1 /* SPDX-License-Identifier: GPL-2.0 */
2 #ifndef _FS_CEPH_OSDMAP_H
3 #define _FS_CEPH_OSDMAP_H
4
5 #include <linux/rbtree.h>
6 #include <linux/ceph/types.h>
7 #include <linux/ceph/decode.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 #define CEPH_SPG_NOSHARD -1
28
29 struct ceph_spg {
30 struct ceph_pg pgid;
31 s8 shard;
32 };
33
34 int ceph_pg_compare(const struct ceph_pg *lhs, const struct ceph_pg *rhs);
35 int ceph_spg_compare(const struct ceph_spg *lhs, const struct ceph_spg *rhs);
36
37 #define CEPH_POOL_FLAG_HASHPSPOOL (1ULL << 0) /* hash pg seed and pool id
38 together */
39 #define CEPH_POOL_FLAG_FULL (1ULL << 1) /* pool is full */
40
41 struct ceph_pg_pool_info {
42 struct rb_node node;
43 s64 id;
44 u8 type; /* CEPH_POOL_TYPE_* */
45 u8 size;
46 u8 min_size;
47 u8 crush_ruleset;
48 u8 object_hash;
49 u32 last_force_request_resend;
50 u32 pg_num, pgp_num;
51 int pg_num_mask, pgp_num_mask;
52 s64 read_tier;
53 s64 write_tier; /* wins for read+write ops */
54 u64 flags; /* CEPH_POOL_FLAG_* */
55 char *name;
56
57 bool was_full; /* for handle_one_map() */
58 };
59
ceph_can_shift_osds(struct ceph_pg_pool_info * pool)60 static inline bool ceph_can_shift_osds(struct ceph_pg_pool_info *pool)
61 {
62 switch (pool->type) {
63 case CEPH_POOL_TYPE_REP:
64 return true;
65 case CEPH_POOL_TYPE_EC:
66 return false;
67 default:
68 BUG();
69 }
70 }
71
72 struct ceph_object_locator {
73 s64 pool;
74 struct ceph_string *pool_ns;
75 };
76
ceph_oloc_init(struct ceph_object_locator * oloc)77 static inline void ceph_oloc_init(struct ceph_object_locator *oloc)
78 {
79 oloc->pool = -1;
80 oloc->pool_ns = NULL;
81 }
82
ceph_oloc_empty(const struct ceph_object_locator * oloc)83 static inline bool ceph_oloc_empty(const struct ceph_object_locator *oloc)
84 {
85 return oloc->pool == -1;
86 }
87
88 void ceph_oloc_copy(struct ceph_object_locator *dest,
89 const struct ceph_object_locator *src);
90 void ceph_oloc_destroy(struct ceph_object_locator *oloc);
91
92 /*
93 * 51-char inline_name is long enough for all cephfs and all but one
94 * rbd requests: <imgname> in "<imgname>.rbd"/"rbd_id.<imgname>" can be
95 * arbitrarily long (~PAGE_SIZE). It's done once during rbd map; all
96 * other rbd requests fit into inline_name.
97 *
98 * Makes ceph_object_id 64 bytes on 64-bit.
99 */
100 #define CEPH_OID_INLINE_LEN 52
101
102 /*
103 * Both inline and external buffers have space for a NUL-terminator,
104 * which is carried around. It's not required though - RADOS object
105 * names don't have to be NUL-terminated and may contain NULs.
106 */
107 struct ceph_object_id {
108 char *name;
109 char inline_name[CEPH_OID_INLINE_LEN];
110 int name_len;
111 };
112
113 #define __CEPH_OID_INITIALIZER(oid) { .name = (oid).inline_name }
114
115 #define CEPH_DEFINE_OID_ONSTACK(oid) \
116 struct ceph_object_id oid = __CEPH_OID_INITIALIZER(oid)
117
ceph_oid_init(struct ceph_object_id * oid)118 static inline void ceph_oid_init(struct ceph_object_id *oid)
119 {
120 *oid = (struct ceph_object_id) __CEPH_OID_INITIALIZER(*oid);
121 }
122
ceph_oid_empty(const struct ceph_object_id * oid)123 static inline bool ceph_oid_empty(const struct ceph_object_id *oid)
124 {
125 return oid->name == oid->inline_name && !oid->name_len;
126 }
127
128 void ceph_oid_copy(struct ceph_object_id *dest,
129 const struct ceph_object_id *src);
130 __printf(2, 3)
131 void ceph_oid_printf(struct ceph_object_id *oid, const char *fmt, ...);
132 __printf(3, 4)
133 int ceph_oid_aprintf(struct ceph_object_id *oid, gfp_t gfp,
134 const char *fmt, ...);
135 void ceph_oid_destroy(struct ceph_object_id *oid);
136
137 struct ceph_pg_mapping {
138 struct rb_node node;
139 struct ceph_pg pgid;
140
141 union {
142 struct {
143 int len;
144 int osds[];
145 } pg_temp, pg_upmap;
146 struct {
147 int osd;
148 } primary_temp;
149 struct {
150 int len;
151 int from_to[][2];
152 } pg_upmap_items;
153 };
154 };
155
156 struct ceph_osdmap {
157 struct ceph_fsid fsid;
158 u32 epoch;
159 struct ceph_timespec created, modified;
160
161 u32 flags; /* CEPH_OSDMAP_* */
162
163 u32 max_osd; /* size of osd_state, _offload, _addr arrays */
164 u32 *osd_state; /* CEPH_OSD_* */
165 u32 *osd_weight; /* 0 = failed, 0x10000 = 100% normal */
166 struct ceph_entity_addr *osd_addr;
167
168 struct rb_root pg_temp;
169 struct rb_root primary_temp;
170
171 /* remap (post-CRUSH, pre-up) */
172 struct rb_root pg_upmap; /* PG := raw set */
173 struct rb_root pg_upmap_items; /* from -> to within raw set */
174
175 u32 *osd_primary_affinity;
176
177 struct rb_root pg_pools;
178 u32 pool_max;
179
180 /* the CRUSH map specifies the mapping of placement groups to
181 * the list of osds that store+replicate them. */
182 struct crush_map *crush;
183
184 struct mutex crush_workspace_mutex;
185 void *crush_workspace;
186 };
187
ceph_osd_exists(struct ceph_osdmap * map,int osd)188 static inline bool ceph_osd_exists(struct ceph_osdmap *map, int osd)
189 {
190 return osd >= 0 && osd < map->max_osd &&
191 (map->osd_state[osd] & CEPH_OSD_EXISTS);
192 }
193
ceph_osd_is_up(struct ceph_osdmap * map,int osd)194 static inline bool ceph_osd_is_up(struct ceph_osdmap *map, int osd)
195 {
196 return ceph_osd_exists(map, osd) &&
197 (map->osd_state[osd] & CEPH_OSD_UP);
198 }
199
ceph_osd_is_down(struct ceph_osdmap * map,int osd)200 static inline bool ceph_osd_is_down(struct ceph_osdmap *map, int osd)
201 {
202 return !ceph_osd_is_up(map, osd);
203 }
204
205 char *ceph_osdmap_state_str(char *str, int len, u32 state);
206 extern u32 ceph_get_primary_affinity(struct ceph_osdmap *map, int osd);
207
ceph_osd_addr(struct ceph_osdmap * map,int osd)208 static inline struct ceph_entity_addr *ceph_osd_addr(struct ceph_osdmap *map,
209 int osd)
210 {
211 if (osd >= map->max_osd)
212 return NULL;
213 return &map->osd_addr[osd];
214 }
215
216 #define CEPH_PGID_ENCODING_LEN (1 + 8 + 4 + 4)
217
ceph_decode_pgid(void ** p,void * end,struct ceph_pg * pgid)218 static inline int ceph_decode_pgid(void **p, void *end, struct ceph_pg *pgid)
219 {
220 __u8 version;
221
222 if (!ceph_has_room(p, end, CEPH_PGID_ENCODING_LEN)) {
223 pr_warn("incomplete pg encoding\n");
224 return -EINVAL;
225 }
226 version = ceph_decode_8(p);
227 if (version > 1) {
228 pr_warn("do not understand pg encoding %d > 1\n",
229 (int)version);
230 return -EINVAL;
231 }
232
233 pgid->pool = ceph_decode_64(p);
234 pgid->seed = ceph_decode_32(p);
235 *p += 4; /* skip deprecated preferred value */
236
237 return 0;
238 }
239
240 struct ceph_osdmap *ceph_osdmap_alloc(void);
241 extern struct ceph_osdmap *ceph_osdmap_decode(void **p, void *end);
242 struct ceph_osdmap *osdmap_apply_incremental(void **p, void *end,
243 struct ceph_osdmap *map);
244 extern void ceph_osdmap_destroy(struct ceph_osdmap *map);
245
246 struct ceph_osds {
247 int osds[CEPH_PG_MAX_SIZE];
248 int size;
249 int primary; /* id, NOT index */
250 };
251
ceph_osds_init(struct ceph_osds * set)252 static inline void ceph_osds_init(struct ceph_osds *set)
253 {
254 set->size = 0;
255 set->primary = -1;
256 }
257
258 void ceph_osds_copy(struct ceph_osds *dest, const struct ceph_osds *src);
259
260 bool ceph_pg_is_split(const struct ceph_pg *pgid, u32 old_pg_num,
261 u32 new_pg_num);
262 bool ceph_is_new_interval(const struct ceph_osds *old_acting,
263 const struct ceph_osds *new_acting,
264 const struct ceph_osds *old_up,
265 const struct ceph_osds *new_up,
266 int old_size,
267 int new_size,
268 int old_min_size,
269 int new_min_size,
270 u32 old_pg_num,
271 u32 new_pg_num,
272 bool old_sort_bitwise,
273 bool new_sort_bitwise,
274 bool old_recovery_deletes,
275 bool new_recovery_deletes,
276 const struct ceph_pg *pgid);
277 bool ceph_osds_changed(const struct ceph_osds *old_acting,
278 const struct ceph_osds *new_acting,
279 bool any_change);
280
281 void __ceph_object_locator_to_pg(struct ceph_pg_pool_info *pi,
282 const struct ceph_object_id *oid,
283 const struct ceph_object_locator *oloc,
284 struct ceph_pg *raw_pgid);
285 int ceph_object_locator_to_pg(struct ceph_osdmap *osdmap,
286 const struct ceph_object_id *oid,
287 const struct ceph_object_locator *oloc,
288 struct ceph_pg *raw_pgid);
289
290 void ceph_pg_to_up_acting_osds(struct ceph_osdmap *osdmap,
291 struct ceph_pg_pool_info *pi,
292 const struct ceph_pg *raw_pgid,
293 struct ceph_osds *up,
294 struct ceph_osds *acting);
295 bool ceph_pg_to_primary_shard(struct ceph_osdmap *osdmap,
296 struct ceph_pg_pool_info *pi,
297 const struct ceph_pg *raw_pgid,
298 struct ceph_spg *spgid);
299 int ceph_pg_to_acting_primary(struct ceph_osdmap *osdmap,
300 const struct ceph_pg *raw_pgid);
301
302 extern struct ceph_pg_pool_info *ceph_pg_pool_by_id(struct ceph_osdmap *map,
303 u64 id);
304
305 extern const char *ceph_pg_pool_name_by_id(struct ceph_osdmap *map, u64 id);
306 extern int ceph_pg_poolid_by_name(struct ceph_osdmap *map, const char *name);
307
308 #endif
309
