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 bversion; 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 bversion; 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, bversion.hi), \ 332 bkey_format_field(VERSION_LO, bversion.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 x(accounting, 34) \ 422 x(inode_alloc_cursor, 35) 423 424 enum bch_bkey_type { 425 #define x(name, nr) KEY_TYPE_##name = nr, 426 BCH_BKEY_TYPES() 427 #undef x 428 KEY_TYPE_MAX, 429 }; 430 431 struct bch_deleted { 432 struct bch_val v; 433 }; 434 435 struct bch_whiteout { 436 struct bch_val v; 437 }; 438 439 struct bch_error { 440 struct bch_val v; 441 }; 442 443 struct bch_cookie { 444 struct bch_val v; 445 __le64 cookie; 446 }; 447 448 struct bch_hash_whiteout { 449 struct bch_val v; 450 }; 451 452 struct bch_set { 453 struct bch_val v; 454 }; 455 456 /* 128 bits, sufficient for cryptographic MACs: */ 457 struct bch_csum { 458 __le64 lo; 459 __le64 hi; 460 } __packed __aligned(8); 461 462 struct bch_backpointer { 463 struct bch_val v; 464 __u8 btree_id; 465 __u8 level; 466 __u8 data_type; 467 __u8 bucket_gen; 468 __u32 pad; 469 __u32 bucket_len; 470 struct bpos pos; 471 } __packed __aligned(8); 472 473 /* Optional/variable size superblock sections: */ 474 475 struct bch_sb_field { 476 __u64 _data[0]; 477 __le32 u64s; 478 __le32 type; 479 }; 480 481 #define BCH_SB_FIELDS() \ 482 x(journal, 0) \ 483 x(members_v1, 1) \ 484 x(crypt, 2) \ 485 x(replicas_v0, 3) \ 486 x(quota, 4) \ 487 x(disk_groups, 5) \ 488 x(clean, 6) \ 489 x(replicas, 7) \ 490 x(journal_seq_blacklist, 8) \ 491 x(journal_v2, 9) \ 492 x(counters, 10) \ 493 x(members_v2, 11) \ 494 x(errors, 12) \ 495 x(ext, 13) \ 496 x(downgrade, 14) 497 498 #include "alloc_background_format.h" 499 #include "dirent_format.h" 500 #include "disk_accounting_format.h" 501 #include "disk_groups_format.h" 502 #include "extents_format.h" 503 #include "ec_format.h" 504 #include "inode_format.h" 505 #include "journal_seq_blacklist_format.h" 506 #include "logged_ops_format.h" 507 #include "lru_format.h" 508 #include "quota_format.h" 509 #include "reflink_format.h" 510 #include "replicas_format.h" 511 #include "snapshot_format.h" 512 #include "subvolume_format.h" 513 #include "sb-counters_format.h" 514 #include "sb-downgrade_format.h" 515 #include "sb-errors_format.h" 516 #include "sb-members_format.h" 517 #include "xattr_format.h" 518 519 enum bch_sb_field_type { 520 #define x(f, nr) BCH_SB_FIELD_##f = nr, 521 BCH_SB_FIELDS() 522 #undef x 523 BCH_SB_FIELD_NR 524 }; 525 526 /* 527 * Most superblock fields are replicated in all device's superblocks - a few are 528 * not: 529 */ 530 #define BCH_SINGLE_DEVICE_SB_FIELDS \ 531 ((1U << BCH_SB_FIELD_journal)| \ 532 (1U << BCH_SB_FIELD_journal_v2)) 533 534 /* BCH_SB_FIELD_journal: */ 535 536 struct bch_sb_field_journal { 537 struct bch_sb_field field; 538 __le64 buckets[]; 539 }; 540 541 struct bch_sb_field_journal_v2 { 542 struct bch_sb_field field; 543 544 struct bch_sb_field_journal_v2_entry { 545 __le64 start; 546 __le64 nr; 547 } d[]; 548 }; 549 550 /* BCH_SB_FIELD_crypt: */ 551 552 struct nonce { 553 __le32 d[4]; 554 }; 555 556 struct bch_key { 557 __le64 key[4]; 558 }; 559 560 #define BCH_KEY_MAGIC \ 561 (((__u64) 'b' << 0)|((__u64) 'c' << 8)| \ 562 ((__u64) 'h' << 16)|((__u64) '*' << 24)| \ 563 ((__u64) '*' << 32)|((__u64) 'k' << 40)| \ 564 ((__u64) 'e' << 48)|((__u64) 'y' << 56)) 565 566 struct bch_encrypted_key { 567 __le64 magic; 568 struct bch_key key; 569 }; 570 571 /* 572 * If this field is present in the superblock, it stores an encryption key which 573 * is used encrypt all other data/metadata. The key will normally be encrypted 574 * with the key userspace provides, but if encryption has been turned off we'll 575 * just store the master key unencrypted in the superblock so we can access the 576 * previously encrypted data. 577 */ 578 struct bch_sb_field_crypt { 579 struct bch_sb_field field; 580 581 __le64 flags; 582 __le64 kdf_flags; 583 struct bch_encrypted_key key; 584 }; 585 586 LE64_BITMASK(BCH_CRYPT_KDF_TYPE, struct bch_sb_field_crypt, flags, 0, 4); 587 588 enum bch_kdf_types { 589 BCH_KDF_SCRYPT = 0, 590 BCH_KDF_NR = 1, 591 }; 592 593 /* stored as base 2 log of scrypt params: */ 594 LE64_BITMASK(BCH_KDF_SCRYPT_N, struct bch_sb_field_crypt, kdf_flags, 0, 16); 595 LE64_BITMASK(BCH_KDF_SCRYPT_R, struct bch_sb_field_crypt, kdf_flags, 16, 32); 596 LE64_BITMASK(BCH_KDF_SCRYPT_P, struct bch_sb_field_crypt, kdf_flags, 32, 48); 597 598 /* 599 * On clean shutdown, store btree roots and current journal sequence number in 600 * the superblock: 601 */ 602 struct jset_entry { 603 __le16 u64s; 604 __u8 btree_id; 605 __u8 level; 606 __u8 type; /* designates what this jset holds */ 607 __u8 pad[3]; 608 609 struct bkey_i start[0]; 610 __u64 _data[]; 611 }; 612 613 struct bch_sb_field_clean { 614 struct bch_sb_field field; 615 616 __le32 flags; 617 __le16 _read_clock; /* no longer used */ 618 __le16 _write_clock; 619 __le64 journal_seq; 620 621 struct jset_entry start[0]; 622 __u64 _data[]; 623 }; 624 625 struct bch_sb_field_ext { 626 struct bch_sb_field field; 627 __le64 recovery_passes_required[2]; 628 __le64 errors_silent[8]; 629 __le64 btrees_lost_data; 630 }; 631 632 /* Superblock: */ 633 634 /* 635 * New versioning scheme: 636 * One common version number for all on disk data structures - superblock, btree 637 * nodes, journal entries 638 */ 639 #define BCH_VERSION_MAJOR(_v) ((__u16) ((_v) >> 10)) 640 #define BCH_VERSION_MINOR(_v) ((__u16) ((_v) & ~(~0U << 10))) 641 #define BCH_VERSION(_major, _minor) (((_major) << 10)|(_minor) << 0) 642 643 /* 644 * field 1: version name 645 * field 2: BCH_VERSION(major, minor) 646 * field 3: recovery passess required on upgrade 647 */ 648 #define BCH_METADATA_VERSIONS() \ 649 x(bkey_renumber, BCH_VERSION(0, 10)) \ 650 x(inode_btree_change, BCH_VERSION(0, 11)) \ 651 x(snapshot, BCH_VERSION(0, 12)) \ 652 x(inode_backpointers, BCH_VERSION(0, 13)) \ 653 x(btree_ptr_sectors_written, BCH_VERSION(0, 14)) \ 654 x(snapshot_2, BCH_VERSION(0, 15)) \ 655 x(reflink_p_fix, BCH_VERSION(0, 16)) \ 656 x(subvol_dirent, BCH_VERSION(0, 17)) \ 657 x(inode_v2, BCH_VERSION(0, 18)) \ 658 x(freespace, BCH_VERSION(0, 19)) \ 659 x(alloc_v4, BCH_VERSION(0, 20)) \ 660 x(new_data_types, BCH_VERSION(0, 21)) \ 661 x(backpointers, BCH_VERSION(0, 22)) \ 662 x(inode_v3, BCH_VERSION(0, 23)) \ 663 x(unwritten_extents, BCH_VERSION(0, 24)) \ 664 x(bucket_gens, BCH_VERSION(0, 25)) \ 665 x(lru_v2, BCH_VERSION(0, 26)) \ 666 x(fragmentation_lru, BCH_VERSION(0, 27)) \ 667 x(no_bps_in_alloc_keys, BCH_VERSION(0, 28)) \ 668 x(snapshot_trees, BCH_VERSION(0, 29)) \ 669 x(major_minor, BCH_VERSION(1, 0)) \ 670 x(snapshot_skiplists, BCH_VERSION(1, 1)) \ 671 x(deleted_inodes, BCH_VERSION(1, 2)) \ 672 x(rebalance_work, BCH_VERSION(1, 3)) \ 673 x(member_seq, BCH_VERSION(1, 4)) \ 674 x(subvolume_fs_parent, BCH_VERSION(1, 5)) \ 675 x(btree_subvolume_children, BCH_VERSION(1, 6)) \ 676 x(mi_btree_bitmap, BCH_VERSION(1, 7)) \ 677 x(bucket_stripe_sectors, BCH_VERSION(1, 8)) \ 678 x(disk_accounting_v2, BCH_VERSION(1, 9)) \ 679 x(disk_accounting_v3, BCH_VERSION(1, 10)) \ 680 x(disk_accounting_inum, BCH_VERSION(1, 11)) \ 681 x(rebalance_work_acct_fix, BCH_VERSION(1, 12)) \ 682 x(inode_has_child_snapshots, BCH_VERSION(1, 13)) \ 683 x(backpointer_bucket_gen, BCH_VERSION(1, 14)) \ 684 x(disk_accounting_big_endian, BCH_VERSION(1, 15)) \ 685 x(reflink_p_may_update_opts, BCH_VERSION(1, 16)) \ 686 x(inode_depth, BCH_VERSION(1, 17)) \ 687 x(persistent_inode_cursors, BCH_VERSION(1, 18)) \ 688 x(autofix_errors, BCH_VERSION(1, 19)) \ 689 x(directory_size, BCH_VERSION(1, 20)) 690 691 enum bcachefs_metadata_version { 692 bcachefs_metadata_version_min = 9, 693 #define x(t, n) bcachefs_metadata_version_##t = n, 694 BCH_METADATA_VERSIONS() 695 #undef x 696 bcachefs_metadata_version_max 697 }; 698 699 static const __maybe_unused 700 unsigned bcachefs_metadata_required_upgrade_below = bcachefs_metadata_version_rebalance_work; 701 702 #define bcachefs_metadata_version_current (bcachefs_metadata_version_max - 1) 703 704 #define BCH_SB_SECTOR 8 705 706 #define BCH_SB_LAYOUT_SIZE_BITS_MAX 16 /* 32 MB */ 707 708 struct bch_sb_layout { 709 __uuid_t magic; /* bcachefs superblock UUID */ 710 __u8 layout_type; 711 __u8 sb_max_size_bits; /* base 2 of 512 byte sectors */ 712 __u8 nr_superblocks; 713 __u8 pad[5]; 714 __le64 sb_offset[61]; 715 } __packed __aligned(8); 716 717 #define BCH_SB_LAYOUT_SECTOR 7 718 719 /* 720 * @offset - sector where this sb was written 721 * @version - on disk format version 722 * @version_min - Oldest metadata version this filesystem contains; so we can 723 * safely drop compatibility code and refuse to mount filesystems 724 * we'd need it for 725 * @magic - identifies as a bcachefs superblock (BCHFS_MAGIC) 726 * @seq - incremented each time superblock is written 727 * @uuid - used for generating various magic numbers and identifying 728 * member devices, never changes 729 * @user_uuid - user visible UUID, may be changed 730 * @label - filesystem label 731 * @seq - identifies most recent superblock, incremented each time 732 * superblock is written 733 * @features - enabled incompatible features 734 */ 735 struct bch_sb { 736 struct bch_csum csum; 737 __le16 version; 738 __le16 version_min; 739 __le16 pad[2]; 740 __uuid_t magic; 741 __uuid_t uuid; 742 __uuid_t user_uuid; 743 __u8 label[BCH_SB_LABEL_SIZE]; 744 __le64 offset; 745 __le64 seq; 746 747 __le16 block_size; 748 __u8 dev_idx; 749 __u8 nr_devices; 750 __le32 u64s; 751 752 __le64 time_base_lo; 753 __le32 time_base_hi; 754 __le32 time_precision; 755 756 __le64 flags[7]; 757 __le64 write_time; 758 __le64 features[2]; 759 __le64 compat[2]; 760 761 struct bch_sb_layout layout; 762 763 struct bch_sb_field start[0]; 764 __le64 _data[]; 765 } __packed __aligned(8); 766 767 /* 768 * Flags: 769 * BCH_SB_INITALIZED - set on first mount 770 * BCH_SB_CLEAN - did we shut down cleanly? Just a hint, doesn't affect 771 * behaviour of mount/recovery path: 772 * BCH_SB_INODE_32BIT - limit inode numbers to 32 bits 773 * BCH_SB_128_BIT_MACS - 128 bit macs instead of 80 774 * BCH_SB_ENCRYPTION_TYPE - if nonzero encryption is enabled; overrides 775 * DATA/META_CSUM_TYPE. Also indicates encryption 776 * algorithm in use, if/when we get more than one 777 */ 778 779 LE16_BITMASK(BCH_SB_BLOCK_SIZE, struct bch_sb, block_size, 0, 16); 780 781 LE64_BITMASK(BCH_SB_INITIALIZED, struct bch_sb, flags[0], 0, 1); 782 LE64_BITMASK(BCH_SB_CLEAN, struct bch_sb, flags[0], 1, 2); 783 LE64_BITMASK(BCH_SB_CSUM_TYPE, struct bch_sb, flags[0], 2, 8); 784 LE64_BITMASK(BCH_SB_ERROR_ACTION, struct bch_sb, flags[0], 8, 12); 785 786 LE64_BITMASK(BCH_SB_BTREE_NODE_SIZE, struct bch_sb, flags[0], 12, 28); 787 788 LE64_BITMASK(BCH_SB_GC_RESERVE, struct bch_sb, flags[0], 28, 33); 789 LE64_BITMASK(BCH_SB_ROOT_RESERVE, struct bch_sb, flags[0], 33, 40); 790 791 LE64_BITMASK(BCH_SB_META_CSUM_TYPE, struct bch_sb, flags[0], 40, 44); 792 LE64_BITMASK(BCH_SB_DATA_CSUM_TYPE, struct bch_sb, flags[0], 44, 48); 793 794 LE64_BITMASK(BCH_SB_META_REPLICAS_WANT, struct bch_sb, flags[0], 48, 52); 795 LE64_BITMASK(BCH_SB_DATA_REPLICAS_WANT, struct bch_sb, flags[0], 52, 56); 796 797 LE64_BITMASK(BCH_SB_POSIX_ACL, struct bch_sb, flags[0], 56, 57); 798 LE64_BITMASK(BCH_SB_USRQUOTA, struct bch_sb, flags[0], 57, 58); 799 LE64_BITMASK(BCH_SB_GRPQUOTA, struct bch_sb, flags[0], 58, 59); 800 LE64_BITMASK(BCH_SB_PRJQUOTA, struct bch_sb, flags[0], 59, 60); 801 802 LE64_BITMASK(BCH_SB_HAS_ERRORS, struct bch_sb, flags[0], 60, 61); 803 LE64_BITMASK(BCH_SB_HAS_TOPOLOGY_ERRORS,struct bch_sb, flags[0], 61, 62); 804 805 LE64_BITMASK(BCH_SB_BIG_ENDIAN, struct bch_sb, flags[0], 62, 63); 806 LE64_BITMASK(BCH_SB_PROMOTE_WHOLE_EXTENTS, 807 struct bch_sb, flags[0], 63, 64); 808 809 LE64_BITMASK(BCH_SB_STR_HASH_TYPE, struct bch_sb, flags[1], 0, 4); 810 LE64_BITMASK(BCH_SB_COMPRESSION_TYPE_LO,struct bch_sb, flags[1], 4, 8); 811 LE64_BITMASK(BCH_SB_INODE_32BIT, struct bch_sb, flags[1], 8, 9); 812 813 LE64_BITMASK(BCH_SB_128_BIT_MACS, struct bch_sb, flags[1], 9, 10); 814 LE64_BITMASK(BCH_SB_ENCRYPTION_TYPE, struct bch_sb, flags[1], 10, 14); 815 816 /* 817 * Max size of an extent that may require bouncing to read or write 818 * (checksummed, compressed): 64k 819 */ 820 LE64_BITMASK(BCH_SB_ENCODED_EXTENT_MAX_BITS, 821 struct bch_sb, flags[1], 14, 20); 822 823 LE64_BITMASK(BCH_SB_META_REPLICAS_REQ, struct bch_sb, flags[1], 20, 24); 824 LE64_BITMASK(BCH_SB_DATA_REPLICAS_REQ, struct bch_sb, flags[1], 24, 28); 825 826 LE64_BITMASK(BCH_SB_PROMOTE_TARGET, struct bch_sb, flags[1], 28, 40); 827 LE64_BITMASK(BCH_SB_FOREGROUND_TARGET, struct bch_sb, flags[1], 40, 52); 828 LE64_BITMASK(BCH_SB_BACKGROUND_TARGET, struct bch_sb, flags[1], 52, 64); 829 830 LE64_BITMASK(BCH_SB_BACKGROUND_COMPRESSION_TYPE_LO, 831 struct bch_sb, flags[2], 0, 4); 832 LE64_BITMASK(BCH_SB_GC_RESERVE_BYTES, struct bch_sb, flags[2], 4, 64); 833 834 LE64_BITMASK(BCH_SB_ERASURE_CODE, struct bch_sb, flags[3], 0, 16); 835 LE64_BITMASK(BCH_SB_METADATA_TARGET, struct bch_sb, flags[3], 16, 28); 836 LE64_BITMASK(BCH_SB_SHARD_INUMS, struct bch_sb, flags[3], 28, 29); 837 LE64_BITMASK(BCH_SB_INODES_USE_KEY_CACHE,struct bch_sb, flags[3], 29, 30); 838 LE64_BITMASK(BCH_SB_JOURNAL_FLUSH_DELAY,struct bch_sb, flags[3], 30, 62); 839 LE64_BITMASK(BCH_SB_JOURNAL_FLUSH_DISABLED,struct bch_sb, flags[3], 62, 63); 840 LE64_BITMASK(BCH_SB_JOURNAL_RECLAIM_DELAY,struct bch_sb, flags[4], 0, 32); 841 LE64_BITMASK(BCH_SB_JOURNAL_TRANSACTION_NAMES,struct bch_sb, flags[4], 32, 33); 842 LE64_BITMASK(BCH_SB_NOCOW, struct bch_sb, flags[4], 33, 34); 843 LE64_BITMASK(BCH_SB_WRITE_BUFFER_SIZE, struct bch_sb, flags[4], 34, 54); 844 LE64_BITMASK(BCH_SB_VERSION_UPGRADE, struct bch_sb, flags[4], 54, 56); 845 846 LE64_BITMASK(BCH_SB_COMPRESSION_TYPE_HI,struct bch_sb, flags[4], 56, 60); 847 LE64_BITMASK(BCH_SB_BACKGROUND_COMPRESSION_TYPE_HI, 848 struct bch_sb, flags[4], 60, 64); 849 850 LE64_BITMASK(BCH_SB_VERSION_UPGRADE_COMPLETE, 851 struct bch_sb, flags[5], 0, 16); 852 LE64_BITMASK(BCH_SB_ALLOCATOR_STUCK_TIMEOUT, 853 struct bch_sb, flags[5], 16, 32); 854 LE64_BITMASK(BCH_SB_VERSION_INCOMPAT, struct bch_sb, flags[5], 32, 48); 855 LE64_BITMASK(BCH_SB_VERSION_INCOMPAT_ALLOWED, 856 struct bch_sb, flags[5], 48, 64); 857 LE64_BITMASK(BCH_SB_SHARD_INUMS_NBITS, struct bch_sb, flags[6], 0, 4); 858 859 static inline __u64 BCH_SB_COMPRESSION_TYPE(const struct bch_sb *sb) 860 { 861 return BCH_SB_COMPRESSION_TYPE_LO(sb) | (BCH_SB_COMPRESSION_TYPE_HI(sb) << 4); 862 } 863 864 static inline void SET_BCH_SB_COMPRESSION_TYPE(struct bch_sb *sb, __u64 v) 865 { 866 SET_BCH_SB_COMPRESSION_TYPE_LO(sb, v); 867 SET_BCH_SB_COMPRESSION_TYPE_HI(sb, v >> 4); 868 } 869 870 static inline __u64 BCH_SB_BACKGROUND_COMPRESSION_TYPE(const struct bch_sb *sb) 871 { 872 return BCH_SB_BACKGROUND_COMPRESSION_TYPE_LO(sb) | 873 (BCH_SB_BACKGROUND_COMPRESSION_TYPE_HI(sb) << 4); 874 } 875 876 static inline void SET_BCH_SB_BACKGROUND_COMPRESSION_TYPE(struct bch_sb *sb, __u64 v) 877 { 878 SET_BCH_SB_BACKGROUND_COMPRESSION_TYPE_LO(sb, v); 879 SET_BCH_SB_BACKGROUND_COMPRESSION_TYPE_HI(sb, v >> 4); 880 } 881 882 /* 883 * Features: 884 * 885 * journal_seq_blacklist_v3: gates BCH_SB_FIELD_journal_seq_blacklist 886 * reflink: gates KEY_TYPE_reflink 887 * inline_data: gates KEY_TYPE_inline_data 888 * new_siphash: gates BCH_STR_HASH_siphash 889 * new_extent_overwrite: gates BTREE_NODE_NEW_EXTENT_OVERWRITE 890 */ 891 #define BCH_SB_FEATURES() \ 892 x(lz4, 0) \ 893 x(gzip, 1) \ 894 x(zstd, 2) \ 895 x(atomic_nlink, 3) \ 896 x(ec, 4) \ 897 x(journal_seq_blacklist_v3, 5) \ 898 x(reflink, 6) \ 899 x(new_siphash, 7) \ 900 x(inline_data, 8) \ 901 x(new_extent_overwrite, 9) \ 902 x(incompressible, 10) \ 903 x(btree_ptr_v2, 11) \ 904 x(extents_above_btree_updates, 12) \ 905 x(btree_updates_journalled, 13) \ 906 x(reflink_inline_data, 14) \ 907 x(new_varint, 15) \ 908 x(journal_no_flush, 16) \ 909 x(alloc_v2, 17) \ 910 x(extents_across_btree_nodes, 18) \ 911 x(incompat_version_field, 19) 912 913 #define BCH_SB_FEATURES_ALWAYS \ 914 (BIT_ULL(BCH_FEATURE_new_extent_overwrite)| \ 915 BIT_ULL(BCH_FEATURE_extents_above_btree_updates)|\ 916 BIT_ULL(BCH_FEATURE_btree_updates_journalled)|\ 917 BIT_ULL(BCH_FEATURE_alloc_v2)|\ 918 BIT_ULL(BCH_FEATURE_extents_across_btree_nodes)) 919 920 #define BCH_SB_FEATURES_ALL \ 921 (BCH_SB_FEATURES_ALWAYS| \ 922 BIT_ULL(BCH_FEATURE_new_siphash)| \ 923 BIT_ULL(BCH_FEATURE_btree_ptr_v2)| \ 924 BIT_ULL(BCH_FEATURE_new_varint)| \ 925 BIT_ULL(BCH_FEATURE_journal_no_flush)) 926 927 enum bch_sb_feature { 928 #define x(f, n) BCH_FEATURE_##f, 929 BCH_SB_FEATURES() 930 #undef x 931 BCH_FEATURE_NR, 932 }; 933 934 #define BCH_SB_COMPAT() \ 935 x(alloc_info, 0) \ 936 x(alloc_metadata, 1) \ 937 x(extents_above_btree_updates_done, 2) \ 938 x(bformat_overflow_done, 3) 939 940 enum bch_sb_compat { 941 #define x(f, n) BCH_COMPAT_##f, 942 BCH_SB_COMPAT() 943 #undef x 944 BCH_COMPAT_NR, 945 }; 946 947 /* options: */ 948 949 #define BCH_VERSION_UPGRADE_OPTS() \ 950 x(compatible, 0) \ 951 x(incompatible, 1) \ 952 x(none, 2) 953 954 enum bch_version_upgrade_opts { 955 #define x(t, n) BCH_VERSION_UPGRADE_##t = n, 956 BCH_VERSION_UPGRADE_OPTS() 957 #undef x 958 }; 959 960 #define BCH_REPLICAS_MAX 4U 961 962 #define BCH_BKEY_PTRS_MAX 16U 963 964 #define BCH_ERROR_ACTIONS() \ 965 x(continue, 0) \ 966 x(fix_safe, 1) \ 967 x(panic, 2) \ 968 x(ro, 3) 969 970 enum bch_error_actions { 971 #define x(t, n) BCH_ON_ERROR_##t = n, 972 BCH_ERROR_ACTIONS() 973 #undef x 974 BCH_ON_ERROR_NR 975 }; 976 977 #define BCH_STR_HASH_TYPES() \ 978 x(crc32c, 0) \ 979 x(crc64, 1) \ 980 x(siphash_old, 2) \ 981 x(siphash, 3) 982 983 enum bch_str_hash_type { 984 #define x(t, n) BCH_STR_HASH_##t = n, 985 BCH_STR_HASH_TYPES() 986 #undef x 987 BCH_STR_HASH_NR 988 }; 989 990 #define BCH_STR_HASH_OPTS() \ 991 x(crc32c, 0) \ 992 x(crc64, 1) \ 993 x(siphash, 2) 994 995 enum bch_str_hash_opts { 996 #define x(t, n) BCH_STR_HASH_OPT_##t = n, 997 BCH_STR_HASH_OPTS() 998 #undef x 999 BCH_STR_HASH_OPT_NR 1000 }; 1001 1002 #define BCH_CSUM_TYPES() \ 1003 x(none, 0) \ 1004 x(crc32c_nonzero, 1) \ 1005 x(crc64_nonzero, 2) \ 1006 x(chacha20_poly1305_80, 3) \ 1007 x(chacha20_poly1305_128, 4) \ 1008 x(crc32c, 5) \ 1009 x(crc64, 6) \ 1010 x(xxhash, 7) 1011 1012 enum bch_csum_type { 1013 #define x(t, n) BCH_CSUM_##t = n, 1014 BCH_CSUM_TYPES() 1015 #undef x 1016 BCH_CSUM_NR 1017 }; 1018 1019 static const __maybe_unused unsigned bch_crc_bytes[] = { 1020 [BCH_CSUM_none] = 0, 1021 [BCH_CSUM_crc32c_nonzero] = 4, 1022 [BCH_CSUM_crc32c] = 4, 1023 [BCH_CSUM_crc64_nonzero] = 8, 1024 [BCH_CSUM_crc64] = 8, 1025 [BCH_CSUM_xxhash] = 8, 1026 [BCH_CSUM_chacha20_poly1305_80] = 10, 1027 [BCH_CSUM_chacha20_poly1305_128] = 16, 1028 }; 1029 1030 static inline _Bool bch2_csum_type_is_encryption(enum bch_csum_type type) 1031 { 1032 switch (type) { 1033 case BCH_CSUM_chacha20_poly1305_80: 1034 case BCH_CSUM_chacha20_poly1305_128: 1035 return true; 1036 default: 1037 return false; 1038 } 1039 } 1040 1041 #define BCH_CSUM_OPTS() \ 1042 x(none, 0) \ 1043 x(crc32c, 1) \ 1044 x(crc64, 2) \ 1045 x(xxhash, 3) 1046 1047 enum bch_csum_opt { 1048 #define x(t, n) BCH_CSUM_OPT_##t = n, 1049 BCH_CSUM_OPTS() 1050 #undef x 1051 BCH_CSUM_OPT_NR 1052 }; 1053 1054 #define BCH_COMPRESSION_TYPES() \ 1055 x(none, 0) \ 1056 x(lz4_old, 1) \ 1057 x(gzip, 2) \ 1058 x(lz4, 3) \ 1059 x(zstd, 4) \ 1060 x(incompressible, 5) 1061 1062 enum bch_compression_type { 1063 #define x(t, n) BCH_COMPRESSION_TYPE_##t = n, 1064 BCH_COMPRESSION_TYPES() 1065 #undef x 1066 BCH_COMPRESSION_TYPE_NR 1067 }; 1068 1069 #define BCH_COMPRESSION_OPTS() \ 1070 x(none, 0) \ 1071 x(lz4, 1) \ 1072 x(gzip, 2) \ 1073 x(zstd, 3) 1074 1075 enum bch_compression_opts { 1076 #define x(t, n) BCH_COMPRESSION_OPT_##t = n, 1077 BCH_COMPRESSION_OPTS() 1078 #undef x 1079 BCH_COMPRESSION_OPT_NR 1080 }; 1081 1082 /* 1083 * Magic numbers 1084 * 1085 * The various other data structures have their own magic numbers, which are 1086 * xored with the first part of the cache set's UUID 1087 */ 1088 1089 #define BCACHE_MAGIC \ 1090 UUID_INIT(0xc68573f6, 0x4e1a, 0x45ca, \ 1091 0x82, 0x65, 0xf5, 0x7f, 0x48, 0xba, 0x6d, 0x81) 1092 #define BCHFS_MAGIC \ 1093 UUID_INIT(0xc68573f6, 0x66ce, 0x90a9, \ 1094 0xd9, 0x6a, 0x60, 0xcf, 0x80, 0x3d, 0xf7, 0xef) 1095 1096 #define BCACHEFS_STATFS_MAGIC BCACHEFS_SUPER_MAGIC 1097 1098 #define JSET_MAGIC __cpu_to_le64(0x245235c1a3625032ULL) 1099 #define BSET_MAGIC __cpu_to_le64(0x90135c78b99e07f5ULL) 1100 1101 static inline __le64 __bch2_sb_magic(struct bch_sb *sb) 1102 { 1103 __le64 ret; 1104 1105 memcpy(&ret, &sb->uuid, sizeof(ret)); 1106 return ret; 1107 } 1108 1109 static inline __u64 __jset_magic(struct bch_sb *sb) 1110 { 1111 return __le64_to_cpu(__bch2_sb_magic(sb) ^ JSET_MAGIC); 1112 } 1113 1114 static inline __u64 __bset_magic(struct bch_sb *sb) 1115 { 1116 return __le64_to_cpu(__bch2_sb_magic(sb) ^ BSET_MAGIC); 1117 } 1118 1119 /* Journal */ 1120 1121 #define JSET_KEYS_U64s (sizeof(struct jset_entry) / sizeof(__u64)) 1122 1123 #define BCH_JSET_ENTRY_TYPES() \ 1124 x(btree_keys, 0) \ 1125 x(btree_root, 1) \ 1126 x(prio_ptrs, 2) \ 1127 x(blacklist, 3) \ 1128 x(blacklist_v2, 4) \ 1129 x(usage, 5) \ 1130 x(data_usage, 6) \ 1131 x(clock, 7) \ 1132 x(dev_usage, 8) \ 1133 x(log, 9) \ 1134 x(overwrite, 10) \ 1135 x(write_buffer_keys, 11) \ 1136 x(datetime, 12) 1137 1138 enum bch_jset_entry_type { 1139 #define x(f, nr) BCH_JSET_ENTRY_##f = nr, 1140 BCH_JSET_ENTRY_TYPES() 1141 #undef x 1142 BCH_JSET_ENTRY_NR 1143 }; 1144 1145 static inline bool jset_entry_is_key(struct jset_entry *e) 1146 { 1147 switch (e->type) { 1148 case BCH_JSET_ENTRY_btree_keys: 1149 case BCH_JSET_ENTRY_btree_root: 1150 case BCH_JSET_ENTRY_write_buffer_keys: 1151 return true; 1152 } 1153 1154 return false; 1155 } 1156 1157 /* 1158 * Journal sequence numbers can be blacklisted: bsets record the max sequence 1159 * number of all the journal entries they contain updates for, so that on 1160 * recovery we can ignore those bsets that contain index updates newer that what 1161 * made it into the journal. 1162 * 1163 * This means that we can't reuse that journal_seq - we have to skip it, and 1164 * then record that we skipped it so that the next time we crash and recover we 1165 * don't think there was a missing journal entry. 1166 */ 1167 struct jset_entry_blacklist { 1168 struct jset_entry entry; 1169 __le64 seq; 1170 }; 1171 1172 struct jset_entry_blacklist_v2 { 1173 struct jset_entry entry; 1174 __le64 start; 1175 __le64 end; 1176 }; 1177 1178 #define BCH_FS_USAGE_TYPES() \ 1179 x(reserved, 0) \ 1180 x(inodes, 1) \ 1181 x(key_version, 2) 1182 1183 enum bch_fs_usage_type { 1184 #define x(f, nr) BCH_FS_USAGE_##f = nr, 1185 BCH_FS_USAGE_TYPES() 1186 #undef x 1187 BCH_FS_USAGE_NR 1188 }; 1189 1190 struct jset_entry_usage { 1191 struct jset_entry entry; 1192 __le64 v; 1193 } __packed; 1194 1195 struct jset_entry_data_usage { 1196 struct jset_entry entry; 1197 __le64 v; 1198 struct bch_replicas_entry_v1 r; 1199 } __packed; 1200 1201 struct jset_entry_clock { 1202 struct jset_entry entry; 1203 __u8 rw; 1204 __u8 pad[7]; 1205 __le64 time; 1206 } __packed; 1207 1208 struct jset_entry_dev_usage_type { 1209 __le64 buckets; 1210 __le64 sectors; 1211 __le64 fragmented; 1212 } __packed; 1213 1214 struct jset_entry_dev_usage { 1215 struct jset_entry entry; 1216 __le32 dev; 1217 __u32 pad; 1218 1219 __le64 _buckets_ec; /* No longer used */ 1220 __le64 _buckets_unavailable; /* No longer used */ 1221 1222 struct jset_entry_dev_usage_type d[]; 1223 }; 1224 1225 static inline unsigned jset_entry_dev_usage_nr_types(struct jset_entry_dev_usage *u) 1226 { 1227 return (vstruct_bytes(&u->entry) - sizeof(struct jset_entry_dev_usage)) / 1228 sizeof(struct jset_entry_dev_usage_type); 1229 } 1230 1231 struct jset_entry_log { 1232 struct jset_entry entry; 1233 u8 d[]; 1234 } __packed __aligned(8); 1235 1236 static inline unsigned jset_entry_log_msg_bytes(struct jset_entry_log *l) 1237 { 1238 unsigned b = vstruct_bytes(&l->entry) - offsetof(struct jset_entry_log, d); 1239 1240 while (b && !l->d[b - 1]) 1241 --b; 1242 return b; 1243 } 1244 1245 struct jset_entry_datetime { 1246 struct jset_entry entry; 1247 __le64 seconds; 1248 } __packed __aligned(8); 1249 1250 /* 1251 * On disk format for a journal entry: 1252 * seq is monotonically increasing; every journal entry has its own unique 1253 * sequence number. 1254 * 1255 * last_seq is the oldest journal entry that still has keys the btree hasn't 1256 * flushed to disk yet. 1257 * 1258 * version is for on disk format changes. 1259 */ 1260 struct jset { 1261 struct bch_csum csum; 1262 1263 __le64 magic; 1264 __le64 seq; 1265 __le32 version; 1266 __le32 flags; 1267 1268 __le32 u64s; /* size of d[] in u64s */ 1269 1270 __u8 encrypted_start[0]; 1271 1272 __le16 _read_clock; /* no longer used */ 1273 __le16 _write_clock; 1274 1275 /* Sequence number of oldest dirty journal entry */ 1276 __le64 last_seq; 1277 1278 1279 struct jset_entry start[0]; 1280 __u64 _data[]; 1281 } __packed __aligned(8); 1282 1283 LE32_BITMASK(JSET_CSUM_TYPE, struct jset, flags, 0, 4); 1284 LE32_BITMASK(JSET_BIG_ENDIAN, struct jset, flags, 4, 5); 1285 LE32_BITMASK(JSET_NO_FLUSH, struct jset, flags, 5, 6); 1286 1287 #define BCH_JOURNAL_BUCKETS_MIN 8 1288 1289 /* Btree: */ 1290 1291 enum btree_id_flags { 1292 BTREE_IS_extents = BIT(0), 1293 BTREE_IS_snapshots = BIT(1), 1294 BTREE_IS_snapshot_field = BIT(2), 1295 BTREE_IS_data = BIT(3), 1296 BTREE_IS_write_buffer = BIT(4), 1297 }; 1298 1299 #define BCH_BTREE_IDS() \ 1300 x(extents, 0, \ 1301 BTREE_IS_extents| \ 1302 BTREE_IS_snapshots| \ 1303 BTREE_IS_data, \ 1304 BIT_ULL(KEY_TYPE_whiteout)| \ 1305 BIT_ULL(KEY_TYPE_error)| \ 1306 BIT_ULL(KEY_TYPE_cookie)| \ 1307 BIT_ULL(KEY_TYPE_extent)| \ 1308 BIT_ULL(KEY_TYPE_reservation)| \ 1309 BIT_ULL(KEY_TYPE_reflink_p)| \ 1310 BIT_ULL(KEY_TYPE_inline_data)) \ 1311 x(inodes, 1, \ 1312 BTREE_IS_snapshots, \ 1313 BIT_ULL(KEY_TYPE_whiteout)| \ 1314 BIT_ULL(KEY_TYPE_inode)| \ 1315 BIT_ULL(KEY_TYPE_inode_v2)| \ 1316 BIT_ULL(KEY_TYPE_inode_v3)| \ 1317 BIT_ULL(KEY_TYPE_inode_generation)) \ 1318 x(dirents, 2, \ 1319 BTREE_IS_snapshots, \ 1320 BIT_ULL(KEY_TYPE_whiteout)| \ 1321 BIT_ULL(KEY_TYPE_hash_whiteout)| \ 1322 BIT_ULL(KEY_TYPE_dirent)) \ 1323 x(xattrs, 3, \ 1324 BTREE_IS_snapshots, \ 1325 BIT_ULL(KEY_TYPE_whiteout)| \ 1326 BIT_ULL(KEY_TYPE_cookie)| \ 1327 BIT_ULL(KEY_TYPE_hash_whiteout)| \ 1328 BIT_ULL(KEY_TYPE_xattr)) \ 1329 x(alloc, 4, 0, \ 1330 BIT_ULL(KEY_TYPE_alloc)| \ 1331 BIT_ULL(KEY_TYPE_alloc_v2)| \ 1332 BIT_ULL(KEY_TYPE_alloc_v3)| \ 1333 BIT_ULL(KEY_TYPE_alloc_v4)) \ 1334 x(quotas, 5, 0, \ 1335 BIT_ULL(KEY_TYPE_quota)) \ 1336 x(stripes, 6, 0, \ 1337 BIT_ULL(KEY_TYPE_stripe)) \ 1338 x(reflink, 7, \ 1339 BTREE_IS_extents| \ 1340 BTREE_IS_data, \ 1341 BIT_ULL(KEY_TYPE_reflink_v)| \ 1342 BIT_ULL(KEY_TYPE_indirect_inline_data)| \ 1343 BIT_ULL(KEY_TYPE_error)) \ 1344 x(subvolumes, 8, 0, \ 1345 BIT_ULL(KEY_TYPE_subvolume)) \ 1346 x(snapshots, 9, 0, \ 1347 BIT_ULL(KEY_TYPE_snapshot)) \ 1348 x(lru, 10, \ 1349 BTREE_IS_write_buffer, \ 1350 BIT_ULL(KEY_TYPE_set)) \ 1351 x(freespace, 11, \ 1352 BTREE_IS_extents, \ 1353 BIT_ULL(KEY_TYPE_set)) \ 1354 x(need_discard, 12, 0, \ 1355 BIT_ULL(KEY_TYPE_set)) \ 1356 x(backpointers, 13, \ 1357 BTREE_IS_write_buffer, \ 1358 BIT_ULL(KEY_TYPE_backpointer)) \ 1359 x(bucket_gens, 14, 0, \ 1360 BIT_ULL(KEY_TYPE_bucket_gens)) \ 1361 x(snapshot_trees, 15, 0, \ 1362 BIT_ULL(KEY_TYPE_snapshot_tree)) \ 1363 x(deleted_inodes, 16, \ 1364 BTREE_IS_snapshot_field| \ 1365 BTREE_IS_write_buffer, \ 1366 BIT_ULL(KEY_TYPE_set)) \ 1367 x(logged_ops, 17, 0, \ 1368 BIT_ULL(KEY_TYPE_logged_op_truncate)| \ 1369 BIT_ULL(KEY_TYPE_logged_op_finsert)| \ 1370 BIT_ULL(KEY_TYPE_inode_alloc_cursor)) \ 1371 x(rebalance_work, 18, \ 1372 BTREE_IS_snapshot_field| \ 1373 BTREE_IS_write_buffer, \ 1374 BIT_ULL(KEY_TYPE_set)|BIT_ULL(KEY_TYPE_cookie)) \ 1375 x(subvolume_children, 19, 0, \ 1376 BIT_ULL(KEY_TYPE_set)) \ 1377 x(accounting, 20, \ 1378 BTREE_IS_snapshot_field| \ 1379 BTREE_IS_write_buffer, \ 1380 BIT_ULL(KEY_TYPE_accounting)) \ 1381 1382 enum btree_id { 1383 #define x(name, nr, ...) BTREE_ID_##name = nr, 1384 BCH_BTREE_IDS() 1385 #undef x 1386 BTREE_ID_NR 1387 }; 1388 1389 /* 1390 * Maximum number of btrees that we will _ever_ have under the current scheme, 1391 * where we refer to them with 64 bit bitfields - and we also need a bit for 1392 * the interior btree node type: 1393 */ 1394 #define BTREE_ID_NR_MAX 63 1395 1396 static inline bool btree_id_is_alloc(enum btree_id id) 1397 { 1398 switch (id) { 1399 case BTREE_ID_alloc: 1400 case BTREE_ID_backpointers: 1401 case BTREE_ID_need_discard: 1402 case BTREE_ID_freespace: 1403 case BTREE_ID_bucket_gens: 1404 case BTREE_ID_lru: 1405 case BTREE_ID_accounting: 1406 return true; 1407 default: 1408 return false; 1409 } 1410 } 1411 1412 #define BTREE_MAX_DEPTH 4U 1413 1414 /* Btree nodes */ 1415 1416 /* 1417 * Btree nodes 1418 * 1419 * On disk a btree node is a list/log of these; within each set the keys are 1420 * sorted 1421 */ 1422 struct bset { 1423 __le64 seq; 1424 1425 /* 1426 * Highest journal entry this bset contains keys for. 1427 * If on recovery we don't see that journal entry, this bset is ignored: 1428 * this allows us to preserve the order of all index updates after a 1429 * crash, since the journal records a total order of all index updates 1430 * and anything that didn't make it to the journal doesn't get used. 1431 */ 1432 __le64 journal_seq; 1433 1434 __le32 flags; 1435 __le16 version; 1436 __le16 u64s; /* count of d[] in u64s */ 1437 1438 struct bkey_packed start[0]; 1439 __u64 _data[]; 1440 } __packed __aligned(8); 1441 1442 LE32_BITMASK(BSET_CSUM_TYPE, struct bset, flags, 0, 4); 1443 1444 LE32_BITMASK(BSET_BIG_ENDIAN, struct bset, flags, 4, 5); 1445 LE32_BITMASK(BSET_SEPARATE_WHITEOUTS, 1446 struct bset, flags, 5, 6); 1447 1448 /* Sector offset within the btree node: */ 1449 LE32_BITMASK(BSET_OFFSET, struct bset, flags, 16, 32); 1450 1451 struct btree_node { 1452 struct bch_csum csum; 1453 __le64 magic; 1454 1455 /* this flags field is encrypted, unlike bset->flags: */ 1456 __le64 flags; 1457 1458 /* Closed interval: */ 1459 struct bpos min_key; 1460 struct bpos max_key; 1461 struct bch_extent_ptr _ptr; /* not used anymore */ 1462 struct bkey_format format; 1463 1464 union { 1465 struct bset keys; 1466 struct { 1467 __u8 pad[22]; 1468 __le16 u64s; 1469 __u64 _data[0]; 1470 1471 }; 1472 }; 1473 } __packed __aligned(8); 1474 1475 LE64_BITMASK(BTREE_NODE_ID_LO, struct btree_node, flags, 0, 4); 1476 LE64_BITMASK(BTREE_NODE_LEVEL, struct btree_node, flags, 4, 8); 1477 LE64_BITMASK(BTREE_NODE_NEW_EXTENT_OVERWRITE, 1478 struct btree_node, flags, 8, 9); 1479 LE64_BITMASK(BTREE_NODE_ID_HI, struct btree_node, flags, 9, 25); 1480 /* 25-32 unused */ 1481 LE64_BITMASK(BTREE_NODE_SEQ, struct btree_node, flags, 32, 64); 1482 1483 static inline __u64 BTREE_NODE_ID(struct btree_node *n) 1484 { 1485 return BTREE_NODE_ID_LO(n) | (BTREE_NODE_ID_HI(n) << 4); 1486 } 1487 1488 static inline void SET_BTREE_NODE_ID(struct btree_node *n, __u64 v) 1489 { 1490 SET_BTREE_NODE_ID_LO(n, v); 1491 SET_BTREE_NODE_ID_HI(n, v >> 4); 1492 } 1493 1494 struct btree_node_entry { 1495 struct bch_csum csum; 1496 1497 union { 1498 struct bset keys; 1499 struct { 1500 __u8 pad[22]; 1501 __le16 u64s; 1502 __u64 _data[0]; 1503 }; 1504 }; 1505 } __packed __aligned(8); 1506 1507 #endif /* _BCACHEFS_FORMAT_H */ 1508