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 "subvolume.h"
   8
   9 /* Snapshot tree: */
  10
  11 static void bch2_delete_dead_snapshots_work(struct work_struct *);
  12 static void bch2_delete_dead_snapshots(struct bch_fs *);
  13
  14 void bch2_snapshot_to_text(struct printbuf *out, struct bch_fs *c,
  15                            struct bkey_s_c k)
  16 {
  17         struct bkey_s_c_snapshot s = bkey_s_c_to_snapshot(k);
  18
  19         pr_buf(out, "is_subvol %llu deleted %llu parent %u children %u %u subvol %u",
  20                BCH_SNAPSHOT_SUBVOL(s.v),
  21                BCH_SNAPSHOT_DELETED(s.v),
  22                le32_to_cpu(s.v->parent),
  23                le32_to_cpu(s.v->children[0]),
  24                le32_to_cpu(s.v->children[1]),
  25                le32_to_cpu(s.v->subvol));
  26 }
  27
  28 const char *bch2_snapshot_invalid(const struct bch_fs *c, struct bkey_s_c k)
  29 {
  30         struct bkey_s_c_snapshot s;
  31         u32 i, id;
  32
  33         if (bkey_cmp(k.k->p, POS(0, U32_MAX)) > 0 ||
  34             bkey_cmp(k.k->p, POS(0, 1)) < 0)
  35                 return "bad pos";
  36
  37         if (bkey_val_bytes(k.k) != sizeof(struct bch_snapshot))
  38                 return "bad val size";
  39
  40         s = bkey_s_c_to_snapshot(k);
  41
  42         id = le32_to_cpu(s.v->parent);
  43         if (id && id <= k.k->p.offset)
  44                 return "bad parent node";
  45
  46         if (le32_to_cpu(s.v->children[0]) < le32_to_cpu(s.v->children[1]))
  47                 return "children not normalized";
  48
  49         if (s.v->children[0] &&
  50             s.v->children[0] == s.v->children[1])
  51                 return "duplicate child nodes";
  52
  53         for (i = 0; i < 2; i++) {
  54                 id = le32_to_cpu(s.v->children[i]);
  55
  56                 if (id >= k.k->p.offset)
  57                         return "bad child node";
  58         }
  59
  60         return NULL;
  61 }
  62
  63 int bch2_mark_snapshot(struct bch_fs *c,
  64                        struct bkey_s_c old, struct bkey_s_c new,
  65                        u64 journal_seq, unsigned flags)
  66 {
  67         struct snapshot_t *t;
  68
  69         t = genradix_ptr_alloc(&c->snapshots,
  70                                U32_MAX - new.k->p.offset,
  71                                GFP_KERNEL);
  72         if (!t)
  73                 return -ENOMEM;
  74
  75         if (new.k->type == KEY_TYPE_snapshot) {
  76                 struct bkey_s_c_snapshot s = bkey_s_c_to_snapshot(new);
  77
  78                 t->parent       = le32_to_cpu(s.v->parent);
  79                 t->children[0]  = le32_to_cpu(s.v->children[0]);
  80                 t->children[1]  = le32_to_cpu(s.v->children[1]);
  81                 t->subvol       = BCH_SNAPSHOT_SUBVOL(s.v) ? le32_to_cpu(s.v->subvol) : 0;
  82         } else {
  83                 t->parent       = 0;
  84                 t->children[0]  = 0;
  85                 t->children[1]  = 0;
  86                 t->subvol       = 0;
  87         }
  88
  89         return 0;
  90 }
  91
  92 static int snapshot_lookup(struct btree_trans *trans, u32 id,
  93                            struct bch_snapshot *s)
  94 {
  95         struct btree_iter iter;
  96         struct bkey_s_c k;
  97         int ret;
  98
  99         bch2_trans_iter_init(trans, &iter, BTREE_ID_snapshots, POS(0, id),
 100                              BTREE_ITER_WITH_UPDATES);
 101         k = bch2_btree_iter_peek_slot(&iter);
 102         ret = bkey_err(k) ?: k.k->type == KEY_TYPE_snapshot ? 0 : -ENOENT;
 103
 104         if (!ret)
 105                 *s = *bkey_s_c_to_snapshot(k).v;
 106
 107         bch2_trans_iter_exit(trans, &iter);
 108         return ret;
 109 }
 110
 111 static int snapshot_live(struct btree_trans *trans, u32 id)
 112 {
 113         struct bch_snapshot v;
 114         int ret;
 115
 116         if (!id)
 117                 return 0;
 118
 119         ret = lockrestart_do(trans, snapshot_lookup(trans, id, &v));
 120         if (ret == -ENOENT)
 121                 bch_err(trans->c, "snapshot node %u not found", id);
 122         if (ret)
 123                 return ret;
 124
 125         return !BCH_SNAPSHOT_DELETED(&v);
 126 }
 127
 128 static int bch2_snapshots_set_equiv(struct btree_trans *trans)
 129 {
 130         struct bch_fs *c = trans->c;
 131         struct btree_iter iter;
 132         struct bkey_s_c k;
 133         struct bkey_s_c_snapshot snap;
 134         unsigned i;
 135         int ret;
 136
 137         for_each_btree_key(trans, iter, BTREE_ID_snapshots,
 138                            POS_MIN, 0, k, ret) {
 139                 u32 id = k.k->p.offset, child[2];
 140                 unsigned nr_live = 0, live_idx;
 141
 142                 if (k.k->type != KEY_TYPE_snapshot)
 143                         continue;
 144
 145                 snap = bkey_s_c_to_snapshot(k);
 146                 child[0] = le32_to_cpu(snap.v->children[0]);
 147                 child[1] = le32_to_cpu(snap.v->children[1]);
 148
 149                 for (i = 0; i < 2; i++) {
 150                         ret = snapshot_live(trans, child[i]);
 151                         if (ret < 0)
 152                                 break;
 153
 154                         if (ret)
 155                                 live_idx = i;
 156                         nr_live += ret;
 157                 }
 158
 159                 snapshot_t(c, id)->equiv = nr_live == 1
 160                         ? snapshot_t(c, child[live_idx])->equiv
 161                         : id;
 162         }
 163         bch2_trans_iter_exit(trans, &iter);
 164
 165         if (ret)
 166                 bch_err(c, "error walking snapshots: %i", ret);
 167
 168         return ret;
 169 }
 170
 171 /* fsck: */
 172 static int bch2_snapshot_check(struct btree_trans *trans,
 173                                struct bkey_s_c_snapshot s)
 174 {
 175         struct bch_subvolume subvol;
 176         struct bch_snapshot v;
 177         u32 i, id;
 178         int ret;
 179
 180         id = le32_to_cpu(s.v->subvol);
 181         ret = lockrestart_do(trans, bch2_subvolume_get(trans, id, 0, false, &subvol));
 182         if (ret == -ENOENT)
 183                 bch_err(trans->c, "snapshot node %llu has nonexistent subvolume %u",
 184                         s.k->p.offset, id);
 185         if (ret)
 186                 return ret;
 187
 188         if (BCH_SNAPSHOT_SUBVOL(s.v) != (le32_to_cpu(subvol.snapshot) == s.k->p.offset)) {
 189                 bch_err(trans->c, "snapshot node %llu has wrong BCH_SNAPSHOT_SUBVOL",
 190                         s.k->p.offset);
 191                 return -EINVAL;
 192         }
 193
 194         id = le32_to_cpu(s.v->parent);
 195         if (id) {
 196                 ret = lockrestart_do(trans, snapshot_lookup(trans, id, &v));
 197                 if (ret == -ENOENT)
 198                         bch_err(trans->c, "snapshot node %llu has nonexistent parent %u",
 199                                 s.k->p.offset, id);
 200                 if (ret)
 201                         return ret;
 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(trans->c, "snapshot parent %u missing pointer to child %llu",
 206                                 id, s.k->p.offset);
 207                         return -EINVAL;
 208                 }
 209         }
 210
 211         for (i = 0; i < 2 && s.v->children[i]; i++) {
 212                 id = le32_to_cpu(s.v->children[i]);
 213
 214                 ret = lockrestart_do(trans, snapshot_lookup(trans, id, &v));
 215                 if (ret == -ENOENT)
 216                         bch_err(trans->c, "snapshot node %llu has nonexistent child %u",
 217                                 s.k->p.offset, id);
 218                 if (ret)
 219                         return ret;
 220
 221                 if (le32_to_cpu(v.parent) != s.k->p.offset) {
 222                         bch_err(trans->c, "snapshot child %u has wrong parent (got %u should be %llu)",
 223                                 id, le32_to_cpu(v.parent), s.k->p.offset);
 224                         return -EINVAL;
 225                 }
 226         }
 227
 228         return 0;
 229 }
 230
 231 int bch2_fs_snapshots_check(struct bch_fs *c)
 232 {
 233         struct btree_trans trans;
 234         struct btree_iter iter;
 235         struct bkey_s_c k;
 236         struct bch_snapshot s;
 237         unsigned id;
 238         int ret;
 239
 240         bch2_trans_init(&trans, c, 0, 0);
 241
 242         for_each_btree_key(&trans, iter, BTREE_ID_snapshots,
 243                            POS_MIN, 0, k, ret) {
 244                 if (k.k->type != KEY_TYPE_snapshot)
 245                         continue;
 246
 247                 ret = bch2_snapshot_check(&trans, bkey_s_c_to_snapshot(k));
 248                 if (ret)
 249                         break;
 250         }
 251         bch2_trans_iter_exit(&trans, &iter);
 252
 253         if (ret) {
 254                 bch_err(c, "error %i checking snapshots", ret);
 255                 goto err;
 256         }
 257
 258         for_each_btree_key(&trans, iter, BTREE_ID_subvolumes,
 259                            POS_MIN, 0, k, ret) {
 260                 if (k.k->type != KEY_TYPE_subvolume)
 261                         continue;
 262 again_2:
 263                 id = le32_to_cpu(bkey_s_c_to_subvolume(k).v->snapshot);
 264                 ret = snapshot_lookup(&trans, id, &s);
 265
 266                 if (ret == -EINTR) {
 267                         k = bch2_btree_iter_peek(&iter);
 268                         goto again_2;
 269                 } else if (ret == -ENOENT)
 270                         bch_err(c, "subvolume %llu points to nonexistent snapshot %u",
 271                                 k.k->p.offset, id);
 272                 else if (ret)
 273                         break;
 274         }
 275         bch2_trans_iter_exit(&trans, &iter);
 276 err:
 277         bch2_trans_exit(&trans);
 278         return ret;
 279 }
 280
 281 void bch2_fs_snapshots_exit(struct bch_fs *c)
 282 {
 283         genradix_free(&c->snapshots);
 284 }
 285
 286 int bch2_fs_snapshots_start(struct bch_fs *c)
 287 {
 288         struct btree_trans trans;
 289         struct btree_iter iter;
 290         struct bkey_s_c k;
 291         bool have_deleted = false;
 292         int ret = 0;
 293
 294         bch2_trans_init(&trans, c, 0, 0);
 295
 296         for_each_btree_key(&trans, iter, BTREE_ID_snapshots,
 297                            POS_MIN, 0, k, ret) {
 298                if (bkey_cmp(k.k->p, POS(0, U32_MAX)) > 0)
 299                        break;
 300
 301                 if (k.k->type != KEY_TYPE_snapshot) {
 302                         bch_err(c, "found wrong key type %u in snapshot node table",
 303                                 k.k->type);
 304                         continue;
 305                 }
 306
 307                 if (BCH_SNAPSHOT_DELETED(bkey_s_c_to_snapshot(k).v))
 308                         have_deleted = true;
 309
 310                 ret = bch2_mark_snapshot(c, bkey_s_c_null, k, 0, 0);
 311                 if (ret)
 312                         break;
 313         }
 314         bch2_trans_iter_exit(&trans, &iter);
 315
 316         if (ret)
 317                 goto err;
 318
 319         ret = bch2_snapshots_set_equiv(&trans);
 320         if (ret)
 321                 goto err;
 322 err:
 323         bch2_trans_exit(&trans);
 324
 325         if (!ret && have_deleted) {
 326                 bch_info(c, "restarting deletion of dead snapshots");
 327                 if (c->opts.fsck) {
 328                         bch2_delete_dead_snapshots_work(&c->snapshot_delete_work);
 329                 } else {
 330                         bch2_delete_dead_snapshots(c);
 331                 }
 332         }
 333
 334         return ret;
 335 }
 336
 337 /*
 338  * Mark a snapshot as deleted, for future cleanup:
 339  */
 340 static int bch2_snapshot_node_set_deleted(struct btree_trans *trans, u32 id)
 341 {
 342         struct btree_iter iter;
 343         struct bkey_s_c k;
 344         struct bkey_i_snapshot *s;
 345         int ret = 0;
 346
 347         bch2_trans_iter_init(trans, &iter, BTREE_ID_snapshots, POS(0, id),
 348                              BTREE_ITER_INTENT);
 349         k = bch2_btree_iter_peek_slot(&iter);
 350         ret = bkey_err(k);
 351         if (ret)
 352                 goto err;
 353
 354         if (k.k->type != KEY_TYPE_snapshot) {
 355                 bch2_fs_inconsistent(trans->c, "missing snapshot %u", id);
 356                 ret = -ENOENT;
 357                 goto err;
 358         }
 359
 360         /* already deleted? */
 361         if (BCH_SNAPSHOT_DELETED(bkey_s_c_to_snapshot(k).v))
 362                 goto err;
 363
 364         s = bch2_trans_kmalloc(trans, sizeof(*s));
 365         ret = PTR_ERR_OR_ZERO(s);
 366         if (ret)
 367                 goto err;
 368
 369         bkey_reassemble(&s->k_i, k);
 370
 371         SET_BCH_SNAPSHOT_DELETED(&s->v, true);
 372         ret = bch2_trans_update(trans, &iter, &s->k_i, 0);
 373         if (ret)
 374                 goto err;
 375 err:
 376         bch2_trans_iter_exit(trans, &iter);
 377         return ret;
 378 }
 379
 380 static int bch2_snapshot_node_delete(struct btree_trans *trans, u32 id)
 381 {
 382         struct btree_iter iter, p_iter = (struct btree_iter) { NULL };
 383         struct bkey_s_c k;
 384         struct bkey_s_c_snapshot s;
 385         struct bkey_i_snapshot *parent;
 386         u32 parent_id;
 387         unsigned i;
 388         int ret = 0;
 389
 390         bch2_trans_iter_init(trans, &iter, BTREE_ID_snapshots, POS(0, id),
 391                              BTREE_ITER_INTENT);
 392         k = bch2_btree_iter_peek_slot(&iter);
 393         ret = bkey_err(k);
 394         if (ret)
 395                 goto err;
 396
 397         if (k.k->type != KEY_TYPE_snapshot) {
 398                 bch2_fs_inconsistent(trans->c, "missing snapshot %u", id);
 399                 ret = -ENOENT;
 400                 goto err;
 401         }
 402
 403         s = bkey_s_c_to_snapshot(k);
 404
 405         BUG_ON(!BCH_SNAPSHOT_DELETED(s.v));
 406         parent_id = le32_to_cpu(s.v->parent);
 407
 408         if (parent_id) {
 409                 bch2_trans_iter_init(trans, &p_iter, BTREE_ID_snapshots,
 410                                      POS(0, parent_id),
 411                                      BTREE_ITER_INTENT);
 412                 k = bch2_btree_iter_peek_slot(&p_iter);
 413                 ret = bkey_err(k);
 414                 if (ret)
 415                         goto err;
 416
 417                 if (k.k->type != KEY_TYPE_snapshot) {
 418                         bch2_fs_inconsistent(trans->c, "missing snapshot %u", parent_id);
 419                         ret = -ENOENT;
 420                         goto err;
 421                 }
 422
 423                 parent = bch2_trans_kmalloc(trans, sizeof(*parent));
 424                 ret = PTR_ERR_OR_ZERO(parent);
 425                 if (ret)
 426                         goto err;
 427
 428                 bkey_reassemble(&parent->k_i, k);
 429
 430                 for (i = 0; i < 2; i++)
 431                         if (le32_to_cpu(parent->v.children[i]) == id)
 432                                 break;
 433
 434                 if (i == 2)
 435                         bch_err(trans->c, "snapshot %u missing child pointer to %u",
 436                                 parent_id, id);
 437                 else
 438                         parent->v.children[i] = 0;
 439
 440                 if (le32_to_cpu(parent->v.children[0]) <
 441                     le32_to_cpu(parent->v.children[1]))
 442                         swap(parent->v.children[0],
 443                              parent->v.children[1]);
 444
 445                 ret = bch2_trans_update(trans, &p_iter, &parent->k_i, 0);
 446                 if (ret)
 447                         goto err;
 448         }
 449
 450         ret = bch2_btree_delete_at(trans, &iter, 0);
 451 err:
 452         bch2_trans_iter_exit(trans, &p_iter);
 453         bch2_trans_iter_exit(trans, &iter);
 454         return ret;
 455 }
 456
 457 static int bch2_snapshot_node_create(struct btree_trans *trans, u32 parent,
 458                                      u32 *new_snapids,
 459                                      u32 *snapshot_subvols,
 460                                      unsigned nr_snapids)
 461 {
 462         struct btree_iter iter;
 463         struct bkey_i_snapshot *n;
 464         struct bkey_s_c k;
 465         unsigned i;
 466         int ret = 0;
 467
 468         bch2_trans_iter_init(trans, &iter, BTREE_ID_snapshots,
 469                              POS_MIN, BTREE_ITER_INTENT);
 470         k = bch2_btree_iter_peek(&iter);
 471         ret = bkey_err(k);
 472         if (ret)
 473                 goto err;
 474
 475         for (i = 0; i < nr_snapids; i++) {
 476                 k = bch2_btree_iter_prev_slot(&iter);
 477                 ret = bkey_err(k);
 478                 if (ret)
 479                         goto err;
 480
 481                 if (!k.k || !k.k->p.offset) {
 482                         ret = -ENOSPC;
 483                         goto err;
 484                 }
 485
 486                 n = bch2_trans_kmalloc(trans, sizeof(*n));
 487                 ret = PTR_ERR_OR_ZERO(n);
 488                 if (ret)
 489                         return ret;
 490
 491                 bkey_snapshot_init(&n->k_i);
 492                 n->k.p          = iter.pos;
 493                 n->v.flags      = 0;
 494                 n->v.parent     = cpu_to_le32(parent);
 495                 n->v.subvol     = cpu_to_le32(snapshot_subvols[i]);
 496                 n->v.pad        = 0;
 497                 SET_BCH_SNAPSHOT_SUBVOL(&n->v, true);
 498
 499                 bch2_trans_update(trans, &iter, &n->k_i, 0);
 500
 501                 ret = bch2_mark_snapshot(trans->c, bkey_s_c_null, bkey_i_to_s_c(&n->k_i), 0, 0);
 502                 if (ret)
 503                         break;
 504
 505                 new_snapids[i]  = iter.pos.offset;
 506         }
 507
 508         if (parent) {
 509                 bch2_btree_iter_set_pos(&iter, POS(0, parent));
 510                 k = bch2_btree_iter_peek(&iter);
 511                 ret = bkey_err(k);
 512                 if (ret)
 513                         goto err;
 514
 515                 if (k.k->type != KEY_TYPE_snapshot) {
 516                         bch_err(trans->c, "snapshot %u not found", parent);
 517                         ret = -ENOENT;
 518                         goto err;
 519                 }
 520
 521                 n = bch2_trans_kmalloc(trans, sizeof(*n));
 522                 ret = PTR_ERR_OR_ZERO(n);
 523                 if (ret)
 524                         return ret;
 525
 526                 bkey_reassemble(&n->k_i, k);
 527
 528                 if (n->v.children[0] || n->v.children[1]) {
 529                         bch_err(trans->c, "Trying to add child snapshot nodes to parent that already has children");
 530                         ret = -EINVAL;
 531                         goto err;
 532                 }
 533
 534                 n->v.children[0] = cpu_to_le32(new_snapids[0]);
 535                 n->v.children[1] = cpu_to_le32(new_snapids[1]);
 536                 SET_BCH_SNAPSHOT_SUBVOL(&n->v, false);
 537                 bch2_trans_update(trans, &iter, &n->k_i, 0);
 538         }
 539 err:
 540         bch2_trans_iter_exit(trans, &iter);
 541         return ret;
 542 }
 543
 544 /* List of snapshot IDs that are being deleted: */
 545 struct snapshot_id_list {
 546         u32             nr;
 547         u32             size;
 548         u32             *d;
 549 };
 550
 551 static bool snapshot_list_has_id(struct snapshot_id_list *s, u32 id)
 552 {
 553         unsigned i;
 554
 555         for (i = 0; i < s->nr; i++)
 556                 if (id == s->d[i])
 557                         return true;
 558         return false;
 559 }
 560
 561 static int snapshot_id_add(struct snapshot_id_list *s, u32 id)
 562 {
 563         BUG_ON(snapshot_list_has_id(s, id));
 564
 565         if (s->nr == s->size) {
 566                 size_t new_size = max(8U, s->size * 2);
 567                 void *n = krealloc(s->d,
 568                                    new_size * sizeof(s->d[0]),
 569                                    GFP_KERNEL);
 570                 if (!n) {
 571                         pr_err("error allocating snapshot ID list");
 572                         return -ENOMEM;
 573                 }
 574
 575                 s->d    = n;
 576                 s->size = new_size;
 577         };
 578
 579         s->d[s->nr++] = id;
 580         return 0;
 581 }
 582
 583 static int bch2_snapshot_delete_keys_btree(struct btree_trans *trans,
 584                                            struct snapshot_id_list *deleted,
 585                                            enum btree_id btree_id)
 586 {
 587         struct bch_fs *c = trans->c;
 588         struct btree_iter iter;
 589         struct bkey_s_c k;
 590         struct snapshot_id_list equiv_seen = { 0 };
 591         struct bpos last_pos = POS_MIN;
 592         int ret = 0;
 593
 594         /*
 595          * XXX: We should also delete whiteouts that no longer overwrite
 596          * anything
 597          */
 598
 599         bch2_trans_iter_init(trans, &iter, btree_id, POS_MIN,
 600                              BTREE_ITER_INTENT|
 601                              BTREE_ITER_PREFETCH|
 602                              BTREE_ITER_NOT_EXTENTS|
 603                              BTREE_ITER_ALL_SNAPSHOTS);
 604
 605         while ((bch2_trans_begin(trans),
 606                 (k = bch2_btree_iter_peek(&iter)).k) &&
 607                !(ret = bkey_err(k))) {
 608                 u32 equiv = snapshot_t(c, k.k->p.snapshot)->equiv;
 609
 610                 if (bkey_cmp(k.k->p, last_pos))
 611                         equiv_seen.nr = 0;
 612                 last_pos = k.k->p;
 613
 614                 if (snapshot_list_has_id(deleted, k.k->p.snapshot) ||
 615                     snapshot_list_has_id(&equiv_seen, equiv)) {
 616                         if (btree_id == BTREE_ID_inodes &&
 617                             bch2_btree_key_cache_flush(trans, btree_id, iter.pos))
 618                                 continue;
 619
 620                         ret = __bch2_trans_do(trans, NULL, NULL,
 621                                               BTREE_INSERT_NOFAIL,
 622                                 bch2_btree_iter_traverse(&iter) ?:
 623                                 bch2_btree_delete_at(trans, &iter,
 624                                         BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE));
 625                         if (ret)
 626                                 break;
 627                 } else {
 628                         ret = snapshot_id_add(&equiv_seen, equiv);
 629                         if (ret)
 630                                 break;
 631                 }
 632
 633                 bch2_btree_iter_advance(&iter);
 634         }
 635         bch2_trans_iter_exit(trans, &iter);
 636
 637         kfree(equiv_seen.d);
 638
 639         return ret;
 640 }
 641
 642 static void bch2_delete_dead_snapshots_work(struct work_struct *work)
 643 {
 644         struct bch_fs *c = container_of(work, struct bch_fs, snapshot_delete_work);
 645         struct btree_trans trans;
 646         struct btree_iter iter;
 647         struct bkey_s_c k;
 648         struct bkey_s_c_snapshot snap;
 649         struct snapshot_id_list deleted = { 0 };
 650         u32 i, id, children[2];
 651         int ret = 0;
 652
 653         bch2_trans_init(&trans, c, 0, 0);
 654
 655         /*
 656          * For every snapshot node: If we have no live children and it's not
 657          * pointed to by a subvolume, delete it:
 658          */
 659         for_each_btree_key(&trans, iter, BTREE_ID_snapshots,
 660                            POS_MIN, 0, k, ret) {
 661                 if (k.k->type != KEY_TYPE_snapshot)
 662                         continue;
 663
 664                 snap = bkey_s_c_to_snapshot(k);
 665                 if (BCH_SNAPSHOT_DELETED(snap.v) ||
 666                     BCH_SNAPSHOT_SUBVOL(snap.v))
 667                         continue;
 668
 669                 children[0] = le32_to_cpu(snap.v->children[0]);
 670                 children[1] = le32_to_cpu(snap.v->children[1]);
 671
 672                 ret   = snapshot_live(&trans, children[0]) ?:
 673                         snapshot_live(&trans, children[1]);
 674                 if (ret < 0)
 675                         break;
 676                 if (ret)
 677                         continue;
 678
 679                 ret = __bch2_trans_do(&trans, NULL, NULL, 0,
 680                         bch2_snapshot_node_set_deleted(&trans, iter.pos.offset));
 681                 if (ret) {
 682                         bch_err(c, "error deleting snapshot %llu: %i", iter.pos.offset, ret);
 683                         break;
 684                 }
 685         }
 686         bch2_trans_iter_exit(&trans, &iter);
 687
 688         if (ret) {
 689                 bch_err(c, "error walking snapshots: %i", ret);
 690                 goto err;
 691         }
 692
 693         ret = bch2_snapshots_set_equiv(&trans);
 694         if (ret)
 695                 goto err;
 696
 697         for_each_btree_key(&trans, iter, BTREE_ID_snapshots,
 698                            POS_MIN, 0, k, ret) {
 699                 if (k.k->type != KEY_TYPE_snapshot)
 700                         continue;
 701
 702                 snap = bkey_s_c_to_snapshot(k);
 703                 if (BCH_SNAPSHOT_DELETED(snap.v)) {
 704                         ret = snapshot_id_add(&deleted, k.k->p.offset);
 705                         if (ret)
 706                                 break;
 707                 }
 708         }
 709         bch2_trans_iter_exit(&trans, &iter);
 710
 711         if (ret) {
 712                 bch_err(c, "error walking snapshots: %i", ret);
 713                 goto err;
 714         }
 715
 716         for (id = 0; id < BTREE_ID_NR; id++) {
 717                 if (!btree_type_has_snapshots(id))
 718                         continue;
 719
 720                 ret = bch2_snapshot_delete_keys_btree(&trans, &deleted, id);
 721                 if (ret) {
 722                         bch_err(c, "error deleting snapshot keys: %i", ret);
 723                         goto err;
 724                 }
 725         }
 726
 727         for (i = 0; i < deleted.nr; i++) {
 728                 ret = __bch2_trans_do(&trans, NULL, NULL, 0,
 729                         bch2_snapshot_node_delete(&trans, deleted.d[i]));
 730                 if (ret) {
 731                         bch_err(c, "error deleting snapshot %u: %i",
 732                                 deleted.d[i], ret);
 733                         goto err;
 734                 }
 735         }
 736 err:
 737         kfree(deleted.d);
 738         bch2_trans_exit(&trans);
 739         percpu_ref_put(&c->writes);
 740 }
 741
 742 static void bch2_delete_dead_snapshots(struct bch_fs *c)
 743 {
 744         if (unlikely(!percpu_ref_tryget(&c->writes)))
 745                 return;
 746
 747         if (!queue_work(system_long_wq, &c->snapshot_delete_work))
 748                 percpu_ref_put(&c->writes);
 749 }
 750
 751 static int bch2_delete_dead_snapshots_hook(struct btree_trans *trans,
 752                                            struct btree_trans_commit_hook *h)
 753 {
 754         bch2_delete_dead_snapshots(trans->c);
 755         return 0;
 756 }
 757
 758 /* Subvolumes: */
 759
 760 const char *bch2_subvolume_invalid(const struct bch_fs *c, struct bkey_s_c k)
 761 {
 762         if (bkey_cmp(k.k->p, SUBVOL_POS_MIN) < 0)
 763                 return "invalid pos";
 764
 765         if (bkey_cmp(k.k->p, SUBVOL_POS_MAX) > 0)
 766                 return "invalid pos";
 767
 768         if (bkey_val_bytes(k.k) != sizeof(struct bch_subvolume))
 769                 return "bad val size";
 770
 771         return NULL;
 772 }
 773
 774 void bch2_subvolume_to_text(struct printbuf *out, struct bch_fs *c,
 775                             struct bkey_s_c k)
 776 {
 777         struct bkey_s_c_subvolume s = bkey_s_c_to_subvolume(k);
 778
 779         pr_buf(out, "root %llu snapshot id %u",
 780                le64_to_cpu(s.v->inode),
 781                le32_to_cpu(s.v->snapshot));
 782 }
 783
 784 int bch2_subvolume_get(struct btree_trans *trans, unsigned subvol,
 785                        bool inconsistent_if_not_found,
 786                        int iter_flags,
 787                        struct bch_subvolume *s)
 788 {
 789         struct btree_iter iter;
 790         struct bkey_s_c k;
 791         int ret;
 792
 793         bch2_trans_iter_init(trans, &iter, BTREE_ID_subvolumes, POS(0, subvol),
 794                              iter_flags);
 795         k = bch2_btree_iter_peek_slot(&iter);
 796         ret = bkey_err(k) ?: k.k->type == KEY_TYPE_subvolume ? 0 : -ENOENT;
 797
 798         if (ret == -ENOENT && inconsistent_if_not_found)
 799                 bch2_fs_inconsistent(trans->c, "missing subvolume %u", subvol);
 800         if (!ret)
 801                 *s = *bkey_s_c_to_subvolume(k).v;
 802
 803         bch2_trans_iter_exit(trans, &iter);
 804         return ret;
 805 }
 806
 807 int bch2_subvolume_get_snapshot(struct btree_trans *trans, u32 subvol,
 808                                 u32 *snapid)
 809 {
 810         struct bch_subvolume s;
 811         int ret;
 812
 813         ret = bch2_subvolume_get(trans, subvol, true,
 814                                  BTREE_ITER_CACHED|
 815                                  BTREE_ITER_WITH_UPDATES,
 816                                  &s);
 817
 818         *snapid = le32_to_cpu(s.snapshot);
 819         return ret;
 820 }
 821
 822 /* XXX: mark snapshot id for deletion, walk btree and delete: */
 823 int bch2_subvolume_delete(struct btree_trans *trans, u32 subvolid,
 824                           int deleting_snapshot)
 825 {
 826         struct btree_iter iter;
 827         struct bkey_s_c k;
 828         struct bkey_s_c_subvolume subvol;
 829         struct btree_trans_commit_hook *h;
 830         struct bkey_i *delete;
 831         u32 snapid;
 832         int ret = 0;
 833
 834         bch2_trans_iter_init(trans, &iter, BTREE_ID_subvolumes,
 835                              POS(0, subvolid),
 836                              BTREE_ITER_CACHED|
 837                              BTREE_ITER_INTENT);
 838         k = bch2_btree_iter_peek_slot(&iter);
 839         ret = bkey_err(k);
 840         if (ret)
 841                 goto err;
 842
 843         if (k.k->type != KEY_TYPE_subvolume) {
 844                 bch2_fs_inconsistent(trans->c, "missing subvolume %u", subvolid);
 845                 ret = -EIO;
 846                 goto err;
 847         }
 848
 849         subvol = bkey_s_c_to_subvolume(k);
 850         snapid = le32_to_cpu(subvol.v->snapshot);
 851
 852         if (deleting_snapshot >= 0 &&
 853             deleting_snapshot != BCH_SUBVOLUME_SNAP(subvol.v)) {
 854                 ret = -ENOENT;
 855                 goto err;
 856         }
 857
 858         delete = bch2_trans_kmalloc(trans, sizeof(*delete));
 859         ret = PTR_ERR_OR_ZERO(delete);
 860         if (ret)
 861                 goto err;
 862
 863         bkey_init(&delete->k);
 864         delete->k.p = iter.pos;
 865         ret = bch2_trans_update(trans, &iter, delete, 0);
 866         if (ret)
 867                 goto err;
 868
 869         ret = bch2_snapshot_node_set_deleted(trans, snapid);
 870
 871         h = bch2_trans_kmalloc(trans, sizeof(*h));
 872         ret = PTR_ERR_OR_ZERO(h);
 873         if (ret)
 874                 goto err;
 875
 876         h->fn = bch2_delete_dead_snapshots_hook;
 877         bch2_trans_commit_hook(trans, h);
 878 err:
 879         bch2_trans_iter_exit(trans, &iter);
 880         return ret;
 881 }
 882
 883 int bch2_subvolume_create(struct btree_trans *trans, u64 inode,
 884                           u32 src_subvolid,
 885                           u32 *new_subvolid,
 886                           u32 *new_snapshotid,
 887                           bool ro)
 888 {
 889         struct bch_fs *c = trans->c;
 890         struct btree_iter dst_iter, src_iter = (struct btree_iter) { NULL };
 891         struct bkey_i_subvolume *new_subvol = NULL;
 892         struct bkey_i_subvolume *src_subvol = NULL;
 893         struct bkey_s_c k;
 894         u32 parent = 0, new_nodes[2], snapshot_subvols[2];
 895         int ret = 0;
 896
 897         for_each_btree_key(trans, dst_iter, BTREE_ID_subvolumes, SUBVOL_POS_MIN,
 898                            BTREE_ITER_SLOTS|BTREE_ITER_INTENT, k, ret) {
 899                 if (bkey_cmp(k.k->p, SUBVOL_POS_MAX) > 0)
 900                         break;
 901
 902                 /*
 903                  * bch2_subvolume_delete() doesn't flush the btree key cache -
 904                  * ideally it would but that's tricky
 905                  */
 906                 if (bkey_deleted(k.k) &&
 907                     !bch2_btree_key_cache_find(c, BTREE_ID_subvolumes, dst_iter.pos))
 908                         goto found_slot;
 909         }
 910
 911         if (!ret)
 912                 ret = -ENOSPC;
 913         goto err;
 914 found_slot:
 915         snapshot_subvols[0] = dst_iter.pos.offset;
 916         snapshot_subvols[1] = src_subvolid;
 917
 918         if (src_subvolid) {
 919                 /* Creating a snapshot: */
 920                 src_subvol = bch2_trans_kmalloc(trans, sizeof(*src_subvol));
 921                 ret = PTR_ERR_OR_ZERO(src_subvol);
 922                 if (ret)
 923                         goto err;
 924
 925                 bch2_trans_iter_init(trans, &src_iter, BTREE_ID_subvolumes,
 926                                      POS(0, src_subvolid),
 927                                      BTREE_ITER_CACHED|
 928                                      BTREE_ITER_INTENT);
 929                 k = bch2_btree_iter_peek_slot(&src_iter);
 930                 ret = bkey_err(k);
 931                 if (ret)
 932                         goto err;
 933
 934                 if (k.k->type != KEY_TYPE_subvolume) {
 935                         bch_err(c, "subvolume %u not found", src_subvolid);
 936                         ret = -ENOENT;
 937                         goto err;
 938                 }
 939
 940                 bkey_reassemble(&src_subvol->k_i, k);
 941                 parent = le32_to_cpu(src_subvol->v.snapshot);
 942         }
 943
 944         ret = bch2_snapshot_node_create(trans, parent, new_nodes,
 945                                         snapshot_subvols,
 946                                         src_subvolid ? 2 : 1);
 947         if (ret)
 948                 goto err;
 949
 950         if (src_subvolid) {
 951                 src_subvol->v.snapshot = cpu_to_le32(new_nodes[1]);
 952                 bch2_trans_update(trans, &src_iter, &src_subvol->k_i, 0);
 953         }
 954
 955         new_subvol = bch2_trans_kmalloc(trans, sizeof(*new_subvol));
 956         ret = PTR_ERR_OR_ZERO(new_subvol);
 957         if (ret)
 958                 goto err;
 959
 960         bkey_subvolume_init(&new_subvol->k_i);
 961         new_subvol->v.flags     = 0;
 962         new_subvol->v.snapshot  = cpu_to_le32(new_nodes[0]);
 963         new_subvol->v.inode     = cpu_to_le64(inode);
 964         SET_BCH_SUBVOLUME_RO(&new_subvol->v, ro);
 965         SET_BCH_SUBVOLUME_SNAP(&new_subvol->v, src_subvolid != 0);
 966         new_subvol->k.p         = dst_iter.pos;
 967         bch2_trans_update(trans, &dst_iter, &new_subvol->k_i, 0);
 968
 969         *new_subvolid   = new_subvol->k.p.offset;
 970         *new_snapshotid = new_nodes[0];
 971 err:
 972         bch2_trans_iter_exit(trans, &src_iter);
 973         bch2_trans_iter_exit(trans, &dst_iter);
 974         return ret;
 975 }
 976
 977 int bch2_fs_subvolumes_init(struct bch_fs *c)
 978 {
 979         INIT_WORK(&c->snapshot_delete_work, bch2_delete_dead_snapshots_work);
 980         return 0;
 981 }