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_32_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 skip_name_map(void **p, void *end)
128 ceph_decode_32_safe(p, end, len ,bad);
132 ceph_decode_32_safe(p, end, strlen, bad);
140 static struct crush_map *crush_decode(void *pbyval, void *end)
146 void *start = pbyval;
150 dout("crush_decode %p to %p len %d\n", *p, end, (int)(end - *p));
152 c = kzalloc(sizeof(*c), GFP_NOFS);
154 return ERR_PTR(-ENOMEM);
156 /* set tunables to default values */
157 c->choose_local_tries = 2;
158 c->choose_local_fallback_tries = 5;
159 c->choose_total_tries = 19;
160 c->chooseleaf_descend_once = 0;
162 ceph_decode_need(p, end, 4*sizeof(u32), bad);
163 magic = ceph_decode_32(p);
164 if (magic != CRUSH_MAGIC) {
165 pr_err("crush_decode magic %x != current %x\n",
166 (unsigned int)magic, (unsigned int)CRUSH_MAGIC);
169 c->max_buckets = ceph_decode_32(p);
170 c->max_rules = ceph_decode_32(p);
171 c->max_devices = ceph_decode_32(p);
173 c->buckets = kcalloc(c->max_buckets, sizeof(*c->buckets), GFP_NOFS);
174 if (c->buckets == NULL)
176 c->rules = kcalloc(c->max_rules, sizeof(*c->rules), GFP_NOFS);
177 if (c->rules == NULL)
181 for (i = 0; i < c->max_buckets; i++) {
184 struct crush_bucket *b;
186 ceph_decode_32_safe(p, end, alg, bad);
188 c->buckets[i] = NULL;
191 dout("crush_decode bucket %d off %x %p to %p\n",
192 i, (int)(*p-start), *p, end);
195 case CRUSH_BUCKET_UNIFORM:
196 size = sizeof(struct crush_bucket_uniform);
198 case CRUSH_BUCKET_LIST:
199 size = sizeof(struct crush_bucket_list);
201 case CRUSH_BUCKET_TREE:
202 size = sizeof(struct crush_bucket_tree);
204 case CRUSH_BUCKET_STRAW:
205 size = sizeof(struct crush_bucket_straw);
212 b = c->buckets[i] = kzalloc(size, GFP_NOFS);
216 ceph_decode_need(p, end, 4*sizeof(u32), bad);
217 b->id = ceph_decode_32(p);
218 b->type = ceph_decode_16(p);
219 b->alg = ceph_decode_8(p);
220 b->hash = ceph_decode_8(p);
221 b->weight = ceph_decode_32(p);
222 b->size = ceph_decode_32(p);
224 dout("crush_decode bucket size %d off %x %p to %p\n",
225 b->size, (int)(*p-start), *p, end);
227 b->items = kcalloc(b->size, sizeof(__s32), GFP_NOFS);
228 if (b->items == NULL)
230 b->perm = kcalloc(b->size, sizeof(u32), GFP_NOFS);
235 ceph_decode_need(p, end, b->size*sizeof(u32), bad);
236 for (j = 0; j < b->size; j++)
237 b->items[j] = ceph_decode_32(p);
240 case CRUSH_BUCKET_UNIFORM:
241 err = crush_decode_uniform_bucket(p, end,
242 (struct crush_bucket_uniform *)b);
246 case CRUSH_BUCKET_LIST:
247 err = crush_decode_list_bucket(p, end,
248 (struct crush_bucket_list *)b);
252 case CRUSH_BUCKET_TREE:
253 err = crush_decode_tree_bucket(p, end,
254 (struct crush_bucket_tree *)b);
258 case CRUSH_BUCKET_STRAW:
259 err = crush_decode_straw_bucket(p, end,
260 (struct crush_bucket_straw *)b);
268 dout("rule vec is %p\n", c->rules);
269 for (i = 0; i < c->max_rules; i++) {
271 struct crush_rule *r;
273 ceph_decode_32_safe(p, end, yes, bad);
275 dout("crush_decode NO rule %d off %x %p to %p\n",
276 i, (int)(*p-start), *p, end);
281 dout("crush_decode rule %d off %x %p to %p\n",
282 i, (int)(*p-start), *p, end);
285 ceph_decode_32_safe(p, end, yes, bad);
286 #if BITS_PER_LONG == 32
288 if (yes > (ULONG_MAX - sizeof(*r))
289 / sizeof(struct crush_rule_step))
292 r = c->rules[i] = kmalloc(sizeof(*r) +
293 yes*sizeof(struct crush_rule_step),
297 dout(" rule %d is at %p\n", i, r);
299 ceph_decode_copy_safe(p, end, &r->mask, 4, bad); /* 4 u8's */
300 ceph_decode_need(p, end, r->len*3*sizeof(u32), bad);
301 for (j = 0; j < r->len; j++) {
302 r->steps[j].op = ceph_decode_32(p);
303 r->steps[j].arg1 = ceph_decode_32(p);
304 r->steps[j].arg2 = ceph_decode_32(p);
308 /* ignore trailing name maps. */
309 for (num_name_maps = 0; num_name_maps < 3; num_name_maps++) {
310 err = skip_name_map(p, end);
316 ceph_decode_need(p, end, 3*sizeof(u32), done);
317 c->choose_local_tries = ceph_decode_32(p);
318 c->choose_local_fallback_tries = ceph_decode_32(p);
319 c->choose_total_tries = ceph_decode_32(p);
320 dout("crush decode tunable choose_local_tries = %d",
321 c->choose_local_tries);
322 dout("crush decode tunable choose_local_fallback_tries = %d",
323 c->choose_local_fallback_tries);
324 dout("crush decode tunable choose_total_tries = %d",
325 c->choose_total_tries);
327 ceph_decode_need(p, end, sizeof(u32), done);
328 c->chooseleaf_descend_once = ceph_decode_32(p);
329 dout("crush decode tunable chooseleaf_descend_once = %d",
330 c->chooseleaf_descend_once);
333 dout("crush_decode success\n");
339 dout("crush_decode fail %d\n", err);
345 * rbtree of pg_mapping for handling pg_temp (explicit mapping of pgid
346 * to a set of osds) and primary_temp (explicit primary setting)
348 static int pgid_cmp(struct ceph_pg l, struct ceph_pg r)
361 static int __insert_pg_mapping(struct ceph_pg_mapping *new,
362 struct rb_root *root)
364 struct rb_node **p = &root->rb_node;
365 struct rb_node *parent = NULL;
366 struct ceph_pg_mapping *pg = NULL;
369 dout("__insert_pg_mapping %llx %p\n", *(u64 *)&new->pgid, new);
372 pg = rb_entry(parent, struct ceph_pg_mapping, node);
373 c = pgid_cmp(new->pgid, pg->pgid);
382 rb_link_node(&new->node, parent, p);
383 rb_insert_color(&new->node, root);
387 static struct ceph_pg_mapping *__lookup_pg_mapping(struct rb_root *root,
390 struct rb_node *n = root->rb_node;
391 struct ceph_pg_mapping *pg;
395 pg = rb_entry(n, struct ceph_pg_mapping, node);
396 c = pgid_cmp(pgid, pg->pgid);
402 dout("__lookup_pg_mapping %lld.%x got %p\n",
403 pgid.pool, pgid.seed, pg);
410 static int __remove_pg_mapping(struct rb_root *root, struct ceph_pg pgid)
412 struct ceph_pg_mapping *pg = __lookup_pg_mapping(root, pgid);
415 dout("__remove_pg_mapping %lld.%x %p\n", pgid.pool, pgid.seed,
417 rb_erase(&pg->node, root);
421 dout("__remove_pg_mapping %lld.%x dne\n", pgid.pool, pgid.seed);
426 * rbtree of pg pool info
428 static int __insert_pg_pool(struct rb_root *root, struct ceph_pg_pool_info *new)
430 struct rb_node **p = &root->rb_node;
431 struct rb_node *parent = NULL;
432 struct ceph_pg_pool_info *pi = NULL;
436 pi = rb_entry(parent, struct ceph_pg_pool_info, node);
437 if (new->id < pi->id)
439 else if (new->id > pi->id)
445 rb_link_node(&new->node, parent, p);
446 rb_insert_color(&new->node, root);
450 static struct ceph_pg_pool_info *__lookup_pg_pool(struct rb_root *root, u64 id)
452 struct ceph_pg_pool_info *pi;
453 struct rb_node *n = root->rb_node;
456 pi = rb_entry(n, struct ceph_pg_pool_info, node);
459 else if (id > pi->id)
467 struct ceph_pg_pool_info *ceph_pg_pool_by_id(struct ceph_osdmap *map, u64 id)
469 return __lookup_pg_pool(&map->pg_pools, id);
472 const char *ceph_pg_pool_name_by_id(struct ceph_osdmap *map, u64 id)
474 struct ceph_pg_pool_info *pi;
476 if (id == CEPH_NOPOOL)
479 if (WARN_ON_ONCE(id > (u64) INT_MAX))
482 pi = __lookup_pg_pool(&map->pg_pools, (int) id);
484 return pi ? pi->name : NULL;
486 EXPORT_SYMBOL(ceph_pg_pool_name_by_id);
488 int ceph_pg_poolid_by_name(struct ceph_osdmap *map, const char *name)
492 for (rbp = rb_first(&map->pg_pools); rbp; rbp = rb_next(rbp)) {
493 struct ceph_pg_pool_info *pi =
494 rb_entry(rbp, struct ceph_pg_pool_info, node);
495 if (pi->name && strcmp(pi->name, name) == 0)
500 EXPORT_SYMBOL(ceph_pg_poolid_by_name);
502 static void __remove_pg_pool(struct rb_root *root, struct ceph_pg_pool_info *pi)
504 rb_erase(&pi->node, root);
509 static int decode_pool(void **p, void *end, struct ceph_pg_pool_info *pi)
515 ceph_decode_need(p, end, 2 + 4, bad);
516 ev = ceph_decode_8(p); /* encoding version */
517 cv = ceph_decode_8(p); /* compat version */
519 pr_warning("got v %d < 5 cv %d of ceph_pg_pool\n", ev, cv);
523 pr_warning("got v %d cv %d > 9 of ceph_pg_pool\n", ev, cv);
526 len = ceph_decode_32(p);
527 ceph_decode_need(p, end, len, bad);
530 pi->type = ceph_decode_8(p);
531 pi->size = ceph_decode_8(p);
532 pi->crush_ruleset = ceph_decode_8(p);
533 pi->object_hash = ceph_decode_8(p);
535 pi->pg_num = ceph_decode_32(p);
536 pi->pgp_num = ceph_decode_32(p);
538 *p += 4 + 4; /* skip lpg* */
539 *p += 4; /* skip last_change */
540 *p += 8 + 4; /* skip snap_seq, snap_epoch */
543 num = ceph_decode_32(p);
545 *p += 8; /* snapid key */
546 *p += 1 + 1; /* versions */
547 len = ceph_decode_32(p);
551 /* skip removed_snaps */
552 num = ceph_decode_32(p);
555 *p += 8; /* skip auid */
556 pi->flags = ceph_decode_64(p);
557 *p += 4; /* skip crash_replay_interval */
560 *p += 1; /* skip min_size */
563 *p += 8 + 8; /* skip quota_max_* */
567 num = ceph_decode_32(p);
570 *p += 8; /* skip tier_of */
571 *p += 1; /* skip cache_mode */
573 pi->read_tier = ceph_decode_64(p);
574 pi->write_tier = ceph_decode_64(p);
580 /* ignore the rest */
590 static int decode_pool_names(void **p, void *end, struct ceph_osdmap *map)
592 struct ceph_pg_pool_info *pi;
596 ceph_decode_32_safe(p, end, num, bad);
597 dout(" %d pool names\n", num);
599 ceph_decode_64_safe(p, end, pool, bad);
600 ceph_decode_32_safe(p, end, len, bad);
601 dout(" pool %llu len %d\n", pool, len);
602 ceph_decode_need(p, end, len, bad);
603 pi = __lookup_pg_pool(&map->pg_pools, pool);
605 char *name = kstrndup(*p, len, GFP_NOFS);
611 dout(" name is %s\n", pi->name);
624 void ceph_osdmap_destroy(struct ceph_osdmap *map)
626 dout("osdmap_destroy %p\n", map);
628 crush_destroy(map->crush);
629 while (!RB_EMPTY_ROOT(&map->pg_temp)) {
630 struct ceph_pg_mapping *pg =
631 rb_entry(rb_first(&map->pg_temp),
632 struct ceph_pg_mapping, node);
633 rb_erase(&pg->node, &map->pg_temp);
636 while (!RB_EMPTY_ROOT(&map->primary_temp)) {
637 struct ceph_pg_mapping *pg =
638 rb_entry(rb_first(&map->primary_temp),
639 struct ceph_pg_mapping, node);
640 rb_erase(&pg->node, &map->primary_temp);
643 while (!RB_EMPTY_ROOT(&map->pg_pools)) {
644 struct ceph_pg_pool_info *pi =
645 rb_entry(rb_first(&map->pg_pools),
646 struct ceph_pg_pool_info, node);
647 __remove_pg_pool(&map->pg_pools, pi);
649 kfree(map->osd_state);
650 kfree(map->osd_weight);
651 kfree(map->osd_addr);
656 * Adjust max_osd value, (re)allocate arrays.
658 * The new elements are properly initialized.
660 static int osdmap_set_max_osd(struct ceph_osdmap *map, int max)
664 struct ceph_entity_addr *addr;
667 state = krealloc(map->osd_state, max*sizeof(*state), GFP_NOFS);
668 weight = krealloc(map->osd_weight, max*sizeof(*weight), GFP_NOFS);
669 addr = krealloc(map->osd_addr, max*sizeof(*addr), GFP_NOFS);
670 if (!state || !weight || !addr) {
678 for (i = map->max_osd; i < max; i++) {
680 weight[i] = CEPH_OSD_OUT;
681 memset(addr + i, 0, sizeof(*addr));
684 map->osd_state = state;
685 map->osd_weight = weight;
686 map->osd_addr = addr;
693 #define OSDMAP_WRAPPER_COMPAT_VER 7
694 #define OSDMAP_CLIENT_DATA_COMPAT_VER 1
697 * Return 0 or error. On success, *v is set to 0 for old (v6) osdmaps,
698 * to struct_v of the client_data section for new (v7 and above)
701 static int get_osdmap_client_data_v(void **p, void *end,
702 const char *prefix, u8 *v)
706 ceph_decode_8_safe(p, end, struct_v, e_inval);
710 ceph_decode_8_safe(p, end, struct_compat, e_inval);
711 if (struct_compat > OSDMAP_WRAPPER_COMPAT_VER) {
712 pr_warning("got v %d cv %d > %d of %s ceph_osdmap\n",
713 struct_v, struct_compat,
714 OSDMAP_WRAPPER_COMPAT_VER, prefix);
717 *p += 4; /* ignore wrapper struct_len */
719 ceph_decode_8_safe(p, end, struct_v, e_inval);
720 ceph_decode_8_safe(p, end, struct_compat, e_inval);
721 if (struct_compat > OSDMAP_CLIENT_DATA_COMPAT_VER) {
722 pr_warning("got v %d cv %d > %d of %s ceph_osdmap client data\n",
723 struct_v, struct_compat,
724 OSDMAP_CLIENT_DATA_COMPAT_VER, prefix);
727 *p += 4; /* ignore client data struct_len */
732 ceph_decode_16_safe(p, end, version, e_inval);
734 pr_warning("got v %d < 6 of %s ceph_osdmap\n", version,
739 /* old osdmap enconding */
750 static int __decode_pools(void **p, void *end, struct ceph_osdmap *map,
755 ceph_decode_32_safe(p, end, n, e_inval);
757 struct ceph_pg_pool_info *pi;
761 ceph_decode_64_safe(p, end, pool, e_inval);
763 pi = __lookup_pg_pool(&map->pg_pools, pool);
764 if (!incremental || !pi) {
765 pi = kzalloc(sizeof(*pi), GFP_NOFS);
771 ret = __insert_pg_pool(&map->pg_pools, pi);
778 ret = decode_pool(p, end, pi);
789 static int decode_pools(void **p, void *end, struct ceph_osdmap *map)
791 return __decode_pools(p, end, map, false);
794 static int decode_new_pools(void **p, void *end, struct ceph_osdmap *map)
796 return __decode_pools(p, end, map, true);
799 static int __decode_pg_temp(void **p, void *end, struct ceph_osdmap *map,
804 ceph_decode_32_safe(p, end, n, e_inval);
810 ret = ceph_decode_pgid(p, end, &pgid);
814 ceph_decode_32_safe(p, end, len, e_inval);
816 ret = __remove_pg_mapping(&map->pg_temp, pgid);
817 BUG_ON(!incremental && ret != -ENOENT);
819 if (!incremental || len > 0) {
820 struct ceph_pg_mapping *pg;
822 ceph_decode_need(p, end, len*sizeof(u32), e_inval);
824 if (len > (UINT_MAX - sizeof(*pg)) / sizeof(u32))
827 pg = kzalloc(sizeof(*pg) + len*sizeof(u32), GFP_NOFS);
832 pg->pg_temp.len = len;
833 for (i = 0; i < len; i++)
834 pg->pg_temp.osds[i] = ceph_decode_32(p);
836 ret = __insert_pg_mapping(pg, &map->pg_temp);
850 static int decode_pg_temp(void **p, void *end, struct ceph_osdmap *map)
852 return __decode_pg_temp(p, end, map, false);
855 static int decode_new_pg_temp(void **p, void *end, struct ceph_osdmap *map)
857 return __decode_pg_temp(p, end, map, true);
863 static int osdmap_decode(void **p, void *end, struct ceph_osdmap *map)
872 dout("%s %p to %p len %d\n", __func__, *p, end, (int)(end - *p));
874 err = get_osdmap_client_data_v(p, end, "full", &struct_v);
878 /* fsid, epoch, created, modified */
879 ceph_decode_need(p, end, sizeof(map->fsid) + sizeof(u32) +
880 sizeof(map->created) + sizeof(map->modified), e_inval);
881 ceph_decode_copy(p, &map->fsid, sizeof(map->fsid));
882 epoch = map->epoch = ceph_decode_32(p);
883 ceph_decode_copy(p, &map->created, sizeof(map->created));
884 ceph_decode_copy(p, &map->modified, sizeof(map->modified));
887 err = decode_pools(p, end, map);
892 err = decode_pool_names(p, end, map);
896 ceph_decode_32_safe(p, end, map->pool_max, e_inval);
898 ceph_decode_32_safe(p, end, map->flags, e_inval);
901 ceph_decode_32_safe(p, end, max, e_inval);
903 /* (re)alloc osd arrays */
904 err = osdmap_set_max_osd(map, max);
908 /* osd_state, osd_weight, osd_addrs->client_addr */
909 ceph_decode_need(p, end, 3*sizeof(u32) +
910 map->max_osd*(1 + sizeof(*map->osd_weight) +
911 sizeof(*map->osd_addr)), e_inval);
913 if (ceph_decode_32(p) != map->max_osd)
916 ceph_decode_copy(p, map->osd_state, map->max_osd);
918 if (ceph_decode_32(p) != map->max_osd)
921 for (i = 0; i < map->max_osd; i++)
922 map->osd_weight[i] = ceph_decode_32(p);
924 if (ceph_decode_32(p) != map->max_osd)
927 ceph_decode_copy(p, map->osd_addr, map->max_osd*sizeof(*map->osd_addr));
928 for (i = 0; i < map->max_osd; i++)
929 ceph_decode_addr(&map->osd_addr[i]);
932 err = decode_pg_temp(p, end, map);
937 ceph_decode_32_safe(p, end, len, e_inval);
938 map->crush = crush_decode(*p, min(*p + len, end));
939 if (IS_ERR(map->crush)) {
940 err = PTR_ERR(map->crush);
946 /* ignore the rest */
949 dout("full osdmap epoch %d max_osd %d\n", map->epoch, map->max_osd);
955 pr_err("corrupt full osdmap (%d) epoch %d off %d (%p of %p-%p)\n",
956 err, epoch, (int)(*p - start), *p, start, end);
957 print_hex_dump(KERN_DEBUG, "osdmap: ",
958 DUMP_PREFIX_OFFSET, 16, 1,
959 start, end - start, true);
964 * Allocate and decode a full map.
966 struct ceph_osdmap *ceph_osdmap_decode(void **p, void *end)
968 struct ceph_osdmap *map;
971 map = kzalloc(sizeof(*map), GFP_NOFS);
973 return ERR_PTR(-ENOMEM);
975 map->pg_temp = RB_ROOT;
976 map->primary_temp = RB_ROOT;
977 mutex_init(&map->crush_scratch_mutex);
979 ret = osdmap_decode(p, end, map);
981 ceph_osdmap_destroy(map);
989 * decode and apply an incremental map update.
991 struct ceph_osdmap *osdmap_apply_incremental(void **p, void *end,
992 struct ceph_osdmap *map,
993 struct ceph_messenger *msgr)
995 struct crush_map *newcrush = NULL;
996 struct ceph_fsid fsid;
998 struct ceph_timespec modified;
1002 __s32 new_flags, max;
1007 dout("%s %p to %p len %d\n", __func__, *p, end, (int)(end - *p));
1009 err = get_osdmap_client_data_v(p, end, "inc", &struct_v);
1013 /* fsid, epoch, modified, new_pool_max, new_flags */
1014 ceph_decode_need(p, end, sizeof(fsid) + sizeof(u32) + sizeof(modified) +
1015 sizeof(u64) + sizeof(u32), e_inval);
1016 ceph_decode_copy(p, &fsid, sizeof(fsid));
1017 epoch = ceph_decode_32(p);
1018 BUG_ON(epoch != map->epoch+1);
1019 ceph_decode_copy(p, &modified, sizeof(modified));
1020 new_pool_max = ceph_decode_64(p);
1021 new_flags = ceph_decode_32(p);
1024 ceph_decode_32_safe(p, end, len, e_inval);
1026 dout("apply_incremental full map len %d, %p to %p\n",
1028 return ceph_osdmap_decode(p, min(*p+len, end));
1032 ceph_decode_32_safe(p, end, len, e_inval);
1034 newcrush = crush_decode(*p, min(*p+len, end));
1035 if (IS_ERR(newcrush)) {
1036 err = PTR_ERR(newcrush);
1045 map->flags = new_flags;
1046 if (new_pool_max >= 0)
1047 map->pool_max = new_pool_max;
1050 ceph_decode_32_safe(p, end, max, e_inval);
1052 err = osdmap_set_max_osd(map, max);
1058 map->modified = modified;
1061 crush_destroy(map->crush);
1062 map->crush = newcrush;
1067 err = decode_new_pools(p, end, map);
1071 /* new_pool_names */
1072 err = decode_pool_names(p, end, map);
1077 ceph_decode_32_safe(p, end, len, e_inval);
1079 struct ceph_pg_pool_info *pi;
1081 ceph_decode_64_safe(p, end, pool, e_inval);
1082 pi = __lookup_pg_pool(&map->pg_pools, pool);
1084 __remove_pg_pool(&map->pg_pools, pi);
1088 ceph_decode_32_safe(p, end, len, e_inval);
1091 struct ceph_entity_addr addr;
1092 ceph_decode_32_safe(p, end, osd, e_inval);
1093 ceph_decode_copy_safe(p, end, &addr, sizeof(addr), e_inval);
1094 ceph_decode_addr(&addr);
1095 pr_info("osd%d up\n", osd);
1096 BUG_ON(osd >= map->max_osd);
1097 map->osd_state[osd] |= CEPH_OSD_UP;
1098 map->osd_addr[osd] = addr;
1102 ceph_decode_32_safe(p, end, len, e_inval);
1106 ceph_decode_32_safe(p, end, osd, e_inval);
1107 xorstate = **(u8 **)p;
1108 (*p)++; /* clean flag */
1110 xorstate = CEPH_OSD_UP;
1111 if (xorstate & CEPH_OSD_UP)
1112 pr_info("osd%d down\n", osd);
1113 if (osd < map->max_osd)
1114 map->osd_state[osd] ^= xorstate;
1118 ceph_decode_32_safe(p, end, len, e_inval);
1121 ceph_decode_need(p, end, sizeof(u32)*2, e_inval);
1122 osd = ceph_decode_32(p);
1123 off = ceph_decode_32(p);
1124 pr_info("osd%d weight 0x%x %s\n", osd, off,
1125 off == CEPH_OSD_IN ? "(in)" :
1126 (off == CEPH_OSD_OUT ? "(out)" : ""));
1127 if (osd < map->max_osd)
1128 map->osd_weight[osd] = off;
1132 err = decode_new_pg_temp(p, end, map);
1136 /* ignore the rest */
1139 dout("inc osdmap epoch %d max_osd %d\n", map->epoch, map->max_osd);
1145 pr_err("corrupt inc osdmap (%d) epoch %d off %d (%p of %p-%p)\n",
1146 err, epoch, (int)(*p - start), *p, start, end);
1147 print_hex_dump(KERN_DEBUG, "osdmap: ",
1148 DUMP_PREFIX_OFFSET, 16, 1,
1149 start, end - start, true);
1151 crush_destroy(newcrush);
1152 return ERR_PTR(err);
1159 * calculate file layout from given offset, length.
1160 * fill in correct oid, logical length, and object extent
1163 * for now, we write only a single su, until we can
1164 * pass a stride back to the caller.
1166 int ceph_calc_file_object_mapping(struct ceph_file_layout *layout,
1169 u64 *oxoff, u64 *oxlen)
1171 u32 osize = le32_to_cpu(layout->fl_object_size);
1172 u32 su = le32_to_cpu(layout->fl_stripe_unit);
1173 u32 sc = le32_to_cpu(layout->fl_stripe_count);
1174 u32 bl, stripeno, stripepos, objsetno;
1178 dout("mapping %llu~%llu osize %u fl_su %u\n", off, len,
1180 if (su == 0 || sc == 0)
1182 su_per_object = osize / su;
1183 if (su_per_object == 0)
1185 dout("osize %u / su %u = su_per_object %u\n", osize, su,
1188 if ((su & ~PAGE_MASK) != 0)
1191 /* bl = *off / su; */
1195 dout("off %llu / su %u = bl %u\n", off, su, bl);
1198 stripepos = bl % sc;
1199 objsetno = stripeno / su_per_object;
1201 *ono = objsetno * sc + stripepos;
1202 dout("objset %u * sc %u = ono %u\n", objsetno, sc, (unsigned int)*ono);
1204 /* *oxoff = *off % layout->fl_stripe_unit; # offset in su */
1206 su_offset = do_div(t, su);
1207 *oxoff = su_offset + (stripeno % su_per_object) * su;
1210 * Calculate the length of the extent being written to the selected
1211 * object. This is the minimum of the full length requested (len) or
1212 * the remainder of the current stripe being written to.
1214 *oxlen = min_t(u64, len, su - su_offset);
1216 dout(" obj extent %llu~%llu\n", *oxoff, *oxlen);
1220 dout(" invalid layout\n");
1226 EXPORT_SYMBOL(ceph_calc_file_object_mapping);
1229 * Calculate mapping of a (oloc, oid) pair to a PG. Should only be
1230 * called with target's (oloc, oid), since tiering isn't taken into
1233 int ceph_oloc_oid_to_pg(struct ceph_osdmap *osdmap,
1234 struct ceph_object_locator *oloc,
1235 struct ceph_object_id *oid,
1236 struct ceph_pg *pg_out)
1238 struct ceph_pg_pool_info *pi;
1240 pi = __lookup_pg_pool(&osdmap->pg_pools, oloc->pool);
1244 pg_out->pool = oloc->pool;
1245 pg_out->seed = ceph_str_hash(pi->object_hash, oid->name,
1248 dout("%s '%.*s' pgid %llu.%x\n", __func__, oid->name_len, oid->name,
1249 pg_out->pool, pg_out->seed);
1252 EXPORT_SYMBOL(ceph_oloc_oid_to_pg);
1254 static int do_crush(struct ceph_osdmap *map, int ruleno, int x,
1255 int *result, int result_max,
1256 const __u32 *weight, int weight_max)
1260 BUG_ON(result_max > CEPH_PG_MAX_SIZE);
1262 mutex_lock(&map->crush_scratch_mutex);
1263 r = crush_do_rule(map->crush, ruleno, x, result, result_max,
1264 weight, weight_max, map->crush_scratch_ary);
1265 mutex_unlock(&map->crush_scratch_mutex);
1271 * Calculate raw osd vector for the given pgid. Return pointer to osd
1272 * array, or NULL on failure.
1274 static int *calc_pg_raw(struct ceph_osdmap *osdmap, struct ceph_pg pgid,
1275 int *osds, int *num)
1277 struct ceph_pg_mapping *pg;
1278 struct ceph_pg_pool_info *pool;
1283 pool = __lookup_pg_pool(&osdmap->pg_pools, pgid.pool);
1288 pgid.seed = ceph_stable_mod(pgid.seed, pool->pg_num,
1290 pg = __lookup_pg_mapping(&osdmap->pg_temp, pgid);
1292 *num = pg->pg_temp.len;
1293 return pg->pg_temp.osds;
1297 ruleno = crush_find_rule(osdmap->crush, pool->crush_ruleset,
1298 pool->type, pool->size);
1300 pr_err("no crush rule pool %lld ruleset %d type %d size %d\n",
1301 pgid.pool, pool->crush_ruleset, pool->type,
1306 if (pool->flags & CEPH_POOL_FLAG_HASHPSPOOL) {
1307 /* hash pool id and seed sothat pool PGs do not overlap */
1308 pps = crush_hash32_2(CRUSH_HASH_RJENKINS1,
1309 ceph_stable_mod(pgid.seed, pool->pgp_num,
1310 pool->pgp_num_mask),
1314 * legacy ehavior: add ps and pool together. this is
1315 * not a great approach because the PGs from each pool
1316 * will overlap on top of each other: 0.5 == 1.4 ==
1319 pps = ceph_stable_mod(pgid.seed, pool->pgp_num,
1320 pool->pgp_num_mask) +
1321 (unsigned)pgid.pool;
1323 r = do_crush(osdmap, ruleno, pps, osds, min_t(int, pool->size, *num),
1324 osdmap->osd_weight, osdmap->max_osd);
1326 pr_err("error %d from crush rule: pool %lld ruleset %d type %d"
1327 " size %d\n", r, pgid.pool, pool->crush_ruleset,
1328 pool->type, pool->size);
1336 * Return acting set for given pgid.
1338 int ceph_calc_pg_acting(struct ceph_osdmap *osdmap, struct ceph_pg pgid,
1341 int rawosds[CEPH_PG_MAX_SIZE], *osds;
1342 int i, o, num = CEPH_PG_MAX_SIZE;
1344 osds = calc_pg_raw(osdmap, pgid, rawosds, &num);
1348 /* primary is first up osd */
1350 for (i = 0; i < num; i++)
1351 if (ceph_osd_is_up(osdmap, osds[i]))
1352 acting[o++] = osds[i];
1357 * Return primary osd for given pgid, or -1 if none.
1359 int ceph_calc_pg_primary(struct ceph_osdmap *osdmap, struct ceph_pg pgid)
1361 int rawosds[CEPH_PG_MAX_SIZE], *osds;
1362 int i, num = CEPH_PG_MAX_SIZE;
1364 osds = calc_pg_raw(osdmap, pgid, rawosds, &num);
1368 /* primary is first up osd */
1369 for (i = 0; i < num; i++)
1370 if (ceph_osd_is_up(osdmap, osds[i]))
1374 EXPORT_SYMBOL(ceph_calc_pg_primary);