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);
413 b = c->buckets[i] = kzalloc(size, GFP_NOFS);
417 ceph_decode_need(p, end, 4*sizeof(u32), bad);
418 b->id = ceph_decode_32(p);
419 b->type = ceph_decode_16(p);
420 b->alg = ceph_decode_8(p);
421 b->hash = ceph_decode_8(p);
422 b->weight = ceph_decode_32(p);
423 b->size = ceph_decode_32(p);
425 dout("crush_decode bucket size %d off %x %p to %p\n",
426 b->size, (int)(*p-start), *p, end);
428 b->items = kcalloc(b->size, sizeof(__s32), GFP_NOFS);
429 if (b->items == NULL)
432 ceph_decode_need(p, end, b->size*sizeof(u32), bad);
433 for (j = 0; j < b->size; j++)
434 b->items[j] = ceph_decode_32(p);
437 case CRUSH_BUCKET_UNIFORM:
438 err = crush_decode_uniform_bucket(p, end,
439 (struct crush_bucket_uniform *)b);
443 case CRUSH_BUCKET_LIST:
444 err = crush_decode_list_bucket(p, end,
445 (struct crush_bucket_list *)b);
449 case CRUSH_BUCKET_TREE:
450 err = crush_decode_tree_bucket(p, end,
451 (struct crush_bucket_tree *)b);
455 case CRUSH_BUCKET_STRAW:
456 err = crush_decode_straw_bucket(p, end,
457 (struct crush_bucket_straw *)b);
461 case CRUSH_BUCKET_STRAW2:
462 err = crush_decode_straw2_bucket(p, end,
463 (struct crush_bucket_straw2 *)b);
471 dout("rule vec is %p\n", c->rules);
472 for (i = 0; i < c->max_rules; i++) {
474 struct crush_rule *r;
476 ceph_decode_32_safe(p, end, yes, bad);
478 dout("crush_decode NO rule %d off %x %p to %p\n",
479 i, (int)(*p-start), *p, end);
484 dout("crush_decode rule %d off %x %p to %p\n",
485 i, (int)(*p-start), *p, end);
488 ceph_decode_32_safe(p, end, yes, bad);
489 #if BITS_PER_LONG == 32
490 if (yes > (ULONG_MAX - sizeof(*r))
491 / sizeof(struct crush_rule_step))
494 r = c->rules[i] = kmalloc(sizeof(*r) +
495 yes*sizeof(struct crush_rule_step),
499 dout(" rule %d is at %p\n", i, r);
501 ceph_decode_copy_safe(p, end, &r->mask, 4, bad); /* 4 u8's */
502 ceph_decode_need(p, end, r->len*3*sizeof(u32), bad);
503 for (j = 0; j < r->len; j++) {
504 r->steps[j].op = ceph_decode_32(p);
505 r->steps[j].arg1 = ceph_decode_32(p);
506 r->steps[j].arg2 = ceph_decode_32(p);
510 ceph_decode_skip_map(p, end, 32, string, bad); /* type_map */
511 ceph_decode_skip_map(p, end, 32, string, bad); /* name_map */
512 ceph_decode_skip_map(p, end, 32, string, bad); /* rule_name_map */
515 ceph_decode_need(p, end, 3*sizeof(u32), done);
516 c->choose_local_tries = ceph_decode_32(p);
517 c->choose_local_fallback_tries = ceph_decode_32(p);
518 c->choose_total_tries = ceph_decode_32(p);
519 dout("crush decode tunable choose_local_tries = %d\n",
520 c->choose_local_tries);
521 dout("crush decode tunable choose_local_fallback_tries = %d\n",
522 c->choose_local_fallback_tries);
523 dout("crush decode tunable choose_total_tries = %d\n",
524 c->choose_total_tries);
526 ceph_decode_need(p, end, sizeof(u32), done);
527 c->chooseleaf_descend_once = ceph_decode_32(p);
528 dout("crush decode tunable chooseleaf_descend_once = %d\n",
529 c->chooseleaf_descend_once);
531 ceph_decode_need(p, end, sizeof(u8), done);
532 c->chooseleaf_vary_r = ceph_decode_8(p);
533 dout("crush decode tunable chooseleaf_vary_r = %d\n",
534 c->chooseleaf_vary_r);
536 /* skip straw_calc_version, allowed_bucket_algs */
537 ceph_decode_need(p, end, sizeof(u8) + sizeof(u32), done);
538 *p += sizeof(u8) + sizeof(u32);
540 ceph_decode_need(p, end, sizeof(u8), done);
541 c->chooseleaf_stable = ceph_decode_8(p);
542 dout("crush decode tunable chooseleaf_stable = %d\n",
543 c->chooseleaf_stable);
547 ceph_decode_skip_map(p, end, 32, 32, bad);
549 ceph_decode_skip_map(p, end, 32, string, bad);
551 ceph_decode_skip_map_of_map(p, end, 32, 32, 32, bad);
555 err = decode_choose_args(p, end, c);
562 dout("crush_decode success\n");
568 dout("crush_decode fail %d\n", err);
577 int ceph_pg_compare(const struct ceph_pg *lhs, const struct ceph_pg *rhs)
579 if (lhs->pool < rhs->pool)
581 if (lhs->pool > rhs->pool)
583 if (lhs->seed < rhs->seed)
585 if (lhs->seed > rhs->seed)
591 int ceph_spg_compare(const struct ceph_spg *lhs, const struct ceph_spg *rhs)
595 ret = ceph_pg_compare(&lhs->pgid, &rhs->pgid);
599 if (lhs->shard < rhs->shard)
601 if (lhs->shard > rhs->shard)
607 static struct ceph_pg_mapping *alloc_pg_mapping(size_t payload_len)
609 struct ceph_pg_mapping *pg;
611 pg = kmalloc(sizeof(*pg) + payload_len, GFP_NOIO);
615 RB_CLEAR_NODE(&pg->node);
619 static void free_pg_mapping(struct ceph_pg_mapping *pg)
621 WARN_ON(!RB_EMPTY_NODE(&pg->node));
627 * rbtree of pg_mapping for handling pg_temp (explicit mapping of pgid
628 * to a set of osds) and primary_temp (explicit primary setting)
630 DEFINE_RB_FUNCS2(pg_mapping, struct ceph_pg_mapping, pgid, ceph_pg_compare,
631 RB_BYPTR, const struct ceph_pg *, node)
634 * rbtree of pg pool info
636 static int __insert_pg_pool(struct rb_root *root, struct ceph_pg_pool_info *new)
638 struct rb_node **p = &root->rb_node;
639 struct rb_node *parent = NULL;
640 struct ceph_pg_pool_info *pi = NULL;
644 pi = rb_entry(parent, struct ceph_pg_pool_info, node);
645 if (new->id < pi->id)
647 else if (new->id > pi->id)
653 rb_link_node(&new->node, parent, p);
654 rb_insert_color(&new->node, root);
658 static struct ceph_pg_pool_info *__lookup_pg_pool(struct rb_root *root, u64 id)
660 struct ceph_pg_pool_info *pi;
661 struct rb_node *n = root->rb_node;
664 pi = rb_entry(n, struct ceph_pg_pool_info, node);
667 else if (id > pi->id)
675 struct ceph_pg_pool_info *ceph_pg_pool_by_id(struct ceph_osdmap *map, u64 id)
677 return __lookup_pg_pool(&map->pg_pools, id);
680 const char *ceph_pg_pool_name_by_id(struct ceph_osdmap *map, u64 id)
682 struct ceph_pg_pool_info *pi;
684 if (id == CEPH_NOPOOL)
687 if (WARN_ON_ONCE(id > (u64) INT_MAX))
690 pi = __lookup_pg_pool(&map->pg_pools, (int) id);
692 return pi ? pi->name : NULL;
694 EXPORT_SYMBOL(ceph_pg_pool_name_by_id);
696 int ceph_pg_poolid_by_name(struct ceph_osdmap *map, const char *name)
700 for (rbp = rb_first(&map->pg_pools); rbp; rbp = rb_next(rbp)) {
701 struct ceph_pg_pool_info *pi =
702 rb_entry(rbp, struct ceph_pg_pool_info, node);
703 if (pi->name && strcmp(pi->name, name) == 0)
708 EXPORT_SYMBOL(ceph_pg_poolid_by_name);
710 static void __remove_pg_pool(struct rb_root *root, struct ceph_pg_pool_info *pi)
712 rb_erase(&pi->node, root);
717 static int decode_pool(void **p, void *end, struct ceph_pg_pool_info *pi)
723 ceph_decode_need(p, end, 2 + 4, bad);
724 ev = ceph_decode_8(p); /* encoding version */
725 cv = ceph_decode_8(p); /* compat version */
727 pr_warn("got v %d < 5 cv %d of ceph_pg_pool\n", ev, cv);
731 pr_warn("got v %d cv %d > 9 of ceph_pg_pool\n", ev, cv);
734 len = ceph_decode_32(p);
735 ceph_decode_need(p, end, len, bad);
738 pi->type = ceph_decode_8(p);
739 pi->size = ceph_decode_8(p);
740 pi->crush_ruleset = ceph_decode_8(p);
741 pi->object_hash = ceph_decode_8(p);
743 pi->pg_num = ceph_decode_32(p);
744 pi->pgp_num = ceph_decode_32(p);
746 *p += 4 + 4; /* skip lpg* */
747 *p += 4; /* skip last_change */
748 *p += 8 + 4; /* skip snap_seq, snap_epoch */
751 num = ceph_decode_32(p);
753 *p += 8; /* snapid key */
754 *p += 1 + 1; /* versions */
755 len = ceph_decode_32(p);
759 /* skip removed_snaps */
760 num = ceph_decode_32(p);
763 *p += 8; /* skip auid */
764 pi->flags = ceph_decode_64(p);
765 *p += 4; /* skip crash_replay_interval */
768 pi->min_size = ceph_decode_8(p);
770 pi->min_size = pi->size - pi->size / 2;
773 *p += 8 + 8; /* skip quota_max_* */
777 num = ceph_decode_32(p);
780 *p += 8; /* skip tier_of */
781 *p += 1; /* skip cache_mode */
783 pi->read_tier = ceph_decode_64(p);
784 pi->write_tier = ceph_decode_64(p);
791 /* skip properties */
792 num = ceph_decode_32(p);
794 len = ceph_decode_32(p);
796 len = ceph_decode_32(p);
802 /* skip hit_set_params */
803 *p += 1 + 1; /* versions */
804 len = ceph_decode_32(p);
807 *p += 4; /* skip hit_set_period */
808 *p += 4; /* skip hit_set_count */
812 *p += 4; /* skip stripe_width */
815 *p += 8; /* skip target_max_bytes */
816 *p += 8; /* skip target_max_objects */
817 *p += 4; /* skip cache_target_dirty_ratio_micro */
818 *p += 4; /* skip cache_target_full_ratio_micro */
819 *p += 4; /* skip cache_min_flush_age */
820 *p += 4; /* skip cache_min_evict_age */
824 /* skip erasure_code_profile */
825 len = ceph_decode_32(p);
830 * last_force_op_resend_preluminous, will be overridden if the
831 * map was encoded with RESEND_ON_SPLIT
834 pi->last_force_request_resend = ceph_decode_32(p);
836 pi->last_force_request_resend = 0;
839 *p += 4; /* skip min_read_recency_for_promote */
842 *p += 8; /* skip expected_num_objects */
845 *p += 4; /* skip cache_target_dirty_high_ratio_micro */
848 *p += 4; /* skip min_write_recency_for_promote */
851 *p += 1; /* skip use_gmt_hitset */
854 *p += 1; /* skip fast_read */
857 *p += 4; /* skip hit_set_grade_decay_rate */
858 *p += 4; /* skip hit_set_search_last_n */
863 *p += 1 + 1; /* versions */
864 len = ceph_decode_32(p);
869 pi->last_force_request_resend = ceph_decode_32(p);
871 /* ignore the rest */
881 static int decode_pool_names(void **p, void *end, struct ceph_osdmap *map)
883 struct ceph_pg_pool_info *pi;
887 ceph_decode_32_safe(p, end, num, bad);
888 dout(" %d pool names\n", num);
890 ceph_decode_64_safe(p, end, pool, bad);
891 ceph_decode_32_safe(p, end, len, bad);
892 dout(" pool %llu len %d\n", pool, len);
893 ceph_decode_need(p, end, len, bad);
894 pi = __lookup_pg_pool(&map->pg_pools, pool);
896 char *name = kstrndup(*p, len, GFP_NOFS);
902 dout(" name is %s\n", pi->name);
915 struct ceph_osdmap *ceph_osdmap_alloc(void)
917 struct ceph_osdmap *map;
919 map = kzalloc(sizeof(*map), GFP_NOIO);
923 map->pg_pools = RB_ROOT;
925 map->pg_temp = RB_ROOT;
926 map->primary_temp = RB_ROOT;
927 map->pg_upmap = RB_ROOT;
928 map->pg_upmap_items = RB_ROOT;
929 mutex_init(&map->crush_workspace_mutex);
934 void ceph_osdmap_destroy(struct ceph_osdmap *map)
936 dout("osdmap_destroy %p\n", map);
938 crush_destroy(map->crush);
939 while (!RB_EMPTY_ROOT(&map->pg_temp)) {
940 struct ceph_pg_mapping *pg =
941 rb_entry(rb_first(&map->pg_temp),
942 struct ceph_pg_mapping, node);
943 erase_pg_mapping(&map->pg_temp, pg);
946 while (!RB_EMPTY_ROOT(&map->primary_temp)) {
947 struct ceph_pg_mapping *pg =
948 rb_entry(rb_first(&map->primary_temp),
949 struct ceph_pg_mapping, node);
950 erase_pg_mapping(&map->primary_temp, pg);
953 while (!RB_EMPTY_ROOT(&map->pg_upmap)) {
954 struct ceph_pg_mapping *pg =
955 rb_entry(rb_first(&map->pg_upmap),
956 struct ceph_pg_mapping, node);
957 rb_erase(&pg->node, &map->pg_upmap);
960 while (!RB_EMPTY_ROOT(&map->pg_upmap_items)) {
961 struct ceph_pg_mapping *pg =
962 rb_entry(rb_first(&map->pg_upmap_items),
963 struct ceph_pg_mapping, node);
964 rb_erase(&pg->node, &map->pg_upmap_items);
967 while (!RB_EMPTY_ROOT(&map->pg_pools)) {
968 struct ceph_pg_pool_info *pi =
969 rb_entry(rb_first(&map->pg_pools),
970 struct ceph_pg_pool_info, node);
971 __remove_pg_pool(&map->pg_pools, pi);
973 kfree(map->osd_state);
974 kfree(map->osd_weight);
975 kfree(map->osd_addr);
976 kfree(map->osd_primary_affinity);
977 kfree(map->crush_workspace);
982 * Adjust max_osd value, (re)allocate arrays.
984 * The new elements are properly initialized.
986 static int osdmap_set_max_osd(struct ceph_osdmap *map, int max)
990 struct ceph_entity_addr *addr;
993 state = krealloc(map->osd_state, max*sizeof(*state), GFP_NOFS);
996 map->osd_state = state;
998 weight = krealloc(map->osd_weight, max*sizeof(*weight), GFP_NOFS);
1001 map->osd_weight = weight;
1003 addr = krealloc(map->osd_addr, max*sizeof(*addr), GFP_NOFS);
1006 map->osd_addr = addr;
1008 for (i = map->max_osd; i < max; i++) {
1009 map->osd_state[i] = 0;
1010 map->osd_weight[i] = CEPH_OSD_OUT;
1011 memset(map->osd_addr + i, 0, sizeof(*map->osd_addr));
1014 if (map->osd_primary_affinity) {
1017 affinity = krealloc(map->osd_primary_affinity,
1018 max*sizeof(*affinity), GFP_NOFS);
1021 map->osd_primary_affinity = affinity;
1023 for (i = map->max_osd; i < max; i++)
1024 map->osd_primary_affinity[i] =
1025 CEPH_OSD_DEFAULT_PRIMARY_AFFINITY;
1033 static int osdmap_set_crush(struct ceph_osdmap *map, struct crush_map *crush)
1039 return PTR_ERR(crush);
1041 work_size = crush_work_size(crush, CEPH_PG_MAX_SIZE);
1042 dout("%s work_size %zu bytes\n", __func__, work_size);
1043 workspace = kmalloc(work_size, GFP_NOIO);
1045 crush_destroy(crush);
1048 crush_init_workspace(crush, workspace);
1051 crush_destroy(map->crush);
1052 kfree(map->crush_workspace);
1054 map->crush_workspace = workspace;
1058 #define OSDMAP_WRAPPER_COMPAT_VER 7
1059 #define OSDMAP_CLIENT_DATA_COMPAT_VER 1
1062 * Return 0 or error. On success, *v is set to 0 for old (v6) osdmaps,
1063 * to struct_v of the client_data section for new (v7 and above)
1066 static int get_osdmap_client_data_v(void **p, void *end,
1067 const char *prefix, u8 *v)
1071 ceph_decode_8_safe(p, end, struct_v, e_inval);
1072 if (struct_v >= 7) {
1075 ceph_decode_8_safe(p, end, struct_compat, e_inval);
1076 if (struct_compat > OSDMAP_WRAPPER_COMPAT_VER) {
1077 pr_warn("got v %d cv %d > %d of %s ceph_osdmap\n",
1078 struct_v, struct_compat,
1079 OSDMAP_WRAPPER_COMPAT_VER, prefix);
1082 *p += 4; /* ignore wrapper struct_len */
1084 ceph_decode_8_safe(p, end, struct_v, e_inval);
1085 ceph_decode_8_safe(p, end, struct_compat, e_inval);
1086 if (struct_compat > OSDMAP_CLIENT_DATA_COMPAT_VER) {
1087 pr_warn("got v %d cv %d > %d of %s ceph_osdmap client data\n",
1088 struct_v, struct_compat,
1089 OSDMAP_CLIENT_DATA_COMPAT_VER, prefix);
1092 *p += 4; /* ignore client data struct_len */
1097 ceph_decode_16_safe(p, end, version, e_inval);
1099 pr_warn("got v %d < 6 of %s ceph_osdmap\n",
1104 /* old osdmap enconding */
1115 static int __decode_pools(void **p, void *end, struct ceph_osdmap *map,
1120 ceph_decode_32_safe(p, end, n, e_inval);
1122 struct ceph_pg_pool_info *pi;
1126 ceph_decode_64_safe(p, end, pool, e_inval);
1128 pi = __lookup_pg_pool(&map->pg_pools, pool);
1129 if (!incremental || !pi) {
1130 pi = kzalloc(sizeof(*pi), GFP_NOFS);
1136 ret = __insert_pg_pool(&map->pg_pools, pi);
1143 ret = decode_pool(p, end, pi);
1154 static int decode_pools(void **p, void *end, struct ceph_osdmap *map)
1156 return __decode_pools(p, end, map, false);
1159 static int decode_new_pools(void **p, void *end, struct ceph_osdmap *map)
1161 return __decode_pools(p, end, map, true);
1164 typedef struct ceph_pg_mapping *(*decode_mapping_fn_t)(void **, void *, bool);
1166 static int decode_pg_mapping(void **p, void *end, struct rb_root *mapping_root,
1167 decode_mapping_fn_t fn, bool incremental)
1171 WARN_ON(!incremental && !fn);
1173 ceph_decode_32_safe(p, end, n, e_inval);
1175 struct ceph_pg_mapping *pg;
1176 struct ceph_pg pgid;
1179 ret = ceph_decode_pgid(p, end, &pgid);
1183 pg = lookup_pg_mapping(mapping_root, &pgid);
1185 WARN_ON(!incremental);
1186 erase_pg_mapping(mapping_root, pg);
1187 free_pg_mapping(pg);
1191 pg = fn(p, end, incremental);
1196 pg->pgid = pgid; /* struct */
1197 insert_pg_mapping(mapping_root, pg);
1208 static struct ceph_pg_mapping *__decode_pg_temp(void **p, void *end,
1211 struct ceph_pg_mapping *pg;
1214 ceph_decode_32_safe(p, end, len, e_inval);
1215 if (len == 0 && incremental)
1216 return NULL; /* new_pg_temp: [] to remove */
1217 if (len > (SIZE_MAX - sizeof(*pg)) / sizeof(u32))
1218 return ERR_PTR(-EINVAL);
1220 ceph_decode_need(p, end, len * sizeof(u32), e_inval);
1221 pg = alloc_pg_mapping(len * sizeof(u32));
1223 return ERR_PTR(-ENOMEM);
1225 pg->pg_temp.len = len;
1226 for (i = 0; i < len; i++)
1227 pg->pg_temp.osds[i] = ceph_decode_32(p);
1232 return ERR_PTR(-EINVAL);
1235 static int decode_pg_temp(void **p, void *end, struct ceph_osdmap *map)
1237 return decode_pg_mapping(p, end, &map->pg_temp, __decode_pg_temp,
1241 static int decode_new_pg_temp(void **p, void *end, struct ceph_osdmap *map)
1243 return decode_pg_mapping(p, end, &map->pg_temp, __decode_pg_temp,
1247 static struct ceph_pg_mapping *__decode_primary_temp(void **p, void *end,
1250 struct ceph_pg_mapping *pg;
1253 ceph_decode_32_safe(p, end, osd, e_inval);
1254 if (osd == (u32)-1 && incremental)
1255 return NULL; /* new_primary_temp: -1 to remove */
1257 pg = alloc_pg_mapping(0);
1259 return ERR_PTR(-ENOMEM);
1261 pg->primary_temp.osd = osd;
1265 return ERR_PTR(-EINVAL);
1268 static int decode_primary_temp(void **p, void *end, struct ceph_osdmap *map)
1270 return decode_pg_mapping(p, end, &map->primary_temp,
1271 __decode_primary_temp, false);
1274 static int decode_new_primary_temp(void **p, void *end,
1275 struct ceph_osdmap *map)
1277 return decode_pg_mapping(p, end, &map->primary_temp,
1278 __decode_primary_temp, true);
1281 u32 ceph_get_primary_affinity(struct ceph_osdmap *map, int osd)
1283 BUG_ON(osd >= map->max_osd);
1285 if (!map->osd_primary_affinity)
1286 return CEPH_OSD_DEFAULT_PRIMARY_AFFINITY;
1288 return map->osd_primary_affinity[osd];
1291 static int set_primary_affinity(struct ceph_osdmap *map, int osd, u32 aff)
1293 BUG_ON(osd >= map->max_osd);
1295 if (!map->osd_primary_affinity) {
1298 map->osd_primary_affinity = kmalloc(map->max_osd*sizeof(u32),
1300 if (!map->osd_primary_affinity)
1303 for (i = 0; i < map->max_osd; i++)
1304 map->osd_primary_affinity[i] =
1305 CEPH_OSD_DEFAULT_PRIMARY_AFFINITY;
1308 map->osd_primary_affinity[osd] = aff;
1313 static int decode_primary_affinity(void **p, void *end,
1314 struct ceph_osdmap *map)
1318 ceph_decode_32_safe(p, end, len, e_inval);
1320 kfree(map->osd_primary_affinity);
1321 map->osd_primary_affinity = NULL;
1324 if (len != map->max_osd)
1327 ceph_decode_need(p, end, map->max_osd*sizeof(u32), e_inval);
1329 for (i = 0; i < map->max_osd; i++) {
1332 ret = set_primary_affinity(map, i, ceph_decode_32(p));
1343 static int decode_new_primary_affinity(void **p, void *end,
1344 struct ceph_osdmap *map)
1348 ceph_decode_32_safe(p, end, n, e_inval);
1353 ceph_decode_32_safe(p, end, osd, e_inval);
1354 ceph_decode_32_safe(p, end, aff, e_inval);
1356 ret = set_primary_affinity(map, osd, aff);
1360 pr_info("osd%d primary-affinity 0x%x\n", osd, aff);
1369 static struct ceph_pg_mapping *__decode_pg_upmap(void **p, void *end,
1372 return __decode_pg_temp(p, end, false);
1375 static int decode_pg_upmap(void **p, void *end, struct ceph_osdmap *map)
1377 return decode_pg_mapping(p, end, &map->pg_upmap, __decode_pg_upmap,
1381 static int decode_new_pg_upmap(void **p, void *end, struct ceph_osdmap *map)
1383 return decode_pg_mapping(p, end, &map->pg_upmap, __decode_pg_upmap,
1387 static int decode_old_pg_upmap(void **p, void *end, struct ceph_osdmap *map)
1389 return decode_pg_mapping(p, end, &map->pg_upmap, NULL, true);
1392 static struct ceph_pg_mapping *__decode_pg_upmap_items(void **p, void *end,
1395 struct ceph_pg_mapping *pg;
1398 ceph_decode_32_safe(p, end, len, e_inval);
1399 if (len > (SIZE_MAX - sizeof(*pg)) / (2 * sizeof(u32)))
1400 return ERR_PTR(-EINVAL);
1402 ceph_decode_need(p, end, 2 * len * sizeof(u32), e_inval);
1403 pg = alloc_pg_mapping(2 * len * sizeof(u32));
1405 return ERR_PTR(-ENOMEM);
1407 pg->pg_upmap_items.len = len;
1408 for (i = 0; i < len; i++) {
1409 pg->pg_upmap_items.from_to[i][0] = ceph_decode_32(p);
1410 pg->pg_upmap_items.from_to[i][1] = ceph_decode_32(p);
1416 return ERR_PTR(-EINVAL);
1419 static int decode_pg_upmap_items(void **p, void *end, struct ceph_osdmap *map)
1421 return decode_pg_mapping(p, end, &map->pg_upmap_items,
1422 __decode_pg_upmap_items, false);
1425 static int decode_new_pg_upmap_items(void **p, void *end,
1426 struct ceph_osdmap *map)
1428 return decode_pg_mapping(p, end, &map->pg_upmap_items,
1429 __decode_pg_upmap_items, true);
1432 static int decode_old_pg_upmap_items(void **p, void *end,
1433 struct ceph_osdmap *map)
1435 return decode_pg_mapping(p, end, &map->pg_upmap_items, NULL, true);
1439 * decode a full map.
1441 static int osdmap_decode(void **p, void *end, struct ceph_osdmap *map)
1450 dout("%s %p to %p len %d\n", __func__, *p, end, (int)(end - *p));
1452 err = get_osdmap_client_data_v(p, end, "full", &struct_v);
1456 /* fsid, epoch, created, modified */
1457 ceph_decode_need(p, end, sizeof(map->fsid) + sizeof(u32) +
1458 sizeof(map->created) + sizeof(map->modified), e_inval);
1459 ceph_decode_copy(p, &map->fsid, sizeof(map->fsid));
1460 epoch = map->epoch = ceph_decode_32(p);
1461 ceph_decode_copy(p, &map->created, sizeof(map->created));
1462 ceph_decode_copy(p, &map->modified, sizeof(map->modified));
1465 err = decode_pools(p, end, map);
1470 err = decode_pool_names(p, end, map);
1474 ceph_decode_32_safe(p, end, map->pool_max, e_inval);
1476 ceph_decode_32_safe(p, end, map->flags, e_inval);
1479 ceph_decode_32_safe(p, end, max, e_inval);
1481 /* (re)alloc osd arrays */
1482 err = osdmap_set_max_osd(map, max);
1486 /* osd_state, osd_weight, osd_addrs->client_addr */
1487 ceph_decode_need(p, end, 3*sizeof(u32) +
1488 map->max_osd*((struct_v >= 5 ? sizeof(u32) :
1490 sizeof(*map->osd_weight) +
1491 sizeof(*map->osd_addr)), e_inval);
1493 if (ceph_decode_32(p) != map->max_osd)
1496 if (struct_v >= 5) {
1497 for (i = 0; i < map->max_osd; i++)
1498 map->osd_state[i] = ceph_decode_32(p);
1500 for (i = 0; i < map->max_osd; i++)
1501 map->osd_state[i] = ceph_decode_8(p);
1504 if (ceph_decode_32(p) != map->max_osd)
1507 for (i = 0; i < map->max_osd; i++)
1508 map->osd_weight[i] = ceph_decode_32(p);
1510 if (ceph_decode_32(p) != map->max_osd)
1513 ceph_decode_copy(p, map->osd_addr, map->max_osd*sizeof(*map->osd_addr));
1514 for (i = 0; i < map->max_osd; i++)
1515 ceph_decode_addr(&map->osd_addr[i]);
1518 err = decode_pg_temp(p, end, map);
1523 if (struct_v >= 1) {
1524 err = decode_primary_temp(p, end, map);
1529 /* primary_affinity */
1530 if (struct_v >= 2) {
1531 err = decode_primary_affinity(p, end, map);
1535 WARN_ON(map->osd_primary_affinity);
1539 ceph_decode_32_safe(p, end, len, e_inval);
1540 err = osdmap_set_crush(map, crush_decode(*p, min(*p + len, end)));
1545 if (struct_v >= 3) {
1546 /* erasure_code_profiles */
1547 ceph_decode_skip_map_of_map(p, end, string, string, string,
1551 if (struct_v >= 4) {
1552 err = decode_pg_upmap(p, end, map);
1556 err = decode_pg_upmap_items(p, end, map);
1560 WARN_ON(!RB_EMPTY_ROOT(&map->pg_upmap));
1561 WARN_ON(!RB_EMPTY_ROOT(&map->pg_upmap_items));
1564 /* ignore the rest */
1567 dout("full osdmap epoch %d max_osd %d\n", map->epoch, map->max_osd);
1573 pr_err("corrupt full osdmap (%d) epoch %d off %d (%p of %p-%p)\n",
1574 err, epoch, (int)(*p - start), *p, start, end);
1575 print_hex_dump(KERN_DEBUG, "osdmap: ",
1576 DUMP_PREFIX_OFFSET, 16, 1,
1577 start, end - start, true);
1582 * Allocate and decode a full map.
1584 struct ceph_osdmap *ceph_osdmap_decode(void **p, void *end)
1586 struct ceph_osdmap *map;
1589 map = ceph_osdmap_alloc();
1591 return ERR_PTR(-ENOMEM);
1593 ret = osdmap_decode(p, end, map);
1595 ceph_osdmap_destroy(map);
1596 return ERR_PTR(ret);
1603 * Encoding order is (new_up_client, new_state, new_weight). Need to
1604 * apply in the (new_weight, new_state, new_up_client) order, because
1605 * an incremental map may look like e.g.
1607 * new_up_client: { osd=6, addr=... } # set osd_state and addr
1608 * new_state: { osd=6, xorstate=EXISTS } # clear osd_state
1610 static int decode_new_up_state_weight(void **p, void *end, u8 struct_v,
1611 struct ceph_osdmap *map)
1613 void *new_up_client;
1615 void *new_weight_end;
1619 ceph_decode_32_safe(p, end, len, e_inval);
1620 len *= sizeof(u32) + sizeof(struct ceph_entity_addr);
1621 ceph_decode_need(p, end, len, e_inval);
1625 ceph_decode_32_safe(p, end, len, e_inval);
1626 len *= sizeof(u32) + (struct_v >= 5 ? sizeof(u32) : sizeof(u8));
1627 ceph_decode_need(p, end, len, e_inval);
1631 ceph_decode_32_safe(p, end, len, e_inval);
1636 ceph_decode_need(p, end, 2*sizeof(u32), e_inval);
1637 osd = ceph_decode_32(p);
1638 w = ceph_decode_32(p);
1639 BUG_ON(osd >= map->max_osd);
1640 pr_info("osd%d weight 0x%x %s\n", osd, w,
1641 w == CEPH_OSD_IN ? "(in)" :
1642 (w == CEPH_OSD_OUT ? "(out)" : ""));
1643 map->osd_weight[osd] = w;
1646 * If we are marking in, set the EXISTS, and clear the
1647 * AUTOOUT and NEW bits.
1650 map->osd_state[osd] |= CEPH_OSD_EXISTS;
1651 map->osd_state[osd] &= ~(CEPH_OSD_AUTOOUT |
1655 new_weight_end = *p;
1657 /* new_state (up/down) */
1659 len = ceph_decode_32(p);
1665 osd = ceph_decode_32(p);
1667 xorstate = ceph_decode_32(p);
1669 xorstate = ceph_decode_8(p);
1671 xorstate = CEPH_OSD_UP;
1672 BUG_ON(osd >= map->max_osd);
1673 if ((map->osd_state[osd] & CEPH_OSD_UP) &&
1674 (xorstate & CEPH_OSD_UP))
1675 pr_info("osd%d down\n", osd);
1676 if ((map->osd_state[osd] & CEPH_OSD_EXISTS) &&
1677 (xorstate & CEPH_OSD_EXISTS)) {
1678 pr_info("osd%d does not exist\n", osd);
1679 ret = set_primary_affinity(map, osd,
1680 CEPH_OSD_DEFAULT_PRIMARY_AFFINITY);
1683 memset(map->osd_addr + osd, 0, sizeof(*map->osd_addr));
1684 map->osd_state[osd] = 0;
1686 map->osd_state[osd] ^= xorstate;
1692 len = ceph_decode_32(p);
1695 struct ceph_entity_addr addr;
1697 osd = ceph_decode_32(p);
1698 ceph_decode_copy(p, &addr, sizeof(addr));
1699 ceph_decode_addr(&addr);
1700 BUG_ON(osd >= map->max_osd);
1701 pr_info("osd%d up\n", osd);
1702 map->osd_state[osd] |= CEPH_OSD_EXISTS | CEPH_OSD_UP;
1703 map->osd_addr[osd] = addr;
1706 *p = new_weight_end;
1714 * decode and apply an incremental map update.
1716 struct ceph_osdmap *osdmap_apply_incremental(void **p, void *end,
1717 struct ceph_osdmap *map)
1719 struct ceph_fsid fsid;
1721 struct ceph_timespec modified;
1725 __s32 new_flags, max;
1730 dout("%s %p to %p len %d\n", __func__, *p, end, (int)(end - *p));
1732 err = get_osdmap_client_data_v(p, end, "inc", &struct_v);
1736 /* fsid, epoch, modified, new_pool_max, new_flags */
1737 ceph_decode_need(p, end, sizeof(fsid) + sizeof(u32) + sizeof(modified) +
1738 sizeof(u64) + sizeof(u32), e_inval);
1739 ceph_decode_copy(p, &fsid, sizeof(fsid));
1740 epoch = ceph_decode_32(p);
1741 BUG_ON(epoch != map->epoch+1);
1742 ceph_decode_copy(p, &modified, sizeof(modified));
1743 new_pool_max = ceph_decode_64(p);
1744 new_flags = ceph_decode_32(p);
1747 ceph_decode_32_safe(p, end, len, e_inval);
1749 dout("apply_incremental full map len %d, %p to %p\n",
1751 return ceph_osdmap_decode(p, min(*p+len, end));
1755 ceph_decode_32_safe(p, end, len, e_inval);
1757 err = osdmap_set_crush(map,
1758 crush_decode(*p, min(*p + len, end)));
1766 map->flags = new_flags;
1767 if (new_pool_max >= 0)
1768 map->pool_max = new_pool_max;
1771 ceph_decode_32_safe(p, end, max, e_inval);
1773 err = osdmap_set_max_osd(map, max);
1779 map->modified = modified;
1782 err = decode_new_pools(p, end, map);
1786 /* new_pool_names */
1787 err = decode_pool_names(p, end, map);
1792 ceph_decode_32_safe(p, end, len, e_inval);
1794 struct ceph_pg_pool_info *pi;
1796 ceph_decode_64_safe(p, end, pool, e_inval);
1797 pi = __lookup_pg_pool(&map->pg_pools, pool);
1799 __remove_pg_pool(&map->pg_pools, pi);
1802 /* new_up_client, new_state, new_weight */
1803 err = decode_new_up_state_weight(p, end, struct_v, map);
1808 err = decode_new_pg_temp(p, end, map);
1812 /* new_primary_temp */
1813 if (struct_v >= 1) {
1814 err = decode_new_primary_temp(p, end, map);
1819 /* new_primary_affinity */
1820 if (struct_v >= 2) {
1821 err = decode_new_primary_affinity(p, end, map);
1826 if (struct_v >= 3) {
1827 /* new_erasure_code_profiles */
1828 ceph_decode_skip_map_of_map(p, end, string, string, string,
1830 /* old_erasure_code_profiles */
1831 ceph_decode_skip_set(p, end, string, e_inval);
1834 if (struct_v >= 4) {
1835 err = decode_new_pg_upmap(p, end, map);
1839 err = decode_old_pg_upmap(p, end, map);
1843 err = decode_new_pg_upmap_items(p, end, map);
1847 err = decode_old_pg_upmap_items(p, end, map);
1852 /* ignore the rest */
1855 dout("inc osdmap epoch %d max_osd %d\n", map->epoch, map->max_osd);
1861 pr_err("corrupt inc osdmap (%d) epoch %d off %d (%p of %p-%p)\n",
1862 err, epoch, (int)(*p - start), *p, start, end);
1863 print_hex_dump(KERN_DEBUG, "osdmap: ",
1864 DUMP_PREFIX_OFFSET, 16, 1,
1865 start, end - start, true);
1866 return ERR_PTR(err);
1869 void ceph_oloc_copy(struct ceph_object_locator *dest,
1870 const struct ceph_object_locator *src)
1872 ceph_oloc_destroy(dest);
1874 dest->pool = src->pool;
1876 dest->pool_ns = ceph_get_string(src->pool_ns);
1878 dest->pool_ns = NULL;
1880 EXPORT_SYMBOL(ceph_oloc_copy);
1882 void ceph_oloc_destroy(struct ceph_object_locator *oloc)
1884 ceph_put_string(oloc->pool_ns);
1886 EXPORT_SYMBOL(ceph_oloc_destroy);
1888 void ceph_oid_copy(struct ceph_object_id *dest,
1889 const struct ceph_object_id *src)
1891 ceph_oid_destroy(dest);
1893 if (src->name != src->inline_name) {
1894 /* very rare, see ceph_object_id definition */
1895 dest->name = kmalloc(src->name_len + 1,
1896 GFP_NOIO | __GFP_NOFAIL);
1898 dest->name = dest->inline_name;
1900 memcpy(dest->name, src->name, src->name_len + 1);
1901 dest->name_len = src->name_len;
1903 EXPORT_SYMBOL(ceph_oid_copy);
1905 static __printf(2, 0)
1906 int oid_printf_vargs(struct ceph_object_id *oid, const char *fmt, va_list ap)
1910 WARN_ON(!ceph_oid_empty(oid));
1912 len = vsnprintf(oid->inline_name, sizeof(oid->inline_name), fmt, ap);
1913 if (len >= sizeof(oid->inline_name))
1916 oid->name_len = len;
1921 * If oid doesn't fit into inline buffer, BUG.
1923 void ceph_oid_printf(struct ceph_object_id *oid, const char *fmt, ...)
1928 BUG_ON(oid_printf_vargs(oid, fmt, ap));
1931 EXPORT_SYMBOL(ceph_oid_printf);
1933 static __printf(3, 0)
1934 int oid_aprintf_vargs(struct ceph_object_id *oid, gfp_t gfp,
1935 const char *fmt, va_list ap)
1941 len = oid_printf_vargs(oid, fmt, aq);
1945 char *external_name;
1947 external_name = kmalloc(len + 1, gfp);
1951 oid->name = external_name;
1952 WARN_ON(vsnprintf(oid->name, len + 1, fmt, ap) != len);
1953 oid->name_len = len;
1960 * If oid doesn't fit into inline buffer, allocate.
1962 int ceph_oid_aprintf(struct ceph_object_id *oid, gfp_t gfp,
1963 const char *fmt, ...)
1969 ret = oid_aprintf_vargs(oid, gfp, fmt, ap);
1974 EXPORT_SYMBOL(ceph_oid_aprintf);
1976 void ceph_oid_destroy(struct ceph_object_id *oid)
1978 if (oid->name != oid->inline_name)
1981 EXPORT_SYMBOL(ceph_oid_destroy);
1986 static bool __osds_equal(const struct ceph_osds *lhs,
1987 const struct ceph_osds *rhs)
1989 if (lhs->size == rhs->size &&
1990 !memcmp(lhs->osds, rhs->osds, rhs->size * sizeof(rhs->osds[0])))
1999 static bool osds_equal(const struct ceph_osds *lhs,
2000 const struct ceph_osds *rhs)
2002 if (__osds_equal(lhs, rhs) &&
2003 lhs->primary == rhs->primary)
2009 static bool osds_valid(const struct ceph_osds *set)
2012 if (set->size > 0 && set->primary >= 0)
2015 /* empty can_shift_osds set */
2016 if (!set->size && set->primary == -1)
2019 /* empty !can_shift_osds set - all NONE */
2020 if (set->size > 0 && set->primary == -1) {
2023 for (i = 0; i < set->size; i++) {
2024 if (set->osds[i] != CRUSH_ITEM_NONE)
2034 void ceph_osds_copy(struct ceph_osds *dest, const struct ceph_osds *src)
2036 memcpy(dest->osds, src->osds, src->size * sizeof(src->osds[0]));
2037 dest->size = src->size;
2038 dest->primary = src->primary;
2041 bool ceph_pg_is_split(const struct ceph_pg *pgid, u32 old_pg_num,
2044 int old_bits = calc_bits_of(old_pg_num);
2045 int old_mask = (1 << old_bits) - 1;
2048 WARN_ON(pgid->seed >= old_pg_num);
2049 if (new_pg_num <= old_pg_num)
2052 for (n = 1; ; n++) {
2053 int next_bit = n << (old_bits - 1);
2054 u32 s = next_bit | pgid->seed;
2056 if (s < old_pg_num || s == pgid->seed)
2058 if (s >= new_pg_num)
2061 s = ceph_stable_mod(s, old_pg_num, old_mask);
2062 if (s == pgid->seed)
2069 bool ceph_is_new_interval(const struct ceph_osds *old_acting,
2070 const struct ceph_osds *new_acting,
2071 const struct ceph_osds *old_up,
2072 const struct ceph_osds *new_up,
2079 bool old_sort_bitwise,
2080 bool new_sort_bitwise,
2081 const struct ceph_pg *pgid)
2083 return !osds_equal(old_acting, new_acting) ||
2084 !osds_equal(old_up, new_up) ||
2085 old_size != new_size ||
2086 old_min_size != new_min_size ||
2087 ceph_pg_is_split(pgid, old_pg_num, new_pg_num) ||
2088 old_sort_bitwise != new_sort_bitwise;
2091 static int calc_pg_rank(int osd, const struct ceph_osds *acting)
2095 for (i = 0; i < acting->size; i++) {
2096 if (acting->osds[i] == osd)
2103 static bool primary_changed(const struct ceph_osds *old_acting,
2104 const struct ceph_osds *new_acting)
2106 if (!old_acting->size && !new_acting->size)
2107 return false; /* both still empty */
2109 if (!old_acting->size ^ !new_acting->size)
2110 return true; /* was empty, now not, or vice versa */
2112 if (old_acting->primary != new_acting->primary)
2113 return true; /* primary changed */
2115 if (calc_pg_rank(old_acting->primary, old_acting) !=
2116 calc_pg_rank(new_acting->primary, new_acting))
2119 return false; /* same primary (tho replicas may have changed) */
2122 bool ceph_osds_changed(const struct ceph_osds *old_acting,
2123 const struct ceph_osds *new_acting,
2126 if (primary_changed(old_acting, new_acting))
2129 if (any_change && !__osds_equal(old_acting, new_acting))
2136 * calculate file layout from given offset, length.
2137 * fill in correct oid, logical length, and object extent
2140 * for now, we write only a single su, until we can
2141 * pass a stride back to the caller.
2143 int ceph_calc_file_object_mapping(struct ceph_file_layout *layout,
2146 u64 *oxoff, u64 *oxlen)
2148 u32 osize = layout->object_size;
2149 u32 su = layout->stripe_unit;
2150 u32 sc = layout->stripe_count;
2151 u32 bl, stripeno, stripepos, objsetno;
2155 dout("mapping %llu~%llu osize %u fl_su %u\n", off, len,
2157 if (su == 0 || sc == 0)
2159 su_per_object = osize / su;
2160 if (su_per_object == 0)
2162 dout("osize %u / su %u = su_per_object %u\n", osize, su,
2165 if ((su & ~PAGE_MASK) != 0)
2168 /* bl = *off / su; */
2172 dout("off %llu / su %u = bl %u\n", off, su, bl);
2175 stripepos = bl % sc;
2176 objsetno = stripeno / su_per_object;
2178 *ono = objsetno * sc + stripepos;
2179 dout("objset %u * sc %u = ono %u\n", objsetno, sc, (unsigned int)*ono);
2181 /* *oxoff = *off % layout->fl_stripe_unit; # offset in su */
2183 su_offset = do_div(t, su);
2184 *oxoff = su_offset + (stripeno % su_per_object) * su;
2187 * Calculate the length of the extent being written to the selected
2188 * object. This is the minimum of the full length requested (len) or
2189 * the remainder of the current stripe being written to.
2191 *oxlen = min_t(u64, len, su - su_offset);
2193 dout(" obj extent %llu~%llu\n", *oxoff, *oxlen);
2197 dout(" invalid layout\n");
2203 EXPORT_SYMBOL(ceph_calc_file_object_mapping);
2206 * Map an object into a PG.
2208 * Should only be called with target_oid and target_oloc (as opposed to
2209 * base_oid and base_oloc), since tiering isn't taken into account.
2211 int __ceph_object_locator_to_pg(struct ceph_pg_pool_info *pi,
2212 const struct ceph_object_id *oid,
2213 const struct ceph_object_locator *oloc,
2214 struct ceph_pg *raw_pgid)
2216 WARN_ON(pi->id != oloc->pool);
2218 if (!oloc->pool_ns) {
2219 raw_pgid->pool = oloc->pool;
2220 raw_pgid->seed = ceph_str_hash(pi->object_hash, oid->name,
2222 dout("%s %s -> raw_pgid %llu.%x\n", __func__, oid->name,
2223 raw_pgid->pool, raw_pgid->seed);
2225 char stack_buf[256];
2226 char *buf = stack_buf;
2227 int nsl = oloc->pool_ns->len;
2228 size_t total = nsl + 1 + oid->name_len;
2230 if (total > sizeof(stack_buf)) {
2231 buf = kmalloc(total, GFP_NOIO);
2235 memcpy(buf, oloc->pool_ns->str, nsl);
2237 memcpy(buf + nsl + 1, oid->name, oid->name_len);
2238 raw_pgid->pool = oloc->pool;
2239 raw_pgid->seed = ceph_str_hash(pi->object_hash, buf, total);
2240 if (buf != stack_buf)
2242 dout("%s %s ns %.*s -> raw_pgid %llu.%x\n", __func__,
2243 oid->name, nsl, oloc->pool_ns->str,
2244 raw_pgid->pool, raw_pgid->seed);
2249 int ceph_object_locator_to_pg(struct ceph_osdmap *osdmap,
2250 const struct ceph_object_id *oid,
2251 const struct ceph_object_locator *oloc,
2252 struct ceph_pg *raw_pgid)
2254 struct ceph_pg_pool_info *pi;
2256 pi = ceph_pg_pool_by_id(osdmap, oloc->pool);
2260 return __ceph_object_locator_to_pg(pi, oid, oloc, raw_pgid);
2262 EXPORT_SYMBOL(ceph_object_locator_to_pg);
2265 * Map a raw PG (full precision ps) into an actual PG.
2267 static void raw_pg_to_pg(struct ceph_pg_pool_info *pi,
2268 const struct ceph_pg *raw_pgid,
2269 struct ceph_pg *pgid)
2271 pgid->pool = raw_pgid->pool;
2272 pgid->seed = ceph_stable_mod(raw_pgid->seed, pi->pg_num,
2277 * Map a raw PG (full precision ps) into a placement ps (placement
2278 * seed). Include pool id in that value so that different pools don't
2279 * use the same seeds.
2281 static u32 raw_pg_to_pps(struct ceph_pg_pool_info *pi,
2282 const struct ceph_pg *raw_pgid)
2284 if (pi->flags & CEPH_POOL_FLAG_HASHPSPOOL) {
2285 /* hash pool id and seed so that pool PGs do not overlap */
2286 return crush_hash32_2(CRUSH_HASH_RJENKINS1,
2287 ceph_stable_mod(raw_pgid->seed,
2293 * legacy behavior: add ps and pool together. this is
2294 * not a great approach because the PGs from each pool
2295 * will overlap on top of each other: 0.5 == 1.4 ==
2298 return ceph_stable_mod(raw_pgid->seed, pi->pgp_num,
2300 (unsigned)raw_pgid->pool;
2304 static int do_crush(struct ceph_osdmap *map, int ruleno, int x,
2305 int *result, int result_max,
2306 const __u32 *weight, int weight_max,
2307 u64 choose_args_index)
2309 struct crush_choose_arg_map *arg_map;
2312 BUG_ON(result_max > CEPH_PG_MAX_SIZE);
2314 arg_map = lookup_choose_arg_map(&map->crush->choose_args,
2317 mutex_lock(&map->crush_workspace_mutex);
2318 r = crush_do_rule(map->crush, ruleno, x, result, result_max,
2319 weight, weight_max, map->crush_workspace,
2320 arg_map ? arg_map->args : NULL);
2321 mutex_unlock(&map->crush_workspace_mutex);
2326 static void remove_nonexistent_osds(struct ceph_osdmap *osdmap,
2327 struct ceph_pg_pool_info *pi,
2328 struct ceph_osds *set)
2332 if (ceph_can_shift_osds(pi)) {
2336 for (i = 0; i < set->size; i++) {
2337 if (!ceph_osd_exists(osdmap, set->osds[i])) {
2342 set->osds[i - removed] = set->osds[i];
2344 set->size -= removed;
2346 /* set dne devices to NONE */
2347 for (i = 0; i < set->size; i++) {
2348 if (!ceph_osd_exists(osdmap, set->osds[i]))
2349 set->osds[i] = CRUSH_ITEM_NONE;
2355 * Calculate raw set (CRUSH output) for given PG and filter out
2356 * nonexistent OSDs. ->primary is undefined for a raw set.
2358 * Placement seed (CRUSH input) is returned through @ppps.
2360 static void pg_to_raw_osds(struct ceph_osdmap *osdmap,
2361 struct ceph_pg_pool_info *pi,
2362 const struct ceph_pg *raw_pgid,
2363 struct ceph_osds *raw,
2366 u32 pps = raw_pg_to_pps(pi, raw_pgid);
2370 ceph_osds_init(raw);
2374 ruleno = crush_find_rule(osdmap->crush, pi->crush_ruleset, pi->type,
2377 pr_err("no crush rule: pool %lld ruleset %d type %d size %d\n",
2378 pi->id, pi->crush_ruleset, pi->type, pi->size);
2382 if (pi->size > ARRAY_SIZE(raw->osds)) {
2383 pr_err_ratelimited("pool %lld ruleset %d type %d too wide: size %d > %zu\n",
2384 pi->id, pi->crush_ruleset, pi->type, pi->size,
2385 ARRAY_SIZE(raw->osds));
2389 len = do_crush(osdmap, ruleno, pps, raw->osds, pi->size,
2390 osdmap->osd_weight, osdmap->max_osd, pi->id);
2392 pr_err("error %d from crush rule %d: pool %lld ruleset %d type %d size %d\n",
2393 len, ruleno, pi->id, pi->crush_ruleset, pi->type,
2399 remove_nonexistent_osds(osdmap, pi, raw);
2402 /* apply pg_upmap[_items] mappings */
2403 static void apply_upmap(struct ceph_osdmap *osdmap,
2404 const struct ceph_pg *pgid,
2405 struct ceph_osds *raw)
2407 struct ceph_pg_mapping *pg;
2410 pg = lookup_pg_mapping(&osdmap->pg_upmap, pgid);
2412 /* make sure targets aren't marked out */
2413 for (i = 0; i < pg->pg_upmap.len; i++) {
2414 int osd = pg->pg_upmap.osds[i];
2416 if (osd != CRUSH_ITEM_NONE &&
2417 osd < osdmap->max_osd &&
2418 osdmap->osd_weight[osd] == 0) {
2419 /* reject/ignore explicit mapping */
2423 for (i = 0; i < pg->pg_upmap.len; i++)
2424 raw->osds[i] = pg->pg_upmap.osds[i];
2425 raw->size = pg->pg_upmap.len;
2429 pg = lookup_pg_mapping(&osdmap->pg_upmap_items, pgid);
2432 * Note: this approach does not allow a bidirectional swap,
2433 * e.g., [[1,2],[2,1]] applied to [0,1,2] -> [0,2,1].
2435 for (i = 0; i < pg->pg_upmap_items.len; i++) {
2436 int from = pg->pg_upmap_items.from_to[i][0];
2437 int to = pg->pg_upmap_items.from_to[i][1];
2439 bool exists = false;
2441 /* make sure replacement doesn't already appear */
2442 for (j = 0; j < raw->size; j++) {
2443 int osd = raw->osds[j];
2449 /* ignore mapping if target is marked out */
2450 if (osd == from && pos < 0 &&
2451 !(to != CRUSH_ITEM_NONE &&
2452 to < osdmap->max_osd &&
2453 osdmap->osd_weight[to] == 0)) {
2457 if (!exists && pos >= 0) {
2458 raw->osds[pos] = to;
2466 * Given raw set, calculate up set and up primary. By definition of an
2467 * up set, the result won't contain nonexistent or down OSDs.
2469 * This is done in-place - on return @set is the up set. If it's
2470 * empty, ->primary will remain undefined.
2472 static void raw_to_up_osds(struct ceph_osdmap *osdmap,
2473 struct ceph_pg_pool_info *pi,
2474 struct ceph_osds *set)
2478 /* ->primary is undefined for a raw set */
2479 BUG_ON(set->primary != -1);
2481 if (ceph_can_shift_osds(pi)) {
2485 for (i = 0; i < set->size; i++) {
2486 if (ceph_osd_is_down(osdmap, set->osds[i])) {
2491 set->osds[i - removed] = set->osds[i];
2493 set->size -= removed;
2495 set->primary = set->osds[0];
2497 /* set down/dne devices to NONE */
2498 for (i = set->size - 1; i >= 0; i--) {
2499 if (ceph_osd_is_down(osdmap, set->osds[i]))
2500 set->osds[i] = CRUSH_ITEM_NONE;
2502 set->primary = set->osds[i];
2507 static void apply_primary_affinity(struct ceph_osdmap *osdmap,
2508 struct ceph_pg_pool_info *pi,
2510 struct ceph_osds *up)
2516 * Do we have any non-default primary_affinity values for these
2519 if (!osdmap->osd_primary_affinity)
2522 for (i = 0; i < up->size; i++) {
2523 int osd = up->osds[i];
2525 if (osd != CRUSH_ITEM_NONE &&
2526 osdmap->osd_primary_affinity[osd] !=
2527 CEPH_OSD_DEFAULT_PRIMARY_AFFINITY) {
2535 * Pick the primary. Feed both the seed (for the pg) and the
2536 * osd into the hash/rng so that a proportional fraction of an
2537 * osd's pgs get rejected as primary.
2539 for (i = 0; i < up->size; i++) {
2540 int osd = up->osds[i];
2543 if (osd == CRUSH_ITEM_NONE)
2546 aff = osdmap->osd_primary_affinity[osd];
2547 if (aff < CEPH_OSD_MAX_PRIMARY_AFFINITY &&
2548 (crush_hash32_2(CRUSH_HASH_RJENKINS1,
2549 pps, osd) >> 16) >= aff) {
2551 * We chose not to use this primary. Note it
2552 * anyway as a fallback in case we don't pick
2553 * anyone else, but keep looking.
2565 up->primary = up->osds[pos];
2567 if (ceph_can_shift_osds(pi) && pos > 0) {
2568 /* move the new primary to the front */
2569 for (i = pos; i > 0; i--)
2570 up->osds[i] = up->osds[i - 1];
2571 up->osds[0] = up->primary;
2576 * Get pg_temp and primary_temp mappings for given PG.
2578 * Note that a PG may have none, only pg_temp, only primary_temp or
2579 * both pg_temp and primary_temp mappings. This means @temp isn't
2580 * always a valid OSD set on return: in the "only primary_temp" case,
2581 * @temp will have its ->primary >= 0 but ->size == 0.
2583 static void get_temp_osds(struct ceph_osdmap *osdmap,
2584 struct ceph_pg_pool_info *pi,
2585 const struct ceph_pg *pgid,
2586 struct ceph_osds *temp)
2588 struct ceph_pg_mapping *pg;
2591 ceph_osds_init(temp);
2594 pg = lookup_pg_mapping(&osdmap->pg_temp, pgid);
2596 for (i = 0; i < pg->pg_temp.len; i++) {
2597 if (ceph_osd_is_down(osdmap, pg->pg_temp.osds[i])) {
2598 if (ceph_can_shift_osds(pi))
2601 temp->osds[temp->size++] = CRUSH_ITEM_NONE;
2603 temp->osds[temp->size++] = pg->pg_temp.osds[i];
2607 /* apply pg_temp's primary */
2608 for (i = 0; i < temp->size; i++) {
2609 if (temp->osds[i] != CRUSH_ITEM_NONE) {
2610 temp->primary = temp->osds[i];
2617 pg = lookup_pg_mapping(&osdmap->primary_temp, pgid);
2619 temp->primary = pg->primary_temp.osd;
2623 * Map a PG to its acting set as well as its up set.
2625 * Acting set is used for data mapping purposes, while up set can be
2626 * recorded for detecting interval changes and deciding whether to
2629 void ceph_pg_to_up_acting_osds(struct ceph_osdmap *osdmap,
2630 struct ceph_pg_pool_info *pi,
2631 const struct ceph_pg *raw_pgid,
2632 struct ceph_osds *up,
2633 struct ceph_osds *acting)
2635 struct ceph_pg pgid;
2638 WARN_ON(pi->id != raw_pgid->pool);
2639 raw_pg_to_pg(pi, raw_pgid, &pgid);
2641 pg_to_raw_osds(osdmap, pi, raw_pgid, up, &pps);
2642 apply_upmap(osdmap, &pgid, up);
2643 raw_to_up_osds(osdmap, pi, up);
2644 apply_primary_affinity(osdmap, pi, pps, up);
2645 get_temp_osds(osdmap, pi, &pgid, acting);
2646 if (!acting->size) {
2647 memcpy(acting->osds, up->osds, up->size * sizeof(up->osds[0]));
2648 acting->size = up->size;
2649 if (acting->primary == -1)
2650 acting->primary = up->primary;
2652 WARN_ON(!osds_valid(up) || !osds_valid(acting));
2655 bool ceph_pg_to_primary_shard(struct ceph_osdmap *osdmap,
2656 struct ceph_pg_pool_info *pi,
2657 const struct ceph_pg *raw_pgid,
2658 struct ceph_spg *spgid)
2660 struct ceph_pg pgid;
2661 struct ceph_osds up, acting;
2664 WARN_ON(pi->id != raw_pgid->pool);
2665 raw_pg_to_pg(pi, raw_pgid, &pgid);
2667 if (ceph_can_shift_osds(pi)) {
2668 spgid->pgid = pgid; /* struct */
2669 spgid->shard = CEPH_SPG_NOSHARD;
2673 ceph_pg_to_up_acting_osds(osdmap, pi, &pgid, &up, &acting);
2674 for (i = 0; i < acting.size; i++) {
2675 if (acting.osds[i] == acting.primary) {
2676 spgid->pgid = pgid; /* struct */
2686 * Return acting primary for given PG, or -1 if none.
2688 int ceph_pg_to_acting_primary(struct ceph_osdmap *osdmap,
2689 const struct ceph_pg *raw_pgid)
2691 struct ceph_pg_pool_info *pi;
2692 struct ceph_osds up, acting;
2694 pi = ceph_pg_pool_by_id(osdmap, raw_pgid->pool);
2698 ceph_pg_to_up_acting_osds(osdmap, pi, raw_pgid, &up, &acting);
2699 return acting.primary;
2701 EXPORT_SYMBOL(ceph_pg_to_acting_primary);