2 #include <linux/ceph/ceph_debug.h>
4 #include <linux/module.h>
5 #include <linux/slab.h>
8 #include <linux/ceph/libceph.h>
9 #include <linux/ceph/osdmap.h>
10 #include <linux/ceph/decode.h>
11 #include <linux/crush/hash.h>
12 #include <linux/crush/mapper.h>
14 char *ceph_osdmap_state_str(char *str, int len, int state)
19 if ((state & CEPH_OSD_EXISTS) && (state & CEPH_OSD_UP))
20 snprintf(str, len, "exists, up");
21 else if (state & CEPH_OSD_EXISTS)
22 snprintf(str, len, "exists");
23 else if (state & CEPH_OSD_UP)
24 snprintf(str, len, "up");
26 snprintf(str, len, "doesn't exist");
33 static int calc_bits_of(unsigned int t)
44 * the foo_mask is the smallest value 2^n-1 that is >= foo.
46 static void calc_pg_masks(struct ceph_pg_pool_info *pi)
48 pi->pg_num_mask = (1 << calc_bits_of(pi->pg_num-1)) - 1;
49 pi->pgp_num_mask = (1 << calc_bits_of(pi->pgp_num-1)) - 1;
55 static int crush_decode_uniform_bucket(void **p, void *end,
56 struct crush_bucket_uniform *b)
58 dout("crush_decode_uniform_bucket %p to %p\n", *p, end);
59 ceph_decode_need(p, end, (1+b->h.size) * sizeof(u32), bad);
60 b->item_weight = ceph_decode_32(p);
66 static int crush_decode_list_bucket(void **p, void *end,
67 struct crush_bucket_list *b)
70 dout("crush_decode_list_bucket %p to %p\n", *p, end);
71 b->item_weights = kcalloc(b->h.size, sizeof(u32), GFP_NOFS);
72 if (b->item_weights == NULL)
74 b->sum_weights = kcalloc(b->h.size, sizeof(u32), GFP_NOFS);
75 if (b->sum_weights == NULL)
77 ceph_decode_need(p, end, 2 * b->h.size * sizeof(u32), bad);
78 for (j = 0; j < b->h.size; j++) {
79 b->item_weights[j] = ceph_decode_32(p);
80 b->sum_weights[j] = ceph_decode_32(p);
87 static int crush_decode_tree_bucket(void **p, void *end,
88 struct crush_bucket_tree *b)
91 dout("crush_decode_tree_bucket %p to %p\n", *p, end);
92 ceph_decode_32_safe(p, end, b->num_nodes, bad);
93 b->node_weights = kcalloc(b->num_nodes, sizeof(u32), GFP_NOFS);
94 if (b->node_weights == NULL)
96 ceph_decode_need(p, end, b->num_nodes * sizeof(u32), bad);
97 for (j = 0; j < b->num_nodes; j++)
98 b->node_weights[j] = ceph_decode_32(p);
104 static int crush_decode_straw_bucket(void **p, void *end,
105 struct crush_bucket_straw *b)
108 dout("crush_decode_straw_bucket %p to %p\n", *p, end);
109 b->item_weights = kcalloc(b->h.size, sizeof(u32), GFP_NOFS);
110 if (b->item_weights == NULL)
112 b->straws = kcalloc(b->h.size, sizeof(u32), GFP_NOFS);
113 if (b->straws == NULL)
115 ceph_decode_need(p, end, 2 * b->h.size * sizeof(u32), bad);
116 for (j = 0; j < b->h.size; j++) {
117 b->item_weights[j] = ceph_decode_32(p);
118 b->straws[j] = ceph_decode_32(p);
125 static int skip_name_map(void **p, void *end)
128 ceph_decode_32_safe(p, end, len ,bad);
132 ceph_decode_32_safe(p, end, strlen, bad);
140 static struct crush_map *crush_decode(void *pbyval, void *end)
146 void *start = pbyval;
150 dout("crush_decode %p to %p len %d\n", *p, end, (int)(end - *p));
152 c = kzalloc(sizeof(*c), GFP_NOFS);
154 return ERR_PTR(-ENOMEM);
156 /* set tunables to default values */
157 c->choose_local_tries = 2;
158 c->choose_local_fallback_tries = 5;
159 c->choose_total_tries = 19;
160 c->chooseleaf_descend_once = 0;
162 ceph_decode_need(p, end, 4*sizeof(u32), bad);
163 magic = ceph_decode_32(p);
164 if (magic != CRUSH_MAGIC) {
165 pr_err("crush_decode magic %x != current %x\n",
166 (unsigned int)magic, (unsigned int)CRUSH_MAGIC);
169 c->max_buckets = ceph_decode_32(p);
170 c->max_rules = ceph_decode_32(p);
171 c->max_devices = ceph_decode_32(p);
173 c->buckets = kcalloc(c->max_buckets, sizeof(*c->buckets), GFP_NOFS);
174 if (c->buckets == NULL)
176 c->rules = kcalloc(c->max_rules, sizeof(*c->rules), GFP_NOFS);
177 if (c->rules == NULL)
181 for (i = 0; i < c->max_buckets; i++) {
184 struct crush_bucket *b;
186 ceph_decode_32_safe(p, end, alg, bad);
188 c->buckets[i] = NULL;
191 dout("crush_decode bucket %d off %x %p to %p\n",
192 i, (int)(*p-start), *p, end);
195 case CRUSH_BUCKET_UNIFORM:
196 size = sizeof(struct crush_bucket_uniform);
198 case CRUSH_BUCKET_LIST:
199 size = sizeof(struct crush_bucket_list);
201 case CRUSH_BUCKET_TREE:
202 size = sizeof(struct crush_bucket_tree);
204 case CRUSH_BUCKET_STRAW:
205 size = sizeof(struct crush_bucket_straw);
212 b = c->buckets[i] = kzalloc(size, GFP_NOFS);
216 ceph_decode_need(p, end, 4*sizeof(u32), bad);
217 b->id = ceph_decode_32(p);
218 b->type = ceph_decode_16(p);
219 b->alg = ceph_decode_8(p);
220 b->hash = ceph_decode_8(p);
221 b->weight = ceph_decode_32(p);
222 b->size = ceph_decode_32(p);
224 dout("crush_decode bucket size %d off %x %p to %p\n",
225 b->size, (int)(*p-start), *p, end);
227 b->items = kcalloc(b->size, sizeof(__s32), GFP_NOFS);
228 if (b->items == NULL)
230 b->perm = kcalloc(b->size, sizeof(u32), GFP_NOFS);
235 ceph_decode_need(p, end, b->size*sizeof(u32), bad);
236 for (j = 0; j < b->size; j++)
237 b->items[j] = ceph_decode_32(p);
240 case CRUSH_BUCKET_UNIFORM:
241 err = crush_decode_uniform_bucket(p, end,
242 (struct crush_bucket_uniform *)b);
246 case CRUSH_BUCKET_LIST:
247 err = crush_decode_list_bucket(p, end,
248 (struct crush_bucket_list *)b);
252 case CRUSH_BUCKET_TREE:
253 err = crush_decode_tree_bucket(p, end,
254 (struct crush_bucket_tree *)b);
258 case CRUSH_BUCKET_STRAW:
259 err = crush_decode_straw_bucket(p, end,
260 (struct crush_bucket_straw *)b);
268 dout("rule vec is %p\n", c->rules);
269 for (i = 0; i < c->max_rules; i++) {
271 struct crush_rule *r;
273 ceph_decode_32_safe(p, end, yes, bad);
275 dout("crush_decode NO rule %d off %x %p to %p\n",
276 i, (int)(*p-start), *p, end);
281 dout("crush_decode rule %d off %x %p to %p\n",
282 i, (int)(*p-start), *p, end);
285 ceph_decode_32_safe(p, end, yes, bad);
286 #if BITS_PER_LONG == 32
288 if (yes > (ULONG_MAX - sizeof(*r))
289 / sizeof(struct crush_rule_step))
292 r = c->rules[i] = kmalloc(sizeof(*r) +
293 yes*sizeof(struct crush_rule_step),
297 dout(" rule %d is at %p\n", i, r);
299 ceph_decode_copy_safe(p, end, &r->mask, 4, bad); /* 4 u8's */
300 ceph_decode_need(p, end, r->len*3*sizeof(u32), bad);
301 for (j = 0; j < r->len; j++) {
302 r->steps[j].op = ceph_decode_32(p);
303 r->steps[j].arg1 = ceph_decode_32(p);
304 r->steps[j].arg2 = ceph_decode_32(p);
308 /* ignore trailing name maps. */
309 for (num_name_maps = 0; num_name_maps < 3; num_name_maps++) {
310 err = skip_name_map(p, end);
316 ceph_decode_need(p, end, 3*sizeof(u32), done);
317 c->choose_local_tries = ceph_decode_32(p);
318 c->choose_local_fallback_tries = ceph_decode_32(p);
319 c->choose_total_tries = ceph_decode_32(p);
320 dout("crush decode tunable choose_local_tries = %d",
321 c->choose_local_tries);
322 dout("crush decode tunable choose_local_fallback_tries = %d",
323 c->choose_local_fallback_tries);
324 dout("crush decode tunable choose_total_tries = %d",
325 c->choose_total_tries);
327 ceph_decode_need(p, end, sizeof(u32), done);
328 c->chooseleaf_descend_once = ceph_decode_32(p);
329 dout("crush decode tunable chooseleaf_descend_once = %d",
330 c->chooseleaf_descend_once);
332 ceph_decode_need(p, end, sizeof(u8), done);
333 c->chooseleaf_vary_r = ceph_decode_8(p);
334 dout("crush decode tunable chooseleaf_vary_r = %d",
335 c->chooseleaf_vary_r);
338 dout("crush_decode success\n");
344 dout("crush_decode fail %d\n", err);
350 * rbtree of pg_mapping for handling pg_temp (explicit mapping of pgid
351 * to a set of osds) and primary_temp (explicit primary setting)
353 static int pgid_cmp(struct ceph_pg l, struct ceph_pg r)
366 static int __insert_pg_mapping(struct ceph_pg_mapping *new,
367 struct rb_root *root)
369 struct rb_node **p = &root->rb_node;
370 struct rb_node *parent = NULL;
371 struct ceph_pg_mapping *pg = NULL;
374 dout("__insert_pg_mapping %llx %p\n", *(u64 *)&new->pgid, new);
377 pg = rb_entry(parent, struct ceph_pg_mapping, node);
378 c = pgid_cmp(new->pgid, pg->pgid);
387 rb_link_node(&new->node, parent, p);
388 rb_insert_color(&new->node, root);
392 static struct ceph_pg_mapping *__lookup_pg_mapping(struct rb_root *root,
395 struct rb_node *n = root->rb_node;
396 struct ceph_pg_mapping *pg;
400 pg = rb_entry(n, struct ceph_pg_mapping, node);
401 c = pgid_cmp(pgid, pg->pgid);
407 dout("__lookup_pg_mapping %lld.%x got %p\n",
408 pgid.pool, pgid.seed, pg);
415 static int __remove_pg_mapping(struct rb_root *root, struct ceph_pg pgid)
417 struct ceph_pg_mapping *pg = __lookup_pg_mapping(root, pgid);
420 dout("__remove_pg_mapping %lld.%x %p\n", pgid.pool, pgid.seed,
422 rb_erase(&pg->node, root);
426 dout("__remove_pg_mapping %lld.%x dne\n", pgid.pool, pgid.seed);
431 * rbtree of pg pool info
433 static int __insert_pg_pool(struct rb_root *root, struct ceph_pg_pool_info *new)
435 struct rb_node **p = &root->rb_node;
436 struct rb_node *parent = NULL;
437 struct ceph_pg_pool_info *pi = NULL;
441 pi = rb_entry(parent, struct ceph_pg_pool_info, node);
442 if (new->id < pi->id)
444 else if (new->id > pi->id)
450 rb_link_node(&new->node, parent, p);
451 rb_insert_color(&new->node, root);
455 static struct ceph_pg_pool_info *__lookup_pg_pool(struct rb_root *root, u64 id)
457 struct ceph_pg_pool_info *pi;
458 struct rb_node *n = root->rb_node;
461 pi = rb_entry(n, struct ceph_pg_pool_info, node);
464 else if (id > pi->id)
472 struct ceph_pg_pool_info *ceph_pg_pool_by_id(struct ceph_osdmap *map, u64 id)
474 return __lookup_pg_pool(&map->pg_pools, id);
477 const char *ceph_pg_pool_name_by_id(struct ceph_osdmap *map, u64 id)
479 struct ceph_pg_pool_info *pi;
481 if (id == CEPH_NOPOOL)
484 if (WARN_ON_ONCE(id > (u64) INT_MAX))
487 pi = __lookup_pg_pool(&map->pg_pools, (int) id);
489 return pi ? pi->name : NULL;
491 EXPORT_SYMBOL(ceph_pg_pool_name_by_id);
493 int ceph_pg_poolid_by_name(struct ceph_osdmap *map, const char *name)
497 for (rbp = rb_first(&map->pg_pools); rbp; rbp = rb_next(rbp)) {
498 struct ceph_pg_pool_info *pi =
499 rb_entry(rbp, struct ceph_pg_pool_info, node);
500 if (pi->name && strcmp(pi->name, name) == 0)
505 EXPORT_SYMBOL(ceph_pg_poolid_by_name);
507 static void __remove_pg_pool(struct rb_root *root, struct ceph_pg_pool_info *pi)
509 rb_erase(&pi->node, root);
514 static int decode_pool(void **p, void *end, struct ceph_pg_pool_info *pi)
520 ceph_decode_need(p, end, 2 + 4, bad);
521 ev = ceph_decode_8(p); /* encoding version */
522 cv = ceph_decode_8(p); /* compat version */
524 pr_warn("got v %d < 5 cv %d of ceph_pg_pool\n", ev, cv);
528 pr_warn("got v %d cv %d > 9 of ceph_pg_pool\n", ev, cv);
531 len = ceph_decode_32(p);
532 ceph_decode_need(p, end, len, bad);
535 pi->type = ceph_decode_8(p);
536 pi->size = ceph_decode_8(p);
537 pi->crush_ruleset = ceph_decode_8(p);
538 pi->object_hash = ceph_decode_8(p);
540 pi->pg_num = ceph_decode_32(p);
541 pi->pgp_num = ceph_decode_32(p);
543 *p += 4 + 4; /* skip lpg* */
544 *p += 4; /* skip last_change */
545 *p += 8 + 4; /* skip snap_seq, snap_epoch */
548 num = ceph_decode_32(p);
550 *p += 8; /* snapid key */
551 *p += 1 + 1; /* versions */
552 len = ceph_decode_32(p);
556 /* skip removed_snaps */
557 num = ceph_decode_32(p);
560 *p += 8; /* skip auid */
561 pi->flags = ceph_decode_64(p);
562 *p += 4; /* skip crash_replay_interval */
565 *p += 1; /* skip min_size */
568 *p += 8 + 8; /* skip quota_max_* */
572 num = ceph_decode_32(p);
575 *p += 8; /* skip tier_of */
576 *p += 1; /* skip cache_mode */
578 pi->read_tier = ceph_decode_64(p);
579 pi->write_tier = ceph_decode_64(p);
585 /* ignore the rest */
595 static int decode_pool_names(void **p, void *end, struct ceph_osdmap *map)
597 struct ceph_pg_pool_info *pi;
601 ceph_decode_32_safe(p, end, num, bad);
602 dout(" %d pool names\n", num);
604 ceph_decode_64_safe(p, end, pool, bad);
605 ceph_decode_32_safe(p, end, len, bad);
606 dout(" pool %llu len %d\n", pool, len);
607 ceph_decode_need(p, end, len, bad);
608 pi = __lookup_pg_pool(&map->pg_pools, pool);
610 char *name = kstrndup(*p, len, GFP_NOFS);
616 dout(" name is %s\n", pi->name);
629 void ceph_osdmap_destroy(struct ceph_osdmap *map)
631 dout("osdmap_destroy %p\n", map);
633 crush_destroy(map->crush);
634 while (!RB_EMPTY_ROOT(&map->pg_temp)) {
635 struct ceph_pg_mapping *pg =
636 rb_entry(rb_first(&map->pg_temp),
637 struct ceph_pg_mapping, node);
638 rb_erase(&pg->node, &map->pg_temp);
641 while (!RB_EMPTY_ROOT(&map->primary_temp)) {
642 struct ceph_pg_mapping *pg =
643 rb_entry(rb_first(&map->primary_temp),
644 struct ceph_pg_mapping, node);
645 rb_erase(&pg->node, &map->primary_temp);
648 while (!RB_EMPTY_ROOT(&map->pg_pools)) {
649 struct ceph_pg_pool_info *pi =
650 rb_entry(rb_first(&map->pg_pools),
651 struct ceph_pg_pool_info, node);
652 __remove_pg_pool(&map->pg_pools, pi);
654 kfree(map->osd_state);
655 kfree(map->osd_weight);
656 kfree(map->osd_addr);
657 kfree(map->osd_primary_affinity);
662 * Adjust max_osd value, (re)allocate arrays.
664 * The new elements are properly initialized.
666 static int osdmap_set_max_osd(struct ceph_osdmap *map, int max)
670 struct ceph_entity_addr *addr;
673 state = krealloc(map->osd_state, max*sizeof(*state), GFP_NOFS);
676 map->osd_state = state;
678 weight = krealloc(map->osd_weight, max*sizeof(*weight), GFP_NOFS);
681 map->osd_weight = weight;
683 addr = krealloc(map->osd_addr, max*sizeof(*addr), GFP_NOFS);
686 map->osd_addr = addr;
688 for (i = map->max_osd; i < max; i++) {
689 map->osd_state[i] = 0;
690 map->osd_weight[i] = CEPH_OSD_OUT;
691 memset(map->osd_addr + i, 0, sizeof(*map->osd_addr));
694 if (map->osd_primary_affinity) {
697 affinity = krealloc(map->osd_primary_affinity,
698 max*sizeof(*affinity), GFP_NOFS);
701 map->osd_primary_affinity = affinity;
703 for (i = map->max_osd; i < max; i++)
704 map->osd_primary_affinity[i] =
705 CEPH_OSD_DEFAULT_PRIMARY_AFFINITY;
713 #define OSDMAP_WRAPPER_COMPAT_VER 7
714 #define OSDMAP_CLIENT_DATA_COMPAT_VER 1
717 * Return 0 or error. On success, *v is set to 0 for old (v6) osdmaps,
718 * to struct_v of the client_data section for new (v7 and above)
721 static int get_osdmap_client_data_v(void **p, void *end,
722 const char *prefix, u8 *v)
726 ceph_decode_8_safe(p, end, struct_v, e_inval);
730 ceph_decode_8_safe(p, end, struct_compat, e_inval);
731 if (struct_compat > OSDMAP_WRAPPER_COMPAT_VER) {
732 pr_warn("got v %d cv %d > %d of %s ceph_osdmap\n",
733 struct_v, struct_compat,
734 OSDMAP_WRAPPER_COMPAT_VER, prefix);
737 *p += 4; /* ignore wrapper struct_len */
739 ceph_decode_8_safe(p, end, struct_v, e_inval);
740 ceph_decode_8_safe(p, end, struct_compat, e_inval);
741 if (struct_compat > OSDMAP_CLIENT_DATA_COMPAT_VER) {
742 pr_warn("got v %d cv %d > %d of %s ceph_osdmap client data\n",
743 struct_v, struct_compat,
744 OSDMAP_CLIENT_DATA_COMPAT_VER, prefix);
747 *p += 4; /* ignore client data struct_len */
752 ceph_decode_16_safe(p, end, version, e_inval);
754 pr_warn("got v %d < 6 of %s ceph_osdmap\n",
759 /* old osdmap enconding */
770 static int __decode_pools(void **p, void *end, struct ceph_osdmap *map,
775 ceph_decode_32_safe(p, end, n, e_inval);
777 struct ceph_pg_pool_info *pi;
781 ceph_decode_64_safe(p, end, pool, e_inval);
783 pi = __lookup_pg_pool(&map->pg_pools, pool);
784 if (!incremental || !pi) {
785 pi = kzalloc(sizeof(*pi), GFP_NOFS);
791 ret = __insert_pg_pool(&map->pg_pools, pi);
798 ret = decode_pool(p, end, pi);
809 static int decode_pools(void **p, void *end, struct ceph_osdmap *map)
811 return __decode_pools(p, end, map, false);
814 static int decode_new_pools(void **p, void *end, struct ceph_osdmap *map)
816 return __decode_pools(p, end, map, true);
819 static int __decode_pg_temp(void **p, void *end, struct ceph_osdmap *map,
824 ceph_decode_32_safe(p, end, n, e_inval);
830 ret = ceph_decode_pgid(p, end, &pgid);
834 ceph_decode_32_safe(p, end, len, e_inval);
836 ret = __remove_pg_mapping(&map->pg_temp, pgid);
837 BUG_ON(!incremental && ret != -ENOENT);
839 if (!incremental || len > 0) {
840 struct ceph_pg_mapping *pg;
842 ceph_decode_need(p, end, len*sizeof(u32), e_inval);
844 if (len > (UINT_MAX - sizeof(*pg)) / sizeof(u32))
847 pg = kzalloc(sizeof(*pg) + len*sizeof(u32), GFP_NOFS);
852 pg->pg_temp.len = len;
853 for (i = 0; i < len; i++)
854 pg->pg_temp.osds[i] = ceph_decode_32(p);
856 ret = __insert_pg_mapping(pg, &map->pg_temp);
870 static int decode_pg_temp(void **p, void *end, struct ceph_osdmap *map)
872 return __decode_pg_temp(p, end, map, false);
875 static int decode_new_pg_temp(void **p, void *end, struct ceph_osdmap *map)
877 return __decode_pg_temp(p, end, map, true);
880 static int __decode_primary_temp(void **p, void *end, struct ceph_osdmap *map,
885 ceph_decode_32_safe(p, end, n, e_inval);
891 ret = ceph_decode_pgid(p, end, &pgid);
895 ceph_decode_32_safe(p, end, osd, e_inval);
897 ret = __remove_pg_mapping(&map->primary_temp, pgid);
898 BUG_ON(!incremental && ret != -ENOENT);
900 if (!incremental || osd != (u32)-1) {
901 struct ceph_pg_mapping *pg;
903 pg = kzalloc(sizeof(*pg), GFP_NOFS);
908 pg->primary_temp.osd = osd;
910 ret = __insert_pg_mapping(pg, &map->primary_temp);
924 static int decode_primary_temp(void **p, void *end, struct ceph_osdmap *map)
926 return __decode_primary_temp(p, end, map, false);
929 static int decode_new_primary_temp(void **p, void *end,
930 struct ceph_osdmap *map)
932 return __decode_primary_temp(p, end, map, true);
935 u32 ceph_get_primary_affinity(struct ceph_osdmap *map, int osd)
937 BUG_ON(osd >= map->max_osd);
939 if (!map->osd_primary_affinity)
940 return CEPH_OSD_DEFAULT_PRIMARY_AFFINITY;
942 return map->osd_primary_affinity[osd];
945 static int set_primary_affinity(struct ceph_osdmap *map, int osd, u32 aff)
947 BUG_ON(osd >= map->max_osd);
949 if (!map->osd_primary_affinity) {
952 map->osd_primary_affinity = kmalloc(map->max_osd*sizeof(u32),
954 if (!map->osd_primary_affinity)
957 for (i = 0; i < map->max_osd; i++)
958 map->osd_primary_affinity[i] =
959 CEPH_OSD_DEFAULT_PRIMARY_AFFINITY;
962 map->osd_primary_affinity[osd] = aff;
967 static int decode_primary_affinity(void **p, void *end,
968 struct ceph_osdmap *map)
972 ceph_decode_32_safe(p, end, len, e_inval);
974 kfree(map->osd_primary_affinity);
975 map->osd_primary_affinity = NULL;
978 if (len != map->max_osd)
981 ceph_decode_need(p, end, map->max_osd*sizeof(u32), e_inval);
983 for (i = 0; i < map->max_osd; i++) {
986 ret = set_primary_affinity(map, i, ceph_decode_32(p));
997 static int decode_new_primary_affinity(void **p, void *end,
998 struct ceph_osdmap *map)
1002 ceph_decode_32_safe(p, end, n, e_inval);
1007 ceph_decode_32_safe(p, end, osd, e_inval);
1008 ceph_decode_32_safe(p, end, aff, e_inval);
1010 ret = set_primary_affinity(map, osd, aff);
1014 pr_info("osd%d primary-affinity 0x%x\n", osd, aff);
1024 * decode a full map.
1026 static int osdmap_decode(void **p, void *end, struct ceph_osdmap *map)
1035 dout("%s %p to %p len %d\n", __func__, *p, end, (int)(end - *p));
1037 err = get_osdmap_client_data_v(p, end, "full", &struct_v);
1041 /* fsid, epoch, created, modified */
1042 ceph_decode_need(p, end, sizeof(map->fsid) + sizeof(u32) +
1043 sizeof(map->created) + sizeof(map->modified), e_inval);
1044 ceph_decode_copy(p, &map->fsid, sizeof(map->fsid));
1045 epoch = map->epoch = ceph_decode_32(p);
1046 ceph_decode_copy(p, &map->created, sizeof(map->created));
1047 ceph_decode_copy(p, &map->modified, sizeof(map->modified));
1050 err = decode_pools(p, end, map);
1055 err = decode_pool_names(p, end, map);
1059 ceph_decode_32_safe(p, end, map->pool_max, e_inval);
1061 ceph_decode_32_safe(p, end, map->flags, e_inval);
1064 ceph_decode_32_safe(p, end, max, e_inval);
1066 /* (re)alloc osd arrays */
1067 err = osdmap_set_max_osd(map, max);
1071 /* osd_state, osd_weight, osd_addrs->client_addr */
1072 ceph_decode_need(p, end, 3*sizeof(u32) +
1073 map->max_osd*(1 + sizeof(*map->osd_weight) +
1074 sizeof(*map->osd_addr)), e_inval);
1076 if (ceph_decode_32(p) != map->max_osd)
1079 ceph_decode_copy(p, map->osd_state, map->max_osd);
1081 if (ceph_decode_32(p) != map->max_osd)
1084 for (i = 0; i < map->max_osd; i++)
1085 map->osd_weight[i] = ceph_decode_32(p);
1087 if (ceph_decode_32(p) != map->max_osd)
1090 ceph_decode_copy(p, map->osd_addr, map->max_osd*sizeof(*map->osd_addr));
1091 for (i = 0; i < map->max_osd; i++)
1092 ceph_decode_addr(&map->osd_addr[i]);
1095 err = decode_pg_temp(p, end, map);
1100 if (struct_v >= 1) {
1101 err = decode_primary_temp(p, end, map);
1106 /* primary_affinity */
1107 if (struct_v >= 2) {
1108 err = decode_primary_affinity(p, end, map);
1112 /* XXX can this happen? */
1113 kfree(map->osd_primary_affinity);
1114 map->osd_primary_affinity = NULL;
1118 ceph_decode_32_safe(p, end, len, e_inval);
1119 map->crush = crush_decode(*p, min(*p + len, end));
1120 if (IS_ERR(map->crush)) {
1121 err = PTR_ERR(map->crush);
1127 /* ignore the rest */
1130 dout("full osdmap epoch %d max_osd %d\n", map->epoch, map->max_osd);
1136 pr_err("corrupt full osdmap (%d) epoch %d off %d (%p of %p-%p)\n",
1137 err, epoch, (int)(*p - start), *p, start, end);
1138 print_hex_dump(KERN_DEBUG, "osdmap: ",
1139 DUMP_PREFIX_OFFSET, 16, 1,
1140 start, end - start, true);
1145 * Allocate and decode a full map.
1147 struct ceph_osdmap *ceph_osdmap_decode(void **p, void *end)
1149 struct ceph_osdmap *map;
1152 map = kzalloc(sizeof(*map), GFP_NOFS);
1154 return ERR_PTR(-ENOMEM);
1156 map->pg_temp = RB_ROOT;
1157 map->primary_temp = RB_ROOT;
1158 mutex_init(&map->crush_scratch_mutex);
1160 ret = osdmap_decode(p, end, map);
1162 ceph_osdmap_destroy(map);
1163 return ERR_PTR(ret);
1170 * decode and apply an incremental map update.
1172 struct ceph_osdmap *osdmap_apply_incremental(void **p, void *end,
1173 struct ceph_osdmap *map,
1174 struct ceph_messenger *msgr)
1176 struct crush_map *newcrush = NULL;
1177 struct ceph_fsid fsid;
1179 struct ceph_timespec modified;
1183 __s32 new_flags, max;
1188 dout("%s %p to %p len %d\n", __func__, *p, end, (int)(end - *p));
1190 err = get_osdmap_client_data_v(p, end, "inc", &struct_v);
1194 /* fsid, epoch, modified, new_pool_max, new_flags */
1195 ceph_decode_need(p, end, sizeof(fsid) + sizeof(u32) + sizeof(modified) +
1196 sizeof(u64) + sizeof(u32), e_inval);
1197 ceph_decode_copy(p, &fsid, sizeof(fsid));
1198 epoch = ceph_decode_32(p);
1199 BUG_ON(epoch != map->epoch+1);
1200 ceph_decode_copy(p, &modified, sizeof(modified));
1201 new_pool_max = ceph_decode_64(p);
1202 new_flags = ceph_decode_32(p);
1205 ceph_decode_32_safe(p, end, len, e_inval);
1207 dout("apply_incremental full map len %d, %p to %p\n",
1209 return ceph_osdmap_decode(p, min(*p+len, end));
1213 ceph_decode_32_safe(p, end, len, e_inval);
1215 newcrush = crush_decode(*p, min(*p+len, end));
1216 if (IS_ERR(newcrush)) {
1217 err = PTR_ERR(newcrush);
1226 map->flags = new_flags;
1227 if (new_pool_max >= 0)
1228 map->pool_max = new_pool_max;
1231 ceph_decode_32_safe(p, end, max, e_inval);
1233 err = osdmap_set_max_osd(map, max);
1239 map->modified = modified;
1242 crush_destroy(map->crush);
1243 map->crush = newcrush;
1248 err = decode_new_pools(p, end, map);
1252 /* new_pool_names */
1253 err = decode_pool_names(p, end, map);
1258 ceph_decode_32_safe(p, end, len, e_inval);
1260 struct ceph_pg_pool_info *pi;
1262 ceph_decode_64_safe(p, end, pool, e_inval);
1263 pi = __lookup_pg_pool(&map->pg_pools, pool);
1265 __remove_pg_pool(&map->pg_pools, pi);
1269 ceph_decode_32_safe(p, end, len, e_inval);
1272 struct ceph_entity_addr addr;
1273 ceph_decode_32_safe(p, end, osd, e_inval);
1274 ceph_decode_copy_safe(p, end, &addr, sizeof(addr), e_inval);
1275 ceph_decode_addr(&addr);
1276 pr_info("osd%d up\n", osd);
1277 BUG_ON(osd >= map->max_osd);
1278 map->osd_state[osd] |= CEPH_OSD_UP;
1279 map->osd_addr[osd] = addr;
1283 ceph_decode_32_safe(p, end, len, e_inval);
1287 ceph_decode_32_safe(p, end, osd, e_inval);
1288 xorstate = **(u8 **)p;
1289 (*p)++; /* clean flag */
1291 xorstate = CEPH_OSD_UP;
1292 if (xorstate & CEPH_OSD_UP)
1293 pr_info("osd%d down\n", osd);
1294 if (osd < map->max_osd)
1295 map->osd_state[osd] ^= xorstate;
1299 ceph_decode_32_safe(p, end, len, e_inval);
1302 ceph_decode_need(p, end, sizeof(u32)*2, e_inval);
1303 osd = ceph_decode_32(p);
1304 off = ceph_decode_32(p);
1305 pr_info("osd%d weight 0x%x %s\n", osd, off,
1306 off == CEPH_OSD_IN ? "(in)" :
1307 (off == CEPH_OSD_OUT ? "(out)" : ""));
1308 if (osd < map->max_osd)
1309 map->osd_weight[osd] = off;
1313 err = decode_new_pg_temp(p, end, map);
1317 /* new_primary_temp */
1318 if (struct_v >= 1) {
1319 err = decode_new_primary_temp(p, end, map);
1324 /* new_primary_affinity */
1325 if (struct_v >= 2) {
1326 err = decode_new_primary_affinity(p, end, map);
1331 /* ignore the rest */
1334 dout("inc osdmap epoch %d max_osd %d\n", map->epoch, map->max_osd);
1340 pr_err("corrupt inc osdmap (%d) epoch %d off %d (%p of %p-%p)\n",
1341 err, epoch, (int)(*p - start), *p, start, end);
1342 print_hex_dump(KERN_DEBUG, "osdmap: ",
1343 DUMP_PREFIX_OFFSET, 16, 1,
1344 start, end - start, true);
1346 crush_destroy(newcrush);
1347 return ERR_PTR(err);
1354 * calculate file layout from given offset, length.
1355 * fill in correct oid, logical length, and object extent
1358 * for now, we write only a single su, until we can
1359 * pass a stride back to the caller.
1361 int ceph_calc_file_object_mapping(struct ceph_file_layout *layout,
1364 u64 *oxoff, u64 *oxlen)
1366 u32 osize = le32_to_cpu(layout->fl_object_size);
1367 u32 su = le32_to_cpu(layout->fl_stripe_unit);
1368 u32 sc = le32_to_cpu(layout->fl_stripe_count);
1369 u32 bl, stripeno, stripepos, objsetno;
1373 dout("mapping %llu~%llu osize %u fl_su %u\n", off, len,
1375 if (su == 0 || sc == 0)
1377 su_per_object = osize / su;
1378 if (su_per_object == 0)
1380 dout("osize %u / su %u = su_per_object %u\n", osize, su,
1383 if ((su & ~PAGE_MASK) != 0)
1386 /* bl = *off / su; */
1390 dout("off %llu / su %u = bl %u\n", off, su, bl);
1393 stripepos = bl % sc;
1394 objsetno = stripeno / su_per_object;
1396 *ono = objsetno * sc + stripepos;
1397 dout("objset %u * sc %u = ono %u\n", objsetno, sc, (unsigned int)*ono);
1399 /* *oxoff = *off % layout->fl_stripe_unit; # offset in su */
1401 su_offset = do_div(t, su);
1402 *oxoff = su_offset + (stripeno % su_per_object) * su;
1405 * Calculate the length of the extent being written to the selected
1406 * object. This is the minimum of the full length requested (len) or
1407 * the remainder of the current stripe being written to.
1409 *oxlen = min_t(u64, len, su - su_offset);
1411 dout(" obj extent %llu~%llu\n", *oxoff, *oxlen);
1415 dout(" invalid layout\n");
1421 EXPORT_SYMBOL(ceph_calc_file_object_mapping);
1424 * Calculate mapping of a (oloc, oid) pair to a PG. Should only be
1425 * called with target's (oloc, oid), since tiering isn't taken into
1428 int ceph_oloc_oid_to_pg(struct ceph_osdmap *osdmap,
1429 struct ceph_object_locator *oloc,
1430 struct ceph_object_id *oid,
1431 struct ceph_pg *pg_out)
1433 struct ceph_pg_pool_info *pi;
1435 pi = __lookup_pg_pool(&osdmap->pg_pools, oloc->pool);
1439 pg_out->pool = oloc->pool;
1440 pg_out->seed = ceph_str_hash(pi->object_hash, oid->name,
1443 dout("%s '%.*s' pgid %llu.%x\n", __func__, oid->name_len, oid->name,
1444 pg_out->pool, pg_out->seed);
1447 EXPORT_SYMBOL(ceph_oloc_oid_to_pg);
1449 static int do_crush(struct ceph_osdmap *map, int ruleno, int x,
1450 int *result, int result_max,
1451 const __u32 *weight, int weight_max)
1455 BUG_ON(result_max > CEPH_PG_MAX_SIZE);
1457 mutex_lock(&map->crush_scratch_mutex);
1458 r = crush_do_rule(map->crush, ruleno, x, result, result_max,
1459 weight, weight_max, map->crush_scratch_ary);
1460 mutex_unlock(&map->crush_scratch_mutex);
1466 * Calculate raw (crush) set for given pgid.
1468 * Return raw set length, or error.
1470 static int pg_to_raw_osds(struct ceph_osdmap *osdmap,
1471 struct ceph_pg_pool_info *pool,
1472 struct ceph_pg pgid, u32 pps, int *osds)
1478 ruleno = crush_find_rule(osdmap->crush, pool->crush_ruleset,
1479 pool->type, pool->size);
1481 pr_err("no crush rule: pool %lld ruleset %d type %d size %d\n",
1482 pgid.pool, pool->crush_ruleset, pool->type,
1487 len = do_crush(osdmap, ruleno, pps, osds,
1488 min_t(int, pool->size, CEPH_PG_MAX_SIZE),
1489 osdmap->osd_weight, osdmap->max_osd);
1491 pr_err("error %d from crush rule %d: pool %lld ruleset %d type %d size %d\n",
1492 len, ruleno, pgid.pool, pool->crush_ruleset,
1493 pool->type, pool->size);
1501 * Given raw set, calculate up set and up primary.
1503 * Return up set length. *primary is set to up primary osd id, or -1
1504 * if up set is empty.
1506 static int raw_to_up_osds(struct ceph_osdmap *osdmap,
1507 struct ceph_pg_pool_info *pool,
1508 int *osds, int len, int *primary)
1510 int up_primary = -1;
1513 if (ceph_can_shift_osds(pool)) {
1516 for (i = 0; i < len; i++) {
1517 if (ceph_osd_is_down(osdmap, osds[i])) {
1522 osds[i - removed] = osds[i];
1527 up_primary = osds[0];
1529 for (i = len - 1; i >= 0; i--) {
1530 if (ceph_osd_is_down(osdmap, osds[i]))
1531 osds[i] = CRUSH_ITEM_NONE;
1533 up_primary = osds[i];
1537 *primary = up_primary;
1541 static void apply_primary_affinity(struct ceph_osdmap *osdmap, u32 pps,
1542 struct ceph_pg_pool_info *pool,
1543 int *osds, int len, int *primary)
1549 * Do we have any non-default primary_affinity values for these
1552 if (!osdmap->osd_primary_affinity)
1555 for (i = 0; i < len; i++) {
1558 if (osd != CRUSH_ITEM_NONE &&
1559 osdmap->osd_primary_affinity[osd] !=
1560 CEPH_OSD_DEFAULT_PRIMARY_AFFINITY) {
1568 * Pick the primary. Feed both the seed (for the pg) and the
1569 * osd into the hash/rng so that a proportional fraction of an
1570 * osd's pgs get rejected as primary.
1572 for (i = 0; i < len; i++) {
1576 if (osd == CRUSH_ITEM_NONE)
1579 aff = osdmap->osd_primary_affinity[osd];
1580 if (aff < CEPH_OSD_MAX_PRIMARY_AFFINITY &&
1581 (crush_hash32_2(CRUSH_HASH_RJENKINS1,
1582 pps, osd) >> 16) >= aff) {
1584 * We chose not to use this primary. Note it
1585 * anyway as a fallback in case we don't pick
1586 * anyone else, but keep looking.
1598 *primary = osds[pos];
1600 if (ceph_can_shift_osds(pool) && pos > 0) {
1601 /* move the new primary to the front */
1602 for (i = pos; i > 0; i--)
1603 osds[i] = osds[i - 1];
1609 * Given up set, apply pg_temp and primary_temp mappings.
1611 * Return acting set length. *primary is set to acting primary osd id,
1612 * or -1 if acting set is empty.
1614 static int apply_temps(struct ceph_osdmap *osdmap,
1615 struct ceph_pg_pool_info *pool, struct ceph_pg pgid,
1616 int *osds, int len, int *primary)
1618 struct ceph_pg_mapping *pg;
1624 pgid.seed = ceph_stable_mod(pgid.seed, pool->pg_num,
1628 pg = __lookup_pg_mapping(&osdmap->pg_temp, pgid);
1633 for (i = 0; i < pg->pg_temp.len; i++) {
1634 if (ceph_osd_is_down(osdmap, pg->pg_temp.osds[i])) {
1635 if (ceph_can_shift_osds(pool))
1638 osds[temp_len++] = CRUSH_ITEM_NONE;
1640 osds[temp_len++] = pg->pg_temp.osds[i];
1644 /* apply pg_temp's primary */
1645 for (i = 0; i < temp_len; i++) {
1646 if (osds[i] != CRUSH_ITEM_NONE) {
1647 temp_primary = osds[i];
1653 temp_primary = *primary;
1657 pg = __lookup_pg_mapping(&osdmap->primary_temp, pgid);
1659 temp_primary = pg->primary_temp.osd;
1661 *primary = temp_primary;
1666 * Calculate acting set for given pgid.
1668 * Return acting set length, or error. *primary is set to acting
1669 * primary osd id, or -1 if acting set is empty or on error.
1671 int ceph_calc_pg_acting(struct ceph_osdmap *osdmap, struct ceph_pg pgid,
1672 int *osds, int *primary)
1674 struct ceph_pg_pool_info *pool;
1678 pool = __lookup_pg_pool(&osdmap->pg_pools, pgid.pool);
1684 if (pool->flags & CEPH_POOL_FLAG_HASHPSPOOL) {
1685 /* hash pool id and seed so that pool PGs do not overlap */
1686 pps = crush_hash32_2(CRUSH_HASH_RJENKINS1,
1687 ceph_stable_mod(pgid.seed, pool->pgp_num,
1688 pool->pgp_num_mask),
1692 * legacy behavior: add ps and pool together. this is
1693 * not a great approach because the PGs from each pool
1694 * will overlap on top of each other: 0.5 == 1.4 ==
1697 pps = ceph_stable_mod(pgid.seed, pool->pgp_num,
1698 pool->pgp_num_mask) +
1699 (unsigned)pgid.pool;
1702 len = pg_to_raw_osds(osdmap, pool, pgid, pps, osds);
1708 len = raw_to_up_osds(osdmap, pool, osds, len, primary);
1710 apply_primary_affinity(osdmap, pps, pool, osds, len, primary);
1712 len = apply_temps(osdmap, pool, pgid, osds, len, primary);
1718 * Return primary osd for given pgid, or -1 if none.
1720 int ceph_calc_pg_primary(struct ceph_osdmap *osdmap, struct ceph_pg pgid)
1722 int osds[CEPH_PG_MAX_SIZE];
1725 ceph_calc_pg_acting(osdmap, pgid, osds, &primary);
1729 EXPORT_SYMBOL(ceph_calc_pg_primary);