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

   1 /* SPDX-License-Identifier: GPL-2.0 */
   2 #ifndef _BCACHEFS_FORMAT_H
   3 #define _BCACHEFS_FORMAT_H
   4
   5 /*
   6  * bcachefs on disk data structures
   7  *
   8  * OVERVIEW:
   9  *
  10  * There are three main types of on disk data structures in bcachefs (this is
  11  * reduced from 5 in bcache)
  12  *
  13  *  - superblock
  14  *  - journal
  15  *  - btree
  16  *
  17  * The btree is the primary structure; most metadata exists as keys in the
  18  * various btrees. There are only a small number of btrees, they're not
  19  * sharded - we have one btree for extents, another for inodes, et cetera.
  20  *
  21  * SUPERBLOCK:
  22  *
  23  * The superblock contains the location of the journal, the list of devices in
  24  * the filesystem, and in general any metadata we need in order to decide
  25  * whether we can start a filesystem or prior to reading the journal/btree
  26  * roots.
  27  *
  28  * The superblock is extensible, and most of the contents of the superblock are
  29  * in variable length, type tagged fields; see struct bch_sb_field.
  30  *
  31  * Backup superblocks do not reside in a fixed location; also, superblocks do
  32  * not have a fixed size. To locate backup superblocks we have struct
  33  * bch_sb_layout; we store a copy of this inside every superblock, and also
  34  * before the first superblock.
  35  *
  36  * JOURNAL:
  37  *
  38  * The journal primarily records btree updates in the order they occurred;
  39  * journal replay consists of just iterating over all the keys in the open
  40  * journal entries and re-inserting them into the btrees.
  41  *
  42  * The journal also contains entry types for the btree roots, and blacklisted
  43  * journal sequence numbers (see journal_seq_blacklist.c).
  44  *
  45  * BTREE:
  46  *
  47  * bcachefs btrees are copy on write b+ trees, where nodes are big (typically
  48  * 128k-256k) and log structured. We use struct btree_node for writing the first
  49  * entry in a given node (offset 0), and struct btree_node_entry for all
  50  * subsequent writes.
  51  *
  52  * After the header, btree node entries contain a list of keys in sorted order.
  53  * Values are stored inline with the keys; since values are variable length (and
  54  * keys effectively are variable length too, due to packing) we can't do random
  55  * access without building up additional in memory tables in the btree node read
  56  * path.
  57  *
  58  * BTREE KEYS (struct bkey):
  59  *
  60  * The various btrees share a common format for the key - so as to avoid
  61  * switching in fastpath lookup/comparison code - but define their own
  62  * structures for the key values.
  63  *
  64  * The size of a key/value pair is stored as a u8 in units of u64s, so the max
  65  * size is just under 2k. The common part also contains a type tag for the
  66  * value, and a format field indicating whether the key is packed or not (and
  67  * also meant to allow adding new key fields in the future, if desired).
  68  *
  69  * bkeys, when stored within a btree node, may also be packed. In that case, the
  70  * bkey_format in that node is used to unpack it. Packed bkeys mean that we can
  71  * be generous with field sizes in the common part of the key format (64 bit
  72  * inode number, 64 bit offset, 96 bit version field, etc.) for negligible cost.
  73  */
  74
  75 #include <asm/types.h>
  76 #include <asm/byteorder.h>
  77 #include <linux/kernel.h>
  78 #include <linux/uuid.h>
  79 #include "vstructs.h"
  80
  81 #define BITMASK(name, type, field, offset, end)                         \
  82 static const unsigned   name##_OFFSET = offset;                         \
  83 static const unsigned   name##_BITS = (end - offset);                   \
  84                                                                         \
  85 static inline __u64 name(const type *k)                                 \
  86 {                                                                       \
  87         return (k->field >> offset) & ~(~0ULL << (end - offset));       \
  88 }                                                                       \
  89                                                                         \
  90 static inline void SET_##name(type *k, __u64 v)                         \
  91 {                                                                       \
  92         k->field &= ~(~(~0ULL << (end - offset)) << offset);            \
  93         k->field |= (v & ~(~0ULL << (end - offset))) << offset;         \
  94 }
  95
  96 #define LE_BITMASK(_bits, name, type, field, offset, end)               \
  97 static const unsigned   name##_OFFSET = offset;                         \
  98 static const unsigned   name##_BITS = (end - offset);                   \
  99 static const __u##_bits name##_MAX = (1ULL << (end - offset)) - 1;      \
 100                                                                         \
 101 static inline __u64 name(const type *k)                                 \
 102 {                                                                       \
 103         return (__le##_bits##_to_cpu(k->field) >> offset) &             \
 104                 ~(~0ULL << (end - offset));                             \
 105 }                                                                       \
 106                                                                         \
 107 static inline void SET_##name(type *k, __u64 v)                         \
 108 {                                                                       \
 109         __u##_bits new = __le##_bits##_to_cpu(k->field);                \
 110                                                                         \
 111         new &= ~(~(~0ULL << (end - offset)) << offset);                 \
 112         new |= (v & ~(~0ULL << (end - offset))) << offset;              \
 113         k->field = __cpu_to_le##_bits(new);                             \
 114 }
 115
 116 #define LE16_BITMASK(n, t, f, o, e)     LE_BITMASK(16, n, t, f, o, e)
 117 #define LE32_BITMASK(n, t, f, o, e)     LE_BITMASK(32, n, t, f, o, e)
 118 #define LE64_BITMASK(n, t, f, o, e)     LE_BITMASK(64, n, t, f, o, e)
 119
 120 struct bkey_format {
 121         __u8            key_u64s;
 122         __u8            nr_fields;
 123         /* One unused slot for now: */
 124         __u8            bits_per_field[6];
 125         __le64          field_offset[6];
 126 };
 127
 128 /* Btree keys - all units are in sectors */
 129
 130 struct bpos {
 131         /*
 132          * Word order matches machine byte order - btree code treats a bpos as a
 133          * single large integer, for search/comparison purposes
 134          *
 135          * Note that wherever a bpos is embedded in another on disk data
 136          * structure, it has to be byte swabbed when reading in metadata that
 137          * wasn't written in native endian order:
 138          */
 139 #if __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__
 140         __u32           snapshot;
 141         __u64           offset;
 142         __u64           inode;
 143 #elif __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__
 144         __u64           inode;
 145         __u64           offset;         /* Points to end of extent - sectors */
 146         __u32           snapshot;
 147 #else
 148 #error edit for your odd byteorder.
 149 #endif
 150 } __packed __aligned(4);
 151
 152 #define KEY_INODE_MAX                   ((__u64)~0ULL)
 153 #define KEY_OFFSET_MAX                  ((__u64)~0ULL)
 154 #define KEY_SNAPSHOT_MAX                ((__u32)~0U)
 155 #define KEY_SIZE_MAX                    ((__u32)~0U)
 156
 157 static inline struct bpos SPOS(__u64 inode, __u64 offset, __u32 snapshot)
 158 {
 159         return (struct bpos) {
 160                 .inode          = inode,
 161                 .offset         = offset,
 162                 .snapshot       = snapshot,
 163         };
 164 }
 165
 166 #define POS_MIN                         SPOS(0, 0, 0)
 167 #define POS_MAX                         SPOS(KEY_INODE_MAX, KEY_OFFSET_MAX, 0)
 168 #define SPOS_MAX                        SPOS(KEY_INODE_MAX, KEY_OFFSET_MAX, KEY_SNAPSHOT_MAX)
 169 #define POS(_inode, _offset)            SPOS(_inode, _offset, 0)
 170
 171 /* Empty placeholder struct, for container_of() */
 172 struct bch_val {
 173         __u64           __nothing[0];
 174 };
 175
 176 struct bversion {
 177 #if __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__
 178         __u64           lo;
 179         __u32           hi;
 180 #elif __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__
 181         __u32           hi;
 182         __u64           lo;
 183 #endif
 184 } __packed __aligned(4);
 185
 186 struct bkey {
 187         /* Size of combined key and value, in u64s */
 188         __u8            u64s;
 189
 190         /* Format of key (0 for format local to btree node) */
 191 #if defined(__LITTLE_ENDIAN_BITFIELD)
 192         __u8            format:7,
 193                         needs_whiteout:1;
 194 #elif defined (__BIG_ENDIAN_BITFIELD)
 195         __u8            needs_whiteout:1,
 196                         format:7;
 197 #else
 198 #error edit for your odd byteorder.
 199 #endif
 200
 201         /* Type of the value */
 202         __u8            type;
 203
 204 #if __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__
 205         __u8            pad[1];
 206
 207         struct bversion version;
 208         __u32           size;           /* extent size, in sectors */
 209         struct bpos     p;
 210 #elif __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__
 211         struct bpos     p;
 212         __u32           size;           /* extent size, in sectors */
 213         struct bversion version;
 214
 215         __u8            pad[1];
 216 #endif
 217 } __packed __aligned(8);
 218
 219 struct bkey_packed {
 220         __u64           _data[0];
 221
 222         /* Size of combined key and value, in u64s */
 223         __u8            u64s;
 224
 225         /* Format of key (0 for format local to btree node) */
 226
 227         /*
 228          * XXX: next incompat on disk format change, switch format and
 229          * needs_whiteout - bkey_packed() will be cheaper if format is the high
 230          * bits of the bitfield
 231          */
 232 #if defined(__LITTLE_ENDIAN_BITFIELD)
 233         __u8            format:7,
 234                         needs_whiteout:1;
 235 #elif defined (__BIG_ENDIAN_BITFIELD)
 236         __u8            needs_whiteout:1,
 237                         format:7;
 238 #endif
 239
 240         /* Type of the value */
 241         __u8            type;
 242         __u8            key_start[0];
 243
 244         /*
 245          * We copy bkeys with struct assignment in various places, and while
 246          * that shouldn't be done with packed bkeys we can't disallow it in C,
 247          * and it's legal to cast a bkey to a bkey_packed  - so padding it out
 248          * to the same size as struct bkey should hopefully be safest.
 249          */
 250         __u8            pad[sizeof(struct bkey) - 3];
 251 } __packed __aligned(8);
 252
 253 #define BKEY_U64s                       (sizeof(struct bkey) / sizeof(__u64))
 254 #define BKEY_U64s_MAX                   U8_MAX
 255 #define BKEY_VAL_U64s_MAX               (BKEY_U64s_MAX - BKEY_U64s)
 256
 257 #define KEY_PACKED_BITS_START           24
 258
 259 #define KEY_FORMAT_LOCAL_BTREE          0
 260 #define KEY_FORMAT_CURRENT              1
 261
 262 enum bch_bkey_fields {
 263         BKEY_FIELD_INODE,
 264         BKEY_FIELD_OFFSET,
 265         BKEY_FIELD_SNAPSHOT,
 266         BKEY_FIELD_SIZE,
 267         BKEY_FIELD_VERSION_HI,
 268         BKEY_FIELD_VERSION_LO,
 269         BKEY_NR_FIELDS,
 270 };
 271
 272 #define bkey_format_field(name, field)                                  \
 273         [BKEY_FIELD_##name] = (sizeof(((struct bkey *) NULL)->field) * 8)
 274
 275 #define BKEY_FORMAT_CURRENT                                             \
 276 ((struct bkey_format) {                                                 \
 277         .key_u64s       = BKEY_U64s,                                    \
 278         .nr_fields      = BKEY_NR_FIELDS,                               \
 279         .bits_per_field = {                                             \
 280                 bkey_format_field(INODE,        p.inode),               \
 281                 bkey_format_field(OFFSET,       p.offset),              \
 282                 bkey_format_field(SNAPSHOT,     p.snapshot),            \
 283                 bkey_format_field(SIZE,         size),                  \
 284                 bkey_format_field(VERSION_HI,   version.hi),            \
 285                 bkey_format_field(VERSION_LO,   version.lo),            \
 286         },                                                              \
 287 })
 288
 289 /* bkey with inline value */
 290 struct bkey_i {
 291         __u64                   _data[0];
 292
 293         union {
 294         struct {
 295                 /* Size of combined key and value, in u64s */
 296                 __u8            u64s;
 297         };
 298         struct {
 299                 struct bkey     k;
 300                 struct bch_val  v;
 301         };
 302         };
 303 };
 304
 305 #define KEY(_inode, _offset, _size)                                     \
 306 ((struct bkey) {                                                        \
 307         .u64s           = BKEY_U64s,                                    \
 308         .format         = KEY_FORMAT_CURRENT,                           \
 309         .p              = POS(_inode, _offset),                         \
 310         .size           = _size,                                        \
 311 })
 312
 313 static inline void bkey_init(struct bkey *k)
 314 {
 315         *k = KEY(0, 0, 0);
 316 }
 317
 318 #define bkey_bytes(_k)          ((_k)->u64s * sizeof(__u64))
 319
 320 #define __BKEY_PADDED(key, pad)                                 \
 321         struct { struct bkey_i key; __u64 key ## _pad[pad]; }
 322
 323 /*
 324  * - DELETED keys are used internally to mark keys that should be ignored but
 325  *   override keys in composition order.  Their version number is ignored.
 326  *
 327  * - DISCARDED keys indicate that the data is all 0s because it has been
 328  *   discarded. DISCARDs may have a version; if the version is nonzero the key
 329  *   will be persistent, otherwise the key will be dropped whenever the btree
 330  *   node is rewritten (like DELETED keys).
 331  *
 332  * - ERROR: any read of the data returns a read error, as the data was lost due
 333  *   to a failing device. Like DISCARDED keys, they can be removed (overridden)
 334  *   by new writes or cluster-wide GC. Node repair can also overwrite them with
 335  *   the same or a more recent version number, but not with an older version
 336  *   number.
 337  *
 338  * - WHITEOUT: for hash table btrees
 339  */
 340 #define BCH_BKEY_TYPES()                                \
 341         x(deleted,              0)                      \
 342         x(whiteout,             1)                      \
 343         x(error,                2)                      \
 344         x(cookie,               3)                      \
 345         x(hash_whiteout,        4)                      \
 346         x(btree_ptr,            5)                      \
 347         x(extent,               6)                      \
 348         x(reservation,          7)                      \
 349         x(inode,                8)                      \
 350         x(inode_generation,     9)                      \
 351         x(dirent,               10)                     \
 352         x(xattr,                11)                     \
 353         x(alloc,                12)                     \
 354         x(quota,                13)                     \
 355         x(stripe,               14)                     \
 356         x(reflink_p,            15)                     \
 357         x(reflink_v,            16)                     \
 358         x(inline_data,          17)                     \
 359         x(btree_ptr_v2,         18)                     \
 360         x(indirect_inline_data, 19)                     \
 361         x(alloc_v2,             20)                     \
 362         x(subvolume,            21)                     \
 363         x(snapshot,             22)                     \
 364         x(inode_v2,             23)                     \
 365         x(alloc_v3,             24)                     \
 366         x(set,                  25)                     \
 367         x(lru,                  26)                     \
 368         x(alloc_v4,             27)                     \
 369         x(backpointer,          28)                     \
 370         x(inode_v3,             29)
 371
 372 enum bch_bkey_type {
 373 #define x(name, nr) KEY_TYPE_##name     = nr,
 374         BCH_BKEY_TYPES()
 375 #undef x
 376         KEY_TYPE_MAX,
 377 };
 378
 379 struct bch_deleted {
 380         struct bch_val          v;
 381 };
 382
 383 struct bch_whiteout {
 384         struct bch_val          v;
 385 };
 386
 387 struct bch_error {
 388         struct bch_val          v;
 389 };
 390
 391 struct bch_cookie {
 392         struct bch_val          v;
 393         __le64                  cookie;
 394 };
 395
 396 struct bch_hash_whiteout {
 397         struct bch_val          v;
 398 };
 399
 400 struct bch_set {
 401         struct bch_val          v;
 402 };
 403
 404 /* Extents */
 405
 406 /*
 407  * In extent bkeys, the value is a list of pointers (bch_extent_ptr), optionally
 408  * preceded by checksum/compression information (bch_extent_crc32 or
 409  * bch_extent_crc64).
 410  *
 411  * One major determining factor in the format of extents is how we handle and
 412  * represent extents that have been partially overwritten and thus trimmed:
 413  *
 414  * If an extent is not checksummed or compressed, when the extent is trimmed we
 415  * don't have to remember the extent we originally allocated and wrote: we can
 416  * merely adjust ptr->offset to point to the start of the data that is currently
 417  * live. The size field in struct bkey records the current (live) size of the
 418  * extent, and is also used to mean "size of region on disk that we point to" in
 419  * this case.
 420  *
 421  * Thus an extent that is not checksummed or compressed will consist only of a
 422  * list of bch_extent_ptrs, with none of the fields in
 423  * bch_extent_crc32/bch_extent_crc64.
 424  *
 425  * When an extent is checksummed or compressed, it's not possible to read only
 426  * the data that is currently live: we have to read the entire extent that was
 427  * originally written, and then return only the part of the extent that is
 428  * currently live.
 429  *
 430  * Thus, in addition to the current size of the extent in struct bkey, we need
 431  * to store the size of the originally allocated space - this is the
 432  * compressed_size and uncompressed_size fields in bch_extent_crc32/64. Also,
 433  * when the extent is trimmed, instead of modifying the offset field of the
 434  * pointer, we keep a second smaller offset field - "offset into the original
 435  * extent of the currently live region".
 436  *
 437  * The other major determining factor is replication and data migration:
 438  *
 439  * Each pointer may have its own bch_extent_crc32/64. When doing a replicated
 440  * write, we will initially write all the replicas in the same format, with the
 441  * same checksum type and compression format - however, when copygc runs later (or
 442  * tiering/cache promotion, anything that moves data), it is not in general
 443  * going to rewrite all the pointers at once - one of the replicas may be in a
 444  * bucket on one device that has very little fragmentation while another lives
 445  * in a bucket that has become heavily fragmented, and thus is being rewritten
 446  * sooner than the rest.
 447  *
 448  * Thus it will only move a subset of the pointers (or in the case of
 449  * tiering/cache promotion perhaps add a single pointer without dropping any
 450  * current pointers), and if the extent has been partially overwritten it must
 451  * write only the currently live portion (or copygc would not be able to reduce
 452  * fragmentation!) - which necessitates a different bch_extent_crc format for
 453  * the new pointer.
 454  *
 455  * But in the interests of space efficiency, we don't want to store one
 456  * bch_extent_crc for each pointer if we don't have to.
 457  *
 458  * Thus, a bch_extent consists of bch_extent_crc32s, bch_extent_crc64s, and
 459  * bch_extent_ptrs appended arbitrarily one after the other. We determine the
 460  * type of a given entry with a scheme similar to utf8 (except we're encoding a
 461  * type, not a size), encoding the type in the position of the first set bit:
 462  *
 463  * bch_extent_crc32     - 0b1
 464  * bch_extent_ptr       - 0b10
 465  * bch_extent_crc64     - 0b100
 466  *
 467  * We do it this way because bch_extent_crc32 is _very_ constrained on bits (and
 468  * bch_extent_crc64 is the least constrained).
 469  *
 470  * Then, each bch_extent_crc32/64 applies to the pointers that follow after it,
 471  * until the next bch_extent_crc32/64.
 472  *
 473  * If there are no bch_extent_crcs preceding a bch_extent_ptr, then that pointer
 474  * is neither checksummed nor compressed.
 475  */
 476
 477 /* 128 bits, sufficient for cryptographic MACs: */
 478 struct bch_csum {
 479         __le64                  lo;
 480         __le64                  hi;
 481 } __packed __aligned(8);
 482
 483 #define BCH_EXTENT_ENTRY_TYPES()                \
 484         x(ptr,                  0)              \
 485         x(crc32,                1)              \
 486         x(crc64,                2)              \
 487         x(crc128,               3)              \
 488         x(stripe_ptr,           4)
 489 #define BCH_EXTENT_ENTRY_MAX    5
 490
 491 enum bch_extent_entry_type {
 492 #define x(f, n) BCH_EXTENT_ENTRY_##f = n,
 493         BCH_EXTENT_ENTRY_TYPES()
 494 #undef x
 495 };
 496
 497 /* Compressed/uncompressed size are stored biased by 1: */
 498 struct bch_extent_crc32 {
 499 #if defined(__LITTLE_ENDIAN_BITFIELD)
 500         __u32                   type:2,
 501                                 _compressed_size:7,
 502                                 _uncompressed_size:7,
 503                                 offset:7,
 504                                 _unused:1,
 505                                 csum_type:4,
 506                                 compression_type:4;
 507         __u32                   csum;
 508 #elif defined (__BIG_ENDIAN_BITFIELD)
 509         __u32                   csum;
 510         __u32                   compression_type:4,
 511                                 csum_type:4,
 512                                 _unused:1,
 513                                 offset:7,
 514                                 _uncompressed_size:7,
 515                                 _compressed_size:7,
 516                                 type:2;
 517 #endif
 518 } __packed __aligned(8);
 519
 520 #define CRC32_SIZE_MAX          (1U << 7)
 521 #define CRC32_NONCE_MAX         0
 522
 523 struct bch_extent_crc64 {
 524 #if defined(__LITTLE_ENDIAN_BITFIELD)
 525         __u64                   type:3,
 526                                 _compressed_size:9,
 527                                 _uncompressed_size:9,
 528                                 offset:9,
 529                                 nonce:10,
 530                                 csum_type:4,
 531                                 compression_type:4,
 532                                 csum_hi:16;
 533 #elif defined (__BIG_ENDIAN_BITFIELD)
 534         __u64                   csum_hi:16,
 535                                 compression_type:4,
 536                                 csum_type:4,
 537                                 nonce:10,
 538                                 offset:9,
 539                                 _uncompressed_size:9,
 540                                 _compressed_size:9,
 541                                 type:3;
 542 #endif
 543         __u64                   csum_lo;
 544 } __packed __aligned(8);
 545
 546 #define CRC64_SIZE_MAX          (1U << 9)
 547 #define CRC64_NONCE_MAX         ((1U << 10) - 1)
 548
 549 struct bch_extent_crc128 {
 550 #if defined(__LITTLE_ENDIAN_BITFIELD)
 551         __u64                   type:4,
 552                                 _compressed_size:13,
 553                                 _uncompressed_size:13,
 554                                 offset:13,
 555                                 nonce:13,
 556                                 csum_type:4,
 557                                 compression_type:4;
 558 #elif defined (__BIG_ENDIAN_BITFIELD)
 559         __u64                   compression_type:4,
 560                                 csum_type:4,
 561                                 nonce:13,
 562                                 offset:13,
 563                                 _uncompressed_size:13,
 564                                 _compressed_size:13,
 565                                 type:4;
 566 #endif
 567         struct bch_csum         csum;
 568 } __packed __aligned(8);
 569
 570 #define CRC128_SIZE_MAX         (1U << 13)
 571 #define CRC128_NONCE_MAX        ((1U << 13) - 1)
 572
 573 /*
 574  * @reservation - pointer hasn't been written to, just reserved
 575  */
 576 struct bch_extent_ptr {
 577 #if defined(__LITTLE_ENDIAN_BITFIELD)
 578         __u64                   type:1,
 579                                 cached:1,
 580                                 unused:1,
 581                                 reservation:1,
 582                                 offset:44, /* 8 petabytes */
 583                                 dev:8,
 584                                 gen:8;
 585 #elif defined (__BIG_ENDIAN_BITFIELD)
 586         __u64                   gen:8,
 587                                 dev:8,
 588                                 offset:44,
 589                                 reservation:1,
 590                                 unused:1,
 591                                 cached:1,
 592                                 type:1;
 593 #endif
 594 } __packed __aligned(8);
 595
 596 struct bch_extent_stripe_ptr {
 597 #if defined(__LITTLE_ENDIAN_BITFIELD)
 598         __u64                   type:5,
 599                                 block:8,
 600                                 redundancy:4,
 601                                 idx:47;
 602 #elif defined (__BIG_ENDIAN_BITFIELD)
 603         __u64                   idx:47,
 604                                 redundancy:4,
 605                                 block:8,
 606                                 type:5;
 607 #endif
 608 };
 609
 610 struct bch_extent_reservation {
 611 #if defined(__LITTLE_ENDIAN_BITFIELD)
 612         __u64                   type:6,
 613                                 unused:22,
 614                                 replicas:4,
 615                                 generation:32;
 616 #elif defined (__BIG_ENDIAN_BITFIELD)
 617         __u64                   generation:32,
 618                                 replicas:4,
 619                                 unused:22,
 620                                 type:6;
 621 #endif
 622 };
 623
 624 union bch_extent_entry {
 625 #if __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__ ||  __BITS_PER_LONG == 64
 626         unsigned long                   type;
 627 #elif __BITS_PER_LONG == 32
 628         struct {
 629                 unsigned long           pad;
 630                 unsigned long           type;
 631         };
 632 #else
 633 #error edit for your odd byteorder.
 634 #endif
 635
 636 #define x(f, n) struct bch_extent_##f   f;
 637         BCH_EXTENT_ENTRY_TYPES()
 638 #undef x
 639 };
 640
 641 struct bch_btree_ptr {
 642         struct bch_val          v;
 643
 644         __u64                   _data[0];
 645         struct bch_extent_ptr   start[];
 646 } __packed __aligned(8);
 647
 648 struct bch_btree_ptr_v2 {
 649         struct bch_val          v;
 650
 651         __u64                   mem_ptr;
 652         __le64                  seq;
 653         __le16                  sectors_written;
 654         __le16                  flags;
 655         struct bpos             min_key;
 656         __u64                   _data[0];
 657         struct bch_extent_ptr   start[];
 658 } __packed __aligned(8);
 659
 660 LE16_BITMASK(BTREE_PTR_RANGE_UPDATED,   struct bch_btree_ptr_v2, flags, 0, 1);
 661
 662 struct bch_extent {
 663         struct bch_val          v;
 664
 665         __u64                   _data[0];
 666         union bch_extent_entry  start[];
 667 } __packed __aligned(8);
 668
 669 struct bch_reservation {
 670         struct bch_val          v;
 671
 672         __le32                  generation;
 673         __u8                    nr_replicas;
 674         __u8                    pad[3];
 675 } __packed __aligned(8);
 676
 677 /* Maximum size (in u64s) a single pointer could be: */
 678 #define BKEY_EXTENT_PTR_U64s_MAX\
 679         ((sizeof(struct bch_extent_crc128) +                    \
 680           sizeof(struct bch_extent_ptr)) / sizeof(u64))
 681
 682 /* Maximum possible size of an entire extent value: */
 683 #define BKEY_EXTENT_VAL_U64s_MAX                                \
 684         (1 + BKEY_EXTENT_PTR_U64s_MAX * (BCH_REPLICAS_MAX + 1))
 685
 686 /* * Maximum possible size of an entire extent, key + value: */
 687 #define BKEY_EXTENT_U64s_MAX            (BKEY_U64s + BKEY_EXTENT_VAL_U64s_MAX)
 688
 689 /* Btree pointers don't carry around checksums: */
 690 #define BKEY_BTREE_PTR_VAL_U64s_MAX                             \
 691         ((sizeof(struct bch_btree_ptr_v2) +                     \
 692           sizeof(struct bch_extent_ptr) * BCH_REPLICAS_MAX) / sizeof(u64))
 693 #define BKEY_BTREE_PTR_U64s_MAX                                 \
 694         (BKEY_U64s + BKEY_BTREE_PTR_VAL_U64s_MAX)
 695
 696 /* Inodes */
 697
 698 #define BLOCKDEV_INODE_MAX      4096
 699
 700 #define BCACHEFS_ROOT_INO       4096
 701
 702 struct bch_inode {
 703         struct bch_val          v;
 704
 705         __le64                  bi_hash_seed;
 706         __le32                  bi_flags;
 707         __le16                  bi_mode;
 708         __u8                    fields[0];
 709 } __packed __aligned(8);
 710
 711 struct bch_inode_v2 {
 712         struct bch_val          v;
 713
 714         __le64                  bi_journal_seq;
 715         __le64                  bi_hash_seed;
 716         __le64                  bi_flags;
 717         __le16                  bi_mode;
 718         __u8                    fields[0];
 719 } __packed __aligned(8);
 720
 721 struct bch_inode_v3 {
 722         struct bch_val          v;
 723
 724         __le64                  bi_journal_seq;
 725         __le64                  bi_hash_seed;
 726         __le64                  bi_flags;
 727         __le64                  bi_sectors;
 728         __le64                  bi_size;
 729         __le64                  bi_version;
 730         __u8                    fields[0];
 731 } __packed __aligned(8);
 732
 733 #define INODEv3_FIELDS_START_INITIAL    6
 734 #define INODEv3_FIELDS_START_CUR        (offsetof(struct bch_inode_v3, fields) / sizeof(u64))
 735
 736 struct bch_inode_generation {
 737         struct bch_val          v;
 738
 739         __le32                  bi_generation;
 740         __le32                  pad;
 741 } __packed __aligned(8);
 742
 743 /*
 744  * bi_subvol and bi_parent_subvol are only set for subvolume roots:
 745  */
 746
 747 #define BCH_INODE_FIELDS_v2()                   \
 748         x(bi_atime,                     96)     \
 749         x(bi_ctime,                     96)     \
 750         x(bi_mtime,                     96)     \
 751         x(bi_otime,                     96)     \
 752         x(bi_size,                      64)     \
 753         x(bi_sectors,                   64)     \
 754         x(bi_uid,                       32)     \
 755         x(bi_gid,                       32)     \
 756         x(bi_nlink,                     32)     \
 757         x(bi_generation,                32)     \
 758         x(bi_dev,                       32)     \
 759         x(bi_data_checksum,             8)      \
 760         x(bi_compression,               8)      \
 761         x(bi_project,                   32)     \
 762         x(bi_background_compression,    8)      \
 763         x(bi_data_replicas,             8)      \
 764         x(bi_promote_target,            16)     \
 765         x(bi_foreground_target,         16)     \
 766         x(bi_background_target,         16)     \
 767         x(bi_erasure_code,              16)     \
 768         x(bi_fields_set,                16)     \
 769         x(bi_dir,                       64)     \
 770         x(bi_dir_offset,                64)     \
 771         x(bi_subvol,                    32)     \
 772         x(bi_parent_subvol,             32)
 773
 774 #define BCH_INODE_FIELDS_v3()                   \
 775         x(bi_atime,                     96)     \
 776         x(bi_ctime,                     96)     \
 777         x(bi_mtime,                     96)     \
 778         x(bi_otime,                     96)     \
 779         x(bi_uid,                       32)     \
 780         x(bi_gid,                       32)     \
 781         x(bi_nlink,                     32)     \
 782         x(bi_generation,                32)     \
 783         x(bi_dev,                       32)     \
 784         x(bi_data_checksum,             8)      \
 785         x(bi_compression,               8)      \
 786         x(bi_project,                   32)     \
 787         x(bi_background_compression,    8)      \
 788         x(bi_data_replicas,             8)      \
 789         x(bi_promote_target,            16)     \
 790         x(bi_foreground_target,         16)     \
 791         x(bi_background_target,         16)     \
 792         x(bi_erasure_code,              16)     \
 793         x(bi_fields_set,                16)     \
 794         x(bi_dir,                       64)     \
 795         x(bi_dir_offset,                64)     \
 796         x(bi_subvol,                    32)     \
 797         x(bi_parent_subvol,             32)
 798
 799 /* subset of BCH_INODE_FIELDS */
 800 #define BCH_INODE_OPTS()                        \
 801         x(data_checksum,                8)      \
 802         x(compression,                  8)      \
 803         x(project,                      32)     \
 804         x(background_compression,       8)      \
 805         x(data_replicas,                8)      \
 806         x(promote_target,               16)     \
 807         x(foreground_target,            16)     \
 808         x(background_target,            16)     \
 809         x(erasure_code,                 16)
 810
 811 enum inode_opt_id {
 812 #define x(name, ...)                            \
 813         Inode_opt_##name,
 814         BCH_INODE_OPTS()
 815 #undef  x
 816         Inode_opt_nr,
 817 };
 818
 819 enum {
 820         /*
 821          * User flags (get/settable with FS_IOC_*FLAGS, correspond to FS_*_FL
 822          * flags)
 823          */
 824         __BCH_INODE_SYNC                = 0,
 825         __BCH_INODE_IMMUTABLE           = 1,
 826         __BCH_INODE_APPEND              = 2,
 827         __BCH_INODE_NODUMP              = 3,
 828         __BCH_INODE_NOATIME             = 4,
 829
 830         __BCH_INODE_I_SIZE_DIRTY        = 5,
 831         __BCH_INODE_I_SECTORS_DIRTY     = 6,
 832         __BCH_INODE_UNLINKED            = 7,
 833         __BCH_INODE_BACKPTR_UNTRUSTED   = 8,
 834
 835         /* bits 20+ reserved for packed fields below: */
 836 };
 837
 838 #define BCH_INODE_SYNC          (1 << __BCH_INODE_SYNC)
 839 #define BCH_INODE_IMMUTABLE     (1 << __BCH_INODE_IMMUTABLE)
 840 #define BCH_INODE_APPEND        (1 << __BCH_INODE_APPEND)
 841 #define BCH_INODE_NODUMP        (1 << __BCH_INODE_NODUMP)
 842 #define BCH_INODE_NOATIME       (1 << __BCH_INODE_NOATIME)
 843 #define BCH_INODE_I_SIZE_DIRTY  (1 << __BCH_INODE_I_SIZE_DIRTY)
 844 #define BCH_INODE_I_SECTORS_DIRTY (1 << __BCH_INODE_I_SECTORS_DIRTY)
 845 #define BCH_INODE_UNLINKED      (1 << __BCH_INODE_UNLINKED)
 846 #define BCH_INODE_BACKPTR_UNTRUSTED (1 << __BCH_INODE_BACKPTR_UNTRUSTED)
 847
 848 LE32_BITMASK(INODE_STR_HASH,    struct bch_inode, bi_flags, 20, 24);
 849 LE32_BITMASK(INODE_NR_FIELDS,   struct bch_inode, bi_flags, 24, 31);
 850 LE32_BITMASK(INODE_NEW_VARINT,  struct bch_inode, bi_flags, 31, 32);
 851
 852 LE64_BITMASK(INODEv2_STR_HASH,  struct bch_inode_v2, bi_flags, 20, 24);
 853 LE64_BITMASK(INODEv2_NR_FIELDS, struct bch_inode_v2, bi_flags, 24, 31);
 854
 855 LE64_BITMASK(INODEv3_STR_HASH,  struct bch_inode_v3, bi_flags, 20, 24);
 856 LE64_BITMASK(INODEv3_NR_FIELDS, struct bch_inode_v3, bi_flags, 24, 31);
 857
 858 LE64_BITMASK(INODEv3_FIELDS_START,
 859                                 struct bch_inode_v3, bi_flags, 31, 36);
 860 LE64_BITMASK(INODEv3_MODE,      struct bch_inode_v3, bi_flags, 36, 52);
 861
 862 /* Dirents */
 863
 864 /*
 865  * Dirents (and xattrs) have to implement string lookups; since our b-tree
 866  * doesn't support arbitrary length strings for the key, we instead index by a
 867  * 64 bit hash (currently truncated sha1) of the string, stored in the offset
 868  * field of the key - using linear probing to resolve hash collisions. This also
 869  * provides us with the readdir cookie posix requires.
 870  *
 871  * Linear probing requires us to use whiteouts for deletions, in the event of a
 872  * collision:
 873  */
 874
 875 struct bch_dirent {
 876         struct bch_val          v;
 877
 878         /* Target inode number: */
 879         union {
 880         __le64                  d_inum;
 881         struct {                /* DT_SUBVOL */
 882         __le32                  d_child_subvol;
 883         __le32                  d_parent_subvol;
 884         };
 885         };
 886
 887         /*
 888          * Copy of mode bits 12-15 from the target inode - so userspace can get
 889          * the filetype without having to do a stat()
 890          */
 891         __u8                    d_type;
 892
 893         __u8                    d_name[];
 894 } __packed __aligned(8);
 895
 896 #define DT_SUBVOL       16
 897 #define BCH_DT_MAX      17
 898
 899 #define BCH_NAME_MAX    ((unsigned) (U8_MAX * sizeof(u64) -             \
 900                          sizeof(struct bkey) -                          \
 901                          offsetof(struct bch_dirent, d_name)))
 902
 903 /* Xattrs */
 904
 905 #define KEY_TYPE_XATTR_INDEX_USER                       0
 906 #define KEY_TYPE_XATTR_INDEX_POSIX_ACL_ACCESS   1
 907 #define KEY_TYPE_XATTR_INDEX_POSIX_ACL_DEFAULT  2
 908 #define KEY_TYPE_XATTR_INDEX_TRUSTED                    3
 909 #define KEY_TYPE_XATTR_INDEX_SECURITY           4
 910
 911 struct bch_xattr {
 912         struct bch_val          v;
 913         __u8                    x_type;
 914         __u8                    x_name_len;
 915         __le16                  x_val_len;
 916         __u8                    x_name[];
 917 } __packed __aligned(8);
 918
 919 /* Bucket/allocation information: */
 920
 921 struct bch_alloc {
 922         struct bch_val          v;
 923         __u8                    fields;
 924         __u8                    gen;
 925         __u8                    data[];
 926 } __packed __aligned(8);
 927
 928 #define BCH_ALLOC_FIELDS_V1()                   \
 929         x(read_time,            16)             \
 930         x(write_time,           16)             \
 931         x(data_type,            8)              \
 932         x(dirty_sectors,        16)             \
 933         x(cached_sectors,       16)             \
 934         x(oldest_gen,           8)              \
 935         x(stripe,               32)             \
 936         x(stripe_redundancy,    8)
 937
 938 enum {
 939 #define x(name, _bits) BCH_ALLOC_FIELD_V1_##name,
 940         BCH_ALLOC_FIELDS_V1()
 941 #undef x
 942 };
 943
 944 struct bch_alloc_v2 {
 945         struct bch_val          v;
 946         __u8                    nr_fields;
 947         __u8                    gen;
 948         __u8                    oldest_gen;
 949         __u8                    data_type;
 950         __u8                    data[];
 951 } __packed __aligned(8);
 952
 953 #define BCH_ALLOC_FIELDS_V2()                   \
 954         x(read_time,            64)             \
 955         x(write_time,           64)             \
 956         x(dirty_sectors,        32)             \
 957         x(cached_sectors,       32)             \
 958         x(stripe,               32)             \
 959         x(stripe_redundancy,    8)
 960
 961 struct bch_alloc_v3 {
 962         struct bch_val          v;
 963         __le64                  journal_seq;
 964         __le32                  flags;
 965         __u8                    nr_fields;
 966         __u8                    gen;
 967         __u8                    oldest_gen;
 968         __u8                    data_type;
 969         __u8                    data[];
 970 } __packed __aligned(8);
 971
 972 LE32_BITMASK(BCH_ALLOC_V3_NEED_DISCARD,struct bch_alloc_v3, flags,  0,  1)
 973 LE32_BITMASK(BCH_ALLOC_V3_NEED_INC_GEN,struct bch_alloc_v3, flags,  1,  2)
 974
 975 struct bch_alloc_v4 {
 976         struct bch_val          v;
 977         __u64                   journal_seq;
 978         __u32                   flags;
 979         __u8                    gen;
 980         __u8                    oldest_gen;
 981         __u8                    data_type;
 982         __u8                    stripe_redundancy;
 983         __u32                   dirty_sectors;
 984         __u32                   cached_sectors;
 985         __u64                   io_time[2];
 986         __u32                   stripe;
 987         __u32                   nr_external_backpointers;
 988 } __packed __aligned(8);
 989
 990 #define BCH_ALLOC_V4_U64s_V0    6
 991 #define BCH_ALLOC_V4_U64s       (sizeof(struct bch_alloc_v4) / sizeof(u64))
 992
 993 BITMASK(BCH_ALLOC_V4_NEED_DISCARD,      struct bch_alloc_v4, flags,  0,  1)
 994 BITMASK(BCH_ALLOC_V4_NEED_INC_GEN,      struct bch_alloc_v4, flags,  1,  2)
 995 BITMASK(BCH_ALLOC_V4_BACKPOINTERS_START,struct bch_alloc_v4, flags,  2,  8)
 996 BITMASK(BCH_ALLOC_V4_NR_BACKPOINTERS,   struct bch_alloc_v4, flags,  8,  14)
 997
 998 #define BCH_ALLOC_V4_NR_BACKPOINTERS_MAX        40
 999
1000 struct bch_backpointer {
1001         struct bch_val          v;
1002         __u8                    btree_id;
1003         __u8                    level;
1004         __u8                    data_type;
1005         __u64                   bucket_offset:40;
1006         __u32                   bucket_len;
1007         struct bpos             pos;
1008 } __packed __aligned(8);
1009
1010 /* Quotas: */
1011
1012 enum quota_types {
1013         QTYP_USR                = 0,
1014         QTYP_GRP                = 1,
1015         QTYP_PRJ                = 2,
1016         QTYP_NR                 = 3,
1017 };
1018
1019 enum quota_counters {
1020         Q_SPC                   = 0,
1021         Q_INO                   = 1,
1022         Q_COUNTERS              = 2,
1023 };
1024
1025 struct bch_quota_counter {
1026         __le64                  hardlimit;
1027         __le64                  softlimit;
1028 };
1029
1030 struct bch_quota {
1031         struct bch_val          v;
1032         struct bch_quota_counter c[Q_COUNTERS];
1033 } __packed __aligned(8);
1034
1035 /* Erasure coding */
1036
1037 struct bch_stripe {
1038         struct bch_val          v;
1039         __le16                  sectors;
1040         __u8                    algorithm;
1041         __u8                    nr_blocks;
1042         __u8                    nr_redundant;
1043
1044         __u8                    csum_granularity_bits;
1045         __u8                    csum_type;
1046         __u8                    pad;
1047
1048         struct bch_extent_ptr   ptrs[];
1049 } __packed __aligned(8);
1050
1051 /* Reflink: */
1052
1053 struct bch_reflink_p {
1054         struct bch_val          v;
1055         __le64                  idx;
1056         /*
1057          * A reflink pointer might point to an indirect extent which is then
1058          * later split (by copygc or rebalance). If we only pointed to part of
1059          * the original indirect extent, and then one of the fragments is
1060          * outside the range we point to, we'd leak a refcount: so when creating
1061          * reflink pointers, we need to store pad values to remember the full
1062          * range we were taking a reference on.
1063          */
1064         __le32                  front_pad;
1065         __le32                  back_pad;
1066 } __packed __aligned(8);
1067
1068 struct bch_reflink_v {
1069         struct bch_val          v;
1070         __le64                  refcount;
1071         union bch_extent_entry  start[0];
1072         __u64                   _data[0];
1073 } __packed __aligned(8);
1074
1075 struct bch_indirect_inline_data {
1076         struct bch_val          v;
1077         __le64                  refcount;
1078         u8                      data[0];
1079 };
1080
1081 /* Inline data */
1082
1083 struct bch_inline_data {
1084         struct bch_val          v;
1085         u8                      data[0];
1086 };
1087
1088 /* Subvolumes: */
1089
1090 #define SUBVOL_POS_MIN          POS(0, 1)
1091 #define SUBVOL_POS_MAX          POS(0, S32_MAX)
1092 #define BCACHEFS_ROOT_SUBVOL    1
1093
1094 struct bch_subvolume {
1095         struct bch_val          v;
1096         __le32                  flags;
1097         __le32                  snapshot;
1098         __le64                  inode;
1099 };
1100
1101 LE32_BITMASK(BCH_SUBVOLUME_RO,          struct bch_subvolume, flags,  0,  1)
1102 /*
1103  * We need to know whether a subvolume is a snapshot so we can know whether we
1104  * can delete it (or whether it should just be rm -rf'd)
1105  */
1106 LE32_BITMASK(BCH_SUBVOLUME_SNAP,        struct bch_subvolume, flags,  1,  2)
1107 LE32_BITMASK(BCH_SUBVOLUME_UNLINKED,    struct bch_subvolume, flags,  2,  3)
1108
1109 /* Snapshots */
1110
1111 struct bch_snapshot {
1112         struct bch_val          v;
1113         __le32                  flags;
1114         __le32                  parent;
1115         __le32                  children[2];
1116         __le32                  subvol;
1117         __le32                  pad;
1118 };
1119
1120 LE32_BITMASK(BCH_SNAPSHOT_DELETED,      struct bch_snapshot, flags,  0,  1)
1121
1122 /* True if a subvolume points to this snapshot node: */
1123 LE32_BITMASK(BCH_SNAPSHOT_SUBVOL,       struct bch_snapshot, flags,  1,  2)
1124
1125 /* LRU btree: */
1126
1127 struct bch_lru {
1128         struct bch_val          v;
1129         __le64                  idx;
1130 } __packed __aligned(8);
1131
1132 #define LRU_ID_STRIPES          (1U << 16)
1133
1134 /* Optional/variable size superblock sections: */
1135
1136 struct bch_sb_field {
1137         __u64                   _data[0];
1138         __le32                  u64s;
1139         __le32                  type;
1140 };
1141
1142 #define BCH_SB_FIELDS()                         \
1143         x(journal,      0)                      \
1144         x(members,      1)                      \
1145         x(crypt,        2)                      \
1146         x(replicas_v0,  3)                      \
1147         x(quota,        4)                      \
1148         x(disk_groups,  5)                      \
1149         x(clean,        6)                      \
1150         x(replicas,     7)                      \
1151         x(journal_seq_blacklist, 8)             \
1152         x(journal_v2,   9)                      \
1153         x(counters,     10)
1154
1155 enum bch_sb_field_type {
1156 #define x(f, nr)        BCH_SB_FIELD_##f = nr,
1157         BCH_SB_FIELDS()
1158 #undef x
1159         BCH_SB_FIELD_NR
1160 };
1161
1162 /*
1163  * Most superblock fields are replicated in all device's superblocks - a few are
1164  * not:
1165  */
1166 #define BCH_SINGLE_DEVICE_SB_FIELDS             \
1167         ((1U << BCH_SB_FIELD_journal)|          \
1168          (1U << BCH_SB_FIELD_journal_v2))
1169
1170 /* BCH_SB_FIELD_journal: */
1171
1172 struct bch_sb_field_journal {
1173         struct bch_sb_field     field;
1174         __le64                  buckets[0];
1175 };
1176
1177 struct bch_sb_field_journal_v2 {
1178         struct bch_sb_field     field;
1179
1180         struct bch_sb_field_journal_v2_entry {
1181                 __le64          start;
1182                 __le64          nr;
1183         }                       d[0];
1184 };
1185
1186 /* BCH_SB_FIELD_members: */
1187
1188 #define BCH_MIN_NR_NBUCKETS     (1 << 6)
1189
1190 struct bch_member {
1191         uuid_le                 uuid;
1192         __le64                  nbuckets;       /* device size */
1193         __le16                  first_bucket;   /* index of first bucket used */
1194         __le16                  bucket_size;    /* sectors */
1195         __le32                  pad;
1196         __le64                  last_mount;     /* time_t */
1197
1198         __le64                  flags[2];
1199 };
1200
1201 LE64_BITMASK(BCH_MEMBER_STATE,          struct bch_member, flags[0],  0,  4)
1202 /* 4-14 unused, was TIER, HAS_(META)DATA, REPLACEMENT */
1203 LE64_BITMASK(BCH_MEMBER_DISCARD,        struct bch_member, flags[0], 14, 15)
1204 LE64_BITMASK(BCH_MEMBER_DATA_ALLOWED,   struct bch_member, flags[0], 15, 20)
1205 LE64_BITMASK(BCH_MEMBER_GROUP,          struct bch_member, flags[0], 20, 28)
1206 LE64_BITMASK(BCH_MEMBER_DURABILITY,     struct bch_member, flags[0], 28, 30)
1207 LE64_BITMASK(BCH_MEMBER_FREESPACE_INITIALIZED,
1208                                         struct bch_member, flags[0], 30, 31)
1209
1210 #if 0
1211 LE64_BITMASK(BCH_MEMBER_NR_READ_ERRORS, struct bch_member, flags[1], 0,  20);
1212 LE64_BITMASK(BCH_MEMBER_NR_WRITE_ERRORS,struct bch_member, flags[1], 20, 40);
1213 #endif
1214
1215 #define BCH_MEMBER_STATES()                     \
1216         x(rw,           0)                      \
1217         x(ro,           1)                      \
1218         x(failed,       2)                      \
1219         x(spare,        3)
1220
1221 enum bch_member_state {
1222 #define x(t, n) BCH_MEMBER_STATE_##t = n,
1223         BCH_MEMBER_STATES()
1224 #undef x
1225         BCH_MEMBER_STATE_NR
1226 };
1227
1228 struct bch_sb_field_members {
1229         struct bch_sb_field     field;
1230         struct bch_member       members[0];
1231 };
1232
1233 /* BCH_SB_FIELD_crypt: */
1234
1235 struct nonce {
1236         __le32                  d[4];
1237 };
1238
1239 struct bch_key {
1240         __le64                  key[4];
1241 };
1242
1243 #define BCH_KEY_MAGIC                                   \
1244         (((u64) 'b' <<  0)|((u64) 'c' <<  8)|           \
1245          ((u64) 'h' << 16)|((u64) '*' << 24)|           \
1246          ((u64) '*' << 32)|((u64) 'k' << 40)|           \
1247          ((u64) 'e' << 48)|((u64) 'y' << 56))
1248
1249 struct bch_encrypted_key {
1250         __le64                  magic;
1251         struct bch_key          key;
1252 };
1253
1254 /*
1255  * If this field is present in the superblock, it stores an encryption key which
1256  * is used encrypt all other data/metadata. The key will normally be encrypted
1257  * with the key userspace provides, but if encryption has been turned off we'll
1258  * just store the master key unencrypted in the superblock so we can access the
1259  * previously encrypted data.
1260  */
1261 struct bch_sb_field_crypt {
1262         struct bch_sb_field     field;
1263
1264         __le64                  flags;
1265         __le64                  kdf_flags;
1266         struct bch_encrypted_key key;
1267 };
1268
1269 LE64_BITMASK(BCH_CRYPT_KDF_TYPE,        struct bch_sb_field_crypt, flags, 0, 4);
1270
1271 enum bch_kdf_types {
1272         BCH_KDF_SCRYPT          = 0,
1273         BCH_KDF_NR              = 1,
1274 };
1275
1276 /* stored as base 2 log of scrypt params: */
1277 LE64_BITMASK(BCH_KDF_SCRYPT_N,  struct bch_sb_field_crypt, kdf_flags,  0, 16);
1278 LE64_BITMASK(BCH_KDF_SCRYPT_R,  struct bch_sb_field_crypt, kdf_flags, 16, 32);
1279 LE64_BITMASK(BCH_KDF_SCRYPT_P,  struct bch_sb_field_crypt, kdf_flags, 32, 48);
1280
1281 /* BCH_SB_FIELD_replicas: */
1282
1283 #define BCH_DATA_TYPES()                \
1284         x(free,         0)              \
1285         x(sb,           1)              \
1286         x(journal,      2)              \
1287         x(btree,        3)              \
1288         x(user,         4)              \
1289         x(cached,       5)              \
1290         x(parity,       6)              \
1291         x(stripe,       7)              \
1292         x(need_gc_gens, 8)              \
1293         x(need_discard, 9)
1294
1295 enum bch_data_type {
1296 #define x(t, n) BCH_DATA_##t,
1297         BCH_DATA_TYPES()
1298 #undef x
1299         BCH_DATA_NR
1300 };
1301
1302 static inline bool data_type_is_empty(enum bch_data_type type)
1303 {
1304         switch (type) {
1305         case BCH_DATA_free:
1306         case BCH_DATA_need_gc_gens:
1307         case BCH_DATA_need_discard:
1308                 return true;
1309         default:
1310                 return false;
1311         }
1312 }
1313
1314 static inline bool data_type_is_hidden(enum bch_data_type type)
1315 {
1316         switch (type) {
1317         case BCH_DATA_sb:
1318         case BCH_DATA_journal:
1319                 return true;
1320         default:
1321                 return false;
1322         }
1323 }
1324
1325 struct bch_replicas_entry_v0 {
1326         __u8                    data_type;
1327         __u8                    nr_devs;
1328         __u8                    devs[];
1329 } __packed;
1330
1331 struct bch_sb_field_replicas_v0 {
1332         struct bch_sb_field     field;
1333         struct bch_replicas_entry_v0 entries[];
1334 } __packed __aligned(8);
1335
1336 struct bch_replicas_entry {
1337         __u8                    data_type;
1338         __u8                    nr_devs;
1339         __u8                    nr_required;
1340         __u8                    devs[];
1341 } __packed;
1342
1343 #define replicas_entry_bytes(_i)                                        \
1344         (offsetof(typeof(*(_i)), devs) + (_i)->nr_devs)
1345
1346 struct bch_sb_field_replicas {
1347         struct bch_sb_field     field;
1348         struct bch_replicas_entry entries[0];
1349 } __packed __aligned(8);
1350
1351 /* BCH_SB_FIELD_quota: */
1352
1353 struct bch_sb_quota_counter {
1354         __le32                          timelimit;
1355         __le32                          warnlimit;
1356 };
1357
1358 struct bch_sb_quota_type {
1359         __le64                          flags;
1360         struct bch_sb_quota_counter     c[Q_COUNTERS];
1361 };
1362
1363 struct bch_sb_field_quota {
1364         struct bch_sb_field             field;
1365         struct bch_sb_quota_type        q[QTYP_NR];
1366 } __packed __aligned(8);
1367
1368 /* BCH_SB_FIELD_disk_groups: */
1369
1370 #define BCH_SB_LABEL_SIZE               32
1371
1372 struct bch_disk_group {
1373         __u8                    label[BCH_SB_LABEL_SIZE];
1374         __le64                  flags[2];
1375 } __packed __aligned(8);
1376
1377 LE64_BITMASK(BCH_GROUP_DELETED,         struct bch_disk_group, flags[0], 0,  1)
1378 LE64_BITMASK(BCH_GROUP_DATA_ALLOWED,    struct bch_disk_group, flags[0], 1,  6)
1379 LE64_BITMASK(BCH_GROUP_PARENT,          struct bch_disk_group, flags[0], 6, 24)
1380
1381 struct bch_sb_field_disk_groups {
1382         struct bch_sb_field     field;
1383         struct bch_disk_group   entries[0];
1384 } __packed __aligned(8);
1385
1386 /* BCH_SB_FIELD_counters */
1387
1388 #define BCH_PERSISTENT_COUNTERS()                               \
1389         x(io_read,                                      0)      \
1390         x(io_write,                                     1)      \
1391         x(io_move,                                      2)      \
1392         x(bucket_invalidate,                            3)      \
1393         x(bucket_discard,                               4)      \
1394         x(bucket_alloc,                                 5)      \
1395         x(bucket_alloc_fail,                            6)      \
1396         x(btree_cache_scan,                             7)      \
1397         x(btree_cache_reap,                             8)      \
1398         x(btree_cache_cannibalize,                      9)      \
1399         x(btree_cache_cannibalize_lock,                 10)     \
1400         x(btree_cache_cannibalize_lock_fail,            11)     \
1401         x(btree_cache_cannibalize_unlock,               12)     \
1402         x(btree_node_write,                             13)     \
1403         x(btree_node_read,                              14)     \
1404         x(btree_node_compact,                           15)     \
1405         x(btree_node_merge,                             16)     \
1406         x(btree_node_split,                             17)     \
1407         x(btree_node_rewrite,                           18)     \
1408         x(btree_node_alloc,                             19)     \
1409         x(btree_node_free,                              20)     \
1410         x(btree_node_set_root,                          21)     \
1411         x(btree_path_relock_fail,                       22)     \
1412         x(btree_path_upgrade_fail,                      23)     \
1413         x(btree_reserve_get_fail,                       24)     \
1414         x(journal_entry_full,                           25)     \
1415         x(journal_full,                                 26)     \
1416         x(journal_reclaim_finish,                       27)     \
1417         x(journal_reclaim_start,                        28)     \
1418         x(journal_write,                                29)     \
1419         x(read_promote,                                 30)     \
1420         x(read_bounce,                                  31)     \
1421         x(read_split,                                   33)     \
1422         x(read_retry,                                   32)     \
1423         x(read_reuse_race,                              34)     \
1424         x(move_extent_read,                             35)     \
1425         x(move_extent_write,                            36)     \
1426         x(move_extent_finish,                           37)     \
1427         x(move_extent_race,                             38)     \
1428         x(move_extent_alloc_mem_fail,                   39)     \
1429         x(copygc,                                       40)     \
1430         x(copygc_wait,                                  41)     \
1431         x(gc_gens_end,                                  42)     \
1432         x(gc_gens_start,                                43)     \
1433         x(trans_blocked_journal_reclaim,                44)     \
1434         x(trans_restart_btree_node_reused,              45)     \
1435         x(trans_restart_btree_node_split,               46)     \
1436         x(trans_restart_fault_inject,                   47)     \
1437         x(trans_restart_iter_upgrade,                   48)     \
1438         x(trans_restart_journal_preres_get,             49)     \
1439         x(trans_restart_journal_reclaim,                50)     \
1440         x(trans_restart_journal_res_get,                51)     \
1441         x(trans_restart_key_cache_key_realloced,        52)     \
1442         x(trans_restart_key_cache_raced,                53)     \
1443         x(trans_restart_mark_replicas,                  54)     \
1444         x(trans_restart_mem_realloced,                  55)     \
1445         x(trans_restart_memory_allocation_failure,      56)     \
1446         x(trans_restart_relock,                         57)     \
1447         x(trans_restart_relock_after_fill,              58)     \
1448         x(trans_restart_relock_key_cache_fill,          59)     \
1449         x(trans_restart_relock_next_node,               60)     \
1450         x(trans_restart_relock_parent_for_fill,         61)     \
1451         x(trans_restart_relock_path,                    62)     \
1452         x(trans_restart_relock_path_intent,             63)     \
1453         x(trans_restart_too_many_iters,                 64)     \
1454         x(trans_restart_traverse,                       65)     \
1455         x(trans_restart_upgrade,                        66)     \
1456         x(trans_restart_would_deadlock,                 67)     \
1457         x(trans_restart_would_deadlock_write,           68)     \
1458         x(trans_restart_injected,                       69)     \
1459         x(trans_restart_key_cache_upgrade,              70)     \
1460         x(trans_traverse_all,                           71)     \
1461         x(transaction_commit,                           72)     \
1462         x(write_super,                                  73)     \
1463         x(trans_restart_would_deadlock_recursion_limit, 74)
1464
1465 enum bch_persistent_counters {
1466 #define x(t, n, ...) BCH_COUNTER_##t,
1467         BCH_PERSISTENT_COUNTERS()
1468 #undef x
1469         BCH_COUNTER_NR
1470 };
1471
1472 struct bch_sb_field_counters {
1473         struct bch_sb_field     field;
1474         __le64                  d[0];
1475 };
1476
1477 /*
1478  * On clean shutdown, store btree roots and current journal sequence number in
1479  * the superblock:
1480  */
1481 struct jset_entry {
1482         __le16                  u64s;
1483         __u8                    btree_id;
1484         __u8                    level;
1485         __u8                    type; /* designates what this jset holds */
1486         __u8                    pad[3];
1487
1488         union {
1489                 struct bkey_i   start[0];
1490                 __u64           _data[0];
1491         };
1492 };
1493
1494 struct bch_sb_field_clean {
1495         struct bch_sb_field     field;
1496
1497         __le32                  flags;
1498         __le16                  _read_clock; /* no longer used */
1499         __le16                  _write_clock;
1500         __le64                  journal_seq;
1501
1502         union {
1503                 struct jset_entry start[0];
1504                 __u64           _data[0];
1505         };
1506 };
1507
1508 struct journal_seq_blacklist_entry {
1509         __le64                  start;
1510         __le64                  end;
1511 };
1512
1513 struct bch_sb_field_journal_seq_blacklist {
1514         struct bch_sb_field     field;
1515
1516         union {
1517                 struct journal_seq_blacklist_entry start[0];
1518                 __u64           _data[0];
1519         };
1520 };
1521
1522 /* Superblock: */
1523
1524 /*
1525  * New versioning scheme:
1526  * One common version number for all on disk data structures - superblock, btree
1527  * nodes, journal entries
1528  */
1529 #define BCH_JSET_VERSION_OLD                    2
1530 #define BCH_BSET_VERSION_OLD                    3
1531
1532 #define BCH_METADATA_VERSIONS()                         \
1533         x(bkey_renumber,                10)             \
1534         x(inode_btree_change,           11)             \
1535         x(snapshot,                     12)             \
1536         x(inode_backpointers,           13)             \
1537         x(btree_ptr_sectors_written,    14)             \
1538         x(snapshot_2,                   15)             \
1539         x(reflink_p_fix,                16)             \
1540         x(subvol_dirent,                17)             \
1541         x(inode_v2,                     18)             \
1542         x(freespace,                    19)             \
1543         x(alloc_v4,                     20)             \
1544         x(new_data_types,               21)             \
1545         x(backpointers,                 22)             \
1546         x(inode_v3,                     23)
1547
1548 enum bcachefs_metadata_version {
1549         bcachefs_metadata_version_min = 9,
1550 #define x(t, n) bcachefs_metadata_version_##t = n,
1551         BCH_METADATA_VERSIONS()
1552 #undef x
1553         bcachefs_metadata_version_max
1554 };
1555
1556 #define bcachefs_metadata_version_current       (bcachefs_metadata_version_max - 1)
1557
1558 #define BCH_SB_SECTOR                   8
1559 #define BCH_SB_MEMBERS_MAX              64 /* XXX kill */
1560
1561 struct bch_sb_layout {
1562         uuid_le                 magic;  /* bcachefs superblock UUID */
1563         __u8                    layout_type;
1564         __u8                    sb_max_size_bits; /* base 2 of 512 byte sectors */
1565         __u8                    nr_superblocks;
1566         __u8                    pad[5];
1567         __le64                  sb_offset[61];
1568 } __packed __aligned(8);
1569
1570 #define BCH_SB_LAYOUT_SECTOR    7
1571
1572 /*
1573  * @offset      - sector where this sb was written
1574  * @version     - on disk format version
1575  * @version_min - Oldest metadata version this filesystem contains; so we can
1576  *                safely drop compatibility code and refuse to mount filesystems
1577  *                we'd need it for
1578  * @magic       - identifies as a bcachefs superblock (BCACHE_MAGIC)
1579  * @seq         - incremented each time superblock is written
1580  * @uuid        - used for generating various magic numbers and identifying
1581  *                member devices, never changes
1582  * @user_uuid   - user visible UUID, may be changed
1583  * @label       - filesystem label
1584  * @seq         - identifies most recent superblock, incremented each time
1585  *                superblock is written
1586  * @features    - enabled incompatible features
1587  */
1588 struct bch_sb {
1589         struct bch_csum         csum;
1590         __le16                  version;
1591         __le16                  version_min;
1592         __le16                  pad[2];
1593         uuid_le                 magic;
1594         uuid_le                 uuid;
1595         uuid_le                 user_uuid;
1596         __u8                    label[BCH_SB_LABEL_SIZE];
1597         __le64                  offset;
1598         __le64                  seq;
1599
1600         __le16                  block_size;
1601         __u8                    dev_idx;
1602         __u8                    nr_devices;
1603         __le32                  u64s;
1604
1605         __le64                  time_base_lo;
1606         __le32                  time_base_hi;
1607         __le32                  time_precision;
1608
1609         __le64                  flags[8];
1610         __le64                  features[2];
1611         __le64                  compat[2];
1612
1613         struct bch_sb_layout    layout;
1614
1615         union {
1616                 struct bch_sb_field start[0];
1617                 __le64          _data[0];
1618         };
1619 } __packed __aligned(8);
1620
1621 /*
1622  * Flags:
1623  * BCH_SB_INITALIZED    - set on first mount
1624  * BCH_SB_CLEAN         - did we shut down cleanly? Just a hint, doesn't affect
1625  *                        behaviour of mount/recovery path:
1626  * BCH_SB_INODE_32BIT   - limit inode numbers to 32 bits
1627  * BCH_SB_128_BIT_MACS  - 128 bit macs instead of 80
1628  * BCH_SB_ENCRYPTION_TYPE - if nonzero encryption is enabled; overrides
1629  *                         DATA/META_CSUM_TYPE. Also indicates encryption
1630  *                         algorithm in use, if/when we get more than one
1631  */
1632
1633 LE16_BITMASK(BCH_SB_BLOCK_SIZE,         struct bch_sb, block_size, 0, 16);
1634
1635 LE64_BITMASK(BCH_SB_INITIALIZED,        struct bch_sb, flags[0],  0,  1);
1636 LE64_BITMASK(BCH_SB_CLEAN,              struct bch_sb, flags[0],  1,  2);
1637 LE64_BITMASK(BCH_SB_CSUM_TYPE,          struct bch_sb, flags[0],  2,  8);
1638 LE64_BITMASK(BCH_SB_ERROR_ACTION,       struct bch_sb, flags[0],  8, 12);
1639
1640 LE64_BITMASK(BCH_SB_BTREE_NODE_SIZE,    struct bch_sb, flags[0], 12, 28);
1641
1642 LE64_BITMASK(BCH_SB_GC_RESERVE,         struct bch_sb, flags[0], 28, 33);
1643 LE64_BITMASK(BCH_SB_ROOT_RESERVE,       struct bch_sb, flags[0], 33, 40);
1644
1645 LE64_BITMASK(BCH_SB_META_CSUM_TYPE,     struct bch_sb, flags[0], 40, 44);
1646 LE64_BITMASK(BCH_SB_DATA_CSUM_TYPE,     struct bch_sb, flags[0], 44, 48);
1647
1648 LE64_BITMASK(BCH_SB_META_REPLICAS_WANT, struct bch_sb, flags[0], 48, 52);
1649 LE64_BITMASK(BCH_SB_DATA_REPLICAS_WANT, struct bch_sb, flags[0], 52, 56);
1650
1651 LE64_BITMASK(BCH_SB_POSIX_ACL,          struct bch_sb, flags[0], 56, 57);
1652 LE64_BITMASK(BCH_SB_USRQUOTA,           struct bch_sb, flags[0], 57, 58);
1653 LE64_BITMASK(BCH_SB_GRPQUOTA,           struct bch_sb, flags[0], 58, 59);
1654 LE64_BITMASK(BCH_SB_PRJQUOTA,           struct bch_sb, flags[0], 59, 60);
1655
1656 LE64_BITMASK(BCH_SB_HAS_ERRORS,         struct bch_sb, flags[0], 60, 61);
1657 LE64_BITMASK(BCH_SB_HAS_TOPOLOGY_ERRORS,struct bch_sb, flags[0], 61, 62);
1658
1659 LE64_BITMASK(BCH_SB_BIG_ENDIAN,         struct bch_sb, flags[0], 62, 63);
1660
1661 LE64_BITMASK(BCH_SB_STR_HASH_TYPE,      struct bch_sb, flags[1],  0,  4);
1662 LE64_BITMASK(BCH_SB_COMPRESSION_TYPE,   struct bch_sb, flags[1],  4,  8);
1663 LE64_BITMASK(BCH_SB_INODE_32BIT,        struct bch_sb, flags[1],  8,  9);
1664
1665 LE64_BITMASK(BCH_SB_128_BIT_MACS,       struct bch_sb, flags[1],  9, 10);
1666 LE64_BITMASK(BCH_SB_ENCRYPTION_TYPE,    struct bch_sb, flags[1], 10, 14);
1667
1668 /*
1669  * Max size of an extent that may require bouncing to read or write
1670  * (checksummed, compressed): 64k
1671  */
1672 LE64_BITMASK(BCH_SB_ENCODED_EXTENT_MAX_BITS,
1673                                         struct bch_sb, flags[1], 14, 20);
1674
1675 LE64_BITMASK(BCH_SB_META_REPLICAS_REQ,  struct bch_sb, flags[1], 20, 24);
1676 LE64_BITMASK(BCH_SB_DATA_REPLICAS_REQ,  struct bch_sb, flags[1], 24, 28);
1677
1678 LE64_BITMASK(BCH_SB_PROMOTE_TARGET,     struct bch_sb, flags[1], 28, 40);
1679 LE64_BITMASK(BCH_SB_FOREGROUND_TARGET,  struct bch_sb, flags[1], 40, 52);
1680 LE64_BITMASK(BCH_SB_BACKGROUND_TARGET,  struct bch_sb, flags[1], 52, 64);
1681
1682 LE64_BITMASK(BCH_SB_BACKGROUND_COMPRESSION_TYPE,
1683                                         struct bch_sb, flags[2],  0,  4);
1684 LE64_BITMASK(BCH_SB_GC_RESERVE_BYTES,   struct bch_sb, flags[2],  4, 64);
1685
1686 LE64_BITMASK(BCH_SB_ERASURE_CODE,       struct bch_sb, flags[3],  0, 16);
1687 LE64_BITMASK(BCH_SB_METADATA_TARGET,    struct bch_sb, flags[3], 16, 28);
1688 LE64_BITMASK(BCH_SB_SHARD_INUMS,        struct bch_sb, flags[3], 28, 29);
1689 LE64_BITMASK(BCH_SB_INODES_USE_KEY_CACHE,struct bch_sb, flags[3], 29, 30);
1690 LE64_BITMASK(BCH_SB_JOURNAL_FLUSH_DELAY,struct bch_sb, flags[3], 30, 62);
1691 LE64_BITMASK(BCH_SB_JOURNAL_FLUSH_DISABLED,struct bch_sb, flags[3], 62, 63);
1692 LE64_BITMASK(BCH_SB_JOURNAL_RECLAIM_DELAY,struct bch_sb, flags[4], 0, 32);
1693 /* Obsolete, always enabled: */
1694 LE64_BITMASK(BCH_SB_JOURNAL_TRANSACTION_NAMES,struct bch_sb, flags[4], 32, 33);
1695
1696 /*
1697  * Features:
1698  *
1699  * journal_seq_blacklist_v3:    gates BCH_SB_FIELD_journal_seq_blacklist
1700  * reflink:                     gates KEY_TYPE_reflink
1701  * inline_data:                 gates KEY_TYPE_inline_data
1702  * new_siphash:                 gates BCH_STR_HASH_siphash
1703  * new_extent_overwrite:        gates BTREE_NODE_NEW_EXTENT_OVERWRITE
1704  */
1705 #define BCH_SB_FEATURES()                       \
1706         x(lz4,                          0)      \
1707         x(gzip,                         1)      \
1708         x(zstd,                         2)      \
1709         x(atomic_nlink,                 3)      \
1710         x(ec,                           4)      \
1711         x(journal_seq_blacklist_v3,     5)      \
1712         x(reflink,                      6)      \
1713         x(new_siphash,                  7)      \
1714         x(inline_data,                  8)      \
1715         x(new_extent_overwrite,         9)      \
1716         x(incompressible,               10)     \
1717         x(btree_ptr_v2,                 11)     \
1718         x(extents_above_btree_updates,  12)     \
1719         x(btree_updates_journalled,     13)     \
1720         x(reflink_inline_data,          14)     \
1721         x(new_varint,                   15)     \
1722         x(journal_no_flush,             16)     \
1723         x(alloc_v2,                     17)     \
1724         x(extents_across_btree_nodes,   18)
1725
1726 #define BCH_SB_FEATURES_ALWAYS                          \
1727         ((1ULL << BCH_FEATURE_new_extent_overwrite)|    \
1728          (1ULL << BCH_FEATURE_extents_above_btree_updates)|\
1729          (1ULL << BCH_FEATURE_btree_updates_journalled)|\
1730          (1ULL << BCH_FEATURE_alloc_v2)|\
1731          (1ULL << BCH_FEATURE_extents_across_btree_nodes))
1732
1733 #define BCH_SB_FEATURES_ALL                             \
1734         (BCH_SB_FEATURES_ALWAYS|                        \
1735          (1ULL << BCH_FEATURE_new_siphash)|             \
1736          (1ULL << BCH_FEATURE_btree_ptr_v2)|            \
1737          (1ULL << BCH_FEATURE_new_varint)|              \
1738          (1ULL << BCH_FEATURE_journal_no_flush))
1739
1740 enum bch_sb_feature {
1741 #define x(f, n) BCH_FEATURE_##f,
1742         BCH_SB_FEATURES()
1743 #undef x
1744         BCH_FEATURE_NR,
1745 };
1746
1747 #define BCH_SB_COMPAT()                                 \
1748         x(alloc_info,                           0)      \
1749         x(alloc_metadata,                       1)      \
1750         x(extents_above_btree_updates_done,     2)      \
1751         x(bformat_overflow_done,                3)
1752
1753 enum bch_sb_compat {
1754 #define x(f, n) BCH_COMPAT_##f,
1755         BCH_SB_COMPAT()
1756 #undef x
1757         BCH_COMPAT_NR,
1758 };
1759
1760 /* options: */
1761
1762 #define BCH_REPLICAS_MAX                4U
1763
1764 #define BCH_BKEY_PTRS_MAX               16U
1765
1766 #define BCH_ERROR_ACTIONS()             \
1767         x(continue,             0)      \
1768         x(ro,                   1)      \
1769         x(panic,                2)
1770
1771 enum bch_error_actions {
1772 #define x(t, n) BCH_ON_ERROR_##t = n,
1773         BCH_ERROR_ACTIONS()
1774 #undef x
1775         BCH_ON_ERROR_NR
1776 };
1777
1778 #define BCH_STR_HASH_TYPES()            \
1779         x(crc32c,               0)      \
1780         x(crc64,                1)      \
1781         x(siphash_old,          2)      \
1782         x(siphash,              3)
1783
1784 enum bch_str_hash_type {
1785 #define x(t, n) BCH_STR_HASH_##t = n,
1786         BCH_STR_HASH_TYPES()
1787 #undef x
1788         BCH_STR_HASH_NR
1789 };
1790
1791 #define BCH_STR_HASH_OPTS()             \
1792         x(crc32c,               0)      \
1793         x(crc64,                1)      \
1794         x(siphash,              2)
1795
1796 enum bch_str_hash_opts {
1797 #define x(t, n) BCH_STR_HASH_OPT_##t = n,
1798         BCH_STR_HASH_OPTS()
1799 #undef x
1800         BCH_STR_HASH_OPT_NR
1801 };
1802
1803 #define BCH_CSUM_TYPES()                        \
1804         x(none,                         0)      \
1805         x(crc32c_nonzero,               1)      \
1806         x(crc64_nonzero,                2)      \
1807         x(chacha20_poly1305_80,         3)      \
1808         x(chacha20_poly1305_128,        4)      \
1809         x(crc32c,                       5)      \
1810         x(crc64,                        6)      \
1811         x(xxhash,                       7)
1812
1813 enum bch_csum_type {
1814 #define x(t, n) BCH_CSUM_##t = n,
1815         BCH_CSUM_TYPES()
1816 #undef x
1817         BCH_CSUM_NR
1818 };
1819
1820 static const unsigned bch_crc_bytes[] = {
1821         [BCH_CSUM_none]                         = 0,
1822         [BCH_CSUM_crc32c_nonzero]               = 4,
1823         [BCH_CSUM_crc32c]                       = 4,
1824         [BCH_CSUM_crc64_nonzero]                = 8,
1825         [BCH_CSUM_crc64]                        = 8,
1826         [BCH_CSUM_xxhash]                       = 8,
1827         [BCH_CSUM_chacha20_poly1305_80]         = 10,
1828         [BCH_CSUM_chacha20_poly1305_128]        = 16,
1829 };
1830
1831 static inline _Bool bch2_csum_type_is_encryption(enum bch_csum_type type)
1832 {
1833         switch (type) {
1834         case BCH_CSUM_chacha20_poly1305_80:
1835         case BCH_CSUM_chacha20_poly1305_128:
1836                 return true;
1837         default:
1838                 return false;
1839         }
1840 }
1841
1842 #define BCH_CSUM_OPTS()                 \
1843         x(none,                 0)      \
1844         x(crc32c,               1)      \
1845         x(crc64,                2)      \
1846         x(xxhash,               3)
1847
1848 enum bch_csum_opts {
1849 #define x(t, n) BCH_CSUM_OPT_##t = n,
1850         BCH_CSUM_OPTS()
1851 #undef x
1852         BCH_CSUM_OPT_NR
1853 };
1854
1855 #define BCH_COMPRESSION_TYPES()         \
1856         x(none,                 0)      \
1857         x(lz4_old,              1)      \
1858         x(gzip,                 2)      \
1859         x(lz4,                  3)      \
1860         x(zstd,                 4)      \
1861         x(incompressible,       5)
1862
1863 enum bch_compression_type {
1864 #define x(t, n) BCH_COMPRESSION_TYPE_##t = n,
1865         BCH_COMPRESSION_TYPES()
1866 #undef x
1867         BCH_COMPRESSION_TYPE_NR
1868 };
1869
1870 #define BCH_COMPRESSION_OPTS()          \
1871         x(none,         0)              \
1872         x(lz4,          1)              \
1873         x(gzip,         2)              \
1874         x(zstd,         3)
1875
1876 enum bch_compression_opts {
1877 #define x(t, n) BCH_COMPRESSION_OPT_##t = n,
1878         BCH_COMPRESSION_OPTS()
1879 #undef x
1880         BCH_COMPRESSION_OPT_NR
1881 };
1882
1883 /*
1884  * Magic numbers
1885  *
1886  * The various other data structures have their own magic numbers, which are
1887  * xored with the first part of the cache set's UUID
1888  */
1889
1890 #define BCACHE_MAGIC                                                    \
1891         UUID_LE(0xf67385c6, 0x1a4e, 0xca45,                             \
1892                 0x82, 0x65, 0xf5, 0x7f, 0x48, 0xba, 0x6d, 0x81)
1893
1894 #define BCACHEFS_STATFS_MAGIC           0xca451a4e
1895
1896 #define JSET_MAGIC              __cpu_to_le64(0x245235c1a3625032ULL)
1897 #define BSET_MAGIC              __cpu_to_le64(0x90135c78b99e07f5ULL)
1898
1899 static inline __le64 __bch2_sb_magic(struct bch_sb *sb)
1900 {
1901         __le64 ret;
1902
1903         memcpy(&ret, &sb->uuid, sizeof(ret));
1904         return ret;
1905 }
1906
1907 static inline __u64 __jset_magic(struct bch_sb *sb)
1908 {
1909         return __le64_to_cpu(__bch2_sb_magic(sb) ^ JSET_MAGIC);
1910 }
1911
1912 static inline __u64 __bset_magic(struct bch_sb *sb)
1913 {
1914         return __le64_to_cpu(__bch2_sb_magic(sb) ^ BSET_MAGIC);
1915 }
1916
1917 /* Journal */
1918
1919 #define JSET_KEYS_U64s  (sizeof(struct jset_entry) / sizeof(__u64))
1920
1921 #define BCH_JSET_ENTRY_TYPES()                  \
1922         x(btree_keys,           0)              \
1923         x(btree_root,           1)              \
1924         x(prio_ptrs,            2)              \
1925         x(blacklist,            3)              \
1926         x(blacklist_v2,         4)              \
1927         x(usage,                5)              \
1928         x(data_usage,           6)              \
1929         x(clock,                7)              \
1930         x(dev_usage,            8)              \
1931         x(log,                  9)              \
1932         x(overwrite,            10)
1933
1934 enum {
1935 #define x(f, nr)        BCH_JSET_ENTRY_##f      = nr,
1936         BCH_JSET_ENTRY_TYPES()
1937 #undef x
1938         BCH_JSET_ENTRY_NR
1939 };
1940
1941 /*
1942  * Journal sequence numbers can be blacklisted: bsets record the max sequence
1943  * number of all the journal entries they contain updates for, so that on
1944  * recovery we can ignore those bsets that contain index updates newer that what
1945  * made it into the journal.
1946  *
1947  * This means that we can't reuse that journal_seq - we have to skip it, and
1948  * then record that we skipped it so that the next time we crash and recover we
1949  * don't think there was a missing journal entry.
1950  */
1951 struct jset_entry_blacklist {
1952         struct jset_entry       entry;
1953         __le64                  seq;
1954 };
1955
1956 struct jset_entry_blacklist_v2 {
1957         struct jset_entry       entry;
1958         __le64                  start;
1959         __le64                  end;
1960 };
1961
1962 #define BCH_FS_USAGE_TYPES()                    \
1963         x(reserved,             0)              \
1964         x(inodes,               1)              \
1965         x(key_version,          2)
1966
1967 enum {
1968 #define x(f, nr)        BCH_FS_USAGE_##f        = nr,
1969         BCH_FS_USAGE_TYPES()
1970 #undef x
1971         BCH_FS_USAGE_NR
1972 };
1973
1974 struct jset_entry_usage {
1975         struct jset_entry       entry;
1976         __le64                  v;
1977 } __packed;
1978
1979 struct jset_entry_data_usage {
1980         struct jset_entry       entry;
1981         __le64                  v;
1982         struct bch_replicas_entry r;
1983 } __packed;
1984
1985 struct jset_entry_clock {
1986         struct jset_entry       entry;
1987         __u8                    rw;
1988         __u8                    pad[7];
1989         __le64                  time;
1990 } __packed;
1991
1992 struct jset_entry_dev_usage_type {
1993         __le64                  buckets;
1994         __le64                  sectors;
1995         __le64                  fragmented;
1996 } __packed;
1997
1998 struct jset_entry_dev_usage {
1999         struct jset_entry       entry;
2000         __le32                  dev;
2001         __u32                   pad;
2002
2003         __le64                  buckets_ec;
2004         __le64                  _buckets_unavailable; /* No longer used */
2005
2006         struct jset_entry_dev_usage_type d[];
2007 } __packed;
2008
2009 static inline unsigned jset_entry_dev_usage_nr_types(struct jset_entry_dev_usage *u)
2010 {
2011         return (vstruct_bytes(&u->entry) - sizeof(struct jset_entry_dev_usage)) /
2012                 sizeof(struct jset_entry_dev_usage_type);
2013 }
2014
2015 struct jset_entry_log {
2016         struct jset_entry       entry;
2017         u8                      d[];
2018 } __packed;
2019
2020 /*
2021  * On disk format for a journal entry:
2022  * seq is monotonically increasing; every journal entry has its own unique
2023  * sequence number.
2024  *
2025  * last_seq is the oldest journal entry that still has keys the btree hasn't
2026  * flushed to disk yet.
2027  *
2028  * version is for on disk format changes.
2029  */
2030 struct jset {
2031         struct bch_csum         csum;
2032
2033         __le64                  magic;
2034         __le64                  seq;
2035         __le32                  version;
2036         __le32                  flags;
2037
2038         __le32                  u64s; /* size of d[] in u64s */
2039
2040         __u8                    encrypted_start[0];
2041
2042         __le16                  _read_clock; /* no longer used */
2043         __le16                  _write_clock;
2044
2045         /* Sequence number of oldest dirty journal entry */
2046         __le64                  last_seq;
2047
2048
2049         union {
2050                 struct jset_entry start[0];
2051                 __u64           _data[0];
2052         };
2053 } __packed __aligned(8);
2054
2055 LE32_BITMASK(JSET_CSUM_TYPE,    struct jset, flags, 0, 4);
2056 LE32_BITMASK(JSET_BIG_ENDIAN,   struct jset, flags, 4, 5);
2057 LE32_BITMASK(JSET_NO_FLUSH,     struct jset, flags, 5, 6);
2058
2059 #define BCH_JOURNAL_BUCKETS_MIN         8
2060
2061 /* Btree: */
2062
2063 #define BCH_BTREE_IDS()                         \
2064         x(extents,      0)                      \
2065         x(inodes,       1)                      \
2066         x(dirents,      2)                      \
2067         x(xattrs,       3)                      \
2068         x(alloc,        4)                      \
2069         x(quotas,       5)                      \
2070         x(stripes,      6)                      \
2071         x(reflink,      7)                      \
2072         x(subvolumes,   8)                      \
2073         x(snapshots,    9)                      \
2074         x(lru,          10)                     \
2075         x(freespace,    11)                     \
2076         x(need_discard, 12)                     \
2077         x(backpointers, 13)
2078
2079 enum btree_id {
2080 #define x(kwd, val) BTREE_ID_##kwd = val,
2081         BCH_BTREE_IDS()
2082 #undef x
2083         BTREE_ID_NR
2084 };
2085
2086 #define BTREE_MAX_DEPTH         4U
2087
2088 /* Btree nodes */
2089
2090 /*
2091  * Btree nodes
2092  *
2093  * On disk a btree node is a list/log of these; within each set the keys are
2094  * sorted
2095  */
2096 struct bset {
2097         __le64                  seq;
2098
2099         /*
2100          * Highest journal entry this bset contains keys for.
2101          * If on recovery we don't see that journal entry, this bset is ignored:
2102          * this allows us to preserve the order of all index updates after a
2103          * crash, since the journal records a total order of all index updates
2104          * and anything that didn't make it to the journal doesn't get used.
2105          */
2106         __le64                  journal_seq;
2107
2108         __le32                  flags;
2109         __le16                  version;
2110         __le16                  u64s; /* count of d[] in u64s */
2111
2112         union {
2113                 struct bkey_packed start[0];
2114                 __u64           _data[0];
2115         };
2116 } __packed __aligned(8);
2117
2118 LE32_BITMASK(BSET_CSUM_TYPE,    struct bset, flags, 0, 4);
2119
2120 LE32_BITMASK(BSET_BIG_ENDIAN,   struct bset, flags, 4, 5);
2121 LE32_BITMASK(BSET_SEPARATE_WHITEOUTS,
2122                                 struct bset, flags, 5, 6);
2123
2124 /* Sector offset within the btree node: */
2125 LE32_BITMASK(BSET_OFFSET,       struct bset, flags, 16, 32);
2126
2127 struct btree_node {
2128         struct bch_csum         csum;
2129         __le64                  magic;
2130
2131         /* this flags field is encrypted, unlike bset->flags: */
2132         __le64                  flags;
2133
2134         /* Closed interval: */
2135         struct bpos             min_key;
2136         struct bpos             max_key;
2137         struct bch_extent_ptr   _ptr; /* not used anymore */
2138         struct bkey_format      format;
2139
2140         union {
2141         struct bset             keys;
2142         struct {
2143                 __u8            pad[22];
2144                 __le16          u64s;
2145                 __u64           _data[0];
2146
2147         };
2148         };
2149 } __packed __aligned(8);
2150
2151 LE64_BITMASK(BTREE_NODE_ID,     struct btree_node, flags,  0,  4);
2152 LE64_BITMASK(BTREE_NODE_LEVEL,  struct btree_node, flags,  4,  8);
2153 LE64_BITMASK(BTREE_NODE_NEW_EXTENT_OVERWRITE,
2154                                 struct btree_node, flags,  8,  9);
2155 /* 9-32 unused */
2156 LE64_BITMASK(BTREE_NODE_SEQ,    struct btree_node, flags, 32, 64);
2157
2158 struct btree_node_entry {
2159         struct bch_csum         csum;
2160
2161         union {
2162         struct bset             keys;
2163         struct {
2164                 __u8            pad[22];
2165                 __le16          u64s;
2166                 __u64           _data[0];
2167
2168         };
2169         };
2170 } __packed __aligned(8);
2171
2172 #endif /* _BCACHEFS_FORMAT_H */