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 #ifdef __KERNEL__
  82 typedef uuid_t __uuid_t;
  83 #endif
  84
  85 #define BITMASK(name, type, field, offset, end)                         \
  86 static const unsigned   name##_OFFSET = offset;                         \
  87 static const unsigned   name##_BITS = (end - offset);                   \
  88                                                                         \
  89 static inline __u64 name(const type *k)                                 \
  90 {                                                                       \
  91         return (k->field >> offset) & ~(~0ULL << (end - offset));       \
  92 }                                                                       \
  93                                                                         \
  94 static inline void SET_##name(type *k, __u64 v)                         \
  95 {                                                                       \
  96         k->field &= ~(~(~0ULL << (end - offset)) << offset);            \
  97         k->field |= (v & ~(~0ULL << (end - offset))) << offset;         \
  98 }
  99
 100 #define LE_BITMASK(_bits, name, type, field, offset, end)               \
 101 static const unsigned   name##_OFFSET = offset;                         \
 102 static const unsigned   name##_BITS = (end - offset);                   \
 103 static const __u##_bits name##_MAX = (1ULL << (end - offset)) - 1;      \
 104                                                                         \
 105 static inline __u64 name(const type *k)                                 \
 106 {                                                                       \
 107         return (__le##_bits##_to_cpu(k->field) >> offset) &             \
 108                 ~(~0ULL << (end - offset));                             \
 109 }                                                                       \
 110                                                                         \
 111 static inline void SET_##name(type *k, __u64 v)                         \
 112 {                                                                       \
 113         __u##_bits new = __le##_bits##_to_cpu(k->field);                \
 114                                                                         \
 115         new &= ~(~(~0ULL << (end - offset)) << offset);                 \
 116         new |= (v & ~(~0ULL << (end - offset))) << offset;              \
 117         k->field = __cpu_to_le##_bits(new);                             \
 118 }
 119
 120 #define LE16_BITMASK(n, t, f, o, e)     LE_BITMASK(16, n, t, f, o, e)
 121 #define LE32_BITMASK(n, t, f, o, e)     LE_BITMASK(32, n, t, f, o, e)
 122 #define LE64_BITMASK(n, t, f, o, e)     LE_BITMASK(64, n, t, f, o, e)
 123
 124 struct bkey_format {
 125         __u8            key_u64s;
 126         __u8            nr_fields;
 127         /* One unused slot for now: */
 128         __u8            bits_per_field[6];
 129         __le64          field_offset[6];
 130 };
 131
 132 /* Btree keys - all units are in sectors */
 133
 134 struct bpos {
 135         /*
 136          * Word order matches machine byte order - btree code treats a bpos as a
 137          * single large integer, for search/comparison purposes
 138          *
 139          * Note that wherever a bpos is embedded in another on disk data
 140          * structure, it has to be byte swabbed when reading in metadata that
 141          * wasn't written in native endian order:
 142          */
 143 #if __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__
 144         __u32           snapshot;
 145         __u64           offset;
 146         __u64           inode;
 147 #elif __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__
 148         __u64           inode;
 149         __u64           offset;         /* Points to end of extent - sectors */
 150         __u32           snapshot;
 151 #else
 152 #error edit for your odd byteorder.
 153 #endif
 154 } __packed __aligned(4);
 155
 156 #define KEY_INODE_MAX                   ((__u64)~0ULL)
 157 #define KEY_OFFSET_MAX                  ((__u64)~0ULL)
 158 #define KEY_SNAPSHOT_MAX                ((__u32)~0U)
 159 #define KEY_SIZE_MAX                    ((__u32)~0U)
 160
 161 static inline struct bpos SPOS(__u64 inode, __u64 offset, __u32 snapshot)
 162 {
 163         return (struct bpos) {
 164                 .inode          = inode,
 165                 .offset         = offset,
 166                 .snapshot       = snapshot,
 167         };
 168 }
 169
 170 #define POS_MIN                         SPOS(0, 0, 0)
 171 #define POS_MAX                         SPOS(KEY_INODE_MAX, KEY_OFFSET_MAX, 0)
 172 #define SPOS_MAX                        SPOS(KEY_INODE_MAX, KEY_OFFSET_MAX, KEY_SNAPSHOT_MAX)
 173 #define POS(_inode, _offset)            SPOS(_inode, _offset, 0)
 174
 175 /* Empty placeholder struct, for container_of() */
 176 struct bch_val {
 177         __u64           __nothing[0];
 178 };
 179
 180 struct bversion {
 181 #if __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__
 182         __u64           lo;
 183         __u32           hi;
 184 #elif __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__
 185         __u32           hi;
 186         __u64           lo;
 187 #endif
 188 } __packed __aligned(4);
 189
 190 struct bkey {
 191         /* Size of combined key and value, in u64s */
 192         __u8            u64s;
 193
 194         /* Format of key (0 for format local to btree node) */
 195 #if defined(__LITTLE_ENDIAN_BITFIELD)
 196         __u8            format:7,
 197                         needs_whiteout:1;
 198 #elif defined (__BIG_ENDIAN_BITFIELD)
 199         __u8            needs_whiteout:1,
 200                         format:7;
 201 #else
 202 #error edit for your odd byteorder.
 203 #endif
 204
 205         /* Type of the value */
 206         __u8            type;
 207
 208 #if __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__
 209         __u8            pad[1];
 210
 211         struct bversion version;
 212         __u32           size;           /* extent size, in sectors */
 213         struct bpos     p;
 214 #elif __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__
 215         struct bpos     p;
 216         __u32           size;           /* extent size, in sectors */
 217         struct bversion version;
 218
 219         __u8            pad[1];
 220 #endif
 221 } __packed __aligned(8);
 222
 223 struct bkey_packed {
 224         __u64           _data[0];
 225
 226         /* Size of combined key and value, in u64s */
 227         __u8            u64s;
 228
 229         /* Format of key (0 for format local to btree node) */
 230
 231         /*
 232          * XXX: next incompat on disk format change, switch format and
 233          * needs_whiteout - bkey_packed() will be cheaper if format is the high
 234          * bits of the bitfield
 235          */
 236 #if defined(__LITTLE_ENDIAN_BITFIELD)
 237         __u8            format:7,
 238                         needs_whiteout:1;
 239 #elif defined (__BIG_ENDIAN_BITFIELD)
 240         __u8            needs_whiteout:1,
 241                         format:7;
 242 #endif
 243
 244         /* Type of the value */
 245         __u8            type;
 246         __u8            key_start[0];
 247
 248         /*
 249          * We copy bkeys with struct assignment in various places, and while
 250          * that shouldn't be done with packed bkeys we can't disallow it in C,
 251          * and it's legal to cast a bkey to a bkey_packed  - so padding it out
 252          * to the same size as struct bkey should hopefully be safest.
 253          */
 254         __u8            pad[sizeof(struct bkey) - 3];
 255 } __packed __aligned(8);
 256
 257 typedef struct {
 258         __le64                  lo;
 259         __le64                  hi;
 260 } bch_le128;
 261
 262 #define BKEY_U64s                       (sizeof(struct bkey) / sizeof(__u64))
 263 #define BKEY_U64s_MAX                   U8_MAX
 264 #define BKEY_VAL_U64s_MAX               (BKEY_U64s_MAX - BKEY_U64s)
 265
 266 #define KEY_PACKED_BITS_START           24
 267
 268 #define KEY_FORMAT_LOCAL_BTREE          0
 269 #define KEY_FORMAT_CURRENT              1
 270
 271 enum bch_bkey_fields {
 272         BKEY_FIELD_INODE,
 273         BKEY_FIELD_OFFSET,
 274         BKEY_FIELD_SNAPSHOT,
 275         BKEY_FIELD_SIZE,
 276         BKEY_FIELD_VERSION_HI,
 277         BKEY_FIELD_VERSION_LO,
 278         BKEY_NR_FIELDS,
 279 };
 280
 281 #define bkey_format_field(name, field)                                  \
 282         [BKEY_FIELD_##name] = (sizeof(((struct bkey *) NULL)->field) * 8)
 283
 284 #define BKEY_FORMAT_CURRENT                                             \
 285 ((struct bkey_format) {                                                 \
 286         .key_u64s       = BKEY_U64s,                                    \
 287         .nr_fields      = BKEY_NR_FIELDS,                               \
 288         .bits_per_field = {                                             \
 289                 bkey_format_field(INODE,        p.inode),               \
 290                 bkey_format_field(OFFSET,       p.offset),              \
 291                 bkey_format_field(SNAPSHOT,     p.snapshot),            \
 292                 bkey_format_field(SIZE,         size),                  \
 293                 bkey_format_field(VERSION_HI,   version.hi),            \
 294                 bkey_format_field(VERSION_LO,   version.lo),            \
 295         },                                                              \
 296 })
 297
 298 /* bkey with inline value */
 299 struct bkey_i {
 300         __u64                   _data[0];
 301
 302         struct bkey     k;
 303         struct bch_val  v;
 304 };
 305
 306 #define KEY(_inode, _offset, _size)                                     \
 307 ((struct bkey) {                                                        \
 308         .u64s           = BKEY_U64s,                                    \
 309         .format         = KEY_FORMAT_CURRENT,                           \
 310         .p              = POS(_inode, _offset),                         \
 311         .size           = _size,                                        \
 312 })
 313
 314 static inline void bkey_init(struct bkey *k)
 315 {
 316         *k = KEY(0, 0, 0);
 317 }
 318
 319 #define bkey_bytes(_k)          ((_k)->u64s * sizeof(__u64))
 320
 321 #define __BKEY_PADDED(key, pad)                                 \
 322         struct bkey_i key; __u64 key ## _pad[pad]
 323
 324 /*
 325  * - DELETED keys are used internally to mark keys that should be ignored but
 326  *   override keys in composition order.  Their version number is ignored.
 327  *
 328  * - DISCARDED keys indicate that the data is all 0s because it has been
 329  *   discarded. DISCARDs may have a version; if the version is nonzero the key
 330  *   will be persistent, otherwise the key will be dropped whenever the btree
 331  *   node is rewritten (like DELETED keys).
 332  *
 333  * - ERROR: any read of the data returns a read error, as the data was lost due
 334  *   to a failing device. Like DISCARDED keys, they can be removed (overridden)
 335  *   by new writes or cluster-wide GC. Node repair can also overwrite them with
 336  *   the same or a more recent version number, but not with an older version
 337  *   number.
 338  *
 339  * - WHITEOUT: for hash table btrees
 340  */
 341 #define BCH_BKEY_TYPES()                                \
 342         x(deleted,              0)                      \
 343         x(whiteout,             1)                      \
 344         x(error,                2)                      \
 345         x(cookie,               3)                      \
 346         x(hash_whiteout,        4)                      \
 347         x(btree_ptr,            5)                      \
 348         x(extent,               6)                      \
 349         x(reservation,          7)                      \
 350         x(inode,                8)                      \
 351         x(inode_generation,     9)                      \
 352         x(dirent,               10)                     \
 353         x(xattr,                11)                     \
 354         x(alloc,                12)                     \
 355         x(quota,                13)                     \
 356         x(stripe,               14)                     \
 357         x(reflink_p,            15)                     \
 358         x(reflink_v,            16)                     \
 359         x(inline_data,          17)                     \
 360         x(btree_ptr_v2,         18)                     \
 361         x(indirect_inline_data, 19)                     \
 362         x(alloc_v2,             20)                     \
 363         x(subvolume,            21)                     \
 364         x(snapshot,             22)                     \
 365         x(inode_v2,             23)                     \
 366         x(alloc_v3,             24)                     \
 367         x(set,                  25)                     \
 368         x(lru,                  26)                     \
 369         x(alloc_v4,             27)                     \
 370         x(backpointer,          28)                     \
 371         x(inode_v3,             29)                     \
 372         x(bucket_gens,          30)                     \
 373         x(snapshot_tree,        31)
 374
 375 enum bch_bkey_type {
 376 #define x(name, nr) KEY_TYPE_##name     = nr,
 377         BCH_BKEY_TYPES()
 378 #undef x
 379         KEY_TYPE_MAX,
 380 };
 381
 382 struct bch_deleted {
 383         struct bch_val          v;
 384 };
 385
 386 struct bch_whiteout {
 387         struct bch_val          v;
 388 };
 389
 390 struct bch_error {
 391         struct bch_val          v;
 392 };
 393
 394 struct bch_cookie {
 395         struct bch_val          v;
 396         __le64                  cookie;
 397 };
 398
 399 struct bch_hash_whiteout {
 400         struct bch_val          v;
 401 };
 402
 403 struct bch_set {
 404         struct bch_val          v;
 405 };
 406
 407 /* Extents */
 408
 409 /*
 410  * In extent bkeys, the value is a list of pointers (bch_extent_ptr), optionally
 411  * preceded by checksum/compression information (bch_extent_crc32 or
 412  * bch_extent_crc64).
 413  *
 414  * One major determining factor in the format of extents is how we handle and
 415  * represent extents that have been partially overwritten and thus trimmed:
 416  *
 417  * If an extent is not checksummed or compressed, when the extent is trimmed we
 418  * don't have to remember the extent we originally allocated and wrote: we can
 419  * merely adjust ptr->offset to point to the start of the data that is currently
 420  * live. The size field in struct bkey records the current (live) size of the
 421  * extent, and is also used to mean "size of region on disk that we point to" in
 422  * this case.
 423  *
 424  * Thus an extent that is not checksummed or compressed will consist only of a
 425  * list of bch_extent_ptrs, with none of the fields in
 426  * bch_extent_crc32/bch_extent_crc64.
 427  *
 428  * When an extent is checksummed or compressed, it's not possible to read only
 429  * the data that is currently live: we have to read the entire extent that was
 430  * originally written, and then return only the part of the extent that is
 431  * currently live.
 432  *
 433  * Thus, in addition to the current size of the extent in struct bkey, we need
 434  * to store the size of the originally allocated space - this is the
 435  * compressed_size and uncompressed_size fields in bch_extent_crc32/64. Also,
 436  * when the extent is trimmed, instead of modifying the offset field of the
 437  * pointer, we keep a second smaller offset field - "offset into the original
 438  * extent of the currently live region".
 439  *
 440  * The other major determining factor is replication and data migration:
 441  *
 442  * Each pointer may have its own bch_extent_crc32/64. When doing a replicated
 443  * write, we will initially write all the replicas in the same format, with the
 444  * same checksum type and compression format - however, when copygc runs later (or
 445  * tiering/cache promotion, anything that moves data), it is not in general
 446  * going to rewrite all the pointers at once - one of the replicas may be in a
 447  * bucket on one device that has very little fragmentation while another lives
 448  * in a bucket that has become heavily fragmented, and thus is being rewritten
 449  * sooner than the rest.
 450  *
 451  * Thus it will only move a subset of the pointers (or in the case of
 452  * tiering/cache promotion perhaps add a single pointer without dropping any
 453  * current pointers), and if the extent has been partially overwritten it must
 454  * write only the currently live portion (or copygc would not be able to reduce
 455  * fragmentation!) - which necessitates a different bch_extent_crc format for
 456  * the new pointer.
 457  *
 458  * But in the interests of space efficiency, we don't want to store one
 459  * bch_extent_crc for each pointer if we don't have to.
 460  *
 461  * Thus, a bch_extent consists of bch_extent_crc32s, bch_extent_crc64s, and
 462  * bch_extent_ptrs appended arbitrarily one after the other. We determine the
 463  * type of a given entry with a scheme similar to utf8 (except we're encoding a
 464  * type, not a size), encoding the type in the position of the first set bit:
 465  *
 466  * bch_extent_crc32     - 0b1
 467  * bch_extent_ptr       - 0b10
 468  * bch_extent_crc64     - 0b100
 469  *
 470  * We do it this way because bch_extent_crc32 is _very_ constrained on bits (and
 471  * bch_extent_crc64 is the least constrained).
 472  *
 473  * Then, each bch_extent_crc32/64 applies to the pointers that follow after it,
 474  * until the next bch_extent_crc32/64.
 475  *
 476  * If there are no bch_extent_crcs preceding a bch_extent_ptr, then that pointer
 477  * is neither checksummed nor compressed.
 478  */
 479
 480 /* 128 bits, sufficient for cryptographic MACs: */
 481 struct bch_csum {
 482         __le64                  lo;
 483         __le64                  hi;
 484 } __packed __aligned(8);
 485
 486 #define BCH_EXTENT_ENTRY_TYPES()                \
 487         x(ptr,                  0)              \
 488         x(crc32,                1)              \
 489         x(crc64,                2)              \
 490         x(crc128,               3)              \
 491         x(stripe_ptr,           4)
 492 #define BCH_EXTENT_ENTRY_MAX    5
 493
 494 enum bch_extent_entry_type {
 495 #define x(f, n) BCH_EXTENT_ENTRY_##f = n,
 496         BCH_EXTENT_ENTRY_TYPES()
 497 #undef x
 498 };
 499
 500 /* Compressed/uncompressed size are stored biased by 1: */
 501 struct bch_extent_crc32 {
 502 #if defined(__LITTLE_ENDIAN_BITFIELD)
 503         __u32                   type:2,
 504                                 _compressed_size:7,
 505                                 _uncompressed_size:7,
 506                                 offset:7,
 507                                 _unused:1,
 508                                 csum_type:4,
 509                                 compression_type:4;
 510         __u32                   csum;
 511 #elif defined (__BIG_ENDIAN_BITFIELD)
 512         __u32                   csum;
 513         __u32                   compression_type:4,
 514                                 csum_type:4,
 515                                 _unused:1,
 516                                 offset:7,
 517                                 _uncompressed_size:7,
 518                                 _compressed_size:7,
 519                                 type:2;
 520 #endif
 521 } __packed __aligned(8);
 522
 523 #define CRC32_SIZE_MAX          (1U << 7)
 524 #define CRC32_NONCE_MAX         0
 525
 526 struct bch_extent_crc64 {
 527 #if defined(__LITTLE_ENDIAN_BITFIELD)
 528         __u64                   type:3,
 529                                 _compressed_size:9,
 530                                 _uncompressed_size:9,
 531                                 offset:9,
 532                                 nonce:10,
 533                                 csum_type:4,
 534                                 compression_type:4,
 535                                 csum_hi:16;
 536 #elif defined (__BIG_ENDIAN_BITFIELD)
 537         __u64                   csum_hi:16,
 538                                 compression_type:4,
 539                                 csum_type:4,
 540                                 nonce:10,
 541                                 offset:9,
 542                                 _uncompressed_size:9,
 543                                 _compressed_size:9,
 544                                 type:3;
 545 #endif
 546         __u64                   csum_lo;
 547 } __packed __aligned(8);
 548
 549 #define CRC64_SIZE_MAX          (1U << 9)
 550 #define CRC64_NONCE_MAX         ((1U << 10) - 1)
 551
 552 struct bch_extent_crc128 {
 553 #if defined(__LITTLE_ENDIAN_BITFIELD)
 554         __u64                   type:4,
 555                                 _compressed_size:13,
 556                                 _uncompressed_size:13,
 557                                 offset:13,
 558                                 nonce:13,
 559                                 csum_type:4,
 560                                 compression_type:4;
 561 #elif defined (__BIG_ENDIAN_BITFIELD)
 562         __u64                   compression_type:4,
 563                                 csum_type:4,
 564                                 nonce:13,
 565                                 offset:13,
 566                                 _uncompressed_size:13,
 567                                 _compressed_size:13,
 568                                 type:4;
 569 #endif
 570         struct bch_csum         csum;
 571 } __packed __aligned(8);
 572
 573 #define CRC128_SIZE_MAX         (1U << 13)
 574 #define CRC128_NONCE_MAX        ((1U << 13) - 1)
 575
 576 /*
 577  * @reservation - pointer hasn't been written to, just reserved
 578  */
 579 struct bch_extent_ptr {
 580 #if defined(__LITTLE_ENDIAN_BITFIELD)
 581         __u64                   type:1,
 582                                 cached:1,
 583                                 unused:1,
 584                                 unwritten:1,
 585                                 offset:44, /* 8 petabytes */
 586                                 dev:8,
 587                                 gen:8;
 588 #elif defined (__BIG_ENDIAN_BITFIELD)
 589         __u64                   gen:8,
 590                                 dev:8,
 591                                 offset:44,
 592                                 unwritten:1,
 593                                 unused:1,
 594                                 cached:1,
 595                                 type:1;
 596 #endif
 597 } __packed __aligned(8);
 598
 599 struct bch_extent_stripe_ptr {
 600 #if defined(__LITTLE_ENDIAN_BITFIELD)
 601         __u64                   type:5,
 602                                 block:8,
 603                                 redundancy:4,
 604                                 idx:47;
 605 #elif defined (__BIG_ENDIAN_BITFIELD)
 606         __u64                   idx:47,
 607                                 redundancy:4,
 608                                 block:8,
 609                                 type:5;
 610 #endif
 611 };
 612
 613 struct bch_extent_reservation {
 614 #if defined(__LITTLE_ENDIAN_BITFIELD)
 615         __u64                   type:6,
 616                                 unused:22,
 617                                 replicas:4,
 618                                 generation:32;
 619 #elif defined (__BIG_ENDIAN_BITFIELD)
 620         __u64                   generation:32,
 621                                 replicas:4,
 622                                 unused:22,
 623                                 type:6;
 624 #endif
 625 };
 626
 627 union bch_extent_entry {
 628 #if __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__ ||  __BITS_PER_LONG == 64
 629         unsigned long                   type;
 630 #elif __BITS_PER_LONG == 32
 631         struct {
 632                 unsigned long           pad;
 633                 unsigned long           type;
 634         };
 635 #else
 636 #error edit for your odd byteorder.
 637 #endif
 638
 639 #define x(f, n) struct bch_extent_##f   f;
 640         BCH_EXTENT_ENTRY_TYPES()
 641 #undef x
 642 };
 643
 644 struct bch_btree_ptr {
 645         struct bch_val          v;
 646
 647         __u64                   _data[0];
 648         struct bch_extent_ptr   start[];
 649 } __packed __aligned(8);
 650
 651 struct bch_btree_ptr_v2 {
 652         struct bch_val          v;
 653
 654         __u64                   mem_ptr;
 655         __le64                  seq;
 656         __le16                  sectors_written;
 657         __le16                  flags;
 658         struct bpos             min_key;
 659         __u64                   _data[0];
 660         struct bch_extent_ptr   start[];
 661 } __packed __aligned(8);
 662
 663 LE16_BITMASK(BTREE_PTR_RANGE_UPDATED,   struct bch_btree_ptr_v2, flags, 0, 1);
 664
 665 struct bch_extent {
 666         struct bch_val          v;
 667
 668         __u64                   _data[0];
 669         union bch_extent_entry  start[];
 670 } __packed __aligned(8);
 671
 672 struct bch_reservation {
 673         struct bch_val          v;
 674
 675         __le32                  generation;
 676         __u8                    nr_replicas;
 677         __u8                    pad[3];
 678 } __packed __aligned(8);
 679
 680 /* Maximum size (in u64s) a single pointer could be: */
 681 #define BKEY_EXTENT_PTR_U64s_MAX\
 682         ((sizeof(struct bch_extent_crc128) +                    \
 683           sizeof(struct bch_extent_ptr)) / sizeof(u64))
 684
 685 /* Maximum possible size of an entire extent value: */
 686 #define BKEY_EXTENT_VAL_U64s_MAX                                \
 687         (1 + BKEY_EXTENT_PTR_U64s_MAX * (BCH_REPLICAS_MAX + 1))
 688
 689 /* * Maximum possible size of an entire extent, key + value: */
 690 #define BKEY_EXTENT_U64s_MAX            (BKEY_U64s + BKEY_EXTENT_VAL_U64s_MAX)
 691
 692 /* Btree pointers don't carry around checksums: */
 693 #define BKEY_BTREE_PTR_VAL_U64s_MAX                             \
 694         ((sizeof(struct bch_btree_ptr_v2) +                     \
 695           sizeof(struct bch_extent_ptr) * BCH_REPLICAS_MAX) / sizeof(u64))
 696 #define BKEY_BTREE_PTR_U64s_MAX                                 \
 697         (BKEY_U64s + BKEY_BTREE_PTR_VAL_U64s_MAX)
 698
 699 /* Inodes */
 700
 701 #define BLOCKDEV_INODE_MAX      4096
 702
 703 #define BCACHEFS_ROOT_INO       4096
 704
 705 struct bch_inode {
 706         struct bch_val          v;
 707
 708         __le64                  bi_hash_seed;
 709         __le32                  bi_flags;
 710         __le16                  bi_mode;
 711         __u8                    fields[0];
 712 } __packed __aligned(8);
 713
 714 struct bch_inode_v2 {
 715         struct bch_val          v;
 716
 717         __le64                  bi_journal_seq;
 718         __le64                  bi_hash_seed;
 719         __le64                  bi_flags;
 720         __le16                  bi_mode;
 721         __u8                    fields[0];
 722 } __packed __aligned(8);
 723
 724 struct bch_inode_v3 {
 725         struct bch_val          v;
 726
 727         __le64                  bi_journal_seq;
 728         __le64                  bi_hash_seed;
 729         __le64                  bi_flags;
 730         __le64                  bi_sectors;
 731         __le64                  bi_size;
 732         __le64                  bi_version;
 733         __u8                    fields[0];
 734 } __packed __aligned(8);
 735
 736 #define INODEv3_FIELDS_START_INITIAL    6
 737 #define INODEv3_FIELDS_START_CUR        (offsetof(struct bch_inode_v3, fields) / sizeof(u64))
 738
 739 struct bch_inode_generation {
 740         struct bch_val          v;
 741
 742         __le32                  bi_generation;
 743         __le32                  pad;
 744 } __packed __aligned(8);
 745
 746 /*
 747  * bi_subvol and bi_parent_subvol are only set for subvolume roots:
 748  */
 749
 750 #define BCH_INODE_FIELDS_v2()                   \
 751         x(bi_atime,                     96)     \
 752         x(bi_ctime,                     96)     \
 753         x(bi_mtime,                     96)     \
 754         x(bi_otime,                     96)     \
 755         x(bi_size,                      64)     \
 756         x(bi_sectors,                   64)     \
 757         x(bi_uid,                       32)     \
 758         x(bi_gid,                       32)     \
 759         x(bi_nlink,                     32)     \
 760         x(bi_generation,                32)     \
 761         x(bi_dev,                       32)     \
 762         x(bi_data_checksum,             8)      \
 763         x(bi_compression,               8)      \
 764         x(bi_project,                   32)     \
 765         x(bi_background_compression,    8)      \
 766         x(bi_data_replicas,             8)      \
 767         x(bi_promote_target,            16)     \
 768         x(bi_foreground_target,         16)     \
 769         x(bi_background_target,         16)     \
 770         x(bi_erasure_code,              16)     \
 771         x(bi_fields_set,                16)     \
 772         x(bi_dir,                       64)     \
 773         x(bi_dir_offset,                64)     \
 774         x(bi_subvol,                    32)     \
 775         x(bi_parent_subvol,             32)
 776
 777 #define BCH_INODE_FIELDS_v3()                   \
 778         x(bi_atime,                     96)     \
 779         x(bi_ctime,                     96)     \
 780         x(bi_mtime,                     96)     \
 781         x(bi_otime,                     96)     \
 782         x(bi_uid,                       32)     \
 783         x(bi_gid,                       32)     \
 784         x(bi_nlink,                     32)     \
 785         x(bi_generation,                32)     \
 786         x(bi_dev,                       32)     \
 787         x(bi_data_checksum,             8)      \
 788         x(bi_compression,               8)      \
 789         x(bi_project,                   32)     \
 790         x(bi_background_compression,    8)      \
 791         x(bi_data_replicas,             8)      \
 792         x(bi_promote_target,            16)     \
 793         x(bi_foreground_target,         16)     \
 794         x(bi_background_target,         16)     \
 795         x(bi_erasure_code,              16)     \
 796         x(bi_fields_set,                16)     \
 797         x(bi_dir,                       64)     \
 798         x(bi_dir_offset,                64)     \
 799         x(bi_subvol,                    32)     \
 800         x(bi_parent_subvol,             32)     \
 801         x(bi_nocow,                     8)
 802
 803 /* subset of BCH_INODE_FIELDS */
 804 #define BCH_INODE_OPTS()                        \
 805         x(data_checksum,                8)      \
 806         x(compression,                  8)      \
 807         x(project,                      32)     \
 808         x(background_compression,       8)      \
 809         x(data_replicas,                8)      \
 810         x(promote_target,               16)     \
 811         x(foreground_target,            16)     \
 812         x(background_target,            16)     \
 813         x(erasure_code,                 16)     \
 814         x(nocow,                        8)
 815
 816 enum inode_opt_id {
 817 #define x(name, ...)                            \
 818         Inode_opt_##name,
 819         BCH_INODE_OPTS()
 820 #undef  x
 821         Inode_opt_nr,
 822 };
 823
 824 enum {
 825         /*
 826          * User flags (get/settable with FS_IOC_*FLAGS, correspond to FS_*_FL
 827          * flags)
 828          */
 829         __BCH_INODE_SYNC                = 0,
 830         __BCH_INODE_IMMUTABLE           = 1,
 831         __BCH_INODE_APPEND              = 2,
 832         __BCH_INODE_NODUMP              = 3,
 833         __BCH_INODE_NOATIME             = 4,
 834
 835         __BCH_INODE_I_SIZE_DIRTY        = 5,
 836         __BCH_INODE_I_SECTORS_DIRTY     = 6,
 837         __BCH_INODE_UNLINKED            = 7,
 838         __BCH_INODE_BACKPTR_UNTRUSTED   = 8,
 839
 840         /* bits 20+ reserved for packed fields below: */
 841 };
 842
 843 #define BCH_INODE_SYNC          (1 << __BCH_INODE_SYNC)
 844 #define BCH_INODE_IMMUTABLE     (1 << __BCH_INODE_IMMUTABLE)
 845 #define BCH_INODE_APPEND        (1 << __BCH_INODE_APPEND)
 846 #define BCH_INODE_NODUMP        (1 << __BCH_INODE_NODUMP)
 847 #define BCH_INODE_NOATIME       (1 << __BCH_INODE_NOATIME)
 848 #define BCH_INODE_I_SIZE_DIRTY  (1 << __BCH_INODE_I_SIZE_DIRTY)
 849 #define BCH_INODE_I_SECTORS_DIRTY (1 << __BCH_INODE_I_SECTORS_DIRTY)
 850 #define BCH_INODE_UNLINKED      (1 << __BCH_INODE_UNLINKED)
 851 #define BCH_INODE_BACKPTR_UNTRUSTED (1 << __BCH_INODE_BACKPTR_UNTRUSTED)
 852
 853 LE32_BITMASK(INODE_STR_HASH,    struct bch_inode, bi_flags, 20, 24);
 854 LE32_BITMASK(INODE_NR_FIELDS,   struct bch_inode, bi_flags, 24, 31);
 855 LE32_BITMASK(INODE_NEW_VARINT,  struct bch_inode, bi_flags, 31, 32);
 856
 857 LE64_BITMASK(INODEv2_STR_HASH,  struct bch_inode_v2, bi_flags, 20, 24);
 858 LE64_BITMASK(INODEv2_NR_FIELDS, struct bch_inode_v2, bi_flags, 24, 31);
 859
 860 LE64_BITMASK(INODEv3_STR_HASH,  struct bch_inode_v3, bi_flags, 20, 24);
 861 LE64_BITMASK(INODEv3_NR_FIELDS, struct bch_inode_v3, bi_flags, 24, 31);
 862
 863 LE64_BITMASK(INODEv3_FIELDS_START,
 864                                 struct bch_inode_v3, bi_flags, 31, 36);
 865 LE64_BITMASK(INODEv3_MODE,      struct bch_inode_v3, bi_flags, 36, 52);
 866
 867 /* Dirents */
 868
 869 /*
 870  * Dirents (and xattrs) have to implement string lookups; since our b-tree
 871  * doesn't support arbitrary length strings for the key, we instead index by a
 872  * 64 bit hash (currently truncated sha1) of the string, stored in the offset
 873  * field of the key - using linear probing to resolve hash collisions. This also
 874  * provides us with the readdir cookie posix requires.
 875  *
 876  * Linear probing requires us to use whiteouts for deletions, in the event of a
 877  * collision:
 878  */
 879
 880 struct bch_dirent {
 881         struct bch_val          v;
 882
 883         /* Target inode number: */
 884         union {
 885         __le64                  d_inum;
 886         struct {                /* DT_SUBVOL */
 887         __le32                  d_child_subvol;
 888         __le32                  d_parent_subvol;
 889         };
 890         };
 891
 892         /*
 893          * Copy of mode bits 12-15 from the target inode - so userspace can get
 894          * the filetype without having to do a stat()
 895          */
 896         __u8                    d_type;
 897
 898         __u8                    d_name[];
 899 } __packed __aligned(8);
 900
 901 #define DT_SUBVOL       16
 902 #define BCH_DT_MAX      17
 903
 904 #define BCH_NAME_MAX    ((unsigned) (U8_MAX * sizeof(u64) -             \
 905                          sizeof(struct bkey) -                          \
 906                          offsetof(struct bch_dirent, d_name)))
 907
 908 /* Xattrs */
 909
 910 #define KEY_TYPE_XATTR_INDEX_USER                       0
 911 #define KEY_TYPE_XATTR_INDEX_POSIX_ACL_ACCESS   1
 912 #define KEY_TYPE_XATTR_INDEX_POSIX_ACL_DEFAULT  2
 913 #define KEY_TYPE_XATTR_INDEX_TRUSTED                    3
 914 #define KEY_TYPE_XATTR_INDEX_SECURITY           4
 915
 916 struct bch_xattr {
 917         struct bch_val          v;
 918         __u8                    x_type;
 919         __u8                    x_name_len;
 920         __le16                  x_val_len;
 921         __u8                    x_name[];
 922 } __packed __aligned(8);
 923
 924 /* Bucket/allocation information: */
 925
 926 struct bch_alloc {
 927         struct bch_val          v;
 928         __u8                    fields;
 929         __u8                    gen;
 930         __u8                    data[];
 931 } __packed __aligned(8);
 932
 933 #define BCH_ALLOC_FIELDS_V1()                   \
 934         x(read_time,            16)             \
 935         x(write_time,           16)             \
 936         x(data_type,            8)              \
 937         x(dirty_sectors,        16)             \
 938         x(cached_sectors,       16)             \
 939         x(oldest_gen,           8)              \
 940         x(stripe,               32)             \
 941         x(stripe_redundancy,    8)
 942
 943 enum {
 944 #define x(name, _bits) BCH_ALLOC_FIELD_V1_##name,
 945         BCH_ALLOC_FIELDS_V1()
 946 #undef x
 947 };
 948
 949 struct bch_alloc_v2 {
 950         struct bch_val          v;
 951         __u8                    nr_fields;
 952         __u8                    gen;
 953         __u8                    oldest_gen;
 954         __u8                    data_type;
 955         __u8                    data[];
 956 } __packed __aligned(8);
 957
 958 #define BCH_ALLOC_FIELDS_V2()                   \
 959         x(read_time,            64)             \
 960         x(write_time,           64)             \
 961         x(dirty_sectors,        32)             \
 962         x(cached_sectors,       32)             \
 963         x(stripe,               32)             \
 964         x(stripe_redundancy,    8)
 965
 966 struct bch_alloc_v3 {
 967         struct bch_val          v;
 968         __le64                  journal_seq;
 969         __le32                  flags;
 970         __u8                    nr_fields;
 971         __u8                    gen;
 972         __u8                    oldest_gen;
 973         __u8                    data_type;
 974         __u8                    data[];
 975 } __packed __aligned(8);
 976
 977 LE32_BITMASK(BCH_ALLOC_V3_NEED_DISCARD,struct bch_alloc_v3, flags,  0,  1)
 978 LE32_BITMASK(BCH_ALLOC_V3_NEED_INC_GEN,struct bch_alloc_v3, flags,  1,  2)
 979
 980 struct bch_alloc_v4 {
 981         struct bch_val          v;
 982         __u64                   journal_seq;
 983         __u32                   flags;
 984         __u8                    gen;
 985         __u8                    oldest_gen;
 986         __u8                    data_type;
 987         __u8                    stripe_redundancy;
 988         __u32                   dirty_sectors;
 989         __u32                   cached_sectors;
 990         __u64                   io_time[2];
 991         __u32                   stripe;
 992         __u32                   nr_external_backpointers;
 993         __u64                   fragmentation_lru;
 994 } __packed __aligned(8);
 995
 996 #define BCH_ALLOC_V4_U64s_V0    6
 997 #define BCH_ALLOC_V4_U64s       (sizeof(struct bch_alloc_v4) / sizeof(u64))
 998
 999 BITMASK(BCH_ALLOC_V4_NEED_DISCARD,      struct bch_alloc_v4, flags,  0,  1)
1000 BITMASK(BCH_ALLOC_V4_NEED_INC_GEN,      struct bch_alloc_v4, flags,  1,  2)
1001 BITMASK(BCH_ALLOC_V4_BACKPOINTERS_START,struct bch_alloc_v4, flags,  2,  8)
1002 BITMASK(BCH_ALLOC_V4_NR_BACKPOINTERS,   struct bch_alloc_v4, flags,  8,  14)
1003
1004 #define BCH_ALLOC_V4_NR_BACKPOINTERS_MAX        40
1005
1006 struct bch_backpointer {
1007         struct bch_val          v;
1008         __u8                    btree_id;
1009         __u8                    level;
1010         __u8                    data_type;
1011         __u64                   bucket_offset:40;
1012         __u32                   bucket_len;
1013         struct bpos             pos;
1014 } __packed __aligned(8);
1015
1016 #define KEY_TYPE_BUCKET_GENS_BITS       8
1017 #define KEY_TYPE_BUCKET_GENS_NR         (1U << KEY_TYPE_BUCKET_GENS_BITS)
1018 #define KEY_TYPE_BUCKET_GENS_MASK       (KEY_TYPE_BUCKET_GENS_NR - 1)
1019
1020 struct bch_bucket_gens {
1021         struct bch_val          v;
1022         u8                      gens[KEY_TYPE_BUCKET_GENS_NR];
1023 } __packed __aligned(8);
1024
1025 /* Quotas: */
1026
1027 enum quota_types {
1028         QTYP_USR                = 0,
1029         QTYP_GRP                = 1,
1030         QTYP_PRJ                = 2,
1031         QTYP_NR                 = 3,
1032 };
1033
1034 enum quota_counters {
1035         Q_SPC                   = 0,
1036         Q_INO                   = 1,
1037         Q_COUNTERS              = 2,
1038 };
1039
1040 struct bch_quota_counter {
1041         __le64                  hardlimit;
1042         __le64                  softlimit;
1043 };
1044
1045 struct bch_quota {
1046         struct bch_val          v;
1047         struct bch_quota_counter c[Q_COUNTERS];
1048 } __packed __aligned(8);
1049
1050 /* Erasure coding */
1051
1052 struct bch_stripe {
1053         struct bch_val          v;
1054         __le16                  sectors;
1055         __u8                    algorithm;
1056         __u8                    nr_blocks;
1057         __u8                    nr_redundant;
1058
1059         __u8                    csum_granularity_bits;
1060         __u8                    csum_type;
1061         __u8                    pad;
1062
1063         struct bch_extent_ptr   ptrs[];
1064 } __packed __aligned(8);
1065
1066 /* Reflink: */
1067
1068 struct bch_reflink_p {
1069         struct bch_val          v;
1070         __le64                  idx;
1071         /*
1072          * A reflink pointer might point to an indirect extent which is then
1073          * later split (by copygc or rebalance). If we only pointed to part of
1074          * the original indirect extent, and then one of the fragments is
1075          * outside the range we point to, we'd leak a refcount: so when creating
1076          * reflink pointers, we need to store pad values to remember the full
1077          * range we were taking a reference on.
1078          */
1079         __le32                  front_pad;
1080         __le32                  back_pad;
1081 } __packed __aligned(8);
1082
1083 struct bch_reflink_v {
1084         struct bch_val          v;
1085         __le64                  refcount;
1086         union bch_extent_entry  start[0];
1087         __u64                   _data[0];
1088 } __packed __aligned(8);
1089
1090 struct bch_indirect_inline_data {
1091         struct bch_val          v;
1092         __le64                  refcount;
1093         u8                      data[0];
1094 };
1095
1096 /* Inline data */
1097
1098 struct bch_inline_data {
1099         struct bch_val          v;
1100         u8                      data[0];
1101 };
1102
1103 /* Subvolumes: */
1104
1105 #define SUBVOL_POS_MIN          POS(0, 1)
1106 #define SUBVOL_POS_MAX          POS(0, S32_MAX)
1107 #define BCACHEFS_ROOT_SUBVOL    1
1108
1109 struct bch_subvolume {
1110         struct bch_val          v;
1111         __le32                  flags;
1112         __le32                  snapshot;
1113         __le64                  inode;
1114         __le32                  parent;
1115         __le32                  pad;
1116         bch_le128               otime;
1117 };
1118
1119 LE32_BITMASK(BCH_SUBVOLUME_RO,          struct bch_subvolume, flags,  0,  1)
1120 /*
1121  * We need to know whether a subvolume is a snapshot so we can know whether we
1122  * can delete it (or whether it should just be rm -rf'd)
1123  */
1124 LE32_BITMASK(BCH_SUBVOLUME_SNAP,        struct bch_subvolume, flags,  1,  2)
1125 LE32_BITMASK(BCH_SUBVOLUME_UNLINKED,    struct bch_subvolume, flags,  2,  3)
1126
1127 /* Snapshots */
1128
1129 struct bch_snapshot {
1130         struct bch_val          v;
1131         __le32                  flags;
1132         __le32                  parent;
1133         __le32                  children[2];
1134         __le32                  subvol;
1135         __le32                  tree;
1136 };
1137
1138 LE32_BITMASK(BCH_SNAPSHOT_DELETED,      struct bch_snapshot, flags,  0,  1)
1139
1140 /* True if a subvolume points to this snapshot node: */
1141 LE32_BITMASK(BCH_SNAPSHOT_SUBVOL,       struct bch_snapshot, flags,  1,  2)
1142
1143 /*
1144  * Snapshot trees:
1145  *
1146  * The snapshot_trees btree gives us persistent indentifier for each tree of
1147  * bch_snapshot nodes, and allow us to record and easily find the root/master
1148  * subvolume that other snapshots were created from:
1149  */
1150 struct bch_snapshot_tree {
1151         struct bch_val          v;
1152         __le32                  master_subvol;
1153         __le32                  root_snapshot;
1154 };
1155
1156 /* LRU btree: */
1157
1158 struct bch_lru {
1159         struct bch_val          v;
1160         __le64                  idx;
1161 } __packed __aligned(8);
1162
1163 #define LRU_ID_STRIPES          (1U << 16)
1164
1165 /* Optional/variable size superblock sections: */
1166
1167 struct bch_sb_field {
1168         __u64                   _data[0];
1169         __le32                  u64s;
1170         __le32                  type;
1171 };
1172
1173 #define BCH_SB_FIELDS()                         \
1174         x(journal,      0)                      \
1175         x(members,      1)                      \
1176         x(crypt,        2)                      \
1177         x(replicas_v0,  3)                      \
1178         x(quota,        4)                      \
1179         x(disk_groups,  5)                      \
1180         x(clean,        6)                      \
1181         x(replicas,     7)                      \
1182         x(journal_seq_blacklist, 8)             \
1183         x(journal_v2,   9)                      \
1184         x(counters,     10)
1185
1186 enum bch_sb_field_type {
1187 #define x(f, nr)        BCH_SB_FIELD_##f = nr,
1188         BCH_SB_FIELDS()
1189 #undef x
1190         BCH_SB_FIELD_NR
1191 };
1192
1193 /*
1194  * Most superblock fields are replicated in all device's superblocks - a few are
1195  * not:
1196  */
1197 #define BCH_SINGLE_DEVICE_SB_FIELDS             \
1198         ((1U << BCH_SB_FIELD_journal)|          \
1199          (1U << BCH_SB_FIELD_journal_v2))
1200
1201 /* BCH_SB_FIELD_journal: */
1202
1203 struct bch_sb_field_journal {
1204         struct bch_sb_field     field;
1205         __le64                  buckets[0];
1206 };
1207
1208 struct bch_sb_field_journal_v2 {
1209         struct bch_sb_field     field;
1210
1211         struct bch_sb_field_journal_v2_entry {
1212                 __le64          start;
1213                 __le64          nr;
1214         }                       d[0];
1215 };
1216
1217 /* BCH_SB_FIELD_members: */
1218
1219 #define BCH_MIN_NR_NBUCKETS     (1 << 6)
1220
1221 struct bch_member {
1222         __uuid_t                uuid;
1223         __le64                  nbuckets;       /* device size */
1224         __le16                  first_bucket;   /* index of first bucket used */
1225         __le16                  bucket_size;    /* sectors */
1226         __le32                  pad;
1227         __le64                  last_mount;     /* time_t */
1228
1229         __le64                  flags[2];
1230 };
1231
1232 LE64_BITMASK(BCH_MEMBER_STATE,          struct bch_member, flags[0],  0,  4)
1233 /* 4-14 unused, was TIER, HAS_(META)DATA, REPLACEMENT */
1234 LE64_BITMASK(BCH_MEMBER_DISCARD,        struct bch_member, flags[0], 14, 15)
1235 LE64_BITMASK(BCH_MEMBER_DATA_ALLOWED,   struct bch_member, flags[0], 15, 20)
1236 LE64_BITMASK(BCH_MEMBER_GROUP,          struct bch_member, flags[0], 20, 28)
1237 LE64_BITMASK(BCH_MEMBER_DURABILITY,     struct bch_member, flags[0], 28, 30)
1238 LE64_BITMASK(BCH_MEMBER_FREESPACE_INITIALIZED,
1239                                         struct bch_member, flags[0], 30, 31)
1240
1241 #if 0
1242 LE64_BITMASK(BCH_MEMBER_NR_READ_ERRORS, struct bch_member, flags[1], 0,  20);
1243 LE64_BITMASK(BCH_MEMBER_NR_WRITE_ERRORS,struct bch_member, flags[1], 20, 40);
1244 #endif
1245
1246 #define BCH_MEMBER_STATES()                     \
1247         x(rw,           0)                      \
1248         x(ro,           1)                      \
1249         x(failed,       2)                      \
1250         x(spare,        3)
1251
1252 enum bch_member_state {
1253 #define x(t, n) BCH_MEMBER_STATE_##t = n,
1254         BCH_MEMBER_STATES()
1255 #undef x
1256         BCH_MEMBER_STATE_NR
1257 };
1258
1259 struct bch_sb_field_members {
1260         struct bch_sb_field     field;
1261         struct bch_member       members[0];
1262 };
1263
1264 /* BCH_SB_FIELD_crypt: */
1265
1266 struct nonce {
1267         __le32                  d[4];
1268 };
1269
1270 struct bch_key {
1271         __le64                  key[4];
1272 };
1273
1274 #define BCH_KEY_MAGIC                                   \
1275         (((u64) 'b' <<  0)|((u64) 'c' <<  8)|           \
1276          ((u64) 'h' << 16)|((u64) '*' << 24)|           \
1277          ((u64) '*' << 32)|((u64) 'k' << 40)|           \
1278          ((u64) 'e' << 48)|((u64) 'y' << 56))
1279
1280 struct bch_encrypted_key {
1281         __le64                  magic;
1282         struct bch_key          key;
1283 };
1284
1285 /*
1286  * If this field is present in the superblock, it stores an encryption key which
1287  * is used encrypt all other data/metadata. The key will normally be encrypted
1288  * with the key userspace provides, but if encryption has been turned off we'll
1289  * just store the master key unencrypted in the superblock so we can access the
1290  * previously encrypted data.
1291  */
1292 struct bch_sb_field_crypt {
1293         struct bch_sb_field     field;
1294
1295         __le64                  flags;
1296         __le64                  kdf_flags;
1297         struct bch_encrypted_key key;
1298 };
1299
1300 LE64_BITMASK(BCH_CRYPT_KDF_TYPE,        struct bch_sb_field_crypt, flags, 0, 4);
1301
1302 enum bch_kdf_types {
1303         BCH_KDF_SCRYPT          = 0,
1304         BCH_KDF_NR              = 1,
1305 };
1306
1307 /* stored as base 2 log of scrypt params: */
1308 LE64_BITMASK(BCH_KDF_SCRYPT_N,  struct bch_sb_field_crypt, kdf_flags,  0, 16);
1309 LE64_BITMASK(BCH_KDF_SCRYPT_R,  struct bch_sb_field_crypt, kdf_flags, 16, 32);
1310 LE64_BITMASK(BCH_KDF_SCRYPT_P,  struct bch_sb_field_crypt, kdf_flags, 32, 48);
1311
1312 /* BCH_SB_FIELD_replicas: */
1313
1314 #define BCH_DATA_TYPES()                \
1315         x(free,         0)              \
1316         x(sb,           1)              \
1317         x(journal,      2)              \
1318         x(btree,        3)              \
1319         x(user,         4)              \
1320         x(cached,       5)              \
1321         x(parity,       6)              \
1322         x(stripe,       7)              \
1323         x(need_gc_gens, 8)              \
1324         x(need_discard, 9)
1325
1326 enum bch_data_type {
1327 #define x(t, n) BCH_DATA_##t,
1328         BCH_DATA_TYPES()
1329 #undef x
1330         BCH_DATA_NR
1331 };
1332
1333 static inline bool data_type_is_empty(enum bch_data_type type)
1334 {
1335         switch (type) {
1336         case BCH_DATA_free:
1337         case BCH_DATA_need_gc_gens:
1338         case BCH_DATA_need_discard:
1339                 return true;
1340         default:
1341                 return false;
1342         }
1343 }
1344
1345 static inline bool data_type_is_hidden(enum bch_data_type type)
1346 {
1347         switch (type) {
1348         case BCH_DATA_sb:
1349         case BCH_DATA_journal:
1350                 return true;
1351         default:
1352                 return false;
1353         }
1354 }
1355
1356 struct bch_replicas_entry_v0 {
1357         __u8                    data_type;
1358         __u8                    nr_devs;
1359         __u8                    devs[];
1360 } __packed;
1361
1362 struct bch_sb_field_replicas_v0 {
1363         struct bch_sb_field     field;
1364         struct bch_replicas_entry_v0 entries[];
1365 } __packed __aligned(8);
1366
1367 struct bch_replicas_entry {
1368         __u8                    data_type;
1369         __u8                    nr_devs;
1370         __u8                    nr_required;
1371         __u8                    devs[];
1372 } __packed;
1373
1374 #define replicas_entry_bytes(_i)                                        \
1375         (offsetof(typeof(*(_i)), devs) + (_i)->nr_devs)
1376
1377 struct bch_sb_field_replicas {
1378         struct bch_sb_field     field;
1379         struct bch_replicas_entry entries[];
1380 } __packed __aligned(8);
1381
1382 /* BCH_SB_FIELD_quota: */
1383
1384 struct bch_sb_quota_counter {
1385         __le32                          timelimit;
1386         __le32                          warnlimit;
1387 };
1388
1389 struct bch_sb_quota_type {
1390         __le64                          flags;
1391         struct bch_sb_quota_counter     c[Q_COUNTERS];
1392 };
1393
1394 struct bch_sb_field_quota {
1395         struct bch_sb_field             field;
1396         struct bch_sb_quota_type        q[QTYP_NR];
1397 } __packed __aligned(8);
1398
1399 /* BCH_SB_FIELD_disk_groups: */
1400
1401 #define BCH_SB_LABEL_SIZE               32
1402
1403 struct bch_disk_group {
1404         __u8                    label[BCH_SB_LABEL_SIZE];
1405         __le64                  flags[2];
1406 } __packed __aligned(8);
1407
1408 LE64_BITMASK(BCH_GROUP_DELETED,         struct bch_disk_group, flags[0], 0,  1)
1409 LE64_BITMASK(BCH_GROUP_DATA_ALLOWED,    struct bch_disk_group, flags[0], 1,  6)
1410 LE64_BITMASK(BCH_GROUP_PARENT,          struct bch_disk_group, flags[0], 6, 24)
1411
1412 struct bch_sb_field_disk_groups {
1413         struct bch_sb_field     field;
1414         struct bch_disk_group   entries[0];
1415 } __packed __aligned(8);
1416
1417 /* BCH_SB_FIELD_counters */
1418
1419 #define BCH_PERSISTENT_COUNTERS()                               \
1420         x(io_read,                                      0)      \
1421         x(io_write,                                     1)      \
1422         x(io_move,                                      2)      \
1423         x(bucket_invalidate,                            3)      \
1424         x(bucket_discard,                               4)      \
1425         x(bucket_alloc,                                 5)      \
1426         x(bucket_alloc_fail,                            6)      \
1427         x(btree_cache_scan,                             7)      \
1428         x(btree_cache_reap,                             8)      \
1429         x(btree_cache_cannibalize,                      9)      \
1430         x(btree_cache_cannibalize_lock,                 10)     \
1431         x(btree_cache_cannibalize_lock_fail,            11)     \
1432         x(btree_cache_cannibalize_unlock,               12)     \
1433         x(btree_node_write,                             13)     \
1434         x(btree_node_read,                              14)     \
1435         x(btree_node_compact,                           15)     \
1436         x(btree_node_merge,                             16)     \
1437         x(btree_node_split,                             17)     \
1438         x(btree_node_rewrite,                           18)     \
1439         x(btree_node_alloc,                             19)     \
1440         x(btree_node_free,                              20)     \
1441         x(btree_node_set_root,                          21)     \
1442         x(btree_path_relock_fail,                       22)     \
1443         x(btree_path_upgrade_fail,                      23)     \
1444         x(btree_reserve_get_fail,                       24)     \
1445         x(journal_entry_full,                           25)     \
1446         x(journal_full,                                 26)     \
1447         x(journal_reclaim_finish,                       27)     \
1448         x(journal_reclaim_start,                        28)     \
1449         x(journal_write,                                29)     \
1450         x(read_promote,                                 30)     \
1451         x(read_bounce,                                  31)     \
1452         x(read_split,                                   33)     \
1453         x(read_retry,                                   32)     \
1454         x(read_reuse_race,                              34)     \
1455         x(move_extent_read,                             35)     \
1456         x(move_extent_write,                            36)     \
1457         x(move_extent_finish,                           37)     \
1458         x(move_extent_fail,                             38)     \
1459         x(move_extent_alloc_mem_fail,                   39)     \
1460         x(copygc,                                       40)     \
1461         x(copygc_wait,                                  41)     \
1462         x(gc_gens_end,                                  42)     \
1463         x(gc_gens_start,                                43)     \
1464         x(trans_blocked_journal_reclaim,                44)     \
1465         x(trans_restart_btree_node_reused,              45)     \
1466         x(trans_restart_btree_node_split,               46)     \
1467         x(trans_restart_fault_inject,                   47)     \
1468         x(trans_restart_iter_upgrade,                   48)     \
1469         x(trans_restart_journal_preres_get,             49)     \
1470         x(trans_restart_journal_reclaim,                50)     \
1471         x(trans_restart_journal_res_get,                51)     \
1472         x(trans_restart_key_cache_key_realloced,        52)     \
1473         x(trans_restart_key_cache_raced,                53)     \
1474         x(trans_restart_mark_replicas,                  54)     \
1475         x(trans_restart_mem_realloced,                  55)     \
1476         x(trans_restart_memory_allocation_failure,      56)     \
1477         x(trans_restart_relock,                         57)     \
1478         x(trans_restart_relock_after_fill,              58)     \
1479         x(trans_restart_relock_key_cache_fill,          59)     \
1480         x(trans_restart_relock_next_node,               60)     \
1481         x(trans_restart_relock_parent_for_fill,         61)     \
1482         x(trans_restart_relock_path,                    62)     \
1483         x(trans_restart_relock_path_intent,             63)     \
1484         x(trans_restart_too_many_iters,                 64)     \
1485         x(trans_restart_traverse,                       65)     \
1486         x(trans_restart_upgrade,                        66)     \
1487         x(trans_restart_would_deadlock,                 67)     \
1488         x(trans_restart_would_deadlock_write,           68)     \
1489         x(trans_restart_injected,                       69)     \
1490         x(trans_restart_key_cache_upgrade,              70)     \
1491         x(trans_traverse_all,                           71)     \
1492         x(transaction_commit,                           72)     \
1493         x(write_super,                                  73)     \
1494         x(trans_restart_would_deadlock_recursion_limit, 74)     \
1495         x(trans_restart_write_buffer_flush,             75)     \
1496         x(trans_restart_split_race,                     76)
1497
1498 enum bch_persistent_counters {
1499 #define x(t, n, ...) BCH_COUNTER_##t,
1500         BCH_PERSISTENT_COUNTERS()
1501 #undef x
1502         BCH_COUNTER_NR
1503 };
1504
1505 struct bch_sb_field_counters {
1506         struct bch_sb_field     field;
1507         __le64                  d[0];
1508 };
1509
1510 /*
1511  * On clean shutdown, store btree roots and current journal sequence number in
1512  * the superblock:
1513  */
1514 struct jset_entry {
1515         __le16                  u64s;
1516         __u8                    btree_id;
1517         __u8                    level;
1518         __u8                    type; /* designates what this jset holds */
1519         __u8                    pad[3];
1520
1521         union {
1522                 struct bkey_i   start[0];
1523                 __u64           _data[0];
1524         };
1525 };
1526
1527 struct bch_sb_field_clean {
1528         struct bch_sb_field     field;
1529
1530         __le32                  flags;
1531         __le16                  _read_clock; /* no longer used */
1532         __le16                  _write_clock;
1533         __le64                  journal_seq;
1534
1535         union {
1536                 struct jset_entry start[0];
1537                 __u64           _data[0];
1538         };
1539 };
1540
1541 struct journal_seq_blacklist_entry {
1542         __le64                  start;
1543         __le64                  end;
1544 };
1545
1546 struct bch_sb_field_journal_seq_blacklist {
1547         struct bch_sb_field     field;
1548
1549         union {
1550                 struct journal_seq_blacklist_entry start[0];
1551                 __u64           _data[0];
1552         };
1553 };
1554
1555 /* Superblock: */
1556
1557 /*
1558  * New versioning scheme:
1559  * One common version number for all on disk data structures - superblock, btree
1560  * nodes, journal entries
1561  */
1562 #define BCH_JSET_VERSION_OLD                    2
1563 #define BCH_BSET_VERSION_OLD                    3
1564
1565 #define BCH_METADATA_VERSIONS()                         \
1566         x(bkey_renumber,                10)             \
1567         x(inode_btree_change,           11)             \
1568         x(snapshot,                     12)             \
1569         x(inode_backpointers,           13)             \
1570         x(btree_ptr_sectors_written,    14)             \
1571         x(snapshot_2,                   15)             \
1572         x(reflink_p_fix,                16)             \
1573         x(subvol_dirent,                17)             \
1574         x(inode_v2,                     18)             \
1575         x(freespace,                    19)             \
1576         x(alloc_v4,                     20)             \
1577         x(new_data_types,               21)             \
1578         x(backpointers,                 22)             \
1579         x(inode_v3,                     23)             \
1580         x(unwritten_extents,            24)             \
1581         x(bucket_gens,                  25)             \
1582         x(lru_v2,                       26)             \
1583         x(fragmentation_lru,            27)             \
1584         x(no_bps_in_alloc_keys,         28)             \
1585         x(snapshot_trees,               29)
1586
1587 enum bcachefs_metadata_version {
1588         bcachefs_metadata_version_min = 9,
1589 #define x(t, n) bcachefs_metadata_version_##t = n,
1590         BCH_METADATA_VERSIONS()
1591 #undef x
1592         bcachefs_metadata_version_max
1593 };
1594
1595 static const unsigned bcachefs_metadata_required_upgrade_below = bcachefs_metadata_version_snapshot_trees;
1596
1597 #define bcachefs_metadata_version_current       (bcachefs_metadata_version_max - 1)
1598
1599 #define BCH_SB_SECTOR                   8
1600 #define BCH_SB_MEMBERS_MAX              64 /* XXX kill */
1601
1602 struct bch_sb_layout {
1603         __uuid_t                magic;  /* bcachefs superblock UUID */
1604         __u8                    layout_type;
1605         __u8                    sb_max_size_bits; /* base 2 of 512 byte sectors */
1606         __u8                    nr_superblocks;
1607         __u8                    pad[5];
1608         __le64                  sb_offset[61];
1609 } __packed __aligned(8);
1610
1611 #define BCH_SB_LAYOUT_SECTOR    7
1612
1613 /*
1614  * @offset      - sector where this sb was written
1615  * @version     - on disk format version
1616  * @version_min - Oldest metadata version this filesystem contains; so we can
1617  *                safely drop compatibility code and refuse to mount filesystems
1618  *                we'd need it for
1619  * @magic       - identifies as a bcachefs superblock (BCHFS_MAGIC)
1620  * @seq         - incremented each time superblock is written
1621  * @uuid        - used for generating various magic numbers and identifying
1622  *                member devices, never changes
1623  * @user_uuid   - user visible UUID, may be changed
1624  * @label       - filesystem label
1625  * @seq         - identifies most recent superblock, incremented each time
1626  *                superblock is written
1627  * @features    - enabled incompatible features
1628  */
1629 struct bch_sb {
1630         struct bch_csum         csum;
1631         __le16                  version;
1632         __le16                  version_min;
1633         __le16                  pad[2];
1634         __uuid_t                magic;
1635         __uuid_t                uuid;
1636         __uuid_t                user_uuid;
1637         __u8                    label[BCH_SB_LABEL_SIZE];
1638         __le64                  offset;
1639         __le64                  seq;
1640
1641         __le16                  block_size;
1642         __u8                    dev_idx;
1643         __u8                    nr_devices;
1644         __le32                  u64s;
1645
1646         __le64                  time_base_lo;
1647         __le32                  time_base_hi;
1648         __le32                  time_precision;
1649
1650         __le64                  flags[8];
1651         __le64                  features[2];
1652         __le64                  compat[2];
1653
1654         struct bch_sb_layout    layout;
1655
1656         union {
1657                 struct bch_sb_field start[0];
1658                 __le64          _data[0];
1659         };
1660 } __packed __aligned(8);
1661
1662 /*
1663  * Flags:
1664  * BCH_SB_INITALIZED    - set on first mount
1665  * BCH_SB_CLEAN         - did we shut down cleanly? Just a hint, doesn't affect
1666  *                        behaviour of mount/recovery path:
1667  * BCH_SB_INODE_32BIT   - limit inode numbers to 32 bits
1668  * BCH_SB_128_BIT_MACS  - 128 bit macs instead of 80
1669  * BCH_SB_ENCRYPTION_TYPE - if nonzero encryption is enabled; overrides
1670  *                         DATA/META_CSUM_TYPE. Also indicates encryption
1671  *                         algorithm in use, if/when we get more than one
1672  */
1673
1674 LE16_BITMASK(BCH_SB_BLOCK_SIZE,         struct bch_sb, block_size, 0, 16);
1675
1676 LE64_BITMASK(BCH_SB_INITIALIZED,        struct bch_sb, flags[0],  0,  1);
1677 LE64_BITMASK(BCH_SB_CLEAN,              struct bch_sb, flags[0],  1,  2);
1678 LE64_BITMASK(BCH_SB_CSUM_TYPE,          struct bch_sb, flags[0],  2,  8);
1679 LE64_BITMASK(BCH_SB_ERROR_ACTION,       struct bch_sb, flags[0],  8, 12);
1680
1681 LE64_BITMASK(BCH_SB_BTREE_NODE_SIZE,    struct bch_sb, flags[0], 12, 28);
1682
1683 LE64_BITMASK(BCH_SB_GC_RESERVE,         struct bch_sb, flags[0], 28, 33);
1684 LE64_BITMASK(BCH_SB_ROOT_RESERVE,       struct bch_sb, flags[0], 33, 40);
1685
1686 LE64_BITMASK(BCH_SB_META_CSUM_TYPE,     struct bch_sb, flags[0], 40, 44);
1687 LE64_BITMASK(BCH_SB_DATA_CSUM_TYPE,     struct bch_sb, flags[0], 44, 48);
1688
1689 LE64_BITMASK(BCH_SB_META_REPLICAS_WANT, struct bch_sb, flags[0], 48, 52);
1690 LE64_BITMASK(BCH_SB_DATA_REPLICAS_WANT, struct bch_sb, flags[0], 52, 56);
1691
1692 LE64_BITMASK(BCH_SB_POSIX_ACL,          struct bch_sb, flags[0], 56, 57);
1693 LE64_BITMASK(BCH_SB_USRQUOTA,           struct bch_sb, flags[0], 57, 58);
1694 LE64_BITMASK(BCH_SB_GRPQUOTA,           struct bch_sb, flags[0], 58, 59);
1695 LE64_BITMASK(BCH_SB_PRJQUOTA,           struct bch_sb, flags[0], 59, 60);
1696
1697 LE64_BITMASK(BCH_SB_HAS_ERRORS,         struct bch_sb, flags[0], 60, 61);
1698 LE64_BITMASK(BCH_SB_HAS_TOPOLOGY_ERRORS,struct bch_sb, flags[0], 61, 62);
1699
1700 LE64_BITMASK(BCH_SB_BIG_ENDIAN,         struct bch_sb, flags[0], 62, 63);
1701
1702 LE64_BITMASK(BCH_SB_STR_HASH_TYPE,      struct bch_sb, flags[1],  0,  4);
1703 LE64_BITMASK(BCH_SB_COMPRESSION_TYPE,   struct bch_sb, flags[1],  4,  8);
1704 LE64_BITMASK(BCH_SB_INODE_32BIT,        struct bch_sb, flags[1],  8,  9);
1705
1706 LE64_BITMASK(BCH_SB_128_BIT_MACS,       struct bch_sb, flags[1],  9, 10);
1707 LE64_BITMASK(BCH_SB_ENCRYPTION_TYPE,    struct bch_sb, flags[1], 10, 14);
1708
1709 /*
1710  * Max size of an extent that may require bouncing to read or write
1711  * (checksummed, compressed): 64k
1712  */
1713 LE64_BITMASK(BCH_SB_ENCODED_EXTENT_MAX_BITS,
1714                                         struct bch_sb, flags[1], 14, 20);
1715
1716 LE64_BITMASK(BCH_SB_META_REPLICAS_REQ,  struct bch_sb, flags[1], 20, 24);
1717 LE64_BITMASK(BCH_SB_DATA_REPLICAS_REQ,  struct bch_sb, flags[1], 24, 28);
1718
1719 LE64_BITMASK(BCH_SB_PROMOTE_TARGET,     struct bch_sb, flags[1], 28, 40);
1720 LE64_BITMASK(BCH_SB_FOREGROUND_TARGET,  struct bch_sb, flags[1], 40, 52);
1721 LE64_BITMASK(BCH_SB_BACKGROUND_TARGET,  struct bch_sb, flags[1], 52, 64);
1722
1723 LE64_BITMASK(BCH_SB_BACKGROUND_COMPRESSION_TYPE,
1724                                         struct bch_sb, flags[2],  0,  4);
1725 LE64_BITMASK(BCH_SB_GC_RESERVE_BYTES,   struct bch_sb, flags[2],  4, 64);
1726
1727 LE64_BITMASK(BCH_SB_ERASURE_CODE,       struct bch_sb, flags[3],  0, 16);
1728 LE64_BITMASK(BCH_SB_METADATA_TARGET,    struct bch_sb, flags[3], 16, 28);
1729 LE64_BITMASK(BCH_SB_SHARD_INUMS,        struct bch_sb, flags[3], 28, 29);
1730 LE64_BITMASK(BCH_SB_INODES_USE_KEY_CACHE,struct bch_sb, flags[3], 29, 30);
1731 LE64_BITMASK(BCH_SB_JOURNAL_FLUSH_DELAY,struct bch_sb, flags[3], 30, 62);
1732 LE64_BITMASK(BCH_SB_JOURNAL_FLUSH_DISABLED,struct bch_sb, flags[3], 62, 63);
1733 LE64_BITMASK(BCH_SB_JOURNAL_RECLAIM_DELAY,struct bch_sb, flags[4], 0, 32);
1734 LE64_BITMASK(BCH_SB_JOURNAL_TRANSACTION_NAMES,struct bch_sb, flags[4], 32, 33);
1735 LE64_BITMASK(BCH_SB_NOCOW,              struct bch_sb, flags[4], 33, 34);
1736 LE64_BITMASK(BCH_SB_WRITE_BUFFER_SIZE,  struct bch_sb, flags[4], 34, 54);
1737
1738 /*
1739  * Features:
1740  *
1741  * journal_seq_blacklist_v3:    gates BCH_SB_FIELD_journal_seq_blacklist
1742  * reflink:                     gates KEY_TYPE_reflink
1743  * inline_data:                 gates KEY_TYPE_inline_data
1744  * new_siphash:                 gates BCH_STR_HASH_siphash
1745  * new_extent_overwrite:        gates BTREE_NODE_NEW_EXTENT_OVERWRITE
1746  */
1747 #define BCH_SB_FEATURES()                       \
1748         x(lz4,                          0)      \
1749         x(gzip,                         1)      \
1750         x(zstd,                         2)      \
1751         x(atomic_nlink,                 3)      \
1752         x(ec,                           4)      \
1753         x(journal_seq_blacklist_v3,     5)      \
1754         x(reflink,                      6)      \
1755         x(new_siphash,                  7)      \
1756         x(inline_data,                  8)      \
1757         x(new_extent_overwrite,         9)      \
1758         x(incompressible,               10)     \
1759         x(btree_ptr_v2,                 11)     \
1760         x(extents_above_btree_updates,  12)     \
1761         x(btree_updates_journalled,     13)     \
1762         x(reflink_inline_data,          14)     \
1763         x(new_varint,                   15)     \
1764         x(journal_no_flush,             16)     \
1765         x(alloc_v2,                     17)     \
1766         x(extents_across_btree_nodes,   18)
1767
1768 #define BCH_SB_FEATURES_ALWAYS                          \
1769         ((1ULL << BCH_FEATURE_new_extent_overwrite)|    \
1770          (1ULL << BCH_FEATURE_extents_above_btree_updates)|\
1771          (1ULL << BCH_FEATURE_btree_updates_journalled)|\
1772          (1ULL << BCH_FEATURE_alloc_v2)|\
1773          (1ULL << BCH_FEATURE_extents_across_btree_nodes))
1774
1775 #define BCH_SB_FEATURES_ALL                             \
1776         (BCH_SB_FEATURES_ALWAYS|                        \
1777          (1ULL << BCH_FEATURE_new_siphash)|             \
1778          (1ULL << BCH_FEATURE_btree_ptr_v2)|            \
1779          (1ULL << BCH_FEATURE_new_varint)|              \
1780          (1ULL << BCH_FEATURE_journal_no_flush))
1781
1782 enum bch_sb_feature {
1783 #define x(f, n) BCH_FEATURE_##f,
1784         BCH_SB_FEATURES()
1785 #undef x
1786         BCH_FEATURE_NR,
1787 };
1788
1789 #define BCH_SB_COMPAT()                                 \
1790         x(alloc_info,                           0)      \
1791         x(alloc_metadata,                       1)      \
1792         x(extents_above_btree_updates_done,     2)      \
1793         x(bformat_overflow_done,                3)
1794
1795 enum bch_sb_compat {
1796 #define x(f, n) BCH_COMPAT_##f,
1797         BCH_SB_COMPAT()
1798 #undef x
1799         BCH_COMPAT_NR,
1800 };
1801
1802 /* options: */
1803
1804 #define BCH_REPLICAS_MAX                4U
1805
1806 #define BCH_BKEY_PTRS_MAX               16U
1807
1808 #define BCH_ERROR_ACTIONS()             \
1809         x(continue,             0)      \
1810         x(ro,                   1)      \
1811         x(panic,                2)
1812
1813 enum bch_error_actions {
1814 #define x(t, n) BCH_ON_ERROR_##t = n,
1815         BCH_ERROR_ACTIONS()
1816 #undef x
1817         BCH_ON_ERROR_NR
1818 };
1819
1820 #define BCH_STR_HASH_TYPES()            \
1821         x(crc32c,               0)      \
1822         x(crc64,                1)      \
1823         x(siphash_old,          2)      \
1824         x(siphash,              3)
1825
1826 enum bch_str_hash_type {
1827 #define x(t, n) BCH_STR_HASH_##t = n,
1828         BCH_STR_HASH_TYPES()
1829 #undef x
1830         BCH_STR_HASH_NR
1831 };
1832
1833 #define BCH_STR_HASH_OPTS()             \
1834         x(crc32c,               0)      \
1835         x(crc64,                1)      \
1836         x(siphash,              2)
1837
1838 enum bch_str_hash_opts {
1839 #define x(t, n) BCH_STR_HASH_OPT_##t = n,
1840         BCH_STR_HASH_OPTS()
1841 #undef x
1842         BCH_STR_HASH_OPT_NR
1843 };
1844
1845 #define BCH_CSUM_TYPES()                        \
1846         x(none,                         0)      \
1847         x(crc32c_nonzero,               1)      \
1848         x(crc64_nonzero,                2)      \
1849         x(chacha20_poly1305_80,         3)      \
1850         x(chacha20_poly1305_128,        4)      \
1851         x(crc32c,                       5)      \
1852         x(crc64,                        6)      \
1853         x(xxhash,                       7)
1854
1855 enum bch_csum_type {
1856 #define x(t, n) BCH_CSUM_##t = n,
1857         BCH_CSUM_TYPES()
1858 #undef x
1859         BCH_CSUM_NR
1860 };
1861
1862 static const unsigned bch_crc_bytes[] = {
1863         [BCH_CSUM_none]                         = 0,
1864         [BCH_CSUM_crc32c_nonzero]               = 4,
1865         [BCH_CSUM_crc32c]                       = 4,
1866         [BCH_CSUM_crc64_nonzero]                = 8,
1867         [BCH_CSUM_crc64]                        = 8,
1868         [BCH_CSUM_xxhash]                       = 8,
1869         [BCH_CSUM_chacha20_poly1305_80]         = 10,
1870         [BCH_CSUM_chacha20_poly1305_128]        = 16,
1871 };
1872
1873 static inline _Bool bch2_csum_type_is_encryption(enum bch_csum_type type)
1874 {
1875         switch (type) {
1876         case BCH_CSUM_chacha20_poly1305_80:
1877         case BCH_CSUM_chacha20_poly1305_128:
1878                 return true;
1879         default:
1880                 return false;
1881         }
1882 }
1883
1884 #define BCH_CSUM_OPTS()                 \
1885         x(none,                 0)      \
1886         x(crc32c,               1)      \
1887         x(crc64,                2)      \
1888         x(xxhash,               3)
1889
1890 enum bch_csum_opts {
1891 #define x(t, n) BCH_CSUM_OPT_##t = n,
1892         BCH_CSUM_OPTS()
1893 #undef x
1894         BCH_CSUM_OPT_NR
1895 };
1896
1897 #define BCH_COMPRESSION_TYPES()         \
1898         x(none,                 0)      \
1899         x(lz4_old,              1)      \
1900         x(gzip,                 2)      \
1901         x(lz4,                  3)      \
1902         x(zstd,                 4)      \
1903         x(incompressible,       5)
1904
1905 enum bch_compression_type {
1906 #define x(t, n) BCH_COMPRESSION_TYPE_##t = n,
1907         BCH_COMPRESSION_TYPES()
1908 #undef x
1909         BCH_COMPRESSION_TYPE_NR
1910 };
1911
1912 #define BCH_COMPRESSION_OPTS()          \
1913         x(none,         0)              \
1914         x(lz4,          1)              \
1915         x(gzip,         2)              \
1916         x(zstd,         3)
1917
1918 enum bch_compression_opts {
1919 #define x(t, n) BCH_COMPRESSION_OPT_##t = n,
1920         BCH_COMPRESSION_OPTS()
1921 #undef x
1922         BCH_COMPRESSION_OPT_NR
1923 };
1924
1925 /*
1926  * Magic numbers
1927  *
1928  * The various other data structures have their own magic numbers, which are
1929  * xored with the first part of the cache set's UUID
1930  */
1931
1932 #define BCACHE_MAGIC                                                    \
1933         UUID_INIT(0xc68573f6, 0x4e1a, 0x45ca,                           \
1934                   0x82, 0x65, 0xf5, 0x7f, 0x48, 0xba, 0x6d, 0x81)
1935 #define BCHFS_MAGIC                                                     \
1936         UUID_INIT(0xc68573f6, 0x66ce, 0x90a9,                           \
1937                   0xd9, 0x6a, 0x60, 0xcf, 0x80, 0x3d, 0xf7, 0xef)
1938
1939 #define BCACHEFS_STATFS_MAGIC           0xca451a4e
1940
1941 #define JSET_MAGIC              __cpu_to_le64(0x245235c1a3625032ULL)
1942 #define BSET_MAGIC              __cpu_to_le64(0x90135c78b99e07f5ULL)
1943
1944 static inline __le64 __bch2_sb_magic(struct bch_sb *sb)
1945 {
1946         __le64 ret;
1947
1948         memcpy(&ret, &sb->uuid, sizeof(ret));
1949         return ret;
1950 }
1951
1952 static inline __u64 __jset_magic(struct bch_sb *sb)
1953 {
1954         return __le64_to_cpu(__bch2_sb_magic(sb) ^ JSET_MAGIC);
1955 }
1956
1957 static inline __u64 __bset_magic(struct bch_sb *sb)
1958 {
1959         return __le64_to_cpu(__bch2_sb_magic(sb) ^ BSET_MAGIC);
1960 }
1961
1962 /* Journal */
1963
1964 #define JSET_KEYS_U64s  (sizeof(struct jset_entry) / sizeof(__u64))
1965
1966 #define BCH_JSET_ENTRY_TYPES()                  \
1967         x(btree_keys,           0)              \
1968         x(btree_root,           1)              \
1969         x(prio_ptrs,            2)              \
1970         x(blacklist,            3)              \
1971         x(blacklist_v2,         4)              \
1972         x(usage,                5)              \
1973         x(data_usage,           6)              \
1974         x(clock,                7)              \
1975         x(dev_usage,            8)              \
1976         x(log,                  9)              \
1977         x(overwrite,            10)
1978
1979 enum {
1980 #define x(f, nr)        BCH_JSET_ENTRY_##f      = nr,
1981         BCH_JSET_ENTRY_TYPES()
1982 #undef x
1983         BCH_JSET_ENTRY_NR
1984 };
1985
1986 /*
1987  * Journal sequence numbers can be blacklisted: bsets record the max sequence
1988  * number of all the journal entries they contain updates for, so that on
1989  * recovery we can ignore those bsets that contain index updates newer that what
1990  * made it into the journal.
1991  *
1992  * This means that we can't reuse that journal_seq - we have to skip it, and
1993  * then record that we skipped it so that the next time we crash and recover we
1994  * don't think there was a missing journal entry.
1995  */
1996 struct jset_entry_blacklist {
1997         struct jset_entry       entry;
1998         __le64                  seq;
1999 };
2000
2001 struct jset_entry_blacklist_v2 {
2002         struct jset_entry       entry;
2003         __le64                  start;
2004         __le64                  end;
2005 };
2006
2007 #define BCH_FS_USAGE_TYPES()                    \
2008         x(reserved,             0)              \
2009         x(inodes,               1)              \
2010         x(key_version,          2)
2011
2012 enum {
2013 #define x(f, nr)        BCH_FS_USAGE_##f        = nr,
2014         BCH_FS_USAGE_TYPES()
2015 #undef x
2016         BCH_FS_USAGE_NR
2017 };
2018
2019 struct jset_entry_usage {
2020         struct jset_entry       entry;
2021         __le64                  v;
2022 } __packed;
2023
2024 struct jset_entry_data_usage {
2025         struct jset_entry       entry;
2026         __le64                  v;
2027         struct bch_replicas_entry r;
2028 } __packed;
2029
2030 struct jset_entry_clock {
2031         struct jset_entry       entry;
2032         __u8                    rw;
2033         __u8                    pad[7];
2034         __le64                  time;
2035 } __packed;
2036
2037 struct jset_entry_dev_usage_type {
2038         __le64                  buckets;
2039         __le64                  sectors;
2040         __le64                  fragmented;
2041 } __packed;
2042
2043 struct jset_entry_dev_usage {
2044         struct jset_entry       entry;
2045         __le32                  dev;
2046         __u32                   pad;
2047
2048         __le64                  buckets_ec;
2049         __le64                  _buckets_unavailable; /* No longer used */
2050
2051         struct jset_entry_dev_usage_type d[];
2052 } __packed;
2053
2054 static inline unsigned jset_entry_dev_usage_nr_types(struct jset_entry_dev_usage *u)
2055 {
2056         return (vstruct_bytes(&u->entry) - sizeof(struct jset_entry_dev_usage)) /
2057                 sizeof(struct jset_entry_dev_usage_type);
2058 }
2059
2060 struct jset_entry_log {
2061         struct jset_entry       entry;
2062         u8                      d[];
2063 } __packed;
2064
2065 /*
2066  * On disk format for a journal entry:
2067  * seq is monotonically increasing; every journal entry has its own unique
2068  * sequence number.
2069  *
2070  * last_seq is the oldest journal entry that still has keys the btree hasn't
2071  * flushed to disk yet.
2072  *
2073  * version is for on disk format changes.
2074  */
2075 struct jset {
2076         struct bch_csum         csum;
2077
2078         __le64                  magic;
2079         __le64                  seq;
2080         __le32                  version;
2081         __le32                  flags;
2082
2083         __le32                  u64s; /* size of d[] in u64s */
2084
2085         __u8                    encrypted_start[0];
2086
2087         __le16                  _read_clock; /* no longer used */
2088         __le16                  _write_clock;
2089
2090         /* Sequence number of oldest dirty journal entry */
2091         __le64                  last_seq;
2092
2093
2094         union {
2095                 struct jset_entry start[0];
2096                 __u64           _data[0];
2097         };
2098 } __packed __aligned(8);
2099
2100 LE32_BITMASK(JSET_CSUM_TYPE,    struct jset, flags, 0, 4);
2101 LE32_BITMASK(JSET_BIG_ENDIAN,   struct jset, flags, 4, 5);
2102 LE32_BITMASK(JSET_NO_FLUSH,     struct jset, flags, 5, 6);
2103
2104 #define BCH_JOURNAL_BUCKETS_MIN         8
2105
2106 /* Btree: */
2107
2108 #define BCH_BTREE_IDS()                         \
2109         x(extents,              0)              \
2110         x(inodes,               1)              \
2111         x(dirents,              2)              \
2112         x(xattrs,               3)              \
2113         x(alloc,                4)              \
2114         x(quotas,               5)              \
2115         x(stripes,              6)              \
2116         x(reflink,              7)              \
2117         x(subvolumes,           8)              \
2118         x(snapshots,            9)              \
2119         x(lru,                  10)             \
2120         x(freespace,            11)             \
2121         x(need_discard,         12)             \
2122         x(backpointers,         13)             \
2123         x(bucket_gens,          14)             \
2124         x(snapshot_trees,       15)
2125
2126 enum btree_id {
2127 #define x(kwd, val) BTREE_ID_##kwd = val,
2128         BCH_BTREE_IDS()
2129 #undef x
2130         BTREE_ID_NR
2131 };
2132
2133 #define BTREE_MAX_DEPTH         4U
2134
2135 /* Btree nodes */
2136
2137 /*
2138  * Btree nodes
2139  *
2140  * On disk a btree node is a list/log of these; within each set the keys are
2141  * sorted
2142  */
2143 struct bset {
2144         __le64                  seq;
2145
2146         /*
2147          * Highest journal entry this bset contains keys for.
2148          * If on recovery we don't see that journal entry, this bset is ignored:
2149          * this allows us to preserve the order of all index updates after a
2150          * crash, since the journal records a total order of all index updates
2151          * and anything that didn't make it to the journal doesn't get used.
2152          */
2153         __le64                  journal_seq;
2154
2155         __le32                  flags;
2156         __le16                  version;
2157         __le16                  u64s; /* count of d[] in u64s */
2158
2159         union {
2160                 struct bkey_packed start[0];
2161                 __u64           _data[0];
2162         };
2163 } __packed __aligned(8);
2164
2165 LE32_BITMASK(BSET_CSUM_TYPE,    struct bset, flags, 0, 4);
2166
2167 LE32_BITMASK(BSET_BIG_ENDIAN,   struct bset, flags, 4, 5);
2168 LE32_BITMASK(BSET_SEPARATE_WHITEOUTS,
2169                                 struct bset, flags, 5, 6);
2170
2171 /* Sector offset within the btree node: */
2172 LE32_BITMASK(BSET_OFFSET,       struct bset, flags, 16, 32);
2173
2174 struct btree_node {
2175         struct bch_csum         csum;
2176         __le64                  magic;
2177
2178         /* this flags field is encrypted, unlike bset->flags: */
2179         __le64                  flags;
2180
2181         /* Closed interval: */
2182         struct bpos             min_key;
2183         struct bpos             max_key;
2184         struct bch_extent_ptr   _ptr; /* not used anymore */
2185         struct bkey_format      format;
2186
2187         union {
2188         struct bset             keys;
2189         struct {
2190                 __u8            pad[22];
2191                 __le16          u64s;
2192                 __u64           _data[0];
2193
2194         };
2195         };
2196 } __packed __aligned(8);
2197
2198 LE64_BITMASK(BTREE_NODE_ID,     struct btree_node, flags,  0,  4);
2199 LE64_BITMASK(BTREE_NODE_LEVEL,  struct btree_node, flags,  4,  8);
2200 LE64_BITMASK(BTREE_NODE_NEW_EXTENT_OVERWRITE,
2201                                 struct btree_node, flags,  8,  9);
2202 /* 9-32 unused */
2203 LE64_BITMASK(BTREE_NODE_SEQ,    struct btree_node, flags, 32, 64);
2204
2205 struct btree_node_entry {
2206         struct bch_csum         csum;
2207
2208         union {
2209         struct bset             keys;
2210         struct {
2211                 __u8            pad[22];
2212                 __le16          u64s;
2213                 __u64           _data[0];
2214
2215         };
2216         };
2217 } __packed __aligned(8);
2218
2219 #endif /* _BCACHEFS_FORMAT_H */