?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
KEYS: add missing permission check for request_key() destination
[linux/fpc-iii.git] / net / ceph / osdmap.c
blob8f35d71da955649422056e9952f4d632d5418f7e
1
2 #include <linux/ceph/ceph_debug.h>
3
4 #include <linux/module.h>
5 #include <linux/slab.h>
6 #include <asm/div64.h>
7
8 #include <linux/ceph/libceph.h>
9 #include <linux/ceph/osdmap.h>
10 #include <linux/ceph/decode.h>
11 #include <linux/crush/hash.h>
12 #include <linux/crush/mapper.h>
13
14 char *ceph_osdmap_state_str(char *str, int len, int state)
15 {
16 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
33 static int calc_bits_of(unsigned int t)
34 {
35 int b = 0;
36 while (t) {
37 t = t >> 1;
38 b++;
39 }
40 return b;
41 }
42
43 /*
44 * the foo_mask is the smallest value 2^n-1 that is >= foo.
45 */
46 static void calc_pg_masks(struct ceph_pg_pool_info *pi)
47 {
48 pi->pg_num_mask = (1 << calc_bits_of(pi->pg_num-1)) - 1;
49 pi->pgp_num_mask = (1 << calc_bits_of(pi->pgp_num-1)) - 1;
50 }
51
52 /*
53 * decode crush map
54 */
55 static int crush_decode_uniform_bucket(void **p, void *end,
56 struct crush_bucket_uniform *b)
57 {
58 dout("crush_decode_uniform_bucket %p to %p\n", *p, end);
59 ceph_decode_need(p, end, (1+b->h.size) * sizeof(u32), bad);
60 b->item_weight = ceph_decode_32(p);
61 return 0;
62 bad:
63 return -EINVAL;
64 }
65
66 static int crush_decode_list_bucket(void **p, void *end,
67 struct crush_bucket_list *b)
68 {
69 int j;
70 dout("crush_decode_list_bucket %p to %p\n", *p, end);
71 b->item_weights = kcalloc(b->h.size, sizeof(u32), GFP_NOFS);
72 if (b->item_weights == NULL)
73 return -ENOMEM;
74 b->sum_weights = kcalloc(b->h.size, sizeof(u32), GFP_NOFS);
75 if (b->sum_weights == NULL)
76 return -ENOMEM;
77 ceph_decode_need(p, end, 2 * b->h.size * sizeof(u32), bad);
78 for (j = 0; j < b->h.size; j++) {
79 b->item_weights[j] = ceph_decode_32(p);
80 b->sum_weights[j] = ceph_decode_32(p);
81 }
82 return 0;
83 bad:
84 return -EINVAL;
85 }
86
87 static int crush_decode_tree_bucket(void **p, void *end,
88 struct crush_bucket_tree *b)
89 {
90 int j;
91 dout("crush_decode_tree_bucket %p to %p\n", *p, end);
92 ceph_decode_8_safe(p, end, b->num_nodes, bad);
93 b->node_weights = kcalloc(b->num_nodes, sizeof(u32), GFP_NOFS);
94 if (b->node_weights == NULL)
95 return -ENOMEM;
96 ceph_decode_need(p, end, b->num_nodes * sizeof(u32), bad);
97 for (j = 0; j < b->num_nodes; j++)
98 b->node_weights[j] = ceph_decode_32(p);
99 return 0;
100 bad:
101 return -EINVAL;
102 }
103
104 static int crush_decode_straw_bucket(void **p, void *end,
105 struct crush_bucket_straw *b)
106 {
107 int j;
108 dout("crush_decode_straw_bucket %p to %p\n", *p, end);
109 b->item_weights = kcalloc(b->h.size, sizeof(u32), GFP_NOFS);
110 if (b->item_weights == NULL)
111 return -ENOMEM;
112 b->straws = kcalloc(b->h.size, sizeof(u32), GFP_NOFS);
113 if (b->straws == NULL)
114 return -ENOMEM;
115 ceph_decode_need(p, end, 2 * b->h.size * sizeof(u32), bad);
116 for (j = 0; j < b->h.size; j++) {
117 b->item_weights[j] = ceph_decode_32(p);
118 b->straws[j] = ceph_decode_32(p);
119 }
120 return 0;
121 bad:
122 return -EINVAL;
123 }
124
125 static int skip_name_map(void **p, void *end)
126 {
127 int len;
128 ceph_decode_32_safe(p, end, len ,bad);
129 while (len--) {
130 int strlen;
131 *p += sizeof(u32);
132 ceph_decode_32_safe(p, end, strlen, bad);
133 *p += strlen;
134 }
135 return 0;
136 bad:
137 return -EINVAL;
138 }
139
140 static struct crush_map *crush_decode(void *pbyval, void *end)
141 {
142 struct crush_map *c;
143 int err = -EINVAL;
144 int i, j;
145 void **p = &pbyval;
146 void *start = pbyval;
147 u32 magic;
148 u32 num_name_maps;
149
150 dout("crush_decode %p to %p len %d\n", *p, end, (int)(end - *p));
151
152 c = kzalloc(sizeof(*c), GFP_NOFS);
153 if (c == NULL)
154 return ERR_PTR(-ENOMEM);
155
156 /* set tunables to default values */
157 c->choose_local_tries = 2;
158 c->choose_local_fallback_tries = 5;
159 c->choose_total_tries = 19;
160 c->chooseleaf_descend_once = 0;
161
162 ceph_decode_need(p, end, 4*sizeof(u32), bad);
163 magic = ceph_decode_32(p);
164 if (magic != CRUSH_MAGIC) {
165 pr_err("crush_decode magic %x != current %x\n",
166 (unsigned int)magic, (unsigned int)CRUSH_MAGIC);
167 goto bad;
168 }
169 c->max_buckets = ceph_decode_32(p);
170 c->max_rules = ceph_decode_32(p);
171 c->max_devices = ceph_decode_32(p);
172
173 c->buckets = kcalloc(c->max_buckets, sizeof(*c->buckets), GFP_NOFS);
174 if (c->buckets == NULL)
175 goto badmem;
176 c->rules = kcalloc(c->max_rules, sizeof(*c->rules), GFP_NOFS);
177 if (c->rules == NULL)
178 goto badmem;
179
180 /* buckets */
181 for (i = 0; i < c->max_buckets; i++) {
182 int size = 0;
183 u32 alg;
184 struct crush_bucket *b;
185
186 ceph_decode_32_safe(p, end, alg, bad);
187 if (alg == 0) {
188 c->buckets[i] = NULL;
189 continue;
190 }
191 dout("crush_decode bucket %d off %x %p to %p\n",
192 i, (int)(*p-start), *p, end);
193
194 switch (alg) {
195 case CRUSH_BUCKET_UNIFORM:
196 size = sizeof(struct crush_bucket_uniform);
197 break;
198 case CRUSH_BUCKET_LIST:
199 size = sizeof(struct crush_bucket_list);
200 break;
201 case CRUSH_BUCKET_TREE:
202 size = sizeof(struct crush_bucket_tree);
203 break;
204 case CRUSH_BUCKET_STRAW:
205 size = sizeof(struct crush_bucket_straw);
206 break;
207 default:
208 err = -EINVAL;
209 goto bad;
210 }
211 BUG_ON(size == 0);
212 b = c->buckets[i] = kzalloc(size, GFP_NOFS);
213 if (b == NULL)
214 goto badmem;
215
216 ceph_decode_need(p, end, 4*sizeof(u32), bad);
217 b->id = ceph_decode_32(p);
218 b->type = ceph_decode_16(p);
219 b->alg = ceph_decode_8(p);
220 b->hash = ceph_decode_8(p);
221 b->weight = ceph_decode_32(p);
222 b->size = ceph_decode_32(p);
223
224 dout("crush_decode bucket size %d off %x %p to %p\n",
225 b->size, (int)(*p-start), *p, end);
226
227 b->items = kcalloc(b->size, sizeof(__s32), GFP_NOFS);
228 if (b->items == NULL)
229 goto badmem;
230 b->perm = kcalloc(b->size, sizeof(u32), GFP_NOFS);
231 if (b->perm == NULL)
232 goto badmem;
233 b->perm_n = 0;
234
235 ceph_decode_need(p, end, b->size*sizeof(u32), bad);
236 for (j = 0; j < b->size; j++)
237 b->items[j] = ceph_decode_32(p);
238
239 switch (b->alg) {
240 case CRUSH_BUCKET_UNIFORM:
241 err = crush_decode_uniform_bucket(p, end,
242 (struct crush_bucket_uniform *)b);
243 if (err < 0)
244 goto bad;
245 break;
246 case CRUSH_BUCKET_LIST:
247 err = crush_decode_list_bucket(p, end,
248 (struct crush_bucket_list *)b);
249 if (err < 0)
250 goto bad;
251 break;
252 case CRUSH_BUCKET_TREE:
253 err = crush_decode_tree_bucket(p, end,
254 (struct crush_bucket_tree *)b);
255 if (err < 0)
256 goto bad;
257 break;
258 case CRUSH_BUCKET_STRAW:
259 err = crush_decode_straw_bucket(p, end,
260 (struct crush_bucket_straw *)b);
261 if (err < 0)
262 goto bad;
263 break;
264 }
265 }
266
267 /* rules */
268 dout("rule vec is %p\n", c->rules);
269 for (i = 0; i < c->max_rules; i++) {
270 u32 yes;
271 struct crush_rule *r;
272
273 err = -EINVAL;
274 ceph_decode_32_safe(p, end, yes, bad);
275 if (!yes) {
276 dout("crush_decode NO rule %d off %x %p to %p\n",
277 i, (int)(*p-start), *p, end);
278 c->rules[i] = NULL;
279 continue;
280 }
281
282 dout("crush_decode rule %d off %x %p to %p\n",
283 i, (int)(*p-start), *p, end);
284
285 /* len */
286 ceph_decode_32_safe(p, end, yes, bad);
287 #if BITS_PER_LONG == 32
288 err = -EINVAL;
289 if (yes > (ULONG_MAX - sizeof(*r))
290 / sizeof(struct crush_rule_step))
291 goto bad;
292 #endif
293 r = c->rules[i] = kmalloc(sizeof(*r) +
294 yes*sizeof(struct crush_rule_step),
295 GFP_NOFS);
296 if (r == NULL)
297 goto badmem;
298 dout(" rule %d is at %p\n", i, r);
299 r->len = yes;
300 ceph_decode_copy_safe(p, end, &r->mask, 4, bad); /* 4 u8's */
301 ceph_decode_need(p, end, r->len*3*sizeof(u32), bad);
302 for (j = 0; j < r->len; j++) {
303 r->steps[j].op = ceph_decode_32(p);
304 r->steps[j].arg1 = ceph_decode_32(p);
305 r->steps[j].arg2 = ceph_decode_32(p);
306 }
307 }
308
309 /* ignore trailing name maps. */
310 for (num_name_maps = 0; num_name_maps < 3; num_name_maps++) {
311 err = skip_name_map(p, end);
312 if (err < 0)
313 goto done;
314 }
315
316 /* tunables */
317 ceph_decode_need(p, end, 3*sizeof(u32), done);
318 c->choose_local_tries = ceph_decode_32(p);
319 c->choose_local_fallback_tries = ceph_decode_32(p);
320 c->choose_total_tries = ceph_decode_32(p);
321 dout("crush decode tunable choose_local_tries = %d",
322 c->choose_local_tries);
323 dout("crush decode tunable choose_local_fallback_tries = %d",
324 c->choose_local_fallback_tries);
325 dout("crush decode tunable choose_total_tries = %d",
326 c->choose_total_tries);
327
328 ceph_decode_need(p, end, sizeof(u32), done);
329 c->chooseleaf_descend_once = ceph_decode_32(p);
330 dout("crush decode tunable chooseleaf_descend_once = %d",
331 c->chooseleaf_descend_once);
332
333 ceph_decode_need(p, end, sizeof(u8), done);
334 c->chooseleaf_vary_r = ceph_decode_8(p);
335 dout("crush decode tunable chooseleaf_vary_r = %d",
336 c->chooseleaf_vary_r);
337
338 done:
339 dout("crush_decode success\n");
340 return c;
341
342 badmem:
343 err = -ENOMEM;
344 bad:
345 dout("crush_decode fail %d\n", err);
346 crush_destroy(c);
347 return ERR_PTR(err);
348 }
349
350 /*
351 * rbtree of pg_mapping for handling pg_temp (explicit mapping of pgid
352 * to a set of osds) and primary_temp (explicit primary setting)
353 */
354 static int pgid_cmp(struct ceph_pg l, struct ceph_pg r)
355 {
356 if (l.pool < r.pool)
357 return -1;
358 if (l.pool > r.pool)
359 return 1;
360 if (l.seed < r.seed)
361 return -1;
362 if (l.seed > r.seed)
363 return 1;
364 return 0;
365 }
366
367 static int __insert_pg_mapping(struct ceph_pg_mapping *new,
368 struct rb_root *root)
369 {
370 struct rb_node **p = &root->rb_node;
371 struct rb_node *parent = NULL;
372 struct ceph_pg_mapping *pg = NULL;
373 int c;
374
375 dout("__insert_pg_mapping %llx %p\n", *(u64 *)&new->pgid, new);
376 while (*p) {
377 parent = *p;
378 pg = rb_entry(parent, struct ceph_pg_mapping, node);
379 c = pgid_cmp(new->pgid, pg->pgid);
380 if (c < 0)
381 p = &(*p)->rb_left;
382 else if (c > 0)
383 p = &(*p)->rb_right;
384 else
385 return -EEXIST;
386 }
387
388 rb_link_node(&new->node, parent, p);
389 rb_insert_color(&new->node, root);
390 return 0;
391 }
392
393 static struct ceph_pg_mapping *__lookup_pg_mapping(struct rb_root *root,
394 struct ceph_pg pgid)
395 {
396 struct rb_node *n = root->rb_node;
397 struct ceph_pg_mapping *pg;
398 int c;
399
400 while (n) {
401 pg = rb_entry(n, struct ceph_pg_mapping, node);
402 c = pgid_cmp(pgid, pg->pgid);
403 if (c < 0) {
404 n = n->rb_left;
405 } else if (c > 0) {
406 n = n->rb_right;
407 } else {
408 dout("__lookup_pg_mapping %lld.%x got %p\n",
409 pgid.pool, pgid.seed, pg);
410 return pg;
411 }
412 }
413 return NULL;
414 }
415
416 static int __remove_pg_mapping(struct rb_root *root, struct ceph_pg pgid)
417 {
418 struct ceph_pg_mapping *pg = __lookup_pg_mapping(root, pgid);
419
420 if (pg) {
421 dout("__remove_pg_mapping %lld.%x %p\n", pgid.pool, pgid.seed,
422 pg);
423 rb_erase(&pg->node, root);
424 kfree(pg);
425 return 0;
426 }
427 dout("__remove_pg_mapping %lld.%x dne\n", pgid.pool, pgid.seed);
428 return -ENOENT;
429 }
430
431 /*
432 * rbtree of pg pool info
433 */
434 static int __insert_pg_pool(struct rb_root *root, struct ceph_pg_pool_info *new)
435 {
436 struct rb_node **p = &root->rb_node;
437 struct rb_node *parent = NULL;
438 struct ceph_pg_pool_info *pi = NULL;
439
440 while (*p) {
441 parent = *p;
442 pi = rb_entry(parent, struct ceph_pg_pool_info, node);
443 if (new->id < pi->id)
444 p = &(*p)->rb_left;
445 else if (new->id > pi->id)
446 p = &(*p)->rb_right;
447 else
448 return -EEXIST;
449 }
450
451 rb_link_node(&new->node, parent, p);
452 rb_insert_color(&new->node, root);
453 return 0;
454 }
455
456 static struct ceph_pg_pool_info *__lookup_pg_pool(struct rb_root *root, u64 id)
457 {
458 struct ceph_pg_pool_info *pi;
459 struct rb_node *n = root->rb_node;
460
461 while (n) {
462 pi = rb_entry(n, struct ceph_pg_pool_info, node);
463 if (id < pi->id)
464 n = n->rb_left;
465 else if (id > pi->id)
466 n = n->rb_right;
467 else
468 return pi;
469 }
470 return NULL;
471 }
472
473 struct ceph_pg_pool_info *ceph_pg_pool_by_id(struct ceph_osdmap *map, u64 id)
474 {
475 return __lookup_pg_pool(&map->pg_pools, id);
476 }
477
478 const char *ceph_pg_pool_name_by_id(struct ceph_osdmap *map, u64 id)
479 {
480 struct ceph_pg_pool_info *pi;
481
482 if (id == CEPH_NOPOOL)
483 return NULL;
484
485 if (WARN_ON_ONCE(id > (u64) INT_MAX))
486 return NULL;
487
488 pi = __lookup_pg_pool(&map->pg_pools, (int) id);
489
490 return pi ? pi->name : NULL;
491 }
492 EXPORT_SYMBOL(ceph_pg_pool_name_by_id);
493
494 int ceph_pg_poolid_by_name(struct ceph_osdmap *map, const char *name)
495 {
496 struct rb_node *rbp;
497
498 for (rbp = rb_first(&map->pg_pools); rbp; rbp = rb_next(rbp)) {
499 struct ceph_pg_pool_info *pi =
500 rb_entry(rbp, struct ceph_pg_pool_info, node);
501 if (pi->name && strcmp(pi->name, name) == 0)
502 return pi->id;
503 }
504 return -ENOENT;
505 }
506 EXPORT_SYMBOL(ceph_pg_poolid_by_name);
507
508 static void __remove_pg_pool(struct rb_root *root, struct ceph_pg_pool_info *pi)
509 {
510 rb_erase(&pi->node, root);
511 kfree(pi->name);
512 kfree(pi);
513 }
514
515 static int decode_pool(void **p, void *end, struct ceph_pg_pool_info *pi)
516 {
517 u8 ev, cv;
518 unsigned len, num;
519 void *pool_end;
520
521 ceph_decode_need(p, end, 2 + 4, bad);
522 ev = ceph_decode_8(p); /* encoding version */
523 cv = ceph_decode_8(p); /* compat version */
524 if (ev < 5) {
525 pr_warning("got v %d < 5 cv %d of ceph_pg_pool\n", ev, cv);
526 return -EINVAL;
527 }
528 if (cv > 9) {
529 pr_warning("got v %d cv %d > 9 of ceph_pg_pool\n", ev, cv);
530 return -EINVAL;
531 }
532 len = ceph_decode_32(p);
533 ceph_decode_need(p, end, len, bad);
534 pool_end = *p + len;
535
536 pi->type = ceph_decode_8(p);
537 pi->size = ceph_decode_8(p);
538 pi->crush_ruleset = ceph_decode_8(p);
539 pi->object_hash = ceph_decode_8(p);
540
541 pi->pg_num = ceph_decode_32(p);
542 pi->pgp_num = ceph_decode_32(p);
543
544 *p += 4 + 4; /* skip lpg* */
545 *p += 4; /* skip last_change */
546 *p += 8 + 4; /* skip snap_seq, snap_epoch */
547
548 /* skip snaps */
549 num = ceph_decode_32(p);
550 while (num--) {
551 *p += 8; /* snapid key */
552 *p += 1 + 1; /* versions */
553 len = ceph_decode_32(p);
554 *p += len;
555 }
556
557 /* skip removed_snaps */
558 num = ceph_decode_32(p);
559 *p += num * (8 + 8);
560
561 *p += 8; /* skip auid */
562 pi->flags = ceph_decode_64(p);
563 *p += 4; /* skip crash_replay_interval */
564
565 if (ev >= 7)
566 *p += 1; /* skip min_size */
567
568 if (ev >= 8)
569 *p += 8 + 8; /* skip quota_max_* */
570
571 if (ev >= 9) {
572 /* skip tiers */
573 num = ceph_decode_32(p);
574 *p += num * 8;
575
576 *p += 8; /* skip tier_of */
577 *p += 1; /* skip cache_mode */
578
579 pi->read_tier = ceph_decode_64(p);
580 pi->write_tier = ceph_decode_64(p);
581 } else {
582 pi->read_tier = -1;
583 pi->write_tier = -1;
584 }
585
586 /* ignore the rest */
587
588 *p = pool_end;
589 calc_pg_masks(pi);
590 return 0;
591
592 bad:
593 return -EINVAL;
594 }
595
596 static int decode_pool_names(void **p, void *end, struct ceph_osdmap *map)
597 {
598 struct ceph_pg_pool_info *pi;
599 u32 num, len;
600 u64 pool;
601
602 ceph_decode_32_safe(p, end, num, bad);
603 dout(" %d pool names\n", num);
604 while (num--) {
605 ceph_decode_64_safe(p, end, pool, bad);
606 ceph_decode_32_safe(p, end, len, bad);
607 dout(" pool %llu len %d\n", pool, len);
608 ceph_decode_need(p, end, len, bad);
609 pi = __lookup_pg_pool(&map->pg_pools, pool);
610 if (pi) {
611 char *name = kstrndup(*p, len, GFP_NOFS);
612
613 if (!name)
614 return -ENOMEM;
615 kfree(pi->name);
616 pi->name = name;
617 dout(" name is %s\n", pi->name);
618 }
619 *p += len;
620 }
621 return 0;
622
623 bad:
624 return -EINVAL;
625 }
626
627 /*
628 * osd map
629 */
630 void ceph_osdmap_destroy(struct ceph_osdmap *map)
631 {
632 dout("osdmap_destroy %p\n", map);
633 if (map->crush)
634 crush_destroy(map->crush);
635 while (!RB_EMPTY_ROOT(&map->pg_temp)) {
636 struct ceph_pg_mapping *pg =
637 rb_entry(rb_first(&map->pg_temp),
638 struct ceph_pg_mapping, node);
639 rb_erase(&pg->node, &map->pg_temp);
640 kfree(pg);
641 }
642 while (!RB_EMPTY_ROOT(&map->primary_temp)) {
643 struct ceph_pg_mapping *pg =
644 rb_entry(rb_first(&map->primary_temp),
645 struct ceph_pg_mapping, node);
646 rb_erase(&pg->node, &map->primary_temp);
647 kfree(pg);
648 }
649 while (!RB_EMPTY_ROOT(&map->pg_pools)) {
650 struct ceph_pg_pool_info *pi =
651 rb_entry(rb_first(&map->pg_pools),
652 struct ceph_pg_pool_info, node);
653 __remove_pg_pool(&map->pg_pools, pi);
654 }
655 kfree(map->osd_state);
656 kfree(map->osd_weight);
657 kfree(map->osd_addr);
658 kfree(map->osd_primary_affinity);
659 kfree(map);
660 }
661
662 /*
663 * Adjust max_osd value, (re)allocate arrays.
664 *
665 * The new elements are properly initialized.
666 */
667 static int osdmap_set_max_osd(struct ceph_osdmap *map, int max)
668 {
669 u8 *state;
670 u32 *weight;
671 struct ceph_entity_addr *addr;
672 int i;
673
674 state = krealloc(map->osd_state, max*sizeof(*state), GFP_NOFS);
675 weight = krealloc(map->osd_weight, max*sizeof(*weight), GFP_NOFS);
676 addr = krealloc(map->osd_addr, max*sizeof(*addr), GFP_NOFS);
677 if (!state || !weight || !addr) {
678 kfree(state);
679 kfree(weight);
680 kfree(addr);
681
682 return -ENOMEM;
683 }
684
685 for (i = map->max_osd; i < max; i++) {
686 state[i] = 0;
687 weight[i] = CEPH_OSD_OUT;
688 memset(addr + i, 0, sizeof(*addr));
689 }
690
691 map->osd_state = state;
692 map->osd_weight = weight;
693 map->osd_addr = addr;
694
695 if (map->osd_primary_affinity) {
696 u32 *affinity;
697
698 affinity = krealloc(map->osd_primary_affinity,
699 max*sizeof(*affinity), GFP_NOFS);
700 if (!affinity)
701 return -ENOMEM;
702
703 for (i = map->max_osd; i < max; i++)
704 affinity[i] = CEPH_OSD_DEFAULT_PRIMARY_AFFINITY;
705
706 map->osd_primary_affinity = affinity;
707 }
708
709 map->max_osd = max;
710
711 return 0;
712 }
713
714 #define OSDMAP_WRAPPER_COMPAT_VER 7
715 #define OSDMAP_CLIENT_DATA_COMPAT_VER 1
716
717 /*
718 * Return 0 or error. On success, *v is set to 0 for old (v6) osdmaps,
719 * to struct_v of the client_data section for new (v7 and above)
720 * osdmaps.
721 */
722 static int get_osdmap_client_data_v(void **p, void *end,
723 const char *prefix, u8 *v)
724 {
725 u8 struct_v;
726
727 ceph_decode_8_safe(p, end, struct_v, e_inval);
728 if (struct_v >= 7) {
729 u8 struct_compat;
730
731 ceph_decode_8_safe(p, end, struct_compat, e_inval);
732 if (struct_compat > OSDMAP_WRAPPER_COMPAT_VER) {
733 pr_warning("got v %d cv %d > %d of %s ceph_osdmap\n",
734 struct_v, struct_compat,
735 OSDMAP_WRAPPER_COMPAT_VER, prefix);
736 return -EINVAL;
737 }
738 *p += 4; /* ignore wrapper struct_len */
739
740 ceph_decode_8_safe(p, end, struct_v, e_inval);
741 ceph_decode_8_safe(p, end, struct_compat, e_inval);
742 if (struct_compat > OSDMAP_CLIENT_DATA_COMPAT_VER) {
743 pr_warning("got v %d cv %d > %d of %s ceph_osdmap client data\n",
744 struct_v, struct_compat,
745 OSDMAP_CLIENT_DATA_COMPAT_VER, prefix);
746 return -EINVAL;
747 }
748 *p += 4; /* ignore client data struct_len */
749 } else {
750 u16 version;
751
752 *p -= 1;
753 ceph_decode_16_safe(p, end, version, e_inval);
754 if (version < 6) {
755 pr_warning("got v %d < 6 of %s ceph_osdmap\n", version,
756 prefix);
757 return -EINVAL;
758 }
759
760 /* old osdmap enconding */
761 struct_v = 0;
762 }
763
764 *v = struct_v;
765 return 0;
766
767 e_inval:
768 return -EINVAL;
769 }
770
771 static int __decode_pools(void **p, void *end, struct ceph_osdmap *map,
772 bool incremental)
773 {
774 u32 n;
775
776 ceph_decode_32_safe(p, end, n, e_inval);
777 while (n--) {
778 struct ceph_pg_pool_info *pi;
779 u64 pool;
780 int ret;
781
782 ceph_decode_64_safe(p, end, pool, e_inval);
783
784 pi = __lookup_pg_pool(&map->pg_pools, pool);
785 if (!incremental || !pi) {
786 pi = kzalloc(sizeof(*pi), GFP_NOFS);
787 if (!pi)
788 return -ENOMEM;
789
790 pi->id = pool;
791
792 ret = __insert_pg_pool(&map->pg_pools, pi);
793 if (ret) {
794 kfree(pi);
795 return ret;
796 }
797 }
798
799 ret = decode_pool(p, end, pi);
800 if (ret)
801 return ret;
802 }
803
804 return 0;
805
806 e_inval:
807 return -EINVAL;
808 }
809
810 static int decode_pools(void **p, void *end, struct ceph_osdmap *map)
811 {
812 return __decode_pools(p, end, map, false);
813 }
814
815 static int decode_new_pools(void **p, void *end, struct ceph_osdmap *map)
816 {
817 return __decode_pools(p, end, map, true);
818 }
819
820 static int __decode_pg_temp(void **p, void *end, struct ceph_osdmap *map,
821 bool incremental)
822 {
823 u32 n;
824
825 ceph_decode_32_safe(p, end, n, e_inval);
826 while (n--) {
827 struct ceph_pg pgid;
828 u32 len, i;
829 int ret;
830
831 ret = ceph_decode_pgid(p, end, &pgid);
832 if (ret)
833 return ret;
834
835 ceph_decode_32_safe(p, end, len, e_inval);
836
837 ret = __remove_pg_mapping(&map->pg_temp, pgid);
838 BUG_ON(!incremental && ret != -ENOENT);
839
840 if (!incremental || len > 0) {
841 struct ceph_pg_mapping *pg;
842
843 ceph_decode_need(p, end, len*sizeof(u32), e_inval);
844
845 if (len > (UINT_MAX - sizeof(*pg)) / sizeof(u32))
846 return -EINVAL;
847
848 pg = kzalloc(sizeof(*pg) + len*sizeof(u32), GFP_NOFS);
849 if (!pg)
850 return -ENOMEM;
851
852 pg->pgid = pgid;
853 pg->pg_temp.len = len;
854 for (i = 0; i < len; i++)
855 pg->pg_temp.osds[i] = ceph_decode_32(p);
856
857 ret = __insert_pg_mapping(pg, &map->pg_temp);
858 if (ret) {
859 kfree(pg);
860 return ret;
861 }
862 }
863 }
864
865 return 0;
866
867 e_inval:
868 return -EINVAL;
869 }
870
871 static int decode_pg_temp(void **p, void *end, struct ceph_osdmap *map)
872 {
873 return __decode_pg_temp(p, end, map, false);
874 }
875
876 static int decode_new_pg_temp(void **p, void *end, struct ceph_osdmap *map)
877 {
878 return __decode_pg_temp(p, end, map, true);
879 }
880
881 static int __decode_primary_temp(void **p, void *end, struct ceph_osdmap *map,
882 bool incremental)
883 {
884 u32 n;
885
886 ceph_decode_32_safe(p, end, n, e_inval);
887 while (n--) {
888 struct ceph_pg pgid;
889 u32 osd;
890 int ret;
891
892 ret = ceph_decode_pgid(p, end, &pgid);
893 if (ret)
894 return ret;
895
896 ceph_decode_32_safe(p, end, osd, e_inval);
897
898 ret = __remove_pg_mapping(&map->primary_temp, pgid);
899 BUG_ON(!incremental && ret != -ENOENT);
900
901 if (!incremental || osd != (u32)-1) {
902 struct ceph_pg_mapping *pg;
903
904 pg = kzalloc(sizeof(*pg), GFP_NOFS);
905 if (!pg)
906 return -ENOMEM;
907
908 pg->pgid = pgid;
909 pg->primary_temp.osd = osd;
910
911 ret = __insert_pg_mapping(pg, &map->primary_temp);
912 if (ret) {
913 kfree(pg);
914 return ret;
915 }
916 }
917 }
918
919 return 0;
920
921 e_inval:
922 return -EINVAL;
923 }
924
925 static int decode_primary_temp(void **p, void *end, struct ceph_osdmap *map)
926 {
927 return __decode_primary_temp(p, end, map, false);
928 }
929
930 static int decode_new_primary_temp(void **p, void *end,
931 struct ceph_osdmap *map)
932 {
933 return __decode_primary_temp(p, end, map, true);
934 }
935
936 u32 ceph_get_primary_affinity(struct ceph_osdmap *map, int osd)
937 {
938 BUG_ON(osd >= map->max_osd);
939
940 if (!map->osd_primary_affinity)
941 return CEPH_OSD_DEFAULT_PRIMARY_AFFINITY;
942
943 return map->osd_primary_affinity[osd];
944 }
945
946 static int set_primary_affinity(struct ceph_osdmap *map, int osd, u32 aff)
947 {
948 BUG_ON(osd >= map->max_osd);
949
950 if (!map->osd_primary_affinity) {
951 int i;
952
953 map->osd_primary_affinity = kmalloc(map->max_osd*sizeof(u32),
954 GFP_NOFS);
955 if (!map->osd_primary_affinity)
956 return -ENOMEM;
957
958 for (i = 0; i < map->max_osd; i++)
959 map->osd_primary_affinity[i] =
960 CEPH_OSD_DEFAULT_PRIMARY_AFFINITY;
961 }
962
963 map->osd_primary_affinity[osd] = aff;
964
965 return 0;
966 }
967
968 static int decode_primary_affinity(void **p, void *end,
969 struct ceph_osdmap *map)
970 {
971 u32 len, i;
972
973 ceph_decode_32_safe(p, end, len, e_inval);
974 if (len == 0) {
975 kfree(map->osd_primary_affinity);
976 map->osd_primary_affinity = NULL;
977 return 0;
978 }
979 if (len != map->max_osd)
980 goto e_inval;
981
982 ceph_decode_need(p, end, map->max_osd*sizeof(u32), e_inval);
983
984 for (i = 0; i < map->max_osd; i++) {
985 int ret;
986
987 ret = set_primary_affinity(map, i, ceph_decode_32(p));
988 if (ret)
989 return ret;
990 }
991
992 return 0;
993
994 e_inval:
995 return -EINVAL;
996 }
997
998 static int decode_new_primary_affinity(void **p, void *end,
999 struct ceph_osdmap *map)
1000 {
1001 u32 n;
1002
1003 ceph_decode_32_safe(p, end, n, e_inval);
1004 while (n--) {
1005 u32 osd, aff;
1006 int ret;
1007
1008 ceph_decode_32_safe(p, end, osd, e_inval);
1009 ceph_decode_32_safe(p, end, aff, e_inval);
1010
1011 ret = set_primary_affinity(map, osd, aff);
1012 if (ret)
1013 return ret;
1014
1015 pr_info("osd%d primary-affinity 0x%x\n", osd, aff);
1016 }
1017
1018 return 0;
1019
1020 e_inval:
1021 return -EINVAL;
1022 }
1023
1024 /*
1025 * decode a full map.
1026 */
1027 static int osdmap_decode(void **p, void *end, struct ceph_osdmap *map)
1028 {
1029 u8 struct_v;
1030 u32 epoch = 0;
1031 void *start = *p;
1032 u32 max;
1033 u32 len, i;
1034 int err;
1035
1036 dout("%s %p to %p len %d\n", __func__, *p, end, (int)(end - *p));
1037
1038 err = get_osdmap_client_data_v(p, end, "full", &struct_v);
1039 if (err)
1040 goto bad;
1041
1042 /* fsid, epoch, created, modified */
1043 ceph_decode_need(p, end, sizeof(map->fsid) + sizeof(u32) +
1044 sizeof(map->created) + sizeof(map->modified), e_inval);
1045 ceph_decode_copy(p, &map->fsid, sizeof(map->fsid));
1046 epoch = map->epoch = ceph_decode_32(p);
1047 ceph_decode_copy(p, &map->created, sizeof(map->created));
1048 ceph_decode_copy(p, &map->modified, sizeof(map->modified));
1049
1050 /* pools */
1051 err = decode_pools(p, end, map);
1052 if (err)
1053 goto bad;
1054
1055 /* pool_name */
1056 err = decode_pool_names(p, end, map);
1057 if (err)
1058 goto bad;
1059
1060 ceph_decode_32_safe(p, end, map->pool_max, e_inval);
1061
1062 ceph_decode_32_safe(p, end, map->flags, e_inval);
1063
1064 /* max_osd */
1065 ceph_decode_32_safe(p, end, max, e_inval);
1066
1067 /* (re)alloc osd arrays */
1068 err = osdmap_set_max_osd(map, max);
1069 if (err)
1070 goto bad;
1071
1072 /* osd_state, osd_weight, osd_addrs->client_addr */
1073 ceph_decode_need(p, end, 3*sizeof(u32) +
1074 map->max_osd*(1 + sizeof(*map->osd_weight) +
1075 sizeof(*map->osd_addr)), e_inval);
1076
1077 if (ceph_decode_32(p) != map->max_osd)
1078 goto e_inval;
1079
1080 ceph_decode_copy(p, map->osd_state, map->max_osd);
1081
1082 if (ceph_decode_32(p) != map->max_osd)
1083 goto e_inval;
1084
1085 for (i = 0; i < map->max_osd; i++)
1086 map->osd_weight[i] = ceph_decode_32(p);
1087
1088 if (ceph_decode_32(p) != map->max_osd)
1089 goto e_inval;
1090
1091 ceph_decode_copy(p, map->osd_addr, map->max_osd*sizeof(*map->osd_addr));
1092 for (i = 0; i < map->max_osd; i++)
1093 ceph_decode_addr(&map->osd_addr[i]);
1094
1095 /* pg_temp */
1096 err = decode_pg_temp(p, end, map);
1097 if (err)
1098 goto bad;
1099
1100 /* primary_temp */
1101 if (struct_v >= 1) {
1102 err = decode_primary_temp(p, end, map);
1103 if (err)
1104 goto bad;
1105 }
1106
1107 /* primary_affinity */
1108 if (struct_v >= 2) {
1109 err = decode_primary_affinity(p, end, map);
1110 if (err)
1111 goto bad;
1112 } else {
1113 /* XXX can this happen? */
1114 kfree(map->osd_primary_affinity);
1115 map->osd_primary_affinity = NULL;
1116 }
1117
1118 /* crush */
1119 ceph_decode_32_safe(p, end, len, e_inval);
1120 map->crush = crush_decode(*p, min(*p + len, end));
1121 if (IS_ERR(map->crush)) {
1122 err = PTR_ERR(map->crush);
1123 map->crush = NULL;
1124 goto bad;
1125 }
1126 *p += len;
1127
1128 /* ignore the rest */
1129 *p = end;
1130
1131 dout("full osdmap epoch %d max_osd %d\n", map->epoch, map->max_osd);
1132 return 0;
1133
1134 e_inval:
1135 err = -EINVAL;
1136 bad:
1137 pr_err("corrupt full osdmap (%d) epoch %d off %d (%p of %p-%p)\n",
1138 err, epoch, (int)(*p - start), *p, start, end);
1139 print_hex_dump(KERN_DEBUG, "osdmap: ",
1140 DUMP_PREFIX_OFFSET, 16, 1,
1141 start, end - start, true);
1142 return err;
1143 }
1144
1145 /*
1146 * Allocate and decode a full map.
1147 */
1148 struct ceph_osdmap *ceph_osdmap_decode(void **p, void *end)
1149 {
1150 struct ceph_osdmap *map;
1151 int ret;
1152
1153 map = kzalloc(sizeof(*map), GFP_NOFS);
1154 if (!map)
1155 return ERR_PTR(-ENOMEM);
1156
1157 map->pg_temp = RB_ROOT;
1158 map->primary_temp = RB_ROOT;
1159 mutex_init(&map->crush_scratch_mutex);
1160
1161 ret = osdmap_decode(p, end, map);
1162 if (ret) {
1163 ceph_osdmap_destroy(map);
1164 return ERR_PTR(ret);
1165 }
1166
1167 return map;
1168 }
1169
1170 /*
1171 * Encoding order is (new_up_client, new_state, new_weight). Need to
1172 * apply in the (new_weight, new_state, new_up_client) order, because
1173 * an incremental map may look like e.g.
1174 *
1175 * new_up_client: { osd=6, addr=... } # set osd_state and addr
1176 * new_state: { osd=6, xorstate=EXISTS } # clear osd_state
1177 */
1178 static int decode_new_up_state_weight(void **p, void *end,
1179 struct ceph_osdmap *map)
1180 {
1181 void *new_up_client;
1182 void *new_state;
1183 void *new_weight_end;
1184 u32 len;
1185
1186 new_up_client = *p;
1187 ceph_decode_32_safe(p, end, len, e_inval);
1188 len *= sizeof(u32) + sizeof(struct ceph_entity_addr);
1189 ceph_decode_need(p, end, len, e_inval);
1190 *p += len;
1191
1192 new_state = *p;
1193 ceph_decode_32_safe(p, end, len, e_inval);
1194 len *= sizeof(u32) + sizeof(u8);
1195 ceph_decode_need(p, end, len, e_inval);
1196 *p += len;
1197
1198 /* new_weight */
1199 ceph_decode_32_safe(p, end, len, e_inval);
1200 while (len--) {
1201 s32 osd;
1202 u32 w;
1203
1204 ceph_decode_need(p, end, 2*sizeof(u32), e_inval);
1205 osd = ceph_decode_32(p);
1206 w = ceph_decode_32(p);
1207 BUG_ON(osd >= map->max_osd);
1208 pr_info("osd%d weight 0x%x %s\n", osd, w,
1209 w == CEPH_OSD_IN ? "(in)" :
1210 (w == CEPH_OSD_OUT ? "(out)" : ""));
1211 map->osd_weight[osd] = w;
1212
1213 /*
1214 * If we are marking in, set the EXISTS, and clear the
1215 * AUTOOUT and NEW bits.
1216 */
1217 if (w) {
1218 map->osd_state[osd] |= CEPH_OSD_EXISTS;
1219 map->osd_state[osd] &= ~(CEPH_OSD_AUTOOUT |
1220 CEPH_OSD_NEW);
1221 }
1222 }
1223 new_weight_end = *p;
1224
1225 /* new_state (up/down) */
1226 *p = new_state;
1227 len = ceph_decode_32(p);
1228 while (len--) {
1229 s32 osd;
1230 u8 xorstate;
1231 int ret;
1232
1233 osd = ceph_decode_32(p);
1234 xorstate = ceph_decode_8(p);
1235 if (xorstate == 0)
1236 xorstate = CEPH_OSD_UP;
1237 BUG_ON(osd >= map->max_osd);
1238 if ((map->osd_state[osd] & CEPH_OSD_UP) &&
1239 (xorstate & CEPH_OSD_UP))
1240 pr_info("osd%d down\n", osd);
1241 if ((map->osd_state[osd] & CEPH_OSD_EXISTS) &&
1242 (xorstate & CEPH_OSD_EXISTS)) {
1243 pr_info("osd%d does not exist\n", osd);
1244 ret = set_primary_affinity(map, osd,
1245 CEPH_OSD_DEFAULT_PRIMARY_AFFINITY);
1246 if (ret)
1247 return ret;
1248 memset(map->osd_addr + osd, 0, sizeof(*map->osd_addr));
1249 map->osd_state[osd] = 0;
1250 } else {
1251 map->osd_state[osd] ^= xorstate;
1252 }
1253 }
1254
1255 /* new_up_client */
1256 *p = new_up_client;
1257 len = ceph_decode_32(p);
1258 while (len--) {
1259 s32 osd;
1260 struct ceph_entity_addr addr;
1261
1262 osd = ceph_decode_32(p);
1263 ceph_decode_copy(p, &addr, sizeof(addr));
1264 ceph_decode_addr(&addr);
1265 BUG_ON(osd >= map->max_osd);
1266 pr_info("osd%d up\n", osd);
1267 map->osd_state[osd] |= CEPH_OSD_EXISTS | CEPH_OSD_UP;
1268 map->osd_addr[osd] = addr;
1269 }
1270
1271 *p = new_weight_end;
1272 return 0;
1273
1274 e_inval:
1275 return -EINVAL;
1276 }
1277
1278 /*
1279 * decode and apply an incremental map update.
1280 */
1281 struct ceph_osdmap *osdmap_apply_incremental(void **p, void *end,
1282 struct ceph_osdmap *map,
1283 struct ceph_messenger *msgr)
1284 {
1285 struct crush_map *newcrush = NULL;
1286 struct ceph_fsid fsid;
1287 u32 epoch = 0;
1288 struct ceph_timespec modified;
1289 s32 len;
1290 u64 pool;
1291 __s64 new_pool_max;
1292 __s32 new_flags, max;
1293 void *start = *p;
1294 int err;
1295 u8 struct_v;
1296
1297 dout("%s %p to %p len %d\n", __func__, *p, end, (int)(end - *p));
1298
1299 err = get_osdmap_client_data_v(p, end, "inc", &struct_v);
1300 if (err)
1301 goto bad;
1302
1303 /* fsid, epoch, modified, new_pool_max, new_flags */
1304 ceph_decode_need(p, end, sizeof(fsid) + sizeof(u32) + sizeof(modified) +
1305 sizeof(u64) + sizeof(u32), e_inval);
1306 ceph_decode_copy(p, &fsid, sizeof(fsid));
1307 epoch = ceph_decode_32(p);
1308 BUG_ON(epoch != map->epoch+1);
1309 ceph_decode_copy(p, &modified, sizeof(modified));
1310 new_pool_max = ceph_decode_64(p);
1311 new_flags = ceph_decode_32(p);
1312
1313 /* full map? */
1314 ceph_decode_32_safe(p, end, len, e_inval);
1315 if (len > 0) {
1316 dout("apply_incremental full map len %d, %p to %p\n",
1317 len, *p, end);
1318 return ceph_osdmap_decode(p, min(*p+len, end));
1319 }
1320
1321 /* new crush? */
1322 ceph_decode_32_safe(p, end, len, e_inval);
1323 if (len > 0) {
1324 newcrush = crush_decode(*p, min(*p+len, end));
1325 if (IS_ERR(newcrush)) {
1326 err = PTR_ERR(newcrush);
1327 newcrush = NULL;
1328 goto bad;
1329 }
1330 *p += len;
1331 }
1332
1333 /* new flags? */
1334 if (new_flags >= 0)
1335 map->flags = new_flags;
1336 if (new_pool_max >= 0)
1337 map->pool_max = new_pool_max;
1338
1339 /* new max? */
1340 ceph_decode_32_safe(p, end, max, e_inval);
1341 if (max >= 0) {
1342 err = osdmap_set_max_osd(map, max);
1343 if (err)
1344 goto bad;
1345 }
1346
1347 map->epoch++;
1348 map->modified = modified;
1349 if (newcrush) {
1350 if (map->crush)
1351 crush_destroy(map->crush);
1352 map->crush = newcrush;
1353 newcrush = NULL;
1354 }
1355
1356 /* new_pools */
1357 err = decode_new_pools(p, end, map);
1358 if (err)
1359 goto bad;
1360
1361 /* new_pool_names */
1362 err = decode_pool_names(p, end, map);
1363 if (err)
1364 goto bad;
1365
1366 /* old_pool */
1367 ceph_decode_32_safe(p, end, len, e_inval);
1368 while (len--) {
1369 struct ceph_pg_pool_info *pi;
1370
1371 ceph_decode_64_safe(p, end, pool, e_inval);
1372 pi = __lookup_pg_pool(&map->pg_pools, pool);
1373 if (pi)
1374 __remove_pg_pool(&map->pg_pools, pi);
1375 }
1376
1377 /* new_up_client, new_state, new_weight */
1378 err = decode_new_up_state_weight(p, end, map);
1379 if (err)
1380 goto bad;
1381
1382 /* new_pg_temp */
1383 err = decode_new_pg_temp(p, end, map);
1384 if (err)
1385 goto bad;
1386
1387 /* new_primary_temp */
1388 if (struct_v >= 1) {
1389 err = decode_new_primary_temp(p, end, map);
1390 if (err)
1391 goto bad;
1392 }
1393
1394 /* new_primary_affinity */
1395 if (struct_v >= 2) {
1396 err = decode_new_primary_affinity(p, end, map);
1397 if (err)
1398 goto bad;
1399 }
1400
1401 /* ignore the rest */
1402 *p = end;
1403
1404 dout("inc osdmap epoch %d max_osd %d\n", map->epoch, map->max_osd);
1405 return map;
1406
1407 e_inval:
1408 err = -EINVAL;
1409 bad:
1410 pr_err("corrupt inc osdmap (%d) epoch %d off %d (%p of %p-%p)\n",
1411 err, epoch, (int)(*p - start), *p, start, end);
1412 print_hex_dump(KERN_DEBUG, "osdmap: ",
1413 DUMP_PREFIX_OFFSET, 16, 1,
1414 start, end - start, true);
1415 if (newcrush)
1416 crush_destroy(newcrush);
1417 return ERR_PTR(err);
1418 }
1419
1420
1421
1422
1423 /*
1424 * calculate file layout from given offset, length.
1425 * fill in correct oid, logical length, and object extent
1426 * offset, length.
1427 *
1428 * for now, we write only a single su, until we can
1429 * pass a stride back to the caller.
1430 */
1431 int ceph_calc_file_object_mapping(struct ceph_file_layout *layout,
1432 u64 off, u64 len,
1433 u64 *ono,
1434 u64 *oxoff, u64 *oxlen)
1435 {
1436 u32 osize = le32_to_cpu(layout->fl_object_size);
1437 u32 su = le32_to_cpu(layout->fl_stripe_unit);
1438 u32 sc = le32_to_cpu(layout->fl_stripe_count);
1439 u32 bl, stripeno, stripepos, objsetno;
1440 u32 su_per_object;
1441 u64 t, su_offset;
1442
1443 dout("mapping %llu~%llu osize %u fl_su %u\n", off, len,
1444 osize, su);
1445 if (su == 0 || sc == 0)
1446 goto invalid;
1447 su_per_object = osize / su;
1448 if (su_per_object == 0)
1449 goto invalid;
1450 dout("osize %u / su %u = su_per_object %u\n", osize, su,
1451 su_per_object);
1452
1453 if ((su & ~PAGE_MASK) != 0)
1454 goto invalid;
1455
1456 /* bl = *off / su; */
1457 t = off;
1458 do_div(t, su);
1459 bl = t;
1460 dout("off %llu / su %u = bl %u\n", off, su, bl);
1461
1462 stripeno = bl / sc;
1463 stripepos = bl % sc;
1464 objsetno = stripeno / su_per_object;
1465
1466 *ono = objsetno * sc + stripepos;
1467 dout("objset %u * sc %u = ono %u\n", objsetno, sc, (unsigned int)*ono);
1468
1469 /* *oxoff = *off % layout->fl_stripe_unit; # offset in su */
1470 t = off;
1471 su_offset = do_div(t, su);
1472 *oxoff = su_offset + (stripeno % su_per_object) * su;
1473
1474 /*
1475 * Calculate the length of the extent being written to the selected
1476 * object. This is the minimum of the full length requested (len) or
1477 * the remainder of the current stripe being written to.
1478 */
1479 *oxlen = min_t(u64, len, su - su_offset);
1480
1481 dout(" obj extent %llu~%llu\n", *oxoff, *oxlen);
1482 return 0;
1483
1484 invalid:
1485 dout(" invalid layout\n");
1486 *ono = 0;
1487 *oxoff = 0;
1488 *oxlen = 0;
1489 return -EINVAL;
1490 }
1491 EXPORT_SYMBOL(ceph_calc_file_object_mapping);
1492
1493 /*
1494 * Calculate mapping of a (oloc, oid) pair to a PG. Should only be
1495 * called with target's (oloc, oid), since tiering isn't taken into
1496 * account.
1497 */
1498 int ceph_oloc_oid_to_pg(struct ceph_osdmap *osdmap,
1499 struct ceph_object_locator *oloc,
1500 struct ceph_object_id *oid,
1501 struct ceph_pg *pg_out)
1502 {
1503 struct ceph_pg_pool_info *pi;
1504
1505 pi = __lookup_pg_pool(&osdmap->pg_pools, oloc->pool);
1506 if (!pi)
1507 return -EIO;
1508
1509 pg_out->pool = oloc->pool;
1510 pg_out->seed = ceph_str_hash(pi->object_hash, oid->name,
1511 oid->name_len);
1512
1513 dout("%s '%.*s' pgid %llu.%x\n", __func__, oid->name_len, oid->name,
1514 pg_out->pool, pg_out->seed);
1515 return 0;
1516 }
1517 EXPORT_SYMBOL(ceph_oloc_oid_to_pg);
1518
1519 static int do_crush(struct ceph_osdmap *map, int ruleno, int x,
1520 int *result, int result_max,
1521 const __u32 *weight, int weight_max)
1522 {
1523 int r;
1524
1525 BUG_ON(result_max > CEPH_PG_MAX_SIZE);
1526
1527 mutex_lock(&map->crush_scratch_mutex);
1528 r = crush_do_rule(map->crush, ruleno, x, result, result_max,
1529 weight, weight_max, map->crush_scratch_ary);
1530 mutex_unlock(&map->crush_scratch_mutex);
1531
1532 return r;
1533 }
1534
1535 /*
1536 * Calculate raw (crush) set for given pgid.
1537 *
1538 * Return raw set length, or error.
1539 */
1540 static int pg_to_raw_osds(struct ceph_osdmap *osdmap,
1541 struct ceph_pg_pool_info *pool,
1542 struct ceph_pg pgid, u32 pps, int *osds)
1543 {
1544 int ruleno;
1545 int len;
1546
1547 /* crush */
1548 ruleno = crush_find_rule(osdmap->crush, pool->crush_ruleset,
1549 pool->type, pool->size);
1550 if (ruleno < 0) {
1551 pr_err("no crush rule: pool %lld ruleset %d type %d size %d\n",
1552 pgid.pool, pool->crush_ruleset, pool->type,
1553 pool->size);
1554 return -ENOENT;
1555 }
1556
1557 len = do_crush(osdmap, ruleno, pps, osds,
1558 min_t(int, pool->size, CEPH_PG_MAX_SIZE),
1559 osdmap->osd_weight, osdmap->max_osd);
1560 if (len < 0) {
1561 pr_err("error %d from crush rule %d: pool %lld ruleset %d type %d size %d\n",
1562 len, ruleno, pgid.pool, pool->crush_ruleset,
1563 pool->type, pool->size);
1564 return len;
1565 }
1566
1567 return len;
1568 }
1569
1570 /*
1571 * Given raw set, calculate up set and up primary.
1572 *
1573 * Return up set length. *primary is set to up primary osd id, or -1
1574 * if up set is empty.
1575 */
1576 static int raw_to_up_osds(struct ceph_osdmap *osdmap,
1577 struct ceph_pg_pool_info *pool,
1578 int *osds, int len, int *primary)
1579 {
1580 int up_primary = -1;
1581 int i;
1582
1583 if (ceph_can_shift_osds(pool)) {
1584 int removed = 0;
1585
1586 for (i = 0; i < len; i++) {
1587 if (ceph_osd_is_down(osdmap, osds[i])) {
1588 removed++;
1589 continue;
1590 }
1591 if (removed)
1592 osds[i - removed] = osds[i];
1593 }
1594
1595 len -= removed;
1596 if (len > 0)
1597 up_primary = osds[0];
1598 } else {
1599 for (i = len - 1; i >= 0; i--) {
1600 if (ceph_osd_is_down(osdmap, osds[i]))
1601 osds[i] = CRUSH_ITEM_NONE;
1602 else
1603 up_primary = osds[i];
1604 }
1605 }
1606
1607 *primary = up_primary;
1608 return len;
1609 }
1610
1611 static void apply_primary_affinity(struct ceph_osdmap *osdmap, u32 pps,
1612 struct ceph_pg_pool_info *pool,
1613 int *osds, int len, int *primary)
1614 {
1615 int i;
1616 int pos = -1;
1617
1618 /*
1619 * Do we have any non-default primary_affinity values for these
1620 * osds?
1621 */
1622 if (!osdmap->osd_primary_affinity)
1623 return;
1624
1625 for (i = 0; i < len; i++) {
1626 int osd = osds[i];
1627
1628 if (osd != CRUSH_ITEM_NONE &&
1629 osdmap->osd_primary_affinity[osd] !=
1630 CEPH_OSD_DEFAULT_PRIMARY_AFFINITY) {
1631 break;
1632 }
1633 }
1634 if (i == len)
1635 return;
1636
1637 /*
1638 * Pick the primary. Feed both the seed (for the pg) and the
1639 * osd into the hash/rng so that a proportional fraction of an
1640 * osd's pgs get rejected as primary.
1641 */
1642 for (i = 0; i < len; i++) {
1643 int osd = osds[i];
1644 u32 aff;
1645
1646 if (osd == CRUSH_ITEM_NONE)
1647 continue;
1648
1649 aff = osdmap->osd_primary_affinity[osd];
1650 if (aff < CEPH_OSD_MAX_PRIMARY_AFFINITY &&
1651 (crush_hash32_2(CRUSH_HASH_RJENKINS1,
1652 pps, osd) >> 16) >= aff) {
1653 /*
1654 * We chose not to use this primary. Note it
1655 * anyway as a fallback in case we don't pick
1656 * anyone else, but keep looking.
1657 */
1658 if (pos < 0)
1659 pos = i;
1660 } else {
1661 pos = i;
1662 break;
1663 }
1664 }
1665 if (pos < 0)
1666 return;
1667
1668 *primary = osds[pos];
1669
1670 if (ceph_can_shift_osds(pool) && pos > 0) {
1671 /* move the new primary to the front */
1672 for (i = pos; i > 0; i--)
1673 osds[i] = osds[i - 1];
1674 osds[0] = *primary;
1675 }
1676 }
1677
1678 /*
1679 * Given up set, apply pg_temp and primary_temp mappings.
1680 *
1681 * Return acting set length. *primary is set to acting primary osd id,
1682 * or -1 if acting set is empty.
1683 */
1684 static int apply_temps(struct ceph_osdmap *osdmap,
1685 struct ceph_pg_pool_info *pool, struct ceph_pg pgid,
1686 int *osds, int len, int *primary)
1687 {
1688 struct ceph_pg_mapping *pg;
1689 int temp_len;
1690 int temp_primary;
1691 int i;
1692
1693 /* raw_pg -> pg */
1694 pgid.seed = ceph_stable_mod(pgid.seed, pool->pg_num,
1695 pool->pg_num_mask);
1696
1697 /* pg_temp? */
1698 pg = __lookup_pg_mapping(&osdmap->pg_temp, pgid);
1699 if (pg) {
1700 temp_len = 0;
1701 temp_primary = -1;
1702
1703 for (i = 0; i < pg->pg_temp.len; i++) {
1704 if (ceph_osd_is_down(osdmap, pg->pg_temp.osds[i])) {
1705 if (ceph_can_shift_osds(pool))
1706 continue;
1707 else
1708 osds[temp_len++] = CRUSH_ITEM_NONE;
1709 } else {
1710 osds[temp_len++] = pg->pg_temp.osds[i];
1711 }
1712 }
1713
1714 /* apply pg_temp's primary */
1715 for (i = 0; i < temp_len; i++) {
1716 if (osds[i] != CRUSH_ITEM_NONE) {
1717 temp_primary = osds[i];
1718 break;
1719 }
1720 }
1721 } else {
1722 temp_len = len;
1723 temp_primary = *primary;
1724 }
1725
1726 /* primary_temp? */
1727 pg = __lookup_pg_mapping(&osdmap->primary_temp, pgid);
1728 if (pg)
1729 temp_primary = pg->primary_temp.osd;
1730
1731 *primary = temp_primary;
1732 return temp_len;
1733 }
1734
1735 /*
1736 * Calculate acting set for given pgid.
1737 *
1738 * Return acting set length, or error. *primary is set to acting
1739 * primary osd id, or -1 if acting set is empty or on error.
1740 */
1741 int ceph_calc_pg_acting(struct ceph_osdmap *osdmap, struct ceph_pg pgid,
1742 int *osds, int *primary)
1743 {
1744 struct ceph_pg_pool_info *pool;
1745 u32 pps;
1746 int len;
1747
1748 pool = __lookup_pg_pool(&osdmap->pg_pools, pgid.pool);
1749 if (!pool) {
1750 *primary = -1;
1751 return -ENOENT;
1752 }
1753
1754 if (pool->flags & CEPH_POOL_FLAG_HASHPSPOOL) {
1755 /* hash pool id and seed so that pool PGs do not overlap */
1756 pps = crush_hash32_2(CRUSH_HASH_RJENKINS1,
1757 ceph_stable_mod(pgid.seed, pool->pgp_num,
1758 pool->pgp_num_mask),
1759 pgid.pool);
1760 } else {
1761 /*
1762 * legacy behavior: add ps and pool together. this is
1763 * not a great approach because the PGs from each pool
1764 * will overlap on top of each other: 0.5 == 1.4 ==
1765 * 2.3 == ...
1766 */
1767 pps = ceph_stable_mod(pgid.seed, pool->pgp_num,
1768 pool->pgp_num_mask) +
1769 (unsigned)pgid.pool;
1770 }
1771
1772 len = pg_to_raw_osds(osdmap, pool, pgid, pps, osds);
1773 if (len < 0) {
1774 *primary = -1;
1775 return len;
1776 }
1777
1778 len = raw_to_up_osds(osdmap, pool, osds, len, primary);
1779
1780 apply_primary_affinity(osdmap, pps, pool, osds, len, primary);
1781
1782 len = apply_temps(osdmap, pool, pgid, osds, len, primary);
1783
1784 return len;
1785 }
1786
1787 /*
1788 * Return primary osd for given pgid, or -1 if none.
1789 */
1790 int ceph_calc_pg_primary(struct ceph_osdmap *osdmap, struct ceph_pg pgid)
1791 {
1792 int osds[CEPH_PG_MAX_SIZE];
1793 int primary;
1794
1795 ceph_calc_pg_acting(osdmap, pgid, osds, &primary);
1796
1797 return primary;
1798 }
1799 EXPORT_SYMBOL(ceph_calc_pg_primary);
Linux with added FPC-III support