1 /* SPDX-License-Identifier: GPL-2.0 */
2 #ifndef _BCACHEFS_FORMAT_H
3 #define _BCACHEFS_FORMAT_H
6 * bcachefs on disk data structures
10 * There are three main types of on disk data structures in bcachefs (this is
11 * reduced from 5 in bcache)
17 * The btree is the primary structure; most metadata exists as keys in the
18 * various btrees. There are only a small number of btrees, they're not
19 * sharded - we have one btree for extents, another for inodes, et cetera.
23 * The superblock contains the location of the journal, the list of devices in
24 * the filesystem, and in general any metadata we need in order to decide
25 * whether we can start a filesystem or prior to reading the journal/btree
28 * The superblock is extensible, and most of the contents of the superblock are
29 * in variable length, type tagged fields; see struct bch_sb_field.
31 * Backup superblocks do not reside in a fixed location; also, superblocks do
32 * not have a fixed size. To locate backup superblocks we have struct
33 * bch_sb_layout; we store a copy of this inside every superblock, and also
34 * before the first superblock.
38 * The journal primarily records btree updates in the order they occurred;
39 * journal replay consists of just iterating over all the keys in the open
40 * journal entries and re-inserting them into the btrees.
42 * The journal also contains entry types for the btree roots, and blacklisted
43 * journal sequence numbers (see journal_seq_blacklist.c).
47 * bcachefs btrees are copy on write b+ trees, where nodes are big (typically
48 * 128k-256k) and log structured. We use struct btree_node for writing the first
49 * entry in a given node (offset 0), and struct btree_node_entry for all
52 * After the header, btree node entries contain a list of keys in sorted order.
53 * Values are stored inline with the keys; since values are variable length (and
54 * keys effectively are variable length too, due to packing) we can't do random
55 * access without building up additional in memory tables in the btree node read
58 * BTREE KEYS (struct bkey):
60 * The various btrees share a common format for the key - so as to avoid
61 * switching in fastpath lookup/comparison code - but define their own
62 * structures for the key values.
64 * The size of a key/value pair is stored as a u8 in units of u64s, so the max
65 * size is just under 2k. The common part also contains a type tag for the
66 * value, and a format field indicating whether the key is packed or not (and
67 * also meant to allow adding new key fields in the future, if desired).
69 * bkeys, when stored within a btree node, may also be packed. In that case, the
70 * bkey_format in that node is used to unpack it. Packed bkeys mean that we can
71 * be generous with field sizes in the common part of the key format (64 bit
72 * inode number, 64 bit offset, 96 bit version field, etc.) for negligible cost.
75 #include <asm/types.h>
76 #include <asm/byteorder.h>
77 #include <linux/kernel.h>
78 #include <linux/uuid.h>
82 typedef uuid_t __uuid_t;
85 #define BITMASK(name, type, field, offset, end) \
86 static const __maybe_unused unsigned name##_OFFSET = offset; \
87 static const __maybe_unused unsigned name##_BITS = (end - offset); \
89 static inline __u64 name(const type *k) \
91 return (k->field >> offset) & ~(~0ULL << (end - offset)); \
94 static inline void SET_##name(type *k, __u64 v) \
96 k->field &= ~(~(~0ULL << (end - offset)) << offset); \
97 k->field |= (v & ~(~0ULL << (end - offset))) << offset; \
100 #define LE_BITMASK(_bits, name, type, field, offset, end) \
101 static const __maybe_unused unsigned name##_OFFSET = offset; \
102 static const __maybe_unused unsigned name##_BITS = (end - offset); \
103 static const __maybe_unused __u##_bits name##_MAX = (1ULL << (end - offset)) - 1;\
105 static inline __u64 name(const type *k) \
107 return (__le##_bits##_to_cpu(k->field) >> offset) & \
108 ~(~0ULL << (end - offset)); \
111 static inline void SET_##name(type *k, __u64 v) \
113 __u##_bits new = __le##_bits##_to_cpu(k->field); \
115 new &= ~(~(~0ULL << (end - offset)) << offset); \
116 new |= (v & ~(~0ULL << (end - offset))) << offset; \
117 k->field = __cpu_to_le##_bits(new); \
120 #define LE16_BITMASK(n, t, f, o, e) LE_BITMASK(16, n, t, f, o, e)
121 #define LE32_BITMASK(n, t, f, o, e) LE_BITMASK(32, n, t, f, o, e)
122 #define LE64_BITMASK(n, t, f, o, e) LE_BITMASK(64, n, t, f, o, e)
127 /* One unused slot for now: */
128 __u8 bits_per_field[6];
129 __le64 field_offset[6];
132 /* Btree keys - all units are in sectors */
136 * Word order matches machine byte order - btree code treats a bpos as a
137 * single large integer, for search/comparison purposes
139 * Note that wherever a bpos is embedded in another on disk data
140 * structure, it has to be byte swabbed when reading in metadata that
141 * wasn't written in native endian order:
143 #if __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__
147 #elif __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__
149 __u64 offset; /* Points to end of extent - sectors */
152 #error edit for your odd byteorder.
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) \
373 x(snapshot_tree, 31) \
374 x(logged_op_truncate, 32) \
375 x(logged_op_finsert, 33)
378 #define x(name, nr) KEY_TYPE_##name = nr,
388 struct bch_whiteout {
401 struct bch_hash_whiteout {
412 * In extent bkeys, the value is a list of pointers (bch_extent_ptr), optionally
413 * preceded by checksum/compression information (bch_extent_crc32 or
416 * One major determining factor in the format of extents is how we handle and
417 * represent extents that have been partially overwritten and thus trimmed:
419 * If an extent is not checksummed or compressed, when the extent is trimmed we
420 * don't have to remember the extent we originally allocated and wrote: we can
421 * merely adjust ptr->offset to point to the start of the data that is currently
422 * live. The size field in struct bkey records the current (live) size of the
423 * extent, and is also used to mean "size of region on disk that we point to" in
426 * Thus an extent that is not checksummed or compressed will consist only of a
427 * list of bch_extent_ptrs, with none of the fields in
428 * bch_extent_crc32/bch_extent_crc64.
430 * When an extent is checksummed or compressed, it's not possible to read only
431 * the data that is currently live: we have to read the entire extent that was
432 * originally written, and then return only the part of the extent that is
435 * Thus, in addition to the current size of the extent in struct bkey, we need
436 * to store the size of the originally allocated space - this is the
437 * compressed_size and uncompressed_size fields in bch_extent_crc32/64. Also,
438 * when the extent is trimmed, instead of modifying the offset field of the
439 * pointer, we keep a second smaller offset field - "offset into the original
440 * extent of the currently live region".
442 * The other major determining factor is replication and data migration:
444 * Each pointer may have its own bch_extent_crc32/64. When doing a replicated
445 * write, we will initially write all the replicas in the same format, with the
446 * same checksum type and compression format - however, when copygc runs later (or
447 * tiering/cache promotion, anything that moves data), it is not in general
448 * going to rewrite all the pointers at once - one of the replicas may be in a
449 * bucket on one device that has very little fragmentation while another lives
450 * in a bucket that has become heavily fragmented, and thus is being rewritten
451 * sooner than the rest.
453 * Thus it will only move a subset of the pointers (or in the case of
454 * tiering/cache promotion perhaps add a single pointer without dropping any
455 * current pointers), and if the extent has been partially overwritten it must
456 * write only the currently live portion (or copygc would not be able to reduce
457 * fragmentation!) - which necessitates a different bch_extent_crc format for
460 * But in the interests of space efficiency, we don't want to store one
461 * bch_extent_crc for each pointer if we don't have to.
463 * Thus, a bch_extent consists of bch_extent_crc32s, bch_extent_crc64s, and
464 * bch_extent_ptrs appended arbitrarily one after the other. We determine the
465 * type of a given entry with a scheme similar to utf8 (except we're encoding a
466 * type, not a size), encoding the type in the position of the first set bit:
468 * bch_extent_crc32 - 0b1
469 * bch_extent_ptr - 0b10
470 * bch_extent_crc64 - 0b100
472 * We do it this way because bch_extent_crc32 is _very_ constrained on bits (and
473 * bch_extent_crc64 is the least constrained).
475 * Then, each bch_extent_crc32/64 applies to the pointers that follow after it,
476 * until the next bch_extent_crc32/64.
478 * If there are no bch_extent_crcs preceding a bch_extent_ptr, then that pointer
479 * is neither checksummed nor compressed.
482 /* 128 bits, sufficient for cryptographic MACs: */
486 } __packed __aligned(8);
488 #define BCH_EXTENT_ENTRY_TYPES() \
495 #define BCH_EXTENT_ENTRY_MAX 6
497 enum bch_extent_entry_type {
498 #define x(f, n) BCH_EXTENT_ENTRY_##f = n,
499 BCH_EXTENT_ENTRY_TYPES()
503 /* Compressed/uncompressed size are stored biased by 1: */
504 struct bch_extent_crc32 {
505 #if defined(__LITTLE_ENDIAN_BITFIELD)
508 _uncompressed_size:7,
514 #elif defined (__BIG_ENDIAN_BITFIELD)
516 __u32 compression_type:4,
520 _uncompressed_size:7,
524 } __packed __aligned(8);
526 #define CRC32_SIZE_MAX (1U << 7)
527 #define CRC32_NONCE_MAX 0
529 struct bch_extent_crc64 {
530 #if defined(__LITTLE_ENDIAN_BITFIELD)
533 _uncompressed_size:9,
539 #elif defined (__BIG_ENDIAN_BITFIELD)
545 _uncompressed_size:9,
550 } __packed __aligned(8);
552 #define CRC64_SIZE_MAX (1U << 9)
553 #define CRC64_NONCE_MAX ((1U << 10) - 1)
555 struct bch_extent_crc128 {
556 #if defined(__LITTLE_ENDIAN_BITFIELD)
559 _uncompressed_size:13,
564 #elif defined (__BIG_ENDIAN_BITFIELD)
565 __u64 compression_type:4,
569 _uncompressed_size:13,
573 struct bch_csum csum;
574 } __packed __aligned(8);
576 #define CRC128_SIZE_MAX (1U << 13)
577 #define CRC128_NONCE_MAX ((1U << 13) - 1)
580 * @reservation - pointer hasn't been written to, just reserved
582 struct bch_extent_ptr {
583 #if defined(__LITTLE_ENDIAN_BITFIELD)
588 offset:44, /* 8 petabytes */
591 #elif defined (__BIG_ENDIAN_BITFIELD)
600 } __packed __aligned(8);
602 struct bch_extent_stripe_ptr {
603 #if defined(__LITTLE_ENDIAN_BITFIELD)
608 #elif defined (__BIG_ENDIAN_BITFIELD)
616 struct bch_extent_rebalance {
617 #if defined(__LITTLE_ENDIAN_BITFIELD)
620 compression:8, /* enum bch_compression_opt */
622 #elif defined (__BIG_ENDIAN_BITFIELD)
630 union bch_extent_entry {
631 #if __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__ || __BITS_PER_LONG == 64
633 #elif __BITS_PER_LONG == 32
639 #error edit for your odd byteorder.
642 #define x(f, n) struct bch_extent_##f f;
643 BCH_EXTENT_ENTRY_TYPES()
647 struct bch_btree_ptr {
651 struct bch_extent_ptr start[];
652 } __packed __aligned(8);
654 struct bch_btree_ptr_v2 {
659 __le16 sectors_written;
663 struct bch_extent_ptr start[];
664 } __packed __aligned(8);
666 LE16_BITMASK(BTREE_PTR_RANGE_UPDATED, struct bch_btree_ptr_v2, flags, 0, 1);
672 union bch_extent_entry start[];
673 } __packed __aligned(8);
675 struct bch_reservation {
681 } __packed __aligned(8);
683 /* Maximum size (in u64s) a single pointer could be: */
684 #define BKEY_EXTENT_PTR_U64s_MAX\
685 ((sizeof(struct bch_extent_crc128) + \
686 sizeof(struct bch_extent_ptr)) / sizeof(__u64))
688 /* Maximum possible size of an entire extent value: */
689 #define BKEY_EXTENT_VAL_U64s_MAX \
690 (1 + BKEY_EXTENT_PTR_U64s_MAX * (BCH_REPLICAS_MAX + 1))
692 /* * Maximum possible size of an entire extent, key + value: */
693 #define BKEY_EXTENT_U64s_MAX (BKEY_U64s + BKEY_EXTENT_VAL_U64s_MAX)
695 /* Btree pointers don't carry around checksums: */
696 #define BKEY_BTREE_PTR_VAL_U64s_MAX \
697 ((sizeof(struct bch_btree_ptr_v2) + \
698 sizeof(struct bch_extent_ptr) * BCH_REPLICAS_MAX) / sizeof(__u64))
699 #define BKEY_BTREE_PTR_U64s_MAX \
700 (BKEY_U64s + BKEY_BTREE_PTR_VAL_U64s_MAX)
704 #define BLOCKDEV_INODE_MAX 4096
706 #define BCACHEFS_ROOT_INO 4096
715 } __packed __aligned(8);
717 struct bch_inode_v2 {
720 __le64 bi_journal_seq;
725 } __packed __aligned(8);
727 struct bch_inode_v3 {
730 __le64 bi_journal_seq;
737 } __packed __aligned(8);
739 #define INODEv3_FIELDS_START_INITIAL 6
740 #define INODEv3_FIELDS_START_CUR (offsetof(struct bch_inode_v3, fields) / sizeof(__u64))
742 struct bch_inode_generation {
745 __le32 bi_generation;
747 } __packed __aligned(8);
750 * bi_subvol and bi_parent_subvol are only set for subvolume roots:
753 #define BCH_INODE_FIELDS_v2() \
763 x(bi_generation, 32) \
765 x(bi_data_checksum, 8) \
766 x(bi_compression, 8) \
768 x(bi_background_compression, 8) \
769 x(bi_data_replicas, 8) \
770 x(bi_promote_target, 16) \
771 x(bi_foreground_target, 16) \
772 x(bi_background_target, 16) \
773 x(bi_erasure_code, 16) \
774 x(bi_fields_set, 16) \
776 x(bi_dir_offset, 64) \
778 x(bi_parent_subvol, 32)
780 #define BCH_INODE_FIELDS_v3() \
788 x(bi_generation, 32) \
790 x(bi_data_checksum, 8) \
791 x(bi_compression, 8) \
793 x(bi_background_compression, 8) \
794 x(bi_data_replicas, 8) \
795 x(bi_promote_target, 16) \
796 x(bi_foreground_target, 16) \
797 x(bi_background_target, 16) \
798 x(bi_erasure_code, 16) \
799 x(bi_fields_set, 16) \
801 x(bi_dir_offset, 64) \
803 x(bi_parent_subvol, 32) \
806 /* subset of BCH_INODE_FIELDS */
807 #define BCH_INODE_OPTS() \
808 x(data_checksum, 8) \
811 x(background_compression, 8) \
812 x(data_replicas, 8) \
813 x(promote_target, 16) \
814 x(foreground_target, 16) \
815 x(background_target, 16) \
816 x(erasure_code, 16) \
820 #define x(name, ...) \
829 * User flags (get/settable with FS_IOC_*FLAGS, correspond to FS_*_FL
832 __BCH_INODE_SYNC = 0,
833 __BCH_INODE_IMMUTABLE = 1,
834 __BCH_INODE_APPEND = 2,
835 __BCH_INODE_NODUMP = 3,
836 __BCH_INODE_NOATIME = 4,
838 __BCH_INODE_I_SIZE_DIRTY = 5, /* obsolete */
839 __BCH_INODE_I_SECTORS_DIRTY = 6, /* obsolete */
840 __BCH_INODE_UNLINKED = 7,
841 __BCH_INODE_BACKPTR_UNTRUSTED = 8,
843 /* bits 20+ reserved for packed fields below: */
846 #define BCH_INODE_SYNC (1 << __BCH_INODE_SYNC)
847 #define BCH_INODE_IMMUTABLE (1 << __BCH_INODE_IMMUTABLE)
848 #define BCH_INODE_APPEND (1 << __BCH_INODE_APPEND)
849 #define BCH_INODE_NODUMP (1 << __BCH_INODE_NODUMP)
850 #define BCH_INODE_NOATIME (1 << __BCH_INODE_NOATIME)
851 #define BCH_INODE_I_SIZE_DIRTY (1 << __BCH_INODE_I_SIZE_DIRTY)
852 #define BCH_INODE_I_SECTORS_DIRTY (1 << __BCH_INODE_I_SECTORS_DIRTY)
853 #define BCH_INODE_UNLINKED (1 << __BCH_INODE_UNLINKED)
854 #define BCH_INODE_BACKPTR_UNTRUSTED (1 << __BCH_INODE_BACKPTR_UNTRUSTED)
856 LE32_BITMASK(INODE_STR_HASH, struct bch_inode, bi_flags, 20, 24);
857 LE32_BITMASK(INODE_NR_FIELDS, struct bch_inode, bi_flags, 24, 31);
858 LE32_BITMASK(INODE_NEW_VARINT, struct bch_inode, bi_flags, 31, 32);
860 LE64_BITMASK(INODEv2_STR_HASH, struct bch_inode_v2, bi_flags, 20, 24);
861 LE64_BITMASK(INODEv2_NR_FIELDS, struct bch_inode_v2, bi_flags, 24, 31);
863 LE64_BITMASK(INODEv3_STR_HASH, struct bch_inode_v3, bi_flags, 20, 24);
864 LE64_BITMASK(INODEv3_NR_FIELDS, struct bch_inode_v3, bi_flags, 24, 31);
866 LE64_BITMASK(INODEv3_FIELDS_START,
867 struct bch_inode_v3, bi_flags, 31, 36);
868 LE64_BITMASK(INODEv3_MODE, struct bch_inode_v3, bi_flags, 36, 52);
873 * Dirents (and xattrs) have to implement string lookups; since our b-tree
874 * doesn't support arbitrary length strings for the key, we instead index by a
875 * 64 bit hash (currently truncated sha1) of the string, stored in the offset
876 * field of the key - using linear probing to resolve hash collisions. This also
877 * provides us with the readdir cookie posix requires.
879 * Linear probing requires us to use whiteouts for deletions, in the event of a
886 /* Target inode number: */
889 struct { /* DT_SUBVOL */
890 __le32 d_child_subvol;
891 __le32 d_parent_subvol;
896 * Copy of mode bits 12-15 from the target inode - so userspace can get
897 * the filetype without having to do a stat()
902 } __packed __aligned(8);
905 #define BCH_DT_MAX 17
907 #define BCH_NAME_MAX 512
911 #define KEY_TYPE_XATTR_INDEX_USER 0
912 #define KEY_TYPE_XATTR_INDEX_POSIX_ACL_ACCESS 1
913 #define KEY_TYPE_XATTR_INDEX_POSIX_ACL_DEFAULT 2
914 #define KEY_TYPE_XATTR_INDEX_TRUSTED 3
915 #define KEY_TYPE_XATTR_INDEX_SECURITY 4
923 } __packed __aligned(8);
925 /* Bucket/allocation information: */
932 } __packed __aligned(8);
934 #define BCH_ALLOC_FIELDS_V1() \
938 x(dirty_sectors, 16) \
939 x(cached_sectors, 16) \
942 x(stripe_redundancy, 8)
945 #define x(name, _bits) BCH_ALLOC_FIELD_V1_##name,
946 BCH_ALLOC_FIELDS_V1()
950 struct bch_alloc_v2 {
957 } __packed __aligned(8);
959 #define BCH_ALLOC_FIELDS_V2() \
962 x(dirty_sectors, 32) \
963 x(cached_sectors, 32) \
965 x(stripe_redundancy, 8)
967 struct bch_alloc_v3 {
976 } __packed __aligned(8);
978 LE32_BITMASK(BCH_ALLOC_V3_NEED_DISCARD,struct bch_alloc_v3, flags, 0, 1)
979 LE32_BITMASK(BCH_ALLOC_V3_NEED_INC_GEN,struct bch_alloc_v3, flags, 1, 2)
981 struct bch_alloc_v4 {
988 __u8 stripe_redundancy;
990 __u32 cached_sectors;
993 __u32 nr_external_backpointers;
994 __u64 fragmentation_lru;
995 } __packed __aligned(8);
997 #define BCH_ALLOC_V4_U64s_V0 6
998 #define BCH_ALLOC_V4_U64s (sizeof(struct bch_alloc_v4) / sizeof(__u64))
1000 BITMASK(BCH_ALLOC_V4_NEED_DISCARD, struct bch_alloc_v4, flags, 0, 1)
1001 BITMASK(BCH_ALLOC_V4_NEED_INC_GEN, struct bch_alloc_v4, flags, 1, 2)
1002 BITMASK(BCH_ALLOC_V4_BACKPOINTERS_START,struct bch_alloc_v4, flags, 2, 8)
1003 BITMASK(BCH_ALLOC_V4_NR_BACKPOINTERS, struct bch_alloc_v4, flags, 8, 14)
1005 #define BCH_ALLOC_V4_NR_BACKPOINTERS_MAX 40
1007 struct bch_backpointer {
1012 __u64 bucket_offset:40;
1015 } __packed __aligned(8);
1017 #define KEY_TYPE_BUCKET_GENS_BITS 8
1018 #define KEY_TYPE_BUCKET_GENS_NR (1U << KEY_TYPE_BUCKET_GENS_BITS)
1019 #define KEY_TYPE_BUCKET_GENS_MASK (KEY_TYPE_BUCKET_GENS_NR - 1)
1021 struct bch_bucket_gens {
1023 u8 gens[KEY_TYPE_BUCKET_GENS_NR];
1024 } __packed __aligned(8);
1035 enum quota_counters {
1041 struct bch_quota_counter {
1048 struct bch_quota_counter c[Q_COUNTERS];
1049 } __packed __aligned(8);
1051 /* Erasure coding */
1060 __u8 csum_granularity_bits;
1064 struct bch_extent_ptr ptrs[];
1065 } __packed __aligned(8);
1069 struct bch_reflink_p {
1073 * A reflink pointer might point to an indirect extent which is then
1074 * later split (by copygc or rebalance). If we only pointed to part of
1075 * the original indirect extent, and then one of the fragments is
1076 * outside the range we point to, we'd leak a refcount: so when creating
1077 * reflink pointers, we need to store pad values to remember the full
1078 * range we were taking a reference on.
1082 } __packed __aligned(8);
1084 struct bch_reflink_v {
1087 union bch_extent_entry start[0];
1089 } __packed __aligned(8);
1091 struct bch_indirect_inline_data {
1099 struct bch_inline_data {
1106 #define SUBVOL_POS_MIN POS(0, 1)
1107 #define SUBVOL_POS_MAX POS(0, S32_MAX)
1108 #define BCACHEFS_ROOT_SUBVOL 1
1110 struct bch_subvolume {
1116 * Snapshot subvolumes form a tree, separate from the snapshot nodes
1117 * tree - if this subvolume is a snapshot, this is the ID of the
1118 * subvolume it was created from:
1125 LE32_BITMASK(BCH_SUBVOLUME_RO, struct bch_subvolume, flags, 0, 1)
1127 * We need to know whether a subvolume is a snapshot so we can know whether we
1128 * can delete it (or whether it should just be rm -rf'd)
1130 LE32_BITMASK(BCH_SUBVOLUME_SNAP, struct bch_subvolume, flags, 1, 2)
1131 LE32_BITMASK(BCH_SUBVOLUME_UNLINKED, struct bch_subvolume, flags, 2, 3)
1135 struct bch_snapshot {
1141 /* corresponds to a bch_snapshot_tree in BTREE_ID_snapshot_trees */
1147 LE32_BITMASK(BCH_SNAPSHOT_DELETED, struct bch_snapshot, flags, 0, 1)
1149 /* True if a subvolume points to this snapshot node: */
1150 LE32_BITMASK(BCH_SNAPSHOT_SUBVOL, struct bch_snapshot, flags, 1, 2)
1155 * The snapshot_trees btree gives us persistent indentifier for each tree of
1156 * bch_snapshot nodes, and allow us to record and easily find the root/master
1157 * subvolume that other snapshots were created from:
1159 struct bch_snapshot_tree {
1161 __le32 master_subvol;
1162 __le32 root_snapshot;
1170 } __packed __aligned(8);
1172 #define LRU_ID_STRIPES (1U << 16)
1174 /* Logged operations btree: */
1176 struct bch_logged_op_truncate {
1184 enum logged_op_finsert_state {
1185 LOGGED_OP_FINSERT_start,
1186 LOGGED_OP_FINSERT_shift_extents,
1187 LOGGED_OP_FINSERT_finish,
1190 struct bch_logged_op_finsert {
1201 /* Optional/variable size superblock sections: */
1203 struct bch_sb_field {
1209 #define BCH_SB_FIELDS() \
1218 x(journal_seq_blacklist, 8) \
1223 enum bch_sb_field_type {
1224 #define x(f, nr) BCH_SB_FIELD_##f = nr,
1231 * Most superblock fields are replicated in all device's superblocks - a few are
1234 #define BCH_SINGLE_DEVICE_SB_FIELDS \
1235 ((1U << BCH_SB_FIELD_journal)| \
1236 (1U << BCH_SB_FIELD_journal_v2))
1238 /* BCH_SB_FIELD_journal: */
1240 struct bch_sb_field_journal {
1241 struct bch_sb_field field;
1245 struct bch_sb_field_journal_v2 {
1246 struct bch_sb_field field;
1248 struct bch_sb_field_journal_v2_entry {
1254 /* BCH_SB_FIELD_members_v1: */
1256 #define BCH_MIN_NR_NBUCKETS (1 << 6)
1258 #define BCH_IOPS_MEASUREMENTS() \
1264 enum bch_iops_measurement {
1265 #define x(t, n) BCH_IOPS_##t = n,
1266 BCH_IOPS_MEASUREMENTS()
1273 __le64 nbuckets; /* device size */
1274 __le16 first_bucket; /* index of first bucket used */
1275 __le16 bucket_size; /* sectors */
1277 __le64 last_mount; /* time_t */
1283 #define BCH_MEMBER_V1_BYTES 56
1285 LE64_BITMASK(BCH_MEMBER_STATE, struct bch_member, flags, 0, 4)
1286 /* 4-14 unused, was TIER, HAS_(META)DATA, REPLACEMENT */
1287 LE64_BITMASK(BCH_MEMBER_DISCARD, struct bch_member, flags, 14, 15)
1288 LE64_BITMASK(BCH_MEMBER_DATA_ALLOWED, struct bch_member, flags, 15, 20)
1289 LE64_BITMASK(BCH_MEMBER_GROUP, struct bch_member, flags, 20, 28)
1290 LE64_BITMASK(BCH_MEMBER_DURABILITY, struct bch_member, flags, 28, 30)
1291 LE64_BITMASK(BCH_MEMBER_FREESPACE_INITIALIZED,
1292 struct bch_member, flags, 30, 31)
1295 LE64_BITMASK(BCH_MEMBER_NR_READ_ERRORS, struct bch_member, flags[1], 0, 20);
1296 LE64_BITMASK(BCH_MEMBER_NR_WRITE_ERRORS,struct bch_member, flags[1], 20, 40);
1299 #define BCH_MEMBER_STATES() \
1305 enum bch_member_state {
1306 #define x(t, n) BCH_MEMBER_STATE_##t = n,
1312 struct bch_sb_field_members_v1 {
1313 struct bch_sb_field field;
1314 struct bch_member _members[]; //Members are now variable size
1317 struct bch_sb_field_members_v2 {
1318 struct bch_sb_field field;
1319 __le16 member_bytes; //size of single member entry
1321 struct bch_member _members[];
1324 /* BCH_SB_FIELD_crypt: */
1334 #define BCH_KEY_MAGIC \
1335 (((__u64) 'b' << 0)|((__u64) 'c' << 8)| \
1336 ((__u64) 'h' << 16)|((__u64) '*' << 24)| \
1337 ((__u64) '*' << 32)|((__u64) 'k' << 40)| \
1338 ((__u64) 'e' << 48)|((__u64) 'y' << 56))
1340 struct bch_encrypted_key {
1346 * If this field is present in the superblock, it stores an encryption key which
1347 * is used encrypt all other data/metadata. The key will normally be encrypted
1348 * with the key userspace provides, but if encryption has been turned off we'll
1349 * just store the master key unencrypted in the superblock so we can access the
1350 * previously encrypted data.
1352 struct bch_sb_field_crypt {
1353 struct bch_sb_field field;
1357 struct bch_encrypted_key key;
1360 LE64_BITMASK(BCH_CRYPT_KDF_TYPE, struct bch_sb_field_crypt, flags, 0, 4);
1362 enum bch_kdf_types {
1367 /* stored as base 2 log of scrypt params: */
1368 LE64_BITMASK(BCH_KDF_SCRYPT_N, struct bch_sb_field_crypt, kdf_flags, 0, 16);
1369 LE64_BITMASK(BCH_KDF_SCRYPT_R, struct bch_sb_field_crypt, kdf_flags, 16, 32);
1370 LE64_BITMASK(BCH_KDF_SCRYPT_P, struct bch_sb_field_crypt, kdf_flags, 32, 48);
1372 /* BCH_SB_FIELD_replicas: */
1374 #define BCH_DATA_TYPES() \
1383 x(need_gc_gens, 8) \
1386 enum bch_data_type {
1387 #define x(t, n) BCH_DATA_##t,
1393 static inline bool data_type_is_empty(enum bch_data_type type)
1397 case BCH_DATA_need_gc_gens:
1398 case BCH_DATA_need_discard:
1405 static inline bool data_type_is_hidden(enum bch_data_type type)
1409 case BCH_DATA_journal:
1416 struct bch_replicas_entry_v0 {
1422 struct bch_sb_field_replicas_v0 {
1423 struct bch_sb_field field;
1424 struct bch_replicas_entry_v0 entries[];
1425 } __packed __aligned(8);
1427 struct bch_replicas_entry {
1434 #define replicas_entry_bytes(_i) \
1435 (offsetof(typeof(*(_i)), devs) + (_i)->nr_devs)
1437 struct bch_sb_field_replicas {
1438 struct bch_sb_field field;
1439 struct bch_replicas_entry entries[];
1440 } __packed __aligned(8);
1442 /* BCH_SB_FIELD_quota: */
1444 struct bch_sb_quota_counter {
1449 struct bch_sb_quota_type {
1451 struct bch_sb_quota_counter c[Q_COUNTERS];
1454 struct bch_sb_field_quota {
1455 struct bch_sb_field field;
1456 struct bch_sb_quota_type q[QTYP_NR];
1457 } __packed __aligned(8);
1459 /* BCH_SB_FIELD_disk_groups: */
1461 #define BCH_SB_LABEL_SIZE 32
1463 struct bch_disk_group {
1464 __u8 label[BCH_SB_LABEL_SIZE];
1466 } __packed __aligned(8);
1468 LE64_BITMASK(BCH_GROUP_DELETED, struct bch_disk_group, flags[0], 0, 1)
1469 LE64_BITMASK(BCH_GROUP_DATA_ALLOWED, struct bch_disk_group, flags[0], 1, 6)
1470 LE64_BITMASK(BCH_GROUP_PARENT, struct bch_disk_group, flags[0], 6, 24)
1472 struct bch_sb_field_disk_groups {
1473 struct bch_sb_field field;
1474 struct bch_disk_group entries[];
1475 } __packed __aligned(8);
1477 /* BCH_SB_FIELD_counters */
1479 #define BCH_PERSISTENT_COUNTERS() \
1483 x(bucket_invalidate, 3) \
1484 x(bucket_discard, 4) \
1485 x(bucket_alloc, 5) \
1486 x(bucket_alloc_fail, 6) \
1487 x(btree_cache_scan, 7) \
1488 x(btree_cache_reap, 8) \
1489 x(btree_cache_cannibalize, 9) \
1490 x(btree_cache_cannibalize_lock, 10) \
1491 x(btree_cache_cannibalize_lock_fail, 11) \
1492 x(btree_cache_cannibalize_unlock, 12) \
1493 x(btree_node_write, 13) \
1494 x(btree_node_read, 14) \
1495 x(btree_node_compact, 15) \
1496 x(btree_node_merge, 16) \
1497 x(btree_node_split, 17) \
1498 x(btree_node_rewrite, 18) \
1499 x(btree_node_alloc, 19) \
1500 x(btree_node_free, 20) \
1501 x(btree_node_set_root, 21) \
1502 x(btree_path_relock_fail, 22) \
1503 x(btree_path_upgrade_fail, 23) \
1504 x(btree_reserve_get_fail, 24) \
1505 x(journal_entry_full, 25) \
1506 x(journal_full, 26) \
1507 x(journal_reclaim_finish, 27) \
1508 x(journal_reclaim_start, 28) \
1509 x(journal_write, 29) \
1510 x(read_promote, 30) \
1511 x(read_bounce, 31) \
1514 x(read_reuse_race, 34) \
1515 x(move_extent_read, 35) \
1516 x(move_extent_write, 36) \
1517 x(move_extent_finish, 37) \
1518 x(move_extent_fail, 38) \
1519 x(move_extent_alloc_mem_fail, 39) \
1521 x(copygc_wait, 41) \
1522 x(gc_gens_end, 42) \
1523 x(gc_gens_start, 43) \
1524 x(trans_blocked_journal_reclaim, 44) \
1525 x(trans_restart_btree_node_reused, 45) \
1526 x(trans_restart_btree_node_split, 46) \
1527 x(trans_restart_fault_inject, 47) \
1528 x(trans_restart_iter_upgrade, 48) \
1529 x(trans_restart_journal_preres_get, 49) \
1530 x(trans_restart_journal_reclaim, 50) \
1531 x(trans_restart_journal_res_get, 51) \
1532 x(trans_restart_key_cache_key_realloced, 52) \
1533 x(trans_restart_key_cache_raced, 53) \
1534 x(trans_restart_mark_replicas, 54) \
1535 x(trans_restart_mem_realloced, 55) \
1536 x(trans_restart_memory_allocation_failure, 56) \
1537 x(trans_restart_relock, 57) \
1538 x(trans_restart_relock_after_fill, 58) \
1539 x(trans_restart_relock_key_cache_fill, 59) \
1540 x(trans_restart_relock_next_node, 60) \
1541 x(trans_restart_relock_parent_for_fill, 61) \
1542 x(trans_restart_relock_path, 62) \
1543 x(trans_restart_relock_path_intent, 63) \
1544 x(trans_restart_too_many_iters, 64) \
1545 x(trans_restart_traverse, 65) \
1546 x(trans_restart_upgrade, 66) \
1547 x(trans_restart_would_deadlock, 67) \
1548 x(trans_restart_would_deadlock_write, 68) \
1549 x(trans_restart_injected, 69) \
1550 x(trans_restart_key_cache_upgrade, 70) \
1551 x(trans_traverse_all, 71) \
1552 x(transaction_commit, 72) \
1553 x(write_super, 73) \
1554 x(trans_restart_would_deadlock_recursion_limit, 74) \
1555 x(trans_restart_write_buffer_flush, 75) \
1556 x(trans_restart_split_race, 76)
1558 enum bch_persistent_counters {
1559 #define x(t, n, ...) BCH_COUNTER_##t,
1560 BCH_PERSISTENT_COUNTERS()
1565 struct bch_sb_field_counters {
1566 struct bch_sb_field field;
1571 * On clean shutdown, store btree roots and current journal sequence number in
1578 __u8 type; /* designates what this jset holds */
1581 struct bkey_i start[0];
1585 struct bch_sb_field_clean {
1586 struct bch_sb_field field;
1589 __le16 _read_clock; /* no longer used */
1590 __le16 _write_clock;
1593 struct jset_entry start[0];
1597 struct journal_seq_blacklist_entry {
1602 struct bch_sb_field_journal_seq_blacklist {
1603 struct bch_sb_field field;
1605 struct journal_seq_blacklist_entry start[0];
1612 * New versioning scheme:
1613 * One common version number for all on disk data structures - superblock, btree
1614 * nodes, journal entries
1616 #define BCH_VERSION_MAJOR(_v) ((__u16) ((_v) >> 10))
1617 #define BCH_VERSION_MINOR(_v) ((__u16) ((_v) & ~(~0U << 10)))
1618 #define BCH_VERSION(_major, _minor) (((_major) << 10)|(_minor) << 0)
1620 #define RECOVERY_PASS_ALL_FSCK (1ULL << 63)
1622 #define BCH_METADATA_VERSIONS() \
1623 x(bkey_renumber, BCH_VERSION(0, 10), \
1624 RECOVERY_PASS_ALL_FSCK) \
1625 x(inode_btree_change, BCH_VERSION(0, 11), \
1626 RECOVERY_PASS_ALL_FSCK) \
1627 x(snapshot, BCH_VERSION(0, 12), \
1628 RECOVERY_PASS_ALL_FSCK) \
1629 x(inode_backpointers, BCH_VERSION(0, 13), \
1630 RECOVERY_PASS_ALL_FSCK) \
1631 x(btree_ptr_sectors_written, BCH_VERSION(0, 14), \
1632 RECOVERY_PASS_ALL_FSCK) \
1633 x(snapshot_2, BCH_VERSION(0, 15), \
1634 BIT_ULL(BCH_RECOVERY_PASS_fs_upgrade_for_subvolumes)| \
1635 BIT_ULL(BCH_RECOVERY_PASS_initialize_subvolumes)| \
1636 RECOVERY_PASS_ALL_FSCK) \
1637 x(reflink_p_fix, BCH_VERSION(0, 16), \
1638 BIT_ULL(BCH_RECOVERY_PASS_fix_reflink_p)) \
1639 x(subvol_dirent, BCH_VERSION(0, 17), \
1640 RECOVERY_PASS_ALL_FSCK) \
1641 x(inode_v2, BCH_VERSION(0, 18), \
1642 RECOVERY_PASS_ALL_FSCK) \
1643 x(freespace, BCH_VERSION(0, 19), \
1644 RECOVERY_PASS_ALL_FSCK) \
1645 x(alloc_v4, BCH_VERSION(0, 20), \
1646 RECOVERY_PASS_ALL_FSCK) \
1647 x(new_data_types, BCH_VERSION(0, 21), \
1648 RECOVERY_PASS_ALL_FSCK) \
1649 x(backpointers, BCH_VERSION(0, 22), \
1650 RECOVERY_PASS_ALL_FSCK) \
1651 x(inode_v3, BCH_VERSION(0, 23), \
1652 RECOVERY_PASS_ALL_FSCK) \
1653 x(unwritten_extents, BCH_VERSION(0, 24), \
1654 RECOVERY_PASS_ALL_FSCK) \
1655 x(bucket_gens, BCH_VERSION(0, 25), \
1656 BIT_ULL(BCH_RECOVERY_PASS_bucket_gens_init)| \
1657 RECOVERY_PASS_ALL_FSCK) \
1658 x(lru_v2, BCH_VERSION(0, 26), \
1659 RECOVERY_PASS_ALL_FSCK) \
1660 x(fragmentation_lru, BCH_VERSION(0, 27), \
1661 RECOVERY_PASS_ALL_FSCK) \
1662 x(no_bps_in_alloc_keys, BCH_VERSION(0, 28), \
1663 RECOVERY_PASS_ALL_FSCK) \
1664 x(snapshot_trees, BCH_VERSION(0, 29), \
1665 RECOVERY_PASS_ALL_FSCK) \
1666 x(major_minor, BCH_VERSION(1, 0), \
1668 x(snapshot_skiplists, BCH_VERSION(1, 1), \
1669 BIT_ULL(BCH_RECOVERY_PASS_check_snapshots)) \
1670 x(deleted_inodes, BCH_VERSION(1, 2), \
1671 BIT_ULL(BCH_RECOVERY_PASS_check_inodes)) \
1672 x(rebalance_work, BCH_VERSION(1, 3), \
1673 BIT_ULL(BCH_RECOVERY_PASS_set_fs_needs_rebalance))
1675 enum bcachefs_metadata_version {
1676 bcachefs_metadata_version_min = 9,
1677 #define x(t, n, upgrade_passes) bcachefs_metadata_version_##t = n,
1678 BCH_METADATA_VERSIONS()
1680 bcachefs_metadata_version_max
1683 static const __maybe_unused
1684 unsigned bcachefs_metadata_required_upgrade_below = bcachefs_metadata_version_rebalance_work;
1686 #define bcachefs_metadata_version_current (bcachefs_metadata_version_max - 1)
1688 #define BCH_SB_SECTOR 8
1689 #define BCH_SB_MEMBERS_MAX 64 /* XXX kill */
1691 struct bch_sb_layout {
1692 __uuid_t magic; /* bcachefs superblock UUID */
1694 __u8 sb_max_size_bits; /* base 2 of 512 byte sectors */
1695 __u8 nr_superblocks;
1697 __le64 sb_offset[61];
1698 } __packed __aligned(8);
1700 #define BCH_SB_LAYOUT_SECTOR 7
1703 * @offset - sector where this sb was written
1704 * @version - on disk format version
1705 * @version_min - Oldest metadata version this filesystem contains; so we can
1706 * safely drop compatibility code and refuse to mount filesystems
1708 * @magic - identifies as a bcachefs superblock (BCHFS_MAGIC)
1709 * @seq - incremented each time superblock is written
1710 * @uuid - used for generating various magic numbers and identifying
1711 * member devices, never changes
1712 * @user_uuid - user visible UUID, may be changed
1713 * @label - filesystem label
1714 * @seq - identifies most recent superblock, incremented each time
1715 * superblock is written
1716 * @features - enabled incompatible features
1719 struct bch_csum csum;
1726 __u8 label[BCH_SB_LABEL_SIZE];
1735 __le64 time_base_lo;
1736 __le32 time_base_hi;
1737 __le32 time_precision;
1743 struct bch_sb_layout layout;
1745 struct bch_sb_field start[0];
1747 } __packed __aligned(8);
1751 * BCH_SB_INITALIZED - set on first mount
1752 * BCH_SB_CLEAN - did we shut down cleanly? Just a hint, doesn't affect
1753 * behaviour of mount/recovery path:
1754 * BCH_SB_INODE_32BIT - limit inode numbers to 32 bits
1755 * BCH_SB_128_BIT_MACS - 128 bit macs instead of 80
1756 * BCH_SB_ENCRYPTION_TYPE - if nonzero encryption is enabled; overrides
1757 * DATA/META_CSUM_TYPE. Also indicates encryption
1758 * algorithm in use, if/when we get more than one
1761 LE16_BITMASK(BCH_SB_BLOCK_SIZE, struct bch_sb, block_size, 0, 16);
1763 LE64_BITMASK(BCH_SB_INITIALIZED, struct bch_sb, flags[0], 0, 1);
1764 LE64_BITMASK(BCH_SB_CLEAN, struct bch_sb, flags[0], 1, 2);
1765 LE64_BITMASK(BCH_SB_CSUM_TYPE, struct bch_sb, flags[0], 2, 8);
1766 LE64_BITMASK(BCH_SB_ERROR_ACTION, struct bch_sb, flags[0], 8, 12);
1768 LE64_BITMASK(BCH_SB_BTREE_NODE_SIZE, struct bch_sb, flags[0], 12, 28);
1770 LE64_BITMASK(BCH_SB_GC_RESERVE, struct bch_sb, flags[0], 28, 33);
1771 LE64_BITMASK(BCH_SB_ROOT_RESERVE, struct bch_sb, flags[0], 33, 40);
1773 LE64_BITMASK(BCH_SB_META_CSUM_TYPE, struct bch_sb, flags[0], 40, 44);
1774 LE64_BITMASK(BCH_SB_DATA_CSUM_TYPE, struct bch_sb, flags[0], 44, 48);
1776 LE64_BITMASK(BCH_SB_META_REPLICAS_WANT, struct bch_sb, flags[0], 48, 52);
1777 LE64_BITMASK(BCH_SB_DATA_REPLICAS_WANT, struct bch_sb, flags[0], 52, 56);
1779 LE64_BITMASK(BCH_SB_POSIX_ACL, struct bch_sb, flags[0], 56, 57);
1780 LE64_BITMASK(BCH_SB_USRQUOTA, struct bch_sb, flags[0], 57, 58);
1781 LE64_BITMASK(BCH_SB_GRPQUOTA, struct bch_sb, flags[0], 58, 59);
1782 LE64_BITMASK(BCH_SB_PRJQUOTA, struct bch_sb, flags[0], 59, 60);
1784 LE64_BITMASK(BCH_SB_HAS_ERRORS, struct bch_sb, flags[0], 60, 61);
1785 LE64_BITMASK(BCH_SB_HAS_TOPOLOGY_ERRORS,struct bch_sb, flags[0], 61, 62);
1787 LE64_BITMASK(BCH_SB_BIG_ENDIAN, struct bch_sb, flags[0], 62, 63);
1789 LE64_BITMASK(BCH_SB_STR_HASH_TYPE, struct bch_sb, flags[1], 0, 4);
1790 LE64_BITMASK(BCH_SB_COMPRESSION_TYPE_LO,struct bch_sb, flags[1], 4, 8);
1791 LE64_BITMASK(BCH_SB_INODE_32BIT, struct bch_sb, flags[1], 8, 9);
1793 LE64_BITMASK(BCH_SB_128_BIT_MACS, struct bch_sb, flags[1], 9, 10);
1794 LE64_BITMASK(BCH_SB_ENCRYPTION_TYPE, struct bch_sb, flags[1], 10, 14);
1797 * Max size of an extent that may require bouncing to read or write
1798 * (checksummed, compressed): 64k
1800 LE64_BITMASK(BCH_SB_ENCODED_EXTENT_MAX_BITS,
1801 struct bch_sb, flags[1], 14, 20);
1803 LE64_BITMASK(BCH_SB_META_REPLICAS_REQ, struct bch_sb, flags[1], 20, 24);
1804 LE64_BITMASK(BCH_SB_DATA_REPLICAS_REQ, struct bch_sb, flags[1], 24, 28);
1806 LE64_BITMASK(BCH_SB_PROMOTE_TARGET, struct bch_sb, flags[1], 28, 40);
1807 LE64_BITMASK(BCH_SB_FOREGROUND_TARGET, struct bch_sb, flags[1], 40, 52);
1808 LE64_BITMASK(BCH_SB_BACKGROUND_TARGET, struct bch_sb, flags[1], 52, 64);
1810 LE64_BITMASK(BCH_SB_BACKGROUND_COMPRESSION_TYPE_LO,
1811 struct bch_sb, flags[2], 0, 4);
1812 LE64_BITMASK(BCH_SB_GC_RESERVE_BYTES, struct bch_sb, flags[2], 4, 64);
1814 LE64_BITMASK(BCH_SB_ERASURE_CODE, struct bch_sb, flags[3], 0, 16);
1815 LE64_BITMASK(BCH_SB_METADATA_TARGET, struct bch_sb, flags[3], 16, 28);
1816 LE64_BITMASK(BCH_SB_SHARD_INUMS, struct bch_sb, flags[3], 28, 29);
1817 LE64_BITMASK(BCH_SB_INODES_USE_KEY_CACHE,struct bch_sb, flags[3], 29, 30);
1818 LE64_BITMASK(BCH_SB_JOURNAL_FLUSH_DELAY,struct bch_sb, flags[3], 30, 62);
1819 LE64_BITMASK(BCH_SB_JOURNAL_FLUSH_DISABLED,struct bch_sb, flags[3], 62, 63);
1820 LE64_BITMASK(BCH_SB_JOURNAL_RECLAIM_DELAY,struct bch_sb, flags[4], 0, 32);
1821 LE64_BITMASK(BCH_SB_JOURNAL_TRANSACTION_NAMES,struct bch_sb, flags[4], 32, 33);
1822 LE64_BITMASK(BCH_SB_NOCOW, struct bch_sb, flags[4], 33, 34);
1823 LE64_BITMASK(BCH_SB_WRITE_BUFFER_SIZE, struct bch_sb, flags[4], 34, 54);
1824 LE64_BITMASK(BCH_SB_VERSION_UPGRADE, struct bch_sb, flags[4], 54, 56);
1826 LE64_BITMASK(BCH_SB_COMPRESSION_TYPE_HI,struct bch_sb, flags[4], 56, 60);
1827 LE64_BITMASK(BCH_SB_BACKGROUND_COMPRESSION_TYPE_HI,
1828 struct bch_sb, flags[4], 60, 64);
1830 LE64_BITMASK(BCH_SB_VERSION_UPGRADE_COMPLETE,
1831 struct bch_sb, flags[5], 0, 16);
1833 static inline __u64 BCH_SB_COMPRESSION_TYPE(const struct bch_sb *sb)
1835 return BCH_SB_COMPRESSION_TYPE_LO(sb) | (BCH_SB_COMPRESSION_TYPE_HI(sb) << 4);
1838 static inline void SET_BCH_SB_COMPRESSION_TYPE(struct bch_sb *sb, __u64 v)
1840 SET_BCH_SB_COMPRESSION_TYPE_LO(sb, v);
1841 SET_BCH_SB_COMPRESSION_TYPE_HI(sb, v >> 4);
1844 static inline __u64 BCH_SB_BACKGROUND_COMPRESSION_TYPE(const struct bch_sb *sb)
1846 return BCH_SB_BACKGROUND_COMPRESSION_TYPE_LO(sb) |
1847 (BCH_SB_BACKGROUND_COMPRESSION_TYPE_HI(sb) << 4);
1850 static inline void SET_BCH_SB_BACKGROUND_COMPRESSION_TYPE(struct bch_sb *sb, __u64 v)
1852 SET_BCH_SB_BACKGROUND_COMPRESSION_TYPE_LO(sb, v);
1853 SET_BCH_SB_BACKGROUND_COMPRESSION_TYPE_HI(sb, v >> 4);
1859 * journal_seq_blacklist_v3: gates BCH_SB_FIELD_journal_seq_blacklist
1860 * reflink: gates KEY_TYPE_reflink
1861 * inline_data: gates KEY_TYPE_inline_data
1862 * new_siphash: gates BCH_STR_HASH_siphash
1863 * new_extent_overwrite: gates BTREE_NODE_NEW_EXTENT_OVERWRITE
1865 #define BCH_SB_FEATURES() \
1869 x(atomic_nlink, 3) \
1871 x(journal_seq_blacklist_v3, 5) \
1875 x(new_extent_overwrite, 9) \
1876 x(incompressible, 10) \
1877 x(btree_ptr_v2, 11) \
1878 x(extents_above_btree_updates, 12) \
1879 x(btree_updates_journalled, 13) \
1880 x(reflink_inline_data, 14) \
1882 x(journal_no_flush, 16) \
1884 x(extents_across_btree_nodes, 18)
1886 #define BCH_SB_FEATURES_ALWAYS \
1887 ((1ULL << BCH_FEATURE_new_extent_overwrite)| \
1888 (1ULL << BCH_FEATURE_extents_above_btree_updates)|\
1889 (1ULL << BCH_FEATURE_btree_updates_journalled)|\
1890 (1ULL << BCH_FEATURE_alloc_v2)|\
1891 (1ULL << BCH_FEATURE_extents_across_btree_nodes))
1893 #define BCH_SB_FEATURES_ALL \
1894 (BCH_SB_FEATURES_ALWAYS| \
1895 (1ULL << BCH_FEATURE_new_siphash)| \
1896 (1ULL << BCH_FEATURE_btree_ptr_v2)| \
1897 (1ULL << BCH_FEATURE_new_varint)| \
1898 (1ULL << BCH_FEATURE_journal_no_flush))
1900 enum bch_sb_feature {
1901 #define x(f, n) BCH_FEATURE_##f,
1907 #define BCH_SB_COMPAT() \
1909 x(alloc_metadata, 1) \
1910 x(extents_above_btree_updates_done, 2) \
1911 x(bformat_overflow_done, 3)
1913 enum bch_sb_compat {
1914 #define x(f, n) BCH_COMPAT_##f,
1922 #define BCH_VERSION_UPGRADE_OPTS() \
1924 x(incompatible, 1) \
1927 enum bch_version_upgrade_opts {
1928 #define x(t, n) BCH_VERSION_UPGRADE_##t = n,
1929 BCH_VERSION_UPGRADE_OPTS()
1933 #define BCH_REPLICAS_MAX 4U
1935 #define BCH_BKEY_PTRS_MAX 16U
1937 #define BCH_ERROR_ACTIONS() \
1942 enum bch_error_actions {
1943 #define x(t, n) BCH_ON_ERROR_##t = n,
1949 #define BCH_STR_HASH_TYPES() \
1955 enum bch_str_hash_type {
1956 #define x(t, n) BCH_STR_HASH_##t = n,
1957 BCH_STR_HASH_TYPES()
1962 #define BCH_STR_HASH_OPTS() \
1967 enum bch_str_hash_opts {
1968 #define x(t, n) BCH_STR_HASH_OPT_##t = n,
1974 #define BCH_CSUM_TYPES() \
1976 x(crc32c_nonzero, 1) \
1977 x(crc64_nonzero, 2) \
1978 x(chacha20_poly1305_80, 3) \
1979 x(chacha20_poly1305_128, 4) \
1984 enum bch_csum_type {
1985 #define x(t, n) BCH_CSUM_##t = n,
1991 static const __maybe_unused unsigned bch_crc_bytes[] = {
1992 [BCH_CSUM_none] = 0,
1993 [BCH_CSUM_crc32c_nonzero] = 4,
1994 [BCH_CSUM_crc32c] = 4,
1995 [BCH_CSUM_crc64_nonzero] = 8,
1996 [BCH_CSUM_crc64] = 8,
1997 [BCH_CSUM_xxhash] = 8,
1998 [BCH_CSUM_chacha20_poly1305_80] = 10,
1999 [BCH_CSUM_chacha20_poly1305_128] = 16,
2002 static inline _Bool bch2_csum_type_is_encryption(enum bch_csum_type type)
2005 case BCH_CSUM_chacha20_poly1305_80:
2006 case BCH_CSUM_chacha20_poly1305_128:
2013 #define BCH_CSUM_OPTS() \
2019 enum bch_csum_opts {
2020 #define x(t, n) BCH_CSUM_OPT_##t = n,
2026 #define BCH_COMPRESSION_TYPES() \
2032 x(incompressible, 5)
2034 enum bch_compression_type {
2035 #define x(t, n) BCH_COMPRESSION_TYPE_##t = n,
2036 BCH_COMPRESSION_TYPES()
2038 BCH_COMPRESSION_TYPE_NR
2041 #define BCH_COMPRESSION_OPTS() \
2047 enum bch_compression_opts {
2048 #define x(t, n) BCH_COMPRESSION_OPT_##t = n,
2049 BCH_COMPRESSION_OPTS()
2051 BCH_COMPRESSION_OPT_NR
2057 * The various other data structures have their own magic numbers, which are
2058 * xored with the first part of the cache set's UUID
2061 #define BCACHE_MAGIC \
2062 UUID_INIT(0xc68573f6, 0x4e1a, 0x45ca, \
2063 0x82, 0x65, 0xf5, 0x7f, 0x48, 0xba, 0x6d, 0x81)
2064 #define BCHFS_MAGIC \
2065 UUID_INIT(0xc68573f6, 0x66ce, 0x90a9, \
2066 0xd9, 0x6a, 0x60, 0xcf, 0x80, 0x3d, 0xf7, 0xef)
2068 #define BCACHEFS_STATFS_MAGIC 0xca451a4e
2070 #define JSET_MAGIC __cpu_to_le64(0x245235c1a3625032ULL)
2071 #define BSET_MAGIC __cpu_to_le64(0x90135c78b99e07f5ULL)
2073 static inline __le64 __bch2_sb_magic(struct bch_sb *sb)
2077 memcpy(&ret, &sb->uuid, sizeof(ret));
2081 static inline __u64 __jset_magic(struct bch_sb *sb)
2083 return __le64_to_cpu(__bch2_sb_magic(sb) ^ JSET_MAGIC);
2086 static inline __u64 __bset_magic(struct bch_sb *sb)
2088 return __le64_to_cpu(__bch2_sb_magic(sb) ^ BSET_MAGIC);
2093 #define JSET_KEYS_U64s (sizeof(struct jset_entry) / sizeof(__u64))
2095 #define BCH_JSET_ENTRY_TYPES() \
2100 x(blacklist_v2, 4) \
2109 #define x(f, nr) BCH_JSET_ENTRY_##f = nr,
2110 BCH_JSET_ENTRY_TYPES()
2116 * Journal sequence numbers can be blacklisted: bsets record the max sequence
2117 * number of all the journal entries they contain updates for, so that on
2118 * recovery we can ignore those bsets that contain index updates newer that what
2119 * made it into the journal.
2121 * This means that we can't reuse that journal_seq - we have to skip it, and
2122 * then record that we skipped it so that the next time we crash and recover we
2123 * don't think there was a missing journal entry.
2125 struct jset_entry_blacklist {
2126 struct jset_entry entry;
2130 struct jset_entry_blacklist_v2 {
2131 struct jset_entry entry;
2136 #define BCH_FS_USAGE_TYPES() \
2142 #define x(f, nr) BCH_FS_USAGE_##f = nr,
2143 BCH_FS_USAGE_TYPES()
2148 struct jset_entry_usage {
2149 struct jset_entry entry;
2153 struct jset_entry_data_usage {
2154 struct jset_entry entry;
2156 struct bch_replicas_entry r;
2159 struct jset_entry_clock {
2160 struct jset_entry entry;
2166 struct jset_entry_dev_usage_type {
2172 struct jset_entry_dev_usage {
2173 struct jset_entry entry;
2178 __le64 _buckets_unavailable; /* No longer used */
2180 struct jset_entry_dev_usage_type d[];
2183 static inline unsigned jset_entry_dev_usage_nr_types(struct jset_entry_dev_usage *u)
2185 return (vstruct_bytes(&u->entry) - sizeof(struct jset_entry_dev_usage)) /
2186 sizeof(struct jset_entry_dev_usage_type);
2189 struct jset_entry_log {
2190 struct jset_entry entry;
2195 * On disk format for a journal entry:
2196 * seq is monotonically increasing; every journal entry has its own unique
2199 * last_seq is the oldest journal entry that still has keys the btree hasn't
2200 * flushed to disk yet.
2202 * version is for on disk format changes.
2205 struct bch_csum csum;
2212 __le32 u64s; /* size of d[] in u64s */
2214 __u8 encrypted_start[0];
2216 __le16 _read_clock; /* no longer used */
2217 __le16 _write_clock;
2219 /* Sequence number of oldest dirty journal entry */
2223 struct jset_entry start[0];
2225 } __packed __aligned(8);
2227 LE32_BITMASK(JSET_CSUM_TYPE, struct jset, flags, 0, 4);
2228 LE32_BITMASK(JSET_BIG_ENDIAN, struct jset, flags, 4, 5);
2229 LE32_BITMASK(JSET_NO_FLUSH, struct jset, flags, 5, 6);
2231 #define BCH_JOURNAL_BUCKETS_MIN 8
2235 enum btree_id_flags {
2236 BTREE_ID_EXTENTS = BIT(0),
2237 BTREE_ID_SNAPSHOTS = BIT(1),
2238 BTREE_ID_DATA = BIT(2),
2241 #define BCH_BTREE_IDS() \
2242 x(extents, 0, BTREE_ID_EXTENTS|BTREE_ID_SNAPSHOTS|BTREE_ID_DATA,\
2243 BIT_ULL(KEY_TYPE_whiteout)| \
2244 BIT_ULL(KEY_TYPE_error)| \
2245 BIT_ULL(KEY_TYPE_cookie)| \
2246 BIT_ULL(KEY_TYPE_extent)| \
2247 BIT_ULL(KEY_TYPE_reservation)| \
2248 BIT_ULL(KEY_TYPE_reflink_p)| \
2249 BIT_ULL(KEY_TYPE_inline_data)) \
2250 x(inodes, 1, BTREE_ID_SNAPSHOTS, \
2251 BIT_ULL(KEY_TYPE_whiteout)| \
2252 BIT_ULL(KEY_TYPE_inode)| \
2253 BIT_ULL(KEY_TYPE_inode_v2)| \
2254 BIT_ULL(KEY_TYPE_inode_v3)| \
2255 BIT_ULL(KEY_TYPE_inode_generation)) \
2256 x(dirents, 2, BTREE_ID_SNAPSHOTS, \
2257 BIT_ULL(KEY_TYPE_whiteout)| \
2258 BIT_ULL(KEY_TYPE_hash_whiteout)| \
2259 BIT_ULL(KEY_TYPE_dirent)) \
2260 x(xattrs, 3, BTREE_ID_SNAPSHOTS, \
2261 BIT_ULL(KEY_TYPE_whiteout)| \
2262 BIT_ULL(KEY_TYPE_cookie)| \
2263 BIT_ULL(KEY_TYPE_hash_whiteout)| \
2264 BIT_ULL(KEY_TYPE_xattr)) \
2266 BIT_ULL(KEY_TYPE_alloc)| \
2267 BIT_ULL(KEY_TYPE_alloc_v2)| \
2268 BIT_ULL(KEY_TYPE_alloc_v3)| \
2269 BIT_ULL(KEY_TYPE_alloc_v4)) \
2271 BIT_ULL(KEY_TYPE_quota)) \
2273 BIT_ULL(KEY_TYPE_stripe)) \
2274 x(reflink, 7, BTREE_ID_EXTENTS|BTREE_ID_DATA, \
2275 BIT_ULL(KEY_TYPE_reflink_v)| \
2276 BIT_ULL(KEY_TYPE_indirect_inline_data)) \
2277 x(subvolumes, 8, 0, \
2278 BIT_ULL(KEY_TYPE_subvolume)) \
2279 x(snapshots, 9, 0, \
2280 BIT_ULL(KEY_TYPE_snapshot)) \
2282 BIT_ULL(KEY_TYPE_set)) \
2283 x(freespace, 11, BTREE_ID_EXTENTS, \
2284 BIT_ULL(KEY_TYPE_set)) \
2285 x(need_discard, 12, 0, \
2286 BIT_ULL(KEY_TYPE_set)) \
2287 x(backpointers, 13, 0, \
2288 BIT_ULL(KEY_TYPE_backpointer)) \
2289 x(bucket_gens, 14, 0, \
2290 BIT_ULL(KEY_TYPE_bucket_gens)) \
2291 x(snapshot_trees, 15, 0, \
2292 BIT_ULL(KEY_TYPE_snapshot_tree)) \
2293 x(deleted_inodes, 16, BTREE_ID_SNAPSHOTS, \
2294 BIT_ULL(KEY_TYPE_set)) \
2295 x(logged_ops, 17, 0, \
2296 BIT_ULL(KEY_TYPE_logged_op_truncate)| \
2297 BIT_ULL(KEY_TYPE_logged_op_finsert)) \
2298 x(rebalance_work, 18, BTREE_ID_SNAPSHOTS, \
2299 BIT_ULL(KEY_TYPE_set)|BIT_ULL(KEY_TYPE_cookie))
2302 #define x(name, nr, ...) BTREE_ID_##name = nr,
2308 #define BTREE_MAX_DEPTH 4U
2315 * On disk a btree node is a list/log of these; within each set the keys are
2322 * Highest journal entry this bset contains keys for.
2323 * If on recovery we don't see that journal entry, this bset is ignored:
2324 * this allows us to preserve the order of all index updates after a
2325 * crash, since the journal records a total order of all index updates
2326 * and anything that didn't make it to the journal doesn't get used.
2332 __le16 u64s; /* count of d[] in u64s */
2334 struct bkey_packed start[0];
2336 } __packed __aligned(8);
2338 LE32_BITMASK(BSET_CSUM_TYPE, struct bset, flags, 0, 4);
2340 LE32_BITMASK(BSET_BIG_ENDIAN, struct bset, flags, 4, 5);
2341 LE32_BITMASK(BSET_SEPARATE_WHITEOUTS,
2342 struct bset, flags, 5, 6);
2344 /* Sector offset within the btree node: */
2345 LE32_BITMASK(BSET_OFFSET, struct bset, flags, 16, 32);
2348 struct bch_csum csum;
2351 /* this flags field is encrypted, unlike bset->flags: */
2354 /* Closed interval: */
2355 struct bpos min_key;
2356 struct bpos max_key;
2357 struct bch_extent_ptr _ptr; /* not used anymore */
2358 struct bkey_format format;
2369 } __packed __aligned(8);
2371 LE64_BITMASK(BTREE_NODE_ID_LO, struct btree_node, flags, 0, 4);
2372 LE64_BITMASK(BTREE_NODE_LEVEL, struct btree_node, flags, 4, 8);
2373 LE64_BITMASK(BTREE_NODE_NEW_EXTENT_OVERWRITE,
2374 struct btree_node, flags, 8, 9);
2375 LE64_BITMASK(BTREE_NODE_ID_HI, struct btree_node, flags, 9, 25);
2377 LE64_BITMASK(BTREE_NODE_SEQ, struct btree_node, flags, 32, 64);
2379 static inline __u64 BTREE_NODE_ID(struct btree_node *n)
2381 return BTREE_NODE_ID_LO(n) | (BTREE_NODE_ID_HI(n) << 4);
2384 static inline void SET_BTREE_NODE_ID(struct btree_node *n, __u64 v)
2386 SET_BTREE_NODE_ID_LO(n, v);
2387 SET_BTREE_NODE_ID_HI(n, v >> 4);
2390 struct btree_node_entry {
2391 struct bch_csum csum;
2401 } __packed __aligned(8);
2403 #endif /* _BCACHEFS_FORMAT_H */