libceph: use cursor for bio reads
[linux-3.10.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, int state)
15 {
16         if (!len)
17                 return str;
18
19         if ((state & CEPH_OSD_EXISTS) && (state & CEPH_OSD_UP))
20                 snprintf(str, len, "exists, up");
21         else if (state & CEPH_OSD_EXISTS)
22                 snprintf(str, len, "exists");
23         else if (state & CEPH_OSD_UP)
24                 snprintf(str, len, "up");
25         else
26                 snprintf(str, len, "doesn't exist");
27
28         return str;
29 }
30
31 /* maps */
32
33 static int calc_bits_of(unsigned int t)
34 {
35         int b = 0;
36         while (t) {
37                 t = t >> 1;
38                 b++;
39         }
40         return b;
41 }
42
43 /*
44  * the foo_mask is the smallest value 2^n-1 that is >= foo.
45  */
46 static void calc_pg_masks(struct ceph_pg_pool_info *pi)
47 {
48         pi->pg_num_mask = (1 << calc_bits_of(pi->pg_num-1)) - 1;
49         pi->pgp_num_mask = (1 << calc_bits_of(pi->pgp_num-1)) - 1;
50 }
51
52 /*
53  * decode crush map
54  */
55 static int crush_decode_uniform_bucket(void **p, void *end,
56                                        struct crush_bucket_uniform *b)
57 {
58         dout("crush_decode_uniform_bucket %p to %p\n", *p, end);
59         ceph_decode_need(p, end, (1+b->h.size) * sizeof(u32), bad);
60         b->item_weight = ceph_decode_32(p);
61         return 0;
62 bad:
63         return -EINVAL;
64 }
65
66 static int crush_decode_list_bucket(void **p, void *end,
67                                     struct crush_bucket_list *b)
68 {
69         int j;
70         dout("crush_decode_list_bucket %p to %p\n", *p, end);
71         b->item_weights = kcalloc(b->h.size, sizeof(u32), GFP_NOFS);
72         if (b->item_weights == NULL)
73                 return -ENOMEM;
74         b->sum_weights = kcalloc(b->h.size, sizeof(u32), GFP_NOFS);
75         if (b->sum_weights == NULL)
76                 return -ENOMEM;
77         ceph_decode_need(p, end, 2 * b->h.size * sizeof(u32), bad);
78         for (j = 0; j < b->h.size; j++) {
79                 b->item_weights[j] = ceph_decode_32(p);
80                 b->sum_weights[j] = ceph_decode_32(p);
81         }
82         return 0;
83 bad:
84         return -EINVAL;
85 }
86
87 static int crush_decode_tree_bucket(void **p, void *end,
88                                     struct crush_bucket_tree *b)
89 {
90         int j;
91         dout("crush_decode_tree_bucket %p to %p\n", *p, end);
92         ceph_decode_32_safe(p, end, b->num_nodes, bad);
93         b->node_weights = kcalloc(b->num_nodes, sizeof(u32), GFP_NOFS);
94         if (b->node_weights == NULL)
95                 return -ENOMEM;
96         ceph_decode_need(p, end, b->num_nodes * sizeof(u32), bad);
97         for (j = 0; j < b->num_nodes; j++)
98                 b->node_weights[j] = ceph_decode_32(p);
99         return 0;
100 bad:
101         return -EINVAL;
102 }
103
104 static int crush_decode_straw_bucket(void **p, void *end,
105                                      struct crush_bucket_straw *b)
106 {
107         int j;
108         dout("crush_decode_straw_bucket %p to %p\n", *p, end);
109         b->item_weights = kcalloc(b->h.size, sizeof(u32), GFP_NOFS);
110         if (b->item_weights == NULL)
111                 return -ENOMEM;
112         b->straws = kcalloc(b->h.size, sizeof(u32), GFP_NOFS);
113         if (b->straws == NULL)
114                 return -ENOMEM;
115         ceph_decode_need(p, end, 2 * b->h.size * sizeof(u32), bad);
116         for (j = 0; j < b->h.size; j++) {
117                 b->item_weights[j] = ceph_decode_32(p);
118                 b->straws[j] = ceph_decode_32(p);
119         }
120         return 0;
121 bad:
122         return -EINVAL;
123 }
124
125 static int skip_name_map(void **p, void *end)
126 {
127         int len;
128         ceph_decode_32_safe(p, end, len ,bad);
129         while (len--) {
130                 int strlen;
131                 *p += sizeof(u32);
132                 ceph_decode_32_safe(p, end, strlen, bad);
133                 *p += strlen;
134 }
135         return 0;
136 bad:
137         return -EINVAL;
138 }
139
140 static struct crush_map *crush_decode(void *pbyval, void *end)
141 {
142         struct crush_map *c;
143         int err = -EINVAL;
144         int i, j;
145         void **p = &pbyval;
146         void *start = pbyval;
147         u32 magic;
148         u32 num_name_maps;
149
150         dout("crush_decode %p to %p len %d\n", *p, end, (int)(end - *p));
151
152         c = kzalloc(sizeof(*c), GFP_NOFS);
153         if (c == NULL)
154                 return ERR_PTR(-ENOMEM);
155
156         /* set tunables to default values */
157         c->choose_local_tries = 2;
158         c->choose_local_fallback_tries = 5;
159         c->choose_total_tries = 19;
160         c->chooseleaf_descend_once = 0;
161
162         ceph_decode_need(p, end, 4*sizeof(u32), bad);
163         magic = ceph_decode_32(p);
164         if (magic != CRUSH_MAGIC) {
165                 pr_err("crush_decode magic %x != current %x\n",
166                        (unsigned int)magic, (unsigned int)CRUSH_MAGIC);
167                 goto bad;
168         }
169         c->max_buckets = ceph_decode_32(p);
170         c->max_rules = ceph_decode_32(p);
171         c->max_devices = ceph_decode_32(p);
172
173         c->buckets = kcalloc(c->max_buckets, sizeof(*c->buckets), GFP_NOFS);
174         if (c->buckets == NULL)
175                 goto badmem;
176         c->rules = kcalloc(c->max_rules, sizeof(*c->rules), GFP_NOFS);
177         if (c->rules == NULL)
178                 goto badmem;
179
180         /* buckets */
181         for (i = 0; i < c->max_buckets; i++) {
182                 int size = 0;
183                 u32 alg;
184                 struct crush_bucket *b;
185
186                 ceph_decode_32_safe(p, end, alg, bad);
187                 if (alg == 0) {
188                         c->buckets[i] = NULL;
189                         continue;
190                 }
191                 dout("crush_decode bucket %d off %x %p to %p\n",
192                      i, (int)(*p-start), *p, end);
193
194                 switch (alg) {
195                 case CRUSH_BUCKET_UNIFORM:
196                         size = sizeof(struct crush_bucket_uniform);
197                         break;
198                 case CRUSH_BUCKET_LIST:
199                         size = sizeof(struct crush_bucket_list);
200                         break;
201                 case CRUSH_BUCKET_TREE:
202                         size = sizeof(struct crush_bucket_tree);
203                         break;
204                 case CRUSH_BUCKET_STRAW:
205                         size = sizeof(struct crush_bucket_straw);
206                         break;
207                 default:
208                         err = -EINVAL;
209                         goto bad;
210                 }
211                 BUG_ON(size == 0);
212                 b = c->buckets[i] = kzalloc(size, GFP_NOFS);
213                 if (b == NULL)
214                         goto badmem;
215
216                 ceph_decode_need(p, end, 4*sizeof(u32), bad);
217                 b->id = ceph_decode_32(p);
218                 b->type = ceph_decode_16(p);
219                 b->alg = ceph_decode_8(p);
220                 b->hash = ceph_decode_8(p);
221                 b->weight = ceph_decode_32(p);
222                 b->size = ceph_decode_32(p);
223
224                 dout("crush_decode bucket size %d off %x %p to %p\n",
225                      b->size, (int)(*p-start), *p, end);
226
227                 b->items = kcalloc(b->size, sizeof(__s32), GFP_NOFS);
228                 if (b->items == NULL)
229                         goto badmem;
230                 b->perm = kcalloc(b->size, sizeof(u32), GFP_NOFS);
231                 if (b->perm == NULL)
232                         goto badmem;
233                 b->perm_n = 0;
234
235                 ceph_decode_need(p, end, b->size*sizeof(u32), bad);
236                 for (j = 0; j < b->size; j++)
237                         b->items[j] = ceph_decode_32(p);
238
239                 switch (b->alg) {
240                 case CRUSH_BUCKET_UNIFORM:
241                         err = crush_decode_uniform_bucket(p, end,
242                                   (struct crush_bucket_uniform *)b);
243                         if (err < 0)
244                                 goto bad;
245                         break;
246                 case CRUSH_BUCKET_LIST:
247                         err = crush_decode_list_bucket(p, end,
248                                (struct crush_bucket_list *)b);
249                         if (err < 0)
250                                 goto bad;
251                         break;
252                 case CRUSH_BUCKET_TREE:
253                         err = crush_decode_tree_bucket(p, end,
254                                 (struct crush_bucket_tree *)b);
255                         if (err < 0)
256                                 goto bad;
257                         break;
258                 case CRUSH_BUCKET_STRAW:
259                         err = crush_decode_straw_bucket(p, end,
260                                 (struct crush_bucket_straw *)b);
261                         if (err < 0)
262                                 goto bad;
263                         break;
264                 }
265         }
266
267         /* rules */
268         dout("rule vec is %p\n", c->rules);
269         for (i = 0; i < c->max_rules; i++) {
270                 u32 yes;
271                 struct crush_rule *r;
272
273                 ceph_decode_32_safe(p, end, yes, bad);
274                 if (!yes) {
275                         dout("crush_decode NO rule %d off %x %p to %p\n",
276                              i, (int)(*p-start), *p, end);
277                         c->rules[i] = NULL;
278                         continue;
279                 }
280
281                 dout("crush_decode rule %d off %x %p to %p\n",
282                      i, (int)(*p-start), *p, end);
283
284                 /* len */
285                 ceph_decode_32_safe(p, end, yes, bad);
286 #if BITS_PER_LONG == 32
287                 err = -EINVAL;
288                 if (yes > (ULONG_MAX - sizeof(*r))
289                           / sizeof(struct crush_rule_step))
290                         goto bad;
291 #endif
292                 r = c->rules[i] = kmalloc(sizeof(*r) +
293                                           yes*sizeof(struct crush_rule_step),
294                                           GFP_NOFS);
295                 if (r == NULL)
296                         goto badmem;
297                 dout(" rule %d is at %p\n", i, r);
298                 r->len = yes;
299                 ceph_decode_copy_safe(p, end, &r->mask, 4, bad); /* 4 u8's */
300                 ceph_decode_need(p, end, r->len*3*sizeof(u32), bad);
301                 for (j = 0; j < r->len; j++) {
302                         r->steps[j].op = ceph_decode_32(p);
303                         r->steps[j].arg1 = ceph_decode_32(p);
304                         r->steps[j].arg2 = ceph_decode_32(p);
305                 }
306         }
307
308         /* ignore trailing name maps. */
309         for (num_name_maps = 0; num_name_maps < 3; num_name_maps++) {
310                 err = skip_name_map(p, end);
311                 if (err < 0)
312                         goto done;
313         }
314
315         /* tunables */
316         ceph_decode_need(p, end, 3*sizeof(u32), done);
317         c->choose_local_tries = ceph_decode_32(p);
318         c->choose_local_fallback_tries =  ceph_decode_32(p);
319         c->choose_total_tries = ceph_decode_32(p);
320         dout("crush decode tunable choose_local_tries = %d",
321              c->choose_local_tries);
322         dout("crush decode tunable choose_local_fallback_tries = %d",
323              c->choose_local_fallback_tries);
324         dout("crush decode tunable choose_total_tries = %d",
325              c->choose_total_tries);
326
327         ceph_decode_need(p, end, sizeof(u32), done);
328         c->chooseleaf_descend_once = ceph_decode_32(p);
329         dout("crush decode tunable chooseleaf_descend_once = %d",
330              c->chooseleaf_descend_once);
331
332 done:
333         dout("crush_decode success\n");
334         return c;
335
336 badmem:
337         err = -ENOMEM;
338 bad:
339         dout("crush_decode fail %d\n", err);
340         crush_destroy(c);
341         return ERR_PTR(err);
342 }
343
344 /*
345  * rbtree of pg_mapping for handling pg_temp (explicit mapping of pgid
346  * to a set of osds)
347  */
348 static int pgid_cmp(struct ceph_pg l, struct ceph_pg r)
349 {
350         if (l.pool < r.pool)
351                 return -1;
352         if (l.pool > r.pool)
353                 return 1;
354         if (l.seed < r.seed)
355                 return -1;
356         if (l.seed > r.seed)
357                 return 1;
358         return 0;
359 }
360
361 static int __insert_pg_mapping(struct ceph_pg_mapping *new,
362                                struct rb_root *root)
363 {
364         struct rb_node **p = &root->rb_node;
365         struct rb_node *parent = NULL;
366         struct ceph_pg_mapping *pg = NULL;
367         int c;
368
369         dout("__insert_pg_mapping %llx %p\n", *(u64 *)&new->pgid, new);
370         while (*p) {
371                 parent = *p;
372                 pg = rb_entry(parent, struct ceph_pg_mapping, node);
373                 c = pgid_cmp(new->pgid, pg->pgid);
374                 if (c < 0)
375                         p = &(*p)->rb_left;
376                 else if (c > 0)
377                         p = &(*p)->rb_right;
378                 else
379                         return -EEXIST;
380         }
381
382         rb_link_node(&new->node, parent, p);
383         rb_insert_color(&new->node, root);
384         return 0;
385 }
386
387 static struct ceph_pg_mapping *__lookup_pg_mapping(struct rb_root *root,
388                                                    struct ceph_pg pgid)
389 {
390         struct rb_node *n = root->rb_node;
391         struct ceph_pg_mapping *pg;
392         int c;
393
394         while (n) {
395                 pg = rb_entry(n, struct ceph_pg_mapping, node);
396                 c = pgid_cmp(pgid, pg->pgid);
397                 if (c < 0) {
398                         n = n->rb_left;
399                 } else if (c > 0) {
400                         n = n->rb_right;
401                 } else {
402                         dout("__lookup_pg_mapping %lld.%x got %p\n",
403                              pgid.pool, pgid.seed, pg);
404                         return pg;
405                 }
406         }
407         return NULL;
408 }
409
410 static int __remove_pg_mapping(struct rb_root *root, struct ceph_pg pgid)
411 {
412         struct ceph_pg_mapping *pg = __lookup_pg_mapping(root, pgid);
413
414         if (pg) {
415                 dout("__remove_pg_mapping %lld.%x %p\n", pgid.pool, pgid.seed,
416                      pg);
417                 rb_erase(&pg->node, root);
418                 kfree(pg);
419                 return 0;
420         }
421         dout("__remove_pg_mapping %lld.%x dne\n", pgid.pool, pgid.seed);
422         return -ENOENT;
423 }
424
425 /*
426  * rbtree of pg pool info
427  */
428 static int __insert_pg_pool(struct rb_root *root, struct ceph_pg_pool_info *new)
429 {
430         struct rb_node **p = &root->rb_node;
431         struct rb_node *parent = NULL;
432         struct ceph_pg_pool_info *pi = NULL;
433
434         while (*p) {
435                 parent = *p;
436                 pi = rb_entry(parent, struct ceph_pg_pool_info, node);
437                 if (new->id < pi->id)
438                         p = &(*p)->rb_left;
439                 else if (new->id > pi->id)
440                         p = &(*p)->rb_right;
441                 else
442                         return -EEXIST;
443         }
444
445         rb_link_node(&new->node, parent, p);
446         rb_insert_color(&new->node, root);
447         return 0;
448 }
449
450 static struct ceph_pg_pool_info *__lookup_pg_pool(struct rb_root *root, u64 id)
451 {
452         struct ceph_pg_pool_info *pi;
453         struct rb_node *n = root->rb_node;
454
455         while (n) {
456                 pi = rb_entry(n, struct ceph_pg_pool_info, node);
457                 if (id < pi->id)
458                         n = n->rb_left;
459                 else if (id > pi->id)
460                         n = n->rb_right;
461                 else
462                         return pi;
463         }
464         return NULL;
465 }
466
467 const char *ceph_pg_pool_name_by_id(struct ceph_osdmap *map, u64 id)
468 {
469         struct ceph_pg_pool_info *pi;
470
471         if (id == CEPH_NOPOOL)
472                 return NULL;
473
474         if (WARN_ON_ONCE(id > (u64) INT_MAX))
475                 return NULL;
476
477         pi = __lookup_pg_pool(&map->pg_pools, (int) id);
478
479         return pi ? pi->name : NULL;
480 }
481 EXPORT_SYMBOL(ceph_pg_pool_name_by_id);
482
483 int ceph_pg_poolid_by_name(struct ceph_osdmap *map, const char *name)
484 {
485         struct rb_node *rbp;
486
487         for (rbp = rb_first(&map->pg_pools); rbp; rbp = rb_next(rbp)) {
488                 struct ceph_pg_pool_info *pi =
489                         rb_entry(rbp, struct ceph_pg_pool_info, node);
490                 if (pi->name && strcmp(pi->name, name) == 0)
491                         return pi->id;
492         }
493         return -ENOENT;
494 }
495 EXPORT_SYMBOL(ceph_pg_poolid_by_name);
496
497 static void __remove_pg_pool(struct rb_root *root, struct ceph_pg_pool_info *pi)
498 {
499         rb_erase(&pi->node, root);
500         kfree(pi->name);
501         kfree(pi);
502 }
503
504 static int __decode_pool(void **p, void *end, struct ceph_pg_pool_info *pi)
505 {
506         u8 ev, cv;
507         unsigned len, num;
508         void *pool_end;
509
510         ceph_decode_need(p, end, 2 + 4, bad);
511         ev = ceph_decode_8(p);  /* encoding version */
512         cv = ceph_decode_8(p); /* compat version */
513         if (ev < 5) {
514                 pr_warning("got v %d < 5 cv %d of ceph_pg_pool\n", ev, cv);
515                 return -EINVAL;
516         }
517         if (cv > 7) {
518                 pr_warning("got v %d cv %d > 7 of ceph_pg_pool\n", ev, cv);
519                 return -EINVAL;
520         }
521         len = ceph_decode_32(p);
522         ceph_decode_need(p, end, len, bad);
523         pool_end = *p + len;
524
525         pi->type = ceph_decode_8(p);
526         pi->size = ceph_decode_8(p);
527         pi->crush_ruleset = ceph_decode_8(p);
528         pi->object_hash = ceph_decode_8(p);
529
530         pi->pg_num = ceph_decode_32(p);
531         pi->pgp_num = ceph_decode_32(p);
532
533         *p += 4 + 4;  /* skip lpg* */
534         *p += 4;      /* skip last_change */
535         *p += 8 + 4;  /* skip snap_seq, snap_epoch */
536
537         /* skip snaps */
538         num = ceph_decode_32(p);
539         while (num--) {
540                 *p += 8;  /* snapid key */
541                 *p += 1 + 1; /* versions */
542                 len = ceph_decode_32(p);
543                 *p += len;
544         }
545
546         /* skip removed snaps */
547         num = ceph_decode_32(p);
548         *p += num * (8 + 8);
549
550         *p += 8;  /* skip auid */
551         pi->flags = ceph_decode_64(p);
552
553         /* ignore the rest */
554
555         *p = pool_end;
556         calc_pg_masks(pi);
557         return 0;
558
559 bad:
560         return -EINVAL;
561 }
562
563 static int __decode_pool_names(void **p, void *end, struct ceph_osdmap *map)
564 {
565         struct ceph_pg_pool_info *pi;
566         u32 num, len;
567         u64 pool;
568
569         ceph_decode_32_safe(p, end, num, bad);
570         dout(" %d pool names\n", num);
571         while (num--) {
572                 ceph_decode_64_safe(p, end, pool, bad);
573                 ceph_decode_32_safe(p, end, len, bad);
574                 dout("  pool %llu len %d\n", pool, len);
575                 ceph_decode_need(p, end, len, bad);
576                 pi = __lookup_pg_pool(&map->pg_pools, pool);
577                 if (pi) {
578                         char *name = kstrndup(*p, len, GFP_NOFS);
579
580                         if (!name)
581                                 return -ENOMEM;
582                         kfree(pi->name);
583                         pi->name = name;
584                         dout("  name is %s\n", pi->name);
585                 }
586                 *p += len;
587         }
588         return 0;
589
590 bad:
591         return -EINVAL;
592 }
593
594 /*
595  * osd map
596  */
597 void ceph_osdmap_destroy(struct ceph_osdmap *map)
598 {
599         dout("osdmap_destroy %p\n", map);
600         if (map->crush)
601                 crush_destroy(map->crush);
602         while (!RB_EMPTY_ROOT(&map->pg_temp)) {
603                 struct ceph_pg_mapping *pg =
604                         rb_entry(rb_first(&map->pg_temp),
605                                  struct ceph_pg_mapping, node);
606                 rb_erase(&pg->node, &map->pg_temp);
607                 kfree(pg);
608         }
609         while (!RB_EMPTY_ROOT(&map->pg_pools)) {
610                 struct ceph_pg_pool_info *pi =
611                         rb_entry(rb_first(&map->pg_pools),
612                                  struct ceph_pg_pool_info, node);
613                 __remove_pg_pool(&map->pg_pools, pi);
614         }
615         kfree(map->osd_state);
616         kfree(map->osd_weight);
617         kfree(map->osd_addr);
618         kfree(map);
619 }
620
621 /*
622  * adjust max osd value.  reallocate arrays.
623  */
624 static int osdmap_set_max_osd(struct ceph_osdmap *map, int max)
625 {
626         u8 *state;
627         struct ceph_entity_addr *addr;
628         u32 *weight;
629
630         state = kcalloc(max, sizeof(*state), GFP_NOFS);
631         addr = kcalloc(max, sizeof(*addr), GFP_NOFS);
632         weight = kcalloc(max, sizeof(*weight), GFP_NOFS);
633         if (state == NULL || addr == NULL || weight == NULL) {
634                 kfree(state);
635                 kfree(addr);
636                 kfree(weight);
637                 return -ENOMEM;
638         }
639
640         /* copy old? */
641         if (map->osd_state) {
642                 memcpy(state, map->osd_state, map->max_osd*sizeof(*state));
643                 memcpy(addr, map->osd_addr, map->max_osd*sizeof(*addr));
644                 memcpy(weight, map->osd_weight, map->max_osd*sizeof(*weight));
645                 kfree(map->osd_state);
646                 kfree(map->osd_addr);
647                 kfree(map->osd_weight);
648         }
649
650         map->osd_state = state;
651         map->osd_weight = weight;
652         map->osd_addr = addr;
653         map->max_osd = max;
654         return 0;
655 }
656
657 static int __decode_pgid(void **p, void *end, struct ceph_pg *pg)
658 {
659         u8 v;
660
661         ceph_decode_need(p, end, 1+8+4+4, bad);
662         v = ceph_decode_8(p);
663         if (v != 1)
664                 goto bad;
665         pg->pool = ceph_decode_64(p);
666         pg->seed = ceph_decode_32(p);
667         *p += 4; /* skip preferred */
668         return 0;
669
670 bad:
671         dout("error decoding pgid\n");
672         return -EINVAL;
673 }
674
675 /*
676  * decode a full map.
677  */
678 struct ceph_osdmap *osdmap_decode(void **p, void *end)
679 {
680         struct ceph_osdmap *map;
681         u16 version;
682         u32 len, max, i;
683         int err = -EINVAL;
684         void *start = *p;
685         struct ceph_pg_pool_info *pi;
686
687         dout("osdmap_decode %p to %p len %d\n", *p, end, (int)(end - *p));
688
689         map = kzalloc(sizeof(*map), GFP_NOFS);
690         if (map == NULL)
691                 return ERR_PTR(-ENOMEM);
692         map->pg_temp = RB_ROOT;
693
694         ceph_decode_16_safe(p, end, version, bad);
695         if (version > 6) {
696                 pr_warning("got unknown v %d > 6 of osdmap\n", version);
697                 goto bad;
698         }
699         if (version < 6) {
700                 pr_warning("got old v %d < 6 of osdmap\n", version);
701                 goto bad;
702         }
703
704         ceph_decode_need(p, end, 2*sizeof(u64)+6*sizeof(u32), bad);
705         ceph_decode_copy(p, &map->fsid, sizeof(map->fsid));
706         map->epoch = ceph_decode_32(p);
707         ceph_decode_copy(p, &map->created, sizeof(map->created));
708         ceph_decode_copy(p, &map->modified, sizeof(map->modified));
709
710         ceph_decode_32_safe(p, end, max, bad);
711         while (max--) {
712                 ceph_decode_need(p, end, 8 + 2, bad);
713                 err = -ENOMEM;
714                 pi = kzalloc(sizeof(*pi), GFP_NOFS);
715                 if (!pi)
716                         goto bad;
717                 pi->id = ceph_decode_64(p);
718                 err = __decode_pool(p, end, pi);
719                 if (err < 0) {
720                         kfree(pi);
721                         goto bad;
722                 }
723                 __insert_pg_pool(&map->pg_pools, pi);
724         }
725
726         err = __decode_pool_names(p, end, map);
727         if (err < 0) {
728                 dout("fail to decode pool names");
729                 goto bad;
730         }
731
732         ceph_decode_32_safe(p, end, map->pool_max, bad);
733
734         ceph_decode_32_safe(p, end, map->flags, bad);
735
736         max = ceph_decode_32(p);
737
738         /* (re)alloc osd arrays */
739         err = osdmap_set_max_osd(map, max);
740         if (err < 0)
741                 goto bad;
742         dout("osdmap_decode max_osd = %d\n", map->max_osd);
743
744         /* osds */
745         err = -EINVAL;
746         ceph_decode_need(p, end, 3*sizeof(u32) +
747                          map->max_osd*(1 + sizeof(*map->osd_weight) +
748                                        sizeof(*map->osd_addr)), bad);
749         *p += 4; /* skip length field (should match max) */
750         ceph_decode_copy(p, map->osd_state, map->max_osd);
751
752         *p += 4; /* skip length field (should match max) */
753         for (i = 0; i < map->max_osd; i++)
754                 map->osd_weight[i] = ceph_decode_32(p);
755
756         *p += 4; /* skip length field (should match max) */
757         ceph_decode_copy(p, map->osd_addr, map->max_osd*sizeof(*map->osd_addr));
758         for (i = 0; i < map->max_osd; i++)
759                 ceph_decode_addr(&map->osd_addr[i]);
760
761         /* pg_temp */
762         ceph_decode_32_safe(p, end, len, bad);
763         for (i = 0; i < len; i++) {
764                 int n, j;
765                 struct ceph_pg pgid;
766                 struct ceph_pg_mapping *pg;
767
768                 err = __decode_pgid(p, end, &pgid);
769                 if (err)
770                         goto bad;
771                 ceph_decode_need(p, end, sizeof(u32), bad);
772                 n = ceph_decode_32(p);
773                 err = -EINVAL;
774                 if (n > (UINT_MAX - sizeof(*pg)) / sizeof(u32))
775                         goto bad;
776                 ceph_decode_need(p, end, n * sizeof(u32), bad);
777                 err = -ENOMEM;
778                 pg = kmalloc(sizeof(*pg) + n*sizeof(u32), GFP_NOFS);
779                 if (!pg)
780                         goto bad;
781                 pg->pgid = pgid;
782                 pg->len = n;
783                 for (j = 0; j < n; j++)
784                         pg->osds[j] = ceph_decode_32(p);
785
786                 err = __insert_pg_mapping(pg, &map->pg_temp);
787                 if (err)
788                         goto bad;
789                 dout(" added pg_temp %lld.%x len %d\n", pgid.pool, pgid.seed,
790                      len);
791         }
792
793         /* crush */
794         ceph_decode_32_safe(p, end, len, bad);
795         dout("osdmap_decode crush len %d from off 0x%x\n", len,
796              (int)(*p - start));
797         ceph_decode_need(p, end, len, bad);
798         map->crush = crush_decode(*p, end);
799         *p += len;
800         if (IS_ERR(map->crush)) {
801                 err = PTR_ERR(map->crush);
802                 map->crush = NULL;
803                 goto bad;
804         }
805
806         /* ignore the rest of the map */
807         *p = end;
808
809         dout("osdmap_decode done %p %p\n", *p, end);
810         return map;
811
812 bad:
813         dout("osdmap_decode fail err %d\n", err);
814         ceph_osdmap_destroy(map);
815         return ERR_PTR(err);
816 }
817
818 /*
819  * decode and apply an incremental map update.
820  */
821 struct ceph_osdmap *osdmap_apply_incremental(void **p, void *end,
822                                              struct ceph_osdmap *map,
823                                              struct ceph_messenger *msgr)
824 {
825         struct crush_map *newcrush = NULL;
826         struct ceph_fsid fsid;
827         u32 epoch = 0;
828         struct ceph_timespec modified;
829         s32 len;
830         u64 pool;
831         __s64 new_pool_max;
832         __s32 new_flags, max;
833         void *start = *p;
834         int err = -EINVAL;
835         u16 version;
836
837         ceph_decode_16_safe(p, end, version, bad);
838         if (version != 6) {
839                 pr_warning("got unknown v %d != 6 of inc osdmap\n", version);
840                 goto bad;
841         }
842
843         ceph_decode_need(p, end, sizeof(fsid)+sizeof(modified)+2*sizeof(u32),
844                          bad);
845         ceph_decode_copy(p, &fsid, sizeof(fsid));
846         epoch = ceph_decode_32(p);
847         BUG_ON(epoch != map->epoch+1);
848         ceph_decode_copy(p, &modified, sizeof(modified));
849         new_pool_max = ceph_decode_64(p);
850         new_flags = ceph_decode_32(p);
851
852         /* full map? */
853         ceph_decode_32_safe(p, end, len, bad);
854         if (len > 0) {
855                 dout("apply_incremental full map len %d, %p to %p\n",
856                      len, *p, end);
857                 return osdmap_decode(p, min(*p+len, end));
858         }
859
860         /* new crush? */
861         ceph_decode_32_safe(p, end, len, bad);
862         if (len > 0) {
863                 dout("apply_incremental new crush map len %d, %p to %p\n",
864                      len, *p, end);
865                 newcrush = crush_decode(*p, min(*p+len, end));
866                 if (IS_ERR(newcrush))
867                         return ERR_CAST(newcrush);
868                 *p += len;
869         }
870
871         /* new flags? */
872         if (new_flags >= 0)
873                 map->flags = new_flags;
874         if (new_pool_max >= 0)
875                 map->pool_max = new_pool_max;
876
877         ceph_decode_need(p, end, 5*sizeof(u32), bad);
878
879         /* new max? */
880         max = ceph_decode_32(p);
881         if (max >= 0) {
882                 err = osdmap_set_max_osd(map, max);
883                 if (err < 0)
884                         goto bad;
885         }
886
887         map->epoch++;
888         map->modified = modified;
889         if (newcrush) {
890                 if (map->crush)
891                         crush_destroy(map->crush);
892                 map->crush = newcrush;
893                 newcrush = NULL;
894         }
895
896         /* new_pool */
897         ceph_decode_32_safe(p, end, len, bad);
898         while (len--) {
899                 struct ceph_pg_pool_info *pi;
900
901                 ceph_decode_64_safe(p, end, pool, bad);
902                 pi = __lookup_pg_pool(&map->pg_pools, pool);
903                 if (!pi) {
904                         pi = kzalloc(sizeof(*pi), GFP_NOFS);
905                         if (!pi) {
906                                 err = -ENOMEM;
907                                 goto bad;
908                         }
909                         pi->id = pool;
910                         __insert_pg_pool(&map->pg_pools, pi);
911                 }
912                 err = __decode_pool(p, end, pi);
913                 if (err < 0)
914                         goto bad;
915         }
916         if (version >= 5) {
917                 err = __decode_pool_names(p, end, map);
918                 if (err < 0)
919                         goto bad;
920         }
921
922         /* old_pool */
923         ceph_decode_32_safe(p, end, len, bad);
924         while (len--) {
925                 struct ceph_pg_pool_info *pi;
926
927                 ceph_decode_64_safe(p, end, pool, bad);
928                 pi = __lookup_pg_pool(&map->pg_pools, pool);
929                 if (pi)
930                         __remove_pg_pool(&map->pg_pools, pi);
931         }
932
933         /* new_up */
934         err = -EINVAL;
935         ceph_decode_32_safe(p, end, len, bad);
936         while (len--) {
937                 u32 osd;
938                 struct ceph_entity_addr addr;
939                 ceph_decode_32_safe(p, end, osd, bad);
940                 ceph_decode_copy_safe(p, end, &addr, sizeof(addr), bad);
941                 ceph_decode_addr(&addr);
942                 pr_info("osd%d up\n", osd);
943                 BUG_ON(osd >= map->max_osd);
944                 map->osd_state[osd] |= CEPH_OSD_UP;
945                 map->osd_addr[osd] = addr;
946         }
947
948         /* new_state */
949         ceph_decode_32_safe(p, end, len, bad);
950         while (len--) {
951                 u32 osd;
952                 u8 xorstate;
953                 ceph_decode_32_safe(p, end, osd, bad);
954                 xorstate = **(u8 **)p;
955                 (*p)++;  /* clean flag */
956                 if (xorstate == 0)
957                         xorstate = CEPH_OSD_UP;
958                 if (xorstate & CEPH_OSD_UP)
959                         pr_info("osd%d down\n", osd);
960                 if (osd < map->max_osd)
961                         map->osd_state[osd] ^= xorstate;
962         }
963
964         /* new_weight */
965         ceph_decode_32_safe(p, end, len, bad);
966         while (len--) {
967                 u32 osd, off;
968                 ceph_decode_need(p, end, sizeof(u32)*2, bad);
969                 osd = ceph_decode_32(p);
970                 off = ceph_decode_32(p);
971                 pr_info("osd%d weight 0x%x %s\n", osd, off,
972                      off == CEPH_OSD_IN ? "(in)" :
973                      (off == CEPH_OSD_OUT ? "(out)" : ""));
974                 if (osd < map->max_osd)
975                         map->osd_weight[osd] = off;
976         }
977
978         /* new_pg_temp */
979         ceph_decode_32_safe(p, end, len, bad);
980         while (len--) {
981                 struct ceph_pg_mapping *pg;
982                 int j;
983                 struct ceph_pg pgid;
984                 u32 pglen;
985
986                 err = __decode_pgid(p, end, &pgid);
987                 if (err)
988                         goto bad;
989                 ceph_decode_need(p, end, sizeof(u32), bad);
990                 pglen = ceph_decode_32(p);
991                 if (pglen) {
992                         ceph_decode_need(p, end, pglen*sizeof(u32), bad);
993
994                         /* removing existing (if any) */
995                         (void) __remove_pg_mapping(&map->pg_temp, pgid);
996
997                         /* insert */
998                         err = -EINVAL;
999                         if (pglen > (UINT_MAX - sizeof(*pg)) / sizeof(u32))
1000                                 goto bad;
1001                         err = -ENOMEM;
1002                         pg = kmalloc(sizeof(*pg) + sizeof(u32)*pglen, GFP_NOFS);
1003                         if (!pg)
1004                                 goto bad;
1005                         pg->pgid = pgid;
1006                         pg->len = pglen;
1007                         for (j = 0; j < pglen; j++)
1008                                 pg->osds[j] = ceph_decode_32(p);
1009                         err = __insert_pg_mapping(pg, &map->pg_temp);
1010                         if (err) {
1011                                 kfree(pg);
1012                                 goto bad;
1013                         }
1014                         dout(" added pg_temp %lld.%x len %d\n", pgid.pool,
1015                              pgid.seed, pglen);
1016                 } else {
1017                         /* remove */
1018                         __remove_pg_mapping(&map->pg_temp, pgid);
1019                 }
1020         }
1021
1022         /* ignore the rest */
1023         *p = end;
1024         return map;
1025
1026 bad:
1027         pr_err("corrupt inc osdmap epoch %d off %d (%p of %p-%p)\n",
1028                epoch, (int)(*p - start), *p, start, end);
1029         print_hex_dump(KERN_DEBUG, "osdmap: ",
1030                        DUMP_PREFIX_OFFSET, 16, 1,
1031                        start, end - start, true);
1032         if (newcrush)
1033                 crush_destroy(newcrush);
1034         return ERR_PTR(err);
1035 }
1036
1037
1038
1039
1040 /*
1041  * calculate file layout from given offset, length.
1042  * fill in correct oid, logical length, and object extent
1043  * offset, length.
1044  *
1045  * for now, we write only a single su, until we can
1046  * pass a stride back to the caller.
1047  */
1048 int ceph_calc_file_object_mapping(struct ceph_file_layout *layout,
1049                                    u64 off, u64 len,
1050                                    u64 *ono,
1051                                    u64 *oxoff, u64 *oxlen)
1052 {
1053         u32 osize = le32_to_cpu(layout->fl_object_size);
1054         u32 su = le32_to_cpu(layout->fl_stripe_unit);
1055         u32 sc = le32_to_cpu(layout->fl_stripe_count);
1056         u32 bl, stripeno, stripepos, objsetno;
1057         u32 su_per_object;
1058         u64 t, su_offset;
1059
1060         dout("mapping %llu~%llu  osize %u fl_su %u\n", off, len,
1061              osize, su);
1062         if (su == 0 || sc == 0)
1063                 goto invalid;
1064         su_per_object = osize / su;
1065         if (su_per_object == 0)
1066                 goto invalid;
1067         dout("osize %u / su %u = su_per_object %u\n", osize, su,
1068              su_per_object);
1069
1070         if ((su & ~PAGE_MASK) != 0)
1071                 goto invalid;
1072
1073         /* bl = *off / su; */
1074         t = off;
1075         do_div(t, su);
1076         bl = t;
1077         dout("off %llu / su %u = bl %u\n", off, su, bl);
1078
1079         stripeno = bl / sc;
1080         stripepos = bl % sc;
1081         objsetno = stripeno / su_per_object;
1082
1083         *ono = objsetno * sc + stripepos;
1084         dout("objset %u * sc %u = ono %u\n", objsetno, sc, (unsigned int)*ono);
1085
1086         /* *oxoff = *off % layout->fl_stripe_unit;  # offset in su */
1087         t = off;
1088         su_offset = do_div(t, su);
1089         *oxoff = su_offset + (stripeno % su_per_object) * su;
1090
1091         /*
1092          * Calculate the length of the extent being written to the selected
1093          * object. This is the minimum of the full length requested (len) or
1094          * the remainder of the current stripe being written to.
1095          */
1096         *oxlen = min_t(u64, len, su - su_offset);
1097
1098         dout(" obj extent %llu~%llu\n", *oxoff, *oxlen);
1099         return 0;
1100
1101 invalid:
1102         dout(" invalid layout\n");
1103         *ono = 0;
1104         *oxoff = 0;
1105         *oxlen = 0;
1106         return -EINVAL;
1107 }
1108 EXPORT_SYMBOL(ceph_calc_file_object_mapping);
1109
1110 /*
1111  * calculate an object layout (i.e. pgid) from an oid,
1112  * file_layout, and osdmap
1113  */
1114 int ceph_calc_ceph_pg(struct ceph_pg *pg, const char *oid,
1115                         struct ceph_osdmap *osdmap, uint64_t pool)
1116 {
1117         struct ceph_pg_pool_info *pool_info;
1118
1119         BUG_ON(!osdmap);
1120         pool_info = __lookup_pg_pool(&osdmap->pg_pools, pool);
1121         if (!pool_info)
1122                 return -EIO;
1123         pg->pool = pool;
1124         pg->seed = ceph_str_hash(pool_info->object_hash, oid, strlen(oid));
1125
1126         dout("%s '%s' pgid %lld.%x\n", __func__, oid, pg->pool, pg->seed);
1127         return 0;
1128 }
1129 EXPORT_SYMBOL(ceph_calc_ceph_pg);
1130
1131 /*
1132  * Calculate raw osd vector for the given pgid.  Return pointer to osd
1133  * array, or NULL on failure.
1134  */
1135 static int *calc_pg_raw(struct ceph_osdmap *osdmap, struct ceph_pg pgid,
1136                         int *osds, int *num)
1137 {
1138         struct ceph_pg_mapping *pg;
1139         struct ceph_pg_pool_info *pool;
1140         int ruleno;
1141         int r;
1142         u32 pps;
1143
1144         pool = __lookup_pg_pool(&osdmap->pg_pools, pgid.pool);
1145         if (!pool)
1146                 return NULL;
1147
1148         /* pg_temp? */
1149         pgid.seed = ceph_stable_mod(pgid.seed, pool->pg_num,
1150                                     pool->pgp_num_mask);
1151         pg = __lookup_pg_mapping(&osdmap->pg_temp, pgid);
1152         if (pg) {
1153                 *num = pg->len;
1154                 return pg->osds;
1155         }
1156
1157         /* crush */
1158         ruleno = crush_find_rule(osdmap->crush, pool->crush_ruleset,
1159                                  pool->type, pool->size);
1160         if (ruleno < 0) {
1161                 pr_err("no crush rule pool %lld ruleset %d type %d size %d\n",
1162                        pgid.pool, pool->crush_ruleset, pool->type,
1163                        pool->size);
1164                 return NULL;
1165         }
1166
1167         if (pool->flags & CEPH_POOL_FLAG_HASHPSPOOL) {
1168                 /* hash pool id and seed sothat pool PGs do not overlap */
1169                 pps = crush_hash32_2(CRUSH_HASH_RJENKINS1,
1170                                      ceph_stable_mod(pgid.seed, pool->pgp_num,
1171                                                      pool->pgp_num_mask),
1172                                      pgid.pool);
1173         } else {
1174                 /*
1175                  * legacy ehavior: add ps and pool together.  this is
1176                  * not a great approach because the PGs from each pool
1177                  * will overlap on top of each other: 0.5 == 1.4 ==
1178                  * 2.3 == ...
1179                  */
1180                 pps = ceph_stable_mod(pgid.seed, pool->pgp_num,
1181                                       pool->pgp_num_mask) +
1182                         (unsigned)pgid.pool;
1183         }
1184         r = crush_do_rule(osdmap->crush, ruleno, pps, osds,
1185                           min_t(int, pool->size, *num),
1186                           osdmap->osd_weight);
1187         if (r < 0) {
1188                 pr_err("error %d from crush rule: pool %lld ruleset %d type %d"
1189                        " size %d\n", r, pgid.pool, pool->crush_ruleset,
1190                        pool->type, pool->size);
1191                 return NULL;
1192         }
1193         *num = r;
1194         return osds;
1195 }
1196
1197 /*
1198  * Return acting set for given pgid.
1199  */
1200 int ceph_calc_pg_acting(struct ceph_osdmap *osdmap, struct ceph_pg pgid,
1201                         int *acting)
1202 {
1203         int rawosds[CEPH_PG_MAX_SIZE], *osds;
1204         int i, o, num = CEPH_PG_MAX_SIZE;
1205
1206         osds = calc_pg_raw(osdmap, pgid, rawosds, &num);
1207         if (!osds)
1208                 return -1;
1209
1210         /* primary is first up osd */
1211         o = 0;
1212         for (i = 0; i < num; i++)
1213                 if (ceph_osd_is_up(osdmap, osds[i]))
1214                         acting[o++] = osds[i];
1215         return o;
1216 }
1217
1218 /*
1219  * Return primary osd for given pgid, or -1 if none.
1220  */
1221 int ceph_calc_pg_primary(struct ceph_osdmap *osdmap, struct ceph_pg pgid)
1222 {
1223         int rawosds[CEPH_PG_MAX_SIZE], *osds;
1224         int i, num = CEPH_PG_MAX_SIZE;
1225
1226         osds = calc_pg_raw(osdmap, pgid, rawosds, &num);
1227         if (!osds)
1228                 return -1;
1229
1230         /* primary is first up osd */
1231         for (i = 0; i < num; i++)
1232                 if (ceph_osd_is_up(osdmap, osds[i]))
1233                         return osds[i];
1234         return -1;
1235 }
1236 EXPORT_SYMBOL(ceph_calc_pg_primary);