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 #define CEPH_POOL_FLAG_FULL_QUOTA (1ULL << 10) /* pool ran out of quota, 41 will set FULL too */ 42 #define CEPH_POOL_FLAG_NEARFULL (1ULL << 11) /* pool is nearfull */ 43 44 struct ceph_pg_pool_info { 45 struct rb_node node; 46 s64 id; 47 u8 type; /* CEPH_POOL_TYPE_* */ 48 u8 size; 49 u8 min_size; 50 u8 crush_ruleset; 51 u8 object_hash; 52 u32 last_force_request_resend; 53 u32 pg_num, pgp_num; 54 int pg_num_mask, pgp_num_mask; 55 s64 read_tier; 56 s64 write_tier; /* wins for read+write ops */ 57 u64 flags; /* CEPH_POOL_FLAG_* */ 58 char *name; 59 60 bool was_full; /* for handle_one_map() */ 61 }; 62 63 static inline bool ceph_can_shift_osds(struct ceph_pg_pool_info *pool) 64 { 65 switch (pool->type) { 66 case CEPH_POOL_TYPE_REP: 67 return true; 68 case CEPH_POOL_TYPE_EC: 69 return false; 70 default: 71 BUG(); 72 } 73 } 74 75 struct ceph_object_locator { 76 s64 pool; 77 struct ceph_string *pool_ns; 78 }; 79 80 static inline void ceph_oloc_init(struct ceph_object_locator *oloc) 81 { 82 oloc->pool = -1; 83 oloc->pool_ns = NULL; 84 } 85 86 static inline bool ceph_oloc_empty(const struct ceph_object_locator *oloc) 87 { 88 return oloc->pool == -1; 89 } 90 91 void ceph_oloc_copy(struct ceph_object_locator *dest, 92 const struct ceph_object_locator *src); 93 void ceph_oloc_destroy(struct ceph_object_locator *oloc); 94 95 /* 96 * 51-char inline_name is long enough for all cephfs and all but one 97 * rbd requests: <imgname> in "<imgname>.rbd"/"rbd_id.<imgname>" can be 98 * arbitrarily long (~PAGE_SIZE). It's done once during rbd map; all 99 * other rbd requests fit into inline_name. 100 * 101 * Makes ceph_object_id 64 bytes on 64-bit. 102 */ 103 #define CEPH_OID_INLINE_LEN 52 104 105 /* 106 * Both inline and external buffers have space for a NUL-terminator, 107 * which is carried around. It's not required though - RADOS object 108 * names don't have to be NUL-terminated and may contain NULs. 109 */ 110 struct ceph_object_id { 111 char *name; 112 char inline_name[CEPH_OID_INLINE_LEN]; 113 int name_len; 114 }; 115 116 #define __CEPH_OID_INITIALIZER(oid) { .name = (oid).inline_name } 117 118 #define CEPH_DEFINE_OID_ONSTACK(oid) \ 119 struct ceph_object_id oid = __CEPH_OID_INITIALIZER(oid) 120 121 static inline void ceph_oid_init(struct ceph_object_id *oid) 122 { 123 *oid = (struct ceph_object_id) __CEPH_OID_INITIALIZER(*oid); 124 } 125 126 static inline bool ceph_oid_empty(const struct ceph_object_id *oid) 127 { 128 return oid->name == oid->inline_name && !oid->name_len; 129 } 130 131 void ceph_oid_copy(struct ceph_object_id *dest, 132 const struct ceph_object_id *src); 133 __printf(2, 3) 134 void ceph_oid_printf(struct ceph_object_id *oid, const char *fmt, ...); 135 __printf(3, 4) 136 int ceph_oid_aprintf(struct ceph_object_id *oid, gfp_t gfp, 137 const char *fmt, ...); 138 void ceph_oid_destroy(struct ceph_object_id *oid); 139 140 struct ceph_pg_mapping { 141 struct rb_node node; 142 struct ceph_pg pgid; 143 144 union { 145 struct { 146 int len; 147 int osds[]; 148 } pg_temp, pg_upmap; 149 struct { 150 int osd; 151 } primary_temp; 152 struct { 153 int len; 154 int from_to[][2]; 155 } pg_upmap_items; 156 }; 157 }; 158 159 struct ceph_osdmap { 160 struct ceph_fsid fsid; 161 u32 epoch; 162 struct ceph_timespec created, modified; 163 164 u32 flags; /* CEPH_OSDMAP_* */ 165 166 u32 max_osd; /* size of osd_state, _offload, _addr arrays */ 167 u32 *osd_state; /* CEPH_OSD_* */ 168 u32 *osd_weight; /* 0 = failed, 0x10000 = 100% normal */ 169 struct ceph_entity_addr *osd_addr; 170 171 struct rb_root pg_temp; 172 struct rb_root primary_temp; 173 174 /* remap (post-CRUSH, pre-up) */ 175 struct rb_root pg_upmap; /* PG := raw set */ 176 struct rb_root pg_upmap_items; /* from -> to within raw set */ 177 178 u32 *osd_primary_affinity; 179 180 struct rb_root pg_pools; 181 u32 pool_max; 182 183 /* the CRUSH map specifies the mapping of placement groups to 184 * the list of osds that store+replicate them. */ 185 struct crush_map *crush; 186 187 struct mutex crush_workspace_mutex; 188 void *crush_workspace; 189 }; 190 191 static inline bool ceph_osd_exists(struct ceph_osdmap *map, int osd) 192 { 193 return osd >= 0 && osd < map->max_osd && 194 (map->osd_state[osd] & CEPH_OSD_EXISTS); 195 } 196 197 static inline bool ceph_osd_is_up(struct ceph_osdmap *map, int osd) 198 { 199 return ceph_osd_exists(map, osd) && 200 (map->osd_state[osd] & CEPH_OSD_UP); 201 } 202 203 static inline bool ceph_osd_is_down(struct ceph_osdmap *map, int osd) 204 { 205 return !ceph_osd_is_up(map, osd); 206 } 207 208 char *ceph_osdmap_state_str(char *str, int len, u32 state); 209 extern u32 ceph_get_primary_affinity(struct ceph_osdmap *map, int osd); 210 211 static inline struct ceph_entity_addr *ceph_osd_addr(struct ceph_osdmap *map, 212 int osd) 213 { 214 if (osd >= map->max_osd) 215 return NULL; 216 return &map->osd_addr[osd]; 217 } 218 219 #define CEPH_PGID_ENCODING_LEN (1 + 8 + 4 + 4) 220 221 static inline int ceph_decode_pgid(void **p, void *end, struct ceph_pg *pgid) 222 { 223 __u8 version; 224 225 if (!ceph_has_room(p, end, CEPH_PGID_ENCODING_LEN)) { 226 pr_warn("incomplete pg encoding\n"); 227 return -EINVAL; 228 } 229 version = ceph_decode_8(p); 230 if (version > 1) { 231 pr_warn("do not understand pg encoding %d > 1\n", 232 (int)version); 233 return -EINVAL; 234 } 235 236 pgid->pool = ceph_decode_64(p); 237 pgid->seed = ceph_decode_32(p); 238 *p += 4; /* skip deprecated preferred value */ 239 240 return 0; 241 } 242 243 struct ceph_osdmap *ceph_osdmap_alloc(void); 244 extern struct ceph_osdmap *ceph_osdmap_decode(void **p, void *end); 245 struct ceph_osdmap *osdmap_apply_incremental(void **p, void *end, 246 struct ceph_osdmap *map); 247 extern void ceph_osdmap_destroy(struct ceph_osdmap *map); 248 249 struct ceph_osds { 250 int osds[CEPH_PG_MAX_SIZE]; 251 int size; 252 int primary; /* id, NOT index */ 253 }; 254 255 static inline void ceph_osds_init(struct ceph_osds *set) 256 { 257 set->size = 0; 258 set->primary = -1; 259 } 260 261 void ceph_osds_copy(struct ceph_osds *dest, const struct ceph_osds *src); 262 263 bool ceph_pg_is_split(const struct ceph_pg *pgid, u32 old_pg_num, 264 u32 new_pg_num); 265 bool ceph_is_new_interval(const struct ceph_osds *old_acting, 266 const struct ceph_osds *new_acting, 267 const struct ceph_osds *old_up, 268 const struct ceph_osds *new_up, 269 int old_size, 270 int new_size, 271 int old_min_size, 272 int new_min_size, 273 u32 old_pg_num, 274 u32 new_pg_num, 275 bool old_sort_bitwise, 276 bool new_sort_bitwise, 277 bool old_recovery_deletes, 278 bool new_recovery_deletes, 279 const struct ceph_pg *pgid); 280 bool ceph_osds_changed(const struct ceph_osds *old_acting, 281 const struct ceph_osds *new_acting, 282 bool any_change); 283 284 void __ceph_object_locator_to_pg(struct ceph_pg_pool_info *pi, 285 const struct ceph_object_id *oid, 286 const struct ceph_object_locator *oloc, 287 struct ceph_pg *raw_pgid); 288 int ceph_object_locator_to_pg(struct ceph_osdmap *osdmap, 289 const struct ceph_object_id *oid, 290 const struct ceph_object_locator *oloc, 291 struct ceph_pg *raw_pgid); 292 293 void ceph_pg_to_up_acting_osds(struct ceph_osdmap *osdmap, 294 struct ceph_pg_pool_info *pi, 295 const struct ceph_pg *raw_pgid, 296 struct ceph_osds *up, 297 struct ceph_osds *acting); 298 bool ceph_pg_to_primary_shard(struct ceph_osdmap *osdmap, 299 struct ceph_pg_pool_info *pi, 300 const struct ceph_pg *raw_pgid, 301 struct ceph_spg *spgid); 302 int ceph_pg_to_acting_primary(struct ceph_osdmap *osdmap, 303 const struct ceph_pg *raw_pgid); 304 305 extern struct ceph_pg_pool_info *ceph_pg_pool_by_id(struct ceph_osdmap *map, 306 u64 id); 307 308 extern const char *ceph_pg_pool_name_by_id(struct ceph_osdmap *map, u64 id); 309 extern int ceph_pg_poolid_by_name(struct ceph_osdmap *map, const char *name); 310 u64 ceph_pg_pool_flags(struct ceph_osdmap *map, u64 id); 311 312 #endif 313