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
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
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 x(accounting, 34)
422
423 enum bch_bkey_type {
424 #define x(name, nr) KEY_TYPE_##name = nr,
425 BCH_BKEY_TYPES()
426 #undef x
427 KEY_TYPE_MAX,
428 };
429
430 struct bch_deleted {
431 struct bch_val v;
432 };
433
434 struct bch_whiteout {
435 struct bch_val v;
436 };
437
438 struct bch_error {
439 struct bch_val v;
440 };
441
442 struct bch_cookie {
443 struct bch_val v;
444 __le64 cookie;
445 };
446
447 struct bch_hash_whiteout {
448 struct bch_val v;
449 };
450
451 struct bch_set {
452 struct bch_val v;
453 };
454
455 /* 128 bits, sufficient for cryptographic MACs: */
456 struct bch_csum {
457 __le64 lo;
458 __le64 hi;
459 } __packed __aligned(8);
460
461 struct bch_backpointer {
462 struct bch_val v;
463 __u8 btree_id;
464 __u8 level;
465 __u8 data_type;
466 __u64 bucket_offset:40;
467 __u32 bucket_len;
468 struct bpos pos;
469 } __packed __aligned(8);
470
471 /* Optional/variable size superblock sections: */
472
473 struct bch_sb_field {
474 __u64 _data[0];
475 __le32 u64s;
476 __le32 type;
477 };
478
479 #define BCH_SB_FIELDS() \
480 x(journal, 0) \
481 x(members_v1, 1) \
482 x(crypt, 2) \
483 x(replicas_v0, 3) \
484 x(quota, 4) \
485 x(disk_groups, 5) \
486 x(clean, 6) \
487 x(replicas, 7) \
488 x(journal_seq_blacklist, 8) \
489 x(journal_v2, 9) \
490 x(counters, 10) \
491 x(members_v2, 11) \
492 x(errors, 12) \
493 x(ext, 13) \
494 x(downgrade, 14)
495
496 #include "alloc_background_format.h"
497 #include "dirent_format.h"
498 #include "disk_accounting_format.h"
499 #include "disk_groups_format.h"
500 #include "extents_format.h"
501 #include "ec_format.h"
502 #include "dirent_format.h"
503 #include "disk_groups_format.h"
504 #include "inode_format.h"
505 #include "journal_seq_blacklist_format.h"
506 #include "logged_ops_format.h"
507 #include "lru_format.h"
508 #include "quota_format.h"
509 #include "reflink_format.h"
510 #include "replicas_format.h"
511 #include "snapshot_format.h"
512 #include "subvolume_format.h"
513 #include "sb-counters_format.h"
514 #include "sb-downgrade_format.h"
515 #include "sb-errors_format.h"
516 #include "sb-members_format.h"
517 #include "xattr_format.h"
518
519 enum bch_sb_field_type {
520 #define x(f, nr) BCH_SB_FIELD_##f = nr,
521 BCH_SB_FIELDS()
522 #undef x
523 BCH_SB_FIELD_NR
524 };
525
526 /*
527 * Most superblock fields are replicated in all device's superblocks - a few are
528 * not:
529 */
530 #define BCH_SINGLE_DEVICE_SB_FIELDS \
531 ((1U << BCH_SB_FIELD_journal)| \
532 (1U << BCH_SB_FIELD_journal_v2))
533
534 /* BCH_SB_FIELD_journal: */
535
536 struct bch_sb_field_journal {
537 struct bch_sb_field field;
538 __le64 buckets[];
539 };
540
541 struct bch_sb_field_journal_v2 {
542 struct bch_sb_field field;
543
544 struct bch_sb_field_journal_v2_entry {
545 __le64 start;
546 __le64 nr;
547 } d[];
548 };
549
550 /* BCH_SB_FIELD_crypt: */
551
552 struct nonce {
553 __le32 d[4];
554 };
555
556 struct bch_key {
557 __le64 key[4];
558 };
559
560 #define BCH_KEY_MAGIC \
561 (((__u64) 'b' << 0)|((__u64) 'c' << 8)| \
562 ((__u64) 'h' << 16)|((__u64) '*' << 24)| \
563 ((__u64) '*' << 32)|((__u64) 'k' << 40)| \
564 ((__u64) 'e' << 48)|((__u64) 'y' << 56))
565
566 struct bch_encrypted_key {
567 __le64 magic;
568 struct bch_key key;
569 };
570
571 /*
572 * If this field is present in the superblock, it stores an encryption key which
573 * is used encrypt all other data/metadata. The key will normally be encrypted
574 * with the key userspace provides, but if encryption has been turned off we'll
575 * just store the master key unencrypted in the superblock so we can access the
576 * previously encrypted data.
577 */
578 struct bch_sb_field_crypt {
579 struct bch_sb_field field;
580
581 __le64 flags;
582 __le64 kdf_flags;
583 struct bch_encrypted_key key;
584 };
585
586 LE64_BITMASK(BCH_CRYPT_KDF_TYPE, struct bch_sb_field_crypt, flags, 0, 4);
587
588 enum bch_kdf_types {
589 BCH_KDF_SCRYPT = 0,
590 BCH_KDF_NR = 1,
591 };
592
593 /* stored as base 2 log of scrypt params: */
594 LE64_BITMASK(BCH_KDF_SCRYPT_N, struct bch_sb_field_crypt, kdf_flags, 0, 16);
595 LE64_BITMASK(BCH_KDF_SCRYPT_R, struct bch_sb_field_crypt, kdf_flags, 16, 32);
596 LE64_BITMASK(BCH_KDF_SCRYPT_P, struct bch_sb_field_crypt, kdf_flags, 32, 48);
597
598 /*
599 * On clean shutdown, store btree roots and current journal sequence number in
600 * the superblock:
601 */
602 struct jset_entry {
603 __le16 u64s;
604 __u8 btree_id;
605 __u8 level;
606 __u8 type; /* designates what this jset holds */
607 __u8 pad[3];
608
609 struct bkey_i start[0];
610 __u64 _data[];
611 };
612
613 struct bch_sb_field_clean {
614 struct bch_sb_field field;
615
616 __le32 flags;
617 __le16 _read_clock; /* no longer used */
618 __le16 _write_clock;
619 __le64 journal_seq;
620
621 struct jset_entry start[0];
622 __u64 _data[];
623 };
624
625 struct bch_sb_field_ext {
626 struct bch_sb_field field;
627 __le64 recovery_passes_required[2];
628 __le64 errors_silent[8];
629 __le64 btrees_lost_data;
630 };
631
632 /* Superblock: */
633
634 /*
635 * New versioning scheme:
636 * One common version number for all on disk data structures - superblock, btree
637 * nodes, journal entries
638 */
639 #define BCH_VERSION_MAJOR(_v) ((__u16) ((_v) >> 10))
640 #define BCH_VERSION_MINOR(_v) ((__u16) ((_v) & ~(~0U << 10)))
641 #define BCH_VERSION(_major, _minor) (((_major) << 10)|(_minor) << 0)
642
643 /*
644 * field 1: version name
645 * field 2: BCH_VERSION(major, minor)
646 * field 3: recovery passess required on upgrade
647 */
648 #define BCH_METADATA_VERSIONS() \
649 x(bkey_renumber, BCH_VERSION(0, 10)) \
650 x(inode_btree_change, BCH_VERSION(0, 11)) \
651 x(snapshot, BCH_VERSION(0, 12)) \
652 x(inode_backpointers, BCH_VERSION(0, 13)) \
653 x(btree_ptr_sectors_written, BCH_VERSION(0, 14)) \
654 x(snapshot_2, BCH_VERSION(0, 15)) \
655 x(reflink_p_fix, BCH_VERSION(0, 16)) \
656 x(subvol_dirent, BCH_VERSION(0, 17)) \
657 x(inode_v2, BCH_VERSION(0, 18)) \
658 x(freespace, BCH_VERSION(0, 19)) \
659 x(alloc_v4, BCH_VERSION(0, 20)) \
660 x(new_data_types, BCH_VERSION(0, 21)) \
661 x(backpointers, BCH_VERSION(0, 22)) \
662 x(inode_v3, BCH_VERSION(0, 23)) \
663 x(unwritten_extents, BCH_VERSION(0, 24)) \
664 x(bucket_gens, BCH_VERSION(0, 25)) \
665 x(lru_v2, BCH_VERSION(0, 26)) \
666 x(fragmentation_lru, BCH_VERSION(0, 27)) \
667 x(no_bps_in_alloc_keys, BCH_VERSION(0, 28)) \
668 x(snapshot_trees, BCH_VERSION(0, 29)) \
669 x(major_minor, BCH_VERSION(1, 0)) \
670 x(snapshot_skiplists, BCH_VERSION(1, 1)) \
671 x(deleted_inodes, BCH_VERSION(1, 2)) \
672 x(rebalance_work, BCH_VERSION(1, 3)) \
673 x(member_seq, BCH_VERSION(1, 4)) \
674 x(subvolume_fs_parent, BCH_VERSION(1, 5)) \
675 x(btree_subvolume_children, BCH_VERSION(1, 6)) \
676 x(mi_btree_bitmap, BCH_VERSION(1, 7)) \
677 x(bucket_stripe_sectors, BCH_VERSION(1, 8)) \
678 x(disk_accounting_v2, BCH_VERSION(1, 9)) \
679 x(disk_accounting_v3, BCH_VERSION(1, 10)) \
680 x(disk_accounting_inum, BCH_VERSION(1, 11)) \
681 x(rebalance_work_acct_fix, BCH_VERSION(1, 12))
682
683 enum bcachefs_metadata_version {
684 bcachefs_metadata_version_min = 9,
685 #define x(t, n) bcachefs_metadata_version_##t = n,
686 BCH_METADATA_VERSIONS()
687 #undef x
688 bcachefs_metadata_version_max
689 };
690
691 static const __maybe_unused
692 unsigned bcachefs_metadata_required_upgrade_below = bcachefs_metadata_version_rebalance_work;
693
694 #define bcachefs_metadata_version_current (bcachefs_metadata_version_max - 1)
695
696 #define BCH_SB_SECTOR 8
697
698 #define BCH_SB_LAYOUT_SIZE_BITS_MAX 16 /* 32 MB */
699
700 struct bch_sb_layout {
701 __uuid_t magic; /* bcachefs superblock UUID */
702 __u8 layout_type;
703 __u8 sb_max_size_bits; /* base 2 of 512 byte sectors */
704 __u8 nr_superblocks;
705 __u8 pad[5];
706 __le64 sb_offset[61];
707 } __packed __aligned(8);
708
709 #define BCH_SB_LAYOUT_SECTOR 7
710
711 /*
712 * @offset - sector where this sb was written
713 * @version - on disk format version
714 * @version_min - Oldest metadata version this filesystem contains; so we can
715 * safely drop compatibility code and refuse to mount filesystems
716 * we'd need it for
717 * @magic - identifies as a bcachefs superblock (BCHFS_MAGIC)
718 * @seq - incremented each time superblock is written
719 * @uuid - used for generating various magic numbers and identifying
720 * member devices, never changes
721 * @user_uuid - user visible UUID, may be changed
722 * @label - filesystem label
723 * @seq - identifies most recent superblock, incremented each time
724 * superblock is written
725 * @features - enabled incompatible features
726 */
727 struct bch_sb {
728 struct bch_csum csum;
729 __le16 version;
730 __le16 version_min;
731 __le16 pad[2];
732 __uuid_t magic;
733 __uuid_t uuid;
734 __uuid_t user_uuid;
735 __u8 label[BCH_SB_LABEL_SIZE];
736 __le64 offset;
737 __le64 seq;
738
739 __le16 block_size;
740 __u8 dev_idx;
741 __u8 nr_devices;
742 __le32 u64s;
743
744 __le64 time_base_lo;
745 __le32 time_base_hi;
746 __le32 time_precision;
747
748 __le64 flags[7];
749 __le64 write_time;
750 __le64 features[2];
751 __le64 compat[2];
752
753 struct bch_sb_layout layout;
754
755 struct bch_sb_field start[0];
756 __le64 _data[];
757 } __packed __aligned(8);
758
759 /*
760 * Flags:
761 * BCH_SB_INITALIZED - set on first mount
762 * BCH_SB_CLEAN - did we shut down cleanly? Just a hint, doesn't affect
763 * behaviour of mount/recovery path:
764 * BCH_SB_INODE_32BIT - limit inode numbers to 32 bits
765 * BCH_SB_128_BIT_MACS - 128 bit macs instead of 80
766 * BCH_SB_ENCRYPTION_TYPE - if nonzero encryption is enabled; overrides
767 * DATA/META_CSUM_TYPE. Also indicates encryption
768 * algorithm in use, if/when we get more than one
769 */
770
771 LE16_BITMASK(BCH_SB_BLOCK_SIZE, struct bch_sb, block_size, 0, 16);
772
773 LE64_BITMASK(BCH_SB_INITIALIZED, struct bch_sb, flags[0], 0, 1);
774 LE64_BITMASK(BCH_SB_CLEAN, struct bch_sb, flags[0], 1, 2);
775 LE64_BITMASK(BCH_SB_CSUM_TYPE, struct bch_sb, flags[0], 2, 8);
776 LE64_BITMASK(BCH_SB_ERROR_ACTION, struct bch_sb, flags[0], 8, 12);
777
778 LE64_BITMASK(BCH_SB_BTREE_NODE_SIZE, struct bch_sb, flags[0], 12, 28);
779
780 LE64_BITMASK(BCH_SB_GC_RESERVE, struct bch_sb, flags[0], 28, 33);
781 LE64_BITMASK(BCH_SB_ROOT_RESERVE, struct bch_sb, flags[0], 33, 40);
782
783 LE64_BITMASK(BCH_SB_META_CSUM_TYPE, struct bch_sb, flags[0], 40, 44);
784 LE64_BITMASK(BCH_SB_DATA_CSUM_TYPE, struct bch_sb, flags[0], 44, 48);
785
786 LE64_BITMASK(BCH_SB_META_REPLICAS_WANT, struct bch_sb, flags[0], 48, 52);
787 LE64_BITMASK(BCH_SB_DATA_REPLICAS_WANT, struct bch_sb, flags[0], 52, 56);
788
789 LE64_BITMASK(BCH_SB_POSIX_ACL, struct bch_sb, flags[0], 56, 57);
790 LE64_BITMASK(BCH_SB_USRQUOTA, struct bch_sb, flags[0], 57, 58);
791 LE64_BITMASK(BCH_SB_GRPQUOTA, struct bch_sb, flags[0], 58, 59);
792 LE64_BITMASK(BCH_SB_PRJQUOTA, struct bch_sb, flags[0], 59, 60);
793
794 LE64_BITMASK(BCH_SB_HAS_ERRORS, struct bch_sb, flags[0], 60, 61);
795 LE64_BITMASK(BCH_SB_HAS_TOPOLOGY_ERRORS,struct bch_sb, flags[0], 61, 62);
796
797 LE64_BITMASK(BCH_SB_BIG_ENDIAN, struct bch_sb, flags[0], 62, 63);
798
799 LE64_BITMASK(BCH_SB_STR_HASH_TYPE, struct bch_sb, flags[1], 0, 4);
800 LE64_BITMASK(BCH_SB_COMPRESSION_TYPE_LO,struct bch_sb, flags[1], 4, 8);
801 LE64_BITMASK(BCH_SB_INODE_32BIT, struct bch_sb, flags[1], 8, 9);
802
803 LE64_BITMASK(BCH_SB_128_BIT_MACS, struct bch_sb, flags[1], 9, 10);
804 LE64_BITMASK(BCH_SB_ENCRYPTION_TYPE, struct bch_sb, flags[1], 10, 14);
805
806 /*
807 * Max size of an extent that may require bouncing to read or write
808 * (checksummed, compressed): 64k
809 */
810 LE64_BITMASK(BCH_SB_ENCODED_EXTENT_MAX_BITS,
811 struct bch_sb, flags[1], 14, 20);
812
813 LE64_BITMASK(BCH_SB_META_REPLICAS_REQ, struct bch_sb, flags[1], 20, 24);
814 LE64_BITMASK(BCH_SB_DATA_REPLICAS_REQ, struct bch_sb, flags[1], 24, 28);
815
816 LE64_BITMASK(BCH_SB_PROMOTE_TARGET, struct bch_sb, flags[1], 28, 40);
817 LE64_BITMASK(BCH_SB_FOREGROUND_TARGET, struct bch_sb, flags[1], 40, 52);
818 LE64_BITMASK(BCH_SB_BACKGROUND_TARGET, struct bch_sb, flags[1], 52, 64);
819
820 LE64_BITMASK(BCH_SB_BACKGROUND_COMPRESSION_TYPE_LO,
821 struct bch_sb, flags[2], 0, 4);
822 LE64_BITMASK(BCH_SB_GC_RESERVE_BYTES, struct bch_sb, flags[2], 4, 64);
823
824 LE64_BITMASK(BCH_SB_ERASURE_CODE, struct bch_sb, flags[3], 0, 16);
825 LE64_BITMASK(BCH_SB_METADATA_TARGET, struct bch_sb, flags[3], 16, 28);
826 LE64_BITMASK(BCH_SB_SHARD_INUMS, struct bch_sb, flags[3], 28, 29);
827 LE64_BITMASK(BCH_SB_INODES_USE_KEY_CACHE,struct bch_sb, flags[3], 29, 30);
828 LE64_BITMASK(BCH_SB_JOURNAL_FLUSH_DELAY,struct bch_sb, flags[3], 30, 62);
829 LE64_BITMASK(BCH_SB_JOURNAL_FLUSH_DISABLED,struct bch_sb, flags[3], 62, 63);
830 LE64_BITMASK(BCH_SB_JOURNAL_RECLAIM_DELAY,struct bch_sb, flags[4], 0, 32);
831 LE64_BITMASK(BCH_SB_JOURNAL_TRANSACTION_NAMES,struct bch_sb, flags[4], 32, 33);
832 LE64_BITMASK(BCH_SB_NOCOW, struct bch_sb, flags[4], 33, 34);
833 LE64_BITMASK(BCH_SB_WRITE_BUFFER_SIZE, struct bch_sb, flags[4], 34, 54);
834 LE64_BITMASK(BCH_SB_VERSION_UPGRADE, struct bch_sb, flags[4], 54, 56);
835
836 LE64_BITMASK(BCH_SB_COMPRESSION_TYPE_HI,struct bch_sb, flags[4], 56, 60);
837 LE64_BITMASK(BCH_SB_BACKGROUND_COMPRESSION_TYPE_HI,
838 struct bch_sb, flags[4], 60, 64);
839
840 LE64_BITMASK(BCH_SB_VERSION_UPGRADE_COMPLETE,
841 struct bch_sb, flags[5], 0, 16);
842 LE64_BITMASK(BCH_SB_ALLOCATOR_STUCK_TIMEOUT,
843 struct bch_sb, flags[5], 16, 32);
844
845 static inline __u64 BCH_SB_COMPRESSION_TYPE(const struct bch_sb *sb)
846 {
847 return BCH_SB_COMPRESSION_TYPE_LO(sb) | (BCH_SB_COMPRESSION_TYPE_HI(sb) << 4);
848 }
849
850 static inline void SET_BCH_SB_COMPRESSION_TYPE(struct bch_sb *sb, __u64 v)
851 {
852 SET_BCH_SB_COMPRESSION_TYPE_LO(sb, v);
853 SET_BCH_SB_COMPRESSION_TYPE_HI(sb, v >> 4);
854 }
855
856 static inline __u64 BCH_SB_BACKGROUND_COMPRESSION_TYPE(const struct bch_sb *sb)
857 {
858 return BCH_SB_BACKGROUND_COMPRESSION_TYPE_LO(sb) |
859 (BCH_SB_BACKGROUND_COMPRESSION_TYPE_HI(sb) << 4);
860 }
861
862 static inline void SET_BCH_SB_BACKGROUND_COMPRESSION_TYPE(struct bch_sb *sb, __u64 v)
863 {
864 SET_BCH_SB_BACKGROUND_COMPRESSION_TYPE_LO(sb, v);
865 SET_BCH_SB_BACKGROUND_COMPRESSION_TYPE_HI(sb, v >> 4);
866 }
867
868 /*
869 * Features:
870 *
871 * journal_seq_blacklist_v3: gates BCH_SB_FIELD_journal_seq_blacklist
872 * reflink: gates KEY_TYPE_reflink
873 * inline_data: gates KEY_TYPE_inline_data
874 * new_siphash: gates BCH_STR_HASH_siphash
875 * new_extent_overwrite: gates BTREE_NODE_NEW_EXTENT_OVERWRITE
876 */
877 #define BCH_SB_FEATURES() \
878 x(lz4, 0) \
879 x(gzip, 1) \
880 x(zstd, 2) \
881 x(atomic_nlink, 3) \
882 x(ec, 4) \
883 x(journal_seq_blacklist_v3, 5) \
884 x(reflink, 6) \
885 x(new_siphash, 7) \
886 x(inline_data, 8) \
887 x(new_extent_overwrite, 9) \
888 x(incompressible, 10) \
889 x(btree_ptr_v2, 11) \
890 x(extents_above_btree_updates, 12) \
891 x(btree_updates_journalled, 13) \
892 x(reflink_inline_data, 14) \
893 x(new_varint, 15) \
894 x(journal_no_flush, 16) \
895 x(alloc_v2, 17) \
896 x(extents_across_btree_nodes, 18)
897
898 #define BCH_SB_FEATURES_ALWAYS \
899 ((1ULL << BCH_FEATURE_new_extent_overwrite)| \
900 (1ULL << BCH_FEATURE_extents_above_btree_updates)|\
901 (1ULL << BCH_FEATURE_btree_updates_journalled)|\
902 (1ULL << BCH_FEATURE_alloc_v2)|\
903 (1ULL << BCH_FEATURE_extents_across_btree_nodes))
904
905 #define BCH_SB_FEATURES_ALL \
906 (BCH_SB_FEATURES_ALWAYS| \
907 (1ULL << BCH_FEATURE_new_siphash)| \
908 (1ULL << BCH_FEATURE_btree_ptr_v2)| \
909 (1ULL << BCH_FEATURE_new_varint)| \
910 (1ULL << BCH_FEATURE_journal_no_flush))
911
912 enum bch_sb_feature {
913 #define x(f, n) BCH_FEATURE_##f,
914 BCH_SB_FEATURES()
915 #undef x
916 BCH_FEATURE_NR,
917 };
918
919 #define BCH_SB_COMPAT() \
920 x(alloc_info, 0) \
921 x(alloc_metadata, 1) \
922 x(extents_above_btree_updates_done, 2) \
923 x(bformat_overflow_done, 3)
924
925 enum bch_sb_compat {
926 #define x(f, n) BCH_COMPAT_##f,
927 BCH_SB_COMPAT()
928 #undef x
929 BCH_COMPAT_NR,
930 };
931
932 /* options: */
933
934 #define BCH_VERSION_UPGRADE_OPTS() \
935 x(compatible, 0) \
936 x(incompatible, 1) \
937 x(none, 2)
938
939 enum bch_version_upgrade_opts {
940 #define x(t, n) BCH_VERSION_UPGRADE_##t = n,
941 BCH_VERSION_UPGRADE_OPTS()
942 #undef x
943 };
944
945 #define BCH_REPLICAS_MAX 4U
946
947 #define BCH_BKEY_PTRS_MAX 16U
948
949 #define BCH_ERROR_ACTIONS() \
950 x(continue, 0) \
951 x(fix_safe, 1) \
952 x(panic, 2) \
953 x(ro, 3)
954
955 enum bch_error_actions {
956 #define x(t, n) BCH_ON_ERROR_##t = n,
957 BCH_ERROR_ACTIONS()
958 #undef x
959 BCH_ON_ERROR_NR
960 };
961
962 #define BCH_STR_HASH_TYPES() \
963 x(crc32c, 0) \
964 x(crc64, 1) \
965 x(siphash_old, 2) \
966 x(siphash, 3)
967
968 enum bch_str_hash_type {
969 #define x(t, n) BCH_STR_HASH_##t = n,
970 BCH_STR_HASH_TYPES()
971 #undef x
972 BCH_STR_HASH_NR
973 };
974
975 #define BCH_STR_HASH_OPTS() \
976 x(crc32c, 0) \
977 x(crc64, 1) \
978 x(siphash, 2)
979
980 enum bch_str_hash_opts {
981 #define x(t, n) BCH_STR_HASH_OPT_##t = n,
982 BCH_STR_HASH_OPTS()
983 #undef x
984 BCH_STR_HASH_OPT_NR
985 };
986
987 #define BCH_CSUM_TYPES() \
988 x(none, 0) \
989 x(crc32c_nonzero, 1) \
990 x(crc64_nonzero, 2) \
991 x(chacha20_poly1305_80, 3) \
992 x(chacha20_poly1305_128, 4) \
993 x(crc32c, 5) \
994 x(crc64, 6) \
995 x(xxhash, 7)
996
997 enum bch_csum_type {
998 #define x(t, n) BCH_CSUM_##t = n,
999 BCH_CSUM_TYPES()
1000 #undef x
1001 BCH_CSUM_NR
1002 };
1003
1004 static const __maybe_unused unsigned bch_crc_bytes[] = {
1005 [BCH_CSUM_none] = 0,
1006 [BCH_CSUM_crc32c_nonzero] = 4,
1007 [BCH_CSUM_crc32c] = 4,
1008 [BCH_CSUM_crc64_nonzero] = 8,
1009 [BCH_CSUM_crc64] = 8,
1010 [BCH_CSUM_xxhash] = 8,
1011 [BCH_CSUM_chacha20_poly1305_80] = 10,
1012 [BCH_CSUM_chacha20_poly1305_128] = 16,
1013 };
1014
1015 static inline _Bool bch2_csum_type_is_encryption(enum bch_csum_type type)
1016 {
1017 switch (type) {
1018 case BCH_CSUM_chacha20_poly1305_80:
1019 case BCH_CSUM_chacha20_poly1305_128:
1020 return true;
1021 default:
1022 return false;
1023 }
1024 }
1025
1026 #define BCH_CSUM_OPTS() \
1027 x(none, 0) \
1028 x(crc32c, 1) \
1029 x(crc64, 2) \
1030 x(xxhash, 3)
1031
1032 enum bch_csum_opts {
1033 #define x(t, n) BCH_CSUM_OPT_##t = n,
1034 BCH_CSUM_OPTS()
1035 #undef x
1036 BCH_CSUM_OPT_NR
1037 };
1038
1039 #define BCH_COMPRESSION_TYPES() \
1040 x(none, 0) \
1041 x(lz4_old, 1) \
1042 x(gzip, 2) \
1043 x(lz4, 3) \
1044 x(zstd, 4) \
1045 x(incompressible, 5)
1046
1047 enum bch_compression_type {
1048 #define x(t, n) BCH_COMPRESSION_TYPE_##t = n,
1049 BCH_COMPRESSION_TYPES()
1050 #undef x
1051 BCH_COMPRESSION_TYPE_NR
1052 };
1053
1054 #define BCH_COMPRESSION_OPTS() \
1055 x(none, 0) \
1056 x(lz4, 1) \
1057 x(gzip, 2) \
1058 x(zstd, 3)
1059
1060 enum bch_compression_opts {
1061 #define x(t, n) BCH_COMPRESSION_OPT_##t = n,
1062 BCH_COMPRESSION_OPTS()
1063 #undef x
1064 BCH_COMPRESSION_OPT_NR
1065 };
1066
1067 /*
1068 * Magic numbers
1069 *
1070 * The various other data structures have their own magic numbers, which are
1071 * xored with the first part of the cache set's UUID
1072 */
1073
1074 #define BCACHE_MAGIC \
1075 UUID_INIT(0xc68573f6, 0x4e1a, 0x45ca, \
1076 0x82, 0x65, 0xf5, 0x7f, 0x48, 0xba, 0x6d, 0x81)
1077 #define BCHFS_MAGIC \
1078 UUID_INIT(0xc68573f6, 0x66ce, 0x90a9, \
1079 0xd9, 0x6a, 0x60, 0xcf, 0x80, 0x3d, 0xf7, 0xef)
1080
1081 #define BCACHEFS_STATFS_MAGIC BCACHEFS_SUPER_MAGIC
1082
1083 #define JSET_MAGIC __cpu_to_le64(0x245235c1a3625032ULL)
1084 #define BSET_MAGIC __cpu_to_le64(0x90135c78b99e07f5ULL)
1085
1086 static inline __le64 __bch2_sb_magic(struct bch_sb *sb)
1087 {
1088 __le64 ret;
1089
1090 memcpy(&ret, &sb->uuid, sizeof(ret));
1091 return ret;
1092 }
1093
1094 static inline __u64 __jset_magic(struct bch_sb *sb)
1095 {
1096 return __le64_to_cpu(__bch2_sb_magic(sb) ^ JSET_MAGIC);
1097 }
1098
1099 static inline __u64 __bset_magic(struct bch_sb *sb)
1100 {
1101 return __le64_to_cpu(__bch2_sb_magic(sb) ^ BSET_MAGIC);
1102 }
1103
1104 /* Journal */
1105
1106 #define JSET_KEYS_U64s (sizeof(struct jset_entry) / sizeof(__u64))
1107
1108 #define BCH_JSET_ENTRY_TYPES() \
1109 x(btree_keys, 0) \
1110 x(btree_root, 1) \
1111 x(prio_ptrs, 2) \
1112 x(blacklist, 3) \
1113 x(blacklist_v2, 4) \
1114 x(usage, 5) \
1115 x(data_usage, 6) \
1116 x(clock, 7) \
1117 x(dev_usage, 8) \
1118 x(log, 9) \
1119 x(overwrite, 10) \
1120 x(write_buffer_keys, 11) \
1121 x(datetime, 12)
1122
1123 enum bch_jset_entry_type {
1124 #define x(f, nr) BCH_JSET_ENTRY_##f = nr,
1125 BCH_JSET_ENTRY_TYPES()
1126 #undef x
1127 BCH_JSET_ENTRY_NR
1128 };
1129
1130 static inline bool jset_entry_is_key(struct jset_entry *e)
1131 {
1132 switch (e->type) {
1133 case BCH_JSET_ENTRY_btree_keys:
1134 case BCH_JSET_ENTRY_btree_root:
1135 case BCH_JSET_ENTRY_write_buffer_keys:
1136 return true;
1137 }
1138
1139 return false;
1140 }
1141
1142 /*
1143 * Journal sequence numbers can be blacklisted: bsets record the max sequence
1144 * number of all the journal entries they contain updates for, so that on
1145 * recovery we can ignore those bsets that contain index updates newer that what
1146 * made it into the journal.
1147 *
1148 * This means that we can't reuse that journal_seq - we have to skip it, and
1149 * then record that we skipped it so that the next time we crash and recover we
1150 * don't think there was a missing journal entry.
1151 */
1152 struct jset_entry_blacklist {
1153 struct jset_entry entry;
1154 __le64 seq;
1155 };
1156
1157 struct jset_entry_blacklist_v2 {
1158 struct jset_entry entry;
1159 __le64 start;
1160 __le64 end;
1161 };
1162
1163 #define BCH_FS_USAGE_TYPES() \
1164 x(reserved, 0) \
1165 x(inodes, 1) \
1166 x(key_version, 2)
1167
1168 enum bch_fs_usage_type {
1169 #define x(f, nr) BCH_FS_USAGE_##f = nr,
1170 BCH_FS_USAGE_TYPES()
1171 #undef x
1172 BCH_FS_USAGE_NR
1173 };
1174
1175 struct jset_entry_usage {
1176 struct jset_entry entry;
1177 __le64 v;
1178 } __packed;
1179
1180 struct jset_entry_data_usage {
1181 struct jset_entry entry;
1182 __le64 v;
1183 struct bch_replicas_entry_v1 r;
1184 } __packed;
1185
1186 struct jset_entry_clock {
1187 struct jset_entry entry;
1188 __u8 rw;
1189 __u8 pad[7];
1190 __le64 time;
1191 } __packed;
1192
1193 struct jset_entry_dev_usage_type {
1194 __le64 buckets;
1195 __le64 sectors;
1196 __le64 fragmented;
1197 } __packed;
1198
1199 struct jset_entry_dev_usage {
1200 struct jset_entry entry;
1201 __le32 dev;
1202 __u32 pad;
1203
1204 __le64 _buckets_ec; /* No longer used */
1205 __le64 _buckets_unavailable; /* No longer used */
1206
1207 struct jset_entry_dev_usage_type d[];
1208 };
1209
1210 static inline unsigned jset_entry_dev_usage_nr_types(struct jset_entry_dev_usage *u)
1211 {
1212 return (vstruct_bytes(&u->entry) - sizeof(struct jset_entry_dev_usage)) /
1213 sizeof(struct jset_entry_dev_usage_type);
1214 }
1215
1216 struct jset_entry_log {
1217 struct jset_entry entry;
1218 u8 d[];
1219 } __packed __aligned(8);
1220
1221 struct jset_entry_datetime {
1222 struct jset_entry entry;
1223 __le64 seconds;
1224 } __packed __aligned(8);
1225
1226 /*
1227 * On disk format for a journal entry:
1228 * seq is monotonically increasing; every journal entry has its own unique
1229 * sequence number.
1230 *
1231 * last_seq is the oldest journal entry that still has keys the btree hasn't
1232 * flushed to disk yet.
1233 *
1234 * version is for on disk format changes.
1235 */
1236 struct jset {
1237 struct bch_csum csum;
1238
1239 __le64 magic;
1240 __le64 seq;
1241 __le32 version;
1242 __le32 flags;
1243
1244 __le32 u64s; /* size of d[] in u64s */
1245
1246 __u8 encrypted_start[0];
1247
1248 __le16 _read_clock; /* no longer used */
1249 __le16 _write_clock;
1250
1251 /* Sequence number of oldest dirty journal entry */
1252 __le64 last_seq;
1253
1254
1255 struct jset_entry start[0];
1256 __u64 _data[];
1257 } __packed __aligned(8);
1258
1259 LE32_BITMASK(JSET_CSUM_TYPE, struct jset, flags, 0, 4);
1260 LE32_BITMASK(JSET_BIG_ENDIAN, struct jset, flags, 4, 5);
1261 LE32_BITMASK(JSET_NO_FLUSH, struct jset, flags, 5, 6);
1262
1263 #define BCH_JOURNAL_BUCKETS_MIN 8
1264
1265 /* Btree: */
1266
1267 enum btree_id_flags {
1268 BTREE_ID_EXTENTS = BIT(0),
1269 BTREE_ID_SNAPSHOTS = BIT(1),
1270 BTREE_ID_SNAPSHOT_FIELD = BIT(2),
1271 BTREE_ID_DATA = BIT(3),
1272 };
1273
1274 #define BCH_BTREE_IDS() \
1275 x(extents, 0, BTREE_ID_EXTENTS|BTREE_ID_SNAPSHOTS|BTREE_ID_DATA,\
1276 BIT_ULL(KEY_TYPE_whiteout)| \
1277 BIT_ULL(KEY_TYPE_error)| \
1278 BIT_ULL(KEY_TYPE_cookie)| \
1279 BIT_ULL(KEY_TYPE_extent)| \
1280 BIT_ULL(KEY_TYPE_reservation)| \
1281 BIT_ULL(KEY_TYPE_reflink_p)| \
1282 BIT_ULL(KEY_TYPE_inline_data)) \
1283 x(inodes, 1, BTREE_ID_SNAPSHOTS, \
1284 BIT_ULL(KEY_TYPE_whiteout)| \
1285 BIT_ULL(KEY_TYPE_inode)| \
1286 BIT_ULL(KEY_TYPE_inode_v2)| \
1287 BIT_ULL(KEY_TYPE_inode_v3)| \
1288 BIT_ULL(KEY_TYPE_inode_generation)) \
1289 x(dirents, 2, BTREE_ID_SNAPSHOTS, \
1290 BIT_ULL(KEY_TYPE_whiteout)| \
1291 BIT_ULL(KEY_TYPE_hash_whiteout)| \
1292 BIT_ULL(KEY_TYPE_dirent)) \
1293 x(xattrs, 3, BTREE_ID_SNAPSHOTS, \
1294 BIT_ULL(KEY_TYPE_whiteout)| \
1295 BIT_ULL(KEY_TYPE_cookie)| \
1296 BIT_ULL(KEY_TYPE_hash_whiteout)| \
1297 BIT_ULL(KEY_TYPE_xattr)) \
1298 x(alloc, 4, 0, \
1299 BIT_ULL(KEY_TYPE_alloc)| \
1300 BIT_ULL(KEY_TYPE_alloc_v2)| \
1301 BIT_ULL(KEY_TYPE_alloc_v3)| \
1302 BIT_ULL(KEY_TYPE_alloc_v4)) \
1303 x(quotas, 5, 0, \
1304 BIT_ULL(KEY_TYPE_quota)) \
1305 x(stripes, 6, 0, \
1306 BIT_ULL(KEY_TYPE_stripe)) \
1307 x(reflink, 7, BTREE_ID_EXTENTS|BTREE_ID_DATA, \
1308 BIT_ULL(KEY_TYPE_reflink_v)| \
1309 BIT_ULL(KEY_TYPE_indirect_inline_data)| \
1310 BIT_ULL(KEY_TYPE_error)) \
1311 x(subvolumes, 8, 0, \
1312 BIT_ULL(KEY_TYPE_subvolume)) \
1313 x(snapshots, 9, 0, \
1314 BIT_ULL(KEY_TYPE_snapshot)) \
1315 x(lru, 10, 0, \
1316 BIT_ULL(KEY_TYPE_set)) \
1317 x(freespace, 11, BTREE_ID_EXTENTS, \
1318 BIT_ULL(KEY_TYPE_set)) \
1319 x(need_discard, 12, 0, \
1320 BIT_ULL(KEY_TYPE_set)) \
1321 x(backpointers, 13, 0, \
1322 BIT_ULL(KEY_TYPE_backpointer)) \
1323 x(bucket_gens, 14, 0, \
1324 BIT_ULL(KEY_TYPE_bucket_gens)) \
1325 x(snapshot_trees, 15, 0, \
1326 BIT_ULL(KEY_TYPE_snapshot_tree)) \
1327 x(deleted_inodes, 16, BTREE_ID_SNAPSHOT_FIELD, \
1328 BIT_ULL(KEY_TYPE_set)) \
1329 x(logged_ops, 17, 0, \
1330 BIT_ULL(KEY_TYPE_logged_op_truncate)| \
1331 BIT_ULL(KEY_TYPE_logged_op_finsert)) \
1332 x(rebalance_work, 18, BTREE_ID_SNAPSHOT_FIELD, \
1333 BIT_ULL(KEY_TYPE_set)|BIT_ULL(KEY_TYPE_cookie)) \
1334 x(subvolume_children, 19, 0, \
1335 BIT_ULL(KEY_TYPE_set)) \
1336 x(accounting, 20, BTREE_ID_SNAPSHOT_FIELD, \
1337 BIT_ULL(KEY_TYPE_accounting)) \
1338
1339 enum btree_id {
1340 #define x(name, nr, ...) BTREE_ID_##name = nr,
1341 BCH_BTREE_IDS()
1342 #undef x
1343 BTREE_ID_NR
1344 };
1345
1346 /*
1347 * Maximum number of btrees that we will _ever_ have under the current scheme,
1348 * where we refer to them with 64 bit bitfields - and we also need a bit for
1349 * the interior btree node type:
1350 */
1351 #define BTREE_ID_NR_MAX 63
1352
1353 static inline bool btree_id_is_alloc(enum btree_id id)
1354 {
1355 switch (id) {
1356 case BTREE_ID_alloc:
1357 case BTREE_ID_backpointers:
1358 case BTREE_ID_need_discard:
1359 case BTREE_ID_freespace:
1360 case BTREE_ID_bucket_gens:
1361 return true;
1362 default:
1363 return false;
1364 }
1365 }
1366
1367 #define BTREE_MAX_DEPTH 4U
1368
1369 /* Btree nodes */
1370
1371 /*
1372 * Btree nodes
1373 *
1374 * On disk a btree node is a list/log of these; within each set the keys are
1375 * sorted
1376 */
1377 struct bset {
1378 __le64 seq;
1379
1380 /*
1381 * Highest journal entry this bset contains keys for.
1382 * If on recovery we don't see that journal entry, this bset is ignored:
1383 * this allows us to preserve the order of all index updates after a
1384 * crash, since the journal records a total order of all index updates
1385 * and anything that didn't make it to the journal doesn't get used.
1386 */
1387 __le64 journal_seq;
1388
1389 __le32 flags;
1390 __le16 version;
1391 __le16 u64s; /* count of d[] in u64s */
1392
1393 struct bkey_packed start[0];
1394 __u64 _data[];
1395 } __packed __aligned(8);
1396
1397 LE32_BITMASK(BSET_CSUM_TYPE, struct bset, flags, 0, 4);
1398
1399 LE32_BITMASK(BSET_BIG_ENDIAN, struct bset, flags, 4, 5);
1400 LE32_BITMASK(BSET_SEPARATE_WHITEOUTS,
1401 struct bset, flags, 5, 6);
1402
1403 /* Sector offset within the btree node: */
1404 LE32_BITMASK(BSET_OFFSET, struct bset, flags, 16, 32);
1405
1406 struct btree_node {
1407 struct bch_csum csum;
1408 __le64 magic;
1409
1410 /* this flags field is encrypted, unlike bset->flags: */
1411 __le64 flags;
1412
1413 /* Closed interval: */
1414 struct bpos min_key;
1415 struct bpos max_key;
1416 struct bch_extent_ptr _ptr; /* not used anymore */
1417 struct bkey_format format;
1418
1419 union {
1420 struct bset keys;
1421 struct {
1422 __u8 pad[22];
1423 __le16 u64s;
1424 __u64 _data[0];
1425
1426 };
1427 };
1428 } __packed __aligned(8);
1429
1430 LE64_BITMASK(BTREE_NODE_ID_LO, struct btree_node, flags, 0, 4);
1431 LE64_BITMASK(BTREE_NODE_LEVEL, struct btree_node, flags, 4, 8);
1432 LE64_BITMASK(BTREE_NODE_NEW_EXTENT_OVERWRITE,
1433 struct btree_node, flags, 8, 9);
1434 LE64_BITMASK(BTREE_NODE_ID_HI, struct btree_node, flags, 9, 25);
1435 /* 25-32 unused */
1436 LE64_BITMASK(BTREE_NODE_SEQ, struct btree_node, flags, 32, 64);
1437
1438 static inline __u64 BTREE_NODE_ID(struct btree_node *n)
1439 {
1440 return BTREE_NODE_ID_LO(n) | (BTREE_NODE_ID_HI(n) << 4);
1441 }
1442
1443 static inline void SET_BTREE_NODE_ID(struct btree_node *n, __u64 v)
1444 {
1445 SET_BTREE_NODE_ID_LO(n, v);
1446 SET_BTREE_NODE_ID_HI(n, v >> 4);
1447 }
1448
1449 struct btree_node_entry {
1450 struct bch_csum csum;
1451
1452 union {
1453 struct bset keys;
1454 struct {
1455 __u8 pad[22];
1456 __le16 u64s;
1457 __u64 _data[0];
1458 };
1459 };
1460 } __packed __aligned(8);
1461
1462 #endif /* _BCACHEFS_FORMAT_H */
1463
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.