include/linux/ceph/osdmap.h

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