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 __u64 name(const type *k) \
13 { return (k->field >> offset) & ~(~0ULL << size); } \
14 \
15 static inline void SET_##name(type *k, __u64 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 __u64 high;
25 __u64 low;
26 __u64 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 __u64 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, __u64 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 __u64 KEY_OFFSET(const struct bkey *k)
57 {
58 return k->low;
59 }
60
SET_KEY_OFFSET(struct bkey * k,__u64 v)61 static inline void SET_KEY_OFFSET(struct bkey *k, __u64 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) | ((__u64) (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 ((((__u64) dev) << 51) | ((__u64) 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(__u64)) + 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(__u64);
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 __u64 *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 __u64 *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; __u64 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_MAX_VERSION 4
149
150 #define SB_SECTOR 8
151 #define SB_SIZE 4096
152 #define SB_LABEL_SIZE 32
153 #define SB_JOURNAL_BUCKETS 256U
154 /* SB_JOURNAL_BUCKETS must be divisible by BITS_PER_LONG */
155 #define MAX_CACHES_PER_SET 8
156
157 #define BDEV_DATA_START_DEFAULT 16 /* sectors */
158
159 struct cache_sb {
160 __u64 csum;
161 __u64 offset; /* sector where this sb was written */
162 __u64 version;
163
164 __u8 magic[16];
165
166 __u8 uuid[16];
167 union {
168 __u8 set_uuid[16];
169 __u64 set_magic;
170 };
171 __u8 label[SB_LABEL_SIZE];
172
173 __u64 flags;
174 __u64 seq;
175 __u64 pad[8];
176
177 union {
178 struct {
179 /* Cache devices */
180 __u64 nbuckets; /* device size */
181
182 __u16 block_size; /* sectors */
183 __u16 bucket_size; /* sectors */
184
185 __u16 nr_in_set;
186 __u16 nr_this_dev;
187 };
188 struct {
189 /* Backing devices */
190 __u64 data_offset;
191
192 /*
193 * block_size from the cache device section is still used by
194 * backing devices, so don't add anything here until we fix
195 * things to not need it for backing devices anymore
196 */
197 };
198 };
199
200 __u32 last_mount; /* time overflow in y2106 */
201
202 __u16 first_bucket;
203 union {
204 __u16 njournal_buckets;
205 __u16 keys;
206 };
207 __u64 d[SB_JOURNAL_BUCKETS]; /* journal buckets */
208 };
209
SB_IS_BDEV(const struct cache_sb * sb)210 static inline _Bool SB_IS_BDEV(const struct cache_sb *sb)
211 {
212 return sb->version == BCACHE_SB_VERSION_BDEV
213 || sb->version == BCACHE_SB_VERSION_BDEV_WITH_OFFSET;
214 }
215
216 BITMASK(CACHE_SYNC, struct cache_sb, flags, 0, 1);
217 BITMASK(CACHE_DISCARD, struct cache_sb, flags, 1, 1);
218 BITMASK(CACHE_REPLACEMENT, struct cache_sb, flags, 2, 3);
219 #define CACHE_REPLACEMENT_LRU 0U
220 #define CACHE_REPLACEMENT_FIFO 1U
221 #define CACHE_REPLACEMENT_RANDOM 2U
222
223 BITMASK(BDEV_CACHE_MODE, struct cache_sb, flags, 0, 4);
224 #define CACHE_MODE_WRITETHROUGH 0U
225 #define CACHE_MODE_WRITEBACK 1U
226 #define CACHE_MODE_WRITEAROUND 2U
227 #define CACHE_MODE_NONE 3U
228 BITMASK(BDEV_STATE, struct cache_sb, flags, 61, 2);
229 #define BDEV_STATE_NONE 0U
230 #define BDEV_STATE_CLEAN 1U
231 #define BDEV_STATE_DIRTY 2U
232 #define BDEV_STATE_STALE 3U
233
234 /*
235 * Magic numbers
236 *
237 * The various other data structures have their own magic numbers, which are
238 * xored with the first part of the cache set's UUID
239 */
240
241 #define JSET_MAGIC 0x245235c1a3625032ULL
242 #define PSET_MAGIC 0x6750e15f87337f91ULL
243 #define BSET_MAGIC 0x90135c78b99e07f5ULL
244
jset_magic(struct cache_sb * sb)245 static inline __u64 jset_magic(struct cache_sb *sb)
246 {
247 return sb->set_magic ^ JSET_MAGIC;
248 }
249
pset_magic(struct cache_sb * sb)250 static inline __u64 pset_magic(struct cache_sb *sb)
251 {
252 return sb->set_magic ^ PSET_MAGIC;
253 }
254
bset_magic(struct cache_sb * sb)255 static inline __u64 bset_magic(struct cache_sb *sb)
256 {
257 return sb->set_magic ^ BSET_MAGIC;
258 }
259
260 /*
261 * Journal
262 *
263 * On disk format for a journal entry:
264 * seq is monotonically increasing; every journal entry has its own unique
265 * sequence number.
266 *
267 * last_seq is the oldest journal entry that still has keys the btree hasn't
268 * flushed to disk yet.
269 *
270 * version is for on disk format changes.
271 */
272
273 #define BCACHE_JSET_VERSION_UUIDv1 1
274 #define BCACHE_JSET_VERSION_UUID 1 /* Always latest UUID format */
275 #define BCACHE_JSET_VERSION 1
276
277 struct jset {
278 __u64 csum;
279 __u64 magic;
280 __u64 seq;
281 __u32 version;
282 __u32 keys;
283
284 __u64 last_seq;
285
286 BKEY_PADDED(uuid_bucket);
287 BKEY_PADDED(btree_root);
288 __u16 btree_level;
289 __u16 pad[3];
290
291 __u64 prio_bucket[MAX_CACHES_PER_SET];
292
293 union {
294 struct bkey start[0];
295 __u64 d[0];
296 };
297 };
298
299 /* Bucket prios/gens */
300
301 struct prio_set {
302 __u64 csum;
303 __u64 magic;
304 __u64 seq;
305 __u32 version;
306 __u32 pad;
307
308 __u64 next_bucket;
309
310 struct bucket_disk {
311 __u16 prio;
312 __u8 gen;
313 } __attribute((packed)) data[];
314 };
315
316 /* UUIDS - per backing device/flash only volume metadata */
317
318 struct uuid_entry {
319 union {
320 struct {
321 __u8 uuid[16];
322 __u8 label[32];
323 __u32 first_reg; /* time overflow in y2106 */
324 __u32 last_reg;
325 __u32 invalidated;
326
327 __u32 flags;
328 /* Size of flash only volumes */
329 __u64 sectors;
330 };
331
332 __u8 pad[128];
333 };
334 };
335
336 BITMASK(UUID_FLASH_ONLY, struct uuid_entry, flags, 0, 1);
337
338 /* Btree nodes */
339
340 /* Version 1: Seed pointer into btree node checksum
341 */
342 #define BCACHE_BSET_CSUM 1
343 #define BCACHE_BSET_VERSION 1
344
345 /*
346 * Btree nodes
347 *
348 * On disk a btree node is a list/log of these; within each set the keys are
349 * sorted
350 */
351 struct bset {
352 __u64 csum;
353 __u64 magic;
354 __u64 seq;
355 __u32 version;
356 __u32 keys;
357
358 union {
359 struct bkey start[0];
360 __u64 d[0];
361 };
362 };
363
364 /* OBSOLETE */
365
366 /* UUIDS - per backing device/flash only volume metadata */
367
368 struct uuid_entry_v0 {
369 __u8 uuid[16];
370 __u8 label[32];
371 __u32 first_reg;
372 __u32 last_reg;
373 __u32 invalidated;
374 __u32 pad;
375 };
376
377 #endif /* _LINUX_BCACHE_H */
378