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 <uapi/linux/magic.h>
80 #include "vstructs.h"
81
82 #ifdef __KERNEL__
83 typedef uuid_t __uuid_t;
84 #endif
85
86 #define BITMASK(name, type, field, offset, end) \
87 static const __maybe_unused unsigned name##_OFFSET = offset; \
88 static const __maybe_unused unsigned name##_BITS = (end - offset); \
89 \
90 static inline __u64 name(const type *k) \
91 { \
92 return (k->field >> offset) & ~(~0ULL << (end - offset)); \
93 } \
94 \
95 static inline void SET_##name(type *k, __u64 v) \
96 { \
97 k->field &= ~(~(~0ULL << (end - offset)) << offset); \
98 k->field |= (v & ~(~0ULL << (end - offset))) << offset; \
99 }
100
101 #define LE_BITMASK(_bits, name, type, field, offset, end) \
102 static const __maybe_unused unsigned name##_OFFSET = offset; \
103 static const __maybe_unused unsigned name##_BITS = (end - offset); \
104 static const __maybe_unused __u##_bits name##_MAX = (1ULL << (end - offset)) - 1;\
105 \
106 static inline __u64 name(const type *k) \
107 { \
108 return (__le##_bits##_to_cpu(k->field) >> offset) & \
109 ~(~0ULL << (end - offset)); \
110 } \
111 \
112 static inline void SET_##name(type *k, __u64 v) \
113 { \
114 __u##_bits new = __le##_bits##_to_cpu(k->field); \
115 \
116 new &= ~(~(~0ULL << (end - offset)) << offset); \
117 new |= (v & ~(~0ULL << (end - offset))) << offset; \
118 k->field = __cpu_to_le##_bits(new); \
119 }
120
121 #define LE16_BITMASK(n, t, f, o, e) LE_BITMASK(16, n, t, f, o, e)
122 #define LE32_BITMASK(n, t, f, o, e) LE_BITMASK(32, n, t, f, o, e)
123 #define LE64_BITMASK(n, t, f, o, e) LE_BITMASK(64, n, t, f, o, e)
124
125 struct bkey_format {
126 __u8 key_u64s;
127 __u8 nr_fields;
128 /* One unused slot for now: */
129 __u8 bits_per_field[6];
130 __le64 field_offset[6];
131 };
132
133 /* Btree keys - all units are in sectors */
134
135 struct bpos {
136 /*
137 * Word order matches machine byte order - btree code treats a bpos as a
138 * single large integer, for search/comparison purposes
139 *
140 * Note that wherever a bpos is embedded in another on disk data
141 * structure, it has to be byte swabbed when reading in metadata that
142 * wasn't written in native endian order:
143 */
144 #if __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__
145 __u32 snapshot;
146 __u64 offset;
147 __u64 inode;
148 #elif __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__
149 __u64 inode;
150 __u64 offset; /* Points to end of extent - sectors */
151 __u32 snapshot;
152 #else
153 #error edit for your odd byteorder.
154 #endif
155 } __packed
156 #if __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__
157 __aligned(4)
158 #endif
159 ;
160
161 #define KEY_INODE_MAX ((__u64)~0ULL)
162 #define KEY_OFFSET_MAX ((__u64)~0ULL)
163 #define KEY_SNAPSHOT_MAX ((__u32)~0U)
164 #define KEY_SIZE_MAX ((__u32)~0U)
165
SPOS(__u64 inode,__u64 offset,__u32 snapshot)166 static inline struct bpos SPOS(__u64 inode, __u64 offset, __u32 snapshot)
167 {
168 return (struct bpos) {
169 .inode = inode,
170 .offset = offset,
171 .snapshot = snapshot,
172 };
173 }
174
175 #define POS_MIN SPOS(0, 0, 0)
176 #define POS_MAX SPOS(KEY_INODE_MAX, KEY_OFFSET_MAX, 0)
177 #define SPOS_MAX SPOS(KEY_INODE_MAX, KEY_OFFSET_MAX, KEY_SNAPSHOT_MAX)
178 #define POS(_inode, _offset) SPOS(_inode, _offset, 0)
179
180 /* Empty placeholder struct, for container_of() */
181 struct bch_val {
182 __u64 __nothing[0];
183 };
184
185 struct bversion {
186 #if __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__
187 __u64 lo;
188 __u32 hi;
189 #elif __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__
190 __u32 hi;
191 __u64 lo;
192 #endif
193 } __packed
194 #if __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__
195 __aligned(4)
196 #endif
197 ;
198
199 struct bkey {
200 /* Size of combined key and value, in u64s */
201 __u8 u64s;
202
203 /* Format of key (0 for format local to btree node) */
204 #if defined(__LITTLE_ENDIAN_BITFIELD)
205 __u8 format:7,
206 needs_whiteout:1;
207 #elif defined (__BIG_ENDIAN_BITFIELD)
208 __u8 needs_whiteout:1,
209 format:7;
210 #else
211 #error edit for your odd byteorder.
212 #endif
213
214 /* Type of the value */
215 __u8 type;
216
217 #if __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__
218 __u8 pad[1];
219
220 struct bversion version;
221 __u32 size; /* extent size, in sectors */
222 struct bpos p;
223 #elif __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__
224 struct bpos p;
225 __u32 size; /* extent size, in sectors */
226 struct bversion version;
227
228 __u8 pad[1];
229 #endif
230 } __packed
231 #if __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__
232 /*
233 * The big-endian version of bkey can't be compiled by rustc with the "aligned"
234 * attr since it doesn't allow types to have both "packed" and "aligned" attrs.
235 * So for Rust compatibility, don't include this. It can be included in the LE
236 * version because the "packed" attr is redundant in that case.
237 *
238 * History: (quoting Kent)
239 *
240 * Specifically, when i was designing bkey, I wanted the header to be no
241 * bigger than necessary so that bkey_packed could use the rest. That means that
242 * decently offten extent keys will fit into only 8 bytes, instead of spilling over
243 * to 16.
244 *
245 * But packed_bkey treats the part after the header - the packed section -
246 * as a single multi word, variable length integer. And bkey, the unpacked
247 * version, is just a special case version of a bkey_packed; all the packed
248 * bkey code will work on keys in any packed format, the in-memory
249 * representation of an unpacked key also is just one type of packed key...
250 *
251 * So that constrains the key part of a bkig endian bkey to start right
252 * after the header.
253 *
254 * If we ever do a bkey_v2 and need to expand the hedaer by another byte for
255 * some reason - that will clean up this wart.
256 */
257 __aligned(8)
258 #endif
259 ;
260
261 struct bkey_packed {
262 __u64 _data[0];
263
264 /* Size of combined key and value, in u64s */
265 __u8 u64s;
266
267 /* Format of key (0 for format local to btree node) */
268
269 /*
270 * XXX: next incompat on disk format change, switch format and
271 * needs_whiteout - bkey_packed() will be cheaper if format is the high
272 * bits of the bitfield
273 */
274 #if defined(__LITTLE_ENDIAN_BITFIELD)
275 __u8 format:7,
276 needs_whiteout:1;
277 #elif defined (__BIG_ENDIAN_BITFIELD)
278 __u8 needs_whiteout:1,
279 format:7;
280 #endif
281
282 /* Type of the value */
283 __u8 type;
284 __u8 key_start[0];
285
286 /*
287 * We copy bkeys with struct assignment in various places, and while
288 * that shouldn't be done with packed bkeys we can't disallow it in C,
289 * and it's legal to cast a bkey to a bkey_packed - so padding it out
290 * to the same size as struct bkey should hopefully be safest.
291 */
292 __u8 pad[sizeof(struct bkey) - 3];
293 } __packed __aligned(8);
294
295 typedef struct {
296 __le64 lo;
297 __le64 hi;
298 } bch_le128;
299
300 #define BKEY_U64s (sizeof(struct bkey) / sizeof(__u64))
301 #define BKEY_U64s_MAX U8_MAX
302 #define BKEY_VAL_U64s_MAX (BKEY_U64s_MAX - BKEY_U64s)
303
304 #define KEY_PACKED_BITS_START 24
305
306 #define KEY_FORMAT_LOCAL_BTREE 0
307 #define KEY_FORMAT_CURRENT 1
308
309 enum bch_bkey_fields {
310 BKEY_FIELD_INODE,
311 BKEY_FIELD_OFFSET,
312 BKEY_FIELD_SNAPSHOT,
313 BKEY_FIELD_SIZE,
314 BKEY_FIELD_VERSION_HI,
315 BKEY_FIELD_VERSION_LO,
316 BKEY_NR_FIELDS,
317 };
318
319 #define bkey_format_field(name, field) \
320 [BKEY_FIELD_##name] = (sizeof(((struct bkey *) NULL)->field) * 8)
321
322 #define BKEY_FORMAT_CURRENT \
323 ((struct bkey_format) { \
324 .key_u64s = BKEY_U64s, \
325 .nr_fields = BKEY_NR_FIELDS, \
326 .bits_per_field = { \
327 bkey_format_field(INODE, p.inode), \
328 bkey_format_field(OFFSET, p.offset), \
329 bkey_format_field(SNAPSHOT, p.snapshot), \
330 bkey_format_field(SIZE, size), \
331 bkey_format_field(VERSION_HI, version.hi), \
332 bkey_format_field(VERSION_LO, version.lo), \
333 }, \
334 })
335
336 /* bkey with inline value */
337 struct bkey_i {
338 __u64 _data[0];
339
340 struct bkey k;
341 struct bch_val v;
342 };
343
344 #define POS_KEY(_pos) \
345 ((struct bkey) { \
346 .u64s = BKEY_U64s, \
347 .format = KEY_FORMAT_CURRENT, \
348 .p = _pos, \
349 })
350
351 #define KEY(_inode, _offset, _size) \
352 ((struct bkey) { \
353 .u64s = BKEY_U64s, \
354 .format = KEY_FORMAT_CURRENT, \
355 .p = POS(_inode, _offset), \
356 .size = _size, \
357 })
358
bkey_init(struct bkey * k)359 static inline void bkey_init(struct bkey *k)
360 {
361 *k = KEY(0, 0, 0);
362 }
363
364 #define bkey_bytes(_k) ((_k)->u64s * sizeof(__u64))
365
366 #define __BKEY_PADDED(key, pad) \
367 struct bkey_i key; __u64 key ## _pad[pad]
368
369 /*
370 * - DELETED keys are used internally to mark keys that should be ignored but
371 * override keys in composition order. Their version number is ignored.
372 *
373 * - DISCARDED keys indicate that the data is all 0s because it has been
374 * discarded. DISCARDs may have a version; if the version is nonzero the key
375 * will be persistent, otherwise the key will be dropped whenever the btree
376 * node is rewritten (like DELETED keys).
377 *
378 * - ERROR: any read of the data returns a read error, as the data was lost due
379 * to a failing device. Like DISCARDED keys, they can be removed (overridden)
380 * by new writes or cluster-wide GC. Node repair can also overwrite them with
381 * the same or a more recent version number, but not with an older version
382 * number.
383 *
384 * - WHITEOUT: for hash table btrees
385 */
386 #define BCH_BKEY_TYPES() \
387 x(deleted, 0) \
388 x(whiteout, 1) \
389 x(error, 2) \
390 x(cookie, 3) \
391 x(hash_whiteout, 4) \
392 x(btree_ptr, 5) \
393 x(extent, 6) \
394 x(reservation, 7) \
395 x(inode, 8) \
396 x(inode_generation, 9) \
397 x(dirent, 10) \
398 x(xattr, 11) \
399 x(alloc, 12) \
400 x(quota, 13) \
401 x(stripe, 14) \
402 x(reflink_p, 15) \
403 x(reflink_v, 16) \
404 x(inline_data, 17) \
405 x(btree_ptr_v2, 18) \
406 x(indirect_inline_data, 19) \
407 x(alloc_v2, 20) \
408 x(subvolume, 21) \
409 x(snapshot, 22) \
410 x(inode_v2, 23) \
411 x(alloc_v3, 24) \
412 x(set, 25) \
413 x(lru, 26) \
414 x(alloc_v4, 27) \
415 x(backpointer, 28) \
416 x(inode_v3, 29) \
417 x(bucket_gens, 30) \
418 x(snapshot_tree, 31) \
419 x(logged_op_truncate, 32) \
420 x(logged_op_finsert, 33)
421
422 enum bch_bkey_type {
423 #define x(name, nr) KEY_TYPE_##name = nr,
424 BCH_BKEY_TYPES()
425 #undef x
426 KEY_TYPE_MAX,
427 };
428
429 struct bch_deleted {
430 struct bch_val v;
431 };
432
433 struct bch_whiteout {
434 struct bch_val v;
435 };
436
437 struct bch_error {
438 struct bch_val v;
439 };
440
441 struct bch_cookie {
442 struct bch_val v;
443 __le64 cookie;
444 };
445
446 struct bch_hash_whiteout {
447 struct bch_val v;
448 };
449
450 struct bch_set {
451 struct bch_val v;
452 };
453
454 /* 128 bits, sufficient for cryptographic MACs: */
455 struct bch_csum {
456 __le64 lo;
457 __le64 hi;
458 } __packed __aligned(8);
459
460 struct bch_backpointer {
461 struct bch_val v;
462 __u8 btree_id;
463 __u8 level;
464 __u8 data_type;
465 __u64 bucket_offset:40;
466 __u32 bucket_len;
467 struct bpos pos;
468 } __packed __aligned(8);
469
470 /* LRU btree: */
471
472 struct bch_lru {
473 struct bch_val v;
474 __le64 idx;
475 } __packed __aligned(8);
476
477 #define LRU_ID_STRIPES (1U << 16)
478
479 /* Optional/variable size superblock sections: */
480
481 struct bch_sb_field {
482 __u64 _data[0];
483 __le32 u64s;
484 __le32 type;
485 };
486
487 #define BCH_SB_FIELDS() \
488 x(journal, 0) \
489 x(members_v1, 1) \
490 x(crypt, 2) \
491 x(replicas_v0, 3) \
492 x(quota, 4) \
493 x(disk_groups, 5) \
494 x(clean, 6) \
495 x(replicas, 7) \
496 x(journal_seq_blacklist, 8) \
497 x(journal_v2, 9) \
498 x(counters, 10) \
499 x(members_v2, 11) \
500 x(errors, 12) \
501 x(ext, 13) \
502 x(downgrade, 14)
503
504 #include "alloc_background_format.h"
505 #include "extents_format.h"
506 #include "ec_format.h"
507 #include "dirent_format.h"
508 #include "disk_groups_format.h"
509 #include "inode_format.h"
510 #include "journal_seq_blacklist_format.h"
511 #include "logged_ops_format.h"
512 #include "quota_format.h"
513 #include "reflink_format.h"
514 #include "replicas_format.h"
515 #include "snapshot_format.h"
516 #include "subvolume_format.h"
517 #include "sb-counters_format.h"
518 #include "sb-downgrade_format.h"
519 #include "sb-errors_format.h"
520 #include "sb-members_format.h"
521 #include "xattr_format.h"
522
523 enum bch_sb_field_type {
524 #define x(f, nr) BCH_SB_FIELD_##f = nr,
525 BCH_SB_FIELDS()
526 #undef x
527 BCH_SB_FIELD_NR
528 };
529
530 /*
531 * Most superblock fields are replicated in all device's superblocks - a few are
532 * not:
533 */
534 #define BCH_SINGLE_DEVICE_SB_FIELDS \
535 ((1U << BCH_SB_FIELD_journal)| \
536 (1U << BCH_SB_FIELD_journal_v2))
537
538 /* BCH_SB_FIELD_journal: */
539
540 struct bch_sb_field_journal {
541 struct bch_sb_field field;
542 __le64 buckets[];
543 };
544
545 struct bch_sb_field_journal_v2 {
546 struct bch_sb_field field;
547
548 struct bch_sb_field_journal_v2_entry {
549 __le64 start;
550 __le64 nr;
551 } d[];
552 };
553
554 /* BCH_SB_FIELD_crypt: */
555
556 struct nonce {
557 __le32 d[4];
558 };
559
560 struct bch_key {
561 __le64 key[4];
562 };
563
564 #define BCH_KEY_MAGIC \
565 (((__u64) 'b' << 0)|((__u64) 'c' << 8)| \
566 ((__u64) 'h' << 16)|((__u64) '*' << 24)| \
567 ((__u64) '*' << 32)|((__u64) 'k' << 40)| \
568 ((__u64) 'e' << 48)|((__u64) 'y' << 56))
569
570 struct bch_encrypted_key {
571 __le64 magic;
572 struct bch_key key;
573 };
574
575 /*
576 * If this field is present in the superblock, it stores an encryption key which
577 * is used encrypt all other data/metadata. The key will normally be encrypted
578 * with the key userspace provides, but if encryption has been turned off we'll
579 * just store the master key unencrypted in the superblock so we can access the
580 * previously encrypted data.
581 */
582 struct bch_sb_field_crypt {
583 struct bch_sb_field field;
584
585 __le64 flags;
586 __le64 kdf_flags;
587 struct bch_encrypted_key key;
588 };
589
590 LE64_BITMASK(BCH_CRYPT_KDF_TYPE, struct bch_sb_field_crypt, flags, 0, 4);
591
592 enum bch_kdf_types {
593 BCH_KDF_SCRYPT = 0,
594 BCH_KDF_NR = 1,
595 };
596
597 /* stored as base 2 log of scrypt params: */
598 LE64_BITMASK(BCH_KDF_SCRYPT_N, struct bch_sb_field_crypt, kdf_flags, 0, 16);
599 LE64_BITMASK(BCH_KDF_SCRYPT_R, struct bch_sb_field_crypt, kdf_flags, 16, 32);
600 LE64_BITMASK(BCH_KDF_SCRYPT_P, struct bch_sb_field_crypt, kdf_flags, 32, 48);
601
602 #define BCH_DATA_TYPES() \
603 x(free, 0) \
604 x(sb, 1) \
605 x(journal, 2) \
606 x(btree, 3) \
607 x(user, 4) \
608 x(cached, 5) \
609 x(parity, 6) \
610 x(stripe, 7) \
611 x(need_gc_gens, 8) \
612 x(need_discard, 9)
613
614 enum bch_data_type {
615 #define x(t, n) BCH_DATA_##t,
616 BCH_DATA_TYPES()
617 #undef x
618 BCH_DATA_NR
619 };
620
data_type_is_empty(enum bch_data_type type)621 static inline bool data_type_is_empty(enum bch_data_type type)
622 {
623 switch (type) {
624 case BCH_DATA_free:
625 case BCH_DATA_need_gc_gens:
626 case BCH_DATA_need_discard:
627 return true;
628 default:
629 return false;
630 }
631 }
632
data_type_is_hidden(enum bch_data_type type)633 static inline bool data_type_is_hidden(enum bch_data_type type)
634 {
635 switch (type) {
636 case BCH_DATA_sb:
637 case BCH_DATA_journal:
638 return true;
639 default:
640 return false;
641 }
642 }
643
644 /*
645 * On clean shutdown, store btree roots and current journal sequence number in
646 * the superblock:
647 */
648 struct jset_entry {
649 __le16 u64s;
650 __u8 btree_id;
651 __u8 level;
652 __u8 type; /* designates what this jset holds */
653 __u8 pad[3];
654
655 struct bkey_i start[0];
656 __u64 _data[];
657 };
658
659 struct bch_sb_field_clean {
660 struct bch_sb_field field;
661
662 __le32 flags;
663 __le16 _read_clock; /* no longer used */
664 __le16 _write_clock;
665 __le64 journal_seq;
666
667 struct jset_entry start[0];
668 __u64 _data[];
669 };
670
671 struct bch_sb_field_ext {
672 struct bch_sb_field field;
673 __le64 recovery_passes_required[2];
674 __le64 errors_silent[8];
675 __le64 btrees_lost_data;
676 };
677
678 /* Superblock: */
679
680 /*
681 * New versioning scheme:
682 * One common version number for all on disk data structures - superblock, btree
683 * nodes, journal entries
684 */
685 #define BCH_VERSION_MAJOR(_v) ((__u16) ((_v) >> 10))
686 #define BCH_VERSION_MINOR(_v) ((__u16) ((_v) & ~(~0U << 10)))
687 #define BCH_VERSION(_major, _minor) (((_major) << 10)|(_minor) << 0)
688
689 /*
690 * field 1: version name
691 * field 2: BCH_VERSION(major, minor)
692 * field 3: recovery passess required on upgrade
693 */
694 #define BCH_METADATA_VERSIONS() \
695 x(bkey_renumber, BCH_VERSION(0, 10)) \
696 x(inode_btree_change, BCH_VERSION(0, 11)) \
697 x(snapshot, BCH_VERSION(0, 12)) \
698 x(inode_backpointers, BCH_VERSION(0, 13)) \
699 x(btree_ptr_sectors_written, BCH_VERSION(0, 14)) \
700 x(snapshot_2, BCH_VERSION(0, 15)) \
701 x(reflink_p_fix, BCH_VERSION(0, 16)) \
702 x(subvol_dirent, BCH_VERSION(0, 17)) \
703 x(inode_v2, BCH_VERSION(0, 18)) \
704 x(freespace, BCH_VERSION(0, 19)) \
705 x(alloc_v4, BCH_VERSION(0, 20)) \
706 x(new_data_types, BCH_VERSION(0, 21)) \
707 x(backpointers, BCH_VERSION(0, 22)) \
708 x(inode_v3, BCH_VERSION(0, 23)) \
709 x(unwritten_extents, BCH_VERSION(0, 24)) \
710 x(bucket_gens, BCH_VERSION(0, 25)) \
711 x(lru_v2, BCH_VERSION(0, 26)) \
712 x(fragmentation_lru, BCH_VERSION(0, 27)) \
713 x(no_bps_in_alloc_keys, BCH_VERSION(0, 28)) \
714 x(snapshot_trees, BCH_VERSION(0, 29)) \
715 x(major_minor, BCH_VERSION(1, 0)) \
716 x(snapshot_skiplists, BCH_VERSION(1, 1)) \
717 x(deleted_inodes, BCH_VERSION(1, 2)) \
718 x(rebalance_work, BCH_VERSION(1, 3)) \
719 x(member_seq, BCH_VERSION(1, 4)) \
720 x(subvolume_fs_parent, BCH_VERSION(1, 5)) \
721 x(btree_subvolume_children, BCH_VERSION(1, 6)) \
722 x(mi_btree_bitmap, BCH_VERSION(1, 7))
723
724 enum bcachefs_metadata_version {
725 bcachefs_metadata_version_min = 9,
726 #define x(t, n) bcachefs_metadata_version_##t = n,
727 BCH_METADATA_VERSIONS()
728 #undef x
729 bcachefs_metadata_version_max
730 };
731
732 static const __maybe_unused
733 unsigned bcachefs_metadata_required_upgrade_below = bcachefs_metadata_version_rebalance_work;
734
735 #define bcachefs_metadata_version_current (bcachefs_metadata_version_max - 1)
736
737 #define BCH_SB_SECTOR 8
738
739 #define BCH_SB_LAYOUT_SIZE_BITS_MAX 16 /* 32 MB */
740
741 struct bch_sb_layout {
742 __uuid_t magic; /* bcachefs superblock UUID */
743 __u8 layout_type;
744 __u8 sb_max_size_bits; /* base 2 of 512 byte sectors */
745 __u8 nr_superblocks;
746 __u8 pad[5];
747 __le64 sb_offset[61];
748 } __packed __aligned(8);
749
750 #define BCH_SB_LAYOUT_SECTOR 7
751
752 /*
753 * @offset - sector where this sb was written
754 * @version - on disk format version
755 * @version_min - Oldest metadata version this filesystem contains; so we can
756 * safely drop compatibility code and refuse to mount filesystems
757 * we'd need it for
758 * @magic - identifies as a bcachefs superblock (BCHFS_MAGIC)
759 * @seq - incremented each time superblock is written
760 * @uuid - used for generating various magic numbers and identifying
761 * member devices, never changes
762 * @user_uuid - user visible UUID, may be changed
763 * @label - filesystem label
764 * @seq - identifies most recent superblock, incremented each time
765 * superblock is written
766 * @features - enabled incompatible features
767 */
768 struct bch_sb {
769 struct bch_csum csum;
770 __le16 version;
771 __le16 version_min;
772 __le16 pad[2];
773 __uuid_t magic;
774 __uuid_t uuid;
775 __uuid_t user_uuid;
776 __u8 label[BCH_SB_LABEL_SIZE];
777 __le64 offset;
778 __le64 seq;
779
780 __le16 block_size;
781 __u8 dev_idx;
782 __u8 nr_devices;
783 __le32 u64s;
784
785 __le64 time_base_lo;
786 __le32 time_base_hi;
787 __le32 time_precision;
788
789 __le64 flags[7];
790 __le64 write_time;
791 __le64 features[2];
792 __le64 compat[2];
793
794 struct bch_sb_layout layout;
795
796 struct bch_sb_field start[0];
797 __le64 _data[];
798 } __packed __aligned(8);
799
800 /*
801 * Flags:
802 * BCH_SB_INITALIZED - set on first mount
803 * BCH_SB_CLEAN - did we shut down cleanly? Just a hint, doesn't affect
804 * behaviour of mount/recovery path:
805 * BCH_SB_INODE_32BIT - limit inode numbers to 32 bits
806 * BCH_SB_128_BIT_MACS - 128 bit macs instead of 80
807 * BCH_SB_ENCRYPTION_TYPE - if nonzero encryption is enabled; overrides
808 * DATA/META_CSUM_TYPE. Also indicates encryption
809 * algorithm in use, if/when we get more than one
810 */
811
812 LE16_BITMASK(BCH_SB_BLOCK_SIZE, struct bch_sb, block_size, 0, 16);
813
814 LE64_BITMASK(BCH_SB_INITIALIZED, struct bch_sb, flags[0], 0, 1);
815 LE64_BITMASK(BCH_SB_CLEAN, struct bch_sb, flags[0], 1, 2);
816 LE64_BITMASK(BCH_SB_CSUM_TYPE, struct bch_sb, flags[0], 2, 8);
817 LE64_BITMASK(BCH_SB_ERROR_ACTION, struct bch_sb, flags[0], 8, 12);
818
819 LE64_BITMASK(BCH_SB_BTREE_NODE_SIZE, struct bch_sb, flags[0], 12, 28);
820
821 LE64_BITMASK(BCH_SB_GC_RESERVE, struct bch_sb, flags[0], 28, 33);
822 LE64_BITMASK(BCH_SB_ROOT_RESERVE, struct bch_sb, flags[0], 33, 40);
823
824 LE64_BITMASK(BCH_SB_META_CSUM_TYPE, struct bch_sb, flags[0], 40, 44);
825 LE64_BITMASK(BCH_SB_DATA_CSUM_TYPE, struct bch_sb, flags[0], 44, 48);
826
827 LE64_BITMASK(BCH_SB_META_REPLICAS_WANT, struct bch_sb, flags[0], 48, 52);
828 LE64_BITMASK(BCH_SB_DATA_REPLICAS_WANT, struct bch_sb, flags[0], 52, 56);
829
830 LE64_BITMASK(BCH_SB_POSIX_ACL, struct bch_sb, flags[0], 56, 57);
831 LE64_BITMASK(BCH_SB_USRQUOTA, struct bch_sb, flags[0], 57, 58);
832 LE64_BITMASK(BCH_SB_GRPQUOTA, struct bch_sb, flags[0], 58, 59);
833 LE64_BITMASK(BCH_SB_PRJQUOTA, struct bch_sb, flags[0], 59, 60);
834
835 LE64_BITMASK(BCH_SB_HAS_ERRORS, struct bch_sb, flags[0], 60, 61);
836 LE64_BITMASK(BCH_SB_HAS_TOPOLOGY_ERRORS,struct bch_sb, flags[0], 61, 62);
837
838 LE64_BITMASK(BCH_SB_BIG_ENDIAN, struct bch_sb, flags[0], 62, 63);
839
840 LE64_BITMASK(BCH_SB_STR_HASH_TYPE, struct bch_sb, flags[1], 0, 4);
841 LE64_BITMASK(BCH_SB_COMPRESSION_TYPE_LO,struct bch_sb, flags[1], 4, 8);
842 LE64_BITMASK(BCH_SB_INODE_32BIT, struct bch_sb, flags[1], 8, 9);
843
844 LE64_BITMASK(BCH_SB_128_BIT_MACS, struct bch_sb, flags[1], 9, 10);
845 LE64_BITMASK(BCH_SB_ENCRYPTION_TYPE, struct bch_sb, flags[1], 10, 14);
846
847 /*
848 * Max size of an extent that may require bouncing to read or write
849 * (checksummed, compressed): 64k
850 */
851 LE64_BITMASK(BCH_SB_ENCODED_EXTENT_MAX_BITS,
852 struct bch_sb, flags[1], 14, 20);
853
854 LE64_BITMASK(BCH_SB_META_REPLICAS_REQ, struct bch_sb, flags[1], 20, 24);
855 LE64_BITMASK(BCH_SB_DATA_REPLICAS_REQ, struct bch_sb, flags[1], 24, 28);
856
857 LE64_BITMASK(BCH_SB_PROMOTE_TARGET, struct bch_sb, flags[1], 28, 40);
858 LE64_BITMASK(BCH_SB_FOREGROUND_TARGET, struct bch_sb, flags[1], 40, 52);
859 LE64_BITMASK(BCH_SB_BACKGROUND_TARGET, struct bch_sb, flags[1], 52, 64);
860
861 LE64_BITMASK(BCH_SB_BACKGROUND_COMPRESSION_TYPE_LO,
862 struct bch_sb, flags[2], 0, 4);
863 LE64_BITMASK(BCH_SB_GC_RESERVE_BYTES, struct bch_sb, flags[2], 4, 64);
864
865 LE64_BITMASK(BCH_SB_ERASURE_CODE, struct bch_sb, flags[3], 0, 16);
866 LE64_BITMASK(BCH_SB_METADATA_TARGET, struct bch_sb, flags[3], 16, 28);
867 LE64_BITMASK(BCH_SB_SHARD_INUMS, struct bch_sb, flags[3], 28, 29);
868 LE64_BITMASK(BCH_SB_INODES_USE_KEY_CACHE,struct bch_sb, flags[3], 29, 30);
869 LE64_BITMASK(BCH_SB_JOURNAL_FLUSH_DELAY,struct bch_sb, flags[3], 30, 62);
870 LE64_BITMASK(BCH_SB_JOURNAL_FLUSH_DISABLED,struct bch_sb, flags[3], 62, 63);
871 LE64_BITMASK(BCH_SB_JOURNAL_RECLAIM_DELAY,struct bch_sb, flags[4], 0, 32);
872 LE64_BITMASK(BCH_SB_JOURNAL_TRANSACTION_NAMES,struct bch_sb, flags[4], 32, 33);
873 LE64_BITMASK(BCH_SB_NOCOW, struct bch_sb, flags[4], 33, 34);
874 LE64_BITMASK(BCH_SB_WRITE_BUFFER_SIZE, struct bch_sb, flags[4], 34, 54);
875 LE64_BITMASK(BCH_SB_VERSION_UPGRADE, struct bch_sb, flags[4], 54, 56);
876
877 LE64_BITMASK(BCH_SB_COMPRESSION_TYPE_HI,struct bch_sb, flags[4], 56, 60);
878 LE64_BITMASK(BCH_SB_BACKGROUND_COMPRESSION_TYPE_HI,
879 struct bch_sb, flags[4], 60, 64);
880
881 LE64_BITMASK(BCH_SB_VERSION_UPGRADE_COMPLETE,
882 struct bch_sb, flags[5], 0, 16);
883
BCH_SB_COMPRESSION_TYPE(const struct bch_sb * sb)884 static inline __u64 BCH_SB_COMPRESSION_TYPE(const struct bch_sb *sb)
885 {
886 return BCH_SB_COMPRESSION_TYPE_LO(sb) | (BCH_SB_COMPRESSION_TYPE_HI(sb) << 4);
887 }
888
SET_BCH_SB_COMPRESSION_TYPE(struct bch_sb * sb,__u64 v)889 static inline void SET_BCH_SB_COMPRESSION_TYPE(struct bch_sb *sb, __u64 v)
890 {
891 SET_BCH_SB_COMPRESSION_TYPE_LO(sb, v);
892 SET_BCH_SB_COMPRESSION_TYPE_HI(sb, v >> 4);
893 }
894
BCH_SB_BACKGROUND_COMPRESSION_TYPE(const struct bch_sb * sb)895 static inline __u64 BCH_SB_BACKGROUND_COMPRESSION_TYPE(const struct bch_sb *sb)
896 {
897 return BCH_SB_BACKGROUND_COMPRESSION_TYPE_LO(sb) |
898 (BCH_SB_BACKGROUND_COMPRESSION_TYPE_HI(sb) << 4);
899 }
900
SET_BCH_SB_BACKGROUND_COMPRESSION_TYPE(struct bch_sb * sb,__u64 v)901 static inline void SET_BCH_SB_BACKGROUND_COMPRESSION_TYPE(struct bch_sb *sb, __u64 v)
902 {
903 SET_BCH_SB_BACKGROUND_COMPRESSION_TYPE_LO(sb, v);
904 SET_BCH_SB_BACKGROUND_COMPRESSION_TYPE_HI(sb, v >> 4);
905 }
906
907 /*
908 * Features:
909 *
910 * journal_seq_blacklist_v3: gates BCH_SB_FIELD_journal_seq_blacklist
911 * reflink: gates KEY_TYPE_reflink
912 * inline_data: gates KEY_TYPE_inline_data
913 * new_siphash: gates BCH_STR_HASH_siphash
914 * new_extent_overwrite: gates BTREE_NODE_NEW_EXTENT_OVERWRITE
915 */
916 #define BCH_SB_FEATURES() \
917 x(lz4, 0) \
918 x(gzip, 1) \
919 x(zstd, 2) \
920 x(atomic_nlink, 3) \
921 x(ec, 4) \
922 x(journal_seq_blacklist_v3, 5) \
923 x(reflink, 6) \
924 x(new_siphash, 7) \
925 x(inline_data, 8) \
926 x(new_extent_overwrite, 9) \
927 x(incompressible, 10) \
928 x(btree_ptr_v2, 11) \
929 x(extents_above_btree_updates, 12) \
930 x(btree_updates_journalled, 13) \
931 x(reflink_inline_data, 14) \
932 x(new_varint, 15) \
933 x(journal_no_flush, 16) \
934 x(alloc_v2, 17) \
935 x(extents_across_btree_nodes, 18)
936
937 #define BCH_SB_FEATURES_ALWAYS \
938 ((1ULL << BCH_FEATURE_new_extent_overwrite)| \
939 (1ULL << BCH_FEATURE_extents_above_btree_updates)|\
940 (1ULL << BCH_FEATURE_btree_updates_journalled)|\
941 (1ULL << BCH_FEATURE_alloc_v2)|\
942 (1ULL << BCH_FEATURE_extents_across_btree_nodes))
943
944 #define BCH_SB_FEATURES_ALL \
945 (BCH_SB_FEATURES_ALWAYS| \
946 (1ULL << BCH_FEATURE_new_siphash)| \
947 (1ULL << BCH_FEATURE_btree_ptr_v2)| \
948 (1ULL << BCH_FEATURE_new_varint)| \
949 (1ULL << BCH_FEATURE_journal_no_flush))
950
951 enum bch_sb_feature {
952 #define x(f, n) BCH_FEATURE_##f,
953 BCH_SB_FEATURES()
954 #undef x
955 BCH_FEATURE_NR,
956 };
957
958 #define BCH_SB_COMPAT() \
959 x(alloc_info, 0) \
960 x(alloc_metadata, 1) \
961 x(extents_above_btree_updates_done, 2) \
962 x(bformat_overflow_done, 3)
963
964 enum bch_sb_compat {
965 #define x(f, n) BCH_COMPAT_##f,
966 BCH_SB_COMPAT()
967 #undef x
968 BCH_COMPAT_NR,
969 };
970
971 /* options: */
972
973 #define BCH_VERSION_UPGRADE_OPTS() \
974 x(compatible, 0) \
975 x(incompatible, 1) \
976 x(none, 2)
977
978 enum bch_version_upgrade_opts {
979 #define x(t, n) BCH_VERSION_UPGRADE_##t = n,
980 BCH_VERSION_UPGRADE_OPTS()
981 #undef x
982 };
983
984 #define BCH_REPLICAS_MAX 4U
985
986 #define BCH_BKEY_PTRS_MAX 16U
987
988 #define BCH_ERROR_ACTIONS() \
989 x(continue, 0) \
990 x(ro, 1) \
991 x(panic, 2)
992
993 enum bch_error_actions {
994 #define x(t, n) BCH_ON_ERROR_##t = n,
995 BCH_ERROR_ACTIONS()
996 #undef x
997 BCH_ON_ERROR_NR
998 };
999
1000 #define BCH_STR_HASH_TYPES() \
1001 x(crc32c, 0) \
1002 x(crc64, 1) \
1003 x(siphash_old, 2) \
1004 x(siphash, 3)
1005
1006 enum bch_str_hash_type {
1007 #define x(t, n) BCH_STR_HASH_##t = n,
1008 BCH_STR_HASH_TYPES()
1009 #undef x
1010 BCH_STR_HASH_NR
1011 };
1012
1013 #define BCH_STR_HASH_OPTS() \
1014 x(crc32c, 0) \
1015 x(crc64, 1) \
1016 x(siphash, 2)
1017
1018 enum bch_str_hash_opts {
1019 #define x(t, n) BCH_STR_HASH_OPT_##t = n,
1020 BCH_STR_HASH_OPTS()
1021 #undef x
1022 BCH_STR_HASH_OPT_NR
1023 };
1024
1025 #define BCH_CSUM_TYPES() \
1026 x(none, 0) \
1027 x(crc32c_nonzero, 1) \
1028 x(crc64_nonzero, 2) \
1029 x(chacha20_poly1305_80, 3) \
1030 x(chacha20_poly1305_128, 4) \
1031 x(crc32c, 5) \
1032 x(crc64, 6) \
1033 x(xxhash, 7)
1034
1035 enum bch_csum_type {
1036 #define x(t, n) BCH_CSUM_##t = n,
1037 BCH_CSUM_TYPES()
1038 #undef x
1039 BCH_CSUM_NR
1040 };
1041
1042 static const __maybe_unused unsigned bch_crc_bytes[] = {
1043 [BCH_CSUM_none] = 0,
1044 [BCH_CSUM_crc32c_nonzero] = 4,
1045 [BCH_CSUM_crc32c] = 4,
1046 [BCH_CSUM_crc64_nonzero] = 8,
1047 [BCH_CSUM_crc64] = 8,
1048 [BCH_CSUM_xxhash] = 8,
1049 [BCH_CSUM_chacha20_poly1305_80] = 10,
1050 [BCH_CSUM_chacha20_poly1305_128] = 16,
1051 };
1052
bch2_csum_type_is_encryption(enum bch_csum_type type)1053 static inline _Bool bch2_csum_type_is_encryption(enum bch_csum_type type)
1054 {
1055 switch (type) {
1056 case BCH_CSUM_chacha20_poly1305_80:
1057 case BCH_CSUM_chacha20_poly1305_128:
1058 return true;
1059 default:
1060 return false;
1061 }
1062 }
1063
1064 #define BCH_CSUM_OPTS() \
1065 x(none, 0) \
1066 x(crc32c, 1) \
1067 x(crc64, 2) \
1068 x(xxhash, 3)
1069
1070 enum bch_csum_opts {
1071 #define x(t, n) BCH_CSUM_OPT_##t = n,
1072 BCH_CSUM_OPTS()
1073 #undef x
1074 BCH_CSUM_OPT_NR
1075 };
1076
1077 #define BCH_COMPRESSION_TYPES() \
1078 x(none, 0) \
1079 x(lz4_old, 1) \
1080 x(gzip, 2) \
1081 x(lz4, 3) \
1082 x(zstd, 4) \
1083 x(incompressible, 5)
1084
1085 enum bch_compression_type {
1086 #define x(t, n) BCH_COMPRESSION_TYPE_##t = n,
1087 BCH_COMPRESSION_TYPES()
1088 #undef x
1089 BCH_COMPRESSION_TYPE_NR
1090 };
1091
1092 #define BCH_COMPRESSION_OPTS() \
1093 x(none, 0) \
1094 x(lz4, 1) \
1095 x(gzip, 2) \
1096 x(zstd, 3)
1097
1098 enum bch_compression_opts {
1099 #define x(t, n) BCH_COMPRESSION_OPT_##t = n,
1100 BCH_COMPRESSION_OPTS()
1101 #undef x
1102 BCH_COMPRESSION_OPT_NR
1103 };
1104
1105 /*
1106 * Magic numbers
1107 *
1108 * The various other data structures have their own magic numbers, which are
1109 * xored with the first part of the cache set's UUID
1110 */
1111
1112 #define BCACHE_MAGIC \
1113 UUID_INIT(0xc68573f6, 0x4e1a, 0x45ca, \
1114 0x82, 0x65, 0xf5, 0x7f, 0x48, 0xba, 0x6d, 0x81)
1115 #define BCHFS_MAGIC \
1116 UUID_INIT(0xc68573f6, 0x66ce, 0x90a9, \
1117 0xd9, 0x6a, 0x60, 0xcf, 0x80, 0x3d, 0xf7, 0xef)
1118
1119 #define BCACHEFS_STATFS_MAGIC BCACHEFS_SUPER_MAGIC
1120
1121 #define JSET_MAGIC __cpu_to_le64(0x245235c1a3625032ULL)
1122 #define BSET_MAGIC __cpu_to_le64(0x90135c78b99e07f5ULL)
1123
__bch2_sb_magic(struct bch_sb * sb)1124 static inline __le64 __bch2_sb_magic(struct bch_sb *sb)
1125 {
1126 __le64 ret;
1127
1128 memcpy(&ret, &sb->uuid, sizeof(ret));
1129 return ret;
1130 }
1131
__jset_magic(struct bch_sb * sb)1132 static inline __u64 __jset_magic(struct bch_sb *sb)
1133 {
1134 return __le64_to_cpu(__bch2_sb_magic(sb) ^ JSET_MAGIC);
1135 }
1136
__bset_magic(struct bch_sb * sb)1137 static inline __u64 __bset_magic(struct bch_sb *sb)
1138 {
1139 return __le64_to_cpu(__bch2_sb_magic(sb) ^ BSET_MAGIC);
1140 }
1141
1142 /* Journal */
1143
1144 #define JSET_KEYS_U64s (sizeof(struct jset_entry) / sizeof(__u64))
1145
1146 #define BCH_JSET_ENTRY_TYPES() \
1147 x(btree_keys, 0) \
1148 x(btree_root, 1) \
1149 x(prio_ptrs, 2) \
1150 x(blacklist, 3) \
1151 x(blacklist_v2, 4) \
1152 x(usage, 5) \
1153 x(data_usage, 6) \
1154 x(clock, 7) \
1155 x(dev_usage, 8) \
1156 x(log, 9) \
1157 x(overwrite, 10) \
1158 x(write_buffer_keys, 11) \
1159 x(datetime, 12)
1160
1161 enum bch_jset_entry_type {
1162 #define x(f, nr) BCH_JSET_ENTRY_##f = nr,
1163 BCH_JSET_ENTRY_TYPES()
1164 #undef x
1165 BCH_JSET_ENTRY_NR
1166 };
1167
jset_entry_is_key(struct jset_entry * e)1168 static inline bool jset_entry_is_key(struct jset_entry *e)
1169 {
1170 switch (e->type) {
1171 case BCH_JSET_ENTRY_btree_keys:
1172 case BCH_JSET_ENTRY_btree_root:
1173 case BCH_JSET_ENTRY_overwrite:
1174 case BCH_JSET_ENTRY_write_buffer_keys:
1175 return true;
1176 }
1177
1178 return false;
1179 }
1180
1181 /*
1182 * Journal sequence numbers can be blacklisted: bsets record the max sequence
1183 * number of all the journal entries they contain updates for, so that on
1184 * recovery we can ignore those bsets that contain index updates newer that what
1185 * made it into the journal.
1186 *
1187 * This means that we can't reuse that journal_seq - we have to skip it, and
1188 * then record that we skipped it so that the next time we crash and recover we
1189 * don't think there was a missing journal entry.
1190 */
1191 struct jset_entry_blacklist {
1192 struct jset_entry entry;
1193 __le64 seq;
1194 };
1195
1196 struct jset_entry_blacklist_v2 {
1197 struct jset_entry entry;
1198 __le64 start;
1199 __le64 end;
1200 };
1201
1202 #define BCH_FS_USAGE_TYPES() \
1203 x(reserved, 0) \
1204 x(inodes, 1) \
1205 x(key_version, 2)
1206
1207 enum bch_fs_usage_type {
1208 #define x(f, nr) BCH_FS_USAGE_##f = nr,
1209 BCH_FS_USAGE_TYPES()
1210 #undef x
1211 BCH_FS_USAGE_NR
1212 };
1213
1214 struct jset_entry_usage {
1215 struct jset_entry entry;
1216 __le64 v;
1217 } __packed;
1218
1219 struct jset_entry_data_usage {
1220 struct jset_entry entry;
1221 __le64 v;
1222 struct bch_replicas_entry_v1 r;
1223 } __packed;
1224
1225 struct jset_entry_clock {
1226 struct jset_entry entry;
1227 __u8 rw;
1228 __u8 pad[7];
1229 __le64 time;
1230 } __packed;
1231
1232 struct jset_entry_dev_usage_type {
1233 __le64 buckets;
1234 __le64 sectors;
1235 __le64 fragmented;
1236 } __packed;
1237
1238 struct jset_entry_dev_usage {
1239 struct jset_entry entry;
1240 __le32 dev;
1241 __u32 pad;
1242
1243 __le64 _buckets_ec; /* No longer used */
1244 __le64 _buckets_unavailable; /* No longer used */
1245
1246 struct jset_entry_dev_usage_type d[];
1247 };
1248
jset_entry_dev_usage_nr_types(struct jset_entry_dev_usage * u)1249 static inline unsigned jset_entry_dev_usage_nr_types(struct jset_entry_dev_usage *u)
1250 {
1251 return (vstruct_bytes(&u->entry) - sizeof(struct jset_entry_dev_usage)) /
1252 sizeof(struct jset_entry_dev_usage_type);
1253 }
1254
1255 struct jset_entry_log {
1256 struct jset_entry entry;
1257 u8 d[];
1258 } __packed __aligned(8);
1259
1260 struct jset_entry_datetime {
1261 struct jset_entry entry;
1262 __le64 seconds;
1263 } __packed __aligned(8);
1264
1265 /*
1266 * On disk format for a journal entry:
1267 * seq is monotonically increasing; every journal entry has its own unique
1268 * sequence number.
1269 *
1270 * last_seq is the oldest journal entry that still has keys the btree hasn't
1271 * flushed to disk yet.
1272 *
1273 * version is for on disk format changes.
1274 */
1275 struct jset {
1276 struct bch_csum csum;
1277
1278 __le64 magic;
1279 __le64 seq;
1280 __le32 version;
1281 __le32 flags;
1282
1283 __le32 u64s; /* size of d[] in u64s */
1284
1285 __u8 encrypted_start[0];
1286
1287 __le16 _read_clock; /* no longer used */
1288 __le16 _write_clock;
1289
1290 /* Sequence number of oldest dirty journal entry */
1291 __le64 last_seq;
1292
1293
1294 struct jset_entry start[0];
1295 __u64 _data[];
1296 } __packed __aligned(8);
1297
1298 LE32_BITMASK(JSET_CSUM_TYPE, struct jset, flags, 0, 4);
1299 LE32_BITMASK(JSET_BIG_ENDIAN, struct jset, flags, 4, 5);
1300 LE32_BITMASK(JSET_NO_FLUSH, struct jset, flags, 5, 6);
1301
1302 #define BCH_JOURNAL_BUCKETS_MIN 8
1303
1304 /* Btree: */
1305
1306 enum btree_id_flags {
1307 BTREE_ID_EXTENTS = BIT(0),
1308 BTREE_ID_SNAPSHOTS = BIT(1),
1309 BTREE_ID_SNAPSHOT_FIELD = BIT(2),
1310 BTREE_ID_DATA = BIT(3),
1311 };
1312
1313 #define BCH_BTREE_IDS() \
1314 x(extents, 0, BTREE_ID_EXTENTS|BTREE_ID_SNAPSHOTS|BTREE_ID_DATA,\
1315 BIT_ULL(KEY_TYPE_whiteout)| \
1316 BIT_ULL(KEY_TYPE_error)| \
1317 BIT_ULL(KEY_TYPE_cookie)| \
1318 BIT_ULL(KEY_TYPE_extent)| \
1319 BIT_ULL(KEY_TYPE_reservation)| \
1320 BIT_ULL(KEY_TYPE_reflink_p)| \
1321 BIT_ULL(KEY_TYPE_inline_data)) \
1322 x(inodes, 1, BTREE_ID_SNAPSHOTS, \
1323 BIT_ULL(KEY_TYPE_whiteout)| \
1324 BIT_ULL(KEY_TYPE_inode)| \
1325 BIT_ULL(KEY_TYPE_inode_v2)| \
1326 BIT_ULL(KEY_TYPE_inode_v3)| \
1327 BIT_ULL(KEY_TYPE_inode_generation)) \
1328 x(dirents, 2, BTREE_ID_SNAPSHOTS, \
1329 BIT_ULL(KEY_TYPE_whiteout)| \
1330 BIT_ULL(KEY_TYPE_hash_whiteout)| \
1331 BIT_ULL(KEY_TYPE_dirent)) \
1332 x(xattrs, 3, BTREE_ID_SNAPSHOTS, \
1333 BIT_ULL(KEY_TYPE_whiteout)| \
1334 BIT_ULL(KEY_TYPE_cookie)| \
1335 BIT_ULL(KEY_TYPE_hash_whiteout)| \
1336 BIT_ULL(KEY_TYPE_xattr)) \
1337 x(alloc, 4, 0, \
1338 BIT_ULL(KEY_TYPE_alloc)| \
1339 BIT_ULL(KEY_TYPE_alloc_v2)| \
1340 BIT_ULL(KEY_TYPE_alloc_v3)| \
1341 BIT_ULL(KEY_TYPE_alloc_v4)) \
1342 x(quotas, 5, 0, \
1343 BIT_ULL(KEY_TYPE_quota)) \
1344 x(stripes, 6, 0, \
1345 BIT_ULL(KEY_TYPE_stripe)) \
1346 x(reflink, 7, BTREE_ID_EXTENTS|BTREE_ID_DATA, \
1347 BIT_ULL(KEY_TYPE_reflink_v)| \
1348 BIT_ULL(KEY_TYPE_indirect_inline_data)| \
1349 BIT_ULL(KEY_TYPE_error)) \
1350 x(subvolumes, 8, 0, \
1351 BIT_ULL(KEY_TYPE_subvolume)) \
1352 x(snapshots, 9, 0, \
1353 BIT_ULL(KEY_TYPE_snapshot)) \
1354 x(lru, 10, 0, \
1355 BIT_ULL(KEY_TYPE_set)) \
1356 x(freespace, 11, BTREE_ID_EXTENTS, \
1357 BIT_ULL(KEY_TYPE_set)) \
1358 x(need_discard, 12, 0, \
1359 BIT_ULL(KEY_TYPE_set)) \
1360 x(backpointers, 13, 0, \
1361 BIT_ULL(KEY_TYPE_backpointer)) \
1362 x(bucket_gens, 14, 0, \
1363 BIT_ULL(KEY_TYPE_bucket_gens)) \
1364 x(snapshot_trees, 15, 0, \
1365 BIT_ULL(KEY_TYPE_snapshot_tree)) \
1366 x(deleted_inodes, 16, BTREE_ID_SNAPSHOT_FIELD, \
1367 BIT_ULL(KEY_TYPE_set)) \
1368 x(logged_ops, 17, 0, \
1369 BIT_ULL(KEY_TYPE_logged_op_truncate)| \
1370 BIT_ULL(KEY_TYPE_logged_op_finsert)) \
1371 x(rebalance_work, 18, BTREE_ID_SNAPSHOT_FIELD, \
1372 BIT_ULL(KEY_TYPE_set)|BIT_ULL(KEY_TYPE_cookie)) \
1373 x(subvolume_children, 19, 0, \
1374 BIT_ULL(KEY_TYPE_set))
1375
1376 enum btree_id {
1377 #define x(name, nr, ...) BTREE_ID_##name = nr,
1378 BCH_BTREE_IDS()
1379 #undef x
1380 BTREE_ID_NR
1381 };
1382
1383 /*
1384 * Maximum number of btrees that we will _ever_ have under the current scheme,
1385 * where we refer to them with bitfields
1386 */
1387 #define BTREE_ID_NR_MAX 64
1388
btree_id_is_alloc(enum btree_id id)1389 static inline bool btree_id_is_alloc(enum btree_id id)
1390 {
1391 switch (id) {
1392 case BTREE_ID_alloc:
1393 case BTREE_ID_backpointers:
1394 case BTREE_ID_need_discard:
1395 case BTREE_ID_freespace:
1396 case BTREE_ID_bucket_gens:
1397 return true;
1398 default:
1399 return false;
1400 }
1401 }
1402
1403 #define BTREE_MAX_DEPTH 4U
1404
1405 /* Btree nodes */
1406
1407 /*
1408 * Btree nodes
1409 *
1410 * On disk a btree node is a list/log of these; within each set the keys are
1411 * sorted
1412 */
1413 struct bset {
1414 __le64 seq;
1415
1416 /*
1417 * Highest journal entry this bset contains keys for.
1418 * If on recovery we don't see that journal entry, this bset is ignored:
1419 * this allows us to preserve the order of all index updates after a
1420 * crash, since the journal records a total order of all index updates
1421 * and anything that didn't make it to the journal doesn't get used.
1422 */
1423 __le64 journal_seq;
1424
1425 __le32 flags;
1426 __le16 version;
1427 __le16 u64s; /* count of d[] in u64s */
1428
1429 struct bkey_packed start[0];
1430 __u64 _data[];
1431 } __packed __aligned(8);
1432
1433 LE32_BITMASK(BSET_CSUM_TYPE, struct bset, flags, 0, 4);
1434
1435 LE32_BITMASK(BSET_BIG_ENDIAN, struct bset, flags, 4, 5);
1436 LE32_BITMASK(BSET_SEPARATE_WHITEOUTS,
1437 struct bset, flags, 5, 6);
1438
1439 /* Sector offset within the btree node: */
1440 LE32_BITMASK(BSET_OFFSET, struct bset, flags, 16, 32);
1441
1442 struct btree_node {
1443 struct bch_csum csum;
1444 __le64 magic;
1445
1446 /* this flags field is encrypted, unlike bset->flags: */
1447 __le64 flags;
1448
1449 /* Closed interval: */
1450 struct bpos min_key;
1451 struct bpos max_key;
1452 struct bch_extent_ptr _ptr; /* not used anymore */
1453 struct bkey_format format;
1454
1455 union {
1456 struct bset keys;
1457 struct {
1458 __u8 pad[22];
1459 __le16 u64s;
1460 __u64 _data[0];
1461
1462 };
1463 };
1464 } __packed __aligned(8);
1465
1466 LE64_BITMASK(BTREE_NODE_ID_LO, struct btree_node, flags, 0, 4);
1467 LE64_BITMASK(BTREE_NODE_LEVEL, struct btree_node, flags, 4, 8);
1468 LE64_BITMASK(BTREE_NODE_NEW_EXTENT_OVERWRITE,
1469 struct btree_node, flags, 8, 9);
1470 LE64_BITMASK(BTREE_NODE_ID_HI, struct btree_node, flags, 9, 25);
1471 /* 25-32 unused */
1472 LE64_BITMASK(BTREE_NODE_SEQ, struct btree_node, flags, 32, 64);
1473
BTREE_NODE_ID(struct btree_node * n)1474 static inline __u64 BTREE_NODE_ID(struct btree_node *n)
1475 {
1476 return BTREE_NODE_ID_LO(n) | (BTREE_NODE_ID_HI(n) << 4);
1477 }
1478
SET_BTREE_NODE_ID(struct btree_node * n,__u64 v)1479 static inline void SET_BTREE_NODE_ID(struct btree_node *n, __u64 v)
1480 {
1481 SET_BTREE_NODE_ID_LO(n, v);
1482 SET_BTREE_NODE_ID_HI(n, v >> 4);
1483 }
1484
1485 struct btree_node_entry {
1486 struct bch_csum csum;
1487
1488 union {
1489 struct bset keys;
1490 struct {
1491 __u8 pad[22];
1492 __le16 u64s;
1493 __u64 _data[0];
1494 };
1495 };
1496 } __packed __aligned(8);
1497
1498 #endif /* _BCACHEFS_FORMAT_H */
1499