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 const struct bkey_ops bch2_bkey_null_ops = { 122 .min_val_size = U8_MAX, 123 }; 124 125 int bch2_bkey_val_invalid(struct bch_fs *c, struct bkey_s_c k, 126 enum bkey_invalid_flags flags, 127 struct printbuf *err) 128 { 129 const struct bkey_ops *ops = bch2_bkey_type_ops(k.k->type); 130 131 if (bkey_val_bytes(k.k) < ops->min_val_size) { 132 prt_printf(err, "bad val size (%zu < %u)", 133 bkey_val_bytes(k.k), ops->min_val_size); 134 return -BCH_ERR_invalid_bkey; 135 } 136 137 if (!ops->key_invalid) 138 return 0; 139 140 return ops->key_invalid(c, k, flags, err); 141 } 142 143 static unsigned bch2_key_types_allowed[] = { 144 [BKEY_TYPE_extents] = 145 (1U << KEY_TYPE_deleted)| 146 (1U << KEY_TYPE_whiteout)| 147 (1U << KEY_TYPE_error)| 148 (1U << KEY_TYPE_cookie)| 149 (1U << KEY_TYPE_extent)| 150 (1U << KEY_TYPE_reservation)| 151 (1U << KEY_TYPE_reflink_p)| 152 (1U << KEY_TYPE_inline_data), 153 [BKEY_TYPE_inodes] = 154 (1U << KEY_TYPE_deleted)| 155 (1U << KEY_TYPE_whiteout)| 156 (1U << KEY_TYPE_inode)| 157 (1U << KEY_TYPE_inode_v2)| 158 (1U << KEY_TYPE_inode_v3)| 159 (1U << KEY_TYPE_inode_generation), 160 [BKEY_TYPE_dirents] = 161 (1U << KEY_TYPE_deleted)| 162 (1U << KEY_TYPE_whiteout)| 163 (1U << KEY_TYPE_hash_whiteout)| 164 (1U << KEY_TYPE_dirent), 165 [BKEY_TYPE_xattrs] = 166 (1U << KEY_TYPE_deleted)| 167 (1U << KEY_TYPE_whiteout)| 168 (1U << KEY_TYPE_cookie)| 169 (1U << KEY_TYPE_hash_whiteout)| 170 (1U << KEY_TYPE_xattr), 171 [BKEY_TYPE_alloc] = 172 (1U << KEY_TYPE_deleted)| 173 (1U << KEY_TYPE_alloc)| 174 (1U << KEY_TYPE_alloc_v2)| 175 (1U << KEY_TYPE_alloc_v3)| 176 (1U << KEY_TYPE_alloc_v4), 177 [BKEY_TYPE_quotas] = 178 (1U << KEY_TYPE_deleted)| 179 (1U << KEY_TYPE_quota), 180 [BKEY_TYPE_stripes] = 181 (1U << KEY_TYPE_deleted)| 182 (1U << KEY_TYPE_stripe), 183 [BKEY_TYPE_reflink] = 184 (1U << KEY_TYPE_deleted)| 185 (1U << KEY_TYPE_reflink_v)| 186 (1U << KEY_TYPE_indirect_inline_data), 187 [BKEY_TYPE_subvolumes] = 188 (1U << KEY_TYPE_deleted)| 189 (1U << KEY_TYPE_subvolume), 190 [BKEY_TYPE_snapshots] = 191 (1U << KEY_TYPE_deleted)| 192 (1U << KEY_TYPE_snapshot), 193 [BKEY_TYPE_lru] = 194 (1U << KEY_TYPE_deleted)| 195 (1U << KEY_TYPE_set), 196 [BKEY_TYPE_freespace] = 197 (1U << KEY_TYPE_deleted)| 198 (1U << KEY_TYPE_set), 199 [BKEY_TYPE_need_discard] = 200 (1U << KEY_TYPE_deleted)| 201 (1U << KEY_TYPE_set), 202 [BKEY_TYPE_backpointers] = 203 (1U << KEY_TYPE_deleted)| 204 (1U << KEY_TYPE_backpointer), 205 [BKEY_TYPE_bucket_gens] = 206 (1U << KEY_TYPE_deleted)| 207 (1U << KEY_TYPE_bucket_gens), 208 [BKEY_TYPE_snapshot_trees] = 209 (1U << KEY_TYPE_deleted)| 210 (1U << KEY_TYPE_snapshot_tree), 211 [BKEY_TYPE_btree] = 212 (1U << KEY_TYPE_deleted)| 213 (1U << KEY_TYPE_btree_ptr)| 214 (1U << KEY_TYPE_btree_ptr_v2), 215 }; 216 217 int __bch2_bkey_invalid(struct bch_fs *c, struct bkey_s_c k, 218 enum btree_node_type type, 219 enum bkey_invalid_flags flags, 220 struct printbuf *err) 221 { 222 if (k.k->u64s < BKEY_U64s) { 223 prt_printf(err, "u64s too small (%u < %zu)", k.k->u64s, BKEY_U64s); 224 return -BCH_ERR_invalid_bkey; 225 } 226 227 if (flags & BKEY_INVALID_COMMIT && 228 !(bch2_key_types_allowed[type] & (1U << k.k->type))) { 229 prt_printf(err, "invalid key type for btree %s (%s)", 230 bch2_btree_ids[type], bch2_bkey_types[k.k->type]); 231 return -BCH_ERR_invalid_bkey; 232 } 233 234 if (btree_node_type_is_extents(type) && !bkey_whiteout(k.k)) { 235 if (k.k->size == 0) { 236 prt_printf(err, "size == 0"); 237 return -BCH_ERR_invalid_bkey; 238 } 239 240 if (k.k->size > k.k->p.offset) { 241 prt_printf(err, "size greater than offset (%u > %llu)", 242 k.k->size, k.k->p.offset); 243 return -BCH_ERR_invalid_bkey; 244 } 245 } else { 246 if (k.k->size) { 247 prt_printf(err, "size != 0"); 248 return -BCH_ERR_invalid_bkey; 249 } 250 } 251 252 if (type != BKEY_TYPE_btree) { 253 if (!btree_type_has_snapshots((enum btree_id) type) && 254 k.k->p.snapshot) { 255 prt_printf(err, "nonzero snapshot"); 256 return -BCH_ERR_invalid_bkey; 257 } 258 259 if (btree_type_has_snapshots((enum btree_id) type) && 260 !k.k->p.snapshot) { 261 prt_printf(err, "snapshot == 0"); 262 return -BCH_ERR_invalid_bkey; 263 } 264 265 if (bkey_eq(k.k->p, POS_MAX)) { 266 prt_printf(err, "key at POS_MAX"); 267 return -BCH_ERR_invalid_bkey; 268 } 269 } 270 271 return 0; 272 } 273 274 int bch2_bkey_invalid(struct bch_fs *c, struct bkey_s_c k, 275 enum btree_node_type type, 276 enum bkey_invalid_flags flags, 277 struct printbuf *err) 278 { 279 return __bch2_bkey_invalid(c, k, type, flags, err) ?: 280 bch2_bkey_val_invalid(c, k, flags, err); 281 } 282 283 int bch2_bkey_in_btree_node(struct btree *b, struct bkey_s_c k, 284 struct printbuf *err) 285 { 286 if (bpos_lt(k.k->p, b->data->min_key)) { 287 prt_printf(err, "key before start of btree node"); 288 return -BCH_ERR_invalid_bkey; 289 } 290 291 if (bpos_gt(k.k->p, b->data->max_key)) { 292 prt_printf(err, "key past end of btree node"); 293 return -BCH_ERR_invalid_bkey; 294 } 295 296 return 0; 297 } 298 299 void bch2_bpos_to_text(struct printbuf *out, struct bpos pos) 300 { 301 if (bpos_eq(pos, POS_MIN)) 302 prt_printf(out, "POS_MIN"); 303 else if (bpos_eq(pos, POS_MAX)) 304 prt_printf(out, "POS_MAX"); 305 else if (bpos_eq(pos, SPOS_MAX)) 306 prt_printf(out, "SPOS_MAX"); 307 else { 308 if (pos.inode == U64_MAX) 309 prt_printf(out, "U64_MAX"); 310 else 311 prt_printf(out, "%llu", pos.inode); 312 prt_printf(out, ":"); 313 if (pos.offset == U64_MAX) 314 prt_printf(out, "U64_MAX"); 315 else 316 prt_printf(out, "%llu", pos.offset); 317 prt_printf(out, ":"); 318 if (pos.snapshot == U32_MAX) 319 prt_printf(out, "U32_MAX"); 320 else 321 prt_printf(out, "%u", pos.snapshot); 322 } 323 } 324 325 void bch2_bkey_to_text(struct printbuf *out, const struct bkey *k) 326 { 327 if (k) { 328 prt_printf(out, "u64s %u type ", k->u64s); 329 330 if (k->type < KEY_TYPE_MAX) 331 prt_printf(out, "%s ", bch2_bkey_types[k->type]); 332 else 333 prt_printf(out, "%u ", k->type); 334 335 bch2_bpos_to_text(out, k->p); 336 337 prt_printf(out, " len %u ver %llu", k->size, k->version.lo); 338 } else { 339 prt_printf(out, "(null)"); 340 } 341 } 342 343 void bch2_val_to_text(struct printbuf *out, struct bch_fs *c, 344 struct bkey_s_c k) 345 { 346 const struct bkey_ops *ops = bch2_bkey_type_ops(k.k->type); 347 348 if (likely(ops->val_to_text)) 349 ops->val_to_text(out, c, k); 350 } 351 352 void bch2_bkey_val_to_text(struct printbuf *out, struct bch_fs *c, 353 struct bkey_s_c k) 354 { 355 bch2_bkey_to_text(out, k.k); 356 357 if (bkey_val_bytes(k.k)) { 358 prt_printf(out, ": "); 359 bch2_val_to_text(out, c, k); 360 } 361 } 362 363 void bch2_bkey_swab_val(struct bkey_s k) 364 { 365 const struct bkey_ops *ops = bch2_bkey_type_ops(k.k->type); 366 367 if (ops->swab) 368 ops->swab(k); 369 } 370 371 bool bch2_bkey_normalize(struct bch_fs *c, struct bkey_s k) 372 { 373 const struct bkey_ops *ops = bch2_bkey_type_ops(k.k->type); 374 375 return ops->key_normalize 376 ? ops->key_normalize(c, k) 377 : false; 378 } 379 380 bool bch2_bkey_merge(struct bch_fs *c, struct bkey_s l, struct bkey_s_c r) 381 { 382 const struct bkey_ops *ops = bch2_bkey_type_ops(l.k->type); 383 384 return ops->key_merge && 385 bch2_bkey_maybe_mergable(l.k, r.k) && 386 (u64) l.k->size + r.k->size <= KEY_SIZE_MAX && 387 !bch2_key_merging_disabled && 388 ops->key_merge(c, l, r); 389 } 390 391 static const struct old_bkey_type { 392 u8 btree_node_type; 393 u8 old; 394 u8 new; 395 } bkey_renumber_table[] = { 396 {BKEY_TYPE_btree, 128, KEY_TYPE_btree_ptr }, 397 {BKEY_TYPE_extents, 128, KEY_TYPE_extent }, 398 {BKEY_TYPE_extents, 129, KEY_TYPE_extent }, 399 {BKEY_TYPE_extents, 130, KEY_TYPE_reservation }, 400 {BKEY_TYPE_inodes, 128, KEY_TYPE_inode }, 401 {BKEY_TYPE_inodes, 130, KEY_TYPE_inode_generation }, 402 {BKEY_TYPE_dirents, 128, KEY_TYPE_dirent }, 403 {BKEY_TYPE_dirents, 129, KEY_TYPE_hash_whiteout }, 404 {BKEY_TYPE_xattrs, 128, KEY_TYPE_xattr }, 405 {BKEY_TYPE_xattrs, 129, KEY_TYPE_hash_whiteout }, 406 {BKEY_TYPE_alloc, 128, KEY_TYPE_alloc }, 407 {BKEY_TYPE_quotas, 128, KEY_TYPE_quota }, 408 }; 409 410 void bch2_bkey_renumber(enum btree_node_type btree_node_type, 411 struct bkey_packed *k, 412 int write) 413 { 414 const struct old_bkey_type *i; 415 416 for (i = bkey_renumber_table; 417 i < bkey_renumber_table + ARRAY_SIZE(bkey_renumber_table); 418 i++) 419 if (btree_node_type == i->btree_node_type && 420 k->type == (write ? i->new : i->old)) { 421 k->type = write ? i->old : i->new; 422 break; 423 } 424 } 425 426 void __bch2_bkey_compat(unsigned level, enum btree_id btree_id, 427 unsigned version, unsigned big_endian, 428 int write, 429 struct bkey_format *f, 430 struct bkey_packed *k) 431 { 432 const struct bkey_ops *ops; 433 struct bkey uk; 434 struct bkey_s u; 435 unsigned nr_compat = 5; 436 int i; 437 438 /* 439 * Do these operations in reverse order in the write path: 440 */ 441 442 for (i = 0; i < nr_compat; i++) 443 switch (!write ? i : nr_compat - 1 - i) { 444 case 0: 445 if (big_endian != CPU_BIG_ENDIAN) 446 bch2_bkey_swab_key(f, k); 447 break; 448 case 1: 449 if (version < bcachefs_metadata_version_bkey_renumber) 450 bch2_bkey_renumber(__btree_node_type(level, btree_id), k, write); 451 break; 452 case 2: 453 if (version < bcachefs_metadata_version_inode_btree_change && 454 btree_id == BTREE_ID_inodes) { 455 if (!bkey_packed(k)) { 456 struct bkey_i *u = packed_to_bkey(k); 457 458 swap(u->k.p.inode, u->k.p.offset); 459 } else if (f->bits_per_field[BKEY_FIELD_INODE] && 460 f->bits_per_field[BKEY_FIELD_OFFSET]) { 461 struct bkey_format tmp = *f, *in = f, *out = &tmp; 462 463 swap(tmp.bits_per_field[BKEY_FIELD_INODE], 464 tmp.bits_per_field[BKEY_FIELD_OFFSET]); 465 swap(tmp.field_offset[BKEY_FIELD_INODE], 466 tmp.field_offset[BKEY_FIELD_OFFSET]); 467 468 if (!write) 469 swap(in, out); 470 471 uk = __bch2_bkey_unpack_key(in, k); 472 swap(uk.p.inode, uk.p.offset); 473 BUG_ON(!bch2_bkey_pack_key(k, &uk, out)); 474 } 475 } 476 break; 477 case 3: 478 if (version < bcachefs_metadata_version_snapshot && 479 (level || btree_type_has_snapshots(btree_id))) { 480 struct bkey_i *u = packed_to_bkey(k); 481 482 if (u) { 483 u->k.p.snapshot = write 484 ? 0 : U32_MAX; 485 } else { 486 u64 min_packed = le64_to_cpu(f->field_offset[BKEY_FIELD_SNAPSHOT]); 487 u64 max_packed = min_packed + 488 ~(~0ULL << f->bits_per_field[BKEY_FIELD_SNAPSHOT]); 489 490 uk = __bch2_bkey_unpack_key(f, k); 491 uk.p.snapshot = write 492 ? min_packed : min_t(u64, U32_MAX, max_packed); 493 494 BUG_ON(!bch2_bkey_pack_key(k, &uk, f)); 495 } 496 } 497 498 break; 499 case 4: 500 if (!bkey_packed(k)) { 501 u = bkey_i_to_s(packed_to_bkey(k)); 502 } else { 503 uk = __bch2_bkey_unpack_key(f, k); 504 u.k = &uk; 505 u.v = bkeyp_val(f, k); 506 } 507 508 if (big_endian != CPU_BIG_ENDIAN) 509 bch2_bkey_swab_val(u); 510 511 ops = bch2_bkey_type_ops(k->type); 512 513 if (ops->compat) 514 ops->compat(btree_id, version, big_endian, write, u); 515 break; 516 default: 517 BUG(); 518 } 519 } 520