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

   1 /* SPDX-License-Identifier: GPL-2.0 */
   2 #ifndef _BCACHEFS_FORMAT_H
   3 #define _BCACHEFS_FORMAT_H
   4
   5 /*
   6  * bcachefs on disk data structures
   7  *
   8  * OVERVIEW:
   9  *
  10  * There are three main types of on disk data structures in bcachefs (this is
  11  * reduced from 5 in bcache)
  12  *
  13  *  - superblock
  14  *  - journal
  15  *  - btree
  16  *
  17  * The btree is the primary structure; most metadata exists as keys in the
  18  * various btrees. There are only a small number of btrees, they're not
  19  * sharded - we have one btree for extents, another for inodes, et cetera.
  20  *
  21  * SUPERBLOCK:
  22  *
  23  * The superblock contains the location of the journal, the list of devices in
  24  * the filesystem, and in general any metadata we need in order to decide
  25  * whether we can start a filesystem or prior to reading the journal/btree
  26  * roots.
  27  *
  28  * The superblock is extensible, and most of the contents of the superblock are
  29  * in variable length, type tagged fields; see struct bch_sb_field.
  30  *
  31  * Backup superblocks do not reside in a fixed location; also, superblocks do
  32  * not have a fixed size. To locate backup superblocks we have struct
  33  * bch_sb_layout; we store a copy of this inside every superblock, and also
  34  * before the first superblock.
  35  *
  36  * JOURNAL:
  37  *
  38  * The journal primarily records btree updates in the order they occurred;
  39  * journal replay consists of just iterating over all the keys in the open
  40  * journal entries and re-inserting them into the btrees.
  41  *
  42  * The journal also contains entry types for the btree roots, and blacklisted
  43  * journal sequence numbers (see journal_seq_blacklist.c).
  44  *
  45  * BTREE:
  46  *
  47  * bcachefs btrees are copy on write b+ trees, where nodes are big (typically
  48  * 128k-256k) and log structured. We use struct btree_node for writing the first
  49  * entry in a given node (offset 0), and struct btree_node_entry for all
  50  * subsequent writes.
  51  *
  52  * After the header, btree node entries contain a list of keys in sorted order.
  53  * Values are stored inline with the keys; since values are variable length (and
  54  * keys effectively are variable length too, due to packing) we can't do random
  55  * access without building up additional in memory tables in the btree node read
  56  * path.
  57  *
  58  * BTREE KEYS (struct bkey):
  59  *
  60  * The various btrees share a common format for the key - so as to avoid
  61  * switching in fastpath lookup/comparison code - but define their own
  62  * structures for the key values.
  63  *
  64  * The size of a key/value pair is stored as a u8 in units of u64s, so the max
  65  * size is just under 2k. The common part also contains a type tag for the
  66  * value, and a format field indicating whether the key is packed or not (and
  67  * also meant to allow adding new key fields in the future, if desired).
  68  *
  69  * bkeys, when stored within a btree node, may also be packed. In that case, the
  70  * bkey_format in that node is used to unpack it. Packed bkeys mean that we can
  71  * be generous with field sizes in the common part of the key format (64 bit
  72  * inode number, 64 bit offset, 96 bit version field, etc.) for negligible cost.
  73  */
  74
  75 #include <asm/types.h>
  76 #include <asm/byteorder.h>
  77 #include <linux/kernel.h>
  78 #include <linux/uuid.h>
  79 #include "vstructs.h"
  80
  81 #ifdef __KERNEL__
  82 typedef uuid_t __uuid_t;
  83 #endif
  84
  85 #define BITMASK(name, type, field, offset, end)                         \
  86 static const unsigned   name##_OFFSET = offset;                         \
  87 static const unsigned   name##_BITS = (end - offset);                   \
  88                                                                         \
  89 static inline __u64 name(const type *k)                                 \
  90 {                                                                       \
  91         return (k->field >> offset) & ~(~0ULL << (end - offset));       \
  92 }                                                                       \
  93                                                                         \
  94 static inline void SET_##name(type *k, __u64 v)                         \
  95 {                                                                       \
  96         k->field &= ~(~(~0ULL << (end - offset)) << offset);            \
  97         k->field |= (v & ~(~0ULL << (end - offset))) << offset;         \
  98 }
  99
 100 #define LE_BITMASK(_bits, name, type, field, offset, end)               \
 101 static const unsigned   name##_OFFSET = offset;                         \
 102 static const unsigned   name##_BITS = (end - offset);                   \
 103 static const __u##_bits name##_MAX = (1ULL << (end - offset)) - 1;      \
 104                                                                         \
 105 static inline __u64 name(const type *k)                                 \
 106 {                                                                       \
 107         return (__le##_bits##_to_cpu(k->field) >> offset) &             \
 108                 ~(~0ULL << (end - offset));                             \
 109 }                                                                       \
 110                                                                         \
 111 static inline void SET_##name(type *k, __u64 v)                         \
 112 {                                                                       \
 113         __u##_bits new = __le##_bits##_to_cpu(k->field);                \
 114                                                                         \
 115         new &= ~(~(~0ULL << (end - offset)) << offset);                 \
 116         new |= (v & ~(~0ULL << (end - offset))) << offset;              \
 117         k->field = __cpu_to_le##_bits(new);                             \
 118 }
 119
 120 #define LE16_BITMASK(n, t, f, o, e)     LE_BITMASK(16, n, t, f, o, e)
 121 #define LE32_BITMASK(n, t, f, o, e)     LE_BITMASK(32, n, t, f, o, e)
 122 #define LE64_BITMASK(n, t, f, o, e)     LE_BITMASK(64, n, t, f, o, e)
 123
 124 struct bkey_format {
 125         __u8            key_u64s;
 126         __u8            nr_fields;
 127         /* One unused slot for now: */
 128         __u8            bits_per_field[6];
 129         __le64          field_offset[6];
 130 };
 131
 132 /* Btree keys - all units are in sectors */
 133
 134 struct bpos {
 135         /*
 136          * Word order matches machine byte order - btree code treats a bpos as a
 137          * single large integer, for search/comparison purposes
 138          *
 139          * Note that wherever a bpos is embedded in another on disk data
 140          * structure, it has to be byte swabbed when reading in metadata that
 141          * wasn't written in native endian order:
 142          */
 143 #if __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__
 144         __u32           snapshot;
 145         __u64           offset;
 146         __u64           inode;
 147 #elif __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__
 148         __u64           inode;
 149         __u64           offset;         /* Points to end of extent - sectors */
 150         __u32           snapshot;
 151 #else
 152 #error edit for your odd byteorder.
 153 #endif
 154 } __packed __aligned(4);
 155
 156 #define KEY_INODE_MAX                   ((__u64)~0ULL)
 157 #define KEY_OFFSET_MAX                  ((__u64)~0ULL)
 158 #define KEY_SNAPSHOT_MAX                ((__u32)~0U)
 159 #define KEY_SIZE_MAX                    ((__u32)~0U)
 160
 161 static inline struct bpos SPOS(__u64 inode, __u64 offset, __u32 snapshot)
 162 {
 163         return (struct bpos) {
 164                 .inode          = inode,
 165                 .offset         = offset,
 166                 .snapshot       = snapshot,
 167         };
 168 }
 169
 170 #define POS_MIN                         SPOS(0, 0, 0)
 171 #define POS_MAX                         SPOS(KEY_INODE_MAX, KEY_OFFSET_MAX, 0)
 172 #define SPOS_MAX                        SPOS(KEY_INODE_MAX, KEY_OFFSET_MAX, KEY_SNAPSHOT_MAX)
 173 #define POS(_inode, _offset)            SPOS(_inode, _offset, 0)
 174
 175 /* Empty placeholder struct, for container_of() */
 176 struct bch_val {
 177         __u64           __nothing[0];
 178 };
 179
 180 struct bversion {
 181 #if __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__
 182         __u64           lo;
 183         __u32           hi;
 184 #elif __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__
 185         __u32           hi;
 186         __u64           lo;
 187 #endif
 188 } __packed __aligned(4);
 189
 190 struct bkey {
 191         /* Size of combined key and value, in u64s */
 192         __u8            u64s;
 193
 194         /* Format of key (0 for format local to btree node) */
 195 #if defined(__LITTLE_ENDIAN_BITFIELD)
 196         __u8            format:7,
 197                         needs_whiteout:1;
 198 #elif defined (__BIG_ENDIAN_BITFIELD)
 199         __u8            needs_whiteout:1,
 200                         format:7;
 201 #else
 202 #error edit for your odd byteorder.
 203 #endif
 204
 205         /* Type of the value */
 206         __u8            type;
 207
 208 #if __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__
 209         __u8            pad[1];
 210
 211         struct bversion version;
 212         __u32           size;           /* extent size, in sectors */
 213         struct bpos     p;
 214 #elif __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__
 215         struct bpos     p;
 216         __u32           size;           /* extent size, in sectors */
 217         struct bversion version;
 218
 219         __u8            pad[1];
 220 #endif
 221 } __packed __aligned(8);
 222
 223 struct bkey_packed {
 224         __u64           _data[0];
 225
 226         /* Size of combined key and value, in u64s */
 227         __u8            u64s;
 228
 229         /* Format of key (0 for format local to btree node) */
 230
 231         /*
 232          * XXX: next incompat on disk format change, switch format and
 233          * needs_whiteout - bkey_packed() will be cheaper if format is the high
 234          * bits of the bitfield
 235          */
 236 #if defined(__LITTLE_ENDIAN_BITFIELD)
 237         __u8            format:7,
 238                         needs_whiteout:1;
 239 #elif defined (__BIG_ENDIAN_BITFIELD)
 240         __u8            needs_whiteout:1,
 241                         format:7;
 242 #endif
 243
 244         /* Type of the value */
 245         __u8            type;
 246         __u8            key_start[0];
 247
 248         /*
 249          * We copy bkeys with struct assignment in various places, and while
 250          * that shouldn't be done with packed bkeys we can't disallow it in C,
 251          * and it's legal to cast a bkey to a bkey_packed  - so padding it out
 252          * to the same size as struct bkey should hopefully be safest.
 253          */
 254         __u8            pad[sizeof(struct bkey) - 3];
 255 } __packed __aligned(8);
 256
 257 typedef struct {
 258         __le64                  lo;
 259         __le64                  hi;
 260 } bch_le128;
 261
 262 #define BKEY_U64s                       (sizeof(struct bkey) / sizeof(__u64))
 263 #define BKEY_U64s_MAX                   U8_MAX
 264 #define BKEY_VAL_U64s_MAX               (BKEY_U64s_MAX - BKEY_U64s)
 265
 266 #define KEY_PACKED_BITS_START           24
 267
 268 #define KEY_FORMAT_LOCAL_BTREE          0
 269 #define KEY_FORMAT_CURRENT              1
 270
 271 enum bch_bkey_fields {
 272         BKEY_FIELD_INODE,
 273         BKEY_FIELD_OFFSET,
 274         BKEY_FIELD_SNAPSHOT,
 275         BKEY_FIELD_SIZE,
 276         BKEY_FIELD_VERSION_HI,
 277         BKEY_FIELD_VERSION_LO,
 278         BKEY_NR_FIELDS,
 279 };
 280
 281 #define bkey_format_field(name, field)                                  \
 282         [BKEY_FIELD_##name] = (sizeof(((struct bkey *) NULL)->field) * 8)
 283
 284 #define BKEY_FORMAT_CURRENT                                             \
 285 ((struct bkey_format) {                                                 \
 286         .key_u64s       = BKEY_U64s,                                    \
 287         .nr_fields      = BKEY_NR_FIELDS,                               \
 288         .bits_per_field = {                                             \
 289                 bkey_format_field(INODE,        p.inode),               \
 290                 bkey_format_field(OFFSET,       p.offset),              \
 291                 bkey_format_field(SNAPSHOT,     p.snapshot),            \
 292                 bkey_format_field(SIZE,         size),                  \
 293                 bkey_format_field(VERSION_HI,   version.hi),            \
 294                 bkey_format_field(VERSION_LO,   version.lo),            \
 295         },                                                              \
 296 })
 297
 298 /* bkey with inline value */
 299 struct bkey_i {
 300         __u64                   _data[0];
 301
 302         struct bkey     k;
 303         struct bch_val  v;
 304 };
 305
 306 #define KEY(_inode, _offset, _size)                                     \
 307 ((struct bkey) {                                                        \
 308         .u64s           = BKEY_U64s,                                    \
 309         .format         = KEY_FORMAT_CURRENT,                           \
 310         .p              = POS(_inode, _offset),                         \
 311         .size           = _size,                                        \
 312 })
 313
 314 static inline void bkey_init(struct bkey *k)
 315 {
 316         *k = KEY(0, 0, 0);
 317 }
 318
 319 #define bkey_bytes(_k)          ((_k)->u64s * sizeof(__u64))
 320
 321 #define __BKEY_PADDED(key, pad)                                 \
 322         struct bkey_i key; __u64 key ## _pad[pad]
 323
 324 /*
 325  * - DELETED keys are used internally to mark keys that should be ignored but
 326  *   override keys in composition order.  Their version number is ignored.
 327  *
 328  * - DISCARDED keys indicate that the data is all 0s because it has been
 329  *   discarded. DISCARDs may have a version; if the version is nonzero the key
 330  *   will be persistent, otherwise the key will be dropped whenever the btree
 331  *   node is rewritten (like DELETED keys).
 332  *
 333  * - ERROR: any read of the data returns a read error, as the data was lost due
 334  *   to a failing device. Like DISCARDED keys, they can be removed (overridden)
 335  *   by new writes or cluster-wide GC. Node repair can also overwrite them with
 336  *   the same or a more recent version number, but not with an older version
 337  *   number.
 338  *
 339  * - WHITEOUT: for hash table btrees
 340  */
 341 #define BCH_BKEY_TYPES()                                \
 342         x(deleted,              0)                      \
 343         x(whiteout,             1)                      \
 344         x(error,                2)                      \
 345         x(cookie,               3)                      \
 346         x(hash_whiteout,        4)                      \
 347         x(btree_ptr,            5)                      \
 348         x(extent,               6)                      \
 349         x(reservation,          7)                      \
 350         x(inode,                8)                      \
 351         x(inode_generation,     9)                      \
 352         x(dirent,               10)                     \
 353         x(xattr,                11)                     \
 354         x(alloc,                12)                     \
 355         x(quota,                13)                     \
 356         x(stripe,               14)                     \
 357         x(reflink_p,            15)                     \
 358         x(reflink_v,            16)                     \
 359         x(inline_data,          17)                     \
 360         x(btree_ptr_v2,         18)                     \
 361         x(indirect_inline_data, 19)                     \
 362         x(alloc_v2,             20)                     \
 363         x(subvolume,            21)                     \
 364         x(snapshot,             22)                     \
 365         x(inode_v2,             23)                     \
 366         x(alloc_v3,             24)                     \
 367         x(set,                  25)                     \
 368         x(lru,                  26)                     \
 369         x(alloc_v4,             27)                     \
 370         x(backpointer,          28)                     \
 371         x(inode_v3,             29)                     \
 372         x(bucket_gens,          30)                     \
 373         x(snapshot_tree,        31)
 374
 375 enum bch_bkey_type {
 376 #define x(name, nr) KEY_TYPE_##name     = nr,
 377         BCH_BKEY_TYPES()
 378 #undef x
 379         KEY_TYPE_MAX,
 380 };
 381
 382 struct bch_deleted {
 383         struct bch_val          v;
 384 };
 385
 386 struct bch_whiteout {
 387         struct bch_val          v;
 388 };
 389
 390 struct bch_error {
 391         struct bch_val          v;
 392 };
 393
 394 struct bch_cookie {
 395         struct bch_val          v;
 396         __le64                  cookie;
 397 };
 398
 399 struct bch_hash_whiteout {
 400         struct bch_val          v;
 401 };
 402
 403 struct bch_set {
 404         struct bch_val          v;
 405 };
 406
 407 /* Extents */
 408
 409 /*
 410  * In extent bkeys, the value is a list of pointers (bch_extent_ptr), optionally
 411  * preceded by checksum/compression information (bch_extent_crc32 or
 412  * bch_extent_crc64).
 413  *
 414  * One major determining factor in the format of extents is how we handle and
 415  * represent extents that have been partially overwritten and thus trimmed:
 416  *
 417  * If an extent is not checksummed or compressed, when the extent is trimmed we
 418  * don't have to remember the extent we originally allocated and wrote: we can
 419  * merely adjust ptr->offset to point to the start of the data that is currently
 420  * live. The size field in struct bkey records the current (live) size of the
 421  * extent, and is also used to mean "size of region on disk that we point to" in
 422  * this case.
 423  *
 424  * Thus an extent that is not checksummed or compressed will consist only of a
 425  * list of bch_extent_ptrs, with none of the fields in
 426  * bch_extent_crc32/bch_extent_crc64.
 427  *
 428  * When an extent is checksummed or compressed, it's not possible to read only
 429  * the data that is currently live: we have to read the entire extent that was
 430  * originally written, and then return only the part of the extent that is
 431  * currently live.
 432  *
 433  * Thus, in addition to the current size of the extent in struct bkey, we need
 434  * to store the size of the originally allocated space - this is the
 435  * compressed_size and uncompressed_size fields in bch_extent_crc32/64. Also,
 436  * when the extent is trimmed, instead of modifying the offset field of the
 437  * pointer, we keep a second smaller offset field - "offset into the original
 438  * extent of the currently live region".
 439  *
 440  * The other major determining factor is replication and data migration:
 441  *
 442  * Each pointer may have its own bch_extent_crc32/64. When doing a replicated
 443  * write, we will initially write all the replicas in the same format, with the
 444  * same checksum type and compression format - however, when copygc runs later (or
 445  * tiering/cache promotion, anything that moves data), it is not in general
 446  * going to rewrite all the pointers at once - one of the replicas may be in a
 447  * bucket on one device that has very little fragmentation while another lives
 448  * in a bucket that has become heavily fragmented, and thus is being rewritten
 449  * sooner than the rest.
 450  *
 451  * Thus it will only move a subset of the pointers (or in the case of
 452  * tiering/cache promotion perhaps add a single pointer without dropping any
 453  * current pointers), and if the extent has been partially overwritten it must
 454  * write only the currently live portion (or copygc would not be able to reduce
 455  * fragmentation!) - which necessitates a different bch_extent_crc format for
 456  * the new pointer.
 457  *
 458  * But in the interests of space efficiency, we don't want to store one
 459  * bch_extent_crc for each pointer if we don't have to.
 460  *
 461  * Thus, a bch_extent consists of bch_extent_crc32s, bch_extent_crc64s, and
 462  * bch_extent_ptrs appended arbitrarily one after the other. We determine the
 463  * type of a given entry with a scheme similar to utf8 (except we're encoding a
 464  * type, not a size), encoding the type in the position of the first set bit:
 465  *
 466  * bch_extent_crc32     - 0b1
 467  * bch_extent_ptr       - 0b10
 468  * bch_extent_crc64     - 0b100
 469  *
 470  * We do it this way because bch_extent_crc32 is _very_ constrained on bits (and
 471  * bch_extent_crc64 is the least constrained).
 472  *
 473  * Then, each bch_extent_crc32/64 applies to the pointers that follow after it,
 474  * until the next bch_extent_crc32/64.
 475  *
 476  * If there are no bch_extent_crcs preceding a bch_extent_ptr, then that pointer
 477  * is neither checksummed nor compressed.
 478  */
 479
 480 /* 128 bits, sufficient for cryptographic MACs: */
 481 struct bch_csum {
 482         __le64                  lo;
 483         __le64                  hi;
 484 } __packed __aligned(8);
 485
 486 #define BCH_EXTENT_ENTRY_TYPES()                \
 487         x(ptr,                  0)              \
 488         x(crc32,                1)              \
 489         x(crc64,                2)              \
 490         x(crc128,               3)              \
 491         x(stripe_ptr,           4)              \
 492         x(rebalance,            5)
 493 #define BCH_EXTENT_ENTRY_MAX    6
 494
 495 enum bch_extent_entry_type {
 496 #define x(f, n) BCH_EXTENT_ENTRY_##f = n,
 497         BCH_EXTENT_ENTRY_TYPES()
 498 #undef x
 499 };
 500
 501 /* Compressed/uncompressed size are stored biased by 1: */
 502 struct bch_extent_crc32 {
 503 #if defined(__LITTLE_ENDIAN_BITFIELD)
 504         __u32                   type:2,
 505                                 _compressed_size:7,
 506                                 _uncompressed_size:7,
 507                                 offset:7,
 508                                 _unused:1,
 509                                 csum_type:4,
 510                                 compression_type:4;
 511         __u32                   csum;
 512 #elif defined (__BIG_ENDIAN_BITFIELD)
 513         __u32                   csum;
 514         __u32                   compression_type:4,
 515                                 csum_type:4,
 516                                 _unused:1,
 517                                 offset:7,
 518                                 _uncompressed_size:7,
 519                                 _compressed_size:7,
 520                                 type:2;
 521 #endif
 522 } __packed __aligned(8);
 523
 524 #define CRC32_SIZE_MAX          (1U << 7)
 525 #define CRC32_NONCE_MAX         0
 526
 527 struct bch_extent_crc64 {
 528 #if defined(__LITTLE_ENDIAN_BITFIELD)
 529         __u64                   type:3,
 530                                 _compressed_size:9,
 531                                 _uncompressed_size:9,
 532                                 offset:9,
 533                                 nonce:10,
 534                                 csum_type:4,
 535                                 compression_type:4,
 536                                 csum_hi:16;
 537 #elif defined (__BIG_ENDIAN_BITFIELD)
 538         __u64                   csum_hi:16,
 539                                 compression_type:4,
 540                                 csum_type:4,
 541                                 nonce:10,
 542                                 offset:9,
 543                                 _uncompressed_size:9,
 544                                 _compressed_size:9,
 545                                 type:3;
 546 #endif
 547         __u64                   csum_lo;
 548 } __packed __aligned(8);
 549
 550 #define CRC64_SIZE_MAX          (1U << 9)
 551 #define CRC64_NONCE_MAX         ((1U << 10) - 1)
 552
 553 struct bch_extent_crc128 {
 554 #if defined(__LITTLE_ENDIAN_BITFIELD)
 555         __u64                   type:4,
 556                                 _compressed_size:13,
 557                                 _uncompressed_size:13,
 558                                 offset:13,
 559                                 nonce:13,
 560                                 csum_type:4,
 561                                 compression_type:4;
 562 #elif defined (__BIG_ENDIAN_BITFIELD)
 563         __u64                   compression_type:4,
 564                                 csum_type:4,
 565                                 nonce:13,
 566                                 offset:13,
 567                                 _uncompressed_size:13,
 568                                 _compressed_size:13,
 569                                 type:4;
 570 #endif
 571         struct bch_csum         csum;
 572 } __packed __aligned(8);
 573
 574 #define CRC128_SIZE_MAX         (1U << 13)
 575 #define CRC128_NONCE_MAX        ((1U << 13) - 1)
 576
 577 /*
 578  * @reservation - pointer hasn't been written to, just reserved
 579  */
 580 struct bch_extent_ptr {
 581 #if defined(__LITTLE_ENDIAN_BITFIELD)
 582         __u64                   type:1,
 583                                 cached:1,
 584                                 unused:1,
 585                                 unwritten:1,
 586                                 offset:44, /* 8 petabytes */
 587                                 dev:8,
 588                                 gen:8;
 589 #elif defined (__BIG_ENDIAN_BITFIELD)
 590         __u64                   gen:8,
 591                                 dev:8,
 592                                 offset:44,
 593                                 unwritten:1,
 594                                 unused:1,
 595                                 cached:1,
 596                                 type:1;
 597 #endif
 598 } __packed __aligned(8);
 599
 600 struct bch_extent_stripe_ptr {
 601 #if defined(__LITTLE_ENDIAN_BITFIELD)
 602         __u64                   type:5,
 603                                 block:8,
 604                                 redundancy:4,
 605                                 idx:47;
 606 #elif defined (__BIG_ENDIAN_BITFIELD)
 607         __u64                   idx:47,
 608                                 redundancy:4,
 609                                 block:8,
 610                                 type:5;
 611 #endif
 612 };
 613
 614 struct bch_extent_reservation {
 615 #if defined(__LITTLE_ENDIAN_BITFIELD)
 616         __u64                   type:6,
 617                                 unused:22,
 618                                 replicas:4,
 619                                 generation:32;
 620 #elif defined (__BIG_ENDIAN_BITFIELD)
 621         __u64                   generation:32,
 622                                 replicas:4,
 623                                 unused:22,
 624                                 type:6;
 625 #endif
 626 };
 627
 628 struct bch_extent_rebalance {
 629 #if defined(__LITTLE_ENDIAN_BITFIELD)
 630         __u64                   type:7,
 631                                 unused:33,
 632                                 compression:8,
 633                                 target:16;
 634 #elif defined (__BIG_ENDIAN_BITFIELD)
 635         __u64                   target:16,
 636                                 compression:8,
 637                                 unused:33,
 638                                 type:7;
 639 #endif
 640 };
 641
 642 union bch_extent_entry {
 643 #if __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__ ||  __BITS_PER_LONG == 64
 644         unsigned long                   type;
 645 #elif __BITS_PER_LONG == 32
 646         struct {
 647                 unsigned long           pad;
 648                 unsigned long           type;
 649         };
 650 #else
 651 #error edit for your odd byteorder.
 652 #endif
 653
 654 #define x(f, n) struct bch_extent_##f   f;
 655         BCH_EXTENT_ENTRY_TYPES()
 656 #undef x
 657 };
 658
 659 struct bch_btree_ptr {
 660         struct bch_val          v;
 661
 662         __u64                   _data[0];
 663         struct bch_extent_ptr   start[];
 664 } __packed __aligned(8);
 665
 666 struct bch_btree_ptr_v2 {
 667         struct bch_val          v;
 668
 669         __u64                   mem_ptr;
 670         __le64                  seq;
 671         __le16                  sectors_written;
 672         __le16                  flags;
 673         struct bpos             min_key;
 674         __u64                   _data[0];
 675         struct bch_extent_ptr   start[];
 676 } __packed __aligned(8);
 677
 678 LE16_BITMASK(BTREE_PTR_RANGE_UPDATED,   struct bch_btree_ptr_v2, flags, 0, 1);
 679
 680 struct bch_extent {
 681         struct bch_val          v;
 682
 683         __u64                   _data[0];
 684         union bch_extent_entry  start[];
 685 } __packed __aligned(8);
 686
 687 struct bch_reservation {
 688         struct bch_val          v;
 689
 690         __le32                  generation;
 691         __u8                    nr_replicas;
 692         __u8                    pad[3];
 693 } __packed __aligned(8);
 694
 695 /* Maximum size (in u64s) a single pointer could be: */
 696 #define BKEY_EXTENT_PTR_U64s_MAX\
 697         ((sizeof(struct bch_extent_crc128) +                    \
 698           sizeof(struct bch_extent_ptr)) / sizeof(__u64))
 699
 700 /* Maximum possible size of an entire extent value: */
 701 #define BKEY_EXTENT_VAL_U64s_MAX                                \
 702         (1 + BKEY_EXTENT_PTR_U64s_MAX * (BCH_REPLICAS_MAX + 1))
 703
 704 /* * Maximum possible size of an entire extent, key + value: */
 705 #define BKEY_EXTENT_U64s_MAX            (BKEY_U64s + BKEY_EXTENT_VAL_U64s_MAX)
 706
 707 /* Btree pointers don't carry around checksums: */
 708 #define BKEY_BTREE_PTR_VAL_U64s_MAX                             \
 709         ((sizeof(struct bch_btree_ptr_v2) +                     \
 710           sizeof(struct bch_extent_ptr) * BCH_REPLICAS_MAX) / sizeof(__u64))
 711 #define BKEY_BTREE_PTR_U64s_MAX                                 \
 712         (BKEY_U64s + BKEY_BTREE_PTR_VAL_U64s_MAX)
 713
 714 /* Inodes */
 715
 716 #define BLOCKDEV_INODE_MAX      4096
 717
 718 #define BCACHEFS_ROOT_INO       4096
 719
 720 struct bch_inode {
 721         struct bch_val          v;
 722
 723         __le64                  bi_hash_seed;
 724         __le32                  bi_flags;
 725         __le16                  bi_mode;
 726         __u8                    fields[0];
 727 } __packed __aligned(8);
 728
 729 struct bch_inode_v2 {
 730         struct bch_val          v;
 731
 732         __le64                  bi_journal_seq;
 733         __le64                  bi_hash_seed;
 734         __le64                  bi_flags;
 735         __le16                  bi_mode;
 736         __u8                    fields[0];
 737 } __packed __aligned(8);
 738
 739 struct bch_inode_v3 {
 740         struct bch_val          v;
 741
 742         __le64                  bi_journal_seq;
 743         __le64                  bi_hash_seed;
 744         __le64                  bi_flags;
 745         __le64                  bi_sectors;
 746         __le64                  bi_size;
 747         __le64                  bi_version;
 748         __u8                    fields[0];
 749 } __packed __aligned(8);
 750
 751 #define INODEv3_FIELDS_START_INITIAL    6
 752 #define INODEv3_FIELDS_START_CUR        (offsetof(struct bch_inode_v3, fields) / sizeof(__u64))
 753
 754 struct bch_inode_generation {
 755         struct bch_val          v;
 756
 757         __le32                  bi_generation;
 758         __le32                  pad;
 759 } __packed __aligned(8);
 760
 761 /*
 762  * bi_subvol and bi_parent_subvol are only set for subvolume roots:
 763  */
 764
 765 #define BCH_INODE_FIELDS_v2()                   \
 766         x(bi_atime,                     96)     \
 767         x(bi_ctime,                     96)     \
 768         x(bi_mtime,                     96)     \
 769         x(bi_otime,                     96)     \
 770         x(bi_size,                      64)     \
 771         x(bi_sectors,                   64)     \
 772         x(bi_uid,                       32)     \
 773         x(bi_gid,                       32)     \
 774         x(bi_nlink,                     32)     \
 775         x(bi_generation,                32)     \
 776         x(bi_dev,                       32)     \
 777         x(bi_data_checksum,             8)      \
 778         x(bi_compression,               8)      \
 779         x(bi_project,                   32)     \
 780         x(bi_background_compression,    8)      \
 781         x(bi_data_replicas,             8)      \
 782         x(bi_promote_target,            16)     \
 783         x(bi_foreground_target,         16)     \
 784         x(bi_background_target,         16)     \
 785         x(bi_erasure_code,              16)     \
 786         x(bi_fields_set,                16)     \
 787         x(bi_dir,                       64)     \
 788         x(bi_dir_offset,                64)     \
 789         x(bi_subvol,                    32)     \
 790         x(bi_parent_subvol,             32)
 791
 792 #define BCH_INODE_FIELDS_v3()                   \
 793         x(bi_atime,                     96)     \
 794         x(bi_ctime,                     96)     \
 795         x(bi_mtime,                     96)     \
 796         x(bi_otime,                     96)     \
 797         x(bi_uid,                       32)     \
 798         x(bi_gid,                       32)     \
 799         x(bi_nlink,                     32)     \
 800         x(bi_generation,                32)     \
 801         x(bi_dev,                       32)     \
 802         x(bi_data_checksum,             8)      \
 803         x(bi_compression,               8)      \
 804         x(bi_project,                   32)     \
 805         x(bi_background_compression,    8)      \
 806         x(bi_data_replicas,             8)      \
 807         x(bi_promote_target,            16)     \
 808         x(bi_foreground_target,         16)     \
 809         x(bi_background_target,         16)     \
 810         x(bi_erasure_code,              16)     \
 811         x(bi_fields_set,                16)     \
 812         x(bi_dir,                       64)     \
 813         x(bi_dir_offset,                64)     \
 814         x(bi_subvol,                    32)     \
 815         x(bi_parent_subvol,             32)     \
 816         x(bi_nocow,                     8)
 817
 818 /* subset of BCH_INODE_FIELDS */
 819 #define BCH_INODE_OPTS()                        \
 820         x(data_checksum,                8)      \
 821         x(compression,                  8)      \
 822         x(project,                      32)     \
 823         x(background_compression,       8)      \
 824         x(data_replicas,                8)      \
 825         x(promote_target,               16)     \
 826         x(foreground_target,            16)     \
 827         x(background_target,            16)     \
 828         x(erasure_code,                 16)     \
 829         x(nocow,                        8)
 830
 831 enum inode_opt_id {
 832 #define x(name, ...)                            \
 833         Inode_opt_##name,
 834         BCH_INODE_OPTS()
 835 #undef  x
 836         Inode_opt_nr,
 837 };
 838
 839 enum {
 840         /*
 841          * User flags (get/settable with FS_IOC_*FLAGS, correspond to FS_*_FL
 842          * flags)
 843          */
 844         __BCH_INODE_SYNC                = 0,
 845         __BCH_INODE_IMMUTABLE           = 1,
 846         __BCH_INODE_APPEND              = 2,
 847         __BCH_INODE_NODUMP              = 3,
 848         __BCH_INODE_NOATIME             = 4,
 849
 850         __BCH_INODE_I_SIZE_DIRTY        = 5,
 851         __BCH_INODE_I_SECTORS_DIRTY     = 6,
 852         __BCH_INODE_UNLINKED            = 7,
 853         __BCH_INODE_BACKPTR_UNTRUSTED   = 8,
 854
 855         /* bits 20+ reserved for packed fields below: */
 856 };
 857
 858 #define BCH_INODE_SYNC          (1 << __BCH_INODE_SYNC)
 859 #define BCH_INODE_IMMUTABLE     (1 << __BCH_INODE_IMMUTABLE)
 860 #define BCH_INODE_APPEND        (1 << __BCH_INODE_APPEND)
 861 #define BCH_INODE_NODUMP        (1 << __BCH_INODE_NODUMP)
 862 #define BCH_INODE_NOATIME       (1 << __BCH_INODE_NOATIME)
 863 #define BCH_INODE_I_SIZE_DIRTY  (1 << __BCH_INODE_I_SIZE_DIRTY)
 864 #define BCH_INODE_I_SECTORS_DIRTY (1 << __BCH_INODE_I_SECTORS_DIRTY)
 865 #define BCH_INODE_UNLINKED      (1 << __BCH_INODE_UNLINKED)
 866 #define BCH_INODE_BACKPTR_UNTRUSTED (1 << __BCH_INODE_BACKPTR_UNTRUSTED)
 867
 868 LE32_BITMASK(INODE_STR_HASH,    struct bch_inode, bi_flags, 20, 24);
 869 LE32_BITMASK(INODE_NR_FIELDS,   struct bch_inode, bi_flags, 24, 31);
 870 LE32_BITMASK(INODE_NEW_VARINT,  struct bch_inode, bi_flags, 31, 32);
 871
 872 LE64_BITMASK(INODEv2_STR_HASH,  struct bch_inode_v2, bi_flags, 20, 24);
 873 LE64_BITMASK(INODEv2_NR_FIELDS, struct bch_inode_v2, bi_flags, 24, 31);
 874
 875 LE64_BITMASK(INODEv3_STR_HASH,  struct bch_inode_v3, bi_flags, 20, 24);
 876 LE64_BITMASK(INODEv3_NR_FIELDS, struct bch_inode_v3, bi_flags, 24, 31);
 877
 878 LE64_BITMASK(INODEv3_FIELDS_START,
 879                                 struct bch_inode_v3, bi_flags, 31, 36);
 880 LE64_BITMASK(INODEv3_MODE,      struct bch_inode_v3, bi_flags, 36, 52);
 881
 882 /* Dirents */
 883
 884 /*
 885  * Dirents (and xattrs) have to implement string lookups; since our b-tree
 886  * doesn't support arbitrary length strings for the key, we instead index by a
 887  * 64 bit hash (currently truncated sha1) of the string, stored in the offset
 888  * field of the key - using linear probing to resolve hash collisions. This also
 889  * provides us with the readdir cookie posix requires.
 890  *
 891  * Linear probing requires us to use whiteouts for deletions, in the event of a
 892  * collision:
 893  */
 894
 895 struct bch_dirent {
 896         struct bch_val          v;
 897
 898         /* Target inode number: */
 899         union {
 900         __le64                  d_inum;
 901         struct {                /* DT_SUBVOL */
 902         __le32                  d_child_subvol;
 903         __le32                  d_parent_subvol;
 904         };
 905         };
 906
 907         /*
 908          * Copy of mode bits 12-15 from the target inode - so userspace can get
 909          * the filetype without having to do a stat()
 910          */
 911         __u8                    d_type;
 912
 913         __u8                    d_name[];
 914 } __packed __aligned(8);
 915
 916 #define DT_SUBVOL       16
 917 #define BCH_DT_MAX      17
 918
 919 #define BCH_NAME_MAX    ((unsigned) (U8_MAX * sizeof(__u64) -           \
 920                          sizeof(struct bkey) -                          \
 921                          offsetof(struct bch_dirent, d_name)))
 922
 923 /* Xattrs */
 924
 925 #define KEY_TYPE_XATTR_INDEX_USER                       0
 926 #define KEY_TYPE_XATTR_INDEX_POSIX_ACL_ACCESS   1
 927 #define KEY_TYPE_XATTR_INDEX_POSIX_ACL_DEFAULT  2
 928 #define KEY_TYPE_XATTR_INDEX_TRUSTED                    3
 929 #define KEY_TYPE_XATTR_INDEX_SECURITY           4
 930
 931 struct bch_xattr {
 932         struct bch_val          v;
 933         __u8                    x_type;
 934         __u8                    x_name_len;
 935         __le16                  x_val_len;
 936         __u8                    x_name[];
 937 } __packed __aligned(8);
 938
 939 /* Bucket/allocation information: */
 940
 941 struct bch_alloc {
 942         struct bch_val          v;
 943         __u8                    fields;
 944         __u8                    gen;
 945         __u8                    data[];
 946 } __packed __aligned(8);
 947
 948 #define BCH_ALLOC_FIELDS_V1()                   \
 949         x(read_time,            16)             \
 950         x(write_time,           16)             \
 951         x(data_type,            8)              \
 952         x(dirty_sectors,        16)             \
 953         x(cached_sectors,       16)             \
 954         x(oldest_gen,           8)              \
 955         x(stripe,               32)             \
 956         x(stripe_redundancy,    8)
 957
 958 enum {
 959 #define x(name, _bits) BCH_ALLOC_FIELD_V1_##name,
 960         BCH_ALLOC_FIELDS_V1()
 961 #undef x
 962 };
 963
 964 struct bch_alloc_v2 {
 965         struct bch_val          v;
 966         __u8                    nr_fields;
 967         __u8                    gen;
 968         __u8                    oldest_gen;
 969         __u8                    data_type;
 970         __u8                    data[];
 971 } __packed __aligned(8);
 972
 973 #define BCH_ALLOC_FIELDS_V2()                   \
 974         x(read_time,            64)             \
 975         x(write_time,           64)             \
 976         x(dirty_sectors,        32)             \
 977         x(cached_sectors,       32)             \
 978         x(stripe,               32)             \
 979         x(stripe_redundancy,    8)
 980
 981 struct bch_alloc_v3 {
 982         struct bch_val          v;
 983         __le64                  journal_seq;
 984         __le32                  flags;
 985         __u8                    nr_fields;
 986         __u8                    gen;
 987         __u8                    oldest_gen;
 988         __u8                    data_type;
 989         __u8                    data[];
 990 } __packed __aligned(8);
 991
 992 LE32_BITMASK(BCH_ALLOC_V3_NEED_DISCARD,struct bch_alloc_v3, flags,  0,  1)
 993 LE32_BITMASK(BCH_ALLOC_V3_NEED_INC_GEN,struct bch_alloc_v3, flags,  1,  2)
 994
 995 struct bch_alloc_v4 {
 996         struct bch_val          v;
 997         __u64                   journal_seq;
 998         __u32                   flags;
 999         __u8                    gen;
1000         __u8                    oldest_gen;
1001         __u8                    data_type;
1002         __u8                    stripe_redundancy;
1003         __u32                   dirty_sectors;
1004         __u32                   cached_sectors;
1005         __u64                   io_time[2];
1006         __u32                   stripe;
1007         __u32                   nr_external_backpointers;
1008         __u64                   fragmentation_lru;
1009 } __packed __aligned(8);
1010
1011 #define BCH_ALLOC_V4_U64s_V0    6
1012 #define BCH_ALLOC_V4_U64s       (sizeof(struct bch_alloc_v4) / sizeof(__u64))
1013
1014 BITMASK(BCH_ALLOC_V4_NEED_DISCARD,      struct bch_alloc_v4, flags,  0,  1)
1015 BITMASK(BCH_ALLOC_V4_NEED_INC_GEN,      struct bch_alloc_v4, flags,  1,  2)
1016 BITMASK(BCH_ALLOC_V4_BACKPOINTERS_START,struct bch_alloc_v4, flags,  2,  8)
1017 BITMASK(BCH_ALLOC_V4_NR_BACKPOINTERS,   struct bch_alloc_v4, flags,  8,  14)
1018
1019 #define BCH_ALLOC_V4_NR_BACKPOINTERS_MAX        40
1020
1021 struct bch_backpointer {
1022         struct bch_val          v;
1023         __u8                    btree_id;
1024         __u8                    level;
1025         __u8                    data_type;
1026         __u64                   bucket_offset:40;
1027         __u32                   bucket_len;
1028         struct bpos             pos;
1029 } __packed __aligned(8);
1030
1031 #define KEY_TYPE_BUCKET_GENS_BITS       8
1032 #define KEY_TYPE_BUCKET_GENS_NR         (1U << KEY_TYPE_BUCKET_GENS_BITS)
1033 #define KEY_TYPE_BUCKET_GENS_MASK       (KEY_TYPE_BUCKET_GENS_NR - 1)
1034
1035 struct bch_bucket_gens {
1036         struct bch_val          v;
1037         u8                      gens[KEY_TYPE_BUCKET_GENS_NR];
1038 } __packed __aligned(8);
1039
1040 /* Quotas: */
1041
1042 enum quota_types {
1043         QTYP_USR                = 0,
1044         QTYP_GRP                = 1,
1045         QTYP_PRJ                = 2,
1046         QTYP_NR                 = 3,
1047 };
1048
1049 enum quota_counters {
1050         Q_SPC                   = 0,
1051         Q_INO                   = 1,
1052         Q_COUNTERS              = 2,
1053 };
1054
1055 struct bch_quota_counter {
1056         __le64                  hardlimit;
1057         __le64                  softlimit;
1058 };
1059
1060 struct bch_quota {
1061         struct bch_val          v;
1062         struct bch_quota_counter c[Q_COUNTERS];
1063 } __packed __aligned(8);
1064
1065 /* Erasure coding */
1066
1067 struct bch_stripe {
1068         struct bch_val          v;
1069         __le16                  sectors;
1070         __u8                    algorithm;
1071         __u8                    nr_blocks;
1072         __u8                    nr_redundant;
1073
1074         __u8                    csum_granularity_bits;
1075         __u8                    csum_type;
1076         __u8                    pad;
1077
1078         struct bch_extent_ptr   ptrs[];
1079 } __packed __aligned(8);
1080
1081 /* Reflink: */
1082
1083 struct bch_reflink_p {
1084         struct bch_val          v;
1085         __le64                  idx;
1086         /*
1087          * A reflink pointer might point to an indirect extent which is then
1088          * later split (by copygc or rebalance). If we only pointed to part of
1089          * the original indirect extent, and then one of the fragments is
1090          * outside the range we point to, we'd leak a refcount: so when creating
1091          * reflink pointers, we need to store pad values to remember the full
1092          * range we were taking a reference on.
1093          */
1094         __le32                  front_pad;
1095         __le32                  back_pad;
1096 } __packed __aligned(8);
1097
1098 struct bch_reflink_v {
1099         struct bch_val          v;
1100         __le64                  refcount;
1101         union bch_extent_entry  start[0];
1102         __u64                   _data[0];
1103 } __packed __aligned(8);
1104
1105 struct bch_indirect_inline_data {
1106         struct bch_val          v;
1107         __le64                  refcount;
1108         u8                      data[0];
1109 };
1110
1111 /* Inline data */
1112
1113 struct bch_inline_data {
1114         struct bch_val          v;
1115         u8                      data[0];
1116 };
1117
1118 /* Subvolumes: */
1119
1120 #define SUBVOL_POS_MIN          POS(0, 1)
1121 #define SUBVOL_POS_MAX          POS(0, S32_MAX)
1122 #define BCACHEFS_ROOT_SUBVOL    1
1123
1124 struct bch_subvolume {
1125         struct bch_val          v;
1126         __le32                  flags;
1127         __le32                  snapshot;
1128         __le64                  inode;
1129         __le32                  parent;
1130         __le32                  pad;
1131         bch_le128               otime;
1132 };
1133
1134 LE32_BITMASK(BCH_SUBVOLUME_RO,          struct bch_subvolume, flags,  0,  1)
1135 /*
1136  * We need to know whether a subvolume is a snapshot so we can know whether we
1137  * can delete it (or whether it should just be rm -rf'd)
1138  */
1139 LE32_BITMASK(BCH_SUBVOLUME_SNAP,        struct bch_subvolume, flags,  1,  2)
1140 LE32_BITMASK(BCH_SUBVOLUME_UNLINKED,    struct bch_subvolume, flags,  2,  3)
1141
1142 /* Snapshots */
1143
1144 struct bch_snapshot {
1145         struct bch_val          v;
1146         __le32                  flags;
1147         __le32                  parent;
1148         __le32                  children[2];
1149         __le32                  subvol;
1150         __le32                  tree;
1151         __le32                  depth;
1152         __le32                  skip[3];
1153 };
1154
1155 LE32_BITMASK(BCH_SNAPSHOT_DELETED,      struct bch_snapshot, flags,  0,  1)
1156
1157 /* True if a subvolume points to this snapshot node: */
1158 LE32_BITMASK(BCH_SNAPSHOT_SUBVOL,       struct bch_snapshot, flags,  1,  2)
1159
1160 /*
1161  * Snapshot trees:
1162  *
1163  * The snapshot_trees btree gives us persistent indentifier for each tree of
1164  * bch_snapshot nodes, and allow us to record and easily find the root/master
1165  * subvolume that other snapshots were created from:
1166  */
1167 struct bch_snapshot_tree {
1168         struct bch_val          v;
1169         __le32                  master_subvol;
1170         __le32                  root_snapshot;
1171 };
1172
1173 /* LRU btree: */
1174
1175 struct bch_lru {
1176         struct bch_val          v;
1177         __le64                  idx;
1178 } __packed __aligned(8);
1179
1180 #define LRU_ID_STRIPES          (1U << 16)
1181
1182 /* Optional/variable size superblock sections: */
1183
1184 struct bch_sb_field {
1185         __u64                   _data[0];
1186         __le32                  u64s;
1187         __le32                  type;
1188 };
1189
1190 #define BCH_SB_FIELDS()                         \
1191         x(journal,      0)                      \
1192         x(members,      1)                      \
1193         x(crypt,        2)                      \
1194         x(replicas_v0,  3)                      \
1195         x(quota,        4)                      \
1196         x(disk_groups,  5)                      \
1197         x(clean,        6)                      \
1198         x(replicas,     7)                      \
1199         x(journal_seq_blacklist, 8)             \
1200         x(journal_v2,   9)                      \
1201         x(counters,     10)
1202
1203 enum bch_sb_field_type {
1204 #define x(f, nr)        BCH_SB_FIELD_##f = nr,
1205         BCH_SB_FIELDS()
1206 #undef x
1207         BCH_SB_FIELD_NR
1208 };
1209
1210 /*
1211  * Most superblock fields are replicated in all device's superblocks - a few are
1212  * not:
1213  */
1214 #define BCH_SINGLE_DEVICE_SB_FIELDS             \
1215         ((1U << BCH_SB_FIELD_journal)|          \
1216          (1U << BCH_SB_FIELD_journal_v2))
1217
1218 /* BCH_SB_FIELD_journal: */
1219
1220 struct bch_sb_field_journal {
1221         struct bch_sb_field     field;
1222         __le64                  buckets[0];
1223 };
1224
1225 struct bch_sb_field_journal_v2 {
1226         struct bch_sb_field     field;
1227
1228         struct bch_sb_field_journal_v2_entry {
1229                 __le64          start;
1230                 __le64          nr;
1231         }                       d[0];
1232 };
1233
1234 /* BCH_SB_FIELD_members: */
1235
1236 #define BCH_MIN_NR_NBUCKETS     (1 << 6)
1237
1238 struct bch_member {
1239         __uuid_t                uuid;
1240         __le64                  nbuckets;       /* device size */
1241         __le16                  first_bucket;   /* index of first bucket used */
1242         __le16                  bucket_size;    /* sectors */
1243         __le32                  pad;
1244         __le64                  last_mount;     /* time_t */
1245
1246         __le64                  flags[2];
1247 };
1248
1249 LE64_BITMASK(BCH_MEMBER_STATE,          struct bch_member, flags[0],  0,  4)
1250 /* 4-14 unused, was TIER, HAS_(META)DATA, REPLACEMENT */
1251 LE64_BITMASK(BCH_MEMBER_DISCARD,        struct bch_member, flags[0], 14, 15)
1252 LE64_BITMASK(BCH_MEMBER_DATA_ALLOWED,   struct bch_member, flags[0], 15, 20)
1253 LE64_BITMASK(BCH_MEMBER_GROUP,          struct bch_member, flags[0], 20, 28)
1254 LE64_BITMASK(BCH_MEMBER_DURABILITY,     struct bch_member, flags[0], 28, 30)
1255 LE64_BITMASK(BCH_MEMBER_FREESPACE_INITIALIZED,
1256                                         struct bch_member, flags[0], 30, 31)
1257
1258 #if 0
1259 LE64_BITMASK(BCH_MEMBER_NR_READ_ERRORS, struct bch_member, flags[1], 0,  20);
1260 LE64_BITMASK(BCH_MEMBER_NR_WRITE_ERRORS,struct bch_member, flags[1], 20, 40);
1261 #endif
1262
1263 #define BCH_MEMBER_STATES()                     \
1264         x(rw,           0)                      \
1265         x(ro,           1)                      \
1266         x(failed,       2)                      \
1267         x(spare,        3)
1268
1269 enum bch_member_state {
1270 #define x(t, n) BCH_MEMBER_STATE_##t = n,
1271         BCH_MEMBER_STATES()
1272 #undef x
1273         BCH_MEMBER_STATE_NR
1274 };
1275
1276 struct bch_sb_field_members {
1277         struct bch_sb_field     field;
1278         struct bch_member       members[0];
1279 };
1280
1281 /* BCH_SB_FIELD_crypt: */
1282
1283 struct nonce {
1284         __le32                  d[4];
1285 };
1286
1287 struct bch_key {
1288         __le64                  key[4];
1289 };
1290
1291 #define BCH_KEY_MAGIC                                   \
1292         (((__u64) 'b' <<  0)|((__u64) 'c' <<  8)|               \
1293          ((__u64) 'h' << 16)|((__u64) '*' << 24)|               \
1294          ((__u64) '*' << 32)|((__u64) 'k' << 40)|               \
1295          ((__u64) 'e' << 48)|((__u64) 'y' << 56))
1296
1297 struct bch_encrypted_key {
1298         __le64                  magic;
1299         struct bch_key          key;
1300 };
1301
1302 /*
1303  * If this field is present in the superblock, it stores an encryption key which
1304  * is used encrypt all other data/metadata. The key will normally be encrypted
1305  * with the key userspace provides, but if encryption has been turned off we'll
1306  * just store the master key unencrypted in the superblock so we can access the
1307  * previously encrypted data.
1308  */
1309 struct bch_sb_field_crypt {
1310         struct bch_sb_field     field;
1311
1312         __le64                  flags;
1313         __le64                  kdf_flags;
1314         struct bch_encrypted_key key;
1315 };
1316
1317 LE64_BITMASK(BCH_CRYPT_KDF_TYPE,        struct bch_sb_field_crypt, flags, 0, 4);
1318
1319 enum bch_kdf_types {
1320         BCH_KDF_SCRYPT          = 0,
1321         BCH_KDF_NR              = 1,
1322 };
1323
1324 /* stored as base 2 log of scrypt params: */
1325 LE64_BITMASK(BCH_KDF_SCRYPT_N,  struct bch_sb_field_crypt, kdf_flags,  0, 16);
1326 LE64_BITMASK(BCH_KDF_SCRYPT_R,  struct bch_sb_field_crypt, kdf_flags, 16, 32);
1327 LE64_BITMASK(BCH_KDF_SCRYPT_P,  struct bch_sb_field_crypt, kdf_flags, 32, 48);
1328
1329 /* BCH_SB_FIELD_replicas: */
1330
1331 #define BCH_DATA_TYPES()                \
1332         x(free,         0)              \
1333         x(sb,           1)              \
1334         x(journal,      2)              \
1335         x(btree,        3)              \
1336         x(user,         4)              \
1337         x(cached,       5)              \
1338         x(parity,       6)              \
1339         x(stripe,       7)              \
1340         x(need_gc_gens, 8)              \
1341         x(need_discard, 9)
1342
1343 enum bch_data_type {
1344 #define x(t, n) BCH_DATA_##t,
1345         BCH_DATA_TYPES()
1346 #undef x
1347         BCH_DATA_NR
1348 };
1349
1350 static inline bool data_type_is_empty(enum bch_data_type type)
1351 {
1352         switch (type) {
1353         case BCH_DATA_free:
1354         case BCH_DATA_need_gc_gens:
1355         case BCH_DATA_need_discard:
1356                 return true;
1357         default:
1358                 return false;
1359         }
1360 }
1361
1362 static inline bool data_type_is_hidden(enum bch_data_type type)
1363 {
1364         switch (type) {
1365         case BCH_DATA_sb:
1366         case BCH_DATA_journal:
1367                 return true;
1368         default:
1369                 return false;
1370         }
1371 }
1372
1373 struct bch_replicas_entry_v0 {
1374         __u8                    data_type;
1375         __u8                    nr_devs;
1376         __u8                    devs[0];
1377 } __packed;
1378
1379 struct bch_sb_field_replicas_v0 {
1380         struct bch_sb_field     field;
1381         struct bch_replicas_entry_v0 entries[0];
1382 } __packed __aligned(8);
1383
1384 struct bch_replicas_entry {
1385         __u8                    data_type;
1386         __u8                    nr_devs;
1387         __u8                    nr_required;
1388         __u8                    devs[0];
1389 } __packed;
1390
1391 #define replicas_entry_bytes(_i)                                        \
1392         (offsetof(typeof(*(_i)), devs) + (_i)->nr_devs)
1393
1394 struct bch_sb_field_replicas {
1395         struct bch_sb_field     field;
1396         struct bch_replicas_entry entries[0];
1397 } __packed __aligned(8);
1398
1399 /* BCH_SB_FIELD_quota: */
1400
1401 struct bch_sb_quota_counter {
1402         __le32                          timelimit;
1403         __le32                          warnlimit;
1404 };
1405
1406 struct bch_sb_quota_type {
1407         __le64                          flags;
1408         struct bch_sb_quota_counter     c[Q_COUNTERS];
1409 };
1410
1411 struct bch_sb_field_quota {
1412         struct bch_sb_field             field;
1413         struct bch_sb_quota_type        q[QTYP_NR];
1414 } __packed __aligned(8);
1415
1416 /* BCH_SB_FIELD_disk_groups: */
1417
1418 #define BCH_SB_LABEL_SIZE               32
1419
1420 struct bch_disk_group {
1421         __u8                    label[BCH_SB_LABEL_SIZE];
1422         __le64                  flags[2];
1423 } __packed __aligned(8);
1424
1425 LE64_BITMASK(BCH_GROUP_DELETED,         struct bch_disk_group, flags[0], 0,  1)
1426 LE64_BITMASK(BCH_GROUP_DATA_ALLOWED,    struct bch_disk_group, flags[0], 1,  6)
1427 LE64_BITMASK(BCH_GROUP_PARENT,          struct bch_disk_group, flags[0], 6, 24)
1428
1429 struct bch_sb_field_disk_groups {
1430         struct bch_sb_field     field;
1431         struct bch_disk_group   entries[0];
1432 } __packed __aligned(8);
1433
1434 /* BCH_SB_FIELD_counters */
1435
1436 #define BCH_PERSISTENT_COUNTERS()                               \
1437         x(io_read,                                      0)      \
1438         x(io_write,                                     1)      \
1439         x(io_move,                                      2)      \
1440         x(bucket_invalidate,                            3)      \
1441         x(bucket_discard,                               4)      \
1442         x(bucket_alloc,                                 5)      \
1443         x(bucket_alloc_fail,                            6)      \
1444         x(btree_cache_scan,                             7)      \
1445         x(btree_cache_reap,                             8)      \
1446         x(btree_cache_cannibalize,                      9)      \
1447         x(btree_cache_cannibalize_lock,                 10)     \
1448         x(btree_cache_cannibalize_lock_fail,            11)     \
1449         x(btree_cache_cannibalize_unlock,               12)     \
1450         x(btree_node_write,                             13)     \
1451         x(btree_node_read,                              14)     \
1452         x(btree_node_compact,                           15)     \
1453         x(btree_node_merge,                             16)     \
1454         x(btree_node_split,                             17)     \
1455         x(btree_node_rewrite,                           18)     \
1456         x(btree_node_alloc,                             19)     \
1457         x(btree_node_free,                              20)     \
1458         x(btree_node_set_root,                          21)     \
1459         x(btree_path_relock_fail,                       22)     \
1460         x(btree_path_upgrade_fail,                      23)     \
1461         x(btree_reserve_get_fail,                       24)     \
1462         x(journal_entry_full,                           25)     \
1463         x(journal_full,                                 26)     \
1464         x(journal_reclaim_finish,                       27)     \
1465         x(journal_reclaim_start,                        28)     \
1466         x(journal_write,                                29)     \
1467         x(read_promote,                                 30)     \
1468         x(read_bounce,                                  31)     \
1469         x(read_split,                                   33)     \
1470         x(read_retry,                                   32)     \
1471         x(read_reuse_race,                              34)     \
1472         x(move_extent_read,                             35)     \
1473         x(move_extent_write,                            36)     \
1474         x(move_extent_finish,                           37)     \
1475         x(move_extent_fail,                             38)     \
1476         x(move_extent_alloc_mem_fail,                   39)     \
1477         x(copygc,                                       40)     \
1478         x(copygc_wait,                                  41)     \
1479         x(gc_gens_end,                                  42)     \
1480         x(gc_gens_start,                                43)     \
1481         x(trans_blocked_journal_reclaim,                44)     \
1482         x(trans_restart_btree_node_reused,              45)     \
1483         x(trans_restart_btree_node_split,               46)     \
1484         x(trans_restart_fault_inject,                   47)     \
1485         x(trans_restart_iter_upgrade,                   48)     \
1486         x(trans_restart_journal_preres_get,             49)     \
1487         x(trans_restart_journal_reclaim,                50)     \
1488         x(trans_restart_journal_res_get,                51)     \
1489         x(trans_restart_key_cache_key_realloced,        52)     \
1490         x(trans_restart_key_cache_raced,                53)     \
1491         x(trans_restart_mark_replicas,                  54)     \
1492         x(trans_restart_mem_realloced,                  55)     \
1493         x(trans_restart_memory_allocation_failure,      56)     \
1494         x(trans_restart_relock,                         57)     \
1495         x(trans_restart_relock_after_fill,              58)     \
1496         x(trans_restart_relock_key_cache_fill,          59)     \
1497         x(trans_restart_relock_next_node,               60)     \
1498         x(trans_restart_relock_parent_for_fill,         61)     \
1499         x(trans_restart_relock_path,                    62)     \
1500         x(trans_restart_relock_path_intent,             63)     \
1501         x(trans_restart_too_many_iters,                 64)     \
1502         x(trans_restart_traverse,                       65)     \
1503         x(trans_restart_upgrade,                        66)     \
1504         x(trans_restart_would_deadlock,                 67)     \
1505         x(trans_restart_would_deadlock_write,           68)     \
1506         x(trans_restart_injected,                       69)     \
1507         x(trans_restart_key_cache_upgrade,              70)     \
1508         x(trans_traverse_all,                           71)     \
1509         x(transaction_commit,                           72)     \
1510         x(write_super,                                  73)     \
1511         x(trans_restart_would_deadlock_recursion_limit, 74)     \
1512         x(trans_restart_write_buffer_flush,             75)     \
1513         x(trans_restart_split_race,                     76)
1514
1515 enum bch_persistent_counters {
1516 #define x(t, n, ...) BCH_COUNTER_##t,
1517         BCH_PERSISTENT_COUNTERS()
1518 #undef x
1519         BCH_COUNTER_NR
1520 };
1521
1522 struct bch_sb_field_counters {
1523         struct bch_sb_field     field;
1524         __le64                  d[0];
1525 };
1526
1527 /*
1528  * On clean shutdown, store btree roots and current journal sequence number in
1529  * the superblock:
1530  */
1531 struct jset_entry {
1532         __le16                  u64s;
1533         __u8                    btree_id;
1534         __u8                    level;
1535         __u8                    type; /* designates what this jset holds */
1536         __u8                    pad[3];
1537
1538         union {
1539                 struct bkey_i   start[0];
1540                 __u64           _data[0];
1541         };
1542 };
1543
1544 struct bch_sb_field_clean {
1545         struct bch_sb_field     field;
1546
1547         __le32                  flags;
1548         __le16                  _read_clock; /* no longer used */
1549         __le16                  _write_clock;
1550         __le64                  journal_seq;
1551
1552         union {
1553                 struct jset_entry start[0];
1554                 __u64           _data[0];
1555         };
1556 };
1557
1558 struct journal_seq_blacklist_entry {
1559         __le64                  start;
1560         __le64                  end;
1561 };
1562
1563 struct bch_sb_field_journal_seq_blacklist {
1564         struct bch_sb_field     field;
1565
1566         union {
1567                 struct journal_seq_blacklist_entry start[0];
1568                 __u64           _data[0];
1569         };
1570 };
1571
1572 /* Superblock: */
1573
1574 /*
1575  * New versioning scheme:
1576  * One common version number for all on disk data structures - superblock, btree
1577  * nodes, journal entries
1578  */
1579 #define BCH_VERSION_MAJOR(_v)           ((__u16) ((_v) >> 10))
1580 #define BCH_VERSION_MINOR(_v)           ((__u16) ((_v) & ~(~0U << 10)))
1581 #define BCH_VERSION(_major, _minor)     (((_major) << 10)|(_minor) << 0)
1582
1583 #define RECOVERY_PASS_ALL_FSCK          (1ULL << 63)
1584
1585 #define BCH_METADATA_VERSIONS()                                         \
1586         x(bkey_renumber,                BCH_VERSION(0, 10),             \
1587           RECOVERY_PASS_ALL_FSCK)                                       \
1588         x(inode_btree_change,           BCH_VERSION(0, 11),             \
1589           RECOVERY_PASS_ALL_FSCK)                                       \
1590         x(snapshot,                     BCH_VERSION(0, 12),             \
1591           RECOVERY_PASS_ALL_FSCK)                                       \
1592         x(inode_backpointers,           BCH_VERSION(0, 13),             \
1593           RECOVERY_PASS_ALL_FSCK)                                       \
1594         x(btree_ptr_sectors_written,    BCH_VERSION(0, 14),             \
1595           RECOVERY_PASS_ALL_FSCK)                                       \
1596         x(snapshot_2,                   BCH_VERSION(0, 15),             \
1597           BIT_ULL(BCH_RECOVERY_PASS_fs_upgrade_for_subvolumes)|         \
1598           BIT_ULL(BCH_RECOVERY_PASS_initialize_subvolumes)|             \
1599           RECOVERY_PASS_ALL_FSCK)                                       \
1600         x(reflink_p_fix,                BCH_VERSION(0, 16),             \
1601           BIT_ULL(BCH_RECOVERY_PASS_fix_reflink_p))                     \
1602         x(subvol_dirent,                BCH_VERSION(0, 17),             \
1603           RECOVERY_PASS_ALL_FSCK)                                       \
1604         x(inode_v2,                     BCH_VERSION(0, 18),             \
1605           RECOVERY_PASS_ALL_FSCK)                                       \
1606         x(freespace,                    BCH_VERSION(0, 19),             \
1607           RECOVERY_PASS_ALL_FSCK)                                       \
1608         x(alloc_v4,                     BCH_VERSION(0, 20),             \
1609           RECOVERY_PASS_ALL_FSCK)                                       \
1610         x(new_data_types,               BCH_VERSION(0, 21),             \
1611           RECOVERY_PASS_ALL_FSCK)                                       \
1612         x(backpointers,                 BCH_VERSION(0, 22),             \
1613           RECOVERY_PASS_ALL_FSCK)                                       \
1614         x(inode_v3,                     BCH_VERSION(0, 23),             \
1615           RECOVERY_PASS_ALL_FSCK)                                       \
1616         x(unwritten_extents,            BCH_VERSION(0, 24),             \
1617           RECOVERY_PASS_ALL_FSCK)                                       \
1618         x(bucket_gens,                  BCH_VERSION(0, 25),             \
1619           BIT_ULL(BCH_RECOVERY_PASS_bucket_gens_init)|                  \
1620           RECOVERY_PASS_ALL_FSCK)                                       \
1621         x(lru_v2,                       BCH_VERSION(0, 26),             \
1622           RECOVERY_PASS_ALL_FSCK)                                       \
1623         x(fragmentation_lru,            BCH_VERSION(0, 27),             \
1624           RECOVERY_PASS_ALL_FSCK)                                       \
1625         x(no_bps_in_alloc_keys,         BCH_VERSION(0, 28),             \
1626           RECOVERY_PASS_ALL_FSCK)                                       \
1627         x(snapshot_trees,               BCH_VERSION(0, 29),             \
1628           RECOVERY_PASS_ALL_FSCK)                                       \
1629         x(major_minor,                  BCH_VERSION(1,  0),             \
1630           0)                                                            \
1631         x(snapshot_skiplists,           BCH_VERSION(1,  1),             \
1632           BIT_ULL(BCH_RECOVERY_PASS_check_snapshots))
1633
1634 enum bcachefs_metadata_version {
1635         bcachefs_metadata_version_min = 9,
1636 #define x(t, n, upgrade_passes) bcachefs_metadata_version_##t = n,
1637         BCH_METADATA_VERSIONS()
1638 #undef x
1639         bcachefs_metadata_version_max
1640 };
1641
1642 static const unsigned bcachefs_metadata_required_upgrade_below = bcachefs_metadata_version_major_minor;
1643
1644 #define bcachefs_metadata_version_current       (bcachefs_metadata_version_max - 1)
1645
1646 #define BCH_SB_SECTOR                   8
1647 #define BCH_SB_MEMBERS_MAX              64 /* XXX kill */
1648
1649 struct bch_sb_layout {
1650         __uuid_t                magic;  /* bcachefs superblock UUID */
1651         __u8                    layout_type;
1652         __u8                    sb_max_size_bits; /* base 2 of 512 byte sectors */
1653         __u8                    nr_superblocks;
1654         __u8                    pad[5];
1655         __le64                  sb_offset[61];
1656 } __packed __aligned(8);
1657
1658 #define BCH_SB_LAYOUT_SECTOR    7
1659
1660 /*
1661  * @offset      - sector where this sb was written
1662  * @version     - on disk format version
1663  * @version_min - Oldest metadata version this filesystem contains; so we can
1664  *                safely drop compatibility code and refuse to mount filesystems
1665  *                we'd need it for
1666  * @magic       - identifies as a bcachefs superblock (BCHFS_MAGIC)
1667  * @seq         - incremented each time superblock is written
1668  * @uuid        - used for generating various magic numbers and identifying
1669  *                member devices, never changes
1670  * @user_uuid   - user visible UUID, may be changed
1671  * @label       - filesystem label
1672  * @seq         - identifies most recent superblock, incremented each time
1673  *                superblock is written
1674  * @features    - enabled incompatible features
1675  */
1676 struct bch_sb {
1677         struct bch_csum         csum;
1678         __le16                  version;
1679         __le16                  version_min;
1680         __le16                  pad[2];
1681         __uuid_t                magic;
1682         __uuid_t                uuid;
1683         __uuid_t                user_uuid;
1684         __u8                    label[BCH_SB_LABEL_SIZE];
1685         __le64                  offset;
1686         __le64                  seq;
1687
1688         __le16                  block_size;
1689         __u8                    dev_idx;
1690         __u8                    nr_devices;
1691         __le32                  u64s;
1692
1693         __le64                  time_base_lo;
1694         __le32                  time_base_hi;
1695         __le32                  time_precision;
1696
1697         __le64                  flags[8];
1698         __le64                  features[2];
1699         __le64                  compat[2];
1700
1701         struct bch_sb_layout    layout;
1702
1703         union {
1704                 struct bch_sb_field start[0];
1705                 __le64          _data[0];
1706         };
1707 } __packed __aligned(8);
1708
1709 /*
1710  * Flags:
1711  * BCH_SB_INITALIZED    - set on first mount
1712  * BCH_SB_CLEAN         - did we shut down cleanly? Just a hint, doesn't affect
1713  *                        behaviour of mount/recovery path:
1714  * BCH_SB_INODE_32BIT   - limit inode numbers to 32 bits
1715  * BCH_SB_128_BIT_MACS  - 128 bit macs instead of 80
1716  * BCH_SB_ENCRYPTION_TYPE - if nonzero encryption is enabled; overrides
1717  *                         DATA/META_CSUM_TYPE. Also indicates encryption
1718  *                         algorithm in use, if/when we get more than one
1719  */
1720
1721 LE16_BITMASK(BCH_SB_BLOCK_SIZE,         struct bch_sb, block_size, 0, 16);
1722
1723 LE64_BITMASK(BCH_SB_INITIALIZED,        struct bch_sb, flags[0],  0,  1);
1724 LE64_BITMASK(BCH_SB_CLEAN,              struct bch_sb, flags[0],  1,  2);
1725 LE64_BITMASK(BCH_SB_CSUM_TYPE,          struct bch_sb, flags[0],  2,  8);
1726 LE64_BITMASK(BCH_SB_ERROR_ACTION,       struct bch_sb, flags[0],  8, 12);
1727
1728 LE64_BITMASK(BCH_SB_BTREE_NODE_SIZE,    struct bch_sb, flags[0], 12, 28);
1729
1730 LE64_BITMASK(BCH_SB_GC_RESERVE,         struct bch_sb, flags[0], 28, 33);
1731 LE64_BITMASK(BCH_SB_ROOT_RESERVE,       struct bch_sb, flags[0], 33, 40);
1732
1733 LE64_BITMASK(BCH_SB_META_CSUM_TYPE,     struct bch_sb, flags[0], 40, 44);
1734 LE64_BITMASK(BCH_SB_DATA_CSUM_TYPE,     struct bch_sb, flags[0], 44, 48);
1735
1736 LE64_BITMASK(BCH_SB_META_REPLICAS_WANT, struct bch_sb, flags[0], 48, 52);
1737 LE64_BITMASK(BCH_SB_DATA_REPLICAS_WANT, struct bch_sb, flags[0], 52, 56);
1738
1739 LE64_BITMASK(BCH_SB_POSIX_ACL,          struct bch_sb, flags[0], 56, 57);
1740 LE64_BITMASK(BCH_SB_USRQUOTA,           struct bch_sb, flags[0], 57, 58);
1741 LE64_BITMASK(BCH_SB_GRPQUOTA,           struct bch_sb, flags[0], 58, 59);
1742 LE64_BITMASK(BCH_SB_PRJQUOTA,           struct bch_sb, flags[0], 59, 60);
1743
1744 LE64_BITMASK(BCH_SB_HAS_ERRORS,         struct bch_sb, flags[0], 60, 61);
1745 LE64_BITMASK(BCH_SB_HAS_TOPOLOGY_ERRORS,struct bch_sb, flags[0], 61, 62);
1746
1747 LE64_BITMASK(BCH_SB_BIG_ENDIAN,         struct bch_sb, flags[0], 62, 63);
1748
1749 LE64_BITMASK(BCH_SB_STR_HASH_TYPE,      struct bch_sb, flags[1],  0,  4);
1750 LE64_BITMASK(BCH_SB_COMPRESSION_TYPE_LO,struct bch_sb, flags[1],  4,  8);
1751 LE64_BITMASK(BCH_SB_INODE_32BIT,        struct bch_sb, flags[1],  8,  9);
1752
1753 LE64_BITMASK(BCH_SB_128_BIT_MACS,       struct bch_sb, flags[1],  9, 10);
1754 LE64_BITMASK(BCH_SB_ENCRYPTION_TYPE,    struct bch_sb, flags[1], 10, 14);
1755
1756 /*
1757  * Max size of an extent that may require bouncing to read or write
1758  * (checksummed, compressed): 64k
1759  */
1760 LE64_BITMASK(BCH_SB_ENCODED_EXTENT_MAX_BITS,
1761                                         struct bch_sb, flags[1], 14, 20);
1762
1763 LE64_BITMASK(BCH_SB_META_REPLICAS_REQ,  struct bch_sb, flags[1], 20, 24);
1764 LE64_BITMASK(BCH_SB_DATA_REPLICAS_REQ,  struct bch_sb, flags[1], 24, 28);
1765
1766 LE64_BITMASK(BCH_SB_PROMOTE_TARGET,     struct bch_sb, flags[1], 28, 40);
1767 LE64_BITMASK(BCH_SB_FOREGROUND_TARGET,  struct bch_sb, flags[1], 40, 52);
1768 LE64_BITMASK(BCH_SB_BACKGROUND_TARGET,  struct bch_sb, flags[1], 52, 64);
1769
1770 LE64_BITMASK(BCH_SB_BACKGROUND_COMPRESSION_TYPE_LO,
1771                                         struct bch_sb, flags[2],  0,  4);
1772 LE64_BITMASK(BCH_SB_GC_RESERVE_BYTES,   struct bch_sb, flags[2],  4, 64);
1773
1774 LE64_BITMASK(BCH_SB_ERASURE_CODE,       struct bch_sb, flags[3],  0, 16);
1775 LE64_BITMASK(BCH_SB_METADATA_TARGET,    struct bch_sb, flags[3], 16, 28);
1776 LE64_BITMASK(BCH_SB_SHARD_INUMS,        struct bch_sb, flags[3], 28, 29);
1777 LE64_BITMASK(BCH_SB_INODES_USE_KEY_CACHE,struct bch_sb, flags[3], 29, 30);
1778 LE64_BITMASK(BCH_SB_JOURNAL_FLUSH_DELAY,struct bch_sb, flags[3], 30, 62);
1779 LE64_BITMASK(BCH_SB_JOURNAL_FLUSH_DISABLED,struct bch_sb, flags[3], 62, 63);
1780 LE64_BITMASK(BCH_SB_JOURNAL_RECLAIM_DELAY,struct bch_sb, flags[4], 0, 32);
1781 LE64_BITMASK(BCH_SB_JOURNAL_TRANSACTION_NAMES,struct bch_sb, flags[4], 32, 33);
1782 LE64_BITMASK(BCH_SB_NOCOW,              struct bch_sb, flags[4], 33, 34);
1783 LE64_BITMASK(BCH_SB_WRITE_BUFFER_SIZE,  struct bch_sb, flags[4], 34, 54);
1784 LE64_BITMASK(BCH_SB_VERSION_UPGRADE,    struct bch_sb, flags[4], 54, 56);
1785
1786 LE64_BITMASK(BCH_SB_COMPRESSION_TYPE_HI,struct bch_sb, flags[4], 56, 60);
1787 LE64_BITMASK(BCH_SB_BACKGROUND_COMPRESSION_TYPE_HI,
1788                                         struct bch_sb, flags[4], 60, 64);
1789
1790 LE64_BITMASK(BCH_SB_VERSION_UPGRADE_COMPLETE,
1791                                         struct bch_sb, flags[5],  0, 16);
1792
1793 static inline __u64 BCH_SB_COMPRESSION_TYPE(const struct bch_sb *sb)
1794 {
1795         return BCH_SB_COMPRESSION_TYPE_LO(sb) | (BCH_SB_COMPRESSION_TYPE_HI(sb) << 4);
1796 }
1797
1798 static inline void SET_BCH_SB_COMPRESSION_TYPE(struct bch_sb *sb, __u64 v)
1799 {
1800         SET_BCH_SB_COMPRESSION_TYPE_LO(sb, v);
1801         SET_BCH_SB_COMPRESSION_TYPE_HI(sb, v >> 4);
1802 }
1803
1804 static inline __u64 BCH_SB_BACKGROUND_COMPRESSION_TYPE(const struct bch_sb *sb)
1805 {
1806         return BCH_SB_BACKGROUND_COMPRESSION_TYPE_LO(sb) |
1807                 (BCH_SB_BACKGROUND_COMPRESSION_TYPE_HI(sb) << 4);
1808 }
1809
1810 static inline void SET_BCH_SB_BACKGROUND_COMPRESSION_TYPE(struct bch_sb *sb, __u64 v)
1811 {
1812         SET_BCH_SB_BACKGROUND_COMPRESSION_TYPE_LO(sb, v);
1813         SET_BCH_SB_BACKGROUND_COMPRESSION_TYPE_HI(sb, v >> 4);
1814 }
1815
1816 /*
1817  * Features:
1818  *
1819  * journal_seq_blacklist_v3:    gates BCH_SB_FIELD_journal_seq_blacklist
1820  * reflink:                     gates KEY_TYPE_reflink
1821  * inline_data:                 gates KEY_TYPE_inline_data
1822  * new_siphash:                 gates BCH_STR_HASH_siphash
1823  * new_extent_overwrite:        gates BTREE_NODE_NEW_EXTENT_OVERWRITE
1824  */
1825 #define BCH_SB_FEATURES()                       \
1826         x(lz4,                          0)      \
1827         x(gzip,                         1)      \
1828         x(zstd,                         2)      \
1829         x(atomic_nlink,                 3)      \
1830         x(ec,                           4)      \
1831         x(journal_seq_blacklist_v3,     5)      \
1832         x(reflink,                      6)      \
1833         x(new_siphash,                  7)      \
1834         x(inline_data,                  8)      \
1835         x(new_extent_overwrite,         9)      \
1836         x(incompressible,               10)     \
1837         x(btree_ptr_v2,                 11)     \
1838         x(extents_above_btree_updates,  12)     \
1839         x(btree_updates_journalled,     13)     \
1840         x(reflink_inline_data,          14)     \
1841         x(new_varint,                   15)     \
1842         x(journal_no_flush,             16)     \
1843         x(alloc_v2,                     17)     \
1844         x(extents_across_btree_nodes,   18)
1845
1846 #define BCH_SB_FEATURES_ALWAYS                          \
1847         ((1ULL << BCH_FEATURE_new_extent_overwrite)|    \
1848          (1ULL << BCH_FEATURE_extents_above_btree_updates)|\
1849          (1ULL << BCH_FEATURE_btree_updates_journalled)|\
1850          (1ULL << BCH_FEATURE_alloc_v2)|\
1851          (1ULL << BCH_FEATURE_extents_across_btree_nodes))
1852
1853 #define BCH_SB_FEATURES_ALL                             \
1854         (BCH_SB_FEATURES_ALWAYS|                        \
1855          (1ULL << BCH_FEATURE_new_siphash)|             \
1856          (1ULL << BCH_FEATURE_btree_ptr_v2)|            \
1857          (1ULL << BCH_FEATURE_new_varint)|              \
1858          (1ULL << BCH_FEATURE_journal_no_flush))
1859
1860 enum bch_sb_feature {
1861 #define x(f, n) BCH_FEATURE_##f,
1862         BCH_SB_FEATURES()
1863 #undef x
1864         BCH_FEATURE_NR,
1865 };
1866
1867 #define BCH_SB_COMPAT()                                 \
1868         x(alloc_info,                           0)      \
1869         x(alloc_metadata,                       1)      \
1870         x(extents_above_btree_updates_done,     2)      \
1871         x(bformat_overflow_done,                3)
1872
1873 enum bch_sb_compat {
1874 #define x(f, n) BCH_COMPAT_##f,
1875         BCH_SB_COMPAT()
1876 #undef x
1877         BCH_COMPAT_NR,
1878 };
1879
1880 /* options: */
1881
1882 #define BCH_VERSION_UPGRADE_OPTS()      \
1883         x(compatible,           0)      \
1884         x(incompatible,         1)      \
1885         x(none,                 2)
1886
1887 enum bch_version_upgrade_opts {
1888 #define x(t, n) BCH_VERSION_UPGRADE_##t = n,
1889         BCH_VERSION_UPGRADE_OPTS()
1890 #undef x
1891 };
1892
1893 #define BCH_REPLICAS_MAX                4U
1894
1895 #define BCH_BKEY_PTRS_MAX               16U
1896
1897 #define BCH_ERROR_ACTIONS()             \
1898         x(continue,             0)      \
1899         x(ro,                   1)      \
1900         x(panic,                2)
1901
1902 enum bch_error_actions {
1903 #define x(t, n) BCH_ON_ERROR_##t = n,
1904         BCH_ERROR_ACTIONS()
1905 #undef x
1906         BCH_ON_ERROR_NR
1907 };
1908
1909 #define BCH_STR_HASH_TYPES()            \
1910         x(crc32c,               0)      \
1911         x(crc64,                1)      \
1912         x(siphash_old,          2)      \
1913         x(siphash,              3)
1914
1915 enum bch_str_hash_type {
1916 #define x(t, n) BCH_STR_HASH_##t = n,
1917         BCH_STR_HASH_TYPES()
1918 #undef x
1919         BCH_STR_HASH_NR
1920 };
1921
1922 #define BCH_STR_HASH_OPTS()             \
1923         x(crc32c,               0)      \
1924         x(crc64,                1)      \
1925         x(siphash,              2)
1926
1927 enum bch_str_hash_opts {
1928 #define x(t, n) BCH_STR_HASH_OPT_##t = n,
1929         BCH_STR_HASH_OPTS()
1930 #undef x
1931         BCH_STR_HASH_OPT_NR
1932 };
1933
1934 #define BCH_CSUM_TYPES()                        \
1935         x(none,                         0)      \
1936         x(crc32c_nonzero,               1)      \
1937         x(crc64_nonzero,                2)      \
1938         x(chacha20_poly1305_80,         3)      \
1939         x(chacha20_poly1305_128,        4)      \
1940         x(crc32c,                       5)      \
1941         x(crc64,                        6)      \
1942         x(xxhash,                       7)
1943
1944 enum bch_csum_type {
1945 #define x(t, n) BCH_CSUM_##t = n,
1946         BCH_CSUM_TYPES()
1947 #undef x
1948         BCH_CSUM_NR
1949 };
1950
1951 static const unsigned bch_crc_bytes[] = {
1952         [BCH_CSUM_none]                         = 0,
1953         [BCH_CSUM_crc32c_nonzero]               = 4,
1954         [BCH_CSUM_crc32c]                       = 4,
1955         [BCH_CSUM_crc64_nonzero]                = 8,
1956         [BCH_CSUM_crc64]                        = 8,
1957         [BCH_CSUM_xxhash]                       = 8,
1958         [BCH_CSUM_chacha20_poly1305_80]         = 10,
1959         [BCH_CSUM_chacha20_poly1305_128]        = 16,
1960 };
1961
1962 static inline _Bool bch2_csum_type_is_encryption(enum bch_csum_type type)
1963 {
1964         switch (type) {
1965         case BCH_CSUM_chacha20_poly1305_80:
1966         case BCH_CSUM_chacha20_poly1305_128:
1967                 return true;
1968         default:
1969                 return false;
1970         }
1971 }
1972
1973 #define BCH_CSUM_OPTS()                 \
1974         x(none,                 0)      \
1975         x(crc32c,               1)      \
1976         x(crc64,                2)      \
1977         x(xxhash,               3)
1978
1979 enum bch_csum_opts {
1980 #define x(t, n) BCH_CSUM_OPT_##t = n,
1981         BCH_CSUM_OPTS()
1982 #undef x
1983         BCH_CSUM_OPT_NR
1984 };
1985
1986 #define BCH_COMPRESSION_TYPES()         \
1987         x(none,                 0)      \
1988         x(lz4_old,              1)      \
1989         x(gzip,                 2)      \
1990         x(lz4,                  3)      \
1991         x(zstd,                 4)      \
1992         x(incompressible,       5)
1993
1994 enum bch_compression_type {
1995 #define x(t, n) BCH_COMPRESSION_TYPE_##t = n,
1996         BCH_COMPRESSION_TYPES()
1997 #undef x
1998         BCH_COMPRESSION_TYPE_NR
1999 };
2000
2001 #define BCH_COMPRESSION_OPTS()          \
2002         x(none,         0)              \
2003         x(lz4,          1)              \
2004         x(gzip,         2)              \
2005         x(zstd,         3)
2006
2007 enum bch_compression_opts {
2008 #define x(t, n) BCH_COMPRESSION_OPT_##t = n,
2009         BCH_COMPRESSION_OPTS()
2010 #undef x
2011         BCH_COMPRESSION_OPT_NR
2012 };
2013
2014 /*
2015  * Magic numbers
2016  *
2017  * The various other data structures have their own magic numbers, which are
2018  * xored with the first part of the cache set's UUID
2019  */
2020
2021 #define BCACHE_MAGIC                                                    \
2022         UUID_INIT(0xc68573f6, 0x4e1a, 0x45ca,                           \
2023                   0x82, 0x65, 0xf5, 0x7f, 0x48, 0xba, 0x6d, 0x81)
2024 #define BCHFS_MAGIC                                                     \
2025         UUID_INIT(0xc68573f6, 0x66ce, 0x90a9,                           \
2026                   0xd9, 0x6a, 0x60, 0xcf, 0x80, 0x3d, 0xf7, 0xef)
2027
2028 #define BCACHEFS_STATFS_MAGIC           0xca451a4e
2029
2030 #define JSET_MAGIC              __cpu_to_le64(0x245235c1a3625032ULL)
2031 #define BSET_MAGIC              __cpu_to_le64(0x90135c78b99e07f5ULL)
2032
2033 static inline __le64 __bch2_sb_magic(struct bch_sb *sb)
2034 {
2035         __le64 ret;
2036
2037         memcpy(&ret, &sb->uuid, sizeof(ret));
2038         return ret;
2039 }
2040
2041 static inline __u64 __jset_magic(struct bch_sb *sb)
2042 {
2043         return __le64_to_cpu(__bch2_sb_magic(sb) ^ JSET_MAGIC);
2044 }
2045
2046 static inline __u64 __bset_magic(struct bch_sb *sb)
2047 {
2048         return __le64_to_cpu(__bch2_sb_magic(sb) ^ BSET_MAGIC);
2049 }
2050
2051 /* Journal */
2052
2053 #define JSET_KEYS_U64s  (sizeof(struct jset_entry) / sizeof(__u64))
2054
2055 #define BCH_JSET_ENTRY_TYPES()                  \
2056         x(btree_keys,           0)              \
2057         x(btree_root,           1)              \
2058         x(prio_ptrs,            2)              \
2059         x(blacklist,            3)              \
2060         x(blacklist_v2,         4)              \
2061         x(usage,                5)              \
2062         x(data_usage,           6)              \
2063         x(clock,                7)              \
2064         x(dev_usage,            8)              \
2065         x(log,                  9)              \
2066         x(overwrite,            10)
2067
2068 enum {
2069 #define x(f, nr)        BCH_JSET_ENTRY_##f      = nr,
2070         BCH_JSET_ENTRY_TYPES()
2071 #undef x
2072         BCH_JSET_ENTRY_NR
2073 };
2074
2075 /*
2076  * Journal sequence numbers can be blacklisted: bsets record the max sequence
2077  * number of all the journal entries they contain updates for, so that on
2078  * recovery we can ignore those bsets that contain index updates newer that what
2079  * made it into the journal.
2080  *
2081  * This means that we can't reuse that journal_seq - we have to skip it, and
2082  * then record that we skipped it so that the next time we crash and recover we
2083  * don't think there was a missing journal entry.
2084  */
2085 struct jset_entry_blacklist {
2086         struct jset_entry       entry;
2087         __le64                  seq;
2088 };
2089
2090 struct jset_entry_blacklist_v2 {
2091         struct jset_entry       entry;
2092         __le64                  start;
2093         __le64                  end;
2094 };
2095
2096 #define BCH_FS_USAGE_TYPES()                    \
2097         x(reserved,             0)              \
2098         x(inodes,               1)              \
2099         x(key_version,          2)
2100
2101 enum {
2102 #define x(f, nr)        BCH_FS_USAGE_##f        = nr,
2103         BCH_FS_USAGE_TYPES()
2104 #undef x
2105         BCH_FS_USAGE_NR
2106 };
2107
2108 struct jset_entry_usage {
2109         struct jset_entry       entry;
2110         __le64                  v;
2111 } __packed;
2112
2113 struct jset_entry_data_usage {
2114         struct jset_entry       entry;
2115         __le64                  v;
2116         struct bch_replicas_entry r;
2117 } __packed;
2118
2119 struct jset_entry_clock {
2120         struct jset_entry       entry;
2121         __u8                    rw;
2122         __u8                    pad[7];
2123         __le64                  time;
2124 } __packed;
2125
2126 struct jset_entry_dev_usage_type {
2127         __le64                  buckets;
2128         __le64                  sectors;
2129         __le64                  fragmented;
2130 } __packed;
2131
2132 struct jset_entry_dev_usage {
2133         struct jset_entry       entry;
2134         __le32                  dev;
2135         __u32                   pad;
2136
2137         __le64                  buckets_ec;
2138         __le64                  _buckets_unavailable; /* No longer used */
2139
2140         struct jset_entry_dev_usage_type d[];
2141 } __packed;
2142
2143 static inline unsigned jset_entry_dev_usage_nr_types(struct jset_entry_dev_usage *u)
2144 {
2145         return (vstruct_bytes(&u->entry) - sizeof(struct jset_entry_dev_usage)) /
2146                 sizeof(struct jset_entry_dev_usage_type);
2147 }
2148
2149 struct jset_entry_log {
2150         struct jset_entry       entry;
2151         u8                      d[];
2152 } __packed;
2153
2154 /*
2155  * On disk format for a journal entry:
2156  * seq is monotonically increasing; every journal entry has its own unique
2157  * sequence number.
2158  *
2159  * last_seq is the oldest journal entry that still has keys the btree hasn't
2160  * flushed to disk yet.
2161  *
2162  * version is for on disk format changes.
2163  */
2164 struct jset {
2165         struct bch_csum         csum;
2166
2167         __le64                  magic;
2168         __le64                  seq;
2169         __le32                  version;
2170         __le32                  flags;
2171
2172         __le32                  u64s; /* size of d[] in u64s */
2173
2174         __u8                    encrypted_start[0];
2175
2176         __le16                  _read_clock; /* no longer used */
2177         __le16                  _write_clock;
2178
2179         /* Sequence number of oldest dirty journal entry */
2180         __le64                  last_seq;
2181
2182
2183         union {
2184                 struct jset_entry start[0];
2185                 __u64           _data[0];
2186         };
2187 } __packed __aligned(8);
2188
2189 LE32_BITMASK(JSET_CSUM_TYPE,    struct jset, flags, 0, 4);
2190 LE32_BITMASK(JSET_BIG_ENDIAN,   struct jset, flags, 4, 5);
2191 LE32_BITMASK(JSET_NO_FLUSH,     struct jset, flags, 5, 6);
2192
2193 #define BCH_JOURNAL_BUCKETS_MIN         8
2194
2195 /* Btree: */
2196
2197 #define BCH_BTREE_IDS()                         \
2198         x(extents,              0)              \
2199         x(inodes,               1)              \
2200         x(dirents,              2)              \
2201         x(xattrs,               3)              \
2202         x(alloc,                4)              \
2203         x(quotas,               5)              \
2204         x(stripes,              6)              \
2205         x(reflink,              7)              \
2206         x(subvolumes,           8)              \
2207         x(snapshots,            9)              \
2208         x(lru,                  10)             \
2209         x(freespace,            11)             \
2210         x(need_discard,         12)             \
2211         x(backpointers,         13)             \
2212         x(bucket_gens,          14)             \
2213         x(snapshot_trees,       15)
2214
2215 enum btree_id {
2216 #define x(kwd, val) BTREE_ID_##kwd = val,
2217         BCH_BTREE_IDS()
2218 #undef x
2219         BTREE_ID_NR
2220 };
2221
2222 #define BTREE_MAX_DEPTH         4U
2223
2224 /* Btree nodes */
2225
2226 /*
2227  * Btree nodes
2228  *
2229  * On disk a btree node is a list/log of these; within each set the keys are
2230  * sorted
2231  */
2232 struct bset {
2233         __le64                  seq;
2234
2235         /*
2236          * Highest journal entry this bset contains keys for.
2237          * If on recovery we don't see that journal entry, this bset is ignored:
2238          * this allows us to preserve the order of all index updates after a
2239          * crash, since the journal records a total order of all index updates
2240          * and anything that didn't make it to the journal doesn't get used.
2241          */
2242         __le64                  journal_seq;
2243
2244         __le32                  flags;
2245         __le16                  version;
2246         __le16                  u64s; /* count of d[] in u64s */
2247
2248         union {
2249                 struct bkey_packed start[0];
2250                 __u64           _data[0];
2251         };
2252 } __packed __aligned(8);
2253
2254 LE32_BITMASK(BSET_CSUM_TYPE,    struct bset, flags, 0, 4);
2255
2256 LE32_BITMASK(BSET_BIG_ENDIAN,   struct bset, flags, 4, 5);
2257 LE32_BITMASK(BSET_SEPARATE_WHITEOUTS,
2258                                 struct bset, flags, 5, 6);
2259
2260 /* Sector offset within the btree node: */
2261 LE32_BITMASK(BSET_OFFSET,       struct bset, flags, 16, 32);
2262
2263 struct btree_node {
2264         struct bch_csum         csum;
2265         __le64                  magic;
2266
2267         /* this flags field is encrypted, unlike bset->flags: */
2268         __le64                  flags;
2269
2270         /* Closed interval: */
2271         struct bpos             min_key;
2272         struct bpos             max_key;
2273         struct bch_extent_ptr   _ptr; /* not used anymore */
2274         struct bkey_format      format;
2275
2276         union {
2277         struct bset             keys;
2278         struct {
2279                 __u8            pad[22];
2280                 __le16          u64s;
2281                 __u64           _data[0];
2282
2283         };
2284         };
2285 } __packed __aligned(8);
2286
2287 LE64_BITMASK(BTREE_NODE_ID_LO,  struct btree_node, flags,  0,  4);
2288 LE64_BITMASK(BTREE_NODE_LEVEL,  struct btree_node, flags,  4,  8);
2289 LE64_BITMASK(BTREE_NODE_NEW_EXTENT_OVERWRITE,
2290                                 struct btree_node, flags,  8,  9);
2291 LE64_BITMASK(BTREE_NODE_ID_HI,  struct btree_node, flags,  9, 25);
2292 /* 25-32 unused */
2293 LE64_BITMASK(BTREE_NODE_SEQ,    struct btree_node, flags, 32, 64);
2294
2295 static inline __u64 BTREE_NODE_ID(struct btree_node *n)
2296 {
2297         return BTREE_NODE_ID_LO(n) | (BTREE_NODE_ID_HI(n) << 4);
2298 }
2299
2300 static inline void SET_BTREE_NODE_ID(struct btree_node *n, __u64 v)
2301 {
2302         SET_BTREE_NODE_ID_LO(n, v);
2303         SET_BTREE_NODE_ID_HI(n, v >> 4);
2304 }
2305
2306 struct btree_node_entry {
2307         struct bch_csum         csum;
2308
2309         union {
2310         struct bset             keys;
2311         struct {
2312                 __u8            pad[22];
2313                 __le16          u64s;
2314                 __u64           _data[0];
2315         };
2316         };
2317 } __packed __aligned(8);
2318
2319 #endif /* _BCACHEFS_FORMAT_H */