blob: ba4c205cbb016a141495e2e872ee6a3cfd57253a [file] [log] [blame]
Sage Weilf24e9982009-10-06 11:31:10 -07001#ifndef _FS_CEPH_OSDMAP_H
2#define _FS_CEPH_OSDMAP_H
3
4#include <linux/rbtree.h>
5#include "types.h"
6#include "ceph_fs.h"
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07007#include <linux/crush/crush.h>
Sage Weilf24e9982009-10-06 11:31:10 -07008
9/*
10 * The osd map describes the current membership of the osd cluster and
11 * specifies the mapping of objects to placement groups and placement
12 * groups to (sets of) osds. That is, it completely specifies the
13 * (desired) distribution of all data objects in the system at some
14 * point in time.
15 *
16 * Each map version is identified by an epoch, which increases monotonically.
17 *
18 * The map can be updated either via an incremental map (diff) describing
19 * the change between two successive epochs, or as a fully encoded map.
20 */
21struct ceph_pg_pool_info {
Sage Weil4fc51be2010-02-16 15:55:03 -080022 struct rb_node node;
23 int id;
Sage Weilf24e9982009-10-06 11:31:10 -070024 struct ceph_pg_pool v;
25 int pg_num_mask, pgp_num_mask, lpg_num_mask, lpgp_num_mask;
Sage Weil2844a762010-04-09 15:46:42 -070026 char *name;
Sage Weilf24e9982009-10-06 11:31:10 -070027};
28
29struct ceph_pg_mapping {
30 struct rb_node node;
Sage Weil51042122009-11-04 11:39:12 -080031 struct ceph_pg pgid;
Sage Weilf24e9982009-10-06 11:31:10 -070032 int len;
33 int osds[];
34};
35
36struct ceph_osdmap {
37 struct ceph_fsid fsid;
38 u32 epoch;
39 u32 mkfs_epoch;
40 struct ceph_timespec created, modified;
41
42 u32 flags; /* CEPH_OSDMAP_* */
43
44 u32 max_osd; /* size of osd_state, _offload, _addr arrays */
45 u8 *osd_state; /* CEPH_OSD_* */
46 u32 *osd_weight; /* 0 = failed, 0x10000 = 100% normal */
47 struct ceph_entity_addr *osd_addr;
48
49 struct rb_root pg_temp;
Sage Weil4fc51be2010-02-16 15:55:03 -080050 struct rb_root pg_pools;
51 u32 pool_max;
Sage Weilf24e9982009-10-06 11:31:10 -070052
53 /* the CRUSH map specifies the mapping of placement groups to
54 * the list of osds that store+replicate them. */
55 struct crush_map *crush;
56};
57
58/*
59 * file layout helpers
60 */
61#define ceph_file_layout_su(l) ((__s32)le32_to_cpu((l).fl_stripe_unit))
62#define ceph_file_layout_stripe_count(l) \
63 ((__s32)le32_to_cpu((l).fl_stripe_count))
64#define ceph_file_layout_object_size(l) ((__s32)le32_to_cpu((l).fl_object_size))
65#define ceph_file_layout_cas_hash(l) ((__s32)le32_to_cpu((l).fl_cas_hash))
66#define ceph_file_layout_object_su(l) \
67 ((__s32)le32_to_cpu((l).fl_object_stripe_unit))
68#define ceph_file_layout_pg_preferred(l) \
69 ((__s32)le32_to_cpu((l).fl_pg_preferred))
70#define ceph_file_layout_pg_pool(l) \
71 ((__s32)le32_to_cpu((l).fl_pg_pool))
72
73static inline unsigned ceph_file_layout_stripe_width(struct ceph_file_layout *l)
74{
75 return le32_to_cpu(l->fl_stripe_unit) *
76 le32_to_cpu(l->fl_stripe_count);
77}
78
79/* "period" == bytes before i start on a new set of objects */
80static inline unsigned ceph_file_layout_period(struct ceph_file_layout *l)
81{
82 return le32_to_cpu(l->fl_object_size) *
83 le32_to_cpu(l->fl_stripe_count);
84}
85
86
87static inline int ceph_osd_is_up(struct ceph_osdmap *map, int osd)
88{
89 return (osd < map->max_osd) && (map->osd_state[osd] & CEPH_OSD_UP);
90}
91
92static inline bool ceph_osdmap_flag(struct ceph_osdmap *map, int flag)
93{
94 return map && (map->flags & flag);
95}
96
97extern char *ceph_osdmap_state_str(char *str, int len, int state);
98
99static inline struct ceph_entity_addr *ceph_osd_addr(struct ceph_osdmap *map,
100 int osd)
101{
102 if (osd >= map->max_osd)
103 return NULL;
104 return &map->osd_addr[osd];
105}
106
107extern struct ceph_osdmap *osdmap_decode(void **p, void *end);
108extern struct ceph_osdmap *osdmap_apply_incremental(void **p, void *end,
109 struct ceph_osdmap *map,
110 struct ceph_messenger *msgr);
111extern void ceph_osdmap_destroy(struct ceph_osdmap *map);
112
113/* calculate mapping of a file extent to an object */
114extern void ceph_calc_file_object_mapping(struct ceph_file_layout *layout,
115 u64 off, u64 *plen,
116 u64 *bno, u64 *oxoff, u64 *oxlen);
117
118/* calculate mapping of object to a placement group */
119extern int ceph_calc_object_layout(struct ceph_object_layout *ol,
120 const char *oid,
121 struct ceph_file_layout *fl,
122 struct ceph_osdmap *osdmap);
Sage Weild85b7052010-05-10 10:24:48 -0700123extern int ceph_calc_pg_acting(struct ceph_osdmap *osdmap, struct ceph_pg pgid,
124 int *acting);
Sage Weil51042122009-11-04 11:39:12 -0800125extern int ceph_calc_pg_primary(struct ceph_osdmap *osdmap,
126 struct ceph_pg pgid);
Sage Weilf24e9982009-10-06 11:31:10 -0700127
Yehuda Sadeh7669a2c2010-05-17 12:31:35 -0700128extern int ceph_pg_poolid_by_name(struct ceph_osdmap *map, const char *name);
129
Sage Weilf24e9982009-10-06 11:31:10 -0700130#endif