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>
81 #define BITMASK(name, type, field, offset, end) \
82 static const unsigned name##_OFFSET = offset; \
83 static const unsigned name##_BITS = (end - offset); \
85 static inline __u64 name(const type *k) \
87 return (k->field >> offset) & ~(~0ULL << (end - offset)); \
90 static inline void SET_##name(type *k, __u64 v) \
92 k->field &= ~(~(~0ULL << (end - offset)) << offset); \
93 k->field |= (v & ~(~0ULL << (end - offset))) << offset; \
96 #define LE_BITMASK(_bits, name, type, field, offset, end) \
97 static const unsigned name##_OFFSET = offset; \
98 static const unsigned name##_BITS = (end - offset); \
99 static const __u##_bits name##_MAX = (1ULL << (end - offset)) - 1; \
101 static inline __u64 name(const type *k) \
103 return (__le##_bits##_to_cpu(k->field) >> offset) & \
104 ~(~0ULL << (end - offset)); \
107 static inline void SET_##name(type *k, __u64 v) \
109 __u##_bits new = __le##_bits##_to_cpu(k->field); \
111 new &= ~(~(~0ULL << (end - offset)) << offset); \
112 new |= (v & ~(~0ULL << (end - offset))) << offset; \
113 k->field = __cpu_to_le##_bits(new); \
116 #define LE16_BITMASK(n, t, f, o, e) LE_BITMASK(16, n, t, f, o, e)
117 #define LE32_BITMASK(n, t, f, o, e) LE_BITMASK(32, n, t, f, o, e)
118 #define LE64_BITMASK(n, t, f, o, e) LE_BITMASK(64, n, t, f, o, e)
123 /* One unused slot for now: */
124 __u8 bits_per_field[6];
125 __le64 field_offset[6];
128 /* Btree keys - all units are in sectors */
132 * Word order matches machine byte order - btree code treats a bpos as a
133 * single large integer, for search/comparison purposes
135 * Note that wherever a bpos is embedded in another on disk data
136 * structure, it has to be byte swabbed when reading in metadata that
137 * wasn't written in native endian order:
139 #if __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__
143 #elif __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__
145 __u64 offset; /* Points to end of extent - sectors */
148 #error edit for your odd byteorder.
150 } __packed __aligned(4);
152 #define KEY_INODE_MAX ((__u64)~0ULL)
153 #define KEY_OFFSET_MAX ((__u64)~0ULL)
154 #define KEY_SNAPSHOT_MAX ((__u32)~0U)
155 #define KEY_SIZE_MAX ((__u32)~0U)
157 static inline struct bpos SPOS(__u64 inode, __u64 offset, __u32 snapshot)
159 return (struct bpos) {
162 .snapshot = snapshot,
166 #define POS_MIN SPOS(0, 0, 0)
167 #define POS_MAX SPOS(KEY_INODE_MAX, KEY_OFFSET_MAX, 0)
168 #define SPOS_MAX SPOS(KEY_INODE_MAX, KEY_OFFSET_MAX, KEY_SNAPSHOT_MAX)
169 #define POS(_inode, _offset) SPOS(_inode, _offset, 0)
171 /* Empty placeholder struct, for container_of() */
177 #if __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__
180 #elif __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__
184 } __packed __aligned(4);
187 /* Size of combined key and value, in u64s */
190 /* Format of key (0 for format local to btree node) */
191 #if defined(__LITTLE_ENDIAN_BITFIELD)
194 #elif defined (__BIG_ENDIAN_BITFIELD)
195 __u8 needs_whiteout:1,
198 #error edit for your odd byteorder.
201 /* Type of the value */
204 #if __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__
207 struct bversion version;
208 __u32 size; /* extent size, in sectors */
210 #elif __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__
212 __u32 size; /* extent size, in sectors */
213 struct bversion version;
217 } __packed __aligned(8);
222 /* Size of combined key and value, in u64s */
225 /* Format of key (0 for format local to btree node) */
228 * XXX: next incompat on disk format change, switch format and
229 * needs_whiteout - bkey_packed() will be cheaper if format is the high
230 * bits of the bitfield
232 #if defined(__LITTLE_ENDIAN_BITFIELD)
235 #elif defined (__BIG_ENDIAN_BITFIELD)
236 __u8 needs_whiteout:1,
240 /* Type of the value */
245 * We copy bkeys with struct assignment in various places, and while
246 * that shouldn't be done with packed bkeys we can't disallow it in C,
247 * and it's legal to cast a bkey to a bkey_packed - so padding it out
248 * to the same size as struct bkey should hopefully be safest.
250 __u8 pad[sizeof(struct bkey) - 3];
251 } __packed __aligned(8);
258 #define BKEY_U64s (sizeof(struct bkey) / sizeof(__u64))
259 #define BKEY_U64s_MAX U8_MAX
260 #define BKEY_VAL_U64s_MAX (BKEY_U64s_MAX - BKEY_U64s)
262 #define KEY_PACKED_BITS_START 24
264 #define KEY_FORMAT_LOCAL_BTREE 0
265 #define KEY_FORMAT_CURRENT 1
267 enum bch_bkey_fields {
272 BKEY_FIELD_VERSION_HI,
273 BKEY_FIELD_VERSION_LO,
277 #define bkey_format_field(name, field) \
278 [BKEY_FIELD_##name] = (sizeof(((struct bkey *) NULL)->field) * 8)
280 #define BKEY_FORMAT_CURRENT \
281 ((struct bkey_format) { \
282 .key_u64s = BKEY_U64s, \
283 .nr_fields = BKEY_NR_FIELDS, \
284 .bits_per_field = { \
285 bkey_format_field(INODE, p.inode), \
286 bkey_format_field(OFFSET, p.offset), \
287 bkey_format_field(SNAPSHOT, p.snapshot), \
288 bkey_format_field(SIZE, size), \
289 bkey_format_field(VERSION_HI, version.hi), \
290 bkey_format_field(VERSION_LO, version.lo), \
294 /* bkey with inline value */
302 #define KEY(_inode, _offset, _size) \
305 .format = KEY_FORMAT_CURRENT, \
306 .p = POS(_inode, _offset), \
310 static inline void bkey_init(struct bkey *k)
315 #define bkey_bytes(_k) ((_k)->u64s * sizeof(__u64))
317 #define __BKEY_PADDED(key, pad) \
318 struct bkey_i key; __u64 key ## _pad[pad]
321 * - DELETED keys are used internally to mark keys that should be ignored but
322 * override keys in composition order. Their version number is ignored.
324 * - DISCARDED keys indicate that the data is all 0s because it has been
325 * discarded. DISCARDs may have a version; if the version is nonzero the key
326 * will be persistent, otherwise the key will be dropped whenever the btree
327 * node is rewritten (like DELETED keys).
329 * - ERROR: any read of the data returns a read error, as the data was lost due
330 * to a failing device. Like DISCARDED keys, they can be removed (overridden)
331 * by new writes or cluster-wide GC. Node repair can also overwrite them with
332 * the same or a more recent version number, but not with an older version
335 * - WHITEOUT: for hash table btrees
337 #define BCH_BKEY_TYPES() \
342 x(hash_whiteout, 4) \
347 x(inode_generation, 9) \
356 x(btree_ptr_v2, 18) \
357 x(indirect_inline_data, 19) \
372 #define x(name, nr) KEY_TYPE_##name = nr,
382 struct bch_whiteout {
395 struct bch_hash_whiteout {
406 * In extent bkeys, the value is a list of pointers (bch_extent_ptr), optionally
407 * preceded by checksum/compression information (bch_extent_crc32 or
410 * One major determining factor in the format of extents is how we handle and
411 * represent extents that have been partially overwritten and thus trimmed:
413 * If an extent is not checksummed or compressed, when the extent is trimmed we
414 * don't have to remember the extent we originally allocated and wrote: we can
415 * merely adjust ptr->offset to point to the start of the data that is currently
416 * live. The size field in struct bkey records the current (live) size of the
417 * extent, and is also used to mean "size of region on disk that we point to" in
420 * Thus an extent that is not checksummed or compressed will consist only of a
421 * list of bch_extent_ptrs, with none of the fields in
422 * bch_extent_crc32/bch_extent_crc64.
424 * When an extent is checksummed or compressed, it's not possible to read only
425 * the data that is currently live: we have to read the entire extent that was
426 * originally written, and then return only the part of the extent that is
429 * Thus, in addition to the current size of the extent in struct bkey, we need
430 * to store the size of the originally allocated space - this is the
431 * compressed_size and uncompressed_size fields in bch_extent_crc32/64. Also,
432 * when the extent is trimmed, instead of modifying the offset field of the
433 * pointer, we keep a second smaller offset field - "offset into the original
434 * extent of the currently live region".
436 * The other major determining factor is replication and data migration:
438 * Each pointer may have its own bch_extent_crc32/64. When doing a replicated
439 * write, we will initially write all the replicas in the same format, with the
440 * same checksum type and compression format - however, when copygc runs later (or
441 * tiering/cache promotion, anything that moves data), it is not in general
442 * going to rewrite all the pointers at once - one of the replicas may be in a
443 * bucket on one device that has very little fragmentation while another lives
444 * in a bucket that has become heavily fragmented, and thus is being rewritten
445 * sooner than the rest.
447 * Thus it will only move a subset of the pointers (or in the case of
448 * tiering/cache promotion perhaps add a single pointer without dropping any
449 * current pointers), and if the extent has been partially overwritten it must
450 * write only the currently live portion (or copygc would not be able to reduce
451 * fragmentation!) - which necessitates a different bch_extent_crc format for
454 * But in the interests of space efficiency, we don't want to store one
455 * bch_extent_crc for each pointer if we don't have to.
457 * Thus, a bch_extent consists of bch_extent_crc32s, bch_extent_crc64s, and
458 * bch_extent_ptrs appended arbitrarily one after the other. We determine the
459 * type of a given entry with a scheme similar to utf8 (except we're encoding a
460 * type, not a size), encoding the type in the position of the first set bit:
462 * bch_extent_crc32 - 0b1
463 * bch_extent_ptr - 0b10
464 * bch_extent_crc64 - 0b100
466 * We do it this way because bch_extent_crc32 is _very_ constrained on bits (and
467 * bch_extent_crc64 is the least constrained).
469 * Then, each bch_extent_crc32/64 applies to the pointers that follow after it,
470 * until the next bch_extent_crc32/64.
472 * If there are no bch_extent_crcs preceding a bch_extent_ptr, then that pointer
473 * is neither checksummed nor compressed.
476 /* 128 bits, sufficient for cryptographic MACs: */
480 } __packed __aligned(8);
482 #define BCH_EXTENT_ENTRY_TYPES() \
488 #define BCH_EXTENT_ENTRY_MAX 5
490 enum bch_extent_entry_type {
491 #define x(f, n) BCH_EXTENT_ENTRY_##f = n,
492 BCH_EXTENT_ENTRY_TYPES()
496 /* Compressed/uncompressed size are stored biased by 1: */
497 struct bch_extent_crc32 {
498 #if defined(__LITTLE_ENDIAN_BITFIELD)
501 _uncompressed_size:7,
507 #elif defined (__BIG_ENDIAN_BITFIELD)
509 __u32 compression_type:4,
513 _uncompressed_size:7,
517 } __packed __aligned(8);
519 #define CRC32_SIZE_MAX (1U << 7)
520 #define CRC32_NONCE_MAX 0
522 struct bch_extent_crc64 {
523 #if defined(__LITTLE_ENDIAN_BITFIELD)
526 _uncompressed_size:9,
532 #elif defined (__BIG_ENDIAN_BITFIELD)
538 _uncompressed_size:9,
543 } __packed __aligned(8);
545 #define CRC64_SIZE_MAX (1U << 9)
546 #define CRC64_NONCE_MAX ((1U << 10) - 1)
548 struct bch_extent_crc128 {
549 #if defined(__LITTLE_ENDIAN_BITFIELD)
552 _uncompressed_size:13,
557 #elif defined (__BIG_ENDIAN_BITFIELD)
558 __u64 compression_type:4,
562 _uncompressed_size:13,
566 struct bch_csum csum;
567 } __packed __aligned(8);
569 #define CRC128_SIZE_MAX (1U << 13)
570 #define CRC128_NONCE_MAX ((1U << 13) - 1)
573 * @reservation - pointer hasn't been written to, just reserved
575 struct bch_extent_ptr {
576 #if defined(__LITTLE_ENDIAN_BITFIELD)
581 offset:44, /* 8 petabytes */
584 #elif defined (__BIG_ENDIAN_BITFIELD)
593 } __packed __aligned(8);
595 struct bch_extent_stripe_ptr {
596 #if defined(__LITTLE_ENDIAN_BITFIELD)
601 #elif defined (__BIG_ENDIAN_BITFIELD)
609 struct bch_extent_reservation {
610 #if defined(__LITTLE_ENDIAN_BITFIELD)
615 #elif defined (__BIG_ENDIAN_BITFIELD)
623 union bch_extent_entry {
624 #if __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__ || __BITS_PER_LONG == 64
626 #elif __BITS_PER_LONG == 32
632 #error edit for your odd byteorder.
635 #define x(f, n) struct bch_extent_##f f;
636 BCH_EXTENT_ENTRY_TYPES()
640 struct bch_btree_ptr {
644 struct bch_extent_ptr start[];
645 } __packed __aligned(8);
647 struct bch_btree_ptr_v2 {
652 __le16 sectors_written;
656 struct bch_extent_ptr start[];
657 } __packed __aligned(8);
659 LE16_BITMASK(BTREE_PTR_RANGE_UPDATED, struct bch_btree_ptr_v2, flags, 0, 1);
665 union bch_extent_entry start[];
666 } __packed __aligned(8);
668 struct bch_reservation {
674 } __packed __aligned(8);
676 /* Maximum size (in u64s) a single pointer could be: */
677 #define BKEY_EXTENT_PTR_U64s_MAX\
678 ((sizeof(struct bch_extent_crc128) + \
679 sizeof(struct bch_extent_ptr)) / sizeof(u64))
681 /* Maximum possible size of an entire extent value: */
682 #define BKEY_EXTENT_VAL_U64s_MAX \
683 (1 + BKEY_EXTENT_PTR_U64s_MAX * (BCH_REPLICAS_MAX + 1))
685 /* * Maximum possible size of an entire extent, key + value: */
686 #define BKEY_EXTENT_U64s_MAX (BKEY_U64s + BKEY_EXTENT_VAL_U64s_MAX)
688 /* Btree pointers don't carry around checksums: */
689 #define BKEY_BTREE_PTR_VAL_U64s_MAX \
690 ((sizeof(struct bch_btree_ptr_v2) + \
691 sizeof(struct bch_extent_ptr) * BCH_REPLICAS_MAX) / sizeof(u64))
692 #define BKEY_BTREE_PTR_U64s_MAX \
693 (BKEY_U64s + BKEY_BTREE_PTR_VAL_U64s_MAX)
697 #define BLOCKDEV_INODE_MAX 4096
699 #define BCACHEFS_ROOT_INO 4096
708 } __packed __aligned(8);
710 struct bch_inode_v2 {
713 __le64 bi_journal_seq;
718 } __packed __aligned(8);
720 struct bch_inode_v3 {
723 __le64 bi_journal_seq;
730 } __packed __aligned(8);
732 #define INODEv3_FIELDS_START_INITIAL 6
733 #define INODEv3_FIELDS_START_CUR (offsetof(struct bch_inode_v3, fields) / sizeof(u64))
735 struct bch_inode_generation {
738 __le32 bi_generation;
740 } __packed __aligned(8);
743 * bi_subvol and bi_parent_subvol are only set for subvolume roots:
746 #define BCH_INODE_FIELDS_v2() \
756 x(bi_generation, 32) \
758 x(bi_data_checksum, 8) \
759 x(bi_compression, 8) \
761 x(bi_background_compression, 8) \
762 x(bi_data_replicas, 8) \
763 x(bi_promote_target, 16) \
764 x(bi_foreground_target, 16) \
765 x(bi_background_target, 16) \
766 x(bi_erasure_code, 16) \
767 x(bi_fields_set, 16) \
769 x(bi_dir_offset, 64) \
771 x(bi_parent_subvol, 32)
773 #define BCH_INODE_FIELDS_v3() \
781 x(bi_generation, 32) \
783 x(bi_data_checksum, 8) \
784 x(bi_compression, 8) \
786 x(bi_background_compression, 8) \
787 x(bi_data_replicas, 8) \
788 x(bi_promote_target, 16) \
789 x(bi_foreground_target, 16) \
790 x(bi_background_target, 16) \
791 x(bi_erasure_code, 16) \
792 x(bi_fields_set, 16) \
794 x(bi_dir_offset, 64) \
796 x(bi_parent_subvol, 32) \
799 /* subset of BCH_INODE_FIELDS */
800 #define BCH_INODE_OPTS() \
801 x(data_checksum, 8) \
804 x(background_compression, 8) \
805 x(data_replicas, 8) \
806 x(promote_target, 16) \
807 x(foreground_target, 16) \
808 x(background_target, 16) \
809 x(erasure_code, 16) \
813 #define x(name, ...) \
822 * User flags (get/settable with FS_IOC_*FLAGS, correspond to FS_*_FL
825 __BCH_INODE_SYNC = 0,
826 __BCH_INODE_IMMUTABLE = 1,
827 __BCH_INODE_APPEND = 2,
828 __BCH_INODE_NODUMP = 3,
829 __BCH_INODE_NOATIME = 4,
831 __BCH_INODE_I_SIZE_DIRTY = 5,
832 __BCH_INODE_I_SECTORS_DIRTY = 6,
833 __BCH_INODE_UNLINKED = 7,
834 __BCH_INODE_BACKPTR_UNTRUSTED = 8,
836 /* bits 20+ reserved for packed fields below: */
839 #define BCH_INODE_SYNC (1 << __BCH_INODE_SYNC)
840 #define BCH_INODE_IMMUTABLE (1 << __BCH_INODE_IMMUTABLE)
841 #define BCH_INODE_APPEND (1 << __BCH_INODE_APPEND)
842 #define BCH_INODE_NODUMP (1 << __BCH_INODE_NODUMP)
843 #define BCH_INODE_NOATIME (1 << __BCH_INODE_NOATIME)
844 #define BCH_INODE_I_SIZE_DIRTY (1 << __BCH_INODE_I_SIZE_DIRTY)
845 #define BCH_INODE_I_SECTORS_DIRTY (1 << __BCH_INODE_I_SECTORS_DIRTY)
846 #define BCH_INODE_UNLINKED (1 << __BCH_INODE_UNLINKED)
847 #define BCH_INODE_BACKPTR_UNTRUSTED (1 << __BCH_INODE_BACKPTR_UNTRUSTED)
849 LE32_BITMASK(INODE_STR_HASH, struct bch_inode, bi_flags, 20, 24);
850 LE32_BITMASK(INODE_NR_FIELDS, struct bch_inode, bi_flags, 24, 31);
851 LE32_BITMASK(INODE_NEW_VARINT, struct bch_inode, bi_flags, 31, 32);
853 LE64_BITMASK(INODEv2_STR_HASH, struct bch_inode_v2, bi_flags, 20, 24);
854 LE64_BITMASK(INODEv2_NR_FIELDS, struct bch_inode_v2, bi_flags, 24, 31);
856 LE64_BITMASK(INODEv3_STR_HASH, struct bch_inode_v3, bi_flags, 20, 24);
857 LE64_BITMASK(INODEv3_NR_FIELDS, struct bch_inode_v3, bi_flags, 24, 31);
859 LE64_BITMASK(INODEv3_FIELDS_START,
860 struct bch_inode_v3, bi_flags, 31, 36);
861 LE64_BITMASK(INODEv3_MODE, struct bch_inode_v3, bi_flags, 36, 52);
866 * Dirents (and xattrs) have to implement string lookups; since our b-tree
867 * doesn't support arbitrary length strings for the key, we instead index by a
868 * 64 bit hash (currently truncated sha1) of the string, stored in the offset
869 * field of the key - using linear probing to resolve hash collisions. This also
870 * provides us with the readdir cookie posix requires.
872 * Linear probing requires us to use whiteouts for deletions, in the event of a
879 /* Target inode number: */
882 struct { /* DT_SUBVOL */
883 __le32 d_child_subvol;
884 __le32 d_parent_subvol;
889 * Copy of mode bits 12-15 from the target inode - so userspace can get
890 * the filetype without having to do a stat()
895 } __packed __aligned(8);
898 #define BCH_DT_MAX 17
900 #define BCH_NAME_MAX ((unsigned) (U8_MAX * sizeof(u64) - \
901 sizeof(struct bkey) - \
902 offsetof(struct bch_dirent, d_name)))
906 #define KEY_TYPE_XATTR_INDEX_USER 0
907 #define KEY_TYPE_XATTR_INDEX_POSIX_ACL_ACCESS 1
908 #define KEY_TYPE_XATTR_INDEX_POSIX_ACL_DEFAULT 2
909 #define KEY_TYPE_XATTR_INDEX_TRUSTED 3
910 #define KEY_TYPE_XATTR_INDEX_SECURITY 4
918 } __packed __aligned(8);
920 /* Bucket/allocation information: */
927 } __packed __aligned(8);
929 #define BCH_ALLOC_FIELDS_V1() \
933 x(dirty_sectors, 16) \
934 x(cached_sectors, 16) \
937 x(stripe_redundancy, 8)
940 #define x(name, _bits) BCH_ALLOC_FIELD_V1_##name,
941 BCH_ALLOC_FIELDS_V1()
945 struct bch_alloc_v2 {
952 } __packed __aligned(8);
954 #define BCH_ALLOC_FIELDS_V2() \
957 x(dirty_sectors, 32) \
958 x(cached_sectors, 32) \
960 x(stripe_redundancy, 8)
962 struct bch_alloc_v3 {
971 } __packed __aligned(8);
973 LE32_BITMASK(BCH_ALLOC_V3_NEED_DISCARD,struct bch_alloc_v3, flags, 0, 1)
974 LE32_BITMASK(BCH_ALLOC_V3_NEED_INC_GEN,struct bch_alloc_v3, flags, 1, 2)
976 struct bch_alloc_v4 {
983 __u8 stripe_redundancy;
985 __u32 cached_sectors;
988 __u32 nr_external_backpointers;
989 __u64 fragmentation_lru;
990 } __packed __aligned(8);
992 #define BCH_ALLOC_V4_U64s_V0 6
993 #define BCH_ALLOC_V4_U64s (sizeof(struct bch_alloc_v4) / sizeof(u64))
995 BITMASK(BCH_ALLOC_V4_NEED_DISCARD, struct bch_alloc_v4, flags, 0, 1)
996 BITMASK(BCH_ALLOC_V4_NEED_INC_GEN, struct bch_alloc_v4, flags, 1, 2)
997 BITMASK(BCH_ALLOC_V4_BACKPOINTERS_START,struct bch_alloc_v4, flags, 2, 8)
998 BITMASK(BCH_ALLOC_V4_NR_BACKPOINTERS, struct bch_alloc_v4, flags, 8, 14)
1000 #define BCH_ALLOC_V4_NR_BACKPOINTERS_MAX 40
1002 struct bch_backpointer {
1007 __u64 bucket_offset:40;
1010 } __packed __aligned(8);
1012 #define KEY_TYPE_BUCKET_GENS_BITS 8
1013 #define KEY_TYPE_BUCKET_GENS_NR (1U << KEY_TYPE_BUCKET_GENS_BITS)
1014 #define KEY_TYPE_BUCKET_GENS_MASK (KEY_TYPE_BUCKET_GENS_NR - 1)
1016 struct bch_bucket_gens {
1018 u8 gens[KEY_TYPE_BUCKET_GENS_NR];
1019 } __packed __aligned(8);
1030 enum quota_counters {
1036 struct bch_quota_counter {
1043 struct bch_quota_counter c[Q_COUNTERS];
1044 } __packed __aligned(8);
1046 /* Erasure coding */
1055 __u8 csum_granularity_bits;
1059 struct bch_extent_ptr ptrs[];
1060 } __packed __aligned(8);
1064 struct bch_reflink_p {
1068 * A reflink pointer might point to an indirect extent which is then
1069 * later split (by copygc or rebalance). If we only pointed to part of
1070 * the original indirect extent, and then one of the fragments is
1071 * outside the range we point to, we'd leak a refcount: so when creating
1072 * reflink pointers, we need to store pad values to remember the full
1073 * range we were taking a reference on.
1077 } __packed __aligned(8);
1079 struct bch_reflink_v {
1082 union bch_extent_entry start[0];
1084 } __packed __aligned(8);
1086 struct bch_indirect_inline_data {
1094 struct bch_inline_data {
1101 #define SUBVOL_POS_MIN POS(0, 1)
1102 #define SUBVOL_POS_MAX POS(0, S32_MAX)
1103 #define BCACHEFS_ROOT_SUBVOL 1
1105 struct bch_subvolume {
1115 LE32_BITMASK(BCH_SUBVOLUME_RO, struct bch_subvolume, flags, 0, 1)
1117 * We need to know whether a subvolume is a snapshot so we can know whether we
1118 * can delete it (or whether it should just be rm -rf'd)
1120 LE32_BITMASK(BCH_SUBVOLUME_SNAP, struct bch_subvolume, flags, 1, 2)
1121 LE32_BITMASK(BCH_SUBVOLUME_UNLINKED, struct bch_subvolume, flags, 2, 3)
1125 struct bch_snapshot {
1134 LE32_BITMASK(BCH_SNAPSHOT_DELETED, struct bch_snapshot, flags, 0, 1)
1136 /* True if a subvolume points to this snapshot node: */
1137 LE32_BITMASK(BCH_SNAPSHOT_SUBVOL, struct bch_snapshot, flags, 1, 2)
1142 * The snapshot_trees btree gives us persistent indentifier for each tree of
1143 * bch_snapshot nodes, and allow us to record and easily find the root/master
1144 * subvolume that other snapshots were created from:
1146 struct bch_snapshot_tree {
1148 __le32 master_subvol;
1149 __le32 root_snapshot;
1157 } __packed __aligned(8);
1159 #define LRU_ID_STRIPES (1U << 16)
1161 /* Optional/variable size superblock sections: */
1163 struct bch_sb_field {
1169 #define BCH_SB_FIELDS() \
1178 x(journal_seq_blacklist, 8) \
1182 enum bch_sb_field_type {
1183 #define x(f, nr) BCH_SB_FIELD_##f = nr,
1190 * Most superblock fields are replicated in all device's superblocks - a few are
1193 #define BCH_SINGLE_DEVICE_SB_FIELDS \
1194 ((1U << BCH_SB_FIELD_journal)| \
1195 (1U << BCH_SB_FIELD_journal_v2))
1197 /* BCH_SB_FIELD_journal: */
1199 struct bch_sb_field_journal {
1200 struct bch_sb_field field;
1204 struct bch_sb_field_journal_v2 {
1205 struct bch_sb_field field;
1207 struct bch_sb_field_journal_v2_entry {
1213 /* BCH_SB_FIELD_members: */
1215 #define BCH_MIN_NR_NBUCKETS (1 << 6)
1219 __le64 nbuckets; /* device size */
1220 __le16 first_bucket; /* index of first bucket used */
1221 __le16 bucket_size; /* sectors */
1223 __le64 last_mount; /* time_t */
1228 LE64_BITMASK(BCH_MEMBER_STATE, struct bch_member, flags[0], 0, 4)
1229 /* 4-14 unused, was TIER, HAS_(META)DATA, REPLACEMENT */
1230 LE64_BITMASK(BCH_MEMBER_DISCARD, struct bch_member, flags[0], 14, 15)
1231 LE64_BITMASK(BCH_MEMBER_DATA_ALLOWED, struct bch_member, flags[0], 15, 20)
1232 LE64_BITMASK(BCH_MEMBER_GROUP, struct bch_member, flags[0], 20, 28)
1233 LE64_BITMASK(BCH_MEMBER_DURABILITY, struct bch_member, flags[0], 28, 30)
1234 LE64_BITMASK(BCH_MEMBER_FREESPACE_INITIALIZED,
1235 struct bch_member, flags[0], 30, 31)
1238 LE64_BITMASK(BCH_MEMBER_NR_READ_ERRORS, struct bch_member, flags[1], 0, 20);
1239 LE64_BITMASK(BCH_MEMBER_NR_WRITE_ERRORS,struct bch_member, flags[1], 20, 40);
1242 #define BCH_MEMBER_STATES() \
1248 enum bch_member_state {
1249 #define x(t, n) BCH_MEMBER_STATE_##t = n,
1255 struct bch_sb_field_members {
1256 struct bch_sb_field field;
1257 struct bch_member members[0];
1260 /* BCH_SB_FIELD_crypt: */
1270 #define BCH_KEY_MAGIC \
1271 (((u64) 'b' << 0)|((u64) 'c' << 8)| \
1272 ((u64) 'h' << 16)|((u64) '*' << 24)| \
1273 ((u64) '*' << 32)|((u64) 'k' << 40)| \
1274 ((u64) 'e' << 48)|((u64) 'y' << 56))
1276 struct bch_encrypted_key {
1282 * If this field is present in the superblock, it stores an encryption key which
1283 * is used encrypt all other data/metadata. The key will normally be encrypted
1284 * with the key userspace provides, but if encryption has been turned off we'll
1285 * just store the master key unencrypted in the superblock so we can access the
1286 * previously encrypted data.
1288 struct bch_sb_field_crypt {
1289 struct bch_sb_field field;
1293 struct bch_encrypted_key key;
1296 LE64_BITMASK(BCH_CRYPT_KDF_TYPE, struct bch_sb_field_crypt, flags, 0, 4);
1298 enum bch_kdf_types {
1303 /* stored as base 2 log of scrypt params: */
1304 LE64_BITMASK(BCH_KDF_SCRYPT_N, struct bch_sb_field_crypt, kdf_flags, 0, 16);
1305 LE64_BITMASK(BCH_KDF_SCRYPT_R, struct bch_sb_field_crypt, kdf_flags, 16, 32);
1306 LE64_BITMASK(BCH_KDF_SCRYPT_P, struct bch_sb_field_crypt, kdf_flags, 32, 48);
1308 /* BCH_SB_FIELD_replicas: */
1310 #define BCH_DATA_TYPES() \
1319 x(need_gc_gens, 8) \
1322 enum bch_data_type {
1323 #define x(t, n) BCH_DATA_##t,
1329 static inline bool data_type_is_empty(enum bch_data_type type)
1333 case BCH_DATA_need_gc_gens:
1334 case BCH_DATA_need_discard:
1341 static inline bool data_type_is_hidden(enum bch_data_type type)
1345 case BCH_DATA_journal:
1352 struct bch_replicas_entry_v0 {
1358 struct bch_sb_field_replicas_v0 {
1359 struct bch_sb_field field;
1360 struct bch_replicas_entry_v0 entries[];
1361 } __packed __aligned(8);
1363 struct bch_replicas_entry {
1370 #define replicas_entry_bytes(_i) \
1371 (offsetof(typeof(*(_i)), devs) + (_i)->nr_devs)
1373 struct bch_sb_field_replicas {
1374 struct bch_sb_field field;
1375 struct bch_replicas_entry entries[];
1376 } __packed __aligned(8);
1378 /* BCH_SB_FIELD_quota: */
1380 struct bch_sb_quota_counter {
1385 struct bch_sb_quota_type {
1387 struct bch_sb_quota_counter c[Q_COUNTERS];
1390 struct bch_sb_field_quota {
1391 struct bch_sb_field field;
1392 struct bch_sb_quota_type q[QTYP_NR];
1393 } __packed __aligned(8);
1395 /* BCH_SB_FIELD_disk_groups: */
1397 #define BCH_SB_LABEL_SIZE 32
1399 struct bch_disk_group {
1400 __u8 label[BCH_SB_LABEL_SIZE];
1402 } __packed __aligned(8);
1404 LE64_BITMASK(BCH_GROUP_DELETED, struct bch_disk_group, flags[0], 0, 1)
1405 LE64_BITMASK(BCH_GROUP_DATA_ALLOWED, struct bch_disk_group, flags[0], 1, 6)
1406 LE64_BITMASK(BCH_GROUP_PARENT, struct bch_disk_group, flags[0], 6, 24)
1408 struct bch_sb_field_disk_groups {
1409 struct bch_sb_field field;
1410 struct bch_disk_group entries[0];
1411 } __packed __aligned(8);
1413 /* BCH_SB_FIELD_counters */
1415 #define BCH_PERSISTENT_COUNTERS() \
1419 x(bucket_invalidate, 3) \
1420 x(bucket_discard, 4) \
1421 x(bucket_alloc, 5) \
1422 x(bucket_alloc_fail, 6) \
1423 x(btree_cache_scan, 7) \
1424 x(btree_cache_reap, 8) \
1425 x(btree_cache_cannibalize, 9) \
1426 x(btree_cache_cannibalize_lock, 10) \
1427 x(btree_cache_cannibalize_lock_fail, 11) \
1428 x(btree_cache_cannibalize_unlock, 12) \
1429 x(btree_node_write, 13) \
1430 x(btree_node_read, 14) \
1431 x(btree_node_compact, 15) \
1432 x(btree_node_merge, 16) \
1433 x(btree_node_split, 17) \
1434 x(btree_node_rewrite, 18) \
1435 x(btree_node_alloc, 19) \
1436 x(btree_node_free, 20) \
1437 x(btree_node_set_root, 21) \
1438 x(btree_path_relock_fail, 22) \
1439 x(btree_path_upgrade_fail, 23) \
1440 x(btree_reserve_get_fail, 24) \
1441 x(journal_entry_full, 25) \
1442 x(journal_full, 26) \
1443 x(journal_reclaim_finish, 27) \
1444 x(journal_reclaim_start, 28) \
1445 x(journal_write, 29) \
1446 x(read_promote, 30) \
1447 x(read_bounce, 31) \
1450 x(read_reuse_race, 34) \
1451 x(move_extent_read, 35) \
1452 x(move_extent_write, 36) \
1453 x(move_extent_finish, 37) \
1454 x(move_extent_fail, 38) \
1455 x(move_extent_alloc_mem_fail, 39) \
1457 x(copygc_wait, 41) \
1458 x(gc_gens_end, 42) \
1459 x(gc_gens_start, 43) \
1460 x(trans_blocked_journal_reclaim, 44) \
1461 x(trans_restart_btree_node_reused, 45) \
1462 x(trans_restart_btree_node_split, 46) \
1463 x(trans_restart_fault_inject, 47) \
1464 x(trans_restart_iter_upgrade, 48) \
1465 x(trans_restart_journal_preres_get, 49) \
1466 x(trans_restart_journal_reclaim, 50) \
1467 x(trans_restart_journal_res_get, 51) \
1468 x(trans_restart_key_cache_key_realloced, 52) \
1469 x(trans_restart_key_cache_raced, 53) \
1470 x(trans_restart_mark_replicas, 54) \
1471 x(trans_restart_mem_realloced, 55) \
1472 x(trans_restart_memory_allocation_failure, 56) \
1473 x(trans_restart_relock, 57) \
1474 x(trans_restart_relock_after_fill, 58) \
1475 x(trans_restart_relock_key_cache_fill, 59) \
1476 x(trans_restart_relock_next_node, 60) \
1477 x(trans_restart_relock_parent_for_fill, 61) \
1478 x(trans_restart_relock_path, 62) \
1479 x(trans_restart_relock_path_intent, 63) \
1480 x(trans_restart_too_many_iters, 64) \
1481 x(trans_restart_traverse, 65) \
1482 x(trans_restart_upgrade, 66) \
1483 x(trans_restart_would_deadlock, 67) \
1484 x(trans_restart_would_deadlock_write, 68) \
1485 x(trans_restart_injected, 69) \
1486 x(trans_restart_key_cache_upgrade, 70) \
1487 x(trans_traverse_all, 71) \
1488 x(transaction_commit, 72) \
1489 x(write_super, 73) \
1490 x(trans_restart_would_deadlock_recursion_limit, 74) \
1491 x(trans_restart_write_buffer_flush, 75) \
1492 x(trans_restart_split_race, 76)
1494 enum bch_persistent_counters {
1495 #define x(t, n, ...) BCH_COUNTER_##t,
1496 BCH_PERSISTENT_COUNTERS()
1501 struct bch_sb_field_counters {
1502 struct bch_sb_field field;
1507 * On clean shutdown, store btree roots and current journal sequence number in
1514 __u8 type; /* designates what this jset holds */
1518 struct bkey_i start[0];
1523 struct bch_sb_field_clean {
1524 struct bch_sb_field field;
1527 __le16 _read_clock; /* no longer used */
1528 __le16 _write_clock;
1532 struct jset_entry start[0];
1537 struct journal_seq_blacklist_entry {
1542 struct bch_sb_field_journal_seq_blacklist {
1543 struct bch_sb_field field;
1546 struct journal_seq_blacklist_entry start[0];
1554 * New versioning scheme:
1555 * One common version number for all on disk data structures - superblock, btree
1556 * nodes, journal entries
1558 #define BCH_JSET_VERSION_OLD 2
1559 #define BCH_BSET_VERSION_OLD 3
1561 #define BCH_METADATA_VERSIONS() \
1562 x(bkey_renumber, 10) \
1563 x(inode_btree_change, 11) \
1565 x(inode_backpointers, 13) \
1566 x(btree_ptr_sectors_written, 14) \
1568 x(reflink_p_fix, 16) \
1569 x(subvol_dirent, 17) \
1573 x(new_data_types, 21) \
1574 x(backpointers, 22) \
1576 x(unwritten_extents, 24) \
1577 x(bucket_gens, 25) \
1579 x(fragmentation_lru, 27) \
1580 x(no_bps_in_alloc_keys, 28) \
1581 x(snapshot_trees, 29)
1583 enum bcachefs_metadata_version {
1584 bcachefs_metadata_version_min = 9,
1585 #define x(t, n) bcachefs_metadata_version_##t = n,
1586 BCH_METADATA_VERSIONS()
1588 bcachefs_metadata_version_max
1591 static const unsigned bcachefs_metadata_required_upgrade_below = bcachefs_metadata_version_snapshot_trees;
1593 #define bcachefs_metadata_version_current (bcachefs_metadata_version_max - 1)
1595 #define BCH_SB_SECTOR 8
1596 #define BCH_SB_MEMBERS_MAX 64 /* XXX kill */
1598 struct bch_sb_layout {
1599 uuid_le magic; /* bcachefs superblock UUID */
1601 __u8 sb_max_size_bits; /* base 2 of 512 byte sectors */
1602 __u8 nr_superblocks;
1604 __le64 sb_offset[61];
1605 } __packed __aligned(8);
1607 #define BCH_SB_LAYOUT_SECTOR 7
1610 * @offset - sector where this sb was written
1611 * @version - on disk format version
1612 * @version_min - Oldest metadata version this filesystem contains; so we can
1613 * safely drop compatibility code and refuse to mount filesystems
1615 * @magic - identifies as a bcachefs superblock (BCHFS_MAGIC)
1616 * @seq - incremented each time superblock is written
1617 * @uuid - used for generating various magic numbers and identifying
1618 * member devices, never changes
1619 * @user_uuid - user visible UUID, may be changed
1620 * @label - filesystem label
1621 * @seq - identifies most recent superblock, incremented each time
1622 * superblock is written
1623 * @features - enabled incompatible features
1626 struct bch_csum csum;
1633 __u8 label[BCH_SB_LABEL_SIZE];
1642 __le64 time_base_lo;
1643 __le32 time_base_hi;
1644 __le32 time_precision;
1650 struct bch_sb_layout layout;
1653 struct bch_sb_field start[0];
1656 } __packed __aligned(8);
1660 * BCH_SB_INITALIZED - set on first mount
1661 * BCH_SB_CLEAN - did we shut down cleanly? Just a hint, doesn't affect
1662 * behaviour of mount/recovery path:
1663 * BCH_SB_INODE_32BIT - limit inode numbers to 32 bits
1664 * BCH_SB_128_BIT_MACS - 128 bit macs instead of 80
1665 * BCH_SB_ENCRYPTION_TYPE - if nonzero encryption is enabled; overrides
1666 * DATA/META_CSUM_TYPE. Also indicates encryption
1667 * algorithm in use, if/when we get more than one
1670 LE16_BITMASK(BCH_SB_BLOCK_SIZE, struct bch_sb, block_size, 0, 16);
1672 LE64_BITMASK(BCH_SB_INITIALIZED, struct bch_sb, flags[0], 0, 1);
1673 LE64_BITMASK(BCH_SB_CLEAN, struct bch_sb, flags[0], 1, 2);
1674 LE64_BITMASK(BCH_SB_CSUM_TYPE, struct bch_sb, flags[0], 2, 8);
1675 LE64_BITMASK(BCH_SB_ERROR_ACTION, struct bch_sb, flags[0], 8, 12);
1677 LE64_BITMASK(BCH_SB_BTREE_NODE_SIZE, struct bch_sb, flags[0], 12, 28);
1679 LE64_BITMASK(BCH_SB_GC_RESERVE, struct bch_sb, flags[0], 28, 33);
1680 LE64_BITMASK(BCH_SB_ROOT_RESERVE, struct bch_sb, flags[0], 33, 40);
1682 LE64_BITMASK(BCH_SB_META_CSUM_TYPE, struct bch_sb, flags[0], 40, 44);
1683 LE64_BITMASK(BCH_SB_DATA_CSUM_TYPE, struct bch_sb, flags[0], 44, 48);
1685 LE64_BITMASK(BCH_SB_META_REPLICAS_WANT, struct bch_sb, flags[0], 48, 52);
1686 LE64_BITMASK(BCH_SB_DATA_REPLICAS_WANT, struct bch_sb, flags[0], 52, 56);
1688 LE64_BITMASK(BCH_SB_POSIX_ACL, struct bch_sb, flags[0], 56, 57);
1689 LE64_BITMASK(BCH_SB_USRQUOTA, struct bch_sb, flags[0], 57, 58);
1690 LE64_BITMASK(BCH_SB_GRPQUOTA, struct bch_sb, flags[0], 58, 59);
1691 LE64_BITMASK(BCH_SB_PRJQUOTA, struct bch_sb, flags[0], 59, 60);
1693 LE64_BITMASK(BCH_SB_HAS_ERRORS, struct bch_sb, flags[0], 60, 61);
1694 LE64_BITMASK(BCH_SB_HAS_TOPOLOGY_ERRORS,struct bch_sb, flags[0], 61, 62);
1696 LE64_BITMASK(BCH_SB_BIG_ENDIAN, struct bch_sb, flags[0], 62, 63);
1698 LE64_BITMASK(BCH_SB_STR_HASH_TYPE, struct bch_sb, flags[1], 0, 4);
1699 LE64_BITMASK(BCH_SB_COMPRESSION_TYPE, struct bch_sb, flags[1], 4, 8);
1700 LE64_BITMASK(BCH_SB_INODE_32BIT, struct bch_sb, flags[1], 8, 9);
1702 LE64_BITMASK(BCH_SB_128_BIT_MACS, struct bch_sb, flags[1], 9, 10);
1703 LE64_BITMASK(BCH_SB_ENCRYPTION_TYPE, struct bch_sb, flags[1], 10, 14);
1706 * Max size of an extent that may require bouncing to read or write
1707 * (checksummed, compressed): 64k
1709 LE64_BITMASK(BCH_SB_ENCODED_EXTENT_MAX_BITS,
1710 struct bch_sb, flags[1], 14, 20);
1712 LE64_BITMASK(BCH_SB_META_REPLICAS_REQ, struct bch_sb, flags[1], 20, 24);
1713 LE64_BITMASK(BCH_SB_DATA_REPLICAS_REQ, struct bch_sb, flags[1], 24, 28);
1715 LE64_BITMASK(BCH_SB_PROMOTE_TARGET, struct bch_sb, flags[1], 28, 40);
1716 LE64_BITMASK(BCH_SB_FOREGROUND_TARGET, struct bch_sb, flags[1], 40, 52);
1717 LE64_BITMASK(BCH_SB_BACKGROUND_TARGET, struct bch_sb, flags[1], 52, 64);
1719 LE64_BITMASK(BCH_SB_BACKGROUND_COMPRESSION_TYPE,
1720 struct bch_sb, flags[2], 0, 4);
1721 LE64_BITMASK(BCH_SB_GC_RESERVE_BYTES, struct bch_sb, flags[2], 4, 64);
1723 LE64_BITMASK(BCH_SB_ERASURE_CODE, struct bch_sb, flags[3], 0, 16);
1724 LE64_BITMASK(BCH_SB_METADATA_TARGET, struct bch_sb, flags[3], 16, 28);
1725 LE64_BITMASK(BCH_SB_SHARD_INUMS, struct bch_sb, flags[3], 28, 29);
1726 LE64_BITMASK(BCH_SB_INODES_USE_KEY_CACHE,struct bch_sb, flags[3], 29, 30);
1727 LE64_BITMASK(BCH_SB_JOURNAL_FLUSH_DELAY,struct bch_sb, flags[3], 30, 62);
1728 LE64_BITMASK(BCH_SB_JOURNAL_FLUSH_DISABLED,struct bch_sb, flags[3], 62, 63);
1729 LE64_BITMASK(BCH_SB_JOURNAL_RECLAIM_DELAY,struct bch_sb, flags[4], 0, 32);
1730 LE64_BITMASK(BCH_SB_JOURNAL_TRANSACTION_NAMES,struct bch_sb, flags[4], 32, 33);
1731 LE64_BITMASK(BCH_SB_NOCOW, struct bch_sb, flags[4], 33, 34);
1732 LE64_BITMASK(BCH_SB_WRITE_BUFFER_SIZE, struct bch_sb, flags[4], 34, 54);
1737 * journal_seq_blacklist_v3: gates BCH_SB_FIELD_journal_seq_blacklist
1738 * reflink: gates KEY_TYPE_reflink
1739 * inline_data: gates KEY_TYPE_inline_data
1740 * new_siphash: gates BCH_STR_HASH_siphash
1741 * new_extent_overwrite: gates BTREE_NODE_NEW_EXTENT_OVERWRITE
1743 #define BCH_SB_FEATURES() \
1747 x(atomic_nlink, 3) \
1749 x(journal_seq_blacklist_v3, 5) \
1753 x(new_extent_overwrite, 9) \
1754 x(incompressible, 10) \
1755 x(btree_ptr_v2, 11) \
1756 x(extents_above_btree_updates, 12) \
1757 x(btree_updates_journalled, 13) \
1758 x(reflink_inline_data, 14) \
1760 x(journal_no_flush, 16) \
1762 x(extents_across_btree_nodes, 18)
1764 #define BCH_SB_FEATURES_ALWAYS \
1765 ((1ULL << BCH_FEATURE_new_extent_overwrite)| \
1766 (1ULL << BCH_FEATURE_extents_above_btree_updates)|\
1767 (1ULL << BCH_FEATURE_btree_updates_journalled)|\
1768 (1ULL << BCH_FEATURE_alloc_v2)|\
1769 (1ULL << BCH_FEATURE_extents_across_btree_nodes))
1771 #define BCH_SB_FEATURES_ALL \
1772 (BCH_SB_FEATURES_ALWAYS| \
1773 (1ULL << BCH_FEATURE_new_siphash)| \
1774 (1ULL << BCH_FEATURE_btree_ptr_v2)| \
1775 (1ULL << BCH_FEATURE_new_varint)| \
1776 (1ULL << BCH_FEATURE_journal_no_flush))
1778 enum bch_sb_feature {
1779 #define x(f, n) BCH_FEATURE_##f,
1785 #define BCH_SB_COMPAT() \
1787 x(alloc_metadata, 1) \
1788 x(extents_above_btree_updates_done, 2) \
1789 x(bformat_overflow_done, 3)
1791 enum bch_sb_compat {
1792 #define x(f, n) BCH_COMPAT_##f,
1800 #define BCH_REPLICAS_MAX 4U
1802 #define BCH_BKEY_PTRS_MAX 16U
1804 #define BCH_ERROR_ACTIONS() \
1809 enum bch_error_actions {
1810 #define x(t, n) BCH_ON_ERROR_##t = n,
1816 #define BCH_STR_HASH_TYPES() \
1822 enum bch_str_hash_type {
1823 #define x(t, n) BCH_STR_HASH_##t = n,
1824 BCH_STR_HASH_TYPES()
1829 #define BCH_STR_HASH_OPTS() \
1834 enum bch_str_hash_opts {
1835 #define x(t, n) BCH_STR_HASH_OPT_##t = n,
1841 #define BCH_CSUM_TYPES() \
1843 x(crc32c_nonzero, 1) \
1844 x(crc64_nonzero, 2) \
1845 x(chacha20_poly1305_80, 3) \
1846 x(chacha20_poly1305_128, 4) \
1851 enum bch_csum_type {
1852 #define x(t, n) BCH_CSUM_##t = n,
1858 static const unsigned bch_crc_bytes[] = {
1859 [BCH_CSUM_none] = 0,
1860 [BCH_CSUM_crc32c_nonzero] = 4,
1861 [BCH_CSUM_crc32c] = 4,
1862 [BCH_CSUM_crc64_nonzero] = 8,
1863 [BCH_CSUM_crc64] = 8,
1864 [BCH_CSUM_xxhash] = 8,
1865 [BCH_CSUM_chacha20_poly1305_80] = 10,
1866 [BCH_CSUM_chacha20_poly1305_128] = 16,
1869 static inline _Bool bch2_csum_type_is_encryption(enum bch_csum_type type)
1872 case BCH_CSUM_chacha20_poly1305_80:
1873 case BCH_CSUM_chacha20_poly1305_128:
1880 #define BCH_CSUM_OPTS() \
1886 enum bch_csum_opts {
1887 #define x(t, n) BCH_CSUM_OPT_##t = n,
1893 #define BCH_COMPRESSION_TYPES() \
1899 x(incompressible, 5)
1901 enum bch_compression_type {
1902 #define x(t, n) BCH_COMPRESSION_TYPE_##t = n,
1903 BCH_COMPRESSION_TYPES()
1905 BCH_COMPRESSION_TYPE_NR
1908 #define BCH_COMPRESSION_OPTS() \
1914 enum bch_compression_opts {
1915 #define x(t, n) BCH_COMPRESSION_OPT_##t = n,
1916 BCH_COMPRESSION_OPTS()
1918 BCH_COMPRESSION_OPT_NR
1924 * The various other data structures have their own magic numbers, which are
1925 * xored with the first part of the cache set's UUID
1928 #define BCACHE_MAGIC \
1929 UUID_LE(0xf67385c6, 0x1a4e, 0xca45, \
1930 0x82, 0x65, 0xf5, 0x7f, 0x48, 0xba, 0x6d, 0x81)
1931 #define BCHFS_MAGIC \
1932 UUID_LE(0xf67385c6, 0xce66, 0xa990, \
1933 0xd9, 0x6a, 0x60, 0xcf, 0x80, 0x3d, 0xf7, 0xef)
1935 #define BCACHEFS_STATFS_MAGIC 0xca451a4e
1937 #define JSET_MAGIC __cpu_to_le64(0x245235c1a3625032ULL)
1938 #define BSET_MAGIC __cpu_to_le64(0x90135c78b99e07f5ULL)
1940 static inline __le64 __bch2_sb_magic(struct bch_sb *sb)
1944 memcpy(&ret, &sb->uuid, sizeof(ret));
1948 static inline __u64 __jset_magic(struct bch_sb *sb)
1950 return __le64_to_cpu(__bch2_sb_magic(sb) ^ JSET_MAGIC);
1953 static inline __u64 __bset_magic(struct bch_sb *sb)
1955 return __le64_to_cpu(__bch2_sb_magic(sb) ^ BSET_MAGIC);
1960 #define JSET_KEYS_U64s (sizeof(struct jset_entry) / sizeof(__u64))
1962 #define BCH_JSET_ENTRY_TYPES() \
1967 x(blacklist_v2, 4) \
1976 #define x(f, nr) BCH_JSET_ENTRY_##f = nr,
1977 BCH_JSET_ENTRY_TYPES()
1983 * Journal sequence numbers can be blacklisted: bsets record the max sequence
1984 * number of all the journal entries they contain updates for, so that on
1985 * recovery we can ignore those bsets that contain index updates newer that what
1986 * made it into the journal.
1988 * This means that we can't reuse that journal_seq - we have to skip it, and
1989 * then record that we skipped it so that the next time we crash and recover we
1990 * don't think there was a missing journal entry.
1992 struct jset_entry_blacklist {
1993 struct jset_entry entry;
1997 struct jset_entry_blacklist_v2 {
1998 struct jset_entry entry;
2003 #define BCH_FS_USAGE_TYPES() \
2009 #define x(f, nr) BCH_FS_USAGE_##f = nr,
2010 BCH_FS_USAGE_TYPES()
2015 struct jset_entry_usage {
2016 struct jset_entry entry;
2020 struct jset_entry_data_usage {
2021 struct jset_entry entry;
2023 struct bch_replicas_entry r;
2026 struct jset_entry_clock {
2027 struct jset_entry entry;
2033 struct jset_entry_dev_usage_type {
2039 struct jset_entry_dev_usage {
2040 struct jset_entry entry;
2045 __le64 _buckets_unavailable; /* No longer used */
2047 struct jset_entry_dev_usage_type d[];
2050 static inline unsigned jset_entry_dev_usage_nr_types(struct jset_entry_dev_usage *u)
2052 return (vstruct_bytes(&u->entry) - sizeof(struct jset_entry_dev_usage)) /
2053 sizeof(struct jset_entry_dev_usage_type);
2056 struct jset_entry_log {
2057 struct jset_entry entry;
2062 * On disk format for a journal entry:
2063 * seq is monotonically increasing; every journal entry has its own unique
2066 * last_seq is the oldest journal entry that still has keys the btree hasn't
2067 * flushed to disk yet.
2069 * version is for on disk format changes.
2072 struct bch_csum csum;
2079 __le32 u64s; /* size of d[] in u64s */
2081 __u8 encrypted_start[0];
2083 __le16 _read_clock; /* no longer used */
2084 __le16 _write_clock;
2086 /* Sequence number of oldest dirty journal entry */
2091 struct jset_entry start[0];
2094 } __packed __aligned(8);
2096 LE32_BITMASK(JSET_CSUM_TYPE, struct jset, flags, 0, 4);
2097 LE32_BITMASK(JSET_BIG_ENDIAN, struct jset, flags, 4, 5);
2098 LE32_BITMASK(JSET_NO_FLUSH, struct jset, flags, 5, 6);
2100 #define BCH_JOURNAL_BUCKETS_MIN 8
2104 #define BCH_BTREE_IDS() \
2117 x(need_discard, 12) \
2118 x(backpointers, 13) \
2119 x(bucket_gens, 14) \
2120 x(snapshot_trees, 15)
2123 #define x(kwd, val) BTREE_ID_##kwd = val,
2129 #define BTREE_MAX_DEPTH 4U
2136 * On disk a btree node is a list/log of these; within each set the keys are
2143 * Highest journal entry this bset contains keys for.
2144 * If on recovery we don't see that journal entry, this bset is ignored:
2145 * this allows us to preserve the order of all index updates after a
2146 * crash, since the journal records a total order of all index updates
2147 * and anything that didn't make it to the journal doesn't get used.
2153 __le16 u64s; /* count of d[] in u64s */
2156 struct bkey_packed start[0];
2159 } __packed __aligned(8);
2161 LE32_BITMASK(BSET_CSUM_TYPE, struct bset, flags, 0, 4);
2163 LE32_BITMASK(BSET_BIG_ENDIAN, struct bset, flags, 4, 5);
2164 LE32_BITMASK(BSET_SEPARATE_WHITEOUTS,
2165 struct bset, flags, 5, 6);
2167 /* Sector offset within the btree node: */
2168 LE32_BITMASK(BSET_OFFSET, struct bset, flags, 16, 32);
2171 struct bch_csum csum;
2174 /* this flags field is encrypted, unlike bset->flags: */
2177 /* Closed interval: */
2178 struct bpos min_key;
2179 struct bpos max_key;
2180 struct bch_extent_ptr _ptr; /* not used anymore */
2181 struct bkey_format format;
2192 } __packed __aligned(8);
2194 LE64_BITMASK(BTREE_NODE_ID, struct btree_node, flags, 0, 4);
2195 LE64_BITMASK(BTREE_NODE_LEVEL, struct btree_node, flags, 4, 8);
2196 LE64_BITMASK(BTREE_NODE_NEW_EXTENT_OVERWRITE,
2197 struct btree_node, flags, 8, 9);
2199 LE64_BITMASK(BTREE_NODE_SEQ, struct btree_node, flags, 32, 64);
2201 struct btree_node_entry {
2202 struct bch_csum csum;
2213 } __packed __aligned(8);
2215 #endif /* _BCACHEFS_FORMAT_H */