1 // SPDX-License-Identifier: GPL-2.0 2 3 #include "bcachefs.h" 4 #include "btree_key_cache.h" 5 #include "btree_write_buffer.h" 6 #include "bkey_methods.h" 7 #include "btree_update.h" 8 #include "buckets.h" 9 #include "compress.h" 10 #include "dirent.h" 11 #include "error.h" 12 #include "extents.h" 13 #include "extent_update.h" 14 #include "inode.h" 15 #include "str_hash.h" 16 #include "snapshot.h" 17 #include "subvolume.h" 18 #include "varint.h" 19 20 #include <linux/random.h> 21 22 #include <asm/unaligned.h> 23 24 #define x(name, ...) #name, 25 const char * const bch2_inode_opts[] = { 26 BCH_INODE_OPTS() 27 NULL, 28 }; 29 30 static const char * const bch2_inode_flag_strs[] = { 31 BCH_INODE_FLAGS() 32 NULL 33 }; 34 #undef x 35 36 static const u8 byte_table[8] = { 1, 2, 3, 4, 6, 8, 10, 13 }; 37 38 static int inode_decode_field(const u8 *in, const u8 *end, 39 u64 out[2], unsigned *out_bits) 40 { 41 __be64 be[2] = { 0, 0 }; 42 unsigned bytes, shift; 43 u8 *p; 44 45 if (in >= end) 46 return -1; 47 48 if (!*in) 49 return -1; 50 51 /* 52 * position of highest set bit indicates number of bytes: 53 * shift = number of bits to remove in high byte: 54 */ 55 shift = 8 - __fls(*in); /* 1 <= shift <= 8 */ 56 bytes = byte_table[shift - 1]; 57 58 if (in + bytes > end) 59 return -1; 60 61 p = (u8 *) be + 16 - bytes; 62 memcpy(p, in, bytes); 63 *p ^= (1 << 8) >> shift; 64 65 out[0] = be64_to_cpu(be[0]); 66 out[1] = be64_to_cpu(be[1]); 67 *out_bits = out[0] ? 64 + fls64(out[0]) : fls64(out[1]); 68 69 return bytes; 70 } 71 72 static inline void bch2_inode_pack_inlined(struct bkey_inode_buf *packed, 73 const struct bch_inode_unpacked *inode) 74 { 75 struct bkey_i_inode_v3 *k = &packed->inode; 76 u8 *out = k->v.fields; 77 u8 *end = (void *) &packed[1]; 78 u8 *last_nonzero_field = out; 79 unsigned nr_fields = 0, last_nonzero_fieldnr = 0; 80 unsigned bytes; 81 int ret; 82 83 bkey_inode_v3_init(&packed->inode.k_i); 84 packed->inode.k.p.offset = inode->bi_inum; 85 packed->inode.v.bi_journal_seq = cpu_to_le64(inode->bi_journal_seq); 86 packed->inode.v.bi_hash_seed = inode->bi_hash_seed; 87 packed->inode.v.bi_flags = cpu_to_le64(inode->bi_flags); 88 packed->inode.v.bi_sectors = cpu_to_le64(inode->bi_sectors); 89 packed->inode.v.bi_size = cpu_to_le64(inode->bi_size); 90 packed->inode.v.bi_version = cpu_to_le64(inode->bi_version); 91 SET_INODEv3_MODE(&packed->inode.v, inode->bi_mode); 92 SET_INODEv3_FIELDS_START(&packed->inode.v, INODEv3_FIELDS_START_CUR); 93 94 95 #define x(_name, _bits) \ 96 nr_fields++; \ 97 \ 98 if (inode->_name) { \ 99 ret = bch2_varint_encode_fast(out, inode->_name); \ 100 out += ret; \ 101 \ 102 if (_bits > 64) \ 103 *out++ = 0; \ 104 \ 105 last_nonzero_field = out; \ 106 last_nonzero_fieldnr = nr_fields; \ 107 } else { \ 108 *out++ = 0; \ 109 \ 110 if (_bits > 64) \ 111 *out++ = 0; \ 112 } 113 114 BCH_INODE_FIELDS_v3() 115 #undef x 116 BUG_ON(out > end); 117 118 out = last_nonzero_field; 119 nr_fields = last_nonzero_fieldnr; 120 121 bytes = out - (u8 *) &packed->inode.v; 122 set_bkey_val_bytes(&packed->inode.k, bytes); 123 memset_u64s_tail(&packed->inode.v, 0, bytes); 124 125 SET_INODEv3_NR_FIELDS(&k->v, nr_fields); 126 127 if (IS_ENABLED(CONFIG_BCACHEFS_DEBUG)) { 128 struct bch_inode_unpacked unpacked; 129 130 ret = bch2_inode_unpack(bkey_i_to_s_c(&packed->inode.k_i), &unpacked); 131 BUG_ON(ret); 132 BUG_ON(unpacked.bi_inum != inode->bi_inum); 133 BUG_ON(unpacked.bi_hash_seed != inode->bi_hash_seed); 134 BUG_ON(unpacked.bi_sectors != inode->bi_sectors); 135 BUG_ON(unpacked.bi_size != inode->bi_size); 136 BUG_ON(unpacked.bi_version != inode->bi_version); 137 BUG_ON(unpacked.bi_mode != inode->bi_mode); 138 139 #define x(_name, _bits) if (unpacked._name != inode->_name) \ 140 panic("unpacked %llu should be %llu", \ 141 (u64) unpacked._name, (u64) inode->_name); 142 BCH_INODE_FIELDS_v3() 143 #undef x 144 } 145 } 146 147 void bch2_inode_pack(struct bkey_inode_buf *packed, 148 const struct bch_inode_unpacked *inode) 149 { 150 bch2_inode_pack_inlined(packed, inode); 151 } 152 153 static noinline int bch2_inode_unpack_v1(struct bkey_s_c_inode inode, 154 struct bch_inode_unpacked *unpacked) 155 { 156 const u8 *in = inode.v->fields; 157 const u8 *end = bkey_val_end(inode); 158 u64 field[2]; 159 unsigned fieldnr = 0, field_bits; 160 int ret; 161 162 #define x(_name, _bits) \ 163 if (fieldnr++ == INODE_NR_FIELDS(inode.v)) { \ 164 unsigned offset = offsetof(struct bch_inode_unpacked, _name);\ 165 memset((void *) unpacked + offset, 0, \ 166 sizeof(*unpacked) - offset); \ 167 return 0; \ 168 } \ 169 \ 170 ret = inode_decode_field(in, end, field, &field_bits); \ 171 if (ret < 0) \ 172 return ret; \ 173 \ 174 if (field_bits > sizeof(unpacked->_name) * 8) \ 175 return -1; \ 176 \ 177 unpacked->_name = field[1]; \ 178 in += ret; 179 180 BCH_INODE_FIELDS_v2() 181 #undef x 182 183 /* XXX: signal if there were more fields than expected? */ 184 return 0; 185 } 186 187 static int bch2_inode_unpack_v2(struct bch_inode_unpacked *unpacked, 188 const u8 *in, const u8 *end, 189 unsigned nr_fields) 190 { 191 unsigned fieldnr = 0; 192 int ret; 193 u64 v[2]; 194 195 #define x(_name, _bits) \ 196 if (fieldnr < nr_fields) { \ 197 ret = bch2_varint_decode_fast(in, end, &v[0]); \ 198 if (ret < 0) \ 199 return ret; \ 200 in += ret; \ 201 \ 202 if (_bits > 64) { \ 203 ret = bch2_varint_decode_fast(in, end, &v[1]); \ 204 if (ret < 0) \ 205 return ret; \ 206 in += ret; \ 207 } else { \ 208 v[1] = 0; \ 209 } \ 210 } else { \ 211 v[0] = v[1] = 0; \ 212 } \ 213 \ 214 unpacked->_name = v[0]; \ 215 if (v[1] || v[0] != unpacked->_name) \ 216 return -1; \ 217 fieldnr++; 218 219 BCH_INODE_FIELDS_v2() 220 #undef x 221 222 /* XXX: signal if there were more fields than expected? */ 223 return 0; 224 } 225 226 static int bch2_inode_unpack_v3(struct bkey_s_c k, 227 struct bch_inode_unpacked *unpacked) 228 { 229 struct bkey_s_c_inode_v3 inode = bkey_s_c_to_inode_v3(k); 230 const u8 *in = inode.v->fields; 231 const u8 *end = bkey_val_end(inode); 232 unsigned nr_fields = INODEv3_NR_FIELDS(inode.v); 233 unsigned fieldnr = 0; 234 int ret; 235 u64 v[2]; 236 237 unpacked->bi_inum = inode.k->p.offset; 238 unpacked->bi_journal_seq= le64_to_cpu(inode.v->bi_journal_seq); 239 unpacked->bi_hash_seed = inode.v->bi_hash_seed; 240 unpacked->bi_flags = le64_to_cpu(inode.v->bi_flags); 241 unpacked->bi_sectors = le64_to_cpu(inode.v->bi_sectors); 242 unpacked->bi_size = le64_to_cpu(inode.v->bi_size); 243 unpacked->bi_version = le64_to_cpu(inode.v->bi_version); 244 unpacked->bi_mode = INODEv3_MODE(inode.v); 245 246 #define x(_name, _bits) \ 247 if (fieldnr < nr_fields) { \ 248 ret = bch2_varint_decode_fast(in, end, &v[0]); \ 249 if (ret < 0) \ 250 return ret; \ 251 in += ret; \ 252 \ 253 if (_bits > 64) { \ 254 ret = bch2_varint_decode_fast(in, end, &v[1]); \ 255 if (ret < 0) \ 256 return ret; \ 257 in += ret; \ 258 } else { \ 259 v[1] = 0; \ 260 } \ 261 } else { \ 262 v[0] = v[1] = 0; \ 263 } \ 264 \ 265 unpacked->_name = v[0]; \ 266 if (v[1] || v[0] != unpacked->_name) \ 267 return -1; \ 268 fieldnr++; 269 270 BCH_INODE_FIELDS_v3() 271 #undef x 272 273 /* XXX: signal if there were more fields than expected? */ 274 return 0; 275 } 276 277 static noinline int bch2_inode_unpack_slowpath(struct bkey_s_c k, 278 struct bch_inode_unpacked *unpacked) 279 { 280 memset(unpacked, 0, sizeof(*unpacked)); 281 282 switch (k.k->type) { 283 case KEY_TYPE_inode: { 284 struct bkey_s_c_inode inode = bkey_s_c_to_inode(k); 285 286 unpacked->bi_inum = inode.k->p.offset; 287 unpacked->bi_journal_seq= 0; 288 unpacked->bi_hash_seed = inode.v->bi_hash_seed; 289 unpacked->bi_flags = le32_to_cpu(inode.v->bi_flags); 290 unpacked->bi_mode = le16_to_cpu(inode.v->bi_mode); 291 292 if (INODE_NEW_VARINT(inode.v)) { 293 return bch2_inode_unpack_v2(unpacked, inode.v->fields, 294 bkey_val_end(inode), 295 INODE_NR_FIELDS(inode.v)); 296 } else { 297 return bch2_inode_unpack_v1(inode, unpacked); 298 } 299 break; 300 } 301 case KEY_TYPE_inode_v2: { 302 struct bkey_s_c_inode_v2 inode = bkey_s_c_to_inode_v2(k); 303 304 unpacked->bi_inum = inode.k->p.offset; 305 unpacked->bi_journal_seq= le64_to_cpu(inode.v->bi_journal_seq); 306 unpacked->bi_hash_seed = inode.v->bi_hash_seed; 307 unpacked->bi_flags = le64_to_cpu(inode.v->bi_flags); 308 unpacked->bi_mode = le16_to_cpu(inode.v->bi_mode); 309 310 return bch2_inode_unpack_v2(unpacked, inode.v->fields, 311 bkey_val_end(inode), 312 INODEv2_NR_FIELDS(inode.v)); 313 } 314 default: 315 BUG(); 316 } 317 } 318 319 int bch2_inode_unpack(struct bkey_s_c k, 320 struct bch_inode_unpacked *unpacked) 321 { 322 if (likely(k.k->type == KEY_TYPE_inode_v3)) 323 return bch2_inode_unpack_v3(k, unpacked); 324 return bch2_inode_unpack_slowpath(k, unpacked); 325 } 326 327 int bch2_inode_peek_nowarn(struct btree_trans *trans, 328 struct btree_iter *iter, 329 struct bch_inode_unpacked *inode, 330 subvol_inum inum, unsigned flags) 331 { 332 struct bkey_s_c k; 333 u32 snapshot; 334 int ret; 335 336 ret = bch2_subvolume_get_snapshot(trans, inum.subvol, &snapshot); 337 if (ret) 338 return ret; 339 340 k = bch2_bkey_get_iter(trans, iter, BTREE_ID_inodes, 341 SPOS(0, inum.inum, snapshot), 342 flags|BTREE_ITER_cached); 343 ret = bkey_err(k); 344 if (ret) 345 return ret; 346 347 ret = bkey_is_inode(k.k) ? 0 : -BCH_ERR_ENOENT_inode; 348 if (ret) 349 goto err; 350 351 ret = bch2_inode_unpack(k, inode); 352 if (ret) 353 goto err; 354 355 return 0; 356 err: 357 bch2_trans_iter_exit(trans, iter); 358 return ret; 359 } 360 361 int bch2_inode_peek(struct btree_trans *trans, 362 struct btree_iter *iter, 363 struct bch_inode_unpacked *inode, 364 subvol_inum inum, unsigned flags) 365 { 366 int ret = bch2_inode_peek_nowarn(trans, iter, inode, inum, flags); 367 bch_err_msg(trans->c, ret, "looking up inum %u:%llu:", inum.subvol, inum.inum); 368 return ret; 369 } 370 371 int bch2_inode_write_flags(struct btree_trans *trans, 372 struct btree_iter *iter, 373 struct bch_inode_unpacked *inode, 374 enum btree_iter_update_trigger_flags flags) 375 { 376 struct bkey_inode_buf *inode_p; 377 378 inode_p = bch2_trans_kmalloc(trans, sizeof(*inode_p)); 379 if (IS_ERR(inode_p)) 380 return PTR_ERR(inode_p); 381 382 bch2_inode_pack_inlined(inode_p, inode); 383 inode_p->inode.k.p.snapshot = iter->snapshot; 384 return bch2_trans_update(trans, iter, &inode_p->inode.k_i, flags); 385 } 386 387 int __bch2_fsck_write_inode(struct btree_trans *trans, 388 struct bch_inode_unpacked *inode, 389 u32 snapshot) 390 { 391 struct bkey_inode_buf *inode_p = 392 bch2_trans_kmalloc(trans, sizeof(*inode_p)); 393 394 if (IS_ERR(inode_p)) 395 return PTR_ERR(inode_p); 396 397 bch2_inode_pack(inode_p, inode); 398 inode_p->inode.k.p.snapshot = snapshot; 399 400 return bch2_btree_insert_nonextent(trans, BTREE_ID_inodes, 401 &inode_p->inode.k_i, 402 BTREE_UPDATE_internal_snapshot_node); 403 } 404 405 int bch2_fsck_write_inode(struct btree_trans *trans, 406 struct bch_inode_unpacked *inode, 407 u32 snapshot) 408 { 409 int ret = commit_do(trans, NULL, NULL, BCH_TRANS_COMMIT_no_enospc, 410 __bch2_fsck_write_inode(trans, inode, snapshot)); 411 bch_err_fn(trans->c, ret); 412 return ret; 413 } 414 415 struct bkey_i *bch2_inode_to_v3(struct btree_trans *trans, struct bkey_i *k) 416 { 417 struct bch_inode_unpacked u; 418 struct bkey_inode_buf *inode_p; 419 int ret; 420 421 if (!bkey_is_inode(&k->k)) 422 return ERR_PTR(-ENOENT); 423 424 inode_p = bch2_trans_kmalloc(trans, sizeof(*inode_p)); 425 if (IS_ERR(inode_p)) 426 return ERR_CAST(inode_p); 427 428 ret = bch2_inode_unpack(bkey_i_to_s_c(k), &u); 429 if (ret) 430 return ERR_PTR(ret); 431 432 bch2_inode_pack(inode_p, &u); 433 return &inode_p->inode.k_i; 434 } 435 436 static int __bch2_inode_invalid(struct bch_fs *c, struct bkey_s_c k, struct printbuf *err) 437 { 438 struct bch_inode_unpacked unpacked; 439 int ret = 0; 440 441 bkey_fsck_err_on(k.k->p.inode, c, err, 442 inode_pos_inode_nonzero, 443 "nonzero k.p.inode"); 444 445 bkey_fsck_err_on(k.k->p.offset < BLOCKDEV_INODE_MAX, c, err, 446 inode_pos_blockdev_range, 447 "fs inode in blockdev range"); 448 449 bkey_fsck_err_on(bch2_inode_unpack(k, &unpacked), c, err, 450 inode_unpack_error, 451 "invalid variable length fields"); 452 453 bkey_fsck_err_on(unpacked.bi_data_checksum >= BCH_CSUM_OPT_NR + 1, c, err, 454 inode_checksum_type_invalid, 455 "invalid data checksum type (%u >= %u", 456 unpacked.bi_data_checksum, BCH_CSUM_OPT_NR + 1); 457 458 bkey_fsck_err_on(unpacked.bi_compression && 459 !bch2_compression_opt_valid(unpacked.bi_compression - 1), c, err, 460 inode_compression_type_invalid, 461 "invalid compression opt %u", unpacked.bi_compression - 1); 462 463 bkey_fsck_err_on((unpacked.bi_flags & BCH_INODE_unlinked) && 464 unpacked.bi_nlink != 0, c, err, 465 inode_unlinked_but_nlink_nonzero, 466 "flagged as unlinked but bi_nlink != 0"); 467 468 bkey_fsck_err_on(unpacked.bi_subvol && !S_ISDIR(unpacked.bi_mode), c, err, 469 inode_subvol_root_but_not_dir, 470 "subvolume root but not a directory"); 471 fsck_err: 472 return ret; 473 } 474 475 int bch2_inode_invalid(struct bch_fs *c, struct bkey_s_c k, 476 enum bch_validate_flags flags, 477 struct printbuf *err) 478 { 479 struct bkey_s_c_inode inode = bkey_s_c_to_inode(k); 480 int ret = 0; 481 482 bkey_fsck_err_on(INODE_STR_HASH(inode.v) >= BCH_STR_HASH_NR, c, err, 483 inode_str_hash_invalid, 484 "invalid str hash type (%llu >= %u)", 485 INODE_STR_HASH(inode.v), BCH_STR_HASH_NR); 486 487 ret = __bch2_inode_invalid(c, k, err); 488 fsck_err: 489 return ret; 490 } 491 492 int bch2_inode_v2_invalid(struct bch_fs *c, struct bkey_s_c k, 493 enum bch_validate_flags flags, 494 struct printbuf *err) 495 { 496 struct bkey_s_c_inode_v2 inode = bkey_s_c_to_inode_v2(k); 497 int ret = 0; 498 499 bkey_fsck_err_on(INODEv2_STR_HASH(inode.v) >= BCH_STR_HASH_NR, c, err, 500 inode_str_hash_invalid, 501 "invalid str hash type (%llu >= %u)", 502 INODEv2_STR_HASH(inode.v), BCH_STR_HASH_NR); 503 504 ret = __bch2_inode_invalid(c, k, err); 505 fsck_err: 506 return ret; 507 } 508 509 int bch2_inode_v3_invalid(struct bch_fs *c, struct bkey_s_c k, 510 enum bch_validate_flags flags, 511 struct printbuf *err) 512 { 513 struct bkey_s_c_inode_v3 inode = bkey_s_c_to_inode_v3(k); 514 int ret = 0; 515 516 bkey_fsck_err_on(INODEv3_FIELDS_START(inode.v) < INODEv3_FIELDS_START_INITIAL || 517 INODEv3_FIELDS_START(inode.v) > bkey_val_u64s(inode.k), c, err, 518 inode_v3_fields_start_bad, 519 "invalid fields_start (got %llu, min %u max %zu)", 520 INODEv3_FIELDS_START(inode.v), 521 INODEv3_FIELDS_START_INITIAL, 522 bkey_val_u64s(inode.k)); 523 524 bkey_fsck_err_on(INODEv3_STR_HASH(inode.v) >= BCH_STR_HASH_NR, c, err, 525 inode_str_hash_invalid, 526 "invalid str hash type (%llu >= %u)", 527 INODEv3_STR_HASH(inode.v), BCH_STR_HASH_NR); 528 529 ret = __bch2_inode_invalid(c, k, err); 530 fsck_err: 531 return ret; 532 } 533 534 static void __bch2_inode_unpacked_to_text(struct printbuf *out, 535 struct bch_inode_unpacked *inode) 536 { 537 printbuf_indent_add(out, 2); 538 prt_printf(out, "mode=%o\n", inode->bi_mode); 539 540 prt_str(out, "flags="); 541 prt_bitflags(out, bch2_inode_flag_strs, inode->bi_flags & ((1U << 20) - 1)); 542 prt_printf(out, " (%x)\n", inode->bi_flags); 543 544 prt_printf(out, "journal_seq=%llu\n", inode->bi_journal_seq); 545 prt_printf(out, "bi_size=%llu\n", inode->bi_size); 546 prt_printf(out, "bi_sectors=%llu\n", inode->bi_sectors); 547 prt_printf(out, "bi_version=%llu\n", inode->bi_version); 548 549 #define x(_name, _bits) \ 550 prt_printf(out, #_name "=%llu\n", (u64) inode->_name); 551 BCH_INODE_FIELDS_v3() 552 #undef x 553 printbuf_indent_sub(out, 2); 554 } 555 556 void bch2_inode_unpacked_to_text(struct printbuf *out, struct bch_inode_unpacked *inode) 557 { 558 prt_printf(out, "inum: %llu ", inode->bi_inum); 559 __bch2_inode_unpacked_to_text(out, inode); 560 } 561 562 void bch2_inode_to_text(struct printbuf *out, struct bch_fs *c, struct bkey_s_c k) 563 { 564 struct bch_inode_unpacked inode; 565 566 if (bch2_inode_unpack(k, &inode)) { 567 prt_printf(out, "(unpack error)"); 568 return; 569 } 570 571 __bch2_inode_unpacked_to_text(out, &inode); 572 } 573 574 static inline u64 bkey_inode_flags(struct bkey_s_c k) 575 { 576 switch (k.k->type) { 577 case KEY_TYPE_inode: 578 return le32_to_cpu(bkey_s_c_to_inode(k).v->bi_flags); 579 case KEY_TYPE_inode_v2: 580 return le64_to_cpu(bkey_s_c_to_inode_v2(k).v->bi_flags); 581 case KEY_TYPE_inode_v3: 582 return le64_to_cpu(bkey_s_c_to_inode_v3(k).v->bi_flags); 583 default: 584 return 0; 585 } 586 } 587 588 static inline bool bkey_is_deleted_inode(struct bkey_s_c k) 589 { 590 return bkey_inode_flags(k) & BCH_INODE_unlinked; 591 } 592 593 int bch2_trigger_inode(struct btree_trans *trans, 594 enum btree_id btree_id, unsigned level, 595 struct bkey_s_c old, 596 struct bkey_s new, 597 enum btree_iter_update_trigger_flags flags) 598 { 599 s64 nr = (s64) bkey_is_inode(new.k) - (s64) bkey_is_inode(old.k); 600 601 if (flags & BTREE_TRIGGER_transactional) { 602 if (nr) { 603 int ret = bch2_replicas_deltas_realloc(trans, 0); 604 if (ret) 605 return ret; 606 607 trans->fs_usage_deltas->nr_inodes += nr; 608 } 609 610 bool old_deleted = bkey_is_deleted_inode(old); 611 bool new_deleted = bkey_is_deleted_inode(new.s_c); 612 if (old_deleted != new_deleted) { 613 int ret = bch2_btree_bit_mod_buffered(trans, BTREE_ID_deleted_inodes, 614 new.k->p, new_deleted); 615 if (ret) 616 return ret; 617 } 618 } 619 620 if ((flags & BTREE_TRIGGER_atomic) && (flags & BTREE_TRIGGER_insert)) { 621 BUG_ON(!trans->journal_res.seq); 622 623 bkey_s_to_inode_v3(new).v->bi_journal_seq = cpu_to_le64(trans->journal_res.seq); 624 } 625 626 if (flags & BTREE_TRIGGER_gc) { 627 struct bch_fs *c = trans->c; 628 629 percpu_down_read(&c->mark_lock); 630 this_cpu_add(c->usage_gc->b.nr_inodes, nr); 631 percpu_up_read(&c->mark_lock); 632 } 633 634 return 0; 635 } 636 637 int bch2_inode_generation_invalid(struct bch_fs *c, struct bkey_s_c k, 638 enum bch_validate_flags flags, 639 struct printbuf *err) 640 { 641 int ret = 0; 642 643 bkey_fsck_err_on(k.k->p.inode, c, err, 644 inode_pos_inode_nonzero, 645 "nonzero k.p.inode"); 646 fsck_err: 647 return ret; 648 } 649 650 void bch2_inode_generation_to_text(struct printbuf *out, struct bch_fs *c, 651 struct bkey_s_c k) 652 { 653 struct bkey_s_c_inode_generation gen = bkey_s_c_to_inode_generation(k); 654 655 prt_printf(out, "generation: %u", le32_to_cpu(gen.v->bi_generation)); 656 } 657 658 void bch2_inode_init_early(struct bch_fs *c, 659 struct bch_inode_unpacked *inode_u) 660 { 661 enum bch_str_hash_type str_hash = 662 bch2_str_hash_opt_to_type(c, c->opts.str_hash); 663 664 memset(inode_u, 0, sizeof(*inode_u)); 665 666 /* ick */ 667 inode_u->bi_flags |= str_hash << INODE_STR_HASH_OFFSET; 668 get_random_bytes(&inode_u->bi_hash_seed, 669 sizeof(inode_u->bi_hash_seed)); 670 } 671 672 void bch2_inode_init_late(struct bch_inode_unpacked *inode_u, u64 now, 673 uid_t uid, gid_t gid, umode_t mode, dev_t rdev, 674 struct bch_inode_unpacked *parent) 675 { 676 inode_u->bi_mode = mode; 677 inode_u->bi_uid = uid; 678 inode_u->bi_gid = gid; 679 inode_u->bi_dev = rdev; 680 inode_u->bi_atime = now; 681 inode_u->bi_mtime = now; 682 inode_u->bi_ctime = now; 683 inode_u->bi_otime = now; 684 685 if (parent && parent->bi_mode & S_ISGID) { 686 inode_u->bi_gid = parent->bi_gid; 687 if (S_ISDIR(mode)) 688 inode_u->bi_mode |= S_ISGID; 689 } 690 691 if (parent) { 692 #define x(_name, ...) inode_u->bi_##_name = parent->bi_##_name; 693 BCH_INODE_OPTS() 694 #undef x 695 } 696 } 697 698 void bch2_inode_init(struct bch_fs *c, struct bch_inode_unpacked *inode_u, 699 uid_t uid, gid_t gid, umode_t mode, dev_t rdev, 700 struct bch_inode_unpacked *parent) 701 { 702 bch2_inode_init_early(c, inode_u); 703 bch2_inode_init_late(inode_u, bch2_current_time(c), 704 uid, gid, mode, rdev, parent); 705 } 706 707 static inline u32 bkey_generation(struct bkey_s_c k) 708 { 709 switch (k.k->type) { 710 case KEY_TYPE_inode: 711 case KEY_TYPE_inode_v2: 712 BUG(); 713 case KEY_TYPE_inode_generation: 714 return le32_to_cpu(bkey_s_c_to_inode_generation(k).v->bi_generation); 715 default: 716 return 0; 717 } 718 } 719 720 /* 721 * This just finds an empty slot: 722 */ 723 int bch2_inode_create(struct btree_trans *trans, 724 struct btree_iter *iter, 725 struct bch_inode_unpacked *inode_u, 726 u32 snapshot, u64 cpu) 727 { 728 struct bch_fs *c = trans->c; 729 struct bkey_s_c k; 730 u64 min, max, start, pos, *hint; 731 int ret = 0; 732 unsigned bits = (c->opts.inodes_32bit ? 31 : 63); 733 734 if (c->opts.shard_inode_numbers) { 735 bits -= c->inode_shard_bits; 736 737 min = (cpu << bits); 738 max = (cpu << bits) | ~(ULLONG_MAX << bits); 739 740 min = max_t(u64, min, BLOCKDEV_INODE_MAX); 741 hint = c->unused_inode_hints + cpu; 742 } else { 743 min = BLOCKDEV_INODE_MAX; 744 max = ~(ULLONG_MAX << bits); 745 hint = c->unused_inode_hints; 746 } 747 748 start = READ_ONCE(*hint); 749 750 if (start >= max || start < min) 751 start = min; 752 753 pos = start; 754 bch2_trans_iter_init(trans, iter, BTREE_ID_inodes, POS(0, pos), 755 BTREE_ITER_all_snapshots| 756 BTREE_ITER_intent); 757 again: 758 while ((k = bch2_btree_iter_peek(iter)).k && 759 !(ret = bkey_err(k)) && 760 bkey_lt(k.k->p, POS(0, max))) { 761 if (pos < iter->pos.offset) 762 goto found_slot; 763 764 /* 765 * We don't need to iterate over keys in every snapshot once 766 * we've found just one: 767 */ 768 pos = iter->pos.offset + 1; 769 bch2_btree_iter_set_pos(iter, POS(0, pos)); 770 } 771 772 if (!ret && pos < max) 773 goto found_slot; 774 775 if (!ret && start == min) 776 ret = -BCH_ERR_ENOSPC_inode_create; 777 778 if (ret) { 779 bch2_trans_iter_exit(trans, iter); 780 return ret; 781 } 782 783 /* Retry from start */ 784 pos = start = min; 785 bch2_btree_iter_set_pos(iter, POS(0, pos)); 786 goto again; 787 found_slot: 788 bch2_btree_iter_set_pos(iter, SPOS(0, pos, snapshot)); 789 k = bch2_btree_iter_peek_slot(iter); 790 ret = bkey_err(k); 791 if (ret) { 792 bch2_trans_iter_exit(trans, iter); 793 return ret; 794 } 795 796 *hint = k.k->p.offset; 797 inode_u->bi_inum = k.k->p.offset; 798 inode_u->bi_generation = bkey_generation(k); 799 return 0; 800 } 801 802 static int bch2_inode_delete_keys(struct btree_trans *trans, 803 subvol_inum inum, enum btree_id id) 804 { 805 struct btree_iter iter; 806 struct bkey_s_c k; 807 struct bkey_i delete; 808 struct bpos end = POS(inum.inum, U64_MAX); 809 u32 snapshot; 810 int ret = 0; 811 812 /* 813 * We're never going to be deleting partial extents, no need to use an 814 * extent iterator: 815 */ 816 bch2_trans_iter_init(trans, &iter, id, POS(inum.inum, 0), 817 BTREE_ITER_intent); 818 819 while (1) { 820 bch2_trans_begin(trans); 821 822 ret = bch2_subvolume_get_snapshot(trans, inum.subvol, &snapshot); 823 if (ret) 824 goto err; 825 826 bch2_btree_iter_set_snapshot(&iter, snapshot); 827 828 k = bch2_btree_iter_peek_upto(&iter, end); 829 ret = bkey_err(k); 830 if (ret) 831 goto err; 832 833 if (!k.k) 834 break; 835 836 bkey_init(&delete.k); 837 delete.k.p = iter.pos; 838 839 if (iter.flags & BTREE_ITER_is_extents) 840 bch2_key_resize(&delete.k, 841 bpos_min(end, k.k->p).offset - 842 iter.pos.offset); 843 844 ret = bch2_trans_update(trans, &iter, &delete, 0) ?: 845 bch2_trans_commit(trans, NULL, NULL, 846 BCH_TRANS_COMMIT_no_enospc); 847 err: 848 if (ret && !bch2_err_matches(ret, BCH_ERR_transaction_restart)) 849 break; 850 } 851 852 bch2_trans_iter_exit(trans, &iter); 853 return ret; 854 } 855 856 int bch2_inode_rm(struct bch_fs *c, subvol_inum inum) 857 { 858 struct btree_trans *trans = bch2_trans_get(c); 859 struct btree_iter iter = { NULL }; 860 struct bkey_i_inode_generation delete; 861 struct bch_inode_unpacked inode_u; 862 struct bkey_s_c k; 863 u32 snapshot; 864 int ret; 865 866 /* 867 * If this was a directory, there shouldn't be any real dirents left - 868 * but there could be whiteouts (from hash collisions) that we should 869 * delete: 870 * 871 * XXX: the dirent could ideally would delete whiteouts when they're no 872 * longer needed 873 */ 874 ret = bch2_inode_delete_keys(trans, inum, BTREE_ID_extents) ?: 875 bch2_inode_delete_keys(trans, inum, BTREE_ID_xattrs) ?: 876 bch2_inode_delete_keys(trans, inum, BTREE_ID_dirents); 877 if (ret) 878 goto err; 879 retry: 880 bch2_trans_begin(trans); 881 882 ret = bch2_subvolume_get_snapshot(trans, inum.subvol, &snapshot); 883 if (ret) 884 goto err; 885 886 k = bch2_bkey_get_iter(trans, &iter, BTREE_ID_inodes, 887 SPOS(0, inum.inum, snapshot), 888 BTREE_ITER_intent|BTREE_ITER_cached); 889 ret = bkey_err(k); 890 if (ret) 891 goto err; 892 893 if (!bkey_is_inode(k.k)) { 894 bch2_fs_inconsistent(c, 895 "inode %llu:%u not found when deleting", 896 inum.inum, snapshot); 897 ret = -EIO; 898 goto err; 899 } 900 901 bch2_inode_unpack(k, &inode_u); 902 903 bkey_inode_generation_init(&delete.k_i); 904 delete.k.p = iter.pos; 905 delete.v.bi_generation = cpu_to_le32(inode_u.bi_generation + 1); 906 907 ret = bch2_trans_update(trans, &iter, &delete.k_i, 0) ?: 908 bch2_trans_commit(trans, NULL, NULL, 909 BCH_TRANS_COMMIT_no_enospc); 910 err: 911 bch2_trans_iter_exit(trans, &iter); 912 if (bch2_err_matches(ret, BCH_ERR_transaction_restart)) 913 goto retry; 914 915 bch2_trans_put(trans); 916 return ret; 917 } 918 919 int bch2_inode_find_by_inum_nowarn_trans(struct btree_trans *trans, 920 subvol_inum inum, 921 struct bch_inode_unpacked *inode) 922 { 923 struct btree_iter iter; 924 int ret; 925 926 ret = bch2_inode_peek_nowarn(trans, &iter, inode, inum, 0); 927 if (!ret) 928 bch2_trans_iter_exit(trans, &iter); 929 return ret; 930 } 931 932 int bch2_inode_find_by_inum_trans(struct btree_trans *trans, 933 subvol_inum inum, 934 struct bch_inode_unpacked *inode) 935 { 936 struct btree_iter iter; 937 int ret; 938 939 ret = bch2_inode_peek(trans, &iter, inode, inum, 0); 940 if (!ret) 941 bch2_trans_iter_exit(trans, &iter); 942 return ret; 943 } 944 945 int bch2_inode_find_by_inum(struct bch_fs *c, subvol_inum inum, 946 struct bch_inode_unpacked *inode) 947 { 948 return bch2_trans_do(c, NULL, NULL, 0, 949 bch2_inode_find_by_inum_trans(trans, inum, inode)); 950 } 951 952 int bch2_inode_nlink_inc(struct bch_inode_unpacked *bi) 953 { 954 if (bi->bi_flags & BCH_INODE_unlinked) 955 bi->bi_flags &= ~BCH_INODE_unlinked; 956 else { 957 if (bi->bi_nlink == U32_MAX) 958 return -EINVAL; 959 960 bi->bi_nlink++; 961 } 962 963 return 0; 964 } 965 966 void bch2_inode_nlink_dec(struct btree_trans *trans, struct bch_inode_unpacked *bi) 967 { 968 if (bi->bi_nlink && (bi->bi_flags & BCH_INODE_unlinked)) { 969 bch2_trans_inconsistent(trans, "inode %llu unlinked but link count nonzero", 970 bi->bi_inum); 971 return; 972 } 973 974 if (bi->bi_flags & BCH_INODE_unlinked) { 975 bch2_trans_inconsistent(trans, "inode %llu link count underflow", bi->bi_inum); 976 return; 977 } 978 979 if (bi->bi_nlink) 980 bi->bi_nlink--; 981 else 982 bi->bi_flags |= BCH_INODE_unlinked; 983 } 984 985 struct bch_opts bch2_inode_opts_to_opts(struct bch_inode_unpacked *inode) 986 { 987 struct bch_opts ret = { 0 }; 988 #define x(_name, _bits) \ 989 if (inode->bi_##_name) \ 990 opt_set(ret, _name, inode->bi_##_name - 1); 991 BCH_INODE_OPTS() 992 #undef x 993 return ret; 994 } 995 996 void bch2_inode_opts_get(struct bch_io_opts *opts, struct bch_fs *c, 997 struct bch_inode_unpacked *inode) 998 { 999 #define x(_name, _bits) opts->_name = inode_opt_get(c, inode, _name); 1000 BCH_INODE_OPTS() 1001 #undef x 1002 1003 if (opts->nocow) 1004 opts->compression = opts->background_compression = opts->data_checksum = opts->erasure_code = 0; 1005 } 1006 1007 int bch2_inum_opts_get(struct btree_trans *trans, subvol_inum inum, struct bch_io_opts *opts) 1008 { 1009 struct bch_inode_unpacked inode; 1010 int ret = lockrestart_do(trans, bch2_inode_find_by_inum_trans(trans, inum, &inode)); 1011 1012 if (ret) 1013 return ret; 1014 1015 bch2_inode_opts_get(opts, trans->c, &inode); 1016 return 0; 1017 } 1018 1019 int bch2_inode_rm_snapshot(struct btree_trans *trans, u64 inum, u32 snapshot) 1020 { 1021 struct bch_fs *c = trans->c; 1022 struct btree_iter iter = { NULL }; 1023 struct bkey_i_inode_generation delete; 1024 struct bch_inode_unpacked inode_u; 1025 struct bkey_s_c k; 1026 int ret; 1027 1028 do { 1029 ret = bch2_btree_delete_range_trans(trans, BTREE_ID_extents, 1030 SPOS(inum, 0, snapshot), 1031 SPOS(inum, U64_MAX, snapshot), 1032 0, NULL) ?: 1033 bch2_btree_delete_range_trans(trans, BTREE_ID_dirents, 1034 SPOS(inum, 0, snapshot), 1035 SPOS(inum, U64_MAX, snapshot), 1036 0, NULL) ?: 1037 bch2_btree_delete_range_trans(trans, BTREE_ID_xattrs, 1038 SPOS(inum, 0, snapshot), 1039 SPOS(inum, U64_MAX, snapshot), 1040 0, NULL); 1041 } while (ret == -BCH_ERR_transaction_restart_nested); 1042 if (ret) 1043 goto err; 1044 retry: 1045 bch2_trans_begin(trans); 1046 1047 k = bch2_bkey_get_iter(trans, &iter, BTREE_ID_inodes, 1048 SPOS(0, inum, snapshot), BTREE_ITER_intent); 1049 ret = bkey_err(k); 1050 if (ret) 1051 goto err; 1052 1053 if (!bkey_is_inode(k.k)) { 1054 bch2_fs_inconsistent(c, 1055 "inode %llu:%u not found when deleting", 1056 inum, snapshot); 1057 ret = -EIO; 1058 goto err; 1059 } 1060 1061 bch2_inode_unpack(k, &inode_u); 1062 1063 /* Subvolume root? */ 1064 if (inode_u.bi_subvol) 1065 bch_warn(c, "deleting inode %llu marked as unlinked, but also a subvolume root!?", inode_u.bi_inum); 1066 1067 bkey_inode_generation_init(&delete.k_i); 1068 delete.k.p = iter.pos; 1069 delete.v.bi_generation = cpu_to_le32(inode_u.bi_generation + 1); 1070 1071 ret = bch2_trans_update(trans, &iter, &delete.k_i, 0) ?: 1072 bch2_trans_commit(trans, NULL, NULL, 1073 BCH_TRANS_COMMIT_no_enospc); 1074 err: 1075 bch2_trans_iter_exit(trans, &iter); 1076 if (bch2_err_matches(ret, BCH_ERR_transaction_restart)) 1077 goto retry; 1078 1079 return ret ?: -BCH_ERR_transaction_restart_nested; 1080 } 1081 1082 static int may_delete_deleted_inode(struct btree_trans *trans, 1083 struct btree_iter *iter, 1084 struct bpos pos, 1085 bool *need_another_pass) 1086 { 1087 struct bch_fs *c = trans->c; 1088 struct btree_iter inode_iter; 1089 struct bkey_s_c k; 1090 struct bch_inode_unpacked inode; 1091 int ret; 1092 1093 k = bch2_bkey_get_iter(trans, &inode_iter, BTREE_ID_inodes, pos, BTREE_ITER_cached); 1094 ret = bkey_err(k); 1095 if (ret) 1096 return ret; 1097 1098 ret = bkey_is_inode(k.k) ? 0 : -BCH_ERR_ENOENT_inode; 1099 if (fsck_err_on(!bkey_is_inode(k.k), c, 1100 deleted_inode_missing, 1101 "nonexistent inode %llu:%u in deleted_inodes btree", 1102 pos.offset, pos.snapshot)) 1103 goto delete; 1104 1105 ret = bch2_inode_unpack(k, &inode); 1106 if (ret) 1107 goto out; 1108 1109 if (S_ISDIR(inode.bi_mode)) { 1110 ret = bch2_empty_dir_snapshot(trans, pos.offset, 0, pos.snapshot); 1111 if (fsck_err_on(bch2_err_matches(ret, ENOTEMPTY), 1112 c, deleted_inode_is_dir, 1113 "non empty directory %llu:%u in deleted_inodes btree", 1114 pos.offset, pos.snapshot)) 1115 goto delete; 1116 if (ret) 1117 goto out; 1118 } 1119 1120 if (fsck_err_on(!(inode.bi_flags & BCH_INODE_unlinked), c, 1121 deleted_inode_not_unlinked, 1122 "non-deleted inode %llu:%u in deleted_inodes btree", 1123 pos.offset, pos.snapshot)) 1124 goto delete; 1125 1126 if (c->sb.clean && 1127 !fsck_err(c, 1128 deleted_inode_but_clean, 1129 "filesystem marked as clean but have deleted inode %llu:%u", 1130 pos.offset, pos.snapshot)) { 1131 ret = 0; 1132 goto out; 1133 } 1134 1135 if (bch2_snapshot_is_internal_node(c, pos.snapshot)) { 1136 struct bpos new_min_pos; 1137 1138 ret = bch2_propagate_key_to_snapshot_leaves(trans, inode_iter.btree_id, k, &new_min_pos); 1139 if (ret) 1140 goto out; 1141 1142 inode.bi_flags &= ~BCH_INODE_unlinked; 1143 1144 ret = bch2_inode_write_flags(trans, &inode_iter, &inode, 1145 BTREE_UPDATE_internal_snapshot_node); 1146 bch_err_msg(c, ret, "clearing inode unlinked flag"); 1147 if (ret) 1148 goto out; 1149 1150 /* 1151 * We'll need another write buffer flush to pick up the new 1152 * unlinked inodes in the snapshot leaves: 1153 */ 1154 *need_another_pass = true; 1155 goto out; 1156 } 1157 1158 ret = 1; 1159 out: 1160 fsck_err: 1161 bch2_trans_iter_exit(trans, &inode_iter); 1162 return ret; 1163 delete: 1164 ret = bch2_btree_bit_mod_buffered(trans, BTREE_ID_deleted_inodes, pos, false); 1165 goto out; 1166 } 1167 1168 int bch2_delete_dead_inodes(struct bch_fs *c) 1169 { 1170 struct btree_trans *trans = bch2_trans_get(c); 1171 bool need_another_pass; 1172 int ret; 1173 again: 1174 /* 1175 * if we ran check_inodes() unlinked inodes will have already been 1176 * cleaned up but the write buffer will be out of sync; therefore we 1177 * alway need a write buffer flush 1178 */ 1179 ret = bch2_btree_write_buffer_flush_sync(trans); 1180 if (ret) 1181 goto err; 1182 1183 need_another_pass = false; 1184 1185 /* 1186 * Weird transaction restart handling here because on successful delete, 1187 * bch2_inode_rm_snapshot() will return a nested transaction restart, 1188 * but we can't retry because the btree write buffer won't have been 1189 * flushed and we'd spin: 1190 */ 1191 ret = for_each_btree_key_commit(trans, iter, BTREE_ID_deleted_inodes, POS_MIN, 1192 BTREE_ITER_prefetch|BTREE_ITER_all_snapshots, k, 1193 NULL, NULL, BCH_TRANS_COMMIT_no_enospc, ({ 1194 ret = may_delete_deleted_inode(trans, &iter, k.k->p, &need_another_pass); 1195 if (ret > 0) { 1196 bch_verbose(c, "deleting unlinked inode %llu:%u", k.k->p.offset, k.k->p.snapshot); 1197 1198 ret = bch2_inode_rm_snapshot(trans, k.k->p.offset, k.k->p.snapshot); 1199 /* 1200 * We don't want to loop here: a transaction restart 1201 * error here means we handled a transaction restart and 1202 * we're actually done, but if we loop we'll retry the 1203 * same key because the write buffer hasn't been flushed 1204 * yet 1205 */ 1206 if (bch2_err_matches(ret, BCH_ERR_transaction_restart)) { 1207 ret = 0; 1208 continue; 1209 } 1210 } 1211 1212 ret; 1213 })); 1214 1215 if (!ret && need_another_pass) 1216 goto again; 1217 err: 1218 bch2_trans_put(trans); 1219 return ret; 1220 } 1221