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