blob: 22b6ad31c706dae59544286faea5808d7b303342 [file] [log] [blame]
Kent Overstreet81ab4192013-10-31 15:46:42 -07001#ifndef _LINUX_BCACHE_H
2#define _LINUX_BCACHE_H
3
4/*
5 * Bcache on disk data structures
6 */
7
8#include <asm/types.h>
9
10#define BITMASK(name, type, field, offset, size) \
11static inline __u64 name(const type *k) \
12{ return (k->field >> offset) & ~(~0ULL << size); } \
13 \
14static inline void SET_##name(type *k, __u64 v) \
15{ \
16 k->field &= ~(~(~0ULL << size) << offset); \
17 k->field |= (v & ~(~0ULL << size)) << offset; \
18}
19
20/* Btree keys - all units are in sectors */
21
22struct bkey {
23 __u64 high;
24 __u64 low;
25 __u64 ptr[];
26};
27
28#define KEY_FIELD(name, field, offset, size) \
29 BITMASK(name, struct bkey, field, offset, size)
30
31#define PTR_FIELD(name, offset, size) \
32static inline __u64 name(const struct bkey *k, unsigned i) \
33{ return (k->ptr[i] >> offset) & ~(~0ULL << size); } \
34 \
35static inline void SET_##name(struct bkey *k, unsigned i, __u64 v) \
36{ \
37 k->ptr[i] &= ~(~(~0ULL << size) << offset); \
38 k->ptr[i] |= (v & ~(~0ULL << size)) << offset; \
39}
40
41#define KEY_SIZE_BITS 16
Kent Overstreet59158fd2013-11-11 19:03:54 -080042#define KEY_MAX_U64S 8
Kent Overstreet81ab4192013-10-31 15:46:42 -070043
44KEY_FIELD(KEY_PTRS, high, 60, 3)
45KEY_FIELD(HEADER_SIZE, high, 58, 2)
46KEY_FIELD(KEY_CSUM, high, 56, 2)
47KEY_FIELD(KEY_PINNED, high, 55, 1)
48KEY_FIELD(KEY_DIRTY, high, 36, 1)
49
50KEY_FIELD(KEY_SIZE, high, 20, KEY_SIZE_BITS)
51KEY_FIELD(KEY_INODE, high, 0, 20)
52
53/* Next time I change the on disk format, KEY_OFFSET() won't be 64 bits */
54
55static inline __u64 KEY_OFFSET(const struct bkey *k)
56{
57 return k->low;
58}
59
60static inline void SET_KEY_OFFSET(struct bkey *k, __u64 v)
61{
62 k->low = v;
63}
64
65/*
66 * The high bit being set is a relic from when we used it to do binary
67 * searches - it told you where a key started. It's not used anymore,
68 * and can probably be safely dropped.
69 */
70#define KEY(inode, offset, size) \
71((struct bkey) { \
72 .high = (1ULL << 63) | ((__u64) (size) << 20) | (inode), \
73 .low = (offset) \
74})
75
76#define ZERO_KEY KEY(0, 0, 0)
77
78#define MAX_KEY_INODE (~(~0 << 20))
79#define MAX_KEY_OFFSET (~0ULL >> 1)
80#define MAX_KEY KEY(MAX_KEY_INODE, MAX_KEY_OFFSET, 0)
81
82#define KEY_START(k) (KEY_OFFSET(k) - KEY_SIZE(k))
83#define START_KEY(k) KEY(KEY_INODE(k), KEY_START(k), 0)
84
85#define PTR_DEV_BITS 12
86
87PTR_FIELD(PTR_DEV, 51, PTR_DEV_BITS)
88PTR_FIELD(PTR_OFFSET, 8, 43)
89PTR_FIELD(PTR_GEN, 0, 8)
90
91#define PTR_CHECK_DEV ((1 << PTR_DEV_BITS) - 1)
92
93#define PTR(gen, offset, dev) \
94 ((((__u64) dev) << 51) | ((__u64) offset) << 8 | gen)
95
96/* Bkey utility code */
97
98static inline unsigned long bkey_u64s(const struct bkey *k)
99{
100 return (sizeof(struct bkey) / sizeof(__u64)) + KEY_PTRS(k);
101}
102
103static inline unsigned long bkey_bytes(const struct bkey *k)
104{
105 return bkey_u64s(k) * sizeof(__u64);
106}
107
108#define bkey_copy(_dest, _src) memcpy(_dest, _src, bkey_bytes(_src))
109
110static inline void bkey_copy_key(struct bkey *dest, const struct bkey *src)
111{
112 SET_KEY_INODE(dest, KEY_INODE(src));
113 SET_KEY_OFFSET(dest, KEY_OFFSET(src));
114}
115
116static inline struct bkey *bkey_next(const struct bkey *k)
117{
118 __u64 *d = (void *) k;
119 return (struct bkey *) (d + bkey_u64s(k));
120}
121
Kent Overstreetfafff812013-12-17 21:56:21 -0800122static inline struct bkey *bkey_idx(const struct bkey *k, unsigned nr_keys)
Kent Overstreet81ab4192013-10-31 15:46:42 -0700123{
124 __u64 *d = (void *) k;
125 return (struct bkey *) (d + nr_keys);
126}
127/* Enough for a key with 6 pointers */
128#define BKEY_PAD 8
129
130#define BKEY_PADDED(key) \
131 union { struct bkey key; __u64 key ## _pad[BKEY_PAD]; }
132
133/* Superblock */
134
135/* Version 0: Cache device
136 * Version 1: Backing device
137 * Version 2: Seed pointer into btree node checksum
138 * Version 3: Cache device with new UUID format
139 * Version 4: Backing device with data offset
140 */
141#define BCACHE_SB_VERSION_CDEV 0
142#define BCACHE_SB_VERSION_BDEV 1
143#define BCACHE_SB_VERSION_CDEV_WITH_UUID 3
144#define BCACHE_SB_VERSION_BDEV_WITH_OFFSET 4
145#define BCACHE_SB_MAX_VERSION 4
146
147#define SB_SECTOR 8
148#define SB_SIZE 4096
149#define SB_LABEL_SIZE 32
150#define SB_JOURNAL_BUCKETS 256U
151/* SB_JOURNAL_BUCKETS must be divisible by BITS_PER_LONG */
152#define MAX_CACHES_PER_SET 8
153
154#define BDEV_DATA_START_DEFAULT 16 /* sectors */
155
156struct cache_sb {
157 __u64 csum;
158 __u64 offset; /* sector where this sb was written */
159 __u64 version;
160
161 __u8 magic[16];
162
163 __u8 uuid[16];
164 union {
165 __u8 set_uuid[16];
166 __u64 set_magic;
167 };
168 __u8 label[SB_LABEL_SIZE];
169
170 __u64 flags;
171 __u64 seq;
172 __u64 pad[8];
173
174 union {
175 struct {
176 /* Cache devices */
177 __u64 nbuckets; /* device size */
178
179 __u16 block_size; /* sectors */
180 __u16 bucket_size; /* sectors */
181
182 __u16 nr_in_set;
183 __u16 nr_this_dev;
184 };
185 struct {
186 /* Backing devices */
187 __u64 data_offset;
188
189 /*
190 * block_size from the cache device section is still used by
191 * backing devices, so don't add anything here until we fix
192 * things to not need it for backing devices anymore
193 */
194 };
195 };
196
197 __u32 last_mount; /* time_t */
198
199 __u16 first_bucket;
200 union {
201 __u16 njournal_buckets;
202 __u16 keys;
203 };
204 __u64 d[SB_JOURNAL_BUCKETS]; /* journal buckets */
205};
206
207static inline _Bool SB_IS_BDEV(const struct cache_sb *sb)
208{
209 return sb->version == BCACHE_SB_VERSION_BDEV
210 || sb->version == BCACHE_SB_VERSION_BDEV_WITH_OFFSET;
211}
212
213BITMASK(CACHE_SYNC, struct cache_sb, flags, 0, 1);
214BITMASK(CACHE_DISCARD, struct cache_sb, flags, 1, 1);
215BITMASK(CACHE_REPLACEMENT, struct cache_sb, flags, 2, 3);
216#define CACHE_REPLACEMENT_LRU 0U
217#define CACHE_REPLACEMENT_FIFO 1U
218#define CACHE_REPLACEMENT_RANDOM 2U
219
220BITMASK(BDEV_CACHE_MODE, struct cache_sb, flags, 0, 4);
221#define CACHE_MODE_WRITETHROUGH 0U
222#define CACHE_MODE_WRITEBACK 1U
223#define CACHE_MODE_WRITEAROUND 2U
224#define CACHE_MODE_NONE 3U
225BITMASK(BDEV_STATE, struct cache_sb, flags, 61, 2);
226#define BDEV_STATE_NONE 0U
227#define BDEV_STATE_CLEAN 1U
228#define BDEV_STATE_DIRTY 2U
229#define BDEV_STATE_STALE 3U
230
231/*
232 * Magic numbers
233 *
234 * The various other data structures have their own magic numbers, which are
235 * xored with the first part of the cache set's UUID
236 */
237
238#define JSET_MAGIC 0x245235c1a3625032ULL
239#define PSET_MAGIC 0x6750e15f87337f91ULL
240#define BSET_MAGIC 0x90135c78b99e07f5ULL
241
242static inline __u64 jset_magic(struct cache_sb *sb)
243{
244 return sb->set_magic ^ JSET_MAGIC;
245}
246
247static inline __u64 pset_magic(struct cache_sb *sb)
248{
249 return sb->set_magic ^ PSET_MAGIC;
250}
251
252static inline __u64 bset_magic(struct cache_sb *sb)
253{
254 return sb->set_magic ^ BSET_MAGIC;
255}
256
257/*
258 * Journal
259 *
260 * On disk format for a journal entry:
261 * seq is monotonically increasing; every journal entry has its own unique
262 * sequence number.
263 *
264 * last_seq is the oldest journal entry that still has keys the btree hasn't
265 * flushed to disk yet.
266 *
267 * version is for on disk format changes.
268 */
269
270#define BCACHE_JSET_VERSION_UUIDv1 1
271#define BCACHE_JSET_VERSION_UUID 1 /* Always latest UUID format */
272#define BCACHE_JSET_VERSION 1
273
274struct jset {
275 __u64 csum;
276 __u64 magic;
277 __u64 seq;
278 __u32 version;
279 __u32 keys;
280
281 __u64 last_seq;
282
283 BKEY_PADDED(uuid_bucket);
284 BKEY_PADDED(btree_root);
285 __u16 btree_level;
286 __u16 pad[3];
287
288 __u64 prio_bucket[MAX_CACHES_PER_SET];
289
290 union {
291 struct bkey start[0];
292 __u64 d[0];
293 };
294};
295
296/* Bucket prios/gens */
297
298struct prio_set {
299 __u64 csum;
300 __u64 magic;
301 __u64 seq;
302 __u32 version;
303 __u32 pad;
304
305 __u64 next_bucket;
306
307 struct bucket_disk {
308 __u16 prio;
309 __u8 gen;
310 } __attribute((packed)) data[];
311};
312
313/* UUIDS - per backing device/flash only volume metadata */
314
315struct uuid_entry {
316 union {
317 struct {
318 __u8 uuid[16];
319 __u8 label[32];
320 __u32 first_reg;
321 __u32 last_reg;
322 __u32 invalidated;
323
324 __u32 flags;
325 /* Size of flash only volumes */
326 __u64 sectors;
327 };
328
329 __u8 pad[128];
330 };
331};
332
333BITMASK(UUID_FLASH_ONLY, struct uuid_entry, flags, 0, 1);
334
335/* Btree nodes */
336
337/* Version 1: Seed pointer into btree node checksum
338 */
339#define BCACHE_BSET_CSUM 1
340#define BCACHE_BSET_VERSION 1
341
342/*
343 * Btree nodes
344 *
345 * On disk a btree node is a list/log of these; within each set the keys are
346 * sorted
347 */
348struct bset {
349 __u64 csum;
350 __u64 magic;
351 __u64 seq;
352 __u32 version;
353 __u32 keys;
354
355 union {
356 struct bkey start[0];
357 __u64 d[0];
358 };
359};
360
361/* OBSOLETE */
362
363/* UUIDS - per backing device/flash only volume metadata */
364
365struct uuid_entry_v0 {
366 __u8 uuid[16];
367 __u8 label[32];
368 __u32 first_reg;
369 __u32 last_reg;
370 __u32 invalidated;
371 __u32 pad;
372};
373
374#endif /* _LINUX_BCACHE_H */