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);
206 b = c->buckets[i] = kzalloc(size, GFP_NOFS);
210 ceph_decode_need(p, end, 4*sizeof(u32), bad);
211 b->id = ceph_decode_32(p);
212 b->type = ceph_decode_16(p);
213 b->alg = ceph_decode_8(p);
214 b->hash = ceph_decode_8(p);
215 b->weight = ceph_decode_32(p);
216 b->size = ceph_decode_32(p);
218 dout("crush_decode bucket size %d off %x %p to %p\n",
219 b->size, (int)(*p-start), *p, end);
221 b->items = kcalloc(b->size, sizeof(__s32), GFP_NOFS);
222 if (b->items == NULL)
224 b->perm = kcalloc(b->size, sizeof(u32), GFP_NOFS);
229 ceph_decode_need(p, end, b->size*sizeof(u32), bad);
230 for (j = 0; j < b->size; j++)
231 b->items[j] = ceph_decode_32(p);
234 case CRUSH_BUCKET_UNIFORM:
235 err = crush_decode_uniform_bucket(p, end,
236 (struct crush_bucket_uniform *)b);
240 case CRUSH_BUCKET_LIST:
241 err = crush_decode_list_bucket(p, end,
242 (struct crush_bucket_list *)b);
246 case CRUSH_BUCKET_TREE:
247 err = crush_decode_tree_bucket(p, end,
248 (struct crush_bucket_tree *)b);
252 case CRUSH_BUCKET_STRAW:
253 err = crush_decode_straw_bucket(p, end,
254 (struct crush_bucket_straw *)b);
262 dout("rule vec is %p\n", c->rules);
263 for (i = 0; i < c->max_rules; i++) {
265 struct crush_rule *r;
267 ceph_decode_32_safe(p, end, yes, bad);
269 dout("crush_decode NO rule %d off %x %p to %p\n",
270 i, (int)(*p-start), *p, end);
275 dout("crush_decode rule %d off %x %p to %p\n",
276 i, (int)(*p-start), *p, end);
279 ceph_decode_32_safe(p, end, yes, bad);
280 #if BITS_PER_LONG == 32
281 if (yes > ULONG_MAX / sizeof(struct crush_rule_step))
284 r = c->rules[i] = kmalloc(sizeof(*r) +
285 yes*sizeof(struct crush_rule_step),
289 dout(" rule %d is at %p\n", i, r);
291 ceph_decode_copy_safe(p, end, &r->mask, 4, bad); /* 4 u8's */
292 ceph_decode_need(p, end, r->len*3*sizeof(u32), bad);
293 for (j = 0; j < r->len; j++) {
294 r->steps[j].op = ceph_decode_32(p);
295 r->steps[j].arg1 = ceph_decode_32(p);
296 r->steps[j].arg2 = ceph_decode_32(p);
300 /* ignore trailing name maps. */
302 dout("crush_decode success\n");
308 dout("crush_decode fail %d\n", err);
317 void ceph_osdmap_destroy(struct ceph_osdmap *map)
319 dout("osdmap_destroy %p\n", map);
321 crush_destroy(map->crush);
322 while (!RB_EMPTY_ROOT(&map->pg_temp))
323 rb_erase(rb_first(&map->pg_temp), &map->pg_temp);
324 kfree(map->osd_state);
325 kfree(map->osd_weight);
327 kfree(map->osd_addr);
332 * adjust max osd value. reallocate arrays.
334 static int osdmap_set_max_osd(struct ceph_osdmap *map, int max)
337 struct ceph_entity_addr *addr;
340 state = kcalloc(max, sizeof(*state), GFP_NOFS);
341 addr = kcalloc(max, sizeof(*addr), GFP_NOFS);
342 weight = kcalloc(max, sizeof(*weight), GFP_NOFS);
343 if (state == NULL || addr == NULL || weight == NULL) {
351 if (map->osd_state) {
352 memcpy(state, map->osd_state, map->max_osd*sizeof(*state));
353 memcpy(addr, map->osd_addr, map->max_osd*sizeof(*addr));
354 memcpy(weight, map->osd_weight, map->max_osd*sizeof(*weight));
355 kfree(map->osd_state);
356 kfree(map->osd_addr);
357 kfree(map->osd_weight);
360 map->osd_state = state;
361 map->osd_weight = weight;
362 map->osd_addr = addr;
368 * Insert a new pg_temp mapping
370 static int pgid_cmp(struct ceph_pg l, struct ceph_pg r)
382 static int __insert_pg_mapping(struct ceph_pg_mapping *new,
383 struct rb_root *root)
385 struct rb_node **p = &root->rb_node;
386 struct rb_node *parent = NULL;
387 struct ceph_pg_mapping *pg = NULL;
392 pg = rb_entry(parent, struct ceph_pg_mapping, node);
393 c = pgid_cmp(new->pgid, pg->pgid);
402 rb_link_node(&new->node, parent, p);
403 rb_insert_color(&new->node, root);
410 struct ceph_osdmap *osdmap_decode(void **p, void *end)
412 struct ceph_osdmap *map;
418 dout("osdmap_decode %p to %p len %d\n", *p, end, (int)(end - *p));
420 map = kzalloc(sizeof(*map), GFP_NOFS);
422 return ERR_PTR(-ENOMEM);
423 map->pg_temp = RB_ROOT;
425 ceph_decode_16_safe(p, end, version, bad);
427 ceph_decode_need(p, end, 2*sizeof(u64)+6*sizeof(u32), bad);
428 ceph_decode_copy(p, &map->fsid, sizeof(map->fsid));
429 map->epoch = ceph_decode_32(p);
430 ceph_decode_copy(p, &map->created, sizeof(map->created));
431 ceph_decode_copy(p, &map->modified, sizeof(map->modified));
433 map->num_pools = ceph_decode_32(p);
434 map->pg_pool = kcalloc(map->num_pools, sizeof(*map->pg_pool),
440 ceph_decode_32_safe(p, end, max, bad);
442 ceph_decode_need(p, end, 4+sizeof(map->pg_pool->v), bad);
443 i = ceph_decode_32(p);
444 if (i >= map->num_pools)
446 ceph_decode_copy(p, &map->pg_pool[i].v,
447 sizeof(map->pg_pool->v));
448 calc_pg_masks(&map->pg_pool[i]);
449 p += le32_to_cpu(map->pg_pool[i].v.num_snaps) * sizeof(u64);
450 p += le32_to_cpu(map->pg_pool[i].v.num_removed_snap_intervals)
454 ceph_decode_32_safe(p, end, map->flags, bad);
456 max = ceph_decode_32(p);
458 /* (re)alloc osd arrays */
459 err = osdmap_set_max_osd(map, max);
462 dout("osdmap_decode max_osd = %d\n", map->max_osd);
466 ceph_decode_need(p, end, 3*sizeof(u32) +
467 map->max_osd*(1 + sizeof(*map->osd_weight) +
468 sizeof(*map->osd_addr)), bad);
469 *p += 4; /* skip length field (should match max) */
470 ceph_decode_copy(p, map->osd_state, map->max_osd);
472 *p += 4; /* skip length field (should match max) */
473 for (i = 0; i < map->max_osd; i++)
474 map->osd_weight[i] = ceph_decode_32(p);
476 *p += 4; /* skip length field (should match max) */
477 ceph_decode_copy(p, map->osd_addr, map->max_osd*sizeof(*map->osd_addr));
478 for (i = 0; i < map->max_osd; i++)
479 ceph_decode_addr(&map->osd_addr[i]);
482 ceph_decode_32_safe(p, end, len, bad);
483 for (i = 0; i < len; i++) {
486 struct ceph_pg_mapping *pg;
488 ceph_decode_need(p, end, sizeof(u32) + sizeof(u64), bad);
489 ceph_decode_copy(p, &pgid, sizeof(pgid));
490 n = ceph_decode_32(p);
491 ceph_decode_need(p, end, n * sizeof(u32), bad);
492 pg = kmalloc(sizeof(*pg) + n*sizeof(u32), GFP_NOFS);
499 for (j = 0; j < n; j++)
500 pg->osds[j] = ceph_decode_32(p);
502 err = __insert_pg_mapping(pg, &map->pg_temp);
505 dout(" added pg_temp %llx len %d\n", *(u64 *)&pgid, len);
509 ceph_decode_32_safe(p, end, len, bad);
510 dout("osdmap_decode crush len %d from off 0x%x\n", len,
512 ceph_decode_need(p, end, len, bad);
513 map->crush = crush_decode(*p, end);
515 if (IS_ERR(map->crush)) {
516 err = PTR_ERR(map->crush);
521 /* ignore the rest of the map */
524 dout("osdmap_decode done %p %p\n", *p, end);
528 dout("osdmap_decode fail\n");
529 ceph_osdmap_destroy(map);
534 * decode and apply an incremental map update.
536 struct ceph_osdmap *osdmap_apply_incremental(void **p, void *end,
537 struct ceph_osdmap *map,
538 struct ceph_messenger *msgr)
540 struct ceph_osdmap *newmap = map;
541 struct crush_map *newcrush = NULL;
542 struct ceph_fsid fsid;
544 struct ceph_timespec modified;
546 __s32 new_flags, max;
552 ceph_decode_16_safe(p, end, version, bad);
554 ceph_decode_need(p, end, sizeof(fsid)+sizeof(modified)+2*sizeof(u32),
556 ceph_decode_copy(p, &fsid, sizeof(fsid));
557 epoch = ceph_decode_32(p);
558 BUG_ON(epoch != map->epoch+1);
559 ceph_decode_copy(p, &modified, sizeof(modified));
560 new_flags = ceph_decode_32(p);
563 ceph_decode_32_safe(p, end, len, bad);
565 dout("apply_incremental full map len %d, %p to %p\n",
567 newmap = osdmap_decode(p, min(*p+len, end));
568 return newmap; /* error or not */
572 ceph_decode_32_safe(p, end, len, bad);
574 dout("apply_incremental new crush map len %d, %p to %p\n",
576 newcrush = crush_decode(*p, min(*p+len, end));
577 if (IS_ERR(newcrush))
578 return ERR_PTR(PTR_ERR(newcrush));
583 map->flags = new_flags;
585 ceph_decode_need(p, end, 5*sizeof(u32), bad);
588 max = ceph_decode_32(p);
590 err = osdmap_set_max_osd(map, max);
596 map->modified = map->modified;
599 crush_destroy(map->crush);
600 map->crush = newcrush;
605 ceph_decode_32_safe(p, end, len, bad);
607 ceph_decode_32_safe(p, end, pool, bad);
608 if (pool >= map->num_pools) {
609 void *pg_pool = kcalloc(pool + 1,
610 sizeof(*map->pg_pool),
616 memcpy(pg_pool, map->pg_pool,
617 map->num_pools * sizeof(*map->pg_pool));
619 map->pg_pool = pg_pool;
620 map->num_pools = pool+1;
622 ceph_decode_copy(p, &map->pg_pool[pool].v,
623 sizeof(map->pg_pool->v));
624 calc_pg_masks(&map->pg_pool[pool]);
627 /* old_pool (ignore) */
628 ceph_decode_32_safe(p, end, len, bad);
629 *p += len * sizeof(u32);
633 ceph_decode_32_safe(p, end, len, bad);
636 struct ceph_entity_addr addr;
637 ceph_decode_32_safe(p, end, osd, bad);
638 ceph_decode_copy_safe(p, end, &addr, sizeof(addr), bad);
639 ceph_decode_addr(&addr);
640 pr_info("osd%d up\n", osd);
641 BUG_ON(osd >= map->max_osd);
642 map->osd_state[osd] |= CEPH_OSD_UP;
643 map->osd_addr[osd] = addr;
647 ceph_decode_32_safe(p, end, len, bad);
650 ceph_decode_32_safe(p, end, osd, bad);
651 (*p)++; /* clean flag */
652 pr_info("osd%d down\n", osd);
653 if (osd < map->max_osd)
654 map->osd_state[osd] &= ~CEPH_OSD_UP;
658 ceph_decode_32_safe(p, end, len, bad);
661 ceph_decode_need(p, end, sizeof(u32)*2, bad);
662 osd = ceph_decode_32(p);
663 off = ceph_decode_32(p);
664 pr_info("osd%d weight 0x%x %s\n", osd, off,
665 off == CEPH_OSD_IN ? "(in)" :
666 (off == CEPH_OSD_OUT ? "(out)" : ""));
667 if (osd < map->max_osd)
668 map->osd_weight[osd] = off;
672 rbp = rb_first(&map->pg_temp);
673 ceph_decode_32_safe(p, end, len, bad);
675 struct ceph_pg_mapping *pg;
679 ceph_decode_need(p, end, sizeof(u64) + sizeof(u32), bad);
680 ceph_decode_copy(p, &pgid, sizeof(pgid));
681 pglen = ceph_decode_32(p);
684 while (rbp && pgid_cmp(rb_entry(rbp, struct ceph_pg_mapping,
685 node)->pgid, pgid) <= 0) {
686 struct rb_node *cur = rbp;
688 dout(" removed pg_temp %llx\n",
689 *(u64 *)&rb_entry(cur, struct ceph_pg_mapping,
691 rb_erase(cur, &map->pg_temp);
696 ceph_decode_need(p, end, pglen*sizeof(u32), bad);
697 pg = kmalloc(sizeof(*pg) + sizeof(u32)*pglen, GFP_NOFS);
704 for (j = 0; j < len; j++)
705 pg->osds[j] = ceph_decode_32(p);
706 err = __insert_pg_mapping(pg, &map->pg_temp);
709 dout(" added pg_temp %llx len %d\n", *(u64 *)&pgid,
714 struct rb_node *cur = rbp;
716 dout(" removed pg_temp %llx\n",
717 *(u64 *)&rb_entry(cur, struct ceph_pg_mapping,
719 rb_erase(cur, &map->pg_temp);
722 /* ignore the rest */
727 pr_err("corrupt inc osdmap epoch %d off %d (%p of %p-%p)\n",
728 epoch, (int)(*p - start), *p, start, end);
729 print_hex_dump(KERN_DEBUG, "osdmap: ",
730 DUMP_PREFIX_OFFSET, 16, 1,
731 start, end - start, true);
733 crush_destroy(newcrush);
741 * calculate file layout from given offset, length.
742 * fill in correct oid, logical length, and object extent
745 * for now, we write only a single su, until we can
746 * pass a stride back to the caller.
748 void ceph_calc_file_object_mapping(struct ceph_file_layout *layout,
751 u64 *oxoff, u64 *oxlen)
753 u32 osize = le32_to_cpu(layout->fl_object_size);
754 u32 su = le32_to_cpu(layout->fl_stripe_unit);
755 u32 sc = le32_to_cpu(layout->fl_stripe_count);
756 u32 bl, stripeno, stripepos, objsetno;
760 dout("mapping %llu~%llu osize %u fl_su %u\n", off, *plen,
762 su_per_object = osize / su;
763 dout("osize %u / su %u = su_per_object %u\n", osize, su,
766 BUG_ON((su & ~PAGE_MASK) != 0);
767 /* bl = *off / su; */
771 dout("off %llu / su %u = bl %u\n", off, su, bl);
775 objsetno = stripeno / su_per_object;
777 *ono = objsetno * sc + stripepos;
778 dout("objset %u * sc %u = ono %u\n", objsetno, sc, (unsigned)*ono);
780 /* *oxoff = *off % layout->fl_stripe_unit; # offset in su */
782 su_offset = do_div(t, su);
783 *oxoff = su_offset + (stripeno % su_per_object) * su;
786 * Calculate the length of the extent being written to the selected
787 * object. This is the minimum of the full length requested (plen) or
788 * the remainder of the current stripe being written to.
790 *oxlen = min_t(u64, *plen, su - su_offset);
793 dout(" obj extent %llu~%llu\n", *oxoff, *oxlen);
797 * calculate an object layout (i.e. pgid) from an oid,
798 * file_layout, and osdmap
800 int ceph_calc_object_layout(struct ceph_object_layout *ol,
802 struct ceph_file_layout *fl,
803 struct ceph_osdmap *osdmap)
805 unsigned num, num_mask;
807 s32 preferred = (s32)le32_to_cpu(fl->fl_pg_preferred);
808 int poolid = le32_to_cpu(fl->fl_pg_pool);
809 struct ceph_pg_pool_info *pool;
812 if (poolid >= osdmap->num_pools)
815 pool = &osdmap->pg_pool[poolid];
816 ps = ceph_str_hash(pool->v.object_hash, oid, strlen(oid));
817 if (preferred >= 0) {
819 num = le32_to_cpu(pool->v.lpg_num);
820 num_mask = pool->lpg_num_mask;
822 num = le32_to_cpu(pool->v.pg_num);
823 num_mask = pool->pg_num_mask;
826 pgid.ps = cpu_to_le16(ps);
827 pgid.preferred = cpu_to_le16(preferred);
828 pgid.pool = fl->fl_pg_pool;
830 dout("calc_object_layout '%s' pgid %d.%xp%d\n", oid, poolid, ps,
833 dout("calc_object_layout '%s' pgid %d.%x\n", oid, poolid, ps);
836 ol->ol_stripe_unit = fl->fl_object_stripe_unit;
841 * Calculate raw osd vector for the given pgid. Return pointer to osd
842 * array, or NULL on failure.
844 static int *calc_pg_raw(struct ceph_osdmap *osdmap, struct ceph_pg pgid,
847 struct rb_node *n = osdmap->pg_temp.rb_node;
848 struct ceph_pg_mapping *pg;
849 struct ceph_pg_pool_info *pool;
851 unsigned poolid, ps, pps;
857 pg = rb_entry(n, struct ceph_pg_mapping, node);
858 c = pgid_cmp(pgid, pg->pgid);
870 poolid = le32_to_cpu(pgid.pool);
871 ps = le16_to_cpu(pgid.ps);
872 preferred = (s16)le16_to_cpu(pgid.preferred);
874 /* don't forcefeed bad device ids to crush */
875 if (preferred >= osdmap->max_osd ||
876 preferred >= osdmap->crush->max_devices)
879 if (poolid >= osdmap->num_pools)
881 pool = &osdmap->pg_pool[poolid];
882 ruleno = crush_find_rule(osdmap->crush, pool->v.crush_ruleset,
883 pool->v.type, pool->v.size);
885 pr_err("no crush rule pool %d type %d size %d\n",
886 poolid, pool->v.type, pool->v.size);
891 pps = ceph_stable_mod(ps,
892 le32_to_cpu(pool->v.lpgp_num),
893 pool->lpgp_num_mask);
895 pps = ceph_stable_mod(ps,
896 le32_to_cpu(pool->v.pgp_num),
899 *num = crush_do_rule(osdmap->crush, ruleno, pps, osds,
900 min_t(int, pool->v.size, *num),
901 preferred, osdmap->osd_weight);
906 * Return primary osd for given pgid, or -1 if none.
908 int ceph_calc_pg_primary(struct ceph_osdmap *osdmap, struct ceph_pg pgid)
910 int rawosds[10], *osds;
911 int i, num = ARRAY_SIZE(rawosds);
913 osds = calc_pg_raw(osdmap, pgid, rawosds, &num);
917 /* primary is first up osd */
918 for (i = 0; i < num; i++)
919 if (ceph_osd_is_up(osdmap, osds[i])) {