xref: /openbmc/linux/net/ceph/osdmap.c (revision 9cfc5c90)
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",
346              c->choose_local_tries);
347         dout("crush decode tunable choose_local_fallback_tries = %d",
348              c->choose_local_fallback_tries);
349         dout("crush decode tunable choose_total_tries = %d",
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",
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",
360 	     c->chooseleaf_vary_r);
361 
362 done:
363 	dout("crush_decode success\n");
364 	return c;
365 
366 badmem:
367 	err = -ENOMEM;
368 bad:
369 	dout("crush_decode fail %d\n", err);
370 	crush_destroy(c);
371 	return ERR_PTR(err);
372 }
373 
374 /*
375  * rbtree of pg_mapping for handling pg_temp (explicit mapping of pgid
376  * to a set of osds) and primary_temp (explicit primary setting)
377  */
378 static int pgid_cmp(struct ceph_pg l, struct ceph_pg r)
379 {
380 	if (l.pool < r.pool)
381 		return -1;
382 	if (l.pool > r.pool)
383 		return 1;
384 	if (l.seed < r.seed)
385 		return -1;
386 	if (l.seed > r.seed)
387 		return 1;
388 	return 0;
389 }
390 
391 static int __insert_pg_mapping(struct ceph_pg_mapping *new,
392 			       struct rb_root *root)
393 {
394 	struct rb_node **p = &root->rb_node;
395 	struct rb_node *parent = NULL;
396 	struct ceph_pg_mapping *pg = NULL;
397 	int c;
398 
399 	dout("__insert_pg_mapping %llx %p\n", *(u64 *)&new->pgid, new);
400 	while (*p) {
401 		parent = *p;
402 		pg = rb_entry(parent, struct ceph_pg_mapping, node);
403 		c = pgid_cmp(new->pgid, pg->pgid);
404 		if (c < 0)
405 			p = &(*p)->rb_left;
406 		else if (c > 0)
407 			p = &(*p)->rb_right;
408 		else
409 			return -EEXIST;
410 	}
411 
412 	rb_link_node(&new->node, parent, p);
413 	rb_insert_color(&new->node, root);
414 	return 0;
415 }
416 
417 static struct ceph_pg_mapping *__lookup_pg_mapping(struct rb_root *root,
418 						   struct ceph_pg pgid)
419 {
420 	struct rb_node *n = root->rb_node;
421 	struct ceph_pg_mapping *pg;
422 	int c;
423 
424 	while (n) {
425 		pg = rb_entry(n, struct ceph_pg_mapping, node);
426 		c = pgid_cmp(pgid, pg->pgid);
427 		if (c < 0) {
428 			n = n->rb_left;
429 		} else if (c > 0) {
430 			n = n->rb_right;
431 		} else {
432 			dout("__lookup_pg_mapping %lld.%x got %p\n",
433 			     pgid.pool, pgid.seed, pg);
434 			return pg;
435 		}
436 	}
437 	return NULL;
438 }
439 
440 static int __remove_pg_mapping(struct rb_root *root, struct ceph_pg pgid)
441 {
442 	struct ceph_pg_mapping *pg = __lookup_pg_mapping(root, pgid);
443 
444 	if (pg) {
445 		dout("__remove_pg_mapping %lld.%x %p\n", pgid.pool, pgid.seed,
446 		     pg);
447 		rb_erase(&pg->node, root);
448 		kfree(pg);
449 		return 0;
450 	}
451 	dout("__remove_pg_mapping %lld.%x dne\n", pgid.pool, pgid.seed);
452 	return -ENOENT;
453 }
454 
455 /*
456  * rbtree of pg pool info
457  */
458 static int __insert_pg_pool(struct rb_root *root, struct ceph_pg_pool_info *new)
459 {
460 	struct rb_node **p = &root->rb_node;
461 	struct rb_node *parent = NULL;
462 	struct ceph_pg_pool_info *pi = NULL;
463 
464 	while (*p) {
465 		parent = *p;
466 		pi = rb_entry(parent, struct ceph_pg_pool_info, node);
467 		if (new->id < pi->id)
468 			p = &(*p)->rb_left;
469 		else if (new->id > pi->id)
470 			p = &(*p)->rb_right;
471 		else
472 			return -EEXIST;
473 	}
474 
475 	rb_link_node(&new->node, parent, p);
476 	rb_insert_color(&new->node, root);
477 	return 0;
478 }
479 
480 static struct ceph_pg_pool_info *__lookup_pg_pool(struct rb_root *root, u64 id)
481 {
482 	struct ceph_pg_pool_info *pi;
483 	struct rb_node *n = root->rb_node;
484 
485 	while (n) {
486 		pi = rb_entry(n, struct ceph_pg_pool_info, node);
487 		if (id < pi->id)
488 			n = n->rb_left;
489 		else if (id > pi->id)
490 			n = n->rb_right;
491 		else
492 			return pi;
493 	}
494 	return NULL;
495 }
496 
497 struct ceph_pg_pool_info *ceph_pg_pool_by_id(struct ceph_osdmap *map, u64 id)
498 {
499 	return __lookup_pg_pool(&map->pg_pools, id);
500 }
501 
502 const char *ceph_pg_pool_name_by_id(struct ceph_osdmap *map, u64 id)
503 {
504 	struct ceph_pg_pool_info *pi;
505 
506 	if (id == CEPH_NOPOOL)
507 		return NULL;
508 
509 	if (WARN_ON_ONCE(id > (u64) INT_MAX))
510 		return NULL;
511 
512 	pi = __lookup_pg_pool(&map->pg_pools, (int) id);
513 
514 	return pi ? pi->name : NULL;
515 }
516 EXPORT_SYMBOL(ceph_pg_pool_name_by_id);
517 
518 int ceph_pg_poolid_by_name(struct ceph_osdmap *map, const char *name)
519 {
520 	struct rb_node *rbp;
521 
522 	for (rbp = rb_first(&map->pg_pools); rbp; rbp = rb_next(rbp)) {
523 		struct ceph_pg_pool_info *pi =
524 			rb_entry(rbp, struct ceph_pg_pool_info, node);
525 		if (pi->name && strcmp(pi->name, name) == 0)
526 			return pi->id;
527 	}
528 	return -ENOENT;
529 }
530 EXPORT_SYMBOL(ceph_pg_poolid_by_name);
531 
532 static void __remove_pg_pool(struct rb_root *root, struct ceph_pg_pool_info *pi)
533 {
534 	rb_erase(&pi->node, root);
535 	kfree(pi->name);
536 	kfree(pi);
537 }
538 
539 static int decode_pool(void **p, void *end, struct ceph_pg_pool_info *pi)
540 {
541 	u8 ev, cv;
542 	unsigned len, num;
543 	void *pool_end;
544 
545 	ceph_decode_need(p, end, 2 + 4, bad);
546 	ev = ceph_decode_8(p);  /* encoding version */
547 	cv = ceph_decode_8(p); /* compat version */
548 	if (ev < 5) {
549 		pr_warn("got v %d < 5 cv %d of ceph_pg_pool\n", ev, cv);
550 		return -EINVAL;
551 	}
552 	if (cv > 9) {
553 		pr_warn("got v %d cv %d > 9 of ceph_pg_pool\n", ev, cv);
554 		return -EINVAL;
555 	}
556 	len = ceph_decode_32(p);
557 	ceph_decode_need(p, end, len, bad);
558 	pool_end = *p + len;
559 
560 	pi->type = ceph_decode_8(p);
561 	pi->size = ceph_decode_8(p);
562 	pi->crush_ruleset = ceph_decode_8(p);
563 	pi->object_hash = ceph_decode_8(p);
564 
565 	pi->pg_num = ceph_decode_32(p);
566 	pi->pgp_num = ceph_decode_32(p);
567 
568 	*p += 4 + 4;  /* skip lpg* */
569 	*p += 4;      /* skip last_change */
570 	*p += 8 + 4;  /* skip snap_seq, snap_epoch */
571 
572 	/* skip snaps */
573 	num = ceph_decode_32(p);
574 	while (num--) {
575 		*p += 8;  /* snapid key */
576 		*p += 1 + 1; /* versions */
577 		len = ceph_decode_32(p);
578 		*p += len;
579 	}
580 
581 	/* skip removed_snaps */
582 	num = ceph_decode_32(p);
583 	*p += num * (8 + 8);
584 
585 	*p += 8;  /* skip auid */
586 	pi->flags = ceph_decode_64(p);
587 	*p += 4;  /* skip crash_replay_interval */
588 
589 	if (ev >= 7)
590 		*p += 1;  /* skip min_size */
591 
592 	if (ev >= 8)
593 		*p += 8 + 8;  /* skip quota_max_* */
594 
595 	if (ev >= 9) {
596 		/* skip tiers */
597 		num = ceph_decode_32(p);
598 		*p += num * 8;
599 
600 		*p += 8;  /* skip tier_of */
601 		*p += 1;  /* skip cache_mode */
602 
603 		pi->read_tier = ceph_decode_64(p);
604 		pi->write_tier = ceph_decode_64(p);
605 	} else {
606 		pi->read_tier = -1;
607 		pi->write_tier = -1;
608 	}
609 
610 	/* ignore the rest */
611 
612 	*p = pool_end;
613 	calc_pg_masks(pi);
614 	return 0;
615 
616 bad:
617 	return -EINVAL;
618 }
619 
620 static int decode_pool_names(void **p, void *end, struct ceph_osdmap *map)
621 {
622 	struct ceph_pg_pool_info *pi;
623 	u32 num, len;
624 	u64 pool;
625 
626 	ceph_decode_32_safe(p, end, num, bad);
627 	dout(" %d pool names\n", num);
628 	while (num--) {
629 		ceph_decode_64_safe(p, end, pool, bad);
630 		ceph_decode_32_safe(p, end, len, bad);
631 		dout("  pool %llu len %d\n", pool, len);
632 		ceph_decode_need(p, end, len, bad);
633 		pi = __lookup_pg_pool(&map->pg_pools, pool);
634 		if (pi) {
635 			char *name = kstrndup(*p, len, GFP_NOFS);
636 
637 			if (!name)
638 				return -ENOMEM;
639 			kfree(pi->name);
640 			pi->name = name;
641 			dout("  name is %s\n", pi->name);
642 		}
643 		*p += len;
644 	}
645 	return 0;
646 
647 bad:
648 	return -EINVAL;
649 }
650 
651 /*
652  * osd map
653  */
654 void ceph_osdmap_destroy(struct ceph_osdmap *map)
655 {
656 	dout("osdmap_destroy %p\n", map);
657 	if (map->crush)
658 		crush_destroy(map->crush);
659 	while (!RB_EMPTY_ROOT(&map->pg_temp)) {
660 		struct ceph_pg_mapping *pg =
661 			rb_entry(rb_first(&map->pg_temp),
662 				 struct ceph_pg_mapping, node);
663 		rb_erase(&pg->node, &map->pg_temp);
664 		kfree(pg);
665 	}
666 	while (!RB_EMPTY_ROOT(&map->primary_temp)) {
667 		struct ceph_pg_mapping *pg =
668 			rb_entry(rb_first(&map->primary_temp),
669 				 struct ceph_pg_mapping, node);
670 		rb_erase(&pg->node, &map->primary_temp);
671 		kfree(pg);
672 	}
673 	while (!RB_EMPTY_ROOT(&map->pg_pools)) {
674 		struct ceph_pg_pool_info *pi =
675 			rb_entry(rb_first(&map->pg_pools),
676 				 struct ceph_pg_pool_info, node);
677 		__remove_pg_pool(&map->pg_pools, pi);
678 	}
679 	kfree(map->osd_state);
680 	kfree(map->osd_weight);
681 	kfree(map->osd_addr);
682 	kfree(map->osd_primary_affinity);
683 	kfree(map);
684 }
685 
686 /*
687  * Adjust max_osd value, (re)allocate arrays.
688  *
689  * The new elements are properly initialized.
690  */
691 static int osdmap_set_max_osd(struct ceph_osdmap *map, int max)
692 {
693 	u8 *state;
694 	u32 *weight;
695 	struct ceph_entity_addr *addr;
696 	int i;
697 
698 	state = krealloc(map->osd_state, max*sizeof(*state), GFP_NOFS);
699 	if (!state)
700 		return -ENOMEM;
701 	map->osd_state = state;
702 
703 	weight = krealloc(map->osd_weight, max*sizeof(*weight), GFP_NOFS);
704 	if (!weight)
705 		return -ENOMEM;
706 	map->osd_weight = weight;
707 
708 	addr = krealloc(map->osd_addr, max*sizeof(*addr), GFP_NOFS);
709 	if (!addr)
710 		return -ENOMEM;
711 	map->osd_addr = addr;
712 
713 	for (i = map->max_osd; i < max; i++) {
714 		map->osd_state[i] = 0;
715 		map->osd_weight[i] = CEPH_OSD_OUT;
716 		memset(map->osd_addr + i, 0, sizeof(*map->osd_addr));
717 	}
718 
719 	if (map->osd_primary_affinity) {
720 		u32 *affinity;
721 
722 		affinity = krealloc(map->osd_primary_affinity,
723 				    max*sizeof(*affinity), GFP_NOFS);
724 		if (!affinity)
725 			return -ENOMEM;
726 		map->osd_primary_affinity = affinity;
727 
728 		for (i = map->max_osd; i < max; i++)
729 			map->osd_primary_affinity[i] =
730 			    CEPH_OSD_DEFAULT_PRIMARY_AFFINITY;
731 	}
732 
733 	map->max_osd = max;
734 
735 	return 0;
736 }
737 
738 #define OSDMAP_WRAPPER_COMPAT_VER	7
739 #define OSDMAP_CLIENT_DATA_COMPAT_VER	1
740 
741 /*
742  * Return 0 or error.  On success, *v is set to 0 for old (v6) osdmaps,
743  * to struct_v of the client_data section for new (v7 and above)
744  * osdmaps.
745  */
746 static int get_osdmap_client_data_v(void **p, void *end,
747 				    const char *prefix, u8 *v)
748 {
749 	u8 struct_v;
750 
751 	ceph_decode_8_safe(p, end, struct_v, e_inval);
752 	if (struct_v >= 7) {
753 		u8 struct_compat;
754 
755 		ceph_decode_8_safe(p, end, struct_compat, e_inval);
756 		if (struct_compat > OSDMAP_WRAPPER_COMPAT_VER) {
757 			pr_warn("got v %d cv %d > %d of %s ceph_osdmap\n",
758 				struct_v, struct_compat,
759 				OSDMAP_WRAPPER_COMPAT_VER, prefix);
760 			return -EINVAL;
761 		}
762 		*p += 4; /* ignore wrapper struct_len */
763 
764 		ceph_decode_8_safe(p, end, struct_v, e_inval);
765 		ceph_decode_8_safe(p, end, struct_compat, e_inval);
766 		if (struct_compat > OSDMAP_CLIENT_DATA_COMPAT_VER) {
767 			pr_warn("got v %d cv %d > %d of %s ceph_osdmap client data\n",
768 				struct_v, struct_compat,
769 				OSDMAP_CLIENT_DATA_COMPAT_VER, prefix);
770 			return -EINVAL;
771 		}
772 		*p += 4; /* ignore client data struct_len */
773 	} else {
774 		u16 version;
775 
776 		*p -= 1;
777 		ceph_decode_16_safe(p, end, version, e_inval);
778 		if (version < 6) {
779 			pr_warn("got v %d < 6 of %s ceph_osdmap\n",
780 				version, prefix);
781 			return -EINVAL;
782 		}
783 
784 		/* old osdmap enconding */
785 		struct_v = 0;
786 	}
787 
788 	*v = struct_v;
789 	return 0;
790 
791 e_inval:
792 	return -EINVAL;
793 }
794 
795 static int __decode_pools(void **p, void *end, struct ceph_osdmap *map,
796 			  bool incremental)
797 {
798 	u32 n;
799 
800 	ceph_decode_32_safe(p, end, n, e_inval);
801 	while (n--) {
802 		struct ceph_pg_pool_info *pi;
803 		u64 pool;
804 		int ret;
805 
806 		ceph_decode_64_safe(p, end, pool, e_inval);
807 
808 		pi = __lookup_pg_pool(&map->pg_pools, pool);
809 		if (!incremental || !pi) {
810 			pi = kzalloc(sizeof(*pi), GFP_NOFS);
811 			if (!pi)
812 				return -ENOMEM;
813 
814 			pi->id = pool;
815 
816 			ret = __insert_pg_pool(&map->pg_pools, pi);
817 			if (ret) {
818 				kfree(pi);
819 				return ret;
820 			}
821 		}
822 
823 		ret = decode_pool(p, end, pi);
824 		if (ret)
825 			return ret;
826 	}
827 
828 	return 0;
829 
830 e_inval:
831 	return -EINVAL;
832 }
833 
834 static int decode_pools(void **p, void *end, struct ceph_osdmap *map)
835 {
836 	return __decode_pools(p, end, map, false);
837 }
838 
839 static int decode_new_pools(void **p, void *end, struct ceph_osdmap *map)
840 {
841 	return __decode_pools(p, end, map, true);
842 }
843 
844 static int __decode_pg_temp(void **p, void *end, struct ceph_osdmap *map,
845 			    bool incremental)
846 {
847 	u32 n;
848 
849 	ceph_decode_32_safe(p, end, n, e_inval);
850 	while (n--) {
851 		struct ceph_pg pgid;
852 		u32 len, i;
853 		int ret;
854 
855 		ret = ceph_decode_pgid(p, end, &pgid);
856 		if (ret)
857 			return ret;
858 
859 		ceph_decode_32_safe(p, end, len, e_inval);
860 
861 		ret = __remove_pg_mapping(&map->pg_temp, pgid);
862 		BUG_ON(!incremental && ret != -ENOENT);
863 
864 		if (!incremental || len > 0) {
865 			struct ceph_pg_mapping *pg;
866 
867 			ceph_decode_need(p, end, len*sizeof(u32), e_inval);
868 
869 			if (len > (UINT_MAX - sizeof(*pg)) / sizeof(u32))
870 				return -EINVAL;
871 
872 			pg = kzalloc(sizeof(*pg) + len*sizeof(u32), GFP_NOFS);
873 			if (!pg)
874 				return -ENOMEM;
875 
876 			pg->pgid = pgid;
877 			pg->pg_temp.len = len;
878 			for (i = 0; i < len; i++)
879 				pg->pg_temp.osds[i] = ceph_decode_32(p);
880 
881 			ret = __insert_pg_mapping(pg, &map->pg_temp);
882 			if (ret) {
883 				kfree(pg);
884 				return ret;
885 			}
886 		}
887 	}
888 
889 	return 0;
890 
891 e_inval:
892 	return -EINVAL;
893 }
894 
895 static int decode_pg_temp(void **p, void *end, struct ceph_osdmap *map)
896 {
897 	return __decode_pg_temp(p, end, map, false);
898 }
899 
900 static int decode_new_pg_temp(void **p, void *end, struct ceph_osdmap *map)
901 {
902 	return __decode_pg_temp(p, end, map, true);
903 }
904 
905 static int __decode_primary_temp(void **p, void *end, struct ceph_osdmap *map,
906 				 bool incremental)
907 {
908 	u32 n;
909 
910 	ceph_decode_32_safe(p, end, n, e_inval);
911 	while (n--) {
912 		struct ceph_pg pgid;
913 		u32 osd;
914 		int ret;
915 
916 		ret = ceph_decode_pgid(p, end, &pgid);
917 		if (ret)
918 			return ret;
919 
920 		ceph_decode_32_safe(p, end, osd, e_inval);
921 
922 		ret = __remove_pg_mapping(&map->primary_temp, pgid);
923 		BUG_ON(!incremental && ret != -ENOENT);
924 
925 		if (!incremental || osd != (u32)-1) {
926 			struct ceph_pg_mapping *pg;
927 
928 			pg = kzalloc(sizeof(*pg), GFP_NOFS);
929 			if (!pg)
930 				return -ENOMEM;
931 
932 			pg->pgid = pgid;
933 			pg->primary_temp.osd = osd;
934 
935 			ret = __insert_pg_mapping(pg, &map->primary_temp);
936 			if (ret) {
937 				kfree(pg);
938 				return ret;
939 			}
940 		}
941 	}
942 
943 	return 0;
944 
945 e_inval:
946 	return -EINVAL;
947 }
948 
949 static int decode_primary_temp(void **p, void *end, struct ceph_osdmap *map)
950 {
951 	return __decode_primary_temp(p, end, map, false);
952 }
953 
954 static int decode_new_primary_temp(void **p, void *end,
955 				   struct ceph_osdmap *map)
956 {
957 	return __decode_primary_temp(p, end, map, true);
958 }
959 
960 u32 ceph_get_primary_affinity(struct ceph_osdmap *map, int osd)
961 {
962 	BUG_ON(osd >= map->max_osd);
963 
964 	if (!map->osd_primary_affinity)
965 		return CEPH_OSD_DEFAULT_PRIMARY_AFFINITY;
966 
967 	return map->osd_primary_affinity[osd];
968 }
969 
970 static int set_primary_affinity(struct ceph_osdmap *map, int osd, u32 aff)
971 {
972 	BUG_ON(osd >= map->max_osd);
973 
974 	if (!map->osd_primary_affinity) {
975 		int i;
976 
977 		map->osd_primary_affinity = kmalloc(map->max_osd*sizeof(u32),
978 						    GFP_NOFS);
979 		if (!map->osd_primary_affinity)
980 			return -ENOMEM;
981 
982 		for (i = 0; i < map->max_osd; i++)
983 			map->osd_primary_affinity[i] =
984 			    CEPH_OSD_DEFAULT_PRIMARY_AFFINITY;
985 	}
986 
987 	map->osd_primary_affinity[osd] = aff;
988 
989 	return 0;
990 }
991 
992 static int decode_primary_affinity(void **p, void *end,
993 				   struct ceph_osdmap *map)
994 {
995 	u32 len, i;
996 
997 	ceph_decode_32_safe(p, end, len, e_inval);
998 	if (len == 0) {
999 		kfree(map->osd_primary_affinity);
1000 		map->osd_primary_affinity = NULL;
1001 		return 0;
1002 	}
1003 	if (len != map->max_osd)
1004 		goto e_inval;
1005 
1006 	ceph_decode_need(p, end, map->max_osd*sizeof(u32), e_inval);
1007 
1008 	for (i = 0; i < map->max_osd; i++) {
1009 		int ret;
1010 
1011 		ret = set_primary_affinity(map, i, ceph_decode_32(p));
1012 		if (ret)
1013 			return ret;
1014 	}
1015 
1016 	return 0;
1017 
1018 e_inval:
1019 	return -EINVAL;
1020 }
1021 
1022 static int decode_new_primary_affinity(void **p, void *end,
1023 				       struct ceph_osdmap *map)
1024 {
1025 	u32 n;
1026 
1027 	ceph_decode_32_safe(p, end, n, e_inval);
1028 	while (n--) {
1029 		u32 osd, aff;
1030 		int ret;
1031 
1032 		ceph_decode_32_safe(p, end, osd, e_inval);
1033 		ceph_decode_32_safe(p, end, aff, e_inval);
1034 
1035 		ret = set_primary_affinity(map, osd, aff);
1036 		if (ret)
1037 			return ret;
1038 
1039 		pr_info("osd%d primary-affinity 0x%x\n", osd, aff);
1040 	}
1041 
1042 	return 0;
1043 
1044 e_inval:
1045 	return -EINVAL;
1046 }
1047 
1048 /*
1049  * decode a full map.
1050  */
1051 static int osdmap_decode(void **p, void *end, struct ceph_osdmap *map)
1052 {
1053 	u8 struct_v;
1054 	u32 epoch = 0;
1055 	void *start = *p;
1056 	u32 max;
1057 	u32 len, i;
1058 	int err;
1059 
1060 	dout("%s %p to %p len %d\n", __func__, *p, end, (int)(end - *p));
1061 
1062 	err = get_osdmap_client_data_v(p, end, "full", &struct_v);
1063 	if (err)
1064 		goto bad;
1065 
1066 	/* fsid, epoch, created, modified */
1067 	ceph_decode_need(p, end, sizeof(map->fsid) + sizeof(u32) +
1068 			 sizeof(map->created) + sizeof(map->modified), e_inval);
1069 	ceph_decode_copy(p, &map->fsid, sizeof(map->fsid));
1070 	epoch = map->epoch = ceph_decode_32(p);
1071 	ceph_decode_copy(p, &map->created, sizeof(map->created));
1072 	ceph_decode_copy(p, &map->modified, sizeof(map->modified));
1073 
1074 	/* pools */
1075 	err = decode_pools(p, end, map);
1076 	if (err)
1077 		goto bad;
1078 
1079 	/* pool_name */
1080 	err = decode_pool_names(p, end, map);
1081 	if (err)
1082 		goto bad;
1083 
1084 	ceph_decode_32_safe(p, end, map->pool_max, e_inval);
1085 
1086 	ceph_decode_32_safe(p, end, map->flags, e_inval);
1087 
1088 	/* max_osd */
1089 	ceph_decode_32_safe(p, end, max, e_inval);
1090 
1091 	/* (re)alloc osd arrays */
1092 	err = osdmap_set_max_osd(map, max);
1093 	if (err)
1094 		goto bad;
1095 
1096 	/* osd_state, osd_weight, osd_addrs->client_addr */
1097 	ceph_decode_need(p, end, 3*sizeof(u32) +
1098 			 map->max_osd*(1 + sizeof(*map->osd_weight) +
1099 				       sizeof(*map->osd_addr)), e_inval);
1100 
1101 	if (ceph_decode_32(p) != map->max_osd)
1102 		goto e_inval;
1103 
1104 	ceph_decode_copy(p, map->osd_state, map->max_osd);
1105 
1106 	if (ceph_decode_32(p) != map->max_osd)
1107 		goto e_inval;
1108 
1109 	for (i = 0; i < map->max_osd; i++)
1110 		map->osd_weight[i] = ceph_decode_32(p);
1111 
1112 	if (ceph_decode_32(p) != map->max_osd)
1113 		goto e_inval;
1114 
1115 	ceph_decode_copy(p, map->osd_addr, map->max_osd*sizeof(*map->osd_addr));
1116 	for (i = 0; i < map->max_osd; i++)
1117 		ceph_decode_addr(&map->osd_addr[i]);
1118 
1119 	/* pg_temp */
1120 	err = decode_pg_temp(p, end, map);
1121 	if (err)
1122 		goto bad;
1123 
1124 	/* primary_temp */
1125 	if (struct_v >= 1) {
1126 		err = decode_primary_temp(p, end, map);
1127 		if (err)
1128 			goto bad;
1129 	}
1130 
1131 	/* primary_affinity */
1132 	if (struct_v >= 2) {
1133 		err = decode_primary_affinity(p, end, map);
1134 		if (err)
1135 			goto bad;
1136 	} else {
1137 		/* XXX can this happen? */
1138 		kfree(map->osd_primary_affinity);
1139 		map->osd_primary_affinity = NULL;
1140 	}
1141 
1142 	/* crush */
1143 	ceph_decode_32_safe(p, end, len, e_inval);
1144 	map->crush = crush_decode(*p, min(*p + len, end));
1145 	if (IS_ERR(map->crush)) {
1146 		err = PTR_ERR(map->crush);
1147 		map->crush = NULL;
1148 		goto bad;
1149 	}
1150 	*p += len;
1151 
1152 	/* ignore the rest */
1153 	*p = end;
1154 
1155 	dout("full osdmap epoch %d max_osd %d\n", map->epoch, map->max_osd);
1156 	return 0;
1157 
1158 e_inval:
1159 	err = -EINVAL;
1160 bad:
1161 	pr_err("corrupt full osdmap (%d) epoch %d off %d (%p of %p-%p)\n",
1162 	       err, epoch, (int)(*p - start), *p, start, end);
1163 	print_hex_dump(KERN_DEBUG, "osdmap: ",
1164 		       DUMP_PREFIX_OFFSET, 16, 1,
1165 		       start, end - start, true);
1166 	return err;
1167 }
1168 
1169 /*
1170  * Allocate and decode a full map.
1171  */
1172 struct ceph_osdmap *ceph_osdmap_decode(void **p, void *end)
1173 {
1174 	struct ceph_osdmap *map;
1175 	int ret;
1176 
1177 	map = kzalloc(sizeof(*map), GFP_NOFS);
1178 	if (!map)
1179 		return ERR_PTR(-ENOMEM);
1180 
1181 	map->pg_temp = RB_ROOT;
1182 	map->primary_temp = RB_ROOT;
1183 	mutex_init(&map->crush_scratch_mutex);
1184 
1185 	ret = osdmap_decode(p, end, map);
1186 	if (ret) {
1187 		ceph_osdmap_destroy(map);
1188 		return ERR_PTR(ret);
1189 	}
1190 
1191 	return map;
1192 }
1193 
1194 /*
1195  * decode and apply an incremental map update.
1196  */
1197 struct ceph_osdmap *osdmap_apply_incremental(void **p, void *end,
1198 					     struct ceph_osdmap *map,
1199 					     struct ceph_messenger *msgr)
1200 {
1201 	struct crush_map *newcrush = NULL;
1202 	struct ceph_fsid fsid;
1203 	u32 epoch = 0;
1204 	struct ceph_timespec modified;
1205 	s32 len;
1206 	u64 pool;
1207 	__s64 new_pool_max;
1208 	__s32 new_flags, max;
1209 	void *start = *p;
1210 	int err;
1211 	u8 struct_v;
1212 
1213 	dout("%s %p to %p len %d\n", __func__, *p, end, (int)(end - *p));
1214 
1215 	err = get_osdmap_client_data_v(p, end, "inc", &struct_v);
1216 	if (err)
1217 		goto bad;
1218 
1219 	/* fsid, epoch, modified, new_pool_max, new_flags */
1220 	ceph_decode_need(p, end, sizeof(fsid) + sizeof(u32) + sizeof(modified) +
1221 			 sizeof(u64) + sizeof(u32), e_inval);
1222 	ceph_decode_copy(p, &fsid, sizeof(fsid));
1223 	epoch = ceph_decode_32(p);
1224 	BUG_ON(epoch != map->epoch+1);
1225 	ceph_decode_copy(p, &modified, sizeof(modified));
1226 	new_pool_max = ceph_decode_64(p);
1227 	new_flags = ceph_decode_32(p);
1228 
1229 	/* full map? */
1230 	ceph_decode_32_safe(p, end, len, e_inval);
1231 	if (len > 0) {
1232 		dout("apply_incremental full map len %d, %p to %p\n",
1233 		     len, *p, end);
1234 		return ceph_osdmap_decode(p, min(*p+len, end));
1235 	}
1236 
1237 	/* new crush? */
1238 	ceph_decode_32_safe(p, end, len, e_inval);
1239 	if (len > 0) {
1240 		newcrush = crush_decode(*p, min(*p+len, end));
1241 		if (IS_ERR(newcrush)) {
1242 			err = PTR_ERR(newcrush);
1243 			newcrush = NULL;
1244 			goto bad;
1245 		}
1246 		*p += len;
1247 	}
1248 
1249 	/* new flags? */
1250 	if (new_flags >= 0)
1251 		map->flags = new_flags;
1252 	if (new_pool_max >= 0)
1253 		map->pool_max = new_pool_max;
1254 
1255 	/* new max? */
1256 	ceph_decode_32_safe(p, end, max, e_inval);
1257 	if (max >= 0) {
1258 		err = osdmap_set_max_osd(map, max);
1259 		if (err)
1260 			goto bad;
1261 	}
1262 
1263 	map->epoch++;
1264 	map->modified = modified;
1265 	if (newcrush) {
1266 		if (map->crush)
1267 			crush_destroy(map->crush);
1268 		map->crush = newcrush;
1269 		newcrush = NULL;
1270 	}
1271 
1272 	/* new_pools */
1273 	err = decode_new_pools(p, end, map);
1274 	if (err)
1275 		goto bad;
1276 
1277 	/* new_pool_names */
1278 	err = decode_pool_names(p, end, map);
1279 	if (err)
1280 		goto bad;
1281 
1282 	/* old_pool */
1283 	ceph_decode_32_safe(p, end, len, e_inval);
1284 	while (len--) {
1285 		struct ceph_pg_pool_info *pi;
1286 
1287 		ceph_decode_64_safe(p, end, pool, e_inval);
1288 		pi = __lookup_pg_pool(&map->pg_pools, pool);
1289 		if (pi)
1290 			__remove_pg_pool(&map->pg_pools, pi);
1291 	}
1292 
1293 	/* new_up */
1294 	ceph_decode_32_safe(p, end, len, e_inval);
1295 	while (len--) {
1296 		u32 osd;
1297 		struct ceph_entity_addr addr;
1298 		ceph_decode_32_safe(p, end, osd, e_inval);
1299 		ceph_decode_copy_safe(p, end, &addr, sizeof(addr), e_inval);
1300 		ceph_decode_addr(&addr);
1301 		pr_info("osd%d up\n", osd);
1302 		BUG_ON(osd >= map->max_osd);
1303 		map->osd_state[osd] |= CEPH_OSD_UP | CEPH_OSD_EXISTS;
1304 		map->osd_addr[osd] = addr;
1305 	}
1306 
1307 	/* new_state */
1308 	ceph_decode_32_safe(p, end, len, e_inval);
1309 	while (len--) {
1310 		u32 osd;
1311 		u8 xorstate;
1312 		ceph_decode_32_safe(p, end, osd, e_inval);
1313 		xorstate = **(u8 **)p;
1314 		(*p)++;  /* clean flag */
1315 		if (xorstate == 0)
1316 			xorstate = CEPH_OSD_UP;
1317 		if (xorstate & CEPH_OSD_UP)
1318 			pr_info("osd%d down\n", osd);
1319 		if (osd < map->max_osd)
1320 			map->osd_state[osd] ^= xorstate;
1321 	}
1322 
1323 	/* new_weight */
1324 	ceph_decode_32_safe(p, end, len, e_inval);
1325 	while (len--) {
1326 		u32 osd, off;
1327 		ceph_decode_need(p, end, sizeof(u32)*2, e_inval);
1328 		osd = ceph_decode_32(p);
1329 		off = ceph_decode_32(p);
1330 		pr_info("osd%d weight 0x%x %s\n", osd, off,
1331 		     off == CEPH_OSD_IN ? "(in)" :
1332 		     (off == CEPH_OSD_OUT ? "(out)" : ""));
1333 		if (osd < map->max_osd)
1334 			map->osd_weight[osd] = off;
1335 	}
1336 
1337 	/* new_pg_temp */
1338 	err = decode_new_pg_temp(p, end, map);
1339 	if (err)
1340 		goto bad;
1341 
1342 	/* new_primary_temp */
1343 	if (struct_v >= 1) {
1344 		err = decode_new_primary_temp(p, end, map);
1345 		if (err)
1346 			goto bad;
1347 	}
1348 
1349 	/* new_primary_affinity */
1350 	if (struct_v >= 2) {
1351 		err = decode_new_primary_affinity(p, end, map);
1352 		if (err)
1353 			goto bad;
1354 	}
1355 
1356 	/* ignore the rest */
1357 	*p = end;
1358 
1359 	dout("inc osdmap epoch %d max_osd %d\n", map->epoch, map->max_osd);
1360 	return map;
1361 
1362 e_inval:
1363 	err = -EINVAL;
1364 bad:
1365 	pr_err("corrupt inc osdmap (%d) epoch %d off %d (%p of %p-%p)\n",
1366 	       err, epoch, (int)(*p - start), *p, start, end);
1367 	print_hex_dump(KERN_DEBUG, "osdmap: ",
1368 		       DUMP_PREFIX_OFFSET, 16, 1,
1369 		       start, end - start, true);
1370 	if (newcrush)
1371 		crush_destroy(newcrush);
1372 	return ERR_PTR(err);
1373 }
1374 
1375 
1376 
1377 
1378 /*
1379  * calculate file layout from given offset, length.
1380  * fill in correct oid, logical length, and object extent
1381  * offset, length.
1382  *
1383  * for now, we write only a single su, until we can
1384  * pass a stride back to the caller.
1385  */
1386 int ceph_calc_file_object_mapping(struct ceph_file_layout *layout,
1387 				   u64 off, u64 len,
1388 				   u64 *ono,
1389 				   u64 *oxoff, u64 *oxlen)
1390 {
1391 	u32 osize = le32_to_cpu(layout->fl_object_size);
1392 	u32 su = le32_to_cpu(layout->fl_stripe_unit);
1393 	u32 sc = le32_to_cpu(layout->fl_stripe_count);
1394 	u32 bl, stripeno, stripepos, objsetno;
1395 	u32 su_per_object;
1396 	u64 t, su_offset;
1397 
1398 	dout("mapping %llu~%llu  osize %u fl_su %u\n", off, len,
1399 	     osize, su);
1400 	if (su == 0 || sc == 0)
1401 		goto invalid;
1402 	su_per_object = osize / su;
1403 	if (su_per_object == 0)
1404 		goto invalid;
1405 	dout("osize %u / su %u = su_per_object %u\n", osize, su,
1406 	     su_per_object);
1407 
1408 	if ((su & ~PAGE_MASK) != 0)
1409 		goto invalid;
1410 
1411 	/* bl = *off / su; */
1412 	t = off;
1413 	do_div(t, su);
1414 	bl = t;
1415 	dout("off %llu / su %u = bl %u\n", off, su, bl);
1416 
1417 	stripeno = bl / sc;
1418 	stripepos = bl % sc;
1419 	objsetno = stripeno / su_per_object;
1420 
1421 	*ono = objsetno * sc + stripepos;
1422 	dout("objset %u * sc %u = ono %u\n", objsetno, sc, (unsigned int)*ono);
1423 
1424 	/* *oxoff = *off % layout->fl_stripe_unit;  # offset in su */
1425 	t = off;
1426 	su_offset = do_div(t, su);
1427 	*oxoff = su_offset + (stripeno % su_per_object) * su;
1428 
1429 	/*
1430 	 * Calculate the length of the extent being written to the selected
1431 	 * object. This is the minimum of the full length requested (len) or
1432 	 * the remainder of the current stripe being written to.
1433 	 */
1434 	*oxlen = min_t(u64, len, su - su_offset);
1435 
1436 	dout(" obj extent %llu~%llu\n", *oxoff, *oxlen);
1437 	return 0;
1438 
1439 invalid:
1440 	dout(" invalid layout\n");
1441 	*ono = 0;
1442 	*oxoff = 0;
1443 	*oxlen = 0;
1444 	return -EINVAL;
1445 }
1446 EXPORT_SYMBOL(ceph_calc_file_object_mapping);
1447 
1448 /*
1449  * Calculate mapping of a (oloc, oid) pair to a PG.  Should only be
1450  * called with target's (oloc, oid), since tiering isn't taken into
1451  * account.
1452  */
1453 int ceph_oloc_oid_to_pg(struct ceph_osdmap *osdmap,
1454 			struct ceph_object_locator *oloc,
1455 			struct ceph_object_id *oid,
1456 			struct ceph_pg *pg_out)
1457 {
1458 	struct ceph_pg_pool_info *pi;
1459 
1460 	pi = __lookup_pg_pool(&osdmap->pg_pools, oloc->pool);
1461 	if (!pi)
1462 		return -EIO;
1463 
1464 	pg_out->pool = oloc->pool;
1465 	pg_out->seed = ceph_str_hash(pi->object_hash, oid->name,
1466 				     oid->name_len);
1467 
1468 	dout("%s '%.*s' pgid %llu.%x\n", __func__, oid->name_len, oid->name,
1469 	     pg_out->pool, pg_out->seed);
1470 	return 0;
1471 }
1472 EXPORT_SYMBOL(ceph_oloc_oid_to_pg);
1473 
1474 static int do_crush(struct ceph_osdmap *map, int ruleno, int x,
1475 		    int *result, int result_max,
1476 		    const __u32 *weight, int weight_max)
1477 {
1478 	int r;
1479 
1480 	BUG_ON(result_max > CEPH_PG_MAX_SIZE);
1481 
1482 	mutex_lock(&map->crush_scratch_mutex);
1483 	r = crush_do_rule(map->crush, ruleno, x, result, result_max,
1484 			  weight, weight_max, map->crush_scratch_ary);
1485 	mutex_unlock(&map->crush_scratch_mutex);
1486 
1487 	return r;
1488 }
1489 
1490 /*
1491  * Calculate raw (crush) set for given pgid.
1492  *
1493  * Return raw set length, or error.
1494  */
1495 static int pg_to_raw_osds(struct ceph_osdmap *osdmap,
1496 			  struct ceph_pg_pool_info *pool,
1497 			  struct ceph_pg pgid, u32 pps, int *osds)
1498 {
1499 	int ruleno;
1500 	int len;
1501 
1502 	/* crush */
1503 	ruleno = crush_find_rule(osdmap->crush, pool->crush_ruleset,
1504 				 pool->type, pool->size);
1505 	if (ruleno < 0) {
1506 		pr_err("no crush rule: pool %lld ruleset %d type %d size %d\n",
1507 		       pgid.pool, pool->crush_ruleset, pool->type,
1508 		       pool->size);
1509 		return -ENOENT;
1510 	}
1511 
1512 	len = do_crush(osdmap, ruleno, pps, osds,
1513 		       min_t(int, pool->size, CEPH_PG_MAX_SIZE),
1514 		       osdmap->osd_weight, osdmap->max_osd);
1515 	if (len < 0) {
1516 		pr_err("error %d from crush rule %d: pool %lld ruleset %d type %d size %d\n",
1517 		       len, ruleno, pgid.pool, pool->crush_ruleset,
1518 		       pool->type, pool->size);
1519 		return len;
1520 	}
1521 
1522 	return len;
1523 }
1524 
1525 /*
1526  * Given raw set, calculate up set and up primary.
1527  *
1528  * Return up set length.  *primary is set to up primary osd id, or -1
1529  * if up set is empty.
1530  */
1531 static int raw_to_up_osds(struct ceph_osdmap *osdmap,
1532 			  struct ceph_pg_pool_info *pool,
1533 			  int *osds, int len, int *primary)
1534 {
1535 	int up_primary = -1;
1536 	int i;
1537 
1538 	if (ceph_can_shift_osds(pool)) {
1539 		int removed = 0;
1540 
1541 		for (i = 0; i < len; i++) {
1542 			if (ceph_osd_is_down(osdmap, osds[i])) {
1543 				removed++;
1544 				continue;
1545 			}
1546 			if (removed)
1547 				osds[i - removed] = osds[i];
1548 		}
1549 
1550 		len -= removed;
1551 		if (len > 0)
1552 			up_primary = osds[0];
1553 	} else {
1554 		for (i = len - 1; i >= 0; i--) {
1555 			if (ceph_osd_is_down(osdmap, osds[i]))
1556 				osds[i] = CRUSH_ITEM_NONE;
1557 			else
1558 				up_primary = osds[i];
1559 		}
1560 	}
1561 
1562 	*primary = up_primary;
1563 	return len;
1564 }
1565 
1566 static void apply_primary_affinity(struct ceph_osdmap *osdmap, u32 pps,
1567 				   struct ceph_pg_pool_info *pool,
1568 				   int *osds, int len, int *primary)
1569 {
1570 	int i;
1571 	int pos = -1;
1572 
1573 	/*
1574 	 * Do we have any non-default primary_affinity values for these
1575 	 * osds?
1576 	 */
1577 	if (!osdmap->osd_primary_affinity)
1578 		return;
1579 
1580 	for (i = 0; i < len; i++) {
1581 		int osd = osds[i];
1582 
1583 		if (osd != CRUSH_ITEM_NONE &&
1584 		    osdmap->osd_primary_affinity[osd] !=
1585 					CEPH_OSD_DEFAULT_PRIMARY_AFFINITY) {
1586 			break;
1587 		}
1588 	}
1589 	if (i == len)
1590 		return;
1591 
1592 	/*
1593 	 * Pick the primary.  Feed both the seed (for the pg) and the
1594 	 * osd into the hash/rng so that a proportional fraction of an
1595 	 * osd's pgs get rejected as primary.
1596 	 */
1597 	for (i = 0; i < len; i++) {
1598 		int osd = osds[i];
1599 		u32 aff;
1600 
1601 		if (osd == CRUSH_ITEM_NONE)
1602 			continue;
1603 
1604 		aff = osdmap->osd_primary_affinity[osd];
1605 		if (aff < CEPH_OSD_MAX_PRIMARY_AFFINITY &&
1606 		    (crush_hash32_2(CRUSH_HASH_RJENKINS1,
1607 				    pps, osd) >> 16) >= aff) {
1608 			/*
1609 			 * We chose not to use this primary.  Note it
1610 			 * anyway as a fallback in case we don't pick
1611 			 * anyone else, but keep looking.
1612 			 */
1613 			if (pos < 0)
1614 				pos = i;
1615 		} else {
1616 			pos = i;
1617 			break;
1618 		}
1619 	}
1620 	if (pos < 0)
1621 		return;
1622 
1623 	*primary = osds[pos];
1624 
1625 	if (ceph_can_shift_osds(pool) && pos > 0) {
1626 		/* move the new primary to the front */
1627 		for (i = pos; i > 0; i--)
1628 			osds[i] = osds[i - 1];
1629 		osds[0] = *primary;
1630 	}
1631 }
1632 
1633 /*
1634  * Given up set, apply pg_temp and primary_temp mappings.
1635  *
1636  * Return acting set length.  *primary is set to acting primary osd id,
1637  * or -1 if acting set is empty.
1638  */
1639 static int apply_temps(struct ceph_osdmap *osdmap,
1640 		       struct ceph_pg_pool_info *pool, struct ceph_pg pgid,
1641 		       int *osds, int len, int *primary)
1642 {
1643 	struct ceph_pg_mapping *pg;
1644 	int temp_len;
1645 	int temp_primary;
1646 	int i;
1647 
1648 	/* raw_pg -> pg */
1649 	pgid.seed = ceph_stable_mod(pgid.seed, pool->pg_num,
1650 				    pool->pg_num_mask);
1651 
1652 	/* pg_temp? */
1653 	pg = __lookup_pg_mapping(&osdmap->pg_temp, pgid);
1654 	if (pg) {
1655 		temp_len = 0;
1656 		temp_primary = -1;
1657 
1658 		for (i = 0; i < pg->pg_temp.len; i++) {
1659 			if (ceph_osd_is_down(osdmap, pg->pg_temp.osds[i])) {
1660 				if (ceph_can_shift_osds(pool))
1661 					continue;
1662 				else
1663 					osds[temp_len++] = CRUSH_ITEM_NONE;
1664 			} else {
1665 				osds[temp_len++] = pg->pg_temp.osds[i];
1666 			}
1667 		}
1668 
1669 		/* apply pg_temp's primary */
1670 		for (i = 0; i < temp_len; i++) {
1671 			if (osds[i] != CRUSH_ITEM_NONE) {
1672 				temp_primary = osds[i];
1673 				break;
1674 			}
1675 		}
1676 	} else {
1677 		temp_len = len;
1678 		temp_primary = *primary;
1679 	}
1680 
1681 	/* primary_temp? */
1682 	pg = __lookup_pg_mapping(&osdmap->primary_temp, pgid);
1683 	if (pg)
1684 		temp_primary = pg->primary_temp.osd;
1685 
1686 	*primary = temp_primary;
1687 	return temp_len;
1688 }
1689 
1690 /*
1691  * Calculate acting set for given pgid.
1692  *
1693  * Return acting set length, or error.  *primary is set to acting
1694  * primary osd id, or -1 if acting set is empty or on error.
1695  */
1696 int ceph_calc_pg_acting(struct ceph_osdmap *osdmap, struct ceph_pg pgid,
1697 			int *osds, int *primary)
1698 {
1699 	struct ceph_pg_pool_info *pool;
1700 	u32 pps;
1701 	int len;
1702 
1703 	pool = __lookup_pg_pool(&osdmap->pg_pools, pgid.pool);
1704 	if (!pool) {
1705 		*primary = -1;
1706 		return -ENOENT;
1707 	}
1708 
1709 	if (pool->flags & CEPH_POOL_FLAG_HASHPSPOOL) {
1710 		/* hash pool id and seed so that pool PGs do not overlap */
1711 		pps = crush_hash32_2(CRUSH_HASH_RJENKINS1,
1712 				     ceph_stable_mod(pgid.seed, pool->pgp_num,
1713 						     pool->pgp_num_mask),
1714 				     pgid.pool);
1715 	} else {
1716 		/*
1717 		 * legacy behavior: add ps and pool together.  this is
1718 		 * not a great approach because the PGs from each pool
1719 		 * will overlap on top of each other: 0.5 == 1.4 ==
1720 		 * 2.3 == ...
1721 		 */
1722 		pps = ceph_stable_mod(pgid.seed, pool->pgp_num,
1723 				      pool->pgp_num_mask) +
1724 			(unsigned)pgid.pool;
1725 	}
1726 
1727 	len = pg_to_raw_osds(osdmap, pool, pgid, pps, osds);
1728 	if (len < 0) {
1729 		*primary = -1;
1730 		return len;
1731 	}
1732 
1733 	len = raw_to_up_osds(osdmap, pool, osds, len, primary);
1734 
1735 	apply_primary_affinity(osdmap, pps, pool, osds, len, primary);
1736 
1737 	len = apply_temps(osdmap, pool, pgid, osds, len, primary);
1738 
1739 	return len;
1740 }
1741 
1742 /*
1743  * Return primary osd for given pgid, or -1 if none.
1744  */
1745 int ceph_calc_pg_primary(struct ceph_osdmap *osdmap, struct ceph_pg pgid)
1746 {
1747 	int osds[CEPH_PG_MAX_SIZE];
1748 	int primary;
1749 
1750 	ceph_calc_pg_acting(osdmap, pgid, osds, &primary);
1751 
1752 	return primary;
1753 }
1754 EXPORT_SYMBOL(ceph_calc_pg_primary);
1755