xref: /linux/net/ceph/osdmap.c (revision 288440de9e5fdb4a3ff73864850f080c1250fc81)
1 // SPDX-License-Identifier: GPL-2.0
2 
3 #include <linux/ceph/ceph_debug.h>
4 
5 #include <linux/module.h>
6 #include <linux/slab.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 static __printf(2, 3)
15 void osdmap_info(const struct ceph_osdmap *map, const char *fmt, ...)
16 {
17 	struct va_format vaf;
18 	va_list args;
19 
20 	va_start(args, fmt);
21 	vaf.fmt = fmt;
22 	vaf.va = &args;
23 
24 	printk(KERN_INFO "%s (%pU e%u): %pV", KBUILD_MODNAME, &map->fsid,
25 	       map->epoch, &vaf);
26 
27 	va_end(args);
28 }
29 
30 char *ceph_osdmap_state_str(char *str, int len, u32 state)
31 {
32 	if (!len)
33 		return str;
34 
35 	if ((state & CEPH_OSD_EXISTS) && (state & CEPH_OSD_UP))
36 		snprintf(str, len, "exists, up");
37 	else if (state & CEPH_OSD_EXISTS)
38 		snprintf(str, len, "exists");
39 	else if (state & CEPH_OSD_UP)
40 		snprintf(str, len, "up");
41 	else
42 		snprintf(str, len, "doesn't exist");
43 
44 	return str;
45 }
46 
47 /* maps */
48 
49 static int calc_bits_of(unsigned int t)
50 {
51 	int b = 0;
52 	while (t) {
53 		t = t >> 1;
54 		b++;
55 	}
56 	return b;
57 }
58 
59 /*
60  * the foo_mask is the smallest value 2^n-1 that is >= foo.
61  */
62 static void calc_pg_masks(struct ceph_pg_pool_info *pi)
63 {
64 	pi->pg_num_mask = (1 << calc_bits_of(pi->pg_num-1)) - 1;
65 	pi->pgp_num_mask = (1 << calc_bits_of(pi->pgp_num-1)) - 1;
66 }
67 
68 /*
69  * decode crush map
70  */
71 static int crush_decode_uniform_bucket(void **p, void *end,
72 				       struct crush_bucket_uniform *b)
73 {
74 	dout("crush_decode_uniform_bucket %p to %p\n", *p, end);
75 	ceph_decode_need(p, end, (1+b->h.size) * sizeof(u32), bad);
76 	b->item_weight = ceph_decode_32(p);
77 	return 0;
78 bad:
79 	return -EINVAL;
80 }
81 
82 static int crush_decode_list_bucket(void **p, void *end,
83 				    struct crush_bucket_list *b)
84 {
85 	int j;
86 	dout("crush_decode_list_bucket %p to %p\n", *p, end);
87 	b->item_weights = kcalloc(b->h.size, sizeof(u32), GFP_NOFS);
88 	if (b->item_weights == NULL)
89 		return -ENOMEM;
90 	b->sum_weights = kcalloc(b->h.size, sizeof(u32), GFP_NOFS);
91 	if (b->sum_weights == NULL)
92 		return -ENOMEM;
93 	ceph_decode_need(p, end, 2 * b->h.size * sizeof(u32), bad);
94 	for (j = 0; j < b->h.size; j++) {
95 		b->item_weights[j] = ceph_decode_32(p);
96 		b->sum_weights[j] = ceph_decode_32(p);
97 	}
98 	return 0;
99 bad:
100 	return -EINVAL;
101 }
102 
103 static int crush_decode_tree_bucket(void **p, void *end,
104 				    struct crush_bucket_tree *b)
105 {
106 	int j;
107 	dout("crush_decode_tree_bucket %p to %p\n", *p, end);
108 	ceph_decode_8_safe(p, end, b->num_nodes, bad);
109 	b->node_weights = kcalloc(b->num_nodes, sizeof(u32), GFP_NOFS);
110 	if (b->node_weights == NULL)
111 		return -ENOMEM;
112 	ceph_decode_need(p, end, b->num_nodes * sizeof(u32), bad);
113 	for (j = 0; j < b->num_nodes; j++)
114 		b->node_weights[j] = ceph_decode_32(p);
115 	return 0;
116 bad:
117 	return -EINVAL;
118 }
119 
120 static int crush_decode_straw_bucket(void **p, void *end,
121 				     struct crush_bucket_straw *b)
122 {
123 	int j;
124 	dout("crush_decode_straw_bucket %p to %p\n", *p, end);
125 	b->item_weights = kcalloc(b->h.size, sizeof(u32), GFP_NOFS);
126 	if (b->item_weights == NULL)
127 		return -ENOMEM;
128 	b->straws = kcalloc(b->h.size, sizeof(u32), GFP_NOFS);
129 	if (b->straws == NULL)
130 		return -ENOMEM;
131 	ceph_decode_need(p, end, 2 * b->h.size * sizeof(u32), bad);
132 	for (j = 0; j < b->h.size; j++) {
133 		b->item_weights[j] = ceph_decode_32(p);
134 		b->straws[j] = ceph_decode_32(p);
135 	}
136 	return 0;
137 bad:
138 	return -EINVAL;
139 }
140 
141 static int crush_decode_straw2_bucket(void **p, void *end,
142 				      struct crush_bucket_straw2 *b)
143 {
144 	int j;
145 	dout("crush_decode_straw2_bucket %p to %p\n", *p, end);
146 	b->item_weights = kcalloc(b->h.size, sizeof(u32), GFP_NOFS);
147 	if (b->item_weights == NULL)
148 		return -ENOMEM;
149 	ceph_decode_need(p, end, b->h.size * sizeof(u32), bad);
150 	for (j = 0; j < b->h.size; j++)
151 		b->item_weights[j] = ceph_decode_32(p);
152 	return 0;
153 bad:
154 	return -EINVAL;
155 }
156 
157 struct crush_name_node {
158 	struct rb_node cn_node;
159 	int cn_id;
160 	char cn_name[];
161 };
162 
163 static struct crush_name_node *alloc_crush_name(size_t name_len)
164 {
165 	struct crush_name_node *cn;
166 
167 	cn = kmalloc(sizeof(*cn) + name_len + 1, GFP_NOIO);
168 	if (!cn)
169 		return NULL;
170 
171 	RB_CLEAR_NODE(&cn->cn_node);
172 	return cn;
173 }
174 
175 static void free_crush_name(struct crush_name_node *cn)
176 {
177 	WARN_ON(!RB_EMPTY_NODE(&cn->cn_node));
178 
179 	kfree(cn);
180 }
181 
182 DEFINE_RB_FUNCS(crush_name, struct crush_name_node, cn_id, cn_node)
183 
184 static int decode_crush_names(void **p, void *end, struct rb_root *root)
185 {
186 	u32 n;
187 
188 	ceph_decode_32_safe(p, end, n, e_inval);
189 	while (n--) {
190 		struct crush_name_node *cn;
191 		int id;
192 		u32 name_len;
193 
194 		ceph_decode_32_safe(p, end, id, e_inval);
195 		ceph_decode_32_safe(p, end, name_len, e_inval);
196 		ceph_decode_need(p, end, name_len, e_inval);
197 
198 		cn = alloc_crush_name(name_len);
199 		if (!cn)
200 			return -ENOMEM;
201 
202 		cn->cn_id = id;
203 		memcpy(cn->cn_name, *p, name_len);
204 		cn->cn_name[name_len] = '\0';
205 		*p += name_len;
206 
207 		if (!__insert_crush_name(root, cn)) {
208 			free_crush_name(cn);
209 			return -EEXIST;
210 		}
211 	}
212 
213 	return 0;
214 
215 e_inval:
216 	return -EINVAL;
217 }
218 
219 void clear_crush_names(struct rb_root *root)
220 {
221 	while (!RB_EMPTY_ROOT(root)) {
222 		struct crush_name_node *cn =
223 		    rb_entry(rb_first(root), struct crush_name_node, cn_node);
224 
225 		erase_crush_name(root, cn);
226 		free_crush_name(cn);
227 	}
228 }
229 
230 static struct crush_choose_arg_map *alloc_choose_arg_map(void)
231 {
232 	struct crush_choose_arg_map *arg_map;
233 
234 	arg_map = kzalloc(sizeof(*arg_map), GFP_NOIO);
235 	if (!arg_map)
236 		return NULL;
237 
238 	RB_CLEAR_NODE(&arg_map->node);
239 	return arg_map;
240 }
241 
242 static void free_choose_arg_map(struct crush_choose_arg_map *arg_map)
243 {
244 	if (arg_map) {
245 		int i, j;
246 
247 		WARN_ON(!RB_EMPTY_NODE(&arg_map->node));
248 
249 		for (i = 0; i < arg_map->size; i++) {
250 			struct crush_choose_arg *arg = &arg_map->args[i];
251 
252 			for (j = 0; j < arg->weight_set_size; j++)
253 				kfree(arg->weight_set[j].weights);
254 			kfree(arg->weight_set);
255 			kfree(arg->ids);
256 		}
257 		kfree(arg_map->args);
258 		kfree(arg_map);
259 	}
260 }
261 
262 DEFINE_RB_FUNCS(choose_arg_map, struct crush_choose_arg_map, choose_args_index,
263 		node);
264 
265 void clear_choose_args(struct crush_map *c)
266 {
267 	while (!RB_EMPTY_ROOT(&c->choose_args)) {
268 		struct crush_choose_arg_map *arg_map =
269 		    rb_entry(rb_first(&c->choose_args),
270 			     struct crush_choose_arg_map, node);
271 
272 		erase_choose_arg_map(&c->choose_args, arg_map);
273 		free_choose_arg_map(arg_map);
274 	}
275 }
276 
277 static u32 *decode_array_32_alloc(void **p, void *end, u32 *plen)
278 {
279 	u32 *a = NULL;
280 	u32 len;
281 	int ret;
282 
283 	ceph_decode_32_safe(p, end, len, e_inval);
284 	if (len) {
285 		u32 i;
286 
287 		a = kmalloc_array(len, sizeof(u32), GFP_NOIO);
288 		if (!a) {
289 			ret = -ENOMEM;
290 			goto fail;
291 		}
292 
293 		ceph_decode_need(p, end, len * sizeof(u32), e_inval);
294 		for (i = 0; i < len; i++)
295 			a[i] = ceph_decode_32(p);
296 	}
297 
298 	*plen = len;
299 	return a;
300 
301 e_inval:
302 	ret = -EINVAL;
303 fail:
304 	kfree(a);
305 	return ERR_PTR(ret);
306 }
307 
308 /*
309  * Assumes @arg is zero-initialized.
310  */
311 static int decode_choose_arg(void **p, void *end, struct crush_choose_arg *arg)
312 {
313 	int ret;
314 
315 	ceph_decode_32_safe(p, end, arg->weight_set_size, e_inval);
316 	if (arg->weight_set_size) {
317 		u32 i;
318 
319 		arg->weight_set = kmalloc_array(arg->weight_set_size,
320 						sizeof(*arg->weight_set),
321 						GFP_NOIO);
322 		if (!arg->weight_set)
323 			return -ENOMEM;
324 
325 		for (i = 0; i < arg->weight_set_size; i++) {
326 			struct crush_weight_set *w = &arg->weight_set[i];
327 
328 			w->weights = decode_array_32_alloc(p, end, &w->size);
329 			if (IS_ERR(w->weights)) {
330 				ret = PTR_ERR(w->weights);
331 				w->weights = NULL;
332 				return ret;
333 			}
334 		}
335 	}
336 
337 	arg->ids = decode_array_32_alloc(p, end, &arg->ids_size);
338 	if (IS_ERR(arg->ids)) {
339 		ret = PTR_ERR(arg->ids);
340 		arg->ids = NULL;
341 		return ret;
342 	}
343 
344 	return 0;
345 
346 e_inval:
347 	return -EINVAL;
348 }
349 
350 static int decode_choose_args(void **p, void *end, struct crush_map *c)
351 {
352 	struct crush_choose_arg_map *arg_map = NULL;
353 	u32 num_choose_arg_maps, num_buckets;
354 	int ret;
355 
356 	ceph_decode_32_safe(p, end, num_choose_arg_maps, e_inval);
357 	while (num_choose_arg_maps--) {
358 		arg_map = alloc_choose_arg_map();
359 		if (!arg_map) {
360 			ret = -ENOMEM;
361 			goto fail;
362 		}
363 
364 		ceph_decode_64_safe(p, end, arg_map->choose_args_index,
365 				    e_inval);
366 		arg_map->size = c->max_buckets;
367 		arg_map->args = kcalloc(arg_map->size, sizeof(*arg_map->args),
368 					GFP_NOIO);
369 		if (!arg_map->args) {
370 			ret = -ENOMEM;
371 			goto fail;
372 		}
373 
374 		ceph_decode_32_safe(p, end, num_buckets, e_inval);
375 		while (num_buckets--) {
376 			struct crush_choose_arg *arg;
377 			u32 bucket_index;
378 
379 			ceph_decode_32_safe(p, end, bucket_index, e_inval);
380 			if (bucket_index >= arg_map->size)
381 				goto e_inval;
382 
383 			arg = &arg_map->args[bucket_index];
384 			ret = decode_choose_arg(p, end, arg);
385 			if (ret)
386 				goto fail;
387 
388 			if (arg->ids_size &&
389 			    arg->ids_size != c->buckets[bucket_index]->size)
390 				goto e_inval;
391 		}
392 
393 		insert_choose_arg_map(&c->choose_args, arg_map);
394 	}
395 
396 	return 0;
397 
398 e_inval:
399 	ret = -EINVAL;
400 fail:
401 	free_choose_arg_map(arg_map);
402 	return ret;
403 }
404 
405 static void crush_finalize(struct crush_map *c)
406 {
407 	__s32 b;
408 
409 	/* Space for the array of pointers to per-bucket workspace */
410 	c->working_size = sizeof(struct crush_work) +
411 	    c->max_buckets * sizeof(struct crush_work_bucket *);
412 
413 	for (b = 0; b < c->max_buckets; b++) {
414 		if (!c->buckets[b])
415 			continue;
416 
417 		switch (c->buckets[b]->alg) {
418 		default:
419 			/*
420 			 * The base case, permutation variables and
421 			 * the pointer to the permutation array.
422 			 */
423 			c->working_size += sizeof(struct crush_work_bucket);
424 			break;
425 		}
426 		/* Every bucket has a permutation array. */
427 		c->working_size += c->buckets[b]->size * sizeof(__u32);
428 	}
429 }
430 
431 static struct crush_map *crush_decode(void *pbyval, void *end)
432 {
433 	struct crush_map *c;
434 	int err;
435 	int i, j;
436 	void **p = &pbyval;
437 	void *start = pbyval;
438 	u32 magic;
439 
440 	dout("crush_decode %p to %p len %d\n", *p, end, (int)(end - *p));
441 
442 	c = kzalloc(sizeof(*c), GFP_NOFS);
443 	if (c == NULL)
444 		return ERR_PTR(-ENOMEM);
445 
446 	c->type_names = RB_ROOT;
447 	c->names = RB_ROOT;
448 	c->choose_args = RB_ROOT;
449 
450         /* set tunables to default values */
451         c->choose_local_tries = 2;
452         c->choose_local_fallback_tries = 5;
453         c->choose_total_tries = 19;
454 	c->chooseleaf_descend_once = 0;
455 
456 	ceph_decode_need(p, end, 4*sizeof(u32), bad);
457 	magic = ceph_decode_32(p);
458 	if (magic != CRUSH_MAGIC) {
459 		pr_err("crush_decode magic %x != current %x\n",
460 		       (unsigned int)magic, (unsigned int)CRUSH_MAGIC);
461 		goto bad;
462 	}
463 	c->max_buckets = ceph_decode_32(p);
464 	c->max_rules = ceph_decode_32(p);
465 	c->max_devices = ceph_decode_32(p);
466 
467 	c->buckets = kcalloc(c->max_buckets, sizeof(*c->buckets), GFP_NOFS);
468 	if (c->buckets == NULL)
469 		goto badmem;
470 	c->rules = kcalloc(c->max_rules, sizeof(*c->rules), GFP_NOFS);
471 	if (c->rules == NULL)
472 		goto badmem;
473 
474 	/* buckets */
475 	for (i = 0; i < c->max_buckets; i++) {
476 		int size = 0;
477 		u32 alg;
478 		struct crush_bucket *b;
479 
480 		ceph_decode_32_safe(p, end, alg, bad);
481 		if (alg == 0) {
482 			c->buckets[i] = NULL;
483 			continue;
484 		}
485 		dout("crush_decode bucket %d off %x %p to %p\n",
486 		     i, (int)(*p-start), *p, end);
487 
488 		switch (alg) {
489 		case CRUSH_BUCKET_UNIFORM:
490 			size = sizeof(struct crush_bucket_uniform);
491 			break;
492 		case CRUSH_BUCKET_LIST:
493 			size = sizeof(struct crush_bucket_list);
494 			break;
495 		case CRUSH_BUCKET_TREE:
496 			size = sizeof(struct crush_bucket_tree);
497 			break;
498 		case CRUSH_BUCKET_STRAW:
499 			size = sizeof(struct crush_bucket_straw);
500 			break;
501 		case CRUSH_BUCKET_STRAW2:
502 			size = sizeof(struct crush_bucket_straw2);
503 			break;
504 		default:
505 			goto bad;
506 		}
507 		BUG_ON(size == 0);
508 		b = c->buckets[i] = kzalloc(size, GFP_NOFS);
509 		if (b == NULL)
510 			goto badmem;
511 
512 		ceph_decode_need(p, end, 4*sizeof(u32), bad);
513 		b->id = ceph_decode_32(p);
514 		b->type = ceph_decode_16(p);
515 		b->alg = ceph_decode_8(p);
516 		b->hash = ceph_decode_8(p);
517 		b->weight = ceph_decode_32(p);
518 		b->size = ceph_decode_32(p);
519 
520 		dout("crush_decode bucket size %d off %x %p to %p\n",
521 		     b->size, (int)(*p-start), *p, end);
522 
523 		b->items = kcalloc(b->size, sizeof(__s32), GFP_NOFS);
524 		if (b->items == NULL)
525 			goto badmem;
526 
527 		ceph_decode_need(p, end, b->size*sizeof(u32), bad);
528 		for (j = 0; j < b->size; j++)
529 			b->items[j] = ceph_decode_32(p);
530 
531 		switch (b->alg) {
532 		case CRUSH_BUCKET_UNIFORM:
533 			err = crush_decode_uniform_bucket(p, end,
534 				  (struct crush_bucket_uniform *)b);
535 			if (err < 0)
536 				goto fail;
537 			break;
538 		case CRUSH_BUCKET_LIST:
539 			err = crush_decode_list_bucket(p, end,
540 			       (struct crush_bucket_list *)b);
541 			if (err < 0)
542 				goto fail;
543 			break;
544 		case CRUSH_BUCKET_TREE:
545 			err = crush_decode_tree_bucket(p, end,
546 				(struct crush_bucket_tree *)b);
547 			if (err < 0)
548 				goto fail;
549 			break;
550 		case CRUSH_BUCKET_STRAW:
551 			err = crush_decode_straw_bucket(p, end,
552 				(struct crush_bucket_straw *)b);
553 			if (err < 0)
554 				goto fail;
555 			break;
556 		case CRUSH_BUCKET_STRAW2:
557 			err = crush_decode_straw2_bucket(p, end,
558 				(struct crush_bucket_straw2 *)b);
559 			if (err < 0)
560 				goto fail;
561 			break;
562 		}
563 	}
564 
565 	/* rules */
566 	dout("rule vec is %p\n", c->rules);
567 	for (i = 0; i < c->max_rules; i++) {
568 		u32 yes;
569 		struct crush_rule *r;
570 
571 		ceph_decode_32_safe(p, end, yes, bad);
572 		if (!yes) {
573 			dout("crush_decode NO rule %d off %x %p to %p\n",
574 			     i, (int)(*p-start), *p, end);
575 			c->rules[i] = NULL;
576 			continue;
577 		}
578 
579 		dout("crush_decode rule %d off %x %p to %p\n",
580 		     i, (int)(*p-start), *p, end);
581 
582 		/* len */
583 		ceph_decode_32_safe(p, end, yes, bad);
584 #if BITS_PER_LONG == 32
585 		if (yes > (ULONG_MAX - sizeof(*r))
586 			  / sizeof(struct crush_rule_step))
587 			goto bad;
588 #endif
589 		r = kmalloc(struct_size(r, steps, yes), GFP_NOFS);
590 		if (r == NULL)
591 			goto badmem;
592 		dout(" rule %d is at %p\n", i, r);
593 		c->rules[i] = r;
594 		r->len = yes;
595 		ceph_decode_copy_safe(p, end, &r->mask, 4, bad); /* 4 u8's */
596 		ceph_decode_need(p, end, r->len*3*sizeof(u32), bad);
597 		for (j = 0; j < r->len; j++) {
598 			r->steps[j].op = ceph_decode_32(p);
599 			r->steps[j].arg1 = ceph_decode_32(p);
600 			r->steps[j].arg2 = ceph_decode_32(p);
601 		}
602 	}
603 
604 	err = decode_crush_names(p, end, &c->type_names);
605 	if (err)
606 		goto fail;
607 
608 	err = decode_crush_names(p, end, &c->names);
609 	if (err)
610 		goto fail;
611 
612 	ceph_decode_skip_map(p, end, 32, string, bad); /* rule_name_map */
613 
614         /* tunables */
615         ceph_decode_need(p, end, 3*sizeof(u32), done);
616         c->choose_local_tries = ceph_decode_32(p);
617         c->choose_local_fallback_tries =  ceph_decode_32(p);
618         c->choose_total_tries = ceph_decode_32(p);
619         dout("crush decode tunable choose_local_tries = %d\n",
620              c->choose_local_tries);
621         dout("crush decode tunable choose_local_fallback_tries = %d\n",
622              c->choose_local_fallback_tries);
623         dout("crush decode tunable choose_total_tries = %d\n",
624              c->choose_total_tries);
625 
626 	ceph_decode_need(p, end, sizeof(u32), done);
627 	c->chooseleaf_descend_once = ceph_decode_32(p);
628 	dout("crush decode tunable chooseleaf_descend_once = %d\n",
629 	     c->chooseleaf_descend_once);
630 
631 	ceph_decode_need(p, end, sizeof(u8), done);
632 	c->chooseleaf_vary_r = ceph_decode_8(p);
633 	dout("crush decode tunable chooseleaf_vary_r = %d\n",
634 	     c->chooseleaf_vary_r);
635 
636 	/* skip straw_calc_version, allowed_bucket_algs */
637 	ceph_decode_need(p, end, sizeof(u8) + sizeof(u32), done);
638 	*p += sizeof(u8) + sizeof(u32);
639 
640 	ceph_decode_need(p, end, sizeof(u8), done);
641 	c->chooseleaf_stable = ceph_decode_8(p);
642 	dout("crush decode tunable chooseleaf_stable = %d\n",
643 	     c->chooseleaf_stable);
644 
645 	if (*p != end) {
646 		/* class_map */
647 		ceph_decode_skip_map(p, end, 32, 32, bad);
648 		/* class_name */
649 		ceph_decode_skip_map(p, end, 32, string, bad);
650 		/* class_bucket */
651 		ceph_decode_skip_map_of_map(p, end, 32, 32, 32, bad);
652 	}
653 
654 	if (*p != end) {
655 		err = decode_choose_args(p, end, c);
656 		if (err)
657 			goto fail;
658 	}
659 
660 done:
661 	crush_finalize(c);
662 	dout("crush_decode success\n");
663 	return c;
664 
665 badmem:
666 	err = -ENOMEM;
667 fail:
668 	dout("crush_decode fail %d\n", err);
669 	crush_destroy(c);
670 	return ERR_PTR(err);
671 
672 bad:
673 	err = -EINVAL;
674 	goto fail;
675 }
676 
677 int ceph_pg_compare(const struct ceph_pg *lhs, const struct ceph_pg *rhs)
678 {
679 	if (lhs->pool < rhs->pool)
680 		return -1;
681 	if (lhs->pool > rhs->pool)
682 		return 1;
683 	if (lhs->seed < rhs->seed)
684 		return -1;
685 	if (lhs->seed > rhs->seed)
686 		return 1;
687 
688 	return 0;
689 }
690 
691 int ceph_spg_compare(const struct ceph_spg *lhs, const struct ceph_spg *rhs)
692 {
693 	int ret;
694 
695 	ret = ceph_pg_compare(&lhs->pgid, &rhs->pgid);
696 	if (ret)
697 		return ret;
698 
699 	if (lhs->shard < rhs->shard)
700 		return -1;
701 	if (lhs->shard > rhs->shard)
702 		return 1;
703 
704 	return 0;
705 }
706 
707 static struct ceph_pg_mapping *alloc_pg_mapping(size_t payload_len)
708 {
709 	struct ceph_pg_mapping *pg;
710 
711 	pg = kmalloc(sizeof(*pg) + payload_len, GFP_NOIO);
712 	if (!pg)
713 		return NULL;
714 
715 	RB_CLEAR_NODE(&pg->node);
716 	return pg;
717 }
718 
719 static void free_pg_mapping(struct ceph_pg_mapping *pg)
720 {
721 	WARN_ON(!RB_EMPTY_NODE(&pg->node));
722 
723 	kfree(pg);
724 }
725 
726 /*
727  * rbtree of pg_mapping for handling pg_temp (explicit mapping of pgid
728  * to a set of osds) and primary_temp (explicit primary setting)
729  */
730 DEFINE_RB_FUNCS2(pg_mapping, struct ceph_pg_mapping, pgid, ceph_pg_compare,
731 		 RB_BYPTR, const struct ceph_pg *, node)
732 
733 /*
734  * rbtree of pg pool info
735  */
736 DEFINE_RB_FUNCS(pg_pool, struct ceph_pg_pool_info, id, node)
737 
738 struct ceph_pg_pool_info *ceph_pg_pool_by_id(struct ceph_osdmap *map, u64 id)
739 {
740 	return lookup_pg_pool(&map->pg_pools, id);
741 }
742 
743 const char *ceph_pg_pool_name_by_id(struct ceph_osdmap *map, u64 id)
744 {
745 	struct ceph_pg_pool_info *pi;
746 
747 	if (id == CEPH_NOPOOL)
748 		return NULL;
749 
750 	if (WARN_ON_ONCE(id > (u64) INT_MAX))
751 		return NULL;
752 
753 	pi = lookup_pg_pool(&map->pg_pools, id);
754 	return pi ? pi->name : NULL;
755 }
756 EXPORT_SYMBOL(ceph_pg_pool_name_by_id);
757 
758 int ceph_pg_poolid_by_name(struct ceph_osdmap *map, const char *name)
759 {
760 	struct rb_node *rbp;
761 
762 	for (rbp = rb_first(&map->pg_pools); rbp; rbp = rb_next(rbp)) {
763 		struct ceph_pg_pool_info *pi =
764 			rb_entry(rbp, struct ceph_pg_pool_info, node);
765 		if (pi->name && strcmp(pi->name, name) == 0)
766 			return pi->id;
767 	}
768 	return -ENOENT;
769 }
770 EXPORT_SYMBOL(ceph_pg_poolid_by_name);
771 
772 u64 ceph_pg_pool_flags(struct ceph_osdmap *map, u64 id)
773 {
774 	struct ceph_pg_pool_info *pi;
775 
776 	pi = lookup_pg_pool(&map->pg_pools, id);
777 	return pi ? pi->flags : 0;
778 }
779 EXPORT_SYMBOL(ceph_pg_pool_flags);
780 
781 static void __remove_pg_pool(struct rb_root *root, struct ceph_pg_pool_info *pi)
782 {
783 	erase_pg_pool(root, pi);
784 	kfree(pi->name);
785 	kfree(pi);
786 }
787 
788 static int decode_pool(void **p, void *end, struct ceph_pg_pool_info *pi)
789 {
790 	u8 ev, cv;
791 	unsigned len, num;
792 	void *pool_end;
793 
794 	ceph_decode_need(p, end, 2 + 4, bad);
795 	ev = ceph_decode_8(p);  /* encoding version */
796 	cv = ceph_decode_8(p); /* compat version */
797 	if (ev < 5) {
798 		pr_warn("got v %d < 5 cv %d of ceph_pg_pool\n", ev, cv);
799 		return -EINVAL;
800 	}
801 	if (cv > 9) {
802 		pr_warn("got v %d cv %d > 9 of ceph_pg_pool\n", ev, cv);
803 		return -EINVAL;
804 	}
805 	len = ceph_decode_32(p);
806 	ceph_decode_need(p, end, len, bad);
807 	pool_end = *p + len;
808 
809 	pi->type = ceph_decode_8(p);
810 	pi->size = ceph_decode_8(p);
811 	pi->crush_ruleset = ceph_decode_8(p);
812 	pi->object_hash = ceph_decode_8(p);
813 
814 	pi->pg_num = ceph_decode_32(p);
815 	pi->pgp_num = ceph_decode_32(p);
816 
817 	*p += 4 + 4;  /* skip lpg* */
818 	*p += 4;      /* skip last_change */
819 	*p += 8 + 4;  /* skip snap_seq, snap_epoch */
820 
821 	/* skip snaps */
822 	num = ceph_decode_32(p);
823 	while (num--) {
824 		*p += 8;  /* snapid key */
825 		*p += 1 + 1; /* versions */
826 		len = ceph_decode_32(p);
827 		*p += len;
828 	}
829 
830 	/* skip removed_snaps */
831 	num = ceph_decode_32(p);
832 	*p += num * (8 + 8);
833 
834 	*p += 8;  /* skip auid */
835 	pi->flags = ceph_decode_64(p);
836 	*p += 4;  /* skip crash_replay_interval */
837 
838 	if (ev >= 7)
839 		pi->min_size = ceph_decode_8(p);
840 	else
841 		pi->min_size = pi->size - pi->size / 2;
842 
843 	if (ev >= 8)
844 		*p += 8 + 8;  /* skip quota_max_* */
845 
846 	if (ev >= 9) {
847 		/* skip tiers */
848 		num = ceph_decode_32(p);
849 		*p += num * 8;
850 
851 		*p += 8;  /* skip tier_of */
852 		*p += 1;  /* skip cache_mode */
853 
854 		pi->read_tier = ceph_decode_64(p);
855 		pi->write_tier = ceph_decode_64(p);
856 	} else {
857 		pi->read_tier = -1;
858 		pi->write_tier = -1;
859 	}
860 
861 	if (ev >= 10) {
862 		/* skip properties */
863 		num = ceph_decode_32(p);
864 		while (num--) {
865 			len = ceph_decode_32(p);
866 			*p += len; /* key */
867 			len = ceph_decode_32(p);
868 			*p += len; /* val */
869 		}
870 	}
871 
872 	if (ev >= 11) {
873 		/* skip hit_set_params */
874 		*p += 1 + 1; /* versions */
875 		len = ceph_decode_32(p);
876 		*p += len;
877 
878 		*p += 4; /* skip hit_set_period */
879 		*p += 4; /* skip hit_set_count */
880 	}
881 
882 	if (ev >= 12)
883 		*p += 4; /* skip stripe_width */
884 
885 	if (ev >= 13) {
886 		*p += 8; /* skip target_max_bytes */
887 		*p += 8; /* skip target_max_objects */
888 		*p += 4; /* skip cache_target_dirty_ratio_micro */
889 		*p += 4; /* skip cache_target_full_ratio_micro */
890 		*p += 4; /* skip cache_min_flush_age */
891 		*p += 4; /* skip cache_min_evict_age */
892 	}
893 
894 	if (ev >=  14) {
895 		/* skip erasure_code_profile */
896 		len = ceph_decode_32(p);
897 		*p += len;
898 	}
899 
900 	/*
901 	 * last_force_op_resend_preluminous, will be overridden if the
902 	 * map was encoded with RESEND_ON_SPLIT
903 	 */
904 	if (ev >= 15)
905 		pi->last_force_request_resend = ceph_decode_32(p);
906 	else
907 		pi->last_force_request_resend = 0;
908 
909 	if (ev >= 16)
910 		*p += 4; /* skip min_read_recency_for_promote */
911 
912 	if (ev >= 17)
913 		*p += 8; /* skip expected_num_objects */
914 
915 	if (ev >= 19)
916 		*p += 4; /* skip cache_target_dirty_high_ratio_micro */
917 
918 	if (ev >= 20)
919 		*p += 4; /* skip min_write_recency_for_promote */
920 
921 	if (ev >= 21)
922 		*p += 1; /* skip use_gmt_hitset */
923 
924 	if (ev >= 22)
925 		*p += 1; /* skip fast_read */
926 
927 	if (ev >= 23) {
928 		*p += 4; /* skip hit_set_grade_decay_rate */
929 		*p += 4; /* skip hit_set_search_last_n */
930 	}
931 
932 	if (ev >= 24) {
933 		/* skip opts */
934 		*p += 1 + 1; /* versions */
935 		len = ceph_decode_32(p);
936 		*p += len;
937 	}
938 
939 	if (ev >= 25)
940 		pi->last_force_request_resend = ceph_decode_32(p);
941 
942 	/* ignore the rest */
943 
944 	*p = pool_end;
945 	calc_pg_masks(pi);
946 	return 0;
947 
948 bad:
949 	return -EINVAL;
950 }
951 
952 static int decode_pool_names(void **p, void *end, struct ceph_osdmap *map)
953 {
954 	struct ceph_pg_pool_info *pi;
955 	u32 num, len;
956 	u64 pool;
957 
958 	ceph_decode_32_safe(p, end, num, bad);
959 	dout(" %d pool names\n", num);
960 	while (num--) {
961 		ceph_decode_64_safe(p, end, pool, bad);
962 		ceph_decode_32_safe(p, end, len, bad);
963 		dout("  pool %llu len %d\n", pool, len);
964 		ceph_decode_need(p, end, len, bad);
965 		pi = lookup_pg_pool(&map->pg_pools, pool);
966 		if (pi) {
967 			char *name = kstrndup(*p, len, GFP_NOFS);
968 
969 			if (!name)
970 				return -ENOMEM;
971 			kfree(pi->name);
972 			pi->name = name;
973 			dout("  name is %s\n", pi->name);
974 		}
975 		*p += len;
976 	}
977 	return 0;
978 
979 bad:
980 	return -EINVAL;
981 }
982 
983 /*
984  * CRUSH workspaces
985  *
986  * workspace_manager framework borrowed from fs/btrfs/compression.c.
987  * Two simplifications: there is only one type of workspace and there
988  * is always at least one workspace.
989  */
990 static struct crush_work *alloc_workspace(const struct crush_map *c)
991 {
992 	struct crush_work *work;
993 	size_t work_size;
994 
995 	WARN_ON(!c->working_size);
996 	work_size = crush_work_size(c, CEPH_PG_MAX_SIZE);
997 	dout("%s work_size %zu bytes\n", __func__, work_size);
998 
999 	work = kvmalloc(work_size, GFP_NOIO);
1000 	if (!work)
1001 		return NULL;
1002 
1003 	INIT_LIST_HEAD(&work->item);
1004 	crush_init_workspace(c, work);
1005 	return work;
1006 }
1007 
1008 static void free_workspace(struct crush_work *work)
1009 {
1010 	WARN_ON(!list_empty(&work->item));
1011 	kvfree(work);
1012 }
1013 
1014 static void init_workspace_manager(struct workspace_manager *wsm)
1015 {
1016 	INIT_LIST_HEAD(&wsm->idle_ws);
1017 	spin_lock_init(&wsm->ws_lock);
1018 	atomic_set(&wsm->total_ws, 0);
1019 	wsm->free_ws = 0;
1020 	init_waitqueue_head(&wsm->ws_wait);
1021 }
1022 
1023 static void add_initial_workspace(struct workspace_manager *wsm,
1024 				  struct crush_work *work)
1025 {
1026 	WARN_ON(!list_empty(&wsm->idle_ws));
1027 
1028 	list_add(&work->item, &wsm->idle_ws);
1029 	atomic_set(&wsm->total_ws, 1);
1030 	wsm->free_ws = 1;
1031 }
1032 
1033 static void cleanup_workspace_manager(struct workspace_manager *wsm)
1034 {
1035 	struct crush_work *work;
1036 
1037 	while (!list_empty(&wsm->idle_ws)) {
1038 		work = list_first_entry(&wsm->idle_ws, struct crush_work,
1039 					item);
1040 		list_del_init(&work->item);
1041 		free_workspace(work);
1042 	}
1043 	atomic_set(&wsm->total_ws, 0);
1044 	wsm->free_ws = 0;
1045 }
1046 
1047 /*
1048  * Finds an available workspace or allocates a new one.  If it's not
1049  * possible to allocate a new one, waits until there is one.
1050  */
1051 static struct crush_work *get_workspace(struct workspace_manager *wsm,
1052 					const struct crush_map *c)
1053 {
1054 	struct crush_work *work;
1055 	int cpus = num_online_cpus();
1056 
1057 again:
1058 	spin_lock(&wsm->ws_lock);
1059 	if (!list_empty(&wsm->idle_ws)) {
1060 		work = list_first_entry(&wsm->idle_ws, struct crush_work,
1061 					item);
1062 		list_del_init(&work->item);
1063 		wsm->free_ws--;
1064 		spin_unlock(&wsm->ws_lock);
1065 		return work;
1066 
1067 	}
1068 	if (atomic_read(&wsm->total_ws) > cpus) {
1069 		DEFINE_WAIT(wait);
1070 
1071 		spin_unlock(&wsm->ws_lock);
1072 		prepare_to_wait(&wsm->ws_wait, &wait, TASK_UNINTERRUPTIBLE);
1073 		if (atomic_read(&wsm->total_ws) > cpus && !wsm->free_ws)
1074 			schedule();
1075 		finish_wait(&wsm->ws_wait, &wait);
1076 		goto again;
1077 	}
1078 	atomic_inc(&wsm->total_ws);
1079 	spin_unlock(&wsm->ws_lock);
1080 
1081 	work = alloc_workspace(c);
1082 	if (!work) {
1083 		atomic_dec(&wsm->total_ws);
1084 		wake_up(&wsm->ws_wait);
1085 
1086 		/*
1087 		 * Do not return the error but go back to waiting.  We
1088 		 * have the initial workspace and the CRUSH computation
1089 		 * time is bounded so we will get it eventually.
1090 		 */
1091 		WARN_ON(atomic_read(&wsm->total_ws) < 1);
1092 		goto again;
1093 	}
1094 	return work;
1095 }
1096 
1097 /*
1098  * Puts a workspace back on the list or frees it if we have enough
1099  * idle ones sitting around.
1100  */
1101 static void put_workspace(struct workspace_manager *wsm,
1102 			  struct crush_work *work)
1103 {
1104 	spin_lock(&wsm->ws_lock);
1105 	if (wsm->free_ws <= num_online_cpus()) {
1106 		list_add(&work->item, &wsm->idle_ws);
1107 		wsm->free_ws++;
1108 		spin_unlock(&wsm->ws_lock);
1109 		goto wake;
1110 	}
1111 	spin_unlock(&wsm->ws_lock);
1112 
1113 	free_workspace(work);
1114 	atomic_dec(&wsm->total_ws);
1115 wake:
1116 	if (wq_has_sleeper(&wsm->ws_wait))
1117 		wake_up(&wsm->ws_wait);
1118 }
1119 
1120 /*
1121  * osd map
1122  */
1123 struct ceph_osdmap *ceph_osdmap_alloc(void)
1124 {
1125 	struct ceph_osdmap *map;
1126 
1127 	map = kzalloc(sizeof(*map), GFP_NOIO);
1128 	if (!map)
1129 		return NULL;
1130 
1131 	map->pg_pools = RB_ROOT;
1132 	map->pool_max = -1;
1133 	map->pg_temp = RB_ROOT;
1134 	map->primary_temp = RB_ROOT;
1135 	map->pg_upmap = RB_ROOT;
1136 	map->pg_upmap_items = RB_ROOT;
1137 
1138 	init_workspace_manager(&map->crush_wsm);
1139 
1140 	return map;
1141 }
1142 
1143 void ceph_osdmap_destroy(struct ceph_osdmap *map)
1144 {
1145 	dout("osdmap_destroy %p\n", map);
1146 
1147 	if (map->crush)
1148 		crush_destroy(map->crush);
1149 	cleanup_workspace_manager(&map->crush_wsm);
1150 
1151 	while (!RB_EMPTY_ROOT(&map->pg_temp)) {
1152 		struct ceph_pg_mapping *pg =
1153 			rb_entry(rb_first(&map->pg_temp),
1154 				 struct ceph_pg_mapping, node);
1155 		erase_pg_mapping(&map->pg_temp, pg);
1156 		free_pg_mapping(pg);
1157 	}
1158 	while (!RB_EMPTY_ROOT(&map->primary_temp)) {
1159 		struct ceph_pg_mapping *pg =
1160 			rb_entry(rb_first(&map->primary_temp),
1161 				 struct ceph_pg_mapping, node);
1162 		erase_pg_mapping(&map->primary_temp, pg);
1163 		free_pg_mapping(pg);
1164 	}
1165 	while (!RB_EMPTY_ROOT(&map->pg_upmap)) {
1166 		struct ceph_pg_mapping *pg =
1167 			rb_entry(rb_first(&map->pg_upmap),
1168 				 struct ceph_pg_mapping, node);
1169 		rb_erase(&pg->node, &map->pg_upmap);
1170 		kfree(pg);
1171 	}
1172 	while (!RB_EMPTY_ROOT(&map->pg_upmap_items)) {
1173 		struct ceph_pg_mapping *pg =
1174 			rb_entry(rb_first(&map->pg_upmap_items),
1175 				 struct ceph_pg_mapping, node);
1176 		rb_erase(&pg->node, &map->pg_upmap_items);
1177 		kfree(pg);
1178 	}
1179 	while (!RB_EMPTY_ROOT(&map->pg_pools)) {
1180 		struct ceph_pg_pool_info *pi =
1181 			rb_entry(rb_first(&map->pg_pools),
1182 				 struct ceph_pg_pool_info, node);
1183 		__remove_pg_pool(&map->pg_pools, pi);
1184 	}
1185 	kvfree(map->osd_state);
1186 	kvfree(map->osd_weight);
1187 	kvfree(map->osd_addr);
1188 	kvfree(map->osd_primary_affinity);
1189 	kfree(map);
1190 }
1191 
1192 /*
1193  * Adjust max_osd value, (re)allocate arrays.
1194  *
1195  * The new elements are properly initialized.
1196  */
1197 static int osdmap_set_max_osd(struct ceph_osdmap *map, u32 max)
1198 {
1199 	u32 *state;
1200 	u32 *weight;
1201 	struct ceph_entity_addr *addr;
1202 	u32 to_copy;
1203 	int i;
1204 
1205 	dout("%s old %u new %u\n", __func__, map->max_osd, max);
1206 	if (max == map->max_osd)
1207 		return 0;
1208 
1209 	state = kvmalloc(array_size(max, sizeof(*state)), GFP_NOFS);
1210 	weight = kvmalloc(array_size(max, sizeof(*weight)), GFP_NOFS);
1211 	addr = kvmalloc(array_size(max, sizeof(*addr)), GFP_NOFS);
1212 	if (!state || !weight || !addr) {
1213 		kvfree(state);
1214 		kvfree(weight);
1215 		kvfree(addr);
1216 		return -ENOMEM;
1217 	}
1218 
1219 	to_copy = min(map->max_osd, max);
1220 	if (map->osd_state) {
1221 		memcpy(state, map->osd_state, to_copy * sizeof(*state));
1222 		memcpy(weight, map->osd_weight, to_copy * sizeof(*weight));
1223 		memcpy(addr, map->osd_addr, to_copy * sizeof(*addr));
1224 		kvfree(map->osd_state);
1225 		kvfree(map->osd_weight);
1226 		kvfree(map->osd_addr);
1227 	}
1228 
1229 	map->osd_state = state;
1230 	map->osd_weight = weight;
1231 	map->osd_addr = addr;
1232 	for (i = map->max_osd; i < max; i++) {
1233 		map->osd_state[i] = 0;
1234 		map->osd_weight[i] = CEPH_OSD_OUT;
1235 		memset(map->osd_addr + i, 0, sizeof(*map->osd_addr));
1236 	}
1237 
1238 	if (map->osd_primary_affinity) {
1239 		u32 *affinity;
1240 
1241 		affinity = kvmalloc(array_size(max, sizeof(*affinity)),
1242 					 GFP_NOFS);
1243 		if (!affinity)
1244 			return -ENOMEM;
1245 
1246 		memcpy(affinity, map->osd_primary_affinity,
1247 		       to_copy * sizeof(*affinity));
1248 		kvfree(map->osd_primary_affinity);
1249 
1250 		map->osd_primary_affinity = affinity;
1251 		for (i = map->max_osd; i < max; i++)
1252 			map->osd_primary_affinity[i] =
1253 			    CEPH_OSD_DEFAULT_PRIMARY_AFFINITY;
1254 	}
1255 
1256 	map->max_osd = max;
1257 
1258 	return 0;
1259 }
1260 
1261 static int osdmap_set_crush(struct ceph_osdmap *map, struct crush_map *crush)
1262 {
1263 	struct crush_work *work;
1264 
1265 	if (IS_ERR(crush))
1266 		return PTR_ERR(crush);
1267 
1268 	work = alloc_workspace(crush);
1269 	if (!work) {
1270 		crush_destroy(crush);
1271 		return -ENOMEM;
1272 	}
1273 
1274 	if (map->crush)
1275 		crush_destroy(map->crush);
1276 	cleanup_workspace_manager(&map->crush_wsm);
1277 	map->crush = crush;
1278 	add_initial_workspace(&map->crush_wsm, work);
1279 	return 0;
1280 }
1281 
1282 #define OSDMAP_WRAPPER_COMPAT_VER	7
1283 #define OSDMAP_CLIENT_DATA_COMPAT_VER	1
1284 
1285 /*
1286  * Return 0 or error.  On success, *v is set to 0 for old (v6) osdmaps,
1287  * to struct_v of the client_data section for new (v7 and above)
1288  * osdmaps.
1289  */
1290 static int get_osdmap_client_data_v(void **p, void *end,
1291 				    const char *prefix, u8 *v)
1292 {
1293 	u8 struct_v;
1294 
1295 	ceph_decode_8_safe(p, end, struct_v, e_inval);
1296 	if (struct_v >= 7) {
1297 		u8 struct_compat;
1298 
1299 		ceph_decode_8_safe(p, end, struct_compat, e_inval);
1300 		if (struct_compat > OSDMAP_WRAPPER_COMPAT_VER) {
1301 			pr_warn("got v %d cv %d > %d of %s ceph_osdmap\n",
1302 				struct_v, struct_compat,
1303 				OSDMAP_WRAPPER_COMPAT_VER, prefix);
1304 			return -EINVAL;
1305 		}
1306 		*p += 4; /* ignore wrapper struct_len */
1307 
1308 		ceph_decode_8_safe(p, end, struct_v, e_inval);
1309 		ceph_decode_8_safe(p, end, struct_compat, e_inval);
1310 		if (struct_compat > OSDMAP_CLIENT_DATA_COMPAT_VER) {
1311 			pr_warn("got v %d cv %d > %d of %s ceph_osdmap client data\n",
1312 				struct_v, struct_compat,
1313 				OSDMAP_CLIENT_DATA_COMPAT_VER, prefix);
1314 			return -EINVAL;
1315 		}
1316 		*p += 4; /* ignore client data struct_len */
1317 	} else {
1318 		u16 version;
1319 
1320 		*p -= 1;
1321 		ceph_decode_16_safe(p, end, version, e_inval);
1322 		if (version < 6) {
1323 			pr_warn("got v %d < 6 of %s ceph_osdmap\n",
1324 				version, prefix);
1325 			return -EINVAL;
1326 		}
1327 
1328 		/* old osdmap encoding */
1329 		struct_v = 0;
1330 	}
1331 
1332 	*v = struct_v;
1333 	return 0;
1334 
1335 e_inval:
1336 	return -EINVAL;
1337 }
1338 
1339 static int __decode_pools(void **p, void *end, struct ceph_osdmap *map,
1340 			  bool incremental)
1341 {
1342 	u32 n;
1343 
1344 	ceph_decode_32_safe(p, end, n, e_inval);
1345 	while (n--) {
1346 		struct ceph_pg_pool_info *pi;
1347 		u64 pool;
1348 		int ret;
1349 
1350 		ceph_decode_64_safe(p, end, pool, e_inval);
1351 
1352 		pi = lookup_pg_pool(&map->pg_pools, pool);
1353 		if (!incremental || !pi) {
1354 			pi = kzalloc(sizeof(*pi), GFP_NOFS);
1355 			if (!pi)
1356 				return -ENOMEM;
1357 
1358 			RB_CLEAR_NODE(&pi->node);
1359 			pi->id = pool;
1360 
1361 			if (!__insert_pg_pool(&map->pg_pools, pi)) {
1362 				kfree(pi);
1363 				return -EEXIST;
1364 			}
1365 		}
1366 
1367 		ret = decode_pool(p, end, pi);
1368 		if (ret)
1369 			return ret;
1370 	}
1371 
1372 	return 0;
1373 
1374 e_inval:
1375 	return -EINVAL;
1376 }
1377 
1378 static int decode_pools(void **p, void *end, struct ceph_osdmap *map)
1379 {
1380 	return __decode_pools(p, end, map, false);
1381 }
1382 
1383 static int decode_new_pools(void **p, void *end, struct ceph_osdmap *map)
1384 {
1385 	return __decode_pools(p, end, map, true);
1386 }
1387 
1388 typedef struct ceph_pg_mapping *(*decode_mapping_fn_t)(void **, void *, bool);
1389 
1390 static int decode_pg_mapping(void **p, void *end, struct rb_root *mapping_root,
1391 			     decode_mapping_fn_t fn, bool incremental)
1392 {
1393 	u32 n;
1394 
1395 	WARN_ON(!incremental && !fn);
1396 
1397 	ceph_decode_32_safe(p, end, n, e_inval);
1398 	while (n--) {
1399 		struct ceph_pg_mapping *pg;
1400 		struct ceph_pg pgid;
1401 		int ret;
1402 
1403 		ret = ceph_decode_pgid(p, end, &pgid);
1404 		if (ret)
1405 			return ret;
1406 
1407 		pg = lookup_pg_mapping(mapping_root, &pgid);
1408 		if (pg) {
1409 			WARN_ON(!incremental);
1410 			erase_pg_mapping(mapping_root, pg);
1411 			free_pg_mapping(pg);
1412 		}
1413 
1414 		if (fn) {
1415 			pg = fn(p, end, incremental);
1416 			if (IS_ERR(pg))
1417 				return PTR_ERR(pg);
1418 
1419 			if (pg) {
1420 				pg->pgid = pgid; /* struct */
1421 				insert_pg_mapping(mapping_root, pg);
1422 			}
1423 		}
1424 	}
1425 
1426 	return 0;
1427 
1428 e_inval:
1429 	return -EINVAL;
1430 }
1431 
1432 static struct ceph_pg_mapping *__decode_pg_temp(void **p, void *end,
1433 						bool incremental)
1434 {
1435 	struct ceph_pg_mapping *pg;
1436 	u32 len, i;
1437 
1438 	ceph_decode_32_safe(p, end, len, e_inval);
1439 	if (len == 0 && incremental)
1440 		return NULL;	/* new_pg_temp: [] to remove */
1441 	if (len > (SIZE_MAX - sizeof(*pg)) / sizeof(u32))
1442 		return ERR_PTR(-EINVAL);
1443 
1444 	ceph_decode_need(p, end, len * sizeof(u32), e_inval);
1445 	pg = alloc_pg_mapping(len * sizeof(u32));
1446 	if (!pg)
1447 		return ERR_PTR(-ENOMEM);
1448 
1449 	pg->pg_temp.len = len;
1450 	for (i = 0; i < len; i++)
1451 		pg->pg_temp.osds[i] = ceph_decode_32(p);
1452 
1453 	return pg;
1454 
1455 e_inval:
1456 	return ERR_PTR(-EINVAL);
1457 }
1458 
1459 static int decode_pg_temp(void **p, void *end, struct ceph_osdmap *map)
1460 {
1461 	return decode_pg_mapping(p, end, &map->pg_temp, __decode_pg_temp,
1462 				 false);
1463 }
1464 
1465 static int decode_new_pg_temp(void **p, void *end, struct ceph_osdmap *map)
1466 {
1467 	return decode_pg_mapping(p, end, &map->pg_temp, __decode_pg_temp,
1468 				 true);
1469 }
1470 
1471 static struct ceph_pg_mapping *__decode_primary_temp(void **p, void *end,
1472 						     bool incremental)
1473 {
1474 	struct ceph_pg_mapping *pg;
1475 	u32 osd;
1476 
1477 	ceph_decode_32_safe(p, end, osd, e_inval);
1478 	if (osd == (u32)-1 && incremental)
1479 		return NULL;	/* new_primary_temp: -1 to remove */
1480 
1481 	pg = alloc_pg_mapping(0);
1482 	if (!pg)
1483 		return ERR_PTR(-ENOMEM);
1484 
1485 	pg->primary_temp.osd = osd;
1486 	return pg;
1487 
1488 e_inval:
1489 	return ERR_PTR(-EINVAL);
1490 }
1491 
1492 static int decode_primary_temp(void **p, void *end, struct ceph_osdmap *map)
1493 {
1494 	return decode_pg_mapping(p, end, &map->primary_temp,
1495 				 __decode_primary_temp, false);
1496 }
1497 
1498 static int decode_new_primary_temp(void **p, void *end,
1499 				   struct ceph_osdmap *map)
1500 {
1501 	return decode_pg_mapping(p, end, &map->primary_temp,
1502 				 __decode_primary_temp, true);
1503 }
1504 
1505 u32 ceph_get_primary_affinity(struct ceph_osdmap *map, int osd)
1506 {
1507 	BUG_ON(osd >= map->max_osd);
1508 
1509 	if (!map->osd_primary_affinity)
1510 		return CEPH_OSD_DEFAULT_PRIMARY_AFFINITY;
1511 
1512 	return map->osd_primary_affinity[osd];
1513 }
1514 
1515 static int set_primary_affinity(struct ceph_osdmap *map, int osd, u32 aff)
1516 {
1517 	BUG_ON(osd >= map->max_osd);
1518 
1519 	if (!map->osd_primary_affinity) {
1520 		int i;
1521 
1522 		map->osd_primary_affinity = kvmalloc(
1523 		    array_size(map->max_osd, sizeof(*map->osd_primary_affinity)),
1524 		    GFP_NOFS);
1525 		if (!map->osd_primary_affinity)
1526 			return -ENOMEM;
1527 
1528 		for (i = 0; i < map->max_osd; i++)
1529 			map->osd_primary_affinity[i] =
1530 			    CEPH_OSD_DEFAULT_PRIMARY_AFFINITY;
1531 	}
1532 
1533 	map->osd_primary_affinity[osd] = aff;
1534 
1535 	return 0;
1536 }
1537 
1538 static int decode_primary_affinity(void **p, void *end,
1539 				   struct ceph_osdmap *map)
1540 {
1541 	u32 len, i;
1542 
1543 	ceph_decode_32_safe(p, end, len, e_inval);
1544 	if (len == 0) {
1545 		kvfree(map->osd_primary_affinity);
1546 		map->osd_primary_affinity = NULL;
1547 		return 0;
1548 	}
1549 	if (len != map->max_osd)
1550 		goto e_inval;
1551 
1552 	ceph_decode_need(p, end, map->max_osd*sizeof(u32), e_inval);
1553 
1554 	for (i = 0; i < map->max_osd; i++) {
1555 		int ret;
1556 
1557 		ret = set_primary_affinity(map, i, ceph_decode_32(p));
1558 		if (ret)
1559 			return ret;
1560 	}
1561 
1562 	return 0;
1563 
1564 e_inval:
1565 	return -EINVAL;
1566 }
1567 
1568 static int decode_new_primary_affinity(void **p, void *end,
1569 				       struct ceph_osdmap *map)
1570 {
1571 	u32 n;
1572 
1573 	ceph_decode_32_safe(p, end, n, e_inval);
1574 	while (n--) {
1575 		u32 osd, aff;
1576 		int ret;
1577 
1578 		ceph_decode_32_safe(p, end, osd, e_inval);
1579 		ceph_decode_32_safe(p, end, aff, e_inval);
1580 
1581 		ret = set_primary_affinity(map, osd, aff);
1582 		if (ret)
1583 			return ret;
1584 
1585 		osdmap_info(map, "osd%d primary-affinity 0x%x\n", osd, aff);
1586 	}
1587 
1588 	return 0;
1589 
1590 e_inval:
1591 	return -EINVAL;
1592 }
1593 
1594 static struct ceph_pg_mapping *__decode_pg_upmap(void **p, void *end,
1595 						 bool __unused)
1596 {
1597 	return __decode_pg_temp(p, end, false);
1598 }
1599 
1600 static int decode_pg_upmap(void **p, void *end, struct ceph_osdmap *map)
1601 {
1602 	return decode_pg_mapping(p, end, &map->pg_upmap, __decode_pg_upmap,
1603 				 false);
1604 }
1605 
1606 static int decode_new_pg_upmap(void **p, void *end, struct ceph_osdmap *map)
1607 {
1608 	return decode_pg_mapping(p, end, &map->pg_upmap, __decode_pg_upmap,
1609 				 true);
1610 }
1611 
1612 static int decode_old_pg_upmap(void **p, void *end, struct ceph_osdmap *map)
1613 {
1614 	return decode_pg_mapping(p, end, &map->pg_upmap, NULL, true);
1615 }
1616 
1617 static struct ceph_pg_mapping *__decode_pg_upmap_items(void **p, void *end,
1618 						       bool __unused)
1619 {
1620 	struct ceph_pg_mapping *pg;
1621 	u32 len, i;
1622 
1623 	ceph_decode_32_safe(p, end, len, e_inval);
1624 	if (len > (SIZE_MAX - sizeof(*pg)) / (2 * sizeof(u32)))
1625 		return ERR_PTR(-EINVAL);
1626 
1627 	ceph_decode_need(p, end, 2 * len * sizeof(u32), e_inval);
1628 	pg = alloc_pg_mapping(2 * len * sizeof(u32));
1629 	if (!pg)
1630 		return ERR_PTR(-ENOMEM);
1631 
1632 	pg->pg_upmap_items.len = len;
1633 	for (i = 0; i < len; i++) {
1634 		pg->pg_upmap_items.from_to[i][0] = ceph_decode_32(p);
1635 		pg->pg_upmap_items.from_to[i][1] = ceph_decode_32(p);
1636 	}
1637 
1638 	return pg;
1639 
1640 e_inval:
1641 	return ERR_PTR(-EINVAL);
1642 }
1643 
1644 static int decode_pg_upmap_items(void **p, void *end, struct ceph_osdmap *map)
1645 {
1646 	return decode_pg_mapping(p, end, &map->pg_upmap_items,
1647 				 __decode_pg_upmap_items, false);
1648 }
1649 
1650 static int decode_new_pg_upmap_items(void **p, void *end,
1651 				     struct ceph_osdmap *map)
1652 {
1653 	return decode_pg_mapping(p, end, &map->pg_upmap_items,
1654 				 __decode_pg_upmap_items, true);
1655 }
1656 
1657 static int decode_old_pg_upmap_items(void **p, void *end,
1658 				     struct ceph_osdmap *map)
1659 {
1660 	return decode_pg_mapping(p, end, &map->pg_upmap_items, NULL, true);
1661 }
1662 
1663 /*
1664  * decode a full map.
1665  */
1666 static int osdmap_decode(void **p, void *end, bool msgr2,
1667 			 struct ceph_osdmap *map)
1668 {
1669 	u8 struct_v;
1670 	u32 epoch = 0;
1671 	void *start = *p;
1672 	u32 max;
1673 	u32 len, i;
1674 	int err;
1675 
1676 	dout("%s %p to %p len %d\n", __func__, *p, end, (int)(end - *p));
1677 
1678 	err = get_osdmap_client_data_v(p, end, "full", &struct_v);
1679 	if (err)
1680 		goto bad;
1681 
1682 	/* fsid, epoch, created, modified */
1683 	ceph_decode_need(p, end, sizeof(map->fsid) + sizeof(u32) +
1684 			 sizeof(map->created) + sizeof(map->modified), e_inval);
1685 	ceph_decode_copy(p, &map->fsid, sizeof(map->fsid));
1686 	epoch = map->epoch = ceph_decode_32(p);
1687 	ceph_decode_copy(p, &map->created, sizeof(map->created));
1688 	ceph_decode_copy(p, &map->modified, sizeof(map->modified));
1689 
1690 	/* pools */
1691 	err = decode_pools(p, end, map);
1692 	if (err)
1693 		goto bad;
1694 
1695 	/* pool_name */
1696 	err = decode_pool_names(p, end, map);
1697 	if (err)
1698 		goto bad;
1699 
1700 	ceph_decode_32_safe(p, end, map->pool_max, e_inval);
1701 
1702 	ceph_decode_32_safe(p, end, map->flags, e_inval);
1703 
1704 	/* max_osd */
1705 	ceph_decode_32_safe(p, end, max, e_inval);
1706 
1707 	/* (re)alloc osd arrays */
1708 	err = osdmap_set_max_osd(map, max);
1709 	if (err)
1710 		goto bad;
1711 
1712 	/* osd_state, osd_weight, osd_addrs->client_addr */
1713 	ceph_decode_need(p, end, 3*sizeof(u32) +
1714 			 map->max_osd*(struct_v >= 5 ? sizeof(u32) :
1715 						       sizeof(u8)) +
1716 				       sizeof(*map->osd_weight), e_inval);
1717 	if (ceph_decode_32(p) != map->max_osd)
1718 		goto e_inval;
1719 
1720 	if (struct_v >= 5) {
1721 		for (i = 0; i < map->max_osd; i++)
1722 			map->osd_state[i] = ceph_decode_32(p);
1723 	} else {
1724 		for (i = 0; i < map->max_osd; i++)
1725 			map->osd_state[i] = ceph_decode_8(p);
1726 	}
1727 
1728 	if (ceph_decode_32(p) != map->max_osd)
1729 		goto e_inval;
1730 
1731 	for (i = 0; i < map->max_osd; i++)
1732 		map->osd_weight[i] = ceph_decode_32(p);
1733 
1734 	if (ceph_decode_32(p) != map->max_osd)
1735 		goto e_inval;
1736 
1737 	for (i = 0; i < map->max_osd; i++) {
1738 		struct ceph_entity_addr *addr = &map->osd_addr[i];
1739 
1740 		if (struct_v >= 8)
1741 			err = ceph_decode_entity_addrvec(p, end, msgr2, addr);
1742 		else
1743 			err = ceph_decode_entity_addr(p, end, addr);
1744 		if (err)
1745 			goto bad;
1746 
1747 		dout("%s osd%d addr %s\n", __func__, i, ceph_pr_addr(addr));
1748 	}
1749 
1750 	/* pg_temp */
1751 	err = decode_pg_temp(p, end, map);
1752 	if (err)
1753 		goto bad;
1754 
1755 	/* primary_temp */
1756 	if (struct_v >= 1) {
1757 		err = decode_primary_temp(p, end, map);
1758 		if (err)
1759 			goto bad;
1760 	}
1761 
1762 	/* primary_affinity */
1763 	if (struct_v >= 2) {
1764 		err = decode_primary_affinity(p, end, map);
1765 		if (err)
1766 			goto bad;
1767 	} else {
1768 		WARN_ON(map->osd_primary_affinity);
1769 	}
1770 
1771 	/* crush */
1772 	ceph_decode_32_safe(p, end, len, e_inval);
1773 	err = osdmap_set_crush(map, crush_decode(*p, min(*p + len, end)));
1774 	if (err)
1775 		goto bad;
1776 
1777 	*p += len;
1778 	if (struct_v >= 3) {
1779 		/* erasure_code_profiles */
1780 		ceph_decode_skip_map_of_map(p, end, string, string, string,
1781 					    e_inval);
1782 	}
1783 
1784 	if (struct_v >= 4) {
1785 		err = decode_pg_upmap(p, end, map);
1786 		if (err)
1787 			goto bad;
1788 
1789 		err = decode_pg_upmap_items(p, end, map);
1790 		if (err)
1791 			goto bad;
1792 	} else {
1793 		WARN_ON(!RB_EMPTY_ROOT(&map->pg_upmap));
1794 		WARN_ON(!RB_EMPTY_ROOT(&map->pg_upmap_items));
1795 	}
1796 
1797 	/* ignore the rest */
1798 	*p = end;
1799 
1800 	dout("full osdmap epoch %d max_osd %d\n", map->epoch, map->max_osd);
1801 	return 0;
1802 
1803 e_inval:
1804 	err = -EINVAL;
1805 bad:
1806 	pr_err("corrupt full osdmap (%d) epoch %d off %d (%p of %p-%p)\n",
1807 	       err, epoch, (int)(*p - start), *p, start, end);
1808 	print_hex_dump(KERN_DEBUG, "osdmap: ",
1809 		       DUMP_PREFIX_OFFSET, 16, 1,
1810 		       start, end - start, true);
1811 	return err;
1812 }
1813 
1814 /*
1815  * Allocate and decode a full map.
1816  */
1817 struct ceph_osdmap *ceph_osdmap_decode(void **p, void *end, bool msgr2)
1818 {
1819 	struct ceph_osdmap *map;
1820 	int ret;
1821 
1822 	map = ceph_osdmap_alloc();
1823 	if (!map)
1824 		return ERR_PTR(-ENOMEM);
1825 
1826 	ret = osdmap_decode(p, end, msgr2, map);
1827 	if (ret) {
1828 		ceph_osdmap_destroy(map);
1829 		return ERR_PTR(ret);
1830 	}
1831 
1832 	return map;
1833 }
1834 
1835 /*
1836  * Encoding order is (new_up_client, new_state, new_weight).  Need to
1837  * apply in the (new_weight, new_state, new_up_client) order, because
1838  * an incremental map may look like e.g.
1839  *
1840  *     new_up_client: { osd=6, addr=... } # set osd_state and addr
1841  *     new_state: { osd=6, xorstate=EXISTS } # clear osd_state
1842  */
1843 static int decode_new_up_state_weight(void **p, void *end, u8 struct_v,
1844 				      bool msgr2, struct ceph_osdmap *map)
1845 {
1846 	void *new_up_client;
1847 	void *new_state;
1848 	void *new_weight_end;
1849 	u32 len;
1850 	int ret;
1851 	int i;
1852 
1853 	new_up_client = *p;
1854 	ceph_decode_32_safe(p, end, len, e_inval);
1855 	for (i = 0; i < len; ++i) {
1856 		struct ceph_entity_addr addr;
1857 
1858 		ceph_decode_skip_32(p, end, e_inval);
1859 		if (struct_v >= 7)
1860 			ret = ceph_decode_entity_addrvec(p, end, msgr2, &addr);
1861 		else
1862 			ret = ceph_decode_entity_addr(p, end, &addr);
1863 		if (ret)
1864 			return ret;
1865 	}
1866 
1867 	new_state = *p;
1868 	ceph_decode_32_safe(p, end, len, e_inval);
1869 	len *= sizeof(u32) + (struct_v >= 5 ? sizeof(u32) : sizeof(u8));
1870 	ceph_decode_need(p, end, len, e_inval);
1871 	*p += len;
1872 
1873 	/* new_weight */
1874 	ceph_decode_32_safe(p, end, len, e_inval);
1875 	while (len--) {
1876 		s32 osd;
1877 		u32 w;
1878 
1879 		ceph_decode_need(p, end, 2*sizeof(u32), e_inval);
1880 		osd = ceph_decode_32(p);
1881 		w = ceph_decode_32(p);
1882 		BUG_ON(osd >= map->max_osd);
1883 		osdmap_info(map, "osd%d weight 0x%x %s\n", osd, w,
1884 			    w == CEPH_OSD_IN ? "(in)" :
1885 			    (w == CEPH_OSD_OUT ? "(out)" : ""));
1886 		map->osd_weight[osd] = w;
1887 
1888 		/*
1889 		 * If we are marking in, set the EXISTS, and clear the
1890 		 * AUTOOUT and NEW bits.
1891 		 */
1892 		if (w) {
1893 			map->osd_state[osd] |= CEPH_OSD_EXISTS;
1894 			map->osd_state[osd] &= ~(CEPH_OSD_AUTOOUT |
1895 						 CEPH_OSD_NEW);
1896 		}
1897 	}
1898 	new_weight_end = *p;
1899 
1900 	/* new_state (up/down) */
1901 	*p = new_state;
1902 	len = ceph_decode_32(p);
1903 	while (len--) {
1904 		s32 osd;
1905 		u32 xorstate;
1906 
1907 		osd = ceph_decode_32(p);
1908 		if (struct_v >= 5)
1909 			xorstate = ceph_decode_32(p);
1910 		else
1911 			xorstate = ceph_decode_8(p);
1912 		if (xorstate == 0)
1913 			xorstate = CEPH_OSD_UP;
1914 		BUG_ON(osd >= map->max_osd);
1915 		if ((map->osd_state[osd] & CEPH_OSD_UP) &&
1916 		    (xorstate & CEPH_OSD_UP))
1917 			osdmap_info(map, "osd%d down\n", osd);
1918 		if ((map->osd_state[osd] & CEPH_OSD_EXISTS) &&
1919 		    (xorstate & CEPH_OSD_EXISTS)) {
1920 			osdmap_info(map, "osd%d does not exist\n", osd);
1921 			ret = set_primary_affinity(map, osd,
1922 						   CEPH_OSD_DEFAULT_PRIMARY_AFFINITY);
1923 			if (ret)
1924 				return ret;
1925 			memset(map->osd_addr + osd, 0, sizeof(*map->osd_addr));
1926 			map->osd_state[osd] = 0;
1927 		} else {
1928 			map->osd_state[osd] ^= xorstate;
1929 		}
1930 	}
1931 
1932 	/* new_up_client */
1933 	*p = new_up_client;
1934 	len = ceph_decode_32(p);
1935 	while (len--) {
1936 		s32 osd;
1937 		struct ceph_entity_addr addr;
1938 
1939 		osd = ceph_decode_32(p);
1940 		BUG_ON(osd >= map->max_osd);
1941 		if (struct_v >= 7)
1942 			ret = ceph_decode_entity_addrvec(p, end, msgr2, &addr);
1943 		else
1944 			ret = ceph_decode_entity_addr(p, end, &addr);
1945 		if (ret)
1946 			return ret;
1947 
1948 		dout("%s osd%d addr %s\n", __func__, osd, ceph_pr_addr(&addr));
1949 
1950 		osdmap_info(map, "osd%d up\n", osd);
1951 		map->osd_state[osd] |= CEPH_OSD_EXISTS | CEPH_OSD_UP;
1952 		map->osd_addr[osd] = addr;
1953 	}
1954 
1955 	*p = new_weight_end;
1956 	return 0;
1957 
1958 e_inval:
1959 	return -EINVAL;
1960 }
1961 
1962 /*
1963  * decode and apply an incremental map update.
1964  */
1965 struct ceph_osdmap *osdmap_apply_incremental(void **p, void *end, bool msgr2,
1966 					     struct ceph_osdmap *map)
1967 {
1968 	struct ceph_fsid fsid;
1969 	u32 epoch = 0;
1970 	struct ceph_timespec modified;
1971 	s32 len;
1972 	u64 pool;
1973 	__s64 new_pool_max;
1974 	__s32 new_flags, max;
1975 	void *start = *p;
1976 	int err;
1977 	u8 struct_v;
1978 
1979 	dout("%s %p to %p len %d\n", __func__, *p, end, (int)(end - *p));
1980 
1981 	err = get_osdmap_client_data_v(p, end, "inc", &struct_v);
1982 	if (err)
1983 		goto bad;
1984 
1985 	/* fsid, epoch, modified, new_pool_max, new_flags */
1986 	ceph_decode_need(p, end, sizeof(fsid) + sizeof(u32) + sizeof(modified) +
1987 			 sizeof(u64) + sizeof(u32), e_inval);
1988 	ceph_decode_copy(p, &fsid, sizeof(fsid));
1989 	epoch = ceph_decode_32(p);
1990 	BUG_ON(epoch != map->epoch+1);
1991 	ceph_decode_copy(p, &modified, sizeof(modified));
1992 	new_pool_max = ceph_decode_64(p);
1993 	new_flags = ceph_decode_32(p);
1994 
1995 	/* full map? */
1996 	ceph_decode_32_safe(p, end, len, e_inval);
1997 	if (len > 0) {
1998 		dout("apply_incremental full map len %d, %p to %p\n",
1999 		     len, *p, end);
2000 		return ceph_osdmap_decode(p, min(*p+len, end), msgr2);
2001 	}
2002 
2003 	/* new crush? */
2004 	ceph_decode_32_safe(p, end, len, e_inval);
2005 	if (len > 0) {
2006 		err = osdmap_set_crush(map,
2007 				       crush_decode(*p, min(*p + len, end)));
2008 		if (err)
2009 			goto bad;
2010 		*p += len;
2011 	}
2012 
2013 	/* new flags? */
2014 	if (new_flags >= 0)
2015 		map->flags = new_flags;
2016 	if (new_pool_max >= 0)
2017 		map->pool_max = new_pool_max;
2018 
2019 	/* new max? */
2020 	ceph_decode_32_safe(p, end, max, e_inval);
2021 	if (max >= 0) {
2022 		err = osdmap_set_max_osd(map, max);
2023 		if (err)
2024 			goto bad;
2025 	}
2026 
2027 	map->epoch++;
2028 	map->modified = modified;
2029 
2030 	/* new_pools */
2031 	err = decode_new_pools(p, end, map);
2032 	if (err)
2033 		goto bad;
2034 
2035 	/* new_pool_names */
2036 	err = decode_pool_names(p, end, map);
2037 	if (err)
2038 		goto bad;
2039 
2040 	/* old_pool */
2041 	ceph_decode_32_safe(p, end, len, e_inval);
2042 	while (len--) {
2043 		struct ceph_pg_pool_info *pi;
2044 
2045 		ceph_decode_64_safe(p, end, pool, e_inval);
2046 		pi = lookup_pg_pool(&map->pg_pools, pool);
2047 		if (pi)
2048 			__remove_pg_pool(&map->pg_pools, pi);
2049 	}
2050 
2051 	/* new_up_client, new_state, new_weight */
2052 	err = decode_new_up_state_weight(p, end, struct_v, msgr2, map);
2053 	if (err)
2054 		goto bad;
2055 
2056 	/* new_pg_temp */
2057 	err = decode_new_pg_temp(p, end, map);
2058 	if (err)
2059 		goto bad;
2060 
2061 	/* new_primary_temp */
2062 	if (struct_v >= 1) {
2063 		err = decode_new_primary_temp(p, end, map);
2064 		if (err)
2065 			goto bad;
2066 	}
2067 
2068 	/* new_primary_affinity */
2069 	if (struct_v >= 2) {
2070 		err = decode_new_primary_affinity(p, end, map);
2071 		if (err)
2072 			goto bad;
2073 	}
2074 
2075 	if (struct_v >= 3) {
2076 		/* new_erasure_code_profiles */
2077 		ceph_decode_skip_map_of_map(p, end, string, string, string,
2078 					    e_inval);
2079 		/* old_erasure_code_profiles */
2080 		ceph_decode_skip_set(p, end, string, e_inval);
2081 	}
2082 
2083 	if (struct_v >= 4) {
2084 		err = decode_new_pg_upmap(p, end, map);
2085 		if (err)
2086 			goto bad;
2087 
2088 		err = decode_old_pg_upmap(p, end, map);
2089 		if (err)
2090 			goto bad;
2091 
2092 		err = decode_new_pg_upmap_items(p, end, map);
2093 		if (err)
2094 			goto bad;
2095 
2096 		err = decode_old_pg_upmap_items(p, end, map);
2097 		if (err)
2098 			goto bad;
2099 	}
2100 
2101 	/* ignore the rest */
2102 	*p = end;
2103 
2104 	dout("inc osdmap epoch %d max_osd %d\n", map->epoch, map->max_osd);
2105 	return map;
2106 
2107 e_inval:
2108 	err = -EINVAL;
2109 bad:
2110 	pr_err("corrupt inc osdmap (%d) epoch %d off %d (%p of %p-%p)\n",
2111 	       err, epoch, (int)(*p - start), *p, start, end);
2112 	print_hex_dump(KERN_DEBUG, "osdmap: ",
2113 		       DUMP_PREFIX_OFFSET, 16, 1,
2114 		       start, end - start, true);
2115 	return ERR_PTR(err);
2116 }
2117 
2118 void ceph_oloc_copy(struct ceph_object_locator *dest,
2119 		    const struct ceph_object_locator *src)
2120 {
2121 	ceph_oloc_destroy(dest);
2122 
2123 	dest->pool = src->pool;
2124 	if (src->pool_ns)
2125 		dest->pool_ns = ceph_get_string(src->pool_ns);
2126 	else
2127 		dest->pool_ns = NULL;
2128 }
2129 EXPORT_SYMBOL(ceph_oloc_copy);
2130 
2131 void ceph_oloc_destroy(struct ceph_object_locator *oloc)
2132 {
2133 	ceph_put_string(oloc->pool_ns);
2134 }
2135 EXPORT_SYMBOL(ceph_oloc_destroy);
2136 
2137 void ceph_oid_copy(struct ceph_object_id *dest,
2138 		   const struct ceph_object_id *src)
2139 {
2140 	ceph_oid_destroy(dest);
2141 
2142 	if (src->name != src->inline_name) {
2143 		/* very rare, see ceph_object_id definition */
2144 		dest->name = kmalloc(src->name_len + 1,
2145 				     GFP_NOIO | __GFP_NOFAIL);
2146 	} else {
2147 		dest->name = dest->inline_name;
2148 	}
2149 	memcpy(dest->name, src->name, src->name_len + 1);
2150 	dest->name_len = src->name_len;
2151 }
2152 EXPORT_SYMBOL(ceph_oid_copy);
2153 
2154 static __printf(2, 0)
2155 int oid_printf_vargs(struct ceph_object_id *oid, const char *fmt, va_list ap)
2156 {
2157 	int len;
2158 
2159 	WARN_ON(!ceph_oid_empty(oid));
2160 
2161 	len = vsnprintf(oid->inline_name, sizeof(oid->inline_name), fmt, ap);
2162 	if (len >= sizeof(oid->inline_name))
2163 		return len;
2164 
2165 	oid->name_len = len;
2166 	return 0;
2167 }
2168 
2169 /*
2170  * If oid doesn't fit into inline buffer, BUG.
2171  */
2172 void ceph_oid_printf(struct ceph_object_id *oid, const char *fmt, ...)
2173 {
2174 	va_list ap;
2175 
2176 	va_start(ap, fmt);
2177 	BUG_ON(oid_printf_vargs(oid, fmt, ap));
2178 	va_end(ap);
2179 }
2180 EXPORT_SYMBOL(ceph_oid_printf);
2181 
2182 static __printf(3, 0)
2183 int oid_aprintf_vargs(struct ceph_object_id *oid, gfp_t gfp,
2184 		      const char *fmt, va_list ap)
2185 {
2186 	va_list aq;
2187 	int len;
2188 
2189 	va_copy(aq, ap);
2190 	len = oid_printf_vargs(oid, fmt, aq);
2191 	va_end(aq);
2192 
2193 	if (len) {
2194 		char *external_name;
2195 
2196 		external_name = kmalloc(len + 1, gfp);
2197 		if (!external_name)
2198 			return -ENOMEM;
2199 
2200 		oid->name = external_name;
2201 		WARN_ON(vsnprintf(oid->name, len + 1, fmt, ap) != len);
2202 		oid->name_len = len;
2203 	}
2204 
2205 	return 0;
2206 }
2207 
2208 /*
2209  * If oid doesn't fit into inline buffer, allocate.
2210  */
2211 int ceph_oid_aprintf(struct ceph_object_id *oid, gfp_t gfp,
2212 		     const char *fmt, ...)
2213 {
2214 	va_list ap;
2215 	int ret;
2216 
2217 	va_start(ap, fmt);
2218 	ret = oid_aprintf_vargs(oid, gfp, fmt, ap);
2219 	va_end(ap);
2220 
2221 	return ret;
2222 }
2223 EXPORT_SYMBOL(ceph_oid_aprintf);
2224 
2225 void ceph_oid_destroy(struct ceph_object_id *oid)
2226 {
2227 	if (oid->name != oid->inline_name)
2228 		kfree(oid->name);
2229 }
2230 EXPORT_SYMBOL(ceph_oid_destroy);
2231 
2232 /*
2233  * osds only
2234  */
2235 static bool __osds_equal(const struct ceph_osds *lhs,
2236 			 const struct ceph_osds *rhs)
2237 {
2238 	if (lhs->size == rhs->size &&
2239 	    !memcmp(lhs->osds, rhs->osds, rhs->size * sizeof(rhs->osds[0])))
2240 		return true;
2241 
2242 	return false;
2243 }
2244 
2245 /*
2246  * osds + primary
2247  */
2248 static bool osds_equal(const struct ceph_osds *lhs,
2249 		       const struct ceph_osds *rhs)
2250 {
2251 	if (__osds_equal(lhs, rhs) &&
2252 	    lhs->primary == rhs->primary)
2253 		return true;
2254 
2255 	return false;
2256 }
2257 
2258 static bool osds_valid(const struct ceph_osds *set)
2259 {
2260 	/* non-empty set */
2261 	if (set->size > 0 && set->primary >= 0)
2262 		return true;
2263 
2264 	/* empty can_shift_osds set */
2265 	if (!set->size && set->primary == -1)
2266 		return true;
2267 
2268 	/* empty !can_shift_osds set - all NONE */
2269 	if (set->size > 0 && set->primary == -1) {
2270 		int i;
2271 
2272 		for (i = 0; i < set->size; i++) {
2273 			if (set->osds[i] != CRUSH_ITEM_NONE)
2274 				break;
2275 		}
2276 		if (i == set->size)
2277 			return true;
2278 	}
2279 
2280 	return false;
2281 }
2282 
2283 void ceph_osds_copy(struct ceph_osds *dest, const struct ceph_osds *src)
2284 {
2285 	memcpy(dest->osds, src->osds, src->size * sizeof(src->osds[0]));
2286 	dest->size = src->size;
2287 	dest->primary = src->primary;
2288 }
2289 
2290 bool ceph_pg_is_split(const struct ceph_pg *pgid, u32 old_pg_num,
2291 		      u32 new_pg_num)
2292 {
2293 	int old_bits = calc_bits_of(old_pg_num);
2294 	int old_mask = (1 << old_bits) - 1;
2295 	int n;
2296 
2297 	WARN_ON(pgid->seed >= old_pg_num);
2298 	if (new_pg_num <= old_pg_num)
2299 		return false;
2300 
2301 	for (n = 1; ; n++) {
2302 		int next_bit = n << (old_bits - 1);
2303 		u32 s = next_bit | pgid->seed;
2304 
2305 		if (s < old_pg_num || s == pgid->seed)
2306 			continue;
2307 		if (s >= new_pg_num)
2308 			break;
2309 
2310 		s = ceph_stable_mod(s, old_pg_num, old_mask);
2311 		if (s == pgid->seed)
2312 			return true;
2313 	}
2314 
2315 	return false;
2316 }
2317 
2318 bool ceph_is_new_interval(const struct ceph_osds *old_acting,
2319 			  const struct ceph_osds *new_acting,
2320 			  const struct ceph_osds *old_up,
2321 			  const struct ceph_osds *new_up,
2322 			  int old_size,
2323 			  int new_size,
2324 			  int old_min_size,
2325 			  int new_min_size,
2326 			  u32 old_pg_num,
2327 			  u32 new_pg_num,
2328 			  bool old_sort_bitwise,
2329 			  bool new_sort_bitwise,
2330 			  bool old_recovery_deletes,
2331 			  bool new_recovery_deletes,
2332 			  const struct ceph_pg *pgid)
2333 {
2334 	return !osds_equal(old_acting, new_acting) ||
2335 	       !osds_equal(old_up, new_up) ||
2336 	       old_size != new_size ||
2337 	       old_min_size != new_min_size ||
2338 	       ceph_pg_is_split(pgid, old_pg_num, new_pg_num) ||
2339 	       old_sort_bitwise != new_sort_bitwise ||
2340 	       old_recovery_deletes != new_recovery_deletes;
2341 }
2342 
2343 static int calc_pg_rank(int osd, const struct ceph_osds *acting)
2344 {
2345 	int i;
2346 
2347 	for (i = 0; i < acting->size; i++) {
2348 		if (acting->osds[i] == osd)
2349 			return i;
2350 	}
2351 
2352 	return -1;
2353 }
2354 
2355 static bool primary_changed(const struct ceph_osds *old_acting,
2356 			    const struct ceph_osds *new_acting)
2357 {
2358 	if (!old_acting->size && !new_acting->size)
2359 		return false; /* both still empty */
2360 
2361 	if (!old_acting->size ^ !new_acting->size)
2362 		return true; /* was empty, now not, or vice versa */
2363 
2364 	if (old_acting->primary != new_acting->primary)
2365 		return true; /* primary changed */
2366 
2367 	if (calc_pg_rank(old_acting->primary, old_acting) !=
2368 	    calc_pg_rank(new_acting->primary, new_acting))
2369 		return true;
2370 
2371 	return false; /* same primary (tho replicas may have changed) */
2372 }
2373 
2374 bool ceph_osds_changed(const struct ceph_osds *old_acting,
2375 		       const struct ceph_osds *new_acting,
2376 		       bool any_change)
2377 {
2378 	if (primary_changed(old_acting, new_acting))
2379 		return true;
2380 
2381 	if (any_change && !__osds_equal(old_acting, new_acting))
2382 		return true;
2383 
2384 	return false;
2385 }
2386 
2387 /*
2388  * Map an object into a PG.
2389  *
2390  * Should only be called with target_oid and target_oloc (as opposed to
2391  * base_oid and base_oloc), since tiering isn't taken into account.
2392  */
2393 void __ceph_object_locator_to_pg(struct ceph_pg_pool_info *pi,
2394 				 const struct ceph_object_id *oid,
2395 				 const struct ceph_object_locator *oloc,
2396 				 struct ceph_pg *raw_pgid)
2397 {
2398 	WARN_ON(pi->id != oloc->pool);
2399 
2400 	if (!oloc->pool_ns) {
2401 		raw_pgid->pool = oloc->pool;
2402 		raw_pgid->seed = ceph_str_hash(pi->object_hash, oid->name,
2403 					     oid->name_len);
2404 		dout("%s %s -> raw_pgid %llu.%x\n", __func__, oid->name,
2405 		     raw_pgid->pool, raw_pgid->seed);
2406 	} else {
2407 		char stack_buf[256];
2408 		char *buf = stack_buf;
2409 		int nsl = oloc->pool_ns->len;
2410 		size_t total = nsl + 1 + oid->name_len;
2411 
2412 		if (total > sizeof(stack_buf))
2413 			buf = kmalloc(total, GFP_NOIO | __GFP_NOFAIL);
2414 		memcpy(buf, oloc->pool_ns->str, nsl);
2415 		buf[nsl] = '\037';
2416 		memcpy(buf + nsl + 1, oid->name, oid->name_len);
2417 		raw_pgid->pool = oloc->pool;
2418 		raw_pgid->seed = ceph_str_hash(pi->object_hash, buf, total);
2419 		if (buf != stack_buf)
2420 			kfree(buf);
2421 		dout("%s %s ns %.*s -> raw_pgid %llu.%x\n", __func__,
2422 		     oid->name, nsl, oloc->pool_ns->str,
2423 		     raw_pgid->pool, raw_pgid->seed);
2424 	}
2425 }
2426 
2427 int ceph_object_locator_to_pg(struct ceph_osdmap *osdmap,
2428 			      const struct ceph_object_id *oid,
2429 			      const struct ceph_object_locator *oloc,
2430 			      struct ceph_pg *raw_pgid)
2431 {
2432 	struct ceph_pg_pool_info *pi;
2433 
2434 	pi = ceph_pg_pool_by_id(osdmap, oloc->pool);
2435 	if (!pi)
2436 		return -ENOENT;
2437 
2438 	__ceph_object_locator_to_pg(pi, oid, oloc, raw_pgid);
2439 	return 0;
2440 }
2441 EXPORT_SYMBOL(ceph_object_locator_to_pg);
2442 
2443 /*
2444  * Map a raw PG (full precision ps) into an actual PG.
2445  */
2446 static void raw_pg_to_pg(struct ceph_pg_pool_info *pi,
2447 			 const struct ceph_pg *raw_pgid,
2448 			 struct ceph_pg *pgid)
2449 {
2450 	pgid->pool = raw_pgid->pool;
2451 	pgid->seed = ceph_stable_mod(raw_pgid->seed, pi->pg_num,
2452 				     pi->pg_num_mask);
2453 }
2454 
2455 /*
2456  * Map a raw PG (full precision ps) into a placement ps (placement
2457  * seed).  Include pool id in that value so that different pools don't
2458  * use the same seeds.
2459  */
2460 static u32 raw_pg_to_pps(struct ceph_pg_pool_info *pi,
2461 			 const struct ceph_pg *raw_pgid)
2462 {
2463 	if (pi->flags & CEPH_POOL_FLAG_HASHPSPOOL) {
2464 		/* hash pool id and seed so that pool PGs do not overlap */
2465 		return crush_hash32_2(CRUSH_HASH_RJENKINS1,
2466 				      ceph_stable_mod(raw_pgid->seed,
2467 						      pi->pgp_num,
2468 						      pi->pgp_num_mask),
2469 				      raw_pgid->pool);
2470 	} else {
2471 		/*
2472 		 * legacy behavior: add ps and pool together.  this is
2473 		 * not a great approach because the PGs from each pool
2474 		 * will overlap on top of each other: 0.5 == 1.4 ==
2475 		 * 2.3 == ...
2476 		 */
2477 		return ceph_stable_mod(raw_pgid->seed, pi->pgp_num,
2478 				       pi->pgp_num_mask) +
2479 		       (unsigned)raw_pgid->pool;
2480 	}
2481 }
2482 
2483 /*
2484  * Magic value used for a "default" fallback choose_args, used if the
2485  * crush_choose_arg_map passed to do_crush() does not exist.  If this
2486  * also doesn't exist, fall back to canonical weights.
2487  */
2488 #define CEPH_DEFAULT_CHOOSE_ARGS	-1
2489 
2490 static int do_crush(struct ceph_osdmap *map, int ruleno, int x,
2491 		    int *result, int result_max,
2492 		    const __u32 *weight, int weight_max,
2493 		    s64 choose_args_index)
2494 {
2495 	struct crush_choose_arg_map *arg_map;
2496 	struct crush_work *work;
2497 	int r;
2498 
2499 	BUG_ON(result_max > CEPH_PG_MAX_SIZE);
2500 
2501 	arg_map = lookup_choose_arg_map(&map->crush->choose_args,
2502 					choose_args_index);
2503 	if (!arg_map)
2504 		arg_map = lookup_choose_arg_map(&map->crush->choose_args,
2505 						CEPH_DEFAULT_CHOOSE_ARGS);
2506 
2507 	work = get_workspace(&map->crush_wsm, map->crush);
2508 	r = crush_do_rule(map->crush, ruleno, x, result, result_max,
2509 			  weight, weight_max, work,
2510 			  arg_map ? arg_map->args : NULL);
2511 	put_workspace(&map->crush_wsm, work);
2512 	return r;
2513 }
2514 
2515 static void remove_nonexistent_osds(struct ceph_osdmap *osdmap,
2516 				    struct ceph_pg_pool_info *pi,
2517 				    struct ceph_osds *set)
2518 {
2519 	int i;
2520 
2521 	if (ceph_can_shift_osds(pi)) {
2522 		int removed = 0;
2523 
2524 		/* shift left */
2525 		for (i = 0; i < set->size; i++) {
2526 			if (!ceph_osd_exists(osdmap, set->osds[i])) {
2527 				removed++;
2528 				continue;
2529 			}
2530 			if (removed)
2531 				set->osds[i - removed] = set->osds[i];
2532 		}
2533 		set->size -= removed;
2534 	} else {
2535 		/* set dne devices to NONE */
2536 		for (i = 0; i < set->size; i++) {
2537 			if (!ceph_osd_exists(osdmap, set->osds[i]))
2538 				set->osds[i] = CRUSH_ITEM_NONE;
2539 		}
2540 	}
2541 }
2542 
2543 /*
2544  * Calculate raw set (CRUSH output) for given PG and filter out
2545  * nonexistent OSDs.  ->primary is undefined for a raw set.
2546  *
2547  * Placement seed (CRUSH input) is returned through @ppps.
2548  */
2549 static void pg_to_raw_osds(struct ceph_osdmap *osdmap,
2550 			   struct ceph_pg_pool_info *pi,
2551 			   const struct ceph_pg *raw_pgid,
2552 			   struct ceph_osds *raw,
2553 			   u32 *ppps)
2554 {
2555 	u32 pps = raw_pg_to_pps(pi, raw_pgid);
2556 	int ruleno;
2557 	int len;
2558 
2559 	ceph_osds_init(raw);
2560 	if (ppps)
2561 		*ppps = pps;
2562 
2563 	ruleno = crush_find_rule(osdmap->crush, pi->crush_ruleset, pi->type,
2564 				 pi->size);
2565 	if (ruleno < 0) {
2566 		pr_err("no crush rule: pool %lld ruleset %d type %d size %d\n",
2567 		       pi->id, pi->crush_ruleset, pi->type, pi->size);
2568 		return;
2569 	}
2570 
2571 	if (pi->size > ARRAY_SIZE(raw->osds)) {
2572 		pr_err_ratelimited("pool %lld ruleset %d type %d too wide: size %d > %zu\n",
2573 		       pi->id, pi->crush_ruleset, pi->type, pi->size,
2574 		       ARRAY_SIZE(raw->osds));
2575 		return;
2576 	}
2577 
2578 	len = do_crush(osdmap, ruleno, pps, raw->osds, pi->size,
2579 		       osdmap->osd_weight, osdmap->max_osd, pi->id);
2580 	if (len < 0) {
2581 		pr_err("error %d from crush rule %d: pool %lld ruleset %d type %d size %d\n",
2582 		       len, ruleno, pi->id, pi->crush_ruleset, pi->type,
2583 		       pi->size);
2584 		return;
2585 	}
2586 
2587 	raw->size = len;
2588 	remove_nonexistent_osds(osdmap, pi, raw);
2589 }
2590 
2591 /* apply pg_upmap[_items] mappings */
2592 static void apply_upmap(struct ceph_osdmap *osdmap,
2593 			const struct ceph_pg *pgid,
2594 			struct ceph_osds *raw)
2595 {
2596 	struct ceph_pg_mapping *pg;
2597 	int i, j;
2598 
2599 	pg = lookup_pg_mapping(&osdmap->pg_upmap, pgid);
2600 	if (pg) {
2601 		/* make sure targets aren't marked out */
2602 		for (i = 0; i < pg->pg_upmap.len; i++) {
2603 			int osd = pg->pg_upmap.osds[i];
2604 
2605 			if (osd != CRUSH_ITEM_NONE &&
2606 			    osd < osdmap->max_osd &&
2607 			    osdmap->osd_weight[osd] == 0) {
2608 				/* reject/ignore explicit mapping */
2609 				return;
2610 			}
2611 		}
2612 		for (i = 0; i < pg->pg_upmap.len; i++)
2613 			raw->osds[i] = pg->pg_upmap.osds[i];
2614 		raw->size = pg->pg_upmap.len;
2615 		/* check and apply pg_upmap_items, if any */
2616 	}
2617 
2618 	pg = lookup_pg_mapping(&osdmap->pg_upmap_items, pgid);
2619 	if (pg) {
2620 		/*
2621 		 * Note: this approach does not allow a bidirectional swap,
2622 		 * e.g., [[1,2],[2,1]] applied to [0,1,2] -> [0,2,1].
2623 		 */
2624 		for (i = 0; i < pg->pg_upmap_items.len; i++) {
2625 			int from = pg->pg_upmap_items.from_to[i][0];
2626 			int to = pg->pg_upmap_items.from_to[i][1];
2627 			int pos = -1;
2628 			bool exists = false;
2629 
2630 			/* make sure replacement doesn't already appear */
2631 			for (j = 0; j < raw->size; j++) {
2632 				int osd = raw->osds[j];
2633 
2634 				if (osd == to) {
2635 					exists = true;
2636 					break;
2637 				}
2638 				/* ignore mapping if target is marked out */
2639 				if (osd == from && pos < 0 &&
2640 				    !(to != CRUSH_ITEM_NONE &&
2641 				      to < osdmap->max_osd &&
2642 				      osdmap->osd_weight[to] == 0)) {
2643 					pos = j;
2644 				}
2645 			}
2646 			if (!exists && pos >= 0)
2647 				raw->osds[pos] = to;
2648 		}
2649 	}
2650 }
2651 
2652 /*
2653  * Given raw set, calculate up set and up primary.  By definition of an
2654  * up set, the result won't contain nonexistent or down OSDs.
2655  *
2656  * This is done in-place - on return @set is the up set.  If it's
2657  * empty, ->primary will remain undefined.
2658  */
2659 static void raw_to_up_osds(struct ceph_osdmap *osdmap,
2660 			   struct ceph_pg_pool_info *pi,
2661 			   struct ceph_osds *set)
2662 {
2663 	int i;
2664 
2665 	/* ->primary is undefined for a raw set */
2666 	BUG_ON(set->primary != -1);
2667 
2668 	if (ceph_can_shift_osds(pi)) {
2669 		int removed = 0;
2670 
2671 		/* shift left */
2672 		for (i = 0; i < set->size; i++) {
2673 			if (ceph_osd_is_down(osdmap, set->osds[i])) {
2674 				removed++;
2675 				continue;
2676 			}
2677 			if (removed)
2678 				set->osds[i - removed] = set->osds[i];
2679 		}
2680 		set->size -= removed;
2681 		if (set->size > 0)
2682 			set->primary = set->osds[0];
2683 	} else {
2684 		/* set down/dne devices to NONE */
2685 		for (i = set->size - 1; i >= 0; i--) {
2686 			if (ceph_osd_is_down(osdmap, set->osds[i]))
2687 				set->osds[i] = CRUSH_ITEM_NONE;
2688 			else
2689 				set->primary = set->osds[i];
2690 		}
2691 	}
2692 }
2693 
2694 static void apply_primary_affinity(struct ceph_osdmap *osdmap,
2695 				   struct ceph_pg_pool_info *pi,
2696 				   u32 pps,
2697 				   struct ceph_osds *up)
2698 {
2699 	int i;
2700 	int pos = -1;
2701 
2702 	/*
2703 	 * Do we have any non-default primary_affinity values for these
2704 	 * osds?
2705 	 */
2706 	if (!osdmap->osd_primary_affinity)
2707 		return;
2708 
2709 	for (i = 0; i < up->size; i++) {
2710 		int osd = up->osds[i];
2711 
2712 		if (osd != CRUSH_ITEM_NONE &&
2713 		    osdmap->osd_primary_affinity[osd] !=
2714 					CEPH_OSD_DEFAULT_PRIMARY_AFFINITY) {
2715 			break;
2716 		}
2717 	}
2718 	if (i == up->size)
2719 		return;
2720 
2721 	/*
2722 	 * Pick the primary.  Feed both the seed (for the pg) and the
2723 	 * osd into the hash/rng so that a proportional fraction of an
2724 	 * osd's pgs get rejected as primary.
2725 	 */
2726 	for (i = 0; i < up->size; i++) {
2727 		int osd = up->osds[i];
2728 		u32 aff;
2729 
2730 		if (osd == CRUSH_ITEM_NONE)
2731 			continue;
2732 
2733 		aff = osdmap->osd_primary_affinity[osd];
2734 		if (aff < CEPH_OSD_MAX_PRIMARY_AFFINITY &&
2735 		    (crush_hash32_2(CRUSH_HASH_RJENKINS1,
2736 				    pps, osd) >> 16) >= aff) {
2737 			/*
2738 			 * We chose not to use this primary.  Note it
2739 			 * anyway as a fallback in case we don't pick
2740 			 * anyone else, but keep looking.
2741 			 */
2742 			if (pos < 0)
2743 				pos = i;
2744 		} else {
2745 			pos = i;
2746 			break;
2747 		}
2748 	}
2749 	if (pos < 0)
2750 		return;
2751 
2752 	up->primary = up->osds[pos];
2753 
2754 	if (ceph_can_shift_osds(pi) && pos > 0) {
2755 		/* move the new primary to the front */
2756 		for (i = pos; i > 0; i--)
2757 			up->osds[i] = up->osds[i - 1];
2758 		up->osds[0] = up->primary;
2759 	}
2760 }
2761 
2762 /*
2763  * Get pg_temp and primary_temp mappings for given PG.
2764  *
2765  * Note that a PG may have none, only pg_temp, only primary_temp or
2766  * both pg_temp and primary_temp mappings.  This means @temp isn't
2767  * always a valid OSD set on return: in the "only primary_temp" case,
2768  * @temp will have its ->primary >= 0 but ->size == 0.
2769  */
2770 static void get_temp_osds(struct ceph_osdmap *osdmap,
2771 			  struct ceph_pg_pool_info *pi,
2772 			  const struct ceph_pg *pgid,
2773 			  struct ceph_osds *temp)
2774 {
2775 	struct ceph_pg_mapping *pg;
2776 	int i;
2777 
2778 	ceph_osds_init(temp);
2779 
2780 	/* pg_temp? */
2781 	pg = lookup_pg_mapping(&osdmap->pg_temp, pgid);
2782 	if (pg) {
2783 		for (i = 0; i < pg->pg_temp.len; i++) {
2784 			if (ceph_osd_is_down(osdmap, pg->pg_temp.osds[i])) {
2785 				if (ceph_can_shift_osds(pi))
2786 					continue;
2787 
2788 				temp->osds[temp->size++] = CRUSH_ITEM_NONE;
2789 			} else {
2790 				temp->osds[temp->size++] = pg->pg_temp.osds[i];
2791 			}
2792 		}
2793 
2794 		/* apply pg_temp's primary */
2795 		for (i = 0; i < temp->size; i++) {
2796 			if (temp->osds[i] != CRUSH_ITEM_NONE) {
2797 				temp->primary = temp->osds[i];
2798 				break;
2799 			}
2800 		}
2801 	}
2802 
2803 	/* primary_temp? */
2804 	pg = lookup_pg_mapping(&osdmap->primary_temp, pgid);
2805 	if (pg)
2806 		temp->primary = pg->primary_temp.osd;
2807 }
2808 
2809 /*
2810  * Map a PG to its acting set as well as its up set.
2811  *
2812  * Acting set is used for data mapping purposes, while up set can be
2813  * recorded for detecting interval changes and deciding whether to
2814  * resend a request.
2815  */
2816 void ceph_pg_to_up_acting_osds(struct ceph_osdmap *osdmap,
2817 			       struct ceph_pg_pool_info *pi,
2818 			       const struct ceph_pg *raw_pgid,
2819 			       struct ceph_osds *up,
2820 			       struct ceph_osds *acting)
2821 {
2822 	struct ceph_pg pgid;
2823 	u32 pps;
2824 
2825 	WARN_ON(pi->id != raw_pgid->pool);
2826 	raw_pg_to_pg(pi, raw_pgid, &pgid);
2827 
2828 	pg_to_raw_osds(osdmap, pi, raw_pgid, up, &pps);
2829 	apply_upmap(osdmap, &pgid, up);
2830 	raw_to_up_osds(osdmap, pi, up);
2831 	apply_primary_affinity(osdmap, pi, pps, up);
2832 	get_temp_osds(osdmap, pi, &pgid, acting);
2833 	if (!acting->size) {
2834 		memcpy(acting->osds, up->osds, up->size * sizeof(up->osds[0]));
2835 		acting->size = up->size;
2836 		if (acting->primary == -1)
2837 			acting->primary = up->primary;
2838 	}
2839 	WARN_ON(!osds_valid(up) || !osds_valid(acting));
2840 }
2841 
2842 bool ceph_pg_to_primary_shard(struct ceph_osdmap *osdmap,
2843 			      struct ceph_pg_pool_info *pi,
2844 			      const struct ceph_pg *raw_pgid,
2845 			      struct ceph_spg *spgid)
2846 {
2847 	struct ceph_pg pgid;
2848 	struct ceph_osds up, acting;
2849 	int i;
2850 
2851 	WARN_ON(pi->id != raw_pgid->pool);
2852 	raw_pg_to_pg(pi, raw_pgid, &pgid);
2853 
2854 	if (ceph_can_shift_osds(pi)) {
2855 		spgid->pgid = pgid; /* struct */
2856 		spgid->shard = CEPH_SPG_NOSHARD;
2857 		return true;
2858 	}
2859 
2860 	ceph_pg_to_up_acting_osds(osdmap, pi, &pgid, &up, &acting);
2861 	for (i = 0; i < acting.size; i++) {
2862 		if (acting.osds[i] == acting.primary) {
2863 			spgid->pgid = pgid; /* struct */
2864 			spgid->shard = i;
2865 			return true;
2866 		}
2867 	}
2868 
2869 	return false;
2870 }
2871 
2872 /*
2873  * Return acting primary for given PG, or -1 if none.
2874  */
2875 int ceph_pg_to_acting_primary(struct ceph_osdmap *osdmap,
2876 			      const struct ceph_pg *raw_pgid)
2877 {
2878 	struct ceph_pg_pool_info *pi;
2879 	struct ceph_osds up, acting;
2880 
2881 	pi = ceph_pg_pool_by_id(osdmap, raw_pgid->pool);
2882 	if (!pi)
2883 		return -1;
2884 
2885 	ceph_pg_to_up_acting_osds(osdmap, pi, raw_pgid, &up, &acting);
2886 	return acting.primary;
2887 }
2888 EXPORT_SYMBOL(ceph_pg_to_acting_primary);
2889 
2890 static struct crush_loc_node *alloc_crush_loc(size_t type_name_len,
2891 					      size_t name_len)
2892 {
2893 	struct crush_loc_node *loc;
2894 
2895 	loc = kmalloc(sizeof(*loc) + type_name_len + name_len + 2, GFP_NOIO);
2896 	if (!loc)
2897 		return NULL;
2898 
2899 	RB_CLEAR_NODE(&loc->cl_node);
2900 	return loc;
2901 }
2902 
2903 static void free_crush_loc(struct crush_loc_node *loc)
2904 {
2905 	WARN_ON(!RB_EMPTY_NODE(&loc->cl_node));
2906 
2907 	kfree(loc);
2908 }
2909 
2910 static int crush_loc_compare(const struct crush_loc *loc1,
2911 			     const struct crush_loc *loc2)
2912 {
2913 	return strcmp(loc1->cl_type_name, loc2->cl_type_name) ?:
2914 	       strcmp(loc1->cl_name, loc2->cl_name);
2915 }
2916 
2917 DEFINE_RB_FUNCS2(crush_loc, struct crush_loc_node, cl_loc, crush_loc_compare,
2918 		 RB_BYPTR, const struct crush_loc *, cl_node)
2919 
2920 /*
2921  * Parses a set of <bucket type name>':'<bucket name> pairs separated
2922  * by '|', e.g. "rack:foo1|rack:foo2|datacenter:bar".
2923  *
2924  * Note that @crush_location is modified by strsep().
2925  */
2926 int ceph_parse_crush_location(char *crush_location, struct rb_root *locs)
2927 {
2928 	struct crush_loc_node *loc;
2929 	const char *type_name, *name, *colon;
2930 	size_t type_name_len, name_len;
2931 
2932 	dout("%s '%s'\n", __func__, crush_location);
2933 	while ((type_name = strsep(&crush_location, "|"))) {
2934 		colon = strchr(type_name, ':');
2935 		if (!colon)
2936 			return -EINVAL;
2937 
2938 		type_name_len = colon - type_name;
2939 		if (type_name_len == 0)
2940 			return -EINVAL;
2941 
2942 		name = colon + 1;
2943 		name_len = strlen(name);
2944 		if (name_len == 0)
2945 			return -EINVAL;
2946 
2947 		loc = alloc_crush_loc(type_name_len, name_len);
2948 		if (!loc)
2949 			return -ENOMEM;
2950 
2951 		loc->cl_loc.cl_type_name = loc->cl_data;
2952 		memcpy(loc->cl_loc.cl_type_name, type_name, type_name_len);
2953 		loc->cl_loc.cl_type_name[type_name_len] = '\0';
2954 
2955 		loc->cl_loc.cl_name = loc->cl_data + type_name_len + 1;
2956 		memcpy(loc->cl_loc.cl_name, name, name_len);
2957 		loc->cl_loc.cl_name[name_len] = '\0';
2958 
2959 		if (!__insert_crush_loc(locs, loc)) {
2960 			free_crush_loc(loc);
2961 			return -EEXIST;
2962 		}
2963 
2964 		dout("%s type_name '%s' name '%s'\n", __func__,
2965 		     loc->cl_loc.cl_type_name, loc->cl_loc.cl_name);
2966 	}
2967 
2968 	return 0;
2969 }
2970 
2971 int ceph_compare_crush_locs(struct rb_root *locs1, struct rb_root *locs2)
2972 {
2973 	struct rb_node *n1 = rb_first(locs1);
2974 	struct rb_node *n2 = rb_first(locs2);
2975 	int ret;
2976 
2977 	for ( ; n1 && n2; n1 = rb_next(n1), n2 = rb_next(n2)) {
2978 		struct crush_loc_node *loc1 =
2979 		    rb_entry(n1, struct crush_loc_node, cl_node);
2980 		struct crush_loc_node *loc2 =
2981 		    rb_entry(n2, struct crush_loc_node, cl_node);
2982 
2983 		ret = crush_loc_compare(&loc1->cl_loc, &loc2->cl_loc);
2984 		if (ret)
2985 			return ret;
2986 	}
2987 
2988 	if (!n1 && n2)
2989 		return -1;
2990 	if (n1 && !n2)
2991 		return 1;
2992 	return 0;
2993 }
2994 
2995 void ceph_clear_crush_locs(struct rb_root *locs)
2996 {
2997 	while (!RB_EMPTY_ROOT(locs)) {
2998 		struct crush_loc_node *loc =
2999 		    rb_entry(rb_first(locs), struct crush_loc_node, cl_node);
3000 
3001 		erase_crush_loc(locs, loc);
3002 		free_crush_loc(loc);
3003 	}
3004 }
3005 
3006 /*
3007  * [a-zA-Z0-9-_.]+
3008  */
3009 static bool is_valid_crush_name(const char *name)
3010 {
3011 	do {
3012 		if (!('a' <= *name && *name <= 'z') &&
3013 		    !('A' <= *name && *name <= 'Z') &&
3014 		    !('0' <= *name && *name <= '9') &&
3015 		    *name != '-' && *name != '_' && *name != '.')
3016 			return false;
3017 	} while (*++name != '\0');
3018 
3019 	return true;
3020 }
3021 
3022 /*
3023  * Gets the parent of an item.  Returns its id (<0 because the
3024  * parent is always a bucket), type id (>0 for the same reason,
3025  * via @parent_type_id) and location (via @parent_loc).  If no
3026  * parent, returns 0.
3027  *
3028  * Does a linear search, as there are no parent pointers of any
3029  * kind.  Note that the result is ambiguous for items that occur
3030  * multiple times in the map.
3031  */
3032 static int get_immediate_parent(struct crush_map *c, int id,
3033 				u16 *parent_type_id,
3034 				struct crush_loc *parent_loc)
3035 {
3036 	struct crush_bucket *b;
3037 	struct crush_name_node *type_cn, *cn;
3038 	int i, j;
3039 
3040 	for (i = 0; i < c->max_buckets; i++) {
3041 		b = c->buckets[i];
3042 		if (!b)
3043 			continue;
3044 
3045 		/* ignore per-class shadow hierarchy */
3046 		cn = lookup_crush_name(&c->names, b->id);
3047 		if (!cn || !is_valid_crush_name(cn->cn_name))
3048 			continue;
3049 
3050 		for (j = 0; j < b->size; j++) {
3051 			if (b->items[j] != id)
3052 				continue;
3053 
3054 			*parent_type_id = b->type;
3055 			type_cn = lookup_crush_name(&c->type_names, b->type);
3056 			parent_loc->cl_type_name = type_cn->cn_name;
3057 			parent_loc->cl_name = cn->cn_name;
3058 			return b->id;
3059 		}
3060 	}
3061 
3062 	return 0;  /* no parent */
3063 }
3064 
3065 /*
3066  * Calculates the locality/distance from an item to a client
3067  * location expressed in terms of CRUSH hierarchy as a set of
3068  * (bucket type name, bucket name) pairs.  Specifically, looks
3069  * for the lowest-valued bucket type for which the location of
3070  * @id matches one of the locations in @locs, so for standard
3071  * bucket types (host = 1, rack = 3, datacenter = 8, zone = 9)
3072  * a matching host is closer than a matching rack and a matching
3073  * data center is closer than a matching zone.
3074  *
3075  * Specifying multiple locations (a "multipath" location) such
3076  * as "rack=foo1 rack=foo2 datacenter=bar" is allowed -- @locs
3077  * is a multimap.  The locality will be:
3078  *
3079  * - 3 for OSDs in racks foo1 and foo2
3080  * - 8 for OSDs in data center bar
3081  * - -1 for all other OSDs
3082  *
3083  * The lowest possible bucket type is 1, so the best locality
3084  * for an OSD is 1 (i.e. a matching host).  Locality 0 would be
3085  * the OSD itself.
3086  */
3087 int ceph_get_crush_locality(struct ceph_osdmap *osdmap, int id,
3088 			    struct rb_root *locs)
3089 {
3090 	struct crush_loc loc;
3091 	u16 type_id;
3092 
3093 	/*
3094 	 * Instead of repeated get_immediate_parent() calls,
3095 	 * the location of @id could be obtained with a single
3096 	 * depth-first traversal.
3097 	 */
3098 	for (;;) {
3099 		id = get_immediate_parent(osdmap->crush, id, &type_id, &loc);
3100 		if (id >= 0)
3101 			return -1;  /* not local */
3102 
3103 		if (lookup_crush_loc(locs, &loc))
3104 			return type_id;
3105 	}
3106 }
3107