1 /* SPDX-License-Identifier: GPL-2.0 */
2 #ifndef _BCACHEFS_FORMAT_H
3 #define _BCACHEFS_FORMAT_H
4
5 /*
6 * bcachefs on disk data structures
7 *
8 * OVERVIEW:
9 *
10 * There are three main types of on disk data structures in bcachefs (this is
11 * reduced from 5 in bcache)
12 *
13 * - superblock
14 * - journal
15 * - btree
16 *
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.
20 *
21 * SUPERBLOCK:
22 *
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
26 * roots.
27 *
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.
30 *
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.
35 *
36 * JOURNAL:
37 *
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.
41 *
42 * The journal also contains entry types for the btree roots, and blacklisted
43 * journal sequence numbers (see journal_seq_blacklist.c).
44 *
45 * BTREE:
46 *
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
50 * subsequent writes.
51 *
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
56 * path.
57 *
58 * BTREE KEYS (struct bkey):
59 *
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.
63 *
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).
68 *
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.
73 */
74
75 #include <asm/types.h>
76 #include <asm/byteorder.h>
77 #include <linux/kernel.h>
78 #include <linux/uuid.h>
79 #include "vstructs.h"
80
81 #ifdef __KERNEL__
82 typedef uuid_t __uuid_t;
83 #endif
84
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); \
88 \
89 static inline __u64 name(const type *k) \
90 { \
91 return (k->field >> offset) & ~(~0ULL << (end - offset)); \
92 } \
93 \
94 static inline void SET_##name(type *k, __u64 v) \
95 { \
96 k->field &= ~(~(~0ULL << (end - offset)) << offset); \
97 k->field |= (v & ~(~0ULL << (end - offset))) << offset; \
98 }
99
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;\
104 \
105 static inline __u64 name(const type *k) \
106 { \
107 return (__le##_bits##_to_cpu(k->field) >> offset) & \
108 ~(~0ULL << (end - offset)); \
109 } \
110 \
111 static inline void SET_##name(type *k, __u64 v) \
112 { \
113 __u##_bits new = __le##_bits##_to_cpu(k->field); \
114 \
115 new &= ~(~(~0ULL << (end - offset)) << offset); \
116 new |= (v & ~(~0ULL << (end - offset))) << offset; \
117 k->field = __cpu_to_le##_bits(new); \
118 }
119
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)
123
124 struct bkey_format {
125 __u8 key_u64s;
126 __u8 nr_fields;
127 /* One unused slot for now: */
128 __u8 bits_per_field[6];
129 __le64 field_offset[6];
130 };
131
132 /* Btree keys - all units are in sectors */
133
134 struct bpos {
135 /*
136 * Word order matches machine byte order - btree code treats a bpos as a
137 * single large integer, for search/comparison purposes
138 *
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:
142 */
143 #if __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__
144 __u32 snapshot;
145 __u64 offset;
146 __u64 inode;
147 #elif __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__
148 __u64 inode;
149 __u64 offset; /* Points to end of extent - sectors */
150 __u32 snapshot;
151 #else
152 #error edit for your odd byteorder.
153 #endif
154 } __packed
155 #if __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__
156 __aligned(4)
157 #endif
158 ;
159
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)
164
SPOS(__u64 inode,__u64 offset,__u32 snapshot)165 static inline struct bpos SPOS(__u64 inode, __u64 offset, __u32 snapshot)
166 {
167 return (struct bpos) {
168 .inode = inode,
169 .offset = offset,
170 .snapshot = snapshot,
171 };
172 }
173
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)
178
179 /* Empty placeholder struct, for container_of() */
180 struct bch_val {
181 __u64 __nothing[0];
182 };
183
184 struct bversion {
185 #if __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__
186 __u64 lo;
187 __u32 hi;
188 #elif __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__
189 __u32 hi;
190 __u64 lo;
191 #endif
192 } __packed __aligned(4);
193
194 struct bkey {
195 /* Size of combined key and value, in u64s */
196 __u8 u64s;
197
198 /* Format of key (0 for format local to btree node) */
199 #if defined(__LITTLE_ENDIAN_BITFIELD)
200 __u8 format:7,
201 needs_whiteout:1;
202 #elif defined (__BIG_ENDIAN_BITFIELD)
203 __u8 needs_whiteout:1,
204 format:7;
205 #else
206 #error edit for your odd byteorder.
207 #endif
208
209 /* Type of the value */
210 __u8 type;
211
212 #if __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__
213 __u8 pad[1];
214
215 struct bversion version;
216 __u32 size; /* extent size, in sectors */
217 struct bpos p;
218 #elif __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__
219 struct bpos p;
220 __u32 size; /* extent size, in sectors */
221 struct bversion version;
222
223 __u8 pad[1];
224 #endif
225 } __packed __aligned(8);
226
227 struct bkey_packed {
228 __u64 _data[0];
229
230 /* Size of combined key and value, in u64s */
231 __u8 u64s;
232
233 /* Format of key (0 for format local to btree node) */
234
235 /*
236 * XXX: next incompat on disk format change, switch format and
237 * needs_whiteout - bkey_packed() will be cheaper if format is the high
238 * bits of the bitfield
239 */
240 #if defined(__LITTLE_ENDIAN_BITFIELD)
241 __u8 format:7,
242 needs_whiteout:1;
243 #elif defined (__BIG_ENDIAN_BITFIELD)
244 __u8 needs_whiteout:1,
245 format:7;
246 #endif
247
248 /* Type of the value */
249 __u8 type;
250 __u8 key_start[0];
251
252 /*
253 * We copy bkeys with struct assignment in various places, and while
254 * that shouldn't be done with packed bkeys we can't disallow it in C,
255 * and it's legal to cast a bkey to a bkey_packed - so padding it out
256 * to the same size as struct bkey should hopefully be safest.
257 */
258 __u8 pad[sizeof(struct bkey) - 3];
259 } __packed __aligned(8);
260
261 typedef struct {
262 __le64 lo;
263 __le64 hi;
264 } bch_le128;
265
266 #define BKEY_U64s (sizeof(struct bkey) / sizeof(__u64))
267 #define BKEY_U64s_MAX U8_MAX
268 #define BKEY_VAL_U64s_MAX (BKEY_U64s_MAX - BKEY_U64s)
269
270 #define KEY_PACKED_BITS_START 24
271
272 #define KEY_FORMAT_LOCAL_BTREE 0
273 #define KEY_FORMAT_CURRENT 1
274
275 enum bch_bkey_fields {
276 BKEY_FIELD_INODE,
277 BKEY_FIELD_OFFSET,
278 BKEY_FIELD_SNAPSHOT,
279 BKEY_FIELD_SIZE,
280 BKEY_FIELD_VERSION_HI,
281 BKEY_FIELD_VERSION_LO,
282 BKEY_NR_FIELDS,
283 };
284
285 #define bkey_format_field(name, field) \
286 [BKEY_FIELD_##name] = (sizeof(((struct bkey *) NULL)->field) * 8)
287
288 #define BKEY_FORMAT_CURRENT \
289 ((struct bkey_format) { \
290 .key_u64s = BKEY_U64s, \
291 .nr_fields = BKEY_NR_FIELDS, \
292 .bits_per_field = { \
293 bkey_format_field(INODE, p.inode), \
294 bkey_format_field(OFFSET, p.offset), \
295 bkey_format_field(SNAPSHOT, p.snapshot), \
296 bkey_format_field(SIZE, size), \
297 bkey_format_field(VERSION_HI, version.hi), \
298 bkey_format_field(VERSION_LO, version.lo), \
299 }, \
300 })
301
302 /* bkey with inline value */
303 struct bkey_i {
304 __u64 _data[0];
305
306 struct bkey k;
307 struct bch_val v;
308 };
309
310 #define POS_KEY(_pos) \
311 ((struct bkey) { \
312 .u64s = BKEY_U64s, \
313 .format = KEY_FORMAT_CURRENT, \
314 .p = _pos, \
315 })
316
317 #define KEY(_inode, _offset, _size) \
318 ((struct bkey) { \
319 .u64s = BKEY_U64s, \
320 .format = KEY_FORMAT_CURRENT, \
321 .p = POS(_inode, _offset), \
322 .size = _size, \
323 })
324
bkey_init(struct bkey * k)325 static inline void bkey_init(struct bkey *k)
326 {
327 *k = KEY(0, 0, 0);
328 }
329
330 #define bkey_bytes(_k) ((_k)->u64s * sizeof(__u64))
331
332 #define __BKEY_PADDED(key, pad) \
333 struct bkey_i key; __u64 key ## _pad[pad]
334
335 /*
336 * - DELETED keys are used internally to mark keys that should be ignored but
337 * override keys in composition order. Their version number is ignored.
338 *
339 * - DISCARDED keys indicate that the data is all 0s because it has been
340 * discarded. DISCARDs may have a version; if the version is nonzero the key
341 * will be persistent, otherwise the key will be dropped whenever the btree
342 * node is rewritten (like DELETED keys).
343 *
344 * - ERROR: any read of the data returns a read error, as the data was lost due
345 * to a failing device. Like DISCARDED keys, they can be removed (overridden)
346 * by new writes or cluster-wide GC. Node repair can also overwrite them with
347 * the same or a more recent version number, but not with an older version
348 * number.
349 *
350 * - WHITEOUT: for hash table btrees
351 */
352 #define BCH_BKEY_TYPES() \
353 x(deleted, 0) \
354 x(whiteout, 1) \
355 x(error, 2) \
356 x(cookie, 3) \
357 x(hash_whiteout, 4) \
358 x(btree_ptr, 5) \
359 x(extent, 6) \
360 x(reservation, 7) \
361 x(inode, 8) \
362 x(inode_generation, 9) \
363 x(dirent, 10) \
364 x(xattr, 11) \
365 x(alloc, 12) \
366 x(quota, 13) \
367 x(stripe, 14) \
368 x(reflink_p, 15) \
369 x(reflink_v, 16) \
370 x(inline_data, 17) \
371 x(btree_ptr_v2, 18) \
372 x(indirect_inline_data, 19) \
373 x(alloc_v2, 20) \
374 x(subvolume, 21) \
375 x(snapshot, 22) \
376 x(inode_v2, 23) \
377 x(alloc_v3, 24) \
378 x(set, 25) \
379 x(lru, 26) \
380 x(alloc_v4, 27) \
381 x(backpointer, 28) \
382 x(inode_v3, 29) \
383 x(bucket_gens, 30) \
384 x(snapshot_tree, 31) \
385 x(logged_op_truncate, 32) \
386 x(logged_op_finsert, 33)
387
388 enum bch_bkey_type {
389 #define x(name, nr) KEY_TYPE_##name = nr,
390 BCH_BKEY_TYPES()
391 #undef x
392 KEY_TYPE_MAX,
393 };
394
395 struct bch_deleted {
396 struct bch_val v;
397 };
398
399 struct bch_whiteout {
400 struct bch_val v;
401 };
402
403 struct bch_error {
404 struct bch_val v;
405 };
406
407 struct bch_cookie {
408 struct bch_val v;
409 __le64 cookie;
410 };
411
412 struct bch_hash_whiteout {
413 struct bch_val v;
414 };
415
416 struct bch_set {
417 struct bch_val v;
418 };
419
420 /* 128 bits, sufficient for cryptographic MACs: */
421 struct bch_csum {
422 __le64 lo;
423 __le64 hi;
424 } __packed __aligned(8);
425
426 struct bch_backpointer {
427 struct bch_val v;
428 __u8 btree_id;
429 __u8 level;
430 __u8 data_type;
431 __u64 bucket_offset:40;
432 __u32 bucket_len;
433 struct bpos pos;
434 } __packed __aligned(8);
435
436 /* LRU btree: */
437
438 struct bch_lru {
439 struct bch_val v;
440 __le64 idx;
441 } __packed __aligned(8);
442
443 #define LRU_ID_STRIPES (1U << 16)
444
445 /* Optional/variable size superblock sections: */
446
447 struct bch_sb_field {
448 __u64 _data[0];
449 __le32 u64s;
450 __le32 type;
451 };
452
453 #define BCH_SB_FIELDS() \
454 x(journal, 0) \
455 x(members_v1, 1) \
456 x(crypt, 2) \
457 x(replicas_v0, 3) \
458 x(quota, 4) \
459 x(disk_groups, 5) \
460 x(clean, 6) \
461 x(replicas, 7) \
462 x(journal_seq_blacklist, 8) \
463 x(journal_v2, 9) \
464 x(counters, 10) \
465 x(members_v2, 11) \
466 x(errors, 12) \
467 x(ext, 13) \
468 x(downgrade, 14)
469
470 #include "alloc_background_format.h"
471 #include "extents_format.h"
472 #include "reflink_format.h"
473 #include "ec_format.h"
474 #include "inode_format.h"
475 #include "dirent_format.h"
476 #include "xattr_format.h"
477 #include "quota_format.h"
478 #include "logged_ops_format.h"
479 #include "snapshot_format.h"
480 #include "subvolume_format.h"
481 #include "sb-counters_format.h"
482
483 enum bch_sb_field_type {
484 #define x(f, nr) BCH_SB_FIELD_##f = nr,
485 BCH_SB_FIELDS()
486 #undef x
487 BCH_SB_FIELD_NR
488 };
489
490 /*
491 * Most superblock fields are replicated in all device's superblocks - a few are
492 * not:
493 */
494 #define BCH_SINGLE_DEVICE_SB_FIELDS \
495 ((1U << BCH_SB_FIELD_journal)| \
496 (1U << BCH_SB_FIELD_journal_v2))
497
498 /* BCH_SB_FIELD_journal: */
499
500 struct bch_sb_field_journal {
501 struct bch_sb_field field;
502 __le64 buckets[];
503 };
504
505 struct bch_sb_field_journal_v2 {
506 struct bch_sb_field field;
507
508 struct bch_sb_field_journal_v2_entry {
509 __le64 start;
510 __le64 nr;
511 } d[];
512 };
513
514 /* BCH_SB_FIELD_members_v1: */
515
516 #define BCH_MIN_NR_NBUCKETS (1 << 6)
517
518 #define BCH_IOPS_MEASUREMENTS() \
519 x(seqread, 0) \
520 x(seqwrite, 1) \
521 x(randread, 2) \
522 x(randwrite, 3)
523
524 enum bch_iops_measurement {
525 #define x(t, n) BCH_IOPS_##t = n,
526 BCH_IOPS_MEASUREMENTS()
527 #undef x
528 BCH_IOPS_NR
529 };
530
531 #define BCH_MEMBER_ERROR_TYPES() \
532 x(read, 0) \
533 x(write, 1) \
534 x(checksum, 2)
535
536 enum bch_member_error_type {
537 #define x(t, n) BCH_MEMBER_ERROR_##t = n,
538 BCH_MEMBER_ERROR_TYPES()
539 #undef x
540 BCH_MEMBER_ERROR_NR
541 };
542
543 struct bch_member {
544 __uuid_t uuid;
545 __le64 nbuckets; /* device size */
546 __le16 first_bucket; /* index of first bucket used */
547 __le16 bucket_size; /* sectors */
548 __le32 pad;
549 __le64 last_mount; /* time_t */
550
551 __le64 flags;
552 __le32 iops[4];
553 __le64 errors[BCH_MEMBER_ERROR_NR];
554 __le64 errors_at_reset[BCH_MEMBER_ERROR_NR];
555 __le64 errors_reset_time;
556 __le64 seq;
557 };
558
559 #define BCH_MEMBER_V1_BYTES 56
560
561 LE64_BITMASK(BCH_MEMBER_STATE, struct bch_member, flags, 0, 4)
562 /* 4-14 unused, was TIER, HAS_(META)DATA, REPLACEMENT */
563 LE64_BITMASK(BCH_MEMBER_DISCARD, struct bch_member, flags, 14, 15)
564 LE64_BITMASK(BCH_MEMBER_DATA_ALLOWED, struct bch_member, flags, 15, 20)
565 LE64_BITMASK(BCH_MEMBER_GROUP, struct bch_member, flags, 20, 28)
566 LE64_BITMASK(BCH_MEMBER_DURABILITY, struct bch_member, flags, 28, 30)
567 LE64_BITMASK(BCH_MEMBER_FREESPACE_INITIALIZED,
568 struct bch_member, flags, 30, 31)
569
570 #if 0
571 LE64_BITMASK(BCH_MEMBER_NR_READ_ERRORS, struct bch_member, flags[1], 0, 20);
572 LE64_BITMASK(BCH_MEMBER_NR_WRITE_ERRORS,struct bch_member, flags[1], 20, 40);
573 #endif
574
575 #define BCH_MEMBER_STATES() \
576 x(rw, 0) \
577 x(ro, 1) \
578 x(failed, 2) \
579 x(spare, 3)
580
581 enum bch_member_state {
582 #define x(t, n) BCH_MEMBER_STATE_##t = n,
583 BCH_MEMBER_STATES()
584 #undef x
585 BCH_MEMBER_STATE_NR
586 };
587
588 struct bch_sb_field_members_v1 {
589 struct bch_sb_field field;
590 struct bch_member _members[]; //Members are now variable size
591 };
592
593 struct bch_sb_field_members_v2 {
594 struct bch_sb_field field;
595 __le16 member_bytes; //size of single member entry
596 u8 pad[6];
597 struct bch_member _members[];
598 };
599
600 /* BCH_SB_FIELD_crypt: */
601
602 struct nonce {
603 __le32 d[4];
604 };
605
606 struct bch_key {
607 __le64 key[4];
608 };
609
610 #define BCH_KEY_MAGIC \
611 (((__u64) 'b' << 0)|((__u64) 'c' << 8)| \
612 ((__u64) 'h' << 16)|((__u64) '*' << 24)| \
613 ((__u64) '*' << 32)|((__u64) 'k' << 40)| \
614 ((__u64) 'e' << 48)|((__u64) 'y' << 56))
615
616 struct bch_encrypted_key {
617 __le64 magic;
618 struct bch_key key;
619 };
620
621 /*
622 * If this field is present in the superblock, it stores an encryption key which
623 * is used encrypt all other data/metadata. The key will normally be encrypted
624 * with the key userspace provides, but if encryption has been turned off we'll
625 * just store the master key unencrypted in the superblock so we can access the
626 * previously encrypted data.
627 */
628 struct bch_sb_field_crypt {
629 struct bch_sb_field field;
630
631 __le64 flags;
632 __le64 kdf_flags;
633 struct bch_encrypted_key key;
634 };
635
636 LE64_BITMASK(BCH_CRYPT_KDF_TYPE, struct bch_sb_field_crypt, flags, 0, 4);
637
638 enum bch_kdf_types {
639 BCH_KDF_SCRYPT = 0,
640 BCH_KDF_NR = 1,
641 };
642
643 /* stored as base 2 log of scrypt params: */
644 LE64_BITMASK(BCH_KDF_SCRYPT_N, struct bch_sb_field_crypt, kdf_flags, 0, 16);
645 LE64_BITMASK(BCH_KDF_SCRYPT_R, struct bch_sb_field_crypt, kdf_flags, 16, 32);
646 LE64_BITMASK(BCH_KDF_SCRYPT_P, struct bch_sb_field_crypt, kdf_flags, 32, 48);
647
648 /* BCH_SB_FIELD_replicas: */
649
650 #define BCH_DATA_TYPES() \
651 x(free, 0) \
652 x(sb, 1) \
653 x(journal, 2) \
654 x(btree, 3) \
655 x(user, 4) \
656 x(cached, 5) \
657 x(parity, 6) \
658 x(stripe, 7) \
659 x(need_gc_gens, 8) \
660 x(need_discard, 9)
661
662 enum bch_data_type {
663 #define x(t, n) BCH_DATA_##t,
664 BCH_DATA_TYPES()
665 #undef x
666 BCH_DATA_NR
667 };
668
data_type_is_empty(enum bch_data_type type)669 static inline bool data_type_is_empty(enum bch_data_type type)
670 {
671 switch (type) {
672 case BCH_DATA_free:
673 case BCH_DATA_need_gc_gens:
674 case BCH_DATA_need_discard:
675 return true;
676 default:
677 return false;
678 }
679 }
680
data_type_is_hidden(enum bch_data_type type)681 static inline bool data_type_is_hidden(enum bch_data_type type)
682 {
683 switch (type) {
684 case BCH_DATA_sb:
685 case BCH_DATA_journal:
686 return true;
687 default:
688 return false;
689 }
690 }
691
692 struct bch_replicas_entry_v0 {
693 __u8 data_type;
694 __u8 nr_devs;
695 __u8 devs[];
696 } __packed;
697
698 struct bch_sb_field_replicas_v0 {
699 struct bch_sb_field field;
700 struct bch_replicas_entry_v0 entries[];
701 } __packed __aligned(8);
702
703 struct bch_replicas_entry_v1 {
704 __u8 data_type;
705 __u8 nr_devs;
706 __u8 nr_required;
707 __u8 devs[];
708 } __packed;
709
710 #define replicas_entry_bytes(_i) \
711 (offsetof(typeof(*(_i)), devs) + (_i)->nr_devs)
712
713 struct bch_sb_field_replicas {
714 struct bch_sb_field field;
715 struct bch_replicas_entry_v1 entries[];
716 } __packed __aligned(8);
717
718 /* BCH_SB_FIELD_disk_groups: */
719
720 #define BCH_SB_LABEL_SIZE 32
721
722 struct bch_disk_group {
723 __u8 label[BCH_SB_LABEL_SIZE];
724 __le64 flags[2];
725 } __packed __aligned(8);
726
727 LE64_BITMASK(BCH_GROUP_DELETED, struct bch_disk_group, flags[0], 0, 1)
728 LE64_BITMASK(BCH_GROUP_DATA_ALLOWED, struct bch_disk_group, flags[0], 1, 6)
729 LE64_BITMASK(BCH_GROUP_PARENT, struct bch_disk_group, flags[0], 6, 24)
730
731 struct bch_sb_field_disk_groups {
732 struct bch_sb_field field;
733 struct bch_disk_group entries[];
734 } __packed __aligned(8);
735
736 /*
737 * On clean shutdown, store btree roots and current journal sequence number in
738 * the superblock:
739 */
740 struct jset_entry {
741 __le16 u64s;
742 __u8 btree_id;
743 __u8 level;
744 __u8 type; /* designates what this jset holds */
745 __u8 pad[3];
746
747 struct bkey_i start[0];
748 __u64 _data[];
749 };
750
751 struct bch_sb_field_clean {
752 struct bch_sb_field field;
753
754 __le32 flags;
755 __le16 _read_clock; /* no longer used */
756 __le16 _write_clock;
757 __le64 journal_seq;
758
759 struct jset_entry start[0];
760 __u64 _data[];
761 };
762
763 struct journal_seq_blacklist_entry {
764 __le64 start;
765 __le64 end;
766 };
767
768 struct bch_sb_field_journal_seq_blacklist {
769 struct bch_sb_field field;
770 struct journal_seq_blacklist_entry start[];
771 };
772
773 struct bch_sb_field_errors {
774 struct bch_sb_field field;
775 struct bch_sb_field_error_entry {
776 __le64 v;
777 __le64 last_error_time;
778 } entries[];
779 };
780
781 LE64_BITMASK(BCH_SB_ERROR_ENTRY_ID, struct bch_sb_field_error_entry, v, 0, 16);
782 LE64_BITMASK(BCH_SB_ERROR_ENTRY_NR, struct bch_sb_field_error_entry, v, 16, 64);
783
784 struct bch_sb_field_ext {
785 struct bch_sb_field field;
786 __le64 recovery_passes_required[2];
787 __le64 errors_silent[8];
788 };
789
790 struct bch_sb_field_downgrade_entry {
791 __le16 version;
792 __le64 recovery_passes[2];
793 __le16 nr_errors;
794 __le16 errors[] __counted_by(nr_errors);
795 } __packed __aligned(2);
796
797 struct bch_sb_field_downgrade {
798 struct bch_sb_field field;
799 struct bch_sb_field_downgrade_entry entries[];
800 };
801
802 /* Superblock: */
803
804 /*
805 * New versioning scheme:
806 * One common version number for all on disk data structures - superblock, btree
807 * nodes, journal entries
808 */
809 #define BCH_VERSION_MAJOR(_v) ((__u16) ((_v) >> 10))
810 #define BCH_VERSION_MINOR(_v) ((__u16) ((_v) & ~(~0U << 10)))
811 #define BCH_VERSION(_major, _minor) (((_major) << 10)|(_minor) << 0)
812
813 /*
814 * field 1: version name
815 * field 2: BCH_VERSION(major, minor)
816 * field 3: recovery passess required on upgrade
817 */
818 #define BCH_METADATA_VERSIONS() \
819 x(bkey_renumber, BCH_VERSION(0, 10)) \
820 x(inode_btree_change, BCH_VERSION(0, 11)) \
821 x(snapshot, BCH_VERSION(0, 12)) \
822 x(inode_backpointers, BCH_VERSION(0, 13)) \
823 x(btree_ptr_sectors_written, BCH_VERSION(0, 14)) \
824 x(snapshot_2, BCH_VERSION(0, 15)) \
825 x(reflink_p_fix, BCH_VERSION(0, 16)) \
826 x(subvol_dirent, BCH_VERSION(0, 17)) \
827 x(inode_v2, BCH_VERSION(0, 18)) \
828 x(freespace, BCH_VERSION(0, 19)) \
829 x(alloc_v4, BCH_VERSION(0, 20)) \
830 x(new_data_types, BCH_VERSION(0, 21)) \
831 x(backpointers, BCH_VERSION(0, 22)) \
832 x(inode_v3, BCH_VERSION(0, 23)) \
833 x(unwritten_extents, BCH_VERSION(0, 24)) \
834 x(bucket_gens, BCH_VERSION(0, 25)) \
835 x(lru_v2, BCH_VERSION(0, 26)) \
836 x(fragmentation_lru, BCH_VERSION(0, 27)) \
837 x(no_bps_in_alloc_keys, BCH_VERSION(0, 28)) \
838 x(snapshot_trees, BCH_VERSION(0, 29)) \
839 x(major_minor, BCH_VERSION(1, 0)) \
840 x(snapshot_skiplists, BCH_VERSION(1, 1)) \
841 x(deleted_inodes, BCH_VERSION(1, 2)) \
842 x(rebalance_work, BCH_VERSION(1, 3)) \
843 x(member_seq, BCH_VERSION(1, 4))
844
845 enum bcachefs_metadata_version {
846 bcachefs_metadata_version_min = 9,
847 #define x(t, n) bcachefs_metadata_version_##t = n,
848 BCH_METADATA_VERSIONS()
849 #undef x
850 bcachefs_metadata_version_max
851 };
852
853 static const __maybe_unused
854 unsigned bcachefs_metadata_required_upgrade_below = bcachefs_metadata_version_rebalance_work;
855
856 #define bcachefs_metadata_version_current (bcachefs_metadata_version_max - 1)
857
858 #define BCH_SB_SECTOR 8
859 #define BCH_SB_MEMBERS_MAX 64 /* XXX kill */
860
861 struct bch_sb_layout {
862 __uuid_t magic; /* bcachefs superblock UUID */
863 __u8 layout_type;
864 __u8 sb_max_size_bits; /* base 2 of 512 byte sectors */
865 __u8 nr_superblocks;
866 __u8 pad[5];
867 __le64 sb_offset[61];
868 } __packed __aligned(8);
869
870 #define BCH_SB_LAYOUT_SECTOR 7
871
872 /*
873 * @offset - sector where this sb was written
874 * @version - on disk format version
875 * @version_min - Oldest metadata version this filesystem contains; so we can
876 * safely drop compatibility code and refuse to mount filesystems
877 * we'd need it for
878 * @magic - identifies as a bcachefs superblock (BCHFS_MAGIC)
879 * @seq - incremented each time superblock is written
880 * @uuid - used for generating various magic numbers and identifying
881 * member devices, never changes
882 * @user_uuid - user visible UUID, may be changed
883 * @label - filesystem label
884 * @seq - identifies most recent superblock, incremented each time
885 * superblock is written
886 * @features - enabled incompatible features
887 */
888 struct bch_sb {
889 struct bch_csum csum;
890 __le16 version;
891 __le16 version_min;
892 __le16 pad[2];
893 __uuid_t magic;
894 __uuid_t uuid;
895 __uuid_t user_uuid;
896 __u8 label[BCH_SB_LABEL_SIZE];
897 __le64 offset;
898 __le64 seq;
899
900 __le16 block_size;
901 __u8 dev_idx;
902 __u8 nr_devices;
903 __le32 u64s;
904
905 __le64 time_base_lo;
906 __le32 time_base_hi;
907 __le32 time_precision;
908
909 __le64 flags[7];
910 __le64 write_time;
911 __le64 features[2];
912 __le64 compat[2];
913
914 struct bch_sb_layout layout;
915
916 struct bch_sb_field start[0];
917 __le64 _data[];
918 } __packed __aligned(8);
919
920 /*
921 * Flags:
922 * BCH_SB_INITALIZED - set on first mount
923 * BCH_SB_CLEAN - did we shut down cleanly? Just a hint, doesn't affect
924 * behaviour of mount/recovery path:
925 * BCH_SB_INODE_32BIT - limit inode numbers to 32 bits
926 * BCH_SB_128_BIT_MACS - 128 bit macs instead of 80
927 * BCH_SB_ENCRYPTION_TYPE - if nonzero encryption is enabled; overrides
928 * DATA/META_CSUM_TYPE. Also indicates encryption
929 * algorithm in use, if/when we get more than one
930 */
931
932 LE16_BITMASK(BCH_SB_BLOCK_SIZE, struct bch_sb, block_size, 0, 16);
933
934 LE64_BITMASK(BCH_SB_INITIALIZED, struct bch_sb, flags[0], 0, 1);
935 LE64_BITMASK(BCH_SB_CLEAN, struct bch_sb, flags[0], 1, 2);
936 LE64_BITMASK(BCH_SB_CSUM_TYPE, struct bch_sb, flags[0], 2, 8);
937 LE64_BITMASK(BCH_SB_ERROR_ACTION, struct bch_sb, flags[0], 8, 12);
938
939 LE64_BITMASK(BCH_SB_BTREE_NODE_SIZE, struct bch_sb, flags[0], 12, 28);
940
941 LE64_BITMASK(BCH_SB_GC_RESERVE, struct bch_sb, flags[0], 28, 33);
942 LE64_BITMASK(BCH_SB_ROOT_RESERVE, struct bch_sb, flags[0], 33, 40);
943
944 LE64_BITMASK(BCH_SB_META_CSUM_TYPE, struct bch_sb, flags[0], 40, 44);
945 LE64_BITMASK(BCH_SB_DATA_CSUM_TYPE, struct bch_sb, flags[0], 44, 48);
946
947 LE64_BITMASK(BCH_SB_META_REPLICAS_WANT, struct bch_sb, flags[0], 48, 52);
948 LE64_BITMASK(BCH_SB_DATA_REPLICAS_WANT, struct bch_sb, flags[0], 52, 56);
949
950 LE64_BITMASK(BCH_SB_POSIX_ACL, struct bch_sb, flags[0], 56, 57);
951 LE64_BITMASK(BCH_SB_USRQUOTA, struct bch_sb, flags[0], 57, 58);
952 LE64_BITMASK(BCH_SB_GRPQUOTA, struct bch_sb, flags[0], 58, 59);
953 LE64_BITMASK(BCH_SB_PRJQUOTA, struct bch_sb, flags[0], 59, 60);
954
955 LE64_BITMASK(BCH_SB_HAS_ERRORS, struct bch_sb, flags[0], 60, 61);
956 LE64_BITMASK(BCH_SB_HAS_TOPOLOGY_ERRORS,struct bch_sb, flags[0], 61, 62);
957
958 LE64_BITMASK(BCH_SB_BIG_ENDIAN, struct bch_sb, flags[0], 62, 63);
959
960 LE64_BITMASK(BCH_SB_STR_HASH_TYPE, struct bch_sb, flags[1], 0, 4);
961 LE64_BITMASK(BCH_SB_COMPRESSION_TYPE_LO,struct bch_sb, flags[1], 4, 8);
962 LE64_BITMASK(BCH_SB_INODE_32BIT, struct bch_sb, flags[1], 8, 9);
963
964 LE64_BITMASK(BCH_SB_128_BIT_MACS, struct bch_sb, flags[1], 9, 10);
965 LE64_BITMASK(BCH_SB_ENCRYPTION_TYPE, struct bch_sb, flags[1], 10, 14);
966
967 /*
968 * Max size of an extent that may require bouncing to read or write
969 * (checksummed, compressed): 64k
970 */
971 LE64_BITMASK(BCH_SB_ENCODED_EXTENT_MAX_BITS,
972 struct bch_sb, flags[1], 14, 20);
973
974 LE64_BITMASK(BCH_SB_META_REPLICAS_REQ, struct bch_sb, flags[1], 20, 24);
975 LE64_BITMASK(BCH_SB_DATA_REPLICAS_REQ, struct bch_sb, flags[1], 24, 28);
976
977 LE64_BITMASK(BCH_SB_PROMOTE_TARGET, struct bch_sb, flags[1], 28, 40);
978 LE64_BITMASK(BCH_SB_FOREGROUND_TARGET, struct bch_sb, flags[1], 40, 52);
979 LE64_BITMASK(BCH_SB_BACKGROUND_TARGET, struct bch_sb, flags[1], 52, 64);
980
981 LE64_BITMASK(BCH_SB_BACKGROUND_COMPRESSION_TYPE_LO,
982 struct bch_sb, flags[2], 0, 4);
983 LE64_BITMASK(BCH_SB_GC_RESERVE_BYTES, struct bch_sb, flags[2], 4, 64);
984
985 LE64_BITMASK(BCH_SB_ERASURE_CODE, struct bch_sb, flags[3], 0, 16);
986 LE64_BITMASK(BCH_SB_METADATA_TARGET, struct bch_sb, flags[3], 16, 28);
987 LE64_BITMASK(BCH_SB_SHARD_INUMS, struct bch_sb, flags[3], 28, 29);
988 LE64_BITMASK(BCH_SB_INODES_USE_KEY_CACHE,struct bch_sb, flags[3], 29, 30);
989 LE64_BITMASK(BCH_SB_JOURNAL_FLUSH_DELAY,struct bch_sb, flags[3], 30, 62);
990 LE64_BITMASK(BCH_SB_JOURNAL_FLUSH_DISABLED,struct bch_sb, flags[3], 62, 63);
991 LE64_BITMASK(BCH_SB_JOURNAL_RECLAIM_DELAY,struct bch_sb, flags[4], 0, 32);
992 LE64_BITMASK(BCH_SB_JOURNAL_TRANSACTION_NAMES,struct bch_sb, flags[4], 32, 33);
993 LE64_BITMASK(BCH_SB_NOCOW, struct bch_sb, flags[4], 33, 34);
994 LE64_BITMASK(BCH_SB_WRITE_BUFFER_SIZE, struct bch_sb, flags[4], 34, 54);
995 LE64_BITMASK(BCH_SB_VERSION_UPGRADE, struct bch_sb, flags[4], 54, 56);
996
997 LE64_BITMASK(BCH_SB_COMPRESSION_TYPE_HI,struct bch_sb, flags[4], 56, 60);
998 LE64_BITMASK(BCH_SB_BACKGROUND_COMPRESSION_TYPE_HI,
999 struct bch_sb, flags[4], 60, 64);
1000
1001 LE64_BITMASK(BCH_SB_VERSION_UPGRADE_COMPLETE,
1002 struct bch_sb, flags[5], 0, 16);
1003
BCH_SB_COMPRESSION_TYPE(const struct bch_sb * sb)1004 static inline __u64 BCH_SB_COMPRESSION_TYPE(const struct bch_sb *sb)
1005 {
1006 return BCH_SB_COMPRESSION_TYPE_LO(sb) | (BCH_SB_COMPRESSION_TYPE_HI(sb) << 4);
1007 }
1008
SET_BCH_SB_COMPRESSION_TYPE(struct bch_sb * sb,__u64 v)1009 static inline void SET_BCH_SB_COMPRESSION_TYPE(struct bch_sb *sb, __u64 v)
1010 {
1011 SET_BCH_SB_COMPRESSION_TYPE_LO(sb, v);
1012 SET_BCH_SB_COMPRESSION_TYPE_HI(sb, v >> 4);
1013 }
1014
BCH_SB_BACKGROUND_COMPRESSION_TYPE(const struct bch_sb * sb)1015 static inline __u64 BCH_SB_BACKGROUND_COMPRESSION_TYPE(const struct bch_sb *sb)
1016 {
1017 return BCH_SB_BACKGROUND_COMPRESSION_TYPE_LO(sb) |
1018 (BCH_SB_BACKGROUND_COMPRESSION_TYPE_HI(sb) << 4);
1019 }
1020
SET_BCH_SB_BACKGROUND_COMPRESSION_TYPE(struct bch_sb * sb,__u64 v)1021 static inline void SET_BCH_SB_BACKGROUND_COMPRESSION_TYPE(struct bch_sb *sb, __u64 v)
1022 {
1023 SET_BCH_SB_BACKGROUND_COMPRESSION_TYPE_LO(sb, v);
1024 SET_BCH_SB_BACKGROUND_COMPRESSION_TYPE_HI(sb, v >> 4);
1025 }
1026
1027 /*
1028 * Features:
1029 *
1030 * journal_seq_blacklist_v3: gates BCH_SB_FIELD_journal_seq_blacklist
1031 * reflink: gates KEY_TYPE_reflink
1032 * inline_data: gates KEY_TYPE_inline_data
1033 * new_siphash: gates BCH_STR_HASH_siphash
1034 * new_extent_overwrite: gates BTREE_NODE_NEW_EXTENT_OVERWRITE
1035 */
1036 #define BCH_SB_FEATURES() \
1037 x(lz4, 0) \
1038 x(gzip, 1) \
1039 x(zstd, 2) \
1040 x(atomic_nlink, 3) \
1041 x(ec, 4) \
1042 x(journal_seq_blacklist_v3, 5) \
1043 x(reflink, 6) \
1044 x(new_siphash, 7) \
1045 x(inline_data, 8) \
1046 x(new_extent_overwrite, 9) \
1047 x(incompressible, 10) \
1048 x(btree_ptr_v2, 11) \
1049 x(extents_above_btree_updates, 12) \
1050 x(btree_updates_journalled, 13) \
1051 x(reflink_inline_data, 14) \
1052 x(new_varint, 15) \
1053 x(journal_no_flush, 16) \
1054 x(alloc_v2, 17) \
1055 x(extents_across_btree_nodes, 18)
1056
1057 #define BCH_SB_FEATURES_ALWAYS \
1058 ((1ULL << BCH_FEATURE_new_extent_overwrite)| \
1059 (1ULL << BCH_FEATURE_extents_above_btree_updates)|\
1060 (1ULL << BCH_FEATURE_btree_updates_journalled)|\
1061 (1ULL << BCH_FEATURE_alloc_v2)|\
1062 (1ULL << BCH_FEATURE_extents_across_btree_nodes))
1063
1064 #define BCH_SB_FEATURES_ALL \
1065 (BCH_SB_FEATURES_ALWAYS| \
1066 (1ULL << BCH_FEATURE_new_siphash)| \
1067 (1ULL << BCH_FEATURE_btree_ptr_v2)| \
1068 (1ULL << BCH_FEATURE_new_varint)| \
1069 (1ULL << BCH_FEATURE_journal_no_flush))
1070
1071 enum bch_sb_feature {
1072 #define x(f, n) BCH_FEATURE_##f,
1073 BCH_SB_FEATURES()
1074 #undef x
1075 BCH_FEATURE_NR,
1076 };
1077
1078 #define BCH_SB_COMPAT() \
1079 x(alloc_info, 0) \
1080 x(alloc_metadata, 1) \
1081 x(extents_above_btree_updates_done, 2) \
1082 x(bformat_overflow_done, 3)
1083
1084 enum bch_sb_compat {
1085 #define x(f, n) BCH_COMPAT_##f,
1086 BCH_SB_COMPAT()
1087 #undef x
1088 BCH_COMPAT_NR,
1089 };
1090
1091 /* options: */
1092
1093 #define BCH_VERSION_UPGRADE_OPTS() \
1094 x(compatible, 0) \
1095 x(incompatible, 1) \
1096 x(none, 2)
1097
1098 enum bch_version_upgrade_opts {
1099 #define x(t, n) BCH_VERSION_UPGRADE_##t = n,
1100 BCH_VERSION_UPGRADE_OPTS()
1101 #undef x
1102 };
1103
1104 #define BCH_REPLICAS_MAX 4U
1105
1106 #define BCH_BKEY_PTRS_MAX 16U
1107
1108 #define BCH_ERROR_ACTIONS() \
1109 x(continue, 0) \
1110 x(ro, 1) \
1111 x(panic, 2)
1112
1113 enum bch_error_actions {
1114 #define x(t, n) BCH_ON_ERROR_##t = n,
1115 BCH_ERROR_ACTIONS()
1116 #undef x
1117 BCH_ON_ERROR_NR
1118 };
1119
1120 #define BCH_STR_HASH_TYPES() \
1121 x(crc32c, 0) \
1122 x(crc64, 1) \
1123 x(siphash_old, 2) \
1124 x(siphash, 3)
1125
1126 enum bch_str_hash_type {
1127 #define x(t, n) BCH_STR_HASH_##t = n,
1128 BCH_STR_HASH_TYPES()
1129 #undef x
1130 BCH_STR_HASH_NR
1131 };
1132
1133 #define BCH_STR_HASH_OPTS() \
1134 x(crc32c, 0) \
1135 x(crc64, 1) \
1136 x(siphash, 2)
1137
1138 enum bch_str_hash_opts {
1139 #define x(t, n) BCH_STR_HASH_OPT_##t = n,
1140 BCH_STR_HASH_OPTS()
1141 #undef x
1142 BCH_STR_HASH_OPT_NR
1143 };
1144
1145 #define BCH_CSUM_TYPES() \
1146 x(none, 0) \
1147 x(crc32c_nonzero, 1) \
1148 x(crc64_nonzero, 2) \
1149 x(chacha20_poly1305_80, 3) \
1150 x(chacha20_poly1305_128, 4) \
1151 x(crc32c, 5) \
1152 x(crc64, 6) \
1153 x(xxhash, 7)
1154
1155 enum bch_csum_type {
1156 #define x(t, n) BCH_CSUM_##t = n,
1157 BCH_CSUM_TYPES()
1158 #undef x
1159 BCH_CSUM_NR
1160 };
1161
1162 static const __maybe_unused unsigned bch_crc_bytes[] = {
1163 [BCH_CSUM_none] = 0,
1164 [BCH_CSUM_crc32c_nonzero] = 4,
1165 [BCH_CSUM_crc32c] = 4,
1166 [BCH_CSUM_crc64_nonzero] = 8,
1167 [BCH_CSUM_crc64] = 8,
1168 [BCH_CSUM_xxhash] = 8,
1169 [BCH_CSUM_chacha20_poly1305_80] = 10,
1170 [BCH_CSUM_chacha20_poly1305_128] = 16,
1171 };
1172
bch2_csum_type_is_encryption(enum bch_csum_type type)1173 static inline _Bool bch2_csum_type_is_encryption(enum bch_csum_type type)
1174 {
1175 switch (type) {
1176 case BCH_CSUM_chacha20_poly1305_80:
1177 case BCH_CSUM_chacha20_poly1305_128:
1178 return true;
1179 default:
1180 return false;
1181 }
1182 }
1183
1184 #define BCH_CSUM_OPTS() \
1185 x(none, 0) \
1186 x(crc32c, 1) \
1187 x(crc64, 2) \
1188 x(xxhash, 3)
1189
1190 enum bch_csum_opts {
1191 #define x(t, n) BCH_CSUM_OPT_##t = n,
1192 BCH_CSUM_OPTS()
1193 #undef x
1194 BCH_CSUM_OPT_NR
1195 };
1196
1197 #define BCH_COMPRESSION_TYPES() \
1198 x(none, 0) \
1199 x(lz4_old, 1) \
1200 x(gzip, 2) \
1201 x(lz4, 3) \
1202 x(zstd, 4) \
1203 x(incompressible, 5)
1204
1205 enum bch_compression_type {
1206 #define x(t, n) BCH_COMPRESSION_TYPE_##t = n,
1207 BCH_COMPRESSION_TYPES()
1208 #undef x
1209 BCH_COMPRESSION_TYPE_NR
1210 };
1211
1212 #define BCH_COMPRESSION_OPTS() \
1213 x(none, 0) \
1214 x(lz4, 1) \
1215 x(gzip, 2) \
1216 x(zstd, 3)
1217
1218 enum bch_compression_opts {
1219 #define x(t, n) BCH_COMPRESSION_OPT_##t = n,
1220 BCH_COMPRESSION_OPTS()
1221 #undef x
1222 BCH_COMPRESSION_OPT_NR
1223 };
1224
1225 /*
1226 * Magic numbers
1227 *
1228 * The various other data structures have their own magic numbers, which are
1229 * xored with the first part of the cache set's UUID
1230 */
1231
1232 #define BCACHE_MAGIC \
1233 UUID_INIT(0xc68573f6, 0x4e1a, 0x45ca, \
1234 0x82, 0x65, 0xf5, 0x7f, 0x48, 0xba, 0x6d, 0x81)
1235 #define BCHFS_MAGIC \
1236 UUID_INIT(0xc68573f6, 0x66ce, 0x90a9, \
1237 0xd9, 0x6a, 0x60, 0xcf, 0x80, 0x3d, 0xf7, 0xef)
1238
1239 #define BCACHEFS_STATFS_MAGIC 0xca451a4e
1240
1241 #define JSET_MAGIC __cpu_to_le64(0x245235c1a3625032ULL)
1242 #define BSET_MAGIC __cpu_to_le64(0x90135c78b99e07f5ULL)
1243
__bch2_sb_magic(struct bch_sb * sb)1244 static inline __le64 __bch2_sb_magic(struct bch_sb *sb)
1245 {
1246 __le64 ret;
1247
1248 memcpy(&ret, &sb->uuid, sizeof(ret));
1249 return ret;
1250 }
1251
__jset_magic(struct bch_sb * sb)1252 static inline __u64 __jset_magic(struct bch_sb *sb)
1253 {
1254 return __le64_to_cpu(__bch2_sb_magic(sb) ^ JSET_MAGIC);
1255 }
1256
__bset_magic(struct bch_sb * sb)1257 static inline __u64 __bset_magic(struct bch_sb *sb)
1258 {
1259 return __le64_to_cpu(__bch2_sb_magic(sb) ^ BSET_MAGIC);
1260 }
1261
1262 /* Journal */
1263
1264 #define JSET_KEYS_U64s (sizeof(struct jset_entry) / sizeof(__u64))
1265
1266 #define BCH_JSET_ENTRY_TYPES() \
1267 x(btree_keys, 0) \
1268 x(btree_root, 1) \
1269 x(prio_ptrs, 2) \
1270 x(blacklist, 3) \
1271 x(blacklist_v2, 4) \
1272 x(usage, 5) \
1273 x(data_usage, 6) \
1274 x(clock, 7) \
1275 x(dev_usage, 8) \
1276 x(log, 9) \
1277 x(overwrite, 10) \
1278 x(write_buffer_keys, 11)
1279
1280 enum {
1281 #define x(f, nr) BCH_JSET_ENTRY_##f = nr,
1282 BCH_JSET_ENTRY_TYPES()
1283 #undef x
1284 BCH_JSET_ENTRY_NR
1285 };
1286
jset_entry_is_key(struct jset_entry * e)1287 static inline bool jset_entry_is_key(struct jset_entry *e)
1288 {
1289 switch (e->type) {
1290 case BCH_JSET_ENTRY_btree_keys:
1291 case BCH_JSET_ENTRY_btree_root:
1292 case BCH_JSET_ENTRY_overwrite:
1293 case BCH_JSET_ENTRY_write_buffer_keys:
1294 return true;
1295 }
1296
1297 return false;
1298 }
1299
1300 /*
1301 * Journal sequence numbers can be blacklisted: bsets record the max sequence
1302 * number of all the journal entries they contain updates for, so that on
1303 * recovery we can ignore those bsets that contain index updates newer that what
1304 * made it into the journal.
1305 *
1306 * This means that we can't reuse that journal_seq - we have to skip it, and
1307 * then record that we skipped it so that the next time we crash and recover we
1308 * don't think there was a missing journal entry.
1309 */
1310 struct jset_entry_blacklist {
1311 struct jset_entry entry;
1312 __le64 seq;
1313 };
1314
1315 struct jset_entry_blacklist_v2 {
1316 struct jset_entry entry;
1317 __le64 start;
1318 __le64 end;
1319 };
1320
1321 #define BCH_FS_USAGE_TYPES() \
1322 x(reserved, 0) \
1323 x(inodes, 1) \
1324 x(key_version, 2)
1325
1326 enum {
1327 #define x(f, nr) BCH_FS_USAGE_##f = nr,
1328 BCH_FS_USAGE_TYPES()
1329 #undef x
1330 BCH_FS_USAGE_NR
1331 };
1332
1333 struct jset_entry_usage {
1334 struct jset_entry entry;
1335 __le64 v;
1336 } __packed;
1337
1338 struct jset_entry_data_usage {
1339 struct jset_entry entry;
1340 __le64 v;
1341 struct bch_replicas_entry_v1 r;
1342 } __packed;
1343
1344 struct jset_entry_clock {
1345 struct jset_entry entry;
1346 __u8 rw;
1347 __u8 pad[7];
1348 __le64 time;
1349 } __packed;
1350
1351 struct jset_entry_dev_usage_type {
1352 __le64 buckets;
1353 __le64 sectors;
1354 __le64 fragmented;
1355 } __packed;
1356
1357 struct jset_entry_dev_usage {
1358 struct jset_entry entry;
1359 __le32 dev;
1360 __u32 pad;
1361
1362 __le64 _buckets_ec; /* No longer used */
1363 __le64 _buckets_unavailable; /* No longer used */
1364
1365 struct jset_entry_dev_usage_type d[];
1366 };
1367
jset_entry_dev_usage_nr_types(struct jset_entry_dev_usage * u)1368 static inline unsigned jset_entry_dev_usage_nr_types(struct jset_entry_dev_usage *u)
1369 {
1370 return (vstruct_bytes(&u->entry) - sizeof(struct jset_entry_dev_usage)) /
1371 sizeof(struct jset_entry_dev_usage_type);
1372 }
1373
1374 struct jset_entry_log {
1375 struct jset_entry entry;
1376 u8 d[];
1377 } __packed __aligned(8);
1378
1379 /*
1380 * On disk format for a journal entry:
1381 * seq is monotonically increasing; every journal entry has its own unique
1382 * sequence number.
1383 *
1384 * last_seq is the oldest journal entry that still has keys the btree hasn't
1385 * flushed to disk yet.
1386 *
1387 * version is for on disk format changes.
1388 */
1389 struct jset {
1390 struct bch_csum csum;
1391
1392 __le64 magic;
1393 __le64 seq;
1394 __le32 version;
1395 __le32 flags;
1396
1397 __le32 u64s; /* size of d[] in u64s */
1398
1399 __u8 encrypted_start[0];
1400
1401 __le16 _read_clock; /* no longer used */
1402 __le16 _write_clock;
1403
1404 /* Sequence number of oldest dirty journal entry */
1405 __le64 last_seq;
1406
1407
1408 struct jset_entry start[0];
1409 __u64 _data[];
1410 } __packed __aligned(8);
1411
1412 LE32_BITMASK(JSET_CSUM_TYPE, struct jset, flags, 0, 4);
1413 LE32_BITMASK(JSET_BIG_ENDIAN, struct jset, flags, 4, 5);
1414 LE32_BITMASK(JSET_NO_FLUSH, struct jset, flags, 5, 6);
1415
1416 #define BCH_JOURNAL_BUCKETS_MIN 8
1417
1418 /* Btree: */
1419
1420 enum btree_id_flags {
1421 BTREE_ID_EXTENTS = BIT(0),
1422 BTREE_ID_SNAPSHOTS = BIT(1),
1423 BTREE_ID_SNAPSHOT_FIELD = BIT(2),
1424 BTREE_ID_DATA = BIT(3),
1425 };
1426
1427 #define BCH_BTREE_IDS() \
1428 x(extents, 0, BTREE_ID_EXTENTS|BTREE_ID_SNAPSHOTS|BTREE_ID_DATA,\
1429 BIT_ULL(KEY_TYPE_whiteout)| \
1430 BIT_ULL(KEY_TYPE_error)| \
1431 BIT_ULL(KEY_TYPE_cookie)| \
1432 BIT_ULL(KEY_TYPE_extent)| \
1433 BIT_ULL(KEY_TYPE_reservation)| \
1434 BIT_ULL(KEY_TYPE_reflink_p)| \
1435 BIT_ULL(KEY_TYPE_inline_data)) \
1436 x(inodes, 1, BTREE_ID_SNAPSHOTS, \
1437 BIT_ULL(KEY_TYPE_whiteout)| \
1438 BIT_ULL(KEY_TYPE_inode)| \
1439 BIT_ULL(KEY_TYPE_inode_v2)| \
1440 BIT_ULL(KEY_TYPE_inode_v3)| \
1441 BIT_ULL(KEY_TYPE_inode_generation)) \
1442 x(dirents, 2, BTREE_ID_SNAPSHOTS, \
1443 BIT_ULL(KEY_TYPE_whiteout)| \
1444 BIT_ULL(KEY_TYPE_hash_whiteout)| \
1445 BIT_ULL(KEY_TYPE_dirent)) \
1446 x(xattrs, 3, BTREE_ID_SNAPSHOTS, \
1447 BIT_ULL(KEY_TYPE_whiteout)| \
1448 BIT_ULL(KEY_TYPE_cookie)| \
1449 BIT_ULL(KEY_TYPE_hash_whiteout)| \
1450 BIT_ULL(KEY_TYPE_xattr)) \
1451 x(alloc, 4, 0, \
1452 BIT_ULL(KEY_TYPE_alloc)| \
1453 BIT_ULL(KEY_TYPE_alloc_v2)| \
1454 BIT_ULL(KEY_TYPE_alloc_v3)| \
1455 BIT_ULL(KEY_TYPE_alloc_v4)) \
1456 x(quotas, 5, 0, \
1457 BIT_ULL(KEY_TYPE_quota)) \
1458 x(stripes, 6, 0, \
1459 BIT_ULL(KEY_TYPE_stripe)) \
1460 x(reflink, 7, BTREE_ID_EXTENTS|BTREE_ID_DATA, \
1461 BIT_ULL(KEY_TYPE_reflink_v)| \
1462 BIT_ULL(KEY_TYPE_indirect_inline_data)) \
1463 x(subvolumes, 8, 0, \
1464 BIT_ULL(KEY_TYPE_subvolume)) \
1465 x(snapshots, 9, 0, \
1466 BIT_ULL(KEY_TYPE_snapshot)) \
1467 x(lru, 10, 0, \
1468 BIT_ULL(KEY_TYPE_set)) \
1469 x(freespace, 11, BTREE_ID_EXTENTS, \
1470 BIT_ULL(KEY_TYPE_set)) \
1471 x(need_discard, 12, 0, \
1472 BIT_ULL(KEY_TYPE_set)) \
1473 x(backpointers, 13, 0, \
1474 BIT_ULL(KEY_TYPE_backpointer)) \
1475 x(bucket_gens, 14, 0, \
1476 BIT_ULL(KEY_TYPE_bucket_gens)) \
1477 x(snapshot_trees, 15, 0, \
1478 BIT_ULL(KEY_TYPE_snapshot_tree)) \
1479 x(deleted_inodes, 16, BTREE_ID_SNAPSHOT_FIELD, \
1480 BIT_ULL(KEY_TYPE_set)) \
1481 x(logged_ops, 17, 0, \
1482 BIT_ULL(KEY_TYPE_logged_op_truncate)| \
1483 BIT_ULL(KEY_TYPE_logged_op_finsert)) \
1484 x(rebalance_work, 18, BTREE_ID_SNAPSHOT_FIELD, \
1485 BIT_ULL(KEY_TYPE_set)|BIT_ULL(KEY_TYPE_cookie))
1486
1487 enum btree_id {
1488 #define x(name, nr, ...) BTREE_ID_##name = nr,
1489 BCH_BTREE_IDS()
1490 #undef x
1491 BTREE_ID_NR
1492 };
1493
1494 #define BTREE_MAX_DEPTH 4U
1495
1496 /* Btree nodes */
1497
1498 /*
1499 * Btree nodes
1500 *
1501 * On disk a btree node is a list/log of these; within each set the keys are
1502 * sorted
1503 */
1504 struct bset {
1505 __le64 seq;
1506
1507 /*
1508 * Highest journal entry this bset contains keys for.
1509 * If on recovery we don't see that journal entry, this bset is ignored:
1510 * this allows us to preserve the order of all index updates after a
1511 * crash, since the journal records a total order of all index updates
1512 * and anything that didn't make it to the journal doesn't get used.
1513 */
1514 __le64 journal_seq;
1515
1516 __le32 flags;
1517 __le16 version;
1518 __le16 u64s; /* count of d[] in u64s */
1519
1520 struct bkey_packed start[0];
1521 __u64 _data[];
1522 } __packed __aligned(8);
1523
1524 LE32_BITMASK(BSET_CSUM_TYPE, struct bset, flags, 0, 4);
1525
1526 LE32_BITMASK(BSET_BIG_ENDIAN, struct bset, flags, 4, 5);
1527 LE32_BITMASK(BSET_SEPARATE_WHITEOUTS,
1528 struct bset, flags, 5, 6);
1529
1530 /* Sector offset within the btree node: */
1531 LE32_BITMASK(BSET_OFFSET, struct bset, flags, 16, 32);
1532
1533 struct btree_node {
1534 struct bch_csum csum;
1535 __le64 magic;
1536
1537 /* this flags field is encrypted, unlike bset->flags: */
1538 __le64 flags;
1539
1540 /* Closed interval: */
1541 struct bpos min_key;
1542 struct bpos max_key;
1543 struct bch_extent_ptr _ptr; /* not used anymore */
1544 struct bkey_format format;
1545
1546 union {
1547 struct bset keys;
1548 struct {
1549 __u8 pad[22];
1550 __le16 u64s;
1551 __u64 _data[0];
1552
1553 };
1554 };
1555 } __packed __aligned(8);
1556
1557 LE64_BITMASK(BTREE_NODE_ID_LO, struct btree_node, flags, 0, 4);
1558 LE64_BITMASK(BTREE_NODE_LEVEL, struct btree_node, flags, 4, 8);
1559 LE64_BITMASK(BTREE_NODE_NEW_EXTENT_OVERWRITE,
1560 struct btree_node, flags, 8, 9);
1561 LE64_BITMASK(BTREE_NODE_ID_HI, struct btree_node, flags, 9, 25);
1562 /* 25-32 unused */
1563 LE64_BITMASK(BTREE_NODE_SEQ, struct btree_node, flags, 32, 64);
1564
BTREE_NODE_ID(struct btree_node * n)1565 static inline __u64 BTREE_NODE_ID(struct btree_node *n)
1566 {
1567 return BTREE_NODE_ID_LO(n) | (BTREE_NODE_ID_HI(n) << 4);
1568 }
1569
SET_BTREE_NODE_ID(struct btree_node * n,__u64 v)1570 static inline void SET_BTREE_NODE_ID(struct btree_node *n, __u64 v)
1571 {
1572 SET_BTREE_NODE_ID_LO(n, v);
1573 SET_BTREE_NODE_ID_HI(n, v >> 4);
1574 }
1575
1576 struct btree_node_entry {
1577 struct bch_csum csum;
1578
1579 union {
1580 struct bset keys;
1581 struct {
1582 __u8 pad[22];
1583 __le16 u64s;
1584 __u64 _data[0];
1585 };
1586 };
1587 } __packed __aligned(8);
1588
1589 #endif /* _BCACHEFS_FORMAT_H */
1590