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/ceph/ceph_fs.h> 9 #include <linux/crush/crush.h> 10 11 /* 12 * The osd map describes the current membership of the osd cluster and 13 * specifies the mapping of objects to placement groups and placement 14 * groups to (sets of) osds. That is, it completely specifies the 15 * (desired) distribution of all data objects in the system at some 16 * point in time. 17 * 18 * Each map version is identified by an epoch, which increases monotonically. 19 * 20 * The map can be updated either via an incremental map (diff) describing 21 * the change between two successive epochs, or as a fully encoded map. 22 */ 23 struct ceph_pg { 24 uint64_t pool; 25 uint32_t seed; 26 }; 27 28 #define CEPH_SPG_NOSHARD -1 29 30 struct ceph_spg { 31 struct ceph_pg pgid; 32 s8 shard; 33 }; 34 35 int ceph_pg_compare(const struct ceph_pg *lhs, const struct ceph_pg *rhs); 36 int ceph_spg_compare(const struct ceph_spg *lhs, const struct ceph_spg *rhs); 37 38 #define CEPH_POOL_FLAG_HASHPSPOOL (1ULL << 0) /* hash pg seed and pool id 39 together */ 40 #define CEPH_POOL_FLAG_FULL (1ULL << 1) /* pool is full */ 41 42 struct ceph_pg_pool_info { 43 struct rb_node node; 44 s64 id; 45 u8 type; /* CEPH_POOL_TYPE_* */ 46 u8 size; 47 u8 min_size; 48 u8 crush_ruleset; 49 u8 object_hash; 50 u32 last_force_request_resend; 51 u32 pg_num, pgp_num; 52 int pg_num_mask, pgp_num_mask; 53 s64 read_tier; 54 s64 write_tier; /* wins for read+write ops */ 55 u64 flags; /* CEPH_POOL_FLAG_* */ 56 char *name; 57 58 bool was_full; /* for handle_one_map() */ 59 }; 60 61 static inline bool ceph_can_shift_osds(struct ceph_pg_pool_info *pool) 62 { 63 switch (pool->type) { 64 case CEPH_POOL_TYPE_REP: 65 return true; 66 case CEPH_POOL_TYPE_EC: 67 return false; 68 default: 69 BUG(); 70 } 71 } 72 73 struct ceph_object_locator { 74 s64 pool; 75 struct ceph_string *pool_ns; 76 }; 77 78 static inline void ceph_oloc_init(struct ceph_object_locator *oloc) 79 { 80 oloc->pool = -1; 81 oloc->pool_ns = NULL; 82 } 83 84 static inline bool ceph_oloc_empty(const struct ceph_object_locator *oloc) 85 { 86 return oloc->pool == -1; 87 } 88 89 void ceph_oloc_copy(struct ceph_object_locator *dest, 90 const struct ceph_object_locator *src); 91 void ceph_oloc_destroy(struct ceph_object_locator *oloc); 92 93 /* 94 * 51-char inline_name is long enough for all cephfs and all but one 95 * rbd requests: <imgname> in "<imgname>.rbd"/"rbd_id.<imgname>" can be 96 * arbitrarily long (~PAGE_SIZE). It's done once during rbd map; all 97 * other rbd requests fit into inline_name. 98 * 99 * Makes ceph_object_id 64 bytes on 64-bit. 100 */ 101 #define CEPH_OID_INLINE_LEN 52 102 103 /* 104 * Both inline and external buffers have space for a NUL-terminator, 105 * which is carried around. It's not required though - RADOS object 106 * names don't have to be NUL-terminated and may contain NULs. 107 */ 108 struct ceph_object_id { 109 char *name; 110 char inline_name[CEPH_OID_INLINE_LEN]; 111 int name_len; 112 }; 113 114 static inline void ceph_oid_init(struct ceph_object_id *oid) 115 { 116 oid->name = oid->inline_name; 117 oid->name_len = 0; 118 } 119 120 #define CEPH_OID_INIT_ONSTACK(oid) \ 121 ({ ceph_oid_init(&oid); oid; }) 122 #define CEPH_DEFINE_OID_ONSTACK(oid) \ 123 struct ceph_object_id oid = CEPH_OID_INIT_ONSTACK(oid) 124 125 static inline bool ceph_oid_empty(const struct ceph_object_id *oid) 126 { 127 return oid->name == oid->inline_name && !oid->name_len; 128 } 129 130 void ceph_oid_copy(struct ceph_object_id *dest, 131 const struct ceph_object_id *src); 132 __printf(2, 3) 133 void ceph_oid_printf(struct ceph_object_id *oid, const char *fmt, ...); 134 __printf(3, 4) 135 int ceph_oid_aprintf(struct ceph_object_id *oid, gfp_t gfp, 136 const char *fmt, ...); 137 void ceph_oid_destroy(struct ceph_object_id *oid); 138 139 struct ceph_pg_mapping { 140 struct rb_node node; 141 struct ceph_pg pgid; 142 143 union { 144 struct { 145 int len; 146 int osds[]; 147 } pg_temp, pg_upmap; 148 struct { 149 int osd; 150 } primary_temp; 151 struct { 152 int len; 153 int from_to[][2]; 154 } pg_upmap_items; 155 }; 156 }; 157 158 struct ceph_osdmap { 159 struct ceph_fsid fsid; 160 u32 epoch; 161 struct ceph_timespec created, modified; 162 163 u32 flags; /* CEPH_OSDMAP_* */ 164 165 u32 max_osd; /* size of osd_state, _offload, _addr arrays */ 166 u32 *osd_state; /* CEPH_OSD_* */ 167 u32 *osd_weight; /* 0 = failed, 0x10000 = 100% normal */ 168 struct ceph_entity_addr *osd_addr; 169 170 struct rb_root pg_temp; 171 struct rb_root primary_temp; 172 173 /* remap (post-CRUSH, pre-up) */ 174 struct rb_root pg_upmap; /* PG := raw set */ 175 struct rb_root pg_upmap_items; /* from -> to within raw set */ 176 177 u32 *osd_primary_affinity; 178 179 struct rb_root pg_pools; 180 u32 pool_max; 181 182 /* the CRUSH map specifies the mapping of placement groups to 183 * the list of osds that store+replicate them. */ 184 struct crush_map *crush; 185 186 struct mutex crush_workspace_mutex; 187 void *crush_workspace; 188 }; 189 190 static inline bool ceph_osd_exists(struct ceph_osdmap *map, int osd) 191 { 192 return osd >= 0 && osd < map->max_osd && 193 (map->osd_state[osd] & CEPH_OSD_EXISTS); 194 } 195 196 static inline bool ceph_osd_is_up(struct ceph_osdmap *map, int osd) 197 { 198 return ceph_osd_exists(map, osd) && 199 (map->osd_state[osd] & CEPH_OSD_UP); 200 } 201 202 static inline bool ceph_osd_is_down(struct ceph_osdmap *map, int osd) 203 { 204 return !ceph_osd_is_up(map, osd); 205 } 206 207 char *ceph_osdmap_state_str(char *str, int len, u32 state); 208 extern u32 ceph_get_primary_affinity(struct ceph_osdmap *map, int osd); 209 210 static inline struct ceph_entity_addr *ceph_osd_addr(struct ceph_osdmap *map, 211 int osd) 212 { 213 if (osd >= map->max_osd) 214 return NULL; 215 return &map->osd_addr[osd]; 216 } 217 218 #define CEPH_PGID_ENCODING_LEN (1 + 8 + 4 + 4) 219 220 static inline int ceph_decode_pgid(void **p, void *end, struct ceph_pg *pgid) 221 { 222 __u8 version; 223 224 if (!ceph_has_room(p, end, CEPH_PGID_ENCODING_LEN)) { 225 pr_warn("incomplete pg encoding\n"); 226 return -EINVAL; 227 } 228 version = ceph_decode_8(p); 229 if (version > 1) { 230 pr_warn("do not understand pg encoding %d > 1\n", 231 (int)version); 232 return -EINVAL; 233 } 234 235 pgid->pool = ceph_decode_64(p); 236 pgid->seed = ceph_decode_32(p); 237 *p += 4; /* skip deprecated preferred value */ 238 239 return 0; 240 } 241 242 struct ceph_osdmap *ceph_osdmap_alloc(void); 243 extern struct ceph_osdmap *ceph_osdmap_decode(void **p, void *end); 244 struct ceph_osdmap *osdmap_apply_incremental(void **p, void *end, 245 struct ceph_osdmap *map); 246 extern void ceph_osdmap_destroy(struct ceph_osdmap *map); 247 248 struct ceph_osds { 249 int osds[CEPH_PG_MAX_SIZE]; 250 int size; 251 int primary; /* id, NOT index */ 252 }; 253 254 static inline void ceph_osds_init(struct ceph_osds *set) 255 { 256 set->size = 0; 257 set->primary = -1; 258 } 259 260 void ceph_osds_copy(struct ceph_osds *dest, const struct ceph_osds *src); 261 262 bool ceph_pg_is_split(const struct ceph_pg *pgid, u32 old_pg_num, 263 u32 new_pg_num); 264 bool ceph_is_new_interval(const struct ceph_osds *old_acting, 265 const struct ceph_osds *new_acting, 266 const struct ceph_osds *old_up, 267 const struct ceph_osds *new_up, 268 int old_size, 269 int new_size, 270 int old_min_size, 271 int new_min_size, 272 u32 old_pg_num, 273 u32 new_pg_num, 274 bool old_sort_bitwise, 275 bool new_sort_bitwise, 276 bool old_recovery_deletes, 277 bool new_recovery_deletes, 278 const struct ceph_pg *pgid); 279 bool ceph_osds_changed(const struct ceph_osds *old_acting, 280 const struct ceph_osds *new_acting, 281 bool any_change); 282 283 void ceph_calc_file_object_mapping(struct ceph_file_layout *l, 284 u64 off, u64 len, 285 u64 *objno, u64 *objoff, u32 *xlen); 286 287 int __ceph_object_locator_to_pg(struct ceph_pg_pool_info *pi, 288 const struct ceph_object_id *oid, 289 const struct ceph_object_locator *oloc, 290 struct ceph_pg *raw_pgid); 291 int ceph_object_locator_to_pg(struct ceph_osdmap *osdmap, 292 const struct ceph_object_id *oid, 293 const struct ceph_object_locator *oloc, 294 struct ceph_pg *raw_pgid); 295 296 void ceph_pg_to_up_acting_osds(struct ceph_osdmap *osdmap, 297 struct ceph_pg_pool_info *pi, 298 const struct ceph_pg *raw_pgid, 299 struct ceph_osds *up, 300 struct ceph_osds *acting); 301 bool ceph_pg_to_primary_shard(struct ceph_osdmap *osdmap, 302 struct ceph_pg_pool_info *pi, 303 const struct ceph_pg *raw_pgid, 304 struct ceph_spg *spgid); 305 int ceph_pg_to_acting_primary(struct ceph_osdmap *osdmap, 306 const struct ceph_pg *raw_pgid); 307 308 extern struct ceph_pg_pool_info *ceph_pg_pool_by_id(struct ceph_osdmap *map, 309 u64 id); 310 311 extern const char *ceph_pg_pool_name_by_id(struct ceph_osdmap *map, u64 id); 312 extern int ceph_pg_poolid_by_name(struct ceph_osdmap *map, const char *name); 313 314 #endif 315