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>
82 typedef uuid_t __uuid_t;
85 #define BITMASK(name, type, field, offset, end) \
86 static const __maybe_unused unsigned name##_OFFSET = offset; \
87 static const __maybe_unused unsigned name##_BITS = (end - offset); \
89 static inline __u64 name(const type *k) \
91 return (k->field >> offset) & ~(~0ULL << (end - offset)); \
94 static inline void SET_##name(type *k, __u64 v) \
96 k->field &= ~(~(~0ULL << (end - offset)) << offset); \
97 k->field |= (v & ~(~0ULL << (end - offset))) << offset; \
100 #define LE_BITMASK(_bits, name, type, field, offset, end) \
101 static const __maybe_unused unsigned name##_OFFSET = offset; \
102 static const __maybe_unused unsigned name##_BITS = (end - offset); \
103 static const __maybe_unused __u##_bits name##_MAX = (1ULL << (end - offset)) - 1;\
105 static inline __u64 name(const type *k) \
107 return (__le##_bits##_to_cpu(k->field) >> offset) & \
108 ~(~0ULL << (end - offset)); \
111 static inline void SET_##name(type *k, __u64 v) \
113 __u##_bits new = __le##_bits##_to_cpu(k->field); \
115 new &= ~(~(~0ULL << (end - offset)) << offset); \
116 new |= (v & ~(~0ULL << (end - offset))) << offset; \
117 k->field = __cpu_to_le##_bits(new); \
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)
127 /* One unused slot for now: */
128 __u8 bits_per_field[6];
129 __le64 field_offset[6];
132 /* Btree keys - all units are in sectors */
136 * Word order matches machine byte order - btree code treats a bpos as a
137 * single large integer, for search/comparison purposes
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:
143 #if __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__
147 #elif __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__
149 __u64 offset; /* Points to end of extent - sectors */
152 #error edit for your odd byteorder.
155 #if __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__
160 #define KEY_INODE_MAX ((__u64)~0ULL)
161 #define KEY_OFFSET_MAX ((__u64)~0ULL)
162 #define KEY_SNAPSHOT_MAX ((__u32)~0U)
163 #define KEY_SIZE_MAX ((__u32)~0U)
165 static inline struct bpos SPOS(__u64 inode, __u64 offset, __u32 snapshot)
167 return (struct bpos) {
170 .snapshot = snapshot,
174 #define POS_MIN SPOS(0, 0, 0)
175 #define POS_MAX SPOS(KEY_INODE_MAX, KEY_OFFSET_MAX, 0)
176 #define SPOS_MAX SPOS(KEY_INODE_MAX, KEY_OFFSET_MAX, KEY_SNAPSHOT_MAX)
177 #define POS(_inode, _offset) SPOS(_inode, _offset, 0)
179 /* Empty placeholder struct, for container_of() */
185 #if __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__
188 #elif __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__
192 } __packed __aligned(4);
195 /* Size of combined key and value, in u64s */
198 /* Format of key (0 for format local to btree node) */
199 #if defined(__LITTLE_ENDIAN_BITFIELD)
202 #elif defined (__BIG_ENDIAN_BITFIELD)
203 __u8 needs_whiteout:1,
206 #error edit for your odd byteorder.
209 /* Type of the value */
212 #if __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__
215 struct bversion version;
216 __u32 size; /* extent size, in sectors */
218 #elif __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__
220 __u32 size; /* extent size, in sectors */
221 struct bversion version;
225 } __packed __aligned(8);
230 /* Size of combined key and value, in u64s */
233 /* Format of key (0 for format local to btree node) */
236 * XXX: next incompat on disk format change, switch format and
237 * needs_whiteout - bkey_packed() will be cheaper if format is the high
238 * bits of the bitfield
240 #if defined(__LITTLE_ENDIAN_BITFIELD)
243 #elif defined (__BIG_ENDIAN_BITFIELD)
244 __u8 needs_whiteout:1,
248 /* Type of the value */
253 * We copy bkeys with struct assignment in various places, and while
254 * that shouldn't be done with packed bkeys we can't disallow it in C,
255 * and it's legal to cast a bkey to a bkey_packed - so padding it out
256 * to the same size as struct bkey should hopefully be safest.
258 __u8 pad[sizeof(struct bkey) - 3];
259 } __packed __aligned(8);
266 #define BKEY_U64s (sizeof(struct bkey) / sizeof(__u64))
267 #define BKEY_U64s_MAX U8_MAX
268 #define BKEY_VAL_U64s_MAX (BKEY_U64s_MAX - BKEY_U64s)
270 #define KEY_PACKED_BITS_START 24
272 #define KEY_FORMAT_LOCAL_BTREE 0
273 #define KEY_FORMAT_CURRENT 1
275 enum bch_bkey_fields {
280 BKEY_FIELD_VERSION_HI,
281 BKEY_FIELD_VERSION_LO,
285 #define bkey_format_field(name, field) \
286 [BKEY_FIELD_##name] = (sizeof(((struct bkey *) NULL)->field) * 8)
288 #define BKEY_FORMAT_CURRENT \
289 ((struct bkey_format) { \
290 .key_u64s = BKEY_U64s, \
291 .nr_fields = BKEY_NR_FIELDS, \
292 .bits_per_field = { \
293 bkey_format_field(INODE, p.inode), \
294 bkey_format_field(OFFSET, p.offset), \
295 bkey_format_field(SNAPSHOT, p.snapshot), \
296 bkey_format_field(SIZE, size), \
297 bkey_format_field(VERSION_HI, version.hi), \
298 bkey_format_field(VERSION_LO, version.lo), \
302 /* bkey with inline value */
310 #define POS_KEY(_pos) \
313 .format = KEY_FORMAT_CURRENT, \
317 #define KEY(_inode, _offset, _size) \
320 .format = KEY_FORMAT_CURRENT, \
321 .p = POS(_inode, _offset), \
325 static inline void bkey_init(struct bkey *k)
330 #define bkey_bytes(_k) ((_k)->u64s * sizeof(__u64))
332 #define __BKEY_PADDED(key, pad) \
333 struct bkey_i key; __u64 key ## _pad[pad]
336 * - DELETED keys are used internally to mark keys that should be ignored but
337 * override keys in composition order. Their version number is ignored.
339 * - DISCARDED keys indicate that the data is all 0s because it has been
340 * discarded. DISCARDs may have a version; if the version is nonzero the key
341 * will be persistent, otherwise the key will be dropped whenever the btree
342 * node is rewritten (like DELETED keys).
344 * - ERROR: any read of the data returns a read error, as the data was lost due
345 * to a failing device. Like DISCARDED keys, they can be removed (overridden)
346 * by new writes or cluster-wide GC. Node repair can also overwrite them with
347 * the same or a more recent version number, but not with an older version
350 * - WHITEOUT: for hash table btrees
352 #define BCH_BKEY_TYPES() \
357 x(hash_whiteout, 4) \
362 x(inode_generation, 9) \
371 x(btree_ptr_v2, 18) \
372 x(indirect_inline_data, 19) \
384 x(snapshot_tree, 31) \
385 x(logged_op_truncate, 32) \
386 x(logged_op_finsert, 33)
389 #define x(name, nr) KEY_TYPE_##name = nr,
399 struct bch_whiteout {
412 struct bch_hash_whiteout {
423 * In extent bkeys, the value is a list of pointers (bch_extent_ptr), optionally
424 * preceded by checksum/compression information (bch_extent_crc32 or
427 * One major determining factor in the format of extents is how we handle and
428 * represent extents that have been partially overwritten and thus trimmed:
430 * If an extent is not checksummed or compressed, when the extent is trimmed we
431 * don't have to remember the extent we originally allocated and wrote: we can
432 * merely adjust ptr->offset to point to the start of the data that is currently
433 * live. The size field in struct bkey records the current (live) size of the
434 * extent, and is also used to mean "size of region on disk that we point to" in
437 * Thus an extent that is not checksummed or compressed will consist only of a
438 * list of bch_extent_ptrs, with none of the fields in
439 * bch_extent_crc32/bch_extent_crc64.
441 * When an extent is checksummed or compressed, it's not possible to read only
442 * the data that is currently live: we have to read the entire extent that was
443 * originally written, and then return only the part of the extent that is
446 * Thus, in addition to the current size of the extent in struct bkey, we need
447 * to store the size of the originally allocated space - this is the
448 * compressed_size and uncompressed_size fields in bch_extent_crc32/64. Also,
449 * when the extent is trimmed, instead of modifying the offset field of the
450 * pointer, we keep a second smaller offset field - "offset into the original
451 * extent of the currently live region".
453 * The other major determining factor is replication and data migration:
455 * Each pointer may have its own bch_extent_crc32/64. When doing a replicated
456 * write, we will initially write all the replicas in the same format, with the
457 * same checksum type and compression format - however, when copygc runs later (or
458 * tiering/cache promotion, anything that moves data), it is not in general
459 * going to rewrite all the pointers at once - one of the replicas may be in a
460 * bucket on one device that has very little fragmentation while another lives
461 * in a bucket that has become heavily fragmented, and thus is being rewritten
462 * sooner than the rest.
464 * Thus it will only move a subset of the pointers (or in the case of
465 * tiering/cache promotion perhaps add a single pointer without dropping any
466 * current pointers), and if the extent has been partially overwritten it must
467 * write only the currently live portion (or copygc would not be able to reduce
468 * fragmentation!) - which necessitates a different bch_extent_crc format for
471 * But in the interests of space efficiency, we don't want to store one
472 * bch_extent_crc for each pointer if we don't have to.
474 * Thus, a bch_extent consists of bch_extent_crc32s, bch_extent_crc64s, and
475 * bch_extent_ptrs appended arbitrarily one after the other. We determine the
476 * type of a given entry with a scheme similar to utf8 (except we're encoding a
477 * type, not a size), encoding the type in the position of the first set bit:
479 * bch_extent_crc32 - 0b1
480 * bch_extent_ptr - 0b10
481 * bch_extent_crc64 - 0b100
483 * We do it this way because bch_extent_crc32 is _very_ constrained on bits (and
484 * bch_extent_crc64 is the least constrained).
486 * Then, each bch_extent_crc32/64 applies to the pointers that follow after it,
487 * until the next bch_extent_crc32/64.
489 * If there are no bch_extent_crcs preceding a bch_extent_ptr, then that pointer
490 * is neither checksummed nor compressed.
493 /* 128 bits, sufficient for cryptographic MACs: */
497 } __packed __aligned(8);
499 #define BCH_EXTENT_ENTRY_TYPES() \
506 #define BCH_EXTENT_ENTRY_MAX 6
508 enum bch_extent_entry_type {
509 #define x(f, n) BCH_EXTENT_ENTRY_##f = n,
510 BCH_EXTENT_ENTRY_TYPES()
514 /* Compressed/uncompressed size are stored biased by 1: */
515 struct bch_extent_crc32 {
516 #if defined(__LITTLE_ENDIAN_BITFIELD)
519 _uncompressed_size:7,
525 #elif defined (__BIG_ENDIAN_BITFIELD)
527 __u32 compression_type:4,
531 _uncompressed_size:7,
535 } __packed __aligned(8);
537 #define CRC32_SIZE_MAX (1U << 7)
538 #define CRC32_NONCE_MAX 0
540 struct bch_extent_crc64 {
541 #if defined(__LITTLE_ENDIAN_BITFIELD)
544 _uncompressed_size:9,
550 #elif defined (__BIG_ENDIAN_BITFIELD)
556 _uncompressed_size:9,
561 } __packed __aligned(8);
563 #define CRC64_SIZE_MAX (1U << 9)
564 #define CRC64_NONCE_MAX ((1U << 10) - 1)
566 struct bch_extent_crc128 {
567 #if defined(__LITTLE_ENDIAN_BITFIELD)
570 _uncompressed_size:13,
575 #elif defined (__BIG_ENDIAN_BITFIELD)
576 __u64 compression_type:4,
580 _uncompressed_size:13,
584 struct bch_csum csum;
585 } __packed __aligned(8);
587 #define CRC128_SIZE_MAX (1U << 13)
588 #define CRC128_NONCE_MAX ((1U << 13) - 1)
591 * @reservation - pointer hasn't been written to, just reserved
593 struct bch_extent_ptr {
594 #if defined(__LITTLE_ENDIAN_BITFIELD)
599 offset:44, /* 8 petabytes */
602 #elif defined (__BIG_ENDIAN_BITFIELD)
611 } __packed __aligned(8);
613 struct bch_extent_stripe_ptr {
614 #if defined(__LITTLE_ENDIAN_BITFIELD)
619 #elif defined (__BIG_ENDIAN_BITFIELD)
627 struct bch_extent_rebalance {
628 #if defined(__LITTLE_ENDIAN_BITFIELD)
631 compression:8, /* enum bch_compression_opt */
633 #elif defined (__BIG_ENDIAN_BITFIELD)
641 union bch_extent_entry {
642 #if __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__ || __BITS_PER_LONG == 64
644 #elif __BITS_PER_LONG == 32
650 #error edit for your odd byteorder.
653 #define x(f, n) struct bch_extent_##f f;
654 BCH_EXTENT_ENTRY_TYPES()
658 struct bch_btree_ptr {
662 struct bch_extent_ptr start[];
663 } __packed __aligned(8);
665 struct bch_btree_ptr_v2 {
670 __le16 sectors_written;
674 struct bch_extent_ptr start[];
675 } __packed __aligned(8);
677 LE16_BITMASK(BTREE_PTR_RANGE_UPDATED, struct bch_btree_ptr_v2, flags, 0, 1);
683 union bch_extent_entry start[];
684 } __packed __aligned(8);
686 struct bch_reservation {
692 } __packed __aligned(8);
694 /* Maximum size (in u64s) a single pointer could be: */
695 #define BKEY_EXTENT_PTR_U64s_MAX\
696 ((sizeof(struct bch_extent_crc128) + \
697 sizeof(struct bch_extent_ptr)) / sizeof(__u64))
699 /* Maximum possible size of an entire extent value: */
700 #define BKEY_EXTENT_VAL_U64s_MAX \
701 (1 + BKEY_EXTENT_PTR_U64s_MAX * (BCH_REPLICAS_MAX + 1))
703 /* * Maximum possible size of an entire extent, key + value: */
704 #define BKEY_EXTENT_U64s_MAX (BKEY_U64s + BKEY_EXTENT_VAL_U64s_MAX)
706 /* Btree pointers don't carry around checksums: */
707 #define BKEY_BTREE_PTR_VAL_U64s_MAX \
708 ((sizeof(struct bch_btree_ptr_v2) + \
709 sizeof(struct bch_extent_ptr) * BCH_REPLICAS_MAX) / sizeof(__u64))
710 #define BKEY_BTREE_PTR_U64s_MAX \
711 (BKEY_U64s + BKEY_BTREE_PTR_VAL_U64s_MAX)
715 #define BLOCKDEV_INODE_MAX 4096
717 #define BCACHEFS_ROOT_INO 4096
726 } __packed __aligned(8);
728 struct bch_inode_v2 {
731 __le64 bi_journal_seq;
736 } __packed __aligned(8);
738 struct bch_inode_v3 {
741 __le64 bi_journal_seq;
748 } __packed __aligned(8);
750 #define INODEv3_FIELDS_START_INITIAL 6
751 #define INODEv3_FIELDS_START_CUR (offsetof(struct bch_inode_v3, fields) / sizeof(__u64))
753 struct bch_inode_generation {
756 __le32 bi_generation;
758 } __packed __aligned(8);
761 * bi_subvol and bi_parent_subvol are only set for subvolume roots:
764 #define BCH_INODE_FIELDS_v2() \
774 x(bi_generation, 32) \
776 x(bi_data_checksum, 8) \
777 x(bi_compression, 8) \
779 x(bi_background_compression, 8) \
780 x(bi_data_replicas, 8) \
781 x(bi_promote_target, 16) \
782 x(bi_foreground_target, 16) \
783 x(bi_background_target, 16) \
784 x(bi_erasure_code, 16) \
785 x(bi_fields_set, 16) \
787 x(bi_dir_offset, 64) \
789 x(bi_parent_subvol, 32)
791 #define BCH_INODE_FIELDS_v3() \
799 x(bi_generation, 32) \
801 x(bi_data_checksum, 8) \
802 x(bi_compression, 8) \
804 x(bi_background_compression, 8) \
805 x(bi_data_replicas, 8) \
806 x(bi_promote_target, 16) \
807 x(bi_foreground_target, 16) \
808 x(bi_background_target, 16) \
809 x(bi_erasure_code, 16) \
810 x(bi_fields_set, 16) \
812 x(bi_dir_offset, 64) \
814 x(bi_parent_subvol, 32) \
817 /* subset of BCH_INODE_FIELDS */
818 #define BCH_INODE_OPTS() \
819 x(data_checksum, 8) \
822 x(background_compression, 8) \
823 x(data_replicas, 8) \
824 x(promote_target, 16) \
825 x(foreground_target, 16) \
826 x(background_target, 16) \
827 x(erasure_code, 16) \
831 #define x(name, ...) \
838 #define BCH_INODE_FLAGS() \
845 x(i_sectors_dirty, 6) \
847 x(backptr_untrusted, 8)
849 /* bits 20+ reserved for packed fields below: */
851 enum bch_inode_flags {
852 #define x(t, n) BCH_INODE_##t = 1U << n,
857 enum __bch_inode_flags {
858 #define x(t, n) __BCH_INODE_##t = n,
863 LE32_BITMASK(INODE_STR_HASH, struct bch_inode, bi_flags, 20, 24);
864 LE32_BITMASK(INODE_NR_FIELDS, struct bch_inode, bi_flags, 24, 31);
865 LE32_BITMASK(INODE_NEW_VARINT, struct bch_inode, bi_flags, 31, 32);
867 LE64_BITMASK(INODEv2_STR_HASH, struct bch_inode_v2, bi_flags, 20, 24);
868 LE64_BITMASK(INODEv2_NR_FIELDS, struct bch_inode_v2, bi_flags, 24, 31);
870 LE64_BITMASK(INODEv3_STR_HASH, struct bch_inode_v3, bi_flags, 20, 24);
871 LE64_BITMASK(INODEv3_NR_FIELDS, struct bch_inode_v3, bi_flags, 24, 31);
873 LE64_BITMASK(INODEv3_FIELDS_START,
874 struct bch_inode_v3, bi_flags, 31, 36);
875 LE64_BITMASK(INODEv3_MODE, struct bch_inode_v3, bi_flags, 36, 52);
880 * Dirents (and xattrs) have to implement string lookups; since our b-tree
881 * doesn't support arbitrary length strings for the key, we instead index by a
882 * 64 bit hash (currently truncated sha1) of the string, stored in the offset
883 * field of the key - using linear probing to resolve hash collisions. This also
884 * provides us with the readdir cookie posix requires.
886 * Linear probing requires us to use whiteouts for deletions, in the event of a
893 /* Target inode number: */
896 struct { /* DT_SUBVOL */
897 __le32 d_child_subvol;
898 __le32 d_parent_subvol;
903 * Copy of mode bits 12-15 from the target inode - so userspace can get
904 * the filetype without having to do a stat()
909 } __packed __aligned(8);
912 #define BCH_DT_MAX 17
914 #define BCH_NAME_MAX 512
918 #define KEY_TYPE_XATTR_INDEX_USER 0
919 #define KEY_TYPE_XATTR_INDEX_POSIX_ACL_ACCESS 1
920 #define KEY_TYPE_XATTR_INDEX_POSIX_ACL_DEFAULT 2
921 #define KEY_TYPE_XATTR_INDEX_TRUSTED 3
922 #define KEY_TYPE_XATTR_INDEX_SECURITY 4
930 } __packed __aligned(8);
932 /* Bucket/allocation information: */
939 } __packed __aligned(8);
941 #define BCH_ALLOC_FIELDS_V1() \
945 x(dirty_sectors, 16) \
946 x(cached_sectors, 16) \
949 x(stripe_redundancy, 8)
952 #define x(name, _bits) BCH_ALLOC_FIELD_V1_##name,
953 BCH_ALLOC_FIELDS_V1()
957 struct bch_alloc_v2 {
964 } __packed __aligned(8);
966 #define BCH_ALLOC_FIELDS_V2() \
969 x(dirty_sectors, 32) \
970 x(cached_sectors, 32) \
972 x(stripe_redundancy, 8)
974 struct bch_alloc_v3 {
983 } __packed __aligned(8);
985 LE32_BITMASK(BCH_ALLOC_V3_NEED_DISCARD,struct bch_alloc_v3, flags, 0, 1)
986 LE32_BITMASK(BCH_ALLOC_V3_NEED_INC_GEN,struct bch_alloc_v3, flags, 1, 2)
988 struct bch_alloc_v4 {
995 __u8 stripe_redundancy;
997 __u32 cached_sectors;
1000 __u32 nr_external_backpointers;
1001 __u64 fragmentation_lru;
1002 } __packed __aligned(8);
1004 #define BCH_ALLOC_V4_U64s_V0 6
1005 #define BCH_ALLOC_V4_U64s (sizeof(struct bch_alloc_v4) / sizeof(__u64))
1007 BITMASK(BCH_ALLOC_V4_NEED_DISCARD, struct bch_alloc_v4, flags, 0, 1)
1008 BITMASK(BCH_ALLOC_V4_NEED_INC_GEN, struct bch_alloc_v4, flags, 1, 2)
1009 BITMASK(BCH_ALLOC_V4_BACKPOINTERS_START,struct bch_alloc_v4, flags, 2, 8)
1010 BITMASK(BCH_ALLOC_V4_NR_BACKPOINTERS, struct bch_alloc_v4, flags, 8, 14)
1012 #define BCH_ALLOC_V4_NR_BACKPOINTERS_MAX 40
1014 struct bch_backpointer {
1019 __u64 bucket_offset:40;
1022 } __packed __aligned(8);
1024 #define KEY_TYPE_BUCKET_GENS_BITS 8
1025 #define KEY_TYPE_BUCKET_GENS_NR (1U << KEY_TYPE_BUCKET_GENS_BITS)
1026 #define KEY_TYPE_BUCKET_GENS_MASK (KEY_TYPE_BUCKET_GENS_NR - 1)
1028 struct bch_bucket_gens {
1030 u8 gens[KEY_TYPE_BUCKET_GENS_NR];
1031 } __packed __aligned(8);
1042 enum quota_counters {
1048 struct bch_quota_counter {
1055 struct bch_quota_counter c[Q_COUNTERS];
1056 } __packed __aligned(8);
1058 /* Erasure coding */
1067 __u8 csum_granularity_bits;
1071 struct bch_extent_ptr ptrs[];
1072 } __packed __aligned(8);
1076 struct bch_reflink_p {
1080 * A reflink pointer might point to an indirect extent which is then
1081 * later split (by copygc or rebalance). If we only pointed to part of
1082 * the original indirect extent, and then one of the fragments is
1083 * outside the range we point to, we'd leak a refcount: so when creating
1084 * reflink pointers, we need to store pad values to remember the full
1085 * range we were taking a reference on.
1089 } __packed __aligned(8);
1091 struct bch_reflink_v {
1094 union bch_extent_entry start[0];
1096 } __packed __aligned(8);
1098 struct bch_indirect_inline_data {
1106 struct bch_inline_data {
1113 #define SUBVOL_POS_MIN POS(0, 1)
1114 #define SUBVOL_POS_MAX POS(0, S32_MAX)
1115 #define BCACHEFS_ROOT_SUBVOL 1
1117 struct bch_subvolume {
1123 * Snapshot subvolumes form a tree, separate from the snapshot nodes
1124 * tree - if this subvolume is a snapshot, this is the ID of the
1125 * subvolume it was created from:
1132 LE32_BITMASK(BCH_SUBVOLUME_RO, struct bch_subvolume, flags, 0, 1)
1134 * We need to know whether a subvolume is a snapshot so we can know whether we
1135 * can delete it (or whether it should just be rm -rf'd)
1137 LE32_BITMASK(BCH_SUBVOLUME_SNAP, struct bch_subvolume, flags, 1, 2)
1138 LE32_BITMASK(BCH_SUBVOLUME_UNLINKED, struct bch_subvolume, flags, 2, 3)
1142 struct bch_snapshot {
1148 /* corresponds to a bch_snapshot_tree in BTREE_ID_snapshot_trees */
1154 LE32_BITMASK(BCH_SNAPSHOT_DELETED, struct bch_snapshot, flags, 0, 1)
1156 /* True if a subvolume points to this snapshot node: */
1157 LE32_BITMASK(BCH_SNAPSHOT_SUBVOL, struct bch_snapshot, flags, 1, 2)
1162 * The snapshot_trees btree gives us persistent indentifier for each tree of
1163 * bch_snapshot nodes, and allow us to record and easily find the root/master
1164 * subvolume that other snapshots were created from:
1166 struct bch_snapshot_tree {
1168 __le32 master_subvol;
1169 __le32 root_snapshot;
1177 } __packed __aligned(8);
1179 #define LRU_ID_STRIPES (1U << 16)
1181 /* Logged operations btree: */
1183 struct bch_logged_op_truncate {
1191 enum logged_op_finsert_state {
1192 LOGGED_OP_FINSERT_start,
1193 LOGGED_OP_FINSERT_shift_extents,
1194 LOGGED_OP_FINSERT_finish,
1197 struct bch_logged_op_finsert {
1208 /* Optional/variable size superblock sections: */
1210 struct bch_sb_field {
1216 #define BCH_SB_FIELDS() \
1225 x(journal_seq_blacklist, 8) \
1233 enum bch_sb_field_type {
1234 #define x(f, nr) BCH_SB_FIELD_##f = nr,
1241 * Most superblock fields are replicated in all device's superblocks - a few are
1244 #define BCH_SINGLE_DEVICE_SB_FIELDS \
1245 ((1U << BCH_SB_FIELD_journal)| \
1246 (1U << BCH_SB_FIELD_journal_v2))
1248 /* BCH_SB_FIELD_journal: */
1250 struct bch_sb_field_journal {
1251 struct bch_sb_field field;
1255 struct bch_sb_field_journal_v2 {
1256 struct bch_sb_field field;
1258 struct bch_sb_field_journal_v2_entry {
1264 /* BCH_SB_FIELD_members_v1: */
1266 #define BCH_MIN_NR_NBUCKETS (1 << 6)
1268 #define BCH_IOPS_MEASUREMENTS() \
1274 enum bch_iops_measurement {
1275 #define x(t, n) BCH_IOPS_##t = n,
1276 BCH_IOPS_MEASUREMENTS()
1281 #define BCH_MEMBER_ERROR_TYPES() \
1286 enum bch_member_error_type {
1287 #define x(t, n) BCH_MEMBER_ERROR_##t = n,
1288 BCH_MEMBER_ERROR_TYPES()
1295 __le64 nbuckets; /* device size */
1296 __le16 first_bucket; /* index of first bucket used */
1297 __le16 bucket_size; /* sectors */
1299 __le64 last_mount; /* time_t */
1303 __le64 errors[BCH_MEMBER_ERROR_NR];
1304 __le64 errors_at_reset[BCH_MEMBER_ERROR_NR];
1305 __le64 errors_reset_time;
1309 #define BCH_MEMBER_V1_BYTES 56
1311 LE64_BITMASK(BCH_MEMBER_STATE, struct bch_member, flags, 0, 4)
1312 /* 4-14 unused, was TIER, HAS_(META)DATA, REPLACEMENT */
1313 LE64_BITMASK(BCH_MEMBER_DISCARD, struct bch_member, flags, 14, 15)
1314 LE64_BITMASK(BCH_MEMBER_DATA_ALLOWED, struct bch_member, flags, 15, 20)
1315 LE64_BITMASK(BCH_MEMBER_GROUP, struct bch_member, flags, 20, 28)
1316 LE64_BITMASK(BCH_MEMBER_DURABILITY, struct bch_member, flags, 28, 30)
1317 LE64_BITMASK(BCH_MEMBER_FREESPACE_INITIALIZED,
1318 struct bch_member, flags, 30, 31)
1321 LE64_BITMASK(BCH_MEMBER_NR_READ_ERRORS, struct bch_member, flags[1], 0, 20);
1322 LE64_BITMASK(BCH_MEMBER_NR_WRITE_ERRORS,struct bch_member, flags[1], 20, 40);
1325 #define BCH_MEMBER_STATES() \
1331 enum bch_member_state {
1332 #define x(t, n) BCH_MEMBER_STATE_##t = n,
1338 struct bch_sb_field_members_v1 {
1339 struct bch_sb_field field;
1340 struct bch_member _members[]; //Members are now variable size
1343 struct bch_sb_field_members_v2 {
1344 struct bch_sb_field field;
1345 __le16 member_bytes; //size of single member entry
1347 struct bch_member _members[];
1350 /* BCH_SB_FIELD_crypt: */
1360 #define BCH_KEY_MAGIC \
1361 (((__u64) 'b' << 0)|((__u64) 'c' << 8)| \
1362 ((__u64) 'h' << 16)|((__u64) '*' << 24)| \
1363 ((__u64) '*' << 32)|((__u64) 'k' << 40)| \
1364 ((__u64) 'e' << 48)|((__u64) 'y' << 56))
1366 struct bch_encrypted_key {
1372 * If this field is present in the superblock, it stores an encryption key which
1373 * is used encrypt all other data/metadata. The key will normally be encrypted
1374 * with the key userspace provides, but if encryption has been turned off we'll
1375 * just store the master key unencrypted in the superblock so we can access the
1376 * previously encrypted data.
1378 struct bch_sb_field_crypt {
1379 struct bch_sb_field field;
1383 struct bch_encrypted_key key;
1386 LE64_BITMASK(BCH_CRYPT_KDF_TYPE, struct bch_sb_field_crypt, flags, 0, 4);
1388 enum bch_kdf_types {
1393 /* stored as base 2 log of scrypt params: */
1394 LE64_BITMASK(BCH_KDF_SCRYPT_N, struct bch_sb_field_crypt, kdf_flags, 0, 16);
1395 LE64_BITMASK(BCH_KDF_SCRYPT_R, struct bch_sb_field_crypt, kdf_flags, 16, 32);
1396 LE64_BITMASK(BCH_KDF_SCRYPT_P, struct bch_sb_field_crypt, kdf_flags, 32, 48);
1398 /* BCH_SB_FIELD_replicas: */
1400 #define BCH_DATA_TYPES() \
1409 x(need_gc_gens, 8) \
1412 enum bch_data_type {
1413 #define x(t, n) BCH_DATA_##t,
1419 static inline bool data_type_is_empty(enum bch_data_type type)
1423 case BCH_DATA_need_gc_gens:
1424 case BCH_DATA_need_discard:
1431 static inline bool data_type_is_hidden(enum bch_data_type type)
1435 case BCH_DATA_journal:
1442 struct bch_replicas_entry_v0 {
1448 struct bch_sb_field_replicas_v0 {
1449 struct bch_sb_field field;
1450 struct bch_replicas_entry_v0 entries[];
1451 } __packed __aligned(8);
1453 struct bch_replicas_entry_v1 {
1460 #define replicas_entry_bytes(_i) \
1461 (offsetof(typeof(*(_i)), devs) + (_i)->nr_devs)
1463 struct bch_sb_field_replicas {
1464 struct bch_sb_field field;
1465 struct bch_replicas_entry_v1 entries[];
1466 } __packed __aligned(8);
1468 /* BCH_SB_FIELD_quota: */
1470 struct bch_sb_quota_counter {
1475 struct bch_sb_quota_type {
1477 struct bch_sb_quota_counter c[Q_COUNTERS];
1480 struct bch_sb_field_quota {
1481 struct bch_sb_field field;
1482 struct bch_sb_quota_type q[QTYP_NR];
1483 } __packed __aligned(8);
1485 /* BCH_SB_FIELD_disk_groups: */
1487 #define BCH_SB_LABEL_SIZE 32
1489 struct bch_disk_group {
1490 __u8 label[BCH_SB_LABEL_SIZE];
1492 } __packed __aligned(8);
1494 LE64_BITMASK(BCH_GROUP_DELETED, struct bch_disk_group, flags[0], 0, 1)
1495 LE64_BITMASK(BCH_GROUP_DATA_ALLOWED, struct bch_disk_group, flags[0], 1, 6)
1496 LE64_BITMASK(BCH_GROUP_PARENT, struct bch_disk_group, flags[0], 6, 24)
1498 struct bch_sb_field_disk_groups {
1499 struct bch_sb_field field;
1500 struct bch_disk_group entries[];
1501 } __packed __aligned(8);
1503 /* BCH_SB_FIELD_counters */
1505 #define BCH_PERSISTENT_COUNTERS() \
1509 x(bucket_invalidate, 3) \
1510 x(bucket_discard, 4) \
1511 x(bucket_alloc, 5) \
1512 x(bucket_alloc_fail, 6) \
1513 x(btree_cache_scan, 7) \
1514 x(btree_cache_reap, 8) \
1515 x(btree_cache_cannibalize, 9) \
1516 x(btree_cache_cannibalize_lock, 10) \
1517 x(btree_cache_cannibalize_lock_fail, 11) \
1518 x(btree_cache_cannibalize_unlock, 12) \
1519 x(btree_node_write, 13) \
1520 x(btree_node_read, 14) \
1521 x(btree_node_compact, 15) \
1522 x(btree_node_merge, 16) \
1523 x(btree_node_split, 17) \
1524 x(btree_node_rewrite, 18) \
1525 x(btree_node_alloc, 19) \
1526 x(btree_node_free, 20) \
1527 x(btree_node_set_root, 21) \
1528 x(btree_path_relock_fail, 22) \
1529 x(btree_path_upgrade_fail, 23) \
1530 x(btree_reserve_get_fail, 24) \
1531 x(journal_entry_full, 25) \
1532 x(journal_full, 26) \
1533 x(journal_reclaim_finish, 27) \
1534 x(journal_reclaim_start, 28) \
1535 x(journal_write, 29) \
1536 x(read_promote, 30) \
1537 x(read_bounce, 31) \
1540 x(read_reuse_race, 34) \
1541 x(move_extent_read, 35) \
1542 x(move_extent_write, 36) \
1543 x(move_extent_finish, 37) \
1544 x(move_extent_fail, 38) \
1545 x(move_extent_start_fail, 39) \
1547 x(copygc_wait, 41) \
1548 x(gc_gens_end, 42) \
1549 x(gc_gens_start, 43) \
1550 x(trans_blocked_journal_reclaim, 44) \
1551 x(trans_restart_btree_node_reused, 45) \
1552 x(trans_restart_btree_node_split, 46) \
1553 x(trans_restart_fault_inject, 47) \
1554 x(trans_restart_iter_upgrade, 48) \
1555 x(trans_restart_journal_preres_get, 49) \
1556 x(trans_restart_journal_reclaim, 50) \
1557 x(trans_restart_journal_res_get, 51) \
1558 x(trans_restart_key_cache_key_realloced, 52) \
1559 x(trans_restart_key_cache_raced, 53) \
1560 x(trans_restart_mark_replicas, 54) \
1561 x(trans_restart_mem_realloced, 55) \
1562 x(trans_restart_memory_allocation_failure, 56) \
1563 x(trans_restart_relock, 57) \
1564 x(trans_restart_relock_after_fill, 58) \
1565 x(trans_restart_relock_key_cache_fill, 59) \
1566 x(trans_restart_relock_next_node, 60) \
1567 x(trans_restart_relock_parent_for_fill, 61) \
1568 x(trans_restart_relock_path, 62) \
1569 x(trans_restart_relock_path_intent, 63) \
1570 x(trans_restart_too_many_iters, 64) \
1571 x(trans_restart_traverse, 65) \
1572 x(trans_restart_upgrade, 66) \
1573 x(trans_restart_would_deadlock, 67) \
1574 x(trans_restart_would_deadlock_write, 68) \
1575 x(trans_restart_injected, 69) \
1576 x(trans_restart_key_cache_upgrade, 70) \
1577 x(trans_traverse_all, 71) \
1578 x(transaction_commit, 72) \
1579 x(write_super, 73) \
1580 x(trans_restart_would_deadlock_recursion_limit, 74) \
1581 x(trans_restart_write_buffer_flush, 75) \
1582 x(trans_restart_split_race, 76) \
1583 x(write_buffer_flush_slowpath, 77) \
1584 x(write_buffer_flush_sync, 78)
1586 enum bch_persistent_counters {
1587 #define x(t, n, ...) BCH_COUNTER_##t,
1588 BCH_PERSISTENT_COUNTERS()
1593 struct bch_sb_field_counters {
1594 struct bch_sb_field field;
1599 * On clean shutdown, store btree roots and current journal sequence number in
1606 __u8 type; /* designates what this jset holds */
1609 struct bkey_i start[0];
1613 struct bch_sb_field_clean {
1614 struct bch_sb_field field;
1617 __le16 _read_clock; /* no longer used */
1618 __le16 _write_clock;
1621 struct jset_entry start[0];
1625 struct journal_seq_blacklist_entry {
1630 struct bch_sb_field_journal_seq_blacklist {
1631 struct bch_sb_field field;
1632 struct journal_seq_blacklist_entry start[];
1635 struct bch_sb_field_errors {
1636 struct bch_sb_field field;
1637 struct bch_sb_field_error_entry {
1639 __le64 last_error_time;
1643 LE64_BITMASK(BCH_SB_ERROR_ENTRY_ID, struct bch_sb_field_error_entry, v, 0, 16);
1644 LE64_BITMASK(BCH_SB_ERROR_ENTRY_NR, struct bch_sb_field_error_entry, v, 16, 64);
1646 struct bch_sb_field_ext {
1647 struct bch_sb_field field;
1648 __le64 recovery_passes_required[2];
1649 __le64 errors_silent[8];
1652 struct bch_sb_field_downgrade_entry {
1654 __le64 recovery_passes[2];
1656 __le16 errors[] __counted_by(nr_errors);
1657 } __packed __aligned(2);
1659 struct bch_sb_field_downgrade {
1660 struct bch_sb_field field;
1661 struct bch_sb_field_downgrade_entry entries[];
1667 * New versioning scheme:
1668 * One common version number for all on disk data structures - superblock, btree
1669 * nodes, journal entries
1671 #define BCH_VERSION_MAJOR(_v) ((__u16) ((_v) >> 10))
1672 #define BCH_VERSION_MINOR(_v) ((__u16) ((_v) & ~(~0U << 10)))
1673 #define BCH_VERSION(_major, _minor) (((_major) << 10)|(_minor) << 0)
1676 * field 1: version name
1677 * field 2: BCH_VERSION(major, minor)
1678 * field 3: recovery passess required on upgrade
1680 #define BCH_METADATA_VERSIONS() \
1681 x(bkey_renumber, BCH_VERSION(0, 10)) \
1682 x(inode_btree_change, BCH_VERSION(0, 11)) \
1683 x(snapshot, BCH_VERSION(0, 12)) \
1684 x(inode_backpointers, BCH_VERSION(0, 13)) \
1685 x(btree_ptr_sectors_written, BCH_VERSION(0, 14)) \
1686 x(snapshot_2, BCH_VERSION(0, 15)) \
1687 x(reflink_p_fix, BCH_VERSION(0, 16)) \
1688 x(subvol_dirent, BCH_VERSION(0, 17)) \
1689 x(inode_v2, BCH_VERSION(0, 18)) \
1690 x(freespace, BCH_VERSION(0, 19)) \
1691 x(alloc_v4, BCH_VERSION(0, 20)) \
1692 x(new_data_types, BCH_VERSION(0, 21)) \
1693 x(backpointers, BCH_VERSION(0, 22)) \
1694 x(inode_v3, BCH_VERSION(0, 23)) \
1695 x(unwritten_extents, BCH_VERSION(0, 24)) \
1696 x(bucket_gens, BCH_VERSION(0, 25)) \
1697 x(lru_v2, BCH_VERSION(0, 26)) \
1698 x(fragmentation_lru, BCH_VERSION(0, 27)) \
1699 x(no_bps_in_alloc_keys, BCH_VERSION(0, 28)) \
1700 x(snapshot_trees, BCH_VERSION(0, 29)) \
1701 x(major_minor, BCH_VERSION(1, 0)) \
1702 x(snapshot_skiplists, BCH_VERSION(1, 1)) \
1703 x(deleted_inodes, BCH_VERSION(1, 2)) \
1704 x(rebalance_work, BCH_VERSION(1, 3)) \
1705 x(member_seq, BCH_VERSION(1, 4))
1707 enum bcachefs_metadata_version {
1708 bcachefs_metadata_version_min = 9,
1709 #define x(t, n) bcachefs_metadata_version_##t = n,
1710 BCH_METADATA_VERSIONS()
1712 bcachefs_metadata_version_max
1715 static const __maybe_unused
1716 unsigned bcachefs_metadata_required_upgrade_below = bcachefs_metadata_version_rebalance_work;
1718 #define bcachefs_metadata_version_current (bcachefs_metadata_version_max - 1)
1720 #define BCH_SB_SECTOR 8
1721 #define BCH_SB_MEMBERS_MAX 64 /* XXX kill */
1723 struct bch_sb_layout {
1724 __uuid_t magic; /* bcachefs superblock UUID */
1726 __u8 sb_max_size_bits; /* base 2 of 512 byte sectors */
1727 __u8 nr_superblocks;
1729 __le64 sb_offset[61];
1730 } __packed __aligned(8);
1732 #define BCH_SB_LAYOUT_SECTOR 7
1735 * @offset - sector where this sb was written
1736 * @version - on disk format version
1737 * @version_min - Oldest metadata version this filesystem contains; so we can
1738 * safely drop compatibility code and refuse to mount filesystems
1740 * @magic - identifies as a bcachefs superblock (BCHFS_MAGIC)
1741 * @seq - incremented each time superblock is written
1742 * @uuid - used for generating various magic numbers and identifying
1743 * member devices, never changes
1744 * @user_uuid - user visible UUID, may be changed
1745 * @label - filesystem label
1746 * @seq - identifies most recent superblock, incremented each time
1747 * superblock is written
1748 * @features - enabled incompatible features
1751 struct bch_csum csum;
1758 __u8 label[BCH_SB_LABEL_SIZE];
1767 __le64 time_base_lo;
1768 __le32 time_base_hi;
1769 __le32 time_precision;
1776 struct bch_sb_layout layout;
1778 struct bch_sb_field start[0];
1780 } __packed __aligned(8);
1784 * BCH_SB_INITALIZED - set on first mount
1785 * BCH_SB_CLEAN - did we shut down cleanly? Just a hint, doesn't affect
1786 * behaviour of mount/recovery path:
1787 * BCH_SB_INODE_32BIT - limit inode numbers to 32 bits
1788 * BCH_SB_128_BIT_MACS - 128 bit macs instead of 80
1789 * BCH_SB_ENCRYPTION_TYPE - if nonzero encryption is enabled; overrides
1790 * DATA/META_CSUM_TYPE. Also indicates encryption
1791 * algorithm in use, if/when we get more than one
1794 LE16_BITMASK(BCH_SB_BLOCK_SIZE, struct bch_sb, block_size, 0, 16);
1796 LE64_BITMASK(BCH_SB_INITIALIZED, struct bch_sb, flags[0], 0, 1);
1797 LE64_BITMASK(BCH_SB_CLEAN, struct bch_sb, flags[0], 1, 2);
1798 LE64_BITMASK(BCH_SB_CSUM_TYPE, struct bch_sb, flags[0], 2, 8);
1799 LE64_BITMASK(BCH_SB_ERROR_ACTION, struct bch_sb, flags[0], 8, 12);
1801 LE64_BITMASK(BCH_SB_BTREE_NODE_SIZE, struct bch_sb, flags[0], 12, 28);
1803 LE64_BITMASK(BCH_SB_GC_RESERVE, struct bch_sb, flags[0], 28, 33);
1804 LE64_BITMASK(BCH_SB_ROOT_RESERVE, struct bch_sb, flags[0], 33, 40);
1806 LE64_BITMASK(BCH_SB_META_CSUM_TYPE, struct bch_sb, flags[0], 40, 44);
1807 LE64_BITMASK(BCH_SB_DATA_CSUM_TYPE, struct bch_sb, flags[0], 44, 48);
1809 LE64_BITMASK(BCH_SB_META_REPLICAS_WANT, struct bch_sb, flags[0], 48, 52);
1810 LE64_BITMASK(BCH_SB_DATA_REPLICAS_WANT, struct bch_sb, flags[0], 52, 56);
1812 LE64_BITMASK(BCH_SB_POSIX_ACL, struct bch_sb, flags[0], 56, 57);
1813 LE64_BITMASK(BCH_SB_USRQUOTA, struct bch_sb, flags[0], 57, 58);
1814 LE64_BITMASK(BCH_SB_GRPQUOTA, struct bch_sb, flags[0], 58, 59);
1815 LE64_BITMASK(BCH_SB_PRJQUOTA, struct bch_sb, flags[0], 59, 60);
1817 LE64_BITMASK(BCH_SB_HAS_ERRORS, struct bch_sb, flags[0], 60, 61);
1818 LE64_BITMASK(BCH_SB_HAS_TOPOLOGY_ERRORS,struct bch_sb, flags[0], 61, 62);
1820 LE64_BITMASK(BCH_SB_BIG_ENDIAN, struct bch_sb, flags[0], 62, 63);
1822 LE64_BITMASK(BCH_SB_STR_HASH_TYPE, struct bch_sb, flags[1], 0, 4);
1823 LE64_BITMASK(BCH_SB_COMPRESSION_TYPE_LO,struct bch_sb, flags[1], 4, 8);
1824 LE64_BITMASK(BCH_SB_INODE_32BIT, struct bch_sb, flags[1], 8, 9);
1826 LE64_BITMASK(BCH_SB_128_BIT_MACS, struct bch_sb, flags[1], 9, 10);
1827 LE64_BITMASK(BCH_SB_ENCRYPTION_TYPE, struct bch_sb, flags[1], 10, 14);
1830 * Max size of an extent that may require bouncing to read or write
1831 * (checksummed, compressed): 64k
1833 LE64_BITMASK(BCH_SB_ENCODED_EXTENT_MAX_BITS,
1834 struct bch_sb, flags[1], 14, 20);
1836 LE64_BITMASK(BCH_SB_META_REPLICAS_REQ, struct bch_sb, flags[1], 20, 24);
1837 LE64_BITMASK(BCH_SB_DATA_REPLICAS_REQ, struct bch_sb, flags[1], 24, 28);
1839 LE64_BITMASK(BCH_SB_PROMOTE_TARGET, struct bch_sb, flags[1], 28, 40);
1840 LE64_BITMASK(BCH_SB_FOREGROUND_TARGET, struct bch_sb, flags[1], 40, 52);
1841 LE64_BITMASK(BCH_SB_BACKGROUND_TARGET, struct bch_sb, flags[1], 52, 64);
1843 LE64_BITMASK(BCH_SB_BACKGROUND_COMPRESSION_TYPE_LO,
1844 struct bch_sb, flags[2], 0, 4);
1845 LE64_BITMASK(BCH_SB_GC_RESERVE_BYTES, struct bch_sb, flags[2], 4, 64);
1847 LE64_BITMASK(BCH_SB_ERASURE_CODE, struct bch_sb, flags[3], 0, 16);
1848 LE64_BITMASK(BCH_SB_METADATA_TARGET, struct bch_sb, flags[3], 16, 28);
1849 LE64_BITMASK(BCH_SB_SHARD_INUMS, struct bch_sb, flags[3], 28, 29);
1850 LE64_BITMASK(BCH_SB_INODES_USE_KEY_CACHE,struct bch_sb, flags[3], 29, 30);
1851 LE64_BITMASK(BCH_SB_JOURNAL_FLUSH_DELAY,struct bch_sb, flags[3], 30, 62);
1852 LE64_BITMASK(BCH_SB_JOURNAL_FLUSH_DISABLED,struct bch_sb, flags[3], 62, 63);
1853 LE64_BITMASK(BCH_SB_JOURNAL_RECLAIM_DELAY,struct bch_sb, flags[4], 0, 32);
1854 LE64_BITMASK(BCH_SB_JOURNAL_TRANSACTION_NAMES,struct bch_sb, flags[4], 32, 33);
1855 LE64_BITMASK(BCH_SB_NOCOW, struct bch_sb, flags[4], 33, 34);
1856 LE64_BITMASK(BCH_SB_WRITE_BUFFER_SIZE, struct bch_sb, flags[4], 34, 54);
1857 LE64_BITMASK(BCH_SB_VERSION_UPGRADE, struct bch_sb, flags[4], 54, 56);
1859 LE64_BITMASK(BCH_SB_COMPRESSION_TYPE_HI,struct bch_sb, flags[4], 56, 60);
1860 LE64_BITMASK(BCH_SB_BACKGROUND_COMPRESSION_TYPE_HI,
1861 struct bch_sb, flags[4], 60, 64);
1863 LE64_BITMASK(BCH_SB_VERSION_UPGRADE_COMPLETE,
1864 struct bch_sb, flags[5], 0, 16);
1866 static inline __u64 BCH_SB_COMPRESSION_TYPE(const struct bch_sb *sb)
1868 return BCH_SB_COMPRESSION_TYPE_LO(sb) | (BCH_SB_COMPRESSION_TYPE_HI(sb) << 4);
1871 static inline void SET_BCH_SB_COMPRESSION_TYPE(struct bch_sb *sb, __u64 v)
1873 SET_BCH_SB_COMPRESSION_TYPE_LO(sb, v);
1874 SET_BCH_SB_COMPRESSION_TYPE_HI(sb, v >> 4);
1877 static inline __u64 BCH_SB_BACKGROUND_COMPRESSION_TYPE(const struct bch_sb *sb)
1879 return BCH_SB_BACKGROUND_COMPRESSION_TYPE_LO(sb) |
1880 (BCH_SB_BACKGROUND_COMPRESSION_TYPE_HI(sb) << 4);
1883 static inline void SET_BCH_SB_BACKGROUND_COMPRESSION_TYPE(struct bch_sb *sb, __u64 v)
1885 SET_BCH_SB_BACKGROUND_COMPRESSION_TYPE_LO(sb, v);
1886 SET_BCH_SB_BACKGROUND_COMPRESSION_TYPE_HI(sb, v >> 4);
1892 * journal_seq_blacklist_v3: gates BCH_SB_FIELD_journal_seq_blacklist
1893 * reflink: gates KEY_TYPE_reflink
1894 * inline_data: gates KEY_TYPE_inline_data
1895 * new_siphash: gates BCH_STR_HASH_siphash
1896 * new_extent_overwrite: gates BTREE_NODE_NEW_EXTENT_OVERWRITE
1898 #define BCH_SB_FEATURES() \
1902 x(atomic_nlink, 3) \
1904 x(journal_seq_blacklist_v3, 5) \
1908 x(new_extent_overwrite, 9) \
1909 x(incompressible, 10) \
1910 x(btree_ptr_v2, 11) \
1911 x(extents_above_btree_updates, 12) \
1912 x(btree_updates_journalled, 13) \
1913 x(reflink_inline_data, 14) \
1915 x(journal_no_flush, 16) \
1917 x(extents_across_btree_nodes, 18)
1919 #define BCH_SB_FEATURES_ALWAYS \
1920 ((1ULL << BCH_FEATURE_new_extent_overwrite)| \
1921 (1ULL << BCH_FEATURE_extents_above_btree_updates)|\
1922 (1ULL << BCH_FEATURE_btree_updates_journalled)|\
1923 (1ULL << BCH_FEATURE_alloc_v2)|\
1924 (1ULL << BCH_FEATURE_extents_across_btree_nodes))
1926 #define BCH_SB_FEATURES_ALL \
1927 (BCH_SB_FEATURES_ALWAYS| \
1928 (1ULL << BCH_FEATURE_new_siphash)| \
1929 (1ULL << BCH_FEATURE_btree_ptr_v2)| \
1930 (1ULL << BCH_FEATURE_new_varint)| \
1931 (1ULL << BCH_FEATURE_journal_no_flush))
1933 enum bch_sb_feature {
1934 #define x(f, n) BCH_FEATURE_##f,
1940 #define BCH_SB_COMPAT() \
1942 x(alloc_metadata, 1) \
1943 x(extents_above_btree_updates_done, 2) \
1944 x(bformat_overflow_done, 3)
1946 enum bch_sb_compat {
1947 #define x(f, n) BCH_COMPAT_##f,
1955 #define BCH_VERSION_UPGRADE_OPTS() \
1957 x(incompatible, 1) \
1960 enum bch_version_upgrade_opts {
1961 #define x(t, n) BCH_VERSION_UPGRADE_##t = n,
1962 BCH_VERSION_UPGRADE_OPTS()
1966 #define BCH_REPLICAS_MAX 4U
1968 #define BCH_BKEY_PTRS_MAX 16U
1970 #define BCH_ERROR_ACTIONS() \
1975 enum bch_error_actions {
1976 #define x(t, n) BCH_ON_ERROR_##t = n,
1982 #define BCH_STR_HASH_TYPES() \
1988 enum bch_str_hash_type {
1989 #define x(t, n) BCH_STR_HASH_##t = n,
1990 BCH_STR_HASH_TYPES()
1995 #define BCH_STR_HASH_OPTS() \
2000 enum bch_str_hash_opts {
2001 #define x(t, n) BCH_STR_HASH_OPT_##t = n,
2007 #define BCH_CSUM_TYPES() \
2009 x(crc32c_nonzero, 1) \
2010 x(crc64_nonzero, 2) \
2011 x(chacha20_poly1305_80, 3) \
2012 x(chacha20_poly1305_128, 4) \
2017 enum bch_csum_type {
2018 #define x(t, n) BCH_CSUM_##t = n,
2024 static const __maybe_unused unsigned bch_crc_bytes[] = {
2025 [BCH_CSUM_none] = 0,
2026 [BCH_CSUM_crc32c_nonzero] = 4,
2027 [BCH_CSUM_crc32c] = 4,
2028 [BCH_CSUM_crc64_nonzero] = 8,
2029 [BCH_CSUM_crc64] = 8,
2030 [BCH_CSUM_xxhash] = 8,
2031 [BCH_CSUM_chacha20_poly1305_80] = 10,
2032 [BCH_CSUM_chacha20_poly1305_128] = 16,
2035 static inline _Bool bch2_csum_type_is_encryption(enum bch_csum_type type)
2038 case BCH_CSUM_chacha20_poly1305_80:
2039 case BCH_CSUM_chacha20_poly1305_128:
2046 #define BCH_CSUM_OPTS() \
2052 enum bch_csum_opts {
2053 #define x(t, n) BCH_CSUM_OPT_##t = n,
2059 #define BCH_COMPRESSION_TYPES() \
2065 x(incompressible, 5)
2067 enum bch_compression_type {
2068 #define x(t, n) BCH_COMPRESSION_TYPE_##t = n,
2069 BCH_COMPRESSION_TYPES()
2071 BCH_COMPRESSION_TYPE_NR
2074 #define BCH_COMPRESSION_OPTS() \
2080 enum bch_compression_opts {
2081 #define x(t, n) BCH_COMPRESSION_OPT_##t = n,
2082 BCH_COMPRESSION_OPTS()
2084 BCH_COMPRESSION_OPT_NR
2090 * The various other data structures have their own magic numbers, which are
2091 * xored with the first part of the cache set's UUID
2094 #define BCACHE_MAGIC \
2095 UUID_INIT(0xc68573f6, 0x4e1a, 0x45ca, \
2096 0x82, 0x65, 0xf5, 0x7f, 0x48, 0xba, 0x6d, 0x81)
2097 #define BCHFS_MAGIC \
2098 UUID_INIT(0xc68573f6, 0x66ce, 0x90a9, \
2099 0xd9, 0x6a, 0x60, 0xcf, 0x80, 0x3d, 0xf7, 0xef)
2101 #define BCACHEFS_STATFS_MAGIC 0xca451a4e
2103 #define JSET_MAGIC __cpu_to_le64(0x245235c1a3625032ULL)
2104 #define BSET_MAGIC __cpu_to_le64(0x90135c78b99e07f5ULL)
2106 static inline __le64 __bch2_sb_magic(struct bch_sb *sb)
2110 memcpy(&ret, &sb->uuid, sizeof(ret));
2114 static inline __u64 __jset_magic(struct bch_sb *sb)
2116 return __le64_to_cpu(__bch2_sb_magic(sb) ^ JSET_MAGIC);
2119 static inline __u64 __bset_magic(struct bch_sb *sb)
2121 return __le64_to_cpu(__bch2_sb_magic(sb) ^ BSET_MAGIC);
2126 #define JSET_KEYS_U64s (sizeof(struct jset_entry) / sizeof(__u64))
2128 #define BCH_JSET_ENTRY_TYPES() \
2133 x(blacklist_v2, 4) \
2140 x(write_buffer_keys, 11)
2143 #define x(f, nr) BCH_JSET_ENTRY_##f = nr,
2144 BCH_JSET_ENTRY_TYPES()
2149 static inline bool jset_entry_is_key(struct jset_entry *e)
2152 case BCH_JSET_ENTRY_btree_keys:
2153 case BCH_JSET_ENTRY_btree_root:
2154 case BCH_JSET_ENTRY_overwrite:
2155 case BCH_JSET_ENTRY_write_buffer_keys:
2163 * Journal sequence numbers can be blacklisted: bsets record the max sequence
2164 * number of all the journal entries they contain updates for, so that on
2165 * recovery we can ignore those bsets that contain index updates newer that what
2166 * made it into the journal.
2168 * This means that we can't reuse that journal_seq - we have to skip it, and
2169 * then record that we skipped it so that the next time we crash and recover we
2170 * don't think there was a missing journal entry.
2172 struct jset_entry_blacklist {
2173 struct jset_entry entry;
2177 struct jset_entry_blacklist_v2 {
2178 struct jset_entry entry;
2183 #define BCH_FS_USAGE_TYPES() \
2189 #define x(f, nr) BCH_FS_USAGE_##f = nr,
2190 BCH_FS_USAGE_TYPES()
2195 struct jset_entry_usage {
2196 struct jset_entry entry;
2200 struct jset_entry_data_usage {
2201 struct jset_entry entry;
2203 struct bch_replicas_entry_v1 r;
2206 struct jset_entry_clock {
2207 struct jset_entry entry;
2213 struct jset_entry_dev_usage_type {
2219 struct jset_entry_dev_usage {
2220 struct jset_entry entry;
2224 __le64 _buckets_ec; /* No longer used */
2225 __le64 _buckets_unavailable; /* No longer used */
2227 struct jset_entry_dev_usage_type d[];
2230 static inline unsigned jset_entry_dev_usage_nr_types(struct jset_entry_dev_usage *u)
2232 return (vstruct_bytes(&u->entry) - sizeof(struct jset_entry_dev_usage)) /
2233 sizeof(struct jset_entry_dev_usage_type);
2236 struct jset_entry_log {
2237 struct jset_entry entry;
2239 } __packed __aligned(8);
2242 * On disk format for a journal entry:
2243 * seq is monotonically increasing; every journal entry has its own unique
2246 * last_seq is the oldest journal entry that still has keys the btree hasn't
2247 * flushed to disk yet.
2249 * version is for on disk format changes.
2252 struct bch_csum csum;
2259 __le32 u64s; /* size of d[] in u64s */
2261 __u8 encrypted_start[0];
2263 __le16 _read_clock; /* no longer used */
2264 __le16 _write_clock;
2266 /* Sequence number of oldest dirty journal entry */
2270 struct jset_entry start[0];
2272 } __packed __aligned(8);
2274 LE32_BITMASK(JSET_CSUM_TYPE, struct jset, flags, 0, 4);
2275 LE32_BITMASK(JSET_BIG_ENDIAN, struct jset, flags, 4, 5);
2276 LE32_BITMASK(JSET_NO_FLUSH, struct jset, flags, 5, 6);
2278 #define BCH_JOURNAL_BUCKETS_MIN 8
2282 enum btree_id_flags {
2283 BTREE_ID_EXTENTS = BIT(0),
2284 BTREE_ID_SNAPSHOTS = BIT(1),
2285 BTREE_ID_SNAPSHOT_FIELD = BIT(2),
2286 BTREE_ID_DATA = BIT(3),
2289 #define BCH_BTREE_IDS() \
2290 x(extents, 0, BTREE_ID_EXTENTS|BTREE_ID_SNAPSHOTS|BTREE_ID_DATA,\
2291 BIT_ULL(KEY_TYPE_whiteout)| \
2292 BIT_ULL(KEY_TYPE_error)| \
2293 BIT_ULL(KEY_TYPE_cookie)| \
2294 BIT_ULL(KEY_TYPE_extent)| \
2295 BIT_ULL(KEY_TYPE_reservation)| \
2296 BIT_ULL(KEY_TYPE_reflink_p)| \
2297 BIT_ULL(KEY_TYPE_inline_data)) \
2298 x(inodes, 1, BTREE_ID_SNAPSHOTS, \
2299 BIT_ULL(KEY_TYPE_whiteout)| \
2300 BIT_ULL(KEY_TYPE_inode)| \
2301 BIT_ULL(KEY_TYPE_inode_v2)| \
2302 BIT_ULL(KEY_TYPE_inode_v3)| \
2303 BIT_ULL(KEY_TYPE_inode_generation)) \
2304 x(dirents, 2, BTREE_ID_SNAPSHOTS, \
2305 BIT_ULL(KEY_TYPE_whiteout)| \
2306 BIT_ULL(KEY_TYPE_hash_whiteout)| \
2307 BIT_ULL(KEY_TYPE_dirent)) \
2308 x(xattrs, 3, BTREE_ID_SNAPSHOTS, \
2309 BIT_ULL(KEY_TYPE_whiteout)| \
2310 BIT_ULL(KEY_TYPE_cookie)| \
2311 BIT_ULL(KEY_TYPE_hash_whiteout)| \
2312 BIT_ULL(KEY_TYPE_xattr)) \
2314 BIT_ULL(KEY_TYPE_alloc)| \
2315 BIT_ULL(KEY_TYPE_alloc_v2)| \
2316 BIT_ULL(KEY_TYPE_alloc_v3)| \
2317 BIT_ULL(KEY_TYPE_alloc_v4)) \
2319 BIT_ULL(KEY_TYPE_quota)) \
2321 BIT_ULL(KEY_TYPE_stripe)) \
2322 x(reflink, 7, BTREE_ID_EXTENTS|BTREE_ID_DATA, \
2323 BIT_ULL(KEY_TYPE_reflink_v)| \
2324 BIT_ULL(KEY_TYPE_indirect_inline_data)) \
2325 x(subvolumes, 8, 0, \
2326 BIT_ULL(KEY_TYPE_subvolume)) \
2327 x(snapshots, 9, 0, \
2328 BIT_ULL(KEY_TYPE_snapshot)) \
2330 BIT_ULL(KEY_TYPE_set)) \
2331 x(freespace, 11, BTREE_ID_EXTENTS, \
2332 BIT_ULL(KEY_TYPE_set)) \
2333 x(need_discard, 12, 0, \
2334 BIT_ULL(KEY_TYPE_set)) \
2335 x(backpointers, 13, 0, \
2336 BIT_ULL(KEY_TYPE_backpointer)) \
2337 x(bucket_gens, 14, 0, \
2338 BIT_ULL(KEY_TYPE_bucket_gens)) \
2339 x(snapshot_trees, 15, 0, \
2340 BIT_ULL(KEY_TYPE_snapshot_tree)) \
2341 x(deleted_inodes, 16, BTREE_ID_SNAPSHOT_FIELD, \
2342 BIT_ULL(KEY_TYPE_set)) \
2343 x(logged_ops, 17, 0, \
2344 BIT_ULL(KEY_TYPE_logged_op_truncate)| \
2345 BIT_ULL(KEY_TYPE_logged_op_finsert)) \
2346 x(rebalance_work, 18, BTREE_ID_SNAPSHOT_FIELD, \
2347 BIT_ULL(KEY_TYPE_set)|BIT_ULL(KEY_TYPE_cookie))
2350 #define x(name, nr, ...) BTREE_ID_##name = nr,
2356 #define BTREE_MAX_DEPTH 4U
2363 * On disk a btree node is a list/log of these; within each set the keys are
2370 * Highest journal entry this bset contains keys for.
2371 * If on recovery we don't see that journal entry, this bset is ignored:
2372 * this allows us to preserve the order of all index updates after a
2373 * crash, since the journal records a total order of all index updates
2374 * and anything that didn't make it to the journal doesn't get used.
2380 __le16 u64s; /* count of d[] in u64s */
2382 struct bkey_packed start[0];
2384 } __packed __aligned(8);
2386 LE32_BITMASK(BSET_CSUM_TYPE, struct bset, flags, 0, 4);
2388 LE32_BITMASK(BSET_BIG_ENDIAN, struct bset, flags, 4, 5);
2389 LE32_BITMASK(BSET_SEPARATE_WHITEOUTS,
2390 struct bset, flags, 5, 6);
2392 /* Sector offset within the btree node: */
2393 LE32_BITMASK(BSET_OFFSET, struct bset, flags, 16, 32);
2396 struct bch_csum csum;
2399 /* this flags field is encrypted, unlike bset->flags: */
2402 /* Closed interval: */
2403 struct bpos min_key;
2404 struct bpos max_key;
2405 struct bch_extent_ptr _ptr; /* not used anymore */
2406 struct bkey_format format;
2417 } __packed __aligned(8);
2419 LE64_BITMASK(BTREE_NODE_ID_LO, struct btree_node, flags, 0, 4);
2420 LE64_BITMASK(BTREE_NODE_LEVEL, struct btree_node, flags, 4, 8);
2421 LE64_BITMASK(BTREE_NODE_NEW_EXTENT_OVERWRITE,
2422 struct btree_node, flags, 8, 9);
2423 LE64_BITMASK(BTREE_NODE_ID_HI, struct btree_node, flags, 9, 25);
2425 LE64_BITMASK(BTREE_NODE_SEQ, struct btree_node, flags, 32, 64);
2427 static inline __u64 BTREE_NODE_ID(struct btree_node *n)
2429 return BTREE_NODE_ID_LO(n) | (BTREE_NODE_ID_HI(n) << 4);
2432 static inline void SET_BTREE_NODE_ID(struct btree_node *n, __u64 v)
2434 SET_BTREE_NODE_ID_LO(n, v);
2435 SET_BTREE_NODE_ID_HI(n, v >> 4);
2438 struct btree_node_entry {
2439 struct bch_csum csum;
2449 } __packed __aligned(8);
2451 #endif /* _BCACHEFS_FORMAT_H */