1 /* SPDX-License-Identifier: GPL-2.0 */
2 #ifndef _BCACHEFS_FORMAT_H
3 #define _BCACHEFS_FORMAT_H
6 * bcachefs on disk data structures
10 * There are three main types of on disk data structures in bcachefs (this is
11 * reduced from 5 in bcache)
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.
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
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.
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.
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.
42 * The journal also contains entry types for the btree roots, and blacklisted
43 * journal sequence numbers (see journal_seq_blacklist.c).
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
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
58 * BTREE KEYS (struct bkey):
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.
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).
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.
75 #include <asm/types.h>
76 #include <asm/byteorder.h>
77 #include <linux/kernel.h>
78 #include <linux/uuid.h>
80 #define LE_BITMASK(_bits, name, type, field, offset, end) \
81 static const unsigned name##_OFFSET = offset; \
82 static const unsigned name##_BITS = (end - offset); \
83 static const __u##_bits name##_MAX = (1ULL << (end - offset)) - 1; \
85 static inline __u64 name(const type *k) \
87 return (__le##_bits##_to_cpu(k->field) >> offset) & \
88 ~(~0ULL << (end - offset)); \
91 static inline void SET_##name(type *k, __u64 v) \
93 __u##_bits new = __le##_bits##_to_cpu(k->field); \
95 new &= ~(~(~0ULL << (end - offset)) << offset); \
96 new |= (v & ~(~0ULL << (end - offset))) << offset; \
97 k->field = __cpu_to_le##_bits(new); \
100 #define LE16_BITMASK(n, t, f, o, e) LE_BITMASK(16, n, t, f, o, e)
101 #define LE32_BITMASK(n, t, f, o, e) LE_BITMASK(32, n, t, f, o, e)
102 #define LE64_BITMASK(n, t, f, o, e) LE_BITMASK(64, n, t, f, o, e)
107 /* One unused slot for now: */
108 __u8 bits_per_field[6];
109 __le64 field_offset[6];
112 /* Btree keys - all units are in sectors */
116 * Word order matches machine byte order - btree code treats a bpos as a
117 * single large integer, for search/comparison purposes
119 * Note that wherever a bpos is embedded in another on disk data
120 * structure, it has to be byte swabbed when reading in metadata that
121 * wasn't written in native endian order:
123 #if __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__
127 #elif __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__
129 __u64 offset; /* Points to end of extent - sectors */
132 #error edit for your odd byteorder.
134 } __attribute__((packed, aligned(4)));
136 #define KEY_INODE_MAX ((__u64)~0ULL)
137 #define KEY_OFFSET_MAX ((__u64)~0ULL)
138 #define KEY_SNAPSHOT_MAX ((__u32)~0U)
139 #define KEY_SIZE_MAX ((__u32)~0U)
141 static inline struct bpos SPOS(__u64 inode, __u64 offset, __u32 snapshot)
143 return (struct bpos) {
146 .snapshot = snapshot,
150 #define POS_MIN SPOS(0, 0, 0)
151 #define POS_MAX SPOS(KEY_INODE_MAX, KEY_OFFSET_MAX, KEY_SNAPSHOT_MAX)
152 #define POS(_inode, _offset) SPOS(_inode, _offset, 0)
154 /* Empty placeholder struct, for container_of() */
160 #if __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__
163 #elif __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__
167 } __attribute__((packed, aligned(4)));
170 /* Size of combined key and value, in u64s */
173 /* Format of key (0 for format local to btree node) */
174 #if defined(__LITTLE_ENDIAN_BITFIELD)
177 #elif defined (__BIG_ENDIAN_BITFIELD)
178 __u8 needs_whiteout:1,
181 #error edit for your odd byteorder.
184 /* Type of the value */
187 #if __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__
190 struct bversion version;
191 __u32 size; /* extent size, in sectors */
193 #elif __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__
195 __u32 size; /* extent size, in sectors */
196 struct bversion version;
200 } __attribute__((packed, aligned(8)));
205 /* Size of combined key and value, in u64s */
208 /* Format of key (0 for format local to btree node) */
211 * XXX: next incompat on disk format change, switch format and
212 * needs_whiteout - bkey_packed() will be cheaper if format is the high
213 * bits of the bitfield
215 #if defined(__LITTLE_ENDIAN_BITFIELD)
218 #elif defined (__BIG_ENDIAN_BITFIELD)
219 __u8 needs_whiteout:1,
223 /* Type of the value */
228 * We copy bkeys with struct assignment in various places, and while
229 * that shouldn't be done with packed bkeys we can't disallow it in C,
230 * and it's legal to cast a bkey to a bkey_packed - so padding it out
231 * to the same size as struct bkey should hopefully be safest.
233 __u8 pad[sizeof(struct bkey) - 3];
234 } __attribute__((packed, aligned(8)));
236 #define BKEY_U64s (sizeof(struct bkey) / sizeof(__u64))
237 #define BKEY_U64s_MAX U8_MAX
238 #define BKEY_VAL_U64s_MAX (BKEY_U64s_MAX - BKEY_U64s)
240 #define KEY_PACKED_BITS_START 24
242 #define KEY_FORMAT_LOCAL_BTREE 0
243 #define KEY_FORMAT_CURRENT 1
245 enum bch_bkey_fields {
250 BKEY_FIELD_VERSION_HI,
251 BKEY_FIELD_VERSION_LO,
255 #define bkey_format_field(name, field) \
256 [BKEY_FIELD_##name] = (sizeof(((struct bkey *) NULL)->field) * 8)
258 #define BKEY_FORMAT_CURRENT \
259 ((struct bkey_format) { \
260 .key_u64s = BKEY_U64s, \
261 .nr_fields = BKEY_NR_FIELDS, \
262 .bits_per_field = { \
263 bkey_format_field(INODE, p.inode), \
264 bkey_format_field(OFFSET, p.offset), \
265 bkey_format_field(SNAPSHOT, p.snapshot), \
266 bkey_format_field(SIZE, size), \
267 bkey_format_field(VERSION_HI, version.hi), \
268 bkey_format_field(VERSION_LO, version.lo), \
272 /* bkey with inline value */
278 /* Size of combined key and value, in u64s */
288 #define KEY(_inode, _offset, _size) \
291 .format = KEY_FORMAT_CURRENT, \
292 .p = POS(_inode, _offset), \
296 static inline void bkey_init(struct bkey *k)
301 #define bkey_bytes(_k) ((_k)->u64s * sizeof(__u64))
303 #define __BKEY_PADDED(key, pad) \
304 struct { struct bkey_i key; __u64 key ## _pad[pad]; }
307 * - DELETED keys are used internally to mark keys that should be ignored but
308 * override keys in composition order. Their version number is ignored.
310 * - DISCARDED keys indicate that the data is all 0s because it has been
311 * discarded. DISCARDs may have a version; if the version is nonzero the key
312 * will be persistent, otherwise the key will be dropped whenever the btree
313 * node is rewritten (like DELETED keys).
315 * - ERROR: any read of the data returns a read error, as the data was lost due
316 * to a failing device. Like DISCARDED keys, they can be removed (overridden)
317 * by new writes or cluster-wide GC. Node repair can also overwrite them with
318 * the same or a more recent version number, but not with an older version
321 * - WHITEOUT: for hash table btrees
323 #define BCH_BKEY_TYPES() \
328 x(hash_whiteout, 4) \
333 x(inode_generation, 9) \
342 x(btree_ptr_v2, 18) \
343 x(indirect_inline_data, 19) \
347 #define x(name, nr) KEY_TYPE_##name = nr,
370 struct bch_hash_whiteout {
377 * In extent bkeys, the value is a list of pointers (bch_extent_ptr), optionally
378 * preceded by checksum/compression information (bch_extent_crc32 or
381 * One major determining factor in the format of extents is how we handle and
382 * represent extents that have been partially overwritten and thus trimmed:
384 * If an extent is not checksummed or compressed, when the extent is trimmed we
385 * don't have to remember the extent we originally allocated and wrote: we can
386 * merely adjust ptr->offset to point to the start of the data that is currently
387 * live. The size field in struct bkey records the current (live) size of the
388 * extent, and is also used to mean "size of region on disk that we point to" in
391 * Thus an extent that is not checksummed or compressed will consist only of a
392 * list of bch_extent_ptrs, with none of the fields in
393 * bch_extent_crc32/bch_extent_crc64.
395 * When an extent is checksummed or compressed, it's not possible to read only
396 * the data that is currently live: we have to read the entire extent that was
397 * originally written, and then return only the part of the extent that is
400 * Thus, in addition to the current size of the extent in struct bkey, we need
401 * to store the size of the originally allocated space - this is the
402 * compressed_size and uncompressed_size fields in bch_extent_crc32/64. Also,
403 * when the extent is trimmed, instead of modifying the offset field of the
404 * pointer, we keep a second smaller offset field - "offset into the original
405 * extent of the currently live region".
407 * The other major determining factor is replication and data migration:
409 * Each pointer may have its own bch_extent_crc32/64. When doing a replicated
410 * write, we will initially write all the replicas in the same format, with the
411 * same checksum type and compression format - however, when copygc runs later (or
412 * tiering/cache promotion, anything that moves data), it is not in general
413 * going to rewrite all the pointers at once - one of the replicas may be in a
414 * bucket on one device that has very little fragmentation while another lives
415 * in a bucket that has become heavily fragmented, and thus is being rewritten
416 * sooner than the rest.
418 * Thus it will only move a subset of the pointers (or in the case of
419 * tiering/cache promotion perhaps add a single pointer without dropping any
420 * current pointers), and if the extent has been partially overwritten it must
421 * write only the currently live portion (or copygc would not be able to reduce
422 * fragmentation!) - which necessitates a different bch_extent_crc format for
425 * But in the interests of space efficiency, we don't want to store one
426 * bch_extent_crc for each pointer if we don't have to.
428 * Thus, a bch_extent consists of bch_extent_crc32s, bch_extent_crc64s, and
429 * bch_extent_ptrs appended arbitrarily one after the other. We determine the
430 * type of a given entry with a scheme similar to utf8 (except we're encoding a
431 * type, not a size), encoding the type in the position of the first set bit:
433 * bch_extent_crc32 - 0b1
434 * bch_extent_ptr - 0b10
435 * bch_extent_crc64 - 0b100
437 * We do it this way because bch_extent_crc32 is _very_ constrained on bits (and
438 * bch_extent_crc64 is the least constrained).
440 * Then, each bch_extent_crc32/64 applies to the pointers that follow after it,
441 * until the next bch_extent_crc32/64.
443 * If there are no bch_extent_crcs preceding a bch_extent_ptr, then that pointer
444 * is neither checksummed nor compressed.
447 /* 128 bits, sufficient for cryptographic MACs: */
451 } __attribute__((packed, aligned(8)));
453 #define BCH_EXTENT_ENTRY_TYPES() \
459 #define BCH_EXTENT_ENTRY_MAX 5
461 enum bch_extent_entry_type {
462 #define x(f, n) BCH_EXTENT_ENTRY_##f = n,
463 BCH_EXTENT_ENTRY_TYPES()
467 /* Compressed/uncompressed size are stored biased by 1: */
468 struct bch_extent_crc32 {
469 #if defined(__LITTLE_ENDIAN_BITFIELD)
472 _uncompressed_size:7,
478 #elif defined (__BIG_ENDIAN_BITFIELD)
480 __u32 compression_type:4,
484 _uncompressed_size:7,
488 } __attribute__((packed, aligned(8)));
490 #define CRC32_SIZE_MAX (1U << 7)
491 #define CRC32_NONCE_MAX 0
493 struct bch_extent_crc64 {
494 #if defined(__LITTLE_ENDIAN_BITFIELD)
497 _uncompressed_size:9,
503 #elif defined (__BIG_ENDIAN_BITFIELD)
509 _uncompressed_size:9,
514 } __attribute__((packed, aligned(8)));
516 #define CRC64_SIZE_MAX (1U << 9)
517 #define CRC64_NONCE_MAX ((1U << 10) - 1)
519 struct bch_extent_crc128 {
520 #if defined(__LITTLE_ENDIAN_BITFIELD)
523 _uncompressed_size:13,
528 #elif defined (__BIG_ENDIAN_BITFIELD)
529 __u64 compression_type:4,
533 _uncompressed_size:13,
537 struct bch_csum csum;
538 } __attribute__((packed, aligned(8)));
540 #define CRC128_SIZE_MAX (1U << 13)
541 #define CRC128_NONCE_MAX ((1U << 13) - 1)
544 * @reservation - pointer hasn't been written to, just reserved
546 struct bch_extent_ptr {
547 #if defined(__LITTLE_ENDIAN_BITFIELD)
552 offset:44, /* 8 petabytes */
555 #elif defined (__BIG_ENDIAN_BITFIELD)
564 } __attribute__((packed, aligned(8)));
566 struct bch_extent_stripe_ptr {
567 #if defined(__LITTLE_ENDIAN_BITFIELD)
572 #elif defined (__BIG_ENDIAN_BITFIELD)
580 struct bch_extent_reservation {
581 #if defined(__LITTLE_ENDIAN_BITFIELD)
586 #elif defined (__BIG_ENDIAN_BITFIELD)
594 union bch_extent_entry {
595 #if __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__ || __BITS_PER_LONG == 64
597 #elif __BITS_PER_LONG == 32
603 #error edit for your odd byteorder.
606 #define x(f, n) struct bch_extent_##f f;
607 BCH_EXTENT_ENTRY_TYPES()
611 struct bch_btree_ptr {
614 struct bch_extent_ptr start[0];
616 } __attribute__((packed, aligned(8)));
618 struct bch_btree_ptr_v2 {
623 __le16 sectors_written;
626 struct bch_extent_ptr start[0];
628 } __attribute__((packed, aligned(8)));
630 LE16_BITMASK(BTREE_PTR_RANGE_UPDATED, struct bch_btree_ptr_v2, flags, 0, 1);
635 union bch_extent_entry start[0];
637 } __attribute__((packed, aligned(8)));
639 struct bch_reservation {
645 } __attribute__((packed, aligned(8)));
647 /* Maximum size (in u64s) a single pointer could be: */
648 #define BKEY_EXTENT_PTR_U64s_MAX\
649 ((sizeof(struct bch_extent_crc128) + \
650 sizeof(struct bch_extent_ptr)) / sizeof(u64))
652 /* Maximum possible size of an entire extent value: */
653 #define BKEY_EXTENT_VAL_U64s_MAX \
654 (1 + BKEY_EXTENT_PTR_U64s_MAX * (BCH_REPLICAS_MAX + 1))
656 /* * Maximum possible size of an entire extent, key + value: */
657 #define BKEY_EXTENT_U64s_MAX (BKEY_U64s + BKEY_EXTENT_VAL_U64s_MAX)
659 /* Btree pointers don't carry around checksums: */
660 #define BKEY_BTREE_PTR_VAL_U64s_MAX \
661 ((sizeof(struct bch_btree_ptr_v2) + \
662 sizeof(struct bch_extent_ptr) * BCH_REPLICAS_MAX) / sizeof(u64))
663 #define BKEY_BTREE_PTR_U64s_MAX \
664 (BKEY_U64s + BKEY_BTREE_PTR_VAL_U64s_MAX)
668 #define BLOCKDEV_INODE_MAX 4096
670 #define BCACHEFS_ROOT_INO 4096
679 } __attribute__((packed, aligned(8)));
681 struct bch_inode_generation {
684 __le32 bi_generation;
686 } __attribute__((packed, aligned(8)));
688 #define BCH_INODE_FIELDS() \
698 x(bi_generation, 32) \
700 x(bi_data_checksum, 8) \
701 x(bi_compression, 8) \
703 x(bi_background_compression, 8) \
704 x(bi_data_replicas, 8) \
705 x(bi_promote_target, 16) \
706 x(bi_foreground_target, 16) \
707 x(bi_background_target, 16) \
708 x(bi_erasure_code, 16) \
709 x(bi_fields_set, 16) \
713 /* subset of BCH_INODE_FIELDS */
714 #define BCH_INODE_OPTS() \
715 x(data_checksum, 8) \
718 x(background_compression, 8) \
719 x(data_replicas, 8) \
720 x(promote_target, 16) \
721 x(foreground_target, 16) \
722 x(background_target, 16) \
726 #define x(name, ...) \
735 * User flags (get/settable with FS_IOC_*FLAGS, correspond to FS_*_FL
738 __BCH_INODE_SYNC = 0,
739 __BCH_INODE_IMMUTABLE = 1,
740 __BCH_INODE_APPEND = 2,
741 __BCH_INODE_NODUMP = 3,
742 __BCH_INODE_NOATIME = 4,
744 __BCH_INODE_I_SIZE_DIRTY= 5,
745 __BCH_INODE_I_SECTORS_DIRTY= 6,
746 __BCH_INODE_UNLINKED = 7,
747 __BCH_INODE_BACKPTR_UNTRUSTED = 8,
749 /* bits 20+ reserved for packed fields below: */
752 #define BCH_INODE_SYNC (1 << __BCH_INODE_SYNC)
753 #define BCH_INODE_IMMUTABLE (1 << __BCH_INODE_IMMUTABLE)
754 #define BCH_INODE_APPEND (1 << __BCH_INODE_APPEND)
755 #define BCH_INODE_NODUMP (1 << __BCH_INODE_NODUMP)
756 #define BCH_INODE_NOATIME (1 << __BCH_INODE_NOATIME)
757 #define BCH_INODE_I_SIZE_DIRTY (1 << __BCH_INODE_I_SIZE_DIRTY)
758 #define BCH_INODE_I_SECTORS_DIRTY (1 << __BCH_INODE_I_SECTORS_DIRTY)
759 #define BCH_INODE_UNLINKED (1 << __BCH_INODE_UNLINKED)
760 #define BCH_INODE_BACKPTR_UNTRUSTED (1 << __BCH_INODE_BACKPTR_UNTRUSTED)
762 LE32_BITMASK(INODE_STR_HASH, struct bch_inode, bi_flags, 20, 24);
763 LE32_BITMASK(INODE_NR_FIELDS, struct bch_inode, bi_flags, 24, 31);
764 LE32_BITMASK(INODE_NEW_VARINT, struct bch_inode, bi_flags, 31, 32);
769 * Dirents (and xattrs) have to implement string lookups; since our b-tree
770 * doesn't support arbitrary length strings for the key, we instead index by a
771 * 64 bit hash (currently truncated sha1) of the string, stored in the offset
772 * field of the key - using linear probing to resolve hash collisions. This also
773 * provides us with the readdir cookie posix requires.
775 * Linear probing requires us to use whiteouts for deletions, in the event of a
782 /* Target inode number: */
786 * Copy of mode bits 12-15 from the target inode - so userspace can get
787 * the filetype without having to do a stat()
792 } __attribute__((packed, aligned(8)));
794 #define BCH_NAME_MAX (U8_MAX * sizeof(u64) - \
795 sizeof(struct bkey) - \
796 offsetof(struct bch_dirent, d_name))
801 #define KEY_TYPE_XATTR_INDEX_USER 0
802 #define KEY_TYPE_XATTR_INDEX_POSIX_ACL_ACCESS 1
803 #define KEY_TYPE_XATTR_INDEX_POSIX_ACL_DEFAULT 2
804 #define KEY_TYPE_XATTR_INDEX_TRUSTED 3
805 #define KEY_TYPE_XATTR_INDEX_SECURITY 4
813 } __attribute__((packed, aligned(8)));
815 /* Bucket/allocation information: */
822 } __attribute__((packed, aligned(8)));
824 #define BCH_ALLOC_FIELDS_V1() \
828 x(dirty_sectors, 16) \
829 x(cached_sectors, 16) \
832 x(stripe_redundancy, 8)
834 struct bch_alloc_v2 {
841 } __attribute__((packed, aligned(8)));
843 #define BCH_ALLOC_FIELDS_V2() \
846 x(dirty_sectors, 16) \
847 x(cached_sectors, 16) \
849 x(stripe_redundancy, 8)
852 #define x(name, _bits) BCH_ALLOC_FIELD_V1_##name,
853 BCH_ALLOC_FIELDS_V1()
867 enum quota_counters {
873 struct bch_quota_counter {
880 struct bch_quota_counter c[Q_COUNTERS];
881 } __attribute__((packed, aligned(8)));
892 __u8 csum_granularity_bits;
896 struct bch_extent_ptr ptrs[0];
897 } __attribute__((packed, aligned(8)));
901 struct bch_reflink_p {
905 __le32 reservation_generation;
910 struct bch_reflink_v {
913 union bch_extent_entry start[0];
917 struct bch_indirect_inline_data {
925 struct bch_inline_data {
930 /* Optional/variable size superblock sections: */
932 struct bch_sb_field {
938 #define BCH_SB_FIELDS() \
947 x(journal_seq_blacklist, 8)
949 enum bch_sb_field_type {
950 #define x(f, nr) BCH_SB_FIELD_##f = nr,
956 /* BCH_SB_FIELD_journal: */
958 struct bch_sb_field_journal {
959 struct bch_sb_field field;
963 /* BCH_SB_FIELD_members: */
965 #define BCH_MIN_NR_NBUCKETS (1 << 6)
969 __le64 nbuckets; /* device size */
970 __le16 first_bucket; /* index of first bucket used */
971 __le16 bucket_size; /* sectors */
973 __le64 last_mount; /* time_t */
978 LE64_BITMASK(BCH_MEMBER_STATE, struct bch_member, flags[0], 0, 4)
979 /* 4-10 unused, was TIER, HAS_(META)DATA */
980 LE64_BITMASK(BCH_MEMBER_REPLACEMENT, struct bch_member, flags[0], 10, 14)
981 LE64_BITMASK(BCH_MEMBER_DISCARD, struct bch_member, flags[0], 14, 15)
982 LE64_BITMASK(BCH_MEMBER_DATA_ALLOWED, struct bch_member, flags[0], 15, 20)
983 LE64_BITMASK(BCH_MEMBER_GROUP, struct bch_member, flags[0], 20, 28)
984 LE64_BITMASK(BCH_MEMBER_DURABILITY, struct bch_member, flags[0], 28, 30)
986 #define BCH_TIER_MAX 4U
989 LE64_BITMASK(BCH_MEMBER_NR_READ_ERRORS, struct bch_member, flags[1], 0, 20);
990 LE64_BITMASK(BCH_MEMBER_NR_WRITE_ERRORS,struct bch_member, flags[1], 20, 40);
993 #define BCH_MEMBER_STATES() \
999 enum bch_member_state {
1000 #define x(t, n) BCH_MEMBER_STATE_##t = n,
1006 #define BCH_CACHE_REPLACEMENT_POLICIES() \
1011 enum bch_cache_replacement_policies {
1012 #define x(t, n) BCH_CACHE_REPLACEMENT_##t = n,
1013 BCH_CACHE_REPLACEMENT_POLICIES()
1015 BCH_CACHE_REPLACEMENT_NR
1018 struct bch_sb_field_members {
1019 struct bch_sb_field field;
1020 struct bch_member members[0];
1023 /* BCH_SB_FIELD_crypt: */
1033 #define BCH_KEY_MAGIC \
1034 (((u64) 'b' << 0)|((u64) 'c' << 8)| \
1035 ((u64) 'h' << 16)|((u64) '*' << 24)| \
1036 ((u64) '*' << 32)|((u64) 'k' << 40)| \
1037 ((u64) 'e' << 48)|((u64) 'y' << 56))
1039 struct bch_encrypted_key {
1045 * If this field is present in the superblock, it stores an encryption key which
1046 * is used encrypt all other data/metadata. The key will normally be encrypted
1047 * with the key userspace provides, but if encryption has been turned off we'll
1048 * just store the master key unencrypted in the superblock so we can access the
1049 * previously encrypted data.
1051 struct bch_sb_field_crypt {
1052 struct bch_sb_field field;
1056 struct bch_encrypted_key key;
1059 LE64_BITMASK(BCH_CRYPT_KDF_TYPE, struct bch_sb_field_crypt, flags, 0, 4);
1061 enum bch_kdf_types {
1066 /* stored as base 2 log of scrypt params: */
1067 LE64_BITMASK(BCH_KDF_SCRYPT_N, struct bch_sb_field_crypt, kdf_flags, 0, 16);
1068 LE64_BITMASK(BCH_KDF_SCRYPT_R, struct bch_sb_field_crypt, kdf_flags, 16, 32);
1069 LE64_BITMASK(BCH_KDF_SCRYPT_P, struct bch_sb_field_crypt, kdf_flags, 32, 48);
1071 /* BCH_SB_FIELD_replicas: */
1073 #define BCH_DATA_TYPES() \
1082 enum bch_data_type {
1083 #define x(t, n) BCH_DATA_##t,
1089 struct bch_replicas_entry_v0 {
1093 } __attribute__((packed));
1095 struct bch_sb_field_replicas_v0 {
1096 struct bch_sb_field field;
1097 struct bch_replicas_entry_v0 entries[0];
1098 } __attribute__((packed, aligned(8)));
1100 struct bch_replicas_entry {
1105 } __attribute__((packed));
1107 #define replicas_entry_bytes(_i) \
1108 (offsetof(typeof(*(_i)), devs) + (_i)->nr_devs)
1110 struct bch_sb_field_replicas {
1111 struct bch_sb_field field;
1112 struct bch_replicas_entry entries[0];
1113 } __attribute__((packed, aligned(8)));
1115 /* BCH_SB_FIELD_quota: */
1117 struct bch_sb_quota_counter {
1122 struct bch_sb_quota_type {
1124 struct bch_sb_quota_counter c[Q_COUNTERS];
1127 struct bch_sb_field_quota {
1128 struct bch_sb_field field;
1129 struct bch_sb_quota_type q[QTYP_NR];
1130 } __attribute__((packed, aligned(8)));
1132 /* BCH_SB_FIELD_disk_groups: */
1134 #define BCH_SB_LABEL_SIZE 32
1136 struct bch_disk_group {
1137 __u8 label[BCH_SB_LABEL_SIZE];
1139 } __attribute__((packed, aligned(8)));
1141 LE64_BITMASK(BCH_GROUP_DELETED, struct bch_disk_group, flags[0], 0, 1)
1142 LE64_BITMASK(BCH_GROUP_DATA_ALLOWED, struct bch_disk_group, flags[0], 1, 6)
1143 LE64_BITMASK(BCH_GROUP_PARENT, struct bch_disk_group, flags[0], 6, 24)
1145 struct bch_sb_field_disk_groups {
1146 struct bch_sb_field field;
1147 struct bch_disk_group entries[0];
1148 } __attribute__((packed, aligned(8)));
1151 * On clean shutdown, store btree roots and current journal sequence number in
1158 __u8 type; /* designates what this jset holds */
1162 struct bkey_i start[0];
1167 struct bch_sb_field_clean {
1168 struct bch_sb_field field;
1171 __le16 _read_clock; /* no longer used */
1172 __le16 _write_clock;
1176 struct jset_entry start[0];
1181 struct journal_seq_blacklist_entry {
1186 struct bch_sb_field_journal_seq_blacklist {
1187 struct bch_sb_field field;
1190 struct journal_seq_blacklist_entry start[0];
1198 * New versioning scheme:
1199 * One common version number for all on disk data structures - superblock, btree
1200 * nodes, journal entries
1202 #define BCH_JSET_VERSION_OLD 2
1203 #define BCH_BSET_VERSION_OLD 3
1205 enum bcachefs_metadata_version {
1206 bcachefs_metadata_version_min = 9,
1207 bcachefs_metadata_version_new_versioning = 10,
1208 bcachefs_metadata_version_bkey_renumber = 10,
1209 bcachefs_metadata_version_inode_btree_change = 11,
1210 bcachefs_metadata_version_snapshot = 12,
1211 bcachefs_metadata_version_inode_backpointers = 13,
1212 bcachefs_metadata_version_max = 14,
1215 #define bcachefs_metadata_version_current (bcachefs_metadata_version_max - 1)
1217 #define BCH_SB_SECTOR 8
1218 #define BCH_SB_MEMBERS_MAX 64 /* XXX kill */
1220 struct bch_sb_layout {
1221 uuid_le magic; /* bcachefs superblock UUID */
1223 __u8 sb_max_size_bits; /* base 2 of 512 byte sectors */
1224 __u8 nr_superblocks;
1226 __le64 sb_offset[61];
1227 } __attribute__((packed, aligned(8)));
1229 #define BCH_SB_LAYOUT_SECTOR 7
1232 * @offset - sector where this sb was written
1233 * @version - on disk format version
1234 * @version_min - Oldest metadata version this filesystem contains; so we can
1235 * safely drop compatibility code and refuse to mount filesystems
1237 * @magic - identifies as a bcachefs superblock (BCACHE_MAGIC)
1238 * @seq - incremented each time superblock is written
1239 * @uuid - used for generating various magic numbers and identifying
1240 * member devices, never changes
1241 * @user_uuid - user visible UUID, may be changed
1242 * @label - filesystem label
1243 * @seq - identifies most recent superblock, incremented each time
1244 * superblock is written
1245 * @features - enabled incompatible features
1248 struct bch_csum csum;
1255 __u8 label[BCH_SB_LABEL_SIZE];
1264 __le64 time_base_lo;
1265 __le32 time_base_hi;
1266 __le32 time_precision;
1272 struct bch_sb_layout layout;
1275 struct bch_sb_field start[0];
1278 } __attribute__((packed, aligned(8)));
1282 * BCH_SB_INITALIZED - set on first mount
1283 * BCH_SB_CLEAN - did we shut down cleanly? Just a hint, doesn't affect
1284 * behaviour of mount/recovery path:
1285 * BCH_SB_INODE_32BIT - limit inode numbers to 32 bits
1286 * BCH_SB_128_BIT_MACS - 128 bit macs instead of 80
1287 * BCH_SB_ENCRYPTION_TYPE - if nonzero encryption is enabled; overrides
1288 * DATA/META_CSUM_TYPE. Also indicates encryption
1289 * algorithm in use, if/when we get more than one
1292 LE16_BITMASK(BCH_SB_BLOCK_SIZE, struct bch_sb, block_size, 0, 16);
1294 LE64_BITMASK(BCH_SB_INITIALIZED, struct bch_sb, flags[0], 0, 1);
1295 LE64_BITMASK(BCH_SB_CLEAN, struct bch_sb, flags[0], 1, 2);
1296 LE64_BITMASK(BCH_SB_CSUM_TYPE, struct bch_sb, flags[0], 2, 8);
1297 LE64_BITMASK(BCH_SB_ERROR_ACTION, struct bch_sb, flags[0], 8, 12);
1299 LE64_BITMASK(BCH_SB_BTREE_NODE_SIZE, struct bch_sb, flags[0], 12, 28);
1301 LE64_BITMASK(BCH_SB_GC_RESERVE, struct bch_sb, flags[0], 28, 33);
1302 LE64_BITMASK(BCH_SB_ROOT_RESERVE, struct bch_sb, flags[0], 33, 40);
1304 LE64_BITMASK(BCH_SB_META_CSUM_TYPE, struct bch_sb, flags[0], 40, 44);
1305 LE64_BITMASK(BCH_SB_DATA_CSUM_TYPE, struct bch_sb, flags[0], 44, 48);
1307 LE64_BITMASK(BCH_SB_META_REPLICAS_WANT, struct bch_sb, flags[0], 48, 52);
1308 LE64_BITMASK(BCH_SB_DATA_REPLICAS_WANT, struct bch_sb, flags[0], 52, 56);
1310 LE64_BITMASK(BCH_SB_POSIX_ACL, struct bch_sb, flags[0], 56, 57);
1311 LE64_BITMASK(BCH_SB_USRQUOTA, struct bch_sb, flags[0], 57, 58);
1312 LE64_BITMASK(BCH_SB_GRPQUOTA, struct bch_sb, flags[0], 58, 59);
1313 LE64_BITMASK(BCH_SB_PRJQUOTA, struct bch_sb, flags[0], 59, 60);
1315 LE64_BITMASK(BCH_SB_HAS_ERRORS, struct bch_sb, flags[0], 60, 61);
1317 /* bit 61 was reflink option */
1318 LE64_BITMASK(BCH_SB_BIG_ENDIAN, struct bch_sb, flags[0], 62, 63);
1322 LE64_BITMASK(BCH_SB_STR_HASH_TYPE, struct bch_sb, flags[1], 0, 4);
1323 LE64_BITMASK(BCH_SB_COMPRESSION_TYPE, struct bch_sb, flags[1], 4, 8);
1324 LE64_BITMASK(BCH_SB_INODE_32BIT, struct bch_sb, flags[1], 8, 9);
1326 LE64_BITMASK(BCH_SB_128_BIT_MACS, struct bch_sb, flags[1], 9, 10);
1327 LE64_BITMASK(BCH_SB_ENCRYPTION_TYPE, struct bch_sb, flags[1], 10, 14);
1330 * Max size of an extent that may require bouncing to read or write
1331 * (checksummed, compressed): 64k
1333 LE64_BITMASK(BCH_SB_ENCODED_EXTENT_MAX_BITS,
1334 struct bch_sb, flags[1], 14, 20);
1336 LE64_BITMASK(BCH_SB_META_REPLICAS_REQ, struct bch_sb, flags[1], 20, 24);
1337 LE64_BITMASK(BCH_SB_DATA_REPLICAS_REQ, struct bch_sb, flags[1], 24, 28);
1339 LE64_BITMASK(BCH_SB_PROMOTE_TARGET, struct bch_sb, flags[1], 28, 40);
1340 LE64_BITMASK(BCH_SB_FOREGROUND_TARGET, struct bch_sb, flags[1], 40, 52);
1341 LE64_BITMASK(BCH_SB_BACKGROUND_TARGET, struct bch_sb, flags[1], 52, 64);
1343 LE64_BITMASK(BCH_SB_BACKGROUND_COMPRESSION_TYPE,
1344 struct bch_sb, flags[2], 0, 4);
1345 LE64_BITMASK(BCH_SB_GC_RESERVE_BYTES, struct bch_sb, flags[2], 4, 64);
1347 LE64_BITMASK(BCH_SB_ERASURE_CODE, struct bch_sb, flags[3], 0, 16);
1348 LE64_BITMASK(BCH_SB_METADATA_TARGET, struct bch_sb, flags[3], 16, 28);
1353 * journal_seq_blacklist_v3: gates BCH_SB_FIELD_journal_seq_blacklist
1354 * reflink: gates KEY_TYPE_reflink
1355 * inline_data: gates KEY_TYPE_inline_data
1356 * new_siphash: gates BCH_STR_HASH_SIPHASH
1357 * new_extent_overwrite: gates BTREE_NODE_NEW_EXTENT_OVERWRITE
1359 #define BCH_SB_FEATURES() \
1363 x(atomic_nlink, 3) \
1365 x(journal_seq_blacklist_v3, 5) \
1369 x(new_extent_overwrite, 9) \
1370 x(incompressible, 10) \
1371 x(btree_ptr_v2, 11) \
1372 x(extents_above_btree_updates, 12) \
1373 x(btree_updates_journalled, 13) \
1374 x(reflink_inline_data, 14) \
1376 x(journal_no_flush, 16) \
1378 x(extents_across_btree_nodes, 18)
1380 #define BCH_SB_FEATURES_ALWAYS \
1381 ((1ULL << BCH_FEATURE_new_extent_overwrite)| \
1382 (1ULL << BCH_FEATURE_extents_above_btree_updates)|\
1383 (1ULL << BCH_FEATURE_btree_updates_journalled)|\
1384 (1ULL << BCH_FEATURE_alloc_v2)|\
1385 (1ULL << BCH_FEATURE_extents_across_btree_nodes))
1387 #define BCH_SB_FEATURES_ALL \
1388 (BCH_SB_FEATURES_ALWAYS| \
1389 (1ULL << BCH_FEATURE_new_siphash)| \
1390 (1ULL << BCH_FEATURE_btree_ptr_v2)| \
1391 (1ULL << BCH_FEATURE_new_varint)| \
1392 (1ULL << BCH_FEATURE_journal_no_flush))
1394 enum bch_sb_feature {
1395 #define x(f, n) BCH_FEATURE_##f,
1401 #define BCH_SB_COMPAT() \
1403 x(alloc_metadata, 1) \
1404 x(extents_above_btree_updates_done, 2) \
1405 x(bformat_overflow_done, 3)
1407 enum bch_sb_compat {
1408 #define x(f, n) BCH_COMPAT_##f,
1416 #define BCH_REPLICAS_MAX 4U
1418 #define BCH_BKEY_PTRS_MAX 16U
1420 #define BCH_ERROR_ACTIONS() \
1425 enum bch_error_actions {
1426 #define x(t, n) BCH_ON_ERROR_##t = n,
1432 enum bch_str_hash_type {
1433 BCH_STR_HASH_CRC32C = 0,
1434 BCH_STR_HASH_CRC64 = 1,
1435 BCH_STR_HASH_SIPHASH_OLD = 2,
1436 BCH_STR_HASH_SIPHASH = 3,
1437 BCH_STR_HASH_NR = 4,
1440 #define BCH_STR_HASH_OPTS() \
1445 enum bch_str_hash_opts {
1446 #define x(t, n) BCH_STR_HASH_OPT_##t = n,
1452 enum bch_csum_type {
1454 BCH_CSUM_CRC32C_NONZERO = 1,
1455 BCH_CSUM_CRC64_NONZERO = 2,
1456 BCH_CSUM_CHACHA20_POLY1305_80 = 3,
1457 BCH_CSUM_CHACHA20_POLY1305_128 = 4,
1458 BCH_CSUM_CRC32C = 5,
1463 static const unsigned bch_crc_bytes[] = {
1464 [BCH_CSUM_NONE] = 0,
1465 [BCH_CSUM_CRC32C_NONZERO] = 4,
1466 [BCH_CSUM_CRC32C] = 4,
1467 [BCH_CSUM_CRC64_NONZERO] = 8,
1468 [BCH_CSUM_CRC64] = 8,
1469 [BCH_CSUM_CHACHA20_POLY1305_80] = 10,
1470 [BCH_CSUM_CHACHA20_POLY1305_128] = 16,
1473 static inline _Bool bch2_csum_type_is_encryption(enum bch_csum_type type)
1476 case BCH_CSUM_CHACHA20_POLY1305_80:
1477 case BCH_CSUM_CHACHA20_POLY1305_128:
1484 #define BCH_CSUM_OPTS() \
1489 enum bch_csum_opts {
1490 #define x(t, n) BCH_CSUM_OPT_##t = n,
1496 #define BCH_COMPRESSION_TYPES() \
1502 x(incompressible, 5)
1504 enum bch_compression_type {
1505 #define x(t, n) BCH_COMPRESSION_TYPE_##t = n,
1506 BCH_COMPRESSION_TYPES()
1508 BCH_COMPRESSION_TYPE_NR
1511 #define BCH_COMPRESSION_OPTS() \
1517 enum bch_compression_opts {
1518 #define x(t, n) BCH_COMPRESSION_OPT_##t = n,
1519 BCH_COMPRESSION_OPTS()
1521 BCH_COMPRESSION_OPT_NR
1527 * The various other data structures have their own magic numbers, which are
1528 * xored with the first part of the cache set's UUID
1531 #define BCACHE_MAGIC \
1532 UUID_LE(0xf67385c6, 0x1a4e, 0xca45, \
1533 0x82, 0x65, 0xf5, 0x7f, 0x48, 0xba, 0x6d, 0x81)
1535 #define BCACHEFS_STATFS_MAGIC 0xca451a4e
1537 #define JSET_MAGIC __cpu_to_le64(0x245235c1a3625032ULL)
1538 #define BSET_MAGIC __cpu_to_le64(0x90135c78b99e07f5ULL)
1540 static inline __le64 __bch2_sb_magic(struct bch_sb *sb)
1543 memcpy(&ret, &sb->uuid, sizeof(ret));
1547 static inline __u64 __jset_magic(struct bch_sb *sb)
1549 return __le64_to_cpu(__bch2_sb_magic(sb) ^ JSET_MAGIC);
1552 static inline __u64 __bset_magic(struct bch_sb *sb)
1554 return __le64_to_cpu(__bch2_sb_magic(sb) ^ BSET_MAGIC);
1559 #define JSET_KEYS_U64s (sizeof(struct jset_entry) / sizeof(__u64))
1561 #define BCH_JSET_ENTRY_TYPES() \
1566 x(blacklist_v2, 4) \
1573 #define x(f, nr) BCH_JSET_ENTRY_##f = nr,
1574 BCH_JSET_ENTRY_TYPES()
1580 * Journal sequence numbers can be blacklisted: bsets record the max sequence
1581 * number of all the journal entries they contain updates for, so that on
1582 * recovery we can ignore those bsets that contain index updates newer that what
1583 * made it into the journal.
1585 * This means that we can't reuse that journal_seq - we have to skip it, and
1586 * then record that we skipped it so that the next time we crash and recover we
1587 * don't think there was a missing journal entry.
1589 struct jset_entry_blacklist {
1590 struct jset_entry entry;
1594 struct jset_entry_blacklist_v2 {
1595 struct jset_entry entry;
1601 FS_USAGE_RESERVED = 0,
1602 FS_USAGE_INODES = 1,
1603 FS_USAGE_KEY_VERSION = 2,
1607 struct jset_entry_usage {
1608 struct jset_entry entry;
1610 } __attribute__((packed));
1612 struct jset_entry_data_usage {
1613 struct jset_entry entry;
1615 struct bch_replicas_entry r;
1616 } __attribute__((packed));
1618 struct jset_entry_clock {
1619 struct jset_entry entry;
1623 } __attribute__((packed));
1625 struct jset_entry_dev_usage_type {
1629 } __attribute__((packed));
1631 struct jset_entry_dev_usage {
1632 struct jset_entry entry;
1637 __le64 buckets_unavailable;
1639 struct jset_entry_dev_usage_type d[];
1640 } __attribute__((packed));
1643 * On disk format for a journal entry:
1644 * seq is monotonically increasing; every journal entry has its own unique
1647 * last_seq is the oldest journal entry that still has keys the btree hasn't
1648 * flushed to disk yet.
1650 * version is for on disk format changes.
1653 struct bch_csum csum;
1660 __le32 u64s; /* size of d[] in u64s */
1662 __u8 encrypted_start[0];
1664 __le16 _read_clock; /* no longer used */
1665 __le16 _write_clock;
1667 /* Sequence number of oldest dirty journal entry */
1672 struct jset_entry start[0];
1675 } __attribute__((packed, aligned(8)));
1677 LE32_BITMASK(JSET_CSUM_TYPE, struct jset, flags, 0, 4);
1678 LE32_BITMASK(JSET_BIG_ENDIAN, struct jset, flags, 4, 5);
1679 LE32_BITMASK(JSET_NO_FLUSH, struct jset, flags, 5, 6);
1681 #define BCH_JOURNAL_BUCKETS_MIN 8
1685 #define BCH_BTREE_IDS() \
1696 #define x(kwd, val) BTREE_ID_##kwd = val,
1702 #define BTREE_MAX_DEPTH 4U
1709 * On disk a btree node is a list/log of these; within each set the keys are
1716 * Highest journal entry this bset contains keys for.
1717 * If on recovery we don't see that journal entry, this bset is ignored:
1718 * this allows us to preserve the order of all index updates after a
1719 * crash, since the journal records a total order of all index updates
1720 * and anything that didn't make it to the journal doesn't get used.
1726 __le16 u64s; /* count of d[] in u64s */
1729 struct bkey_packed start[0];
1732 } __attribute__((packed, aligned(8)));
1734 LE32_BITMASK(BSET_CSUM_TYPE, struct bset, flags, 0, 4);
1736 LE32_BITMASK(BSET_BIG_ENDIAN, struct bset, flags, 4, 5);
1737 LE32_BITMASK(BSET_SEPARATE_WHITEOUTS,
1738 struct bset, flags, 5, 6);
1741 struct bch_csum csum;
1744 /* this flags field is encrypted, unlike bset->flags: */
1747 /* Closed interval: */
1748 struct bpos min_key;
1749 struct bpos max_key;
1750 struct bch_extent_ptr _ptr; /* not used anymore */
1751 struct bkey_format format;
1762 } __attribute__((packed, aligned(8)));
1764 LE64_BITMASK(BTREE_NODE_ID, struct btree_node, flags, 0, 4);
1765 LE64_BITMASK(BTREE_NODE_LEVEL, struct btree_node, flags, 4, 8);
1766 LE64_BITMASK(BTREE_NODE_NEW_EXTENT_OVERWRITE,
1767 struct btree_node, flags, 8, 9);
1769 LE64_BITMASK(BTREE_NODE_SEQ, struct btree_node, flags, 32, 64);
1771 struct btree_node_entry {
1772 struct bch_csum csum;
1783 } __attribute__((packed, aligned(8)));
1785 #endif /* _BCACHEFS_FORMAT_H */