2 #include <linux/ceph/ceph_debug.h>
4 #include <linux/module.h>
5 #include <linux/slab.h>
8 #include <linux/ceph/libceph.h>
9 #include <linux/ceph/osdmap.h>
10 #include <linux/ceph/decode.h>
11 #include <linux/crush/hash.h>
12 #include <linux/crush/mapper.h>
14 char *ceph_osdmap_state_str(char *str, int len, int state)
19 if ((state & CEPH_OSD_EXISTS) && (state & CEPH_OSD_UP))
20 snprintf(str, len, "exists, up");
21 else if (state & CEPH_OSD_EXISTS)
22 snprintf(str, len, "exists");
23 else if (state & CEPH_OSD_UP)
24 snprintf(str, len, "up");
26 snprintf(str, len, "doesn't exist");
33 static int calc_bits_of(unsigned int t)
44 * the foo_mask is the smallest value 2^n-1 that is >= foo.
46 static void calc_pg_masks(struct ceph_pg_pool_info *pi)
48 pi->pg_num_mask = (1 << calc_bits_of(pi->pg_num-1)) - 1;
49 pi->pgp_num_mask = (1 << calc_bits_of(pi->pgp_num-1)) - 1;
55 static int crush_decode_uniform_bucket(void **p, void *end,
56 struct crush_bucket_uniform *b)
58 dout("crush_decode_uniform_bucket %p to %p\n", *p, end);
59 ceph_decode_need(p, end, (1+b->h.size) * sizeof(u32), bad);
60 b->item_weight = ceph_decode_32(p);
66 static int crush_decode_list_bucket(void **p, void *end,
67 struct crush_bucket_list *b)
70 dout("crush_decode_list_bucket %p to %p\n", *p, end);
71 b->item_weights = kcalloc(b->h.size, sizeof(u32), GFP_NOFS);
72 if (b->item_weights == NULL)
74 b->sum_weights = kcalloc(b->h.size, sizeof(u32), GFP_NOFS);
75 if (b->sum_weights == NULL)
77 ceph_decode_need(p, end, 2 * b->h.size * sizeof(u32), bad);
78 for (j = 0; j < b->h.size; j++) {
79 b->item_weights[j] = ceph_decode_32(p);
80 b->sum_weights[j] = ceph_decode_32(p);
87 static int crush_decode_tree_bucket(void **p, void *end,
88 struct crush_bucket_tree *b)
91 dout("crush_decode_tree_bucket %p to %p\n", *p, end);
92 ceph_decode_8_safe(p, end, b->num_nodes, bad);
93 b->node_weights = kcalloc(b->num_nodes, sizeof(u32), GFP_NOFS);
94 if (b->node_weights == NULL)
96 ceph_decode_need(p, end, b->num_nodes * sizeof(u32), bad);
97 for (j = 0; j < b->num_nodes; j++)
98 b->node_weights[j] = ceph_decode_32(p);
104 static int crush_decode_straw_bucket(void **p, void *end,
105 struct crush_bucket_straw *b)
108 dout("crush_decode_straw_bucket %p to %p\n", *p, end);
109 b->item_weights = kcalloc(b->h.size, sizeof(u32), GFP_NOFS);
110 if (b->item_weights == NULL)
112 b->straws = kcalloc(b->h.size, sizeof(u32), GFP_NOFS);
113 if (b->straws == NULL)
115 ceph_decode_need(p, end, 2 * b->h.size * sizeof(u32), bad);
116 for (j = 0; j < b->h.size; j++) {
117 b->item_weights[j] = ceph_decode_32(p);
118 b->straws[j] = ceph_decode_32(p);
125 static int crush_decode_straw2_bucket(void **p, void *end,
126 struct crush_bucket_straw2 *b)
129 dout("crush_decode_straw2_bucket %p to %p\n", *p, end);
130 b->item_weights = kcalloc(b->h.size, sizeof(u32), GFP_NOFS);
131 if (b->item_weights == NULL)
133 ceph_decode_need(p, end, b->h.size * sizeof(u32), bad);
134 for (j = 0; j < b->h.size; j++)
135 b->item_weights[j] = ceph_decode_32(p);
141 static int skip_name_map(void **p, void *end)
144 ceph_decode_32_safe(p, end, len ,bad);
148 ceph_decode_32_safe(p, end, strlen, bad);
156 static struct crush_map *crush_decode(void *pbyval, void *end)
162 void *start = pbyval;
166 dout("crush_decode %p to %p len %d\n", *p, end, (int)(end - *p));
168 c = kzalloc(sizeof(*c), GFP_NOFS);
170 return ERR_PTR(-ENOMEM);
172 /* set tunables to default values */
173 c->choose_local_tries = 2;
174 c->choose_local_fallback_tries = 5;
175 c->choose_total_tries = 19;
176 c->chooseleaf_descend_once = 0;
178 ceph_decode_need(p, end, 4*sizeof(u32), bad);
179 magic = ceph_decode_32(p);
180 if (magic != CRUSH_MAGIC) {
181 pr_err("crush_decode magic %x != current %x\n",
182 (unsigned int)magic, (unsigned int)CRUSH_MAGIC);
185 c->max_buckets = ceph_decode_32(p);
186 c->max_rules = ceph_decode_32(p);
187 c->max_devices = ceph_decode_32(p);
189 c->buckets = kcalloc(c->max_buckets, sizeof(*c->buckets), GFP_NOFS);
190 if (c->buckets == NULL)
192 c->rules = kcalloc(c->max_rules, sizeof(*c->rules), GFP_NOFS);
193 if (c->rules == NULL)
197 for (i = 0; i < c->max_buckets; i++) {
200 struct crush_bucket *b;
202 ceph_decode_32_safe(p, end, alg, bad);
204 c->buckets[i] = NULL;
207 dout("crush_decode bucket %d off %x %p to %p\n",
208 i, (int)(*p-start), *p, end);
211 case CRUSH_BUCKET_UNIFORM:
212 size = sizeof(struct crush_bucket_uniform);
214 case CRUSH_BUCKET_LIST:
215 size = sizeof(struct crush_bucket_list);
217 case CRUSH_BUCKET_TREE:
218 size = sizeof(struct crush_bucket_tree);
220 case CRUSH_BUCKET_STRAW:
221 size = sizeof(struct crush_bucket_straw);
223 case CRUSH_BUCKET_STRAW2:
224 size = sizeof(struct crush_bucket_straw2);
231 b = c->buckets[i] = kzalloc(size, GFP_NOFS);
235 ceph_decode_need(p, end, 4*sizeof(u32), bad);
236 b->id = ceph_decode_32(p);
237 b->type = ceph_decode_16(p);
238 b->alg = ceph_decode_8(p);
239 b->hash = ceph_decode_8(p);
240 b->weight = ceph_decode_32(p);
241 b->size = ceph_decode_32(p);
243 dout("crush_decode bucket size %d off %x %p to %p\n",
244 b->size, (int)(*p-start), *p, end);
246 b->items = kcalloc(b->size, sizeof(__s32), GFP_NOFS);
247 if (b->items == NULL)
249 b->perm = kcalloc(b->size, sizeof(u32), GFP_NOFS);
254 ceph_decode_need(p, end, b->size*sizeof(u32), bad);
255 for (j = 0; j < b->size; j++)
256 b->items[j] = ceph_decode_32(p);
259 case CRUSH_BUCKET_UNIFORM:
260 err = crush_decode_uniform_bucket(p, end,
261 (struct crush_bucket_uniform *)b);
265 case CRUSH_BUCKET_LIST:
266 err = crush_decode_list_bucket(p, end,
267 (struct crush_bucket_list *)b);
271 case CRUSH_BUCKET_TREE:
272 err = crush_decode_tree_bucket(p, end,
273 (struct crush_bucket_tree *)b);
277 case CRUSH_BUCKET_STRAW:
278 err = crush_decode_straw_bucket(p, end,
279 (struct crush_bucket_straw *)b);
283 case CRUSH_BUCKET_STRAW2:
284 err = crush_decode_straw2_bucket(p, end,
285 (struct crush_bucket_straw2 *)b);
293 dout("rule vec is %p\n", c->rules);
294 for (i = 0; i < c->max_rules; i++) {
296 struct crush_rule *r;
298 ceph_decode_32_safe(p, end, yes, bad);
300 dout("crush_decode NO rule %d off %x %p to %p\n",
301 i, (int)(*p-start), *p, end);
306 dout("crush_decode rule %d off %x %p to %p\n",
307 i, (int)(*p-start), *p, end);
310 ceph_decode_32_safe(p, end, yes, bad);
311 #if BITS_PER_LONG == 32
313 if (yes > (ULONG_MAX - sizeof(*r))
314 / sizeof(struct crush_rule_step))
317 r = c->rules[i] = kmalloc(sizeof(*r) +
318 yes*sizeof(struct crush_rule_step),
322 dout(" rule %d is at %p\n", i, r);
324 ceph_decode_copy_safe(p, end, &r->mask, 4, bad); /* 4 u8's */
325 ceph_decode_need(p, end, r->len*3*sizeof(u32), bad);
326 for (j = 0; j < r->len; j++) {
327 r->steps[j].op = ceph_decode_32(p);
328 r->steps[j].arg1 = ceph_decode_32(p);
329 r->steps[j].arg2 = ceph_decode_32(p);
333 /* ignore trailing name maps. */
334 for (num_name_maps = 0; num_name_maps < 3; num_name_maps++) {
335 err = skip_name_map(p, end);
341 ceph_decode_need(p, end, 3*sizeof(u32), done);
342 c->choose_local_tries = ceph_decode_32(p);
343 c->choose_local_fallback_tries = ceph_decode_32(p);
344 c->choose_total_tries = ceph_decode_32(p);
345 dout("crush decode tunable choose_local_tries = %d\n",
346 c->choose_local_tries);
347 dout("crush decode tunable choose_local_fallback_tries = %d\n",
348 c->choose_local_fallback_tries);
349 dout("crush decode tunable choose_total_tries = %d\n",
350 c->choose_total_tries);
352 ceph_decode_need(p, end, sizeof(u32), done);
353 c->chooseleaf_descend_once = ceph_decode_32(p);
354 dout("crush decode tunable chooseleaf_descend_once = %d\n",
355 c->chooseleaf_descend_once);
357 ceph_decode_need(p, end, sizeof(u8), done);
358 c->chooseleaf_vary_r = ceph_decode_8(p);
359 dout("crush decode tunable chooseleaf_vary_r = %d\n",
360 c->chooseleaf_vary_r);
362 /* skip straw_calc_version, allowed_bucket_algs */
363 ceph_decode_need(p, end, sizeof(u8) + sizeof(u32), done);
364 *p += sizeof(u8) + sizeof(u32);
366 ceph_decode_need(p, end, sizeof(u8), done);
367 c->chooseleaf_stable = ceph_decode_8(p);
368 dout("crush decode tunable chooseleaf_stable = %d\n",
369 c->chooseleaf_stable);
372 dout("crush_decode success\n");
378 dout("crush_decode fail %d\n", err);
384 * rbtree of pg_mapping for handling pg_temp (explicit mapping of pgid
385 * to a set of osds) and primary_temp (explicit primary setting)
387 static int pgid_cmp(struct ceph_pg l, struct ceph_pg r)
400 static int __insert_pg_mapping(struct ceph_pg_mapping *new,
401 struct rb_root *root)
403 struct rb_node **p = &root->rb_node;
404 struct rb_node *parent = NULL;
405 struct ceph_pg_mapping *pg = NULL;
408 dout("__insert_pg_mapping %llx %p\n", *(u64 *)&new->pgid, new);
411 pg = rb_entry(parent, struct ceph_pg_mapping, node);
412 c = pgid_cmp(new->pgid, pg->pgid);
421 rb_link_node(&new->node, parent, p);
422 rb_insert_color(&new->node, root);
426 static struct ceph_pg_mapping *__lookup_pg_mapping(struct rb_root *root,
429 struct rb_node *n = root->rb_node;
430 struct ceph_pg_mapping *pg;
434 pg = rb_entry(n, struct ceph_pg_mapping, node);
435 c = pgid_cmp(pgid, pg->pgid);
441 dout("__lookup_pg_mapping %lld.%x got %p\n",
442 pgid.pool, pgid.seed, pg);
449 static int __remove_pg_mapping(struct rb_root *root, struct ceph_pg pgid)
451 struct ceph_pg_mapping *pg = __lookup_pg_mapping(root, pgid);
454 dout("__remove_pg_mapping %lld.%x %p\n", pgid.pool, pgid.seed,
456 rb_erase(&pg->node, root);
460 dout("__remove_pg_mapping %lld.%x dne\n", pgid.pool, pgid.seed);
465 * rbtree of pg pool info
467 static int __insert_pg_pool(struct rb_root *root, struct ceph_pg_pool_info *new)
469 struct rb_node **p = &root->rb_node;
470 struct rb_node *parent = NULL;
471 struct ceph_pg_pool_info *pi = NULL;
475 pi = rb_entry(parent, struct ceph_pg_pool_info, node);
476 if (new->id < pi->id)
478 else if (new->id > pi->id)
484 rb_link_node(&new->node, parent, p);
485 rb_insert_color(&new->node, root);
489 static struct ceph_pg_pool_info *__lookup_pg_pool(struct rb_root *root, u64 id)
491 struct ceph_pg_pool_info *pi;
492 struct rb_node *n = root->rb_node;
495 pi = rb_entry(n, struct ceph_pg_pool_info, node);
498 else if (id > pi->id)
506 struct ceph_pg_pool_info *ceph_pg_pool_by_id(struct ceph_osdmap *map, u64 id)
508 return __lookup_pg_pool(&map->pg_pools, id);
511 const char *ceph_pg_pool_name_by_id(struct ceph_osdmap *map, u64 id)
513 struct ceph_pg_pool_info *pi;
515 if (id == CEPH_NOPOOL)
518 if (WARN_ON_ONCE(id > (u64) INT_MAX))
521 pi = __lookup_pg_pool(&map->pg_pools, (int) id);
523 return pi ? pi->name : NULL;
525 EXPORT_SYMBOL(ceph_pg_pool_name_by_id);
527 int ceph_pg_poolid_by_name(struct ceph_osdmap *map, const char *name)
531 for (rbp = rb_first(&map->pg_pools); rbp; rbp = rb_next(rbp)) {
532 struct ceph_pg_pool_info *pi =
533 rb_entry(rbp, struct ceph_pg_pool_info, node);
534 if (pi->name && strcmp(pi->name, name) == 0)
539 EXPORT_SYMBOL(ceph_pg_poolid_by_name);
541 static void __remove_pg_pool(struct rb_root *root, struct ceph_pg_pool_info *pi)
543 rb_erase(&pi->node, root);
548 static int decode_pool(void **p, void *end, struct ceph_pg_pool_info *pi)
554 ceph_decode_need(p, end, 2 + 4, bad);
555 ev = ceph_decode_8(p); /* encoding version */
556 cv = ceph_decode_8(p); /* compat version */
558 pr_warn("got v %d < 5 cv %d of ceph_pg_pool\n", ev, cv);
562 pr_warn("got v %d cv %d > 9 of ceph_pg_pool\n", ev, cv);
565 len = ceph_decode_32(p);
566 ceph_decode_need(p, end, len, bad);
569 pi->type = ceph_decode_8(p);
570 pi->size = ceph_decode_8(p);
571 pi->crush_ruleset = ceph_decode_8(p);
572 pi->object_hash = ceph_decode_8(p);
574 pi->pg_num = ceph_decode_32(p);
575 pi->pgp_num = ceph_decode_32(p);
577 *p += 4 + 4; /* skip lpg* */
578 *p += 4; /* skip last_change */
579 *p += 8 + 4; /* skip snap_seq, snap_epoch */
582 num = ceph_decode_32(p);
584 *p += 8; /* snapid key */
585 *p += 1 + 1; /* versions */
586 len = ceph_decode_32(p);
590 /* skip removed_snaps */
591 num = ceph_decode_32(p);
594 *p += 8; /* skip auid */
595 pi->flags = ceph_decode_64(p);
596 *p += 4; /* skip crash_replay_interval */
599 *p += 1; /* skip min_size */
602 *p += 8 + 8; /* skip quota_max_* */
606 num = ceph_decode_32(p);
609 *p += 8; /* skip tier_of */
610 *p += 1; /* skip cache_mode */
612 pi->read_tier = ceph_decode_64(p);
613 pi->write_tier = ceph_decode_64(p);
619 /* ignore the rest */
629 static int decode_pool_names(void **p, void *end, struct ceph_osdmap *map)
631 struct ceph_pg_pool_info *pi;
635 ceph_decode_32_safe(p, end, num, bad);
636 dout(" %d pool names\n", num);
638 ceph_decode_64_safe(p, end, pool, bad);
639 ceph_decode_32_safe(p, end, len, bad);
640 dout(" pool %llu len %d\n", pool, len);
641 ceph_decode_need(p, end, len, bad);
642 pi = __lookup_pg_pool(&map->pg_pools, pool);
644 char *name = kstrndup(*p, len, GFP_NOFS);
650 dout(" name is %s\n", pi->name);
663 void ceph_osdmap_destroy(struct ceph_osdmap *map)
665 dout("osdmap_destroy %p\n", map);
667 crush_destroy(map->crush);
668 while (!RB_EMPTY_ROOT(&map->pg_temp)) {
669 struct ceph_pg_mapping *pg =
670 rb_entry(rb_first(&map->pg_temp),
671 struct ceph_pg_mapping, node);
672 rb_erase(&pg->node, &map->pg_temp);
675 while (!RB_EMPTY_ROOT(&map->primary_temp)) {
676 struct ceph_pg_mapping *pg =
677 rb_entry(rb_first(&map->primary_temp),
678 struct ceph_pg_mapping, node);
679 rb_erase(&pg->node, &map->primary_temp);
682 while (!RB_EMPTY_ROOT(&map->pg_pools)) {
683 struct ceph_pg_pool_info *pi =
684 rb_entry(rb_first(&map->pg_pools),
685 struct ceph_pg_pool_info, node);
686 __remove_pg_pool(&map->pg_pools, pi);
688 kfree(map->osd_state);
689 kfree(map->osd_weight);
690 kfree(map->osd_addr);
691 kfree(map->osd_primary_affinity);
696 * Adjust max_osd value, (re)allocate arrays.
698 * The new elements are properly initialized.
700 static int osdmap_set_max_osd(struct ceph_osdmap *map, int max)
704 struct ceph_entity_addr *addr;
707 state = krealloc(map->osd_state, max*sizeof(*state), GFP_NOFS);
710 map->osd_state = state;
712 weight = krealloc(map->osd_weight, max*sizeof(*weight), GFP_NOFS);
715 map->osd_weight = weight;
717 addr = krealloc(map->osd_addr, max*sizeof(*addr), GFP_NOFS);
720 map->osd_addr = addr;
722 for (i = map->max_osd; i < max; i++) {
723 map->osd_state[i] = 0;
724 map->osd_weight[i] = CEPH_OSD_OUT;
725 memset(map->osd_addr + i, 0, sizeof(*map->osd_addr));
728 if (map->osd_primary_affinity) {
731 affinity = krealloc(map->osd_primary_affinity,
732 max*sizeof(*affinity), GFP_NOFS);
735 map->osd_primary_affinity = affinity;
737 for (i = map->max_osd; i < max; i++)
738 map->osd_primary_affinity[i] =
739 CEPH_OSD_DEFAULT_PRIMARY_AFFINITY;
747 #define OSDMAP_WRAPPER_COMPAT_VER 7
748 #define OSDMAP_CLIENT_DATA_COMPAT_VER 1
751 * Return 0 or error. On success, *v is set to 0 for old (v6) osdmaps,
752 * to struct_v of the client_data section for new (v7 and above)
755 static int get_osdmap_client_data_v(void **p, void *end,
756 const char *prefix, u8 *v)
760 ceph_decode_8_safe(p, end, struct_v, e_inval);
764 ceph_decode_8_safe(p, end, struct_compat, e_inval);
765 if (struct_compat > OSDMAP_WRAPPER_COMPAT_VER) {
766 pr_warn("got v %d cv %d > %d of %s ceph_osdmap\n",
767 struct_v, struct_compat,
768 OSDMAP_WRAPPER_COMPAT_VER, prefix);
771 *p += 4; /* ignore wrapper struct_len */
773 ceph_decode_8_safe(p, end, struct_v, e_inval);
774 ceph_decode_8_safe(p, end, struct_compat, e_inval);
775 if (struct_compat > OSDMAP_CLIENT_DATA_COMPAT_VER) {
776 pr_warn("got v %d cv %d > %d of %s ceph_osdmap client data\n",
777 struct_v, struct_compat,
778 OSDMAP_CLIENT_DATA_COMPAT_VER, prefix);
781 *p += 4; /* ignore client data struct_len */
786 ceph_decode_16_safe(p, end, version, e_inval);
788 pr_warn("got v %d < 6 of %s ceph_osdmap\n",
793 /* old osdmap enconding */
804 static int __decode_pools(void **p, void *end, struct ceph_osdmap *map,
809 ceph_decode_32_safe(p, end, n, e_inval);
811 struct ceph_pg_pool_info *pi;
815 ceph_decode_64_safe(p, end, pool, e_inval);
817 pi = __lookup_pg_pool(&map->pg_pools, pool);
818 if (!incremental || !pi) {
819 pi = kzalloc(sizeof(*pi), GFP_NOFS);
825 ret = __insert_pg_pool(&map->pg_pools, pi);
832 ret = decode_pool(p, end, pi);
843 static int decode_pools(void **p, void *end, struct ceph_osdmap *map)
845 return __decode_pools(p, end, map, false);
848 static int decode_new_pools(void **p, void *end, struct ceph_osdmap *map)
850 return __decode_pools(p, end, map, true);
853 static int __decode_pg_temp(void **p, void *end, struct ceph_osdmap *map,
858 ceph_decode_32_safe(p, end, n, e_inval);
864 ret = ceph_decode_pgid(p, end, &pgid);
868 ceph_decode_32_safe(p, end, len, e_inval);
870 ret = __remove_pg_mapping(&map->pg_temp, pgid);
871 BUG_ON(!incremental && ret != -ENOENT);
873 if (!incremental || len > 0) {
874 struct ceph_pg_mapping *pg;
876 ceph_decode_need(p, end, len*sizeof(u32), e_inval);
878 if (len > (UINT_MAX - sizeof(*pg)) / sizeof(u32))
881 pg = kzalloc(sizeof(*pg) + len*sizeof(u32), GFP_NOFS);
886 pg->pg_temp.len = len;
887 for (i = 0; i < len; i++)
888 pg->pg_temp.osds[i] = ceph_decode_32(p);
890 ret = __insert_pg_mapping(pg, &map->pg_temp);
904 static int decode_pg_temp(void **p, void *end, struct ceph_osdmap *map)
906 return __decode_pg_temp(p, end, map, false);
909 static int decode_new_pg_temp(void **p, void *end, struct ceph_osdmap *map)
911 return __decode_pg_temp(p, end, map, true);
914 static int __decode_primary_temp(void **p, void *end, struct ceph_osdmap *map,
919 ceph_decode_32_safe(p, end, n, e_inval);
925 ret = ceph_decode_pgid(p, end, &pgid);
929 ceph_decode_32_safe(p, end, osd, e_inval);
931 ret = __remove_pg_mapping(&map->primary_temp, pgid);
932 BUG_ON(!incremental && ret != -ENOENT);
934 if (!incremental || osd != (u32)-1) {
935 struct ceph_pg_mapping *pg;
937 pg = kzalloc(sizeof(*pg), GFP_NOFS);
942 pg->primary_temp.osd = osd;
944 ret = __insert_pg_mapping(pg, &map->primary_temp);
958 static int decode_primary_temp(void **p, void *end, struct ceph_osdmap *map)
960 return __decode_primary_temp(p, end, map, false);
963 static int decode_new_primary_temp(void **p, void *end,
964 struct ceph_osdmap *map)
966 return __decode_primary_temp(p, end, map, true);
969 u32 ceph_get_primary_affinity(struct ceph_osdmap *map, int osd)
971 BUG_ON(osd >= map->max_osd);
973 if (!map->osd_primary_affinity)
974 return CEPH_OSD_DEFAULT_PRIMARY_AFFINITY;
976 return map->osd_primary_affinity[osd];
979 static int set_primary_affinity(struct ceph_osdmap *map, int osd, u32 aff)
981 BUG_ON(osd >= map->max_osd);
983 if (!map->osd_primary_affinity) {
986 map->osd_primary_affinity = kmalloc(map->max_osd*sizeof(u32),
988 if (!map->osd_primary_affinity)
991 for (i = 0; i < map->max_osd; i++)
992 map->osd_primary_affinity[i] =
993 CEPH_OSD_DEFAULT_PRIMARY_AFFINITY;
996 map->osd_primary_affinity[osd] = aff;
1001 static int decode_primary_affinity(void **p, void *end,
1002 struct ceph_osdmap *map)
1006 ceph_decode_32_safe(p, end, len, e_inval);
1008 kfree(map->osd_primary_affinity);
1009 map->osd_primary_affinity = NULL;
1012 if (len != map->max_osd)
1015 ceph_decode_need(p, end, map->max_osd*sizeof(u32), e_inval);
1017 for (i = 0; i < map->max_osd; i++) {
1020 ret = set_primary_affinity(map, i, ceph_decode_32(p));
1031 static int decode_new_primary_affinity(void **p, void *end,
1032 struct ceph_osdmap *map)
1036 ceph_decode_32_safe(p, end, n, e_inval);
1041 ceph_decode_32_safe(p, end, osd, e_inval);
1042 ceph_decode_32_safe(p, end, aff, e_inval);
1044 ret = set_primary_affinity(map, osd, aff);
1048 pr_info("osd%d primary-affinity 0x%x\n", osd, aff);
1058 * decode a full map.
1060 static int osdmap_decode(void **p, void *end, struct ceph_osdmap *map)
1069 dout("%s %p to %p len %d\n", __func__, *p, end, (int)(end - *p));
1071 err = get_osdmap_client_data_v(p, end, "full", &struct_v);
1075 /* fsid, epoch, created, modified */
1076 ceph_decode_need(p, end, sizeof(map->fsid) + sizeof(u32) +
1077 sizeof(map->created) + sizeof(map->modified), e_inval);
1078 ceph_decode_copy(p, &map->fsid, sizeof(map->fsid));
1079 epoch = map->epoch = ceph_decode_32(p);
1080 ceph_decode_copy(p, &map->created, sizeof(map->created));
1081 ceph_decode_copy(p, &map->modified, sizeof(map->modified));
1084 err = decode_pools(p, end, map);
1089 err = decode_pool_names(p, end, map);
1093 ceph_decode_32_safe(p, end, map->pool_max, e_inval);
1095 ceph_decode_32_safe(p, end, map->flags, e_inval);
1098 ceph_decode_32_safe(p, end, max, e_inval);
1100 /* (re)alloc osd arrays */
1101 err = osdmap_set_max_osd(map, max);
1105 /* osd_state, osd_weight, osd_addrs->client_addr */
1106 ceph_decode_need(p, end, 3*sizeof(u32) +
1107 map->max_osd*(1 + sizeof(*map->osd_weight) +
1108 sizeof(*map->osd_addr)), e_inval);
1110 if (ceph_decode_32(p) != map->max_osd)
1113 ceph_decode_copy(p, map->osd_state, map->max_osd);
1115 if (ceph_decode_32(p) != map->max_osd)
1118 for (i = 0; i < map->max_osd; i++)
1119 map->osd_weight[i] = ceph_decode_32(p);
1121 if (ceph_decode_32(p) != map->max_osd)
1124 ceph_decode_copy(p, map->osd_addr, map->max_osd*sizeof(*map->osd_addr));
1125 for (i = 0; i < map->max_osd; i++)
1126 ceph_decode_addr(&map->osd_addr[i]);
1129 err = decode_pg_temp(p, end, map);
1134 if (struct_v >= 1) {
1135 err = decode_primary_temp(p, end, map);
1140 /* primary_affinity */
1141 if (struct_v >= 2) {
1142 err = decode_primary_affinity(p, end, map);
1146 /* XXX can this happen? */
1147 kfree(map->osd_primary_affinity);
1148 map->osd_primary_affinity = NULL;
1152 ceph_decode_32_safe(p, end, len, e_inval);
1153 map->crush = crush_decode(*p, min(*p + len, end));
1154 if (IS_ERR(map->crush)) {
1155 err = PTR_ERR(map->crush);
1161 /* ignore the rest */
1164 dout("full osdmap epoch %d max_osd %d\n", map->epoch, map->max_osd);
1170 pr_err("corrupt full osdmap (%d) epoch %d off %d (%p of %p-%p)\n",
1171 err, epoch, (int)(*p - start), *p, start, end);
1172 print_hex_dump(KERN_DEBUG, "osdmap: ",
1173 DUMP_PREFIX_OFFSET, 16, 1,
1174 start, end - start, true);
1179 * Allocate and decode a full map.
1181 struct ceph_osdmap *ceph_osdmap_decode(void **p, void *end)
1183 struct ceph_osdmap *map;
1186 map = kzalloc(sizeof(*map), GFP_NOFS);
1188 return ERR_PTR(-ENOMEM);
1190 map->pg_temp = RB_ROOT;
1191 map->primary_temp = RB_ROOT;
1192 mutex_init(&map->crush_scratch_mutex);
1194 ret = osdmap_decode(p, end, map);
1196 ceph_osdmap_destroy(map);
1197 return ERR_PTR(ret);
1204 * decode and apply an incremental map update.
1206 struct ceph_osdmap *osdmap_apply_incremental(void **p, void *end,
1207 struct ceph_osdmap *map)
1209 struct crush_map *newcrush = NULL;
1210 struct ceph_fsid fsid;
1212 struct ceph_timespec modified;
1216 __s32 new_flags, max;
1221 dout("%s %p to %p len %d\n", __func__, *p, end, (int)(end - *p));
1223 err = get_osdmap_client_data_v(p, end, "inc", &struct_v);
1227 /* fsid, epoch, modified, new_pool_max, new_flags */
1228 ceph_decode_need(p, end, sizeof(fsid) + sizeof(u32) + sizeof(modified) +
1229 sizeof(u64) + sizeof(u32), e_inval);
1230 ceph_decode_copy(p, &fsid, sizeof(fsid));
1231 epoch = ceph_decode_32(p);
1232 BUG_ON(epoch != map->epoch+1);
1233 ceph_decode_copy(p, &modified, sizeof(modified));
1234 new_pool_max = ceph_decode_64(p);
1235 new_flags = ceph_decode_32(p);
1238 ceph_decode_32_safe(p, end, len, e_inval);
1240 dout("apply_incremental full map len %d, %p to %p\n",
1242 return ceph_osdmap_decode(p, min(*p+len, end));
1246 ceph_decode_32_safe(p, end, len, e_inval);
1248 newcrush = crush_decode(*p, min(*p+len, end));
1249 if (IS_ERR(newcrush)) {
1250 err = PTR_ERR(newcrush);
1259 map->flags = new_flags;
1260 if (new_pool_max >= 0)
1261 map->pool_max = new_pool_max;
1264 ceph_decode_32_safe(p, end, max, e_inval);
1266 err = osdmap_set_max_osd(map, max);
1272 map->modified = modified;
1275 crush_destroy(map->crush);
1276 map->crush = newcrush;
1281 err = decode_new_pools(p, end, map);
1285 /* new_pool_names */
1286 err = decode_pool_names(p, end, map);
1291 ceph_decode_32_safe(p, end, len, e_inval);
1293 struct ceph_pg_pool_info *pi;
1295 ceph_decode_64_safe(p, end, pool, e_inval);
1296 pi = __lookup_pg_pool(&map->pg_pools, pool);
1298 __remove_pg_pool(&map->pg_pools, pi);
1302 ceph_decode_32_safe(p, end, len, e_inval);
1305 struct ceph_entity_addr addr;
1306 ceph_decode_32_safe(p, end, osd, e_inval);
1307 ceph_decode_copy_safe(p, end, &addr, sizeof(addr), e_inval);
1308 ceph_decode_addr(&addr);
1309 pr_info("osd%d up\n", osd);
1310 BUG_ON(osd >= map->max_osd);
1311 map->osd_state[osd] |= CEPH_OSD_UP | CEPH_OSD_EXISTS;
1312 map->osd_addr[osd] = addr;
1316 ceph_decode_32_safe(p, end, len, e_inval);
1320 ceph_decode_32_safe(p, end, osd, e_inval);
1321 xorstate = **(u8 **)p;
1322 (*p)++; /* clean flag */
1324 xorstate = CEPH_OSD_UP;
1325 if (xorstate & CEPH_OSD_UP)
1326 pr_info("osd%d down\n", osd);
1327 if (osd < map->max_osd)
1328 map->osd_state[osd] ^= xorstate;
1332 ceph_decode_32_safe(p, end, len, e_inval);
1335 ceph_decode_need(p, end, sizeof(u32)*2, e_inval);
1336 osd = ceph_decode_32(p);
1337 off = ceph_decode_32(p);
1338 pr_info("osd%d weight 0x%x %s\n", osd, off,
1339 off == CEPH_OSD_IN ? "(in)" :
1340 (off == CEPH_OSD_OUT ? "(out)" : ""));
1341 if (osd < map->max_osd)
1342 map->osd_weight[osd] = off;
1346 err = decode_new_pg_temp(p, end, map);
1350 /* new_primary_temp */
1351 if (struct_v >= 1) {
1352 err = decode_new_primary_temp(p, end, map);
1357 /* new_primary_affinity */
1358 if (struct_v >= 2) {
1359 err = decode_new_primary_affinity(p, end, map);
1364 /* ignore the rest */
1367 dout("inc osdmap epoch %d max_osd %d\n", map->epoch, map->max_osd);
1373 pr_err("corrupt inc osdmap (%d) epoch %d off %d (%p of %p-%p)\n",
1374 err, epoch, (int)(*p - start), *p, start, end);
1375 print_hex_dump(KERN_DEBUG, "osdmap: ",
1376 DUMP_PREFIX_OFFSET, 16, 1,
1377 start, end - start, true);
1379 crush_destroy(newcrush);
1380 return ERR_PTR(err);
1383 void ceph_oid_copy(struct ceph_object_id *dest,
1384 const struct ceph_object_id *src)
1386 WARN_ON(!ceph_oid_empty(dest));
1388 if (src->name != src->inline_name) {
1389 /* very rare, see ceph_object_id definition */
1390 dest->name = kmalloc(src->name_len + 1,
1391 GFP_NOIO | __GFP_NOFAIL);
1394 memcpy(dest->name, src->name, src->name_len + 1);
1395 dest->name_len = src->name_len;
1397 EXPORT_SYMBOL(ceph_oid_copy);
1399 static __printf(2, 0)
1400 int oid_printf_vargs(struct ceph_object_id *oid, const char *fmt, va_list ap)
1404 WARN_ON(!ceph_oid_empty(oid));
1406 len = vsnprintf(oid->inline_name, sizeof(oid->inline_name), fmt, ap);
1407 if (len >= sizeof(oid->inline_name))
1410 oid->name_len = len;
1415 * If oid doesn't fit into inline buffer, BUG.
1417 void ceph_oid_printf(struct ceph_object_id *oid, const char *fmt, ...)
1422 BUG_ON(oid_printf_vargs(oid, fmt, ap));
1425 EXPORT_SYMBOL(ceph_oid_printf);
1427 static __printf(3, 0)
1428 int oid_aprintf_vargs(struct ceph_object_id *oid, gfp_t gfp,
1429 const char *fmt, va_list ap)
1435 len = oid_printf_vargs(oid, fmt, aq);
1439 char *external_name;
1441 external_name = kmalloc(len + 1, gfp);
1445 oid->name = external_name;
1446 WARN_ON(vsnprintf(oid->name, len + 1, fmt, ap) != len);
1447 oid->name_len = len;
1454 * If oid doesn't fit into inline buffer, allocate.
1456 int ceph_oid_aprintf(struct ceph_object_id *oid, gfp_t gfp,
1457 const char *fmt, ...)
1463 ret = oid_aprintf_vargs(oid, gfp, fmt, ap);
1468 EXPORT_SYMBOL(ceph_oid_aprintf);
1470 void ceph_oid_destroy(struct ceph_object_id *oid)
1472 if (oid->name != oid->inline_name)
1475 EXPORT_SYMBOL(ceph_oid_destroy);
1478 * calculate file layout from given offset, length.
1479 * fill in correct oid, logical length, and object extent
1482 * for now, we write only a single su, until we can
1483 * pass a stride back to the caller.
1485 int ceph_calc_file_object_mapping(struct ceph_file_layout *layout,
1488 u64 *oxoff, u64 *oxlen)
1490 u32 osize = le32_to_cpu(layout->fl_object_size);
1491 u32 su = le32_to_cpu(layout->fl_stripe_unit);
1492 u32 sc = le32_to_cpu(layout->fl_stripe_count);
1493 u32 bl, stripeno, stripepos, objsetno;
1497 dout("mapping %llu~%llu osize %u fl_su %u\n", off, len,
1499 if (su == 0 || sc == 0)
1501 su_per_object = osize / su;
1502 if (su_per_object == 0)
1504 dout("osize %u / su %u = su_per_object %u\n", osize, su,
1507 if ((su & ~PAGE_MASK) != 0)
1510 /* bl = *off / su; */
1514 dout("off %llu / su %u = bl %u\n", off, su, bl);
1517 stripepos = bl % sc;
1518 objsetno = stripeno / su_per_object;
1520 *ono = objsetno * sc + stripepos;
1521 dout("objset %u * sc %u = ono %u\n", objsetno, sc, (unsigned int)*ono);
1523 /* *oxoff = *off % layout->fl_stripe_unit; # offset in su */
1525 su_offset = do_div(t, su);
1526 *oxoff = su_offset + (stripeno % su_per_object) * su;
1529 * Calculate the length of the extent being written to the selected
1530 * object. This is the minimum of the full length requested (len) or
1531 * the remainder of the current stripe being written to.
1533 *oxlen = min_t(u64, len, su - su_offset);
1535 dout(" obj extent %llu~%llu\n", *oxoff, *oxlen);
1539 dout(" invalid layout\n");
1545 EXPORT_SYMBOL(ceph_calc_file_object_mapping);
1548 * Map an object into a PG.
1550 * Should only be called with target_oid and target_oloc (as opposed to
1551 * base_oid and base_oloc), since tiering isn't taken into account.
1553 int ceph_object_locator_to_pg(struct ceph_osdmap *osdmap,
1554 struct ceph_object_id *oid,
1555 struct ceph_object_locator *oloc,
1556 struct ceph_pg *raw_pgid)
1558 struct ceph_pg_pool_info *pi;
1560 pi = ceph_pg_pool_by_id(osdmap, oloc->pool);
1564 raw_pgid->pool = oloc->pool;
1565 raw_pgid->seed = ceph_str_hash(pi->object_hash, oid->name,
1568 dout("%s %*pE -> raw_pgid %llu.%x\n", __func__, oid->name_len,
1569 oid->name, raw_pgid->pool, raw_pgid->seed);
1572 EXPORT_SYMBOL(ceph_object_locator_to_pg);
1574 static int do_crush(struct ceph_osdmap *map, int ruleno, int x,
1575 int *result, int result_max,
1576 const __u32 *weight, int weight_max)
1580 BUG_ON(result_max > CEPH_PG_MAX_SIZE);
1582 mutex_lock(&map->crush_scratch_mutex);
1583 r = crush_do_rule(map->crush, ruleno, x, result, result_max,
1584 weight, weight_max, map->crush_scratch_ary);
1585 mutex_unlock(&map->crush_scratch_mutex);
1591 * Calculate raw (crush) set for given pgid.
1593 * Return raw set length, or error.
1595 static int pg_to_raw_osds(struct ceph_osdmap *osdmap,
1596 struct ceph_pg_pool_info *pool,
1597 struct ceph_pg pgid, u32 pps, int *osds)
1603 ruleno = crush_find_rule(osdmap->crush, pool->crush_ruleset,
1604 pool->type, pool->size);
1606 pr_err("no crush rule: pool %lld ruleset %d type %d size %d\n",
1607 pgid.pool, pool->crush_ruleset, pool->type,
1612 len = do_crush(osdmap, ruleno, pps, osds,
1613 min_t(int, pool->size, CEPH_PG_MAX_SIZE),
1614 osdmap->osd_weight, osdmap->max_osd);
1616 pr_err("error %d from crush rule %d: pool %lld ruleset %d type %d size %d\n",
1617 len, ruleno, pgid.pool, pool->crush_ruleset,
1618 pool->type, pool->size);
1626 * Given raw set, calculate up set and up primary.
1628 * Return up set length. *primary is set to up primary osd id, or -1
1629 * if up set is empty.
1631 static int raw_to_up_osds(struct ceph_osdmap *osdmap,
1632 struct ceph_pg_pool_info *pool,
1633 int *osds, int len, int *primary)
1635 int up_primary = -1;
1638 if (ceph_can_shift_osds(pool)) {
1641 for (i = 0; i < len; i++) {
1642 if (ceph_osd_is_down(osdmap, osds[i])) {
1647 osds[i - removed] = osds[i];
1652 up_primary = osds[0];
1654 for (i = len - 1; i >= 0; i--) {
1655 if (ceph_osd_is_down(osdmap, osds[i]))
1656 osds[i] = CRUSH_ITEM_NONE;
1658 up_primary = osds[i];
1662 *primary = up_primary;
1666 static void apply_primary_affinity(struct ceph_osdmap *osdmap, u32 pps,
1667 struct ceph_pg_pool_info *pool,
1668 int *osds, int len, int *primary)
1674 * Do we have any non-default primary_affinity values for these
1677 if (!osdmap->osd_primary_affinity)
1680 for (i = 0; i < len; i++) {
1683 if (osd != CRUSH_ITEM_NONE &&
1684 osdmap->osd_primary_affinity[osd] !=
1685 CEPH_OSD_DEFAULT_PRIMARY_AFFINITY) {
1693 * Pick the primary. Feed both the seed (for the pg) and the
1694 * osd into the hash/rng so that a proportional fraction of an
1695 * osd's pgs get rejected as primary.
1697 for (i = 0; i < len; i++) {
1701 if (osd == CRUSH_ITEM_NONE)
1704 aff = osdmap->osd_primary_affinity[osd];
1705 if (aff < CEPH_OSD_MAX_PRIMARY_AFFINITY &&
1706 (crush_hash32_2(CRUSH_HASH_RJENKINS1,
1707 pps, osd) >> 16) >= aff) {
1709 * We chose not to use this primary. Note it
1710 * anyway as a fallback in case we don't pick
1711 * anyone else, but keep looking.
1723 *primary = osds[pos];
1725 if (ceph_can_shift_osds(pool) && pos > 0) {
1726 /* move the new primary to the front */
1727 for (i = pos; i > 0; i--)
1728 osds[i] = osds[i - 1];
1734 * Given up set, apply pg_temp and primary_temp mappings.
1736 * Return acting set length. *primary is set to acting primary osd id,
1737 * or -1 if acting set is empty.
1739 static int apply_temps(struct ceph_osdmap *osdmap,
1740 struct ceph_pg_pool_info *pool, struct ceph_pg pgid,
1741 int *osds, int len, int *primary)
1743 struct ceph_pg_mapping *pg;
1749 pgid.seed = ceph_stable_mod(pgid.seed, pool->pg_num,
1753 pg = __lookup_pg_mapping(&osdmap->pg_temp, pgid);
1758 for (i = 0; i < pg->pg_temp.len; i++) {
1759 if (ceph_osd_is_down(osdmap, pg->pg_temp.osds[i])) {
1760 if (ceph_can_shift_osds(pool))
1763 osds[temp_len++] = CRUSH_ITEM_NONE;
1765 osds[temp_len++] = pg->pg_temp.osds[i];
1769 /* apply pg_temp's primary */
1770 for (i = 0; i < temp_len; i++) {
1771 if (osds[i] != CRUSH_ITEM_NONE) {
1772 temp_primary = osds[i];
1778 temp_primary = *primary;
1782 pg = __lookup_pg_mapping(&osdmap->primary_temp, pgid);
1784 temp_primary = pg->primary_temp.osd;
1786 *primary = temp_primary;
1791 * Calculate acting set for given pgid.
1793 * Return acting set length, or error. *primary is set to acting
1794 * primary osd id, or -1 if acting set is empty or on error.
1796 int ceph_calc_pg_acting(struct ceph_osdmap *osdmap, struct ceph_pg pgid,
1797 int *osds, int *primary)
1799 struct ceph_pg_pool_info *pool;
1803 pool = __lookup_pg_pool(&osdmap->pg_pools, pgid.pool);
1809 if (pool->flags & CEPH_POOL_FLAG_HASHPSPOOL) {
1810 /* hash pool id and seed so that pool PGs do not overlap */
1811 pps = crush_hash32_2(CRUSH_HASH_RJENKINS1,
1812 ceph_stable_mod(pgid.seed, pool->pgp_num,
1813 pool->pgp_num_mask),
1817 * legacy behavior: add ps and pool together. this is
1818 * not a great approach because the PGs from each pool
1819 * will overlap on top of each other: 0.5 == 1.4 ==
1822 pps = ceph_stable_mod(pgid.seed, pool->pgp_num,
1823 pool->pgp_num_mask) +
1824 (unsigned)pgid.pool;
1827 len = pg_to_raw_osds(osdmap, pool, pgid, pps, osds);
1833 len = raw_to_up_osds(osdmap, pool, osds, len, primary);
1835 apply_primary_affinity(osdmap, pps, pool, osds, len, primary);
1837 len = apply_temps(osdmap, pool, pgid, osds, len, primary);
1843 * Return primary osd for given pgid, or -1 if none.
1845 int ceph_calc_pg_primary(struct ceph_osdmap *osdmap, struct ceph_pg pgid)
1847 int osds[CEPH_PG_MAX_SIZE];
1850 ceph_calc_pg_acting(osdmap, pgid, osds, &primary);
1854 EXPORT_SYMBOL(ceph_calc_pg_primary);