xref: /linux/net/ceph/osdmap.c (revision f884ab15afdc5514e88105c92a4e2e1e6539869a)
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_32_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 skip_name_map(void **p, void *end)
126 {
127         int len;
128         ceph_decode_32_safe(p, end, len ,bad);
129         while (len--) {
130                 int strlen;
131                 *p += sizeof(u32);
132                 ceph_decode_32_safe(p, end, strlen, bad);
133                 *p += strlen;
134 }
135         return 0;
136 bad:
137         return -EINVAL;
138 }
139 
140 static struct crush_map *crush_decode(void *pbyval, void *end)
141 {
142 	struct crush_map *c;
143 	int err = -EINVAL;
144 	int i, j;
145 	void **p = &pbyval;
146 	void *start = pbyval;
147 	u32 magic;
148 	u32 num_name_maps;
149 
150 	dout("crush_decode %p to %p len %d\n", *p, end, (int)(end - *p));
151 
152 	c = kzalloc(sizeof(*c), GFP_NOFS);
153 	if (c == NULL)
154 		return ERR_PTR(-ENOMEM);
155 
156         /* set tunables to default values */
157         c->choose_local_tries = 2;
158         c->choose_local_fallback_tries = 5;
159         c->choose_total_tries = 19;
160 	c->chooseleaf_descend_once = 0;
161 
162 	ceph_decode_need(p, end, 4*sizeof(u32), bad);
163 	magic = ceph_decode_32(p);
164 	if (magic != CRUSH_MAGIC) {
165 		pr_err("crush_decode magic %x != current %x\n",
166 		       (unsigned int)magic, (unsigned int)CRUSH_MAGIC);
167 		goto bad;
168 	}
169 	c->max_buckets = ceph_decode_32(p);
170 	c->max_rules = ceph_decode_32(p);
171 	c->max_devices = ceph_decode_32(p);
172 
173 	c->buckets = kcalloc(c->max_buckets, sizeof(*c->buckets), GFP_NOFS);
174 	if (c->buckets == NULL)
175 		goto badmem;
176 	c->rules = kcalloc(c->max_rules, sizeof(*c->rules), GFP_NOFS);
177 	if (c->rules == NULL)
178 		goto badmem;
179 
180 	/* buckets */
181 	for (i = 0; i < c->max_buckets; i++) {
182 		int size = 0;
183 		u32 alg;
184 		struct crush_bucket *b;
185 
186 		ceph_decode_32_safe(p, end, alg, bad);
187 		if (alg == 0) {
188 			c->buckets[i] = NULL;
189 			continue;
190 		}
191 		dout("crush_decode bucket %d off %x %p to %p\n",
192 		     i, (int)(*p-start), *p, end);
193 
194 		switch (alg) {
195 		case CRUSH_BUCKET_UNIFORM:
196 			size = sizeof(struct crush_bucket_uniform);
197 			break;
198 		case CRUSH_BUCKET_LIST:
199 			size = sizeof(struct crush_bucket_list);
200 			break;
201 		case CRUSH_BUCKET_TREE:
202 			size = sizeof(struct crush_bucket_tree);
203 			break;
204 		case CRUSH_BUCKET_STRAW:
205 			size = sizeof(struct crush_bucket_straw);
206 			break;
207 		default:
208 			err = -EINVAL;
209 			goto bad;
210 		}
211 		BUG_ON(size == 0);
212 		b = c->buckets[i] = kzalloc(size, GFP_NOFS);
213 		if (b == NULL)
214 			goto badmem;
215 
216 		ceph_decode_need(p, end, 4*sizeof(u32), bad);
217 		b->id = ceph_decode_32(p);
218 		b->type = ceph_decode_16(p);
219 		b->alg = ceph_decode_8(p);
220 		b->hash = ceph_decode_8(p);
221 		b->weight = ceph_decode_32(p);
222 		b->size = ceph_decode_32(p);
223 
224 		dout("crush_decode bucket size %d off %x %p to %p\n",
225 		     b->size, (int)(*p-start), *p, end);
226 
227 		b->items = kcalloc(b->size, sizeof(__s32), GFP_NOFS);
228 		if (b->items == NULL)
229 			goto badmem;
230 		b->perm = kcalloc(b->size, sizeof(u32), GFP_NOFS);
231 		if (b->perm == NULL)
232 			goto badmem;
233 		b->perm_n = 0;
234 
235 		ceph_decode_need(p, end, b->size*sizeof(u32), bad);
236 		for (j = 0; j < b->size; j++)
237 			b->items[j] = ceph_decode_32(p);
238 
239 		switch (b->alg) {
240 		case CRUSH_BUCKET_UNIFORM:
241 			err = crush_decode_uniform_bucket(p, end,
242 				  (struct crush_bucket_uniform *)b);
243 			if (err < 0)
244 				goto bad;
245 			break;
246 		case CRUSH_BUCKET_LIST:
247 			err = crush_decode_list_bucket(p, end,
248 			       (struct crush_bucket_list *)b);
249 			if (err < 0)
250 				goto bad;
251 			break;
252 		case CRUSH_BUCKET_TREE:
253 			err = crush_decode_tree_bucket(p, end,
254 				(struct crush_bucket_tree *)b);
255 			if (err < 0)
256 				goto bad;
257 			break;
258 		case CRUSH_BUCKET_STRAW:
259 			err = crush_decode_straw_bucket(p, end,
260 				(struct crush_bucket_straw *)b);
261 			if (err < 0)
262 				goto bad;
263 			break;
264 		}
265 	}
266 
267 	/* rules */
268 	dout("rule vec is %p\n", c->rules);
269 	for (i = 0; i < c->max_rules; i++) {
270 		u32 yes;
271 		struct crush_rule *r;
272 
273 		ceph_decode_32_safe(p, end, yes, bad);
274 		if (!yes) {
275 			dout("crush_decode NO rule %d off %x %p to %p\n",
276 			     i, (int)(*p-start), *p, end);
277 			c->rules[i] = NULL;
278 			continue;
279 		}
280 
281 		dout("crush_decode rule %d off %x %p to %p\n",
282 		     i, (int)(*p-start), *p, end);
283 
284 		/* len */
285 		ceph_decode_32_safe(p, end, yes, bad);
286 #if BITS_PER_LONG == 32
287 		err = -EINVAL;
288 		if (yes > (ULONG_MAX - sizeof(*r))
289 			  / sizeof(struct crush_rule_step))
290 			goto bad;
291 #endif
292 		r = c->rules[i] = kmalloc(sizeof(*r) +
293 					  yes*sizeof(struct crush_rule_step),
294 					  GFP_NOFS);
295 		if (r == NULL)
296 			goto badmem;
297 		dout(" rule %d is at %p\n", i, r);
298 		r->len = yes;
299 		ceph_decode_copy_safe(p, end, &r->mask, 4, bad); /* 4 u8's */
300 		ceph_decode_need(p, end, r->len*3*sizeof(u32), bad);
301 		for (j = 0; j < r->len; j++) {
302 			r->steps[j].op = ceph_decode_32(p);
303 			r->steps[j].arg1 = ceph_decode_32(p);
304 			r->steps[j].arg2 = ceph_decode_32(p);
305 		}
306 	}
307 
308 	/* ignore trailing name maps. */
309         for (num_name_maps = 0; num_name_maps < 3; num_name_maps++) {
310                 err = skip_name_map(p, end);
311                 if (err < 0)
312                         goto done;
313         }
314 
315         /* tunables */
316         ceph_decode_need(p, end, 3*sizeof(u32), done);
317         c->choose_local_tries = ceph_decode_32(p);
318         c->choose_local_fallback_tries =  ceph_decode_32(p);
319         c->choose_total_tries = ceph_decode_32(p);
320         dout("crush decode tunable choose_local_tries = %d",
321              c->choose_local_tries);
322         dout("crush decode tunable choose_local_fallback_tries = %d",
323              c->choose_local_fallback_tries);
324         dout("crush decode tunable choose_total_tries = %d",
325              c->choose_total_tries);
326 
327 	ceph_decode_need(p, end, sizeof(u32), done);
328 	c->chooseleaf_descend_once = ceph_decode_32(p);
329 	dout("crush decode tunable chooseleaf_descend_once = %d",
330 	     c->chooseleaf_descend_once);
331 
332 done:
333 	dout("crush_decode success\n");
334 	return c;
335 
336 badmem:
337 	err = -ENOMEM;
338 bad:
339 	dout("crush_decode fail %d\n", err);
340 	crush_destroy(c);
341 	return ERR_PTR(err);
342 }
343 
344 /*
345  * rbtree of pg_mapping for handling pg_temp (explicit mapping of pgid
346  * to a set of osds)
347  */
348 static int pgid_cmp(struct ceph_pg l, struct ceph_pg r)
349 {
350 	if (l.pool < r.pool)
351 		return -1;
352 	if (l.pool > r.pool)
353 		return 1;
354 	if (l.seed < r.seed)
355 		return -1;
356 	if (l.seed > r.seed)
357 		return 1;
358 	return 0;
359 }
360 
361 static int __insert_pg_mapping(struct ceph_pg_mapping *new,
362 			       struct rb_root *root)
363 {
364 	struct rb_node **p = &root->rb_node;
365 	struct rb_node *parent = NULL;
366 	struct ceph_pg_mapping *pg = NULL;
367 	int c;
368 
369 	dout("__insert_pg_mapping %llx %p\n", *(u64 *)&new->pgid, new);
370 	while (*p) {
371 		parent = *p;
372 		pg = rb_entry(parent, struct ceph_pg_mapping, node);
373 		c = pgid_cmp(new->pgid, pg->pgid);
374 		if (c < 0)
375 			p = &(*p)->rb_left;
376 		else if (c > 0)
377 			p = &(*p)->rb_right;
378 		else
379 			return -EEXIST;
380 	}
381 
382 	rb_link_node(&new->node, parent, p);
383 	rb_insert_color(&new->node, root);
384 	return 0;
385 }
386 
387 static struct ceph_pg_mapping *__lookup_pg_mapping(struct rb_root *root,
388 						   struct ceph_pg pgid)
389 {
390 	struct rb_node *n = root->rb_node;
391 	struct ceph_pg_mapping *pg;
392 	int c;
393 
394 	while (n) {
395 		pg = rb_entry(n, struct ceph_pg_mapping, node);
396 		c = pgid_cmp(pgid, pg->pgid);
397 		if (c < 0) {
398 			n = n->rb_left;
399 		} else if (c > 0) {
400 			n = n->rb_right;
401 		} else {
402 			dout("__lookup_pg_mapping %lld.%x got %p\n",
403 			     pgid.pool, pgid.seed, pg);
404 			return pg;
405 		}
406 	}
407 	return NULL;
408 }
409 
410 static int __remove_pg_mapping(struct rb_root *root, struct ceph_pg pgid)
411 {
412 	struct ceph_pg_mapping *pg = __lookup_pg_mapping(root, pgid);
413 
414 	if (pg) {
415 		dout("__remove_pg_mapping %lld.%x %p\n", pgid.pool, pgid.seed,
416 		     pg);
417 		rb_erase(&pg->node, root);
418 		kfree(pg);
419 		return 0;
420 	}
421 	dout("__remove_pg_mapping %lld.%x dne\n", pgid.pool, pgid.seed);
422 	return -ENOENT;
423 }
424 
425 /*
426  * rbtree of pg pool info
427  */
428 static int __insert_pg_pool(struct rb_root *root, struct ceph_pg_pool_info *new)
429 {
430 	struct rb_node **p = &root->rb_node;
431 	struct rb_node *parent = NULL;
432 	struct ceph_pg_pool_info *pi = NULL;
433 
434 	while (*p) {
435 		parent = *p;
436 		pi = rb_entry(parent, struct ceph_pg_pool_info, node);
437 		if (new->id < pi->id)
438 			p = &(*p)->rb_left;
439 		else if (new->id > pi->id)
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_pool_info *__lookup_pg_pool(struct rb_root *root, u64 id)
451 {
452 	struct ceph_pg_pool_info *pi;
453 	struct rb_node *n = root->rb_node;
454 
455 	while (n) {
456 		pi = rb_entry(n, struct ceph_pg_pool_info, node);
457 		if (id < pi->id)
458 			n = n->rb_left;
459 		else if (id > pi->id)
460 			n = n->rb_right;
461 		else
462 			return pi;
463 	}
464 	return NULL;
465 }
466 
467 const char *ceph_pg_pool_name_by_id(struct ceph_osdmap *map, u64 id)
468 {
469 	struct ceph_pg_pool_info *pi;
470 
471 	if (id == CEPH_NOPOOL)
472 		return NULL;
473 
474 	if (WARN_ON_ONCE(id > (u64) INT_MAX))
475 		return NULL;
476 
477 	pi = __lookup_pg_pool(&map->pg_pools, (int) id);
478 
479 	return pi ? pi->name : NULL;
480 }
481 EXPORT_SYMBOL(ceph_pg_pool_name_by_id);
482 
483 int ceph_pg_poolid_by_name(struct ceph_osdmap *map, const char *name)
484 {
485 	struct rb_node *rbp;
486 
487 	for (rbp = rb_first(&map->pg_pools); rbp; rbp = rb_next(rbp)) {
488 		struct ceph_pg_pool_info *pi =
489 			rb_entry(rbp, struct ceph_pg_pool_info, node);
490 		if (pi->name && strcmp(pi->name, name) == 0)
491 			return pi->id;
492 	}
493 	return -ENOENT;
494 }
495 EXPORT_SYMBOL(ceph_pg_poolid_by_name);
496 
497 static void __remove_pg_pool(struct rb_root *root, struct ceph_pg_pool_info *pi)
498 {
499 	rb_erase(&pi->node, root);
500 	kfree(pi->name);
501 	kfree(pi);
502 }
503 
504 static int __decode_pool(void **p, void *end, struct ceph_pg_pool_info *pi)
505 {
506 	u8 ev, cv;
507 	unsigned len, num;
508 	void *pool_end;
509 
510 	ceph_decode_need(p, end, 2 + 4, bad);
511 	ev = ceph_decode_8(p);  /* encoding version */
512 	cv = ceph_decode_8(p); /* compat version */
513 	if (ev < 5) {
514 		pr_warning("got v %d < 5 cv %d of ceph_pg_pool\n", ev, cv);
515 		return -EINVAL;
516 	}
517 	if (cv > 7) {
518 		pr_warning("got v %d cv %d > 7 of ceph_pg_pool\n", ev, cv);
519 		return -EINVAL;
520 	}
521 	len = ceph_decode_32(p);
522 	ceph_decode_need(p, end, len, bad);
523 	pool_end = *p + len;
524 
525 	pi->type = ceph_decode_8(p);
526 	pi->size = ceph_decode_8(p);
527 	pi->crush_ruleset = ceph_decode_8(p);
528 	pi->object_hash = ceph_decode_8(p);
529 
530 	pi->pg_num = ceph_decode_32(p);
531 	pi->pgp_num = ceph_decode_32(p);
532 
533 	*p += 4 + 4;  /* skip lpg* */
534 	*p += 4;      /* skip last_change */
535 	*p += 8 + 4;  /* skip snap_seq, snap_epoch */
536 
537 	/* skip snaps */
538 	num = ceph_decode_32(p);
539 	while (num--) {
540 		*p += 8;  /* snapid key */
541 		*p += 1 + 1; /* versions */
542 		len = ceph_decode_32(p);
543 		*p += len;
544 	}
545 
546 	/* skip removed snaps */
547 	num = ceph_decode_32(p);
548 	*p += num * (8 + 8);
549 
550 	*p += 8;  /* skip auid */
551 	pi->flags = ceph_decode_64(p);
552 
553 	/* ignore the rest */
554 
555 	*p = pool_end;
556 	calc_pg_masks(pi);
557 	return 0;
558 
559 bad:
560 	return -EINVAL;
561 }
562 
563 static int __decode_pool_names(void **p, void *end, struct ceph_osdmap *map)
564 {
565 	struct ceph_pg_pool_info *pi;
566 	u32 num, len;
567 	u64 pool;
568 
569 	ceph_decode_32_safe(p, end, num, bad);
570 	dout(" %d pool names\n", num);
571 	while (num--) {
572 		ceph_decode_64_safe(p, end, pool, bad);
573 		ceph_decode_32_safe(p, end, len, bad);
574 		dout("  pool %llu len %d\n", pool, len);
575 		ceph_decode_need(p, end, len, bad);
576 		pi = __lookup_pg_pool(&map->pg_pools, pool);
577 		if (pi) {
578 			char *name = kstrndup(*p, len, GFP_NOFS);
579 
580 			if (!name)
581 				return -ENOMEM;
582 			kfree(pi->name);
583 			pi->name = name;
584 			dout("  name is %s\n", pi->name);
585 		}
586 		*p += len;
587 	}
588 	return 0;
589 
590 bad:
591 	return -EINVAL;
592 }
593 
594 /*
595  * osd map
596  */
597 void ceph_osdmap_destroy(struct ceph_osdmap *map)
598 {
599 	dout("osdmap_destroy %p\n", map);
600 	if (map->crush)
601 		crush_destroy(map->crush);
602 	while (!RB_EMPTY_ROOT(&map->pg_temp)) {
603 		struct ceph_pg_mapping *pg =
604 			rb_entry(rb_first(&map->pg_temp),
605 				 struct ceph_pg_mapping, node);
606 		rb_erase(&pg->node, &map->pg_temp);
607 		kfree(pg);
608 	}
609 	while (!RB_EMPTY_ROOT(&map->pg_pools)) {
610 		struct ceph_pg_pool_info *pi =
611 			rb_entry(rb_first(&map->pg_pools),
612 				 struct ceph_pg_pool_info, node);
613 		__remove_pg_pool(&map->pg_pools, pi);
614 	}
615 	kfree(map->osd_state);
616 	kfree(map->osd_weight);
617 	kfree(map->osd_addr);
618 	kfree(map);
619 }
620 
621 /*
622  * adjust max osd value.  reallocate arrays.
623  */
624 static int osdmap_set_max_osd(struct ceph_osdmap *map, int max)
625 {
626 	u8 *state;
627 	struct ceph_entity_addr *addr;
628 	u32 *weight;
629 
630 	state = kcalloc(max, sizeof(*state), GFP_NOFS);
631 	addr = kcalloc(max, sizeof(*addr), GFP_NOFS);
632 	weight = kcalloc(max, sizeof(*weight), GFP_NOFS);
633 	if (state == NULL || addr == NULL || weight == NULL) {
634 		kfree(state);
635 		kfree(addr);
636 		kfree(weight);
637 		return -ENOMEM;
638 	}
639 
640 	/* copy old? */
641 	if (map->osd_state) {
642 		memcpy(state, map->osd_state, map->max_osd*sizeof(*state));
643 		memcpy(addr, map->osd_addr, map->max_osd*sizeof(*addr));
644 		memcpy(weight, map->osd_weight, map->max_osd*sizeof(*weight));
645 		kfree(map->osd_state);
646 		kfree(map->osd_addr);
647 		kfree(map->osd_weight);
648 	}
649 
650 	map->osd_state = state;
651 	map->osd_weight = weight;
652 	map->osd_addr = addr;
653 	map->max_osd = max;
654 	return 0;
655 }
656 
657 /*
658  * decode a full map.
659  */
660 struct ceph_osdmap *osdmap_decode(void **p, void *end)
661 {
662 	struct ceph_osdmap *map;
663 	u16 version;
664 	u32 len, max, i;
665 	int err = -EINVAL;
666 	void *start = *p;
667 	struct ceph_pg_pool_info *pi;
668 
669 	dout("osdmap_decode %p to %p len %d\n", *p, end, (int)(end - *p));
670 
671 	map = kzalloc(sizeof(*map), GFP_NOFS);
672 	if (map == NULL)
673 		return ERR_PTR(-ENOMEM);
674 	map->pg_temp = RB_ROOT;
675 
676 	ceph_decode_16_safe(p, end, version, bad);
677 	if (version > 6) {
678 		pr_warning("got unknown v %d > 6 of osdmap\n", version);
679 		goto bad;
680 	}
681 	if (version < 6) {
682 		pr_warning("got old v %d < 6 of osdmap\n", version);
683 		goto bad;
684 	}
685 
686 	ceph_decode_need(p, end, 2*sizeof(u64)+6*sizeof(u32), bad);
687 	ceph_decode_copy(p, &map->fsid, sizeof(map->fsid));
688 	map->epoch = ceph_decode_32(p);
689 	ceph_decode_copy(p, &map->created, sizeof(map->created));
690 	ceph_decode_copy(p, &map->modified, sizeof(map->modified));
691 
692 	ceph_decode_32_safe(p, end, max, bad);
693 	while (max--) {
694 		ceph_decode_need(p, end, 8 + 2, bad);
695 		err = -ENOMEM;
696 		pi = kzalloc(sizeof(*pi), GFP_NOFS);
697 		if (!pi)
698 			goto bad;
699 		pi->id = ceph_decode_64(p);
700 		err = __decode_pool(p, end, pi);
701 		if (err < 0) {
702 			kfree(pi);
703 			goto bad;
704 		}
705 		__insert_pg_pool(&map->pg_pools, pi);
706 	}
707 
708 	err = __decode_pool_names(p, end, map);
709 	if (err < 0) {
710 		dout("fail to decode pool names");
711 		goto bad;
712 	}
713 
714 	ceph_decode_32_safe(p, end, map->pool_max, bad);
715 
716 	ceph_decode_32_safe(p, end, map->flags, bad);
717 
718 	max = ceph_decode_32(p);
719 
720 	/* (re)alloc osd arrays */
721 	err = osdmap_set_max_osd(map, max);
722 	if (err < 0)
723 		goto bad;
724 	dout("osdmap_decode max_osd = %d\n", map->max_osd);
725 
726 	/* osds */
727 	err = -EINVAL;
728 	ceph_decode_need(p, end, 3*sizeof(u32) +
729 			 map->max_osd*(1 + sizeof(*map->osd_weight) +
730 				       sizeof(*map->osd_addr)), bad);
731 	*p += 4; /* skip length field (should match max) */
732 	ceph_decode_copy(p, map->osd_state, map->max_osd);
733 
734 	*p += 4; /* skip length field (should match max) */
735 	for (i = 0; i < map->max_osd; i++)
736 		map->osd_weight[i] = ceph_decode_32(p);
737 
738 	*p += 4; /* skip length field (should match max) */
739 	ceph_decode_copy(p, map->osd_addr, map->max_osd*sizeof(*map->osd_addr));
740 	for (i = 0; i < map->max_osd; i++)
741 		ceph_decode_addr(&map->osd_addr[i]);
742 
743 	/* pg_temp */
744 	ceph_decode_32_safe(p, end, len, bad);
745 	for (i = 0; i < len; i++) {
746 		int n, j;
747 		struct ceph_pg pgid;
748 		struct ceph_pg_mapping *pg;
749 
750 		err = ceph_decode_pgid(p, end, &pgid);
751 		if (err)
752 			goto bad;
753 		ceph_decode_need(p, end, sizeof(u32), bad);
754 		n = ceph_decode_32(p);
755 		err = -EINVAL;
756 		if (n > (UINT_MAX - sizeof(*pg)) / sizeof(u32))
757 			goto bad;
758 		ceph_decode_need(p, end, n * sizeof(u32), bad);
759 		err = -ENOMEM;
760 		pg = kmalloc(sizeof(*pg) + n*sizeof(u32), GFP_NOFS);
761 		if (!pg)
762 			goto bad;
763 		pg->pgid = pgid;
764 		pg->len = n;
765 		for (j = 0; j < n; j++)
766 			pg->osds[j] = ceph_decode_32(p);
767 
768 		err = __insert_pg_mapping(pg, &map->pg_temp);
769 		if (err)
770 			goto bad;
771 		dout(" added pg_temp %lld.%x len %d\n", pgid.pool, pgid.seed,
772 		     len);
773 	}
774 
775 	/* crush */
776 	ceph_decode_32_safe(p, end, len, bad);
777 	dout("osdmap_decode crush len %d from off 0x%x\n", len,
778 	     (int)(*p - start));
779 	ceph_decode_need(p, end, len, bad);
780 	map->crush = crush_decode(*p, end);
781 	*p += len;
782 	if (IS_ERR(map->crush)) {
783 		err = PTR_ERR(map->crush);
784 		map->crush = NULL;
785 		goto bad;
786 	}
787 
788 	/* ignore the rest of the map */
789 	*p = end;
790 
791 	dout("osdmap_decode done %p %p\n", *p, end);
792 	return map;
793 
794 bad:
795 	dout("osdmap_decode fail err %d\n", err);
796 	ceph_osdmap_destroy(map);
797 	return ERR_PTR(err);
798 }
799 
800 /*
801  * decode and apply an incremental map update.
802  */
803 struct ceph_osdmap *osdmap_apply_incremental(void **p, void *end,
804 					     struct ceph_osdmap *map,
805 					     struct ceph_messenger *msgr)
806 {
807 	struct crush_map *newcrush = NULL;
808 	struct ceph_fsid fsid;
809 	u32 epoch = 0;
810 	struct ceph_timespec modified;
811 	s32 len;
812 	u64 pool;
813 	__s64 new_pool_max;
814 	__s32 new_flags, max;
815 	void *start = *p;
816 	int err = -EINVAL;
817 	u16 version;
818 
819 	ceph_decode_16_safe(p, end, version, bad);
820 	if (version != 6) {
821 		pr_warning("got unknown v %d != 6 of inc osdmap\n", version);
822 		goto bad;
823 	}
824 
825 	ceph_decode_need(p, end, sizeof(fsid)+sizeof(modified)+2*sizeof(u32),
826 			 bad);
827 	ceph_decode_copy(p, &fsid, sizeof(fsid));
828 	epoch = ceph_decode_32(p);
829 	BUG_ON(epoch != map->epoch+1);
830 	ceph_decode_copy(p, &modified, sizeof(modified));
831 	new_pool_max = ceph_decode_64(p);
832 	new_flags = ceph_decode_32(p);
833 
834 	/* full map? */
835 	ceph_decode_32_safe(p, end, len, bad);
836 	if (len > 0) {
837 		dout("apply_incremental full map len %d, %p to %p\n",
838 		     len, *p, end);
839 		return osdmap_decode(p, min(*p+len, end));
840 	}
841 
842 	/* new crush? */
843 	ceph_decode_32_safe(p, end, len, bad);
844 	if (len > 0) {
845 		dout("apply_incremental new crush map len %d, %p to %p\n",
846 		     len, *p, end);
847 		newcrush = crush_decode(*p, min(*p+len, end));
848 		if (IS_ERR(newcrush))
849 			return ERR_CAST(newcrush);
850 		*p += len;
851 	}
852 
853 	/* new flags? */
854 	if (new_flags >= 0)
855 		map->flags = new_flags;
856 	if (new_pool_max >= 0)
857 		map->pool_max = new_pool_max;
858 
859 	ceph_decode_need(p, end, 5*sizeof(u32), bad);
860 
861 	/* new max? */
862 	max = ceph_decode_32(p);
863 	if (max >= 0) {
864 		err = osdmap_set_max_osd(map, max);
865 		if (err < 0)
866 			goto bad;
867 	}
868 
869 	map->epoch++;
870 	map->modified = modified;
871 	if (newcrush) {
872 		if (map->crush)
873 			crush_destroy(map->crush);
874 		map->crush = newcrush;
875 		newcrush = NULL;
876 	}
877 
878 	/* new_pool */
879 	ceph_decode_32_safe(p, end, len, bad);
880 	while (len--) {
881 		struct ceph_pg_pool_info *pi;
882 
883 		ceph_decode_64_safe(p, end, pool, bad);
884 		pi = __lookup_pg_pool(&map->pg_pools, pool);
885 		if (!pi) {
886 			pi = kzalloc(sizeof(*pi), GFP_NOFS);
887 			if (!pi) {
888 				err = -ENOMEM;
889 				goto bad;
890 			}
891 			pi->id = pool;
892 			__insert_pg_pool(&map->pg_pools, pi);
893 		}
894 		err = __decode_pool(p, end, pi);
895 		if (err < 0)
896 			goto bad;
897 	}
898 	if (version >= 5) {
899 		err = __decode_pool_names(p, end, map);
900 		if (err < 0)
901 			goto bad;
902 	}
903 
904 	/* old_pool */
905 	ceph_decode_32_safe(p, end, len, bad);
906 	while (len--) {
907 		struct ceph_pg_pool_info *pi;
908 
909 		ceph_decode_64_safe(p, end, pool, bad);
910 		pi = __lookup_pg_pool(&map->pg_pools, pool);
911 		if (pi)
912 			__remove_pg_pool(&map->pg_pools, pi);
913 	}
914 
915 	/* new_up */
916 	err = -EINVAL;
917 	ceph_decode_32_safe(p, end, len, bad);
918 	while (len--) {
919 		u32 osd;
920 		struct ceph_entity_addr addr;
921 		ceph_decode_32_safe(p, end, osd, bad);
922 		ceph_decode_copy_safe(p, end, &addr, sizeof(addr), bad);
923 		ceph_decode_addr(&addr);
924 		pr_info("osd%d up\n", osd);
925 		BUG_ON(osd >= map->max_osd);
926 		map->osd_state[osd] |= CEPH_OSD_UP;
927 		map->osd_addr[osd] = addr;
928 	}
929 
930 	/* new_state */
931 	ceph_decode_32_safe(p, end, len, bad);
932 	while (len--) {
933 		u32 osd;
934 		u8 xorstate;
935 		ceph_decode_32_safe(p, end, osd, bad);
936 		xorstate = **(u8 **)p;
937 		(*p)++;  /* clean flag */
938 		if (xorstate == 0)
939 			xorstate = CEPH_OSD_UP;
940 		if (xorstate & CEPH_OSD_UP)
941 			pr_info("osd%d down\n", osd);
942 		if (osd < map->max_osd)
943 			map->osd_state[osd] ^= xorstate;
944 	}
945 
946 	/* new_weight */
947 	ceph_decode_32_safe(p, end, len, bad);
948 	while (len--) {
949 		u32 osd, off;
950 		ceph_decode_need(p, end, sizeof(u32)*2, bad);
951 		osd = ceph_decode_32(p);
952 		off = ceph_decode_32(p);
953 		pr_info("osd%d weight 0x%x %s\n", osd, off,
954 		     off == CEPH_OSD_IN ? "(in)" :
955 		     (off == CEPH_OSD_OUT ? "(out)" : ""));
956 		if (osd < map->max_osd)
957 			map->osd_weight[osd] = off;
958 	}
959 
960 	/* new_pg_temp */
961 	ceph_decode_32_safe(p, end, len, bad);
962 	while (len--) {
963 		struct ceph_pg_mapping *pg;
964 		int j;
965 		struct ceph_pg pgid;
966 		u32 pglen;
967 
968 		err = ceph_decode_pgid(p, end, &pgid);
969 		if (err)
970 			goto bad;
971 		ceph_decode_need(p, end, sizeof(u32), bad);
972 		pglen = ceph_decode_32(p);
973 		if (pglen) {
974 			ceph_decode_need(p, end, pglen*sizeof(u32), bad);
975 
976 			/* removing existing (if any) */
977 			(void) __remove_pg_mapping(&map->pg_temp, pgid);
978 
979 			/* insert */
980 			err = -EINVAL;
981 			if (pglen > (UINT_MAX - sizeof(*pg)) / sizeof(u32))
982 				goto bad;
983 			err = -ENOMEM;
984 			pg = kmalloc(sizeof(*pg) + sizeof(u32)*pglen, GFP_NOFS);
985 			if (!pg)
986 				goto bad;
987 			pg->pgid = pgid;
988 			pg->len = pglen;
989 			for (j = 0; j < pglen; j++)
990 				pg->osds[j] = ceph_decode_32(p);
991 			err = __insert_pg_mapping(pg, &map->pg_temp);
992 			if (err) {
993 				kfree(pg);
994 				goto bad;
995 			}
996 			dout(" added pg_temp %lld.%x len %d\n", pgid.pool,
997 			     pgid.seed, pglen);
998 		} else {
999 			/* remove */
1000 			__remove_pg_mapping(&map->pg_temp, pgid);
1001 		}
1002 	}
1003 
1004 	/* ignore the rest */
1005 	*p = end;
1006 	return map;
1007 
1008 bad:
1009 	pr_err("corrupt inc osdmap epoch %d off %d (%p of %p-%p)\n",
1010 	       epoch, (int)(*p - start), *p, start, end);
1011 	print_hex_dump(KERN_DEBUG, "osdmap: ",
1012 		       DUMP_PREFIX_OFFSET, 16, 1,
1013 		       start, end - start, true);
1014 	if (newcrush)
1015 		crush_destroy(newcrush);
1016 	return ERR_PTR(err);
1017 }
1018 
1019 
1020 
1021 
1022 /*
1023  * calculate file layout from given offset, length.
1024  * fill in correct oid, logical length, and object extent
1025  * offset, length.
1026  *
1027  * for now, we write only a single su, until we can
1028  * pass a stride back to the caller.
1029  */
1030 int ceph_calc_file_object_mapping(struct ceph_file_layout *layout,
1031 				   u64 off, u64 len,
1032 				   u64 *ono,
1033 				   u64 *oxoff, u64 *oxlen)
1034 {
1035 	u32 osize = le32_to_cpu(layout->fl_object_size);
1036 	u32 su = le32_to_cpu(layout->fl_stripe_unit);
1037 	u32 sc = le32_to_cpu(layout->fl_stripe_count);
1038 	u32 bl, stripeno, stripepos, objsetno;
1039 	u32 su_per_object;
1040 	u64 t, su_offset;
1041 
1042 	dout("mapping %llu~%llu  osize %u fl_su %u\n", off, len,
1043 	     osize, su);
1044 	if (su == 0 || sc == 0)
1045 		goto invalid;
1046 	su_per_object = osize / su;
1047 	if (su_per_object == 0)
1048 		goto invalid;
1049 	dout("osize %u / su %u = su_per_object %u\n", osize, su,
1050 	     su_per_object);
1051 
1052 	if ((su & ~PAGE_MASK) != 0)
1053 		goto invalid;
1054 
1055 	/* bl = *off / su; */
1056 	t = off;
1057 	do_div(t, su);
1058 	bl = t;
1059 	dout("off %llu / su %u = bl %u\n", off, su, bl);
1060 
1061 	stripeno = bl / sc;
1062 	stripepos = bl % sc;
1063 	objsetno = stripeno / su_per_object;
1064 
1065 	*ono = objsetno * sc + stripepos;
1066 	dout("objset %u * sc %u = ono %u\n", objsetno, sc, (unsigned int)*ono);
1067 
1068 	/* *oxoff = *off % layout->fl_stripe_unit;  # offset in su */
1069 	t = off;
1070 	su_offset = do_div(t, su);
1071 	*oxoff = su_offset + (stripeno % su_per_object) * su;
1072 
1073 	/*
1074 	 * Calculate the length of the extent being written to the selected
1075 	 * object. This is the minimum of the full length requested (len) or
1076 	 * the remainder of the current stripe being written to.
1077 	 */
1078 	*oxlen = min_t(u64, len, su - su_offset);
1079 
1080 	dout(" obj extent %llu~%llu\n", *oxoff, *oxlen);
1081 	return 0;
1082 
1083 invalid:
1084 	dout(" invalid layout\n");
1085 	*ono = 0;
1086 	*oxoff = 0;
1087 	*oxlen = 0;
1088 	return -EINVAL;
1089 }
1090 EXPORT_SYMBOL(ceph_calc_file_object_mapping);
1091 
1092 /*
1093  * calculate an object layout (i.e. pgid) from an oid,
1094  * file_layout, and osdmap
1095  */
1096 int ceph_calc_ceph_pg(struct ceph_pg *pg, const char *oid,
1097 			struct ceph_osdmap *osdmap, uint64_t pool)
1098 {
1099 	struct ceph_pg_pool_info *pool_info;
1100 
1101 	BUG_ON(!osdmap);
1102 	pool_info = __lookup_pg_pool(&osdmap->pg_pools, pool);
1103 	if (!pool_info)
1104 		return -EIO;
1105 	pg->pool = pool;
1106 	pg->seed = ceph_str_hash(pool_info->object_hash, oid, strlen(oid));
1107 
1108 	dout("%s '%s' pgid %lld.%x\n", __func__, oid, pg->pool, pg->seed);
1109 	return 0;
1110 }
1111 EXPORT_SYMBOL(ceph_calc_ceph_pg);
1112 
1113 /*
1114  * Calculate raw osd vector for the given pgid.  Return pointer to osd
1115  * array, or NULL on failure.
1116  */
1117 static int *calc_pg_raw(struct ceph_osdmap *osdmap, struct ceph_pg pgid,
1118 			int *osds, int *num)
1119 {
1120 	struct ceph_pg_mapping *pg;
1121 	struct ceph_pg_pool_info *pool;
1122 	int ruleno;
1123 	int r;
1124 	u32 pps;
1125 
1126 	pool = __lookup_pg_pool(&osdmap->pg_pools, pgid.pool);
1127 	if (!pool)
1128 		return NULL;
1129 
1130 	/* pg_temp? */
1131 	pgid.seed = ceph_stable_mod(pgid.seed, pool->pg_num,
1132 				    pool->pgp_num_mask);
1133 	pg = __lookup_pg_mapping(&osdmap->pg_temp, pgid);
1134 	if (pg) {
1135 		*num = pg->len;
1136 		return pg->osds;
1137 	}
1138 
1139 	/* crush */
1140 	ruleno = crush_find_rule(osdmap->crush, pool->crush_ruleset,
1141 				 pool->type, pool->size);
1142 	if (ruleno < 0) {
1143 		pr_err("no crush rule pool %lld ruleset %d type %d size %d\n",
1144 		       pgid.pool, pool->crush_ruleset, pool->type,
1145 		       pool->size);
1146 		return NULL;
1147 	}
1148 
1149 	if (pool->flags & CEPH_POOL_FLAG_HASHPSPOOL) {
1150 		/* hash pool id and seed sothat pool PGs do not overlap */
1151 		pps = crush_hash32_2(CRUSH_HASH_RJENKINS1,
1152 				     ceph_stable_mod(pgid.seed, pool->pgp_num,
1153 						     pool->pgp_num_mask),
1154 				     pgid.pool);
1155 	} else {
1156 		/*
1157 		 * legacy ehavior: add ps and pool together.  this is
1158 		 * not a great approach because the PGs from each pool
1159 		 * will overlap on top of each other: 0.5 == 1.4 ==
1160 		 * 2.3 == ...
1161 		 */
1162 		pps = ceph_stable_mod(pgid.seed, pool->pgp_num,
1163 				      pool->pgp_num_mask) +
1164 			(unsigned)pgid.pool;
1165 	}
1166 	r = crush_do_rule(osdmap->crush, ruleno, pps, osds,
1167 			  min_t(int, pool->size, *num),
1168 			  osdmap->osd_weight);
1169 	if (r < 0) {
1170 		pr_err("error %d from crush rule: pool %lld ruleset %d type %d"
1171 		       " size %d\n", r, pgid.pool, pool->crush_ruleset,
1172 		       pool->type, pool->size);
1173 		return NULL;
1174 	}
1175 	*num = r;
1176 	return osds;
1177 }
1178 
1179 /*
1180  * Return acting set for given pgid.
1181  */
1182 int ceph_calc_pg_acting(struct ceph_osdmap *osdmap, struct ceph_pg pgid,
1183 			int *acting)
1184 {
1185 	int rawosds[CEPH_PG_MAX_SIZE], *osds;
1186 	int i, o, num = CEPH_PG_MAX_SIZE;
1187 
1188 	osds = calc_pg_raw(osdmap, pgid, rawosds, &num);
1189 	if (!osds)
1190 		return -1;
1191 
1192 	/* primary is first up osd */
1193 	o = 0;
1194 	for (i = 0; i < num; i++)
1195 		if (ceph_osd_is_up(osdmap, osds[i]))
1196 			acting[o++] = osds[i];
1197 	return o;
1198 }
1199 
1200 /*
1201  * Return primary osd for given pgid, or -1 if none.
1202  */
1203 int ceph_calc_pg_primary(struct ceph_osdmap *osdmap, struct ceph_pg pgid)
1204 {
1205 	int rawosds[CEPH_PG_MAX_SIZE], *osds;
1206 	int i, num = CEPH_PG_MAX_SIZE;
1207 
1208 	osds = calc_pg_raw(osdmap, pgid, rawosds, &num);
1209 	if (!osds)
1210 		return -1;
1211 
1212 	/* primary is first up osd */
1213 	for (i = 0; i < num; i++)
1214 		if (ceph_osd_is_up(osdmap, osds[i]))
1215 			return osds[i];
1216 	return -1;
1217 }
1218 EXPORT_SYMBOL(ceph_calc_pg_primary);
1219