xref: /linux/net/ceph/osdmap.c (revision 83bce9c2baa51e439480a713119a73d3c8b61083)
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 void crush_finalize(struct crush_map *c)
157 {
158 	__s32 b;
159 
160 	/* Space for the array of pointers to per-bucket workspace */
161 	c->working_size = sizeof(struct crush_work) +
162 	    c->max_buckets * sizeof(struct crush_work_bucket *);
163 
164 	for (b = 0; b < c->max_buckets; b++) {
165 		if (!c->buckets[b])
166 			continue;
167 
168 		switch (c->buckets[b]->alg) {
169 		default:
170 			/*
171 			 * The base case, permutation variables and
172 			 * the pointer to the permutation array.
173 			 */
174 			c->working_size += sizeof(struct crush_work_bucket);
175 			break;
176 		}
177 		/* Every bucket has a permutation array. */
178 		c->working_size += c->buckets[b]->size * sizeof(__u32);
179 	}
180 }
181 
182 static struct crush_map *crush_decode(void *pbyval, void *end)
183 {
184 	struct crush_map *c;
185 	int err = -EINVAL;
186 	int i, j;
187 	void **p = &pbyval;
188 	void *start = pbyval;
189 	u32 magic;
190 	u32 num_name_maps;
191 
192 	dout("crush_decode %p to %p len %d\n", *p, end, (int)(end - *p));
193 
194 	c = kzalloc(sizeof(*c), GFP_NOFS);
195 	if (c == NULL)
196 		return ERR_PTR(-ENOMEM);
197 
198         /* set tunables to default values */
199         c->choose_local_tries = 2;
200         c->choose_local_fallback_tries = 5;
201         c->choose_total_tries = 19;
202 	c->chooseleaf_descend_once = 0;
203 
204 	ceph_decode_need(p, end, 4*sizeof(u32), bad);
205 	magic = ceph_decode_32(p);
206 	if (magic != CRUSH_MAGIC) {
207 		pr_err("crush_decode magic %x != current %x\n",
208 		       (unsigned int)magic, (unsigned int)CRUSH_MAGIC);
209 		goto bad;
210 	}
211 	c->max_buckets = ceph_decode_32(p);
212 	c->max_rules = ceph_decode_32(p);
213 	c->max_devices = ceph_decode_32(p);
214 
215 	c->buckets = kcalloc(c->max_buckets, sizeof(*c->buckets), GFP_NOFS);
216 	if (c->buckets == NULL)
217 		goto badmem;
218 	c->rules = kcalloc(c->max_rules, sizeof(*c->rules), GFP_NOFS);
219 	if (c->rules == NULL)
220 		goto badmem;
221 
222 	/* buckets */
223 	for (i = 0; i < c->max_buckets; i++) {
224 		int size = 0;
225 		u32 alg;
226 		struct crush_bucket *b;
227 
228 		ceph_decode_32_safe(p, end, alg, bad);
229 		if (alg == 0) {
230 			c->buckets[i] = NULL;
231 			continue;
232 		}
233 		dout("crush_decode bucket %d off %x %p to %p\n",
234 		     i, (int)(*p-start), *p, end);
235 
236 		switch (alg) {
237 		case CRUSH_BUCKET_UNIFORM:
238 			size = sizeof(struct crush_bucket_uniform);
239 			break;
240 		case CRUSH_BUCKET_LIST:
241 			size = sizeof(struct crush_bucket_list);
242 			break;
243 		case CRUSH_BUCKET_TREE:
244 			size = sizeof(struct crush_bucket_tree);
245 			break;
246 		case CRUSH_BUCKET_STRAW:
247 			size = sizeof(struct crush_bucket_straw);
248 			break;
249 		case CRUSH_BUCKET_STRAW2:
250 			size = sizeof(struct crush_bucket_straw2);
251 			break;
252 		default:
253 			err = -EINVAL;
254 			goto bad;
255 		}
256 		BUG_ON(size == 0);
257 		b = c->buckets[i] = kzalloc(size, GFP_NOFS);
258 		if (b == NULL)
259 			goto badmem;
260 
261 		ceph_decode_need(p, end, 4*sizeof(u32), bad);
262 		b->id = ceph_decode_32(p);
263 		b->type = ceph_decode_16(p);
264 		b->alg = ceph_decode_8(p);
265 		b->hash = ceph_decode_8(p);
266 		b->weight = ceph_decode_32(p);
267 		b->size = ceph_decode_32(p);
268 
269 		dout("crush_decode bucket size %d off %x %p to %p\n",
270 		     b->size, (int)(*p-start), *p, end);
271 
272 		b->items = kcalloc(b->size, sizeof(__s32), GFP_NOFS);
273 		if (b->items == NULL)
274 			goto badmem;
275 
276 		ceph_decode_need(p, end, b->size*sizeof(u32), bad);
277 		for (j = 0; j < b->size; j++)
278 			b->items[j] = ceph_decode_32(p);
279 
280 		switch (b->alg) {
281 		case CRUSH_BUCKET_UNIFORM:
282 			err = crush_decode_uniform_bucket(p, end,
283 				  (struct crush_bucket_uniform *)b);
284 			if (err < 0)
285 				goto bad;
286 			break;
287 		case CRUSH_BUCKET_LIST:
288 			err = crush_decode_list_bucket(p, end,
289 			       (struct crush_bucket_list *)b);
290 			if (err < 0)
291 				goto bad;
292 			break;
293 		case CRUSH_BUCKET_TREE:
294 			err = crush_decode_tree_bucket(p, end,
295 				(struct crush_bucket_tree *)b);
296 			if (err < 0)
297 				goto bad;
298 			break;
299 		case CRUSH_BUCKET_STRAW:
300 			err = crush_decode_straw_bucket(p, end,
301 				(struct crush_bucket_straw *)b);
302 			if (err < 0)
303 				goto bad;
304 			break;
305 		case CRUSH_BUCKET_STRAW2:
306 			err = crush_decode_straw2_bucket(p, end,
307 				(struct crush_bucket_straw2 *)b);
308 			if (err < 0)
309 				goto bad;
310 			break;
311 		}
312 	}
313 
314 	/* rules */
315 	dout("rule vec is %p\n", c->rules);
316 	for (i = 0; i < c->max_rules; i++) {
317 		u32 yes;
318 		struct crush_rule *r;
319 
320 		ceph_decode_32_safe(p, end, yes, bad);
321 		if (!yes) {
322 			dout("crush_decode NO rule %d off %x %p to %p\n",
323 			     i, (int)(*p-start), *p, end);
324 			c->rules[i] = NULL;
325 			continue;
326 		}
327 
328 		dout("crush_decode rule %d off %x %p to %p\n",
329 		     i, (int)(*p-start), *p, end);
330 
331 		/* len */
332 		ceph_decode_32_safe(p, end, yes, bad);
333 #if BITS_PER_LONG == 32
334 		err = -EINVAL;
335 		if (yes > (ULONG_MAX - sizeof(*r))
336 			  / sizeof(struct crush_rule_step))
337 			goto bad;
338 #endif
339 		r = c->rules[i] = kmalloc(sizeof(*r) +
340 					  yes*sizeof(struct crush_rule_step),
341 					  GFP_NOFS);
342 		if (r == NULL)
343 			goto badmem;
344 		dout(" rule %d is at %p\n", i, r);
345 		r->len = yes;
346 		ceph_decode_copy_safe(p, end, &r->mask, 4, bad); /* 4 u8's */
347 		ceph_decode_need(p, end, r->len*3*sizeof(u32), bad);
348 		for (j = 0; j < r->len; j++) {
349 			r->steps[j].op = ceph_decode_32(p);
350 			r->steps[j].arg1 = ceph_decode_32(p);
351 			r->steps[j].arg2 = ceph_decode_32(p);
352 		}
353 	}
354 
355 	/* ignore trailing name maps. */
356         for (num_name_maps = 0; num_name_maps < 3; num_name_maps++) {
357                 err = skip_name_map(p, end);
358                 if (err < 0)
359                         goto done;
360         }
361 
362         /* tunables */
363         ceph_decode_need(p, end, 3*sizeof(u32), done);
364         c->choose_local_tries = ceph_decode_32(p);
365         c->choose_local_fallback_tries =  ceph_decode_32(p);
366         c->choose_total_tries = ceph_decode_32(p);
367         dout("crush decode tunable choose_local_tries = %d\n",
368              c->choose_local_tries);
369         dout("crush decode tunable choose_local_fallback_tries = %d\n",
370              c->choose_local_fallback_tries);
371         dout("crush decode tunable choose_total_tries = %d\n",
372              c->choose_total_tries);
373 
374 	ceph_decode_need(p, end, sizeof(u32), done);
375 	c->chooseleaf_descend_once = ceph_decode_32(p);
376 	dout("crush decode tunable chooseleaf_descend_once = %d\n",
377 	     c->chooseleaf_descend_once);
378 
379 	ceph_decode_need(p, end, sizeof(u8), done);
380 	c->chooseleaf_vary_r = ceph_decode_8(p);
381 	dout("crush decode tunable chooseleaf_vary_r = %d\n",
382 	     c->chooseleaf_vary_r);
383 
384 	/* skip straw_calc_version, allowed_bucket_algs */
385 	ceph_decode_need(p, end, sizeof(u8) + sizeof(u32), done);
386 	*p += sizeof(u8) + sizeof(u32);
387 
388 	ceph_decode_need(p, end, sizeof(u8), done);
389 	c->chooseleaf_stable = ceph_decode_8(p);
390 	dout("crush decode tunable chooseleaf_stable = %d\n",
391 	     c->chooseleaf_stable);
392 
393 done:
394 	crush_finalize(c);
395 	dout("crush_decode success\n");
396 	return c;
397 
398 badmem:
399 	err = -ENOMEM;
400 bad:
401 	dout("crush_decode fail %d\n", err);
402 	crush_destroy(c);
403 	return ERR_PTR(err);
404 }
405 
406 int ceph_pg_compare(const struct ceph_pg *lhs, const struct ceph_pg *rhs)
407 {
408 	if (lhs->pool < rhs->pool)
409 		return -1;
410 	if (lhs->pool > rhs->pool)
411 		return 1;
412 	if (lhs->seed < rhs->seed)
413 		return -1;
414 	if (lhs->seed > rhs->seed)
415 		return 1;
416 
417 	return 0;
418 }
419 
420 /*
421  * rbtree of pg_mapping for handling pg_temp (explicit mapping of pgid
422  * to a set of osds) and primary_temp (explicit primary setting)
423  */
424 static int __insert_pg_mapping(struct ceph_pg_mapping *new,
425 			       struct rb_root *root)
426 {
427 	struct rb_node **p = &root->rb_node;
428 	struct rb_node *parent = NULL;
429 	struct ceph_pg_mapping *pg = NULL;
430 	int c;
431 
432 	dout("__insert_pg_mapping %llx %p\n", *(u64 *)&new->pgid, new);
433 	while (*p) {
434 		parent = *p;
435 		pg = rb_entry(parent, struct ceph_pg_mapping, node);
436 		c = ceph_pg_compare(&new->pgid, &pg->pgid);
437 		if (c < 0)
438 			p = &(*p)->rb_left;
439 		else if (c > 0)
440 			p = &(*p)->rb_right;
441 		else
442 			return -EEXIST;
443 	}
444 
445 	rb_link_node(&new->node, parent, p);
446 	rb_insert_color(&new->node, root);
447 	return 0;
448 }
449 
450 static struct ceph_pg_mapping *__lookup_pg_mapping(struct rb_root *root,
451 						   struct ceph_pg pgid)
452 {
453 	struct rb_node *n = root->rb_node;
454 	struct ceph_pg_mapping *pg;
455 	int c;
456 
457 	while (n) {
458 		pg = rb_entry(n, struct ceph_pg_mapping, node);
459 		c = ceph_pg_compare(&pgid, &pg->pgid);
460 		if (c < 0) {
461 			n = n->rb_left;
462 		} else if (c > 0) {
463 			n = n->rb_right;
464 		} else {
465 			dout("__lookup_pg_mapping %lld.%x got %p\n",
466 			     pgid.pool, pgid.seed, pg);
467 			return pg;
468 		}
469 	}
470 	return NULL;
471 }
472 
473 static int __remove_pg_mapping(struct rb_root *root, struct ceph_pg pgid)
474 {
475 	struct ceph_pg_mapping *pg = __lookup_pg_mapping(root, pgid);
476 
477 	if (pg) {
478 		dout("__remove_pg_mapping %lld.%x %p\n", pgid.pool, pgid.seed,
479 		     pg);
480 		rb_erase(&pg->node, root);
481 		kfree(pg);
482 		return 0;
483 	}
484 	dout("__remove_pg_mapping %lld.%x dne\n", pgid.pool, pgid.seed);
485 	return -ENOENT;
486 }
487 
488 /*
489  * rbtree of pg pool info
490  */
491 static int __insert_pg_pool(struct rb_root *root, struct ceph_pg_pool_info *new)
492 {
493 	struct rb_node **p = &root->rb_node;
494 	struct rb_node *parent = NULL;
495 	struct ceph_pg_pool_info *pi = NULL;
496 
497 	while (*p) {
498 		parent = *p;
499 		pi = rb_entry(parent, struct ceph_pg_pool_info, node);
500 		if (new->id < pi->id)
501 			p = &(*p)->rb_left;
502 		else if (new->id > pi->id)
503 			p = &(*p)->rb_right;
504 		else
505 			return -EEXIST;
506 	}
507 
508 	rb_link_node(&new->node, parent, p);
509 	rb_insert_color(&new->node, root);
510 	return 0;
511 }
512 
513 static struct ceph_pg_pool_info *__lookup_pg_pool(struct rb_root *root, u64 id)
514 {
515 	struct ceph_pg_pool_info *pi;
516 	struct rb_node *n = root->rb_node;
517 
518 	while (n) {
519 		pi = rb_entry(n, struct ceph_pg_pool_info, node);
520 		if (id < pi->id)
521 			n = n->rb_left;
522 		else if (id > pi->id)
523 			n = n->rb_right;
524 		else
525 			return pi;
526 	}
527 	return NULL;
528 }
529 
530 struct ceph_pg_pool_info *ceph_pg_pool_by_id(struct ceph_osdmap *map, u64 id)
531 {
532 	return __lookup_pg_pool(&map->pg_pools, id);
533 }
534 
535 const char *ceph_pg_pool_name_by_id(struct ceph_osdmap *map, u64 id)
536 {
537 	struct ceph_pg_pool_info *pi;
538 
539 	if (id == CEPH_NOPOOL)
540 		return NULL;
541 
542 	if (WARN_ON_ONCE(id > (u64) INT_MAX))
543 		return NULL;
544 
545 	pi = __lookup_pg_pool(&map->pg_pools, (int) id);
546 
547 	return pi ? pi->name : NULL;
548 }
549 EXPORT_SYMBOL(ceph_pg_pool_name_by_id);
550 
551 int ceph_pg_poolid_by_name(struct ceph_osdmap *map, const char *name)
552 {
553 	struct rb_node *rbp;
554 
555 	for (rbp = rb_first(&map->pg_pools); rbp; rbp = rb_next(rbp)) {
556 		struct ceph_pg_pool_info *pi =
557 			rb_entry(rbp, struct ceph_pg_pool_info, node);
558 		if (pi->name && strcmp(pi->name, name) == 0)
559 			return pi->id;
560 	}
561 	return -ENOENT;
562 }
563 EXPORT_SYMBOL(ceph_pg_poolid_by_name);
564 
565 static void __remove_pg_pool(struct rb_root *root, struct ceph_pg_pool_info *pi)
566 {
567 	rb_erase(&pi->node, root);
568 	kfree(pi->name);
569 	kfree(pi);
570 }
571 
572 static int decode_pool(void **p, void *end, struct ceph_pg_pool_info *pi)
573 {
574 	u8 ev, cv;
575 	unsigned len, num;
576 	void *pool_end;
577 
578 	ceph_decode_need(p, end, 2 + 4, bad);
579 	ev = ceph_decode_8(p);  /* encoding version */
580 	cv = ceph_decode_8(p); /* compat version */
581 	if (ev < 5) {
582 		pr_warn("got v %d < 5 cv %d of ceph_pg_pool\n", ev, cv);
583 		return -EINVAL;
584 	}
585 	if (cv > 9) {
586 		pr_warn("got v %d cv %d > 9 of ceph_pg_pool\n", ev, cv);
587 		return -EINVAL;
588 	}
589 	len = ceph_decode_32(p);
590 	ceph_decode_need(p, end, len, bad);
591 	pool_end = *p + len;
592 
593 	pi->type = ceph_decode_8(p);
594 	pi->size = ceph_decode_8(p);
595 	pi->crush_ruleset = ceph_decode_8(p);
596 	pi->object_hash = ceph_decode_8(p);
597 
598 	pi->pg_num = ceph_decode_32(p);
599 	pi->pgp_num = ceph_decode_32(p);
600 
601 	*p += 4 + 4;  /* skip lpg* */
602 	*p += 4;      /* skip last_change */
603 	*p += 8 + 4;  /* skip snap_seq, snap_epoch */
604 
605 	/* skip snaps */
606 	num = ceph_decode_32(p);
607 	while (num--) {
608 		*p += 8;  /* snapid key */
609 		*p += 1 + 1; /* versions */
610 		len = ceph_decode_32(p);
611 		*p += len;
612 	}
613 
614 	/* skip removed_snaps */
615 	num = ceph_decode_32(p);
616 	*p += num * (8 + 8);
617 
618 	*p += 8;  /* skip auid */
619 	pi->flags = ceph_decode_64(p);
620 	*p += 4;  /* skip crash_replay_interval */
621 
622 	if (ev >= 7)
623 		pi->min_size = ceph_decode_8(p);
624 	else
625 		pi->min_size = pi->size - pi->size / 2;
626 
627 	if (ev >= 8)
628 		*p += 8 + 8;  /* skip quota_max_* */
629 
630 	if (ev >= 9) {
631 		/* skip tiers */
632 		num = ceph_decode_32(p);
633 		*p += num * 8;
634 
635 		*p += 8;  /* skip tier_of */
636 		*p += 1;  /* skip cache_mode */
637 
638 		pi->read_tier = ceph_decode_64(p);
639 		pi->write_tier = ceph_decode_64(p);
640 	} else {
641 		pi->read_tier = -1;
642 		pi->write_tier = -1;
643 	}
644 
645 	if (ev >= 10) {
646 		/* skip properties */
647 		num = ceph_decode_32(p);
648 		while (num--) {
649 			len = ceph_decode_32(p);
650 			*p += len; /* key */
651 			len = ceph_decode_32(p);
652 			*p += len; /* val */
653 		}
654 	}
655 
656 	if (ev >= 11) {
657 		/* skip hit_set_params */
658 		*p += 1 + 1; /* versions */
659 		len = ceph_decode_32(p);
660 		*p += len;
661 
662 		*p += 4; /* skip hit_set_period */
663 		*p += 4; /* skip hit_set_count */
664 	}
665 
666 	if (ev >= 12)
667 		*p += 4; /* skip stripe_width */
668 
669 	if (ev >= 13) {
670 		*p += 8; /* skip target_max_bytes */
671 		*p += 8; /* skip target_max_objects */
672 		*p += 4; /* skip cache_target_dirty_ratio_micro */
673 		*p += 4; /* skip cache_target_full_ratio_micro */
674 		*p += 4; /* skip cache_min_flush_age */
675 		*p += 4; /* skip cache_min_evict_age */
676 	}
677 
678 	if (ev >=  14) {
679 		/* skip erasure_code_profile */
680 		len = ceph_decode_32(p);
681 		*p += len;
682 	}
683 
684 	if (ev >= 15)
685 		pi->last_force_request_resend = ceph_decode_32(p);
686 	else
687 		pi->last_force_request_resend = 0;
688 
689 	/* ignore the rest */
690 
691 	*p = pool_end;
692 	calc_pg_masks(pi);
693 	return 0;
694 
695 bad:
696 	return -EINVAL;
697 }
698 
699 static int decode_pool_names(void **p, void *end, struct ceph_osdmap *map)
700 {
701 	struct ceph_pg_pool_info *pi;
702 	u32 num, len;
703 	u64 pool;
704 
705 	ceph_decode_32_safe(p, end, num, bad);
706 	dout(" %d pool names\n", num);
707 	while (num--) {
708 		ceph_decode_64_safe(p, end, pool, bad);
709 		ceph_decode_32_safe(p, end, len, bad);
710 		dout("  pool %llu len %d\n", pool, len);
711 		ceph_decode_need(p, end, len, bad);
712 		pi = __lookup_pg_pool(&map->pg_pools, pool);
713 		if (pi) {
714 			char *name = kstrndup(*p, len, GFP_NOFS);
715 
716 			if (!name)
717 				return -ENOMEM;
718 			kfree(pi->name);
719 			pi->name = name;
720 			dout("  name is %s\n", pi->name);
721 		}
722 		*p += len;
723 	}
724 	return 0;
725 
726 bad:
727 	return -EINVAL;
728 }
729 
730 /*
731  * osd map
732  */
733 struct ceph_osdmap *ceph_osdmap_alloc(void)
734 {
735 	struct ceph_osdmap *map;
736 
737 	map = kzalloc(sizeof(*map), GFP_NOIO);
738 	if (!map)
739 		return NULL;
740 
741 	map->pg_pools = RB_ROOT;
742 	map->pool_max = -1;
743 	map->pg_temp = RB_ROOT;
744 	map->primary_temp = RB_ROOT;
745 	mutex_init(&map->crush_workspace_mutex);
746 
747 	return map;
748 }
749 
750 void ceph_osdmap_destroy(struct ceph_osdmap *map)
751 {
752 	dout("osdmap_destroy %p\n", map);
753 	if (map->crush)
754 		crush_destroy(map->crush);
755 	while (!RB_EMPTY_ROOT(&map->pg_temp)) {
756 		struct ceph_pg_mapping *pg =
757 			rb_entry(rb_first(&map->pg_temp),
758 				 struct ceph_pg_mapping, node);
759 		rb_erase(&pg->node, &map->pg_temp);
760 		kfree(pg);
761 	}
762 	while (!RB_EMPTY_ROOT(&map->primary_temp)) {
763 		struct ceph_pg_mapping *pg =
764 			rb_entry(rb_first(&map->primary_temp),
765 				 struct ceph_pg_mapping, node);
766 		rb_erase(&pg->node, &map->primary_temp);
767 		kfree(pg);
768 	}
769 	while (!RB_EMPTY_ROOT(&map->pg_pools)) {
770 		struct ceph_pg_pool_info *pi =
771 			rb_entry(rb_first(&map->pg_pools),
772 				 struct ceph_pg_pool_info, node);
773 		__remove_pg_pool(&map->pg_pools, pi);
774 	}
775 	kfree(map->osd_state);
776 	kfree(map->osd_weight);
777 	kfree(map->osd_addr);
778 	kfree(map->osd_primary_affinity);
779 	kfree(map->crush_workspace);
780 	kfree(map);
781 }
782 
783 /*
784  * Adjust max_osd value, (re)allocate arrays.
785  *
786  * The new elements are properly initialized.
787  */
788 static int osdmap_set_max_osd(struct ceph_osdmap *map, int max)
789 {
790 	u8 *state;
791 	u32 *weight;
792 	struct ceph_entity_addr *addr;
793 	int i;
794 
795 	state = krealloc(map->osd_state, max*sizeof(*state), GFP_NOFS);
796 	if (!state)
797 		return -ENOMEM;
798 	map->osd_state = state;
799 
800 	weight = krealloc(map->osd_weight, max*sizeof(*weight), GFP_NOFS);
801 	if (!weight)
802 		return -ENOMEM;
803 	map->osd_weight = weight;
804 
805 	addr = krealloc(map->osd_addr, max*sizeof(*addr), GFP_NOFS);
806 	if (!addr)
807 		return -ENOMEM;
808 	map->osd_addr = addr;
809 
810 	for (i = map->max_osd; i < max; i++) {
811 		map->osd_state[i] = 0;
812 		map->osd_weight[i] = CEPH_OSD_OUT;
813 		memset(map->osd_addr + i, 0, sizeof(*map->osd_addr));
814 	}
815 
816 	if (map->osd_primary_affinity) {
817 		u32 *affinity;
818 
819 		affinity = krealloc(map->osd_primary_affinity,
820 				    max*sizeof(*affinity), GFP_NOFS);
821 		if (!affinity)
822 			return -ENOMEM;
823 		map->osd_primary_affinity = affinity;
824 
825 		for (i = map->max_osd; i < max; i++)
826 			map->osd_primary_affinity[i] =
827 			    CEPH_OSD_DEFAULT_PRIMARY_AFFINITY;
828 	}
829 
830 	map->max_osd = max;
831 
832 	return 0;
833 }
834 
835 static int osdmap_set_crush(struct ceph_osdmap *map, struct crush_map *crush)
836 {
837 	void *workspace;
838 	size_t work_size;
839 
840 	if (IS_ERR(crush))
841 		return PTR_ERR(crush);
842 
843 	work_size = crush_work_size(crush, CEPH_PG_MAX_SIZE);
844 	dout("%s work_size %zu bytes\n", __func__, work_size);
845 	workspace = kmalloc(work_size, GFP_NOIO);
846 	if (!workspace) {
847 		crush_destroy(crush);
848 		return -ENOMEM;
849 	}
850 	crush_init_workspace(crush, workspace);
851 
852 	if (map->crush)
853 		crush_destroy(map->crush);
854 	kfree(map->crush_workspace);
855 	map->crush = crush;
856 	map->crush_workspace = workspace;
857 	return 0;
858 }
859 
860 #define OSDMAP_WRAPPER_COMPAT_VER	7
861 #define OSDMAP_CLIENT_DATA_COMPAT_VER	1
862 
863 /*
864  * Return 0 or error.  On success, *v is set to 0 for old (v6) osdmaps,
865  * to struct_v of the client_data section for new (v7 and above)
866  * osdmaps.
867  */
868 static int get_osdmap_client_data_v(void **p, void *end,
869 				    const char *prefix, u8 *v)
870 {
871 	u8 struct_v;
872 
873 	ceph_decode_8_safe(p, end, struct_v, e_inval);
874 	if (struct_v >= 7) {
875 		u8 struct_compat;
876 
877 		ceph_decode_8_safe(p, end, struct_compat, e_inval);
878 		if (struct_compat > OSDMAP_WRAPPER_COMPAT_VER) {
879 			pr_warn("got v %d cv %d > %d of %s ceph_osdmap\n",
880 				struct_v, struct_compat,
881 				OSDMAP_WRAPPER_COMPAT_VER, prefix);
882 			return -EINVAL;
883 		}
884 		*p += 4; /* ignore wrapper struct_len */
885 
886 		ceph_decode_8_safe(p, end, struct_v, e_inval);
887 		ceph_decode_8_safe(p, end, struct_compat, e_inval);
888 		if (struct_compat > OSDMAP_CLIENT_DATA_COMPAT_VER) {
889 			pr_warn("got v %d cv %d > %d of %s ceph_osdmap client data\n",
890 				struct_v, struct_compat,
891 				OSDMAP_CLIENT_DATA_COMPAT_VER, prefix);
892 			return -EINVAL;
893 		}
894 		*p += 4; /* ignore client data struct_len */
895 	} else {
896 		u16 version;
897 
898 		*p -= 1;
899 		ceph_decode_16_safe(p, end, version, e_inval);
900 		if (version < 6) {
901 			pr_warn("got v %d < 6 of %s ceph_osdmap\n",
902 				version, prefix);
903 			return -EINVAL;
904 		}
905 
906 		/* old osdmap enconding */
907 		struct_v = 0;
908 	}
909 
910 	*v = struct_v;
911 	return 0;
912 
913 e_inval:
914 	return -EINVAL;
915 }
916 
917 static int __decode_pools(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_pool_info *pi;
925 		u64 pool;
926 		int ret;
927 
928 		ceph_decode_64_safe(p, end, pool, e_inval);
929 
930 		pi = __lookup_pg_pool(&map->pg_pools, pool);
931 		if (!incremental || !pi) {
932 			pi = kzalloc(sizeof(*pi), GFP_NOFS);
933 			if (!pi)
934 				return -ENOMEM;
935 
936 			pi->id = pool;
937 
938 			ret = __insert_pg_pool(&map->pg_pools, pi);
939 			if (ret) {
940 				kfree(pi);
941 				return ret;
942 			}
943 		}
944 
945 		ret = decode_pool(p, end, pi);
946 		if (ret)
947 			return ret;
948 	}
949 
950 	return 0;
951 
952 e_inval:
953 	return -EINVAL;
954 }
955 
956 static int decode_pools(void **p, void *end, struct ceph_osdmap *map)
957 {
958 	return __decode_pools(p, end, map, false);
959 }
960 
961 static int decode_new_pools(void **p, void *end, struct ceph_osdmap *map)
962 {
963 	return __decode_pools(p, end, map, true);
964 }
965 
966 static int __decode_pg_temp(void **p, void *end, struct ceph_osdmap *map,
967 			    bool incremental)
968 {
969 	u32 n;
970 
971 	ceph_decode_32_safe(p, end, n, e_inval);
972 	while (n--) {
973 		struct ceph_pg pgid;
974 		u32 len, i;
975 		int ret;
976 
977 		ret = ceph_decode_pgid(p, end, &pgid);
978 		if (ret)
979 			return ret;
980 
981 		ceph_decode_32_safe(p, end, len, e_inval);
982 
983 		ret = __remove_pg_mapping(&map->pg_temp, pgid);
984 		BUG_ON(!incremental && ret != -ENOENT);
985 
986 		if (!incremental || len > 0) {
987 			struct ceph_pg_mapping *pg;
988 
989 			ceph_decode_need(p, end, len*sizeof(u32), e_inval);
990 
991 			if (len > (UINT_MAX - sizeof(*pg)) / sizeof(u32))
992 				return -EINVAL;
993 
994 			pg = kzalloc(sizeof(*pg) + len*sizeof(u32), GFP_NOFS);
995 			if (!pg)
996 				return -ENOMEM;
997 
998 			pg->pgid = pgid;
999 			pg->pg_temp.len = len;
1000 			for (i = 0; i < len; i++)
1001 				pg->pg_temp.osds[i] = ceph_decode_32(p);
1002 
1003 			ret = __insert_pg_mapping(pg, &map->pg_temp);
1004 			if (ret) {
1005 				kfree(pg);
1006 				return ret;
1007 			}
1008 		}
1009 	}
1010 
1011 	return 0;
1012 
1013 e_inval:
1014 	return -EINVAL;
1015 }
1016 
1017 static int decode_pg_temp(void **p, void *end, struct ceph_osdmap *map)
1018 {
1019 	return __decode_pg_temp(p, end, map, false);
1020 }
1021 
1022 static int decode_new_pg_temp(void **p, void *end, struct ceph_osdmap *map)
1023 {
1024 	return __decode_pg_temp(p, end, map, true);
1025 }
1026 
1027 static int __decode_primary_temp(void **p, void *end, struct ceph_osdmap *map,
1028 				 bool incremental)
1029 {
1030 	u32 n;
1031 
1032 	ceph_decode_32_safe(p, end, n, e_inval);
1033 	while (n--) {
1034 		struct ceph_pg pgid;
1035 		u32 osd;
1036 		int ret;
1037 
1038 		ret = ceph_decode_pgid(p, end, &pgid);
1039 		if (ret)
1040 			return ret;
1041 
1042 		ceph_decode_32_safe(p, end, osd, e_inval);
1043 
1044 		ret = __remove_pg_mapping(&map->primary_temp, pgid);
1045 		BUG_ON(!incremental && ret != -ENOENT);
1046 
1047 		if (!incremental || osd != (u32)-1) {
1048 			struct ceph_pg_mapping *pg;
1049 
1050 			pg = kzalloc(sizeof(*pg), GFP_NOFS);
1051 			if (!pg)
1052 				return -ENOMEM;
1053 
1054 			pg->pgid = pgid;
1055 			pg->primary_temp.osd = osd;
1056 
1057 			ret = __insert_pg_mapping(pg, &map->primary_temp);
1058 			if (ret) {
1059 				kfree(pg);
1060 				return ret;
1061 			}
1062 		}
1063 	}
1064 
1065 	return 0;
1066 
1067 e_inval:
1068 	return -EINVAL;
1069 }
1070 
1071 static int decode_primary_temp(void **p, void *end, struct ceph_osdmap *map)
1072 {
1073 	return __decode_primary_temp(p, end, map, false);
1074 }
1075 
1076 static int decode_new_primary_temp(void **p, void *end,
1077 				   struct ceph_osdmap *map)
1078 {
1079 	return __decode_primary_temp(p, end, map, true);
1080 }
1081 
1082 u32 ceph_get_primary_affinity(struct ceph_osdmap *map, int osd)
1083 {
1084 	BUG_ON(osd >= map->max_osd);
1085 
1086 	if (!map->osd_primary_affinity)
1087 		return CEPH_OSD_DEFAULT_PRIMARY_AFFINITY;
1088 
1089 	return map->osd_primary_affinity[osd];
1090 }
1091 
1092 static int set_primary_affinity(struct ceph_osdmap *map, int osd, u32 aff)
1093 {
1094 	BUG_ON(osd >= map->max_osd);
1095 
1096 	if (!map->osd_primary_affinity) {
1097 		int i;
1098 
1099 		map->osd_primary_affinity = kmalloc(map->max_osd*sizeof(u32),
1100 						    GFP_NOFS);
1101 		if (!map->osd_primary_affinity)
1102 			return -ENOMEM;
1103 
1104 		for (i = 0; i < map->max_osd; i++)
1105 			map->osd_primary_affinity[i] =
1106 			    CEPH_OSD_DEFAULT_PRIMARY_AFFINITY;
1107 	}
1108 
1109 	map->osd_primary_affinity[osd] = aff;
1110 
1111 	return 0;
1112 }
1113 
1114 static int decode_primary_affinity(void **p, void *end,
1115 				   struct ceph_osdmap *map)
1116 {
1117 	u32 len, i;
1118 
1119 	ceph_decode_32_safe(p, end, len, e_inval);
1120 	if (len == 0) {
1121 		kfree(map->osd_primary_affinity);
1122 		map->osd_primary_affinity = NULL;
1123 		return 0;
1124 	}
1125 	if (len != map->max_osd)
1126 		goto e_inval;
1127 
1128 	ceph_decode_need(p, end, map->max_osd*sizeof(u32), e_inval);
1129 
1130 	for (i = 0; i < map->max_osd; i++) {
1131 		int ret;
1132 
1133 		ret = set_primary_affinity(map, i, ceph_decode_32(p));
1134 		if (ret)
1135 			return ret;
1136 	}
1137 
1138 	return 0;
1139 
1140 e_inval:
1141 	return -EINVAL;
1142 }
1143 
1144 static int decode_new_primary_affinity(void **p, void *end,
1145 				       struct ceph_osdmap *map)
1146 {
1147 	u32 n;
1148 
1149 	ceph_decode_32_safe(p, end, n, e_inval);
1150 	while (n--) {
1151 		u32 osd, aff;
1152 		int ret;
1153 
1154 		ceph_decode_32_safe(p, end, osd, e_inval);
1155 		ceph_decode_32_safe(p, end, aff, e_inval);
1156 
1157 		ret = set_primary_affinity(map, osd, aff);
1158 		if (ret)
1159 			return ret;
1160 
1161 		pr_info("osd%d primary-affinity 0x%x\n", osd, aff);
1162 	}
1163 
1164 	return 0;
1165 
1166 e_inval:
1167 	return -EINVAL;
1168 }
1169 
1170 /*
1171  * decode a full map.
1172  */
1173 static int osdmap_decode(void **p, void *end, struct ceph_osdmap *map)
1174 {
1175 	u8 struct_v;
1176 	u32 epoch = 0;
1177 	void *start = *p;
1178 	u32 max;
1179 	u32 len, i;
1180 	int err;
1181 
1182 	dout("%s %p to %p len %d\n", __func__, *p, end, (int)(end - *p));
1183 
1184 	err = get_osdmap_client_data_v(p, end, "full", &struct_v);
1185 	if (err)
1186 		goto bad;
1187 
1188 	/* fsid, epoch, created, modified */
1189 	ceph_decode_need(p, end, sizeof(map->fsid) + sizeof(u32) +
1190 			 sizeof(map->created) + sizeof(map->modified), e_inval);
1191 	ceph_decode_copy(p, &map->fsid, sizeof(map->fsid));
1192 	epoch = map->epoch = ceph_decode_32(p);
1193 	ceph_decode_copy(p, &map->created, sizeof(map->created));
1194 	ceph_decode_copy(p, &map->modified, sizeof(map->modified));
1195 
1196 	/* pools */
1197 	err = decode_pools(p, end, map);
1198 	if (err)
1199 		goto bad;
1200 
1201 	/* pool_name */
1202 	err = decode_pool_names(p, end, map);
1203 	if (err)
1204 		goto bad;
1205 
1206 	ceph_decode_32_safe(p, end, map->pool_max, e_inval);
1207 
1208 	ceph_decode_32_safe(p, end, map->flags, e_inval);
1209 
1210 	/* max_osd */
1211 	ceph_decode_32_safe(p, end, max, e_inval);
1212 
1213 	/* (re)alloc osd arrays */
1214 	err = osdmap_set_max_osd(map, max);
1215 	if (err)
1216 		goto bad;
1217 
1218 	/* osd_state, osd_weight, osd_addrs->client_addr */
1219 	ceph_decode_need(p, end, 3*sizeof(u32) +
1220 			 map->max_osd*(1 + sizeof(*map->osd_weight) +
1221 				       sizeof(*map->osd_addr)), e_inval);
1222 
1223 	if (ceph_decode_32(p) != map->max_osd)
1224 		goto e_inval;
1225 
1226 	ceph_decode_copy(p, map->osd_state, map->max_osd);
1227 
1228 	if (ceph_decode_32(p) != map->max_osd)
1229 		goto e_inval;
1230 
1231 	for (i = 0; i < map->max_osd; i++)
1232 		map->osd_weight[i] = ceph_decode_32(p);
1233 
1234 	if (ceph_decode_32(p) != map->max_osd)
1235 		goto e_inval;
1236 
1237 	ceph_decode_copy(p, map->osd_addr, map->max_osd*sizeof(*map->osd_addr));
1238 	for (i = 0; i < map->max_osd; i++)
1239 		ceph_decode_addr(&map->osd_addr[i]);
1240 
1241 	/* pg_temp */
1242 	err = decode_pg_temp(p, end, map);
1243 	if (err)
1244 		goto bad;
1245 
1246 	/* primary_temp */
1247 	if (struct_v >= 1) {
1248 		err = decode_primary_temp(p, end, map);
1249 		if (err)
1250 			goto bad;
1251 	}
1252 
1253 	/* primary_affinity */
1254 	if (struct_v >= 2) {
1255 		err = decode_primary_affinity(p, end, map);
1256 		if (err)
1257 			goto bad;
1258 	} else {
1259 		/* XXX can this happen? */
1260 		kfree(map->osd_primary_affinity);
1261 		map->osd_primary_affinity = NULL;
1262 	}
1263 
1264 	/* crush */
1265 	ceph_decode_32_safe(p, end, len, e_inval);
1266 	err = osdmap_set_crush(map, crush_decode(*p, min(*p + len, end)));
1267 	if (err)
1268 		goto bad;
1269 
1270 	/* ignore the rest */
1271 	*p = end;
1272 
1273 	dout("full osdmap epoch %d max_osd %d\n", map->epoch, map->max_osd);
1274 	return 0;
1275 
1276 e_inval:
1277 	err = -EINVAL;
1278 bad:
1279 	pr_err("corrupt full osdmap (%d) epoch %d off %d (%p of %p-%p)\n",
1280 	       err, epoch, (int)(*p - start), *p, start, end);
1281 	print_hex_dump(KERN_DEBUG, "osdmap: ",
1282 		       DUMP_PREFIX_OFFSET, 16, 1,
1283 		       start, end - start, true);
1284 	return err;
1285 }
1286 
1287 /*
1288  * Allocate and decode a full map.
1289  */
1290 struct ceph_osdmap *ceph_osdmap_decode(void **p, void *end)
1291 {
1292 	struct ceph_osdmap *map;
1293 	int ret;
1294 
1295 	map = ceph_osdmap_alloc();
1296 	if (!map)
1297 		return ERR_PTR(-ENOMEM);
1298 
1299 	ret = osdmap_decode(p, end, map);
1300 	if (ret) {
1301 		ceph_osdmap_destroy(map);
1302 		return ERR_PTR(ret);
1303 	}
1304 
1305 	return map;
1306 }
1307 
1308 /*
1309  * Encoding order is (new_up_client, new_state, new_weight).  Need to
1310  * apply in the (new_weight, new_state, new_up_client) order, because
1311  * an incremental map may look like e.g.
1312  *
1313  *     new_up_client: { osd=6, addr=... } # set osd_state and addr
1314  *     new_state: { osd=6, xorstate=EXISTS } # clear osd_state
1315  */
1316 static int decode_new_up_state_weight(void **p, void *end,
1317 				      struct ceph_osdmap *map)
1318 {
1319 	void *new_up_client;
1320 	void *new_state;
1321 	void *new_weight_end;
1322 	u32 len;
1323 
1324 	new_up_client = *p;
1325 	ceph_decode_32_safe(p, end, len, e_inval);
1326 	len *= sizeof(u32) + sizeof(struct ceph_entity_addr);
1327 	ceph_decode_need(p, end, len, e_inval);
1328 	*p += len;
1329 
1330 	new_state = *p;
1331 	ceph_decode_32_safe(p, end, len, e_inval);
1332 	len *= sizeof(u32) + sizeof(u8);
1333 	ceph_decode_need(p, end, len, e_inval);
1334 	*p += len;
1335 
1336 	/* new_weight */
1337 	ceph_decode_32_safe(p, end, len, e_inval);
1338 	while (len--) {
1339 		s32 osd;
1340 		u32 w;
1341 
1342 		ceph_decode_need(p, end, 2*sizeof(u32), e_inval);
1343 		osd = ceph_decode_32(p);
1344 		w = ceph_decode_32(p);
1345 		BUG_ON(osd >= map->max_osd);
1346 		pr_info("osd%d weight 0x%x %s\n", osd, w,
1347 		     w == CEPH_OSD_IN ? "(in)" :
1348 		     (w == CEPH_OSD_OUT ? "(out)" : ""));
1349 		map->osd_weight[osd] = w;
1350 
1351 		/*
1352 		 * If we are marking in, set the EXISTS, and clear the
1353 		 * AUTOOUT and NEW bits.
1354 		 */
1355 		if (w) {
1356 			map->osd_state[osd] |= CEPH_OSD_EXISTS;
1357 			map->osd_state[osd] &= ~(CEPH_OSD_AUTOOUT |
1358 						 CEPH_OSD_NEW);
1359 		}
1360 	}
1361 	new_weight_end = *p;
1362 
1363 	/* new_state (up/down) */
1364 	*p = new_state;
1365 	len = ceph_decode_32(p);
1366 	while (len--) {
1367 		s32 osd;
1368 		u8 xorstate;
1369 		int ret;
1370 
1371 		osd = ceph_decode_32(p);
1372 		xorstate = ceph_decode_8(p);
1373 		if (xorstate == 0)
1374 			xorstate = CEPH_OSD_UP;
1375 		BUG_ON(osd >= map->max_osd);
1376 		if ((map->osd_state[osd] & CEPH_OSD_UP) &&
1377 		    (xorstate & CEPH_OSD_UP))
1378 			pr_info("osd%d down\n", osd);
1379 		if ((map->osd_state[osd] & CEPH_OSD_EXISTS) &&
1380 		    (xorstate & CEPH_OSD_EXISTS)) {
1381 			pr_info("osd%d does not exist\n", osd);
1382 			ret = set_primary_affinity(map, osd,
1383 						   CEPH_OSD_DEFAULT_PRIMARY_AFFINITY);
1384 			if (ret)
1385 				return ret;
1386 			memset(map->osd_addr + osd, 0, sizeof(*map->osd_addr));
1387 			map->osd_state[osd] = 0;
1388 		} else {
1389 			map->osd_state[osd] ^= xorstate;
1390 		}
1391 	}
1392 
1393 	/* new_up_client */
1394 	*p = new_up_client;
1395 	len = ceph_decode_32(p);
1396 	while (len--) {
1397 		s32 osd;
1398 		struct ceph_entity_addr addr;
1399 
1400 		osd = ceph_decode_32(p);
1401 		ceph_decode_copy(p, &addr, sizeof(addr));
1402 		ceph_decode_addr(&addr);
1403 		BUG_ON(osd >= map->max_osd);
1404 		pr_info("osd%d up\n", osd);
1405 		map->osd_state[osd] |= CEPH_OSD_EXISTS | CEPH_OSD_UP;
1406 		map->osd_addr[osd] = addr;
1407 	}
1408 
1409 	*p = new_weight_end;
1410 	return 0;
1411 
1412 e_inval:
1413 	return -EINVAL;
1414 }
1415 
1416 /*
1417  * decode and apply an incremental map update.
1418  */
1419 struct ceph_osdmap *osdmap_apply_incremental(void **p, void *end,
1420 					     struct ceph_osdmap *map)
1421 {
1422 	struct ceph_fsid fsid;
1423 	u32 epoch = 0;
1424 	struct ceph_timespec modified;
1425 	s32 len;
1426 	u64 pool;
1427 	__s64 new_pool_max;
1428 	__s32 new_flags, max;
1429 	void *start = *p;
1430 	int err;
1431 	u8 struct_v;
1432 
1433 	dout("%s %p to %p len %d\n", __func__, *p, end, (int)(end - *p));
1434 
1435 	err = get_osdmap_client_data_v(p, end, "inc", &struct_v);
1436 	if (err)
1437 		goto bad;
1438 
1439 	/* fsid, epoch, modified, new_pool_max, new_flags */
1440 	ceph_decode_need(p, end, sizeof(fsid) + sizeof(u32) + sizeof(modified) +
1441 			 sizeof(u64) + sizeof(u32), e_inval);
1442 	ceph_decode_copy(p, &fsid, sizeof(fsid));
1443 	epoch = ceph_decode_32(p);
1444 	BUG_ON(epoch != map->epoch+1);
1445 	ceph_decode_copy(p, &modified, sizeof(modified));
1446 	new_pool_max = ceph_decode_64(p);
1447 	new_flags = ceph_decode_32(p);
1448 
1449 	/* full map? */
1450 	ceph_decode_32_safe(p, end, len, e_inval);
1451 	if (len > 0) {
1452 		dout("apply_incremental full map len %d, %p to %p\n",
1453 		     len, *p, end);
1454 		return ceph_osdmap_decode(p, min(*p+len, end));
1455 	}
1456 
1457 	/* new crush? */
1458 	ceph_decode_32_safe(p, end, len, e_inval);
1459 	if (len > 0) {
1460 		err = osdmap_set_crush(map,
1461 				       crush_decode(*p, min(*p + len, end)));
1462 		if (err)
1463 			goto bad;
1464 		*p += len;
1465 	}
1466 
1467 	/* new flags? */
1468 	if (new_flags >= 0)
1469 		map->flags = new_flags;
1470 	if (new_pool_max >= 0)
1471 		map->pool_max = new_pool_max;
1472 
1473 	/* new max? */
1474 	ceph_decode_32_safe(p, end, max, e_inval);
1475 	if (max >= 0) {
1476 		err = osdmap_set_max_osd(map, max);
1477 		if (err)
1478 			goto bad;
1479 	}
1480 
1481 	map->epoch++;
1482 	map->modified = modified;
1483 
1484 	/* new_pools */
1485 	err = decode_new_pools(p, end, map);
1486 	if (err)
1487 		goto bad;
1488 
1489 	/* new_pool_names */
1490 	err = decode_pool_names(p, end, map);
1491 	if (err)
1492 		goto bad;
1493 
1494 	/* old_pool */
1495 	ceph_decode_32_safe(p, end, len, e_inval);
1496 	while (len--) {
1497 		struct ceph_pg_pool_info *pi;
1498 
1499 		ceph_decode_64_safe(p, end, pool, e_inval);
1500 		pi = __lookup_pg_pool(&map->pg_pools, pool);
1501 		if (pi)
1502 			__remove_pg_pool(&map->pg_pools, pi);
1503 	}
1504 
1505 	/* new_up_client, new_state, new_weight */
1506 	err = decode_new_up_state_weight(p, end, map);
1507 	if (err)
1508 		goto bad;
1509 
1510 	/* new_pg_temp */
1511 	err = decode_new_pg_temp(p, end, map);
1512 	if (err)
1513 		goto bad;
1514 
1515 	/* new_primary_temp */
1516 	if (struct_v >= 1) {
1517 		err = decode_new_primary_temp(p, end, map);
1518 		if (err)
1519 			goto bad;
1520 	}
1521 
1522 	/* new_primary_affinity */
1523 	if (struct_v >= 2) {
1524 		err = decode_new_primary_affinity(p, end, map);
1525 		if (err)
1526 			goto bad;
1527 	}
1528 
1529 	/* ignore the rest */
1530 	*p = end;
1531 
1532 	dout("inc osdmap epoch %d max_osd %d\n", map->epoch, map->max_osd);
1533 	return map;
1534 
1535 e_inval:
1536 	err = -EINVAL;
1537 bad:
1538 	pr_err("corrupt inc osdmap (%d) epoch %d off %d (%p of %p-%p)\n",
1539 	       err, epoch, (int)(*p - start), *p, start, end);
1540 	print_hex_dump(KERN_DEBUG, "osdmap: ",
1541 		       DUMP_PREFIX_OFFSET, 16, 1,
1542 		       start, end - start, true);
1543 	return ERR_PTR(err);
1544 }
1545 
1546 void ceph_oloc_copy(struct ceph_object_locator *dest,
1547 		    const struct ceph_object_locator *src)
1548 {
1549 	WARN_ON(!ceph_oloc_empty(dest));
1550 	WARN_ON(dest->pool_ns); /* empty() only covers ->pool */
1551 
1552 	dest->pool = src->pool;
1553 	if (src->pool_ns)
1554 		dest->pool_ns = ceph_get_string(src->pool_ns);
1555 }
1556 EXPORT_SYMBOL(ceph_oloc_copy);
1557 
1558 void ceph_oloc_destroy(struct ceph_object_locator *oloc)
1559 {
1560 	ceph_put_string(oloc->pool_ns);
1561 }
1562 EXPORT_SYMBOL(ceph_oloc_destroy);
1563 
1564 void ceph_oid_copy(struct ceph_object_id *dest,
1565 		   const struct ceph_object_id *src)
1566 {
1567 	WARN_ON(!ceph_oid_empty(dest));
1568 
1569 	if (src->name != src->inline_name) {
1570 		/* very rare, see ceph_object_id definition */
1571 		dest->name = kmalloc(src->name_len + 1,
1572 				     GFP_NOIO | __GFP_NOFAIL);
1573 	}
1574 
1575 	memcpy(dest->name, src->name, src->name_len + 1);
1576 	dest->name_len = src->name_len;
1577 }
1578 EXPORT_SYMBOL(ceph_oid_copy);
1579 
1580 static __printf(2, 0)
1581 int oid_printf_vargs(struct ceph_object_id *oid, const char *fmt, va_list ap)
1582 {
1583 	int len;
1584 
1585 	WARN_ON(!ceph_oid_empty(oid));
1586 
1587 	len = vsnprintf(oid->inline_name, sizeof(oid->inline_name), fmt, ap);
1588 	if (len >= sizeof(oid->inline_name))
1589 		return len;
1590 
1591 	oid->name_len = len;
1592 	return 0;
1593 }
1594 
1595 /*
1596  * If oid doesn't fit into inline buffer, BUG.
1597  */
1598 void ceph_oid_printf(struct ceph_object_id *oid, const char *fmt, ...)
1599 {
1600 	va_list ap;
1601 
1602 	va_start(ap, fmt);
1603 	BUG_ON(oid_printf_vargs(oid, fmt, ap));
1604 	va_end(ap);
1605 }
1606 EXPORT_SYMBOL(ceph_oid_printf);
1607 
1608 static __printf(3, 0)
1609 int oid_aprintf_vargs(struct ceph_object_id *oid, gfp_t gfp,
1610 		      const char *fmt, va_list ap)
1611 {
1612 	va_list aq;
1613 	int len;
1614 
1615 	va_copy(aq, ap);
1616 	len = oid_printf_vargs(oid, fmt, aq);
1617 	va_end(aq);
1618 
1619 	if (len) {
1620 		char *external_name;
1621 
1622 		external_name = kmalloc(len + 1, gfp);
1623 		if (!external_name)
1624 			return -ENOMEM;
1625 
1626 		oid->name = external_name;
1627 		WARN_ON(vsnprintf(oid->name, len + 1, fmt, ap) != len);
1628 		oid->name_len = len;
1629 	}
1630 
1631 	return 0;
1632 }
1633 
1634 /*
1635  * If oid doesn't fit into inline buffer, allocate.
1636  */
1637 int ceph_oid_aprintf(struct ceph_object_id *oid, gfp_t gfp,
1638 		     const char *fmt, ...)
1639 {
1640 	va_list ap;
1641 	int ret;
1642 
1643 	va_start(ap, fmt);
1644 	ret = oid_aprintf_vargs(oid, gfp, fmt, ap);
1645 	va_end(ap);
1646 
1647 	return ret;
1648 }
1649 EXPORT_SYMBOL(ceph_oid_aprintf);
1650 
1651 void ceph_oid_destroy(struct ceph_object_id *oid)
1652 {
1653 	if (oid->name != oid->inline_name)
1654 		kfree(oid->name);
1655 }
1656 EXPORT_SYMBOL(ceph_oid_destroy);
1657 
1658 /*
1659  * osds only
1660  */
1661 static bool __osds_equal(const struct ceph_osds *lhs,
1662 			 const struct ceph_osds *rhs)
1663 {
1664 	if (lhs->size == rhs->size &&
1665 	    !memcmp(lhs->osds, rhs->osds, rhs->size * sizeof(rhs->osds[0])))
1666 		return true;
1667 
1668 	return false;
1669 }
1670 
1671 /*
1672  * osds + primary
1673  */
1674 static bool osds_equal(const struct ceph_osds *lhs,
1675 		       const struct ceph_osds *rhs)
1676 {
1677 	if (__osds_equal(lhs, rhs) &&
1678 	    lhs->primary == rhs->primary)
1679 		return true;
1680 
1681 	return false;
1682 }
1683 
1684 static bool osds_valid(const struct ceph_osds *set)
1685 {
1686 	/* non-empty set */
1687 	if (set->size > 0 && set->primary >= 0)
1688 		return true;
1689 
1690 	/* empty can_shift_osds set */
1691 	if (!set->size && set->primary == -1)
1692 		return true;
1693 
1694 	/* empty !can_shift_osds set - all NONE */
1695 	if (set->size > 0 && set->primary == -1) {
1696 		int i;
1697 
1698 		for (i = 0; i < set->size; i++) {
1699 			if (set->osds[i] != CRUSH_ITEM_NONE)
1700 				break;
1701 		}
1702 		if (i == set->size)
1703 			return true;
1704 	}
1705 
1706 	return false;
1707 }
1708 
1709 void ceph_osds_copy(struct ceph_osds *dest, const struct ceph_osds *src)
1710 {
1711 	memcpy(dest->osds, src->osds, src->size * sizeof(src->osds[0]));
1712 	dest->size = src->size;
1713 	dest->primary = src->primary;
1714 }
1715 
1716 static bool is_split(const struct ceph_pg *pgid,
1717 		     u32 old_pg_num,
1718 		     u32 new_pg_num)
1719 {
1720 	int old_bits = calc_bits_of(old_pg_num);
1721 	int old_mask = (1 << old_bits) - 1;
1722 	int n;
1723 
1724 	WARN_ON(pgid->seed >= old_pg_num);
1725 	if (new_pg_num <= old_pg_num)
1726 		return false;
1727 
1728 	for (n = 1; ; n++) {
1729 		int next_bit = n << (old_bits - 1);
1730 		u32 s = next_bit | pgid->seed;
1731 
1732 		if (s < old_pg_num || s == pgid->seed)
1733 			continue;
1734 		if (s >= new_pg_num)
1735 			break;
1736 
1737 		s = ceph_stable_mod(s, old_pg_num, old_mask);
1738 		if (s == pgid->seed)
1739 			return true;
1740 	}
1741 
1742 	return false;
1743 }
1744 
1745 bool ceph_is_new_interval(const struct ceph_osds *old_acting,
1746 			  const struct ceph_osds *new_acting,
1747 			  const struct ceph_osds *old_up,
1748 			  const struct ceph_osds *new_up,
1749 			  int old_size,
1750 			  int new_size,
1751 			  int old_min_size,
1752 			  int new_min_size,
1753 			  u32 old_pg_num,
1754 			  u32 new_pg_num,
1755 			  bool old_sort_bitwise,
1756 			  bool new_sort_bitwise,
1757 			  const struct ceph_pg *pgid)
1758 {
1759 	return !osds_equal(old_acting, new_acting) ||
1760 	       !osds_equal(old_up, new_up) ||
1761 	       old_size != new_size ||
1762 	       old_min_size != new_min_size ||
1763 	       is_split(pgid, old_pg_num, new_pg_num) ||
1764 	       old_sort_bitwise != new_sort_bitwise;
1765 }
1766 
1767 static int calc_pg_rank(int osd, const struct ceph_osds *acting)
1768 {
1769 	int i;
1770 
1771 	for (i = 0; i < acting->size; i++) {
1772 		if (acting->osds[i] == osd)
1773 			return i;
1774 	}
1775 
1776 	return -1;
1777 }
1778 
1779 static bool primary_changed(const struct ceph_osds *old_acting,
1780 			    const struct ceph_osds *new_acting)
1781 {
1782 	if (!old_acting->size && !new_acting->size)
1783 		return false; /* both still empty */
1784 
1785 	if (!old_acting->size ^ !new_acting->size)
1786 		return true; /* was empty, now not, or vice versa */
1787 
1788 	if (old_acting->primary != new_acting->primary)
1789 		return true; /* primary changed */
1790 
1791 	if (calc_pg_rank(old_acting->primary, old_acting) !=
1792 	    calc_pg_rank(new_acting->primary, new_acting))
1793 		return true;
1794 
1795 	return false; /* same primary (tho replicas may have changed) */
1796 }
1797 
1798 bool ceph_osds_changed(const struct ceph_osds *old_acting,
1799 		       const struct ceph_osds *new_acting,
1800 		       bool any_change)
1801 {
1802 	if (primary_changed(old_acting, new_acting))
1803 		return true;
1804 
1805 	if (any_change && !__osds_equal(old_acting, new_acting))
1806 		return true;
1807 
1808 	return false;
1809 }
1810 
1811 /*
1812  * calculate file layout from given offset, length.
1813  * fill in correct oid, logical length, and object extent
1814  * offset, length.
1815  *
1816  * for now, we write only a single su, until we can
1817  * pass a stride back to the caller.
1818  */
1819 int ceph_calc_file_object_mapping(struct ceph_file_layout *layout,
1820 				   u64 off, u64 len,
1821 				   u64 *ono,
1822 				   u64 *oxoff, u64 *oxlen)
1823 {
1824 	u32 osize = layout->object_size;
1825 	u32 su = layout->stripe_unit;
1826 	u32 sc = layout->stripe_count;
1827 	u32 bl, stripeno, stripepos, objsetno;
1828 	u32 su_per_object;
1829 	u64 t, su_offset;
1830 
1831 	dout("mapping %llu~%llu  osize %u fl_su %u\n", off, len,
1832 	     osize, su);
1833 	if (su == 0 || sc == 0)
1834 		goto invalid;
1835 	su_per_object = osize / su;
1836 	if (su_per_object == 0)
1837 		goto invalid;
1838 	dout("osize %u / su %u = su_per_object %u\n", osize, su,
1839 	     su_per_object);
1840 
1841 	if ((su & ~PAGE_MASK) != 0)
1842 		goto invalid;
1843 
1844 	/* bl = *off / su; */
1845 	t = off;
1846 	do_div(t, su);
1847 	bl = t;
1848 	dout("off %llu / su %u = bl %u\n", off, su, bl);
1849 
1850 	stripeno = bl / sc;
1851 	stripepos = bl % sc;
1852 	objsetno = stripeno / su_per_object;
1853 
1854 	*ono = objsetno * sc + stripepos;
1855 	dout("objset %u * sc %u = ono %u\n", objsetno, sc, (unsigned int)*ono);
1856 
1857 	/* *oxoff = *off % layout->fl_stripe_unit;  # offset in su */
1858 	t = off;
1859 	su_offset = do_div(t, su);
1860 	*oxoff = su_offset + (stripeno % su_per_object) * su;
1861 
1862 	/*
1863 	 * Calculate the length of the extent being written to the selected
1864 	 * object. This is the minimum of the full length requested (len) or
1865 	 * the remainder of the current stripe being written to.
1866 	 */
1867 	*oxlen = min_t(u64, len, su - su_offset);
1868 
1869 	dout(" obj extent %llu~%llu\n", *oxoff, *oxlen);
1870 	return 0;
1871 
1872 invalid:
1873 	dout(" invalid layout\n");
1874 	*ono = 0;
1875 	*oxoff = 0;
1876 	*oxlen = 0;
1877 	return -EINVAL;
1878 }
1879 EXPORT_SYMBOL(ceph_calc_file_object_mapping);
1880 
1881 /*
1882  * Map an object into a PG.
1883  *
1884  * Should only be called with target_oid and target_oloc (as opposed to
1885  * base_oid and base_oloc), since tiering isn't taken into account.
1886  */
1887 int ceph_object_locator_to_pg(struct ceph_osdmap *osdmap,
1888 			      struct ceph_object_id *oid,
1889 			      struct ceph_object_locator *oloc,
1890 			      struct ceph_pg *raw_pgid)
1891 {
1892 	struct ceph_pg_pool_info *pi;
1893 
1894 	pi = ceph_pg_pool_by_id(osdmap, oloc->pool);
1895 	if (!pi)
1896 		return -ENOENT;
1897 
1898 	if (!oloc->pool_ns) {
1899 		raw_pgid->pool = oloc->pool;
1900 		raw_pgid->seed = ceph_str_hash(pi->object_hash, oid->name,
1901 					     oid->name_len);
1902 		dout("%s %s -> raw_pgid %llu.%x\n", __func__, oid->name,
1903 		     raw_pgid->pool, raw_pgid->seed);
1904 	} else {
1905 		char stack_buf[256];
1906 		char *buf = stack_buf;
1907 		int nsl = oloc->pool_ns->len;
1908 		size_t total = nsl + 1 + oid->name_len;
1909 
1910 		if (total > sizeof(stack_buf)) {
1911 			buf = kmalloc(total, GFP_NOIO);
1912 			if (!buf)
1913 				return -ENOMEM;
1914 		}
1915 		memcpy(buf, oloc->pool_ns->str, nsl);
1916 		buf[nsl] = '\037';
1917 		memcpy(buf + nsl + 1, oid->name, oid->name_len);
1918 		raw_pgid->pool = oloc->pool;
1919 		raw_pgid->seed = ceph_str_hash(pi->object_hash, buf, total);
1920 		if (buf != stack_buf)
1921 			kfree(buf);
1922 		dout("%s %s ns %.*s -> raw_pgid %llu.%x\n", __func__,
1923 		     oid->name, nsl, oloc->pool_ns->str,
1924 		     raw_pgid->pool, raw_pgid->seed);
1925 	}
1926 	return 0;
1927 }
1928 EXPORT_SYMBOL(ceph_object_locator_to_pg);
1929 
1930 /*
1931  * Map a raw PG (full precision ps) into an actual PG.
1932  */
1933 static void raw_pg_to_pg(struct ceph_pg_pool_info *pi,
1934 			 const struct ceph_pg *raw_pgid,
1935 			 struct ceph_pg *pgid)
1936 {
1937 	pgid->pool = raw_pgid->pool;
1938 	pgid->seed = ceph_stable_mod(raw_pgid->seed, pi->pg_num,
1939 				     pi->pg_num_mask);
1940 }
1941 
1942 /*
1943  * Map a raw PG (full precision ps) into a placement ps (placement
1944  * seed).  Include pool id in that value so that different pools don't
1945  * use the same seeds.
1946  */
1947 static u32 raw_pg_to_pps(struct ceph_pg_pool_info *pi,
1948 			 const struct ceph_pg *raw_pgid)
1949 {
1950 	if (pi->flags & CEPH_POOL_FLAG_HASHPSPOOL) {
1951 		/* hash pool id and seed so that pool PGs do not overlap */
1952 		return crush_hash32_2(CRUSH_HASH_RJENKINS1,
1953 				      ceph_stable_mod(raw_pgid->seed,
1954 						      pi->pgp_num,
1955 						      pi->pgp_num_mask),
1956 				      raw_pgid->pool);
1957 	} else {
1958 		/*
1959 		 * legacy behavior: add ps and pool together.  this is
1960 		 * not a great approach because the PGs from each pool
1961 		 * will overlap on top of each other: 0.5 == 1.4 ==
1962 		 * 2.3 == ...
1963 		 */
1964 		return ceph_stable_mod(raw_pgid->seed, pi->pgp_num,
1965 				       pi->pgp_num_mask) +
1966 		       (unsigned)raw_pgid->pool;
1967 	}
1968 }
1969 
1970 static int do_crush(struct ceph_osdmap *map, int ruleno, int x,
1971 		    int *result, int result_max,
1972 		    const __u32 *weight, int weight_max)
1973 {
1974 	int r;
1975 
1976 	BUG_ON(result_max > CEPH_PG_MAX_SIZE);
1977 
1978 	mutex_lock(&map->crush_workspace_mutex);
1979 	r = crush_do_rule(map->crush, ruleno, x, result, result_max,
1980 			  weight, weight_max, map->crush_workspace);
1981 	mutex_unlock(&map->crush_workspace_mutex);
1982 
1983 	return r;
1984 }
1985 
1986 /*
1987  * Calculate raw set (CRUSH output) for given PG.  The result may
1988  * contain nonexistent OSDs.  ->primary is undefined for a raw set.
1989  *
1990  * Placement seed (CRUSH input) is returned through @ppps.
1991  */
1992 static void pg_to_raw_osds(struct ceph_osdmap *osdmap,
1993 			   struct ceph_pg_pool_info *pi,
1994 			   const struct ceph_pg *raw_pgid,
1995 			   struct ceph_osds *raw,
1996 			   u32 *ppps)
1997 {
1998 	u32 pps = raw_pg_to_pps(pi, raw_pgid);
1999 	int ruleno;
2000 	int len;
2001 
2002 	ceph_osds_init(raw);
2003 	if (ppps)
2004 		*ppps = pps;
2005 
2006 	ruleno = crush_find_rule(osdmap->crush, pi->crush_ruleset, pi->type,
2007 				 pi->size);
2008 	if (ruleno < 0) {
2009 		pr_err("no crush rule: pool %lld ruleset %d type %d size %d\n",
2010 		       pi->id, pi->crush_ruleset, pi->type, pi->size);
2011 		return;
2012 	}
2013 
2014 	if (pi->size > ARRAY_SIZE(raw->osds)) {
2015 		pr_err_ratelimited("pool %lld ruleset %d type %d too wide: size %d > %zu\n",
2016 		       pi->id, pi->crush_ruleset, pi->type, pi->size,
2017 		       ARRAY_SIZE(raw->osds));
2018 		return;
2019 	}
2020 
2021 	len = do_crush(osdmap, ruleno, pps, raw->osds, pi->size,
2022 		       osdmap->osd_weight, osdmap->max_osd);
2023 	if (len < 0) {
2024 		pr_err("error %d from crush rule %d: pool %lld ruleset %d type %d size %d\n",
2025 		       len, ruleno, pi->id, pi->crush_ruleset, pi->type,
2026 		       pi->size);
2027 		return;
2028 	}
2029 
2030 	raw->size = len;
2031 }
2032 
2033 /*
2034  * Given raw set, calculate up set and up primary.  By definition of an
2035  * up set, the result won't contain nonexistent or down OSDs.
2036  *
2037  * This is done in-place - on return @set is the up set.  If it's
2038  * empty, ->primary will remain undefined.
2039  */
2040 static void raw_to_up_osds(struct ceph_osdmap *osdmap,
2041 			   struct ceph_pg_pool_info *pi,
2042 			   struct ceph_osds *set)
2043 {
2044 	int i;
2045 
2046 	/* ->primary is undefined for a raw set */
2047 	BUG_ON(set->primary != -1);
2048 
2049 	if (ceph_can_shift_osds(pi)) {
2050 		int removed = 0;
2051 
2052 		/* shift left */
2053 		for (i = 0; i < set->size; i++) {
2054 			if (ceph_osd_is_down(osdmap, set->osds[i])) {
2055 				removed++;
2056 				continue;
2057 			}
2058 			if (removed)
2059 				set->osds[i - removed] = set->osds[i];
2060 		}
2061 		set->size -= removed;
2062 		if (set->size > 0)
2063 			set->primary = set->osds[0];
2064 	} else {
2065 		/* set down/dne devices to NONE */
2066 		for (i = set->size - 1; i >= 0; i--) {
2067 			if (ceph_osd_is_down(osdmap, set->osds[i]))
2068 				set->osds[i] = CRUSH_ITEM_NONE;
2069 			else
2070 				set->primary = set->osds[i];
2071 		}
2072 	}
2073 }
2074 
2075 static void apply_primary_affinity(struct ceph_osdmap *osdmap,
2076 				   struct ceph_pg_pool_info *pi,
2077 				   u32 pps,
2078 				   struct ceph_osds *up)
2079 {
2080 	int i;
2081 	int pos = -1;
2082 
2083 	/*
2084 	 * Do we have any non-default primary_affinity values for these
2085 	 * osds?
2086 	 */
2087 	if (!osdmap->osd_primary_affinity)
2088 		return;
2089 
2090 	for (i = 0; i < up->size; i++) {
2091 		int osd = up->osds[i];
2092 
2093 		if (osd != CRUSH_ITEM_NONE &&
2094 		    osdmap->osd_primary_affinity[osd] !=
2095 					CEPH_OSD_DEFAULT_PRIMARY_AFFINITY) {
2096 			break;
2097 		}
2098 	}
2099 	if (i == up->size)
2100 		return;
2101 
2102 	/*
2103 	 * Pick the primary.  Feed both the seed (for the pg) and the
2104 	 * osd into the hash/rng so that a proportional fraction of an
2105 	 * osd's pgs get rejected as primary.
2106 	 */
2107 	for (i = 0; i < up->size; i++) {
2108 		int osd = up->osds[i];
2109 		u32 aff;
2110 
2111 		if (osd == CRUSH_ITEM_NONE)
2112 			continue;
2113 
2114 		aff = osdmap->osd_primary_affinity[osd];
2115 		if (aff < CEPH_OSD_MAX_PRIMARY_AFFINITY &&
2116 		    (crush_hash32_2(CRUSH_HASH_RJENKINS1,
2117 				    pps, osd) >> 16) >= aff) {
2118 			/*
2119 			 * We chose not to use this primary.  Note it
2120 			 * anyway as a fallback in case we don't pick
2121 			 * anyone else, but keep looking.
2122 			 */
2123 			if (pos < 0)
2124 				pos = i;
2125 		} else {
2126 			pos = i;
2127 			break;
2128 		}
2129 	}
2130 	if (pos < 0)
2131 		return;
2132 
2133 	up->primary = up->osds[pos];
2134 
2135 	if (ceph_can_shift_osds(pi) && pos > 0) {
2136 		/* move the new primary to the front */
2137 		for (i = pos; i > 0; i--)
2138 			up->osds[i] = up->osds[i - 1];
2139 		up->osds[0] = up->primary;
2140 	}
2141 }
2142 
2143 /*
2144  * Get pg_temp and primary_temp mappings for given PG.
2145  *
2146  * Note that a PG may have none, only pg_temp, only primary_temp or
2147  * both pg_temp and primary_temp mappings.  This means @temp isn't
2148  * always a valid OSD set on return: in the "only primary_temp" case,
2149  * @temp will have its ->primary >= 0 but ->size == 0.
2150  */
2151 static void get_temp_osds(struct ceph_osdmap *osdmap,
2152 			  struct ceph_pg_pool_info *pi,
2153 			  const struct ceph_pg *raw_pgid,
2154 			  struct ceph_osds *temp)
2155 {
2156 	struct ceph_pg pgid;
2157 	struct ceph_pg_mapping *pg;
2158 	int i;
2159 
2160 	raw_pg_to_pg(pi, raw_pgid, &pgid);
2161 	ceph_osds_init(temp);
2162 
2163 	/* pg_temp? */
2164 	pg = __lookup_pg_mapping(&osdmap->pg_temp, pgid);
2165 	if (pg) {
2166 		for (i = 0; i < pg->pg_temp.len; i++) {
2167 			if (ceph_osd_is_down(osdmap, pg->pg_temp.osds[i])) {
2168 				if (ceph_can_shift_osds(pi))
2169 					continue;
2170 
2171 				temp->osds[temp->size++] = CRUSH_ITEM_NONE;
2172 			} else {
2173 				temp->osds[temp->size++] = pg->pg_temp.osds[i];
2174 			}
2175 		}
2176 
2177 		/* apply pg_temp's primary */
2178 		for (i = 0; i < temp->size; i++) {
2179 			if (temp->osds[i] != CRUSH_ITEM_NONE) {
2180 				temp->primary = temp->osds[i];
2181 				break;
2182 			}
2183 		}
2184 	}
2185 
2186 	/* primary_temp? */
2187 	pg = __lookup_pg_mapping(&osdmap->primary_temp, pgid);
2188 	if (pg)
2189 		temp->primary = pg->primary_temp.osd;
2190 }
2191 
2192 /*
2193  * Map a PG to its acting set as well as its up set.
2194  *
2195  * Acting set is used for data mapping purposes, while up set can be
2196  * recorded for detecting interval changes and deciding whether to
2197  * resend a request.
2198  */
2199 void ceph_pg_to_up_acting_osds(struct ceph_osdmap *osdmap,
2200 			       const struct ceph_pg *raw_pgid,
2201 			       struct ceph_osds *up,
2202 			       struct ceph_osds *acting)
2203 {
2204 	struct ceph_pg_pool_info *pi;
2205 	u32 pps;
2206 
2207 	pi = ceph_pg_pool_by_id(osdmap, raw_pgid->pool);
2208 	if (!pi) {
2209 		ceph_osds_init(up);
2210 		ceph_osds_init(acting);
2211 		goto out;
2212 	}
2213 
2214 	pg_to_raw_osds(osdmap, pi, raw_pgid, up, &pps);
2215 	raw_to_up_osds(osdmap, pi, up);
2216 	apply_primary_affinity(osdmap, pi, pps, up);
2217 	get_temp_osds(osdmap, pi, raw_pgid, acting);
2218 	if (!acting->size) {
2219 		memcpy(acting->osds, up->osds, up->size * sizeof(up->osds[0]));
2220 		acting->size = up->size;
2221 		if (acting->primary == -1)
2222 			acting->primary = up->primary;
2223 	}
2224 out:
2225 	WARN_ON(!osds_valid(up) || !osds_valid(acting));
2226 }
2227 
2228 /*
2229  * Return acting primary for given PG, or -1 if none.
2230  */
2231 int ceph_pg_to_acting_primary(struct ceph_osdmap *osdmap,
2232 			      const struct ceph_pg *raw_pgid)
2233 {
2234 	struct ceph_osds up, acting;
2235 
2236 	ceph_pg_to_up_acting_osds(osdmap, raw_pgid, &up, &acting);
2237 	return acting.primary;
2238 }
2239 EXPORT_SYMBOL(ceph_pg_to_acting_primary);
2240