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pool; 82 } 83 84 /* 85 * Maximum supported by kernel client object name length 86 * 87 * (probably outdated: must be 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 static inline bool ceph_osdmap_flag(struct ceph_osdmap *map, int flag) 193 { 194 return map && (map->flags & flag); 195 } 196 197 extern char *ceph_osdmap_state_str(char *str, int len, int state); 198 extern u32 ceph_get_primary_affinity(struct ceph_osdmap *map, int osd); 199 200 static inline struct ceph_entity_addr *ceph_osd_addr(struct ceph_osdmap *map, 201 int osd) 202 { 203 if (osd >= map->max_osd) 204 return NULL; 205 return &map->osd_addr[osd]; 206 } 207 208 static inline int ceph_decode_pgid(void **p, void *end, struct ceph_pg *pgid) 209 { 210 __u8 version; 211 212 if (!ceph_has_room(p, end, 1 + 8 + 4 + 4)) { 213 pr_warn("incomplete pg encoding\n"); 214 return -EINVAL; 215 } 216 version = ceph_decode_8(p); 217 if (version > 1) { 218 pr_warn("do not understand pg encoding %d > 1\n", 219 (int)version); 220 return -EINVAL; 221 } 222 223 pgid->pool = ceph_decode_64(p); 224 pgid->seed = ceph_decode_32(p); 225 *p += 4; /* skip deprecated preferred value */ 226 227 return 0; 228 } 229 230 struct ceph_osdmap *ceph_osdmap_alloc(void); 231 extern struct ceph_osdmap *ceph_osdmap_decode(void **p, void *end); 232 struct ceph_osdmap *osdmap_apply_incremental(void **p, void *end, 233 struct ceph_osdmap *map); 234 extern void ceph_osdmap_destroy(struct ceph_osdmap *map); 235 236 struct ceph_osds { 237 int osds[CEPH_PG_MAX_SIZE]; 238 int size; 239 int primary; /* id, NOT index */ 240 }; 241 242 static inline void ceph_osds_init(struct ceph_osds *set) 243 { 244 set->size = 0; 245 set->primary = -1; 246 } 247 248 void ceph_osds_copy(struct ceph_osds *dest, const struct ceph_osds *src); 249 250 bool ceph_is_new_interval(const struct ceph_osds *old_acting, 251 const struct ceph_osds *new_acting, 252 const struct ceph_osds *old_up, 253 const struct ceph_osds *new_up, 254 int old_size, 255 int new_size, 256 int old_min_size, 257 int new_min_size, 258 u32 old_pg_num, 259 u32 new_pg_num, 260 bool old_sort_bitwise, 261 bool new_sort_bitwise, 262 const struct ceph_pg *pgid); 263 bool ceph_osds_changed(const struct ceph_osds *old_acting, 264 const struct ceph_osds *new_acting, 265 bool any_change); 266 267 /* calculate mapping of a file extent to an object */ 268 extern int ceph_calc_file_object_mapping(struct ceph_file_layout *layout, 269 u64 off, u64 len, 270 u64 *bno, u64 *oxoff, u64 *oxlen); 271 272 int ceph_object_locator_to_pg(struct ceph_osdmap *osdmap, 273 struct ceph_object_id *oid, 274 struct ceph_object_locator *oloc, 275 struct ceph_pg *raw_pgid); 276 277 void ceph_pg_to_up_acting_osds(struct ceph_osdmap *osdmap, 278 const struct ceph_pg *raw_pgid, 279 struct ceph_osds *up, 280 struct ceph_osds *acting); 281 int ceph_pg_to_acting_primary(struct ceph_osdmap *osdmap, 282 const struct ceph_pg *raw_pgid); 283 284 extern struct ceph_pg_pool_info *ceph_pg_pool_by_id(struct ceph_osdmap *map, 285 u64 id); 286 287 extern const char *ceph_pg_pool_name_by_id(struct ceph_osdmap *map, u64 id); 288 extern int ceph_pg_poolid_by_name(struct ceph_osdmap *map, const char *name); 289 290 #endif 291