xref: /openbmc/linux/net/ceph/osdmap.c (revision 7490ca1e)
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 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)magic, (unsigned)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->device_parents = kcalloc(c->max_devices, sizeof(u32), GFP_NOFS);
165 	if (c->device_parents == NULL)
166 		goto badmem;
167 	c->bucket_parents = kcalloc(c->max_buckets, sizeof(u32), GFP_NOFS);
168 	if (c->bucket_parents == NULL)
169 		goto badmem;
170 
171 	c->buckets = kcalloc(c->max_buckets, sizeof(*c->buckets), GFP_NOFS);
172 	if (c->buckets == NULL)
173 		goto badmem;
174 	c->rules = kcalloc(c->max_rules, sizeof(*c->rules), GFP_NOFS);
175 	if (c->rules == NULL)
176 		goto badmem;
177 
178 	/* buckets */
179 	for (i = 0; i < c->max_buckets; i++) {
180 		int size = 0;
181 		u32 alg;
182 		struct crush_bucket *b;
183 
184 		ceph_decode_32_safe(p, end, alg, bad);
185 		if (alg == 0) {
186 			c->buckets[i] = NULL;
187 			continue;
188 		}
189 		dout("crush_decode bucket %d off %x %p to %p\n",
190 		     i, (int)(*p-start), *p, end);
191 
192 		switch (alg) {
193 		case CRUSH_BUCKET_UNIFORM:
194 			size = sizeof(struct crush_bucket_uniform);
195 			break;
196 		case CRUSH_BUCKET_LIST:
197 			size = sizeof(struct crush_bucket_list);
198 			break;
199 		case CRUSH_BUCKET_TREE:
200 			size = sizeof(struct crush_bucket_tree);
201 			break;
202 		case CRUSH_BUCKET_STRAW:
203 			size = sizeof(struct crush_bucket_straw);
204 			break;
205 		default:
206 			err = -EINVAL;
207 			goto bad;
208 		}
209 		BUG_ON(size == 0);
210 		b = c->buckets[i] = kzalloc(size, GFP_NOFS);
211 		if (b == NULL)
212 			goto badmem;
213 
214 		ceph_decode_need(p, end, 4*sizeof(u32), bad);
215 		b->id = ceph_decode_32(p);
216 		b->type = ceph_decode_16(p);
217 		b->alg = ceph_decode_8(p);
218 		b->hash = ceph_decode_8(p);
219 		b->weight = ceph_decode_32(p);
220 		b->size = ceph_decode_32(p);
221 
222 		dout("crush_decode bucket size %d off %x %p to %p\n",
223 		     b->size, (int)(*p-start), *p, end);
224 
225 		b->items = kcalloc(b->size, sizeof(__s32), GFP_NOFS);
226 		if (b->items == NULL)
227 			goto badmem;
228 		b->perm = kcalloc(b->size, sizeof(u32), GFP_NOFS);
229 		if (b->perm == NULL)
230 			goto badmem;
231 		b->perm_n = 0;
232 
233 		ceph_decode_need(p, end, b->size*sizeof(u32), bad);
234 		for (j = 0; j < b->size; j++)
235 			b->items[j] = ceph_decode_32(p);
236 
237 		switch (b->alg) {
238 		case CRUSH_BUCKET_UNIFORM:
239 			err = crush_decode_uniform_bucket(p, end,
240 				  (struct crush_bucket_uniform *)b);
241 			if (err < 0)
242 				goto bad;
243 			break;
244 		case CRUSH_BUCKET_LIST:
245 			err = crush_decode_list_bucket(p, end,
246 			       (struct crush_bucket_list *)b);
247 			if (err < 0)
248 				goto bad;
249 			break;
250 		case CRUSH_BUCKET_TREE:
251 			err = crush_decode_tree_bucket(p, end,
252 				(struct crush_bucket_tree *)b);
253 			if (err < 0)
254 				goto bad;
255 			break;
256 		case CRUSH_BUCKET_STRAW:
257 			err = crush_decode_straw_bucket(p, end,
258 				(struct crush_bucket_straw *)b);
259 			if (err < 0)
260 				goto bad;
261 			break;
262 		}
263 	}
264 
265 	/* rules */
266 	dout("rule vec is %p\n", c->rules);
267 	for (i = 0; i < c->max_rules; i++) {
268 		u32 yes;
269 		struct crush_rule *r;
270 
271 		ceph_decode_32_safe(p, end, yes, bad);
272 		if (!yes) {
273 			dout("crush_decode NO rule %d off %x %p to %p\n",
274 			     i, (int)(*p-start), *p, end);
275 			c->rules[i] = NULL;
276 			continue;
277 		}
278 
279 		dout("crush_decode rule %d off %x %p to %p\n",
280 		     i, (int)(*p-start), *p, end);
281 
282 		/* len */
283 		ceph_decode_32_safe(p, end, yes, bad);
284 #if BITS_PER_LONG == 32
285 		err = -EINVAL;
286 		if (yes > ULONG_MAX / sizeof(struct crush_rule_step))
287 			goto bad;
288 #endif
289 		r = c->rules[i] = kmalloc(sizeof(*r) +
290 					  yes*sizeof(struct crush_rule_step),
291 					  GFP_NOFS);
292 		if (r == NULL)
293 			goto badmem;
294 		dout(" rule %d is at %p\n", i, r);
295 		r->len = yes;
296 		ceph_decode_copy_safe(p, end, &r->mask, 4, bad); /* 4 u8's */
297 		ceph_decode_need(p, end, r->len*3*sizeof(u32), bad);
298 		for (j = 0; j < r->len; j++) {
299 			r->steps[j].op = ceph_decode_32(p);
300 			r->steps[j].arg1 = ceph_decode_32(p);
301 			r->steps[j].arg2 = ceph_decode_32(p);
302 		}
303 	}
304 
305 	/* ignore trailing name maps. */
306 
307 	dout("crush_decode success\n");
308 	return c;
309 
310 badmem:
311 	err = -ENOMEM;
312 bad:
313 	dout("crush_decode fail %d\n", err);
314 	crush_destroy(c);
315 	return ERR_PTR(err);
316 }
317 
318 /*
319  * rbtree of pg_mapping for handling pg_temp (explicit mapping of pgid
320  * to a set of osds)
321  */
322 static int pgid_cmp(struct ceph_pg l, struct ceph_pg r)
323 {
324 	u64 a = *(u64 *)&l;
325 	u64 b = *(u64 *)&r;
326 
327 	if (a < b)
328 		return -1;
329 	if (a > b)
330 		return 1;
331 	return 0;
332 }
333 
334 static int __insert_pg_mapping(struct ceph_pg_mapping *new,
335 			       struct rb_root *root)
336 {
337 	struct rb_node **p = &root->rb_node;
338 	struct rb_node *parent = NULL;
339 	struct ceph_pg_mapping *pg = NULL;
340 	int c;
341 
342 	dout("__insert_pg_mapping %llx %p\n", *(u64 *)&new->pgid, new);
343 	while (*p) {
344 		parent = *p;
345 		pg = rb_entry(parent, struct ceph_pg_mapping, node);
346 		c = pgid_cmp(new->pgid, pg->pgid);
347 		if (c < 0)
348 			p = &(*p)->rb_left;
349 		else if (c > 0)
350 			p = &(*p)->rb_right;
351 		else
352 			return -EEXIST;
353 	}
354 
355 	rb_link_node(&new->node, parent, p);
356 	rb_insert_color(&new->node, root);
357 	return 0;
358 }
359 
360 static struct ceph_pg_mapping *__lookup_pg_mapping(struct rb_root *root,
361 						   struct ceph_pg pgid)
362 {
363 	struct rb_node *n = root->rb_node;
364 	struct ceph_pg_mapping *pg;
365 	int c;
366 
367 	while (n) {
368 		pg = rb_entry(n, struct ceph_pg_mapping, node);
369 		c = pgid_cmp(pgid, pg->pgid);
370 		if (c < 0) {
371 			n = n->rb_left;
372 		} else if (c > 0) {
373 			n = n->rb_right;
374 		} else {
375 			dout("__lookup_pg_mapping %llx got %p\n",
376 			     *(u64 *)&pgid, pg);
377 			return pg;
378 		}
379 	}
380 	return NULL;
381 }
382 
383 static int __remove_pg_mapping(struct rb_root *root, struct ceph_pg pgid)
384 {
385 	struct ceph_pg_mapping *pg = __lookup_pg_mapping(root, pgid);
386 
387 	if (pg) {
388 		dout("__remove_pg_mapping %llx %p\n", *(u64 *)&pgid, pg);
389 		rb_erase(&pg->node, root);
390 		kfree(pg);
391 		return 0;
392 	}
393 	dout("__remove_pg_mapping %llx dne\n", *(u64 *)&pgid);
394 	return -ENOENT;
395 }
396 
397 /*
398  * rbtree of pg pool info
399  */
400 static int __insert_pg_pool(struct rb_root *root, struct ceph_pg_pool_info *new)
401 {
402 	struct rb_node **p = &root->rb_node;
403 	struct rb_node *parent = NULL;
404 	struct ceph_pg_pool_info *pi = NULL;
405 
406 	while (*p) {
407 		parent = *p;
408 		pi = rb_entry(parent, struct ceph_pg_pool_info, node);
409 		if (new->id < pi->id)
410 			p = &(*p)->rb_left;
411 		else if (new->id > pi->id)
412 			p = &(*p)->rb_right;
413 		else
414 			return -EEXIST;
415 	}
416 
417 	rb_link_node(&new->node, parent, p);
418 	rb_insert_color(&new->node, root);
419 	return 0;
420 }
421 
422 static struct ceph_pg_pool_info *__lookup_pg_pool(struct rb_root *root, int id)
423 {
424 	struct ceph_pg_pool_info *pi;
425 	struct rb_node *n = root->rb_node;
426 
427 	while (n) {
428 		pi = rb_entry(n, struct ceph_pg_pool_info, node);
429 		if (id < pi->id)
430 			n = n->rb_left;
431 		else if (id > pi->id)
432 			n = n->rb_right;
433 		else
434 			return pi;
435 	}
436 	return NULL;
437 }
438 
439 int ceph_pg_poolid_by_name(struct ceph_osdmap *map, const char *name)
440 {
441 	struct rb_node *rbp;
442 
443 	for (rbp = rb_first(&map->pg_pools); rbp; rbp = rb_next(rbp)) {
444 		struct ceph_pg_pool_info *pi =
445 			rb_entry(rbp, struct ceph_pg_pool_info, node);
446 		if (pi->name && strcmp(pi->name, name) == 0)
447 			return pi->id;
448 	}
449 	return -ENOENT;
450 }
451 EXPORT_SYMBOL(ceph_pg_poolid_by_name);
452 
453 static void __remove_pg_pool(struct rb_root *root, struct ceph_pg_pool_info *pi)
454 {
455 	rb_erase(&pi->node, root);
456 	kfree(pi->name);
457 	kfree(pi);
458 }
459 
460 static int __decode_pool(void **p, void *end, struct ceph_pg_pool_info *pi)
461 {
462 	unsigned n, m;
463 
464 	ceph_decode_copy(p, &pi->v, sizeof(pi->v));
465 	calc_pg_masks(pi);
466 
467 	/* num_snaps * snap_info_t */
468 	n = le32_to_cpu(pi->v.num_snaps);
469 	while (n--) {
470 		ceph_decode_need(p, end, sizeof(u64) + 1 + sizeof(u64) +
471 				 sizeof(struct ceph_timespec), bad);
472 		*p += sizeof(u64) +       /* key */
473 			1 + sizeof(u64) + /* u8, snapid */
474 			sizeof(struct ceph_timespec);
475 		m = ceph_decode_32(p);    /* snap name */
476 		*p += m;
477 	}
478 
479 	*p += le32_to_cpu(pi->v.num_removed_snap_intervals) * sizeof(u64) * 2;
480 	return 0;
481 
482 bad:
483 	return -EINVAL;
484 }
485 
486 static int __decode_pool_names(void **p, void *end, struct ceph_osdmap *map)
487 {
488 	struct ceph_pg_pool_info *pi;
489 	u32 num, len, pool;
490 
491 	ceph_decode_32_safe(p, end, num, bad);
492 	dout(" %d pool names\n", num);
493 	while (num--) {
494 		ceph_decode_32_safe(p, end, pool, bad);
495 		ceph_decode_32_safe(p, end, len, bad);
496 		dout("  pool %d len %d\n", pool, len);
497 		pi = __lookup_pg_pool(&map->pg_pools, pool);
498 		if (pi) {
499 			kfree(pi->name);
500 			pi->name = kmalloc(len + 1, GFP_NOFS);
501 			if (pi->name) {
502 				memcpy(pi->name, *p, len);
503 				pi->name[len] = '\0';
504 				dout("  name is %s\n", pi->name);
505 			}
506 		}
507 		*p += len;
508 	}
509 	return 0;
510 
511 bad:
512 	return -EINVAL;
513 }
514 
515 /*
516  * osd map
517  */
518 void ceph_osdmap_destroy(struct ceph_osdmap *map)
519 {
520 	dout("osdmap_destroy %p\n", map);
521 	if (map->crush)
522 		crush_destroy(map->crush);
523 	while (!RB_EMPTY_ROOT(&map->pg_temp)) {
524 		struct ceph_pg_mapping *pg =
525 			rb_entry(rb_first(&map->pg_temp),
526 				 struct ceph_pg_mapping, node);
527 		rb_erase(&pg->node, &map->pg_temp);
528 		kfree(pg);
529 	}
530 	while (!RB_EMPTY_ROOT(&map->pg_pools)) {
531 		struct ceph_pg_pool_info *pi =
532 			rb_entry(rb_first(&map->pg_pools),
533 				 struct ceph_pg_pool_info, node);
534 		__remove_pg_pool(&map->pg_pools, pi);
535 	}
536 	kfree(map->osd_state);
537 	kfree(map->osd_weight);
538 	kfree(map->osd_addr);
539 	kfree(map);
540 }
541 
542 /*
543  * adjust max osd value.  reallocate arrays.
544  */
545 static int osdmap_set_max_osd(struct ceph_osdmap *map, int max)
546 {
547 	u8 *state;
548 	struct ceph_entity_addr *addr;
549 	u32 *weight;
550 
551 	state = kcalloc(max, sizeof(*state), GFP_NOFS);
552 	addr = kcalloc(max, sizeof(*addr), GFP_NOFS);
553 	weight = kcalloc(max, sizeof(*weight), GFP_NOFS);
554 	if (state == NULL || addr == NULL || weight == NULL) {
555 		kfree(state);
556 		kfree(addr);
557 		kfree(weight);
558 		return -ENOMEM;
559 	}
560 
561 	/* copy old? */
562 	if (map->osd_state) {
563 		memcpy(state, map->osd_state, map->max_osd*sizeof(*state));
564 		memcpy(addr, map->osd_addr, map->max_osd*sizeof(*addr));
565 		memcpy(weight, map->osd_weight, map->max_osd*sizeof(*weight));
566 		kfree(map->osd_state);
567 		kfree(map->osd_addr);
568 		kfree(map->osd_weight);
569 	}
570 
571 	map->osd_state = state;
572 	map->osd_weight = weight;
573 	map->osd_addr = addr;
574 	map->max_osd = max;
575 	return 0;
576 }
577 
578 /*
579  * decode a full map.
580  */
581 struct ceph_osdmap *osdmap_decode(void **p, void *end)
582 {
583 	struct ceph_osdmap *map;
584 	u16 version;
585 	u32 len, max, i;
586 	u8 ev;
587 	int err = -EINVAL;
588 	void *start = *p;
589 	struct ceph_pg_pool_info *pi;
590 
591 	dout("osdmap_decode %p to %p len %d\n", *p, end, (int)(end - *p));
592 
593 	map = kzalloc(sizeof(*map), GFP_NOFS);
594 	if (map == NULL)
595 		return ERR_PTR(-ENOMEM);
596 	map->pg_temp = RB_ROOT;
597 
598 	ceph_decode_16_safe(p, end, version, bad);
599 	if (version > CEPH_OSDMAP_VERSION) {
600 		pr_warning("got unknown v %d > %d of osdmap\n", version,
601 			   CEPH_OSDMAP_VERSION);
602 		goto bad;
603 	}
604 
605 	ceph_decode_need(p, end, 2*sizeof(u64)+6*sizeof(u32), bad);
606 	ceph_decode_copy(p, &map->fsid, sizeof(map->fsid));
607 	map->epoch = ceph_decode_32(p);
608 	ceph_decode_copy(p, &map->created, sizeof(map->created));
609 	ceph_decode_copy(p, &map->modified, sizeof(map->modified));
610 
611 	ceph_decode_32_safe(p, end, max, bad);
612 	while (max--) {
613 		ceph_decode_need(p, end, 4 + 1 + sizeof(pi->v), bad);
614 		pi = kzalloc(sizeof(*pi), GFP_NOFS);
615 		if (!pi)
616 			goto bad;
617 		pi->id = ceph_decode_32(p);
618 		ev = ceph_decode_8(p); /* encoding version */
619 		if (ev > CEPH_PG_POOL_VERSION) {
620 			pr_warning("got unknown v %d > %d of ceph_pg_pool\n",
621 				   ev, CEPH_PG_POOL_VERSION);
622 			kfree(pi);
623 			goto bad;
624 		}
625 		err = __decode_pool(p, end, pi);
626 		if (err < 0) {
627 			kfree(pi);
628 			goto bad;
629 		}
630 		__insert_pg_pool(&map->pg_pools, pi);
631 	}
632 
633 	if (version >= 5 && __decode_pool_names(p, end, map) < 0)
634 		goto bad;
635 
636 	ceph_decode_32_safe(p, end, map->pool_max, bad);
637 
638 	ceph_decode_32_safe(p, end, map->flags, bad);
639 
640 	max = ceph_decode_32(p);
641 
642 	/* (re)alloc osd arrays */
643 	err = osdmap_set_max_osd(map, max);
644 	if (err < 0)
645 		goto bad;
646 	dout("osdmap_decode max_osd = %d\n", map->max_osd);
647 
648 	/* osds */
649 	err = -EINVAL;
650 	ceph_decode_need(p, end, 3*sizeof(u32) +
651 			 map->max_osd*(1 + sizeof(*map->osd_weight) +
652 				       sizeof(*map->osd_addr)), bad);
653 	*p += 4; /* skip length field (should match max) */
654 	ceph_decode_copy(p, map->osd_state, map->max_osd);
655 
656 	*p += 4; /* skip length field (should match max) */
657 	for (i = 0; i < map->max_osd; i++)
658 		map->osd_weight[i] = ceph_decode_32(p);
659 
660 	*p += 4; /* skip length field (should match max) */
661 	ceph_decode_copy(p, map->osd_addr, map->max_osd*sizeof(*map->osd_addr));
662 	for (i = 0; i < map->max_osd; i++)
663 		ceph_decode_addr(&map->osd_addr[i]);
664 
665 	/* pg_temp */
666 	ceph_decode_32_safe(p, end, len, bad);
667 	for (i = 0; i < len; i++) {
668 		int n, j;
669 		struct ceph_pg pgid;
670 		struct ceph_pg_mapping *pg;
671 
672 		ceph_decode_need(p, end, sizeof(u32) + sizeof(u64), bad);
673 		ceph_decode_copy(p, &pgid, sizeof(pgid));
674 		n = ceph_decode_32(p);
675 		ceph_decode_need(p, end, n * sizeof(u32), bad);
676 		err = -ENOMEM;
677 		pg = kmalloc(sizeof(*pg) + n*sizeof(u32), GFP_NOFS);
678 		if (!pg)
679 			goto bad;
680 		pg->pgid = pgid;
681 		pg->len = n;
682 		for (j = 0; j < n; j++)
683 			pg->osds[j] = ceph_decode_32(p);
684 
685 		err = __insert_pg_mapping(pg, &map->pg_temp);
686 		if (err)
687 			goto bad;
688 		dout(" added pg_temp %llx len %d\n", *(u64 *)&pgid, len);
689 	}
690 
691 	/* crush */
692 	ceph_decode_32_safe(p, end, len, bad);
693 	dout("osdmap_decode crush len %d from off 0x%x\n", len,
694 	     (int)(*p - start));
695 	ceph_decode_need(p, end, len, bad);
696 	map->crush = crush_decode(*p, end);
697 	*p += len;
698 	if (IS_ERR(map->crush)) {
699 		err = PTR_ERR(map->crush);
700 		map->crush = NULL;
701 		goto bad;
702 	}
703 
704 	/* ignore the rest of the map */
705 	*p = end;
706 
707 	dout("osdmap_decode done %p %p\n", *p, end);
708 	return map;
709 
710 bad:
711 	dout("osdmap_decode fail\n");
712 	ceph_osdmap_destroy(map);
713 	return ERR_PTR(err);
714 }
715 
716 /*
717  * decode and apply an incremental map update.
718  */
719 struct ceph_osdmap *osdmap_apply_incremental(void **p, void *end,
720 					     struct ceph_osdmap *map,
721 					     struct ceph_messenger *msgr)
722 {
723 	struct crush_map *newcrush = NULL;
724 	struct ceph_fsid fsid;
725 	u32 epoch = 0;
726 	struct ceph_timespec modified;
727 	u32 len, pool;
728 	__s32 new_pool_max, new_flags, max;
729 	void *start = *p;
730 	int err = -EINVAL;
731 	u16 version;
732 
733 	ceph_decode_16_safe(p, end, version, bad);
734 	if (version > CEPH_OSDMAP_INC_VERSION) {
735 		pr_warning("got unknown v %d > %d of inc osdmap\n", version,
736 			   CEPH_OSDMAP_INC_VERSION);
737 		goto bad;
738 	}
739 
740 	ceph_decode_need(p, end, sizeof(fsid)+sizeof(modified)+2*sizeof(u32),
741 			 bad);
742 	ceph_decode_copy(p, &fsid, sizeof(fsid));
743 	epoch = ceph_decode_32(p);
744 	BUG_ON(epoch != map->epoch+1);
745 	ceph_decode_copy(p, &modified, sizeof(modified));
746 	new_pool_max = ceph_decode_32(p);
747 	new_flags = ceph_decode_32(p);
748 
749 	/* full map? */
750 	ceph_decode_32_safe(p, end, len, bad);
751 	if (len > 0) {
752 		dout("apply_incremental full map len %d, %p to %p\n",
753 		     len, *p, end);
754 		return osdmap_decode(p, min(*p+len, end));
755 	}
756 
757 	/* new crush? */
758 	ceph_decode_32_safe(p, end, len, bad);
759 	if (len > 0) {
760 		dout("apply_incremental new crush map len %d, %p to %p\n",
761 		     len, *p, end);
762 		newcrush = crush_decode(*p, min(*p+len, end));
763 		if (IS_ERR(newcrush))
764 			return ERR_CAST(newcrush);
765 		*p += len;
766 	}
767 
768 	/* new flags? */
769 	if (new_flags >= 0)
770 		map->flags = new_flags;
771 	if (new_pool_max >= 0)
772 		map->pool_max = new_pool_max;
773 
774 	ceph_decode_need(p, end, 5*sizeof(u32), bad);
775 
776 	/* new max? */
777 	max = ceph_decode_32(p);
778 	if (max >= 0) {
779 		err = osdmap_set_max_osd(map, max);
780 		if (err < 0)
781 			goto bad;
782 	}
783 
784 	map->epoch++;
785 	map->modified = modified;
786 	if (newcrush) {
787 		if (map->crush)
788 			crush_destroy(map->crush);
789 		map->crush = newcrush;
790 		newcrush = NULL;
791 	}
792 
793 	/* new_pool */
794 	ceph_decode_32_safe(p, end, len, bad);
795 	while (len--) {
796 		__u8 ev;
797 		struct ceph_pg_pool_info *pi;
798 
799 		ceph_decode_32_safe(p, end, pool, bad);
800 		ceph_decode_need(p, end, 1 + sizeof(pi->v), bad);
801 		ev = ceph_decode_8(p);  /* encoding version */
802 		if (ev > CEPH_PG_POOL_VERSION) {
803 			pr_warning("got unknown v %d > %d of ceph_pg_pool\n",
804 				   ev, CEPH_PG_POOL_VERSION);
805 			goto bad;
806 		}
807 		pi = __lookup_pg_pool(&map->pg_pools, pool);
808 		if (!pi) {
809 			pi = kzalloc(sizeof(*pi), GFP_NOFS);
810 			if (!pi) {
811 				err = -ENOMEM;
812 				goto bad;
813 			}
814 			pi->id = pool;
815 			__insert_pg_pool(&map->pg_pools, pi);
816 		}
817 		err = __decode_pool(p, end, pi);
818 		if (err < 0)
819 			goto bad;
820 	}
821 	if (version >= 5 && __decode_pool_names(p, end, map) < 0)
822 		goto bad;
823 
824 	/* old_pool */
825 	ceph_decode_32_safe(p, end, len, bad);
826 	while (len--) {
827 		struct ceph_pg_pool_info *pi;
828 
829 		ceph_decode_32_safe(p, end, pool, bad);
830 		pi = __lookup_pg_pool(&map->pg_pools, pool);
831 		if (pi)
832 			__remove_pg_pool(&map->pg_pools, pi);
833 	}
834 
835 	/* new_up */
836 	err = -EINVAL;
837 	ceph_decode_32_safe(p, end, len, bad);
838 	while (len--) {
839 		u32 osd;
840 		struct ceph_entity_addr addr;
841 		ceph_decode_32_safe(p, end, osd, bad);
842 		ceph_decode_copy_safe(p, end, &addr, sizeof(addr), bad);
843 		ceph_decode_addr(&addr);
844 		pr_info("osd%d up\n", osd);
845 		BUG_ON(osd >= map->max_osd);
846 		map->osd_state[osd] |= CEPH_OSD_UP;
847 		map->osd_addr[osd] = addr;
848 	}
849 
850 	/* new_state */
851 	ceph_decode_32_safe(p, end, len, bad);
852 	while (len--) {
853 		u32 osd;
854 		u8 xorstate;
855 		ceph_decode_32_safe(p, end, osd, bad);
856 		xorstate = **(u8 **)p;
857 		(*p)++;  /* clean flag */
858 		if (xorstate == 0)
859 			xorstate = CEPH_OSD_UP;
860 		if (xorstate & CEPH_OSD_UP)
861 			pr_info("osd%d down\n", osd);
862 		if (osd < map->max_osd)
863 			map->osd_state[osd] ^= xorstate;
864 	}
865 
866 	/* new_weight */
867 	ceph_decode_32_safe(p, end, len, bad);
868 	while (len--) {
869 		u32 osd, off;
870 		ceph_decode_need(p, end, sizeof(u32)*2, bad);
871 		osd = ceph_decode_32(p);
872 		off = ceph_decode_32(p);
873 		pr_info("osd%d weight 0x%x %s\n", osd, off,
874 		     off == CEPH_OSD_IN ? "(in)" :
875 		     (off == CEPH_OSD_OUT ? "(out)" : ""));
876 		if (osd < map->max_osd)
877 			map->osd_weight[osd] = off;
878 	}
879 
880 	/* new_pg_temp */
881 	ceph_decode_32_safe(p, end, len, bad);
882 	while (len--) {
883 		struct ceph_pg_mapping *pg;
884 		int j;
885 		struct ceph_pg pgid;
886 		u32 pglen;
887 		ceph_decode_need(p, end, sizeof(u64) + sizeof(u32), bad);
888 		ceph_decode_copy(p, &pgid, sizeof(pgid));
889 		pglen = ceph_decode_32(p);
890 
891 		if (pglen) {
892 			/* insert */
893 			ceph_decode_need(p, end, pglen*sizeof(u32), bad);
894 			pg = kmalloc(sizeof(*pg) + sizeof(u32)*pglen, GFP_NOFS);
895 			if (!pg) {
896 				err = -ENOMEM;
897 				goto bad;
898 			}
899 			pg->pgid = pgid;
900 			pg->len = pglen;
901 			for (j = 0; j < pglen; j++)
902 				pg->osds[j] = ceph_decode_32(p);
903 			err = __insert_pg_mapping(pg, &map->pg_temp);
904 			if (err) {
905 				kfree(pg);
906 				goto bad;
907 			}
908 			dout(" added pg_temp %llx len %d\n", *(u64 *)&pgid,
909 			     pglen);
910 		} else {
911 			/* remove */
912 			__remove_pg_mapping(&map->pg_temp, pgid);
913 		}
914 	}
915 
916 	/* ignore the rest */
917 	*p = end;
918 	return map;
919 
920 bad:
921 	pr_err("corrupt inc osdmap epoch %d off %d (%p of %p-%p)\n",
922 	       epoch, (int)(*p - start), *p, start, end);
923 	print_hex_dump(KERN_DEBUG, "osdmap: ",
924 		       DUMP_PREFIX_OFFSET, 16, 1,
925 		       start, end - start, true);
926 	if (newcrush)
927 		crush_destroy(newcrush);
928 	return ERR_PTR(err);
929 }
930 
931 
932 
933 
934 /*
935  * calculate file layout from given offset, length.
936  * fill in correct oid, logical length, and object extent
937  * offset, length.
938  *
939  * for now, we write only a single su, until we can
940  * pass a stride back to the caller.
941  */
942 void ceph_calc_file_object_mapping(struct ceph_file_layout *layout,
943 				   u64 off, u64 *plen,
944 				   u64 *ono,
945 				   u64 *oxoff, u64 *oxlen)
946 {
947 	u32 osize = le32_to_cpu(layout->fl_object_size);
948 	u32 su = le32_to_cpu(layout->fl_stripe_unit);
949 	u32 sc = le32_to_cpu(layout->fl_stripe_count);
950 	u32 bl, stripeno, stripepos, objsetno;
951 	u32 su_per_object;
952 	u64 t, su_offset;
953 
954 	dout("mapping %llu~%llu  osize %u fl_su %u\n", off, *plen,
955 	     osize, su);
956 	su_per_object = osize / su;
957 	dout("osize %u / su %u = su_per_object %u\n", osize, su,
958 	     su_per_object);
959 
960 	BUG_ON((su & ~PAGE_MASK) != 0);
961 	/* bl = *off / su; */
962 	t = off;
963 	do_div(t, su);
964 	bl = t;
965 	dout("off %llu / su %u = bl %u\n", off, su, bl);
966 
967 	stripeno = bl / sc;
968 	stripepos = bl % sc;
969 	objsetno = stripeno / su_per_object;
970 
971 	*ono = objsetno * sc + stripepos;
972 	dout("objset %u * sc %u = ono %u\n", objsetno, sc, (unsigned)*ono);
973 
974 	/* *oxoff = *off % layout->fl_stripe_unit;  # offset in su */
975 	t = off;
976 	su_offset = do_div(t, su);
977 	*oxoff = su_offset + (stripeno % su_per_object) * su;
978 
979 	/*
980 	 * Calculate the length of the extent being written to the selected
981 	 * object. This is the minimum of the full length requested (plen) or
982 	 * the remainder of the current stripe being written to.
983 	 */
984 	*oxlen = min_t(u64, *plen, su - su_offset);
985 	*plen = *oxlen;
986 
987 	dout(" obj extent %llu~%llu\n", *oxoff, *oxlen);
988 }
989 EXPORT_SYMBOL(ceph_calc_file_object_mapping);
990 
991 /*
992  * calculate an object layout (i.e. pgid) from an oid,
993  * file_layout, and osdmap
994  */
995 int ceph_calc_object_layout(struct ceph_object_layout *ol,
996 			    const char *oid,
997 			    struct ceph_file_layout *fl,
998 			    struct ceph_osdmap *osdmap)
999 {
1000 	unsigned num, num_mask;
1001 	struct ceph_pg pgid;
1002 	s32 preferred = (s32)le32_to_cpu(fl->fl_pg_preferred);
1003 	int poolid = le32_to_cpu(fl->fl_pg_pool);
1004 	struct ceph_pg_pool_info *pool;
1005 	unsigned ps;
1006 
1007 	BUG_ON(!osdmap);
1008 
1009 	pool = __lookup_pg_pool(&osdmap->pg_pools, poolid);
1010 	if (!pool)
1011 		return -EIO;
1012 	ps = ceph_str_hash(pool->v.object_hash, oid, strlen(oid));
1013 	if (preferred >= 0) {
1014 		ps += preferred;
1015 		num = le32_to_cpu(pool->v.lpg_num);
1016 		num_mask = pool->lpg_num_mask;
1017 	} else {
1018 		num = le32_to_cpu(pool->v.pg_num);
1019 		num_mask = pool->pg_num_mask;
1020 	}
1021 
1022 	pgid.ps = cpu_to_le16(ps);
1023 	pgid.preferred = cpu_to_le16(preferred);
1024 	pgid.pool = fl->fl_pg_pool;
1025 	if (preferred >= 0)
1026 		dout("calc_object_layout '%s' pgid %d.%xp%d\n", oid, poolid, ps,
1027 		     (int)preferred);
1028 	else
1029 		dout("calc_object_layout '%s' pgid %d.%x\n", oid, poolid, ps);
1030 
1031 	ol->ol_pgid = pgid;
1032 	ol->ol_stripe_unit = fl->fl_object_stripe_unit;
1033 	return 0;
1034 }
1035 EXPORT_SYMBOL(ceph_calc_object_layout);
1036 
1037 /*
1038  * Calculate raw osd vector for the given pgid.  Return pointer to osd
1039  * array, or NULL on failure.
1040  */
1041 static int *calc_pg_raw(struct ceph_osdmap *osdmap, struct ceph_pg pgid,
1042 			int *osds, int *num)
1043 {
1044 	struct ceph_pg_mapping *pg;
1045 	struct ceph_pg_pool_info *pool;
1046 	int ruleno;
1047 	unsigned poolid, ps, pps, t;
1048 	int preferred;
1049 
1050 	poolid = le32_to_cpu(pgid.pool);
1051 	ps = le16_to_cpu(pgid.ps);
1052 	preferred = (s16)le16_to_cpu(pgid.preferred);
1053 
1054 	pool = __lookup_pg_pool(&osdmap->pg_pools, poolid);
1055 	if (!pool)
1056 		return NULL;
1057 
1058 	/* pg_temp? */
1059 	if (preferred >= 0)
1060 		t = ceph_stable_mod(ps, le32_to_cpu(pool->v.lpg_num),
1061 				    pool->lpgp_num_mask);
1062 	else
1063 		t = ceph_stable_mod(ps, le32_to_cpu(pool->v.pg_num),
1064 				    pool->pgp_num_mask);
1065 	pgid.ps = cpu_to_le16(t);
1066 	pg = __lookup_pg_mapping(&osdmap->pg_temp, pgid);
1067 	if (pg) {
1068 		*num = pg->len;
1069 		return pg->osds;
1070 	}
1071 
1072 	/* crush */
1073 	ruleno = crush_find_rule(osdmap->crush, pool->v.crush_ruleset,
1074 				 pool->v.type, pool->v.size);
1075 	if (ruleno < 0) {
1076 		pr_err("no crush rule pool %d ruleset %d type %d size %d\n",
1077 		       poolid, pool->v.crush_ruleset, pool->v.type,
1078 		       pool->v.size);
1079 		return NULL;
1080 	}
1081 
1082 	/* don't forcefeed bad device ids to crush */
1083 	if (preferred >= osdmap->max_osd ||
1084 	    preferred >= osdmap->crush->max_devices)
1085 		preferred = -1;
1086 
1087 	if (preferred >= 0)
1088 		pps = ceph_stable_mod(ps,
1089 				      le32_to_cpu(pool->v.lpgp_num),
1090 				      pool->lpgp_num_mask);
1091 	else
1092 		pps = ceph_stable_mod(ps,
1093 				      le32_to_cpu(pool->v.pgp_num),
1094 				      pool->pgp_num_mask);
1095 	pps += poolid;
1096 	*num = crush_do_rule(osdmap->crush, ruleno, pps, osds,
1097 			     min_t(int, pool->v.size, *num),
1098 			     preferred, osdmap->osd_weight);
1099 	return osds;
1100 }
1101 
1102 /*
1103  * Return acting set for given pgid.
1104  */
1105 int ceph_calc_pg_acting(struct ceph_osdmap *osdmap, struct ceph_pg pgid,
1106 			int *acting)
1107 {
1108 	int rawosds[CEPH_PG_MAX_SIZE], *osds;
1109 	int i, o, num = CEPH_PG_MAX_SIZE;
1110 
1111 	osds = calc_pg_raw(osdmap, pgid, rawosds, &num);
1112 	if (!osds)
1113 		return -1;
1114 
1115 	/* primary is first up osd */
1116 	o = 0;
1117 	for (i = 0; i < num; i++)
1118 		if (ceph_osd_is_up(osdmap, osds[i]))
1119 			acting[o++] = osds[i];
1120 	return o;
1121 }
1122 
1123 /*
1124  * Return primary osd for given pgid, or -1 if none.
1125  */
1126 int ceph_calc_pg_primary(struct ceph_osdmap *osdmap, struct ceph_pg pgid)
1127 {
1128 	int rawosds[CEPH_PG_MAX_SIZE], *osds;
1129 	int i, num = CEPH_PG_MAX_SIZE;
1130 
1131 	osds = calc_pg_raw(osdmap, pgid, rawosds, &num);
1132 	if (!osds)
1133 		return -1;
1134 
1135 	/* primary is first up osd */
1136 	for (i = 0; i < num; i++)
1137 		if (ceph_osd_is_up(osdmap, osds[i]))
1138 			return osds[i];
1139 	return -1;
1140 }
1141 EXPORT_SYMBOL(ceph_calc_pg_primary);
1142