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 if (btree_type_has_snapshots(btree)) { 190 bkey_fsck_err_on(!k.k->p.snapshot, c, err, 191 bkey_snapshot_zero, 192 "snapshot == 0"); 193 } else if (!btree_type_has_snapshot_field(btree)) { 194 bkey_fsck_err_on(k.k->p.snapshot, c, err, 195 bkey_snapshot_nonzero, 196 "nonzero snapshot"); 197 } else { 198 /* 199 * btree uses snapshot field but it's not required to be 200 * nonzero 201 */ 202 } 203 204 bkey_fsck_err_on(bkey_eq(k.k->p, POS_MAX), c, err, 205 bkey_at_pos_max, 206 "key at POS_MAX"); 207 } 208 fsck_err: 209 return ret; 210 } 211 212 int bch2_bkey_invalid(struct bch_fs *c, struct bkey_s_c k, 213 enum btree_node_type type, 214 enum bkey_invalid_flags flags, 215 struct printbuf *err) 216 { 217 return __bch2_bkey_invalid(c, k, type, flags, err) ?: 218 bch2_bkey_val_invalid(c, k, flags, err); 219 } 220 221 int bch2_bkey_in_btree_node(struct bch_fs *c, struct btree *b, 222 struct bkey_s_c k, struct printbuf *err) 223 { 224 int ret = 0; 225 226 bkey_fsck_err_on(bpos_lt(k.k->p, b->data->min_key), c, err, 227 bkey_before_start_of_btree_node, 228 "key before start of btree node"); 229 230 bkey_fsck_err_on(bpos_gt(k.k->p, b->data->max_key), c, err, 231 bkey_after_end_of_btree_node, 232 "key past end of btree node"); 233 fsck_err: 234 return ret; 235 } 236 237 void bch2_bpos_to_text(struct printbuf *out, struct bpos pos) 238 { 239 if (bpos_eq(pos, POS_MIN)) 240 prt_printf(out, "POS_MIN"); 241 else if (bpos_eq(pos, POS_MAX)) 242 prt_printf(out, "POS_MAX"); 243 else if (bpos_eq(pos, SPOS_MAX)) 244 prt_printf(out, "SPOS_MAX"); 245 else { 246 if (pos.inode == U64_MAX) 247 prt_printf(out, "U64_MAX"); 248 else 249 prt_printf(out, "%llu", pos.inode); 250 prt_printf(out, ":"); 251 if (pos.offset == U64_MAX) 252 prt_printf(out, "U64_MAX"); 253 else 254 prt_printf(out, "%llu", pos.offset); 255 prt_printf(out, ":"); 256 if (pos.snapshot == U32_MAX) 257 prt_printf(out, "U32_MAX"); 258 else 259 prt_printf(out, "%u", pos.snapshot); 260 } 261 } 262 263 void bch2_bkey_to_text(struct printbuf *out, const struct bkey *k) 264 { 265 if (k) { 266 prt_printf(out, "u64s %u type ", k->u64s); 267 268 if (k->type < KEY_TYPE_MAX) 269 prt_printf(out, "%s ", bch2_bkey_types[k->type]); 270 else 271 prt_printf(out, "%u ", k->type); 272 273 bch2_bpos_to_text(out, k->p); 274 275 prt_printf(out, " len %u ver %llu", k->size, k->version.lo); 276 } else { 277 prt_printf(out, "(null)"); 278 } 279 } 280 281 void bch2_val_to_text(struct printbuf *out, struct bch_fs *c, 282 struct bkey_s_c k) 283 { 284 const struct bkey_ops *ops = bch2_bkey_type_ops(k.k->type); 285 286 if (likely(ops->val_to_text)) 287 ops->val_to_text(out, c, k); 288 } 289 290 void bch2_bkey_val_to_text(struct printbuf *out, struct bch_fs *c, 291 struct bkey_s_c k) 292 { 293 bch2_bkey_to_text(out, k.k); 294 295 if (bkey_val_bytes(k.k)) { 296 prt_printf(out, ": "); 297 bch2_val_to_text(out, c, k); 298 } 299 } 300 301 void bch2_bkey_swab_val(struct bkey_s k) 302 { 303 const struct bkey_ops *ops = bch2_bkey_type_ops(k.k->type); 304 305 if (ops->swab) 306 ops->swab(k); 307 } 308 309 bool bch2_bkey_normalize(struct bch_fs *c, struct bkey_s k) 310 { 311 const struct bkey_ops *ops = bch2_bkey_type_ops(k.k->type); 312 313 return ops->key_normalize 314 ? ops->key_normalize(c, k) 315 : false; 316 } 317 318 bool bch2_bkey_merge(struct bch_fs *c, struct bkey_s l, struct bkey_s_c r) 319 { 320 const struct bkey_ops *ops = bch2_bkey_type_ops(l.k->type); 321 322 return ops->key_merge && 323 bch2_bkey_maybe_mergable(l.k, r.k) && 324 (u64) l.k->size + r.k->size <= KEY_SIZE_MAX && 325 !bch2_key_merging_disabled && 326 ops->key_merge(c, l, r); 327 } 328 329 static const struct old_bkey_type { 330 u8 btree_node_type; 331 u8 old; 332 u8 new; 333 } bkey_renumber_table[] = { 334 {BKEY_TYPE_btree, 128, KEY_TYPE_btree_ptr }, 335 {BKEY_TYPE_extents, 128, KEY_TYPE_extent }, 336 {BKEY_TYPE_extents, 129, KEY_TYPE_extent }, 337 {BKEY_TYPE_extents, 130, KEY_TYPE_reservation }, 338 {BKEY_TYPE_inodes, 128, KEY_TYPE_inode }, 339 {BKEY_TYPE_inodes, 130, KEY_TYPE_inode_generation }, 340 {BKEY_TYPE_dirents, 128, KEY_TYPE_dirent }, 341 {BKEY_TYPE_dirents, 129, KEY_TYPE_hash_whiteout }, 342 {BKEY_TYPE_xattrs, 128, KEY_TYPE_xattr }, 343 {BKEY_TYPE_xattrs, 129, KEY_TYPE_hash_whiteout }, 344 {BKEY_TYPE_alloc, 128, KEY_TYPE_alloc }, 345 {BKEY_TYPE_quotas, 128, KEY_TYPE_quota }, 346 }; 347 348 void bch2_bkey_renumber(enum btree_node_type btree_node_type, 349 struct bkey_packed *k, 350 int write) 351 { 352 const struct old_bkey_type *i; 353 354 for (i = bkey_renumber_table; 355 i < bkey_renumber_table + ARRAY_SIZE(bkey_renumber_table); 356 i++) 357 if (btree_node_type == i->btree_node_type && 358 k->type == (write ? i->new : i->old)) { 359 k->type = write ? i->old : i->new; 360 break; 361 } 362 } 363 364 void __bch2_bkey_compat(unsigned level, enum btree_id btree_id, 365 unsigned version, unsigned big_endian, 366 int write, 367 struct bkey_format *f, 368 struct bkey_packed *k) 369 { 370 const struct bkey_ops *ops; 371 struct bkey uk; 372 unsigned nr_compat = 5; 373 int i; 374 375 /* 376 * Do these operations in reverse order in the write path: 377 */ 378 379 for (i = 0; i < nr_compat; i++) 380 switch (!write ? i : nr_compat - 1 - i) { 381 case 0: 382 if (big_endian != CPU_BIG_ENDIAN) 383 bch2_bkey_swab_key(f, k); 384 break; 385 case 1: 386 if (version < bcachefs_metadata_version_bkey_renumber) 387 bch2_bkey_renumber(__btree_node_type(level, btree_id), k, write); 388 break; 389 case 2: 390 if (version < bcachefs_metadata_version_inode_btree_change && 391 btree_id == BTREE_ID_inodes) { 392 if (!bkey_packed(k)) { 393 struct bkey_i *u = packed_to_bkey(k); 394 395 swap(u->k.p.inode, u->k.p.offset); 396 } else if (f->bits_per_field[BKEY_FIELD_INODE] && 397 f->bits_per_field[BKEY_FIELD_OFFSET]) { 398 struct bkey_format tmp = *f, *in = f, *out = &tmp; 399 400 swap(tmp.bits_per_field[BKEY_FIELD_INODE], 401 tmp.bits_per_field[BKEY_FIELD_OFFSET]); 402 swap(tmp.field_offset[BKEY_FIELD_INODE], 403 tmp.field_offset[BKEY_FIELD_OFFSET]); 404 405 if (!write) 406 swap(in, out); 407 408 uk = __bch2_bkey_unpack_key(in, k); 409 swap(uk.p.inode, uk.p.offset); 410 BUG_ON(!bch2_bkey_pack_key(k, &uk, out)); 411 } 412 } 413 break; 414 case 3: 415 if (version < bcachefs_metadata_version_snapshot && 416 (level || btree_type_has_snapshots(btree_id))) { 417 struct bkey_i *u = packed_to_bkey(k); 418 419 if (u) { 420 u->k.p.snapshot = write 421 ? 0 : U32_MAX; 422 } else { 423 u64 min_packed = le64_to_cpu(f->field_offset[BKEY_FIELD_SNAPSHOT]); 424 u64 max_packed = min_packed + 425 ~(~0ULL << f->bits_per_field[BKEY_FIELD_SNAPSHOT]); 426 427 uk = __bch2_bkey_unpack_key(f, k); 428 uk.p.snapshot = write 429 ? min_packed : min_t(u64, U32_MAX, max_packed); 430 431 BUG_ON(!bch2_bkey_pack_key(k, &uk, f)); 432 } 433 } 434 435 break; 436 case 4: { 437 struct bkey_s u; 438 439 if (!bkey_packed(k)) { 440 u = bkey_i_to_s(packed_to_bkey(k)); 441 } else { 442 uk = __bch2_bkey_unpack_key(f, k); 443 u.k = &uk; 444 u.v = bkeyp_val(f, k); 445 } 446 447 if (big_endian != CPU_BIG_ENDIAN) 448 bch2_bkey_swab_val(u); 449 450 ops = bch2_bkey_type_ops(k->type); 451 452 if (ops->compat) 453 ops->compat(btree_id, version, big_endian, write, u); 454 break; 455 } 456 default: 457 BUG(); 458 } 459 } 460