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_16(p);
214 b->weight = ceph_decode_32(p);
215 b->size = ceph_decode_32(p);
217 dout("crush_decode bucket size %d off %x %p to %p\n",
218 b->size, (int)(*p-start), *p, end);
220 b->items = kcalloc(b->size, sizeof(__s32), GFP_NOFS);
221 if (b->items == NULL)
223 b->perm = kcalloc(b->size, sizeof(u32), GFP_NOFS);
228 ceph_decode_need(p, end, b->size*sizeof(u32), bad);
229 for (j = 0; j < b->size; j++)
230 b->items[j] = ceph_decode_32(p);
233 case CRUSH_BUCKET_UNIFORM:
234 err = crush_decode_uniform_bucket(p, end,
235 (struct crush_bucket_uniform *)b);
239 case CRUSH_BUCKET_LIST:
240 err = crush_decode_list_bucket(p, end,
241 (struct crush_bucket_list *)b);
245 case CRUSH_BUCKET_TREE:
246 err = crush_decode_tree_bucket(p, end,
247 (struct crush_bucket_tree *)b);
251 case CRUSH_BUCKET_STRAW:
252 err = crush_decode_straw_bucket(p, end,
253 (struct crush_bucket_straw *)b);
261 dout("rule vec is %p\n", c->rules);
262 for (i = 0; i < c->max_rules; i++) {
264 struct crush_rule *r;
266 ceph_decode_32_safe(p, end, yes, bad);
268 dout("crush_decode NO rule %d off %x %p to %p\n",
269 i, (int)(*p-start), *p, end);
274 dout("crush_decode rule %d off %x %p to %p\n",
275 i, (int)(*p-start), *p, end);
278 ceph_decode_32_safe(p, end, yes, bad);
279 #if BITS_PER_LONG == 32
280 if (yes > ULONG_MAX / sizeof(struct crush_rule_step))
283 r = c->rules[i] = kmalloc(sizeof(*r) +
284 yes*sizeof(struct crush_rule_step),
288 dout(" rule %d is at %p\n", i, r);
290 ceph_decode_copy_safe(p, end, &r->mask, 4, bad); /* 4 u8's */
291 ceph_decode_need(p, end, r->len*3*sizeof(u32), bad);
292 for (j = 0; j < r->len; j++) {
293 r->steps[j].op = ceph_decode_32(p);
294 r->steps[j].arg1 = ceph_decode_32(p);
295 r->steps[j].arg2 = ceph_decode_32(p);
299 /* ignore trailing name maps. */
301 dout("crush_decode success\n");
307 dout("crush_decode fail %d\n", err);
316 void ceph_osdmap_destroy(struct ceph_osdmap *map)
318 dout("osdmap_destroy %p\n", map);
320 crush_destroy(map->crush);
321 while (!RB_EMPTY_ROOT(&map->pg_temp))
322 rb_erase(rb_first(&map->pg_temp), &map->pg_temp);
323 kfree(map->osd_state);
324 kfree(map->osd_weight);
326 kfree(map->osd_addr);
331 * adjust max osd value. reallocate arrays.
333 static int osdmap_set_max_osd(struct ceph_osdmap *map, int max)
336 struct ceph_entity_addr *addr;
339 state = kcalloc(max, sizeof(*state), GFP_NOFS);
340 addr = kcalloc(max, sizeof(*addr), GFP_NOFS);
341 weight = kcalloc(max, sizeof(*weight), GFP_NOFS);
342 if (state == NULL || addr == NULL || weight == NULL) {
350 if (map->osd_state) {
351 memcpy(state, map->osd_state, map->max_osd*sizeof(*state));
352 memcpy(addr, map->osd_addr, map->max_osd*sizeof(*addr));
353 memcpy(weight, map->osd_weight, map->max_osd*sizeof(*weight));
354 kfree(map->osd_state);
355 kfree(map->osd_addr);
356 kfree(map->osd_weight);
359 map->osd_state = state;
360 map->osd_weight = weight;
361 map->osd_addr = addr;
367 * Insert a new pg_temp mapping
369 static int __insert_pg_mapping(struct ceph_pg_mapping *new,
370 struct rb_root *root)
372 struct rb_node **p = &root->rb_node;
373 struct rb_node *parent = NULL;
374 struct ceph_pg_mapping *pg = NULL;
378 pg = rb_entry(parent, struct ceph_pg_mapping, node);
379 if (new->pgid < pg->pgid)
381 else if (new->pgid > pg->pgid)
387 rb_link_node(&new->node, parent, p);
388 rb_insert_color(&new->node, root);
395 struct ceph_osdmap *osdmap_decode(void **p, void *end)
397 struct ceph_osdmap *map;
403 dout("osdmap_decode %p to %p len %d\n", *p, end, (int)(end - *p));
405 map = kzalloc(sizeof(*map), GFP_NOFS);
407 return ERR_PTR(-ENOMEM);
408 map->pg_temp = RB_ROOT;
410 ceph_decode_16_safe(p, end, version, bad);
412 ceph_decode_need(p, end, 2*sizeof(u64)+6*sizeof(u32), bad);
413 ceph_decode_copy(p, &map->fsid, sizeof(map->fsid));
414 map->epoch = ceph_decode_32(p);
415 ceph_decode_copy(p, &map->created, sizeof(map->created));
416 ceph_decode_copy(p, &map->modified, sizeof(map->modified));
418 map->num_pools = ceph_decode_32(p);
419 map->pg_pool = kcalloc(map->num_pools, sizeof(*map->pg_pool),
425 ceph_decode_32_safe(p, end, max, bad);
427 ceph_decode_need(p, end, 4+sizeof(map->pg_pool->v), bad);
428 i = ceph_decode_32(p);
429 if (i >= map->num_pools)
431 ceph_decode_copy(p, &map->pg_pool[i].v,
432 sizeof(map->pg_pool->v));
433 calc_pg_masks(&map->pg_pool[i]);
434 p += le32_to_cpu(map->pg_pool[i].v.num_snaps) * sizeof(u64);
435 p += le32_to_cpu(map->pg_pool[i].v.num_removed_snap_intervals)
439 ceph_decode_32_safe(p, end, map->flags, bad);
441 max = ceph_decode_32(p);
443 /* (re)alloc osd arrays */
444 err = osdmap_set_max_osd(map, max);
447 dout("osdmap_decode max_osd = %d\n", map->max_osd);
451 ceph_decode_need(p, end, 3*sizeof(u32) +
452 map->max_osd*(1 + sizeof(*map->osd_weight) +
453 sizeof(*map->osd_addr)), bad);
454 *p += 4; /* skip length field (should match max) */
455 ceph_decode_copy(p, map->osd_state, map->max_osd);
457 *p += 4; /* skip length field (should match max) */
458 for (i = 0; i < map->max_osd; i++)
459 map->osd_weight[i] = ceph_decode_32(p);
461 *p += 4; /* skip length field (should match max) */
462 ceph_decode_copy(p, map->osd_addr, map->max_osd*sizeof(*map->osd_addr));
465 ceph_decode_32_safe(p, end, len, bad);
466 for (i = 0; i < len; i++) {
469 struct ceph_pg_mapping *pg;
471 ceph_decode_need(p, end, sizeof(u32) + sizeof(u64), bad);
472 pgid = ceph_decode_64(p);
473 n = ceph_decode_32(p);
474 ceph_decode_need(p, end, n * sizeof(u32), bad);
475 pg = kmalloc(sizeof(*pg) + n*sizeof(u32), GFP_NOFS);
482 for (j = 0; j < n; j++)
483 pg->osds[j] = ceph_decode_32(p);
485 err = __insert_pg_mapping(pg, &map->pg_temp);
488 dout(" added pg_temp %llx len %d\n", pgid, len);
492 ceph_decode_32_safe(p, end, len, bad);
493 dout("osdmap_decode crush len %d from off 0x%x\n", len,
495 ceph_decode_need(p, end, len, bad);
496 map->crush = crush_decode(*p, end);
498 if (IS_ERR(map->crush)) {
499 err = PTR_ERR(map->crush);
504 /* ignore the rest of the map */
507 dout("osdmap_decode done %p %p\n", *p, end);
511 dout("osdmap_decode fail\n");
512 ceph_osdmap_destroy(map);
517 * decode and apply an incremental map update.
519 struct ceph_osdmap *osdmap_apply_incremental(void **p, void *end,
520 struct ceph_osdmap *map,
521 struct ceph_messenger *msgr)
523 struct ceph_osdmap *newmap = map;
524 struct crush_map *newcrush = NULL;
525 struct ceph_fsid fsid;
527 struct ceph_timespec modified;
529 __s32 new_flags, max;
535 ceph_decode_16_safe(p, end, version, bad);
537 ceph_decode_need(p, end, sizeof(fsid)+sizeof(modified)+2*sizeof(u32),
539 ceph_decode_copy(p, &fsid, sizeof(fsid));
540 epoch = ceph_decode_32(p);
541 BUG_ON(epoch != map->epoch+1);
542 ceph_decode_copy(p, &modified, sizeof(modified));
543 new_flags = ceph_decode_32(p);
546 ceph_decode_32_safe(p, end, len, bad);
548 dout("apply_incremental full map len %d, %p to %p\n",
550 newmap = osdmap_decode(p, min(*p+len, end));
551 return newmap; /* error or not */
555 ceph_decode_32_safe(p, end, len, bad);
557 dout("apply_incremental new crush map len %d, %p to %p\n",
559 newcrush = crush_decode(*p, min(*p+len, end));
560 if (IS_ERR(newcrush))
561 return ERR_PTR(PTR_ERR(newcrush));
566 map->flags = new_flags;
568 ceph_decode_need(p, end, 5*sizeof(u32), bad);
571 max = ceph_decode_32(p);
573 err = osdmap_set_max_osd(map, max);
579 map->modified = map->modified;
582 crush_destroy(map->crush);
583 map->crush = newcrush;
588 ceph_decode_32_safe(p, end, len, bad);
590 ceph_decode_32_safe(p, end, pool, bad);
591 if (pool >= map->num_pools) {
592 void *pg_pool = kcalloc(pool + 1,
593 sizeof(*map->pg_pool),
599 memcpy(pg_pool, map->pg_pool,
600 map->num_pools * sizeof(*map->pg_pool));
602 map->pg_pool = pg_pool;
603 map->num_pools = pool+1;
605 ceph_decode_copy(p, &map->pg_pool[pool].v,
606 sizeof(map->pg_pool->v));
607 calc_pg_masks(&map->pg_pool[pool]);
610 /* old_pool (ignore) */
611 ceph_decode_32_safe(p, end, len, bad);
612 *p += len * sizeof(u32);
616 ceph_decode_32_safe(p, end, len, bad);
619 struct ceph_entity_addr addr;
620 ceph_decode_32_safe(p, end, osd, bad);
621 ceph_decode_copy_safe(p, end, &addr, sizeof(addr), bad);
622 pr_info("osd%d up\n", osd);
623 BUG_ON(osd >= map->max_osd);
624 map->osd_state[osd] |= CEPH_OSD_UP;
625 map->osd_addr[osd] = addr;
629 ceph_decode_32_safe(p, end, len, bad);
632 ceph_decode_32_safe(p, end, osd, bad);
633 (*p)++; /* clean flag */
634 pr_info("ceph osd%d down\n", osd);
635 if (osd < map->max_osd)
636 map->osd_state[osd] &= ~CEPH_OSD_UP;
640 ceph_decode_32_safe(p, end, len, bad);
643 ceph_decode_need(p, end, sizeof(u32)*2, bad);
644 osd = ceph_decode_32(p);
645 off = ceph_decode_32(p);
646 pr_info("osd%d weight 0x%x %s\n", osd, off,
647 off == CEPH_OSD_IN ? "(in)" :
648 (off == CEPH_OSD_OUT ? "(out)" : ""));
649 if (osd < map->max_osd)
650 map->osd_weight[osd] = off;
654 rbp = rb_first(&map->pg_temp);
655 ceph_decode_32_safe(p, end, len, bad);
657 struct ceph_pg_mapping *pg;
661 ceph_decode_need(p, end, sizeof(u64) + sizeof(u32), bad);
662 pgid = ceph_decode_64(p);
663 pglen = ceph_decode_32(p);
666 while (rbp && rb_entry(rbp, struct ceph_pg_mapping,
667 node)->pgid <= pgid) {
668 struct rb_node *cur = rbp;
670 dout(" removed pg_temp %llx\n",
671 rb_entry(cur, struct ceph_pg_mapping, node)->pgid);
672 rb_erase(cur, &map->pg_temp);
677 ceph_decode_need(p, end, pglen*sizeof(u32), bad);
678 pg = kmalloc(sizeof(*pg) + sizeof(u32)*pglen, GFP_NOFS);
685 for (j = 0; j < len; j++)
686 pg->osds[j] = ceph_decode_32(p);
687 err = __insert_pg_mapping(pg, &map->pg_temp);
690 dout(" added pg_temp %llx len %d\n", pgid, pglen);
694 struct rb_node *cur = rbp;
696 dout(" removed pg_temp %llx\n",
697 rb_entry(cur, struct ceph_pg_mapping, node)->pgid);
698 rb_erase(cur, &map->pg_temp);
701 /* ignore the rest */
706 pr_err("corrupt inc osdmap epoch %d off %d (%p of %p-%p)\n",
707 epoch, (int)(*p - start), *p, start, end);
709 crush_destroy(newcrush);
717 * calculate file layout from given offset, length.
718 * fill in correct oid, logical length, and object extent
721 * for now, we write only a single su, until we can
722 * pass a stride back to the caller.
724 void ceph_calc_file_object_mapping(struct ceph_file_layout *layout,
727 u64 *oxoff, u64 *oxlen)
729 u32 osize = le32_to_cpu(layout->fl_object_size);
730 u32 su = le32_to_cpu(layout->fl_stripe_unit);
731 u32 sc = le32_to_cpu(layout->fl_stripe_count);
732 u32 bl, stripeno, stripepos, objsetno;
736 dout("mapping %llu~%llu osize %u fl_su %u\n", off, *plen,
738 su_per_object = osize / su;
739 dout("osize %u / su %u = su_per_object %u\n", osize, su,
742 BUG_ON((su & ~PAGE_MASK) != 0);
743 /* bl = *off / su; */
747 dout("off %llu / su %u = bl %u\n", off, su, bl);
751 objsetno = stripeno / su_per_object;
753 *ono = objsetno * sc + stripepos;
754 dout("objset %u * sc %u = ono %u\n", objsetno, sc, (unsigned)*ono);
756 /* *oxoff = *off % layout->fl_stripe_unit; # offset in su */
758 su_offset = do_div(t, su);
759 *oxoff = su_offset + (stripeno % su_per_object) * su;
762 * Calculate the length of the extent being written to the selected
763 * object. This is the minimum of the full length requested (plen) or
764 * the remainder of the current stripe being written to.
766 *oxlen = min_t(u64, *plen, su - su_offset);
769 dout(" obj extent %llu~%llu\n", *oxoff, *oxlen);
773 * calculate an object layout (i.e. pgid) from an oid,
774 * file_layout, and osdmap
776 int ceph_calc_object_layout(struct ceph_object_layout *ol,
778 struct ceph_file_layout *fl,
779 struct ceph_osdmap *osdmap)
781 unsigned num, num_mask;
783 s32 preferred = (s32)le32_to_cpu(fl->fl_pg_preferred);
784 int poolid = le32_to_cpu(fl->fl_pg_pool);
785 struct ceph_pg_pool_info *pool;
787 if (poolid >= osdmap->num_pools)
789 pool = &osdmap->pg_pool[poolid];
791 if (preferred >= 0) {
792 num = le32_to_cpu(pool->v.lpg_num);
793 num_mask = pool->lpg_num_mask;
795 num = le32_to_cpu(pool->v.pg_num);
796 num_mask = pool->pg_num_mask;
799 pgid.pg64 = 0; /* start with it zeroed out */
800 pgid.pg.ps = ceph_full_name_hash(oid, strlen(oid));
801 pgid.pg.preferred = preferred;
803 pgid.pg.ps += preferred;
804 pgid.pg.pool = le32_to_cpu(fl->fl_pg_pool);
806 dout("calc_object_layout '%s' pgid %d.%xp%d (%llx)\n", oid,
807 pgid.pg.pool, pgid.pg.ps, (int)preferred, pgid.pg64);
809 dout("calc_object_layout '%s' pgid %d.%x (%llx)\n", oid,
810 pgid.pg.pool, pgid.pg.ps, pgid.pg64);
812 ol->ol_pgid = cpu_to_le64(pgid.pg64);
813 ol->ol_stripe_unit = fl->fl_object_stripe_unit;
819 * Calculate raw osd vector for the given pgid. Return pointer to osd
820 * array, or NULL on failure.
822 static int *calc_pg_raw(struct ceph_osdmap *osdmap, union ceph_pg pgid,
825 struct rb_node *n = osdmap->pg_temp.rb_node;
826 struct ceph_pg_mapping *pg;
827 struct ceph_pg_pool_info *pool;
829 unsigned pps; /* placement ps */
833 pg = rb_entry(n, struct ceph_pg_mapping, node);
834 if (pgid.pg64 < pg->pgid)
836 else if (pgid.pg64 > pg->pgid)
845 if (pgid.pg.pool >= osdmap->num_pools)
847 pool = &osdmap->pg_pool[pgid.pg.pool];
848 ruleno = crush_find_rule(osdmap->crush, pool->v.crush_ruleset,
849 pool->v.type, pool->v.size);
851 pr_err("no crush rule pool %d type %d size %d\n",
852 pgid.pg.pool, pool->v.type, pool->v.size);
856 if (pgid.pg.preferred >= 0)
857 pps = ceph_stable_mod(pgid.pg.ps,
858 le32_to_cpu(pool->v.lpgp_num),
859 pool->lpgp_num_mask);
861 pps = ceph_stable_mod(pgid.pg.ps,
862 le32_to_cpu(pool->v.pgp_num),
865 *num = crush_do_rule(osdmap->crush, ruleno, pps, osds,
866 min_t(int, pool->v.size, *num),
867 pgid.pg.preferred, osdmap->osd_weight);
872 * Return primary osd for given pgid, or -1 if none.
874 int ceph_calc_pg_primary(struct ceph_osdmap *osdmap, union ceph_pg pgid)
876 int rawosds[10], *osds;
877 int i, num = ARRAY_SIZE(rawosds);
879 osds = calc_pg_raw(osdmap, pgid, rawosds, &num);
883 /* primary is first up osd */
884 for (i = 0; i < num; i++)
885 if (ceph_osd_is_up(osdmap, osds[i])) {