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 bversion;
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 bversion;
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, bversion.hi), \
332 bkey_format_field(VERSION_LO, bversion.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 enum bch_bkey_type_flags {
370 BKEY_TYPE_strict_btree_checks = BIT(0),
371 };
372
373 /*
374 * - DELETED keys are used internally to mark keys that should be ignored but
375 * override keys in composition order. Their version number is ignored.
376 *
377 * - DISCARDED keys indicate that the data is all 0s because it has been
378 * discarded. DISCARDs may have a version; if the version is nonzero the key
379 * will be persistent, otherwise the key will be dropped whenever the btree
380 * node is rewritten (like DELETED keys).
381 *
382 * - ERROR: any read of the data returns a read error, as the data was lost due
383 * to a failing device. Like DISCARDED keys, they can be removed (overridden)
384 * by new writes or cluster-wide GC. Node repair can also overwrite them with
385 * the same or a more recent version number, but not with an older version
386 * number.
387 *
388 * - WHITEOUT: for hash table btrees
389 */
390 #define BCH_BKEY_TYPES() \
391 x(deleted, 0, 0) \
392 x(whiteout, 1, 0) \
393 x(error, 2, 0) \
394 x(cookie, 3, 0) \
395 x(hash_whiteout, 4, BKEY_TYPE_strict_btree_checks) \
396 x(btree_ptr, 5, BKEY_TYPE_strict_btree_checks) \
397 x(extent, 6, BKEY_TYPE_strict_btree_checks) \
398 x(reservation, 7, BKEY_TYPE_strict_btree_checks) \
399 x(inode, 8, BKEY_TYPE_strict_btree_checks) \
400 x(inode_generation, 9, BKEY_TYPE_strict_btree_checks) \
401 x(dirent, 10, BKEY_TYPE_strict_btree_checks) \
402 x(xattr, 11, BKEY_TYPE_strict_btree_checks) \
403 x(alloc, 12, BKEY_TYPE_strict_btree_checks) \
404 x(quota, 13, BKEY_TYPE_strict_btree_checks) \
405 x(stripe, 14, BKEY_TYPE_strict_btree_checks) \
406 x(reflink_p, 15, BKEY_TYPE_strict_btree_checks) \
407 x(reflink_v, 16, BKEY_TYPE_strict_btree_checks) \
408 x(inline_data, 17, BKEY_TYPE_strict_btree_checks) \
409 x(btree_ptr_v2, 18, BKEY_TYPE_strict_btree_checks) \
410 x(indirect_inline_data, 19, BKEY_TYPE_strict_btree_checks) \
411 x(alloc_v2, 20, BKEY_TYPE_strict_btree_checks) \
412 x(subvolume, 21, BKEY_TYPE_strict_btree_checks) \
413 x(snapshot, 22, BKEY_TYPE_strict_btree_checks) \
414 x(inode_v2, 23, BKEY_TYPE_strict_btree_checks) \
415 x(alloc_v3, 24, BKEY_TYPE_strict_btree_checks) \
416 x(set, 25, 0) \
417 x(lru, 26, BKEY_TYPE_strict_btree_checks) \
418 x(alloc_v4, 27, BKEY_TYPE_strict_btree_checks) \
419 x(backpointer, 28, BKEY_TYPE_strict_btree_checks) \
420 x(inode_v3, 29, BKEY_TYPE_strict_btree_checks) \
421 x(bucket_gens, 30, BKEY_TYPE_strict_btree_checks) \
422 x(snapshot_tree, 31, BKEY_TYPE_strict_btree_checks) \
423 x(logged_op_truncate, 32, BKEY_TYPE_strict_btree_checks) \
424 x(logged_op_finsert, 33, BKEY_TYPE_strict_btree_checks) \
425 x(accounting, 34, BKEY_TYPE_strict_btree_checks) \
426 x(inode_alloc_cursor, 35, BKEY_TYPE_strict_btree_checks)
427
428 enum bch_bkey_type {
429 #define x(name, nr, ...) KEY_TYPE_##name = nr,
430 BCH_BKEY_TYPES()
431 #undef x
432 KEY_TYPE_MAX,
433 };
434
435 struct bch_deleted {
436 struct bch_val v;
437 };
438
439 struct bch_whiteout {
440 struct bch_val v;
441 };
442
443 struct bch_error {
444 struct bch_val v;
445 };
446
447 struct bch_cookie {
448 struct bch_val v;
449 __le64 cookie;
450 };
451
452 struct bch_hash_whiteout {
453 struct bch_val v;
454 };
455
456 struct bch_set {
457 struct bch_val v;
458 };
459
460 /* 128 bits, sufficient for cryptographic MACs: */
461 struct bch_csum {
462 __le64 lo;
463 __le64 hi;
464 } __packed __aligned(8);
465
466 struct bch_backpointer {
467 struct bch_val v;
468 __u8 btree_id;
469 __u8 level;
470 __u8 data_type;
471 __u8 bucket_gen;
472 __u32 pad;
473 __u32 bucket_len;
474 struct bpos pos;
475 } __packed __aligned(8);
476
477 /* Optional/variable size superblock sections: */
478
479 struct bch_sb_field {
480 __u64 _data[0];
481 __le32 u64s;
482 __le32 type;
483 };
484
485 #define BCH_SB_FIELDS() \
486 x(journal, 0) \
487 x(members_v1, 1) \
488 x(crypt, 2) \
489 x(replicas_v0, 3) \
490 x(quota, 4) \
491 x(disk_groups, 5) \
492 x(clean, 6) \
493 x(replicas, 7) \
494 x(journal_seq_blacklist, 8) \
495 x(journal_v2, 9) \
496 x(counters, 10) \
497 x(members_v2, 11) \
498 x(errors, 12) \
499 x(ext, 13) \
500 x(downgrade, 14) \
501 x(recovery_passes, 15)
502
503 #include "alloc_background_format.h"
504 #include "dirent_format.h"
505 #include "disk_accounting_format.h"
506 #include "disk_groups_format.h"
507 #include "extents_format.h"
508 #include "ec_format.h"
509 #include "inode_format.h"
510 #include "journal_seq_blacklist_format.h"
511 #include "logged_ops_format.h"
512 #include "lru_format.h"
513 #include "quota_format.h"
514 #include "recovery_passes_format.h"
515 #include "reflink_format.h"
516 #include "replicas_format.h"
517 #include "snapshot_format.h"
518 #include "subvolume_format.h"
519 #include "sb-counters_format.h"
520 #include "sb-downgrade_format.h"
521 #include "sb-errors_format.h"
522 #include "sb-members_format.h"
523 #include "xattr_format.h"
524
525 enum bch_sb_field_type {
526 #define x(f, nr) BCH_SB_FIELD_##f = nr,
527 BCH_SB_FIELDS()
528 #undef x
529 BCH_SB_FIELD_NR
530 };
531
532 /*
533 * Most superblock fields are replicated in all device's superblocks - a few are
534 * not:
535 */
536 #define BCH_SINGLE_DEVICE_SB_FIELDS \
537 ((1U << BCH_SB_FIELD_journal)| \
538 (1U << BCH_SB_FIELD_journal_v2))
539
540 /* BCH_SB_FIELD_journal: */
541
542 struct bch_sb_field_journal {
543 struct bch_sb_field field;
544 __le64 buckets[];
545 };
546
547 struct bch_sb_field_journal_v2 {
548 struct bch_sb_field field;
549
550 struct bch_sb_field_journal_v2_entry {
551 __le64 start;
552 __le64 nr;
553 } d[];
554 };
555
556 /* BCH_SB_FIELD_crypt: */
557
558 struct nonce {
559 __le32 d[4];
560 };
561
562 struct bch_key {
563 __le64 key[4];
564 };
565
566 #define BCH_KEY_MAGIC \
567 (((__u64) 'b' << 0)|((__u64) 'c' << 8)| \
568 ((__u64) 'h' << 16)|((__u64) '*' << 24)| \
569 ((__u64) '*' << 32)|((__u64) 'k' << 40)| \
570 ((__u64) 'e' << 48)|((__u64) 'y' << 56))
571
572 struct bch_encrypted_key {
573 __le64 magic;
574 struct bch_key key;
575 };
576
577 /*
578 * If this field is present in the superblock, it stores an encryption key which
579 * is used encrypt all other data/metadata. The key will normally be encrypted
580 * with the key userspace provides, but if encryption has been turned off we'll
581 * just store the master key unencrypted in the superblock so we can access the
582 * previously encrypted data.
583 */
584 struct bch_sb_field_crypt {
585 struct bch_sb_field field;
586
587 __le64 flags;
588 __le64 kdf_flags;
589 struct bch_encrypted_key key;
590 };
591
592 LE64_BITMASK(BCH_CRYPT_KDF_TYPE, struct bch_sb_field_crypt, flags, 0, 4);
593
594 enum bch_kdf_types {
595 BCH_KDF_SCRYPT = 0,
596 BCH_KDF_NR = 1,
597 };
598
599 /* stored as base 2 log of scrypt params: */
600 LE64_BITMASK(BCH_KDF_SCRYPT_N, struct bch_sb_field_crypt, kdf_flags, 0, 16);
601 LE64_BITMASK(BCH_KDF_SCRYPT_R, struct bch_sb_field_crypt, kdf_flags, 16, 32);
602 LE64_BITMASK(BCH_KDF_SCRYPT_P, struct bch_sb_field_crypt, kdf_flags, 32, 48);
603
604 /*
605 * On clean shutdown, store btree roots and current journal sequence number in
606 * the superblock:
607 */
608 struct jset_entry {
609 __le16 u64s;
610 __u8 btree_id;
611 __u8 level;
612 __u8 type; /* designates what this jset holds */
613 __u8 pad[3];
614
615 struct bkey_i start[0];
616 __u64 _data[];
617 };
618
619 struct bch_sb_field_clean {
620 struct bch_sb_field field;
621
622 __le32 flags;
623 __le16 _read_clock; /* no longer used */
624 __le16 _write_clock;
625 __le64 journal_seq;
626
627 struct jset_entry start[0];
628 __u64 _data[];
629 };
630
631 struct bch_sb_field_ext {
632 struct bch_sb_field field;
633 __le64 recovery_passes_required[2];
634 __le64 errors_silent[8];
635 __le64 btrees_lost_data;
636 };
637
638 /* Superblock: */
639
640 /*
641 * New versioning scheme:
642 * One common version number for all on disk data structures - superblock, btree
643 * nodes, journal entries
644 */
645 #define BCH_VERSION_MAJOR(_v) ((__u16) ((_v) >> 10))
646 #define BCH_VERSION_MINOR(_v) ((__u16) ((_v) & ~(~0U << 10)))
647 #define BCH_VERSION(_major, _minor) (((_major) << 10)|(_minor) << 0)
648
649 /*
650 * field 1: version name
651 * field 2: BCH_VERSION(major, minor)
652 * field 3: recovery passess required on upgrade
653 */
654 #define BCH_METADATA_VERSIONS() \
655 x(bkey_renumber, BCH_VERSION(0, 10)) \
656 x(inode_btree_change, BCH_VERSION(0, 11)) \
657 x(snapshot, BCH_VERSION(0, 12)) \
658 x(inode_backpointers, BCH_VERSION(0, 13)) \
659 x(btree_ptr_sectors_written, BCH_VERSION(0, 14)) \
660 x(snapshot_2, BCH_VERSION(0, 15)) \
661 x(reflink_p_fix, BCH_VERSION(0, 16)) \
662 x(subvol_dirent, BCH_VERSION(0, 17)) \
663 x(inode_v2, BCH_VERSION(0, 18)) \
664 x(freespace, BCH_VERSION(0, 19)) \
665 x(alloc_v4, BCH_VERSION(0, 20)) \
666 x(new_data_types, BCH_VERSION(0, 21)) \
667 x(backpointers, BCH_VERSION(0, 22)) \
668 x(inode_v3, BCH_VERSION(0, 23)) \
669 x(unwritten_extents, BCH_VERSION(0, 24)) \
670 x(bucket_gens, BCH_VERSION(0, 25)) \
671 x(lru_v2, BCH_VERSION(0, 26)) \
672 x(fragmentation_lru, BCH_VERSION(0, 27)) \
673 x(no_bps_in_alloc_keys, BCH_VERSION(0, 28)) \
674 x(snapshot_trees, BCH_VERSION(0, 29)) \
675 x(major_minor, BCH_VERSION(1, 0)) \
676 x(snapshot_skiplists, BCH_VERSION(1, 1)) \
677 x(deleted_inodes, BCH_VERSION(1, 2)) \
678 x(rebalance_work, BCH_VERSION(1, 3)) \
679 x(member_seq, BCH_VERSION(1, 4)) \
680 x(subvolume_fs_parent, BCH_VERSION(1, 5)) \
681 x(btree_subvolume_children, BCH_VERSION(1, 6)) \
682 x(mi_btree_bitmap, BCH_VERSION(1, 7)) \
683 x(bucket_stripe_sectors, BCH_VERSION(1, 8)) \
684 x(disk_accounting_v2, BCH_VERSION(1, 9)) \
685 x(disk_accounting_v3, BCH_VERSION(1, 10)) \
686 x(disk_accounting_inum, BCH_VERSION(1, 11)) \
687 x(rebalance_work_acct_fix, BCH_VERSION(1, 12)) \
688 x(inode_has_child_snapshots, BCH_VERSION(1, 13)) \
689 x(backpointer_bucket_gen, BCH_VERSION(1, 14)) \
690 x(disk_accounting_big_endian, BCH_VERSION(1, 15)) \
691 x(reflink_p_may_update_opts, BCH_VERSION(1, 16)) \
692 x(inode_depth, BCH_VERSION(1, 17)) \
693 x(persistent_inode_cursors, BCH_VERSION(1, 18)) \
694 x(autofix_errors, BCH_VERSION(1, 19)) \
695 x(directory_size, BCH_VERSION(1, 20)) \
696 x(cached_backpointers, BCH_VERSION(1, 21)) \
697 x(stripe_backpointers, BCH_VERSION(1, 22)) \
698 x(stripe_lru, BCH_VERSION(1, 23)) \
699 x(casefolding, BCH_VERSION(1, 24)) \
700 x(extent_flags, BCH_VERSION(1, 25)) \
701 x(snapshot_deletion_v2, BCH_VERSION(1, 26)) \
702 x(fast_device_removal, BCH_VERSION(1, 27)) \
703 x(inode_has_case_insensitive, BCH_VERSION(1, 28))
704
705 enum bcachefs_metadata_version {
706 bcachefs_metadata_version_min = 9,
707 #define x(t, n) bcachefs_metadata_version_##t = n,
708 BCH_METADATA_VERSIONS()
709 #undef x
710 bcachefs_metadata_version_max
711 };
712
713 static const __maybe_unused
714 unsigned bcachefs_metadata_required_upgrade_below = bcachefs_metadata_version_rebalance_work;
715
716 #define bcachefs_metadata_version_current (bcachefs_metadata_version_max - 1)
717
718 #define BCH_SB_SECTOR 8
719
720 #define BCH_SB_LAYOUT_SIZE_BITS_MAX 16 /* 32 MB */
721
722 struct bch_sb_layout {
723 __uuid_t magic; /* bcachefs superblock UUID */
724 __u8 layout_type;
725 __u8 sb_max_size_bits; /* base 2 of 512 byte sectors */
726 __u8 nr_superblocks;
727 __u8 pad[5];
728 __le64 sb_offset[61];
729 } __packed __aligned(8);
730
731 #define BCH_SB_LAYOUT_SECTOR 7
732
733 /*
734 * @offset - sector where this sb was written
735 * @version - on disk format version
736 * @version_min - Oldest metadata version this filesystem contains; so we can
737 * safely drop compatibility code and refuse to mount filesystems
738 * we'd need it for
739 * @magic - identifies as a bcachefs superblock (BCHFS_MAGIC)
740 * @seq - incremented each time superblock is written
741 * @uuid - used for generating various magic numbers and identifying
742 * member devices, never changes
743 * @user_uuid - user visible UUID, may be changed
744 * @label - filesystem label
745 * @seq - identifies most recent superblock, incremented each time
746 * superblock is written
747 * @features - enabled incompatible features
748 */
749 struct bch_sb {
750 struct bch_csum csum;
751 __le16 version;
752 __le16 version_min;
753 __le16 pad[2];
754 __uuid_t magic;
755 __uuid_t uuid;
756 __uuid_t user_uuid;
757 __u8 label[BCH_SB_LABEL_SIZE];
758 __le64 offset;
759 __le64 seq;
760
761 __le16 block_size;
762 __u8 dev_idx;
763 __u8 nr_devices;
764 __le32 u64s;
765
766 __le64 time_base_lo;
767 __le32 time_base_hi;
768 __le32 time_precision;
769
770 __le64 flags[7];
771 __le64 write_time;
772 __le64 features[2];
773 __le64 compat[2];
774
775 struct bch_sb_layout layout;
776
777 struct bch_sb_field start[0];
778 __le64 _data[];
779 } __packed __aligned(8);
780
781 /*
782 * Flags:
783 * BCH_SB_INITALIZED - set on first mount
784 * BCH_SB_CLEAN - did we shut down cleanly? Just a hint, doesn't affect
785 * behaviour of mount/recovery path:
786 * BCH_SB_INODE_32BIT - limit inode numbers to 32 bits
787 * BCH_SB_128_BIT_MACS - 128 bit macs instead of 80
788 * BCH_SB_ENCRYPTION_TYPE - if nonzero encryption is enabled; overrides
789 * DATA/META_CSUM_TYPE. Also indicates encryption
790 * algorithm in use, if/when we get more than one
791 */
792
793 LE16_BITMASK(BCH_SB_BLOCK_SIZE, struct bch_sb, block_size, 0, 16);
794
795 LE64_BITMASK(BCH_SB_INITIALIZED, struct bch_sb, flags[0], 0, 1);
796 LE64_BITMASK(BCH_SB_CLEAN, struct bch_sb, flags[0], 1, 2);
797 LE64_BITMASK(BCH_SB_CSUM_TYPE, struct bch_sb, flags[0], 2, 8);
798 LE64_BITMASK(BCH_SB_ERROR_ACTION, struct bch_sb, flags[0], 8, 12);
799
800 LE64_BITMASK(BCH_SB_BTREE_NODE_SIZE, struct bch_sb, flags[0], 12, 28);
801
802 LE64_BITMASK(BCH_SB_GC_RESERVE, struct bch_sb, flags[0], 28, 33);
803 LE64_BITMASK(BCH_SB_ROOT_RESERVE, struct bch_sb, flags[0], 33, 40);
804
805 LE64_BITMASK(BCH_SB_META_CSUM_TYPE, struct bch_sb, flags[0], 40, 44);
806 LE64_BITMASK(BCH_SB_DATA_CSUM_TYPE, struct bch_sb, flags[0], 44, 48);
807
808 LE64_BITMASK(BCH_SB_META_REPLICAS_WANT, struct bch_sb, flags[0], 48, 52);
809 LE64_BITMASK(BCH_SB_DATA_REPLICAS_WANT, struct bch_sb, flags[0], 52, 56);
810
811 LE64_BITMASK(BCH_SB_POSIX_ACL, struct bch_sb, flags[0], 56, 57);
812 LE64_BITMASK(BCH_SB_USRQUOTA, struct bch_sb, flags[0], 57, 58);
813 LE64_BITMASK(BCH_SB_GRPQUOTA, struct bch_sb, flags[0], 58, 59);
814 LE64_BITMASK(BCH_SB_PRJQUOTA, struct bch_sb, flags[0], 59, 60);
815
816 LE64_BITMASK(BCH_SB_HAS_ERRORS, struct bch_sb, flags[0], 60, 61);
817 LE64_BITMASK(BCH_SB_HAS_TOPOLOGY_ERRORS,struct bch_sb, flags[0], 61, 62);
818
819 LE64_BITMASK(BCH_SB_BIG_ENDIAN, struct bch_sb, flags[0], 62, 63);
820 LE64_BITMASK(BCH_SB_PROMOTE_WHOLE_EXTENTS,
821 struct bch_sb, flags[0], 63, 64);
822
823 LE64_BITMASK(BCH_SB_STR_HASH_TYPE, struct bch_sb, flags[1], 0, 4);
824 LE64_BITMASK(BCH_SB_COMPRESSION_TYPE_LO,struct bch_sb, flags[1], 4, 8);
825 LE64_BITMASK(BCH_SB_INODE_32BIT, struct bch_sb, flags[1], 8, 9);
826
827 LE64_BITMASK(BCH_SB_128_BIT_MACS, struct bch_sb, flags[1], 9, 10);
828 LE64_BITMASK(BCH_SB_ENCRYPTION_TYPE, struct bch_sb, flags[1], 10, 14);
829
830 /*
831 * Max size of an extent that may require bouncing to read or write
832 * (checksummed, compressed): 64k
833 */
834 LE64_BITMASK(BCH_SB_ENCODED_EXTENT_MAX_BITS,
835 struct bch_sb, flags[1], 14, 20);
836
837 LE64_BITMASK(BCH_SB_META_REPLICAS_REQ, struct bch_sb, flags[1], 20, 24);
838 LE64_BITMASK(BCH_SB_DATA_REPLICAS_REQ, struct bch_sb, flags[1], 24, 28);
839
840 LE64_BITMASK(BCH_SB_PROMOTE_TARGET, struct bch_sb, flags[1], 28, 40);
841 LE64_BITMASK(BCH_SB_FOREGROUND_TARGET, struct bch_sb, flags[1], 40, 52);
842 LE64_BITMASK(BCH_SB_BACKGROUND_TARGET, struct bch_sb, flags[1], 52, 64);
843
844 LE64_BITMASK(BCH_SB_BACKGROUND_COMPRESSION_TYPE_LO,
845 struct bch_sb, flags[2], 0, 4);
846 LE64_BITMASK(BCH_SB_GC_RESERVE_BYTES, struct bch_sb, flags[2], 4, 64);
847
848 LE64_BITMASK(BCH_SB_ERASURE_CODE, struct bch_sb, flags[3], 0, 16);
849 LE64_BITMASK(BCH_SB_METADATA_TARGET, struct bch_sb, flags[3], 16, 28);
850 LE64_BITMASK(BCH_SB_SHARD_INUMS, struct bch_sb, flags[3], 28, 29);
851 LE64_BITMASK(BCH_SB_INODES_USE_KEY_CACHE,struct bch_sb, flags[3], 29, 30);
852 LE64_BITMASK(BCH_SB_JOURNAL_FLUSH_DELAY,struct bch_sb, flags[3], 30, 62);
853 LE64_BITMASK(BCH_SB_JOURNAL_FLUSH_DISABLED,struct bch_sb, flags[3], 62, 63);
854 LE64_BITMASK(BCH_SB_MULTI_DEVICE, struct bch_sb, flags[3], 63, 64);
855 LE64_BITMASK(BCH_SB_JOURNAL_RECLAIM_DELAY,struct bch_sb, flags[4], 0, 32);
856 LE64_BITMASK(BCH_SB_JOURNAL_TRANSACTION_NAMES,struct bch_sb, flags[4], 32, 33);
857 LE64_BITMASK(BCH_SB_NOCOW, struct bch_sb, flags[4], 33, 34);
858 LE64_BITMASK(BCH_SB_WRITE_BUFFER_SIZE, struct bch_sb, flags[4], 34, 54);
859 LE64_BITMASK(BCH_SB_VERSION_UPGRADE, struct bch_sb, flags[4], 54, 56);
860
861 LE64_BITMASK(BCH_SB_COMPRESSION_TYPE_HI,struct bch_sb, flags[4], 56, 60);
862 LE64_BITMASK(BCH_SB_BACKGROUND_COMPRESSION_TYPE_HI,
863 struct bch_sb, flags[4], 60, 64);
864
865 LE64_BITMASK(BCH_SB_VERSION_UPGRADE_COMPLETE,
866 struct bch_sb, flags[5], 0, 16);
867 LE64_BITMASK(BCH_SB_ALLOCATOR_STUCK_TIMEOUT,
868 struct bch_sb, flags[5], 16, 32);
869 LE64_BITMASK(BCH_SB_VERSION_INCOMPAT, struct bch_sb, flags[5], 32, 48);
870 LE64_BITMASK(BCH_SB_VERSION_INCOMPAT_ALLOWED,
871 struct bch_sb, flags[5], 48, 64);
872 LE64_BITMASK(BCH_SB_SHARD_INUMS_NBITS, struct bch_sb, flags[6], 0, 4);
873 LE64_BITMASK(BCH_SB_WRITE_ERROR_TIMEOUT,struct bch_sb, flags[6], 4, 14);
874 LE64_BITMASK(BCH_SB_CSUM_ERR_RETRY_NR, struct bch_sb, flags[6], 14, 20);
875 LE64_BITMASK(BCH_SB_DEGRADED_ACTION, struct bch_sb, flags[6], 20, 22);
876 LE64_BITMASK(BCH_SB_CASEFOLD, struct bch_sb, flags[6], 22, 23);
877 LE64_BITMASK(BCH_SB_REBALANCE_AC_ONLY, struct bch_sb, flags[6], 23, 24);
878
BCH_SB_COMPRESSION_TYPE(const struct bch_sb * sb)879 static inline __u64 BCH_SB_COMPRESSION_TYPE(const struct bch_sb *sb)
880 {
881 return BCH_SB_COMPRESSION_TYPE_LO(sb) | (BCH_SB_COMPRESSION_TYPE_HI(sb) << 4);
882 }
883
SET_BCH_SB_COMPRESSION_TYPE(struct bch_sb * sb,__u64 v)884 static inline void SET_BCH_SB_COMPRESSION_TYPE(struct bch_sb *sb, __u64 v)
885 {
886 SET_BCH_SB_COMPRESSION_TYPE_LO(sb, v);
887 SET_BCH_SB_COMPRESSION_TYPE_HI(sb, v >> 4);
888 }
889
BCH_SB_BACKGROUND_COMPRESSION_TYPE(const struct bch_sb * sb)890 static inline __u64 BCH_SB_BACKGROUND_COMPRESSION_TYPE(const struct bch_sb *sb)
891 {
892 return BCH_SB_BACKGROUND_COMPRESSION_TYPE_LO(sb) |
893 (BCH_SB_BACKGROUND_COMPRESSION_TYPE_HI(sb) << 4);
894 }
895
SET_BCH_SB_BACKGROUND_COMPRESSION_TYPE(struct bch_sb * sb,__u64 v)896 static inline void SET_BCH_SB_BACKGROUND_COMPRESSION_TYPE(struct bch_sb *sb, __u64 v)
897 {
898 SET_BCH_SB_BACKGROUND_COMPRESSION_TYPE_LO(sb, v);
899 SET_BCH_SB_BACKGROUND_COMPRESSION_TYPE_HI(sb, v >> 4);
900 }
901
902 /*
903 * Features:
904 *
905 * journal_seq_blacklist_v3: gates BCH_SB_FIELD_journal_seq_blacklist
906 * reflink: gates KEY_TYPE_reflink
907 * inline_data: gates KEY_TYPE_inline_data
908 * new_siphash: gates BCH_STR_HASH_siphash
909 * new_extent_overwrite: gates BTREE_NODE_NEW_EXTENT_OVERWRITE
910 */
911 #define BCH_SB_FEATURES() \
912 x(lz4, 0) \
913 x(gzip, 1) \
914 x(zstd, 2) \
915 x(atomic_nlink, 3) \
916 x(ec, 4) \
917 x(journal_seq_blacklist_v3, 5) \
918 x(reflink, 6) \
919 x(new_siphash, 7) \
920 x(inline_data, 8) \
921 x(new_extent_overwrite, 9) \
922 x(incompressible, 10) \
923 x(btree_ptr_v2, 11) \
924 x(extents_above_btree_updates, 12) \
925 x(btree_updates_journalled, 13) \
926 x(reflink_inline_data, 14) \
927 x(new_varint, 15) \
928 x(journal_no_flush, 16) \
929 x(alloc_v2, 17) \
930 x(extents_across_btree_nodes, 18) \
931 x(incompat_version_field, 19) \
932 x(casefolding, 20) \
933 x(no_alloc_info, 21) \
934 x(small_image, 22)
935
936 #define BCH_SB_FEATURES_ALWAYS \
937 (BIT_ULL(BCH_FEATURE_new_extent_overwrite)| \
938 BIT_ULL(BCH_FEATURE_extents_above_btree_updates)|\
939 BIT_ULL(BCH_FEATURE_btree_updates_journalled)|\
940 BIT_ULL(BCH_FEATURE_alloc_v2)|\
941 BIT_ULL(BCH_FEATURE_extents_across_btree_nodes))
942
943 #define BCH_SB_FEATURES_ALL \
944 (BCH_SB_FEATURES_ALWAYS| \
945 BIT_ULL(BCH_FEATURE_new_siphash)| \
946 BIT_ULL(BCH_FEATURE_btree_ptr_v2)| \
947 BIT_ULL(BCH_FEATURE_new_varint)| \
948 BIT_ULL(BCH_FEATURE_journal_no_flush)| \
949 BIT_ULL(BCH_FEATURE_incompat_version_field))
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(fix_safe, 1) \
991 x(panic, 2) \
992 x(ro, 3)
993
994 enum bch_error_actions {
995 #define x(t, n) BCH_ON_ERROR_##t = n,
996 BCH_ERROR_ACTIONS()
997 #undef x
998 BCH_ON_ERROR_NR
999 };
1000
1001 #define BCH_DEGRADED_ACTIONS() \
1002 x(ask, 0) \
1003 x(yes, 1) \
1004 x(very, 2) \
1005 x(no, 3)
1006
1007 enum bch_degraded_actions {
1008 #define x(t, n) BCH_DEGRADED_##t = n,
1009 BCH_DEGRADED_ACTIONS()
1010 #undef x
1011 BCH_DEGRADED_ACTIONS_NR
1012 };
1013
1014 #define BCH_STR_HASH_TYPES() \
1015 x(crc32c, 0) \
1016 x(crc64, 1) \
1017 x(siphash_old, 2) \
1018 x(siphash, 3)
1019
1020 enum bch_str_hash_type {
1021 #define x(t, n) BCH_STR_HASH_##t = n,
1022 BCH_STR_HASH_TYPES()
1023 #undef x
1024 BCH_STR_HASH_NR
1025 };
1026
1027 #define BCH_STR_HASH_OPTS() \
1028 x(crc32c, 0) \
1029 x(crc64, 1) \
1030 x(siphash, 2)
1031
1032 enum bch_str_hash_opts {
1033 #define x(t, n) BCH_STR_HASH_OPT_##t = n,
1034 BCH_STR_HASH_OPTS()
1035 #undef x
1036 BCH_STR_HASH_OPT_NR
1037 };
1038
1039 #define BCH_CSUM_TYPES() \
1040 x(none, 0) \
1041 x(crc32c_nonzero, 1) \
1042 x(crc64_nonzero, 2) \
1043 x(chacha20_poly1305_80, 3) \
1044 x(chacha20_poly1305_128, 4) \
1045 x(crc32c, 5) \
1046 x(crc64, 6) \
1047 x(xxhash, 7)
1048
1049 enum bch_csum_type {
1050 #define x(t, n) BCH_CSUM_##t = n,
1051 BCH_CSUM_TYPES()
1052 #undef x
1053 BCH_CSUM_NR
1054 };
1055
1056 static const __maybe_unused unsigned bch_crc_bytes[] = {
1057 [BCH_CSUM_none] = 0,
1058 [BCH_CSUM_crc32c_nonzero] = 4,
1059 [BCH_CSUM_crc32c] = 4,
1060 [BCH_CSUM_crc64_nonzero] = 8,
1061 [BCH_CSUM_crc64] = 8,
1062 [BCH_CSUM_xxhash] = 8,
1063 [BCH_CSUM_chacha20_poly1305_80] = 10,
1064 [BCH_CSUM_chacha20_poly1305_128] = 16,
1065 };
1066
bch2_csum_type_is_encryption(enum bch_csum_type type)1067 static inline _Bool bch2_csum_type_is_encryption(enum bch_csum_type type)
1068 {
1069 switch (type) {
1070 case BCH_CSUM_chacha20_poly1305_80:
1071 case BCH_CSUM_chacha20_poly1305_128:
1072 return true;
1073 default:
1074 return false;
1075 }
1076 }
1077
1078 #define BCH_CSUM_OPTS() \
1079 x(none, 0) \
1080 x(crc32c, 1) \
1081 x(crc64, 2) \
1082 x(xxhash, 3)
1083
1084 enum bch_csum_opt {
1085 #define x(t, n) BCH_CSUM_OPT_##t = n,
1086 BCH_CSUM_OPTS()
1087 #undef x
1088 BCH_CSUM_OPT_NR
1089 };
1090
1091 #define BCH_COMPRESSION_TYPES() \
1092 x(none, 0) \
1093 x(lz4_old, 1) \
1094 x(gzip, 2) \
1095 x(lz4, 3) \
1096 x(zstd, 4) \
1097 x(incompressible, 5)
1098
1099 enum bch_compression_type {
1100 #define x(t, n) BCH_COMPRESSION_TYPE_##t = n,
1101 BCH_COMPRESSION_TYPES()
1102 #undef x
1103 BCH_COMPRESSION_TYPE_NR
1104 };
1105
1106 #define BCH_COMPRESSION_OPTS() \
1107 x(none, 0) \
1108 x(lz4, 1) \
1109 x(gzip, 2) \
1110 x(zstd, 3)
1111
1112 enum bch_compression_opts {
1113 #define x(t, n) BCH_COMPRESSION_OPT_##t = n,
1114 BCH_COMPRESSION_OPTS()
1115 #undef x
1116 BCH_COMPRESSION_OPT_NR
1117 };
1118
1119 /*
1120 * Magic numbers
1121 *
1122 * The various other data structures have their own magic numbers, which are
1123 * xored with the first part of the cache set's UUID
1124 */
1125
1126 #define BCACHE_MAGIC \
1127 UUID_INIT(0xc68573f6, 0x4e1a, 0x45ca, \
1128 0x82, 0x65, 0xf5, 0x7f, 0x48, 0xba, 0x6d, 0x81)
1129 #define BCHFS_MAGIC \
1130 UUID_INIT(0xc68573f6, 0x66ce, 0x90a9, \
1131 0xd9, 0x6a, 0x60, 0xcf, 0x80, 0x3d, 0xf7, 0xef)
1132
1133 #define BCACHEFS_STATFS_MAGIC BCACHEFS_SUPER_MAGIC
1134
1135 #define JSET_MAGIC __cpu_to_le64(0x245235c1a3625032ULL)
1136 #define BSET_MAGIC __cpu_to_le64(0x90135c78b99e07f5ULL)
1137
__bch2_sb_magic(struct bch_sb * sb)1138 static inline __le64 __bch2_sb_magic(struct bch_sb *sb)
1139 {
1140 __le64 ret;
1141
1142 memcpy(&ret, &sb->uuid, sizeof(ret));
1143 return ret;
1144 }
1145
__jset_magic(struct bch_sb * sb)1146 static inline __u64 __jset_magic(struct bch_sb *sb)
1147 {
1148 return __le64_to_cpu(__bch2_sb_magic(sb) ^ JSET_MAGIC);
1149 }
1150
__bset_magic(struct bch_sb * sb)1151 static inline __u64 __bset_magic(struct bch_sb *sb)
1152 {
1153 return __le64_to_cpu(__bch2_sb_magic(sb) ^ BSET_MAGIC);
1154 }
1155
1156 /* Journal */
1157
1158 #define JSET_KEYS_U64s (sizeof(struct jset_entry) / sizeof(__u64))
1159
1160 #define BCH_JSET_ENTRY_TYPES() \
1161 x(btree_keys, 0) \
1162 x(btree_root, 1) \
1163 x(prio_ptrs, 2) \
1164 x(blacklist, 3) \
1165 x(blacklist_v2, 4) \
1166 x(usage, 5) \
1167 x(data_usage, 6) \
1168 x(clock, 7) \
1169 x(dev_usage, 8) \
1170 x(log, 9) \
1171 x(overwrite, 10) \
1172 x(write_buffer_keys, 11) \
1173 x(datetime, 12) \
1174 x(log_bkey, 13)
1175
1176 enum bch_jset_entry_type {
1177 #define x(f, nr) BCH_JSET_ENTRY_##f = nr,
1178 BCH_JSET_ENTRY_TYPES()
1179 #undef x
1180 BCH_JSET_ENTRY_NR
1181 };
1182
jset_entry_is_key(struct jset_entry * e)1183 static inline bool jset_entry_is_key(struct jset_entry *e)
1184 {
1185 switch (e->type) {
1186 case BCH_JSET_ENTRY_btree_keys:
1187 case BCH_JSET_ENTRY_btree_root:
1188 case BCH_JSET_ENTRY_write_buffer_keys:
1189 return true;
1190 }
1191
1192 return false;
1193 }
1194
1195 /*
1196 * Journal sequence numbers can be blacklisted: bsets record the max sequence
1197 * number of all the journal entries they contain updates for, so that on
1198 * recovery we can ignore those bsets that contain index updates newer that what
1199 * made it into the journal.
1200 *
1201 * This means that we can't reuse that journal_seq - we have to skip it, and
1202 * then record that we skipped it so that the next time we crash and recover we
1203 * don't think there was a missing journal entry.
1204 */
1205 struct jset_entry_blacklist {
1206 struct jset_entry entry;
1207 __le64 seq;
1208 };
1209
1210 struct jset_entry_blacklist_v2 {
1211 struct jset_entry entry;
1212 __le64 start;
1213 __le64 end;
1214 };
1215
1216 #define BCH_FS_USAGE_TYPES() \
1217 x(reserved, 0) \
1218 x(inodes, 1) \
1219 x(key_version, 2)
1220
1221 enum bch_fs_usage_type {
1222 #define x(f, nr) BCH_FS_USAGE_##f = nr,
1223 BCH_FS_USAGE_TYPES()
1224 #undef x
1225 BCH_FS_USAGE_NR
1226 };
1227
1228 struct jset_entry_usage {
1229 struct jset_entry entry;
1230 __le64 v;
1231 } __packed;
1232
1233 struct jset_entry_data_usage {
1234 struct jset_entry entry;
1235 __le64 v;
1236 struct bch_replicas_entry_v1 r;
1237 } __packed;
1238
1239 struct jset_entry_clock {
1240 struct jset_entry entry;
1241 __u8 rw;
1242 __u8 pad[7];
1243 __le64 time;
1244 } __packed;
1245
1246 struct jset_entry_dev_usage_type {
1247 __le64 buckets;
1248 __le64 sectors;
1249 __le64 fragmented;
1250 } __packed;
1251
1252 struct jset_entry_dev_usage {
1253 struct jset_entry entry;
1254 __le32 dev;
1255 __u32 pad;
1256
1257 __le64 _buckets_ec; /* No longer used */
1258 __le64 _buckets_unavailable; /* No longer used */
1259
1260 struct jset_entry_dev_usage_type d[];
1261 };
1262
jset_entry_dev_usage_nr_types(struct jset_entry_dev_usage * u)1263 static inline unsigned jset_entry_dev_usage_nr_types(struct jset_entry_dev_usage *u)
1264 {
1265 return (vstruct_bytes(&u->entry) - sizeof(struct jset_entry_dev_usage)) /
1266 sizeof(struct jset_entry_dev_usage_type);
1267 }
1268
1269 struct jset_entry_log {
1270 struct jset_entry entry;
1271 u8 d[];
1272 } __packed __aligned(8);
1273
jset_entry_log_msg_bytes(struct jset_entry_log * l)1274 static inline unsigned jset_entry_log_msg_bytes(struct jset_entry_log *l)
1275 {
1276 unsigned b = vstruct_bytes(&l->entry) - offsetof(struct jset_entry_log, d);
1277
1278 while (b && !l->d[b - 1])
1279 --b;
1280 return b;
1281 }
1282
1283 struct jset_entry_datetime {
1284 struct jset_entry entry;
1285 __le64 seconds;
1286 } __packed __aligned(8);
1287
1288 /*
1289 * On disk format for a journal entry:
1290 * seq is monotonically increasing; every journal entry has its own unique
1291 * sequence number.
1292 *
1293 * last_seq is the oldest journal entry that still has keys the btree hasn't
1294 * flushed to disk yet.
1295 *
1296 * version is for on disk format changes.
1297 */
1298 struct jset {
1299 struct bch_csum csum;
1300
1301 __le64 magic;
1302 __le64 seq;
1303 __le32 version;
1304 __le32 flags;
1305
1306 __le32 u64s; /* size of d[] in u64s */
1307
1308 __u8 encrypted_start[0];
1309
1310 __le16 _read_clock; /* no longer used */
1311 __le16 _write_clock;
1312
1313 /* Sequence number of oldest dirty journal entry */
1314 __le64 last_seq;
1315
1316
1317 struct jset_entry start[0];
1318 __u64 _data[];
1319 } __packed __aligned(8);
1320
1321 LE32_BITMASK(JSET_CSUM_TYPE, struct jset, flags, 0, 4);
1322 LE32_BITMASK(JSET_BIG_ENDIAN, struct jset, flags, 4, 5);
1323 LE32_BITMASK(JSET_NO_FLUSH, struct jset, flags, 5, 6);
1324
1325 #define BCH_JOURNAL_BUCKETS_MIN 8
1326
1327 /* Btree: */
1328
1329 enum btree_id_flags {
1330 BTREE_IS_extents = BIT(0),
1331 BTREE_IS_snapshots = BIT(1),
1332 BTREE_IS_snapshot_field = BIT(2),
1333 BTREE_IS_data = BIT(3),
1334 BTREE_IS_write_buffer = BIT(4),
1335 };
1336
1337 #define BCH_BTREE_IDS() \
1338 x(extents, 0, \
1339 BTREE_IS_extents| \
1340 BTREE_IS_snapshots| \
1341 BTREE_IS_data, \
1342 BIT_ULL(KEY_TYPE_whiteout)| \
1343 BIT_ULL(KEY_TYPE_error)| \
1344 BIT_ULL(KEY_TYPE_cookie)| \
1345 BIT_ULL(KEY_TYPE_extent)| \
1346 BIT_ULL(KEY_TYPE_reservation)| \
1347 BIT_ULL(KEY_TYPE_reflink_p)| \
1348 BIT_ULL(KEY_TYPE_inline_data)) \
1349 x(inodes, 1, \
1350 BTREE_IS_snapshots, \
1351 BIT_ULL(KEY_TYPE_whiteout)| \
1352 BIT_ULL(KEY_TYPE_inode)| \
1353 BIT_ULL(KEY_TYPE_inode_v2)| \
1354 BIT_ULL(KEY_TYPE_inode_v3)| \
1355 BIT_ULL(KEY_TYPE_inode_generation)) \
1356 x(dirents, 2, \
1357 BTREE_IS_snapshots, \
1358 BIT_ULL(KEY_TYPE_whiteout)| \
1359 BIT_ULL(KEY_TYPE_hash_whiteout)| \
1360 BIT_ULL(KEY_TYPE_dirent)) \
1361 x(xattrs, 3, \
1362 BTREE_IS_snapshots, \
1363 BIT_ULL(KEY_TYPE_whiteout)| \
1364 BIT_ULL(KEY_TYPE_cookie)| \
1365 BIT_ULL(KEY_TYPE_hash_whiteout)| \
1366 BIT_ULL(KEY_TYPE_xattr)) \
1367 x(alloc, 4, 0, \
1368 BIT_ULL(KEY_TYPE_alloc)| \
1369 BIT_ULL(KEY_TYPE_alloc_v2)| \
1370 BIT_ULL(KEY_TYPE_alloc_v3)| \
1371 BIT_ULL(KEY_TYPE_alloc_v4)) \
1372 x(quotas, 5, 0, \
1373 BIT_ULL(KEY_TYPE_quota)) \
1374 x(stripes, 6, 0, \
1375 BIT_ULL(KEY_TYPE_stripe)) \
1376 x(reflink, 7, \
1377 BTREE_IS_extents| \
1378 BTREE_IS_data, \
1379 BIT_ULL(KEY_TYPE_reflink_v)| \
1380 BIT_ULL(KEY_TYPE_indirect_inline_data)| \
1381 BIT_ULL(KEY_TYPE_error)) \
1382 x(subvolumes, 8, 0, \
1383 BIT_ULL(KEY_TYPE_subvolume)) \
1384 x(snapshots, 9, 0, \
1385 BIT_ULL(KEY_TYPE_snapshot)) \
1386 x(lru, 10, \
1387 BTREE_IS_write_buffer, \
1388 BIT_ULL(KEY_TYPE_set)) \
1389 x(freespace, 11, \
1390 BTREE_IS_extents, \
1391 BIT_ULL(KEY_TYPE_set)) \
1392 x(need_discard, 12, 0, \
1393 BIT_ULL(KEY_TYPE_set)) \
1394 x(backpointers, 13, \
1395 BTREE_IS_write_buffer, \
1396 BIT_ULL(KEY_TYPE_backpointer)) \
1397 x(bucket_gens, 14, 0, \
1398 BIT_ULL(KEY_TYPE_bucket_gens)) \
1399 x(snapshot_trees, 15, 0, \
1400 BIT_ULL(KEY_TYPE_snapshot_tree)) \
1401 x(deleted_inodes, 16, \
1402 BTREE_IS_snapshot_field| \
1403 BTREE_IS_write_buffer, \
1404 BIT_ULL(KEY_TYPE_set)) \
1405 x(logged_ops, 17, 0, \
1406 BIT_ULL(KEY_TYPE_logged_op_truncate)| \
1407 BIT_ULL(KEY_TYPE_logged_op_finsert)| \
1408 BIT_ULL(KEY_TYPE_inode_alloc_cursor)) \
1409 x(rebalance_work, 18, \
1410 BTREE_IS_snapshot_field| \
1411 BTREE_IS_write_buffer, \
1412 BIT_ULL(KEY_TYPE_set)|BIT_ULL(KEY_TYPE_cookie)) \
1413 x(subvolume_children, 19, 0, \
1414 BIT_ULL(KEY_TYPE_set)) \
1415 x(accounting, 20, \
1416 BTREE_IS_snapshot_field| \
1417 BTREE_IS_write_buffer, \
1418 BIT_ULL(KEY_TYPE_accounting)) \
1419
1420 enum btree_id {
1421 #define x(name, nr, ...) BTREE_ID_##name = nr,
1422 BCH_BTREE_IDS()
1423 #undef x
1424 BTREE_ID_NR
1425 };
1426
1427 /*
1428 * Maximum number of btrees that we will _ever_ have under the current scheme,
1429 * where we refer to them with 64 bit bitfields - and we also need a bit for
1430 * the interior btree node type:
1431 */
1432 #define BTREE_ID_NR_MAX 63
1433
btree_id_is_alloc(enum btree_id id)1434 static inline bool btree_id_is_alloc(enum btree_id id)
1435 {
1436 switch (id) {
1437 case BTREE_ID_alloc:
1438 case BTREE_ID_backpointers:
1439 case BTREE_ID_need_discard:
1440 case BTREE_ID_freespace:
1441 case BTREE_ID_bucket_gens:
1442 case BTREE_ID_lru:
1443 case BTREE_ID_accounting:
1444 return true;
1445 default:
1446 return false;
1447 }
1448 }
1449
1450 #define BTREE_MAX_DEPTH 4U
1451
1452 /* Btree nodes */
1453
1454 /*
1455 * Btree nodes
1456 *
1457 * On disk a btree node is a list/log of these; within each set the keys are
1458 * sorted
1459 */
1460 struct bset {
1461 __le64 seq;
1462
1463 /*
1464 * Highest journal entry this bset contains keys for.
1465 * If on recovery we don't see that journal entry, this bset is ignored:
1466 * this allows us to preserve the order of all index updates after a
1467 * crash, since the journal records a total order of all index updates
1468 * and anything that didn't make it to the journal doesn't get used.
1469 */
1470 __le64 journal_seq;
1471
1472 __le32 flags;
1473 __le16 version;
1474 __le16 u64s; /* count of d[] in u64s */
1475
1476 struct bkey_packed start[0];
1477 __u64 _data[];
1478 } __packed __aligned(8);
1479
1480 LE32_BITMASK(BSET_CSUM_TYPE, struct bset, flags, 0, 4);
1481
1482 LE32_BITMASK(BSET_BIG_ENDIAN, struct bset, flags, 4, 5);
1483 LE32_BITMASK(BSET_SEPARATE_WHITEOUTS,
1484 struct bset, flags, 5, 6);
1485
1486 /* Sector offset within the btree node: */
1487 LE32_BITMASK(BSET_OFFSET, struct bset, flags, 16, 32);
1488
1489 struct btree_node {
1490 struct bch_csum csum;
1491 __le64 magic;
1492
1493 /* this flags field is encrypted, unlike bset->flags: */
1494 __le64 flags;
1495
1496 /* Closed interval: */
1497 struct bpos min_key;
1498 struct bpos max_key;
1499 struct bch_extent_ptr _ptr; /* not used anymore */
1500 struct bkey_format format;
1501
1502 union {
1503 struct bset keys;
1504 struct {
1505 __u8 pad[22];
1506 __le16 u64s;
1507 __u64 _data[0];
1508
1509 };
1510 };
1511 } __packed __aligned(8);
1512
1513 LE64_BITMASK(BTREE_NODE_ID_LO, struct btree_node, flags, 0, 4);
1514 LE64_BITMASK(BTREE_NODE_LEVEL, struct btree_node, flags, 4, 8);
1515 LE64_BITMASK(BTREE_NODE_NEW_EXTENT_OVERWRITE,
1516 struct btree_node, flags, 8, 9);
1517 LE64_BITMASK(BTREE_NODE_ID_HI, struct btree_node, flags, 9, 25);
1518 /* 25-32 unused */
1519 LE64_BITMASK(BTREE_NODE_SEQ, struct btree_node, flags, 32, 64);
1520
BTREE_NODE_ID(struct btree_node * n)1521 static inline __u64 BTREE_NODE_ID(struct btree_node *n)
1522 {
1523 return BTREE_NODE_ID_LO(n) | (BTREE_NODE_ID_HI(n) << 4);
1524 }
1525
SET_BTREE_NODE_ID(struct btree_node * n,__u64 v)1526 static inline void SET_BTREE_NODE_ID(struct btree_node *n, __u64 v)
1527 {
1528 SET_BTREE_NODE_ID_LO(n, v);
1529 SET_BTREE_NODE_ID_HI(n, v >> 4);
1530 }
1531
1532 struct btree_node_entry {
1533 struct bch_csum csum;
1534
1535 union {
1536 struct bset keys;
1537 struct {
1538 __u8 pad[22];
1539 __le16 u64s;
1540 __u64 _data[0];
1541 };
1542 };
1543 } __packed __aligned(8);
1544
1545 #endif /* _BCACHEFS_FORMAT_H */
1546