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