GFS2: Use kmalloc when possible for ->readdir()
[linux-2.6.git] / fs / ceph / osdmap.c
1
2 #include "ceph_debug.h"
3
4 #include <linux/slab.h>
5 #include <asm/div64.h>
6
7 #include "super.h"
8 #include "osdmap.h"
9 #include "crush/hash.h"
10 #include "crush/mapper.h"
11 #include "decode.h"
12
13 char *ceph_osdmap_state_str(char *str, int len, int state)
14 {
15         int flag = 0;
16
17         if (!len)
18                 goto done;
19
20         *str = '\0';
21         if (state) {
22                 if (state & CEPH_OSD_EXISTS) {
23                         snprintf(str, len, "exists");
24                         flag = 1;
25                 }
26                 if (state & CEPH_OSD_UP) {
27                         snprintf(str, len, "%s%s%s", str, (flag ? ", " : ""),
28                                  "up");
29                         flag = 1;
30                 }
31         } else {
32                 snprintf(str, len, "doesn't exist");
33         }
34 done:
35         return str;
36 }
37
38 /* maps */
39
40 static int calc_bits_of(unsigned t)
41 {
42         int b = 0;
43         while (t) {
44                 t = t >> 1;
45                 b++;
46         }
47         return b;
48 }
49
50 /*
51  * the foo_mask is the smallest value 2^n-1 that is >= foo.
52  */
53 static void calc_pg_masks(struct ceph_pg_pool_info *pi)
54 {
55         pi->pg_num_mask = (1 << calc_bits_of(le32_to_cpu(pi->v.pg_num)-1)) - 1;
56         pi->pgp_num_mask =
57                 (1 << calc_bits_of(le32_to_cpu(pi->v.pgp_num)-1)) - 1;
58         pi->lpg_num_mask =
59                 (1 << calc_bits_of(le32_to_cpu(pi->v.lpg_num)-1)) - 1;
60         pi->lpgp_num_mask =
61                 (1 << calc_bits_of(le32_to_cpu(pi->v.lpgp_num)-1)) - 1;
62 }
63
64 /*
65  * decode crush map
66  */
67 static int crush_decode_uniform_bucket(void **p, void *end,
68                                        struct crush_bucket_uniform *b)
69 {
70         dout("crush_decode_uniform_bucket %p to %p\n", *p, end);
71         ceph_decode_need(p, end, (1+b->h.size) * sizeof(u32), bad);
72         b->item_weight = ceph_decode_32(p);
73         return 0;
74 bad:
75         return -EINVAL;
76 }
77
78 static int crush_decode_list_bucket(void **p, void *end,
79                                     struct crush_bucket_list *b)
80 {
81         int j;
82         dout("crush_decode_list_bucket %p to %p\n", *p, end);
83         b->item_weights = kcalloc(b->h.size, sizeof(u32), GFP_NOFS);
84         if (b->item_weights == NULL)
85                 return -ENOMEM;
86         b->sum_weights = kcalloc(b->h.size, sizeof(u32), GFP_NOFS);
87         if (b->sum_weights == NULL)
88                 return -ENOMEM;
89         ceph_decode_need(p, end, 2 * b->h.size * sizeof(u32), bad);
90         for (j = 0; j < b->h.size; j++) {
91                 b->item_weights[j] = ceph_decode_32(p);
92                 b->sum_weights[j] = ceph_decode_32(p);
93         }
94         return 0;
95 bad:
96         return -EINVAL;
97 }
98
99 static int crush_decode_tree_bucket(void **p, void *end,
100                                     struct crush_bucket_tree *b)
101 {
102         int j;
103         dout("crush_decode_tree_bucket %p to %p\n", *p, end);
104         ceph_decode_32_safe(p, end, b->num_nodes, bad);
105         b->node_weights = kcalloc(b->num_nodes, sizeof(u32), GFP_NOFS);
106         if (b->node_weights == NULL)
107                 return -ENOMEM;
108         ceph_decode_need(p, end, b->num_nodes * sizeof(u32), bad);
109         for (j = 0; j < b->num_nodes; j++)
110                 b->node_weights[j] = ceph_decode_32(p);
111         return 0;
112 bad:
113         return -EINVAL;
114 }
115
116 static int crush_decode_straw_bucket(void **p, void *end,
117                                      struct crush_bucket_straw *b)
118 {
119         int j;
120         dout("crush_decode_straw_bucket %p to %p\n", *p, end);
121         b->item_weights = kcalloc(b->h.size, sizeof(u32), GFP_NOFS);
122         if (b->item_weights == NULL)
123                 return -ENOMEM;
124         b->straws = kcalloc(b->h.size, sizeof(u32), GFP_NOFS);
125         if (b->straws == NULL)
126                 return -ENOMEM;
127         ceph_decode_need(p, end, 2 * b->h.size * sizeof(u32), bad);
128         for (j = 0; j < b->h.size; j++) {
129                 b->item_weights[j] = ceph_decode_32(p);
130                 b->straws[j] = ceph_decode_32(p);
131         }
132         return 0;
133 bad:
134         return -EINVAL;
135 }
136
137 static struct crush_map *crush_decode(void *pbyval, void *end)
138 {
139         struct crush_map *c;
140         int err = -EINVAL;
141         int i, j;
142         void **p = &pbyval;
143         void *start = pbyval;
144         u32 magic;
145
146         dout("crush_decode %p to %p len %d\n", *p, end, (int)(end - *p));
147
148         c = kzalloc(sizeof(*c), GFP_NOFS);
149         if (c == NULL)
150                 return ERR_PTR(-ENOMEM);
151
152         ceph_decode_need(p, end, 4*sizeof(u32), bad);
153         magic = ceph_decode_32(p);
154         if (magic != CRUSH_MAGIC) {
155                 pr_err("crush_decode magic %x != current %x\n",
156                        (unsigned)magic, (unsigned)CRUSH_MAGIC);
157                 goto bad;
158         }
159         c->max_buckets = ceph_decode_32(p);
160         c->max_rules = ceph_decode_32(p);
161         c->max_devices = ceph_decode_32(p);
162
163         c->device_parents = kcalloc(c->max_devices, sizeof(u32), GFP_NOFS);
164         if (c->device_parents == NULL)
165                 goto badmem;
166         c->bucket_parents = kcalloc(c->max_buckets, sizeof(u32), GFP_NOFS);
167         if (c->bucket_parents == NULL)
168                 goto badmem;
169
170         c->buckets = kcalloc(c->max_buckets, sizeof(*c->buckets), GFP_NOFS);
171         if (c->buckets == NULL)
172                 goto badmem;
173         c->rules = kcalloc(c->max_rules, sizeof(*c->rules), GFP_NOFS);
174         if (c->rules == NULL)
175                 goto badmem;
176
177         /* buckets */
178         for (i = 0; i < c->max_buckets; i++) {
179                 int size = 0;
180                 u32 alg;
181                 struct crush_bucket *b;
182
183                 ceph_decode_32_safe(p, end, alg, bad);
184                 if (alg == 0) {
185                         c->buckets[i] = NULL;
186                         continue;
187                 }
188                 dout("crush_decode bucket %d off %x %p to %p\n",
189                      i, (int)(*p-start), *p, end);
190
191                 switch (alg) {
192                 case CRUSH_BUCKET_UNIFORM:
193                         size = sizeof(struct crush_bucket_uniform);
194                         break;
195                 case CRUSH_BUCKET_LIST:
196                         size = sizeof(struct crush_bucket_list);
197                         break;
198                 case CRUSH_BUCKET_TREE:
199                         size = sizeof(struct crush_bucket_tree);
200                         break;
201                 case CRUSH_BUCKET_STRAW:
202                         size = sizeof(struct crush_bucket_straw);
203                         break;
204                 default:
205                         err = -EINVAL;
206                         goto bad;
207                 }
208                 BUG_ON(size == 0);
209                 b = c->buckets[i] = kzalloc(size, GFP_NOFS);
210                 if (b == NULL)
211                         goto badmem;
212
213                 ceph_decode_need(p, end, 4*sizeof(u32), bad);
214                 b->id = ceph_decode_32(p);
215                 b->type = ceph_decode_16(p);
216                 b->alg = ceph_decode_8(p);
217                 b->hash = ceph_decode_8(p);
218                 b->weight = ceph_decode_32(p);
219                 b->size = ceph_decode_32(p);
220
221                 dout("crush_decode bucket size %d off %x %p to %p\n",
222                      b->size, (int)(*p-start), *p, end);
223
224                 b->items = kcalloc(b->size, sizeof(__s32), GFP_NOFS);
225                 if (b->items == NULL)
226                         goto badmem;
227                 b->perm = kcalloc(b->size, sizeof(u32), GFP_NOFS);
228                 if (b->perm == NULL)
229                         goto badmem;
230                 b->perm_n = 0;
231
232                 ceph_decode_need(p, end, b->size*sizeof(u32), bad);
233                 for (j = 0; j < b->size; j++)
234                         b->items[j] = ceph_decode_32(p);
235
236                 switch (b->alg) {
237                 case CRUSH_BUCKET_UNIFORM:
238                         err = crush_decode_uniform_bucket(p, end,
239                                   (struct crush_bucket_uniform *)b);
240                         if (err < 0)
241                                 goto bad;
242                         break;
243                 case CRUSH_BUCKET_LIST:
244                         err = crush_decode_list_bucket(p, end,
245                                (struct crush_bucket_list *)b);
246                         if (err < 0)
247                                 goto bad;
248                         break;
249                 case CRUSH_BUCKET_TREE:
250                         err = crush_decode_tree_bucket(p, end,
251                                 (struct crush_bucket_tree *)b);
252                         if (err < 0)
253                                 goto bad;
254                         break;
255                 case CRUSH_BUCKET_STRAW:
256                         err = crush_decode_straw_bucket(p, end,
257                                 (struct crush_bucket_straw *)b);
258                         if (err < 0)
259                                 goto bad;
260                         break;
261                 }
262         }
263
264         /* rules */
265         dout("rule vec is %p\n", c->rules);
266         for (i = 0; i < c->max_rules; i++) {
267                 u32 yes;
268                 struct crush_rule *r;
269
270                 ceph_decode_32_safe(p, end, yes, bad);
271                 if (!yes) {
272                         dout("crush_decode NO rule %d off %x %p to %p\n",
273                              i, (int)(*p-start), *p, end);
274                         c->rules[i] = NULL;
275                         continue;
276                 }
277
278                 dout("crush_decode rule %d off %x %p to %p\n",
279                      i, (int)(*p-start), *p, end);
280
281                 /* len */
282                 ceph_decode_32_safe(p, end, yes, bad);
283 #if BITS_PER_LONG == 32
284                 err = -EINVAL;
285                 if (yes > ULONG_MAX / sizeof(struct crush_rule_step))
286                         goto bad;
287 #endif
288                 r = c->rules[i] = kmalloc(sizeof(*r) +
289                                           yes*sizeof(struct crush_rule_step),
290                                           GFP_NOFS);
291                 if (r == NULL)
292                         goto badmem;
293                 dout(" rule %d is at %p\n", i, r);
294                 r->len = yes;
295                 ceph_decode_copy_safe(p, end, &r->mask, 4, bad); /* 4 u8's */
296                 ceph_decode_need(p, end, r->len*3*sizeof(u32), bad);
297                 for (j = 0; j < r->len; j++) {
298                         r->steps[j].op = ceph_decode_32(p);
299                         r->steps[j].arg1 = ceph_decode_32(p);
300                         r->steps[j].arg2 = ceph_decode_32(p);
301                 }
302         }
303
304         /* ignore trailing name maps. */
305
306         dout("crush_decode success\n");
307         return c;
308
309 badmem:
310         err = -ENOMEM;
311 bad:
312         dout("crush_decode fail %d\n", err);
313         crush_destroy(c);
314         return ERR_PTR(err);
315 }
316
317 /*
318  * rbtree of pg_mapping for handling pg_temp (explicit mapping of pgid
319  * to a set of osds)
320  */
321 static int pgid_cmp(struct ceph_pg l, struct ceph_pg r)
322 {
323         u64 a = *(u64 *)&l;
324         u64 b = *(u64 *)&r;
325
326         if (a < b)
327                 return -1;
328         if (a > b)
329                 return 1;
330         return 0;
331 }
332
333 static int __insert_pg_mapping(struct ceph_pg_mapping *new,
334                                struct rb_root *root)
335 {
336         struct rb_node **p = &root->rb_node;
337         struct rb_node *parent = NULL;
338         struct ceph_pg_mapping *pg = NULL;
339         int c;
340
341         while (*p) {
342                 parent = *p;
343                 pg = rb_entry(parent, struct ceph_pg_mapping, node);
344                 c = pgid_cmp(new->pgid, pg->pgid);
345                 if (c < 0)
346                         p = &(*p)->rb_left;
347                 else if (c > 0)
348                         p = &(*p)->rb_right;
349                 else
350                         return -EEXIST;
351         }
352
353         rb_link_node(&new->node, parent, p);
354         rb_insert_color(&new->node, root);
355         return 0;
356 }
357
358 static struct ceph_pg_mapping *__lookup_pg_mapping(struct rb_root *root,
359                                                    struct ceph_pg pgid)
360 {
361         struct rb_node *n = root->rb_node;
362         struct ceph_pg_mapping *pg;
363         int c;
364
365         while (n) {
366                 pg = rb_entry(n, struct ceph_pg_mapping, node);
367                 c = pgid_cmp(pgid, pg->pgid);
368                 if (c < 0)
369                         n = n->rb_left;
370                 else if (c > 0)
371                         n = n->rb_right;
372                 else
373                         return pg;
374         }
375         return NULL;
376 }
377
378 /*
379  * rbtree of pg pool info
380  */
381 static int __insert_pg_pool(struct rb_root *root, struct ceph_pg_pool_info *new)
382 {
383         struct rb_node **p = &root->rb_node;
384         struct rb_node *parent = NULL;
385         struct ceph_pg_pool_info *pi = NULL;
386
387         while (*p) {
388                 parent = *p;
389                 pi = rb_entry(parent, struct ceph_pg_pool_info, node);
390                 if (new->id < pi->id)
391                         p = &(*p)->rb_left;
392                 else if (new->id > pi->id)
393                         p = &(*p)->rb_right;
394                 else
395                         return -EEXIST;
396         }
397
398         rb_link_node(&new->node, parent, p);
399         rb_insert_color(&new->node, root);
400         return 0;
401 }
402
403 static struct ceph_pg_pool_info *__lookup_pg_pool(struct rb_root *root, int id)
404 {
405         struct ceph_pg_pool_info *pi;
406         struct rb_node *n = root->rb_node;
407
408         while (n) {
409                 pi = rb_entry(n, struct ceph_pg_pool_info, node);
410                 if (id < pi->id)
411                         n = n->rb_left;
412                 else if (id > pi->id)
413                         n = n->rb_right;
414                 else
415                         return pi;
416         }
417         return NULL;
418 }
419
420 static void __remove_pg_pool(struct rb_root *root, struct ceph_pg_pool_info *pi)
421 {
422         rb_erase(&pi->node, root);
423         kfree(pi->name);
424         kfree(pi);
425 }
426
427 void __decode_pool(void **p, struct ceph_pg_pool_info *pi)
428 {
429         ceph_decode_copy(p, &pi->v, sizeof(pi->v));
430         calc_pg_masks(pi);
431         *p += le32_to_cpu(pi->v.num_snaps) * sizeof(u64);
432         *p += le32_to_cpu(pi->v.num_removed_snap_intervals) * sizeof(u64) * 2;
433 }
434
435 static int __decode_pool_names(void **p, void *end, struct ceph_osdmap *map)
436 {
437         struct ceph_pg_pool_info *pi;
438         u32 num, len, pool;
439
440         ceph_decode_32_safe(p, end, num, bad);
441         dout(" %d pool names\n", num);
442         while (num--) {
443                 ceph_decode_32_safe(p, end, pool, bad);
444                 ceph_decode_32_safe(p, end, len, bad);
445                 dout("  pool %d len %d\n", pool, len);
446                 pi = __lookup_pg_pool(&map->pg_pools, pool);
447                 if (pi) {
448                         kfree(pi->name);
449                         pi->name = kmalloc(len + 1, GFP_NOFS);
450                         if (pi->name) {
451                                 memcpy(pi->name, *p, len);
452                                 pi->name[len] = '\0';
453                                 dout("  name is %s\n", pi->name);
454                         }
455                 }
456                 *p += len;
457         }
458         return 0;
459
460 bad:
461         return -EINVAL;
462 }
463
464 /*
465  * osd map
466  */
467 void ceph_osdmap_destroy(struct ceph_osdmap *map)
468 {
469         dout("osdmap_destroy %p\n", map);
470         if (map->crush)
471                 crush_destroy(map->crush);
472         while (!RB_EMPTY_ROOT(&map->pg_temp)) {
473                 struct ceph_pg_mapping *pg =
474                         rb_entry(rb_first(&map->pg_temp),
475                                  struct ceph_pg_mapping, node);
476                 rb_erase(&pg->node, &map->pg_temp);
477                 kfree(pg);
478         }
479         while (!RB_EMPTY_ROOT(&map->pg_pools)) {
480                 struct ceph_pg_pool_info *pi =
481                         rb_entry(rb_first(&map->pg_pools),
482                                  struct ceph_pg_pool_info, node);
483                 __remove_pg_pool(&map->pg_pools, pi);
484         }
485         kfree(map->osd_state);
486         kfree(map->osd_weight);
487         kfree(map->osd_addr);
488         kfree(map);
489 }
490
491 /*
492  * adjust max osd value.  reallocate arrays.
493  */
494 static int osdmap_set_max_osd(struct ceph_osdmap *map, int max)
495 {
496         u8 *state;
497         struct ceph_entity_addr *addr;
498         u32 *weight;
499
500         state = kcalloc(max, sizeof(*state), GFP_NOFS);
501         addr = kcalloc(max, sizeof(*addr), GFP_NOFS);
502         weight = kcalloc(max, sizeof(*weight), GFP_NOFS);
503         if (state == NULL || addr == NULL || weight == NULL) {
504                 kfree(state);
505                 kfree(addr);
506                 kfree(weight);
507                 return -ENOMEM;
508         }
509
510         /* copy old? */
511         if (map->osd_state) {
512                 memcpy(state, map->osd_state, map->max_osd*sizeof(*state));
513                 memcpy(addr, map->osd_addr, map->max_osd*sizeof(*addr));
514                 memcpy(weight, map->osd_weight, map->max_osd*sizeof(*weight));
515                 kfree(map->osd_state);
516                 kfree(map->osd_addr);
517                 kfree(map->osd_weight);
518         }
519
520         map->osd_state = state;
521         map->osd_weight = weight;
522         map->osd_addr = addr;
523         map->max_osd = max;
524         return 0;
525 }
526
527 /*
528  * decode a full map.
529  */
530 struct ceph_osdmap *osdmap_decode(void **p, void *end)
531 {
532         struct ceph_osdmap *map;
533         u16 version;
534         u32 len, max, i;
535         u8 ev;
536         int err = -EINVAL;
537         void *start = *p;
538         struct ceph_pg_pool_info *pi;
539
540         dout("osdmap_decode %p to %p len %d\n", *p, end, (int)(end - *p));
541
542         map = kzalloc(sizeof(*map), GFP_NOFS);
543         if (map == NULL)
544                 return ERR_PTR(-ENOMEM);
545         map->pg_temp = RB_ROOT;
546
547         ceph_decode_16_safe(p, end, version, bad);
548         if (version > CEPH_OSDMAP_VERSION) {
549                 pr_warning("got unknown v %d > %d of osdmap\n", version,
550                            CEPH_OSDMAP_VERSION);
551                 goto bad;
552         }
553
554         ceph_decode_need(p, end, 2*sizeof(u64)+6*sizeof(u32), bad);
555         ceph_decode_copy(p, &map->fsid, sizeof(map->fsid));
556         map->epoch = ceph_decode_32(p);
557         ceph_decode_copy(p, &map->created, sizeof(map->created));
558         ceph_decode_copy(p, &map->modified, sizeof(map->modified));
559
560         ceph_decode_32_safe(p, end, max, bad);
561         while (max--) {
562                 ceph_decode_need(p, end, 4 + 1 + sizeof(pi->v), bad);
563                 pi = kzalloc(sizeof(*pi), GFP_NOFS);
564                 if (!pi)
565                         goto bad;
566                 pi->id = ceph_decode_32(p);
567                 ev = ceph_decode_8(p); /* encoding version */
568                 if (ev > CEPH_PG_POOL_VERSION) {
569                         pr_warning("got unknown v %d > %d of ceph_pg_pool\n",
570                                    ev, CEPH_PG_POOL_VERSION);
571                         kfree(pi);
572                         goto bad;
573                 }
574                 __decode_pool(p, pi);
575                 __insert_pg_pool(&map->pg_pools, pi);
576         }
577
578         if (version >= 5 && __decode_pool_names(p, end, map) < 0)
579                 goto bad;
580
581         ceph_decode_32_safe(p, end, map->pool_max, bad);
582
583         ceph_decode_32_safe(p, end, map->flags, bad);
584
585         max = ceph_decode_32(p);
586
587         /* (re)alloc osd arrays */
588         err = osdmap_set_max_osd(map, max);
589         if (err < 0)
590                 goto bad;
591         dout("osdmap_decode max_osd = %d\n", map->max_osd);
592
593         /* osds */
594         err = -EINVAL;
595         ceph_decode_need(p, end, 3*sizeof(u32) +
596                          map->max_osd*(1 + sizeof(*map->osd_weight) +
597                                        sizeof(*map->osd_addr)), bad);
598         *p += 4; /* skip length field (should match max) */
599         ceph_decode_copy(p, map->osd_state, map->max_osd);
600
601         *p += 4; /* skip length field (should match max) */
602         for (i = 0; i < map->max_osd; i++)
603                 map->osd_weight[i] = ceph_decode_32(p);
604
605         *p += 4; /* skip length field (should match max) */
606         ceph_decode_copy(p, map->osd_addr, map->max_osd*sizeof(*map->osd_addr));
607         for (i = 0; i < map->max_osd; i++)
608                 ceph_decode_addr(&map->osd_addr[i]);
609
610         /* pg_temp */
611         ceph_decode_32_safe(p, end, len, bad);
612         for (i = 0; i < len; i++) {
613                 int n, j;
614                 struct ceph_pg pgid;
615                 struct ceph_pg_mapping *pg;
616
617                 ceph_decode_need(p, end, sizeof(u32) + sizeof(u64), bad);
618                 ceph_decode_copy(p, &pgid, sizeof(pgid));
619                 n = ceph_decode_32(p);
620                 ceph_decode_need(p, end, n * sizeof(u32), bad);
621                 err = -ENOMEM;
622                 pg = kmalloc(sizeof(*pg) + n*sizeof(u32), GFP_NOFS);
623                 if (!pg)
624                         goto bad;
625                 pg->pgid = pgid;
626                 pg->len = n;
627                 for (j = 0; j < n; j++)
628                         pg->osds[j] = ceph_decode_32(p);
629
630                 err = __insert_pg_mapping(pg, &map->pg_temp);
631                 if (err)
632                         goto bad;
633                 dout(" added pg_temp %llx len %d\n", *(u64 *)&pgid, len);
634         }
635
636         /* crush */
637         ceph_decode_32_safe(p, end, len, bad);
638         dout("osdmap_decode crush len %d from off 0x%x\n", len,
639              (int)(*p - start));
640         ceph_decode_need(p, end, len, bad);
641         map->crush = crush_decode(*p, end);
642         *p += len;
643         if (IS_ERR(map->crush)) {
644                 err = PTR_ERR(map->crush);
645                 map->crush = NULL;
646                 goto bad;
647         }
648
649         /* ignore the rest of the map */
650         *p = end;
651
652         dout("osdmap_decode done %p %p\n", *p, end);
653         return map;
654
655 bad:
656         dout("osdmap_decode fail\n");
657         ceph_osdmap_destroy(map);
658         return ERR_PTR(err);
659 }
660
661 /*
662  * decode and apply an incremental map update.
663  */
664 struct ceph_osdmap *osdmap_apply_incremental(void **p, void *end,
665                                              struct ceph_osdmap *map,
666                                              struct ceph_messenger *msgr)
667 {
668         struct crush_map *newcrush = NULL;
669         struct ceph_fsid fsid;
670         u32 epoch = 0;
671         struct ceph_timespec modified;
672         u32 len, pool;
673         __s32 new_pool_max, new_flags, max;
674         void *start = *p;
675         int err = -EINVAL;
676         u16 version;
677         struct rb_node *rbp;
678
679         ceph_decode_16_safe(p, end, version, bad);
680         if (version > CEPH_OSDMAP_INC_VERSION) {
681                 pr_warning("got unknown v %d > %d of inc osdmap\n", version,
682                            CEPH_OSDMAP_INC_VERSION);
683                 goto bad;
684         }
685
686         ceph_decode_need(p, end, sizeof(fsid)+sizeof(modified)+2*sizeof(u32),
687                          bad);
688         ceph_decode_copy(p, &fsid, sizeof(fsid));
689         epoch = ceph_decode_32(p);
690         BUG_ON(epoch != map->epoch+1);
691         ceph_decode_copy(p, &modified, sizeof(modified));
692         new_pool_max = ceph_decode_32(p);
693         new_flags = ceph_decode_32(p);
694
695         /* full map? */
696         ceph_decode_32_safe(p, end, len, bad);
697         if (len > 0) {
698                 dout("apply_incremental full map len %d, %p to %p\n",
699                      len, *p, end);
700                 return osdmap_decode(p, min(*p+len, end));
701         }
702
703         /* new crush? */
704         ceph_decode_32_safe(p, end, len, bad);
705         if (len > 0) {
706                 dout("apply_incremental new crush map len %d, %p to %p\n",
707                      len, *p, end);
708                 newcrush = crush_decode(*p, min(*p+len, end));
709                 if (IS_ERR(newcrush))
710                         return ERR_CAST(newcrush);
711                 *p += len;
712         }
713
714         /* new flags? */
715         if (new_flags >= 0)
716                 map->flags = new_flags;
717         if (new_pool_max >= 0)
718                 map->pool_max = new_pool_max;
719
720         ceph_decode_need(p, end, 5*sizeof(u32), bad);
721
722         /* new max? */
723         max = ceph_decode_32(p);
724         if (max >= 0) {
725                 err = osdmap_set_max_osd(map, max);
726                 if (err < 0)
727                         goto bad;
728         }
729
730         map->epoch++;
731         map->modified = map->modified;
732         if (newcrush) {
733                 if (map->crush)
734                         crush_destroy(map->crush);
735                 map->crush = newcrush;
736                 newcrush = NULL;
737         }
738
739         /* new_pool */
740         ceph_decode_32_safe(p, end, len, bad);
741         while (len--) {
742                 __u8 ev;
743                 struct ceph_pg_pool_info *pi;
744
745                 ceph_decode_32_safe(p, end, pool, bad);
746                 ceph_decode_need(p, end, 1 + sizeof(pi->v), bad);
747                 ev = ceph_decode_8(p);  /* encoding version */
748                 if (ev > CEPH_PG_POOL_VERSION) {
749                         pr_warning("got unknown v %d > %d of ceph_pg_pool\n",
750                                    ev, CEPH_PG_POOL_VERSION);
751                         goto bad;
752                 }
753                 pi = __lookup_pg_pool(&map->pg_pools, pool);
754                 if (!pi) {
755                         pi = kzalloc(sizeof(*pi), GFP_NOFS);
756                         if (!pi) {
757                                 err = -ENOMEM;
758                                 goto bad;
759                         }
760                         pi->id = pool;
761                         __insert_pg_pool(&map->pg_pools, pi);
762                 }
763                 __decode_pool(p, pi);
764         }
765         if (version >= 5 && __decode_pool_names(p, end, map) < 0)
766                 goto bad;
767
768         /* old_pool */
769         ceph_decode_32_safe(p, end, len, bad);
770         while (len--) {
771                 struct ceph_pg_pool_info *pi;
772
773                 ceph_decode_32_safe(p, end, pool, bad);
774                 pi = __lookup_pg_pool(&map->pg_pools, pool);
775                 if (pi)
776                         __remove_pg_pool(&map->pg_pools, pi);
777         }
778
779         /* new_up */
780         err = -EINVAL;
781         ceph_decode_32_safe(p, end, len, bad);
782         while (len--) {
783                 u32 osd;
784                 struct ceph_entity_addr addr;
785                 ceph_decode_32_safe(p, end, osd, bad);
786                 ceph_decode_copy_safe(p, end, &addr, sizeof(addr), bad);
787                 ceph_decode_addr(&addr);
788                 pr_info("osd%d up\n", osd);
789                 BUG_ON(osd >= map->max_osd);
790                 map->osd_state[osd] |= CEPH_OSD_UP;
791                 map->osd_addr[osd] = addr;
792         }
793
794         /* new_down */
795         ceph_decode_32_safe(p, end, len, bad);
796         while (len--) {
797                 u32 osd;
798                 ceph_decode_32_safe(p, end, osd, bad);
799                 (*p)++;  /* clean flag */
800                 pr_info("osd%d down\n", osd);
801                 if (osd < map->max_osd)
802                         map->osd_state[osd] &= ~CEPH_OSD_UP;
803         }
804
805         /* new_weight */
806         ceph_decode_32_safe(p, end, len, bad);
807         while (len--) {
808                 u32 osd, off;
809                 ceph_decode_need(p, end, sizeof(u32)*2, bad);
810                 osd = ceph_decode_32(p);
811                 off = ceph_decode_32(p);
812                 pr_info("osd%d weight 0x%x %s\n", osd, off,
813                      off == CEPH_OSD_IN ? "(in)" :
814                      (off == CEPH_OSD_OUT ? "(out)" : ""));
815                 if (osd < map->max_osd)
816                         map->osd_weight[osd] = off;
817         }
818
819         /* new_pg_temp */
820         rbp = rb_first(&map->pg_temp);
821         ceph_decode_32_safe(p, end, len, bad);
822         while (len--) {
823                 struct ceph_pg_mapping *pg;
824                 int j;
825                 struct ceph_pg pgid;
826                 u32 pglen;
827                 ceph_decode_need(p, end, sizeof(u64) + sizeof(u32), bad);
828                 ceph_decode_copy(p, &pgid, sizeof(pgid));
829                 pglen = ceph_decode_32(p);
830
831                 /* remove any? */
832                 while (rbp && pgid_cmp(rb_entry(rbp, struct ceph_pg_mapping,
833                                                 node)->pgid, pgid) <= 0) {
834                         struct rb_node *cur = rbp;
835                         rbp = rb_next(rbp);
836                         dout(" removed pg_temp %llx\n",
837                              *(u64 *)&rb_entry(cur, struct ceph_pg_mapping,
838                                                node)->pgid);
839                         rb_erase(cur, &map->pg_temp);
840                 }
841
842                 if (pglen) {
843                         /* insert */
844                         ceph_decode_need(p, end, pglen*sizeof(u32), bad);
845                         pg = kmalloc(sizeof(*pg) + sizeof(u32)*pglen, GFP_NOFS);
846                         if (!pg) {
847                                 err = -ENOMEM;
848                                 goto bad;
849                         }
850                         pg->pgid = pgid;
851                         pg->len = pglen;
852                         for (j = 0; j < pglen; j++)
853                                 pg->osds[j] = ceph_decode_32(p);
854                         err = __insert_pg_mapping(pg, &map->pg_temp);
855                         if (err)
856                                 goto bad;
857                         dout(" added pg_temp %llx len %d\n", *(u64 *)&pgid,
858                              pglen);
859                 }
860         }
861         while (rbp) {
862                 struct rb_node *cur = rbp;
863                 rbp = rb_next(rbp);
864                 dout(" removed pg_temp %llx\n",
865                      *(u64 *)&rb_entry(cur, struct ceph_pg_mapping,
866                                        node)->pgid);
867                 rb_erase(cur, &map->pg_temp);
868         }
869
870         /* ignore the rest */
871         *p = end;
872         return map;
873
874 bad:
875         pr_err("corrupt inc osdmap epoch %d off %d (%p of %p-%p)\n",
876                epoch, (int)(*p - start), *p, start, end);
877         print_hex_dump(KERN_DEBUG, "osdmap: ",
878                        DUMP_PREFIX_OFFSET, 16, 1,
879                        start, end - start, true);
880         if (newcrush)
881                 crush_destroy(newcrush);
882         return ERR_PTR(err);
883 }
884
885
886
887
888 /*
889  * calculate file layout from given offset, length.
890  * fill in correct oid, logical length, and object extent
891  * offset, length.
892  *
893  * for now, we write only a single su, until we can
894  * pass a stride back to the caller.
895  */
896 void ceph_calc_file_object_mapping(struct ceph_file_layout *layout,
897                                    u64 off, u64 *plen,
898                                    u64 *ono,
899                                    u64 *oxoff, u64 *oxlen)
900 {
901         u32 osize = le32_to_cpu(layout->fl_object_size);
902         u32 su = le32_to_cpu(layout->fl_stripe_unit);
903         u32 sc = le32_to_cpu(layout->fl_stripe_count);
904         u32 bl, stripeno, stripepos, objsetno;
905         u32 su_per_object;
906         u64 t, su_offset;
907
908         dout("mapping %llu~%llu  osize %u fl_su %u\n", off, *plen,
909              osize, su);
910         su_per_object = osize / su;
911         dout("osize %u / su %u = su_per_object %u\n", osize, su,
912              su_per_object);
913
914         BUG_ON((su & ~PAGE_MASK) != 0);
915         /* bl = *off / su; */
916         t = off;
917         do_div(t, su);
918         bl = t;
919         dout("off %llu / su %u = bl %u\n", off, su, bl);
920
921         stripeno = bl / sc;
922         stripepos = bl % sc;
923         objsetno = stripeno / su_per_object;
924
925         *ono = objsetno * sc + stripepos;
926         dout("objset %u * sc %u = ono %u\n", objsetno, sc, (unsigned)*ono);
927
928         /* *oxoff = *off % layout->fl_stripe_unit;  # offset in su */
929         t = off;
930         su_offset = do_div(t, su);
931         *oxoff = su_offset + (stripeno % su_per_object) * su;
932
933         /*
934          * Calculate the length of the extent being written to the selected
935          * object. This is the minimum of the full length requested (plen) or
936          * the remainder of the current stripe being written to.
937          */
938         *oxlen = min_t(u64, *plen, su - su_offset);
939         *plen = *oxlen;
940
941         dout(" obj extent %llu~%llu\n", *oxoff, *oxlen);
942 }
943
944 /*
945  * calculate an object layout (i.e. pgid) from an oid,
946  * file_layout, and osdmap
947  */
948 int ceph_calc_object_layout(struct ceph_object_layout *ol,
949                             const char *oid,
950                             struct ceph_file_layout *fl,
951                             struct ceph_osdmap *osdmap)
952 {
953         unsigned num, num_mask;
954         struct ceph_pg pgid;
955         s32 preferred = (s32)le32_to_cpu(fl->fl_pg_preferred);
956         int poolid = le32_to_cpu(fl->fl_pg_pool);
957         struct ceph_pg_pool_info *pool;
958         unsigned ps;
959
960         BUG_ON(!osdmap);
961
962         pool = __lookup_pg_pool(&osdmap->pg_pools, poolid);
963         if (!pool)
964                 return -EIO;
965         ps = ceph_str_hash(pool->v.object_hash, oid, strlen(oid));
966         if (preferred >= 0) {
967                 ps += preferred;
968                 num = le32_to_cpu(pool->v.lpg_num);
969                 num_mask = pool->lpg_num_mask;
970         } else {
971                 num = le32_to_cpu(pool->v.pg_num);
972                 num_mask = pool->pg_num_mask;
973         }
974
975         pgid.ps = cpu_to_le16(ps);
976         pgid.preferred = cpu_to_le16(preferred);
977         pgid.pool = fl->fl_pg_pool;
978         if (preferred >= 0)
979                 dout("calc_object_layout '%s' pgid %d.%xp%d\n", oid, poolid, ps,
980                      (int)preferred);
981         else
982                 dout("calc_object_layout '%s' pgid %d.%x\n", oid, poolid, ps);
983
984         ol->ol_pgid = pgid;
985         ol->ol_stripe_unit = fl->fl_object_stripe_unit;
986         return 0;
987 }
988
989 /*
990  * Calculate raw osd vector for the given pgid.  Return pointer to osd
991  * array, or NULL on failure.
992  */
993 static int *calc_pg_raw(struct ceph_osdmap *osdmap, struct ceph_pg pgid,
994                         int *osds, int *num)
995 {
996         struct ceph_pg_mapping *pg;
997         struct ceph_pg_pool_info *pool;
998         int ruleno;
999         unsigned poolid, ps, pps;
1000         int preferred;
1001
1002         /* pg_temp? */
1003         pg = __lookup_pg_mapping(&osdmap->pg_temp, pgid);
1004         if (pg) {
1005                 *num = pg->len;
1006                 return pg->osds;
1007         }
1008
1009         /* crush */
1010         poolid = le32_to_cpu(pgid.pool);
1011         ps = le16_to_cpu(pgid.ps);
1012         preferred = (s16)le16_to_cpu(pgid.preferred);
1013
1014         /* don't forcefeed bad device ids to crush */
1015         if (preferred >= osdmap->max_osd ||
1016             preferred >= osdmap->crush->max_devices)
1017                 preferred = -1;
1018
1019         pool = __lookup_pg_pool(&osdmap->pg_pools, poolid);
1020         if (!pool)
1021                 return NULL;
1022         ruleno = crush_find_rule(osdmap->crush, pool->v.crush_ruleset,
1023                                  pool->v.type, pool->v.size);
1024         if (ruleno < 0) {
1025                 pr_err("no crush rule pool %d type %d size %d\n",
1026                        poolid, pool->v.type, pool->v.size);
1027                 return NULL;
1028         }
1029
1030         if (preferred >= 0)
1031                 pps = ceph_stable_mod(ps,
1032                                       le32_to_cpu(pool->v.lpgp_num),
1033                                       pool->lpgp_num_mask);
1034         else
1035                 pps = ceph_stable_mod(ps,
1036                                       le32_to_cpu(pool->v.pgp_num),
1037                                       pool->pgp_num_mask);
1038         pps += poolid;
1039         *num = crush_do_rule(osdmap->crush, ruleno, pps, osds,
1040                              min_t(int, pool->v.size, *num),
1041                              preferred, osdmap->osd_weight);
1042         return osds;
1043 }
1044
1045 /*
1046  * Return acting set for given pgid.
1047  */
1048 int ceph_calc_pg_acting(struct ceph_osdmap *osdmap, struct ceph_pg pgid,
1049                         int *acting)
1050 {
1051         int rawosds[CEPH_PG_MAX_SIZE], *osds;
1052         int i, o, num = CEPH_PG_MAX_SIZE;
1053
1054         osds = calc_pg_raw(osdmap, pgid, rawosds, &num);
1055         if (!osds)
1056                 return -1;
1057
1058         /* primary is first up osd */
1059         o = 0;
1060         for (i = 0; i < num; i++)
1061                 if (ceph_osd_is_up(osdmap, osds[i]))
1062                         acting[o++] = osds[i];
1063         return o;
1064 }
1065
1066 /*
1067  * Return primary osd for given pgid, or -1 if none.
1068  */
1069 int ceph_calc_pg_primary(struct ceph_osdmap *osdmap, struct ceph_pg pgid)
1070 {
1071         int rawosds[CEPH_PG_MAX_SIZE], *osds;
1072         int i, num = CEPH_PG_MAX_SIZE;
1073
1074         osds = calc_pg_raw(osdmap, pgid, rawosds, &num);
1075         if (!osds)
1076                 return -1;
1077
1078         /* primary is first up osd */
1079         for (i = 0; i < num; i++)
1080                 if (ceph_osd_is_up(osdmap, osds[i]))
1081                         return osds[i];
1082         return -1;
1083 }