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 "disk_accounting.h" 12 #include "error.h" 13 #include "extents.h" 14 #include "extent_update.h" 15 #include "inode.h" 16 #include "str_hash.h" 17 #include "snapshot.h" 18 #include "subvolume.h" 19 #include "varint.h" 20 21 #include <linux/random.h> 22 23 #include <asm/unaligned.h> 24 25 #define x(name, ...) #name, 26 const char * const bch2_inode_opts[] = { 27 BCH_INODE_OPTS() 28 NULL, 29 }; 30 31 static const char * const bch2_inode_flag_strs[] = { 32 BCH_INODE_FLAGS() 33 NULL 34 }; 35 #undef x 36 37 static const u8 byte_table[8] = { 1, 2, 3, 4, 6, 8, 10, 13 }; 38 39 static int inode_decode_field(const u8 *in, const u8 *end, 40 u64 out[2], unsigned *out_bits) 41 { 42 __be64 be[2] = { 0, 0 }; 43 unsigned bytes, shift; 44 u8 *p; 45 46 if (in >= end) 47 return -1; 48 49 if (!*in) 50 return -1; 51 52 /* 53 * position of highest set bit indicates number of bytes: 54 * shift = number of bits to remove in high byte: 55 */ 56 shift = 8 - __fls(*in); /* 1 <= shift <= 8 */ 57 bytes = byte_table[shift - 1]; 58 59 if (in + bytes > end) 60 return -1; 61 62 p = (u8 *) be + 16 - bytes; 63 memcpy(p, in, bytes); 64 *p ^= (1 << 8) >> shift; 65 66 out[0] = be64_to_cpu(be[0]); 67 out[1] = be64_to_cpu(be[1]); 68 *out_bits = out[0] ? 64 + fls64(out[0]) : fls64(out[1]); 69 70 return bytes; 71 } 72 73 static inline void bch2_inode_pack_inlined(struct bkey_inode_buf *packed, 74 const struct bch_inode_unpacked *inode) 75 { 76 struct bkey_i_inode_v3 *k = &packed->inode; 77 u8 *out = k->v.fields; 78 u8 *end = (void *) &packed[1]; 79 u8 *last_nonzero_field = out; 80 unsigned nr_fields = 0, last_nonzero_fieldnr = 0; 81 unsigned bytes; 82 int ret; 83 84 bkey_inode_v3_init(&packed->inode.k_i); 85 packed->inode.k.p.offset = inode->bi_inum; 86 packed->inode.v.bi_journal_seq = cpu_to_le64(inode->bi_journal_seq); 87 packed->inode.v.bi_hash_seed = inode->bi_hash_seed; 88 packed->inode.v.bi_flags = cpu_to_le64(inode->bi_flags); 89 packed->inode.v.bi_sectors = cpu_to_le64(inode->bi_sectors); 90 packed->inode.v.bi_size = cpu_to_le64(inode->bi_size); 91 packed->inode.v.bi_version = cpu_to_le64(inode->bi_version); 92 SET_INODEv3_MODE(&packed->inode.v, inode->bi_mode); 93 SET_INODEv3_FIELDS_START(&packed->inode.v, INODEv3_FIELDS_START_CUR); 94 95 96 #define x(_name, _bits) \ 97 nr_fields++; \ 98 \ 99 if (inode->_name) { \ 100 ret = bch2_varint_encode_fast(out, inode->_name); \ 101 out += ret; \ 102 \ 103 if (_bits > 64) \ 104 *out++ = 0; \ 105 \ 106 last_nonzero_field = out; \ 107 last_nonzero_fieldnr = nr_fields; \ 108 } else { \ 109 *out++ = 0; \ 110 \ 111 if (_bits > 64) \ 112 *out++ = 0; \ 113 } 114 115 BCH_INODE_FIELDS_v3() 116 #undef x 117 BUG_ON(out > end); 118 119 out = last_nonzero_field; 120 nr_fields = last_nonzero_fieldnr; 121 122 bytes = out - (u8 *) &packed->inode.v; 123 set_bkey_val_bytes(&packed->inode.k, bytes); 124 memset_u64s_tail(&packed->inode.v, 0, bytes); 125 126 SET_INODEv3_NR_FIELDS(&k->v, nr_fields); 127 128 if (IS_ENABLED(CONFIG_BCACHEFS_DEBUG)) { 129 struct bch_inode_unpacked unpacked; 130 131 ret = bch2_inode_unpack(bkey_i_to_s_c(&packed->inode.k_i), &unpacked); 132 BUG_ON(ret); 133 BUG_ON(unpacked.bi_inum != inode->bi_inum); 134 BUG_ON(unpacked.bi_hash_seed != inode->bi_hash_seed); 135 BUG_ON(unpacked.bi_sectors != inode->bi_sectors); 136 BUG_ON(unpacked.bi_size != inode->bi_size); 137 BUG_ON(unpacked.bi_version != inode->bi_version); 138 BUG_ON(unpacked.bi_mode != inode->bi_mode); 139 140 #define x(_name, _bits) if (unpacked._name != inode->_name) \ 141 panic("unpacked %llu should be %llu", \ 142 (u64) unpacked._name, (u64) inode->_name); 143 BCH_INODE_FIELDS_v3() 144 #undef x 145 } 146 } 147 148 void bch2_inode_pack(struct bkey_inode_buf *packed, 149 const struct bch_inode_unpacked *inode) 150 { 151 bch2_inode_pack_inlined(packed, inode); 152 } 153 154 static noinline int bch2_inode_unpack_v1(struct bkey_s_c_inode inode, 155 struct bch_inode_unpacked *unpacked) 156 { 157 const u8 *in = inode.v->fields; 158 const u8 *end = bkey_val_end(inode); 159 u64 field[2]; 160 unsigned fieldnr = 0, field_bits; 161 int ret; 162 163 #define x(_name, _bits) \ 164 if (fieldnr++ == INODE_NR_FIELDS(inode.v)) { \ 165 unsigned offset = offsetof(struct bch_inode_unpacked, _name);\ 166 memset((void *) unpacked + offset, 0, \ 167 sizeof(*unpacked) - offset); \ 168 return 0; \ 169 } \ 170 \ 171 ret = inode_decode_field(in, end, field, &field_bits); \ 172 if (ret < 0) \ 173 return ret; \ 174 \ 175 if (field_bits > sizeof(unpacked->_name) * 8) \ 176 return -1; \ 177 \ 178 unpacked->_name = field[1]; \ 179 in += ret; 180 181 BCH_INODE_FIELDS_v2() 182 #undef x 183 184 /* XXX: signal if there were more fields than expected? */ 185 return 0; 186 } 187 188 static int bch2_inode_unpack_v2(struct bch_inode_unpacked *unpacked, 189 const u8 *in, const u8 *end, 190 unsigned nr_fields) 191 { 192 unsigned fieldnr = 0; 193 int ret; 194 u64 v[2]; 195 196 #define x(_name, _bits) \ 197 if (fieldnr < nr_fields) { \ 198 ret = bch2_varint_decode_fast(in, end, &v[0]); \ 199 if (ret < 0) \ 200 return ret; \ 201 in += ret; \ 202 \ 203 if (_bits > 64) { \ 204 ret = bch2_varint_decode_fast(in, end, &v[1]); \ 205 if (ret < 0) \ 206 return ret; \ 207 in += ret; \ 208 } else { \ 209 v[1] = 0; \ 210 } \ 211 } else { \ 212 v[0] = v[1] = 0; \ 213 } \ 214 \ 215 unpacked->_name = v[0]; \ 216 if (v[1] || v[0] != unpacked->_name) \ 217 return -1; \ 218 fieldnr++; 219 220 BCH_INODE_FIELDS_v2() 221 #undef x 222 223 /* XXX: signal if there were more fields than expected? */ 224 return 0; 225 } 226 227 static int bch2_inode_unpack_v3(struct bkey_s_c k, 228 struct bch_inode_unpacked *unpacked) 229 { 230 struct bkey_s_c_inode_v3 inode = bkey_s_c_to_inode_v3(k); 231 const u8 *in = inode.v->fields; 232 const u8 *end = bkey_val_end(inode); 233 unsigned nr_fields = INODEv3_NR_FIELDS(inode.v); 234 unsigned fieldnr = 0; 235 int ret; 236 u64 v[2]; 237 238 unpacked->bi_inum = inode.k->p.offset; 239 unpacked->bi_journal_seq= le64_to_cpu(inode.v->bi_journal_seq); 240 unpacked->bi_hash_seed = inode.v->bi_hash_seed; 241 unpacked->bi_flags = le64_to_cpu(inode.v->bi_flags); 242 unpacked->bi_sectors = le64_to_cpu(inode.v->bi_sectors); 243 unpacked->bi_size = le64_to_cpu(inode.v->bi_size); 244 unpacked->bi_version = le64_to_cpu(inode.v->bi_version); 245 unpacked->bi_mode = INODEv3_MODE(inode.v); 246 247 #define x(_name, _bits) \ 248 if (fieldnr < nr_fields) { \ 249 ret = bch2_varint_decode_fast(in, end, &v[0]); \ 250 if (ret < 0) \ 251 return ret; \ 252 in += ret; \ 253 \ 254 if (_bits > 64) { \ 255 ret = bch2_varint_decode_fast(in, end, &v[1]); \ 256 if (ret < 0) \ 257 return ret; \ 258 in += ret; \ 259 } else { \ 260 v[1] = 0; \ 261 } \ 262 } else { \ 263 v[0] = v[1] = 0; \ 264 } \ 265 \ 266 unpacked->_name = v[0]; \ 267 if (v[1] || v[0] != unpacked->_name) \ 268 return -1; \ 269 fieldnr++; 270 271 BCH_INODE_FIELDS_v3() 272 #undef x 273 274 /* XXX: signal if there were more fields than expected? */ 275 return 0; 276 } 277 278 static noinline int bch2_inode_unpack_slowpath(struct bkey_s_c k, 279 struct bch_inode_unpacked *unpacked) 280 { 281 memset(unpacked, 0, sizeof(*unpacked)); 282 283 switch (k.k->type) { 284 case KEY_TYPE_inode: { 285 struct bkey_s_c_inode inode = bkey_s_c_to_inode(k); 286 287 unpacked->bi_inum = inode.k->p.offset; 288 unpacked->bi_journal_seq= 0; 289 unpacked->bi_hash_seed = inode.v->bi_hash_seed; 290 unpacked->bi_flags = le32_to_cpu(inode.v->bi_flags); 291 unpacked->bi_mode = le16_to_cpu(inode.v->bi_mode); 292 293 if (INODE_NEW_VARINT(inode.v)) { 294 return bch2_inode_unpack_v2(unpacked, inode.v->fields, 295 bkey_val_end(inode), 296 INODE_NR_FIELDS(inode.v)); 297 } else { 298 return bch2_inode_unpack_v1(inode, unpacked); 299 } 300 break; 301 } 302 case KEY_TYPE_inode_v2: { 303 struct bkey_s_c_inode_v2 inode = bkey_s_c_to_inode_v2(k); 304 305 unpacked->bi_inum = inode.k->p.offset; 306 unpacked->bi_journal_seq= le64_to_cpu(inode.v->bi_journal_seq); 307 unpacked->bi_hash_seed = inode.v->bi_hash_seed; 308 unpacked->bi_flags = le64_to_cpu(inode.v->bi_flags); 309 unpacked->bi_mode = le16_to_cpu(inode.v->bi_mode); 310 311 return bch2_inode_unpack_v2(unpacked, inode.v->fields, 312 bkey_val_end(inode), 313 INODEv2_NR_FIELDS(inode.v)); 314 } 315 default: 316 BUG(); 317 } 318 } 319 320 int bch2_inode_unpack(struct bkey_s_c k, 321 struct bch_inode_unpacked *unpacked) 322 { 323 unpacked->bi_snapshot = k.k->p.snapshot; 324 325 return likely(k.k->type == KEY_TYPE_inode_v3) 326 ? bch2_inode_unpack_v3(k, unpacked) 327 : bch2_inode_unpack_slowpath(k, unpacked); 328 } 329 330 int bch2_inode_peek_nowarn(struct btree_trans *trans, 331 struct btree_iter *iter, 332 struct bch_inode_unpacked *inode, 333 subvol_inum inum, unsigned flags) 334 { 335 struct bkey_s_c k; 336 u32 snapshot; 337 int ret; 338 339 ret = bch2_subvolume_get_snapshot(trans, inum.subvol, &snapshot); 340 if (ret) 341 return ret; 342 343 k = bch2_bkey_get_iter(trans, iter, BTREE_ID_inodes, 344 SPOS(0, inum.inum, snapshot), 345 flags|BTREE_ITER_cached); 346 ret = bkey_err(k); 347 if (ret) 348 return ret; 349 350 ret = bkey_is_inode(k.k) ? 0 : -BCH_ERR_ENOENT_inode; 351 if (ret) 352 goto err; 353 354 ret = bch2_inode_unpack(k, inode); 355 if (ret) 356 goto err; 357 358 return 0; 359 err: 360 bch2_trans_iter_exit(trans, iter); 361 return ret; 362 } 363 364 int bch2_inode_peek(struct btree_trans *trans, 365 struct btree_iter *iter, 366 struct bch_inode_unpacked *inode, 367 subvol_inum inum, unsigned flags) 368 { 369 int ret = bch2_inode_peek_nowarn(trans, iter, inode, inum, flags); 370 bch_err_msg(trans->c, ret, "looking up inum %llu:%llu:", inum.subvol, inum.inum); 371 return ret; 372 } 373 374 int bch2_inode_write_flags(struct btree_trans *trans, 375 struct btree_iter *iter, 376 struct bch_inode_unpacked *inode, 377 enum btree_iter_update_trigger_flags flags) 378 { 379 struct bkey_inode_buf *inode_p; 380 381 inode_p = bch2_trans_kmalloc(trans, sizeof(*inode_p)); 382 if (IS_ERR(inode_p)) 383 return PTR_ERR(inode_p); 384 385 bch2_inode_pack_inlined(inode_p, inode); 386 inode_p->inode.k.p.snapshot = iter->snapshot; 387 return bch2_trans_update(trans, iter, &inode_p->inode.k_i, flags); 388 } 389 390 int __bch2_fsck_write_inode(struct btree_trans *trans, 391 struct bch_inode_unpacked *inode, 392 u32 snapshot) 393 { 394 struct bkey_inode_buf *inode_p = 395 bch2_trans_kmalloc(trans, sizeof(*inode_p)); 396 397 if (IS_ERR(inode_p)) 398 return PTR_ERR(inode_p); 399 400 bch2_inode_pack(inode_p, inode); 401 inode_p->inode.k.p.snapshot = snapshot; 402 403 return bch2_btree_insert_nonextent(trans, BTREE_ID_inodes, 404 &inode_p->inode.k_i, 405 BTREE_UPDATE_internal_snapshot_node); 406 } 407 408 int bch2_fsck_write_inode(struct btree_trans *trans, 409 struct bch_inode_unpacked *inode, 410 u32 snapshot) 411 { 412 int ret = commit_do(trans, NULL, NULL, BCH_TRANS_COMMIT_no_enospc, 413 __bch2_fsck_write_inode(trans, inode, snapshot)); 414 bch_err_fn(trans->c, ret); 415 return ret; 416 } 417 418 struct bkey_i *bch2_inode_to_v3(struct btree_trans *trans, struct bkey_i *k) 419 { 420 struct bch_inode_unpacked u; 421 struct bkey_inode_buf *inode_p; 422 int ret; 423 424 if (!bkey_is_inode(&k->k)) 425 return ERR_PTR(-ENOENT); 426 427 inode_p = bch2_trans_kmalloc(trans, sizeof(*inode_p)); 428 if (IS_ERR(inode_p)) 429 return ERR_CAST(inode_p); 430 431 ret = bch2_inode_unpack(bkey_i_to_s_c(k), &u); 432 if (ret) 433 return ERR_PTR(ret); 434 435 bch2_inode_pack(inode_p, &u); 436 return &inode_p->inode.k_i; 437 } 438 439 static int __bch2_inode_validate(struct bch_fs *c, struct bkey_s_c k, 440 enum bch_validate_flags flags) 441 { 442 struct bch_inode_unpacked unpacked; 443 int ret = 0; 444 445 bkey_fsck_err_on(k.k->p.inode, 446 c, inode_pos_inode_nonzero, 447 "nonzero k.p.inode"); 448 449 bkey_fsck_err_on(k.k->p.offset < BLOCKDEV_INODE_MAX, 450 c, inode_pos_blockdev_range, 451 "fs inode in blockdev range"); 452 453 bkey_fsck_err_on(bch2_inode_unpack(k, &unpacked), 454 c, inode_unpack_error, 455 "invalid variable length fields"); 456 457 bkey_fsck_err_on(unpacked.bi_data_checksum >= BCH_CSUM_OPT_NR + 1, 458 c, inode_checksum_type_invalid, 459 "invalid data checksum type (%u >= %u", 460 unpacked.bi_data_checksum, BCH_CSUM_OPT_NR + 1); 461 462 bkey_fsck_err_on(unpacked.bi_compression && 463 !bch2_compression_opt_valid(unpacked.bi_compression - 1), 464 c, inode_compression_type_invalid, 465 "invalid compression opt %u", unpacked.bi_compression - 1); 466 467 bkey_fsck_err_on((unpacked.bi_flags & BCH_INODE_unlinked) && 468 unpacked.bi_nlink != 0, 469 c, inode_unlinked_but_nlink_nonzero, 470 "flagged as unlinked but bi_nlink != 0"); 471 472 bkey_fsck_err_on(unpacked.bi_subvol && !S_ISDIR(unpacked.bi_mode), 473 c, inode_subvol_root_but_not_dir, 474 "subvolume root but not a directory"); 475 fsck_err: 476 return ret; 477 } 478 479 int bch2_inode_validate(struct bch_fs *c, struct bkey_s_c k, 480 enum bch_validate_flags flags) 481 { 482 struct bkey_s_c_inode inode = bkey_s_c_to_inode(k); 483 int ret = 0; 484 485 bkey_fsck_err_on(INODE_STR_HASH(inode.v) >= BCH_STR_HASH_NR, 486 c, inode_str_hash_invalid, 487 "invalid str hash type (%llu >= %u)", 488 INODE_STR_HASH(inode.v), BCH_STR_HASH_NR); 489 490 ret = __bch2_inode_validate(c, k, flags); 491 fsck_err: 492 return ret; 493 } 494 495 int bch2_inode_v2_validate(struct bch_fs *c, struct bkey_s_c k, 496 enum bch_validate_flags flags) 497 { 498 struct bkey_s_c_inode_v2 inode = bkey_s_c_to_inode_v2(k); 499 int ret = 0; 500 501 bkey_fsck_err_on(INODEv2_STR_HASH(inode.v) >= BCH_STR_HASH_NR, 502 c, inode_str_hash_invalid, 503 "invalid str hash type (%llu >= %u)", 504 INODEv2_STR_HASH(inode.v), BCH_STR_HASH_NR); 505 506 ret = __bch2_inode_validate(c, k, flags); 507 fsck_err: 508 return ret; 509 } 510 511 int bch2_inode_v3_validate(struct bch_fs *c, struct bkey_s_c k, 512 enum bch_validate_flags flags) 513 { 514 struct bkey_s_c_inode_v3 inode = bkey_s_c_to_inode_v3(k); 515 int ret = 0; 516 517 bkey_fsck_err_on(INODEv3_FIELDS_START(inode.v) < INODEv3_FIELDS_START_INITIAL || 518 INODEv3_FIELDS_START(inode.v) > bkey_val_u64s(inode.k), 519 c, inode_v3_fields_start_bad, 520 "invalid fields_start (got %llu, min %u max %zu)", 521 INODEv3_FIELDS_START(inode.v), 522 INODEv3_FIELDS_START_INITIAL, 523 bkey_val_u64s(inode.k)); 524 525 bkey_fsck_err_on(INODEv3_STR_HASH(inode.v) >= BCH_STR_HASH_NR, 526 c, inode_str_hash_invalid, 527 "invalid str hash type (%llu >= %u)", 528 INODEv3_STR_HASH(inode.v), BCH_STR_HASH_NR); 529 530 ret = __bch2_inode_validate(c, k, flags); 531 fsck_err: 532 return ret; 533 } 534 535 static void __bch2_inode_unpacked_to_text(struct printbuf *out, 536 struct bch_inode_unpacked *inode) 537 { 538 prt_printf(out, "\n"); 539 printbuf_indent_add(out, 2); 540 prt_printf(out, "mode=%o\n", inode->bi_mode); 541 542 prt_str(out, "flags="); 543 prt_bitflags(out, bch2_inode_flag_strs, inode->bi_flags & ((1U << 20) - 1)); 544 prt_printf(out, "(%x)\n", inode->bi_flags); 545 546 prt_printf(out, "journal_seq=%llu\n", inode->bi_journal_seq); 547 prt_printf(out, "bi_size=%llu\n", inode->bi_size); 548 prt_printf(out, "bi_sectors=%llu\n", inode->bi_sectors); 549 prt_printf(out, "bi_version=%llu\n", inode->bi_version); 550 551 #define x(_name, _bits) \ 552 prt_printf(out, #_name "=%llu\n", (u64) inode->_name); 553 BCH_INODE_FIELDS_v3() 554 #undef x 555 556 bch2_printbuf_strip_trailing_newline(out); 557 printbuf_indent_sub(out, 2); 558 } 559 560 void bch2_inode_unpacked_to_text(struct printbuf *out, struct bch_inode_unpacked *inode) 561 { 562 prt_printf(out, "inum: %llu:%u ", inode->bi_inum, inode->bi_snapshot); 563 __bch2_inode_unpacked_to_text(out, inode); 564 } 565 566 void bch2_inode_to_text(struct printbuf *out, struct bch_fs *c, struct bkey_s_c k) 567 { 568 struct bch_inode_unpacked inode; 569 570 if (bch2_inode_unpack(k, &inode)) { 571 prt_printf(out, "(unpack error)"); 572 return; 573 } 574 575 __bch2_inode_unpacked_to_text(out, &inode); 576 } 577 578 static inline u64 bkey_inode_flags(struct bkey_s_c k) 579 { 580 switch (k.k->type) { 581 case KEY_TYPE_inode: 582 return le32_to_cpu(bkey_s_c_to_inode(k).v->bi_flags); 583 case KEY_TYPE_inode_v2: 584 return le64_to_cpu(bkey_s_c_to_inode_v2(k).v->bi_flags); 585 case KEY_TYPE_inode_v3: 586 return le64_to_cpu(bkey_s_c_to_inode_v3(k).v->bi_flags); 587 default: 588 return 0; 589 } 590 } 591 592 static inline bool bkey_is_deleted_inode(struct bkey_s_c k) 593 { 594 return bkey_inode_flags(k) & BCH_INODE_unlinked; 595 } 596 597 int bch2_trigger_inode(struct btree_trans *trans, 598 enum btree_id btree_id, unsigned level, 599 struct bkey_s_c old, 600 struct bkey_s new, 601 enum btree_iter_update_trigger_flags flags) 602 { 603 if ((flags & BTREE_TRIGGER_atomic) && (flags & BTREE_TRIGGER_insert)) { 604 BUG_ON(!trans->journal_res.seq); 605 bkey_s_to_inode_v3(new).v->bi_journal_seq = cpu_to_le64(trans->journal_res.seq); 606 } 607 608 s64 nr = bkey_is_inode(new.k) - bkey_is_inode(old.k); 609 if ((flags & (BTREE_TRIGGER_transactional|BTREE_TRIGGER_gc)) && nr) { 610 struct disk_accounting_pos acc = { .type = BCH_DISK_ACCOUNTING_nr_inodes }; 611 int ret = bch2_disk_accounting_mod(trans, &acc, &nr, 1, flags & BTREE_TRIGGER_gc); 612 if (ret) 613 return ret; 614 } 615 616 int deleted_delta = (int) bkey_is_deleted_inode(new.s_c) - 617 (int) bkey_is_deleted_inode(old); 618 if ((flags & BTREE_TRIGGER_transactional) && deleted_delta) { 619 int ret = bch2_btree_bit_mod_buffered(trans, BTREE_ID_deleted_inodes, 620 new.k->p, deleted_delta > 0); 621 if (ret) 622 return ret; 623 } 624 625 return 0; 626 } 627 628 int bch2_inode_generation_validate(struct bch_fs *c, struct bkey_s_c k, 629 enum bch_validate_flags flags) 630 { 631 int ret = 0; 632 633 bkey_fsck_err_on(k.k->p.inode, 634 c, inode_pos_inode_nonzero, 635 "nonzero k.p.inode"); 636 fsck_err: 637 return ret; 638 } 639 640 void bch2_inode_generation_to_text(struct printbuf *out, struct bch_fs *c, 641 struct bkey_s_c k) 642 { 643 struct bkey_s_c_inode_generation gen = bkey_s_c_to_inode_generation(k); 644 645 prt_printf(out, "generation: %u", le32_to_cpu(gen.v->bi_generation)); 646 } 647 648 void bch2_inode_init_early(struct bch_fs *c, 649 struct bch_inode_unpacked *inode_u) 650 { 651 enum bch_str_hash_type str_hash = 652 bch2_str_hash_opt_to_type(c, c->opts.str_hash); 653 654 memset(inode_u, 0, sizeof(*inode_u)); 655 656 /* ick */ 657 inode_u->bi_flags |= str_hash << INODE_STR_HASH_OFFSET; 658 get_random_bytes(&inode_u->bi_hash_seed, 659 sizeof(inode_u->bi_hash_seed)); 660 } 661 662 void bch2_inode_init_late(struct bch_inode_unpacked *inode_u, u64 now, 663 uid_t uid, gid_t gid, umode_t mode, dev_t rdev, 664 struct bch_inode_unpacked *parent) 665 { 666 inode_u->bi_mode = mode; 667 inode_u->bi_uid = uid; 668 inode_u->bi_gid = gid; 669 inode_u->bi_dev = rdev; 670 inode_u->bi_atime = now; 671 inode_u->bi_mtime = now; 672 inode_u->bi_ctime = now; 673 inode_u->bi_otime = now; 674 675 if (parent && parent->bi_mode & S_ISGID) { 676 inode_u->bi_gid = parent->bi_gid; 677 if (S_ISDIR(mode)) 678 inode_u->bi_mode |= S_ISGID; 679 } 680 681 if (parent) { 682 #define x(_name, ...) inode_u->bi_##_name = parent->bi_##_name; 683 BCH_INODE_OPTS() 684 #undef x 685 } 686 } 687 688 void bch2_inode_init(struct bch_fs *c, struct bch_inode_unpacked *inode_u, 689 uid_t uid, gid_t gid, umode_t mode, dev_t rdev, 690 struct bch_inode_unpacked *parent) 691 { 692 bch2_inode_init_early(c, inode_u); 693 bch2_inode_init_late(inode_u, bch2_current_time(c), 694 uid, gid, mode, rdev, parent); 695 } 696 697 static inline u32 bkey_generation(struct bkey_s_c k) 698 { 699 switch (k.k->type) { 700 case KEY_TYPE_inode: 701 case KEY_TYPE_inode_v2: 702 BUG(); 703 case KEY_TYPE_inode_generation: 704 return le32_to_cpu(bkey_s_c_to_inode_generation(k).v->bi_generation); 705 default: 706 return 0; 707 } 708 } 709 710 /* 711 * This just finds an empty slot: 712 */ 713 int bch2_inode_create(struct btree_trans *trans, 714 struct btree_iter *iter, 715 struct bch_inode_unpacked *inode_u, 716 u32 snapshot, u64 cpu) 717 { 718 struct bch_fs *c = trans->c; 719 struct bkey_s_c k; 720 u64 min, max, start, pos, *hint; 721 int ret = 0; 722 unsigned bits = (c->opts.inodes_32bit ? 31 : 63); 723 724 if (c->opts.shard_inode_numbers) { 725 bits -= c->inode_shard_bits; 726 727 min = (cpu << bits); 728 max = (cpu << bits) | ~(ULLONG_MAX << bits); 729 730 min = max_t(u64, min, BLOCKDEV_INODE_MAX); 731 hint = c->unused_inode_hints + cpu; 732 } else { 733 min = BLOCKDEV_INODE_MAX; 734 max = ~(ULLONG_MAX << bits); 735 hint = c->unused_inode_hints; 736 } 737 738 start = READ_ONCE(*hint); 739 740 if (start >= max || start < min) 741 start = min; 742 743 pos = start; 744 bch2_trans_iter_init(trans, iter, BTREE_ID_inodes, POS(0, pos), 745 BTREE_ITER_all_snapshots| 746 BTREE_ITER_intent); 747 again: 748 while ((k = bch2_btree_iter_peek(iter)).k && 749 !(ret = bkey_err(k)) && 750 bkey_lt(k.k->p, POS(0, max))) { 751 if (pos < iter->pos.offset) 752 goto found_slot; 753 754 /* 755 * We don't need to iterate over keys in every snapshot once 756 * we've found just one: 757 */ 758 pos = iter->pos.offset + 1; 759 bch2_btree_iter_set_pos(iter, POS(0, pos)); 760 } 761 762 if (!ret && pos < max) 763 goto found_slot; 764 765 if (!ret && start == min) 766 ret = -BCH_ERR_ENOSPC_inode_create; 767 768 if (ret) { 769 bch2_trans_iter_exit(trans, iter); 770 return ret; 771 } 772 773 /* Retry from start */ 774 pos = start = min; 775 bch2_btree_iter_set_pos(iter, POS(0, pos)); 776 goto again; 777 found_slot: 778 bch2_btree_iter_set_pos(iter, SPOS(0, pos, snapshot)); 779 k = bch2_btree_iter_peek_slot(iter); 780 ret = bkey_err(k); 781 if (ret) { 782 bch2_trans_iter_exit(trans, iter); 783 return ret; 784 } 785 786 *hint = k.k->p.offset; 787 inode_u->bi_inum = k.k->p.offset; 788 inode_u->bi_generation = bkey_generation(k); 789 return 0; 790 } 791 792 static int bch2_inode_delete_keys(struct btree_trans *trans, 793 subvol_inum inum, enum btree_id id) 794 { 795 struct btree_iter iter; 796 struct bkey_s_c k; 797 struct bkey_i delete; 798 struct bpos end = POS(inum.inum, U64_MAX); 799 u32 snapshot; 800 int ret = 0; 801 802 /* 803 * We're never going to be deleting partial extents, no need to use an 804 * extent iterator: 805 */ 806 bch2_trans_iter_init(trans, &iter, id, POS(inum.inum, 0), 807 BTREE_ITER_intent); 808 809 while (1) { 810 bch2_trans_begin(trans); 811 812 ret = bch2_subvolume_get_snapshot(trans, inum.subvol, &snapshot); 813 if (ret) 814 goto err; 815 816 bch2_btree_iter_set_snapshot(&iter, snapshot); 817 818 k = bch2_btree_iter_peek_upto(&iter, end); 819 ret = bkey_err(k); 820 if (ret) 821 goto err; 822 823 if (!k.k) 824 break; 825 826 bkey_init(&delete.k); 827 delete.k.p = iter.pos; 828 829 if (iter.flags & BTREE_ITER_is_extents) 830 bch2_key_resize(&delete.k, 831 bpos_min(end, k.k->p).offset - 832 iter.pos.offset); 833 834 ret = bch2_trans_update(trans, &iter, &delete, 0) ?: 835 bch2_trans_commit(trans, NULL, NULL, 836 BCH_TRANS_COMMIT_no_enospc); 837 err: 838 if (ret && !bch2_err_matches(ret, BCH_ERR_transaction_restart)) 839 break; 840 } 841 842 bch2_trans_iter_exit(trans, &iter); 843 return ret; 844 } 845 846 int bch2_inode_rm(struct bch_fs *c, subvol_inum inum) 847 { 848 struct btree_trans *trans = bch2_trans_get(c); 849 struct btree_iter iter = { NULL }; 850 struct bkey_i_inode_generation delete; 851 struct bch_inode_unpacked inode_u; 852 struct bkey_s_c k; 853 u32 snapshot; 854 int ret; 855 856 /* 857 * If this was a directory, there shouldn't be any real dirents left - 858 * but there could be whiteouts (from hash collisions) that we should 859 * delete: 860 * 861 * XXX: the dirent could ideally would delete whiteouts when they're no 862 * longer needed 863 */ 864 ret = bch2_inode_delete_keys(trans, inum, BTREE_ID_extents) ?: 865 bch2_inode_delete_keys(trans, inum, BTREE_ID_xattrs) ?: 866 bch2_inode_delete_keys(trans, inum, BTREE_ID_dirents); 867 if (ret) 868 goto err; 869 retry: 870 bch2_trans_begin(trans); 871 872 ret = bch2_subvolume_get_snapshot(trans, inum.subvol, &snapshot); 873 if (ret) 874 goto err; 875 876 k = bch2_bkey_get_iter(trans, &iter, BTREE_ID_inodes, 877 SPOS(0, inum.inum, snapshot), 878 BTREE_ITER_intent|BTREE_ITER_cached); 879 ret = bkey_err(k); 880 if (ret) 881 goto err; 882 883 if (!bkey_is_inode(k.k)) { 884 bch2_fs_inconsistent(c, 885 "inode %llu:%u not found when deleting", 886 inum.inum, snapshot); 887 ret = -EIO; 888 goto err; 889 } 890 891 bch2_inode_unpack(k, &inode_u); 892 893 bkey_inode_generation_init(&delete.k_i); 894 delete.k.p = iter.pos; 895 delete.v.bi_generation = cpu_to_le32(inode_u.bi_generation + 1); 896 897 ret = bch2_trans_update(trans, &iter, &delete.k_i, 0) ?: 898 bch2_trans_commit(trans, NULL, NULL, 899 BCH_TRANS_COMMIT_no_enospc); 900 err: 901 bch2_trans_iter_exit(trans, &iter); 902 if (bch2_err_matches(ret, BCH_ERR_transaction_restart)) 903 goto retry; 904 905 bch2_trans_put(trans); 906 return ret; 907 } 908 909 int bch2_inode_find_by_inum_nowarn_trans(struct btree_trans *trans, 910 subvol_inum inum, 911 struct bch_inode_unpacked *inode) 912 { 913 struct btree_iter iter; 914 int ret; 915 916 ret = bch2_inode_peek_nowarn(trans, &iter, inode, inum, 0); 917 if (!ret) 918 bch2_trans_iter_exit(trans, &iter); 919 return ret; 920 } 921 922 int bch2_inode_find_by_inum_trans(struct btree_trans *trans, 923 subvol_inum inum, 924 struct bch_inode_unpacked *inode) 925 { 926 struct btree_iter iter; 927 int ret; 928 929 ret = bch2_inode_peek(trans, &iter, inode, inum, 0); 930 if (!ret) 931 bch2_trans_iter_exit(trans, &iter); 932 return ret; 933 } 934 935 int bch2_inode_find_by_inum(struct bch_fs *c, subvol_inum inum, 936 struct bch_inode_unpacked *inode) 937 { 938 return bch2_trans_do(c, NULL, NULL, 0, 939 bch2_inode_find_by_inum_trans(trans, inum, inode)); 940 } 941 942 int bch2_inode_nlink_inc(struct bch_inode_unpacked *bi) 943 { 944 if (bi->bi_flags & BCH_INODE_unlinked) 945 bi->bi_flags &= ~BCH_INODE_unlinked; 946 else { 947 if (bi->bi_nlink == U32_MAX) 948 return -EINVAL; 949 950 bi->bi_nlink++; 951 } 952 953 return 0; 954 } 955 956 void bch2_inode_nlink_dec(struct btree_trans *trans, struct bch_inode_unpacked *bi) 957 { 958 if (bi->bi_nlink && (bi->bi_flags & BCH_INODE_unlinked)) { 959 bch2_trans_inconsistent(trans, "inode %llu unlinked but link count nonzero", 960 bi->bi_inum); 961 return; 962 } 963 964 if (bi->bi_flags & BCH_INODE_unlinked) { 965 bch2_trans_inconsistent(trans, "inode %llu link count underflow", bi->bi_inum); 966 return; 967 } 968 969 if (bi->bi_nlink) 970 bi->bi_nlink--; 971 else 972 bi->bi_flags |= BCH_INODE_unlinked; 973 } 974 975 struct bch_opts bch2_inode_opts_to_opts(struct bch_inode_unpacked *inode) 976 { 977 struct bch_opts ret = { 0 }; 978 #define x(_name, _bits) \ 979 if (inode->bi_##_name) \ 980 opt_set(ret, _name, inode->bi_##_name - 1); 981 BCH_INODE_OPTS() 982 #undef x 983 return ret; 984 } 985 986 void bch2_inode_opts_get(struct bch_io_opts *opts, struct bch_fs *c, 987 struct bch_inode_unpacked *inode) 988 { 989 #define x(_name, _bits) opts->_name = inode_opt_get(c, inode, _name); 990 BCH_INODE_OPTS() 991 #undef x 992 993 if (opts->nocow) 994 opts->compression = opts->background_compression = opts->data_checksum = opts->erasure_code = 0; 995 } 996 997 int bch2_inum_opts_get(struct btree_trans *trans, subvol_inum inum, struct bch_io_opts *opts) 998 { 999 struct bch_inode_unpacked inode; 1000 int ret = lockrestart_do(trans, bch2_inode_find_by_inum_trans(trans, inum, &inode)); 1001 1002 if (ret) 1003 return ret; 1004 1005 bch2_inode_opts_get(opts, trans->c, &inode); 1006 return 0; 1007 } 1008 1009 int bch2_inode_rm_snapshot(struct btree_trans *trans, u64 inum, u32 snapshot) 1010 { 1011 struct bch_fs *c = trans->c; 1012 struct btree_iter iter = { NULL }; 1013 struct bkey_i_inode_generation delete; 1014 struct bch_inode_unpacked inode_u; 1015 struct bkey_s_c k; 1016 int ret; 1017 1018 do { 1019 ret = bch2_btree_delete_range_trans(trans, BTREE_ID_extents, 1020 SPOS(inum, 0, snapshot), 1021 SPOS(inum, U64_MAX, snapshot), 1022 0, NULL) ?: 1023 bch2_btree_delete_range_trans(trans, BTREE_ID_dirents, 1024 SPOS(inum, 0, snapshot), 1025 SPOS(inum, U64_MAX, snapshot), 1026 0, NULL) ?: 1027 bch2_btree_delete_range_trans(trans, BTREE_ID_xattrs, 1028 SPOS(inum, 0, snapshot), 1029 SPOS(inum, U64_MAX, snapshot), 1030 0, NULL); 1031 } while (ret == -BCH_ERR_transaction_restart_nested); 1032 if (ret) 1033 goto err; 1034 retry: 1035 bch2_trans_begin(trans); 1036 1037 k = bch2_bkey_get_iter(trans, &iter, BTREE_ID_inodes, 1038 SPOS(0, inum, snapshot), BTREE_ITER_intent); 1039 ret = bkey_err(k); 1040 if (ret) 1041 goto err; 1042 1043 if (!bkey_is_inode(k.k)) { 1044 bch2_fs_inconsistent(c, 1045 "inode %llu:%u not found when deleting", 1046 inum, snapshot); 1047 ret = -EIO; 1048 goto err; 1049 } 1050 1051 bch2_inode_unpack(k, &inode_u); 1052 1053 /* Subvolume root? */ 1054 if (inode_u.bi_subvol) 1055 bch_warn(c, "deleting inode %llu marked as unlinked, but also a subvolume root!?", inode_u.bi_inum); 1056 1057 bkey_inode_generation_init(&delete.k_i); 1058 delete.k.p = iter.pos; 1059 delete.v.bi_generation = cpu_to_le32(inode_u.bi_generation + 1); 1060 1061 ret = bch2_trans_update(trans, &iter, &delete.k_i, 0) ?: 1062 bch2_trans_commit(trans, NULL, NULL, 1063 BCH_TRANS_COMMIT_no_enospc); 1064 err: 1065 bch2_trans_iter_exit(trans, &iter); 1066 if (bch2_err_matches(ret, BCH_ERR_transaction_restart)) 1067 goto retry; 1068 1069 return ret ?: -BCH_ERR_transaction_restart_nested; 1070 } 1071 1072 static int may_delete_deleted_inode(struct btree_trans *trans, 1073 struct btree_iter *iter, 1074 struct bpos pos, 1075 bool *need_another_pass) 1076 { 1077 struct bch_fs *c = trans->c; 1078 struct btree_iter inode_iter; 1079 struct bkey_s_c k; 1080 struct bch_inode_unpacked inode; 1081 int ret; 1082 1083 k = bch2_bkey_get_iter(trans, &inode_iter, BTREE_ID_inodes, pos, BTREE_ITER_cached); 1084 ret = bkey_err(k); 1085 if (ret) 1086 return ret; 1087 1088 ret = bkey_is_inode(k.k) ? 0 : -BCH_ERR_ENOENT_inode; 1089 if (fsck_err_on(!bkey_is_inode(k.k), 1090 trans, deleted_inode_missing, 1091 "nonexistent inode %llu:%u in deleted_inodes btree", 1092 pos.offset, pos.snapshot)) 1093 goto delete; 1094 1095 ret = bch2_inode_unpack(k, &inode); 1096 if (ret) 1097 goto out; 1098 1099 if (S_ISDIR(inode.bi_mode)) { 1100 ret = bch2_empty_dir_snapshot(trans, pos.offset, 0, pos.snapshot); 1101 if (fsck_err_on(bch2_err_matches(ret, ENOTEMPTY), 1102 trans, deleted_inode_is_dir, 1103 "non empty directory %llu:%u in deleted_inodes btree", 1104 pos.offset, pos.snapshot)) 1105 goto delete; 1106 if (ret) 1107 goto out; 1108 } 1109 1110 if (fsck_err_on(!(inode.bi_flags & BCH_INODE_unlinked), 1111 trans, deleted_inode_not_unlinked, 1112 "non-deleted inode %llu:%u in deleted_inodes btree", 1113 pos.offset, pos.snapshot)) 1114 goto delete; 1115 1116 if (test_bit(BCH_FS_clean_recovery, &c->flags) && 1117 !fsck_err(trans, deleted_inode_but_clean, 1118 "filesystem marked as clean but have deleted inode %llu:%u", 1119 pos.offset, pos.snapshot)) { 1120 ret = 0; 1121 goto out; 1122 } 1123 1124 if (bch2_snapshot_is_internal_node(c, pos.snapshot)) { 1125 struct bpos new_min_pos; 1126 1127 ret = bch2_propagate_key_to_snapshot_leaves(trans, inode_iter.btree_id, k, &new_min_pos); 1128 if (ret) 1129 goto out; 1130 1131 inode.bi_flags &= ~BCH_INODE_unlinked; 1132 1133 ret = bch2_inode_write_flags(trans, &inode_iter, &inode, 1134 BTREE_UPDATE_internal_snapshot_node); 1135 bch_err_msg(c, ret, "clearing inode unlinked flag"); 1136 if (ret) 1137 goto out; 1138 1139 /* 1140 * We'll need another write buffer flush to pick up the new 1141 * unlinked inodes in the snapshot leaves: 1142 */ 1143 *need_another_pass = true; 1144 goto out; 1145 } 1146 1147 ret = 1; 1148 out: 1149 fsck_err: 1150 bch2_trans_iter_exit(trans, &inode_iter); 1151 return ret; 1152 delete: 1153 ret = bch2_btree_bit_mod_buffered(trans, BTREE_ID_deleted_inodes, pos, false); 1154 goto out; 1155 } 1156 1157 int bch2_delete_dead_inodes(struct bch_fs *c) 1158 { 1159 struct btree_trans *trans = bch2_trans_get(c); 1160 bool need_another_pass; 1161 int ret; 1162 again: 1163 /* 1164 * if we ran check_inodes() unlinked inodes will have already been 1165 * cleaned up but the write buffer will be out of sync; therefore we 1166 * alway need a write buffer flush 1167 */ 1168 ret = bch2_btree_write_buffer_flush_sync(trans); 1169 if (ret) 1170 goto err; 1171 1172 need_another_pass = false; 1173 1174 /* 1175 * Weird transaction restart handling here because on successful delete, 1176 * bch2_inode_rm_snapshot() will return a nested transaction restart, 1177 * but we can't retry because the btree write buffer won't have been 1178 * flushed and we'd spin: 1179 */ 1180 ret = for_each_btree_key_commit(trans, iter, BTREE_ID_deleted_inodes, POS_MIN, 1181 BTREE_ITER_prefetch|BTREE_ITER_all_snapshots, k, 1182 NULL, NULL, BCH_TRANS_COMMIT_no_enospc, ({ 1183 ret = may_delete_deleted_inode(trans, &iter, k.k->p, &need_another_pass); 1184 if (ret > 0) { 1185 bch_verbose(c, "deleting unlinked inode %llu:%u", k.k->p.offset, k.k->p.snapshot); 1186 1187 ret = bch2_inode_rm_snapshot(trans, k.k->p.offset, k.k->p.snapshot); 1188 /* 1189 * We don't want to loop here: a transaction restart 1190 * error here means we handled a transaction restart and 1191 * we're actually done, but if we loop we'll retry the 1192 * same key because the write buffer hasn't been flushed 1193 * yet 1194 */ 1195 if (bch2_err_matches(ret, BCH_ERR_transaction_restart)) { 1196 ret = 0; 1197 continue; 1198 } 1199 } 1200 1201 ret; 1202 })); 1203 1204 if (!ret && need_another_pass) 1205 goto again; 1206 err: 1207 bch2_trans_put(trans); 1208 return ret; 1209 } 1210