linux - net/ceph/osdmap.c

3121 lines · 71.1 KiB

1	// SPDX-License-Identifier: GPL-2.0
2
3	#include <linux/ceph/ceph_debug.h>
4
5	#include <linux/module.h>
6	#include <linux/slab.h>
7
8	#include <linux/ceph/libceph.h>
9	#include <linux/ceph/osdmap.h>
10	#include <linux/ceph/decode.h>
11	#include <linux/crush/hash.h>
12	#include <linux/crush/mapper.h>
13
14	static __printf(2, 3)
15	void osdmap_info(const struct ceph_osdmap map, const char fmt, ...)
16	{
17	struct va_format vaf;
18	va_list args;
19
20	va_start(args, fmt);
21	vaf.fmt = fmt;
22	vaf.va = &args;
23
24	printk(KERN_INFO "%s (%pU e%u): %pV", KBUILD_MODNAME, &map->fsid,
25	map->epoch, &vaf);
26
27	va_end(args);
28	}
29
30	char ceph_osdmap_state_str(char str, int len, u32 state)
31	{
32	if (!len)
33	return str;
34
35	if ((state & CEPH_OSD_EXISTS) && (state & CEPH_OSD_UP))
36	snprintf(str, len, "exists, up");
37	else if (state & CEPH_OSD_EXISTS)
38	snprintf(str, len, "exists");
39	else if (state & CEPH_OSD_UP)
40	snprintf(str, len, "up");
41	else
42	snprintf(str, len, "doesn't exist");
43
44	return str;
45	}
46
47	/* maps */
48
49	static int calc_bits_of(unsigned int t)
50	{
51	int b = 0;
52	while (t) {
53	t = t >> 1;
54	b++;
55	}
56	return b;
57	}
58
59	/*
60	* the foo_mask is the smallest value 2^n-1 that is >= foo.
61	*/
62	static void calc_pg_masks(struct ceph_pg_pool_info *pi)
63	{
64	pi->pg_num_mask = (1 << calc_bits_of(pi->pg_num-1)) - 1;
65	pi->pgp_num_mask = (1 << calc_bits_of(pi->pgp_num-1)) - 1;
66	}
67
68	/*
69	* decode crush map
70	*/
71	static int crush_decode_uniform_bucket(void *p, void end,
72	struct crush_bucket_uniform *b)
73	{
74	dout("crush_decode_uniform_bucket %p to %p\n", *p, end);
75	ceph_decode_32_safe(p, end, b->item_weight, bad);
76	return 0;
77	bad:
78	return -EINVAL;
79	}
80
81	static int crush_decode_list_bucket(void *p, void end,
82	struct crush_bucket_list *b)
83	{
84	int j;
85	dout("crush_decode_list_bucket %p to %p\n", *p, end);
86	b->item_weights = kcalloc(b->h.size, sizeof(u32), GFP_NOFS);
87	if (b->item_weights == NULL)
88	return -ENOMEM;
89	b->sum_weights = kcalloc(b->h.size, sizeof(u32), GFP_NOFS);
90	if (b->sum_weights == NULL)
91	return -ENOMEM;
92	ceph_decode_need(p, end, 2 * b->h.size * sizeof(u32), bad);
93	for (j = 0; j < b->h.size; j++) {
94	b->item_weights[j] = ceph_decode_32(p);
95	b->sum_weights[j] = ceph_decode_32(p);
96	}
97	return 0;
98	bad:
99	return -EINVAL;
100	}
101
102	static int crush_decode_tree_bucket(void *p, void end,
103	struct crush_bucket_tree *b)
104	{
105	int j;
106	dout("crush_decode_tree_bucket %p to %p\n", *p, end);
107	ceph_decode_8_safe(p, end, b->num_nodes, bad);
108	b->node_weights = kcalloc(b->num_nodes, sizeof(u32), GFP_NOFS);
109	if (b->node_weights == NULL)
110	return -ENOMEM;
111	ceph_decode_need(p, end, b->num_nodes * sizeof(u32), bad);
112	for (j = 0; j < b->num_nodes; j++)
113	b->node_weights[j] = ceph_decode_32(p);
114	return 0;
115	bad:
116	return -EINVAL;
117	}
118
119	static int crush_decode_straw_bucket(void *p, void end,
120	struct crush_bucket_straw *b)
121	{
122	int j;
123	dout("crush_decode_straw_bucket %p to %p\n", *p, end);
124	b->item_weights = kcalloc(b->h.size, sizeof(u32), GFP_NOFS);
125	if (b->item_weights == NULL)
126	return -ENOMEM;
127	b->straws = kcalloc(b->h.size, sizeof(u32), GFP_NOFS);
128	if (b->straws == NULL)
129	return -ENOMEM;
130	ceph_decode_need(p, end, 2 * b->h.size * sizeof(u32), bad);
131	for (j = 0; j < b->h.size; j++) {
132	b->item_weights[j] = ceph_decode_32(p);
133	b->straws[j] = ceph_decode_32(p);
134	}
135	return 0;
136	bad:
137	return -EINVAL;
138	}
139
140	static int crush_decode_straw2_bucket(void *p, void end,
141	struct crush_bucket_straw2 *b)
142	{
143	int j;
144	dout("crush_decode_straw2_bucket %p to %p\n", *p, end);
145	b->item_weights = kcalloc(b->h.size, sizeof(u32), GFP_NOFS);
146	if (b->item_weights == NULL)
147	return -ENOMEM;
148	ceph_decode_need(p, end, b->h.size * sizeof(u32), bad);
149	for (j = 0; j < b->h.size; j++)
150	b->item_weights[j] = ceph_decode_32(p);
151	return 0;
152	bad:
153	return -EINVAL;
154	}
155
156	struct crush_name_node {
157	struct rb_node cn_node;
158	int cn_id;
159	char cn_name[];
160	};
161
162	static struct crush_name_node *alloc_crush_name(size_t name_len)
163	{
164	struct crush_name_node *cn;
165
166	cn = kmalloc(sizeof(*cn) + name_len + 1, GFP_NOIO);
167	if (!cn)
168	return NULL;
169
170	RB_CLEAR_NODE(&cn->cn_node);
171	return cn;
172	}
173
174	static void free_crush_name(struct crush_name_node *cn)
175	{
176	WARN_ON(!RB_EMPTY_NODE(&cn->cn_node));
177
178	kfree(cn);
179	}
180
181	DEFINE_RB_FUNCS(crush_name, struct crush_name_node, cn_id, cn_node)
182
183	static int decode_crush_names(void *p, void end, struct rb_root *root)
184	{
185	u32 n;
186
187	ceph_decode_32_safe(p, end, n, e_inval);
188	while (n--) {
189	struct crush_name_node *cn;
190	int id;
191	u32 name_len;
192
193	ceph_decode_32_safe(p, end, id, e_inval);
194	ceph_decode_32_safe(p, end, name_len, e_inval);
195	ceph_decode_need(p, end, name_len, e_inval);
196
197	cn = alloc_crush_name(name_len);
198	if (!cn)
199	return -ENOMEM;
200
201	cn->cn_id = id;
202	memcpy(cn->cn_name, *p, name_len);
203	cn->cn_name[name_len] = '\0';
204	*p += name_len;
205
206	if (!__insert_crush_name(root, cn)) {
207	free_crush_name(cn);
208	return -EEXIST;
209	}
210	}
211
212	return 0;
213
214	e_inval:
215	return -EINVAL;
216	}
217
218	void clear_crush_names(struct rb_root *root)
219	{
220	while (!RB_EMPTY_ROOT(root)) {
221	struct crush_name_node *cn =
222	rb_entry(rb_first(root), struct crush_name_node, cn_node);
223
224	erase_crush_name(root, cn);
225	free_crush_name(cn);
226	}
227	}
228
229	static struct crush_choose_arg_map *alloc_choose_arg_map(void)
230	{
231	struct crush_choose_arg_map *arg_map;
232
233	arg_map = kzalloc_obj(*arg_map, GFP_NOIO);
234	if (!arg_map)
235	return NULL;
236
237	RB_CLEAR_NODE(&arg_map->node);
238	return arg_map;
239	}
240
241	static void free_choose_arg_map(struct crush_choose_arg_map *arg_map)
242	{
243	int i, j;
244
245	if (!arg_map)
246	return;
247
248	WARN_ON(!RB_EMPTY_NODE(&arg_map->node));
249
250	if (arg_map->args) {
251	for (i = 0; i < arg_map->size; i++) {
252	struct crush_choose_arg *arg = &arg_map->args[i];
253	if (arg->weight_set) {
254	for (j = 0; j < arg->weight_set_size; j++)
255	kfree(arg->weight_set[j].weights);
256	kfree(arg->weight_set);
257	}
258	kfree(arg->ids);
259	}
260	kfree(arg_map->args);
261	}
262	kfree(arg_map);
263	}
264
265	DEFINE_RB_FUNCS(choose_arg_map, struct crush_choose_arg_map, choose_args_index,
266	node);
267
268	void clear_choose_args(struct crush_map *c)
269	{
270	while (!RB_EMPTY_ROOT(&c->choose_args)) {
271	struct crush_choose_arg_map *arg_map =
272	rb_entry(rb_first(&c->choose_args),
273	struct crush_choose_arg_map, node);
274
275	erase_choose_arg_map(&c->choose_args, arg_map);
276	free_choose_arg_map(arg_map);
277	}
278	}
279
280	static u32 decode_array_32_alloc(void p, void end, u32 *plen)
281	{
282	u32 *a = NULL;
283	u32 len;
284	int ret;
285
286	ceph_decode_32_safe(p, end, len, e_inval);
287	if (len) {
288	u32 i;
289
290	a = kmalloc_array(len, sizeof(u32), GFP_NOIO);
291	if (!a) {
292	ret = -ENOMEM;
293	goto fail;
294	}
295
296	ceph_decode_need(p, end, len * sizeof(u32), e_inval);
297	for (i = 0; i < len; i++)
298	a[i] = ceph_decode_32(p);
299	}
300
301	*plen = len;
302	return a;
303
304	e_inval:
305	ret = -EINVAL;
306	fail:
307	kfree(a);
308	return ERR_PTR(ret);
309	}
310
311	/*
312	* Assumes @arg is zero-initialized.
313	*/
314	static int decode_choose_arg(void *p, void end, struct crush_choose_arg *arg)
315	{
316	int ret;
317
318	ceph_decode_32_safe(p, end, arg->weight_set_size, e_inval);
319	if (arg->weight_set_size) {
320	u32 i;
321
322	arg->weight_set = kmalloc_objs(*arg->weight_set,
323	arg->weight_set_size, GFP_NOIO);
324	if (!arg->weight_set)
325	return -ENOMEM;
326
327	for (i = 0; i < arg->weight_set_size; i++) {
328	struct crush_weight_set *w = &arg->weight_set[i];
329
330	w->weights = decode_array_32_alloc(p, end, &w->size);
331	if (IS_ERR(w->weights)) {
332	ret = PTR_ERR(w->weights);
333	w->weights = NULL;
334	return ret;
335	}
336	}
337	}
338
339	arg->ids = decode_array_32_alloc(p, end, &arg->ids_size);
340	if (IS_ERR(arg->ids)) {
341	ret = PTR_ERR(arg->ids);
342	arg->ids = NULL;
343	return ret;
344	}
345
346	return 0;
347
348	e_inval:
349	return -EINVAL;
350	}
351
352	static int decode_choose_args(void *p, void end, struct crush_map *c)
353	{
354	struct crush_choose_arg_map *arg_map = NULL;
355	u32 num_choose_arg_maps, num_buckets;
356	int ret;
357
358	ceph_decode_32_safe(p, end, num_choose_arg_maps, e_inval);
359	while (num_choose_arg_maps--) {
360	arg_map = alloc_choose_arg_map();
361	if (!arg_map) {
362	ret = -ENOMEM;
363	goto fail;
364	}
365
366	ceph_decode_64_safe(p, end, arg_map->choose_args_index,
367	e_inval);
368	arg_map->size = c->max_buckets;
369	arg_map->args = kzalloc_objs(*arg_map->args, arg_map->size,
370	GFP_NOIO);
371	if (!arg_map->args) {
372	ret = -ENOMEM;
373	goto fail;
374	}
375
376	ceph_decode_32_safe(p, end, num_buckets, e_inval);
377	while (num_buckets--) {
378	struct crush_choose_arg *arg;
379	u32 bucket_index;
380
381	ceph_decode_32_safe(p, end, bucket_index, e_inval);
382	if (bucket_index >= arg_map->size)
383	goto e_inval;
384
385	arg = &arg_map->args[bucket_index];
386	ret = decode_choose_arg(p, end, arg);
387	if (ret)
388	goto fail;
389
390	if (arg->ids_size &&
391	(!c->buckets[bucket_index] \|\|
392	arg->ids_size != c->buckets[bucket_index]->size))
393	goto e_inval;
394	}
395
396	if (!__insert_choose_arg_map(&c->choose_args, arg_map)) {
397	ret = -EEXIST;
398	goto fail;
399	}
400	}
401
402	return 0;
403
404	e_inval:
405	ret = -EINVAL;
406	fail:
407	free_choose_arg_map(arg_map);
408	return ret;
409	}
410
411	static void crush_finalize(struct crush_map *c)
412	{
413	__s32 b;
414
415	/* Space for the array of pointers to per-bucket workspace */
416	c->working_size = sizeof(struct crush_work) +
417	c->max_buckets * sizeof(struct crush_work_bucket *);
418
419	for (b = 0; b < c->max_buckets; b++) {
420	if (!c->buckets[b])
421	continue;
422
423	switch (c->buckets[b]->alg) {
424	default:
425	/*
426	* The base case, permutation variables and
427	* the pointer to the permutation array.
428	*/
429	c->working_size += sizeof(struct crush_work_bucket);
430	break;
431	}
432	/* Every bucket has a permutation array. */
433	c->working_size += c->buckets[b]->size * sizeof(__u32);
434	}
435	}
436
437	static struct crush_map crush_decode(void pbyval, void *end)
438	{
439	struct crush_map *c;
440	int err;
441	int i, j;
442	void **p = &pbyval;
443	void *start = pbyval;
444	u32 magic;
445
446	dout("crush_decode %p to %p len %d\n", p, end, (int)(end - p));
447
448	c = kzalloc_obj(*c, GFP_NOFS);
449	if (c == NULL)
450	return ERR_PTR(-ENOMEM);
451
452	c->type_names = RB_ROOT;
453	c->names = RB_ROOT;
454	c->choose_args = RB_ROOT;
455
456	/* set tunables to default values */
457	c->choose_local_tries = 2;
458	c->choose_local_fallback_tries = 5;
459	c->choose_total_tries = 19;
460	c->chooseleaf_descend_once = 0;
461
462	ceph_decode_need(p, end, 4*sizeof(u32), bad);
463	magic = ceph_decode_32(p);
464	if (magic != CRUSH_MAGIC) {
465	pr_err("crush_decode magic %x != current %x\n",
466	(unsigned int)magic, (unsigned int)CRUSH_MAGIC);
467	goto bad;
468	}
469	c->max_buckets = ceph_decode_32(p);
470	c->max_rules = ceph_decode_32(p);
471	c->max_devices = ceph_decode_32(p);
472
473	c->buckets = kzalloc_objs(*c->buckets, c->max_buckets, GFP_NOFS);
474	if (c->buckets == NULL)
475	goto badmem;
476	c->rules = kzalloc_objs(*c->rules, c->max_rules, GFP_NOFS);
477	if (c->rules == NULL)
478	goto badmem;
479
480	/* buckets */
481	for (i = 0; i < c->max_buckets; i++) {
482	int size = 0;
483	u32 alg;
484	struct crush_bucket *b;
485
486	ceph_decode_32_safe(p, end, alg, bad);
487	if (alg == 0) {
488	c->buckets[i] = NULL;
489	continue;
490	}
491	dout("crush_decode bucket %d off %x %p to %p\n",
492	i, (int)(p-start), p, end);
493
494	switch (alg) {
495	case CRUSH_BUCKET_UNIFORM:
496	size = sizeof(struct crush_bucket_uniform);
497	break;
498	case CRUSH_BUCKET_LIST:
499	size = sizeof(struct crush_bucket_list);
500	break;
501	case CRUSH_BUCKET_TREE:
502	size = sizeof(struct crush_bucket_tree);
503	break;
504	case CRUSH_BUCKET_STRAW:
505	size = sizeof(struct crush_bucket_straw);
506	break;
507	case CRUSH_BUCKET_STRAW2:
508	size = sizeof(struct crush_bucket_straw2);
509	break;
510	default:
511	goto bad;
512	}
513	BUG_ON(size == 0);
514	b = c->buckets[i] = kzalloc(size, GFP_NOFS);
515	if (b == NULL)
516	goto badmem;
517
518	ceph_decode_need(p, end, 4*sizeof(u32), bad);
519	b->id = ceph_decode_32(p);
520	if (b->id != -1 - i)
521	goto bad;
522	b->type = ceph_decode_16(p);
523	if (b->type == 0)
524	goto bad;
525	b->alg = ceph_decode_8(p);
526	if (b->alg != alg) {
527	b->alg = 0;
528	goto bad;
529	}
530	b->hash = ceph_decode_8(p);
531	b->weight = ceph_decode_32(p);
532	b->size = ceph_decode_32(p);
533
534	dout("crush_decode bucket size %d off %x %p to %p\n",
535	b->size, (int)(p-start), p, end);
536
537	b->items = kzalloc_objs(__s32, b->size, GFP_NOFS);
538	if (b->items == NULL)
539	goto badmem;
540
541	ceph_decode_need(p, end, b->size*sizeof(u32), bad);
542	for (j = 0; j < b->size; j++)
543	b->items[j] = ceph_decode_32(p);
544
545	switch (b->alg) {
546	case CRUSH_BUCKET_UNIFORM:
547	err = crush_decode_uniform_bucket(p, end,
548	(struct crush_bucket_uniform *)b);
549	if (err < 0)
550	goto fail;
551	break;
552	case CRUSH_BUCKET_LIST:
553	err = crush_decode_list_bucket(p, end,
554	(struct crush_bucket_list *)b);
555	if (err < 0)
556	goto fail;
557	break;
558	case CRUSH_BUCKET_TREE:
559	err = crush_decode_tree_bucket(p, end,
560	(struct crush_bucket_tree *)b);
561	if (err < 0)
562	goto fail;
563	break;
564	case CRUSH_BUCKET_STRAW:
565	err = crush_decode_straw_bucket(p, end,
566	(struct crush_bucket_straw *)b);
567	if (err < 0)
568	goto fail;
569	break;
570	case CRUSH_BUCKET_STRAW2:
571	err = crush_decode_straw2_bucket(p, end,
572	(struct crush_bucket_straw2 *)b);
573	if (err < 0)
574	goto fail;
575	break;
576	}
577	}
578
579	/* rules */
580	dout("rule vec is %p\n", c->rules);
581	for (i = 0; i < c->max_rules; i++) {
582	u32 yes;
583	struct crush_rule *r;
584
585	ceph_decode_32_safe(p, end, yes, bad);
586	if (!yes) {
587	dout("crush_decode NO rule %d off %x %p to %p\n",
588	i, (int)(p-start), p, end);
589	c->rules[i] = NULL;
590	continue;
591	}
592
593	dout("crush_decode rule %d off %x %p to %p\n",
594	i, (int)(p-start), p, end);
595
596	/* len */
597	ceph_decode_32_safe(p, end, yes, bad);
598	#if BITS_PER_LONG == 32
599	if (yes > (ULONG_MAX - sizeof(*r))
600	/ sizeof(struct crush_rule_step))
601	goto bad;
602	#endif
603	r = kmalloc_flex(*r, steps, yes, GFP_NOFS);
604	if (r == NULL)
605	goto badmem;
606	dout(" rule %d is at %p\n", i, r);
607	c->rules[i] = r;
608	r->len = yes;
609	ceph_decode_copy_safe(p, end, &r->mask, 4, bad); /* 4 u8's */
610	ceph_decode_need(p, end, r->len3sizeof(u32), bad);
611	for (j = 0; j < r->len; j++) {
612	r->steps[j].op = ceph_decode_32(p);
613	r->steps[j].arg1 = ceph_decode_32(p);
614	r->steps[j].arg2 = ceph_decode_32(p);
615	}
616	}
617
618	err = decode_crush_names(p, end, &c->type_names);
619	if (err)
620	goto fail;
621
622	err = decode_crush_names(p, end, &c->names);
623	if (err)
624	goto fail;
625
626	ceph_decode_skip_map(p, end, 32, string, bad); /* rule_name_map */
627
628	/* tunables */
629	ceph_decode_need(p, end, 3*sizeof(u32), done);
630	c->choose_local_tries = ceph_decode_32(p);
631	c->choose_local_fallback_tries = ceph_decode_32(p);
632	c->choose_total_tries = ceph_decode_32(p);
633	dout("crush decode tunable choose_local_tries = %d\n",
634	c->choose_local_tries);
635	dout("crush decode tunable choose_local_fallback_tries = %d\n",
636	c->choose_local_fallback_tries);
637	dout("crush decode tunable choose_total_tries = %d\n",
638	c->choose_total_tries);
639
640	ceph_decode_need(p, end, sizeof(u32), done);
641	c->chooseleaf_descend_once = ceph_decode_32(p);
642	dout("crush decode tunable chooseleaf_descend_once = %d\n",
643	c->chooseleaf_descend_once);
644
645	ceph_decode_need(p, end, sizeof(u8), done);
646	c->chooseleaf_vary_r = ceph_decode_8(p);
647	dout("crush decode tunable chooseleaf_vary_r = %d\n",
648	c->chooseleaf_vary_r);
649
650	/* skip straw_calc_version, allowed_bucket_algs */
651	ceph_decode_need(p, end, sizeof(u8) + sizeof(u32), done);
652	*p += sizeof(u8) + sizeof(u32);
653
654	ceph_decode_need(p, end, sizeof(u8), done);
655	c->chooseleaf_stable = ceph_decode_8(p);
656	dout("crush decode tunable chooseleaf_stable = %d\n",
657	c->chooseleaf_stable);
658
659	if (*p != end) {
660	/* class_map */
661	ceph_decode_skip_map(p, end, 32, 32, bad);
662	/* class_name */
663	ceph_decode_skip_map(p, end, 32, string, bad);
664	/* class_bucket */
665	ceph_decode_skip_map_of_map(p, end, 32, 32, 32, bad);
666	}
667
668	if (*p != end) {
669	err = decode_choose_args(p, end, c);
670	if (err)
671	goto fail;
672	}
673
674	done:
675	crush_finalize(c);
676	dout("crush_decode success\n");
677	return c;
678
679	badmem:
680	err = -ENOMEM;
681	fail:
682	dout("crush_decode fail %d\n", err);
683	crush_destroy(c);
684	return ERR_PTR(err);
685
686	bad:
687	err = -EINVAL;
688	goto fail;
689	}
690
691	int ceph_pg_compare(const struct ceph_pg lhs, const struct ceph_pg rhs)
692	{
693	if (lhs->pool < rhs->pool)
694	return -1;
695	if (lhs->pool > rhs->pool)
696	return 1;
697	if (lhs->seed < rhs->seed)
698	return -1;
699	if (lhs->seed > rhs->seed)
700	return 1;
701
702	return 0;
703	}
704
705	int ceph_spg_compare(const struct ceph_spg lhs, const struct ceph_spg rhs)
706	{
707	int ret;
708
709	ret = ceph_pg_compare(&lhs->pgid, &rhs->pgid);
710	if (ret)
711	return ret;
712
713	if (lhs->shard < rhs->shard)
714	return -1;
715	if (lhs->shard > rhs->shard)
716	return 1;
717
718	return 0;
719	}
720
721	static struct ceph_pg_mapping *alloc_pg_mapping(size_t payload_len)
722	{
723	struct ceph_pg_mapping *pg;
724
725	pg = kmalloc(sizeof(*pg) + payload_len, GFP_NOIO);
726	if (!pg)
727	return NULL;
728
729	RB_CLEAR_NODE(&pg->node);
730	return pg;
731	}
732
733	static void free_pg_mapping(struct ceph_pg_mapping *pg)
734	{
735	WARN_ON(!RB_EMPTY_NODE(&pg->node));
736
737	kfree(pg);
738	}
739
740	/*
741	* rbtree of pg_mapping for handling pg_temp (explicit mapping of pgid
742	* to a set of osds) and primary_temp (explicit primary setting)
743	*/
744	DEFINE_RB_FUNCS2(pg_mapping, struct ceph_pg_mapping, pgid, ceph_pg_compare,
745	RB_BYPTR, const struct ceph_pg *, node)
746
747	/*
748	* rbtree of pg pool info
749	*/
750	DEFINE_RB_FUNCS(pg_pool, struct ceph_pg_pool_info, id, node)
751
752	struct ceph_pg_pool_info ceph_pg_pool_by_id(struct ceph_osdmap map, u64 id)
753	{
754	return lookup_pg_pool(&map->pg_pools, id);
755	}
756
757	const char ceph_pg_pool_name_by_id(struct ceph_osdmap map, u64 id)
758	{
759	struct ceph_pg_pool_info *pi;
760
761	if (id == CEPH_NOPOOL)
762	return NULL;
763
764	if (WARN_ON_ONCE(id > (u64) INT_MAX))
765	return NULL;
766
767	pi = lookup_pg_pool(&map->pg_pools, id);
768	return pi ? pi->name : NULL;
769	}
770	EXPORT_SYMBOL(ceph_pg_pool_name_by_id);
771
772	int ceph_pg_poolid_by_name(struct ceph_osdmap map, const char name)
773	{
774	struct rb_node *rbp;
775
776	for (rbp = rb_first(&map->pg_pools); rbp; rbp = rb_next(rbp)) {
777	struct ceph_pg_pool_info *pi =
778	rb_entry(rbp, struct ceph_pg_pool_info, node);
779	if (pi->name && strcmp(pi->name, name) == 0)
780	return pi->id;
781	}
782	return -ENOENT;
783	}
784	EXPORT_SYMBOL(ceph_pg_poolid_by_name);
785
786	u64 ceph_pg_pool_flags(struct ceph_osdmap *map, u64 id)
787	{
788	struct ceph_pg_pool_info *pi;
789
790	pi = lookup_pg_pool(&map->pg_pools, id);
791	return pi ? pi->flags : 0;
792	}
793	EXPORT_SYMBOL(ceph_pg_pool_flags);
794
795	static void __remove_pg_pool(struct rb_root root, struct ceph_pg_pool_info pi)
796	{
797	erase_pg_pool(root, pi);
798	kfree(pi->name);
799	kfree(pi);
800	}
801
802	static int decode_pool(void *p, void end, struct ceph_pg_pool_info *pi)
803	{
804	u8 ev, cv;
805	unsigned len, num;
806	void *pool_end;
807
808	ceph_decode_need(p, end, 2 + 4, bad);
809	ev = ceph_decode_8(p); /* encoding version */
810	cv = ceph_decode_8(p); /* compat version */
811	if (ev < 5) {
812	pr_warn("got v %d < 5 cv %d of ceph_pg_pool\n", ev, cv);
813	return -EINVAL;
814	}
815	if (cv > 9) {
816	pr_warn("got v %d cv %d > 9 of ceph_pg_pool\n", ev, cv);
817	return -EINVAL;
818	}
819	len = ceph_decode_32(p);
820	ceph_decode_need(p, end, len, bad);
821	pool_end = *p + len;
822
823	ceph_decode_need(p, end, 4 + 4 + 4, bad);
824	pi->type = ceph_decode_8(p);
825	pi->size = ceph_decode_8(p);
826	pi->crush_ruleset = ceph_decode_8(p);
827	pi->object_hash = ceph_decode_8(p);
828	pi->pg_num = ceph_decode_32(p);
829	pi->pgp_num = ceph_decode_32(p);
830
831	/* lpg, last_change, snap_seq, snap_epoch /
832	ceph_decode_skip_n(p, end, 8 + 4 + 8 + 4, bad);
833
834	/* skip snaps */
835	ceph_decode_32_safe(p, end, num, bad);
836	while (num--) {
837	/* snapid key, pool snap (with versions) */
838	ceph_decode_skip_n(p, end, 8 + 2, bad);
839	ceph_decode_skip_string(p, end, bad);
840	}
841
842	/* removed_snaps */
843	ceph_decode_skip_map(p, end, 64, 64, bad);
844
845	ceph_decode_need(p, end, 8 + 8 + 4, bad);
846	p += 8; / skip auid */
847	pi->flags = ceph_decode_64(p);
848	p += 4; / skip crash_replay_interval */
849
850	if (ev >= 7)
851	ceph_decode_8_safe(p, end, pi->min_size, bad);
852	else
853	pi->min_size = pi->size - pi->size / 2;
854
855	if (ev >= 8)
856	/* quota_max_* */
857	ceph_decode_skip_n(p, end, 8 + 8, bad);
858
859	if (ev >= 9) {
860	/* tiers */
861	ceph_decode_skip_set(p, end, 64, bad);
862
863	ceph_decode_need(p, end, 8 + 1 + 8 + 8, bad);
864	p += 8; / skip tier_of */
865	p += 1; / skip cache_mode */
866	pi->read_tier = ceph_decode_64(p);
867	pi->write_tier = ceph_decode_64(p);
868	} else {
869	pi->read_tier = -1;
870	pi->write_tier = -1;
871	}
872
873	if (ev >= 10)
874	/* properties */
875	ceph_decode_skip_map(p, end, string, string, bad);
876
877	if (ev >= 11) {
878	/* hit_set_params (with versions) */
879	ceph_decode_skip_n(p, end, 2, bad);
880	ceph_decode_skip_string(p, end, bad);
881
882	/* hit_set_period, hit_set_count */
883	ceph_decode_skip_n(p, end, 4 + 4, bad);
884	}
885
886	if (ev >= 12)
887	/* stripe_width */
888	ceph_decode_skip_32(p, end, bad);
889
890	if (ev >= 13)
891	/* target_max_, cache_target_, cache_min_* */
892	ceph_decode_skip_n(p, end, 16 + 8 + 8, bad);
893
894	if (ev >= 14)
895	/* erasure_code_profile */
896	ceph_decode_skip_string(p, end, bad);
897
898	/*
899	* last_force_op_resend_preluminous, will be overridden if the
900	* map was encoded with RESEND_ON_SPLIT
901	*/
902	if (ev >= 15)
903	ceph_decode_32_safe(p, end, pi->last_force_request_resend, bad);
904	else
905	pi->last_force_request_resend = 0;
906
907	if (ev >= 16)
908	/* min_read_recency_for_promote */
909	ceph_decode_skip_32(p, end, bad);
910
911	if (ev >= 17)
912	/* expected_num_objects */
913	ceph_decode_skip_64(p, end, bad);
914
915	if (ev >= 19)
916	/* cache_target_dirty_high_ratio_micro */
917	ceph_decode_skip_32(p, end, bad);
918
919	if (ev >= 20)
920	/* min_write_recency_for_promote */
921	ceph_decode_skip_32(p, end, bad);
922
923	if (ev >= 21)
924	/* use_gmt_hitset */
925	ceph_decode_skip_8(p, end, bad);
926
927	if (ev >= 22)
928	/* fast_read */
929	ceph_decode_skip_8(p, end, bad);
930
931	if (ev >= 23)
932	/* hit_set_grade_decay_rate, hit_set_search_last_n */
933	ceph_decode_skip_n(p, end, 4 + 4, bad);
934
935	if (ev >= 24) {
936	/* opts (with versions) */
937	ceph_decode_skip_n(p, end, 2, bad);
938	ceph_decode_skip_string(p, end, bad);
939	}
940
941	if (ev >= 25)
942	ceph_decode_32_safe(p, end, pi->last_force_request_resend, bad);
943
944	/* ignore the rest */
945
946	*p = pool_end;
947	calc_pg_masks(pi);
948	return 0;
949
950	bad:
951	return -EINVAL;
952	}
953
954	static int decode_pool_names(void *p, void end, struct ceph_osdmap *map)
955	{
956	struct ceph_pg_pool_info *pi;
957	u32 num, len;
958	u64 pool;
959
960	ceph_decode_32_safe(p, end, num, bad);
961	dout(" %d pool names\n", num);
962	while (num--) {
963	ceph_decode_64_safe(p, end, pool, bad);
964	ceph_decode_32_safe(p, end, len, bad);
965	dout(" pool %llu len %d\n", pool, len);
966	ceph_decode_need(p, end, len, bad);
967	pi = lookup_pg_pool(&map->pg_pools, pool);
968	if (pi) {
969	char name = kstrndup(p, len, GFP_NOFS);
970
971	if (!name)
972	return -ENOMEM;
973	kfree(pi->name);
974	pi->name = name;
975	dout(" name is %s\n", pi->name);
976	}
977	*p += len;
978	}
979	return 0;
980
981	bad:
982	return -EINVAL;
983	}
984
985	/*
986	* CRUSH workspaces
987	*
988	* workspace_manager framework borrowed from fs/btrfs/compression.c.
989	* Two simplifications: there is only one type of workspace and there
990	* is always at least one workspace.
991	*/
992	static struct crush_work alloc_workspace(const struct crush_map c)
993	{
994	struct crush_work *work;
995	size_t work_size;
996
997	WARN_ON(!c->working_size);
998	work_size = crush_work_size(c, CEPH_PG_MAX_SIZE);
999	dout("%s work_size %zu bytes\n", __func__, work_size);
1000
1001	work = kvmalloc(work_size, GFP_NOIO);
1002	if (!work)
1003	return NULL;
1004
1005	INIT_LIST_HEAD(&work->item);
1006	crush_init_workspace(c, work);
1007	return work;
1008	}
1009
1010	static void free_workspace(struct crush_work *work)
1011	{
1012	WARN_ON(!list_empty(&work->item));
1013	kvfree(work);
1014	}
1015
1016	static void init_workspace_manager(struct workspace_manager *wsm)
1017	{
1018	INIT_LIST_HEAD(&wsm->idle_ws);
1019	spin_lock_init(&wsm->ws_lock);
1020	atomic_set(&wsm->total_ws, 0);
1021	wsm->free_ws = 0;
1022	init_waitqueue_head(&wsm->ws_wait);
1023	}
1024
1025	static void add_initial_workspace(struct workspace_manager *wsm,
1026	struct crush_work *work)
1027	{
1028	WARN_ON(!list_empty(&wsm->idle_ws));
1029
1030	list_add(&work->item, &wsm->idle_ws);
1031	atomic_set(&wsm->total_ws, 1);
1032	wsm->free_ws = 1;
1033	}
1034
1035	static void cleanup_workspace_manager(struct workspace_manager *wsm)
1036	{
1037	struct crush_work *work;
1038
1039	while (!list_empty(&wsm->idle_ws)) {
1040	work = list_first_entry(&wsm->idle_ws, struct crush_work,
1041	item);
1042	list_del_init(&work->item);
1043	free_workspace(work);
1044	}
1045	atomic_set(&wsm->total_ws, 0);
1046	wsm->free_ws = 0;
1047	}
1048
1049	/*
1050	* Finds an available workspace or allocates a new one. If it's not
1051	* possible to allocate a new one, waits until there is one.
1052	*/
1053	static struct crush_work get_workspace(struct workspace_manager wsm,
1054	const struct crush_map *c)
1055	{
1056	struct crush_work *work;
1057	int cpus = num_online_cpus();
1058
1059	again:
1060	spin_lock(&wsm->ws_lock);
1061	if (!list_empty(&wsm->idle_ws)) {
1062	work = list_first_entry(&wsm->idle_ws, struct crush_work,
1063	item);
1064	list_del_init(&work->item);
1065	wsm->free_ws--;
1066	spin_unlock(&wsm->ws_lock);
1067	return work;
1068
1069	}
1070	if (atomic_read(&wsm->total_ws) > cpus) {
1071	DEFINE_WAIT(wait);
1072
1073	spin_unlock(&wsm->ws_lock);
1074	prepare_to_wait(&wsm->ws_wait, &wait, TASK_UNINTERRUPTIBLE);
1075	if (atomic_read(&wsm->total_ws) > cpus && !wsm->free_ws)
1076	schedule();
1077	finish_wait(&wsm->ws_wait, &wait);
1078	goto again;
1079	}
1080	atomic_inc(&wsm->total_ws);
1081	spin_unlock(&wsm->ws_lock);
1082
1083	work = alloc_workspace(c);
1084	if (!work) {
1085	atomic_dec(&wsm->total_ws);
1086	wake_up(&wsm->ws_wait);
1087
1088	/*
1089	* Do not return the error but go back to waiting. We
1090	* have the initial workspace and the CRUSH computation
1091	* time is bounded so we will get it eventually.
1092	*/
1093	WARN_ON(atomic_read(&wsm->total_ws) < 1);
1094	goto again;
1095	}
1096	return work;
1097	}
1098
1099	/*
1100	* Puts a workspace back on the list or frees it if we have enough
1101	* idle ones sitting around.
1102	*/
1103	static void put_workspace(struct workspace_manager *wsm,
1104	struct crush_work *work)
1105	{
1106	spin_lock(&wsm->ws_lock);
1107	if (wsm->free_ws <= num_online_cpus()) {
1108	list_add(&work->item, &wsm->idle_ws);
1109	wsm->free_ws++;
1110	spin_unlock(&wsm->ws_lock);
1111	goto wake;
1112	}
1113	spin_unlock(&wsm->ws_lock);
1114
1115	free_workspace(work);
1116	atomic_dec(&wsm->total_ws);
1117	wake:
1118	if (wq_has_sleeper(&wsm->ws_wait))
1119	wake_up(&wsm->ws_wait);
1120	}
1121
1122	/*
1123	* osd map
1124	*/
1125	struct ceph_osdmap *ceph_osdmap_alloc(void)
1126	{
1127	struct ceph_osdmap *map;
1128
1129	map = kzalloc_obj(*map, GFP_NOIO);
1130	if (!map)
1131	return NULL;
1132
1133	map->pg_pools = RB_ROOT;
1134	map->pool_max = -1;
1135	map->pg_temp = RB_ROOT;
1136	map->primary_temp = RB_ROOT;
1137	map->pg_upmap = RB_ROOT;
1138	map->pg_upmap_items = RB_ROOT;
1139
1140	init_workspace_manager(&map->crush_wsm);
1141
1142	return map;
1143	}
1144
1145	void ceph_osdmap_destroy(struct ceph_osdmap *map)
1146	{
1147	dout("osdmap_destroy %p\n", map);
1148
1149	if (map->crush)
1150	crush_destroy(map->crush);
1151	cleanup_workspace_manager(&map->crush_wsm);
1152
1153	while (!RB_EMPTY_ROOT(&map->pg_temp)) {
1154	struct ceph_pg_mapping *pg =
1155	rb_entry(rb_first(&map->pg_temp),
1156	struct ceph_pg_mapping, node);
1157	erase_pg_mapping(&map->pg_temp, pg);
1158	free_pg_mapping(pg);
1159	}
1160	while (!RB_EMPTY_ROOT(&map->primary_temp)) {
1161	struct ceph_pg_mapping *pg =
1162	rb_entry(rb_first(&map->primary_temp),
1163	struct ceph_pg_mapping, node);
1164	erase_pg_mapping(&map->primary_temp, pg);
1165	free_pg_mapping(pg);
1166	}
1167	while (!RB_EMPTY_ROOT(&map->pg_upmap)) {
1168	struct ceph_pg_mapping *pg =
1169	rb_entry(rb_first(&map->pg_upmap),
1170	struct ceph_pg_mapping, node);
1171	rb_erase(&pg->node, &map->pg_upmap);
1172	kfree(pg);
1173	}
1174	while (!RB_EMPTY_ROOT(&map->pg_upmap_items)) {
1175	struct ceph_pg_mapping *pg =
1176	rb_entry(rb_first(&map->pg_upmap_items),
1177	struct ceph_pg_mapping, node);
1178	rb_erase(&pg->node, &map->pg_upmap_items);
1179	kfree(pg);
1180	}
1181	while (!RB_EMPTY_ROOT(&map->pg_pools)) {
1182	struct ceph_pg_pool_info *pi =
1183	rb_entry(rb_first(&map->pg_pools),
1184	struct ceph_pg_pool_info, node);
1185	__remove_pg_pool(&map->pg_pools, pi);
1186	}
1187	kvfree(map->osd_state);
1188	kvfree(map->osd_weight);
1189	kvfree(map->osd_addr);
1190	kvfree(map->osd_primary_affinity);
1191	kfree(map);
1192	}
1193
1194	/*
1195	* Adjust max_osd value, (re)allocate arrays.
1196	*
1197	* The new elements are properly initialized.
1198	*/
1199	static int osdmap_set_max_osd(struct ceph_osdmap *map, u32 max)
1200	{
1201	u32 *state;
1202	u32 *weight;
1203	struct ceph_entity_addr *addr;
1204	u32 to_copy;
1205	int i;
1206
1207	dout("%s old %u new %u\n", __func__, map->max_osd, max);
1208	if (max == map->max_osd)
1209	return 0;
1210
1211	state = kvmalloc(array_size(max, sizeof(*state)), GFP_NOFS);
1212	weight = kvmalloc(array_size(max, sizeof(*weight)), GFP_NOFS);
1213	addr = kvmalloc(array_size(max, sizeof(*addr)), GFP_NOFS);
1214	if (!state \|\| !weight \|\| !addr) {
1215	kvfree(state);
1216	kvfree(weight);
1217	kvfree(addr);
1218	return -ENOMEM;
1219	}
1220
1221	to_copy = min(map->max_osd, max);
1222	if (map->osd_state) {
1223	memcpy(state, map->osd_state, to_copy * sizeof(*state));
1224	memcpy(weight, map->osd_weight, to_copy * sizeof(*weight));
1225	memcpy(addr, map->osd_addr, to_copy * sizeof(*addr));
1226	kvfree(map->osd_state);
1227	kvfree(map->osd_weight);
1228	kvfree(map->osd_addr);
1229	}
1230
1231	map->osd_state = state;
1232	map->osd_weight = weight;
1233	map->osd_addr = addr;
1234	for (i = map->max_osd; i < max; i++) {
1235	map->osd_state[i] = 0;
1236	map->osd_weight[i] = CEPH_OSD_OUT;
1237	memset(map->osd_addr + i, 0, sizeof(*map->osd_addr));
1238	}
1239
1240	if (map->osd_primary_affinity) {
1241	u32 *affinity;
1242
1243	affinity = kvmalloc(array_size(max, sizeof(*affinity)),
1244	GFP_NOFS);
1245	if (!affinity)
1246	return -ENOMEM;
1247
1248	memcpy(affinity, map->osd_primary_affinity,
1249	to_copy * sizeof(*affinity));
1250	kvfree(map->osd_primary_affinity);
1251
1252	map->osd_primary_affinity = affinity;
1253	for (i = map->max_osd; i < max; i++)
1254	map->osd_primary_affinity[i] =
1255	CEPH_OSD_DEFAULT_PRIMARY_AFFINITY;
1256	}
1257
1258	map->max_osd = max;
1259
1260	return 0;
1261	}
1262
1263	static int osdmap_set_crush(struct ceph_osdmap map, struct crush_map crush)
1264	{
1265	struct crush_work *work;
1266
1267	if (IS_ERR(crush))
1268	return PTR_ERR(crush);
1269
1270	work = alloc_workspace(crush);
1271	if (!work) {
1272	crush_destroy(crush);
1273	return -ENOMEM;
1274	}
1275
1276	if (map->crush)
1277	crush_destroy(map->crush);
1278	cleanup_workspace_manager(&map->crush_wsm);
1279	map->crush = crush;
1280	add_initial_workspace(&map->crush_wsm, work);
1281	return 0;
1282	}
1283
1284	#define OSDMAP_WRAPPER_COMPAT_VER 7
1285	#define OSDMAP_CLIENT_DATA_COMPAT_VER 1
1286
1287	/*
1288	* Return 0 or error. On success, *v is set to 0 for old (v6) osdmaps,
1289	* to struct_v of the client_data section for new (v7 and above)
1290	* osdmaps.
1291	*/
1292	static int get_osdmap_client_data_v(void *p, void end,
1293	const char prefix, u8 v)
1294	{
1295	u8 struct_v;
1296
1297	ceph_decode_8_safe(p, end, struct_v, e_inval);
1298	if (struct_v >= 7) {
1299	u8 struct_compat;
1300
1301	ceph_decode_8_safe(p, end, struct_compat, e_inval);
1302	if (struct_compat > OSDMAP_WRAPPER_COMPAT_VER) {
1303	pr_warn("got v %d cv %d > %d of %s ceph_osdmap\n",
1304	struct_v, struct_compat,
1305	OSDMAP_WRAPPER_COMPAT_VER, prefix);
1306	return -EINVAL;
1307	}
1308	p += 4; / ignore wrapper struct_len */
1309
1310	ceph_decode_8_safe(p, end, struct_v, e_inval);
1311	ceph_decode_8_safe(p, end, struct_compat, e_inval);
1312	if (struct_compat > OSDMAP_CLIENT_DATA_COMPAT_VER) {
1313	pr_warn("got v %d cv %d > %d of %s ceph_osdmap client data\n",
1314	struct_v, struct_compat,
1315	OSDMAP_CLIENT_DATA_COMPAT_VER, prefix);
1316	return -EINVAL;
1317	}
1318	p += 4; / ignore client data struct_len */
1319	} else {
1320	u16 version;
1321
1322	*p -= 1;
1323	ceph_decode_16_safe(p, end, version, e_inval);
1324	if (version < 6) {
1325	pr_warn("got v %d < 6 of %s ceph_osdmap\n",
1326	version, prefix);
1327	return -EINVAL;
1328	}
1329
1330	/* old osdmap encoding */
1331	struct_v = 0;
1332	}
1333
1334	*v = struct_v;
1335	return 0;
1336
1337	e_inval:
1338	return -EINVAL;
1339	}
1340
1341	static int __decode_pools(void *p, void end, struct ceph_osdmap *map,
1342	bool incremental)
1343	{
1344	u32 n;
1345
1346	ceph_decode_32_safe(p, end, n, e_inval);
1347	while (n--) {
1348	struct ceph_pg_pool_info *pi;
1349	u64 pool;
1350	int ret;
1351
1352	ceph_decode_64_safe(p, end, pool, e_inval);
1353
1354	pi = lookup_pg_pool(&map->pg_pools, pool);
1355	if (!incremental \|\| !pi) {
1356	pi = kzalloc_obj(*pi, GFP_NOFS);
1357	if (!pi)
1358	return -ENOMEM;
1359
1360	RB_CLEAR_NODE(&pi->node);
1361	pi->id = pool;
1362
1363	if (!__insert_pg_pool(&map->pg_pools, pi)) {
1364	kfree(pi);
1365	return -EEXIST;
1366	}
1367	}
1368
1369	ret = decode_pool(p, end, pi);
1370	if (ret)
1371	return ret;
1372	}
1373
1374	return 0;
1375
1376	e_inval:
1377	return -EINVAL;
1378	}
1379
1380	static int decode_pools(void *p, void end, struct ceph_osdmap *map)
1381	{
1382	return __decode_pools(p, end, map, false);
1383	}
1384
1385	static int decode_new_pools(void *p, void end, struct ceph_osdmap *map)
1386	{
1387	return __decode_pools(p, end, map, true);
1388	}
1389
1390	typedef struct ceph_pg_mapping (decode_mapping_fn_t)(void *, void , bool);
1391
1392	static int decode_pg_mapping(void *p, void end, struct rb_root *mapping_root,
1393	decode_mapping_fn_t fn, bool incremental)
1394	{
1395	u32 n;
1396
1397	WARN_ON(!incremental && !fn);
1398
1399	ceph_decode_32_safe(p, end, n, e_inval);
1400	while (n--) {
1401	struct ceph_pg_mapping *pg;
1402	struct ceph_pg pgid;
1403	int ret;
1404
1405	ret = ceph_decode_pgid(p, end, &pgid);
1406	if (ret)
1407	return ret;
1408
1409	pg = lookup_pg_mapping(mapping_root, &pgid);
1410	if (pg) {
1411	WARN_ON(!incremental);
1412	erase_pg_mapping(mapping_root, pg);
1413	free_pg_mapping(pg);
1414	}
1415
1416	if (fn) {
1417	pg = fn(p, end, incremental);
1418	if (IS_ERR(pg))
1419	return PTR_ERR(pg);
1420
1421	if (pg) {
1422	pg->pgid = pgid; /* struct */
1423	insert_pg_mapping(mapping_root, pg);
1424	}
1425	}
1426	}
1427
1428	return 0;
1429
1430	e_inval:
1431	return -EINVAL;
1432	}
1433
1434	static struct ceph_pg_mapping __decode_pg_temp(void p, void end,
1435	bool incremental)
1436	{
1437	struct ceph_pg_mapping *pg;
1438	u32 len, i;
1439
1440	ceph_decode_32_safe(p, end, len, e_inval);
1441	if (len == 0 && incremental)
1442	return NULL; /* new_pg_temp: [] to remove */
1443	if (len > CEPH_PG_MAX_SIZE)
1444	return ERR_PTR(-EINVAL);
1445
1446	ceph_decode_need(p, end, len * sizeof(u32), e_inval);
1447	pg = alloc_pg_mapping(len * sizeof(u32));
1448	if (!pg)
1449	return ERR_PTR(-ENOMEM);
1450
1451	pg->pg_temp.len = len;
1452	for (i = 0; i < len; i++)
1453	pg->pg_temp.osds[i] = ceph_decode_32(p);
1454
1455	return pg;
1456
1457	e_inval:
1458	return ERR_PTR(-EINVAL);
1459	}
1460
1461	static int decode_pg_temp(void *p, void end, struct ceph_osdmap *map)
1462	{
1463	return decode_pg_mapping(p, end, &map->pg_temp, __decode_pg_temp,
1464	false);
1465	}
1466
1467	static int decode_new_pg_temp(void *p, void end, struct ceph_osdmap *map)
1468	{
1469	return decode_pg_mapping(p, end, &map->pg_temp, __decode_pg_temp,
1470	true);
1471	}
1472
1473	static struct ceph_pg_mapping __decode_primary_temp(void p, void end,
1474	bool incremental)
1475	{
1476	struct ceph_pg_mapping *pg;
1477	u32 osd;
1478
1479	ceph_decode_32_safe(p, end, osd, e_inval);
1480	if (osd == (u32)-1 && incremental)
1481	return NULL; /* new_primary_temp: -1 to remove */
1482
1483	pg = alloc_pg_mapping(0);
1484	if (!pg)
1485	return ERR_PTR(-ENOMEM);
1486
1487	pg->primary_temp.osd = osd;
1488	return pg;
1489
1490	e_inval:
1491	return ERR_PTR(-EINVAL);
1492	}
1493
1494	static int decode_primary_temp(void *p, void end, struct ceph_osdmap *map)
1495	{
1496	return decode_pg_mapping(p, end, &map->primary_temp,
1497	__decode_primary_temp, false);
1498	}
1499
1500	static int decode_new_primary_temp(void *p, void end,
1501	struct ceph_osdmap *map)
1502	{
1503	return decode_pg_mapping(p, end, &map->primary_temp,
1504	__decode_primary_temp, true);
1505	}
1506
1507	u32 ceph_get_primary_affinity(struct ceph_osdmap *map, int osd)
1508	{
1509	if (!map->osd_primary_affinity)
1510	return CEPH_OSD_DEFAULT_PRIMARY_AFFINITY;
1511
1512	return map->osd_primary_affinity[osd];
1513	}
1514
1515	static int set_primary_affinity(struct ceph_osdmap *map, int osd, u32 aff)
1516	{
1517	if (!map->osd_primary_affinity) {
1518	int i;
1519
1520	map->osd_primary_affinity = kvmalloc(
1521	array_size(map->max_osd, sizeof(*map->osd_primary_affinity)),
1522	GFP_NOFS);
1523	if (!map->osd_primary_affinity)
1524	return -ENOMEM;
1525
1526	for (i = 0; i < map->max_osd; i++)
1527	map->osd_primary_affinity[i] =
1528	CEPH_OSD_DEFAULT_PRIMARY_AFFINITY;
1529	}
1530
1531	map->osd_primary_affinity[osd] = aff;
1532
1533	return 0;
1534	}
1535
1536	static int decode_primary_affinity(void *p, void end,
1537	struct ceph_osdmap *map)
1538	{
1539	u32 len, i;
1540
1541	ceph_decode_32_safe(p, end, len, e_inval);
1542	if (len == 0) {
1543	kvfree(map->osd_primary_affinity);
1544	map->osd_primary_affinity = NULL;
1545	return 0;
1546	}
1547	if (len != map->max_osd)
1548	goto e_inval;
1549
1550	ceph_decode_need(p, end, map->max_osd*sizeof(u32), e_inval);
1551
1552	for (i = 0; i < map->max_osd; i++) {
1553	int ret;
1554
1555	ret = set_primary_affinity(map, i, ceph_decode_32(p));
1556	if (ret)
1557	return ret;
1558	}
1559
1560	return 0;
1561
1562	e_inval:
1563	return -EINVAL;
1564	}
1565
1566	static int decode_new_primary_affinity(void *p, void end,
1567	struct ceph_osdmap *map)
1568	{
1569	u32 n;
1570
1571	ceph_decode_32_safe(p, end, n, e_inval);
1572	while (n--) {
1573	u32 osd, aff;
1574	int ret;
1575
1576	ceph_decode_32_safe(p, end, osd, e_inval);
1577	ceph_decode_32_safe(p, end, aff, e_inval);
1578	if (osd >= map->max_osd)
1579	goto e_inval;
1580
1581	ret = set_primary_affinity(map, osd, aff);
1582	if (ret)
1583	return ret;
1584
1585	osdmap_info(map, "osd%d primary-affinity 0x%x\n", osd, aff);
1586	}
1587
1588	return 0;
1589
1590	e_inval:
1591	return -EINVAL;
1592	}
1593
1594	static struct ceph_pg_mapping __decode_pg_upmap(void p, void end,
1595	bool __unused)
1596	{
1597	return __decode_pg_temp(p, end, false);
1598	}
1599
1600	static int decode_pg_upmap(void *p, void end, struct ceph_osdmap *map)
1601	{
1602	return decode_pg_mapping(p, end, &map->pg_upmap, __decode_pg_upmap,
1603	false);
1604	}
1605
1606	static int decode_new_pg_upmap(void *p, void end, struct ceph_osdmap *map)
1607	{
1608	return decode_pg_mapping(p, end, &map->pg_upmap, __decode_pg_upmap,
1609	true);
1610	}
1611
1612	static int decode_old_pg_upmap(void *p, void end, struct ceph_osdmap *map)
1613	{
1614	return decode_pg_mapping(p, end, &map->pg_upmap, NULL, true);
1615	}
1616
1617	static struct ceph_pg_mapping __decode_pg_upmap_items(void p, void end,
1618	bool __unused)
1619	{
1620	struct ceph_pg_mapping *pg;
1621	u32 len, i;
1622
1623	ceph_decode_32_safe(p, end, len, e_inval);
1624	if (len > CEPH_PG_MAX_SIZE)
1625	return ERR_PTR(-EINVAL);
1626
1627	ceph_decode_need(p, end, 2 * len * sizeof(u32), e_inval);
1628	pg = alloc_pg_mapping(2 * len * sizeof(u32));
1629	if (!pg)
1630	return ERR_PTR(-ENOMEM);
1631
1632	pg->pg_upmap_items.len = len;
1633	for (i = 0; i < len; i++) {
1634	pg->pg_upmap_items.from_to[i][0] = ceph_decode_32(p);
1635	pg->pg_upmap_items.from_to[i][1] = ceph_decode_32(p);
1636	}
1637
1638	return pg;
1639
1640	e_inval:
1641	return ERR_PTR(-EINVAL);
1642	}
1643
1644	static int decode_pg_upmap_items(void *p, void end, struct ceph_osdmap *map)
1645	{
1646	return decode_pg_mapping(p, end, &map->pg_upmap_items,
1647	__decode_pg_upmap_items, false);
1648	}
1649
1650	static int decode_new_pg_upmap_items(void *p, void end,
1651	struct ceph_osdmap *map)
1652	{
1653	return decode_pg_mapping(p, end, &map->pg_upmap_items,
1654	__decode_pg_upmap_items, true);
1655	}
1656
1657	static int decode_old_pg_upmap_items(void *p, void end,
1658	struct ceph_osdmap *map)
1659	{
1660	return decode_pg_mapping(p, end, &map->pg_upmap_items, NULL, true);
1661	}
1662
1663	/*
1664	* decode a full map.
1665	*/
1666	static int osdmap_decode(void *p, void end, bool msgr2,
1667	struct ceph_osdmap *map)
1668	{
1669	u8 struct_v;
1670	u32 epoch = 0;
1671	void start = p;
1672	u32 max;
1673	u32 len, i;
1674	int err;
1675
1676	dout("%s %p to %p len %d\n", __func__, p, end, (int)(end - p));
1677
1678	err = get_osdmap_client_data_v(p, end, "full", &struct_v);
1679	if (err)
1680	goto bad;
1681
1682	/* fsid, epoch, created, modified */
1683	ceph_decode_need(p, end, sizeof(map->fsid) + sizeof(u32) +
1684	sizeof(map->created) + sizeof(map->modified), e_inval);
1685	ceph_decode_copy(p, &map->fsid, sizeof(map->fsid));
1686	epoch = map->epoch = ceph_decode_32(p);
1687	ceph_decode_copy(p, &map->created, sizeof(map->created));
1688	ceph_decode_copy(p, &map->modified, sizeof(map->modified));
1689
1690	/* pools */
1691	err = decode_pools(p, end, map);
1692	if (err)
1693	goto bad;
1694
1695	/* pool_name */
1696	err = decode_pool_names(p, end, map);
1697	if (err)
1698	goto bad;
1699
1700	ceph_decode_32_safe(p, end, map->pool_max, e_inval);
1701
1702	ceph_decode_32_safe(p, end, map->flags, e_inval);
1703
1704	/* max_osd */
1705	ceph_decode_32_safe(p, end, max, e_inval);
1706
1707	/* (re)alloc osd arrays */
1708	err = osdmap_set_max_osd(map, max);
1709	if (err)
1710	goto bad;
1711
1712	/* osd_state, osd_weight, osd_addrs->client_addr */
1713	ceph_decode_need(p, end, 3*sizeof(u32) +
1714	map->max_osd*(struct_v >= 5 ? sizeof(u32) :
1715	sizeof(u8)) +
1716	map->max_osdsizeof(map->osd_weight), e_inval);
1717	if (ceph_decode_32(p) != map->max_osd)
1718	goto e_inval;
1719
1720	if (struct_v >= 5) {
1721	for (i = 0; i < map->max_osd; i++)
1722	map->osd_state[i] = ceph_decode_32(p);
1723	} else {
1724	for (i = 0; i < map->max_osd; i++)
1725	map->osd_state[i] = ceph_decode_8(p);
1726	}
1727
1728	if (ceph_decode_32(p) != map->max_osd)
1729	goto e_inval;
1730
1731	for (i = 0; i < map->max_osd; i++)
1732	map->osd_weight[i] = ceph_decode_32(p);
1733
1734	if (ceph_decode_32(p) != map->max_osd)
1735	goto e_inval;
1736
1737	for (i = 0; i < map->max_osd; i++) {
1738	struct ceph_entity_addr *addr = &map->osd_addr[i];
1739
1740	if (struct_v >= 8)
1741	err = ceph_decode_entity_addrvec(p, end, msgr2, addr);
1742	else
1743	err = ceph_decode_entity_addr(p, end, addr);
1744	if (err)
1745	goto bad;
1746
1747	dout("%s osd%d addr %s\n", __func__, i, ceph_pr_addr(addr));
1748	}
1749
1750	/* pg_temp */
1751	err = decode_pg_temp(p, end, map);
1752	if (err)
1753	goto bad;
1754
1755	/* primary_temp */
1756	if (struct_v >= 1) {
1757	err = decode_primary_temp(p, end, map);
1758	if (err)
1759	goto bad;
1760	}
1761
1762	/* primary_affinity */
1763	if (struct_v >= 2) {
1764	err = decode_primary_affinity(p, end, map);
1765	if (err)
1766	goto bad;
1767	} else {
1768	WARN_ON(map->osd_primary_affinity);
1769	}
1770
1771	/* crush */
1772	ceph_decode_32_safe(p, end, len, e_inval);
1773	err = osdmap_set_crush(map, crush_decode(p, min(p + len, end)));
1774	if (err)
1775	goto bad;
1776
1777	*p += len;
1778	if (struct_v >= 3) {
1779	/* erasure_code_profiles */
1780	ceph_decode_skip_map_of_map(p, end, string, string, string,
1781	e_inval);
1782	}
1783
1784	if (struct_v >= 4) {
1785	err = decode_pg_upmap(p, end, map);
1786	if (err)
1787	goto bad;
1788
1789	err = decode_pg_upmap_items(p, end, map);
1790	if (err)
1791	goto bad;
1792	} else {
1793	WARN_ON(!RB_EMPTY_ROOT(&map->pg_upmap));
1794	WARN_ON(!RB_EMPTY_ROOT(&map->pg_upmap_items));
1795	}
1796
1797	/* ignore the rest */
1798	*p = end;
1799
1800	dout("full osdmap epoch %d max_osd %d\n", map->epoch, map->max_osd);
1801	return 0;
1802
1803	e_inval:
1804	err = -EINVAL;
1805	bad:
1806	pr_err("corrupt full osdmap (%d) epoch %d off %d (%p of %p-%p)\n",
1807	err, epoch, (int)(p - start), p, start, end);
1808	print_hex_dump(KERN_DEBUG, "osdmap: ",
1809	DUMP_PREFIX_OFFSET, 16, 1,
1810	start, end - start, true);
1811	return err;
1812	}
1813
1814	/*
1815	* Allocate and decode a full map.
1816	*/
1817	struct ceph_osdmap ceph_osdmap_decode(void p, void end, bool msgr2)
1818	{
1819	struct ceph_osdmap *map;
1820	int ret;
1821
1822	map = ceph_osdmap_alloc();
1823	if (!map)
1824	return ERR_PTR(-ENOMEM);
1825
1826	ret = osdmap_decode(p, end, msgr2, map);
1827	if (ret) {
1828	ceph_osdmap_destroy(map);
1829	return ERR_PTR(ret);
1830	}
1831
1832	return map;
1833	}
1834
1835	/*
1836	* Encoding order is (new_up_client, new_state, new_weight). Need to
1837	* apply in the (new_weight, new_state, new_up_client) order, because
1838	* an incremental map may look like e.g.
1839	*
1840	* new_up_client: { osd=6, addr=... } # set osd_state and addr
1841	* new_state: { osd=6, xorstate=EXISTS } # clear osd_state
1842	*/
1843	static int decode_new_up_state_weight(void *p, void end, u8 struct_v,
1844	bool msgr2, struct ceph_osdmap *map)
1845	{
1846	void *new_up_client;
1847	void *new_state;
1848	void *new_weight_end;
1849	const u32 new_state_item_size =
1850	sizeof(u32) + (struct_v >= 5 ? sizeof(u32) : sizeof(u8));
1851	u32 len;
1852	int ret;
1853	int i;
1854
1855	new_up_client = *p;
1856	ceph_decode_32_safe(p, end, len, e_inval);
1857	for (i = 0; i < len; ++i) {
1858	struct ceph_entity_addr addr;
1859
1860	ceph_decode_skip_32(p, end, e_inval);
1861	if (struct_v >= 7)
1862	ret = ceph_decode_entity_addrvec(p, end, msgr2, &addr);
1863	else
1864	ret = ceph_decode_entity_addr(p, end, &addr);
1865	if (ret)
1866	return ret;
1867	}
1868
1869	new_state = *p;
1870	ceph_decode_32_safe(p, end, len, e_inval);
1871	if (check_mul_overflow(len, new_state_item_size, &len))
1872	goto e_inval;
1873	ceph_decode_need(p, end, len, e_inval);
1874	*p += len;
1875
1876	/* new_weight */
1877	ceph_decode_32_safe(p, end, len, e_inval);
1878	while (len--) {
1879	s32 osd;
1880	u32 w;
1881
1882	ceph_decode_need(p, end, 2*sizeof(u32), e_inval);
1883	osd = ceph_decode_32(p);
1884	w = ceph_decode_32(p);
1885	if (osd >= map->max_osd)
1886	goto e_inval;
1887
1888	osdmap_info(map, "osd%d weight 0x%x %s\n", osd, w,
1889	w == CEPH_OSD_IN ? "(in)" :
1890	(w == CEPH_OSD_OUT ? "(out)" : ""));
1891	map->osd_weight[osd] = w;
1892
1893	/*
1894	* If we are marking in, set the EXISTS, and clear the
1895	* AUTOOUT and NEW bits.
1896	*/
1897	if (w) {
1898	map->osd_state[osd] \|= CEPH_OSD_EXISTS;
1899	map->osd_state[osd] &= ~(CEPH_OSD_AUTOOUT \|
1900	CEPH_OSD_NEW);
1901	}
1902	}
1903	new_weight_end = *p;
1904
1905	/* new_state (up/down) */
1906	*p = new_state;
1907	len = ceph_decode_32(p);
1908	while (len--) {
1909	s32 osd;
1910	u32 xorstate;
1911
1912	osd = ceph_decode_32(p);
1913	if (osd >= map->max_osd)
1914	goto e_inval;
1915
1916	if (struct_v >= 5)
1917	xorstate = ceph_decode_32(p);
1918	else
1919	xorstate = ceph_decode_8(p);
1920	if (xorstate == 0)
1921	xorstate = CEPH_OSD_UP;
1922	if ((map->osd_state[osd] & CEPH_OSD_UP) &&
1923	(xorstate & CEPH_OSD_UP))
1924	osdmap_info(map, "osd%d down\n", osd);
1925	if ((map->osd_state[osd] & CEPH_OSD_EXISTS) &&
1926	(xorstate & CEPH_OSD_EXISTS)) {
1927	osdmap_info(map, "osd%d does not exist\n", osd);
1928	ret = set_primary_affinity(map, osd,
1929	CEPH_OSD_DEFAULT_PRIMARY_AFFINITY);
1930	if (ret)
1931	return ret;
1932	memset(map->osd_addr + osd, 0, sizeof(*map->osd_addr));
1933	map->osd_state[osd] = 0;
1934	} else {
1935	map->osd_state[osd] ^= xorstate;
1936	}
1937	}
1938
1939	/* new_up_client */
1940	*p = new_up_client;
1941	len = ceph_decode_32(p);
1942	while (len--) {
1943	s32 osd;
1944	struct ceph_entity_addr addr;
1945
1946	osd = ceph_decode_32(p);
1947	if (osd >= map->max_osd)
1948	goto e_inval;
1949
1950	if (struct_v >= 7)
1951	ret = ceph_decode_entity_addrvec(p, end, msgr2, &addr);
1952	else
1953	ret = ceph_decode_entity_addr(p, end, &addr);
1954	if (ret)
1955	return ret;
1956
1957	dout("%s osd%d addr %s\n", __func__, osd, ceph_pr_addr(&addr));
1958
1959	osdmap_info(map, "osd%d up\n", osd);
1960	map->osd_state[osd] \|= CEPH_OSD_EXISTS \| CEPH_OSD_UP;
1961	map->osd_addr[osd] = addr;
1962	}
1963
1964	*p = new_weight_end;
1965	return 0;
1966
1967	e_inval:
1968	return -EINVAL;
1969	}
1970
1971	/*
1972	* decode and apply an incremental map update.
1973	*/
1974	struct ceph_osdmap osdmap_apply_incremental(void p, void end, bool msgr2,
1975	struct ceph_osdmap *map)
1976	{
1977	struct ceph_fsid fsid;
1978	u32 epoch = 0;
1979	struct ceph_timespec modified;
1980	s32 len;
1981	u64 pool;
1982	__s64 new_pool_max;
1983	__s32 new_flags, max;
1984	void start = p;
1985	int err;
1986	u8 struct_v;
1987
1988	dout("%s %p to %p len %d\n", __func__, p, end, (int)(end - p));
1989
1990	err = get_osdmap_client_data_v(p, end, "inc", &struct_v);
1991	if (err)
1992	goto bad;
1993
1994	/* fsid, epoch, modified, new_pool_max, new_flags */
1995	ceph_decode_need(p, end, sizeof(fsid) + sizeof(u32) + sizeof(modified) +
1996	sizeof(u64) + sizeof(u32), e_inval);
1997	ceph_decode_copy(p, &fsid, sizeof(fsid));
1998	epoch = ceph_decode_32(p);
1999	ceph_decode_copy(p, &modified, sizeof(modified));
2000	new_pool_max = ceph_decode_64(p);
2001	new_flags = ceph_decode_32(p);
2002
2003	if (epoch != map->epoch + 1)
2004	goto e_inval;
2005
2006	/* full map? */
2007	ceph_decode_32_safe(p, end, len, e_inval);
2008	if (len > 0) {
2009	dout("apply_incremental full map len %d, %p to %p\n",
2010	len, *p, end);
2011	return ceph_osdmap_decode(p, min(*p+len, end), msgr2);
2012	}
2013
2014	/* new crush? */
2015	ceph_decode_32_safe(p, end, len, e_inval);
2016	if (len > 0) {
2017	err = osdmap_set_crush(map,
2018	crush_decode(p, min(p + len, end)));
2019	if (err)
2020	goto bad;
2021	*p += len;
2022	}
2023
2024	/* new flags? */
2025	if (new_flags >= 0)
2026	map->flags = new_flags;
2027	if (new_pool_max >= 0)
2028	map->pool_max = new_pool_max;
2029
2030	/* new max? */
2031	ceph_decode_32_safe(p, end, max, e_inval);
2032	if (max >= 0) {
2033	err = osdmap_set_max_osd(map, max);
2034	if (err)
2035	goto bad;
2036	}
2037
2038	map->epoch++;
2039	map->modified = modified;
2040
2041	/* new_pools */
2042	err = decode_new_pools(p, end, map);
2043	if (err)
2044	goto bad;
2045
2046	/* new_pool_names */
2047	err = decode_pool_names(p, end, map);
2048	if (err)
2049	goto bad;
2050
2051	/* old_pool */
2052	ceph_decode_32_safe(p, end, len, e_inval);
2053	while (len--) {
2054	struct ceph_pg_pool_info *pi;
2055
2056	ceph_decode_64_safe(p, end, pool, e_inval);
2057	pi = lookup_pg_pool(&map->pg_pools, pool);
2058	if (pi)
2059	__remove_pg_pool(&map->pg_pools, pi);
2060	}
2061
2062	/* new_up_client, new_state, new_weight */
2063	err = decode_new_up_state_weight(p, end, struct_v, msgr2, map);
2064	if (err)
2065	goto bad;
2066
2067	/* new_pg_temp */
2068	err = decode_new_pg_temp(p, end, map);
2069	if (err)
2070	goto bad;
2071
2072	/* new_primary_temp */
2073	if (struct_v >= 1) {
2074	err = decode_new_primary_temp(p, end, map);
2075	if (err)
2076	goto bad;
2077	}
2078
2079	/* new_primary_affinity */
2080	if (struct_v >= 2) {
2081	err = decode_new_primary_affinity(p, end, map);
2082	if (err)
2083	goto bad;
2084	}
2085
2086	if (struct_v >= 3) {
2087	/* new_erasure_code_profiles */
2088	ceph_decode_skip_map_of_map(p, end, string, string, string,
2089	e_inval);
2090	/* old_erasure_code_profiles */
2091	ceph_decode_skip_set(p, end, string, e_inval);
2092	}
2093
2094	if (struct_v >= 4) {
2095	err = decode_new_pg_upmap(p, end, map);
2096	if (err)
2097	goto bad;
2098
2099	err = decode_old_pg_upmap(p, end, map);
2100	if (err)
2101	goto bad;
2102
2103	err = decode_new_pg_upmap_items(p, end, map);
2104	if (err)
2105	goto bad;
2106
2107	err = decode_old_pg_upmap_items(p, end, map);
2108	if (err)
2109	goto bad;
2110	}
2111
2112	/* ignore the rest */
2113	*p = end;
2114
2115	dout("inc osdmap epoch %d max_osd %d\n", map->epoch, map->max_osd);
2116	return map;
2117
2118	e_inval:
2119	err = -EINVAL;
2120	bad:
2121	pr_err("corrupt inc osdmap (%d) epoch %d off %d (%p of %p-%p)\n",
2122	err, epoch, (int)(p - start), p, start, end);
2123	print_hex_dump(KERN_DEBUG, "osdmap: ",
2124	DUMP_PREFIX_OFFSET, 16, 1,
2125	start, end - start, true);
2126	return ERR_PTR(err);
2127	}
2128
2129	void ceph_oloc_copy(struct ceph_object_locator *dest,
2130	const struct ceph_object_locator *src)
2131	{
2132	ceph_oloc_destroy(dest);
2133
2134	dest->pool = src->pool;
2135	if (src->pool_ns)
2136	dest->pool_ns = ceph_get_string(src->pool_ns);
2137	else
2138	dest->pool_ns = NULL;
2139	}
2140	EXPORT_SYMBOL(ceph_oloc_copy);
2141
2142	void ceph_oloc_destroy(struct ceph_object_locator *oloc)
2143	{
2144	ceph_put_string(oloc->pool_ns);
2145	}
2146	EXPORT_SYMBOL(ceph_oloc_destroy);
2147
2148	void ceph_oid_copy(struct ceph_object_id *dest,
2149	const struct ceph_object_id *src)
2150	{
2151	ceph_oid_destroy(dest);
2152
2153	if (src->name != src->inline_name) {
2154	/* very rare, see ceph_object_id definition */
2155	dest->name = kmalloc(src->name_len + 1,
2156	GFP_NOIO \| __GFP_NOFAIL);
2157	} else {
2158	dest->name = dest->inline_name;
2159	}
2160	memcpy(dest->name, src->name, src->name_len + 1);
2161	dest->name_len = src->name_len;
2162	}
2163	EXPORT_SYMBOL(ceph_oid_copy);
2164
2165	static __printf(2, 0)
2166	int oid_printf_vargs(struct ceph_object_id oid, const char fmt, va_list ap)
2167	{
2168	int len;
2169
2170	WARN_ON(!ceph_oid_empty(oid));
2171
2172	len = vsnprintf(oid->inline_name, sizeof(oid->inline_name), fmt, ap);
2173	if (len >= sizeof(oid->inline_name))
2174	return len;
2175
2176	oid->name_len = len;
2177	return 0;
2178	}
2179
2180	/*
2181	* If oid doesn't fit into inline buffer, BUG.
2182	*/
2183	void ceph_oid_printf(struct ceph_object_id oid, const char fmt, ...)
2184	{
2185	va_list ap;
2186
2187	va_start(ap, fmt);
2188	BUG_ON(oid_printf_vargs(oid, fmt, ap));
2189	va_end(ap);
2190	}
2191	EXPORT_SYMBOL(ceph_oid_printf);
2192
2193	static __printf(3, 0)
2194	int oid_aprintf_vargs(struct ceph_object_id *oid, gfp_t gfp,
2195	const char *fmt, va_list ap)
2196	{
2197	va_list aq;
2198	int len;
2199
2200	va_copy(aq, ap);
2201	len = oid_printf_vargs(oid, fmt, aq);
2202	va_end(aq);
2203
2204	if (len) {
2205	char *external_name;
2206
2207	external_name = kmalloc(len + 1, gfp);
2208	if (!external_name)
2209	return -ENOMEM;
2210
2211	oid->name = external_name;
2212	WARN_ON(vsnprintf(oid->name, len + 1, fmt, ap) != len);
2213	oid->name_len = len;
2214	}
2215
2216	return 0;
2217	}
2218
2219	/*
2220	* If oid doesn't fit into inline buffer, allocate.
2221	*/
2222	int ceph_oid_aprintf(struct ceph_object_id *oid, gfp_t gfp,
2223	const char *fmt, ...)
2224	{
2225	va_list ap;
2226	int ret;
2227
2228	va_start(ap, fmt);
2229	ret = oid_aprintf_vargs(oid, gfp, fmt, ap);
2230	va_end(ap);
2231
2232	return ret;
2233	}
2234	EXPORT_SYMBOL(ceph_oid_aprintf);
2235
2236	void ceph_oid_destroy(struct ceph_object_id *oid)
2237	{
2238	if (oid->name != oid->inline_name)
2239	kfree(oid->name);
2240	}
2241	EXPORT_SYMBOL(ceph_oid_destroy);
2242
2243	/*
2244	* osds only
2245	*/
2246	static bool __osds_equal(const struct ceph_osds *lhs,
2247	const struct ceph_osds *rhs)
2248	{
2249	if (lhs->size == rhs->size &&
2250	!memcmp(lhs->osds, rhs->osds, rhs->size * sizeof(rhs->osds[0])))
2251	return true;
2252
2253	return false;
2254	}
2255
2256	/*
2257	* osds + primary
2258	*/
2259	static bool osds_equal(const struct ceph_osds *lhs,
2260	const struct ceph_osds *rhs)
2261	{
2262	if (__osds_equal(lhs, rhs) &&
2263	lhs->primary == rhs->primary)
2264	return true;
2265
2266	return false;
2267	}
2268
2269	static bool osds_valid(const struct ceph_osds *set)
2270	{
2271	/* non-empty set */
2272	if (set->size > 0 && set->primary >= 0)
2273	return true;
2274
2275	/* empty can_shift_osds set */
2276	if (!set->size && set->primary == -1)
2277	return true;
2278
2279	/* empty !can_shift_osds set - all NONE */
2280	if (set->size > 0 && set->primary == -1) {
2281	int i;
2282
2283	for (i = 0; i < set->size; i++) {
2284	if (set->osds[i] != CRUSH_ITEM_NONE)
2285	break;
2286	}
2287	if (i == set->size)
2288	return true;
2289	}
2290
2291	return false;
2292	}
2293
2294	void ceph_osds_copy(struct ceph_osds dest, const struct ceph_osds src)
2295	{
2296	memcpy(dest->osds, src->osds, src->size * sizeof(src->osds[0]));
2297	dest->size = src->size;
2298	dest->primary = src->primary;
2299	}
2300
2301	bool ceph_pg_is_split(const struct ceph_pg *pgid, u32 old_pg_num,
2302	u32 new_pg_num)
2303	{
2304	int old_bits = calc_bits_of(old_pg_num);
2305	int old_mask = (1 << old_bits) - 1;
2306	int n;
2307
2308	WARN_ON(pgid->seed >= old_pg_num);
2309	if (new_pg_num <= old_pg_num)
2310	return false;
2311
2312	for (n = 1; ; n++) {
2313	int next_bit = n << (old_bits - 1);
2314	u32 s = next_bit \| pgid->seed;
2315
2316	if (s < old_pg_num \|\| s == pgid->seed)
2317	continue;
2318	if (s >= new_pg_num)
2319	break;
2320
2321	s = ceph_stable_mod(s, old_pg_num, old_mask);
2322	if (s == pgid->seed)
2323	return true;
2324	}
2325
2326	return false;
2327	}
2328
2329	bool ceph_is_new_interval(const struct ceph_osds *old_acting,
2330	const struct ceph_osds *new_acting,
2331	const struct ceph_osds *old_up,
2332	const struct ceph_osds *new_up,
2333	int old_size,
2334	int new_size,
2335	int old_min_size,
2336	int new_min_size,
2337	u32 old_pg_num,
2338	u32 new_pg_num,
2339	bool old_sort_bitwise,
2340	bool new_sort_bitwise,
2341	bool old_recovery_deletes,
2342	bool new_recovery_deletes,
2343	const struct ceph_pg *pgid)
2344	{
2345	return !osds_equal(old_acting, new_acting) \|\|
2346	!osds_equal(old_up, new_up) \|\|
2347	old_size != new_size \|\|
2348	old_min_size != new_min_size \|\|
2349	ceph_pg_is_split(pgid, old_pg_num, new_pg_num) \|\|
2350	old_sort_bitwise != new_sort_bitwise \|\|
2351	old_recovery_deletes != new_recovery_deletes;
2352	}
2353
2354	static int calc_pg_rank(int osd, const struct ceph_osds *acting)
2355	{
2356	int i;
2357
2358	for (i = 0; i < acting->size; i++) {
2359	if (acting->osds[i] == osd)
2360	return i;
2361	}
2362
2363	return -1;
2364	}
2365
2366	static bool primary_changed(const struct ceph_osds *old_acting,
2367	const struct ceph_osds *new_acting)
2368	{
2369	if (!old_acting->size && !new_acting->size)
2370	return false; /* both still empty */
2371
2372	if (!old_acting->size ^ !new_acting->size)
2373	return true; /* was empty, now not, or vice versa */
2374
2375	if (old_acting->primary != new_acting->primary)
2376	return true; /* primary changed */
2377
2378	if (calc_pg_rank(old_acting->primary, old_acting) !=
2379	calc_pg_rank(new_acting->primary, new_acting))
2380	return true;
2381
2382	return false; /* same primary (tho replicas may have changed) */
2383	}
2384
2385	bool ceph_osds_changed(const struct ceph_osds *old_acting,
2386	const struct ceph_osds *new_acting,
2387	bool any_change)
2388	{
2389	if (primary_changed(old_acting, new_acting))
2390	return true;
2391
2392	if (any_change && !__osds_equal(old_acting, new_acting))
2393	return true;
2394
2395	return false;
2396	}
2397
2398	/*
2399	* Map an object into a PG.
2400	*
2401	* Should only be called with target_oid and target_oloc (as opposed to
2402	* base_oid and base_oloc), since tiering isn't taken into account.
2403	*/
2404	void __ceph_object_locator_to_pg(struct ceph_pg_pool_info *pi,
2405	const struct ceph_object_id *oid,
2406	const struct ceph_object_locator *oloc,
2407	struct ceph_pg *raw_pgid)
2408	{
2409	WARN_ON(pi->id != oloc->pool);
2410
2411	if (!oloc->pool_ns) {
2412	raw_pgid->pool = oloc->pool;
2413	raw_pgid->seed = ceph_str_hash(pi->object_hash, oid->name,
2414	oid->name_len);
2415	dout("%s %s -> raw_pgid %llu.%x\n", __func__, oid->name,
2416	raw_pgid->pool, raw_pgid->seed);
2417	} else {
2418	char stack_buf[256];
2419	char *buf = stack_buf;
2420	int nsl = oloc->pool_ns->len;
2421	size_t total = nsl + 1 + oid->name_len;
2422
2423	if (total > sizeof(stack_buf))
2424	buf = kmalloc(total, GFP_NOIO \| __GFP_NOFAIL);
2425	memcpy(buf, oloc->pool_ns->str, nsl);
2426	buf[nsl] = '\037';
2427	memcpy(buf + nsl + 1, oid->name, oid->name_len);
2428	raw_pgid->pool = oloc->pool;
2429	raw_pgid->seed = ceph_str_hash(pi->object_hash, buf, total);
2430	if (buf != stack_buf)
2431	kfree(buf);
2432	dout("%s %s ns %.*s -> raw_pgid %llu.%x\n", __func__,
2433	oid->name, nsl, oloc->pool_ns->str,
2434	raw_pgid->pool, raw_pgid->seed);
2435	}
2436	}
2437
2438	int ceph_object_locator_to_pg(struct ceph_osdmap *osdmap,
2439	const struct ceph_object_id *oid,
2440	const struct ceph_object_locator *oloc,
2441	struct ceph_pg *raw_pgid)
2442	{
2443	struct ceph_pg_pool_info *pi;
2444
2445	pi = ceph_pg_pool_by_id(osdmap, oloc->pool);
2446	if (!pi)
2447	return -ENOENT;
2448
2449	__ceph_object_locator_to_pg(pi, oid, oloc, raw_pgid);
2450	return 0;
2451	}
2452	EXPORT_SYMBOL(ceph_object_locator_to_pg);
2453
2454	/*
2455	* Map a raw PG (full precision ps) into an actual PG.
2456	*/
2457	static void raw_pg_to_pg(struct ceph_pg_pool_info *pi,
2458	const struct ceph_pg *raw_pgid,
2459	struct ceph_pg *pgid)
2460	{
2461	pgid->pool = raw_pgid->pool;
2462	pgid->seed = ceph_stable_mod(raw_pgid->seed, pi->pg_num,
2463	pi->pg_num_mask);
2464	}
2465
2466	/*
2467	* Map a raw PG (full precision ps) into a placement ps (placement
2468	* seed). Include pool id in that value so that different pools don't
2469	* use the same seeds.
2470	*/
2471	static u32 raw_pg_to_pps(struct ceph_pg_pool_info *pi,
2472	const struct ceph_pg *raw_pgid)
2473	{
2474	if (pi->flags & CEPH_POOL_FLAG_HASHPSPOOL) {
2475	/* hash pool id and seed so that pool PGs do not overlap */
2476	return crush_hash32_2(CRUSH_HASH_RJENKINS1,
2477	ceph_stable_mod(raw_pgid->seed,
2478	pi->pgp_num,
2479	pi->pgp_num_mask),
2480	raw_pgid->pool);
2481	} else {
2482	/*
2483	* legacy behavior: add ps and pool together. this is
2484	* not a great approach because the PGs from each pool
2485	* will overlap on top of each other: 0.5 == 1.4 ==
2486	* 2.3 == ...
2487	*/
2488	return ceph_stable_mod(raw_pgid->seed, pi->pgp_num,
2489	pi->pgp_num_mask) +
2490	(unsigned)raw_pgid->pool;
2491	}
2492	}
2493
2494	/*
2495	* Magic value used for a "default" fallback choose_args, used if the
2496	* crush_choose_arg_map passed to do_crush() does not exist. If this
2497	* also doesn't exist, fall back to canonical weights.
2498	*/
2499	#define CEPH_DEFAULT_CHOOSE_ARGS -1
2500
2501	static int do_crush(struct ceph_osdmap *map, int ruleno, int x,
2502	int *result, int result_max,
2503	const __u32 *weight, int weight_max,
2504	s64 choose_args_index)
2505	{
2506	struct crush_choose_arg_map *arg_map;
2507	struct crush_work *work;
2508	int r;
2509
2510	BUG_ON(result_max > CEPH_PG_MAX_SIZE);
2511
2512	arg_map = lookup_choose_arg_map(&map->crush->choose_args,
2513	choose_args_index);
2514	if (!arg_map)
2515	arg_map = lookup_choose_arg_map(&map->crush->choose_args,
2516	CEPH_DEFAULT_CHOOSE_ARGS);
2517
2518	work = get_workspace(&map->crush_wsm, map->crush);
2519	r = crush_do_rule(map->crush, ruleno, x, result, result_max,
2520	weight, weight_max, work,
2521	arg_map ? arg_map->args : NULL);
2522	put_workspace(&map->crush_wsm, work);
2523	return r;
2524	}
2525
2526	static void remove_nonexistent_osds(struct ceph_osdmap *osdmap,
2527	struct ceph_pg_pool_info *pi,
2528	struct ceph_osds *set)
2529	{
2530	int i;
2531
2532	if (ceph_can_shift_osds(pi)) {
2533	int removed = 0;
2534
2535	/* shift left */
2536	for (i = 0; i < set->size; i++) {
2537	if (!ceph_osd_exists(osdmap, set->osds[i])) {
2538	removed++;
2539	continue;
2540	}
2541	if (removed)
2542	set->osds[i - removed] = set->osds[i];
2543	}
2544	set->size -= removed;
2545	} else {
2546	/* set dne devices to NONE */
2547	for (i = 0; i < set->size; i++) {
2548	if (!ceph_osd_exists(osdmap, set->osds[i]))
2549	set->osds[i] = CRUSH_ITEM_NONE;
2550	}
2551	}
2552	}
2553
2554	/*
2555	* Calculate raw set (CRUSH output) for given PG and filter out
2556	* nonexistent OSDs. ->primary is undefined for a raw set.
2557	*
2558	* Placement seed (CRUSH input) is returned through @ppps.
2559	*/
2560	static void pg_to_raw_osds(struct ceph_osdmap *osdmap,
2561	struct ceph_pg_pool_info *pi,
2562	const struct ceph_pg *raw_pgid,
2563	struct ceph_osds *raw,
2564	u32 *ppps)
2565	{
2566	u32 pps = raw_pg_to_pps(pi, raw_pgid);
2567	int ruleno;
2568	int len;
2569
2570	ceph_osds_init(raw);
2571	if (ppps)
2572	*ppps = pps;
2573
2574	ruleno = crush_find_rule(osdmap->crush, pi->crush_ruleset, pi->type,
2575	pi->size);
2576	if (ruleno < 0) {
2577	pr_err("no crush rule: pool %lld ruleset %d type %d size %d\n",
2578	pi->id, pi->crush_ruleset, pi->type, pi->size);
2579	return;
2580	}
2581
2582	if (pi->size > ARRAY_SIZE(raw->osds)) {
2583	pr_err_ratelimited("pool %lld ruleset %d type %d too wide: size %d > %zu\n",
2584	pi->id, pi->crush_ruleset, pi->type, pi->size,
2585	ARRAY_SIZE(raw->osds));
2586	return;
2587	}
2588
2589	len = do_crush(osdmap, ruleno, pps, raw->osds, pi->size,
2590	osdmap->osd_weight, osdmap->max_osd, pi->id);
2591	if (len < 0) {
2592	pr_err("error %d from crush rule %d: pool %lld ruleset %d type %d size %d\n",
2593	len, ruleno, pi->id, pi->crush_ruleset, pi->type,
2594	pi->size);
2595	return;
2596	}
2597
2598	raw->size = len;
2599	remove_nonexistent_osds(osdmap, pi, raw);
2600	}
2601
2602	/* apply pg_upmap[_items] mappings */
2603	static void apply_upmap(struct ceph_osdmap *osdmap,
2604	const struct ceph_pg *pgid,
2605	struct ceph_osds *raw)
2606	{
2607	struct ceph_pg_mapping *pg;
2608	int i, j;
2609
2610	pg = lookup_pg_mapping(&osdmap->pg_upmap, pgid);
2611	if (pg) {
2612	/* make sure targets aren't marked out */
2613	for (i = 0; i < pg->pg_upmap.len; i++) {
2614	int osd = pg->pg_upmap.osds[i];
2615
2616	if (osd != CRUSH_ITEM_NONE &&
2617	osd < osdmap->max_osd &&
2618	osdmap->osd_weight[osd] == 0) {
2619	/* reject/ignore explicit mapping */
2620	return;
2621	}
2622	}
2623	for (i = 0; i < pg->pg_upmap.len; i++)
2624	raw->osds[i] = pg->pg_upmap.osds[i];
2625	raw->size = pg->pg_upmap.len;
2626	/* check and apply pg_upmap_items, if any */
2627	}
2628
2629	pg = lookup_pg_mapping(&osdmap->pg_upmap_items, pgid);
2630	if (pg) {
2631	/*
2632	* Note: this approach does not allow a bidirectional swap,
2633	* e.g., [[1,2],[2,1]] applied to [0,1,2] -> [0,2,1].
2634	*/
2635	for (i = 0; i < pg->pg_upmap_items.len; i++) {
2636	int from = pg->pg_upmap_items.from_to[i][0];
2637	int to = pg->pg_upmap_items.from_to[i][1];
2638	int pos = -1;
2639	bool exists = false;
2640
2641	/* make sure replacement doesn't already appear */
2642	for (j = 0; j < raw->size; j++) {
2643	int osd = raw->osds[j];
2644
2645	if (osd == to) {
2646	exists = true;
2647	break;
2648	}
2649	/* ignore mapping if target is marked out */
2650	if (osd == from && pos < 0 &&
2651	!(to != CRUSH_ITEM_NONE &&
2652	to < osdmap->max_osd &&
2653	osdmap->osd_weight[to] == 0)) {
2654	pos = j;
2655	}
2656	}
2657	if (!exists && pos >= 0)
2658	raw->osds[pos] = to;
2659	}
2660	}
2661	}
2662
2663	/*
2664	* Given raw set, calculate up set and up primary. By definition of an
2665	* up set, the result won't contain nonexistent or down OSDs.
2666	*
2667	* This is done in-place - on return @set is the up set. If it's
2668	* empty, ->primary will remain undefined.
2669	*/
2670	static void raw_to_up_osds(struct ceph_osdmap *osdmap,
2671	struct ceph_pg_pool_info *pi,
2672	struct ceph_osds *set)
2673	{
2674	int i;
2675
2676	/* ->primary is undefined for a raw set */
2677	BUG_ON(set->primary != -1);
2678
2679	if (ceph_can_shift_osds(pi)) {
2680	int removed = 0;
2681
2682	/* shift left */
2683	for (i = 0; i < set->size; i++) {
2684	if (ceph_osd_is_down(osdmap, set->osds[i])) {
2685	removed++;
2686	continue;
2687	}
2688	if (removed)
2689	set->osds[i - removed] = set->osds[i];
2690	}
2691	set->size -= removed;
2692	if (set->size > 0)
2693	set->primary = set->osds[0];
2694	} else {
2695	/* set down/dne devices to NONE */
2696	for (i = set->size - 1; i >= 0; i--) {
2697	if (ceph_osd_is_down(osdmap, set->osds[i]))
2698	set->osds[i] = CRUSH_ITEM_NONE;
2699	else
2700	set->primary = set->osds[i];
2701	}
2702	}
2703	}
2704
2705	static void apply_primary_affinity(struct ceph_osdmap *osdmap,
2706	struct ceph_pg_pool_info *pi,
2707	u32 pps,
2708	struct ceph_osds *up)
2709	{
2710	int i;
2711	int pos = -1;
2712
2713	/*
2714	* Do we have any non-default primary_affinity values for these
2715	* osds?
2716	*/
2717	if (!osdmap->osd_primary_affinity)
2718	return;
2719
2720	for (i = 0; i < up->size; i++) {
2721	int osd = up->osds[i];
2722
2723	if (osd != CRUSH_ITEM_NONE &&
2724	osdmap->osd_primary_affinity[osd] !=
2725	CEPH_OSD_DEFAULT_PRIMARY_AFFINITY) {
2726	break;
2727	}
2728	}
2729	if (i == up->size)
2730	return;
2731
2732	/*
2733	* Pick the primary. Feed both the seed (for the pg) and the
2734	* osd into the hash/rng so that a proportional fraction of an
2735	* osd's pgs get rejected as primary.
2736	*/
2737	for (i = 0; i < up->size; i++) {
2738	int osd = up->osds[i];
2739	u32 aff;
2740
2741	if (osd == CRUSH_ITEM_NONE)
2742	continue;
2743
2744	aff = osdmap->osd_primary_affinity[osd];
2745	if (aff < CEPH_OSD_MAX_PRIMARY_AFFINITY &&
2746	(crush_hash32_2(CRUSH_HASH_RJENKINS1,
2747	pps, osd) >> 16) >= aff) {
2748	/*
2749	* We chose not to use this primary. Note it
2750	* anyway as a fallback in case we don't pick
2751	* anyone else, but keep looking.
2752	*/
2753	if (pos < 0)
2754	pos = i;
2755	} else {
2756	pos = i;
2757	break;
2758	}
2759	}
2760	if (pos < 0)
2761	return;
2762
2763	up->primary = up->osds[pos];
2764
2765	if (ceph_can_shift_osds(pi) && pos > 0) {
2766	/* move the new primary to the front */
2767	for (i = pos; i > 0; i--)
2768	up->osds[i] = up->osds[i - 1];
2769	up->osds[0] = up->primary;
2770	}
2771	}
2772
2773	/*
2774	* Get pg_temp and primary_temp mappings for given PG.
2775	*
2776	* Note that a PG may have none, only pg_temp, only primary_temp or
2777	* both pg_temp and primary_temp mappings. This means @temp isn't
2778	* always a valid OSD set on return: in the "only primary_temp" case,
2779	* @temp will have its ->primary >= 0 but ->size == 0.
2780	*/
2781	static void get_temp_osds(struct ceph_osdmap *osdmap,
2782	struct ceph_pg_pool_info *pi,
2783	const struct ceph_pg *pgid,
2784	struct ceph_osds *temp)
2785	{
2786	struct ceph_pg_mapping *pg;
2787	int i;
2788
2789	ceph_osds_init(temp);
2790
2791	/* pg_temp? */
2792	pg = lookup_pg_mapping(&osdmap->pg_temp, pgid);
2793	if (pg) {
2794	for (i = 0; i < pg->pg_temp.len; i++) {
2795	if (ceph_osd_is_down(osdmap, pg->pg_temp.osds[i])) {
2796	if (ceph_can_shift_osds(pi))
2797	continue;
2798
2799	temp->osds[temp->size++] = CRUSH_ITEM_NONE;
2800	} else {
2801	temp->osds[temp->size++] = pg->pg_temp.osds[i];
2802	}
2803	}
2804
2805	/* apply pg_temp's primary */
2806	for (i = 0; i < temp->size; i++) {
2807	if (temp->osds[i] != CRUSH_ITEM_NONE) {
2808	temp->primary = temp->osds[i];
2809	break;
2810	}
2811	}
2812	}
2813
2814	/* primary_temp? (shouldn't ever be a nonexistent or down OSD) */
2815	pg = lookup_pg_mapping(&osdmap->primary_temp, pgid);
2816	if (pg && !WARN_ON_ONCE(ceph_osd_is_down(osdmap,
2817	pg->primary_temp.osd)))
2818	temp->primary = pg->primary_temp.osd;
2819	}
2820
2821	/*
2822	* Map a PG to its acting set as well as its up set.
2823	*
2824	* Acting set is used for data mapping purposes, while up set can be
2825	* recorded for detecting interval changes and deciding whether to
2826	* resend a request.
2827	*/
2828	void ceph_pg_to_up_acting_osds(struct ceph_osdmap *osdmap,
2829	struct ceph_pg_pool_info *pi,
2830	const struct ceph_pg *raw_pgid,
2831	struct ceph_osds *up,
2832	struct ceph_osds *acting)
2833	{
2834	struct ceph_pg pgid;
2835	u32 pps;
2836
2837	WARN_ON(pi->id != raw_pgid->pool);
2838	raw_pg_to_pg(pi, raw_pgid, &pgid);
2839
2840	pg_to_raw_osds(osdmap, pi, raw_pgid, up, &pps);
2841	apply_upmap(osdmap, &pgid, up);
2842	raw_to_up_osds(osdmap, pi, up);
2843	apply_primary_affinity(osdmap, pi, pps, up);
2844	get_temp_osds(osdmap, pi, &pgid, acting);
2845	if (!acting->size) {
2846	memcpy(acting->osds, up->osds, up->size * sizeof(up->osds[0]));
2847	acting->size = up->size;
2848	if (acting->primary == -1)
2849	acting->primary = up->primary;
2850	}
2851	WARN_ON(!osds_valid(up) \|\| !osds_valid(acting));
2852	}
2853
2854	bool ceph_pg_to_primary_shard(struct ceph_osdmap *osdmap,
2855	struct ceph_pg_pool_info *pi,
2856	const struct ceph_pg *raw_pgid,
2857	struct ceph_spg *spgid)
2858	{
2859	struct ceph_pg pgid;
2860	struct ceph_osds up, acting;
2861	int i;
2862
2863	WARN_ON(pi->id != raw_pgid->pool);
2864	raw_pg_to_pg(pi, raw_pgid, &pgid);
2865
2866	if (ceph_can_shift_osds(pi)) {
2867	spgid->pgid = pgid; /* struct */
2868	spgid->shard = CEPH_SPG_NOSHARD;
2869	return true;
2870	}
2871
2872	ceph_pg_to_up_acting_osds(osdmap, pi, &pgid, &up, &acting);
2873	for (i = 0; i < acting.size; i++) {
2874	if (acting.osds[i] == acting.primary) {
2875	spgid->pgid = pgid; /* struct */
2876	spgid->shard = i;
2877	return true;
2878	}
2879	}
2880
2881	return false;
2882	}
2883
2884	/*
2885	* Return acting primary for given PG, or -1 if none.
2886	*/
2887	int ceph_pg_to_acting_primary(struct ceph_osdmap *osdmap,
2888	const struct ceph_pg *raw_pgid)
2889	{
2890	struct ceph_pg_pool_info *pi;
2891	struct ceph_osds up, acting;
2892
2893	pi = ceph_pg_pool_by_id(osdmap, raw_pgid->pool);
2894	if (!pi)
2895	return -1;
2896
2897	ceph_pg_to_up_acting_osds(osdmap, pi, raw_pgid, &up, &acting);
2898	return acting.primary;
2899	}
2900	EXPORT_SYMBOL(ceph_pg_to_acting_primary);
2901
2902	static struct crush_loc_node *alloc_crush_loc(size_t type_name_len,
2903	size_t name_len)
2904	{
2905	struct crush_loc_node *loc;
2906
2907	loc = kmalloc(sizeof(*loc) + type_name_len + name_len + 2, GFP_NOIO);
2908	if (!loc)
2909	return NULL;
2910
2911	RB_CLEAR_NODE(&loc->cl_node);
2912	return loc;
2913	}
2914
2915	static void free_crush_loc(struct crush_loc_node *loc)
2916	{
2917	WARN_ON(!RB_EMPTY_NODE(&loc->cl_node));
2918
2919	kfree(loc);
2920	}
2921
2922	static int crush_loc_compare(const struct crush_loc *loc1,
2923	const struct crush_loc *loc2)
2924	{
2925	return strcmp(loc1->cl_type_name, loc2->cl_type_name) ?:
2926	strcmp(loc1->cl_name, loc2->cl_name);
2927	}
2928
2929	DEFINE_RB_FUNCS2(crush_loc, struct crush_loc_node, cl_loc, crush_loc_compare,
2930	RB_BYPTR, const struct crush_loc *, cl_node)
2931
2932	/*
2933	* Parses a set of <bucket type name>':'<bucket name> pairs separated
2934	* by '\|', e.g. "rack:foo1\|rack:foo2\|datacenter:bar".
2935	*
2936	* Note that @crush_location is modified by strsep().
2937	*/
2938	int ceph_parse_crush_location(char crush_location, struct rb_root locs)
2939	{
2940	struct crush_loc_node *loc;
2941	const char type_name, name, *colon;
2942	size_t type_name_len, name_len;
2943
2944	dout("%s '%s'\n", __func__, crush_location);
2945	while ((type_name = strsep(&crush_location, "\|"))) {
2946	colon = strchr(type_name, ':');
2947	if (!colon)
2948	return -EINVAL;
2949
2950	type_name_len = colon - type_name;
2951	if (type_name_len == 0)
2952	return -EINVAL;
2953
2954	name = colon + 1;
2955	name_len = strlen(name);
2956	if (name_len == 0)
2957	return -EINVAL;
2958
2959	loc = alloc_crush_loc(type_name_len, name_len);
2960	if (!loc)
2961	return -ENOMEM;
2962
2963	loc->cl_loc.cl_type_name = loc->cl_data;
2964	memcpy(loc->cl_loc.cl_type_name, type_name, type_name_len);
2965	loc->cl_loc.cl_type_name[type_name_len] = '\0';
2966
2967	loc->cl_loc.cl_name = loc->cl_data + type_name_len + 1;
2968	memcpy(loc->cl_loc.cl_name, name, name_len);
2969	loc->cl_loc.cl_name[name_len] = '\0';
2970
2971	if (!__insert_crush_loc(locs, loc)) {
2972	free_crush_loc(loc);
2973	return -EEXIST;
2974	}
2975
2976	dout("%s type_name '%s' name '%s'\n", __func__,
2977	loc->cl_loc.cl_type_name, loc->cl_loc.cl_name);
2978	}
2979
2980	return 0;
2981	}
2982
2983	int ceph_compare_crush_locs(struct rb_root locs1, struct rb_root locs2)
2984	{
2985	struct rb_node *n1 = rb_first(locs1);
2986	struct rb_node *n2 = rb_first(locs2);
2987	int ret;
2988
2989	for ( ; n1 && n2; n1 = rb_next(n1), n2 = rb_next(n2)) {
2990	struct crush_loc_node *loc1 =
2991	rb_entry(n1, struct crush_loc_node, cl_node);
2992	struct crush_loc_node *loc2 =
2993	rb_entry(n2, struct crush_loc_node, cl_node);
2994
2995	ret = crush_loc_compare(&loc1->cl_loc, &loc2->cl_loc);
2996	if (ret)
2997	return ret;
2998	}
2999
3000	if (!n1 && n2)
3001	return -1;
3002	if (n1 && !n2)
3003	return 1;
3004	return 0;
3005	}
3006
3007	void ceph_clear_crush_locs(struct rb_root *locs)
3008	{
3009	while (!RB_EMPTY_ROOT(locs)) {
3010	struct crush_loc_node *loc =
3011	rb_entry(rb_first(locs), struct crush_loc_node, cl_node);
3012
3013	erase_crush_loc(locs, loc);
3014	free_crush_loc(loc);
3015	}
3016	}
3017
3018	/*
3019	* [a-zA-Z0-9-_.]+
3020	*/
3021	static bool is_valid_crush_name(const char *name)
3022	{
3023	do {
3024	if (!('a' <= name && name <= 'z') &&
3025	!('A' <= name && name <= 'Z') &&
3026	!('0' <= name && name <= '9') &&
3027	name != '-' && name != '_' && *name != '.')
3028	return false;
3029	} while (*++name != '\0');
3030
3031	return true;
3032	}
3033
3034	/*
3035	* Gets the parent of an item. Returns its id (<0 because the
3036	* parent is always a bucket), type id (>0 for the same reason,
3037	* via @parent_type_id) and location (via @parent_loc). If no
3038	* parent, returns 0.
3039	*
3040	* Does a linear search, as there are no parent pointers of any
3041	* kind. Note that the result is ambiguous for items that occur
3042	* multiple times in the map.
3043	*/
3044	static int get_immediate_parent(struct crush_map *c, int id,
3045	u16 *parent_type_id,
3046	struct crush_loc *parent_loc)
3047	{
3048	struct crush_bucket *b;
3049	struct crush_name_node type_cn, cn;
3050	int i, j;
3051
3052	for (i = 0; i < c->max_buckets; i++) {
3053	b = c->buckets[i];
3054	if (!b)
3055	continue;
3056
3057	/* ignore per-class shadow hierarchy */
3058	cn = lookup_crush_name(&c->names, b->id);
3059	if (!cn \|\| !is_valid_crush_name(cn->cn_name))
3060	continue;
3061
3062	for (j = 0; j < b->size; j++) {
3063	if (b->items[j] != id)
3064	continue;
3065
3066	type_cn = lookup_crush_name(&c->type_names, b->type);
3067	if (WARN_ON_ONCE(!type_cn))
3068	continue;
3069
3070	*parent_type_id = b->type;
3071	parent_loc->cl_type_name = type_cn->cn_name;
3072	parent_loc->cl_name = cn->cn_name;
3073	return b->id;
3074	}
3075	}
3076
3077	return 0; /* no parent */
3078	}
3079
3080	/*
3081	* Calculates the locality/distance from an item to a client
3082	* location expressed in terms of CRUSH hierarchy as a set of
3083	* (bucket type name, bucket name) pairs. Specifically, looks
3084	* for the lowest-valued bucket type for which the location of
3085	* @id matches one of the locations in @locs, so for standard
3086	* bucket types (host = 1, rack = 3, datacenter = 8, zone = 9)
3087	* a matching host is closer than a matching rack and a matching
3088	* data center is closer than a matching zone.
3089	*
3090	* Specifying multiple locations (a "multipath" location) such
3091	* as "rack=foo1 rack=foo2 datacenter=bar" is allowed -- @locs
3092	* is a multimap. The locality will be:
3093	*
3094	* - 3 for OSDs in racks foo1 and foo2
3095	* - 8 for OSDs in data center bar
3096	* - -1 for all other OSDs
3097	*
3098	* The lowest possible bucket type is 1, so the best locality
3099	* for an OSD is 1 (i.e. a matching host). Locality 0 would be
3100	* the OSD itself.
3101	*/
3102	int ceph_get_crush_locality(struct ceph_osdmap *osdmap, int id,
3103	struct rb_root *locs)
3104	{
3105	struct crush_loc loc;
3106	u16 type_id;
3107
3108	/*
3109	* Instead of repeated get_immediate_parent() calls,
3110	* the location of @id could be obtained with a single
3111	* depth-first traversal.
3112	*/
3113	for (;;) {
3114	id = get_immediate_parent(osdmap->crush, id, &type_id, &loc);
3115	if (id >= 0)
3116	return -1; /* not local */
3117
3118	if (lookup_crush_loc(locs, &loc))
3119	return type_id;
3120	}
3121	}