git.sesse.net Git - bcachefs-tools-debian/blob - libbcachefs/bkey_methods.c

   1 // SPDX-License-Identifier: GPL-2.0
   2
   3 #include "bcachefs.h"
   4 #include "backpointers.h"
   5 #include "bkey_methods.h"
   6 #include "btree_types.h"
   7 #include "alloc_background.h"
   8 #include "dirent.h"
   9 #include "ec.h"
  10 #include "error.h"
  11 #include "extents.h"
  12 #include "inode.h"
  13 #include "lru.h"
  14 #include "quota.h"
  15 #include "reflink.h"
  16 #include "subvolume.h"
  17 #include "xattr.h"
  18
  19 const char * const bch2_bkey_types[] = {
  20 #define x(name, nr) #name,
  21         BCH_BKEY_TYPES()
  22 #undef x
  23         NULL
  24 };
  25
  26 static int deleted_key_invalid(const struct bch_fs *c, struct bkey_s_c k,
  27                                unsigned flags, struct printbuf *err)
  28 {
  29         return 0;
  30 }
  31
  32 #define bch2_bkey_ops_deleted ((struct bkey_ops) {      \
  33         .key_invalid = deleted_key_invalid,             \
  34 })
  35
  36 #define bch2_bkey_ops_whiteout ((struct bkey_ops) {     \
  37         .key_invalid = deleted_key_invalid,             \
  38 })
  39
  40 static int empty_val_key_invalid(const struct bch_fs *c, struct bkey_s_c k,
  41                                  unsigned flags, struct printbuf *err)
  42 {
  43         if (bkey_val_bytes(k.k)) {
  44                 prt_printf(err, "incorrect value size (%zu != 0)",
  45                        bkey_val_bytes(k.k));
  46                 return -BCH_ERR_invalid_bkey;
  47         }
  48
  49         return 0;
  50 }
  51
  52 #define bch2_bkey_ops_error ((struct bkey_ops) {        \
  53         .key_invalid = empty_val_key_invalid,           \
  54 })
  55
  56 static int key_type_cookie_invalid(const struct bch_fs *c, struct bkey_s_c k,
  57                                    unsigned flags, struct printbuf *err)
  58 {
  59         return 0;
  60 }
  61
  62 #define bch2_bkey_ops_cookie ((struct bkey_ops) {       \
  63         .key_invalid    = key_type_cookie_invalid,      \
  64         .min_val_size   = 8,                            \
  65 })
  66
  67 #define bch2_bkey_ops_hash_whiteout ((struct bkey_ops) {\
  68         .key_invalid = empty_val_key_invalid,           \
  69 })
  70
  71 static int key_type_inline_data_invalid(const struct bch_fs *c, struct bkey_s_c k,
  72                                         unsigned flags, struct printbuf *err)
  73 {
  74         return 0;
  75 }
  76
  77 static void key_type_inline_data_to_text(struct printbuf *out, struct bch_fs *c,
  78                                          struct bkey_s_c k)
  79 {
  80         struct bkey_s_c_inline_data d = bkey_s_c_to_inline_data(k);
  81         unsigned datalen = bkey_inline_data_bytes(k.k);
  82
  83         prt_printf(out, "datalen %u: %*phN",
  84                datalen, min(datalen, 32U), d.v->data);
  85 }
  86
  87 #define bch2_bkey_ops_inline_data ((struct bkey_ops) {  \
  88         .key_invalid    = key_type_inline_data_invalid, \
  89         .val_to_text    = key_type_inline_data_to_text, \
  90 })
  91
  92 static int key_type_set_invalid(const struct bch_fs *c, struct bkey_s_c k,
  93                                 unsigned flags, struct printbuf *err)
  94 {
  95         if (bkey_val_bytes(k.k)) {
  96                 prt_printf(err, "incorrect value size (%zu != %zu)",
  97                        bkey_val_bytes(k.k), sizeof(struct bch_cookie));
  98                 return -BCH_ERR_invalid_bkey;
  99         }
 100
 101         return 0;
 102 }
 103
 104 static bool key_type_set_merge(struct bch_fs *c, struct bkey_s l, struct bkey_s_c r)
 105 {
 106         bch2_key_resize(l.k, l.k->size + r.k->size);
 107         return true;
 108 }
 109
 110 #define bch2_bkey_ops_set ((struct bkey_ops) {          \
 111         .key_invalid    = key_type_set_invalid,         \
 112         .key_merge      = key_type_set_merge,           \
 113 })
 114
 115 const struct bkey_ops bch2_bkey_ops[] = {
 116 #define x(name, nr) [KEY_TYPE_##name]   = bch2_bkey_ops_##name,
 117         BCH_BKEY_TYPES()
 118 #undef x
 119 };
 120
 121 int bch2_bkey_val_invalid(struct bch_fs *c, struct bkey_s_c k,
 122                           unsigned flags, struct printbuf *err)
 123 {
 124         const struct bkey_ops *ops;
 125
 126         if (k.k->type >= KEY_TYPE_MAX) {
 127                 prt_printf(err, "invalid type (%u >= %u)", k.k->type, KEY_TYPE_MAX);
 128                 return -BCH_ERR_invalid_bkey;
 129         }
 130
 131         ops = &bch2_bkey_ops[k.k->type];
 132
 133         if (bkey_val_bytes(k.k) < ops->min_val_size) {
 134                 prt_printf(err, "bad val size (%zu < %u)",
 135                            bkey_val_bytes(k.k), ops->min_val_size);
 136                 return -BCH_ERR_invalid_bkey;
 137         }
 138
 139         return ops->key_invalid(c, k, flags, err);
 140 }
 141
 142 static unsigned bch2_key_types_allowed[] = {
 143         [BKEY_TYPE_extents] =
 144                 (1U << KEY_TYPE_deleted)|
 145                 (1U << KEY_TYPE_whiteout)|
 146                 (1U << KEY_TYPE_error)|
 147                 (1U << KEY_TYPE_cookie)|
 148                 (1U << KEY_TYPE_extent)|
 149                 (1U << KEY_TYPE_reservation)|
 150                 (1U << KEY_TYPE_reflink_p)|
 151                 (1U << KEY_TYPE_inline_data),
 152         [BKEY_TYPE_inodes] =
 153                 (1U << KEY_TYPE_deleted)|
 154                 (1U << KEY_TYPE_whiteout)|
 155                 (1U << KEY_TYPE_inode)|
 156                 (1U << KEY_TYPE_inode_v2)|
 157                 (1U << KEY_TYPE_inode_v3)|
 158                 (1U << KEY_TYPE_inode_generation),
 159         [BKEY_TYPE_dirents] =
 160                 (1U << KEY_TYPE_deleted)|
 161                 (1U << KEY_TYPE_whiteout)|
 162                 (1U << KEY_TYPE_hash_whiteout)|
 163                 (1U << KEY_TYPE_dirent),
 164         [BKEY_TYPE_xattrs] =
 165                 (1U << KEY_TYPE_deleted)|
 166                 (1U << KEY_TYPE_whiteout)|
 167                 (1U << KEY_TYPE_cookie)|
 168                 (1U << KEY_TYPE_hash_whiteout)|
 169                 (1U << KEY_TYPE_xattr),
 170         [BKEY_TYPE_alloc] =
 171                 (1U << KEY_TYPE_deleted)|
 172                 (1U << KEY_TYPE_alloc)|
 173                 (1U << KEY_TYPE_alloc_v2)|
 174                 (1U << KEY_TYPE_alloc_v3)|
 175                 (1U << KEY_TYPE_alloc_v4),
 176         [BKEY_TYPE_quotas] =
 177                 (1U << KEY_TYPE_deleted)|
 178                 (1U << KEY_TYPE_quota),
 179         [BKEY_TYPE_stripes] =
 180                 (1U << KEY_TYPE_deleted)|
 181                 (1U << KEY_TYPE_stripe),
 182         [BKEY_TYPE_reflink] =
 183                 (1U << KEY_TYPE_deleted)|
 184                 (1U << KEY_TYPE_reflink_v)|
 185                 (1U << KEY_TYPE_indirect_inline_data),
 186         [BKEY_TYPE_subvolumes] =
 187                 (1U << KEY_TYPE_deleted)|
 188                 (1U << KEY_TYPE_subvolume),
 189         [BKEY_TYPE_snapshots] =
 190                 (1U << KEY_TYPE_deleted)|
 191                 (1U << KEY_TYPE_snapshot),
 192         [BKEY_TYPE_lru] =
 193                 (1U << KEY_TYPE_deleted)|
 194                 (1U << KEY_TYPE_set),
 195         [BKEY_TYPE_freespace] =
 196                 (1U << KEY_TYPE_deleted)|
 197                 (1U << KEY_TYPE_set),
 198         [BKEY_TYPE_need_discard] =
 199                 (1U << KEY_TYPE_deleted)|
 200                 (1U << KEY_TYPE_set),
 201         [BKEY_TYPE_backpointers] =
 202                 (1U << KEY_TYPE_deleted)|
 203                 (1U << KEY_TYPE_backpointer),
 204         [BKEY_TYPE_bucket_gens] =
 205                 (1U << KEY_TYPE_deleted)|
 206                 (1U << KEY_TYPE_bucket_gens),
 207         [BKEY_TYPE_snapshot_trees] =
 208                 (1U << KEY_TYPE_deleted)|
 209                 (1U << KEY_TYPE_snapshot_tree),
 210         [BKEY_TYPE_btree] =
 211                 (1U << KEY_TYPE_deleted)|
 212                 (1U << KEY_TYPE_btree_ptr)|
 213                 (1U << KEY_TYPE_btree_ptr_v2),
 214 };
 215
 216 int __bch2_bkey_invalid(struct bch_fs *c, struct bkey_s_c k,
 217                         enum btree_node_type type,
 218                         unsigned flags, struct printbuf *err)
 219 {
 220         if (k.k->u64s < BKEY_U64s) {
 221                 prt_printf(err, "u64s too small (%u < %zu)", k.k->u64s, BKEY_U64s);
 222                 return -BCH_ERR_invalid_bkey;
 223         }
 224
 225         if (!(bch2_key_types_allowed[type] & (1U << k.k->type))) {
 226                 prt_printf(err, "invalid key type for btree %s (%s)",
 227                            bch2_btree_ids[type], bch2_bkey_types[k.k->type]);
 228                 return -BCH_ERR_invalid_bkey;
 229         }
 230
 231         if (btree_node_type_is_extents(type) && !bkey_whiteout(k.k)) {
 232                 if (k.k->size == 0) {
 233                         prt_printf(err, "size == 0");
 234                         return -BCH_ERR_invalid_bkey;
 235                 }
 236
 237                 if (k.k->size > k.k->p.offset) {
 238                         prt_printf(err, "size greater than offset (%u > %llu)",
 239                                k.k->size, k.k->p.offset);
 240                         return -BCH_ERR_invalid_bkey;
 241                 }
 242         } else {
 243                 if (k.k->size) {
 244                         prt_printf(err, "size != 0");
 245                         return -BCH_ERR_invalid_bkey;
 246                 }
 247         }
 248
 249         if (type != BKEY_TYPE_btree &&
 250             !btree_type_has_snapshots(type) &&
 251             k.k->p.snapshot) {
 252                 prt_printf(err, "nonzero snapshot");
 253                 return -BCH_ERR_invalid_bkey;
 254         }
 255
 256         if (type != BKEY_TYPE_btree &&
 257             btree_type_has_snapshots(type) &&
 258             !k.k->p.snapshot) {
 259                 prt_printf(err, "snapshot == 0");
 260                 return -BCH_ERR_invalid_bkey;
 261         }
 262
 263         if (type != BKEY_TYPE_btree &&
 264             bkey_eq(k.k->p, POS_MAX)) {
 265                 prt_printf(err, "key at POS_MAX");
 266                 return -BCH_ERR_invalid_bkey;
 267         }
 268
 269         return 0;
 270 }
 271
 272 int bch2_bkey_invalid(struct bch_fs *c, struct bkey_s_c k,
 273                       enum btree_node_type type,
 274                       unsigned flags, struct printbuf *err)
 275 {
 276         return __bch2_bkey_invalid(c, k, type, flags, err) ?:
 277                 bch2_bkey_val_invalid(c, k, flags, err);
 278 }
 279
 280 int bch2_bkey_in_btree_node(struct btree *b, struct bkey_s_c k,
 281                             struct printbuf *err)
 282 {
 283         if (bpos_lt(k.k->p, b->data->min_key)) {
 284                 prt_printf(err, "key before start of btree node");
 285                 return -BCH_ERR_invalid_bkey;
 286         }
 287
 288         if (bpos_gt(k.k->p, b->data->max_key)) {
 289                 prt_printf(err, "key past end of btree node");
 290                 return -BCH_ERR_invalid_bkey;
 291         }
 292
 293         return 0;
 294 }
 295
 296 void bch2_bpos_to_text(struct printbuf *out, struct bpos pos)
 297 {
 298         if (bpos_eq(pos, POS_MIN))
 299                 prt_printf(out, "POS_MIN");
 300         else if (bpos_eq(pos, POS_MAX))
 301                 prt_printf(out, "POS_MAX");
 302         else if (bpos_eq(pos, SPOS_MAX))
 303                 prt_printf(out, "SPOS_MAX");
 304         else {
 305                 if (pos.inode == U64_MAX)
 306                         prt_printf(out, "U64_MAX");
 307                 else
 308                         prt_printf(out, "%llu", pos.inode);
 309                 prt_printf(out, ":");
 310                 if (pos.offset == U64_MAX)
 311                         prt_printf(out, "U64_MAX");
 312                 else
 313                         prt_printf(out, "%llu", pos.offset);
 314                 prt_printf(out, ":");
 315                 if (pos.snapshot == U32_MAX)
 316                         prt_printf(out, "U32_MAX");
 317                 else
 318                         prt_printf(out, "%u", pos.snapshot);
 319         }
 320 }
 321
 322 void bch2_bkey_to_text(struct printbuf *out, const struct bkey *k)
 323 {
 324         if (k) {
 325                 prt_printf(out, "u64s %u type ", k->u64s);
 326
 327                 if (k->type < KEY_TYPE_MAX)
 328                         prt_printf(out, "%s ", bch2_bkey_types[k->type]);
 329                 else
 330                         prt_printf(out, "%u ", k->type);
 331
 332                 bch2_bpos_to_text(out, k->p);
 333
 334                 prt_printf(out, " len %u ver %llu", k->size, k->version.lo);
 335         } else {
 336                 prt_printf(out, "(null)");
 337         }
 338 }
 339
 340 void bch2_val_to_text(struct printbuf *out, struct bch_fs *c,
 341                       struct bkey_s_c k)
 342 {
 343         if (k.k->type < KEY_TYPE_MAX) {
 344                 const struct bkey_ops *ops = &bch2_bkey_ops[k.k->type];
 345
 346                 if (likely(ops->val_to_text))
 347                         ops->val_to_text(out, c, k);
 348         } else {
 349                 prt_printf(out, "(invalid type %u)", k.k->type);
 350         }
 351 }
 352
 353 void bch2_bkey_val_to_text(struct printbuf *out, struct bch_fs *c,
 354                            struct bkey_s_c k)
 355 {
 356         bch2_bkey_to_text(out, k.k);
 357
 358         if (bkey_val_bytes(k.k)) {
 359                 prt_printf(out, ": ");
 360                 bch2_val_to_text(out, c, k);
 361         }
 362 }
 363
 364 void bch2_bkey_swab_val(struct bkey_s k)
 365 {
 366         const struct bkey_ops *ops = &bch2_bkey_ops[k.k->type];
 367
 368         if (ops->swab)
 369                 ops->swab(k);
 370 }
 371
 372 bool bch2_bkey_normalize(struct bch_fs *c, struct bkey_s k)
 373 {
 374         const struct bkey_ops *ops = &bch2_bkey_ops[k.k->type];
 375
 376         return ops->key_normalize
 377                 ? ops->key_normalize(c, k)
 378                 : false;
 379 }
 380
 381 bool bch2_bkey_merge(struct bch_fs *c, struct bkey_s l, struct bkey_s_c r)
 382 {
 383         const struct bkey_ops *ops = &bch2_bkey_ops[l.k->type];
 384
 385         return bch2_bkey_maybe_mergable(l.k, r.k) &&
 386                 (u64) l.k->size + r.k->size <= KEY_SIZE_MAX &&
 387                 bch2_bkey_ops[l.k->type].key_merge &&
 388                 !bch2_key_merging_disabled &&
 389                 ops->key_merge(c, l, r);
 390 }
 391
 392 static const struct old_bkey_type {
 393         u8              btree_node_type;
 394         u8              old;
 395         u8              new;
 396 } bkey_renumber_table[] = {
 397         {BKEY_TYPE_btree,       128, KEY_TYPE_btree_ptr         },
 398         {BKEY_TYPE_extents,     128, KEY_TYPE_extent            },
 399         {BKEY_TYPE_extents,     129, KEY_TYPE_extent            },
 400         {BKEY_TYPE_extents,     130, KEY_TYPE_reservation       },
 401         {BKEY_TYPE_inodes,      128, KEY_TYPE_inode             },
 402         {BKEY_TYPE_inodes,      130, KEY_TYPE_inode_generation  },
 403         {BKEY_TYPE_dirents,     128, KEY_TYPE_dirent            },
 404         {BKEY_TYPE_dirents,     129, KEY_TYPE_hash_whiteout     },
 405         {BKEY_TYPE_xattrs,      128, KEY_TYPE_xattr             },
 406         {BKEY_TYPE_xattrs,      129, KEY_TYPE_hash_whiteout     },
 407         {BKEY_TYPE_alloc,       128, KEY_TYPE_alloc             },
 408         {BKEY_TYPE_quotas,      128, KEY_TYPE_quota             },
 409 };
 410
 411 void bch2_bkey_renumber(enum btree_node_type btree_node_type,
 412                         struct bkey_packed *k,
 413                         int write)
 414 {
 415         const struct old_bkey_type *i;
 416
 417         for (i = bkey_renumber_table;
 418              i < bkey_renumber_table + ARRAY_SIZE(bkey_renumber_table);
 419              i++)
 420                 if (btree_node_type == i->btree_node_type &&
 421                     k->type == (write ? i->new : i->old)) {
 422                         k->type = write ? i->old : i->new;
 423                         break;
 424                 }
 425 }
 426
 427 void __bch2_bkey_compat(unsigned level, enum btree_id btree_id,
 428                         unsigned version, unsigned big_endian,
 429                         int write,
 430                         struct bkey_format *f,
 431                         struct bkey_packed *k)
 432 {
 433         const struct bkey_ops *ops;
 434         struct bkey uk;
 435         struct bkey_s u;
 436         unsigned nr_compat = 5;
 437         int i;
 438
 439         /*
 440          * Do these operations in reverse order in the write path:
 441          */
 442
 443         for (i = 0; i < nr_compat; i++)
 444         switch (!write ? i : nr_compat - 1 - i) {
 445         case 0:
 446                 if (big_endian != CPU_BIG_ENDIAN)
 447                         bch2_bkey_swab_key(f, k);
 448                 break;
 449         case 1:
 450                 if (version < bcachefs_metadata_version_bkey_renumber)
 451                         bch2_bkey_renumber(__btree_node_type(level, btree_id), k, write);
 452                 break;
 453         case 2:
 454                 if (version < bcachefs_metadata_version_inode_btree_change &&
 455                     btree_id == BTREE_ID_inodes) {
 456                         if (!bkey_packed(k)) {
 457                                 struct bkey_i *u = packed_to_bkey(k);
 458
 459                                 swap(u->k.p.inode, u->k.p.offset);
 460                         } else if (f->bits_per_field[BKEY_FIELD_INODE] &&
 461                                    f->bits_per_field[BKEY_FIELD_OFFSET]) {
 462                                 struct bkey_format tmp = *f, *in = f, *out = &tmp;
 463
 464                                 swap(tmp.bits_per_field[BKEY_FIELD_INODE],
 465                                      tmp.bits_per_field[BKEY_FIELD_OFFSET]);
 466                                 swap(tmp.field_offset[BKEY_FIELD_INODE],
 467                                      tmp.field_offset[BKEY_FIELD_OFFSET]);
 468
 469                                 if (!write)
 470                                         swap(in, out);
 471
 472                                 uk = __bch2_bkey_unpack_key(in, k);
 473                                 swap(uk.p.inode, uk.p.offset);
 474                                 BUG_ON(!bch2_bkey_pack_key(k, &uk, out));
 475                         }
 476                 }
 477                 break;
 478         case 3:
 479                 if (version < bcachefs_metadata_version_snapshot &&
 480                     (level || btree_type_has_snapshots(btree_id))) {
 481                         struct bkey_i *u = packed_to_bkey(k);
 482
 483                         if (u) {
 484                                 u->k.p.snapshot = write
 485                                         ? 0 : U32_MAX;
 486                         } else {
 487                                 u64 min_packed = f->field_offset[BKEY_FIELD_SNAPSHOT];
 488                                 u64 max_packed = min_packed +
 489                                         ~(~0ULL << f->bits_per_field[BKEY_FIELD_SNAPSHOT]);
 490
 491                                 uk = __bch2_bkey_unpack_key(f, k);
 492                                 uk.p.snapshot = write
 493                                         ? min_packed : min_t(u64, U32_MAX, max_packed);
 494
 495                                 BUG_ON(!bch2_bkey_pack_key(k, &uk, f));
 496                         }
 497                 }
 498
 499                 break;
 500         case 4:
 501                 if (!bkey_packed(k)) {
 502                         u = bkey_i_to_s(packed_to_bkey(k));
 503                 } else {
 504                         uk = __bch2_bkey_unpack_key(f, k);
 505                         u.k = &uk;
 506                         u.v = bkeyp_val(f, k);
 507                 }
 508
 509                 if (big_endian != CPU_BIG_ENDIAN)
 510                         bch2_bkey_swab_val(u);
 511
 512                 ops = &bch2_bkey_ops[k->type];
 513
 514                 if (ops->compat)
 515                         ops->compat(btree_id, version, big_endian, write, u);
 516                 break;
 517         default:
 518                 BUG();
 519         }
 520 }