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 "ec.h" 11 #include "error.h" 12 #include "extents.h" 13 #include "inode.h" 14 #include "io_misc.h" 15 #include "lru.h" 16 #include "quota.h" 17 #include "reflink.h" 18 #include "snapshot.h" 19 #include "subvolume.h" 20 #include "xattr.h" 21 22 const char * const bch2_bkey_types[] = { 23 #define x(name, nr) #name, 24 BCH_BKEY_TYPES() 25 #undef x 26 NULL 27 }; 28 29 static int deleted_key_invalid(struct bch_fs *c, struct bkey_s_c k, 30 enum bch_validate_flags flags, struct printbuf *err) 31 { 32 return 0; 33 } 34 35 #define bch2_bkey_ops_deleted ((struct bkey_ops) { \ 36 .key_invalid = deleted_key_invalid, \ 37 }) 38 39 #define bch2_bkey_ops_whiteout ((struct bkey_ops) { \ 40 .key_invalid = deleted_key_invalid, \ 41 }) 42 43 static int empty_val_key_invalid(struct bch_fs *c, struct bkey_s_c k, 44 enum bch_validate_flags flags, struct printbuf *err) 45 { 46 int ret = 0; 47 48 bkey_fsck_err_on(bkey_val_bytes(k.k), c, err, 49 bkey_val_size_nonzero, 50 "incorrect value size (%zu != 0)", 51 bkey_val_bytes(k.k)); 52 fsck_err: 53 return ret; 54 } 55 56 #define bch2_bkey_ops_error ((struct bkey_ops) { \ 57 .key_invalid = empty_val_key_invalid, \ 58 }) 59 60 static int key_type_cookie_invalid(struct bch_fs *c, struct bkey_s_c k, 61 enum bch_validate_flags flags, struct printbuf *err) 62 { 63 return 0; 64 } 65 66 static void key_type_cookie_to_text(struct printbuf *out, struct bch_fs *c, 67 struct bkey_s_c k) 68 { 69 struct bkey_s_c_cookie ck = bkey_s_c_to_cookie(k); 70 71 prt_printf(out, "%llu", le64_to_cpu(ck.v->cookie)); 72 } 73 74 #define bch2_bkey_ops_cookie ((struct bkey_ops) { \ 75 .key_invalid = key_type_cookie_invalid, \ 76 .val_to_text = key_type_cookie_to_text, \ 77 .min_val_size = 8, \ 78 }) 79 80 #define bch2_bkey_ops_hash_whiteout ((struct bkey_ops) {\ 81 .key_invalid = empty_val_key_invalid, \ 82 }) 83 84 static int key_type_inline_data_invalid(struct bch_fs *c, struct bkey_s_c k, 85 enum bch_validate_flags flags, struct printbuf *err) 86 { 87 return 0; 88 } 89 90 static void key_type_inline_data_to_text(struct printbuf *out, struct bch_fs *c, 91 struct bkey_s_c k) 92 { 93 struct bkey_s_c_inline_data d = bkey_s_c_to_inline_data(k); 94 unsigned datalen = bkey_inline_data_bytes(k.k); 95 96 prt_printf(out, "datalen %u: %*phN", 97 datalen, min(datalen, 32U), d.v->data); 98 } 99 100 #define bch2_bkey_ops_inline_data ((struct bkey_ops) { \ 101 .key_invalid = key_type_inline_data_invalid, \ 102 .val_to_text = key_type_inline_data_to_text, \ 103 }) 104 105 static bool key_type_set_merge(struct bch_fs *c, struct bkey_s l, struct bkey_s_c r) 106 { 107 bch2_key_resize(l.k, l.k->size + r.k->size); 108 return true; 109 } 110 111 #define bch2_bkey_ops_set ((struct bkey_ops) { \ 112 .key_invalid = empty_val_key_invalid, \ 113 .key_merge = key_type_set_merge, \ 114 }) 115 116 const struct bkey_ops bch2_bkey_ops[] = { 117 #define x(name, nr) [KEY_TYPE_##name] = bch2_bkey_ops_##name, 118 BCH_BKEY_TYPES() 119 #undef x 120 }; 121 122 const struct bkey_ops bch2_bkey_null_ops = { 123 }; 124 125 int bch2_bkey_val_invalid(struct bch_fs *c, struct bkey_s_c k, 126 enum bch_validate_flags flags, 127 struct printbuf *err) 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, c, err, 136 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_invalid) 141 return 0; 142 143 ret = ops->key_invalid(c, k, flags, err); 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 const char *bch2_btree_node_type_str(enum btree_node_type type) 159 { 160 return type == BKEY_TYPE_btree ? "internal btree node" : bch2_btree_id_str(type - 1); 161 } 162 163 int __bch2_bkey_invalid(struct bch_fs *c, struct bkey_s_c k, 164 enum btree_node_type type, 165 enum bch_validate_flags flags, 166 struct printbuf *err) 167 { 168 if (test_bit(BCH_FS_no_invalid_checks, &c->flags)) 169 return 0; 170 171 int ret = 0; 172 173 bkey_fsck_err_on(k.k->u64s < BKEY_U64s, c, err, 174 bkey_u64s_too_small, 175 "u64s too small (%u < %zu)", k.k->u64s, BKEY_U64s); 176 177 if (type >= BKEY_TYPE_NR) 178 return 0; 179 180 bkey_fsck_err_on(k.k->type < KEY_TYPE_MAX && 181 (type == BKEY_TYPE_btree || (flags & BCH_VALIDATE_commit)) && 182 !(bch2_key_types_allowed[type] & BIT_ULL(k.k->type)), c, err, 183 bkey_invalid_type_for_btree, 184 "invalid key type for btree %s (%s)", 185 bch2_btree_node_type_str(type), 186 k.k->type < KEY_TYPE_MAX 187 ? bch2_bkey_types[k.k->type] 188 : "(unknown)"); 189 190 if (btree_node_type_is_extents(type) && !bkey_whiteout(k.k)) { 191 bkey_fsck_err_on(k.k->size == 0, c, err, 192 bkey_extent_size_zero, 193 "size == 0"); 194 195 bkey_fsck_err_on(k.k->size > k.k->p.offset, c, err, 196 bkey_extent_size_greater_than_offset, 197 "size greater than offset (%u > %llu)", 198 k.k->size, k.k->p.offset); 199 } else { 200 bkey_fsck_err_on(k.k->size, c, err, 201 bkey_size_nonzero, 202 "size != 0"); 203 } 204 205 if (type != BKEY_TYPE_btree) { 206 enum btree_id btree = type - 1; 207 208 if (btree_type_has_snapshots(btree)) { 209 bkey_fsck_err_on(!k.k->p.snapshot, c, err, 210 bkey_snapshot_zero, 211 "snapshot == 0"); 212 } else if (!btree_type_has_snapshot_field(btree)) { 213 bkey_fsck_err_on(k.k->p.snapshot, c, err, 214 bkey_snapshot_nonzero, 215 "nonzero snapshot"); 216 } else { 217 /* 218 * btree uses snapshot field but it's not required to be 219 * nonzero 220 */ 221 } 222 223 bkey_fsck_err_on(bkey_eq(k.k->p, POS_MAX), c, err, 224 bkey_at_pos_max, 225 "key at POS_MAX"); 226 } 227 fsck_err: 228 return ret; 229 } 230 231 int bch2_bkey_invalid(struct bch_fs *c, struct bkey_s_c k, 232 enum btree_node_type type, 233 enum bch_validate_flags flags, 234 struct printbuf *err) 235 { 236 return __bch2_bkey_invalid(c, k, type, flags, err) ?: 237 bch2_bkey_val_invalid(c, k, flags, err); 238 } 239 240 int bch2_bkey_in_btree_node(struct bch_fs *c, struct btree *b, 241 struct bkey_s_c k, struct printbuf *err) 242 { 243 int ret = 0; 244 245 bkey_fsck_err_on(bpos_lt(k.k->p, b->data->min_key), c, err, 246 bkey_before_start_of_btree_node, 247 "key before start of btree node"); 248 249 bkey_fsck_err_on(bpos_gt(k.k->p, b->data->max_key), c, err, 250 bkey_after_end_of_btree_node, 251 "key past end of btree node"); 252 fsck_err: 253 return ret; 254 } 255 256 void bch2_bpos_to_text(struct printbuf *out, struct bpos pos) 257 { 258 if (bpos_eq(pos, POS_MIN)) 259 prt_printf(out, "POS_MIN"); 260 else if (bpos_eq(pos, POS_MAX)) 261 prt_printf(out, "POS_MAX"); 262 else if (bpos_eq(pos, SPOS_MAX)) 263 prt_printf(out, "SPOS_MAX"); 264 else { 265 if (pos.inode == U64_MAX) 266 prt_printf(out, "U64_MAX"); 267 else 268 prt_printf(out, "%llu", pos.inode); 269 prt_printf(out, ":"); 270 if (pos.offset == U64_MAX) 271 prt_printf(out, "U64_MAX"); 272 else 273 prt_printf(out, "%llu", pos.offset); 274 prt_printf(out, ":"); 275 if (pos.snapshot == U32_MAX) 276 prt_printf(out, "U32_MAX"); 277 else 278 prt_printf(out, "%u", pos.snapshot); 279 } 280 } 281 282 void bch2_bkey_to_text(struct printbuf *out, const struct bkey *k) 283 { 284 if (k) { 285 prt_printf(out, "u64s %u type ", k->u64s); 286 287 if (k->type < KEY_TYPE_MAX) 288 prt_printf(out, "%s ", bch2_bkey_types[k->type]); 289 else 290 prt_printf(out, "%u ", k->type); 291 292 bch2_bpos_to_text(out, k->p); 293 294 prt_printf(out, " len %u ver %llu", k->size, k->version.lo); 295 } else { 296 prt_printf(out, "(null)"); 297 } 298 } 299 300 void bch2_val_to_text(struct printbuf *out, struct bch_fs *c, 301 struct bkey_s_c k) 302 { 303 const struct bkey_ops *ops = bch2_bkey_type_ops(k.k->type); 304 305 if (likely(ops->val_to_text)) 306 ops->val_to_text(out, c, k); 307 } 308 309 void bch2_bkey_val_to_text(struct printbuf *out, struct bch_fs *c, 310 struct bkey_s_c k) 311 { 312 bch2_bkey_to_text(out, k.k); 313 314 if (bkey_val_bytes(k.k)) { 315 prt_printf(out, ": "); 316 bch2_val_to_text(out, c, k); 317 } 318 } 319 320 void bch2_bkey_swab_val(struct bkey_s k) 321 { 322 const struct bkey_ops *ops = bch2_bkey_type_ops(k.k->type); 323 324 if (ops->swab) 325 ops->swab(k); 326 } 327 328 bool bch2_bkey_normalize(struct bch_fs *c, struct bkey_s k) 329 { 330 const struct bkey_ops *ops = bch2_bkey_type_ops(k.k->type); 331 332 return ops->key_normalize 333 ? ops->key_normalize(c, k) 334 : false; 335 } 336 337 bool bch2_bkey_merge(struct bch_fs *c, struct bkey_s l, struct bkey_s_c r) 338 { 339 const struct bkey_ops *ops = bch2_bkey_type_ops(l.k->type); 340 341 return ops->key_merge && 342 bch2_bkey_maybe_mergable(l.k, r.k) && 343 (u64) l.k->size + r.k->size <= KEY_SIZE_MAX && 344 !bch2_key_merging_disabled && 345 ops->key_merge(c, l, r); 346 } 347 348 static const struct old_bkey_type { 349 u8 btree_node_type; 350 u8 old; 351 u8 new; 352 } bkey_renumber_table[] = { 353 {BKEY_TYPE_btree, 128, KEY_TYPE_btree_ptr }, 354 {BKEY_TYPE_extents, 128, KEY_TYPE_extent }, 355 {BKEY_TYPE_extents, 129, KEY_TYPE_extent }, 356 {BKEY_TYPE_extents, 130, KEY_TYPE_reservation }, 357 {BKEY_TYPE_inodes, 128, KEY_TYPE_inode }, 358 {BKEY_TYPE_inodes, 130, KEY_TYPE_inode_generation }, 359 {BKEY_TYPE_dirents, 128, KEY_TYPE_dirent }, 360 {BKEY_TYPE_dirents, 129, KEY_TYPE_hash_whiteout }, 361 {BKEY_TYPE_xattrs, 128, KEY_TYPE_xattr }, 362 {BKEY_TYPE_xattrs, 129, KEY_TYPE_hash_whiteout }, 363 {BKEY_TYPE_alloc, 128, KEY_TYPE_alloc }, 364 {BKEY_TYPE_quotas, 128, KEY_TYPE_quota }, 365 }; 366 367 void bch2_bkey_renumber(enum btree_node_type btree_node_type, 368 struct bkey_packed *k, 369 int write) 370 { 371 const struct old_bkey_type *i; 372 373 for (i = bkey_renumber_table; 374 i < bkey_renumber_table + ARRAY_SIZE(bkey_renumber_table); 375 i++) 376 if (btree_node_type == i->btree_node_type && 377 k->type == (write ? i->new : i->old)) { 378 k->type = write ? i->old : i->new; 379 break; 380 } 381 } 382 383 void __bch2_bkey_compat(unsigned level, enum btree_id btree_id, 384 unsigned version, unsigned big_endian, 385 int write, 386 struct bkey_format *f, 387 struct bkey_packed *k) 388 { 389 const struct bkey_ops *ops; 390 struct bkey uk; 391 unsigned nr_compat = 5; 392 int i; 393 394 /* 395 * Do these operations in reverse order in the write path: 396 */ 397 398 for (i = 0; i < nr_compat; i++) 399 switch (!write ? i : nr_compat - 1 - i) { 400 case 0: 401 if (big_endian != CPU_BIG_ENDIAN) { 402 bch2_bkey_swab_key(f, k); 403 } else if (IS_ENABLED(CONFIG_BCACHEFS_DEBUG)) { 404 bch2_bkey_swab_key(f, k); 405 bch2_bkey_swab_key(f, k); 406 } 407 break; 408 case 1: 409 if (version < bcachefs_metadata_version_bkey_renumber) 410 bch2_bkey_renumber(__btree_node_type(level, btree_id), k, write); 411 break; 412 case 2: 413 if (version < bcachefs_metadata_version_inode_btree_change && 414 btree_id == BTREE_ID_inodes) { 415 if (!bkey_packed(k)) { 416 struct bkey_i *u = packed_to_bkey(k); 417 418 swap(u->k.p.inode, u->k.p.offset); 419 } else if (f->bits_per_field[BKEY_FIELD_INODE] && 420 f->bits_per_field[BKEY_FIELD_OFFSET]) { 421 struct bkey_format tmp = *f, *in = f, *out = &tmp; 422 423 swap(tmp.bits_per_field[BKEY_FIELD_INODE], 424 tmp.bits_per_field[BKEY_FIELD_OFFSET]); 425 swap(tmp.field_offset[BKEY_FIELD_INODE], 426 tmp.field_offset[BKEY_FIELD_OFFSET]); 427 428 if (!write) 429 swap(in, out); 430 431 uk = __bch2_bkey_unpack_key(in, k); 432 swap(uk.p.inode, uk.p.offset); 433 BUG_ON(!bch2_bkey_pack_key(k, &uk, out)); 434 } 435 } 436 break; 437 case 3: 438 if (version < bcachefs_metadata_version_snapshot && 439 (level || btree_type_has_snapshots(btree_id))) { 440 struct bkey_i *u = packed_to_bkey(k); 441 442 if (u) { 443 u->k.p.snapshot = write 444 ? 0 : U32_MAX; 445 } else { 446 u64 min_packed = le64_to_cpu(f->field_offset[BKEY_FIELD_SNAPSHOT]); 447 u64 max_packed = min_packed + 448 ~(~0ULL << f->bits_per_field[BKEY_FIELD_SNAPSHOT]); 449 450 uk = __bch2_bkey_unpack_key(f, k); 451 uk.p.snapshot = write 452 ? min_packed : min_t(u64, U32_MAX, max_packed); 453 454 BUG_ON(!bch2_bkey_pack_key(k, &uk, f)); 455 } 456 } 457 458 break; 459 case 4: { 460 struct bkey_s u; 461 462 if (!bkey_packed(k)) { 463 u = bkey_i_to_s(packed_to_bkey(k)); 464 } else { 465 uk = __bch2_bkey_unpack_key(f, k); 466 u.k = &uk; 467 u.v = bkeyp_val(f, k); 468 } 469 470 if (big_endian != CPU_BIG_ENDIAN) 471 bch2_bkey_swab_val(u); 472 473 ops = bch2_bkey_type_ops(k->type); 474 475 if (ops->compat) 476 ops->compat(btree_id, version, big_endian, write, u); 477 break; 478 } 479 default: 480 BUG(); 481 } 482 } 483