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 bkey_invalid_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 bkey_invalid_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 bkey_invalid_flags flags, struct printbuf *err) 62 { 63 return 0; 64 } 65 66 #define bch2_bkey_ops_cookie ((struct bkey_ops) { \ 67 .key_invalid = key_type_cookie_invalid, \ 68 .min_val_size = 8, \ 69 }) 70 71 #define bch2_bkey_ops_hash_whiteout ((struct bkey_ops) {\ 72 .key_invalid = empty_val_key_invalid, \ 73 }) 74 75 static int key_type_inline_data_invalid(struct bch_fs *c, struct bkey_s_c k, 76 enum bkey_invalid_flags flags, struct printbuf *err) 77 { 78 return 0; 79 } 80 81 static void key_type_inline_data_to_text(struct printbuf *out, struct bch_fs *c, 82 struct bkey_s_c k) 83 { 84 struct bkey_s_c_inline_data d = bkey_s_c_to_inline_data(k); 85 unsigned datalen = bkey_inline_data_bytes(k.k); 86 87 prt_printf(out, "datalen %u: %*phN", 88 datalen, min(datalen, 32U), d.v->data); 89 } 90 91 #define bch2_bkey_ops_inline_data ((struct bkey_ops) { \ 92 .key_invalid = key_type_inline_data_invalid, \ 93 .val_to_text = key_type_inline_data_to_text, \ 94 }) 95 96 static bool key_type_set_merge(struct bch_fs *c, struct bkey_s l, struct bkey_s_c r) 97 { 98 bch2_key_resize(l.k, l.k->size + r.k->size); 99 return true; 100 } 101 102 #define bch2_bkey_ops_set ((struct bkey_ops) { \ 103 .key_invalid = empty_val_key_invalid, \ 104 .key_merge = key_type_set_merge, \ 105 }) 106 107 const struct bkey_ops bch2_bkey_ops[] = { 108 #define x(name, nr) [KEY_TYPE_##name] = bch2_bkey_ops_##name, 109 BCH_BKEY_TYPES() 110 #undef x 111 }; 112 113 const struct bkey_ops bch2_bkey_null_ops = { 114 }; 115 116 int bch2_bkey_val_invalid(struct bch_fs *c, struct bkey_s_c k, 117 enum bkey_invalid_flags flags, 118 struct printbuf *err) 119 { 120 const struct bkey_ops *ops = bch2_bkey_type_ops(k.k->type); 121 int ret = 0; 122 123 bkey_fsck_err_on(bkey_val_bytes(k.k) < ops->min_val_size, c, err, 124 bkey_val_size_too_small, 125 "bad val size (%zu < %u)", 126 bkey_val_bytes(k.k), ops->min_val_size); 127 128 if (!ops->key_invalid) 129 return 0; 130 131 ret = ops->key_invalid(c, k, flags, err); 132 fsck_err: 133 return ret; 134 } 135 136 static u64 bch2_key_types_allowed[] = { 137 [BKEY_TYPE_btree] = 138 BIT_ULL(KEY_TYPE_deleted)| 139 BIT_ULL(KEY_TYPE_btree_ptr)| 140 BIT_ULL(KEY_TYPE_btree_ptr_v2), 141 #define x(name, nr, flags, keys) [BKEY_TYPE_##name] = BIT_ULL(KEY_TYPE_deleted)|keys, 142 BCH_BTREE_IDS() 143 #undef x 144 }; 145 146 const char *bch2_btree_node_type_str(enum btree_node_type type) 147 { 148 return type == BKEY_TYPE_btree ? "internal btree node" : bch2_btree_id_str(type - 1); 149 } 150 151 int __bch2_bkey_invalid(struct bch_fs *c, struct bkey_s_c k, 152 enum btree_node_type type, 153 enum bkey_invalid_flags flags, 154 struct printbuf *err) 155 { 156 int ret = 0; 157 158 bkey_fsck_err_on(k.k->u64s < BKEY_U64s, c, err, 159 bkey_u64s_too_small, 160 "u64s too small (%u < %zu)", k.k->u64s, BKEY_U64s); 161 162 if (type >= BKEY_TYPE_NR) 163 return 0; 164 165 bkey_fsck_err_on((flags & BKEY_INVALID_COMMIT) && 166 !(bch2_key_types_allowed[type] & BIT_ULL(k.k->type)), c, err, 167 bkey_invalid_type_for_btree, 168 "invalid key type for btree %s (%s)", 169 bch2_btree_node_type_str(type), bch2_bkey_types[k.k->type]); 170 171 if (btree_node_type_is_extents(type) && !bkey_whiteout(k.k)) { 172 bkey_fsck_err_on(k.k->size == 0, c, err, 173 bkey_extent_size_zero, 174 "size == 0"); 175 176 bkey_fsck_err_on(k.k->size > k.k->p.offset, c, err, 177 bkey_extent_size_greater_than_offset, 178 "size greater than offset (%u > %llu)", 179 k.k->size, k.k->p.offset); 180 } else { 181 bkey_fsck_err_on(k.k->size, c, err, 182 bkey_size_nonzero, 183 "size != 0"); 184 } 185 186 if (type != BKEY_TYPE_btree) { 187 enum btree_id btree = type - 1; 188 189 bkey_fsck_err_on(!btree_type_has_snapshots(btree) && 190 k.k->p.snapshot, c, err, 191 bkey_snapshot_nonzero, 192 "nonzero snapshot"); 193 194 bkey_fsck_err_on(btree_type_has_snapshots(btree) && 195 !k.k->p.snapshot, c, err, 196 bkey_snapshot_zero, 197 "snapshot == 0"); 198 199 bkey_fsck_err_on(bkey_eq(k.k->p, POS_MAX), c, err, 200 bkey_at_pos_max, 201 "key at POS_MAX"); 202 } 203 fsck_err: 204 return ret; 205 } 206 207 int bch2_bkey_invalid(struct bch_fs *c, struct bkey_s_c k, 208 enum btree_node_type type, 209 enum bkey_invalid_flags flags, 210 struct printbuf *err) 211 { 212 return __bch2_bkey_invalid(c, k, type, flags, err) ?: 213 bch2_bkey_val_invalid(c, k, flags, err); 214 } 215 216 int bch2_bkey_in_btree_node(struct bch_fs *c, struct btree *b, 217 struct bkey_s_c k, struct printbuf *err) 218 { 219 int ret = 0; 220 221 bkey_fsck_err_on(bpos_lt(k.k->p, b->data->min_key), c, err, 222 bkey_before_start_of_btree_node, 223 "key before start of btree node"); 224 225 bkey_fsck_err_on(bpos_gt(k.k->p, b->data->max_key), c, err, 226 bkey_after_end_of_btree_node, 227 "key past end of btree node"); 228 fsck_err: 229 return ret; 230 } 231 232 void bch2_bpos_to_text(struct printbuf *out, struct bpos pos) 233 { 234 if (bpos_eq(pos, POS_MIN)) 235 prt_printf(out, "POS_MIN"); 236 else if (bpos_eq(pos, POS_MAX)) 237 prt_printf(out, "POS_MAX"); 238 else if (bpos_eq(pos, SPOS_MAX)) 239 prt_printf(out, "SPOS_MAX"); 240 else { 241 if (pos.inode == U64_MAX) 242 prt_printf(out, "U64_MAX"); 243 else 244 prt_printf(out, "%llu", pos.inode); 245 prt_printf(out, ":"); 246 if (pos.offset == U64_MAX) 247 prt_printf(out, "U64_MAX"); 248 else 249 prt_printf(out, "%llu", pos.offset); 250 prt_printf(out, ":"); 251 if (pos.snapshot == U32_MAX) 252 prt_printf(out, "U32_MAX"); 253 else 254 prt_printf(out, "%u", pos.snapshot); 255 } 256 } 257 258 void bch2_bkey_to_text(struct printbuf *out, const struct bkey *k) 259 { 260 if (k) { 261 prt_printf(out, "u64s %u type ", k->u64s); 262 263 if (k->type < KEY_TYPE_MAX) 264 prt_printf(out, "%s ", bch2_bkey_types[k->type]); 265 else 266 prt_printf(out, "%u ", k->type); 267 268 bch2_bpos_to_text(out, k->p); 269 270 prt_printf(out, " len %u ver %llu", k->size, k->version.lo); 271 } else { 272 prt_printf(out, "(null)"); 273 } 274 } 275 276 void bch2_val_to_text(struct printbuf *out, struct bch_fs *c, 277 struct bkey_s_c k) 278 { 279 const struct bkey_ops *ops = bch2_bkey_type_ops(k.k->type); 280 281 if (likely(ops->val_to_text)) 282 ops->val_to_text(out, c, k); 283 } 284 285 void bch2_bkey_val_to_text(struct printbuf *out, struct bch_fs *c, 286 struct bkey_s_c k) 287 { 288 bch2_bkey_to_text(out, k.k); 289 290 if (bkey_val_bytes(k.k)) { 291 prt_printf(out, ": "); 292 bch2_val_to_text(out, c, k); 293 } 294 } 295 296 void bch2_bkey_swab_val(struct bkey_s k) 297 { 298 const struct bkey_ops *ops = bch2_bkey_type_ops(k.k->type); 299 300 if (ops->swab) 301 ops->swab(k); 302 } 303 304 bool bch2_bkey_normalize(struct bch_fs *c, struct bkey_s k) 305 { 306 const struct bkey_ops *ops = bch2_bkey_type_ops(k.k->type); 307 308 return ops->key_normalize 309 ? ops->key_normalize(c, k) 310 : false; 311 } 312 313 bool bch2_bkey_merge(struct bch_fs *c, struct bkey_s l, struct bkey_s_c r) 314 { 315 const struct bkey_ops *ops = bch2_bkey_type_ops(l.k->type); 316 317 return ops->key_merge && 318 bch2_bkey_maybe_mergable(l.k, r.k) && 319 (u64) l.k->size + r.k->size <= KEY_SIZE_MAX && 320 !bch2_key_merging_disabled && 321 ops->key_merge(c, l, r); 322 } 323 324 static const struct old_bkey_type { 325 u8 btree_node_type; 326 u8 old; 327 u8 new; 328 } bkey_renumber_table[] = { 329 {BKEY_TYPE_btree, 128, KEY_TYPE_btree_ptr }, 330 {BKEY_TYPE_extents, 128, KEY_TYPE_extent }, 331 {BKEY_TYPE_extents, 129, KEY_TYPE_extent }, 332 {BKEY_TYPE_extents, 130, KEY_TYPE_reservation }, 333 {BKEY_TYPE_inodes, 128, KEY_TYPE_inode }, 334 {BKEY_TYPE_inodes, 130, KEY_TYPE_inode_generation }, 335 {BKEY_TYPE_dirents, 128, KEY_TYPE_dirent }, 336 {BKEY_TYPE_dirents, 129, KEY_TYPE_hash_whiteout }, 337 {BKEY_TYPE_xattrs, 128, KEY_TYPE_xattr }, 338 {BKEY_TYPE_xattrs, 129, KEY_TYPE_hash_whiteout }, 339 {BKEY_TYPE_alloc, 128, KEY_TYPE_alloc }, 340 {BKEY_TYPE_quotas, 128, KEY_TYPE_quota }, 341 }; 342 343 void bch2_bkey_renumber(enum btree_node_type btree_node_type, 344 struct bkey_packed *k, 345 int write) 346 { 347 const struct old_bkey_type *i; 348 349 for (i = bkey_renumber_table; 350 i < bkey_renumber_table + ARRAY_SIZE(bkey_renumber_table); 351 i++) 352 if (btree_node_type == i->btree_node_type && 353 k->type == (write ? i->new : i->old)) { 354 k->type = write ? i->old : i->new; 355 break; 356 } 357 } 358 359 void __bch2_bkey_compat(unsigned level, enum btree_id btree_id, 360 unsigned version, unsigned big_endian, 361 int write, 362 struct bkey_format *f, 363 struct bkey_packed *k) 364 { 365 const struct bkey_ops *ops; 366 struct bkey uk; 367 unsigned nr_compat = 5; 368 int i; 369 370 /* 371 * Do these operations in reverse order in the write path: 372 */ 373 374 for (i = 0; i < nr_compat; i++) 375 switch (!write ? i : nr_compat - 1 - i) { 376 case 0: 377 if (big_endian != CPU_BIG_ENDIAN) 378 bch2_bkey_swab_key(f, k); 379 break; 380 case 1: 381 if (version < bcachefs_metadata_version_bkey_renumber) 382 bch2_bkey_renumber(__btree_node_type(level, btree_id), k, write); 383 break; 384 case 2: 385 if (version < bcachefs_metadata_version_inode_btree_change && 386 btree_id == BTREE_ID_inodes) { 387 if (!bkey_packed(k)) { 388 struct bkey_i *u = packed_to_bkey(k); 389 390 swap(u->k.p.inode, u->k.p.offset); 391 } else if (f->bits_per_field[BKEY_FIELD_INODE] && 392 f->bits_per_field[BKEY_FIELD_OFFSET]) { 393 struct bkey_format tmp = *f, *in = f, *out = &tmp; 394 395 swap(tmp.bits_per_field[BKEY_FIELD_INODE], 396 tmp.bits_per_field[BKEY_FIELD_OFFSET]); 397 swap(tmp.field_offset[BKEY_FIELD_INODE], 398 tmp.field_offset[BKEY_FIELD_OFFSET]); 399 400 if (!write) 401 swap(in, out); 402 403 uk = __bch2_bkey_unpack_key(in, k); 404 swap(uk.p.inode, uk.p.offset); 405 BUG_ON(!bch2_bkey_pack_key(k, &uk, out)); 406 } 407 } 408 break; 409 case 3: 410 if (version < bcachefs_metadata_version_snapshot && 411 (level || btree_type_has_snapshots(btree_id))) { 412 struct bkey_i *u = packed_to_bkey(k); 413 414 if (u) { 415 u->k.p.snapshot = write 416 ? 0 : U32_MAX; 417 } else { 418 u64 min_packed = le64_to_cpu(f->field_offset[BKEY_FIELD_SNAPSHOT]); 419 u64 max_packed = min_packed + 420 ~(~0ULL << f->bits_per_field[BKEY_FIELD_SNAPSHOT]); 421 422 uk = __bch2_bkey_unpack_key(f, k); 423 uk.p.snapshot = write 424 ? min_packed : min_t(u64, U32_MAX, max_packed); 425 426 BUG_ON(!bch2_bkey_pack_key(k, &uk, f)); 427 } 428 } 429 430 break; 431 case 4: { 432 struct bkey_s u; 433 434 if (!bkey_packed(k)) { 435 u = bkey_i_to_s(packed_to_bkey(k)); 436 } else { 437 uk = __bch2_bkey_unpack_key(f, k); 438 u.k = &uk; 439 u.v = bkeyp_val(f, k); 440 } 441 442 if (big_endian != CPU_BIG_ENDIAN) 443 bch2_bkey_swab_val(u); 444 445 ops = bch2_bkey_type_ops(k->type); 446 447 if (ops->compat) 448 ops->compat(btree_id, version, big_endian, write, u); 449 break; 450 } 451 default: 452 BUG(); 453 } 454 } 455