6 #include "crush/hash.h"
7 #include "crush/mapper.h"
9 #include "ceph_debug.h"
11 char *ceph_osdmap_state_str(char *str, int len, int state)
20 if (state & CEPH_OSD_EXISTS) {
21 snprintf(str, len, "exists");
24 if (state & CEPH_OSD_UP) {
25 snprintf(str, len, "%s%s%s", str, (flag ? ", " : ""),
30 snprintf(str, len, "doesn't exist");
38 static int calc_bits_of(unsigned t)
49 * the foo_mask is the smallest value 2^n-1 that is >= foo.
51 static void calc_pg_masks(struct ceph_pg_pool_info *pi)
53 pi->pg_num_mask = (1 << calc_bits_of(le32_to_cpu(pi->v.pg_num)-1)) - 1;
55 (1 << calc_bits_of(le32_to_cpu(pi->v.pgp_num)-1)) - 1;
57 (1 << calc_bits_of(le32_to_cpu(pi->v.lpg_num)-1)) - 1;
59 (1 << calc_bits_of(le32_to_cpu(pi->v.lpgp_num)-1)) - 1;
65 static int crush_decode_uniform_bucket(void **p, void *end,
66 struct crush_bucket_uniform *b)
68 dout("crush_decode_uniform_bucket %p to %p\n", *p, end);
69 ceph_decode_need(p, end, (1+b->h.size) * sizeof(u32), bad);
70 b->item_weight = ceph_decode_32(p);
76 static int crush_decode_list_bucket(void **p, void *end,
77 struct crush_bucket_list *b)
80 dout("crush_decode_list_bucket %p to %p\n", *p, end);
81 b->item_weights = kcalloc(b->h.size, sizeof(u32), GFP_NOFS);
82 if (b->item_weights == NULL)
84 b->sum_weights = kcalloc(b->h.size, sizeof(u32), GFP_NOFS);
85 if (b->sum_weights == NULL)
87 ceph_decode_need(p, end, 2 * b->h.size * sizeof(u32), bad);
88 for (j = 0; j < b->h.size; j++) {
89 b->item_weights[j] = ceph_decode_32(p);
90 b->sum_weights[j] = ceph_decode_32(p);
97 static int crush_decode_tree_bucket(void **p, void *end,
98 struct crush_bucket_tree *b)
101 dout("crush_decode_tree_bucket %p to %p\n", *p, end);
102 ceph_decode_32_safe(p, end, b->num_nodes, bad);
103 b->node_weights = kcalloc(b->num_nodes, sizeof(u32), GFP_NOFS);
104 if (b->node_weights == NULL)
106 ceph_decode_need(p, end, b->num_nodes * sizeof(u32), bad);
107 for (j = 0; j < b->num_nodes; j++)
108 b->node_weights[j] = ceph_decode_32(p);
114 static int crush_decode_straw_bucket(void **p, void *end,
115 struct crush_bucket_straw *b)
118 dout("crush_decode_straw_bucket %p to %p\n", *p, end);
119 b->item_weights = kcalloc(b->h.size, sizeof(u32), GFP_NOFS);
120 if (b->item_weights == NULL)
122 b->straws = kcalloc(b->h.size, sizeof(u32), GFP_NOFS);
123 if (b->straws == NULL)
125 ceph_decode_need(p, end, 2 * b->h.size * sizeof(u32), bad);
126 for (j = 0; j < b->h.size; j++) {
127 b->item_weights[j] = ceph_decode_32(p);
128 b->straws[j] = ceph_decode_32(p);
135 static struct crush_map *crush_decode(void *pbyval, void *end)
141 void *start = pbyval;
144 dout("crush_decode %p to %p len %d\n", *p, end, (int)(end - *p));
146 c = kzalloc(sizeof(*c), GFP_NOFS);
148 return ERR_PTR(-ENOMEM);
150 ceph_decode_need(p, end, 4*sizeof(u32), bad);
151 magic = ceph_decode_32(p);
152 if (magic != CRUSH_MAGIC) {
153 pr_err("crush_decode magic %x != current %x\n",
154 (unsigned)magic, (unsigned)CRUSH_MAGIC);
157 c->max_buckets = ceph_decode_32(p);
158 c->max_rules = ceph_decode_32(p);
159 c->max_devices = ceph_decode_32(p);
161 c->device_parents = kcalloc(c->max_devices, sizeof(u32), GFP_NOFS);
162 if (c->device_parents == NULL)
164 c->bucket_parents = kcalloc(c->max_buckets, sizeof(u32), GFP_NOFS);
165 if (c->bucket_parents == NULL)
168 c->buckets = kcalloc(c->max_buckets, sizeof(*c->buckets), GFP_NOFS);
169 if (c->buckets == NULL)
171 c->rules = kcalloc(c->max_rules, sizeof(*c->rules), GFP_NOFS);
172 if (c->rules == NULL)
176 for (i = 0; i < c->max_buckets; i++) {
179 struct crush_bucket *b;
181 ceph_decode_32_safe(p, end, alg, bad);
183 c->buckets[i] = NULL;
186 dout("crush_decode bucket %d off %x %p to %p\n",
187 i, (int)(*p-start), *p, end);
190 case CRUSH_BUCKET_UNIFORM:
191 size = sizeof(struct crush_bucket_uniform);
193 case CRUSH_BUCKET_LIST:
194 size = sizeof(struct crush_bucket_list);
196 case CRUSH_BUCKET_TREE:
197 size = sizeof(struct crush_bucket_tree);
199 case CRUSH_BUCKET_STRAW:
200 size = sizeof(struct crush_bucket_straw);
207 b = c->buckets[i] = kzalloc(size, GFP_NOFS);
211 ceph_decode_need(p, end, 4*sizeof(u32), bad);
212 b->id = ceph_decode_32(p);
213 b->type = ceph_decode_16(p);
214 b->alg = ceph_decode_8(p);
215 b->hash = ceph_decode_8(p);
216 b->weight = ceph_decode_32(p);
217 b->size = ceph_decode_32(p);
219 dout("crush_decode bucket size %d off %x %p to %p\n",
220 b->size, (int)(*p-start), *p, end);
222 b->items = kcalloc(b->size, sizeof(__s32), GFP_NOFS);
223 if (b->items == NULL)
225 b->perm = kcalloc(b->size, sizeof(u32), GFP_NOFS);
230 ceph_decode_need(p, end, b->size*sizeof(u32), bad);
231 for (j = 0; j < b->size; j++)
232 b->items[j] = ceph_decode_32(p);
235 case CRUSH_BUCKET_UNIFORM:
236 err = crush_decode_uniform_bucket(p, end,
237 (struct crush_bucket_uniform *)b);
241 case CRUSH_BUCKET_LIST:
242 err = crush_decode_list_bucket(p, end,
243 (struct crush_bucket_list *)b);
247 case CRUSH_BUCKET_TREE:
248 err = crush_decode_tree_bucket(p, end,
249 (struct crush_bucket_tree *)b);
253 case CRUSH_BUCKET_STRAW:
254 err = crush_decode_straw_bucket(p, end,
255 (struct crush_bucket_straw *)b);
263 dout("rule vec is %p\n", c->rules);
264 for (i = 0; i < c->max_rules; i++) {
266 struct crush_rule *r;
268 ceph_decode_32_safe(p, end, yes, bad);
270 dout("crush_decode NO rule %d off %x %p to %p\n",
271 i, (int)(*p-start), *p, end);
276 dout("crush_decode rule %d off %x %p to %p\n",
277 i, (int)(*p-start), *p, end);
280 ceph_decode_32_safe(p, end, yes, bad);
281 #if BITS_PER_LONG == 32
283 if (yes > ULONG_MAX / sizeof(struct crush_rule_step))
286 r = c->rules[i] = kmalloc(sizeof(*r) +
287 yes*sizeof(struct crush_rule_step),
291 dout(" rule %d is at %p\n", i, r);
293 ceph_decode_copy_safe(p, end, &r->mask, 4, bad); /* 4 u8's */
294 ceph_decode_need(p, end, r->len*3*sizeof(u32), bad);
295 for (j = 0; j < r->len; j++) {
296 r->steps[j].op = ceph_decode_32(p);
297 r->steps[j].arg1 = ceph_decode_32(p);
298 r->steps[j].arg2 = ceph_decode_32(p);
302 /* ignore trailing name maps. */
304 dout("crush_decode success\n");
310 dout("crush_decode fail %d\n", err);
316 * rbtree of pg_mapping for handling pg_temp (explicit mapping of pgid
319 static int pgid_cmp(struct ceph_pg l, struct ceph_pg r)
331 static int __insert_pg_mapping(struct ceph_pg_mapping *new,
332 struct rb_root *root)
334 struct rb_node **p = &root->rb_node;
335 struct rb_node *parent = NULL;
336 struct ceph_pg_mapping *pg = NULL;
341 pg = rb_entry(parent, struct ceph_pg_mapping, node);
342 c = pgid_cmp(new->pgid, pg->pgid);
351 rb_link_node(&new->node, parent, p);
352 rb_insert_color(&new->node, root);
356 static struct ceph_pg_mapping *__lookup_pg_mapping(struct rb_root *root,
359 struct rb_node *n = root->rb_node;
360 struct ceph_pg_mapping *pg;
364 pg = rb_entry(n, struct ceph_pg_mapping, node);
365 c = pgid_cmp(pgid, pg->pgid);
377 * rbtree of pg pool info
379 static int __insert_pg_pool(struct rb_root *root, struct ceph_pg_pool_info *new)
381 struct rb_node **p = &root->rb_node;
382 struct rb_node *parent = NULL;
383 struct ceph_pg_pool_info *pi = NULL;
387 pi = rb_entry(parent, struct ceph_pg_pool_info, node);
388 if (new->id < pi->id)
390 else if (new->id > pi->id)
396 rb_link_node(&new->node, parent, p);
397 rb_insert_color(&new->node, root);
401 static struct ceph_pg_pool_info *__lookup_pg_pool(struct rb_root *root, int id)
403 struct ceph_pg_pool_info *pi;
404 struct rb_node *n = root->rb_node;
407 pi = rb_entry(n, struct ceph_pg_pool_info, node);
410 else if (id > pi->id)
418 static void __remove_pg_pool(struct rb_root *root, struct ceph_pg_pool_info *pi)
420 rb_erase(&pi->node, root);
425 void __decode_pool(void **p, struct ceph_pg_pool_info *pi)
427 ceph_decode_copy(p, &pi->v, sizeof(pi->v));
429 *p += le32_to_cpu(pi->v.num_snaps) * sizeof(u64);
430 *p += le32_to_cpu(pi->v.num_removed_snap_intervals) * sizeof(u64) * 2;
433 static int __decode_pool_names(void **p, void *end, struct ceph_osdmap *map)
435 struct ceph_pg_pool_info *pi;
438 ceph_decode_32_safe(p, end, num, bad);
439 dout(" %d pool names\n", num);
441 ceph_decode_32_safe(p, end, pool, bad);
442 ceph_decode_32_safe(p, end, len, bad);
443 dout(" pool %d len %d\n", pool, len);
444 pi = __lookup_pg_pool(&map->pg_pools, pool);
447 pi->name = kmalloc(len + 1, GFP_NOFS);
449 memcpy(pi->name, *p, len);
450 pi->name[len] = '\0';
451 dout(" name is %s\n", pi->name);
465 void ceph_osdmap_destroy(struct ceph_osdmap *map)
467 dout("osdmap_destroy %p\n", map);
469 crush_destroy(map->crush);
470 while (!RB_EMPTY_ROOT(&map->pg_temp)) {
471 struct ceph_pg_mapping *pg =
472 rb_entry(rb_first(&map->pg_temp),
473 struct ceph_pg_mapping, node);
474 rb_erase(&pg->node, &map->pg_temp);
477 while (!RB_EMPTY_ROOT(&map->pg_pools)) {
478 struct ceph_pg_pool_info *pi =
479 rb_entry(rb_first(&map->pg_pools),
480 struct ceph_pg_pool_info, node);
481 __remove_pg_pool(&map->pg_pools, pi);
483 kfree(map->osd_state);
484 kfree(map->osd_weight);
485 kfree(map->osd_addr);
490 * adjust max osd value. reallocate arrays.
492 static int osdmap_set_max_osd(struct ceph_osdmap *map, int max)
495 struct ceph_entity_addr *addr;
498 state = kcalloc(max, sizeof(*state), GFP_NOFS);
499 addr = kcalloc(max, sizeof(*addr), GFP_NOFS);
500 weight = kcalloc(max, sizeof(*weight), GFP_NOFS);
501 if (state == NULL || addr == NULL || weight == NULL) {
509 if (map->osd_state) {
510 memcpy(state, map->osd_state, map->max_osd*sizeof(*state));
511 memcpy(addr, map->osd_addr, map->max_osd*sizeof(*addr));
512 memcpy(weight, map->osd_weight, map->max_osd*sizeof(*weight));
513 kfree(map->osd_state);
514 kfree(map->osd_addr);
515 kfree(map->osd_weight);
518 map->osd_state = state;
519 map->osd_weight = weight;
520 map->osd_addr = addr;
528 struct ceph_osdmap *osdmap_decode(void **p, void *end)
530 struct ceph_osdmap *map;
536 struct ceph_pg_pool_info *pi;
538 dout("osdmap_decode %p to %p len %d\n", *p, end, (int)(end - *p));
540 map = kzalloc(sizeof(*map), GFP_NOFS);
542 return ERR_PTR(-ENOMEM);
543 map->pg_temp = RB_ROOT;
545 ceph_decode_16_safe(p, end, version, bad);
546 if (version > CEPH_OSDMAP_VERSION) {
547 pr_warning("got unknown v %d > %d of osdmap\n", version,
548 CEPH_OSDMAP_VERSION);
552 ceph_decode_need(p, end, 2*sizeof(u64)+6*sizeof(u32), bad);
553 ceph_decode_copy(p, &map->fsid, sizeof(map->fsid));
554 map->epoch = ceph_decode_32(p);
555 ceph_decode_copy(p, &map->created, sizeof(map->created));
556 ceph_decode_copy(p, &map->modified, sizeof(map->modified));
558 ceph_decode_32_safe(p, end, max, bad);
560 ceph_decode_need(p, end, 4 + 1 + sizeof(pi->v), bad);
561 pi = kzalloc(sizeof(*pi), GFP_NOFS);
564 pi->id = ceph_decode_32(p);
565 ev = ceph_decode_8(p); /* encoding version */
566 if (ev > CEPH_PG_POOL_VERSION) {
567 pr_warning("got unknown v %d > %d of ceph_pg_pool\n",
568 ev, CEPH_PG_POOL_VERSION);
571 __decode_pool(p, pi);
572 __insert_pg_pool(&map->pg_pools, pi);
575 if (version >= 5 && __decode_pool_names(p, end, map) < 0)
578 ceph_decode_32_safe(p, end, map->pool_max, bad);
580 ceph_decode_32_safe(p, end, map->flags, bad);
582 max = ceph_decode_32(p);
584 /* (re)alloc osd arrays */
585 err = osdmap_set_max_osd(map, max);
588 dout("osdmap_decode max_osd = %d\n", map->max_osd);
592 ceph_decode_need(p, end, 3*sizeof(u32) +
593 map->max_osd*(1 + sizeof(*map->osd_weight) +
594 sizeof(*map->osd_addr)), bad);
595 *p += 4; /* skip length field (should match max) */
596 ceph_decode_copy(p, map->osd_state, map->max_osd);
598 *p += 4; /* skip length field (should match max) */
599 for (i = 0; i < map->max_osd; i++)
600 map->osd_weight[i] = ceph_decode_32(p);
602 *p += 4; /* skip length field (should match max) */
603 ceph_decode_copy(p, map->osd_addr, map->max_osd*sizeof(*map->osd_addr));
604 for (i = 0; i < map->max_osd; i++)
605 ceph_decode_addr(&map->osd_addr[i]);
608 ceph_decode_32_safe(p, end, len, bad);
609 for (i = 0; i < len; i++) {
612 struct ceph_pg_mapping *pg;
614 ceph_decode_need(p, end, sizeof(u32) + sizeof(u64), bad);
615 ceph_decode_copy(p, &pgid, sizeof(pgid));
616 n = ceph_decode_32(p);
617 ceph_decode_need(p, end, n * sizeof(u32), bad);
619 pg = kmalloc(sizeof(*pg) + n*sizeof(u32), GFP_NOFS);
624 for (j = 0; j < n; j++)
625 pg->osds[j] = ceph_decode_32(p);
627 err = __insert_pg_mapping(pg, &map->pg_temp);
630 dout(" added pg_temp %llx len %d\n", *(u64 *)&pgid, len);
634 ceph_decode_32_safe(p, end, len, bad);
635 dout("osdmap_decode crush len %d from off 0x%x\n", len,
637 ceph_decode_need(p, end, len, bad);
638 map->crush = crush_decode(*p, end);
640 if (IS_ERR(map->crush)) {
641 err = PTR_ERR(map->crush);
646 /* ignore the rest of the map */
649 dout("osdmap_decode done %p %p\n", *p, end);
653 dout("osdmap_decode fail\n");
654 ceph_osdmap_destroy(map);
659 * decode and apply an incremental map update.
661 struct ceph_osdmap *osdmap_apply_incremental(void **p, void *end,
662 struct ceph_osdmap *map,
663 struct ceph_messenger *msgr)
665 struct crush_map *newcrush = NULL;
666 struct ceph_fsid fsid;
668 struct ceph_timespec modified;
670 __s32 new_pool_max, new_flags, max;
676 ceph_decode_16_safe(p, end, version, bad);
677 if (version > CEPH_OSDMAP_INC_VERSION) {
678 pr_warning("got unknown v %d > %d of inc osdmap\n", version,
679 CEPH_OSDMAP_INC_VERSION);
683 ceph_decode_need(p, end, sizeof(fsid)+sizeof(modified)+2*sizeof(u32),
685 ceph_decode_copy(p, &fsid, sizeof(fsid));
686 epoch = ceph_decode_32(p);
687 BUG_ON(epoch != map->epoch+1);
688 ceph_decode_copy(p, &modified, sizeof(modified));
689 new_pool_max = ceph_decode_32(p);
690 new_flags = ceph_decode_32(p);
693 ceph_decode_32_safe(p, end, len, bad);
695 dout("apply_incremental full map len %d, %p to %p\n",
697 return osdmap_decode(p, min(*p+len, end));
701 ceph_decode_32_safe(p, end, len, bad);
703 dout("apply_incremental new crush map len %d, %p to %p\n",
705 newcrush = crush_decode(*p, min(*p+len, end));
706 if (IS_ERR(newcrush))
707 return ERR_PTR(PTR_ERR(newcrush));
712 map->flags = new_flags;
713 if (new_pool_max >= 0)
714 map->pool_max = new_pool_max;
716 ceph_decode_need(p, end, 5*sizeof(u32), bad);
719 max = ceph_decode_32(p);
721 err = osdmap_set_max_osd(map, max);
727 map->modified = map->modified;
730 crush_destroy(map->crush);
731 map->crush = newcrush;
736 ceph_decode_32_safe(p, end, len, bad);
739 struct ceph_pg_pool_info *pi;
741 ceph_decode_32_safe(p, end, pool, bad);
742 ceph_decode_need(p, end, 1 + sizeof(pi->v), bad);
743 ev = ceph_decode_8(p); /* encoding version */
744 if (ev > CEPH_PG_POOL_VERSION) {
745 pr_warning("got unknown v %d > %d of ceph_pg_pool\n",
746 ev, CEPH_PG_POOL_VERSION);
749 pi = __lookup_pg_pool(&map->pg_pools, pool);
751 pi = kzalloc(sizeof(*pi), GFP_NOFS);
757 __insert_pg_pool(&map->pg_pools, pi);
759 __decode_pool(p, pi);
761 if (version >= 5 && __decode_pool_names(p, end, map) < 0)
765 ceph_decode_32_safe(p, end, len, bad);
767 struct ceph_pg_pool_info *pi;
769 ceph_decode_32_safe(p, end, pool, bad);
770 pi = __lookup_pg_pool(&map->pg_pools, pool);
772 __remove_pg_pool(&map->pg_pools, pi);
777 ceph_decode_32_safe(p, end, len, bad);
780 struct ceph_entity_addr addr;
781 ceph_decode_32_safe(p, end, osd, bad);
782 ceph_decode_copy_safe(p, end, &addr, sizeof(addr), bad);
783 ceph_decode_addr(&addr);
784 pr_info("osd%d up\n", osd);
785 BUG_ON(osd >= map->max_osd);
786 map->osd_state[osd] |= CEPH_OSD_UP;
787 map->osd_addr[osd] = addr;
791 ceph_decode_32_safe(p, end, len, bad);
794 ceph_decode_32_safe(p, end, osd, bad);
795 (*p)++; /* clean flag */
796 pr_info("osd%d down\n", osd);
797 if (osd < map->max_osd)
798 map->osd_state[osd] &= ~CEPH_OSD_UP;
802 ceph_decode_32_safe(p, end, len, bad);
805 ceph_decode_need(p, end, sizeof(u32)*2, bad);
806 osd = ceph_decode_32(p);
807 off = ceph_decode_32(p);
808 pr_info("osd%d weight 0x%x %s\n", osd, off,
809 off == CEPH_OSD_IN ? "(in)" :
810 (off == CEPH_OSD_OUT ? "(out)" : ""));
811 if (osd < map->max_osd)
812 map->osd_weight[osd] = off;
816 rbp = rb_first(&map->pg_temp);
817 ceph_decode_32_safe(p, end, len, bad);
819 struct ceph_pg_mapping *pg;
823 ceph_decode_need(p, end, sizeof(u64) + sizeof(u32), bad);
824 ceph_decode_copy(p, &pgid, sizeof(pgid));
825 pglen = ceph_decode_32(p);
828 while (rbp && pgid_cmp(rb_entry(rbp, struct ceph_pg_mapping,
829 node)->pgid, pgid) <= 0) {
830 struct rb_node *cur = rbp;
832 dout(" removed pg_temp %llx\n",
833 *(u64 *)&rb_entry(cur, struct ceph_pg_mapping,
835 rb_erase(cur, &map->pg_temp);
840 ceph_decode_need(p, end, pglen*sizeof(u32), bad);
841 pg = kmalloc(sizeof(*pg) + sizeof(u32)*pglen, GFP_NOFS);
848 for (j = 0; j < pglen; j++)
849 pg->osds[j] = ceph_decode_32(p);
850 err = __insert_pg_mapping(pg, &map->pg_temp);
853 dout(" added pg_temp %llx len %d\n", *(u64 *)&pgid,
858 struct rb_node *cur = rbp;
860 dout(" removed pg_temp %llx\n",
861 *(u64 *)&rb_entry(cur, struct ceph_pg_mapping,
863 rb_erase(cur, &map->pg_temp);
866 /* ignore the rest */
871 pr_err("corrupt inc osdmap epoch %d off %d (%p of %p-%p)\n",
872 epoch, (int)(*p - start), *p, start, end);
873 print_hex_dump(KERN_DEBUG, "osdmap: ",
874 DUMP_PREFIX_OFFSET, 16, 1,
875 start, end - start, true);
877 crush_destroy(newcrush);
885 * calculate file layout from given offset, length.
886 * fill in correct oid, logical length, and object extent
889 * for now, we write only a single su, until we can
890 * pass a stride back to the caller.
892 void ceph_calc_file_object_mapping(struct ceph_file_layout *layout,
895 u64 *oxoff, u64 *oxlen)
897 u32 osize = le32_to_cpu(layout->fl_object_size);
898 u32 su = le32_to_cpu(layout->fl_stripe_unit);
899 u32 sc = le32_to_cpu(layout->fl_stripe_count);
900 u32 bl, stripeno, stripepos, objsetno;
904 dout("mapping %llu~%llu osize %u fl_su %u\n", off, *plen,
906 su_per_object = osize / su;
907 dout("osize %u / su %u = su_per_object %u\n", osize, su,
910 BUG_ON((su & ~PAGE_MASK) != 0);
911 /* bl = *off / su; */
915 dout("off %llu / su %u = bl %u\n", off, su, bl);
919 objsetno = stripeno / su_per_object;
921 *ono = objsetno * sc + stripepos;
922 dout("objset %u * sc %u = ono %u\n", objsetno, sc, (unsigned)*ono);
924 /* *oxoff = *off % layout->fl_stripe_unit; # offset in su */
926 su_offset = do_div(t, su);
927 *oxoff = su_offset + (stripeno % su_per_object) * su;
930 * Calculate the length of the extent being written to the selected
931 * object. This is the minimum of the full length requested (plen) or
932 * the remainder of the current stripe being written to.
934 *oxlen = min_t(u64, *plen, su - su_offset);
937 dout(" obj extent %llu~%llu\n", *oxoff, *oxlen);
941 * calculate an object layout (i.e. pgid) from an oid,
942 * file_layout, and osdmap
944 int ceph_calc_object_layout(struct ceph_object_layout *ol,
946 struct ceph_file_layout *fl,
947 struct ceph_osdmap *osdmap)
949 unsigned num, num_mask;
951 s32 preferred = (s32)le32_to_cpu(fl->fl_pg_preferred);
952 int poolid = le32_to_cpu(fl->fl_pg_pool);
953 struct ceph_pg_pool_info *pool;
958 pool = __lookup_pg_pool(&osdmap->pg_pools, poolid);
961 ps = ceph_str_hash(pool->v.object_hash, oid, strlen(oid));
962 if (preferred >= 0) {
964 num = le32_to_cpu(pool->v.lpg_num);
965 num_mask = pool->lpg_num_mask;
967 num = le32_to_cpu(pool->v.pg_num);
968 num_mask = pool->pg_num_mask;
971 pgid.ps = cpu_to_le16(ps);
972 pgid.preferred = cpu_to_le16(preferred);
973 pgid.pool = fl->fl_pg_pool;
975 dout("calc_object_layout '%s' pgid %d.%xp%d\n", oid, poolid, ps,
978 dout("calc_object_layout '%s' pgid %d.%x\n", oid, poolid, ps);
981 ol->ol_stripe_unit = fl->fl_object_stripe_unit;
986 * Calculate raw osd vector for the given pgid. Return pointer to osd
987 * array, or NULL on failure.
989 static int *calc_pg_raw(struct ceph_osdmap *osdmap, struct ceph_pg pgid,
992 struct ceph_pg_mapping *pg;
993 struct ceph_pg_pool_info *pool;
995 unsigned poolid, ps, pps;
999 pg = __lookup_pg_mapping(&osdmap->pg_temp, pgid);
1006 poolid = le32_to_cpu(pgid.pool);
1007 ps = le16_to_cpu(pgid.ps);
1008 preferred = (s16)le16_to_cpu(pgid.preferred);
1010 /* don't forcefeed bad device ids to crush */
1011 if (preferred >= osdmap->max_osd ||
1012 preferred >= osdmap->crush->max_devices)
1015 pool = __lookup_pg_pool(&osdmap->pg_pools, poolid);
1018 ruleno = crush_find_rule(osdmap->crush, pool->v.crush_ruleset,
1019 pool->v.type, pool->v.size);
1021 pr_err("no crush rule pool %d type %d size %d\n",
1022 poolid, pool->v.type, pool->v.size);
1027 pps = ceph_stable_mod(ps,
1028 le32_to_cpu(pool->v.lpgp_num),
1029 pool->lpgp_num_mask);
1031 pps = ceph_stable_mod(ps,
1032 le32_to_cpu(pool->v.pgp_num),
1033 pool->pgp_num_mask);
1035 *num = crush_do_rule(osdmap->crush, ruleno, pps, osds,
1036 min_t(int, pool->v.size, *num),
1037 preferred, osdmap->osd_weight);
1042 * Return primary osd for given pgid, or -1 if none.
1044 int ceph_calc_pg_primary(struct ceph_osdmap *osdmap, struct ceph_pg pgid)
1046 int rawosds[10], *osds;
1047 int i, num = ARRAY_SIZE(rawosds);
1049 osds = calc_pg_raw(osdmap, pgid, rawosds, &num);
1053 /* primary is first up osd */
1054 for (i = 0; i < num; i++)
1055 if (ceph_osd_is_up(osdmap, osds[i])) {