]> git.kernelconcepts.de Git - karo-tx-linux.git/blob - net/ceph/osdmap.c
Merge tag 'nfs-for-4.13-5' of git://git.linux-nfs.org/projects/anna/linux-nfs
[karo-tx-linux.git] / net / ceph / osdmap.c
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, 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
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 crush_decode_straw2_bucket(void **p, void *end,
126                                       struct crush_bucket_straw2 *b)
127 {
128         int j;
129         dout("crush_decode_straw2_bucket %p to %p\n", *p, end);
130         b->item_weights = kcalloc(b->h.size, sizeof(u32), GFP_NOFS);
131         if (b->item_weights == NULL)
132                 return -ENOMEM;
133         ceph_decode_need(p, end, b->h.size * sizeof(u32), bad);
134         for (j = 0; j < b->h.size; j++)
135                 b->item_weights[j] = ceph_decode_32(p);
136         return 0;
137 bad:
138         return -EINVAL;
139 }
140
141 static 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
153 static 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
173 DEFINE_RB_FUNCS(choose_arg_map, struct crush_choose_arg_map, choose_args_index,
174                 node);
175
176 void 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
188 static 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
212 e_inval:
213         ret = -EINVAL;
214 fail:
215         kfree(a);
216         return ERR_PTR(ret);
217 }
218
219 /*
220  * Assumes @arg is zero-initialized.
221  */
222 static 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
257 e_inval:
258         return -EINVAL;
259 }
260
261 static 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
309 e_inval:
310         ret = -EINVAL;
311 fail:
312         free_choose_arg_map(arg_map);
313         return ret;
314 }
315
316 static 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
342 static 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 = c->rules[i] = kmalloc(sizeof(*r) +
499                                           yes*sizeof(struct crush_rule_step),
500                                           GFP_NOFS);
501                 if (r == NULL)
502                         goto badmem;
503                 dout(" rule %d is at %p\n", i, r);
504                 r->len = yes;
505                 ceph_decode_copy_safe(p, end, &r->mask, 4, bad); /* 4 u8's */
506                 ceph_decode_need(p, end, r->len*3*sizeof(u32), bad);
507                 for (j = 0; j < r->len; j++) {
508                         r->steps[j].op = ceph_decode_32(p);
509                         r->steps[j].arg1 = ceph_decode_32(p);
510                         r->steps[j].arg2 = ceph_decode_32(p);
511                 }
512         }
513
514         ceph_decode_skip_map(p, end, 32, string, bad); /* type_map */
515         ceph_decode_skip_map(p, end, 32, string, bad); /* name_map */
516         ceph_decode_skip_map(p, end, 32, string, bad); /* rule_name_map */
517
518         /* tunables */
519         ceph_decode_need(p, end, 3*sizeof(u32), done);
520         c->choose_local_tries = ceph_decode_32(p);
521         c->choose_local_fallback_tries =  ceph_decode_32(p);
522         c->choose_total_tries = ceph_decode_32(p);
523         dout("crush decode tunable choose_local_tries = %d\n",
524              c->choose_local_tries);
525         dout("crush decode tunable choose_local_fallback_tries = %d\n",
526              c->choose_local_fallback_tries);
527         dout("crush decode tunable choose_total_tries = %d\n",
528              c->choose_total_tries);
529
530         ceph_decode_need(p, end, sizeof(u32), done);
531         c->chooseleaf_descend_once = ceph_decode_32(p);
532         dout("crush decode tunable chooseleaf_descend_once = %d\n",
533              c->chooseleaf_descend_once);
534
535         ceph_decode_need(p, end, sizeof(u8), done);
536         c->chooseleaf_vary_r = ceph_decode_8(p);
537         dout("crush decode tunable chooseleaf_vary_r = %d\n",
538              c->chooseleaf_vary_r);
539
540         /* skip straw_calc_version, allowed_bucket_algs */
541         ceph_decode_need(p, end, sizeof(u8) + sizeof(u32), done);
542         *p += sizeof(u8) + sizeof(u32);
543
544         ceph_decode_need(p, end, sizeof(u8), done);
545         c->chooseleaf_stable = ceph_decode_8(p);
546         dout("crush decode tunable chooseleaf_stable = %d\n",
547              c->chooseleaf_stable);
548
549         if (*p != end) {
550                 /* class_map */
551                 ceph_decode_skip_map(p, end, 32, 32, bad);
552                 /* class_name */
553                 ceph_decode_skip_map(p, end, 32, string, bad);
554                 /* class_bucket */
555                 ceph_decode_skip_map_of_map(p, end, 32, 32, 32, bad);
556         }
557
558         if (*p != end) {
559                 err = decode_choose_args(p, end, c);
560                 if (err)
561                         goto fail;
562         }
563
564 done:
565         crush_finalize(c);
566         dout("crush_decode success\n");
567         return c;
568
569 badmem:
570         err = -ENOMEM;
571 fail:
572         dout("crush_decode fail %d\n", err);
573         crush_destroy(c);
574         return ERR_PTR(err);
575
576 bad:
577         err = -EINVAL;
578         goto fail;
579 }
580
581 int ceph_pg_compare(const struct ceph_pg *lhs, const struct ceph_pg *rhs)
582 {
583         if (lhs->pool < rhs->pool)
584                 return -1;
585         if (lhs->pool > rhs->pool)
586                 return 1;
587         if (lhs->seed < rhs->seed)
588                 return -1;
589         if (lhs->seed > rhs->seed)
590                 return 1;
591
592         return 0;
593 }
594
595 int ceph_spg_compare(const struct ceph_spg *lhs, const struct ceph_spg *rhs)
596 {
597         int ret;
598
599         ret = ceph_pg_compare(&lhs->pgid, &rhs->pgid);
600         if (ret)
601                 return ret;
602
603         if (lhs->shard < rhs->shard)
604                 return -1;
605         if (lhs->shard > rhs->shard)
606                 return 1;
607
608         return 0;
609 }
610
611 static struct ceph_pg_mapping *alloc_pg_mapping(size_t payload_len)
612 {
613         struct ceph_pg_mapping *pg;
614
615         pg = kmalloc(sizeof(*pg) + payload_len, GFP_NOIO);
616         if (!pg)
617                 return NULL;
618
619         RB_CLEAR_NODE(&pg->node);
620         return pg;
621 }
622
623 static void free_pg_mapping(struct ceph_pg_mapping *pg)
624 {
625         WARN_ON(!RB_EMPTY_NODE(&pg->node));
626
627         kfree(pg);
628 }
629
630 /*
631  * rbtree of pg_mapping for handling pg_temp (explicit mapping of pgid
632  * to a set of osds) and primary_temp (explicit primary setting)
633  */
634 DEFINE_RB_FUNCS2(pg_mapping, struct ceph_pg_mapping, pgid, ceph_pg_compare,
635                  RB_BYPTR, const struct ceph_pg *, node)
636
637 /*
638  * rbtree of pg pool info
639  */
640 static int __insert_pg_pool(struct rb_root *root, struct ceph_pg_pool_info *new)
641 {
642         struct rb_node **p = &root->rb_node;
643         struct rb_node *parent = NULL;
644         struct ceph_pg_pool_info *pi = NULL;
645
646         while (*p) {
647                 parent = *p;
648                 pi = rb_entry(parent, struct ceph_pg_pool_info, node);
649                 if (new->id < pi->id)
650                         p = &(*p)->rb_left;
651                 else if (new->id > pi->id)
652                         p = &(*p)->rb_right;
653                 else
654                         return -EEXIST;
655         }
656
657         rb_link_node(&new->node, parent, p);
658         rb_insert_color(&new->node, root);
659         return 0;
660 }
661
662 static struct ceph_pg_pool_info *__lookup_pg_pool(struct rb_root *root, u64 id)
663 {
664         struct ceph_pg_pool_info *pi;
665         struct rb_node *n = root->rb_node;
666
667         while (n) {
668                 pi = rb_entry(n, struct ceph_pg_pool_info, node);
669                 if (id < pi->id)
670                         n = n->rb_left;
671                 else if (id > pi->id)
672                         n = n->rb_right;
673                 else
674                         return pi;
675         }
676         return NULL;
677 }
678
679 struct ceph_pg_pool_info *ceph_pg_pool_by_id(struct ceph_osdmap *map, u64 id)
680 {
681         return __lookup_pg_pool(&map->pg_pools, id);
682 }
683
684 const char *ceph_pg_pool_name_by_id(struct ceph_osdmap *map, u64 id)
685 {
686         struct ceph_pg_pool_info *pi;
687
688         if (id == CEPH_NOPOOL)
689                 return NULL;
690
691         if (WARN_ON_ONCE(id > (u64) INT_MAX))
692                 return NULL;
693
694         pi = __lookup_pg_pool(&map->pg_pools, (int) id);
695
696         return pi ? pi->name : NULL;
697 }
698 EXPORT_SYMBOL(ceph_pg_pool_name_by_id);
699
700 int ceph_pg_poolid_by_name(struct ceph_osdmap *map, const char *name)
701 {
702         struct rb_node *rbp;
703
704         for (rbp = rb_first(&map->pg_pools); rbp; rbp = rb_next(rbp)) {
705                 struct ceph_pg_pool_info *pi =
706                         rb_entry(rbp, struct ceph_pg_pool_info, node);
707                 if (pi->name && strcmp(pi->name, name) == 0)
708                         return pi->id;
709         }
710         return -ENOENT;
711 }
712 EXPORT_SYMBOL(ceph_pg_poolid_by_name);
713
714 static void __remove_pg_pool(struct rb_root *root, struct ceph_pg_pool_info *pi)
715 {
716         rb_erase(&pi->node, root);
717         kfree(pi->name);
718         kfree(pi);
719 }
720
721 static int decode_pool(void **p, void *end, struct ceph_pg_pool_info *pi)
722 {
723         u8 ev, cv;
724         unsigned len, num;
725         void *pool_end;
726
727         ceph_decode_need(p, end, 2 + 4, bad);
728         ev = ceph_decode_8(p);  /* encoding version */
729         cv = ceph_decode_8(p); /* compat version */
730         if (ev < 5) {
731                 pr_warn("got v %d < 5 cv %d of ceph_pg_pool\n", ev, cv);
732                 return -EINVAL;
733         }
734         if (cv > 9) {
735                 pr_warn("got v %d cv %d > 9 of ceph_pg_pool\n", ev, cv);
736                 return -EINVAL;
737         }
738         len = ceph_decode_32(p);
739         ceph_decode_need(p, end, len, bad);
740         pool_end = *p + len;
741
742         pi->type = ceph_decode_8(p);
743         pi->size = ceph_decode_8(p);
744         pi->crush_ruleset = ceph_decode_8(p);
745         pi->object_hash = ceph_decode_8(p);
746
747         pi->pg_num = ceph_decode_32(p);
748         pi->pgp_num = ceph_decode_32(p);
749
750         *p += 4 + 4;  /* skip lpg* */
751         *p += 4;      /* skip last_change */
752         *p += 8 + 4;  /* skip snap_seq, snap_epoch */
753
754         /* skip snaps */
755         num = ceph_decode_32(p);
756         while (num--) {
757                 *p += 8;  /* snapid key */
758                 *p += 1 + 1; /* versions */
759                 len = ceph_decode_32(p);
760                 *p += len;
761         }
762
763         /* skip removed_snaps */
764         num = ceph_decode_32(p);
765         *p += num * (8 + 8);
766
767         *p += 8;  /* skip auid */
768         pi->flags = ceph_decode_64(p);
769         *p += 4;  /* skip crash_replay_interval */
770
771         if (ev >= 7)
772                 pi->min_size = ceph_decode_8(p);
773         else
774                 pi->min_size = pi->size - pi->size / 2;
775
776         if (ev >= 8)
777                 *p += 8 + 8;  /* skip quota_max_* */
778
779         if (ev >= 9) {
780                 /* skip tiers */
781                 num = ceph_decode_32(p);
782                 *p += num * 8;
783
784                 *p += 8;  /* skip tier_of */
785                 *p += 1;  /* skip cache_mode */
786
787                 pi->read_tier = ceph_decode_64(p);
788                 pi->write_tier = ceph_decode_64(p);
789         } else {
790                 pi->read_tier = -1;
791                 pi->write_tier = -1;
792         }
793
794         if (ev >= 10) {
795                 /* skip properties */
796                 num = ceph_decode_32(p);
797                 while (num--) {
798                         len = ceph_decode_32(p);
799                         *p += len; /* key */
800                         len = ceph_decode_32(p);
801                         *p += len; /* val */
802                 }
803         }
804
805         if (ev >= 11) {
806                 /* skip hit_set_params */
807                 *p += 1 + 1; /* versions */
808                 len = ceph_decode_32(p);
809                 *p += len;
810
811                 *p += 4; /* skip hit_set_period */
812                 *p += 4; /* skip hit_set_count */
813         }
814
815         if (ev >= 12)
816                 *p += 4; /* skip stripe_width */
817
818         if (ev >= 13) {
819                 *p += 8; /* skip target_max_bytes */
820                 *p += 8; /* skip target_max_objects */
821                 *p += 4; /* skip cache_target_dirty_ratio_micro */
822                 *p += 4; /* skip cache_target_full_ratio_micro */
823                 *p += 4; /* skip cache_min_flush_age */
824                 *p += 4; /* skip cache_min_evict_age */
825         }
826
827         if (ev >=  14) {
828                 /* skip erasure_code_profile */
829                 len = ceph_decode_32(p);
830                 *p += len;
831         }
832
833         /*
834          * last_force_op_resend_preluminous, will be overridden if the
835          * map was encoded with RESEND_ON_SPLIT
836          */
837         if (ev >= 15)
838                 pi->last_force_request_resend = ceph_decode_32(p);
839         else
840                 pi->last_force_request_resend = 0;
841
842         if (ev >= 16)
843                 *p += 4; /* skip min_read_recency_for_promote */
844
845         if (ev >= 17)
846                 *p += 8; /* skip expected_num_objects */
847
848         if (ev >= 19)
849                 *p += 4; /* skip cache_target_dirty_high_ratio_micro */
850
851         if (ev >= 20)
852                 *p += 4; /* skip min_write_recency_for_promote */
853
854         if (ev >= 21)
855                 *p += 1; /* skip use_gmt_hitset */
856
857         if (ev >= 22)
858                 *p += 1; /* skip fast_read */
859
860         if (ev >= 23) {
861                 *p += 4; /* skip hit_set_grade_decay_rate */
862                 *p += 4; /* skip hit_set_search_last_n */
863         }
864
865         if (ev >= 24) {
866                 /* skip opts */
867                 *p += 1 + 1; /* versions */
868                 len = ceph_decode_32(p);
869                 *p += len;
870         }
871
872         if (ev >= 25)
873                 pi->last_force_request_resend = ceph_decode_32(p);
874
875         /* ignore the rest */
876
877         *p = pool_end;
878         calc_pg_masks(pi);
879         return 0;
880
881 bad:
882         return -EINVAL;
883 }
884
885 static int decode_pool_names(void **p, void *end, struct ceph_osdmap *map)
886 {
887         struct ceph_pg_pool_info *pi;
888         u32 num, len;
889         u64 pool;
890
891         ceph_decode_32_safe(p, end, num, bad);
892         dout(" %d pool names\n", num);
893         while (num--) {
894                 ceph_decode_64_safe(p, end, pool, bad);
895                 ceph_decode_32_safe(p, end, len, bad);
896                 dout("  pool %llu len %d\n", pool, len);
897                 ceph_decode_need(p, end, len, bad);
898                 pi = __lookup_pg_pool(&map->pg_pools, pool);
899                 if (pi) {
900                         char *name = kstrndup(*p, len, GFP_NOFS);
901
902                         if (!name)
903                                 return -ENOMEM;
904                         kfree(pi->name);
905                         pi->name = name;
906                         dout("  name is %s\n", pi->name);
907                 }
908                 *p += len;
909         }
910         return 0;
911
912 bad:
913         return -EINVAL;
914 }
915
916 /*
917  * osd map
918  */
919 struct ceph_osdmap *ceph_osdmap_alloc(void)
920 {
921         struct ceph_osdmap *map;
922
923         map = kzalloc(sizeof(*map), GFP_NOIO);
924         if (!map)
925                 return NULL;
926
927         map->pg_pools = RB_ROOT;
928         map->pool_max = -1;
929         map->pg_temp = RB_ROOT;
930         map->primary_temp = RB_ROOT;
931         map->pg_upmap = RB_ROOT;
932         map->pg_upmap_items = RB_ROOT;
933         mutex_init(&map->crush_workspace_mutex);
934
935         return map;
936 }
937
938 void ceph_osdmap_destroy(struct ceph_osdmap *map)
939 {
940         dout("osdmap_destroy %p\n", map);
941         if (map->crush)
942                 crush_destroy(map->crush);
943         while (!RB_EMPTY_ROOT(&map->pg_temp)) {
944                 struct ceph_pg_mapping *pg =
945                         rb_entry(rb_first(&map->pg_temp),
946                                  struct ceph_pg_mapping, node);
947                 erase_pg_mapping(&map->pg_temp, pg);
948                 free_pg_mapping(pg);
949         }
950         while (!RB_EMPTY_ROOT(&map->primary_temp)) {
951                 struct ceph_pg_mapping *pg =
952                         rb_entry(rb_first(&map->primary_temp),
953                                  struct ceph_pg_mapping, node);
954                 erase_pg_mapping(&map->primary_temp, pg);
955                 free_pg_mapping(pg);
956         }
957         while (!RB_EMPTY_ROOT(&map->pg_upmap)) {
958                 struct ceph_pg_mapping *pg =
959                         rb_entry(rb_first(&map->pg_upmap),
960                                  struct ceph_pg_mapping, node);
961                 rb_erase(&pg->node, &map->pg_upmap);
962                 kfree(pg);
963         }
964         while (!RB_EMPTY_ROOT(&map->pg_upmap_items)) {
965                 struct ceph_pg_mapping *pg =
966                         rb_entry(rb_first(&map->pg_upmap_items),
967                                  struct ceph_pg_mapping, node);
968                 rb_erase(&pg->node, &map->pg_upmap_items);
969                 kfree(pg);
970         }
971         while (!RB_EMPTY_ROOT(&map->pg_pools)) {
972                 struct ceph_pg_pool_info *pi =
973                         rb_entry(rb_first(&map->pg_pools),
974                                  struct ceph_pg_pool_info, node);
975                 __remove_pg_pool(&map->pg_pools, pi);
976         }
977         kfree(map->osd_state);
978         kfree(map->osd_weight);
979         kfree(map->osd_addr);
980         kfree(map->osd_primary_affinity);
981         kfree(map->crush_workspace);
982         kfree(map);
983 }
984
985 /*
986  * Adjust max_osd value, (re)allocate arrays.
987  *
988  * The new elements are properly initialized.
989  */
990 static int osdmap_set_max_osd(struct ceph_osdmap *map, int max)
991 {
992         u32 *state;
993         u32 *weight;
994         struct ceph_entity_addr *addr;
995         int i;
996
997         state = krealloc(map->osd_state, max*sizeof(*state), GFP_NOFS);
998         if (!state)
999                 return -ENOMEM;
1000         map->osd_state = state;
1001
1002         weight = krealloc(map->osd_weight, max*sizeof(*weight), GFP_NOFS);
1003         if (!weight)
1004                 return -ENOMEM;
1005         map->osd_weight = weight;
1006
1007         addr = krealloc(map->osd_addr, max*sizeof(*addr), GFP_NOFS);
1008         if (!addr)
1009                 return -ENOMEM;
1010         map->osd_addr = addr;
1011
1012         for (i = map->max_osd; i < max; i++) {
1013                 map->osd_state[i] = 0;
1014                 map->osd_weight[i] = CEPH_OSD_OUT;
1015                 memset(map->osd_addr + i, 0, sizeof(*map->osd_addr));
1016         }
1017
1018         if (map->osd_primary_affinity) {
1019                 u32 *affinity;
1020
1021                 affinity = krealloc(map->osd_primary_affinity,
1022                                     max*sizeof(*affinity), GFP_NOFS);
1023                 if (!affinity)
1024                         return -ENOMEM;
1025                 map->osd_primary_affinity = affinity;
1026
1027                 for (i = map->max_osd; i < max; i++)
1028                         map->osd_primary_affinity[i] =
1029                             CEPH_OSD_DEFAULT_PRIMARY_AFFINITY;
1030         }
1031
1032         map->max_osd = max;
1033
1034         return 0;
1035 }
1036
1037 static int osdmap_set_crush(struct ceph_osdmap *map, struct crush_map *crush)
1038 {
1039         void *workspace;
1040         size_t work_size;
1041
1042         if (IS_ERR(crush))
1043                 return PTR_ERR(crush);
1044
1045         work_size = crush_work_size(crush, CEPH_PG_MAX_SIZE);
1046         dout("%s work_size %zu bytes\n", __func__, work_size);
1047         workspace = kmalloc(work_size, GFP_NOIO);
1048         if (!workspace) {
1049                 crush_destroy(crush);
1050                 return -ENOMEM;
1051         }
1052         crush_init_workspace(crush, workspace);
1053
1054         if (map->crush)
1055                 crush_destroy(map->crush);
1056         kfree(map->crush_workspace);
1057         map->crush = crush;
1058         map->crush_workspace = workspace;
1059         return 0;
1060 }
1061
1062 #define OSDMAP_WRAPPER_COMPAT_VER       7
1063 #define OSDMAP_CLIENT_DATA_COMPAT_VER   1
1064
1065 /*
1066  * Return 0 or error.  On success, *v is set to 0 for old (v6) osdmaps,
1067  * to struct_v of the client_data section for new (v7 and above)
1068  * osdmaps.
1069  */
1070 static int get_osdmap_client_data_v(void **p, void *end,
1071                                     const char *prefix, u8 *v)
1072 {
1073         u8 struct_v;
1074
1075         ceph_decode_8_safe(p, end, struct_v, e_inval);
1076         if (struct_v >= 7) {
1077                 u8 struct_compat;
1078
1079                 ceph_decode_8_safe(p, end, struct_compat, e_inval);
1080                 if (struct_compat > OSDMAP_WRAPPER_COMPAT_VER) {
1081                         pr_warn("got v %d cv %d > %d of %s ceph_osdmap\n",
1082                                 struct_v, struct_compat,
1083                                 OSDMAP_WRAPPER_COMPAT_VER, prefix);
1084                         return -EINVAL;
1085                 }
1086                 *p += 4; /* ignore wrapper struct_len */
1087
1088                 ceph_decode_8_safe(p, end, struct_v, e_inval);
1089                 ceph_decode_8_safe(p, end, struct_compat, e_inval);
1090                 if (struct_compat > OSDMAP_CLIENT_DATA_COMPAT_VER) {
1091                         pr_warn("got v %d cv %d > %d of %s ceph_osdmap client data\n",
1092                                 struct_v, struct_compat,
1093                                 OSDMAP_CLIENT_DATA_COMPAT_VER, prefix);
1094                         return -EINVAL;
1095                 }
1096                 *p += 4; /* ignore client data struct_len */
1097         } else {
1098                 u16 version;
1099
1100                 *p -= 1;
1101                 ceph_decode_16_safe(p, end, version, e_inval);
1102                 if (version < 6) {
1103                         pr_warn("got v %d < 6 of %s ceph_osdmap\n",
1104                                 version, prefix);
1105                         return -EINVAL;
1106                 }
1107
1108                 /* old osdmap enconding */
1109                 struct_v = 0;
1110         }
1111
1112         *v = struct_v;
1113         return 0;
1114
1115 e_inval:
1116         return -EINVAL;
1117 }
1118
1119 static int __decode_pools(void **p, void *end, struct ceph_osdmap *map,
1120                           bool incremental)
1121 {
1122         u32 n;
1123
1124         ceph_decode_32_safe(p, end, n, e_inval);
1125         while (n--) {
1126                 struct ceph_pg_pool_info *pi;
1127                 u64 pool;
1128                 int ret;
1129
1130                 ceph_decode_64_safe(p, end, pool, e_inval);
1131
1132                 pi = __lookup_pg_pool(&map->pg_pools, pool);
1133                 if (!incremental || !pi) {
1134                         pi = kzalloc(sizeof(*pi), GFP_NOFS);
1135                         if (!pi)
1136                                 return -ENOMEM;
1137
1138                         pi->id = pool;
1139
1140                         ret = __insert_pg_pool(&map->pg_pools, pi);
1141                         if (ret) {
1142                                 kfree(pi);
1143                                 return ret;
1144                         }
1145                 }
1146
1147                 ret = decode_pool(p, end, pi);
1148                 if (ret)
1149                         return ret;
1150         }
1151
1152         return 0;
1153
1154 e_inval:
1155         return -EINVAL;
1156 }
1157
1158 static int decode_pools(void **p, void *end, struct ceph_osdmap *map)
1159 {
1160         return __decode_pools(p, end, map, false);
1161 }
1162
1163 static int decode_new_pools(void **p, void *end, struct ceph_osdmap *map)
1164 {
1165         return __decode_pools(p, end, map, true);
1166 }
1167
1168 typedef struct ceph_pg_mapping *(*decode_mapping_fn_t)(void **, void *, bool);
1169
1170 static int decode_pg_mapping(void **p, void *end, struct rb_root *mapping_root,
1171                              decode_mapping_fn_t fn, bool incremental)
1172 {
1173         u32 n;
1174
1175         WARN_ON(!incremental && !fn);
1176
1177         ceph_decode_32_safe(p, end, n, e_inval);
1178         while (n--) {
1179                 struct ceph_pg_mapping *pg;
1180                 struct ceph_pg pgid;
1181                 int ret;
1182
1183                 ret = ceph_decode_pgid(p, end, &pgid);
1184                 if (ret)
1185                         return ret;
1186
1187                 pg = lookup_pg_mapping(mapping_root, &pgid);
1188                 if (pg) {
1189                         WARN_ON(!incremental);
1190                         erase_pg_mapping(mapping_root, pg);
1191                         free_pg_mapping(pg);
1192                 }
1193
1194                 if (fn) {
1195                         pg = fn(p, end, incremental);
1196                         if (IS_ERR(pg))
1197                                 return PTR_ERR(pg);
1198
1199                         if (pg) {
1200                                 pg->pgid = pgid; /* struct */
1201                                 insert_pg_mapping(mapping_root, pg);
1202                         }
1203                 }
1204         }
1205
1206         return 0;
1207
1208 e_inval:
1209         return -EINVAL;
1210 }
1211
1212 static struct ceph_pg_mapping *__decode_pg_temp(void **p, void *end,
1213                                                 bool incremental)
1214 {
1215         struct ceph_pg_mapping *pg;
1216         u32 len, i;
1217
1218         ceph_decode_32_safe(p, end, len, e_inval);
1219         if (len == 0 && incremental)
1220                 return NULL;    /* new_pg_temp: [] to remove */
1221         if (len > (SIZE_MAX - sizeof(*pg)) / sizeof(u32))
1222                 return ERR_PTR(-EINVAL);
1223
1224         ceph_decode_need(p, end, len * sizeof(u32), e_inval);
1225         pg = alloc_pg_mapping(len * sizeof(u32));
1226         if (!pg)
1227                 return ERR_PTR(-ENOMEM);
1228
1229         pg->pg_temp.len = len;
1230         for (i = 0; i < len; i++)
1231                 pg->pg_temp.osds[i] = ceph_decode_32(p);
1232
1233         return pg;
1234
1235 e_inval:
1236         return ERR_PTR(-EINVAL);
1237 }
1238
1239 static int decode_pg_temp(void **p, void *end, struct ceph_osdmap *map)
1240 {
1241         return decode_pg_mapping(p, end, &map->pg_temp, __decode_pg_temp,
1242                                  false);
1243 }
1244
1245 static int decode_new_pg_temp(void **p, void *end, struct ceph_osdmap *map)
1246 {
1247         return decode_pg_mapping(p, end, &map->pg_temp, __decode_pg_temp,
1248                                  true);
1249 }
1250
1251 static struct ceph_pg_mapping *__decode_primary_temp(void **p, void *end,
1252                                                      bool incremental)
1253 {
1254         struct ceph_pg_mapping *pg;
1255         u32 osd;
1256
1257         ceph_decode_32_safe(p, end, osd, e_inval);
1258         if (osd == (u32)-1 && incremental)
1259                 return NULL;    /* new_primary_temp: -1 to remove */
1260
1261         pg = alloc_pg_mapping(0);
1262         if (!pg)
1263                 return ERR_PTR(-ENOMEM);
1264
1265         pg->primary_temp.osd = osd;
1266         return pg;
1267
1268 e_inval:
1269         return ERR_PTR(-EINVAL);
1270 }
1271
1272 static int decode_primary_temp(void **p, void *end, struct ceph_osdmap *map)
1273 {
1274         return decode_pg_mapping(p, end, &map->primary_temp,
1275                                  __decode_primary_temp, false);
1276 }
1277
1278 static int decode_new_primary_temp(void **p, void *end,
1279                                    struct ceph_osdmap *map)
1280 {
1281         return decode_pg_mapping(p, end, &map->primary_temp,
1282                                  __decode_primary_temp, true);
1283 }
1284
1285 u32 ceph_get_primary_affinity(struct ceph_osdmap *map, int osd)
1286 {
1287         BUG_ON(osd >= map->max_osd);
1288
1289         if (!map->osd_primary_affinity)
1290                 return CEPH_OSD_DEFAULT_PRIMARY_AFFINITY;
1291
1292         return map->osd_primary_affinity[osd];
1293 }
1294
1295 static int set_primary_affinity(struct ceph_osdmap *map, int osd, u32 aff)
1296 {
1297         BUG_ON(osd >= map->max_osd);
1298
1299         if (!map->osd_primary_affinity) {
1300                 int i;
1301
1302                 map->osd_primary_affinity = kmalloc(map->max_osd*sizeof(u32),
1303                                                     GFP_NOFS);
1304                 if (!map->osd_primary_affinity)
1305                         return -ENOMEM;
1306
1307                 for (i = 0; i < map->max_osd; i++)
1308                         map->osd_primary_affinity[i] =
1309                             CEPH_OSD_DEFAULT_PRIMARY_AFFINITY;
1310         }
1311
1312         map->osd_primary_affinity[osd] = aff;
1313
1314         return 0;
1315 }
1316
1317 static int decode_primary_affinity(void **p, void *end,
1318                                    struct ceph_osdmap *map)
1319 {
1320         u32 len, i;
1321
1322         ceph_decode_32_safe(p, end, len, e_inval);
1323         if (len == 0) {
1324                 kfree(map->osd_primary_affinity);
1325                 map->osd_primary_affinity = NULL;
1326                 return 0;
1327         }
1328         if (len != map->max_osd)
1329                 goto e_inval;
1330
1331         ceph_decode_need(p, end, map->max_osd*sizeof(u32), e_inval);
1332
1333         for (i = 0; i < map->max_osd; i++) {
1334                 int ret;
1335
1336                 ret = set_primary_affinity(map, i, ceph_decode_32(p));
1337                 if (ret)
1338                         return ret;
1339         }
1340
1341         return 0;
1342
1343 e_inval:
1344         return -EINVAL;
1345 }
1346
1347 static int decode_new_primary_affinity(void **p, void *end,
1348                                        struct ceph_osdmap *map)
1349 {
1350         u32 n;
1351
1352         ceph_decode_32_safe(p, end, n, e_inval);
1353         while (n--) {
1354                 u32 osd, aff;
1355                 int ret;
1356
1357                 ceph_decode_32_safe(p, end, osd, e_inval);
1358                 ceph_decode_32_safe(p, end, aff, e_inval);
1359
1360                 ret = set_primary_affinity(map, osd, aff);
1361                 if (ret)
1362                         return ret;
1363
1364                 pr_info("osd%d primary-affinity 0x%x\n", osd, aff);
1365         }
1366
1367         return 0;
1368
1369 e_inval:
1370         return -EINVAL;
1371 }
1372
1373 static struct ceph_pg_mapping *__decode_pg_upmap(void **p, void *end,
1374                                                  bool __unused)
1375 {
1376         return __decode_pg_temp(p, end, false);
1377 }
1378
1379 static int decode_pg_upmap(void **p, void *end, struct ceph_osdmap *map)
1380 {
1381         return decode_pg_mapping(p, end, &map->pg_upmap, __decode_pg_upmap,
1382                                  false);
1383 }
1384
1385 static int decode_new_pg_upmap(void **p, void *end, struct ceph_osdmap *map)
1386 {
1387         return decode_pg_mapping(p, end, &map->pg_upmap, __decode_pg_upmap,
1388                                  true);
1389 }
1390
1391 static int decode_old_pg_upmap(void **p, void *end, struct ceph_osdmap *map)
1392 {
1393         return decode_pg_mapping(p, end, &map->pg_upmap, NULL, true);
1394 }
1395
1396 static struct ceph_pg_mapping *__decode_pg_upmap_items(void **p, void *end,
1397                                                        bool __unused)
1398 {
1399         struct ceph_pg_mapping *pg;
1400         u32 len, i;
1401
1402         ceph_decode_32_safe(p, end, len, e_inval);
1403         if (len > (SIZE_MAX - sizeof(*pg)) / (2 * sizeof(u32)))
1404                 return ERR_PTR(-EINVAL);
1405
1406         ceph_decode_need(p, end, 2 * len * sizeof(u32), e_inval);
1407         pg = alloc_pg_mapping(2 * len * sizeof(u32));
1408         if (!pg)
1409                 return ERR_PTR(-ENOMEM);
1410
1411         pg->pg_upmap_items.len = len;
1412         for (i = 0; i < len; i++) {
1413                 pg->pg_upmap_items.from_to[i][0] = ceph_decode_32(p);
1414                 pg->pg_upmap_items.from_to[i][1] = ceph_decode_32(p);
1415         }
1416
1417         return pg;
1418
1419 e_inval:
1420         return ERR_PTR(-EINVAL);
1421 }
1422
1423 static int decode_pg_upmap_items(void **p, void *end, struct ceph_osdmap *map)
1424 {
1425         return decode_pg_mapping(p, end, &map->pg_upmap_items,
1426                                  __decode_pg_upmap_items, false);
1427 }
1428
1429 static int decode_new_pg_upmap_items(void **p, void *end,
1430                                      struct ceph_osdmap *map)
1431 {
1432         return decode_pg_mapping(p, end, &map->pg_upmap_items,
1433                                  __decode_pg_upmap_items, true);
1434 }
1435
1436 static int decode_old_pg_upmap_items(void **p, void *end,
1437                                      struct ceph_osdmap *map)
1438 {
1439         return decode_pg_mapping(p, end, &map->pg_upmap_items, NULL, true);
1440 }
1441
1442 /*
1443  * decode a full map.
1444  */
1445 static int osdmap_decode(void **p, void *end, struct ceph_osdmap *map)
1446 {
1447         u8 struct_v;
1448         u32 epoch = 0;
1449         void *start = *p;
1450         u32 max;
1451         u32 len, i;
1452         int err;
1453
1454         dout("%s %p to %p len %d\n", __func__, *p, end, (int)(end - *p));
1455
1456         err = get_osdmap_client_data_v(p, end, "full", &struct_v);
1457         if (err)
1458                 goto bad;
1459
1460         /* fsid, epoch, created, modified */
1461         ceph_decode_need(p, end, sizeof(map->fsid) + sizeof(u32) +
1462                          sizeof(map->created) + sizeof(map->modified), e_inval);
1463         ceph_decode_copy(p, &map->fsid, sizeof(map->fsid));
1464         epoch = map->epoch = ceph_decode_32(p);
1465         ceph_decode_copy(p, &map->created, sizeof(map->created));
1466         ceph_decode_copy(p, &map->modified, sizeof(map->modified));
1467
1468         /* pools */
1469         err = decode_pools(p, end, map);
1470         if (err)
1471                 goto bad;
1472
1473         /* pool_name */
1474         err = decode_pool_names(p, end, map);
1475         if (err)
1476                 goto bad;
1477
1478         ceph_decode_32_safe(p, end, map->pool_max, e_inval);
1479
1480         ceph_decode_32_safe(p, end, map->flags, e_inval);
1481
1482         /* max_osd */
1483         ceph_decode_32_safe(p, end, max, e_inval);
1484
1485         /* (re)alloc osd arrays */
1486         err = osdmap_set_max_osd(map, max);
1487         if (err)
1488                 goto bad;
1489
1490         /* osd_state, osd_weight, osd_addrs->client_addr */
1491         ceph_decode_need(p, end, 3*sizeof(u32) +
1492                          map->max_osd*((struct_v >= 5 ? sizeof(u32) :
1493                                                         sizeof(u8)) +
1494                                        sizeof(*map->osd_weight) +
1495                                        sizeof(*map->osd_addr)), e_inval);
1496
1497         if (ceph_decode_32(p) != map->max_osd)
1498                 goto e_inval;
1499
1500         if (struct_v >= 5) {
1501                 for (i = 0; i < map->max_osd; i++)
1502                         map->osd_state[i] = ceph_decode_32(p);
1503         } else {
1504                 for (i = 0; i < map->max_osd; i++)
1505                         map->osd_state[i] = ceph_decode_8(p);
1506         }
1507
1508         if (ceph_decode_32(p) != map->max_osd)
1509                 goto e_inval;
1510
1511         for (i = 0; i < map->max_osd; i++)
1512                 map->osd_weight[i] = ceph_decode_32(p);
1513
1514         if (ceph_decode_32(p) != map->max_osd)
1515                 goto e_inval;
1516
1517         ceph_decode_copy(p, map->osd_addr, map->max_osd*sizeof(*map->osd_addr));
1518         for (i = 0; i < map->max_osd; i++)
1519                 ceph_decode_addr(&map->osd_addr[i]);
1520
1521         /* pg_temp */
1522         err = decode_pg_temp(p, end, map);
1523         if (err)
1524                 goto bad;
1525
1526         /* primary_temp */
1527         if (struct_v >= 1) {
1528                 err = decode_primary_temp(p, end, map);
1529                 if (err)
1530                         goto bad;
1531         }
1532
1533         /* primary_affinity */
1534         if (struct_v >= 2) {
1535                 err = decode_primary_affinity(p, end, map);
1536                 if (err)
1537                         goto bad;
1538         } else {
1539                 WARN_ON(map->osd_primary_affinity);
1540         }
1541
1542         /* crush */
1543         ceph_decode_32_safe(p, end, len, e_inval);
1544         err = osdmap_set_crush(map, crush_decode(*p, min(*p + len, end)));
1545         if (err)
1546                 goto bad;
1547
1548         *p += len;
1549         if (struct_v >= 3) {
1550                 /* erasure_code_profiles */
1551                 ceph_decode_skip_map_of_map(p, end, string, string, string,
1552                                             e_inval);
1553         }
1554
1555         if (struct_v >= 4) {
1556                 err = decode_pg_upmap(p, end, map);
1557                 if (err)
1558                         goto bad;
1559
1560                 err = decode_pg_upmap_items(p, end, map);
1561                 if (err)
1562                         goto bad;
1563         } else {
1564                 WARN_ON(!RB_EMPTY_ROOT(&map->pg_upmap));
1565                 WARN_ON(!RB_EMPTY_ROOT(&map->pg_upmap_items));
1566         }
1567
1568         /* ignore the rest */
1569         *p = end;
1570
1571         dout("full osdmap epoch %d max_osd %d\n", map->epoch, map->max_osd);
1572         return 0;
1573
1574 e_inval:
1575         err = -EINVAL;
1576 bad:
1577         pr_err("corrupt full osdmap (%d) epoch %d off %d (%p of %p-%p)\n",
1578                err, epoch, (int)(*p - start), *p, start, end);
1579         print_hex_dump(KERN_DEBUG, "osdmap: ",
1580                        DUMP_PREFIX_OFFSET, 16, 1,
1581                        start, end - start, true);
1582         return err;
1583 }
1584
1585 /*
1586  * Allocate and decode a full map.
1587  */
1588 struct ceph_osdmap *ceph_osdmap_decode(void **p, void *end)
1589 {
1590         struct ceph_osdmap *map;
1591         int ret;
1592
1593         map = ceph_osdmap_alloc();
1594         if (!map)
1595                 return ERR_PTR(-ENOMEM);
1596
1597         ret = osdmap_decode(p, end, map);
1598         if (ret) {
1599                 ceph_osdmap_destroy(map);
1600                 return ERR_PTR(ret);
1601         }
1602
1603         return map;
1604 }
1605
1606 /*
1607  * Encoding order is (new_up_client, new_state, new_weight).  Need to
1608  * apply in the (new_weight, new_state, new_up_client) order, because
1609  * an incremental map may look like e.g.
1610  *
1611  *     new_up_client: { osd=6, addr=... } # set osd_state and addr
1612  *     new_state: { osd=6, xorstate=EXISTS } # clear osd_state
1613  */
1614 static int decode_new_up_state_weight(void **p, void *end, u8 struct_v,
1615                                       struct ceph_osdmap *map)
1616 {
1617         void *new_up_client;
1618         void *new_state;
1619         void *new_weight_end;
1620         u32 len;
1621
1622         new_up_client = *p;
1623         ceph_decode_32_safe(p, end, len, e_inval);
1624         len *= sizeof(u32) + sizeof(struct ceph_entity_addr);
1625         ceph_decode_need(p, end, len, e_inval);
1626         *p += len;
1627
1628         new_state = *p;
1629         ceph_decode_32_safe(p, end, len, e_inval);
1630         len *= sizeof(u32) + (struct_v >= 5 ? sizeof(u32) : sizeof(u8));
1631         ceph_decode_need(p, end, len, e_inval);
1632         *p += len;
1633
1634         /* new_weight */
1635         ceph_decode_32_safe(p, end, len, e_inval);
1636         while (len--) {
1637                 s32 osd;
1638                 u32 w;
1639
1640                 ceph_decode_need(p, end, 2*sizeof(u32), e_inval);
1641                 osd = ceph_decode_32(p);
1642                 w = ceph_decode_32(p);
1643                 BUG_ON(osd >= map->max_osd);
1644                 pr_info("osd%d weight 0x%x %s\n", osd, w,
1645                      w == CEPH_OSD_IN ? "(in)" :
1646                      (w == CEPH_OSD_OUT ? "(out)" : ""));
1647                 map->osd_weight[osd] = w;
1648
1649                 /*
1650                  * If we are marking in, set the EXISTS, and clear the
1651                  * AUTOOUT and NEW bits.
1652                  */
1653                 if (w) {
1654                         map->osd_state[osd] |= CEPH_OSD_EXISTS;
1655                         map->osd_state[osd] &= ~(CEPH_OSD_AUTOOUT |
1656                                                  CEPH_OSD_NEW);
1657                 }
1658         }
1659         new_weight_end = *p;
1660
1661         /* new_state (up/down) */
1662         *p = new_state;
1663         len = ceph_decode_32(p);
1664         while (len--) {
1665                 s32 osd;
1666                 u32 xorstate;
1667                 int ret;
1668
1669                 osd = ceph_decode_32(p);
1670                 if (struct_v >= 5)
1671                         xorstate = ceph_decode_32(p);
1672                 else
1673                         xorstate = ceph_decode_8(p);
1674                 if (xorstate == 0)
1675                         xorstate = CEPH_OSD_UP;
1676                 BUG_ON(osd >= map->max_osd);
1677                 if ((map->osd_state[osd] & CEPH_OSD_UP) &&
1678                     (xorstate & CEPH_OSD_UP))
1679                         pr_info("osd%d down\n", osd);
1680                 if ((map->osd_state[osd] & CEPH_OSD_EXISTS) &&
1681                     (xorstate & CEPH_OSD_EXISTS)) {
1682                         pr_info("osd%d does not exist\n", osd);
1683                         ret = set_primary_affinity(map, osd,
1684                                                    CEPH_OSD_DEFAULT_PRIMARY_AFFINITY);
1685                         if (ret)
1686                                 return ret;
1687                         memset(map->osd_addr + osd, 0, sizeof(*map->osd_addr));
1688                         map->osd_state[osd] = 0;
1689                 } else {
1690                         map->osd_state[osd] ^= xorstate;
1691                 }
1692         }
1693
1694         /* new_up_client */
1695         *p = new_up_client;
1696         len = ceph_decode_32(p);
1697         while (len--) {
1698                 s32 osd;
1699                 struct ceph_entity_addr addr;
1700
1701                 osd = ceph_decode_32(p);
1702                 ceph_decode_copy(p, &addr, sizeof(addr));
1703                 ceph_decode_addr(&addr);
1704                 BUG_ON(osd >= map->max_osd);
1705                 pr_info("osd%d up\n", osd);
1706                 map->osd_state[osd] |= CEPH_OSD_EXISTS | CEPH_OSD_UP;
1707                 map->osd_addr[osd] = addr;
1708         }
1709
1710         *p = new_weight_end;
1711         return 0;
1712
1713 e_inval:
1714         return -EINVAL;
1715 }
1716
1717 /*
1718  * decode and apply an incremental map update.
1719  */
1720 struct ceph_osdmap *osdmap_apply_incremental(void **p, void *end,
1721                                              struct ceph_osdmap *map)
1722 {
1723         struct ceph_fsid fsid;
1724         u32 epoch = 0;
1725         struct ceph_timespec modified;
1726         s32 len;
1727         u64 pool;
1728         __s64 new_pool_max;
1729         __s32 new_flags, max;
1730         void *start = *p;
1731         int err;
1732         u8 struct_v;
1733
1734         dout("%s %p to %p len %d\n", __func__, *p, end, (int)(end - *p));
1735
1736         err = get_osdmap_client_data_v(p, end, "inc", &struct_v);
1737         if (err)
1738                 goto bad;
1739
1740         /* fsid, epoch, modified, new_pool_max, new_flags */
1741         ceph_decode_need(p, end, sizeof(fsid) + sizeof(u32) + sizeof(modified) +
1742                          sizeof(u64) + sizeof(u32), e_inval);
1743         ceph_decode_copy(p, &fsid, sizeof(fsid));
1744         epoch = ceph_decode_32(p);
1745         BUG_ON(epoch != map->epoch+1);
1746         ceph_decode_copy(p, &modified, sizeof(modified));
1747         new_pool_max = ceph_decode_64(p);
1748         new_flags = ceph_decode_32(p);
1749
1750         /* full map? */
1751         ceph_decode_32_safe(p, end, len, e_inval);
1752         if (len > 0) {
1753                 dout("apply_incremental full map len %d, %p to %p\n",
1754                      len, *p, end);
1755                 return ceph_osdmap_decode(p, min(*p+len, end));
1756         }
1757
1758         /* new crush? */
1759         ceph_decode_32_safe(p, end, len, e_inval);
1760         if (len > 0) {
1761                 err = osdmap_set_crush(map,
1762                                        crush_decode(*p, min(*p + len, end)));
1763                 if (err)
1764                         goto bad;
1765                 *p += len;
1766         }
1767
1768         /* new flags? */
1769         if (new_flags >= 0)
1770                 map->flags = new_flags;
1771         if (new_pool_max >= 0)
1772                 map->pool_max = new_pool_max;
1773
1774         /* new max? */
1775         ceph_decode_32_safe(p, end, max, e_inval);
1776         if (max >= 0) {
1777                 err = osdmap_set_max_osd(map, max);
1778                 if (err)
1779                         goto bad;
1780         }
1781
1782         map->epoch++;
1783         map->modified = modified;
1784
1785         /* new_pools */
1786         err = decode_new_pools(p, end, map);
1787         if (err)
1788                 goto bad;
1789
1790         /* new_pool_names */
1791         err = decode_pool_names(p, end, map);
1792         if (err)
1793                 goto bad;
1794
1795         /* old_pool */
1796         ceph_decode_32_safe(p, end, len, e_inval);
1797         while (len--) {
1798                 struct ceph_pg_pool_info *pi;
1799
1800                 ceph_decode_64_safe(p, end, pool, e_inval);
1801                 pi = __lookup_pg_pool(&map->pg_pools, pool);
1802                 if (pi)
1803                         __remove_pg_pool(&map->pg_pools, pi);
1804         }
1805
1806         /* new_up_client, new_state, new_weight */
1807         err = decode_new_up_state_weight(p, end, struct_v, map);
1808         if (err)
1809                 goto bad;
1810
1811         /* new_pg_temp */
1812         err = decode_new_pg_temp(p, end, map);
1813         if (err)
1814                 goto bad;
1815
1816         /* new_primary_temp */
1817         if (struct_v >= 1) {
1818                 err = decode_new_primary_temp(p, end, map);
1819                 if (err)
1820                         goto bad;
1821         }
1822
1823         /* new_primary_affinity */
1824         if (struct_v >= 2) {
1825                 err = decode_new_primary_affinity(p, end, map);
1826                 if (err)
1827                         goto bad;
1828         }
1829
1830         if (struct_v >= 3) {
1831                 /* new_erasure_code_profiles */
1832                 ceph_decode_skip_map_of_map(p, end, string, string, string,
1833                                             e_inval);
1834                 /* old_erasure_code_profiles */
1835                 ceph_decode_skip_set(p, end, string, e_inval);
1836         }
1837
1838         if (struct_v >= 4) {
1839                 err = decode_new_pg_upmap(p, end, map);
1840                 if (err)
1841                         goto bad;
1842
1843                 err = decode_old_pg_upmap(p, end, map);
1844                 if (err)
1845                         goto bad;
1846
1847                 err = decode_new_pg_upmap_items(p, end, map);
1848                 if (err)
1849                         goto bad;
1850
1851                 err = decode_old_pg_upmap_items(p, end, map);
1852                 if (err)
1853                         goto bad;
1854         }
1855
1856         /* ignore the rest */
1857         *p = end;
1858
1859         dout("inc osdmap epoch %d max_osd %d\n", map->epoch, map->max_osd);
1860         return map;
1861
1862 e_inval:
1863         err = -EINVAL;
1864 bad:
1865         pr_err("corrupt inc osdmap (%d) epoch %d off %d (%p of %p-%p)\n",
1866                err, epoch, (int)(*p - start), *p, start, end);
1867         print_hex_dump(KERN_DEBUG, "osdmap: ",
1868                        DUMP_PREFIX_OFFSET, 16, 1,
1869                        start, end - start, true);
1870         return ERR_PTR(err);
1871 }
1872
1873 void ceph_oloc_copy(struct ceph_object_locator *dest,
1874                     const struct ceph_object_locator *src)
1875 {
1876         ceph_oloc_destroy(dest);
1877
1878         dest->pool = src->pool;
1879         if (src->pool_ns)
1880                 dest->pool_ns = ceph_get_string(src->pool_ns);
1881         else
1882                 dest->pool_ns = NULL;
1883 }
1884 EXPORT_SYMBOL(ceph_oloc_copy);
1885
1886 void ceph_oloc_destroy(struct ceph_object_locator *oloc)
1887 {
1888         ceph_put_string(oloc->pool_ns);
1889 }
1890 EXPORT_SYMBOL(ceph_oloc_destroy);
1891
1892 void ceph_oid_copy(struct ceph_object_id *dest,
1893                    const struct ceph_object_id *src)
1894 {
1895         ceph_oid_destroy(dest);
1896
1897         if (src->name != src->inline_name) {
1898                 /* very rare, see ceph_object_id definition */
1899                 dest->name = kmalloc(src->name_len + 1,
1900                                      GFP_NOIO | __GFP_NOFAIL);
1901         } else {
1902                 dest->name = dest->inline_name;
1903         }
1904         memcpy(dest->name, src->name, src->name_len + 1);
1905         dest->name_len = src->name_len;
1906 }
1907 EXPORT_SYMBOL(ceph_oid_copy);
1908
1909 static __printf(2, 0)
1910 int oid_printf_vargs(struct ceph_object_id *oid, const char *fmt, va_list ap)
1911 {
1912         int len;
1913
1914         WARN_ON(!ceph_oid_empty(oid));
1915
1916         len = vsnprintf(oid->inline_name, sizeof(oid->inline_name), fmt, ap);
1917         if (len >= sizeof(oid->inline_name))
1918                 return len;
1919
1920         oid->name_len = len;
1921         return 0;
1922 }
1923
1924 /*
1925  * If oid doesn't fit into inline buffer, BUG.
1926  */
1927 void ceph_oid_printf(struct ceph_object_id *oid, const char *fmt, ...)
1928 {
1929         va_list ap;
1930
1931         va_start(ap, fmt);
1932         BUG_ON(oid_printf_vargs(oid, fmt, ap));
1933         va_end(ap);
1934 }
1935 EXPORT_SYMBOL(ceph_oid_printf);
1936
1937 static __printf(3, 0)
1938 int oid_aprintf_vargs(struct ceph_object_id *oid, gfp_t gfp,
1939                       const char *fmt, va_list ap)
1940 {
1941         va_list aq;
1942         int len;
1943
1944         va_copy(aq, ap);
1945         len = oid_printf_vargs(oid, fmt, aq);
1946         va_end(aq);
1947
1948         if (len) {
1949                 char *external_name;
1950
1951                 external_name = kmalloc(len + 1, gfp);
1952                 if (!external_name)
1953                         return -ENOMEM;
1954
1955                 oid->name = external_name;
1956                 WARN_ON(vsnprintf(oid->name, len + 1, fmt, ap) != len);
1957                 oid->name_len = len;
1958         }
1959
1960         return 0;
1961 }
1962
1963 /*
1964  * If oid doesn't fit into inline buffer, allocate.
1965  */
1966 int ceph_oid_aprintf(struct ceph_object_id *oid, gfp_t gfp,
1967                      const char *fmt, ...)
1968 {
1969         va_list ap;
1970         int ret;
1971
1972         va_start(ap, fmt);
1973         ret = oid_aprintf_vargs(oid, gfp, fmt, ap);
1974         va_end(ap);
1975
1976         return ret;
1977 }
1978 EXPORT_SYMBOL(ceph_oid_aprintf);
1979
1980 void ceph_oid_destroy(struct ceph_object_id *oid)
1981 {
1982         if (oid->name != oid->inline_name)
1983                 kfree(oid->name);
1984 }
1985 EXPORT_SYMBOL(ceph_oid_destroy);
1986
1987 /*
1988  * osds only
1989  */
1990 static bool __osds_equal(const struct ceph_osds *lhs,
1991                          const struct ceph_osds *rhs)
1992 {
1993         if (lhs->size == rhs->size &&
1994             !memcmp(lhs->osds, rhs->osds, rhs->size * sizeof(rhs->osds[0])))
1995                 return true;
1996
1997         return false;
1998 }
1999
2000 /*
2001  * osds + primary
2002  */
2003 static bool osds_equal(const struct ceph_osds *lhs,
2004                        const struct ceph_osds *rhs)
2005 {
2006         if (__osds_equal(lhs, rhs) &&
2007             lhs->primary == rhs->primary)
2008                 return true;
2009
2010         return false;
2011 }
2012
2013 static bool osds_valid(const struct ceph_osds *set)
2014 {
2015         /* non-empty set */
2016         if (set->size > 0 && set->primary >= 0)
2017                 return true;
2018
2019         /* empty can_shift_osds set */
2020         if (!set->size && set->primary == -1)
2021                 return true;
2022
2023         /* empty !can_shift_osds set - all NONE */
2024         if (set->size > 0 && set->primary == -1) {
2025                 int i;
2026
2027                 for (i = 0; i < set->size; i++) {
2028                         if (set->osds[i] != CRUSH_ITEM_NONE)
2029                                 break;
2030                 }
2031                 if (i == set->size)
2032                         return true;
2033         }
2034
2035         return false;
2036 }
2037
2038 void ceph_osds_copy(struct ceph_osds *dest, const struct ceph_osds *src)
2039 {
2040         memcpy(dest->osds, src->osds, src->size * sizeof(src->osds[0]));
2041         dest->size = src->size;
2042         dest->primary = src->primary;
2043 }
2044
2045 bool ceph_pg_is_split(const struct ceph_pg *pgid, u32 old_pg_num,
2046                       u32 new_pg_num)
2047 {
2048         int old_bits = calc_bits_of(old_pg_num);
2049         int old_mask = (1 << old_bits) - 1;
2050         int n;
2051
2052         WARN_ON(pgid->seed >= old_pg_num);
2053         if (new_pg_num <= old_pg_num)
2054                 return false;
2055
2056         for (n = 1; ; n++) {
2057                 int next_bit = n << (old_bits - 1);
2058                 u32 s = next_bit | pgid->seed;
2059
2060                 if (s < old_pg_num || s == pgid->seed)
2061                         continue;
2062                 if (s >= new_pg_num)
2063                         break;
2064
2065                 s = ceph_stable_mod(s, old_pg_num, old_mask);
2066                 if (s == pgid->seed)
2067                         return true;
2068         }
2069
2070         return false;
2071 }
2072
2073 bool ceph_is_new_interval(const struct ceph_osds *old_acting,
2074                           const struct ceph_osds *new_acting,
2075                           const struct ceph_osds *old_up,
2076                           const struct ceph_osds *new_up,
2077                           int old_size,
2078                           int new_size,
2079                           int old_min_size,
2080                           int new_min_size,
2081                           u32 old_pg_num,
2082                           u32 new_pg_num,
2083                           bool old_sort_bitwise,
2084                           bool new_sort_bitwise,
2085                           bool old_recovery_deletes,
2086                           bool new_recovery_deletes,
2087                           const struct ceph_pg *pgid)
2088 {
2089         return !osds_equal(old_acting, new_acting) ||
2090                !osds_equal(old_up, new_up) ||
2091                old_size != new_size ||
2092                old_min_size != new_min_size ||
2093                ceph_pg_is_split(pgid, old_pg_num, new_pg_num) ||
2094                old_sort_bitwise != new_sort_bitwise ||
2095                old_recovery_deletes != new_recovery_deletes;
2096 }
2097
2098 static int calc_pg_rank(int osd, const struct ceph_osds *acting)
2099 {
2100         int i;
2101
2102         for (i = 0; i < acting->size; i++) {
2103                 if (acting->osds[i] == osd)
2104                         return i;
2105         }
2106
2107         return -1;
2108 }
2109
2110 static bool primary_changed(const struct ceph_osds *old_acting,
2111                             const struct ceph_osds *new_acting)
2112 {
2113         if (!old_acting->size && !new_acting->size)
2114                 return false; /* both still empty */
2115
2116         if (!old_acting->size ^ !new_acting->size)
2117                 return true; /* was empty, now not, or vice versa */
2118
2119         if (old_acting->primary != new_acting->primary)
2120                 return true; /* primary changed */
2121
2122         if (calc_pg_rank(old_acting->primary, old_acting) !=
2123             calc_pg_rank(new_acting->primary, new_acting))
2124                 return true;
2125
2126         return false; /* same primary (tho replicas may have changed) */
2127 }
2128
2129 bool ceph_osds_changed(const struct ceph_osds *old_acting,
2130                        const struct ceph_osds *new_acting,
2131                        bool any_change)
2132 {
2133         if (primary_changed(old_acting, new_acting))
2134                 return true;
2135
2136         if (any_change && !__osds_equal(old_acting, new_acting))
2137                 return true;
2138
2139         return false;
2140 }
2141
2142 /*
2143  * calculate file layout from given offset, length.
2144  * fill in correct oid, logical length, and object extent
2145  * offset, length.
2146  *
2147  * for now, we write only a single su, until we can
2148  * pass a stride back to the caller.
2149  */
2150 int ceph_calc_file_object_mapping(struct ceph_file_layout *layout,
2151                                    u64 off, u64 len,
2152                                    u64 *ono,
2153                                    u64 *oxoff, u64 *oxlen)
2154 {
2155         u32 osize = layout->object_size;
2156         u32 su = layout->stripe_unit;
2157         u32 sc = layout->stripe_count;
2158         u32 bl, stripeno, stripepos, objsetno;
2159         u32 su_per_object;
2160         u64 t, su_offset;
2161
2162         dout("mapping %llu~%llu  osize %u fl_su %u\n", off, len,
2163              osize, su);
2164         if (su == 0 || sc == 0)
2165                 goto invalid;
2166         su_per_object = osize / su;
2167         if (su_per_object == 0)
2168                 goto invalid;
2169         dout("osize %u / su %u = su_per_object %u\n", osize, su,
2170              su_per_object);
2171
2172         if ((su & ~PAGE_MASK) != 0)
2173                 goto invalid;
2174
2175         /* bl = *off / su; */
2176         t = off;
2177         do_div(t, su);
2178         bl = t;
2179         dout("off %llu / su %u = bl %u\n", off, su, bl);
2180
2181         stripeno = bl / sc;
2182         stripepos = bl % sc;
2183         objsetno = stripeno / su_per_object;
2184
2185         *ono = objsetno * sc + stripepos;
2186         dout("objset %u * sc %u = ono %u\n", objsetno, sc, (unsigned int)*ono);
2187
2188         /* *oxoff = *off % layout->fl_stripe_unit;  # offset in su */
2189         t = off;
2190         su_offset = do_div(t, su);
2191         *oxoff = su_offset + (stripeno % su_per_object) * su;
2192
2193         /*
2194          * Calculate the length of the extent being written to the selected
2195          * object. This is the minimum of the full length requested (len) or
2196          * the remainder of the current stripe being written to.
2197          */
2198         *oxlen = min_t(u64, len, su - su_offset);
2199
2200         dout(" obj extent %llu~%llu\n", *oxoff, *oxlen);
2201         return 0;
2202
2203 invalid:
2204         dout(" invalid layout\n");
2205         *ono = 0;
2206         *oxoff = 0;
2207         *oxlen = 0;
2208         return -EINVAL;
2209 }
2210 EXPORT_SYMBOL(ceph_calc_file_object_mapping);
2211
2212 /*
2213  * Map an object into a PG.
2214  *
2215  * Should only be called with target_oid and target_oloc (as opposed to
2216  * base_oid and base_oloc), since tiering isn't taken into account.
2217  */
2218 int __ceph_object_locator_to_pg(struct ceph_pg_pool_info *pi,
2219                                 const struct ceph_object_id *oid,
2220                                 const struct ceph_object_locator *oloc,
2221                                 struct ceph_pg *raw_pgid)
2222 {
2223         WARN_ON(pi->id != oloc->pool);
2224
2225         if (!oloc->pool_ns) {
2226                 raw_pgid->pool = oloc->pool;
2227                 raw_pgid->seed = ceph_str_hash(pi->object_hash, oid->name,
2228                                              oid->name_len);
2229                 dout("%s %s -> raw_pgid %llu.%x\n", __func__, oid->name,
2230                      raw_pgid->pool, raw_pgid->seed);
2231         } else {
2232                 char stack_buf[256];
2233                 char *buf = stack_buf;
2234                 int nsl = oloc->pool_ns->len;
2235                 size_t total = nsl + 1 + oid->name_len;
2236
2237                 if (total > sizeof(stack_buf)) {
2238                         buf = kmalloc(total, GFP_NOIO);
2239                         if (!buf)
2240                                 return -ENOMEM;
2241                 }
2242                 memcpy(buf, oloc->pool_ns->str, nsl);
2243                 buf[nsl] = '\037';
2244                 memcpy(buf + nsl + 1, oid->name, oid->name_len);
2245                 raw_pgid->pool = oloc->pool;
2246                 raw_pgid->seed = ceph_str_hash(pi->object_hash, buf, total);
2247                 if (buf != stack_buf)
2248                         kfree(buf);
2249                 dout("%s %s ns %.*s -> raw_pgid %llu.%x\n", __func__,
2250                      oid->name, nsl, oloc->pool_ns->str,
2251                      raw_pgid->pool, raw_pgid->seed);
2252         }
2253         return 0;
2254 }
2255
2256 int ceph_object_locator_to_pg(struct ceph_osdmap *osdmap,
2257                               const struct ceph_object_id *oid,
2258                               const struct ceph_object_locator *oloc,
2259                               struct ceph_pg *raw_pgid)
2260 {
2261         struct ceph_pg_pool_info *pi;
2262
2263         pi = ceph_pg_pool_by_id(osdmap, oloc->pool);
2264         if (!pi)
2265                 return -ENOENT;
2266
2267         return __ceph_object_locator_to_pg(pi, oid, oloc, raw_pgid);
2268 }
2269 EXPORT_SYMBOL(ceph_object_locator_to_pg);
2270
2271 /*
2272  * Map a raw PG (full precision ps) into an actual PG.
2273  */
2274 static void raw_pg_to_pg(struct ceph_pg_pool_info *pi,
2275                          const struct ceph_pg *raw_pgid,
2276                          struct ceph_pg *pgid)
2277 {
2278         pgid->pool = raw_pgid->pool;
2279         pgid->seed = ceph_stable_mod(raw_pgid->seed, pi->pg_num,
2280                                      pi->pg_num_mask);
2281 }
2282
2283 /*
2284  * Map a raw PG (full precision ps) into a placement ps (placement
2285  * seed).  Include pool id in that value so that different pools don't
2286  * use the same seeds.
2287  */
2288 static u32 raw_pg_to_pps(struct ceph_pg_pool_info *pi,
2289                          const struct ceph_pg *raw_pgid)
2290 {
2291         if (pi->flags & CEPH_POOL_FLAG_HASHPSPOOL) {
2292                 /* hash pool id and seed so that pool PGs do not overlap */
2293                 return crush_hash32_2(CRUSH_HASH_RJENKINS1,
2294                                       ceph_stable_mod(raw_pgid->seed,
2295                                                       pi->pgp_num,
2296                                                       pi->pgp_num_mask),
2297                                       raw_pgid->pool);
2298         } else {
2299                 /*
2300                  * legacy behavior: add ps and pool together.  this is
2301                  * not a great approach because the PGs from each pool
2302                  * will overlap on top of each other: 0.5 == 1.4 ==
2303                  * 2.3 == ...
2304                  */
2305                 return ceph_stable_mod(raw_pgid->seed, pi->pgp_num,
2306                                        pi->pgp_num_mask) +
2307                        (unsigned)raw_pgid->pool;
2308         }
2309 }
2310
2311 /*
2312  * Magic value used for a "default" fallback choose_args, used if the
2313  * crush_choose_arg_map passed to do_crush() does not exist.  If this
2314  * also doesn't exist, fall back to canonical weights.
2315  */
2316 #define CEPH_DEFAULT_CHOOSE_ARGS        -1
2317
2318 static int do_crush(struct ceph_osdmap *map, int ruleno, int x,
2319                     int *result, int result_max,
2320                     const __u32 *weight, int weight_max,
2321                     s64 choose_args_index)
2322 {
2323         struct crush_choose_arg_map *arg_map;
2324         int r;
2325
2326         BUG_ON(result_max > CEPH_PG_MAX_SIZE);
2327
2328         arg_map = lookup_choose_arg_map(&map->crush->choose_args,
2329                                         choose_args_index);
2330         if (!arg_map)
2331                 arg_map = lookup_choose_arg_map(&map->crush->choose_args,
2332                                                 CEPH_DEFAULT_CHOOSE_ARGS);
2333
2334         mutex_lock(&map->crush_workspace_mutex);
2335         r = crush_do_rule(map->crush, ruleno, x, result, result_max,
2336                           weight, weight_max, map->crush_workspace,
2337                           arg_map ? arg_map->args : NULL);
2338         mutex_unlock(&map->crush_workspace_mutex);
2339
2340         return r;
2341 }
2342
2343 static void remove_nonexistent_osds(struct ceph_osdmap *osdmap,
2344                                     struct ceph_pg_pool_info *pi,
2345                                     struct ceph_osds *set)
2346 {
2347         int i;
2348
2349         if (ceph_can_shift_osds(pi)) {
2350                 int removed = 0;
2351
2352                 /* shift left */
2353                 for (i = 0; i < set->size; i++) {
2354                         if (!ceph_osd_exists(osdmap, set->osds[i])) {
2355                                 removed++;
2356                                 continue;
2357                         }
2358                         if (removed)
2359                                 set->osds[i - removed] = set->osds[i];
2360                 }
2361                 set->size -= removed;
2362         } else {
2363                 /* set dne devices to NONE */
2364                 for (i = 0; i < set->size; i++) {
2365                         if (!ceph_osd_exists(osdmap, set->osds[i]))
2366                                 set->osds[i] = CRUSH_ITEM_NONE;
2367                 }
2368         }
2369 }
2370
2371 /*
2372  * Calculate raw set (CRUSH output) for given PG and filter out
2373  * nonexistent OSDs.  ->primary is undefined for a raw set.
2374  *
2375  * Placement seed (CRUSH input) is returned through @ppps.
2376  */
2377 static void pg_to_raw_osds(struct ceph_osdmap *osdmap,
2378                            struct ceph_pg_pool_info *pi,
2379                            const struct ceph_pg *raw_pgid,
2380                            struct ceph_osds *raw,
2381                            u32 *ppps)
2382 {
2383         u32 pps = raw_pg_to_pps(pi, raw_pgid);
2384         int ruleno;
2385         int len;
2386
2387         ceph_osds_init(raw);
2388         if (ppps)
2389                 *ppps = pps;
2390
2391         ruleno = crush_find_rule(osdmap->crush, pi->crush_ruleset, pi->type,
2392                                  pi->size);
2393         if (ruleno < 0) {
2394                 pr_err("no crush rule: pool %lld ruleset %d type %d size %d\n",
2395                        pi->id, pi->crush_ruleset, pi->type, pi->size);
2396                 return;
2397         }
2398
2399         if (pi->size > ARRAY_SIZE(raw->osds)) {
2400                 pr_err_ratelimited("pool %lld ruleset %d type %d too wide: size %d > %zu\n",
2401                        pi->id, pi->crush_ruleset, pi->type, pi->size,
2402                        ARRAY_SIZE(raw->osds));
2403                 return;
2404         }
2405
2406         len = do_crush(osdmap, ruleno, pps, raw->osds, pi->size,
2407                        osdmap->osd_weight, osdmap->max_osd, pi->id);
2408         if (len < 0) {
2409                 pr_err("error %d from crush rule %d: pool %lld ruleset %d type %d size %d\n",
2410                        len, ruleno, pi->id, pi->crush_ruleset, pi->type,
2411                        pi->size);
2412                 return;
2413         }
2414
2415         raw->size = len;
2416         remove_nonexistent_osds(osdmap, pi, raw);
2417 }
2418
2419 /* apply pg_upmap[_items] mappings */
2420 static void apply_upmap(struct ceph_osdmap *osdmap,
2421                         const struct ceph_pg *pgid,
2422                         struct ceph_osds *raw)
2423 {
2424         struct ceph_pg_mapping *pg;
2425         int i, j;
2426
2427         pg = lookup_pg_mapping(&osdmap->pg_upmap, pgid);
2428         if (pg) {
2429                 /* make sure targets aren't marked out */
2430                 for (i = 0; i < pg->pg_upmap.len; i++) {
2431                         int osd = pg->pg_upmap.osds[i];
2432
2433                         if (osd != CRUSH_ITEM_NONE &&
2434                             osd < osdmap->max_osd &&
2435                             osdmap->osd_weight[osd] == 0) {
2436                                 /* reject/ignore explicit mapping */
2437                                 return;
2438                         }
2439                 }
2440                 for (i = 0; i < pg->pg_upmap.len; i++)
2441                         raw->osds[i] = pg->pg_upmap.osds[i];
2442                 raw->size = pg->pg_upmap.len;
2443                 /* check and apply pg_upmap_items, if any */
2444         }
2445
2446         pg = lookup_pg_mapping(&osdmap->pg_upmap_items, pgid);
2447         if (pg) {
2448                 for (i = 0; i < raw->size; i++) {
2449                         for (j = 0; j < pg->pg_upmap_items.len; j++) {
2450                                 int from = pg->pg_upmap_items.from_to[j][0];
2451                                 int to = pg->pg_upmap_items.from_to[j][1];
2452
2453                                 if (from == raw->osds[i]) {
2454                                         if (!(to != CRUSH_ITEM_NONE &&
2455                                               to < osdmap->max_osd &&
2456                                               osdmap->osd_weight[to] == 0))
2457                                                 raw->osds[i] = to;
2458                                         break;
2459                                 }
2460                         }
2461                 }
2462         }
2463 }
2464
2465 /*
2466  * Given raw set, calculate up set and up primary.  By definition of an
2467  * up set, the result won't contain nonexistent or down OSDs.
2468  *
2469  * This is done in-place - on return @set is the up set.  If it's
2470  * empty, ->primary will remain undefined.
2471  */
2472 static void raw_to_up_osds(struct ceph_osdmap *osdmap,
2473                            struct ceph_pg_pool_info *pi,
2474                            struct ceph_osds *set)
2475 {
2476         int i;
2477
2478         /* ->primary is undefined for a raw set */
2479         BUG_ON(set->primary != -1);
2480
2481         if (ceph_can_shift_osds(pi)) {
2482                 int removed = 0;
2483
2484                 /* shift left */
2485                 for (i = 0; i < set->size; i++) {
2486                         if (ceph_osd_is_down(osdmap, set->osds[i])) {
2487                                 removed++;
2488                                 continue;
2489                         }
2490                         if (removed)
2491                                 set->osds[i - removed] = set->osds[i];
2492                 }
2493                 set->size -= removed;
2494                 if (set->size > 0)
2495                         set->primary = set->osds[0];
2496         } else {
2497                 /* set down/dne devices to NONE */
2498                 for (i = set->size - 1; i >= 0; i--) {
2499                         if (ceph_osd_is_down(osdmap, set->osds[i]))
2500                                 set->osds[i] = CRUSH_ITEM_NONE;
2501                         else
2502                                 set->primary = set->osds[i];
2503                 }
2504         }
2505 }
2506
2507 static void apply_primary_affinity(struct ceph_osdmap *osdmap,
2508                                    struct ceph_pg_pool_info *pi,
2509                                    u32 pps,
2510                                    struct ceph_osds *up)
2511 {
2512         int i;
2513         int pos = -1;
2514
2515         /*
2516          * Do we have any non-default primary_affinity values for these
2517          * osds?
2518          */
2519         if (!osdmap->osd_primary_affinity)
2520                 return;
2521
2522         for (i = 0; i < up->size; i++) {
2523                 int osd = up->osds[i];
2524
2525                 if (osd != CRUSH_ITEM_NONE &&
2526                     osdmap->osd_primary_affinity[osd] !=
2527                                         CEPH_OSD_DEFAULT_PRIMARY_AFFINITY) {
2528                         break;
2529                 }
2530         }
2531         if (i == up->size)
2532                 return;
2533
2534         /*
2535          * Pick the primary.  Feed both the seed (for the pg) and the
2536          * osd into the hash/rng so that a proportional fraction of an
2537          * osd's pgs get rejected as primary.
2538          */
2539         for (i = 0; i < up->size; i++) {
2540                 int osd = up->osds[i];
2541                 u32 aff;
2542
2543                 if (osd == CRUSH_ITEM_NONE)
2544                         continue;
2545
2546                 aff = osdmap->osd_primary_affinity[osd];
2547                 if (aff < CEPH_OSD_MAX_PRIMARY_AFFINITY &&
2548                     (crush_hash32_2(CRUSH_HASH_RJENKINS1,
2549                                     pps, osd) >> 16) >= aff) {
2550                         /*
2551                          * We chose not to use this primary.  Note it
2552                          * anyway as a fallback in case we don't pick
2553                          * anyone else, but keep looking.
2554                          */
2555                         if (pos < 0)
2556                                 pos = i;
2557                 } else {
2558                         pos = i;
2559                         break;
2560                 }
2561         }
2562         if (pos < 0)
2563                 return;
2564
2565         up->primary = up->osds[pos];
2566
2567         if (ceph_can_shift_osds(pi) && pos > 0) {
2568                 /* move the new primary to the front */
2569                 for (i = pos; i > 0; i--)
2570                         up->osds[i] = up->osds[i - 1];
2571                 up->osds[0] = up->primary;
2572         }
2573 }
2574
2575 /*
2576  * Get pg_temp and primary_temp mappings for given PG.
2577  *
2578  * Note that a PG may have none, only pg_temp, only primary_temp or
2579  * both pg_temp and primary_temp mappings.  This means @temp isn't
2580  * always a valid OSD set on return: in the "only primary_temp" case,
2581  * @temp will have its ->primary >= 0 but ->size == 0.
2582  */
2583 static void get_temp_osds(struct ceph_osdmap *osdmap,
2584                           struct ceph_pg_pool_info *pi,
2585                           const struct ceph_pg *pgid,
2586                           struct ceph_osds *temp)
2587 {
2588         struct ceph_pg_mapping *pg;
2589         int i;
2590
2591         ceph_osds_init(temp);
2592
2593         /* pg_temp? */
2594         pg = lookup_pg_mapping(&osdmap->pg_temp, pgid);
2595         if (pg) {
2596                 for (i = 0; i < pg->pg_temp.len; i++) {
2597                         if (ceph_osd_is_down(osdmap, pg->pg_temp.osds[i])) {
2598                                 if (ceph_can_shift_osds(pi))
2599                                         continue;
2600
2601                                 temp->osds[temp->size++] = CRUSH_ITEM_NONE;
2602                         } else {
2603                                 temp->osds[temp->size++] = pg->pg_temp.osds[i];
2604                         }
2605                 }
2606
2607                 /* apply pg_temp's primary */
2608                 for (i = 0; i < temp->size; i++) {
2609                         if (temp->osds[i] != CRUSH_ITEM_NONE) {
2610                                 temp->primary = temp->osds[i];
2611                                 break;
2612                         }
2613                 }
2614         }
2615
2616         /* primary_temp? */
2617         pg = lookup_pg_mapping(&osdmap->primary_temp, pgid);
2618         if (pg)
2619                 temp->primary = pg->primary_temp.osd;
2620 }
2621
2622 /*
2623  * Map a PG to its acting set as well as its up set.
2624  *
2625  * Acting set is used for data mapping purposes, while up set can be
2626  * recorded for detecting interval changes and deciding whether to
2627  * resend a request.
2628  */
2629 void ceph_pg_to_up_acting_osds(struct ceph_osdmap *osdmap,
2630                                struct ceph_pg_pool_info *pi,
2631                                const struct ceph_pg *raw_pgid,
2632                                struct ceph_osds *up,
2633                                struct ceph_osds *acting)
2634 {
2635         struct ceph_pg pgid;
2636         u32 pps;
2637
2638         WARN_ON(pi->id != raw_pgid->pool);
2639         raw_pg_to_pg(pi, raw_pgid, &pgid);
2640
2641         pg_to_raw_osds(osdmap, pi, raw_pgid, up, &pps);
2642         apply_upmap(osdmap, &pgid, up);
2643         raw_to_up_osds(osdmap, pi, up);
2644         apply_primary_affinity(osdmap, pi, pps, up);
2645         get_temp_osds(osdmap, pi, &pgid, acting);
2646         if (!acting->size) {
2647                 memcpy(acting->osds, up->osds, up->size * sizeof(up->osds[0]));
2648                 acting->size = up->size;
2649                 if (acting->primary == -1)
2650                         acting->primary = up->primary;
2651         }
2652         WARN_ON(!osds_valid(up) || !osds_valid(acting));
2653 }
2654
2655 bool ceph_pg_to_primary_shard(struct ceph_osdmap *osdmap,
2656                               struct ceph_pg_pool_info *pi,
2657                               const struct ceph_pg *raw_pgid,
2658                               struct ceph_spg *spgid)
2659 {
2660         struct ceph_pg pgid;
2661         struct ceph_osds up, acting;
2662         int i;
2663
2664         WARN_ON(pi->id != raw_pgid->pool);
2665         raw_pg_to_pg(pi, raw_pgid, &pgid);
2666
2667         if (ceph_can_shift_osds(pi)) {
2668                 spgid->pgid = pgid; /* struct */
2669                 spgid->shard = CEPH_SPG_NOSHARD;
2670                 return true;
2671         }
2672
2673         ceph_pg_to_up_acting_osds(osdmap, pi, &pgid, &up, &acting);
2674         for (i = 0; i < acting.size; i++) {
2675                 if (acting.osds[i] == acting.primary) {
2676                         spgid->pgid = pgid; /* struct */
2677                         spgid->shard = i;
2678                         return true;
2679                 }
2680         }
2681
2682         return false;
2683 }
2684
2685 /*
2686  * Return acting primary for given PG, or -1 if none.
2687  */
2688 int ceph_pg_to_acting_primary(struct ceph_osdmap *osdmap,
2689                               const struct ceph_pg *raw_pgid)
2690 {
2691         struct ceph_pg_pool_info *pi;
2692         struct ceph_osds up, acting;
2693
2694         pi = ceph_pg_pool_by_id(osdmap, raw_pgid->pool);
2695         if (!pi)
2696                 return -1;
2697
2698         ceph_pg_to_up_acting_osds(osdmap, pi, raw_pgid, &up, &acting);
2699         return acting.primary;
2700 }
2701 EXPORT_SYMBOL(ceph_pg_to_acting_primary);