1 /* SPDX-License-Identifier: GPL-2.0 */ 2 #ifndef _BCACHEFS_FORMAT_H 3 #define _BCACHEFS_FORMAT_H 4 5 /* 6 * bcachefs on disk data structures 7 * 8 * OVERVIEW: 9 * 10 * There are three main types of on disk data structures in bcachefs (this is 11 * reduced from 5 in bcache) 12 * 13 * - superblock 14 * - journal 15 * - btree 16 * 17 * The btree is the primary structure; most metadata exists as keys in the 18 * various btrees. There are only a small number of btrees, they're not 19 * sharded - we have one btree for extents, another for inodes, et cetera. 20 * 21 * SUPERBLOCK: 22 * 23 * The superblock contains the location of the journal, the list of devices in 24 * the filesystem, and in general any metadata we need in order to decide 25 * whether we can start a filesystem or prior to reading the journal/btree 26 * roots. 27 * 28 * The superblock is extensible, and most of the contents of the superblock are 29 * in variable length, type tagged fields; see struct bch_sb_field. 30 * 31 * Backup superblocks do not reside in a fixed location; also, superblocks do 32 * not have a fixed size. To locate backup superblocks we have struct 33 * bch_sb_layout; we store a copy of this inside every superblock, and also 34 * before the first superblock. 35 * 36 * JOURNAL: 37 * 38 * The journal primarily records btree updates in the order they occurred; 39 * journal replay consists of just iterating over all the keys in the open 40 * journal entries and re-inserting them into the btrees. 41 * 42 * The journal also contains entry types for the btree roots, and blacklisted 43 * journal sequence numbers (see journal_seq_blacklist.c). 44 * 45 * BTREE: 46 * 47 * bcachefs btrees are copy on write b+ trees, where nodes are big (typically 48 * 128k-256k) and log structured. We use struct btree_node for writing the first 49 * entry in a given node (offset 0), and struct btree_node_entry for all 50 * subsequent writes. 51 * 52 * After the header, btree node entries contain a list of keys in sorted order. 53 * Values are stored inline with the keys; since values are variable length (and 54 * keys effectively are variable length too, due to packing) we can't do random 55 * access without building up additional in memory tables in the btree node read 56 * path. 57 * 58 * BTREE KEYS (struct bkey): 59 * 60 * The various btrees share a common format for the key - so as to avoid 61 * switching in fastpath lookup/comparison code - but define their own 62 * structures for the key values. 63 * 64 * The size of a key/value pair is stored as a u8 in units of u64s, so the max 65 * size is just under 2k. The common part also contains a type tag for the 66 * value, and a format field indicating whether the key is packed or not (and 67 * also meant to allow adding new key fields in the future, if desired). 68 * 69 * bkeys, when stored within a btree node, may also be packed. In that case, the 70 * bkey_format in that node is used to unpack it. Packed bkeys mean that we can 71 * be generous with field sizes in the common part of the key format (64 bit 72 * inode number, 64 bit offset, 96 bit version field, etc.) for negligible cost. 73 */ 74 75 #include <asm/types.h> 76 #include <asm/byteorder.h> 77 #include <linux/kernel.h> 78 #include <linux/uuid.h> 79 #include <uapi/linux/magic.h> 80 #include "vstructs.h" 81 82 #ifdef __KERNEL__ 83 typedef uuid_t __uuid_t; 84 #endif 85 86 #define BITMASK(name, type, field, offset, end) \ 87 static const __maybe_unused unsigned name##_OFFSET = offset; \ 88 static const __maybe_unused unsigned name##_BITS = (end - offset); \ 89 \ 90 static inline __u64 name(const type *k) \ 91 { \ 92 return (k->field >> offset) & ~(~0ULL << (end - offset)); \ 93 } \ 94 \ 95 static inline void SET_##name(type *k, __u64 v) \ 96 { \ 97 k->field &= ~(~(~0ULL << (end - offset)) << offset); \ 98 k->field |= (v & ~(~0ULL << (end - offset))) << offset; \ 99 } 100 101 #define LE_BITMASK(_bits, name, type, field, offset, end) \ 102 static const __maybe_unused unsigned name##_OFFSET = offset; \ 103 static const __maybe_unused unsigned name##_BITS = (end - offset); \ 104 static const __maybe_unused __u##_bits name##_MAX = (1ULL << (end - offset)) - 1;\ 105 \ 106 static inline __u64 name(const type *k) \ 107 { \ 108 return (__le##_bits##_to_cpu(k->field) >> offset) & \ 109 ~(~0ULL << (end - offset)); \ 110 } \ 111 \ 112 static inline void SET_##name(type *k, __u64 v) \ 113 { \ 114 __u##_bits new = __le##_bits##_to_cpu(k->field); \ 115 \ 116 new &= ~(~(~0ULL << (end - offset)) << offset); \ 117 new |= (v & ~(~0ULL << (end - offset))) << offset; \ 118 k->field = __cpu_to_le##_bits(new); \ 119 } 120 121 #define LE16_BITMASK(n, t, f, o, e) LE_BITMASK(16, n, t, f, o, e) 122 #define LE32_BITMASK(n, t, f, o, e) LE_BITMASK(32, n, t, f, o, e) 123 #define LE64_BITMASK(n, t, f, o, e) LE_BITMASK(64, n, t, f, o, e) 124 125 struct bkey_format { 126 __u8 key_u64s; 127 __u8 nr_fields; 128 /* One unused slot for now: */ 129 __u8 bits_per_field[6]; 130 __le64 field_offset[6]; 131 }; 132 133 /* Btree keys - all units are in sectors */ 134 135 struct bpos { 136 /* 137 * Word order matches machine byte order - btree code treats a bpos as a 138 * single large integer, for search/comparison purposes 139 * 140 * Note that wherever a bpos is embedded in another on disk data 141 * structure, it has to be byte swabbed when reading in metadata that 142 * wasn't written in native endian order: 143 */ 144 #if __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__ 145 __u32 snapshot; 146 __u64 offset; 147 __u64 inode; 148 #elif __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__ 149 __u64 inode; 150 __u64 offset; /* Points to end of extent - sectors */ 151 __u32 snapshot; 152 #else 153 #error edit for your odd byteorder. 154 #endif 155 } __packed 156 #if __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__ 157 __aligned(4) 158 #endif 159 ; 160 161 #define KEY_INODE_MAX ((__u64)~0ULL) 162 #define KEY_OFFSET_MAX ((__u64)~0ULL) 163 #define KEY_SNAPSHOT_MAX ((__u32)~0U) 164 #define KEY_SIZE_MAX ((__u32)~0U) 165 166 static inline struct bpos SPOS(__u64 inode, __u64 offset, __u32 snapshot) 167 { 168 return (struct bpos) { 169 .inode = inode, 170 .offset = offset, 171 .snapshot = snapshot, 172 }; 173 } 174 175 #define POS_MIN SPOS(0, 0, 0) 176 #define POS_MAX SPOS(KEY_INODE_MAX, KEY_OFFSET_MAX, 0) 177 #define SPOS_MAX SPOS(KEY_INODE_MAX, KEY_OFFSET_MAX, KEY_SNAPSHOT_MAX) 178 #define POS(_inode, _offset) SPOS(_inode, _offset, 0) 179 180 /* Empty placeholder struct, for container_of() */ 181 struct bch_val { 182 __u64 __nothing[0]; 183 }; 184 185 struct bversion { 186 #if __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__ 187 __u64 lo; 188 __u32 hi; 189 #elif __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__ 190 __u32 hi; 191 __u64 lo; 192 #endif 193 } __packed 194 #if __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__ 195 __aligned(4) 196 #endif 197 ; 198 199 struct bkey { 200 /* Size of combined key and value, in u64s */ 201 __u8 u64s; 202 203 /* Format of key (0 for format local to btree node) */ 204 #if defined(__LITTLE_ENDIAN_BITFIELD) 205 __u8 format:7, 206 needs_whiteout:1; 207 #elif defined (__BIG_ENDIAN_BITFIELD) 208 __u8 needs_whiteout:1, 209 format:7; 210 #else 211 #error edit for your odd byteorder. 212 #endif 213 214 /* Type of the value */ 215 __u8 type; 216 217 #if __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__ 218 __u8 pad[1]; 219 220 struct bversion version; 221 __u32 size; /* extent size, in sectors */ 222 struct bpos p; 223 #elif __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__ 224 struct bpos p; 225 __u32 size; /* extent size, in sectors */ 226 struct bversion version; 227 228 __u8 pad[1]; 229 #endif 230 } __packed 231 #if __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__ 232 /* 233 * The big-endian version of bkey can't be compiled by rustc with the "aligned" 234 * attr since it doesn't allow types to have both "packed" and "aligned" attrs. 235 * So for Rust compatibility, don't include this. It can be included in the LE 236 * version because the "packed" attr is redundant in that case. 237 * 238 * History: (quoting Kent) 239 * 240 * Specifically, when i was designing bkey, I wanted the header to be no 241 * bigger than necessary so that bkey_packed could use the rest. That means that 242 * decently offten extent keys will fit into only 8 bytes, instead of spilling over 243 * to 16. 244 * 245 * But packed_bkey treats the part after the header - the packed section - 246 * as a single multi word, variable length integer. And bkey, the unpacked 247 * version, is just a special case version of a bkey_packed; all the packed 248 * bkey code will work on keys in any packed format, the in-memory 249 * representation of an unpacked key also is just one type of packed key... 250 * 251 * So that constrains the key part of a bkig endian bkey to start right 252 * after the header. 253 * 254 * If we ever do a bkey_v2 and need to expand the hedaer by another byte for 255 * some reason - that will clean up this wart. 256 */ 257 __aligned(8) 258 #endif 259 ; 260 261 struct bkey_packed { 262 __u64 _data[0]; 263 264 /* Size of combined key and value, in u64s */ 265 __u8 u64s; 266 267 /* Format of key (0 for format local to btree node) */ 268 269 /* 270 * XXX: next incompat on disk format change, switch format and 271 * needs_whiteout - bkey_packed() will be cheaper if format is the high 272 * bits of the bitfield 273 */ 274 #if defined(__LITTLE_ENDIAN_BITFIELD) 275 __u8 format:7, 276 needs_whiteout:1; 277 #elif defined (__BIG_ENDIAN_BITFIELD) 278 __u8 needs_whiteout:1, 279 format:7; 280 #endif 281 282 /* Type of the value */ 283 __u8 type; 284 __u8 key_start[0]; 285 286 /* 287 * We copy bkeys with struct assignment in various places, and while 288 * that shouldn't be done with packed bkeys we can't disallow it in C, 289 * and it's legal to cast a bkey to a bkey_packed - so padding it out 290 * to the same size as struct bkey should hopefully be safest. 291 */ 292 __u8 pad[sizeof(struct bkey) - 3]; 293 } __packed __aligned(8); 294 295 typedef struct { 296 __le64 lo; 297 __le64 hi; 298 } bch_le128; 299 300 #define BKEY_U64s (sizeof(struct bkey) / sizeof(__u64)) 301 #define BKEY_U64s_MAX U8_MAX 302 #define BKEY_VAL_U64s_MAX (BKEY_U64s_MAX - BKEY_U64s) 303 304 #define KEY_PACKED_BITS_START 24 305 306 #define KEY_FORMAT_LOCAL_BTREE 0 307 #define KEY_FORMAT_CURRENT 1 308 309 enum bch_bkey_fields { 310 BKEY_FIELD_INODE, 311 BKEY_FIELD_OFFSET, 312 BKEY_FIELD_SNAPSHOT, 313 BKEY_FIELD_SIZE, 314 BKEY_FIELD_VERSION_HI, 315 BKEY_FIELD_VERSION_LO, 316 BKEY_NR_FIELDS, 317 }; 318 319 #define bkey_format_field(name, field) \ 320 [BKEY_FIELD_##name] = (sizeof(((struct bkey *) NULL)->field) * 8) 321 322 #define BKEY_FORMAT_CURRENT \ 323 ((struct bkey_format) { \ 324 .key_u64s = BKEY_U64s, \ 325 .nr_fields = BKEY_NR_FIELDS, \ 326 .bits_per_field = { \ 327 bkey_format_field(INODE, p.inode), \ 328 bkey_format_field(OFFSET, p.offset), \ 329 bkey_format_field(SNAPSHOT, p.snapshot), \ 330 bkey_format_field(SIZE, size), \ 331 bkey_format_field(VERSION_HI, version.hi), \ 332 bkey_format_field(VERSION_LO, version.lo), \ 333 }, \ 334 }) 335 336 /* bkey with inline value */ 337 struct bkey_i { 338 __u64 _data[0]; 339 340 struct bkey k; 341 struct bch_val v; 342 }; 343 344 #define POS_KEY(_pos) \ 345 ((struct bkey) { \ 346 .u64s = BKEY_U64s, \ 347 .format = KEY_FORMAT_CURRENT, \ 348 .p = _pos, \ 349 }) 350 351 #define KEY(_inode, _offset, _size) \ 352 ((struct bkey) { \ 353 .u64s = BKEY_U64s, \ 354 .format = KEY_FORMAT_CURRENT, \ 355 .p = POS(_inode, _offset), \ 356 .size = _size, \ 357 }) 358 359 static inline void bkey_init(struct bkey *k) 360 { 361 *k = KEY(0, 0, 0); 362 } 363 364 #define bkey_bytes(_k) ((_k)->u64s * sizeof(__u64)) 365 366 #define __BKEY_PADDED(key, pad) \ 367 struct bkey_i key; __u64 key ## _pad[pad] 368 369 /* 370 * - DELETED keys are used internally to mark keys that should be ignored but 371 * override keys in composition order. Their version number is ignored. 372 * 373 * - DISCARDED keys indicate that the data is all 0s because it has been 374 * discarded. DISCARDs may have a version; if the version is nonzero the key 375 * will be persistent, otherwise the key will be dropped whenever the btree 376 * node is rewritten (like DELETED keys). 377 * 378 * - ERROR: any read of the data returns a read error, as the data was lost due 379 * to a failing device. Like DISCARDED keys, they can be removed (overridden) 380 * by new writes or cluster-wide GC. Node repair can also overwrite them with 381 * the same or a more recent version number, but not with an older version 382 * number. 383 * 384 * - WHITEOUT: for hash table btrees 385 */ 386 #define BCH_BKEY_TYPES() \ 387 x(deleted, 0) \ 388 x(whiteout, 1) \ 389 x(error, 2) \ 390 x(cookie, 3) \ 391 x(hash_whiteout, 4) \ 392 x(btree_ptr, 5) \ 393 x(extent, 6) \ 394 x(reservation, 7) \ 395 x(inode, 8) \ 396 x(inode_generation, 9) \ 397 x(dirent, 10) \ 398 x(xattr, 11) \ 399 x(alloc, 12) \ 400 x(quota, 13) \ 401 x(stripe, 14) \ 402 x(reflink_p, 15) \ 403 x(reflink_v, 16) \ 404 x(inline_data, 17) \ 405 x(btree_ptr_v2, 18) \ 406 x(indirect_inline_data, 19) \ 407 x(alloc_v2, 20) \ 408 x(subvolume, 21) \ 409 x(snapshot, 22) \ 410 x(inode_v2, 23) \ 411 x(alloc_v3, 24) \ 412 x(set, 25) \ 413 x(lru, 26) \ 414 x(alloc_v4, 27) \ 415 x(backpointer, 28) \ 416 x(inode_v3, 29) \ 417 x(bucket_gens, 30) \ 418 x(snapshot_tree, 31) \ 419 x(logged_op_truncate, 32) \ 420 x(logged_op_finsert, 33) 421 422 enum bch_bkey_type { 423 #define x(name, nr) KEY_TYPE_##name = nr, 424 BCH_BKEY_TYPES() 425 #undef x 426 KEY_TYPE_MAX, 427 }; 428 429 struct bch_deleted { 430 struct bch_val v; 431 }; 432 433 struct bch_whiteout { 434 struct bch_val v; 435 }; 436 437 struct bch_error { 438 struct bch_val v; 439 }; 440 441 struct bch_cookie { 442 struct bch_val v; 443 __le64 cookie; 444 }; 445 446 struct bch_hash_whiteout { 447 struct bch_val v; 448 }; 449 450 struct bch_set { 451 struct bch_val v; 452 }; 453 454 /* 128 bits, sufficient for cryptographic MACs: */ 455 struct bch_csum { 456 __le64 lo; 457 __le64 hi; 458 } __packed __aligned(8); 459 460 struct bch_backpointer { 461 struct bch_val v; 462 __u8 btree_id; 463 __u8 level; 464 __u8 data_type; 465 __u64 bucket_offset:40; 466 __u32 bucket_len; 467 struct bpos pos; 468 } __packed __aligned(8); 469 470 /* LRU btree: */ 471 472 struct bch_lru { 473 struct bch_val v; 474 __le64 idx; 475 } __packed __aligned(8); 476 477 #define LRU_ID_STRIPES (1U << 16) 478 479 #define LRU_TIME_BITS 48 480 #define LRU_TIME_MAX ((1ULL << LRU_TIME_BITS) - 1) 481 482 /* Optional/variable size superblock sections: */ 483 484 struct bch_sb_field { 485 __u64 _data[0]; 486 __le32 u64s; 487 __le32 type; 488 }; 489 490 #define BCH_SB_FIELDS() \ 491 x(journal, 0) \ 492 x(members_v1, 1) \ 493 x(crypt, 2) \ 494 x(replicas_v0, 3) \ 495 x(quota, 4) \ 496 x(disk_groups, 5) \ 497 x(clean, 6) \ 498 x(replicas, 7) \ 499 x(journal_seq_blacklist, 8) \ 500 x(journal_v2, 9) \ 501 x(counters, 10) \ 502 x(members_v2, 11) \ 503 x(errors, 12) \ 504 x(ext, 13) \ 505 x(downgrade, 14) 506 507 #include "alloc_background_format.h" 508 #include "extents_format.h" 509 #include "ec_format.h" 510 #include "dirent_format.h" 511 #include "disk_groups_format.h" 512 #include "inode_format.h" 513 #include "journal_seq_blacklist_format.h" 514 #include "logged_ops_format.h" 515 #include "quota_format.h" 516 #include "reflink_format.h" 517 #include "replicas_format.h" 518 #include "snapshot_format.h" 519 #include "subvolume_format.h" 520 #include "sb-counters_format.h" 521 #include "sb-downgrade_format.h" 522 #include "sb-errors_format.h" 523 #include "sb-members_format.h" 524 #include "xattr_format.h" 525 526 enum bch_sb_field_type { 527 #define x(f, nr) BCH_SB_FIELD_##f = nr, 528 BCH_SB_FIELDS() 529 #undef x 530 BCH_SB_FIELD_NR 531 }; 532 533 /* 534 * Most superblock fields are replicated in all device's superblocks - a few are 535 * not: 536 */ 537 #define BCH_SINGLE_DEVICE_SB_FIELDS \ 538 ((1U << BCH_SB_FIELD_journal)| \ 539 (1U << BCH_SB_FIELD_journal_v2)) 540 541 /* BCH_SB_FIELD_journal: */ 542 543 struct bch_sb_field_journal { 544 struct bch_sb_field field; 545 __le64 buckets[]; 546 }; 547 548 struct bch_sb_field_journal_v2 { 549 struct bch_sb_field field; 550 551 struct bch_sb_field_journal_v2_entry { 552 __le64 start; 553 __le64 nr; 554 } d[]; 555 }; 556 557 /* BCH_SB_FIELD_crypt: */ 558 559 struct nonce { 560 __le32 d[4]; 561 }; 562 563 struct bch_key { 564 __le64 key[4]; 565 }; 566 567 #define BCH_KEY_MAGIC \ 568 (((__u64) 'b' << 0)|((__u64) 'c' << 8)| \ 569 ((__u64) 'h' << 16)|((__u64) '*' << 24)| \ 570 ((__u64) '*' << 32)|((__u64) 'k' << 40)| \ 571 ((__u64) 'e' << 48)|((__u64) 'y' << 56)) 572 573 struct bch_encrypted_key { 574 __le64 magic; 575 struct bch_key key; 576 }; 577 578 /* 579 * If this field is present in the superblock, it stores an encryption key which 580 * is used encrypt all other data/metadata. The key will normally be encrypted 581 * with the key userspace provides, but if encryption has been turned off we'll 582 * just store the master key unencrypted in the superblock so we can access the 583 * previously encrypted data. 584 */ 585 struct bch_sb_field_crypt { 586 struct bch_sb_field field; 587 588 __le64 flags; 589 __le64 kdf_flags; 590 struct bch_encrypted_key key; 591 }; 592 593 LE64_BITMASK(BCH_CRYPT_KDF_TYPE, struct bch_sb_field_crypt, flags, 0, 4); 594 595 enum bch_kdf_types { 596 BCH_KDF_SCRYPT = 0, 597 BCH_KDF_NR = 1, 598 }; 599 600 /* stored as base 2 log of scrypt params: */ 601 LE64_BITMASK(BCH_KDF_SCRYPT_N, struct bch_sb_field_crypt, kdf_flags, 0, 16); 602 LE64_BITMASK(BCH_KDF_SCRYPT_R, struct bch_sb_field_crypt, kdf_flags, 16, 32); 603 LE64_BITMASK(BCH_KDF_SCRYPT_P, struct bch_sb_field_crypt, kdf_flags, 32, 48); 604 605 #define BCH_DATA_TYPES() \ 606 x(free, 0) \ 607 x(sb, 1) \ 608 x(journal, 2) \ 609 x(btree, 3) \ 610 x(user, 4) \ 611 x(cached, 5) \ 612 x(parity, 6) \ 613 x(stripe, 7) \ 614 x(need_gc_gens, 8) \ 615 x(need_discard, 9) 616 617 enum bch_data_type { 618 #define x(t, n) BCH_DATA_##t, 619 BCH_DATA_TYPES() 620 #undef x 621 BCH_DATA_NR 622 }; 623 624 static inline bool data_type_is_empty(enum bch_data_type type) 625 { 626 switch (type) { 627 case BCH_DATA_free: 628 case BCH_DATA_need_gc_gens: 629 case BCH_DATA_need_discard: 630 return true; 631 default: 632 return false; 633 } 634 } 635 636 static inline bool data_type_is_hidden(enum bch_data_type type) 637 { 638 switch (type) { 639 case BCH_DATA_sb: 640 case BCH_DATA_journal: 641 return true; 642 default: 643 return false; 644 } 645 } 646 647 /* 648 * On clean shutdown, store btree roots and current journal sequence number in 649 * the superblock: 650 */ 651 struct jset_entry { 652 __le16 u64s; 653 __u8 btree_id; 654 __u8 level; 655 __u8 type; /* designates what this jset holds */ 656 __u8 pad[3]; 657 658 struct bkey_i start[0]; 659 __u64 _data[]; 660 }; 661 662 struct bch_sb_field_clean { 663 struct bch_sb_field field; 664 665 __le32 flags; 666 __le16 _read_clock; /* no longer used */ 667 __le16 _write_clock; 668 __le64 journal_seq; 669 670 struct jset_entry start[0]; 671 __u64 _data[]; 672 }; 673 674 struct bch_sb_field_ext { 675 struct bch_sb_field field; 676 __le64 recovery_passes_required[2]; 677 __le64 errors_silent[8]; 678 __le64 btrees_lost_data; 679 }; 680 681 /* Superblock: */ 682 683 /* 684 * New versioning scheme: 685 * One common version number for all on disk data structures - superblock, btree 686 * nodes, journal entries 687 */ 688 #define BCH_VERSION_MAJOR(_v) ((__u16) ((_v) >> 10)) 689 #define BCH_VERSION_MINOR(_v) ((__u16) ((_v) & ~(~0U << 10))) 690 #define BCH_VERSION(_major, _minor) (((_major) << 10)|(_minor) << 0) 691 692 /* 693 * field 1: version name 694 * field 2: BCH_VERSION(major, minor) 695 * field 3: recovery passess required on upgrade 696 */ 697 #define BCH_METADATA_VERSIONS() \ 698 x(bkey_renumber, BCH_VERSION(0, 10)) \ 699 x(inode_btree_change, BCH_VERSION(0, 11)) \ 700 x(snapshot, BCH_VERSION(0, 12)) \ 701 x(inode_backpointers, BCH_VERSION(0, 13)) \ 702 x(btree_ptr_sectors_written, BCH_VERSION(0, 14)) \ 703 x(snapshot_2, BCH_VERSION(0, 15)) \ 704 x(reflink_p_fix, BCH_VERSION(0, 16)) \ 705 x(subvol_dirent, BCH_VERSION(0, 17)) \ 706 x(inode_v2, BCH_VERSION(0, 18)) \ 707 x(freespace, BCH_VERSION(0, 19)) \ 708 x(alloc_v4, BCH_VERSION(0, 20)) \ 709 x(new_data_types, BCH_VERSION(0, 21)) \ 710 x(backpointers, BCH_VERSION(0, 22)) \ 711 x(inode_v3, BCH_VERSION(0, 23)) \ 712 x(unwritten_extents, BCH_VERSION(0, 24)) \ 713 x(bucket_gens, BCH_VERSION(0, 25)) \ 714 x(lru_v2, BCH_VERSION(0, 26)) \ 715 x(fragmentation_lru, BCH_VERSION(0, 27)) \ 716 x(no_bps_in_alloc_keys, BCH_VERSION(0, 28)) \ 717 x(snapshot_trees, BCH_VERSION(0, 29)) \ 718 x(major_minor, BCH_VERSION(1, 0)) \ 719 x(snapshot_skiplists, BCH_VERSION(1, 1)) \ 720 x(deleted_inodes, BCH_VERSION(1, 2)) \ 721 x(rebalance_work, BCH_VERSION(1, 3)) \ 722 x(member_seq, BCH_VERSION(1, 4)) \ 723 x(subvolume_fs_parent, BCH_VERSION(1, 5)) \ 724 x(btree_subvolume_children, BCH_VERSION(1, 6)) \ 725 x(mi_btree_bitmap, BCH_VERSION(1, 7)) 726 727 enum bcachefs_metadata_version { 728 bcachefs_metadata_version_min = 9, 729 #define x(t, n) bcachefs_metadata_version_##t = n, 730 BCH_METADATA_VERSIONS() 731 #undef x 732 bcachefs_metadata_version_max 733 }; 734 735 static const __maybe_unused 736 unsigned bcachefs_metadata_required_upgrade_below = bcachefs_metadata_version_rebalance_work; 737 738 #define bcachefs_metadata_version_current (bcachefs_metadata_version_max - 1) 739 740 #define BCH_SB_SECTOR 8 741 742 #define BCH_SB_LAYOUT_SIZE_BITS_MAX 16 /* 32 MB */ 743 744 struct bch_sb_layout { 745 __uuid_t magic; /* bcachefs superblock UUID */ 746 __u8 layout_type; 747 __u8 sb_max_size_bits; /* base 2 of 512 byte sectors */ 748 __u8 nr_superblocks; 749 __u8 pad[5]; 750 __le64 sb_offset[61]; 751 } __packed __aligned(8); 752 753 #define BCH_SB_LAYOUT_SECTOR 7 754 755 /* 756 * @offset - sector where this sb was written 757 * @version - on disk format version 758 * @version_min - Oldest metadata version this filesystem contains; so we can 759 * safely drop compatibility code and refuse to mount filesystems 760 * we'd need it for 761 * @magic - identifies as a bcachefs superblock (BCHFS_MAGIC) 762 * @seq - incremented each time superblock is written 763 * @uuid - used for generating various magic numbers and identifying 764 * member devices, never changes 765 * @user_uuid - user visible UUID, may be changed 766 * @label - filesystem label 767 * @seq - identifies most recent superblock, incremented each time 768 * superblock is written 769 * @features - enabled incompatible features 770 */ 771 struct bch_sb { 772 struct bch_csum csum; 773 __le16 version; 774 __le16 version_min; 775 __le16 pad[2]; 776 __uuid_t magic; 777 __uuid_t uuid; 778 __uuid_t user_uuid; 779 __u8 label[BCH_SB_LABEL_SIZE]; 780 __le64 offset; 781 __le64 seq; 782 783 __le16 block_size; 784 __u8 dev_idx; 785 __u8 nr_devices; 786 __le32 u64s; 787 788 __le64 time_base_lo; 789 __le32 time_base_hi; 790 __le32 time_precision; 791 792 __le64 flags[7]; 793 __le64 write_time; 794 __le64 features[2]; 795 __le64 compat[2]; 796 797 struct bch_sb_layout layout; 798 799 struct bch_sb_field start[0]; 800 __le64 _data[]; 801 } __packed __aligned(8); 802 803 /* 804 * Flags: 805 * BCH_SB_INITALIZED - set on first mount 806 * BCH_SB_CLEAN - did we shut down cleanly? Just a hint, doesn't affect 807 * behaviour of mount/recovery path: 808 * BCH_SB_INODE_32BIT - limit inode numbers to 32 bits 809 * BCH_SB_128_BIT_MACS - 128 bit macs instead of 80 810 * BCH_SB_ENCRYPTION_TYPE - if nonzero encryption is enabled; overrides 811 * DATA/META_CSUM_TYPE. Also indicates encryption 812 * algorithm in use, if/when we get more than one 813 */ 814 815 LE16_BITMASK(BCH_SB_BLOCK_SIZE, struct bch_sb, block_size, 0, 16); 816 817 LE64_BITMASK(BCH_SB_INITIALIZED, struct bch_sb, flags[0], 0, 1); 818 LE64_BITMASK(BCH_SB_CLEAN, struct bch_sb, flags[0], 1, 2); 819 LE64_BITMASK(BCH_SB_CSUM_TYPE, struct bch_sb, flags[0], 2, 8); 820 LE64_BITMASK(BCH_SB_ERROR_ACTION, struct bch_sb, flags[0], 8, 12); 821 822 LE64_BITMASK(BCH_SB_BTREE_NODE_SIZE, struct bch_sb, flags[0], 12, 28); 823 824 LE64_BITMASK(BCH_SB_GC_RESERVE, struct bch_sb, flags[0], 28, 33); 825 LE64_BITMASK(BCH_SB_ROOT_RESERVE, struct bch_sb, flags[0], 33, 40); 826 827 LE64_BITMASK(BCH_SB_META_CSUM_TYPE, struct bch_sb, flags[0], 40, 44); 828 LE64_BITMASK(BCH_SB_DATA_CSUM_TYPE, struct bch_sb, flags[0], 44, 48); 829 830 LE64_BITMASK(BCH_SB_META_REPLICAS_WANT, struct bch_sb, flags[0], 48, 52); 831 LE64_BITMASK(BCH_SB_DATA_REPLICAS_WANT, struct bch_sb, flags[0], 52, 56); 832 833 LE64_BITMASK(BCH_SB_POSIX_ACL, struct bch_sb, flags[0], 56, 57); 834 LE64_BITMASK(BCH_SB_USRQUOTA, struct bch_sb, flags[0], 57, 58); 835 LE64_BITMASK(BCH_SB_GRPQUOTA, struct bch_sb, flags[0], 58, 59); 836 LE64_BITMASK(BCH_SB_PRJQUOTA, struct bch_sb, flags[0], 59, 60); 837 838 LE64_BITMASK(BCH_SB_HAS_ERRORS, struct bch_sb, flags[0], 60, 61); 839 LE64_BITMASK(BCH_SB_HAS_TOPOLOGY_ERRORS,struct bch_sb, flags[0], 61, 62); 840 841 LE64_BITMASK(BCH_SB_BIG_ENDIAN, struct bch_sb, flags[0], 62, 63); 842 843 LE64_BITMASK(BCH_SB_STR_HASH_TYPE, struct bch_sb, flags[1], 0, 4); 844 LE64_BITMASK(BCH_SB_COMPRESSION_TYPE_LO,struct bch_sb, flags[1], 4, 8); 845 LE64_BITMASK(BCH_SB_INODE_32BIT, struct bch_sb, flags[1], 8, 9); 846 847 LE64_BITMASK(BCH_SB_128_BIT_MACS, struct bch_sb, flags[1], 9, 10); 848 LE64_BITMASK(BCH_SB_ENCRYPTION_TYPE, struct bch_sb, flags[1], 10, 14); 849 850 /* 851 * Max size of an extent that may require bouncing to read or write 852 * (checksummed, compressed): 64k 853 */ 854 LE64_BITMASK(BCH_SB_ENCODED_EXTENT_MAX_BITS, 855 struct bch_sb, flags[1], 14, 20); 856 857 LE64_BITMASK(BCH_SB_META_REPLICAS_REQ, struct bch_sb, flags[1], 20, 24); 858 LE64_BITMASK(BCH_SB_DATA_REPLICAS_REQ, struct bch_sb, flags[1], 24, 28); 859 860 LE64_BITMASK(BCH_SB_PROMOTE_TARGET, struct bch_sb, flags[1], 28, 40); 861 LE64_BITMASK(BCH_SB_FOREGROUND_TARGET, struct bch_sb, flags[1], 40, 52); 862 LE64_BITMASK(BCH_SB_BACKGROUND_TARGET, struct bch_sb, flags[1], 52, 64); 863 864 LE64_BITMASK(BCH_SB_BACKGROUND_COMPRESSION_TYPE_LO, 865 struct bch_sb, flags[2], 0, 4); 866 LE64_BITMASK(BCH_SB_GC_RESERVE_BYTES, struct bch_sb, flags[2], 4, 64); 867 868 LE64_BITMASK(BCH_SB_ERASURE_CODE, struct bch_sb, flags[3], 0, 16); 869 LE64_BITMASK(BCH_SB_METADATA_TARGET, struct bch_sb, flags[3], 16, 28); 870 LE64_BITMASK(BCH_SB_SHARD_INUMS, struct bch_sb, flags[3], 28, 29); 871 LE64_BITMASK(BCH_SB_INODES_USE_KEY_CACHE,struct bch_sb, flags[3], 29, 30); 872 LE64_BITMASK(BCH_SB_JOURNAL_FLUSH_DELAY,struct bch_sb, flags[3], 30, 62); 873 LE64_BITMASK(BCH_SB_JOURNAL_FLUSH_DISABLED,struct bch_sb, flags[3], 62, 63); 874 LE64_BITMASK(BCH_SB_JOURNAL_RECLAIM_DELAY,struct bch_sb, flags[4], 0, 32); 875 LE64_BITMASK(BCH_SB_JOURNAL_TRANSACTION_NAMES,struct bch_sb, flags[4], 32, 33); 876 LE64_BITMASK(BCH_SB_NOCOW, struct bch_sb, flags[4], 33, 34); 877 LE64_BITMASK(BCH_SB_WRITE_BUFFER_SIZE, struct bch_sb, flags[4], 34, 54); 878 LE64_BITMASK(BCH_SB_VERSION_UPGRADE, struct bch_sb, flags[4], 54, 56); 879 880 LE64_BITMASK(BCH_SB_COMPRESSION_TYPE_HI,struct bch_sb, flags[4], 56, 60); 881 LE64_BITMASK(BCH_SB_BACKGROUND_COMPRESSION_TYPE_HI, 882 struct bch_sb, flags[4], 60, 64); 883 884 LE64_BITMASK(BCH_SB_VERSION_UPGRADE_COMPLETE, 885 struct bch_sb, flags[5], 0, 16); 886 887 static inline __u64 BCH_SB_COMPRESSION_TYPE(const struct bch_sb *sb) 888 { 889 return BCH_SB_COMPRESSION_TYPE_LO(sb) | (BCH_SB_COMPRESSION_TYPE_HI(sb) << 4); 890 } 891 892 static inline void SET_BCH_SB_COMPRESSION_TYPE(struct bch_sb *sb, __u64 v) 893 { 894 SET_BCH_SB_COMPRESSION_TYPE_LO(sb, v); 895 SET_BCH_SB_COMPRESSION_TYPE_HI(sb, v >> 4); 896 } 897 898 static inline __u64 BCH_SB_BACKGROUND_COMPRESSION_TYPE(const struct bch_sb *sb) 899 { 900 return BCH_SB_BACKGROUND_COMPRESSION_TYPE_LO(sb) | 901 (BCH_SB_BACKGROUND_COMPRESSION_TYPE_HI(sb) << 4); 902 } 903 904 static inline void SET_BCH_SB_BACKGROUND_COMPRESSION_TYPE(struct bch_sb *sb, __u64 v) 905 { 906 SET_BCH_SB_BACKGROUND_COMPRESSION_TYPE_LO(sb, v); 907 SET_BCH_SB_BACKGROUND_COMPRESSION_TYPE_HI(sb, v >> 4); 908 } 909 910 /* 911 * Features: 912 * 913 * journal_seq_blacklist_v3: gates BCH_SB_FIELD_journal_seq_blacklist 914 * reflink: gates KEY_TYPE_reflink 915 * inline_data: gates KEY_TYPE_inline_data 916 * new_siphash: gates BCH_STR_HASH_siphash 917 * new_extent_overwrite: gates BTREE_NODE_NEW_EXTENT_OVERWRITE 918 */ 919 #define BCH_SB_FEATURES() \ 920 x(lz4, 0) \ 921 x(gzip, 1) \ 922 x(zstd, 2) \ 923 x(atomic_nlink, 3) \ 924 x(ec, 4) \ 925 x(journal_seq_blacklist_v3, 5) \ 926 x(reflink, 6) \ 927 x(new_siphash, 7) \ 928 x(inline_data, 8) \ 929 x(new_extent_overwrite, 9) \ 930 x(incompressible, 10) \ 931 x(btree_ptr_v2, 11) \ 932 x(extents_above_btree_updates, 12) \ 933 x(btree_updates_journalled, 13) \ 934 x(reflink_inline_data, 14) \ 935 x(new_varint, 15) \ 936 x(journal_no_flush, 16) \ 937 x(alloc_v2, 17) \ 938 x(extents_across_btree_nodes, 18) 939 940 #define BCH_SB_FEATURES_ALWAYS \ 941 ((1ULL << BCH_FEATURE_new_extent_overwrite)| \ 942 (1ULL << BCH_FEATURE_extents_above_btree_updates)|\ 943 (1ULL << BCH_FEATURE_btree_updates_journalled)|\ 944 (1ULL << BCH_FEATURE_alloc_v2)|\ 945 (1ULL << BCH_FEATURE_extents_across_btree_nodes)) 946 947 #define BCH_SB_FEATURES_ALL \ 948 (BCH_SB_FEATURES_ALWAYS| \ 949 (1ULL << BCH_FEATURE_new_siphash)| \ 950 (1ULL << BCH_FEATURE_btree_ptr_v2)| \ 951 (1ULL << BCH_FEATURE_new_varint)| \ 952 (1ULL << BCH_FEATURE_journal_no_flush)) 953 954 enum bch_sb_feature { 955 #define x(f, n) BCH_FEATURE_##f, 956 BCH_SB_FEATURES() 957 #undef x 958 BCH_FEATURE_NR, 959 }; 960 961 #define BCH_SB_COMPAT() \ 962 x(alloc_info, 0) \ 963 x(alloc_metadata, 1) \ 964 x(extents_above_btree_updates_done, 2) \ 965 x(bformat_overflow_done, 3) 966 967 enum bch_sb_compat { 968 #define x(f, n) BCH_COMPAT_##f, 969 BCH_SB_COMPAT() 970 #undef x 971 BCH_COMPAT_NR, 972 }; 973 974 /* options: */ 975 976 #define BCH_VERSION_UPGRADE_OPTS() \ 977 x(compatible, 0) \ 978 x(incompatible, 1) \ 979 x(none, 2) 980 981 enum bch_version_upgrade_opts { 982 #define x(t, n) BCH_VERSION_UPGRADE_##t = n, 983 BCH_VERSION_UPGRADE_OPTS() 984 #undef x 985 }; 986 987 #define BCH_REPLICAS_MAX 4U 988 989 #define BCH_BKEY_PTRS_MAX 16U 990 991 #define BCH_ERROR_ACTIONS() \ 992 x(continue, 0) \ 993 x(fix_safe, 1) \ 994 x(panic, 2) \ 995 x(ro, 3) 996 997 enum bch_error_actions { 998 #define x(t, n) BCH_ON_ERROR_##t = n, 999 BCH_ERROR_ACTIONS() 1000 #undef x 1001 BCH_ON_ERROR_NR 1002 }; 1003 1004 #define BCH_STR_HASH_TYPES() \ 1005 x(crc32c, 0) \ 1006 x(crc64, 1) \ 1007 x(siphash_old, 2) \ 1008 x(siphash, 3) 1009 1010 enum bch_str_hash_type { 1011 #define x(t, n) BCH_STR_HASH_##t = n, 1012 BCH_STR_HASH_TYPES() 1013 #undef x 1014 BCH_STR_HASH_NR 1015 }; 1016 1017 #define BCH_STR_HASH_OPTS() \ 1018 x(crc32c, 0) \ 1019 x(crc64, 1) \ 1020 x(siphash, 2) 1021 1022 enum bch_str_hash_opts { 1023 #define x(t, n) BCH_STR_HASH_OPT_##t = n, 1024 BCH_STR_HASH_OPTS() 1025 #undef x 1026 BCH_STR_HASH_OPT_NR 1027 }; 1028 1029 #define BCH_CSUM_TYPES() \ 1030 x(none, 0) \ 1031 x(crc32c_nonzero, 1) \ 1032 x(crc64_nonzero, 2) \ 1033 x(chacha20_poly1305_80, 3) \ 1034 x(chacha20_poly1305_128, 4) \ 1035 x(crc32c, 5) \ 1036 x(crc64, 6) \ 1037 x(xxhash, 7) 1038 1039 enum bch_csum_type { 1040 #define x(t, n) BCH_CSUM_##t = n, 1041 BCH_CSUM_TYPES() 1042 #undef x 1043 BCH_CSUM_NR 1044 }; 1045 1046 static const __maybe_unused unsigned bch_crc_bytes[] = { 1047 [BCH_CSUM_none] = 0, 1048 [BCH_CSUM_crc32c_nonzero] = 4, 1049 [BCH_CSUM_crc32c] = 4, 1050 [BCH_CSUM_crc64_nonzero] = 8, 1051 [BCH_CSUM_crc64] = 8, 1052 [BCH_CSUM_xxhash] = 8, 1053 [BCH_CSUM_chacha20_poly1305_80] = 10, 1054 [BCH_CSUM_chacha20_poly1305_128] = 16, 1055 }; 1056 1057 static inline _Bool bch2_csum_type_is_encryption(enum bch_csum_type type) 1058 { 1059 switch (type) { 1060 case BCH_CSUM_chacha20_poly1305_80: 1061 case BCH_CSUM_chacha20_poly1305_128: 1062 return true; 1063 default: 1064 return false; 1065 } 1066 } 1067 1068 #define BCH_CSUM_OPTS() \ 1069 x(none, 0) \ 1070 x(crc32c, 1) \ 1071 x(crc64, 2) \ 1072 x(xxhash, 3) 1073 1074 enum bch_csum_opts { 1075 #define x(t, n) BCH_CSUM_OPT_##t = n, 1076 BCH_CSUM_OPTS() 1077 #undef x 1078 BCH_CSUM_OPT_NR 1079 }; 1080 1081 #define BCH_COMPRESSION_TYPES() \ 1082 x(none, 0) \ 1083 x(lz4_old, 1) \ 1084 x(gzip, 2) \ 1085 x(lz4, 3) \ 1086 x(zstd, 4) \ 1087 x(incompressible, 5) 1088 1089 enum bch_compression_type { 1090 #define x(t, n) BCH_COMPRESSION_TYPE_##t = n, 1091 BCH_COMPRESSION_TYPES() 1092 #undef x 1093 BCH_COMPRESSION_TYPE_NR 1094 }; 1095 1096 #define BCH_COMPRESSION_OPTS() \ 1097 x(none, 0) \ 1098 x(lz4, 1) \ 1099 x(gzip, 2) \ 1100 x(zstd, 3) 1101 1102 enum bch_compression_opts { 1103 #define x(t, n) BCH_COMPRESSION_OPT_##t = n, 1104 BCH_COMPRESSION_OPTS() 1105 #undef x 1106 BCH_COMPRESSION_OPT_NR 1107 }; 1108 1109 /* 1110 * Magic numbers 1111 * 1112 * The various other data structures have their own magic numbers, which are 1113 * xored with the first part of the cache set's UUID 1114 */ 1115 1116 #define BCACHE_MAGIC \ 1117 UUID_INIT(0xc68573f6, 0x4e1a, 0x45ca, \ 1118 0x82, 0x65, 0xf5, 0x7f, 0x48, 0xba, 0x6d, 0x81) 1119 #define BCHFS_MAGIC \ 1120 UUID_INIT(0xc68573f6, 0x66ce, 0x90a9, \ 1121 0xd9, 0x6a, 0x60, 0xcf, 0x80, 0x3d, 0xf7, 0xef) 1122 1123 #define BCACHEFS_STATFS_MAGIC BCACHEFS_SUPER_MAGIC 1124 1125 #define JSET_MAGIC __cpu_to_le64(0x245235c1a3625032ULL) 1126 #define BSET_MAGIC __cpu_to_le64(0x90135c78b99e07f5ULL) 1127 1128 static inline __le64 __bch2_sb_magic(struct bch_sb *sb) 1129 { 1130 __le64 ret; 1131 1132 memcpy(&ret, &sb->uuid, sizeof(ret)); 1133 return ret; 1134 } 1135 1136 static inline __u64 __jset_magic(struct bch_sb *sb) 1137 { 1138 return __le64_to_cpu(__bch2_sb_magic(sb) ^ JSET_MAGIC); 1139 } 1140 1141 static inline __u64 __bset_magic(struct bch_sb *sb) 1142 { 1143 return __le64_to_cpu(__bch2_sb_magic(sb) ^ BSET_MAGIC); 1144 } 1145 1146 /* Journal */ 1147 1148 #define JSET_KEYS_U64s (sizeof(struct jset_entry) / sizeof(__u64)) 1149 1150 #define BCH_JSET_ENTRY_TYPES() \ 1151 x(btree_keys, 0) \ 1152 x(btree_root, 1) \ 1153 x(prio_ptrs, 2) \ 1154 x(blacklist, 3) \ 1155 x(blacklist_v2, 4) \ 1156 x(usage, 5) \ 1157 x(data_usage, 6) \ 1158 x(clock, 7) \ 1159 x(dev_usage, 8) \ 1160 x(log, 9) \ 1161 x(overwrite, 10) \ 1162 x(write_buffer_keys, 11) \ 1163 x(datetime, 12) 1164 1165 enum bch_jset_entry_type { 1166 #define x(f, nr) BCH_JSET_ENTRY_##f = nr, 1167 BCH_JSET_ENTRY_TYPES() 1168 #undef x 1169 BCH_JSET_ENTRY_NR 1170 }; 1171 1172 static inline bool jset_entry_is_key(struct jset_entry *e) 1173 { 1174 switch (e->type) { 1175 case BCH_JSET_ENTRY_btree_keys: 1176 case BCH_JSET_ENTRY_btree_root: 1177 case BCH_JSET_ENTRY_overwrite: 1178 case BCH_JSET_ENTRY_write_buffer_keys: 1179 return true; 1180 } 1181 1182 return false; 1183 } 1184 1185 /* 1186 * Journal sequence numbers can be blacklisted: bsets record the max sequence 1187 * number of all the journal entries they contain updates for, so that on 1188 * recovery we can ignore those bsets that contain index updates newer that what 1189 * made it into the journal. 1190 * 1191 * This means that we can't reuse that journal_seq - we have to skip it, and 1192 * then record that we skipped it so that the next time we crash and recover we 1193 * don't think there was a missing journal entry. 1194 */ 1195 struct jset_entry_blacklist { 1196 struct jset_entry entry; 1197 __le64 seq; 1198 }; 1199 1200 struct jset_entry_blacklist_v2 { 1201 struct jset_entry entry; 1202 __le64 start; 1203 __le64 end; 1204 }; 1205 1206 #define BCH_FS_USAGE_TYPES() \ 1207 x(reserved, 0) \ 1208 x(inodes, 1) \ 1209 x(key_version, 2) 1210 1211 enum bch_fs_usage_type { 1212 #define x(f, nr) BCH_FS_USAGE_##f = nr, 1213 BCH_FS_USAGE_TYPES() 1214 #undef x 1215 BCH_FS_USAGE_NR 1216 }; 1217 1218 struct jset_entry_usage { 1219 struct jset_entry entry; 1220 __le64 v; 1221 } __packed; 1222 1223 struct jset_entry_data_usage { 1224 struct jset_entry entry; 1225 __le64 v; 1226 struct bch_replicas_entry_v1 r; 1227 } __packed; 1228 1229 struct jset_entry_clock { 1230 struct jset_entry entry; 1231 __u8 rw; 1232 __u8 pad[7]; 1233 __le64 time; 1234 } __packed; 1235 1236 struct jset_entry_dev_usage_type { 1237 __le64 buckets; 1238 __le64 sectors; 1239 __le64 fragmented; 1240 } __packed; 1241 1242 struct jset_entry_dev_usage { 1243 struct jset_entry entry; 1244 __le32 dev; 1245 __u32 pad; 1246 1247 __le64 _buckets_ec; /* No longer used */ 1248 __le64 _buckets_unavailable; /* No longer used */ 1249 1250 struct jset_entry_dev_usage_type d[]; 1251 }; 1252 1253 static inline unsigned jset_entry_dev_usage_nr_types(struct jset_entry_dev_usage *u) 1254 { 1255 return (vstruct_bytes(&u->entry) - sizeof(struct jset_entry_dev_usage)) / 1256 sizeof(struct jset_entry_dev_usage_type); 1257 } 1258 1259 struct jset_entry_log { 1260 struct jset_entry entry; 1261 u8 d[]; 1262 } __packed __aligned(8); 1263 1264 struct jset_entry_datetime { 1265 struct jset_entry entry; 1266 __le64 seconds; 1267 } __packed __aligned(8); 1268 1269 /* 1270 * On disk format for a journal entry: 1271 * seq is monotonically increasing; every journal entry has its own unique 1272 * sequence number. 1273 * 1274 * last_seq is the oldest journal entry that still has keys the btree hasn't 1275 * flushed to disk yet. 1276 * 1277 * version is for on disk format changes. 1278 */ 1279 struct jset { 1280 struct bch_csum csum; 1281 1282 __le64 magic; 1283 __le64 seq; 1284 __le32 version; 1285 __le32 flags; 1286 1287 __le32 u64s; /* size of d[] in u64s */ 1288 1289 __u8 encrypted_start[0]; 1290 1291 __le16 _read_clock; /* no longer used */ 1292 __le16 _write_clock; 1293 1294 /* Sequence number of oldest dirty journal entry */ 1295 __le64 last_seq; 1296 1297 1298 struct jset_entry start[0]; 1299 __u64 _data[]; 1300 } __packed __aligned(8); 1301 1302 LE32_BITMASK(JSET_CSUM_TYPE, struct jset, flags, 0, 4); 1303 LE32_BITMASK(JSET_BIG_ENDIAN, struct jset, flags, 4, 5); 1304 LE32_BITMASK(JSET_NO_FLUSH, struct jset, flags, 5, 6); 1305 1306 #define BCH_JOURNAL_BUCKETS_MIN 8 1307 1308 /* Btree: */ 1309 1310 enum btree_id_flags { 1311 BTREE_ID_EXTENTS = BIT(0), 1312 BTREE_ID_SNAPSHOTS = BIT(1), 1313 BTREE_ID_SNAPSHOT_FIELD = BIT(2), 1314 BTREE_ID_DATA = BIT(3), 1315 }; 1316 1317 #define BCH_BTREE_IDS() \ 1318 x(extents, 0, BTREE_ID_EXTENTS|BTREE_ID_SNAPSHOTS|BTREE_ID_DATA,\ 1319 BIT_ULL(KEY_TYPE_whiteout)| \ 1320 BIT_ULL(KEY_TYPE_error)| \ 1321 BIT_ULL(KEY_TYPE_cookie)| \ 1322 BIT_ULL(KEY_TYPE_extent)| \ 1323 BIT_ULL(KEY_TYPE_reservation)| \ 1324 BIT_ULL(KEY_TYPE_reflink_p)| \ 1325 BIT_ULL(KEY_TYPE_inline_data)) \ 1326 x(inodes, 1, BTREE_ID_SNAPSHOTS, \ 1327 BIT_ULL(KEY_TYPE_whiteout)| \ 1328 BIT_ULL(KEY_TYPE_inode)| \ 1329 BIT_ULL(KEY_TYPE_inode_v2)| \ 1330 BIT_ULL(KEY_TYPE_inode_v3)| \ 1331 BIT_ULL(KEY_TYPE_inode_generation)) \ 1332 x(dirents, 2, BTREE_ID_SNAPSHOTS, \ 1333 BIT_ULL(KEY_TYPE_whiteout)| \ 1334 BIT_ULL(KEY_TYPE_hash_whiteout)| \ 1335 BIT_ULL(KEY_TYPE_dirent)) \ 1336 x(xattrs, 3, BTREE_ID_SNAPSHOTS, \ 1337 BIT_ULL(KEY_TYPE_whiteout)| \ 1338 BIT_ULL(KEY_TYPE_cookie)| \ 1339 BIT_ULL(KEY_TYPE_hash_whiteout)| \ 1340 BIT_ULL(KEY_TYPE_xattr)) \ 1341 x(alloc, 4, 0, \ 1342 BIT_ULL(KEY_TYPE_alloc)| \ 1343 BIT_ULL(KEY_TYPE_alloc_v2)| \ 1344 BIT_ULL(KEY_TYPE_alloc_v3)| \ 1345 BIT_ULL(KEY_TYPE_alloc_v4)) \ 1346 x(quotas, 5, 0, \ 1347 BIT_ULL(KEY_TYPE_quota)) \ 1348 x(stripes, 6, 0, \ 1349 BIT_ULL(KEY_TYPE_stripe)) \ 1350 x(reflink, 7, BTREE_ID_EXTENTS|BTREE_ID_DATA, \ 1351 BIT_ULL(KEY_TYPE_reflink_v)| \ 1352 BIT_ULL(KEY_TYPE_indirect_inline_data)| \ 1353 BIT_ULL(KEY_TYPE_error)) \ 1354 x(subvolumes, 8, 0, \ 1355 BIT_ULL(KEY_TYPE_subvolume)) \ 1356 x(snapshots, 9, 0, \ 1357 BIT_ULL(KEY_TYPE_snapshot)) \ 1358 x(lru, 10, 0, \ 1359 BIT_ULL(KEY_TYPE_set)) \ 1360 x(freespace, 11, BTREE_ID_EXTENTS, \ 1361 BIT_ULL(KEY_TYPE_set)) \ 1362 x(need_discard, 12, 0, \ 1363 BIT_ULL(KEY_TYPE_set)) \ 1364 x(backpointers, 13, 0, \ 1365 BIT_ULL(KEY_TYPE_backpointer)) \ 1366 x(bucket_gens, 14, 0, \ 1367 BIT_ULL(KEY_TYPE_bucket_gens)) \ 1368 x(snapshot_trees, 15, 0, \ 1369 BIT_ULL(KEY_TYPE_snapshot_tree)) \ 1370 x(deleted_inodes, 16, BTREE_ID_SNAPSHOT_FIELD, \ 1371 BIT_ULL(KEY_TYPE_set)) \ 1372 x(logged_ops, 17, 0, \ 1373 BIT_ULL(KEY_TYPE_logged_op_truncate)| \ 1374 BIT_ULL(KEY_TYPE_logged_op_finsert)) \ 1375 x(rebalance_work, 18, BTREE_ID_SNAPSHOT_FIELD, \ 1376 BIT_ULL(KEY_TYPE_set)|BIT_ULL(KEY_TYPE_cookie)) \ 1377 x(subvolume_children, 19, 0, \ 1378 BIT_ULL(KEY_TYPE_set)) 1379 1380 enum btree_id { 1381 #define x(name, nr, ...) BTREE_ID_##name = nr, 1382 BCH_BTREE_IDS() 1383 #undef x 1384 BTREE_ID_NR 1385 }; 1386 1387 /* 1388 * Maximum number of btrees that we will _ever_ have under the current scheme, 1389 * where we refer to them with 64 bit bitfields - and we also need a bit for 1390 * the interior btree node type: 1391 */ 1392 #define BTREE_ID_NR_MAX 63 1393 1394 static inline bool btree_id_is_alloc(enum btree_id id) 1395 { 1396 switch (id) { 1397 case BTREE_ID_alloc: 1398 case BTREE_ID_backpointers: 1399 case BTREE_ID_need_discard: 1400 case BTREE_ID_freespace: 1401 case BTREE_ID_bucket_gens: 1402 return true; 1403 default: 1404 return false; 1405 } 1406 } 1407 1408 #define BTREE_MAX_DEPTH 4U 1409 1410 /* Btree nodes */ 1411 1412 /* 1413 * Btree nodes 1414 * 1415 * On disk a btree node is a list/log of these; within each set the keys are 1416 * sorted 1417 */ 1418 struct bset { 1419 __le64 seq; 1420 1421 /* 1422 * Highest journal entry this bset contains keys for. 1423 * If on recovery we don't see that journal entry, this bset is ignored: 1424 * this allows us to preserve the order of all index updates after a 1425 * crash, since the journal records a total order of all index updates 1426 * and anything that didn't make it to the journal doesn't get used. 1427 */ 1428 __le64 journal_seq; 1429 1430 __le32 flags; 1431 __le16 version; 1432 __le16 u64s; /* count of d[] in u64s */ 1433 1434 struct bkey_packed start[0]; 1435 __u64 _data[]; 1436 } __packed __aligned(8); 1437 1438 LE32_BITMASK(BSET_CSUM_TYPE, struct bset, flags, 0, 4); 1439 1440 LE32_BITMASK(BSET_BIG_ENDIAN, struct bset, flags, 4, 5); 1441 LE32_BITMASK(BSET_SEPARATE_WHITEOUTS, 1442 struct bset, flags, 5, 6); 1443 1444 /* Sector offset within the btree node: */ 1445 LE32_BITMASK(BSET_OFFSET, struct bset, flags, 16, 32); 1446 1447 struct btree_node { 1448 struct bch_csum csum; 1449 __le64 magic; 1450 1451 /* this flags field is encrypted, unlike bset->flags: */ 1452 __le64 flags; 1453 1454 /* Closed interval: */ 1455 struct bpos min_key; 1456 struct bpos max_key; 1457 struct bch_extent_ptr _ptr; /* not used anymore */ 1458 struct bkey_format format; 1459 1460 union { 1461 struct bset keys; 1462 struct { 1463 __u8 pad[22]; 1464 __le16 u64s; 1465 __u64 _data[0]; 1466 1467 }; 1468 }; 1469 } __packed __aligned(8); 1470 1471 LE64_BITMASK(BTREE_NODE_ID_LO, struct btree_node, flags, 0, 4); 1472 LE64_BITMASK(BTREE_NODE_LEVEL, struct btree_node, flags, 4, 8); 1473 LE64_BITMASK(BTREE_NODE_NEW_EXTENT_OVERWRITE, 1474 struct btree_node, flags, 8, 9); 1475 LE64_BITMASK(BTREE_NODE_ID_HI, struct btree_node, flags, 9, 25); 1476 /* 25-32 unused */ 1477 LE64_BITMASK(BTREE_NODE_SEQ, struct btree_node, flags, 32, 64); 1478 1479 static inline __u64 BTREE_NODE_ID(struct btree_node *n) 1480 { 1481 return BTREE_NODE_ID_LO(n) | (BTREE_NODE_ID_HI(n) << 4); 1482 } 1483 1484 static inline void SET_BTREE_NODE_ID(struct btree_node *n, __u64 v) 1485 { 1486 SET_BTREE_NODE_ID_LO(n, v); 1487 SET_BTREE_NODE_ID_HI(n, v >> 4); 1488 } 1489 1490 struct btree_node_entry { 1491 struct bch_csum csum; 1492 1493 union { 1494 struct bset keys; 1495 struct { 1496 __u8 pad[22]; 1497 __le16 u64s; 1498 __u64 _data[0]; 1499 }; 1500 }; 1501 } __packed __aligned(8); 1502 1503 #endif /* _BCACHEFS_FORMAT_H */ 1504