Merge branch 'bnxt_en-next'
[linux/fpc-iii.git] / net / ceph / osdmap.c
blob03062bb763b34cc64eb7c0b38c042801e0be0b1c
2 #include <linux/ceph/ceph_debug.h>
4 #include <linux/module.h>
5 #include <linux/slab.h>
6 #include <asm/div64.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)
16 if (!len)
17 return str;
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");
25 else
26 snprintf(str, len, "doesn't exist");
28 return str;
31 /* maps */
33 static int calc_bits_of(unsigned int t)
35 int b = 0;
36 while (t) {
37 t = t >> 1;
38 b++;
40 return b;
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;
53 * decode crush map
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);
61 return 0;
62 bad:
63 return -EINVAL;
66 static int crush_decode_list_bucket(void **p, void *end,
67 struct crush_bucket_list *b)
69 int j;
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)
73 return -ENOMEM;
74 b->sum_weights = kcalloc(b->h.size, sizeof(u32), GFP_NOFS);
75 if (b->sum_weights == NULL)
76 return -ENOMEM;
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);
82 return 0;
83 bad:
84 return -EINVAL;
87 static int crush_decode_tree_bucket(void **p, void *end,
88 struct crush_bucket_tree *b)
90 int j;
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)
95 return -ENOMEM;
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);
99 return 0;
100 bad:
101 return -EINVAL;
104 static int crush_decode_straw_bucket(void **p, void *end,
105 struct crush_bucket_straw *b)
107 int j;
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)
111 return -ENOMEM;
112 b->straws = kcalloc(b->h.size, sizeof(u32), GFP_NOFS);
113 if (b->straws == NULL)
114 return -ENOMEM;
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);
120 return 0;
121 bad:
122 return -EINVAL;
125 static int crush_decode_straw2_bucket(void **p, void *end,
126 struct crush_bucket_straw2 *b)
128 int j;
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)
132 return -ENOMEM;
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);
136 return 0;
137 bad:
138 return -EINVAL;
141 static int skip_name_map(void **p, void *end)
143 int len;
144 ceph_decode_32_safe(p, end, len ,bad);
145 while (len--) {
146 int strlen;
147 *p += sizeof(u32);
148 ceph_decode_32_safe(p, end, strlen, bad);
149 *p += strlen;
151 return 0;
152 bad:
153 return -EINVAL;
156 static struct crush_map *crush_decode(void *pbyval, void *end)
158 struct crush_map *c;
159 int err = -EINVAL;
160 int i, j;
161 void **p = &pbyval;
162 void *start = pbyval;
163 u32 magic;
164 u32 num_name_maps;
166 dout("crush_decode %p to %p len %d\n", *p, end, (int)(end - *p));
168 c = kzalloc(sizeof(*c), GFP_NOFS);
169 if (c == NULL)
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);
183 goto bad;
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)
191 goto badmem;
192 c->rules = kcalloc(c->max_rules, sizeof(*c->rules), GFP_NOFS);
193 if (c->rules == NULL)
194 goto badmem;
196 /* buckets */
197 for (i = 0; i < c->max_buckets; i++) {
198 int size = 0;
199 u32 alg;
200 struct crush_bucket *b;
202 ceph_decode_32_safe(p, end, alg, bad);
203 if (alg == 0) {
204 c->buckets[i] = NULL;
205 continue;
207 dout("crush_decode bucket %d off %x %p to %p\n",
208 i, (int)(*p-start), *p, end);
210 switch (alg) {
211 case CRUSH_BUCKET_UNIFORM:
212 size = sizeof(struct crush_bucket_uniform);
213 break;
214 case CRUSH_BUCKET_LIST:
215 size = sizeof(struct crush_bucket_list);
216 break;
217 case CRUSH_BUCKET_TREE:
218 size = sizeof(struct crush_bucket_tree);
219 break;
220 case CRUSH_BUCKET_STRAW:
221 size = sizeof(struct crush_bucket_straw);
222 break;
223 case CRUSH_BUCKET_STRAW2:
224 size = sizeof(struct crush_bucket_straw2);
225 break;
226 default:
227 err = -EINVAL;
228 goto bad;
230 BUG_ON(size == 0);
231 b = c->buckets[i] = kzalloc(size, GFP_NOFS);
232 if (b == NULL)
233 goto badmem;
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)
248 goto badmem;
249 b->perm = kcalloc(b->size, sizeof(u32), GFP_NOFS);
250 if (b->perm == NULL)
251 goto badmem;
252 b->perm_n = 0;
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);
258 switch (b->alg) {
259 case CRUSH_BUCKET_UNIFORM:
260 err = crush_decode_uniform_bucket(p, end,
261 (struct crush_bucket_uniform *)b);
262 if (err < 0)
263 goto bad;
264 break;
265 case CRUSH_BUCKET_LIST:
266 err = crush_decode_list_bucket(p, end,
267 (struct crush_bucket_list *)b);
268 if (err < 0)
269 goto bad;
270 break;
271 case CRUSH_BUCKET_TREE:
272 err = crush_decode_tree_bucket(p, end,
273 (struct crush_bucket_tree *)b);
274 if (err < 0)
275 goto bad;
276 break;
277 case CRUSH_BUCKET_STRAW:
278 err = crush_decode_straw_bucket(p, end,
279 (struct crush_bucket_straw *)b);
280 if (err < 0)
281 goto bad;
282 break;
283 case CRUSH_BUCKET_STRAW2:
284 err = crush_decode_straw2_bucket(p, end,
285 (struct crush_bucket_straw2 *)b);
286 if (err < 0)
287 goto bad;
288 break;
292 /* rules */
293 dout("rule vec is %p\n", c->rules);
294 for (i = 0; i < c->max_rules; i++) {
295 u32 yes;
296 struct crush_rule *r;
298 ceph_decode_32_safe(p, end, yes, bad);
299 if (!yes) {
300 dout("crush_decode NO rule %d off %x %p to %p\n",
301 i, (int)(*p-start), *p, end);
302 c->rules[i] = NULL;
303 continue;
306 dout("crush_decode rule %d off %x %p to %p\n",
307 i, (int)(*p-start), *p, end);
309 /* len */
310 ceph_decode_32_safe(p, end, yes, bad);
311 #if BITS_PER_LONG == 32
312 err = -EINVAL;
313 if (yes > (ULONG_MAX - sizeof(*r))
314 / sizeof(struct crush_rule_step))
315 goto bad;
316 #endif
317 r = c->rules[i] = kmalloc(sizeof(*r) +
318 yes*sizeof(struct crush_rule_step),
319 GFP_NOFS);
320 if (r == NULL)
321 goto badmem;
322 dout(" rule %d is at %p\n", i, r);
323 r->len = yes;
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);
336 if (err < 0)
337 goto done;
340 /* tunables */
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);
371 done:
372 dout("crush_decode success\n");
373 return c;
375 badmem:
376 err = -ENOMEM;
377 bad:
378 dout("crush_decode fail %d\n", err);
379 crush_destroy(c);
380 return ERR_PTR(err);
383 int ceph_pg_compare(const struct ceph_pg *lhs, const struct ceph_pg *rhs)
385 if (lhs->pool < rhs->pool)
386 return -1;
387 if (lhs->pool > rhs->pool)
388 return 1;
389 if (lhs->seed < rhs->seed)
390 return -1;
391 if (lhs->seed > rhs->seed)
392 return 1;
394 return 0;
398 * rbtree of pg_mapping for handling pg_temp (explicit mapping of pgid
399 * to a set of osds) and primary_temp (explicit primary setting)
401 static int __insert_pg_mapping(struct ceph_pg_mapping *new,
402 struct rb_root *root)
404 struct rb_node **p = &root->rb_node;
405 struct rb_node *parent = NULL;
406 struct ceph_pg_mapping *pg = NULL;
407 int c;
409 dout("__insert_pg_mapping %llx %p\n", *(u64 *)&new->pgid, new);
410 while (*p) {
411 parent = *p;
412 pg = rb_entry(parent, struct ceph_pg_mapping, node);
413 c = ceph_pg_compare(&new->pgid, &pg->pgid);
414 if (c < 0)
415 p = &(*p)->rb_left;
416 else if (c > 0)
417 p = &(*p)->rb_right;
418 else
419 return -EEXIST;
422 rb_link_node(&new->node, parent, p);
423 rb_insert_color(&new->node, root);
424 return 0;
427 static struct ceph_pg_mapping *__lookup_pg_mapping(struct rb_root *root,
428 struct ceph_pg pgid)
430 struct rb_node *n = root->rb_node;
431 struct ceph_pg_mapping *pg;
432 int c;
434 while (n) {
435 pg = rb_entry(n, struct ceph_pg_mapping, node);
436 c = ceph_pg_compare(&pgid, &pg->pgid);
437 if (c < 0) {
438 n = n->rb_left;
439 } else if (c > 0) {
440 n = n->rb_right;
441 } else {
442 dout("__lookup_pg_mapping %lld.%x got %p\n",
443 pgid.pool, pgid.seed, pg);
444 return pg;
447 return NULL;
450 static int __remove_pg_mapping(struct rb_root *root, struct ceph_pg pgid)
452 struct ceph_pg_mapping *pg = __lookup_pg_mapping(root, pgid);
454 if (pg) {
455 dout("__remove_pg_mapping %lld.%x %p\n", pgid.pool, pgid.seed,
456 pg);
457 rb_erase(&pg->node, root);
458 kfree(pg);
459 return 0;
461 dout("__remove_pg_mapping %lld.%x dne\n", pgid.pool, pgid.seed);
462 return -ENOENT;
466 * rbtree of pg pool info
468 static int __insert_pg_pool(struct rb_root *root, struct ceph_pg_pool_info *new)
470 struct rb_node **p = &root->rb_node;
471 struct rb_node *parent = NULL;
472 struct ceph_pg_pool_info *pi = NULL;
474 while (*p) {
475 parent = *p;
476 pi = rb_entry(parent, struct ceph_pg_pool_info, node);
477 if (new->id < pi->id)
478 p = &(*p)->rb_left;
479 else if (new->id > pi->id)
480 p = &(*p)->rb_right;
481 else
482 return -EEXIST;
485 rb_link_node(&new->node, parent, p);
486 rb_insert_color(&new->node, root);
487 return 0;
490 static struct ceph_pg_pool_info *__lookup_pg_pool(struct rb_root *root, u64 id)
492 struct ceph_pg_pool_info *pi;
493 struct rb_node *n = root->rb_node;
495 while (n) {
496 pi = rb_entry(n, struct ceph_pg_pool_info, node);
497 if (id < pi->id)
498 n = n->rb_left;
499 else if (id > pi->id)
500 n = n->rb_right;
501 else
502 return pi;
504 return NULL;
507 struct ceph_pg_pool_info *ceph_pg_pool_by_id(struct ceph_osdmap *map, u64 id)
509 return __lookup_pg_pool(&map->pg_pools, id);
512 const char *ceph_pg_pool_name_by_id(struct ceph_osdmap *map, u64 id)
514 struct ceph_pg_pool_info *pi;
516 if (id == CEPH_NOPOOL)
517 return NULL;
519 if (WARN_ON_ONCE(id > (u64) INT_MAX))
520 return NULL;
522 pi = __lookup_pg_pool(&map->pg_pools, (int) id);
524 return pi ? pi->name : NULL;
526 EXPORT_SYMBOL(ceph_pg_pool_name_by_id);
528 int ceph_pg_poolid_by_name(struct ceph_osdmap *map, const char *name)
530 struct rb_node *rbp;
532 for (rbp = rb_first(&map->pg_pools); rbp; rbp = rb_next(rbp)) {
533 struct ceph_pg_pool_info *pi =
534 rb_entry(rbp, struct ceph_pg_pool_info, node);
535 if (pi->name && strcmp(pi->name, name) == 0)
536 return pi->id;
538 return -ENOENT;
540 EXPORT_SYMBOL(ceph_pg_poolid_by_name);
542 static void __remove_pg_pool(struct rb_root *root, struct ceph_pg_pool_info *pi)
544 rb_erase(&pi->node, root);
545 kfree(pi->name);
546 kfree(pi);
549 static int decode_pool(void **p, void *end, struct ceph_pg_pool_info *pi)
551 u8 ev, cv;
552 unsigned len, num;
553 void *pool_end;
555 ceph_decode_need(p, end, 2 + 4, bad);
556 ev = ceph_decode_8(p); /* encoding version */
557 cv = ceph_decode_8(p); /* compat version */
558 if (ev < 5) {
559 pr_warn("got v %d < 5 cv %d of ceph_pg_pool\n", ev, cv);
560 return -EINVAL;
562 if (cv > 9) {
563 pr_warn("got v %d cv %d > 9 of ceph_pg_pool\n", ev, cv);
564 return -EINVAL;
566 len = ceph_decode_32(p);
567 ceph_decode_need(p, end, len, bad);
568 pool_end = *p + len;
570 pi->type = ceph_decode_8(p);
571 pi->size = ceph_decode_8(p);
572 pi->crush_ruleset = ceph_decode_8(p);
573 pi->object_hash = ceph_decode_8(p);
575 pi->pg_num = ceph_decode_32(p);
576 pi->pgp_num = ceph_decode_32(p);
578 *p += 4 + 4; /* skip lpg* */
579 *p += 4; /* skip last_change */
580 *p += 8 + 4; /* skip snap_seq, snap_epoch */
582 /* skip snaps */
583 num = ceph_decode_32(p);
584 while (num--) {
585 *p += 8; /* snapid key */
586 *p += 1 + 1; /* versions */
587 len = ceph_decode_32(p);
588 *p += len;
591 /* skip removed_snaps */
592 num = ceph_decode_32(p);
593 *p += num * (8 + 8);
595 *p += 8; /* skip auid */
596 pi->flags = ceph_decode_64(p);
597 *p += 4; /* skip crash_replay_interval */
599 if (ev >= 7)
600 pi->min_size = ceph_decode_8(p);
601 else
602 pi->min_size = pi->size - pi->size / 2;
604 if (ev >= 8)
605 *p += 8 + 8; /* skip quota_max_* */
607 if (ev >= 9) {
608 /* skip tiers */
609 num = ceph_decode_32(p);
610 *p += num * 8;
612 *p += 8; /* skip tier_of */
613 *p += 1; /* skip cache_mode */
615 pi->read_tier = ceph_decode_64(p);
616 pi->write_tier = ceph_decode_64(p);
617 } else {
618 pi->read_tier = -1;
619 pi->write_tier = -1;
622 if (ev >= 10) {
623 /* skip properties */
624 num = ceph_decode_32(p);
625 while (num--) {
626 len = ceph_decode_32(p);
627 *p += len; /* key */
628 len = ceph_decode_32(p);
629 *p += len; /* val */
633 if (ev >= 11) {
634 /* skip hit_set_params */
635 *p += 1 + 1; /* versions */
636 len = ceph_decode_32(p);
637 *p += len;
639 *p += 4; /* skip hit_set_period */
640 *p += 4; /* skip hit_set_count */
643 if (ev >= 12)
644 *p += 4; /* skip stripe_width */
646 if (ev >= 13) {
647 *p += 8; /* skip target_max_bytes */
648 *p += 8; /* skip target_max_objects */
649 *p += 4; /* skip cache_target_dirty_ratio_micro */
650 *p += 4; /* skip cache_target_full_ratio_micro */
651 *p += 4; /* skip cache_min_flush_age */
652 *p += 4; /* skip cache_min_evict_age */
655 if (ev >= 14) {
656 /* skip erasure_code_profile */
657 len = ceph_decode_32(p);
658 *p += len;
661 if (ev >= 15)
662 pi->last_force_request_resend = ceph_decode_32(p);
663 else
664 pi->last_force_request_resend = 0;
666 /* ignore the rest */
668 *p = pool_end;
669 calc_pg_masks(pi);
670 return 0;
672 bad:
673 return -EINVAL;
676 static int decode_pool_names(void **p, void *end, struct ceph_osdmap *map)
678 struct ceph_pg_pool_info *pi;
679 u32 num, len;
680 u64 pool;
682 ceph_decode_32_safe(p, end, num, bad);
683 dout(" %d pool names\n", num);
684 while (num--) {
685 ceph_decode_64_safe(p, end, pool, bad);
686 ceph_decode_32_safe(p, end, len, bad);
687 dout(" pool %llu len %d\n", pool, len);
688 ceph_decode_need(p, end, len, bad);
689 pi = __lookup_pg_pool(&map->pg_pools, pool);
690 if (pi) {
691 char *name = kstrndup(*p, len, GFP_NOFS);
693 if (!name)
694 return -ENOMEM;
695 kfree(pi->name);
696 pi->name = name;
697 dout(" name is %s\n", pi->name);
699 *p += len;
701 return 0;
703 bad:
704 return -EINVAL;
708 * osd map
710 struct ceph_osdmap *ceph_osdmap_alloc(void)
712 struct ceph_osdmap *map;
714 map = kzalloc(sizeof(*map), GFP_NOIO);
715 if (!map)
716 return NULL;
718 map->pg_pools = RB_ROOT;
719 map->pool_max = -1;
720 map->pg_temp = RB_ROOT;
721 map->primary_temp = RB_ROOT;
722 mutex_init(&map->crush_scratch_mutex);
724 return map;
727 void ceph_osdmap_destroy(struct ceph_osdmap *map)
729 dout("osdmap_destroy %p\n", map);
730 if (map->crush)
731 crush_destroy(map->crush);
732 while (!RB_EMPTY_ROOT(&map->pg_temp)) {
733 struct ceph_pg_mapping *pg =
734 rb_entry(rb_first(&map->pg_temp),
735 struct ceph_pg_mapping, node);
736 rb_erase(&pg->node, &map->pg_temp);
737 kfree(pg);
739 while (!RB_EMPTY_ROOT(&map->primary_temp)) {
740 struct ceph_pg_mapping *pg =
741 rb_entry(rb_first(&map->primary_temp),
742 struct ceph_pg_mapping, node);
743 rb_erase(&pg->node, &map->primary_temp);
744 kfree(pg);
746 while (!RB_EMPTY_ROOT(&map->pg_pools)) {
747 struct ceph_pg_pool_info *pi =
748 rb_entry(rb_first(&map->pg_pools),
749 struct ceph_pg_pool_info, node);
750 __remove_pg_pool(&map->pg_pools, pi);
752 kfree(map->osd_state);
753 kfree(map->osd_weight);
754 kfree(map->osd_addr);
755 kfree(map->osd_primary_affinity);
756 kfree(map);
760 * Adjust max_osd value, (re)allocate arrays.
762 * The new elements are properly initialized.
764 static int osdmap_set_max_osd(struct ceph_osdmap *map, int max)
766 u8 *state;
767 u32 *weight;
768 struct ceph_entity_addr *addr;
769 int i;
771 state = krealloc(map->osd_state, max*sizeof(*state), GFP_NOFS);
772 if (!state)
773 return -ENOMEM;
774 map->osd_state = state;
776 weight = krealloc(map->osd_weight, max*sizeof(*weight), GFP_NOFS);
777 if (!weight)
778 return -ENOMEM;
779 map->osd_weight = weight;
781 addr = krealloc(map->osd_addr, max*sizeof(*addr), GFP_NOFS);
782 if (!addr)
783 return -ENOMEM;
784 map->osd_addr = addr;
786 for (i = map->max_osd; i < max; i++) {
787 map->osd_state[i] = 0;
788 map->osd_weight[i] = CEPH_OSD_OUT;
789 memset(map->osd_addr + i, 0, sizeof(*map->osd_addr));
792 if (map->osd_primary_affinity) {
793 u32 *affinity;
795 affinity = krealloc(map->osd_primary_affinity,
796 max*sizeof(*affinity), GFP_NOFS);
797 if (!affinity)
798 return -ENOMEM;
799 map->osd_primary_affinity = affinity;
801 for (i = map->max_osd; i < max; i++)
802 map->osd_primary_affinity[i] =
803 CEPH_OSD_DEFAULT_PRIMARY_AFFINITY;
806 map->max_osd = max;
808 return 0;
811 #define OSDMAP_WRAPPER_COMPAT_VER 7
812 #define OSDMAP_CLIENT_DATA_COMPAT_VER 1
815 * Return 0 or error. On success, *v is set to 0 for old (v6) osdmaps,
816 * to struct_v of the client_data section for new (v7 and above)
817 * osdmaps.
819 static int get_osdmap_client_data_v(void **p, void *end,
820 const char *prefix, u8 *v)
822 u8 struct_v;
824 ceph_decode_8_safe(p, end, struct_v, e_inval);
825 if (struct_v >= 7) {
826 u8 struct_compat;
828 ceph_decode_8_safe(p, end, struct_compat, e_inval);
829 if (struct_compat > OSDMAP_WRAPPER_COMPAT_VER) {
830 pr_warn("got v %d cv %d > %d of %s ceph_osdmap\n",
831 struct_v, struct_compat,
832 OSDMAP_WRAPPER_COMPAT_VER, prefix);
833 return -EINVAL;
835 *p += 4; /* ignore wrapper struct_len */
837 ceph_decode_8_safe(p, end, struct_v, e_inval);
838 ceph_decode_8_safe(p, end, struct_compat, e_inval);
839 if (struct_compat > OSDMAP_CLIENT_DATA_COMPAT_VER) {
840 pr_warn("got v %d cv %d > %d of %s ceph_osdmap client data\n",
841 struct_v, struct_compat,
842 OSDMAP_CLIENT_DATA_COMPAT_VER, prefix);
843 return -EINVAL;
845 *p += 4; /* ignore client data struct_len */
846 } else {
847 u16 version;
849 *p -= 1;
850 ceph_decode_16_safe(p, end, version, e_inval);
851 if (version < 6) {
852 pr_warn("got v %d < 6 of %s ceph_osdmap\n",
853 version, prefix);
854 return -EINVAL;
857 /* old osdmap enconding */
858 struct_v = 0;
861 *v = struct_v;
862 return 0;
864 e_inval:
865 return -EINVAL;
868 static int __decode_pools(void **p, void *end, struct ceph_osdmap *map,
869 bool incremental)
871 u32 n;
873 ceph_decode_32_safe(p, end, n, e_inval);
874 while (n--) {
875 struct ceph_pg_pool_info *pi;
876 u64 pool;
877 int ret;
879 ceph_decode_64_safe(p, end, pool, e_inval);
881 pi = __lookup_pg_pool(&map->pg_pools, pool);
882 if (!incremental || !pi) {
883 pi = kzalloc(sizeof(*pi), GFP_NOFS);
884 if (!pi)
885 return -ENOMEM;
887 pi->id = pool;
889 ret = __insert_pg_pool(&map->pg_pools, pi);
890 if (ret) {
891 kfree(pi);
892 return ret;
896 ret = decode_pool(p, end, pi);
897 if (ret)
898 return ret;
901 return 0;
903 e_inval:
904 return -EINVAL;
907 static int decode_pools(void **p, void *end, struct ceph_osdmap *map)
909 return __decode_pools(p, end, map, false);
912 static int decode_new_pools(void **p, void *end, struct ceph_osdmap *map)
914 return __decode_pools(p, end, map, true);
917 static int __decode_pg_temp(void **p, void *end, struct ceph_osdmap *map,
918 bool incremental)
920 u32 n;
922 ceph_decode_32_safe(p, end, n, e_inval);
923 while (n--) {
924 struct ceph_pg pgid;
925 u32 len, i;
926 int ret;
928 ret = ceph_decode_pgid(p, end, &pgid);
929 if (ret)
930 return ret;
932 ceph_decode_32_safe(p, end, len, e_inval);
934 ret = __remove_pg_mapping(&map->pg_temp, pgid);
935 BUG_ON(!incremental && ret != -ENOENT);
937 if (!incremental || len > 0) {
938 struct ceph_pg_mapping *pg;
940 ceph_decode_need(p, end, len*sizeof(u32), e_inval);
942 if (len > (UINT_MAX - sizeof(*pg)) / sizeof(u32))
943 return -EINVAL;
945 pg = kzalloc(sizeof(*pg) + len*sizeof(u32), GFP_NOFS);
946 if (!pg)
947 return -ENOMEM;
949 pg->pgid = pgid;
950 pg->pg_temp.len = len;
951 for (i = 0; i < len; i++)
952 pg->pg_temp.osds[i] = ceph_decode_32(p);
954 ret = __insert_pg_mapping(pg, &map->pg_temp);
955 if (ret) {
956 kfree(pg);
957 return ret;
962 return 0;
964 e_inval:
965 return -EINVAL;
968 static int decode_pg_temp(void **p, void *end, struct ceph_osdmap *map)
970 return __decode_pg_temp(p, end, map, false);
973 static int decode_new_pg_temp(void **p, void *end, struct ceph_osdmap *map)
975 return __decode_pg_temp(p, end, map, true);
978 static int __decode_primary_temp(void **p, void *end, struct ceph_osdmap *map,
979 bool incremental)
981 u32 n;
983 ceph_decode_32_safe(p, end, n, e_inval);
984 while (n--) {
985 struct ceph_pg pgid;
986 u32 osd;
987 int ret;
989 ret = ceph_decode_pgid(p, end, &pgid);
990 if (ret)
991 return ret;
993 ceph_decode_32_safe(p, end, osd, e_inval);
995 ret = __remove_pg_mapping(&map->primary_temp, pgid);
996 BUG_ON(!incremental && ret != -ENOENT);
998 if (!incremental || osd != (u32)-1) {
999 struct ceph_pg_mapping *pg;
1001 pg = kzalloc(sizeof(*pg), GFP_NOFS);
1002 if (!pg)
1003 return -ENOMEM;
1005 pg->pgid = pgid;
1006 pg->primary_temp.osd = osd;
1008 ret = __insert_pg_mapping(pg, &map->primary_temp);
1009 if (ret) {
1010 kfree(pg);
1011 return ret;
1016 return 0;
1018 e_inval:
1019 return -EINVAL;
1022 static int decode_primary_temp(void **p, void *end, struct ceph_osdmap *map)
1024 return __decode_primary_temp(p, end, map, false);
1027 static int decode_new_primary_temp(void **p, void *end,
1028 struct ceph_osdmap *map)
1030 return __decode_primary_temp(p, end, map, true);
1033 u32 ceph_get_primary_affinity(struct ceph_osdmap *map, int osd)
1035 BUG_ON(osd >= map->max_osd);
1037 if (!map->osd_primary_affinity)
1038 return CEPH_OSD_DEFAULT_PRIMARY_AFFINITY;
1040 return map->osd_primary_affinity[osd];
1043 static int set_primary_affinity(struct ceph_osdmap *map, int osd, u32 aff)
1045 BUG_ON(osd >= map->max_osd);
1047 if (!map->osd_primary_affinity) {
1048 int i;
1050 map->osd_primary_affinity = kmalloc(map->max_osd*sizeof(u32),
1051 GFP_NOFS);
1052 if (!map->osd_primary_affinity)
1053 return -ENOMEM;
1055 for (i = 0; i < map->max_osd; i++)
1056 map->osd_primary_affinity[i] =
1057 CEPH_OSD_DEFAULT_PRIMARY_AFFINITY;
1060 map->osd_primary_affinity[osd] = aff;
1062 return 0;
1065 static int decode_primary_affinity(void **p, void *end,
1066 struct ceph_osdmap *map)
1068 u32 len, i;
1070 ceph_decode_32_safe(p, end, len, e_inval);
1071 if (len == 0) {
1072 kfree(map->osd_primary_affinity);
1073 map->osd_primary_affinity = NULL;
1074 return 0;
1076 if (len != map->max_osd)
1077 goto e_inval;
1079 ceph_decode_need(p, end, map->max_osd*sizeof(u32), e_inval);
1081 for (i = 0; i < map->max_osd; i++) {
1082 int ret;
1084 ret = set_primary_affinity(map, i, ceph_decode_32(p));
1085 if (ret)
1086 return ret;
1089 return 0;
1091 e_inval:
1092 return -EINVAL;
1095 static int decode_new_primary_affinity(void **p, void *end,
1096 struct ceph_osdmap *map)
1098 u32 n;
1100 ceph_decode_32_safe(p, end, n, e_inval);
1101 while (n--) {
1102 u32 osd, aff;
1103 int ret;
1105 ceph_decode_32_safe(p, end, osd, e_inval);
1106 ceph_decode_32_safe(p, end, aff, e_inval);
1108 ret = set_primary_affinity(map, osd, aff);
1109 if (ret)
1110 return ret;
1112 pr_info("osd%d primary-affinity 0x%x\n", osd, aff);
1115 return 0;
1117 e_inval:
1118 return -EINVAL;
1122 * decode a full map.
1124 static int osdmap_decode(void **p, void *end, struct ceph_osdmap *map)
1126 u8 struct_v;
1127 u32 epoch = 0;
1128 void *start = *p;
1129 u32 max;
1130 u32 len, i;
1131 int err;
1133 dout("%s %p to %p len %d\n", __func__, *p, end, (int)(end - *p));
1135 err = get_osdmap_client_data_v(p, end, "full", &struct_v);
1136 if (err)
1137 goto bad;
1139 /* fsid, epoch, created, modified */
1140 ceph_decode_need(p, end, sizeof(map->fsid) + sizeof(u32) +
1141 sizeof(map->created) + sizeof(map->modified), e_inval);
1142 ceph_decode_copy(p, &map->fsid, sizeof(map->fsid));
1143 epoch = map->epoch = ceph_decode_32(p);
1144 ceph_decode_copy(p, &map->created, sizeof(map->created));
1145 ceph_decode_copy(p, &map->modified, sizeof(map->modified));
1147 /* pools */
1148 err = decode_pools(p, end, map);
1149 if (err)
1150 goto bad;
1152 /* pool_name */
1153 err = decode_pool_names(p, end, map);
1154 if (err)
1155 goto bad;
1157 ceph_decode_32_safe(p, end, map->pool_max, e_inval);
1159 ceph_decode_32_safe(p, end, map->flags, e_inval);
1161 /* max_osd */
1162 ceph_decode_32_safe(p, end, max, e_inval);
1164 /* (re)alloc osd arrays */
1165 err = osdmap_set_max_osd(map, max);
1166 if (err)
1167 goto bad;
1169 /* osd_state, osd_weight, osd_addrs->client_addr */
1170 ceph_decode_need(p, end, 3*sizeof(u32) +
1171 map->max_osd*(1 + sizeof(*map->osd_weight) +
1172 sizeof(*map->osd_addr)), e_inval);
1174 if (ceph_decode_32(p) != map->max_osd)
1175 goto e_inval;
1177 ceph_decode_copy(p, map->osd_state, map->max_osd);
1179 if (ceph_decode_32(p) != map->max_osd)
1180 goto e_inval;
1182 for (i = 0; i < map->max_osd; i++)
1183 map->osd_weight[i] = ceph_decode_32(p);
1185 if (ceph_decode_32(p) != map->max_osd)
1186 goto e_inval;
1188 ceph_decode_copy(p, map->osd_addr, map->max_osd*sizeof(*map->osd_addr));
1189 for (i = 0; i < map->max_osd; i++)
1190 ceph_decode_addr(&map->osd_addr[i]);
1192 /* pg_temp */
1193 err = decode_pg_temp(p, end, map);
1194 if (err)
1195 goto bad;
1197 /* primary_temp */
1198 if (struct_v >= 1) {
1199 err = decode_primary_temp(p, end, map);
1200 if (err)
1201 goto bad;
1204 /* primary_affinity */
1205 if (struct_v >= 2) {
1206 err = decode_primary_affinity(p, end, map);
1207 if (err)
1208 goto bad;
1209 } else {
1210 /* XXX can this happen? */
1211 kfree(map->osd_primary_affinity);
1212 map->osd_primary_affinity = NULL;
1215 /* crush */
1216 ceph_decode_32_safe(p, end, len, e_inval);
1217 map->crush = crush_decode(*p, min(*p + len, end));
1218 if (IS_ERR(map->crush)) {
1219 err = PTR_ERR(map->crush);
1220 map->crush = NULL;
1221 goto bad;
1223 *p += len;
1225 /* ignore the rest */
1226 *p = end;
1228 dout("full osdmap epoch %d max_osd %d\n", map->epoch, map->max_osd);
1229 return 0;
1231 e_inval:
1232 err = -EINVAL;
1233 bad:
1234 pr_err("corrupt full osdmap (%d) epoch %d off %d (%p of %p-%p)\n",
1235 err, epoch, (int)(*p - start), *p, start, end);
1236 print_hex_dump(KERN_DEBUG, "osdmap: ",
1237 DUMP_PREFIX_OFFSET, 16, 1,
1238 start, end - start, true);
1239 return err;
1243 * Allocate and decode a full map.
1245 struct ceph_osdmap *ceph_osdmap_decode(void **p, void *end)
1247 struct ceph_osdmap *map;
1248 int ret;
1250 map = ceph_osdmap_alloc();
1251 if (!map)
1252 return ERR_PTR(-ENOMEM);
1254 ret = osdmap_decode(p, end, map);
1255 if (ret) {
1256 ceph_osdmap_destroy(map);
1257 return ERR_PTR(ret);
1260 return map;
1264 * decode and apply an incremental map update.
1266 struct ceph_osdmap *osdmap_apply_incremental(void **p, void *end,
1267 struct ceph_osdmap *map)
1269 struct crush_map *newcrush = NULL;
1270 struct ceph_fsid fsid;
1271 u32 epoch = 0;
1272 struct ceph_timespec modified;
1273 s32 len;
1274 u64 pool;
1275 __s64 new_pool_max;
1276 __s32 new_flags, max;
1277 void *start = *p;
1278 int err;
1279 u8 struct_v;
1281 dout("%s %p to %p len %d\n", __func__, *p, end, (int)(end - *p));
1283 err = get_osdmap_client_data_v(p, end, "inc", &struct_v);
1284 if (err)
1285 goto bad;
1287 /* fsid, epoch, modified, new_pool_max, new_flags */
1288 ceph_decode_need(p, end, sizeof(fsid) + sizeof(u32) + sizeof(modified) +
1289 sizeof(u64) + sizeof(u32), e_inval);
1290 ceph_decode_copy(p, &fsid, sizeof(fsid));
1291 epoch = ceph_decode_32(p);
1292 BUG_ON(epoch != map->epoch+1);
1293 ceph_decode_copy(p, &modified, sizeof(modified));
1294 new_pool_max = ceph_decode_64(p);
1295 new_flags = ceph_decode_32(p);
1297 /* full map? */
1298 ceph_decode_32_safe(p, end, len, e_inval);
1299 if (len > 0) {
1300 dout("apply_incremental full map len %d, %p to %p\n",
1301 len, *p, end);
1302 return ceph_osdmap_decode(p, min(*p+len, end));
1305 /* new crush? */
1306 ceph_decode_32_safe(p, end, len, e_inval);
1307 if (len > 0) {
1308 newcrush = crush_decode(*p, min(*p+len, end));
1309 if (IS_ERR(newcrush)) {
1310 err = PTR_ERR(newcrush);
1311 newcrush = NULL;
1312 goto bad;
1314 *p += len;
1317 /* new flags? */
1318 if (new_flags >= 0)
1319 map->flags = new_flags;
1320 if (new_pool_max >= 0)
1321 map->pool_max = new_pool_max;
1323 /* new max? */
1324 ceph_decode_32_safe(p, end, max, e_inval);
1325 if (max >= 0) {
1326 err = osdmap_set_max_osd(map, max);
1327 if (err)
1328 goto bad;
1331 map->epoch++;
1332 map->modified = modified;
1333 if (newcrush) {
1334 if (map->crush)
1335 crush_destroy(map->crush);
1336 map->crush = newcrush;
1337 newcrush = NULL;
1340 /* new_pools */
1341 err = decode_new_pools(p, end, map);
1342 if (err)
1343 goto bad;
1345 /* new_pool_names */
1346 err = decode_pool_names(p, end, map);
1347 if (err)
1348 goto bad;
1350 /* old_pool */
1351 ceph_decode_32_safe(p, end, len, e_inval);
1352 while (len--) {
1353 struct ceph_pg_pool_info *pi;
1355 ceph_decode_64_safe(p, end, pool, e_inval);
1356 pi = __lookup_pg_pool(&map->pg_pools, pool);
1357 if (pi)
1358 __remove_pg_pool(&map->pg_pools, pi);
1361 /* new_up */
1362 ceph_decode_32_safe(p, end, len, e_inval);
1363 while (len--) {
1364 u32 osd;
1365 struct ceph_entity_addr addr;
1366 ceph_decode_32_safe(p, end, osd, e_inval);
1367 ceph_decode_copy_safe(p, end, &addr, sizeof(addr), e_inval);
1368 ceph_decode_addr(&addr);
1369 pr_info("osd%d up\n", osd);
1370 BUG_ON(osd >= map->max_osd);
1371 map->osd_state[osd] |= CEPH_OSD_UP | CEPH_OSD_EXISTS;
1372 map->osd_addr[osd] = addr;
1375 /* new_state */
1376 ceph_decode_32_safe(p, end, len, e_inval);
1377 while (len--) {
1378 u32 osd;
1379 u8 xorstate;
1380 ceph_decode_32_safe(p, end, osd, e_inval);
1381 xorstate = **(u8 **)p;
1382 (*p)++; /* clean flag */
1383 if (xorstate == 0)
1384 xorstate = CEPH_OSD_UP;
1385 if (xorstate & CEPH_OSD_UP)
1386 pr_info("osd%d down\n", osd);
1387 if (osd < map->max_osd)
1388 map->osd_state[osd] ^= xorstate;
1391 /* new_weight */
1392 ceph_decode_32_safe(p, end, len, e_inval);
1393 while (len--) {
1394 u32 osd, off;
1395 ceph_decode_need(p, end, sizeof(u32)*2, e_inval);
1396 osd = ceph_decode_32(p);
1397 off = ceph_decode_32(p);
1398 pr_info("osd%d weight 0x%x %s\n", osd, off,
1399 off == CEPH_OSD_IN ? "(in)" :
1400 (off == CEPH_OSD_OUT ? "(out)" : ""));
1401 if (osd < map->max_osd)
1402 map->osd_weight[osd] = off;
1405 /* new_pg_temp */
1406 err = decode_new_pg_temp(p, end, map);
1407 if (err)
1408 goto bad;
1410 /* new_primary_temp */
1411 if (struct_v >= 1) {
1412 err = decode_new_primary_temp(p, end, map);
1413 if (err)
1414 goto bad;
1417 /* new_primary_affinity */
1418 if (struct_v >= 2) {
1419 err = decode_new_primary_affinity(p, end, map);
1420 if (err)
1421 goto bad;
1424 /* ignore the rest */
1425 *p = end;
1427 dout("inc osdmap epoch %d max_osd %d\n", map->epoch, map->max_osd);
1428 return map;
1430 e_inval:
1431 err = -EINVAL;
1432 bad:
1433 pr_err("corrupt inc osdmap (%d) epoch %d off %d (%p of %p-%p)\n",
1434 err, epoch, (int)(*p - start), *p, start, end);
1435 print_hex_dump(KERN_DEBUG, "osdmap: ",
1436 DUMP_PREFIX_OFFSET, 16, 1,
1437 start, end - start, true);
1438 if (newcrush)
1439 crush_destroy(newcrush);
1440 return ERR_PTR(err);
1443 void ceph_oid_copy(struct ceph_object_id *dest,
1444 const struct ceph_object_id *src)
1446 WARN_ON(!ceph_oid_empty(dest));
1448 if (src->name != src->inline_name) {
1449 /* very rare, see ceph_object_id definition */
1450 dest->name = kmalloc(src->name_len + 1,
1451 GFP_NOIO | __GFP_NOFAIL);
1454 memcpy(dest->name, src->name, src->name_len + 1);
1455 dest->name_len = src->name_len;
1457 EXPORT_SYMBOL(ceph_oid_copy);
1459 static __printf(2, 0)
1460 int oid_printf_vargs(struct ceph_object_id *oid, const char *fmt, va_list ap)
1462 int len;
1464 WARN_ON(!ceph_oid_empty(oid));
1466 len = vsnprintf(oid->inline_name, sizeof(oid->inline_name), fmt, ap);
1467 if (len >= sizeof(oid->inline_name))
1468 return len;
1470 oid->name_len = len;
1471 return 0;
1475 * If oid doesn't fit into inline buffer, BUG.
1477 void ceph_oid_printf(struct ceph_object_id *oid, const char *fmt, ...)
1479 va_list ap;
1481 va_start(ap, fmt);
1482 BUG_ON(oid_printf_vargs(oid, fmt, ap));
1483 va_end(ap);
1485 EXPORT_SYMBOL(ceph_oid_printf);
1487 static __printf(3, 0)
1488 int oid_aprintf_vargs(struct ceph_object_id *oid, gfp_t gfp,
1489 const char *fmt, va_list ap)
1491 va_list aq;
1492 int len;
1494 va_copy(aq, ap);
1495 len = oid_printf_vargs(oid, fmt, aq);
1496 va_end(aq);
1498 if (len) {
1499 char *external_name;
1501 external_name = kmalloc(len + 1, gfp);
1502 if (!external_name)
1503 return -ENOMEM;
1505 oid->name = external_name;
1506 WARN_ON(vsnprintf(oid->name, len + 1, fmt, ap) != len);
1507 oid->name_len = len;
1510 return 0;
1514 * If oid doesn't fit into inline buffer, allocate.
1516 int ceph_oid_aprintf(struct ceph_object_id *oid, gfp_t gfp,
1517 const char *fmt, ...)
1519 va_list ap;
1520 int ret;
1522 va_start(ap, fmt);
1523 ret = oid_aprintf_vargs(oid, gfp, fmt, ap);
1524 va_end(ap);
1526 return ret;
1528 EXPORT_SYMBOL(ceph_oid_aprintf);
1530 void ceph_oid_destroy(struct ceph_object_id *oid)
1532 if (oid->name != oid->inline_name)
1533 kfree(oid->name);
1535 EXPORT_SYMBOL(ceph_oid_destroy);
1538 * osds only
1540 static bool __osds_equal(const struct ceph_osds *lhs,
1541 const struct ceph_osds *rhs)
1543 if (lhs->size == rhs->size &&
1544 !memcmp(lhs->osds, rhs->osds, rhs->size * sizeof(rhs->osds[0])))
1545 return true;
1547 return false;
1551 * osds + primary
1553 static bool osds_equal(const struct ceph_osds *lhs,
1554 const struct ceph_osds *rhs)
1556 if (__osds_equal(lhs, rhs) &&
1557 lhs->primary == rhs->primary)
1558 return true;
1560 return false;
1563 static bool osds_valid(const struct ceph_osds *set)
1565 /* non-empty set */
1566 if (set->size > 0 && set->primary >= 0)
1567 return true;
1569 /* empty can_shift_osds set */
1570 if (!set->size && set->primary == -1)
1571 return true;
1573 /* empty !can_shift_osds set - all NONE */
1574 if (set->size > 0 && set->primary == -1) {
1575 int i;
1577 for (i = 0; i < set->size; i++) {
1578 if (set->osds[i] != CRUSH_ITEM_NONE)
1579 break;
1581 if (i == set->size)
1582 return true;
1585 return false;
1588 void ceph_osds_copy(struct ceph_osds *dest, const struct ceph_osds *src)
1590 memcpy(dest->osds, src->osds, src->size * sizeof(src->osds[0]));
1591 dest->size = src->size;
1592 dest->primary = src->primary;
1595 static bool is_split(const struct ceph_pg *pgid,
1596 u32 old_pg_num,
1597 u32 new_pg_num)
1599 int old_bits = calc_bits_of(old_pg_num);
1600 int old_mask = (1 << old_bits) - 1;
1601 int n;
1603 WARN_ON(pgid->seed >= old_pg_num);
1604 if (new_pg_num <= old_pg_num)
1605 return false;
1607 for (n = 1; ; n++) {
1608 int next_bit = n << (old_bits - 1);
1609 u32 s = next_bit | pgid->seed;
1611 if (s < old_pg_num || s == pgid->seed)
1612 continue;
1613 if (s >= new_pg_num)
1614 break;
1616 s = ceph_stable_mod(s, old_pg_num, old_mask);
1617 if (s == pgid->seed)
1618 return true;
1621 return false;
1624 bool ceph_is_new_interval(const struct ceph_osds *old_acting,
1625 const struct ceph_osds *new_acting,
1626 const struct ceph_osds *old_up,
1627 const struct ceph_osds *new_up,
1628 int old_size,
1629 int new_size,
1630 int old_min_size,
1631 int new_min_size,
1632 u32 old_pg_num,
1633 u32 new_pg_num,
1634 bool old_sort_bitwise,
1635 bool new_sort_bitwise,
1636 const struct ceph_pg *pgid)
1638 return !osds_equal(old_acting, new_acting) ||
1639 !osds_equal(old_up, new_up) ||
1640 old_size != new_size ||
1641 old_min_size != new_min_size ||
1642 is_split(pgid, old_pg_num, new_pg_num) ||
1643 old_sort_bitwise != new_sort_bitwise;
1646 static int calc_pg_rank(int osd, const struct ceph_osds *acting)
1648 int i;
1650 for (i = 0; i < acting->size; i++) {
1651 if (acting->osds[i] == osd)
1652 return i;
1655 return -1;
1658 static bool primary_changed(const struct ceph_osds *old_acting,
1659 const struct ceph_osds *new_acting)
1661 if (!old_acting->size && !new_acting->size)
1662 return false; /* both still empty */
1664 if (!old_acting->size ^ !new_acting->size)
1665 return true; /* was empty, now not, or vice versa */
1667 if (old_acting->primary != new_acting->primary)
1668 return true; /* primary changed */
1670 if (calc_pg_rank(old_acting->primary, old_acting) !=
1671 calc_pg_rank(new_acting->primary, new_acting))
1672 return true;
1674 return false; /* same primary (tho replicas may have changed) */
1677 bool ceph_osds_changed(const struct ceph_osds *old_acting,
1678 const struct ceph_osds *new_acting,
1679 bool any_change)
1681 if (primary_changed(old_acting, new_acting))
1682 return true;
1684 if (any_change && !__osds_equal(old_acting, new_acting))
1685 return true;
1687 return false;
1691 * calculate file layout from given offset, length.
1692 * fill in correct oid, logical length, and object extent
1693 * offset, length.
1695 * for now, we write only a single su, until we can
1696 * pass a stride back to the caller.
1698 int ceph_calc_file_object_mapping(struct ceph_file_layout *layout,
1699 u64 off, u64 len,
1700 u64 *ono,
1701 u64 *oxoff, u64 *oxlen)
1703 u32 osize = le32_to_cpu(layout->fl_object_size);
1704 u32 su = le32_to_cpu(layout->fl_stripe_unit);
1705 u32 sc = le32_to_cpu(layout->fl_stripe_count);
1706 u32 bl, stripeno, stripepos, objsetno;
1707 u32 su_per_object;
1708 u64 t, su_offset;
1710 dout("mapping %llu~%llu osize %u fl_su %u\n", off, len,
1711 osize, su);
1712 if (su == 0 || sc == 0)
1713 goto invalid;
1714 su_per_object = osize / su;
1715 if (su_per_object == 0)
1716 goto invalid;
1717 dout("osize %u / su %u = su_per_object %u\n", osize, su,
1718 su_per_object);
1720 if ((su & ~PAGE_MASK) != 0)
1721 goto invalid;
1723 /* bl = *off / su; */
1724 t = off;
1725 do_div(t, su);
1726 bl = t;
1727 dout("off %llu / su %u = bl %u\n", off, su, bl);
1729 stripeno = bl / sc;
1730 stripepos = bl % sc;
1731 objsetno = stripeno / su_per_object;
1733 *ono = objsetno * sc + stripepos;
1734 dout("objset %u * sc %u = ono %u\n", objsetno, sc, (unsigned int)*ono);
1736 /* *oxoff = *off % layout->fl_stripe_unit; # offset in su */
1737 t = off;
1738 su_offset = do_div(t, su);
1739 *oxoff = su_offset + (stripeno % su_per_object) * su;
1742 * Calculate the length of the extent being written to the selected
1743 * object. This is the minimum of the full length requested (len) or
1744 * the remainder of the current stripe being written to.
1746 *oxlen = min_t(u64, len, su - su_offset);
1748 dout(" obj extent %llu~%llu\n", *oxoff, *oxlen);
1749 return 0;
1751 invalid:
1752 dout(" invalid layout\n");
1753 *ono = 0;
1754 *oxoff = 0;
1755 *oxlen = 0;
1756 return -EINVAL;
1758 EXPORT_SYMBOL(ceph_calc_file_object_mapping);
1761 * Map an object into a PG.
1763 * Should only be called with target_oid and target_oloc (as opposed to
1764 * base_oid and base_oloc), since tiering isn't taken into account.
1766 int ceph_object_locator_to_pg(struct ceph_osdmap *osdmap,
1767 struct ceph_object_id *oid,
1768 struct ceph_object_locator *oloc,
1769 struct ceph_pg *raw_pgid)
1771 struct ceph_pg_pool_info *pi;
1773 pi = ceph_pg_pool_by_id(osdmap, oloc->pool);
1774 if (!pi)
1775 return -ENOENT;
1777 raw_pgid->pool = oloc->pool;
1778 raw_pgid->seed = ceph_str_hash(pi->object_hash, oid->name,
1779 oid->name_len);
1781 dout("%s %s -> raw_pgid %llu.%x\n", __func__, oid->name,
1782 raw_pgid->pool, raw_pgid->seed);
1783 return 0;
1785 EXPORT_SYMBOL(ceph_object_locator_to_pg);
1788 * Map a raw PG (full precision ps) into an actual PG.
1790 static void raw_pg_to_pg(struct ceph_pg_pool_info *pi,
1791 const struct ceph_pg *raw_pgid,
1792 struct ceph_pg *pgid)
1794 pgid->pool = raw_pgid->pool;
1795 pgid->seed = ceph_stable_mod(raw_pgid->seed, pi->pg_num,
1796 pi->pg_num_mask);
1800 * Map a raw PG (full precision ps) into a placement ps (placement
1801 * seed). Include pool id in that value so that different pools don't
1802 * use the same seeds.
1804 static u32 raw_pg_to_pps(struct ceph_pg_pool_info *pi,
1805 const struct ceph_pg *raw_pgid)
1807 if (pi->flags & CEPH_POOL_FLAG_HASHPSPOOL) {
1808 /* hash pool id and seed so that pool PGs do not overlap */
1809 return crush_hash32_2(CRUSH_HASH_RJENKINS1,
1810 ceph_stable_mod(raw_pgid->seed,
1811 pi->pgp_num,
1812 pi->pgp_num_mask),
1813 raw_pgid->pool);
1814 } else {
1816 * legacy behavior: add ps and pool together. this is
1817 * not a great approach because the PGs from each pool
1818 * will overlap on top of each other: 0.5 == 1.4 ==
1819 * 2.3 == ...
1821 return ceph_stable_mod(raw_pgid->seed, pi->pgp_num,
1822 pi->pgp_num_mask) +
1823 (unsigned)raw_pgid->pool;
1827 static int do_crush(struct ceph_osdmap *map, int ruleno, int x,
1828 int *result, int result_max,
1829 const __u32 *weight, int weight_max)
1831 int r;
1833 BUG_ON(result_max > CEPH_PG_MAX_SIZE);
1835 mutex_lock(&map->crush_scratch_mutex);
1836 r = crush_do_rule(map->crush, ruleno, x, result, result_max,
1837 weight, weight_max, map->crush_scratch_ary);
1838 mutex_unlock(&map->crush_scratch_mutex);
1840 return r;
1844 * Calculate raw set (CRUSH output) for given PG. The result may
1845 * contain nonexistent OSDs. ->primary is undefined for a raw set.
1847 * Placement seed (CRUSH input) is returned through @ppps.
1849 static void pg_to_raw_osds(struct ceph_osdmap *osdmap,
1850 struct ceph_pg_pool_info *pi,
1851 const struct ceph_pg *raw_pgid,
1852 struct ceph_osds *raw,
1853 u32 *ppps)
1855 u32 pps = raw_pg_to_pps(pi, raw_pgid);
1856 int ruleno;
1857 int len;
1859 ceph_osds_init(raw);
1860 if (ppps)
1861 *ppps = pps;
1863 ruleno = crush_find_rule(osdmap->crush, pi->crush_ruleset, pi->type,
1864 pi->size);
1865 if (ruleno < 0) {
1866 pr_err("no crush rule: pool %lld ruleset %d type %d size %d\n",
1867 pi->id, pi->crush_ruleset, pi->type, pi->size);
1868 return;
1871 len = do_crush(osdmap, ruleno, pps, raw->osds,
1872 min_t(int, pi->size, ARRAY_SIZE(raw->osds)),
1873 osdmap->osd_weight, osdmap->max_osd);
1874 if (len < 0) {
1875 pr_err("error %d from crush rule %d: pool %lld ruleset %d type %d size %d\n",
1876 len, ruleno, pi->id, pi->crush_ruleset, pi->type,
1877 pi->size);
1878 return;
1881 raw->size = len;
1885 * Given raw set, calculate up set and up primary. By definition of an
1886 * up set, the result won't contain nonexistent or down OSDs.
1888 * This is done in-place - on return @set is the up set. If it's
1889 * empty, ->primary will remain undefined.
1891 static void raw_to_up_osds(struct ceph_osdmap *osdmap,
1892 struct ceph_pg_pool_info *pi,
1893 struct ceph_osds *set)
1895 int i;
1897 /* ->primary is undefined for a raw set */
1898 BUG_ON(set->primary != -1);
1900 if (ceph_can_shift_osds(pi)) {
1901 int removed = 0;
1903 /* shift left */
1904 for (i = 0; i < set->size; i++) {
1905 if (ceph_osd_is_down(osdmap, set->osds[i])) {
1906 removed++;
1907 continue;
1909 if (removed)
1910 set->osds[i - removed] = set->osds[i];
1912 set->size -= removed;
1913 if (set->size > 0)
1914 set->primary = set->osds[0];
1915 } else {
1916 /* set down/dne devices to NONE */
1917 for (i = set->size - 1; i >= 0; i--) {
1918 if (ceph_osd_is_down(osdmap, set->osds[i]))
1919 set->osds[i] = CRUSH_ITEM_NONE;
1920 else
1921 set->primary = set->osds[i];
1926 static void apply_primary_affinity(struct ceph_osdmap *osdmap,
1927 struct ceph_pg_pool_info *pi,
1928 u32 pps,
1929 struct ceph_osds *up)
1931 int i;
1932 int pos = -1;
1935 * Do we have any non-default primary_affinity values for these
1936 * osds?
1938 if (!osdmap->osd_primary_affinity)
1939 return;
1941 for (i = 0; i < up->size; i++) {
1942 int osd = up->osds[i];
1944 if (osd != CRUSH_ITEM_NONE &&
1945 osdmap->osd_primary_affinity[osd] !=
1946 CEPH_OSD_DEFAULT_PRIMARY_AFFINITY) {
1947 break;
1950 if (i == up->size)
1951 return;
1954 * Pick the primary. Feed both the seed (for the pg) and the
1955 * osd into the hash/rng so that a proportional fraction of an
1956 * osd's pgs get rejected as primary.
1958 for (i = 0; i < up->size; i++) {
1959 int osd = up->osds[i];
1960 u32 aff;
1962 if (osd == CRUSH_ITEM_NONE)
1963 continue;
1965 aff = osdmap->osd_primary_affinity[osd];
1966 if (aff < CEPH_OSD_MAX_PRIMARY_AFFINITY &&
1967 (crush_hash32_2(CRUSH_HASH_RJENKINS1,
1968 pps, osd) >> 16) >= aff) {
1970 * We chose not to use this primary. Note it
1971 * anyway as a fallback in case we don't pick
1972 * anyone else, but keep looking.
1974 if (pos < 0)
1975 pos = i;
1976 } else {
1977 pos = i;
1978 break;
1981 if (pos < 0)
1982 return;
1984 up->primary = up->osds[pos];
1986 if (ceph_can_shift_osds(pi) && pos > 0) {
1987 /* move the new primary to the front */
1988 for (i = pos; i > 0; i--)
1989 up->osds[i] = up->osds[i - 1];
1990 up->osds[0] = up->primary;
1995 * Get pg_temp and primary_temp mappings for given PG.
1997 * Note that a PG may have none, only pg_temp, only primary_temp or
1998 * both pg_temp and primary_temp mappings. This means @temp isn't
1999 * always a valid OSD set on return: in the "only primary_temp" case,
2000 * @temp will have its ->primary >= 0 but ->size == 0.
2002 static void get_temp_osds(struct ceph_osdmap *osdmap,
2003 struct ceph_pg_pool_info *pi,
2004 const struct ceph_pg *raw_pgid,
2005 struct ceph_osds *temp)
2007 struct ceph_pg pgid;
2008 struct ceph_pg_mapping *pg;
2009 int i;
2011 raw_pg_to_pg(pi, raw_pgid, &pgid);
2012 ceph_osds_init(temp);
2014 /* pg_temp? */
2015 pg = __lookup_pg_mapping(&osdmap->pg_temp, pgid);
2016 if (pg) {
2017 for (i = 0; i < pg->pg_temp.len; i++) {
2018 if (ceph_osd_is_down(osdmap, pg->pg_temp.osds[i])) {
2019 if (ceph_can_shift_osds(pi))
2020 continue;
2022 temp->osds[temp->size++] = CRUSH_ITEM_NONE;
2023 } else {
2024 temp->osds[temp->size++] = pg->pg_temp.osds[i];
2028 /* apply pg_temp's primary */
2029 for (i = 0; i < temp->size; i++) {
2030 if (temp->osds[i] != CRUSH_ITEM_NONE) {
2031 temp->primary = temp->osds[i];
2032 break;
2037 /* primary_temp? */
2038 pg = __lookup_pg_mapping(&osdmap->primary_temp, pgid);
2039 if (pg)
2040 temp->primary = pg->primary_temp.osd;
2044 * Map a PG to its acting set as well as its up set.
2046 * Acting set is used for data mapping purposes, while up set can be
2047 * recorded for detecting interval changes and deciding whether to
2048 * resend a request.
2050 void ceph_pg_to_up_acting_osds(struct ceph_osdmap *osdmap,
2051 const struct ceph_pg *raw_pgid,
2052 struct ceph_osds *up,
2053 struct ceph_osds *acting)
2055 struct ceph_pg_pool_info *pi;
2056 u32 pps;
2058 pi = ceph_pg_pool_by_id(osdmap, raw_pgid->pool);
2059 if (!pi) {
2060 ceph_osds_init(up);
2061 ceph_osds_init(acting);
2062 goto out;
2065 pg_to_raw_osds(osdmap, pi, raw_pgid, up, &pps);
2066 raw_to_up_osds(osdmap, pi, up);
2067 apply_primary_affinity(osdmap, pi, pps, up);
2068 get_temp_osds(osdmap, pi, raw_pgid, acting);
2069 if (!acting->size) {
2070 memcpy(acting->osds, up->osds, up->size * sizeof(up->osds[0]));
2071 acting->size = up->size;
2072 if (acting->primary == -1)
2073 acting->primary = up->primary;
2075 out:
2076 WARN_ON(!osds_valid(up) || !osds_valid(acting));
2080 * Return acting primary for given PG, or -1 if none.
2082 int ceph_pg_to_acting_primary(struct ceph_osdmap *osdmap,
2083 const struct ceph_pg *raw_pgid)
2085 struct ceph_osds up, acting;
2087 ceph_pg_to_up_acting_osds(osdmap, raw_pgid, &up, &acting);
2088 return acting.primary;
2090 EXPORT_SYMBOL(ceph_pg_to_acting_primary);