Merge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/sage/ceph...
[linux-2.6.git] / fs / ceph / osdmap.c
1
2 #include <asm/div64.h>
3
4 #include "super.h"
5 #include "osdmap.h"
6 #include "crush/hash.h"
7 #include "crush/mapper.h"
8 #include "decode.h"
9 #include "ceph_debug.h"
10
11 char *ceph_osdmap_state_str(char *str, int len, int state)
12 {
13         int flag = 0;
14
15         if (!len)
16                 goto done;
17
18         *str = '\0';
19         if (state) {
20                 if (state & CEPH_OSD_EXISTS) {
21                         snprintf(str, len, "exists");
22                         flag = 1;
23                 }
24                 if (state & CEPH_OSD_UP) {
25                         snprintf(str, len, "%s%s%s", str, (flag ? ", " : ""),
26                                  "up");
27                         flag = 1;
28                 }
29         } else {
30                 snprintf(str, len, "doesn't exist");
31         }
32 done:
33         return str;
34 }
35
36 /* maps */
37
38 static int calc_bits_of(unsigned t)
39 {
40         int b = 0;
41         while (t) {
42                 t = t >> 1;
43                 b++;
44         }
45         return b;
46 }
47
48 /*
49  * the foo_mask is the smallest value 2^n-1 that is >= foo.
50  */
51 static void calc_pg_masks(struct ceph_pg_pool_info *pi)
52 {
53         pi->pg_num_mask = (1 << calc_bits_of(le32_to_cpu(pi->v.pg_num)-1)) - 1;
54         pi->pgp_num_mask =
55                 (1 << calc_bits_of(le32_to_cpu(pi->v.pgp_num)-1)) - 1;
56         pi->lpg_num_mask =
57                 (1 << calc_bits_of(le32_to_cpu(pi->v.lpg_num)-1)) - 1;
58         pi->lpgp_num_mask =
59                 (1 << calc_bits_of(le32_to_cpu(pi->v.lpgp_num)-1)) - 1;
60 }
61
62 /*
63  * decode crush map
64  */
65 static int crush_decode_uniform_bucket(void **p, void *end,
66                                        struct crush_bucket_uniform *b)
67 {
68         dout("crush_decode_uniform_bucket %p to %p\n", *p, end);
69         ceph_decode_need(p, end, (1+b->h.size) * sizeof(u32), bad);
70         b->item_weight = ceph_decode_32(p);
71         return 0;
72 bad:
73         return -EINVAL;
74 }
75
76 static int crush_decode_list_bucket(void **p, void *end,
77                                     struct crush_bucket_list *b)
78 {
79         int j;
80         dout("crush_decode_list_bucket %p to %p\n", *p, end);
81         b->item_weights = kcalloc(b->h.size, sizeof(u32), GFP_NOFS);
82         if (b->item_weights == NULL)
83                 return -ENOMEM;
84         b->sum_weights = kcalloc(b->h.size, sizeof(u32), GFP_NOFS);
85         if (b->sum_weights == NULL)
86                 return -ENOMEM;
87         ceph_decode_need(p, end, 2 * b->h.size * sizeof(u32), bad);
88         for (j = 0; j < b->h.size; j++) {
89                 b->item_weights[j] = ceph_decode_32(p);
90                 b->sum_weights[j] = ceph_decode_32(p);
91         }
92         return 0;
93 bad:
94         return -EINVAL;
95 }
96
97 static int crush_decode_tree_bucket(void **p, void *end,
98                                     struct crush_bucket_tree *b)
99 {
100         int j;
101         dout("crush_decode_tree_bucket %p to %p\n", *p, end);
102         ceph_decode_32_safe(p, end, b->num_nodes, bad);
103         b->node_weights = kcalloc(b->num_nodes, sizeof(u32), GFP_NOFS);
104         if (b->node_weights == NULL)
105                 return -ENOMEM;
106         ceph_decode_need(p, end, b->num_nodes * sizeof(u32), bad);
107         for (j = 0; j < b->num_nodes; j++)
108                 b->node_weights[j] = ceph_decode_32(p);
109         return 0;
110 bad:
111         return -EINVAL;
112 }
113
114 static int crush_decode_straw_bucket(void **p, void *end,
115                                      struct crush_bucket_straw *b)
116 {
117         int j;
118         dout("crush_decode_straw_bucket %p to %p\n", *p, end);
119         b->item_weights = kcalloc(b->h.size, sizeof(u32), GFP_NOFS);
120         if (b->item_weights == NULL)
121                 return -ENOMEM;
122         b->straws = kcalloc(b->h.size, sizeof(u32), GFP_NOFS);
123         if (b->straws == NULL)
124                 return -ENOMEM;
125         ceph_decode_need(p, end, 2 * b->h.size * sizeof(u32), bad);
126         for (j = 0; j < b->h.size; j++) {
127                 b->item_weights[j] = ceph_decode_32(p);
128                 b->straws[j] = ceph_decode_32(p);
129         }
130         return 0;
131 bad:
132         return -EINVAL;
133 }
134
135 static struct crush_map *crush_decode(void *pbyval, void *end)
136 {
137         struct crush_map *c;
138         int err = -EINVAL;
139         int i, j;
140         void **p = &pbyval;
141         void *start = pbyval;
142         u32 magic;
143
144         dout("crush_decode %p to %p len %d\n", *p, end, (int)(end - *p));
145
146         c = kzalloc(sizeof(*c), GFP_NOFS);
147         if (c == NULL)
148                 return ERR_PTR(-ENOMEM);
149
150         ceph_decode_need(p, end, 4*sizeof(u32), bad);
151         magic = ceph_decode_32(p);
152         if (magic != CRUSH_MAGIC) {
153                 pr_err("crush_decode magic %x != current %x\n",
154                        (unsigned)magic, (unsigned)CRUSH_MAGIC);
155                 goto bad;
156         }
157         c->max_buckets = ceph_decode_32(p);
158         c->max_rules = ceph_decode_32(p);
159         c->max_devices = ceph_decode_32(p);
160
161         c->device_parents = kcalloc(c->max_devices, sizeof(u32), GFP_NOFS);
162         if (c->device_parents == NULL)
163                 goto badmem;
164         c->bucket_parents = kcalloc(c->max_buckets, sizeof(u32), GFP_NOFS);
165         if (c->bucket_parents == NULL)
166                 goto badmem;
167
168         c->buckets = kcalloc(c->max_buckets, sizeof(*c->buckets), GFP_NOFS);
169         if (c->buckets == NULL)
170                 goto badmem;
171         c->rules = kcalloc(c->max_rules, sizeof(*c->rules), GFP_NOFS);
172         if (c->rules == NULL)
173                 goto badmem;
174
175         /* buckets */
176         for (i = 0; i < c->max_buckets; i++) {
177                 int size = 0;
178                 u32 alg;
179                 struct crush_bucket *b;
180
181                 ceph_decode_32_safe(p, end, alg, bad);
182                 if (alg == 0) {
183                         c->buckets[i] = NULL;
184                         continue;
185                 }
186                 dout("crush_decode bucket %d off %x %p to %p\n",
187                      i, (int)(*p-start), *p, end);
188
189                 switch (alg) {
190                 case CRUSH_BUCKET_UNIFORM:
191                         size = sizeof(struct crush_bucket_uniform);
192                         break;
193                 case CRUSH_BUCKET_LIST:
194                         size = sizeof(struct crush_bucket_list);
195                         break;
196                 case CRUSH_BUCKET_TREE:
197                         size = sizeof(struct crush_bucket_tree);
198                         break;
199                 case CRUSH_BUCKET_STRAW:
200                         size = sizeof(struct crush_bucket_straw);
201                         break;
202                 default:
203                         err = -EINVAL;
204                         goto bad;
205                 }
206                 BUG_ON(size == 0);
207                 b = c->buckets[i] = kzalloc(size, GFP_NOFS);
208                 if (b == NULL)
209                         goto badmem;
210
211                 ceph_decode_need(p, end, 4*sizeof(u32), bad);
212                 b->id = ceph_decode_32(p);
213                 b->type = ceph_decode_16(p);
214                 b->alg = ceph_decode_8(p);
215                 b->hash = ceph_decode_8(p);
216                 b->weight = ceph_decode_32(p);
217                 b->size = ceph_decode_32(p);
218
219                 dout("crush_decode bucket size %d off %x %p to %p\n",
220                      b->size, (int)(*p-start), *p, end);
221
222                 b->items = kcalloc(b->size, sizeof(__s32), GFP_NOFS);
223                 if (b->items == NULL)
224                         goto badmem;
225                 b->perm = kcalloc(b->size, sizeof(u32), GFP_NOFS);
226                 if (b->perm == NULL)
227                         goto badmem;
228                 b->perm_n = 0;
229
230                 ceph_decode_need(p, end, b->size*sizeof(u32), bad);
231                 for (j = 0; j < b->size; j++)
232                         b->items[j] = ceph_decode_32(p);
233
234                 switch (b->alg) {
235                 case CRUSH_BUCKET_UNIFORM:
236                         err = crush_decode_uniform_bucket(p, end,
237                                   (struct crush_bucket_uniform *)b);
238                         if (err < 0)
239                                 goto bad;
240                         break;
241                 case CRUSH_BUCKET_LIST:
242                         err = crush_decode_list_bucket(p, end,
243                                (struct crush_bucket_list *)b);
244                         if (err < 0)
245                                 goto bad;
246                         break;
247                 case CRUSH_BUCKET_TREE:
248                         err = crush_decode_tree_bucket(p, end,
249                                 (struct crush_bucket_tree *)b);
250                         if (err < 0)
251                                 goto bad;
252                         break;
253                 case CRUSH_BUCKET_STRAW:
254                         err = crush_decode_straw_bucket(p, end,
255                                 (struct crush_bucket_straw *)b);
256                         if (err < 0)
257                                 goto bad;
258                         break;
259                 }
260         }
261
262         /* rules */
263         dout("rule vec is %p\n", c->rules);
264         for (i = 0; i < c->max_rules; i++) {
265                 u32 yes;
266                 struct crush_rule *r;
267
268                 ceph_decode_32_safe(p, end, yes, bad);
269                 if (!yes) {
270                         dout("crush_decode NO rule %d off %x %p to %p\n",
271                              i, (int)(*p-start), *p, end);
272                         c->rules[i] = NULL;
273                         continue;
274                 }
275
276                 dout("crush_decode rule %d off %x %p to %p\n",
277                      i, (int)(*p-start), *p, end);
278
279                 /* len */
280                 ceph_decode_32_safe(p, end, yes, bad);
281 #if BITS_PER_LONG == 32
282                 err = -EINVAL;
283                 if (yes > ULONG_MAX / sizeof(struct crush_rule_step))
284                         goto bad;
285 #endif
286                 r = c->rules[i] = kmalloc(sizeof(*r) +
287                                           yes*sizeof(struct crush_rule_step),
288                                           GFP_NOFS);
289                 if (r == NULL)
290                         goto badmem;
291                 dout(" rule %d is at %p\n", i, r);
292                 r->len = yes;
293                 ceph_decode_copy_safe(p, end, &r->mask, 4, bad); /* 4 u8's */
294                 ceph_decode_need(p, end, r->len*3*sizeof(u32), bad);
295                 for (j = 0; j < r->len; j++) {
296                         r->steps[j].op = ceph_decode_32(p);
297                         r->steps[j].arg1 = ceph_decode_32(p);
298                         r->steps[j].arg2 = ceph_decode_32(p);
299                 }
300         }
301
302         /* ignore trailing name maps. */
303
304         dout("crush_decode success\n");
305         return c;
306
307 badmem:
308         err = -ENOMEM;
309 bad:
310         dout("crush_decode fail %d\n", err);
311         crush_destroy(c);
312         return ERR_PTR(err);
313 }
314
315
316 /*
317  * osd map
318  */
319 void ceph_osdmap_destroy(struct ceph_osdmap *map)
320 {
321         dout("osdmap_destroy %p\n", map);
322         if (map->crush)
323                 crush_destroy(map->crush);
324         while (!RB_EMPTY_ROOT(&map->pg_temp)) {
325                 struct ceph_pg_mapping *pg =
326                         rb_entry(rb_first(&map->pg_temp),
327                                  struct ceph_pg_mapping, node);
328                 rb_erase(&pg->node, &map->pg_temp);
329                 kfree(pg);
330         }
331         while (!RB_EMPTY_ROOT(&map->pg_pools)) {
332                 struct ceph_pg_pool_info *pi =
333                         rb_entry(rb_first(&map->pg_pools),
334                                  struct ceph_pg_pool_info, node);
335                 rb_erase(&pi->node, &map->pg_pools);
336                 kfree(pi);
337         }
338         kfree(map->osd_state);
339         kfree(map->osd_weight);
340         kfree(map->osd_addr);
341         kfree(map);
342 }
343
344 /*
345  * adjust max osd value.  reallocate arrays.
346  */
347 static int osdmap_set_max_osd(struct ceph_osdmap *map, int max)
348 {
349         u8 *state;
350         struct ceph_entity_addr *addr;
351         u32 *weight;
352
353         state = kcalloc(max, sizeof(*state), GFP_NOFS);
354         addr = kcalloc(max, sizeof(*addr), GFP_NOFS);
355         weight = kcalloc(max, sizeof(*weight), GFP_NOFS);
356         if (state == NULL || addr == NULL || weight == NULL) {
357                 kfree(state);
358                 kfree(addr);
359                 kfree(weight);
360                 return -ENOMEM;
361         }
362
363         /* copy old? */
364         if (map->osd_state) {
365                 memcpy(state, map->osd_state, map->max_osd*sizeof(*state));
366                 memcpy(addr, map->osd_addr, map->max_osd*sizeof(*addr));
367                 memcpy(weight, map->osd_weight, map->max_osd*sizeof(*weight));
368                 kfree(map->osd_state);
369                 kfree(map->osd_addr);
370                 kfree(map->osd_weight);
371         }
372
373         map->osd_state = state;
374         map->osd_weight = weight;
375         map->osd_addr = addr;
376         map->max_osd = max;
377         return 0;
378 }
379
380 /*
381  * rbtree of pg_mapping for handling pg_temp (explicit mapping of pgid
382  * to a set of osds)
383  */
384 static int pgid_cmp(struct ceph_pg l, struct ceph_pg r)
385 {
386         u64 a = *(u64 *)&l;
387         u64 b = *(u64 *)&r;
388
389         if (a < b)
390                 return -1;
391         if (a > b)
392                 return 1;
393         return 0;
394 }
395
396 static int __insert_pg_mapping(struct ceph_pg_mapping *new,
397                                struct rb_root *root)
398 {
399         struct rb_node **p = &root->rb_node;
400         struct rb_node *parent = NULL;
401         struct ceph_pg_mapping *pg = NULL;
402         int c;
403
404         while (*p) {
405                 parent = *p;
406                 pg = rb_entry(parent, struct ceph_pg_mapping, node);
407                 c = pgid_cmp(new->pgid, pg->pgid);
408                 if (c < 0)
409                         p = &(*p)->rb_left;
410                 else if (c > 0)
411                         p = &(*p)->rb_right;
412                 else
413                         return -EEXIST;
414         }
415
416         rb_link_node(&new->node, parent, p);
417         rb_insert_color(&new->node, root);
418         return 0;
419 }
420
421 static struct ceph_pg_mapping *__lookup_pg_mapping(struct rb_root *root,
422                                                    struct ceph_pg pgid)
423 {
424         struct rb_node *n = root->rb_node;
425         struct ceph_pg_mapping *pg;
426         int c;
427
428         while (n) {
429                 pg = rb_entry(n, struct ceph_pg_mapping, node);
430                 c = pgid_cmp(pgid, pg->pgid);
431                 if (c < 0)
432                         n = n->rb_left;
433                 else if (c > 0)
434                         n = n->rb_right;
435                 else
436                         return pg;
437         }
438         return NULL;
439 }
440
441 /*
442  * rbtree of pg pool info
443  */
444 static int __insert_pg_pool(struct rb_root *root, struct ceph_pg_pool_info *new)
445 {
446         struct rb_node **p = &root->rb_node;
447         struct rb_node *parent = NULL;
448         struct ceph_pg_pool_info *pi = NULL;
449
450         while (*p) {
451                 parent = *p;
452                 pi = rb_entry(parent, struct ceph_pg_pool_info, node);
453                 if (new->id < pi->id)
454                         p = &(*p)->rb_left;
455                 else if (new->id > pi->id)
456                         p = &(*p)->rb_right;
457                 else
458                         return -EEXIST;
459         }
460
461         rb_link_node(&new->node, parent, p);
462         rb_insert_color(&new->node, root);
463         return 0;
464 }
465
466 static struct ceph_pg_pool_info *__lookup_pg_pool(struct rb_root *root, int id)
467 {
468         struct ceph_pg_pool_info *pi;
469         struct rb_node *n = root->rb_node;
470
471         while (n) {
472                 pi = rb_entry(n, struct ceph_pg_pool_info, node);
473                 if (id < pi->id)
474                         n = n->rb_left;
475                 else if (id > pi->id)
476                         n = n->rb_right;
477                 else
478                         return pi;
479         }
480         return NULL;
481 }
482
483 void __decode_pool(void **p, struct ceph_pg_pool_info *pi)
484 {
485         ceph_decode_copy(p, &pi->v, sizeof(pi->v));
486         calc_pg_masks(pi);
487         *p += le32_to_cpu(pi->v.num_snaps) * sizeof(u64);
488         *p += le32_to_cpu(pi->v.num_removed_snap_intervals) * sizeof(u64) * 2;
489 }
490
491 /*
492  * decode a full map.
493  */
494 struct ceph_osdmap *osdmap_decode(void **p, void *end)
495 {
496         struct ceph_osdmap *map;
497         u16 version;
498         u32 len, max, i;
499         u8 ev;
500         int err = -EINVAL;
501         void *start = *p;
502         struct ceph_pg_pool_info *pi;
503
504         dout("osdmap_decode %p to %p len %d\n", *p, end, (int)(end - *p));
505
506         map = kzalloc(sizeof(*map), GFP_NOFS);
507         if (map == NULL)
508                 return ERR_PTR(-ENOMEM);
509         map->pg_temp = RB_ROOT;
510
511         ceph_decode_16_safe(p, end, version, bad);
512         if (version > CEPH_OSDMAP_VERSION) {
513                 pr_warning("got unknown v %d > %d of osdmap\n", version,
514                            CEPH_OSDMAP_VERSION);
515                 goto bad;
516         }
517
518         ceph_decode_need(p, end, 2*sizeof(u64)+6*sizeof(u32), bad);
519         ceph_decode_copy(p, &map->fsid, sizeof(map->fsid));
520         map->epoch = ceph_decode_32(p);
521         ceph_decode_copy(p, &map->created, sizeof(map->created));
522         ceph_decode_copy(p, &map->modified, sizeof(map->modified));
523
524         ceph_decode_32_safe(p, end, max, bad);
525         while (max--) {
526                 ceph_decode_need(p, end, 4 + 1 + sizeof(pi->v), bad);
527                 pi = kmalloc(sizeof(*pi), GFP_NOFS);
528                 if (!pi)
529                         goto bad;
530                 pi->id = ceph_decode_32(p);
531                 ev = ceph_decode_8(p); /* encoding version */
532                 if (ev > CEPH_PG_POOL_VERSION) {
533                         pr_warning("got unknown v %d > %d of ceph_pg_pool\n",
534                                    ev, CEPH_PG_POOL_VERSION);
535                         goto bad;
536                 }
537                 __decode_pool(p, pi);
538                 __insert_pg_pool(&map->pg_pools, pi);
539         }
540         ceph_decode_32_safe(p, end, map->pool_max, bad);
541
542         ceph_decode_32_safe(p, end, map->flags, bad);
543
544         max = ceph_decode_32(p);
545
546         /* (re)alloc osd arrays */
547         err = osdmap_set_max_osd(map, max);
548         if (err < 0)
549                 goto bad;
550         dout("osdmap_decode max_osd = %d\n", map->max_osd);
551
552         /* osds */
553         err = -EINVAL;
554         ceph_decode_need(p, end, 3*sizeof(u32) +
555                          map->max_osd*(1 + sizeof(*map->osd_weight) +
556                                        sizeof(*map->osd_addr)), bad);
557         *p += 4; /* skip length field (should match max) */
558         ceph_decode_copy(p, map->osd_state, map->max_osd);
559
560         *p += 4; /* skip length field (should match max) */
561         for (i = 0; i < map->max_osd; i++)
562                 map->osd_weight[i] = ceph_decode_32(p);
563
564         *p += 4; /* skip length field (should match max) */
565         ceph_decode_copy(p, map->osd_addr, map->max_osd*sizeof(*map->osd_addr));
566         for (i = 0; i < map->max_osd; i++)
567                 ceph_decode_addr(&map->osd_addr[i]);
568
569         /* pg_temp */
570         ceph_decode_32_safe(p, end, len, bad);
571         for (i = 0; i < len; i++) {
572                 int n, j;
573                 struct ceph_pg pgid;
574                 struct ceph_pg_mapping *pg;
575
576                 ceph_decode_need(p, end, sizeof(u32) + sizeof(u64), bad);
577                 ceph_decode_copy(p, &pgid, sizeof(pgid));
578                 n = ceph_decode_32(p);
579                 ceph_decode_need(p, end, n * sizeof(u32), bad);
580                 err = -ENOMEM;
581                 pg = kmalloc(sizeof(*pg) + n*sizeof(u32), GFP_NOFS);
582                 if (!pg)
583                         goto bad;
584                 pg->pgid = pgid;
585                 pg->len = n;
586                 for (j = 0; j < n; j++)
587                         pg->osds[j] = ceph_decode_32(p);
588
589                 err = __insert_pg_mapping(pg, &map->pg_temp);
590                 if (err)
591                         goto bad;
592                 dout(" added pg_temp %llx len %d\n", *(u64 *)&pgid, len);
593         }
594
595         /* crush */
596         ceph_decode_32_safe(p, end, len, bad);
597         dout("osdmap_decode crush len %d from off 0x%x\n", len,
598              (int)(*p - start));
599         ceph_decode_need(p, end, len, bad);
600         map->crush = crush_decode(*p, end);
601         *p += len;
602         if (IS_ERR(map->crush)) {
603                 err = PTR_ERR(map->crush);
604                 map->crush = NULL;
605                 goto bad;
606         }
607
608         /* ignore the rest of the map */
609         *p = end;
610
611         dout("osdmap_decode done %p %p\n", *p, end);
612         return map;
613
614 bad:
615         dout("osdmap_decode fail\n");
616         ceph_osdmap_destroy(map);
617         return ERR_PTR(err);
618 }
619
620 /*
621  * decode and apply an incremental map update.
622  */
623 struct ceph_osdmap *osdmap_apply_incremental(void **p, void *end,
624                                              struct ceph_osdmap *map,
625                                              struct ceph_messenger *msgr)
626 {
627         struct crush_map *newcrush = NULL;
628         struct ceph_fsid fsid;
629         u32 epoch = 0;
630         struct ceph_timespec modified;
631         u32 len, pool;
632         __s32 new_pool_max, new_flags, max;
633         void *start = *p;
634         int err = -EINVAL;
635         u16 version;
636         struct rb_node *rbp;
637
638         ceph_decode_16_safe(p, end, version, bad);
639         if (version > CEPH_OSDMAP_INC_VERSION) {
640                 pr_warning("got unknown v %d > %d of inc osdmap\n", version,
641                            CEPH_OSDMAP_INC_VERSION);
642                 goto bad;
643         }
644
645         ceph_decode_need(p, end, sizeof(fsid)+sizeof(modified)+2*sizeof(u32),
646                          bad);
647         ceph_decode_copy(p, &fsid, sizeof(fsid));
648         epoch = ceph_decode_32(p);
649         BUG_ON(epoch != map->epoch+1);
650         ceph_decode_copy(p, &modified, sizeof(modified));
651         new_pool_max = ceph_decode_32(p);
652         new_flags = ceph_decode_32(p);
653
654         /* full map? */
655         ceph_decode_32_safe(p, end, len, bad);
656         if (len > 0) {
657                 dout("apply_incremental full map len %d, %p to %p\n",
658                      len, *p, end);
659                 return osdmap_decode(p, min(*p+len, end));
660         }
661
662         /* new crush? */
663         ceph_decode_32_safe(p, end, len, bad);
664         if (len > 0) {
665                 dout("apply_incremental new crush map len %d, %p to %p\n",
666                      len, *p, end);
667                 newcrush = crush_decode(*p, min(*p+len, end));
668                 if (IS_ERR(newcrush))
669                         return ERR_PTR(PTR_ERR(newcrush));
670         }
671
672         /* new flags? */
673         if (new_flags >= 0)
674                 map->flags = new_flags;
675         if (new_pool_max >= 0)
676                 map->pool_max = new_pool_max;
677
678         ceph_decode_need(p, end, 5*sizeof(u32), bad);
679
680         /* new max? */
681         max = ceph_decode_32(p);
682         if (max >= 0) {
683                 err = osdmap_set_max_osd(map, max);
684                 if (err < 0)
685                         goto bad;
686         }
687
688         map->epoch++;
689         map->modified = map->modified;
690         if (newcrush) {
691                 if (map->crush)
692                         crush_destroy(map->crush);
693                 map->crush = newcrush;
694                 newcrush = NULL;
695         }
696
697         /* new_pool */
698         ceph_decode_32_safe(p, end, len, bad);
699         while (len--) {
700                 __u8 ev;
701                 struct ceph_pg_pool_info *pi;
702
703                 ceph_decode_32_safe(p, end, pool, bad);
704                 ceph_decode_need(p, end, 1 + sizeof(pi->v), bad);
705                 ev = ceph_decode_8(p);  /* encoding version */
706                 if (ev > CEPH_PG_POOL_VERSION) {
707                         pr_warning("got unknown v %d > %d of ceph_pg_pool\n",
708                                    ev, CEPH_PG_POOL_VERSION);
709                         goto bad;
710                 }
711                 pi = __lookup_pg_pool(&map->pg_pools, pool);
712                 if (!pi) {
713                         pi = kmalloc(sizeof(*pi), GFP_NOFS);
714                         if (!pi) {
715                                 err = -ENOMEM;
716                                 goto bad;
717                         }
718                         pi->id = pool;
719                         __insert_pg_pool(&map->pg_pools, pi);
720                 }
721                 __decode_pool(p, pi);
722         }
723
724         /* old_pool */
725         ceph_decode_32_safe(p, end, len, bad);
726         while (len--) {
727                 struct ceph_pg_pool_info *pi;
728
729                 ceph_decode_32_safe(p, end, pool, bad);
730                 pi = __lookup_pg_pool(&map->pg_pools, pool);
731                 if (pi) {
732                         rb_erase(&pi->node, &map->pg_pools);
733                         kfree(pi);
734                 }
735         }
736
737         /* new_up */
738         err = -EINVAL;
739         ceph_decode_32_safe(p, end, len, bad);
740         while (len--) {
741                 u32 osd;
742                 struct ceph_entity_addr addr;
743                 ceph_decode_32_safe(p, end, osd, bad);
744                 ceph_decode_copy_safe(p, end, &addr, sizeof(addr), bad);
745                 ceph_decode_addr(&addr);
746                 pr_info("osd%d up\n", osd);
747                 BUG_ON(osd >= map->max_osd);
748                 map->osd_state[osd] |= CEPH_OSD_UP;
749                 map->osd_addr[osd] = addr;
750         }
751
752         /* new_down */
753         ceph_decode_32_safe(p, end, len, bad);
754         while (len--) {
755                 u32 osd;
756                 ceph_decode_32_safe(p, end, osd, bad);
757                 (*p)++;  /* clean flag */
758                 pr_info("osd%d down\n", osd);
759                 if (osd < map->max_osd)
760                         map->osd_state[osd] &= ~CEPH_OSD_UP;
761         }
762
763         /* new_weight */
764         ceph_decode_32_safe(p, end, len, bad);
765         while (len--) {
766                 u32 osd, off;
767                 ceph_decode_need(p, end, sizeof(u32)*2, bad);
768                 osd = ceph_decode_32(p);
769                 off = ceph_decode_32(p);
770                 pr_info("osd%d weight 0x%x %s\n", osd, off,
771                      off == CEPH_OSD_IN ? "(in)" :
772                      (off == CEPH_OSD_OUT ? "(out)" : ""));
773                 if (osd < map->max_osd)
774                         map->osd_weight[osd] = off;
775         }
776
777         /* new_pg_temp */
778         rbp = rb_first(&map->pg_temp);
779         ceph_decode_32_safe(p, end, len, bad);
780         while (len--) {
781                 struct ceph_pg_mapping *pg;
782                 int j;
783                 struct ceph_pg pgid;
784                 u32 pglen;
785                 ceph_decode_need(p, end, sizeof(u64) + sizeof(u32), bad);
786                 ceph_decode_copy(p, &pgid, sizeof(pgid));
787                 pglen = ceph_decode_32(p);
788
789                 /* remove any? */
790                 while (rbp && pgid_cmp(rb_entry(rbp, struct ceph_pg_mapping,
791                                                 node)->pgid, pgid) <= 0) {
792                         struct rb_node *cur = rbp;
793                         rbp = rb_next(rbp);
794                         dout(" removed pg_temp %llx\n",
795                              *(u64 *)&rb_entry(cur, struct ceph_pg_mapping,
796                                                node)->pgid);
797                         rb_erase(cur, &map->pg_temp);
798                 }
799
800                 if (pglen) {
801                         /* insert */
802                         ceph_decode_need(p, end, pglen*sizeof(u32), bad);
803                         pg = kmalloc(sizeof(*pg) + sizeof(u32)*pglen, GFP_NOFS);
804                         if (!pg) {
805                                 err = -ENOMEM;
806                                 goto bad;
807                         }
808                         pg->pgid = pgid;
809                         pg->len = pglen;
810                         for (j = 0; j < pglen; j++)
811                                 pg->osds[j] = ceph_decode_32(p);
812                         err = __insert_pg_mapping(pg, &map->pg_temp);
813                         if (err)
814                                 goto bad;
815                         dout(" added pg_temp %llx len %d\n", *(u64 *)&pgid,
816                              pglen);
817                 }
818         }
819         while (rbp) {
820                 struct rb_node *cur = rbp;
821                 rbp = rb_next(rbp);
822                 dout(" removed pg_temp %llx\n",
823                      *(u64 *)&rb_entry(cur, struct ceph_pg_mapping,
824                                        node)->pgid);
825                 rb_erase(cur, &map->pg_temp);
826         }
827
828         /* ignore the rest */
829         *p = end;
830         return map;
831
832 bad:
833         pr_err("corrupt inc osdmap epoch %d off %d (%p of %p-%p)\n",
834                epoch, (int)(*p - start), *p, start, end);
835         print_hex_dump(KERN_DEBUG, "osdmap: ",
836                        DUMP_PREFIX_OFFSET, 16, 1,
837                        start, end - start, true);
838         if (newcrush)
839                 crush_destroy(newcrush);
840         return ERR_PTR(err);
841 }
842
843
844
845
846 /*
847  * calculate file layout from given offset, length.
848  * fill in correct oid, logical length, and object extent
849  * offset, length.
850  *
851  * for now, we write only a single su, until we can
852  * pass a stride back to the caller.
853  */
854 void ceph_calc_file_object_mapping(struct ceph_file_layout *layout,
855                                    u64 off, u64 *plen,
856                                    u64 *ono,
857                                    u64 *oxoff, u64 *oxlen)
858 {
859         u32 osize = le32_to_cpu(layout->fl_object_size);
860         u32 su = le32_to_cpu(layout->fl_stripe_unit);
861         u32 sc = le32_to_cpu(layout->fl_stripe_count);
862         u32 bl, stripeno, stripepos, objsetno;
863         u32 su_per_object;
864         u64 t, su_offset;
865
866         dout("mapping %llu~%llu  osize %u fl_su %u\n", off, *plen,
867              osize, su);
868         su_per_object = osize / su;
869         dout("osize %u / su %u = su_per_object %u\n", osize, su,
870              su_per_object);
871
872         BUG_ON((su & ~PAGE_MASK) != 0);
873         /* bl = *off / su; */
874         t = off;
875         do_div(t, su);
876         bl = t;
877         dout("off %llu / su %u = bl %u\n", off, su, bl);
878
879         stripeno = bl / sc;
880         stripepos = bl % sc;
881         objsetno = stripeno / su_per_object;
882
883         *ono = objsetno * sc + stripepos;
884         dout("objset %u * sc %u = ono %u\n", objsetno, sc, (unsigned)*ono);
885
886         /* *oxoff = *off % layout->fl_stripe_unit;  # offset in su */
887         t = off;
888         su_offset = do_div(t, su);
889         *oxoff = su_offset + (stripeno % su_per_object) * su;
890
891         /*
892          * Calculate the length of the extent being written to the selected
893          * object. This is the minimum of the full length requested (plen) or
894          * the remainder of the current stripe being written to.
895          */
896         *oxlen = min_t(u64, *plen, su - su_offset);
897         *plen = *oxlen;
898
899         dout(" obj extent %llu~%llu\n", *oxoff, *oxlen);
900 }
901
902 /*
903  * calculate an object layout (i.e. pgid) from an oid,
904  * file_layout, and osdmap
905  */
906 int ceph_calc_object_layout(struct ceph_object_layout *ol,
907                             const char *oid,
908                             struct ceph_file_layout *fl,
909                             struct ceph_osdmap *osdmap)
910 {
911         unsigned num, num_mask;
912         struct ceph_pg pgid;
913         s32 preferred = (s32)le32_to_cpu(fl->fl_pg_preferred);
914         int poolid = le32_to_cpu(fl->fl_pg_pool);
915         struct ceph_pg_pool_info *pool;
916         unsigned ps;
917
918         BUG_ON(!osdmap);
919
920         pool = __lookup_pg_pool(&osdmap->pg_pools, poolid);
921         if (!pool)
922                 return -EIO;
923         ps = ceph_str_hash(pool->v.object_hash, oid, strlen(oid));
924         if (preferred >= 0) {
925                 ps += preferred;
926                 num = le32_to_cpu(pool->v.lpg_num);
927                 num_mask = pool->lpg_num_mask;
928         } else {
929                 num = le32_to_cpu(pool->v.pg_num);
930                 num_mask = pool->pg_num_mask;
931         }
932
933         pgid.ps = cpu_to_le16(ps);
934         pgid.preferred = cpu_to_le16(preferred);
935         pgid.pool = fl->fl_pg_pool;
936         if (preferred >= 0)
937                 dout("calc_object_layout '%s' pgid %d.%xp%d\n", oid, poolid, ps,
938                      (int)preferred);
939         else
940                 dout("calc_object_layout '%s' pgid %d.%x\n", oid, poolid, ps);
941
942         ol->ol_pgid = pgid;
943         ol->ol_stripe_unit = fl->fl_object_stripe_unit;
944         return 0;
945 }
946
947 /*
948  * Calculate raw osd vector for the given pgid.  Return pointer to osd
949  * array, or NULL on failure.
950  */
951 static int *calc_pg_raw(struct ceph_osdmap *osdmap, struct ceph_pg pgid,
952                         int *osds, int *num)
953 {
954         struct ceph_pg_mapping *pg;
955         struct ceph_pg_pool_info *pool;
956         int ruleno;
957         unsigned poolid, ps, pps;
958         int preferred;
959
960         /* pg_temp? */
961         pg = __lookup_pg_mapping(&osdmap->pg_temp, pgid);
962         if (pg) {
963                 *num = pg->len;
964                 return pg->osds;
965         }
966
967         /* crush */
968         poolid = le32_to_cpu(pgid.pool);
969         ps = le16_to_cpu(pgid.ps);
970         preferred = (s16)le16_to_cpu(pgid.preferred);
971
972         /* don't forcefeed bad device ids to crush */
973         if (preferred >= osdmap->max_osd ||
974             preferred >= osdmap->crush->max_devices)
975                 preferred = -1;
976
977         pool = __lookup_pg_pool(&osdmap->pg_pools, poolid);
978         if (!pool)
979                 return NULL;
980         ruleno = crush_find_rule(osdmap->crush, pool->v.crush_ruleset,
981                                  pool->v.type, pool->v.size);
982         if (ruleno < 0) {
983                 pr_err("no crush rule pool %d type %d size %d\n",
984                        poolid, pool->v.type, pool->v.size);
985                 return NULL;
986         }
987
988         if (preferred >= 0)
989                 pps = ceph_stable_mod(ps,
990                                       le32_to_cpu(pool->v.lpgp_num),
991                                       pool->lpgp_num_mask);
992         else
993                 pps = ceph_stable_mod(ps,
994                                       le32_to_cpu(pool->v.pgp_num),
995                                       pool->pgp_num_mask);
996         pps += poolid;
997         *num = crush_do_rule(osdmap->crush, ruleno, pps, osds,
998                              min_t(int, pool->v.size, *num),
999                              preferred, osdmap->osd_weight);
1000         return osds;
1001 }
1002
1003 /*
1004  * Return primary osd for given pgid, or -1 if none.
1005  */
1006 int ceph_calc_pg_primary(struct ceph_osdmap *osdmap, struct ceph_pg pgid)
1007 {
1008         int rawosds[10], *osds;
1009         int i, num = ARRAY_SIZE(rawosds);
1010
1011         osds = calc_pg_raw(osdmap, pgid, rawosds, &num);
1012         if (!osds)
1013                 return -1;
1014
1015         /* primary is first up osd */
1016         for (i = 0; i < num; i++)
1017                 if (ceph_osd_is_up(osdmap, osds[i])) {
1018                         return osds[i];
1019                         break;
1020                 }
1021         return -1;
1022 }