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_8_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 crush_decode_straw2_bucket(void **p, void *end,
126 struct crush_bucket_straw2 *b)
129 dout("crush_decode_straw2_bucket %p to %p\n", *p, end);
130 b->item_weights = kcalloc(b->h.size, sizeof(u32), GFP_NOFS);
131 if (b->item_weights == NULL)
133 ceph_decode_need(p, end, b->h.size * sizeof(u32), bad);
134 for (j = 0; j < b->h.size; j++)
135 b->item_weights[j] = ceph_decode_32(p);
141 static int skip_name_map(void **p, void *end)
144 ceph_decode_32_safe(p, end, len ,bad);
148 ceph_decode_32_safe(p, end, strlen, bad);
156 static struct crush_map *crush_decode(void *pbyval, void *end)
162 void *start = pbyval;
166 dout("crush_decode %p to %p len %d\n", *p, end, (int)(end - *p));
168 c = kzalloc(sizeof(*c), GFP_NOFS);
170 return ERR_PTR(-ENOMEM);
172 /* set tunables to default values */
173 c->choose_local_tries = 2;
174 c->choose_local_fallback_tries = 5;
175 c->choose_total_tries = 19;
176 c->chooseleaf_descend_once = 0;
178 ceph_decode_need(p, end, 4*sizeof(u32), bad);
179 magic = ceph_decode_32(p);
180 if (magic != CRUSH_MAGIC) {
181 pr_err("crush_decode magic %x != current %x\n",
182 (unsigned int)magic, (unsigned int)CRUSH_MAGIC);
185 c->max_buckets = ceph_decode_32(p);
186 c->max_rules = ceph_decode_32(p);
187 c->max_devices = ceph_decode_32(p);
189 c->buckets = kcalloc(c->max_buckets, sizeof(*c->buckets), GFP_NOFS);
190 if (c->buckets == NULL)
192 c->rules = kcalloc(c->max_rules, sizeof(*c->rules), GFP_NOFS);
193 if (c->rules == NULL)
197 for (i = 0; i < c->max_buckets; i++) {
200 struct crush_bucket *b;
202 ceph_decode_32_safe(p, end, alg, bad);
204 c->buckets[i] = NULL;
207 dout("crush_decode bucket %d off %x %p to %p\n",
208 i, (int)(*p-start), *p, end);
211 case CRUSH_BUCKET_UNIFORM:
212 size = sizeof(struct crush_bucket_uniform);
214 case CRUSH_BUCKET_LIST:
215 size = sizeof(struct crush_bucket_list);
217 case CRUSH_BUCKET_TREE:
218 size = sizeof(struct crush_bucket_tree);
220 case CRUSH_BUCKET_STRAW:
221 size = sizeof(struct crush_bucket_straw);
223 case CRUSH_BUCKET_STRAW2:
224 size = sizeof(struct crush_bucket_straw2);
231 b = c->buckets[i] = kzalloc(size, GFP_NOFS);
235 ceph_decode_need(p, end, 4*sizeof(u32), bad);
236 b->id = ceph_decode_32(p);
237 b->type = ceph_decode_16(p);
238 b->alg = ceph_decode_8(p);
239 b->hash = ceph_decode_8(p);
240 b->weight = ceph_decode_32(p);
241 b->size = ceph_decode_32(p);
243 dout("crush_decode bucket size %d off %x %p to %p\n",
244 b->size, (int)(*p-start), *p, end);
246 b->items = kcalloc(b->size, sizeof(__s32), GFP_NOFS);
247 if (b->items == NULL)
249 b->perm = kcalloc(b->size, sizeof(u32), GFP_NOFS);
254 ceph_decode_need(p, end, b->size*sizeof(u32), bad);
255 for (j = 0; j < b->size; j++)
256 b->items[j] = ceph_decode_32(p);
259 case CRUSH_BUCKET_UNIFORM:
260 err = crush_decode_uniform_bucket(p, end,
261 (struct crush_bucket_uniform *)b);
265 case CRUSH_BUCKET_LIST:
266 err = crush_decode_list_bucket(p, end,
267 (struct crush_bucket_list *)b);
271 case CRUSH_BUCKET_TREE:
272 err = crush_decode_tree_bucket(p, end,
273 (struct crush_bucket_tree *)b);
277 case CRUSH_BUCKET_STRAW:
278 err = crush_decode_straw_bucket(p, end,
279 (struct crush_bucket_straw *)b);
283 case CRUSH_BUCKET_STRAW2:
284 err = crush_decode_straw2_bucket(p, end,
285 (struct crush_bucket_straw2 *)b);
293 dout("rule vec is %p\n", c->rules);
294 for (i = 0; i < c->max_rules; i++) {
296 struct crush_rule *r;
298 ceph_decode_32_safe(p, end, yes, bad);
300 dout("crush_decode NO rule %d off %x %p to %p\n",
301 i, (int)(*p-start), *p, end);
306 dout("crush_decode rule %d off %x %p to %p\n",
307 i, (int)(*p-start), *p, end);
310 ceph_decode_32_safe(p, end, yes, bad);
311 #if BITS_PER_LONG == 32
313 if (yes > (ULONG_MAX - sizeof(*r))
314 / sizeof(struct crush_rule_step))
317 r = c->rules[i] = kmalloc(sizeof(*r) +
318 yes*sizeof(struct crush_rule_step),
322 dout(" rule %d is at %p\n", i, r);
324 ceph_decode_copy_safe(p, end, &r->mask, 4, bad); /* 4 u8's */
325 ceph_decode_need(p, end, r->len*3*sizeof(u32), bad);
326 for (j = 0; j < r->len; j++) {
327 r->steps[j].op = ceph_decode_32(p);
328 r->steps[j].arg1 = ceph_decode_32(p);
329 r->steps[j].arg2 = ceph_decode_32(p);
333 /* ignore trailing name maps. */
334 for (num_name_maps = 0; num_name_maps < 3; num_name_maps++) {
335 err = skip_name_map(p, end);
341 ceph_decode_need(p, end, 3*sizeof(u32), done);
342 c->choose_local_tries = ceph_decode_32(p);
343 c->choose_local_fallback_tries = ceph_decode_32(p);
344 c->choose_total_tries = ceph_decode_32(p);
345 dout("crush decode tunable choose_local_tries = %d",
346 c->choose_local_tries);
347 dout("crush decode tunable choose_local_fallback_tries = %d",
348 c->choose_local_fallback_tries);
349 dout("crush decode tunable choose_total_tries = %d",
350 c->choose_total_tries);
352 ceph_decode_need(p, end, sizeof(u32), done);
353 c->chooseleaf_descend_once = ceph_decode_32(p);
354 dout("crush decode tunable chooseleaf_descend_once = %d",
355 c->chooseleaf_descend_once);
357 ceph_decode_need(p, end, sizeof(u8), done);
358 c->chooseleaf_vary_r = ceph_decode_8(p);
359 dout("crush decode tunable chooseleaf_vary_r = %d",
360 c->chooseleaf_vary_r);
363 dout("crush_decode success\n");
369 dout("crush_decode fail %d\n", err);
375 * rbtree of pg_mapping for handling pg_temp (explicit mapping of pgid
376 * to a set of osds) and primary_temp (explicit primary setting)
378 static int pgid_cmp(struct ceph_pg l, struct ceph_pg r)
391 static int __insert_pg_mapping(struct ceph_pg_mapping *new,
392 struct rb_root *root)
394 struct rb_node **p = &root->rb_node;
395 struct rb_node *parent = NULL;
396 struct ceph_pg_mapping *pg = NULL;
399 dout("__insert_pg_mapping %llx %p\n", *(u64 *)&new->pgid, new);
402 pg = rb_entry(parent, struct ceph_pg_mapping, node);
403 c = pgid_cmp(new->pgid, pg->pgid);
412 rb_link_node(&new->node, parent, p);
413 rb_insert_color(&new->node, root);
417 static struct ceph_pg_mapping *__lookup_pg_mapping(struct rb_root *root,
420 struct rb_node *n = root->rb_node;
421 struct ceph_pg_mapping *pg;
425 pg = rb_entry(n, struct ceph_pg_mapping, node);
426 c = pgid_cmp(pgid, pg->pgid);
432 dout("__lookup_pg_mapping %lld.%x got %p\n",
433 pgid.pool, pgid.seed, pg);
440 static int __remove_pg_mapping(struct rb_root *root, struct ceph_pg pgid)
442 struct ceph_pg_mapping *pg = __lookup_pg_mapping(root, pgid);
445 dout("__remove_pg_mapping %lld.%x %p\n", pgid.pool, pgid.seed,
447 rb_erase(&pg->node, root);
451 dout("__remove_pg_mapping %lld.%x dne\n", pgid.pool, pgid.seed);
456 * rbtree of pg pool info
458 static int __insert_pg_pool(struct rb_root *root, struct ceph_pg_pool_info *new)
460 struct rb_node **p = &root->rb_node;
461 struct rb_node *parent = NULL;
462 struct ceph_pg_pool_info *pi = NULL;
466 pi = rb_entry(parent, struct ceph_pg_pool_info, node);
467 if (new->id < pi->id)
469 else if (new->id > pi->id)
475 rb_link_node(&new->node, parent, p);
476 rb_insert_color(&new->node, root);
480 static struct ceph_pg_pool_info *__lookup_pg_pool(struct rb_root *root, u64 id)
482 struct ceph_pg_pool_info *pi;
483 struct rb_node *n = root->rb_node;
486 pi = rb_entry(n, struct ceph_pg_pool_info, node);
489 else if (id > pi->id)
497 struct ceph_pg_pool_info *ceph_pg_pool_by_id(struct ceph_osdmap *map, u64 id)
499 return __lookup_pg_pool(&map->pg_pools, id);
502 const char *ceph_pg_pool_name_by_id(struct ceph_osdmap *map, u64 id)
504 struct ceph_pg_pool_info *pi;
506 if (id == CEPH_NOPOOL)
509 if (WARN_ON_ONCE(id > (u64) INT_MAX))
512 pi = __lookup_pg_pool(&map->pg_pools, (int) id);
514 return pi ? pi->name : NULL;
516 EXPORT_SYMBOL(ceph_pg_pool_name_by_id);
518 int ceph_pg_poolid_by_name(struct ceph_osdmap *map, const char *name)
522 for (rbp = rb_first(&map->pg_pools); rbp; rbp = rb_next(rbp)) {
523 struct ceph_pg_pool_info *pi =
524 rb_entry(rbp, struct ceph_pg_pool_info, node);
525 if (pi->name && strcmp(pi->name, name) == 0)
530 EXPORT_SYMBOL(ceph_pg_poolid_by_name);
532 static void __remove_pg_pool(struct rb_root *root, struct ceph_pg_pool_info *pi)
534 rb_erase(&pi->node, root);
539 static int decode_pool(void **p, void *end, struct ceph_pg_pool_info *pi)
545 ceph_decode_need(p, end, 2 + 4, bad);
546 ev = ceph_decode_8(p); /* encoding version */
547 cv = ceph_decode_8(p); /* compat version */
549 pr_warn("got v %d < 5 cv %d of ceph_pg_pool\n", ev, cv);
553 pr_warn("got v %d cv %d > 9 of ceph_pg_pool\n", ev, cv);
556 len = ceph_decode_32(p);
557 ceph_decode_need(p, end, len, bad);
560 pi->type = ceph_decode_8(p);
561 pi->size = ceph_decode_8(p);
562 pi->crush_ruleset = ceph_decode_8(p);
563 pi->object_hash = ceph_decode_8(p);
565 pi->pg_num = ceph_decode_32(p);
566 pi->pgp_num = ceph_decode_32(p);
568 *p += 4 + 4; /* skip lpg* */
569 *p += 4; /* skip last_change */
570 *p += 8 + 4; /* skip snap_seq, snap_epoch */
573 num = ceph_decode_32(p);
575 *p += 8; /* snapid key */
576 *p += 1 + 1; /* versions */
577 len = ceph_decode_32(p);
581 /* skip removed_snaps */
582 num = ceph_decode_32(p);
585 *p += 8; /* skip auid */
586 pi->flags = ceph_decode_64(p);
587 *p += 4; /* skip crash_replay_interval */
590 *p += 1; /* skip min_size */
593 *p += 8 + 8; /* skip quota_max_* */
597 num = ceph_decode_32(p);
600 *p += 8; /* skip tier_of */
601 *p += 1; /* skip cache_mode */
603 pi->read_tier = ceph_decode_64(p);
604 pi->write_tier = ceph_decode_64(p);
610 /* ignore the rest */
620 static int decode_pool_names(void **p, void *end, struct ceph_osdmap *map)
622 struct ceph_pg_pool_info *pi;
626 ceph_decode_32_safe(p, end, num, bad);
627 dout(" %d pool names\n", num);
629 ceph_decode_64_safe(p, end, pool, bad);
630 ceph_decode_32_safe(p, end, len, bad);
631 dout(" pool %llu len %d\n", pool, len);
632 ceph_decode_need(p, end, len, bad);
633 pi = __lookup_pg_pool(&map->pg_pools, pool);
635 char *name = kstrndup(*p, len, GFP_NOFS);
641 dout(" name is %s\n", pi->name);
654 void ceph_osdmap_destroy(struct ceph_osdmap *map)
656 dout("osdmap_destroy %p\n", map);
658 crush_destroy(map->crush);
659 while (!RB_EMPTY_ROOT(&map->pg_temp)) {
660 struct ceph_pg_mapping *pg =
661 rb_entry(rb_first(&map->pg_temp),
662 struct ceph_pg_mapping, node);
663 rb_erase(&pg->node, &map->pg_temp);
666 while (!RB_EMPTY_ROOT(&map->primary_temp)) {
667 struct ceph_pg_mapping *pg =
668 rb_entry(rb_first(&map->primary_temp),
669 struct ceph_pg_mapping, node);
670 rb_erase(&pg->node, &map->primary_temp);
673 while (!RB_EMPTY_ROOT(&map->pg_pools)) {
674 struct ceph_pg_pool_info *pi =
675 rb_entry(rb_first(&map->pg_pools),
676 struct ceph_pg_pool_info, node);
677 __remove_pg_pool(&map->pg_pools, pi);
679 kfree(map->osd_state);
680 kfree(map->osd_weight);
681 kfree(map->osd_addr);
682 kfree(map->osd_primary_affinity);
687 * Adjust max_osd value, (re)allocate arrays.
689 * The new elements are properly initialized.
691 static int osdmap_set_max_osd(struct ceph_osdmap *map, int max)
695 struct ceph_entity_addr *addr;
698 state = krealloc(map->osd_state, max*sizeof(*state), GFP_NOFS);
701 map->osd_state = state;
703 weight = krealloc(map->osd_weight, max*sizeof(*weight), GFP_NOFS);
706 map->osd_weight = weight;
708 addr = krealloc(map->osd_addr, max*sizeof(*addr), GFP_NOFS);
711 map->osd_addr = addr;
713 for (i = map->max_osd; i < max; i++) {
714 map->osd_state[i] = 0;
715 map->osd_weight[i] = CEPH_OSD_OUT;
716 memset(map->osd_addr + i, 0, sizeof(*map->osd_addr));
719 if (map->osd_primary_affinity) {
722 affinity = krealloc(map->osd_primary_affinity,
723 max*sizeof(*affinity), GFP_NOFS);
726 map->osd_primary_affinity = affinity;
728 for (i = map->max_osd; i < max; i++)
729 map->osd_primary_affinity[i] =
730 CEPH_OSD_DEFAULT_PRIMARY_AFFINITY;
738 #define OSDMAP_WRAPPER_COMPAT_VER 7
739 #define OSDMAP_CLIENT_DATA_COMPAT_VER 1
742 * Return 0 or error. On success, *v is set to 0 for old (v6) osdmaps,
743 * to struct_v of the client_data section for new (v7 and above)
746 static int get_osdmap_client_data_v(void **p, void *end,
747 const char *prefix, u8 *v)
751 ceph_decode_8_safe(p, end, struct_v, e_inval);
755 ceph_decode_8_safe(p, end, struct_compat, e_inval);
756 if (struct_compat > OSDMAP_WRAPPER_COMPAT_VER) {
757 pr_warn("got v %d cv %d > %d of %s ceph_osdmap\n",
758 struct_v, struct_compat,
759 OSDMAP_WRAPPER_COMPAT_VER, prefix);
762 *p += 4; /* ignore wrapper struct_len */
764 ceph_decode_8_safe(p, end, struct_v, e_inval);
765 ceph_decode_8_safe(p, end, struct_compat, e_inval);
766 if (struct_compat > OSDMAP_CLIENT_DATA_COMPAT_VER) {
767 pr_warn("got v %d cv %d > %d of %s ceph_osdmap client data\n",
768 struct_v, struct_compat,
769 OSDMAP_CLIENT_DATA_COMPAT_VER, prefix);
772 *p += 4; /* ignore client data struct_len */
777 ceph_decode_16_safe(p, end, version, e_inval);
779 pr_warn("got v %d < 6 of %s ceph_osdmap\n",
784 /* old osdmap enconding */
795 static int __decode_pools(void **p, void *end, struct ceph_osdmap *map,
800 ceph_decode_32_safe(p, end, n, e_inval);
802 struct ceph_pg_pool_info *pi;
806 ceph_decode_64_safe(p, end, pool, e_inval);
808 pi = __lookup_pg_pool(&map->pg_pools, pool);
809 if (!incremental || !pi) {
810 pi = kzalloc(sizeof(*pi), GFP_NOFS);
816 ret = __insert_pg_pool(&map->pg_pools, pi);
823 ret = decode_pool(p, end, pi);
834 static int decode_pools(void **p, void *end, struct ceph_osdmap *map)
836 return __decode_pools(p, end, map, false);
839 static int decode_new_pools(void **p, void *end, struct ceph_osdmap *map)
841 return __decode_pools(p, end, map, true);
844 static int __decode_pg_temp(void **p, void *end, struct ceph_osdmap *map,
849 ceph_decode_32_safe(p, end, n, e_inval);
855 ret = ceph_decode_pgid(p, end, &pgid);
859 ceph_decode_32_safe(p, end, len, e_inval);
861 ret = __remove_pg_mapping(&map->pg_temp, pgid);
862 BUG_ON(!incremental && ret != -ENOENT);
864 if (!incremental || len > 0) {
865 struct ceph_pg_mapping *pg;
867 ceph_decode_need(p, end, len*sizeof(u32), e_inval);
869 if (len > (UINT_MAX - sizeof(*pg)) / sizeof(u32))
872 pg = kzalloc(sizeof(*pg) + len*sizeof(u32), GFP_NOFS);
877 pg->pg_temp.len = len;
878 for (i = 0; i < len; i++)
879 pg->pg_temp.osds[i] = ceph_decode_32(p);
881 ret = __insert_pg_mapping(pg, &map->pg_temp);
895 static int decode_pg_temp(void **p, void *end, struct ceph_osdmap *map)
897 return __decode_pg_temp(p, end, map, false);
900 static int decode_new_pg_temp(void **p, void *end, struct ceph_osdmap *map)
902 return __decode_pg_temp(p, end, map, true);
905 static int __decode_primary_temp(void **p, void *end, struct ceph_osdmap *map,
910 ceph_decode_32_safe(p, end, n, e_inval);
916 ret = ceph_decode_pgid(p, end, &pgid);
920 ceph_decode_32_safe(p, end, osd, e_inval);
922 ret = __remove_pg_mapping(&map->primary_temp, pgid);
923 BUG_ON(!incremental && ret != -ENOENT);
925 if (!incremental || osd != (u32)-1) {
926 struct ceph_pg_mapping *pg;
928 pg = kzalloc(sizeof(*pg), GFP_NOFS);
933 pg->primary_temp.osd = osd;
935 ret = __insert_pg_mapping(pg, &map->primary_temp);
949 static int decode_primary_temp(void **p, void *end, struct ceph_osdmap *map)
951 return __decode_primary_temp(p, end, map, false);
954 static int decode_new_primary_temp(void **p, void *end,
955 struct ceph_osdmap *map)
957 return __decode_primary_temp(p, end, map, true);
960 u32 ceph_get_primary_affinity(struct ceph_osdmap *map, int osd)
962 BUG_ON(osd >= map->max_osd);
964 if (!map->osd_primary_affinity)
965 return CEPH_OSD_DEFAULT_PRIMARY_AFFINITY;
967 return map->osd_primary_affinity[osd];
970 static int set_primary_affinity(struct ceph_osdmap *map, int osd, u32 aff)
972 BUG_ON(osd >= map->max_osd);
974 if (!map->osd_primary_affinity) {
977 map->osd_primary_affinity = kmalloc(map->max_osd*sizeof(u32),
979 if (!map->osd_primary_affinity)
982 for (i = 0; i < map->max_osd; i++)
983 map->osd_primary_affinity[i] =
984 CEPH_OSD_DEFAULT_PRIMARY_AFFINITY;
987 map->osd_primary_affinity[osd] = aff;
992 static int decode_primary_affinity(void **p, void *end,
993 struct ceph_osdmap *map)
997 ceph_decode_32_safe(p, end, len, e_inval);
999 kfree(map->osd_primary_affinity);
1000 map->osd_primary_affinity = NULL;
1003 if (len != map->max_osd)
1006 ceph_decode_need(p, end, map->max_osd*sizeof(u32), e_inval);
1008 for (i = 0; i < map->max_osd; i++) {
1011 ret = set_primary_affinity(map, i, ceph_decode_32(p));
1022 static int decode_new_primary_affinity(void **p, void *end,
1023 struct ceph_osdmap *map)
1027 ceph_decode_32_safe(p, end, n, e_inval);
1032 ceph_decode_32_safe(p, end, osd, e_inval);
1033 ceph_decode_32_safe(p, end, aff, e_inval);
1035 ret = set_primary_affinity(map, osd, aff);
1039 pr_info("osd%d primary-affinity 0x%x\n", osd, aff);
1049 * decode a full map.
1051 static int osdmap_decode(void **p, void *end, struct ceph_osdmap *map)
1060 dout("%s %p to %p len %d\n", __func__, *p, end, (int)(end - *p));
1062 err = get_osdmap_client_data_v(p, end, "full", &struct_v);
1066 /* fsid, epoch, created, modified */
1067 ceph_decode_need(p, end, sizeof(map->fsid) + sizeof(u32) +
1068 sizeof(map->created) + sizeof(map->modified), e_inval);
1069 ceph_decode_copy(p, &map->fsid, sizeof(map->fsid));
1070 epoch = map->epoch = ceph_decode_32(p);
1071 ceph_decode_copy(p, &map->created, sizeof(map->created));
1072 ceph_decode_copy(p, &map->modified, sizeof(map->modified));
1075 err = decode_pools(p, end, map);
1080 err = decode_pool_names(p, end, map);
1084 ceph_decode_32_safe(p, end, map->pool_max, e_inval);
1086 ceph_decode_32_safe(p, end, map->flags, e_inval);
1089 ceph_decode_32_safe(p, end, max, e_inval);
1091 /* (re)alloc osd arrays */
1092 err = osdmap_set_max_osd(map, max);
1096 /* osd_state, osd_weight, osd_addrs->client_addr */
1097 ceph_decode_need(p, end, 3*sizeof(u32) +
1098 map->max_osd*(1 + sizeof(*map->osd_weight) +
1099 sizeof(*map->osd_addr)), e_inval);
1101 if (ceph_decode_32(p) != map->max_osd)
1104 ceph_decode_copy(p, map->osd_state, map->max_osd);
1106 if (ceph_decode_32(p) != map->max_osd)
1109 for (i = 0; i < map->max_osd; i++)
1110 map->osd_weight[i] = ceph_decode_32(p);
1112 if (ceph_decode_32(p) != map->max_osd)
1115 ceph_decode_copy(p, map->osd_addr, map->max_osd*sizeof(*map->osd_addr));
1116 for (i = 0; i < map->max_osd; i++)
1117 ceph_decode_addr(&map->osd_addr[i]);
1120 err = decode_pg_temp(p, end, map);
1125 if (struct_v >= 1) {
1126 err = decode_primary_temp(p, end, map);
1131 /* primary_affinity */
1132 if (struct_v >= 2) {
1133 err = decode_primary_affinity(p, end, map);
1137 /* XXX can this happen? */
1138 kfree(map->osd_primary_affinity);
1139 map->osd_primary_affinity = NULL;
1143 ceph_decode_32_safe(p, end, len, e_inval);
1144 map->crush = crush_decode(*p, min(*p + len, end));
1145 if (IS_ERR(map->crush)) {
1146 err = PTR_ERR(map->crush);
1152 /* ignore the rest */
1155 dout("full osdmap epoch %d max_osd %d\n", map->epoch, map->max_osd);
1161 pr_err("corrupt full osdmap (%d) epoch %d off %d (%p of %p-%p)\n",
1162 err, epoch, (int)(*p - start), *p, start, end);
1163 print_hex_dump(KERN_DEBUG, "osdmap: ",
1164 DUMP_PREFIX_OFFSET, 16, 1,
1165 start, end - start, true);
1170 * Allocate and decode a full map.
1172 struct ceph_osdmap *ceph_osdmap_decode(void **p, void *end)
1174 struct ceph_osdmap *map;
1177 map = kzalloc(sizeof(*map), GFP_NOFS);
1179 return ERR_PTR(-ENOMEM);
1181 map->pg_temp = RB_ROOT;
1182 map->primary_temp = RB_ROOT;
1183 mutex_init(&map->crush_scratch_mutex);
1185 ret = osdmap_decode(p, end, map);
1187 ceph_osdmap_destroy(map);
1188 return ERR_PTR(ret);
1195 * decode and apply an incremental map update.
1197 struct ceph_osdmap *osdmap_apply_incremental(void **p, void *end,
1198 struct ceph_osdmap *map,
1199 struct ceph_messenger *msgr)
1201 struct crush_map *newcrush = NULL;
1202 struct ceph_fsid fsid;
1204 struct ceph_timespec modified;
1208 __s32 new_flags, max;
1213 dout("%s %p to %p len %d\n", __func__, *p, end, (int)(end - *p));
1215 err = get_osdmap_client_data_v(p, end, "inc", &struct_v);
1219 /* fsid, epoch, modified, new_pool_max, new_flags */
1220 ceph_decode_need(p, end, sizeof(fsid) + sizeof(u32) + sizeof(modified) +
1221 sizeof(u64) + sizeof(u32), e_inval);
1222 ceph_decode_copy(p, &fsid, sizeof(fsid));
1223 epoch = ceph_decode_32(p);
1224 BUG_ON(epoch != map->epoch+1);
1225 ceph_decode_copy(p, &modified, sizeof(modified));
1226 new_pool_max = ceph_decode_64(p);
1227 new_flags = ceph_decode_32(p);
1230 ceph_decode_32_safe(p, end, len, e_inval);
1232 dout("apply_incremental full map len %d, %p to %p\n",
1234 return ceph_osdmap_decode(p, min(*p+len, end));
1238 ceph_decode_32_safe(p, end, len, e_inval);
1240 newcrush = crush_decode(*p, min(*p+len, end));
1241 if (IS_ERR(newcrush)) {
1242 err = PTR_ERR(newcrush);
1251 map->flags = new_flags;
1252 if (new_pool_max >= 0)
1253 map->pool_max = new_pool_max;
1256 ceph_decode_32_safe(p, end, max, e_inval);
1258 err = osdmap_set_max_osd(map, max);
1264 map->modified = modified;
1267 crush_destroy(map->crush);
1268 map->crush = newcrush;
1273 err = decode_new_pools(p, end, map);
1277 /* new_pool_names */
1278 err = decode_pool_names(p, end, map);
1283 ceph_decode_32_safe(p, end, len, e_inval);
1285 struct ceph_pg_pool_info *pi;
1287 ceph_decode_64_safe(p, end, pool, e_inval);
1288 pi = __lookup_pg_pool(&map->pg_pools, pool);
1290 __remove_pg_pool(&map->pg_pools, pi);
1294 ceph_decode_32_safe(p, end, len, e_inval);
1297 struct ceph_entity_addr addr;
1298 ceph_decode_32_safe(p, end, osd, e_inval);
1299 ceph_decode_copy_safe(p, end, &addr, sizeof(addr), e_inval);
1300 ceph_decode_addr(&addr);
1301 pr_info("osd%d up\n", osd);
1302 BUG_ON(osd >= map->max_osd);
1303 map->osd_state[osd] |= CEPH_OSD_UP | CEPH_OSD_EXISTS;
1304 map->osd_addr[osd] = addr;
1308 ceph_decode_32_safe(p, end, len, e_inval);
1312 ceph_decode_32_safe(p, end, osd, e_inval);
1313 xorstate = **(u8 **)p;
1314 (*p)++; /* clean flag */
1316 xorstate = CEPH_OSD_UP;
1317 if (xorstate & CEPH_OSD_UP)
1318 pr_info("osd%d down\n", osd);
1319 if (osd < map->max_osd)
1320 map->osd_state[osd] ^= xorstate;
1324 ceph_decode_32_safe(p, end, len, e_inval);
1327 ceph_decode_need(p, end, sizeof(u32)*2, e_inval);
1328 osd = ceph_decode_32(p);
1329 off = ceph_decode_32(p);
1330 pr_info("osd%d weight 0x%x %s\n", osd, off,
1331 off == CEPH_OSD_IN ? "(in)" :
1332 (off == CEPH_OSD_OUT ? "(out)" : ""));
1333 if (osd < map->max_osd)
1334 map->osd_weight[osd] = off;
1338 err = decode_new_pg_temp(p, end, map);
1342 /* new_primary_temp */
1343 if (struct_v >= 1) {
1344 err = decode_new_primary_temp(p, end, map);
1349 /* new_primary_affinity */
1350 if (struct_v >= 2) {
1351 err = decode_new_primary_affinity(p, end, map);
1356 /* ignore the rest */
1359 dout("inc osdmap epoch %d max_osd %d\n", map->epoch, map->max_osd);
1365 pr_err("corrupt inc osdmap (%d) epoch %d off %d (%p of %p-%p)\n",
1366 err, epoch, (int)(*p - start), *p, start, end);
1367 print_hex_dump(KERN_DEBUG, "osdmap: ",
1368 DUMP_PREFIX_OFFSET, 16, 1,
1369 start, end - start, true);
1371 crush_destroy(newcrush);
1372 return ERR_PTR(err);
1379 * calculate file layout from given offset, length.
1380 * fill in correct oid, logical length, and object extent
1383 * for now, we write only a single su, until we can
1384 * pass a stride back to the caller.
1386 int ceph_calc_file_object_mapping(struct ceph_file_layout *layout,
1389 u64 *oxoff, u64 *oxlen)
1391 u32 osize = le32_to_cpu(layout->fl_object_size);
1392 u32 su = le32_to_cpu(layout->fl_stripe_unit);
1393 u32 sc = le32_to_cpu(layout->fl_stripe_count);
1394 u32 bl, stripeno, stripepos, objsetno;
1398 dout("mapping %llu~%llu osize %u fl_su %u\n", off, len,
1400 if (su == 0 || sc == 0)
1402 su_per_object = osize / su;
1403 if (su_per_object == 0)
1405 dout("osize %u / su %u = su_per_object %u\n", osize, su,
1408 if ((su & ~PAGE_MASK) != 0)
1411 /* bl = *off / su; */
1415 dout("off %llu / su %u = bl %u\n", off, su, bl);
1418 stripepos = bl % sc;
1419 objsetno = stripeno / su_per_object;
1421 *ono = objsetno * sc + stripepos;
1422 dout("objset %u * sc %u = ono %u\n", objsetno, sc, (unsigned int)*ono);
1424 /* *oxoff = *off % layout->fl_stripe_unit; # offset in su */
1426 su_offset = do_div(t, su);
1427 *oxoff = su_offset + (stripeno % su_per_object) * su;
1430 * Calculate the length of the extent being written to the selected
1431 * object. This is the minimum of the full length requested (len) or
1432 * the remainder of the current stripe being written to.
1434 *oxlen = min_t(u64, len, su - su_offset);
1436 dout(" obj extent %llu~%llu\n", *oxoff, *oxlen);
1440 dout(" invalid layout\n");
1446 EXPORT_SYMBOL(ceph_calc_file_object_mapping);
1449 * Calculate mapping of a (oloc, oid) pair to a PG. Should only be
1450 * called with target's (oloc, oid), since tiering isn't taken into
1453 int ceph_oloc_oid_to_pg(struct ceph_osdmap *osdmap,
1454 struct ceph_object_locator *oloc,
1455 struct ceph_object_id *oid,
1456 struct ceph_pg *pg_out)
1458 struct ceph_pg_pool_info *pi;
1460 pi = __lookup_pg_pool(&osdmap->pg_pools, oloc->pool);
1464 pg_out->pool = oloc->pool;
1465 pg_out->seed = ceph_str_hash(pi->object_hash, oid->name,
1468 dout("%s '%.*s' pgid %llu.%x\n", __func__, oid->name_len, oid->name,
1469 pg_out->pool, pg_out->seed);
1472 EXPORT_SYMBOL(ceph_oloc_oid_to_pg);
1474 static int do_crush(struct ceph_osdmap *map, int ruleno, int x,
1475 int *result, int result_max,
1476 const __u32 *weight, int weight_max)
1480 BUG_ON(result_max > CEPH_PG_MAX_SIZE);
1482 mutex_lock(&map->crush_scratch_mutex);
1483 r = crush_do_rule(map->crush, ruleno, x, result, result_max,
1484 weight, weight_max, map->crush_scratch_ary);
1485 mutex_unlock(&map->crush_scratch_mutex);
1491 * Calculate raw (crush) set for given pgid.
1493 * Return raw set length, or error.
1495 static int pg_to_raw_osds(struct ceph_osdmap *osdmap,
1496 struct ceph_pg_pool_info *pool,
1497 struct ceph_pg pgid, u32 pps, int *osds)
1503 ruleno = crush_find_rule(osdmap->crush, pool->crush_ruleset,
1504 pool->type, pool->size);
1506 pr_err("no crush rule: pool %lld ruleset %d type %d size %d\n",
1507 pgid.pool, pool->crush_ruleset, pool->type,
1512 len = do_crush(osdmap, ruleno, pps, osds,
1513 min_t(int, pool->size, CEPH_PG_MAX_SIZE),
1514 osdmap->osd_weight, osdmap->max_osd);
1516 pr_err("error %d from crush rule %d: pool %lld ruleset %d type %d size %d\n",
1517 len, ruleno, pgid.pool, pool->crush_ruleset,
1518 pool->type, pool->size);
1526 * Given raw set, calculate up set and up primary.
1528 * Return up set length. *primary is set to up primary osd id, or -1
1529 * if up set is empty.
1531 static int raw_to_up_osds(struct ceph_osdmap *osdmap,
1532 struct ceph_pg_pool_info *pool,
1533 int *osds, int len, int *primary)
1535 int up_primary = -1;
1538 if (ceph_can_shift_osds(pool)) {
1541 for (i = 0; i < len; i++) {
1542 if (ceph_osd_is_down(osdmap, osds[i])) {
1547 osds[i - removed] = osds[i];
1552 up_primary = osds[0];
1554 for (i = len - 1; i >= 0; i--) {
1555 if (ceph_osd_is_down(osdmap, osds[i]))
1556 osds[i] = CRUSH_ITEM_NONE;
1558 up_primary = osds[i];
1562 *primary = up_primary;
1566 static void apply_primary_affinity(struct ceph_osdmap *osdmap, u32 pps,
1567 struct ceph_pg_pool_info *pool,
1568 int *osds, int len, int *primary)
1574 * Do we have any non-default primary_affinity values for these
1577 if (!osdmap->osd_primary_affinity)
1580 for (i = 0; i < len; i++) {
1583 if (osd != CRUSH_ITEM_NONE &&
1584 osdmap->osd_primary_affinity[osd] !=
1585 CEPH_OSD_DEFAULT_PRIMARY_AFFINITY) {
1593 * Pick the primary. Feed both the seed (for the pg) and the
1594 * osd into the hash/rng so that a proportional fraction of an
1595 * osd's pgs get rejected as primary.
1597 for (i = 0; i < len; i++) {
1601 if (osd == CRUSH_ITEM_NONE)
1604 aff = osdmap->osd_primary_affinity[osd];
1605 if (aff < CEPH_OSD_MAX_PRIMARY_AFFINITY &&
1606 (crush_hash32_2(CRUSH_HASH_RJENKINS1,
1607 pps, osd) >> 16) >= aff) {
1609 * We chose not to use this primary. Note it
1610 * anyway as a fallback in case we don't pick
1611 * anyone else, but keep looking.
1623 *primary = osds[pos];
1625 if (ceph_can_shift_osds(pool) && pos > 0) {
1626 /* move the new primary to the front */
1627 for (i = pos; i > 0; i--)
1628 osds[i] = osds[i - 1];
1634 * Given up set, apply pg_temp and primary_temp mappings.
1636 * Return acting set length. *primary is set to acting primary osd id,
1637 * or -1 if acting set is empty.
1639 static int apply_temps(struct ceph_osdmap *osdmap,
1640 struct ceph_pg_pool_info *pool, struct ceph_pg pgid,
1641 int *osds, int len, int *primary)
1643 struct ceph_pg_mapping *pg;
1649 pgid.seed = ceph_stable_mod(pgid.seed, pool->pg_num,
1653 pg = __lookup_pg_mapping(&osdmap->pg_temp, pgid);
1658 for (i = 0; i < pg->pg_temp.len; i++) {
1659 if (ceph_osd_is_down(osdmap, pg->pg_temp.osds[i])) {
1660 if (ceph_can_shift_osds(pool))
1663 osds[temp_len++] = CRUSH_ITEM_NONE;
1665 osds[temp_len++] = pg->pg_temp.osds[i];
1669 /* apply pg_temp's primary */
1670 for (i = 0; i < temp_len; i++) {
1671 if (osds[i] != CRUSH_ITEM_NONE) {
1672 temp_primary = osds[i];
1678 temp_primary = *primary;
1682 pg = __lookup_pg_mapping(&osdmap->primary_temp, pgid);
1684 temp_primary = pg->primary_temp.osd;
1686 *primary = temp_primary;
1691 * Calculate acting set for given pgid.
1693 * Return acting set length, or error. *primary is set to acting
1694 * primary osd id, or -1 if acting set is empty or on error.
1696 int ceph_calc_pg_acting(struct ceph_osdmap *osdmap, struct ceph_pg pgid,
1697 int *osds, int *primary)
1699 struct ceph_pg_pool_info *pool;
1703 pool = __lookup_pg_pool(&osdmap->pg_pools, pgid.pool);
1709 if (pool->flags & CEPH_POOL_FLAG_HASHPSPOOL) {
1710 /* hash pool id and seed so that pool PGs do not overlap */
1711 pps = crush_hash32_2(CRUSH_HASH_RJENKINS1,
1712 ceph_stable_mod(pgid.seed, pool->pgp_num,
1713 pool->pgp_num_mask),
1717 * legacy behavior: add ps and pool together. this is
1718 * not a great approach because the PGs from each pool
1719 * will overlap on top of each other: 0.5 == 1.4 ==
1722 pps = ceph_stable_mod(pgid.seed, pool->pgp_num,
1723 pool->pgp_num_mask) +
1724 (unsigned)pgid.pool;
1727 len = pg_to_raw_osds(osdmap, pool, pgid, pps, osds);
1733 len = raw_to_up_osds(osdmap, pool, osds, len, primary);
1735 apply_primary_affinity(osdmap, pps, pool, osds, len, primary);
1737 len = apply_temps(osdmap, pool, pgid, osds, len, primary);
1743 * Return primary osd for given pgid, or -1 if none.
1745 int ceph_calc_pg_primary(struct ceph_osdmap *osdmap, struct ceph_pg pgid)
1747 int osds[CEPH_PG_MAX_SIZE];
1750 ceph_calc_pg_acting(osdmap, pgid, osds, &primary);
1754 EXPORT_SYMBOL(ceph_calc_pg_primary);