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 "vstructs.h" 80 81#ifdef __KERNEL__ 82typedef uuid_t __uuid_t; 83#endif 84 85#define BITMASK(name, type, field, offset, end) \ 86static const __maybe_unused unsigned name##_OFFSET = offset; \ 87static const __maybe_unused unsigned name##_BITS = (end - offset); \ 88 \ 89static inline __u64 name(const type *k) \ 90{ \ 91 return (k->field >> offset) & ~(~0ULL << (end - offset)); \ 92} \ 93 \ 94static inline void SET_##name(type *k, __u64 v) \ 95{ \ 96 k->field &= ~(~(~0ULL << (end - offset)) << offset); \ 97 k->field |= (v & ~(~0ULL << (end - offset))) << offset; \ 98} 99 100#define LE_BITMASK(_bits, name, type, field, offset, end) \ 101static const __maybe_unused unsigned name##_OFFSET = offset; \ 102static const __maybe_unused unsigned name##_BITS = (end - offset); \ 103static const __maybe_unused __u##_bits name##_MAX = (1ULL << (end - offset)) - 1;\ 104 \ 105static inline __u64 name(const type *k) \ 106{ \ 107 return (__le##_bits##_to_cpu(k->field) >> offset) & \ 108 ~(~0ULL << (end - offset)); \ 109} \ 110 \ 111static inline void SET_##name(type *k, __u64 v) \ 112{ \ 113 __u##_bits new = __le##_bits##_to_cpu(k->field); \ 114 \ 115 new &= ~(~(~0ULL << (end - offset)) << offset); \ 116 new |= (v & ~(~0ULL << (end - offset))) << offset; \ 117 k->field = __cpu_to_le##_bits(new); \ 118} 119 120#define LE16_BITMASK(n, t, f, o, e) LE_BITMASK(16, n, t, f, o, e) 121#define LE32_BITMASK(n, t, f, o, e) LE_BITMASK(32, n, t, f, o, e) 122#define LE64_BITMASK(n, t, f, o, e) LE_BITMASK(64, n, t, f, o, e) 123 124struct bkey_format { 125 __u8 key_u64s; 126 __u8 nr_fields; 127 /* One unused slot for now: */ 128 __u8 bits_per_field[6]; 129 __le64 field_offset[6]; 130}; 131 132/* Btree keys - all units are in sectors */ 133 134struct bpos { 135 /* 136 * Word order matches machine byte order - btree code treats a bpos as a 137 * single large integer, for search/comparison purposes 138 * 139 * Note that wherever a bpos is embedded in another on disk data 140 * structure, it has to be byte swabbed when reading in metadata that 141 * wasn't written in native endian order: 142 */ 143#if __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__ 144 __u32 snapshot; 145 __u64 offset; 146 __u64 inode; 147#elif __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__ 148 __u64 inode; 149 __u64 offset; /* Points to end of extent - sectors */ 150 __u32 snapshot; 151#else 152#error edit for your odd byteorder. 153#endif 154} __packed 155#if __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__ 156__aligned(4) 157#endif 158; 159 160#define KEY_INODE_MAX ((__u64)~0ULL) 161#define KEY_OFFSET_MAX ((__u64)~0ULL) 162#define KEY_SNAPSHOT_MAX ((__u32)~0U) 163#define KEY_SIZE_MAX ((__u32)~0U) 164 165static inline struct bpos SPOS(__u64 inode, __u64 offset, __u32 snapshot) 166{ 167 return (struct bpos) { 168 .inode = inode, 169 .offset = offset, 170 .snapshot = snapshot, 171 }; 172} 173 174#define POS_MIN SPOS(0, 0, 0) 175#define POS_MAX SPOS(KEY_INODE_MAX, KEY_OFFSET_MAX, 0) 176#define SPOS_MAX SPOS(KEY_INODE_MAX, KEY_OFFSET_MAX, KEY_SNAPSHOT_MAX) 177#define POS(_inode, _offset) SPOS(_inode, _offset, 0) 178 179/* Empty placeholder struct, for container_of() */ 180struct bch_val { 181 __u64 __nothing[0]; 182}; 183 184struct bversion { 185#if __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__ 186 __u64 lo; 187 __u32 hi; 188#elif __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__ 189 __u32 hi; 190 __u64 lo; 191#endif 192} __packed 193#if __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__ 194__aligned(4) 195#endif 196; 197 198struct bkey { 199 /* Size of combined key and value, in u64s */ 200 __u8 u64s; 201 202 /* Format of key (0 for format local to btree node) */ 203#if defined(__LITTLE_ENDIAN_BITFIELD) 204 __u8 format:7, 205 needs_whiteout:1; 206#elif defined (__BIG_ENDIAN_BITFIELD) 207 __u8 needs_whiteout:1, 208 format:7; 209#else 210#error edit for your odd byteorder. 211#endif 212 213 /* Type of the value */ 214 __u8 type; 215 216#if __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__ 217 __u8 pad[1]; 218 219 struct bversion version; 220 __u32 size; /* extent size, in sectors */ 221 struct bpos p; 222#elif __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__ 223 struct bpos p; 224 __u32 size; /* extent size, in sectors */ 225 struct bversion version; 226 227 __u8 pad[1]; 228#endif 229} __packed 230#if __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__ 231/* 232 * The big-endian version of bkey can't be compiled by rustc with the "aligned" 233 * attr since it doesn't allow types to have both "packed" and "aligned" attrs. 234 * So for Rust compatibility, don't include this. It can be included in the LE 235 * version because the "packed" attr is redundant in that case. 236 * 237 * History: (quoting Kent) 238 * 239 * Specifically, when i was designing bkey, I wanted the header to be no 240 * bigger than necessary so that bkey_packed could use the rest. That means that 241 * decently offten extent keys will fit into only 8 bytes, instead of spilling over 242 * to 16. 243 * 244 * But packed_bkey treats the part after the header - the packed section - 245 * as a single multi word, variable length integer. And bkey, the unpacked 246 * version, is just a special case version of a bkey_packed; all the packed 247 * bkey code will work on keys in any packed format, the in-memory 248 * representation of an unpacked key also is just one type of packed key... 249 * 250 * So that constrains the key part of a bkig endian bkey to start right 251 * after the header. 252 * 253 * If we ever do a bkey_v2 and need to expand the hedaer by another byte for 254 * some reason - that will clean up this wart. 255 */ 256__aligned(8) 257#endif 258; 259 260struct bkey_packed { 261 __u64 _data[0]; 262 263 /* Size of combined key and value, in u64s */ 264 __u8 u64s; 265 266 /* Format of key (0 for format local to btree node) */ 267 268 /* 269 * XXX: next incompat on disk format change, switch format and 270 * needs_whiteout - bkey_packed() will be cheaper if format is the high 271 * bits of the bitfield 272 */ 273#if defined(__LITTLE_ENDIAN_BITFIELD) 274 __u8 format:7, 275 needs_whiteout:1; 276#elif defined (__BIG_ENDIAN_BITFIELD) 277 __u8 needs_whiteout:1, 278 format:7; 279#endif 280 281 /* Type of the value */ 282 __u8 type; 283 __u8 key_start[0]; 284 285 /* 286 * We copy bkeys with struct assignment in various places, and while 287 * that shouldn't be done with packed bkeys we can't disallow it in C, 288 * and it's legal to cast a bkey to a bkey_packed - so padding it out 289 * to the same size as struct bkey should hopefully be safest. 290 */ 291 __u8 pad[sizeof(struct bkey) - 3]; 292} __packed __aligned(8); 293 294typedef struct { 295 __le64 lo; 296 __le64 hi; 297} bch_le128; 298 299#define BKEY_U64s (sizeof(struct bkey) / sizeof(__u64)) 300#define BKEY_U64s_MAX U8_MAX 301#define BKEY_VAL_U64s_MAX (BKEY_U64s_MAX - BKEY_U64s) 302 303#define KEY_PACKED_BITS_START 24 304 305#define KEY_FORMAT_LOCAL_BTREE 0 306#define KEY_FORMAT_CURRENT 1 307 308enum bch_bkey_fields { 309 BKEY_FIELD_INODE, 310 BKEY_FIELD_OFFSET, 311 BKEY_FIELD_SNAPSHOT, 312 BKEY_FIELD_SIZE, 313 BKEY_FIELD_VERSION_HI, 314 BKEY_FIELD_VERSION_LO, 315 BKEY_NR_FIELDS, 316}; 317 318#define bkey_format_field(name, field) \ 319 [BKEY_FIELD_##name] = (sizeof(((struct bkey *) NULL)->field) * 8) 320 321#define BKEY_FORMAT_CURRENT \ 322((struct bkey_format) { \ 323 .key_u64s = BKEY_U64s, \ 324 .nr_fields = BKEY_NR_FIELDS, \ 325 .bits_per_field = { \ 326 bkey_format_field(INODE, p.inode), \ 327 bkey_format_field(OFFSET, p.offset), \ 328 bkey_format_field(SNAPSHOT, p.snapshot), \ 329 bkey_format_field(SIZE, size), \ 330 bkey_format_field(VERSION_HI, version.hi), \ 331 bkey_format_field(VERSION_LO, version.lo), \ 332 }, \ 333}) 334 335/* bkey with inline value */ 336struct bkey_i { 337 __u64 _data[0]; 338 339 struct bkey k; 340 struct bch_val v; 341}; 342 343#define POS_KEY(_pos) \ 344((struct bkey) { \ 345 .u64s = BKEY_U64s, \ 346 .format = KEY_FORMAT_CURRENT, \ 347 .p = _pos, \ 348}) 349 350#define KEY(_inode, _offset, _size) \ 351((struct bkey) { \ 352 .u64s = BKEY_U64s, \ 353 .format = KEY_FORMAT_CURRENT, \ 354 .p = POS(_inode, _offset), \ 355 .size = _size, \ 356}) 357 358static inline void bkey_init(struct bkey *k) 359{ 360 *k = KEY(0, 0, 0); 361} 362 363#define bkey_bytes(_k) ((_k)->u64s * sizeof(__u64)) 364 365#define __BKEY_PADDED(key, pad) \ 366 struct bkey_i key; __u64 key ## _pad[pad] 367 368/* 369 * - DELETED keys are used internally to mark keys that should be ignored but 370 * override keys in composition order. Their version number is ignored. 371 * 372 * - DISCARDED keys indicate that the data is all 0s because it has been 373 * discarded. DISCARDs may have a version; if the version is nonzero the key 374 * will be persistent, otherwise the key will be dropped whenever the btree 375 * node is rewritten (like DELETED keys). 376 * 377 * - ERROR: any read of the data returns a read error, as the data was lost due 378 * to a failing device. Like DISCARDED keys, they can be removed (overridden) 379 * by new writes or cluster-wide GC. Node repair can also overwrite them with 380 * the same or a more recent version number, but not with an older version 381 * number. 382 * 383 * - WHITEOUT: for hash table btrees 384 */ 385#define BCH_BKEY_TYPES() \ 386 x(deleted, 0) \ 387 x(whiteout, 1) \ 388 x(error, 2) \ 389 x(cookie, 3) \ 390 x(hash_whiteout, 4) \ 391 x(btree_ptr, 5) \ 392 x(extent, 6) \ 393 x(reservation, 7) \ 394 x(inode, 8) \ 395 x(inode_generation, 9) \ 396 x(dirent, 10) \ 397 x(xattr, 11) \ 398 x(alloc, 12) \ 399 x(quota, 13) \ 400 x(stripe, 14) \ 401 x(reflink_p, 15) \ 402 x(reflink_v, 16) \ 403 x(inline_data, 17) \ 404 x(btree_ptr_v2, 18) \ 405 x(indirect_inline_data, 19) \ 406 x(alloc_v2, 20) \ 407 x(subvolume, 21) \ 408 x(snapshot, 22) \ 409 x(inode_v2, 23) \ 410 x(alloc_v3, 24) \ 411 x(set, 25) \ 412 x(lru, 26) \ 413 x(alloc_v4, 27) \ 414 x(backpointer, 28) \ 415 x(inode_v3, 29) \ 416 x(bucket_gens, 30) \ 417 x(snapshot_tree, 31) \ 418 x(logged_op_truncate, 32) \ 419 x(logged_op_finsert, 33) 420 421enum bch_bkey_type { 422#define x(name, nr) KEY_TYPE_##name = nr, 423 BCH_BKEY_TYPES() 424#undef x 425 KEY_TYPE_MAX, 426}; 427 428struct bch_deleted { 429 struct bch_val v; 430}; 431 432struct bch_whiteout { 433 struct bch_val v; 434}; 435 436struct bch_error { 437 struct bch_val v; 438}; 439 440struct bch_cookie { 441 struct bch_val v; 442 __le64 cookie; 443}; 444 445struct bch_hash_whiteout { 446 struct bch_val v; 447}; 448 449struct bch_set { 450 struct bch_val v; 451}; 452 453/* 128 bits, sufficient for cryptographic MACs: */ 454struct bch_csum { 455 __le64 lo; 456 __le64 hi; 457} __packed __aligned(8); 458 459struct bch_backpointer { 460 struct bch_val v; 461 __u8 btree_id; 462 __u8 level; 463 __u8 data_type; 464 __u64 bucket_offset:40; 465 __u32 bucket_len; 466 struct bpos pos; 467} __packed __aligned(8); 468 469/* LRU btree: */ 470 471struct bch_lru { 472 struct bch_val v; 473 __le64 idx; 474} __packed __aligned(8); 475 476#define LRU_ID_STRIPES (1U << 16) 477 478/* Optional/variable size superblock sections: */ 479 480struct bch_sb_field { 481 __u64 _data[0]; 482 __le32 u64s; 483 __le32 type; 484}; 485 486#define BCH_SB_FIELDS() \ 487 x(journal, 0) \ 488 x(members_v1, 1) \ 489 x(crypt, 2) \ 490 x(replicas_v0, 3) \ 491 x(quota, 4) \ 492 x(disk_groups, 5) \ 493 x(clean, 6) \ 494 x(replicas, 7) \ 495 x(journal_seq_blacklist, 8) \ 496 x(journal_v2, 9) \ 497 x(counters, 10) \ 498 x(members_v2, 11) \ 499 x(errors, 12) \ 500 x(ext, 13) \ 501 x(downgrade, 14) 502 503#include "alloc_background_format.h" 504#include "extents_format.h" 505#include "reflink_format.h" 506#include "ec_format.h" 507#include "inode_format.h" 508#include "dirent_format.h" 509#include "xattr_format.h" 510#include "quota_format.h" 511#include "logged_ops_format.h" 512#include "snapshot_format.h" 513#include "subvolume_format.h" 514#include "sb-counters_format.h" 515 516enum bch_sb_field_type { 517#define x(f, nr) BCH_SB_FIELD_##f = nr, 518 BCH_SB_FIELDS() 519#undef x 520 BCH_SB_FIELD_NR 521}; 522 523/* 524 * Most superblock fields are replicated in all device's superblocks - a few are 525 * not: 526 */ 527#define BCH_SINGLE_DEVICE_SB_FIELDS \ 528 ((1U << BCH_SB_FIELD_journal)| \ 529 (1U << BCH_SB_FIELD_journal_v2)) 530 531/* BCH_SB_FIELD_journal: */ 532 533struct bch_sb_field_journal { 534 struct bch_sb_field field; 535 __le64 buckets[]; 536}; 537 538struct bch_sb_field_journal_v2 { 539 struct bch_sb_field field; 540 541 struct bch_sb_field_journal_v2_entry { 542 __le64 start; 543 __le64 nr; 544 } d[]; 545}; 546 547/* BCH_SB_FIELD_members_v1: */ 548 549#define BCH_MIN_NR_NBUCKETS (1 << 6) 550 551#define BCH_IOPS_MEASUREMENTS() \ 552 x(seqread, 0) \ 553 x(seqwrite, 1) \ 554 x(randread, 2) \ 555 x(randwrite, 3) 556 557enum bch_iops_measurement { 558#define x(t, n) BCH_IOPS_##t = n, 559 BCH_IOPS_MEASUREMENTS() 560#undef x 561 BCH_IOPS_NR 562}; 563 564#define BCH_MEMBER_ERROR_TYPES() \ 565 x(read, 0) \ 566 x(write, 1) \ 567 x(checksum, 2) 568 569enum bch_member_error_type { 570#define x(t, n) BCH_MEMBER_ERROR_##t = n, 571 BCH_MEMBER_ERROR_TYPES() 572#undef x 573 BCH_MEMBER_ERROR_NR 574}; 575 576struct bch_member { 577 __uuid_t uuid; 578 __le64 nbuckets; /* device size */ 579 __le16 first_bucket; /* index of first bucket used */ 580 __le16 bucket_size; /* sectors */ 581 __u8 btree_bitmap_shift; 582 __u8 pad[3]; 583 __le64 last_mount; /* time_t */ 584 585 __le64 flags; 586 __le32 iops[4]; 587 __le64 errors[BCH_MEMBER_ERROR_NR]; 588 __le64 errors_at_reset[BCH_MEMBER_ERROR_NR]; 589 __le64 errors_reset_time; 590 __le64 seq; 591 __le64 btree_allocated_bitmap; 592}; 593 594#define BCH_MEMBER_V1_BYTES 56 595 596LE64_BITMASK(BCH_MEMBER_STATE, struct bch_member, flags, 0, 4) 597/* 4-14 unused, was TIER, HAS_(META)DATA, REPLACEMENT */ 598LE64_BITMASK(BCH_MEMBER_DISCARD, struct bch_member, flags, 14, 15) 599LE64_BITMASK(BCH_MEMBER_DATA_ALLOWED, struct bch_member, flags, 15, 20) 600LE64_BITMASK(BCH_MEMBER_GROUP, struct bch_member, flags, 20, 28) 601LE64_BITMASK(BCH_MEMBER_DURABILITY, struct bch_member, flags, 28, 30) 602LE64_BITMASK(BCH_MEMBER_FREESPACE_INITIALIZED, 603 struct bch_member, flags, 30, 31) 604 605#if 0 606LE64_BITMASK(BCH_MEMBER_NR_READ_ERRORS, struct bch_member, flags[1], 0, 20); 607LE64_BITMASK(BCH_MEMBER_NR_WRITE_ERRORS,struct bch_member, flags[1], 20, 40); 608#endif 609 610#define BCH_MEMBER_STATES() \ 611 x(rw, 0) \ 612 x(ro, 1) \ 613 x(failed, 2) \ 614 x(spare, 3) 615 616enum bch_member_state { 617#define x(t, n) BCH_MEMBER_STATE_##t = n, 618 BCH_MEMBER_STATES() 619#undef x 620 BCH_MEMBER_STATE_NR 621}; 622 623struct bch_sb_field_members_v1 { 624 struct bch_sb_field field; 625 struct bch_member _members[]; //Members are now variable size 626}; 627 628struct bch_sb_field_members_v2 { 629 struct bch_sb_field field; 630 __le16 member_bytes; //size of single member entry 631 u8 pad[6]; 632 struct bch_member _members[]; 633}; 634 635/* BCH_SB_FIELD_crypt: */ 636 637struct nonce { 638 __le32 d[4]; 639}; 640 641struct bch_key { 642 __le64 key[4]; 643}; 644 645#define BCH_KEY_MAGIC \ 646 (((__u64) 'b' << 0)|((__u64) 'c' << 8)| \ 647 ((__u64) 'h' << 16)|((__u64) '*' << 24)| \ 648 ((__u64) '*' << 32)|((__u64) 'k' << 40)| \ 649 ((__u64) 'e' << 48)|((__u64) 'y' << 56)) 650 651struct bch_encrypted_key { 652 __le64 magic; 653 struct bch_key key; 654}; 655 656/* 657 * If this field is present in the superblock, it stores an encryption key which 658 * is used encrypt all other data/metadata. The key will normally be encrypted 659 * with the key userspace provides, but if encryption has been turned off we'll 660 * just store the master key unencrypted in the superblock so we can access the 661 * previously encrypted data. 662 */ 663struct bch_sb_field_crypt { 664 struct bch_sb_field field; 665 666 __le64 flags; 667 __le64 kdf_flags; 668 struct bch_encrypted_key key; 669}; 670 671LE64_BITMASK(BCH_CRYPT_KDF_TYPE, struct bch_sb_field_crypt, flags, 0, 4); 672 673enum bch_kdf_types { 674 BCH_KDF_SCRYPT = 0, 675 BCH_KDF_NR = 1, 676}; 677 678/* stored as base 2 log of scrypt params: */ 679LE64_BITMASK(BCH_KDF_SCRYPT_N, struct bch_sb_field_crypt, kdf_flags, 0, 16); 680LE64_BITMASK(BCH_KDF_SCRYPT_R, struct bch_sb_field_crypt, kdf_flags, 16, 32); 681LE64_BITMASK(BCH_KDF_SCRYPT_P, struct bch_sb_field_crypt, kdf_flags, 32, 48); 682 683/* BCH_SB_FIELD_replicas: */ 684 685#define BCH_DATA_TYPES() \ 686 x(free, 0) \ 687 x(sb, 1) \ 688 x(journal, 2) \ 689 x(btree, 3) \ 690 x(user, 4) \ 691 x(cached, 5) \ 692 x(parity, 6) \ 693 x(stripe, 7) \ 694 x(need_gc_gens, 8) \ 695 x(need_discard, 9) 696 697enum bch_data_type { 698#define x(t, n) BCH_DATA_##t, 699 BCH_DATA_TYPES() 700#undef x 701 BCH_DATA_NR 702}; 703 704static inline bool data_type_is_empty(enum bch_data_type type) 705{ 706 switch (type) { 707 case BCH_DATA_free: 708 case BCH_DATA_need_gc_gens: 709 case BCH_DATA_need_discard: 710 return true; 711 default: 712 return false; 713 } 714} 715 716static inline bool data_type_is_hidden(enum bch_data_type type) 717{ 718 switch (type) { 719 case BCH_DATA_sb: 720 case BCH_DATA_journal: 721 return true; 722 default: 723 return false; 724 } 725} 726 727struct bch_replicas_entry_v0 { 728 __u8 data_type; 729 __u8 nr_devs; 730 __u8 devs[]; 731} __packed; 732 733struct bch_sb_field_replicas_v0 { 734 struct bch_sb_field field; 735 struct bch_replicas_entry_v0 entries[]; 736} __packed __aligned(8); 737 738struct bch_replicas_entry_v1 { 739 __u8 data_type; 740 __u8 nr_devs; 741 __u8 nr_required; 742 __u8 devs[]; 743} __packed; 744 745#define replicas_entry_bytes(_i) \ 746 (offsetof(typeof(*(_i)), devs) + (_i)->nr_devs) 747 748struct bch_sb_field_replicas { 749 struct bch_sb_field field; 750 struct bch_replicas_entry_v1 entries[]; 751} __packed __aligned(8); 752 753/* BCH_SB_FIELD_disk_groups: */ 754 755#define BCH_SB_LABEL_SIZE 32 756 757struct bch_disk_group { 758 __u8 label[BCH_SB_LABEL_SIZE]; 759 __le64 flags[2]; 760} __packed __aligned(8); 761 762LE64_BITMASK(BCH_GROUP_DELETED, struct bch_disk_group, flags[0], 0, 1) 763LE64_BITMASK(BCH_GROUP_DATA_ALLOWED, struct bch_disk_group, flags[0], 1, 6) 764LE64_BITMASK(BCH_GROUP_PARENT, struct bch_disk_group, flags[0], 6, 24) 765 766struct bch_sb_field_disk_groups { 767 struct bch_sb_field field; 768 struct bch_disk_group entries[]; 769} __packed __aligned(8); 770 771/* 772 * On clean shutdown, store btree roots and current journal sequence number in 773 * the superblock: 774 */ 775struct jset_entry { 776 __le16 u64s; 777 __u8 btree_id; 778 __u8 level; 779 __u8 type; /* designates what this jset holds */ 780 __u8 pad[3]; 781 782 struct bkey_i start[0]; 783 __u64 _data[]; 784}; 785 786struct bch_sb_field_clean { 787 struct bch_sb_field field; 788 789 __le32 flags; 790 __le16 _read_clock; /* no longer used */ 791 __le16 _write_clock; 792 __le64 journal_seq; 793 794 struct jset_entry start[0]; 795 __u64 _data[]; 796}; 797 798struct journal_seq_blacklist_entry { 799 __le64 start; 800 __le64 end; 801}; 802 803struct bch_sb_field_journal_seq_blacklist { 804 struct bch_sb_field field; 805 struct journal_seq_blacklist_entry start[]; 806}; 807 808struct bch_sb_field_errors { 809 struct bch_sb_field field; 810 struct bch_sb_field_error_entry { 811 __le64 v; 812 __le64 last_error_time; 813 } entries[]; 814}; 815 816LE64_BITMASK(BCH_SB_ERROR_ENTRY_ID, struct bch_sb_field_error_entry, v, 0, 16); 817LE64_BITMASK(BCH_SB_ERROR_ENTRY_NR, struct bch_sb_field_error_entry, v, 16, 64); 818 819struct bch_sb_field_ext { 820 struct bch_sb_field field; 821 __le64 recovery_passes_required[2]; 822 __le64 errors_silent[8]; 823 __le64 btrees_lost_data; 824}; 825 826struct bch_sb_field_downgrade_entry { 827 __le16 version; 828 __le64 recovery_passes[2]; 829 __le16 nr_errors; 830 __le16 errors[] __counted_by(nr_errors); 831} __packed __aligned(2); 832 833struct bch_sb_field_downgrade { 834 struct bch_sb_field field; 835 struct bch_sb_field_downgrade_entry entries[]; 836}; 837 838/* Superblock: */ 839 840/* 841 * New versioning scheme: 842 * One common version number for all on disk data structures - superblock, btree 843 * nodes, journal entries 844 */ 845#define BCH_VERSION_MAJOR(_v) ((__u16) ((_v) >> 10)) 846#define BCH_VERSION_MINOR(_v) ((__u16) ((_v) & ~(~0U << 10))) 847#define BCH_VERSION(_major, _minor) (((_major) << 10)|(_minor) << 0) 848 849/* 850 * field 1: version name 851 * field 2: BCH_VERSION(major, minor) 852 * field 3: recovery passess required on upgrade 853 */ 854#define BCH_METADATA_VERSIONS() \ 855 x(bkey_renumber, BCH_VERSION(0, 10)) \ 856 x(inode_btree_change, BCH_VERSION(0, 11)) \ 857 x(snapshot, BCH_VERSION(0, 12)) \ 858 x(inode_backpointers, BCH_VERSION(0, 13)) \ 859 x(btree_ptr_sectors_written, BCH_VERSION(0, 14)) \ 860 x(snapshot_2, BCH_VERSION(0, 15)) \ 861 x(reflink_p_fix, BCH_VERSION(0, 16)) \ 862 x(subvol_dirent, BCH_VERSION(0, 17)) \ 863 x(inode_v2, BCH_VERSION(0, 18)) \ 864 x(freespace, BCH_VERSION(0, 19)) \ 865 x(alloc_v4, BCH_VERSION(0, 20)) \ 866 x(new_data_types, BCH_VERSION(0, 21)) \ 867 x(backpointers, BCH_VERSION(0, 22)) \ 868 x(inode_v3, BCH_VERSION(0, 23)) \ 869 x(unwritten_extents, BCH_VERSION(0, 24)) \ 870 x(bucket_gens, BCH_VERSION(0, 25)) \ 871 x(lru_v2, BCH_VERSION(0, 26)) \ 872 x(fragmentation_lru, BCH_VERSION(0, 27)) \ 873 x(no_bps_in_alloc_keys, BCH_VERSION(0, 28)) \ 874 x(snapshot_trees, BCH_VERSION(0, 29)) \ 875 x(major_minor, BCH_VERSION(1, 0)) \ 876 x(snapshot_skiplists, BCH_VERSION(1, 1)) \ 877 x(deleted_inodes, BCH_VERSION(1, 2)) \ 878 x(rebalance_work, BCH_VERSION(1, 3)) \ 879 x(member_seq, BCH_VERSION(1, 4)) \ 880 x(subvolume_fs_parent, BCH_VERSION(1, 5)) \ 881 x(btree_subvolume_children, BCH_VERSION(1, 6)) \ 882 x(mi_btree_bitmap, BCH_VERSION(1, 7)) 883 884enum bcachefs_metadata_version { 885 bcachefs_metadata_version_min = 9, 886#define x(t, n) bcachefs_metadata_version_##t = n, 887 BCH_METADATA_VERSIONS() 888#undef x 889 bcachefs_metadata_version_max 890}; 891 892static const __maybe_unused 893unsigned bcachefs_metadata_required_upgrade_below = bcachefs_metadata_version_rebalance_work; 894 895#define bcachefs_metadata_version_current (bcachefs_metadata_version_max - 1) 896 897#define BCH_SB_SECTOR 8 898#define BCH_SB_MEMBERS_MAX 64 /* XXX kill */ 899 900struct bch_sb_layout { 901 __uuid_t magic; /* bcachefs superblock UUID */ 902 __u8 layout_type; 903 __u8 sb_max_size_bits; /* base 2 of 512 byte sectors */ 904 __u8 nr_superblocks; 905 __u8 pad[5]; 906 __le64 sb_offset[61]; 907} __packed __aligned(8); 908 909#define BCH_SB_LAYOUT_SECTOR 7 910 911/* 912 * @offset - sector where this sb was written 913 * @version - on disk format version 914 * @version_min - Oldest metadata version this filesystem contains; so we can 915 * safely drop compatibility code and refuse to mount filesystems 916 * we'd need it for 917 * @magic - identifies as a bcachefs superblock (BCHFS_MAGIC) 918 * @seq - incremented each time superblock is written 919 * @uuid - used for generating various magic numbers and identifying 920 * member devices, never changes 921 * @user_uuid - user visible UUID, may be changed 922 * @label - filesystem label 923 * @seq - identifies most recent superblock, incremented each time 924 * superblock is written 925 * @features - enabled incompatible features 926 */ 927struct bch_sb { 928 struct bch_csum csum; 929 __le16 version; 930 __le16 version_min; 931 __le16 pad[2]; 932 __uuid_t magic; 933 __uuid_t uuid; 934 __uuid_t user_uuid; 935 __u8 label[BCH_SB_LABEL_SIZE]; 936 __le64 offset; 937 __le64 seq; 938 939 __le16 block_size; 940 __u8 dev_idx; 941 __u8 nr_devices; 942 __le32 u64s; 943 944 __le64 time_base_lo; 945 __le32 time_base_hi; 946 __le32 time_precision; 947 948 __le64 flags[7]; 949 __le64 write_time; 950 __le64 features[2]; 951 __le64 compat[2]; 952 953 struct bch_sb_layout layout; 954 955 struct bch_sb_field start[0]; 956 __le64 _data[]; 957} __packed __aligned(8); 958 959/* 960 * Flags: 961 * BCH_SB_INITALIZED - set on first mount 962 * BCH_SB_CLEAN - did we shut down cleanly? Just a hint, doesn't affect 963 * behaviour of mount/recovery path: 964 * BCH_SB_INODE_32BIT - limit inode numbers to 32 bits 965 * BCH_SB_128_BIT_MACS - 128 bit macs instead of 80 966 * BCH_SB_ENCRYPTION_TYPE - if nonzero encryption is enabled; overrides 967 * DATA/META_CSUM_TYPE. Also indicates encryption 968 * algorithm in use, if/when we get more than one 969 */ 970 971LE16_BITMASK(BCH_SB_BLOCK_SIZE, struct bch_sb, block_size, 0, 16); 972 973LE64_BITMASK(BCH_SB_INITIALIZED, struct bch_sb, flags[0], 0, 1); 974LE64_BITMASK(BCH_SB_CLEAN, struct bch_sb, flags[0], 1, 2); 975LE64_BITMASK(BCH_SB_CSUM_TYPE, struct bch_sb, flags[0], 2, 8); 976LE64_BITMASK(BCH_SB_ERROR_ACTION, struct bch_sb, flags[0], 8, 12); 977 978LE64_BITMASK(BCH_SB_BTREE_NODE_SIZE, struct bch_sb, flags[0], 12, 28); 979 980LE64_BITMASK(BCH_SB_GC_RESERVE, struct bch_sb, flags[0], 28, 33); 981LE64_BITMASK(BCH_SB_ROOT_RESERVE, struct bch_sb, flags[0], 33, 40); 982 983LE64_BITMASK(BCH_SB_META_CSUM_TYPE, struct bch_sb, flags[0], 40, 44); 984LE64_BITMASK(BCH_SB_DATA_CSUM_TYPE, struct bch_sb, flags[0], 44, 48); 985 986LE64_BITMASK(BCH_SB_META_REPLICAS_WANT, struct bch_sb, flags[0], 48, 52); 987LE64_BITMASK(BCH_SB_DATA_REPLICAS_WANT, struct bch_sb, flags[0], 52, 56); 988 989LE64_BITMASK(BCH_SB_POSIX_ACL, struct bch_sb, flags[0], 56, 57); 990LE64_BITMASK(BCH_SB_USRQUOTA, struct bch_sb, flags[0], 57, 58); 991LE64_BITMASK(BCH_SB_GRPQUOTA, struct bch_sb, flags[0], 58, 59); 992LE64_BITMASK(BCH_SB_PRJQUOTA, struct bch_sb, flags[0], 59, 60); 993 994LE64_BITMASK(BCH_SB_HAS_ERRORS, struct bch_sb, flags[0], 60, 61); 995LE64_BITMASK(BCH_SB_HAS_TOPOLOGY_ERRORS,struct bch_sb, flags[0], 61, 62); 996 997LE64_BITMASK(BCH_SB_BIG_ENDIAN, struct bch_sb, flags[0], 62, 63); 998 999LE64_BITMASK(BCH_SB_STR_HASH_TYPE, struct bch_sb, flags[1], 0, 4); 1000LE64_BITMASK(BCH_SB_COMPRESSION_TYPE_LO,struct bch_sb, flags[1], 4, 8); 1001LE64_BITMASK(BCH_SB_INODE_32BIT, struct bch_sb, flags[1], 8, 9); 1002 1003LE64_BITMASK(BCH_SB_128_BIT_MACS, struct bch_sb, flags[1], 9, 10); 1004LE64_BITMASK(BCH_SB_ENCRYPTION_TYPE, struct bch_sb, flags[1], 10, 14); 1005 1006/* 1007 * Max size of an extent that may require bouncing to read or write 1008 * (checksummed, compressed): 64k 1009 */ 1010LE64_BITMASK(BCH_SB_ENCODED_EXTENT_MAX_BITS, 1011 struct bch_sb, flags[1], 14, 20); 1012 1013LE64_BITMASK(BCH_SB_META_REPLICAS_REQ, struct bch_sb, flags[1], 20, 24); 1014LE64_BITMASK(BCH_SB_DATA_REPLICAS_REQ, struct bch_sb, flags[1], 24, 28); 1015 1016LE64_BITMASK(BCH_SB_PROMOTE_TARGET, struct bch_sb, flags[1], 28, 40); 1017LE64_BITMASK(BCH_SB_FOREGROUND_TARGET, struct bch_sb, flags[1], 40, 52); 1018LE64_BITMASK(BCH_SB_BACKGROUND_TARGET, struct bch_sb, flags[1], 52, 64); 1019 1020LE64_BITMASK(BCH_SB_BACKGROUND_COMPRESSION_TYPE_LO, 1021 struct bch_sb, flags[2], 0, 4); 1022LE64_BITMASK(BCH_SB_GC_RESERVE_BYTES, struct bch_sb, flags[2], 4, 64); 1023 1024LE64_BITMASK(BCH_SB_ERASURE_CODE, struct bch_sb, flags[3], 0, 16); 1025LE64_BITMASK(BCH_SB_METADATA_TARGET, struct bch_sb, flags[3], 16, 28); 1026LE64_BITMASK(BCH_SB_SHARD_INUMS, struct bch_sb, flags[3], 28, 29); 1027LE64_BITMASK(BCH_SB_INODES_USE_KEY_CACHE,struct bch_sb, flags[3], 29, 30); 1028LE64_BITMASK(BCH_SB_JOURNAL_FLUSH_DELAY,struct bch_sb, flags[3], 30, 62); 1029LE64_BITMASK(BCH_SB_JOURNAL_FLUSH_DISABLED,struct bch_sb, flags[3], 62, 63); 1030LE64_BITMASK(BCH_SB_JOURNAL_RECLAIM_DELAY,struct bch_sb, flags[4], 0, 32); 1031LE64_BITMASK(BCH_SB_JOURNAL_TRANSACTION_NAMES,struct bch_sb, flags[4], 32, 33); 1032LE64_BITMASK(BCH_SB_NOCOW, struct bch_sb, flags[4], 33, 34); 1033LE64_BITMASK(BCH_SB_WRITE_BUFFER_SIZE, struct bch_sb, flags[4], 34, 54); 1034LE64_BITMASK(BCH_SB_VERSION_UPGRADE, struct bch_sb, flags[4], 54, 56); 1035 1036LE64_BITMASK(BCH_SB_COMPRESSION_TYPE_HI,struct bch_sb, flags[4], 56, 60); 1037LE64_BITMASK(BCH_SB_BACKGROUND_COMPRESSION_TYPE_HI, 1038 struct bch_sb, flags[4], 60, 64); 1039 1040LE64_BITMASK(BCH_SB_VERSION_UPGRADE_COMPLETE, 1041 struct bch_sb, flags[5], 0, 16); 1042 1043static inline __u64 BCH_SB_COMPRESSION_TYPE(const struct bch_sb *sb) 1044{ 1045 return BCH_SB_COMPRESSION_TYPE_LO(sb) | (BCH_SB_COMPRESSION_TYPE_HI(sb) << 4); 1046} 1047 1048static inline void SET_BCH_SB_COMPRESSION_TYPE(struct bch_sb *sb, __u64 v) 1049{ 1050 SET_BCH_SB_COMPRESSION_TYPE_LO(sb, v); 1051 SET_BCH_SB_COMPRESSION_TYPE_HI(sb, v >> 4); 1052} 1053 1054static inline __u64 BCH_SB_BACKGROUND_COMPRESSION_TYPE(const struct bch_sb *sb) 1055{ 1056 return BCH_SB_BACKGROUND_COMPRESSION_TYPE_LO(sb) | 1057 (BCH_SB_BACKGROUND_COMPRESSION_TYPE_HI(sb) << 4); 1058} 1059 1060static inline void SET_BCH_SB_BACKGROUND_COMPRESSION_TYPE(struct bch_sb *sb, __u64 v) 1061{ 1062 SET_BCH_SB_BACKGROUND_COMPRESSION_TYPE_LO(sb, v); 1063 SET_BCH_SB_BACKGROUND_COMPRESSION_TYPE_HI(sb, v >> 4); 1064} 1065 1066/* 1067 * Features: 1068 * 1069 * journal_seq_blacklist_v3: gates BCH_SB_FIELD_journal_seq_blacklist 1070 * reflink: gates KEY_TYPE_reflink 1071 * inline_data: gates KEY_TYPE_inline_data 1072 * new_siphash: gates BCH_STR_HASH_siphash 1073 * new_extent_overwrite: gates BTREE_NODE_NEW_EXTENT_OVERWRITE 1074 */ 1075#define BCH_SB_FEATURES() \ 1076 x(lz4, 0) \ 1077 x(gzip, 1) \ 1078 x(zstd, 2) \ 1079 x(atomic_nlink, 3) \ 1080 x(ec, 4) \ 1081 x(journal_seq_blacklist_v3, 5) \ 1082 x(reflink, 6) \ 1083 x(new_siphash, 7) \ 1084 x(inline_data, 8) \ 1085 x(new_extent_overwrite, 9) \ 1086 x(incompressible, 10) \ 1087 x(btree_ptr_v2, 11) \ 1088 x(extents_above_btree_updates, 12) \ 1089 x(btree_updates_journalled, 13) \ 1090 x(reflink_inline_data, 14) \ 1091 x(new_varint, 15) \ 1092 x(journal_no_flush, 16) \ 1093 x(alloc_v2, 17) \ 1094 x(extents_across_btree_nodes, 18) 1095 1096#define BCH_SB_FEATURES_ALWAYS \ 1097 ((1ULL << BCH_FEATURE_new_extent_overwrite)| \ 1098 (1ULL << BCH_FEATURE_extents_above_btree_updates)|\ 1099 (1ULL << BCH_FEATURE_btree_updates_journalled)|\ 1100 (1ULL << BCH_FEATURE_alloc_v2)|\ 1101 (1ULL << BCH_FEATURE_extents_across_btree_nodes)) 1102 1103#define BCH_SB_FEATURES_ALL \ 1104 (BCH_SB_FEATURES_ALWAYS| \ 1105 (1ULL << BCH_FEATURE_new_siphash)| \ 1106 (1ULL << BCH_FEATURE_btree_ptr_v2)| \ 1107 (1ULL << BCH_FEATURE_new_varint)| \ 1108 (1ULL << BCH_FEATURE_journal_no_flush)) 1109 1110enum bch_sb_feature { 1111#define x(f, n) BCH_FEATURE_##f, 1112 BCH_SB_FEATURES() 1113#undef x 1114 BCH_FEATURE_NR, 1115}; 1116 1117#define BCH_SB_COMPAT() \ 1118 x(alloc_info, 0) \ 1119 x(alloc_metadata, 1) \ 1120 x(extents_above_btree_updates_done, 2) \ 1121 x(bformat_overflow_done, 3) 1122 1123enum bch_sb_compat { 1124#define x(f, n) BCH_COMPAT_##f, 1125 BCH_SB_COMPAT() 1126#undef x 1127 BCH_COMPAT_NR, 1128}; 1129 1130/* options: */ 1131 1132#define BCH_VERSION_UPGRADE_OPTS() \ 1133 x(compatible, 0) \ 1134 x(incompatible, 1) \ 1135 x(none, 2) 1136 1137enum bch_version_upgrade_opts { 1138#define x(t, n) BCH_VERSION_UPGRADE_##t = n, 1139 BCH_VERSION_UPGRADE_OPTS() 1140#undef x 1141}; 1142 1143#define BCH_REPLICAS_MAX 4U 1144 1145#define BCH_BKEY_PTRS_MAX 16U 1146 1147#define BCH_ERROR_ACTIONS() \ 1148 x(continue, 0) \ 1149 x(ro, 1) \ 1150 x(panic, 2) 1151 1152enum bch_error_actions { 1153#define x(t, n) BCH_ON_ERROR_##t = n, 1154 BCH_ERROR_ACTIONS() 1155#undef x 1156 BCH_ON_ERROR_NR 1157}; 1158 1159#define BCH_STR_HASH_TYPES() \ 1160 x(crc32c, 0) \ 1161 x(crc64, 1) \ 1162 x(siphash_old, 2) \ 1163 x(siphash, 3) 1164 1165enum bch_str_hash_type { 1166#define x(t, n) BCH_STR_HASH_##t = n, 1167 BCH_STR_HASH_TYPES() 1168#undef x 1169 BCH_STR_HASH_NR 1170}; 1171 1172#define BCH_STR_HASH_OPTS() \ 1173 x(crc32c, 0) \ 1174 x(crc64, 1) \ 1175 x(siphash, 2) 1176 1177enum bch_str_hash_opts { 1178#define x(t, n) BCH_STR_HASH_OPT_##t = n, 1179 BCH_STR_HASH_OPTS() 1180#undef x 1181 BCH_STR_HASH_OPT_NR 1182}; 1183 1184#define BCH_CSUM_TYPES() \ 1185 x(none, 0) \ 1186 x(crc32c_nonzero, 1) \ 1187 x(crc64_nonzero, 2) \ 1188 x(chacha20_poly1305_80, 3) \ 1189 x(chacha20_poly1305_128, 4) \ 1190 x(crc32c, 5) \ 1191 x(crc64, 6) \ 1192 x(xxhash, 7) 1193 1194enum bch_csum_type { 1195#define x(t, n) BCH_CSUM_##t = n, 1196 BCH_CSUM_TYPES() 1197#undef x 1198 BCH_CSUM_NR 1199}; 1200 1201static const __maybe_unused unsigned bch_crc_bytes[] = { 1202 [BCH_CSUM_none] = 0, 1203 [BCH_CSUM_crc32c_nonzero] = 4, 1204 [BCH_CSUM_crc32c] = 4, 1205 [BCH_CSUM_crc64_nonzero] = 8, 1206 [BCH_CSUM_crc64] = 8, 1207 [BCH_CSUM_xxhash] = 8, 1208 [BCH_CSUM_chacha20_poly1305_80] = 10, 1209 [BCH_CSUM_chacha20_poly1305_128] = 16, 1210}; 1211 1212static inline _Bool bch2_csum_type_is_encryption(enum bch_csum_type type) 1213{ 1214 switch (type) { 1215 case BCH_CSUM_chacha20_poly1305_80: 1216 case BCH_CSUM_chacha20_poly1305_128: 1217 return true; 1218 default: 1219 return false; 1220 } 1221} 1222 1223#define BCH_CSUM_OPTS() \ 1224 x(none, 0) \ 1225 x(crc32c, 1) \ 1226 x(crc64, 2) \ 1227 x(xxhash, 3) 1228 1229enum bch_csum_opts { 1230#define x(t, n) BCH_CSUM_OPT_##t = n, 1231 BCH_CSUM_OPTS() 1232#undef x 1233 BCH_CSUM_OPT_NR 1234}; 1235 1236#define BCH_COMPRESSION_TYPES() \ 1237 x(none, 0) \ 1238 x(lz4_old, 1) \ 1239 x(gzip, 2) \ 1240 x(lz4, 3) \ 1241 x(zstd, 4) \ 1242 x(incompressible, 5) 1243 1244enum bch_compression_type { 1245#define x(t, n) BCH_COMPRESSION_TYPE_##t = n, 1246 BCH_COMPRESSION_TYPES() 1247#undef x 1248 BCH_COMPRESSION_TYPE_NR 1249}; 1250 1251#define BCH_COMPRESSION_OPTS() \ 1252 x(none, 0) \ 1253 x(lz4, 1) \ 1254 x(gzip, 2) \ 1255 x(zstd, 3) 1256 1257enum bch_compression_opts { 1258#define x(t, n) BCH_COMPRESSION_OPT_##t = n, 1259 BCH_COMPRESSION_OPTS() 1260#undef x 1261 BCH_COMPRESSION_OPT_NR 1262}; 1263 1264/* 1265 * Magic numbers 1266 * 1267 * The various other data structures have their own magic numbers, which are 1268 * xored with the first part of the cache set's UUID 1269 */ 1270 1271#define BCACHE_MAGIC \ 1272 UUID_INIT(0xc68573f6, 0x4e1a, 0x45ca, \ 1273 0x82, 0x65, 0xf5, 0x7f, 0x48, 0xba, 0x6d, 0x81) 1274#define BCHFS_MAGIC \ 1275 UUID_INIT(0xc68573f6, 0x66ce, 0x90a9, \ 1276 0xd9, 0x6a, 0x60, 0xcf, 0x80, 0x3d, 0xf7, 0xef) 1277 1278#define BCACHEFS_STATFS_MAGIC 0xca451a4e 1279 1280#define JSET_MAGIC __cpu_to_le64(0x245235c1a3625032ULL) 1281#define BSET_MAGIC __cpu_to_le64(0x90135c78b99e07f5ULL) 1282 1283static inline __le64 __bch2_sb_magic(struct bch_sb *sb) 1284{ 1285 __le64 ret; 1286 1287 memcpy(&ret, &sb->uuid, sizeof(ret)); 1288 return ret; 1289} 1290 1291static inline __u64 __jset_magic(struct bch_sb *sb) 1292{ 1293 return __le64_to_cpu(__bch2_sb_magic(sb) ^ JSET_MAGIC); 1294} 1295 1296static inline __u64 __bset_magic(struct bch_sb *sb) 1297{ 1298 return __le64_to_cpu(__bch2_sb_magic(sb) ^ BSET_MAGIC); 1299} 1300 1301/* Journal */ 1302 1303#define JSET_KEYS_U64s (sizeof(struct jset_entry) / sizeof(__u64)) 1304 1305#define BCH_JSET_ENTRY_TYPES() \ 1306 x(btree_keys, 0) \ 1307 x(btree_root, 1) \ 1308 x(prio_ptrs, 2) \ 1309 x(blacklist, 3) \ 1310 x(blacklist_v2, 4) \ 1311 x(usage, 5) \ 1312 x(data_usage, 6) \ 1313 x(clock, 7) \ 1314 x(dev_usage, 8) \ 1315 x(log, 9) \ 1316 x(overwrite, 10) \ 1317 x(write_buffer_keys, 11) \ 1318 x(datetime, 12) 1319 1320enum bch_jset_entry_type { 1321#define x(f, nr) BCH_JSET_ENTRY_##f = nr, 1322 BCH_JSET_ENTRY_TYPES() 1323#undef x 1324 BCH_JSET_ENTRY_NR 1325}; 1326 1327static inline bool jset_entry_is_key(struct jset_entry *e) 1328{ 1329 switch (e->type) { 1330 case BCH_JSET_ENTRY_btree_keys: 1331 case BCH_JSET_ENTRY_btree_root: 1332 case BCH_JSET_ENTRY_overwrite: 1333 case BCH_JSET_ENTRY_write_buffer_keys: 1334 return true; 1335 } 1336 1337 return false; 1338} 1339 1340/* 1341 * Journal sequence numbers can be blacklisted: bsets record the max sequence 1342 * number of all the journal entries they contain updates for, so that on 1343 * recovery we can ignore those bsets that contain index updates newer that what 1344 * made it into the journal. 1345 * 1346 * This means that we can't reuse that journal_seq - we have to skip it, and 1347 * then record that we skipped it so that the next time we crash and recover we 1348 * don't think there was a missing journal entry. 1349 */ 1350struct jset_entry_blacklist { 1351 struct jset_entry entry; 1352 __le64 seq; 1353}; 1354 1355struct jset_entry_blacklist_v2 { 1356 struct jset_entry entry; 1357 __le64 start; 1358 __le64 end; 1359}; 1360 1361#define BCH_FS_USAGE_TYPES() \ 1362 x(reserved, 0) \ 1363 x(inodes, 1) \ 1364 x(key_version, 2) 1365 1366enum bch_fs_usage_type { 1367#define x(f, nr) BCH_FS_USAGE_##f = nr, 1368 BCH_FS_USAGE_TYPES() 1369#undef x 1370 BCH_FS_USAGE_NR 1371}; 1372 1373struct jset_entry_usage { 1374 struct jset_entry entry; 1375 __le64 v; 1376} __packed; 1377 1378struct jset_entry_data_usage { 1379 struct jset_entry entry; 1380 __le64 v; 1381 struct bch_replicas_entry_v1 r; 1382} __packed; 1383 1384struct jset_entry_clock { 1385 struct jset_entry entry; 1386 __u8 rw; 1387 __u8 pad[7]; 1388 __le64 time; 1389} __packed; 1390 1391struct jset_entry_dev_usage_type { 1392 __le64 buckets; 1393 __le64 sectors; 1394 __le64 fragmented; 1395} __packed; 1396 1397struct jset_entry_dev_usage { 1398 struct jset_entry entry; 1399 __le32 dev; 1400 __u32 pad; 1401 1402 __le64 _buckets_ec; /* No longer used */ 1403 __le64 _buckets_unavailable; /* No longer used */ 1404 1405 struct jset_entry_dev_usage_type d[]; 1406}; 1407 1408static inline unsigned jset_entry_dev_usage_nr_types(struct jset_entry_dev_usage *u) 1409{ 1410 return (vstruct_bytes(&u->entry) - sizeof(struct jset_entry_dev_usage)) / 1411 sizeof(struct jset_entry_dev_usage_type); 1412} 1413 1414struct jset_entry_log { 1415 struct jset_entry entry; 1416 u8 d[]; 1417} __packed __aligned(8); 1418 1419struct jset_entry_datetime { 1420 struct jset_entry entry; 1421 __le64 seconds; 1422} __packed __aligned(8); 1423 1424/* 1425 * On disk format for a journal entry: 1426 * seq is monotonically increasing; every journal entry has its own unique 1427 * sequence number. 1428 * 1429 * last_seq is the oldest journal entry that still has keys the btree hasn't 1430 * flushed to disk yet. 1431 * 1432 * version is for on disk format changes. 1433 */ 1434struct jset { 1435 struct bch_csum csum; 1436 1437 __le64 magic; 1438 __le64 seq; 1439 __le32 version; 1440 __le32 flags; 1441 1442 __le32 u64s; /* size of d[] in u64s */ 1443 1444 __u8 encrypted_start[0]; 1445 1446 __le16 _read_clock; /* no longer used */ 1447 __le16 _write_clock; 1448 1449 /* Sequence number of oldest dirty journal entry */ 1450 __le64 last_seq; 1451 1452 1453 struct jset_entry start[0]; 1454 __u64 _data[]; 1455} __packed __aligned(8); 1456 1457LE32_BITMASK(JSET_CSUM_TYPE, struct jset, flags, 0, 4); 1458LE32_BITMASK(JSET_BIG_ENDIAN, struct jset, flags, 4, 5); 1459LE32_BITMASK(JSET_NO_FLUSH, struct jset, flags, 5, 6); 1460 1461#define BCH_JOURNAL_BUCKETS_MIN 8 1462 1463/* Btree: */ 1464 1465enum btree_id_flags { 1466 BTREE_ID_EXTENTS = BIT(0), 1467 BTREE_ID_SNAPSHOTS = BIT(1), 1468 BTREE_ID_SNAPSHOT_FIELD = BIT(2), 1469 BTREE_ID_DATA = BIT(3), 1470}; 1471 1472#define BCH_BTREE_IDS() \ 1473 x(extents, 0, BTREE_ID_EXTENTS|BTREE_ID_SNAPSHOTS|BTREE_ID_DATA,\ 1474 BIT_ULL(KEY_TYPE_whiteout)| \ 1475 BIT_ULL(KEY_TYPE_error)| \ 1476 BIT_ULL(KEY_TYPE_cookie)| \ 1477 BIT_ULL(KEY_TYPE_extent)| \ 1478 BIT_ULL(KEY_TYPE_reservation)| \ 1479 BIT_ULL(KEY_TYPE_reflink_p)| \ 1480 BIT_ULL(KEY_TYPE_inline_data)) \ 1481 x(inodes, 1, BTREE_ID_SNAPSHOTS, \ 1482 BIT_ULL(KEY_TYPE_whiteout)| \ 1483 BIT_ULL(KEY_TYPE_inode)| \ 1484 BIT_ULL(KEY_TYPE_inode_v2)| \ 1485 BIT_ULL(KEY_TYPE_inode_v3)| \ 1486 BIT_ULL(KEY_TYPE_inode_generation)) \ 1487 x(dirents, 2, BTREE_ID_SNAPSHOTS, \ 1488 BIT_ULL(KEY_TYPE_whiteout)| \ 1489 BIT_ULL(KEY_TYPE_hash_whiteout)| \ 1490 BIT_ULL(KEY_TYPE_dirent)) \ 1491 x(xattrs, 3, BTREE_ID_SNAPSHOTS, \ 1492 BIT_ULL(KEY_TYPE_whiteout)| \ 1493 BIT_ULL(KEY_TYPE_cookie)| \ 1494 BIT_ULL(KEY_TYPE_hash_whiteout)| \ 1495 BIT_ULL(KEY_TYPE_xattr)) \ 1496 x(alloc, 4, 0, \ 1497 BIT_ULL(KEY_TYPE_alloc)| \ 1498 BIT_ULL(KEY_TYPE_alloc_v2)| \ 1499 BIT_ULL(KEY_TYPE_alloc_v3)| \ 1500 BIT_ULL(KEY_TYPE_alloc_v4)) \ 1501 x(quotas, 5, 0, \ 1502 BIT_ULL(KEY_TYPE_quota)) \ 1503 x(stripes, 6, 0, \ 1504 BIT_ULL(KEY_TYPE_stripe)) \ 1505 x(reflink, 7, BTREE_ID_EXTENTS|BTREE_ID_DATA, \ 1506 BIT_ULL(KEY_TYPE_reflink_v)| \ 1507 BIT_ULL(KEY_TYPE_indirect_inline_data)| \ 1508 BIT_ULL(KEY_TYPE_error)) \ 1509 x(subvolumes, 8, 0, \ 1510 BIT_ULL(KEY_TYPE_subvolume)) \ 1511 x(snapshots, 9, 0, \ 1512 BIT_ULL(KEY_TYPE_snapshot)) \ 1513 x(lru, 10, 0, \ 1514 BIT_ULL(KEY_TYPE_set)) \ 1515 x(freespace, 11, BTREE_ID_EXTENTS, \ 1516 BIT_ULL(KEY_TYPE_set)) \ 1517 x(need_discard, 12, 0, \ 1518 BIT_ULL(KEY_TYPE_set)) \ 1519 x(backpointers, 13, 0, \ 1520 BIT_ULL(KEY_TYPE_backpointer)) \ 1521 x(bucket_gens, 14, 0, \ 1522 BIT_ULL(KEY_TYPE_bucket_gens)) \ 1523 x(snapshot_trees, 15, 0, \ 1524 BIT_ULL(KEY_TYPE_snapshot_tree)) \ 1525 x(deleted_inodes, 16, BTREE_ID_SNAPSHOT_FIELD, \ 1526 BIT_ULL(KEY_TYPE_set)) \ 1527 x(logged_ops, 17, 0, \ 1528 BIT_ULL(KEY_TYPE_logged_op_truncate)| \ 1529 BIT_ULL(KEY_TYPE_logged_op_finsert)) \ 1530 x(rebalance_work, 18, BTREE_ID_SNAPSHOT_FIELD, \ 1531 BIT_ULL(KEY_TYPE_set)|BIT_ULL(KEY_TYPE_cookie)) \ 1532 x(subvolume_children, 19, 0, \ 1533 BIT_ULL(KEY_TYPE_set)) 1534 1535enum btree_id { 1536#define x(name, nr, ...) BTREE_ID_##name = nr, 1537 BCH_BTREE_IDS() 1538#undef x 1539 BTREE_ID_NR 1540}; 1541 1542static inline bool btree_id_is_alloc(enum btree_id id) 1543{ 1544 switch (id) { 1545 case BTREE_ID_alloc: 1546 case BTREE_ID_backpointers: 1547 case BTREE_ID_need_discard: 1548 case BTREE_ID_freespace: 1549 case BTREE_ID_bucket_gens: 1550 return true; 1551 default: 1552 return false; 1553 } 1554} 1555 1556#define BTREE_MAX_DEPTH 4U 1557 1558/* Btree nodes */ 1559 1560/* 1561 * Btree nodes 1562 * 1563 * On disk a btree node is a list/log of these; within each set the keys are 1564 * sorted 1565 */ 1566struct bset { 1567 __le64 seq; 1568 1569 /* 1570 * Highest journal entry this bset contains keys for. 1571 * If on recovery we don't see that journal entry, this bset is ignored: 1572 * this allows us to preserve the order of all index updates after a 1573 * crash, since the journal records a total order of all index updates 1574 * and anything that didn't make it to the journal doesn't get used. 1575 */ 1576 __le64 journal_seq; 1577 1578 __le32 flags; 1579 __le16 version; 1580 __le16 u64s; /* count of d[] in u64s */ 1581 1582 struct bkey_packed start[0]; 1583 __u64 _data[]; 1584} __packed __aligned(8); 1585 1586LE32_BITMASK(BSET_CSUM_TYPE, struct bset, flags, 0, 4); 1587 1588LE32_BITMASK(BSET_BIG_ENDIAN, struct bset, flags, 4, 5); 1589LE32_BITMASK(BSET_SEPARATE_WHITEOUTS, 1590 struct bset, flags, 5, 6); 1591 1592/* Sector offset within the btree node: */ 1593LE32_BITMASK(BSET_OFFSET, struct bset, flags, 16, 32); 1594 1595struct btree_node { 1596 struct bch_csum csum; 1597 __le64 magic; 1598 1599 /* this flags field is encrypted, unlike bset->flags: */ 1600 __le64 flags; 1601 1602 /* Closed interval: */ 1603 struct bpos min_key; 1604 struct bpos max_key; 1605 struct bch_extent_ptr _ptr; /* not used anymore */ 1606 struct bkey_format format; 1607 1608 union { 1609 struct bset keys; 1610 struct { 1611 __u8 pad[22]; 1612 __le16 u64s; 1613 __u64 _data[0]; 1614 1615 }; 1616 }; 1617} __packed __aligned(8); 1618 1619LE64_BITMASK(BTREE_NODE_ID_LO, struct btree_node, flags, 0, 4); 1620LE64_BITMASK(BTREE_NODE_LEVEL, struct btree_node, flags, 4, 8); 1621LE64_BITMASK(BTREE_NODE_NEW_EXTENT_OVERWRITE, 1622 struct btree_node, flags, 8, 9); 1623LE64_BITMASK(BTREE_NODE_ID_HI, struct btree_node, flags, 9, 25); 1624/* 25-32 unused */ 1625LE64_BITMASK(BTREE_NODE_SEQ, struct btree_node, flags, 32, 64); 1626 1627static inline __u64 BTREE_NODE_ID(struct btree_node *n) 1628{ 1629 return BTREE_NODE_ID_LO(n) | (BTREE_NODE_ID_HI(n) << 4); 1630} 1631 1632static inline void SET_BTREE_NODE_ID(struct btree_node *n, __u64 v) 1633{ 1634 SET_BTREE_NODE_ID_LO(n, v); 1635 SET_BTREE_NODE_ID_HI(n, v >> 4); 1636} 1637 1638struct btree_node_entry { 1639 struct bch_csum csum; 1640 1641 union { 1642 struct bset keys; 1643 struct { 1644 __u8 pad[22]; 1645 __le16 u64s; 1646 __u64 _data[0]; 1647 }; 1648 }; 1649} __packed __aligned(8); 1650 1651#endif /* _BCACHEFS_FORMAT_H */ 1652