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, u32 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 struct crush_choose_arg_map *alloc_choose_arg_map(void)
143 struct crush_choose_arg_map *arg_map;
145 arg_map = kzalloc(sizeof(*arg_map), GFP_NOIO);
149 RB_CLEAR_NODE(&arg_map->node);
153 static void free_choose_arg_map(struct crush_choose_arg_map *arg_map)
158 WARN_ON(!RB_EMPTY_NODE(&arg_map->node));
160 for (i = 0; i < arg_map->size; i++) {
161 struct crush_choose_arg *arg = &arg_map->args[i];
163 for (j = 0; j < arg->weight_set_size; j++)
164 kfree(arg->weight_set[j].weights);
165 kfree(arg->weight_set);
168 kfree(arg_map->args);
173 DEFINE_RB_FUNCS(choose_arg_map, struct crush_choose_arg_map, choose_args_index,
176 void clear_choose_args(struct crush_map *c)
178 while (!RB_EMPTY_ROOT(&c->choose_args)) {
179 struct crush_choose_arg_map *arg_map =
180 rb_entry(rb_first(&c->choose_args),
181 struct crush_choose_arg_map, node);
183 erase_choose_arg_map(&c->choose_args, arg_map);
184 free_choose_arg_map(arg_map);
188 static u32 *decode_array_32_alloc(void **p, void *end, u32 *plen)
194 ceph_decode_32_safe(p, end, len, e_inval);
198 a = kmalloc_array(len, sizeof(u32), GFP_NOIO);
204 ceph_decode_need(p, end, len * sizeof(u32), e_inval);
205 for (i = 0; i < len; i++)
206 a[i] = ceph_decode_32(p);
220 * Assumes @arg is zero-initialized.
222 static int decode_choose_arg(void **p, void *end, struct crush_choose_arg *arg)
226 ceph_decode_32_safe(p, end, arg->weight_set_size, e_inval);
227 if (arg->weight_set_size) {
230 arg->weight_set = kmalloc_array(arg->weight_set_size,
231 sizeof(*arg->weight_set),
233 if (!arg->weight_set)
236 for (i = 0; i < arg->weight_set_size; i++) {
237 struct crush_weight_set *w = &arg->weight_set[i];
239 w->weights = decode_array_32_alloc(p, end, &w->size);
240 if (IS_ERR(w->weights)) {
241 ret = PTR_ERR(w->weights);
248 arg->ids = decode_array_32_alloc(p, end, &arg->ids_size);
249 if (IS_ERR(arg->ids)) {
250 ret = PTR_ERR(arg->ids);
261 static int decode_choose_args(void **p, void *end, struct crush_map *c)
263 struct crush_choose_arg_map *arg_map = NULL;
264 u32 num_choose_arg_maps, num_buckets;
267 ceph_decode_32_safe(p, end, num_choose_arg_maps, e_inval);
268 while (num_choose_arg_maps--) {
269 arg_map = alloc_choose_arg_map();
275 ceph_decode_64_safe(p, end, arg_map->choose_args_index,
277 arg_map->size = c->max_buckets;
278 arg_map->args = kcalloc(arg_map->size, sizeof(*arg_map->args),
280 if (!arg_map->args) {
285 ceph_decode_32_safe(p, end, num_buckets, e_inval);
286 while (num_buckets--) {
287 struct crush_choose_arg *arg;
290 ceph_decode_32_safe(p, end, bucket_index, e_inval);
291 if (bucket_index >= arg_map->size)
294 arg = &arg_map->args[bucket_index];
295 ret = decode_choose_arg(p, end, arg);
300 insert_choose_arg_map(&c->choose_args, arg_map);
308 free_choose_arg_map(arg_map);
312 static void crush_finalize(struct crush_map *c)
316 /* Space for the array of pointers to per-bucket workspace */
317 c->working_size = sizeof(struct crush_work) +
318 c->max_buckets * sizeof(struct crush_work_bucket *);
320 for (b = 0; b < c->max_buckets; b++) {
324 switch (c->buckets[b]->alg) {
327 * The base case, permutation variables and
328 * the pointer to the permutation array.
330 c->working_size += sizeof(struct crush_work_bucket);
333 /* Every bucket has a permutation array. */
334 c->working_size += c->buckets[b]->size * sizeof(__u32);
338 static struct crush_map *crush_decode(void *pbyval, void *end)
344 void *start = pbyval;
347 dout("crush_decode %p to %p len %d\n", *p, end, (int)(end - *p));
349 c = kzalloc(sizeof(*c), GFP_NOFS);
351 return ERR_PTR(-ENOMEM);
353 c->choose_args = RB_ROOT;
355 /* set tunables to default values */
356 c->choose_local_tries = 2;
357 c->choose_local_fallback_tries = 5;
358 c->choose_total_tries = 19;
359 c->chooseleaf_descend_once = 0;
361 ceph_decode_need(p, end, 4*sizeof(u32), bad);
362 magic = ceph_decode_32(p);
363 if (magic != CRUSH_MAGIC) {
364 pr_err("crush_decode magic %x != current %x\n",
365 (unsigned int)magic, (unsigned int)CRUSH_MAGIC);
368 c->max_buckets = ceph_decode_32(p);
369 c->max_rules = ceph_decode_32(p);
370 c->max_devices = ceph_decode_32(p);
372 c->buckets = kcalloc(c->max_buckets, sizeof(*c->buckets), GFP_NOFS);
373 if (c->buckets == NULL)
375 c->rules = kcalloc(c->max_rules, sizeof(*c->rules), GFP_NOFS);
376 if (c->rules == NULL)
380 for (i = 0; i < c->max_buckets; i++) {
383 struct crush_bucket *b;
385 ceph_decode_32_safe(p, end, alg, bad);
387 c->buckets[i] = NULL;
390 dout("crush_decode bucket %d off %x %p to %p\n",
391 i, (int)(*p-start), *p, end);
394 case CRUSH_BUCKET_UNIFORM:
395 size = sizeof(struct crush_bucket_uniform);
397 case CRUSH_BUCKET_LIST:
398 size = sizeof(struct crush_bucket_list);
400 case CRUSH_BUCKET_TREE:
401 size = sizeof(struct crush_bucket_tree);
403 case CRUSH_BUCKET_STRAW:
404 size = sizeof(struct crush_bucket_straw);
406 case CRUSH_BUCKET_STRAW2:
407 size = sizeof(struct crush_bucket_straw2);
414 b = c->buckets[i] = kzalloc(size, GFP_NOFS);
418 ceph_decode_need(p, end, 4*sizeof(u32), bad);
419 b->id = ceph_decode_32(p);
420 b->type = ceph_decode_16(p);
421 b->alg = ceph_decode_8(p);
422 b->hash = ceph_decode_8(p);
423 b->weight = ceph_decode_32(p);
424 b->size = ceph_decode_32(p);
426 dout("crush_decode bucket size %d off %x %p to %p\n",
427 b->size, (int)(*p-start), *p, end);
429 b->items = kcalloc(b->size, sizeof(__s32), GFP_NOFS);
430 if (b->items == NULL)
433 ceph_decode_need(p, end, b->size*sizeof(u32), bad);
434 for (j = 0; j < b->size; j++)
435 b->items[j] = ceph_decode_32(p);
438 case CRUSH_BUCKET_UNIFORM:
439 err = crush_decode_uniform_bucket(p, end,
440 (struct crush_bucket_uniform *)b);
444 case CRUSH_BUCKET_LIST:
445 err = crush_decode_list_bucket(p, end,
446 (struct crush_bucket_list *)b);
450 case CRUSH_BUCKET_TREE:
451 err = crush_decode_tree_bucket(p, end,
452 (struct crush_bucket_tree *)b);
456 case CRUSH_BUCKET_STRAW:
457 err = crush_decode_straw_bucket(p, end,
458 (struct crush_bucket_straw *)b);
462 case CRUSH_BUCKET_STRAW2:
463 err = crush_decode_straw2_bucket(p, end,
464 (struct crush_bucket_straw2 *)b);
472 dout("rule vec is %p\n", c->rules);
473 for (i = 0; i < c->max_rules; i++) {
475 struct crush_rule *r;
478 ceph_decode_32_safe(p, end, yes, bad);
480 dout("crush_decode NO rule %d off %x %p to %p\n",
481 i, (int)(*p-start), *p, end);
486 dout("crush_decode rule %d off %x %p to %p\n",
487 i, (int)(*p-start), *p, end);
490 ceph_decode_32_safe(p, end, yes, bad);
491 #if BITS_PER_LONG == 32
493 if (yes > (ULONG_MAX - sizeof(*r))
494 / sizeof(struct crush_rule_step))
497 r = c->rules[i] = kmalloc(sizeof(*r) +
498 yes*sizeof(struct crush_rule_step),
502 dout(" rule %d is at %p\n", i, r);
504 ceph_decode_copy_safe(p, end, &r->mask, 4, bad); /* 4 u8's */
505 ceph_decode_need(p, end, r->len*3*sizeof(u32), bad);
506 for (j = 0; j < r->len; j++) {
507 r->steps[j].op = ceph_decode_32(p);
508 r->steps[j].arg1 = ceph_decode_32(p);
509 r->steps[j].arg2 = ceph_decode_32(p);
513 ceph_decode_skip_map(p, end, 32, string, bad); /* type_map */
514 ceph_decode_skip_map(p, end, 32, string, bad); /* name_map */
515 ceph_decode_skip_map(p, end, 32, string, bad); /* rule_name_map */
518 ceph_decode_need(p, end, 3*sizeof(u32), done);
519 c->choose_local_tries = ceph_decode_32(p);
520 c->choose_local_fallback_tries = ceph_decode_32(p);
521 c->choose_total_tries = ceph_decode_32(p);
522 dout("crush decode tunable choose_local_tries = %d\n",
523 c->choose_local_tries);
524 dout("crush decode tunable choose_local_fallback_tries = %d\n",
525 c->choose_local_fallback_tries);
526 dout("crush decode tunable choose_total_tries = %d\n",
527 c->choose_total_tries);
529 ceph_decode_need(p, end, sizeof(u32), done);
530 c->chooseleaf_descend_once = ceph_decode_32(p);
531 dout("crush decode tunable chooseleaf_descend_once = %d\n",
532 c->chooseleaf_descend_once);
534 ceph_decode_need(p, end, sizeof(u8), done);
535 c->chooseleaf_vary_r = ceph_decode_8(p);
536 dout("crush decode tunable chooseleaf_vary_r = %d\n",
537 c->chooseleaf_vary_r);
539 /* skip straw_calc_version, allowed_bucket_algs */
540 ceph_decode_need(p, end, sizeof(u8) + sizeof(u32), done);
541 *p += sizeof(u8) + sizeof(u32);
543 ceph_decode_need(p, end, sizeof(u8), done);
544 c->chooseleaf_stable = ceph_decode_8(p);
545 dout("crush decode tunable chooseleaf_stable = %d\n",
546 c->chooseleaf_stable);
550 ceph_decode_skip_map(p, end, 32, 32, bad);
552 ceph_decode_skip_map(p, end, 32, string, bad);
554 ceph_decode_skip_map_of_map(p, end, 32, 32, 32, bad);
558 err = decode_choose_args(p, end, c);
565 dout("crush_decode success\n");
571 dout("crush_decode fail %d\n", err);
576 int ceph_pg_compare(const struct ceph_pg *lhs, const struct ceph_pg *rhs)
578 if (lhs->pool < rhs->pool)
580 if (lhs->pool > rhs->pool)
582 if (lhs->seed < rhs->seed)
584 if (lhs->seed > rhs->seed)
590 int ceph_spg_compare(const struct ceph_spg *lhs, const struct ceph_spg *rhs)
594 ret = ceph_pg_compare(&lhs->pgid, &rhs->pgid);
598 if (lhs->shard < rhs->shard)
600 if (lhs->shard > rhs->shard)
606 static struct ceph_pg_mapping *alloc_pg_mapping(size_t payload_len)
608 struct ceph_pg_mapping *pg;
610 pg = kmalloc(sizeof(*pg) + payload_len, GFP_NOIO);
614 RB_CLEAR_NODE(&pg->node);
618 static void free_pg_mapping(struct ceph_pg_mapping *pg)
620 WARN_ON(!RB_EMPTY_NODE(&pg->node));
626 * rbtree of pg_mapping for handling pg_temp (explicit mapping of pgid
627 * to a set of osds) and primary_temp (explicit primary setting)
629 DEFINE_RB_FUNCS2(pg_mapping, struct ceph_pg_mapping, pgid, ceph_pg_compare,
630 RB_BYPTR, const struct ceph_pg *, node)
633 * rbtree of pg pool info
635 static int __insert_pg_pool(struct rb_root *root, struct ceph_pg_pool_info *new)
637 struct rb_node **p = &root->rb_node;
638 struct rb_node *parent = NULL;
639 struct ceph_pg_pool_info *pi = NULL;
643 pi = rb_entry(parent, struct ceph_pg_pool_info, node);
644 if (new->id < pi->id)
646 else if (new->id > pi->id)
652 rb_link_node(&new->node, parent, p);
653 rb_insert_color(&new->node, root);
657 static struct ceph_pg_pool_info *__lookup_pg_pool(struct rb_root *root, u64 id)
659 struct ceph_pg_pool_info *pi;
660 struct rb_node *n = root->rb_node;
663 pi = rb_entry(n, struct ceph_pg_pool_info, node);
666 else if (id > pi->id)
674 struct ceph_pg_pool_info *ceph_pg_pool_by_id(struct ceph_osdmap *map, u64 id)
676 return __lookup_pg_pool(&map->pg_pools, id);
679 const char *ceph_pg_pool_name_by_id(struct ceph_osdmap *map, u64 id)
681 struct ceph_pg_pool_info *pi;
683 if (id == CEPH_NOPOOL)
686 if (WARN_ON_ONCE(id > (u64) INT_MAX))
689 pi = __lookup_pg_pool(&map->pg_pools, (int) id);
691 return pi ? pi->name : NULL;
693 EXPORT_SYMBOL(ceph_pg_pool_name_by_id);
695 int ceph_pg_poolid_by_name(struct ceph_osdmap *map, const char *name)
699 for (rbp = rb_first(&map->pg_pools); rbp; rbp = rb_next(rbp)) {
700 struct ceph_pg_pool_info *pi =
701 rb_entry(rbp, struct ceph_pg_pool_info, node);
702 if (pi->name && strcmp(pi->name, name) == 0)
707 EXPORT_SYMBOL(ceph_pg_poolid_by_name);
709 static void __remove_pg_pool(struct rb_root *root, struct ceph_pg_pool_info *pi)
711 rb_erase(&pi->node, root);
716 static int decode_pool(void **p, void *end, struct ceph_pg_pool_info *pi)
722 ceph_decode_need(p, end, 2 + 4, bad);
723 ev = ceph_decode_8(p); /* encoding version */
724 cv = ceph_decode_8(p); /* compat version */
726 pr_warn("got v %d < 5 cv %d of ceph_pg_pool\n", ev, cv);
730 pr_warn("got v %d cv %d > 9 of ceph_pg_pool\n", ev, cv);
733 len = ceph_decode_32(p);
734 ceph_decode_need(p, end, len, bad);
737 pi->type = ceph_decode_8(p);
738 pi->size = ceph_decode_8(p);
739 pi->crush_ruleset = ceph_decode_8(p);
740 pi->object_hash = ceph_decode_8(p);
742 pi->pg_num = ceph_decode_32(p);
743 pi->pgp_num = ceph_decode_32(p);
745 *p += 4 + 4; /* skip lpg* */
746 *p += 4; /* skip last_change */
747 *p += 8 + 4; /* skip snap_seq, snap_epoch */
750 num = ceph_decode_32(p);
752 *p += 8; /* snapid key */
753 *p += 1 + 1; /* versions */
754 len = ceph_decode_32(p);
758 /* skip removed_snaps */
759 num = ceph_decode_32(p);
762 *p += 8; /* skip auid */
763 pi->flags = ceph_decode_64(p);
764 *p += 4; /* skip crash_replay_interval */
767 pi->min_size = ceph_decode_8(p);
769 pi->min_size = pi->size - pi->size / 2;
772 *p += 8 + 8; /* skip quota_max_* */
776 num = ceph_decode_32(p);
779 *p += 8; /* skip tier_of */
780 *p += 1; /* skip cache_mode */
782 pi->read_tier = ceph_decode_64(p);
783 pi->write_tier = ceph_decode_64(p);
790 /* skip properties */
791 num = ceph_decode_32(p);
793 len = ceph_decode_32(p);
795 len = ceph_decode_32(p);
801 /* skip hit_set_params */
802 *p += 1 + 1; /* versions */
803 len = ceph_decode_32(p);
806 *p += 4; /* skip hit_set_period */
807 *p += 4; /* skip hit_set_count */
811 *p += 4; /* skip stripe_width */
814 *p += 8; /* skip target_max_bytes */
815 *p += 8; /* skip target_max_objects */
816 *p += 4; /* skip cache_target_dirty_ratio_micro */
817 *p += 4; /* skip cache_target_full_ratio_micro */
818 *p += 4; /* skip cache_min_flush_age */
819 *p += 4; /* skip cache_min_evict_age */
823 /* skip erasure_code_profile */
824 len = ceph_decode_32(p);
829 * last_force_op_resend_preluminous, will be overridden if the
830 * map was encoded with RESEND_ON_SPLIT
833 pi->last_force_request_resend = ceph_decode_32(p);
835 pi->last_force_request_resend = 0;
838 *p += 4; /* skip min_read_recency_for_promote */
841 *p += 8; /* skip expected_num_objects */
844 *p += 4; /* skip cache_target_dirty_high_ratio_micro */
847 *p += 4; /* skip min_write_recency_for_promote */
850 *p += 1; /* skip use_gmt_hitset */
853 *p += 1; /* skip fast_read */
856 *p += 4; /* skip hit_set_grade_decay_rate */
857 *p += 4; /* skip hit_set_search_last_n */
862 *p += 1 + 1; /* versions */
863 len = ceph_decode_32(p);
868 pi->last_force_request_resend = ceph_decode_32(p);
870 /* ignore the rest */
880 static int decode_pool_names(void **p, void *end, struct ceph_osdmap *map)
882 struct ceph_pg_pool_info *pi;
886 ceph_decode_32_safe(p, end, num, bad);
887 dout(" %d pool names\n", num);
889 ceph_decode_64_safe(p, end, pool, bad);
890 ceph_decode_32_safe(p, end, len, bad);
891 dout(" pool %llu len %d\n", pool, len);
892 ceph_decode_need(p, end, len, bad);
893 pi = __lookup_pg_pool(&map->pg_pools, pool);
895 char *name = kstrndup(*p, len, GFP_NOFS);
901 dout(" name is %s\n", pi->name);
914 struct ceph_osdmap *ceph_osdmap_alloc(void)
916 struct ceph_osdmap *map;
918 map = kzalloc(sizeof(*map), GFP_NOIO);
922 map->pg_pools = RB_ROOT;
924 map->pg_temp = RB_ROOT;
925 map->primary_temp = RB_ROOT;
926 map->pg_upmap = RB_ROOT;
927 map->pg_upmap_items = RB_ROOT;
928 mutex_init(&map->crush_workspace_mutex);
933 void ceph_osdmap_destroy(struct ceph_osdmap *map)
935 dout("osdmap_destroy %p\n", map);
937 crush_destroy(map->crush);
938 while (!RB_EMPTY_ROOT(&map->pg_temp)) {
939 struct ceph_pg_mapping *pg =
940 rb_entry(rb_first(&map->pg_temp),
941 struct ceph_pg_mapping, node);
942 erase_pg_mapping(&map->pg_temp, pg);
945 while (!RB_EMPTY_ROOT(&map->primary_temp)) {
946 struct ceph_pg_mapping *pg =
947 rb_entry(rb_first(&map->primary_temp),
948 struct ceph_pg_mapping, node);
949 erase_pg_mapping(&map->primary_temp, pg);
952 while (!RB_EMPTY_ROOT(&map->pg_upmap)) {
953 struct ceph_pg_mapping *pg =
954 rb_entry(rb_first(&map->pg_upmap),
955 struct ceph_pg_mapping, node);
956 rb_erase(&pg->node, &map->pg_upmap);
959 while (!RB_EMPTY_ROOT(&map->pg_upmap_items)) {
960 struct ceph_pg_mapping *pg =
961 rb_entry(rb_first(&map->pg_upmap_items),
962 struct ceph_pg_mapping, node);
963 rb_erase(&pg->node, &map->pg_upmap_items);
966 while (!RB_EMPTY_ROOT(&map->pg_pools)) {
967 struct ceph_pg_pool_info *pi =
968 rb_entry(rb_first(&map->pg_pools),
969 struct ceph_pg_pool_info, node);
970 __remove_pg_pool(&map->pg_pools, pi);
972 kfree(map->osd_state);
973 kfree(map->osd_weight);
974 kfree(map->osd_addr);
975 kfree(map->osd_primary_affinity);
976 kfree(map->crush_workspace);
981 * Adjust max_osd value, (re)allocate arrays.
983 * The new elements are properly initialized.
985 static int osdmap_set_max_osd(struct ceph_osdmap *map, int max)
989 struct ceph_entity_addr *addr;
992 state = krealloc(map->osd_state, max*sizeof(*state), GFP_NOFS);
995 map->osd_state = state;
997 weight = krealloc(map->osd_weight, max*sizeof(*weight), GFP_NOFS);
1000 map->osd_weight = weight;
1002 addr = krealloc(map->osd_addr, max*sizeof(*addr), GFP_NOFS);
1005 map->osd_addr = addr;
1007 for (i = map->max_osd; i < max; i++) {
1008 map->osd_state[i] = 0;
1009 map->osd_weight[i] = CEPH_OSD_OUT;
1010 memset(map->osd_addr + i, 0, sizeof(*map->osd_addr));
1013 if (map->osd_primary_affinity) {
1016 affinity = krealloc(map->osd_primary_affinity,
1017 max*sizeof(*affinity), GFP_NOFS);
1020 map->osd_primary_affinity = affinity;
1022 for (i = map->max_osd; i < max; i++)
1023 map->osd_primary_affinity[i] =
1024 CEPH_OSD_DEFAULT_PRIMARY_AFFINITY;
1032 static int osdmap_set_crush(struct ceph_osdmap *map, struct crush_map *crush)
1038 return PTR_ERR(crush);
1040 work_size = crush_work_size(crush, CEPH_PG_MAX_SIZE);
1041 dout("%s work_size %zu bytes\n", __func__, work_size);
1042 workspace = kmalloc(work_size, GFP_NOIO);
1044 crush_destroy(crush);
1047 crush_init_workspace(crush, workspace);
1050 crush_destroy(map->crush);
1051 kfree(map->crush_workspace);
1053 map->crush_workspace = workspace;
1057 #define OSDMAP_WRAPPER_COMPAT_VER 7
1058 #define OSDMAP_CLIENT_DATA_COMPAT_VER 1
1061 * Return 0 or error. On success, *v is set to 0 for old (v6) osdmaps,
1062 * to struct_v of the client_data section for new (v7 and above)
1065 static int get_osdmap_client_data_v(void **p, void *end,
1066 const char *prefix, u8 *v)
1070 ceph_decode_8_safe(p, end, struct_v, e_inval);
1071 if (struct_v >= 7) {
1074 ceph_decode_8_safe(p, end, struct_compat, e_inval);
1075 if (struct_compat > OSDMAP_WRAPPER_COMPAT_VER) {
1076 pr_warn("got v %d cv %d > %d of %s ceph_osdmap\n",
1077 struct_v, struct_compat,
1078 OSDMAP_WRAPPER_COMPAT_VER, prefix);
1081 *p += 4; /* ignore wrapper struct_len */
1083 ceph_decode_8_safe(p, end, struct_v, e_inval);
1084 ceph_decode_8_safe(p, end, struct_compat, e_inval);
1085 if (struct_compat > OSDMAP_CLIENT_DATA_COMPAT_VER) {
1086 pr_warn("got v %d cv %d > %d of %s ceph_osdmap client data\n",
1087 struct_v, struct_compat,
1088 OSDMAP_CLIENT_DATA_COMPAT_VER, prefix);
1091 *p += 4; /* ignore client data struct_len */
1096 ceph_decode_16_safe(p, end, version, e_inval);
1098 pr_warn("got v %d < 6 of %s ceph_osdmap\n",
1103 /* old osdmap enconding */
1114 static int __decode_pools(void **p, void *end, struct ceph_osdmap *map,
1119 ceph_decode_32_safe(p, end, n, e_inval);
1121 struct ceph_pg_pool_info *pi;
1125 ceph_decode_64_safe(p, end, pool, e_inval);
1127 pi = __lookup_pg_pool(&map->pg_pools, pool);
1128 if (!incremental || !pi) {
1129 pi = kzalloc(sizeof(*pi), GFP_NOFS);
1135 ret = __insert_pg_pool(&map->pg_pools, pi);
1142 ret = decode_pool(p, end, pi);
1153 static int decode_pools(void **p, void *end, struct ceph_osdmap *map)
1155 return __decode_pools(p, end, map, false);
1158 static int decode_new_pools(void **p, void *end, struct ceph_osdmap *map)
1160 return __decode_pools(p, end, map, true);
1163 typedef struct ceph_pg_mapping *(*decode_mapping_fn_t)(void **, void *, bool);
1165 static int decode_pg_mapping(void **p, void *end, struct rb_root *mapping_root,
1166 decode_mapping_fn_t fn, bool incremental)
1170 WARN_ON(!incremental && !fn);
1172 ceph_decode_32_safe(p, end, n, e_inval);
1174 struct ceph_pg_mapping *pg;
1175 struct ceph_pg pgid;
1178 ret = ceph_decode_pgid(p, end, &pgid);
1182 pg = lookup_pg_mapping(mapping_root, &pgid);
1184 WARN_ON(!incremental);
1185 erase_pg_mapping(mapping_root, pg);
1186 free_pg_mapping(pg);
1190 pg = fn(p, end, incremental);
1195 pg->pgid = pgid; /* struct */
1196 insert_pg_mapping(mapping_root, pg);
1207 static struct ceph_pg_mapping *__decode_pg_temp(void **p, void *end,
1210 struct ceph_pg_mapping *pg;
1213 ceph_decode_32_safe(p, end, len, e_inval);
1214 if (len == 0 && incremental)
1215 return NULL; /* new_pg_temp: [] to remove */
1216 if (len > (SIZE_MAX - sizeof(*pg)) / sizeof(u32))
1217 return ERR_PTR(-EINVAL);
1219 ceph_decode_need(p, end, len * sizeof(u32), e_inval);
1220 pg = alloc_pg_mapping(len * sizeof(u32));
1222 return ERR_PTR(-ENOMEM);
1224 pg->pg_temp.len = len;
1225 for (i = 0; i < len; i++)
1226 pg->pg_temp.osds[i] = ceph_decode_32(p);
1231 return ERR_PTR(-EINVAL);
1234 static int decode_pg_temp(void **p, void *end, struct ceph_osdmap *map)
1236 return decode_pg_mapping(p, end, &map->pg_temp, __decode_pg_temp,
1240 static int decode_new_pg_temp(void **p, void *end, struct ceph_osdmap *map)
1242 return decode_pg_mapping(p, end, &map->pg_temp, __decode_pg_temp,
1246 static struct ceph_pg_mapping *__decode_primary_temp(void **p, void *end,
1249 struct ceph_pg_mapping *pg;
1252 ceph_decode_32_safe(p, end, osd, e_inval);
1253 if (osd == (u32)-1 && incremental)
1254 return NULL; /* new_primary_temp: -1 to remove */
1256 pg = alloc_pg_mapping(0);
1258 return ERR_PTR(-ENOMEM);
1260 pg->primary_temp.osd = osd;
1264 return ERR_PTR(-EINVAL);
1267 static int decode_primary_temp(void **p, void *end, struct ceph_osdmap *map)
1269 return decode_pg_mapping(p, end, &map->primary_temp,
1270 __decode_primary_temp, false);
1273 static int decode_new_primary_temp(void **p, void *end,
1274 struct ceph_osdmap *map)
1276 return decode_pg_mapping(p, end, &map->primary_temp,
1277 __decode_primary_temp, true);
1280 u32 ceph_get_primary_affinity(struct ceph_osdmap *map, int osd)
1282 BUG_ON(osd >= map->max_osd);
1284 if (!map->osd_primary_affinity)
1285 return CEPH_OSD_DEFAULT_PRIMARY_AFFINITY;
1287 return map->osd_primary_affinity[osd];
1290 static int set_primary_affinity(struct ceph_osdmap *map, int osd, u32 aff)
1292 BUG_ON(osd >= map->max_osd);
1294 if (!map->osd_primary_affinity) {
1297 map->osd_primary_affinity = kmalloc(map->max_osd*sizeof(u32),
1299 if (!map->osd_primary_affinity)
1302 for (i = 0; i < map->max_osd; i++)
1303 map->osd_primary_affinity[i] =
1304 CEPH_OSD_DEFAULT_PRIMARY_AFFINITY;
1307 map->osd_primary_affinity[osd] = aff;
1312 static int decode_primary_affinity(void **p, void *end,
1313 struct ceph_osdmap *map)
1317 ceph_decode_32_safe(p, end, len, e_inval);
1319 kfree(map->osd_primary_affinity);
1320 map->osd_primary_affinity = NULL;
1323 if (len != map->max_osd)
1326 ceph_decode_need(p, end, map->max_osd*sizeof(u32), e_inval);
1328 for (i = 0; i < map->max_osd; i++) {
1331 ret = set_primary_affinity(map, i, ceph_decode_32(p));
1342 static int decode_new_primary_affinity(void **p, void *end,
1343 struct ceph_osdmap *map)
1347 ceph_decode_32_safe(p, end, n, e_inval);
1352 ceph_decode_32_safe(p, end, osd, e_inval);
1353 ceph_decode_32_safe(p, end, aff, e_inval);
1355 ret = set_primary_affinity(map, osd, aff);
1359 pr_info("osd%d primary-affinity 0x%x\n", osd, aff);
1368 static struct ceph_pg_mapping *__decode_pg_upmap(void **p, void *end,
1371 return __decode_pg_temp(p, end, false);
1374 static int decode_pg_upmap(void **p, void *end, struct ceph_osdmap *map)
1376 return decode_pg_mapping(p, end, &map->pg_upmap, __decode_pg_upmap,
1380 static int decode_new_pg_upmap(void **p, void *end, struct ceph_osdmap *map)
1382 return decode_pg_mapping(p, end, &map->pg_upmap, __decode_pg_upmap,
1386 static int decode_old_pg_upmap(void **p, void *end, struct ceph_osdmap *map)
1388 return decode_pg_mapping(p, end, &map->pg_upmap, NULL, true);
1391 static struct ceph_pg_mapping *__decode_pg_upmap_items(void **p, void *end,
1394 struct ceph_pg_mapping *pg;
1397 ceph_decode_32_safe(p, end, len, e_inval);
1398 if (len > (SIZE_MAX - sizeof(*pg)) / (2 * sizeof(u32)))
1399 return ERR_PTR(-EINVAL);
1401 ceph_decode_need(p, end, 2 * len * sizeof(u32), e_inval);
1402 pg = kzalloc(sizeof(*pg) + 2 * len * sizeof(u32), GFP_NOIO);
1404 return ERR_PTR(-ENOMEM);
1406 pg->pg_upmap_items.len = len;
1407 for (i = 0; i < len; i++) {
1408 pg->pg_upmap_items.from_to[i][0] = ceph_decode_32(p);
1409 pg->pg_upmap_items.from_to[i][1] = ceph_decode_32(p);
1415 return ERR_PTR(-EINVAL);
1418 static int decode_pg_upmap_items(void **p, void *end, struct ceph_osdmap *map)
1420 return decode_pg_mapping(p, end, &map->pg_upmap_items,
1421 __decode_pg_upmap_items, false);
1424 static int decode_new_pg_upmap_items(void **p, void *end,
1425 struct ceph_osdmap *map)
1427 return decode_pg_mapping(p, end, &map->pg_upmap_items,
1428 __decode_pg_upmap_items, true);
1431 static int decode_old_pg_upmap_items(void **p, void *end,
1432 struct ceph_osdmap *map)
1434 return decode_pg_mapping(p, end, &map->pg_upmap_items, NULL, true);
1438 * decode a full map.
1440 static int osdmap_decode(void **p, void *end, struct ceph_osdmap *map)
1449 dout("%s %p to %p len %d\n", __func__, *p, end, (int)(end - *p));
1451 err = get_osdmap_client_data_v(p, end, "full", &struct_v);
1455 /* fsid, epoch, created, modified */
1456 ceph_decode_need(p, end, sizeof(map->fsid) + sizeof(u32) +
1457 sizeof(map->created) + sizeof(map->modified), e_inval);
1458 ceph_decode_copy(p, &map->fsid, sizeof(map->fsid));
1459 epoch = map->epoch = ceph_decode_32(p);
1460 ceph_decode_copy(p, &map->created, sizeof(map->created));
1461 ceph_decode_copy(p, &map->modified, sizeof(map->modified));
1464 err = decode_pools(p, end, map);
1469 err = decode_pool_names(p, end, map);
1473 ceph_decode_32_safe(p, end, map->pool_max, e_inval);
1475 ceph_decode_32_safe(p, end, map->flags, e_inval);
1478 ceph_decode_32_safe(p, end, max, e_inval);
1480 /* (re)alloc osd arrays */
1481 err = osdmap_set_max_osd(map, max);
1485 /* osd_state, osd_weight, osd_addrs->client_addr */
1486 ceph_decode_need(p, end, 3*sizeof(u32) +
1487 map->max_osd*((struct_v >= 5 ? sizeof(u32) :
1489 sizeof(*map->osd_weight) +
1490 sizeof(*map->osd_addr)), e_inval);
1492 if (ceph_decode_32(p) != map->max_osd)
1495 if (struct_v >= 5) {
1496 for (i = 0; i < map->max_osd; i++)
1497 map->osd_state[i] = ceph_decode_32(p);
1499 for (i = 0; i < map->max_osd; i++)
1500 map->osd_state[i] = ceph_decode_8(p);
1503 if (ceph_decode_32(p) != map->max_osd)
1506 for (i = 0; i < map->max_osd; i++)
1507 map->osd_weight[i] = ceph_decode_32(p);
1509 if (ceph_decode_32(p) != map->max_osd)
1512 ceph_decode_copy(p, map->osd_addr, map->max_osd*sizeof(*map->osd_addr));
1513 for (i = 0; i < map->max_osd; i++)
1514 ceph_decode_addr(&map->osd_addr[i]);
1517 err = decode_pg_temp(p, end, map);
1522 if (struct_v >= 1) {
1523 err = decode_primary_temp(p, end, map);
1528 /* primary_affinity */
1529 if (struct_v >= 2) {
1530 err = decode_primary_affinity(p, end, map);
1534 WARN_ON(map->osd_primary_affinity);
1538 ceph_decode_32_safe(p, end, len, e_inval);
1539 err = osdmap_set_crush(map, crush_decode(*p, min(*p + len, end)));
1544 if (struct_v >= 3) {
1545 /* erasure_code_profiles */
1546 ceph_decode_skip_map_of_map(p, end, string, string, string,
1550 if (struct_v >= 4) {
1551 err = decode_pg_upmap(p, end, map);
1555 err = decode_pg_upmap_items(p, end, map);
1559 WARN_ON(!RB_EMPTY_ROOT(&map->pg_upmap));
1560 WARN_ON(!RB_EMPTY_ROOT(&map->pg_upmap_items));
1563 /* ignore the rest */
1566 dout("full osdmap epoch %d max_osd %d\n", map->epoch, map->max_osd);
1572 pr_err("corrupt full osdmap (%d) epoch %d off %d (%p of %p-%p)\n",
1573 err, epoch, (int)(*p - start), *p, start, end);
1574 print_hex_dump(KERN_DEBUG, "osdmap: ",
1575 DUMP_PREFIX_OFFSET, 16, 1,
1576 start, end - start, true);
1581 * Allocate and decode a full map.
1583 struct ceph_osdmap *ceph_osdmap_decode(void **p, void *end)
1585 struct ceph_osdmap *map;
1588 map = ceph_osdmap_alloc();
1590 return ERR_PTR(-ENOMEM);
1592 ret = osdmap_decode(p, end, map);
1594 ceph_osdmap_destroy(map);
1595 return ERR_PTR(ret);
1602 * Encoding order is (new_up_client, new_state, new_weight). Need to
1603 * apply in the (new_weight, new_state, new_up_client) order, because
1604 * an incremental map may look like e.g.
1606 * new_up_client: { osd=6, addr=... } # set osd_state and addr
1607 * new_state: { osd=6, xorstate=EXISTS } # clear osd_state
1609 static int decode_new_up_state_weight(void **p, void *end, u8 struct_v,
1610 struct ceph_osdmap *map)
1612 void *new_up_client;
1614 void *new_weight_end;
1618 ceph_decode_32_safe(p, end, len, e_inval);
1619 len *= sizeof(u32) + sizeof(struct ceph_entity_addr);
1620 ceph_decode_need(p, end, len, e_inval);
1624 ceph_decode_32_safe(p, end, len, e_inval);
1625 len *= sizeof(u32) + (struct_v >= 5 ? sizeof(u32) : sizeof(u8));
1626 ceph_decode_need(p, end, len, e_inval);
1630 ceph_decode_32_safe(p, end, len, e_inval);
1635 ceph_decode_need(p, end, 2*sizeof(u32), e_inval);
1636 osd = ceph_decode_32(p);
1637 w = ceph_decode_32(p);
1638 BUG_ON(osd >= map->max_osd);
1639 pr_info("osd%d weight 0x%x %s\n", osd, w,
1640 w == CEPH_OSD_IN ? "(in)" :
1641 (w == CEPH_OSD_OUT ? "(out)" : ""));
1642 map->osd_weight[osd] = w;
1645 * If we are marking in, set the EXISTS, and clear the
1646 * AUTOOUT and NEW bits.
1649 map->osd_state[osd] |= CEPH_OSD_EXISTS;
1650 map->osd_state[osd] &= ~(CEPH_OSD_AUTOOUT |
1654 new_weight_end = *p;
1656 /* new_state (up/down) */
1658 len = ceph_decode_32(p);
1664 osd = ceph_decode_32(p);
1666 xorstate = ceph_decode_32(p);
1668 xorstate = ceph_decode_8(p);
1670 xorstate = CEPH_OSD_UP;
1671 BUG_ON(osd >= map->max_osd);
1672 if ((map->osd_state[osd] & CEPH_OSD_UP) &&
1673 (xorstate & CEPH_OSD_UP))
1674 pr_info("osd%d down\n", osd);
1675 if ((map->osd_state[osd] & CEPH_OSD_EXISTS) &&
1676 (xorstate & CEPH_OSD_EXISTS)) {
1677 pr_info("osd%d does not exist\n", osd);
1678 ret = set_primary_affinity(map, osd,
1679 CEPH_OSD_DEFAULT_PRIMARY_AFFINITY);
1682 memset(map->osd_addr + osd, 0, sizeof(*map->osd_addr));
1683 map->osd_state[osd] = 0;
1685 map->osd_state[osd] ^= xorstate;
1691 len = ceph_decode_32(p);
1694 struct ceph_entity_addr addr;
1696 osd = ceph_decode_32(p);
1697 ceph_decode_copy(p, &addr, sizeof(addr));
1698 ceph_decode_addr(&addr);
1699 BUG_ON(osd >= map->max_osd);
1700 pr_info("osd%d up\n", osd);
1701 map->osd_state[osd] |= CEPH_OSD_EXISTS | CEPH_OSD_UP;
1702 map->osd_addr[osd] = addr;
1705 *p = new_weight_end;
1713 * decode and apply an incremental map update.
1715 struct ceph_osdmap *osdmap_apply_incremental(void **p, void *end,
1716 struct ceph_osdmap *map)
1718 struct ceph_fsid fsid;
1720 struct ceph_timespec modified;
1724 __s32 new_flags, max;
1729 dout("%s %p to %p len %d\n", __func__, *p, end, (int)(end - *p));
1731 err = get_osdmap_client_data_v(p, end, "inc", &struct_v);
1735 /* fsid, epoch, modified, new_pool_max, new_flags */
1736 ceph_decode_need(p, end, sizeof(fsid) + sizeof(u32) + sizeof(modified) +
1737 sizeof(u64) + sizeof(u32), e_inval);
1738 ceph_decode_copy(p, &fsid, sizeof(fsid));
1739 epoch = ceph_decode_32(p);
1740 BUG_ON(epoch != map->epoch+1);
1741 ceph_decode_copy(p, &modified, sizeof(modified));
1742 new_pool_max = ceph_decode_64(p);
1743 new_flags = ceph_decode_32(p);
1746 ceph_decode_32_safe(p, end, len, e_inval);
1748 dout("apply_incremental full map len %d, %p to %p\n",
1750 return ceph_osdmap_decode(p, min(*p+len, end));
1754 ceph_decode_32_safe(p, end, len, e_inval);
1756 err = osdmap_set_crush(map,
1757 crush_decode(*p, min(*p + len, end)));
1765 map->flags = new_flags;
1766 if (new_pool_max >= 0)
1767 map->pool_max = new_pool_max;
1770 ceph_decode_32_safe(p, end, max, e_inval);
1772 err = osdmap_set_max_osd(map, max);
1778 map->modified = modified;
1781 err = decode_new_pools(p, end, map);
1785 /* new_pool_names */
1786 err = decode_pool_names(p, end, map);
1791 ceph_decode_32_safe(p, end, len, e_inval);
1793 struct ceph_pg_pool_info *pi;
1795 ceph_decode_64_safe(p, end, pool, e_inval);
1796 pi = __lookup_pg_pool(&map->pg_pools, pool);
1798 __remove_pg_pool(&map->pg_pools, pi);
1801 /* new_up_client, new_state, new_weight */
1802 err = decode_new_up_state_weight(p, end, struct_v, map);
1807 err = decode_new_pg_temp(p, end, map);
1811 /* new_primary_temp */
1812 if (struct_v >= 1) {
1813 err = decode_new_primary_temp(p, end, map);
1818 /* new_primary_affinity */
1819 if (struct_v >= 2) {
1820 err = decode_new_primary_affinity(p, end, map);
1825 if (struct_v >= 3) {
1826 /* new_erasure_code_profiles */
1827 ceph_decode_skip_map_of_map(p, end, string, string, string,
1829 /* old_erasure_code_profiles */
1830 ceph_decode_skip_set(p, end, string, bad);
1833 if (struct_v >= 4) {
1834 err = decode_new_pg_upmap(p, end, map);
1838 err = decode_old_pg_upmap(p, end, map);
1842 err = decode_new_pg_upmap_items(p, end, map);
1846 err = decode_old_pg_upmap_items(p, end, map);
1851 /* ignore the rest */
1854 dout("inc osdmap epoch %d max_osd %d\n", map->epoch, map->max_osd);
1860 pr_err("corrupt inc osdmap (%d) epoch %d off %d (%p of %p-%p)\n",
1861 err, epoch, (int)(*p - start), *p, start, end);
1862 print_hex_dump(KERN_DEBUG, "osdmap: ",
1863 DUMP_PREFIX_OFFSET, 16, 1,
1864 start, end - start, true);
1865 return ERR_PTR(err);
1868 void ceph_oloc_copy(struct ceph_object_locator *dest,
1869 const struct ceph_object_locator *src)
1871 ceph_oloc_destroy(dest);
1873 dest->pool = src->pool;
1875 dest->pool_ns = ceph_get_string(src->pool_ns);
1877 dest->pool_ns = NULL;
1879 EXPORT_SYMBOL(ceph_oloc_copy);
1881 void ceph_oloc_destroy(struct ceph_object_locator *oloc)
1883 ceph_put_string(oloc->pool_ns);
1885 EXPORT_SYMBOL(ceph_oloc_destroy);
1887 void ceph_oid_copy(struct ceph_object_id *dest,
1888 const struct ceph_object_id *src)
1890 ceph_oid_destroy(dest);
1892 if (src->name != src->inline_name) {
1893 /* very rare, see ceph_object_id definition */
1894 dest->name = kmalloc(src->name_len + 1,
1895 GFP_NOIO | __GFP_NOFAIL);
1897 dest->name = dest->inline_name;
1899 memcpy(dest->name, src->name, src->name_len + 1);
1900 dest->name_len = src->name_len;
1902 EXPORT_SYMBOL(ceph_oid_copy);
1904 static __printf(2, 0)
1905 int oid_printf_vargs(struct ceph_object_id *oid, const char *fmt, va_list ap)
1909 WARN_ON(!ceph_oid_empty(oid));
1911 len = vsnprintf(oid->inline_name, sizeof(oid->inline_name), fmt, ap);
1912 if (len >= sizeof(oid->inline_name))
1915 oid->name_len = len;
1920 * If oid doesn't fit into inline buffer, BUG.
1922 void ceph_oid_printf(struct ceph_object_id *oid, const char *fmt, ...)
1927 BUG_ON(oid_printf_vargs(oid, fmt, ap));
1930 EXPORT_SYMBOL(ceph_oid_printf);
1932 static __printf(3, 0)
1933 int oid_aprintf_vargs(struct ceph_object_id *oid, gfp_t gfp,
1934 const char *fmt, va_list ap)
1940 len = oid_printf_vargs(oid, fmt, aq);
1944 char *external_name;
1946 external_name = kmalloc(len + 1, gfp);
1950 oid->name = external_name;
1951 WARN_ON(vsnprintf(oid->name, len + 1, fmt, ap) != len);
1952 oid->name_len = len;
1959 * If oid doesn't fit into inline buffer, allocate.
1961 int ceph_oid_aprintf(struct ceph_object_id *oid, gfp_t gfp,
1962 const char *fmt, ...)
1968 ret = oid_aprintf_vargs(oid, gfp, fmt, ap);
1973 EXPORT_SYMBOL(ceph_oid_aprintf);
1975 void ceph_oid_destroy(struct ceph_object_id *oid)
1977 if (oid->name != oid->inline_name)
1980 EXPORT_SYMBOL(ceph_oid_destroy);
1985 static bool __osds_equal(const struct ceph_osds *lhs,
1986 const struct ceph_osds *rhs)
1988 if (lhs->size == rhs->size &&
1989 !memcmp(lhs->osds, rhs->osds, rhs->size * sizeof(rhs->osds[0])))
1998 static bool osds_equal(const struct ceph_osds *lhs,
1999 const struct ceph_osds *rhs)
2001 if (__osds_equal(lhs, rhs) &&
2002 lhs->primary == rhs->primary)
2008 static bool osds_valid(const struct ceph_osds *set)
2011 if (set->size > 0 && set->primary >= 0)
2014 /* empty can_shift_osds set */
2015 if (!set->size && set->primary == -1)
2018 /* empty !can_shift_osds set - all NONE */
2019 if (set->size > 0 && set->primary == -1) {
2022 for (i = 0; i < set->size; i++) {
2023 if (set->osds[i] != CRUSH_ITEM_NONE)
2033 void ceph_osds_copy(struct ceph_osds *dest, const struct ceph_osds *src)
2035 memcpy(dest->osds, src->osds, src->size * sizeof(src->osds[0]));
2036 dest->size = src->size;
2037 dest->primary = src->primary;
2040 bool ceph_pg_is_split(const struct ceph_pg *pgid, u32 old_pg_num,
2043 int old_bits = calc_bits_of(old_pg_num);
2044 int old_mask = (1 << old_bits) - 1;
2047 WARN_ON(pgid->seed >= old_pg_num);
2048 if (new_pg_num <= old_pg_num)
2051 for (n = 1; ; n++) {
2052 int next_bit = n << (old_bits - 1);
2053 u32 s = next_bit | pgid->seed;
2055 if (s < old_pg_num || s == pgid->seed)
2057 if (s >= new_pg_num)
2060 s = ceph_stable_mod(s, old_pg_num, old_mask);
2061 if (s == pgid->seed)
2068 bool ceph_is_new_interval(const struct ceph_osds *old_acting,
2069 const struct ceph_osds *new_acting,
2070 const struct ceph_osds *old_up,
2071 const struct ceph_osds *new_up,
2078 bool old_sort_bitwise,
2079 bool new_sort_bitwise,
2080 const struct ceph_pg *pgid)
2082 return !osds_equal(old_acting, new_acting) ||
2083 !osds_equal(old_up, new_up) ||
2084 old_size != new_size ||
2085 old_min_size != new_min_size ||
2086 ceph_pg_is_split(pgid, old_pg_num, new_pg_num) ||
2087 old_sort_bitwise != new_sort_bitwise;
2090 static int calc_pg_rank(int osd, const struct ceph_osds *acting)
2094 for (i = 0; i < acting->size; i++) {
2095 if (acting->osds[i] == osd)
2102 static bool primary_changed(const struct ceph_osds *old_acting,
2103 const struct ceph_osds *new_acting)
2105 if (!old_acting->size && !new_acting->size)
2106 return false; /* both still empty */
2108 if (!old_acting->size ^ !new_acting->size)
2109 return true; /* was empty, now not, or vice versa */
2111 if (old_acting->primary != new_acting->primary)
2112 return true; /* primary changed */
2114 if (calc_pg_rank(old_acting->primary, old_acting) !=
2115 calc_pg_rank(new_acting->primary, new_acting))
2118 return false; /* same primary (tho replicas may have changed) */
2121 bool ceph_osds_changed(const struct ceph_osds *old_acting,
2122 const struct ceph_osds *new_acting,
2125 if (primary_changed(old_acting, new_acting))
2128 if (any_change && !__osds_equal(old_acting, new_acting))
2135 * calculate file layout from given offset, length.
2136 * fill in correct oid, logical length, and object extent
2139 * for now, we write only a single su, until we can
2140 * pass a stride back to the caller.
2142 int ceph_calc_file_object_mapping(struct ceph_file_layout *layout,
2145 u64 *oxoff, u64 *oxlen)
2147 u32 osize = layout->object_size;
2148 u32 su = layout->stripe_unit;
2149 u32 sc = layout->stripe_count;
2150 u32 bl, stripeno, stripepos, objsetno;
2154 dout("mapping %llu~%llu osize %u fl_su %u\n", off, len,
2156 if (su == 0 || sc == 0)
2158 su_per_object = osize / su;
2159 if (su_per_object == 0)
2161 dout("osize %u / su %u = su_per_object %u\n", osize, su,
2164 if ((su & ~PAGE_MASK) != 0)
2167 /* bl = *off / su; */
2171 dout("off %llu / su %u = bl %u\n", off, su, bl);
2174 stripepos = bl % sc;
2175 objsetno = stripeno / su_per_object;
2177 *ono = objsetno * sc + stripepos;
2178 dout("objset %u * sc %u = ono %u\n", objsetno, sc, (unsigned int)*ono);
2180 /* *oxoff = *off % layout->fl_stripe_unit; # offset in su */
2182 su_offset = do_div(t, su);
2183 *oxoff = su_offset + (stripeno % su_per_object) * su;
2186 * Calculate the length of the extent being written to the selected
2187 * object. This is the minimum of the full length requested (len) or
2188 * the remainder of the current stripe being written to.
2190 *oxlen = min_t(u64, len, su - su_offset);
2192 dout(" obj extent %llu~%llu\n", *oxoff, *oxlen);
2196 dout(" invalid layout\n");
2202 EXPORT_SYMBOL(ceph_calc_file_object_mapping);
2205 * Map an object into a PG.
2207 * Should only be called with target_oid and target_oloc (as opposed to
2208 * base_oid and base_oloc), since tiering isn't taken into account.
2210 int __ceph_object_locator_to_pg(struct ceph_pg_pool_info *pi,
2211 const struct ceph_object_id *oid,
2212 const struct ceph_object_locator *oloc,
2213 struct ceph_pg *raw_pgid)
2215 WARN_ON(pi->id != oloc->pool);
2217 if (!oloc->pool_ns) {
2218 raw_pgid->pool = oloc->pool;
2219 raw_pgid->seed = ceph_str_hash(pi->object_hash, oid->name,
2221 dout("%s %s -> raw_pgid %llu.%x\n", __func__, oid->name,
2222 raw_pgid->pool, raw_pgid->seed);
2224 char stack_buf[256];
2225 char *buf = stack_buf;
2226 int nsl = oloc->pool_ns->len;
2227 size_t total = nsl + 1 + oid->name_len;
2229 if (total > sizeof(stack_buf)) {
2230 buf = kmalloc(total, GFP_NOIO);
2234 memcpy(buf, oloc->pool_ns->str, nsl);
2236 memcpy(buf + nsl + 1, oid->name, oid->name_len);
2237 raw_pgid->pool = oloc->pool;
2238 raw_pgid->seed = ceph_str_hash(pi->object_hash, buf, total);
2239 if (buf != stack_buf)
2241 dout("%s %s ns %.*s -> raw_pgid %llu.%x\n", __func__,
2242 oid->name, nsl, oloc->pool_ns->str,
2243 raw_pgid->pool, raw_pgid->seed);
2248 int ceph_object_locator_to_pg(struct ceph_osdmap *osdmap,
2249 const struct ceph_object_id *oid,
2250 const struct ceph_object_locator *oloc,
2251 struct ceph_pg *raw_pgid)
2253 struct ceph_pg_pool_info *pi;
2255 pi = ceph_pg_pool_by_id(osdmap, oloc->pool);
2259 return __ceph_object_locator_to_pg(pi, oid, oloc, raw_pgid);
2261 EXPORT_SYMBOL(ceph_object_locator_to_pg);
2264 * Map a raw PG (full precision ps) into an actual PG.
2266 static void raw_pg_to_pg(struct ceph_pg_pool_info *pi,
2267 const struct ceph_pg *raw_pgid,
2268 struct ceph_pg *pgid)
2270 pgid->pool = raw_pgid->pool;
2271 pgid->seed = ceph_stable_mod(raw_pgid->seed, pi->pg_num,
2276 * Map a raw PG (full precision ps) into a placement ps (placement
2277 * seed). Include pool id in that value so that different pools don't
2278 * use the same seeds.
2280 static u32 raw_pg_to_pps(struct ceph_pg_pool_info *pi,
2281 const struct ceph_pg *raw_pgid)
2283 if (pi->flags & CEPH_POOL_FLAG_HASHPSPOOL) {
2284 /* hash pool id and seed so that pool PGs do not overlap */
2285 return crush_hash32_2(CRUSH_HASH_RJENKINS1,
2286 ceph_stable_mod(raw_pgid->seed,
2292 * legacy behavior: add ps and pool together. this is
2293 * not a great approach because the PGs from each pool
2294 * will overlap on top of each other: 0.5 == 1.4 ==
2297 return ceph_stable_mod(raw_pgid->seed, pi->pgp_num,
2299 (unsigned)raw_pgid->pool;
2303 static int do_crush(struct ceph_osdmap *map, int ruleno, int x,
2304 int *result, int result_max,
2305 const __u32 *weight, int weight_max,
2306 u64 choose_args_index)
2308 struct crush_choose_arg_map *arg_map;
2311 BUG_ON(result_max > CEPH_PG_MAX_SIZE);
2313 arg_map = lookup_choose_arg_map(&map->crush->choose_args,
2316 mutex_lock(&map->crush_workspace_mutex);
2317 r = crush_do_rule(map->crush, ruleno, x, result, result_max,
2318 weight, weight_max, map->crush_workspace,
2319 arg_map ? arg_map->args : NULL);
2320 mutex_unlock(&map->crush_workspace_mutex);
2325 static void remove_nonexistent_osds(struct ceph_osdmap *osdmap,
2326 struct ceph_pg_pool_info *pi,
2327 struct ceph_osds *set)
2331 if (ceph_can_shift_osds(pi)) {
2335 for (i = 0; i < set->size; i++) {
2336 if (!ceph_osd_exists(osdmap, set->osds[i])) {
2341 set->osds[i - removed] = set->osds[i];
2343 set->size -= removed;
2345 /* set dne devices to NONE */
2346 for (i = 0; i < set->size; i++) {
2347 if (!ceph_osd_exists(osdmap, set->osds[i]))
2348 set->osds[i] = CRUSH_ITEM_NONE;
2354 * Calculate raw set (CRUSH output) for given PG and filter out
2355 * nonexistent OSDs. ->primary is undefined for a raw set.
2357 * Placement seed (CRUSH input) is returned through @ppps.
2359 static void pg_to_raw_osds(struct ceph_osdmap *osdmap,
2360 struct ceph_pg_pool_info *pi,
2361 const struct ceph_pg *raw_pgid,
2362 struct ceph_osds *raw,
2365 u32 pps = raw_pg_to_pps(pi, raw_pgid);
2369 ceph_osds_init(raw);
2373 ruleno = crush_find_rule(osdmap->crush, pi->crush_ruleset, pi->type,
2376 pr_err("no crush rule: pool %lld ruleset %d type %d size %d\n",
2377 pi->id, pi->crush_ruleset, pi->type, pi->size);
2381 if (pi->size > ARRAY_SIZE(raw->osds)) {
2382 pr_err_ratelimited("pool %lld ruleset %d type %d too wide: size %d > %zu\n",
2383 pi->id, pi->crush_ruleset, pi->type, pi->size,
2384 ARRAY_SIZE(raw->osds));
2388 len = do_crush(osdmap, ruleno, pps, raw->osds, pi->size,
2389 osdmap->osd_weight, osdmap->max_osd, pi->id);
2391 pr_err("error %d from crush rule %d: pool %lld ruleset %d type %d size %d\n",
2392 len, ruleno, pi->id, pi->crush_ruleset, pi->type,
2398 remove_nonexistent_osds(osdmap, pi, raw);
2401 /* apply pg_upmap[_items] mappings */
2402 static void apply_upmap(struct ceph_osdmap *osdmap,
2403 const struct ceph_pg *pgid,
2404 struct ceph_osds *raw)
2406 struct ceph_pg_mapping *pg;
2409 pg = lookup_pg_mapping(&osdmap->pg_upmap, pgid);
2411 /* make sure targets aren't marked out */
2412 for (i = 0; i < pg->pg_upmap.len; i++) {
2413 int osd = pg->pg_upmap.osds[i];
2415 if (osd != CRUSH_ITEM_NONE &&
2416 osd < osdmap->max_osd &&
2417 osdmap->osd_weight[osd] == 0) {
2418 /* reject/ignore explicit mapping */
2422 for (i = 0; i < pg->pg_upmap.len; i++)
2423 raw->osds[i] = pg->pg_upmap.osds[i];
2424 raw->size = pg->pg_upmap.len;
2428 pg = lookup_pg_mapping(&osdmap->pg_upmap_items, pgid);
2431 * Note: this approach does not allow a bidirectional swap,
2432 * e.g., [[1,2],[2,1]] applied to [0,1,2] -> [0,2,1].
2434 for (i = 0; i < pg->pg_upmap_items.len; i++) {
2435 int from = pg->pg_upmap_items.from_to[i][0];
2436 int to = pg->pg_upmap_items.from_to[i][1];
2438 bool exists = false;
2440 /* make sure replacement doesn't already appear */
2441 for (j = 0; j < raw->size; j++) {
2442 int osd = raw->osds[j];
2448 /* ignore mapping if target is marked out */
2449 if (osd == from && pos < 0 &&
2450 !(to != CRUSH_ITEM_NONE &&
2451 to < osdmap->max_osd &&
2452 osdmap->osd_weight[to] == 0)) {
2456 if (!exists && pos >= 0) {
2457 raw->osds[pos] = to;
2465 * Given raw set, calculate up set and up primary. By definition of an
2466 * up set, the result won't contain nonexistent or down OSDs.
2468 * This is done in-place - on return @set is the up set. If it's
2469 * empty, ->primary will remain undefined.
2471 static void raw_to_up_osds(struct ceph_osdmap *osdmap,
2472 struct ceph_pg_pool_info *pi,
2473 struct ceph_osds *set)
2477 /* ->primary is undefined for a raw set */
2478 BUG_ON(set->primary != -1);
2480 if (ceph_can_shift_osds(pi)) {
2484 for (i = 0; i < set->size; i++) {
2485 if (ceph_osd_is_down(osdmap, set->osds[i])) {
2490 set->osds[i - removed] = set->osds[i];
2492 set->size -= removed;
2494 set->primary = set->osds[0];
2496 /* set down/dne devices to NONE */
2497 for (i = set->size - 1; i >= 0; i--) {
2498 if (ceph_osd_is_down(osdmap, set->osds[i]))
2499 set->osds[i] = CRUSH_ITEM_NONE;
2501 set->primary = set->osds[i];
2506 static void apply_primary_affinity(struct ceph_osdmap *osdmap,
2507 struct ceph_pg_pool_info *pi,
2509 struct ceph_osds *up)
2515 * Do we have any non-default primary_affinity values for these
2518 if (!osdmap->osd_primary_affinity)
2521 for (i = 0; i < up->size; i++) {
2522 int osd = up->osds[i];
2524 if (osd != CRUSH_ITEM_NONE &&
2525 osdmap->osd_primary_affinity[osd] !=
2526 CEPH_OSD_DEFAULT_PRIMARY_AFFINITY) {
2534 * Pick the primary. Feed both the seed (for the pg) and the
2535 * osd into the hash/rng so that a proportional fraction of an
2536 * osd's pgs get rejected as primary.
2538 for (i = 0; i < up->size; i++) {
2539 int osd = up->osds[i];
2542 if (osd == CRUSH_ITEM_NONE)
2545 aff = osdmap->osd_primary_affinity[osd];
2546 if (aff < CEPH_OSD_MAX_PRIMARY_AFFINITY &&
2547 (crush_hash32_2(CRUSH_HASH_RJENKINS1,
2548 pps, osd) >> 16) >= aff) {
2550 * We chose not to use this primary. Note it
2551 * anyway as a fallback in case we don't pick
2552 * anyone else, but keep looking.
2564 up->primary = up->osds[pos];
2566 if (ceph_can_shift_osds(pi) && pos > 0) {
2567 /* move the new primary to the front */
2568 for (i = pos; i > 0; i--)
2569 up->osds[i] = up->osds[i - 1];
2570 up->osds[0] = up->primary;
2575 * Get pg_temp and primary_temp mappings for given PG.
2577 * Note that a PG may have none, only pg_temp, only primary_temp or
2578 * both pg_temp and primary_temp mappings. This means @temp isn't
2579 * always a valid OSD set on return: in the "only primary_temp" case,
2580 * @temp will have its ->primary >= 0 but ->size == 0.
2582 static void get_temp_osds(struct ceph_osdmap *osdmap,
2583 struct ceph_pg_pool_info *pi,
2584 const struct ceph_pg *pgid,
2585 struct ceph_osds *temp)
2587 struct ceph_pg_mapping *pg;
2590 ceph_osds_init(temp);
2593 pg = lookup_pg_mapping(&osdmap->pg_temp, pgid);
2595 for (i = 0; i < pg->pg_temp.len; i++) {
2596 if (ceph_osd_is_down(osdmap, pg->pg_temp.osds[i])) {
2597 if (ceph_can_shift_osds(pi))
2600 temp->osds[temp->size++] = CRUSH_ITEM_NONE;
2602 temp->osds[temp->size++] = pg->pg_temp.osds[i];
2606 /* apply pg_temp's primary */
2607 for (i = 0; i < temp->size; i++) {
2608 if (temp->osds[i] != CRUSH_ITEM_NONE) {
2609 temp->primary = temp->osds[i];
2616 pg = lookup_pg_mapping(&osdmap->primary_temp, pgid);
2618 temp->primary = pg->primary_temp.osd;
2622 * Map a PG to its acting set as well as its up set.
2624 * Acting set is used for data mapping purposes, while up set can be
2625 * recorded for detecting interval changes and deciding whether to
2628 void ceph_pg_to_up_acting_osds(struct ceph_osdmap *osdmap,
2629 struct ceph_pg_pool_info *pi,
2630 const struct ceph_pg *raw_pgid,
2631 struct ceph_osds *up,
2632 struct ceph_osds *acting)
2634 struct ceph_pg pgid;
2637 WARN_ON(pi->id != raw_pgid->pool);
2638 raw_pg_to_pg(pi, raw_pgid, &pgid);
2640 pg_to_raw_osds(osdmap, pi, raw_pgid, up, &pps);
2641 apply_upmap(osdmap, &pgid, up);
2642 raw_to_up_osds(osdmap, pi, up);
2643 apply_primary_affinity(osdmap, pi, pps, up);
2644 get_temp_osds(osdmap, pi, &pgid, acting);
2645 if (!acting->size) {
2646 memcpy(acting->osds, up->osds, up->size * sizeof(up->osds[0]));
2647 acting->size = up->size;
2648 if (acting->primary == -1)
2649 acting->primary = up->primary;
2651 WARN_ON(!osds_valid(up) || !osds_valid(acting));
2654 bool ceph_pg_to_primary_shard(struct ceph_osdmap *osdmap,
2655 struct ceph_pg_pool_info *pi,
2656 const struct ceph_pg *raw_pgid,
2657 struct ceph_spg *spgid)
2659 struct ceph_pg pgid;
2660 struct ceph_osds up, acting;
2663 WARN_ON(pi->id != raw_pgid->pool);
2664 raw_pg_to_pg(pi, raw_pgid, &pgid);
2666 if (ceph_can_shift_osds(pi)) {
2667 spgid->pgid = pgid; /* struct */
2668 spgid->shard = CEPH_SPG_NOSHARD;
2672 ceph_pg_to_up_acting_osds(osdmap, pi, &pgid, &up, &acting);
2673 for (i = 0; i < acting.size; i++) {
2674 if (acting.osds[i] == acting.primary) {
2675 spgid->pgid = pgid; /* struct */
2685 * Return acting primary for given PG, or -1 if none.
2687 int ceph_pg_to_acting_primary(struct ceph_osdmap *osdmap,
2688 const struct ceph_pg *raw_pgid)
2690 struct ceph_pg_pool_info *pi;
2691 struct ceph_osds up, acting;
2693 pi = ceph_pg_pool_by_id(osdmap, raw_pgid->pool);
2697 ceph_pg_to_up_acting_osds(osdmap, pi, raw_pgid, &up, &acting);
2698 return acting.primary;
2700 EXPORT_SYMBOL(ceph_pg_to_acting_primary);