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                                 unwritten: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                                 unwritten: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         x(bi_nocow,                     8)
 799
 800 /* subset of BCH_INODE_FIELDS */
 801 #define BCH_INODE_OPTS()                        \
 802         x(data_checksum,                8)      \
 803         x(compression,                  8)      \
 804         x(project,                      32)     \
 805         x(background_compression,       8)      \
 806         x(data_replicas,                8)      \
 807         x(promote_target,               16)     \
 808         x(foreground_target,            16)     \
 809         x(background_target,            16)     \
 810         x(erasure_code,                 16)     \
 811         x(nocow,                        8)
 812
 813 enum inode_opt_id {
 814 #define x(name, ...)                            \
 815         Inode_opt_##name,
 816         BCH_INODE_OPTS()
 817 #undef  x
 818         Inode_opt_nr,
 819 };
 820
 821 enum {
 822         /*
 823          * User flags (get/settable with FS_IOC_*FLAGS, correspond to FS_*_FL
 824          * flags)
 825          */
 826         __BCH_INODE_SYNC                = 0,
 827         __BCH_INODE_IMMUTABLE           = 1,
 828         __BCH_INODE_APPEND              = 2,
 829         __BCH_INODE_NODUMP              = 3,
 830         __BCH_INODE_NOATIME             = 4,
 831
 832         __BCH_INODE_I_SIZE_DIRTY        = 5,
 833         __BCH_INODE_I_SECTORS_DIRTY     = 6,
 834         __BCH_INODE_UNLINKED            = 7,
 835         __BCH_INODE_BACKPTR_UNTRUSTED   = 8,
 836
 837         /* bits 20+ reserved for packed fields below: */
 838 };
 839
 840 #define BCH_INODE_SYNC          (1 << __BCH_INODE_SYNC)
 841 #define BCH_INODE_IMMUTABLE     (1 << __BCH_INODE_IMMUTABLE)
 842 #define BCH_INODE_APPEND        (1 << __BCH_INODE_APPEND)
 843 #define BCH_INODE_NODUMP        (1 << __BCH_INODE_NODUMP)
 844 #define BCH_INODE_NOATIME       (1 << __BCH_INODE_NOATIME)
 845 #define BCH_INODE_I_SIZE_DIRTY  (1 << __BCH_INODE_I_SIZE_DIRTY)
 846 #define BCH_INODE_I_SECTORS_DIRTY (1 << __BCH_INODE_I_SECTORS_DIRTY)
 847 #define BCH_INODE_UNLINKED      (1 << __BCH_INODE_UNLINKED)
 848 #define BCH_INODE_BACKPTR_UNTRUSTED (1 << __BCH_INODE_BACKPTR_UNTRUSTED)
 849
 850 LE32_BITMASK(INODE_STR_HASH,    struct bch_inode, bi_flags, 20, 24);
 851 LE32_BITMASK(INODE_NR_FIELDS,   struct bch_inode, bi_flags, 24, 31);
 852 LE32_BITMASK(INODE_NEW_VARINT,  struct bch_inode, bi_flags, 31, 32);
 853
 854 LE64_BITMASK(INODEv2_STR_HASH,  struct bch_inode_v2, bi_flags, 20, 24);
 855 LE64_BITMASK(INODEv2_NR_FIELDS, struct bch_inode_v2, bi_flags, 24, 31);
 856
 857 LE64_BITMASK(INODEv3_STR_HASH,  struct bch_inode_v3, bi_flags, 20, 24);
 858 LE64_BITMASK(INODEv3_NR_FIELDS, struct bch_inode_v3, bi_flags, 24, 31);
 859
 860 LE64_BITMASK(INODEv3_FIELDS_START,
 861                                 struct bch_inode_v3, bi_flags, 31, 36);
 862 LE64_BITMASK(INODEv3_MODE,      struct bch_inode_v3, bi_flags, 36, 52);
 863
 864 /* Dirents */
 865
 866 /*
 867  * Dirents (and xattrs) have to implement string lookups; since our b-tree
 868  * doesn't support arbitrary length strings for the key, we instead index by a
 869  * 64 bit hash (currently truncated sha1) of the string, stored in the offset
 870  * field of the key - using linear probing to resolve hash collisions. This also
 871  * provides us with the readdir cookie posix requires.
 872  *
 873  * Linear probing requires us to use whiteouts for deletions, in the event of a
 874  * collision:
 875  */
 876
 877 struct bch_dirent {
 878         struct bch_val          v;
 879
 880         /* Target inode number: */
 881         union {
 882         __le64                  d_inum;
 883         struct {                /* DT_SUBVOL */
 884         __le32                  d_child_subvol;
 885         __le32                  d_parent_subvol;
 886         };
 887         };
 888
 889         /*
 890          * Copy of mode bits 12-15 from the target inode - so userspace can get
 891          * the filetype without having to do a stat()
 892          */
 893         __u8                    d_type;
 894
 895         __u8                    d_name[];
 896 } __packed __aligned(8);
 897
 898 #define DT_SUBVOL       16
 899 #define BCH_DT_MAX      17
 900
 901 #define BCH_NAME_MAX    ((unsigned) (U8_MAX * sizeof(u64) -             \
 902                          sizeof(struct bkey) -                          \
 903                          offsetof(struct bch_dirent, d_name)))
 904
 905 /* Xattrs */
 906
 907 #define KEY_TYPE_XATTR_INDEX_USER                       0
 908 #define KEY_TYPE_XATTR_INDEX_POSIX_ACL_ACCESS   1
 909 #define KEY_TYPE_XATTR_INDEX_POSIX_ACL_DEFAULT  2
 910 #define KEY_TYPE_XATTR_INDEX_TRUSTED                    3
 911 #define KEY_TYPE_XATTR_INDEX_SECURITY           4
 912
 913 struct bch_xattr {
 914         struct bch_val          v;
 915         __u8                    x_type;
 916         __u8                    x_name_len;
 917         __le16                  x_val_len;
 918         __u8                    x_name[];
 919 } __packed __aligned(8);
 920
 921 /* Bucket/allocation information: */
 922
 923 struct bch_alloc {
 924         struct bch_val          v;
 925         __u8                    fields;
 926         __u8                    gen;
 927         __u8                    data[];
 928 } __packed __aligned(8);
 929
 930 #define BCH_ALLOC_FIELDS_V1()                   \
 931         x(read_time,            16)             \
 932         x(write_time,           16)             \
 933         x(data_type,            8)              \
 934         x(dirty_sectors,        16)             \
 935         x(cached_sectors,       16)             \
 936         x(oldest_gen,           8)              \
 937         x(stripe,               32)             \
 938         x(stripe_redundancy,    8)
 939
 940 enum {
 941 #define x(name, _bits) BCH_ALLOC_FIELD_V1_##name,
 942         BCH_ALLOC_FIELDS_V1()
 943 #undef x
 944 };
 945
 946 struct bch_alloc_v2 {
 947         struct bch_val          v;
 948         __u8                    nr_fields;
 949         __u8                    gen;
 950         __u8                    oldest_gen;
 951         __u8                    data_type;
 952         __u8                    data[];
 953 } __packed __aligned(8);
 954
 955 #define BCH_ALLOC_FIELDS_V2()                   \
 956         x(read_time,            64)             \
 957         x(write_time,           64)             \
 958         x(dirty_sectors,        32)             \
 959         x(cached_sectors,       32)             \
 960         x(stripe,               32)             \
 961         x(stripe_redundancy,    8)
 962
 963 struct bch_alloc_v3 {
 964         struct bch_val          v;
 965         __le64                  journal_seq;
 966         __le32                  flags;
 967         __u8                    nr_fields;
 968         __u8                    gen;
 969         __u8                    oldest_gen;
 970         __u8                    data_type;
 971         __u8                    data[];
 972 } __packed __aligned(8);
 973
 974 LE32_BITMASK(BCH_ALLOC_V3_NEED_DISCARD,struct bch_alloc_v3, flags,  0,  1)
 975 LE32_BITMASK(BCH_ALLOC_V3_NEED_INC_GEN,struct bch_alloc_v3, flags,  1,  2)
 976
 977 struct bch_alloc_v4 {
 978         struct bch_val          v;
 979         __u64                   journal_seq;
 980         __u32                   flags;
 981         __u8                    gen;
 982         __u8                    oldest_gen;
 983         __u8                    data_type;
 984         __u8                    stripe_redundancy;
 985         __u32                   dirty_sectors;
 986         __u32                   cached_sectors;
 987         __u64                   io_time[2];
 988         __u32                   stripe;
 989         __u32                   nr_external_backpointers;
 990 } __packed __aligned(8);
 991
 992 #define BCH_ALLOC_V4_U64s_V0    6
 993 #define BCH_ALLOC_V4_U64s       (sizeof(struct bch_alloc_v4) / sizeof(u64))
 994
 995 BITMASK(BCH_ALLOC_V4_NEED_DISCARD,      struct bch_alloc_v4, flags,  0,  1)
 996 BITMASK(BCH_ALLOC_V4_NEED_INC_GEN,      struct bch_alloc_v4, flags,  1,  2)
 997 BITMASK(BCH_ALLOC_V4_BACKPOINTERS_START,struct bch_alloc_v4, flags,  2,  8)
 998 BITMASK(BCH_ALLOC_V4_NR_BACKPOINTERS,   struct bch_alloc_v4, flags,  8,  14)
 999
1000 #define BCH_ALLOC_V4_NR_BACKPOINTERS_MAX        40
1001
1002 struct bch_backpointer {
1003         struct bch_val          v;
1004         __u8                    btree_id;
1005         __u8                    level;
1006         __u8                    data_type;
1007         __u64                   bucket_offset:40;
1008         __u32                   bucket_len;
1009         struct bpos             pos;
1010 } __packed __aligned(8);
1011
1012 /* Quotas: */
1013
1014 enum quota_types {
1015         QTYP_USR                = 0,
1016         QTYP_GRP                = 1,
1017         QTYP_PRJ                = 2,
1018         QTYP_NR                 = 3,
1019 };
1020
1021 enum quota_counters {
1022         Q_SPC                   = 0,
1023         Q_INO                   = 1,
1024         Q_COUNTERS              = 2,
1025 };
1026
1027 struct bch_quota_counter {
1028         __le64                  hardlimit;
1029         __le64                  softlimit;
1030 };
1031
1032 struct bch_quota {
1033         struct bch_val          v;
1034         struct bch_quota_counter c[Q_COUNTERS];
1035 } __packed __aligned(8);
1036
1037 /* Erasure coding */
1038
1039 struct bch_stripe {
1040         struct bch_val          v;
1041         __le16                  sectors;
1042         __u8                    algorithm;
1043         __u8                    nr_blocks;
1044         __u8                    nr_redundant;
1045
1046         __u8                    csum_granularity_bits;
1047         __u8                    csum_type;
1048         __u8                    pad;
1049
1050         struct bch_extent_ptr   ptrs[];
1051 } __packed __aligned(8);
1052
1053 /* Reflink: */
1054
1055 struct bch_reflink_p {
1056         struct bch_val          v;
1057         __le64                  idx;
1058         /*
1059          * A reflink pointer might point to an indirect extent which is then
1060          * later split (by copygc or rebalance). If we only pointed to part of
1061          * the original indirect extent, and then one of the fragments is
1062          * outside the range we point to, we'd leak a refcount: so when creating
1063          * reflink pointers, we need to store pad values to remember the full
1064          * range we were taking a reference on.
1065          */
1066         __le32                  front_pad;
1067         __le32                  back_pad;
1068 } __packed __aligned(8);
1069
1070 struct bch_reflink_v {
1071         struct bch_val          v;
1072         __le64                  refcount;
1073         union bch_extent_entry  start[0];
1074         __u64                   _data[0];
1075 } __packed __aligned(8);
1076
1077 struct bch_indirect_inline_data {
1078         struct bch_val          v;
1079         __le64                  refcount;
1080         u8                      data[0];
1081 };
1082
1083 /* Inline data */
1084
1085 struct bch_inline_data {
1086         struct bch_val          v;
1087         u8                      data[0];
1088 };
1089
1090 /* Subvolumes: */
1091
1092 #define SUBVOL_POS_MIN          POS(0, 1)
1093 #define SUBVOL_POS_MAX          POS(0, S32_MAX)
1094 #define BCACHEFS_ROOT_SUBVOL    1
1095
1096 struct bch_subvolume {
1097         struct bch_val          v;
1098         __le32                  flags;
1099         __le32                  snapshot;
1100         __le64                  inode;
1101 };
1102
1103 LE32_BITMASK(BCH_SUBVOLUME_RO,          struct bch_subvolume, flags,  0,  1)
1104 /*
1105  * We need to know whether a subvolume is a snapshot so we can know whether we
1106  * can delete it (or whether it should just be rm -rf'd)
1107  */
1108 LE32_BITMASK(BCH_SUBVOLUME_SNAP,        struct bch_subvolume, flags,  1,  2)
1109 LE32_BITMASK(BCH_SUBVOLUME_UNLINKED,    struct bch_subvolume, flags,  2,  3)
1110
1111 /* Snapshots */
1112
1113 struct bch_snapshot {
1114         struct bch_val          v;
1115         __le32                  flags;
1116         __le32                  parent;
1117         __le32                  children[2];
1118         __le32                  subvol;
1119         __le32                  pad;
1120 };
1121
1122 LE32_BITMASK(BCH_SNAPSHOT_DELETED,      struct bch_snapshot, flags,  0,  1)
1123
1124 /* True if a subvolume points to this snapshot node: */
1125 LE32_BITMASK(BCH_SNAPSHOT_SUBVOL,       struct bch_snapshot, flags,  1,  2)
1126
1127 /* LRU btree: */
1128
1129 struct bch_lru {
1130         struct bch_val          v;
1131         __le64                  idx;
1132 } __packed __aligned(8);
1133
1134 #define LRU_ID_STRIPES          (1U << 16)
1135
1136 /* Optional/variable size superblock sections: */
1137
1138 struct bch_sb_field {
1139         __u64                   _data[0];
1140         __le32                  u64s;
1141         __le32                  type;
1142 };
1143
1144 #define BCH_SB_FIELDS()                         \
1145         x(journal,      0)                      \
1146         x(members,      1)                      \
1147         x(crypt,        2)                      \
1148         x(replicas_v0,  3)                      \
1149         x(quota,        4)                      \
1150         x(disk_groups,  5)                      \
1151         x(clean,        6)                      \
1152         x(replicas,     7)                      \
1153         x(journal_seq_blacklist, 8)             \
1154         x(journal_v2,   9)                      \
1155         x(counters,     10)
1156
1157 enum bch_sb_field_type {
1158 #define x(f, nr)        BCH_SB_FIELD_##f = nr,
1159         BCH_SB_FIELDS()
1160 #undef x
1161         BCH_SB_FIELD_NR
1162 };
1163
1164 /*
1165  * Most superblock fields are replicated in all device's superblocks - a few are
1166  * not:
1167  */
1168 #define BCH_SINGLE_DEVICE_SB_FIELDS             \
1169         ((1U << BCH_SB_FIELD_journal)|          \
1170          (1U << BCH_SB_FIELD_journal_v2))
1171
1172 /* BCH_SB_FIELD_journal: */
1173
1174 struct bch_sb_field_journal {
1175         struct bch_sb_field     field;
1176         __le64                  buckets[0];
1177 };
1178
1179 struct bch_sb_field_journal_v2 {
1180         struct bch_sb_field     field;
1181
1182         struct bch_sb_field_journal_v2_entry {
1183                 __le64          start;
1184                 __le64          nr;
1185         }                       d[0];
1186 };
1187
1188 /* BCH_SB_FIELD_members: */
1189
1190 #define BCH_MIN_NR_NBUCKETS     (1 << 6)
1191
1192 struct bch_member {
1193         uuid_le                 uuid;
1194         __le64                  nbuckets;       /* device size */
1195         __le16                  first_bucket;   /* index of first bucket used */
1196         __le16                  bucket_size;    /* sectors */
1197         __le32                  pad;
1198         __le64                  last_mount;     /* time_t */
1199
1200         __le64                  flags[2];
1201 };
1202
1203 LE64_BITMASK(BCH_MEMBER_STATE,          struct bch_member, flags[0],  0,  4)
1204 /* 4-14 unused, was TIER, HAS_(META)DATA, REPLACEMENT */
1205 LE64_BITMASK(BCH_MEMBER_DISCARD,        struct bch_member, flags[0], 14, 15)
1206 LE64_BITMASK(BCH_MEMBER_DATA_ALLOWED,   struct bch_member, flags[0], 15, 20)
1207 LE64_BITMASK(BCH_MEMBER_GROUP,          struct bch_member, flags[0], 20, 28)
1208 LE64_BITMASK(BCH_MEMBER_DURABILITY,     struct bch_member, flags[0], 28, 30)
1209 LE64_BITMASK(BCH_MEMBER_FREESPACE_INITIALIZED,
1210                                         struct bch_member, flags[0], 30, 31)
1211
1212 #if 0
1213 LE64_BITMASK(BCH_MEMBER_NR_READ_ERRORS, struct bch_member, flags[1], 0,  20);
1214 LE64_BITMASK(BCH_MEMBER_NR_WRITE_ERRORS,struct bch_member, flags[1], 20, 40);
1215 #endif
1216
1217 #define BCH_MEMBER_STATES()                     \
1218         x(rw,           0)                      \
1219         x(ro,           1)                      \
1220         x(failed,       2)                      \
1221         x(spare,        3)
1222
1223 enum bch_member_state {
1224 #define x(t, n) BCH_MEMBER_STATE_##t = n,
1225         BCH_MEMBER_STATES()
1226 #undef x
1227         BCH_MEMBER_STATE_NR
1228 };
1229
1230 struct bch_sb_field_members {
1231         struct bch_sb_field     field;
1232         struct bch_member       members[0];
1233 };
1234
1235 /* BCH_SB_FIELD_crypt: */
1236
1237 struct nonce {
1238         __le32                  d[4];
1239 };
1240
1241 struct bch_key {
1242         __le64                  key[4];
1243 };
1244
1245 #define BCH_KEY_MAGIC                                   \
1246         (((u64) 'b' <<  0)|((u64) 'c' <<  8)|           \
1247          ((u64) 'h' << 16)|((u64) '*' << 24)|           \
1248          ((u64) '*' << 32)|((u64) 'k' << 40)|           \
1249          ((u64) 'e' << 48)|((u64) 'y' << 56))
1250
1251 struct bch_encrypted_key {
1252         __le64                  magic;
1253         struct bch_key          key;
1254 };
1255
1256 /*
1257  * If this field is present in the superblock, it stores an encryption key which
1258  * is used encrypt all other data/metadata. The key will normally be encrypted
1259  * with the key userspace provides, but if encryption has been turned off we'll
1260  * just store the master key unencrypted in the superblock so we can access the
1261  * previously encrypted data.
1262  */
1263 struct bch_sb_field_crypt {
1264         struct bch_sb_field     field;
1265
1266         __le64                  flags;
1267         __le64                  kdf_flags;
1268         struct bch_encrypted_key key;
1269 };
1270
1271 LE64_BITMASK(BCH_CRYPT_KDF_TYPE,        struct bch_sb_field_crypt, flags, 0, 4);
1272
1273 enum bch_kdf_types {
1274         BCH_KDF_SCRYPT          = 0,
1275         BCH_KDF_NR              = 1,
1276 };
1277
1278 /* stored as base 2 log of scrypt params: */
1279 LE64_BITMASK(BCH_KDF_SCRYPT_N,  struct bch_sb_field_crypt, kdf_flags,  0, 16);
1280 LE64_BITMASK(BCH_KDF_SCRYPT_R,  struct bch_sb_field_crypt, kdf_flags, 16, 32);
1281 LE64_BITMASK(BCH_KDF_SCRYPT_P,  struct bch_sb_field_crypt, kdf_flags, 32, 48);
1282
1283 /* BCH_SB_FIELD_replicas: */
1284
1285 #define BCH_DATA_TYPES()                \
1286         x(free,         0)              \
1287         x(sb,           1)              \
1288         x(journal,      2)              \
1289         x(btree,        3)              \
1290         x(user,         4)              \
1291         x(cached,       5)              \
1292         x(parity,       6)              \
1293         x(stripe,       7)              \
1294         x(need_gc_gens, 8)              \
1295         x(need_discard, 9)
1296
1297 enum bch_data_type {
1298 #define x(t, n) BCH_DATA_##t,
1299         BCH_DATA_TYPES()
1300 #undef x
1301         BCH_DATA_NR
1302 };
1303
1304 static inline bool data_type_is_empty(enum bch_data_type type)
1305 {
1306         switch (type) {
1307         case BCH_DATA_free:
1308         case BCH_DATA_need_gc_gens:
1309         case BCH_DATA_need_discard:
1310                 return true;
1311         default:
1312                 return false;
1313         }
1314 }
1315
1316 static inline bool data_type_is_hidden(enum bch_data_type type)
1317 {
1318         switch (type) {
1319         case BCH_DATA_sb:
1320         case BCH_DATA_journal:
1321                 return true;
1322         default:
1323                 return false;
1324         }
1325 }
1326
1327 struct bch_replicas_entry_v0 {
1328         __u8                    data_type;
1329         __u8                    nr_devs;
1330         __u8                    devs[];
1331 } __packed;
1332
1333 struct bch_sb_field_replicas_v0 {
1334         struct bch_sb_field     field;
1335         struct bch_replicas_entry_v0 entries[];
1336 } __packed __aligned(8);
1337
1338 struct bch_replicas_entry {
1339         __u8                    data_type;
1340         __u8                    nr_devs;
1341         __u8                    nr_required;
1342         __u8                    devs[];
1343 } __packed;
1344
1345 #define replicas_entry_bytes(_i)                                        \
1346         (offsetof(typeof(*(_i)), devs) + (_i)->nr_devs)
1347
1348 struct bch_sb_field_replicas {
1349         struct bch_sb_field     field;
1350         struct bch_replicas_entry entries[0];
1351 } __packed __aligned(8);
1352
1353 /* BCH_SB_FIELD_quota: */
1354
1355 struct bch_sb_quota_counter {
1356         __le32                          timelimit;
1357         __le32                          warnlimit;
1358 };
1359
1360 struct bch_sb_quota_type {
1361         __le64                          flags;
1362         struct bch_sb_quota_counter     c[Q_COUNTERS];
1363 };
1364
1365 struct bch_sb_field_quota {
1366         struct bch_sb_field             field;
1367         struct bch_sb_quota_type        q[QTYP_NR];
1368 } __packed __aligned(8);
1369
1370 /* BCH_SB_FIELD_disk_groups: */
1371
1372 #define BCH_SB_LABEL_SIZE               32
1373
1374 struct bch_disk_group {
1375         __u8                    label[BCH_SB_LABEL_SIZE];
1376         __le64                  flags[2];
1377 } __packed __aligned(8);
1378
1379 LE64_BITMASK(BCH_GROUP_DELETED,         struct bch_disk_group, flags[0], 0,  1)
1380 LE64_BITMASK(BCH_GROUP_DATA_ALLOWED,    struct bch_disk_group, flags[0], 1,  6)
1381 LE64_BITMASK(BCH_GROUP_PARENT,          struct bch_disk_group, flags[0], 6, 24)
1382
1383 struct bch_sb_field_disk_groups {
1384         struct bch_sb_field     field;
1385         struct bch_disk_group   entries[0];
1386 } __packed __aligned(8);
1387
1388 /* BCH_SB_FIELD_counters */
1389
1390 #define BCH_PERSISTENT_COUNTERS()                               \
1391         x(io_read,                                      0)      \
1392         x(io_write,                                     1)      \
1393         x(io_move,                                      2)      \
1394         x(bucket_invalidate,                            3)      \
1395         x(bucket_discard,                               4)      \
1396         x(bucket_alloc,                                 5)      \
1397         x(bucket_alloc_fail,                            6)      \
1398         x(btree_cache_scan,                             7)      \
1399         x(btree_cache_reap,                             8)      \
1400         x(btree_cache_cannibalize,                      9)      \
1401         x(btree_cache_cannibalize_lock,                 10)     \
1402         x(btree_cache_cannibalize_lock_fail,            11)     \
1403         x(btree_cache_cannibalize_unlock,               12)     \
1404         x(btree_node_write,                             13)     \
1405         x(btree_node_read,                              14)     \
1406         x(btree_node_compact,                           15)     \
1407         x(btree_node_merge,                             16)     \
1408         x(btree_node_split,                             17)     \
1409         x(btree_node_rewrite,                           18)     \
1410         x(btree_node_alloc,                             19)     \
1411         x(btree_node_free,                              20)     \
1412         x(btree_node_set_root,                          21)     \
1413         x(btree_path_relock_fail,                       22)     \
1414         x(btree_path_upgrade_fail,                      23)     \
1415         x(btree_reserve_get_fail,                       24)     \
1416         x(journal_entry_full,                           25)     \
1417         x(journal_full,                                 26)     \
1418         x(journal_reclaim_finish,                       27)     \
1419         x(journal_reclaim_start,                        28)     \
1420         x(journal_write,                                29)     \
1421         x(read_promote,                                 30)     \
1422         x(read_bounce,                                  31)     \
1423         x(read_split,                                   33)     \
1424         x(read_retry,                                   32)     \
1425         x(read_reuse_race,                              34)     \
1426         x(move_extent_read,                             35)     \
1427         x(move_extent_write,                            36)     \
1428         x(move_extent_finish,                           37)     \
1429         x(move_extent_race,                             38)     \
1430         x(move_extent_alloc_mem_fail,                   39)     \
1431         x(copygc,                                       40)     \
1432         x(copygc_wait,                                  41)     \
1433         x(gc_gens_end,                                  42)     \
1434         x(gc_gens_start,                                43)     \
1435         x(trans_blocked_journal_reclaim,                44)     \
1436         x(trans_restart_btree_node_reused,              45)     \
1437         x(trans_restart_btree_node_split,               46)     \
1438         x(trans_restart_fault_inject,                   47)     \
1439         x(trans_restart_iter_upgrade,                   48)     \
1440         x(trans_restart_journal_preres_get,             49)     \
1441         x(trans_restart_journal_reclaim,                50)     \
1442         x(trans_restart_journal_res_get,                51)     \
1443         x(trans_restart_key_cache_key_realloced,        52)     \
1444         x(trans_restart_key_cache_raced,                53)     \
1445         x(trans_restart_mark_replicas,                  54)     \
1446         x(trans_restart_mem_realloced,                  55)     \
1447         x(trans_restart_memory_allocation_failure,      56)     \
1448         x(trans_restart_relock,                         57)     \
1449         x(trans_restart_relock_after_fill,              58)     \
1450         x(trans_restart_relock_key_cache_fill,          59)     \
1451         x(trans_restart_relock_next_node,               60)     \
1452         x(trans_restart_relock_parent_for_fill,         61)     \
1453         x(trans_restart_relock_path,                    62)     \
1454         x(trans_restart_relock_path_intent,             63)     \
1455         x(trans_restart_too_many_iters,                 64)     \
1456         x(trans_restart_traverse,                       65)     \
1457         x(trans_restart_upgrade,                        66)     \
1458         x(trans_restart_would_deadlock,                 67)     \
1459         x(trans_restart_would_deadlock_write,           68)     \
1460         x(trans_restart_injected,                       69)     \
1461         x(trans_restart_key_cache_upgrade,              70)     \
1462         x(trans_traverse_all,                           71)     \
1463         x(transaction_commit,                           72)     \
1464         x(write_super,                                  73)     \
1465         x(trans_restart_would_deadlock_recursion_limit, 74)
1466
1467 enum bch_persistent_counters {
1468 #define x(t, n, ...) BCH_COUNTER_##t,
1469         BCH_PERSISTENT_COUNTERS()
1470 #undef x
1471         BCH_COUNTER_NR
1472 };
1473
1474 struct bch_sb_field_counters {
1475         struct bch_sb_field     field;
1476         __le64                  d[0];
1477 };
1478
1479 /*
1480  * On clean shutdown, store btree roots and current journal sequence number in
1481  * the superblock:
1482  */
1483 struct jset_entry {
1484         __le16                  u64s;
1485         __u8                    btree_id;
1486         __u8                    level;
1487         __u8                    type; /* designates what this jset holds */
1488         __u8                    pad[3];
1489
1490         union {
1491                 struct bkey_i   start[0];
1492                 __u64           _data[0];
1493         };
1494 };
1495
1496 struct bch_sb_field_clean {
1497         struct bch_sb_field     field;
1498
1499         __le32                  flags;
1500         __le16                  _read_clock; /* no longer used */
1501         __le16                  _write_clock;
1502         __le64                  journal_seq;
1503
1504         union {
1505                 struct jset_entry start[0];
1506                 __u64           _data[0];
1507         };
1508 };
1509
1510 struct journal_seq_blacklist_entry {
1511         __le64                  start;
1512         __le64                  end;
1513 };
1514
1515 struct bch_sb_field_journal_seq_blacklist {
1516         struct bch_sb_field     field;
1517
1518         union {
1519                 struct journal_seq_blacklist_entry start[0];
1520                 __u64           _data[0];
1521         };
1522 };
1523
1524 /* Superblock: */
1525
1526 /*
1527  * New versioning scheme:
1528  * One common version number for all on disk data structures - superblock, btree
1529  * nodes, journal entries
1530  */
1531 #define BCH_JSET_VERSION_OLD                    2
1532 #define BCH_BSET_VERSION_OLD                    3
1533
1534 #define BCH_METADATA_VERSIONS()                         \
1535         x(bkey_renumber,                10)             \
1536         x(inode_btree_change,           11)             \
1537         x(snapshot,                     12)             \
1538         x(inode_backpointers,           13)             \
1539         x(btree_ptr_sectors_written,    14)             \
1540         x(snapshot_2,                   15)             \
1541         x(reflink_p_fix,                16)             \
1542         x(subvol_dirent,                17)             \
1543         x(inode_v2,                     18)             \
1544         x(freespace,                    19)             \
1545         x(alloc_v4,                     20)             \
1546         x(new_data_types,               21)             \
1547         x(backpointers,                 22)             \
1548         x(inode_v3,                     23)             \
1549         x(unwritten_extents,            24)
1550
1551 enum bcachefs_metadata_version {
1552         bcachefs_metadata_version_min = 9,
1553 #define x(t, n) bcachefs_metadata_version_##t = n,
1554         BCH_METADATA_VERSIONS()
1555 #undef x
1556         bcachefs_metadata_version_max
1557 };
1558
1559 #define bcachefs_metadata_version_current       (bcachefs_metadata_version_max - 1)
1560
1561 #define BCH_SB_SECTOR                   8
1562 #define BCH_SB_MEMBERS_MAX              64 /* XXX kill */
1563
1564 struct bch_sb_layout {
1565         uuid_le                 magic;  /* bcachefs superblock UUID */
1566         __u8                    layout_type;
1567         __u8                    sb_max_size_bits; /* base 2 of 512 byte sectors */
1568         __u8                    nr_superblocks;
1569         __u8                    pad[5];
1570         __le64                  sb_offset[61];
1571 } __packed __aligned(8);
1572
1573 #define BCH_SB_LAYOUT_SECTOR    7
1574
1575 /*
1576  * @offset      - sector where this sb was written
1577  * @version     - on disk format version
1578  * @version_min - Oldest metadata version this filesystem contains; so we can
1579  *                safely drop compatibility code and refuse to mount filesystems
1580  *                we'd need it for
1581  * @magic       - identifies as a bcachefs superblock (BCACHE_MAGIC)
1582  * @seq         - incremented each time superblock is written
1583  * @uuid        - used for generating various magic numbers and identifying
1584  *                member devices, never changes
1585  * @user_uuid   - user visible UUID, may be changed
1586  * @label       - filesystem label
1587  * @seq         - identifies most recent superblock, incremented each time
1588  *                superblock is written
1589  * @features    - enabled incompatible features
1590  */
1591 struct bch_sb {
1592         struct bch_csum         csum;
1593         __le16                  version;
1594         __le16                  version_min;
1595         __le16                  pad[2];
1596         uuid_le                 magic;
1597         uuid_le                 uuid;
1598         uuid_le                 user_uuid;
1599         __u8                    label[BCH_SB_LABEL_SIZE];
1600         __le64                  offset;
1601         __le64                  seq;
1602
1603         __le16                  block_size;
1604         __u8                    dev_idx;
1605         __u8                    nr_devices;
1606         __le32                  u64s;
1607
1608         __le64                  time_base_lo;
1609         __le32                  time_base_hi;
1610         __le32                  time_precision;
1611
1612         __le64                  flags[8];
1613         __le64                  features[2];
1614         __le64                  compat[2];
1615
1616         struct bch_sb_layout    layout;
1617
1618         union {
1619                 struct bch_sb_field start[0];
1620                 __le64          _data[0];
1621         };
1622 } __packed __aligned(8);
1623
1624 /*
1625  * Flags:
1626  * BCH_SB_INITALIZED    - set on first mount
1627  * BCH_SB_CLEAN         - did we shut down cleanly? Just a hint, doesn't affect
1628  *                        behaviour of mount/recovery path:
1629  * BCH_SB_INODE_32BIT   - limit inode numbers to 32 bits
1630  * BCH_SB_128_BIT_MACS  - 128 bit macs instead of 80
1631  * BCH_SB_ENCRYPTION_TYPE - if nonzero encryption is enabled; overrides
1632  *                         DATA/META_CSUM_TYPE. Also indicates encryption
1633  *                         algorithm in use, if/when we get more than one
1634  */
1635
1636 LE16_BITMASK(BCH_SB_BLOCK_SIZE,         struct bch_sb, block_size, 0, 16);
1637
1638 LE64_BITMASK(BCH_SB_INITIALIZED,        struct bch_sb, flags[0],  0,  1);
1639 LE64_BITMASK(BCH_SB_CLEAN,              struct bch_sb, flags[0],  1,  2);
1640 LE64_BITMASK(BCH_SB_CSUM_TYPE,          struct bch_sb, flags[0],  2,  8);
1641 LE64_BITMASK(BCH_SB_ERROR_ACTION,       struct bch_sb, flags[0],  8, 12);
1642
1643 LE64_BITMASK(BCH_SB_BTREE_NODE_SIZE,    struct bch_sb, flags[0], 12, 28);
1644
1645 LE64_BITMASK(BCH_SB_GC_RESERVE,         struct bch_sb, flags[0], 28, 33);
1646 LE64_BITMASK(BCH_SB_ROOT_RESERVE,       struct bch_sb, flags[0], 33, 40);
1647
1648 LE64_BITMASK(BCH_SB_META_CSUM_TYPE,     struct bch_sb, flags[0], 40, 44);
1649 LE64_BITMASK(BCH_SB_DATA_CSUM_TYPE,     struct bch_sb, flags[0], 44, 48);
1650
1651 LE64_BITMASK(BCH_SB_META_REPLICAS_WANT, struct bch_sb, flags[0], 48, 52);
1652 LE64_BITMASK(BCH_SB_DATA_REPLICAS_WANT, struct bch_sb, flags[0], 52, 56);
1653
1654 LE64_BITMASK(BCH_SB_POSIX_ACL,          struct bch_sb, flags[0], 56, 57);
1655 LE64_BITMASK(BCH_SB_USRQUOTA,           struct bch_sb, flags[0], 57, 58);
1656 LE64_BITMASK(BCH_SB_GRPQUOTA,           struct bch_sb, flags[0], 58, 59);
1657 LE64_BITMASK(BCH_SB_PRJQUOTA,           struct bch_sb, flags[0], 59, 60);
1658
1659 LE64_BITMASK(BCH_SB_HAS_ERRORS,         struct bch_sb, flags[0], 60, 61);
1660 LE64_BITMASK(BCH_SB_HAS_TOPOLOGY_ERRORS,struct bch_sb, flags[0], 61, 62);
1661
1662 LE64_BITMASK(BCH_SB_BIG_ENDIAN,         struct bch_sb, flags[0], 62, 63);
1663
1664 LE64_BITMASK(BCH_SB_STR_HASH_TYPE,      struct bch_sb, flags[1],  0,  4);
1665 LE64_BITMASK(BCH_SB_COMPRESSION_TYPE,   struct bch_sb, flags[1],  4,  8);
1666 LE64_BITMASK(BCH_SB_INODE_32BIT,        struct bch_sb, flags[1],  8,  9);
1667
1668 LE64_BITMASK(BCH_SB_128_BIT_MACS,       struct bch_sb, flags[1],  9, 10);
1669 LE64_BITMASK(BCH_SB_ENCRYPTION_TYPE,    struct bch_sb, flags[1], 10, 14);
1670
1671 /*
1672  * Max size of an extent that may require bouncing to read or write
1673  * (checksummed, compressed): 64k
1674  */
1675 LE64_BITMASK(BCH_SB_ENCODED_EXTENT_MAX_BITS,
1676                                         struct bch_sb, flags[1], 14, 20);
1677
1678 LE64_BITMASK(BCH_SB_META_REPLICAS_REQ,  struct bch_sb, flags[1], 20, 24);
1679 LE64_BITMASK(BCH_SB_DATA_REPLICAS_REQ,  struct bch_sb, flags[1], 24, 28);
1680
1681 LE64_BITMASK(BCH_SB_PROMOTE_TARGET,     struct bch_sb, flags[1], 28, 40);
1682 LE64_BITMASK(BCH_SB_FOREGROUND_TARGET,  struct bch_sb, flags[1], 40, 52);
1683 LE64_BITMASK(BCH_SB_BACKGROUND_TARGET,  struct bch_sb, flags[1], 52, 64);
1684
1685 LE64_BITMASK(BCH_SB_BACKGROUND_COMPRESSION_TYPE,
1686                                         struct bch_sb, flags[2],  0,  4);
1687 LE64_BITMASK(BCH_SB_GC_RESERVE_BYTES,   struct bch_sb, flags[2],  4, 64);
1688
1689 LE64_BITMASK(BCH_SB_ERASURE_CODE,       struct bch_sb, flags[3],  0, 16);
1690 LE64_BITMASK(BCH_SB_METADATA_TARGET,    struct bch_sb, flags[3], 16, 28);
1691 LE64_BITMASK(BCH_SB_SHARD_INUMS,        struct bch_sb, flags[3], 28, 29);
1692 LE64_BITMASK(BCH_SB_INODES_USE_KEY_CACHE,struct bch_sb, flags[3], 29, 30);
1693 LE64_BITMASK(BCH_SB_JOURNAL_FLUSH_DELAY,struct bch_sb, flags[3], 30, 62);
1694 LE64_BITMASK(BCH_SB_JOURNAL_FLUSH_DISABLED,struct bch_sb, flags[3], 62, 63);
1695 LE64_BITMASK(BCH_SB_JOURNAL_RECLAIM_DELAY,struct bch_sb, flags[4], 0, 32);
1696 /* Obsolete, always enabled: */
1697 LE64_BITMASK(BCH_SB_JOURNAL_TRANSACTION_NAMES,struct bch_sb, flags[4], 32, 33);
1698 LE64_BITMASK(BCH_SB_NOCOW,              struct bch_sb, flags[4], 33, 34);
1699
1700 /*
1701  * Features:
1702  *
1703  * journal_seq_blacklist_v3:    gates BCH_SB_FIELD_journal_seq_blacklist
1704  * reflink:                     gates KEY_TYPE_reflink
1705  * inline_data:                 gates KEY_TYPE_inline_data
1706  * new_siphash:                 gates BCH_STR_HASH_siphash
1707  * new_extent_overwrite:        gates BTREE_NODE_NEW_EXTENT_OVERWRITE
1708  */
1709 #define BCH_SB_FEATURES()                       \
1710         x(lz4,                          0)      \
1711         x(gzip,                         1)      \
1712         x(zstd,                         2)      \
1713         x(atomic_nlink,                 3)      \
1714         x(ec,                           4)      \
1715         x(journal_seq_blacklist_v3,     5)      \
1716         x(reflink,                      6)      \
1717         x(new_siphash,                  7)      \
1718         x(inline_data,                  8)      \
1719         x(new_extent_overwrite,         9)      \
1720         x(incompressible,               10)     \
1721         x(btree_ptr_v2,                 11)     \
1722         x(extents_above_btree_updates,  12)     \
1723         x(btree_updates_journalled,     13)     \
1724         x(reflink_inline_data,          14)     \
1725         x(new_varint,                   15)     \
1726         x(journal_no_flush,             16)     \
1727         x(alloc_v2,                     17)     \
1728         x(extents_across_btree_nodes,   18)
1729
1730 #define BCH_SB_FEATURES_ALWAYS                          \
1731         ((1ULL << BCH_FEATURE_new_extent_overwrite)|    \
1732          (1ULL << BCH_FEATURE_extents_above_btree_updates)|\
1733          (1ULL << BCH_FEATURE_btree_updates_journalled)|\
1734          (1ULL << BCH_FEATURE_alloc_v2)|\
1735          (1ULL << BCH_FEATURE_extents_across_btree_nodes))
1736
1737 #define BCH_SB_FEATURES_ALL                             \
1738         (BCH_SB_FEATURES_ALWAYS|                        \
1739          (1ULL << BCH_FEATURE_new_siphash)|             \
1740          (1ULL << BCH_FEATURE_btree_ptr_v2)|            \
1741          (1ULL << BCH_FEATURE_new_varint)|              \
1742          (1ULL << BCH_FEATURE_journal_no_flush))
1743
1744 enum bch_sb_feature {
1745 #define x(f, n) BCH_FEATURE_##f,
1746         BCH_SB_FEATURES()
1747 #undef x
1748         BCH_FEATURE_NR,
1749 };
1750
1751 #define BCH_SB_COMPAT()                                 \
1752         x(alloc_info,                           0)      \
1753         x(alloc_metadata,                       1)      \
1754         x(extents_above_btree_updates_done,     2)      \
1755         x(bformat_overflow_done,                3)
1756
1757 enum bch_sb_compat {
1758 #define x(f, n) BCH_COMPAT_##f,
1759         BCH_SB_COMPAT()
1760 #undef x
1761         BCH_COMPAT_NR,
1762 };
1763
1764 /* options: */
1765
1766 #define BCH_REPLICAS_MAX                4U
1767
1768 #define BCH_BKEY_PTRS_MAX               16U
1769
1770 #define BCH_ERROR_ACTIONS()             \
1771         x(continue,             0)      \
1772         x(ro,                   1)      \
1773         x(panic,                2)
1774
1775 enum bch_error_actions {
1776 #define x(t, n) BCH_ON_ERROR_##t = n,
1777         BCH_ERROR_ACTIONS()
1778 #undef x
1779         BCH_ON_ERROR_NR
1780 };
1781
1782 #define BCH_STR_HASH_TYPES()            \
1783         x(crc32c,               0)      \
1784         x(crc64,                1)      \
1785         x(siphash_old,          2)      \
1786         x(siphash,              3)
1787
1788 enum bch_str_hash_type {
1789 #define x(t, n) BCH_STR_HASH_##t = n,
1790         BCH_STR_HASH_TYPES()
1791 #undef x
1792         BCH_STR_HASH_NR
1793 };
1794
1795 #define BCH_STR_HASH_OPTS()             \
1796         x(crc32c,               0)      \
1797         x(crc64,                1)      \
1798         x(siphash,              2)
1799
1800 enum bch_str_hash_opts {
1801 #define x(t, n) BCH_STR_HASH_OPT_##t = n,
1802         BCH_STR_HASH_OPTS()
1803 #undef x
1804         BCH_STR_HASH_OPT_NR
1805 };
1806
1807 #define BCH_CSUM_TYPES()                        \
1808         x(none,                         0)      \
1809         x(crc32c_nonzero,               1)      \
1810         x(crc64_nonzero,                2)      \
1811         x(chacha20_poly1305_80,         3)      \
1812         x(chacha20_poly1305_128,        4)      \
1813         x(crc32c,                       5)      \
1814         x(crc64,                        6)      \
1815         x(xxhash,                       7)
1816
1817 enum bch_csum_type {
1818 #define x(t, n) BCH_CSUM_##t = n,
1819         BCH_CSUM_TYPES()
1820 #undef x
1821         BCH_CSUM_NR
1822 };
1823
1824 static const unsigned bch_crc_bytes[] = {
1825         [BCH_CSUM_none]                         = 0,
1826         [BCH_CSUM_crc32c_nonzero]               = 4,
1827         [BCH_CSUM_crc32c]                       = 4,
1828         [BCH_CSUM_crc64_nonzero]                = 8,
1829         [BCH_CSUM_crc64]                        = 8,
1830         [BCH_CSUM_xxhash]                       = 8,
1831         [BCH_CSUM_chacha20_poly1305_80]         = 10,
1832         [BCH_CSUM_chacha20_poly1305_128]        = 16,
1833 };
1834
1835 static inline _Bool bch2_csum_type_is_encryption(enum bch_csum_type type)
1836 {
1837         switch (type) {
1838         case BCH_CSUM_chacha20_poly1305_80:
1839         case BCH_CSUM_chacha20_poly1305_128:
1840                 return true;
1841         default:
1842                 return false;
1843         }
1844 }
1845
1846 #define BCH_CSUM_OPTS()                 \
1847         x(none,                 0)      \
1848         x(crc32c,               1)      \
1849         x(crc64,                2)      \
1850         x(xxhash,               3)
1851
1852 enum bch_csum_opts {
1853 #define x(t, n) BCH_CSUM_OPT_##t = n,
1854         BCH_CSUM_OPTS()
1855 #undef x
1856         BCH_CSUM_OPT_NR
1857 };
1858
1859 #define BCH_COMPRESSION_TYPES()         \
1860         x(none,                 0)      \
1861         x(lz4_old,              1)      \
1862         x(gzip,                 2)      \
1863         x(lz4,                  3)      \
1864         x(zstd,                 4)      \
1865         x(incompressible,       5)
1866
1867 enum bch_compression_type {
1868 #define x(t, n) BCH_COMPRESSION_TYPE_##t = n,
1869         BCH_COMPRESSION_TYPES()
1870 #undef x
1871         BCH_COMPRESSION_TYPE_NR
1872 };
1873
1874 #define BCH_COMPRESSION_OPTS()          \
1875         x(none,         0)              \
1876         x(lz4,          1)              \
1877         x(gzip,         2)              \
1878         x(zstd,         3)
1879
1880 enum bch_compression_opts {
1881 #define x(t, n) BCH_COMPRESSION_OPT_##t = n,
1882         BCH_COMPRESSION_OPTS()
1883 #undef x
1884         BCH_COMPRESSION_OPT_NR
1885 };
1886
1887 /*
1888  * Magic numbers
1889  *
1890  * The various other data structures have their own magic numbers, which are
1891  * xored with the first part of the cache set's UUID
1892  */
1893
1894 #define BCACHE_MAGIC                                                    \
1895         UUID_LE(0xf67385c6, 0x1a4e, 0xca45,                             \
1896                 0x82, 0x65, 0xf5, 0x7f, 0x48, 0xba, 0x6d, 0x81)
1897
1898 #define BCACHEFS_STATFS_MAGIC           0xca451a4e
1899
1900 #define JSET_MAGIC              __cpu_to_le64(0x245235c1a3625032ULL)
1901 #define BSET_MAGIC              __cpu_to_le64(0x90135c78b99e07f5ULL)
1902
1903 static inline __le64 __bch2_sb_magic(struct bch_sb *sb)
1904 {
1905         __le64 ret;
1906
1907         memcpy(&ret, &sb->uuid, sizeof(ret));
1908         return ret;
1909 }
1910
1911 static inline __u64 __jset_magic(struct bch_sb *sb)
1912 {
1913         return __le64_to_cpu(__bch2_sb_magic(sb) ^ JSET_MAGIC);
1914 }
1915
1916 static inline __u64 __bset_magic(struct bch_sb *sb)
1917 {
1918         return __le64_to_cpu(__bch2_sb_magic(sb) ^ BSET_MAGIC);
1919 }
1920
1921 /* Journal */
1922
1923 #define JSET_KEYS_U64s  (sizeof(struct jset_entry) / sizeof(__u64))
1924
1925 #define BCH_JSET_ENTRY_TYPES()                  \
1926         x(btree_keys,           0)              \
1927         x(btree_root,           1)              \
1928         x(prio_ptrs,            2)              \
1929         x(blacklist,            3)              \
1930         x(blacklist_v2,         4)              \
1931         x(usage,                5)              \
1932         x(data_usage,           6)              \
1933         x(clock,                7)              \
1934         x(dev_usage,            8)              \
1935         x(log,                  9)              \
1936         x(overwrite,            10)
1937
1938 enum {
1939 #define x(f, nr)        BCH_JSET_ENTRY_##f      = nr,
1940         BCH_JSET_ENTRY_TYPES()
1941 #undef x
1942         BCH_JSET_ENTRY_NR
1943 };
1944
1945 /*
1946  * Journal sequence numbers can be blacklisted: bsets record the max sequence
1947  * number of all the journal entries they contain updates for, so that on
1948  * recovery we can ignore those bsets that contain index updates newer that what
1949  * made it into the journal.
1950  *
1951  * This means that we can't reuse that journal_seq - we have to skip it, and
1952  * then record that we skipped it so that the next time we crash and recover we
1953  * don't think there was a missing journal entry.
1954  */
1955 struct jset_entry_blacklist {
1956         struct jset_entry       entry;
1957         __le64                  seq;
1958 };
1959
1960 struct jset_entry_blacklist_v2 {
1961         struct jset_entry       entry;
1962         __le64                  start;
1963         __le64                  end;
1964 };
1965
1966 #define BCH_FS_USAGE_TYPES()                    \
1967         x(reserved,             0)              \
1968         x(inodes,               1)              \
1969         x(key_version,          2)
1970
1971 enum {
1972 #define x(f, nr)        BCH_FS_USAGE_##f        = nr,
1973         BCH_FS_USAGE_TYPES()
1974 #undef x
1975         BCH_FS_USAGE_NR
1976 };
1977
1978 struct jset_entry_usage {
1979         struct jset_entry       entry;
1980         __le64                  v;
1981 } __packed;
1982
1983 struct jset_entry_data_usage {
1984         struct jset_entry       entry;
1985         __le64                  v;
1986         struct bch_replicas_entry r;
1987 } __packed;
1988
1989 struct jset_entry_clock {
1990         struct jset_entry       entry;
1991         __u8                    rw;
1992         __u8                    pad[7];
1993         __le64                  time;
1994 } __packed;
1995
1996 struct jset_entry_dev_usage_type {
1997         __le64                  buckets;
1998         __le64                  sectors;
1999         __le64                  fragmented;
2000 } __packed;
2001
2002 struct jset_entry_dev_usage {
2003         struct jset_entry       entry;
2004         __le32                  dev;
2005         __u32                   pad;
2006
2007         __le64                  buckets_ec;
2008         __le64                  _buckets_unavailable; /* No longer used */
2009
2010         struct jset_entry_dev_usage_type d[];
2011 } __packed;
2012
2013 static inline unsigned jset_entry_dev_usage_nr_types(struct jset_entry_dev_usage *u)
2014 {
2015         return (vstruct_bytes(&u->entry) - sizeof(struct jset_entry_dev_usage)) /
2016                 sizeof(struct jset_entry_dev_usage_type);
2017 }
2018
2019 struct jset_entry_log {
2020         struct jset_entry       entry;
2021         u8                      d[];
2022 } __packed;
2023
2024 /*
2025  * On disk format for a journal entry:
2026  * seq is monotonically increasing; every journal entry has its own unique
2027  * sequence number.
2028  *
2029  * last_seq is the oldest journal entry that still has keys the btree hasn't
2030  * flushed to disk yet.
2031  *
2032  * version is for on disk format changes.
2033  */
2034 struct jset {
2035         struct bch_csum         csum;
2036
2037         __le64                  magic;
2038         __le64                  seq;
2039         __le32                  version;
2040         __le32                  flags;
2041
2042         __le32                  u64s; /* size of d[] in u64s */
2043
2044         __u8                    encrypted_start[0];
2045
2046         __le16                  _read_clock; /* no longer used */
2047         __le16                  _write_clock;
2048
2049         /* Sequence number of oldest dirty journal entry */
2050         __le64                  last_seq;
2051
2052
2053         union {
2054                 struct jset_entry start[0];
2055                 __u64           _data[0];
2056         };
2057 } __packed __aligned(8);
2058
2059 LE32_BITMASK(JSET_CSUM_TYPE,    struct jset, flags, 0, 4);
2060 LE32_BITMASK(JSET_BIG_ENDIAN,   struct jset, flags, 4, 5);
2061 LE32_BITMASK(JSET_NO_FLUSH,     struct jset, flags, 5, 6);
2062
2063 #define BCH_JOURNAL_BUCKETS_MIN         8
2064
2065 /* Btree: */
2066
2067 #define BCH_BTREE_IDS()                         \
2068         x(extents,      0)                      \
2069         x(inodes,       1)                      \
2070         x(dirents,      2)                      \
2071         x(xattrs,       3)                      \
2072         x(alloc,        4)                      \
2073         x(quotas,       5)                      \
2074         x(stripes,      6)                      \
2075         x(reflink,      7)                      \
2076         x(subvolumes,   8)                      \
2077         x(snapshots,    9)                      \
2078         x(lru,          10)                     \
2079         x(freespace,    11)                     \
2080         x(need_discard, 12)                     \
2081         x(backpointers, 13)
2082
2083 enum btree_id {
2084 #define x(kwd, val) BTREE_ID_##kwd = val,
2085         BCH_BTREE_IDS()
2086 #undef x
2087         BTREE_ID_NR
2088 };
2089
2090 #define BTREE_MAX_DEPTH         4U
2091
2092 /* Btree nodes */
2093
2094 /*
2095  * Btree nodes
2096  *
2097  * On disk a btree node is a list/log of these; within each set the keys are
2098  * sorted
2099  */
2100 struct bset {
2101         __le64                  seq;
2102
2103         /*
2104          * Highest journal entry this bset contains keys for.
2105          * If on recovery we don't see that journal entry, this bset is ignored:
2106          * this allows us to preserve the order of all index updates after a
2107          * crash, since the journal records a total order of all index updates
2108          * and anything that didn't make it to the journal doesn't get used.
2109          */
2110         __le64                  journal_seq;
2111
2112         __le32                  flags;
2113         __le16                  version;
2114         __le16                  u64s; /* count of d[] in u64s */
2115
2116         union {
2117                 struct bkey_packed start[0];
2118                 __u64           _data[0];
2119         };
2120 } __packed __aligned(8);
2121
2122 LE32_BITMASK(BSET_CSUM_TYPE,    struct bset, flags, 0, 4);
2123
2124 LE32_BITMASK(BSET_BIG_ENDIAN,   struct bset, flags, 4, 5);
2125 LE32_BITMASK(BSET_SEPARATE_WHITEOUTS,
2126                                 struct bset, flags, 5, 6);
2127
2128 /* Sector offset within the btree node: */
2129 LE32_BITMASK(BSET_OFFSET,       struct bset, flags, 16, 32);
2130
2131 struct btree_node {
2132         struct bch_csum         csum;
2133         __le64                  magic;
2134
2135         /* this flags field is encrypted, unlike bset->flags: */
2136         __le64                  flags;
2137
2138         /* Closed interval: */
2139         struct bpos             min_key;
2140         struct bpos             max_key;
2141         struct bch_extent_ptr   _ptr; /* not used anymore */
2142         struct bkey_format      format;
2143
2144         union {
2145         struct bset             keys;
2146         struct {
2147                 __u8            pad[22];
2148                 __le16          u64s;
2149                 __u64           _data[0];
2150
2151         };
2152         };
2153 } __packed __aligned(8);
2154
2155 LE64_BITMASK(BTREE_NODE_ID,     struct btree_node, flags,  0,  4);
2156 LE64_BITMASK(BTREE_NODE_LEVEL,  struct btree_node, flags,  4,  8);
2157 LE64_BITMASK(BTREE_NODE_NEW_EXTENT_OVERWRITE,
2158                                 struct btree_node, flags,  8,  9);
2159 /* 9-32 unused */
2160 LE64_BITMASK(BTREE_NODE_SEQ,    struct btree_node, flags, 32, 64);
2161
2162 struct btree_node_entry {
2163         struct bch_csum         csum;
2164
2165         union {
2166         struct bset             keys;
2167         struct {
2168                 __u8            pad[22];
2169                 __le16          u64s;
2170                 __u64           _data[0];
2171
2172         };
2173         };
2174 } __packed __aligned(8);
2175
2176 #endif /* _BCACHEFS_FORMAT_H */