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