snapshot);
1020         }
1021
1022         ret = bch2_snapshot_node_create(trans, parent, new_nodes,
1023                                         snapshot_subvols,
1024                                         src_subvolid ? 2 : 1);
1025         if (ret)
1026                 goto err;
1027
1028         if (src_subvolid) {
1029                 src_subvol->v.snapshot = cpu_to_le32(new_nodes[1]);
1030                 ret = bch2_trans_update(trans, &src_iter, &src_subvol->k_i, 0);
1031                 if (ret)
1032                         goto err;
1033         }
1034
1035         new_subvol = bch2_bkey_alloc(trans, &dst_iter, subvolume);
1036         ret = PTR_ERR_OR_ZERO(new_subvol);
1037         if (ret)
1038                 goto err;
1039
1040         new_subvol->v.flags     = 0;
1041         new_subvol->v.snapshot  = cpu_to_le32(new_nodes[0]);
1042         new_subvol->v.inode     = cpu_to_le64(inode);
1043         SET_BCH_SUBVOLUME_RO(&new_subvol->v, ro);
1044         SET_BCH_SUBVOLUME_SNAP(&new_subvol->v, src_subvolid != 0);
1045         ret = bch2_trans_update(trans, &dst_iter, &new_subvol->k_i, 0);
1046         if (ret)
1047                 goto err;
1048
1049         *new_subvolid   = new_subvol->k.p.offset;
1050         *new_snapshotid = new_nodes[0];
1051 err:
1052         bch2_trans_iter_exit(trans, &src_iter);
1053         bch2_trans_iter_exit(trans, &dst_iter);
1054         return ret;
1055 }
1056
1057 int bch2_fs_subvolumes_init(struct bch_fs *c)
1058 {
1059         INIT_WORK(&c->snapshot_delete_work, bch2_delete_dead_snapshots_work);
1060         INIT_WORK(&c->snapshot_wait_for_pagecache_and_delete_work,
1061                   bch2_subvolume_wait_for_pagecache_and_delete);
1062         mutex_init(&c->snapshots_unlinked_lock);
1063         return 0;
1064 }