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 "io_misc.h" 14 #include "lru.h" 15 #include "quota.h" 16 #include "reflink.h" 17 #include "snapshot.h" 18 #include "subvolume.h" 19 #include "xattr.h" 20 21 const char * const bch2_bkey_types[] = { 22 #define x(name, nr) #name, 23 BCH_BKEY_TYPES() 24 #undef x 25 NULL 26 }; 27 28 static int deleted_key_invalid(const struct bch_fs *c, struct bkey_s_c k, 29 enum bkey_invalid_flags flags, struct printbuf *err) 30 { 31 return 0; 32 } 33 34 #define bch2_bkey_ops_deleted ((struct bkey_ops) { \ 35 .key_invalid = deleted_key_invalid, \ 36 }) 37 38 #define bch2_bkey_ops_whiteout ((struct bkey_ops) { \ 39 .key_invalid = deleted_key_invalid, \ 40 }) 41 42 static int empty_val_key_invalid(const struct bch_fs *c, struct bkey_s_c k, 43 enum bkey_invalid_flags flags, struct printbuf *err) 44 { 45 if (bkey_val_bytes(k.k)) { 46 prt_printf(err, "incorrect value size (%zu != 0)", 47 bkey_val_bytes(k.k)); 48 return -BCH_ERR_invalid_bkey; 49 } 50 51 return 0; 52 } 53 54 #define bch2_bkey_ops_error ((struct bkey_ops) { \ 55 .key_invalid = empty_val_key_invalid, \ 56 }) 57 58 static int key_type_cookie_invalid(const struct bch_fs *c, struct bkey_s_c k, 59 enum bkey_invalid_flags flags, struct printbuf *err) 60 { 61 return 0; 62 } 63 64 #define bch2_bkey_ops_cookie ((struct bkey_ops) { \ 65 .key_invalid = key_type_cookie_invalid, \ 66 .min_val_size = 8, \ 67 }) 68 69 #define bch2_bkey_ops_hash_whiteout ((struct bkey_ops) {\ 70 .key_invalid = empty_val_key_invalid, \ 71 }) 72 73 static int key_type_inline_data_invalid(const struct bch_fs *c, struct bkey_s_c k, 74 enum bkey_invalid_flags flags, struct printbuf *err) 75 { 76 return 0; 77 } 78 79 static void key_type_inline_data_to_text(struct printbuf *out, struct bch_fs *c, 80 struct bkey_s_c k) 81 { 82 struct bkey_s_c_inline_data d = bkey_s_c_to_inline_data(k); 83 unsigned datalen = bkey_inline_data_bytes(k.k); 84 85 prt_printf(out, "datalen %u: %*phN", 86 datalen, min(datalen, 32U), d.v->data); 87 } 88 89 #define bch2_bkey_ops_inline_data ((struct bkey_ops) { \ 90 .key_invalid = key_type_inline_data_invalid, \ 91 .val_to_text = key_type_inline_data_to_text, \ 92 }) 93 94 static int key_type_set_invalid(const struct bch_fs *c, struct bkey_s_c k, 95 enum bkey_invalid_flags flags, struct printbuf *err) 96 { 97 if (bkey_val_bytes(k.k)) { 98 prt_printf(err, "incorrect value size (%zu != %zu)", 99 bkey_val_bytes(k.k), sizeof(struct bch_cookie)); 100 return -BCH_ERR_invalid_bkey; 101 } 102 103 return 0; 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_invalid = key_type_set_invalid, \ 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_invalid(struct bch_fs *c, struct bkey_s_c k, 127 enum bkey_invalid_flags flags, 128 struct printbuf *err) 129 { 130 const struct bkey_ops *ops = bch2_bkey_type_ops(k.k->type); 131 132 if (bkey_val_bytes(k.k) < ops->min_val_size) { 133 prt_printf(err, "bad val size (%zu < %u)", 134 bkey_val_bytes(k.k), ops->min_val_size); 135 return -BCH_ERR_invalid_bkey; 136 } 137 138 if (!ops->key_invalid) 139 return 0; 140 141 return ops->key_invalid(c, k, flags, err); 142 } 143 144 static u64 bch2_key_types_allowed[] = { 145 #define x(name, nr, flags, keys) [BKEY_TYPE_##name] = BIT_ULL(KEY_TYPE_deleted)|keys, 146 BCH_BTREE_IDS() 147 #undef x 148 [BKEY_TYPE_btree] = 149 BIT_ULL(KEY_TYPE_deleted)| 150 BIT_ULL(KEY_TYPE_btree_ptr)| 151 BIT_ULL(KEY_TYPE_btree_ptr_v2), 152 }; 153 154 int __bch2_bkey_invalid(struct bch_fs *c, struct bkey_s_c k, 155 enum btree_node_type type, 156 enum bkey_invalid_flags flags, 157 struct printbuf *err) 158 { 159 if (k.k->u64s < BKEY_U64s) { 160 prt_printf(err, "u64s too small (%u < %zu)", k.k->u64s, BKEY_U64s); 161 return -BCH_ERR_invalid_bkey; 162 } 163 164 if (flags & BKEY_INVALID_COMMIT && 165 !(bch2_key_types_allowed[type] & BIT_ULL(k.k->type))) { 166 prt_printf(err, "invalid key type for btree %s (%s)", 167 bch2_btree_ids[type], bch2_bkey_types[k.k->type]); 168 return -BCH_ERR_invalid_bkey; 169 } 170 171 if (btree_node_type_is_extents(type) && !bkey_whiteout(k.k)) { 172 if (k.k->size == 0) { 173 prt_printf(err, "size == 0"); 174 return -BCH_ERR_invalid_bkey; 175 } 176 177 if (k.k->size > k.k->p.offset) { 178 prt_printf(err, "size greater than offset (%u > %llu)", 179 k.k->size, k.k->p.offset); 180 return -BCH_ERR_invalid_bkey; 181 } 182 } else { 183 if (k.k->size) { 184 prt_printf(err, "size != 0"); 185 return -BCH_ERR_invalid_bkey; 186 } 187 } 188 189 if (type != BKEY_TYPE_btree) { 190 if (!btree_type_has_snapshots((enum btree_id) type) && 191 k.k->p.snapshot) { 192 prt_printf(err, "nonzero snapshot"); 193 return -BCH_ERR_invalid_bkey; 194 } 195 196 if (btree_type_has_snapshots((enum btree_id) type) && 197 !k.k->p.snapshot) { 198 prt_printf(err, "snapshot == 0"); 199 return -BCH_ERR_invalid_bkey; 200 } 201 202 if (bkey_eq(k.k->p, POS_MAX)) { 203 prt_printf(err, "key at POS_MAX"); 204 return -BCH_ERR_invalid_bkey; 205 } 206 } 207 208 return 0; 209 } 210 211 int bch2_bkey_invalid(struct bch_fs *c, struct bkey_s_c k, 212 enum btree_node_type type, 213 enum bkey_invalid_flags flags, 214 struct printbuf *err) 215 { 216 return __bch2_bkey_invalid(c, k, type, flags, err) ?: 217 bch2_bkey_val_invalid(c, k, flags, err); 218 } 219 220 int bch2_bkey_in_btree_node(struct btree *b, struct bkey_s_c k, 221 struct printbuf *err) 222 { 223 if (bpos_lt(k.k->p, b->data->min_key)) { 224 prt_printf(err, "key before start of btree node"); 225 return -BCH_ERR_invalid_bkey; 226 } 227 228 if (bpos_gt(k.k->p, b->data->max_key)) { 229 prt_printf(err, "key past end of btree node"); 230 return -BCH_ERR_invalid_bkey; 231 } 232 233 return 0; 234 } 235 236 void bch2_bpos_to_text(struct printbuf *out, struct bpos pos) 237 { 238 if (bpos_eq(pos, POS_MIN)) 239 prt_printf(out, "POS_MIN"); 240 else if (bpos_eq(pos, POS_MAX)) 241 prt_printf(out, "POS_MAX"); 242 else if (bpos_eq(pos, SPOS_MAX)) 243 prt_printf(out, "SPOS_MAX"); 244 else { 245 if (pos.inode == U64_MAX) 246 prt_printf(out, "U64_MAX"); 247 else 248 prt_printf(out, "%llu", pos.inode); 249 prt_printf(out, ":"); 250 if (pos.offset == U64_MAX) 251 prt_printf(out, "U64_MAX"); 252 else 253 prt_printf(out, "%llu", pos.offset); 254 prt_printf(out, ":"); 255 if (pos.snapshot == U32_MAX) 256 prt_printf(out, "U32_MAX"); 257 else 258 prt_printf(out, "%u", pos.snapshot); 259 } 260 } 261 262 void bch2_bkey_to_text(struct printbuf *out, const struct bkey *k) 263 { 264 if (k) { 265 prt_printf(out, "u64s %u type ", k->u64s); 266 267 if (k->type < KEY_TYPE_MAX) 268 prt_printf(out, "%s ", bch2_bkey_types[k->type]); 269 else 270 prt_printf(out, "%u ", k->type); 271 272 bch2_bpos_to_text(out, k->p); 273 274 prt_printf(out, " len %u ver %llu", k->size, k->version.lo); 275 } else { 276 prt_printf(out, "(null)"); 277 } 278 } 279 280 void bch2_val_to_text(struct printbuf *out, struct bch_fs *c, 281 struct bkey_s_c k) 282 { 283 const struct bkey_ops *ops = bch2_bkey_type_ops(k.k->type); 284 285 if (likely(ops->val_to_text)) 286 ops->val_to_text(out, c, k); 287 } 288 289 void bch2_bkey_val_to_text(struct printbuf *out, struct bch_fs *c, 290 struct bkey_s_c k) 291 { 292 bch2_bkey_to_text(out, k.k); 293 294 if (bkey_val_bytes(k.k)) { 295 prt_printf(out, ": "); 296 bch2_val_to_text(out, c, k); 297 } 298 } 299 300 void bch2_bkey_swab_val(struct bkey_s k) 301 { 302 const struct bkey_ops *ops = bch2_bkey_type_ops(k.k->type); 303 304 if (ops->swab) 305 ops->swab(k); 306 } 307 308 bool bch2_bkey_normalize(struct bch_fs *c, struct bkey_s k) 309 { 310 const struct bkey_ops *ops = bch2_bkey_type_ops(k.k->type); 311 312 return ops->key_normalize 313 ? ops->key_normalize(c, k) 314 : false; 315 } 316 317 bool bch2_bkey_merge(struct bch_fs *c, struct bkey_s l, struct bkey_s_c r) 318 { 319 const struct bkey_ops *ops = bch2_bkey_type_ops(l.k->type); 320 321 return ops->key_merge && 322 bch2_bkey_maybe_mergable(l.k, r.k) && 323 (u64) l.k->size + r.k->size <= KEY_SIZE_MAX && 324 !bch2_key_merging_disabled && 325 ops->key_merge(c, l, r); 326 } 327 328 static const struct old_bkey_type { 329 u8 btree_node_type; 330 u8 old; 331 u8 new; 332 } bkey_renumber_table[] = { 333 {BKEY_TYPE_btree, 128, KEY_TYPE_btree_ptr }, 334 {BKEY_TYPE_extents, 128, KEY_TYPE_extent }, 335 {BKEY_TYPE_extents, 129, KEY_TYPE_extent }, 336 {BKEY_TYPE_extents, 130, KEY_TYPE_reservation }, 337 {BKEY_TYPE_inodes, 128, KEY_TYPE_inode }, 338 {BKEY_TYPE_inodes, 130, KEY_TYPE_inode_generation }, 339 {BKEY_TYPE_dirents, 128, KEY_TYPE_dirent }, 340 {BKEY_TYPE_dirents, 129, KEY_TYPE_hash_whiteout }, 341 {BKEY_TYPE_xattrs, 128, KEY_TYPE_xattr }, 342 {BKEY_TYPE_xattrs, 129, KEY_TYPE_hash_whiteout }, 343 {BKEY_TYPE_alloc, 128, KEY_TYPE_alloc }, 344 {BKEY_TYPE_quotas, 128, KEY_TYPE_quota }, 345 }; 346 347 void bch2_bkey_renumber(enum btree_node_type btree_node_type, 348 struct bkey_packed *k, 349 int write) 350 { 351 const struct old_bkey_type *i; 352 353 for (i = bkey_renumber_table; 354 i < bkey_renumber_table + ARRAY_SIZE(bkey_renumber_table); 355 i++) 356 if (btree_node_type == i->btree_node_type && 357 k->type == (write ? i->new : i->old)) { 358 k->type = write ? i->old : i->new; 359 break; 360 } 361 } 362 363 void __bch2_bkey_compat(unsigned level, enum btree_id btree_id, 364 unsigned version, unsigned big_endian, 365 int write, 366 struct bkey_format *f, 367 struct bkey_packed *k) 368 { 369 const struct bkey_ops *ops; 370 struct bkey uk; 371 unsigned nr_compat = 5; 372 int i; 373 374 /* 375 * Do these operations in reverse order in the write path: 376 */ 377 378 for (i = 0; i < nr_compat; i++) 379 switch (!write ? i : nr_compat - 1 - i) { 380 case 0: 381 if (big_endian != CPU_BIG_ENDIAN) 382 bch2_bkey_swab_key(f, k); 383 break; 384 case 1: 385 if (version < bcachefs_metadata_version_bkey_renumber) 386 bch2_bkey_renumber(__btree_node_type(level, btree_id), k, write); 387 break; 388 case 2: 389 if (version < bcachefs_metadata_version_inode_btree_change && 390 btree_id == BTREE_ID_inodes) { 391 if (!bkey_packed(k)) { 392 struct bkey_i *u = packed_to_bkey(k); 393 394 swap(u->k.p.inode, u->k.p.offset); 395 } else if (f->bits_per_field[BKEY_FIELD_INODE] && 396 f->bits_per_field[BKEY_FIELD_OFFSET]) { 397 struct bkey_format tmp = *f, *in = f, *out = &tmp; 398 399 swap(tmp.bits_per_field[BKEY_FIELD_INODE], 400 tmp.bits_per_field[BKEY_FIELD_OFFSET]); 401 swap(tmp.field_offset[BKEY_FIELD_INODE], 402 tmp.field_offset[BKEY_FIELD_OFFSET]); 403 404 if (!write) 405 swap(in, out); 406 407 uk = __bch2_bkey_unpack_key(in, k); 408 swap(uk.p.inode, uk.p.offset); 409 BUG_ON(!bch2_bkey_pack_key(k, &uk, out)); 410 } 411 } 412 break; 413 case 3: 414 if (version < bcachefs_metadata_version_snapshot && 415 (level || btree_type_has_snapshots(btree_id))) { 416 struct bkey_i *u = packed_to_bkey(k); 417 418 if (u) { 419 u->k.p.snapshot = write 420 ? 0 : U32_MAX; 421 } else { 422 u64 min_packed = le64_to_cpu(f->field_offset[BKEY_FIELD_SNAPSHOT]); 423 u64 max_packed = min_packed + 424 ~(~0ULL << f->bits_per_field[BKEY_FIELD_SNAPSHOT]); 425 426 uk = __bch2_bkey_unpack_key(f, k); 427 uk.p.snapshot = write 428 ? min_packed : min_t(u64, U32_MAX, max_packed); 429 430 BUG_ON(!bch2_bkey_pack_key(k, &uk, f)); 431 } 432 } 433 434 break; 435 case 4: { 436 struct bkey_s u; 437 438 if (!bkey_packed(k)) { 439 u = bkey_i_to_s(packed_to_bkey(k)); 440 } else { 441 uk = __bch2_bkey_unpack_key(f, k); 442 u.k = &uk; 443 u.v = bkeyp_val(f, k); 444 } 445 446 if (big_endian != CPU_BIG_ENDIAN) 447 bch2_bkey_swab_val(u); 448 449 ops = bch2_bkey_type_ops(k->type); 450 451 if (ops->compat) 452 ops->compat(btree_id, version, big_endian, write, u); 453 break; 454 } 455 default: 456 BUG(); 457 } 458 } 459