1 // SPDX-License-Identifier: GPL-2.0 2 3 #include "bcachefs.h" 4 #include "backpointers.h" 5 #include "bkey_methods.h" 6 #include "btree_cache.h" 7 #include "btree_types.h" 8 #include "alloc_background.h" 9 #include "dirent.h" 10 #include "disk_accounting.h" 11 #include "ec.h" 12 #include "error.h" 13 #include "extents.h" 14 #include "inode.h" 15 #include "io_misc.h" 16 #include "lru.h" 17 #include "quota.h" 18 #include "reflink.h" 19 #include "snapshot.h" 20 #include "subvolume.h" 21 #include "xattr.h" 22 23 const char * const bch2_bkey_types[] = { 24 #define x(name, nr, ...) #name, 25 BCH_BKEY_TYPES() 26 #undef x 27 NULL 28 }; 29 30 static int deleted_key_validate(struct bch_fs *c, struct bkey_s_c k, 31 struct bkey_validate_context from) 32 { 33 return 0; 34 } 35 36 #define bch2_bkey_ops_deleted ((struct bkey_ops) { \ 37 .key_validate = deleted_key_validate, \ 38 }) 39 40 #define bch2_bkey_ops_whiteout ((struct bkey_ops) { \ 41 .key_validate = deleted_key_validate, \ 42 }) 43 44 static int empty_val_key_validate(struct bch_fs *c, struct bkey_s_c k, 45 struct bkey_validate_context from) 46 { 47 int ret = 0; 48 49 bkey_fsck_err_on(bkey_val_bytes(k.k), 50 c, bkey_val_size_nonzero, 51 "incorrect value size (%zu != 0)", 52 bkey_val_bytes(k.k)); 53 fsck_err: 54 return ret; 55 } 56 57 #define bch2_bkey_ops_error ((struct bkey_ops) { \ 58 .key_validate = empty_val_key_validate, \ 59 }) 60 61 static int key_type_cookie_validate(struct bch_fs *c, struct bkey_s_c k, 62 struct bkey_validate_context from) 63 { 64 return 0; 65 } 66 67 static void key_type_cookie_to_text(struct printbuf *out, struct bch_fs *c, 68 struct bkey_s_c k) 69 { 70 struct bkey_s_c_cookie ck = bkey_s_c_to_cookie(k); 71 72 prt_printf(out, "%llu", le64_to_cpu(ck.v->cookie)); 73 } 74 75 #define bch2_bkey_ops_cookie ((struct bkey_ops) { \ 76 .key_validate = key_type_cookie_validate, \ 77 .val_to_text = key_type_cookie_to_text, \ 78 .min_val_size = 8, \ 79 }) 80 81 #define bch2_bkey_ops_hash_whiteout ((struct bkey_ops) {\ 82 .key_validate = empty_val_key_validate, \ 83 }) 84 85 static int key_type_inline_data_validate(struct bch_fs *c, struct bkey_s_c k, 86 struct bkey_validate_context from) 87 { 88 return 0; 89 } 90 91 static void key_type_inline_data_to_text(struct printbuf *out, struct bch_fs *c, 92 struct bkey_s_c k) 93 { 94 struct bkey_s_c_inline_data d = bkey_s_c_to_inline_data(k); 95 unsigned datalen = bkey_inline_data_bytes(k.k); 96 97 prt_printf(out, "datalen %u: %*phN", 98 datalen, min(datalen, 32U), d.v->data); 99 } 100 101 #define bch2_bkey_ops_inline_data ((struct bkey_ops) { \ 102 .key_validate = key_type_inline_data_validate, \ 103 .val_to_text = key_type_inline_data_to_text, \ 104 }) 105 106 static bool key_type_set_merge(struct bch_fs *c, struct bkey_s l, struct bkey_s_c r) 107 { 108 bch2_key_resize(l.k, l.k->size + r.k->size); 109 return true; 110 } 111 112 #define bch2_bkey_ops_set ((struct bkey_ops) { \ 113 .key_validate = empty_val_key_validate, \ 114 .key_merge = key_type_set_merge, \ 115 }) 116 117 const struct bkey_ops bch2_bkey_ops[] = { 118 #define x(name, nr, ...) [KEY_TYPE_##name] = bch2_bkey_ops_##name, 119 BCH_BKEY_TYPES() 120 #undef x 121 }; 122 123 const struct bkey_ops bch2_bkey_null_ops = { 124 }; 125 126 int bch2_bkey_val_validate(struct bch_fs *c, struct bkey_s_c k, 127 struct bkey_validate_context from) 128 { 129 if (test_bit(BCH_FS_no_invalid_checks, &c->flags)) 130 return 0; 131 132 const struct bkey_ops *ops = bch2_bkey_type_ops(k.k->type); 133 int ret = 0; 134 135 bkey_fsck_err_on(bkey_val_bytes(k.k) < ops->min_val_size, 136 c, bkey_val_size_too_small, 137 "bad val size (%zu < %u)", 138 bkey_val_bytes(k.k), ops->min_val_size); 139 140 if (!ops->key_validate) 141 return 0; 142 143 ret = ops->key_validate(c, k, from); 144 fsck_err: 145 return ret; 146 } 147 148 static u64 bch2_key_types_allowed[] = { 149 [BKEY_TYPE_btree] = 150 BIT_ULL(KEY_TYPE_deleted)| 151 BIT_ULL(KEY_TYPE_btree_ptr)| 152 BIT_ULL(KEY_TYPE_btree_ptr_v2), 153 #define x(name, nr, flags, keys) [BKEY_TYPE_##name] = BIT_ULL(KEY_TYPE_deleted)|keys, 154 BCH_BTREE_IDS() 155 #undef x 156 }; 157 158 static const enum bch_bkey_type_flags bch2_bkey_type_flags[] = { 159 #define x(name, nr, flags) [KEY_TYPE_##name] = flags, 160 BCH_BKEY_TYPES() 161 #undef x 162 }; 163 164 const char *bch2_btree_node_type_str(enum btree_node_type type) 165 { 166 return type == BKEY_TYPE_btree ? "internal btree node" : bch2_btree_id_str(type - 1); 167 } 168 169 int __bch2_bkey_validate(struct bch_fs *c, struct bkey_s_c k, 170 struct bkey_validate_context from) 171 { 172 enum btree_node_type type = __btree_node_type(from.level, from.btree); 173 174 if (test_bit(BCH_FS_no_invalid_checks, &c->flags)) 175 return 0; 176 177 int ret = 0; 178 179 bkey_fsck_err_on(k.k->u64s < BKEY_U64s, 180 c, bkey_u64s_too_small, 181 "u64s too small (%u < %zu)", k.k->u64s, BKEY_U64s); 182 183 if (type >= BKEY_TYPE_NR) 184 return 0; 185 186 enum bch_bkey_type_flags bkey_flags = k.k->type < KEY_TYPE_MAX 187 ? bch2_bkey_type_flags[k.k->type] 188 : 0; 189 190 bool strict_key_type_allowed = 191 (from.flags & BCH_VALIDATE_commit) || 192 type == BKEY_TYPE_btree || 193 (from.btree < BTREE_ID_NR && 194 (bkey_flags & BKEY_TYPE_strict_btree_checks)); 195 196 bkey_fsck_err_on(strict_key_type_allowed && 197 k.k->type < KEY_TYPE_MAX && 198 !(bch2_key_types_allowed[type] & BIT_ULL(k.k->type)), 199 c, bkey_invalid_type_for_btree, 200 "invalid key type for btree %s (%s)", 201 bch2_btree_node_type_str(type), 202 k.k->type < KEY_TYPE_MAX 203 ? bch2_bkey_types[k.k->type] 204 : "(unknown)"); 205 206 if (btree_node_type_is_extents(type) && !bkey_whiteout(k.k)) { 207 bkey_fsck_err_on(k.k->size == 0, 208 c, bkey_extent_size_zero, 209 "size == 0"); 210 211 bkey_fsck_err_on(k.k->size > k.k->p.offset, 212 c, bkey_extent_size_greater_than_offset, 213 "size greater than offset (%u > %llu)", 214 k.k->size, k.k->p.offset); 215 } else { 216 bkey_fsck_err_on(k.k->size, 217 c, bkey_size_nonzero, 218 "size != 0"); 219 } 220 221 if (type != BKEY_TYPE_btree) { 222 enum btree_id btree = type - 1; 223 224 if (btree_type_has_snapshots(btree)) { 225 bkey_fsck_err_on(!k.k->p.snapshot, 226 c, bkey_snapshot_zero, 227 "snapshot == 0"); 228 } else if (!btree_type_has_snapshot_field(btree)) { 229 bkey_fsck_err_on(k.k->p.snapshot, 230 c, bkey_snapshot_nonzero, 231 "nonzero snapshot"); 232 } else { 233 /* 234 * btree uses snapshot field but it's not required to be 235 * nonzero 236 */ 237 } 238 239 bkey_fsck_err_on(bkey_eq(k.k->p, POS_MAX), 240 c, bkey_at_pos_max, 241 "key at POS_MAX"); 242 } 243 fsck_err: 244 return ret; 245 } 246 247 int bch2_bkey_validate(struct bch_fs *c, struct bkey_s_c k, 248 struct bkey_validate_context from) 249 { 250 return __bch2_bkey_validate(c, k, from) ?: 251 bch2_bkey_val_validate(c, k, from); 252 } 253 254 int bch2_bkey_in_btree_node(struct bch_fs *c, struct btree *b, 255 struct bkey_s_c k, 256 struct bkey_validate_context from) 257 { 258 int ret = 0; 259 260 bkey_fsck_err_on(bpos_lt(k.k->p, b->data->min_key), 261 c, bkey_before_start_of_btree_node, 262 "key before start of btree node"); 263 264 bkey_fsck_err_on(bpos_gt(k.k->p, b->data->max_key), 265 c, bkey_after_end_of_btree_node, 266 "key past end of btree node"); 267 fsck_err: 268 return ret; 269 } 270 271 void bch2_bpos_to_text(struct printbuf *out, struct bpos pos) 272 { 273 if (bpos_eq(pos, POS_MIN)) 274 prt_printf(out, "POS_MIN"); 275 else if (bpos_eq(pos, POS_MAX)) 276 prt_printf(out, "POS_MAX"); 277 else if (bpos_eq(pos, SPOS_MAX)) 278 prt_printf(out, "SPOS_MAX"); 279 else { 280 if (pos.inode == U64_MAX) 281 prt_printf(out, "U64_MAX"); 282 else 283 prt_printf(out, "%llu", pos.inode); 284 prt_printf(out, ":"); 285 if (pos.offset == U64_MAX) 286 prt_printf(out, "U64_MAX"); 287 else 288 prt_printf(out, "%llu", pos.offset); 289 prt_printf(out, ":"); 290 if (pos.snapshot == U32_MAX) 291 prt_printf(out, "U32_MAX"); 292 else 293 prt_printf(out, "%u", pos.snapshot); 294 } 295 } 296 297 void bch2_bkey_to_text(struct printbuf *out, const struct bkey *k) 298 { 299 if (k) { 300 prt_printf(out, "u64s %u type ", k->u64s); 301 302 if (k->type < KEY_TYPE_MAX) 303 prt_printf(out, "%s ", bch2_bkey_types[k->type]); 304 else 305 prt_printf(out, "%u ", k->type); 306 307 bch2_bpos_to_text(out, k->p); 308 309 prt_printf(out, " len %u ver %llu", k->size, k->bversion.lo); 310 } else { 311 prt_printf(out, "(null)"); 312 } 313 } 314 315 void bch2_val_to_text(struct printbuf *out, struct bch_fs *c, 316 struct bkey_s_c k) 317 { 318 const struct bkey_ops *ops = bch2_bkey_type_ops(k.k->type); 319 320 if (likely(ops->val_to_text)) 321 ops->val_to_text(out, c, k); 322 } 323 324 void bch2_bkey_val_to_text(struct printbuf *out, struct bch_fs *c, 325 struct bkey_s_c k) 326 { 327 bch2_bkey_to_text(out, k.k); 328 329 if (bkey_val_bytes(k.k)) { 330 prt_printf(out, ": "); 331 bch2_val_to_text(out, c, k); 332 } 333 } 334 335 void bch2_bkey_swab_val(struct bkey_s k) 336 { 337 const struct bkey_ops *ops = bch2_bkey_type_ops(k.k->type); 338 339 if (ops->swab) 340 ops->swab(k); 341 } 342 343 bool bch2_bkey_normalize(struct bch_fs *c, struct bkey_s k) 344 { 345 const struct bkey_ops *ops = bch2_bkey_type_ops(k.k->type); 346 347 return ops->key_normalize 348 ? ops->key_normalize(c, k) 349 : false; 350 } 351 352 bool bch2_bkey_merge(struct bch_fs *c, struct bkey_s l, struct bkey_s_c r) 353 { 354 const struct bkey_ops *ops = bch2_bkey_type_ops(l.k->type); 355 356 return ops->key_merge && 357 bch2_bkey_maybe_mergable(l.k, r.k) && 358 (u64) l.k->size + r.k->size <= KEY_SIZE_MAX && 359 !bch2_key_merging_disabled && 360 ops->key_merge(c, l, r); 361 } 362 363 static const struct old_bkey_type { 364 u8 btree_node_type; 365 u8 old; 366 u8 new; 367 } bkey_renumber_table[] = { 368 {BKEY_TYPE_btree, 128, KEY_TYPE_btree_ptr }, 369 {BKEY_TYPE_extents, 128, KEY_TYPE_extent }, 370 {BKEY_TYPE_extents, 129, KEY_TYPE_extent }, 371 {BKEY_TYPE_extents, 130, KEY_TYPE_reservation }, 372 {BKEY_TYPE_inodes, 128, KEY_TYPE_inode }, 373 {BKEY_TYPE_inodes, 130, KEY_TYPE_inode_generation }, 374 {BKEY_TYPE_dirents, 128, KEY_TYPE_dirent }, 375 {BKEY_TYPE_dirents, 129, KEY_TYPE_hash_whiteout }, 376 {BKEY_TYPE_xattrs, 128, KEY_TYPE_xattr }, 377 {BKEY_TYPE_xattrs, 129, KEY_TYPE_hash_whiteout }, 378 {BKEY_TYPE_alloc, 128, KEY_TYPE_alloc }, 379 {BKEY_TYPE_quotas, 128, KEY_TYPE_quota }, 380 }; 381 382 void bch2_bkey_renumber(enum btree_node_type btree_node_type, 383 struct bkey_packed *k, 384 int write) 385 { 386 const struct old_bkey_type *i; 387 388 for (i = bkey_renumber_table; 389 i < bkey_renumber_table + ARRAY_SIZE(bkey_renumber_table); 390 i++) 391 if (btree_node_type == i->btree_node_type && 392 k->type == (write ? i->new : i->old)) { 393 k->type = write ? i->old : i->new; 394 break; 395 } 396 } 397 398 void __bch2_bkey_compat(unsigned level, enum btree_id btree_id, 399 unsigned version, unsigned big_endian, 400 int write, 401 struct bkey_format *f, 402 struct bkey_packed *k) 403 { 404 const struct bkey_ops *ops; 405 struct bkey uk; 406 unsigned nr_compat = 5; 407 int i; 408 409 /* 410 * Do these operations in reverse order in the write path: 411 */ 412 413 for (i = 0; i < nr_compat; i++) 414 switch (!write ? i : nr_compat - 1 - i) { 415 case 0: 416 if (big_endian != CPU_BIG_ENDIAN) { 417 bch2_bkey_swab_key(f, k); 418 } else if (IS_ENABLED(CONFIG_BCACHEFS_DEBUG)) { 419 bch2_bkey_swab_key(f, k); 420 bch2_bkey_swab_key(f, k); 421 } 422 break; 423 case 1: 424 if (version < bcachefs_metadata_version_bkey_renumber) 425 bch2_bkey_renumber(__btree_node_type(level, btree_id), k, write); 426 break; 427 case 2: 428 if (version < bcachefs_metadata_version_inode_btree_change && 429 btree_id == BTREE_ID_inodes) { 430 if (!bkey_packed(k)) { 431 struct bkey_i *u = packed_to_bkey(k); 432 433 swap(u->k.p.inode, u->k.p.offset); 434 } else if (f->bits_per_field[BKEY_FIELD_INODE] && 435 f->bits_per_field[BKEY_FIELD_OFFSET]) { 436 struct bkey_format tmp = *f, *in = f, *out = &tmp; 437 438 swap(tmp.bits_per_field[BKEY_FIELD_INODE], 439 tmp.bits_per_field[BKEY_FIELD_OFFSET]); 440 swap(tmp.field_offset[BKEY_FIELD_INODE], 441 tmp.field_offset[BKEY_FIELD_OFFSET]); 442 443 if (!write) 444 swap(in, out); 445 446 uk = __bch2_bkey_unpack_key(in, k); 447 swap(uk.p.inode, uk.p.offset); 448 BUG_ON(!bch2_bkey_pack_key(k, &uk, out)); 449 } 450 } 451 break; 452 case 3: 453 if (version < bcachefs_metadata_version_snapshot && 454 (level || btree_type_has_snapshots(btree_id))) { 455 struct bkey_i *u = packed_to_bkey(k); 456 457 if (u) { 458 u->k.p.snapshot = write 459 ? 0 : U32_MAX; 460 } else { 461 u64 min_packed = le64_to_cpu(f->field_offset[BKEY_FIELD_SNAPSHOT]); 462 u64 max_packed = min_packed + 463 ~(~0ULL << f->bits_per_field[BKEY_FIELD_SNAPSHOT]); 464 465 uk = __bch2_bkey_unpack_key(f, k); 466 uk.p.snapshot = write 467 ? min_packed : min_t(u64, U32_MAX, max_packed); 468 469 BUG_ON(!bch2_bkey_pack_key(k, &uk, f)); 470 } 471 } 472 473 break; 474 case 4: { 475 struct bkey_s u; 476 477 if (!bkey_packed(k)) { 478 u = bkey_i_to_s(packed_to_bkey(k)); 479 } else { 480 uk = __bch2_bkey_unpack_key(f, k); 481 u.k = &uk; 482 u.v = bkeyp_val(f, k); 483 } 484 485 if (big_endian != CPU_BIG_ENDIAN) 486 bch2_bkey_swab_val(u); 487 488 ops = bch2_bkey_type_ops(k->type); 489 490 if (ops->compat) 491 ops->compat(btree_id, version, big_endian, write, u); 492 break; 493 } 494 default: 495 BUG(); 496 } 497 } 498