xref: /linux/net/ceph/osdmap.c (revision b6ebbac51bedf9e98e837688bc838f400196da5e)
1 
2 #include <linux/ceph/ceph_debug.h>
3 
4 #include <linux/module.h>
5 #include <linux/slab.h>
6 #include <asm/div64.h>
7 
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>
13 
14 char *ceph_osdmap_state_str(char *str, int len, int state)
15 {
16 	if (!len)
17 		return str;
18 
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");
27 
28 	return str;
29 }
30 
31 /* maps */
32 
33 static int calc_bits_of(unsigned int t)
34 {
35 	int b = 0;
36 	while (t) {
37 		t = t >> 1;
38 		b++;
39 	}
40 	return b;
41 }
42 
43 /*
44  * the foo_mask is the smallest value 2^n-1 that is >= foo.
45  */
46 static void calc_pg_masks(struct ceph_pg_pool_info *pi)
47 {
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;
50 }
51 
52 /*
53  * decode crush map
54  */
55 static int crush_decode_uniform_bucket(void **p, void *end,
56 				       struct crush_bucket_uniform *b)
57 {
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;
64 }
65 
66 static int crush_decode_list_bucket(void **p, void *end,
67 				    struct crush_bucket_list *b)
68 {
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);
81 	}
82 	return 0;
83 bad:
84 	return -EINVAL;
85 }
86 
87 static int crush_decode_tree_bucket(void **p, void *end,
88 				    struct crush_bucket_tree *b)
89 {
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;
102 }
103 
104 static int crush_decode_straw_bucket(void **p, void *end,
105 				     struct crush_bucket_straw *b)
106 {
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);
119 	}
120 	return 0;
121 bad:
122 	return -EINVAL;
123 }
124 
125 static int crush_decode_straw2_bucket(void **p, void *end,
126 				      struct crush_bucket_straw2 *b)
127 {
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;
139 }
140 
141 static int skip_name_map(void **p, void *end)
142 {
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;
150 }
151         return 0;
152 bad:
153         return -EINVAL;
154 }
155 
156 static struct crush_map *crush_decode(void *pbyval, void *end)
157 {
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;
165 
166 	dout("crush_decode %p to %p len %d\n", *p, end, (int)(end - *p));
167 
168 	c = kzalloc(sizeof(*c), GFP_NOFS);
169 	if (c == NULL)
170 		return ERR_PTR(-ENOMEM);
171 
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;
177 
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;
184 	}
185 	c->max_buckets = ceph_decode_32(p);
186 	c->max_rules = ceph_decode_32(p);
187 	c->max_devices = ceph_decode_32(p);
188 
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;
195 
196 	/* buckets */
197 	for (i = 0; i < c->max_buckets; i++) {
198 		int size = 0;
199 		u32 alg;
200 		struct crush_bucket *b;
201 
202 		ceph_decode_32_safe(p, end, alg, bad);
203 		if (alg == 0) {
204 			c->buckets[i] = NULL;
205 			continue;
206 		}
207 		dout("crush_decode bucket %d off %x %p to %p\n",
208 		     i, (int)(*p-start), *p, end);
209 
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;
229 		}
230 		BUG_ON(size == 0);
231 		b = c->buckets[i] = kzalloc(size, GFP_NOFS);
232 		if (b == NULL)
233 			goto badmem;
234 
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);
242 
243 		dout("crush_decode bucket size %d off %x %p to %p\n",
244 		     b->size, (int)(*p-start), *p, end);
245 
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;
253 
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);
257 
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;
289 		}
290 	}
291 
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;
297 
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;
304 		}
305 
306 		dout("crush_decode rule %d off %x %p to %p\n",
307 		     i, (int)(*p-start), *p, end);
308 
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);
330 		}
331 	}
332 
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;
338         }
339 
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);
351 
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);
356 
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);
361 
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);
365 
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);
370 
371 done:
372 	dout("crush_decode success\n");
373 	return c;
374 
375 badmem:
376 	err = -ENOMEM;
377 bad:
378 	dout("crush_decode fail %d\n", err);
379 	crush_destroy(c);
380 	return ERR_PTR(err);
381 }
382 
383 int ceph_pg_compare(const struct ceph_pg *lhs, const struct ceph_pg *rhs)
384 {
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;
393 
394 	return 0;
395 }
396 
397 /*
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)
400  */
401 static int __insert_pg_mapping(struct ceph_pg_mapping *new,
402 			       struct rb_root *root)
403 {
404 	struct rb_node **p = &root->rb_node;
405 	struct rb_node *parent = NULL;
406 	struct ceph_pg_mapping *pg = NULL;
407 	int c;
408 
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;
420 	}
421 
422 	rb_link_node(&new->node, parent, p);
423 	rb_insert_color(&new->node, root);
424 	return 0;
425 }
426 
427 static struct ceph_pg_mapping *__lookup_pg_mapping(struct rb_root *root,
428 						   struct ceph_pg pgid)
429 {
430 	struct rb_node *n = root->rb_node;
431 	struct ceph_pg_mapping *pg;
432 	int c;
433 
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;
445 		}
446 	}
447 	return NULL;
448 }
449 
450 static int __remove_pg_mapping(struct rb_root *root, struct ceph_pg pgid)
451 {
452 	struct ceph_pg_mapping *pg = __lookup_pg_mapping(root, pgid);
453 
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;
460 	}
461 	dout("__remove_pg_mapping %lld.%x dne\n", pgid.pool, pgid.seed);
462 	return -ENOENT;
463 }
464 
465 /*
466  * rbtree of pg pool info
467  */
468 static int __insert_pg_pool(struct rb_root *root, struct ceph_pg_pool_info *new)
469 {
470 	struct rb_node **p = &root->rb_node;
471 	struct rb_node *parent = NULL;
472 	struct ceph_pg_pool_info *pi = NULL;
473 
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;
483 	}
484 
485 	rb_link_node(&new->node, parent, p);
486 	rb_insert_color(&new->node, root);
487 	return 0;
488 }
489 
490 static struct ceph_pg_pool_info *__lookup_pg_pool(struct rb_root *root, u64 id)
491 {
492 	struct ceph_pg_pool_info *pi;
493 	struct rb_node *n = root->rb_node;
494 
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;
503 	}
504 	return NULL;
505 }
506 
507 struct ceph_pg_pool_info *ceph_pg_pool_by_id(struct ceph_osdmap *map, u64 id)
508 {
509 	return __lookup_pg_pool(&map->pg_pools, id);
510 }
511 
512 const char *ceph_pg_pool_name_by_id(struct ceph_osdmap *map, u64 id)
513 {
514 	struct ceph_pg_pool_info *pi;
515 
516 	if (id == CEPH_NOPOOL)
517 		return NULL;
518 
519 	if (WARN_ON_ONCE(id > (u64) INT_MAX))
520 		return NULL;
521 
522 	pi = __lookup_pg_pool(&map->pg_pools, (int) id);
523 
524 	return pi ? pi->name : NULL;
525 }
526 EXPORT_SYMBOL(ceph_pg_pool_name_by_id);
527 
528 int ceph_pg_poolid_by_name(struct ceph_osdmap *map, const char *name)
529 {
530 	struct rb_node *rbp;
531 
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;
537 	}
538 	return -ENOENT;
539 }
540 EXPORT_SYMBOL(ceph_pg_poolid_by_name);
541 
542 static void __remove_pg_pool(struct rb_root *root, struct ceph_pg_pool_info *pi)
543 {
544 	rb_erase(&pi->node, root);
545 	kfree(pi->name);
546 	kfree(pi);
547 }
548 
549 static int decode_pool(void **p, void *end, struct ceph_pg_pool_info *pi)
550 {
551 	u8 ev, cv;
552 	unsigned len, num;
553 	void *pool_end;
554 
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;
561 	}
562 	if (cv > 9) {
563 		pr_warn("got v %d cv %d > 9 of ceph_pg_pool\n", ev, cv);
564 		return -EINVAL;
565 	}
566 	len = ceph_decode_32(p);
567 	ceph_decode_need(p, end, len, bad);
568 	pool_end = *p + len;
569 
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);
574 
575 	pi->pg_num = ceph_decode_32(p);
576 	pi->pgp_num = ceph_decode_32(p);
577 
578 	*p += 4 + 4;  /* skip lpg* */
579 	*p += 4;      /* skip last_change */
580 	*p += 8 + 4;  /* skip snap_seq, snap_epoch */
581 
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;
589 	}
590 
591 	/* skip removed_snaps */
592 	num = ceph_decode_32(p);
593 	*p += num * (8 + 8);
594 
595 	*p += 8;  /* skip auid */
596 	pi->flags = ceph_decode_64(p);
597 	*p += 4;  /* skip crash_replay_interval */
598 
599 	if (ev >= 7)
600 		pi->min_size = ceph_decode_8(p);
601 	else
602 		pi->min_size = pi->size - pi->size / 2;
603 
604 	if (ev >= 8)
605 		*p += 8 + 8;  /* skip quota_max_* */
606 
607 	if (ev >= 9) {
608 		/* skip tiers */
609 		num = ceph_decode_32(p);
610 		*p += num * 8;
611 
612 		*p += 8;  /* skip tier_of */
613 		*p += 1;  /* skip cache_mode */
614 
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;
620 	}
621 
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 */
630 		}
631 	}
632 
633 	if (ev >= 11) {
634 		/* skip hit_set_params */
635 		*p += 1 + 1; /* versions */
636 		len = ceph_decode_32(p);
637 		*p += len;
638 
639 		*p += 4; /* skip hit_set_period */
640 		*p += 4; /* skip hit_set_count */
641 	}
642 
643 	if (ev >= 12)
644 		*p += 4; /* skip stripe_width */
645 
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 */
653 	}
654 
655 	if (ev >=  14) {
656 		/* skip erasure_code_profile */
657 		len = ceph_decode_32(p);
658 		*p += len;
659 	}
660 
661 	if (ev >= 15)
662 		pi->last_force_request_resend = ceph_decode_32(p);
663 	else
664 		pi->last_force_request_resend = 0;
665 
666 	/* ignore the rest */
667 
668 	*p = pool_end;
669 	calc_pg_masks(pi);
670 	return 0;
671 
672 bad:
673 	return -EINVAL;
674 }
675 
676 static int decode_pool_names(void **p, void *end, struct ceph_osdmap *map)
677 {
678 	struct ceph_pg_pool_info *pi;
679 	u32 num, len;
680 	u64 pool;
681 
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);
692 
693 			if (!name)
694 				return -ENOMEM;
695 			kfree(pi->name);
696 			pi->name = name;
697 			dout("  name is %s\n", pi->name);
698 		}
699 		*p += len;
700 	}
701 	return 0;
702 
703 bad:
704 	return -EINVAL;
705 }
706 
707 /*
708  * osd map
709  */
710 struct ceph_osdmap *ceph_osdmap_alloc(void)
711 {
712 	struct ceph_osdmap *map;
713 
714 	map = kzalloc(sizeof(*map), GFP_NOIO);
715 	if (!map)
716 		return NULL;
717 
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);
723 
724 	return map;
725 }
726 
727 void ceph_osdmap_destroy(struct ceph_osdmap *map)
728 {
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);
738 	}
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);
745 	}
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);
751 	}
752 	kfree(map->osd_state);
753 	kfree(map->osd_weight);
754 	kfree(map->osd_addr);
755 	kfree(map->osd_primary_affinity);
756 	kfree(map);
757 }
758 
759 /*
760  * Adjust max_osd value, (re)allocate arrays.
761  *
762  * The new elements are properly initialized.
763  */
764 static int osdmap_set_max_osd(struct ceph_osdmap *map, int max)
765 {
766 	u8 *state;
767 	u32 *weight;
768 	struct ceph_entity_addr *addr;
769 	int i;
770 
771 	state = krealloc(map->osd_state, max*sizeof(*state), GFP_NOFS);
772 	if (!state)
773 		return -ENOMEM;
774 	map->osd_state = state;
775 
776 	weight = krealloc(map->osd_weight, max*sizeof(*weight), GFP_NOFS);
777 	if (!weight)
778 		return -ENOMEM;
779 	map->osd_weight = weight;
780 
781 	addr = krealloc(map->osd_addr, max*sizeof(*addr), GFP_NOFS);
782 	if (!addr)
783 		return -ENOMEM;
784 	map->osd_addr = addr;
785 
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));
790 	}
791 
792 	if (map->osd_primary_affinity) {
793 		u32 *affinity;
794 
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;
800 
801 		for (i = map->max_osd; i < max; i++)
802 			map->osd_primary_affinity[i] =
803 			    CEPH_OSD_DEFAULT_PRIMARY_AFFINITY;
804 	}
805 
806 	map->max_osd = max;
807 
808 	return 0;
809 }
810 
811 #define OSDMAP_WRAPPER_COMPAT_VER	7
812 #define OSDMAP_CLIENT_DATA_COMPAT_VER	1
813 
814 /*
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.
818  */
819 static int get_osdmap_client_data_v(void **p, void *end,
820 				    const char *prefix, u8 *v)
821 {
822 	u8 struct_v;
823 
824 	ceph_decode_8_safe(p, end, struct_v, e_inval);
825 	if (struct_v >= 7) {
826 		u8 struct_compat;
827 
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;
834 		}
835 		*p += 4; /* ignore wrapper struct_len */
836 
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;
844 		}
845 		*p += 4; /* ignore client data struct_len */
846 	} else {
847 		u16 version;
848 
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;
855 		}
856 
857 		/* old osdmap enconding */
858 		struct_v = 0;
859 	}
860 
861 	*v = struct_v;
862 	return 0;
863 
864 e_inval:
865 	return -EINVAL;
866 }
867 
868 static int __decode_pools(void **p, void *end, struct ceph_osdmap *map,
869 			  bool incremental)
870 {
871 	u32 n;
872 
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;
878 
879 		ceph_decode_64_safe(p, end, pool, e_inval);
880 
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;
886 
887 			pi->id = pool;
888 
889 			ret = __insert_pg_pool(&map->pg_pools, pi);
890 			if (ret) {
891 				kfree(pi);
892 				return ret;
893 			}
894 		}
895 
896 		ret = decode_pool(p, end, pi);
897 		if (ret)
898 			return ret;
899 	}
900 
901 	return 0;
902 
903 e_inval:
904 	return -EINVAL;
905 }
906 
907 static int decode_pools(void **p, void *end, struct ceph_osdmap *map)
908 {
909 	return __decode_pools(p, end, map, false);
910 }
911 
912 static int decode_new_pools(void **p, void *end, struct ceph_osdmap *map)
913 {
914 	return __decode_pools(p, end, map, true);
915 }
916 
917 static int __decode_pg_temp(void **p, void *end, struct ceph_osdmap *map,
918 			    bool incremental)
919 {
920 	u32 n;
921 
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;
927 
928 		ret = ceph_decode_pgid(p, end, &pgid);
929 		if (ret)
930 			return ret;
931 
932 		ceph_decode_32_safe(p, end, len, e_inval);
933 
934 		ret = __remove_pg_mapping(&map->pg_temp, pgid);
935 		BUG_ON(!incremental && ret != -ENOENT);
936 
937 		if (!incremental || len > 0) {
938 			struct ceph_pg_mapping *pg;
939 
940 			ceph_decode_need(p, end, len*sizeof(u32), e_inval);
941 
942 			if (len > (UINT_MAX - sizeof(*pg)) / sizeof(u32))
943 				return -EINVAL;
944 
945 			pg = kzalloc(sizeof(*pg) + len*sizeof(u32), GFP_NOFS);
946 			if (!pg)
947 				return -ENOMEM;
948 
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);
953 
954 			ret = __insert_pg_mapping(pg, &map->pg_temp);
955 			if (ret) {
956 				kfree(pg);
957 				return ret;
958 			}
959 		}
960 	}
961 
962 	return 0;
963 
964 e_inval:
965 	return -EINVAL;
966 }
967 
968 static int decode_pg_temp(void **p, void *end, struct ceph_osdmap *map)
969 {
970 	return __decode_pg_temp(p, end, map, false);
971 }
972 
973 static int decode_new_pg_temp(void **p, void *end, struct ceph_osdmap *map)
974 {
975 	return __decode_pg_temp(p, end, map, true);
976 }
977 
978 static int __decode_primary_temp(void **p, void *end, struct ceph_osdmap *map,
979 				 bool incremental)
980 {
981 	u32 n;
982 
983 	ceph_decode_32_safe(p, end, n, e_inval);
984 	while (n--) {
985 		struct ceph_pg pgid;
986 		u32 osd;
987 		int ret;
988 
989 		ret = ceph_decode_pgid(p, end, &pgid);
990 		if (ret)
991 			return ret;
992 
993 		ceph_decode_32_safe(p, end, osd, e_inval);
994 
995 		ret = __remove_pg_mapping(&map->primary_temp, pgid);
996 		BUG_ON(!incremental && ret != -ENOENT);
997 
998 		if (!incremental || osd != (u32)-1) {
999 			struct ceph_pg_mapping *pg;
1000 
1001 			pg = kzalloc(sizeof(*pg), GFP_NOFS);
1002 			if (!pg)
1003 				return -ENOMEM;
1004 
1005 			pg->pgid = pgid;
1006 			pg->primary_temp.osd = osd;
1007 
1008 			ret = __insert_pg_mapping(pg, &map->primary_temp);
1009 			if (ret) {
1010 				kfree(pg);
1011 				return ret;
1012 			}
1013 		}
1014 	}
1015 
1016 	return 0;
1017 
1018 e_inval:
1019 	return -EINVAL;
1020 }
1021 
1022 static int decode_primary_temp(void **p, void *end, struct ceph_osdmap *map)
1023 {
1024 	return __decode_primary_temp(p, end, map, false);
1025 }
1026 
1027 static int decode_new_primary_temp(void **p, void *end,
1028 				   struct ceph_osdmap *map)
1029 {
1030 	return __decode_primary_temp(p, end, map, true);
1031 }
1032 
1033 u32 ceph_get_primary_affinity(struct ceph_osdmap *map, int osd)
1034 {
1035 	BUG_ON(osd >= map->max_osd);
1036 
1037 	if (!map->osd_primary_affinity)
1038 		return CEPH_OSD_DEFAULT_PRIMARY_AFFINITY;
1039 
1040 	return map->osd_primary_affinity[osd];
1041 }
1042 
1043 static int set_primary_affinity(struct ceph_osdmap *map, int osd, u32 aff)
1044 {
1045 	BUG_ON(osd >= map->max_osd);
1046 
1047 	if (!map->osd_primary_affinity) {
1048 		int i;
1049 
1050 		map->osd_primary_affinity = kmalloc(map->max_osd*sizeof(u32),
1051 						    GFP_NOFS);
1052 		if (!map->osd_primary_affinity)
1053 			return -ENOMEM;
1054 
1055 		for (i = 0; i < map->max_osd; i++)
1056 			map->osd_primary_affinity[i] =
1057 			    CEPH_OSD_DEFAULT_PRIMARY_AFFINITY;
1058 	}
1059 
1060 	map->osd_primary_affinity[osd] = aff;
1061 
1062 	return 0;
1063 }
1064 
1065 static int decode_primary_affinity(void **p, void *end,
1066 				   struct ceph_osdmap *map)
1067 {
1068 	u32 len, i;
1069 
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;
1075 	}
1076 	if (len != map->max_osd)
1077 		goto e_inval;
1078 
1079 	ceph_decode_need(p, end, map->max_osd*sizeof(u32), e_inval);
1080 
1081 	for (i = 0; i < map->max_osd; i++) {
1082 		int ret;
1083 
1084 		ret = set_primary_affinity(map, i, ceph_decode_32(p));
1085 		if (ret)
1086 			return ret;
1087 	}
1088 
1089 	return 0;
1090 
1091 e_inval:
1092 	return -EINVAL;
1093 }
1094 
1095 static int decode_new_primary_affinity(void **p, void *end,
1096 				       struct ceph_osdmap *map)
1097 {
1098 	u32 n;
1099 
1100 	ceph_decode_32_safe(p, end, n, e_inval);
1101 	while (n--) {
1102 		u32 osd, aff;
1103 		int ret;
1104 
1105 		ceph_decode_32_safe(p, end, osd, e_inval);
1106 		ceph_decode_32_safe(p, end, aff, e_inval);
1107 
1108 		ret = set_primary_affinity(map, osd, aff);
1109 		if (ret)
1110 			return ret;
1111 
1112 		pr_info("osd%d primary-affinity 0x%x\n", osd, aff);
1113 	}
1114 
1115 	return 0;
1116 
1117 e_inval:
1118 	return -EINVAL;
1119 }
1120 
1121 /*
1122  * decode a full map.
1123  */
1124 static int osdmap_decode(void **p, void *end, struct ceph_osdmap *map)
1125 {
1126 	u8 struct_v;
1127 	u32 epoch = 0;
1128 	void *start = *p;
1129 	u32 max;
1130 	u32 len, i;
1131 	int err;
1132 
1133 	dout("%s %p to %p len %d\n", __func__, *p, end, (int)(end - *p));
1134 
1135 	err = get_osdmap_client_data_v(p, end, "full", &struct_v);
1136 	if (err)
1137 		goto bad;
1138 
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));
1146 
1147 	/* pools */
1148 	err = decode_pools(p, end, map);
1149 	if (err)
1150 		goto bad;
1151 
1152 	/* pool_name */
1153 	err = decode_pool_names(p, end, map);
1154 	if (err)
1155 		goto bad;
1156 
1157 	ceph_decode_32_safe(p, end, map->pool_max, e_inval);
1158 
1159 	ceph_decode_32_safe(p, end, map->flags, e_inval);
1160 
1161 	/* max_osd */
1162 	ceph_decode_32_safe(p, end, max, e_inval);
1163 
1164 	/* (re)alloc osd arrays */
1165 	err = osdmap_set_max_osd(map, max);
1166 	if (err)
1167 		goto bad;
1168 
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);
1173 
1174 	if (ceph_decode_32(p) != map->max_osd)
1175 		goto e_inval;
1176 
1177 	ceph_decode_copy(p, map->osd_state, map->max_osd);
1178 
1179 	if (ceph_decode_32(p) != map->max_osd)
1180 		goto e_inval;
1181 
1182 	for (i = 0; i < map->max_osd; i++)
1183 		map->osd_weight[i] = ceph_decode_32(p);
1184 
1185 	if (ceph_decode_32(p) != map->max_osd)
1186 		goto e_inval;
1187 
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]);
1191 
1192 	/* pg_temp */
1193 	err = decode_pg_temp(p, end, map);
1194 	if (err)
1195 		goto bad;
1196 
1197 	/* primary_temp */
1198 	if (struct_v >= 1) {
1199 		err = decode_primary_temp(p, end, map);
1200 		if (err)
1201 			goto bad;
1202 	}
1203 
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;
1213 	}
1214 
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;
1222 	}
1223 	*p += len;
1224 
1225 	/* ignore the rest */
1226 	*p = end;
1227 
1228 	dout("full osdmap epoch %d max_osd %d\n", map->epoch, map->max_osd);
1229 	return 0;
1230 
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;
1240 }
1241 
1242 /*
1243  * Allocate and decode a full map.
1244  */
1245 struct ceph_osdmap *ceph_osdmap_decode(void **p, void *end)
1246 {
1247 	struct ceph_osdmap *map;
1248 	int ret;
1249 
1250 	map = ceph_osdmap_alloc();
1251 	if (!map)
1252 		return ERR_PTR(-ENOMEM);
1253 
1254 	ret = osdmap_decode(p, end, map);
1255 	if (ret) {
1256 		ceph_osdmap_destroy(map);
1257 		return ERR_PTR(ret);
1258 	}
1259 
1260 	return map;
1261 }
1262 
1263 /*
1264  * Encoding order is (new_up_client, new_state, new_weight).  Need to
1265  * apply in the (new_weight, new_state, new_up_client) order, because
1266  * an incremental map may look like e.g.
1267  *
1268  *     new_up_client: { osd=6, addr=... } # set osd_state and addr
1269  *     new_state: { osd=6, xorstate=EXISTS } # clear osd_state
1270  */
1271 static int decode_new_up_state_weight(void **p, void *end,
1272 				      struct ceph_osdmap *map)
1273 {
1274 	void *new_up_client;
1275 	void *new_state;
1276 	void *new_weight_end;
1277 	u32 len;
1278 
1279 	new_up_client = *p;
1280 	ceph_decode_32_safe(p, end, len, e_inval);
1281 	len *= sizeof(u32) + sizeof(struct ceph_entity_addr);
1282 	ceph_decode_need(p, end, len, e_inval);
1283 	*p += len;
1284 
1285 	new_state = *p;
1286 	ceph_decode_32_safe(p, end, len, e_inval);
1287 	len *= sizeof(u32) + sizeof(u8);
1288 	ceph_decode_need(p, end, len, e_inval);
1289 	*p += len;
1290 
1291 	/* new_weight */
1292 	ceph_decode_32_safe(p, end, len, e_inval);
1293 	while (len--) {
1294 		s32 osd;
1295 		u32 w;
1296 
1297 		ceph_decode_need(p, end, 2*sizeof(u32), e_inval);
1298 		osd = ceph_decode_32(p);
1299 		w = ceph_decode_32(p);
1300 		BUG_ON(osd >= map->max_osd);
1301 		pr_info("osd%d weight 0x%x %s\n", osd, w,
1302 		     w == CEPH_OSD_IN ? "(in)" :
1303 		     (w == CEPH_OSD_OUT ? "(out)" : ""));
1304 		map->osd_weight[osd] = w;
1305 
1306 		/*
1307 		 * If we are marking in, set the EXISTS, and clear the
1308 		 * AUTOOUT and NEW bits.
1309 		 */
1310 		if (w) {
1311 			map->osd_state[osd] |= CEPH_OSD_EXISTS;
1312 			map->osd_state[osd] &= ~(CEPH_OSD_AUTOOUT |
1313 						 CEPH_OSD_NEW);
1314 		}
1315 	}
1316 	new_weight_end = *p;
1317 
1318 	/* new_state (up/down) */
1319 	*p = new_state;
1320 	len = ceph_decode_32(p);
1321 	while (len--) {
1322 		s32 osd;
1323 		u8 xorstate;
1324 		int ret;
1325 
1326 		osd = ceph_decode_32(p);
1327 		xorstate = ceph_decode_8(p);
1328 		if (xorstate == 0)
1329 			xorstate = CEPH_OSD_UP;
1330 		BUG_ON(osd >= map->max_osd);
1331 		if ((map->osd_state[osd] & CEPH_OSD_UP) &&
1332 		    (xorstate & CEPH_OSD_UP))
1333 			pr_info("osd%d down\n", osd);
1334 		if ((map->osd_state[osd] & CEPH_OSD_EXISTS) &&
1335 		    (xorstate & CEPH_OSD_EXISTS)) {
1336 			pr_info("osd%d does not exist\n", osd);
1337 			map->osd_weight[osd] = CEPH_OSD_IN;
1338 			ret = set_primary_affinity(map, osd,
1339 						   CEPH_OSD_DEFAULT_PRIMARY_AFFINITY);
1340 			if (ret)
1341 				return ret;
1342 			memset(map->osd_addr + osd, 0, sizeof(*map->osd_addr));
1343 			map->osd_state[osd] = 0;
1344 		} else {
1345 			map->osd_state[osd] ^= xorstate;
1346 		}
1347 	}
1348 
1349 	/* new_up_client */
1350 	*p = new_up_client;
1351 	len = ceph_decode_32(p);
1352 	while (len--) {
1353 		s32 osd;
1354 		struct ceph_entity_addr addr;
1355 
1356 		osd = ceph_decode_32(p);
1357 		ceph_decode_copy(p, &addr, sizeof(addr));
1358 		ceph_decode_addr(&addr);
1359 		BUG_ON(osd >= map->max_osd);
1360 		pr_info("osd%d up\n", osd);
1361 		map->osd_state[osd] |= CEPH_OSD_EXISTS | CEPH_OSD_UP;
1362 		map->osd_addr[osd] = addr;
1363 	}
1364 
1365 	*p = new_weight_end;
1366 	return 0;
1367 
1368 e_inval:
1369 	return -EINVAL;
1370 }
1371 
1372 /*
1373  * decode and apply an incremental map update.
1374  */
1375 struct ceph_osdmap *osdmap_apply_incremental(void **p, void *end,
1376 					     struct ceph_osdmap *map)
1377 {
1378 	struct crush_map *newcrush = NULL;
1379 	struct ceph_fsid fsid;
1380 	u32 epoch = 0;
1381 	struct ceph_timespec modified;
1382 	s32 len;
1383 	u64 pool;
1384 	__s64 new_pool_max;
1385 	__s32 new_flags, max;
1386 	void *start = *p;
1387 	int err;
1388 	u8 struct_v;
1389 
1390 	dout("%s %p to %p len %d\n", __func__, *p, end, (int)(end - *p));
1391 
1392 	err = get_osdmap_client_data_v(p, end, "inc", &struct_v);
1393 	if (err)
1394 		goto bad;
1395 
1396 	/* fsid, epoch, modified, new_pool_max, new_flags */
1397 	ceph_decode_need(p, end, sizeof(fsid) + sizeof(u32) + sizeof(modified) +
1398 			 sizeof(u64) + sizeof(u32), e_inval);
1399 	ceph_decode_copy(p, &fsid, sizeof(fsid));
1400 	epoch = ceph_decode_32(p);
1401 	BUG_ON(epoch != map->epoch+1);
1402 	ceph_decode_copy(p, &modified, sizeof(modified));
1403 	new_pool_max = ceph_decode_64(p);
1404 	new_flags = ceph_decode_32(p);
1405 
1406 	/* full map? */
1407 	ceph_decode_32_safe(p, end, len, e_inval);
1408 	if (len > 0) {
1409 		dout("apply_incremental full map len %d, %p to %p\n",
1410 		     len, *p, end);
1411 		return ceph_osdmap_decode(p, min(*p+len, end));
1412 	}
1413 
1414 	/* new crush? */
1415 	ceph_decode_32_safe(p, end, len, e_inval);
1416 	if (len > 0) {
1417 		newcrush = crush_decode(*p, min(*p+len, end));
1418 		if (IS_ERR(newcrush)) {
1419 			err = PTR_ERR(newcrush);
1420 			newcrush = NULL;
1421 			goto bad;
1422 		}
1423 		*p += len;
1424 	}
1425 
1426 	/* new flags? */
1427 	if (new_flags >= 0)
1428 		map->flags = new_flags;
1429 	if (new_pool_max >= 0)
1430 		map->pool_max = new_pool_max;
1431 
1432 	/* new max? */
1433 	ceph_decode_32_safe(p, end, max, e_inval);
1434 	if (max >= 0) {
1435 		err = osdmap_set_max_osd(map, max);
1436 		if (err)
1437 			goto bad;
1438 	}
1439 
1440 	map->epoch++;
1441 	map->modified = modified;
1442 	if (newcrush) {
1443 		if (map->crush)
1444 			crush_destroy(map->crush);
1445 		map->crush = newcrush;
1446 		newcrush = NULL;
1447 	}
1448 
1449 	/* new_pools */
1450 	err = decode_new_pools(p, end, map);
1451 	if (err)
1452 		goto bad;
1453 
1454 	/* new_pool_names */
1455 	err = decode_pool_names(p, end, map);
1456 	if (err)
1457 		goto bad;
1458 
1459 	/* old_pool */
1460 	ceph_decode_32_safe(p, end, len, e_inval);
1461 	while (len--) {
1462 		struct ceph_pg_pool_info *pi;
1463 
1464 		ceph_decode_64_safe(p, end, pool, e_inval);
1465 		pi = __lookup_pg_pool(&map->pg_pools, pool);
1466 		if (pi)
1467 			__remove_pg_pool(&map->pg_pools, pi);
1468 	}
1469 
1470 	/* new_up_client, new_state, new_weight */
1471 	err = decode_new_up_state_weight(p, end, map);
1472 	if (err)
1473 		goto bad;
1474 
1475 	/* new_pg_temp */
1476 	err = decode_new_pg_temp(p, end, map);
1477 	if (err)
1478 		goto bad;
1479 
1480 	/* new_primary_temp */
1481 	if (struct_v >= 1) {
1482 		err = decode_new_primary_temp(p, end, map);
1483 		if (err)
1484 			goto bad;
1485 	}
1486 
1487 	/* new_primary_affinity */
1488 	if (struct_v >= 2) {
1489 		err = decode_new_primary_affinity(p, end, map);
1490 		if (err)
1491 			goto bad;
1492 	}
1493 
1494 	/* ignore the rest */
1495 	*p = end;
1496 
1497 	dout("inc osdmap epoch %d max_osd %d\n", map->epoch, map->max_osd);
1498 	return map;
1499 
1500 e_inval:
1501 	err = -EINVAL;
1502 bad:
1503 	pr_err("corrupt inc osdmap (%d) epoch %d off %d (%p of %p-%p)\n",
1504 	       err, epoch, (int)(*p - start), *p, start, end);
1505 	print_hex_dump(KERN_DEBUG, "osdmap: ",
1506 		       DUMP_PREFIX_OFFSET, 16, 1,
1507 		       start, end - start, true);
1508 	if (newcrush)
1509 		crush_destroy(newcrush);
1510 	return ERR_PTR(err);
1511 }
1512 
1513 void ceph_oloc_copy(struct ceph_object_locator *dest,
1514 		    const struct ceph_object_locator *src)
1515 {
1516 	WARN_ON(!ceph_oloc_empty(dest));
1517 	WARN_ON(dest->pool_ns); /* empty() only covers ->pool */
1518 
1519 	dest->pool = src->pool;
1520 	if (src->pool_ns)
1521 		dest->pool_ns = ceph_get_string(src->pool_ns);
1522 }
1523 EXPORT_SYMBOL(ceph_oloc_copy);
1524 
1525 void ceph_oloc_destroy(struct ceph_object_locator *oloc)
1526 {
1527 	ceph_put_string(oloc->pool_ns);
1528 }
1529 EXPORT_SYMBOL(ceph_oloc_destroy);
1530 
1531 void ceph_oid_copy(struct ceph_object_id *dest,
1532 		   const struct ceph_object_id *src)
1533 {
1534 	WARN_ON(!ceph_oid_empty(dest));
1535 
1536 	if (src->name != src->inline_name) {
1537 		/* very rare, see ceph_object_id definition */
1538 		dest->name = kmalloc(src->name_len + 1,
1539 				     GFP_NOIO | __GFP_NOFAIL);
1540 	}
1541 
1542 	memcpy(dest->name, src->name, src->name_len + 1);
1543 	dest->name_len = src->name_len;
1544 }
1545 EXPORT_SYMBOL(ceph_oid_copy);
1546 
1547 static __printf(2, 0)
1548 int oid_printf_vargs(struct ceph_object_id *oid, const char *fmt, va_list ap)
1549 {
1550 	int len;
1551 
1552 	WARN_ON(!ceph_oid_empty(oid));
1553 
1554 	len = vsnprintf(oid->inline_name, sizeof(oid->inline_name), fmt, ap);
1555 	if (len >= sizeof(oid->inline_name))
1556 		return len;
1557 
1558 	oid->name_len = len;
1559 	return 0;
1560 }
1561 
1562 /*
1563  * If oid doesn't fit into inline buffer, BUG.
1564  */
1565 void ceph_oid_printf(struct ceph_object_id *oid, const char *fmt, ...)
1566 {
1567 	va_list ap;
1568 
1569 	va_start(ap, fmt);
1570 	BUG_ON(oid_printf_vargs(oid, fmt, ap));
1571 	va_end(ap);
1572 }
1573 EXPORT_SYMBOL(ceph_oid_printf);
1574 
1575 static __printf(3, 0)
1576 int oid_aprintf_vargs(struct ceph_object_id *oid, gfp_t gfp,
1577 		      const char *fmt, va_list ap)
1578 {
1579 	va_list aq;
1580 	int len;
1581 
1582 	va_copy(aq, ap);
1583 	len = oid_printf_vargs(oid, fmt, aq);
1584 	va_end(aq);
1585 
1586 	if (len) {
1587 		char *external_name;
1588 
1589 		external_name = kmalloc(len + 1, gfp);
1590 		if (!external_name)
1591 			return -ENOMEM;
1592 
1593 		oid->name = external_name;
1594 		WARN_ON(vsnprintf(oid->name, len + 1, fmt, ap) != len);
1595 		oid->name_len = len;
1596 	}
1597 
1598 	return 0;
1599 }
1600 
1601 /*
1602  * If oid doesn't fit into inline buffer, allocate.
1603  */
1604 int ceph_oid_aprintf(struct ceph_object_id *oid, gfp_t gfp,
1605 		     const char *fmt, ...)
1606 {
1607 	va_list ap;
1608 	int ret;
1609 
1610 	va_start(ap, fmt);
1611 	ret = oid_aprintf_vargs(oid, gfp, fmt, ap);
1612 	va_end(ap);
1613 
1614 	return ret;
1615 }
1616 EXPORT_SYMBOL(ceph_oid_aprintf);
1617 
1618 void ceph_oid_destroy(struct ceph_object_id *oid)
1619 {
1620 	if (oid->name != oid->inline_name)
1621 		kfree(oid->name);
1622 }
1623 EXPORT_SYMBOL(ceph_oid_destroy);
1624 
1625 /*
1626  * osds only
1627  */
1628 static bool __osds_equal(const struct ceph_osds *lhs,
1629 			 const struct ceph_osds *rhs)
1630 {
1631 	if (lhs->size == rhs->size &&
1632 	    !memcmp(lhs->osds, rhs->osds, rhs->size * sizeof(rhs->osds[0])))
1633 		return true;
1634 
1635 	return false;
1636 }
1637 
1638 /*
1639  * osds + primary
1640  */
1641 static bool osds_equal(const struct ceph_osds *lhs,
1642 		       const struct ceph_osds *rhs)
1643 {
1644 	if (__osds_equal(lhs, rhs) &&
1645 	    lhs->primary == rhs->primary)
1646 		return true;
1647 
1648 	return false;
1649 }
1650 
1651 static bool osds_valid(const struct ceph_osds *set)
1652 {
1653 	/* non-empty set */
1654 	if (set->size > 0 && set->primary >= 0)
1655 		return true;
1656 
1657 	/* empty can_shift_osds set */
1658 	if (!set->size && set->primary == -1)
1659 		return true;
1660 
1661 	/* empty !can_shift_osds set - all NONE */
1662 	if (set->size > 0 && set->primary == -1) {
1663 		int i;
1664 
1665 		for (i = 0; i < set->size; i++) {
1666 			if (set->osds[i] != CRUSH_ITEM_NONE)
1667 				break;
1668 		}
1669 		if (i == set->size)
1670 			return true;
1671 	}
1672 
1673 	return false;
1674 }
1675 
1676 void ceph_osds_copy(struct ceph_osds *dest, const struct ceph_osds *src)
1677 {
1678 	memcpy(dest->osds, src->osds, src->size * sizeof(src->osds[0]));
1679 	dest->size = src->size;
1680 	dest->primary = src->primary;
1681 }
1682 
1683 static bool is_split(const struct ceph_pg *pgid,
1684 		     u32 old_pg_num,
1685 		     u32 new_pg_num)
1686 {
1687 	int old_bits = calc_bits_of(old_pg_num);
1688 	int old_mask = (1 << old_bits) - 1;
1689 	int n;
1690 
1691 	WARN_ON(pgid->seed >= old_pg_num);
1692 	if (new_pg_num <= old_pg_num)
1693 		return false;
1694 
1695 	for (n = 1; ; n++) {
1696 		int next_bit = n << (old_bits - 1);
1697 		u32 s = next_bit | pgid->seed;
1698 
1699 		if (s < old_pg_num || s == pgid->seed)
1700 			continue;
1701 		if (s >= new_pg_num)
1702 			break;
1703 
1704 		s = ceph_stable_mod(s, old_pg_num, old_mask);
1705 		if (s == pgid->seed)
1706 			return true;
1707 	}
1708 
1709 	return false;
1710 }
1711 
1712 bool ceph_is_new_interval(const struct ceph_osds *old_acting,
1713 			  const struct ceph_osds *new_acting,
1714 			  const struct ceph_osds *old_up,
1715 			  const struct ceph_osds *new_up,
1716 			  int old_size,
1717 			  int new_size,
1718 			  int old_min_size,
1719 			  int new_min_size,
1720 			  u32 old_pg_num,
1721 			  u32 new_pg_num,
1722 			  bool old_sort_bitwise,
1723 			  bool new_sort_bitwise,
1724 			  const struct ceph_pg *pgid)
1725 {
1726 	return !osds_equal(old_acting, new_acting) ||
1727 	       !osds_equal(old_up, new_up) ||
1728 	       old_size != new_size ||
1729 	       old_min_size != new_min_size ||
1730 	       is_split(pgid, old_pg_num, new_pg_num) ||
1731 	       old_sort_bitwise != new_sort_bitwise;
1732 }
1733 
1734 static int calc_pg_rank(int osd, const struct ceph_osds *acting)
1735 {
1736 	int i;
1737 
1738 	for (i = 0; i < acting->size; i++) {
1739 		if (acting->osds[i] == osd)
1740 			return i;
1741 	}
1742 
1743 	return -1;
1744 }
1745 
1746 static bool primary_changed(const struct ceph_osds *old_acting,
1747 			    const struct ceph_osds *new_acting)
1748 {
1749 	if (!old_acting->size && !new_acting->size)
1750 		return false; /* both still empty */
1751 
1752 	if (!old_acting->size ^ !new_acting->size)
1753 		return true; /* was empty, now not, or vice versa */
1754 
1755 	if (old_acting->primary != new_acting->primary)
1756 		return true; /* primary changed */
1757 
1758 	if (calc_pg_rank(old_acting->primary, old_acting) !=
1759 	    calc_pg_rank(new_acting->primary, new_acting))
1760 		return true;
1761 
1762 	return false; /* same primary (tho replicas may have changed) */
1763 }
1764 
1765 bool ceph_osds_changed(const struct ceph_osds *old_acting,
1766 		       const struct ceph_osds *new_acting,
1767 		       bool any_change)
1768 {
1769 	if (primary_changed(old_acting, new_acting))
1770 		return true;
1771 
1772 	if (any_change && !__osds_equal(old_acting, new_acting))
1773 		return true;
1774 
1775 	return false;
1776 }
1777 
1778 /*
1779  * calculate file layout from given offset, length.
1780  * fill in correct oid, logical length, and object extent
1781  * offset, length.
1782  *
1783  * for now, we write only a single su, until we can
1784  * pass a stride back to the caller.
1785  */
1786 int ceph_calc_file_object_mapping(struct ceph_file_layout *layout,
1787 				   u64 off, u64 len,
1788 				   u64 *ono,
1789 				   u64 *oxoff, u64 *oxlen)
1790 {
1791 	u32 osize = layout->object_size;
1792 	u32 su = layout->stripe_unit;
1793 	u32 sc = layout->stripe_count;
1794 	u32 bl, stripeno, stripepos, objsetno;
1795 	u32 su_per_object;
1796 	u64 t, su_offset;
1797 
1798 	dout("mapping %llu~%llu  osize %u fl_su %u\n", off, len,
1799 	     osize, su);
1800 	if (su == 0 || sc == 0)
1801 		goto invalid;
1802 	su_per_object = osize / su;
1803 	if (su_per_object == 0)
1804 		goto invalid;
1805 	dout("osize %u / su %u = su_per_object %u\n", osize, su,
1806 	     su_per_object);
1807 
1808 	if ((su & ~PAGE_MASK) != 0)
1809 		goto invalid;
1810 
1811 	/* bl = *off / su; */
1812 	t = off;
1813 	do_div(t, su);
1814 	bl = t;
1815 	dout("off %llu / su %u = bl %u\n", off, su, bl);
1816 
1817 	stripeno = bl / sc;
1818 	stripepos = bl % sc;
1819 	objsetno = stripeno / su_per_object;
1820 
1821 	*ono = objsetno * sc + stripepos;
1822 	dout("objset %u * sc %u = ono %u\n", objsetno, sc, (unsigned int)*ono);
1823 
1824 	/* *oxoff = *off % layout->fl_stripe_unit;  # offset in su */
1825 	t = off;
1826 	su_offset = do_div(t, su);
1827 	*oxoff = su_offset + (stripeno % su_per_object) * su;
1828 
1829 	/*
1830 	 * Calculate the length of the extent being written to the selected
1831 	 * object. This is the minimum of the full length requested (len) or
1832 	 * the remainder of the current stripe being written to.
1833 	 */
1834 	*oxlen = min_t(u64, len, su - su_offset);
1835 
1836 	dout(" obj extent %llu~%llu\n", *oxoff, *oxlen);
1837 	return 0;
1838 
1839 invalid:
1840 	dout(" invalid layout\n");
1841 	*ono = 0;
1842 	*oxoff = 0;
1843 	*oxlen = 0;
1844 	return -EINVAL;
1845 }
1846 EXPORT_SYMBOL(ceph_calc_file_object_mapping);
1847 
1848 /*
1849  * Map an object into a PG.
1850  *
1851  * Should only be called with target_oid and target_oloc (as opposed to
1852  * base_oid and base_oloc), since tiering isn't taken into account.
1853  */
1854 int ceph_object_locator_to_pg(struct ceph_osdmap *osdmap,
1855 			      struct ceph_object_id *oid,
1856 			      struct ceph_object_locator *oloc,
1857 			      struct ceph_pg *raw_pgid)
1858 {
1859 	struct ceph_pg_pool_info *pi;
1860 
1861 	pi = ceph_pg_pool_by_id(osdmap, oloc->pool);
1862 	if (!pi)
1863 		return -ENOENT;
1864 
1865 	if (!oloc->pool_ns) {
1866 		raw_pgid->pool = oloc->pool;
1867 		raw_pgid->seed = ceph_str_hash(pi->object_hash, oid->name,
1868 					     oid->name_len);
1869 		dout("%s %s -> raw_pgid %llu.%x\n", __func__, oid->name,
1870 		     raw_pgid->pool, raw_pgid->seed);
1871 	} else {
1872 		char stack_buf[256];
1873 		char *buf = stack_buf;
1874 		int nsl = oloc->pool_ns->len;
1875 		size_t total = nsl + 1 + oid->name_len;
1876 
1877 		if (total > sizeof(stack_buf)) {
1878 			buf = kmalloc(total, GFP_NOIO);
1879 			if (!buf)
1880 				return -ENOMEM;
1881 		}
1882 		memcpy(buf, oloc->pool_ns->str, nsl);
1883 		buf[nsl] = '\037';
1884 		memcpy(buf + nsl + 1, oid->name, oid->name_len);
1885 		raw_pgid->pool = oloc->pool;
1886 		raw_pgid->seed = ceph_str_hash(pi->object_hash, buf, total);
1887 		if (buf != stack_buf)
1888 			kfree(buf);
1889 		dout("%s %s ns %.*s -> raw_pgid %llu.%x\n", __func__,
1890 		     oid->name, nsl, oloc->pool_ns->str,
1891 		     raw_pgid->pool, raw_pgid->seed);
1892 	}
1893 	return 0;
1894 }
1895 EXPORT_SYMBOL(ceph_object_locator_to_pg);
1896 
1897 /*
1898  * Map a raw PG (full precision ps) into an actual PG.
1899  */
1900 static void raw_pg_to_pg(struct ceph_pg_pool_info *pi,
1901 			 const struct ceph_pg *raw_pgid,
1902 			 struct ceph_pg *pgid)
1903 {
1904 	pgid->pool = raw_pgid->pool;
1905 	pgid->seed = ceph_stable_mod(raw_pgid->seed, pi->pg_num,
1906 				     pi->pg_num_mask);
1907 }
1908 
1909 /*
1910  * Map a raw PG (full precision ps) into a placement ps (placement
1911  * seed).  Include pool id in that value so that different pools don't
1912  * use the same seeds.
1913  */
1914 static u32 raw_pg_to_pps(struct ceph_pg_pool_info *pi,
1915 			 const struct ceph_pg *raw_pgid)
1916 {
1917 	if (pi->flags & CEPH_POOL_FLAG_HASHPSPOOL) {
1918 		/* hash pool id and seed so that pool PGs do not overlap */
1919 		return crush_hash32_2(CRUSH_HASH_RJENKINS1,
1920 				      ceph_stable_mod(raw_pgid->seed,
1921 						      pi->pgp_num,
1922 						      pi->pgp_num_mask),
1923 				      raw_pgid->pool);
1924 	} else {
1925 		/*
1926 		 * legacy behavior: add ps and pool together.  this is
1927 		 * not a great approach because the PGs from each pool
1928 		 * will overlap on top of each other: 0.5 == 1.4 ==
1929 		 * 2.3 == ...
1930 		 */
1931 		return ceph_stable_mod(raw_pgid->seed, pi->pgp_num,
1932 				       pi->pgp_num_mask) +
1933 		       (unsigned)raw_pgid->pool;
1934 	}
1935 }
1936 
1937 static int do_crush(struct ceph_osdmap *map, int ruleno, int x,
1938 		    int *result, int result_max,
1939 		    const __u32 *weight, int weight_max)
1940 {
1941 	int r;
1942 
1943 	BUG_ON(result_max > CEPH_PG_MAX_SIZE);
1944 
1945 	mutex_lock(&map->crush_scratch_mutex);
1946 	r = crush_do_rule(map->crush, ruleno, x, result, result_max,
1947 			  weight, weight_max, map->crush_scratch_ary);
1948 	mutex_unlock(&map->crush_scratch_mutex);
1949 
1950 	return r;
1951 }
1952 
1953 /*
1954  * Calculate raw set (CRUSH output) for given PG.  The result may
1955  * contain nonexistent OSDs.  ->primary is undefined for a raw set.
1956  *
1957  * Placement seed (CRUSH input) is returned through @ppps.
1958  */
1959 static void pg_to_raw_osds(struct ceph_osdmap *osdmap,
1960 			   struct ceph_pg_pool_info *pi,
1961 			   const struct ceph_pg *raw_pgid,
1962 			   struct ceph_osds *raw,
1963 			   u32 *ppps)
1964 {
1965 	u32 pps = raw_pg_to_pps(pi, raw_pgid);
1966 	int ruleno;
1967 	int len;
1968 
1969 	ceph_osds_init(raw);
1970 	if (ppps)
1971 		*ppps = pps;
1972 
1973 	ruleno = crush_find_rule(osdmap->crush, pi->crush_ruleset, pi->type,
1974 				 pi->size);
1975 	if (ruleno < 0) {
1976 		pr_err("no crush rule: pool %lld ruleset %d type %d size %d\n",
1977 		       pi->id, pi->crush_ruleset, pi->type, pi->size);
1978 		return;
1979 	}
1980 
1981 	len = do_crush(osdmap, ruleno, pps, raw->osds,
1982 		       min_t(int, pi->size, ARRAY_SIZE(raw->osds)),
1983 		       osdmap->osd_weight, osdmap->max_osd);
1984 	if (len < 0) {
1985 		pr_err("error %d from crush rule %d: pool %lld ruleset %d type %d size %d\n",
1986 		       len, ruleno, pi->id, pi->crush_ruleset, pi->type,
1987 		       pi->size);
1988 		return;
1989 	}
1990 
1991 	raw->size = len;
1992 }
1993 
1994 /*
1995  * Given raw set, calculate up set and up primary.  By definition of an
1996  * up set, the result won't contain nonexistent or down OSDs.
1997  *
1998  * This is done in-place - on return @set is the up set.  If it's
1999  * empty, ->primary will remain undefined.
2000  */
2001 static void raw_to_up_osds(struct ceph_osdmap *osdmap,
2002 			   struct ceph_pg_pool_info *pi,
2003 			   struct ceph_osds *set)
2004 {
2005 	int i;
2006 
2007 	/* ->primary is undefined for a raw set */
2008 	BUG_ON(set->primary != -1);
2009 
2010 	if (ceph_can_shift_osds(pi)) {
2011 		int removed = 0;
2012 
2013 		/* shift left */
2014 		for (i = 0; i < set->size; i++) {
2015 			if (ceph_osd_is_down(osdmap, set->osds[i])) {
2016 				removed++;
2017 				continue;
2018 			}
2019 			if (removed)
2020 				set->osds[i - removed] = set->osds[i];
2021 		}
2022 		set->size -= removed;
2023 		if (set->size > 0)
2024 			set->primary = set->osds[0];
2025 	} else {
2026 		/* set down/dne devices to NONE */
2027 		for (i = set->size - 1; i >= 0; i--) {
2028 			if (ceph_osd_is_down(osdmap, set->osds[i]))
2029 				set->osds[i] = CRUSH_ITEM_NONE;
2030 			else
2031 				set->primary = set->osds[i];
2032 		}
2033 	}
2034 }
2035 
2036 static void apply_primary_affinity(struct ceph_osdmap *osdmap,
2037 				   struct ceph_pg_pool_info *pi,
2038 				   u32 pps,
2039 				   struct ceph_osds *up)
2040 {
2041 	int i;
2042 	int pos = -1;
2043 
2044 	/*
2045 	 * Do we have any non-default primary_affinity values for these
2046 	 * osds?
2047 	 */
2048 	if (!osdmap->osd_primary_affinity)
2049 		return;
2050 
2051 	for (i = 0; i < up->size; i++) {
2052 		int osd = up->osds[i];
2053 
2054 		if (osd != CRUSH_ITEM_NONE &&
2055 		    osdmap->osd_primary_affinity[osd] !=
2056 					CEPH_OSD_DEFAULT_PRIMARY_AFFINITY) {
2057 			break;
2058 		}
2059 	}
2060 	if (i == up->size)
2061 		return;
2062 
2063 	/*
2064 	 * Pick the primary.  Feed both the seed (for the pg) and the
2065 	 * osd into the hash/rng so that a proportional fraction of an
2066 	 * osd's pgs get rejected as primary.
2067 	 */
2068 	for (i = 0; i < up->size; i++) {
2069 		int osd = up->osds[i];
2070 		u32 aff;
2071 
2072 		if (osd == CRUSH_ITEM_NONE)
2073 			continue;
2074 
2075 		aff = osdmap->osd_primary_affinity[osd];
2076 		if (aff < CEPH_OSD_MAX_PRIMARY_AFFINITY &&
2077 		    (crush_hash32_2(CRUSH_HASH_RJENKINS1,
2078 				    pps, osd) >> 16) >= aff) {
2079 			/*
2080 			 * We chose not to use this primary.  Note it
2081 			 * anyway as a fallback in case we don't pick
2082 			 * anyone else, but keep looking.
2083 			 */
2084 			if (pos < 0)
2085 				pos = i;
2086 		} else {
2087 			pos = i;
2088 			break;
2089 		}
2090 	}
2091 	if (pos < 0)
2092 		return;
2093 
2094 	up->primary = up->osds[pos];
2095 
2096 	if (ceph_can_shift_osds(pi) && pos > 0) {
2097 		/* move the new primary to the front */
2098 		for (i = pos; i > 0; i--)
2099 			up->osds[i] = up->osds[i - 1];
2100 		up->osds[0] = up->primary;
2101 	}
2102 }
2103 
2104 /*
2105  * Get pg_temp and primary_temp mappings for given PG.
2106  *
2107  * Note that a PG may have none, only pg_temp, only primary_temp or
2108  * both pg_temp and primary_temp mappings.  This means @temp isn't
2109  * always a valid OSD set on return: in the "only primary_temp" case,
2110  * @temp will have its ->primary >= 0 but ->size == 0.
2111  */
2112 static void get_temp_osds(struct ceph_osdmap *osdmap,
2113 			  struct ceph_pg_pool_info *pi,
2114 			  const struct ceph_pg *raw_pgid,
2115 			  struct ceph_osds *temp)
2116 {
2117 	struct ceph_pg pgid;
2118 	struct ceph_pg_mapping *pg;
2119 	int i;
2120 
2121 	raw_pg_to_pg(pi, raw_pgid, &pgid);
2122 	ceph_osds_init(temp);
2123 
2124 	/* pg_temp? */
2125 	pg = __lookup_pg_mapping(&osdmap->pg_temp, pgid);
2126 	if (pg) {
2127 		for (i = 0; i < pg->pg_temp.len; i++) {
2128 			if (ceph_osd_is_down(osdmap, pg->pg_temp.osds[i])) {
2129 				if (ceph_can_shift_osds(pi))
2130 					continue;
2131 
2132 				temp->osds[temp->size++] = CRUSH_ITEM_NONE;
2133 			} else {
2134 				temp->osds[temp->size++] = pg->pg_temp.osds[i];
2135 			}
2136 		}
2137 
2138 		/* apply pg_temp's primary */
2139 		for (i = 0; i < temp->size; i++) {
2140 			if (temp->osds[i] != CRUSH_ITEM_NONE) {
2141 				temp->primary = temp->osds[i];
2142 				break;
2143 			}
2144 		}
2145 	}
2146 
2147 	/* primary_temp? */
2148 	pg = __lookup_pg_mapping(&osdmap->primary_temp, pgid);
2149 	if (pg)
2150 		temp->primary = pg->primary_temp.osd;
2151 }
2152 
2153 /*
2154  * Map a PG to its acting set as well as its up set.
2155  *
2156  * Acting set is used for data mapping purposes, while up set can be
2157  * recorded for detecting interval changes and deciding whether to
2158  * resend a request.
2159  */
2160 void ceph_pg_to_up_acting_osds(struct ceph_osdmap *osdmap,
2161 			       const struct ceph_pg *raw_pgid,
2162 			       struct ceph_osds *up,
2163 			       struct ceph_osds *acting)
2164 {
2165 	struct ceph_pg_pool_info *pi;
2166 	u32 pps;
2167 
2168 	pi = ceph_pg_pool_by_id(osdmap, raw_pgid->pool);
2169 	if (!pi) {
2170 		ceph_osds_init(up);
2171 		ceph_osds_init(acting);
2172 		goto out;
2173 	}
2174 
2175 	pg_to_raw_osds(osdmap, pi, raw_pgid, up, &pps);
2176 	raw_to_up_osds(osdmap, pi, up);
2177 	apply_primary_affinity(osdmap, pi, pps, up);
2178 	get_temp_osds(osdmap, pi, raw_pgid, acting);
2179 	if (!acting->size) {
2180 		memcpy(acting->osds, up->osds, up->size * sizeof(up->osds[0]));
2181 		acting->size = up->size;
2182 		if (acting->primary == -1)
2183 			acting->primary = up->primary;
2184 	}
2185 out:
2186 	WARN_ON(!osds_valid(up) || !osds_valid(acting));
2187 }
2188 
2189 /*
2190  * Return acting primary for given PG, or -1 if none.
2191  */
2192 int ceph_pg_to_acting_primary(struct ceph_osdmap *osdmap,
2193 			      const struct ceph_pg *raw_pgid)
2194 {
2195 	struct ceph_osds up, acting;
2196 
2197 	ceph_pg_to_up_acting_osds(osdmap, raw_pgid, &up, &acting);
2198 	return acting.primary;
2199 }
2200 EXPORT_SYMBOL(ceph_pg_to_acting_primary);
2201