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 "error.h"
   7 #include "fs.h"
   8 #include "subvolume.h"
   9
  10 /* Snapshot tree: */
  11
  12 static void bch2_delete_dead_snapshots_work(struct work_struct *);
  13 static void bch2_delete_dead_snapshots(struct bch_fs *);
  14
  15 void bch2_snapshot_to_text(struct printbuf *out, struct bch_fs *c,
  16                            struct bkey_s_c k)
  17 {
  18         struct bkey_s_c_snapshot s = bkey_s_c_to_snapshot(k);
  19
  20         pr_buf(out, "is_subvol %llu deleted %llu parent %u children %u %u subvol %u",
  21                BCH_SNAPSHOT_SUBVOL(s.v),
  22                BCH_SNAPSHOT_DELETED(s.v),
  23                le32_to_cpu(s.v->parent),
  24                le32_to_cpu(s.v->children[0]),
  25                le32_to_cpu(s.v->children[1]),
  26                le32_to_cpu(s.v->subvol));
  27 }
  28
  29 const char *bch2_snapshot_invalid(const struct bch_fs *c, struct bkey_s_c k)
  30 {
  31         struct bkey_s_c_snapshot s;
  32         u32 i, id;
  33
  34         if (bkey_cmp(k.k->p, POS(0, U32_MAX)) > 0 ||
  35             bkey_cmp(k.k->p, POS(0, 1)) < 0)
  36                 return "bad pos";
  37
  38         if (bkey_val_bytes(k.k) != sizeof(struct bch_snapshot))
  39                 return "bad val size";
  40
  41         s = bkey_s_c_to_snapshot(k);
  42
  43         id = le32_to_cpu(s.v->parent);
  44         if (id && id <= k.k->p.offset)
  45                 return "bad parent node";
  46
  47         if (le32_to_cpu(s.v->children[0]) < le32_to_cpu(s.v->children[1]))
  48                 return "children not normalized";
  49
  50         if (s.v->children[0] &&
  51             s.v->children[0] == s.v->children[1])
  52                 return "duplicate child nodes";
  53
  54         for (i = 0; i < 2; i++) {
  55                 id = le32_to_cpu(s.v->children[i]);
  56
  57                 if (id >= k.k->p.offset)
  58                         return "bad child node";
  59         }
  60
  61         return NULL;
  62 }
  63
  64 int bch2_mark_snapshot(struct btree_trans *trans,
  65                        struct bkey_s_c old, struct bkey_s_c new,
  66                        unsigned flags)
  67 {
  68         struct bch_fs *c = trans->c;
  69         struct snapshot_t *t;
  70
  71         t = genradix_ptr_alloc(&c->snapshots,
  72                                U32_MAX - new.k->p.offset,
  73                                GFP_KERNEL);
  74         if (!t)
  75                 return -ENOMEM;
  76
  77         if (new.k->type == KEY_TYPE_snapshot) {
  78                 struct bkey_s_c_snapshot s = bkey_s_c_to_snapshot(new);
  79
  80                 t->parent       = le32_to_cpu(s.v->parent);
  81                 t->children[0]  = le32_to_cpu(s.v->children[0]);
  82                 t->children[1]  = le32_to_cpu(s.v->children[1]);
  83                 t->subvol       = BCH_SNAPSHOT_SUBVOL(s.v) ? le32_to_cpu(s.v->subvol) : 0;
  84         } else {
  85                 t->parent       = 0;
  86                 t->children[0]  = 0;
  87                 t->children[1]  = 0;
  88                 t->subvol       = 0;
  89         }
  90
  91         return 0;
  92 }
  93
  94 static int snapshot_lookup(struct btree_trans *trans, u32 id,
  95                            struct bch_snapshot *s)
  96 {
  97         struct btree_iter iter;
  98         struct bkey_s_c k;
  99         int ret;
 100
 101         bch2_trans_iter_init(trans, &iter, BTREE_ID_snapshots, POS(0, id),
 102                              BTREE_ITER_WITH_UPDATES);
 103         k = bch2_btree_iter_peek_slot(&iter);
 104         ret = bkey_err(k) ?: k.k->type == KEY_TYPE_snapshot ? 0 : -ENOENT;
 105
 106         if (!ret)
 107                 *s = *bkey_s_c_to_snapshot(k).v;
 108
 109         bch2_trans_iter_exit(trans, &iter);
 110         return ret;
 111 }
 112
 113 static int snapshot_live(struct btree_trans *trans, u32 id)
 114 {
 115         struct bch_snapshot v;
 116         int ret;
 117
 118         if (!id)
 119                 return 0;
 120
 121         ret = lockrestart_do(trans, snapshot_lookup(trans, id, &v));
 122         if (ret == -ENOENT)
 123                 bch_err(trans->c, "snapshot node %u not found", id);
 124         if (ret)
 125                 return ret;
 126
 127         return !BCH_SNAPSHOT_DELETED(&v);
 128 }
 129
 130 static int bch2_snapshots_set_equiv(struct btree_trans *trans)
 131 {
 132         struct bch_fs *c = trans->c;
 133         struct btree_iter iter;
 134         struct bkey_s_c k;
 135         struct bkey_s_c_snapshot snap;
 136         unsigned i;
 137         int ret;
 138
 139         for_each_btree_key(trans, iter, BTREE_ID_snapshots,
 140                            POS_MIN, 0, k, ret) {
 141                 u32 id = k.k->p.offset, child[2];
 142                 unsigned nr_live = 0, live_idx;
 143
 144                 if (k.k->type != KEY_TYPE_snapshot)
 145                         continue;
 146
 147                 snap = bkey_s_c_to_snapshot(k);
 148                 child[0] = le32_to_cpu(snap.v->children[0]);
 149                 child[1] = le32_to_cpu(snap.v->children[1]);
 150
 151                 for (i = 0; i < 2; i++) {
 152                         ret = snapshot_live(trans, child[i]);
 153                         if (ret < 0)
 154                                 break;
 155
 156                         if (ret)
 157                                 live_idx = i;
 158                         nr_live += ret;
 159                 }
 160
 161                 snapshot_t(c, id)->equiv = nr_live == 1
 162                         ? snapshot_t(c, child[live_idx])->equiv
 163                         : id;
 164         }
 165         bch2_trans_iter_exit(trans, &iter);
 166
 167         if (ret)
 168                 bch_err(c, "error walking snapshots: %i", ret);
 169
 170         return ret;
 171 }
 172
 173 /* fsck: */
 174 static int bch2_snapshot_check(struct btree_trans *trans,
 175                                struct bkey_s_c_snapshot s)
 176 {
 177         struct bch_subvolume subvol;
 178         struct bch_snapshot v;
 179         u32 i, id;
 180         int ret;
 181
 182         id = le32_to_cpu(s.v->subvol);
 183         ret = lockrestart_do(trans, bch2_subvolume_get(trans, id, 0, false, &subvol));
 184         if (ret == -ENOENT)
 185                 bch_err(trans->c, "snapshot node %llu has nonexistent subvolume %u",
 186                         s.k->p.offset, id);
 187         if (ret)
 188                 return ret;
 189
 190         if (BCH_SNAPSHOT_SUBVOL(s.v) != (le32_to_cpu(subvol.snapshot) == s.k->p.offset)) {
 191                 bch_err(trans->c, "snapshot node %llu has wrong BCH_SNAPSHOT_SUBVOL",
 192                         s.k->p.offset);
 193                 return -EINVAL;
 194         }
 195
 196         id = le32_to_cpu(s.v->parent);
 197         if (id) {
 198                 ret = lockrestart_do(trans, snapshot_lookup(trans, id, &v));
 199                 if (ret == -ENOENT)
 200                         bch_err(trans->c, "snapshot node %llu has nonexistent parent %u",
 201                                 s.k->p.offset, id);
 202                 if (ret)
 203                         return ret;
 204
 205                 if (le32_to_cpu(v.children[0]) != s.k->p.offset &&
 206                     le32_to_cpu(v.children[1]) != s.k->p.offset) {
 207                         bch_err(trans->c, "snapshot parent %u missing pointer to child %llu",
 208                                 id, s.k->p.offset);
 209                         return -EINVAL;
 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 = lockrestart_do(trans, snapshot_lookup(trans, id, &v));
 217                 if (ret == -ENOENT)
 218                         bch_err(trans->c, "snapshot node %llu has nonexistent child %u",
 219                                 s.k->p.offset, id);
 220                 if (ret)
 221                         return ret;
 222
 223                 if (le32_to_cpu(v.parent) != s.k->p.offset) {
 224                         bch_err(trans->c, "snapshot child %u has wrong parent (got %u should be %llu)",
 225                                 id, le32_to_cpu(v.parent), s.k->p.offset);
 226                         return -EINVAL;
 227                 }
 228         }
 229
 230         return 0;
 231 }
 232
 233 int bch2_fs_snapshots_check(struct bch_fs *c)
 234 {
 235         struct btree_trans trans;
 236         struct btree_iter iter;
 237         struct bkey_s_c k;
 238         struct bch_snapshot s;
 239         unsigned id;
 240         int ret;
 241
 242         bch2_trans_init(&trans, c, 0, 0);
 243
 244         for_each_btree_key(&trans, iter, BTREE_ID_snapshots,
 245                            POS_MIN, 0, k, ret) {
 246                 if (k.k->type != KEY_TYPE_snapshot)
 247                         continue;
 248
 249                 ret = bch2_snapshot_check(&trans, bkey_s_c_to_snapshot(k));
 250                 if (ret)
 251                         break;
 252         }
 253         bch2_trans_iter_exit(&trans, &iter);
 254
 255         if (ret) {
 256                 bch_err(c, "error %i checking snapshots", ret);
 257                 goto err;
 258         }
 259
 260         for_each_btree_key(&trans, iter, BTREE_ID_subvolumes,
 261                            POS_MIN, 0, k, ret) {
 262                 if (k.k->type != KEY_TYPE_subvolume)
 263                         continue;
 264 again_2:
 265                 id = le32_to_cpu(bkey_s_c_to_subvolume(k).v->snapshot);
 266                 ret = snapshot_lookup(&trans, id, &s);
 267
 268                 if (ret == -EINTR) {
 269                         k = bch2_btree_iter_peek(&iter);
 270                         goto again_2;
 271                 } else if (ret == -ENOENT)
 272                         bch_err(c, "subvolume %llu points to nonexistent snapshot %u",
 273                                 k.k->p.offset, id);
 274                 else if (ret)
 275                         break;
 276         }
 277         bch2_trans_iter_exit(&trans, &iter);
 278 err:
 279         bch2_trans_exit(&trans);
 280         return ret;
 281 }
 282
 283 void bch2_fs_snapshots_exit(struct bch_fs *c)
 284 {
 285         genradix_free(&c->snapshots);
 286 }
 287
 288 int bch2_fs_snapshots_start(struct bch_fs *c)
 289 {
 290         struct btree_trans trans;
 291         struct btree_iter iter;
 292         struct bkey_s_c k;
 293         bool have_deleted = false;
 294         int ret = 0;
 295
 296         bch2_trans_init(&trans, c, 0, 0);
 297
 298         for_each_btree_key(&trans, iter, BTREE_ID_snapshots,
 299                            POS_MIN, 0, k, ret) {
 300                if (bkey_cmp(k.k->p, POS(0, U32_MAX)) > 0)
 301                        break;
 302
 303                 if (k.k->type != KEY_TYPE_snapshot) {
 304                         bch_err(c, "found wrong key type %u in snapshot node table",
 305                                 k.k->type);
 306                         continue;
 307                 }
 308
 309                 if (BCH_SNAPSHOT_DELETED(bkey_s_c_to_snapshot(k).v))
 310                         have_deleted = true;
 311
 312                 ret = bch2_mark_snapshot(&trans, bkey_s_c_null, k, 0);
 313                 if (ret)
 314                         break;
 315         }
 316         bch2_trans_iter_exit(&trans, &iter);
 317
 318         if (ret)
 319                 goto err;
 320
 321         ret = bch2_snapshots_set_equiv(&trans);
 322         if (ret)
 323                 goto err;
 324 err:
 325         bch2_trans_exit(&trans);
 326
 327         if (!ret && have_deleted) {
 328                 bch_info(c, "restarting deletion of dead snapshots");
 329                 if (c->opts.fsck) {
 330                         bch2_delete_dead_snapshots_work(&c->snapshot_delete_work);
 331                 } else {
 332                         bch2_delete_dead_snapshots(c);
 333                 }
 334         }
 335
 336         return ret;
 337 }
 338
 339 /*
 340  * Mark a snapshot as deleted, for future cleanup:
 341  */
 342 static int bch2_snapshot_node_set_deleted(struct btree_trans *trans, u32 id)
 343 {
 344         struct btree_iter iter;
 345         struct bkey_s_c k;
 346         struct bkey_i_snapshot *s;
 347         int ret = 0;
 348
 349         bch2_trans_iter_init(trans, &iter, BTREE_ID_snapshots, POS(0, id),
 350                              BTREE_ITER_INTENT);
 351         k = bch2_btree_iter_peek_slot(&iter);
 352         ret = bkey_err(k);
 353         if (ret)
 354                 goto err;
 355
 356         if (k.k->type != KEY_TYPE_snapshot) {
 357                 bch2_fs_inconsistent(trans->c, "missing snapshot %u", id);
 358                 ret = -ENOENT;
 359                 goto err;
 360         }
 361
 362         /* already deleted? */
 363         if (BCH_SNAPSHOT_DELETED(bkey_s_c_to_snapshot(k).v))
 364                 goto err;
 365
 366         s = bch2_trans_kmalloc(trans, sizeof(*s));
 367         ret = PTR_ERR_OR_ZERO(s);
 368         if (ret)
 369                 goto err;
 370
 371         bkey_reassemble(&s->k_i, k);
 372
 373         SET_BCH_SNAPSHOT_DELETED(&s->v, true);
 374         ret = bch2_trans_update(trans, &iter, &s->k_i, 0);
 375         if (ret)
 376                 goto err;
 377 err:
 378         bch2_trans_iter_exit(trans, &iter);
 379         return ret;
 380 }
 381
 382 static int bch2_snapshot_node_delete(struct btree_trans *trans, u32 id)
 383 {
 384         struct btree_iter iter, p_iter = (struct btree_iter) { NULL };
 385         struct bkey_s_c k;
 386         struct bkey_s_c_snapshot s;
 387         struct bkey_i_snapshot *parent;
 388         u32 parent_id;
 389         unsigned i;
 390         int ret = 0;
 391
 392         bch2_trans_iter_init(trans, &iter, BTREE_ID_snapshots, POS(0, id),
 393                              BTREE_ITER_INTENT);
 394         k = bch2_btree_iter_peek_slot(&iter);
 395         ret = bkey_err(k);
 396         if (ret)
 397                 goto err;
 398
 399         if (k.k->type != KEY_TYPE_snapshot) {
 400                 bch2_fs_inconsistent(trans->c, "missing snapshot %u", id);
 401                 ret = -ENOENT;
 402                 goto err;
 403         }
 404
 405         s = bkey_s_c_to_snapshot(k);
 406
 407         BUG_ON(!BCH_SNAPSHOT_DELETED(s.v));
 408         parent_id = le32_to_cpu(s.v->parent);
 409
 410         if (parent_id) {
 411                 bch2_trans_iter_init(trans, &p_iter, BTREE_ID_snapshots,
 412                                      POS(0, parent_id),
 413                                      BTREE_ITER_INTENT);
 414                 k = bch2_btree_iter_peek_slot(&p_iter);
 415                 ret = bkey_err(k);
 416                 if (ret)
 417                         goto err;
 418
 419                 if (k.k->type != KEY_TYPE_snapshot) {
 420                         bch2_fs_inconsistent(trans->c, "missing snapshot %u", parent_id);
 421                         ret = -ENOENT;
 422                         goto err;
 423                 }
 424
 425                 parent = bch2_trans_kmalloc(trans, sizeof(*parent));
 426                 ret = PTR_ERR_OR_ZERO(parent);
 427                 if (ret)
 428                         goto err;
 429
 430                 bkey_reassemble(&parent->k_i, k);
 431
 432                 for (i = 0; i < 2; i++)
 433                         if (le32_to_cpu(parent->v.children[i]) == id)
 434                                 break;
 435
 436                 if (i == 2)
 437                         bch_err(trans->c, "snapshot %u missing child pointer to %u",
 438                                 parent_id, id);
 439                 else
 440                         parent->v.children[i] = 0;
 441
 442                 if (le32_to_cpu(parent->v.children[0]) <
 443                     le32_to_cpu(parent->v.children[1]))
 444                         swap(parent->v.children[0],
 445                              parent->v.children[1]);
 446
 447                 ret = bch2_trans_update(trans, &p_iter, &parent->k_i, 0);
 448                 if (ret)
 449                         goto err;
 450         }
 451
 452         ret = bch2_btree_delete_at(trans, &iter, 0);
 453 err:
 454         bch2_trans_iter_exit(trans, &p_iter);
 455         bch2_trans_iter_exit(trans, &iter);
 456         return ret;
 457 }
 458
 459 static int bch2_snapshot_node_create(struct btree_trans *trans, u32 parent,
 460                                      u32 *new_snapids,
 461                                      u32 *snapshot_subvols,
 462                                      unsigned nr_snapids)
 463 {
 464         struct btree_iter iter;
 465         struct bkey_i_snapshot *n;
 466         struct bkey_s_c k;
 467         unsigned i;
 468         int ret = 0;
 469
 470         bch2_trans_iter_init(trans, &iter, BTREE_ID_snapshots,
 471                              POS_MIN, BTREE_ITER_INTENT);
 472         k = bch2_btree_iter_peek(&iter);
 473         ret = bkey_err(k);
 474         if (ret)
 475                 goto err;
 476
 477         for (i = 0; i < nr_snapids; i++) {
 478                 k = bch2_btree_iter_prev_slot(&iter);
 479                 ret = bkey_err(k);
 480                 if (ret)
 481                         goto err;
 482
 483                 if (!k.k || !k.k->p.offset) {
 484                         ret = -ENOSPC;
 485                         goto err;
 486                 }
 487
 488                 n = bch2_trans_kmalloc(trans, sizeof(*n));
 489                 ret = PTR_ERR_OR_ZERO(n);
 490                 if (ret)
 491                         return ret;
 492
 493                 bkey_snapshot_init(&n->k_i);
 494                 n->k.p          = iter.pos;
 495                 n->v.flags      = 0;
 496                 n->v.parent     = cpu_to_le32(parent);
 497                 n->v.subvol     = cpu_to_le32(snapshot_subvols[i]);
 498                 n->v.pad        = 0;
 499                 SET_BCH_SNAPSHOT_SUBVOL(&n->v, true);
 500
 501                 bch2_trans_update(trans, &iter, &n->k_i, 0);
 502
 503                 ret = bch2_mark_snapshot(trans, bkey_s_c_null, bkey_i_to_s_c(&n->k_i), 0);
 504                 if (ret)
 505                         break;
 506
 507                 new_snapids[i]  = iter.pos.offset;
 508         }
 509
 510         if (parent) {
 511                 bch2_btree_iter_set_pos(&iter, POS(0, parent));
 512                 k = bch2_btree_iter_peek(&iter);
 513                 ret = bkey_err(k);
 514                 if (ret)
 515                         goto err;
 516
 517                 if (k.k->type != KEY_TYPE_snapshot) {
 518                         bch_err(trans->c, "snapshot %u not found", parent);
 519                         ret = -ENOENT;
 520                         goto err;
 521                 }
 522
 523                 n = bch2_trans_kmalloc(trans, sizeof(*n));
 524                 ret = PTR_ERR_OR_ZERO(n);
 525                 if (ret)
 526                         return ret;
 527
 528                 bkey_reassemble(&n->k_i, k);
 529
 530                 if (n->v.children[0] || n->v.children[1]) {
 531                         bch_err(trans->c, "Trying to add child snapshot nodes to parent that already has children");
 532                         ret = -EINVAL;
 533                         goto err;
 534                 }
 535
 536                 n->v.children[0] = cpu_to_le32(new_snapids[0]);
 537                 n->v.children[1] = cpu_to_le32(new_snapids[1]);
 538                 SET_BCH_SNAPSHOT_SUBVOL(&n->v, false);
 539                 bch2_trans_update(trans, &iter, &n->k_i, 0);
 540         }
 541 err:
 542         bch2_trans_iter_exit(trans, &iter);
 543         return ret;
 544 }
 545
 546 static int snapshot_id_add(struct snapshot_id_list *s, u32 id)
 547 {
 548         BUG_ON(snapshot_list_has_id(s, id));
 549
 550         if (s->nr == s->size) {
 551                 size_t new_size = max(8U, s->size * 2);
 552                 void *n = krealloc(s->d,
 553                                    new_size * sizeof(s->d[0]),
 554                                    GFP_KERNEL);
 555                 if (!n) {
 556                         pr_err("error allocating snapshot ID list");
 557                         return -ENOMEM;
 558                 }
 559
 560                 s->d    = n;
 561                 s->size = new_size;
 562         };
 563
 564         s->d[s->nr++] = id;
 565         return 0;
 566 }
 567
 568 static int bch2_snapshot_delete_keys_btree(struct btree_trans *trans,
 569                                            struct snapshot_id_list *deleted,
 570                                            enum btree_id btree_id)
 571 {
 572         struct bch_fs *c = trans->c;
 573         struct btree_iter iter;
 574         struct bkey_s_c k;
 575         struct snapshot_id_list equiv_seen = { 0 };
 576         struct bpos last_pos = POS_MIN;
 577         int ret = 0;
 578
 579         /*
 580          * XXX: We should also delete whiteouts that no longer overwrite
 581          * anything
 582          */
 583
 584         bch2_trans_iter_init(trans, &iter, btree_id, POS_MIN,
 585                              BTREE_ITER_INTENT|
 586                              BTREE_ITER_PREFETCH|
 587                              BTREE_ITER_NOT_EXTENTS|
 588                              BTREE_ITER_ALL_SNAPSHOTS);
 589
 590         while ((bch2_trans_begin(trans),
 591                 (k = bch2_btree_iter_peek(&iter)).k) &&
 592                !(ret = bkey_err(k))) {
 593                 u32 equiv = snapshot_t(c, k.k->p.snapshot)->equiv;
 594
 595                 if (bkey_cmp(k.k->p, last_pos))
 596                         equiv_seen.nr = 0;
 597                 last_pos = k.k->p;
 598
 599                 if (snapshot_list_has_id(deleted, k.k->p.snapshot) ||
 600                     snapshot_list_has_id(&equiv_seen, equiv)) {
 601                         if (btree_id == BTREE_ID_inodes &&
 602                             bch2_btree_key_cache_flush(trans, btree_id, iter.pos))
 603                                 continue;
 604
 605                         ret = __bch2_trans_do(trans, NULL, NULL,
 606                                               BTREE_INSERT_NOFAIL,
 607                                 bch2_btree_iter_traverse(&iter) ?:
 608                                 bch2_btree_delete_at(trans, &iter,
 609                                         BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE));
 610                         if (ret)
 611                                 break;
 612                 } else {
 613                         ret = snapshot_id_add(&equiv_seen, equiv);
 614                         if (ret)
 615                                 break;
 616                 }
 617
 618                 bch2_btree_iter_advance(&iter);
 619         }
 620         bch2_trans_iter_exit(trans, &iter);
 621
 622         kfree(equiv_seen.d);
 623
 624         return ret;
 625 }
 626
 627 static void bch2_delete_dead_snapshots_work(struct work_struct *work)
 628 {
 629         struct bch_fs *c = container_of(work, struct bch_fs, snapshot_delete_work);
 630         struct btree_trans trans;
 631         struct btree_iter iter;
 632         struct bkey_s_c k;
 633         struct bkey_s_c_snapshot snap;
 634         struct snapshot_id_list deleted = { 0 };
 635         u32 i, id, children[2];
 636         int ret = 0;
 637
 638         bch2_trans_init(&trans, c, 0, 0);
 639
 640         /*
 641          * For every snapshot node: If we have no live children and it's not
 642          * pointed to by a subvolume, delete it:
 643          */
 644         for_each_btree_key(&trans, iter, BTREE_ID_snapshots,
 645                            POS_MIN, 0, k, ret) {
 646                 if (k.k->type != KEY_TYPE_snapshot)
 647                         continue;
 648
 649                 snap = bkey_s_c_to_snapshot(k);
 650                 if (BCH_SNAPSHOT_DELETED(snap.v) ||
 651                     BCH_SNAPSHOT_SUBVOL(snap.v))
 652                         continue;
 653
 654                 children[0] = le32_to_cpu(snap.v->children[0]);
 655                 children[1] = le32_to_cpu(snap.v->children[1]);
 656
 657                 ret   = snapshot_live(&trans, children[0]) ?:
 658                         snapshot_live(&trans, children[1]);
 659                 if (ret < 0)
 660                         break;
 661                 if (ret)
 662                         continue;
 663
 664                 ret = __bch2_trans_do(&trans, NULL, NULL, 0,
 665                         bch2_snapshot_node_set_deleted(&trans, iter.pos.offset));
 666                 if (ret) {
 667                         bch_err(c, "error deleting snapshot %llu: %i", iter.pos.offset, ret);
 668                         break;
 669                 }
 670         }
 671         bch2_trans_iter_exit(&trans, &iter);
 672
 673         if (ret) {
 674                 bch_err(c, "error walking snapshots: %i", ret);
 675                 goto err;
 676         }
 677
 678         ret = bch2_snapshots_set_equiv(&trans);
 679         if (ret)
 680                 goto err;
 681
 682         for_each_btree_key(&trans, iter, BTREE_ID_snapshots,
 683                            POS_MIN, 0, k, ret) {
 684                 if (k.k->type != KEY_TYPE_snapshot)
 685                         continue;
 686
 687                 snap = bkey_s_c_to_snapshot(k);
 688                 if (BCH_SNAPSHOT_DELETED(snap.v)) {
 689                         ret = snapshot_id_add(&deleted, k.k->p.offset);
 690                         if (ret)
 691                                 break;
 692                 }
 693         }
 694         bch2_trans_iter_exit(&trans, &iter);
 695
 696         if (ret) {
 697                 bch_err(c, "error walking snapshots: %i", ret);
 698                 goto err;
 699         }
 700
 701         for (id = 0; id < BTREE_ID_NR; id++) {
 702                 if (!btree_type_has_snapshots(id))
 703                         continue;
 704
 705                 ret = bch2_snapshot_delete_keys_btree(&trans, &deleted, id);
 706                 if (ret) {
 707                         bch_err(c, "error deleting snapshot keys: %i", ret);
 708                         goto err;
 709                 }
 710         }
 711
 712         for (i = 0; i < deleted.nr; i++) {
 713                 ret = __bch2_trans_do(&trans, NULL, NULL, 0,
 714                         bch2_snapshot_node_delete(&trans, deleted.d[i]));
 715                 if (ret) {
 716                         bch_err(c, "error deleting snapshot %u: %i",
 717                                 deleted.d[i], ret);
 718                         goto err;
 719                 }
 720         }
 721 err:
 722         kfree(deleted.d);
 723         bch2_trans_exit(&trans);
 724         percpu_ref_put(&c->writes);
 725 }
 726
 727 static void bch2_delete_dead_snapshots(struct bch_fs *c)
 728 {
 729         if (unlikely(!percpu_ref_tryget(&c->writes)))
 730                 return;
 731
 732         if (!queue_work(system_long_wq, &c->snapshot_delete_work))
 733                 percpu_ref_put(&c->writes);
 734 }
 735
 736 static int bch2_delete_dead_snapshots_hook(struct btree_trans *trans,
 737                                            struct btree_trans_commit_hook *h)
 738 {
 739         bch2_delete_dead_snapshots(trans->c);
 740         return 0;
 741 }
 742
 743 /* Subvolumes: */
 744
 745 const char *bch2_subvolume_invalid(const struct bch_fs *c, struct bkey_s_c k)
 746 {
 747         if (bkey_cmp(k.k->p, SUBVOL_POS_MIN) < 0)
 748                 return "invalid pos";
 749
 750         if (bkey_cmp(k.k->p, SUBVOL_POS_MAX) > 0)
 751                 return "invalid pos";
 752
 753         if (bkey_val_bytes(k.k) != sizeof(struct bch_subvolume))
 754                 return "bad val size";
 755
 756         return NULL;
 757 }
 758
 759 void bch2_subvolume_to_text(struct printbuf *out, struct bch_fs *c,
 760                             struct bkey_s_c k)
 761 {
 762         struct bkey_s_c_subvolume s = bkey_s_c_to_subvolume(k);
 763
 764         pr_buf(out, "root %llu snapshot id %u",
 765                le64_to_cpu(s.v->inode),
 766                le32_to_cpu(s.v->snapshot));
 767 }
 768
 769 int bch2_subvolume_get(struct btree_trans *trans, unsigned subvol,
 770                        bool inconsistent_if_not_found,
 771                        int iter_flags,
 772                        struct bch_subvolume *s)
 773 {
 774         struct btree_iter iter;
 775         struct bkey_s_c k;
 776         int ret;
 777
 778         bch2_trans_iter_init(trans, &iter, BTREE_ID_subvolumes, POS(0, subvol),
 779                              iter_flags);
 780         k = bch2_btree_iter_peek_slot(&iter);
 781         ret = bkey_err(k) ?: k.k->type == KEY_TYPE_subvolume ? 0 : -ENOENT;
 782
 783         if (ret == -ENOENT && inconsistent_if_not_found)
 784                 bch2_fs_inconsistent(trans->c, "missing subvolume %u", subvol);
 785         if (!ret)
 786                 *s = *bkey_s_c_to_subvolume(k).v;
 787
 788         bch2_trans_iter_exit(trans, &iter);
 789         return ret;
 790 }
 791
 792 int bch2_snapshot_get_subvol(struct btree_trans *trans, u32 snapshot,
 793                              struct bch_subvolume *subvol)
 794 {
 795         struct bch_snapshot snap;
 796
 797         return  snapshot_lookup(trans, snapshot, &snap) ?:
 798                 bch2_subvolume_get(trans, le32_to_cpu(snap.subvol), true, 0, subvol);
 799 }
 800
 801 int bch2_subvolume_get_snapshot(struct btree_trans *trans, u32 subvol,
 802                                 u32 *snapid)
 803 {
 804         struct bch_subvolume s;
 805         int ret;
 806
 807         ret = bch2_subvolume_get(trans, subvol, true,
 808                                  BTREE_ITER_CACHED|
 809                                  BTREE_ITER_WITH_UPDATES,
 810                                  &s);
 811
 812         *snapid = le32_to_cpu(s.snapshot);
 813         return ret;
 814 }
 815
 816 /*
 817  * Delete subvolume, mark snapshot ID as deleted, queue up snapshot
 818  * deletion/cleanup:
 819  */
 820 int bch2_subvolume_delete(struct btree_trans *trans, u32 subvolid)
 821 {
 822         struct btree_iter iter;
 823         struct bkey_s_c k;
 824         struct bkey_s_c_subvolume subvol;
 825         struct btree_trans_commit_hook *h;
 826         struct bkey_i *delete;
 827         u32 snapid;
 828         int ret = 0;
 829
 830         bch2_trans_iter_init(trans, &iter, BTREE_ID_subvolumes,
 831                              POS(0, subvolid),
 832                              BTREE_ITER_CACHED|
 833                              BTREE_ITER_INTENT);
 834         k = bch2_btree_iter_peek_slot(&iter);
 835         ret = bkey_err(k);
 836         if (ret)
 837                 goto err;
 838
 839         if (k.k->type != KEY_TYPE_subvolume) {
 840                 bch2_fs_inconsistent(trans->c, "missing subvolume %u", subvolid);
 841                 ret = -EIO;
 842                 goto err;
 843         }
 844
 845         subvol = bkey_s_c_to_subvolume(k);
 846         snapid = le32_to_cpu(subvol.v->snapshot);
 847
 848         delete = bch2_trans_kmalloc(trans, sizeof(*delete));
 849         ret = PTR_ERR_OR_ZERO(delete);
 850         if (ret)
 851                 goto err;
 852
 853         bkey_init(&delete->k);
 854         delete->k.p = iter.pos;
 855         ret = bch2_trans_update(trans, &iter, delete, 0);
 856         if (ret)
 857                 goto err;
 858
 859         ret = bch2_snapshot_node_set_deleted(trans, snapid);
 860
 861         h = bch2_trans_kmalloc(trans, sizeof(*h));
 862         ret = PTR_ERR_OR_ZERO(h);
 863         if (ret)
 864                 goto err;
 865
 866         h->fn = bch2_delete_dead_snapshots_hook;
 867         bch2_trans_commit_hook(trans, h);
 868 err:
 869         bch2_trans_iter_exit(trans, &iter);
 870         return ret;
 871 }
 872
 873 void bch2_subvolume_wait_for_pagecache_and_delete(struct work_struct *work)
 874 {
 875         struct bch_fs *c = container_of(work, struct bch_fs,
 876                                 snapshot_wait_for_pagecache_and_delete_work);
 877         struct snapshot_id_list s;
 878         u32 *id;
 879         int ret = 0;
 880
 881         while (!ret) {
 882                 mutex_lock(&c->snapshots_unlinked_lock);
 883                 s = c->snapshots_unlinked;
 884                 memset(&c->snapshots_unlinked, 0, sizeof(c->snapshots_unlinked));
 885                 mutex_unlock(&c->snapshots_unlinked_lock);
 886
 887                 if (!s.nr)
 888                         break;
 889
 890                 bch2_evict_subvolume_inodes(c, &s);
 891
 892                 for (id = s.d; id < s.d + s.nr; id++) {
 893                         ret = bch2_trans_do(c, NULL, NULL, BTREE_INSERT_NOFAIL,
 894                                       bch2_subvolume_delete(&trans, *id));
 895                         if (ret) {
 896                                 bch_err(c, "error %i deleting subvolume %u", ret, *id);
 897                                 break;
 898                         }
 899                 }
 900
 901                 kfree(s.d);
 902         }
 903
 904         percpu_ref_put(&c->writes);
 905 }
 906
 907 struct subvolume_unlink_hook {
 908         struct btree_trans_commit_hook  h;
 909         u32                             subvol;
 910 };
 911
 912 int bch2_subvolume_wait_for_pagecache_and_delete_hook(struct btree_trans *trans,
 913                                                       struct btree_trans_commit_hook *_h)
 914 {
 915         struct subvolume_unlink_hook *h = container_of(_h, struct subvolume_unlink_hook, h);
 916         struct bch_fs *c = trans->c;
 917         int ret = 0;
 918
 919         mutex_lock(&c->snapshots_unlinked_lock);
 920         if (!snapshot_list_has_id(&c->snapshots_unlinked, h->subvol))
 921                 ret = snapshot_id_add(&c->snapshots_unlinked, h->subvol);
 922         mutex_unlock(&c->snapshots_unlinked_lock);
 923
 924         if (ret)
 925                 return ret;
 926
 927         if (unlikely(!percpu_ref_tryget(&c->writes)))
 928                 return -EROFS;
 929
 930         if (!queue_work(system_long_wq, &c->snapshot_wait_for_pagecache_and_delete_work))
 931                 percpu_ref_put(&c->writes);
 932         return 0;
 933 }
 934
 935 int bch2_subvolume_unlink(struct btree_trans *trans, u32 subvolid)
 936 {
 937         struct btree_iter iter;
 938         struct bkey_s_c k;
 939         struct bkey_i_subvolume *n;
 940         struct subvolume_unlink_hook *h;
 941         int ret = 0;
 942
 943         bch2_trans_iter_init(trans, &iter, BTREE_ID_subvolumes,
 944                              POS(0, subvolid),
 945                              BTREE_ITER_CACHED|
 946                              BTREE_ITER_INTENT);
 947         k = bch2_btree_iter_peek_slot(&iter);
 948         ret = bkey_err(k);
 949         if (ret)
 950                 goto err;
 951
 952         if (k.k->type != KEY_TYPE_subvolume) {
 953                 bch2_fs_inconsistent(trans->c, "missing subvolume %u", subvolid);
 954                 ret = -EIO;
 955                 goto err;
 956         }
 957
 958         n = bch2_trans_kmalloc(trans, sizeof(*n));
 959         ret = PTR_ERR_OR_ZERO(n);
 960         if (ret)
 961                 goto err;
 962
 963         bkey_reassemble(&n->k_i, k);
 964         SET_BCH_SUBVOLUME_UNLINKED(&n->v, true);
 965
 966         ret = bch2_trans_update(trans, &iter, &n->k_i, 0);
 967         if (ret)
 968                 goto err;
 969
 970         h = bch2_trans_kmalloc(trans, sizeof(*h));
 971         ret = PTR_ERR_OR_ZERO(h);
 972         if (ret)
 973                 goto err;
 974
 975         h->h.fn         = bch2_subvolume_wait_for_pagecache_and_delete_hook;
 976         h->subvol       = subvolid;
 977         bch2_trans_commit_hook(trans, &h->h);
 978 err:
 979         bch2_trans_iter_exit(trans, &iter);
 980         return ret;
 981 }
 982
 983 int bch2_subvolume_create(struct btree_trans *trans, u64 inode,
 984                           u32 src_subvolid,
 985                           u32 *new_subvolid,
 986                           u32 *new_snapshotid,
 987                           bool ro)
 988 {
 989         struct bch_fs *c = trans->c;
 990         struct btree_iter dst_iter, src_iter = (struct btree_iter) { NULL };
 991         struct bkey_i_subvolume *new_subvol = NULL;
 992         struct bkey_i_subvolume *src_subvol = NULL;
 993         struct bkey_s_c k;
 994         u32 parent = 0, new_nodes[2], snapshot_subvols[2];
 995         int ret = 0;
 996
 997         for_each_btree_key(trans, dst_iter, BTREE_ID_subvolumes, SUBVOL_POS_MIN,
 998                            BTREE_ITER_SLOTS|BTREE_ITER_INTENT, k, ret) {
 999                 if (bkey_cmp(k.k->p, SUBVOL_POS_MAX) > 0)
1000                         break;
1001
1002                 /*
1003                  * bch2_subvolume_delete() doesn't flush the btree key cache -
1004                  * ideally it would but that's tricky
1005                  */
1006                 if (bkey_deleted(k.k) &&
1007                     !bch2_btree_key_cache_find(c, BTREE_ID_subvolumes, dst_iter.pos))
1008                         goto found_slot;
1009         }
1010
1011         if (!ret)
1012                 ret = -ENOSPC;
1013         goto err;
1014 found_slot:
1015         snapshot_subvols[0] = dst_iter.pos.offset;
1016         snapshot_subvols[1] = src_subvolid;
1017
1018         if (src_subvolid) {
1019                 /* Creating a snapshot: */
1020                 src_subvol = bch2_trans_kmalloc(trans, sizeof(*src_subvol));
1021                 ret = PTR_ERR_OR_ZERO(src_subvol);
1022                 if (ret)
1023                         goto err;
1024
1025                 bch2_trans_iter_init(trans, &src_iter, BTREE_ID_subvolumes,
1026                                      POS(0, src_subvolid),
1027                                      BTREE_ITER_CACHED|
1028                                      BTREE_ITER_INTENT);
1029                 k = bch2_btree_iter_peek_slot(&src_iter);
1030                 ret = bkey_err(k);
1031                 if (ret)
1032                         goto err;
1033
1034                 if (k.k->type != KEY_TYPE_subvolume) {
1035                         bch_err(c, "subvolume %u not found", src_subvolid);
1036                         ret = -ENOENT;
1037                         goto err;
1038                 }
1039
1040                 bkey_reassemble(&src_subvol->k_i, k);
1041                 parent = le32_to_cpu(src_subvol->v.snapshot);
1042         }
1043
1044         ret = bch2_snapshot_node_create(trans, parent, new_nodes,
1045                                         snapshot_subvols,
1046                                         src_subvolid ? 2 : 1);
1047         if (ret)
1048                 goto err;
1049
1050         if (src_subvolid) {
1051                 src_subvol->v.snapshot = cpu_to_le32(new_nodes[1]);
1052                 bch2_trans_update(trans, &src_iter, &src_subvol->k_i, 0);
1053         }
1054
1055         new_subvol = bch2_trans_kmalloc(trans, sizeof(*new_subvol));
1056         ret = PTR_ERR_OR_ZERO(new_subvol);
1057         if (ret)
1058                 goto err;
1059
1060         bkey_subvolume_init(&new_subvol->k_i);
1061         new_subvol->v.flags     = 0;
1062         new_subvol->v.snapshot  = cpu_to_le32(new_nodes[0]);
1063         new_subvol->v.inode     = cpu_to_le64(inode);
1064         SET_BCH_SUBVOLUME_RO(&new_subvol->v, ro);
1065         SET_BCH_SUBVOLUME_SNAP(&new_subvol->v, src_subvolid != 0);
1066         new_subvol->k.p         = dst_iter.pos;
1067         bch2_trans_update(trans, &dst_iter, &new_subvol->k_i, 0);
1068
1069         *new_subvolid   = new_subvol->k.p.offset;
1070         *new_snapshotid = new_nodes[0];
1071 err:
1072         bch2_trans_iter_exit(trans, &src_iter);
1073         bch2_trans_iter_exit(trans, &dst_iter);
1074         return ret;
1075 }
1076
1077 int bch2_fs_subvolumes_init(struct bch_fs *c)
1078 {
1079         INIT_WORK(&c->snapshot_delete_work, bch2_delete_dead_snapshots_work);
1080         INIT_WORK(&c->snapshot_wait_for_pagecache_and_delete_work,
1081                   bch2_subvolume_wait_for_pagecache_and_delete);
1082         mutex_init(&c->snapshots_unlinked_lock);
1083         return 0;
1084 }