1 /* SPDX-License-Identifier: GPL-2.0 */
2 #ifndef _BCACHEFS_FORMAT_H
3 #define _BCACHEFS_FORMAT_H
6 * bcachefs on disk data structures
10 * There are three main types of on disk data structures in bcachefs (this is
11 * reduced from 5 in bcache)
17 * The btree is the primary structure; most metadata exists as keys in the
18 * various btrees. There are only a small number of btrees, they're not
19 * sharded - we have one btree for extents, another for inodes, et cetera.
23 * The superblock contains the location of the journal, the list of devices in
24 * the filesystem, and in general any metadata we need in order to decide
25 * whether we can start a filesystem or prior to reading the journal/btree
28 * The superblock is extensible, and most of the contents of the superblock are
29 * in variable length, type tagged fields; see struct bch_sb_field.
31 * Backup superblocks do not reside in a fixed location; also, superblocks do
32 * not have a fixed size. To locate backup superblocks we have struct
33 * bch_sb_layout; we store a copy of this inside every superblock, and also
34 * before the first superblock.
38 * The journal primarily records btree updates in the order they occurred;
39 * journal replay consists of just iterating over all the keys in the open
40 * journal entries and re-inserting them into the btrees.
42 * The journal also contains entry types for the btree roots, and blacklisted
43 * journal sequence numbers (see journal_seq_blacklist.c).
47 * bcachefs btrees are copy on write b+ trees, where nodes are big (typically
48 * 128k-256k) and log structured. We use struct btree_node for writing the first
49 * entry in a given node (offset 0), and struct btree_node_entry for all
52 * After the header, btree node entries contain a list of keys in sorted order.
53 * Values are stored inline with the keys; since values are variable length (and
54 * keys effectively are variable length too, due to packing) we can't do random
55 * access without building up additional in memory tables in the btree node read
58 * BTREE KEYS (struct bkey):
60 * The various btrees share a common format for the key - so as to avoid
61 * switching in fastpath lookup/comparison code - but define their own
62 * structures for the key values.
64 * The size of a key/value pair is stored as a u8 in units of u64s, so the max
65 * size is just under 2k. The common part also contains a type tag for the
66 * value, and a format field indicating whether the key is packed or not (and
67 * also meant to allow adding new key fields in the future, if desired).
69 * bkeys, when stored within a btree node, may also be packed. In that case, the
70 * bkey_format in that node is used to unpack it. Packed bkeys mean that we can
71 * be generous with field sizes in the common part of the key format (64 bit
72 * inode number, 64 bit offset, 96 bit version field, etc.) for negligible cost.
75 #include <asm/types.h>
76 #include <asm/byteorder.h>
77 #include <linux/kernel.h>
78 #include <linux/uuid.h>
82 typedef uuid_t __uuid_t;
85 #define BITMASK(name, type, field, offset, end) \
86 static const __maybe_unused unsigned name##_OFFSET = offset; \
87 static const __maybe_unused unsigned name##_BITS = (end - offset); \
89 static inline __u64 name(const type *k) \
91 return (k->field >> offset) & ~(~0ULL << (end - offset)); \
94 static inline void SET_##name(type *k, __u64 v) \
96 k->field &= ~(~(~0ULL << (end - offset)) << offset); \
97 k->field |= (v & ~(~0ULL << (end - offset))) << offset; \
100 #define LE_BITMASK(_bits, name, type, field, offset, end) \
101 static const __maybe_unused unsigned name##_OFFSET = offset; \
102 static const __maybe_unused unsigned name##_BITS = (end - offset); \
103 static const __maybe_unused __u##_bits name##_MAX = (1ULL << (end - offset)) - 1;\
105 static inline __u64 name(const type *k) \
107 return (__le##_bits##_to_cpu(k->field) >> offset) & \
108 ~(~0ULL << (end - offset)); \
111 static inline void SET_##name(type *k, __u64 v) \
113 __u##_bits new = __le##_bits##_to_cpu(k->field); \
115 new &= ~(~(~0ULL << (end - offset)) << offset); \
116 new |= (v & ~(~0ULL << (end - offset))) << offset; \
117 k->field = __cpu_to_le##_bits(new); \
120 #define LE16_BITMASK(n, t, f, o, e) LE_BITMASK(16, n, t, f, o, e)
121 #define LE32_BITMASK(n, t, f, o, e) LE_BITMASK(32, n, t, f, o, e)
122 #define LE64_BITMASK(n, t, f, o, e) LE_BITMASK(64, n, t, f, o, e)
127 /* One unused slot for now: */
128 __u8 bits_per_field[6];
129 __le64 field_offset[6];
132 /* Btree keys - all units are in sectors */
136 * Word order matches machine byte order - btree code treats a bpos as a
137 * single large integer, for search/comparison purposes
139 * Note that wherever a bpos is embedded in another on disk data
140 * structure, it has to be byte swabbed when reading in metadata that
141 * wasn't written in native endian order:
143 #if __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__
147 #elif __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__
149 __u64 offset; /* Points to end of extent - sectors */
152 #error edit for your odd byteorder.
155 #if __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__
160 #define KEY_INODE_MAX ((__u64)~0ULL)
161 #define KEY_OFFSET_MAX ((__u64)~0ULL)
162 #define KEY_SNAPSHOT_MAX ((__u32)~0U)
163 #define KEY_SIZE_MAX ((__u32)~0U)
165 static inline struct bpos SPOS(__u64 inode, __u64 offset, __u32 snapshot)
167 return (struct bpos) {
170 .snapshot = snapshot,
174 #define POS_MIN SPOS(0, 0, 0)
175 #define POS_MAX SPOS(KEY_INODE_MAX, KEY_OFFSET_MAX, 0)
176 #define SPOS_MAX SPOS(KEY_INODE_MAX, KEY_OFFSET_MAX, KEY_SNAPSHOT_MAX)
177 #define POS(_inode, _offset) SPOS(_inode, _offset, 0)
179 /* Empty placeholder struct, for container_of() */
185 #if __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__
188 #elif __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__
192 } __packed __aligned(4);
195 /* Size of combined key and value, in u64s */
198 /* Format of key (0 for format local to btree node) */
199 #if defined(__LITTLE_ENDIAN_BITFIELD)
202 #elif defined (__BIG_ENDIAN_BITFIELD)
203 __u8 needs_whiteout:1,
206 #error edit for your odd byteorder.
209 /* Type of the value */
212 #if __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__
215 struct bversion version;
216 __u32 size; /* extent size, in sectors */
218 #elif __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__
220 __u32 size; /* extent size, in sectors */
221 struct bversion version;
226 #if __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__
228 * The big-endian version of bkey can't be compiled by rustc with the "aligned"
229 * attr since it doesn't allow types to have both "packed" and "aligned" attrs.
230 * So for Rust compatibility, don't include this. It can be included in the LE
231 * version because the "packed" attr is redundant in that case.
233 * History: (quoting Kent)
235 * Specifically, when i was designing bkey, I wanted the header to be no
236 * bigger than necessary so that bkey_packed could use the rest. That means that
237 * decently offten extent keys will fit into only 8 bytes, instead of spilling over
240 * But packed_bkey treats the part after the header - the packed section -
241 * as a single multi word, variable length integer. And bkey, the unpacked
242 * version, is just a special case version of a bkey_packed; all the packed
243 * bkey code will work on keys in any packed format, the in-memory
244 * representation of an unpacked key also is just one type of packed key...
246 * So that constrains the key part of a bkig endian bkey to start right
249 * If we ever do a bkey_v2 and need to expand the hedaer by another byte for
250 * some reason - that will clean up this wart.
259 /* Size of combined key and value, in u64s */
262 /* Format of key (0 for format local to btree node) */
265 * XXX: next incompat on disk format change, switch format and
266 * needs_whiteout - bkey_packed() will be cheaper if format is the high
267 * bits of the bitfield
269 #if defined(__LITTLE_ENDIAN_BITFIELD)
272 #elif defined (__BIG_ENDIAN_BITFIELD)
273 __u8 needs_whiteout:1,
277 /* Type of the value */
282 * We copy bkeys with struct assignment in various places, and while
283 * that shouldn't be done with packed bkeys we can't disallow it in C,
284 * and it's legal to cast a bkey to a bkey_packed - so padding it out
285 * to the same size as struct bkey should hopefully be safest.
287 __u8 pad[sizeof(struct bkey) - 3];
288 } __packed __aligned(8);
295 #define BKEY_U64s (sizeof(struct bkey) / sizeof(__u64))
296 #define BKEY_U64s_MAX U8_MAX
297 #define BKEY_VAL_U64s_MAX (BKEY_U64s_MAX - BKEY_U64s)
299 #define KEY_PACKED_BITS_START 24
301 #define KEY_FORMAT_LOCAL_BTREE 0
302 #define KEY_FORMAT_CURRENT 1
304 enum bch_bkey_fields {
309 BKEY_FIELD_VERSION_HI,
310 BKEY_FIELD_VERSION_LO,
314 #define bkey_format_field(name, field) \
315 [BKEY_FIELD_##name] = (sizeof(((struct bkey *) NULL)->field) * 8)
317 #define BKEY_FORMAT_CURRENT \
318 ((struct bkey_format) { \
319 .key_u64s = BKEY_U64s, \
320 .nr_fields = BKEY_NR_FIELDS, \
321 .bits_per_field = { \
322 bkey_format_field(INODE, p.inode), \
323 bkey_format_field(OFFSET, p.offset), \
324 bkey_format_field(SNAPSHOT, p.snapshot), \
325 bkey_format_field(SIZE, size), \
326 bkey_format_field(VERSION_HI, version.hi), \
327 bkey_format_field(VERSION_LO, version.lo), \
331 /* bkey with inline value */
339 #define POS_KEY(_pos) \
342 .format = KEY_FORMAT_CURRENT, \
346 #define KEY(_inode, _offset, _size) \
349 .format = KEY_FORMAT_CURRENT, \
350 .p = POS(_inode, _offset), \
354 static inline void bkey_init(struct bkey *k)
359 #define bkey_bytes(_k) ((_k)->u64s * sizeof(__u64))
361 #define __BKEY_PADDED(key, pad) \
362 struct bkey_i key; __u64 key ## _pad[pad]
365 * - DELETED keys are used internally to mark keys that should be ignored but
366 * override keys in composition order. Their version number is ignored.
368 * - DISCARDED keys indicate that the data is all 0s because it has been
369 * discarded. DISCARDs may have a version; if the version is nonzero the key
370 * will be persistent, otherwise the key will be dropped whenever the btree
371 * node is rewritten (like DELETED keys).
373 * - ERROR: any read of the data returns a read error, as the data was lost due
374 * to a failing device. Like DISCARDED keys, they can be removed (overridden)
375 * by new writes or cluster-wide GC. Node repair can also overwrite them with
376 * the same or a more recent version number, but not with an older version
379 * - WHITEOUT: for hash table btrees
381 #define BCH_BKEY_TYPES() \
386 x(hash_whiteout, 4) \
391 x(inode_generation, 9) \
400 x(btree_ptr_v2, 18) \
401 x(indirect_inline_data, 19) \
413 x(snapshot_tree, 31) \
414 x(logged_op_truncate, 32) \
415 x(logged_op_finsert, 33)
418 #define x(name, nr) KEY_TYPE_##name = nr,
428 struct bch_whiteout {
441 struct bch_hash_whiteout {
449 /* 128 bits, sufficient for cryptographic MACs: */
453 } __packed __aligned(8);
455 struct bch_backpointer {
460 __u64 bucket_offset:40;
463 } __packed __aligned(8);
470 } __packed __aligned(8);
472 #define LRU_ID_STRIPES (1U << 16)
474 /* Optional/variable size superblock sections: */
476 struct bch_sb_field {
482 #define BCH_SB_FIELDS() \
491 x(journal_seq_blacklist, 8) \
499 #include "alloc_background_format.h"
500 #include "extents_format.h"
501 #include "reflink_format.h"
502 #include "ec_format.h"
503 #include "inode_format.h"
504 #include "dirent_format.h"
505 #include "xattr_format.h"
506 #include "quota_format.h"
507 #include "logged_ops_format.h"
508 #include "snapshot_format.h"
509 #include "subvolume_format.h"
510 #include "sb-counters_format.h"
512 enum bch_sb_field_type {
513 #define x(f, nr) BCH_SB_FIELD_##f = nr,
520 * Most superblock fields are replicated in all device's superblocks - a few are
523 #define BCH_SINGLE_DEVICE_SB_FIELDS \
524 ((1U << BCH_SB_FIELD_journal)| \
525 (1U << BCH_SB_FIELD_journal_v2))
527 /* BCH_SB_FIELD_journal: */
529 struct bch_sb_field_journal {
530 struct bch_sb_field field;
534 struct bch_sb_field_journal_v2 {
535 struct bch_sb_field field;
537 struct bch_sb_field_journal_v2_entry {
543 /* BCH_SB_FIELD_members_v1: */
545 #define BCH_MIN_NR_NBUCKETS (1 << 6)
547 #define BCH_IOPS_MEASUREMENTS() \
553 enum bch_iops_measurement {
554 #define x(t, n) BCH_IOPS_##t = n,
555 BCH_IOPS_MEASUREMENTS()
560 #define BCH_MEMBER_ERROR_TYPES() \
565 enum bch_member_error_type {
566 #define x(t, n) BCH_MEMBER_ERROR_##t = n,
567 BCH_MEMBER_ERROR_TYPES()
574 __le64 nbuckets; /* device size */
575 __le16 first_bucket; /* index of first bucket used */
576 __le16 bucket_size; /* sectors */
578 __le64 last_mount; /* time_t */
582 __le64 errors[BCH_MEMBER_ERROR_NR];
583 __le64 errors_at_reset[BCH_MEMBER_ERROR_NR];
584 __le64 errors_reset_time;
588 #define BCH_MEMBER_V1_BYTES 56
590 LE64_BITMASK(BCH_MEMBER_STATE, struct bch_member, flags, 0, 4)
591 /* 4-14 unused, was TIER, HAS_(META)DATA, REPLACEMENT */
592 LE64_BITMASK(BCH_MEMBER_DISCARD, struct bch_member, flags, 14, 15)
593 LE64_BITMASK(BCH_MEMBER_DATA_ALLOWED, struct bch_member, flags, 15, 20)
594 LE64_BITMASK(BCH_MEMBER_GROUP, struct bch_member, flags, 20, 28)
595 LE64_BITMASK(BCH_MEMBER_DURABILITY, struct bch_member, flags, 28, 30)
596 LE64_BITMASK(BCH_MEMBER_FREESPACE_INITIALIZED,
597 struct bch_member, flags, 30, 31)
600 LE64_BITMASK(BCH_MEMBER_NR_READ_ERRORS, struct bch_member, flags[1], 0, 20);
601 LE64_BITMASK(BCH_MEMBER_NR_WRITE_ERRORS,struct bch_member, flags[1], 20, 40);
604 #define BCH_MEMBER_STATES() \
610 enum bch_member_state {
611 #define x(t, n) BCH_MEMBER_STATE_##t = n,
617 struct bch_sb_field_members_v1 {
618 struct bch_sb_field field;
619 struct bch_member _members[]; //Members are now variable size
622 struct bch_sb_field_members_v2 {
623 struct bch_sb_field field;
624 __le16 member_bytes; //size of single member entry
626 struct bch_member _members[];
629 /* BCH_SB_FIELD_crypt: */
639 #define BCH_KEY_MAGIC \
640 (((__u64) 'b' << 0)|((__u64) 'c' << 8)| \
641 ((__u64) 'h' << 16)|((__u64) '*' << 24)| \
642 ((__u64) '*' << 32)|((__u64) 'k' << 40)| \
643 ((__u64) 'e' << 48)|((__u64) 'y' << 56))
645 struct bch_encrypted_key {
651 * If this field is present in the superblock, it stores an encryption key which
652 * is used encrypt all other data/metadata. The key will normally be encrypted
653 * with the key userspace provides, but if encryption has been turned off we'll
654 * just store the master key unencrypted in the superblock so we can access the
655 * previously encrypted data.
657 struct bch_sb_field_crypt {
658 struct bch_sb_field field;
662 struct bch_encrypted_key key;
665 LE64_BITMASK(BCH_CRYPT_KDF_TYPE, struct bch_sb_field_crypt, flags, 0, 4);
672 /* stored as base 2 log of scrypt params: */
673 LE64_BITMASK(BCH_KDF_SCRYPT_N, struct bch_sb_field_crypt, kdf_flags, 0, 16);
674 LE64_BITMASK(BCH_KDF_SCRYPT_R, struct bch_sb_field_crypt, kdf_flags, 16, 32);
675 LE64_BITMASK(BCH_KDF_SCRYPT_P, struct bch_sb_field_crypt, kdf_flags, 32, 48);
677 /* BCH_SB_FIELD_replicas: */
679 #define BCH_DATA_TYPES() \
692 #define x(t, n) BCH_DATA_##t,
698 static inline bool data_type_is_empty(enum bch_data_type type)
702 case BCH_DATA_need_gc_gens:
703 case BCH_DATA_need_discard:
710 static inline bool data_type_is_hidden(enum bch_data_type type)
714 case BCH_DATA_journal:
721 struct bch_replicas_entry_v0 {
727 struct bch_sb_field_replicas_v0 {
728 struct bch_sb_field field;
729 struct bch_replicas_entry_v0 entries[];
730 } __packed __aligned(8);
732 struct bch_replicas_entry_v1 {
739 #define replicas_entry_bytes(_i) \
740 (offsetof(typeof(*(_i)), devs) + (_i)->nr_devs)
742 struct bch_sb_field_replicas {
743 struct bch_sb_field field;
744 struct bch_replicas_entry_v1 entries[];
745 } __packed __aligned(8);
747 /* BCH_SB_FIELD_disk_groups: */
749 #define BCH_SB_LABEL_SIZE 32
751 struct bch_disk_group {
752 __u8 label[BCH_SB_LABEL_SIZE];
754 } __packed __aligned(8);
756 LE64_BITMASK(BCH_GROUP_DELETED, struct bch_disk_group, flags[0], 0, 1)
757 LE64_BITMASK(BCH_GROUP_DATA_ALLOWED, struct bch_disk_group, flags[0], 1, 6)
758 LE64_BITMASK(BCH_GROUP_PARENT, struct bch_disk_group, flags[0], 6, 24)
760 struct bch_sb_field_disk_groups {
761 struct bch_sb_field field;
762 struct bch_disk_group entries[];
763 } __packed __aligned(8);
766 * On clean shutdown, store btree roots and current journal sequence number in
773 __u8 type; /* designates what this jset holds */
776 struct bkey_i start[0];
780 struct bch_sb_field_clean {
781 struct bch_sb_field field;
784 __le16 _read_clock; /* no longer used */
788 struct jset_entry start[0];
792 struct journal_seq_blacklist_entry {
797 struct bch_sb_field_journal_seq_blacklist {
798 struct bch_sb_field field;
799 struct journal_seq_blacklist_entry start[];
802 struct bch_sb_field_errors {
803 struct bch_sb_field field;
804 struct bch_sb_field_error_entry {
806 __le64 last_error_time;
810 LE64_BITMASK(BCH_SB_ERROR_ENTRY_ID, struct bch_sb_field_error_entry, v, 0, 16);
811 LE64_BITMASK(BCH_SB_ERROR_ENTRY_NR, struct bch_sb_field_error_entry, v, 16, 64);
813 struct bch_sb_field_ext {
814 struct bch_sb_field field;
815 __le64 recovery_passes_required[2];
816 __le64 errors_silent[8];
819 struct bch_sb_field_downgrade_entry {
821 __le64 recovery_passes[2];
823 __le16 errors[] __counted_by(nr_errors);
824 } __packed __aligned(2);
826 struct bch_sb_field_downgrade {
827 struct bch_sb_field field;
828 struct bch_sb_field_downgrade_entry entries[];
834 * New versioning scheme:
835 * One common version number for all on disk data structures - superblock, btree
836 * nodes, journal entries
838 #define BCH_VERSION_MAJOR(_v) ((__u16) ((_v) >> 10))
839 #define BCH_VERSION_MINOR(_v) ((__u16) ((_v) & ~(~0U << 10)))
840 #define BCH_VERSION(_major, _minor) (((_major) << 10)|(_minor) << 0)
843 * field 1: version name
844 * field 2: BCH_VERSION(major, minor)
845 * field 3: recovery passess required on upgrade
847 #define BCH_METADATA_VERSIONS() \
848 x(bkey_renumber, BCH_VERSION(0, 10)) \
849 x(inode_btree_change, BCH_VERSION(0, 11)) \
850 x(snapshot, BCH_VERSION(0, 12)) \
851 x(inode_backpointers, BCH_VERSION(0, 13)) \
852 x(btree_ptr_sectors_written, BCH_VERSION(0, 14)) \
853 x(snapshot_2, BCH_VERSION(0, 15)) \
854 x(reflink_p_fix, BCH_VERSION(0, 16)) \
855 x(subvol_dirent, BCH_VERSION(0, 17)) \
856 x(inode_v2, BCH_VERSION(0, 18)) \
857 x(freespace, BCH_VERSION(0, 19)) \
858 x(alloc_v4, BCH_VERSION(0, 20)) \
859 x(new_data_types, BCH_VERSION(0, 21)) \
860 x(backpointers, BCH_VERSION(0, 22)) \
861 x(inode_v3, BCH_VERSION(0, 23)) \
862 x(unwritten_extents, BCH_VERSION(0, 24)) \
863 x(bucket_gens, BCH_VERSION(0, 25)) \
864 x(lru_v2, BCH_VERSION(0, 26)) \
865 x(fragmentation_lru, BCH_VERSION(0, 27)) \
866 x(no_bps_in_alloc_keys, BCH_VERSION(0, 28)) \
867 x(snapshot_trees, BCH_VERSION(0, 29)) \
868 x(major_minor, BCH_VERSION(1, 0)) \
869 x(snapshot_skiplists, BCH_VERSION(1, 1)) \
870 x(deleted_inodes, BCH_VERSION(1, 2)) \
871 x(rebalance_work, BCH_VERSION(1, 3)) \
872 x(member_seq, BCH_VERSION(1, 4)) \
873 x(subvolume_fs_parent, BCH_VERSION(1, 5)) \
874 x(btree_subvolume_children, BCH_VERSION(1, 6))
876 enum bcachefs_metadata_version {
877 bcachefs_metadata_version_min = 9,
878 #define x(t, n) bcachefs_metadata_version_##t = n,
879 BCH_METADATA_VERSIONS()
881 bcachefs_metadata_version_max
884 static const __maybe_unused
885 unsigned bcachefs_metadata_required_upgrade_below = bcachefs_metadata_version_rebalance_work;
887 #define bcachefs_metadata_version_current (bcachefs_metadata_version_max - 1)
889 #define BCH_SB_SECTOR 8
890 #define BCH_SB_MEMBERS_MAX 64 /* XXX kill */
892 struct bch_sb_layout {
893 __uuid_t magic; /* bcachefs superblock UUID */
895 __u8 sb_max_size_bits; /* base 2 of 512 byte sectors */
898 __le64 sb_offset[61];
899 } __packed __aligned(8);
901 #define BCH_SB_LAYOUT_SECTOR 7
904 * @offset - sector where this sb was written
905 * @version - on disk format version
906 * @version_min - Oldest metadata version this filesystem contains; so we can
907 * safely drop compatibility code and refuse to mount filesystems
909 * @magic - identifies as a bcachefs superblock (BCHFS_MAGIC)
910 * @seq - incremented each time superblock is written
911 * @uuid - used for generating various magic numbers and identifying
912 * member devices, never changes
913 * @user_uuid - user visible UUID, may be changed
914 * @label - filesystem label
915 * @seq - identifies most recent superblock, incremented each time
916 * superblock is written
917 * @features - enabled incompatible features
920 struct bch_csum csum;
927 __u8 label[BCH_SB_LABEL_SIZE];
938 __le32 time_precision;
945 struct bch_sb_layout layout;
947 struct bch_sb_field start[0];
949 } __packed __aligned(8);
953 * BCH_SB_INITALIZED - set on first mount
954 * BCH_SB_CLEAN - did we shut down cleanly? Just a hint, doesn't affect
955 * behaviour of mount/recovery path:
956 * BCH_SB_INODE_32BIT - limit inode numbers to 32 bits
957 * BCH_SB_128_BIT_MACS - 128 bit macs instead of 80
958 * BCH_SB_ENCRYPTION_TYPE - if nonzero encryption is enabled; overrides
959 * DATA/META_CSUM_TYPE. Also indicates encryption
960 * algorithm in use, if/when we get more than one
963 LE16_BITMASK(BCH_SB_BLOCK_SIZE, struct bch_sb, block_size, 0, 16);
965 LE64_BITMASK(BCH_SB_INITIALIZED, struct bch_sb, flags[0], 0, 1);
966 LE64_BITMASK(BCH_SB_CLEAN, struct bch_sb, flags[0], 1, 2);
967 LE64_BITMASK(BCH_SB_CSUM_TYPE, struct bch_sb, flags[0], 2, 8);
968 LE64_BITMASK(BCH_SB_ERROR_ACTION, struct bch_sb, flags[0], 8, 12);
970 LE64_BITMASK(BCH_SB_BTREE_NODE_SIZE, struct bch_sb, flags[0], 12, 28);
972 LE64_BITMASK(BCH_SB_GC_RESERVE, struct bch_sb, flags[0], 28, 33);
973 LE64_BITMASK(BCH_SB_ROOT_RESERVE, struct bch_sb, flags[0], 33, 40);
975 LE64_BITMASK(BCH_SB_META_CSUM_TYPE, struct bch_sb, flags[0], 40, 44);
976 LE64_BITMASK(BCH_SB_DATA_CSUM_TYPE, struct bch_sb, flags[0], 44, 48);
978 LE64_BITMASK(BCH_SB_META_REPLICAS_WANT, struct bch_sb, flags[0], 48, 52);
979 LE64_BITMASK(BCH_SB_DATA_REPLICAS_WANT, struct bch_sb, flags[0], 52, 56);
981 LE64_BITMASK(BCH_SB_POSIX_ACL, struct bch_sb, flags[0], 56, 57);
982 LE64_BITMASK(BCH_SB_USRQUOTA, struct bch_sb, flags[0], 57, 58);
983 LE64_BITMASK(BCH_SB_GRPQUOTA, struct bch_sb, flags[0], 58, 59);
984 LE64_BITMASK(BCH_SB_PRJQUOTA, struct bch_sb, flags[0], 59, 60);
986 LE64_BITMASK(BCH_SB_HAS_ERRORS, struct bch_sb, flags[0], 60, 61);
987 LE64_BITMASK(BCH_SB_HAS_TOPOLOGY_ERRORS,struct bch_sb, flags[0], 61, 62);
989 LE64_BITMASK(BCH_SB_BIG_ENDIAN, struct bch_sb, flags[0], 62, 63);
991 LE64_BITMASK(BCH_SB_STR_HASH_TYPE, struct bch_sb, flags[1], 0, 4);
992 LE64_BITMASK(BCH_SB_COMPRESSION_TYPE_LO,struct bch_sb, flags[1], 4, 8);
993 LE64_BITMASK(BCH_SB_INODE_32BIT, struct bch_sb, flags[1], 8, 9);
995 LE64_BITMASK(BCH_SB_128_BIT_MACS, struct bch_sb, flags[1], 9, 10);
996 LE64_BITMASK(BCH_SB_ENCRYPTION_TYPE, struct bch_sb, flags[1], 10, 14);
999 * Max size of an extent that may require bouncing to read or write
1000 * (checksummed, compressed): 64k
1002 LE64_BITMASK(BCH_SB_ENCODED_EXTENT_MAX_BITS,
1003 struct bch_sb, flags[1], 14, 20);
1005 LE64_BITMASK(BCH_SB_META_REPLICAS_REQ, struct bch_sb, flags[1], 20, 24);
1006 LE64_BITMASK(BCH_SB_DATA_REPLICAS_REQ, struct bch_sb, flags[1], 24, 28);
1008 LE64_BITMASK(BCH_SB_PROMOTE_TARGET, struct bch_sb, flags[1], 28, 40);
1009 LE64_BITMASK(BCH_SB_FOREGROUND_TARGET, struct bch_sb, flags[1], 40, 52);
1010 LE64_BITMASK(BCH_SB_BACKGROUND_TARGET, struct bch_sb, flags[1], 52, 64);
1012 LE64_BITMASK(BCH_SB_BACKGROUND_COMPRESSION_TYPE_LO,
1013 struct bch_sb, flags[2], 0, 4);
1014 LE64_BITMASK(BCH_SB_GC_RESERVE_BYTES, struct bch_sb, flags[2], 4, 64);
1016 LE64_BITMASK(BCH_SB_ERASURE_CODE, struct bch_sb, flags[3], 0, 16);
1017 LE64_BITMASK(BCH_SB_METADATA_TARGET, struct bch_sb, flags[3], 16, 28);
1018 LE64_BITMASK(BCH_SB_SHARD_INUMS, struct bch_sb, flags[3], 28, 29);
1019 LE64_BITMASK(BCH_SB_INODES_USE_KEY_CACHE,struct bch_sb, flags[3], 29, 30);
1020 LE64_BITMASK(BCH_SB_JOURNAL_FLUSH_DELAY,struct bch_sb, flags[3], 30, 62);
1021 LE64_BITMASK(BCH_SB_JOURNAL_FLUSH_DISABLED,struct bch_sb, flags[3], 62, 63);
1022 LE64_BITMASK(BCH_SB_JOURNAL_RECLAIM_DELAY,struct bch_sb, flags[4], 0, 32);
1023 LE64_BITMASK(BCH_SB_JOURNAL_TRANSACTION_NAMES,struct bch_sb, flags[4], 32, 33);
1024 LE64_BITMASK(BCH_SB_NOCOW, struct bch_sb, flags[4], 33, 34);
1025 LE64_BITMASK(BCH_SB_WRITE_BUFFER_SIZE, struct bch_sb, flags[4], 34, 54);
1026 LE64_BITMASK(BCH_SB_VERSION_UPGRADE, struct bch_sb, flags[4], 54, 56);
1028 LE64_BITMASK(BCH_SB_COMPRESSION_TYPE_HI,struct bch_sb, flags[4], 56, 60);
1029 LE64_BITMASK(BCH_SB_BACKGROUND_COMPRESSION_TYPE_HI,
1030 struct bch_sb, flags[4], 60, 64);
1032 LE64_BITMASK(BCH_SB_VERSION_UPGRADE_COMPLETE,
1033 struct bch_sb, flags[5], 0, 16);
1035 static inline __u64 BCH_SB_COMPRESSION_TYPE(const struct bch_sb *sb)
1037 return BCH_SB_COMPRESSION_TYPE_LO(sb) | (BCH_SB_COMPRESSION_TYPE_HI(sb) << 4);
1040 static inline void SET_BCH_SB_COMPRESSION_TYPE(struct bch_sb *sb, __u64 v)
1042 SET_BCH_SB_COMPRESSION_TYPE_LO(sb, v);
1043 SET_BCH_SB_COMPRESSION_TYPE_HI(sb, v >> 4);
1046 static inline __u64 BCH_SB_BACKGROUND_COMPRESSION_TYPE(const struct bch_sb *sb)
1048 return BCH_SB_BACKGROUND_COMPRESSION_TYPE_LO(sb) |
1049 (BCH_SB_BACKGROUND_COMPRESSION_TYPE_HI(sb) << 4);
1052 static inline void SET_BCH_SB_BACKGROUND_COMPRESSION_TYPE(struct bch_sb *sb, __u64 v)
1054 SET_BCH_SB_BACKGROUND_COMPRESSION_TYPE_LO(sb, v);
1055 SET_BCH_SB_BACKGROUND_COMPRESSION_TYPE_HI(sb, v >> 4);
1061 * journal_seq_blacklist_v3: gates BCH_SB_FIELD_journal_seq_blacklist
1062 * reflink: gates KEY_TYPE_reflink
1063 * inline_data: gates KEY_TYPE_inline_data
1064 * new_siphash: gates BCH_STR_HASH_siphash
1065 * new_extent_overwrite: gates BTREE_NODE_NEW_EXTENT_OVERWRITE
1067 #define BCH_SB_FEATURES() \
1071 x(atomic_nlink, 3) \
1073 x(journal_seq_blacklist_v3, 5) \
1077 x(new_extent_overwrite, 9) \
1078 x(incompressible, 10) \
1079 x(btree_ptr_v2, 11) \
1080 x(extents_above_btree_updates, 12) \
1081 x(btree_updates_journalled, 13) \
1082 x(reflink_inline_data, 14) \
1084 x(journal_no_flush, 16) \
1086 x(extents_across_btree_nodes, 18)
1088 #define BCH_SB_FEATURES_ALWAYS \
1089 ((1ULL << BCH_FEATURE_new_extent_overwrite)| \
1090 (1ULL << BCH_FEATURE_extents_above_btree_updates)|\
1091 (1ULL << BCH_FEATURE_btree_updates_journalled)|\
1092 (1ULL << BCH_FEATURE_alloc_v2)|\
1093 (1ULL << BCH_FEATURE_extents_across_btree_nodes))
1095 #define BCH_SB_FEATURES_ALL \
1096 (BCH_SB_FEATURES_ALWAYS| \
1097 (1ULL << BCH_FEATURE_new_siphash)| \
1098 (1ULL << BCH_FEATURE_btree_ptr_v2)| \
1099 (1ULL << BCH_FEATURE_new_varint)| \
1100 (1ULL << BCH_FEATURE_journal_no_flush))
1102 enum bch_sb_feature {
1103 #define x(f, n) BCH_FEATURE_##f,
1109 #define BCH_SB_COMPAT() \
1111 x(alloc_metadata, 1) \
1112 x(extents_above_btree_updates_done, 2) \
1113 x(bformat_overflow_done, 3)
1115 enum bch_sb_compat {
1116 #define x(f, n) BCH_COMPAT_##f,
1124 #define BCH_VERSION_UPGRADE_OPTS() \
1126 x(incompatible, 1) \
1129 enum bch_version_upgrade_opts {
1130 #define x(t, n) BCH_VERSION_UPGRADE_##t = n,
1131 BCH_VERSION_UPGRADE_OPTS()
1135 #define BCH_REPLICAS_MAX 4U
1137 #define BCH_BKEY_PTRS_MAX 16U
1139 #define BCH_ERROR_ACTIONS() \
1144 enum bch_error_actions {
1145 #define x(t, n) BCH_ON_ERROR_##t = n,
1151 #define BCH_STR_HASH_TYPES() \
1157 enum bch_str_hash_type {
1158 #define x(t, n) BCH_STR_HASH_##t = n,
1159 BCH_STR_HASH_TYPES()
1164 #define BCH_STR_HASH_OPTS() \
1169 enum bch_str_hash_opts {
1170 #define x(t, n) BCH_STR_HASH_OPT_##t = n,
1176 #define BCH_CSUM_TYPES() \
1178 x(crc32c_nonzero, 1) \
1179 x(crc64_nonzero, 2) \
1180 x(chacha20_poly1305_80, 3) \
1181 x(chacha20_poly1305_128, 4) \
1186 enum bch_csum_type {
1187 #define x(t, n) BCH_CSUM_##t = n,
1193 static const __maybe_unused unsigned bch_crc_bytes[] = {
1194 [BCH_CSUM_none] = 0,
1195 [BCH_CSUM_crc32c_nonzero] = 4,
1196 [BCH_CSUM_crc32c] = 4,
1197 [BCH_CSUM_crc64_nonzero] = 8,
1198 [BCH_CSUM_crc64] = 8,
1199 [BCH_CSUM_xxhash] = 8,
1200 [BCH_CSUM_chacha20_poly1305_80] = 10,
1201 [BCH_CSUM_chacha20_poly1305_128] = 16,
1204 static inline _Bool bch2_csum_type_is_encryption(enum bch_csum_type type)
1207 case BCH_CSUM_chacha20_poly1305_80:
1208 case BCH_CSUM_chacha20_poly1305_128:
1215 #define BCH_CSUM_OPTS() \
1221 enum bch_csum_opts {
1222 #define x(t, n) BCH_CSUM_OPT_##t = n,
1228 #define BCH_COMPRESSION_TYPES() \
1234 x(incompressible, 5)
1236 enum bch_compression_type {
1237 #define x(t, n) BCH_COMPRESSION_TYPE_##t = n,
1238 BCH_COMPRESSION_TYPES()
1240 BCH_COMPRESSION_TYPE_NR
1243 #define BCH_COMPRESSION_OPTS() \
1249 enum bch_compression_opts {
1250 #define x(t, n) BCH_COMPRESSION_OPT_##t = n,
1251 BCH_COMPRESSION_OPTS()
1253 BCH_COMPRESSION_OPT_NR
1259 * The various other data structures have their own magic numbers, which are
1260 * xored with the first part of the cache set's UUID
1263 #define BCACHE_MAGIC \
1264 UUID_INIT(0xc68573f6, 0x4e1a, 0x45ca, \
1265 0x82, 0x65, 0xf5, 0x7f, 0x48, 0xba, 0x6d, 0x81)
1266 #define BCHFS_MAGIC \
1267 UUID_INIT(0xc68573f6, 0x66ce, 0x90a9, \
1268 0xd9, 0x6a, 0x60, 0xcf, 0x80, 0x3d, 0xf7, 0xef)
1270 #define BCACHEFS_STATFS_MAGIC 0xca451a4e
1272 #define JSET_MAGIC __cpu_to_le64(0x245235c1a3625032ULL)
1273 #define BSET_MAGIC __cpu_to_le64(0x90135c78b99e07f5ULL)
1275 static inline __le64 __bch2_sb_magic(struct bch_sb *sb)
1279 memcpy(&ret, &sb->uuid, sizeof(ret));
1283 static inline __u64 __jset_magic(struct bch_sb *sb)
1285 return __le64_to_cpu(__bch2_sb_magic(sb) ^ JSET_MAGIC);
1288 static inline __u64 __bset_magic(struct bch_sb *sb)
1290 return __le64_to_cpu(__bch2_sb_magic(sb) ^ BSET_MAGIC);
1295 #define JSET_KEYS_U64s (sizeof(struct jset_entry) / sizeof(__u64))
1297 #define BCH_JSET_ENTRY_TYPES() \
1302 x(blacklist_v2, 4) \
1309 x(write_buffer_keys, 11) \
1313 #define x(f, nr) BCH_JSET_ENTRY_##f = nr,
1314 BCH_JSET_ENTRY_TYPES()
1319 static inline bool jset_entry_is_key(struct jset_entry *e)
1322 case BCH_JSET_ENTRY_btree_keys:
1323 case BCH_JSET_ENTRY_btree_root:
1324 case BCH_JSET_ENTRY_overwrite:
1325 case BCH_JSET_ENTRY_write_buffer_keys:
1333 * Journal sequence numbers can be blacklisted: bsets record the max sequence
1334 * number of all the journal entries they contain updates for, so that on
1335 * recovery we can ignore those bsets that contain index updates newer that what
1336 * made it into the journal.
1338 * This means that we can't reuse that journal_seq - we have to skip it, and
1339 * then record that we skipped it so that the next time we crash and recover we
1340 * don't think there was a missing journal entry.
1342 struct jset_entry_blacklist {
1343 struct jset_entry entry;
1347 struct jset_entry_blacklist_v2 {
1348 struct jset_entry entry;
1353 #define BCH_FS_USAGE_TYPES() \
1359 #define x(f, nr) BCH_FS_USAGE_##f = nr,
1360 BCH_FS_USAGE_TYPES()
1365 struct jset_entry_usage {
1366 struct jset_entry entry;
1370 struct jset_entry_data_usage {
1371 struct jset_entry entry;
1373 struct bch_replicas_entry_v1 r;
1376 struct jset_entry_clock {
1377 struct jset_entry entry;
1383 struct jset_entry_dev_usage_type {
1389 struct jset_entry_dev_usage {
1390 struct jset_entry entry;
1394 __le64 _buckets_ec; /* No longer used */
1395 __le64 _buckets_unavailable; /* No longer used */
1397 struct jset_entry_dev_usage_type d[];
1400 static inline unsigned jset_entry_dev_usage_nr_types(struct jset_entry_dev_usage *u)
1402 return (vstruct_bytes(&u->entry) - sizeof(struct jset_entry_dev_usage)) /
1403 sizeof(struct jset_entry_dev_usage_type);
1406 struct jset_entry_log {
1407 struct jset_entry entry;
1409 } __packed __aligned(8);
1411 struct jset_entry_datetime {
1412 struct jset_entry entry;
1414 } __packed __aligned(8);
1417 * On disk format for a journal entry:
1418 * seq is monotonically increasing; every journal entry has its own unique
1421 * last_seq is the oldest journal entry that still has keys the btree hasn't
1422 * flushed to disk yet.
1424 * version is for on disk format changes.
1427 struct bch_csum csum;
1434 __le32 u64s; /* size of d[] in u64s */
1436 __u8 encrypted_start[0];
1438 __le16 _read_clock; /* no longer used */
1439 __le16 _write_clock;
1441 /* Sequence number of oldest dirty journal entry */
1445 struct jset_entry start[0];
1447 } __packed __aligned(8);
1449 LE32_BITMASK(JSET_CSUM_TYPE, struct jset, flags, 0, 4);
1450 LE32_BITMASK(JSET_BIG_ENDIAN, struct jset, flags, 4, 5);
1451 LE32_BITMASK(JSET_NO_FLUSH, struct jset, flags, 5, 6);
1453 #define BCH_JOURNAL_BUCKETS_MIN 8
1457 enum btree_id_flags {
1458 BTREE_ID_EXTENTS = BIT(0),
1459 BTREE_ID_SNAPSHOTS = BIT(1),
1460 BTREE_ID_SNAPSHOT_FIELD = BIT(2),
1461 BTREE_ID_DATA = BIT(3),
1464 #define BCH_BTREE_IDS() \
1465 x(extents, 0, BTREE_ID_EXTENTS|BTREE_ID_SNAPSHOTS|BTREE_ID_DATA,\
1466 BIT_ULL(KEY_TYPE_whiteout)| \
1467 BIT_ULL(KEY_TYPE_error)| \
1468 BIT_ULL(KEY_TYPE_cookie)| \
1469 BIT_ULL(KEY_TYPE_extent)| \
1470 BIT_ULL(KEY_TYPE_reservation)| \
1471 BIT_ULL(KEY_TYPE_reflink_p)| \
1472 BIT_ULL(KEY_TYPE_inline_data)) \
1473 x(inodes, 1, BTREE_ID_SNAPSHOTS, \
1474 BIT_ULL(KEY_TYPE_whiteout)| \
1475 BIT_ULL(KEY_TYPE_inode)| \
1476 BIT_ULL(KEY_TYPE_inode_v2)| \
1477 BIT_ULL(KEY_TYPE_inode_v3)| \
1478 BIT_ULL(KEY_TYPE_inode_generation)) \
1479 x(dirents, 2, BTREE_ID_SNAPSHOTS, \
1480 BIT_ULL(KEY_TYPE_whiteout)| \
1481 BIT_ULL(KEY_TYPE_hash_whiteout)| \
1482 BIT_ULL(KEY_TYPE_dirent)) \
1483 x(xattrs, 3, BTREE_ID_SNAPSHOTS, \
1484 BIT_ULL(KEY_TYPE_whiteout)| \
1485 BIT_ULL(KEY_TYPE_cookie)| \
1486 BIT_ULL(KEY_TYPE_hash_whiteout)| \
1487 BIT_ULL(KEY_TYPE_xattr)) \
1489 BIT_ULL(KEY_TYPE_alloc)| \
1490 BIT_ULL(KEY_TYPE_alloc_v2)| \
1491 BIT_ULL(KEY_TYPE_alloc_v3)| \
1492 BIT_ULL(KEY_TYPE_alloc_v4)) \
1494 BIT_ULL(KEY_TYPE_quota)) \
1496 BIT_ULL(KEY_TYPE_stripe)) \
1497 x(reflink, 7, BTREE_ID_EXTENTS|BTREE_ID_DATA, \
1498 BIT_ULL(KEY_TYPE_reflink_v)| \
1499 BIT_ULL(KEY_TYPE_indirect_inline_data)) \
1500 x(subvolumes, 8, 0, \
1501 BIT_ULL(KEY_TYPE_subvolume)) \
1502 x(snapshots, 9, 0, \
1503 BIT_ULL(KEY_TYPE_snapshot)) \
1505 BIT_ULL(KEY_TYPE_set)) \
1506 x(freespace, 11, BTREE_ID_EXTENTS, \
1507 BIT_ULL(KEY_TYPE_set)) \
1508 x(need_discard, 12, 0, \
1509 BIT_ULL(KEY_TYPE_set)) \
1510 x(backpointers, 13, 0, \
1511 BIT_ULL(KEY_TYPE_backpointer)) \
1512 x(bucket_gens, 14, 0, \
1513 BIT_ULL(KEY_TYPE_bucket_gens)) \
1514 x(snapshot_trees, 15, 0, \
1515 BIT_ULL(KEY_TYPE_snapshot_tree)) \
1516 x(deleted_inodes, 16, BTREE_ID_SNAPSHOT_FIELD, \
1517 BIT_ULL(KEY_TYPE_set)) \
1518 x(logged_ops, 17, 0, \
1519 BIT_ULL(KEY_TYPE_logged_op_truncate)| \
1520 BIT_ULL(KEY_TYPE_logged_op_finsert)) \
1521 x(rebalance_work, 18, BTREE_ID_SNAPSHOT_FIELD, \
1522 BIT_ULL(KEY_TYPE_set)|BIT_ULL(KEY_TYPE_cookie)) \
1523 x(subvolume_children, 19, 0, \
1524 BIT_ULL(KEY_TYPE_set))
1527 #define x(name, nr, ...) BTREE_ID_##name = nr,
1533 #define BTREE_MAX_DEPTH 4U
1540 * On disk a btree node is a list/log of these; within each set the keys are
1547 * Highest journal entry this bset contains keys for.
1548 * If on recovery we don't see that journal entry, this bset is ignored:
1549 * this allows us to preserve the order of all index updates after a
1550 * crash, since the journal records a total order of all index updates
1551 * and anything that didn't make it to the journal doesn't get used.
1557 __le16 u64s; /* count of d[] in u64s */
1559 struct bkey_packed start[0];
1561 } __packed __aligned(8);
1563 LE32_BITMASK(BSET_CSUM_TYPE, struct bset, flags, 0, 4);
1565 LE32_BITMASK(BSET_BIG_ENDIAN, struct bset, flags, 4, 5);
1566 LE32_BITMASK(BSET_SEPARATE_WHITEOUTS,
1567 struct bset, flags, 5, 6);
1569 /* Sector offset within the btree node: */
1570 LE32_BITMASK(BSET_OFFSET, struct bset, flags, 16, 32);
1573 struct bch_csum csum;
1576 /* this flags field is encrypted, unlike bset->flags: */
1579 /* Closed interval: */
1580 struct bpos min_key;
1581 struct bpos max_key;
1582 struct bch_extent_ptr _ptr; /* not used anymore */
1583 struct bkey_format format;
1594 } __packed __aligned(8);
1596 LE64_BITMASK(BTREE_NODE_ID_LO, struct btree_node, flags, 0, 4);
1597 LE64_BITMASK(BTREE_NODE_LEVEL, struct btree_node, flags, 4, 8);
1598 LE64_BITMASK(BTREE_NODE_NEW_EXTENT_OVERWRITE,
1599 struct btree_node, flags, 8, 9);
1600 LE64_BITMASK(BTREE_NODE_ID_HI, struct btree_node, flags, 9, 25);
1602 LE64_BITMASK(BTREE_NODE_SEQ, struct btree_node, flags, 32, 64);
1604 static inline __u64 BTREE_NODE_ID(struct btree_node *n)
1606 return BTREE_NODE_ID_LO(n) | (BTREE_NODE_ID_HI(n) << 4);
1609 static inline void SET_BTREE_NODE_ID(struct btree_node *n, __u64 v)
1611 SET_BTREE_NODE_ID_LO(n, v);
1612 SET_BTREE_NODE_ID_HI(n, v >> 4);
1615 struct btree_node_entry {
1616 struct bch_csum csum;
1626 } __packed __aligned(8);
1628 #endif /* _BCACHEFS_FORMAT_H */