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ceph_object_locator *oloc); 83 84 /* 85 * Maximum supported by kernel client object name length 86 * 87 * (probably outdated: must be 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 #define CEPH_OID_INIT_ONSTACK(oid) \ 119 ({ ceph_oid_init(&oid); oid; }) 120 #define CEPH_DEFINE_OID_ONSTACK(oid) \ 121 struct ceph_object_id oid = CEPH_OID_INIT_ONSTACK(oid) 122 123 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; 146 struct { 147 int osd; 148 } primary_temp; 149 }; 150 }; 151 152 struct ceph_osdmap { 153 struct ceph_fsid fsid; 154 u32 epoch; 155 struct ceph_timespec created, modified; 156 157 u32 flags; /* CEPH_OSDMAP_* */ 158 159 u32 max_osd; /* size of osd_state, _offload, _addr arrays */ 160 u8 *osd_state; /* CEPH_OSD_* */ 161 u32 *osd_weight; /* 0 = failed, 0x10000 = 100% normal */ 162 struct ceph_entity_addr *osd_addr; 163 164 struct rb_root pg_temp; 165 struct rb_root primary_temp; 166 167 u32 *osd_primary_affinity; 168 169 struct rb_root pg_pools; 170 u32 pool_max; 171 172 /* the CRUSH map specifies the mapping of placement groups to 173 * the list of osds that store+replicate them. */ 174 struct crush_map *crush; 175 176 struct mutex crush_scratch_mutex; 177 int crush_scratch_ary[CEPH_PG_MAX_SIZE * 3]; 178 void *crush_workspace; 179 }; 180 181 static inline bool ceph_osd_exists(struct ceph_osdmap *map, int osd) 182 { 183 return osd >= 0 && osd < map->max_osd && 184 (map->osd_state[osd] & CEPH_OSD_EXISTS); 185 } 186 187 static inline bool ceph_osd_is_up(struct ceph_osdmap *map, int osd) 188 { 189 return ceph_osd_exists(map, osd) && 190 (map->osd_state[osd] & CEPH_OSD_UP); 191 } 192 193 static inline bool ceph_osd_is_down(struct ceph_osdmap *map, int osd) 194 { 195 return !ceph_osd_is_up(map, osd); 196 } 197 198 extern char *ceph_osdmap_state_str(char *str, int len, int state); 199 extern u32 ceph_get_primary_affinity(struct ceph_osdmap *map, int osd); 200 201 static inline struct ceph_entity_addr *ceph_osd_addr(struct ceph_osdmap *map, 202 int osd) 203 { 204 if (osd >= map->max_osd) 205 return NULL; 206 return &map->osd_addr[osd]; 207 } 208 209 static inline int ceph_decode_pgid(void **p, void *end, struct ceph_pg *pgid) 210 { 211 __u8 version; 212 213 if (!ceph_has_room(p, end, 1 + 8 + 4 + 4)) { 214 pr_warn("incomplete pg encoding\n"); 215 return -EINVAL; 216 } 217 version = ceph_decode_8(p); 218 if (version > 1) { 219 pr_warn("do not understand pg encoding %d > 1\n", 220 (int)version); 221 return -EINVAL; 222 } 223 224 pgid->pool = ceph_decode_64(p); 225 pgid->seed = ceph_decode_32(p); 226 *p += 4; /* skip deprecated preferred value */ 227 228 return 0; 229 } 230 231 struct ceph_osdmap *ceph_osdmap_alloc(void); 232 extern struct ceph_osdmap *ceph_osdmap_decode(void **p, void *end); 233 struct ceph_osdmap *osdmap_apply_incremental(void **p, void *end, 234 struct ceph_osdmap *map); 235 extern void ceph_osdmap_destroy(struct ceph_osdmap *map); 236 237 struct ceph_osds { 238 int osds[CEPH_PG_MAX_SIZE]; 239 int size; 240 int primary; /* id, NOT index */ 241 }; 242 243 static inline void ceph_osds_init(struct ceph_osds *set) 244 { 245 set->size = 0; 246 set->primary = -1; 247 } 248 249 void ceph_osds_copy(struct ceph_osds *dest, const struct ceph_osds *src); 250 251 bool ceph_is_new_interval(const struct ceph_osds *old_acting, 252 const struct ceph_osds *new_acting, 253 const struct ceph_osds *old_up, 254 const struct ceph_osds *new_up, 255 int old_size, 256 int new_size, 257 int old_min_size, 258 int new_min_size, 259 u32 old_pg_num, 260 u32 new_pg_num, 261 bool old_sort_bitwise, 262 bool new_sort_bitwise, 263 const struct ceph_pg *pgid); 264 bool ceph_osds_changed(const struct ceph_osds *old_acting, 265 const struct ceph_osds *new_acting, 266 bool any_change); 267 268 /* calculate mapping of a file extent to an object */ 269 extern int ceph_calc_file_object_mapping(struct ceph_file_layout *layout, 270 u64 off, u64 len, 271 u64 *bno, u64 *oxoff, u64 *oxlen); 272 273 int ceph_object_locator_to_pg(struct ceph_osdmap *osdmap, 274 struct ceph_object_id *oid, 275 struct ceph_object_locator *oloc, 276 struct ceph_pg *raw_pgid); 277 278 void ceph_pg_to_up_acting_osds(struct ceph_osdmap *osdmap, 279 const struct ceph_pg *raw_pgid, 280 struct ceph_osds *up, 281 struct ceph_osds *acting); 282 int ceph_pg_to_acting_primary(struct ceph_osdmap *osdmap, 283 const struct ceph_pg *raw_pgid); 284 285 extern struct ceph_pg_pool_info *ceph_pg_pool_by_id(struct ceph_osdmap *map, 286 u64 id); 287 288 extern const char *ceph_pg_pool_name_by_id(struct ceph_osdmap *map, u64 id); 289 extern int ceph_pg_poolid_by_name(struct ceph_osdmap *map, const char *name); 290 291 #endif 292