xref: /linux/net/ceph/osdmap.c (revision 861e10be08c69808065d755d3e3cab5d520a2d32)
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 	int flag = 0;
17 
18 	if (!len)
19 		goto done;
20 
21 	*str = '\0';
22 	if (state) {
23 		if (state & CEPH_OSD_EXISTS) {
24 			snprintf(str, len, "exists");
25 			flag = 1;
26 		}
27 		if (state & CEPH_OSD_UP) {
28 			snprintf(str, len, "%s%s%s", str, (flag ? ", " : ""),
29 				 "up");
30 			flag = 1;
31 		}
32 	} else {
33 		snprintf(str, len, "doesn't exist");
34 	}
35 done:
36 	return str;
37 }
38 
39 /* maps */
40 
41 static int calc_bits_of(unsigned int t)
42 {
43 	int b = 0;
44 	while (t) {
45 		t = t >> 1;
46 		b++;
47 	}
48 	return b;
49 }
50 
51 /*
52  * the foo_mask is the smallest value 2^n-1 that is >= foo.
53  */
54 static void calc_pg_masks(struct ceph_pg_pool_info *pi)
55 {
56 	pi->pg_num_mask = (1 << calc_bits_of(le32_to_cpu(pi->v.pg_num)-1)) - 1;
57 	pi->pgp_num_mask =
58 		(1 << calc_bits_of(le32_to_cpu(pi->v.pgp_num)-1)) - 1;
59 	pi->lpg_num_mask =
60 		(1 << calc_bits_of(le32_to_cpu(pi->v.lpg_num)-1)) - 1;
61 	pi->lpgp_num_mask =
62 		(1 << calc_bits_of(le32_to_cpu(pi->v.lpgp_num)-1)) - 1;
63 }
64 
65 /*
66  * decode crush map
67  */
68 static int crush_decode_uniform_bucket(void **p, void *end,
69 				       struct crush_bucket_uniform *b)
70 {
71 	dout("crush_decode_uniform_bucket %p to %p\n", *p, end);
72 	ceph_decode_need(p, end, (1+b->h.size) * sizeof(u32), bad);
73 	b->item_weight = ceph_decode_32(p);
74 	return 0;
75 bad:
76 	return -EINVAL;
77 }
78 
79 static int crush_decode_list_bucket(void **p, void *end,
80 				    struct crush_bucket_list *b)
81 {
82 	int j;
83 	dout("crush_decode_list_bucket %p to %p\n", *p, end);
84 	b->item_weights = kcalloc(b->h.size, sizeof(u32), GFP_NOFS);
85 	if (b->item_weights == NULL)
86 		return -ENOMEM;
87 	b->sum_weights = kcalloc(b->h.size, sizeof(u32), GFP_NOFS);
88 	if (b->sum_weights == NULL)
89 		return -ENOMEM;
90 	ceph_decode_need(p, end, 2 * b->h.size * sizeof(u32), bad);
91 	for (j = 0; j < b->h.size; j++) {
92 		b->item_weights[j] = ceph_decode_32(p);
93 		b->sum_weights[j] = ceph_decode_32(p);
94 	}
95 	return 0;
96 bad:
97 	return -EINVAL;
98 }
99 
100 static int crush_decode_tree_bucket(void **p, void *end,
101 				    struct crush_bucket_tree *b)
102 {
103 	int j;
104 	dout("crush_decode_tree_bucket %p to %p\n", *p, end);
105 	ceph_decode_32_safe(p, end, b->num_nodes, bad);
106 	b->node_weights = kcalloc(b->num_nodes, sizeof(u32), GFP_NOFS);
107 	if (b->node_weights == NULL)
108 		return -ENOMEM;
109 	ceph_decode_need(p, end, b->num_nodes * sizeof(u32), bad);
110 	for (j = 0; j < b->num_nodes; j++)
111 		b->node_weights[j] = ceph_decode_32(p);
112 	return 0;
113 bad:
114 	return -EINVAL;
115 }
116 
117 static int crush_decode_straw_bucket(void **p, void *end,
118 				     struct crush_bucket_straw *b)
119 {
120 	int j;
121 	dout("crush_decode_straw_bucket %p to %p\n", *p, end);
122 	b->item_weights = kcalloc(b->h.size, sizeof(u32), GFP_NOFS);
123 	if (b->item_weights == NULL)
124 		return -ENOMEM;
125 	b->straws = kcalloc(b->h.size, sizeof(u32), GFP_NOFS);
126 	if (b->straws == NULL)
127 		return -ENOMEM;
128 	ceph_decode_need(p, end, 2 * b->h.size * sizeof(u32), bad);
129 	for (j = 0; j < b->h.size; j++) {
130 		b->item_weights[j] = ceph_decode_32(p);
131 		b->straws[j] = ceph_decode_32(p);
132 	}
133 	return 0;
134 bad:
135 	return -EINVAL;
136 }
137 
138 static int skip_name_map(void **p, void *end)
139 {
140         int len;
141         ceph_decode_32_safe(p, end, len ,bad);
142         while (len--) {
143                 int strlen;
144                 *p += sizeof(u32);
145                 ceph_decode_32_safe(p, end, strlen, bad);
146                 *p += strlen;
147 }
148         return 0;
149 bad:
150         return -EINVAL;
151 }
152 
153 static struct crush_map *crush_decode(void *pbyval, void *end)
154 {
155 	struct crush_map *c;
156 	int err = -EINVAL;
157 	int i, j;
158 	void **p = &pbyval;
159 	void *start = pbyval;
160 	u32 magic;
161 	u32 num_name_maps;
162 
163 	dout("crush_decode %p to %p len %d\n", *p, end, (int)(end - *p));
164 
165 	c = kzalloc(sizeof(*c), GFP_NOFS);
166 	if (c == NULL)
167 		return ERR_PTR(-ENOMEM);
168 
169         /* set tunables to default values */
170         c->choose_local_tries = 2;
171         c->choose_local_fallback_tries = 5;
172         c->choose_total_tries = 19;
173 
174 	ceph_decode_need(p, end, 4*sizeof(u32), bad);
175 	magic = ceph_decode_32(p);
176 	if (magic != CRUSH_MAGIC) {
177 		pr_err("crush_decode magic %x != current %x\n",
178 		       (unsigned int)magic, (unsigned int)CRUSH_MAGIC);
179 		goto bad;
180 	}
181 	c->max_buckets = ceph_decode_32(p);
182 	c->max_rules = ceph_decode_32(p);
183 	c->max_devices = ceph_decode_32(p);
184 
185 	c->buckets = kcalloc(c->max_buckets, sizeof(*c->buckets), GFP_NOFS);
186 	if (c->buckets == NULL)
187 		goto badmem;
188 	c->rules = kcalloc(c->max_rules, sizeof(*c->rules), GFP_NOFS);
189 	if (c->rules == NULL)
190 		goto badmem;
191 
192 	/* buckets */
193 	for (i = 0; i < c->max_buckets; i++) {
194 		int size = 0;
195 		u32 alg;
196 		struct crush_bucket *b;
197 
198 		ceph_decode_32_safe(p, end, alg, bad);
199 		if (alg == 0) {
200 			c->buckets[i] = NULL;
201 			continue;
202 		}
203 		dout("crush_decode bucket %d off %x %p to %p\n",
204 		     i, (int)(*p-start), *p, end);
205 
206 		switch (alg) {
207 		case CRUSH_BUCKET_UNIFORM:
208 			size = sizeof(struct crush_bucket_uniform);
209 			break;
210 		case CRUSH_BUCKET_LIST:
211 			size = sizeof(struct crush_bucket_list);
212 			break;
213 		case CRUSH_BUCKET_TREE:
214 			size = sizeof(struct crush_bucket_tree);
215 			break;
216 		case CRUSH_BUCKET_STRAW:
217 			size = sizeof(struct crush_bucket_straw);
218 			break;
219 		default:
220 			err = -EINVAL;
221 			goto bad;
222 		}
223 		BUG_ON(size == 0);
224 		b = c->buckets[i] = kzalloc(size, GFP_NOFS);
225 		if (b == NULL)
226 			goto badmem;
227 
228 		ceph_decode_need(p, end, 4*sizeof(u32), bad);
229 		b->id = ceph_decode_32(p);
230 		b->type = ceph_decode_16(p);
231 		b->alg = ceph_decode_8(p);
232 		b->hash = ceph_decode_8(p);
233 		b->weight = ceph_decode_32(p);
234 		b->size = ceph_decode_32(p);
235 
236 		dout("crush_decode bucket size %d off %x %p to %p\n",
237 		     b->size, (int)(*p-start), *p, end);
238 
239 		b->items = kcalloc(b->size, sizeof(__s32), GFP_NOFS);
240 		if (b->items == NULL)
241 			goto badmem;
242 		b->perm = kcalloc(b->size, sizeof(u32), GFP_NOFS);
243 		if (b->perm == NULL)
244 			goto badmem;
245 		b->perm_n = 0;
246 
247 		ceph_decode_need(p, end, b->size*sizeof(u32), bad);
248 		for (j = 0; j < b->size; j++)
249 			b->items[j] = ceph_decode_32(p);
250 
251 		switch (b->alg) {
252 		case CRUSH_BUCKET_UNIFORM:
253 			err = crush_decode_uniform_bucket(p, end,
254 				  (struct crush_bucket_uniform *)b);
255 			if (err < 0)
256 				goto bad;
257 			break;
258 		case CRUSH_BUCKET_LIST:
259 			err = crush_decode_list_bucket(p, end,
260 			       (struct crush_bucket_list *)b);
261 			if (err < 0)
262 				goto bad;
263 			break;
264 		case CRUSH_BUCKET_TREE:
265 			err = crush_decode_tree_bucket(p, end,
266 				(struct crush_bucket_tree *)b);
267 			if (err < 0)
268 				goto bad;
269 			break;
270 		case CRUSH_BUCKET_STRAW:
271 			err = crush_decode_straw_bucket(p, end,
272 				(struct crush_bucket_straw *)b);
273 			if (err < 0)
274 				goto bad;
275 			break;
276 		}
277 	}
278 
279 	/* rules */
280 	dout("rule vec is %p\n", c->rules);
281 	for (i = 0; i < c->max_rules; i++) {
282 		u32 yes;
283 		struct crush_rule *r;
284 
285 		ceph_decode_32_safe(p, end, yes, bad);
286 		if (!yes) {
287 			dout("crush_decode NO rule %d off %x %p to %p\n",
288 			     i, (int)(*p-start), *p, end);
289 			c->rules[i] = NULL;
290 			continue;
291 		}
292 
293 		dout("crush_decode rule %d off %x %p to %p\n",
294 		     i, (int)(*p-start), *p, end);
295 
296 		/* len */
297 		ceph_decode_32_safe(p, end, yes, bad);
298 #if BITS_PER_LONG == 32
299 		err = -EINVAL;
300 		if (yes > (ULONG_MAX - sizeof(*r))
301 			  / sizeof(struct crush_rule_step))
302 			goto bad;
303 #endif
304 		r = c->rules[i] = kmalloc(sizeof(*r) +
305 					  yes*sizeof(struct crush_rule_step),
306 					  GFP_NOFS);
307 		if (r == NULL)
308 			goto badmem;
309 		dout(" rule %d is at %p\n", i, r);
310 		r->len = yes;
311 		ceph_decode_copy_safe(p, end, &r->mask, 4, bad); /* 4 u8's */
312 		ceph_decode_need(p, end, r->len*3*sizeof(u32), bad);
313 		for (j = 0; j < r->len; j++) {
314 			r->steps[j].op = ceph_decode_32(p);
315 			r->steps[j].arg1 = ceph_decode_32(p);
316 			r->steps[j].arg2 = ceph_decode_32(p);
317 		}
318 	}
319 
320 	/* ignore trailing name maps. */
321         for (num_name_maps = 0; num_name_maps < 3; num_name_maps++) {
322                 err = skip_name_map(p, end);
323                 if (err < 0)
324                         goto done;
325         }
326 
327         /* tunables */
328         ceph_decode_need(p, end, 3*sizeof(u32), done);
329         c->choose_local_tries = ceph_decode_32(p);
330         c->choose_local_fallback_tries =  ceph_decode_32(p);
331         c->choose_total_tries = ceph_decode_32(p);
332         dout("crush decode tunable choose_local_tries = %d",
333              c->choose_local_tries);
334         dout("crush decode tunable choose_local_fallback_tries = %d",
335              c->choose_local_fallback_tries);
336         dout("crush decode tunable choose_total_tries = %d",
337              c->choose_total_tries);
338 
339 done:
340 	dout("crush_decode success\n");
341 	return c;
342 
343 badmem:
344 	err = -ENOMEM;
345 bad:
346 	dout("crush_decode fail %d\n", err);
347 	crush_destroy(c);
348 	return ERR_PTR(err);
349 }
350 
351 /*
352  * rbtree of pg_mapping for handling pg_temp (explicit mapping of pgid
353  * to a set of osds)
354  */
355 static int pgid_cmp(struct ceph_pg l, struct ceph_pg r)
356 {
357 	u64 a = *(u64 *)&l;
358 	u64 b = *(u64 *)&r;
359 
360 	if (a < b)
361 		return -1;
362 	if (a > b)
363 		return 1;
364 	return 0;
365 }
366 
367 static int __insert_pg_mapping(struct ceph_pg_mapping *new,
368 			       struct rb_root *root)
369 {
370 	struct rb_node **p = &root->rb_node;
371 	struct rb_node *parent = NULL;
372 	struct ceph_pg_mapping *pg = NULL;
373 	int c;
374 
375 	dout("__insert_pg_mapping %llx %p\n", *(u64 *)&new->pgid, new);
376 	while (*p) {
377 		parent = *p;
378 		pg = rb_entry(parent, struct ceph_pg_mapping, node);
379 		c = pgid_cmp(new->pgid, pg->pgid);
380 		if (c < 0)
381 			p = &(*p)->rb_left;
382 		else if (c > 0)
383 			p = &(*p)->rb_right;
384 		else
385 			return -EEXIST;
386 	}
387 
388 	rb_link_node(&new->node, parent, p);
389 	rb_insert_color(&new->node, root);
390 	return 0;
391 }
392 
393 static struct ceph_pg_mapping *__lookup_pg_mapping(struct rb_root *root,
394 						   struct ceph_pg pgid)
395 {
396 	struct rb_node *n = root->rb_node;
397 	struct ceph_pg_mapping *pg;
398 	int c;
399 
400 	while (n) {
401 		pg = rb_entry(n, struct ceph_pg_mapping, node);
402 		c = pgid_cmp(pgid, pg->pgid);
403 		if (c < 0) {
404 			n = n->rb_left;
405 		} else if (c > 0) {
406 			n = n->rb_right;
407 		} else {
408 			dout("__lookup_pg_mapping %llx got %p\n",
409 			     *(u64 *)&pgid, pg);
410 			return pg;
411 		}
412 	}
413 	return NULL;
414 }
415 
416 static int __remove_pg_mapping(struct rb_root *root, struct ceph_pg pgid)
417 {
418 	struct ceph_pg_mapping *pg = __lookup_pg_mapping(root, pgid);
419 
420 	if (pg) {
421 		dout("__remove_pg_mapping %llx %p\n", *(u64 *)&pgid, pg);
422 		rb_erase(&pg->node, root);
423 		kfree(pg);
424 		return 0;
425 	}
426 	dout("__remove_pg_mapping %llx dne\n", *(u64 *)&pgid);
427 	return -ENOENT;
428 }
429 
430 /*
431  * rbtree of pg pool info
432  */
433 static int __insert_pg_pool(struct rb_root *root, struct ceph_pg_pool_info *new)
434 {
435 	struct rb_node **p = &root->rb_node;
436 	struct rb_node *parent = NULL;
437 	struct ceph_pg_pool_info *pi = NULL;
438 
439 	while (*p) {
440 		parent = *p;
441 		pi = rb_entry(parent, struct ceph_pg_pool_info, node);
442 		if (new->id < pi->id)
443 			p = &(*p)->rb_left;
444 		else if (new->id > pi->id)
445 			p = &(*p)->rb_right;
446 		else
447 			return -EEXIST;
448 	}
449 
450 	rb_link_node(&new->node, parent, p);
451 	rb_insert_color(&new->node, root);
452 	return 0;
453 }
454 
455 static struct ceph_pg_pool_info *__lookup_pg_pool(struct rb_root *root, int id)
456 {
457 	struct ceph_pg_pool_info *pi;
458 	struct rb_node *n = root->rb_node;
459 
460 	while (n) {
461 		pi = rb_entry(n, struct ceph_pg_pool_info, node);
462 		if (id < pi->id)
463 			n = n->rb_left;
464 		else if (id > pi->id)
465 			n = n->rb_right;
466 		else
467 			return pi;
468 	}
469 	return NULL;
470 }
471 
472 const char *ceph_pg_pool_name_by_id(struct ceph_osdmap *map, u64 id)
473 {
474 	struct ceph_pg_pool_info *pi;
475 
476 	if (id == CEPH_NOPOOL)
477 		return NULL;
478 
479 	if (WARN_ON_ONCE(id > (u64) INT_MAX))
480 		return NULL;
481 
482 	pi = __lookup_pg_pool(&map->pg_pools, (int) id);
483 
484 	return pi ? pi->name : NULL;
485 }
486 EXPORT_SYMBOL(ceph_pg_pool_name_by_id);
487 
488 int ceph_pg_poolid_by_name(struct ceph_osdmap *map, const char *name)
489 {
490 	struct rb_node *rbp;
491 
492 	for (rbp = rb_first(&map->pg_pools); rbp; rbp = rb_next(rbp)) {
493 		struct ceph_pg_pool_info *pi =
494 			rb_entry(rbp, struct ceph_pg_pool_info, node);
495 		if (pi->name && strcmp(pi->name, name) == 0)
496 			return pi->id;
497 	}
498 	return -ENOENT;
499 }
500 EXPORT_SYMBOL(ceph_pg_poolid_by_name);
501 
502 static void __remove_pg_pool(struct rb_root *root, struct ceph_pg_pool_info *pi)
503 {
504 	rb_erase(&pi->node, root);
505 	kfree(pi->name);
506 	kfree(pi);
507 }
508 
509 static int __decode_pool(void **p, void *end, struct ceph_pg_pool_info *pi)
510 {
511 	unsigned int n, m;
512 
513 	ceph_decode_copy(p, &pi->v, sizeof(pi->v));
514 	calc_pg_masks(pi);
515 
516 	/* num_snaps * snap_info_t */
517 	n = le32_to_cpu(pi->v.num_snaps);
518 	while (n--) {
519 		ceph_decode_need(p, end, sizeof(u64) + 1 + sizeof(u64) +
520 				 sizeof(struct ceph_timespec), bad);
521 		*p += sizeof(u64) +       /* key */
522 			1 + sizeof(u64) + /* u8, snapid */
523 			sizeof(struct ceph_timespec);
524 		m = ceph_decode_32(p);    /* snap name */
525 		*p += m;
526 	}
527 
528 	*p += le32_to_cpu(pi->v.num_removed_snap_intervals) * sizeof(u64) * 2;
529 	return 0;
530 
531 bad:
532 	return -EINVAL;
533 }
534 
535 static int __decode_pool_names(void **p, void *end, struct ceph_osdmap *map)
536 {
537 	struct ceph_pg_pool_info *pi;
538 	u32 num, len, pool;
539 
540 	ceph_decode_32_safe(p, end, num, bad);
541 	dout(" %d pool names\n", num);
542 	while (num--) {
543 		ceph_decode_32_safe(p, end, pool, bad);
544 		ceph_decode_32_safe(p, end, len, bad);
545 		dout("  pool %d len %d\n", pool, len);
546 		ceph_decode_need(p, end, len, bad);
547 		pi = __lookup_pg_pool(&map->pg_pools, pool);
548 		if (pi) {
549 			char *name = kstrndup(*p, len, GFP_NOFS);
550 
551 			if (!name)
552 				return -ENOMEM;
553 			kfree(pi->name);
554 			pi->name = name;
555 			dout("  name is %s\n", pi->name);
556 		}
557 		*p += len;
558 	}
559 	return 0;
560 
561 bad:
562 	return -EINVAL;
563 }
564 
565 /*
566  * osd map
567  */
568 void ceph_osdmap_destroy(struct ceph_osdmap *map)
569 {
570 	dout("osdmap_destroy %p\n", map);
571 	if (map->crush)
572 		crush_destroy(map->crush);
573 	while (!RB_EMPTY_ROOT(&map->pg_temp)) {
574 		struct ceph_pg_mapping *pg =
575 			rb_entry(rb_first(&map->pg_temp),
576 				 struct ceph_pg_mapping, node);
577 		rb_erase(&pg->node, &map->pg_temp);
578 		kfree(pg);
579 	}
580 	while (!RB_EMPTY_ROOT(&map->pg_pools)) {
581 		struct ceph_pg_pool_info *pi =
582 			rb_entry(rb_first(&map->pg_pools),
583 				 struct ceph_pg_pool_info, node);
584 		__remove_pg_pool(&map->pg_pools, pi);
585 	}
586 	kfree(map->osd_state);
587 	kfree(map->osd_weight);
588 	kfree(map->osd_addr);
589 	kfree(map);
590 }
591 
592 /*
593  * adjust max osd value.  reallocate arrays.
594  */
595 static int osdmap_set_max_osd(struct ceph_osdmap *map, int max)
596 {
597 	u8 *state;
598 	struct ceph_entity_addr *addr;
599 	u32 *weight;
600 
601 	state = kcalloc(max, sizeof(*state), GFP_NOFS);
602 	addr = kcalloc(max, sizeof(*addr), GFP_NOFS);
603 	weight = kcalloc(max, sizeof(*weight), GFP_NOFS);
604 	if (state == NULL || addr == NULL || weight == NULL) {
605 		kfree(state);
606 		kfree(addr);
607 		kfree(weight);
608 		return -ENOMEM;
609 	}
610 
611 	/* copy old? */
612 	if (map->osd_state) {
613 		memcpy(state, map->osd_state, map->max_osd*sizeof(*state));
614 		memcpy(addr, map->osd_addr, map->max_osd*sizeof(*addr));
615 		memcpy(weight, map->osd_weight, map->max_osd*sizeof(*weight));
616 		kfree(map->osd_state);
617 		kfree(map->osd_addr);
618 		kfree(map->osd_weight);
619 	}
620 
621 	map->osd_state = state;
622 	map->osd_weight = weight;
623 	map->osd_addr = addr;
624 	map->max_osd = max;
625 	return 0;
626 }
627 
628 /*
629  * decode a full map.
630  */
631 struct ceph_osdmap *osdmap_decode(void **p, void *end)
632 {
633 	struct ceph_osdmap *map;
634 	u16 version;
635 	u32 len, max, i;
636 	u8 ev;
637 	int err = -EINVAL;
638 	void *start = *p;
639 	struct ceph_pg_pool_info *pi;
640 
641 	dout("osdmap_decode %p to %p len %d\n", *p, end, (int)(end - *p));
642 
643 	map = kzalloc(sizeof(*map), GFP_NOFS);
644 	if (map == NULL)
645 		return ERR_PTR(-ENOMEM);
646 	map->pg_temp = RB_ROOT;
647 
648 	ceph_decode_16_safe(p, end, version, bad);
649 	if (version > CEPH_OSDMAP_VERSION) {
650 		pr_warning("got unknown v %d > %d of osdmap\n", version,
651 			   CEPH_OSDMAP_VERSION);
652 		goto bad;
653 	}
654 
655 	ceph_decode_need(p, end, 2*sizeof(u64)+6*sizeof(u32), bad);
656 	ceph_decode_copy(p, &map->fsid, sizeof(map->fsid));
657 	map->epoch = ceph_decode_32(p);
658 	ceph_decode_copy(p, &map->created, sizeof(map->created));
659 	ceph_decode_copy(p, &map->modified, sizeof(map->modified));
660 
661 	ceph_decode_32_safe(p, end, max, bad);
662 	while (max--) {
663 		ceph_decode_need(p, end, 4 + 1 + sizeof(pi->v), bad);
664 		err = -ENOMEM;
665 		pi = kzalloc(sizeof(*pi), GFP_NOFS);
666 		if (!pi)
667 			goto bad;
668 		pi->id = ceph_decode_32(p);
669 		err = -EINVAL;
670 		ev = ceph_decode_8(p); /* encoding version */
671 		if (ev > CEPH_PG_POOL_VERSION) {
672 			pr_warning("got unknown v %d > %d of ceph_pg_pool\n",
673 				   ev, CEPH_PG_POOL_VERSION);
674 			kfree(pi);
675 			goto bad;
676 		}
677 		err = __decode_pool(p, end, pi);
678 		if (err < 0) {
679 			kfree(pi);
680 			goto bad;
681 		}
682 		__insert_pg_pool(&map->pg_pools, pi);
683 	}
684 
685 	if (version >= 5) {
686 		err = __decode_pool_names(p, end, map);
687 		if (err < 0) {
688 			dout("fail to decode pool names");
689 			goto bad;
690 		}
691 	}
692 
693 	ceph_decode_32_safe(p, end, map->pool_max, bad);
694 
695 	ceph_decode_32_safe(p, end, map->flags, bad);
696 
697 	max = ceph_decode_32(p);
698 
699 	/* (re)alloc osd arrays */
700 	err = osdmap_set_max_osd(map, max);
701 	if (err < 0)
702 		goto bad;
703 	dout("osdmap_decode max_osd = %d\n", map->max_osd);
704 
705 	/* osds */
706 	err = -EINVAL;
707 	ceph_decode_need(p, end, 3*sizeof(u32) +
708 			 map->max_osd*(1 + sizeof(*map->osd_weight) +
709 				       sizeof(*map->osd_addr)), bad);
710 	*p += 4; /* skip length field (should match max) */
711 	ceph_decode_copy(p, map->osd_state, map->max_osd);
712 
713 	*p += 4; /* skip length field (should match max) */
714 	for (i = 0; i < map->max_osd; i++)
715 		map->osd_weight[i] = ceph_decode_32(p);
716 
717 	*p += 4; /* skip length field (should match max) */
718 	ceph_decode_copy(p, map->osd_addr, map->max_osd*sizeof(*map->osd_addr));
719 	for (i = 0; i < map->max_osd; i++)
720 		ceph_decode_addr(&map->osd_addr[i]);
721 
722 	/* pg_temp */
723 	ceph_decode_32_safe(p, end, len, bad);
724 	for (i = 0; i < len; i++) {
725 		int n, j;
726 		struct ceph_pg pgid;
727 		struct ceph_pg_mapping *pg;
728 
729 		ceph_decode_need(p, end, sizeof(u32) + sizeof(u64), bad);
730 		ceph_decode_copy(p, &pgid, sizeof(pgid));
731 		n = ceph_decode_32(p);
732 		err = -EINVAL;
733 		if (n > (UINT_MAX - sizeof(*pg)) / sizeof(u32))
734 			goto bad;
735 		ceph_decode_need(p, end, n * sizeof(u32), bad);
736 		err = -ENOMEM;
737 		pg = kmalloc(sizeof(*pg) + n*sizeof(u32), GFP_NOFS);
738 		if (!pg)
739 			goto bad;
740 		pg->pgid = pgid;
741 		pg->len = n;
742 		for (j = 0; j < n; j++)
743 			pg->osds[j] = ceph_decode_32(p);
744 
745 		err = __insert_pg_mapping(pg, &map->pg_temp);
746 		if (err)
747 			goto bad;
748 		dout(" added pg_temp %llx len %d\n", *(u64 *)&pgid, len);
749 	}
750 
751 	/* crush */
752 	ceph_decode_32_safe(p, end, len, bad);
753 	dout("osdmap_decode crush len %d from off 0x%x\n", len,
754 	     (int)(*p - start));
755 	ceph_decode_need(p, end, len, bad);
756 	map->crush = crush_decode(*p, end);
757 	*p += len;
758 	if (IS_ERR(map->crush)) {
759 		err = PTR_ERR(map->crush);
760 		map->crush = NULL;
761 		goto bad;
762 	}
763 
764 	/* ignore the rest of the map */
765 	*p = end;
766 
767 	dout("osdmap_decode done %p %p\n", *p, end);
768 	return map;
769 
770 bad:
771 	dout("osdmap_decode fail err %d\n", err);
772 	ceph_osdmap_destroy(map);
773 	return ERR_PTR(err);
774 }
775 
776 /*
777  * decode and apply an incremental map update.
778  */
779 struct ceph_osdmap *osdmap_apply_incremental(void **p, void *end,
780 					     struct ceph_osdmap *map,
781 					     struct ceph_messenger *msgr)
782 {
783 	struct crush_map *newcrush = NULL;
784 	struct ceph_fsid fsid;
785 	u32 epoch = 0;
786 	struct ceph_timespec modified;
787 	u32 len, pool;
788 	__s32 new_pool_max, new_flags, max;
789 	void *start = *p;
790 	int err = -EINVAL;
791 	u16 version;
792 
793 	ceph_decode_16_safe(p, end, version, bad);
794 	if (version > CEPH_OSDMAP_INC_VERSION) {
795 		pr_warning("got unknown v %d > %d of inc osdmap\n", version,
796 			   CEPH_OSDMAP_INC_VERSION);
797 		goto bad;
798 	}
799 
800 	ceph_decode_need(p, end, sizeof(fsid)+sizeof(modified)+2*sizeof(u32),
801 			 bad);
802 	ceph_decode_copy(p, &fsid, sizeof(fsid));
803 	epoch = ceph_decode_32(p);
804 	BUG_ON(epoch != map->epoch+1);
805 	ceph_decode_copy(p, &modified, sizeof(modified));
806 	new_pool_max = ceph_decode_32(p);
807 	new_flags = ceph_decode_32(p);
808 
809 	/* full map? */
810 	ceph_decode_32_safe(p, end, len, bad);
811 	if (len > 0) {
812 		dout("apply_incremental full map len %d, %p to %p\n",
813 		     len, *p, end);
814 		return osdmap_decode(p, min(*p+len, end));
815 	}
816 
817 	/* new crush? */
818 	ceph_decode_32_safe(p, end, len, bad);
819 	if (len > 0) {
820 		dout("apply_incremental new crush map len %d, %p to %p\n",
821 		     len, *p, end);
822 		newcrush = crush_decode(*p, min(*p+len, end));
823 		if (IS_ERR(newcrush))
824 			return ERR_CAST(newcrush);
825 		*p += len;
826 	}
827 
828 	/* new flags? */
829 	if (new_flags >= 0)
830 		map->flags = new_flags;
831 	if (new_pool_max >= 0)
832 		map->pool_max = new_pool_max;
833 
834 	ceph_decode_need(p, end, 5*sizeof(u32), bad);
835 
836 	/* new max? */
837 	max = ceph_decode_32(p);
838 	if (max >= 0) {
839 		err = osdmap_set_max_osd(map, max);
840 		if (err < 0)
841 			goto bad;
842 	}
843 
844 	map->epoch++;
845 	map->modified = modified;
846 	if (newcrush) {
847 		if (map->crush)
848 			crush_destroy(map->crush);
849 		map->crush = newcrush;
850 		newcrush = NULL;
851 	}
852 
853 	/* new_pool */
854 	ceph_decode_32_safe(p, end, len, bad);
855 	while (len--) {
856 		__u8 ev;
857 		struct ceph_pg_pool_info *pi;
858 
859 		ceph_decode_32_safe(p, end, pool, bad);
860 		ceph_decode_need(p, end, 1 + sizeof(pi->v), bad);
861 		ev = ceph_decode_8(p);  /* encoding version */
862 		if (ev > CEPH_PG_POOL_VERSION) {
863 			pr_warning("got unknown v %d > %d of ceph_pg_pool\n",
864 				   ev, CEPH_PG_POOL_VERSION);
865 			err = -EINVAL;
866 			goto bad;
867 		}
868 		pi = __lookup_pg_pool(&map->pg_pools, pool);
869 		if (!pi) {
870 			pi = kzalloc(sizeof(*pi), GFP_NOFS);
871 			if (!pi) {
872 				err = -ENOMEM;
873 				goto bad;
874 			}
875 			pi->id = pool;
876 			__insert_pg_pool(&map->pg_pools, pi);
877 		}
878 		err = __decode_pool(p, end, pi);
879 		if (err < 0)
880 			goto bad;
881 	}
882 	if (version >= 5) {
883 		err = __decode_pool_names(p, end, map);
884 		if (err < 0)
885 			goto bad;
886 	}
887 
888 	/* old_pool */
889 	ceph_decode_32_safe(p, end, len, bad);
890 	while (len--) {
891 		struct ceph_pg_pool_info *pi;
892 
893 		ceph_decode_32_safe(p, end, pool, bad);
894 		pi = __lookup_pg_pool(&map->pg_pools, pool);
895 		if (pi)
896 			__remove_pg_pool(&map->pg_pools, pi);
897 	}
898 
899 	/* new_up */
900 	err = -EINVAL;
901 	ceph_decode_32_safe(p, end, len, bad);
902 	while (len--) {
903 		u32 osd;
904 		struct ceph_entity_addr addr;
905 		ceph_decode_32_safe(p, end, osd, bad);
906 		ceph_decode_copy_safe(p, end, &addr, sizeof(addr), bad);
907 		ceph_decode_addr(&addr);
908 		pr_info("osd%d up\n", osd);
909 		BUG_ON(osd >= map->max_osd);
910 		map->osd_state[osd] |= CEPH_OSD_UP;
911 		map->osd_addr[osd] = addr;
912 	}
913 
914 	/* new_state */
915 	ceph_decode_32_safe(p, end, len, bad);
916 	while (len--) {
917 		u32 osd;
918 		u8 xorstate;
919 		ceph_decode_32_safe(p, end, osd, bad);
920 		xorstate = **(u8 **)p;
921 		(*p)++;  /* clean flag */
922 		if (xorstate == 0)
923 			xorstate = CEPH_OSD_UP;
924 		if (xorstate & CEPH_OSD_UP)
925 			pr_info("osd%d down\n", osd);
926 		if (osd < map->max_osd)
927 			map->osd_state[osd] ^= xorstate;
928 	}
929 
930 	/* new_weight */
931 	ceph_decode_32_safe(p, end, len, bad);
932 	while (len--) {
933 		u32 osd, off;
934 		ceph_decode_need(p, end, sizeof(u32)*2, bad);
935 		osd = ceph_decode_32(p);
936 		off = ceph_decode_32(p);
937 		pr_info("osd%d weight 0x%x %s\n", osd, off,
938 		     off == CEPH_OSD_IN ? "(in)" :
939 		     (off == CEPH_OSD_OUT ? "(out)" : ""));
940 		if (osd < map->max_osd)
941 			map->osd_weight[osd] = off;
942 	}
943 
944 	/* new_pg_temp */
945 	ceph_decode_32_safe(p, end, len, bad);
946 	while (len--) {
947 		struct ceph_pg_mapping *pg;
948 		int j;
949 		struct ceph_pg pgid;
950 		u32 pglen;
951 		ceph_decode_need(p, end, sizeof(u64) + sizeof(u32), bad);
952 		ceph_decode_copy(p, &pgid, sizeof(pgid));
953 		pglen = ceph_decode_32(p);
954 
955 		if (pglen) {
956 			ceph_decode_need(p, end, pglen*sizeof(u32), bad);
957 
958 			/* removing existing (if any) */
959 			(void) __remove_pg_mapping(&map->pg_temp, pgid);
960 
961 			/* insert */
962 			err = -EINVAL;
963 			if (pglen > (UINT_MAX - sizeof(*pg)) / sizeof(u32))
964 				goto bad;
965 			err = -ENOMEM;
966 			pg = kmalloc(sizeof(*pg) + sizeof(u32)*pglen, GFP_NOFS);
967 			if (!pg)
968 				goto bad;
969 			pg->pgid = pgid;
970 			pg->len = pglen;
971 			for (j = 0; j < pglen; j++)
972 				pg->osds[j] = ceph_decode_32(p);
973 			err = __insert_pg_mapping(pg, &map->pg_temp);
974 			if (err) {
975 				kfree(pg);
976 				goto bad;
977 			}
978 			dout(" added pg_temp %llx len %d\n", *(u64 *)&pgid,
979 			     pglen);
980 		} else {
981 			/* remove */
982 			__remove_pg_mapping(&map->pg_temp, pgid);
983 		}
984 	}
985 
986 	/* ignore the rest */
987 	*p = end;
988 	return map;
989 
990 bad:
991 	pr_err("corrupt inc osdmap epoch %d off %d (%p of %p-%p)\n",
992 	       epoch, (int)(*p - start), *p, start, end);
993 	print_hex_dump(KERN_DEBUG, "osdmap: ",
994 		       DUMP_PREFIX_OFFSET, 16, 1,
995 		       start, end - start, true);
996 	if (newcrush)
997 		crush_destroy(newcrush);
998 	return ERR_PTR(err);
999 }
1000 
1001 
1002 
1003 
1004 /*
1005  * calculate file layout from given offset, length.
1006  * fill in correct oid, logical length, and object extent
1007  * offset, length.
1008  *
1009  * for now, we write only a single su, until we can
1010  * pass a stride back to the caller.
1011  */
1012 int ceph_calc_file_object_mapping(struct ceph_file_layout *layout,
1013 				   u64 off, u64 *plen,
1014 				   u64 *ono,
1015 				   u64 *oxoff, u64 *oxlen)
1016 {
1017 	u32 osize = le32_to_cpu(layout->fl_object_size);
1018 	u32 su = le32_to_cpu(layout->fl_stripe_unit);
1019 	u32 sc = le32_to_cpu(layout->fl_stripe_count);
1020 	u32 bl, stripeno, stripepos, objsetno;
1021 	u32 su_per_object;
1022 	u64 t, su_offset;
1023 
1024 	dout("mapping %llu~%llu  osize %u fl_su %u\n", off, *plen,
1025 	     osize, su);
1026 	if (su == 0 || sc == 0)
1027 		goto invalid;
1028 	su_per_object = osize / su;
1029 	if (su_per_object == 0)
1030 		goto invalid;
1031 	dout("osize %u / su %u = su_per_object %u\n", osize, su,
1032 	     su_per_object);
1033 
1034 	if ((su & ~PAGE_MASK) != 0)
1035 		goto invalid;
1036 
1037 	/* bl = *off / su; */
1038 	t = off;
1039 	do_div(t, su);
1040 	bl = t;
1041 	dout("off %llu / su %u = bl %u\n", off, su, bl);
1042 
1043 	stripeno = bl / sc;
1044 	stripepos = bl % sc;
1045 	objsetno = stripeno / su_per_object;
1046 
1047 	*ono = objsetno * sc + stripepos;
1048 	dout("objset %u * sc %u = ono %u\n", objsetno, sc, (unsigned int)*ono);
1049 
1050 	/* *oxoff = *off % layout->fl_stripe_unit;  # offset in su */
1051 	t = off;
1052 	su_offset = do_div(t, su);
1053 	*oxoff = su_offset + (stripeno % su_per_object) * su;
1054 
1055 	/*
1056 	 * Calculate the length of the extent being written to the selected
1057 	 * object. This is the minimum of the full length requested (plen) or
1058 	 * the remainder of the current stripe being written to.
1059 	 */
1060 	*oxlen = min_t(u64, *plen, su - su_offset);
1061 	*plen = *oxlen;
1062 
1063 	dout(" obj extent %llu~%llu\n", *oxoff, *oxlen);
1064 	return 0;
1065 
1066 invalid:
1067 	dout(" invalid layout\n");
1068 	*ono = 0;
1069 	*oxoff = 0;
1070 	*oxlen = 0;
1071 	return -EINVAL;
1072 }
1073 EXPORT_SYMBOL(ceph_calc_file_object_mapping);
1074 
1075 /*
1076  * calculate an object layout (i.e. pgid) from an oid,
1077  * file_layout, and osdmap
1078  */
1079 int ceph_calc_object_layout(struct ceph_object_layout *ol,
1080 			    const char *oid,
1081 			    struct ceph_file_layout *fl,
1082 			    struct ceph_osdmap *osdmap)
1083 {
1084 	unsigned int num, num_mask;
1085 	struct ceph_pg pgid;
1086 	int poolid = le32_to_cpu(fl->fl_pg_pool);
1087 	struct ceph_pg_pool_info *pool;
1088 	unsigned int ps;
1089 
1090 	BUG_ON(!osdmap);
1091 
1092 	pool = __lookup_pg_pool(&osdmap->pg_pools, poolid);
1093 	if (!pool)
1094 		return -EIO;
1095 	ps = ceph_str_hash(pool->v.object_hash, oid, strlen(oid));
1096 	num = le32_to_cpu(pool->v.pg_num);
1097 	num_mask = pool->pg_num_mask;
1098 
1099 	pgid.ps = cpu_to_le16(ps);
1100 	pgid.preferred = cpu_to_le16(-1);
1101 	pgid.pool = fl->fl_pg_pool;
1102 	dout("calc_object_layout '%s' pgid %d.%x\n", oid, poolid, ps);
1103 
1104 	ol->ol_pgid = pgid;
1105 	ol->ol_stripe_unit = fl->fl_object_stripe_unit;
1106 	return 0;
1107 }
1108 EXPORT_SYMBOL(ceph_calc_object_layout);
1109 
1110 /*
1111  * Calculate raw osd vector for the given pgid.  Return pointer to osd
1112  * array, or NULL on failure.
1113  */
1114 static int *calc_pg_raw(struct ceph_osdmap *osdmap, struct ceph_pg pgid,
1115 			int *osds, int *num)
1116 {
1117 	struct ceph_pg_mapping *pg;
1118 	struct ceph_pg_pool_info *pool;
1119 	int ruleno;
1120 	unsigned int poolid, ps, pps, t, r;
1121 
1122 	poolid = le32_to_cpu(pgid.pool);
1123 	ps = le16_to_cpu(pgid.ps);
1124 
1125 	pool = __lookup_pg_pool(&osdmap->pg_pools, poolid);
1126 	if (!pool)
1127 		return NULL;
1128 
1129 	/* pg_temp? */
1130 	t = ceph_stable_mod(ps, le32_to_cpu(pool->v.pg_num),
1131 			    pool->pgp_num_mask);
1132 	pgid.ps = cpu_to_le16(t);
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->v.crush_ruleset,
1141 				 pool->v.type, pool->v.size);
1142 	if (ruleno < 0) {
1143 		pr_err("no crush rule pool %d ruleset %d type %d size %d\n",
1144 		       poolid, pool->v.crush_ruleset, pool->v.type,
1145 		       pool->v.size);
1146 		return NULL;
1147 	}
1148 
1149 	pps = ceph_stable_mod(ps,
1150 			      le32_to_cpu(pool->v.pgp_num),
1151 			      pool->pgp_num_mask);
1152 	pps += poolid;
1153 	r = crush_do_rule(osdmap->crush, ruleno, pps, osds,
1154 			  min_t(int, pool->v.size, *num),
1155 			  osdmap->osd_weight);
1156 	if (r < 0) {
1157 		pr_err("error %d from crush rule: pool %d ruleset %d type %d"
1158 		       " size %d\n", r, poolid, pool->v.crush_ruleset,
1159 		       pool->v.type, pool->v.size);
1160 		return NULL;
1161 	}
1162 	*num = r;
1163 	return osds;
1164 }
1165 
1166 /*
1167  * Return acting set for given pgid.
1168  */
1169 int ceph_calc_pg_acting(struct ceph_osdmap *osdmap, struct ceph_pg pgid,
1170 			int *acting)
1171 {
1172 	int rawosds[CEPH_PG_MAX_SIZE], *osds;
1173 	int i, o, num = CEPH_PG_MAX_SIZE;
1174 
1175 	osds = calc_pg_raw(osdmap, pgid, rawosds, &num);
1176 	if (!osds)
1177 		return -1;
1178 
1179 	/* primary is first up osd */
1180 	o = 0;
1181 	for (i = 0; i < num; i++)
1182 		if (ceph_osd_is_up(osdmap, osds[i]))
1183 			acting[o++] = osds[i];
1184 	return o;
1185 }
1186 
1187 /*
1188  * Return primary osd for given pgid, or -1 if none.
1189  */
1190 int ceph_calc_pg_primary(struct ceph_osdmap *osdmap, struct ceph_pg pgid)
1191 {
1192 	int rawosds[CEPH_PG_MAX_SIZE], *osds;
1193 	int i, num = CEPH_PG_MAX_SIZE;
1194 
1195 	osds = calc_pg_raw(osdmap, pgid, rawosds, &num);
1196 	if (!osds)
1197 		return -1;
1198 
1199 	/* primary is first up osd */
1200 	for (i = 0; i < num; i++)
1201 		if (ceph_osd_is_up(osdmap, osds[i]))
1202 			return osds[i];
1203 	return -1;
1204 }
1205 EXPORT_SYMBOL(ceph_calc_pg_primary);
1206