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/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 */
22struct ceph_pg {
23	uint64_t pool;
24	uint32_t seed;
25};
26
27#define CEPH_SPG_NOSHARD	-1
28
29struct ceph_spg {
30	struct ceph_pg pgid;
31	s8 shard;
32};
33
34int ceph_pg_compare(const struct ceph_pg *lhs, const struct ceph_pg *rhs);
35int 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#define CEPH_POOL_FLAG_FULL_QUOTA	(1ULL << 10) /* pool ran out of quota,
41							will set FULL too */
42#define CEPH_POOL_FLAG_NEARFULL		(1ULL << 11) /* pool is nearfull */
43
44struct ceph_pg_pool_info {
45	struct rb_node node;
46	s64 id;
47	u8 type; /* CEPH_POOL_TYPE_* */
48	u8 size;
49	u8 min_size;
50	u8 crush_ruleset;
51	u8 object_hash;
52	u32 last_force_request_resend;
53	u32 pg_num, pgp_num;
54	int pg_num_mask, pgp_num_mask;
55	s64 read_tier;
56	s64 write_tier; /* wins for read+write ops */
57	u64 flags; /* CEPH_POOL_FLAG_* */
58	char *name;
59
60	bool was_full;  /* for handle_one_map() */
61};
62
63static inline bool ceph_can_shift_osds(struct ceph_pg_pool_info *pool)
64{
65	switch (pool->type) {
66	case CEPH_POOL_TYPE_REP:
67		return true;
68	case CEPH_POOL_TYPE_EC:
69		return false;
70	default:
71		BUG();
72	}
73}
74
75struct ceph_object_locator {
76	s64 pool;
77	struct ceph_string *pool_ns;
78};
79
80static inline void ceph_oloc_init(struct ceph_object_locator *oloc)
81{
82	oloc->pool = -1;
83	oloc->pool_ns = NULL;
84}
85
86static inline bool ceph_oloc_empty(const struct ceph_object_locator *oloc)
87{
88	return oloc->pool == -1;
89}
90
91void ceph_oloc_copy(struct ceph_object_locator *dest,
92		    const struct ceph_object_locator *src);
93void ceph_oloc_destroy(struct ceph_object_locator *oloc);
94
95/*
96 * 51-char inline_name is long enough for all cephfs and all but one
97 * rbd requests: <imgname> in "<imgname>.rbd"/"rbd_id.<imgname>" can be
98 * arbitrarily long (~PAGE_SIZE).  It's done once during rbd map; all
99 * other rbd requests fit into inline_name.
100 *
101 * Makes ceph_object_id 64 bytes on 64-bit.
102 */
103#define CEPH_OID_INLINE_LEN 52
104
105/*
106 * Both inline and external buffers have space for a NUL-terminator,
107 * which is carried around.  It's not required though - RADOS object
108 * names don't have to be NUL-terminated and may contain NULs.
109 */
110struct ceph_object_id {
111	char *name;
112	char inline_name[CEPH_OID_INLINE_LEN];
113	int name_len;
114};
115
116#define __CEPH_OID_INITIALIZER(oid) { .name = (oid).inline_name }
117
118#define CEPH_DEFINE_OID_ONSTACK(oid)				\
119	struct ceph_object_id oid = __CEPH_OID_INITIALIZER(oid)
120
121static inline void ceph_oid_init(struct ceph_object_id *oid)
122{
123	*oid = (struct ceph_object_id) __CEPH_OID_INITIALIZER(*oid);
124}
125
126static inline bool ceph_oid_empty(const struct ceph_object_id *oid)
127{
128	return oid->name == oid->inline_name && !oid->name_len;
129}
130
131void ceph_oid_copy(struct ceph_object_id *dest,
132		   const struct ceph_object_id *src);
133__printf(2, 3)
134void ceph_oid_printf(struct ceph_object_id *oid, const char *fmt, ...);
135__printf(3, 4)
136int ceph_oid_aprintf(struct ceph_object_id *oid, gfp_t gfp,
137		     const char *fmt, ...);
138void ceph_oid_destroy(struct ceph_object_id *oid);
139
140struct workspace_manager {
141	struct list_head idle_ws;
142	spinlock_t ws_lock;
143	/* Number of free workspaces */
144	int free_ws;
145	/* Total number of allocated workspaces */
146	atomic_t total_ws;
147	/* Waiters for a free workspace */
148	wait_queue_head_t ws_wait;
149};
150
151struct ceph_pg_mapping {
152	struct rb_node node;
153	struct ceph_pg pgid;
154
155	union {
156		struct {
157			int len;
158			int osds[];
159		} pg_temp, pg_upmap;
160		struct {
161			int osd;
162		} primary_temp;
163		struct {
164			int len;
165			int from_to[][2];
166		} pg_upmap_items;
167	};
168};
169
170struct ceph_osdmap {
171	struct ceph_fsid fsid;
172	u32 epoch;
173	struct ceph_timespec created, modified;
174
175	u32 flags;         /* CEPH_OSDMAP_* */
176
177	u32 max_osd;       /* size of osd_state, _offload, _addr arrays */
178	u32 *osd_state;    /* CEPH_OSD_* */
179	u32 *osd_weight;   /* 0 = failed, 0x10000 = 100% normal */
180	struct ceph_entity_addr *osd_addr;
181
182	struct rb_root pg_temp;
183	struct rb_root primary_temp;
184
185	/* remap (post-CRUSH, pre-up) */
186	struct rb_root pg_upmap;	/* PG := raw set */
187	struct rb_root pg_upmap_items;	/* from -> to within raw set */
188
189	u32 *osd_primary_affinity;
190
191	struct rb_root pg_pools;
192	u32 pool_max;
193
194	/* the CRUSH map specifies the mapping of placement groups to
195	 * the list of osds that store+replicate them. */
196	struct crush_map *crush;
197
198	struct workspace_manager crush_wsm;
199};
200
201static inline bool ceph_osd_exists(struct ceph_osdmap *map, int osd)
202{
203	return osd >= 0 && osd < map->max_osd &&
204	       (map->osd_state[osd] & CEPH_OSD_EXISTS);
205}
206
207static inline bool ceph_osd_is_up(struct ceph_osdmap *map, int osd)
208{
209	return ceph_osd_exists(map, osd) &&
210	       (map->osd_state[osd] & CEPH_OSD_UP);
211}
212
213static inline bool ceph_osd_is_down(struct ceph_osdmap *map, int osd)
214{
215	return !ceph_osd_is_up(map, osd);
216}
217
218char *ceph_osdmap_state_str(char *str, int len, u32 state);
219extern u32 ceph_get_primary_affinity(struct ceph_osdmap *map, int osd);
220
221static inline struct ceph_entity_addr *ceph_osd_addr(struct ceph_osdmap *map,
222						     int osd)
223{
224	if (osd >= map->max_osd)
225		return NULL;
226	return &map->osd_addr[osd];
227}
228
229#define CEPH_PGID_ENCODING_LEN		(1 + 8 + 4 + 4)
230
231static inline int ceph_decode_pgid(void **p, void *end, struct ceph_pg *pgid)
232{
233	__u8 version;
234
235	if (!ceph_has_room(p, end, CEPH_PGID_ENCODING_LEN)) {
236		pr_warn("incomplete pg encoding\n");
237		return -EINVAL;
238	}
239	version = ceph_decode_8(p);
240	if (version > 1) {
241		pr_warn("do not understand pg encoding %d > 1\n",
242			(int)version);
243		return -EINVAL;
244	}
245
246	pgid->pool = ceph_decode_64(p);
247	pgid->seed = ceph_decode_32(p);
248	*p += 4;	/* skip deprecated preferred value */
249
250	return 0;
251}
252
253struct ceph_osdmap *ceph_osdmap_alloc(void);
254struct ceph_osdmap *ceph_osdmap_decode(void **p, void *end, bool msgr2);
255struct ceph_osdmap *osdmap_apply_incremental(void **p, void *end, bool msgr2,
256					     struct ceph_osdmap *map);
257extern void ceph_osdmap_destroy(struct ceph_osdmap *map);
258
259struct ceph_osds {
260	int osds[CEPH_PG_MAX_SIZE];
261	int size;
262	int primary; /* id, NOT index */
263};
264
265static inline void ceph_osds_init(struct ceph_osds *set)
266{
267	set->size = 0;
268	set->primary = -1;
269}
270
271void ceph_osds_copy(struct ceph_osds *dest, const struct ceph_osds *src);
272
273bool ceph_pg_is_split(const struct ceph_pg *pgid, u32 old_pg_num,
274		      u32 new_pg_num);
275bool ceph_is_new_interval(const struct ceph_osds *old_acting,
276			  const struct ceph_osds *new_acting,
277			  const struct ceph_osds *old_up,
278			  const struct ceph_osds *new_up,
279			  int old_size,
280			  int new_size,
281			  int old_min_size,
282			  int new_min_size,
283			  u32 old_pg_num,
284			  u32 new_pg_num,
285			  bool old_sort_bitwise,
286			  bool new_sort_bitwise,
287			  bool old_recovery_deletes,
288			  bool new_recovery_deletes,
289			  const struct ceph_pg *pgid);
290bool ceph_osds_changed(const struct ceph_osds *old_acting,
291		       const struct ceph_osds *new_acting,
292		       bool any_change);
293
294void __ceph_object_locator_to_pg(struct ceph_pg_pool_info *pi,
295				 const struct ceph_object_id *oid,
296				 const struct ceph_object_locator *oloc,
297				 struct ceph_pg *raw_pgid);
298int ceph_object_locator_to_pg(struct ceph_osdmap *osdmap,
299			      const struct ceph_object_id *oid,
300			      const struct ceph_object_locator *oloc,
301			      struct ceph_pg *raw_pgid);
302
303void ceph_pg_to_up_acting_osds(struct ceph_osdmap *osdmap,
304			       struct ceph_pg_pool_info *pi,
305			       const struct ceph_pg *raw_pgid,
306			       struct ceph_osds *up,
307			       struct ceph_osds *acting);
308bool ceph_pg_to_primary_shard(struct ceph_osdmap *osdmap,
309			      struct ceph_pg_pool_info *pi,
310			      const struct ceph_pg *raw_pgid,
311			      struct ceph_spg *spgid);
312int ceph_pg_to_acting_primary(struct ceph_osdmap *osdmap,
313			      const struct ceph_pg *raw_pgid);
314
315struct crush_loc {
316	char *cl_type_name;
317	char *cl_name;
318};
319
320struct crush_loc_node {
321	struct rb_node cl_node;
322	struct crush_loc cl_loc;  /* pointers into cl_data */
323	char cl_data[];
324};
325
326int ceph_parse_crush_location(char *crush_location, struct rb_root *locs);
327int ceph_compare_crush_locs(struct rb_root *locs1, struct rb_root *locs2);
328void ceph_clear_crush_locs(struct rb_root *locs);
329
330int ceph_get_crush_locality(struct ceph_osdmap *osdmap, int id,
331			    struct rb_root *locs);
332
333extern struct ceph_pg_pool_info *ceph_pg_pool_by_id(struct ceph_osdmap *map,
334						    u64 id);
335extern const char *ceph_pg_pool_name_by_id(struct ceph_osdmap *map, u64 id);
336extern int ceph_pg_poolid_by_name(struct ceph_osdmap *map, const char *name);
337u64 ceph_pg_pool_flags(struct ceph_osdmap *map, u64 id);
338
339#endif
340