xref: /linux-6.15/include/linux/ceph/osdmap.h (revision bb970707)
1 #ifndef _FS_CEPH_OSDMAP_H
2 #define _FS_CEPH_OSDMAP_H
3 
4 #include <linux/rbtree.h>
5 #include <linux/ceph/types.h>
6 #include <linux/ceph/decode.h>
7 #include <linux/ceph/ceph_fs.h>
8 #include <linux/crush/crush.h>
9 
10 /*
11  * The osd map describes the current membership of the osd cluster and
12  * specifies the mapping of objects to placement groups and placement
13  * groups to (sets of) osds.  That is, it completely specifies the
14  * (desired) distribution of all data objects in the system at some
15  * point in time.
16  *
17  * Each map version is identified by an epoch, which increases monotonically.
18  *
19  * The map can be updated either via an incremental map (diff) describing
20  * the change between two successive epochs, or as a fully encoded map.
21  */
22 struct ceph_pg {
23 	uint64_t pool;
24 	uint32_t seed;
25 };
26 
27 int ceph_pg_compare(const struct ceph_pg *lhs, const struct ceph_pg *rhs);
28 
29 #define CEPH_POOL_FLAG_HASHPSPOOL	(1ULL << 0) /* hash pg seed and pool id
30 						       together */
31 #define CEPH_POOL_FLAG_FULL		(1ULL << 1) /* pool is full */
32 
33 struct ceph_pg_pool_info {
34 	struct rb_node node;
35 	s64 id;
36 	u8 type; /* CEPH_POOL_TYPE_* */
37 	u8 size;
38 	u8 min_size;
39 	u8 crush_ruleset;
40 	u8 object_hash;
41 	u32 last_force_request_resend;
42 	u32 pg_num, pgp_num;
43 	int pg_num_mask, pgp_num_mask;
44 	s64 read_tier;
45 	s64 write_tier; /* wins for read+write ops */
46 	u64 flags; /* CEPH_POOL_FLAG_* */
47 	char *name;
48 
49 	bool was_full;  /* for handle_one_map() */
50 };
51 
52 static inline bool ceph_can_shift_osds(struct ceph_pg_pool_info *pool)
53 {
54 	switch (pool->type) {
55 	case CEPH_POOL_TYPE_REP:
56 		return true;
57 	case CEPH_POOL_TYPE_EC:
58 		return false;
59 	default:
60 		BUG_ON(1);
61 	}
62 }
63 
64 struct ceph_object_locator {
65 	s64 pool;
66 };
67 
68 static inline void ceph_oloc_init(struct ceph_object_locator *oloc)
69 {
70 	oloc->pool = -1;
71 }
72 
73 static inline bool ceph_oloc_empty(const struct ceph_object_locator *oloc)
74 {
75 	return oloc->pool == -1;
76 }
77 
78 static inline void ceph_oloc_copy(struct ceph_object_locator *dest,
79 				  const struct ceph_object_locator *src)
80 {
81 	dest->pool = src->pool;
82 }
83 
84 /*
85  * Maximum supported by kernel client object name length
86  *
87  * (probably outdated: must be >= RBD_MAX_MD_NAME_LEN -- currently 100)
88  */
89 #define CEPH_MAX_OID_NAME_LEN 100
90 
91 /*
92  * 51-char inline_name is long enough for all cephfs and all but one
93  * rbd requests: <imgname> in "<imgname>.rbd"/"rbd_id.<imgname>" can be
94  * arbitrarily long (~PAGE_SIZE).  It's done once during rbd map; all
95  * other rbd requests fit into inline_name.
96  *
97  * Makes ceph_object_id 64 bytes on 64-bit.
98  */
99 #define CEPH_OID_INLINE_LEN 52
100 
101 /*
102  * Both inline and external buffers have space for a NUL-terminator,
103  * which is carried around.  It's not required though - RADOS object
104  * names don't have to be NUL-terminated and may contain NULs.
105  */
106 struct ceph_object_id {
107 	char *name;
108 	char inline_name[CEPH_OID_INLINE_LEN];
109 	int name_len;
110 };
111 
112 static inline void ceph_oid_init(struct ceph_object_id *oid)
113 {
114 	oid->name = oid->inline_name;
115 	oid->name_len = 0;
116 }
117 
118 static inline bool ceph_oid_empty(const struct ceph_object_id *oid)
119 {
120 	return oid->name == oid->inline_name && !oid->name_len;
121 }
122 
123 void ceph_oid_copy(struct ceph_object_id *dest,
124 		   const struct ceph_object_id *src);
125 __printf(2, 3)
126 void ceph_oid_printf(struct ceph_object_id *oid, const char *fmt, ...);
127 __printf(3, 4)
128 int ceph_oid_aprintf(struct ceph_object_id *oid, gfp_t gfp,
129 		     const char *fmt, ...);
130 void ceph_oid_destroy(struct ceph_object_id *oid);
131 
132 struct ceph_pg_mapping {
133 	struct rb_node node;
134 	struct ceph_pg pgid;
135 
136 	union {
137 		struct {
138 			int len;
139 			int osds[];
140 		} pg_temp;
141 		struct {
142 			int osd;
143 		} primary_temp;
144 	};
145 };
146 
147 struct ceph_osdmap {
148 	struct ceph_fsid fsid;
149 	u32 epoch;
150 	struct ceph_timespec created, modified;
151 
152 	u32 flags;         /* CEPH_OSDMAP_* */
153 
154 	u32 max_osd;       /* size of osd_state, _offload, _addr arrays */
155 	u8 *osd_state;     /* CEPH_OSD_* */
156 	u32 *osd_weight;   /* 0 = failed, 0x10000 = 100% normal */
157 	struct ceph_entity_addr *osd_addr;
158 
159 	struct rb_root pg_temp;
160 	struct rb_root primary_temp;
161 
162 	u32 *osd_primary_affinity;
163 
164 	struct rb_root pg_pools;
165 	u32 pool_max;
166 
167 	/* the CRUSH map specifies the mapping of placement groups to
168 	 * the list of osds that store+replicate them. */
169 	struct crush_map *crush;
170 
171 	struct mutex crush_scratch_mutex;
172 	int crush_scratch_ary[CEPH_PG_MAX_SIZE * 3];
173 };
174 
175 static inline bool ceph_osd_exists(struct ceph_osdmap *map, int osd)
176 {
177 	return osd >= 0 && osd < map->max_osd &&
178 	       (map->osd_state[osd] & CEPH_OSD_EXISTS);
179 }
180 
181 static inline bool ceph_osd_is_up(struct ceph_osdmap *map, int osd)
182 {
183 	return ceph_osd_exists(map, osd) &&
184 	       (map->osd_state[osd] & CEPH_OSD_UP);
185 }
186 
187 static inline bool ceph_osd_is_down(struct ceph_osdmap *map, int osd)
188 {
189 	return !ceph_osd_is_up(map, osd);
190 }
191 
192 extern char *ceph_osdmap_state_str(char *str, int len, int state);
193 extern u32 ceph_get_primary_affinity(struct ceph_osdmap *map, int osd);
194 
195 static inline struct ceph_entity_addr *ceph_osd_addr(struct ceph_osdmap *map,
196 						     int osd)
197 {
198 	if (osd >= map->max_osd)
199 		return NULL;
200 	return &map->osd_addr[osd];
201 }
202 
203 static inline int ceph_decode_pgid(void **p, void *end, struct ceph_pg *pgid)
204 {
205 	__u8 version;
206 
207 	if (!ceph_has_room(p, end, 1 + 8 + 4 + 4)) {
208 		pr_warn("incomplete pg encoding\n");
209 		return -EINVAL;
210 	}
211 	version = ceph_decode_8(p);
212 	if (version > 1) {
213 		pr_warn("do not understand pg encoding %d > 1\n",
214 			(int)version);
215 		return -EINVAL;
216 	}
217 
218 	pgid->pool = ceph_decode_64(p);
219 	pgid->seed = ceph_decode_32(p);
220 	*p += 4;	/* skip deprecated preferred value */
221 
222 	return 0;
223 }
224 
225 struct ceph_osdmap *ceph_osdmap_alloc(void);
226 extern struct ceph_osdmap *ceph_osdmap_decode(void **p, void *end);
227 struct ceph_osdmap *osdmap_apply_incremental(void **p, void *end,
228 					     struct ceph_osdmap *map);
229 extern void ceph_osdmap_destroy(struct ceph_osdmap *map);
230 
231 struct ceph_osds {
232 	int osds[CEPH_PG_MAX_SIZE];
233 	int size;
234 	int primary; /* id, NOT index */
235 };
236 
237 static inline void ceph_osds_init(struct ceph_osds *set)
238 {
239 	set->size = 0;
240 	set->primary = -1;
241 }
242 
243 void ceph_osds_copy(struct ceph_osds *dest, const struct ceph_osds *src);
244 
245 bool ceph_is_new_interval(const struct ceph_osds *old_acting,
246 			  const struct ceph_osds *new_acting,
247 			  const struct ceph_osds *old_up,
248 			  const struct ceph_osds *new_up,
249 			  int old_size,
250 			  int new_size,
251 			  int old_min_size,
252 			  int new_min_size,
253 			  u32 old_pg_num,
254 			  u32 new_pg_num,
255 			  bool old_sort_bitwise,
256 			  bool new_sort_bitwise,
257 			  const struct ceph_pg *pgid);
258 bool ceph_osds_changed(const struct ceph_osds *old_acting,
259 		       const struct ceph_osds *new_acting,
260 		       bool any_change);
261 
262 /* calculate mapping of a file extent to an object */
263 extern int ceph_calc_file_object_mapping(struct ceph_file_layout *layout,
264 					 u64 off, u64 len,
265 					 u64 *bno, u64 *oxoff, u64 *oxlen);
266 
267 int ceph_object_locator_to_pg(struct ceph_osdmap *osdmap,
268 			      struct ceph_object_id *oid,
269 			      struct ceph_object_locator *oloc,
270 			      struct ceph_pg *raw_pgid);
271 
272 void ceph_pg_to_up_acting_osds(struct ceph_osdmap *osdmap,
273 			       const struct ceph_pg *raw_pgid,
274 			       struct ceph_osds *up,
275 			       struct ceph_osds *acting);
276 int ceph_pg_to_acting_primary(struct ceph_osdmap *osdmap,
277 			      const struct ceph_pg *raw_pgid);
278 
279 extern struct ceph_pg_pool_info *ceph_pg_pool_by_id(struct ceph_osdmap *map,
280 						    u64 id);
281 
282 extern const char *ceph_pg_pool_name_by_id(struct ceph_osdmap *map, u64 id);
283 extern int ceph_pg_poolid_by_name(struct ceph_osdmap *map, const char *name);
284 
285 #endif
286