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