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

   1 /* SPDX-License-Identifier: GPL-2.0 */
   2 #ifndef _BCACHEFS_FORMAT_H
   3 #define _BCACHEFS_FORMAT_H
   4
   5 /*
   6  * bcachefs on disk data structures
   7  *
   8  * OVERVIEW:
   9  *
  10  * There are three main types of on disk data structures in bcachefs (this is
  11  * reduced from 5 in bcache)
  12  *
  13  *  - superblock
  14  *  - journal
  15  *  - btree
  16  *
  17  * The btree is the primary structure; most metadata exists as keys in the
  18  * various btrees. There are only a small number of btrees, they're not
  19  * sharded - we have one btree for extents, another for inodes, et cetera.
  20  *
  21  * SUPERBLOCK:
  22  *
  23  * The superblock contains the location of the journal, the list of devices in
  24  * the filesystem, and in general any metadata we need in order to decide
  25  * whether we can start a filesystem or prior to reading the journal/btree
  26  * roots.
  27  *
  28  * The superblock is extensible, and most of the contents of the superblock are
  29  * in variable length, type tagged fields; see struct bch_sb_field.
  30  *
  31  * Backup superblocks do not reside in a fixed location; also, superblocks do
  32  * not have a fixed size. To locate backup superblocks we have struct
  33  * bch_sb_layout; we store a copy of this inside every superblock, and also
  34  * before the first superblock.
  35  *
  36  * JOURNAL:
  37  *
  38  * The journal primarily records btree updates in the order they occurred;
  39  * journal replay consists of just iterating over all the keys in the open
  40  * journal entries and re-inserting them into the btrees.
  41  *
  42  * The journal also contains entry types for the btree roots, and blacklisted
  43  * journal sequence numbers (see journal_seq_blacklist.c).
  44  *
  45  * BTREE:
  46  *
  47  * bcachefs btrees are copy on write b+ trees, where nodes are big (typically
  48  * 128k-256k) and log structured. We use struct btree_node for writing the first
  49  * entry in a given node (offset 0), and struct btree_node_entry for all
  50  * subsequent writes.
  51  *
  52  * After the header, btree node entries contain a list of keys in sorted order.
  53  * Values are stored inline with the keys; since values are variable length (and
  54  * keys effectively are variable length too, due to packing) we can't do random
  55  * access without building up additional in memory tables in the btree node read
  56  * path.
  57  *
  58  * BTREE KEYS (struct bkey):
  59  *
  60  * The various btrees share a common format for the key - so as to avoid
  61  * switching in fastpath lookup/comparison code - but define their own
  62  * structures for the key values.
  63  *
  64  * The size of a key/value pair is stored as a u8 in units of u64s, so the max
  65  * size is just under 2k. The common part also contains a type tag for the
  66  * value, and a format field indicating whether the key is packed or not (and
  67  * also meant to allow adding new key fields in the future, if desired).
  68  *
  69  * bkeys, when stored within a btree node, may also be packed. In that case, the
  70  * bkey_format in that node is used to unpack it. Packed bkeys mean that we can
  71  * be generous with field sizes in the common part of the key format (64 bit
  72  * inode number, 64 bit offset, 96 bit version field, etc.) for negligible cost.
  73  */
  74
  75 #include <asm/types.h>
  76 #include <asm/byteorder.h>
  77 #include <linux/kernel.h>
  78 #include <linux/uuid.h>
  79 #include "vstructs.h"
  80
  81 #ifdef __KERNEL__
  82 typedef uuid_t __uuid_t;
  83 #endif
  84
  85 #define BITMASK(name, type, field, offset, end)                         \
  86 static const unsigned   name##_OFFSET = offset;                         \
  87 static const unsigned   name##_BITS = (end - offset);                   \
  88                                                                         \
  89 static inline __u64 name(const type *k)                                 \
  90 {                                                                       \
  91         return (k->field >> offset) & ~(~0ULL << (end - offset));       \
  92 }                                                                       \
  93                                                                         \
  94 static inline void SET_##name(type *k, __u64 v)                         \
  95 {                                                                       \
  96         k->field &= ~(~(~0ULL << (end - offset)) << offset);            \
  97         k->field |= (v & ~(~0ULL << (end - offset))) << offset;         \
  98 }
  99
 100 #define LE_BITMASK(_bits, name, type, field, offset, end)               \
 101 static const unsigned   name##_OFFSET = offset;                         \
 102 static const unsigned   name##_BITS = (end - offset);                   \
 103 static const __u##_bits name##_MAX = (1ULL << (end - offset)) - 1;      \
 104                                                                         \
 105 static inline __u64 name(const type *k)                                 \
 106 {                                                                       \
 107         return (__le##_bits##_to_cpu(k->field) >> offset) &             \
 108                 ~(~0ULL << (end - offset));                             \
 109 }                                                                       \
 110                                                                         \
 111 static inline void SET_##name(type *k, __u64 v)                         \
 112 {                                                                       \
 113         __u##_bits new = __le##_bits##_to_cpu(k->field);                \
 114                                                                         \
 115         new &= ~(~(~0ULL << (end - offset)) << offset);                 \
 116         new |= (v & ~(~0ULL << (end - offset))) << offset;              \
 117         k->field = __cpu_to_le##_bits(new);                             \
 118 }
 119
 120 #define LE16_BITMASK(n, t, f, o, e)     LE_BITMASK(16, n, t, f, o, e)
 121 #define LE32_BITMASK(n, t, f, o, e)     LE_BITMASK(32, n, t, f, o, e)
 122 #define LE64_BITMASK(n, t, f, o, e)     LE_BITMASK(64, n, t, f, o, e)
 123
 124 struct bkey_format {
 125         __u8            key_u64s;
 126         __u8            nr_fields;
 127         /* One unused slot for now: */
 128         __u8            bits_per_field[6];
 129         __le64          field_offset[6];
 130 };
 131
 132 /* Btree keys - all units are in sectors */
 133
 134 struct bpos {
 135         /*
 136          * Word order matches machine byte order - btree code treats a bpos as a
 137          * single large integer, for search/comparison purposes
 138          *
 139          * Note that wherever a bpos is embedded in another on disk data
 140          * structure, it has to be byte swabbed when reading in metadata that
 141          * wasn't written in native endian order:
 142          */
 143 #if __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__
 144         __u32           snapshot;
 145         __u64           offset;
 146         __u64           inode;
 147 #elif __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__
 148         __u64           inode;
 149         __u64           offset;         /* Points to end of extent - sectors */
 150         __u32           snapshot;
 151 #else
 152 #error edit for your odd byteorder.
 153 #endif
 154 } __packed __aligned(4);
 155
 156 #define KEY_INODE_MAX                   ((__u64)~0ULL)
 157 #define KEY_OFFSET_MAX                  ((__u64)~0ULL)
 158 #define KEY_SNAPSHOT_MAX                ((__u32)~0U)
 159 #define KEY_SIZE_MAX                    ((__u32)~0U)
 160
 161 static inline struct bpos SPOS(__u64 inode, __u64 offset, __u32 snapshot)
 162 {
 163         return (struct bpos) {
 164                 .inode          = inode,
 165                 .offset         = offset,
 166                 .snapshot       = snapshot,
 167         };
 168 }
 169
 170 #define POS_MIN                         SPOS(0, 0, 0)
 171 #define POS_MAX                         SPOS(KEY_INODE_MAX, KEY_OFFSET_MAX, 0)
 172 #define SPOS_MAX                        SPOS(KEY_INODE_MAX, KEY_OFFSET_MAX, KEY_SNAPSHOT_MAX)
 173 #define POS(_inode, _offset)            SPOS(_inode, _offset, 0)
 174
 175 /* Empty placeholder struct, for container_of() */
 176 struct bch_val {
 177         __u64           __nothing[0];
 178 };
 179
 180 struct bversion {
 181 #if __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__
 182         __u64           lo;
 183         __u32           hi;
 184 #elif __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__
 185         __u32           hi;
 186         __u64           lo;
 187 #endif
 188 } __packed __aligned(4);
 189
 190 struct bkey {
 191         /* Size of combined key and value, in u64s */
 192         __u8            u64s;
 193
 194         /* Format of key (0 for format local to btree node) */
 195 #if defined(__LITTLE_ENDIAN_BITFIELD)
 196         __u8            format:7,
 197                         needs_whiteout:1;
 198 #elif defined (__BIG_ENDIAN_BITFIELD)
 199         __u8            needs_whiteout:1,
 200                         format:7;
 201 #else
 202 #error edit for your odd byteorder.
 203 #endif
 204
 205         /* Type of the value */
 206         __u8            type;
 207
 208 #if __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__
 209         __u8            pad[1];
 210
 211         struct bversion version;
 212         __u32           size;           /* extent size, in sectors */
 213         struct bpos     p;
 214 #elif __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__
 215         struct bpos     p;
 216         __u32           size;           /* extent size, in sectors */
 217         struct bversion version;
 218
 219         __u8            pad[1];
 220 #endif
 221 } __packed __aligned(8);
 222
 223 struct bkey_packed {
 224         __u64           _data[0];
 225
 226         /* Size of combined key and value, in u64s */
 227         __u8            u64s;
 228
 229         /* Format of key (0 for format local to btree node) */
 230
 231         /*
 232          * XXX: next incompat on disk format change, switch format and
 233          * needs_whiteout - bkey_packed() will be cheaper if format is the high
 234          * bits of the bitfield
 235          */
 236 #if defined(__LITTLE_ENDIAN_BITFIELD)
 237         __u8            format:7,
 238                         needs_whiteout:1;
 239 #elif defined (__BIG_ENDIAN_BITFIELD)
 240         __u8            needs_whiteout:1,
 241                         format:7;
 242 #endif
 243
 244         /* Type of the value */
 245         __u8            type;
 246         __u8            key_start[0];
 247
 248         /*
 249          * We copy bkeys with struct assignment in various places, and while
 250          * that shouldn't be done with packed bkeys we can't disallow it in C,
 251          * and it's legal to cast a bkey to a bkey_packed  - so padding it out
 252          * to the same size as struct bkey should hopefully be safest.
 253          */
 254         __u8            pad[sizeof(struct bkey) - 3];
 255 } __packed __aligned(8);
 256
 257 typedef struct {
 258         __le64                  lo;
 259         __le64                  hi;
 260 } bch_le128;
 261
 262 #define BKEY_U64s                       (sizeof(struct bkey) / sizeof(__u64))
 263 #define BKEY_U64s_MAX                   U8_MAX
 264 #define BKEY_VAL_U64s_MAX               (BKEY_U64s_MAX - BKEY_U64s)
 265
 266 #define KEY_PACKED_BITS_START           24
 267
 268 #define KEY_FORMAT_LOCAL_BTREE          0
 269 #define KEY_FORMAT_CURRENT              1
 270
 271 enum bch_bkey_fields {
 272         BKEY_FIELD_INODE,
 273         BKEY_FIELD_OFFSET,
 274         BKEY_FIELD_SNAPSHOT,
 275         BKEY_FIELD_SIZE,
 276         BKEY_FIELD_VERSION_HI,
 277         BKEY_FIELD_VERSION_LO,
 278         BKEY_NR_FIELDS,
 279 };
 280
 281 #define bkey_format_field(name, field)                                  \
 282         [BKEY_FIELD_##name] = (sizeof(((struct bkey *) NULL)->field) * 8)
 283
 284 #define BKEY_FORMAT_CURRENT                                             \
 285 ((struct bkey_format) {                                                 \
 286         .key_u64s       = BKEY_U64s,                                    \
 287         .nr_fields      = BKEY_NR_FIELDS,                               \
 288         .bits_per_field = {                                             \
 289                 bkey_format_field(INODE,        p.inode),               \
 290                 bkey_format_field(OFFSET,       p.offset),              \
 291                 bkey_format_field(SNAPSHOT,     p.snapshot),            \
 292                 bkey_format_field(SIZE,         size),                  \
 293                 bkey_format_field(VERSION_HI,   version.hi),            \
 294                 bkey_format_field(VERSION_LO,   version.lo),            \
 295         },                                                              \
 296 })
 297
 298 /* bkey with inline value */
 299 struct bkey_i {
 300         __u64                   _data[0];
 301
 302         struct bkey     k;
 303         struct bch_val  v;
 304 };
 305
 306 #define KEY(_inode, _offset, _size)                                     \
 307 ((struct bkey) {                                                        \
 308         .u64s           = BKEY_U64s,                                    \
 309         .format         = KEY_FORMAT_CURRENT,                           \
 310         .p              = POS(_inode, _offset),                         \
 311         .size           = _size,                                        \
 312 })
 313
 314 static inline void bkey_init(struct bkey *k)
 315 {
 316         *k = KEY(0, 0, 0);
 317 }
 318
 319 #define bkey_bytes(_k)          ((_k)->u64s * sizeof(__u64))
 320
 321 #define __BKEY_PADDED(key, pad)                                 \
 322         struct bkey_i key; __u64 key ## _pad[pad]
 323
 324 /*
 325  * - DELETED keys are used internally to mark keys that should be ignored but
 326  *   override keys in composition order.  Their version number is ignored.
 327  *
 328  * - DISCARDED keys indicate that the data is all 0s because it has been
 329  *   discarded. DISCARDs may have a version; if the version is nonzero the key
 330  *   will be persistent, otherwise the key will be dropped whenever the btree
 331  *   node is rewritten (like DELETED keys).
 332  *
 333  * - ERROR: any read of the data returns a read error, as the data was lost due
 334  *   to a failing device. Like DISCARDED keys, they can be removed (overridden)
 335  *   by new writes or cluster-wide GC. Node repair can also overwrite them with
 336  *   the same or a more recent version number, but not with an older version
 337  *   number.
 338  *
 339  * - WHITEOUT: for hash table btrees
 340  */
 341 #define BCH_BKEY_TYPES()                                \
 342         x(deleted,              0)                      \
 343         x(whiteout,             1)                      \
 344         x(error,                2)                      \
 345         x(cookie,               3)                      \
 346         x(hash_whiteout,        4)                      \
 347         x(btree_ptr,            5)                      \
 348         x(extent,               6)                      \
 349         x(reservation,          7)                      \
 350         x(inode,                8)                      \
 351         x(inode_generation,     9)                      \
 352         x(dirent,               10)                     \
 353         x(xattr,                11)                     \
 354         x(alloc,                12)                     \
 355         x(quota,                13)                     \
 356         x(stripe,               14)                     \
 357         x(reflink_p,            15)                     \
 358         x(reflink_v,            16)                     \
 359         x(inline_data,          17)                     \
 360         x(btree_ptr_v2,         18)                     \
 361         x(indirect_inline_data, 19)                     \
 362         x(alloc_v2,             20)                     \
 363         x(subvolume,            21)                     \
 364         x(snapshot,             22)                     \
 365         x(inode_v2,             23)                     \
 366         x(alloc_v3,             24)                     \
 367         x(set,                  25)                     \
 368         x(lru,                  26)                     \
 369         x(alloc_v4,             27)                     \
 370         x(backpointer,          28)                     \
 371         x(inode_v3,             29)                     \
 372         x(bucket_gens,          30)                     \
 373         x(snapshot_tree,        31)
 374
 375 enum bch_bkey_type {
 376 #define x(name, nr) KEY_TYPE_##name     = nr,
 377         BCH_BKEY_TYPES()
 378 #undef x
 379         KEY_TYPE_MAX,
 380 };
 381
 382 struct bch_deleted {
 383         struct bch_val          v;
 384 };
 385
 386 struct bch_whiteout {
 387         struct bch_val          v;
 388 };
 389
 390 struct bch_error {
 391         struct bch_val          v;
 392 };
 393
 394 struct bch_cookie {
 395         struct bch_val          v;
 396         __le64                  cookie;
 397 };
 398
 399 struct bch_hash_whiteout {
 400         struct bch_val          v;
 401 };
 402
 403 struct bch_set {
 404         struct bch_val          v;
 405 };
 406
 407 /* Extents */
 408
 409 /*
 410  * In extent bkeys, the value is a list of pointers (bch_extent_ptr), optionally
 411  * preceded by checksum/compression information (bch_extent_crc32 or
 412  * bch_extent_crc64).
 413  *
 414  * One major determining factor in the format of extents is how we handle and
 415  * represent extents that have been partially overwritten and thus trimmed:
 416  *
 417  * If an extent is not checksummed or compressed, when the extent is trimmed we
 418  * don't have to remember the extent we originally allocated and wrote: we can
 419  * merely adjust ptr->offset to point to the start of the data that is currently
 420  * live. The size field in struct bkey records the current (live) size of the
 421  * extent, and is also used to mean "size of region on disk that we point to" in
 422  * this case.
 423  *
 424  * Thus an extent that is not checksummed or compressed will consist only of a
 425  * list of bch_extent_ptrs, with none of the fields in
 426  * bch_extent_crc32/bch_extent_crc64.
 427  *
 428  * When an extent is checksummed or compressed, it's not possible to read only
 429  * the data that is currently live: we have to read the entire extent that was
 430  * originally written, and then return only the part of the extent that is
 431  * currently live.
 432  *
 433  * Thus, in addition to the current size of the extent in struct bkey, we need
 434  * to store the size of the originally allocated space - this is the
 435  * compressed_size and uncompressed_size fields in bch_extent_crc32/64. Also,
 436  * when the extent is trimmed, instead of modifying the offset field of the
 437  * pointer, we keep a second smaller offset field - "offset into the original
 438  * extent of the currently live region".
 439  *
 440  * The other major determining factor is replication and data migration:
 441  *
 442  * Each pointer may have its own bch_extent_crc32/64. When doing a replicated
 443  * write, we will initially write all the replicas in the same format, with the
 444  * same checksum type and compression format - however, when copygc runs later (or
 445  * tiering/cache promotion, anything that moves data), it is not in general
 446  * going to rewrite all the pointers at once - one of the replicas may be in a
 447  * bucket on one device that has very little fragmentation while another lives
 448  * in a bucket that has become heavily fragmented, and thus is being rewritten
 449  * sooner than the rest.
 450  *
 451  * Thus it will only move a subset of the pointers (or in the case of
 452  * tiering/cache promotion perhaps add a single pointer without dropping any
 453  * current pointers), and if the extent has been partially overwritten it must
 454  * write only the currently live portion (or copygc would not be able to reduce
 455  * fragmentation!) - which necessitates a different bch_extent_crc format for
 456  * the new pointer.
 457  *
 458  * But in the interests of space efficiency, we don't want to store one
 459  * bch_extent_crc for each pointer if we don't have to.
 460  *
 461  * Thus, a bch_extent consists of bch_extent_crc32s, bch_extent_crc64s, and
 462  * bch_extent_ptrs appended arbitrarily one after the other. We determine the
 463  * type of a given entry with a scheme similar to utf8 (except we're encoding a
 464  * type, not a size), encoding the type in the position of the first set bit:
 465  *
 466  * bch_extent_crc32     - 0b1
 467  * bch_extent_ptr       - 0b10
 468  * bch_extent_crc64     - 0b100
 469  *
 470  * We do it this way because bch_extent_crc32 is _very_ constrained on bits (and
 471  * bch_extent_crc64 is the least constrained).
 472  *
 473  * Then, each bch_extent_crc32/64 applies to the pointers that follow after it,
 474  * until the next bch_extent_crc32/64.
 475  *
 476  * If there are no bch_extent_crcs preceding a bch_extent_ptr, then that pointer
 477  * is neither checksummed nor compressed.
 478  */
 479
 480 /* 128 bits, sufficient for cryptographic MACs: */
 481 struct bch_csum {
 482         __le64                  lo;
 483         __le64                  hi;
 484 } __packed __aligned(8);
 485
 486 #define BCH_EXTENT_ENTRY_TYPES()                \
 487         x(ptr,                  0)              \
 488         x(crc32,                1)              \
 489         x(crc64,                2)              \
 490         x(crc128,               3)              \
 491         x(stripe_ptr,           4)              \
 492         x(rebalance,            5)
 493 #define BCH_EXTENT_ENTRY_MAX    6
 494
 495 enum bch_extent_entry_type {
 496 #define x(f, n) BCH_EXTENT_ENTRY_##f = n,
 497         BCH_EXTENT_ENTRY_TYPES()
 498 #undef x
 499 };
 500
 501 /* Compressed/uncompressed size are stored biased by 1: */
 502 struct bch_extent_crc32 {
 503 #if defined(__LITTLE_ENDIAN_BITFIELD)
 504         __u32                   type:2,
 505                                 _compressed_size:7,
 506                                 _uncompressed_size:7,
 507                                 offset:7,
 508                                 _unused:1,
 509                                 csum_type:4,
 510                                 compression_type:4;
 511         __u32                   csum;
 512 #elif defined (__BIG_ENDIAN_BITFIELD)
 513         __u32                   csum;
 514         __u32                   compression_type:4,
 515                                 csum_type:4,
 516                                 _unused:1,
 517                                 offset:7,
 518                                 _uncompressed_size:7,
 519                                 _compressed_size:7,
 520                                 type:2;
 521 #endif
 522 } __packed __aligned(8);
 523
 524 #define CRC32_SIZE_MAX          (1U << 7)
 525 #define CRC32_NONCE_MAX         0
 526
 527 struct bch_extent_crc64 {
 528 #if defined(__LITTLE_ENDIAN_BITFIELD)
 529         __u64                   type:3,
 530                                 _compressed_size:9,
 531                                 _uncompressed_size:9,
 532                                 offset:9,
 533                                 nonce:10,
 534                                 csum_type:4,
 535                                 compression_type:4,
 536                                 csum_hi:16;
 537 #elif defined (__BIG_ENDIAN_BITFIELD)
 538         __u64                   csum_hi:16,
 539                                 compression_type:4,
 540                                 csum_type:4,
 541                                 nonce:10,
 542                                 offset:9,
 543                                 _uncompressed_size:9,
 544                                 _compressed_size:9,
 545                                 type:3;
 546 #endif
 547         __u64                   csum_lo;
 548 } __packed __aligned(8);
 549
 550 #define CRC64_SIZE_MAX          (1U << 9)
 551 #define CRC64_NONCE_MAX         ((1U << 10) - 1)
 552
 553 struct bch_extent_crc128 {
 554 #if defined(__LITTLE_ENDIAN_BITFIELD)
 555         __u64                   type:4,
 556                                 _compressed_size:13,
 557                                 _uncompressed_size:13,
 558                                 offset:13,
 559                                 nonce:13,
 560                                 csum_type:4,
 561                                 compression_type:4;
 562 #elif defined (__BIG_ENDIAN_BITFIELD)
 563         __u64                   compression_type:4,
 564                                 csum_type:4,
 565                                 nonce:13,
 566                                 offset:13,
 567                                 _uncompressed_size:13,
 568                                 _compressed_size:13,
 569                                 type:4;
 570 #endif
 571         struct bch_csum         csum;
 572 } __packed __aligned(8);
 573
 574 #define CRC128_SIZE_MAX         (1U << 13)
 575 #define CRC128_NONCE_MAX        ((1U << 13) - 1)
 576
 577 /*
 578  * @reservation - pointer hasn't been written to, just reserved
 579  */
 580 struct bch_extent_ptr {
 581 #if defined(__LITTLE_ENDIAN_BITFIELD)
 582         __u64                   type:1,
 583                                 cached:1,
 584                                 unused:1,
 585                                 unwritten:1,
 586                                 offset:44, /* 8 petabytes */
 587                                 dev:8,
 588                                 gen:8;
 589 #elif defined (__BIG_ENDIAN_BITFIELD)
 590         __u64                   gen:8,
 591                                 dev:8,
 592                                 offset:44,
 593                                 unwritten:1,
 594                                 unused:1,
 595                                 cached:1,
 596                                 type:1;
 597 #endif
 598 } __packed __aligned(8);
 599
 600 struct bch_extent_stripe_ptr {
 601 #if defined(__LITTLE_ENDIAN_BITFIELD)
 602         __u64                   type:5,
 603                                 block:8,
 604                                 redundancy:4,
 605                                 idx:47;
 606 #elif defined (__BIG_ENDIAN_BITFIELD)
 607         __u64                   idx:47,
 608                                 redundancy:4,
 609                                 block:8,
 610                                 type:5;
 611 #endif
 612 };
 613
 614 struct bch_extent_reservation {
 615 #if defined(__LITTLE_ENDIAN_BITFIELD)
 616         __u64                   type:6,
 617                                 unused:22,
 618                                 replicas:4,
 619                                 generation:32;
 620 #elif defined (__BIG_ENDIAN_BITFIELD)
 621         __u64                   generation:32,
 622                                 replicas:4,
 623                                 unused:22,
 624                                 type:6;
 625 #endif
 626 };
 627
 628 struct bch_extent_rebalance {
 629 #if defined(__LITTLE_ENDIAN_BITFIELD)
 630         __u64                   type:7,
 631                                 unused:33,
 632                                 compression:8,
 633                                 target:16;
 634 #elif defined (__BIG_ENDIAN_BITFIELD)
 635         __u64                   target:16,
 636                                 compression:8,
 637                                 unused:33,
 638                                 type:7;
 639 #endif
 640 };
 641
 642 union bch_extent_entry {
 643 #if __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__ ||  __BITS_PER_LONG == 64
 644         unsigned long                   type;
 645 #elif __BITS_PER_LONG == 32
 646         struct {
 647                 unsigned long           pad;
 648                 unsigned long           type;
 649         };
 650 #else
 651 #error edit for your odd byteorder.
 652 #endif
 653
 654 #define x(f, n) struct bch_extent_##f   f;
 655         BCH_EXTENT_ENTRY_TYPES()
 656 #undef x
 657 };
 658
 659 struct bch_btree_ptr {
 660         struct bch_val          v;
 661
 662         __u64                   _data[0];
 663         struct bch_extent_ptr   start[];
 664 } __packed __aligned(8);
 665
 666 struct bch_btree_ptr_v2 {
 667         struct bch_val          v;
 668
 669         __u64                   mem_ptr;
 670         __le64                  seq;
 671         __le16                  sectors_written;
 672         __le16                  flags;
 673         struct bpos             min_key;
 674         __u64                   _data[0];
 675         struct bch_extent_ptr   start[];
 676 } __packed __aligned(8);
 677
 678 LE16_BITMASK(BTREE_PTR_RANGE_UPDATED,   struct bch_btree_ptr_v2, flags, 0, 1);
 679
 680 struct bch_extent {
 681         struct bch_val          v;
 682
 683         __u64                   _data[0];
 684         union bch_extent_entry  start[];
 685 } __packed __aligned(8);
 686
 687 struct bch_reservation {
 688         struct bch_val          v;
 689
 690         __le32                  generation;
 691         __u8                    nr_replicas;
 692         __u8                    pad[3];
 693 } __packed __aligned(8);
 694
 695 /* Maximum size (in u64s) a single pointer could be: */
 696 #define BKEY_EXTENT_PTR_U64s_MAX\
 697         ((sizeof(struct bch_extent_crc128) +                    \
 698           sizeof(struct bch_extent_ptr)) / sizeof(__u64))
 699
 700 /* Maximum possible size of an entire extent value: */
 701 #define BKEY_EXTENT_VAL_U64s_MAX                                \
 702         (1 + BKEY_EXTENT_PTR_U64s_MAX * (BCH_REPLICAS_MAX + 1))
 703
 704 /* * Maximum possible size of an entire extent, key + value: */
 705 #define BKEY_EXTENT_U64s_MAX            (BKEY_U64s + BKEY_EXTENT_VAL_U64s_MAX)
 706
 707 /* Btree pointers don't carry around checksums: */
 708 #define BKEY_BTREE_PTR_VAL_U64s_MAX                             \
 709         ((sizeof(struct bch_btree_ptr_v2) +                     \
 710           sizeof(struct bch_extent_ptr) * BCH_REPLICAS_MAX) / sizeof(__u64))
 711 #define BKEY_BTREE_PTR_U64s_MAX                                 \
 712         (BKEY_U64s + BKEY_BTREE_PTR_VAL_U64s_MAX)
 713
 714 /* Inodes */
 715
 716 #define BLOCKDEV_INODE_MAX      4096
 717
 718 #define BCACHEFS_ROOT_INO       4096
 719
 720 struct bch_inode {
 721         struct bch_val          v;
 722
 723         __le64                  bi_hash_seed;
 724         __le32                  bi_flags;
 725         __le16                  bi_mode;
 726         __u8                    fields[0];
 727 } __packed __aligned(8);
 728
 729 struct bch_inode_v2 {
 730         struct bch_val          v;
 731
 732         __le64                  bi_journal_seq;
 733         __le64                  bi_hash_seed;
 734         __le64                  bi_flags;
 735         __le16                  bi_mode;
 736         __u8                    fields[0];
 737 } __packed __aligned(8);
 738
 739 struct bch_inode_v3 {
 740         struct bch_val          v;
 741
 742         __le64                  bi_journal_seq;
 743         __le64                  bi_hash_seed;
 744         __le64                  bi_flags;
 745         __le64                  bi_sectors;
 746         __le64                  bi_size;
 747         __le64                  bi_version;
 748         __u8                    fields[0];
 749 } __packed __aligned(8);
 750
 751 #define INODEv3_FIELDS_START_INITIAL    6
 752 #define INODEv3_FIELDS_START_CUR        (offsetof(struct bch_inode_v3, fields) / sizeof(__u64))
 753
 754 struct bch_inode_generation {
 755         struct bch_val          v;
 756
 757         __le32                  bi_generation;
 758         __le32                  pad;
 759 } __packed __aligned(8);
 760
 761 /*
 762  * bi_subvol and bi_parent_subvol are only set for subvolume roots:
 763  */
 764
 765 #define BCH_INODE_FIELDS_v2()                   \
 766         x(bi_atime,                     96)     \
 767         x(bi_ctime,                     96)     \
 768         x(bi_mtime,                     96)     \
 769         x(bi_otime,                     96)     \
 770         x(bi_size,                      64)     \
 771         x(bi_sectors,                   64)     \
 772         x(bi_uid,                       32)     \
 773         x(bi_gid,                       32)     \
 774         x(bi_nlink,                     32)     \
 775         x(bi_generation,                32)     \
 776         x(bi_dev,                       32)     \
 777         x(bi_data_checksum,             8)      \
 778         x(bi_compression,               8)      \
 779         x(bi_project,                   32)     \
 780         x(bi_background_compression,    8)      \
 781         x(bi_data_replicas,             8)      \
 782         x(bi_promote_target,            16)     \
 783         x(bi_foreground_target,         16)     \
 784         x(bi_background_target,         16)     \
 785         x(bi_erasure_code,              16)     \
 786         x(bi_fields_set,                16)     \
 787         x(bi_dir,                       64)     \
 788         x(bi_dir_offset,                64)     \
 789         x(bi_subvol,                    32)     \
 790         x(bi_parent_subvol,             32)
 791
 792 #define BCH_INODE_FIELDS_v3()                   \
 793         x(bi_atime,                     96)     \
 794         x(bi_ctime,                     96)     \
 795         x(bi_mtime,                     96)     \
 796         x(bi_otime,                     96)     \
 797         x(bi_uid,                       32)     \
 798         x(bi_gid,                       32)     \
 799         x(bi_nlink,                     32)     \
 800         x(bi_generation,                32)     \
 801         x(bi_dev,                       32)     \
 802         x(bi_data_checksum,             8)      \
 803         x(bi_compression,               8)      \
 804         x(bi_project,                   32)     \
 805         x(bi_background_compression,    8)      \
 806         x(bi_data_replicas,             8)      \
 807         x(bi_promote_target,            16)     \
 808         x(bi_foreground_target,         16)     \
 809         x(bi_background_target,         16)     \
 810         x(bi_erasure_code,              16)     \
 811         x(bi_fields_set,                16)     \
 812         x(bi_dir,                       64)     \
 813         x(bi_dir_offset,                64)     \
 814         x(bi_subvol,                    32)     \
 815         x(bi_parent_subvol,             32)     \
 816         x(bi_nocow,                     8)
 817
 818 /* subset of BCH_INODE_FIELDS */
 819 #define BCH_INODE_OPTS()                        \
 820         x(data_checksum,                8)      \
 821         x(compression,                  8)      \
 822         x(project,                      32)     \
 823         x(background_compression,       8)      \
 824         x(data_replicas,                8)      \
 825         x(promote_target,               16)     \
 826         x(foreground_target,            16)     \
 827         x(background_target,            16)     \
 828         x(erasure_code,                 16)     \
 829         x(nocow,                        8)
 830
 831 enum inode_opt_id {
 832 #define x(name, ...)                            \
 833         Inode_opt_##name,
 834         BCH_INODE_OPTS()
 835 #undef  x
 836         Inode_opt_nr,
 837 };
 838
 839 enum {
 840         /*
 841          * User flags (get/settable with FS_IOC_*FLAGS, correspond to FS_*_FL
 842          * flags)
 843          */
 844         __BCH_INODE_SYNC                = 0,
 845         __BCH_INODE_IMMUTABLE           = 1,
 846         __BCH_INODE_APPEND              = 2,
 847         __BCH_INODE_NODUMP              = 3,
 848         __BCH_INODE_NOATIME             = 4,
 849
 850         __BCH_INODE_I_SIZE_DIRTY        = 5,
 851         __BCH_INODE_I_SECTORS_DIRTY     = 6,
 852         __BCH_INODE_UNLINKED            = 7,
 853         __BCH_INODE_BACKPTR_UNTRUSTED   = 8,
 854
 855         /* bits 20+ reserved for packed fields below: */
 856 };
 857
 858 #define BCH_INODE_SYNC          (1 << __BCH_INODE_SYNC)
 859 #define BCH_INODE_IMMUTABLE     (1 << __BCH_INODE_IMMUTABLE)
 860 #define BCH_INODE_APPEND        (1 << __BCH_INODE_APPEND)
 861 #define BCH_INODE_NODUMP        (1 << __BCH_INODE_NODUMP)
 862 #define BCH_INODE_NOATIME       (1 << __BCH_INODE_NOATIME)
 863 #define BCH_INODE_I_SIZE_DIRTY  (1 << __BCH_INODE_I_SIZE_DIRTY)
 864 #define BCH_INODE_I_SECTORS_DIRTY (1 << __BCH_INODE_I_SECTORS_DIRTY)
 865 #define BCH_INODE_UNLINKED      (1 << __BCH_INODE_UNLINKED)
 866 #define BCH_INODE_BACKPTR_UNTRUSTED (1 << __BCH_INODE_BACKPTR_UNTRUSTED)
 867
 868 LE32_BITMASK(INODE_STR_HASH,    struct bch_inode, bi_flags, 20, 24);
 869 LE32_BITMASK(INODE_NR_FIELDS,   struct bch_inode, bi_flags, 24, 31);
 870 LE32_BITMASK(INODE_NEW_VARINT,  struct bch_inode, bi_flags, 31, 32);
 871
 872 LE64_BITMASK(INODEv2_STR_HASH,  struct bch_inode_v2, bi_flags, 20, 24);
 873 LE64_BITMASK(INODEv2_NR_FIELDS, struct bch_inode_v2, bi_flags, 24, 31);
 874
 875 LE64_BITMASK(INODEv3_STR_HASH,  struct bch_inode_v3, bi_flags, 20, 24);
 876 LE64_BITMASK(INODEv3_NR_FIELDS, struct bch_inode_v3, bi_flags, 24, 31);
 877
 878 LE64_BITMASK(INODEv3_FIELDS_START,
 879                                 struct bch_inode_v3, bi_flags, 31, 36);
 880 LE64_BITMASK(INODEv3_MODE,      struct bch_inode_v3, bi_flags, 36, 52);
 881
 882 /* Dirents */
 883
 884 /*
 885  * Dirents (and xattrs) have to implement string lookups; since our b-tree
 886  * doesn't support arbitrary length strings for the key, we instead index by a
 887  * 64 bit hash (currently truncated sha1) of the string, stored in the offset
 888  * field of the key - using linear probing to resolve hash collisions. This also
 889  * provides us with the readdir cookie posix requires.
 890  *
 891  * Linear probing requires us to use whiteouts for deletions, in the event of a
 892  * collision:
 893  */
 894
 895 struct bch_dirent {
 896         struct bch_val          v;
 897
 898         /* Target inode number: */
 899         union {
 900         __le64                  d_inum;
 901         struct {                /* DT_SUBVOL */
 902         __le32                  d_child_subvol;
 903         __le32                  d_parent_subvol;
 904         };
 905         };
 906
 907         /*
 908          * Copy of mode bits 12-15 from the target inode - so userspace can get
 909          * the filetype without having to do a stat()
 910          */
 911         __u8                    d_type;
 912
 913         __u8                    d_name[];
 914 } __packed __aligned(8);
 915
 916 #define DT_SUBVOL       16
 917 #define BCH_DT_MAX      17
 918
 919 #define BCH_NAME_MAX    512
 920
 921 /* Xattrs */
 922
 923 #define KEY_TYPE_XATTR_INDEX_USER                       0
 924 #define KEY_TYPE_XATTR_INDEX_POSIX_ACL_ACCESS   1
 925 #define KEY_TYPE_XATTR_INDEX_POSIX_ACL_DEFAULT  2
 926 #define KEY_TYPE_XATTR_INDEX_TRUSTED                    3
 927 #define KEY_TYPE_XATTR_INDEX_SECURITY           4
 928
 929 struct bch_xattr {
 930         struct bch_val          v;
 931         __u8                    x_type;
 932         __u8                    x_name_len;
 933         __le16                  x_val_len;
 934         __u8                    x_name[];
 935 } __packed __aligned(8);
 936
 937 /* Bucket/allocation information: */
 938
 939 struct bch_alloc {
 940         struct bch_val          v;
 941         __u8                    fields;
 942         __u8                    gen;
 943         __u8                    data[];
 944 } __packed __aligned(8);
 945
 946 #define BCH_ALLOC_FIELDS_V1()                   \
 947         x(read_time,            16)             \
 948         x(write_time,           16)             \
 949         x(data_type,            8)              \
 950         x(dirty_sectors,        16)             \
 951         x(cached_sectors,       16)             \
 952         x(oldest_gen,           8)              \
 953         x(stripe,               32)             \
 954         x(stripe_redundancy,    8)
 955
 956 enum {
 957 #define x(name, _bits) BCH_ALLOC_FIELD_V1_##name,
 958         BCH_ALLOC_FIELDS_V1()
 959 #undef x
 960 };
 961
 962 struct bch_alloc_v2 {
 963         struct bch_val          v;
 964         __u8                    nr_fields;
 965         __u8                    gen;
 966         __u8                    oldest_gen;
 967         __u8                    data_type;
 968         __u8                    data[];
 969 } __packed __aligned(8);
 970
 971 #define BCH_ALLOC_FIELDS_V2()                   \
 972         x(read_time,            64)             \
 973         x(write_time,           64)             \
 974         x(dirty_sectors,        32)             \
 975         x(cached_sectors,       32)             \
 976         x(stripe,               32)             \
 977         x(stripe_redundancy,    8)
 978
 979 struct bch_alloc_v3 {
 980         struct bch_val          v;
 981         __le64                  journal_seq;
 982         __le32                  flags;
 983         __u8                    nr_fields;
 984         __u8                    gen;
 985         __u8                    oldest_gen;
 986         __u8                    data_type;
 987         __u8                    data[];
 988 } __packed __aligned(8);
 989
 990 LE32_BITMASK(BCH_ALLOC_V3_NEED_DISCARD,struct bch_alloc_v3, flags,  0,  1)
 991 LE32_BITMASK(BCH_ALLOC_V3_NEED_INC_GEN,struct bch_alloc_v3, flags,  1,  2)
 992
 993 struct bch_alloc_v4 {
 994         struct bch_val          v;
 995         __u64                   journal_seq;
 996         __u32                   flags;
 997         __u8                    gen;
 998         __u8                    oldest_gen;
 999         __u8                    data_type;
1000         __u8                    stripe_redundancy;
1001         __u32                   dirty_sectors;
1002         __u32                   cached_sectors;
1003         __u64                   io_time[2];
1004         __u32                   stripe;
1005         __u32                   nr_external_backpointers;
1006         __u64                   fragmentation_lru;
1007 } __packed __aligned(8);
1008
1009 #define BCH_ALLOC_V4_U64s_V0    6
1010 #define BCH_ALLOC_V4_U64s       (sizeof(struct bch_alloc_v4) / sizeof(__u64))
1011
1012 BITMASK(BCH_ALLOC_V4_NEED_DISCARD,      struct bch_alloc_v4, flags,  0,  1)
1013 BITMASK(BCH_ALLOC_V4_NEED_INC_GEN,      struct bch_alloc_v4, flags,  1,  2)
1014 BITMASK(BCH_ALLOC_V4_BACKPOINTERS_START,struct bch_alloc_v4, flags,  2,  8)
1015 BITMASK(BCH_ALLOC_V4_NR_BACKPOINTERS,   struct bch_alloc_v4, flags,  8,  14)
1016
1017 #define BCH_ALLOC_V4_NR_BACKPOINTERS_MAX        40
1018
1019 struct bch_backpointer {
1020         struct bch_val          v;
1021         __u8                    btree_id;
1022         __u8                    level;
1023         __u8                    data_type;
1024         __u64                   bucket_offset:40;
1025         __u32                   bucket_len;
1026         struct bpos             pos;
1027 } __packed __aligned(8);
1028
1029 #define KEY_TYPE_BUCKET_GENS_BITS       8
1030 #define KEY_TYPE_BUCKET_GENS_NR         (1U << KEY_TYPE_BUCKET_GENS_BITS)
1031 #define KEY_TYPE_BUCKET_GENS_MASK       (KEY_TYPE_BUCKET_GENS_NR - 1)
1032
1033 struct bch_bucket_gens {
1034         struct bch_val          v;
1035         u8                      gens[KEY_TYPE_BUCKET_GENS_NR];
1036 } __packed __aligned(8);
1037
1038 /* Quotas: */
1039
1040 enum quota_types {
1041         QTYP_USR                = 0,
1042         QTYP_GRP                = 1,
1043         QTYP_PRJ                = 2,
1044         QTYP_NR                 = 3,
1045 };
1046
1047 enum quota_counters {
1048         Q_SPC                   = 0,
1049         Q_INO                   = 1,
1050         Q_COUNTERS              = 2,
1051 };
1052
1053 struct bch_quota_counter {
1054         __le64                  hardlimit;
1055         __le64                  softlimit;
1056 };
1057
1058 struct bch_quota {
1059         struct bch_val          v;
1060         struct bch_quota_counter c[Q_COUNTERS];
1061 } __packed __aligned(8);
1062
1063 /* Erasure coding */
1064
1065 struct bch_stripe {
1066         struct bch_val          v;
1067         __le16                  sectors;
1068         __u8                    algorithm;
1069         __u8                    nr_blocks;
1070         __u8                    nr_redundant;
1071
1072         __u8                    csum_granularity_bits;
1073         __u8                    csum_type;
1074         __u8                    pad;
1075
1076         struct bch_extent_ptr   ptrs[];
1077 } __packed __aligned(8);
1078
1079 /* Reflink: */
1080
1081 struct bch_reflink_p {
1082         struct bch_val          v;
1083         __le64                  idx;
1084         /*
1085          * A reflink pointer might point to an indirect extent which is then
1086          * later split (by copygc or rebalance). If we only pointed to part of
1087          * the original indirect extent, and then one of the fragments is
1088          * outside the range we point to, we'd leak a refcount: so when creating
1089          * reflink pointers, we need to store pad values to remember the full
1090          * range we were taking a reference on.
1091          */
1092         __le32                  front_pad;
1093         __le32                  back_pad;
1094 } __packed __aligned(8);
1095
1096 struct bch_reflink_v {
1097         struct bch_val          v;
1098         __le64                  refcount;
1099         union bch_extent_entry  start[0];
1100         __u64                   _data[0];
1101 } __packed __aligned(8);
1102
1103 struct bch_indirect_inline_data {
1104         struct bch_val          v;
1105         __le64                  refcount;
1106         u8                      data[0];
1107 };
1108
1109 /* Inline data */
1110
1111 struct bch_inline_data {
1112         struct bch_val          v;
1113         u8                      data[0];
1114 };
1115
1116 /* Subvolumes: */
1117
1118 #define SUBVOL_POS_MIN          POS(0, 1)
1119 #define SUBVOL_POS_MAX          POS(0, S32_MAX)
1120 #define BCACHEFS_ROOT_SUBVOL    1
1121
1122 struct bch_subvolume {
1123         struct bch_val          v;
1124         __le32                  flags;
1125         __le32                  snapshot;
1126         __le64                  inode;
1127         /*
1128          * Snapshot subvolumes form a tree, separate from the snapshot nodes
1129          * tree - if this subvolume is a snapshot, this is the ID of the
1130          * subvolume it was created from:
1131          */
1132         __le32                  parent;
1133         __le32                  pad;
1134         bch_le128               otime;
1135 };
1136
1137 LE32_BITMASK(BCH_SUBVOLUME_RO,          struct bch_subvolume, flags,  0,  1)
1138 /*
1139  * We need to know whether a subvolume is a snapshot so we can know whether we
1140  * can delete it (or whether it should just be rm -rf'd)
1141  */
1142 LE32_BITMASK(BCH_SUBVOLUME_SNAP,        struct bch_subvolume, flags,  1,  2)
1143 LE32_BITMASK(BCH_SUBVOLUME_UNLINKED,    struct bch_subvolume, flags,  2,  3)
1144
1145 /* Snapshots */
1146
1147 struct bch_snapshot {
1148         struct bch_val          v;
1149         __le32                  flags;
1150         __le32                  parent;
1151         __le32                  children[2];
1152         __le32                  subvol;
1153         /* corresponds to a bch_snapshot_tree in BTREE_ID_snapshot_trees */
1154         __le32                  tree;
1155         __le32                  depth;
1156         __le32                  skip[3];
1157 };
1158
1159 LE32_BITMASK(BCH_SNAPSHOT_DELETED,      struct bch_snapshot, flags,  0,  1)
1160
1161 /* True if a subvolume points to this snapshot node: */
1162 LE32_BITMASK(BCH_SNAPSHOT_SUBVOL,       struct bch_snapshot, flags,  1,  2)
1163
1164 /*
1165  * Snapshot trees:
1166  *
1167  * The snapshot_trees btree gives us persistent indentifier for each tree of
1168  * bch_snapshot nodes, and allow us to record and easily find the root/master
1169  * subvolume that other snapshots were created from:
1170  */
1171 struct bch_snapshot_tree {
1172         struct bch_val          v;
1173         __le32                  master_subvol;
1174         __le32                  root_snapshot;
1175 };
1176
1177 /* LRU btree: */
1178
1179 struct bch_lru {
1180         struct bch_val          v;
1181         __le64                  idx;
1182 } __packed __aligned(8);
1183
1184 #define LRU_ID_STRIPES          (1U << 16)
1185
1186 /* Optional/variable size superblock sections: */
1187
1188 struct bch_sb_field {
1189         __u64                   _data[0];
1190         __le32                  u64s;
1191         __le32                  type;
1192 };
1193
1194 #define BCH_SB_FIELDS()                         \
1195         x(journal,      0)                      \
1196         x(members,      1)                      \
1197         x(crypt,        2)                      \
1198         x(replicas_v0,  3)                      \
1199         x(quota,        4)                      \
1200         x(disk_groups,  5)                      \
1201         x(clean,        6)                      \
1202         x(replicas,     7)                      \
1203         x(journal_seq_blacklist, 8)             \
1204         x(journal_v2,   9)                      \
1205         x(counters,     10)
1206
1207 enum bch_sb_field_type {
1208 #define x(f, nr)        BCH_SB_FIELD_##f = nr,
1209         BCH_SB_FIELDS()
1210 #undef x
1211         BCH_SB_FIELD_NR
1212 };
1213
1214 /*
1215  * Most superblock fields are replicated in all device's superblocks - a few are
1216  * not:
1217  */
1218 #define BCH_SINGLE_DEVICE_SB_FIELDS             \
1219         ((1U << BCH_SB_FIELD_journal)|          \
1220          (1U << BCH_SB_FIELD_journal_v2))
1221
1222 /* BCH_SB_FIELD_journal: */
1223
1224 struct bch_sb_field_journal {
1225         struct bch_sb_field     field;
1226         __le64                  buckets[0];
1227 };
1228
1229 struct bch_sb_field_journal_v2 {
1230         struct bch_sb_field     field;
1231
1232         struct bch_sb_field_journal_v2_entry {
1233                 __le64          start;
1234                 __le64          nr;
1235         }                       d[0];
1236 };
1237
1238 /* BCH_SB_FIELD_members: */
1239
1240 #define BCH_MIN_NR_NBUCKETS     (1 << 6)
1241
1242 struct bch_member {
1243         __uuid_t                uuid;
1244         __le64                  nbuckets;       /* device size */
1245         __le16                  first_bucket;   /* index of first bucket used */
1246         __le16                  bucket_size;    /* sectors */
1247         __le32                  pad;
1248         __le64                  last_mount;     /* time_t */
1249
1250         __le64                  flags[2];
1251 };
1252
1253 LE64_BITMASK(BCH_MEMBER_STATE,          struct bch_member, flags[0],  0,  4)
1254 /* 4-14 unused, was TIER, HAS_(META)DATA, REPLACEMENT */
1255 LE64_BITMASK(BCH_MEMBER_DISCARD,        struct bch_member, flags[0], 14, 15)
1256 LE64_BITMASK(BCH_MEMBER_DATA_ALLOWED,   struct bch_member, flags[0], 15, 20)
1257 LE64_BITMASK(BCH_MEMBER_GROUP,          struct bch_member, flags[0], 20, 28)
1258 LE64_BITMASK(BCH_MEMBER_DURABILITY,     struct bch_member, flags[0], 28, 30)
1259 LE64_BITMASK(BCH_MEMBER_FREESPACE_INITIALIZED,
1260                                         struct bch_member, flags[0], 30, 31)
1261
1262 #if 0
1263 LE64_BITMASK(BCH_MEMBER_NR_READ_ERRORS, struct bch_member, flags[1], 0,  20);
1264 LE64_BITMASK(BCH_MEMBER_NR_WRITE_ERRORS,struct bch_member, flags[1], 20, 40);
1265 #endif
1266
1267 #define BCH_MEMBER_STATES()                     \
1268         x(rw,           0)                      \
1269         x(ro,           1)                      \
1270         x(failed,       2)                      \
1271         x(spare,        3)
1272
1273 enum bch_member_state {
1274 #define x(t, n) BCH_MEMBER_STATE_##t = n,
1275         BCH_MEMBER_STATES()
1276 #undef x
1277         BCH_MEMBER_STATE_NR
1278 };
1279
1280 struct bch_sb_field_members {
1281         struct bch_sb_field     field;
1282         struct bch_member       members[0];
1283 };
1284
1285 /* BCH_SB_FIELD_crypt: */
1286
1287 struct nonce {
1288         __le32                  d[4];
1289 };
1290
1291 struct bch_key {
1292         __le64                  key[4];
1293 };
1294
1295 #define BCH_KEY_MAGIC                                   \
1296         (((__u64) 'b' <<  0)|((__u64) 'c' <<  8)|               \
1297          ((__u64) 'h' << 16)|((__u64) '*' << 24)|               \
1298          ((__u64) '*' << 32)|((__u64) 'k' << 40)|               \
1299          ((__u64) 'e' << 48)|((__u64) 'y' << 56))
1300
1301 struct bch_encrypted_key {
1302         __le64                  magic;
1303         struct bch_key          key;
1304 };
1305
1306 /*
1307  * If this field is present in the superblock, it stores an encryption key which
1308  * is used encrypt all other data/metadata. The key will normally be encrypted
1309  * with the key userspace provides, but if encryption has been turned off we'll
1310  * just store the master key unencrypted in the superblock so we can access the
1311  * previously encrypted data.
1312  */
1313 struct bch_sb_field_crypt {
1314         struct bch_sb_field     field;
1315
1316         __le64                  flags;
1317         __le64                  kdf_flags;
1318         struct bch_encrypted_key key;
1319 };
1320
1321 LE64_BITMASK(BCH_CRYPT_KDF_TYPE,        struct bch_sb_field_crypt, flags, 0, 4);
1322
1323 enum bch_kdf_types {
1324         BCH_KDF_SCRYPT          = 0,
1325         BCH_KDF_NR              = 1,
1326 };
1327
1328 /* stored as base 2 log of scrypt params: */
1329 LE64_BITMASK(BCH_KDF_SCRYPT_N,  struct bch_sb_field_crypt, kdf_flags,  0, 16);
1330 LE64_BITMASK(BCH_KDF_SCRYPT_R,  struct bch_sb_field_crypt, kdf_flags, 16, 32);
1331 LE64_BITMASK(BCH_KDF_SCRYPT_P,  struct bch_sb_field_crypt, kdf_flags, 32, 48);
1332
1333 /* BCH_SB_FIELD_replicas: */
1334
1335 #define BCH_DATA_TYPES()                \
1336         x(free,         0)              \
1337         x(sb,           1)              \
1338         x(journal,      2)              \
1339         x(btree,        3)              \
1340         x(user,         4)              \
1341         x(cached,       5)              \
1342         x(parity,       6)              \
1343         x(stripe,       7)              \
1344         x(need_gc_gens, 8)              \
1345         x(need_discard, 9)
1346
1347 enum bch_data_type {
1348 #define x(t, n) BCH_DATA_##t,
1349         BCH_DATA_TYPES()
1350 #undef x
1351         BCH_DATA_NR
1352 };
1353
1354 static inline bool data_type_is_empty(enum bch_data_type type)
1355 {
1356         switch (type) {
1357         case BCH_DATA_free:
1358         case BCH_DATA_need_gc_gens:
1359         case BCH_DATA_need_discard:
1360                 return true;
1361         default:
1362                 return false;
1363         }
1364 }
1365
1366 static inline bool data_type_is_hidden(enum bch_data_type type)
1367 {
1368         switch (type) {
1369         case BCH_DATA_sb:
1370         case BCH_DATA_journal:
1371                 return true;
1372         default:
1373                 return false;
1374         }
1375 }
1376
1377 struct bch_replicas_entry_v0 {
1378         __u8                    data_type;
1379         __u8                    nr_devs;
1380         __u8                    devs[0];
1381 } __packed;
1382
1383 struct bch_sb_field_replicas_v0 {
1384         struct bch_sb_field     field;
1385         struct bch_replicas_entry_v0 entries[0];
1386 } __packed __aligned(8);
1387
1388 struct bch_replicas_entry {
1389         __u8                    data_type;
1390         __u8                    nr_devs;
1391         __u8                    nr_required;
1392         __u8                    devs[0];
1393 } __packed;
1394
1395 #define replicas_entry_bytes(_i)                                        \
1396         (offsetof(typeof(*(_i)), devs) + (_i)->nr_devs)
1397
1398 struct bch_sb_field_replicas {
1399         struct bch_sb_field     field;
1400         struct bch_replicas_entry entries[0];
1401 } __packed __aligned(8);
1402
1403 /* BCH_SB_FIELD_quota: */
1404
1405 struct bch_sb_quota_counter {
1406         __le32                          timelimit;
1407         __le32                          warnlimit;
1408 };
1409
1410 struct bch_sb_quota_type {
1411         __le64                          flags;
1412         struct bch_sb_quota_counter     c[Q_COUNTERS];
1413 };
1414
1415 struct bch_sb_field_quota {
1416         struct bch_sb_field             field;
1417         struct bch_sb_quota_type        q[QTYP_NR];
1418 } __packed __aligned(8);
1419
1420 /* BCH_SB_FIELD_disk_groups: */
1421
1422 #define BCH_SB_LABEL_SIZE               32
1423
1424 struct bch_disk_group {
1425         __u8                    label[BCH_SB_LABEL_SIZE];
1426         __le64                  flags[2];
1427 } __packed __aligned(8);
1428
1429 LE64_BITMASK(BCH_GROUP_DELETED,         struct bch_disk_group, flags[0], 0,  1)
1430 LE64_BITMASK(BCH_GROUP_DATA_ALLOWED,    struct bch_disk_group, flags[0], 1,  6)
1431 LE64_BITMASK(BCH_GROUP_PARENT,          struct bch_disk_group, flags[0], 6, 24)
1432
1433 struct bch_sb_field_disk_groups {
1434         struct bch_sb_field     field;
1435         struct bch_disk_group   entries[0];
1436 } __packed __aligned(8);
1437
1438 /* BCH_SB_FIELD_counters */
1439
1440 #define BCH_PERSISTENT_COUNTERS()                               \
1441         x(io_read,                                      0)      \
1442         x(io_write,                                     1)      \
1443         x(io_move,                                      2)      \
1444         x(bucket_invalidate,                            3)      \
1445         x(bucket_discard,                               4)      \
1446         x(bucket_alloc,                                 5)      \
1447         x(bucket_alloc_fail,                            6)      \
1448         x(btree_cache_scan,                             7)      \
1449         x(btree_cache_reap,                             8)      \
1450         x(btree_cache_cannibalize,                      9)      \
1451         x(btree_cache_cannibalize_lock,                 10)     \
1452         x(btree_cache_cannibalize_lock_fail,            11)     \
1453         x(btree_cache_cannibalize_unlock,               12)     \
1454         x(btree_node_write,                             13)     \
1455         x(btree_node_read,                              14)     \
1456         x(btree_node_compact,                           15)     \
1457         x(btree_node_merge,                             16)     \
1458         x(btree_node_split,                             17)     \
1459         x(btree_node_rewrite,                           18)     \
1460         x(btree_node_alloc,                             19)     \
1461         x(btree_node_free,                              20)     \
1462         x(btree_node_set_root,                          21)     \
1463         x(btree_path_relock_fail,                       22)     \
1464         x(btree_path_upgrade_fail,                      23)     \
1465         x(btree_reserve_get_fail,                       24)     \
1466         x(journal_entry_full,                           25)     \
1467         x(journal_full,                                 26)     \
1468         x(journal_reclaim_finish,                       27)     \
1469         x(journal_reclaim_start,                        28)     \
1470         x(journal_write,                                29)     \
1471         x(read_promote,                                 30)     \
1472         x(read_bounce,                                  31)     \
1473         x(read_split,                                   33)     \
1474         x(read_retry,                                   32)     \
1475         x(read_reuse_race,                              34)     \
1476         x(move_extent_read,                             35)     \
1477         x(move_extent_write,                            36)     \
1478         x(move_extent_finish,                           37)     \
1479         x(move_extent_fail,                             38)     \
1480         x(move_extent_alloc_mem_fail,                   39)     \
1481         x(copygc,                                       40)     \
1482         x(copygc_wait,                                  41)     \
1483         x(gc_gens_end,                                  42)     \
1484         x(gc_gens_start,                                43)     \
1485         x(trans_blocked_journal_reclaim,                44)     \
1486         x(trans_restart_btree_node_reused,              45)     \
1487         x(trans_restart_btree_node_split,               46)     \
1488         x(trans_restart_fault_inject,                   47)     \
1489         x(trans_restart_iter_upgrade,                   48)     \
1490         x(trans_restart_journal_preres_get,             49)     \
1491         x(trans_restart_journal_reclaim,                50)     \
1492         x(trans_restart_journal_res_get,                51)     \
1493         x(trans_restart_key_cache_key_realloced,        52)     \
1494         x(trans_restart_key_cache_raced,                53)     \
1495         x(trans_restart_mark_replicas,                  54)     \
1496         x(trans_restart_mem_realloced,                  55)     \
1497         x(trans_restart_memory_allocation_failure,      56)     \
1498         x(trans_restart_relock,                         57)     \
1499         x(trans_restart_relock_after_fill,              58)     \
1500         x(trans_restart_relock_key_cache_fill,          59)     \
1501         x(trans_restart_relock_next_node,               60)     \
1502         x(trans_restart_relock_parent_for_fill,         61)     \
1503         x(trans_restart_relock_path,                    62)     \
1504         x(trans_restart_relock_path_intent,             63)     \
1505         x(trans_restart_too_many_iters,                 64)     \
1506         x(trans_restart_traverse,                       65)     \
1507         x(trans_restart_upgrade,                        66)     \
1508         x(trans_restart_would_deadlock,                 67)     \
1509         x(trans_restart_would_deadlock_write,           68)     \
1510         x(trans_restart_injected,                       69)     \
1511         x(trans_restart_key_cache_upgrade,              70)     \
1512         x(trans_traverse_all,                           71)     \
1513         x(transaction_commit,                           72)     \
1514         x(write_super,                                  73)     \
1515         x(trans_restart_would_deadlock_recursion_limit, 74)     \
1516         x(trans_restart_write_buffer_flush,             75)     \
1517         x(trans_restart_split_race,                     76)
1518
1519 enum bch_persistent_counters {
1520 #define x(t, n, ...) BCH_COUNTER_##t,
1521         BCH_PERSISTENT_COUNTERS()
1522 #undef x
1523         BCH_COUNTER_NR
1524 };
1525
1526 struct bch_sb_field_counters {
1527         struct bch_sb_field     field;
1528         __le64                  d[0];
1529 };
1530
1531 /*
1532  * On clean shutdown, store btree roots and current journal sequence number in
1533  * the superblock:
1534  */
1535 struct jset_entry {
1536         __le16                  u64s;
1537         __u8                    btree_id;
1538         __u8                    level;
1539         __u8                    type; /* designates what this jset holds */
1540         __u8                    pad[3];
1541
1542         union {
1543                 struct bkey_i   start[0];
1544                 __u64           _data[0];
1545         };
1546 };
1547
1548 struct bch_sb_field_clean {
1549         struct bch_sb_field     field;
1550
1551         __le32                  flags;
1552         __le16                  _read_clock; /* no longer used */
1553         __le16                  _write_clock;
1554         __le64                  journal_seq;
1555
1556         union {
1557                 struct jset_entry start[0];
1558                 __u64           _data[0];
1559         };
1560 };
1561
1562 struct journal_seq_blacklist_entry {
1563         __le64                  start;
1564         __le64                  end;
1565 };
1566
1567 struct bch_sb_field_journal_seq_blacklist {
1568         struct bch_sb_field     field;
1569
1570         union {
1571                 struct journal_seq_blacklist_entry start[0];
1572                 __u64           _data[0];
1573         };
1574 };
1575
1576 /* Superblock: */
1577
1578 /*
1579  * New versioning scheme:
1580  * One common version number for all on disk data structures - superblock, btree
1581  * nodes, journal entries
1582  */
1583 #define BCH_VERSION_MAJOR(_v)           ((__u16) ((_v) >> 10))
1584 #define BCH_VERSION_MINOR(_v)           ((__u16) ((_v) & ~(~0U << 10)))
1585 #define BCH_VERSION(_major, _minor)     (((_major) << 10)|(_minor) << 0)
1586
1587 #define RECOVERY_PASS_ALL_FSCK          (1ULL << 63)
1588
1589 #define BCH_METADATA_VERSIONS()                                         \
1590         x(bkey_renumber,                BCH_VERSION(0, 10),             \
1591           RECOVERY_PASS_ALL_FSCK)                                       \
1592         x(inode_btree_change,           BCH_VERSION(0, 11),             \
1593           RECOVERY_PASS_ALL_FSCK)                                       \
1594         x(snapshot,                     BCH_VERSION(0, 12),             \
1595           RECOVERY_PASS_ALL_FSCK)                                       \
1596         x(inode_backpointers,           BCH_VERSION(0, 13),             \
1597           RECOVERY_PASS_ALL_FSCK)                                       \
1598         x(btree_ptr_sectors_written,    BCH_VERSION(0, 14),             \
1599           RECOVERY_PASS_ALL_FSCK)                                       \
1600         x(snapshot_2,                   BCH_VERSION(0, 15),             \
1601           BIT_ULL(BCH_RECOVERY_PASS_fs_upgrade_for_subvolumes)|         \
1602           BIT_ULL(BCH_RECOVERY_PASS_initialize_subvolumes)|             \
1603           RECOVERY_PASS_ALL_FSCK)                                       \
1604         x(reflink_p_fix,                BCH_VERSION(0, 16),             \
1605           BIT_ULL(BCH_RECOVERY_PASS_fix_reflink_p))                     \
1606         x(subvol_dirent,                BCH_VERSION(0, 17),             \
1607           RECOVERY_PASS_ALL_FSCK)                                       \
1608         x(inode_v2,                     BCH_VERSION(0, 18),             \
1609           RECOVERY_PASS_ALL_FSCK)                                       \
1610         x(freespace,                    BCH_VERSION(0, 19),             \
1611           RECOVERY_PASS_ALL_FSCK)                                       \
1612         x(alloc_v4,                     BCH_VERSION(0, 20),             \
1613           RECOVERY_PASS_ALL_FSCK)                                       \
1614         x(new_data_types,               BCH_VERSION(0, 21),             \
1615           RECOVERY_PASS_ALL_FSCK)                                       \
1616         x(backpointers,                 BCH_VERSION(0, 22),             \
1617           RECOVERY_PASS_ALL_FSCK)                                       \
1618         x(inode_v3,                     BCH_VERSION(0, 23),             \
1619           RECOVERY_PASS_ALL_FSCK)                                       \
1620         x(unwritten_extents,            BCH_VERSION(0, 24),             \
1621           RECOVERY_PASS_ALL_FSCK)                                       \
1622         x(bucket_gens,                  BCH_VERSION(0, 25),             \
1623           BIT_ULL(BCH_RECOVERY_PASS_bucket_gens_init)|                  \
1624           RECOVERY_PASS_ALL_FSCK)                                       \
1625         x(lru_v2,                       BCH_VERSION(0, 26),             \
1626           RECOVERY_PASS_ALL_FSCK)                                       \
1627         x(fragmentation_lru,            BCH_VERSION(0, 27),             \
1628           RECOVERY_PASS_ALL_FSCK)                                       \
1629         x(no_bps_in_alloc_keys,         BCH_VERSION(0, 28),             \
1630           RECOVERY_PASS_ALL_FSCK)                                       \
1631         x(snapshot_trees,               BCH_VERSION(0, 29),             \
1632           RECOVERY_PASS_ALL_FSCK)                                       \
1633         x(major_minor,                  BCH_VERSION(1,  0),             \
1634           0)                                                            \
1635         x(snapshot_skiplists,           BCH_VERSION(1,  1),             \
1636           BIT_ULL(BCH_RECOVERY_PASS_check_snapshots))                   \
1637         x(deleted_inodes,               BCH_VERSION(1,  2),             \
1638           BIT_ULL(BCH_RECOVERY_PASS_check_inodes))
1639
1640 enum bcachefs_metadata_version {
1641         bcachefs_metadata_version_min = 9,
1642 #define x(t, n, upgrade_passes) bcachefs_metadata_version_##t = n,
1643         BCH_METADATA_VERSIONS()
1644 #undef x
1645         bcachefs_metadata_version_max
1646 };
1647
1648 static const unsigned bcachefs_metadata_required_upgrade_below = bcachefs_metadata_version_major_minor;
1649
1650 #define bcachefs_metadata_version_current       (bcachefs_metadata_version_max - 1)
1651
1652 #define BCH_SB_SECTOR                   8
1653 #define BCH_SB_MEMBERS_MAX              64 /* XXX kill */
1654
1655 struct bch_sb_layout {
1656         __uuid_t                magic;  /* bcachefs superblock UUID */
1657         __u8                    layout_type;
1658         __u8                    sb_max_size_bits; /* base 2 of 512 byte sectors */
1659         __u8                    nr_superblocks;
1660         __u8                    pad[5];
1661         __le64                  sb_offset[61];
1662 } __packed __aligned(8);
1663
1664 #define BCH_SB_LAYOUT_SECTOR    7
1665
1666 /*
1667  * @offset      - sector where this sb was written
1668  * @version     - on disk format version
1669  * @version_min - Oldest metadata version this filesystem contains; so we can
1670  *                safely drop compatibility code and refuse to mount filesystems
1671  *                we'd need it for
1672  * @magic       - identifies as a bcachefs superblock (BCHFS_MAGIC)
1673  * @seq         - incremented each time superblock is written
1674  * @uuid        - used for generating various magic numbers and identifying
1675  *                member devices, never changes
1676  * @user_uuid   - user visible UUID, may be changed
1677  * @label       - filesystem label
1678  * @seq         - identifies most recent superblock, incremented each time
1679  *                superblock is written
1680  * @features    - enabled incompatible features
1681  */
1682 struct bch_sb {
1683         struct bch_csum         csum;
1684         __le16                  version;
1685         __le16                  version_min;
1686         __le16                  pad[2];
1687         __uuid_t                magic;
1688         __uuid_t                uuid;
1689         __uuid_t                user_uuid;
1690         __u8                    label[BCH_SB_LABEL_SIZE];
1691         __le64                  offset;
1692         __le64                  seq;
1693
1694         __le16                  block_size;
1695         __u8                    dev_idx;
1696         __u8                    nr_devices;
1697         __le32                  u64s;
1698
1699         __le64                  time_base_lo;
1700         __le32                  time_base_hi;
1701         __le32                  time_precision;
1702
1703         __le64                  flags[8];
1704         __le64                  features[2];
1705         __le64                  compat[2];
1706
1707         struct bch_sb_layout    layout;
1708
1709         union {
1710                 struct bch_sb_field start[0];
1711                 __le64          _data[0];
1712         };
1713 } __packed __aligned(8);
1714
1715 /*
1716  * Flags:
1717  * BCH_SB_INITALIZED    - set on first mount
1718  * BCH_SB_CLEAN         - did we shut down cleanly? Just a hint, doesn't affect
1719  *                        behaviour of mount/recovery path:
1720  * BCH_SB_INODE_32BIT   - limit inode numbers to 32 bits
1721  * BCH_SB_128_BIT_MACS  - 128 bit macs instead of 80
1722  * BCH_SB_ENCRYPTION_TYPE - if nonzero encryption is enabled; overrides
1723  *                         DATA/META_CSUM_TYPE. Also indicates encryption
1724  *                         algorithm in use, if/when we get more than one
1725  */
1726
1727 LE16_BITMASK(BCH_SB_BLOCK_SIZE,         struct bch_sb, block_size, 0, 16);
1728
1729 LE64_BITMASK(BCH_SB_INITIALIZED,        struct bch_sb, flags[0],  0,  1);
1730 LE64_BITMASK(BCH_SB_CLEAN,              struct bch_sb, flags[0],  1,  2);
1731 LE64_BITMASK(BCH_SB_CSUM_TYPE,          struct bch_sb, flags[0],  2,  8);
1732 LE64_BITMASK(BCH_SB_ERROR_ACTION,       struct bch_sb, flags[0],  8, 12);
1733
1734 LE64_BITMASK(BCH_SB_BTREE_NODE_SIZE,    struct bch_sb, flags[0], 12, 28);
1735
1736 LE64_BITMASK(BCH_SB_GC_RESERVE,         struct bch_sb, flags[0], 28, 33);
1737 LE64_BITMASK(BCH_SB_ROOT_RESERVE,       struct bch_sb, flags[0], 33, 40);
1738
1739 LE64_BITMASK(BCH_SB_META_CSUM_TYPE,     struct bch_sb, flags[0], 40, 44);
1740 LE64_BITMASK(BCH_SB_DATA_CSUM_TYPE,     struct bch_sb, flags[0], 44, 48);
1741
1742 LE64_BITMASK(BCH_SB_META_REPLICAS_WANT, struct bch_sb, flags[0], 48, 52);
1743 LE64_BITMASK(BCH_SB_DATA_REPLICAS_WANT, struct bch_sb, flags[0], 52, 56);
1744
1745 LE64_BITMASK(BCH_SB_POSIX_ACL,          struct bch_sb, flags[0], 56, 57);
1746 LE64_BITMASK(BCH_SB_USRQUOTA,           struct bch_sb, flags[0], 57, 58);
1747 LE64_BITMASK(BCH_SB_GRPQUOTA,           struct bch_sb, flags[0], 58, 59);
1748 LE64_BITMASK(BCH_SB_PRJQUOTA,           struct bch_sb, flags[0], 59, 60);
1749
1750 LE64_BITMASK(BCH_SB_HAS_ERRORS,         struct bch_sb, flags[0], 60, 61);
1751 LE64_BITMASK(BCH_SB_HAS_TOPOLOGY_ERRORS,struct bch_sb, flags[0], 61, 62);
1752
1753 LE64_BITMASK(BCH_SB_BIG_ENDIAN,         struct bch_sb, flags[0], 62, 63);
1754
1755 LE64_BITMASK(BCH_SB_STR_HASH_TYPE,      struct bch_sb, flags[1],  0,  4);
1756 LE64_BITMASK(BCH_SB_COMPRESSION_TYPE_LO,struct bch_sb, flags[1],  4,  8);
1757 LE64_BITMASK(BCH_SB_INODE_32BIT,        struct bch_sb, flags[1],  8,  9);
1758
1759 LE64_BITMASK(BCH_SB_128_BIT_MACS,       struct bch_sb, flags[1],  9, 10);
1760 LE64_BITMASK(BCH_SB_ENCRYPTION_TYPE,    struct bch_sb, flags[1], 10, 14);
1761
1762 /*
1763  * Max size of an extent that may require bouncing to read or write
1764  * (checksummed, compressed): 64k
1765  */
1766 LE64_BITMASK(BCH_SB_ENCODED_EXTENT_MAX_BITS,
1767                                         struct bch_sb, flags[1], 14, 20);
1768
1769 LE64_BITMASK(BCH_SB_META_REPLICAS_REQ,  struct bch_sb, flags[1], 20, 24);
1770 LE64_BITMASK(BCH_SB_DATA_REPLICAS_REQ,  struct bch_sb, flags[1], 24, 28);
1771
1772 LE64_BITMASK(BCH_SB_PROMOTE_TARGET,     struct bch_sb, flags[1], 28, 40);
1773 LE64_BITMASK(BCH_SB_FOREGROUND_TARGET,  struct bch_sb, flags[1], 40, 52);
1774 LE64_BITMASK(BCH_SB_BACKGROUND_TARGET,  struct bch_sb, flags[1], 52, 64);
1775
1776 LE64_BITMASK(BCH_SB_BACKGROUND_COMPRESSION_TYPE_LO,
1777                                         struct bch_sb, flags[2],  0,  4);
1778 LE64_BITMASK(BCH_SB_GC_RESERVE_BYTES,   struct bch_sb, flags[2],  4, 64);
1779
1780 LE64_BITMASK(BCH_SB_ERASURE_CODE,       struct bch_sb, flags[3],  0, 16);
1781 LE64_BITMASK(BCH_SB_METADATA_TARGET,    struct bch_sb, flags[3], 16, 28);
1782 LE64_BITMASK(BCH_SB_SHARD_INUMS,        struct bch_sb, flags[3], 28, 29);
1783 LE64_BITMASK(BCH_SB_INODES_USE_KEY_CACHE,struct bch_sb, flags[3], 29, 30);
1784 LE64_BITMASK(BCH_SB_JOURNAL_FLUSH_DELAY,struct bch_sb, flags[3], 30, 62);
1785 LE64_BITMASK(BCH_SB_JOURNAL_FLUSH_DISABLED,struct bch_sb, flags[3], 62, 63);
1786 LE64_BITMASK(BCH_SB_JOURNAL_RECLAIM_DELAY,struct bch_sb, flags[4], 0, 32);
1787 LE64_BITMASK(BCH_SB_JOURNAL_TRANSACTION_NAMES,struct bch_sb, flags[4], 32, 33);
1788 LE64_BITMASK(BCH_SB_NOCOW,              struct bch_sb, flags[4], 33, 34);
1789 LE64_BITMASK(BCH_SB_WRITE_BUFFER_SIZE,  struct bch_sb, flags[4], 34, 54);
1790 LE64_BITMASK(BCH_SB_VERSION_UPGRADE,    struct bch_sb, flags[4], 54, 56);
1791
1792 LE64_BITMASK(BCH_SB_COMPRESSION_TYPE_HI,struct bch_sb, flags[4], 56, 60);
1793 LE64_BITMASK(BCH_SB_BACKGROUND_COMPRESSION_TYPE_HI,
1794                                         struct bch_sb, flags[4], 60, 64);
1795
1796 LE64_BITMASK(BCH_SB_VERSION_UPGRADE_COMPLETE,
1797                                         struct bch_sb, flags[5],  0, 16);
1798
1799 static inline __u64 BCH_SB_COMPRESSION_TYPE(const struct bch_sb *sb)
1800 {
1801         return BCH_SB_COMPRESSION_TYPE_LO(sb) | (BCH_SB_COMPRESSION_TYPE_HI(sb) << 4);
1802 }
1803
1804 static inline void SET_BCH_SB_COMPRESSION_TYPE(struct bch_sb *sb, __u64 v)
1805 {
1806         SET_BCH_SB_COMPRESSION_TYPE_LO(sb, v);
1807         SET_BCH_SB_COMPRESSION_TYPE_HI(sb, v >> 4);
1808 }
1809
1810 static inline __u64 BCH_SB_BACKGROUND_COMPRESSION_TYPE(const struct bch_sb *sb)
1811 {
1812         return BCH_SB_BACKGROUND_COMPRESSION_TYPE_LO(sb) |
1813                 (BCH_SB_BACKGROUND_COMPRESSION_TYPE_HI(sb) << 4);
1814 }
1815
1816 static inline void SET_BCH_SB_BACKGROUND_COMPRESSION_TYPE(struct bch_sb *sb, __u64 v)
1817 {
1818         SET_BCH_SB_BACKGROUND_COMPRESSION_TYPE_LO(sb, v);
1819         SET_BCH_SB_BACKGROUND_COMPRESSION_TYPE_HI(sb, v >> 4);
1820 }
1821
1822 /*
1823  * Features:
1824  *
1825  * journal_seq_blacklist_v3:    gates BCH_SB_FIELD_journal_seq_blacklist
1826  * reflink:                     gates KEY_TYPE_reflink
1827  * inline_data:                 gates KEY_TYPE_inline_data
1828  * new_siphash:                 gates BCH_STR_HASH_siphash
1829  * new_extent_overwrite:        gates BTREE_NODE_NEW_EXTENT_OVERWRITE
1830  */
1831 #define BCH_SB_FEATURES()                       \
1832         x(lz4,                          0)      \
1833         x(gzip,                         1)      \
1834         x(zstd,                         2)      \
1835         x(atomic_nlink,                 3)      \
1836         x(ec,                           4)      \
1837         x(journal_seq_blacklist_v3,     5)      \
1838         x(reflink,                      6)      \
1839         x(new_siphash,                  7)      \
1840         x(inline_data,                  8)      \
1841         x(new_extent_overwrite,         9)      \
1842         x(incompressible,               10)     \
1843         x(btree_ptr_v2,                 11)     \
1844         x(extents_above_btree_updates,  12)     \
1845         x(btree_updates_journalled,     13)     \
1846         x(reflink_inline_data,          14)     \
1847         x(new_varint,                   15)     \
1848         x(journal_no_flush,             16)     \
1849         x(alloc_v2,                     17)     \
1850         x(extents_across_btree_nodes,   18)
1851
1852 #define BCH_SB_FEATURES_ALWAYS                          \
1853         ((1ULL << BCH_FEATURE_new_extent_overwrite)|    \
1854          (1ULL << BCH_FEATURE_extents_above_btree_updates)|\
1855          (1ULL << BCH_FEATURE_btree_updates_journalled)|\
1856          (1ULL << BCH_FEATURE_alloc_v2)|\
1857          (1ULL << BCH_FEATURE_extents_across_btree_nodes))
1858
1859 #define BCH_SB_FEATURES_ALL                             \
1860         (BCH_SB_FEATURES_ALWAYS|                        \
1861          (1ULL << BCH_FEATURE_new_siphash)|             \
1862          (1ULL << BCH_FEATURE_btree_ptr_v2)|            \
1863          (1ULL << BCH_FEATURE_new_varint)|              \
1864          (1ULL << BCH_FEATURE_journal_no_flush))
1865
1866 enum bch_sb_feature {
1867 #define x(f, n) BCH_FEATURE_##f,
1868         BCH_SB_FEATURES()
1869 #undef x
1870         BCH_FEATURE_NR,
1871 };
1872
1873 #define BCH_SB_COMPAT()                                 \
1874         x(alloc_info,                           0)      \
1875         x(alloc_metadata,                       1)      \
1876         x(extents_above_btree_updates_done,     2)      \
1877         x(bformat_overflow_done,                3)
1878
1879 enum bch_sb_compat {
1880 #define x(f, n) BCH_COMPAT_##f,
1881         BCH_SB_COMPAT()
1882 #undef x
1883         BCH_COMPAT_NR,
1884 };
1885
1886 /* options: */
1887
1888 #define BCH_VERSION_UPGRADE_OPTS()      \
1889         x(compatible,           0)      \
1890         x(incompatible,         1)      \
1891         x(none,                 2)
1892
1893 enum bch_version_upgrade_opts {
1894 #define x(t, n) BCH_VERSION_UPGRADE_##t = n,
1895         BCH_VERSION_UPGRADE_OPTS()
1896 #undef x
1897 };
1898
1899 #define BCH_REPLICAS_MAX                4U
1900
1901 #define BCH_BKEY_PTRS_MAX               16U
1902
1903 #define BCH_ERROR_ACTIONS()             \
1904         x(continue,             0)      \
1905         x(ro,                   1)      \
1906         x(panic,                2)
1907
1908 enum bch_error_actions {
1909 #define x(t, n) BCH_ON_ERROR_##t = n,
1910         BCH_ERROR_ACTIONS()
1911 #undef x
1912         BCH_ON_ERROR_NR
1913 };
1914
1915 #define BCH_STR_HASH_TYPES()            \
1916         x(crc32c,               0)      \
1917         x(crc64,                1)      \
1918         x(siphash_old,          2)      \
1919         x(siphash,              3)
1920
1921 enum bch_str_hash_type {
1922 #define x(t, n) BCH_STR_HASH_##t = n,
1923         BCH_STR_HASH_TYPES()
1924 #undef x
1925         BCH_STR_HASH_NR
1926 };
1927
1928 #define BCH_STR_HASH_OPTS()             \
1929         x(crc32c,               0)      \
1930         x(crc64,                1)      \
1931         x(siphash,              2)
1932
1933 enum bch_str_hash_opts {
1934 #define x(t, n) BCH_STR_HASH_OPT_##t = n,
1935         BCH_STR_HASH_OPTS()
1936 #undef x
1937         BCH_STR_HASH_OPT_NR
1938 };
1939
1940 #define BCH_CSUM_TYPES()                        \
1941         x(none,                         0)      \
1942         x(crc32c_nonzero,               1)      \
1943         x(crc64_nonzero,                2)      \
1944         x(chacha20_poly1305_80,         3)      \
1945         x(chacha20_poly1305_128,        4)      \
1946         x(crc32c,                       5)      \
1947         x(crc64,                        6)      \
1948         x(xxhash,                       7)
1949
1950 enum bch_csum_type {
1951 #define x(t, n) BCH_CSUM_##t = n,
1952         BCH_CSUM_TYPES()
1953 #undef x
1954         BCH_CSUM_NR
1955 };
1956
1957 static const unsigned bch_crc_bytes[] = {
1958         [BCH_CSUM_none]                         = 0,
1959         [BCH_CSUM_crc32c_nonzero]               = 4,
1960         [BCH_CSUM_crc32c]                       = 4,
1961         [BCH_CSUM_crc64_nonzero]                = 8,
1962         [BCH_CSUM_crc64]                        = 8,
1963         [BCH_CSUM_xxhash]                       = 8,
1964         [BCH_CSUM_chacha20_poly1305_80]         = 10,
1965         [BCH_CSUM_chacha20_poly1305_128]        = 16,
1966 };
1967
1968 static inline _Bool bch2_csum_type_is_encryption(enum bch_csum_type type)
1969 {
1970         switch (type) {
1971         case BCH_CSUM_chacha20_poly1305_80:
1972         case BCH_CSUM_chacha20_poly1305_128:
1973                 return true;
1974         default:
1975                 return false;
1976         }
1977 }
1978
1979 #define BCH_CSUM_OPTS()                 \
1980         x(none,                 0)      \
1981         x(crc32c,               1)      \
1982         x(crc64,                2)      \
1983         x(xxhash,               3)
1984
1985 enum bch_csum_opts {
1986 #define x(t, n) BCH_CSUM_OPT_##t = n,
1987         BCH_CSUM_OPTS()
1988 #undef x
1989         BCH_CSUM_OPT_NR
1990 };
1991
1992 #define BCH_COMPRESSION_TYPES()         \
1993         x(none,                 0)      \
1994         x(lz4_old,              1)      \
1995         x(gzip,                 2)      \
1996         x(lz4,                  3)      \
1997         x(zstd,                 4)      \
1998         x(incompressible,       5)
1999
2000 enum bch_compression_type {
2001 #define x(t, n) BCH_COMPRESSION_TYPE_##t = n,
2002         BCH_COMPRESSION_TYPES()
2003 #undef x
2004         BCH_COMPRESSION_TYPE_NR
2005 };
2006
2007 #define BCH_COMPRESSION_OPTS()          \
2008         x(none,         0)              \
2009         x(lz4,          1)              \
2010         x(gzip,         2)              \
2011         x(zstd,         3)
2012
2013 enum bch_compression_opts {
2014 #define x(t, n) BCH_COMPRESSION_OPT_##t = n,
2015         BCH_COMPRESSION_OPTS()
2016 #undef x
2017         BCH_COMPRESSION_OPT_NR
2018 };
2019
2020 /*
2021  * Magic numbers
2022  *
2023  * The various other data structures have their own magic numbers, which are
2024  * xored with the first part of the cache set's UUID
2025  */
2026
2027 #define BCACHE_MAGIC                                                    \
2028         UUID_INIT(0xc68573f6, 0x4e1a, 0x45ca,                           \
2029                   0x82, 0x65, 0xf5, 0x7f, 0x48, 0xba, 0x6d, 0x81)
2030 #define BCHFS_MAGIC                                                     \
2031         UUID_INIT(0xc68573f6, 0x66ce, 0x90a9,                           \
2032                   0xd9, 0x6a, 0x60, 0xcf, 0x80, 0x3d, 0xf7, 0xef)
2033
2034 #define BCACHEFS_STATFS_MAGIC           0xca451a4e
2035
2036 #define JSET_MAGIC              __cpu_to_le64(0x245235c1a3625032ULL)
2037 #define BSET_MAGIC              __cpu_to_le64(0x90135c78b99e07f5ULL)
2038
2039 static inline __le64 __bch2_sb_magic(struct bch_sb *sb)
2040 {
2041         __le64 ret;
2042
2043         memcpy(&ret, &sb->uuid, sizeof(ret));
2044         return ret;
2045 }
2046
2047 static inline __u64 __jset_magic(struct bch_sb *sb)
2048 {
2049         return __le64_to_cpu(__bch2_sb_magic(sb) ^ JSET_MAGIC);
2050 }
2051
2052 static inline __u64 __bset_magic(struct bch_sb *sb)
2053 {
2054         return __le64_to_cpu(__bch2_sb_magic(sb) ^ BSET_MAGIC);
2055 }
2056
2057 /* Journal */
2058
2059 #define JSET_KEYS_U64s  (sizeof(struct jset_entry) / sizeof(__u64))
2060
2061 #define BCH_JSET_ENTRY_TYPES()                  \
2062         x(btree_keys,           0)              \
2063         x(btree_root,           1)              \
2064         x(prio_ptrs,            2)              \
2065         x(blacklist,            3)              \
2066         x(blacklist_v2,         4)              \
2067         x(usage,                5)              \
2068         x(data_usage,           6)              \
2069         x(clock,                7)              \
2070         x(dev_usage,            8)              \
2071         x(log,                  9)              \
2072         x(overwrite,            10)
2073
2074 enum {
2075 #define x(f, nr)        BCH_JSET_ENTRY_##f      = nr,
2076         BCH_JSET_ENTRY_TYPES()
2077 #undef x
2078         BCH_JSET_ENTRY_NR
2079 };
2080
2081 /*
2082  * Journal sequence numbers can be blacklisted: bsets record the max sequence
2083  * number of all the journal entries they contain updates for, so that on
2084  * recovery we can ignore those bsets that contain index updates newer that what
2085  * made it into the journal.
2086  *
2087  * This means that we can't reuse that journal_seq - we have to skip it, and
2088  * then record that we skipped it so that the next time we crash and recover we
2089  * don't think there was a missing journal entry.
2090  */
2091 struct jset_entry_blacklist {
2092         struct jset_entry       entry;
2093         __le64                  seq;
2094 };
2095
2096 struct jset_entry_blacklist_v2 {
2097         struct jset_entry       entry;
2098         __le64                  start;
2099         __le64                  end;
2100 };
2101
2102 #define BCH_FS_USAGE_TYPES()                    \
2103         x(reserved,             0)              \
2104         x(inodes,               1)              \
2105         x(key_version,          2)
2106
2107 enum {
2108 #define x(f, nr)        BCH_FS_USAGE_##f        = nr,
2109         BCH_FS_USAGE_TYPES()
2110 #undef x
2111         BCH_FS_USAGE_NR
2112 };
2113
2114 struct jset_entry_usage {
2115         struct jset_entry       entry;
2116         __le64                  v;
2117 } __packed;
2118
2119 struct jset_entry_data_usage {
2120         struct jset_entry       entry;
2121         __le64                  v;
2122         struct bch_replicas_entry r;
2123 } __packed;
2124
2125 struct jset_entry_clock {
2126         struct jset_entry       entry;
2127         __u8                    rw;
2128         __u8                    pad[7];
2129         __le64                  time;
2130 } __packed;
2131
2132 struct jset_entry_dev_usage_type {
2133         __le64                  buckets;
2134         __le64                  sectors;
2135         __le64                  fragmented;
2136 } __packed;
2137
2138 struct jset_entry_dev_usage {
2139         struct jset_entry       entry;
2140         __le32                  dev;
2141         __u32                   pad;
2142
2143         __le64                  buckets_ec;
2144         __le64                  _buckets_unavailable; /* No longer used */
2145
2146         struct jset_entry_dev_usage_type d[];
2147 };
2148
2149 static inline unsigned jset_entry_dev_usage_nr_types(struct jset_entry_dev_usage *u)
2150 {
2151         return (vstruct_bytes(&u->entry) - sizeof(struct jset_entry_dev_usage)) /
2152                 sizeof(struct jset_entry_dev_usage_type);
2153 }
2154
2155 struct jset_entry_log {
2156         struct jset_entry       entry;
2157         u8                      d[];
2158 } __packed;
2159
2160 /*
2161  * On disk format for a journal entry:
2162  * seq is monotonically increasing; every journal entry has its own unique
2163  * sequence number.
2164  *
2165  * last_seq is the oldest journal entry that still has keys the btree hasn't
2166  * flushed to disk yet.
2167  *
2168  * version is for on disk format changes.
2169  */
2170 struct jset {
2171         struct bch_csum         csum;
2172
2173         __le64                  magic;
2174         __le64                  seq;
2175         __le32                  version;
2176         __le32                  flags;
2177
2178         __le32                  u64s; /* size of d[] in u64s */
2179
2180         __u8                    encrypted_start[0];
2181
2182         __le16                  _read_clock; /* no longer used */
2183         __le16                  _write_clock;
2184
2185         /* Sequence number of oldest dirty journal entry */
2186         __le64                  last_seq;
2187
2188
2189         union {
2190                 struct jset_entry start[0];
2191                 __u64           _data[0];
2192         };
2193 } __packed __aligned(8);
2194
2195 LE32_BITMASK(JSET_CSUM_TYPE,    struct jset, flags, 0, 4);
2196 LE32_BITMASK(JSET_BIG_ENDIAN,   struct jset, flags, 4, 5);
2197 LE32_BITMASK(JSET_NO_FLUSH,     struct jset, flags, 5, 6);
2198
2199 #define BCH_JOURNAL_BUCKETS_MIN         8
2200
2201 /* Btree: */
2202
2203 enum btree_id_flags {
2204         BTREE_ID_EXTENTS        = BIT(0),
2205         BTREE_ID_SNAPSHOTS      = BIT(1),
2206         BTREE_ID_DATA           = BIT(2),
2207 };
2208
2209 #define BCH_BTREE_IDS()                                                         \
2210         x(extents,              0,      BTREE_ID_EXTENTS|BTREE_ID_SNAPSHOTS|BTREE_ID_DATA,\
2211           BIT_ULL(KEY_TYPE_whiteout)|                                           \
2212           BIT_ULL(KEY_TYPE_error)|                                              \
2213           BIT_ULL(KEY_TYPE_cookie)|                                             \
2214           BIT_ULL(KEY_TYPE_extent)|                                             \
2215           BIT_ULL(KEY_TYPE_reservation)|                                        \
2216           BIT_ULL(KEY_TYPE_reflink_p)|                                          \
2217           BIT_ULL(KEY_TYPE_inline_data))                                        \
2218         x(inodes,               1,      BTREE_ID_SNAPSHOTS,                     \
2219           BIT_ULL(KEY_TYPE_whiteout)|                                           \
2220           BIT_ULL(KEY_TYPE_inode)|                                              \
2221           BIT_ULL(KEY_TYPE_inode_v2)|                                           \
2222           BIT_ULL(KEY_TYPE_inode_v3)|                                           \
2223           BIT_ULL(KEY_TYPE_inode_generation))                                   \
2224         x(dirents,              2,      BTREE_ID_SNAPSHOTS,                     \
2225           BIT_ULL(KEY_TYPE_whiteout)|                                           \
2226           BIT_ULL(KEY_TYPE_hash_whiteout)|                                      \
2227           BIT_ULL(KEY_TYPE_dirent))                                             \
2228         x(xattrs,               3,      BTREE_ID_SNAPSHOTS,                     \
2229           BIT_ULL(KEY_TYPE_whiteout)|                                           \
2230           BIT_ULL(KEY_TYPE_cookie)|                                             \
2231           BIT_ULL(KEY_TYPE_hash_whiteout)|                                      \
2232           BIT_ULL(KEY_TYPE_xattr))                                              \
2233         x(alloc,                4,      0,                                      \
2234           BIT_ULL(KEY_TYPE_alloc)|                                              \
2235           BIT_ULL(KEY_TYPE_alloc_v2)|                                           \
2236           BIT_ULL(KEY_TYPE_alloc_v3)|                                           \
2237           BIT_ULL(KEY_TYPE_alloc_v4))                                           \
2238         x(quotas,               5,      0,                                      \
2239           BIT_ULL(KEY_TYPE_quota))                                              \
2240         x(stripes,              6,      0,                                      \
2241           BIT_ULL(KEY_TYPE_stripe))                                             \
2242         x(reflink,              7,      BTREE_ID_EXTENTS|BTREE_ID_DATA,         \
2243           BIT_ULL(KEY_TYPE_reflink_v)|                                          \
2244           BIT_ULL(KEY_TYPE_indirect_inline_data))                               \
2245         x(subvolumes,           8,      0,                                      \
2246           BIT_ULL(KEY_TYPE_subvolume))                                          \
2247         x(snapshots,            9,      0,                                      \
2248           BIT_ULL(KEY_TYPE_snapshot))                                           \
2249         x(lru,                  10,     0,                                      \
2250           BIT_ULL(KEY_TYPE_set))                                                \
2251         x(freespace,            11,     BTREE_ID_EXTENTS,                       \
2252           BIT_ULL(KEY_TYPE_set))                                                \
2253         x(need_discard,         12,     0,                                      \
2254           BIT_ULL(KEY_TYPE_set))                                                \
2255         x(backpointers,         13,     0,                                      \
2256           BIT_ULL(KEY_TYPE_backpointer))                                        \
2257         x(bucket_gens,          14,     0,                                      \
2258           BIT_ULL(KEY_TYPE_bucket_gens))                                        \
2259         x(snapshot_trees,       15,     0,                                      \
2260           BIT_ULL(KEY_TYPE_snapshot_tree))                                      \
2261         x(deleted_inodes,       16,     BTREE_ID_SNAPSHOTS,                     \
2262           BIT_ULL(KEY_TYPE_set))
2263
2264 enum btree_id {
2265 #define x(name, nr, ...) BTREE_ID_##name = nr,
2266         BCH_BTREE_IDS()
2267 #undef x
2268         BTREE_ID_NR
2269 };
2270
2271 #define BTREE_MAX_DEPTH         4U
2272
2273 /* Btree nodes */
2274
2275 /*
2276  * Btree nodes
2277  *
2278  * On disk a btree node is a list/log of these; within each set the keys are
2279  * sorted
2280  */
2281 struct bset {
2282         __le64                  seq;
2283
2284         /*
2285          * Highest journal entry this bset contains keys for.
2286          * If on recovery we don't see that journal entry, this bset is ignored:
2287          * this allows us to preserve the order of all index updates after a
2288          * crash, since the journal records a total order of all index updates
2289          * and anything that didn't make it to the journal doesn't get used.
2290          */
2291         __le64                  journal_seq;
2292
2293         __le32                  flags;
2294         __le16                  version;
2295         __le16                  u64s; /* count of d[] in u64s */
2296
2297         union {
2298                 struct bkey_packed start[0];
2299                 __u64           _data[0];
2300         };
2301 } __packed __aligned(8);
2302
2303 LE32_BITMASK(BSET_CSUM_TYPE,    struct bset, flags, 0, 4);
2304
2305 LE32_BITMASK(BSET_BIG_ENDIAN,   struct bset, flags, 4, 5);
2306 LE32_BITMASK(BSET_SEPARATE_WHITEOUTS,
2307                                 struct bset, flags, 5, 6);
2308
2309 /* Sector offset within the btree node: */
2310 LE32_BITMASK(BSET_OFFSET,       struct bset, flags, 16, 32);
2311
2312 struct btree_node {
2313         struct bch_csum         csum;
2314         __le64                  magic;
2315
2316         /* this flags field is encrypted, unlike bset->flags: */
2317         __le64                  flags;
2318
2319         /* Closed interval: */
2320         struct bpos             min_key;
2321         struct bpos             max_key;
2322         struct bch_extent_ptr   _ptr; /* not used anymore */
2323         struct bkey_format      format;
2324
2325         union {
2326         struct bset             keys;
2327         struct {
2328                 __u8            pad[22];
2329                 __le16          u64s;
2330                 __u64           _data[0];
2331
2332         };
2333         };
2334 } __packed __aligned(8);
2335
2336 LE64_BITMASK(BTREE_NODE_ID_LO,  struct btree_node, flags,  0,  4);
2337 LE64_BITMASK(BTREE_NODE_LEVEL,  struct btree_node, flags,  4,  8);
2338 LE64_BITMASK(BTREE_NODE_NEW_EXTENT_OVERWRITE,
2339                                 struct btree_node, flags,  8,  9);
2340 LE64_BITMASK(BTREE_NODE_ID_HI,  struct btree_node, flags,  9, 25);
2341 /* 25-32 unused */
2342 LE64_BITMASK(BTREE_NODE_SEQ,    struct btree_node, flags, 32, 64);
2343
2344 static inline __u64 BTREE_NODE_ID(struct btree_node *n)
2345 {
2346         return BTREE_NODE_ID_LO(n) | (BTREE_NODE_ID_HI(n) << 4);
2347 }
2348
2349 static inline void SET_BTREE_NODE_ID(struct btree_node *n, __u64 v)
2350 {
2351         SET_BTREE_NODE_ID_LO(n, v);
2352         SET_BTREE_NODE_ID_HI(n, v >> 4);
2353 }
2354
2355 struct btree_node_entry {
2356         struct bch_csum         csum;
2357
2358         union {
2359         struct bset             keys;
2360         struct {
2361                 __u8            pad[22];
2362                 __le16          u64s;
2363                 __u64           _data[0];
2364         };
2365         };
2366 } __packed __aligned(8);
2367
2368 #endif /* _BCACHEFS_FORMAT_H */