libceph: check for invalid mapping
[linux-2.6.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         int flag = 0;
17
18         if (!len)
19                 goto done;
20
21         *str = '\0';
22         if (state) {
23                 if (state & CEPH_OSD_EXISTS) {
24                         snprintf(str, len, "exists");
25                         flag = 1;
26                 }
27                 if (state & CEPH_OSD_UP) {
28                         snprintf(str, len, "%s%s%s", str, (flag ? ", " : ""),
29                                  "up");
30                         flag = 1;
31                 }
32         } else {
33                 snprintf(str, len, "doesn't exist");
34         }
35 done:
36         return str;
37 }
38
39 /* maps */
40
41 static int calc_bits_of(unsigned t)
42 {
43         int b = 0;
44         while (t) {
45                 t = t >> 1;
46                 b++;
47         }
48         return b;
49 }
50
51 /*
52  * the foo_mask is the smallest value 2^n-1 that is >= foo.
53  */
54 static void calc_pg_masks(struct ceph_pg_pool_info *pi)
55 {
56         pi->pg_num_mask = (1 << calc_bits_of(le32_to_cpu(pi->v.pg_num)-1)) - 1;
57         pi->pgp_num_mask =
58                 (1 << calc_bits_of(le32_to_cpu(pi->v.pgp_num)-1)) - 1;
59         pi->lpg_num_mask =
60                 (1 << calc_bits_of(le32_to_cpu(pi->v.lpg_num)-1)) - 1;
61         pi->lpgp_num_mask =
62                 (1 << calc_bits_of(le32_to_cpu(pi->v.lpgp_num)-1)) - 1;
63 }
64
65 /*
66  * decode crush map
67  */
68 static int crush_decode_uniform_bucket(void **p, void *end,
69                                        struct crush_bucket_uniform *b)
70 {
71         dout("crush_decode_uniform_bucket %p to %p\n", *p, end);
72         ceph_decode_need(p, end, (1+b->h.size) * sizeof(u32), bad);
73         b->item_weight = ceph_decode_32(p);
74         return 0;
75 bad:
76         return -EINVAL;
77 }
78
79 static int crush_decode_list_bucket(void **p, void *end,
80                                     struct crush_bucket_list *b)
81 {
82         int j;
83         dout("crush_decode_list_bucket %p to %p\n", *p, end);
84         b->item_weights = kcalloc(b->h.size, sizeof(u32), GFP_NOFS);
85         if (b->item_weights == NULL)
86                 return -ENOMEM;
87         b->sum_weights = kcalloc(b->h.size, sizeof(u32), GFP_NOFS);
88         if (b->sum_weights == NULL)
89                 return -ENOMEM;
90         ceph_decode_need(p, end, 2 * b->h.size * sizeof(u32), bad);
91         for (j = 0; j < b->h.size; j++) {
92                 b->item_weights[j] = ceph_decode_32(p);
93                 b->sum_weights[j] = ceph_decode_32(p);
94         }
95         return 0;
96 bad:
97         return -EINVAL;
98 }
99
100 static int crush_decode_tree_bucket(void **p, void *end,
101                                     struct crush_bucket_tree *b)
102 {
103         int j;
104         dout("crush_decode_tree_bucket %p to %p\n", *p, end);
105         ceph_decode_32_safe(p, end, b->num_nodes, bad);
106         b->node_weights = kcalloc(b->num_nodes, sizeof(u32), GFP_NOFS);
107         if (b->node_weights == NULL)
108                 return -ENOMEM;
109         ceph_decode_need(p, end, b->num_nodes * sizeof(u32), bad);
110         for (j = 0; j < b->num_nodes; j++)
111                 b->node_weights[j] = ceph_decode_32(p);
112         return 0;
113 bad:
114         return -EINVAL;
115 }
116
117 static int crush_decode_straw_bucket(void **p, void *end,
118                                      struct crush_bucket_straw *b)
119 {
120         int j;
121         dout("crush_decode_straw_bucket %p to %p\n", *p, end);
122         b->item_weights = kcalloc(b->h.size, sizeof(u32), GFP_NOFS);
123         if (b->item_weights == NULL)
124                 return -ENOMEM;
125         b->straws = kcalloc(b->h.size, sizeof(u32), GFP_NOFS);
126         if (b->straws == NULL)
127                 return -ENOMEM;
128         ceph_decode_need(p, end, 2 * b->h.size * sizeof(u32), bad);
129         for (j = 0; j < b->h.size; j++) {
130                 b->item_weights[j] = ceph_decode_32(p);
131                 b->straws[j] = ceph_decode_32(p);
132         }
133         return 0;
134 bad:
135         return -EINVAL;
136 }
137
138 static struct crush_map *crush_decode(void *pbyval, void *end)
139 {
140         struct crush_map *c;
141         int err = -EINVAL;
142         int i, j;
143         void **p = &pbyval;
144         void *start = pbyval;
145         u32 magic;
146
147         dout("crush_decode %p to %p len %d\n", *p, end, (int)(end - *p));
148
149         c = kzalloc(sizeof(*c), GFP_NOFS);
150         if (c == NULL)
151                 return ERR_PTR(-ENOMEM);
152
153         ceph_decode_need(p, end, 4*sizeof(u32), bad);
154         magic = ceph_decode_32(p);
155         if (magic != CRUSH_MAGIC) {
156                 pr_err("crush_decode magic %x != current %x\n",
157                        (unsigned)magic, (unsigned)CRUSH_MAGIC);
158                 goto bad;
159         }
160         c->max_buckets = ceph_decode_32(p);
161         c->max_rules = ceph_decode_32(p);
162         c->max_devices = ceph_decode_32(p);
163
164         c->device_parents = kcalloc(c->max_devices, sizeof(u32), GFP_NOFS);
165         if (c->device_parents == NULL)
166                 goto badmem;
167         c->bucket_parents = kcalloc(c->max_buckets, sizeof(u32), GFP_NOFS);
168         if (c->bucket_parents == NULL)
169                 goto badmem;
170
171         c->buckets = kcalloc(c->max_buckets, sizeof(*c->buckets), GFP_NOFS);
172         if (c->buckets == NULL)
173                 goto badmem;
174         c->rules = kcalloc(c->max_rules, sizeof(*c->rules), GFP_NOFS);
175         if (c->rules == NULL)
176                 goto badmem;
177
178         /* buckets */
179         for (i = 0; i < c->max_buckets; i++) {
180                 int size = 0;
181                 u32 alg;
182                 struct crush_bucket *b;
183
184                 ceph_decode_32_safe(p, end, alg, bad);
185                 if (alg == 0) {
186                         c->buckets[i] = NULL;
187                         continue;
188                 }
189                 dout("crush_decode bucket %d off %x %p to %p\n",
190                      i, (int)(*p-start), *p, end);
191
192                 switch (alg) {
193                 case CRUSH_BUCKET_UNIFORM:
194                         size = sizeof(struct crush_bucket_uniform);
195                         break;
196                 case CRUSH_BUCKET_LIST:
197                         size = sizeof(struct crush_bucket_list);
198                         break;
199                 case CRUSH_BUCKET_TREE:
200                         size = sizeof(struct crush_bucket_tree);
201                         break;
202                 case CRUSH_BUCKET_STRAW:
203                         size = sizeof(struct crush_bucket_straw);
204                         break;
205                 default:
206                         err = -EINVAL;
207                         goto bad;
208                 }
209                 BUG_ON(size == 0);
210                 b = c->buckets[i] = kzalloc(size, GFP_NOFS);
211                 if (b == NULL)
212                         goto badmem;
213
214                 ceph_decode_need(p, end, 4*sizeof(u32), bad);
215                 b->id = ceph_decode_32(p);
216                 b->type = ceph_decode_16(p);
217                 b->alg = ceph_decode_8(p);
218                 b->hash = ceph_decode_8(p);
219                 b->weight = ceph_decode_32(p);
220                 b->size = ceph_decode_32(p);
221
222                 dout("crush_decode bucket size %d off %x %p to %p\n",
223                      b->size, (int)(*p-start), *p, end);
224
225                 b->items = kcalloc(b->size, sizeof(__s32), GFP_NOFS);
226                 if (b->items == NULL)
227                         goto badmem;
228                 b->perm = kcalloc(b->size, sizeof(u32), GFP_NOFS);
229                 if (b->perm == NULL)
230                         goto badmem;
231                 b->perm_n = 0;
232
233                 ceph_decode_need(p, end, b->size*sizeof(u32), bad);
234                 for (j = 0; j < b->size; j++)
235                         b->items[j] = ceph_decode_32(p);
236
237                 switch (b->alg) {
238                 case CRUSH_BUCKET_UNIFORM:
239                         err = crush_decode_uniform_bucket(p, end,
240                                   (struct crush_bucket_uniform *)b);
241                         if (err < 0)
242                                 goto bad;
243                         break;
244                 case CRUSH_BUCKET_LIST:
245                         err = crush_decode_list_bucket(p, end,
246                                (struct crush_bucket_list *)b);
247                         if (err < 0)
248                                 goto bad;
249                         break;
250                 case CRUSH_BUCKET_TREE:
251                         err = crush_decode_tree_bucket(p, end,
252                                 (struct crush_bucket_tree *)b);
253                         if (err < 0)
254                                 goto bad;
255                         break;
256                 case CRUSH_BUCKET_STRAW:
257                         err = crush_decode_straw_bucket(p, end,
258                                 (struct crush_bucket_straw *)b);
259                         if (err < 0)
260                                 goto bad;
261                         break;
262                 }
263         }
264
265         /* rules */
266         dout("rule vec is %p\n", c->rules);
267         for (i = 0; i < c->max_rules; i++) {
268                 u32 yes;
269                 struct crush_rule *r;
270
271                 ceph_decode_32_safe(p, end, yes, bad);
272                 if (!yes) {
273                         dout("crush_decode NO rule %d off %x %p to %p\n",
274                              i, (int)(*p-start), *p, end);
275                         c->rules[i] = NULL;
276                         continue;
277                 }
278
279                 dout("crush_decode rule %d off %x %p to %p\n",
280                      i, (int)(*p-start), *p, end);
281
282                 /* len */
283                 ceph_decode_32_safe(p, end, yes, bad);
284 #if BITS_PER_LONG == 32
285                 err = -EINVAL;
286                 if (yes > (ULONG_MAX - sizeof(*r))
287                           / sizeof(struct crush_rule_step))
288                         goto bad;
289 #endif
290                 r = c->rules[i] = kmalloc(sizeof(*r) +
291                                           yes*sizeof(struct crush_rule_step),
292                                           GFP_NOFS);
293                 if (r == NULL)
294                         goto badmem;
295                 dout(" rule %d is at %p\n", i, r);
296                 r->len = yes;
297                 ceph_decode_copy_safe(p, end, &r->mask, 4, bad); /* 4 u8's */
298                 ceph_decode_need(p, end, r->len*3*sizeof(u32), bad);
299                 for (j = 0; j < r->len; j++) {
300                         r->steps[j].op = ceph_decode_32(p);
301                         r->steps[j].arg1 = ceph_decode_32(p);
302                         r->steps[j].arg2 = ceph_decode_32(p);
303                 }
304         }
305
306         /* ignore trailing name maps. */
307
308         dout("crush_decode success\n");
309         return c;
310
311 badmem:
312         err = -ENOMEM;
313 bad:
314         dout("crush_decode fail %d\n", err);
315         crush_destroy(c);
316         return ERR_PTR(err);
317 }
318
319 /*
320  * rbtree of pg_mapping for handling pg_temp (explicit mapping of pgid
321  * to a set of osds)
322  */
323 static int pgid_cmp(struct ceph_pg l, struct ceph_pg r)
324 {
325         u64 a = *(u64 *)&l;
326         u64 b = *(u64 *)&r;
327
328         if (a < b)
329                 return -1;
330         if (a > b)
331                 return 1;
332         return 0;
333 }
334
335 static int __insert_pg_mapping(struct ceph_pg_mapping *new,
336                                struct rb_root *root)
337 {
338         struct rb_node **p = &root->rb_node;
339         struct rb_node *parent = NULL;
340         struct ceph_pg_mapping *pg = NULL;
341         int c;
342
343         dout("__insert_pg_mapping %llx %p\n", *(u64 *)&new->pgid, new);
344         while (*p) {
345                 parent = *p;
346                 pg = rb_entry(parent, struct ceph_pg_mapping, node);
347                 c = pgid_cmp(new->pgid, pg->pgid);
348                 if (c < 0)
349                         p = &(*p)->rb_left;
350                 else if (c > 0)
351                         p = &(*p)->rb_right;
352                 else
353                         return -EEXIST;
354         }
355
356         rb_link_node(&new->node, parent, p);
357         rb_insert_color(&new->node, root);
358         return 0;
359 }
360
361 static struct ceph_pg_mapping *__lookup_pg_mapping(struct rb_root *root,
362                                                    struct ceph_pg pgid)
363 {
364         struct rb_node *n = root->rb_node;
365         struct ceph_pg_mapping *pg;
366         int c;
367
368         while (n) {
369                 pg = rb_entry(n, struct ceph_pg_mapping, node);
370                 c = pgid_cmp(pgid, pg->pgid);
371                 if (c < 0) {
372                         n = n->rb_left;
373                 } else if (c > 0) {
374                         n = n->rb_right;
375                 } else {
376                         dout("__lookup_pg_mapping %llx got %p\n",
377                              *(u64 *)&pgid, pg);
378                         return pg;
379                 }
380         }
381         return NULL;
382 }
383
384 static int __remove_pg_mapping(struct rb_root *root, struct ceph_pg pgid)
385 {
386         struct ceph_pg_mapping *pg = __lookup_pg_mapping(root, pgid);
387
388         if (pg) {
389                 dout("__remove_pg_mapping %llx %p\n", *(u64 *)&pgid, pg);
390                 rb_erase(&pg->node, root);
391                 kfree(pg);
392                 return 0;
393         }
394         dout("__remove_pg_mapping %llx dne\n", *(u64 *)&pgid);
395         return -ENOENT;
396 }
397
398 /*
399  * rbtree of pg pool info
400  */
401 static int __insert_pg_pool(struct rb_root *root, struct ceph_pg_pool_info *new)
402 {
403         struct rb_node **p = &root->rb_node;
404         struct rb_node *parent = NULL;
405         struct ceph_pg_pool_info *pi = NULL;
406
407         while (*p) {
408                 parent = *p;
409                 pi = rb_entry(parent, struct ceph_pg_pool_info, node);
410                 if (new->id < pi->id)
411                         p = &(*p)->rb_left;
412                 else if (new->id > pi->id)
413                         p = &(*p)->rb_right;
414                 else
415                         return -EEXIST;
416         }
417
418         rb_link_node(&new->node, parent, p);
419         rb_insert_color(&new->node, root);
420         return 0;
421 }
422
423 static struct ceph_pg_pool_info *__lookup_pg_pool(struct rb_root *root, int id)
424 {
425         struct ceph_pg_pool_info *pi;
426         struct rb_node *n = root->rb_node;
427
428         while (n) {
429                 pi = rb_entry(n, struct ceph_pg_pool_info, node);
430                 if (id < pi->id)
431                         n = n->rb_left;
432                 else if (id > pi->id)
433                         n = n->rb_right;
434                 else
435                         return pi;
436         }
437         return NULL;
438 }
439
440 int ceph_pg_poolid_by_name(struct ceph_osdmap *map, const char *name)
441 {
442         struct rb_node *rbp;
443
444         for (rbp = rb_first(&map->pg_pools); rbp; rbp = rb_next(rbp)) {
445                 struct ceph_pg_pool_info *pi =
446                         rb_entry(rbp, struct ceph_pg_pool_info, node);
447                 if (pi->name && strcmp(pi->name, name) == 0)
448                         return pi->id;
449         }
450         return -ENOENT;
451 }
452 EXPORT_SYMBOL(ceph_pg_poolid_by_name);
453
454 static void __remove_pg_pool(struct rb_root *root, struct ceph_pg_pool_info *pi)
455 {
456         rb_erase(&pi->node, root);
457         kfree(pi->name);
458         kfree(pi);
459 }
460
461 static int __decode_pool(void **p, void *end, struct ceph_pg_pool_info *pi)
462 {
463         unsigned n, m;
464
465         ceph_decode_copy(p, &pi->v, sizeof(pi->v));
466         calc_pg_masks(pi);
467
468         /* num_snaps * snap_info_t */
469         n = le32_to_cpu(pi->v.num_snaps);
470         while (n--) {
471                 ceph_decode_need(p, end, sizeof(u64) + 1 + sizeof(u64) +
472                                  sizeof(struct ceph_timespec), bad);
473                 *p += sizeof(u64) +       /* key */
474                         1 + sizeof(u64) + /* u8, snapid */
475                         sizeof(struct ceph_timespec);
476                 m = ceph_decode_32(p);    /* snap name */
477                 *p += m;
478         }
479
480         *p += le32_to_cpu(pi->v.num_removed_snap_intervals) * sizeof(u64) * 2;
481         return 0;
482
483 bad:
484         return -EINVAL;
485 }
486
487 static int __decode_pool_names(void **p, void *end, struct ceph_osdmap *map)
488 {
489         struct ceph_pg_pool_info *pi;
490         u32 num, len, pool;
491
492         ceph_decode_32_safe(p, end, num, bad);
493         dout(" %d pool names\n", num);
494         while (num--) {
495                 ceph_decode_32_safe(p, end, pool, bad);
496                 ceph_decode_32_safe(p, end, len, bad);
497                 dout("  pool %d len %d\n", pool, len);
498                 ceph_decode_need(p, end, len, bad);
499                 pi = __lookup_pg_pool(&map->pg_pools, pool);
500                 if (pi) {
501                         char *name = kstrndup(*p, len, GFP_NOFS);
502
503                         if (!name)
504                                 return -ENOMEM;
505                         kfree(pi->name);
506                         pi->name = name;
507                         dout("  name is %s\n", pi->name);
508                 }
509                 *p += len;
510         }
511         return 0;
512
513 bad:
514         return -EINVAL;
515 }
516
517 /*
518  * osd map
519  */
520 void ceph_osdmap_destroy(struct ceph_osdmap *map)
521 {
522         dout("osdmap_destroy %p\n", map);
523         if (map->crush)
524                 crush_destroy(map->crush);
525         while (!RB_EMPTY_ROOT(&map->pg_temp)) {
526                 struct ceph_pg_mapping *pg =
527                         rb_entry(rb_first(&map->pg_temp),
528                                  struct ceph_pg_mapping, node);
529                 rb_erase(&pg->node, &map->pg_temp);
530                 kfree(pg);
531         }
532         while (!RB_EMPTY_ROOT(&map->pg_pools)) {
533                 struct ceph_pg_pool_info *pi =
534                         rb_entry(rb_first(&map->pg_pools),
535                                  struct ceph_pg_pool_info, node);
536                 __remove_pg_pool(&map->pg_pools, pi);
537         }
538         kfree(map->osd_state);
539         kfree(map->osd_weight);
540         kfree(map->osd_addr);
541         kfree(map);
542 }
543
544 /*
545  * adjust max osd value.  reallocate arrays.
546  */
547 static int osdmap_set_max_osd(struct ceph_osdmap *map, int max)
548 {
549         u8 *state;
550         struct ceph_entity_addr *addr;
551         u32 *weight;
552
553         state = kcalloc(max, sizeof(*state), GFP_NOFS);
554         addr = kcalloc(max, sizeof(*addr), GFP_NOFS);
555         weight = kcalloc(max, sizeof(*weight), GFP_NOFS);
556         if (state == NULL || addr == NULL || weight == NULL) {
557                 kfree(state);
558                 kfree(addr);
559                 kfree(weight);
560                 return -ENOMEM;
561         }
562
563         /* copy old? */
564         if (map->osd_state) {
565                 memcpy(state, map->osd_state, map->max_osd*sizeof(*state));
566                 memcpy(addr, map->osd_addr, map->max_osd*sizeof(*addr));
567                 memcpy(weight, map->osd_weight, map->max_osd*sizeof(*weight));
568                 kfree(map->osd_state);
569                 kfree(map->osd_addr);
570                 kfree(map->osd_weight);
571         }
572
573         map->osd_state = state;
574         map->osd_weight = weight;
575         map->osd_addr = addr;
576         map->max_osd = max;
577         return 0;
578 }
579
580 /*
581  * decode a full map.
582  */
583 struct ceph_osdmap *osdmap_decode(void **p, void *end)
584 {
585         struct ceph_osdmap *map;
586         u16 version;
587         u32 len, max, i;
588         u8 ev;
589         int err = -EINVAL;
590         void *start = *p;
591         struct ceph_pg_pool_info *pi;
592
593         dout("osdmap_decode %p to %p len %d\n", *p, end, (int)(end - *p));
594
595         map = kzalloc(sizeof(*map), GFP_NOFS);
596         if (map == NULL)
597                 return ERR_PTR(-ENOMEM);
598         map->pg_temp = RB_ROOT;
599
600         ceph_decode_16_safe(p, end, version, bad);
601         if (version > CEPH_OSDMAP_VERSION) {
602                 pr_warning("got unknown v %d > %d of osdmap\n", version,
603                            CEPH_OSDMAP_VERSION);
604                 goto bad;
605         }
606
607         ceph_decode_need(p, end, 2*sizeof(u64)+6*sizeof(u32), bad);
608         ceph_decode_copy(p, &map->fsid, sizeof(map->fsid));
609         map->epoch = ceph_decode_32(p);
610         ceph_decode_copy(p, &map->created, sizeof(map->created));
611         ceph_decode_copy(p, &map->modified, sizeof(map->modified));
612
613         ceph_decode_32_safe(p, end, max, bad);
614         while (max--) {
615                 ceph_decode_need(p, end, 4 + 1 + sizeof(pi->v), bad);
616                 pi = kzalloc(sizeof(*pi), GFP_NOFS);
617                 if (!pi)
618                         goto bad;
619                 pi->id = ceph_decode_32(p);
620                 ev = ceph_decode_8(p); /* encoding version */
621                 if (ev > CEPH_PG_POOL_VERSION) {
622                         pr_warning("got unknown v %d > %d of ceph_pg_pool\n",
623                                    ev, CEPH_PG_POOL_VERSION);
624                         kfree(pi);
625                         goto bad;
626                 }
627                 err = __decode_pool(p, end, pi);
628                 if (err < 0) {
629                         kfree(pi);
630                         goto bad;
631                 }
632                 __insert_pg_pool(&map->pg_pools, pi);
633         }
634
635         if (version >= 5 && __decode_pool_names(p, end, map) < 0)
636                 goto bad;
637
638         ceph_decode_32_safe(p, end, map->pool_max, bad);
639
640         ceph_decode_32_safe(p, end, map->flags, bad);
641
642         max = ceph_decode_32(p);
643
644         /* (re)alloc osd arrays */
645         err = osdmap_set_max_osd(map, max);
646         if (err < 0)
647                 goto bad;
648         dout("osdmap_decode max_osd = %d\n", map->max_osd);
649
650         /* osds */
651         err = -EINVAL;
652         ceph_decode_need(p, end, 3*sizeof(u32) +
653                          map->max_osd*(1 + sizeof(*map->osd_weight) +
654                                        sizeof(*map->osd_addr)), bad);
655         *p += 4; /* skip length field (should match max) */
656         ceph_decode_copy(p, map->osd_state, map->max_osd);
657
658         *p += 4; /* skip length field (should match max) */
659         for (i = 0; i < map->max_osd; i++)
660                 map->osd_weight[i] = ceph_decode_32(p);
661
662         *p += 4; /* skip length field (should match max) */
663         ceph_decode_copy(p, map->osd_addr, map->max_osd*sizeof(*map->osd_addr));
664         for (i = 0; i < map->max_osd; i++)
665                 ceph_decode_addr(&map->osd_addr[i]);
666
667         /* pg_temp */
668         ceph_decode_32_safe(p, end, len, bad);
669         for (i = 0; i < len; i++) {
670                 int n, j;
671                 struct ceph_pg pgid;
672                 struct ceph_pg_mapping *pg;
673
674                 ceph_decode_need(p, end, sizeof(u32) + sizeof(u64), bad);
675                 ceph_decode_copy(p, &pgid, sizeof(pgid));
676                 n = ceph_decode_32(p);
677                 err = -EINVAL;
678                 if (n > (UINT_MAX - sizeof(*pg)) / sizeof(u32))
679                         goto bad;
680                 ceph_decode_need(p, end, n * sizeof(u32), bad);
681                 err = -ENOMEM;
682                 pg = kmalloc(sizeof(*pg) + n*sizeof(u32), GFP_NOFS);
683                 if (!pg)
684                         goto bad;
685                 pg->pgid = pgid;
686                 pg->len = n;
687                 for (j = 0; j < n; j++)
688                         pg->osds[j] = ceph_decode_32(p);
689
690                 err = __insert_pg_mapping(pg, &map->pg_temp);
691                 if (err)
692                         goto bad;
693                 dout(" added pg_temp %llx len %d\n", *(u64 *)&pgid, len);
694         }
695
696         /* crush */
697         ceph_decode_32_safe(p, end, len, bad);
698         dout("osdmap_decode crush len %d from off 0x%x\n", len,
699              (int)(*p - start));
700         ceph_decode_need(p, end, len, bad);
701         map->crush = crush_decode(*p, end);
702         *p += len;
703         if (IS_ERR(map->crush)) {
704                 err = PTR_ERR(map->crush);
705                 map->crush = NULL;
706                 goto bad;
707         }
708
709         /* ignore the rest of the map */
710         *p = end;
711
712         dout("osdmap_decode done %p %p\n", *p, end);
713         return map;
714
715 bad:
716         dout("osdmap_decode fail\n");
717         ceph_osdmap_destroy(map);
718         return ERR_PTR(err);
719 }
720
721 /*
722  * decode and apply an incremental map update.
723  */
724 struct ceph_osdmap *osdmap_apply_incremental(void **p, void *end,
725                                              struct ceph_osdmap *map,
726                                              struct ceph_messenger *msgr)
727 {
728         struct crush_map *newcrush = NULL;
729         struct ceph_fsid fsid;
730         u32 epoch = 0;
731         struct ceph_timespec modified;
732         u32 len, pool;
733         __s32 new_pool_max, new_flags, max;
734         void *start = *p;
735         int err = -EINVAL;
736         u16 version;
737
738         ceph_decode_16_safe(p, end, version, bad);
739         if (version > CEPH_OSDMAP_INC_VERSION) {
740                 pr_warning("got unknown v %d > %d of inc osdmap\n", version,
741                            CEPH_OSDMAP_INC_VERSION);
742                 goto bad;
743         }
744
745         ceph_decode_need(p, end, sizeof(fsid)+sizeof(modified)+2*sizeof(u32),
746                          bad);
747         ceph_decode_copy(p, &fsid, sizeof(fsid));
748         epoch = ceph_decode_32(p);
749         BUG_ON(epoch != map->epoch+1);
750         ceph_decode_copy(p, &modified, sizeof(modified));
751         new_pool_max = ceph_decode_32(p);
752         new_flags = ceph_decode_32(p);
753
754         /* full map? */
755         ceph_decode_32_safe(p, end, len, bad);
756         if (len > 0) {
757                 dout("apply_incremental full map len %d, %p to %p\n",
758                      len, *p, end);
759                 return osdmap_decode(p, min(*p+len, end));
760         }
761
762         /* new crush? */
763         ceph_decode_32_safe(p, end, len, bad);
764         if (len > 0) {
765                 dout("apply_incremental new crush map len %d, %p to %p\n",
766                      len, *p, end);
767                 newcrush = crush_decode(*p, min(*p+len, end));
768                 if (IS_ERR(newcrush))
769                         return ERR_CAST(newcrush);
770                 *p += len;
771         }
772
773         /* new flags? */
774         if (new_flags >= 0)
775                 map->flags = new_flags;
776         if (new_pool_max >= 0)
777                 map->pool_max = new_pool_max;
778
779         ceph_decode_need(p, end, 5*sizeof(u32), bad);
780
781         /* new max? */
782         max = ceph_decode_32(p);
783         if (max >= 0) {
784                 err = osdmap_set_max_osd(map, max);
785                 if (err < 0)
786                         goto bad;
787         }
788
789         map->epoch++;
790         map->modified = modified;
791         if (newcrush) {
792                 if (map->crush)
793                         crush_destroy(map->crush);
794                 map->crush = newcrush;
795                 newcrush = NULL;
796         }
797
798         /* new_pool */
799         ceph_decode_32_safe(p, end, len, bad);
800         while (len--) {
801                 __u8 ev;
802                 struct ceph_pg_pool_info *pi;
803
804                 ceph_decode_32_safe(p, end, pool, bad);
805                 ceph_decode_need(p, end, 1 + sizeof(pi->v), bad);
806                 ev = ceph_decode_8(p);  /* encoding version */
807                 if (ev > CEPH_PG_POOL_VERSION) {
808                         pr_warning("got unknown v %d > %d of ceph_pg_pool\n",
809                                    ev, CEPH_PG_POOL_VERSION);
810                         goto bad;
811                 }
812                 pi = __lookup_pg_pool(&map->pg_pools, pool);
813                 if (!pi) {
814                         pi = kzalloc(sizeof(*pi), GFP_NOFS);
815                         if (!pi) {
816                                 err = -ENOMEM;
817                                 goto bad;
818                         }
819                         pi->id = pool;
820                         __insert_pg_pool(&map->pg_pools, pi);
821                 }
822                 err = __decode_pool(p, end, pi);
823                 if (err < 0)
824                         goto bad;
825         }
826         if (version >= 5 && __decode_pool_names(p, end, map) < 0)
827                 goto bad;
828
829         /* old_pool */
830         ceph_decode_32_safe(p, end, len, bad);
831         while (len--) {
832                 struct ceph_pg_pool_info *pi;
833
834                 ceph_decode_32_safe(p, end, pool, bad);
835                 pi = __lookup_pg_pool(&map->pg_pools, pool);
836                 if (pi)
837                         __remove_pg_pool(&map->pg_pools, pi);
838         }
839
840         /* new_up */
841         err = -EINVAL;
842         ceph_decode_32_safe(p, end, len, bad);
843         while (len--) {
844                 u32 osd;
845                 struct ceph_entity_addr addr;
846                 ceph_decode_32_safe(p, end, osd, bad);
847                 ceph_decode_copy_safe(p, end, &addr, sizeof(addr), bad);
848                 ceph_decode_addr(&addr);
849                 pr_info("osd%d up\n", osd);
850                 BUG_ON(osd >= map->max_osd);
851                 map->osd_state[osd] |= CEPH_OSD_UP;
852                 map->osd_addr[osd] = addr;
853         }
854
855         /* new_state */
856         ceph_decode_32_safe(p, end, len, bad);
857         while (len--) {
858                 u32 osd;
859                 u8 xorstate;
860                 ceph_decode_32_safe(p, end, osd, bad);
861                 xorstate = **(u8 **)p;
862                 (*p)++;  /* clean flag */
863                 if (xorstate == 0)
864                         xorstate = CEPH_OSD_UP;
865                 if (xorstate & CEPH_OSD_UP)
866                         pr_info("osd%d down\n", osd);
867                 if (osd < map->max_osd)
868                         map->osd_state[osd] ^= xorstate;
869         }
870
871         /* new_weight */
872         ceph_decode_32_safe(p, end, len, bad);
873         while (len--) {
874                 u32 osd, off;
875                 ceph_decode_need(p, end, sizeof(u32)*2, bad);
876                 osd = ceph_decode_32(p);
877                 off = ceph_decode_32(p);
878                 pr_info("osd%d weight 0x%x %s\n", osd, off,
879                      off == CEPH_OSD_IN ? "(in)" :
880                      (off == CEPH_OSD_OUT ? "(out)" : ""));
881                 if (osd < map->max_osd)
882                         map->osd_weight[osd] = off;
883         }
884
885         /* new_pg_temp */
886         ceph_decode_32_safe(p, end, len, bad);
887         while (len--) {
888                 struct ceph_pg_mapping *pg;
889                 int j;
890                 struct ceph_pg pgid;
891                 u32 pglen;
892                 ceph_decode_need(p, end, sizeof(u64) + sizeof(u32), bad);
893                 ceph_decode_copy(p, &pgid, sizeof(pgid));
894                 pglen = ceph_decode_32(p);
895
896                 if (pglen) {
897                         ceph_decode_need(p, end, pglen*sizeof(u32), bad);
898
899                         /* removing existing (if any) */
900                         (void) __remove_pg_mapping(&map->pg_temp, pgid);
901
902                         /* insert */
903                         if (pglen > (UINT_MAX - sizeof(*pg)) / sizeof(u32)) {
904                                 err = -EINVAL;
905                                 goto bad;
906                         }
907                         pg = kmalloc(sizeof(*pg) + sizeof(u32)*pglen, GFP_NOFS);
908                         if (!pg) {
909                                 err = -ENOMEM;
910                                 goto bad;
911                         }
912                         pg->pgid = pgid;
913                         pg->len = pglen;
914                         for (j = 0; j < pglen; j++)
915                                 pg->osds[j] = ceph_decode_32(p);
916                         err = __insert_pg_mapping(pg, &map->pg_temp);
917                         if (err) {
918                                 kfree(pg);
919                                 goto bad;
920                         }
921                         dout(" added pg_temp %llx len %d\n", *(u64 *)&pgid,
922                              pglen);
923                 } else {
924                         /* remove */
925                         __remove_pg_mapping(&map->pg_temp, pgid);
926                 }
927         }
928
929         /* ignore the rest */
930         *p = end;
931         return map;
932
933 bad:
934         pr_err("corrupt inc osdmap epoch %d off %d (%p of %p-%p)\n",
935                epoch, (int)(*p - start), *p, start, end);
936         print_hex_dump(KERN_DEBUG, "osdmap: ",
937                        DUMP_PREFIX_OFFSET, 16, 1,
938                        start, end - start, true);
939         if (newcrush)
940                 crush_destroy(newcrush);
941         return ERR_PTR(err);
942 }
943
944
945
946
947 /*
948  * calculate file layout from given offset, length.
949  * fill in correct oid, logical length, and object extent
950  * offset, length.
951  *
952  * for now, we write only a single su, until we can
953  * pass a stride back to the caller.
954  */
955 int ceph_calc_file_object_mapping(struct ceph_file_layout *layout,
956                                    u64 off, u64 *plen,
957                                    u64 *ono,
958                                    u64 *oxoff, u64 *oxlen)
959 {
960         u32 osize = le32_to_cpu(layout->fl_object_size);
961         u32 su = le32_to_cpu(layout->fl_stripe_unit);
962         u32 sc = le32_to_cpu(layout->fl_stripe_count);
963         u32 bl, stripeno, stripepos, objsetno;
964         u32 su_per_object;
965         u64 t, su_offset;
966
967         dout("mapping %llu~%llu  osize %u fl_su %u\n", off, *plen,
968              osize, su);
969         if (su == 0 || sc == 0)
970                 goto invalid;
971         su_per_object = osize / su;
972         if (su_per_object == 0)
973                 goto invalid;
974         dout("osize %u / su %u = su_per_object %u\n", osize, su,
975              su_per_object);
976
977         if ((su & ~PAGE_MASK) != 0)
978                 goto invalid;
979
980         /* bl = *off / su; */
981         t = off;
982         do_div(t, su);
983         bl = t;
984         dout("off %llu / su %u = bl %u\n", off, su, bl);
985
986         stripeno = bl / sc;
987         stripepos = bl % sc;
988         objsetno = stripeno / su_per_object;
989
990         *ono = objsetno * sc + stripepos;
991         dout("objset %u * sc %u = ono %u\n", objsetno, sc, (unsigned)*ono);
992
993         /* *oxoff = *off % layout->fl_stripe_unit;  # offset in su */
994         t = off;
995         su_offset = do_div(t, su);
996         *oxoff = su_offset + (stripeno % su_per_object) * su;
997
998         /*
999          * Calculate the length of the extent being written to the selected
1000          * object. This is the minimum of the full length requested (plen) or
1001          * the remainder of the current stripe being written to.
1002          */
1003         *oxlen = min_t(u64, *plen, su - su_offset);
1004         *plen = *oxlen;
1005
1006         dout(" obj extent %llu~%llu\n", *oxoff, *oxlen);
1007         return 0;
1008
1009 invalid:
1010         dout(" invalid layout\n");
1011         *ono = 0;
1012         *oxoff = 0;
1013         *oxlen = 0;
1014         return -EINVAL;
1015 }
1016 EXPORT_SYMBOL(ceph_calc_file_object_mapping);
1017
1018 /*
1019  * calculate an object layout (i.e. pgid) from an oid,
1020  * file_layout, and osdmap
1021  */
1022 int ceph_calc_object_layout(struct ceph_object_layout *ol,
1023                             const char *oid,
1024                             struct ceph_file_layout *fl,
1025                             struct ceph_osdmap *osdmap)
1026 {
1027         unsigned num, num_mask;
1028         struct ceph_pg pgid;
1029         s32 preferred = (s32)le32_to_cpu(fl->fl_pg_preferred);
1030         int poolid = le32_to_cpu(fl->fl_pg_pool);
1031         struct ceph_pg_pool_info *pool;
1032         unsigned ps;
1033
1034         BUG_ON(!osdmap);
1035
1036         pool = __lookup_pg_pool(&osdmap->pg_pools, poolid);
1037         if (!pool)
1038                 return -EIO;
1039         ps = ceph_str_hash(pool->v.object_hash, oid, strlen(oid));
1040         if (preferred >= 0) {
1041                 ps += preferred;
1042                 num = le32_to_cpu(pool->v.lpg_num);
1043                 num_mask = pool->lpg_num_mask;
1044         } else {
1045                 num = le32_to_cpu(pool->v.pg_num);
1046                 num_mask = pool->pg_num_mask;
1047         }
1048
1049         pgid.ps = cpu_to_le16(ps);
1050         pgid.preferred = cpu_to_le16(preferred);
1051         pgid.pool = fl->fl_pg_pool;
1052         if (preferred >= 0)
1053                 dout("calc_object_layout '%s' pgid %d.%xp%d\n", oid, poolid, ps,
1054                      (int)preferred);
1055         else
1056                 dout("calc_object_layout '%s' pgid %d.%x\n", oid, poolid, ps);
1057
1058         ol->ol_pgid = pgid;
1059         ol->ol_stripe_unit = fl->fl_object_stripe_unit;
1060         return 0;
1061 }
1062 EXPORT_SYMBOL(ceph_calc_object_layout);
1063
1064 /*
1065  * Calculate raw osd vector for the given pgid.  Return pointer to osd
1066  * array, or NULL on failure.
1067  */
1068 static int *calc_pg_raw(struct ceph_osdmap *osdmap, struct ceph_pg pgid,
1069                         int *osds, int *num)
1070 {
1071         struct ceph_pg_mapping *pg;
1072         struct ceph_pg_pool_info *pool;
1073         int ruleno;
1074         unsigned poolid, ps, pps, t;
1075         int preferred;
1076
1077         poolid = le32_to_cpu(pgid.pool);
1078         ps = le16_to_cpu(pgid.ps);
1079         preferred = (s16)le16_to_cpu(pgid.preferred);
1080
1081         pool = __lookup_pg_pool(&osdmap->pg_pools, poolid);
1082         if (!pool)
1083                 return NULL;
1084
1085         /* pg_temp? */
1086         if (preferred >= 0)
1087                 t = ceph_stable_mod(ps, le32_to_cpu(pool->v.lpg_num),
1088                                     pool->lpgp_num_mask);
1089         else
1090                 t = ceph_stable_mod(ps, le32_to_cpu(pool->v.pg_num),
1091                                     pool->pgp_num_mask);
1092         pgid.ps = cpu_to_le16(t);
1093         pg = __lookup_pg_mapping(&osdmap->pg_temp, pgid);
1094         if (pg) {
1095                 *num = pg->len;
1096                 return pg->osds;
1097         }
1098
1099         /* crush */
1100         ruleno = crush_find_rule(osdmap->crush, pool->v.crush_ruleset,
1101                                  pool->v.type, pool->v.size);
1102         if (ruleno < 0) {
1103                 pr_err("no crush rule pool %d ruleset %d type %d size %d\n",
1104                        poolid, pool->v.crush_ruleset, pool->v.type,
1105                        pool->v.size);
1106                 return NULL;
1107         }
1108
1109         /* don't forcefeed bad device ids to crush */
1110         if (preferred >= osdmap->max_osd ||
1111             preferred >= osdmap->crush->max_devices)
1112                 preferred = -1;
1113
1114         if (preferred >= 0)
1115                 pps = ceph_stable_mod(ps,
1116                                       le32_to_cpu(pool->v.lpgp_num),
1117                                       pool->lpgp_num_mask);
1118         else
1119                 pps = ceph_stable_mod(ps,
1120                                       le32_to_cpu(pool->v.pgp_num),
1121                                       pool->pgp_num_mask);
1122         pps += poolid;
1123         *num = crush_do_rule(osdmap->crush, ruleno, pps, osds,
1124                              min_t(int, pool->v.size, *num),
1125                              preferred, osdmap->osd_weight);
1126         return osds;
1127 }
1128
1129 /*
1130  * Return acting set for given pgid.
1131  */
1132 int ceph_calc_pg_acting(struct ceph_osdmap *osdmap, struct ceph_pg pgid,
1133                         int *acting)
1134 {
1135         int rawosds[CEPH_PG_MAX_SIZE], *osds;
1136         int i, o, num = CEPH_PG_MAX_SIZE;
1137
1138         osds = calc_pg_raw(osdmap, pgid, rawosds, &num);
1139         if (!osds)
1140                 return -1;
1141
1142         /* primary is first up osd */
1143         o = 0;
1144         for (i = 0; i < num; i++)
1145                 if (ceph_osd_is_up(osdmap, osds[i]))
1146                         acting[o++] = osds[i];
1147         return o;
1148 }
1149
1150 /*
1151  * Return primary osd for given pgid, or -1 if none.
1152  */
1153 int ceph_calc_pg_primary(struct ceph_osdmap *osdmap, struct ceph_pg pgid)
1154 {
1155         int rawosds[CEPH_PG_MAX_SIZE], *osds;
1156         int i, num = CEPH_PG_MAX_SIZE;
1157
1158         osds = calc_pg_raw(osdmap, pgid, rawosds, &num);
1159         if (!osds)
1160                 return -1;
1161
1162         /* primary is first up osd */
1163         for (i = 0; i < num; i++)
1164                 if (ceph_osd_is_up(osdmap, osds[i]))
1165                         return osds[i];
1166         return -1;
1167 }
1168 EXPORT_SYMBOL(ceph_calc_pg_primary);