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 void crush_finalize(struct crush_map *c)
160 /* Space for the array of pointers to per-bucket workspace */
161 c->working_size = sizeof(struct crush_work) +
162 c->max_buckets * sizeof(struct crush_work_bucket *);
164 for (b = 0; b < c->max_buckets; b++) {
168 switch (c->buckets[b]->alg) {
171 * The base case, permutation variables and
172 * the pointer to the permutation array.
174 c->working_size += sizeof(struct crush_work_bucket);
177 /* Every bucket has a permutation array. */
178 c->working_size += c->buckets[b]->size * sizeof(__u32);
182 static struct crush_map *crush_decode(void *pbyval, void *end)
188 void *start = pbyval;
192 dout("crush_decode %p to %p len %d\n", *p, end, (int)(end - *p));
194 c = kzalloc(sizeof(*c), GFP_NOFS);
196 return ERR_PTR(-ENOMEM);
198 /* set tunables to default values */
199 c->choose_local_tries = 2;
200 c->choose_local_fallback_tries = 5;
201 c->choose_total_tries = 19;
202 c->chooseleaf_descend_once = 0;
204 ceph_decode_need(p, end, 4*sizeof(u32), bad);
205 magic = ceph_decode_32(p);
206 if (magic != CRUSH_MAGIC) {
207 pr_err("crush_decode magic %x != current %x\n",
208 (unsigned int)magic, (unsigned int)CRUSH_MAGIC);
211 c->max_buckets = ceph_decode_32(p);
212 c->max_rules = ceph_decode_32(p);
213 c->max_devices = ceph_decode_32(p);
215 c->buckets = kcalloc(c->max_buckets, sizeof(*c->buckets), GFP_NOFS);
216 if (c->buckets == NULL)
218 c->rules = kcalloc(c->max_rules, sizeof(*c->rules), GFP_NOFS);
219 if (c->rules == NULL)
223 for (i = 0; i < c->max_buckets; i++) {
226 struct crush_bucket *b;
228 ceph_decode_32_safe(p, end, alg, bad);
230 c->buckets[i] = NULL;
233 dout("crush_decode bucket %d off %x %p to %p\n",
234 i, (int)(*p-start), *p, end);
237 case CRUSH_BUCKET_UNIFORM:
238 size = sizeof(struct crush_bucket_uniform);
240 case CRUSH_BUCKET_LIST:
241 size = sizeof(struct crush_bucket_list);
243 case CRUSH_BUCKET_TREE:
244 size = sizeof(struct crush_bucket_tree);
246 case CRUSH_BUCKET_STRAW:
247 size = sizeof(struct crush_bucket_straw);
249 case CRUSH_BUCKET_STRAW2:
250 size = sizeof(struct crush_bucket_straw2);
257 b = c->buckets[i] = kzalloc(size, GFP_NOFS);
261 ceph_decode_need(p, end, 4*sizeof(u32), bad);
262 b->id = ceph_decode_32(p);
263 b->type = ceph_decode_16(p);
264 b->alg = ceph_decode_8(p);
265 b->hash = ceph_decode_8(p);
266 b->weight = ceph_decode_32(p);
267 b->size = ceph_decode_32(p);
269 dout("crush_decode bucket size %d off %x %p to %p\n",
270 b->size, (int)(*p-start), *p, end);
272 b->items = kcalloc(b->size, sizeof(__s32), GFP_NOFS);
273 if (b->items == NULL)
276 ceph_decode_need(p, end, b->size*sizeof(u32), bad);
277 for (j = 0; j < b->size; j++)
278 b->items[j] = ceph_decode_32(p);
281 case CRUSH_BUCKET_UNIFORM:
282 err = crush_decode_uniform_bucket(p, end,
283 (struct crush_bucket_uniform *)b);
287 case CRUSH_BUCKET_LIST:
288 err = crush_decode_list_bucket(p, end,
289 (struct crush_bucket_list *)b);
293 case CRUSH_BUCKET_TREE:
294 err = crush_decode_tree_bucket(p, end,
295 (struct crush_bucket_tree *)b);
299 case CRUSH_BUCKET_STRAW:
300 err = crush_decode_straw_bucket(p, end,
301 (struct crush_bucket_straw *)b);
305 case CRUSH_BUCKET_STRAW2:
306 err = crush_decode_straw2_bucket(p, end,
307 (struct crush_bucket_straw2 *)b);
315 dout("rule vec is %p\n", c->rules);
316 for (i = 0; i < c->max_rules; i++) {
318 struct crush_rule *r;
320 ceph_decode_32_safe(p, end, yes, bad);
322 dout("crush_decode NO rule %d off %x %p to %p\n",
323 i, (int)(*p-start), *p, end);
328 dout("crush_decode rule %d off %x %p to %p\n",
329 i, (int)(*p-start), *p, end);
332 ceph_decode_32_safe(p, end, yes, bad);
333 #if BITS_PER_LONG == 32
335 if (yes > (ULONG_MAX - sizeof(*r))
336 / sizeof(struct crush_rule_step))
339 r = c->rules[i] = kmalloc(sizeof(*r) +
340 yes*sizeof(struct crush_rule_step),
344 dout(" rule %d is at %p\n", i, r);
346 ceph_decode_copy_safe(p, end, &r->mask, 4, bad); /* 4 u8's */
347 ceph_decode_need(p, end, r->len*3*sizeof(u32), bad);
348 for (j = 0; j < r->len; j++) {
349 r->steps[j].op = ceph_decode_32(p);
350 r->steps[j].arg1 = ceph_decode_32(p);
351 r->steps[j].arg2 = ceph_decode_32(p);
355 /* ignore trailing name maps. */
356 for (num_name_maps = 0; num_name_maps < 3; num_name_maps++) {
357 err = skip_name_map(p, end);
363 ceph_decode_need(p, end, 3*sizeof(u32), done);
364 c->choose_local_tries = ceph_decode_32(p);
365 c->choose_local_fallback_tries = ceph_decode_32(p);
366 c->choose_total_tries = ceph_decode_32(p);
367 dout("crush decode tunable choose_local_tries = %d\n",
368 c->choose_local_tries);
369 dout("crush decode tunable choose_local_fallback_tries = %d\n",
370 c->choose_local_fallback_tries);
371 dout("crush decode tunable choose_total_tries = %d\n",
372 c->choose_total_tries);
374 ceph_decode_need(p, end, sizeof(u32), done);
375 c->chooseleaf_descend_once = ceph_decode_32(p);
376 dout("crush decode tunable chooseleaf_descend_once = %d\n",
377 c->chooseleaf_descend_once);
379 ceph_decode_need(p, end, sizeof(u8), done);
380 c->chooseleaf_vary_r = ceph_decode_8(p);
381 dout("crush decode tunable chooseleaf_vary_r = %d\n",
382 c->chooseleaf_vary_r);
384 /* skip straw_calc_version, allowed_bucket_algs */
385 ceph_decode_need(p, end, sizeof(u8) + sizeof(u32), done);
386 *p += sizeof(u8) + sizeof(u32);
388 ceph_decode_need(p, end, sizeof(u8), done);
389 c->chooseleaf_stable = ceph_decode_8(p);
390 dout("crush decode tunable chooseleaf_stable = %d\n",
391 c->chooseleaf_stable);
395 dout("crush_decode success\n");
401 dout("crush_decode fail %d\n", err);
406 int ceph_pg_compare(const struct ceph_pg *lhs, const struct ceph_pg *rhs)
408 if (lhs->pool < rhs->pool)
410 if (lhs->pool > rhs->pool)
412 if (lhs->seed < rhs->seed)
414 if (lhs->seed > rhs->seed)
421 * rbtree of pg_mapping for handling pg_temp (explicit mapping of pgid
422 * to a set of osds) and primary_temp (explicit primary setting)
424 static int __insert_pg_mapping(struct ceph_pg_mapping *new,
425 struct rb_root *root)
427 struct rb_node **p = &root->rb_node;
428 struct rb_node *parent = NULL;
429 struct ceph_pg_mapping *pg = NULL;
432 dout("__insert_pg_mapping %llx %p\n", *(u64 *)&new->pgid, new);
435 pg = rb_entry(parent, struct ceph_pg_mapping, node);
436 c = ceph_pg_compare(&new->pgid, &pg->pgid);
445 rb_link_node(&new->node, parent, p);
446 rb_insert_color(&new->node, root);
450 static struct ceph_pg_mapping *__lookup_pg_mapping(struct rb_root *root,
453 struct rb_node *n = root->rb_node;
454 struct ceph_pg_mapping *pg;
458 pg = rb_entry(n, struct ceph_pg_mapping, node);
459 c = ceph_pg_compare(&pgid, &pg->pgid);
465 dout("__lookup_pg_mapping %lld.%x got %p\n",
466 pgid.pool, pgid.seed, pg);
473 static int __remove_pg_mapping(struct rb_root *root, struct ceph_pg pgid)
475 struct ceph_pg_mapping *pg = __lookup_pg_mapping(root, pgid);
478 dout("__remove_pg_mapping %lld.%x %p\n", pgid.pool, pgid.seed,
480 rb_erase(&pg->node, root);
484 dout("__remove_pg_mapping %lld.%x dne\n", pgid.pool, pgid.seed);
489 * rbtree of pg pool info
491 static int __insert_pg_pool(struct rb_root *root, struct ceph_pg_pool_info *new)
493 struct rb_node **p = &root->rb_node;
494 struct rb_node *parent = NULL;
495 struct ceph_pg_pool_info *pi = NULL;
499 pi = rb_entry(parent, struct ceph_pg_pool_info, node);
500 if (new->id < pi->id)
502 else if (new->id > pi->id)
508 rb_link_node(&new->node, parent, p);
509 rb_insert_color(&new->node, root);
513 static struct ceph_pg_pool_info *__lookup_pg_pool(struct rb_root *root, u64 id)
515 struct ceph_pg_pool_info *pi;
516 struct rb_node *n = root->rb_node;
519 pi = rb_entry(n, struct ceph_pg_pool_info, node);
522 else if (id > pi->id)
530 struct ceph_pg_pool_info *ceph_pg_pool_by_id(struct ceph_osdmap *map, u64 id)
532 return __lookup_pg_pool(&map->pg_pools, id);
535 const char *ceph_pg_pool_name_by_id(struct ceph_osdmap *map, u64 id)
537 struct ceph_pg_pool_info *pi;
539 if (id == CEPH_NOPOOL)
542 if (WARN_ON_ONCE(id > (u64) INT_MAX))
545 pi = __lookup_pg_pool(&map->pg_pools, (int) id);
547 return pi ? pi->name : NULL;
549 EXPORT_SYMBOL(ceph_pg_pool_name_by_id);
551 int ceph_pg_poolid_by_name(struct ceph_osdmap *map, const char *name)
555 for (rbp = rb_first(&map->pg_pools); rbp; rbp = rb_next(rbp)) {
556 struct ceph_pg_pool_info *pi =
557 rb_entry(rbp, struct ceph_pg_pool_info, node);
558 if (pi->name && strcmp(pi->name, name) == 0)
563 EXPORT_SYMBOL(ceph_pg_poolid_by_name);
565 static void __remove_pg_pool(struct rb_root *root, struct ceph_pg_pool_info *pi)
567 rb_erase(&pi->node, root);
572 static int decode_pool(void **p, void *end, struct ceph_pg_pool_info *pi)
578 ceph_decode_need(p, end, 2 + 4, bad);
579 ev = ceph_decode_8(p); /* encoding version */
580 cv = ceph_decode_8(p); /* compat version */
582 pr_warn("got v %d < 5 cv %d of ceph_pg_pool\n", ev, cv);
586 pr_warn("got v %d cv %d > 9 of ceph_pg_pool\n", ev, cv);
589 len = ceph_decode_32(p);
590 ceph_decode_need(p, end, len, bad);
593 pi->type = ceph_decode_8(p);
594 pi->size = ceph_decode_8(p);
595 pi->crush_ruleset = ceph_decode_8(p);
596 pi->object_hash = ceph_decode_8(p);
598 pi->pg_num = ceph_decode_32(p);
599 pi->pgp_num = ceph_decode_32(p);
601 *p += 4 + 4; /* skip lpg* */
602 *p += 4; /* skip last_change */
603 *p += 8 + 4; /* skip snap_seq, snap_epoch */
606 num = ceph_decode_32(p);
608 *p += 8; /* snapid key */
609 *p += 1 + 1; /* versions */
610 len = ceph_decode_32(p);
614 /* skip removed_snaps */
615 num = ceph_decode_32(p);
618 *p += 8; /* skip auid */
619 pi->flags = ceph_decode_64(p);
620 *p += 4; /* skip crash_replay_interval */
623 pi->min_size = ceph_decode_8(p);
625 pi->min_size = pi->size - pi->size / 2;
628 *p += 8 + 8; /* skip quota_max_* */
632 num = ceph_decode_32(p);
635 *p += 8; /* skip tier_of */
636 *p += 1; /* skip cache_mode */
638 pi->read_tier = ceph_decode_64(p);
639 pi->write_tier = ceph_decode_64(p);
646 /* skip properties */
647 num = ceph_decode_32(p);
649 len = ceph_decode_32(p);
651 len = ceph_decode_32(p);
657 /* skip hit_set_params */
658 *p += 1 + 1; /* versions */
659 len = ceph_decode_32(p);
662 *p += 4; /* skip hit_set_period */
663 *p += 4; /* skip hit_set_count */
667 *p += 4; /* skip stripe_width */
670 *p += 8; /* skip target_max_bytes */
671 *p += 8; /* skip target_max_objects */
672 *p += 4; /* skip cache_target_dirty_ratio_micro */
673 *p += 4; /* skip cache_target_full_ratio_micro */
674 *p += 4; /* skip cache_min_flush_age */
675 *p += 4; /* skip cache_min_evict_age */
679 /* skip erasure_code_profile */
680 len = ceph_decode_32(p);
685 pi->last_force_request_resend = ceph_decode_32(p);
687 pi->last_force_request_resend = 0;
689 /* ignore the rest */
699 static int decode_pool_names(void **p, void *end, struct ceph_osdmap *map)
701 struct ceph_pg_pool_info *pi;
705 ceph_decode_32_safe(p, end, num, bad);
706 dout(" %d pool names\n", num);
708 ceph_decode_64_safe(p, end, pool, bad);
709 ceph_decode_32_safe(p, end, len, bad);
710 dout(" pool %llu len %d\n", pool, len);
711 ceph_decode_need(p, end, len, bad);
712 pi = __lookup_pg_pool(&map->pg_pools, pool);
714 char *name = kstrndup(*p, len, GFP_NOFS);
720 dout(" name is %s\n", pi->name);
733 struct ceph_osdmap *ceph_osdmap_alloc(void)
735 struct ceph_osdmap *map;
737 map = kzalloc(sizeof(*map), GFP_NOIO);
741 map->pg_pools = RB_ROOT;
743 map->pg_temp = RB_ROOT;
744 map->primary_temp = RB_ROOT;
745 mutex_init(&map->crush_workspace_mutex);
750 void ceph_osdmap_destroy(struct ceph_osdmap *map)
752 dout("osdmap_destroy %p\n", map);
754 crush_destroy(map->crush);
755 while (!RB_EMPTY_ROOT(&map->pg_temp)) {
756 struct ceph_pg_mapping *pg =
757 rb_entry(rb_first(&map->pg_temp),
758 struct ceph_pg_mapping, node);
759 rb_erase(&pg->node, &map->pg_temp);
762 while (!RB_EMPTY_ROOT(&map->primary_temp)) {
763 struct ceph_pg_mapping *pg =
764 rb_entry(rb_first(&map->primary_temp),
765 struct ceph_pg_mapping, node);
766 rb_erase(&pg->node, &map->primary_temp);
769 while (!RB_EMPTY_ROOT(&map->pg_pools)) {
770 struct ceph_pg_pool_info *pi =
771 rb_entry(rb_first(&map->pg_pools),
772 struct ceph_pg_pool_info, node);
773 __remove_pg_pool(&map->pg_pools, pi);
775 kfree(map->osd_state);
776 kfree(map->osd_weight);
777 kfree(map->osd_addr);
778 kfree(map->osd_primary_affinity);
779 kfree(map->crush_workspace);
784 * Adjust max_osd value, (re)allocate arrays.
786 * The new elements are properly initialized.
788 static int osdmap_set_max_osd(struct ceph_osdmap *map, int max)
792 struct ceph_entity_addr *addr;
795 state = krealloc(map->osd_state, max*sizeof(*state), GFP_NOFS);
798 map->osd_state = state;
800 weight = krealloc(map->osd_weight, max*sizeof(*weight), GFP_NOFS);
803 map->osd_weight = weight;
805 addr = krealloc(map->osd_addr, max*sizeof(*addr), GFP_NOFS);
808 map->osd_addr = addr;
810 for (i = map->max_osd; i < max; i++) {
811 map->osd_state[i] = 0;
812 map->osd_weight[i] = CEPH_OSD_OUT;
813 memset(map->osd_addr + i, 0, sizeof(*map->osd_addr));
816 if (map->osd_primary_affinity) {
819 affinity = krealloc(map->osd_primary_affinity,
820 max*sizeof(*affinity), GFP_NOFS);
823 map->osd_primary_affinity = affinity;
825 for (i = map->max_osd; i < max; i++)
826 map->osd_primary_affinity[i] =
827 CEPH_OSD_DEFAULT_PRIMARY_AFFINITY;
835 static int osdmap_set_crush(struct ceph_osdmap *map, struct crush_map *crush)
841 return PTR_ERR(crush);
843 work_size = crush_work_size(crush, CEPH_PG_MAX_SIZE);
844 dout("%s work_size %zu bytes\n", __func__, work_size);
845 workspace = kmalloc(work_size, GFP_NOIO);
847 crush_destroy(crush);
850 crush_init_workspace(crush, workspace);
853 crush_destroy(map->crush);
854 kfree(map->crush_workspace);
856 map->crush_workspace = workspace;
860 #define OSDMAP_WRAPPER_COMPAT_VER 7
861 #define OSDMAP_CLIENT_DATA_COMPAT_VER 1
864 * Return 0 or error. On success, *v is set to 0 for old (v6) osdmaps,
865 * to struct_v of the client_data section for new (v7 and above)
868 static int get_osdmap_client_data_v(void **p, void *end,
869 const char *prefix, u8 *v)
873 ceph_decode_8_safe(p, end, struct_v, e_inval);
877 ceph_decode_8_safe(p, end, struct_compat, e_inval);
878 if (struct_compat > OSDMAP_WRAPPER_COMPAT_VER) {
879 pr_warn("got v %d cv %d > %d of %s ceph_osdmap\n",
880 struct_v, struct_compat,
881 OSDMAP_WRAPPER_COMPAT_VER, prefix);
884 *p += 4; /* ignore wrapper struct_len */
886 ceph_decode_8_safe(p, end, struct_v, e_inval);
887 ceph_decode_8_safe(p, end, struct_compat, e_inval);
888 if (struct_compat > OSDMAP_CLIENT_DATA_COMPAT_VER) {
889 pr_warn("got v %d cv %d > %d of %s ceph_osdmap client data\n",
890 struct_v, struct_compat,
891 OSDMAP_CLIENT_DATA_COMPAT_VER, prefix);
894 *p += 4; /* ignore client data struct_len */
899 ceph_decode_16_safe(p, end, version, e_inval);
901 pr_warn("got v %d < 6 of %s ceph_osdmap\n",
906 /* old osdmap enconding */
917 static int __decode_pools(void **p, void *end, struct ceph_osdmap *map,
922 ceph_decode_32_safe(p, end, n, e_inval);
924 struct ceph_pg_pool_info *pi;
928 ceph_decode_64_safe(p, end, pool, e_inval);
930 pi = __lookup_pg_pool(&map->pg_pools, pool);
931 if (!incremental || !pi) {
932 pi = kzalloc(sizeof(*pi), GFP_NOFS);
938 ret = __insert_pg_pool(&map->pg_pools, pi);
945 ret = decode_pool(p, end, pi);
956 static int decode_pools(void **p, void *end, struct ceph_osdmap *map)
958 return __decode_pools(p, end, map, false);
961 static int decode_new_pools(void **p, void *end, struct ceph_osdmap *map)
963 return __decode_pools(p, end, map, true);
966 static int __decode_pg_temp(void **p, void *end, struct ceph_osdmap *map,
971 ceph_decode_32_safe(p, end, n, e_inval);
977 ret = ceph_decode_pgid(p, end, &pgid);
981 ceph_decode_32_safe(p, end, len, e_inval);
983 ret = __remove_pg_mapping(&map->pg_temp, pgid);
984 BUG_ON(!incremental && ret != -ENOENT);
986 if (!incremental || len > 0) {
987 struct ceph_pg_mapping *pg;
989 ceph_decode_need(p, end, len*sizeof(u32), e_inval);
991 if (len > (UINT_MAX - sizeof(*pg)) / sizeof(u32))
994 pg = kzalloc(sizeof(*pg) + len*sizeof(u32), GFP_NOFS);
999 pg->pg_temp.len = len;
1000 for (i = 0; i < len; i++)
1001 pg->pg_temp.osds[i] = ceph_decode_32(p);
1003 ret = __insert_pg_mapping(pg, &map->pg_temp);
1017 static int decode_pg_temp(void **p, void *end, struct ceph_osdmap *map)
1019 return __decode_pg_temp(p, end, map, false);
1022 static int decode_new_pg_temp(void **p, void *end, struct ceph_osdmap *map)
1024 return __decode_pg_temp(p, end, map, true);
1027 static int __decode_primary_temp(void **p, void *end, struct ceph_osdmap *map,
1032 ceph_decode_32_safe(p, end, n, e_inval);
1034 struct ceph_pg pgid;
1038 ret = ceph_decode_pgid(p, end, &pgid);
1042 ceph_decode_32_safe(p, end, osd, e_inval);
1044 ret = __remove_pg_mapping(&map->primary_temp, pgid);
1045 BUG_ON(!incremental && ret != -ENOENT);
1047 if (!incremental || osd != (u32)-1) {
1048 struct ceph_pg_mapping *pg;
1050 pg = kzalloc(sizeof(*pg), GFP_NOFS);
1055 pg->primary_temp.osd = osd;
1057 ret = __insert_pg_mapping(pg, &map->primary_temp);
1071 static int decode_primary_temp(void **p, void *end, struct ceph_osdmap *map)
1073 return __decode_primary_temp(p, end, map, false);
1076 static int decode_new_primary_temp(void **p, void *end,
1077 struct ceph_osdmap *map)
1079 return __decode_primary_temp(p, end, map, true);
1082 u32 ceph_get_primary_affinity(struct ceph_osdmap *map, int osd)
1084 BUG_ON(osd >= map->max_osd);
1086 if (!map->osd_primary_affinity)
1087 return CEPH_OSD_DEFAULT_PRIMARY_AFFINITY;
1089 return map->osd_primary_affinity[osd];
1092 static int set_primary_affinity(struct ceph_osdmap *map, int osd, u32 aff)
1094 BUG_ON(osd >= map->max_osd);
1096 if (!map->osd_primary_affinity) {
1099 map->osd_primary_affinity = kmalloc(map->max_osd*sizeof(u32),
1101 if (!map->osd_primary_affinity)
1104 for (i = 0; i < map->max_osd; i++)
1105 map->osd_primary_affinity[i] =
1106 CEPH_OSD_DEFAULT_PRIMARY_AFFINITY;
1109 map->osd_primary_affinity[osd] = aff;
1114 static int decode_primary_affinity(void **p, void *end,
1115 struct ceph_osdmap *map)
1119 ceph_decode_32_safe(p, end, len, e_inval);
1121 kfree(map->osd_primary_affinity);
1122 map->osd_primary_affinity = NULL;
1125 if (len != map->max_osd)
1128 ceph_decode_need(p, end, map->max_osd*sizeof(u32), e_inval);
1130 for (i = 0; i < map->max_osd; i++) {
1133 ret = set_primary_affinity(map, i, ceph_decode_32(p));
1144 static int decode_new_primary_affinity(void **p, void *end,
1145 struct ceph_osdmap *map)
1149 ceph_decode_32_safe(p, end, n, e_inval);
1154 ceph_decode_32_safe(p, end, osd, e_inval);
1155 ceph_decode_32_safe(p, end, aff, e_inval);
1157 ret = set_primary_affinity(map, osd, aff);
1161 pr_info("osd%d primary-affinity 0x%x\n", osd, aff);
1171 * decode a full map.
1173 static int osdmap_decode(void **p, void *end, struct ceph_osdmap *map)
1182 dout("%s %p to %p len %d\n", __func__, *p, end, (int)(end - *p));
1184 err = get_osdmap_client_data_v(p, end, "full", &struct_v);
1188 /* fsid, epoch, created, modified */
1189 ceph_decode_need(p, end, sizeof(map->fsid) + sizeof(u32) +
1190 sizeof(map->created) + sizeof(map->modified), e_inval);
1191 ceph_decode_copy(p, &map->fsid, sizeof(map->fsid));
1192 epoch = map->epoch = ceph_decode_32(p);
1193 ceph_decode_copy(p, &map->created, sizeof(map->created));
1194 ceph_decode_copy(p, &map->modified, sizeof(map->modified));
1197 err = decode_pools(p, end, map);
1202 err = decode_pool_names(p, end, map);
1206 ceph_decode_32_safe(p, end, map->pool_max, e_inval);
1208 ceph_decode_32_safe(p, end, map->flags, e_inval);
1211 ceph_decode_32_safe(p, end, max, e_inval);
1213 /* (re)alloc osd arrays */
1214 err = osdmap_set_max_osd(map, max);
1218 /* osd_state, osd_weight, osd_addrs->client_addr */
1219 ceph_decode_need(p, end, 3*sizeof(u32) +
1220 map->max_osd*(1 + sizeof(*map->osd_weight) +
1221 sizeof(*map->osd_addr)), e_inval);
1223 if (ceph_decode_32(p) != map->max_osd)
1226 ceph_decode_copy(p, map->osd_state, map->max_osd);
1228 if (ceph_decode_32(p) != map->max_osd)
1231 for (i = 0; i < map->max_osd; i++)
1232 map->osd_weight[i] = ceph_decode_32(p);
1234 if (ceph_decode_32(p) != map->max_osd)
1237 ceph_decode_copy(p, map->osd_addr, map->max_osd*sizeof(*map->osd_addr));
1238 for (i = 0; i < map->max_osd; i++)
1239 ceph_decode_addr(&map->osd_addr[i]);
1242 err = decode_pg_temp(p, end, map);
1247 if (struct_v >= 1) {
1248 err = decode_primary_temp(p, end, map);
1253 /* primary_affinity */
1254 if (struct_v >= 2) {
1255 err = decode_primary_affinity(p, end, map);
1259 /* XXX can this happen? */
1260 kfree(map->osd_primary_affinity);
1261 map->osd_primary_affinity = NULL;
1265 ceph_decode_32_safe(p, end, len, e_inval);
1266 err = osdmap_set_crush(map, crush_decode(*p, min(*p + len, end)));
1270 /* ignore the rest */
1273 dout("full osdmap epoch %d max_osd %d\n", map->epoch, map->max_osd);
1279 pr_err("corrupt full osdmap (%d) epoch %d off %d (%p of %p-%p)\n",
1280 err, epoch, (int)(*p - start), *p, start, end);
1281 print_hex_dump(KERN_DEBUG, "osdmap: ",
1282 DUMP_PREFIX_OFFSET, 16, 1,
1283 start, end - start, true);
1288 * Allocate and decode a full map.
1290 struct ceph_osdmap *ceph_osdmap_decode(void **p, void *end)
1292 struct ceph_osdmap *map;
1295 map = ceph_osdmap_alloc();
1297 return ERR_PTR(-ENOMEM);
1299 ret = osdmap_decode(p, end, map);
1301 ceph_osdmap_destroy(map);
1302 return ERR_PTR(ret);
1309 * Encoding order is (new_up_client, new_state, new_weight). Need to
1310 * apply in the (new_weight, new_state, new_up_client) order, because
1311 * an incremental map may look like e.g.
1313 * new_up_client: { osd=6, addr=... } # set osd_state and addr
1314 * new_state: { osd=6, xorstate=EXISTS } # clear osd_state
1316 static int decode_new_up_state_weight(void **p, void *end,
1317 struct ceph_osdmap *map)
1319 void *new_up_client;
1321 void *new_weight_end;
1325 ceph_decode_32_safe(p, end, len, e_inval);
1326 len *= sizeof(u32) + sizeof(struct ceph_entity_addr);
1327 ceph_decode_need(p, end, len, e_inval);
1331 ceph_decode_32_safe(p, end, len, e_inval);
1332 len *= sizeof(u32) + sizeof(u8);
1333 ceph_decode_need(p, end, len, e_inval);
1337 ceph_decode_32_safe(p, end, len, e_inval);
1342 ceph_decode_need(p, end, 2*sizeof(u32), e_inval);
1343 osd = ceph_decode_32(p);
1344 w = ceph_decode_32(p);
1345 BUG_ON(osd >= map->max_osd);
1346 pr_info("osd%d weight 0x%x %s\n", osd, w,
1347 w == CEPH_OSD_IN ? "(in)" :
1348 (w == CEPH_OSD_OUT ? "(out)" : ""));
1349 map->osd_weight[osd] = w;
1352 * If we are marking in, set the EXISTS, and clear the
1353 * AUTOOUT and NEW bits.
1356 map->osd_state[osd] |= CEPH_OSD_EXISTS;
1357 map->osd_state[osd] &= ~(CEPH_OSD_AUTOOUT |
1361 new_weight_end = *p;
1363 /* new_state (up/down) */
1365 len = ceph_decode_32(p);
1371 osd = ceph_decode_32(p);
1372 xorstate = ceph_decode_8(p);
1374 xorstate = CEPH_OSD_UP;
1375 BUG_ON(osd >= map->max_osd);
1376 if ((map->osd_state[osd] & CEPH_OSD_UP) &&
1377 (xorstate & CEPH_OSD_UP))
1378 pr_info("osd%d down\n", osd);
1379 if ((map->osd_state[osd] & CEPH_OSD_EXISTS) &&
1380 (xorstate & CEPH_OSD_EXISTS)) {
1381 pr_info("osd%d does not exist\n", osd);
1382 ret = set_primary_affinity(map, osd,
1383 CEPH_OSD_DEFAULT_PRIMARY_AFFINITY);
1386 memset(map->osd_addr + osd, 0, sizeof(*map->osd_addr));
1387 map->osd_state[osd] = 0;
1389 map->osd_state[osd] ^= xorstate;
1395 len = ceph_decode_32(p);
1398 struct ceph_entity_addr addr;
1400 osd = ceph_decode_32(p);
1401 ceph_decode_copy(p, &addr, sizeof(addr));
1402 ceph_decode_addr(&addr);
1403 BUG_ON(osd >= map->max_osd);
1404 pr_info("osd%d up\n", osd);
1405 map->osd_state[osd] |= CEPH_OSD_EXISTS | CEPH_OSD_UP;
1406 map->osd_addr[osd] = addr;
1409 *p = new_weight_end;
1417 * decode and apply an incremental map update.
1419 struct ceph_osdmap *osdmap_apply_incremental(void **p, void *end,
1420 struct ceph_osdmap *map)
1422 struct ceph_fsid fsid;
1424 struct ceph_timespec modified;
1428 __s32 new_flags, max;
1433 dout("%s %p to %p len %d\n", __func__, *p, end, (int)(end - *p));
1435 err = get_osdmap_client_data_v(p, end, "inc", &struct_v);
1439 /* fsid, epoch, modified, new_pool_max, new_flags */
1440 ceph_decode_need(p, end, sizeof(fsid) + sizeof(u32) + sizeof(modified) +
1441 sizeof(u64) + sizeof(u32), e_inval);
1442 ceph_decode_copy(p, &fsid, sizeof(fsid));
1443 epoch = ceph_decode_32(p);
1444 BUG_ON(epoch != map->epoch+1);
1445 ceph_decode_copy(p, &modified, sizeof(modified));
1446 new_pool_max = ceph_decode_64(p);
1447 new_flags = ceph_decode_32(p);
1450 ceph_decode_32_safe(p, end, len, e_inval);
1452 dout("apply_incremental full map len %d, %p to %p\n",
1454 return ceph_osdmap_decode(p, min(*p+len, end));
1458 ceph_decode_32_safe(p, end, len, e_inval);
1460 err = osdmap_set_crush(map,
1461 crush_decode(*p, min(*p + len, end)));
1469 map->flags = new_flags;
1470 if (new_pool_max >= 0)
1471 map->pool_max = new_pool_max;
1474 ceph_decode_32_safe(p, end, max, e_inval);
1476 err = osdmap_set_max_osd(map, max);
1482 map->modified = modified;
1485 err = decode_new_pools(p, end, map);
1489 /* new_pool_names */
1490 err = decode_pool_names(p, end, map);
1495 ceph_decode_32_safe(p, end, len, e_inval);
1497 struct ceph_pg_pool_info *pi;
1499 ceph_decode_64_safe(p, end, pool, e_inval);
1500 pi = __lookup_pg_pool(&map->pg_pools, pool);
1502 __remove_pg_pool(&map->pg_pools, pi);
1505 /* new_up_client, new_state, new_weight */
1506 err = decode_new_up_state_weight(p, end, map);
1511 err = decode_new_pg_temp(p, end, map);
1515 /* new_primary_temp */
1516 if (struct_v >= 1) {
1517 err = decode_new_primary_temp(p, end, map);
1522 /* new_primary_affinity */
1523 if (struct_v >= 2) {
1524 err = decode_new_primary_affinity(p, end, map);
1529 /* ignore the rest */
1532 dout("inc osdmap epoch %d max_osd %d\n", map->epoch, map->max_osd);
1538 pr_err("corrupt inc osdmap (%d) epoch %d off %d (%p of %p-%p)\n",
1539 err, epoch, (int)(*p - start), *p, start, end);
1540 print_hex_dump(KERN_DEBUG, "osdmap: ",
1541 DUMP_PREFIX_OFFSET, 16, 1,
1542 start, end - start, true);
1543 return ERR_PTR(err);
1546 void ceph_oloc_copy(struct ceph_object_locator *dest,
1547 const struct ceph_object_locator *src)
1549 WARN_ON(!ceph_oloc_empty(dest));
1550 WARN_ON(dest->pool_ns); /* empty() only covers ->pool */
1552 dest->pool = src->pool;
1554 dest->pool_ns = ceph_get_string(src->pool_ns);
1556 EXPORT_SYMBOL(ceph_oloc_copy);
1558 void ceph_oloc_destroy(struct ceph_object_locator *oloc)
1560 ceph_put_string(oloc->pool_ns);
1562 EXPORT_SYMBOL(ceph_oloc_destroy);
1564 void ceph_oid_copy(struct ceph_object_id *dest,
1565 const struct ceph_object_id *src)
1567 WARN_ON(!ceph_oid_empty(dest));
1569 if (src->name != src->inline_name) {
1570 /* very rare, see ceph_object_id definition */
1571 dest->name = kmalloc(src->name_len + 1,
1572 GFP_NOIO | __GFP_NOFAIL);
1575 memcpy(dest->name, src->name, src->name_len + 1);
1576 dest->name_len = src->name_len;
1578 EXPORT_SYMBOL(ceph_oid_copy);
1580 static __printf(2, 0)
1581 int oid_printf_vargs(struct ceph_object_id *oid, const char *fmt, va_list ap)
1585 WARN_ON(!ceph_oid_empty(oid));
1587 len = vsnprintf(oid->inline_name, sizeof(oid->inline_name), fmt, ap);
1588 if (len >= sizeof(oid->inline_name))
1591 oid->name_len = len;
1596 * If oid doesn't fit into inline buffer, BUG.
1598 void ceph_oid_printf(struct ceph_object_id *oid, const char *fmt, ...)
1603 BUG_ON(oid_printf_vargs(oid, fmt, ap));
1606 EXPORT_SYMBOL(ceph_oid_printf);
1608 static __printf(3, 0)
1609 int oid_aprintf_vargs(struct ceph_object_id *oid, gfp_t gfp,
1610 const char *fmt, va_list ap)
1616 len = oid_printf_vargs(oid, fmt, aq);
1620 char *external_name;
1622 external_name = kmalloc(len + 1, gfp);
1626 oid->name = external_name;
1627 WARN_ON(vsnprintf(oid->name, len + 1, fmt, ap) != len);
1628 oid->name_len = len;
1635 * If oid doesn't fit into inline buffer, allocate.
1637 int ceph_oid_aprintf(struct ceph_object_id *oid, gfp_t gfp,
1638 const char *fmt, ...)
1644 ret = oid_aprintf_vargs(oid, gfp, fmt, ap);
1649 EXPORT_SYMBOL(ceph_oid_aprintf);
1651 void ceph_oid_destroy(struct ceph_object_id *oid)
1653 if (oid->name != oid->inline_name)
1656 EXPORT_SYMBOL(ceph_oid_destroy);
1661 static bool __osds_equal(const struct ceph_osds *lhs,
1662 const struct ceph_osds *rhs)
1664 if (lhs->size == rhs->size &&
1665 !memcmp(lhs->osds, rhs->osds, rhs->size * sizeof(rhs->osds[0])))
1674 static bool osds_equal(const struct ceph_osds *lhs,
1675 const struct ceph_osds *rhs)
1677 if (__osds_equal(lhs, rhs) &&
1678 lhs->primary == rhs->primary)
1684 static bool osds_valid(const struct ceph_osds *set)
1687 if (set->size > 0 && set->primary >= 0)
1690 /* empty can_shift_osds set */
1691 if (!set->size && set->primary == -1)
1694 /* empty !can_shift_osds set - all NONE */
1695 if (set->size > 0 && set->primary == -1) {
1698 for (i = 0; i < set->size; i++) {
1699 if (set->osds[i] != CRUSH_ITEM_NONE)
1709 void ceph_osds_copy(struct ceph_osds *dest, const struct ceph_osds *src)
1711 memcpy(dest->osds, src->osds, src->size * sizeof(src->osds[0]));
1712 dest->size = src->size;
1713 dest->primary = src->primary;
1716 static bool is_split(const struct ceph_pg *pgid,
1720 int old_bits = calc_bits_of(old_pg_num);
1721 int old_mask = (1 << old_bits) - 1;
1724 WARN_ON(pgid->seed >= old_pg_num);
1725 if (new_pg_num <= old_pg_num)
1728 for (n = 1; ; n++) {
1729 int next_bit = n << (old_bits - 1);
1730 u32 s = next_bit | pgid->seed;
1732 if (s < old_pg_num || s == pgid->seed)
1734 if (s >= new_pg_num)
1737 s = ceph_stable_mod(s, old_pg_num, old_mask);
1738 if (s == pgid->seed)
1745 bool ceph_is_new_interval(const struct ceph_osds *old_acting,
1746 const struct ceph_osds *new_acting,
1747 const struct ceph_osds *old_up,
1748 const struct ceph_osds *new_up,
1755 bool old_sort_bitwise,
1756 bool new_sort_bitwise,
1757 const struct ceph_pg *pgid)
1759 return !osds_equal(old_acting, new_acting) ||
1760 !osds_equal(old_up, new_up) ||
1761 old_size != new_size ||
1762 old_min_size != new_min_size ||
1763 is_split(pgid, old_pg_num, new_pg_num) ||
1764 old_sort_bitwise != new_sort_bitwise;
1767 static int calc_pg_rank(int osd, const struct ceph_osds *acting)
1771 for (i = 0; i < acting->size; i++) {
1772 if (acting->osds[i] == osd)
1779 static bool primary_changed(const struct ceph_osds *old_acting,
1780 const struct ceph_osds *new_acting)
1782 if (!old_acting->size && !new_acting->size)
1783 return false; /* both still empty */
1785 if (!old_acting->size ^ !new_acting->size)
1786 return true; /* was empty, now not, or vice versa */
1788 if (old_acting->primary != new_acting->primary)
1789 return true; /* primary changed */
1791 if (calc_pg_rank(old_acting->primary, old_acting) !=
1792 calc_pg_rank(new_acting->primary, new_acting))
1795 return false; /* same primary (tho replicas may have changed) */
1798 bool ceph_osds_changed(const struct ceph_osds *old_acting,
1799 const struct ceph_osds *new_acting,
1802 if (primary_changed(old_acting, new_acting))
1805 if (any_change && !__osds_equal(old_acting, new_acting))
1812 * calculate file layout from given offset, length.
1813 * fill in correct oid, logical length, and object extent
1816 * for now, we write only a single su, until we can
1817 * pass a stride back to the caller.
1819 int ceph_calc_file_object_mapping(struct ceph_file_layout *layout,
1822 u64 *oxoff, u64 *oxlen)
1824 u32 osize = layout->object_size;
1825 u32 su = layout->stripe_unit;
1826 u32 sc = layout->stripe_count;
1827 u32 bl, stripeno, stripepos, objsetno;
1831 dout("mapping %llu~%llu osize %u fl_su %u\n", off, len,
1833 if (su == 0 || sc == 0)
1835 su_per_object = osize / su;
1836 if (su_per_object == 0)
1838 dout("osize %u / su %u = su_per_object %u\n", osize, su,
1841 if ((su & ~PAGE_MASK) != 0)
1844 /* bl = *off / su; */
1848 dout("off %llu / su %u = bl %u\n", off, su, bl);
1851 stripepos = bl % sc;
1852 objsetno = stripeno / su_per_object;
1854 *ono = objsetno * sc + stripepos;
1855 dout("objset %u * sc %u = ono %u\n", objsetno, sc, (unsigned int)*ono);
1857 /* *oxoff = *off % layout->fl_stripe_unit; # offset in su */
1859 su_offset = do_div(t, su);
1860 *oxoff = su_offset + (stripeno % su_per_object) * su;
1863 * Calculate the length of the extent being written to the selected
1864 * object. This is the minimum of the full length requested (len) or
1865 * the remainder of the current stripe being written to.
1867 *oxlen = min_t(u64, len, su - su_offset);
1869 dout(" obj extent %llu~%llu\n", *oxoff, *oxlen);
1873 dout(" invalid layout\n");
1879 EXPORT_SYMBOL(ceph_calc_file_object_mapping);
1882 * Map an object into a PG.
1884 * Should only be called with target_oid and target_oloc (as opposed to
1885 * base_oid and base_oloc), since tiering isn't taken into account.
1887 int ceph_object_locator_to_pg(struct ceph_osdmap *osdmap,
1888 struct ceph_object_id *oid,
1889 struct ceph_object_locator *oloc,
1890 struct ceph_pg *raw_pgid)
1892 struct ceph_pg_pool_info *pi;
1894 pi = ceph_pg_pool_by_id(osdmap, oloc->pool);
1898 if (!oloc->pool_ns) {
1899 raw_pgid->pool = oloc->pool;
1900 raw_pgid->seed = ceph_str_hash(pi->object_hash, oid->name,
1902 dout("%s %s -> raw_pgid %llu.%x\n", __func__, oid->name,
1903 raw_pgid->pool, raw_pgid->seed);
1905 char stack_buf[256];
1906 char *buf = stack_buf;
1907 int nsl = oloc->pool_ns->len;
1908 size_t total = nsl + 1 + oid->name_len;
1910 if (total > sizeof(stack_buf)) {
1911 buf = kmalloc(total, GFP_NOIO);
1915 memcpy(buf, oloc->pool_ns->str, nsl);
1917 memcpy(buf + nsl + 1, oid->name, oid->name_len);
1918 raw_pgid->pool = oloc->pool;
1919 raw_pgid->seed = ceph_str_hash(pi->object_hash, buf, total);
1920 if (buf != stack_buf)
1922 dout("%s %s ns %.*s -> raw_pgid %llu.%x\n", __func__,
1923 oid->name, nsl, oloc->pool_ns->str,
1924 raw_pgid->pool, raw_pgid->seed);
1928 EXPORT_SYMBOL(ceph_object_locator_to_pg);
1931 * Map a raw PG (full precision ps) into an actual PG.
1933 static void raw_pg_to_pg(struct ceph_pg_pool_info *pi,
1934 const struct ceph_pg *raw_pgid,
1935 struct ceph_pg *pgid)
1937 pgid->pool = raw_pgid->pool;
1938 pgid->seed = ceph_stable_mod(raw_pgid->seed, pi->pg_num,
1943 * Map a raw PG (full precision ps) into a placement ps (placement
1944 * seed). Include pool id in that value so that different pools don't
1945 * use the same seeds.
1947 static u32 raw_pg_to_pps(struct ceph_pg_pool_info *pi,
1948 const struct ceph_pg *raw_pgid)
1950 if (pi->flags & CEPH_POOL_FLAG_HASHPSPOOL) {
1951 /* hash pool id and seed so that pool PGs do not overlap */
1952 return crush_hash32_2(CRUSH_HASH_RJENKINS1,
1953 ceph_stable_mod(raw_pgid->seed,
1959 * legacy behavior: add ps and pool together. this is
1960 * not a great approach because the PGs from each pool
1961 * will overlap on top of each other: 0.5 == 1.4 ==
1964 return ceph_stable_mod(raw_pgid->seed, pi->pgp_num,
1966 (unsigned)raw_pgid->pool;
1970 static int do_crush(struct ceph_osdmap *map, int ruleno, int x,
1971 int *result, int result_max,
1972 const __u32 *weight, int weight_max)
1976 BUG_ON(result_max > CEPH_PG_MAX_SIZE);
1978 mutex_lock(&map->crush_workspace_mutex);
1979 r = crush_do_rule(map->crush, ruleno, x, result, result_max,
1980 weight, weight_max, map->crush_workspace);
1981 mutex_unlock(&map->crush_workspace_mutex);
1987 * Calculate raw set (CRUSH output) for given PG. The result may
1988 * contain nonexistent OSDs. ->primary is undefined for a raw set.
1990 * Placement seed (CRUSH input) is returned through @ppps.
1992 static void pg_to_raw_osds(struct ceph_osdmap *osdmap,
1993 struct ceph_pg_pool_info *pi,
1994 const struct ceph_pg *raw_pgid,
1995 struct ceph_osds *raw,
1998 u32 pps = raw_pg_to_pps(pi, raw_pgid);
2002 ceph_osds_init(raw);
2006 ruleno = crush_find_rule(osdmap->crush, pi->crush_ruleset, pi->type,
2009 pr_err("no crush rule: pool %lld ruleset %d type %d size %d\n",
2010 pi->id, pi->crush_ruleset, pi->type, pi->size);
2014 if (pi->size > ARRAY_SIZE(raw->osds)) {
2015 pr_err_ratelimited("pool %lld ruleset %d type %d too wide: size %d > %zu\n",
2016 pi->id, pi->crush_ruleset, pi->type, pi->size,
2017 ARRAY_SIZE(raw->osds));
2021 len = do_crush(osdmap, ruleno, pps, raw->osds, pi->size,
2022 osdmap->osd_weight, osdmap->max_osd);
2024 pr_err("error %d from crush rule %d: pool %lld ruleset %d type %d size %d\n",
2025 len, ruleno, pi->id, pi->crush_ruleset, pi->type,
2034 * Given raw set, calculate up set and up primary. By definition of an
2035 * up set, the result won't contain nonexistent or down OSDs.
2037 * This is done in-place - on return @set is the up set. If it's
2038 * empty, ->primary will remain undefined.
2040 static void raw_to_up_osds(struct ceph_osdmap *osdmap,
2041 struct ceph_pg_pool_info *pi,
2042 struct ceph_osds *set)
2046 /* ->primary is undefined for a raw set */
2047 BUG_ON(set->primary != -1);
2049 if (ceph_can_shift_osds(pi)) {
2053 for (i = 0; i < set->size; i++) {
2054 if (ceph_osd_is_down(osdmap, set->osds[i])) {
2059 set->osds[i - removed] = set->osds[i];
2061 set->size -= removed;
2063 set->primary = set->osds[0];
2065 /* set down/dne devices to NONE */
2066 for (i = set->size - 1; i >= 0; i--) {
2067 if (ceph_osd_is_down(osdmap, set->osds[i]))
2068 set->osds[i] = CRUSH_ITEM_NONE;
2070 set->primary = set->osds[i];
2075 static void apply_primary_affinity(struct ceph_osdmap *osdmap,
2076 struct ceph_pg_pool_info *pi,
2078 struct ceph_osds *up)
2084 * Do we have any non-default primary_affinity values for these
2087 if (!osdmap->osd_primary_affinity)
2090 for (i = 0; i < up->size; i++) {
2091 int osd = up->osds[i];
2093 if (osd != CRUSH_ITEM_NONE &&
2094 osdmap->osd_primary_affinity[osd] !=
2095 CEPH_OSD_DEFAULT_PRIMARY_AFFINITY) {
2103 * Pick the primary. Feed both the seed (for the pg) and the
2104 * osd into the hash/rng so that a proportional fraction of an
2105 * osd's pgs get rejected as primary.
2107 for (i = 0; i < up->size; i++) {
2108 int osd = up->osds[i];
2111 if (osd == CRUSH_ITEM_NONE)
2114 aff = osdmap->osd_primary_affinity[osd];
2115 if (aff < CEPH_OSD_MAX_PRIMARY_AFFINITY &&
2116 (crush_hash32_2(CRUSH_HASH_RJENKINS1,
2117 pps, osd) >> 16) >= aff) {
2119 * We chose not to use this primary. Note it
2120 * anyway as a fallback in case we don't pick
2121 * anyone else, but keep looking.
2133 up->primary = up->osds[pos];
2135 if (ceph_can_shift_osds(pi) && pos > 0) {
2136 /* move the new primary to the front */
2137 for (i = pos; i > 0; i--)
2138 up->osds[i] = up->osds[i - 1];
2139 up->osds[0] = up->primary;
2144 * Get pg_temp and primary_temp mappings for given PG.
2146 * Note that a PG may have none, only pg_temp, only primary_temp or
2147 * both pg_temp and primary_temp mappings. This means @temp isn't
2148 * always a valid OSD set on return: in the "only primary_temp" case,
2149 * @temp will have its ->primary >= 0 but ->size == 0.
2151 static void get_temp_osds(struct ceph_osdmap *osdmap,
2152 struct ceph_pg_pool_info *pi,
2153 const struct ceph_pg *raw_pgid,
2154 struct ceph_osds *temp)
2156 struct ceph_pg pgid;
2157 struct ceph_pg_mapping *pg;
2160 raw_pg_to_pg(pi, raw_pgid, &pgid);
2161 ceph_osds_init(temp);
2164 pg = __lookup_pg_mapping(&osdmap->pg_temp, pgid);
2166 for (i = 0; i < pg->pg_temp.len; i++) {
2167 if (ceph_osd_is_down(osdmap, pg->pg_temp.osds[i])) {
2168 if (ceph_can_shift_osds(pi))
2171 temp->osds[temp->size++] = CRUSH_ITEM_NONE;
2173 temp->osds[temp->size++] = pg->pg_temp.osds[i];
2177 /* apply pg_temp's primary */
2178 for (i = 0; i < temp->size; i++) {
2179 if (temp->osds[i] != CRUSH_ITEM_NONE) {
2180 temp->primary = temp->osds[i];
2187 pg = __lookup_pg_mapping(&osdmap->primary_temp, pgid);
2189 temp->primary = pg->primary_temp.osd;
2193 * Map a PG to its acting set as well as its up set.
2195 * Acting set is used for data mapping purposes, while up set can be
2196 * recorded for detecting interval changes and deciding whether to
2199 void ceph_pg_to_up_acting_osds(struct ceph_osdmap *osdmap,
2200 const struct ceph_pg *raw_pgid,
2201 struct ceph_osds *up,
2202 struct ceph_osds *acting)
2204 struct ceph_pg_pool_info *pi;
2207 pi = ceph_pg_pool_by_id(osdmap, raw_pgid->pool);
2210 ceph_osds_init(acting);
2214 pg_to_raw_osds(osdmap, pi, raw_pgid, up, &pps);
2215 raw_to_up_osds(osdmap, pi, up);
2216 apply_primary_affinity(osdmap, pi, pps, up);
2217 get_temp_osds(osdmap, pi, raw_pgid, acting);
2218 if (!acting->size) {
2219 memcpy(acting->osds, up->osds, up->size * sizeof(up->osds[0]));
2220 acting->size = up->size;
2221 if (acting->primary == -1)
2222 acting->primary = up->primary;
2225 WARN_ON(!osds_valid(up) || !osds_valid(acting));
2229 * Return acting primary for given PG, or -1 if none.
2231 int ceph_pg_to_acting_primary(struct ceph_osdmap *osdmap,
2232 const struct ceph_pg *raw_pgid)
2234 struct ceph_osds up, acting;
2236 ceph_pg_to_up_acting_osds(osdmap, raw_pgid, &up, &acting);
2237 return acting.primary;
2239 EXPORT_SYMBOL(ceph_pg_to_acting_primary);