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