xref: /linux/drivers/md/bcache/bcache_ondisk.h (revision 9a87ffc99ec8eb8d35eed7c4f816d75f5cc9662e)
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(__PAD0,	high, 58, 2)
47 KEY_FIELD(KEY_CSUM,	high, 56, 2)
48 KEY_FIELD(__PAD1,	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)	unsafe_memcpy(_dest, _src, bkey_bytes(_src), \
110 					/* bkey is always padded */)
111 
bkey_copy_key(struct bkey * dest,const struct bkey * src)112 static inline void bkey_copy_key(struct bkey *dest, const struct bkey *src)
113 {
114 	SET_KEY_INODE(dest, KEY_INODE(src));
115 	SET_KEY_OFFSET(dest, KEY_OFFSET(src));
116 }
117 
bkey_next(const struct bkey * k)118 static inline struct bkey *bkey_next(const struct bkey *k)
119 {
120 	__u64 *d = (void *) k;
121 
122 	return (struct bkey *) (d + bkey_u64s(k));
123 }
124 
bkey_idx(const struct bkey * k,unsigned int nr_keys)125 static inline struct bkey *bkey_idx(const struct bkey *k, unsigned int nr_keys)
126 {
127 	__u64 *d = (void *) k;
128 
129 	return (struct bkey *) (d + nr_keys);
130 }
131 /* Enough for a key with 6 pointers */
132 #define BKEY_PAD		8
133 
134 #define BKEY_PADDED(key)					\
135 	union { struct bkey key; __u64 key ## _pad[BKEY_PAD]; }
136 
137 /* Superblock */
138 
139 /* Version 0: Cache device
140  * Version 1: Backing device
141  * Version 2: Seed pointer into btree node checksum
142  * Version 3: Cache device with new UUID format
143  * Version 4: Backing device with data offset
144  */
145 #define BCACHE_SB_VERSION_CDEV			0
146 #define BCACHE_SB_VERSION_BDEV			1
147 #define BCACHE_SB_VERSION_CDEV_WITH_UUID	3
148 #define BCACHE_SB_VERSION_BDEV_WITH_OFFSET	4
149 #define BCACHE_SB_VERSION_CDEV_WITH_FEATURES	5
150 #define BCACHE_SB_VERSION_BDEV_WITH_FEATURES	6
151 #define BCACHE_SB_MAX_VERSION			6
152 
153 #define SB_SECTOR			8
154 #define SB_OFFSET			(SB_SECTOR << SECTOR_SHIFT)
155 #define SB_SIZE				4096
156 #define SB_LABEL_SIZE			32
157 #define SB_JOURNAL_BUCKETS		256U
158 /* SB_JOURNAL_BUCKETS must be divisible by BITS_PER_LONG */
159 #define MAX_CACHES_PER_SET		8
160 
161 #define BDEV_DATA_START_DEFAULT		16	/* sectors */
162 
163 struct cache_sb_disk {
164 	__le64			csum;
165 	__le64			offset;	/* sector where this sb was written */
166 	__le64			version;
167 
168 	__u8			magic[16];
169 
170 	__u8			uuid[16];
171 	union {
172 		__u8		set_uuid[16];
173 		__le64		set_magic;
174 	};
175 	__u8			label[SB_LABEL_SIZE];
176 
177 	__le64			flags;
178 	__le64			seq;
179 
180 	__le64			feature_compat;
181 	__le64			feature_incompat;
182 	__le64			feature_ro_compat;
183 
184 	__le64			pad[5];
185 
186 	union {
187 	struct {
188 		/* Cache devices */
189 		__le64		nbuckets;	/* device size */
190 
191 		__le16		block_size;	/* sectors */
192 		__le16		bucket_size;	/* sectors */
193 
194 		__le16		nr_in_set;
195 		__le16		nr_this_dev;
196 	};
197 	struct {
198 		/* Backing devices */
199 		__le64		data_offset;
200 
201 		/*
202 		 * block_size from the cache device section is still used by
203 		 * backing devices, so don't add anything here until we fix
204 		 * things to not need it for backing devices anymore
205 		 */
206 	};
207 	};
208 
209 	__le32			last_mount;	/* time overflow in y2106 */
210 
211 	__le16			first_bucket;
212 	union {
213 		__le16		njournal_buckets;
214 		__le16		keys;
215 	};
216 	__le64			d[SB_JOURNAL_BUCKETS];	/* journal buckets */
217 	__le16			obso_bucket_size_hi;	/* obsoleted */
218 };
219 
220 /*
221  * This is for in-memory bcache super block.
222  * NOTE: cache_sb is NOT exactly mapping to cache_sb_disk, the member
223  *       size, ordering and even whole struct size may be different
224  *       from cache_sb_disk.
225  */
226 struct cache_sb {
227 	__u64			offset;	/* sector where this sb was written */
228 	__u64			version;
229 
230 	__u8			magic[16];
231 
232 	__u8			uuid[16];
233 	union {
234 		__u8		set_uuid[16];
235 		__u64		set_magic;
236 	};
237 	__u8			label[SB_LABEL_SIZE];
238 
239 	__u64			flags;
240 	__u64			seq;
241 
242 	__u64			feature_compat;
243 	__u64			feature_incompat;
244 	__u64			feature_ro_compat;
245 
246 	union {
247 	struct {
248 		/* Cache devices */
249 		__u64		nbuckets;	/* device size */
250 
251 		__u16		block_size;	/* sectors */
252 		__u16		nr_in_set;
253 		__u16		nr_this_dev;
254 		__u32		bucket_size;	/* sectors */
255 	};
256 	struct {
257 		/* Backing devices */
258 		__u64		data_offset;
259 
260 		/*
261 		 * block_size from the cache device section is still used by
262 		 * backing devices, so don't add anything here until we fix
263 		 * things to not need it for backing devices anymore
264 		 */
265 	};
266 	};
267 
268 	__u32			last_mount;	/* time overflow in y2106 */
269 
270 	__u16			first_bucket;
271 	union {
272 		__u16		njournal_buckets;
273 		__u16		keys;
274 	};
275 	__u64			d[SB_JOURNAL_BUCKETS];	/* journal buckets */
276 };
277 
SB_IS_BDEV(const struct cache_sb * sb)278 static inline _Bool SB_IS_BDEV(const struct cache_sb *sb)
279 {
280 	return sb->version == BCACHE_SB_VERSION_BDEV
281 		|| sb->version == BCACHE_SB_VERSION_BDEV_WITH_OFFSET
282 		|| sb->version == BCACHE_SB_VERSION_BDEV_WITH_FEATURES;
283 }
284 
285 BITMASK(CACHE_SYNC,			struct cache_sb, flags, 0, 1);
286 BITMASK(CACHE_DISCARD,			struct cache_sb, flags, 1, 1);
287 BITMASK(CACHE_REPLACEMENT,		struct cache_sb, flags, 2, 3);
288 #define CACHE_REPLACEMENT_LRU		0U
289 #define CACHE_REPLACEMENT_FIFO		1U
290 #define CACHE_REPLACEMENT_RANDOM	2U
291 
292 BITMASK(BDEV_CACHE_MODE,		struct cache_sb, flags, 0, 4);
293 #define CACHE_MODE_WRITETHROUGH		0U
294 #define CACHE_MODE_WRITEBACK		1U
295 #define CACHE_MODE_WRITEAROUND		2U
296 #define CACHE_MODE_NONE			3U
297 BITMASK(BDEV_STATE,			struct cache_sb, flags, 61, 2);
298 #define BDEV_STATE_NONE			0U
299 #define BDEV_STATE_CLEAN		1U
300 #define BDEV_STATE_DIRTY		2U
301 #define BDEV_STATE_STALE		3U
302 
303 /*
304  * Magic numbers
305  *
306  * The various other data structures have their own magic numbers, which are
307  * xored with the first part of the cache set's UUID
308  */
309 
310 #define JSET_MAGIC			0x245235c1a3625032ULL
311 #define PSET_MAGIC			0x6750e15f87337f91ULL
312 #define BSET_MAGIC			0x90135c78b99e07f5ULL
313 
jset_magic(struct cache_sb * sb)314 static inline __u64 jset_magic(struct cache_sb *sb)
315 {
316 	return sb->set_magic ^ JSET_MAGIC;
317 }
318 
pset_magic(struct cache_sb * sb)319 static inline __u64 pset_magic(struct cache_sb *sb)
320 {
321 	return sb->set_magic ^ PSET_MAGIC;
322 }
323 
bset_magic(struct cache_sb * sb)324 static inline __u64 bset_magic(struct cache_sb *sb)
325 {
326 	return sb->set_magic ^ BSET_MAGIC;
327 }
328 
329 /*
330  * Journal
331  *
332  * On disk format for a journal entry:
333  * seq is monotonically increasing; every journal entry has its own unique
334  * sequence number.
335  *
336  * last_seq is the oldest journal entry that still has keys the btree hasn't
337  * flushed to disk yet.
338  *
339  * version is for on disk format changes.
340  */
341 
342 #define BCACHE_JSET_VERSION_UUIDv1	1
343 #define BCACHE_JSET_VERSION_UUID	1	/* Always latest UUID format */
344 #define BCACHE_JSET_VERSION		1
345 
346 struct jset {
347 	__u64			csum;
348 	__u64			magic;
349 	__u64			seq;
350 	__u32			version;
351 	__u32			keys;
352 
353 	__u64			last_seq;
354 
355 	BKEY_PADDED(uuid_bucket);
356 	BKEY_PADDED(btree_root);
357 	__u16			btree_level;
358 	__u16			pad[3];
359 
360 	__u64			prio_bucket[MAX_CACHES_PER_SET];
361 
362 	union {
363 		DECLARE_FLEX_ARRAY(struct bkey, start);
364 		DECLARE_FLEX_ARRAY(__u64, d);
365 	};
366 };
367 
368 /* Bucket prios/gens */
369 
370 struct prio_set {
371 	__u64			csum;
372 	__u64			magic;
373 	__u64			seq;
374 	__u32			version;
375 	__u32			pad;
376 
377 	__u64			next_bucket;
378 
379 	struct bucket_disk {
380 		__u16		prio;
381 		__u8		gen;
382 	} __attribute((packed)) data[];
383 };
384 
385 /* UUIDS - per backing device/flash only volume metadata */
386 
387 struct uuid_entry {
388 	union {
389 		struct {
390 			__u8	uuid[16];
391 			__u8	label[32];
392 			__u32	first_reg; /* time overflow in y2106 */
393 			__u32	last_reg;
394 			__u32	invalidated;
395 
396 			__u32	flags;
397 			/* Size of flash only volumes */
398 			__u64	sectors;
399 		};
400 
401 		__u8		pad[128];
402 	};
403 };
404 
405 BITMASK(UUID_FLASH_ONLY,	struct uuid_entry, flags, 0, 1);
406 
407 /* Btree nodes */
408 
409 /* Version 1: Seed pointer into btree node checksum
410  */
411 #define BCACHE_BSET_CSUM		1
412 #define BCACHE_BSET_VERSION		1
413 
414 /*
415  * Btree nodes
416  *
417  * On disk a btree node is a list/log of these; within each set the keys are
418  * sorted
419  */
420 struct bset {
421 	__u64			csum;
422 	__u64			magic;
423 	__u64			seq;
424 	__u32			version;
425 	__u32			keys;
426 
427 	union {
428 		DECLARE_FLEX_ARRAY(struct bkey, start);
429 		DECLARE_FLEX_ARRAY(__u64, d);
430 	};
431 };
432 
433 /* OBSOLETE */
434 
435 /* UUIDS - per backing device/flash only volume metadata */
436 
437 struct uuid_entry_v0 {
438 	__u8		uuid[16];
439 	__u8		label[32];
440 	__u32		first_reg;
441 	__u32		last_reg;
442 	__u32		invalidated;
443 	__u32		pad;
444 };
445 
446 #endif /* _LINUX_BCACHE_H */
447