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 __aligned(4); 155 156 #define KEY_INODE_MAX ((__u64)~0ULL) 157 #define KEY_OFFSET_MAX ((__u64)~0ULL) 158 #define KEY_SNAPSHOT_MAX ((__u32)~0U) 159 #define KEY_SIZE_MAX ((__u32)~0U) 160 161 static inline struct bpos SPOS(__u64 inode, __u64 offset, __u32 snapshot) 162 { 163 return (struct bpos) { 164 .inode = inode, 165 .offset = offset, 166 .snapshot = snapshot, 167 }; 168 } 169 170 #define POS_MIN SPOS(0, 0, 0) 171 #define POS_MAX SPOS(KEY_INODE_MAX, KEY_OFFSET_MAX, 0) 172 #define SPOS_MAX SPOS(KEY_INODE_MAX, KEY_OFFSET_MAX, KEY_SNAPSHOT_MAX) 173 #define POS(_inode, _offset) SPOS(_inode, _offset, 0) 174 175 /* Empty placeholder struct, for container_of() */ 176 struct bch_val { 177 __u64 __nothing[0]; 178 }; 179 180 struct bversion { 181 #if __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__ 182 __u64 lo; 183 __u32 hi; 184 #elif __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__ 185 __u32 hi; 186 __u64 lo; 187 #endif 188 } __packed __aligned(4); 189 190 struct bkey { 191 /* Size of combined key and value, in u64s */ 192 __u8 u64s; 193 194 /* Format of key (0 for format local to btree node) */ 195 #if defined(__LITTLE_ENDIAN_BITFIELD) 196 __u8 format:7, 197 needs_whiteout:1; 198 #elif defined (__BIG_ENDIAN_BITFIELD) 199 __u8 needs_whiteout:1, 200 format:7; 201 #else 202 #error edit for your odd byteorder. 203 #endif 204 205 /* Type of the value */ 206 __u8 type; 207 208 #if __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__ 209 __u8 pad[1]; 210 211 struct bversion version; 212 __u32 size; /* extent size, in sectors */ 213 struct bpos p; 214 #elif __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__ 215 struct bpos p; 216 __u32 size; /* extent size, in sectors */ 217 struct bversion version; 218 219 __u8 pad[1]; 220 #endif 221 } __packed __aligned(8); 222 223 struct bkey_packed { 224 __u64 _data[0]; 225 226 /* Size of combined key and value, in u64s */ 227 __u8 u64s; 228 229 /* Format of key (0 for format local to btree node) */ 230 231 /* 232 * XXX: next incompat on disk format change, switch format and 233 * needs_whiteout - bkey_packed() will be cheaper if format is the high 234 * bits of the bitfield 235 */ 236 #if defined(__LITTLE_ENDIAN_BITFIELD) 237 __u8 format:7, 238 needs_whiteout:1; 239 #elif defined (__BIG_ENDIAN_BITFIELD) 240 __u8 needs_whiteout:1, 241 format:7; 242 #endif 243 244 /* Type of the value */ 245 __u8 type; 246 __u8 key_start[0]; 247 248 /* 249 * We copy bkeys with struct assignment in various places, and while 250 * that shouldn't be done with packed bkeys we can't disallow it in C, 251 * and it's legal to cast a bkey to a bkey_packed - so padding it out 252 * to the same size as struct bkey should hopefully be safest. 253 */ 254 __u8 pad[sizeof(struct bkey) - 3]; 255 } __packed __aligned(8); 256 257 typedef struct { 258 __le64 lo; 259 __le64 hi; 260 } bch_le128; 261 262 #define BKEY_U64s (sizeof(struct bkey) / sizeof(__u64)) 263 #define BKEY_U64s_MAX U8_MAX 264 #define BKEY_VAL_U64s_MAX (BKEY_U64s_MAX - BKEY_U64s) 265 266 #define KEY_PACKED_BITS_START 24 267 268 #define KEY_FORMAT_LOCAL_BTREE 0 269 #define KEY_FORMAT_CURRENT 1 270 271 enum bch_bkey_fields { 272 BKEY_FIELD_INODE, 273 BKEY_FIELD_OFFSET, 274 BKEY_FIELD_SNAPSHOT, 275 BKEY_FIELD_SIZE, 276 BKEY_FIELD_VERSION_HI, 277 BKEY_FIELD_VERSION_LO, 278 BKEY_NR_FIELDS, 279 }; 280 281 #define bkey_format_field(name, field) \ 282 [BKEY_FIELD_##name] = (sizeof(((struct bkey *) NULL)->field) * 8) 283 284 #define BKEY_FORMAT_CURRENT \ 285 ((struct bkey_format) { \ 286 .key_u64s = BKEY_U64s, \ 287 .nr_fields = BKEY_NR_FIELDS, \ 288 .bits_per_field = { \ 289 bkey_format_field(INODE, p.inode), \ 290 bkey_format_field(OFFSET, p.offset), \ 291 bkey_format_field(SNAPSHOT, p.snapshot), \ 292 bkey_format_field(SIZE, size), \ 293 bkey_format_field(VERSION_HI, version.hi), \ 294 bkey_format_field(VERSION_LO, version.lo), \ 295 }, \ 296 }) 297 298 /* bkey with inline value */ 299 struct bkey_i { 300 __u64 _data[0]; 301 302 struct bkey k; 303 struct bch_val v; 304 }; 305 306 #define KEY(_inode, _offset, _size) \ 307 ((struct bkey) { \ 308 .u64s = BKEY_U64s, \ 309 .format = KEY_FORMAT_CURRENT, \ 310 .p = POS(_inode, _offset), \ 311 .size = _size, \ 312 }) 313 314 static inline void bkey_init(struct bkey *k) 315 { 316 *k = KEY(0, 0, 0); 317 } 318 319 #define bkey_bytes(_k) ((_k)->u64s * sizeof(__u64)) 320 321 #define __BKEY_PADDED(key, pad) \ 322 struct bkey_i key; __u64 key ## _pad[pad] 323 324 /* 325 * - DELETED keys are used internally to mark keys that should be ignored but 326 * override keys in composition order. Their version number is ignored. 327 * 328 * - DISCARDED keys indicate that the data is all 0s because it has been 329 * discarded. DISCARDs may have a version; if the version is nonzero the key 330 * will be persistent, otherwise the key will be dropped whenever the btree 331 * node is rewritten (like DELETED keys). 332 * 333 * - ERROR: any read of the data returns a read error, as the data was lost due 334 * to a failing device. Like DISCARDED keys, they can be removed (overridden) 335 * by new writes or cluster-wide GC. Node repair can also overwrite them with 336 * the same or a more recent version number, but not with an older version 337 * number. 338 * 339 * - WHITEOUT: for hash table btrees 340 */ 341 #define BCH_BKEY_TYPES() \ 342 x(deleted, 0) \ 343 x(whiteout, 1) \ 344 x(error, 2) \ 345 x(cookie, 3) \ 346 x(hash_whiteout, 4) \ 347 x(btree_ptr, 5) \ 348 x(extent, 6) \ 349 x(reservation, 7) \ 350 x(inode, 8) \ 351 x(inode_generation, 9) \ 352 x(dirent, 10) \ 353 x(xattr, 11) \ 354 x(alloc, 12) \ 355 x(quota, 13) \ 356 x(stripe, 14) \ 357 x(reflink_p, 15) \ 358 x(reflink_v, 16) \ 359 x(inline_data, 17) \ 360 x(btree_ptr_v2, 18) \ 361 x(indirect_inline_data, 19) \ 362 x(alloc_v2, 20) \ 363 x(subvolume, 21) \ 364 x(snapshot, 22) \ 365 x(inode_v2, 23) \ 366 x(alloc_v3, 24) \ 367 x(set, 25) \ 368 x(lru, 26) \ 369 x(alloc_v4, 27) \ 370 x(backpointer, 28) \ 371 x(inode_v3, 29) \ 372 x(bucket_gens, 30) \ 373 x(snapshot_tree, 31) \ 374 x(logged_op_truncate, 32) \ 375 x(logged_op_finsert, 33) 376 377 enum bch_bkey_type { 378 #define x(name, nr) KEY_TYPE_##name = nr, 379 BCH_BKEY_TYPES() 380 #undef x 381 KEY_TYPE_MAX, 382 }; 383 384 struct bch_deleted { 385 struct bch_val v; 386 }; 387 388 struct bch_whiteout { 389 struct bch_val v; 390 }; 391 392 struct bch_error { 393 struct bch_val v; 394 }; 395 396 struct bch_cookie { 397 struct bch_val v; 398 __le64 cookie; 399 }; 400 401 struct bch_hash_whiteout { 402 struct bch_val v; 403 }; 404 405 struct bch_set { 406 struct bch_val v; 407 }; 408 409 /* Extents */ 410 411 /* 412 * In extent bkeys, the value is a list of pointers (bch_extent_ptr), optionally 413 * preceded by checksum/compression information (bch_extent_crc32 or 414 * bch_extent_crc64). 415 * 416 * One major determining factor in the format of extents is how we handle and 417 * represent extents that have been partially overwritten and thus trimmed: 418 * 419 * If an extent is not checksummed or compressed, when the extent is trimmed we 420 * don't have to remember the extent we originally allocated and wrote: we can 421 * merely adjust ptr->offset to point to the start of the data that is currently 422 * live. The size field in struct bkey records the current (live) size of the 423 * extent, and is also used to mean "size of region on disk that we point to" in 424 * this case. 425 * 426 * Thus an extent that is not checksummed or compressed will consist only of a 427 * list of bch_extent_ptrs, with none of the fields in 428 * bch_extent_crc32/bch_extent_crc64. 429 * 430 * When an extent is checksummed or compressed, it's not possible to read only 431 * the data that is currently live: we have to read the entire extent that was 432 * originally written, and then return only the part of the extent that is 433 * currently live. 434 * 435 * Thus, in addition to the current size of the extent in struct bkey, we need 436 * to store the size of the originally allocated space - this is the 437 * compressed_size and uncompressed_size fields in bch_extent_crc32/64. Also, 438 * when the extent is trimmed, instead of modifying the offset field of the 439 * pointer, we keep a second smaller offset field - "offset into the original 440 * extent of the currently live region". 441 * 442 * The other major determining factor is replication and data migration: 443 * 444 * Each pointer may have its own bch_extent_crc32/64. When doing a replicated 445 * write, we will initially write all the replicas in the same format, with the 446 * same checksum type and compression format - however, when copygc runs later (or 447 * tiering/cache promotion, anything that moves data), it is not in general 448 * going to rewrite all the pointers at once - one of the replicas may be in a 449 * bucket on one device that has very little fragmentation while another lives 450 * in a bucket that has become heavily fragmented, and thus is being rewritten 451 * sooner than the rest. 452 * 453 * Thus it will only move a subset of the pointers (or in the case of 454 * tiering/cache promotion perhaps add a single pointer without dropping any 455 * current pointers), and if the extent has been partially overwritten it must 456 * write only the currently live portion (or copygc would not be able to reduce 457 * fragmentation!) - which necessitates a different bch_extent_crc format for 458 * the new pointer. 459 * 460 * But in the interests of space efficiency, we don't want to store one 461 * bch_extent_crc for each pointer if we don't have to. 462 * 463 * Thus, a bch_extent consists of bch_extent_crc32s, bch_extent_crc64s, and 464 * bch_extent_ptrs appended arbitrarily one after the other. We determine the 465 * type of a given entry with a scheme similar to utf8 (except we're encoding a 466 * type, not a size), encoding the type in the position of the first set bit: 467 * 468 * bch_extent_crc32 - 0b1 469 * bch_extent_ptr - 0b10 470 * bch_extent_crc64 - 0b100 471 * 472 * We do it this way because bch_extent_crc32 is _very_ constrained on bits (and 473 * bch_extent_crc64 is the least constrained). 474 * 475 * Then, each bch_extent_crc32/64 applies to the pointers that follow after it, 476 * until the next bch_extent_crc32/64. 477 * 478 * If there are no bch_extent_crcs preceding a bch_extent_ptr, then that pointer 479 * is neither checksummed nor compressed. 480 */ 481 482 /* 128 bits, sufficient for cryptographic MACs: */ 483 struct bch_csum { 484 __le64 lo; 485 __le64 hi; 486 } __packed __aligned(8); 487 488 #define BCH_EXTENT_ENTRY_TYPES() \ 489 x(ptr, 0) \ 490 x(crc32, 1) \ 491 x(crc64, 2) \ 492 x(crc128, 3) \ 493 x(stripe_ptr, 4) \ 494 x(rebalance, 5) 495 #define BCH_EXTENT_ENTRY_MAX 6 496 497 enum bch_extent_entry_type { 498 #define x(f, n) BCH_EXTENT_ENTRY_##f = n, 499 BCH_EXTENT_ENTRY_TYPES() 500 #undef x 501 }; 502 503 /* Compressed/uncompressed size are stored biased by 1: */ 504 struct bch_extent_crc32 { 505 #if defined(__LITTLE_ENDIAN_BITFIELD) 506 __u32 type:2, 507 _compressed_size:7, 508 _uncompressed_size:7, 509 offset:7, 510 _unused:1, 511 csum_type:4, 512 compression_type:4; 513 __u32 csum; 514 #elif defined (__BIG_ENDIAN_BITFIELD) 515 __u32 csum; 516 __u32 compression_type:4, 517 csum_type:4, 518 _unused:1, 519 offset:7, 520 _uncompressed_size:7, 521 _compressed_size:7, 522 type:2; 523 #endif 524 } __packed __aligned(8); 525 526 #define CRC32_SIZE_MAX (1U << 7) 527 #define CRC32_NONCE_MAX 0 528 529 struct bch_extent_crc64 { 530 #if defined(__LITTLE_ENDIAN_BITFIELD) 531 __u64 type:3, 532 _compressed_size:9, 533 _uncompressed_size:9, 534 offset:9, 535 nonce:10, 536 csum_type:4, 537 compression_type:4, 538 csum_hi:16; 539 #elif defined (__BIG_ENDIAN_BITFIELD) 540 __u64 csum_hi:16, 541 compression_type:4, 542 csum_type:4, 543 nonce:10, 544 offset:9, 545 _uncompressed_size:9, 546 _compressed_size:9, 547 type:3; 548 #endif 549 __u64 csum_lo; 550 } __packed __aligned(8); 551 552 #define CRC64_SIZE_MAX (1U << 9) 553 #define CRC64_NONCE_MAX ((1U << 10) - 1) 554 555 struct bch_extent_crc128 { 556 #if defined(__LITTLE_ENDIAN_BITFIELD) 557 __u64 type:4, 558 _compressed_size:13, 559 _uncompressed_size:13, 560 offset:13, 561 nonce:13, 562 csum_type:4, 563 compression_type:4; 564 #elif defined (__BIG_ENDIAN_BITFIELD) 565 __u64 compression_type:4, 566 csum_type:4, 567 nonce:13, 568 offset:13, 569 _uncompressed_size:13, 570 _compressed_size:13, 571 type:4; 572 #endif 573 struct bch_csum csum; 574 } __packed __aligned(8); 575 576 #define CRC128_SIZE_MAX (1U << 13) 577 #define CRC128_NONCE_MAX ((1U << 13) - 1) 578 579 /* 580 * @reservation - pointer hasn't been written to, just reserved 581 */ 582 struct bch_extent_ptr { 583 #if defined(__LITTLE_ENDIAN_BITFIELD) 584 __u64 type:1, 585 cached:1, 586 unused:1, 587 unwritten:1, 588 offset:44, /* 8 petabytes */ 589 dev:8, 590 gen:8; 591 #elif defined (__BIG_ENDIAN_BITFIELD) 592 __u64 gen:8, 593 dev:8, 594 offset:44, 595 unwritten:1, 596 unused:1, 597 cached:1, 598 type:1; 599 #endif 600 } __packed __aligned(8); 601 602 struct bch_extent_stripe_ptr { 603 #if defined(__LITTLE_ENDIAN_BITFIELD) 604 __u64 type:5, 605 block:8, 606 redundancy:4, 607 idx:47; 608 #elif defined (__BIG_ENDIAN_BITFIELD) 609 __u64 idx:47, 610 redundancy:4, 611 block:8, 612 type:5; 613 #endif 614 }; 615 616 struct bch_extent_rebalance { 617 #if defined(__LITTLE_ENDIAN_BITFIELD) 618 __u64 type:6, 619 unused:34, 620 compression:8, /* enum bch_compression_opt */ 621 target:16; 622 #elif defined (__BIG_ENDIAN_BITFIELD) 623 __u64 target:16, 624 compression:8, 625 unused:34, 626 type:6; 627 #endif 628 }; 629 630 union bch_extent_entry { 631 #if __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__ || __BITS_PER_LONG == 64 632 unsigned long type; 633 #elif __BITS_PER_LONG == 32 634 struct { 635 unsigned long pad; 636 unsigned long type; 637 }; 638 #else 639 #error edit for your odd byteorder. 640 #endif 641 642 #define x(f, n) struct bch_extent_##f f; 643 BCH_EXTENT_ENTRY_TYPES() 644 #undef x 645 }; 646 647 struct bch_btree_ptr { 648 struct bch_val v; 649 650 __u64 _data[0]; 651 struct bch_extent_ptr start[]; 652 } __packed __aligned(8); 653 654 struct bch_btree_ptr_v2 { 655 struct bch_val v; 656 657 __u64 mem_ptr; 658 __le64 seq; 659 __le16 sectors_written; 660 __le16 flags; 661 struct bpos min_key; 662 __u64 _data[0]; 663 struct bch_extent_ptr start[]; 664 } __packed __aligned(8); 665 666 LE16_BITMASK(BTREE_PTR_RANGE_UPDATED, struct bch_btree_ptr_v2, flags, 0, 1); 667 668 struct bch_extent { 669 struct bch_val v; 670 671 __u64 _data[0]; 672 union bch_extent_entry start[]; 673 } __packed __aligned(8); 674 675 struct bch_reservation { 676 struct bch_val v; 677 678 __le32 generation; 679 __u8 nr_replicas; 680 __u8 pad[3]; 681 } __packed __aligned(8); 682 683 /* Maximum size (in u64s) a single pointer could be: */ 684 #define BKEY_EXTENT_PTR_U64s_MAX\ 685 ((sizeof(struct bch_extent_crc128) + \ 686 sizeof(struct bch_extent_ptr)) / sizeof(__u64)) 687 688 /* Maximum possible size of an entire extent value: */ 689 #define BKEY_EXTENT_VAL_U64s_MAX \ 690 (1 + BKEY_EXTENT_PTR_U64s_MAX * (BCH_REPLICAS_MAX + 1)) 691 692 /* * Maximum possible size of an entire extent, key + value: */ 693 #define BKEY_EXTENT_U64s_MAX (BKEY_U64s + BKEY_EXTENT_VAL_U64s_MAX) 694 695 /* Btree pointers don't carry around checksums: */ 696 #define BKEY_BTREE_PTR_VAL_U64s_MAX \ 697 ((sizeof(struct bch_btree_ptr_v2) + \ 698 sizeof(struct bch_extent_ptr) * BCH_REPLICAS_MAX) / sizeof(__u64)) 699 #define BKEY_BTREE_PTR_U64s_MAX \ 700 (BKEY_U64s + BKEY_BTREE_PTR_VAL_U64s_MAX) 701 702 /* Inodes */ 703 704 #define BLOCKDEV_INODE_MAX 4096 705 706 #define BCACHEFS_ROOT_INO 4096 707 708 struct bch_inode { 709 struct bch_val v; 710 711 __le64 bi_hash_seed; 712 __le32 bi_flags; 713 __le16 bi_mode; 714 __u8 fields[]; 715 } __packed __aligned(8); 716 717 struct bch_inode_v2 { 718 struct bch_val v; 719 720 __le64 bi_journal_seq; 721 __le64 bi_hash_seed; 722 __le64 bi_flags; 723 __le16 bi_mode; 724 __u8 fields[]; 725 } __packed __aligned(8); 726 727 struct bch_inode_v3 { 728 struct bch_val v; 729 730 __le64 bi_journal_seq; 731 __le64 bi_hash_seed; 732 __le64 bi_flags; 733 __le64 bi_sectors; 734 __le64 bi_size; 735 __le64 bi_version; 736 __u8 fields[]; 737 } __packed __aligned(8); 738 739 #define INODEv3_FIELDS_START_INITIAL 6 740 #define INODEv3_FIELDS_START_CUR (offsetof(struct bch_inode_v3, fields) / sizeof(__u64)) 741 742 struct bch_inode_generation { 743 struct bch_val v; 744 745 __le32 bi_generation; 746 __le32 pad; 747 } __packed __aligned(8); 748 749 /* 750 * bi_subvol and bi_parent_subvol are only set for subvolume roots: 751 */ 752 753 #define BCH_INODE_FIELDS_v2() \ 754 x(bi_atime, 96) \ 755 x(bi_ctime, 96) \ 756 x(bi_mtime, 96) \ 757 x(bi_otime, 96) \ 758 x(bi_size, 64) \ 759 x(bi_sectors, 64) \ 760 x(bi_uid, 32) \ 761 x(bi_gid, 32) \ 762 x(bi_nlink, 32) \ 763 x(bi_generation, 32) \ 764 x(bi_dev, 32) \ 765 x(bi_data_checksum, 8) \ 766 x(bi_compression, 8) \ 767 x(bi_project, 32) \ 768 x(bi_background_compression, 8) \ 769 x(bi_data_replicas, 8) \ 770 x(bi_promote_target, 16) \ 771 x(bi_foreground_target, 16) \ 772 x(bi_background_target, 16) \ 773 x(bi_erasure_code, 16) \ 774 x(bi_fields_set, 16) \ 775 x(bi_dir, 64) \ 776 x(bi_dir_offset, 64) \ 777 x(bi_subvol, 32) \ 778 x(bi_parent_subvol, 32) 779 780 #define BCH_INODE_FIELDS_v3() \ 781 x(bi_atime, 96) \ 782 x(bi_ctime, 96) \ 783 x(bi_mtime, 96) \ 784 x(bi_otime, 96) \ 785 x(bi_uid, 32) \ 786 x(bi_gid, 32) \ 787 x(bi_nlink, 32) \ 788 x(bi_generation, 32) \ 789 x(bi_dev, 32) \ 790 x(bi_data_checksum, 8) \ 791 x(bi_compression, 8) \ 792 x(bi_project, 32) \ 793 x(bi_background_compression, 8) \ 794 x(bi_data_replicas, 8) \ 795 x(bi_promote_target, 16) \ 796 x(bi_foreground_target, 16) \ 797 x(bi_background_target, 16) \ 798 x(bi_erasure_code, 16) \ 799 x(bi_fields_set, 16) \ 800 x(bi_dir, 64) \ 801 x(bi_dir_offset, 64) \ 802 x(bi_subvol, 32) \ 803 x(bi_parent_subvol, 32) \ 804 x(bi_nocow, 8) 805 806 /* subset of BCH_INODE_FIELDS */ 807 #define BCH_INODE_OPTS() \ 808 x(data_checksum, 8) \ 809 x(compression, 8) \ 810 x(project, 32) \ 811 x(background_compression, 8) \ 812 x(data_replicas, 8) \ 813 x(promote_target, 16) \ 814 x(foreground_target, 16) \ 815 x(background_target, 16) \ 816 x(erasure_code, 16) \ 817 x(nocow, 8) 818 819 enum inode_opt_id { 820 #define x(name, ...) \ 821 Inode_opt_##name, 822 BCH_INODE_OPTS() 823 #undef x 824 Inode_opt_nr, 825 }; 826 827 #define BCH_INODE_FLAGS() \ 828 x(sync, 0) \ 829 x(immutable, 1) \ 830 x(append, 2) \ 831 x(nodump, 3) \ 832 x(noatime, 4) \ 833 x(i_size_dirty, 5) \ 834 x(i_sectors_dirty, 6) \ 835 x(unlinked, 7) \ 836 x(backptr_untrusted, 8) 837 838 /* bits 20+ reserved for packed fields below: */ 839 840 enum bch_inode_flags { 841 #define x(t, n) BCH_INODE_##t = 1U << n, 842 BCH_INODE_FLAGS() 843 #undef x 844 }; 845 846 enum __bch_inode_flags { 847 #define x(t, n) __BCH_INODE_##t = n, 848 BCH_INODE_FLAGS() 849 #undef x 850 }; 851 852 LE32_BITMASK(INODE_STR_HASH, struct bch_inode, bi_flags, 20, 24); 853 LE32_BITMASK(INODE_NR_FIELDS, struct bch_inode, bi_flags, 24, 31); 854 LE32_BITMASK(INODE_NEW_VARINT, struct bch_inode, bi_flags, 31, 32); 855 856 LE64_BITMASK(INODEv2_STR_HASH, struct bch_inode_v2, bi_flags, 20, 24); 857 LE64_BITMASK(INODEv2_NR_FIELDS, struct bch_inode_v2, bi_flags, 24, 31); 858 859 LE64_BITMASK(INODEv3_STR_HASH, struct bch_inode_v3, bi_flags, 20, 24); 860 LE64_BITMASK(INODEv3_NR_FIELDS, struct bch_inode_v3, bi_flags, 24, 31); 861 862 LE64_BITMASK(INODEv3_FIELDS_START, 863 struct bch_inode_v3, bi_flags, 31, 36); 864 LE64_BITMASK(INODEv3_MODE, struct bch_inode_v3, bi_flags, 36, 52); 865 866 /* Dirents */ 867 868 /* 869 * Dirents (and xattrs) have to implement string lookups; since our b-tree 870 * doesn't support arbitrary length strings for the key, we instead index by a 871 * 64 bit hash (currently truncated sha1) of the string, stored in the offset 872 * field of the key - using linear probing to resolve hash collisions. This also 873 * provides us with the readdir cookie posix requires. 874 * 875 * Linear probing requires us to use whiteouts for deletions, in the event of a 876 * collision: 877 */ 878 879 struct bch_dirent { 880 struct bch_val v; 881 882 /* Target inode number: */ 883 union { 884 __le64 d_inum; 885 struct { /* DT_SUBVOL */ 886 __le32 d_child_subvol; 887 __le32 d_parent_subvol; 888 }; 889 }; 890 891 /* 892 * Copy of mode bits 12-15 from the target inode - so userspace can get 893 * the filetype without having to do a stat() 894 */ 895 __u8 d_type; 896 897 __u8 d_name[]; 898 } __packed __aligned(8); 899 900 #define DT_SUBVOL 16 901 #define BCH_DT_MAX 17 902 903 #define BCH_NAME_MAX 512 904 905 /* Xattrs */ 906 907 #define KEY_TYPE_XATTR_INDEX_USER 0 908 #define KEY_TYPE_XATTR_INDEX_POSIX_ACL_ACCESS 1 909 #define KEY_TYPE_XATTR_INDEX_POSIX_ACL_DEFAULT 2 910 #define KEY_TYPE_XATTR_INDEX_TRUSTED 3 911 #define KEY_TYPE_XATTR_INDEX_SECURITY 4 912 913 struct bch_xattr { 914 struct bch_val v; 915 __u8 x_type; 916 __u8 x_name_len; 917 __le16 x_val_len; 918 __u8 x_name[]; 919 } __packed __aligned(8); 920 921 /* Bucket/allocation information: */ 922 923 struct bch_alloc { 924 struct bch_val v; 925 __u8 fields; 926 __u8 gen; 927 __u8 data[]; 928 } __packed __aligned(8); 929 930 #define BCH_ALLOC_FIELDS_V1() \ 931 x(read_time, 16) \ 932 x(write_time, 16) \ 933 x(data_type, 8) \ 934 x(dirty_sectors, 16) \ 935 x(cached_sectors, 16) \ 936 x(oldest_gen, 8) \ 937 x(stripe, 32) \ 938 x(stripe_redundancy, 8) 939 940 enum { 941 #define x(name, _bits) BCH_ALLOC_FIELD_V1_##name, 942 BCH_ALLOC_FIELDS_V1() 943 #undef x 944 }; 945 946 struct bch_alloc_v2 { 947 struct bch_val v; 948 __u8 nr_fields; 949 __u8 gen; 950 __u8 oldest_gen; 951 __u8 data_type; 952 __u8 data[]; 953 } __packed __aligned(8); 954 955 #define BCH_ALLOC_FIELDS_V2() \ 956 x(read_time, 64) \ 957 x(write_time, 64) \ 958 x(dirty_sectors, 32) \ 959 x(cached_sectors, 32) \ 960 x(stripe, 32) \ 961 x(stripe_redundancy, 8) 962 963 struct bch_alloc_v3 { 964 struct bch_val v; 965 __le64 journal_seq; 966 __le32 flags; 967 __u8 nr_fields; 968 __u8 gen; 969 __u8 oldest_gen; 970 __u8 data_type; 971 __u8 data[]; 972 } __packed __aligned(8); 973 974 LE32_BITMASK(BCH_ALLOC_V3_NEED_DISCARD,struct bch_alloc_v3, flags, 0, 1) 975 LE32_BITMASK(BCH_ALLOC_V3_NEED_INC_GEN,struct bch_alloc_v3, flags, 1, 2) 976 977 struct bch_alloc_v4 { 978 struct bch_val v; 979 __u64 journal_seq; 980 __u32 flags; 981 __u8 gen; 982 __u8 oldest_gen; 983 __u8 data_type; 984 __u8 stripe_redundancy; 985 __u32 dirty_sectors; 986 __u32 cached_sectors; 987 __u64 io_time[2]; 988 __u32 stripe; 989 __u32 nr_external_backpointers; 990 __u64 fragmentation_lru; 991 } __packed __aligned(8); 992 993 #define BCH_ALLOC_V4_U64s_V0 6 994 #define BCH_ALLOC_V4_U64s (sizeof(struct bch_alloc_v4) / sizeof(__u64)) 995 996 BITMASK(BCH_ALLOC_V4_NEED_DISCARD, struct bch_alloc_v4, flags, 0, 1) 997 BITMASK(BCH_ALLOC_V4_NEED_INC_GEN, struct bch_alloc_v4, flags, 1, 2) 998 BITMASK(BCH_ALLOC_V4_BACKPOINTERS_START,struct bch_alloc_v4, flags, 2, 8) 999 BITMASK(BCH_ALLOC_V4_NR_BACKPOINTERS, struct bch_alloc_v4, flags, 8, 14) 1000 1001 #define BCH_ALLOC_V4_NR_BACKPOINTERS_MAX 40 1002 1003 struct bch_backpointer { 1004 struct bch_val v; 1005 __u8 btree_id; 1006 __u8 level; 1007 __u8 data_type; 1008 __u64 bucket_offset:40; 1009 __u32 bucket_len; 1010 struct bpos pos; 1011 } __packed __aligned(8); 1012 1013 #define KEY_TYPE_BUCKET_GENS_BITS 8 1014 #define KEY_TYPE_BUCKET_GENS_NR (1U << KEY_TYPE_BUCKET_GENS_BITS) 1015 #define KEY_TYPE_BUCKET_GENS_MASK (KEY_TYPE_BUCKET_GENS_NR - 1) 1016 1017 struct bch_bucket_gens { 1018 struct bch_val v; 1019 u8 gens[KEY_TYPE_BUCKET_GENS_NR]; 1020 } __packed __aligned(8); 1021 1022 /* Quotas: */ 1023 1024 enum quota_types { 1025 QTYP_USR = 0, 1026 QTYP_GRP = 1, 1027 QTYP_PRJ = 2, 1028 QTYP_NR = 3, 1029 }; 1030 1031 enum quota_counters { 1032 Q_SPC = 0, 1033 Q_INO = 1, 1034 Q_COUNTERS = 2, 1035 }; 1036 1037 struct bch_quota_counter { 1038 __le64 hardlimit; 1039 __le64 softlimit; 1040 }; 1041 1042 struct bch_quota { 1043 struct bch_val v; 1044 struct bch_quota_counter c[Q_COUNTERS]; 1045 } __packed __aligned(8); 1046 1047 /* Erasure coding */ 1048 1049 struct bch_stripe { 1050 struct bch_val v; 1051 __le16 sectors; 1052 __u8 algorithm; 1053 __u8 nr_blocks; 1054 __u8 nr_redundant; 1055 1056 __u8 csum_granularity_bits; 1057 __u8 csum_type; 1058 __u8 pad; 1059 1060 struct bch_extent_ptr ptrs[]; 1061 } __packed __aligned(8); 1062 1063 /* Reflink: */ 1064 1065 struct bch_reflink_p { 1066 struct bch_val v; 1067 __le64 idx; 1068 /* 1069 * A reflink pointer might point to an indirect extent which is then 1070 * later split (by copygc or rebalance). If we only pointed to part of 1071 * the original indirect extent, and then one of the fragments is 1072 * outside the range we point to, we'd leak a refcount: so when creating 1073 * reflink pointers, we need to store pad values to remember the full 1074 * range we were taking a reference on. 1075 */ 1076 __le32 front_pad; 1077 __le32 back_pad; 1078 } __packed __aligned(8); 1079 1080 struct bch_reflink_v { 1081 struct bch_val v; 1082 __le64 refcount; 1083 union bch_extent_entry start[0]; 1084 __u64 _data[]; 1085 } __packed __aligned(8); 1086 1087 struct bch_indirect_inline_data { 1088 struct bch_val v; 1089 __le64 refcount; 1090 u8 data[]; 1091 }; 1092 1093 /* Inline data */ 1094 1095 struct bch_inline_data { 1096 struct bch_val v; 1097 u8 data[]; 1098 }; 1099 1100 /* Subvolumes: */ 1101 1102 #define SUBVOL_POS_MIN POS(0, 1) 1103 #define SUBVOL_POS_MAX POS(0, S32_MAX) 1104 #define BCACHEFS_ROOT_SUBVOL 1 1105 1106 struct bch_subvolume { 1107 struct bch_val v; 1108 __le32 flags; 1109 __le32 snapshot; 1110 __le64 inode; 1111 /* 1112 * Snapshot subvolumes form a tree, separate from the snapshot nodes 1113 * tree - if this subvolume is a snapshot, this is the ID of the 1114 * subvolume it was created from: 1115 */ 1116 __le32 parent; 1117 __le32 pad; 1118 bch_le128 otime; 1119 }; 1120 1121 LE32_BITMASK(BCH_SUBVOLUME_RO, struct bch_subvolume, flags, 0, 1) 1122 /* 1123 * We need to know whether a subvolume is a snapshot so we can know whether we 1124 * can delete it (or whether it should just be rm -rf'd) 1125 */ 1126 LE32_BITMASK(BCH_SUBVOLUME_SNAP, struct bch_subvolume, flags, 1, 2) 1127 LE32_BITMASK(BCH_SUBVOLUME_UNLINKED, struct bch_subvolume, flags, 2, 3) 1128 1129 /* Snapshots */ 1130 1131 struct bch_snapshot { 1132 struct bch_val v; 1133 __le32 flags; 1134 __le32 parent; 1135 __le32 children[2]; 1136 __le32 subvol; 1137 /* corresponds to a bch_snapshot_tree in BTREE_ID_snapshot_trees */ 1138 __le32 tree; 1139 __le32 depth; 1140 __le32 skip[3]; 1141 }; 1142 1143 LE32_BITMASK(BCH_SNAPSHOT_DELETED, struct bch_snapshot, flags, 0, 1) 1144 1145 /* True if a subvolume points to this snapshot node: */ 1146 LE32_BITMASK(BCH_SNAPSHOT_SUBVOL, struct bch_snapshot, flags, 1, 2) 1147 1148 /* 1149 * Snapshot trees: 1150 * 1151 * The snapshot_trees btree gives us persistent indentifier for each tree of 1152 * bch_snapshot nodes, and allow us to record and easily find the root/master 1153 * subvolume that other snapshots were created from: 1154 */ 1155 struct bch_snapshot_tree { 1156 struct bch_val v; 1157 __le32 master_subvol; 1158 __le32 root_snapshot; 1159 }; 1160 1161 /* LRU btree: */ 1162 1163 struct bch_lru { 1164 struct bch_val v; 1165 __le64 idx; 1166 } __packed __aligned(8); 1167 1168 #define LRU_ID_STRIPES (1U << 16) 1169 1170 /* Logged operations btree: */ 1171 1172 struct bch_logged_op_truncate { 1173 struct bch_val v; 1174 __le32 subvol; 1175 __le32 pad; 1176 __le64 inum; 1177 __le64 new_i_size; 1178 }; 1179 1180 enum logged_op_finsert_state { 1181 LOGGED_OP_FINSERT_start, 1182 LOGGED_OP_FINSERT_shift_extents, 1183 LOGGED_OP_FINSERT_finish, 1184 }; 1185 1186 struct bch_logged_op_finsert { 1187 struct bch_val v; 1188 __u8 state; 1189 __u8 pad[3]; 1190 __le32 subvol; 1191 __le64 inum; 1192 __le64 dst_offset; 1193 __le64 src_offset; 1194 __le64 pos; 1195 }; 1196 1197 /* Optional/variable size superblock sections: */ 1198 1199 struct bch_sb_field { 1200 __u64 _data[0]; 1201 __le32 u64s; 1202 __le32 type; 1203 }; 1204 1205 #define BCH_SB_FIELDS() \ 1206 x(journal, 0) \ 1207 x(members_v1, 1) \ 1208 x(crypt, 2) \ 1209 x(replicas_v0, 3) \ 1210 x(quota, 4) \ 1211 x(disk_groups, 5) \ 1212 x(clean, 6) \ 1213 x(replicas, 7) \ 1214 x(journal_seq_blacklist, 8) \ 1215 x(journal_v2, 9) \ 1216 x(counters, 10) \ 1217 x(members_v2, 11) \ 1218 x(errors, 12) 1219 1220 enum bch_sb_field_type { 1221 #define x(f, nr) BCH_SB_FIELD_##f = nr, 1222 BCH_SB_FIELDS() 1223 #undef x 1224 BCH_SB_FIELD_NR 1225 }; 1226 1227 /* 1228 * Most superblock fields are replicated in all device's superblocks - a few are 1229 * not: 1230 */ 1231 #define BCH_SINGLE_DEVICE_SB_FIELDS \ 1232 ((1U << BCH_SB_FIELD_journal)| \ 1233 (1U << BCH_SB_FIELD_journal_v2)) 1234 1235 /* BCH_SB_FIELD_journal: */ 1236 1237 struct bch_sb_field_journal { 1238 struct bch_sb_field field; 1239 __le64 buckets[]; 1240 }; 1241 1242 struct bch_sb_field_journal_v2 { 1243 struct bch_sb_field field; 1244 1245 struct bch_sb_field_journal_v2_entry { 1246 __le64 start; 1247 __le64 nr; 1248 } d[]; 1249 }; 1250 1251 /* BCH_SB_FIELD_members_v1: */ 1252 1253 #define BCH_MIN_NR_NBUCKETS (1 << 6) 1254 1255 #define BCH_IOPS_MEASUREMENTS() \ 1256 x(seqread, 0) \ 1257 x(seqwrite, 1) \ 1258 x(randread, 2) \ 1259 x(randwrite, 3) 1260 1261 enum bch_iops_measurement { 1262 #define x(t, n) BCH_IOPS_##t = n, 1263 BCH_IOPS_MEASUREMENTS() 1264 #undef x 1265 BCH_IOPS_NR 1266 }; 1267 1268 #define BCH_MEMBER_ERROR_TYPES() \ 1269 x(read, 0) \ 1270 x(write, 1) \ 1271 x(checksum, 2) 1272 1273 enum bch_member_error_type { 1274 #define x(t, n) BCH_MEMBER_ERROR_##t = n, 1275 BCH_MEMBER_ERROR_TYPES() 1276 #undef x 1277 BCH_MEMBER_ERROR_NR 1278 }; 1279 1280 struct bch_member { 1281 __uuid_t uuid; 1282 __le64 nbuckets; /* device size */ 1283 __le16 first_bucket; /* index of first bucket used */ 1284 __le16 bucket_size; /* sectors */ 1285 __le32 pad; 1286 __le64 last_mount; /* time_t */ 1287 1288 __le64 flags; 1289 __le32 iops[4]; 1290 __le64 errors[BCH_MEMBER_ERROR_NR]; 1291 __le64 errors_at_reset[BCH_MEMBER_ERROR_NR]; 1292 __le64 errors_reset_time; 1293 }; 1294 1295 #define BCH_MEMBER_V1_BYTES 56 1296 1297 LE64_BITMASK(BCH_MEMBER_STATE, struct bch_member, flags, 0, 4) 1298 /* 4-14 unused, was TIER, HAS_(META)DATA, REPLACEMENT */ 1299 LE64_BITMASK(BCH_MEMBER_DISCARD, struct bch_member, flags, 14, 15) 1300 LE64_BITMASK(BCH_MEMBER_DATA_ALLOWED, struct bch_member, flags, 15, 20) 1301 LE64_BITMASK(BCH_MEMBER_GROUP, struct bch_member, flags, 20, 28) 1302 LE64_BITMASK(BCH_MEMBER_DURABILITY, struct bch_member, flags, 28, 30) 1303 LE64_BITMASK(BCH_MEMBER_FREESPACE_INITIALIZED, 1304 struct bch_member, flags, 30, 31) 1305 1306 #if 0 1307 LE64_BITMASK(BCH_MEMBER_NR_READ_ERRORS, struct bch_member, flags[1], 0, 20); 1308 LE64_BITMASK(BCH_MEMBER_NR_WRITE_ERRORS,struct bch_member, flags[1], 20, 40); 1309 #endif 1310 1311 #define BCH_MEMBER_STATES() \ 1312 x(rw, 0) \ 1313 x(ro, 1) \ 1314 x(failed, 2) \ 1315 x(spare, 3) 1316 1317 enum bch_member_state { 1318 #define x(t, n) BCH_MEMBER_STATE_##t = n, 1319 BCH_MEMBER_STATES() 1320 #undef x 1321 BCH_MEMBER_STATE_NR 1322 }; 1323 1324 struct bch_sb_field_members_v1 { 1325 struct bch_sb_field field; 1326 struct bch_member _members[]; //Members are now variable size 1327 }; 1328 1329 struct bch_sb_field_members_v2 { 1330 struct bch_sb_field field; 1331 __le16 member_bytes; //size of single member entry 1332 u8 pad[6]; 1333 struct bch_member _members[]; 1334 }; 1335 1336 /* BCH_SB_FIELD_crypt: */ 1337 1338 struct nonce { 1339 __le32 d[4]; 1340 }; 1341 1342 struct bch_key { 1343 __le64 key[4]; 1344 }; 1345 1346 #define BCH_KEY_MAGIC \ 1347 (((__u64) 'b' << 0)|((__u64) 'c' << 8)| \ 1348 ((__u64) 'h' << 16)|((__u64) '*' << 24)| \ 1349 ((__u64) '*' << 32)|((__u64) 'k' << 40)| \ 1350 ((__u64) 'e' << 48)|((__u64) 'y' << 56)) 1351 1352 struct bch_encrypted_key { 1353 __le64 magic; 1354 struct bch_key key; 1355 }; 1356 1357 /* 1358 * If this field is present in the superblock, it stores an encryption key which 1359 * is used encrypt all other data/metadata. The key will normally be encrypted 1360 * with the key userspace provides, but if encryption has been turned off we'll 1361 * just store the master key unencrypted in the superblock so we can access the 1362 * previously encrypted data. 1363 */ 1364 struct bch_sb_field_crypt { 1365 struct bch_sb_field field; 1366 1367 __le64 flags; 1368 __le64 kdf_flags; 1369 struct bch_encrypted_key key; 1370 }; 1371 1372 LE64_BITMASK(BCH_CRYPT_KDF_TYPE, struct bch_sb_field_crypt, flags, 0, 4); 1373 1374 enum bch_kdf_types { 1375 BCH_KDF_SCRYPT = 0, 1376 BCH_KDF_NR = 1, 1377 }; 1378 1379 /* stored as base 2 log of scrypt params: */ 1380 LE64_BITMASK(BCH_KDF_SCRYPT_N, struct bch_sb_field_crypt, kdf_flags, 0, 16); 1381 LE64_BITMASK(BCH_KDF_SCRYPT_R, struct bch_sb_field_crypt, kdf_flags, 16, 32); 1382 LE64_BITMASK(BCH_KDF_SCRYPT_P, struct bch_sb_field_crypt, kdf_flags, 32, 48); 1383 1384 /* BCH_SB_FIELD_replicas: */ 1385 1386 #define BCH_DATA_TYPES() \ 1387 x(free, 0) \ 1388 x(sb, 1) \ 1389 x(journal, 2) \ 1390 x(btree, 3) \ 1391 x(user, 4) \ 1392 x(cached, 5) \ 1393 x(parity, 6) \ 1394 x(stripe, 7) \ 1395 x(need_gc_gens, 8) \ 1396 x(need_discard, 9) 1397 1398 enum bch_data_type { 1399 #define x(t, n) BCH_DATA_##t, 1400 BCH_DATA_TYPES() 1401 #undef x 1402 BCH_DATA_NR 1403 }; 1404 1405 static inline bool data_type_is_empty(enum bch_data_type type) 1406 { 1407 switch (type) { 1408 case BCH_DATA_free: 1409 case BCH_DATA_need_gc_gens: 1410 case BCH_DATA_need_discard: 1411 return true; 1412 default: 1413 return false; 1414 } 1415 } 1416 1417 static inline bool data_type_is_hidden(enum bch_data_type type) 1418 { 1419 switch (type) { 1420 case BCH_DATA_sb: 1421 case BCH_DATA_journal: 1422 return true; 1423 default: 1424 return false; 1425 } 1426 } 1427 1428 struct bch_replicas_entry_v0 { 1429 __u8 data_type; 1430 __u8 nr_devs; 1431 __u8 devs[]; 1432 } __packed; 1433 1434 struct bch_sb_field_replicas_v0 { 1435 struct bch_sb_field field; 1436 struct bch_replicas_entry_v0 entries[]; 1437 } __packed __aligned(8); 1438 1439 struct bch_replicas_entry { 1440 __u8 data_type; 1441 __u8 nr_devs; 1442 __u8 nr_required; 1443 __u8 devs[]; 1444 } __packed; 1445 1446 #define replicas_entry_bytes(_i) \ 1447 (offsetof(typeof(*(_i)), devs) + (_i)->nr_devs) 1448 1449 struct bch_sb_field_replicas { 1450 struct bch_sb_field field; 1451 struct bch_replicas_entry entries[]; 1452 } __packed __aligned(8); 1453 1454 /* BCH_SB_FIELD_quota: */ 1455 1456 struct bch_sb_quota_counter { 1457 __le32 timelimit; 1458 __le32 warnlimit; 1459 }; 1460 1461 struct bch_sb_quota_type { 1462 __le64 flags; 1463 struct bch_sb_quota_counter c[Q_COUNTERS]; 1464 }; 1465 1466 struct bch_sb_field_quota { 1467 struct bch_sb_field field; 1468 struct bch_sb_quota_type q[QTYP_NR]; 1469 } __packed __aligned(8); 1470 1471 /* BCH_SB_FIELD_disk_groups: */ 1472 1473 #define BCH_SB_LABEL_SIZE 32 1474 1475 struct bch_disk_group { 1476 __u8 label[BCH_SB_LABEL_SIZE]; 1477 __le64 flags[2]; 1478 } __packed __aligned(8); 1479 1480 LE64_BITMASK(BCH_GROUP_DELETED, struct bch_disk_group, flags[0], 0, 1) 1481 LE64_BITMASK(BCH_GROUP_DATA_ALLOWED, struct bch_disk_group, flags[0], 1, 6) 1482 LE64_BITMASK(BCH_GROUP_PARENT, struct bch_disk_group, flags[0], 6, 24) 1483 1484 struct bch_sb_field_disk_groups { 1485 struct bch_sb_field field; 1486 struct bch_disk_group entries[]; 1487 } __packed __aligned(8); 1488 1489 /* BCH_SB_FIELD_counters */ 1490 1491 #define BCH_PERSISTENT_COUNTERS() \ 1492 x(io_read, 0) \ 1493 x(io_write, 1) \ 1494 x(io_move, 2) \ 1495 x(bucket_invalidate, 3) \ 1496 x(bucket_discard, 4) \ 1497 x(bucket_alloc, 5) \ 1498 x(bucket_alloc_fail, 6) \ 1499 x(btree_cache_scan, 7) \ 1500 x(btree_cache_reap, 8) \ 1501 x(btree_cache_cannibalize, 9) \ 1502 x(btree_cache_cannibalize_lock, 10) \ 1503 x(btree_cache_cannibalize_lock_fail, 11) \ 1504 x(btree_cache_cannibalize_unlock, 12) \ 1505 x(btree_node_write, 13) \ 1506 x(btree_node_read, 14) \ 1507 x(btree_node_compact, 15) \ 1508 x(btree_node_merge, 16) \ 1509 x(btree_node_split, 17) \ 1510 x(btree_node_rewrite, 18) \ 1511 x(btree_node_alloc, 19) \ 1512 x(btree_node_free, 20) \ 1513 x(btree_node_set_root, 21) \ 1514 x(btree_path_relock_fail, 22) \ 1515 x(btree_path_upgrade_fail, 23) \ 1516 x(btree_reserve_get_fail, 24) \ 1517 x(journal_entry_full, 25) \ 1518 x(journal_full, 26) \ 1519 x(journal_reclaim_finish, 27) \ 1520 x(journal_reclaim_start, 28) \ 1521 x(journal_write, 29) \ 1522 x(read_promote, 30) \ 1523 x(read_bounce, 31) \ 1524 x(read_split, 33) \ 1525 x(read_retry, 32) \ 1526 x(read_reuse_race, 34) \ 1527 x(move_extent_read, 35) \ 1528 x(move_extent_write, 36) \ 1529 x(move_extent_finish, 37) \ 1530 x(move_extent_fail, 38) \ 1531 x(move_extent_alloc_mem_fail, 39) \ 1532 x(copygc, 40) \ 1533 x(copygc_wait, 41) \ 1534 x(gc_gens_end, 42) \ 1535 x(gc_gens_start, 43) \ 1536 x(trans_blocked_journal_reclaim, 44) \ 1537 x(trans_restart_btree_node_reused, 45) \ 1538 x(trans_restart_btree_node_split, 46) \ 1539 x(trans_restart_fault_inject, 47) \ 1540 x(trans_restart_iter_upgrade, 48) \ 1541 x(trans_restart_journal_preres_get, 49) \ 1542 x(trans_restart_journal_reclaim, 50) \ 1543 x(trans_restart_journal_res_get, 51) \ 1544 x(trans_restart_key_cache_key_realloced, 52) \ 1545 x(trans_restart_key_cache_raced, 53) \ 1546 x(trans_restart_mark_replicas, 54) \ 1547 x(trans_restart_mem_realloced, 55) \ 1548 x(trans_restart_memory_allocation_failure, 56) \ 1549 x(trans_restart_relock, 57) \ 1550 x(trans_restart_relock_after_fill, 58) \ 1551 x(trans_restart_relock_key_cache_fill, 59) \ 1552 x(trans_restart_relock_next_node, 60) \ 1553 x(trans_restart_relock_parent_for_fill, 61) \ 1554 x(trans_restart_relock_path, 62) \ 1555 x(trans_restart_relock_path_intent, 63) \ 1556 x(trans_restart_too_many_iters, 64) \ 1557 x(trans_restart_traverse, 65) \ 1558 x(trans_restart_upgrade, 66) \ 1559 x(trans_restart_would_deadlock, 67) \ 1560 x(trans_restart_would_deadlock_write, 68) \ 1561 x(trans_restart_injected, 69) \ 1562 x(trans_restart_key_cache_upgrade, 70) \ 1563 x(trans_traverse_all, 71) \ 1564 x(transaction_commit, 72) \ 1565 x(write_super, 73) \ 1566 x(trans_restart_would_deadlock_recursion_limit, 74) \ 1567 x(trans_restart_write_buffer_flush, 75) \ 1568 x(trans_restart_split_race, 76) 1569 1570 enum bch_persistent_counters { 1571 #define x(t, n, ...) BCH_COUNTER_##t, 1572 BCH_PERSISTENT_COUNTERS() 1573 #undef x 1574 BCH_COUNTER_NR 1575 }; 1576 1577 struct bch_sb_field_counters { 1578 struct bch_sb_field field; 1579 __le64 d[]; 1580 }; 1581 1582 /* 1583 * On clean shutdown, store btree roots and current journal sequence number in 1584 * the superblock: 1585 */ 1586 struct jset_entry { 1587 __le16 u64s; 1588 __u8 btree_id; 1589 __u8 level; 1590 __u8 type; /* designates what this jset holds */ 1591 __u8 pad[3]; 1592 1593 struct bkey_i start[0]; 1594 __u64 _data[]; 1595 }; 1596 1597 struct bch_sb_field_clean { 1598 struct bch_sb_field field; 1599 1600 __le32 flags; 1601 __le16 _read_clock; /* no longer used */ 1602 __le16 _write_clock; 1603 __le64 journal_seq; 1604 1605 struct jset_entry start[0]; 1606 __u64 _data[]; 1607 }; 1608 1609 struct journal_seq_blacklist_entry { 1610 __le64 start; 1611 __le64 end; 1612 }; 1613 1614 struct bch_sb_field_journal_seq_blacklist { 1615 struct bch_sb_field field; 1616 struct journal_seq_blacklist_entry start[]; 1617 }; 1618 1619 struct bch_sb_field_errors { 1620 struct bch_sb_field field; 1621 struct bch_sb_field_error_entry { 1622 __le64 v; 1623 __le64 last_error_time; 1624 } entries[]; 1625 }; 1626 1627 LE64_BITMASK(BCH_SB_ERROR_ENTRY_ID, struct bch_sb_field_error_entry, v, 0, 16); 1628 LE64_BITMASK(BCH_SB_ERROR_ENTRY_NR, struct bch_sb_field_error_entry, v, 16, 64); 1629 1630 /* Superblock: */ 1631 1632 /* 1633 * New versioning scheme: 1634 * One common version number for all on disk data structures - superblock, btree 1635 * nodes, journal entries 1636 */ 1637 #define BCH_VERSION_MAJOR(_v) ((__u16) ((_v) >> 10)) 1638 #define BCH_VERSION_MINOR(_v) ((__u16) ((_v) & ~(~0U << 10))) 1639 #define BCH_VERSION(_major, _minor) (((_major) << 10)|(_minor) << 0) 1640 1641 #define RECOVERY_PASS_ALL_FSCK (1ULL << 63) 1642 1643 #define BCH_METADATA_VERSIONS() \ 1644 x(bkey_renumber, BCH_VERSION(0, 10), \ 1645 RECOVERY_PASS_ALL_FSCK) \ 1646 x(inode_btree_change, BCH_VERSION(0, 11), \ 1647 RECOVERY_PASS_ALL_FSCK) \ 1648 x(snapshot, BCH_VERSION(0, 12), \ 1649 RECOVERY_PASS_ALL_FSCK) \ 1650 x(inode_backpointers, BCH_VERSION(0, 13), \ 1651 RECOVERY_PASS_ALL_FSCK) \ 1652 x(btree_ptr_sectors_written, BCH_VERSION(0, 14), \ 1653 RECOVERY_PASS_ALL_FSCK) \ 1654 x(snapshot_2, BCH_VERSION(0, 15), \ 1655 BIT_ULL(BCH_RECOVERY_PASS_fs_upgrade_for_subvolumes)| \ 1656 BIT_ULL(BCH_RECOVERY_PASS_initialize_subvolumes)| \ 1657 RECOVERY_PASS_ALL_FSCK) \ 1658 x(reflink_p_fix, BCH_VERSION(0, 16), \ 1659 BIT_ULL(BCH_RECOVERY_PASS_fix_reflink_p)) \ 1660 x(subvol_dirent, BCH_VERSION(0, 17), \ 1661 RECOVERY_PASS_ALL_FSCK) \ 1662 x(inode_v2, BCH_VERSION(0, 18), \ 1663 RECOVERY_PASS_ALL_FSCK) \ 1664 x(freespace, BCH_VERSION(0, 19), \ 1665 RECOVERY_PASS_ALL_FSCK) \ 1666 x(alloc_v4, BCH_VERSION(0, 20), \ 1667 RECOVERY_PASS_ALL_FSCK) \ 1668 x(new_data_types, BCH_VERSION(0, 21), \ 1669 RECOVERY_PASS_ALL_FSCK) \ 1670 x(backpointers, BCH_VERSION(0, 22), \ 1671 RECOVERY_PASS_ALL_FSCK) \ 1672 x(inode_v3, BCH_VERSION(0, 23), \ 1673 RECOVERY_PASS_ALL_FSCK) \ 1674 x(unwritten_extents, BCH_VERSION(0, 24), \ 1675 RECOVERY_PASS_ALL_FSCK) \ 1676 x(bucket_gens, BCH_VERSION(0, 25), \ 1677 BIT_ULL(BCH_RECOVERY_PASS_bucket_gens_init)| \ 1678 RECOVERY_PASS_ALL_FSCK) \ 1679 x(lru_v2, BCH_VERSION(0, 26), \ 1680 RECOVERY_PASS_ALL_FSCK) \ 1681 x(fragmentation_lru, BCH_VERSION(0, 27), \ 1682 RECOVERY_PASS_ALL_FSCK) \ 1683 x(no_bps_in_alloc_keys, BCH_VERSION(0, 28), \ 1684 RECOVERY_PASS_ALL_FSCK) \ 1685 x(snapshot_trees, BCH_VERSION(0, 29), \ 1686 RECOVERY_PASS_ALL_FSCK) \ 1687 x(major_minor, BCH_VERSION(1, 0), \ 1688 0) \ 1689 x(snapshot_skiplists, BCH_VERSION(1, 1), \ 1690 BIT_ULL(BCH_RECOVERY_PASS_check_snapshots)) \ 1691 x(deleted_inodes, BCH_VERSION(1, 2), \ 1692 BIT_ULL(BCH_RECOVERY_PASS_check_inodes)) \ 1693 x(rebalance_work, BCH_VERSION(1, 3), \ 1694 BIT_ULL(BCH_RECOVERY_PASS_set_fs_needs_rebalance)) 1695 1696 enum bcachefs_metadata_version { 1697 bcachefs_metadata_version_min = 9, 1698 #define x(t, n, upgrade_passes) bcachefs_metadata_version_##t = n, 1699 BCH_METADATA_VERSIONS() 1700 #undef x 1701 bcachefs_metadata_version_max 1702 }; 1703 1704 static const __maybe_unused 1705 unsigned bcachefs_metadata_required_upgrade_below = bcachefs_metadata_version_rebalance_work; 1706 1707 #define bcachefs_metadata_version_current (bcachefs_metadata_version_max - 1) 1708 1709 #define BCH_SB_SECTOR 8 1710 #define BCH_SB_MEMBERS_MAX 64 /* XXX kill */ 1711 1712 struct bch_sb_layout { 1713 __uuid_t magic; /* bcachefs superblock UUID */ 1714 __u8 layout_type; 1715 __u8 sb_max_size_bits; /* base 2 of 512 byte sectors */ 1716 __u8 nr_superblocks; 1717 __u8 pad[5]; 1718 __le64 sb_offset[61]; 1719 } __packed __aligned(8); 1720 1721 #define BCH_SB_LAYOUT_SECTOR 7 1722 1723 /* 1724 * @offset - sector where this sb was written 1725 * @version - on disk format version 1726 * @version_min - Oldest metadata version this filesystem contains; so we can 1727 * safely drop compatibility code and refuse to mount filesystems 1728 * we'd need it for 1729 * @magic - identifies as a bcachefs superblock (BCHFS_MAGIC) 1730 * @seq - incremented each time superblock is written 1731 * @uuid - used for generating various magic numbers and identifying 1732 * member devices, never changes 1733 * @user_uuid - user visible UUID, may be changed 1734 * @label - filesystem label 1735 * @seq - identifies most recent superblock, incremented each time 1736 * superblock is written 1737 * @features - enabled incompatible features 1738 */ 1739 struct bch_sb { 1740 struct bch_csum csum; 1741 __le16 version; 1742 __le16 version_min; 1743 __le16 pad[2]; 1744 __uuid_t magic; 1745 __uuid_t uuid; 1746 __uuid_t user_uuid; 1747 __u8 label[BCH_SB_LABEL_SIZE]; 1748 __le64 offset; 1749 __le64 seq; 1750 1751 __le16 block_size; 1752 __u8 dev_idx; 1753 __u8 nr_devices; 1754 __le32 u64s; 1755 1756 __le64 time_base_lo; 1757 __le32 time_base_hi; 1758 __le32 time_precision; 1759 1760 __le64 flags[8]; 1761 __le64 features[2]; 1762 __le64 compat[2]; 1763 1764 struct bch_sb_layout layout; 1765 1766 struct bch_sb_field start[0]; 1767 __le64 _data[]; 1768 } __packed __aligned(8); 1769 1770 /* 1771 * Flags: 1772 * BCH_SB_INITALIZED - set on first mount 1773 * BCH_SB_CLEAN - did we shut down cleanly? Just a hint, doesn't affect 1774 * behaviour of mount/recovery path: 1775 * BCH_SB_INODE_32BIT - limit inode numbers to 32 bits 1776 * BCH_SB_128_BIT_MACS - 128 bit macs instead of 80 1777 * BCH_SB_ENCRYPTION_TYPE - if nonzero encryption is enabled; overrides 1778 * DATA/META_CSUM_TYPE. Also indicates encryption 1779 * algorithm in use, if/when we get more than one 1780 */ 1781 1782 LE16_BITMASK(BCH_SB_BLOCK_SIZE, struct bch_sb, block_size, 0, 16); 1783 1784 LE64_BITMASK(BCH_SB_INITIALIZED, struct bch_sb, flags[0], 0, 1); 1785 LE64_BITMASK(BCH_SB_CLEAN, struct bch_sb, flags[0], 1, 2); 1786 LE64_BITMASK(BCH_SB_CSUM_TYPE, struct bch_sb, flags[0], 2, 8); 1787 LE64_BITMASK(BCH_SB_ERROR_ACTION, struct bch_sb, flags[0], 8, 12); 1788 1789 LE64_BITMASK(BCH_SB_BTREE_NODE_SIZE, struct bch_sb, flags[0], 12, 28); 1790 1791 LE64_BITMASK(BCH_SB_GC_RESERVE, struct bch_sb, flags[0], 28, 33); 1792 LE64_BITMASK(BCH_SB_ROOT_RESERVE, struct bch_sb, flags[0], 33, 40); 1793 1794 LE64_BITMASK(BCH_SB_META_CSUM_TYPE, struct bch_sb, flags[0], 40, 44); 1795 LE64_BITMASK(BCH_SB_DATA_CSUM_TYPE, struct bch_sb, flags[0], 44, 48); 1796 1797 LE64_BITMASK(BCH_SB_META_REPLICAS_WANT, struct bch_sb, flags[0], 48, 52); 1798 LE64_BITMASK(BCH_SB_DATA_REPLICAS_WANT, struct bch_sb, flags[0], 52, 56); 1799 1800 LE64_BITMASK(BCH_SB_POSIX_ACL, struct bch_sb, flags[0], 56, 57); 1801 LE64_BITMASK(BCH_SB_USRQUOTA, struct bch_sb, flags[0], 57, 58); 1802 LE64_BITMASK(BCH_SB_GRPQUOTA, struct bch_sb, flags[0], 58, 59); 1803 LE64_BITMASK(BCH_SB_PRJQUOTA, struct bch_sb, flags[0], 59, 60); 1804 1805 LE64_BITMASK(BCH_SB_HAS_ERRORS, struct bch_sb, flags[0], 60, 61); 1806 LE64_BITMASK(BCH_SB_HAS_TOPOLOGY_ERRORS,struct bch_sb, flags[0], 61, 62); 1807 1808 LE64_BITMASK(BCH_SB_BIG_ENDIAN, struct bch_sb, flags[0], 62, 63); 1809 1810 LE64_BITMASK(BCH_SB_STR_HASH_TYPE, struct bch_sb, flags[1], 0, 4); 1811 LE64_BITMASK(BCH_SB_COMPRESSION_TYPE_LO,struct bch_sb, flags[1], 4, 8); 1812 LE64_BITMASK(BCH_SB_INODE_32BIT, struct bch_sb, flags[1], 8, 9); 1813 1814 LE64_BITMASK(BCH_SB_128_BIT_MACS, struct bch_sb, flags[1], 9, 10); 1815 LE64_BITMASK(BCH_SB_ENCRYPTION_TYPE, struct bch_sb, flags[1], 10, 14); 1816 1817 /* 1818 * Max size of an extent that may require bouncing to read or write 1819 * (checksummed, compressed): 64k 1820 */ 1821 LE64_BITMASK(BCH_SB_ENCODED_EXTENT_MAX_BITS, 1822 struct bch_sb, flags[1], 14, 20); 1823 1824 LE64_BITMASK(BCH_SB_META_REPLICAS_REQ, struct bch_sb, flags[1], 20, 24); 1825 LE64_BITMASK(BCH_SB_DATA_REPLICAS_REQ, struct bch_sb, flags[1], 24, 28); 1826 1827 LE64_BITMASK(BCH_SB_PROMOTE_TARGET, struct bch_sb, flags[1], 28, 40); 1828 LE64_BITMASK(BCH_SB_FOREGROUND_TARGET, struct bch_sb, flags[1], 40, 52); 1829 LE64_BITMASK(BCH_SB_BACKGROUND_TARGET, struct bch_sb, flags[1], 52, 64); 1830 1831 LE64_BITMASK(BCH_SB_BACKGROUND_COMPRESSION_TYPE_LO, 1832 struct bch_sb, flags[2], 0, 4); 1833 LE64_BITMASK(BCH_SB_GC_RESERVE_BYTES, struct bch_sb, flags[2], 4, 64); 1834 1835 LE64_BITMASK(BCH_SB_ERASURE_CODE, struct bch_sb, flags[3], 0, 16); 1836 LE64_BITMASK(BCH_SB_METADATA_TARGET, struct bch_sb, flags[3], 16, 28); 1837 LE64_BITMASK(BCH_SB_SHARD_INUMS, struct bch_sb, flags[3], 28, 29); 1838 LE64_BITMASK(BCH_SB_INODES_USE_KEY_CACHE,struct bch_sb, flags[3], 29, 30); 1839 LE64_BITMASK(BCH_SB_JOURNAL_FLUSH_DELAY,struct bch_sb, flags[3], 30, 62); 1840 LE64_BITMASK(BCH_SB_JOURNAL_FLUSH_DISABLED,struct bch_sb, flags[3], 62, 63); 1841 LE64_BITMASK(BCH_SB_JOURNAL_RECLAIM_DELAY,struct bch_sb, flags[4], 0, 32); 1842 LE64_BITMASK(BCH_SB_JOURNAL_TRANSACTION_NAMES,struct bch_sb, flags[4], 32, 33); 1843 LE64_BITMASK(BCH_SB_NOCOW, struct bch_sb, flags[4], 33, 34); 1844 LE64_BITMASK(BCH_SB_WRITE_BUFFER_SIZE, struct bch_sb, flags[4], 34, 54); 1845 LE64_BITMASK(BCH_SB_VERSION_UPGRADE, struct bch_sb, flags[4], 54, 56); 1846 1847 LE64_BITMASK(BCH_SB_COMPRESSION_TYPE_HI,struct bch_sb, flags[4], 56, 60); 1848 LE64_BITMASK(BCH_SB_BACKGROUND_COMPRESSION_TYPE_HI, 1849 struct bch_sb, flags[4], 60, 64); 1850 1851 LE64_BITMASK(BCH_SB_VERSION_UPGRADE_COMPLETE, 1852 struct bch_sb, flags[5], 0, 16); 1853 1854 static inline __u64 BCH_SB_COMPRESSION_TYPE(const struct bch_sb *sb) 1855 { 1856 return BCH_SB_COMPRESSION_TYPE_LO(sb) | (BCH_SB_COMPRESSION_TYPE_HI(sb) << 4); 1857 } 1858 1859 static inline void SET_BCH_SB_COMPRESSION_TYPE(struct bch_sb *sb, __u64 v) 1860 { 1861 SET_BCH_SB_COMPRESSION_TYPE_LO(sb, v); 1862 SET_BCH_SB_COMPRESSION_TYPE_HI(sb, v >> 4); 1863 } 1864 1865 static inline __u64 BCH_SB_BACKGROUND_COMPRESSION_TYPE(const struct bch_sb *sb) 1866 { 1867 return BCH_SB_BACKGROUND_COMPRESSION_TYPE_LO(sb) | 1868 (BCH_SB_BACKGROUND_COMPRESSION_TYPE_HI(sb) << 4); 1869 } 1870 1871 static inline void SET_BCH_SB_BACKGROUND_COMPRESSION_TYPE(struct bch_sb *sb, __u64 v) 1872 { 1873 SET_BCH_SB_BACKGROUND_COMPRESSION_TYPE_LO(sb, v); 1874 SET_BCH_SB_BACKGROUND_COMPRESSION_TYPE_HI(sb, v >> 4); 1875 } 1876 1877 /* 1878 * Features: 1879 * 1880 * journal_seq_blacklist_v3: gates BCH_SB_FIELD_journal_seq_blacklist 1881 * reflink: gates KEY_TYPE_reflink 1882 * inline_data: gates KEY_TYPE_inline_data 1883 * new_siphash: gates BCH_STR_HASH_siphash 1884 * new_extent_overwrite: gates BTREE_NODE_NEW_EXTENT_OVERWRITE 1885 */ 1886 #define BCH_SB_FEATURES() \ 1887 x(lz4, 0) \ 1888 x(gzip, 1) \ 1889 x(zstd, 2) \ 1890 x(atomic_nlink, 3) \ 1891 x(ec, 4) \ 1892 x(journal_seq_blacklist_v3, 5) \ 1893 x(reflink, 6) \ 1894 x(new_siphash, 7) \ 1895 x(inline_data, 8) \ 1896 x(new_extent_overwrite, 9) \ 1897 x(incompressible, 10) \ 1898 x(btree_ptr_v2, 11) \ 1899 x(extents_above_btree_updates, 12) \ 1900 x(btree_updates_journalled, 13) \ 1901 x(reflink_inline_data, 14) \ 1902 x(new_varint, 15) \ 1903 x(journal_no_flush, 16) \ 1904 x(alloc_v2, 17) \ 1905 x(extents_across_btree_nodes, 18) 1906 1907 #define BCH_SB_FEATURES_ALWAYS \ 1908 ((1ULL << BCH_FEATURE_new_extent_overwrite)| \ 1909 (1ULL << BCH_FEATURE_extents_above_btree_updates)|\ 1910 (1ULL << BCH_FEATURE_btree_updates_journalled)|\ 1911 (1ULL << BCH_FEATURE_alloc_v2)|\ 1912 (1ULL << BCH_FEATURE_extents_across_btree_nodes)) 1913 1914 #define BCH_SB_FEATURES_ALL \ 1915 (BCH_SB_FEATURES_ALWAYS| \ 1916 (1ULL << BCH_FEATURE_new_siphash)| \ 1917 (1ULL << BCH_FEATURE_btree_ptr_v2)| \ 1918 (1ULL << BCH_FEATURE_new_varint)| \ 1919 (1ULL << BCH_FEATURE_journal_no_flush)) 1920 1921 enum bch_sb_feature { 1922 #define x(f, n) BCH_FEATURE_##f, 1923 BCH_SB_FEATURES() 1924 #undef x 1925 BCH_FEATURE_NR, 1926 }; 1927 1928 #define BCH_SB_COMPAT() \ 1929 x(alloc_info, 0) \ 1930 x(alloc_metadata, 1) \ 1931 x(extents_above_btree_updates_done, 2) \ 1932 x(bformat_overflow_done, 3) 1933 1934 enum bch_sb_compat { 1935 #define x(f, n) BCH_COMPAT_##f, 1936 BCH_SB_COMPAT() 1937 #undef x 1938 BCH_COMPAT_NR, 1939 }; 1940 1941 /* options: */ 1942 1943 #define BCH_VERSION_UPGRADE_OPTS() \ 1944 x(compatible, 0) \ 1945 x(incompatible, 1) \ 1946 x(none, 2) 1947 1948 enum bch_version_upgrade_opts { 1949 #define x(t, n) BCH_VERSION_UPGRADE_##t = n, 1950 BCH_VERSION_UPGRADE_OPTS() 1951 #undef x 1952 }; 1953 1954 #define BCH_REPLICAS_MAX 4U 1955 1956 #define BCH_BKEY_PTRS_MAX 16U 1957 1958 #define BCH_ERROR_ACTIONS() \ 1959 x(continue, 0) \ 1960 x(ro, 1) \ 1961 x(panic, 2) 1962 1963 enum bch_error_actions { 1964 #define x(t, n) BCH_ON_ERROR_##t = n, 1965 BCH_ERROR_ACTIONS() 1966 #undef x 1967 BCH_ON_ERROR_NR 1968 }; 1969 1970 #define BCH_STR_HASH_TYPES() \ 1971 x(crc32c, 0) \ 1972 x(crc64, 1) \ 1973 x(siphash_old, 2) \ 1974 x(siphash, 3) 1975 1976 enum bch_str_hash_type { 1977 #define x(t, n) BCH_STR_HASH_##t = n, 1978 BCH_STR_HASH_TYPES() 1979 #undef x 1980 BCH_STR_HASH_NR 1981 }; 1982 1983 #define BCH_STR_HASH_OPTS() \ 1984 x(crc32c, 0) \ 1985 x(crc64, 1) \ 1986 x(siphash, 2) 1987 1988 enum bch_str_hash_opts { 1989 #define x(t, n) BCH_STR_HASH_OPT_##t = n, 1990 BCH_STR_HASH_OPTS() 1991 #undef x 1992 BCH_STR_HASH_OPT_NR 1993 }; 1994 1995 #define BCH_CSUM_TYPES() \ 1996 x(none, 0) \ 1997 x(crc32c_nonzero, 1) \ 1998 x(crc64_nonzero, 2) \ 1999 x(chacha20_poly1305_80, 3) \ 2000 x(chacha20_poly1305_128, 4) \ 2001 x(crc32c, 5) \ 2002 x(crc64, 6) \ 2003 x(xxhash, 7) 2004 2005 enum bch_csum_type { 2006 #define x(t, n) BCH_CSUM_##t = n, 2007 BCH_CSUM_TYPES() 2008 #undef x 2009 BCH_CSUM_NR 2010 }; 2011 2012 static const __maybe_unused unsigned bch_crc_bytes[] = { 2013 [BCH_CSUM_none] = 0, 2014 [BCH_CSUM_crc32c_nonzero] = 4, 2015 [BCH_CSUM_crc32c] = 4, 2016 [BCH_CSUM_crc64_nonzero] = 8, 2017 [BCH_CSUM_crc64] = 8, 2018 [BCH_CSUM_xxhash] = 8, 2019 [BCH_CSUM_chacha20_poly1305_80] = 10, 2020 [BCH_CSUM_chacha20_poly1305_128] = 16, 2021 }; 2022 2023 static inline _Bool bch2_csum_type_is_encryption(enum bch_csum_type type) 2024 { 2025 switch (type) { 2026 case BCH_CSUM_chacha20_poly1305_80: 2027 case BCH_CSUM_chacha20_poly1305_128: 2028 return true; 2029 default: 2030 return false; 2031 } 2032 } 2033 2034 #define BCH_CSUM_OPTS() \ 2035 x(none, 0) \ 2036 x(crc32c, 1) \ 2037 x(crc64, 2) \ 2038 x(xxhash, 3) 2039 2040 enum bch_csum_opts { 2041 #define x(t, n) BCH_CSUM_OPT_##t = n, 2042 BCH_CSUM_OPTS() 2043 #undef x 2044 BCH_CSUM_OPT_NR 2045 }; 2046 2047 #define BCH_COMPRESSION_TYPES() \ 2048 x(none, 0) \ 2049 x(lz4_old, 1) \ 2050 x(gzip, 2) \ 2051 x(lz4, 3) \ 2052 x(zstd, 4) \ 2053 x(incompressible, 5) 2054 2055 enum bch_compression_type { 2056 #define x(t, n) BCH_COMPRESSION_TYPE_##t = n, 2057 BCH_COMPRESSION_TYPES() 2058 #undef x 2059 BCH_COMPRESSION_TYPE_NR 2060 }; 2061 2062 #define BCH_COMPRESSION_OPTS() \ 2063 x(none, 0) \ 2064 x(lz4, 1) \ 2065 x(gzip, 2) \ 2066 x(zstd, 3) 2067 2068 enum bch_compression_opts { 2069 #define x(t, n) BCH_COMPRESSION_OPT_##t = n, 2070 BCH_COMPRESSION_OPTS() 2071 #undef x 2072 BCH_COMPRESSION_OPT_NR 2073 }; 2074 2075 /* 2076 * Magic numbers 2077 * 2078 * The various other data structures have their own magic numbers, which are 2079 * xored with the first part of the cache set's UUID 2080 */ 2081 2082 #define BCACHE_MAGIC \ 2083 UUID_INIT(0xc68573f6, 0x4e1a, 0x45ca, \ 2084 0x82, 0x65, 0xf5, 0x7f, 0x48, 0xba, 0x6d, 0x81) 2085 #define BCHFS_MAGIC \ 2086 UUID_INIT(0xc68573f6, 0x66ce, 0x90a9, \ 2087 0xd9, 0x6a, 0x60, 0xcf, 0x80, 0x3d, 0xf7, 0xef) 2088 2089 #define BCACHEFS_STATFS_MAGIC 0xca451a4e 2090 2091 #define JSET_MAGIC __cpu_to_le64(0x245235c1a3625032ULL) 2092 #define BSET_MAGIC __cpu_to_le64(0x90135c78b99e07f5ULL) 2093 2094 static inline __le64 __bch2_sb_magic(struct bch_sb *sb) 2095 { 2096 __le64 ret; 2097 2098 memcpy(&ret, &sb->uuid, sizeof(ret)); 2099 return ret; 2100 } 2101 2102 static inline __u64 __jset_magic(struct bch_sb *sb) 2103 { 2104 return __le64_to_cpu(__bch2_sb_magic(sb) ^ JSET_MAGIC); 2105 } 2106 2107 static inline __u64 __bset_magic(struct bch_sb *sb) 2108 { 2109 return __le64_to_cpu(__bch2_sb_magic(sb) ^ BSET_MAGIC); 2110 } 2111 2112 /* Journal */ 2113 2114 #define JSET_KEYS_U64s (sizeof(struct jset_entry) / sizeof(__u64)) 2115 2116 #define BCH_JSET_ENTRY_TYPES() \ 2117 x(btree_keys, 0) \ 2118 x(btree_root, 1) \ 2119 x(prio_ptrs, 2) \ 2120 x(blacklist, 3) \ 2121 x(blacklist_v2, 4) \ 2122 x(usage, 5) \ 2123 x(data_usage, 6) \ 2124 x(clock, 7) \ 2125 x(dev_usage, 8) \ 2126 x(log, 9) \ 2127 x(overwrite, 10) 2128 2129 enum { 2130 #define x(f, nr) BCH_JSET_ENTRY_##f = nr, 2131 BCH_JSET_ENTRY_TYPES() 2132 #undef x 2133 BCH_JSET_ENTRY_NR 2134 }; 2135 2136 /* 2137 * Journal sequence numbers can be blacklisted: bsets record the max sequence 2138 * number of all the journal entries they contain updates for, so that on 2139 * recovery we can ignore those bsets that contain index updates newer that what 2140 * made it into the journal. 2141 * 2142 * This means that we can't reuse that journal_seq - we have to skip it, and 2143 * then record that we skipped it so that the next time we crash and recover we 2144 * don't think there was a missing journal entry. 2145 */ 2146 struct jset_entry_blacklist { 2147 struct jset_entry entry; 2148 __le64 seq; 2149 }; 2150 2151 struct jset_entry_blacklist_v2 { 2152 struct jset_entry entry; 2153 __le64 start; 2154 __le64 end; 2155 }; 2156 2157 #define BCH_FS_USAGE_TYPES() \ 2158 x(reserved, 0) \ 2159 x(inodes, 1) \ 2160 x(key_version, 2) 2161 2162 enum { 2163 #define x(f, nr) BCH_FS_USAGE_##f = nr, 2164 BCH_FS_USAGE_TYPES() 2165 #undef x 2166 BCH_FS_USAGE_NR 2167 }; 2168 2169 struct jset_entry_usage { 2170 struct jset_entry entry; 2171 __le64 v; 2172 } __packed; 2173 2174 struct jset_entry_data_usage { 2175 struct jset_entry entry; 2176 __le64 v; 2177 struct bch_replicas_entry r; 2178 } __packed; 2179 2180 struct jset_entry_clock { 2181 struct jset_entry entry; 2182 __u8 rw; 2183 __u8 pad[7]; 2184 __le64 time; 2185 } __packed; 2186 2187 struct jset_entry_dev_usage_type { 2188 __le64 buckets; 2189 __le64 sectors; 2190 __le64 fragmented; 2191 } __packed; 2192 2193 struct jset_entry_dev_usage { 2194 struct jset_entry entry; 2195 __le32 dev; 2196 __u32 pad; 2197 2198 __le64 buckets_ec; 2199 __le64 _buckets_unavailable; /* No longer used */ 2200 2201 struct jset_entry_dev_usage_type d[]; 2202 }; 2203 2204 static inline unsigned jset_entry_dev_usage_nr_types(struct jset_entry_dev_usage *u) 2205 { 2206 return (vstruct_bytes(&u->entry) - sizeof(struct jset_entry_dev_usage)) / 2207 sizeof(struct jset_entry_dev_usage_type); 2208 } 2209 2210 struct jset_entry_log { 2211 struct jset_entry entry; 2212 u8 d[]; 2213 } __packed; 2214 2215 /* 2216 * On disk format for a journal entry: 2217 * seq is monotonically increasing; every journal entry has its own unique 2218 * sequence number. 2219 * 2220 * last_seq is the oldest journal entry that still has keys the btree hasn't 2221 * flushed to disk yet. 2222 * 2223 * version is for on disk format changes. 2224 */ 2225 struct jset { 2226 struct bch_csum csum; 2227 2228 __le64 magic; 2229 __le64 seq; 2230 __le32 version; 2231 __le32 flags; 2232 2233 __le32 u64s; /* size of d[] in u64s */ 2234 2235 __u8 encrypted_start[0]; 2236 2237 __le16 _read_clock; /* no longer used */ 2238 __le16 _write_clock; 2239 2240 /* Sequence number of oldest dirty journal entry */ 2241 __le64 last_seq; 2242 2243 2244 struct jset_entry start[0]; 2245 __u64 _data[]; 2246 } __packed __aligned(8); 2247 2248 LE32_BITMASK(JSET_CSUM_TYPE, struct jset, flags, 0, 4); 2249 LE32_BITMASK(JSET_BIG_ENDIAN, struct jset, flags, 4, 5); 2250 LE32_BITMASK(JSET_NO_FLUSH, struct jset, flags, 5, 6); 2251 2252 #define BCH_JOURNAL_BUCKETS_MIN 8 2253 2254 /* Btree: */ 2255 2256 enum btree_id_flags { 2257 BTREE_ID_EXTENTS = BIT(0), 2258 BTREE_ID_SNAPSHOTS = BIT(1), 2259 BTREE_ID_SNAPSHOT_FIELD = BIT(2), 2260 BTREE_ID_DATA = BIT(3), 2261 }; 2262 2263 #define BCH_BTREE_IDS() \ 2264 x(extents, 0, BTREE_ID_EXTENTS|BTREE_ID_SNAPSHOTS|BTREE_ID_DATA,\ 2265 BIT_ULL(KEY_TYPE_whiteout)| \ 2266 BIT_ULL(KEY_TYPE_error)| \ 2267 BIT_ULL(KEY_TYPE_cookie)| \ 2268 BIT_ULL(KEY_TYPE_extent)| \ 2269 BIT_ULL(KEY_TYPE_reservation)| \ 2270 BIT_ULL(KEY_TYPE_reflink_p)| \ 2271 BIT_ULL(KEY_TYPE_inline_data)) \ 2272 x(inodes, 1, BTREE_ID_SNAPSHOTS, \ 2273 BIT_ULL(KEY_TYPE_whiteout)| \ 2274 BIT_ULL(KEY_TYPE_inode)| \ 2275 BIT_ULL(KEY_TYPE_inode_v2)| \ 2276 BIT_ULL(KEY_TYPE_inode_v3)| \ 2277 BIT_ULL(KEY_TYPE_inode_generation)) \ 2278 x(dirents, 2, BTREE_ID_SNAPSHOTS, \ 2279 BIT_ULL(KEY_TYPE_whiteout)| \ 2280 BIT_ULL(KEY_TYPE_hash_whiteout)| \ 2281 BIT_ULL(KEY_TYPE_dirent)) \ 2282 x(xattrs, 3, BTREE_ID_SNAPSHOTS, \ 2283 BIT_ULL(KEY_TYPE_whiteout)| \ 2284 BIT_ULL(KEY_TYPE_cookie)| \ 2285 BIT_ULL(KEY_TYPE_hash_whiteout)| \ 2286 BIT_ULL(KEY_TYPE_xattr)) \ 2287 x(alloc, 4, 0, \ 2288 BIT_ULL(KEY_TYPE_alloc)| \ 2289 BIT_ULL(KEY_TYPE_alloc_v2)| \ 2290 BIT_ULL(KEY_TYPE_alloc_v3)| \ 2291 BIT_ULL(KEY_TYPE_alloc_v4)) \ 2292 x(quotas, 5, 0, \ 2293 BIT_ULL(KEY_TYPE_quota)) \ 2294 x(stripes, 6, 0, \ 2295 BIT_ULL(KEY_TYPE_stripe)) \ 2296 x(reflink, 7, BTREE_ID_EXTENTS|BTREE_ID_DATA, \ 2297 BIT_ULL(KEY_TYPE_reflink_v)| \ 2298 BIT_ULL(KEY_TYPE_indirect_inline_data)) \ 2299 x(subvolumes, 8, 0, \ 2300 BIT_ULL(KEY_TYPE_subvolume)) \ 2301 x(snapshots, 9, 0, \ 2302 BIT_ULL(KEY_TYPE_snapshot)) \ 2303 x(lru, 10, 0, \ 2304 BIT_ULL(KEY_TYPE_set)) \ 2305 x(freespace, 11, BTREE_ID_EXTENTS, \ 2306 BIT_ULL(KEY_TYPE_set)) \ 2307 x(need_discard, 12, 0, \ 2308 BIT_ULL(KEY_TYPE_set)) \ 2309 x(backpointers, 13, 0, \ 2310 BIT_ULL(KEY_TYPE_backpointer)) \ 2311 x(bucket_gens, 14, 0, \ 2312 BIT_ULL(KEY_TYPE_bucket_gens)) \ 2313 x(snapshot_trees, 15, 0, \ 2314 BIT_ULL(KEY_TYPE_snapshot_tree)) \ 2315 x(deleted_inodes, 16, BTREE_ID_SNAPSHOT_FIELD, \ 2316 BIT_ULL(KEY_TYPE_set)) \ 2317 x(logged_ops, 17, 0, \ 2318 BIT_ULL(KEY_TYPE_logged_op_truncate)| \ 2319 BIT_ULL(KEY_TYPE_logged_op_finsert)) \ 2320 x(rebalance_work, 18, BTREE_ID_SNAPSHOT_FIELD, \ 2321 BIT_ULL(KEY_TYPE_set)|BIT_ULL(KEY_TYPE_cookie)) 2322 2323 enum btree_id { 2324 #define x(name, nr, ...) BTREE_ID_##name = nr, 2325 BCH_BTREE_IDS() 2326 #undef x 2327 BTREE_ID_NR 2328 }; 2329 2330 #define BTREE_MAX_DEPTH 4U 2331 2332 /* Btree nodes */ 2333 2334 /* 2335 * Btree nodes 2336 * 2337 * On disk a btree node is a list/log of these; within each set the keys are 2338 * sorted 2339 */ 2340 struct bset { 2341 __le64 seq; 2342 2343 /* 2344 * Highest journal entry this bset contains keys for. 2345 * If on recovery we don't see that journal entry, this bset is ignored: 2346 * this allows us to preserve the order of all index updates after a 2347 * crash, since the journal records a total order of all index updates 2348 * and anything that didn't make it to the journal doesn't get used. 2349 */ 2350 __le64 journal_seq; 2351 2352 __le32 flags; 2353 __le16 version; 2354 __le16 u64s; /* count of d[] in u64s */ 2355 2356 struct bkey_packed start[0]; 2357 __u64 _data[]; 2358 } __packed __aligned(8); 2359 2360 LE32_BITMASK(BSET_CSUM_TYPE, struct bset, flags, 0, 4); 2361 2362 LE32_BITMASK(BSET_BIG_ENDIAN, struct bset, flags, 4, 5); 2363 LE32_BITMASK(BSET_SEPARATE_WHITEOUTS, 2364 struct bset, flags, 5, 6); 2365 2366 /* Sector offset within the btree node: */ 2367 LE32_BITMASK(BSET_OFFSET, struct bset, flags, 16, 32); 2368 2369 struct btree_node { 2370 struct bch_csum csum; 2371 __le64 magic; 2372 2373 /* this flags field is encrypted, unlike bset->flags: */ 2374 __le64 flags; 2375 2376 /* Closed interval: */ 2377 struct bpos min_key; 2378 struct bpos max_key; 2379 struct bch_extent_ptr _ptr; /* not used anymore */ 2380 struct bkey_format format; 2381 2382 union { 2383 struct bset keys; 2384 struct { 2385 __u8 pad[22]; 2386 __le16 u64s; 2387 __u64 _data[0]; 2388 2389 }; 2390 }; 2391 } __packed __aligned(8); 2392 2393 LE64_BITMASK(BTREE_NODE_ID_LO, struct btree_node, flags, 0, 4); 2394 LE64_BITMASK(BTREE_NODE_LEVEL, struct btree_node, flags, 4, 8); 2395 LE64_BITMASK(BTREE_NODE_NEW_EXTENT_OVERWRITE, 2396 struct btree_node, flags, 8, 9); 2397 LE64_BITMASK(BTREE_NODE_ID_HI, struct btree_node, flags, 9, 25); 2398 /* 25-32 unused */ 2399 LE64_BITMASK(BTREE_NODE_SEQ, struct btree_node, flags, 32, 64); 2400 2401 static inline __u64 BTREE_NODE_ID(struct btree_node *n) 2402 { 2403 return BTREE_NODE_ID_LO(n) | (BTREE_NODE_ID_HI(n) << 4); 2404 } 2405 2406 static inline void SET_BTREE_NODE_ID(struct btree_node *n, __u64 v) 2407 { 2408 SET_BTREE_NODE_ID_LO(n, v); 2409 SET_BTREE_NODE_ID_HI(n, v >> 4); 2410 } 2411 2412 struct btree_node_entry { 2413 struct bch_csum csum; 2414 2415 union { 2416 struct bset keys; 2417 struct { 2418 __u8 pad[22]; 2419 __le16 u64s; 2420 __u64 _data[0]; 2421 }; 2422 }; 2423 } __packed __aligned(8); 2424 2425 #endif /* _BCACHEFS_FORMAT_H */ 2426