1 /* SPDX-License-Identifier: GPL-2.0 */
2 #ifndef _BCACHEFS_FORMAT_H
3 #define _BCACHEFS_FORMAT_H
6 * bcachefs on disk data structures
10 * There are three main types of on disk data structures in bcachefs (this is
11 * reduced from 5 in bcache)
17 * The btree is the primary structure; most metadata exists as keys in the
18 * various btrees. There are only a small number of btrees, they're not
19 * sharded - we have one btree for extents, another for inodes, et cetera.
23 * The superblock contains the location of the journal, the list of devices in
24 * the filesystem, and in general any metadata we need in order to decide
25 * whether we can start a filesystem or prior to reading the journal/btree
28 * The superblock is extensible, and most of the contents of the superblock are
29 * in variable length, type tagged fields; see struct bch_sb_field.
31 * Backup superblocks do not reside in a fixed location; also, superblocks do
32 * not have a fixed size. To locate backup superblocks we have struct
33 * bch_sb_layout; we store a copy of this inside every superblock, and also
34 * before the first superblock.
38 * The journal primarily records btree updates in the order they occurred;
39 * journal replay consists of just iterating over all the keys in the open
40 * journal entries and re-inserting them into the btrees.
42 * The journal also contains entry types for the btree roots, and blacklisted
43 * journal sequence numbers (see journal_seq_blacklist.c).
47 * bcachefs btrees are copy on write b+ trees, where nodes are big (typically
48 * 128k-256k) and log structured. We use struct btree_node for writing the first
49 * entry in a given node (offset 0), and struct btree_node_entry for all
52 * After the header, btree node entries contain a list of keys in sorted order.
53 * Values are stored inline with the keys; since values are variable length (and
54 * keys effectively are variable length too, due to packing) we can't do random
55 * access without building up additional in memory tables in the btree node read
58 * BTREE KEYS (struct bkey):
60 * The various btrees share a common format for the key - so as to avoid
61 * switching in fastpath lookup/comparison code - but define their own
62 * structures for the key values.
64 * The size of a key/value pair is stored as a u8 in units of u64s, so the max
65 * size is just under 2k. The common part also contains a type tag for the
66 * value, and a format field indicating whether the key is packed or not (and
67 * also meant to allow adding new key fields in the future, if desired).
69 * bkeys, when stored within a btree node, may also be packed. In that case, the
70 * bkey_format in that node is used to unpack it. Packed bkeys mean that we can
71 * be generous with field sizes in the common part of the key format (64 bit
72 * inode number, 64 bit offset, 96 bit version field, etc.) for negligible cost.
75 #include <asm/types.h>
76 #include <asm/byteorder.h>
77 #include <linux/kernel.h>
78 #include <linux/uuid.h>
82 typedef uuid_t __uuid_t;
85 #define BITMASK(name, type, field, offset, end) \
86 static const __maybe_unused unsigned name##_OFFSET = offset; \
87 static const __maybe_unused unsigned name##_BITS = (end - offset); \
89 static inline __u64 name(const type *k) \
91 return (k->field >> offset) & ~(~0ULL << (end - offset)); \
94 static inline void SET_##name(type *k, __u64 v) \
96 k->field &= ~(~(~0ULL << (end - offset)) << offset); \
97 k->field |= (v & ~(~0ULL << (end - offset))) << offset; \
100 #define LE_BITMASK(_bits, name, type, field, offset, end) \
101 static const __maybe_unused unsigned name##_OFFSET = offset; \
102 static const __maybe_unused unsigned name##_BITS = (end - offset); \
103 static const __maybe_unused __u##_bits name##_MAX = (1ULL << (end - offset)) - 1;\
105 static inline __u64 name(const type *k) \
107 return (__le##_bits##_to_cpu(k->field) >> offset) & \
108 ~(~0ULL << (end - offset)); \
111 static inline void SET_##name(type *k, __u64 v) \
113 __u##_bits new = __le##_bits##_to_cpu(k->field); \
115 new &= ~(~(~0ULL << (end - offset)) << offset); \
116 new |= (v & ~(~0ULL << (end - offset))) << offset; \
117 k->field = __cpu_to_le##_bits(new); \
120 #define LE16_BITMASK(n, t, f, o, e) LE_BITMASK(16, n, t, f, o, e)
121 #define LE32_BITMASK(n, t, f, o, e) LE_BITMASK(32, n, t, f, o, e)
122 #define LE64_BITMASK(n, t, f, o, e) LE_BITMASK(64, n, t, f, o, e)
127 /* One unused slot for now: */
128 __u8 bits_per_field[6];
129 __le64 field_offset[6];
132 /* Btree keys - all units are in sectors */
136 * Word order matches machine byte order - btree code treats a bpos as a
137 * single large integer, for search/comparison purposes
139 * Note that wherever a bpos is embedded in another on disk data
140 * structure, it has to be byte swabbed when reading in metadata that
141 * wasn't written in native endian order:
143 #if __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__
147 #elif __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__
149 __u64 offset; /* Points to end of extent - sectors */
152 #error edit for your odd byteorder.
155 #if __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__
160 #define KEY_INODE_MAX ((__u64)~0ULL)
161 #define KEY_OFFSET_MAX ((__u64)~0ULL)
162 #define KEY_SNAPSHOT_MAX ((__u32)~0U)
163 #define KEY_SIZE_MAX ((__u32)~0U)
165 static inline struct bpos SPOS(__u64 inode, __u64 offset, __u32 snapshot)
167 return (struct bpos) {
170 .snapshot = snapshot,
174 #define POS_MIN SPOS(0, 0, 0)
175 #define POS_MAX SPOS(KEY_INODE_MAX, KEY_OFFSET_MAX, 0)
176 #define SPOS_MAX SPOS(KEY_INODE_MAX, KEY_OFFSET_MAX, KEY_SNAPSHOT_MAX)
177 #define POS(_inode, _offset) SPOS(_inode, _offset, 0)
179 /* Empty placeholder struct, for container_of() */
185 #if __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__
188 #elif __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__
193 #if __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__
199 /* Size of combined key and value, in u64s */
202 /* Format of key (0 for format local to btree node) */
203 #if defined(__LITTLE_ENDIAN_BITFIELD)
206 #elif defined (__BIG_ENDIAN_BITFIELD)
207 __u8 needs_whiteout:1,
210 #error edit for your odd byteorder.
213 /* Type of the value */
216 #if __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__
219 struct bversion version;
220 __u32 size; /* extent size, in sectors */
222 #elif __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__
224 __u32 size; /* extent size, in sectors */
225 struct bversion version;
230 #if __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__
232 * The big-endian version of bkey can't be compiled by rustc with the "aligned"
233 * attr since it doesn't allow types to have both "packed" and "aligned" attrs.
234 * So for Rust compatibility, don't include this. It can be included in the LE
235 * version because the "packed" attr is redundant in that case.
237 * History: (quoting Kent)
239 * Specifically, when i was designing bkey, I wanted the header to be no
240 * bigger than necessary so that bkey_packed could use the rest. That means that
241 * decently offten extent keys will fit into only 8 bytes, instead of spilling over
244 * But packed_bkey treats the part after the header - the packed section -
245 * as a single multi word, variable length integer. And bkey, the unpacked
246 * version, is just a special case version of a bkey_packed; all the packed
247 * bkey code will work on keys in any packed format, the in-memory
248 * representation of an unpacked key also is just one type of packed key...
250 * So that constrains the key part of a bkig endian bkey to start right
253 * If we ever do a bkey_v2 and need to expand the hedaer by another byte for
254 * some reason - that will clean up this wart.
263 /* Size of combined key and value, in u64s */
266 /* Format of key (0 for format local to btree node) */
269 * XXX: next incompat on disk format change, switch format and
270 * needs_whiteout - bkey_packed() will be cheaper if format is the high
271 * bits of the bitfield
273 #if defined(__LITTLE_ENDIAN_BITFIELD)
276 #elif defined (__BIG_ENDIAN_BITFIELD)
277 __u8 needs_whiteout:1,
281 /* Type of the value */
286 * We copy bkeys with struct assignment in various places, and while
287 * that shouldn't be done with packed bkeys we can't disallow it in C,
288 * and it's legal to cast a bkey to a bkey_packed - so padding it out
289 * to the same size as struct bkey should hopefully be safest.
291 __u8 pad[sizeof(struct bkey) - 3];
292 } __packed __aligned(8);
299 #define BKEY_U64s (sizeof(struct bkey) / sizeof(__u64))
300 #define BKEY_U64s_MAX U8_MAX
301 #define BKEY_VAL_U64s_MAX (BKEY_U64s_MAX - BKEY_U64s)
303 #define KEY_PACKED_BITS_START 24
305 #define KEY_FORMAT_LOCAL_BTREE 0
306 #define KEY_FORMAT_CURRENT 1
308 enum bch_bkey_fields {
313 BKEY_FIELD_VERSION_HI,
314 BKEY_FIELD_VERSION_LO,
318 #define bkey_format_field(name, field) \
319 [BKEY_FIELD_##name] = (sizeof(((struct bkey *) NULL)->field) * 8)
321 #define BKEY_FORMAT_CURRENT \
322 ((struct bkey_format) { \
323 .key_u64s = BKEY_U64s, \
324 .nr_fields = BKEY_NR_FIELDS, \
325 .bits_per_field = { \
326 bkey_format_field(INODE, p.inode), \
327 bkey_format_field(OFFSET, p.offset), \
328 bkey_format_field(SNAPSHOT, p.snapshot), \
329 bkey_format_field(SIZE, size), \
330 bkey_format_field(VERSION_HI, version.hi), \
331 bkey_format_field(VERSION_LO, version.lo), \
335 /* bkey with inline value */
343 #define POS_KEY(_pos) \
346 .format = KEY_FORMAT_CURRENT, \
350 #define KEY(_inode, _offset, _size) \
353 .format = KEY_FORMAT_CURRENT, \
354 .p = POS(_inode, _offset), \
358 static inline void bkey_init(struct bkey *k)
363 #define bkey_bytes(_k) ((_k)->u64s * sizeof(__u64))
365 #define __BKEY_PADDED(key, pad) \
366 struct bkey_i key; __u64 key ## _pad[pad]
369 * - DELETED keys are used internally to mark keys that should be ignored but
370 * override keys in composition order. Their version number is ignored.
372 * - DISCARDED keys indicate that the data is all 0s because it has been
373 * discarded. DISCARDs may have a version; if the version is nonzero the key
374 * will be persistent, otherwise the key will be dropped whenever the btree
375 * node is rewritten (like DELETED keys).
377 * - ERROR: any read of the data returns a read error, as the data was lost due
378 * to a failing device. Like DISCARDED keys, they can be removed (overridden)
379 * by new writes or cluster-wide GC. Node repair can also overwrite them with
380 * the same or a more recent version number, but not with an older version
383 * - WHITEOUT: for hash table btrees
385 #define BCH_BKEY_TYPES() \
390 x(hash_whiteout, 4) \
395 x(inode_generation, 9) \
404 x(btree_ptr_v2, 18) \
405 x(indirect_inline_data, 19) \
417 x(snapshot_tree, 31) \
418 x(logged_op_truncate, 32) \
419 x(logged_op_finsert, 33)
422 #define x(name, nr) KEY_TYPE_##name = nr,
432 struct bch_whiteout {
445 struct bch_hash_whiteout {
453 /* 128 bits, sufficient for cryptographic MACs: */
457 } __packed __aligned(8);
459 struct bch_backpointer {
464 __u64 bucket_offset:40;
467 } __packed __aligned(8);
474 } __packed __aligned(8);
476 #define LRU_ID_STRIPES (1U << 16)
478 /* Optional/variable size superblock sections: */
480 struct bch_sb_field {
486 #define BCH_SB_FIELDS() \
495 x(journal_seq_blacklist, 8) \
503 #include "alloc_background_format.h"
504 #include "extents_format.h"
505 #include "reflink_format.h"
506 #include "ec_format.h"
507 #include "inode_format.h"
508 #include "dirent_format.h"
509 #include "xattr_format.h"
510 #include "quota_format.h"
511 #include "logged_ops_format.h"
512 #include "snapshot_format.h"
513 #include "subvolume_format.h"
514 #include "sb-counters_format.h"
516 enum bch_sb_field_type {
517 #define x(f, nr) BCH_SB_FIELD_##f = nr,
524 * Most superblock fields are replicated in all device's superblocks - a few are
527 #define BCH_SINGLE_DEVICE_SB_FIELDS \
528 ((1U << BCH_SB_FIELD_journal)| \
529 (1U << BCH_SB_FIELD_journal_v2))
531 /* BCH_SB_FIELD_journal: */
533 struct bch_sb_field_journal {
534 struct bch_sb_field field;
538 struct bch_sb_field_journal_v2 {
539 struct bch_sb_field field;
541 struct bch_sb_field_journal_v2_entry {
547 /* BCH_SB_FIELD_members_v1: */
549 #define BCH_MIN_NR_NBUCKETS (1 << 6)
551 #define BCH_IOPS_MEASUREMENTS() \
557 enum bch_iops_measurement {
558 #define x(t, n) BCH_IOPS_##t = n,
559 BCH_IOPS_MEASUREMENTS()
564 #define BCH_MEMBER_ERROR_TYPES() \
569 enum bch_member_error_type {
570 #define x(t, n) BCH_MEMBER_ERROR_##t = n,
571 BCH_MEMBER_ERROR_TYPES()
578 __le64 nbuckets; /* device size */
579 __le16 first_bucket; /* index of first bucket used */
580 __le16 bucket_size; /* sectors */
582 __le64 last_mount; /* time_t */
586 __le64 errors[BCH_MEMBER_ERROR_NR];
587 __le64 errors_at_reset[BCH_MEMBER_ERROR_NR];
588 __le64 errors_reset_time;
592 #define BCH_MEMBER_V1_BYTES 56
594 LE64_BITMASK(BCH_MEMBER_STATE, struct bch_member, flags, 0, 4)
595 /* 4-14 unused, was TIER, HAS_(META)DATA, REPLACEMENT */
596 LE64_BITMASK(BCH_MEMBER_DISCARD, struct bch_member, flags, 14, 15)
597 LE64_BITMASK(BCH_MEMBER_DATA_ALLOWED, struct bch_member, flags, 15, 20)
598 LE64_BITMASK(BCH_MEMBER_GROUP, struct bch_member, flags, 20, 28)
599 LE64_BITMASK(BCH_MEMBER_DURABILITY, struct bch_member, flags, 28, 30)
600 LE64_BITMASK(BCH_MEMBER_FREESPACE_INITIALIZED,
601 struct bch_member, flags, 30, 31)
604 LE64_BITMASK(BCH_MEMBER_NR_READ_ERRORS, struct bch_member, flags[1], 0, 20);
605 LE64_BITMASK(BCH_MEMBER_NR_WRITE_ERRORS,struct bch_member, flags[1], 20, 40);
608 #define BCH_MEMBER_STATES() \
614 enum bch_member_state {
615 #define x(t, n) BCH_MEMBER_STATE_##t = n,
621 struct bch_sb_field_members_v1 {
622 struct bch_sb_field field;
623 struct bch_member _members[]; //Members are now variable size
626 struct bch_sb_field_members_v2 {
627 struct bch_sb_field field;
628 __le16 member_bytes; //size of single member entry
630 struct bch_member _members[];
633 /* BCH_SB_FIELD_crypt: */
643 #define BCH_KEY_MAGIC \
644 (((__u64) 'b' << 0)|((__u64) 'c' << 8)| \
645 ((__u64) 'h' << 16)|((__u64) '*' << 24)| \
646 ((__u64) '*' << 32)|((__u64) 'k' << 40)| \
647 ((__u64) 'e' << 48)|((__u64) 'y' << 56))
649 struct bch_encrypted_key {
655 * If this field is present in the superblock, it stores an encryption key which
656 * is used encrypt all other data/metadata. The key will normally be encrypted
657 * with the key userspace provides, but if encryption has been turned off we'll
658 * just store the master key unencrypted in the superblock so we can access the
659 * previously encrypted data.
661 struct bch_sb_field_crypt {
662 struct bch_sb_field field;
666 struct bch_encrypted_key key;
669 LE64_BITMASK(BCH_CRYPT_KDF_TYPE, struct bch_sb_field_crypt, flags, 0, 4);
676 /* stored as base 2 log of scrypt params: */
677 LE64_BITMASK(BCH_KDF_SCRYPT_N, struct bch_sb_field_crypt, kdf_flags, 0, 16);
678 LE64_BITMASK(BCH_KDF_SCRYPT_R, struct bch_sb_field_crypt, kdf_flags, 16, 32);
679 LE64_BITMASK(BCH_KDF_SCRYPT_P, struct bch_sb_field_crypt, kdf_flags, 32, 48);
681 /* BCH_SB_FIELD_replicas: */
683 #define BCH_DATA_TYPES() \
696 #define x(t, n) BCH_DATA_##t,
702 static inline bool data_type_is_empty(enum bch_data_type type)
706 case BCH_DATA_need_gc_gens:
707 case BCH_DATA_need_discard:
714 static inline bool data_type_is_hidden(enum bch_data_type type)
718 case BCH_DATA_journal:
725 struct bch_replicas_entry_v0 {
731 struct bch_sb_field_replicas_v0 {
732 struct bch_sb_field field;
733 struct bch_replicas_entry_v0 entries[];
734 } __packed __aligned(8);
736 struct bch_replicas_entry_v1 {
743 #define replicas_entry_bytes(_i) \
744 (offsetof(typeof(*(_i)), devs) + (_i)->nr_devs)
746 struct bch_sb_field_replicas {
747 struct bch_sb_field field;
748 struct bch_replicas_entry_v1 entries[];
749 } __packed __aligned(8);
751 /* BCH_SB_FIELD_disk_groups: */
753 #define BCH_SB_LABEL_SIZE 32
755 struct bch_disk_group {
756 __u8 label[BCH_SB_LABEL_SIZE];
758 } __packed __aligned(8);
760 LE64_BITMASK(BCH_GROUP_DELETED, struct bch_disk_group, flags[0], 0, 1)
761 LE64_BITMASK(BCH_GROUP_DATA_ALLOWED, struct bch_disk_group, flags[0], 1, 6)
762 LE64_BITMASK(BCH_GROUP_PARENT, struct bch_disk_group, flags[0], 6, 24)
764 struct bch_sb_field_disk_groups {
765 struct bch_sb_field field;
766 struct bch_disk_group entries[];
767 } __packed __aligned(8);
770 * On clean shutdown, store btree roots and current journal sequence number in
777 __u8 type; /* designates what this jset holds */
780 struct bkey_i start[0];
784 struct bch_sb_field_clean {
785 struct bch_sb_field field;
788 __le16 _read_clock; /* no longer used */
792 struct jset_entry start[0];
796 struct journal_seq_blacklist_entry {
801 struct bch_sb_field_journal_seq_blacklist {
802 struct bch_sb_field field;
803 struct journal_seq_blacklist_entry start[];
806 struct bch_sb_field_errors {
807 struct bch_sb_field field;
808 struct bch_sb_field_error_entry {
810 __le64 last_error_time;
814 LE64_BITMASK(BCH_SB_ERROR_ENTRY_ID, struct bch_sb_field_error_entry, v, 0, 16);
815 LE64_BITMASK(BCH_SB_ERROR_ENTRY_NR, struct bch_sb_field_error_entry, v, 16, 64);
817 struct bch_sb_field_ext {
818 struct bch_sb_field field;
819 __le64 recovery_passes_required[2];
820 __le64 errors_silent[8];
823 struct bch_sb_field_downgrade_entry {
825 __le64 recovery_passes[2];
827 __le16 errors[] __counted_by(nr_errors);
828 } __packed __aligned(2);
830 struct bch_sb_field_downgrade {
831 struct bch_sb_field field;
832 struct bch_sb_field_downgrade_entry entries[];
838 * New versioning scheme:
839 * One common version number for all on disk data structures - superblock, btree
840 * nodes, journal entries
842 #define BCH_VERSION_MAJOR(_v) ((__u16) ((_v) >> 10))
843 #define BCH_VERSION_MINOR(_v) ((__u16) ((_v) & ~(~0U << 10)))
844 #define BCH_VERSION(_major, _minor) (((_major) << 10)|(_minor) << 0)
847 * field 1: version name
848 * field 2: BCH_VERSION(major, minor)
849 * field 3: recovery passess required on upgrade
851 #define BCH_METADATA_VERSIONS() \
852 x(bkey_renumber, BCH_VERSION(0, 10)) \
853 x(inode_btree_change, BCH_VERSION(0, 11)) \
854 x(snapshot, BCH_VERSION(0, 12)) \
855 x(inode_backpointers, BCH_VERSION(0, 13)) \
856 x(btree_ptr_sectors_written, BCH_VERSION(0, 14)) \
857 x(snapshot_2, BCH_VERSION(0, 15)) \
858 x(reflink_p_fix, BCH_VERSION(0, 16)) \
859 x(subvol_dirent, BCH_VERSION(0, 17)) \
860 x(inode_v2, BCH_VERSION(0, 18)) \
861 x(freespace, BCH_VERSION(0, 19)) \
862 x(alloc_v4, BCH_VERSION(0, 20)) \
863 x(new_data_types, BCH_VERSION(0, 21)) \
864 x(backpointers, BCH_VERSION(0, 22)) \
865 x(inode_v3, BCH_VERSION(0, 23)) \
866 x(unwritten_extents, BCH_VERSION(0, 24)) \
867 x(bucket_gens, BCH_VERSION(0, 25)) \
868 x(lru_v2, BCH_VERSION(0, 26)) \
869 x(fragmentation_lru, BCH_VERSION(0, 27)) \
870 x(no_bps_in_alloc_keys, BCH_VERSION(0, 28)) \
871 x(snapshot_trees, BCH_VERSION(0, 29)) \
872 x(major_minor, BCH_VERSION(1, 0)) \
873 x(snapshot_skiplists, BCH_VERSION(1, 1)) \
874 x(deleted_inodes, BCH_VERSION(1, 2)) \
875 x(rebalance_work, BCH_VERSION(1, 3)) \
876 x(member_seq, BCH_VERSION(1, 4)) \
877 x(subvolume_fs_parent, BCH_VERSION(1, 5)) \
878 x(btree_subvolume_children, BCH_VERSION(1, 6))
880 enum bcachefs_metadata_version {
881 bcachefs_metadata_version_min = 9,
882 #define x(t, n) bcachefs_metadata_version_##t = n,
883 BCH_METADATA_VERSIONS()
885 bcachefs_metadata_version_max
888 static const __maybe_unused
889 unsigned bcachefs_metadata_required_upgrade_below = bcachefs_metadata_version_rebalance_work;
891 #define bcachefs_metadata_version_current (bcachefs_metadata_version_max - 1)
893 #define BCH_SB_SECTOR 8
894 #define BCH_SB_MEMBERS_MAX 64 /* XXX kill */
896 struct bch_sb_layout {
897 __uuid_t magic; /* bcachefs superblock UUID */
899 __u8 sb_max_size_bits; /* base 2 of 512 byte sectors */
902 __le64 sb_offset[61];
903 } __packed __aligned(8);
905 #define BCH_SB_LAYOUT_SECTOR 7
908 * @offset - sector where this sb was written
909 * @version - on disk format version
910 * @version_min - Oldest metadata version this filesystem contains; so we can
911 * safely drop compatibility code and refuse to mount filesystems
913 * @magic - identifies as a bcachefs superblock (BCHFS_MAGIC)
914 * @seq - incremented each time superblock is written
915 * @uuid - used for generating various magic numbers and identifying
916 * member devices, never changes
917 * @user_uuid - user visible UUID, may be changed
918 * @label - filesystem label
919 * @seq - identifies most recent superblock, incremented each time
920 * superblock is written
921 * @features - enabled incompatible features
924 struct bch_csum csum;
931 __u8 label[BCH_SB_LABEL_SIZE];
942 __le32 time_precision;
949 struct bch_sb_layout layout;
951 struct bch_sb_field start[0];
953 } __packed __aligned(8);
957 * BCH_SB_INITALIZED - set on first mount
958 * BCH_SB_CLEAN - did we shut down cleanly? Just a hint, doesn't affect
959 * behaviour of mount/recovery path:
960 * BCH_SB_INODE_32BIT - limit inode numbers to 32 bits
961 * BCH_SB_128_BIT_MACS - 128 bit macs instead of 80
962 * BCH_SB_ENCRYPTION_TYPE - if nonzero encryption is enabled; overrides
963 * DATA/META_CSUM_TYPE. Also indicates encryption
964 * algorithm in use, if/when we get more than one
967 LE16_BITMASK(BCH_SB_BLOCK_SIZE, struct bch_sb, block_size, 0, 16);
969 LE64_BITMASK(BCH_SB_INITIALIZED, struct bch_sb, flags[0], 0, 1);
970 LE64_BITMASK(BCH_SB_CLEAN, struct bch_sb, flags[0], 1, 2);
971 LE64_BITMASK(BCH_SB_CSUM_TYPE, struct bch_sb, flags[0], 2, 8);
972 LE64_BITMASK(BCH_SB_ERROR_ACTION, struct bch_sb, flags[0], 8, 12);
974 LE64_BITMASK(BCH_SB_BTREE_NODE_SIZE, struct bch_sb, flags[0], 12, 28);
976 LE64_BITMASK(BCH_SB_GC_RESERVE, struct bch_sb, flags[0], 28, 33);
977 LE64_BITMASK(BCH_SB_ROOT_RESERVE, struct bch_sb, flags[0], 33, 40);
979 LE64_BITMASK(BCH_SB_META_CSUM_TYPE, struct bch_sb, flags[0], 40, 44);
980 LE64_BITMASK(BCH_SB_DATA_CSUM_TYPE, struct bch_sb, flags[0], 44, 48);
982 LE64_BITMASK(BCH_SB_META_REPLICAS_WANT, struct bch_sb, flags[0], 48, 52);
983 LE64_BITMASK(BCH_SB_DATA_REPLICAS_WANT, struct bch_sb, flags[0], 52, 56);
985 LE64_BITMASK(BCH_SB_POSIX_ACL, struct bch_sb, flags[0], 56, 57);
986 LE64_BITMASK(BCH_SB_USRQUOTA, struct bch_sb, flags[0], 57, 58);
987 LE64_BITMASK(BCH_SB_GRPQUOTA, struct bch_sb, flags[0], 58, 59);
988 LE64_BITMASK(BCH_SB_PRJQUOTA, struct bch_sb, flags[0], 59, 60);
990 LE64_BITMASK(BCH_SB_HAS_ERRORS, struct bch_sb, flags[0], 60, 61);
991 LE64_BITMASK(BCH_SB_HAS_TOPOLOGY_ERRORS,struct bch_sb, flags[0], 61, 62);
993 LE64_BITMASK(BCH_SB_BIG_ENDIAN, struct bch_sb, flags[0], 62, 63);
995 LE64_BITMASK(BCH_SB_STR_HASH_TYPE, struct bch_sb, flags[1], 0, 4);
996 LE64_BITMASK(BCH_SB_COMPRESSION_TYPE_LO,struct bch_sb, flags[1], 4, 8);
997 LE64_BITMASK(BCH_SB_INODE_32BIT, struct bch_sb, flags[1], 8, 9);
999 LE64_BITMASK(BCH_SB_128_BIT_MACS, struct bch_sb, flags[1], 9, 10);
1000 LE64_BITMASK(BCH_SB_ENCRYPTION_TYPE, struct bch_sb, flags[1], 10, 14);
1003 * Max size of an extent that may require bouncing to read or write
1004 * (checksummed, compressed): 64k
1006 LE64_BITMASK(BCH_SB_ENCODED_EXTENT_MAX_BITS,
1007 struct bch_sb, flags[1], 14, 20);
1009 LE64_BITMASK(BCH_SB_META_REPLICAS_REQ, struct bch_sb, flags[1], 20, 24);
1010 LE64_BITMASK(BCH_SB_DATA_REPLICAS_REQ, struct bch_sb, flags[1], 24, 28);
1012 LE64_BITMASK(BCH_SB_PROMOTE_TARGET, struct bch_sb, flags[1], 28, 40);
1013 LE64_BITMASK(BCH_SB_FOREGROUND_TARGET, struct bch_sb, flags[1], 40, 52);
1014 LE64_BITMASK(BCH_SB_BACKGROUND_TARGET, struct bch_sb, flags[1], 52, 64);
1016 LE64_BITMASK(BCH_SB_BACKGROUND_COMPRESSION_TYPE_LO,
1017 struct bch_sb, flags[2], 0, 4);
1018 LE64_BITMASK(BCH_SB_GC_RESERVE_BYTES, struct bch_sb, flags[2], 4, 64);
1020 LE64_BITMASK(BCH_SB_ERASURE_CODE, struct bch_sb, flags[3], 0, 16);
1021 LE64_BITMASK(BCH_SB_METADATA_TARGET, struct bch_sb, flags[3], 16, 28);
1022 LE64_BITMASK(BCH_SB_SHARD_INUMS, struct bch_sb, flags[3], 28, 29);
1023 LE64_BITMASK(BCH_SB_INODES_USE_KEY_CACHE,struct bch_sb, flags[3], 29, 30);
1024 LE64_BITMASK(BCH_SB_JOURNAL_FLUSH_DELAY,struct bch_sb, flags[3], 30, 62);
1025 LE64_BITMASK(BCH_SB_JOURNAL_FLUSH_DISABLED,struct bch_sb, flags[3], 62, 63);
1026 LE64_BITMASK(BCH_SB_JOURNAL_RECLAIM_DELAY,struct bch_sb, flags[4], 0, 32);
1027 LE64_BITMASK(BCH_SB_JOURNAL_TRANSACTION_NAMES,struct bch_sb, flags[4], 32, 33);
1028 LE64_BITMASK(BCH_SB_NOCOW, struct bch_sb, flags[4], 33, 34);
1029 LE64_BITMASK(BCH_SB_WRITE_BUFFER_SIZE, struct bch_sb, flags[4], 34, 54);
1030 LE64_BITMASK(BCH_SB_VERSION_UPGRADE, struct bch_sb, flags[4], 54, 56);
1032 LE64_BITMASK(BCH_SB_COMPRESSION_TYPE_HI,struct bch_sb, flags[4], 56, 60);
1033 LE64_BITMASK(BCH_SB_BACKGROUND_COMPRESSION_TYPE_HI,
1034 struct bch_sb, flags[4], 60, 64);
1036 LE64_BITMASK(BCH_SB_VERSION_UPGRADE_COMPLETE,
1037 struct bch_sb, flags[5], 0, 16);
1039 static inline __u64 BCH_SB_COMPRESSION_TYPE(const struct bch_sb *sb)
1041 return BCH_SB_COMPRESSION_TYPE_LO(sb) | (BCH_SB_COMPRESSION_TYPE_HI(sb) << 4);
1044 static inline void SET_BCH_SB_COMPRESSION_TYPE(struct bch_sb *sb, __u64 v)
1046 SET_BCH_SB_COMPRESSION_TYPE_LO(sb, v);
1047 SET_BCH_SB_COMPRESSION_TYPE_HI(sb, v >> 4);
1050 static inline __u64 BCH_SB_BACKGROUND_COMPRESSION_TYPE(const struct bch_sb *sb)
1052 return BCH_SB_BACKGROUND_COMPRESSION_TYPE_LO(sb) |
1053 (BCH_SB_BACKGROUND_COMPRESSION_TYPE_HI(sb) << 4);
1056 static inline void SET_BCH_SB_BACKGROUND_COMPRESSION_TYPE(struct bch_sb *sb, __u64 v)
1058 SET_BCH_SB_BACKGROUND_COMPRESSION_TYPE_LO(sb, v);
1059 SET_BCH_SB_BACKGROUND_COMPRESSION_TYPE_HI(sb, v >> 4);
1065 * journal_seq_blacklist_v3: gates BCH_SB_FIELD_journal_seq_blacklist
1066 * reflink: gates KEY_TYPE_reflink
1067 * inline_data: gates KEY_TYPE_inline_data
1068 * new_siphash: gates BCH_STR_HASH_siphash
1069 * new_extent_overwrite: gates BTREE_NODE_NEW_EXTENT_OVERWRITE
1071 #define BCH_SB_FEATURES() \
1075 x(atomic_nlink, 3) \
1077 x(journal_seq_blacklist_v3, 5) \
1081 x(new_extent_overwrite, 9) \
1082 x(incompressible, 10) \
1083 x(btree_ptr_v2, 11) \
1084 x(extents_above_btree_updates, 12) \
1085 x(btree_updates_journalled, 13) \
1086 x(reflink_inline_data, 14) \
1088 x(journal_no_flush, 16) \
1090 x(extents_across_btree_nodes, 18)
1092 #define BCH_SB_FEATURES_ALWAYS \
1093 ((1ULL << BCH_FEATURE_new_extent_overwrite)| \
1094 (1ULL << BCH_FEATURE_extents_above_btree_updates)|\
1095 (1ULL << BCH_FEATURE_btree_updates_journalled)|\
1096 (1ULL << BCH_FEATURE_alloc_v2)|\
1097 (1ULL << BCH_FEATURE_extents_across_btree_nodes))
1099 #define BCH_SB_FEATURES_ALL \
1100 (BCH_SB_FEATURES_ALWAYS| \
1101 (1ULL << BCH_FEATURE_new_siphash)| \
1102 (1ULL << BCH_FEATURE_btree_ptr_v2)| \
1103 (1ULL << BCH_FEATURE_new_varint)| \
1104 (1ULL << BCH_FEATURE_journal_no_flush))
1106 enum bch_sb_feature {
1107 #define x(f, n) BCH_FEATURE_##f,
1113 #define BCH_SB_COMPAT() \
1115 x(alloc_metadata, 1) \
1116 x(extents_above_btree_updates_done, 2) \
1117 x(bformat_overflow_done, 3)
1119 enum bch_sb_compat {
1120 #define x(f, n) BCH_COMPAT_##f,
1128 #define BCH_VERSION_UPGRADE_OPTS() \
1130 x(incompatible, 1) \
1133 enum bch_version_upgrade_opts {
1134 #define x(t, n) BCH_VERSION_UPGRADE_##t = n,
1135 BCH_VERSION_UPGRADE_OPTS()
1139 #define BCH_REPLICAS_MAX 4U
1141 #define BCH_BKEY_PTRS_MAX 16U
1143 #define BCH_ERROR_ACTIONS() \
1148 enum bch_error_actions {
1149 #define x(t, n) BCH_ON_ERROR_##t = n,
1155 #define BCH_STR_HASH_TYPES() \
1161 enum bch_str_hash_type {
1162 #define x(t, n) BCH_STR_HASH_##t = n,
1163 BCH_STR_HASH_TYPES()
1168 #define BCH_STR_HASH_OPTS() \
1173 enum bch_str_hash_opts {
1174 #define x(t, n) BCH_STR_HASH_OPT_##t = n,
1180 #define BCH_CSUM_TYPES() \
1182 x(crc32c_nonzero, 1) \
1183 x(crc64_nonzero, 2) \
1184 x(chacha20_poly1305_80, 3) \
1185 x(chacha20_poly1305_128, 4) \
1190 enum bch_csum_type {
1191 #define x(t, n) BCH_CSUM_##t = n,
1197 static const __maybe_unused unsigned bch_crc_bytes[] = {
1198 [BCH_CSUM_none] = 0,
1199 [BCH_CSUM_crc32c_nonzero] = 4,
1200 [BCH_CSUM_crc32c] = 4,
1201 [BCH_CSUM_crc64_nonzero] = 8,
1202 [BCH_CSUM_crc64] = 8,
1203 [BCH_CSUM_xxhash] = 8,
1204 [BCH_CSUM_chacha20_poly1305_80] = 10,
1205 [BCH_CSUM_chacha20_poly1305_128] = 16,
1208 static inline _Bool bch2_csum_type_is_encryption(enum bch_csum_type type)
1211 case BCH_CSUM_chacha20_poly1305_80:
1212 case BCH_CSUM_chacha20_poly1305_128:
1219 #define BCH_CSUM_OPTS() \
1225 enum bch_csum_opts {
1226 #define x(t, n) BCH_CSUM_OPT_##t = n,
1232 #define BCH_COMPRESSION_TYPES() \
1238 x(incompressible, 5)
1240 enum bch_compression_type {
1241 #define x(t, n) BCH_COMPRESSION_TYPE_##t = n,
1242 BCH_COMPRESSION_TYPES()
1244 BCH_COMPRESSION_TYPE_NR
1247 #define BCH_COMPRESSION_OPTS() \
1253 enum bch_compression_opts {
1254 #define x(t, n) BCH_COMPRESSION_OPT_##t = n,
1255 BCH_COMPRESSION_OPTS()
1257 BCH_COMPRESSION_OPT_NR
1263 * The various other data structures have their own magic numbers, which are
1264 * xored with the first part of the cache set's UUID
1267 #define BCACHE_MAGIC \
1268 UUID_INIT(0xc68573f6, 0x4e1a, 0x45ca, \
1269 0x82, 0x65, 0xf5, 0x7f, 0x48, 0xba, 0x6d, 0x81)
1270 #define BCHFS_MAGIC \
1271 UUID_INIT(0xc68573f6, 0x66ce, 0x90a9, \
1272 0xd9, 0x6a, 0x60, 0xcf, 0x80, 0x3d, 0xf7, 0xef)
1274 #define BCACHEFS_STATFS_MAGIC 0xca451a4e
1276 #define JSET_MAGIC __cpu_to_le64(0x245235c1a3625032ULL)
1277 #define BSET_MAGIC __cpu_to_le64(0x90135c78b99e07f5ULL)
1279 static inline __le64 __bch2_sb_magic(struct bch_sb *sb)
1283 memcpy(&ret, &sb->uuid, sizeof(ret));
1287 static inline __u64 __jset_magic(struct bch_sb *sb)
1289 return __le64_to_cpu(__bch2_sb_magic(sb) ^ JSET_MAGIC);
1292 static inline __u64 __bset_magic(struct bch_sb *sb)
1294 return __le64_to_cpu(__bch2_sb_magic(sb) ^ BSET_MAGIC);
1299 #define JSET_KEYS_U64s (sizeof(struct jset_entry) / sizeof(__u64))
1301 #define BCH_JSET_ENTRY_TYPES() \
1306 x(blacklist_v2, 4) \
1313 x(write_buffer_keys, 11) \
1317 #define x(f, nr) BCH_JSET_ENTRY_##f = nr,
1318 BCH_JSET_ENTRY_TYPES()
1323 static inline bool jset_entry_is_key(struct jset_entry *e)
1326 case BCH_JSET_ENTRY_btree_keys:
1327 case BCH_JSET_ENTRY_btree_root:
1328 case BCH_JSET_ENTRY_overwrite:
1329 case BCH_JSET_ENTRY_write_buffer_keys:
1337 * Journal sequence numbers can be blacklisted: bsets record the max sequence
1338 * number of all the journal entries they contain updates for, so that on
1339 * recovery we can ignore those bsets that contain index updates newer that what
1340 * made it into the journal.
1342 * This means that we can't reuse that journal_seq - we have to skip it, and
1343 * then record that we skipped it so that the next time we crash and recover we
1344 * don't think there was a missing journal entry.
1346 struct jset_entry_blacklist {
1347 struct jset_entry entry;
1351 struct jset_entry_blacklist_v2 {
1352 struct jset_entry entry;
1357 #define BCH_FS_USAGE_TYPES() \
1363 #define x(f, nr) BCH_FS_USAGE_##f = nr,
1364 BCH_FS_USAGE_TYPES()
1369 struct jset_entry_usage {
1370 struct jset_entry entry;
1374 struct jset_entry_data_usage {
1375 struct jset_entry entry;
1377 struct bch_replicas_entry_v1 r;
1380 struct jset_entry_clock {
1381 struct jset_entry entry;
1387 struct jset_entry_dev_usage_type {
1393 struct jset_entry_dev_usage {
1394 struct jset_entry entry;
1398 __le64 _buckets_ec; /* No longer used */
1399 __le64 _buckets_unavailable; /* No longer used */
1401 struct jset_entry_dev_usage_type d[];
1404 static inline unsigned jset_entry_dev_usage_nr_types(struct jset_entry_dev_usage *u)
1406 return (vstruct_bytes(&u->entry) - sizeof(struct jset_entry_dev_usage)) /
1407 sizeof(struct jset_entry_dev_usage_type);
1410 struct jset_entry_log {
1411 struct jset_entry entry;
1413 } __packed __aligned(8);
1415 struct jset_entry_datetime {
1416 struct jset_entry entry;
1418 } __packed __aligned(8);
1421 * On disk format for a journal entry:
1422 * seq is monotonically increasing; every journal entry has its own unique
1425 * last_seq is the oldest journal entry that still has keys the btree hasn't
1426 * flushed to disk yet.
1428 * version is for on disk format changes.
1431 struct bch_csum csum;
1438 __le32 u64s; /* size of d[] in u64s */
1440 __u8 encrypted_start[0];
1442 __le16 _read_clock; /* no longer used */
1443 __le16 _write_clock;
1445 /* Sequence number of oldest dirty journal entry */
1449 struct jset_entry start[0];
1451 } __packed __aligned(8);
1453 LE32_BITMASK(JSET_CSUM_TYPE, struct jset, flags, 0, 4);
1454 LE32_BITMASK(JSET_BIG_ENDIAN, struct jset, flags, 4, 5);
1455 LE32_BITMASK(JSET_NO_FLUSH, struct jset, flags, 5, 6);
1457 #define BCH_JOURNAL_BUCKETS_MIN 8
1461 enum btree_id_flags {
1462 BTREE_ID_EXTENTS = BIT(0),
1463 BTREE_ID_SNAPSHOTS = BIT(1),
1464 BTREE_ID_SNAPSHOT_FIELD = BIT(2),
1465 BTREE_ID_DATA = BIT(3),
1468 #define BCH_BTREE_IDS() \
1469 x(extents, 0, BTREE_ID_EXTENTS|BTREE_ID_SNAPSHOTS|BTREE_ID_DATA,\
1470 BIT_ULL(KEY_TYPE_whiteout)| \
1471 BIT_ULL(KEY_TYPE_error)| \
1472 BIT_ULL(KEY_TYPE_cookie)| \
1473 BIT_ULL(KEY_TYPE_extent)| \
1474 BIT_ULL(KEY_TYPE_reservation)| \
1475 BIT_ULL(KEY_TYPE_reflink_p)| \
1476 BIT_ULL(KEY_TYPE_inline_data)) \
1477 x(inodes, 1, BTREE_ID_SNAPSHOTS, \
1478 BIT_ULL(KEY_TYPE_whiteout)| \
1479 BIT_ULL(KEY_TYPE_inode)| \
1480 BIT_ULL(KEY_TYPE_inode_v2)| \
1481 BIT_ULL(KEY_TYPE_inode_v3)| \
1482 BIT_ULL(KEY_TYPE_inode_generation)) \
1483 x(dirents, 2, BTREE_ID_SNAPSHOTS, \
1484 BIT_ULL(KEY_TYPE_whiteout)| \
1485 BIT_ULL(KEY_TYPE_hash_whiteout)| \
1486 BIT_ULL(KEY_TYPE_dirent)) \
1487 x(xattrs, 3, BTREE_ID_SNAPSHOTS, \
1488 BIT_ULL(KEY_TYPE_whiteout)| \
1489 BIT_ULL(KEY_TYPE_cookie)| \
1490 BIT_ULL(KEY_TYPE_hash_whiteout)| \
1491 BIT_ULL(KEY_TYPE_xattr)) \
1493 BIT_ULL(KEY_TYPE_alloc)| \
1494 BIT_ULL(KEY_TYPE_alloc_v2)| \
1495 BIT_ULL(KEY_TYPE_alloc_v3)| \
1496 BIT_ULL(KEY_TYPE_alloc_v4)) \
1498 BIT_ULL(KEY_TYPE_quota)) \
1500 BIT_ULL(KEY_TYPE_stripe)) \
1501 x(reflink, 7, BTREE_ID_EXTENTS|BTREE_ID_DATA, \
1502 BIT_ULL(KEY_TYPE_reflink_v)| \
1503 BIT_ULL(KEY_TYPE_indirect_inline_data)) \
1504 x(subvolumes, 8, 0, \
1505 BIT_ULL(KEY_TYPE_subvolume)) \
1506 x(snapshots, 9, 0, \
1507 BIT_ULL(KEY_TYPE_snapshot)) \
1509 BIT_ULL(KEY_TYPE_set)) \
1510 x(freespace, 11, BTREE_ID_EXTENTS, \
1511 BIT_ULL(KEY_TYPE_set)) \
1512 x(need_discard, 12, 0, \
1513 BIT_ULL(KEY_TYPE_set)) \
1514 x(backpointers, 13, 0, \
1515 BIT_ULL(KEY_TYPE_backpointer)) \
1516 x(bucket_gens, 14, 0, \
1517 BIT_ULL(KEY_TYPE_bucket_gens)) \
1518 x(snapshot_trees, 15, 0, \
1519 BIT_ULL(KEY_TYPE_snapshot_tree)) \
1520 x(deleted_inodes, 16, BTREE_ID_SNAPSHOT_FIELD, \
1521 BIT_ULL(KEY_TYPE_set)) \
1522 x(logged_ops, 17, 0, \
1523 BIT_ULL(KEY_TYPE_logged_op_truncate)| \
1524 BIT_ULL(KEY_TYPE_logged_op_finsert)) \
1525 x(rebalance_work, 18, BTREE_ID_SNAPSHOT_FIELD, \
1526 BIT_ULL(KEY_TYPE_set)|BIT_ULL(KEY_TYPE_cookie)) \
1527 x(subvolume_children, 19, 0, \
1528 BIT_ULL(KEY_TYPE_set))
1531 #define x(name, nr, ...) BTREE_ID_##name = nr,
1537 #define BTREE_MAX_DEPTH 4U
1544 * On disk a btree node is a list/log of these; within each set the keys are
1551 * Highest journal entry this bset contains keys for.
1552 * If on recovery we don't see that journal entry, this bset is ignored:
1553 * this allows us to preserve the order of all index updates after a
1554 * crash, since the journal records a total order of all index updates
1555 * and anything that didn't make it to the journal doesn't get used.
1561 __le16 u64s; /* count of d[] in u64s */
1563 struct bkey_packed start[0];
1565 } __packed __aligned(8);
1567 LE32_BITMASK(BSET_CSUM_TYPE, struct bset, flags, 0, 4);
1569 LE32_BITMASK(BSET_BIG_ENDIAN, struct bset, flags, 4, 5);
1570 LE32_BITMASK(BSET_SEPARATE_WHITEOUTS,
1571 struct bset, flags, 5, 6);
1573 /* Sector offset within the btree node: */
1574 LE32_BITMASK(BSET_OFFSET, struct bset, flags, 16, 32);
1577 struct bch_csum csum;
1580 /* this flags field is encrypted, unlike bset->flags: */
1583 /* Closed interval: */
1584 struct bpos min_key;
1585 struct bpos max_key;
1586 struct bch_extent_ptr _ptr; /* not used anymore */
1587 struct bkey_format format;
1598 } __packed __aligned(8);
1600 LE64_BITMASK(BTREE_NODE_ID_LO, struct btree_node, flags, 0, 4);
1601 LE64_BITMASK(BTREE_NODE_LEVEL, struct btree_node, flags, 4, 8);
1602 LE64_BITMASK(BTREE_NODE_NEW_EXTENT_OVERWRITE,
1603 struct btree_node, flags, 8, 9);
1604 LE64_BITMASK(BTREE_NODE_ID_HI, struct btree_node, flags, 9, 25);
1606 LE64_BITMASK(BTREE_NODE_SEQ, struct btree_node, flags, 32, 64);
1608 static inline __u64 BTREE_NODE_ID(struct btree_node *n)
1610 return BTREE_NODE_ID_LO(n) | (BTREE_NODE_ID_HI(n) << 4);
1613 static inline void SET_BTREE_NODE_ID(struct btree_node *n, __u64 v)
1615 SET_BTREE_NODE_ID_LO(n, v);
1616 SET_BTREE_NODE_ID_HI(n, v >> 4);
1619 struct btree_node_entry {
1620 struct bch_csum csum;
1630 } __packed __aligned(8);
1632 #endif /* _BCACHEFS_FORMAT_H */