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__ 82 typedef uuid_t __uuid_t; 83 #endif 84 85 #define BITMASK(name, type, field, offset, end) \ 86 static const __maybe_unused unsigned name##_OFFSET = offset; \ 87 static const __maybe_unused unsigned name##_BITS = (end - offset); \ 88 \ 89 static inline __u64 name(const type *k) \ 90 { \ 91 return (k->field >> offset) & ~(~0ULL << (end - offset)); \ 92 } \ 93 \ 94 static 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) \ 101 static const __maybe_unused unsigned name##_OFFSET = offset; \ 102 static const __maybe_unused unsigned name##_BITS = (end - offset); \ 103 static const __maybe_unused __u##_bits name##_MAX = (1ULL << (end - offset)) - 1;\ 104 \ 105 static inline __u64 name(const type *k) \ 106 { \ 107 return (__le##_bits##_to_cpu(k->field) >> offset) & \ 108 ~(~0ULL << (end - offset)); \ 109 } \ 110 \ 111 static 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) 123 124 struct 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 */ 133 134 struct 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) 164 165 static 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() */ 180 struct bch_val { 181 __u64 __nothing[0]; 182 }; 183 184 struct 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 __aligned(4); 193 194 struct bkey { 195 /* Size of combined key and value, in u64s */ 196 __u8 u64s; 197 198 /* Format of key (0 for format local to btree node) */ 199 #if defined(__LITTLE_ENDIAN_BITFIELD) 200 __u8 format:7, 201 needs_whiteout:1; 202 #elif defined (__BIG_ENDIAN_BITFIELD) 203 __u8 needs_whiteout:1, 204 format:7; 205 #else 206 #error edit for your odd byteorder. 207 #endif 208 209 /* Type of the value */ 210 __u8 type; 211 212 #if __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__ 213 __u8 pad[1]; 214 215 struct bversion version; 216 __u32 size; /* extent size, in sectors */ 217 struct bpos p; 218 #elif __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__ 219 struct bpos p; 220 __u32 size; /* extent size, in sectors */ 221 struct bversion version; 222 223 __u8 pad[1]; 224 #endif 225 } __packed __aligned(8); 226 227 struct bkey_packed { 228 __u64 _data[0]; 229 230 /* Size of combined key and value, in u64s */ 231 __u8 u64s; 232 233 /* Format of key (0 for format local to btree node) */ 234 235 /* 236 * XXX: next incompat on disk format change, switch format and 237 * needs_whiteout - bkey_packed() will be cheaper if format is the high 238 * bits of the bitfield 239 */ 240 #if defined(__LITTLE_ENDIAN_BITFIELD) 241 __u8 format:7, 242 needs_whiteout:1; 243 #elif defined (__BIG_ENDIAN_BITFIELD) 244 __u8 needs_whiteout:1, 245 format:7; 246 #endif 247 248 /* Type of the value */ 249 __u8 type; 250 __u8 key_start[0]; 251 252 /* 253 * We copy bkeys with struct assignment in various places, and while 254 * that shouldn't be done with packed bkeys we can't disallow it in C, 255 * and it's legal to cast a bkey to a bkey_packed - so padding it out 256 * to the same size as struct bkey should hopefully be safest. 257 */ 258 __u8 pad[sizeof(struct bkey) - 3]; 259 } __packed __aligned(8); 260 261 typedef struct { 262 __le64 lo; 263 __le64 hi; 264 } bch_le128; 265 266 #define BKEY_U64s (sizeof(struct bkey) / sizeof(__u64)) 267 #define BKEY_U64s_MAX U8_MAX 268 #define BKEY_VAL_U64s_MAX (BKEY_U64s_MAX - BKEY_U64s) 269 270 #define KEY_PACKED_BITS_START 24 271 272 #define KEY_FORMAT_LOCAL_BTREE 0 273 #define KEY_FORMAT_CURRENT 1 274 275 enum bch_bkey_fields { 276 BKEY_FIELD_INODE, 277 BKEY_FIELD_OFFSET, 278 BKEY_FIELD_SNAPSHOT, 279 BKEY_FIELD_SIZE, 280 BKEY_FIELD_VERSION_HI, 281 BKEY_FIELD_VERSION_LO, 282 BKEY_NR_FIELDS, 283 }; 284 285 #define bkey_format_field(name, field) \ 286 [BKEY_FIELD_##name] = (sizeof(((struct bkey *) NULL)->field) * 8) 287 288 #define BKEY_FORMAT_CURRENT \ 289 ((struct bkey_format) { \ 290 .key_u64s = BKEY_U64s, \ 291 .nr_fields = BKEY_NR_FIELDS, \ 292 .bits_per_field = { \ 293 bkey_format_field(INODE, p.inode), \ 294 bkey_format_field(OFFSET, p.offset), \ 295 bkey_format_field(SNAPSHOT, p.snapshot), \ 296 bkey_format_field(SIZE, size), \ 297 bkey_format_field(VERSION_HI, version.hi), \ 298 bkey_format_field(VERSION_LO, version.lo), \ 299 }, \ 300 }) 301 302 /* bkey with inline value */ 303 struct bkey_i { 304 __u64 _data[0]; 305 306 struct bkey k; 307 struct bch_val v; 308 }; 309 310 #define KEY(_inode, _offset, _size) \ 311 ((struct bkey) { \ 312 .u64s = BKEY_U64s, \ 313 .format = KEY_FORMAT_CURRENT, \ 314 .p = POS(_inode, _offset), \ 315 .size = _size, \ 316 }) 317 318 static inline void bkey_init(struct bkey *k) 319 { 320 *k = KEY(0, 0, 0); 321 } 322 323 #define bkey_bytes(_k) ((_k)->u64s * sizeof(__u64)) 324 325 #define __BKEY_PADDED(key, pad) \ 326 struct bkey_i key; __u64 key ## _pad[pad] 327 328 /* 329 * - DELETED keys are used internally to mark keys that should be ignored but 330 * override keys in composition order. Their version number is ignored. 331 * 332 * - DISCARDED keys indicate that the data is all 0s because it has been 333 * discarded. DISCARDs may have a version; if the version is nonzero the key 334 * will be persistent, otherwise the key will be dropped whenever the btree 335 * node is rewritten (like DELETED keys). 336 * 337 * - ERROR: any read of the data returns a read error, as the data was lost due 338 * to a failing device. Like DISCARDED keys, they can be removed (overridden) 339 * by new writes or cluster-wide GC. Node repair can also overwrite them with 340 * the same or a more recent version number, but not with an older version 341 * number. 342 * 343 * - WHITEOUT: for hash table btrees 344 */ 345 #define BCH_BKEY_TYPES() \ 346 x(deleted, 0) \ 347 x(whiteout, 1) \ 348 x(error, 2) \ 349 x(cookie, 3) \ 350 x(hash_whiteout, 4) \ 351 x(btree_ptr, 5) \ 352 x(extent, 6) \ 353 x(reservation, 7) \ 354 x(inode, 8) \ 355 x(inode_generation, 9) \ 356 x(dirent, 10) \ 357 x(xattr, 11) \ 358 x(alloc, 12) \ 359 x(quota, 13) \ 360 x(stripe, 14) \ 361 x(reflink_p, 15) \ 362 x(reflink_v, 16) \ 363 x(inline_data, 17) \ 364 x(btree_ptr_v2, 18) \ 365 x(indirect_inline_data, 19) \ 366 x(alloc_v2, 20) \ 367 x(subvolume, 21) \ 368 x(snapshot, 22) \ 369 x(inode_v2, 23) \ 370 x(alloc_v3, 24) \ 371 x(set, 25) \ 372 x(lru, 26) \ 373 x(alloc_v4, 27) \ 374 x(backpointer, 28) \ 375 x(inode_v3, 29) \ 376 x(bucket_gens, 30) \ 377 x(snapshot_tree, 31) \ 378 x(logged_op_truncate, 32) \ 379 x(logged_op_finsert, 33) 380 381 enum bch_bkey_type { 382 #define x(name, nr) KEY_TYPE_##name = nr, 383 BCH_BKEY_TYPES() 384 #undef x 385 KEY_TYPE_MAX, 386 }; 387 388 struct bch_deleted { 389 struct bch_val v; 390 }; 391 392 struct bch_whiteout { 393 struct bch_val v; 394 }; 395 396 struct bch_error { 397 struct bch_val v; 398 }; 399 400 struct bch_cookie { 401 struct bch_val v; 402 __le64 cookie; 403 }; 404 405 struct bch_hash_whiteout { 406 struct bch_val v; 407 }; 408 409 struct bch_set { 410 struct bch_val v; 411 }; 412 413 /* Extents */ 414 415 /* 416 * In extent bkeys, the value is a list of pointers (bch_extent_ptr), optionally 417 * preceded by checksum/compression information (bch_extent_crc32 or 418 * bch_extent_crc64). 419 * 420 * One major determining factor in the format of extents is how we handle and 421 * represent extents that have been partially overwritten and thus trimmed: 422 * 423 * If an extent is not checksummed or compressed, when the extent is trimmed we 424 * don't have to remember the extent we originally allocated and wrote: we can 425 * merely adjust ptr->offset to point to the start of the data that is currently 426 * live. The size field in struct bkey records the current (live) size of the 427 * extent, and is also used to mean "size of region on disk that we point to" in 428 * this case. 429 * 430 * Thus an extent that is not checksummed or compressed will consist only of a 431 * list of bch_extent_ptrs, with none of the fields in 432 * bch_extent_crc32/bch_extent_crc64. 433 * 434 * When an extent is checksummed or compressed, it's not possible to read only 435 * the data that is currently live: we have to read the entire extent that was 436 * originally written, and then return only the part of the extent that is 437 * currently live. 438 * 439 * Thus, in addition to the current size of the extent in struct bkey, we need 440 * to store the size of the originally allocated space - this is the 441 * compressed_size and uncompressed_size fields in bch_extent_crc32/64. Also, 442 * when the extent is trimmed, instead of modifying the offset field of the 443 * pointer, we keep a second smaller offset field - "offset into the original 444 * extent of the currently live region". 445 * 446 * The other major determining factor is replication and data migration: 447 * 448 * Each pointer may have its own bch_extent_crc32/64. When doing a replicated 449 * write, we will initially write all the replicas in the same format, with the 450 * same checksum type and compression format - however, when copygc runs later (or 451 * tiering/cache promotion, anything that moves data), it is not in general 452 * going to rewrite all the pointers at once - one of the replicas may be in a 453 * bucket on one device that has very little fragmentation while another lives 454 * in a bucket that has become heavily fragmented, and thus is being rewritten 455 * sooner than the rest. 456 * 457 * Thus it will only move a subset of the pointers (or in the case of 458 * tiering/cache promotion perhaps add a single pointer without dropping any 459 * current pointers), and if the extent has been partially overwritten it must 460 * write only the currently live portion (or copygc would not be able to reduce 461 * fragmentation!) - which necessitates a different bch_extent_crc format for 462 * the new pointer. 463 * 464 * But in the interests of space efficiency, we don't want to store one 465 * bch_extent_crc for each pointer if we don't have to. 466 * 467 * Thus, a bch_extent consists of bch_extent_crc32s, bch_extent_crc64s, and 468 * bch_extent_ptrs appended arbitrarily one after the other. We determine the 469 * type of a given entry with a scheme similar to utf8 (except we're encoding a 470 * type, not a size), encoding the type in the position of the first set bit: 471 * 472 * bch_extent_crc32 - 0b1 473 * bch_extent_ptr - 0b10 474 * bch_extent_crc64 - 0b100 475 * 476 * We do it this way because bch_extent_crc32 is _very_ constrained on bits (and 477 * bch_extent_crc64 is the least constrained). 478 * 479 * Then, each bch_extent_crc32/64 applies to the pointers that follow after it, 480 * until the next bch_extent_crc32/64. 481 * 482 * If there are no bch_extent_crcs preceding a bch_extent_ptr, then that pointer 483 * is neither checksummed nor compressed. 484 */ 485 486 /* 128 bits, sufficient for cryptographic MACs: */ 487 struct bch_csum { 488 __le64 lo; 489 __le64 hi; 490 } __packed __aligned(8); 491 492 #define BCH_EXTENT_ENTRY_TYPES() \ 493 x(ptr, 0) \ 494 x(crc32, 1) \ 495 x(crc64, 2) \ 496 x(crc128, 3) \ 497 x(stripe_ptr, 4) \ 498 x(rebalance, 5) 499 #define BCH_EXTENT_ENTRY_MAX 6 500 501 enum bch_extent_entry_type { 502 #define x(f, n) BCH_EXTENT_ENTRY_##f = n, 503 BCH_EXTENT_ENTRY_TYPES() 504 #undef x 505 }; 506 507 /* Compressed/uncompressed size are stored biased by 1: */ 508 struct bch_extent_crc32 { 509 #if defined(__LITTLE_ENDIAN_BITFIELD) 510 __u32 type:2, 511 _compressed_size:7, 512 _uncompressed_size:7, 513 offset:7, 514 _unused:1, 515 csum_type:4, 516 compression_type:4; 517 __u32 csum; 518 #elif defined (__BIG_ENDIAN_BITFIELD) 519 __u32 csum; 520 __u32 compression_type:4, 521 csum_type:4, 522 _unused:1, 523 offset:7, 524 _uncompressed_size:7, 525 _compressed_size:7, 526 type:2; 527 #endif 528 } __packed __aligned(8); 529 530 #define CRC32_SIZE_MAX (1U << 7) 531 #define CRC32_NONCE_MAX 0 532 533 struct bch_extent_crc64 { 534 #if defined(__LITTLE_ENDIAN_BITFIELD) 535 __u64 type:3, 536 _compressed_size:9, 537 _uncompressed_size:9, 538 offset:9, 539 nonce:10, 540 csum_type:4, 541 compression_type:4, 542 csum_hi:16; 543 #elif defined (__BIG_ENDIAN_BITFIELD) 544 __u64 csum_hi:16, 545 compression_type:4, 546 csum_type:4, 547 nonce:10, 548 offset:9, 549 _uncompressed_size:9, 550 _compressed_size:9, 551 type:3; 552 #endif 553 __u64 csum_lo; 554 } __packed __aligned(8); 555 556 #define CRC64_SIZE_MAX (1U << 9) 557 #define CRC64_NONCE_MAX ((1U << 10) - 1) 558 559 struct bch_extent_crc128 { 560 #if defined(__LITTLE_ENDIAN_BITFIELD) 561 __u64 type:4, 562 _compressed_size:13, 563 _uncompressed_size:13, 564 offset:13, 565 nonce:13, 566 csum_type:4, 567 compression_type:4; 568 #elif defined (__BIG_ENDIAN_BITFIELD) 569 __u64 compression_type:4, 570 csum_type:4, 571 nonce:13, 572 offset:13, 573 _uncompressed_size:13, 574 _compressed_size:13, 575 type:4; 576 #endif 577 struct bch_csum csum; 578 } __packed __aligned(8); 579 580 #define CRC128_SIZE_MAX (1U << 13) 581 #define CRC128_NONCE_MAX ((1U << 13) - 1) 582 583 /* 584 * @reservation - pointer hasn't been written to, just reserved 585 */ 586 struct bch_extent_ptr { 587 #if defined(__LITTLE_ENDIAN_BITFIELD) 588 __u64 type:1, 589 cached:1, 590 unused:1, 591 unwritten:1, 592 offset:44, /* 8 petabytes */ 593 dev:8, 594 gen:8; 595 #elif defined (__BIG_ENDIAN_BITFIELD) 596 __u64 gen:8, 597 dev:8, 598 offset:44, 599 unwritten:1, 600 unused:1, 601 cached:1, 602 type:1; 603 #endif 604 } __packed __aligned(8); 605 606 struct bch_extent_stripe_ptr { 607 #if defined(__LITTLE_ENDIAN_BITFIELD) 608 __u64 type:5, 609 block:8, 610 redundancy:4, 611 idx:47; 612 #elif defined (__BIG_ENDIAN_BITFIELD) 613 __u64 idx:47, 614 redundancy:4, 615 block:8, 616 type:5; 617 #endif 618 }; 619 620 struct bch_extent_rebalance { 621 #if defined(__LITTLE_ENDIAN_BITFIELD) 622 __u64 type:6, 623 unused:34, 624 compression:8, /* enum bch_compression_opt */ 625 target:16; 626 #elif defined (__BIG_ENDIAN_BITFIELD) 627 __u64 target:16, 628 compression:8, 629 unused:34, 630 type:6; 631 #endif 632 }; 633 634 union bch_extent_entry { 635 #if __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__ || __BITS_PER_LONG == 64 636 unsigned long type; 637 #elif __BITS_PER_LONG == 32 638 struct { 639 unsigned long pad; 640 unsigned long type; 641 }; 642 #else 643 #error edit for your odd byteorder. 644 #endif 645 646 #define x(f, n) struct bch_extent_##f f; 647 BCH_EXTENT_ENTRY_TYPES() 648 #undef x 649 }; 650 651 struct bch_btree_ptr { 652 struct bch_val v; 653 654 __u64 _data[0]; 655 struct bch_extent_ptr start[]; 656 } __packed __aligned(8); 657 658 struct bch_btree_ptr_v2 { 659 struct bch_val v; 660 661 __u64 mem_ptr; 662 __le64 seq; 663 __le16 sectors_written; 664 __le16 flags; 665 struct bpos min_key; 666 __u64 _data[0]; 667 struct bch_extent_ptr start[]; 668 } __packed __aligned(8); 669 670 LE16_BITMASK(BTREE_PTR_RANGE_UPDATED, struct bch_btree_ptr_v2, flags, 0, 1); 671 672 struct bch_extent { 673 struct bch_val v; 674 675 __u64 _data[0]; 676 union bch_extent_entry start[]; 677 } __packed __aligned(8); 678 679 struct bch_reservation { 680 struct bch_val v; 681 682 __le32 generation; 683 __u8 nr_replicas; 684 __u8 pad[3]; 685 } __packed __aligned(8); 686 687 /* Maximum size (in u64s) a single pointer could be: */ 688 #define BKEY_EXTENT_PTR_U64s_MAX\ 689 ((sizeof(struct bch_extent_crc128) + \ 690 sizeof(struct bch_extent_ptr)) / sizeof(__u64)) 691 692 /* Maximum possible size of an entire extent value: */ 693 #define BKEY_EXTENT_VAL_U64s_MAX \ 694 (1 + BKEY_EXTENT_PTR_U64s_MAX * (BCH_REPLICAS_MAX + 1)) 695 696 /* * Maximum possible size of an entire extent, key + value: */ 697 #define BKEY_EXTENT_U64s_MAX (BKEY_U64s + BKEY_EXTENT_VAL_U64s_MAX) 698 699 /* Btree pointers don't carry around checksums: */ 700 #define BKEY_BTREE_PTR_VAL_U64s_MAX \ 701 ((sizeof(struct bch_btree_ptr_v2) + \ 702 sizeof(struct bch_extent_ptr) * BCH_REPLICAS_MAX) / sizeof(__u64)) 703 #define BKEY_BTREE_PTR_U64s_MAX \ 704 (BKEY_U64s + BKEY_BTREE_PTR_VAL_U64s_MAX) 705 706 /* Inodes */ 707 708 #define BLOCKDEV_INODE_MAX 4096 709 710 #define BCACHEFS_ROOT_INO 4096 711 712 struct bch_inode { 713 struct bch_val v; 714 715 __le64 bi_hash_seed; 716 __le32 bi_flags; 717 __le16 bi_mode; 718 __u8 fields[]; 719 } __packed __aligned(8); 720 721 struct bch_inode_v2 { 722 struct bch_val v; 723 724 __le64 bi_journal_seq; 725 __le64 bi_hash_seed; 726 __le64 bi_flags; 727 __le16 bi_mode; 728 __u8 fields[]; 729 } __packed __aligned(8); 730 731 struct bch_inode_v3 { 732 struct bch_val v; 733 734 __le64 bi_journal_seq; 735 __le64 bi_hash_seed; 736 __le64 bi_flags; 737 __le64 bi_sectors; 738 __le64 bi_size; 739 __le64 bi_version; 740 __u8 fields[]; 741 } __packed __aligned(8); 742 743 #define INODEv3_FIELDS_START_INITIAL 6 744 #define INODEv3_FIELDS_START_CUR (offsetof(struct bch_inode_v3, fields) / sizeof(__u64)) 745 746 struct bch_inode_generation { 747 struct bch_val v; 748 749 __le32 bi_generation; 750 __le32 pad; 751 } __packed __aligned(8); 752 753 /* 754 * bi_subvol and bi_parent_subvol are only set for subvolume roots: 755 */ 756 757 #define BCH_INODE_FIELDS_v2() \ 758 x(bi_atime, 96) \ 759 x(bi_ctime, 96) \ 760 x(bi_mtime, 96) \ 761 x(bi_otime, 96) \ 762 x(bi_size, 64) \ 763 x(bi_sectors, 64) \ 764 x(bi_uid, 32) \ 765 x(bi_gid, 32) \ 766 x(bi_nlink, 32) \ 767 x(bi_generation, 32) \ 768 x(bi_dev, 32) \ 769 x(bi_data_checksum, 8) \ 770 x(bi_compression, 8) \ 771 x(bi_project, 32) \ 772 x(bi_background_compression, 8) \ 773 x(bi_data_replicas, 8) \ 774 x(bi_promote_target, 16) \ 775 x(bi_foreground_target, 16) \ 776 x(bi_background_target, 16) \ 777 x(bi_erasure_code, 16) \ 778 x(bi_fields_set, 16) \ 779 x(bi_dir, 64) \ 780 x(bi_dir_offset, 64) \ 781 x(bi_subvol, 32) \ 782 x(bi_parent_subvol, 32) 783 784 #define BCH_INODE_FIELDS_v3() \ 785 x(bi_atime, 96) \ 786 x(bi_ctime, 96) \ 787 x(bi_mtime, 96) \ 788 x(bi_otime, 96) \ 789 x(bi_uid, 32) \ 790 x(bi_gid, 32) \ 791 x(bi_nlink, 32) \ 792 x(bi_generation, 32) \ 793 x(bi_dev, 32) \ 794 x(bi_data_checksum, 8) \ 795 x(bi_compression, 8) \ 796 x(bi_project, 32) \ 797 x(bi_background_compression, 8) \ 798 x(bi_data_replicas, 8) \ 799 x(bi_promote_target, 16) \ 800 x(bi_foreground_target, 16) \ 801 x(bi_background_target, 16) \ 802 x(bi_erasure_code, 16) \ 803 x(bi_fields_set, 16) \ 804 x(bi_dir, 64) \ 805 x(bi_dir_offset, 64) \ 806 x(bi_subvol, 32) \ 807 x(bi_parent_subvol, 32) \ 808 x(bi_nocow, 8) 809 810 /* subset of BCH_INODE_FIELDS */ 811 #define BCH_INODE_OPTS() \ 812 x(data_checksum, 8) \ 813 x(compression, 8) \ 814 x(project, 32) \ 815 x(background_compression, 8) \ 816 x(data_replicas, 8) \ 817 x(promote_target, 16) \ 818 x(foreground_target, 16) \ 819 x(background_target, 16) \ 820 x(erasure_code, 16) \ 821 x(nocow, 8) 822 823 enum inode_opt_id { 824 #define x(name, ...) \ 825 Inode_opt_##name, 826 BCH_INODE_OPTS() 827 #undef x 828 Inode_opt_nr, 829 }; 830 831 #define BCH_INODE_FLAGS() \ 832 x(sync, 0) \ 833 x(immutable, 1) \ 834 x(append, 2) \ 835 x(nodump, 3) \ 836 x(noatime, 4) \ 837 x(i_size_dirty, 5) \ 838 x(i_sectors_dirty, 6) \ 839 x(unlinked, 7) \ 840 x(backptr_untrusted, 8) 841 842 /* bits 20+ reserved for packed fields below: */ 843 844 enum bch_inode_flags { 845 #define x(t, n) BCH_INODE_##t = 1U << n, 846 BCH_INODE_FLAGS() 847 #undef x 848 }; 849 850 enum __bch_inode_flags { 851 #define x(t, n) __BCH_INODE_##t = n, 852 BCH_INODE_FLAGS() 853 #undef x 854 }; 855 856 LE32_BITMASK(INODE_STR_HASH, struct bch_inode, bi_flags, 20, 24); 857 LE32_BITMASK(INODE_NR_FIELDS, struct bch_inode, bi_flags, 24, 31); 858 LE32_BITMASK(INODE_NEW_VARINT, struct bch_inode, bi_flags, 31, 32); 859 860 LE64_BITMASK(INODEv2_STR_HASH, struct bch_inode_v2, bi_flags, 20, 24); 861 LE64_BITMASK(INODEv2_NR_FIELDS, struct bch_inode_v2, bi_flags, 24, 31); 862 863 LE64_BITMASK(INODEv3_STR_HASH, struct bch_inode_v3, bi_flags, 20, 24); 864 LE64_BITMASK(INODEv3_NR_FIELDS, struct bch_inode_v3, bi_flags, 24, 31); 865 866 LE64_BITMASK(INODEv3_FIELDS_START, 867 struct bch_inode_v3, bi_flags, 31, 36); 868 LE64_BITMASK(INODEv3_MODE, struct bch_inode_v3, bi_flags, 36, 52); 869 870 /* Dirents */ 871 872 /* 873 * Dirents (and xattrs) have to implement string lookups; since our b-tree 874 * doesn't support arbitrary length strings for the key, we instead index by a 875 * 64 bit hash (currently truncated sha1) of the string, stored in the offset 876 * field of the key - using linear probing to resolve hash collisions. This also 877 * provides us with the readdir cookie posix requires. 878 * 879 * Linear probing requires us to use whiteouts for deletions, in the event of a 880 * collision: 881 */ 882 883 struct bch_dirent { 884 struct bch_val v; 885 886 /* Target inode number: */ 887 union { 888 __le64 d_inum; 889 struct { /* DT_SUBVOL */ 890 __le32 d_child_subvol; 891 __le32 d_parent_subvol; 892 }; 893 }; 894 895 /* 896 * Copy of mode bits 12-15 from the target inode - so userspace can get 897 * the filetype without having to do a stat() 898 */ 899 __u8 d_type; 900 901 __u8 d_name[]; 902 } __packed __aligned(8); 903 904 #define DT_SUBVOL 16 905 #define BCH_DT_MAX 17 906 907 #define BCH_NAME_MAX 512 908 909 /* Xattrs */ 910 911 #define KEY_TYPE_XATTR_INDEX_USER 0 912 #define KEY_TYPE_XATTR_INDEX_POSIX_ACL_ACCESS 1 913 #define KEY_TYPE_XATTR_INDEX_POSIX_ACL_DEFAULT 2 914 #define KEY_TYPE_XATTR_INDEX_TRUSTED 3 915 #define KEY_TYPE_XATTR_INDEX_SECURITY 4 916 917 struct bch_xattr { 918 struct bch_val v; 919 __u8 x_type; 920 __u8 x_name_len; 921 __le16 x_val_len; 922 __u8 x_name[]; 923 } __packed __aligned(8); 924 925 /* Bucket/allocation information: */ 926 927 struct bch_alloc { 928 struct bch_val v; 929 __u8 fields; 930 __u8 gen; 931 __u8 data[]; 932 } __packed __aligned(8); 933 934 #define BCH_ALLOC_FIELDS_V1() \ 935 x(read_time, 16) \ 936 x(write_time, 16) \ 937 x(data_type, 8) \ 938 x(dirty_sectors, 16) \ 939 x(cached_sectors, 16) \ 940 x(oldest_gen, 8) \ 941 x(stripe, 32) \ 942 x(stripe_redundancy, 8) 943 944 enum { 945 #define x(name, _bits) BCH_ALLOC_FIELD_V1_##name, 946 BCH_ALLOC_FIELDS_V1() 947 #undef x 948 }; 949 950 struct bch_alloc_v2 { 951 struct bch_val v; 952 __u8 nr_fields; 953 __u8 gen; 954 __u8 oldest_gen; 955 __u8 data_type; 956 __u8 data[]; 957 } __packed __aligned(8); 958 959 #define BCH_ALLOC_FIELDS_V2() \ 960 x(read_time, 64) \ 961 x(write_time, 64) \ 962 x(dirty_sectors, 32) \ 963 x(cached_sectors, 32) \ 964 x(stripe, 32) \ 965 x(stripe_redundancy, 8) 966 967 struct bch_alloc_v3 { 968 struct bch_val v; 969 __le64 journal_seq; 970 __le32 flags; 971 __u8 nr_fields; 972 __u8 gen; 973 __u8 oldest_gen; 974 __u8 data_type; 975 __u8 data[]; 976 } __packed __aligned(8); 977 978 LE32_BITMASK(BCH_ALLOC_V3_NEED_DISCARD,struct bch_alloc_v3, flags, 0, 1) 979 LE32_BITMASK(BCH_ALLOC_V3_NEED_INC_GEN,struct bch_alloc_v3, flags, 1, 2) 980 981 struct bch_alloc_v4 { 982 struct bch_val v; 983 __u64 journal_seq; 984 __u32 flags; 985 __u8 gen; 986 __u8 oldest_gen; 987 __u8 data_type; 988 __u8 stripe_redundancy; 989 __u32 dirty_sectors; 990 __u32 cached_sectors; 991 __u64 io_time[2]; 992 __u32 stripe; 993 __u32 nr_external_backpointers; 994 __u64 fragmentation_lru; 995 } __packed __aligned(8); 996 997 #define BCH_ALLOC_V4_U64s_V0 6 998 #define BCH_ALLOC_V4_U64s (sizeof(struct bch_alloc_v4) / sizeof(__u64)) 999 1000 BITMASK(BCH_ALLOC_V4_NEED_DISCARD, struct bch_alloc_v4, flags, 0, 1) 1001 BITMASK(BCH_ALLOC_V4_NEED_INC_GEN, struct bch_alloc_v4, flags, 1, 2) 1002 BITMASK(BCH_ALLOC_V4_BACKPOINTERS_START,struct bch_alloc_v4, flags, 2, 8) 1003 BITMASK(BCH_ALLOC_V4_NR_BACKPOINTERS, struct bch_alloc_v4, flags, 8, 14) 1004 1005 #define BCH_ALLOC_V4_NR_BACKPOINTERS_MAX 40 1006 1007 struct bch_backpointer { 1008 struct bch_val v; 1009 __u8 btree_id; 1010 __u8 level; 1011 __u8 data_type; 1012 __u64 bucket_offset:40; 1013 __u32 bucket_len; 1014 struct bpos pos; 1015 } __packed __aligned(8); 1016 1017 #define KEY_TYPE_BUCKET_GENS_BITS 8 1018 #define KEY_TYPE_BUCKET_GENS_NR (1U << KEY_TYPE_BUCKET_GENS_BITS) 1019 #define KEY_TYPE_BUCKET_GENS_MASK (KEY_TYPE_BUCKET_GENS_NR - 1) 1020 1021 struct bch_bucket_gens { 1022 struct bch_val v; 1023 u8 gens[KEY_TYPE_BUCKET_GENS_NR]; 1024 } __packed __aligned(8); 1025 1026 /* Quotas: */ 1027 1028 enum quota_types { 1029 QTYP_USR = 0, 1030 QTYP_GRP = 1, 1031 QTYP_PRJ = 2, 1032 QTYP_NR = 3, 1033 }; 1034 1035 enum quota_counters { 1036 Q_SPC = 0, 1037 Q_INO = 1, 1038 Q_COUNTERS = 2, 1039 }; 1040 1041 struct bch_quota_counter { 1042 __le64 hardlimit; 1043 __le64 softlimit; 1044 }; 1045 1046 struct bch_quota { 1047 struct bch_val v; 1048 struct bch_quota_counter c[Q_COUNTERS]; 1049 } __packed __aligned(8); 1050 1051 /* Erasure coding */ 1052 1053 struct bch_stripe { 1054 struct bch_val v; 1055 __le16 sectors; 1056 __u8 algorithm; 1057 __u8 nr_blocks; 1058 __u8 nr_redundant; 1059 1060 __u8 csum_granularity_bits; 1061 __u8 csum_type; 1062 __u8 pad; 1063 1064 struct bch_extent_ptr ptrs[]; 1065 } __packed __aligned(8); 1066 1067 /* Reflink: */ 1068 1069 struct bch_reflink_p { 1070 struct bch_val v; 1071 __le64 idx; 1072 /* 1073 * A reflink pointer might point to an indirect extent which is then 1074 * later split (by copygc or rebalance). If we only pointed to part of 1075 * the original indirect extent, and then one of the fragments is 1076 * outside the range we point to, we'd leak a refcount: so when creating 1077 * reflink pointers, we need to store pad values to remember the full 1078 * range we were taking a reference on. 1079 */ 1080 __le32 front_pad; 1081 __le32 back_pad; 1082 } __packed __aligned(8); 1083 1084 struct bch_reflink_v { 1085 struct bch_val v; 1086 __le64 refcount; 1087 union bch_extent_entry start[0]; 1088 __u64 _data[]; 1089 } __packed __aligned(8); 1090 1091 struct bch_indirect_inline_data { 1092 struct bch_val v; 1093 __le64 refcount; 1094 u8 data[]; 1095 }; 1096 1097 /* Inline data */ 1098 1099 struct bch_inline_data { 1100 struct bch_val v; 1101 u8 data[]; 1102 }; 1103 1104 /* Subvolumes: */ 1105 1106 #define SUBVOL_POS_MIN POS(0, 1) 1107 #define SUBVOL_POS_MAX POS(0, S32_MAX) 1108 #define BCACHEFS_ROOT_SUBVOL 1 1109 1110 struct bch_subvolume { 1111 struct bch_val v; 1112 __le32 flags; 1113 __le32 snapshot; 1114 __le64 inode; 1115 /* 1116 * Snapshot subvolumes form a tree, separate from the snapshot nodes 1117 * tree - if this subvolume is a snapshot, this is the ID of the 1118 * subvolume it was created from: 1119 */ 1120 __le32 parent; 1121 __le32 pad; 1122 bch_le128 otime; 1123 }; 1124 1125 LE32_BITMASK(BCH_SUBVOLUME_RO, struct bch_subvolume, flags, 0, 1) 1126 /* 1127 * We need to know whether a subvolume is a snapshot so we can know whether we 1128 * can delete it (or whether it should just be rm -rf'd) 1129 */ 1130 LE32_BITMASK(BCH_SUBVOLUME_SNAP, struct bch_subvolume, flags, 1, 2) 1131 LE32_BITMASK(BCH_SUBVOLUME_UNLINKED, struct bch_subvolume, flags, 2, 3) 1132 1133 /* Snapshots */ 1134 1135 struct bch_snapshot { 1136 struct bch_val v; 1137 __le32 flags; 1138 __le32 parent; 1139 __le32 children[2]; 1140 __le32 subvol; 1141 /* corresponds to a bch_snapshot_tree in BTREE_ID_snapshot_trees */ 1142 __le32 tree; 1143 __le32 depth; 1144 __le32 skip[3]; 1145 }; 1146 1147 LE32_BITMASK(BCH_SNAPSHOT_DELETED, struct bch_snapshot, flags, 0, 1) 1148 1149 /* True if a subvolume points to this snapshot node: */ 1150 LE32_BITMASK(BCH_SNAPSHOT_SUBVOL, struct bch_snapshot, flags, 1, 2) 1151 1152 /* 1153 * Snapshot trees: 1154 * 1155 * The snapshot_trees btree gives us persistent indentifier for each tree of 1156 * bch_snapshot nodes, and allow us to record and easily find the root/master 1157 * subvolume that other snapshots were created from: 1158 */ 1159 struct bch_snapshot_tree { 1160 struct bch_val v; 1161 __le32 master_subvol; 1162 __le32 root_snapshot; 1163 }; 1164 1165 /* LRU btree: */ 1166 1167 struct bch_lru { 1168 struct bch_val v; 1169 __le64 idx; 1170 } __packed __aligned(8); 1171 1172 #define LRU_ID_STRIPES (1U << 16) 1173 1174 /* Logged operations btree: */ 1175 1176 struct bch_logged_op_truncate { 1177 struct bch_val v; 1178 __le32 subvol; 1179 __le32 pad; 1180 __le64 inum; 1181 __le64 new_i_size; 1182 }; 1183 1184 enum logged_op_finsert_state { 1185 LOGGED_OP_FINSERT_start, 1186 LOGGED_OP_FINSERT_shift_extents, 1187 LOGGED_OP_FINSERT_finish, 1188 }; 1189 1190 struct bch_logged_op_finsert { 1191 struct bch_val v; 1192 __u8 state; 1193 __u8 pad[3]; 1194 __le32 subvol; 1195 __le64 inum; 1196 __le64 dst_offset; 1197 __le64 src_offset; 1198 __le64 pos; 1199 }; 1200 1201 /* Optional/variable size superblock sections: */ 1202 1203 struct bch_sb_field { 1204 __u64 _data[0]; 1205 __le32 u64s; 1206 __le32 type; 1207 }; 1208 1209 #define BCH_SB_FIELDS() \ 1210 x(journal, 0) \ 1211 x(members_v1, 1) \ 1212 x(crypt, 2) \ 1213 x(replicas_v0, 3) \ 1214 x(quota, 4) \ 1215 x(disk_groups, 5) \ 1216 x(clean, 6) \ 1217 x(replicas, 7) \ 1218 x(journal_seq_blacklist, 8) \ 1219 x(journal_v2, 9) \ 1220 x(counters, 10) \ 1221 x(members_v2, 11) \ 1222 x(errors, 12) \ 1223 x(ext, 13) \ 1224 x(downgrade, 14) 1225 1226 enum bch_sb_field_type { 1227 #define x(f, nr) BCH_SB_FIELD_##f = nr, 1228 BCH_SB_FIELDS() 1229 #undef x 1230 BCH_SB_FIELD_NR 1231 }; 1232 1233 /* 1234 * Most superblock fields are replicated in all device's superblocks - a few are 1235 * not: 1236 */ 1237 #define BCH_SINGLE_DEVICE_SB_FIELDS \ 1238 ((1U << BCH_SB_FIELD_journal)| \ 1239 (1U << BCH_SB_FIELD_journal_v2)) 1240 1241 /* BCH_SB_FIELD_journal: */ 1242 1243 struct bch_sb_field_journal { 1244 struct bch_sb_field field; 1245 __le64 buckets[]; 1246 }; 1247 1248 struct bch_sb_field_journal_v2 { 1249 struct bch_sb_field field; 1250 1251 struct bch_sb_field_journal_v2_entry { 1252 __le64 start; 1253 __le64 nr; 1254 } d[]; 1255 }; 1256 1257 /* BCH_SB_FIELD_members_v1: */ 1258 1259 #define BCH_MIN_NR_NBUCKETS (1 << 6) 1260 1261 #define BCH_IOPS_MEASUREMENTS() \ 1262 x(seqread, 0) \ 1263 x(seqwrite, 1) \ 1264 x(randread, 2) \ 1265 x(randwrite, 3) 1266 1267 enum bch_iops_measurement { 1268 #define x(t, n) BCH_IOPS_##t = n, 1269 BCH_IOPS_MEASUREMENTS() 1270 #undef x 1271 BCH_IOPS_NR 1272 }; 1273 1274 #define BCH_MEMBER_ERROR_TYPES() \ 1275 x(read, 0) \ 1276 x(write, 1) \ 1277 x(checksum, 2) 1278 1279 enum bch_member_error_type { 1280 #define x(t, n) BCH_MEMBER_ERROR_##t = n, 1281 BCH_MEMBER_ERROR_TYPES() 1282 #undef x 1283 BCH_MEMBER_ERROR_NR 1284 }; 1285 1286 struct bch_member { 1287 __uuid_t uuid; 1288 __le64 nbuckets; /* device size */ 1289 __le16 first_bucket; /* index of first bucket used */ 1290 __le16 bucket_size; /* sectors */ 1291 __le32 pad; 1292 __le64 last_mount; /* time_t */ 1293 1294 __le64 flags; 1295 __le32 iops[4]; 1296 __le64 errors[BCH_MEMBER_ERROR_NR]; 1297 __le64 errors_at_reset[BCH_MEMBER_ERROR_NR]; 1298 __le64 errors_reset_time; 1299 }; 1300 1301 #define BCH_MEMBER_V1_BYTES 56 1302 1303 LE64_BITMASK(BCH_MEMBER_STATE, struct bch_member, flags, 0, 4) 1304 /* 4-14 unused, was TIER, HAS_(META)DATA, REPLACEMENT */ 1305 LE64_BITMASK(BCH_MEMBER_DISCARD, struct bch_member, flags, 14, 15) 1306 LE64_BITMASK(BCH_MEMBER_DATA_ALLOWED, struct bch_member, flags, 15, 20) 1307 LE64_BITMASK(BCH_MEMBER_GROUP, struct bch_member, flags, 20, 28) 1308 LE64_BITMASK(BCH_MEMBER_DURABILITY, struct bch_member, flags, 28, 30) 1309 LE64_BITMASK(BCH_MEMBER_FREESPACE_INITIALIZED, 1310 struct bch_member, flags, 30, 31) 1311 1312 #if 0 1313 LE64_BITMASK(BCH_MEMBER_NR_READ_ERRORS, struct bch_member, flags[1], 0, 20); 1314 LE64_BITMASK(BCH_MEMBER_NR_WRITE_ERRORS,struct bch_member, flags[1], 20, 40); 1315 #endif 1316 1317 #define BCH_MEMBER_STATES() \ 1318 x(rw, 0) \ 1319 x(ro, 1) \ 1320 x(failed, 2) \ 1321 x(spare, 3) 1322 1323 enum bch_member_state { 1324 #define x(t, n) BCH_MEMBER_STATE_##t = n, 1325 BCH_MEMBER_STATES() 1326 #undef x 1327 BCH_MEMBER_STATE_NR 1328 }; 1329 1330 struct bch_sb_field_members_v1 { 1331 struct bch_sb_field field; 1332 struct bch_member _members[]; //Members are now variable size 1333 }; 1334 1335 struct bch_sb_field_members_v2 { 1336 struct bch_sb_field field; 1337 __le16 member_bytes; //size of single member entry 1338 u8 pad[6]; 1339 struct bch_member _members[]; 1340 }; 1341 1342 /* BCH_SB_FIELD_crypt: */ 1343 1344 struct nonce { 1345 __le32 d[4]; 1346 }; 1347 1348 struct bch_key { 1349 __le64 key[4]; 1350 }; 1351 1352 #define BCH_KEY_MAGIC \ 1353 (((__u64) 'b' << 0)|((__u64) 'c' << 8)| \ 1354 ((__u64) 'h' << 16)|((__u64) '*' << 24)| \ 1355 ((__u64) '*' << 32)|((__u64) 'k' << 40)| \ 1356 ((__u64) 'e' << 48)|((__u64) 'y' << 56)) 1357 1358 struct bch_encrypted_key { 1359 __le64 magic; 1360 struct bch_key key; 1361 }; 1362 1363 /* 1364 * If this field is present in the superblock, it stores an encryption key which 1365 * is used encrypt all other data/metadata. The key will normally be encrypted 1366 * with the key userspace provides, but if encryption has been turned off we'll 1367 * just store the master key unencrypted in the superblock so we can access the 1368 * previously encrypted data. 1369 */ 1370 struct bch_sb_field_crypt { 1371 struct bch_sb_field field; 1372 1373 __le64 flags; 1374 __le64 kdf_flags; 1375 struct bch_encrypted_key key; 1376 }; 1377 1378 LE64_BITMASK(BCH_CRYPT_KDF_TYPE, struct bch_sb_field_crypt, flags, 0, 4); 1379 1380 enum bch_kdf_types { 1381 BCH_KDF_SCRYPT = 0, 1382 BCH_KDF_NR = 1, 1383 }; 1384 1385 /* stored as base 2 log of scrypt params: */ 1386 LE64_BITMASK(BCH_KDF_SCRYPT_N, struct bch_sb_field_crypt, kdf_flags, 0, 16); 1387 LE64_BITMASK(BCH_KDF_SCRYPT_R, struct bch_sb_field_crypt, kdf_flags, 16, 32); 1388 LE64_BITMASK(BCH_KDF_SCRYPT_P, struct bch_sb_field_crypt, kdf_flags, 32, 48); 1389 1390 /* BCH_SB_FIELD_replicas: */ 1391 1392 #define BCH_DATA_TYPES() \ 1393 x(free, 0) \ 1394 x(sb, 1) \ 1395 x(journal, 2) \ 1396 x(btree, 3) \ 1397 x(user, 4) \ 1398 x(cached, 5) \ 1399 x(parity, 6) \ 1400 x(stripe, 7) \ 1401 x(need_gc_gens, 8) \ 1402 x(need_discard, 9) 1403 1404 enum bch_data_type { 1405 #define x(t, n) BCH_DATA_##t, 1406 BCH_DATA_TYPES() 1407 #undef x 1408 BCH_DATA_NR 1409 }; 1410 1411 static inline bool data_type_is_empty(enum bch_data_type type) 1412 { 1413 switch (type) { 1414 case BCH_DATA_free: 1415 case BCH_DATA_need_gc_gens: 1416 case BCH_DATA_need_discard: 1417 return true; 1418 default: 1419 return false; 1420 } 1421 } 1422 1423 static inline bool data_type_is_hidden(enum bch_data_type type) 1424 { 1425 switch (type) { 1426 case BCH_DATA_sb: 1427 case BCH_DATA_journal: 1428 return true; 1429 default: 1430 return false; 1431 } 1432 } 1433 1434 struct bch_replicas_entry_v0 { 1435 __u8 data_type; 1436 __u8 nr_devs; 1437 __u8 devs[]; 1438 } __packed; 1439 1440 struct bch_sb_field_replicas_v0 { 1441 struct bch_sb_field field; 1442 struct bch_replicas_entry_v0 entries[]; 1443 } __packed __aligned(8); 1444 1445 struct bch_replicas_entry { 1446 __u8 data_type; 1447 __u8 nr_devs; 1448 __u8 nr_required; 1449 __u8 devs[]; 1450 } __packed; 1451 1452 #define replicas_entry_bytes(_i) \ 1453 (offsetof(typeof(*(_i)), devs) + (_i)->nr_devs) 1454 1455 struct bch_sb_field_replicas { 1456 struct bch_sb_field field; 1457 struct bch_replicas_entry entries[]; 1458 } __packed __aligned(8); 1459 1460 /* BCH_SB_FIELD_quota: */ 1461 1462 struct bch_sb_quota_counter { 1463 __le32 timelimit; 1464 __le32 warnlimit; 1465 }; 1466 1467 struct bch_sb_quota_type { 1468 __le64 flags; 1469 struct bch_sb_quota_counter c[Q_COUNTERS]; 1470 }; 1471 1472 struct bch_sb_field_quota { 1473 struct bch_sb_field field; 1474 struct bch_sb_quota_type q[QTYP_NR]; 1475 } __packed __aligned(8); 1476 1477 /* BCH_SB_FIELD_disk_groups: */ 1478 1479 #define BCH_SB_LABEL_SIZE 32 1480 1481 struct bch_disk_group { 1482 __u8 label[BCH_SB_LABEL_SIZE]; 1483 __le64 flags[2]; 1484 } __packed __aligned(8); 1485 1486 LE64_BITMASK(BCH_GROUP_DELETED, struct bch_disk_group, flags[0], 0, 1) 1487 LE64_BITMASK(BCH_GROUP_DATA_ALLOWED, struct bch_disk_group, flags[0], 1, 6) 1488 LE64_BITMASK(BCH_GROUP_PARENT, struct bch_disk_group, flags[0], 6, 24) 1489 1490 struct bch_sb_field_disk_groups { 1491 struct bch_sb_field field; 1492 struct bch_disk_group entries[]; 1493 } __packed __aligned(8); 1494 1495 /* BCH_SB_FIELD_counters */ 1496 1497 #define BCH_PERSISTENT_COUNTERS() \ 1498 x(io_read, 0) \ 1499 x(io_write, 1) \ 1500 x(io_move, 2) \ 1501 x(bucket_invalidate, 3) \ 1502 x(bucket_discard, 4) \ 1503 x(bucket_alloc, 5) \ 1504 x(bucket_alloc_fail, 6) \ 1505 x(btree_cache_scan, 7) \ 1506 x(btree_cache_reap, 8) \ 1507 x(btree_cache_cannibalize, 9) \ 1508 x(btree_cache_cannibalize_lock, 10) \ 1509 x(btree_cache_cannibalize_lock_fail, 11) \ 1510 x(btree_cache_cannibalize_unlock, 12) \ 1511 x(btree_node_write, 13) \ 1512 x(btree_node_read, 14) \ 1513 x(btree_node_compact, 15) \ 1514 x(btree_node_merge, 16) \ 1515 x(btree_node_split, 17) \ 1516 x(btree_node_rewrite, 18) \ 1517 x(btree_node_alloc, 19) \ 1518 x(btree_node_free, 20) \ 1519 x(btree_node_set_root, 21) \ 1520 x(btree_path_relock_fail, 22) \ 1521 x(btree_path_upgrade_fail, 23) \ 1522 x(btree_reserve_get_fail, 24) \ 1523 x(journal_entry_full, 25) \ 1524 x(journal_full, 26) \ 1525 x(journal_reclaim_finish, 27) \ 1526 x(journal_reclaim_start, 28) \ 1527 x(journal_write, 29) \ 1528 x(read_promote, 30) \ 1529 x(read_bounce, 31) \ 1530 x(read_split, 33) \ 1531 x(read_retry, 32) \ 1532 x(read_reuse_race, 34) \ 1533 x(move_extent_read, 35) \ 1534 x(move_extent_write, 36) \ 1535 x(move_extent_finish, 37) \ 1536 x(move_extent_fail, 38) \ 1537 x(move_extent_start_fail, 39) \ 1538 x(copygc, 40) \ 1539 x(copygc_wait, 41) \ 1540 x(gc_gens_end, 42) \ 1541 x(gc_gens_start, 43) \ 1542 x(trans_blocked_journal_reclaim, 44) \ 1543 x(trans_restart_btree_node_reused, 45) \ 1544 x(trans_restart_btree_node_split, 46) \ 1545 x(trans_restart_fault_inject, 47) \ 1546 x(trans_restart_iter_upgrade, 48) \ 1547 x(trans_restart_journal_preres_get, 49) \ 1548 x(trans_restart_journal_reclaim, 50) \ 1549 x(trans_restart_journal_res_get, 51) \ 1550 x(trans_restart_key_cache_key_realloced, 52) \ 1551 x(trans_restart_key_cache_raced, 53) \ 1552 x(trans_restart_mark_replicas, 54) \ 1553 x(trans_restart_mem_realloced, 55) \ 1554 x(trans_restart_memory_allocation_failure, 56) \ 1555 x(trans_restart_relock, 57) \ 1556 x(trans_restart_relock_after_fill, 58) \ 1557 x(trans_restart_relock_key_cache_fill, 59) \ 1558 x(trans_restart_relock_next_node, 60) \ 1559 x(trans_restart_relock_parent_for_fill, 61) \ 1560 x(trans_restart_relock_path, 62) \ 1561 x(trans_restart_relock_path_intent, 63) \ 1562 x(trans_restart_too_many_iters, 64) \ 1563 x(trans_restart_traverse, 65) \ 1564 x(trans_restart_upgrade, 66) \ 1565 x(trans_restart_would_deadlock, 67) \ 1566 x(trans_restart_would_deadlock_write, 68) \ 1567 x(trans_restart_injected, 69) \ 1568 x(trans_restart_key_cache_upgrade, 70) \ 1569 x(trans_traverse_all, 71) \ 1570 x(transaction_commit, 72) \ 1571 x(write_super, 73) \ 1572 x(trans_restart_would_deadlock_recursion_limit, 74) \ 1573 x(trans_restart_write_buffer_flush, 75) \ 1574 x(trans_restart_split_race, 76) 1575 1576 enum bch_persistent_counters { 1577 #define x(t, n, ...) BCH_COUNTER_##t, 1578 BCH_PERSISTENT_COUNTERS() 1579 #undef x 1580 BCH_COUNTER_NR 1581 }; 1582 1583 struct bch_sb_field_counters { 1584 struct bch_sb_field field; 1585 __le64 d[]; 1586 }; 1587 1588 /* 1589 * On clean shutdown, store btree roots and current journal sequence number in 1590 * the superblock: 1591 */ 1592 struct jset_entry { 1593 __le16 u64s; 1594 __u8 btree_id; 1595 __u8 level; 1596 __u8 type; /* designates what this jset holds */ 1597 __u8 pad[3]; 1598 1599 struct bkey_i start[0]; 1600 __u64 _data[]; 1601 }; 1602 1603 struct bch_sb_field_clean { 1604 struct bch_sb_field field; 1605 1606 __le32 flags; 1607 __le16 _read_clock; /* no longer used */ 1608 __le16 _write_clock; 1609 __le64 journal_seq; 1610 1611 struct jset_entry start[0]; 1612 __u64 _data[]; 1613 }; 1614 1615 struct journal_seq_blacklist_entry { 1616 __le64 start; 1617 __le64 end; 1618 }; 1619 1620 struct bch_sb_field_journal_seq_blacklist { 1621 struct bch_sb_field field; 1622 struct journal_seq_blacklist_entry start[]; 1623 }; 1624 1625 struct bch_sb_field_errors { 1626 struct bch_sb_field field; 1627 struct bch_sb_field_error_entry { 1628 __le64 v; 1629 __le64 last_error_time; 1630 } entries[]; 1631 }; 1632 1633 LE64_BITMASK(BCH_SB_ERROR_ENTRY_ID, struct bch_sb_field_error_entry, v, 0, 16); 1634 LE64_BITMASK(BCH_SB_ERROR_ENTRY_NR, struct bch_sb_field_error_entry, v, 16, 64); 1635 1636 struct bch_sb_field_ext { 1637 struct bch_sb_field field; 1638 __le64 recovery_passes_required[2]; 1639 __le64 errors_silent[8]; 1640 }; 1641 1642 struct bch_sb_field_downgrade_entry { 1643 __le16 version; 1644 __le64 recovery_passes[2]; 1645 __le16 nr_errors; 1646 __le16 errors[] __counted_by(nr_errors); 1647 } __packed __aligned(2); 1648 1649 struct bch_sb_field_downgrade { 1650 struct bch_sb_field field; 1651 struct bch_sb_field_downgrade_entry entries[]; 1652 }; 1653 1654 /* Superblock: */ 1655 1656 /* 1657 * New versioning scheme: 1658 * One common version number for all on disk data structures - superblock, btree 1659 * nodes, journal entries 1660 */ 1661 #define BCH_VERSION_MAJOR(_v) ((__u16) ((_v) >> 10)) 1662 #define BCH_VERSION_MINOR(_v) ((__u16) ((_v) & ~(~0U << 10))) 1663 #define BCH_VERSION(_major, _minor) (((_major) << 10)|(_minor) << 0) 1664 1665 #define RECOVERY_PASS_ALL_FSCK (1ULL << 63) 1666 1667 /* 1668 * field 1: version name 1669 * field 2: BCH_VERSION(major, minor) 1670 * field 3: recovery passess required on upgrade 1671 */ 1672 #define BCH_METADATA_VERSIONS() \ 1673 x(bkey_renumber, BCH_VERSION(0, 10), \ 1674 RECOVERY_PASS_ALL_FSCK) \ 1675 x(inode_btree_change, BCH_VERSION(0, 11), \ 1676 RECOVERY_PASS_ALL_FSCK) \ 1677 x(snapshot, BCH_VERSION(0, 12), \ 1678 RECOVERY_PASS_ALL_FSCK) \ 1679 x(inode_backpointers, BCH_VERSION(0, 13), \ 1680 RECOVERY_PASS_ALL_FSCK) \ 1681 x(btree_ptr_sectors_written, BCH_VERSION(0, 14), \ 1682 RECOVERY_PASS_ALL_FSCK) \ 1683 x(snapshot_2, BCH_VERSION(0, 15), \ 1684 BIT_ULL(BCH_RECOVERY_PASS_fs_upgrade_for_subvolumes)| \ 1685 BIT_ULL(BCH_RECOVERY_PASS_initialize_subvolumes)| \ 1686 RECOVERY_PASS_ALL_FSCK) \ 1687 x(reflink_p_fix, BCH_VERSION(0, 16), \ 1688 BIT_ULL(BCH_RECOVERY_PASS_fix_reflink_p)) \ 1689 x(subvol_dirent, BCH_VERSION(0, 17), \ 1690 RECOVERY_PASS_ALL_FSCK) \ 1691 x(inode_v2, BCH_VERSION(0, 18), \ 1692 RECOVERY_PASS_ALL_FSCK) \ 1693 x(freespace, BCH_VERSION(0, 19), \ 1694 RECOVERY_PASS_ALL_FSCK) \ 1695 x(alloc_v4, BCH_VERSION(0, 20), \ 1696 RECOVERY_PASS_ALL_FSCK) \ 1697 x(new_data_types, BCH_VERSION(0, 21), \ 1698 RECOVERY_PASS_ALL_FSCK) \ 1699 x(backpointers, BCH_VERSION(0, 22), \ 1700 RECOVERY_PASS_ALL_FSCK) \ 1701 x(inode_v3, BCH_VERSION(0, 23), \ 1702 RECOVERY_PASS_ALL_FSCK) \ 1703 x(unwritten_extents, BCH_VERSION(0, 24), \ 1704 RECOVERY_PASS_ALL_FSCK) \ 1705 x(bucket_gens, BCH_VERSION(0, 25), \ 1706 BIT_ULL(BCH_RECOVERY_PASS_bucket_gens_init)| \ 1707 RECOVERY_PASS_ALL_FSCK) \ 1708 x(lru_v2, BCH_VERSION(0, 26), \ 1709 RECOVERY_PASS_ALL_FSCK) \ 1710 x(fragmentation_lru, BCH_VERSION(0, 27), \ 1711 RECOVERY_PASS_ALL_FSCK) \ 1712 x(no_bps_in_alloc_keys, BCH_VERSION(0, 28), \ 1713 RECOVERY_PASS_ALL_FSCK) \ 1714 x(snapshot_trees, BCH_VERSION(0, 29), \ 1715 RECOVERY_PASS_ALL_FSCK) \ 1716 x(major_minor, BCH_VERSION(1, 0), \ 1717 0) \ 1718 x(snapshot_skiplists, BCH_VERSION(1, 1), \ 1719 BIT_ULL(BCH_RECOVERY_PASS_check_snapshots)) \ 1720 x(deleted_inodes, BCH_VERSION(1, 2), \ 1721 BIT_ULL(BCH_RECOVERY_PASS_check_inodes)) \ 1722 x(rebalance_work, BCH_VERSION(1, 3), \ 1723 BIT_ULL(BCH_RECOVERY_PASS_set_fs_needs_rebalance)) 1724 1725 enum bcachefs_metadata_version { 1726 bcachefs_metadata_version_min = 9, 1727 #define x(t, n, upgrade_passes) bcachefs_metadata_version_##t = n, 1728 BCH_METADATA_VERSIONS() 1729 #undef x 1730 bcachefs_metadata_version_max 1731 }; 1732 1733 static const __maybe_unused 1734 unsigned bcachefs_metadata_required_upgrade_below = bcachefs_metadata_version_rebalance_work; 1735 1736 #define bcachefs_metadata_version_current (bcachefs_metadata_version_max - 1) 1737 1738 #define BCH_SB_SECTOR 8 1739 #define BCH_SB_MEMBERS_MAX 64 /* XXX kill */ 1740 1741 struct bch_sb_layout { 1742 __uuid_t magic; /* bcachefs superblock UUID */ 1743 __u8 layout_type; 1744 __u8 sb_max_size_bits; /* base 2 of 512 byte sectors */ 1745 __u8 nr_superblocks; 1746 __u8 pad[5]; 1747 __le64 sb_offset[61]; 1748 } __packed __aligned(8); 1749 1750 #define BCH_SB_LAYOUT_SECTOR 7 1751 1752 /* 1753 * @offset - sector where this sb was written 1754 * @version - on disk format version 1755 * @version_min - Oldest metadata version this filesystem contains; so we can 1756 * safely drop compatibility code and refuse to mount filesystems 1757 * we'd need it for 1758 * @magic - identifies as a bcachefs superblock (BCHFS_MAGIC) 1759 * @seq - incremented each time superblock is written 1760 * @uuid - used for generating various magic numbers and identifying 1761 * member devices, never changes 1762 * @user_uuid - user visible UUID, may be changed 1763 * @label - filesystem label 1764 * @seq - identifies most recent superblock, incremented each time 1765 * superblock is written 1766 * @features - enabled incompatible features 1767 */ 1768 struct bch_sb { 1769 struct bch_csum csum; 1770 __le16 version; 1771 __le16 version_min; 1772 __le16 pad[2]; 1773 __uuid_t magic; 1774 __uuid_t uuid; 1775 __uuid_t user_uuid; 1776 __u8 label[BCH_SB_LABEL_SIZE]; 1777 __le64 offset; 1778 __le64 seq; 1779 1780 __le16 block_size; 1781 __u8 dev_idx; 1782 __u8 nr_devices; 1783 __le32 u64s; 1784 1785 __le64 time_base_lo; 1786 __le32 time_base_hi; 1787 __le32 time_precision; 1788 1789 __le64 flags[8]; 1790 __le64 features[2]; 1791 __le64 compat[2]; 1792 1793 struct bch_sb_layout layout; 1794 1795 struct bch_sb_field start[0]; 1796 __le64 _data[]; 1797 } __packed __aligned(8); 1798 1799 /* 1800 * Flags: 1801 * BCH_SB_INITALIZED - set on first mount 1802 * BCH_SB_CLEAN - did we shut down cleanly? Just a hint, doesn't affect 1803 * behaviour of mount/recovery path: 1804 * BCH_SB_INODE_32BIT - limit inode numbers to 32 bits 1805 * BCH_SB_128_BIT_MACS - 128 bit macs instead of 80 1806 * BCH_SB_ENCRYPTION_TYPE - if nonzero encryption is enabled; overrides 1807 * DATA/META_CSUM_TYPE. Also indicates encryption 1808 * algorithm in use, if/when we get more than one 1809 */ 1810 1811 LE16_BITMASK(BCH_SB_BLOCK_SIZE, struct bch_sb, block_size, 0, 16); 1812 1813 LE64_BITMASK(BCH_SB_INITIALIZED, struct bch_sb, flags[0], 0, 1); 1814 LE64_BITMASK(BCH_SB_CLEAN, struct bch_sb, flags[0], 1, 2); 1815 LE64_BITMASK(BCH_SB_CSUM_TYPE, struct bch_sb, flags[0], 2, 8); 1816 LE64_BITMASK(BCH_SB_ERROR_ACTION, struct bch_sb, flags[0], 8, 12); 1817 1818 LE64_BITMASK(BCH_SB_BTREE_NODE_SIZE, struct bch_sb, flags[0], 12, 28); 1819 1820 LE64_BITMASK(BCH_SB_GC_RESERVE, struct bch_sb, flags[0], 28, 33); 1821 LE64_BITMASK(BCH_SB_ROOT_RESERVE, struct bch_sb, flags[0], 33, 40); 1822 1823 LE64_BITMASK(BCH_SB_META_CSUM_TYPE, struct bch_sb, flags[0], 40, 44); 1824 LE64_BITMASK(BCH_SB_DATA_CSUM_TYPE, struct bch_sb, flags[0], 44, 48); 1825 1826 LE64_BITMASK(BCH_SB_META_REPLICAS_WANT, struct bch_sb, flags[0], 48, 52); 1827 LE64_BITMASK(BCH_SB_DATA_REPLICAS_WANT, struct bch_sb, flags[0], 52, 56); 1828 1829 LE64_BITMASK(BCH_SB_POSIX_ACL, struct bch_sb, flags[0], 56, 57); 1830 LE64_BITMASK(BCH_SB_USRQUOTA, struct bch_sb, flags[0], 57, 58); 1831 LE64_BITMASK(BCH_SB_GRPQUOTA, struct bch_sb, flags[0], 58, 59); 1832 LE64_BITMASK(BCH_SB_PRJQUOTA, struct bch_sb, flags[0], 59, 60); 1833 1834 LE64_BITMASK(BCH_SB_HAS_ERRORS, struct bch_sb, flags[0], 60, 61); 1835 LE64_BITMASK(BCH_SB_HAS_TOPOLOGY_ERRORS,struct bch_sb, flags[0], 61, 62); 1836 1837 LE64_BITMASK(BCH_SB_BIG_ENDIAN, struct bch_sb, flags[0], 62, 63); 1838 1839 LE64_BITMASK(BCH_SB_STR_HASH_TYPE, struct bch_sb, flags[1], 0, 4); 1840 LE64_BITMASK(BCH_SB_COMPRESSION_TYPE_LO,struct bch_sb, flags[1], 4, 8); 1841 LE64_BITMASK(BCH_SB_INODE_32BIT, struct bch_sb, flags[1], 8, 9); 1842 1843 LE64_BITMASK(BCH_SB_128_BIT_MACS, struct bch_sb, flags[1], 9, 10); 1844 LE64_BITMASK(BCH_SB_ENCRYPTION_TYPE, struct bch_sb, flags[1], 10, 14); 1845 1846 /* 1847 * Max size of an extent that may require bouncing to read or write 1848 * (checksummed, compressed): 64k 1849 */ 1850 LE64_BITMASK(BCH_SB_ENCODED_EXTENT_MAX_BITS, 1851 struct bch_sb, flags[1], 14, 20); 1852 1853 LE64_BITMASK(BCH_SB_META_REPLICAS_REQ, struct bch_sb, flags[1], 20, 24); 1854 LE64_BITMASK(BCH_SB_DATA_REPLICAS_REQ, struct bch_sb, flags[1], 24, 28); 1855 1856 LE64_BITMASK(BCH_SB_PROMOTE_TARGET, struct bch_sb, flags[1], 28, 40); 1857 LE64_BITMASK(BCH_SB_FOREGROUND_TARGET, struct bch_sb, flags[1], 40, 52); 1858 LE64_BITMASK(BCH_SB_BACKGROUND_TARGET, struct bch_sb, flags[1], 52, 64); 1859 1860 LE64_BITMASK(BCH_SB_BACKGROUND_COMPRESSION_TYPE_LO, 1861 struct bch_sb, flags[2], 0, 4); 1862 LE64_BITMASK(BCH_SB_GC_RESERVE_BYTES, struct bch_sb, flags[2], 4, 64); 1863 1864 LE64_BITMASK(BCH_SB_ERASURE_CODE, struct bch_sb, flags[3], 0, 16); 1865 LE64_BITMASK(BCH_SB_METADATA_TARGET, struct bch_sb, flags[3], 16, 28); 1866 LE64_BITMASK(BCH_SB_SHARD_INUMS, struct bch_sb, flags[3], 28, 29); 1867 LE64_BITMASK(BCH_SB_INODES_USE_KEY_CACHE,struct bch_sb, flags[3], 29, 30); 1868 LE64_BITMASK(BCH_SB_JOURNAL_FLUSH_DELAY,struct bch_sb, flags[3], 30, 62); 1869 LE64_BITMASK(BCH_SB_JOURNAL_FLUSH_DISABLED,struct bch_sb, flags[3], 62, 63); 1870 LE64_BITMASK(BCH_SB_JOURNAL_RECLAIM_DELAY,struct bch_sb, flags[4], 0, 32); 1871 LE64_BITMASK(BCH_SB_JOURNAL_TRANSACTION_NAMES,struct bch_sb, flags[4], 32, 33); 1872 LE64_BITMASK(BCH_SB_NOCOW, struct bch_sb, flags[4], 33, 34); 1873 LE64_BITMASK(BCH_SB_WRITE_BUFFER_SIZE, struct bch_sb, flags[4], 34, 54); 1874 LE64_BITMASK(BCH_SB_VERSION_UPGRADE, struct bch_sb, flags[4], 54, 56); 1875 1876 LE64_BITMASK(BCH_SB_COMPRESSION_TYPE_HI,struct bch_sb, flags[4], 56, 60); 1877 LE64_BITMASK(BCH_SB_BACKGROUND_COMPRESSION_TYPE_HI, 1878 struct bch_sb, flags[4], 60, 64); 1879 1880 LE64_BITMASK(BCH_SB_VERSION_UPGRADE_COMPLETE, 1881 struct bch_sb, flags[5], 0, 16); 1882 1883 static inline __u64 BCH_SB_COMPRESSION_TYPE(const struct bch_sb *sb) 1884 { 1885 return BCH_SB_COMPRESSION_TYPE_LO(sb) | (BCH_SB_COMPRESSION_TYPE_HI(sb) << 4); 1886 } 1887 1888 static inline void SET_BCH_SB_COMPRESSION_TYPE(struct bch_sb *sb, __u64 v) 1889 { 1890 SET_BCH_SB_COMPRESSION_TYPE_LO(sb, v); 1891 SET_BCH_SB_COMPRESSION_TYPE_HI(sb, v >> 4); 1892 } 1893 1894 static inline __u64 BCH_SB_BACKGROUND_COMPRESSION_TYPE(const struct bch_sb *sb) 1895 { 1896 return BCH_SB_BACKGROUND_COMPRESSION_TYPE_LO(sb) | 1897 (BCH_SB_BACKGROUND_COMPRESSION_TYPE_HI(sb) << 4); 1898 } 1899 1900 static inline void SET_BCH_SB_BACKGROUND_COMPRESSION_TYPE(struct bch_sb *sb, __u64 v) 1901 { 1902 SET_BCH_SB_BACKGROUND_COMPRESSION_TYPE_LO(sb, v); 1903 SET_BCH_SB_BACKGROUND_COMPRESSION_TYPE_HI(sb, v >> 4); 1904 } 1905 1906 /* 1907 * Features: 1908 * 1909 * journal_seq_blacklist_v3: gates BCH_SB_FIELD_journal_seq_blacklist 1910 * reflink: gates KEY_TYPE_reflink 1911 * inline_data: gates KEY_TYPE_inline_data 1912 * new_siphash: gates BCH_STR_HASH_siphash 1913 * new_extent_overwrite: gates BTREE_NODE_NEW_EXTENT_OVERWRITE 1914 */ 1915 #define BCH_SB_FEATURES() \ 1916 x(lz4, 0) \ 1917 x(gzip, 1) \ 1918 x(zstd, 2) \ 1919 x(atomic_nlink, 3) \ 1920 x(ec, 4) \ 1921 x(journal_seq_blacklist_v3, 5) \ 1922 x(reflink, 6) \ 1923 x(new_siphash, 7) \ 1924 x(inline_data, 8) \ 1925 x(new_extent_overwrite, 9) \ 1926 x(incompressible, 10) \ 1927 x(btree_ptr_v2, 11) \ 1928 x(extents_above_btree_updates, 12) \ 1929 x(btree_updates_journalled, 13) \ 1930 x(reflink_inline_data, 14) \ 1931 x(new_varint, 15) \ 1932 x(journal_no_flush, 16) \ 1933 x(alloc_v2, 17) \ 1934 x(extents_across_btree_nodes, 18) 1935 1936 #define BCH_SB_FEATURES_ALWAYS \ 1937 ((1ULL << BCH_FEATURE_new_extent_overwrite)| \ 1938 (1ULL << BCH_FEATURE_extents_above_btree_updates)|\ 1939 (1ULL << BCH_FEATURE_btree_updates_journalled)|\ 1940 (1ULL << BCH_FEATURE_alloc_v2)|\ 1941 (1ULL << BCH_FEATURE_extents_across_btree_nodes)) 1942 1943 #define BCH_SB_FEATURES_ALL \ 1944 (BCH_SB_FEATURES_ALWAYS| \ 1945 (1ULL << BCH_FEATURE_new_siphash)| \ 1946 (1ULL << BCH_FEATURE_btree_ptr_v2)| \ 1947 (1ULL << BCH_FEATURE_new_varint)| \ 1948 (1ULL << BCH_FEATURE_journal_no_flush)) 1949 1950 enum bch_sb_feature { 1951 #define x(f, n) BCH_FEATURE_##f, 1952 BCH_SB_FEATURES() 1953 #undef x 1954 BCH_FEATURE_NR, 1955 }; 1956 1957 #define BCH_SB_COMPAT() \ 1958 x(alloc_info, 0) \ 1959 x(alloc_metadata, 1) \ 1960 x(extents_above_btree_updates_done, 2) \ 1961 x(bformat_overflow_done, 3) 1962 1963 enum bch_sb_compat { 1964 #define x(f, n) BCH_COMPAT_##f, 1965 BCH_SB_COMPAT() 1966 #undef x 1967 BCH_COMPAT_NR, 1968 }; 1969 1970 /* options: */ 1971 1972 #define BCH_VERSION_UPGRADE_OPTS() \ 1973 x(compatible, 0) \ 1974 x(incompatible, 1) \ 1975 x(none, 2) 1976 1977 enum bch_version_upgrade_opts { 1978 #define x(t, n) BCH_VERSION_UPGRADE_##t = n, 1979 BCH_VERSION_UPGRADE_OPTS() 1980 #undef x 1981 }; 1982 1983 #define BCH_REPLICAS_MAX 4U 1984 1985 #define BCH_BKEY_PTRS_MAX 16U 1986 1987 #define BCH_ERROR_ACTIONS() \ 1988 x(continue, 0) \ 1989 x(ro, 1) \ 1990 x(panic, 2) 1991 1992 enum bch_error_actions { 1993 #define x(t, n) BCH_ON_ERROR_##t = n, 1994 BCH_ERROR_ACTIONS() 1995 #undef x 1996 BCH_ON_ERROR_NR 1997 }; 1998 1999 #define BCH_STR_HASH_TYPES() \ 2000 x(crc32c, 0) \ 2001 x(crc64, 1) \ 2002 x(siphash_old, 2) \ 2003 x(siphash, 3) 2004 2005 enum bch_str_hash_type { 2006 #define x(t, n) BCH_STR_HASH_##t = n, 2007 BCH_STR_HASH_TYPES() 2008 #undef x 2009 BCH_STR_HASH_NR 2010 }; 2011 2012 #define BCH_STR_HASH_OPTS() \ 2013 x(crc32c, 0) \ 2014 x(crc64, 1) \ 2015 x(siphash, 2) 2016 2017 enum bch_str_hash_opts { 2018 #define x(t, n) BCH_STR_HASH_OPT_##t = n, 2019 BCH_STR_HASH_OPTS() 2020 #undef x 2021 BCH_STR_HASH_OPT_NR 2022 }; 2023 2024 #define BCH_CSUM_TYPES() \ 2025 x(none, 0) \ 2026 x(crc32c_nonzero, 1) \ 2027 x(crc64_nonzero, 2) \ 2028 x(chacha20_poly1305_80, 3) \ 2029 x(chacha20_poly1305_128, 4) \ 2030 x(crc32c, 5) \ 2031 x(crc64, 6) \ 2032 x(xxhash, 7) 2033 2034 enum bch_csum_type { 2035 #define x(t, n) BCH_CSUM_##t = n, 2036 BCH_CSUM_TYPES() 2037 #undef x 2038 BCH_CSUM_NR 2039 }; 2040 2041 static const __maybe_unused unsigned bch_crc_bytes[] = { 2042 [BCH_CSUM_none] = 0, 2043 [BCH_CSUM_crc32c_nonzero] = 4, 2044 [BCH_CSUM_crc32c] = 4, 2045 [BCH_CSUM_crc64_nonzero] = 8, 2046 [BCH_CSUM_crc64] = 8, 2047 [BCH_CSUM_xxhash] = 8, 2048 [BCH_CSUM_chacha20_poly1305_80] = 10, 2049 [BCH_CSUM_chacha20_poly1305_128] = 16, 2050 }; 2051 2052 static inline _Bool bch2_csum_type_is_encryption(enum bch_csum_type type) 2053 { 2054 switch (type) { 2055 case BCH_CSUM_chacha20_poly1305_80: 2056 case BCH_CSUM_chacha20_poly1305_128: 2057 return true; 2058 default: 2059 return false; 2060 } 2061 } 2062 2063 #define BCH_CSUM_OPTS() \ 2064 x(none, 0) \ 2065 x(crc32c, 1) \ 2066 x(crc64, 2) \ 2067 x(xxhash, 3) 2068 2069 enum bch_csum_opts { 2070 #define x(t, n) BCH_CSUM_OPT_##t = n, 2071 BCH_CSUM_OPTS() 2072 #undef x 2073 BCH_CSUM_OPT_NR 2074 }; 2075 2076 #define BCH_COMPRESSION_TYPES() \ 2077 x(none, 0) \ 2078 x(lz4_old, 1) \ 2079 x(gzip, 2) \ 2080 x(lz4, 3) \ 2081 x(zstd, 4) \ 2082 x(incompressible, 5) 2083 2084 enum bch_compression_type { 2085 #define x(t, n) BCH_COMPRESSION_TYPE_##t = n, 2086 BCH_COMPRESSION_TYPES() 2087 #undef x 2088 BCH_COMPRESSION_TYPE_NR 2089 }; 2090 2091 #define BCH_COMPRESSION_OPTS() \ 2092 x(none, 0) \ 2093 x(lz4, 1) \ 2094 x(gzip, 2) \ 2095 x(zstd, 3) 2096 2097 enum bch_compression_opts { 2098 #define x(t, n) BCH_COMPRESSION_OPT_##t = n, 2099 BCH_COMPRESSION_OPTS() 2100 #undef x 2101 BCH_COMPRESSION_OPT_NR 2102 }; 2103 2104 /* 2105 * Magic numbers 2106 * 2107 * The various other data structures have their own magic numbers, which are 2108 * xored with the first part of the cache set's UUID 2109 */ 2110 2111 #define BCACHE_MAGIC \ 2112 UUID_INIT(0xc68573f6, 0x4e1a, 0x45ca, \ 2113 0x82, 0x65, 0xf5, 0x7f, 0x48, 0xba, 0x6d, 0x81) 2114 #define BCHFS_MAGIC \ 2115 UUID_INIT(0xc68573f6, 0x66ce, 0x90a9, \ 2116 0xd9, 0x6a, 0x60, 0xcf, 0x80, 0x3d, 0xf7, 0xef) 2117 2118 #define BCACHEFS_STATFS_MAGIC 0xca451a4e 2119 2120 #define JSET_MAGIC __cpu_to_le64(0x245235c1a3625032ULL) 2121 #define BSET_MAGIC __cpu_to_le64(0x90135c78b99e07f5ULL) 2122 2123 static inline __le64 __bch2_sb_magic(struct bch_sb *sb) 2124 { 2125 __le64 ret; 2126 2127 memcpy(&ret, &sb->uuid, sizeof(ret)); 2128 return ret; 2129 } 2130 2131 static inline __u64 __jset_magic(struct bch_sb *sb) 2132 { 2133 return __le64_to_cpu(__bch2_sb_magic(sb) ^ JSET_MAGIC); 2134 } 2135 2136 static inline __u64 __bset_magic(struct bch_sb *sb) 2137 { 2138 return __le64_to_cpu(__bch2_sb_magic(sb) ^ BSET_MAGIC); 2139 } 2140 2141 /* Journal */ 2142 2143 #define JSET_KEYS_U64s (sizeof(struct jset_entry) / sizeof(__u64)) 2144 2145 #define BCH_JSET_ENTRY_TYPES() \ 2146 x(btree_keys, 0) \ 2147 x(btree_root, 1) \ 2148 x(prio_ptrs, 2) \ 2149 x(blacklist, 3) \ 2150 x(blacklist_v2, 4) \ 2151 x(usage, 5) \ 2152 x(data_usage, 6) \ 2153 x(clock, 7) \ 2154 x(dev_usage, 8) \ 2155 x(log, 9) \ 2156 x(overwrite, 10) 2157 2158 enum { 2159 #define x(f, nr) BCH_JSET_ENTRY_##f = nr, 2160 BCH_JSET_ENTRY_TYPES() 2161 #undef x 2162 BCH_JSET_ENTRY_NR 2163 }; 2164 2165 /* 2166 * Journal sequence numbers can be blacklisted: bsets record the max sequence 2167 * number of all the journal entries they contain updates for, so that on 2168 * recovery we can ignore those bsets that contain index updates newer that what 2169 * made it into the journal. 2170 * 2171 * This means that we can't reuse that journal_seq - we have to skip it, and 2172 * then record that we skipped it so that the next time we crash and recover we 2173 * don't think there was a missing journal entry. 2174 */ 2175 struct jset_entry_blacklist { 2176 struct jset_entry entry; 2177 __le64 seq; 2178 }; 2179 2180 struct jset_entry_blacklist_v2 { 2181 struct jset_entry entry; 2182 __le64 start; 2183 __le64 end; 2184 }; 2185 2186 #define BCH_FS_USAGE_TYPES() \ 2187 x(reserved, 0) \ 2188 x(inodes, 1) \ 2189 x(key_version, 2) 2190 2191 enum { 2192 #define x(f, nr) BCH_FS_USAGE_##f = nr, 2193 BCH_FS_USAGE_TYPES() 2194 #undef x 2195 BCH_FS_USAGE_NR 2196 }; 2197 2198 struct jset_entry_usage { 2199 struct jset_entry entry; 2200 __le64 v; 2201 } __packed; 2202 2203 struct jset_entry_data_usage { 2204 struct jset_entry entry; 2205 __le64 v; 2206 struct bch_replicas_entry r; 2207 } __packed; 2208 2209 struct jset_entry_clock { 2210 struct jset_entry entry; 2211 __u8 rw; 2212 __u8 pad[7]; 2213 __le64 time; 2214 } __packed; 2215 2216 struct jset_entry_dev_usage_type { 2217 __le64 buckets; 2218 __le64 sectors; 2219 __le64 fragmented; 2220 } __packed; 2221 2222 struct jset_entry_dev_usage { 2223 struct jset_entry entry; 2224 __le32 dev; 2225 __u32 pad; 2226 2227 __le64 buckets_ec; 2228 __le64 _buckets_unavailable; /* No longer used */ 2229 2230 struct jset_entry_dev_usage_type d[]; 2231 }; 2232 2233 static inline unsigned jset_entry_dev_usage_nr_types(struct jset_entry_dev_usage *u) 2234 { 2235 return (vstruct_bytes(&u->entry) - sizeof(struct jset_entry_dev_usage)) / 2236 sizeof(struct jset_entry_dev_usage_type); 2237 } 2238 2239 struct jset_entry_log { 2240 struct jset_entry entry; 2241 u8 d[]; 2242 } __packed; 2243 2244 /* 2245 * On disk format for a journal entry: 2246 * seq is monotonically increasing; every journal entry has its own unique 2247 * sequence number. 2248 * 2249 * last_seq is the oldest journal entry that still has keys the btree hasn't 2250 * flushed to disk yet. 2251 * 2252 * version is for on disk format changes. 2253 */ 2254 struct jset { 2255 struct bch_csum csum; 2256 2257 __le64 magic; 2258 __le64 seq; 2259 __le32 version; 2260 __le32 flags; 2261 2262 __le32 u64s; /* size of d[] in u64s */ 2263 2264 __u8 encrypted_start[0]; 2265 2266 __le16 _read_clock; /* no longer used */ 2267 __le16 _write_clock; 2268 2269 /* Sequence number of oldest dirty journal entry */ 2270 __le64 last_seq; 2271 2272 2273 struct jset_entry start[0]; 2274 __u64 _data[]; 2275 } __packed __aligned(8); 2276 2277 LE32_BITMASK(JSET_CSUM_TYPE, struct jset, flags, 0, 4); 2278 LE32_BITMASK(JSET_BIG_ENDIAN, struct jset, flags, 4, 5); 2279 LE32_BITMASK(JSET_NO_FLUSH, struct jset, flags, 5, 6); 2280 2281 #define BCH_JOURNAL_BUCKETS_MIN 8 2282 2283 /* Btree: */ 2284 2285 enum btree_id_flags { 2286 BTREE_ID_EXTENTS = BIT(0), 2287 BTREE_ID_SNAPSHOTS = BIT(1), 2288 BTREE_ID_SNAPSHOT_FIELD = BIT(2), 2289 BTREE_ID_DATA = BIT(3), 2290 }; 2291 2292 #define BCH_BTREE_IDS() \ 2293 x(extents, 0, BTREE_ID_EXTENTS|BTREE_ID_SNAPSHOTS|BTREE_ID_DATA,\ 2294 BIT_ULL(KEY_TYPE_whiteout)| \ 2295 BIT_ULL(KEY_TYPE_error)| \ 2296 BIT_ULL(KEY_TYPE_cookie)| \ 2297 BIT_ULL(KEY_TYPE_extent)| \ 2298 BIT_ULL(KEY_TYPE_reservation)| \ 2299 BIT_ULL(KEY_TYPE_reflink_p)| \ 2300 BIT_ULL(KEY_TYPE_inline_data)) \ 2301 x(inodes, 1, BTREE_ID_SNAPSHOTS, \ 2302 BIT_ULL(KEY_TYPE_whiteout)| \ 2303 BIT_ULL(KEY_TYPE_inode)| \ 2304 BIT_ULL(KEY_TYPE_inode_v2)| \ 2305 BIT_ULL(KEY_TYPE_inode_v3)| \ 2306 BIT_ULL(KEY_TYPE_inode_generation)) \ 2307 x(dirents, 2, BTREE_ID_SNAPSHOTS, \ 2308 BIT_ULL(KEY_TYPE_whiteout)| \ 2309 BIT_ULL(KEY_TYPE_hash_whiteout)| \ 2310 BIT_ULL(KEY_TYPE_dirent)) \ 2311 x(xattrs, 3, BTREE_ID_SNAPSHOTS, \ 2312 BIT_ULL(KEY_TYPE_whiteout)| \ 2313 BIT_ULL(KEY_TYPE_cookie)| \ 2314 BIT_ULL(KEY_TYPE_hash_whiteout)| \ 2315 BIT_ULL(KEY_TYPE_xattr)) \ 2316 x(alloc, 4, 0, \ 2317 BIT_ULL(KEY_TYPE_alloc)| \ 2318 BIT_ULL(KEY_TYPE_alloc_v2)| \ 2319 BIT_ULL(KEY_TYPE_alloc_v3)| \ 2320 BIT_ULL(KEY_TYPE_alloc_v4)) \ 2321 x(quotas, 5, 0, \ 2322 BIT_ULL(KEY_TYPE_quota)) \ 2323 x(stripes, 6, 0, \ 2324 BIT_ULL(KEY_TYPE_stripe)) \ 2325 x(reflink, 7, BTREE_ID_EXTENTS|BTREE_ID_DATA, \ 2326 BIT_ULL(KEY_TYPE_reflink_v)| \ 2327 BIT_ULL(KEY_TYPE_indirect_inline_data)) \ 2328 x(subvolumes, 8, 0, \ 2329 BIT_ULL(KEY_TYPE_subvolume)) \ 2330 x(snapshots, 9, 0, \ 2331 BIT_ULL(KEY_TYPE_snapshot)) \ 2332 x(lru, 10, 0, \ 2333 BIT_ULL(KEY_TYPE_set)) \ 2334 x(freespace, 11, BTREE_ID_EXTENTS, \ 2335 BIT_ULL(KEY_TYPE_set)) \ 2336 x(need_discard, 12, 0, \ 2337 BIT_ULL(KEY_TYPE_set)) \ 2338 x(backpointers, 13, 0, \ 2339 BIT_ULL(KEY_TYPE_backpointer)) \ 2340 x(bucket_gens, 14, 0, \ 2341 BIT_ULL(KEY_TYPE_bucket_gens)) \ 2342 x(snapshot_trees, 15, 0, \ 2343 BIT_ULL(KEY_TYPE_snapshot_tree)) \ 2344 x(deleted_inodes, 16, BTREE_ID_SNAPSHOT_FIELD, \ 2345 BIT_ULL(KEY_TYPE_set)) \ 2346 x(logged_ops, 17, 0, \ 2347 BIT_ULL(KEY_TYPE_logged_op_truncate)| \ 2348 BIT_ULL(KEY_TYPE_logged_op_finsert)) \ 2349 x(rebalance_work, 18, BTREE_ID_SNAPSHOT_FIELD, \ 2350 BIT_ULL(KEY_TYPE_set)|BIT_ULL(KEY_TYPE_cookie)) 2351 2352 enum btree_id { 2353 #define x(name, nr, ...) BTREE_ID_##name = nr, 2354 BCH_BTREE_IDS() 2355 #undef x 2356 BTREE_ID_NR 2357 }; 2358 2359 #define BTREE_MAX_DEPTH 4U 2360 2361 /* Btree nodes */ 2362 2363 /* 2364 * Btree nodes 2365 * 2366 * On disk a btree node is a list/log of these; within each set the keys are 2367 * sorted 2368 */ 2369 struct bset { 2370 __le64 seq; 2371 2372 /* 2373 * Highest journal entry this bset contains keys for. 2374 * If on recovery we don't see that journal entry, this bset is ignored: 2375 * this allows us to preserve the order of all index updates after a 2376 * crash, since the journal records a total order of all index updates 2377 * and anything that didn't make it to the journal doesn't get used. 2378 */ 2379 __le64 journal_seq; 2380 2381 __le32 flags; 2382 __le16 version; 2383 __le16 u64s; /* count of d[] in u64s */ 2384 2385 struct bkey_packed start[0]; 2386 __u64 _data[]; 2387 } __packed __aligned(8); 2388 2389 LE32_BITMASK(BSET_CSUM_TYPE, struct bset, flags, 0, 4); 2390 2391 LE32_BITMASK(BSET_BIG_ENDIAN, struct bset, flags, 4, 5); 2392 LE32_BITMASK(BSET_SEPARATE_WHITEOUTS, 2393 struct bset, flags, 5, 6); 2394 2395 /* Sector offset within the btree node: */ 2396 LE32_BITMASK(BSET_OFFSET, struct bset, flags, 16, 32); 2397 2398 struct btree_node { 2399 struct bch_csum csum; 2400 __le64 magic; 2401 2402 /* this flags field is encrypted, unlike bset->flags: */ 2403 __le64 flags; 2404 2405 /* Closed interval: */ 2406 struct bpos min_key; 2407 struct bpos max_key; 2408 struct bch_extent_ptr _ptr; /* not used anymore */ 2409 struct bkey_format format; 2410 2411 union { 2412 struct bset keys; 2413 struct { 2414 __u8 pad[22]; 2415 __le16 u64s; 2416 __u64 _data[0]; 2417 2418 }; 2419 }; 2420 } __packed __aligned(8); 2421 2422 LE64_BITMASK(BTREE_NODE_ID_LO, struct btree_node, flags, 0, 4); 2423 LE64_BITMASK(BTREE_NODE_LEVEL, struct btree_node, flags, 4, 8); 2424 LE64_BITMASK(BTREE_NODE_NEW_EXTENT_OVERWRITE, 2425 struct btree_node, flags, 8, 9); 2426 LE64_BITMASK(BTREE_NODE_ID_HI, struct btree_node, flags, 9, 25); 2427 /* 25-32 unused */ 2428 LE64_BITMASK(BTREE_NODE_SEQ, struct btree_node, flags, 32, 64); 2429 2430 static inline __u64 BTREE_NODE_ID(struct btree_node *n) 2431 { 2432 return BTREE_NODE_ID_LO(n) | (BTREE_NODE_ID_HI(n) << 4); 2433 } 2434 2435 static inline void SET_BTREE_NODE_ID(struct btree_node *n, __u64 v) 2436 { 2437 SET_BTREE_NODE_ID_LO(n, v); 2438 SET_BTREE_NODE_ID_HI(n, v >> 4); 2439 } 2440 2441 struct btree_node_entry { 2442 struct bch_csum csum; 2443 2444 union { 2445 struct bset keys; 2446 struct { 2447 __u8 pad[22]; 2448 __le16 u64s; 2449 __u64 _data[0]; 2450 }; 2451 }; 2452 } __packed __aligned(8); 2453 2454 #endif /* _BCACHEFS_FORMAT_H */ 2455