1 // SPDX-License-Identifier: GPL-2.0
3 #include <linux/ceph/ceph_debug.h>
5 #include <linux/module.h>
6 #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 struct crush_name_node {
142 struct rb_node cn_node;
147 static struct crush_name_node *alloc_crush_name(size_t name_len)
149 struct crush_name_node *cn;
151 cn = kmalloc(sizeof(*cn) + name_len + 1, GFP_NOIO);
155 RB_CLEAR_NODE(&cn->cn_node);
159 static void free_crush_name(struct crush_name_node *cn)
161 WARN_ON(!RB_EMPTY_NODE(&cn->cn_node));
166 DEFINE_RB_FUNCS(crush_name, struct crush_name_node, cn_id, cn_node)
168 static int decode_crush_names(void **p, void *end, struct rb_root *root)
172 ceph_decode_32_safe(p, end, n, e_inval);
174 struct crush_name_node *cn;
178 ceph_decode_32_safe(p, end, id, e_inval);
179 ceph_decode_32_safe(p, end, name_len, e_inval);
180 ceph_decode_need(p, end, name_len, e_inval);
182 cn = alloc_crush_name(name_len);
187 memcpy(cn->cn_name, *p, name_len);
188 cn->cn_name[name_len] = '\0';
191 if (!__insert_crush_name(root, cn)) {
203 void clear_crush_names(struct rb_root *root)
205 while (!RB_EMPTY_ROOT(root)) {
206 struct crush_name_node *cn =
207 rb_entry(rb_first(root), struct crush_name_node, cn_node);
209 erase_crush_name(root, cn);
214 static struct crush_choose_arg_map *alloc_choose_arg_map(void)
216 struct crush_choose_arg_map *arg_map;
218 arg_map = kzalloc(sizeof(*arg_map), GFP_NOIO);
222 RB_CLEAR_NODE(&arg_map->node);
226 static void free_choose_arg_map(struct crush_choose_arg_map *arg_map)
231 WARN_ON(!RB_EMPTY_NODE(&arg_map->node));
233 for (i = 0; i < arg_map->size; i++) {
234 struct crush_choose_arg *arg = &arg_map->args[i];
236 for (j = 0; j < arg->weight_set_size; j++)
237 kfree(arg->weight_set[j].weights);
238 kfree(arg->weight_set);
241 kfree(arg_map->args);
246 DEFINE_RB_FUNCS(choose_arg_map, struct crush_choose_arg_map, choose_args_index,
249 void clear_choose_args(struct crush_map *c)
251 while (!RB_EMPTY_ROOT(&c->choose_args)) {
252 struct crush_choose_arg_map *arg_map =
253 rb_entry(rb_first(&c->choose_args),
254 struct crush_choose_arg_map, node);
256 erase_choose_arg_map(&c->choose_args, arg_map);
257 free_choose_arg_map(arg_map);
261 static u32 *decode_array_32_alloc(void **p, void *end, u32 *plen)
267 ceph_decode_32_safe(p, end, len, e_inval);
271 a = kmalloc_array(len, sizeof(u32), GFP_NOIO);
277 ceph_decode_need(p, end, len * sizeof(u32), e_inval);
278 for (i = 0; i < len; i++)
279 a[i] = ceph_decode_32(p);
293 * Assumes @arg is zero-initialized.
295 static int decode_choose_arg(void **p, void *end, struct crush_choose_arg *arg)
299 ceph_decode_32_safe(p, end, arg->weight_set_size, e_inval);
300 if (arg->weight_set_size) {
303 arg->weight_set = kmalloc_array(arg->weight_set_size,
304 sizeof(*arg->weight_set),
306 if (!arg->weight_set)
309 for (i = 0; i < arg->weight_set_size; i++) {
310 struct crush_weight_set *w = &arg->weight_set[i];
312 w->weights = decode_array_32_alloc(p, end, &w->size);
313 if (IS_ERR(w->weights)) {
314 ret = PTR_ERR(w->weights);
321 arg->ids = decode_array_32_alloc(p, end, &arg->ids_size);
322 if (IS_ERR(arg->ids)) {
323 ret = PTR_ERR(arg->ids);
334 static int decode_choose_args(void **p, void *end, struct crush_map *c)
336 struct crush_choose_arg_map *arg_map = NULL;
337 u32 num_choose_arg_maps, num_buckets;
340 ceph_decode_32_safe(p, end, num_choose_arg_maps, e_inval);
341 while (num_choose_arg_maps--) {
342 arg_map = alloc_choose_arg_map();
348 ceph_decode_64_safe(p, end, arg_map->choose_args_index,
350 arg_map->size = c->max_buckets;
351 arg_map->args = kcalloc(arg_map->size, sizeof(*arg_map->args),
353 if (!arg_map->args) {
358 ceph_decode_32_safe(p, end, num_buckets, e_inval);
359 while (num_buckets--) {
360 struct crush_choose_arg *arg;
363 ceph_decode_32_safe(p, end, bucket_index, e_inval);
364 if (bucket_index >= arg_map->size)
367 arg = &arg_map->args[bucket_index];
368 ret = decode_choose_arg(p, end, arg);
373 arg->ids_size != c->buckets[bucket_index]->size)
377 insert_choose_arg_map(&c->choose_args, arg_map);
385 free_choose_arg_map(arg_map);
389 static void crush_finalize(struct crush_map *c)
393 /* Space for the array of pointers to per-bucket workspace */
394 c->working_size = sizeof(struct crush_work) +
395 c->max_buckets * sizeof(struct crush_work_bucket *);
397 for (b = 0; b < c->max_buckets; b++) {
401 switch (c->buckets[b]->alg) {
404 * The base case, permutation variables and
405 * the pointer to the permutation array.
407 c->working_size += sizeof(struct crush_work_bucket);
410 /* Every bucket has a permutation array. */
411 c->working_size += c->buckets[b]->size * sizeof(__u32);
415 static struct crush_map *crush_decode(void *pbyval, void *end)
421 void *start = pbyval;
424 dout("crush_decode %p to %p len %d\n", *p, end, (int)(end - *p));
426 c = kzalloc(sizeof(*c), GFP_NOFS);
428 return ERR_PTR(-ENOMEM);
430 c->type_names = RB_ROOT;
432 c->choose_args = RB_ROOT;
434 /* set tunables to default values */
435 c->choose_local_tries = 2;
436 c->choose_local_fallback_tries = 5;
437 c->choose_total_tries = 19;
438 c->chooseleaf_descend_once = 0;
440 ceph_decode_need(p, end, 4*sizeof(u32), bad);
441 magic = ceph_decode_32(p);
442 if (magic != CRUSH_MAGIC) {
443 pr_err("crush_decode magic %x != current %x\n",
444 (unsigned int)magic, (unsigned int)CRUSH_MAGIC);
447 c->max_buckets = ceph_decode_32(p);
448 c->max_rules = ceph_decode_32(p);
449 c->max_devices = ceph_decode_32(p);
451 c->buckets = kcalloc(c->max_buckets, sizeof(*c->buckets), GFP_NOFS);
452 if (c->buckets == NULL)
454 c->rules = kcalloc(c->max_rules, sizeof(*c->rules), GFP_NOFS);
455 if (c->rules == NULL)
459 for (i = 0; i < c->max_buckets; i++) {
462 struct crush_bucket *b;
464 ceph_decode_32_safe(p, end, alg, bad);
466 c->buckets[i] = NULL;
469 dout("crush_decode bucket %d off %x %p to %p\n",
470 i, (int)(*p-start), *p, end);
473 case CRUSH_BUCKET_UNIFORM:
474 size = sizeof(struct crush_bucket_uniform);
476 case CRUSH_BUCKET_LIST:
477 size = sizeof(struct crush_bucket_list);
479 case CRUSH_BUCKET_TREE:
480 size = sizeof(struct crush_bucket_tree);
482 case CRUSH_BUCKET_STRAW:
483 size = sizeof(struct crush_bucket_straw);
485 case CRUSH_BUCKET_STRAW2:
486 size = sizeof(struct crush_bucket_straw2);
492 b = c->buckets[i] = kzalloc(size, GFP_NOFS);
496 ceph_decode_need(p, end, 4*sizeof(u32), bad);
497 b->id = ceph_decode_32(p);
498 b->type = ceph_decode_16(p);
499 b->alg = ceph_decode_8(p);
500 b->hash = ceph_decode_8(p);
501 b->weight = ceph_decode_32(p);
502 b->size = ceph_decode_32(p);
504 dout("crush_decode bucket size %d off %x %p to %p\n",
505 b->size, (int)(*p-start), *p, end);
507 b->items = kcalloc(b->size, sizeof(__s32), GFP_NOFS);
508 if (b->items == NULL)
511 ceph_decode_need(p, end, b->size*sizeof(u32), bad);
512 for (j = 0; j < b->size; j++)
513 b->items[j] = ceph_decode_32(p);
516 case CRUSH_BUCKET_UNIFORM:
517 err = crush_decode_uniform_bucket(p, end,
518 (struct crush_bucket_uniform *)b);
522 case CRUSH_BUCKET_LIST:
523 err = crush_decode_list_bucket(p, end,
524 (struct crush_bucket_list *)b);
528 case CRUSH_BUCKET_TREE:
529 err = crush_decode_tree_bucket(p, end,
530 (struct crush_bucket_tree *)b);
534 case CRUSH_BUCKET_STRAW:
535 err = crush_decode_straw_bucket(p, end,
536 (struct crush_bucket_straw *)b);
540 case CRUSH_BUCKET_STRAW2:
541 err = crush_decode_straw2_bucket(p, end,
542 (struct crush_bucket_straw2 *)b);
550 dout("rule vec is %p\n", c->rules);
551 for (i = 0; i < c->max_rules; i++) {
553 struct crush_rule *r;
555 ceph_decode_32_safe(p, end, yes, bad);
557 dout("crush_decode NO rule %d off %x %p to %p\n",
558 i, (int)(*p-start), *p, end);
563 dout("crush_decode rule %d off %x %p to %p\n",
564 i, (int)(*p-start), *p, end);
567 ceph_decode_32_safe(p, end, yes, bad);
568 #if BITS_PER_LONG == 32
569 if (yes > (ULONG_MAX - sizeof(*r))
570 / sizeof(struct crush_rule_step))
573 r = kmalloc(struct_size(r, steps, yes), GFP_NOFS);
577 dout(" rule %d is at %p\n", i, r);
579 ceph_decode_copy_safe(p, end, &r->mask, 4, bad); /* 4 u8's */
580 ceph_decode_need(p, end, r->len*3*sizeof(u32), bad);
581 for (j = 0; j < r->len; j++) {
582 r->steps[j].op = ceph_decode_32(p);
583 r->steps[j].arg1 = ceph_decode_32(p);
584 r->steps[j].arg2 = ceph_decode_32(p);
588 err = decode_crush_names(p, end, &c->type_names);
592 err = decode_crush_names(p, end, &c->names);
596 ceph_decode_skip_map(p, end, 32, string, bad); /* rule_name_map */
599 ceph_decode_need(p, end, 3*sizeof(u32), done);
600 c->choose_local_tries = ceph_decode_32(p);
601 c->choose_local_fallback_tries = ceph_decode_32(p);
602 c->choose_total_tries = ceph_decode_32(p);
603 dout("crush decode tunable choose_local_tries = %d\n",
604 c->choose_local_tries);
605 dout("crush decode tunable choose_local_fallback_tries = %d\n",
606 c->choose_local_fallback_tries);
607 dout("crush decode tunable choose_total_tries = %d\n",
608 c->choose_total_tries);
610 ceph_decode_need(p, end, sizeof(u32), done);
611 c->chooseleaf_descend_once = ceph_decode_32(p);
612 dout("crush decode tunable chooseleaf_descend_once = %d\n",
613 c->chooseleaf_descend_once);
615 ceph_decode_need(p, end, sizeof(u8), done);
616 c->chooseleaf_vary_r = ceph_decode_8(p);
617 dout("crush decode tunable chooseleaf_vary_r = %d\n",
618 c->chooseleaf_vary_r);
620 /* skip straw_calc_version, allowed_bucket_algs */
621 ceph_decode_need(p, end, sizeof(u8) + sizeof(u32), done);
622 *p += sizeof(u8) + sizeof(u32);
624 ceph_decode_need(p, end, sizeof(u8), done);
625 c->chooseleaf_stable = ceph_decode_8(p);
626 dout("crush decode tunable chooseleaf_stable = %d\n",
627 c->chooseleaf_stable);
631 ceph_decode_skip_map(p, end, 32, 32, bad);
633 ceph_decode_skip_map(p, end, 32, string, bad);
635 ceph_decode_skip_map_of_map(p, end, 32, 32, 32, bad);
639 err = decode_choose_args(p, end, c);
646 dout("crush_decode success\n");
652 dout("crush_decode fail %d\n", err);
661 int ceph_pg_compare(const struct ceph_pg *lhs, const struct ceph_pg *rhs)
663 if (lhs->pool < rhs->pool)
665 if (lhs->pool > rhs->pool)
667 if (lhs->seed < rhs->seed)
669 if (lhs->seed > rhs->seed)
675 int ceph_spg_compare(const struct ceph_spg *lhs, const struct ceph_spg *rhs)
679 ret = ceph_pg_compare(&lhs->pgid, &rhs->pgid);
683 if (lhs->shard < rhs->shard)
685 if (lhs->shard > rhs->shard)
691 static struct ceph_pg_mapping *alloc_pg_mapping(size_t payload_len)
693 struct ceph_pg_mapping *pg;
695 pg = kmalloc(sizeof(*pg) + payload_len, GFP_NOIO);
699 RB_CLEAR_NODE(&pg->node);
703 static void free_pg_mapping(struct ceph_pg_mapping *pg)
705 WARN_ON(!RB_EMPTY_NODE(&pg->node));
711 * rbtree of pg_mapping for handling pg_temp (explicit mapping of pgid
712 * to a set of osds) and primary_temp (explicit primary setting)
714 DEFINE_RB_FUNCS2(pg_mapping, struct ceph_pg_mapping, pgid, ceph_pg_compare,
715 RB_BYPTR, const struct ceph_pg *, node)
718 * rbtree of pg pool info
720 DEFINE_RB_FUNCS(pg_pool, struct ceph_pg_pool_info, id, node)
722 struct ceph_pg_pool_info *ceph_pg_pool_by_id(struct ceph_osdmap *map, u64 id)
724 return lookup_pg_pool(&map->pg_pools, id);
727 const char *ceph_pg_pool_name_by_id(struct ceph_osdmap *map, u64 id)
729 struct ceph_pg_pool_info *pi;
731 if (id == CEPH_NOPOOL)
734 if (WARN_ON_ONCE(id > (u64) INT_MAX))
737 pi = lookup_pg_pool(&map->pg_pools, id);
738 return pi ? pi->name : NULL;
740 EXPORT_SYMBOL(ceph_pg_pool_name_by_id);
742 int ceph_pg_poolid_by_name(struct ceph_osdmap *map, const char *name)
746 for (rbp = rb_first(&map->pg_pools); rbp; rbp = rb_next(rbp)) {
747 struct ceph_pg_pool_info *pi =
748 rb_entry(rbp, struct ceph_pg_pool_info, node);
749 if (pi->name && strcmp(pi->name, name) == 0)
754 EXPORT_SYMBOL(ceph_pg_poolid_by_name);
756 u64 ceph_pg_pool_flags(struct ceph_osdmap *map, u64 id)
758 struct ceph_pg_pool_info *pi;
760 pi = lookup_pg_pool(&map->pg_pools, id);
761 return pi ? pi->flags : 0;
763 EXPORT_SYMBOL(ceph_pg_pool_flags);
765 static void __remove_pg_pool(struct rb_root *root, struct ceph_pg_pool_info *pi)
767 erase_pg_pool(root, pi);
772 static int decode_pool(void **p, void *end, struct ceph_pg_pool_info *pi)
778 ceph_decode_need(p, end, 2 + 4, bad);
779 ev = ceph_decode_8(p); /* encoding version */
780 cv = ceph_decode_8(p); /* compat version */
782 pr_warn("got v %d < 5 cv %d of ceph_pg_pool\n", ev, cv);
786 pr_warn("got v %d cv %d > 9 of ceph_pg_pool\n", ev, cv);
789 len = ceph_decode_32(p);
790 ceph_decode_need(p, end, len, bad);
793 pi->type = ceph_decode_8(p);
794 pi->size = ceph_decode_8(p);
795 pi->crush_ruleset = ceph_decode_8(p);
796 pi->object_hash = ceph_decode_8(p);
798 pi->pg_num = ceph_decode_32(p);
799 pi->pgp_num = ceph_decode_32(p);
801 *p += 4 + 4; /* skip lpg* */
802 *p += 4; /* skip last_change */
803 *p += 8 + 4; /* skip snap_seq, snap_epoch */
806 num = ceph_decode_32(p);
808 *p += 8; /* snapid key */
809 *p += 1 + 1; /* versions */
810 len = ceph_decode_32(p);
814 /* skip removed_snaps */
815 num = ceph_decode_32(p);
818 *p += 8; /* skip auid */
819 pi->flags = ceph_decode_64(p);
820 *p += 4; /* skip crash_replay_interval */
823 pi->min_size = ceph_decode_8(p);
825 pi->min_size = pi->size - pi->size / 2;
828 *p += 8 + 8; /* skip quota_max_* */
832 num = ceph_decode_32(p);
835 *p += 8; /* skip tier_of */
836 *p += 1; /* skip cache_mode */
838 pi->read_tier = ceph_decode_64(p);
839 pi->write_tier = ceph_decode_64(p);
846 /* skip properties */
847 num = ceph_decode_32(p);
849 len = ceph_decode_32(p);
851 len = ceph_decode_32(p);
857 /* skip hit_set_params */
858 *p += 1 + 1; /* versions */
859 len = ceph_decode_32(p);
862 *p += 4; /* skip hit_set_period */
863 *p += 4; /* skip hit_set_count */
867 *p += 4; /* skip stripe_width */
870 *p += 8; /* skip target_max_bytes */
871 *p += 8; /* skip target_max_objects */
872 *p += 4; /* skip cache_target_dirty_ratio_micro */
873 *p += 4; /* skip cache_target_full_ratio_micro */
874 *p += 4; /* skip cache_min_flush_age */
875 *p += 4; /* skip cache_min_evict_age */
879 /* skip erasure_code_profile */
880 len = ceph_decode_32(p);
885 * last_force_op_resend_preluminous, will be overridden if the
886 * map was encoded with RESEND_ON_SPLIT
889 pi->last_force_request_resend = ceph_decode_32(p);
891 pi->last_force_request_resend = 0;
894 *p += 4; /* skip min_read_recency_for_promote */
897 *p += 8; /* skip expected_num_objects */
900 *p += 4; /* skip cache_target_dirty_high_ratio_micro */
903 *p += 4; /* skip min_write_recency_for_promote */
906 *p += 1; /* skip use_gmt_hitset */
909 *p += 1; /* skip fast_read */
912 *p += 4; /* skip hit_set_grade_decay_rate */
913 *p += 4; /* skip hit_set_search_last_n */
918 *p += 1 + 1; /* versions */
919 len = ceph_decode_32(p);
924 pi->last_force_request_resend = ceph_decode_32(p);
926 /* ignore the rest */
936 static int decode_pool_names(void **p, void *end, struct ceph_osdmap *map)
938 struct ceph_pg_pool_info *pi;
942 ceph_decode_32_safe(p, end, num, bad);
943 dout(" %d pool names\n", num);
945 ceph_decode_64_safe(p, end, pool, bad);
946 ceph_decode_32_safe(p, end, len, bad);
947 dout(" pool %llu len %d\n", pool, len);
948 ceph_decode_need(p, end, len, bad);
949 pi = lookup_pg_pool(&map->pg_pools, pool);
951 char *name = kstrndup(*p, len, GFP_NOFS);
957 dout(" name is %s\n", pi->name);
970 * workspace_manager framework borrowed from fs/btrfs/compression.c.
971 * Two simplifications: there is only one type of workspace and there
972 * is always at least one workspace.
974 static struct crush_work *alloc_workspace(const struct crush_map *c)
976 struct crush_work *work;
979 WARN_ON(!c->working_size);
980 work_size = crush_work_size(c, CEPH_PG_MAX_SIZE);
981 dout("%s work_size %zu bytes\n", __func__, work_size);
983 work = ceph_kvmalloc(work_size, GFP_NOIO);
987 INIT_LIST_HEAD(&work->item);
988 crush_init_workspace(c, work);
992 static void free_workspace(struct crush_work *work)
994 WARN_ON(!list_empty(&work->item));
998 static void init_workspace_manager(struct workspace_manager *wsm)
1000 INIT_LIST_HEAD(&wsm->idle_ws);
1001 spin_lock_init(&wsm->ws_lock);
1002 atomic_set(&wsm->total_ws, 0);
1004 init_waitqueue_head(&wsm->ws_wait);
1007 static void add_initial_workspace(struct workspace_manager *wsm,
1008 struct crush_work *work)
1010 WARN_ON(!list_empty(&wsm->idle_ws));
1012 list_add(&work->item, &wsm->idle_ws);
1013 atomic_set(&wsm->total_ws, 1);
1017 static void cleanup_workspace_manager(struct workspace_manager *wsm)
1019 struct crush_work *work;
1021 while (!list_empty(&wsm->idle_ws)) {
1022 work = list_first_entry(&wsm->idle_ws, struct crush_work,
1024 list_del_init(&work->item);
1025 free_workspace(work);
1027 atomic_set(&wsm->total_ws, 0);
1032 * Finds an available workspace or allocates a new one. If it's not
1033 * possible to allocate a new one, waits until there is one.
1035 static struct crush_work *get_workspace(struct workspace_manager *wsm,
1036 const struct crush_map *c)
1038 struct crush_work *work;
1039 int cpus = num_online_cpus();
1042 spin_lock(&wsm->ws_lock);
1043 if (!list_empty(&wsm->idle_ws)) {
1044 work = list_first_entry(&wsm->idle_ws, struct crush_work,
1046 list_del_init(&work->item);
1048 spin_unlock(&wsm->ws_lock);
1052 if (atomic_read(&wsm->total_ws) > cpus) {
1055 spin_unlock(&wsm->ws_lock);
1056 prepare_to_wait(&wsm->ws_wait, &wait, TASK_UNINTERRUPTIBLE);
1057 if (atomic_read(&wsm->total_ws) > cpus && !wsm->free_ws)
1059 finish_wait(&wsm->ws_wait, &wait);
1062 atomic_inc(&wsm->total_ws);
1063 spin_unlock(&wsm->ws_lock);
1065 work = alloc_workspace(c);
1067 atomic_dec(&wsm->total_ws);
1068 wake_up(&wsm->ws_wait);
1071 * Do not return the error but go back to waiting. We
1072 * have the inital workspace and the CRUSH computation
1073 * time is bounded so we will get it eventually.
1075 WARN_ON(atomic_read(&wsm->total_ws) < 1);
1082 * Puts a workspace back on the list or frees it if we have enough
1083 * idle ones sitting around.
1085 static void put_workspace(struct workspace_manager *wsm,
1086 struct crush_work *work)
1088 spin_lock(&wsm->ws_lock);
1089 if (wsm->free_ws <= num_online_cpus()) {
1090 list_add(&work->item, &wsm->idle_ws);
1092 spin_unlock(&wsm->ws_lock);
1095 spin_unlock(&wsm->ws_lock);
1097 free_workspace(work);
1098 atomic_dec(&wsm->total_ws);
1100 if (wq_has_sleeper(&wsm->ws_wait))
1101 wake_up(&wsm->ws_wait);
1107 struct ceph_osdmap *ceph_osdmap_alloc(void)
1109 struct ceph_osdmap *map;
1111 map = kzalloc(sizeof(*map), GFP_NOIO);
1115 map->pg_pools = RB_ROOT;
1117 map->pg_temp = RB_ROOT;
1118 map->primary_temp = RB_ROOT;
1119 map->pg_upmap = RB_ROOT;
1120 map->pg_upmap_items = RB_ROOT;
1122 init_workspace_manager(&map->crush_wsm);
1127 void ceph_osdmap_destroy(struct ceph_osdmap *map)
1129 dout("osdmap_destroy %p\n", map);
1132 crush_destroy(map->crush);
1133 cleanup_workspace_manager(&map->crush_wsm);
1135 while (!RB_EMPTY_ROOT(&map->pg_temp)) {
1136 struct ceph_pg_mapping *pg =
1137 rb_entry(rb_first(&map->pg_temp),
1138 struct ceph_pg_mapping, node);
1139 erase_pg_mapping(&map->pg_temp, pg);
1140 free_pg_mapping(pg);
1142 while (!RB_EMPTY_ROOT(&map->primary_temp)) {
1143 struct ceph_pg_mapping *pg =
1144 rb_entry(rb_first(&map->primary_temp),
1145 struct ceph_pg_mapping, node);
1146 erase_pg_mapping(&map->primary_temp, pg);
1147 free_pg_mapping(pg);
1149 while (!RB_EMPTY_ROOT(&map->pg_upmap)) {
1150 struct ceph_pg_mapping *pg =
1151 rb_entry(rb_first(&map->pg_upmap),
1152 struct ceph_pg_mapping, node);
1153 rb_erase(&pg->node, &map->pg_upmap);
1156 while (!RB_EMPTY_ROOT(&map->pg_upmap_items)) {
1157 struct ceph_pg_mapping *pg =
1158 rb_entry(rb_first(&map->pg_upmap_items),
1159 struct ceph_pg_mapping, node);
1160 rb_erase(&pg->node, &map->pg_upmap_items);
1163 while (!RB_EMPTY_ROOT(&map->pg_pools)) {
1164 struct ceph_pg_pool_info *pi =
1165 rb_entry(rb_first(&map->pg_pools),
1166 struct ceph_pg_pool_info, node);
1167 __remove_pg_pool(&map->pg_pools, pi);
1169 kvfree(map->osd_state);
1170 kvfree(map->osd_weight);
1171 kvfree(map->osd_addr);
1172 kvfree(map->osd_primary_affinity);
1177 * Adjust max_osd value, (re)allocate arrays.
1179 * The new elements are properly initialized.
1181 static int osdmap_set_max_osd(struct ceph_osdmap *map, u32 max)
1185 struct ceph_entity_addr *addr;
1189 dout("%s old %u new %u\n", __func__, map->max_osd, max);
1190 if (max == map->max_osd)
1193 state = ceph_kvmalloc(array_size(max, sizeof(*state)), GFP_NOFS);
1194 weight = ceph_kvmalloc(array_size(max, sizeof(*weight)), GFP_NOFS);
1195 addr = ceph_kvmalloc(array_size(max, sizeof(*addr)), GFP_NOFS);
1196 if (!state || !weight || !addr) {
1203 to_copy = min(map->max_osd, max);
1204 if (map->osd_state) {
1205 memcpy(state, map->osd_state, to_copy * sizeof(*state));
1206 memcpy(weight, map->osd_weight, to_copy * sizeof(*weight));
1207 memcpy(addr, map->osd_addr, to_copy * sizeof(*addr));
1208 kvfree(map->osd_state);
1209 kvfree(map->osd_weight);
1210 kvfree(map->osd_addr);
1213 map->osd_state = state;
1214 map->osd_weight = weight;
1215 map->osd_addr = addr;
1216 for (i = map->max_osd; i < max; i++) {
1217 map->osd_state[i] = 0;
1218 map->osd_weight[i] = CEPH_OSD_OUT;
1219 memset(map->osd_addr + i, 0, sizeof(*map->osd_addr));
1222 if (map->osd_primary_affinity) {
1225 affinity = ceph_kvmalloc(array_size(max, sizeof(*affinity)),
1230 memcpy(affinity, map->osd_primary_affinity,
1231 to_copy * sizeof(*affinity));
1232 kvfree(map->osd_primary_affinity);
1234 map->osd_primary_affinity = affinity;
1235 for (i = map->max_osd; i < max; i++)
1236 map->osd_primary_affinity[i] =
1237 CEPH_OSD_DEFAULT_PRIMARY_AFFINITY;
1245 static int osdmap_set_crush(struct ceph_osdmap *map, struct crush_map *crush)
1247 struct crush_work *work;
1250 return PTR_ERR(crush);
1252 work = alloc_workspace(crush);
1254 crush_destroy(crush);
1259 crush_destroy(map->crush);
1260 cleanup_workspace_manager(&map->crush_wsm);
1262 add_initial_workspace(&map->crush_wsm, work);
1266 #define OSDMAP_WRAPPER_COMPAT_VER 7
1267 #define OSDMAP_CLIENT_DATA_COMPAT_VER 1
1270 * Return 0 or error. On success, *v is set to 0 for old (v6) osdmaps,
1271 * to struct_v of the client_data section for new (v7 and above)
1274 static int get_osdmap_client_data_v(void **p, void *end,
1275 const char *prefix, u8 *v)
1279 ceph_decode_8_safe(p, end, struct_v, e_inval);
1280 if (struct_v >= 7) {
1283 ceph_decode_8_safe(p, end, struct_compat, e_inval);
1284 if (struct_compat > OSDMAP_WRAPPER_COMPAT_VER) {
1285 pr_warn("got v %d cv %d > %d of %s ceph_osdmap\n",
1286 struct_v, struct_compat,
1287 OSDMAP_WRAPPER_COMPAT_VER, prefix);
1290 *p += 4; /* ignore wrapper struct_len */
1292 ceph_decode_8_safe(p, end, struct_v, e_inval);
1293 ceph_decode_8_safe(p, end, struct_compat, e_inval);
1294 if (struct_compat > OSDMAP_CLIENT_DATA_COMPAT_VER) {
1295 pr_warn("got v %d cv %d > %d of %s ceph_osdmap client data\n",
1296 struct_v, struct_compat,
1297 OSDMAP_CLIENT_DATA_COMPAT_VER, prefix);
1300 *p += 4; /* ignore client data struct_len */
1305 ceph_decode_16_safe(p, end, version, e_inval);
1307 pr_warn("got v %d < 6 of %s ceph_osdmap\n",
1312 /* old osdmap enconding */
1323 static int __decode_pools(void **p, void *end, struct ceph_osdmap *map,
1328 ceph_decode_32_safe(p, end, n, e_inval);
1330 struct ceph_pg_pool_info *pi;
1334 ceph_decode_64_safe(p, end, pool, e_inval);
1336 pi = lookup_pg_pool(&map->pg_pools, pool);
1337 if (!incremental || !pi) {
1338 pi = kzalloc(sizeof(*pi), GFP_NOFS);
1342 RB_CLEAR_NODE(&pi->node);
1345 if (!__insert_pg_pool(&map->pg_pools, pi)) {
1351 ret = decode_pool(p, end, pi);
1362 static int decode_pools(void **p, void *end, struct ceph_osdmap *map)
1364 return __decode_pools(p, end, map, false);
1367 static int decode_new_pools(void **p, void *end, struct ceph_osdmap *map)
1369 return __decode_pools(p, end, map, true);
1372 typedef struct ceph_pg_mapping *(*decode_mapping_fn_t)(void **, void *, bool);
1374 static int decode_pg_mapping(void **p, void *end, struct rb_root *mapping_root,
1375 decode_mapping_fn_t fn, bool incremental)
1379 WARN_ON(!incremental && !fn);
1381 ceph_decode_32_safe(p, end, n, e_inval);
1383 struct ceph_pg_mapping *pg;
1384 struct ceph_pg pgid;
1387 ret = ceph_decode_pgid(p, end, &pgid);
1391 pg = lookup_pg_mapping(mapping_root, &pgid);
1393 WARN_ON(!incremental);
1394 erase_pg_mapping(mapping_root, pg);
1395 free_pg_mapping(pg);
1399 pg = fn(p, end, incremental);
1404 pg->pgid = pgid; /* struct */
1405 insert_pg_mapping(mapping_root, pg);
1416 static struct ceph_pg_mapping *__decode_pg_temp(void **p, void *end,
1419 struct ceph_pg_mapping *pg;
1422 ceph_decode_32_safe(p, end, len, e_inval);
1423 if (len == 0 && incremental)
1424 return NULL; /* new_pg_temp: [] to remove */
1425 if (len > (SIZE_MAX - sizeof(*pg)) / sizeof(u32))
1426 return ERR_PTR(-EINVAL);
1428 ceph_decode_need(p, end, len * sizeof(u32), e_inval);
1429 pg = alloc_pg_mapping(len * sizeof(u32));
1431 return ERR_PTR(-ENOMEM);
1433 pg->pg_temp.len = len;
1434 for (i = 0; i < len; i++)
1435 pg->pg_temp.osds[i] = ceph_decode_32(p);
1440 return ERR_PTR(-EINVAL);
1443 static int decode_pg_temp(void **p, void *end, struct ceph_osdmap *map)
1445 return decode_pg_mapping(p, end, &map->pg_temp, __decode_pg_temp,
1449 static int decode_new_pg_temp(void **p, void *end, struct ceph_osdmap *map)
1451 return decode_pg_mapping(p, end, &map->pg_temp, __decode_pg_temp,
1455 static struct ceph_pg_mapping *__decode_primary_temp(void **p, void *end,
1458 struct ceph_pg_mapping *pg;
1461 ceph_decode_32_safe(p, end, osd, e_inval);
1462 if (osd == (u32)-1 && incremental)
1463 return NULL; /* new_primary_temp: -1 to remove */
1465 pg = alloc_pg_mapping(0);
1467 return ERR_PTR(-ENOMEM);
1469 pg->primary_temp.osd = osd;
1473 return ERR_PTR(-EINVAL);
1476 static int decode_primary_temp(void **p, void *end, struct ceph_osdmap *map)
1478 return decode_pg_mapping(p, end, &map->primary_temp,
1479 __decode_primary_temp, false);
1482 static int decode_new_primary_temp(void **p, void *end,
1483 struct ceph_osdmap *map)
1485 return decode_pg_mapping(p, end, &map->primary_temp,
1486 __decode_primary_temp, true);
1489 u32 ceph_get_primary_affinity(struct ceph_osdmap *map, int osd)
1491 BUG_ON(osd >= map->max_osd);
1493 if (!map->osd_primary_affinity)
1494 return CEPH_OSD_DEFAULT_PRIMARY_AFFINITY;
1496 return map->osd_primary_affinity[osd];
1499 static int set_primary_affinity(struct ceph_osdmap *map, int osd, u32 aff)
1501 BUG_ON(osd >= map->max_osd);
1503 if (!map->osd_primary_affinity) {
1506 map->osd_primary_affinity = ceph_kvmalloc(
1507 array_size(map->max_osd, sizeof(*map->osd_primary_affinity)),
1509 if (!map->osd_primary_affinity)
1512 for (i = 0; i < map->max_osd; i++)
1513 map->osd_primary_affinity[i] =
1514 CEPH_OSD_DEFAULT_PRIMARY_AFFINITY;
1517 map->osd_primary_affinity[osd] = aff;
1522 static int decode_primary_affinity(void **p, void *end,
1523 struct ceph_osdmap *map)
1527 ceph_decode_32_safe(p, end, len, e_inval);
1529 kvfree(map->osd_primary_affinity);
1530 map->osd_primary_affinity = NULL;
1533 if (len != map->max_osd)
1536 ceph_decode_need(p, end, map->max_osd*sizeof(u32), e_inval);
1538 for (i = 0; i < map->max_osd; i++) {
1541 ret = set_primary_affinity(map, i, ceph_decode_32(p));
1552 static int decode_new_primary_affinity(void **p, void *end,
1553 struct ceph_osdmap *map)
1557 ceph_decode_32_safe(p, end, n, e_inval);
1562 ceph_decode_32_safe(p, end, osd, e_inval);
1563 ceph_decode_32_safe(p, end, aff, e_inval);
1565 ret = set_primary_affinity(map, osd, aff);
1569 pr_info("osd%d primary-affinity 0x%x\n", osd, aff);
1578 static struct ceph_pg_mapping *__decode_pg_upmap(void **p, void *end,
1581 return __decode_pg_temp(p, end, false);
1584 static int decode_pg_upmap(void **p, void *end, struct ceph_osdmap *map)
1586 return decode_pg_mapping(p, end, &map->pg_upmap, __decode_pg_upmap,
1590 static int decode_new_pg_upmap(void **p, void *end, struct ceph_osdmap *map)
1592 return decode_pg_mapping(p, end, &map->pg_upmap, __decode_pg_upmap,
1596 static int decode_old_pg_upmap(void **p, void *end, struct ceph_osdmap *map)
1598 return decode_pg_mapping(p, end, &map->pg_upmap, NULL, true);
1601 static struct ceph_pg_mapping *__decode_pg_upmap_items(void **p, void *end,
1604 struct ceph_pg_mapping *pg;
1607 ceph_decode_32_safe(p, end, len, e_inval);
1608 if (len > (SIZE_MAX - sizeof(*pg)) / (2 * sizeof(u32)))
1609 return ERR_PTR(-EINVAL);
1611 ceph_decode_need(p, end, 2 * len * sizeof(u32), e_inval);
1612 pg = alloc_pg_mapping(2 * len * sizeof(u32));
1614 return ERR_PTR(-ENOMEM);
1616 pg->pg_upmap_items.len = len;
1617 for (i = 0; i < len; i++) {
1618 pg->pg_upmap_items.from_to[i][0] = ceph_decode_32(p);
1619 pg->pg_upmap_items.from_to[i][1] = ceph_decode_32(p);
1625 return ERR_PTR(-EINVAL);
1628 static int decode_pg_upmap_items(void **p, void *end, struct ceph_osdmap *map)
1630 return decode_pg_mapping(p, end, &map->pg_upmap_items,
1631 __decode_pg_upmap_items, false);
1634 static int decode_new_pg_upmap_items(void **p, void *end,
1635 struct ceph_osdmap *map)
1637 return decode_pg_mapping(p, end, &map->pg_upmap_items,
1638 __decode_pg_upmap_items, true);
1641 static int decode_old_pg_upmap_items(void **p, void *end,
1642 struct ceph_osdmap *map)
1644 return decode_pg_mapping(p, end, &map->pg_upmap_items, NULL, true);
1648 * decode a full map.
1650 static int osdmap_decode(void **p, void *end, struct ceph_osdmap *map)
1659 dout("%s %p to %p len %d\n", __func__, *p, end, (int)(end - *p));
1661 err = get_osdmap_client_data_v(p, end, "full", &struct_v);
1665 /* fsid, epoch, created, modified */
1666 ceph_decode_need(p, end, sizeof(map->fsid) + sizeof(u32) +
1667 sizeof(map->created) + sizeof(map->modified), e_inval);
1668 ceph_decode_copy(p, &map->fsid, sizeof(map->fsid));
1669 epoch = map->epoch = ceph_decode_32(p);
1670 ceph_decode_copy(p, &map->created, sizeof(map->created));
1671 ceph_decode_copy(p, &map->modified, sizeof(map->modified));
1674 err = decode_pools(p, end, map);
1679 err = decode_pool_names(p, end, map);
1683 ceph_decode_32_safe(p, end, map->pool_max, e_inval);
1685 ceph_decode_32_safe(p, end, map->flags, e_inval);
1688 ceph_decode_32_safe(p, end, max, e_inval);
1690 /* (re)alloc osd arrays */
1691 err = osdmap_set_max_osd(map, max);
1695 /* osd_state, osd_weight, osd_addrs->client_addr */
1696 ceph_decode_need(p, end, 3*sizeof(u32) +
1697 map->max_osd*(struct_v >= 5 ? sizeof(u32) :
1699 sizeof(*map->osd_weight), e_inval);
1700 if (ceph_decode_32(p) != map->max_osd)
1703 if (struct_v >= 5) {
1704 for (i = 0; i < map->max_osd; i++)
1705 map->osd_state[i] = ceph_decode_32(p);
1707 for (i = 0; i < map->max_osd; i++)
1708 map->osd_state[i] = ceph_decode_8(p);
1711 if (ceph_decode_32(p) != map->max_osd)
1714 for (i = 0; i < map->max_osd; i++)
1715 map->osd_weight[i] = ceph_decode_32(p);
1717 if (ceph_decode_32(p) != map->max_osd)
1720 for (i = 0; i < map->max_osd; i++) {
1721 err = ceph_decode_entity_addr(p, end, &map->osd_addr[i]);
1727 err = decode_pg_temp(p, end, map);
1732 if (struct_v >= 1) {
1733 err = decode_primary_temp(p, end, map);
1738 /* primary_affinity */
1739 if (struct_v >= 2) {
1740 err = decode_primary_affinity(p, end, map);
1744 WARN_ON(map->osd_primary_affinity);
1748 ceph_decode_32_safe(p, end, len, e_inval);
1749 err = osdmap_set_crush(map, crush_decode(*p, min(*p + len, end)));
1754 if (struct_v >= 3) {
1755 /* erasure_code_profiles */
1756 ceph_decode_skip_map_of_map(p, end, string, string, string,
1760 if (struct_v >= 4) {
1761 err = decode_pg_upmap(p, end, map);
1765 err = decode_pg_upmap_items(p, end, map);
1769 WARN_ON(!RB_EMPTY_ROOT(&map->pg_upmap));
1770 WARN_ON(!RB_EMPTY_ROOT(&map->pg_upmap_items));
1773 /* ignore the rest */
1776 dout("full osdmap epoch %d max_osd %d\n", map->epoch, map->max_osd);
1782 pr_err("corrupt full osdmap (%d) epoch %d off %d (%p of %p-%p)\n",
1783 err, epoch, (int)(*p - start), *p, start, end);
1784 print_hex_dump(KERN_DEBUG, "osdmap: ",
1785 DUMP_PREFIX_OFFSET, 16, 1,
1786 start, end - start, true);
1791 * Allocate and decode a full map.
1793 struct ceph_osdmap *ceph_osdmap_decode(void **p, void *end)
1795 struct ceph_osdmap *map;
1798 map = ceph_osdmap_alloc();
1800 return ERR_PTR(-ENOMEM);
1802 ret = osdmap_decode(p, end, map);
1804 ceph_osdmap_destroy(map);
1805 return ERR_PTR(ret);
1812 * Encoding order is (new_up_client, new_state, new_weight). Need to
1813 * apply in the (new_weight, new_state, new_up_client) order, because
1814 * an incremental map may look like e.g.
1816 * new_up_client: { osd=6, addr=... } # set osd_state and addr
1817 * new_state: { osd=6, xorstate=EXISTS } # clear osd_state
1819 static int decode_new_up_state_weight(void **p, void *end, u8 struct_v,
1820 struct ceph_osdmap *map)
1822 void *new_up_client;
1824 void *new_weight_end;
1829 ceph_decode_32_safe(p, end, len, e_inval);
1830 for (i = 0; i < len; ++i) {
1831 struct ceph_entity_addr addr;
1833 ceph_decode_skip_32(p, end, e_inval);
1834 if (ceph_decode_entity_addr(p, end, &addr))
1839 ceph_decode_32_safe(p, end, len, e_inval);
1840 len *= sizeof(u32) + (struct_v >= 5 ? sizeof(u32) : sizeof(u8));
1841 ceph_decode_need(p, end, len, e_inval);
1845 ceph_decode_32_safe(p, end, len, e_inval);
1850 ceph_decode_need(p, end, 2*sizeof(u32), e_inval);
1851 osd = ceph_decode_32(p);
1852 w = ceph_decode_32(p);
1853 BUG_ON(osd >= map->max_osd);
1854 pr_info("osd%d weight 0x%x %s\n", osd, w,
1855 w == CEPH_OSD_IN ? "(in)" :
1856 (w == CEPH_OSD_OUT ? "(out)" : ""));
1857 map->osd_weight[osd] = w;
1860 * If we are marking in, set the EXISTS, and clear the
1861 * AUTOOUT and NEW bits.
1864 map->osd_state[osd] |= CEPH_OSD_EXISTS;
1865 map->osd_state[osd] &= ~(CEPH_OSD_AUTOOUT |
1869 new_weight_end = *p;
1871 /* new_state (up/down) */
1873 len = ceph_decode_32(p);
1879 osd = ceph_decode_32(p);
1881 xorstate = ceph_decode_32(p);
1883 xorstate = ceph_decode_8(p);
1885 xorstate = CEPH_OSD_UP;
1886 BUG_ON(osd >= map->max_osd);
1887 if ((map->osd_state[osd] & CEPH_OSD_UP) &&
1888 (xorstate & CEPH_OSD_UP))
1889 pr_info("osd%d down\n", osd);
1890 if ((map->osd_state[osd] & CEPH_OSD_EXISTS) &&
1891 (xorstate & CEPH_OSD_EXISTS)) {
1892 pr_info("osd%d does not exist\n", osd);
1893 ret = set_primary_affinity(map, osd,
1894 CEPH_OSD_DEFAULT_PRIMARY_AFFINITY);
1897 memset(map->osd_addr + osd, 0, sizeof(*map->osd_addr));
1898 map->osd_state[osd] = 0;
1900 map->osd_state[osd] ^= xorstate;
1906 len = ceph_decode_32(p);
1909 struct ceph_entity_addr addr;
1911 osd = ceph_decode_32(p);
1912 BUG_ON(osd >= map->max_osd);
1913 if (ceph_decode_entity_addr(p, end, &addr))
1915 pr_info("osd%d up\n", osd);
1916 map->osd_state[osd] |= CEPH_OSD_EXISTS | CEPH_OSD_UP;
1917 map->osd_addr[osd] = addr;
1920 *p = new_weight_end;
1928 * decode and apply an incremental map update.
1930 struct ceph_osdmap *osdmap_apply_incremental(void **p, void *end,
1931 struct ceph_osdmap *map)
1933 struct ceph_fsid fsid;
1935 struct ceph_timespec modified;
1939 __s32 new_flags, max;
1944 dout("%s %p to %p len %d\n", __func__, *p, end, (int)(end - *p));
1946 err = get_osdmap_client_data_v(p, end, "inc", &struct_v);
1950 /* fsid, epoch, modified, new_pool_max, new_flags */
1951 ceph_decode_need(p, end, sizeof(fsid) + sizeof(u32) + sizeof(modified) +
1952 sizeof(u64) + sizeof(u32), e_inval);
1953 ceph_decode_copy(p, &fsid, sizeof(fsid));
1954 epoch = ceph_decode_32(p);
1955 BUG_ON(epoch != map->epoch+1);
1956 ceph_decode_copy(p, &modified, sizeof(modified));
1957 new_pool_max = ceph_decode_64(p);
1958 new_flags = ceph_decode_32(p);
1961 ceph_decode_32_safe(p, end, len, e_inval);
1963 dout("apply_incremental full map len %d, %p to %p\n",
1965 return ceph_osdmap_decode(p, min(*p+len, end));
1969 ceph_decode_32_safe(p, end, len, e_inval);
1971 err = osdmap_set_crush(map,
1972 crush_decode(*p, min(*p + len, end)));
1980 map->flags = new_flags;
1981 if (new_pool_max >= 0)
1982 map->pool_max = new_pool_max;
1985 ceph_decode_32_safe(p, end, max, e_inval);
1987 err = osdmap_set_max_osd(map, max);
1993 map->modified = modified;
1996 err = decode_new_pools(p, end, map);
2000 /* new_pool_names */
2001 err = decode_pool_names(p, end, map);
2006 ceph_decode_32_safe(p, end, len, e_inval);
2008 struct ceph_pg_pool_info *pi;
2010 ceph_decode_64_safe(p, end, pool, e_inval);
2011 pi = lookup_pg_pool(&map->pg_pools, pool);
2013 __remove_pg_pool(&map->pg_pools, pi);
2016 /* new_up_client, new_state, new_weight */
2017 err = decode_new_up_state_weight(p, end, struct_v, map);
2022 err = decode_new_pg_temp(p, end, map);
2026 /* new_primary_temp */
2027 if (struct_v >= 1) {
2028 err = decode_new_primary_temp(p, end, map);
2033 /* new_primary_affinity */
2034 if (struct_v >= 2) {
2035 err = decode_new_primary_affinity(p, end, map);
2040 if (struct_v >= 3) {
2041 /* new_erasure_code_profiles */
2042 ceph_decode_skip_map_of_map(p, end, string, string, string,
2044 /* old_erasure_code_profiles */
2045 ceph_decode_skip_set(p, end, string, e_inval);
2048 if (struct_v >= 4) {
2049 err = decode_new_pg_upmap(p, end, map);
2053 err = decode_old_pg_upmap(p, end, map);
2057 err = decode_new_pg_upmap_items(p, end, map);
2061 err = decode_old_pg_upmap_items(p, end, map);
2066 /* ignore the rest */
2069 dout("inc osdmap epoch %d max_osd %d\n", map->epoch, map->max_osd);
2075 pr_err("corrupt inc osdmap (%d) epoch %d off %d (%p of %p-%p)\n",
2076 err, epoch, (int)(*p - start), *p, start, end);
2077 print_hex_dump(KERN_DEBUG, "osdmap: ",
2078 DUMP_PREFIX_OFFSET, 16, 1,
2079 start, end - start, true);
2080 return ERR_PTR(err);
2083 void ceph_oloc_copy(struct ceph_object_locator *dest,
2084 const struct ceph_object_locator *src)
2086 ceph_oloc_destroy(dest);
2088 dest->pool = src->pool;
2090 dest->pool_ns = ceph_get_string(src->pool_ns);
2092 dest->pool_ns = NULL;
2094 EXPORT_SYMBOL(ceph_oloc_copy);
2096 void ceph_oloc_destroy(struct ceph_object_locator *oloc)
2098 ceph_put_string(oloc->pool_ns);
2100 EXPORT_SYMBOL(ceph_oloc_destroy);
2102 void ceph_oid_copy(struct ceph_object_id *dest,
2103 const struct ceph_object_id *src)
2105 ceph_oid_destroy(dest);
2107 if (src->name != src->inline_name) {
2108 /* very rare, see ceph_object_id definition */
2109 dest->name = kmalloc(src->name_len + 1,
2110 GFP_NOIO | __GFP_NOFAIL);
2112 dest->name = dest->inline_name;
2114 memcpy(dest->name, src->name, src->name_len + 1);
2115 dest->name_len = src->name_len;
2117 EXPORT_SYMBOL(ceph_oid_copy);
2119 static __printf(2, 0)
2120 int oid_printf_vargs(struct ceph_object_id *oid, const char *fmt, va_list ap)
2124 WARN_ON(!ceph_oid_empty(oid));
2126 len = vsnprintf(oid->inline_name, sizeof(oid->inline_name), fmt, ap);
2127 if (len >= sizeof(oid->inline_name))
2130 oid->name_len = len;
2135 * If oid doesn't fit into inline buffer, BUG.
2137 void ceph_oid_printf(struct ceph_object_id *oid, const char *fmt, ...)
2142 BUG_ON(oid_printf_vargs(oid, fmt, ap));
2145 EXPORT_SYMBOL(ceph_oid_printf);
2147 static __printf(3, 0)
2148 int oid_aprintf_vargs(struct ceph_object_id *oid, gfp_t gfp,
2149 const char *fmt, va_list ap)
2155 len = oid_printf_vargs(oid, fmt, aq);
2159 char *external_name;
2161 external_name = kmalloc(len + 1, gfp);
2165 oid->name = external_name;
2166 WARN_ON(vsnprintf(oid->name, len + 1, fmt, ap) != len);
2167 oid->name_len = len;
2174 * If oid doesn't fit into inline buffer, allocate.
2176 int ceph_oid_aprintf(struct ceph_object_id *oid, gfp_t gfp,
2177 const char *fmt, ...)
2183 ret = oid_aprintf_vargs(oid, gfp, fmt, ap);
2188 EXPORT_SYMBOL(ceph_oid_aprintf);
2190 void ceph_oid_destroy(struct ceph_object_id *oid)
2192 if (oid->name != oid->inline_name)
2195 EXPORT_SYMBOL(ceph_oid_destroy);
2200 static bool __osds_equal(const struct ceph_osds *lhs,
2201 const struct ceph_osds *rhs)
2203 if (lhs->size == rhs->size &&
2204 !memcmp(lhs->osds, rhs->osds, rhs->size * sizeof(rhs->osds[0])))
2213 static bool osds_equal(const struct ceph_osds *lhs,
2214 const struct ceph_osds *rhs)
2216 if (__osds_equal(lhs, rhs) &&
2217 lhs->primary == rhs->primary)
2223 static bool osds_valid(const struct ceph_osds *set)
2226 if (set->size > 0 && set->primary >= 0)
2229 /* empty can_shift_osds set */
2230 if (!set->size && set->primary == -1)
2233 /* empty !can_shift_osds set - all NONE */
2234 if (set->size > 0 && set->primary == -1) {
2237 for (i = 0; i < set->size; i++) {
2238 if (set->osds[i] != CRUSH_ITEM_NONE)
2248 void ceph_osds_copy(struct ceph_osds *dest, const struct ceph_osds *src)
2250 memcpy(dest->osds, src->osds, src->size * sizeof(src->osds[0]));
2251 dest->size = src->size;
2252 dest->primary = src->primary;
2255 bool ceph_pg_is_split(const struct ceph_pg *pgid, u32 old_pg_num,
2258 int old_bits = calc_bits_of(old_pg_num);
2259 int old_mask = (1 << old_bits) - 1;
2262 WARN_ON(pgid->seed >= old_pg_num);
2263 if (new_pg_num <= old_pg_num)
2266 for (n = 1; ; n++) {
2267 int next_bit = n << (old_bits - 1);
2268 u32 s = next_bit | pgid->seed;
2270 if (s < old_pg_num || s == pgid->seed)
2272 if (s >= new_pg_num)
2275 s = ceph_stable_mod(s, old_pg_num, old_mask);
2276 if (s == pgid->seed)
2283 bool ceph_is_new_interval(const struct ceph_osds *old_acting,
2284 const struct ceph_osds *new_acting,
2285 const struct ceph_osds *old_up,
2286 const struct ceph_osds *new_up,
2293 bool old_sort_bitwise,
2294 bool new_sort_bitwise,
2295 bool old_recovery_deletes,
2296 bool new_recovery_deletes,
2297 const struct ceph_pg *pgid)
2299 return !osds_equal(old_acting, new_acting) ||
2300 !osds_equal(old_up, new_up) ||
2301 old_size != new_size ||
2302 old_min_size != new_min_size ||
2303 ceph_pg_is_split(pgid, old_pg_num, new_pg_num) ||
2304 old_sort_bitwise != new_sort_bitwise ||
2305 old_recovery_deletes != new_recovery_deletes;
2308 static int calc_pg_rank(int osd, const struct ceph_osds *acting)
2312 for (i = 0; i < acting->size; i++) {
2313 if (acting->osds[i] == osd)
2320 static bool primary_changed(const struct ceph_osds *old_acting,
2321 const struct ceph_osds *new_acting)
2323 if (!old_acting->size && !new_acting->size)
2324 return false; /* both still empty */
2326 if (!old_acting->size ^ !new_acting->size)
2327 return true; /* was empty, now not, or vice versa */
2329 if (old_acting->primary != new_acting->primary)
2330 return true; /* primary changed */
2332 if (calc_pg_rank(old_acting->primary, old_acting) !=
2333 calc_pg_rank(new_acting->primary, new_acting))
2336 return false; /* same primary (tho replicas may have changed) */
2339 bool ceph_osds_changed(const struct ceph_osds *old_acting,
2340 const struct ceph_osds *new_acting,
2343 if (primary_changed(old_acting, new_acting))
2346 if (any_change && !__osds_equal(old_acting, new_acting))
2353 * Map an object into a PG.
2355 * Should only be called with target_oid and target_oloc (as opposed to
2356 * base_oid and base_oloc), since tiering isn't taken into account.
2358 void __ceph_object_locator_to_pg(struct ceph_pg_pool_info *pi,
2359 const struct ceph_object_id *oid,
2360 const struct ceph_object_locator *oloc,
2361 struct ceph_pg *raw_pgid)
2363 WARN_ON(pi->id != oloc->pool);
2365 if (!oloc->pool_ns) {
2366 raw_pgid->pool = oloc->pool;
2367 raw_pgid->seed = ceph_str_hash(pi->object_hash, oid->name,
2369 dout("%s %s -> raw_pgid %llu.%x\n", __func__, oid->name,
2370 raw_pgid->pool, raw_pgid->seed);
2372 char stack_buf[256];
2373 char *buf = stack_buf;
2374 int nsl = oloc->pool_ns->len;
2375 size_t total = nsl + 1 + oid->name_len;
2377 if (total > sizeof(stack_buf))
2378 buf = kmalloc(total, GFP_NOIO | __GFP_NOFAIL);
2379 memcpy(buf, oloc->pool_ns->str, nsl);
2381 memcpy(buf + nsl + 1, oid->name, oid->name_len);
2382 raw_pgid->pool = oloc->pool;
2383 raw_pgid->seed = ceph_str_hash(pi->object_hash, buf, total);
2384 if (buf != stack_buf)
2386 dout("%s %s ns %.*s -> raw_pgid %llu.%x\n", __func__,
2387 oid->name, nsl, oloc->pool_ns->str,
2388 raw_pgid->pool, raw_pgid->seed);
2392 int ceph_object_locator_to_pg(struct ceph_osdmap *osdmap,
2393 const struct ceph_object_id *oid,
2394 const struct ceph_object_locator *oloc,
2395 struct ceph_pg *raw_pgid)
2397 struct ceph_pg_pool_info *pi;
2399 pi = ceph_pg_pool_by_id(osdmap, oloc->pool);
2403 __ceph_object_locator_to_pg(pi, oid, oloc, raw_pgid);
2406 EXPORT_SYMBOL(ceph_object_locator_to_pg);
2409 * Map a raw PG (full precision ps) into an actual PG.
2411 static void raw_pg_to_pg(struct ceph_pg_pool_info *pi,
2412 const struct ceph_pg *raw_pgid,
2413 struct ceph_pg *pgid)
2415 pgid->pool = raw_pgid->pool;
2416 pgid->seed = ceph_stable_mod(raw_pgid->seed, pi->pg_num,
2421 * Map a raw PG (full precision ps) into a placement ps (placement
2422 * seed). Include pool id in that value so that different pools don't
2423 * use the same seeds.
2425 static u32 raw_pg_to_pps(struct ceph_pg_pool_info *pi,
2426 const struct ceph_pg *raw_pgid)
2428 if (pi->flags & CEPH_POOL_FLAG_HASHPSPOOL) {
2429 /* hash pool id and seed so that pool PGs do not overlap */
2430 return crush_hash32_2(CRUSH_HASH_RJENKINS1,
2431 ceph_stable_mod(raw_pgid->seed,
2437 * legacy behavior: add ps and pool together. this is
2438 * not a great approach because the PGs from each pool
2439 * will overlap on top of each other: 0.5 == 1.4 ==
2442 return ceph_stable_mod(raw_pgid->seed, pi->pgp_num,
2444 (unsigned)raw_pgid->pool;
2449 * Magic value used for a "default" fallback choose_args, used if the
2450 * crush_choose_arg_map passed to do_crush() does not exist. If this
2451 * also doesn't exist, fall back to canonical weights.
2453 #define CEPH_DEFAULT_CHOOSE_ARGS -1
2455 static int do_crush(struct ceph_osdmap *map, int ruleno, int x,
2456 int *result, int result_max,
2457 const __u32 *weight, int weight_max,
2458 s64 choose_args_index)
2460 struct crush_choose_arg_map *arg_map;
2461 struct crush_work *work;
2464 BUG_ON(result_max > CEPH_PG_MAX_SIZE);
2466 arg_map = lookup_choose_arg_map(&map->crush->choose_args,
2469 arg_map = lookup_choose_arg_map(&map->crush->choose_args,
2470 CEPH_DEFAULT_CHOOSE_ARGS);
2472 work = get_workspace(&map->crush_wsm, map->crush);
2473 r = crush_do_rule(map->crush, ruleno, x, result, result_max,
2474 weight, weight_max, work,
2475 arg_map ? arg_map->args : NULL);
2476 put_workspace(&map->crush_wsm, work);
2480 static void remove_nonexistent_osds(struct ceph_osdmap *osdmap,
2481 struct ceph_pg_pool_info *pi,
2482 struct ceph_osds *set)
2486 if (ceph_can_shift_osds(pi)) {
2490 for (i = 0; i < set->size; i++) {
2491 if (!ceph_osd_exists(osdmap, set->osds[i])) {
2496 set->osds[i - removed] = set->osds[i];
2498 set->size -= removed;
2500 /* set dne devices to NONE */
2501 for (i = 0; i < set->size; i++) {
2502 if (!ceph_osd_exists(osdmap, set->osds[i]))
2503 set->osds[i] = CRUSH_ITEM_NONE;
2509 * Calculate raw set (CRUSH output) for given PG and filter out
2510 * nonexistent OSDs. ->primary is undefined for a raw set.
2512 * Placement seed (CRUSH input) is returned through @ppps.
2514 static void pg_to_raw_osds(struct ceph_osdmap *osdmap,
2515 struct ceph_pg_pool_info *pi,
2516 const struct ceph_pg *raw_pgid,
2517 struct ceph_osds *raw,
2520 u32 pps = raw_pg_to_pps(pi, raw_pgid);
2524 ceph_osds_init(raw);
2528 ruleno = crush_find_rule(osdmap->crush, pi->crush_ruleset, pi->type,
2531 pr_err("no crush rule: pool %lld ruleset %d type %d size %d\n",
2532 pi->id, pi->crush_ruleset, pi->type, pi->size);
2536 if (pi->size > ARRAY_SIZE(raw->osds)) {
2537 pr_err_ratelimited("pool %lld ruleset %d type %d too wide: size %d > %zu\n",
2538 pi->id, pi->crush_ruleset, pi->type, pi->size,
2539 ARRAY_SIZE(raw->osds));
2543 len = do_crush(osdmap, ruleno, pps, raw->osds, pi->size,
2544 osdmap->osd_weight, osdmap->max_osd, pi->id);
2546 pr_err("error %d from crush rule %d: pool %lld ruleset %d type %d size %d\n",
2547 len, ruleno, pi->id, pi->crush_ruleset, pi->type,
2553 remove_nonexistent_osds(osdmap, pi, raw);
2556 /* apply pg_upmap[_items] mappings */
2557 static void apply_upmap(struct ceph_osdmap *osdmap,
2558 const struct ceph_pg *pgid,
2559 struct ceph_osds *raw)
2561 struct ceph_pg_mapping *pg;
2564 pg = lookup_pg_mapping(&osdmap->pg_upmap, pgid);
2566 /* make sure targets aren't marked out */
2567 for (i = 0; i < pg->pg_upmap.len; i++) {
2568 int osd = pg->pg_upmap.osds[i];
2570 if (osd != CRUSH_ITEM_NONE &&
2571 osd < osdmap->max_osd &&
2572 osdmap->osd_weight[osd] == 0) {
2573 /* reject/ignore explicit mapping */
2577 for (i = 0; i < pg->pg_upmap.len; i++)
2578 raw->osds[i] = pg->pg_upmap.osds[i];
2579 raw->size = pg->pg_upmap.len;
2580 /* check and apply pg_upmap_items, if any */
2583 pg = lookup_pg_mapping(&osdmap->pg_upmap_items, pgid);
2586 * Note: this approach does not allow a bidirectional swap,
2587 * e.g., [[1,2],[2,1]] applied to [0,1,2] -> [0,2,1].
2589 for (i = 0; i < pg->pg_upmap_items.len; i++) {
2590 int from = pg->pg_upmap_items.from_to[i][0];
2591 int to = pg->pg_upmap_items.from_to[i][1];
2593 bool exists = false;
2595 /* make sure replacement doesn't already appear */
2596 for (j = 0; j < raw->size; j++) {
2597 int osd = raw->osds[j];
2603 /* ignore mapping if target is marked out */
2604 if (osd == from && pos < 0 &&
2605 !(to != CRUSH_ITEM_NONE &&
2606 to < osdmap->max_osd &&
2607 osdmap->osd_weight[to] == 0)) {
2611 if (!exists && pos >= 0)
2612 raw->osds[pos] = to;
2618 * Given raw set, calculate up set and up primary. By definition of an
2619 * up set, the result won't contain nonexistent or down OSDs.
2621 * This is done in-place - on return @set is the up set. If it's
2622 * empty, ->primary will remain undefined.
2624 static void raw_to_up_osds(struct ceph_osdmap *osdmap,
2625 struct ceph_pg_pool_info *pi,
2626 struct ceph_osds *set)
2630 /* ->primary is undefined for a raw set */
2631 BUG_ON(set->primary != -1);
2633 if (ceph_can_shift_osds(pi)) {
2637 for (i = 0; i < set->size; i++) {
2638 if (ceph_osd_is_down(osdmap, set->osds[i])) {
2643 set->osds[i - removed] = set->osds[i];
2645 set->size -= removed;
2647 set->primary = set->osds[0];
2649 /* set down/dne devices to NONE */
2650 for (i = set->size - 1; i >= 0; i--) {
2651 if (ceph_osd_is_down(osdmap, set->osds[i]))
2652 set->osds[i] = CRUSH_ITEM_NONE;
2654 set->primary = set->osds[i];
2659 static void apply_primary_affinity(struct ceph_osdmap *osdmap,
2660 struct ceph_pg_pool_info *pi,
2662 struct ceph_osds *up)
2668 * Do we have any non-default primary_affinity values for these
2671 if (!osdmap->osd_primary_affinity)
2674 for (i = 0; i < up->size; i++) {
2675 int osd = up->osds[i];
2677 if (osd != CRUSH_ITEM_NONE &&
2678 osdmap->osd_primary_affinity[osd] !=
2679 CEPH_OSD_DEFAULT_PRIMARY_AFFINITY) {
2687 * Pick the primary. Feed both the seed (for the pg) and the
2688 * osd into the hash/rng so that a proportional fraction of an
2689 * osd's pgs get rejected as primary.
2691 for (i = 0; i < up->size; i++) {
2692 int osd = up->osds[i];
2695 if (osd == CRUSH_ITEM_NONE)
2698 aff = osdmap->osd_primary_affinity[osd];
2699 if (aff < CEPH_OSD_MAX_PRIMARY_AFFINITY &&
2700 (crush_hash32_2(CRUSH_HASH_RJENKINS1,
2701 pps, osd) >> 16) >= aff) {
2703 * We chose not to use this primary. Note it
2704 * anyway as a fallback in case we don't pick
2705 * anyone else, but keep looking.
2717 up->primary = up->osds[pos];
2719 if (ceph_can_shift_osds(pi) && pos > 0) {
2720 /* move the new primary to the front */
2721 for (i = pos; i > 0; i--)
2722 up->osds[i] = up->osds[i - 1];
2723 up->osds[0] = up->primary;
2728 * Get pg_temp and primary_temp mappings for given PG.
2730 * Note that a PG may have none, only pg_temp, only primary_temp or
2731 * both pg_temp and primary_temp mappings. This means @temp isn't
2732 * always a valid OSD set on return: in the "only primary_temp" case,
2733 * @temp will have its ->primary >= 0 but ->size == 0.
2735 static void get_temp_osds(struct ceph_osdmap *osdmap,
2736 struct ceph_pg_pool_info *pi,
2737 const struct ceph_pg *pgid,
2738 struct ceph_osds *temp)
2740 struct ceph_pg_mapping *pg;
2743 ceph_osds_init(temp);
2746 pg = lookup_pg_mapping(&osdmap->pg_temp, pgid);
2748 for (i = 0; i < pg->pg_temp.len; i++) {
2749 if (ceph_osd_is_down(osdmap, pg->pg_temp.osds[i])) {
2750 if (ceph_can_shift_osds(pi))
2753 temp->osds[temp->size++] = CRUSH_ITEM_NONE;
2755 temp->osds[temp->size++] = pg->pg_temp.osds[i];
2759 /* apply pg_temp's primary */
2760 for (i = 0; i < temp->size; i++) {
2761 if (temp->osds[i] != CRUSH_ITEM_NONE) {
2762 temp->primary = temp->osds[i];
2769 pg = lookup_pg_mapping(&osdmap->primary_temp, pgid);
2771 temp->primary = pg->primary_temp.osd;
2775 * Map a PG to its acting set as well as its up set.
2777 * Acting set is used for data mapping purposes, while up set can be
2778 * recorded for detecting interval changes and deciding whether to
2781 void ceph_pg_to_up_acting_osds(struct ceph_osdmap *osdmap,
2782 struct ceph_pg_pool_info *pi,
2783 const struct ceph_pg *raw_pgid,
2784 struct ceph_osds *up,
2785 struct ceph_osds *acting)
2787 struct ceph_pg pgid;
2790 WARN_ON(pi->id != raw_pgid->pool);
2791 raw_pg_to_pg(pi, raw_pgid, &pgid);
2793 pg_to_raw_osds(osdmap, pi, raw_pgid, up, &pps);
2794 apply_upmap(osdmap, &pgid, up);
2795 raw_to_up_osds(osdmap, pi, up);
2796 apply_primary_affinity(osdmap, pi, pps, up);
2797 get_temp_osds(osdmap, pi, &pgid, acting);
2798 if (!acting->size) {
2799 memcpy(acting->osds, up->osds, up->size * sizeof(up->osds[0]));
2800 acting->size = up->size;
2801 if (acting->primary == -1)
2802 acting->primary = up->primary;
2804 WARN_ON(!osds_valid(up) || !osds_valid(acting));
2807 bool ceph_pg_to_primary_shard(struct ceph_osdmap *osdmap,
2808 struct ceph_pg_pool_info *pi,
2809 const struct ceph_pg *raw_pgid,
2810 struct ceph_spg *spgid)
2812 struct ceph_pg pgid;
2813 struct ceph_osds up, acting;
2816 WARN_ON(pi->id != raw_pgid->pool);
2817 raw_pg_to_pg(pi, raw_pgid, &pgid);
2819 if (ceph_can_shift_osds(pi)) {
2820 spgid->pgid = pgid; /* struct */
2821 spgid->shard = CEPH_SPG_NOSHARD;
2825 ceph_pg_to_up_acting_osds(osdmap, pi, &pgid, &up, &acting);
2826 for (i = 0; i < acting.size; i++) {
2827 if (acting.osds[i] == acting.primary) {
2828 spgid->pgid = pgid; /* struct */
2838 * Return acting primary for given PG, or -1 if none.
2840 int ceph_pg_to_acting_primary(struct ceph_osdmap *osdmap,
2841 const struct ceph_pg *raw_pgid)
2843 struct ceph_pg_pool_info *pi;
2844 struct ceph_osds up, acting;
2846 pi = ceph_pg_pool_by_id(osdmap, raw_pgid->pool);
2850 ceph_pg_to_up_acting_osds(osdmap, pi, raw_pgid, &up, &acting);
2851 return acting.primary;
2853 EXPORT_SYMBOL(ceph_pg_to_acting_primary);
2855 static struct crush_loc_node *alloc_crush_loc(size_t type_name_len,
2858 struct crush_loc_node *loc;
2860 loc = kmalloc(sizeof(*loc) + type_name_len + name_len + 2, GFP_NOIO);
2864 RB_CLEAR_NODE(&loc->cl_node);
2868 static void free_crush_loc(struct crush_loc_node *loc)
2870 WARN_ON(!RB_EMPTY_NODE(&loc->cl_node));
2875 static int crush_loc_compare(const struct crush_loc *loc1,
2876 const struct crush_loc *loc2)
2878 return strcmp(loc1->cl_type_name, loc2->cl_type_name) ?:
2879 strcmp(loc1->cl_name, loc2->cl_name);
2882 DEFINE_RB_FUNCS2(crush_loc, struct crush_loc_node, cl_loc, crush_loc_compare,
2883 RB_BYPTR, const struct crush_loc *, cl_node)
2886 * Parses a set of <bucket type name>':'<bucket name> pairs separated
2887 * by '|', e.g. "rack:foo1|rack:foo2|datacenter:bar".
2889 * Note that @crush_location is modified by strsep().
2891 int ceph_parse_crush_location(char *crush_location, struct rb_root *locs)
2893 struct crush_loc_node *loc;
2894 const char *type_name, *name, *colon;
2895 size_t type_name_len, name_len;
2897 dout("%s '%s'\n", __func__, crush_location);
2898 while ((type_name = strsep(&crush_location, "|"))) {
2899 colon = strchr(type_name, ':');
2903 type_name_len = colon - type_name;
2904 if (type_name_len == 0)
2908 name_len = strlen(name);
2912 loc = alloc_crush_loc(type_name_len, name_len);
2916 loc->cl_loc.cl_type_name = loc->cl_data;
2917 memcpy(loc->cl_loc.cl_type_name, type_name, type_name_len);
2918 loc->cl_loc.cl_type_name[type_name_len] = '\0';
2920 loc->cl_loc.cl_name = loc->cl_data + type_name_len + 1;
2921 memcpy(loc->cl_loc.cl_name, name, name_len);
2922 loc->cl_loc.cl_name[name_len] = '\0';
2924 if (!__insert_crush_loc(locs, loc)) {
2925 free_crush_loc(loc);
2929 dout("%s type_name '%s' name '%s'\n", __func__,
2930 loc->cl_loc.cl_type_name, loc->cl_loc.cl_name);
2936 int ceph_compare_crush_locs(struct rb_root *locs1, struct rb_root *locs2)
2938 struct rb_node *n1 = rb_first(locs1);
2939 struct rb_node *n2 = rb_first(locs2);
2942 for ( ; n1 && n2; n1 = rb_next(n1), n2 = rb_next(n2)) {
2943 struct crush_loc_node *loc1 =
2944 rb_entry(n1, struct crush_loc_node, cl_node);
2945 struct crush_loc_node *loc2 =
2946 rb_entry(n2, struct crush_loc_node, cl_node);
2948 ret = crush_loc_compare(&loc1->cl_loc, &loc2->cl_loc);
2960 void ceph_clear_crush_locs(struct rb_root *locs)
2962 while (!RB_EMPTY_ROOT(locs)) {
2963 struct crush_loc_node *loc =
2964 rb_entry(rb_first(locs), struct crush_loc_node, cl_node);
2966 erase_crush_loc(locs, loc);
2967 free_crush_loc(loc);
2974 static bool is_valid_crush_name(const char *name)
2977 if (!('a' <= *name && *name <= 'z') &&
2978 !('A' <= *name && *name <= 'Z') &&
2979 !('0' <= *name && *name <= '9') &&
2980 *name != '-' && *name != '_' && *name != '.')
2982 } while (*++name != '\0');
2988 * Gets the parent of an item. Returns its id (<0 because the
2989 * parent is always a bucket), type id (>0 for the same reason,
2990 * via @parent_type_id) and location (via @parent_loc). If no
2991 * parent, returns 0.
2993 * Does a linear search, as there are no parent pointers of any
2994 * kind. Note that the result is ambigous for items that occur
2995 * multiple times in the map.
2997 static int get_immediate_parent(struct crush_map *c, int id,
2998 u16 *parent_type_id,
2999 struct crush_loc *parent_loc)
3001 struct crush_bucket *b;
3002 struct crush_name_node *type_cn, *cn;
3005 for (i = 0; i < c->max_buckets; i++) {
3010 /* ignore per-class shadow hierarchy */
3011 cn = lookup_crush_name(&c->names, b->id);
3012 if (!cn || !is_valid_crush_name(cn->cn_name))
3015 for (j = 0; j < b->size; j++) {
3016 if (b->items[j] != id)
3019 *parent_type_id = b->type;
3020 type_cn = lookup_crush_name(&c->type_names, b->type);
3021 parent_loc->cl_type_name = type_cn->cn_name;
3022 parent_loc->cl_name = cn->cn_name;
3027 return 0; /* no parent */
3031 * Calculates the locality/distance from an item to a client
3032 * location expressed in terms of CRUSH hierarchy as a set of
3033 * (bucket type name, bucket name) pairs. Specifically, looks
3034 * for the lowest-valued bucket type for which the location of
3035 * @id matches one of the locations in @locs, so for standard
3036 * bucket types (host = 1, rack = 3, datacenter = 8, zone = 9)
3037 * a matching host is closer than a matching rack and a matching
3038 * data center is closer than a matching zone.
3040 * Specifying multiple locations (a "multipath" location) such
3041 * as "rack=foo1 rack=foo2 datacenter=bar" is allowed -- @locs
3042 * is a multimap. The locality will be:
3044 * - 3 for OSDs in racks foo1 and foo2
3045 * - 8 for OSDs in data center bar
3046 * - -1 for all other OSDs
3048 * The lowest possible bucket type is 1, so the best locality
3049 * for an OSD is 1 (i.e. a matching host). Locality 0 would be
3052 int ceph_get_crush_locality(struct ceph_osdmap *osdmap, int id,
3053 struct rb_root *locs)
3055 struct crush_loc loc;
3059 * Instead of repeated get_immediate_parent() calls,
3060 * the location of @id could be obtained with a single
3061 * depth-first traversal.
3064 id = get_immediate_parent(osdmap->crush, id, &type_id, &loc);
3066 return -1; /* not local */
3068 if (lookup_crush_loc(locs, &loc))