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 __maybe_unused unsigned    name##_OFFSET = offset;         \
  87 static const __maybe_unused 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 __maybe_unused unsigned    name##_OFFSET = offset;         \
 102 static const __maybe_unused unsigned    name##_BITS = (end - offset);   \
 103 static const __maybe_unused __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 POS_KEY(_pos)                                                   \
 307 ((struct bkey) {                                                        \
 308         .u64s           = BKEY_U64s,                                    \
 309         .format         = KEY_FORMAT_CURRENT,                           \
 310         .p              = _pos,                                         \
 311 })
 312
 313 #define KEY(_inode, _offset, _size)                                     \
 314 ((struct bkey) {                                                        \
 315         .u64s           = BKEY_U64s,                                    \
 316         .format         = KEY_FORMAT_CURRENT,                           \
 317         .p              = POS(_inode, _offset),                         \
 318         .size           = _size,                                        \
 319 })
 320
 321 static inline void bkey_init(struct bkey *k)
 322 {
 323         *k = KEY(0, 0, 0);
 324 }
 325
 326 #define bkey_bytes(_k)          ((_k)->u64s * sizeof(__u64))
 327
 328 #define __BKEY_PADDED(key, pad)                                 \
 329         struct bkey_i key; __u64 key ## _pad[pad]
 330
 331 /*
 332  * - DELETED keys are used internally to mark keys that should be ignored but
 333  *   override keys in composition order.  Their version number is ignored.
 334  *
 335  * - DISCARDED keys indicate that the data is all 0s because it has been
 336  *   discarded. DISCARDs may have a version; if the version is nonzero the key
 337  *   will be persistent, otherwise the key will be dropped whenever the btree
 338  *   node is rewritten (like DELETED keys).
 339  *
 340  * - ERROR: any read of the data returns a read error, as the data was lost due
 341  *   to a failing device. Like DISCARDED keys, they can be removed (overridden)
 342  *   by new writes or cluster-wide GC. Node repair can also overwrite them with
 343  *   the same or a more recent version number, but not with an older version
 344  *   number.
 345  *
 346  * - WHITEOUT: for hash table btrees
 347  */
 348 #define BCH_BKEY_TYPES()                                \
 349         x(deleted,              0)                      \
 350         x(whiteout,             1)                      \
 351         x(error,                2)                      \
 352         x(cookie,               3)                      \
 353         x(hash_whiteout,        4)                      \
 354         x(btree_ptr,            5)                      \
 355         x(extent,               6)                      \
 356         x(reservation,          7)                      \
 357         x(inode,                8)                      \
 358         x(inode_generation,     9)                      \
 359         x(dirent,               10)                     \
 360         x(xattr,                11)                     \
 361         x(alloc,                12)                     \
 362         x(quota,                13)                     \
 363         x(stripe,               14)                     \
 364         x(reflink_p,            15)                     \
 365         x(reflink_v,            16)                     \
 366         x(inline_data,          17)                     \
 367         x(btree_ptr_v2,         18)                     \
 368         x(indirect_inline_data, 19)                     \
 369         x(alloc_v2,             20)                     \
 370         x(subvolume,            21)                     \
 371         x(snapshot,             22)                     \
 372         x(inode_v2,             23)                     \
 373         x(alloc_v3,             24)                     \
 374         x(set,                  25)                     \
 375         x(lru,                  26)                     \
 376         x(alloc_v4,             27)                     \
 377         x(backpointer,          28)                     \
 378         x(inode_v3,             29)                     \
 379         x(bucket_gens,          30)                     \
 380         x(snapshot_tree,        31)                     \
 381         x(logged_op_truncate,   32)                     \
 382         x(logged_op_finsert,    33)
 383
 384 enum bch_bkey_type {
 385 #define x(name, nr) KEY_TYPE_##name     = nr,
 386         BCH_BKEY_TYPES()
 387 #undef x
 388         KEY_TYPE_MAX,
 389 };
 390
 391 struct bch_deleted {
 392         struct bch_val          v;
 393 };
 394
 395 struct bch_whiteout {
 396         struct bch_val          v;
 397 };
 398
 399 struct bch_error {
 400         struct bch_val          v;
 401 };
 402
 403 struct bch_cookie {
 404         struct bch_val          v;
 405         __le64                  cookie;
 406 };
 407
 408 struct bch_hash_whiteout {
 409         struct bch_val          v;
 410 };
 411
 412 struct bch_set {
 413         struct bch_val          v;
 414 };
 415
 416 /* Extents */
 417
 418 /*
 419  * In extent bkeys, the value is a list of pointers (bch_extent_ptr), optionally
 420  * preceded by checksum/compression information (bch_extent_crc32 or
 421  * bch_extent_crc64).
 422  *
 423  * One major determining factor in the format of extents is how we handle and
 424  * represent extents that have been partially overwritten and thus trimmed:
 425  *
 426  * If an extent is not checksummed or compressed, when the extent is trimmed we
 427  * don't have to remember the extent we originally allocated and wrote: we can
 428  * merely adjust ptr->offset to point to the start of the data that is currently
 429  * live. The size field in struct bkey records the current (live) size of the
 430  * extent, and is also used to mean "size of region on disk that we point to" in
 431  * this case.
 432  *
 433  * Thus an extent that is not checksummed or compressed will consist only of a
 434  * list of bch_extent_ptrs, with none of the fields in
 435  * bch_extent_crc32/bch_extent_crc64.
 436  *
 437  * When an extent is checksummed or compressed, it's not possible to read only
 438  * the data that is currently live: we have to read the entire extent that was
 439  * originally written, and then return only the part of the extent that is
 440  * currently live.
 441  *
 442  * Thus, in addition to the current size of the extent in struct bkey, we need
 443  * to store the size of the originally allocated space - this is the
 444  * compressed_size and uncompressed_size fields in bch_extent_crc32/64. Also,
 445  * when the extent is trimmed, instead of modifying the offset field of the
 446  * pointer, we keep a second smaller offset field - "offset into the original
 447  * extent of the currently live region".
 448  *
 449  * The other major determining factor is replication and data migration:
 450  *
 451  * Each pointer may have its own bch_extent_crc32/64. When doing a replicated
 452  * write, we will initially write all the replicas in the same format, with the
 453  * same checksum type and compression format - however, when copygc runs later (or
 454  * tiering/cache promotion, anything that moves data), it is not in general
 455  * going to rewrite all the pointers at once - one of the replicas may be in a
 456  * bucket on one device that has very little fragmentation while another lives
 457  * in a bucket that has become heavily fragmented, and thus is being rewritten
 458  * sooner than the rest.
 459  *
 460  * Thus it will only move a subset of the pointers (or in the case of
 461  * tiering/cache promotion perhaps add a single pointer without dropping any
 462  * current pointers), and if the extent has been partially overwritten it must
 463  * write only the currently live portion (or copygc would not be able to reduce
 464  * fragmentation!) - which necessitates a different bch_extent_crc format for
 465  * the new pointer.
 466  *
 467  * But in the interests of space efficiency, we don't want to store one
 468  * bch_extent_crc for each pointer if we don't have to.
 469  *
 470  * Thus, a bch_extent consists of bch_extent_crc32s, bch_extent_crc64s, and
 471  * bch_extent_ptrs appended arbitrarily one after the other. We determine the
 472  * type of a given entry with a scheme similar to utf8 (except we're encoding a
 473  * type, not a size), encoding the type in the position of the first set bit:
 474  *
 475  * bch_extent_crc32     - 0b1
 476  * bch_extent_ptr       - 0b10
 477  * bch_extent_crc64     - 0b100
 478  *
 479  * We do it this way because bch_extent_crc32 is _very_ constrained on bits (and
 480  * bch_extent_crc64 is the least constrained).
 481  *
 482  * Then, each bch_extent_crc32/64 applies to the pointers that follow after it,
 483  * until the next bch_extent_crc32/64.
 484  *
 485  * If there are no bch_extent_crcs preceding a bch_extent_ptr, then that pointer
 486  * is neither checksummed nor compressed.
 487  */
 488
 489 /* 128 bits, sufficient for cryptographic MACs: */
 490 struct bch_csum {
 491         __le64                  lo;
 492         __le64                  hi;
 493 } __packed __aligned(8);
 494
 495 #define BCH_EXTENT_ENTRY_TYPES()                \
 496         x(ptr,                  0)              \
 497         x(crc32,                1)              \
 498         x(crc64,                2)              \
 499         x(crc128,               3)              \
 500         x(stripe_ptr,           4)              \
 501         x(rebalance,            5)
 502 #define BCH_EXTENT_ENTRY_MAX    6
 503
 504 enum bch_extent_entry_type {
 505 #define x(f, n) BCH_EXTENT_ENTRY_##f = n,
 506         BCH_EXTENT_ENTRY_TYPES()
 507 #undef x
 508 };
 509
 510 /* Compressed/uncompressed size are stored biased by 1: */
 511 struct bch_extent_crc32 {
 512 #if defined(__LITTLE_ENDIAN_BITFIELD)
 513         __u32                   type:2,
 514                                 _compressed_size:7,
 515                                 _uncompressed_size:7,
 516                                 offset:7,
 517                                 _unused:1,
 518                                 csum_type:4,
 519                                 compression_type:4;
 520         __u32                   csum;
 521 #elif defined (__BIG_ENDIAN_BITFIELD)
 522         __u32                   csum;
 523         __u32                   compression_type:4,
 524                                 csum_type:4,
 525                                 _unused:1,
 526                                 offset:7,
 527                                 _uncompressed_size:7,
 528                                 _compressed_size:7,
 529                                 type:2;
 530 #endif
 531 } __packed __aligned(8);
 532
 533 #define CRC32_SIZE_MAX          (1U << 7)
 534 #define CRC32_NONCE_MAX         0
 535
 536 struct bch_extent_crc64 {
 537 #if defined(__LITTLE_ENDIAN_BITFIELD)
 538         __u64                   type:3,
 539                                 _compressed_size:9,
 540                                 _uncompressed_size:9,
 541                                 offset:9,
 542                                 nonce:10,
 543                                 csum_type:4,
 544                                 compression_type:4,
 545                                 csum_hi:16;
 546 #elif defined (__BIG_ENDIAN_BITFIELD)
 547         __u64                   csum_hi:16,
 548                                 compression_type:4,
 549                                 csum_type:4,
 550                                 nonce:10,
 551                                 offset:9,
 552                                 _uncompressed_size:9,
 553                                 _compressed_size:9,
 554                                 type:3;
 555 #endif
 556         __u64                   csum_lo;
 557 } __packed __aligned(8);
 558
 559 #define CRC64_SIZE_MAX          (1U << 9)
 560 #define CRC64_NONCE_MAX         ((1U << 10) - 1)
 561
 562 struct bch_extent_crc128 {
 563 #if defined(__LITTLE_ENDIAN_BITFIELD)
 564         __u64                   type:4,
 565                                 _compressed_size:13,
 566                                 _uncompressed_size:13,
 567                                 offset:13,
 568                                 nonce:13,
 569                                 csum_type:4,
 570                                 compression_type:4;
 571 #elif defined (__BIG_ENDIAN_BITFIELD)
 572         __u64                   compression_type:4,
 573                                 csum_type:4,
 574                                 nonce:13,
 575                                 offset:13,
 576                                 _uncompressed_size:13,
 577                                 _compressed_size:13,
 578                                 type:4;
 579 #endif
 580         struct bch_csum         csum;
 581 } __packed __aligned(8);
 582
 583 #define CRC128_SIZE_MAX         (1U << 13)
 584 #define CRC128_NONCE_MAX        ((1U << 13) - 1)
 585
 586 /*
 587  * @reservation - pointer hasn't been written to, just reserved
 588  */
 589 struct bch_extent_ptr {
 590 #if defined(__LITTLE_ENDIAN_BITFIELD)
 591         __u64                   type:1,
 592                                 cached:1,
 593                                 unused:1,
 594                                 unwritten:1,
 595                                 offset:44, /* 8 petabytes */
 596                                 dev:8,
 597                                 gen:8;
 598 #elif defined (__BIG_ENDIAN_BITFIELD)
 599         __u64                   gen:8,
 600                                 dev:8,
 601                                 offset:44,
 602                                 unwritten:1,
 603                                 unused:1,
 604                                 cached:1,
 605                                 type:1;
 606 #endif
 607 } __packed __aligned(8);
 608
 609 struct bch_extent_stripe_ptr {
 610 #if defined(__LITTLE_ENDIAN_BITFIELD)
 611         __u64                   type:5,
 612                                 block:8,
 613                                 redundancy:4,
 614                                 idx:47;
 615 #elif defined (__BIG_ENDIAN_BITFIELD)
 616         __u64                   idx:47,
 617                                 redundancy:4,
 618                                 block:8,
 619                                 type:5;
 620 #endif
 621 };
 622
 623 struct bch_extent_rebalance {
 624 #if defined(__LITTLE_ENDIAN_BITFIELD)
 625         __u64                   type:6,
 626                                 unused:34,
 627                                 compression:8, /* enum bch_compression_opt */
 628                                 target:16;
 629 #elif defined (__BIG_ENDIAN_BITFIELD)
 630         __u64                   target:16,
 631                                 compression:8,
 632                                 unused:34,
 633                                 type:6;
 634 #endif
 635 };
 636
 637 union bch_extent_entry {
 638 #if __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__ ||  __BITS_PER_LONG == 64
 639         unsigned long                   type;
 640 #elif __BITS_PER_LONG == 32
 641         struct {
 642                 unsigned long           pad;
 643                 unsigned long           type;
 644         };
 645 #else
 646 #error edit for your odd byteorder.
 647 #endif
 648
 649 #define x(f, n) struct bch_extent_##f   f;
 650         BCH_EXTENT_ENTRY_TYPES()
 651 #undef x
 652 };
 653
 654 struct bch_btree_ptr {
 655         struct bch_val          v;
 656
 657         __u64                   _data[0];
 658         struct bch_extent_ptr   start[];
 659 } __packed __aligned(8);
 660
 661 struct bch_btree_ptr_v2 {
 662         struct bch_val          v;
 663
 664         __u64                   mem_ptr;
 665         __le64                  seq;
 666         __le16                  sectors_written;
 667         __le16                  flags;
 668         struct bpos             min_key;
 669         __u64                   _data[0];
 670         struct bch_extent_ptr   start[];
 671 } __packed __aligned(8);
 672
 673 LE16_BITMASK(BTREE_PTR_RANGE_UPDATED,   struct bch_btree_ptr_v2, flags, 0, 1);
 674
 675 struct bch_extent {
 676         struct bch_val          v;
 677
 678         __u64                   _data[0];
 679         union bch_extent_entry  start[];
 680 } __packed __aligned(8);
 681
 682 struct bch_reservation {
 683         struct bch_val          v;
 684
 685         __le32                  generation;
 686         __u8                    nr_replicas;
 687         __u8                    pad[3];
 688 } __packed __aligned(8);
 689
 690 /* Maximum size (in u64s) a single pointer could be: */
 691 #define BKEY_EXTENT_PTR_U64s_MAX\
 692         ((sizeof(struct bch_extent_crc128) +                    \
 693           sizeof(struct bch_extent_ptr)) / sizeof(__u64))
 694
 695 /* Maximum possible size of an entire extent value: */
 696 #define BKEY_EXTENT_VAL_U64s_MAX                                \
 697         (1 + BKEY_EXTENT_PTR_U64s_MAX * (BCH_REPLICAS_MAX + 1))
 698
 699 /* * Maximum possible size of an entire extent, key + value: */
 700 #define BKEY_EXTENT_U64s_MAX            (BKEY_U64s + BKEY_EXTENT_VAL_U64s_MAX)
 701
 702 /* Btree pointers don't carry around checksums: */
 703 #define BKEY_BTREE_PTR_VAL_U64s_MAX                             \
 704         ((sizeof(struct bch_btree_ptr_v2) +                     \
 705           sizeof(struct bch_extent_ptr) * BCH_REPLICAS_MAX) / sizeof(__u64))
 706 #define BKEY_BTREE_PTR_U64s_MAX                                 \
 707         (BKEY_U64s + BKEY_BTREE_PTR_VAL_U64s_MAX)
 708
 709 /* Inodes */
 710
 711 #define BLOCKDEV_INODE_MAX      4096
 712
 713 #define BCACHEFS_ROOT_INO       4096
 714
 715 struct bch_inode {
 716         struct bch_val          v;
 717
 718         __le64                  bi_hash_seed;
 719         __le32                  bi_flags;
 720         __le16                  bi_mode;
 721         __u8                    fields[];
 722 } __packed __aligned(8);
 723
 724 struct bch_inode_v2 {
 725         struct bch_val          v;
 726
 727         __le64                  bi_journal_seq;
 728         __le64                  bi_hash_seed;
 729         __le64                  bi_flags;
 730         __le16                  bi_mode;
 731         __u8                    fields[];
 732 } __packed __aligned(8);
 733
 734 struct bch_inode_v3 {
 735         struct bch_val          v;
 736
 737         __le64                  bi_journal_seq;
 738         __le64                  bi_hash_seed;
 739         __le64                  bi_flags;
 740         __le64                  bi_sectors;
 741         __le64                  bi_size;
 742         __le64                  bi_version;
 743         __u8                    fields[];
 744 } __packed __aligned(8);
 745
 746 #define INODEv3_FIELDS_START_INITIAL    6
 747 #define INODEv3_FIELDS_START_CUR        (offsetof(struct bch_inode_v3, fields) / sizeof(__u64))
 748
 749 struct bch_inode_generation {
 750         struct bch_val          v;
 751
 752         __le32                  bi_generation;
 753         __le32                  pad;
 754 } __packed __aligned(8);
 755
 756 /*
 757  * bi_subvol and bi_parent_subvol are only set for subvolume roots:
 758  */
 759
 760 #define BCH_INODE_FIELDS_v2()                   \
 761         x(bi_atime,                     96)     \
 762         x(bi_ctime,                     96)     \
 763         x(bi_mtime,                     96)     \
 764         x(bi_otime,                     96)     \
 765         x(bi_size,                      64)     \
 766         x(bi_sectors,                   64)     \
 767         x(bi_uid,                       32)     \
 768         x(bi_gid,                       32)     \
 769         x(bi_nlink,                     32)     \
 770         x(bi_generation,                32)     \
 771         x(bi_dev,                       32)     \
 772         x(bi_data_checksum,             8)      \
 773         x(bi_compression,               8)      \
 774         x(bi_project,                   32)     \
 775         x(bi_background_compression,    8)      \
 776         x(bi_data_replicas,             8)      \
 777         x(bi_promote_target,            16)     \
 778         x(bi_foreground_target,         16)     \
 779         x(bi_background_target,         16)     \
 780         x(bi_erasure_code,              16)     \
 781         x(bi_fields_set,                16)     \
 782         x(bi_dir,                       64)     \
 783         x(bi_dir_offset,                64)     \
 784         x(bi_subvol,                    32)     \
 785         x(bi_parent_subvol,             32)
 786
 787 #define BCH_INODE_FIELDS_v3()                   \
 788         x(bi_atime,                     96)     \
 789         x(bi_ctime,                     96)     \
 790         x(bi_mtime,                     96)     \
 791         x(bi_otime,                     96)     \
 792         x(bi_uid,                       32)     \
 793         x(bi_gid,                       32)     \
 794         x(bi_nlink,                     32)     \
 795         x(bi_generation,                32)     \
 796         x(bi_dev,                       32)     \
 797         x(bi_data_checksum,             8)      \
 798         x(bi_compression,               8)      \
 799         x(bi_project,                   32)     \
 800         x(bi_background_compression,    8)      \
 801         x(bi_data_replicas,             8)      \
 802         x(bi_promote_target,            16)     \
 803         x(bi_foreground_target,         16)     \
 804         x(bi_background_target,         16)     \
 805         x(bi_erasure_code,              16)     \
 806         x(bi_fields_set,                16)     \
 807         x(bi_dir,                       64)     \
 808         x(bi_dir_offset,                64)     \
 809         x(bi_subvol,                    32)     \
 810         x(bi_parent_subvol,             32)     \
 811         x(bi_nocow,                     8)
 812
 813 /* subset of BCH_INODE_FIELDS */
 814 #define BCH_INODE_OPTS()                        \
 815         x(data_checksum,                8)      \
 816         x(compression,                  8)      \
 817         x(project,                      32)     \
 818         x(background_compression,       8)      \
 819         x(data_replicas,                8)      \
 820         x(promote_target,               16)     \
 821         x(foreground_target,            16)     \
 822         x(background_target,            16)     \
 823         x(erasure_code,                 16)     \
 824         x(nocow,                        8)
 825
 826 enum inode_opt_id {
 827 #define x(name, ...)                            \
 828         Inode_opt_##name,
 829         BCH_INODE_OPTS()
 830 #undef  x
 831         Inode_opt_nr,
 832 };
 833
 834 #define BCH_INODE_FLAGS()                       \
 835         x(sync,                         0)      \
 836         x(immutable,                    1)      \
 837         x(append,                       2)      \
 838         x(nodump,                       3)      \
 839         x(noatime,                      4)      \
 840         x(i_size_dirty,                 5)      \
 841         x(i_sectors_dirty,              6)      \
 842         x(unlinked,                     7)      \
 843         x(backptr_untrusted,            8)
 844
 845 /* bits 20+ reserved for packed fields below: */
 846
 847 enum bch_inode_flags {
 848 #define x(t, n) BCH_INODE_##t = 1U << n,
 849         BCH_INODE_FLAGS()
 850 #undef x
 851 };
 852
 853 enum __bch_inode_flags {
 854 #define x(t, n) __BCH_INODE_##t = n,
 855         BCH_INODE_FLAGS()
 856 #undef x
 857 };
 858
 859 LE32_BITMASK(INODE_STR_HASH,    struct bch_inode, bi_flags, 20, 24);
 860 LE32_BITMASK(INODE_NR_FIELDS,   struct bch_inode, bi_flags, 24, 31);
 861 LE32_BITMASK(INODE_NEW_VARINT,  struct bch_inode, bi_flags, 31, 32);
 862
 863 LE64_BITMASK(INODEv2_STR_HASH,  struct bch_inode_v2, bi_flags, 20, 24);
 864 LE64_BITMASK(INODEv2_NR_FIELDS, struct bch_inode_v2, bi_flags, 24, 31);
 865
 866 LE64_BITMASK(INODEv3_STR_HASH,  struct bch_inode_v3, bi_flags, 20, 24);
 867 LE64_BITMASK(INODEv3_NR_FIELDS, struct bch_inode_v3, bi_flags, 24, 31);
 868
 869 LE64_BITMASK(INODEv3_FIELDS_START,
 870                                 struct bch_inode_v3, bi_flags, 31, 36);
 871 LE64_BITMASK(INODEv3_MODE,      struct bch_inode_v3, bi_flags, 36, 52);
 872
 873 /* Dirents */
 874
 875 /*
 876  * Dirents (and xattrs) have to implement string lookups; since our b-tree
 877  * doesn't support arbitrary length strings for the key, we instead index by a
 878  * 64 bit hash (currently truncated sha1) of the string, stored in the offset
 879  * field of the key - using linear probing to resolve hash collisions. This also
 880  * provides us with the readdir cookie posix requires.
 881  *
 882  * Linear probing requires us to use whiteouts for deletions, in the event of a
 883  * collision:
 884  */
 885
 886 struct bch_dirent {
 887         struct bch_val          v;
 888
 889         /* Target inode number: */
 890         union {
 891         __le64                  d_inum;
 892         struct {                /* DT_SUBVOL */
 893         __le32                  d_child_subvol;
 894         __le32                  d_parent_subvol;
 895         };
 896         };
 897
 898         /*
 899          * Copy of mode bits 12-15 from the target inode - so userspace can get
 900          * the filetype without having to do a stat()
 901          */
 902         __u8                    d_type;
 903
 904         __u8                    d_name[];
 905 } __packed __aligned(8);
 906
 907 #define DT_SUBVOL       16
 908 #define BCH_DT_MAX      17
 909
 910 #define BCH_NAME_MAX    512
 911
 912 /* Xattrs */
 913
 914 #define KEY_TYPE_XATTR_INDEX_USER                       0
 915 #define KEY_TYPE_XATTR_INDEX_POSIX_ACL_ACCESS   1
 916 #define KEY_TYPE_XATTR_INDEX_POSIX_ACL_DEFAULT  2
 917 #define KEY_TYPE_XATTR_INDEX_TRUSTED                    3
 918 #define KEY_TYPE_XATTR_INDEX_SECURITY           4
 919
 920 struct bch_xattr {
 921         struct bch_val          v;
 922         __u8                    x_type;
 923         __u8                    x_name_len;
 924         __le16                  x_val_len;
 925         __u8                    x_name[];
 926 } __packed __aligned(8);
 927
 928 /* Bucket/allocation information: */
 929
 930 struct bch_alloc {
 931         struct bch_val          v;
 932         __u8                    fields;
 933         __u8                    gen;
 934         __u8                    data[];
 935 } __packed __aligned(8);
 936
 937 #define BCH_ALLOC_FIELDS_V1()                   \
 938         x(read_time,            16)             \
 939         x(write_time,           16)             \
 940         x(data_type,            8)              \
 941         x(dirty_sectors,        16)             \
 942         x(cached_sectors,       16)             \
 943         x(oldest_gen,           8)              \
 944         x(stripe,               32)             \
 945         x(stripe_redundancy,    8)
 946
 947 enum {
 948 #define x(name, _bits) BCH_ALLOC_FIELD_V1_##name,
 949         BCH_ALLOC_FIELDS_V1()
 950 #undef x
 951 };
 952
 953 struct bch_alloc_v2 {
 954         struct bch_val          v;
 955         __u8                    nr_fields;
 956         __u8                    gen;
 957         __u8                    oldest_gen;
 958         __u8                    data_type;
 959         __u8                    data[];
 960 } __packed __aligned(8);
 961
 962 #define BCH_ALLOC_FIELDS_V2()                   \
 963         x(read_time,            64)             \
 964         x(write_time,           64)             \
 965         x(dirty_sectors,        32)             \
 966         x(cached_sectors,       32)             \
 967         x(stripe,               32)             \
 968         x(stripe_redundancy,    8)
 969
 970 struct bch_alloc_v3 {
 971         struct bch_val          v;
 972         __le64                  journal_seq;
 973         __le32                  flags;
 974         __u8                    nr_fields;
 975         __u8                    gen;
 976         __u8                    oldest_gen;
 977         __u8                    data_type;
 978         __u8                    data[];
 979 } __packed __aligned(8);
 980
 981 LE32_BITMASK(BCH_ALLOC_V3_NEED_DISCARD,struct bch_alloc_v3, flags,  0,  1)
 982 LE32_BITMASK(BCH_ALLOC_V3_NEED_INC_GEN,struct bch_alloc_v3, flags,  1,  2)
 983
 984 struct bch_alloc_v4 {
 985         struct bch_val          v;
 986         __u64                   journal_seq;
 987         __u32                   flags;
 988         __u8                    gen;
 989         __u8                    oldest_gen;
 990         __u8                    data_type;
 991         __u8                    stripe_redundancy;
 992         __u32                   dirty_sectors;
 993         __u32                   cached_sectors;
 994         __u64                   io_time[2];
 995         __u32                   stripe;
 996         __u32                   nr_external_backpointers;
 997         __u64                   fragmentation_lru;
 998 } __packed __aligned(8);
 999
1000 #define BCH_ALLOC_V4_U64s_V0    6
1001 #define BCH_ALLOC_V4_U64s       (sizeof(struct bch_alloc_v4) / sizeof(__u64))
1002
1003 BITMASK(BCH_ALLOC_V4_NEED_DISCARD,      struct bch_alloc_v4, flags,  0,  1)
1004 BITMASK(BCH_ALLOC_V4_NEED_INC_GEN,      struct bch_alloc_v4, flags,  1,  2)
1005 BITMASK(BCH_ALLOC_V4_BACKPOINTERS_START,struct bch_alloc_v4, flags,  2,  8)
1006 BITMASK(BCH_ALLOC_V4_NR_BACKPOINTERS,   struct bch_alloc_v4, flags,  8,  14)
1007
1008 #define BCH_ALLOC_V4_NR_BACKPOINTERS_MAX        40
1009
1010 struct bch_backpointer {
1011         struct bch_val          v;
1012         __u8                    btree_id;
1013         __u8                    level;
1014         __u8                    data_type;
1015         __u64                   bucket_offset:40;
1016         __u32                   bucket_len;
1017         struct bpos             pos;
1018 } __packed __aligned(8);
1019
1020 #define KEY_TYPE_BUCKET_GENS_BITS       8
1021 #define KEY_TYPE_BUCKET_GENS_NR         (1U << KEY_TYPE_BUCKET_GENS_BITS)
1022 #define KEY_TYPE_BUCKET_GENS_MASK       (KEY_TYPE_BUCKET_GENS_NR - 1)
1023
1024 struct bch_bucket_gens {
1025         struct bch_val          v;
1026         u8                      gens[KEY_TYPE_BUCKET_GENS_NR];
1027 } __packed __aligned(8);
1028
1029 /* Quotas: */
1030
1031 enum quota_types {
1032         QTYP_USR                = 0,
1033         QTYP_GRP                = 1,
1034         QTYP_PRJ                = 2,
1035         QTYP_NR                 = 3,
1036 };
1037
1038 enum quota_counters {
1039         Q_SPC                   = 0,
1040         Q_INO                   = 1,
1041         Q_COUNTERS              = 2,
1042 };
1043
1044 struct bch_quota_counter {
1045         __le64                  hardlimit;
1046         __le64                  softlimit;
1047 };
1048
1049 struct bch_quota {
1050         struct bch_val          v;
1051         struct bch_quota_counter c[Q_COUNTERS];
1052 } __packed __aligned(8);
1053
1054 /* Erasure coding */
1055
1056 struct bch_stripe {
1057         struct bch_val          v;
1058         __le16                  sectors;
1059         __u8                    algorithm;
1060         __u8                    nr_blocks;
1061         __u8                    nr_redundant;
1062
1063         __u8                    csum_granularity_bits;
1064         __u8                    csum_type;
1065         __u8                    pad;
1066
1067         struct bch_extent_ptr   ptrs[];
1068 } __packed __aligned(8);
1069
1070 /* Reflink: */
1071
1072 struct bch_reflink_p {
1073         struct bch_val          v;
1074         __le64                  idx;
1075         /*
1076          * A reflink pointer might point to an indirect extent which is then
1077          * later split (by copygc or rebalance). If we only pointed to part of
1078          * the original indirect extent, and then one of the fragments is
1079          * outside the range we point to, we'd leak a refcount: so when creating
1080          * reflink pointers, we need to store pad values to remember the full
1081          * range we were taking a reference on.
1082          */
1083         __le32                  front_pad;
1084         __le32                  back_pad;
1085 } __packed __aligned(8);
1086
1087 struct bch_reflink_v {
1088         struct bch_val          v;
1089         __le64                  refcount;
1090         union bch_extent_entry  start[0];
1091         __u64                   _data[];
1092 } __packed __aligned(8);
1093
1094 struct bch_indirect_inline_data {
1095         struct bch_val          v;
1096         __le64                  refcount;
1097         u8                      data[];
1098 };
1099
1100 /* Inline data */
1101
1102 struct bch_inline_data {
1103         struct bch_val          v;
1104         u8                      data[];
1105 };
1106
1107 /* Subvolumes: */
1108
1109 #define SUBVOL_POS_MIN          POS(0, 1)
1110 #define SUBVOL_POS_MAX          POS(0, S32_MAX)
1111 #define BCACHEFS_ROOT_SUBVOL    1
1112
1113 struct bch_subvolume {
1114         struct bch_val          v;
1115         __le32                  flags;
1116         __le32                  snapshot;
1117         __le64                  inode;
1118         /*
1119          * Snapshot subvolumes form a tree, separate from the snapshot nodes
1120          * tree - if this subvolume is a snapshot, this is the ID of the
1121          * subvolume it was created from:
1122          */
1123         __le32                  parent;
1124         __le32                  pad;
1125         bch_le128               otime;
1126 };
1127
1128 LE32_BITMASK(BCH_SUBVOLUME_RO,          struct bch_subvolume, flags,  0,  1)
1129 /*
1130  * We need to know whether a subvolume is a snapshot so we can know whether we
1131  * can delete it (or whether it should just be rm -rf'd)
1132  */
1133 LE32_BITMASK(BCH_SUBVOLUME_SNAP,        struct bch_subvolume, flags,  1,  2)
1134 LE32_BITMASK(BCH_SUBVOLUME_UNLINKED,    struct bch_subvolume, flags,  2,  3)
1135
1136 /* Snapshots */
1137
1138 struct bch_snapshot {
1139         struct bch_val          v;
1140         __le32                  flags;
1141         __le32                  parent;
1142         __le32                  children[2];
1143         __le32                  subvol;
1144         /* corresponds to a bch_snapshot_tree in BTREE_ID_snapshot_trees */
1145         __le32                  tree;
1146         __le32                  depth;
1147         __le32                  skip[3];
1148 };
1149
1150 LE32_BITMASK(BCH_SNAPSHOT_DELETED,      struct bch_snapshot, flags,  0,  1)
1151
1152 /* True if a subvolume points to this snapshot node: */
1153 LE32_BITMASK(BCH_SNAPSHOT_SUBVOL,       struct bch_snapshot, flags,  1,  2)
1154
1155 /*
1156  * Snapshot trees:
1157  *
1158  * The snapshot_trees btree gives us persistent indentifier for each tree of
1159  * bch_snapshot nodes, and allow us to record and easily find the root/master
1160  * subvolume that other snapshots were created from:
1161  */
1162 struct bch_snapshot_tree {
1163         struct bch_val          v;
1164         __le32                  master_subvol;
1165         __le32                  root_snapshot;
1166 };
1167
1168 /* LRU btree: */
1169
1170 struct bch_lru {
1171         struct bch_val          v;
1172         __le64                  idx;
1173 } __packed __aligned(8);
1174
1175 #define LRU_ID_STRIPES          (1U << 16)
1176
1177 /* Logged operations btree: */
1178
1179 struct bch_logged_op_truncate {
1180         struct bch_val          v;
1181         __le32                  subvol;
1182         __le32                  pad;
1183         __le64                  inum;
1184         __le64                  new_i_size;
1185 };
1186
1187 enum logged_op_finsert_state {
1188         LOGGED_OP_FINSERT_start,
1189         LOGGED_OP_FINSERT_shift_extents,
1190         LOGGED_OP_FINSERT_finish,
1191 };
1192
1193 struct bch_logged_op_finsert {
1194         struct bch_val          v;
1195         __u8                    state;
1196         __u8                    pad[3];
1197         __le32                  subvol;
1198         __le64                  inum;
1199         __le64                  dst_offset;
1200         __le64                  src_offset;
1201         __le64                  pos;
1202 };
1203
1204 /* Optional/variable size superblock sections: */
1205
1206 struct bch_sb_field {
1207         __u64                   _data[0];
1208         __le32                  u64s;
1209         __le32                  type;
1210 };
1211
1212 #define BCH_SB_FIELDS()                         \
1213         x(journal,      0)                      \
1214         x(members_v1,   1)                      \
1215         x(crypt,        2)                      \
1216         x(replicas_v0,  3)                      \
1217         x(quota,        4)                      \
1218         x(disk_groups,  5)                      \
1219         x(clean,        6)                      \
1220         x(replicas,     7)                      \
1221         x(journal_seq_blacklist, 8)             \
1222         x(journal_v2,   9)                      \
1223         x(counters,     10)                     \
1224         x(members_v2,   11)                     \
1225         x(errors,       12)
1226
1227 enum bch_sb_field_type {
1228 #define x(f, nr)        BCH_SB_FIELD_##f = nr,
1229         BCH_SB_FIELDS()
1230 #undef x
1231         BCH_SB_FIELD_NR
1232 };
1233
1234 /*
1235  * Most superblock fields are replicated in all device's superblocks - a few are
1236  * not:
1237  */
1238 #define BCH_SINGLE_DEVICE_SB_FIELDS             \
1239         ((1U << BCH_SB_FIELD_journal)|          \
1240          (1U << BCH_SB_FIELD_journal_v2))
1241
1242 /* BCH_SB_FIELD_journal: */
1243
1244 struct bch_sb_field_journal {
1245         struct bch_sb_field     field;
1246         __le64                  buckets[];
1247 };
1248
1249 struct bch_sb_field_journal_v2 {
1250         struct bch_sb_field     field;
1251
1252         struct bch_sb_field_journal_v2_entry {
1253                 __le64          start;
1254                 __le64          nr;
1255         }                       d[];
1256 };
1257
1258 /* BCH_SB_FIELD_members_v1: */
1259
1260 #define BCH_MIN_NR_NBUCKETS     (1 << 6)
1261
1262 #define BCH_IOPS_MEASUREMENTS()                 \
1263         x(seqread,      0)                      \
1264         x(seqwrite,     1)                      \
1265         x(randread,     2)                      \
1266         x(randwrite,    3)
1267
1268 enum bch_iops_measurement {
1269 #define x(t, n) BCH_IOPS_##t = n,
1270         BCH_IOPS_MEASUREMENTS()
1271 #undef x
1272         BCH_IOPS_NR
1273 };
1274
1275 #define BCH_MEMBER_ERROR_TYPES()                \
1276         x(read,         0)                      \
1277         x(write,        1)                      \
1278         x(checksum,     2)
1279
1280 enum bch_member_error_type {
1281 #define x(t, n) BCH_MEMBER_ERROR_##t = n,
1282         BCH_MEMBER_ERROR_TYPES()
1283 #undef x
1284         BCH_MEMBER_ERROR_NR
1285 };
1286
1287 struct bch_member {
1288         __uuid_t                uuid;
1289         __le64                  nbuckets;       /* device size */
1290         __le16                  first_bucket;   /* index of first bucket used */
1291         __le16                  bucket_size;    /* sectors */
1292         __le32                  pad;
1293         __le64                  last_mount;     /* time_t */
1294
1295         __le64                  flags;
1296         __le32                  iops[4];
1297         __le64                  errors[BCH_MEMBER_ERROR_NR];
1298         __le64                  errors_at_reset[BCH_MEMBER_ERROR_NR];
1299         __le64                  errors_reset_time;
1300 };
1301
1302 #define BCH_MEMBER_V1_BYTES     56
1303
1304 LE64_BITMASK(BCH_MEMBER_STATE,          struct bch_member, flags,  0,  4)
1305 /* 4-14 unused, was TIER, HAS_(META)DATA, REPLACEMENT */
1306 LE64_BITMASK(BCH_MEMBER_DISCARD,        struct bch_member, flags, 14, 15)
1307 LE64_BITMASK(BCH_MEMBER_DATA_ALLOWED,   struct bch_member, flags, 15, 20)
1308 LE64_BITMASK(BCH_MEMBER_GROUP,          struct bch_member, flags, 20, 28)
1309 LE64_BITMASK(BCH_MEMBER_DURABILITY,     struct bch_member, flags, 28, 30)
1310 LE64_BITMASK(BCH_MEMBER_FREESPACE_INITIALIZED,
1311                                         struct bch_member, flags, 30, 31)
1312
1313 #if 0
1314 LE64_BITMASK(BCH_MEMBER_NR_READ_ERRORS, struct bch_member, flags[1], 0,  20);
1315 LE64_BITMASK(BCH_MEMBER_NR_WRITE_ERRORS,struct bch_member, flags[1], 20, 40);
1316 #endif
1317
1318 #define BCH_MEMBER_STATES()                     \
1319         x(rw,           0)                      \
1320         x(ro,           1)                      \
1321         x(failed,       2)                      \
1322         x(spare,        3)
1323
1324 enum bch_member_state {
1325 #define x(t, n) BCH_MEMBER_STATE_##t = n,
1326         BCH_MEMBER_STATES()
1327 #undef x
1328         BCH_MEMBER_STATE_NR
1329 };
1330
1331 struct bch_sb_field_members_v1 {
1332         struct bch_sb_field     field;
1333         struct bch_member       _members[]; //Members are now variable size
1334 };
1335
1336 struct bch_sb_field_members_v2 {
1337         struct bch_sb_field     field;
1338         __le16                  member_bytes; //size of single member entry
1339         u8                      pad[6];
1340         struct bch_member       _members[];
1341 };
1342
1343 /* BCH_SB_FIELD_crypt: */
1344
1345 struct nonce {
1346         __le32                  d[4];
1347 };
1348
1349 struct bch_key {
1350         __le64                  key[4];
1351 };
1352
1353 #define BCH_KEY_MAGIC                                   \
1354         (((__u64) 'b' <<  0)|((__u64) 'c' <<  8)|               \
1355          ((__u64) 'h' << 16)|((__u64) '*' << 24)|               \
1356          ((__u64) '*' << 32)|((__u64) 'k' << 40)|               \
1357          ((__u64) 'e' << 48)|((__u64) 'y' << 56))
1358
1359 struct bch_encrypted_key {
1360         __le64                  magic;
1361         struct bch_key          key;
1362 };
1363
1364 /*
1365  * If this field is present in the superblock, it stores an encryption key which
1366  * is used encrypt all other data/metadata. The key will normally be encrypted
1367  * with the key userspace provides, but if encryption has been turned off we'll
1368  * just store the master key unencrypted in the superblock so we can access the
1369  * previously encrypted data.
1370  */
1371 struct bch_sb_field_crypt {
1372         struct bch_sb_field     field;
1373
1374         __le64                  flags;
1375         __le64                  kdf_flags;
1376         struct bch_encrypted_key key;
1377 };
1378
1379 LE64_BITMASK(BCH_CRYPT_KDF_TYPE,        struct bch_sb_field_crypt, flags, 0, 4);
1380
1381 enum bch_kdf_types {
1382         BCH_KDF_SCRYPT          = 0,
1383         BCH_KDF_NR              = 1,
1384 };
1385
1386 /* stored as base 2 log of scrypt params: */
1387 LE64_BITMASK(BCH_KDF_SCRYPT_N,  struct bch_sb_field_crypt, kdf_flags,  0, 16);
1388 LE64_BITMASK(BCH_KDF_SCRYPT_R,  struct bch_sb_field_crypt, kdf_flags, 16, 32);
1389 LE64_BITMASK(BCH_KDF_SCRYPT_P,  struct bch_sb_field_crypt, kdf_flags, 32, 48);
1390
1391 /* BCH_SB_FIELD_replicas: */
1392
1393 #define BCH_DATA_TYPES()                \
1394         x(free,         0)              \
1395         x(sb,           1)              \
1396         x(journal,      2)              \
1397         x(btree,        3)              \
1398         x(user,         4)              \
1399         x(cached,       5)              \
1400         x(parity,       6)              \
1401         x(stripe,       7)              \
1402         x(need_gc_gens, 8)              \
1403         x(need_discard, 9)
1404
1405 enum bch_data_type {
1406 #define x(t, n) BCH_DATA_##t,
1407         BCH_DATA_TYPES()
1408 #undef x
1409         BCH_DATA_NR
1410 };
1411
1412 static inline bool data_type_is_empty(enum bch_data_type type)
1413 {
1414         switch (type) {
1415         case BCH_DATA_free:
1416         case BCH_DATA_need_gc_gens:
1417         case BCH_DATA_need_discard:
1418                 return true;
1419         default:
1420                 return false;
1421         }
1422 }
1423
1424 static inline bool data_type_is_hidden(enum bch_data_type type)
1425 {
1426         switch (type) {
1427         case BCH_DATA_sb:
1428         case BCH_DATA_journal:
1429                 return true;
1430         default:
1431                 return false;
1432         }
1433 }
1434
1435 struct bch_replicas_entry_v0 {
1436         __u8                    data_type;
1437         __u8                    nr_devs;
1438         __u8                    devs[];
1439 } __packed;
1440
1441 struct bch_sb_field_replicas_v0 {
1442         struct bch_sb_field     field;
1443         struct bch_replicas_entry_v0 entries[];
1444 } __packed __aligned(8);
1445
1446 struct bch_replicas_entry_v1 {
1447         __u8                    data_type;
1448         __u8                    nr_devs;
1449         __u8                    nr_required;
1450         __u8                    devs[];
1451 } __packed;
1452
1453 #define replicas_entry_bytes(_i)                                        \
1454         (offsetof(typeof(*(_i)), devs) + (_i)->nr_devs)
1455
1456 struct bch_sb_field_replicas {
1457         struct bch_sb_field     field;
1458         struct bch_replicas_entry_v1 entries[];
1459 } __packed __aligned(8);
1460
1461 /* BCH_SB_FIELD_quota: */
1462
1463 struct bch_sb_quota_counter {
1464         __le32                          timelimit;
1465         __le32                          warnlimit;
1466 };
1467
1468 struct bch_sb_quota_type {
1469         __le64                          flags;
1470         struct bch_sb_quota_counter     c[Q_COUNTERS];
1471 };
1472
1473 struct bch_sb_field_quota {
1474         struct bch_sb_field             field;
1475         struct bch_sb_quota_type        q[QTYP_NR];
1476 } __packed __aligned(8);
1477
1478 /* BCH_SB_FIELD_disk_groups: */
1479
1480 #define BCH_SB_LABEL_SIZE               32
1481
1482 struct bch_disk_group {
1483         __u8                    label[BCH_SB_LABEL_SIZE];
1484         __le64                  flags[2];
1485 } __packed __aligned(8);
1486
1487 LE64_BITMASK(BCH_GROUP_DELETED,         struct bch_disk_group, flags[0], 0,  1)
1488 LE64_BITMASK(BCH_GROUP_DATA_ALLOWED,    struct bch_disk_group, flags[0], 1,  6)
1489 LE64_BITMASK(BCH_GROUP_PARENT,          struct bch_disk_group, flags[0], 6, 24)
1490
1491 struct bch_sb_field_disk_groups {
1492         struct bch_sb_field     field;
1493         struct bch_disk_group   entries[];
1494 } __packed __aligned(8);
1495
1496 /* BCH_SB_FIELD_counters */
1497
1498 #define BCH_PERSISTENT_COUNTERS()                               \
1499         x(io_read,                                      0)      \
1500         x(io_write,                                     1)      \
1501         x(io_move,                                      2)      \
1502         x(bucket_invalidate,                            3)      \
1503         x(bucket_discard,                               4)      \
1504         x(bucket_alloc,                                 5)      \
1505         x(bucket_alloc_fail,                            6)      \
1506         x(btree_cache_scan,                             7)      \
1507         x(btree_cache_reap,                             8)      \
1508         x(btree_cache_cannibalize,                      9)      \
1509         x(btree_cache_cannibalize_lock,                 10)     \
1510         x(btree_cache_cannibalize_lock_fail,            11)     \
1511         x(btree_cache_cannibalize_unlock,               12)     \
1512         x(btree_node_write,                             13)     \
1513         x(btree_node_read,                              14)     \
1514         x(btree_node_compact,                           15)     \
1515         x(btree_node_merge,                             16)     \
1516         x(btree_node_split,                             17)     \
1517         x(btree_node_rewrite,                           18)     \
1518         x(btree_node_alloc,                             19)     \
1519         x(btree_node_free,                              20)     \
1520         x(btree_node_set_root,                          21)     \
1521         x(btree_path_relock_fail,                       22)     \
1522         x(btree_path_upgrade_fail,                      23)     \
1523         x(btree_reserve_get_fail,                       24)     \
1524         x(journal_entry_full,                           25)     \
1525         x(journal_full,                                 26)     \
1526         x(journal_reclaim_finish,                       27)     \
1527         x(journal_reclaim_start,                        28)     \
1528         x(journal_write,                                29)     \
1529         x(read_promote,                                 30)     \
1530         x(read_bounce,                                  31)     \
1531         x(read_split,                                   33)     \
1532         x(read_retry,                                   32)     \
1533         x(read_reuse_race,                              34)     \
1534         x(move_extent_read,                             35)     \
1535         x(move_extent_write,                            36)     \
1536         x(move_extent_finish,                           37)     \
1537         x(move_extent_fail,                             38)     \
1538         x(move_extent_alloc_mem_fail,                   39)     \
1539         x(copygc,                                       40)     \
1540         x(copygc_wait,                                  41)     \
1541         x(gc_gens_end,                                  42)     \
1542         x(gc_gens_start,                                43)     \
1543         x(trans_blocked_journal_reclaim,                44)     \
1544         x(trans_restart_btree_node_reused,              45)     \
1545         x(trans_restart_btree_node_split,               46)     \
1546         x(trans_restart_fault_inject,                   47)     \
1547         x(trans_restart_iter_upgrade,                   48)     \
1548         x(trans_restart_journal_preres_get,             49)     \
1549         x(trans_restart_journal_reclaim,                50)     \
1550         x(trans_restart_journal_res_get,                51)     \
1551         x(trans_restart_key_cache_key_realloced,        52)     \
1552         x(trans_restart_key_cache_raced,                53)     \
1553         x(trans_restart_mark_replicas,                  54)     \
1554         x(trans_restart_mem_realloced,                  55)     \
1555         x(trans_restart_memory_allocation_failure,      56)     \
1556         x(trans_restart_relock,                         57)     \
1557         x(trans_restart_relock_after_fill,              58)     \
1558         x(trans_restart_relock_key_cache_fill,          59)     \
1559         x(trans_restart_relock_next_node,               60)     \
1560         x(trans_restart_relock_parent_for_fill,         61)     \
1561         x(trans_restart_relock_path,                    62)     \
1562         x(trans_restart_relock_path_intent,             63)     \
1563         x(trans_restart_too_many_iters,                 64)     \
1564         x(trans_restart_traverse,                       65)     \
1565         x(trans_restart_upgrade,                        66)     \
1566         x(trans_restart_would_deadlock,                 67)     \
1567         x(trans_restart_would_deadlock_write,           68)     \
1568         x(trans_restart_injected,                       69)     \
1569         x(trans_restart_key_cache_upgrade,              70)     \
1570         x(trans_traverse_all,                           71)     \
1571         x(transaction_commit,                           72)     \
1572         x(write_super,                                  73)     \
1573         x(trans_restart_would_deadlock_recursion_limit, 74)     \
1574         x(trans_restart_write_buffer_flush,             75)     \
1575         x(trans_restart_split_race,                     76)
1576
1577 enum bch_persistent_counters {
1578 #define x(t, n, ...) BCH_COUNTER_##t,
1579         BCH_PERSISTENT_COUNTERS()
1580 #undef x
1581         BCH_COUNTER_NR
1582 };
1583
1584 struct bch_sb_field_counters {
1585         struct bch_sb_field     field;
1586         __le64                  d[];
1587 };
1588
1589 /*
1590  * On clean shutdown, store btree roots and current journal sequence number in
1591  * the superblock:
1592  */
1593 struct jset_entry {
1594         __le16                  u64s;
1595         __u8                    btree_id;
1596         __u8                    level;
1597         __u8                    type; /* designates what this jset holds */
1598         __u8                    pad[3];
1599
1600         struct bkey_i           start[0];
1601         __u64                   _data[];
1602 };
1603
1604 struct bch_sb_field_clean {
1605         struct bch_sb_field     field;
1606
1607         __le32                  flags;
1608         __le16                  _read_clock; /* no longer used */
1609         __le16                  _write_clock;
1610         __le64                  journal_seq;
1611
1612         struct jset_entry       start[0];
1613         __u64                   _data[];
1614 };
1615
1616 struct journal_seq_blacklist_entry {
1617         __le64                  start;
1618         __le64                  end;
1619 };
1620
1621 struct bch_sb_field_journal_seq_blacklist {
1622         struct bch_sb_field     field;
1623         struct journal_seq_blacklist_entry start[];
1624 };
1625
1626 struct bch_sb_field_errors {
1627         struct bch_sb_field     field;
1628         struct bch_sb_field_error_entry {
1629                 __le64          v;
1630                 __le64          last_error_time;
1631         }                       entries[];
1632 };
1633
1634 LE64_BITMASK(BCH_SB_ERROR_ENTRY_ID,     struct bch_sb_field_error_entry, v,  0, 16);
1635 LE64_BITMASK(BCH_SB_ERROR_ENTRY_NR,     struct bch_sb_field_error_entry, v, 16, 64);
1636
1637 /* Superblock: */
1638
1639 /*
1640  * New versioning scheme:
1641  * One common version number for all on disk data structures - superblock, btree
1642  * nodes, journal entries
1643  */
1644 #define BCH_VERSION_MAJOR(_v)           ((__u16) ((_v) >> 10))
1645 #define BCH_VERSION_MINOR(_v)           ((__u16) ((_v) & ~(~0U << 10)))
1646 #define BCH_VERSION(_major, _minor)     (((_major) << 10)|(_minor) << 0)
1647
1648 #define RECOVERY_PASS_ALL_FSCK          (1ULL << 63)
1649
1650 #define BCH_METADATA_VERSIONS()                                         \
1651         x(bkey_renumber,                BCH_VERSION(0, 10),             \
1652           RECOVERY_PASS_ALL_FSCK)                                       \
1653         x(inode_btree_change,           BCH_VERSION(0, 11),             \
1654           RECOVERY_PASS_ALL_FSCK)                                       \
1655         x(snapshot,                     BCH_VERSION(0, 12),             \
1656           RECOVERY_PASS_ALL_FSCK)                                       \
1657         x(inode_backpointers,           BCH_VERSION(0, 13),             \
1658           RECOVERY_PASS_ALL_FSCK)                                       \
1659         x(btree_ptr_sectors_written,    BCH_VERSION(0, 14),             \
1660           RECOVERY_PASS_ALL_FSCK)                                       \
1661         x(snapshot_2,                   BCH_VERSION(0, 15),             \
1662           BIT_ULL(BCH_RECOVERY_PASS_fs_upgrade_for_subvolumes)|         \
1663           BIT_ULL(BCH_RECOVERY_PASS_initialize_subvolumes)|             \
1664           RECOVERY_PASS_ALL_FSCK)                                       \
1665         x(reflink_p_fix,                BCH_VERSION(0, 16),             \
1666           BIT_ULL(BCH_RECOVERY_PASS_fix_reflink_p))                     \
1667         x(subvol_dirent,                BCH_VERSION(0, 17),             \
1668           RECOVERY_PASS_ALL_FSCK)                                       \
1669         x(inode_v2,                     BCH_VERSION(0, 18),             \
1670           RECOVERY_PASS_ALL_FSCK)                                       \
1671         x(freespace,                    BCH_VERSION(0, 19),             \
1672           RECOVERY_PASS_ALL_FSCK)                                       \
1673         x(alloc_v4,                     BCH_VERSION(0, 20),             \
1674           RECOVERY_PASS_ALL_FSCK)                                       \
1675         x(new_data_types,               BCH_VERSION(0, 21),             \
1676           RECOVERY_PASS_ALL_FSCK)                                       \
1677         x(backpointers,                 BCH_VERSION(0, 22),             \
1678           RECOVERY_PASS_ALL_FSCK)                                       \
1679         x(inode_v3,                     BCH_VERSION(0, 23),             \
1680           RECOVERY_PASS_ALL_FSCK)                                       \
1681         x(unwritten_extents,            BCH_VERSION(0, 24),             \
1682           RECOVERY_PASS_ALL_FSCK)                                       \
1683         x(bucket_gens,                  BCH_VERSION(0, 25),             \
1684           BIT_ULL(BCH_RECOVERY_PASS_bucket_gens_init)|                  \
1685           RECOVERY_PASS_ALL_FSCK)                                       \
1686         x(lru_v2,                       BCH_VERSION(0, 26),             \
1687           RECOVERY_PASS_ALL_FSCK)                                       \
1688         x(fragmentation_lru,            BCH_VERSION(0, 27),             \
1689           RECOVERY_PASS_ALL_FSCK)                                       \
1690         x(no_bps_in_alloc_keys,         BCH_VERSION(0, 28),             \
1691           RECOVERY_PASS_ALL_FSCK)                                       \
1692         x(snapshot_trees,               BCH_VERSION(0, 29),             \
1693           RECOVERY_PASS_ALL_FSCK)                                       \
1694         x(major_minor,                  BCH_VERSION(1,  0),             \
1695           0)                                                            \
1696         x(snapshot_skiplists,           BCH_VERSION(1,  1),             \
1697           BIT_ULL(BCH_RECOVERY_PASS_check_snapshots))                   \
1698         x(deleted_inodes,               BCH_VERSION(1,  2),             \
1699           BIT_ULL(BCH_RECOVERY_PASS_check_inodes))                      \
1700         x(rebalance_work,               BCH_VERSION(1,  3),             \
1701           BIT_ULL(BCH_RECOVERY_PASS_set_fs_needs_rebalance))
1702
1703 enum bcachefs_metadata_version {
1704         bcachefs_metadata_version_min = 9,
1705 #define x(t, n, upgrade_passes) bcachefs_metadata_version_##t = n,
1706         BCH_METADATA_VERSIONS()
1707 #undef x
1708         bcachefs_metadata_version_max
1709 };
1710
1711 static const __maybe_unused
1712 unsigned bcachefs_metadata_required_upgrade_below = bcachefs_metadata_version_rebalance_work;
1713
1714 #define bcachefs_metadata_version_current       (bcachefs_metadata_version_max - 1)
1715
1716 #define BCH_SB_SECTOR                   8
1717 #define BCH_SB_MEMBERS_MAX              64 /* XXX kill */
1718
1719 struct bch_sb_layout {
1720         __uuid_t                magic;  /* bcachefs superblock UUID */
1721         __u8                    layout_type;
1722         __u8                    sb_max_size_bits; /* base 2 of 512 byte sectors */
1723         __u8                    nr_superblocks;
1724         __u8                    pad[5];
1725         __le64                  sb_offset[61];
1726 } __packed __aligned(8);
1727
1728 #define BCH_SB_LAYOUT_SECTOR    7
1729
1730 /*
1731  * @offset      - sector where this sb was written
1732  * @version     - on disk format version
1733  * @version_min - Oldest metadata version this filesystem contains; so we can
1734  *                safely drop compatibility code and refuse to mount filesystems
1735  *                we'd need it for
1736  * @magic       - identifies as a bcachefs superblock (BCHFS_MAGIC)
1737  * @seq         - incremented each time superblock is written
1738  * @uuid        - used for generating various magic numbers and identifying
1739  *                member devices, never changes
1740  * @user_uuid   - user visible UUID, may be changed
1741  * @label       - filesystem label
1742  * @seq         - identifies most recent superblock, incremented each time
1743  *                superblock is written
1744  * @features    - enabled incompatible features
1745  */
1746 struct bch_sb {
1747         struct bch_csum         csum;
1748         __le16                  version;
1749         __le16                  version_min;
1750         __le16                  pad[2];
1751         __uuid_t                magic;
1752         __uuid_t                uuid;
1753         __uuid_t                user_uuid;
1754         __u8                    label[BCH_SB_LABEL_SIZE];
1755         __le64                  offset;
1756         __le64                  seq;
1757
1758         __le16                  block_size;
1759         __u8                    dev_idx;
1760         __u8                    nr_devices;
1761         __le32                  u64s;
1762
1763         __le64                  time_base_lo;
1764         __le32                  time_base_hi;
1765         __le32                  time_precision;
1766
1767         __le64                  flags[8];
1768         __le64                  features[2];
1769         __le64                  compat[2];
1770
1771         struct bch_sb_layout    layout;
1772
1773         struct bch_sb_field     start[0];
1774         __le64                  _data[];
1775 } __packed __aligned(8);
1776
1777 /*
1778  * Flags:
1779  * BCH_SB_INITALIZED    - set on first mount
1780  * BCH_SB_CLEAN         - did we shut down cleanly? Just a hint, doesn't affect
1781  *                        behaviour of mount/recovery path:
1782  * BCH_SB_INODE_32BIT   - limit inode numbers to 32 bits
1783  * BCH_SB_128_BIT_MACS  - 128 bit macs instead of 80
1784  * BCH_SB_ENCRYPTION_TYPE - if nonzero encryption is enabled; overrides
1785  *                         DATA/META_CSUM_TYPE. Also indicates encryption
1786  *                         algorithm in use, if/when we get more than one
1787  */
1788
1789 LE16_BITMASK(BCH_SB_BLOCK_SIZE,         struct bch_sb, block_size, 0, 16);
1790
1791 LE64_BITMASK(BCH_SB_INITIALIZED,        struct bch_sb, flags[0],  0,  1);
1792 LE64_BITMASK(BCH_SB_CLEAN,              struct bch_sb, flags[0],  1,  2);
1793 LE64_BITMASK(BCH_SB_CSUM_TYPE,          struct bch_sb, flags[0],  2,  8);
1794 LE64_BITMASK(BCH_SB_ERROR_ACTION,       struct bch_sb, flags[0],  8, 12);
1795
1796 LE64_BITMASK(BCH_SB_BTREE_NODE_SIZE,    struct bch_sb, flags[0], 12, 28);
1797
1798 LE64_BITMASK(BCH_SB_GC_RESERVE,         struct bch_sb, flags[0], 28, 33);
1799 LE64_BITMASK(BCH_SB_ROOT_RESERVE,       struct bch_sb, flags[0], 33, 40);
1800
1801 LE64_BITMASK(BCH_SB_META_CSUM_TYPE,     struct bch_sb, flags[0], 40, 44);
1802 LE64_BITMASK(BCH_SB_DATA_CSUM_TYPE,     struct bch_sb, flags[0], 44, 48);
1803
1804 LE64_BITMASK(BCH_SB_META_REPLICAS_WANT, struct bch_sb, flags[0], 48, 52);
1805 LE64_BITMASK(BCH_SB_DATA_REPLICAS_WANT, struct bch_sb, flags[0], 52, 56);
1806
1807 LE64_BITMASK(BCH_SB_POSIX_ACL,          struct bch_sb, flags[0], 56, 57);
1808 LE64_BITMASK(BCH_SB_USRQUOTA,           struct bch_sb, flags[0], 57, 58);
1809 LE64_BITMASK(BCH_SB_GRPQUOTA,           struct bch_sb, flags[0], 58, 59);
1810 LE64_BITMASK(BCH_SB_PRJQUOTA,           struct bch_sb, flags[0], 59, 60);
1811
1812 LE64_BITMASK(BCH_SB_HAS_ERRORS,         struct bch_sb, flags[0], 60, 61);
1813 LE64_BITMASK(BCH_SB_HAS_TOPOLOGY_ERRORS,struct bch_sb, flags[0], 61, 62);
1814
1815 LE64_BITMASK(BCH_SB_BIG_ENDIAN,         struct bch_sb, flags[0], 62, 63);
1816
1817 LE64_BITMASK(BCH_SB_STR_HASH_TYPE,      struct bch_sb, flags[1],  0,  4);
1818 LE64_BITMASK(BCH_SB_COMPRESSION_TYPE_LO,struct bch_sb, flags[1],  4,  8);
1819 LE64_BITMASK(BCH_SB_INODE_32BIT,        struct bch_sb, flags[1],  8,  9);
1820
1821 LE64_BITMASK(BCH_SB_128_BIT_MACS,       struct bch_sb, flags[1],  9, 10);
1822 LE64_BITMASK(BCH_SB_ENCRYPTION_TYPE,    struct bch_sb, flags[1], 10, 14);
1823
1824 /*
1825  * Max size of an extent that may require bouncing to read or write
1826  * (checksummed, compressed): 64k
1827  */
1828 LE64_BITMASK(BCH_SB_ENCODED_EXTENT_MAX_BITS,
1829                                         struct bch_sb, flags[1], 14, 20);
1830
1831 LE64_BITMASK(BCH_SB_META_REPLICAS_REQ,  struct bch_sb, flags[1], 20, 24);
1832 LE64_BITMASK(BCH_SB_DATA_REPLICAS_REQ,  struct bch_sb, flags[1], 24, 28);
1833
1834 LE64_BITMASK(BCH_SB_PROMOTE_TARGET,     struct bch_sb, flags[1], 28, 40);
1835 LE64_BITMASK(BCH_SB_FOREGROUND_TARGET,  struct bch_sb, flags[1], 40, 52);
1836 LE64_BITMASK(BCH_SB_BACKGROUND_TARGET,  struct bch_sb, flags[1], 52, 64);
1837
1838 LE64_BITMASK(BCH_SB_BACKGROUND_COMPRESSION_TYPE_LO,
1839                                         struct bch_sb, flags[2],  0,  4);
1840 LE64_BITMASK(BCH_SB_GC_RESERVE_BYTES,   struct bch_sb, flags[2],  4, 64);
1841
1842 LE64_BITMASK(BCH_SB_ERASURE_CODE,       struct bch_sb, flags[3],  0, 16);
1843 LE64_BITMASK(BCH_SB_METADATA_TARGET,    struct bch_sb, flags[3], 16, 28);
1844 LE64_BITMASK(BCH_SB_SHARD_INUMS,        struct bch_sb, flags[3], 28, 29);
1845 LE64_BITMASK(BCH_SB_INODES_USE_KEY_CACHE,struct bch_sb, flags[3], 29, 30);
1846 LE64_BITMASK(BCH_SB_JOURNAL_FLUSH_DELAY,struct bch_sb, flags[3], 30, 62);
1847 LE64_BITMASK(BCH_SB_JOURNAL_FLUSH_DISABLED,struct bch_sb, flags[3], 62, 63);
1848 LE64_BITMASK(BCH_SB_JOURNAL_RECLAIM_DELAY,struct bch_sb, flags[4], 0, 32);
1849 LE64_BITMASK(BCH_SB_JOURNAL_TRANSACTION_NAMES,struct bch_sb, flags[4], 32, 33);
1850 LE64_BITMASK(BCH_SB_NOCOW,              struct bch_sb, flags[4], 33, 34);
1851 LE64_BITMASK(BCH_SB_WRITE_BUFFER_SIZE,  struct bch_sb, flags[4], 34, 54);
1852 LE64_BITMASK(BCH_SB_VERSION_UPGRADE,    struct bch_sb, flags[4], 54, 56);
1853
1854 LE64_BITMASK(BCH_SB_COMPRESSION_TYPE_HI,struct bch_sb, flags[4], 56, 60);
1855 LE64_BITMASK(BCH_SB_BACKGROUND_COMPRESSION_TYPE_HI,
1856                                         struct bch_sb, flags[4], 60, 64);
1857
1858 LE64_BITMASK(BCH_SB_VERSION_UPGRADE_COMPLETE,
1859                                         struct bch_sb, flags[5],  0, 16);
1860
1861 static inline __u64 BCH_SB_COMPRESSION_TYPE(const struct bch_sb *sb)
1862 {
1863         return BCH_SB_COMPRESSION_TYPE_LO(sb) | (BCH_SB_COMPRESSION_TYPE_HI(sb) << 4);
1864 }
1865
1866 static inline void SET_BCH_SB_COMPRESSION_TYPE(struct bch_sb *sb, __u64 v)
1867 {
1868         SET_BCH_SB_COMPRESSION_TYPE_LO(sb, v);
1869         SET_BCH_SB_COMPRESSION_TYPE_HI(sb, v >> 4);
1870 }
1871
1872 static inline __u64 BCH_SB_BACKGROUND_COMPRESSION_TYPE(const struct bch_sb *sb)
1873 {
1874         return BCH_SB_BACKGROUND_COMPRESSION_TYPE_LO(sb) |
1875                 (BCH_SB_BACKGROUND_COMPRESSION_TYPE_HI(sb) << 4);
1876 }
1877
1878 static inline void SET_BCH_SB_BACKGROUND_COMPRESSION_TYPE(struct bch_sb *sb, __u64 v)
1879 {
1880         SET_BCH_SB_BACKGROUND_COMPRESSION_TYPE_LO(sb, v);
1881         SET_BCH_SB_BACKGROUND_COMPRESSION_TYPE_HI(sb, v >> 4);
1882 }
1883
1884 /*
1885  * Features:
1886  *
1887  * journal_seq_blacklist_v3:    gates BCH_SB_FIELD_journal_seq_blacklist
1888  * reflink:                     gates KEY_TYPE_reflink
1889  * inline_data:                 gates KEY_TYPE_inline_data
1890  * new_siphash:                 gates BCH_STR_HASH_siphash
1891  * new_extent_overwrite:        gates BTREE_NODE_NEW_EXTENT_OVERWRITE
1892  */
1893 #define BCH_SB_FEATURES()                       \
1894         x(lz4,                          0)      \
1895         x(gzip,                         1)      \
1896         x(zstd,                         2)      \
1897         x(atomic_nlink,                 3)      \
1898         x(ec,                           4)      \
1899         x(journal_seq_blacklist_v3,     5)      \
1900         x(reflink,                      6)      \
1901         x(new_siphash,                  7)      \
1902         x(inline_data,                  8)      \
1903         x(new_extent_overwrite,         9)      \
1904         x(incompressible,               10)     \
1905         x(btree_ptr_v2,                 11)     \
1906         x(extents_above_btree_updates,  12)     \
1907         x(btree_updates_journalled,     13)     \
1908         x(reflink_inline_data,          14)     \
1909         x(new_varint,                   15)     \
1910         x(journal_no_flush,             16)     \
1911         x(alloc_v2,                     17)     \
1912         x(extents_across_btree_nodes,   18)
1913
1914 #define BCH_SB_FEATURES_ALWAYS                          \
1915         ((1ULL << BCH_FEATURE_new_extent_overwrite)|    \
1916          (1ULL << BCH_FEATURE_extents_above_btree_updates)|\
1917          (1ULL << BCH_FEATURE_btree_updates_journalled)|\
1918          (1ULL << BCH_FEATURE_alloc_v2)|\
1919          (1ULL << BCH_FEATURE_extents_across_btree_nodes))
1920
1921 #define BCH_SB_FEATURES_ALL                             \
1922         (BCH_SB_FEATURES_ALWAYS|                        \
1923          (1ULL << BCH_FEATURE_new_siphash)|             \
1924          (1ULL << BCH_FEATURE_btree_ptr_v2)|            \
1925          (1ULL << BCH_FEATURE_new_varint)|              \
1926          (1ULL << BCH_FEATURE_journal_no_flush))
1927
1928 enum bch_sb_feature {
1929 #define x(f, n) BCH_FEATURE_##f,
1930         BCH_SB_FEATURES()
1931 #undef x
1932         BCH_FEATURE_NR,
1933 };
1934
1935 #define BCH_SB_COMPAT()                                 \
1936         x(alloc_info,                           0)      \
1937         x(alloc_metadata,                       1)      \
1938         x(extents_above_btree_updates_done,     2)      \
1939         x(bformat_overflow_done,                3)
1940
1941 enum bch_sb_compat {
1942 #define x(f, n) BCH_COMPAT_##f,
1943         BCH_SB_COMPAT()
1944 #undef x
1945         BCH_COMPAT_NR,
1946 };
1947
1948 /* options: */
1949
1950 #define BCH_VERSION_UPGRADE_OPTS()      \
1951         x(compatible,           0)      \
1952         x(incompatible,         1)      \
1953         x(none,                 2)
1954
1955 enum bch_version_upgrade_opts {
1956 #define x(t, n) BCH_VERSION_UPGRADE_##t = n,
1957         BCH_VERSION_UPGRADE_OPTS()
1958 #undef x
1959 };
1960
1961 #define BCH_REPLICAS_MAX                4U
1962
1963 #define BCH_BKEY_PTRS_MAX               16U
1964
1965 #define BCH_ERROR_ACTIONS()             \
1966         x(continue,             0)      \
1967         x(ro,                   1)      \
1968         x(panic,                2)
1969
1970 enum bch_error_actions {
1971 #define x(t, n) BCH_ON_ERROR_##t = n,
1972         BCH_ERROR_ACTIONS()
1973 #undef x
1974         BCH_ON_ERROR_NR
1975 };
1976
1977 #define BCH_STR_HASH_TYPES()            \
1978         x(crc32c,               0)      \
1979         x(crc64,                1)      \
1980         x(siphash_old,          2)      \
1981         x(siphash,              3)
1982
1983 enum bch_str_hash_type {
1984 #define x(t, n) BCH_STR_HASH_##t = n,
1985         BCH_STR_HASH_TYPES()
1986 #undef x
1987         BCH_STR_HASH_NR
1988 };
1989
1990 #define BCH_STR_HASH_OPTS()             \
1991         x(crc32c,               0)      \
1992         x(crc64,                1)      \
1993         x(siphash,              2)
1994
1995 enum bch_str_hash_opts {
1996 #define x(t, n) BCH_STR_HASH_OPT_##t = n,
1997         BCH_STR_HASH_OPTS()
1998 #undef x
1999         BCH_STR_HASH_OPT_NR
2000 };
2001
2002 #define BCH_CSUM_TYPES()                        \
2003         x(none,                         0)      \
2004         x(crc32c_nonzero,               1)      \
2005         x(crc64_nonzero,                2)      \
2006         x(chacha20_poly1305_80,         3)      \
2007         x(chacha20_poly1305_128,        4)      \
2008         x(crc32c,                       5)      \
2009         x(crc64,                        6)      \
2010         x(xxhash,                       7)
2011
2012 enum bch_csum_type {
2013 #define x(t, n) BCH_CSUM_##t = n,
2014         BCH_CSUM_TYPES()
2015 #undef x
2016         BCH_CSUM_NR
2017 };
2018
2019 static const __maybe_unused unsigned bch_crc_bytes[] = {
2020         [BCH_CSUM_none]                         = 0,
2021         [BCH_CSUM_crc32c_nonzero]               = 4,
2022         [BCH_CSUM_crc32c]                       = 4,
2023         [BCH_CSUM_crc64_nonzero]                = 8,
2024         [BCH_CSUM_crc64]                        = 8,
2025         [BCH_CSUM_xxhash]                       = 8,
2026         [BCH_CSUM_chacha20_poly1305_80]         = 10,
2027         [BCH_CSUM_chacha20_poly1305_128]        = 16,
2028 };
2029
2030 static inline _Bool bch2_csum_type_is_encryption(enum bch_csum_type type)
2031 {
2032         switch (type) {
2033         case BCH_CSUM_chacha20_poly1305_80:
2034         case BCH_CSUM_chacha20_poly1305_128:
2035                 return true;
2036         default:
2037                 return false;
2038         }
2039 }
2040
2041 #define BCH_CSUM_OPTS()                 \
2042         x(none,                 0)      \
2043         x(crc32c,               1)      \
2044         x(crc64,                2)      \
2045         x(xxhash,               3)
2046
2047 enum bch_csum_opts {
2048 #define x(t, n) BCH_CSUM_OPT_##t = n,
2049         BCH_CSUM_OPTS()
2050 #undef x
2051         BCH_CSUM_OPT_NR
2052 };
2053
2054 #define BCH_COMPRESSION_TYPES()         \
2055         x(none,                 0)      \
2056         x(lz4_old,              1)      \
2057         x(gzip,                 2)      \
2058         x(lz4,                  3)      \
2059         x(zstd,                 4)      \
2060         x(incompressible,       5)
2061
2062 enum bch_compression_type {
2063 #define x(t, n) BCH_COMPRESSION_TYPE_##t = n,
2064         BCH_COMPRESSION_TYPES()
2065 #undef x
2066         BCH_COMPRESSION_TYPE_NR
2067 };
2068
2069 #define BCH_COMPRESSION_OPTS()          \
2070         x(none,         0)              \
2071         x(lz4,          1)              \
2072         x(gzip,         2)              \
2073         x(zstd,         3)
2074
2075 enum bch_compression_opts {
2076 #define x(t, n) BCH_COMPRESSION_OPT_##t = n,
2077         BCH_COMPRESSION_OPTS()
2078 #undef x
2079         BCH_COMPRESSION_OPT_NR
2080 };
2081
2082 /*
2083  * Magic numbers
2084  *
2085  * The various other data structures have their own magic numbers, which are
2086  * xored with the first part of the cache set's UUID
2087  */
2088
2089 #define BCACHE_MAGIC                                                    \
2090         UUID_INIT(0xc68573f6, 0x4e1a, 0x45ca,                           \
2091                   0x82, 0x65, 0xf5, 0x7f, 0x48, 0xba, 0x6d, 0x81)
2092 #define BCHFS_MAGIC                                                     \
2093         UUID_INIT(0xc68573f6, 0x66ce, 0x90a9,                           \
2094                   0xd9, 0x6a, 0x60, 0xcf, 0x80, 0x3d, 0xf7, 0xef)
2095
2096 #define BCACHEFS_STATFS_MAGIC           0xca451a4e
2097
2098 #define JSET_MAGIC              __cpu_to_le64(0x245235c1a3625032ULL)
2099 #define BSET_MAGIC              __cpu_to_le64(0x90135c78b99e07f5ULL)
2100
2101 static inline __le64 __bch2_sb_magic(struct bch_sb *sb)
2102 {
2103         __le64 ret;
2104
2105         memcpy(&ret, &sb->uuid, sizeof(ret));
2106         return ret;
2107 }
2108
2109 static inline __u64 __jset_magic(struct bch_sb *sb)
2110 {
2111         return __le64_to_cpu(__bch2_sb_magic(sb) ^ JSET_MAGIC);
2112 }
2113
2114 static inline __u64 __bset_magic(struct bch_sb *sb)
2115 {
2116         return __le64_to_cpu(__bch2_sb_magic(sb) ^ BSET_MAGIC);
2117 }
2118
2119 /* Journal */
2120
2121 #define JSET_KEYS_U64s  (sizeof(struct jset_entry) / sizeof(__u64))
2122
2123 #define BCH_JSET_ENTRY_TYPES()                  \
2124         x(btree_keys,           0)              \
2125         x(btree_root,           1)              \
2126         x(prio_ptrs,            2)              \
2127         x(blacklist,            3)              \
2128         x(blacklist_v2,         4)              \
2129         x(usage,                5)              \
2130         x(data_usage,           6)              \
2131         x(clock,                7)              \
2132         x(dev_usage,            8)              \
2133         x(log,                  9)              \
2134         x(overwrite,            10)             \
2135         x(write_buffer_keys,    11)
2136
2137 enum {
2138 #define x(f, nr)        BCH_JSET_ENTRY_##f      = nr,
2139         BCH_JSET_ENTRY_TYPES()
2140 #undef x
2141         BCH_JSET_ENTRY_NR
2142 };
2143
2144 /*
2145  * Journal sequence numbers can be blacklisted: bsets record the max sequence
2146  * number of all the journal entries they contain updates for, so that on
2147  * recovery we can ignore those bsets that contain index updates newer that what
2148  * made it into the journal.
2149  *
2150  * This means that we can't reuse that journal_seq - we have to skip it, and
2151  * then record that we skipped it so that the next time we crash and recover we
2152  * don't think there was a missing journal entry.
2153  */
2154 struct jset_entry_blacklist {
2155         struct jset_entry       entry;
2156         __le64                  seq;
2157 };
2158
2159 struct jset_entry_blacklist_v2 {
2160         struct jset_entry       entry;
2161         __le64                  start;
2162         __le64                  end;
2163 };
2164
2165 #define BCH_FS_USAGE_TYPES()                    \
2166         x(reserved,             0)              \
2167         x(inodes,               1)              \
2168         x(key_version,          2)
2169
2170 enum {
2171 #define x(f, nr)        BCH_FS_USAGE_##f        = nr,
2172         BCH_FS_USAGE_TYPES()
2173 #undef x
2174         BCH_FS_USAGE_NR
2175 };
2176
2177 struct jset_entry_usage {
2178         struct jset_entry       entry;
2179         __le64                  v;
2180 } __packed;
2181
2182 struct jset_entry_data_usage {
2183         struct jset_entry       entry;
2184         __le64                  v;
2185         struct bch_replicas_entry_v1 r;
2186 } __packed;
2187
2188 struct jset_entry_clock {
2189         struct jset_entry       entry;
2190         __u8                    rw;
2191         __u8                    pad[7];
2192         __le64                  time;
2193 } __packed;
2194
2195 struct jset_entry_dev_usage_type {
2196         __le64                  buckets;
2197         __le64                  sectors;
2198         __le64                  fragmented;
2199 } __packed;
2200
2201 struct jset_entry_dev_usage {
2202         struct jset_entry       entry;
2203         __le32                  dev;
2204         __u32                   pad;
2205
2206         __le64                  buckets_ec;
2207         __le64                  _buckets_unavailable; /* No longer used */
2208
2209         struct jset_entry_dev_usage_type d[];
2210 };
2211
2212 static inline unsigned jset_entry_dev_usage_nr_types(struct jset_entry_dev_usage *u)
2213 {
2214         return (vstruct_bytes(&u->entry) - sizeof(struct jset_entry_dev_usage)) /
2215                 sizeof(struct jset_entry_dev_usage_type);
2216 }
2217
2218 struct jset_entry_log {
2219         struct jset_entry       entry;
2220         u8                      d[];
2221 } __packed;
2222
2223 /*
2224  * On disk format for a journal entry:
2225  * seq is monotonically increasing; every journal entry has its own unique
2226  * sequence number.
2227  *
2228  * last_seq is the oldest journal entry that still has keys the btree hasn't
2229  * flushed to disk yet.
2230  *
2231  * version is for on disk format changes.
2232  */
2233 struct jset {
2234         struct bch_csum         csum;
2235
2236         __le64                  magic;
2237         __le64                  seq;
2238         __le32                  version;
2239         __le32                  flags;
2240
2241         __le32                  u64s; /* size of d[] in u64s */
2242
2243         __u8                    encrypted_start[0];
2244
2245         __le16                  _read_clock; /* no longer used */
2246         __le16                  _write_clock;
2247
2248         /* Sequence number of oldest dirty journal entry */
2249         __le64                  last_seq;
2250
2251
2252         struct jset_entry       start[0];
2253         __u64                   _data[];
2254 } __packed __aligned(8);
2255
2256 LE32_BITMASK(JSET_CSUM_TYPE,    struct jset, flags, 0, 4);
2257 LE32_BITMASK(JSET_BIG_ENDIAN,   struct jset, flags, 4, 5);
2258 LE32_BITMASK(JSET_NO_FLUSH,     struct jset, flags, 5, 6);
2259
2260 #define BCH_JOURNAL_BUCKETS_MIN         8
2261
2262 /* Btree: */
2263
2264 enum btree_id_flags {
2265         BTREE_ID_EXTENTS        = BIT(0),
2266         BTREE_ID_SNAPSHOTS      = BIT(1),
2267         BTREE_ID_SNAPSHOT_FIELD = BIT(2),
2268         BTREE_ID_DATA           = BIT(3),
2269 };
2270
2271 #define BCH_BTREE_IDS()                                                         \
2272         x(extents,              0,      BTREE_ID_EXTENTS|BTREE_ID_SNAPSHOTS|BTREE_ID_DATA,\
2273           BIT_ULL(KEY_TYPE_whiteout)|                                           \
2274           BIT_ULL(KEY_TYPE_error)|                                              \
2275           BIT_ULL(KEY_TYPE_cookie)|                                             \
2276           BIT_ULL(KEY_TYPE_extent)|                                             \
2277           BIT_ULL(KEY_TYPE_reservation)|                                        \
2278           BIT_ULL(KEY_TYPE_reflink_p)|                                          \
2279           BIT_ULL(KEY_TYPE_inline_data))                                        \
2280         x(inodes,               1,      BTREE_ID_SNAPSHOTS,                     \
2281           BIT_ULL(KEY_TYPE_whiteout)|                                           \
2282           BIT_ULL(KEY_TYPE_inode)|                                              \
2283           BIT_ULL(KEY_TYPE_inode_v2)|                                           \
2284           BIT_ULL(KEY_TYPE_inode_v3)|                                           \
2285           BIT_ULL(KEY_TYPE_inode_generation))                                   \
2286         x(dirents,              2,      BTREE_ID_SNAPSHOTS,                     \
2287           BIT_ULL(KEY_TYPE_whiteout)|                                           \
2288           BIT_ULL(KEY_TYPE_hash_whiteout)|                                      \
2289           BIT_ULL(KEY_TYPE_dirent))                                             \
2290         x(xattrs,               3,      BTREE_ID_SNAPSHOTS,                     \
2291           BIT_ULL(KEY_TYPE_whiteout)|                                           \
2292           BIT_ULL(KEY_TYPE_cookie)|                                             \
2293           BIT_ULL(KEY_TYPE_hash_whiteout)|                                      \
2294           BIT_ULL(KEY_TYPE_xattr))                                              \
2295         x(alloc,                4,      0,                                      \
2296           BIT_ULL(KEY_TYPE_alloc)|                                              \
2297           BIT_ULL(KEY_TYPE_alloc_v2)|                                           \
2298           BIT_ULL(KEY_TYPE_alloc_v3)|                                           \
2299           BIT_ULL(KEY_TYPE_alloc_v4))                                           \
2300         x(quotas,               5,      0,                                      \
2301           BIT_ULL(KEY_TYPE_quota))                                              \
2302         x(stripes,              6,      0,                                      \
2303           BIT_ULL(KEY_TYPE_stripe))                                             \
2304         x(reflink,              7,      BTREE_ID_EXTENTS|BTREE_ID_DATA,         \
2305           BIT_ULL(KEY_TYPE_reflink_v)|                                          \
2306           BIT_ULL(KEY_TYPE_indirect_inline_data))                               \
2307         x(subvolumes,           8,      0,                                      \
2308           BIT_ULL(KEY_TYPE_subvolume))                                          \
2309         x(snapshots,            9,      0,                                      \
2310           BIT_ULL(KEY_TYPE_snapshot))                                           \
2311         x(lru,                  10,     0,                                      \
2312           BIT_ULL(KEY_TYPE_set))                                                \
2313         x(freespace,            11,     BTREE_ID_EXTENTS,                       \
2314           BIT_ULL(KEY_TYPE_set))                                                \
2315         x(need_discard,         12,     0,                                      \
2316           BIT_ULL(KEY_TYPE_set))                                                \
2317         x(backpointers,         13,     0,                                      \
2318           BIT_ULL(KEY_TYPE_backpointer))                                        \
2319         x(bucket_gens,          14,     0,                                      \
2320           BIT_ULL(KEY_TYPE_bucket_gens))                                        \
2321         x(snapshot_trees,       15,     0,                                      \
2322           BIT_ULL(KEY_TYPE_snapshot_tree))                                      \
2323         x(deleted_inodes,       16,     BTREE_ID_SNAPSHOT_FIELD,                \
2324           BIT_ULL(KEY_TYPE_set))                                                \
2325         x(logged_ops,           17,     0,                                      \
2326           BIT_ULL(KEY_TYPE_logged_op_truncate)|                                 \
2327           BIT_ULL(KEY_TYPE_logged_op_finsert))                                  \
2328         x(rebalance_work,       18,     BTREE_ID_SNAPSHOT_FIELD,                \
2329           BIT_ULL(KEY_TYPE_set)|BIT_ULL(KEY_TYPE_cookie))
2330
2331 enum btree_id {
2332 #define x(name, nr, ...) BTREE_ID_##name = nr,
2333         BCH_BTREE_IDS()
2334 #undef x
2335         BTREE_ID_NR
2336 };
2337
2338 #define BTREE_MAX_DEPTH         4U
2339
2340 /* Btree nodes */
2341
2342 /*
2343  * Btree nodes
2344  *
2345  * On disk a btree node is a list/log of these; within each set the keys are
2346  * sorted
2347  */
2348 struct bset {
2349         __le64                  seq;
2350
2351         /*
2352          * Highest journal entry this bset contains keys for.
2353          * If on recovery we don't see that journal entry, this bset is ignored:
2354          * this allows us to preserve the order of all index updates after a
2355          * crash, since the journal records a total order of all index updates
2356          * and anything that didn't make it to the journal doesn't get used.
2357          */
2358         __le64                  journal_seq;
2359
2360         __le32                  flags;
2361         __le16                  version;
2362         __le16                  u64s; /* count of d[] in u64s */
2363
2364         struct bkey_packed      start[0];
2365         __u64                   _data[];
2366 } __packed __aligned(8);
2367
2368 LE32_BITMASK(BSET_CSUM_TYPE,    struct bset, flags, 0, 4);
2369
2370 LE32_BITMASK(BSET_BIG_ENDIAN,   struct bset, flags, 4, 5);
2371 LE32_BITMASK(BSET_SEPARATE_WHITEOUTS,
2372                                 struct bset, flags, 5, 6);
2373
2374 /* Sector offset within the btree node: */
2375 LE32_BITMASK(BSET_OFFSET,       struct bset, flags, 16, 32);
2376
2377 struct btree_node {
2378         struct bch_csum         csum;
2379         __le64                  magic;
2380
2381         /* this flags field is encrypted, unlike bset->flags: */
2382         __le64                  flags;
2383
2384         /* Closed interval: */
2385         struct bpos             min_key;
2386         struct bpos             max_key;
2387         struct bch_extent_ptr   _ptr; /* not used anymore */
2388         struct bkey_format      format;
2389
2390         union {
2391         struct bset             keys;
2392         struct {
2393                 __u8            pad[22];
2394                 __le16          u64s;
2395                 __u64           _data[0];
2396
2397         };
2398         };
2399 } __packed __aligned(8);
2400
2401 LE64_BITMASK(BTREE_NODE_ID_LO,  struct btree_node, flags,  0,  4);
2402 LE64_BITMASK(BTREE_NODE_LEVEL,  struct btree_node, flags,  4,  8);
2403 LE64_BITMASK(BTREE_NODE_NEW_EXTENT_OVERWRITE,
2404                                 struct btree_node, flags,  8,  9);
2405 LE64_BITMASK(BTREE_NODE_ID_HI,  struct btree_node, flags,  9, 25);
2406 /* 25-32 unused */
2407 LE64_BITMASK(BTREE_NODE_SEQ,    struct btree_node, flags, 32, 64);
2408
2409 static inline __u64 BTREE_NODE_ID(struct btree_node *n)
2410 {
2411         return BTREE_NODE_ID_LO(n) | (BTREE_NODE_ID_HI(n) << 4);
2412 }
2413
2414 static inline void SET_BTREE_NODE_ID(struct btree_node *n, __u64 v)
2415 {
2416         SET_BTREE_NODE_ID_LO(n, v);
2417         SET_BTREE_NODE_ID_HI(n, v >> 4);
2418 }
2419
2420 struct btree_node_entry {
2421         struct bch_csum         csum;
2422
2423         union {
2424         struct bset             keys;
2425         struct {
2426                 __u8            pad[22];
2427                 __le16          u64s;
2428                 __u64           _data[0];
2429         };
2430         };
2431 } __packed __aligned(8);
2432
2433 #endif /* _BCACHEFS_FORMAT_H */