git.sesse.net Git - bcachefs-tools-debian/blob - libbcachefs/bcachefs_format.h

   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 #define BITMASK(name, type, field, offset, end)                         \
  82 static const unsigned   name##_OFFSET = offset;                         \
  83 static const unsigned   name##_BITS = (end - offset);                   \
  84                                                                         \
  85 static inline __u64 name(const type *k)                                 \
  86 {                                                                       \
  87         return (k->field >> offset) & ~(~0ULL << (end - offset));       \
  88 }                                                                       \
  89                                                                         \
  90 static inline void SET_##name(type *k, __u64 v)                         \
  91 {                                                                       \
  92         k->field &= ~(~(~0ULL << (end - offset)) << offset);            \
  93         k->field |= (v & ~(~0ULL << (end - offset))) << offset;         \
  94 }
  95
  96 #define LE_BITMASK(_bits, name, type, field, offset, end)               \
  97 static const unsigned   name##_OFFSET = offset;                         \
  98 static const unsigned   name##_BITS = (end - offset);                   \
  99 static const __u##_bits name##_MAX = (1ULL << (end - offset)) - 1;      \
 100                                                                         \
 101 static inline __u64 name(const type *k)                                 \
 102 {                                                                       \
 103         return (__le##_bits##_to_cpu(k->field) >> offset) &             \
 104                 ~(~0ULL << (end - offset));                             \
 105 }                                                                       \
 106                                                                         \
 107 static inline void SET_##name(type *k, __u64 v)                         \
 108 {                                                                       \
 109         __u##_bits new = __le##_bits##_to_cpu(k->field);                \
 110                                                                         \
 111         new &= ~(~(~0ULL << (end - offset)) << offset);                 \
 112         new |= (v & ~(~0ULL << (end - offset))) << offset;              \
 113         k->field = __cpu_to_le##_bits(new);                             \
 114 }
 115
 116 #define LE16_BITMASK(n, t, f, o, e)     LE_BITMASK(16, n, t, f, o, e)
 117 #define LE32_BITMASK(n, t, f, o, e)     LE_BITMASK(32, n, t, f, o, e)
 118 #define LE64_BITMASK(n, t, f, o, e)     LE_BITMASK(64, n, t, f, o, e)
 119
 120 struct bkey_format {
 121         __u8            key_u64s;
 122         __u8            nr_fields;
 123         /* One unused slot for now: */
 124         __u8            bits_per_field[6];
 125         __le64          field_offset[6];
 126 };
 127
 128 /* Btree keys - all units are in sectors */
 129
 130 struct bpos {
 131         /*
 132          * Word order matches machine byte order - btree code treats a bpos as a
 133          * single large integer, for search/comparison purposes
 134          *
 135          * Note that wherever a bpos is embedded in another on disk data
 136          * structure, it has to be byte swabbed when reading in metadata that
 137          * wasn't written in native endian order:
 138          */
 139 #if __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__
 140         __u32           snapshot;
 141         __u64           offset;
 142         __u64           inode;
 143 #elif __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__
 144         __u64           inode;
 145         __u64           offset;         /* Points to end of extent - sectors */
 146         __u32           snapshot;
 147 #else
 148 #error edit for your odd byteorder.
 149 #endif
 150 } __packed __aligned(4);
 151
 152 #define KEY_INODE_MAX                   ((__u64)~0ULL)
 153 #define KEY_OFFSET_MAX                  ((__u64)~0ULL)
 154 #define KEY_SNAPSHOT_MAX                ((__u32)~0U)
 155 #define KEY_SIZE_MAX                    ((__u32)~0U)
 156
 157 static inline struct bpos SPOS(__u64 inode, __u64 offset, __u32 snapshot)
 158 {
 159         return (struct bpos) {
 160                 .inode          = inode,
 161                 .offset         = offset,
 162                 .snapshot       = snapshot,
 163         };
 164 }
 165
 166 #define POS_MIN                         SPOS(0, 0, 0)
 167 #define POS_MAX                         SPOS(KEY_INODE_MAX, KEY_OFFSET_MAX, 0)
 168 #define SPOS_MAX                        SPOS(KEY_INODE_MAX, KEY_OFFSET_MAX, KEY_SNAPSHOT_MAX)
 169 #define POS(_inode, _offset)            SPOS(_inode, _offset, 0)
 170
 171 /* Empty placeholder struct, for container_of() */
 172 struct bch_val {
 173         __u64           __nothing[0];
 174 };
 175
 176 struct bversion {
 177 #if __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__
 178         __u64           lo;
 179         __u32           hi;
 180 #elif __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__
 181         __u32           hi;
 182         __u64           lo;
 183 #endif
 184 } __packed __aligned(4);
 185
 186 struct bkey {
 187         /* Size of combined key and value, in u64s */
 188         __u8            u64s;
 189
 190         /* Format of key (0 for format local to btree node) */
 191 #if defined(__LITTLE_ENDIAN_BITFIELD)
 192         __u8            format:7,
 193                         needs_whiteout:1;
 194 #elif defined (__BIG_ENDIAN_BITFIELD)
 195         __u8            needs_whiteout:1,
 196                         format:7;
 197 #else
 198 #error edit for your odd byteorder.
 199 #endif
 200
 201         /* Type of the value */
 202         __u8            type;
 203
 204 #if __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__
 205         __u8            pad[1];
 206
 207         struct bversion version;
 208         __u32           size;           /* extent size, in sectors */
 209         struct bpos     p;
 210 #elif __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__
 211         struct bpos     p;
 212         __u32           size;           /* extent size, in sectors */
 213         struct bversion version;
 214
 215         __u8            pad[1];
 216 #endif
 217 } __packed __aligned(8);
 218
 219 struct bkey_packed {
 220         __u64           _data[0];
 221
 222         /* Size of combined key and value, in u64s */
 223         __u8            u64s;
 224
 225         /* Format of key (0 for format local to btree node) */
 226
 227         /*
 228          * XXX: next incompat on disk format change, switch format and
 229          * needs_whiteout - bkey_packed() will be cheaper if format is the high
 230          * bits of the bitfield
 231          */
 232 #if defined(__LITTLE_ENDIAN_BITFIELD)
 233         __u8            format:7,
 234                         needs_whiteout:1;
 235 #elif defined (__BIG_ENDIAN_BITFIELD)
 236         __u8            needs_whiteout:1,
 237                         format:7;
 238 #endif
 239
 240         /* Type of the value */
 241         __u8            type;
 242         __u8            key_start[0];
 243
 244         /*
 245          * We copy bkeys with struct assignment in various places, and while
 246          * that shouldn't be done with packed bkeys we can't disallow it in C,
 247          * and it's legal to cast a bkey to a bkey_packed  - so padding it out
 248          * to the same size as struct bkey should hopefully be safest.
 249          */
 250         __u8            pad[sizeof(struct bkey) - 3];
 251 } __packed __aligned(8);
 252
 253 #define BKEY_U64s                       (sizeof(struct bkey) / sizeof(__u64))
 254 #define BKEY_U64s_MAX                   U8_MAX
 255 #define BKEY_VAL_U64s_MAX               (BKEY_U64s_MAX - BKEY_U64s)
 256
 257 #define KEY_PACKED_BITS_START           24
 258
 259 #define KEY_FORMAT_LOCAL_BTREE          0
 260 #define KEY_FORMAT_CURRENT              1
 261
 262 enum bch_bkey_fields {
 263         BKEY_FIELD_INODE,
 264         BKEY_FIELD_OFFSET,
 265         BKEY_FIELD_SNAPSHOT,
 266         BKEY_FIELD_SIZE,
 267         BKEY_FIELD_VERSION_HI,
 268         BKEY_FIELD_VERSION_LO,
 269         BKEY_NR_FIELDS,
 270 };
 271
 272 #define bkey_format_field(name, field)                                  \
 273         [BKEY_FIELD_##name] = (sizeof(((struct bkey *) NULL)->field) * 8)
 274
 275 #define BKEY_FORMAT_CURRENT                                             \
 276 ((struct bkey_format) {                                                 \
 277         .key_u64s       = BKEY_U64s,                                    \
 278         .nr_fields      = BKEY_NR_FIELDS,                               \
 279         .bits_per_field = {                                             \
 280                 bkey_format_field(INODE,        p.inode),               \
 281                 bkey_format_field(OFFSET,       p.offset),              \
 282                 bkey_format_field(SNAPSHOT,     p.snapshot),            \
 283                 bkey_format_field(SIZE,         size),                  \
 284                 bkey_format_field(VERSION_HI,   version.hi),            \
 285                 bkey_format_field(VERSION_LO,   version.lo),            \
 286         },                                                              \
 287 })
 288
 289 /* bkey with inline value */
 290 struct bkey_i {
 291         __u64                   _data[0];
 292
 293         union {
 294         struct {
 295                 /* Size of combined key and value, in u64s */
 296                 __u8            u64s;
 297         };
 298         struct {
 299                 struct bkey     k;
 300                 struct bch_val  v;
 301         };
 302         };
 303 };
 304
 305 #define KEY(_inode, _offset, _size)                                     \
 306 ((struct bkey) {                                                        \
 307         .u64s           = BKEY_U64s,                                    \
 308         .format         = KEY_FORMAT_CURRENT,                           \
 309         .p              = POS(_inode, _offset),                         \
 310         .size           = _size,                                        \
 311 })
 312
 313 static inline void bkey_init(struct bkey *k)
 314 {
 315         *k = KEY(0, 0, 0);
 316 }
 317
 318 #define bkey_bytes(_k)          ((_k)->u64s * sizeof(__u64))
 319
 320 #define __BKEY_PADDED(key, pad)                                 \
 321         struct { struct bkey_i key; __u64 key ## _pad[pad]; }
 322
 323 /*
 324  * - DELETED keys are used internally to mark keys that should be ignored but
 325  *   override keys in composition order.  Their version number is ignored.
 326  *
 327  * - DISCARDED keys indicate that the data is all 0s because it has been
 328  *   discarded. DISCARDs may have a version; if the version is nonzero the key
 329  *   will be persistent, otherwise the key will be dropped whenever the btree
 330  *   node is rewritten (like DELETED keys).
 331  *
 332  * - ERROR: any read of the data returns a read error, as the data was lost due
 333  *   to a failing device. Like DISCARDED keys, they can be removed (overridden)
 334  *   by new writes or cluster-wide GC. Node repair can also overwrite them with
 335  *   the same or a more recent version number, but not with an older version
 336  *   number.
 337  *
 338  * - WHITEOUT: for hash table btrees
 339  */
 340 #define BCH_BKEY_TYPES()                                \
 341         x(deleted,              0)                      \
 342         x(whiteout,             1)                      \
 343         x(error,                2)                      \
 344         x(cookie,               3)                      \
 345         x(hash_whiteout,        4)                      \
 346         x(btree_ptr,            5)                      \
 347         x(extent,               6)                      \
 348         x(reservation,          7)                      \
 349         x(inode,                8)                      \
 350         x(inode_generation,     9)                      \
 351         x(dirent,               10)                     \
 352         x(xattr,                11)                     \
 353         x(alloc,                12)                     \
 354         x(quota,                13)                     \
 355         x(stripe,               14)                     \
 356         x(reflink_p,            15)                     \
 357         x(reflink_v,            16)                     \
 358         x(inline_data,          17)                     \
 359         x(btree_ptr_v2,         18)                     \
 360         x(indirect_inline_data, 19)                     \
 361         x(alloc_v2,             20)                     \
 362         x(subvolume,            21)                     \
 363         x(snapshot,             22)                     \
 364         x(inode_v2,             23)                     \
 365         x(alloc_v3,             24)                     \
 366         x(set,                  25)                     \
 367         x(lru,                  26)                     \
 368         x(alloc_v4,             27)                     \
 369         x(backpointer,          28)                     \
 370         x(inode_v3,             29)                     \
 371         x(bucket_gens,          30)
 372
 373 enum bch_bkey_type {
 374 #define x(name, nr) KEY_TYPE_##name     = nr,
 375         BCH_BKEY_TYPES()
 376 #undef x
 377         KEY_TYPE_MAX,
 378 };
 379
 380 struct bch_deleted {
 381         struct bch_val          v;
 382 };
 383
 384 struct bch_whiteout {
 385         struct bch_val          v;
 386 };
 387
 388 struct bch_error {
 389         struct bch_val          v;
 390 };
 391
 392 struct bch_cookie {
 393         struct bch_val          v;
 394         __le64                  cookie;
 395 };
 396
 397 struct bch_hash_whiteout {
 398         struct bch_val          v;
 399 };
 400
 401 struct bch_set {
 402         struct bch_val          v;
 403 };
 404
 405 /* Extents */
 406
 407 /*
 408  * In extent bkeys, the value is a list of pointers (bch_extent_ptr), optionally
 409  * preceded by checksum/compression information (bch_extent_crc32 or
 410  * bch_extent_crc64).
 411  *
 412  * One major determining factor in the format of extents is how we handle and
 413  * represent extents that have been partially overwritten and thus trimmed:
 414  *
 415  * If an extent is not checksummed or compressed, when the extent is trimmed we
 416  * don't have to remember the extent we originally allocated and wrote: we can
 417  * merely adjust ptr->offset to point to the start of the data that is currently
 418  * live. The size field in struct bkey records the current (live) size of the
 419  * extent, and is also used to mean "size of region on disk that we point to" in
 420  * this case.
 421  *
 422  * Thus an extent that is not checksummed or compressed will consist only of a
 423  * list of bch_extent_ptrs, with none of the fields in
 424  * bch_extent_crc32/bch_extent_crc64.
 425  *
 426  * When an extent is checksummed or compressed, it's not possible to read only
 427  * the data that is currently live: we have to read the entire extent that was
 428  * originally written, and then return only the part of the extent that is
 429  * currently live.
 430  *
 431  * Thus, in addition to the current size of the extent in struct bkey, we need
 432  * to store the size of the originally allocated space - this is the
 433  * compressed_size and uncompressed_size fields in bch_extent_crc32/64. Also,
 434  * when the extent is trimmed, instead of modifying the offset field of the
 435  * pointer, we keep a second smaller offset field - "offset into the original
 436  * extent of the currently live region".
 437  *
 438  * The other major determining factor is replication and data migration:
 439  *
 440  * Each pointer may have its own bch_extent_crc32/64. When doing a replicated
 441  * write, we will initially write all the replicas in the same format, with the
 442  * same checksum type and compression format - however, when copygc runs later (or
 443  * tiering/cache promotion, anything that moves data), it is not in general
 444  * going to rewrite all the pointers at once - one of the replicas may be in a
 445  * bucket on one device that has very little fragmentation while another lives
 446  * in a bucket that has become heavily fragmented, and thus is being rewritten
 447  * sooner than the rest.
 448  *
 449  * Thus it will only move a subset of the pointers (or in the case of
 450  * tiering/cache promotion perhaps add a single pointer without dropping any
 451  * current pointers), and if the extent has been partially overwritten it must
 452  * write only the currently live portion (or copygc would not be able to reduce
 453  * fragmentation!) - which necessitates a different bch_extent_crc format for
 454  * the new pointer.
 455  *
 456  * But in the interests of space efficiency, we don't want to store one
 457  * bch_extent_crc for each pointer if we don't have to.
 458  *
 459  * Thus, a bch_extent consists of bch_extent_crc32s, bch_extent_crc64s, and
 460  * bch_extent_ptrs appended arbitrarily one after the other. We determine the
 461  * type of a given entry with a scheme similar to utf8 (except we're encoding a
 462  * type, not a size), encoding the type in the position of the first set bit:
 463  *
 464  * bch_extent_crc32     - 0b1
 465  * bch_extent_ptr       - 0b10
 466  * bch_extent_crc64     - 0b100
 467  *
 468  * We do it this way because bch_extent_crc32 is _very_ constrained on bits (and
 469  * bch_extent_crc64 is the least constrained).
 470  *
 471  * Then, each bch_extent_crc32/64 applies to the pointers that follow after it,
 472  * until the next bch_extent_crc32/64.
 473  *
 474  * If there are no bch_extent_crcs preceding a bch_extent_ptr, then that pointer
 475  * is neither checksummed nor compressed.
 476  */
 477
 478 /* 128 bits, sufficient for cryptographic MACs: */
 479 struct bch_csum {
 480         __le64                  lo;
 481         __le64                  hi;
 482 } __packed __aligned(8);
 483
 484 #define BCH_EXTENT_ENTRY_TYPES()                \
 485         x(ptr,                  0)              \
 486         x(crc32,                1)              \
 487         x(crc64,                2)              \
 488         x(crc128,               3)              \
 489         x(stripe_ptr,           4)
 490 #define BCH_EXTENT_ENTRY_MAX    5
 491
 492 enum bch_extent_entry_type {
 493 #define x(f, n) BCH_EXTENT_ENTRY_##f = n,
 494         BCH_EXTENT_ENTRY_TYPES()
 495 #undef x
 496 };
 497
 498 /* Compressed/uncompressed size are stored biased by 1: */
 499 struct bch_extent_crc32 {
 500 #if defined(__LITTLE_ENDIAN_BITFIELD)
 501         __u32                   type:2,
 502                                 _compressed_size:7,
 503                                 _uncompressed_size:7,
 504                                 offset:7,
 505                                 _unused:1,
 506                                 csum_type:4,
 507                                 compression_type:4;
 508         __u32                   csum;
 509 #elif defined (__BIG_ENDIAN_BITFIELD)
 510         __u32                   csum;
 511         __u32                   compression_type:4,
 512                                 csum_type:4,
 513                                 _unused:1,
 514                                 offset:7,
 515                                 _uncompressed_size:7,
 516                                 _compressed_size:7,
 517                                 type:2;
 518 #endif
 519 } __packed __aligned(8);
 520
 521 #define CRC32_SIZE_MAX          (1U << 7)
 522 #define CRC32_NONCE_MAX         0
 523
 524 struct bch_extent_crc64 {
 525 #if defined(__LITTLE_ENDIAN_BITFIELD)
 526         __u64                   type:3,
 527                                 _compressed_size:9,
 528                                 _uncompressed_size:9,
 529                                 offset:9,
 530                                 nonce:10,
 531                                 csum_type:4,
 532                                 compression_type:4,
 533                                 csum_hi:16;
 534 #elif defined (__BIG_ENDIAN_BITFIELD)
 535         __u64                   csum_hi:16,
 536                                 compression_type:4,
 537                                 csum_type:4,
 538                                 nonce:10,
 539                                 offset:9,
 540                                 _uncompressed_size:9,
 541                                 _compressed_size:9,
 542                                 type:3;
 543 #endif
 544         __u64                   csum_lo;
 545 } __packed __aligned(8);
 546
 547 #define CRC64_SIZE_MAX          (1U << 9)
 548 #define CRC64_NONCE_MAX         ((1U << 10) - 1)
 549
 550 struct bch_extent_crc128 {
 551 #if defined(__LITTLE_ENDIAN_BITFIELD)
 552         __u64                   type:4,
 553                                 _compressed_size:13,
 554                                 _uncompressed_size:13,
 555                                 offset:13,
 556                                 nonce:13,
 557                                 csum_type:4,
 558                                 compression_type:4;
 559 #elif defined (__BIG_ENDIAN_BITFIELD)
 560         __u64                   compression_type:4,
 561                                 csum_type:4,
 562                                 nonce:13,
 563                                 offset:13,
 564                                 _uncompressed_size:13,
 565                                 _compressed_size:13,
 566                                 type:4;
 567 #endif
 568         struct bch_csum         csum;
 569 } __packed __aligned(8);
 570
 571 #define CRC128_SIZE_MAX         (1U << 13)
 572 #define CRC128_NONCE_MAX        ((1U << 13) - 1)
 573
 574 /*
 575  * @reservation - pointer hasn't been written to, just reserved
 576  */
 577 struct bch_extent_ptr {
 578 #if defined(__LITTLE_ENDIAN_BITFIELD)
 579         __u64                   type:1,
 580                                 cached:1,
 581                                 unused:1,
 582                                 unwritten:1,
 583                                 offset:44, /* 8 petabytes */
 584                                 dev:8,
 585                                 gen:8;
 586 #elif defined (__BIG_ENDIAN_BITFIELD)
 587         __u64                   gen:8,
 588                                 dev:8,
 589                                 offset:44,
 590                                 unwritten:1,
 591                                 unused:1,
 592                                 cached:1,
 593                                 type:1;
 594 #endif
 595 } __packed __aligned(8);
 596
 597 struct bch_extent_stripe_ptr {
 598 #if defined(__LITTLE_ENDIAN_BITFIELD)
 599         __u64                   type:5,
 600                                 block:8,
 601                                 redundancy:4,
 602                                 idx:47;
 603 #elif defined (__BIG_ENDIAN_BITFIELD)
 604         __u64                   idx:47,
 605                                 redundancy:4,
 606                                 block:8,
 607                                 type:5;
 608 #endif
 609 };
 610
 611 struct bch_extent_reservation {
 612 #if defined(__LITTLE_ENDIAN_BITFIELD)
 613         __u64                   type:6,
 614                                 unused:22,
 615                                 replicas:4,
 616                                 generation:32;
 617 #elif defined (__BIG_ENDIAN_BITFIELD)
 618         __u64                   generation:32,
 619                                 replicas:4,
 620                                 unused:22,
 621                                 type:6;
 622 #endif
 623 };
 624
 625 union bch_extent_entry {
 626 #if __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__ ||  __BITS_PER_LONG == 64
 627         unsigned long                   type;
 628 #elif __BITS_PER_LONG == 32
 629         struct {
 630                 unsigned long           pad;
 631                 unsigned long           type;
 632         };
 633 #else
 634 #error edit for your odd byteorder.
 635 #endif
 636
 637 #define x(f, n) struct bch_extent_##f   f;
 638         BCH_EXTENT_ENTRY_TYPES()
 639 #undef x
 640 };
 641
 642 struct bch_btree_ptr {
 643         struct bch_val          v;
 644
 645         __u64                   _data[0];
 646         struct bch_extent_ptr   start[];
 647 } __packed __aligned(8);
 648
 649 struct bch_btree_ptr_v2 {
 650         struct bch_val          v;
 651
 652         __u64                   mem_ptr;
 653         __le64                  seq;
 654         __le16                  sectors_written;
 655         __le16                  flags;
 656         struct bpos             min_key;
 657         __u64                   _data[0];
 658         struct bch_extent_ptr   start[];
 659 } __packed __aligned(8);
 660
 661 LE16_BITMASK(BTREE_PTR_RANGE_UPDATED,   struct bch_btree_ptr_v2, flags, 0, 1);
 662
 663 struct bch_extent {
 664         struct bch_val          v;
 665
 666         __u64                   _data[0];
 667         union bch_extent_entry  start[];
 668 } __packed __aligned(8);
 669
 670 struct bch_reservation {
 671         struct bch_val          v;
 672
 673         __le32                  generation;
 674         __u8                    nr_replicas;
 675         __u8                    pad[3];
 676 } __packed __aligned(8);
 677
 678 /* Maximum size (in u64s) a single pointer could be: */
 679 #define BKEY_EXTENT_PTR_U64s_MAX\
 680         ((sizeof(struct bch_extent_crc128) +                    \
 681           sizeof(struct bch_extent_ptr)) / sizeof(u64))
 682
 683 /* Maximum possible size of an entire extent value: */
 684 #define BKEY_EXTENT_VAL_U64s_MAX                                \
 685         (1 + BKEY_EXTENT_PTR_U64s_MAX * (BCH_REPLICAS_MAX + 1))
 686
 687 /* * Maximum possible size of an entire extent, key + value: */
 688 #define BKEY_EXTENT_U64s_MAX            (BKEY_U64s + BKEY_EXTENT_VAL_U64s_MAX)
 689
 690 /* Btree pointers don't carry around checksums: */
 691 #define BKEY_BTREE_PTR_VAL_U64s_MAX                             \
 692         ((sizeof(struct bch_btree_ptr_v2) +                     \
 693           sizeof(struct bch_extent_ptr) * BCH_REPLICAS_MAX) / sizeof(u64))
 694 #define BKEY_BTREE_PTR_U64s_MAX                                 \
 695         (BKEY_U64s + BKEY_BTREE_PTR_VAL_U64s_MAX)
 696
 697 /* Inodes */
 698
 699 #define BLOCKDEV_INODE_MAX      4096
 700
 701 #define BCACHEFS_ROOT_INO       4096
 702
 703 struct bch_inode {
 704         struct bch_val          v;
 705
 706         __le64                  bi_hash_seed;
 707         __le32                  bi_flags;
 708         __le16                  bi_mode;
 709         __u8                    fields[0];
 710 } __packed __aligned(8);
 711
 712 struct bch_inode_v2 {
 713         struct bch_val          v;
 714
 715         __le64                  bi_journal_seq;
 716         __le64                  bi_hash_seed;
 717         __le64                  bi_flags;
 718         __le16                  bi_mode;
 719         __u8                    fields[0];
 720 } __packed __aligned(8);
 721
 722 struct bch_inode_v3 {
 723         struct bch_val          v;
 724
 725         __le64                  bi_journal_seq;
 726         __le64                  bi_hash_seed;
 727         __le64                  bi_flags;
 728         __le64                  bi_sectors;
 729         __le64                  bi_size;
 730         __le64                  bi_version;
 731         __u8                    fields[0];
 732 } __packed __aligned(8);
 733
 734 #define INODEv3_FIELDS_START_INITIAL    6
 735 #define INODEv3_FIELDS_START_CUR        (offsetof(struct bch_inode_v3, fields) / sizeof(u64))
 736
 737 struct bch_inode_generation {
 738         struct bch_val          v;
 739
 740         __le32                  bi_generation;
 741         __le32                  pad;
 742 } __packed __aligned(8);
 743
 744 /*
 745  * bi_subvol and bi_parent_subvol are only set for subvolume roots:
 746  */
 747
 748 #define BCH_INODE_FIELDS_v2()                   \
 749         x(bi_atime,                     96)     \
 750         x(bi_ctime,                     96)     \
 751         x(bi_mtime,                     96)     \
 752         x(bi_otime,                     96)     \
 753         x(bi_size,                      64)     \
 754         x(bi_sectors,                   64)     \
 755         x(bi_uid,                       32)     \
 756         x(bi_gid,                       32)     \
 757         x(bi_nlink,                     32)     \
 758         x(bi_generation,                32)     \
 759         x(bi_dev,                       32)     \
 760         x(bi_data_checksum,             8)      \
 761         x(bi_compression,               8)      \
 762         x(bi_project,                   32)     \
 763         x(bi_background_compression,    8)      \
 764         x(bi_data_replicas,             8)      \
 765         x(bi_promote_target,            16)     \
 766         x(bi_foreground_target,         16)     \
 767         x(bi_background_target,         16)     \
 768         x(bi_erasure_code,              16)     \
 769         x(bi_fields_set,                16)     \
 770         x(bi_dir,                       64)     \
 771         x(bi_dir_offset,                64)     \
 772         x(bi_subvol,                    32)     \
 773         x(bi_parent_subvol,             32)
 774
 775 #define BCH_INODE_FIELDS_v3()                   \
 776         x(bi_atime,                     96)     \
 777         x(bi_ctime,                     96)     \
 778         x(bi_mtime,                     96)     \
 779         x(bi_otime,                     96)     \
 780         x(bi_uid,                       32)     \
 781         x(bi_gid,                       32)     \
 782         x(bi_nlink,                     32)     \
 783         x(bi_generation,                32)     \
 784         x(bi_dev,                       32)     \
 785         x(bi_data_checksum,             8)      \
 786         x(bi_compression,               8)      \
 787         x(bi_project,                   32)     \
 788         x(bi_background_compression,    8)      \
 789         x(bi_data_replicas,             8)      \
 790         x(bi_promote_target,            16)     \
 791         x(bi_foreground_target,         16)     \
 792         x(bi_background_target,         16)     \
 793         x(bi_erasure_code,              16)     \
 794         x(bi_fields_set,                16)     \
 795         x(bi_dir,                       64)     \
 796         x(bi_dir_offset,                64)     \
 797         x(bi_subvol,                    32)     \
 798         x(bi_parent_subvol,             32)     \
 799         x(bi_nocow,                     8)
 800
 801 /* subset of BCH_INODE_FIELDS */
 802 #define BCH_INODE_OPTS()                        \
 803         x(data_checksum,                8)      \
 804         x(compression,                  8)      \
 805         x(project,                      32)     \
 806         x(background_compression,       8)      \
 807         x(data_replicas,                8)      \
 808         x(promote_target,               16)     \
 809         x(foreground_target,            16)     \
 810         x(background_target,            16)     \
 811         x(erasure_code,                 16)     \
 812         x(nocow,                        8)
 813
 814 enum inode_opt_id {
 815 #define x(name, ...)                            \
 816         Inode_opt_##name,
 817         BCH_INODE_OPTS()
 818 #undef  x
 819         Inode_opt_nr,
 820 };
 821
 822 enum {
 823         /*
 824          * User flags (get/settable with FS_IOC_*FLAGS, correspond to FS_*_FL
 825          * flags)
 826          */
 827         __BCH_INODE_SYNC                = 0,
 828         __BCH_INODE_IMMUTABLE           = 1,
 829         __BCH_INODE_APPEND              = 2,
 830         __BCH_INODE_NODUMP              = 3,
 831         __BCH_INODE_NOATIME             = 4,
 832
 833         __BCH_INODE_I_SIZE_DIRTY        = 5,
 834         __BCH_INODE_I_SECTORS_DIRTY     = 6,
 835         __BCH_INODE_UNLINKED            = 7,
 836         __BCH_INODE_BACKPTR_UNTRUSTED   = 8,
 837
 838         /* bits 20+ reserved for packed fields below: */
 839 };
 840
 841 #define BCH_INODE_SYNC          (1 << __BCH_INODE_SYNC)
 842 #define BCH_INODE_IMMUTABLE     (1 << __BCH_INODE_IMMUTABLE)
 843 #define BCH_INODE_APPEND        (1 << __BCH_INODE_APPEND)
 844 #define BCH_INODE_NODUMP        (1 << __BCH_INODE_NODUMP)
 845 #define BCH_INODE_NOATIME       (1 << __BCH_INODE_NOATIME)
 846 #define BCH_INODE_I_SIZE_DIRTY  (1 << __BCH_INODE_I_SIZE_DIRTY)
 847 #define BCH_INODE_I_SECTORS_DIRTY (1 << __BCH_INODE_I_SECTORS_DIRTY)
 848 #define BCH_INODE_UNLINKED      (1 << __BCH_INODE_UNLINKED)
 849 #define BCH_INODE_BACKPTR_UNTRUSTED (1 << __BCH_INODE_BACKPTR_UNTRUSTED)
 850
 851 LE32_BITMASK(INODE_STR_HASH,    struct bch_inode, bi_flags, 20, 24);
 852 LE32_BITMASK(INODE_NR_FIELDS,   struct bch_inode, bi_flags, 24, 31);
 853 LE32_BITMASK(INODE_NEW_VARINT,  struct bch_inode, bi_flags, 31, 32);
 854
 855 LE64_BITMASK(INODEv2_STR_HASH,  struct bch_inode_v2, bi_flags, 20, 24);
 856 LE64_BITMASK(INODEv2_NR_FIELDS, struct bch_inode_v2, bi_flags, 24, 31);
 857
 858 LE64_BITMASK(INODEv3_STR_HASH,  struct bch_inode_v3, bi_flags, 20, 24);
 859 LE64_BITMASK(INODEv3_NR_FIELDS, struct bch_inode_v3, bi_flags, 24, 31);
 860
 861 LE64_BITMASK(INODEv3_FIELDS_START,
 862                                 struct bch_inode_v3, bi_flags, 31, 36);
 863 LE64_BITMASK(INODEv3_MODE,      struct bch_inode_v3, bi_flags, 36, 52);
 864
 865 /* Dirents */
 866
 867 /*
 868  * Dirents (and xattrs) have to implement string lookups; since our b-tree
 869  * doesn't support arbitrary length strings for the key, we instead index by a
 870  * 64 bit hash (currently truncated sha1) of the string, stored in the offset
 871  * field of the key - using linear probing to resolve hash collisions. This also
 872  * provides us with the readdir cookie posix requires.
 873  *
 874  * Linear probing requires us to use whiteouts for deletions, in the event of a
 875  * collision:
 876  */
 877
 878 struct bch_dirent {
 879         struct bch_val          v;
 880
 881         /* Target inode number: */
 882         union {
 883         __le64                  d_inum;
 884         struct {                /* DT_SUBVOL */
 885         __le32                  d_child_subvol;
 886         __le32                  d_parent_subvol;
 887         };
 888         };
 889
 890         /*
 891          * Copy of mode bits 12-15 from the target inode - so userspace can get
 892          * the filetype without having to do a stat()
 893          */
 894         __u8                    d_type;
 895
 896         __u8                    d_name[];
 897 } __packed __aligned(8);
 898
 899 #define DT_SUBVOL       16
 900 #define BCH_DT_MAX      17
 901
 902 #define BCH_NAME_MAX    ((unsigned) (U8_MAX * sizeof(u64) -             \
 903                          sizeof(struct bkey) -                          \
 904                          offsetof(struct bch_dirent, d_name)))
 905
 906 /* Xattrs */
 907
 908 #define KEY_TYPE_XATTR_INDEX_USER                       0
 909 #define KEY_TYPE_XATTR_INDEX_POSIX_ACL_ACCESS   1
 910 #define KEY_TYPE_XATTR_INDEX_POSIX_ACL_DEFAULT  2
 911 #define KEY_TYPE_XATTR_INDEX_TRUSTED                    3
 912 #define KEY_TYPE_XATTR_INDEX_SECURITY           4
 913
 914 struct bch_xattr {
 915         struct bch_val          v;
 916         __u8                    x_type;
 917         __u8                    x_name_len;
 918         __le16                  x_val_len;
 919         __u8                    x_name[];
 920 } __packed __aligned(8);
 921
 922 /* Bucket/allocation information: */
 923
 924 struct bch_alloc {
 925         struct bch_val          v;
 926         __u8                    fields;
 927         __u8                    gen;
 928         __u8                    data[];
 929 } __packed __aligned(8);
 930
 931 #define BCH_ALLOC_FIELDS_V1()                   \
 932         x(read_time,            16)             \
 933         x(write_time,           16)             \
 934         x(data_type,            8)              \
 935         x(dirty_sectors,        16)             \
 936         x(cached_sectors,       16)             \
 937         x(oldest_gen,           8)              \
 938         x(stripe,               32)             \
 939         x(stripe_redundancy,    8)
 940
 941 enum {
 942 #define x(name, _bits) BCH_ALLOC_FIELD_V1_##name,
 943         BCH_ALLOC_FIELDS_V1()
 944 #undef x
 945 };
 946
 947 struct bch_alloc_v2 {
 948         struct bch_val          v;
 949         __u8                    nr_fields;
 950         __u8                    gen;
 951         __u8                    oldest_gen;
 952         __u8                    data_type;
 953         __u8                    data[];
 954 } __packed __aligned(8);
 955
 956 #define BCH_ALLOC_FIELDS_V2()                   \
 957         x(read_time,            64)             \
 958         x(write_time,           64)             \
 959         x(dirty_sectors,        32)             \
 960         x(cached_sectors,       32)             \
 961         x(stripe,               32)             \
 962         x(stripe_redundancy,    8)
 963
 964 struct bch_alloc_v3 {
 965         struct bch_val          v;
 966         __le64                  journal_seq;
 967         __le32                  flags;
 968         __u8                    nr_fields;
 969         __u8                    gen;
 970         __u8                    oldest_gen;
 971         __u8                    data_type;
 972         __u8                    data[];
 973 } __packed __aligned(8);
 974
 975 LE32_BITMASK(BCH_ALLOC_V3_NEED_DISCARD,struct bch_alloc_v3, flags,  0,  1)
 976 LE32_BITMASK(BCH_ALLOC_V3_NEED_INC_GEN,struct bch_alloc_v3, flags,  1,  2)
 977
 978 struct bch_alloc_v4 {
 979         struct bch_val          v;
 980         __u64                   journal_seq;
 981         __u32                   flags;
 982         __u8                    gen;
 983         __u8                    oldest_gen;
 984         __u8                    data_type;
 985         __u8                    stripe_redundancy;
 986         __u32                   dirty_sectors;
 987         __u32                   cached_sectors;
 988         __u64                   io_time[2];
 989         __u32                   stripe;
 990         __u32                   nr_external_backpointers;
 991         __u64                   fragmentation_lru;
 992 } __packed __aligned(8);
 993
 994 #define BCH_ALLOC_V4_U64s_V0    6
 995 #define BCH_ALLOC_V4_U64s       (sizeof(struct bch_alloc_v4) / sizeof(u64))
 996
 997 BITMASK(BCH_ALLOC_V4_NEED_DISCARD,      struct bch_alloc_v4, flags,  0,  1)
 998 BITMASK(BCH_ALLOC_V4_NEED_INC_GEN,      struct bch_alloc_v4, flags,  1,  2)
 999 BITMASK(BCH_ALLOC_V4_BACKPOINTERS_START,struct bch_alloc_v4, flags,  2,  8)
1000 BITMASK(BCH_ALLOC_V4_NR_BACKPOINTERS,   struct bch_alloc_v4, flags,  8,  14)
1001
1002 #define BCH_ALLOC_V4_NR_BACKPOINTERS_MAX        40
1003
1004 struct bch_backpointer {
1005         struct bch_val          v;
1006         __u8                    btree_id;
1007         __u8                    level;
1008         __u8                    data_type;
1009         __u64                   bucket_offset:40;
1010         __u32                   bucket_len;
1011         struct bpos             pos;
1012 } __packed __aligned(8);
1013
1014 #define KEY_TYPE_BUCKET_GENS_BITS       8
1015 #define KEY_TYPE_BUCKET_GENS_NR         (1U << KEY_TYPE_BUCKET_GENS_BITS)
1016 #define KEY_TYPE_BUCKET_GENS_MASK       (KEY_TYPE_BUCKET_GENS_NR - 1)
1017
1018 struct bch_bucket_gens {
1019         struct bch_val          v;
1020         u8                      gens[KEY_TYPE_BUCKET_GENS_NR];
1021 } __packed __aligned(8);
1022
1023 /* Quotas: */
1024
1025 enum quota_types {
1026         QTYP_USR                = 0,
1027         QTYP_GRP                = 1,
1028         QTYP_PRJ                = 2,
1029         QTYP_NR                 = 3,
1030 };
1031
1032 enum quota_counters {
1033         Q_SPC                   = 0,
1034         Q_INO                   = 1,
1035         Q_COUNTERS              = 2,
1036 };
1037
1038 struct bch_quota_counter {
1039         __le64                  hardlimit;
1040         __le64                  softlimit;
1041 };
1042
1043 struct bch_quota {
1044         struct bch_val          v;
1045         struct bch_quota_counter c[Q_COUNTERS];
1046 } __packed __aligned(8);
1047
1048 /* Erasure coding */
1049
1050 struct bch_stripe {
1051         struct bch_val          v;
1052         __le16                  sectors;
1053         __u8                    algorithm;
1054         __u8                    nr_blocks;
1055         __u8                    nr_redundant;
1056
1057         __u8                    csum_granularity_bits;
1058         __u8                    csum_type;
1059         __u8                    pad;
1060
1061         struct bch_extent_ptr   ptrs[];
1062 } __packed __aligned(8);
1063
1064 /* Reflink: */
1065
1066 struct bch_reflink_p {
1067         struct bch_val          v;
1068         __le64                  idx;
1069         /*
1070          * A reflink pointer might point to an indirect extent which is then
1071          * later split (by copygc or rebalance). If we only pointed to part of
1072          * the original indirect extent, and then one of the fragments is
1073          * outside the range we point to, we'd leak a refcount: so when creating
1074          * reflink pointers, we need to store pad values to remember the full
1075          * range we were taking a reference on.
1076          */
1077         __le32                  front_pad;
1078         __le32                  back_pad;
1079 } __packed __aligned(8);
1080
1081 struct bch_reflink_v {
1082         struct bch_val          v;
1083         __le64                  refcount;
1084         union bch_extent_entry  start[0];
1085         __u64                   _data[0];
1086 } __packed __aligned(8);
1087
1088 struct bch_indirect_inline_data {
1089         struct bch_val          v;
1090         __le64                  refcount;
1091         u8                      data[0];
1092 };
1093
1094 /* Inline data */
1095
1096 struct bch_inline_data {
1097         struct bch_val          v;
1098         u8                      data[0];
1099 };
1100
1101 /* Subvolumes: */
1102
1103 #define SUBVOL_POS_MIN          POS(0, 1)
1104 #define SUBVOL_POS_MAX          POS(0, S32_MAX)
1105 #define BCACHEFS_ROOT_SUBVOL    1
1106
1107 struct bch_subvolume {
1108         struct bch_val          v;
1109         __le32                  flags;
1110         __le32                  snapshot;
1111         __le64                  inode;
1112 };
1113
1114 LE32_BITMASK(BCH_SUBVOLUME_RO,          struct bch_subvolume, flags,  0,  1)
1115 /*
1116  * We need to know whether a subvolume is a snapshot so we can know whether we
1117  * can delete it (or whether it should just be rm -rf'd)
1118  */
1119 LE32_BITMASK(BCH_SUBVOLUME_SNAP,        struct bch_subvolume, flags,  1,  2)
1120 LE32_BITMASK(BCH_SUBVOLUME_UNLINKED,    struct bch_subvolume, flags,  2,  3)
1121
1122 /* Snapshots */
1123
1124 struct bch_snapshot {
1125         struct bch_val          v;
1126         __le32                  flags;
1127         __le32                  parent;
1128         __le32                  children[2];
1129         __le32                  subvol;
1130         __le32                  pad;
1131 };
1132
1133 LE32_BITMASK(BCH_SNAPSHOT_DELETED,      struct bch_snapshot, flags,  0,  1)
1134
1135 /* True if a subvolume points to this snapshot node: */
1136 LE32_BITMASK(BCH_SNAPSHOT_SUBVOL,       struct bch_snapshot, flags,  1,  2)
1137
1138 /* LRU btree: */
1139
1140 struct bch_lru {
1141         struct bch_val          v;
1142         __le64                  idx;
1143 } __packed __aligned(8);
1144
1145 #define LRU_ID_STRIPES          (1U << 16)
1146
1147 /* Optional/variable size superblock sections: */
1148
1149 struct bch_sb_field {
1150         __u64                   _data[0];
1151         __le32                  u64s;
1152         __le32                  type;
1153 };
1154
1155 #define BCH_SB_FIELDS()                         \
1156         x(journal,      0)                      \
1157         x(members,      1)                      \
1158         x(crypt,        2)                      \
1159         x(replicas_v0,  3)                      \
1160         x(quota,        4)                      \
1161         x(disk_groups,  5)                      \
1162         x(clean,        6)                      \
1163         x(replicas,     7)                      \
1164         x(journal_seq_blacklist, 8)             \
1165         x(journal_v2,   9)                      \
1166         x(counters,     10)
1167
1168 enum bch_sb_field_type {
1169 #define x(f, nr)        BCH_SB_FIELD_##f = nr,
1170         BCH_SB_FIELDS()
1171 #undef x
1172         BCH_SB_FIELD_NR
1173 };
1174
1175 /*
1176  * Most superblock fields are replicated in all device's superblocks - a few are
1177  * not:
1178  */
1179 #define BCH_SINGLE_DEVICE_SB_FIELDS             \
1180         ((1U << BCH_SB_FIELD_journal)|          \
1181          (1U << BCH_SB_FIELD_journal_v2))
1182
1183 /* BCH_SB_FIELD_journal: */
1184
1185 struct bch_sb_field_journal {
1186         struct bch_sb_field     field;
1187         __le64                  buckets[0];
1188 };
1189
1190 struct bch_sb_field_journal_v2 {
1191         struct bch_sb_field     field;
1192
1193         struct bch_sb_field_journal_v2_entry {
1194                 __le64          start;
1195                 __le64          nr;
1196         }                       d[0];
1197 };
1198
1199 /* BCH_SB_FIELD_members: */
1200
1201 #define BCH_MIN_NR_NBUCKETS     (1 << 6)
1202
1203 struct bch_member {
1204         uuid_le                 uuid;
1205         __le64                  nbuckets;       /* device size */
1206         __le16                  first_bucket;   /* index of first bucket used */
1207         __le16                  bucket_size;    /* sectors */
1208         __le32                  pad;
1209         __le64                  last_mount;     /* time_t */
1210
1211         __le64                  flags[2];
1212 };
1213
1214 LE64_BITMASK(BCH_MEMBER_STATE,          struct bch_member, flags[0],  0,  4)
1215 /* 4-14 unused, was TIER, HAS_(META)DATA, REPLACEMENT */
1216 LE64_BITMASK(BCH_MEMBER_DISCARD,        struct bch_member, flags[0], 14, 15)
1217 LE64_BITMASK(BCH_MEMBER_DATA_ALLOWED,   struct bch_member, flags[0], 15, 20)
1218 LE64_BITMASK(BCH_MEMBER_GROUP,          struct bch_member, flags[0], 20, 28)
1219 LE64_BITMASK(BCH_MEMBER_DURABILITY,     struct bch_member, flags[0], 28, 30)
1220 LE64_BITMASK(BCH_MEMBER_FREESPACE_INITIALIZED,
1221                                         struct bch_member, flags[0], 30, 31)
1222
1223 #if 0
1224 LE64_BITMASK(BCH_MEMBER_NR_READ_ERRORS, struct bch_member, flags[1], 0,  20);
1225 LE64_BITMASK(BCH_MEMBER_NR_WRITE_ERRORS,struct bch_member, flags[1], 20, 40);
1226 #endif
1227
1228 #define BCH_MEMBER_STATES()                     \
1229         x(rw,           0)                      \
1230         x(ro,           1)                      \
1231         x(failed,       2)                      \
1232         x(spare,        3)
1233
1234 enum bch_member_state {
1235 #define x(t, n) BCH_MEMBER_STATE_##t = n,
1236         BCH_MEMBER_STATES()
1237 #undef x
1238         BCH_MEMBER_STATE_NR
1239 };
1240
1241 struct bch_sb_field_members {
1242         struct bch_sb_field     field;
1243         struct bch_member       members[0];
1244 };
1245
1246 /* BCH_SB_FIELD_crypt: */
1247
1248 struct nonce {
1249         __le32                  d[4];
1250 };
1251
1252 struct bch_key {
1253         __le64                  key[4];
1254 };
1255
1256 #define BCH_KEY_MAGIC                                   \
1257         (((u64) 'b' <<  0)|((u64) 'c' <<  8)|           \
1258          ((u64) 'h' << 16)|((u64) '*' << 24)|           \
1259          ((u64) '*' << 32)|((u64) 'k' << 40)|           \
1260          ((u64) 'e' << 48)|((u64) 'y' << 56))
1261
1262 struct bch_encrypted_key {
1263         __le64                  magic;
1264         struct bch_key          key;
1265 };
1266
1267 /*
1268  * If this field is present in the superblock, it stores an encryption key which
1269  * is used encrypt all other data/metadata. The key will normally be encrypted
1270  * with the key userspace provides, but if encryption has been turned off we'll
1271  * just store the master key unencrypted in the superblock so we can access the
1272  * previously encrypted data.
1273  */
1274 struct bch_sb_field_crypt {
1275         struct bch_sb_field     field;
1276
1277         __le64                  flags;
1278         __le64                  kdf_flags;
1279         struct bch_encrypted_key key;
1280 };
1281
1282 LE64_BITMASK(BCH_CRYPT_KDF_TYPE,        struct bch_sb_field_crypt, flags, 0, 4);
1283
1284 enum bch_kdf_types {
1285         BCH_KDF_SCRYPT          = 0,
1286         BCH_KDF_NR              = 1,
1287 };
1288
1289 /* stored as base 2 log of scrypt params: */
1290 LE64_BITMASK(BCH_KDF_SCRYPT_N,  struct bch_sb_field_crypt, kdf_flags,  0, 16);
1291 LE64_BITMASK(BCH_KDF_SCRYPT_R,  struct bch_sb_field_crypt, kdf_flags, 16, 32);
1292 LE64_BITMASK(BCH_KDF_SCRYPT_P,  struct bch_sb_field_crypt, kdf_flags, 32, 48);
1293
1294 /* BCH_SB_FIELD_replicas: */
1295
1296 #define BCH_DATA_TYPES()                \
1297         x(free,         0)              \
1298         x(sb,           1)              \
1299         x(journal,      2)              \
1300         x(btree,        3)              \
1301         x(user,         4)              \
1302         x(cached,       5)              \
1303         x(parity,       6)              \
1304         x(stripe,       7)              \
1305         x(need_gc_gens, 8)              \
1306         x(need_discard, 9)
1307
1308 enum bch_data_type {
1309 #define x(t, n) BCH_DATA_##t,
1310         BCH_DATA_TYPES()
1311 #undef x
1312         BCH_DATA_NR
1313 };
1314
1315 static inline bool data_type_is_empty(enum bch_data_type type)
1316 {
1317         switch (type) {
1318         case BCH_DATA_free:
1319         case BCH_DATA_need_gc_gens:
1320         case BCH_DATA_need_discard:
1321                 return true;
1322         default:
1323                 return false;
1324         }
1325 }
1326
1327 static inline bool data_type_is_hidden(enum bch_data_type type)
1328 {
1329         switch (type) {
1330         case BCH_DATA_sb:
1331         case BCH_DATA_journal:
1332                 return true;
1333         default:
1334                 return false;
1335         }
1336 }
1337
1338 struct bch_replicas_entry_v0 {
1339         __u8                    data_type;
1340         __u8                    nr_devs;
1341         __u8                    devs[];
1342 } __packed;
1343
1344 struct bch_sb_field_replicas_v0 {
1345         struct bch_sb_field     field;
1346         struct bch_replicas_entry_v0 entries[];
1347 } __packed __aligned(8);
1348
1349 struct bch_replicas_entry {
1350         __u8                    data_type;
1351         __u8                    nr_devs;
1352         __u8                    nr_required;
1353         __u8                    devs[];
1354 } __packed;
1355
1356 #define replicas_entry_bytes(_i)                                        \
1357         (offsetof(typeof(*(_i)), devs) + (_i)->nr_devs)
1358
1359 struct bch_sb_field_replicas {
1360         struct bch_sb_field     field;
1361         struct bch_replicas_entry entries[];
1362 } __packed __aligned(8);
1363
1364 /* BCH_SB_FIELD_quota: */
1365
1366 struct bch_sb_quota_counter {
1367         __le32                          timelimit;
1368         __le32                          warnlimit;
1369 };
1370
1371 struct bch_sb_quota_type {
1372         __le64                          flags;
1373         struct bch_sb_quota_counter     c[Q_COUNTERS];
1374 };
1375
1376 struct bch_sb_field_quota {
1377         struct bch_sb_field             field;
1378         struct bch_sb_quota_type        q[QTYP_NR];
1379 } __packed __aligned(8);
1380
1381 /* BCH_SB_FIELD_disk_groups: */
1382
1383 #define BCH_SB_LABEL_SIZE               32
1384
1385 struct bch_disk_group {
1386         __u8                    label[BCH_SB_LABEL_SIZE];
1387         __le64                  flags[2];
1388 } __packed __aligned(8);
1389
1390 LE64_BITMASK(BCH_GROUP_DELETED,         struct bch_disk_group, flags[0], 0,  1)
1391 LE64_BITMASK(BCH_GROUP_DATA_ALLOWED,    struct bch_disk_group, flags[0], 1,  6)
1392 LE64_BITMASK(BCH_GROUP_PARENT,          struct bch_disk_group, flags[0], 6, 24)
1393
1394 struct bch_sb_field_disk_groups {
1395         struct bch_sb_field     field;
1396         struct bch_disk_group   entries[0];
1397 } __packed __aligned(8);
1398
1399 /* BCH_SB_FIELD_counters */
1400
1401 #define BCH_PERSISTENT_COUNTERS()                               \
1402         x(io_read,                                      0)      \
1403         x(io_write,                                     1)      \
1404         x(io_move,                                      2)      \
1405         x(bucket_invalidate,                            3)      \
1406         x(bucket_discard,                               4)      \
1407         x(bucket_alloc,                                 5)      \
1408         x(bucket_alloc_fail,                            6)      \
1409         x(btree_cache_scan,                             7)      \
1410         x(btree_cache_reap,                             8)      \
1411         x(btree_cache_cannibalize,                      9)      \
1412         x(btree_cache_cannibalize_lock,                 10)     \
1413         x(btree_cache_cannibalize_lock_fail,            11)     \
1414         x(btree_cache_cannibalize_unlock,               12)     \
1415         x(btree_node_write,                             13)     \
1416         x(btree_node_read,                              14)     \
1417         x(btree_node_compact,                           15)     \
1418         x(btree_node_merge,                             16)     \
1419         x(btree_node_split,                             17)     \
1420         x(btree_node_rewrite,                           18)     \
1421         x(btree_node_alloc,                             19)     \
1422         x(btree_node_free,                              20)     \
1423         x(btree_node_set_root,                          21)     \
1424         x(btree_path_relock_fail,                       22)     \
1425         x(btree_path_upgrade_fail,                      23)     \
1426         x(btree_reserve_get_fail,                       24)     \
1427         x(journal_entry_full,                           25)     \
1428         x(journal_full,                                 26)     \
1429         x(journal_reclaim_finish,                       27)     \
1430         x(journal_reclaim_start,                        28)     \
1431         x(journal_write,                                29)     \
1432         x(read_promote,                                 30)     \
1433         x(read_bounce,                                  31)     \
1434         x(read_split,                                   33)     \
1435         x(read_retry,                                   32)     \
1436         x(read_reuse_race,                              34)     \
1437         x(move_extent_read,                             35)     \
1438         x(move_extent_write,                            36)     \
1439         x(move_extent_finish,                           37)     \
1440         x(move_extent_fail,                             38)     \
1441         x(move_extent_alloc_mem_fail,                   39)     \
1442         x(copygc,                                       40)     \
1443         x(copygc_wait,                                  41)     \
1444         x(gc_gens_end,                                  42)     \
1445         x(gc_gens_start,                                43)     \
1446         x(trans_blocked_journal_reclaim,                44)     \
1447         x(trans_restart_btree_node_reused,              45)     \
1448         x(trans_restart_btree_node_split,               46)     \
1449         x(trans_restart_fault_inject,                   47)     \
1450         x(trans_restart_iter_upgrade,                   48)     \
1451         x(trans_restart_journal_preres_get,             49)     \
1452         x(trans_restart_journal_reclaim,                50)     \
1453         x(trans_restart_journal_res_get,                51)     \
1454         x(trans_restart_key_cache_key_realloced,        52)     \
1455         x(trans_restart_key_cache_raced,                53)     \
1456         x(trans_restart_mark_replicas,                  54)     \
1457         x(trans_restart_mem_realloced,                  55)     \
1458         x(trans_restart_memory_allocation_failure,      56)     \
1459         x(trans_restart_relock,                         57)     \
1460         x(trans_restart_relock_after_fill,              58)     \
1461         x(trans_restart_relock_key_cache_fill,          59)     \
1462         x(trans_restart_relock_next_node,               60)     \
1463         x(trans_restart_relock_parent_for_fill,         61)     \
1464         x(trans_restart_relock_path,                    62)     \
1465         x(trans_restart_relock_path_intent,             63)     \
1466         x(trans_restart_too_many_iters,                 64)     \
1467         x(trans_restart_traverse,                       65)     \
1468         x(trans_restart_upgrade,                        66)     \
1469         x(trans_restart_would_deadlock,                 67)     \
1470         x(trans_restart_would_deadlock_write,           68)     \
1471         x(trans_restart_injected,                       69)     \
1472         x(trans_restart_key_cache_upgrade,              70)     \
1473         x(trans_traverse_all,                           71)     \
1474         x(transaction_commit,                           72)     \
1475         x(write_super,                                  73)     \
1476         x(trans_restart_would_deadlock_recursion_limit, 74)     \
1477         x(trans_restart_write_buffer_flush,             75)     \
1478         x(trans_restart_split_race,                     76)
1479
1480 enum bch_persistent_counters {
1481 #define x(t, n, ...) BCH_COUNTER_##t,
1482         BCH_PERSISTENT_COUNTERS()
1483 #undef x
1484         BCH_COUNTER_NR
1485 };
1486
1487 struct bch_sb_field_counters {
1488         struct bch_sb_field     field;
1489         __le64                  d[0];
1490 };
1491
1492 /*
1493  * On clean shutdown, store btree roots and current journal sequence number in
1494  * the superblock:
1495  */
1496 struct jset_entry {
1497         __le16                  u64s;
1498         __u8                    btree_id;
1499         __u8                    level;
1500         __u8                    type; /* designates what this jset holds */
1501         __u8                    pad[3];
1502
1503         union {
1504                 struct bkey_i   start[0];
1505                 __u64           _data[0];
1506         };
1507 };
1508
1509 struct bch_sb_field_clean {
1510         struct bch_sb_field     field;
1511
1512         __le32                  flags;
1513         __le16                  _read_clock; /* no longer used */
1514         __le16                  _write_clock;
1515         __le64                  journal_seq;
1516
1517         union {
1518                 struct jset_entry start[0];
1519                 __u64           _data[0];
1520         };
1521 };
1522
1523 struct journal_seq_blacklist_entry {
1524         __le64                  start;
1525         __le64                  end;
1526 };
1527
1528 struct bch_sb_field_journal_seq_blacklist {
1529         struct bch_sb_field     field;
1530
1531         union {
1532                 struct journal_seq_blacklist_entry start[0];
1533                 __u64           _data[0];
1534         };
1535 };
1536
1537 /* Superblock: */
1538
1539 /*
1540  * New versioning scheme:
1541  * One common version number for all on disk data structures - superblock, btree
1542  * nodes, journal entries
1543  */
1544 #define BCH_JSET_VERSION_OLD                    2
1545 #define BCH_BSET_VERSION_OLD                    3
1546
1547 #define BCH_METADATA_VERSIONS()                         \
1548         x(bkey_renumber,                10)             \
1549         x(inode_btree_change,           11)             \
1550         x(snapshot,                     12)             \
1551         x(inode_backpointers,           13)             \
1552         x(btree_ptr_sectors_written,    14)             \
1553         x(snapshot_2,                   15)             \
1554         x(reflink_p_fix,                16)             \
1555         x(subvol_dirent,                17)             \
1556         x(inode_v2,                     18)             \
1557         x(freespace,                    19)             \
1558         x(alloc_v4,                     20)             \
1559         x(new_data_types,               21)             \
1560         x(backpointers,                 22)             \
1561         x(inode_v3,                     23)             \
1562         x(unwritten_extents,            24)             \
1563         x(bucket_gens,                  25)             \
1564         x(lru_v2,                       26)             \
1565         x(fragmentation_lru,            27)
1566
1567 enum bcachefs_metadata_version {
1568         bcachefs_metadata_version_min = 9,
1569 #define x(t, n) bcachefs_metadata_version_##t = n,
1570         BCH_METADATA_VERSIONS()
1571 #undef x
1572         bcachefs_metadata_version_max
1573 };
1574
1575 #define bcachefs_metadata_version_current       (bcachefs_metadata_version_max - 1)
1576
1577 #define BCH_SB_SECTOR                   8
1578 #define BCH_SB_MEMBERS_MAX              64 /* XXX kill */
1579
1580 struct bch_sb_layout {
1581         uuid_le                 magic;  /* bcachefs superblock UUID */
1582         __u8                    layout_type;
1583         __u8                    sb_max_size_bits; /* base 2 of 512 byte sectors */
1584         __u8                    nr_superblocks;
1585         __u8                    pad[5];
1586         __le64                  sb_offset[61];
1587 } __packed __aligned(8);
1588
1589 #define BCH_SB_LAYOUT_SECTOR    7
1590
1591 /*
1592  * @offset      - sector where this sb was written
1593  * @version     - on disk format version
1594  * @version_min - Oldest metadata version this filesystem contains; so we can
1595  *                safely drop compatibility code and refuse to mount filesystems
1596  *                we'd need it for
1597  * @magic       - identifies as a bcachefs superblock (BCHFS_MAGIC)
1598  * @seq         - incremented each time superblock is written
1599  * @uuid        - used for generating various magic numbers and identifying
1600  *                member devices, never changes
1601  * @user_uuid   - user visible UUID, may be changed
1602  * @label       - filesystem label
1603  * @seq         - identifies most recent superblock, incremented each time
1604  *                superblock is written
1605  * @features    - enabled incompatible features
1606  */
1607 struct bch_sb {
1608         struct bch_csum         csum;
1609         __le16                  version;
1610         __le16                  version_min;
1611         __le16                  pad[2];
1612         uuid_le                 magic;
1613         uuid_le                 uuid;
1614         uuid_le                 user_uuid;
1615         __u8                    label[BCH_SB_LABEL_SIZE];
1616         __le64                  offset;
1617         __le64                  seq;
1618
1619         __le16                  block_size;
1620         __u8                    dev_idx;
1621         __u8                    nr_devices;
1622         __le32                  u64s;
1623
1624         __le64                  time_base_lo;
1625         __le32                  time_base_hi;
1626         __le32                  time_precision;
1627
1628         __le64                  flags[8];
1629         __le64                  features[2];
1630         __le64                  compat[2];
1631
1632         struct bch_sb_layout    layout;
1633
1634         union {
1635                 struct bch_sb_field start[0];
1636                 __le64          _data[0];
1637         };
1638 } __packed __aligned(8);
1639
1640 /*
1641  * Flags:
1642  * BCH_SB_INITALIZED    - set on first mount
1643  * BCH_SB_CLEAN         - did we shut down cleanly? Just a hint, doesn't affect
1644  *                        behaviour of mount/recovery path:
1645  * BCH_SB_INODE_32BIT   - limit inode numbers to 32 bits
1646  * BCH_SB_128_BIT_MACS  - 128 bit macs instead of 80
1647  * BCH_SB_ENCRYPTION_TYPE - if nonzero encryption is enabled; overrides
1648  *                         DATA/META_CSUM_TYPE. Also indicates encryption
1649  *                         algorithm in use, if/when we get more than one
1650  */
1651
1652 LE16_BITMASK(BCH_SB_BLOCK_SIZE,         struct bch_sb, block_size, 0, 16);
1653
1654 LE64_BITMASK(BCH_SB_INITIALIZED,        struct bch_sb, flags[0],  0,  1);
1655 LE64_BITMASK(BCH_SB_CLEAN,              struct bch_sb, flags[0],  1,  2);
1656 LE64_BITMASK(BCH_SB_CSUM_TYPE,          struct bch_sb, flags[0],  2,  8);
1657 LE64_BITMASK(BCH_SB_ERROR_ACTION,       struct bch_sb, flags[0],  8, 12);
1658
1659 LE64_BITMASK(BCH_SB_BTREE_NODE_SIZE,    struct bch_sb, flags[0], 12, 28);
1660
1661 LE64_BITMASK(BCH_SB_GC_RESERVE,         struct bch_sb, flags[0], 28, 33);
1662 LE64_BITMASK(BCH_SB_ROOT_RESERVE,       struct bch_sb, flags[0], 33, 40);
1663
1664 LE64_BITMASK(BCH_SB_META_CSUM_TYPE,     struct bch_sb, flags[0], 40, 44);
1665 LE64_BITMASK(BCH_SB_DATA_CSUM_TYPE,     struct bch_sb, flags[0], 44, 48);
1666
1667 LE64_BITMASK(BCH_SB_META_REPLICAS_WANT, struct bch_sb, flags[0], 48, 52);
1668 LE64_BITMASK(BCH_SB_DATA_REPLICAS_WANT, struct bch_sb, flags[0], 52, 56);
1669
1670 LE64_BITMASK(BCH_SB_POSIX_ACL,          struct bch_sb, flags[0], 56, 57);
1671 LE64_BITMASK(BCH_SB_USRQUOTA,           struct bch_sb, flags[0], 57, 58);
1672 LE64_BITMASK(BCH_SB_GRPQUOTA,           struct bch_sb, flags[0], 58, 59);
1673 LE64_BITMASK(BCH_SB_PRJQUOTA,           struct bch_sb, flags[0], 59, 60);
1674
1675 LE64_BITMASK(BCH_SB_HAS_ERRORS,         struct bch_sb, flags[0], 60, 61);
1676 LE64_BITMASK(BCH_SB_HAS_TOPOLOGY_ERRORS,struct bch_sb, flags[0], 61, 62);
1677
1678 LE64_BITMASK(BCH_SB_BIG_ENDIAN,         struct bch_sb, flags[0], 62, 63);
1679
1680 LE64_BITMASK(BCH_SB_STR_HASH_TYPE,      struct bch_sb, flags[1],  0,  4);
1681 LE64_BITMASK(BCH_SB_COMPRESSION_TYPE,   struct bch_sb, flags[1],  4,  8);
1682 LE64_BITMASK(BCH_SB_INODE_32BIT,        struct bch_sb, flags[1],  8,  9);
1683
1684 LE64_BITMASK(BCH_SB_128_BIT_MACS,       struct bch_sb, flags[1],  9, 10);
1685 LE64_BITMASK(BCH_SB_ENCRYPTION_TYPE,    struct bch_sb, flags[1], 10, 14);
1686
1687 /*
1688  * Max size of an extent that may require bouncing to read or write
1689  * (checksummed, compressed): 64k
1690  */
1691 LE64_BITMASK(BCH_SB_ENCODED_EXTENT_MAX_BITS,
1692                                         struct bch_sb, flags[1], 14, 20);
1693
1694 LE64_BITMASK(BCH_SB_META_REPLICAS_REQ,  struct bch_sb, flags[1], 20, 24);
1695 LE64_BITMASK(BCH_SB_DATA_REPLICAS_REQ,  struct bch_sb, flags[1], 24, 28);
1696
1697 LE64_BITMASK(BCH_SB_PROMOTE_TARGET,     struct bch_sb, flags[1], 28, 40);
1698 LE64_BITMASK(BCH_SB_FOREGROUND_TARGET,  struct bch_sb, flags[1], 40, 52);
1699 LE64_BITMASK(BCH_SB_BACKGROUND_TARGET,  struct bch_sb, flags[1], 52, 64);
1700
1701 LE64_BITMASK(BCH_SB_BACKGROUND_COMPRESSION_TYPE,
1702                                         struct bch_sb, flags[2],  0,  4);
1703 LE64_BITMASK(BCH_SB_GC_RESERVE_BYTES,   struct bch_sb, flags[2],  4, 64);
1704
1705 LE64_BITMASK(BCH_SB_ERASURE_CODE,       struct bch_sb, flags[3],  0, 16);
1706 LE64_BITMASK(BCH_SB_METADATA_TARGET,    struct bch_sb, flags[3], 16, 28);
1707 LE64_BITMASK(BCH_SB_SHARD_INUMS,        struct bch_sb, flags[3], 28, 29);
1708 LE64_BITMASK(BCH_SB_INODES_USE_KEY_CACHE,struct bch_sb, flags[3], 29, 30);
1709 LE64_BITMASK(BCH_SB_JOURNAL_FLUSH_DELAY,struct bch_sb, flags[3], 30, 62);
1710 LE64_BITMASK(BCH_SB_JOURNAL_FLUSH_DISABLED,struct bch_sb, flags[3], 62, 63);
1711 LE64_BITMASK(BCH_SB_JOURNAL_RECLAIM_DELAY,struct bch_sb, flags[4], 0, 32);
1712 LE64_BITMASK(BCH_SB_JOURNAL_TRANSACTION_NAMES,struct bch_sb, flags[4], 32, 33);
1713 LE64_BITMASK(BCH_SB_NOCOW,              struct bch_sb, flags[4], 33, 34);
1714 LE64_BITMASK(BCH_SB_WRITE_BUFFER_SIZE,  struct bch_sb, flags[4], 34, 54);
1715
1716 /*
1717  * Features:
1718  *
1719  * journal_seq_blacklist_v3:    gates BCH_SB_FIELD_journal_seq_blacklist
1720  * reflink:                     gates KEY_TYPE_reflink
1721  * inline_data:                 gates KEY_TYPE_inline_data
1722  * new_siphash:                 gates BCH_STR_HASH_siphash
1723  * new_extent_overwrite:        gates BTREE_NODE_NEW_EXTENT_OVERWRITE
1724  */
1725 #define BCH_SB_FEATURES()                       \
1726         x(lz4,                          0)      \
1727         x(gzip,                         1)      \
1728         x(zstd,                         2)      \
1729         x(atomic_nlink,                 3)      \
1730         x(ec,                           4)      \
1731         x(journal_seq_blacklist_v3,     5)      \
1732         x(reflink,                      6)      \
1733         x(new_siphash,                  7)      \
1734         x(inline_data,                  8)      \
1735         x(new_extent_overwrite,         9)      \
1736         x(incompressible,               10)     \
1737         x(btree_ptr_v2,                 11)     \
1738         x(extents_above_btree_updates,  12)     \
1739         x(btree_updates_journalled,     13)     \
1740         x(reflink_inline_data,          14)     \
1741         x(new_varint,                   15)     \
1742         x(journal_no_flush,             16)     \
1743         x(alloc_v2,                     17)     \
1744         x(extents_across_btree_nodes,   18)
1745
1746 #define BCH_SB_FEATURES_ALWAYS                          \
1747         ((1ULL << BCH_FEATURE_new_extent_overwrite)|    \
1748          (1ULL << BCH_FEATURE_extents_above_btree_updates)|\
1749          (1ULL << BCH_FEATURE_btree_updates_journalled)|\
1750          (1ULL << BCH_FEATURE_alloc_v2)|\
1751          (1ULL << BCH_FEATURE_extents_across_btree_nodes))
1752
1753 #define BCH_SB_FEATURES_ALL                             \
1754         (BCH_SB_FEATURES_ALWAYS|                        \
1755          (1ULL << BCH_FEATURE_new_siphash)|             \
1756          (1ULL << BCH_FEATURE_btree_ptr_v2)|            \
1757          (1ULL << BCH_FEATURE_new_varint)|              \
1758          (1ULL << BCH_FEATURE_journal_no_flush))
1759
1760 enum bch_sb_feature {
1761 #define x(f, n) BCH_FEATURE_##f,
1762         BCH_SB_FEATURES()
1763 #undef x
1764         BCH_FEATURE_NR,
1765 };
1766
1767 #define BCH_SB_COMPAT()                                 \
1768         x(alloc_info,                           0)      \
1769         x(alloc_metadata,                       1)      \
1770         x(extents_above_btree_updates_done,     2)      \
1771         x(bformat_overflow_done,                3)
1772
1773 enum bch_sb_compat {
1774 #define x(f, n) BCH_COMPAT_##f,
1775         BCH_SB_COMPAT()
1776 #undef x
1777         BCH_COMPAT_NR,
1778 };
1779
1780 /* options: */
1781
1782 #define BCH_REPLICAS_MAX                4U
1783
1784 #define BCH_BKEY_PTRS_MAX               16U
1785
1786 #define BCH_ERROR_ACTIONS()             \
1787         x(continue,             0)      \
1788         x(ro,                   1)      \
1789         x(panic,                2)
1790
1791 enum bch_error_actions {
1792 #define x(t, n) BCH_ON_ERROR_##t = n,
1793         BCH_ERROR_ACTIONS()
1794 #undef x
1795         BCH_ON_ERROR_NR
1796 };
1797
1798 #define BCH_STR_HASH_TYPES()            \
1799         x(crc32c,               0)      \
1800         x(crc64,                1)      \
1801         x(siphash_old,          2)      \
1802         x(siphash,              3)
1803
1804 enum bch_str_hash_type {
1805 #define x(t, n) BCH_STR_HASH_##t = n,
1806         BCH_STR_HASH_TYPES()
1807 #undef x
1808         BCH_STR_HASH_NR
1809 };
1810
1811 #define BCH_STR_HASH_OPTS()             \
1812         x(crc32c,               0)      \
1813         x(crc64,                1)      \
1814         x(siphash,              2)
1815
1816 enum bch_str_hash_opts {
1817 #define x(t, n) BCH_STR_HASH_OPT_##t = n,
1818         BCH_STR_HASH_OPTS()
1819 #undef x
1820         BCH_STR_HASH_OPT_NR
1821 };
1822
1823 #define BCH_CSUM_TYPES()                        \
1824         x(none,                         0)      \
1825         x(crc32c_nonzero,               1)      \
1826         x(crc64_nonzero,                2)      \
1827         x(chacha20_poly1305_80,         3)      \
1828         x(chacha20_poly1305_128,        4)      \
1829         x(crc32c,                       5)      \
1830         x(crc64,                        6)      \
1831         x(xxhash,                       7)
1832
1833 enum bch_csum_type {
1834 #define x(t, n) BCH_CSUM_##t = n,
1835         BCH_CSUM_TYPES()
1836 #undef x
1837         BCH_CSUM_NR
1838 };
1839
1840 static const unsigned bch_crc_bytes[] = {
1841         [BCH_CSUM_none]                         = 0,
1842         [BCH_CSUM_crc32c_nonzero]               = 4,
1843         [BCH_CSUM_crc32c]                       = 4,
1844         [BCH_CSUM_crc64_nonzero]                = 8,
1845         [BCH_CSUM_crc64]                        = 8,
1846         [BCH_CSUM_xxhash]                       = 8,
1847         [BCH_CSUM_chacha20_poly1305_80]         = 10,
1848         [BCH_CSUM_chacha20_poly1305_128]        = 16,
1849 };
1850
1851 static inline _Bool bch2_csum_type_is_encryption(enum bch_csum_type type)
1852 {
1853         switch (type) {
1854         case BCH_CSUM_chacha20_poly1305_80:
1855         case BCH_CSUM_chacha20_poly1305_128:
1856                 return true;
1857         default:
1858                 return false;
1859         }
1860 }
1861
1862 #define BCH_CSUM_OPTS()                 \
1863         x(none,                 0)      \
1864         x(crc32c,               1)      \
1865         x(crc64,                2)      \
1866         x(xxhash,               3)
1867
1868 enum bch_csum_opts {
1869 #define x(t, n) BCH_CSUM_OPT_##t = n,
1870         BCH_CSUM_OPTS()
1871 #undef x
1872         BCH_CSUM_OPT_NR
1873 };
1874
1875 #define BCH_COMPRESSION_TYPES()         \
1876         x(none,                 0)      \
1877         x(lz4_old,              1)      \
1878         x(gzip,                 2)      \
1879         x(lz4,                  3)      \
1880         x(zstd,                 4)      \
1881         x(incompressible,       5)
1882
1883 enum bch_compression_type {
1884 #define x(t, n) BCH_COMPRESSION_TYPE_##t = n,
1885         BCH_COMPRESSION_TYPES()
1886 #undef x
1887         BCH_COMPRESSION_TYPE_NR
1888 };
1889
1890 #define BCH_COMPRESSION_OPTS()          \
1891         x(none,         0)              \
1892         x(lz4,          1)              \
1893         x(gzip,         2)              \
1894         x(zstd,         3)
1895
1896 enum bch_compression_opts {
1897 #define x(t, n) BCH_COMPRESSION_OPT_##t = n,
1898         BCH_COMPRESSION_OPTS()
1899 #undef x
1900         BCH_COMPRESSION_OPT_NR
1901 };
1902
1903 /*
1904  * Magic numbers
1905  *
1906  * The various other data structures have their own magic numbers, which are
1907  * xored with the first part of the cache set's UUID
1908  */
1909
1910 #define BCACHE_MAGIC                                                    \
1911         UUID_LE(0xf67385c6, 0x1a4e, 0xca45,                             \
1912                 0x82, 0x65, 0xf5, 0x7f, 0x48, 0xba, 0x6d, 0x81)
1913 #define BCHFS_MAGIC                                                     \
1914         UUID_LE(0xf67385c6, 0xce66, 0xa990,                             \
1915                 0xd9, 0x6a, 0x60, 0xcf, 0x80, 0x3d, 0xf7, 0xef)
1916
1917 #define BCACHEFS_STATFS_MAGIC           0xca451a4e
1918
1919 #define JSET_MAGIC              __cpu_to_le64(0x245235c1a3625032ULL)
1920 #define BSET_MAGIC              __cpu_to_le64(0x90135c78b99e07f5ULL)
1921
1922 static inline __le64 __bch2_sb_magic(struct bch_sb *sb)
1923 {
1924         __le64 ret;
1925
1926         memcpy(&ret, &sb->uuid, sizeof(ret));
1927         return ret;
1928 }
1929
1930 static inline __u64 __jset_magic(struct bch_sb *sb)
1931 {
1932         return __le64_to_cpu(__bch2_sb_magic(sb) ^ JSET_MAGIC);
1933 }
1934
1935 static inline __u64 __bset_magic(struct bch_sb *sb)
1936 {
1937         return __le64_to_cpu(__bch2_sb_magic(sb) ^ BSET_MAGIC);
1938 }
1939
1940 /* Journal */
1941
1942 #define JSET_KEYS_U64s  (sizeof(struct jset_entry) / sizeof(__u64))
1943
1944 #define BCH_JSET_ENTRY_TYPES()                  \
1945         x(btree_keys,           0)              \
1946         x(btree_root,           1)              \
1947         x(prio_ptrs,            2)              \
1948         x(blacklist,            3)              \
1949         x(blacklist_v2,         4)              \
1950         x(usage,                5)              \
1951         x(data_usage,           6)              \
1952         x(clock,                7)              \
1953         x(dev_usage,            8)              \
1954         x(log,                  9)              \
1955         x(overwrite,            10)
1956
1957 enum {
1958 #define x(f, nr)        BCH_JSET_ENTRY_##f      = nr,
1959         BCH_JSET_ENTRY_TYPES()
1960 #undef x
1961         BCH_JSET_ENTRY_NR
1962 };
1963
1964 /*
1965  * Journal sequence numbers can be blacklisted: bsets record the max sequence
1966  * number of all the journal entries they contain updates for, so that on
1967  * recovery we can ignore those bsets that contain index updates newer that what
1968  * made it into the journal.
1969  *
1970  * This means that we can't reuse that journal_seq - we have to skip it, and
1971  * then record that we skipped it so that the next time we crash and recover we
1972  * don't think there was a missing journal entry.
1973  */
1974 struct jset_entry_blacklist {
1975         struct jset_entry       entry;
1976         __le64                  seq;
1977 };
1978
1979 struct jset_entry_blacklist_v2 {
1980         struct jset_entry       entry;
1981         __le64                  start;
1982         __le64                  end;
1983 };
1984
1985 #define BCH_FS_USAGE_TYPES()                    \
1986         x(reserved,             0)              \
1987         x(inodes,               1)              \
1988         x(key_version,          2)
1989
1990 enum {
1991 #define x(f, nr)        BCH_FS_USAGE_##f        = nr,
1992         BCH_FS_USAGE_TYPES()
1993 #undef x
1994         BCH_FS_USAGE_NR
1995 };
1996
1997 struct jset_entry_usage {
1998         struct jset_entry       entry;
1999         __le64                  v;
2000 } __packed;
2001
2002 struct jset_entry_data_usage {
2003         struct jset_entry       entry;
2004         __le64                  v;
2005         struct bch_replicas_entry r;
2006 } __packed;
2007
2008 struct jset_entry_clock {
2009         struct jset_entry       entry;
2010         __u8                    rw;
2011         __u8                    pad[7];
2012         __le64                  time;
2013 } __packed;
2014
2015 struct jset_entry_dev_usage_type {
2016         __le64                  buckets;
2017         __le64                  sectors;
2018         __le64                  fragmented;
2019 } __packed;
2020
2021 struct jset_entry_dev_usage {
2022         struct jset_entry       entry;
2023         __le32                  dev;
2024         __u32                   pad;
2025
2026         __le64                  buckets_ec;
2027         __le64                  _buckets_unavailable; /* No longer used */
2028
2029         struct jset_entry_dev_usage_type d[];
2030 } __packed;
2031
2032 static inline unsigned jset_entry_dev_usage_nr_types(struct jset_entry_dev_usage *u)
2033 {
2034         return (vstruct_bytes(&u->entry) - sizeof(struct jset_entry_dev_usage)) /
2035                 sizeof(struct jset_entry_dev_usage_type);
2036 }
2037
2038 struct jset_entry_log {
2039         struct jset_entry       entry;
2040         u8                      d[];
2041 } __packed;
2042
2043 /*
2044  * On disk format for a journal entry:
2045  * seq is monotonically increasing; every journal entry has its own unique
2046  * sequence number.
2047  *
2048  * last_seq is the oldest journal entry that still has keys the btree hasn't
2049  * flushed to disk yet.
2050  *
2051  * version is for on disk format changes.
2052  */
2053 struct jset {
2054         struct bch_csum         csum;
2055
2056         __le64                  magic;
2057         __le64                  seq;
2058         __le32                  version;
2059         __le32                  flags;
2060
2061         __le32                  u64s; /* size of d[] in u64s */
2062
2063         __u8                    encrypted_start[0];
2064
2065         __le16                  _read_clock; /* no longer used */
2066         __le16                  _write_clock;
2067
2068         /* Sequence number of oldest dirty journal entry */
2069         __le64                  last_seq;
2070
2071
2072         union {
2073                 struct jset_entry start[0];
2074                 __u64           _data[0];
2075         };
2076 } __packed __aligned(8);
2077
2078 LE32_BITMASK(JSET_CSUM_TYPE,    struct jset, flags, 0, 4);
2079 LE32_BITMASK(JSET_BIG_ENDIAN,   struct jset, flags, 4, 5);
2080 LE32_BITMASK(JSET_NO_FLUSH,     struct jset, flags, 5, 6);
2081
2082 #define BCH_JOURNAL_BUCKETS_MIN         8
2083
2084 /* Btree: */
2085
2086 #define BCH_BTREE_IDS()                         \
2087         x(extents,              0)              \
2088         x(inodes,               1)              \
2089         x(dirents,              2)              \
2090         x(xattrs,               3)              \
2091         x(alloc,                4)              \
2092         x(quotas,               5)              \
2093         x(stripes,              6)              \
2094         x(reflink,              7)              \
2095         x(subvolumes,           8)              \
2096         x(snapshots,            9)              \
2097         x(lru,                  10)             \
2098         x(freespace,            11)             \
2099         x(need_discard,         12)             \
2100         x(backpointers,         13)             \
2101         x(bucket_gens,          14)
2102
2103 enum btree_id {
2104 #define x(kwd, val) BTREE_ID_##kwd = val,
2105         BCH_BTREE_IDS()
2106 #undef x
2107         BTREE_ID_NR
2108 };
2109
2110 #define BTREE_MAX_DEPTH         4U
2111
2112 /* Btree nodes */
2113
2114 /*
2115  * Btree nodes
2116  *
2117  * On disk a btree node is a list/log of these; within each set the keys are
2118  * sorted
2119  */
2120 struct bset {
2121         __le64                  seq;
2122
2123         /*
2124          * Highest journal entry this bset contains keys for.
2125          * If on recovery we don't see that journal entry, this bset is ignored:
2126          * this allows us to preserve the order of all index updates after a
2127          * crash, since the journal records a total order of all index updates
2128          * and anything that didn't make it to the journal doesn't get used.
2129          */
2130         __le64                  journal_seq;
2131
2132         __le32                  flags;
2133         __le16                  version;
2134         __le16                  u64s; /* count of d[] in u64s */
2135
2136         union {
2137                 struct bkey_packed start[0];
2138                 __u64           _data[0];
2139         };
2140 } __packed __aligned(8);
2141
2142 LE32_BITMASK(BSET_CSUM_TYPE,    struct bset, flags, 0, 4);
2143
2144 LE32_BITMASK(BSET_BIG_ENDIAN,   struct bset, flags, 4, 5);
2145 LE32_BITMASK(BSET_SEPARATE_WHITEOUTS,
2146                                 struct bset, flags, 5, 6);
2147
2148 /* Sector offset within the btree node: */
2149 LE32_BITMASK(BSET_OFFSET,       struct bset, flags, 16, 32);
2150
2151 struct btree_node {
2152         struct bch_csum         csum;
2153         __le64                  magic;
2154
2155         /* this flags field is encrypted, unlike bset->flags: */
2156         __le64                  flags;
2157
2158         /* Closed interval: */
2159         struct bpos             min_key;
2160         struct bpos             max_key;
2161         struct bch_extent_ptr   _ptr; /* not used anymore */
2162         struct bkey_format      format;
2163
2164         union {
2165         struct bset             keys;
2166         struct {
2167                 __u8            pad[22];
2168                 __le16          u64s;
2169                 __u64           _data[0];
2170
2171         };
2172         };
2173 } __packed __aligned(8);
2174
2175 LE64_BITMASK(BTREE_NODE_ID,     struct btree_node, flags,  0,  4);
2176 LE64_BITMASK(BTREE_NODE_LEVEL,  struct btree_node, flags,  4,  8);
2177 LE64_BITMASK(BTREE_NODE_NEW_EXTENT_OVERWRITE,
2178                                 struct btree_node, flags,  8,  9);
2179 /* 9-32 unused */
2180 LE64_BITMASK(BTREE_NODE_SEQ,    struct btree_node, flags, 32, 64);
2181
2182 struct btree_node_entry {
2183         struct bch_csum         csum;
2184
2185         union {
2186         struct bset             keys;
2187         struct {
2188                 __u8            pad[22];
2189                 __le16          u64s;
2190                 __u64           _data[0];
2191
2192         };
2193         };
2194 } __packed __aligned(8);
2195
2196 #endif /* _BCACHEFS_FORMAT_H */