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 unsigned name##_OFFSET = offset; \
87 static const 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 unsigned name##_OFFSET = offset; \
102 static const unsigned name##_BITS = (end - offset); \
103 static const __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.
154 } __packed __aligned(4);
156 #define KEY_INODE_MAX ((__u64)~0ULL)
157 #define KEY_OFFSET_MAX ((__u64)~0ULL)
158 #define KEY_SNAPSHOT_MAX ((__u32)~0U)
159 #define KEY_SIZE_MAX ((__u32)~0U)
161 static inline struct bpos SPOS(__u64 inode, __u64 offset, __u32 snapshot)
163 return (struct bpos) {
166 .snapshot = snapshot,
170 #define POS_MIN SPOS(0, 0, 0)
171 #define POS_MAX SPOS(KEY_INODE_MAX, KEY_OFFSET_MAX, 0)
172 #define SPOS_MAX SPOS(KEY_INODE_MAX, KEY_OFFSET_MAX, KEY_SNAPSHOT_MAX)
173 #define POS(_inode, _offset) SPOS(_inode, _offset, 0)
175 /* Empty placeholder struct, for container_of() */
181 #if __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__
184 #elif __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__
188 } __packed __aligned(4);
191 /* Size of combined key and value, in u64s */
194 /* Format of key (0 for format local to btree node) */
195 #if defined(__LITTLE_ENDIAN_BITFIELD)
198 #elif defined (__BIG_ENDIAN_BITFIELD)
199 __u8 needs_whiteout:1,
202 #error edit for your odd byteorder.
205 /* Type of the value */
208 #if __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__
211 struct bversion version;
212 __u32 size; /* extent size, in sectors */
214 #elif __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__
216 __u32 size; /* extent size, in sectors */
217 struct bversion version;
221 } __packed __aligned(8);
226 /* Size of combined key and value, in u64s */
229 /* Format of key (0 for format local to btree node) */
232 * XXX: next incompat on disk format change, switch format and
233 * needs_whiteout - bkey_packed() will be cheaper if format is the high
234 * bits of the bitfield
236 #if defined(__LITTLE_ENDIAN_BITFIELD)
239 #elif defined (__BIG_ENDIAN_BITFIELD)
240 __u8 needs_whiteout:1,
244 /* Type of the value */
249 * We copy bkeys with struct assignment in various places, and while
250 * that shouldn't be done with packed bkeys we can't disallow it in C,
251 * and it's legal to cast a bkey to a bkey_packed - so padding it out
252 * to the same size as struct bkey should hopefully be safest.
254 __u8 pad[sizeof(struct bkey) - 3];
255 } __packed __aligned(8);
262 #define BKEY_U64s (sizeof(struct bkey) / sizeof(__u64))
263 #define BKEY_U64s_MAX U8_MAX
264 #define BKEY_VAL_U64s_MAX (BKEY_U64s_MAX - BKEY_U64s)
266 #define KEY_PACKED_BITS_START 24
268 #define KEY_FORMAT_LOCAL_BTREE 0
269 #define KEY_FORMAT_CURRENT 1
271 enum bch_bkey_fields {
276 BKEY_FIELD_VERSION_HI,
277 BKEY_FIELD_VERSION_LO,
281 #define bkey_format_field(name, field) \
282 [BKEY_FIELD_##name] = (sizeof(((struct bkey *) NULL)->field) * 8)
284 #define BKEY_FORMAT_CURRENT \
285 ((struct bkey_format) { \
286 .key_u64s = BKEY_U64s, \
287 .nr_fields = BKEY_NR_FIELDS, \
288 .bits_per_field = { \
289 bkey_format_field(INODE, p.inode), \
290 bkey_format_field(OFFSET, p.offset), \
291 bkey_format_field(SNAPSHOT, p.snapshot), \
292 bkey_format_field(SIZE, size), \
293 bkey_format_field(VERSION_HI, version.hi), \
294 bkey_format_field(VERSION_LO, version.lo), \
298 /* bkey with inline value */
306 #define KEY(_inode, _offset, _size) \
309 .format = KEY_FORMAT_CURRENT, \
310 .p = POS(_inode, _offset), \
314 static inline void bkey_init(struct bkey *k)
319 #define bkey_bytes(_k) ((_k)->u64s * sizeof(__u64))
321 #define __BKEY_PADDED(key, pad) \
322 struct bkey_i key; __u64 key ## _pad[pad]
325 * - DELETED keys are used internally to mark keys that should be ignored but
326 * override keys in composition order. Their version number is ignored.
328 * - DISCARDED keys indicate that the data is all 0s because it has been
329 * discarded. DISCARDs may have a version; if the version is nonzero the key
330 * will be persistent, otherwise the key will be dropped whenever the btree
331 * node is rewritten (like DELETED keys).
333 * - ERROR: any read of the data returns a read error, as the data was lost due
334 * to a failing device. Like DISCARDED keys, they can be removed (overridden)
335 * by new writes or cluster-wide GC. Node repair can also overwrite them with
336 * the same or a more recent version number, but not with an older version
339 * - WHITEOUT: for hash table btrees
341 #define BCH_BKEY_TYPES() \
346 x(hash_whiteout, 4) \
351 x(inode_generation, 9) \
360 x(btree_ptr_v2, 18) \
361 x(indirect_inline_data, 19) \
376 #define x(name, nr) KEY_TYPE_##name = nr,
386 struct bch_whiteout {
399 struct bch_hash_whiteout {
410 * In extent bkeys, the value is a list of pointers (bch_extent_ptr), optionally
411 * preceded by checksum/compression information (bch_extent_crc32 or
414 * One major determining factor in the format of extents is how we handle and
415 * represent extents that have been partially overwritten and thus trimmed:
417 * If an extent is not checksummed or compressed, when the extent is trimmed we
418 * don't have to remember the extent we originally allocated and wrote: we can
419 * merely adjust ptr->offset to point to the start of the data that is currently
420 * live. The size field in struct bkey records the current (live) size of the
421 * extent, and is also used to mean "size of region on disk that we point to" in
424 * Thus an extent that is not checksummed or compressed will consist only of a
425 * list of bch_extent_ptrs, with none of the fields in
426 * bch_extent_crc32/bch_extent_crc64.
428 * When an extent is checksummed or compressed, it's not possible to read only
429 * the data that is currently live: we have to read the entire extent that was
430 * originally written, and then return only the part of the extent that is
433 * Thus, in addition to the current size of the extent in struct bkey, we need
434 * to store the size of the originally allocated space - this is the
435 * compressed_size and uncompressed_size fields in bch_extent_crc32/64. Also,
436 * when the extent is trimmed, instead of modifying the offset field of the
437 * pointer, we keep a second smaller offset field - "offset into the original
438 * extent of the currently live region".
440 * The other major determining factor is replication and data migration:
442 * Each pointer may have its own bch_extent_crc32/64. When doing a replicated
443 * write, we will initially write all the replicas in the same format, with the
444 * same checksum type and compression format - however, when copygc runs later (or
445 * tiering/cache promotion, anything that moves data), it is not in general
446 * going to rewrite all the pointers at once - one of the replicas may be in a
447 * bucket on one device that has very little fragmentation while another lives
448 * in a bucket that has become heavily fragmented, and thus is being rewritten
449 * sooner than the rest.
451 * Thus it will only move a subset of the pointers (or in the case of
452 * tiering/cache promotion perhaps add a single pointer without dropping any
453 * current pointers), and if the extent has been partially overwritten it must
454 * write only the currently live portion (or copygc would not be able to reduce
455 * fragmentation!) - which necessitates a different bch_extent_crc format for
458 * But in the interests of space efficiency, we don't want to store one
459 * bch_extent_crc for each pointer if we don't have to.
461 * Thus, a bch_extent consists of bch_extent_crc32s, bch_extent_crc64s, and
462 * bch_extent_ptrs appended arbitrarily one after the other. We determine the
463 * type of a given entry with a scheme similar to utf8 (except we're encoding a
464 * type, not a size), encoding the type in the position of the first set bit:
466 * bch_extent_crc32 - 0b1
467 * bch_extent_ptr - 0b10
468 * bch_extent_crc64 - 0b100
470 * We do it this way because bch_extent_crc32 is _very_ constrained on bits (and
471 * bch_extent_crc64 is the least constrained).
473 * Then, each bch_extent_crc32/64 applies to the pointers that follow after it,
474 * until the next bch_extent_crc32/64.
476 * If there are no bch_extent_crcs preceding a bch_extent_ptr, then that pointer
477 * is neither checksummed nor compressed.
480 /* 128 bits, sufficient for cryptographic MACs: */
484 } __packed __aligned(8);
486 #define BCH_EXTENT_ENTRY_TYPES() \
493 #define BCH_EXTENT_ENTRY_MAX 6
495 enum bch_extent_entry_type {
496 #define x(f, n) BCH_EXTENT_ENTRY_##f = n,
497 BCH_EXTENT_ENTRY_TYPES()
501 /* Compressed/uncompressed size are stored biased by 1: */
502 struct bch_extent_crc32 {
503 #if defined(__LITTLE_ENDIAN_BITFIELD)
506 _uncompressed_size:7,
512 #elif defined (__BIG_ENDIAN_BITFIELD)
514 __u32 compression_type:4,
518 _uncompressed_size:7,
522 } __packed __aligned(8);
524 #define CRC32_SIZE_MAX (1U << 7)
525 #define CRC32_NONCE_MAX 0
527 struct bch_extent_crc64 {
528 #if defined(__LITTLE_ENDIAN_BITFIELD)
531 _uncompressed_size:9,
537 #elif defined (__BIG_ENDIAN_BITFIELD)
543 _uncompressed_size:9,
548 } __packed __aligned(8);
550 #define CRC64_SIZE_MAX (1U << 9)
551 #define CRC64_NONCE_MAX ((1U << 10) - 1)
553 struct bch_extent_crc128 {
554 #if defined(__LITTLE_ENDIAN_BITFIELD)
557 _uncompressed_size:13,
562 #elif defined (__BIG_ENDIAN_BITFIELD)
563 __u64 compression_type:4,
567 _uncompressed_size:13,
571 struct bch_csum csum;
572 } __packed __aligned(8);
574 #define CRC128_SIZE_MAX (1U << 13)
575 #define CRC128_NONCE_MAX ((1U << 13) - 1)
578 * @reservation - pointer hasn't been written to, just reserved
580 struct bch_extent_ptr {
581 #if defined(__LITTLE_ENDIAN_BITFIELD)
586 offset:44, /* 8 petabytes */
589 #elif defined (__BIG_ENDIAN_BITFIELD)
598 } __packed __aligned(8);
600 struct bch_extent_stripe_ptr {
601 #if defined(__LITTLE_ENDIAN_BITFIELD)
606 #elif defined (__BIG_ENDIAN_BITFIELD)
614 struct bch_extent_reservation {
615 #if defined(__LITTLE_ENDIAN_BITFIELD)
620 #elif defined (__BIG_ENDIAN_BITFIELD)
628 struct bch_extent_rebalance {
629 #if defined(__LITTLE_ENDIAN_BITFIELD)
634 #elif defined (__BIG_ENDIAN_BITFIELD)
642 union bch_extent_entry {
643 #if __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__ || __BITS_PER_LONG == 64
645 #elif __BITS_PER_LONG == 32
651 #error edit for your odd byteorder.
654 #define x(f, n) struct bch_extent_##f f;
655 BCH_EXTENT_ENTRY_TYPES()
659 struct bch_btree_ptr {
663 struct bch_extent_ptr start[];
664 } __packed __aligned(8);
666 struct bch_btree_ptr_v2 {
671 __le16 sectors_written;
675 struct bch_extent_ptr start[];
676 } __packed __aligned(8);
678 LE16_BITMASK(BTREE_PTR_RANGE_UPDATED, struct bch_btree_ptr_v2, flags, 0, 1);
684 union bch_extent_entry start[];
685 } __packed __aligned(8);
687 struct bch_reservation {
693 } __packed __aligned(8);
695 /* Maximum size (in u64s) a single pointer could be: */
696 #define BKEY_EXTENT_PTR_U64s_MAX\
697 ((sizeof(struct bch_extent_crc128) + \
698 sizeof(struct bch_extent_ptr)) / sizeof(__u64))
700 /* Maximum possible size of an entire extent value: */
701 #define BKEY_EXTENT_VAL_U64s_MAX \
702 (1 + BKEY_EXTENT_PTR_U64s_MAX * (BCH_REPLICAS_MAX + 1))
704 /* * Maximum possible size of an entire extent, key + value: */
705 #define BKEY_EXTENT_U64s_MAX (BKEY_U64s + BKEY_EXTENT_VAL_U64s_MAX)
707 /* Btree pointers don't carry around checksums: */
708 #define BKEY_BTREE_PTR_VAL_U64s_MAX \
709 ((sizeof(struct bch_btree_ptr_v2) + \
710 sizeof(struct bch_extent_ptr) * BCH_REPLICAS_MAX) / sizeof(__u64))
711 #define BKEY_BTREE_PTR_U64s_MAX \
712 (BKEY_U64s + BKEY_BTREE_PTR_VAL_U64s_MAX)
716 #define BLOCKDEV_INODE_MAX 4096
718 #define BCACHEFS_ROOT_INO 4096
727 } __packed __aligned(8);
729 struct bch_inode_v2 {
732 __le64 bi_journal_seq;
737 } __packed __aligned(8);
739 struct bch_inode_v3 {
742 __le64 bi_journal_seq;
749 } __packed __aligned(8);
751 #define INODEv3_FIELDS_START_INITIAL 6
752 #define INODEv3_FIELDS_START_CUR (offsetof(struct bch_inode_v3, fields) / sizeof(__u64))
754 struct bch_inode_generation {
757 __le32 bi_generation;
759 } __packed __aligned(8);
762 * bi_subvol and bi_parent_subvol are only set for subvolume roots:
765 #define BCH_INODE_FIELDS_v2() \
775 x(bi_generation, 32) \
777 x(bi_data_checksum, 8) \
778 x(bi_compression, 8) \
780 x(bi_background_compression, 8) \
781 x(bi_data_replicas, 8) \
782 x(bi_promote_target, 16) \
783 x(bi_foreground_target, 16) \
784 x(bi_background_target, 16) \
785 x(bi_erasure_code, 16) \
786 x(bi_fields_set, 16) \
788 x(bi_dir_offset, 64) \
790 x(bi_parent_subvol, 32)
792 #define BCH_INODE_FIELDS_v3() \
800 x(bi_generation, 32) \
802 x(bi_data_checksum, 8) \
803 x(bi_compression, 8) \
805 x(bi_background_compression, 8) \
806 x(bi_data_replicas, 8) \
807 x(bi_promote_target, 16) \
808 x(bi_foreground_target, 16) \
809 x(bi_background_target, 16) \
810 x(bi_erasure_code, 16) \
811 x(bi_fields_set, 16) \
813 x(bi_dir_offset, 64) \
815 x(bi_parent_subvol, 32) \
818 /* subset of BCH_INODE_FIELDS */
819 #define BCH_INODE_OPTS() \
820 x(data_checksum, 8) \
823 x(background_compression, 8) \
824 x(data_replicas, 8) \
825 x(promote_target, 16) \
826 x(foreground_target, 16) \
827 x(background_target, 16) \
828 x(erasure_code, 16) \
832 #define x(name, ...) \
841 * User flags (get/settable with FS_IOC_*FLAGS, correspond to FS_*_FL
844 __BCH_INODE_SYNC = 0,
845 __BCH_INODE_IMMUTABLE = 1,
846 __BCH_INODE_APPEND = 2,
847 __BCH_INODE_NODUMP = 3,
848 __BCH_INODE_NOATIME = 4,
850 __BCH_INODE_I_SIZE_DIRTY = 5,
851 __BCH_INODE_I_SECTORS_DIRTY = 6,
852 __BCH_INODE_UNLINKED = 7,
853 __BCH_INODE_BACKPTR_UNTRUSTED = 8,
855 /* bits 20+ reserved for packed fields below: */
858 #define BCH_INODE_SYNC (1 << __BCH_INODE_SYNC)
859 #define BCH_INODE_IMMUTABLE (1 << __BCH_INODE_IMMUTABLE)
860 #define BCH_INODE_APPEND (1 << __BCH_INODE_APPEND)
861 #define BCH_INODE_NODUMP (1 << __BCH_INODE_NODUMP)
862 #define BCH_INODE_NOATIME (1 << __BCH_INODE_NOATIME)
863 #define BCH_INODE_I_SIZE_DIRTY (1 << __BCH_INODE_I_SIZE_DIRTY)
864 #define BCH_INODE_I_SECTORS_DIRTY (1 << __BCH_INODE_I_SECTORS_DIRTY)
865 #define BCH_INODE_UNLINKED (1 << __BCH_INODE_UNLINKED)
866 #define BCH_INODE_BACKPTR_UNTRUSTED (1 << __BCH_INODE_BACKPTR_UNTRUSTED)
868 LE32_BITMASK(INODE_STR_HASH, struct bch_inode, bi_flags, 20, 24);
869 LE32_BITMASK(INODE_NR_FIELDS, struct bch_inode, bi_flags, 24, 31);
870 LE32_BITMASK(INODE_NEW_VARINT, struct bch_inode, bi_flags, 31, 32);
872 LE64_BITMASK(INODEv2_STR_HASH, struct bch_inode_v2, bi_flags, 20, 24);
873 LE64_BITMASK(INODEv2_NR_FIELDS, struct bch_inode_v2, bi_flags, 24, 31);
875 LE64_BITMASK(INODEv3_STR_HASH, struct bch_inode_v3, bi_flags, 20, 24);
876 LE64_BITMASK(INODEv3_NR_FIELDS, struct bch_inode_v3, bi_flags, 24, 31);
878 LE64_BITMASK(INODEv3_FIELDS_START,
879 struct bch_inode_v3, bi_flags, 31, 36);
880 LE64_BITMASK(INODEv3_MODE, struct bch_inode_v3, bi_flags, 36, 52);
885 * Dirents (and xattrs) have to implement string lookups; since our b-tree
886 * doesn't support arbitrary length strings for the key, we instead index by a
887 * 64 bit hash (currently truncated sha1) of the string, stored in the offset
888 * field of the key - using linear probing to resolve hash collisions. This also
889 * provides us with the readdir cookie posix requires.
891 * Linear probing requires us to use whiteouts for deletions, in the event of a
898 /* Target inode number: */
901 struct { /* DT_SUBVOL */
902 __le32 d_child_subvol;
903 __le32 d_parent_subvol;
908 * Copy of mode bits 12-15 from the target inode - so userspace can get
909 * the filetype without having to do a stat()
914 } __packed __aligned(8);
917 #define BCH_DT_MAX 17
919 #define BCH_NAME_MAX 512
923 #define KEY_TYPE_XATTR_INDEX_USER 0
924 #define KEY_TYPE_XATTR_INDEX_POSIX_ACL_ACCESS 1
925 #define KEY_TYPE_XATTR_INDEX_POSIX_ACL_DEFAULT 2
926 #define KEY_TYPE_XATTR_INDEX_TRUSTED 3
927 #define KEY_TYPE_XATTR_INDEX_SECURITY 4
935 } __packed __aligned(8);
937 /* Bucket/allocation information: */
944 } __packed __aligned(8);
946 #define BCH_ALLOC_FIELDS_V1() \
950 x(dirty_sectors, 16) \
951 x(cached_sectors, 16) \
954 x(stripe_redundancy, 8)
957 #define x(name, _bits) BCH_ALLOC_FIELD_V1_##name,
958 BCH_ALLOC_FIELDS_V1()
962 struct bch_alloc_v2 {
969 } __packed __aligned(8);
971 #define BCH_ALLOC_FIELDS_V2() \
974 x(dirty_sectors, 32) \
975 x(cached_sectors, 32) \
977 x(stripe_redundancy, 8)
979 struct bch_alloc_v3 {
988 } __packed __aligned(8);
990 LE32_BITMASK(BCH_ALLOC_V3_NEED_DISCARD,struct bch_alloc_v3, flags, 0, 1)
991 LE32_BITMASK(BCH_ALLOC_V3_NEED_INC_GEN,struct bch_alloc_v3, flags, 1, 2)
993 struct bch_alloc_v4 {
1000 __u8 stripe_redundancy;
1001 __u32 dirty_sectors;
1002 __u32 cached_sectors;
1005 __u32 nr_external_backpointers;
1006 __u64 fragmentation_lru;
1007 } __packed __aligned(8);
1009 #define BCH_ALLOC_V4_U64s_V0 6
1010 #define BCH_ALLOC_V4_U64s (sizeof(struct bch_alloc_v4) / sizeof(__u64))
1012 BITMASK(BCH_ALLOC_V4_NEED_DISCARD, struct bch_alloc_v4, flags, 0, 1)
1013 BITMASK(BCH_ALLOC_V4_NEED_INC_GEN, struct bch_alloc_v4, flags, 1, 2)
1014 BITMASK(BCH_ALLOC_V4_BACKPOINTERS_START,struct bch_alloc_v4, flags, 2, 8)
1015 BITMASK(BCH_ALLOC_V4_NR_BACKPOINTERS, struct bch_alloc_v4, flags, 8, 14)
1017 #define BCH_ALLOC_V4_NR_BACKPOINTERS_MAX 40
1019 struct bch_backpointer {
1024 __u64 bucket_offset:40;
1027 } __packed __aligned(8);
1029 #define KEY_TYPE_BUCKET_GENS_BITS 8
1030 #define KEY_TYPE_BUCKET_GENS_NR (1U << KEY_TYPE_BUCKET_GENS_BITS)
1031 #define KEY_TYPE_BUCKET_GENS_MASK (KEY_TYPE_BUCKET_GENS_NR - 1)
1033 struct bch_bucket_gens {
1035 u8 gens[KEY_TYPE_BUCKET_GENS_NR];
1036 } __packed __aligned(8);
1047 enum quota_counters {
1053 struct bch_quota_counter {
1060 struct bch_quota_counter c[Q_COUNTERS];
1061 } __packed __aligned(8);
1063 /* Erasure coding */
1072 __u8 csum_granularity_bits;
1076 struct bch_extent_ptr ptrs[];
1077 } __packed __aligned(8);
1081 struct bch_reflink_p {
1085 * A reflink pointer might point to an indirect extent which is then
1086 * later split (by copygc or rebalance). If we only pointed to part of
1087 * the original indirect extent, and then one of the fragments is
1088 * outside the range we point to, we'd leak a refcount: so when creating
1089 * reflink pointers, we need to store pad values to remember the full
1090 * range we were taking a reference on.
1094 } __packed __aligned(8);
1096 struct bch_reflink_v {
1099 union bch_extent_entry start[0];
1101 } __packed __aligned(8);
1103 struct bch_indirect_inline_data {
1111 struct bch_inline_data {
1118 #define SUBVOL_POS_MIN POS(0, 1)
1119 #define SUBVOL_POS_MAX POS(0, S32_MAX)
1120 #define BCACHEFS_ROOT_SUBVOL 1
1122 struct bch_subvolume {
1128 * Snapshot subvolumes form a tree, separate from the snapshot nodes
1129 * tree - if this subvolume is a snapshot, this is the ID of the
1130 * subvolume it was created from:
1137 LE32_BITMASK(BCH_SUBVOLUME_RO, struct bch_subvolume, flags, 0, 1)
1139 * We need to know whether a subvolume is a snapshot so we can know whether we
1140 * can delete it (or whether it should just be rm -rf'd)
1142 LE32_BITMASK(BCH_SUBVOLUME_SNAP, struct bch_subvolume, flags, 1, 2)
1143 LE32_BITMASK(BCH_SUBVOLUME_UNLINKED, struct bch_subvolume, flags, 2, 3)
1147 struct bch_snapshot {
1158 LE32_BITMASK(BCH_SNAPSHOT_DELETED, struct bch_snapshot, flags, 0, 1)
1160 /* True if a subvolume points to this snapshot node: */
1161 LE32_BITMASK(BCH_SNAPSHOT_SUBVOL, struct bch_snapshot, flags, 1, 2)
1166 * The snapshot_trees btree gives us persistent indentifier for each tree of
1167 * bch_snapshot nodes, and allow us to record and easily find the root/master
1168 * subvolume that other snapshots were created from:
1170 struct bch_snapshot_tree {
1172 __le32 master_subvol;
1173 __le32 root_snapshot;
1181 } __packed __aligned(8);
1183 #define LRU_ID_STRIPES (1U << 16)
1185 /* Optional/variable size superblock sections: */
1187 struct bch_sb_field {
1193 #define BCH_SB_FIELDS() \
1202 x(journal_seq_blacklist, 8) \
1206 enum bch_sb_field_type {
1207 #define x(f, nr) BCH_SB_FIELD_##f = nr,
1214 * Most superblock fields are replicated in all device's superblocks - a few are
1217 #define BCH_SINGLE_DEVICE_SB_FIELDS \
1218 ((1U << BCH_SB_FIELD_journal)| \
1219 (1U << BCH_SB_FIELD_journal_v2))
1221 /* BCH_SB_FIELD_journal: */
1223 struct bch_sb_field_journal {
1224 struct bch_sb_field field;
1228 struct bch_sb_field_journal_v2 {
1229 struct bch_sb_field field;
1231 struct bch_sb_field_journal_v2_entry {
1237 /* BCH_SB_FIELD_members: */
1239 #define BCH_MIN_NR_NBUCKETS (1 << 6)
1243 __le64 nbuckets; /* device size */
1244 __le16 first_bucket; /* index of first bucket used */
1245 __le16 bucket_size; /* sectors */
1247 __le64 last_mount; /* time_t */
1252 LE64_BITMASK(BCH_MEMBER_STATE, struct bch_member, flags[0], 0, 4)
1253 /* 4-14 unused, was TIER, HAS_(META)DATA, REPLACEMENT */
1254 LE64_BITMASK(BCH_MEMBER_DISCARD, struct bch_member, flags[0], 14, 15)
1255 LE64_BITMASK(BCH_MEMBER_DATA_ALLOWED, struct bch_member, flags[0], 15, 20)
1256 LE64_BITMASK(BCH_MEMBER_GROUP, struct bch_member, flags[0], 20, 28)
1257 LE64_BITMASK(BCH_MEMBER_DURABILITY, struct bch_member, flags[0], 28, 30)
1258 LE64_BITMASK(BCH_MEMBER_FREESPACE_INITIALIZED,
1259 struct bch_member, flags[0], 30, 31)
1262 LE64_BITMASK(BCH_MEMBER_NR_READ_ERRORS, struct bch_member, flags[1], 0, 20);
1263 LE64_BITMASK(BCH_MEMBER_NR_WRITE_ERRORS,struct bch_member, flags[1], 20, 40);
1266 #define BCH_MEMBER_STATES() \
1272 enum bch_member_state {
1273 #define x(t, n) BCH_MEMBER_STATE_##t = n,
1279 struct bch_sb_field_members {
1280 struct bch_sb_field field;
1281 struct bch_member members[0];
1284 /* BCH_SB_FIELD_crypt: */
1294 #define BCH_KEY_MAGIC \
1295 (((__u64) 'b' << 0)|((__u64) 'c' << 8)| \
1296 ((__u64) 'h' << 16)|((__u64) '*' << 24)| \
1297 ((__u64) '*' << 32)|((__u64) 'k' << 40)| \
1298 ((__u64) 'e' << 48)|((__u64) 'y' << 56))
1300 struct bch_encrypted_key {
1306 * If this field is present in the superblock, it stores an encryption key which
1307 * is used encrypt all other data/metadata. The key will normally be encrypted
1308 * with the key userspace provides, but if encryption has been turned off we'll
1309 * just store the master key unencrypted in the superblock so we can access the
1310 * previously encrypted data.
1312 struct bch_sb_field_crypt {
1313 struct bch_sb_field field;
1317 struct bch_encrypted_key key;
1320 LE64_BITMASK(BCH_CRYPT_KDF_TYPE, struct bch_sb_field_crypt, flags, 0, 4);
1322 enum bch_kdf_types {
1327 /* stored as base 2 log of scrypt params: */
1328 LE64_BITMASK(BCH_KDF_SCRYPT_N, struct bch_sb_field_crypt, kdf_flags, 0, 16);
1329 LE64_BITMASK(BCH_KDF_SCRYPT_R, struct bch_sb_field_crypt, kdf_flags, 16, 32);
1330 LE64_BITMASK(BCH_KDF_SCRYPT_P, struct bch_sb_field_crypt, kdf_flags, 32, 48);
1332 /* BCH_SB_FIELD_replicas: */
1334 #define BCH_DATA_TYPES() \
1343 x(need_gc_gens, 8) \
1346 enum bch_data_type {
1347 #define x(t, n) BCH_DATA_##t,
1353 static inline bool data_type_is_empty(enum bch_data_type type)
1357 case BCH_DATA_need_gc_gens:
1358 case BCH_DATA_need_discard:
1365 static inline bool data_type_is_hidden(enum bch_data_type type)
1369 case BCH_DATA_journal:
1376 struct bch_replicas_entry_v0 {
1382 struct bch_sb_field_replicas_v0 {
1383 struct bch_sb_field field;
1384 struct bch_replicas_entry_v0 entries[0];
1385 } __packed __aligned(8);
1387 struct bch_replicas_entry {
1394 #define replicas_entry_bytes(_i) \
1395 (offsetof(typeof(*(_i)), devs) + (_i)->nr_devs)
1397 struct bch_sb_field_replicas {
1398 struct bch_sb_field field;
1399 struct bch_replicas_entry entries[0];
1400 } __packed __aligned(8);
1402 /* BCH_SB_FIELD_quota: */
1404 struct bch_sb_quota_counter {
1409 struct bch_sb_quota_type {
1411 struct bch_sb_quota_counter c[Q_COUNTERS];
1414 struct bch_sb_field_quota {
1415 struct bch_sb_field field;
1416 struct bch_sb_quota_type q[QTYP_NR];
1417 } __packed __aligned(8);
1419 /* BCH_SB_FIELD_disk_groups: */
1421 #define BCH_SB_LABEL_SIZE 32
1423 struct bch_disk_group {
1424 __u8 label[BCH_SB_LABEL_SIZE];
1426 } __packed __aligned(8);
1428 LE64_BITMASK(BCH_GROUP_DELETED, struct bch_disk_group, flags[0], 0, 1)
1429 LE64_BITMASK(BCH_GROUP_DATA_ALLOWED, struct bch_disk_group, flags[0], 1, 6)
1430 LE64_BITMASK(BCH_GROUP_PARENT, struct bch_disk_group, flags[0], 6, 24)
1432 struct bch_sb_field_disk_groups {
1433 struct bch_sb_field field;
1434 struct bch_disk_group entries[0];
1435 } __packed __aligned(8);
1437 /* BCH_SB_FIELD_counters */
1439 #define BCH_PERSISTENT_COUNTERS() \
1443 x(bucket_invalidate, 3) \
1444 x(bucket_discard, 4) \
1445 x(bucket_alloc, 5) \
1446 x(bucket_alloc_fail, 6) \
1447 x(btree_cache_scan, 7) \
1448 x(btree_cache_reap, 8) \
1449 x(btree_cache_cannibalize, 9) \
1450 x(btree_cache_cannibalize_lock, 10) \
1451 x(btree_cache_cannibalize_lock_fail, 11) \
1452 x(btree_cache_cannibalize_unlock, 12) \
1453 x(btree_node_write, 13) \
1454 x(btree_node_read, 14) \
1455 x(btree_node_compact, 15) \
1456 x(btree_node_merge, 16) \
1457 x(btree_node_split, 17) \
1458 x(btree_node_rewrite, 18) \
1459 x(btree_node_alloc, 19) \
1460 x(btree_node_free, 20) \
1461 x(btree_node_set_root, 21) \
1462 x(btree_path_relock_fail, 22) \
1463 x(btree_path_upgrade_fail, 23) \
1464 x(btree_reserve_get_fail, 24) \
1465 x(journal_entry_full, 25) \
1466 x(journal_full, 26) \
1467 x(journal_reclaim_finish, 27) \
1468 x(journal_reclaim_start, 28) \
1469 x(journal_write, 29) \
1470 x(read_promote, 30) \
1471 x(read_bounce, 31) \
1474 x(read_reuse_race, 34) \
1475 x(move_extent_read, 35) \
1476 x(move_extent_write, 36) \
1477 x(move_extent_finish, 37) \
1478 x(move_extent_fail, 38) \
1479 x(move_extent_alloc_mem_fail, 39) \
1481 x(copygc_wait, 41) \
1482 x(gc_gens_end, 42) \
1483 x(gc_gens_start, 43) \
1484 x(trans_blocked_journal_reclaim, 44) \
1485 x(trans_restart_btree_node_reused, 45) \
1486 x(trans_restart_btree_node_split, 46) \
1487 x(trans_restart_fault_inject, 47) \
1488 x(trans_restart_iter_upgrade, 48) \
1489 x(trans_restart_journal_preres_get, 49) \
1490 x(trans_restart_journal_reclaim, 50) \
1491 x(trans_restart_journal_res_get, 51) \
1492 x(trans_restart_key_cache_key_realloced, 52) \
1493 x(trans_restart_key_cache_raced, 53) \
1494 x(trans_restart_mark_replicas, 54) \
1495 x(trans_restart_mem_realloced, 55) \
1496 x(trans_restart_memory_allocation_failure, 56) \
1497 x(trans_restart_relock, 57) \
1498 x(trans_restart_relock_after_fill, 58) \
1499 x(trans_restart_relock_key_cache_fill, 59) \
1500 x(trans_restart_relock_next_node, 60) \
1501 x(trans_restart_relock_parent_for_fill, 61) \
1502 x(trans_restart_relock_path, 62) \
1503 x(trans_restart_relock_path_intent, 63) \
1504 x(trans_restart_too_many_iters, 64) \
1505 x(trans_restart_traverse, 65) \
1506 x(trans_restart_upgrade, 66) \
1507 x(trans_restart_would_deadlock, 67) \
1508 x(trans_restart_would_deadlock_write, 68) \
1509 x(trans_restart_injected, 69) \
1510 x(trans_restart_key_cache_upgrade, 70) \
1511 x(trans_traverse_all, 71) \
1512 x(transaction_commit, 72) \
1513 x(write_super, 73) \
1514 x(trans_restart_would_deadlock_recursion_limit, 74) \
1515 x(trans_restart_write_buffer_flush, 75) \
1516 x(trans_restart_split_race, 76)
1518 enum bch_persistent_counters {
1519 #define x(t, n, ...) BCH_COUNTER_##t,
1520 BCH_PERSISTENT_COUNTERS()
1525 struct bch_sb_field_counters {
1526 struct bch_sb_field field;
1531 * On clean shutdown, store btree roots and current journal sequence number in
1538 __u8 type; /* designates what this jset holds */
1542 struct bkey_i start[0];
1547 struct bch_sb_field_clean {
1548 struct bch_sb_field field;
1551 __le16 _read_clock; /* no longer used */
1552 __le16 _write_clock;
1556 struct jset_entry start[0];
1561 struct journal_seq_blacklist_entry {
1566 struct bch_sb_field_journal_seq_blacklist {
1567 struct bch_sb_field field;
1570 struct journal_seq_blacklist_entry start[0];
1578 * New versioning scheme:
1579 * One common version number for all on disk data structures - superblock, btree
1580 * nodes, journal entries
1582 #define BCH_VERSION_MAJOR(_v) ((__u16) ((_v) >> 10))
1583 #define BCH_VERSION_MINOR(_v) ((__u16) ((_v) & ~(~0U << 10)))
1584 #define BCH_VERSION(_major, _minor) (((_major) << 10)|(_minor) << 0)
1586 #define RECOVERY_PASS_ALL_FSCK (1ULL << 63)
1588 #define BCH_METADATA_VERSIONS() \
1589 x(bkey_renumber, BCH_VERSION(0, 10), \
1590 RECOVERY_PASS_ALL_FSCK) \
1591 x(inode_btree_change, BCH_VERSION(0, 11), \
1592 RECOVERY_PASS_ALL_FSCK) \
1593 x(snapshot, BCH_VERSION(0, 12), \
1594 RECOVERY_PASS_ALL_FSCK) \
1595 x(inode_backpointers, BCH_VERSION(0, 13), \
1596 RECOVERY_PASS_ALL_FSCK) \
1597 x(btree_ptr_sectors_written, BCH_VERSION(0, 14), \
1598 RECOVERY_PASS_ALL_FSCK) \
1599 x(snapshot_2, BCH_VERSION(0, 15), \
1600 BIT_ULL(BCH_RECOVERY_PASS_fs_upgrade_for_subvolumes)| \
1601 BIT_ULL(BCH_RECOVERY_PASS_initialize_subvolumes)| \
1602 RECOVERY_PASS_ALL_FSCK) \
1603 x(reflink_p_fix, BCH_VERSION(0, 16), \
1604 BIT_ULL(BCH_RECOVERY_PASS_fix_reflink_p)) \
1605 x(subvol_dirent, BCH_VERSION(0, 17), \
1606 RECOVERY_PASS_ALL_FSCK) \
1607 x(inode_v2, BCH_VERSION(0, 18), \
1608 RECOVERY_PASS_ALL_FSCK) \
1609 x(freespace, BCH_VERSION(0, 19), \
1610 RECOVERY_PASS_ALL_FSCK) \
1611 x(alloc_v4, BCH_VERSION(0, 20), \
1612 RECOVERY_PASS_ALL_FSCK) \
1613 x(new_data_types, BCH_VERSION(0, 21), \
1614 RECOVERY_PASS_ALL_FSCK) \
1615 x(backpointers, BCH_VERSION(0, 22), \
1616 RECOVERY_PASS_ALL_FSCK) \
1617 x(inode_v3, BCH_VERSION(0, 23), \
1618 RECOVERY_PASS_ALL_FSCK) \
1619 x(unwritten_extents, BCH_VERSION(0, 24), \
1620 RECOVERY_PASS_ALL_FSCK) \
1621 x(bucket_gens, BCH_VERSION(0, 25), \
1622 BIT_ULL(BCH_RECOVERY_PASS_bucket_gens_init)| \
1623 RECOVERY_PASS_ALL_FSCK) \
1624 x(lru_v2, BCH_VERSION(0, 26), \
1625 RECOVERY_PASS_ALL_FSCK) \
1626 x(fragmentation_lru, BCH_VERSION(0, 27), \
1627 RECOVERY_PASS_ALL_FSCK) \
1628 x(no_bps_in_alloc_keys, BCH_VERSION(0, 28), \
1629 RECOVERY_PASS_ALL_FSCK) \
1630 x(snapshot_trees, BCH_VERSION(0, 29), \
1631 RECOVERY_PASS_ALL_FSCK) \
1632 x(major_minor, BCH_VERSION(1, 0), \
1634 x(snapshot_skiplists, BCH_VERSION(1, 1), \
1635 BIT_ULL(BCH_RECOVERY_PASS_check_snapshots)) \
1636 x(deleted_inodes, BCH_VERSION(1, 2), \
1637 BIT_ULL(BCH_RECOVERY_PASS_check_inodes))
1639 enum bcachefs_metadata_version {
1640 bcachefs_metadata_version_min = 9,
1641 #define x(t, n, upgrade_passes) bcachefs_metadata_version_##t = n,
1642 BCH_METADATA_VERSIONS()
1644 bcachefs_metadata_version_max
1647 static const unsigned bcachefs_metadata_required_upgrade_below = bcachefs_metadata_version_major_minor;
1649 #define bcachefs_metadata_version_current (bcachefs_metadata_version_max - 1)
1651 #define BCH_SB_SECTOR 8
1652 #define BCH_SB_MEMBERS_MAX 64 /* XXX kill */
1654 struct bch_sb_layout {
1655 __uuid_t magic; /* bcachefs superblock UUID */
1657 __u8 sb_max_size_bits; /* base 2 of 512 byte sectors */
1658 __u8 nr_superblocks;
1660 __le64 sb_offset[61];
1661 } __packed __aligned(8);
1663 #define BCH_SB_LAYOUT_SECTOR 7
1666 * @offset - sector where this sb was written
1667 * @version - on disk format version
1668 * @version_min - Oldest metadata version this filesystem contains; so we can
1669 * safely drop compatibility code and refuse to mount filesystems
1671 * @magic - identifies as a bcachefs superblock (BCHFS_MAGIC)
1672 * @seq - incremented each time superblock is written
1673 * @uuid - used for generating various magic numbers and identifying
1674 * member devices, never changes
1675 * @user_uuid - user visible UUID, may be changed
1676 * @label - filesystem label
1677 * @seq - identifies most recent superblock, incremented each time
1678 * superblock is written
1679 * @features - enabled incompatible features
1682 struct bch_csum csum;
1689 __u8 label[BCH_SB_LABEL_SIZE];
1698 __le64 time_base_lo;
1699 __le32 time_base_hi;
1700 __le32 time_precision;
1706 struct bch_sb_layout layout;
1709 struct bch_sb_field start[0];
1712 } __packed __aligned(8);
1716 * BCH_SB_INITALIZED - set on first mount
1717 * BCH_SB_CLEAN - did we shut down cleanly? Just a hint, doesn't affect
1718 * behaviour of mount/recovery path:
1719 * BCH_SB_INODE_32BIT - limit inode numbers to 32 bits
1720 * BCH_SB_128_BIT_MACS - 128 bit macs instead of 80
1721 * BCH_SB_ENCRYPTION_TYPE - if nonzero encryption is enabled; overrides
1722 * DATA/META_CSUM_TYPE. Also indicates encryption
1723 * algorithm in use, if/when we get more than one
1726 LE16_BITMASK(BCH_SB_BLOCK_SIZE, struct bch_sb, block_size, 0, 16);
1728 LE64_BITMASK(BCH_SB_INITIALIZED, struct bch_sb, flags[0], 0, 1);
1729 LE64_BITMASK(BCH_SB_CLEAN, struct bch_sb, flags[0], 1, 2);
1730 LE64_BITMASK(BCH_SB_CSUM_TYPE, struct bch_sb, flags[0], 2, 8);
1731 LE64_BITMASK(BCH_SB_ERROR_ACTION, struct bch_sb, flags[0], 8, 12);
1733 LE64_BITMASK(BCH_SB_BTREE_NODE_SIZE, struct bch_sb, flags[0], 12, 28);
1735 LE64_BITMASK(BCH_SB_GC_RESERVE, struct bch_sb, flags[0], 28, 33);
1736 LE64_BITMASK(BCH_SB_ROOT_RESERVE, struct bch_sb, flags[0], 33, 40);
1738 LE64_BITMASK(BCH_SB_META_CSUM_TYPE, struct bch_sb, flags[0], 40, 44);
1739 LE64_BITMASK(BCH_SB_DATA_CSUM_TYPE, struct bch_sb, flags[0], 44, 48);
1741 LE64_BITMASK(BCH_SB_META_REPLICAS_WANT, struct bch_sb, flags[0], 48, 52);
1742 LE64_BITMASK(BCH_SB_DATA_REPLICAS_WANT, struct bch_sb, flags[0], 52, 56);
1744 LE64_BITMASK(BCH_SB_POSIX_ACL, struct bch_sb, flags[0], 56, 57);
1745 LE64_BITMASK(BCH_SB_USRQUOTA, struct bch_sb, flags[0], 57, 58);
1746 LE64_BITMASK(BCH_SB_GRPQUOTA, struct bch_sb, flags[0], 58, 59);
1747 LE64_BITMASK(BCH_SB_PRJQUOTA, struct bch_sb, flags[0], 59, 60);
1749 LE64_BITMASK(BCH_SB_HAS_ERRORS, struct bch_sb, flags[0], 60, 61);
1750 LE64_BITMASK(BCH_SB_HAS_TOPOLOGY_ERRORS,struct bch_sb, flags[0], 61, 62);
1752 LE64_BITMASK(BCH_SB_BIG_ENDIAN, struct bch_sb, flags[0], 62, 63);
1754 LE64_BITMASK(BCH_SB_STR_HASH_TYPE, struct bch_sb, flags[1], 0, 4);
1755 LE64_BITMASK(BCH_SB_COMPRESSION_TYPE_LO,struct bch_sb, flags[1], 4, 8);
1756 LE64_BITMASK(BCH_SB_INODE_32BIT, struct bch_sb, flags[1], 8, 9);
1758 LE64_BITMASK(BCH_SB_128_BIT_MACS, struct bch_sb, flags[1], 9, 10);
1759 LE64_BITMASK(BCH_SB_ENCRYPTION_TYPE, struct bch_sb, flags[1], 10, 14);
1762 * Max size of an extent that may require bouncing to read or write
1763 * (checksummed, compressed): 64k
1765 LE64_BITMASK(BCH_SB_ENCODED_EXTENT_MAX_BITS,
1766 struct bch_sb, flags[1], 14, 20);
1768 LE64_BITMASK(BCH_SB_META_REPLICAS_REQ, struct bch_sb, flags[1], 20, 24);
1769 LE64_BITMASK(BCH_SB_DATA_REPLICAS_REQ, struct bch_sb, flags[1], 24, 28);
1771 LE64_BITMASK(BCH_SB_PROMOTE_TARGET, struct bch_sb, flags[1], 28, 40);
1772 LE64_BITMASK(BCH_SB_FOREGROUND_TARGET, struct bch_sb, flags[1], 40, 52);
1773 LE64_BITMASK(BCH_SB_BACKGROUND_TARGET, struct bch_sb, flags[1], 52, 64);
1775 LE64_BITMASK(BCH_SB_BACKGROUND_COMPRESSION_TYPE_LO,
1776 struct bch_sb, flags[2], 0, 4);
1777 LE64_BITMASK(BCH_SB_GC_RESERVE_BYTES, struct bch_sb, flags[2], 4, 64);
1779 LE64_BITMASK(BCH_SB_ERASURE_CODE, struct bch_sb, flags[3], 0, 16);
1780 LE64_BITMASK(BCH_SB_METADATA_TARGET, struct bch_sb, flags[3], 16, 28);
1781 LE64_BITMASK(BCH_SB_SHARD_INUMS, struct bch_sb, flags[3], 28, 29);
1782 LE64_BITMASK(BCH_SB_INODES_USE_KEY_CACHE,struct bch_sb, flags[3], 29, 30);
1783 LE64_BITMASK(BCH_SB_JOURNAL_FLUSH_DELAY,struct bch_sb, flags[3], 30, 62);
1784 LE64_BITMASK(BCH_SB_JOURNAL_FLUSH_DISABLED,struct bch_sb, flags[3], 62, 63);
1785 LE64_BITMASK(BCH_SB_JOURNAL_RECLAIM_DELAY,struct bch_sb, flags[4], 0, 32);
1786 LE64_BITMASK(BCH_SB_JOURNAL_TRANSACTION_NAMES,struct bch_sb, flags[4], 32, 33);
1787 LE64_BITMASK(BCH_SB_NOCOW, struct bch_sb, flags[4], 33, 34);
1788 LE64_BITMASK(BCH_SB_WRITE_BUFFER_SIZE, struct bch_sb, flags[4], 34, 54);
1789 LE64_BITMASK(BCH_SB_VERSION_UPGRADE, struct bch_sb, flags[4], 54, 56);
1791 LE64_BITMASK(BCH_SB_COMPRESSION_TYPE_HI,struct bch_sb, flags[4], 56, 60);
1792 LE64_BITMASK(BCH_SB_BACKGROUND_COMPRESSION_TYPE_HI,
1793 struct bch_sb, flags[4], 60, 64);
1795 LE64_BITMASK(BCH_SB_VERSION_UPGRADE_COMPLETE,
1796 struct bch_sb, flags[5], 0, 16);
1798 static inline __u64 BCH_SB_COMPRESSION_TYPE(const struct bch_sb *sb)
1800 return BCH_SB_COMPRESSION_TYPE_LO(sb) | (BCH_SB_COMPRESSION_TYPE_HI(sb) << 4);
1803 static inline void SET_BCH_SB_COMPRESSION_TYPE(struct bch_sb *sb, __u64 v)
1805 SET_BCH_SB_COMPRESSION_TYPE_LO(sb, v);
1806 SET_BCH_SB_COMPRESSION_TYPE_HI(sb, v >> 4);
1809 static inline __u64 BCH_SB_BACKGROUND_COMPRESSION_TYPE(const struct bch_sb *sb)
1811 return BCH_SB_BACKGROUND_COMPRESSION_TYPE_LO(sb) |
1812 (BCH_SB_BACKGROUND_COMPRESSION_TYPE_HI(sb) << 4);
1815 static inline void SET_BCH_SB_BACKGROUND_COMPRESSION_TYPE(struct bch_sb *sb, __u64 v)
1817 SET_BCH_SB_BACKGROUND_COMPRESSION_TYPE_LO(sb, v);
1818 SET_BCH_SB_BACKGROUND_COMPRESSION_TYPE_HI(sb, v >> 4);
1824 * journal_seq_blacklist_v3: gates BCH_SB_FIELD_journal_seq_blacklist
1825 * reflink: gates KEY_TYPE_reflink
1826 * inline_data: gates KEY_TYPE_inline_data
1827 * new_siphash: gates BCH_STR_HASH_siphash
1828 * new_extent_overwrite: gates BTREE_NODE_NEW_EXTENT_OVERWRITE
1830 #define BCH_SB_FEATURES() \
1834 x(atomic_nlink, 3) \
1836 x(journal_seq_blacklist_v3, 5) \
1840 x(new_extent_overwrite, 9) \
1841 x(incompressible, 10) \
1842 x(btree_ptr_v2, 11) \
1843 x(extents_above_btree_updates, 12) \
1844 x(btree_updates_journalled, 13) \
1845 x(reflink_inline_data, 14) \
1847 x(journal_no_flush, 16) \
1849 x(extents_across_btree_nodes, 18)
1851 #define BCH_SB_FEATURES_ALWAYS \
1852 ((1ULL << BCH_FEATURE_new_extent_overwrite)| \
1853 (1ULL << BCH_FEATURE_extents_above_btree_updates)|\
1854 (1ULL << BCH_FEATURE_btree_updates_journalled)|\
1855 (1ULL << BCH_FEATURE_alloc_v2)|\
1856 (1ULL << BCH_FEATURE_extents_across_btree_nodes))
1858 #define BCH_SB_FEATURES_ALL \
1859 (BCH_SB_FEATURES_ALWAYS| \
1860 (1ULL << BCH_FEATURE_new_siphash)| \
1861 (1ULL << BCH_FEATURE_btree_ptr_v2)| \
1862 (1ULL << BCH_FEATURE_new_varint)| \
1863 (1ULL << BCH_FEATURE_journal_no_flush))
1865 enum bch_sb_feature {
1866 #define x(f, n) BCH_FEATURE_##f,
1872 #define BCH_SB_COMPAT() \
1874 x(alloc_metadata, 1) \
1875 x(extents_above_btree_updates_done, 2) \
1876 x(bformat_overflow_done, 3)
1878 enum bch_sb_compat {
1879 #define x(f, n) BCH_COMPAT_##f,
1887 #define BCH_VERSION_UPGRADE_OPTS() \
1889 x(incompatible, 1) \
1892 enum bch_version_upgrade_opts {
1893 #define x(t, n) BCH_VERSION_UPGRADE_##t = n,
1894 BCH_VERSION_UPGRADE_OPTS()
1898 #define BCH_REPLICAS_MAX 4U
1900 #define BCH_BKEY_PTRS_MAX 16U
1902 #define BCH_ERROR_ACTIONS() \
1907 enum bch_error_actions {
1908 #define x(t, n) BCH_ON_ERROR_##t = n,
1914 #define BCH_STR_HASH_TYPES() \
1920 enum bch_str_hash_type {
1921 #define x(t, n) BCH_STR_HASH_##t = n,
1922 BCH_STR_HASH_TYPES()
1927 #define BCH_STR_HASH_OPTS() \
1932 enum bch_str_hash_opts {
1933 #define x(t, n) BCH_STR_HASH_OPT_##t = n,
1939 #define BCH_CSUM_TYPES() \
1941 x(crc32c_nonzero, 1) \
1942 x(crc64_nonzero, 2) \
1943 x(chacha20_poly1305_80, 3) \
1944 x(chacha20_poly1305_128, 4) \
1949 enum bch_csum_type {
1950 #define x(t, n) BCH_CSUM_##t = n,
1956 static const unsigned bch_crc_bytes[] = {
1957 [BCH_CSUM_none] = 0,
1958 [BCH_CSUM_crc32c_nonzero] = 4,
1959 [BCH_CSUM_crc32c] = 4,
1960 [BCH_CSUM_crc64_nonzero] = 8,
1961 [BCH_CSUM_crc64] = 8,
1962 [BCH_CSUM_xxhash] = 8,
1963 [BCH_CSUM_chacha20_poly1305_80] = 10,
1964 [BCH_CSUM_chacha20_poly1305_128] = 16,
1967 static inline _Bool bch2_csum_type_is_encryption(enum bch_csum_type type)
1970 case BCH_CSUM_chacha20_poly1305_80:
1971 case BCH_CSUM_chacha20_poly1305_128:
1978 #define BCH_CSUM_OPTS() \
1984 enum bch_csum_opts {
1985 #define x(t, n) BCH_CSUM_OPT_##t = n,
1991 #define BCH_COMPRESSION_TYPES() \
1997 x(incompressible, 5)
1999 enum bch_compression_type {
2000 #define x(t, n) BCH_COMPRESSION_TYPE_##t = n,
2001 BCH_COMPRESSION_TYPES()
2003 BCH_COMPRESSION_TYPE_NR
2006 #define BCH_COMPRESSION_OPTS() \
2012 enum bch_compression_opts {
2013 #define x(t, n) BCH_COMPRESSION_OPT_##t = n,
2014 BCH_COMPRESSION_OPTS()
2016 BCH_COMPRESSION_OPT_NR
2022 * The various other data structures have their own magic numbers, which are
2023 * xored with the first part of the cache set's UUID
2026 #define BCACHE_MAGIC \
2027 UUID_INIT(0xc68573f6, 0x4e1a, 0x45ca, \
2028 0x82, 0x65, 0xf5, 0x7f, 0x48, 0xba, 0x6d, 0x81)
2029 #define BCHFS_MAGIC \
2030 UUID_INIT(0xc68573f6, 0x66ce, 0x90a9, \
2031 0xd9, 0x6a, 0x60, 0xcf, 0x80, 0x3d, 0xf7, 0xef)
2033 #define BCACHEFS_STATFS_MAGIC 0xca451a4e
2035 #define JSET_MAGIC __cpu_to_le64(0x245235c1a3625032ULL)
2036 #define BSET_MAGIC __cpu_to_le64(0x90135c78b99e07f5ULL)
2038 static inline __le64 __bch2_sb_magic(struct bch_sb *sb)
2042 memcpy(&ret, &sb->uuid, sizeof(ret));
2046 static inline __u64 __jset_magic(struct bch_sb *sb)
2048 return __le64_to_cpu(__bch2_sb_magic(sb) ^ JSET_MAGIC);
2051 static inline __u64 __bset_magic(struct bch_sb *sb)
2053 return __le64_to_cpu(__bch2_sb_magic(sb) ^ BSET_MAGIC);
2058 #define JSET_KEYS_U64s (sizeof(struct jset_entry) / sizeof(__u64))
2060 #define BCH_JSET_ENTRY_TYPES() \
2065 x(blacklist_v2, 4) \
2074 #define x(f, nr) BCH_JSET_ENTRY_##f = nr,
2075 BCH_JSET_ENTRY_TYPES()
2081 * Journal sequence numbers can be blacklisted: bsets record the max sequence
2082 * number of all the journal entries they contain updates for, so that on
2083 * recovery we can ignore those bsets that contain index updates newer that what
2084 * made it into the journal.
2086 * This means that we can't reuse that journal_seq - we have to skip it, and
2087 * then record that we skipped it so that the next time we crash and recover we
2088 * don't think there was a missing journal entry.
2090 struct jset_entry_blacklist {
2091 struct jset_entry entry;
2095 struct jset_entry_blacklist_v2 {
2096 struct jset_entry entry;
2101 #define BCH_FS_USAGE_TYPES() \
2107 #define x(f, nr) BCH_FS_USAGE_##f = nr,
2108 BCH_FS_USAGE_TYPES()
2113 struct jset_entry_usage {
2114 struct jset_entry entry;
2118 struct jset_entry_data_usage {
2119 struct jset_entry entry;
2121 struct bch_replicas_entry r;
2124 struct jset_entry_clock {
2125 struct jset_entry entry;
2131 struct jset_entry_dev_usage_type {
2137 struct jset_entry_dev_usage {
2138 struct jset_entry entry;
2143 __le64 _buckets_unavailable; /* No longer used */
2145 struct jset_entry_dev_usage_type d[];
2148 static inline unsigned jset_entry_dev_usage_nr_types(struct jset_entry_dev_usage *u)
2150 return (vstruct_bytes(&u->entry) - sizeof(struct jset_entry_dev_usage)) /
2151 sizeof(struct jset_entry_dev_usage_type);
2154 struct jset_entry_log {
2155 struct jset_entry entry;
2160 * On disk format for a journal entry:
2161 * seq is monotonically increasing; every journal entry has its own unique
2164 * last_seq is the oldest journal entry that still has keys the btree hasn't
2165 * flushed to disk yet.
2167 * version is for on disk format changes.
2170 struct bch_csum csum;
2177 __le32 u64s; /* size of d[] in u64s */
2179 __u8 encrypted_start[0];
2181 __le16 _read_clock; /* no longer used */
2182 __le16 _write_clock;
2184 /* Sequence number of oldest dirty journal entry */
2189 struct jset_entry start[0];
2192 } __packed __aligned(8);
2194 LE32_BITMASK(JSET_CSUM_TYPE, struct jset, flags, 0, 4);
2195 LE32_BITMASK(JSET_BIG_ENDIAN, struct jset, flags, 4, 5);
2196 LE32_BITMASK(JSET_NO_FLUSH, struct jset, flags, 5, 6);
2198 #define BCH_JOURNAL_BUCKETS_MIN 8
2202 enum btree_id_flags {
2203 BTREE_ID_EXTENTS = BIT(0),
2204 BTREE_ID_SNAPSHOTS = BIT(1),
2205 BTREE_ID_DATA = BIT(2),
2208 #define BCH_BTREE_IDS() \
2209 x(extents, 0, BTREE_ID_EXTENTS|BTREE_ID_SNAPSHOTS|BTREE_ID_DATA,\
2210 BIT_ULL(KEY_TYPE_whiteout)| \
2211 BIT_ULL(KEY_TYPE_error)| \
2212 BIT_ULL(KEY_TYPE_cookie)| \
2213 BIT_ULL(KEY_TYPE_extent)| \
2214 BIT_ULL(KEY_TYPE_reservation)| \
2215 BIT_ULL(KEY_TYPE_reflink_p)| \
2216 BIT_ULL(KEY_TYPE_inline_data)) \
2217 x(inodes, 1, BTREE_ID_SNAPSHOTS, \
2218 BIT_ULL(KEY_TYPE_whiteout)| \
2219 BIT_ULL(KEY_TYPE_inode)| \
2220 BIT_ULL(KEY_TYPE_inode_v2)| \
2221 BIT_ULL(KEY_TYPE_inode_v3)| \
2222 BIT_ULL(KEY_TYPE_inode_generation)) \
2223 x(dirents, 2, BTREE_ID_SNAPSHOTS, \
2224 BIT_ULL(KEY_TYPE_whiteout)| \
2225 BIT_ULL(KEY_TYPE_hash_whiteout)| \
2226 BIT_ULL(KEY_TYPE_dirent)) \
2227 x(xattrs, 3, BTREE_ID_SNAPSHOTS, \
2228 BIT_ULL(KEY_TYPE_whiteout)| \
2229 BIT_ULL(KEY_TYPE_cookie)| \
2230 BIT_ULL(KEY_TYPE_hash_whiteout)| \
2231 BIT_ULL(KEY_TYPE_xattr)) \
2233 BIT_ULL(KEY_TYPE_alloc)| \
2234 BIT_ULL(KEY_TYPE_alloc_v2)| \
2235 BIT_ULL(KEY_TYPE_alloc_v3)| \
2236 BIT_ULL(KEY_TYPE_alloc_v4)) \
2238 BIT_ULL(KEY_TYPE_quota)) \
2240 BIT_ULL(KEY_TYPE_stripe)) \
2241 x(reflink, 7, BTREE_ID_EXTENTS|BTREE_ID_DATA, \
2242 BIT_ULL(KEY_TYPE_reflink_v)| \
2243 BIT_ULL(KEY_TYPE_indirect_inline_data)) \
2244 x(subvolumes, 8, 0, \
2245 BIT_ULL(KEY_TYPE_subvolume)) \
2246 x(snapshots, 9, 0, \
2247 BIT_ULL(KEY_TYPE_snapshot)) \
2249 BIT_ULL(KEY_TYPE_set)) \
2250 x(freespace, 11, BTREE_ID_EXTENTS, \
2251 BIT_ULL(KEY_TYPE_set)) \
2252 x(need_discard, 12, 0, \
2253 BIT_ULL(KEY_TYPE_set)) \
2254 x(backpointers, 13, 0, \
2255 BIT_ULL(KEY_TYPE_backpointer)) \
2256 x(bucket_gens, 14, 0, \
2257 BIT_ULL(KEY_TYPE_bucket_gens)) \
2258 x(snapshot_trees, 15, 0, \
2259 BIT_ULL(KEY_TYPE_snapshot_tree)) \
2260 x(deleted_inodes, 16, BTREE_ID_SNAPSHOTS, \
2261 BIT_ULL(KEY_TYPE_set))
2264 #define x(name, nr, ...) BTREE_ID_##name = nr,
2270 #define BTREE_MAX_DEPTH 4U
2277 * On disk a btree node is a list/log of these; within each set the keys are
2284 * Highest journal entry this bset contains keys for.
2285 * If on recovery we don't see that journal entry, this bset is ignored:
2286 * this allows us to preserve the order of all index updates after a
2287 * crash, since the journal records a total order of all index updates
2288 * and anything that didn't make it to the journal doesn't get used.
2294 __le16 u64s; /* count of d[] in u64s */
2297 struct bkey_packed start[0];
2300 } __packed __aligned(8);
2302 LE32_BITMASK(BSET_CSUM_TYPE, struct bset, flags, 0, 4);
2304 LE32_BITMASK(BSET_BIG_ENDIAN, struct bset, flags, 4, 5);
2305 LE32_BITMASK(BSET_SEPARATE_WHITEOUTS,
2306 struct bset, flags, 5, 6);
2308 /* Sector offset within the btree node: */
2309 LE32_BITMASK(BSET_OFFSET, struct bset, flags, 16, 32);
2312 struct bch_csum csum;
2315 /* this flags field is encrypted, unlike bset->flags: */
2318 /* Closed interval: */
2319 struct bpos min_key;
2320 struct bpos max_key;
2321 struct bch_extent_ptr _ptr; /* not used anymore */
2322 struct bkey_format format;
2333 } __packed __aligned(8);
2335 LE64_BITMASK(BTREE_NODE_ID_LO, struct btree_node, flags, 0, 4);
2336 LE64_BITMASK(BTREE_NODE_LEVEL, struct btree_node, flags, 4, 8);
2337 LE64_BITMASK(BTREE_NODE_NEW_EXTENT_OVERWRITE,
2338 struct btree_node, flags, 8, 9);
2339 LE64_BITMASK(BTREE_NODE_ID_HI, struct btree_node, flags, 9, 25);
2341 LE64_BITMASK(BTREE_NODE_SEQ, struct btree_node, flags, 32, 64);
2343 static inline __u64 BTREE_NODE_ID(struct btree_node *n)
2345 return BTREE_NODE_ID_LO(n) | (BTREE_NODE_ID_HI(n) << 4);
2348 static inline void SET_BTREE_NODE_ID(struct btree_node *n, __u64 v)
2350 SET_BTREE_NODE_ID_LO(n, v);
2351 SET_BTREE_NODE_ID_HI(n, v >> 4);
2354 struct btree_node_entry {
2355 struct bch_csum csum;
2365 } __packed __aligned(8);
2367 #endif /* _BCACHEFS_FORMAT_H */