1 /* SPDX-License-Identifier: GPL-2.0 WITH Linux-syscall-note */
2 #ifndef _LINUX_BCACHE_H
3 #define _LINUX_BCACHE_H
4
5 /*
6 * Bcache on disk data structures
7 */
8
9 #include <linux/types.h>
10
11 #define BITMASK(name, type, field, offset, size) \
12 static inline uint64_t name(const type *k) \
13 { return (k->field >> offset) & ~(~0ULL << size); } \
14 \
15 static inline void SET_##name(type *k, uint64_t v) \
16 { \
17 k->field &= ~(~(~0ULL << size) << offset); \
18 k->field |= (v & ~(~0ULL << size)) << offset; \
19 }
20
21 /* Btree keys - all units are in sectors */
22
23 struct bkey {
24 uint64_t high;
25 uint64_t low;
26 uint64_t ptr[];
27 };
28
29 #define KEY_FIELD(name, field, offset, size) \
30 BITMASK(name, struct bkey, field, offset, size)
31
32 #define PTR_FIELD(name, offset, size) \
33 static inline uint64_t name(const struct bkey *k, unsigned int i) \
34 { return (k->ptr[i] >> offset) & ~(~0ULL << size); } \
35 \
36 static inline void SET_##name(struct bkey *k, unsigned int i, uint64_t v) \
37 { \
38 k->ptr[i] &= ~(~(~0ULL << size) << offset); \
39 k->ptr[i] |= (v & ~(~0ULL << size)) << offset; \
40 }
41
42 #define KEY_SIZE_BITS 16
43 #define KEY_MAX_U64S 8
44
45 KEY_FIELD(KEY_PTRS, high, 60, 3)
46 KEY_FIELD(HEADER_SIZE, high, 58, 2)
47 KEY_FIELD(KEY_CSUM, high, 56, 2)
48 KEY_FIELD(KEY_PINNED, high, 55, 1)
49 KEY_FIELD(KEY_DIRTY, high, 36, 1)
50
51 KEY_FIELD(KEY_SIZE, high, 20, KEY_SIZE_BITS)
52 KEY_FIELD(KEY_INODE, high, 0, 20)
53
54 /* Next time I change the on disk format, KEY_OFFSET() won't be 64 bits */
55
KEY_OFFSET(const struct bkey * k)56 static inline uint64_t KEY_OFFSET(const struct bkey *k)
57 {
58 return k->low;
59 }
60
SET_KEY_OFFSET(struct bkey * k,uint64_t v)61 static inline void SET_KEY_OFFSET(struct bkey *k, uint64_t v)
62 {
63 k->low = v;
64 }
65
66 /*
67 * The high bit being set is a relic from when we used it to do binary
68 * searches - it told you where a key started. It's not used anymore,
69 * and can probably be safely dropped.
70 */
71 #define KEY(inode, offset, size) \
72 ((struct bkey) { \
73 .high = (1ULL << 63) | ((uint64_t) (size) << 20) | (inode), \
74 .low = (offset) \
75 })
76
77 #define ZERO_KEY KEY(0, 0, 0)
78
79 #define MAX_KEY_INODE (~(~0 << 20))
80 #define MAX_KEY_OFFSET (~0ULL >> 1)
81 #define MAX_KEY KEY(MAX_KEY_INODE, MAX_KEY_OFFSET, 0)
82
83 #define KEY_START(k) (KEY_OFFSET(k) - KEY_SIZE(k))
84 #define START_KEY(k) KEY(KEY_INODE(k), KEY_START(k), 0)
85
86 #define PTR_DEV_BITS 12
87
88 PTR_FIELD(PTR_DEV, 51, PTR_DEV_BITS)
89 PTR_FIELD(PTR_OFFSET, 8, 43)
90 PTR_FIELD(PTR_GEN, 0, 8)
91
92 #define PTR_CHECK_DEV ((1 << PTR_DEV_BITS) - 1)
93
94 #define MAKE_PTR(gen, offset, dev) \
95 ((((uint64_t) dev) << 51) | ((uint64_t) offset) << 8 | gen)
96
97 /* Bkey utility code */
98
bkey_u64s(const struct bkey * k)99 static inline unsigned long bkey_u64s(const struct bkey *k)
100 {
101 return (sizeof(struct bkey) / sizeof(uint64_t)) + KEY_PTRS(k);
102 }
103
bkey_bytes(const struct bkey * k)104 static inline unsigned long bkey_bytes(const struct bkey *k)
105 {
106 return bkey_u64s(k) * sizeof(uint64_t);
107 }
108
109 #define bkey_copy(_dest, _src) memcpy(_dest, _src, bkey_bytes(_src))
110
bkey_copy_key(struct bkey * dest,const struct bkey * src)111 static inline void bkey_copy_key(struct bkey *dest, const struct bkey *src)
112 {
113 SET_KEY_INODE(dest, KEY_INODE(src));
114 SET_KEY_OFFSET(dest, KEY_OFFSET(src));
115 }
116
bkey_next(const struct bkey * k)117 static inline struct bkey *bkey_next(const struct bkey *k)
118 {
119 uint64_t *d = (void *) k;
120
121 return (struct bkey *) (d + bkey_u64s(k));
122 }
123
bkey_idx(const struct bkey * k,unsigned int nr_keys)124 static inline struct bkey *bkey_idx(const struct bkey *k, unsigned int nr_keys)
125 {
126 uint64_t *d = (void *) k;
127
128 return (struct bkey *) (d + nr_keys);
129 }
130 /* Enough for a key with 6 pointers */
131 #define BKEY_PAD 8
132
133 #define BKEY_PADDED(key) \
134 union { struct bkey key; uint64_t key ## _pad[BKEY_PAD]; }
135
136 /* Superblock */
137
138 /* Version 0: Cache device
139 * Version 1: Backing device
140 * Version 2: Seed pointer into btree node checksum
141 * Version 3: Cache device with new UUID format
142 * Version 4: Backing device with data offset
143 */
144 #define BCACHE_SB_VERSION_CDEV 0
145 #define BCACHE_SB_VERSION_BDEV 1
146 #define BCACHE_SB_VERSION_CDEV_WITH_UUID 3
147 #define BCACHE_SB_VERSION_BDEV_WITH_OFFSET 4
148 #define BCACHE_SB_VERSION_CDEV_WITH_FEATURES 5
149 #define BCACHE_SB_VERSION_BDEV_WITH_FEATURES 6
150 #define BCACHE_SB_MAX_VERSION 6
151
152 #define SB_SECTOR 8
153 #define SB_OFFSET (SB_SECTOR << SECTOR_SHIFT)
154 #define SB_SIZE 4096
155 #define SB_LABEL_SIZE 32
156 #define SB_JOURNAL_BUCKETS 256U
157 /* SB_JOURNAL_BUCKETS must be divisible by BITS_PER_LONG */
158 #define MAX_CACHES_PER_SET 8
159
160 #define BDEV_DATA_START_DEFAULT 16 /* sectors */
161
162 struct cache_sb_disk {
163 uint64_t csum;
164 uint64_t offset; /* sector where this sb was written */
165 uint64_t version;
166
167 uint8_t magic[16];
168
169 uint8_t uuid[16];
170 union {
171 uint8_t set_uuid[16];
172 uint64_t set_magic;
173 };
174 uint8_t label[SB_LABEL_SIZE];
175
176 uint64_t flags;
177 uint64_t seq;
178
179 uint64_t feature_compat;
180 uint64_t feature_incompat;
181 uint64_t feature_ro_compat;
182
183 uint64_t pad[5];
184
185 union {
186 struct {
187 /* Cache devices */
188 uint64_t nbuckets; /* device size */
189
190 uint16_t block_size; /* sectors */
191 uint16_t bucket_size; /* sectors */
192
193 uint16_t nr_in_set;
194 uint16_t nr_this_dev;
195 };
196 struct {
197 /* Backing devices */
198 uint64_t data_offset;
199
200 /*
201 * block_size from the cache device section is still used by
202 * backing devices, so don't add anything here until we fix
203 * things to not need it for backing devices anymore
204 */
205 };
206 };
207
208 uint32_t last_mount; /* time overflow in y2106 */
209
210 uint16_t first_bucket;
211 union {
212 uint16_t njournal_buckets;
213 uint16_t keys;
214 };
215 uint64_t d[SB_JOURNAL_BUCKETS]; /* journal buckets */
216 uint16_t obso_bucket_size_hi; /* obsoleted */
217 };
218
219 /*
220 * This is for in-memory bcache super block.
221 * NOTE: cache_sb is NOT exactly mapping to cache_sb_disk, the member
222 * size, ordering and even whole struct size may be different
223 * from cache_sb_disk.
224 */
225 struct cache_sb {
226 uint64_t offset; /* sector where this sb was written */
227 uint64_t version;
228
229 uint8_t magic[16];
230
231 uint8_t uuid[16];
232 union {
233 uint8_t set_uuid[16];
234 uint64_t set_magic;
235 };
236 uint8_t label[SB_LABEL_SIZE];
237
238 uint64_t flags;
239 uint64_t seq;
240
241 uint64_t feature_compat;
242 uint64_t feature_incompat;
243 uint64_t feature_ro_compat;
244
245 union {
246 struct {
247 /* Cache devices */
248 uint64_t nbuckets; /* device size */
249
250 uint16_t block_size; /* sectors */
251 uint16_t nr_in_set;
252 uint16_t nr_this_dev;
253 uint32_t bucket_size; /* sectors */
254 };
255 struct {
256 /* Backing devices */
257 uint64_t data_offset;
258
259 /*
260 * block_size from the cache device section is still used by
261 * backing devices, so don't add anything here until we fix
262 * things to not need it for backing devices anymore
263 */
264 };
265 };
266
267 uint32_t last_mount; /* time overflow in y2106 */
268
269 uint16_t first_bucket;
270 union {
271 uint16_t njournal_buckets;
272 uint16_t keys;
273 };
274 uint64_t d[SB_JOURNAL_BUCKETS]; /* journal buckets */
275 };
276
SB_IS_BDEV(const struct cache_sb * sb)277 static inline _Bool SB_IS_BDEV(const struct cache_sb *sb)
278 {
279 return sb->version == BCACHE_SB_VERSION_BDEV
280 || sb->version == BCACHE_SB_VERSION_BDEV_WITH_OFFSET
281 || sb->version == BCACHE_SB_VERSION_BDEV_WITH_FEATURES;
282 }
283
284 BITMASK(CACHE_SYNC, struct cache_sb, flags, 0, 1);
285 BITMASK(CACHE_DISCARD, struct cache_sb, flags, 1, 1);
286 BITMASK(CACHE_REPLACEMENT, struct cache_sb, flags, 2, 3);
287 #define CACHE_REPLACEMENT_LRU 0U
288 #define CACHE_REPLACEMENT_FIFO 1U
289 #define CACHE_REPLACEMENT_RANDOM 2U
290
291 BITMASK(BDEV_CACHE_MODE, struct cache_sb, flags, 0, 4);
292 #define CACHE_MODE_WRITETHROUGH 0U
293 #define CACHE_MODE_WRITEBACK 1U
294 #define CACHE_MODE_WRITEAROUND 2U
295 #define CACHE_MODE_NONE 3U
296 BITMASK(BDEV_STATE, struct cache_sb, flags, 61, 2);
297 #define BDEV_STATE_NONE 0U
298 #define BDEV_STATE_CLEAN 1U
299 #define BDEV_STATE_DIRTY 2U
300 #define BDEV_STATE_STALE 3U
301
302 /*
303 * Magic numbers
304 *
305 * The various other data structures have their own magic numbers, which are
306 * xored with the first part of the cache set's UUID
307 */
308
309 #define JSET_MAGIC 0x245235c1a3625032ULL
310 #define PSET_MAGIC 0x6750e15f87337f91ULL
311 #define BSET_MAGIC 0x90135c78b99e07f5ULL
312
jset_magic(struct cache_sb * sb)313 static inline uint64_t jset_magic(struct cache_sb *sb)
314 {
315 return sb->set_magic ^ JSET_MAGIC;
316 }
317
pset_magic(struct cache_sb * sb)318 static inline uint64_t pset_magic(struct cache_sb *sb)
319 {
320 return sb->set_magic ^ PSET_MAGIC;
321 }
322
bset_magic(struct cache_sb * sb)323 static inline uint64_t bset_magic(struct cache_sb *sb)
324 {
325 return sb->set_magic ^ BSET_MAGIC;
326 }
327
328 /*
329 * Journal
330 *
331 * On disk format for a journal entry:
332 * seq is monotonically increasing; every journal entry has its own unique
333 * sequence number.
334 *
335 * last_seq is the oldest journal entry that still has keys the btree hasn't
336 * flushed to disk yet.
337 *
338 * version is for on disk format changes.
339 */
340
341 #define BCACHE_JSET_VERSION_UUIDv1 1
342 #define BCACHE_JSET_VERSION_UUID 1 /* Always latest UUID format */
343 #define BCACHE_JSET_VERSION 1
344
345 struct jset {
346 uint64_t csum;
347 uint64_t magic;
348 uint64_t seq;
349 uint32_t version;
350 uint32_t keys;
351
352 uint64_t last_seq;
353
354 BKEY_PADDED(uuid_bucket);
355 BKEY_PADDED(btree_root);
356 uint16_t btree_level;
357 uint16_t pad[3];
358
359 uint64_t prio_bucket[MAX_CACHES_PER_SET];
360
361 union {
362 struct bkey start[0];
363 uint64_t d[0];
364 };
365 };
366
367 /* Bucket prios/gens */
368
369 struct prio_set {
370 uint64_t csum;
371 uint64_t magic;
372 uint64_t seq;
373 uint32_t version;
374 uint32_t pad;
375
376 uint64_t next_bucket;
377
378 struct bucket_disk {
379 uint16_t prio;
380 uint8_t gen;
381 } __attribute((packed)) data[];
382 };
383
384 /* UUIDS - per backing device/flash only volume metadata */
385
386 struct uuid_entry {
387 union {
388 struct {
389 uint8_t uuid[16];
390 uint8_t label[32];
391 uint32_t first_reg; /* time overflow in y2106 */
392 uint32_t last_reg;
393 uint32_t invalidated;
394
395 uint32_t flags;
396 /* Size of flash only volumes */
397 uint64_t sectors;
398 };
399
400 uint8_t pad[128];
401 };
402 };
403
404 BITMASK(UUID_FLASH_ONLY, struct uuid_entry, flags, 0, 1);
405
406 /* Btree nodes */
407
408 /* Version 1: Seed pointer into btree node checksum
409 */
410 #define BCACHE_BSET_CSUM 1
411 #define BCACHE_BSET_VERSION 1
412
413 /*
414 * Btree nodes
415 *
416 * On disk a btree node is a list/log of these; within each set the keys are
417 * sorted
418 */
419 struct bset {
420 uint64_t csum;
421 uint64_t magic;
422 uint64_t seq;
423 uint32_t version;
424 uint32_t keys;
425
426 union {
427 struct bkey start[0];
428 uint64_t d[0];
429 };
430 };
431
432 /* OBSOLETE */
433
434 /* UUIDS - per backing device/flash only volume metadata */
435
436 struct uuid_entry_v0 {
437 uint8_t uuid[16];
438 uint8_t label[32];
439 uint32_t first_reg;
440 uint32_t last_reg;
441 uint32_t invalidated;
442 uint32_t pad;
443 };
444
445 #endif /* _LINUX_BCACHE_H */
446