xref: /linux/net/ceph/osdmap.c (revision 26b0d14106954ae46d2f4f7eec3481828a210f7d)
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 struct crush_map *crush_decode(void *pbyval, void *end)
139 {
140 	struct crush_map *c;
141 	int err = -EINVAL;
142 	int i, j;
143 	void **p = &pbyval;
144 	void *start = pbyval;
145 	u32 magic;
146 
147 	dout("crush_decode %p to %p len %d\n", *p, end, (int)(end - *p));
148 
149 	c = kzalloc(sizeof(*c), GFP_NOFS);
150 	if (c == NULL)
151 		return ERR_PTR(-ENOMEM);
152 
153 	ceph_decode_need(p, end, 4*sizeof(u32), bad);
154 	magic = ceph_decode_32(p);
155 	if (magic != CRUSH_MAGIC) {
156 		pr_err("crush_decode magic %x != current %x\n",
157 		       (unsigned int)magic, (unsigned int)CRUSH_MAGIC);
158 		goto bad;
159 	}
160 	c->max_buckets = ceph_decode_32(p);
161 	c->max_rules = ceph_decode_32(p);
162 	c->max_devices = ceph_decode_32(p);
163 
164 	c->buckets = kcalloc(c->max_buckets, sizeof(*c->buckets), GFP_NOFS);
165 	if (c->buckets == NULL)
166 		goto badmem;
167 	c->rules = kcalloc(c->max_rules, sizeof(*c->rules), GFP_NOFS);
168 	if (c->rules == NULL)
169 		goto badmem;
170 
171 	/* buckets */
172 	for (i = 0; i < c->max_buckets; i++) {
173 		int size = 0;
174 		u32 alg;
175 		struct crush_bucket *b;
176 
177 		ceph_decode_32_safe(p, end, alg, bad);
178 		if (alg == 0) {
179 			c->buckets[i] = NULL;
180 			continue;
181 		}
182 		dout("crush_decode bucket %d off %x %p to %p\n",
183 		     i, (int)(*p-start), *p, end);
184 
185 		switch (alg) {
186 		case CRUSH_BUCKET_UNIFORM:
187 			size = sizeof(struct crush_bucket_uniform);
188 			break;
189 		case CRUSH_BUCKET_LIST:
190 			size = sizeof(struct crush_bucket_list);
191 			break;
192 		case CRUSH_BUCKET_TREE:
193 			size = sizeof(struct crush_bucket_tree);
194 			break;
195 		case CRUSH_BUCKET_STRAW:
196 			size = sizeof(struct crush_bucket_straw);
197 			break;
198 		default:
199 			err = -EINVAL;
200 			goto bad;
201 		}
202 		BUG_ON(size == 0);
203 		b = c->buckets[i] = kzalloc(size, GFP_NOFS);
204 		if (b == NULL)
205 			goto badmem;
206 
207 		ceph_decode_need(p, end, 4*sizeof(u32), bad);
208 		b->id = ceph_decode_32(p);
209 		b->type = ceph_decode_16(p);
210 		b->alg = ceph_decode_8(p);
211 		b->hash = ceph_decode_8(p);
212 		b->weight = ceph_decode_32(p);
213 		b->size = ceph_decode_32(p);
214 
215 		dout("crush_decode bucket size %d off %x %p to %p\n",
216 		     b->size, (int)(*p-start), *p, end);
217 
218 		b->items = kcalloc(b->size, sizeof(__s32), GFP_NOFS);
219 		if (b->items == NULL)
220 			goto badmem;
221 		b->perm = kcalloc(b->size, sizeof(u32), GFP_NOFS);
222 		if (b->perm == NULL)
223 			goto badmem;
224 		b->perm_n = 0;
225 
226 		ceph_decode_need(p, end, b->size*sizeof(u32), bad);
227 		for (j = 0; j < b->size; j++)
228 			b->items[j] = ceph_decode_32(p);
229 
230 		switch (b->alg) {
231 		case CRUSH_BUCKET_UNIFORM:
232 			err = crush_decode_uniform_bucket(p, end,
233 				  (struct crush_bucket_uniform *)b);
234 			if (err < 0)
235 				goto bad;
236 			break;
237 		case CRUSH_BUCKET_LIST:
238 			err = crush_decode_list_bucket(p, end,
239 			       (struct crush_bucket_list *)b);
240 			if (err < 0)
241 				goto bad;
242 			break;
243 		case CRUSH_BUCKET_TREE:
244 			err = crush_decode_tree_bucket(p, end,
245 				(struct crush_bucket_tree *)b);
246 			if (err < 0)
247 				goto bad;
248 			break;
249 		case CRUSH_BUCKET_STRAW:
250 			err = crush_decode_straw_bucket(p, end,
251 				(struct crush_bucket_straw *)b);
252 			if (err < 0)
253 				goto bad;
254 			break;
255 		}
256 	}
257 
258 	/* rules */
259 	dout("rule vec is %p\n", c->rules);
260 	for (i = 0; i < c->max_rules; i++) {
261 		u32 yes;
262 		struct crush_rule *r;
263 
264 		ceph_decode_32_safe(p, end, yes, bad);
265 		if (!yes) {
266 			dout("crush_decode NO rule %d off %x %p to %p\n",
267 			     i, (int)(*p-start), *p, end);
268 			c->rules[i] = NULL;
269 			continue;
270 		}
271 
272 		dout("crush_decode rule %d off %x %p to %p\n",
273 		     i, (int)(*p-start), *p, end);
274 
275 		/* len */
276 		ceph_decode_32_safe(p, end, yes, bad);
277 #if BITS_PER_LONG == 32
278 		err = -EINVAL;
279 		if (yes > (ULONG_MAX - sizeof(*r))
280 			  / sizeof(struct crush_rule_step))
281 			goto bad;
282 #endif
283 		r = c->rules[i] = kmalloc(sizeof(*r) +
284 					  yes*sizeof(struct crush_rule_step),
285 					  GFP_NOFS);
286 		if (r == NULL)
287 			goto badmem;
288 		dout(" rule %d is at %p\n", i, r);
289 		r->len = yes;
290 		ceph_decode_copy_safe(p, end, &r->mask, 4, bad); /* 4 u8's */
291 		ceph_decode_need(p, end, r->len*3*sizeof(u32), bad);
292 		for (j = 0; j < r->len; j++) {
293 			r->steps[j].op = ceph_decode_32(p);
294 			r->steps[j].arg1 = ceph_decode_32(p);
295 			r->steps[j].arg2 = ceph_decode_32(p);
296 		}
297 	}
298 
299 	/* ignore trailing name maps. */
300 
301 	dout("crush_decode success\n");
302 	return c;
303 
304 badmem:
305 	err = -ENOMEM;
306 bad:
307 	dout("crush_decode fail %d\n", err);
308 	crush_destroy(c);
309 	return ERR_PTR(err);
310 }
311 
312 /*
313  * rbtree of pg_mapping for handling pg_temp (explicit mapping of pgid
314  * to a set of osds)
315  */
316 static int pgid_cmp(struct ceph_pg l, struct ceph_pg r)
317 {
318 	u64 a = *(u64 *)&l;
319 	u64 b = *(u64 *)&r;
320 
321 	if (a < b)
322 		return -1;
323 	if (a > b)
324 		return 1;
325 	return 0;
326 }
327 
328 static int __insert_pg_mapping(struct ceph_pg_mapping *new,
329 			       struct rb_root *root)
330 {
331 	struct rb_node **p = &root->rb_node;
332 	struct rb_node *parent = NULL;
333 	struct ceph_pg_mapping *pg = NULL;
334 	int c;
335 
336 	dout("__insert_pg_mapping %llx %p\n", *(u64 *)&new->pgid, new);
337 	while (*p) {
338 		parent = *p;
339 		pg = rb_entry(parent, struct ceph_pg_mapping, node);
340 		c = pgid_cmp(new->pgid, pg->pgid);
341 		if (c < 0)
342 			p = &(*p)->rb_left;
343 		else if (c > 0)
344 			p = &(*p)->rb_right;
345 		else
346 			return -EEXIST;
347 	}
348 
349 	rb_link_node(&new->node, parent, p);
350 	rb_insert_color(&new->node, root);
351 	return 0;
352 }
353 
354 static struct ceph_pg_mapping *__lookup_pg_mapping(struct rb_root *root,
355 						   struct ceph_pg pgid)
356 {
357 	struct rb_node *n = root->rb_node;
358 	struct ceph_pg_mapping *pg;
359 	int c;
360 
361 	while (n) {
362 		pg = rb_entry(n, struct ceph_pg_mapping, node);
363 		c = pgid_cmp(pgid, pg->pgid);
364 		if (c < 0) {
365 			n = n->rb_left;
366 		} else if (c > 0) {
367 			n = n->rb_right;
368 		} else {
369 			dout("__lookup_pg_mapping %llx got %p\n",
370 			     *(u64 *)&pgid, pg);
371 			return pg;
372 		}
373 	}
374 	return NULL;
375 }
376 
377 static int __remove_pg_mapping(struct rb_root *root, struct ceph_pg pgid)
378 {
379 	struct ceph_pg_mapping *pg = __lookup_pg_mapping(root, pgid);
380 
381 	if (pg) {
382 		dout("__remove_pg_mapping %llx %p\n", *(u64 *)&pgid, pg);
383 		rb_erase(&pg->node, root);
384 		kfree(pg);
385 		return 0;
386 	}
387 	dout("__remove_pg_mapping %llx dne\n", *(u64 *)&pgid);
388 	return -ENOENT;
389 }
390 
391 /*
392  * rbtree of pg pool info
393  */
394 static int __insert_pg_pool(struct rb_root *root, struct ceph_pg_pool_info *new)
395 {
396 	struct rb_node **p = &root->rb_node;
397 	struct rb_node *parent = NULL;
398 	struct ceph_pg_pool_info *pi = NULL;
399 
400 	while (*p) {
401 		parent = *p;
402 		pi = rb_entry(parent, struct ceph_pg_pool_info, node);
403 		if (new->id < pi->id)
404 			p = &(*p)->rb_left;
405 		else if (new->id > pi->id)
406 			p = &(*p)->rb_right;
407 		else
408 			return -EEXIST;
409 	}
410 
411 	rb_link_node(&new->node, parent, p);
412 	rb_insert_color(&new->node, root);
413 	return 0;
414 }
415 
416 static struct ceph_pg_pool_info *__lookup_pg_pool(struct rb_root *root, int id)
417 {
418 	struct ceph_pg_pool_info *pi;
419 	struct rb_node *n = root->rb_node;
420 
421 	while (n) {
422 		pi = rb_entry(n, struct ceph_pg_pool_info, node);
423 		if (id < pi->id)
424 			n = n->rb_left;
425 		else if (id > pi->id)
426 			n = n->rb_right;
427 		else
428 			return pi;
429 	}
430 	return NULL;
431 }
432 
433 int ceph_pg_poolid_by_name(struct ceph_osdmap *map, const char *name)
434 {
435 	struct rb_node *rbp;
436 
437 	for (rbp = rb_first(&map->pg_pools); rbp; rbp = rb_next(rbp)) {
438 		struct ceph_pg_pool_info *pi =
439 			rb_entry(rbp, struct ceph_pg_pool_info, node);
440 		if (pi->name && strcmp(pi->name, name) == 0)
441 			return pi->id;
442 	}
443 	return -ENOENT;
444 }
445 EXPORT_SYMBOL(ceph_pg_poolid_by_name);
446 
447 static void __remove_pg_pool(struct rb_root *root, struct ceph_pg_pool_info *pi)
448 {
449 	rb_erase(&pi->node, root);
450 	kfree(pi->name);
451 	kfree(pi);
452 }
453 
454 static int __decode_pool(void **p, void *end, struct ceph_pg_pool_info *pi)
455 {
456 	unsigned int n, m;
457 
458 	ceph_decode_copy(p, &pi->v, sizeof(pi->v));
459 	calc_pg_masks(pi);
460 
461 	/* num_snaps * snap_info_t */
462 	n = le32_to_cpu(pi->v.num_snaps);
463 	while (n--) {
464 		ceph_decode_need(p, end, sizeof(u64) + 1 + sizeof(u64) +
465 				 sizeof(struct ceph_timespec), bad);
466 		*p += sizeof(u64) +       /* key */
467 			1 + sizeof(u64) + /* u8, snapid */
468 			sizeof(struct ceph_timespec);
469 		m = ceph_decode_32(p);    /* snap name */
470 		*p += m;
471 	}
472 
473 	*p += le32_to_cpu(pi->v.num_removed_snap_intervals) * sizeof(u64) * 2;
474 	return 0;
475 
476 bad:
477 	return -EINVAL;
478 }
479 
480 static int __decode_pool_names(void **p, void *end, struct ceph_osdmap *map)
481 {
482 	struct ceph_pg_pool_info *pi;
483 	u32 num, len, pool;
484 
485 	ceph_decode_32_safe(p, end, num, bad);
486 	dout(" %d pool names\n", num);
487 	while (num--) {
488 		ceph_decode_32_safe(p, end, pool, bad);
489 		ceph_decode_32_safe(p, end, len, bad);
490 		dout("  pool %d len %d\n", pool, len);
491 		pi = __lookup_pg_pool(&map->pg_pools, pool);
492 		if (pi) {
493 			kfree(pi->name);
494 			pi->name = kmalloc(len + 1, GFP_NOFS);
495 			if (pi->name) {
496 				memcpy(pi->name, *p, len);
497 				pi->name[len] = '\0';
498 				dout("  name is %s\n", pi->name);
499 			}
500 		}
501 		*p += len;
502 	}
503 	return 0;
504 
505 bad:
506 	return -EINVAL;
507 }
508 
509 /*
510  * osd map
511  */
512 void ceph_osdmap_destroy(struct ceph_osdmap *map)
513 {
514 	dout("osdmap_destroy %p\n", map);
515 	if (map->crush)
516 		crush_destroy(map->crush);
517 	while (!RB_EMPTY_ROOT(&map->pg_temp)) {
518 		struct ceph_pg_mapping *pg =
519 			rb_entry(rb_first(&map->pg_temp),
520 				 struct ceph_pg_mapping, node);
521 		rb_erase(&pg->node, &map->pg_temp);
522 		kfree(pg);
523 	}
524 	while (!RB_EMPTY_ROOT(&map->pg_pools)) {
525 		struct ceph_pg_pool_info *pi =
526 			rb_entry(rb_first(&map->pg_pools),
527 				 struct ceph_pg_pool_info, node);
528 		__remove_pg_pool(&map->pg_pools, pi);
529 	}
530 	kfree(map->osd_state);
531 	kfree(map->osd_weight);
532 	kfree(map->osd_addr);
533 	kfree(map);
534 }
535 
536 /*
537  * adjust max osd value.  reallocate arrays.
538  */
539 static int osdmap_set_max_osd(struct ceph_osdmap *map, int max)
540 {
541 	u8 *state;
542 	struct ceph_entity_addr *addr;
543 	u32 *weight;
544 
545 	state = kcalloc(max, sizeof(*state), GFP_NOFS);
546 	addr = kcalloc(max, sizeof(*addr), GFP_NOFS);
547 	weight = kcalloc(max, sizeof(*weight), GFP_NOFS);
548 	if (state == NULL || addr == NULL || weight == NULL) {
549 		kfree(state);
550 		kfree(addr);
551 		kfree(weight);
552 		return -ENOMEM;
553 	}
554 
555 	/* copy old? */
556 	if (map->osd_state) {
557 		memcpy(state, map->osd_state, map->max_osd*sizeof(*state));
558 		memcpy(addr, map->osd_addr, map->max_osd*sizeof(*addr));
559 		memcpy(weight, map->osd_weight, map->max_osd*sizeof(*weight));
560 		kfree(map->osd_state);
561 		kfree(map->osd_addr);
562 		kfree(map->osd_weight);
563 	}
564 
565 	map->osd_state = state;
566 	map->osd_weight = weight;
567 	map->osd_addr = addr;
568 	map->max_osd = max;
569 	return 0;
570 }
571 
572 /*
573  * decode a full map.
574  */
575 struct ceph_osdmap *osdmap_decode(void **p, void *end)
576 {
577 	struct ceph_osdmap *map;
578 	u16 version;
579 	u32 len, max, i;
580 	u8 ev;
581 	int err = -EINVAL;
582 	void *start = *p;
583 	struct ceph_pg_pool_info *pi;
584 
585 	dout("osdmap_decode %p to %p len %d\n", *p, end, (int)(end - *p));
586 
587 	map = kzalloc(sizeof(*map), GFP_NOFS);
588 	if (map == NULL)
589 		return ERR_PTR(-ENOMEM);
590 	map->pg_temp = RB_ROOT;
591 
592 	ceph_decode_16_safe(p, end, version, bad);
593 	if (version > CEPH_OSDMAP_VERSION) {
594 		pr_warning("got unknown v %d > %d of osdmap\n", version,
595 			   CEPH_OSDMAP_VERSION);
596 		goto bad;
597 	}
598 
599 	ceph_decode_need(p, end, 2*sizeof(u64)+6*sizeof(u32), bad);
600 	ceph_decode_copy(p, &map->fsid, sizeof(map->fsid));
601 	map->epoch = ceph_decode_32(p);
602 	ceph_decode_copy(p, &map->created, sizeof(map->created));
603 	ceph_decode_copy(p, &map->modified, sizeof(map->modified));
604 
605 	ceph_decode_32_safe(p, end, max, bad);
606 	while (max--) {
607 		ceph_decode_need(p, end, 4 + 1 + sizeof(pi->v), bad);
608 		pi = kzalloc(sizeof(*pi), GFP_NOFS);
609 		if (!pi)
610 			goto bad;
611 		pi->id = ceph_decode_32(p);
612 		ev = ceph_decode_8(p); /* encoding version */
613 		if (ev > CEPH_PG_POOL_VERSION) {
614 			pr_warning("got unknown v %d > %d of ceph_pg_pool\n",
615 				   ev, CEPH_PG_POOL_VERSION);
616 			kfree(pi);
617 			goto bad;
618 		}
619 		err = __decode_pool(p, end, pi);
620 		if (err < 0) {
621 			kfree(pi);
622 			goto bad;
623 		}
624 		__insert_pg_pool(&map->pg_pools, pi);
625 	}
626 
627 	if (version >= 5 && __decode_pool_names(p, end, map) < 0)
628 		goto bad;
629 
630 	ceph_decode_32_safe(p, end, map->pool_max, bad);
631 
632 	ceph_decode_32_safe(p, end, map->flags, bad);
633 
634 	max = ceph_decode_32(p);
635 
636 	/* (re)alloc osd arrays */
637 	err = osdmap_set_max_osd(map, max);
638 	if (err < 0)
639 		goto bad;
640 	dout("osdmap_decode max_osd = %d\n", map->max_osd);
641 
642 	/* osds */
643 	err = -EINVAL;
644 	ceph_decode_need(p, end, 3*sizeof(u32) +
645 			 map->max_osd*(1 + sizeof(*map->osd_weight) +
646 				       sizeof(*map->osd_addr)), bad);
647 	*p += 4; /* skip length field (should match max) */
648 	ceph_decode_copy(p, map->osd_state, map->max_osd);
649 
650 	*p += 4; /* skip length field (should match max) */
651 	for (i = 0; i < map->max_osd; i++)
652 		map->osd_weight[i] = ceph_decode_32(p);
653 
654 	*p += 4; /* skip length field (should match max) */
655 	ceph_decode_copy(p, map->osd_addr, map->max_osd*sizeof(*map->osd_addr));
656 	for (i = 0; i < map->max_osd; i++)
657 		ceph_decode_addr(&map->osd_addr[i]);
658 
659 	/* pg_temp */
660 	ceph_decode_32_safe(p, end, len, bad);
661 	for (i = 0; i < len; i++) {
662 		int n, j;
663 		struct ceph_pg pgid;
664 		struct ceph_pg_mapping *pg;
665 
666 		ceph_decode_need(p, end, sizeof(u32) + sizeof(u64), bad);
667 		ceph_decode_copy(p, &pgid, sizeof(pgid));
668 		n = ceph_decode_32(p);
669 		ceph_decode_need(p, end, n * sizeof(u32), bad);
670 		err = -ENOMEM;
671 		pg = kmalloc(sizeof(*pg) + n*sizeof(u32), GFP_NOFS);
672 		if (!pg)
673 			goto bad;
674 		pg->pgid = pgid;
675 		pg->len = n;
676 		for (j = 0; j < n; j++)
677 			pg->osds[j] = ceph_decode_32(p);
678 
679 		err = __insert_pg_mapping(pg, &map->pg_temp);
680 		if (err)
681 			goto bad;
682 		dout(" added pg_temp %llx len %d\n", *(u64 *)&pgid, len);
683 	}
684 
685 	/* crush */
686 	ceph_decode_32_safe(p, end, len, bad);
687 	dout("osdmap_decode crush len %d from off 0x%x\n", len,
688 	     (int)(*p - start));
689 	ceph_decode_need(p, end, len, bad);
690 	map->crush = crush_decode(*p, end);
691 	*p += len;
692 	if (IS_ERR(map->crush)) {
693 		err = PTR_ERR(map->crush);
694 		map->crush = NULL;
695 		goto bad;
696 	}
697 
698 	/* ignore the rest of the map */
699 	*p = end;
700 
701 	dout("osdmap_decode done %p %p\n", *p, end);
702 	return map;
703 
704 bad:
705 	dout("osdmap_decode fail\n");
706 	ceph_osdmap_destroy(map);
707 	return ERR_PTR(err);
708 }
709 
710 /*
711  * decode and apply an incremental map update.
712  */
713 struct ceph_osdmap *osdmap_apply_incremental(void **p, void *end,
714 					     struct ceph_osdmap *map,
715 					     struct ceph_messenger *msgr)
716 {
717 	struct crush_map *newcrush = NULL;
718 	struct ceph_fsid fsid;
719 	u32 epoch = 0;
720 	struct ceph_timespec modified;
721 	u32 len, pool;
722 	__s32 new_pool_max, new_flags, max;
723 	void *start = *p;
724 	int err = -EINVAL;
725 	u16 version;
726 
727 	ceph_decode_16_safe(p, end, version, bad);
728 	if (version > CEPH_OSDMAP_INC_VERSION) {
729 		pr_warning("got unknown v %d > %d of inc osdmap\n", version,
730 			   CEPH_OSDMAP_INC_VERSION);
731 		goto bad;
732 	}
733 
734 	ceph_decode_need(p, end, sizeof(fsid)+sizeof(modified)+2*sizeof(u32),
735 			 bad);
736 	ceph_decode_copy(p, &fsid, sizeof(fsid));
737 	epoch = ceph_decode_32(p);
738 	BUG_ON(epoch != map->epoch+1);
739 	ceph_decode_copy(p, &modified, sizeof(modified));
740 	new_pool_max = ceph_decode_32(p);
741 	new_flags = ceph_decode_32(p);
742 
743 	/* full map? */
744 	ceph_decode_32_safe(p, end, len, bad);
745 	if (len > 0) {
746 		dout("apply_incremental full map len %d, %p to %p\n",
747 		     len, *p, end);
748 		return osdmap_decode(p, min(*p+len, end));
749 	}
750 
751 	/* new crush? */
752 	ceph_decode_32_safe(p, end, len, bad);
753 	if (len > 0) {
754 		dout("apply_incremental new crush map len %d, %p to %p\n",
755 		     len, *p, end);
756 		newcrush = crush_decode(*p, min(*p+len, end));
757 		if (IS_ERR(newcrush))
758 			return ERR_CAST(newcrush);
759 		*p += len;
760 	}
761 
762 	/* new flags? */
763 	if (new_flags >= 0)
764 		map->flags = new_flags;
765 	if (new_pool_max >= 0)
766 		map->pool_max = new_pool_max;
767 
768 	ceph_decode_need(p, end, 5*sizeof(u32), bad);
769 
770 	/* new max? */
771 	max = ceph_decode_32(p);
772 	if (max >= 0) {
773 		err = osdmap_set_max_osd(map, max);
774 		if (err < 0)
775 			goto bad;
776 	}
777 
778 	map->epoch++;
779 	map->modified = modified;
780 	if (newcrush) {
781 		if (map->crush)
782 			crush_destroy(map->crush);
783 		map->crush = newcrush;
784 		newcrush = NULL;
785 	}
786 
787 	/* new_pool */
788 	ceph_decode_32_safe(p, end, len, bad);
789 	while (len--) {
790 		__u8 ev;
791 		struct ceph_pg_pool_info *pi;
792 
793 		ceph_decode_32_safe(p, end, pool, bad);
794 		ceph_decode_need(p, end, 1 + sizeof(pi->v), bad);
795 		ev = ceph_decode_8(p);  /* encoding version */
796 		if (ev > CEPH_PG_POOL_VERSION) {
797 			pr_warning("got unknown v %d > %d of ceph_pg_pool\n",
798 				   ev, CEPH_PG_POOL_VERSION);
799 			goto bad;
800 		}
801 		pi = __lookup_pg_pool(&map->pg_pools, pool);
802 		if (!pi) {
803 			pi = kzalloc(sizeof(*pi), GFP_NOFS);
804 			if (!pi) {
805 				err = -ENOMEM;
806 				goto bad;
807 			}
808 			pi->id = pool;
809 			__insert_pg_pool(&map->pg_pools, pi);
810 		}
811 		err = __decode_pool(p, end, pi);
812 		if (err < 0)
813 			goto bad;
814 	}
815 	if (version >= 5 && __decode_pool_names(p, end, map) < 0)
816 		goto bad;
817 
818 	/* old_pool */
819 	ceph_decode_32_safe(p, end, len, bad);
820 	while (len--) {
821 		struct ceph_pg_pool_info *pi;
822 
823 		ceph_decode_32_safe(p, end, pool, bad);
824 		pi = __lookup_pg_pool(&map->pg_pools, pool);
825 		if (pi)
826 			__remove_pg_pool(&map->pg_pools, pi);
827 	}
828 
829 	/* new_up */
830 	err = -EINVAL;
831 	ceph_decode_32_safe(p, end, len, bad);
832 	while (len--) {
833 		u32 osd;
834 		struct ceph_entity_addr addr;
835 		ceph_decode_32_safe(p, end, osd, bad);
836 		ceph_decode_copy_safe(p, end, &addr, sizeof(addr), bad);
837 		ceph_decode_addr(&addr);
838 		pr_info("osd%d up\n", osd);
839 		BUG_ON(osd >= map->max_osd);
840 		map->osd_state[osd] |= CEPH_OSD_UP;
841 		map->osd_addr[osd] = addr;
842 	}
843 
844 	/* new_state */
845 	ceph_decode_32_safe(p, end, len, bad);
846 	while (len--) {
847 		u32 osd;
848 		u8 xorstate;
849 		ceph_decode_32_safe(p, end, osd, bad);
850 		xorstate = **(u8 **)p;
851 		(*p)++;  /* clean flag */
852 		if (xorstate == 0)
853 			xorstate = CEPH_OSD_UP;
854 		if (xorstate & CEPH_OSD_UP)
855 			pr_info("osd%d down\n", osd);
856 		if (osd < map->max_osd)
857 			map->osd_state[osd] ^= xorstate;
858 	}
859 
860 	/* new_weight */
861 	ceph_decode_32_safe(p, end, len, bad);
862 	while (len--) {
863 		u32 osd, off;
864 		ceph_decode_need(p, end, sizeof(u32)*2, bad);
865 		osd = ceph_decode_32(p);
866 		off = ceph_decode_32(p);
867 		pr_info("osd%d weight 0x%x %s\n", osd, off,
868 		     off == CEPH_OSD_IN ? "(in)" :
869 		     (off == CEPH_OSD_OUT ? "(out)" : ""));
870 		if (osd < map->max_osd)
871 			map->osd_weight[osd] = off;
872 	}
873 
874 	/* new_pg_temp */
875 	ceph_decode_32_safe(p, end, len, bad);
876 	while (len--) {
877 		struct ceph_pg_mapping *pg;
878 		int j;
879 		struct ceph_pg pgid;
880 		u32 pglen;
881 		ceph_decode_need(p, end, sizeof(u64) + sizeof(u32), bad);
882 		ceph_decode_copy(p, &pgid, sizeof(pgid));
883 		pglen = ceph_decode_32(p);
884 
885 		if (pglen) {
886 			ceph_decode_need(p, end, pglen*sizeof(u32), bad);
887 
888 			/* removing existing (if any) */
889 			(void) __remove_pg_mapping(&map->pg_temp, pgid);
890 
891 			/* insert */
892 			pg = kmalloc(sizeof(*pg) + sizeof(u32)*pglen, GFP_NOFS);
893 			if (!pg) {
894 				err = -ENOMEM;
895 				goto bad;
896 			}
897 			pg->pgid = pgid;
898 			pg->len = pglen;
899 			for (j = 0; j < pglen; j++)
900 				pg->osds[j] = ceph_decode_32(p);
901 			err = __insert_pg_mapping(pg, &map->pg_temp);
902 			if (err) {
903 				kfree(pg);
904 				goto bad;
905 			}
906 			dout(" added pg_temp %llx len %d\n", *(u64 *)&pgid,
907 			     pglen);
908 		} else {
909 			/* remove */
910 			__remove_pg_mapping(&map->pg_temp, pgid);
911 		}
912 	}
913 
914 	/* ignore the rest */
915 	*p = end;
916 	return map;
917 
918 bad:
919 	pr_err("corrupt inc osdmap epoch %d off %d (%p of %p-%p)\n",
920 	       epoch, (int)(*p - start), *p, start, end);
921 	print_hex_dump(KERN_DEBUG, "osdmap: ",
922 		       DUMP_PREFIX_OFFSET, 16, 1,
923 		       start, end - start, true);
924 	if (newcrush)
925 		crush_destroy(newcrush);
926 	return ERR_PTR(err);
927 }
928 
929 
930 
931 
932 /*
933  * calculate file layout from given offset, length.
934  * fill in correct oid, logical length, and object extent
935  * offset, length.
936  *
937  * for now, we write only a single su, until we can
938  * pass a stride back to the caller.
939  */
940 void ceph_calc_file_object_mapping(struct ceph_file_layout *layout,
941 				   u64 off, u64 *plen,
942 				   u64 *ono,
943 				   u64 *oxoff, u64 *oxlen)
944 {
945 	u32 osize = le32_to_cpu(layout->fl_object_size);
946 	u32 su = le32_to_cpu(layout->fl_stripe_unit);
947 	u32 sc = le32_to_cpu(layout->fl_stripe_count);
948 	u32 bl, stripeno, stripepos, objsetno;
949 	u32 su_per_object;
950 	u64 t, su_offset;
951 
952 	dout("mapping %llu~%llu  osize %u fl_su %u\n", off, *plen,
953 	     osize, su);
954 	su_per_object = osize / su;
955 	dout("osize %u / su %u = su_per_object %u\n", osize, su,
956 	     su_per_object);
957 
958 	BUG_ON((su & ~PAGE_MASK) != 0);
959 	/* bl = *off / su; */
960 	t = off;
961 	do_div(t, su);
962 	bl = t;
963 	dout("off %llu / su %u = bl %u\n", off, su, bl);
964 
965 	stripeno = bl / sc;
966 	stripepos = bl % sc;
967 	objsetno = stripeno / su_per_object;
968 
969 	*ono = objsetno * sc + stripepos;
970 	dout("objset %u * sc %u = ono %u\n", objsetno, sc, (unsigned int)*ono);
971 
972 	/* *oxoff = *off % layout->fl_stripe_unit;  # offset in su */
973 	t = off;
974 	su_offset = do_div(t, su);
975 	*oxoff = su_offset + (stripeno % su_per_object) * su;
976 
977 	/*
978 	 * Calculate the length of the extent being written to the selected
979 	 * object. This is the minimum of the full length requested (plen) or
980 	 * the remainder of the current stripe being written to.
981 	 */
982 	*oxlen = min_t(u64, *plen, su - su_offset);
983 	*plen = *oxlen;
984 
985 	dout(" obj extent %llu~%llu\n", *oxoff, *oxlen);
986 }
987 EXPORT_SYMBOL(ceph_calc_file_object_mapping);
988 
989 /*
990  * calculate an object layout (i.e. pgid) from an oid,
991  * file_layout, and osdmap
992  */
993 int ceph_calc_object_layout(struct ceph_object_layout *ol,
994 			    const char *oid,
995 			    struct ceph_file_layout *fl,
996 			    struct ceph_osdmap *osdmap)
997 {
998 	unsigned int num, num_mask;
999 	struct ceph_pg pgid;
1000 	int poolid = le32_to_cpu(fl->fl_pg_pool);
1001 	struct ceph_pg_pool_info *pool;
1002 	unsigned int ps;
1003 
1004 	BUG_ON(!osdmap);
1005 
1006 	pool = __lookup_pg_pool(&osdmap->pg_pools, poolid);
1007 	if (!pool)
1008 		return -EIO;
1009 	ps = ceph_str_hash(pool->v.object_hash, oid, strlen(oid));
1010 	num = le32_to_cpu(pool->v.pg_num);
1011 	num_mask = pool->pg_num_mask;
1012 
1013 	pgid.ps = cpu_to_le16(ps);
1014 	pgid.preferred = cpu_to_le16(-1);
1015 	pgid.pool = fl->fl_pg_pool;
1016 	dout("calc_object_layout '%s' pgid %d.%x\n", oid, poolid, ps);
1017 
1018 	ol->ol_pgid = pgid;
1019 	ol->ol_stripe_unit = fl->fl_object_stripe_unit;
1020 	return 0;
1021 }
1022 EXPORT_SYMBOL(ceph_calc_object_layout);
1023 
1024 /*
1025  * Calculate raw osd vector for the given pgid.  Return pointer to osd
1026  * array, or NULL on failure.
1027  */
1028 static int *calc_pg_raw(struct ceph_osdmap *osdmap, struct ceph_pg pgid,
1029 			int *osds, int *num)
1030 {
1031 	struct ceph_pg_mapping *pg;
1032 	struct ceph_pg_pool_info *pool;
1033 	int ruleno;
1034 	unsigned int poolid, ps, pps, t, r;
1035 
1036 	poolid = le32_to_cpu(pgid.pool);
1037 	ps = le16_to_cpu(pgid.ps);
1038 
1039 	pool = __lookup_pg_pool(&osdmap->pg_pools, poolid);
1040 	if (!pool)
1041 		return NULL;
1042 
1043 	/* pg_temp? */
1044 	t = ceph_stable_mod(ps, le32_to_cpu(pool->v.pg_num),
1045 			    pool->pgp_num_mask);
1046 	pgid.ps = cpu_to_le16(t);
1047 	pg = __lookup_pg_mapping(&osdmap->pg_temp, pgid);
1048 	if (pg) {
1049 		*num = pg->len;
1050 		return pg->osds;
1051 	}
1052 
1053 	/* crush */
1054 	ruleno = crush_find_rule(osdmap->crush, pool->v.crush_ruleset,
1055 				 pool->v.type, pool->v.size);
1056 	if (ruleno < 0) {
1057 		pr_err("no crush rule pool %d ruleset %d type %d size %d\n",
1058 		       poolid, pool->v.crush_ruleset, pool->v.type,
1059 		       pool->v.size);
1060 		return NULL;
1061 	}
1062 
1063 	pps = ceph_stable_mod(ps,
1064 			      le32_to_cpu(pool->v.pgp_num),
1065 			      pool->pgp_num_mask);
1066 	pps += poolid;
1067 	r = crush_do_rule(osdmap->crush, ruleno, pps, osds,
1068 			  min_t(int, pool->v.size, *num),
1069 			  osdmap->osd_weight);
1070 	if (r < 0) {
1071 		pr_err("error %d from crush rule: pool %d ruleset %d type %d"
1072 		       " size %d\n", r, poolid, pool->v.crush_ruleset,
1073 		       pool->v.type, pool->v.size);
1074 		return NULL;
1075 	}
1076 	*num = r;
1077 	return osds;
1078 }
1079 
1080 /*
1081  * Return acting set for given pgid.
1082  */
1083 int ceph_calc_pg_acting(struct ceph_osdmap *osdmap, struct ceph_pg pgid,
1084 			int *acting)
1085 {
1086 	int rawosds[CEPH_PG_MAX_SIZE], *osds;
1087 	int i, o, num = CEPH_PG_MAX_SIZE;
1088 
1089 	osds = calc_pg_raw(osdmap, pgid, rawosds, &num);
1090 	if (!osds)
1091 		return -1;
1092 
1093 	/* primary is first up osd */
1094 	o = 0;
1095 	for (i = 0; i < num; i++)
1096 		if (ceph_osd_is_up(osdmap, osds[i]))
1097 			acting[o++] = osds[i];
1098 	return o;
1099 }
1100 
1101 /*
1102  * Return primary osd for given pgid, or -1 if none.
1103  */
1104 int ceph_calc_pg_primary(struct ceph_osdmap *osdmap, struct ceph_pg pgid)
1105 {
1106 	int rawosds[CEPH_PG_MAX_SIZE], *osds;
1107 	int i, num = CEPH_PG_MAX_SIZE;
1108 
1109 	osds = calc_pg_raw(osdmap, pgid, rawosds, &num);
1110 	if (!osds)
1111 		return -1;
1112 
1113 	/* primary is first up osd */
1114 	for (i = 0; i < num; i++)
1115 		if (ceph_osd_is_up(osdmap, osds[i]))
1116 			return osds[i];
1117 	return -1;
1118 }
1119 EXPORT_SYMBOL(ceph_calc_pg_primary);
1120