ceph: fix memory leak when destroying osdmap with pg_temp mappings
[linux-2.6-block.git] / fs / ceph / osdmap.c
CommitLineData
f24e9980
SW
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
11char *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 }
32done:
33 return str;
34}
35
36/* maps */
37
38static 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 */
51static 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 */
65static 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);
c89136ea 70 b->item_weight = ceph_decode_32(p);
f24e9980
SW
71 return 0;
72bad:
73 return -EINVAL;
74}
75
76static 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++) {
c89136ea
SW
89 b->item_weights[j] = ceph_decode_32(p);
90 b->sum_weights[j] = ceph_decode_32(p);
f24e9980
SW
91 }
92 return 0;
93bad:
94 return -EINVAL;
95}
96
97static 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++)
c89136ea 108 b->node_weights[j] = ceph_decode_32(p);
f24e9980
SW
109 return 0;
110bad:
111 return -EINVAL;
112}
113
114static 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++) {
c89136ea
SW
127 b->item_weights[j] = ceph_decode_32(p);
128 b->straws[j] = ceph_decode_32(p);
f24e9980
SW
129 }
130 return 0;
131bad:
132 return -EINVAL;
133}
134
135static 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);
c89136ea 151 magic = ceph_decode_32(p);
f24e9980
SW
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 }
c89136ea
SW
157 c->max_buckets = ceph_decode_32(p);
158 c->max_rules = ceph_decode_32(p);
159 c->max_devices = ceph_decode_32(p);
f24e9980
SW
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:
30dc6381 203 err = -EINVAL;
f24e9980
SW
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);
c89136ea
SW
212 b->id = ceph_decode_32(p);
213 b->type = ceph_decode_16(p);
fb690390
SW
214 b->alg = ceph_decode_8(p);
215 b->hash = ceph_decode_8(p);
c89136ea
SW
216 b->weight = ceph_decode_32(p);
217 b->size = ceph_decode_32(p);
f24e9980
SW
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++)
c89136ea 232 b->items[j] = ceph_decode_32(p);
f24e9980
SW
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
30dc6381 282 err = -EINVAL;
f24e9980
SW
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++) {
c89136ea
SW
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);
f24e9980
SW
299 }
300 }
301
302 /* ignore trailing name maps. */
303
304 dout("crush_decode success\n");
305 return c;
306
307badmem:
308 err = -ENOMEM;
309bad:
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 */
319void ceph_osdmap_destroy(struct ceph_osdmap *map)
320{
321 dout("osdmap_destroy %p\n", map);
322 if (map->crush)
323 crush_destroy(map->crush);
9794b146
SW
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 }
f24e9980
SW
331 kfree(map->osd_state);
332 kfree(map->osd_weight);
333 kfree(map->pg_pool);
334 kfree(map->osd_addr);
335 kfree(map);
336}
337
338/*
339 * adjust max osd value. reallocate arrays.
340 */
341static int osdmap_set_max_osd(struct ceph_osdmap *map, int max)
342{
343 u8 *state;
344 struct ceph_entity_addr *addr;
345 u32 *weight;
346
347 state = kcalloc(max, sizeof(*state), GFP_NOFS);
348 addr = kcalloc(max, sizeof(*addr), GFP_NOFS);
349 weight = kcalloc(max, sizeof(*weight), GFP_NOFS);
350 if (state == NULL || addr == NULL || weight == NULL) {
351 kfree(state);
352 kfree(addr);
353 kfree(weight);
354 return -ENOMEM;
355 }
356
357 /* copy old? */
358 if (map->osd_state) {
359 memcpy(state, map->osd_state, map->max_osd*sizeof(*state));
360 memcpy(addr, map->osd_addr, map->max_osd*sizeof(*addr));
361 memcpy(weight, map->osd_weight, map->max_osd*sizeof(*weight));
362 kfree(map->osd_state);
363 kfree(map->osd_addr);
364 kfree(map->osd_weight);
365 }
366
367 map->osd_state = state;
368 map->osd_weight = weight;
369 map->osd_addr = addr;
370 map->max_osd = max;
371 return 0;
372}
373
374/*
9794b146
SW
375 * rbtree of pg_mapping for handling pg_temp (explicit mapping of pgid
376 * to a set of osds)
f24e9980 377 */
51042122
SW
378static int pgid_cmp(struct ceph_pg l, struct ceph_pg r)
379{
380 u64 a = *(u64 *)&l;
381 u64 b = *(u64 *)&r;
382
383 if (a < b)
384 return -1;
385 if (a > b)
386 return 1;
387 return 0;
388}
389
991abb6e
SW
390static int __insert_pg_mapping(struct ceph_pg_mapping *new,
391 struct rb_root *root)
f24e9980
SW
392{
393 struct rb_node **p = &root->rb_node;
394 struct rb_node *parent = NULL;
395 struct ceph_pg_mapping *pg = NULL;
51042122 396 int c;
f24e9980
SW
397
398 while (*p) {
399 parent = *p;
400 pg = rb_entry(parent, struct ceph_pg_mapping, node);
51042122
SW
401 c = pgid_cmp(new->pgid, pg->pgid);
402 if (c < 0)
f24e9980 403 p = &(*p)->rb_left;
51042122 404 else if (c > 0)
f24e9980
SW
405 p = &(*p)->rb_right;
406 else
991abb6e 407 return -EEXIST;
f24e9980
SW
408 }
409
410 rb_link_node(&new->node, parent, p);
411 rb_insert_color(&new->node, root);
991abb6e 412 return 0;
f24e9980
SW
413}
414
9794b146
SW
415static struct ceph_pg_mapping *__lookup_pg_mapping(struct rb_root *root,
416 struct ceph_pg pgid)
417{
418 struct rb_node *n = root->rb_node;
419 struct ceph_pg_mapping *pg;
420 int c;
421
422 while (n) {
423 pg = rb_entry(n, struct ceph_pg_mapping, node);
424 c = pgid_cmp(pgid, pg->pgid);
425 if (c < 0)
426 n = n->rb_left;
427 else if (c > 0)
428 n = n->rb_right;
429 else
430 return pg;
431 }
432 return NULL;
433}
434
f24e9980
SW
435/*
436 * decode a full map.
437 */
438struct ceph_osdmap *osdmap_decode(void **p, void *end)
439{
440 struct ceph_osdmap *map;
441 u16 version;
442 u32 len, max, i;
361be860 443 u8 ev;
f24e9980
SW
444 int err = -EINVAL;
445 void *start = *p;
446
447 dout("osdmap_decode %p to %p len %d\n", *p, end, (int)(end - *p));
448
449 map = kzalloc(sizeof(*map), GFP_NOFS);
450 if (map == NULL)
451 return ERR_PTR(-ENOMEM);
452 map->pg_temp = RB_ROOT;
453
454 ceph_decode_16_safe(p, end, version, bad);
02f90c61
SW
455 if (version > CEPH_OSDMAP_VERSION) {
456 pr_warning("got unknown v %d > %d of osdmap\n", version,
457 CEPH_OSDMAP_VERSION);
458 goto bad;
459 }
f24e9980
SW
460
461 ceph_decode_need(p, end, 2*sizeof(u64)+6*sizeof(u32), bad);
462 ceph_decode_copy(p, &map->fsid, sizeof(map->fsid));
c89136ea 463 map->epoch = ceph_decode_32(p);
f24e9980
SW
464 ceph_decode_copy(p, &map->created, sizeof(map->created));
465 ceph_decode_copy(p, &map->modified, sizeof(map->modified));
466
c89136ea 467 map->num_pools = ceph_decode_32(p);
f24e9980
SW
468 map->pg_pool = kcalloc(map->num_pools, sizeof(*map->pg_pool),
469 GFP_NOFS);
470 if (!map->pg_pool) {
471 err = -ENOMEM;
472 goto bad;
473 }
474 ceph_decode_32_safe(p, end, max, bad);
475 while (max--) {
361be860 476 ceph_decode_need(p, end, 4+1+sizeof(map->pg_pool->v), bad);
c89136ea 477 i = ceph_decode_32(p);
f24e9980
SW
478 if (i >= map->num_pools)
479 goto bad;
361be860 480 ev = ceph_decode_8(p); /* encoding version */
02f90c61
SW
481 if (ev > CEPH_PG_POOL_VERSION) {
482 pr_warning("got unknown v %d > %d of ceph_pg_pool\n",
483 ev, CEPH_PG_POOL_VERSION);
484 goto bad;
485 }
f24e9980
SW
486 ceph_decode_copy(p, &map->pg_pool[i].v,
487 sizeof(map->pg_pool->v));
488 calc_pg_masks(&map->pg_pool[i]);
489 p += le32_to_cpu(map->pg_pool[i].v.num_snaps) * sizeof(u64);
490 p += le32_to_cpu(map->pg_pool[i].v.num_removed_snap_intervals)
491 * sizeof(u64) * 2;
492 }
493
494 ceph_decode_32_safe(p, end, map->flags, bad);
495
c89136ea 496 max = ceph_decode_32(p);
f24e9980
SW
497
498 /* (re)alloc osd arrays */
499 err = osdmap_set_max_osd(map, max);
500 if (err < 0)
501 goto bad;
502 dout("osdmap_decode max_osd = %d\n", map->max_osd);
503
504 /* osds */
505 err = -EINVAL;
506 ceph_decode_need(p, end, 3*sizeof(u32) +
507 map->max_osd*(1 + sizeof(*map->osd_weight) +
508 sizeof(*map->osd_addr)), bad);
509 *p += 4; /* skip length field (should match max) */
510 ceph_decode_copy(p, map->osd_state, map->max_osd);
511
512 *p += 4; /* skip length field (should match max) */
513 for (i = 0; i < map->max_osd; i++)
c89136ea 514 map->osd_weight[i] = ceph_decode_32(p);
f24e9980
SW
515
516 *p += 4; /* skip length field (should match max) */
517 ceph_decode_copy(p, map->osd_addr, map->max_osd*sizeof(*map->osd_addr));
63f2d211
SW
518 for (i = 0; i < map->max_osd; i++)
519 ceph_decode_addr(&map->osd_addr[i]);
f24e9980
SW
520
521 /* pg_temp */
522 ceph_decode_32_safe(p, end, len, bad);
523 for (i = 0; i < len; i++) {
524 int n, j;
51042122 525 struct ceph_pg pgid;
f24e9980
SW
526 struct ceph_pg_mapping *pg;
527
528 ceph_decode_need(p, end, sizeof(u32) + sizeof(u64), bad);
51042122 529 ceph_decode_copy(p, &pgid, sizeof(pgid));
c89136ea 530 n = ceph_decode_32(p);
f24e9980 531 ceph_decode_need(p, end, n * sizeof(u32), bad);
30dc6381 532 err = -ENOMEM;
f24e9980 533 pg = kmalloc(sizeof(*pg) + n*sizeof(u32), GFP_NOFS);
30dc6381 534 if (!pg)
f24e9980 535 goto bad;
f24e9980
SW
536 pg->pgid = pgid;
537 pg->len = n;
538 for (j = 0; j < n; j++)
c89136ea 539 pg->osds[j] = ceph_decode_32(p);
f24e9980 540
991abb6e
SW
541 err = __insert_pg_mapping(pg, &map->pg_temp);
542 if (err)
543 goto bad;
51042122 544 dout(" added pg_temp %llx len %d\n", *(u64 *)&pgid, len);
f24e9980
SW
545 }
546
547 /* crush */
548 ceph_decode_32_safe(p, end, len, bad);
549 dout("osdmap_decode crush len %d from off 0x%x\n", len,
550 (int)(*p - start));
551 ceph_decode_need(p, end, len, bad);
552 map->crush = crush_decode(*p, end);
553 *p += len;
554 if (IS_ERR(map->crush)) {
555 err = PTR_ERR(map->crush);
556 map->crush = NULL;
557 goto bad;
558 }
559
560 /* ignore the rest of the map */
561 *p = end;
562
563 dout("osdmap_decode done %p %p\n", *p, end);
564 return map;
565
566bad:
567 dout("osdmap_decode fail\n");
568 ceph_osdmap_destroy(map);
569 return ERR_PTR(err);
570}
571
572/*
573 * decode and apply an incremental map update.
574 */
575struct ceph_osdmap *osdmap_apply_incremental(void **p, void *end,
576 struct ceph_osdmap *map,
577 struct ceph_messenger *msgr)
578{
f24e9980
SW
579 struct crush_map *newcrush = NULL;
580 struct ceph_fsid fsid;
581 u32 epoch = 0;
582 struct ceph_timespec modified;
583 u32 len, pool;
584 __s32 new_flags, max;
585 void *start = *p;
586 int err = -EINVAL;
587 u16 version;
588 struct rb_node *rbp;
589
590 ceph_decode_16_safe(p, end, version, bad);
02f90c61
SW
591 if (version > CEPH_OSDMAP_INC_VERSION) {
592 pr_warning("got unknown v %d > %d of inc osdmap\n", version,
593 CEPH_OSDMAP_INC_VERSION);
594 goto bad;
595 }
f24e9980
SW
596
597 ceph_decode_need(p, end, sizeof(fsid)+sizeof(modified)+2*sizeof(u32),
598 bad);
599 ceph_decode_copy(p, &fsid, sizeof(fsid));
c89136ea 600 epoch = ceph_decode_32(p);
f24e9980
SW
601 BUG_ON(epoch != map->epoch+1);
602 ceph_decode_copy(p, &modified, sizeof(modified));
c89136ea 603 new_flags = ceph_decode_32(p);
f24e9980
SW
604
605 /* full map? */
606 ceph_decode_32_safe(p, end, len, bad);
607 if (len > 0) {
608 dout("apply_incremental full map len %d, %p to %p\n",
609 len, *p, end);
30dc6381 610 return osdmap_decode(p, min(*p+len, end));
f24e9980
SW
611 }
612
613 /* new crush? */
614 ceph_decode_32_safe(p, end, len, bad);
615 if (len > 0) {
616 dout("apply_incremental new crush map len %d, %p to %p\n",
617 len, *p, end);
618 newcrush = crush_decode(*p, min(*p+len, end));
619 if (IS_ERR(newcrush))
620 return ERR_PTR(PTR_ERR(newcrush));
621 }
622
623 /* new flags? */
624 if (new_flags >= 0)
625 map->flags = new_flags;
626
627 ceph_decode_need(p, end, 5*sizeof(u32), bad);
628
629 /* new max? */
c89136ea 630 max = ceph_decode_32(p);
f24e9980
SW
631 if (max >= 0) {
632 err = osdmap_set_max_osd(map, max);
633 if (err < 0)
634 goto bad;
635 }
636
637 map->epoch++;
638 map->modified = map->modified;
639 if (newcrush) {
640 if (map->crush)
641 crush_destroy(map->crush);
642 map->crush = newcrush;
643 newcrush = NULL;
644 }
645
646 /* new_pool */
647 ceph_decode_32_safe(p, end, len, bad);
648 while (len--) {
361be860
SW
649 __u8 ev;
650
f24e9980
SW
651 ceph_decode_32_safe(p, end, pool, bad);
652 if (pool >= map->num_pools) {
653 void *pg_pool = kcalloc(pool + 1,
654 sizeof(*map->pg_pool),
655 GFP_NOFS);
656 if (!pg_pool) {
657 err = -ENOMEM;
658 goto bad;
659 }
660 memcpy(pg_pool, map->pg_pool,
661 map->num_pools * sizeof(*map->pg_pool));
662 kfree(map->pg_pool);
663 map->pg_pool = pg_pool;
664 map->num_pools = pool+1;
665 }
361be860
SW
666 ceph_decode_need(p, end, 1 + sizeof(map->pg_pool->v), bad);
667 ev = ceph_decode_8(p); /* encoding version */
02f90c61
SW
668 if (ev > CEPH_PG_POOL_VERSION) {
669 pr_warning("got unknown v %d > %d of ceph_pg_pool\n",
670 ev, CEPH_PG_POOL_VERSION);
671 goto bad;
672 }
f24e9980
SW
673 ceph_decode_copy(p, &map->pg_pool[pool].v,
674 sizeof(map->pg_pool->v));
675 calc_pg_masks(&map->pg_pool[pool]);
676 }
677
678 /* old_pool (ignore) */
679 ceph_decode_32_safe(p, end, len, bad);
680 *p += len * sizeof(u32);
681
682 /* new_up */
683 err = -EINVAL;
684 ceph_decode_32_safe(p, end, len, bad);
685 while (len--) {
686 u32 osd;
687 struct ceph_entity_addr addr;
688 ceph_decode_32_safe(p, end, osd, bad);
689 ceph_decode_copy_safe(p, end, &addr, sizeof(addr), bad);
63f2d211 690 ceph_decode_addr(&addr);
f24e9980
SW
691 pr_info("osd%d up\n", osd);
692 BUG_ON(osd >= map->max_osd);
693 map->osd_state[osd] |= CEPH_OSD_UP;
694 map->osd_addr[osd] = addr;
695 }
696
697 /* new_down */
698 ceph_decode_32_safe(p, end, len, bad);
699 while (len--) {
700 u32 osd;
701 ceph_decode_32_safe(p, end, osd, bad);
702 (*p)++; /* clean flag */
1bdb70e5 703 pr_info("osd%d down\n", osd);
f24e9980
SW
704 if (osd < map->max_osd)
705 map->osd_state[osd] &= ~CEPH_OSD_UP;
706 }
707
708 /* new_weight */
709 ceph_decode_32_safe(p, end, len, bad);
710 while (len--) {
711 u32 osd, off;
712 ceph_decode_need(p, end, sizeof(u32)*2, bad);
c89136ea
SW
713 osd = ceph_decode_32(p);
714 off = ceph_decode_32(p);
f24e9980
SW
715 pr_info("osd%d weight 0x%x %s\n", osd, off,
716 off == CEPH_OSD_IN ? "(in)" :
717 (off == CEPH_OSD_OUT ? "(out)" : ""));
718 if (osd < map->max_osd)
719 map->osd_weight[osd] = off;
720 }
721
722 /* new_pg_temp */
723 rbp = rb_first(&map->pg_temp);
724 ceph_decode_32_safe(p, end, len, bad);
725 while (len--) {
726 struct ceph_pg_mapping *pg;
727 int j;
51042122 728 struct ceph_pg pgid;
f24e9980
SW
729 u32 pglen;
730 ceph_decode_need(p, end, sizeof(u64) + sizeof(u32), bad);
51042122 731 ceph_decode_copy(p, &pgid, sizeof(pgid));
c89136ea 732 pglen = ceph_decode_32(p);
f24e9980
SW
733
734 /* remove any? */
51042122
SW
735 while (rbp && pgid_cmp(rb_entry(rbp, struct ceph_pg_mapping,
736 node)->pgid, pgid) <= 0) {
f24e9980
SW
737 struct rb_node *cur = rbp;
738 rbp = rb_next(rbp);
739 dout(" removed pg_temp %llx\n",
51042122
SW
740 *(u64 *)&rb_entry(cur, struct ceph_pg_mapping,
741 node)->pgid);
f24e9980
SW
742 rb_erase(cur, &map->pg_temp);
743 }
744
745 if (pglen) {
746 /* insert */
747 ceph_decode_need(p, end, pglen*sizeof(u32), bad);
748 pg = kmalloc(sizeof(*pg) + sizeof(u32)*pglen, GFP_NOFS);
749 if (!pg) {
750 err = -ENOMEM;
751 goto bad;
752 }
753 pg->pgid = pgid;
754 pg->len = pglen;
7067f797 755 for (j = 0; j < pglen; j++)
c89136ea 756 pg->osds[j] = ceph_decode_32(p);
991abb6e
SW
757 err = __insert_pg_mapping(pg, &map->pg_temp);
758 if (err)
759 goto bad;
51042122
SW
760 dout(" added pg_temp %llx len %d\n", *(u64 *)&pgid,
761 pglen);
f24e9980
SW
762 }
763 }
764 while (rbp) {
765 struct rb_node *cur = rbp;
766 rbp = rb_next(rbp);
767 dout(" removed pg_temp %llx\n",
51042122
SW
768 *(u64 *)&rb_entry(cur, struct ceph_pg_mapping,
769 node)->pgid);
f24e9980
SW
770 rb_erase(cur, &map->pg_temp);
771 }
772
773 /* ignore the rest */
774 *p = end;
775 return map;
776
777bad:
778 pr_err("corrupt inc osdmap epoch %d off %d (%p of %p-%p)\n",
779 epoch, (int)(*p - start), *p, start, end);
9ec7cab1
SW
780 print_hex_dump(KERN_DEBUG, "osdmap: ",
781 DUMP_PREFIX_OFFSET, 16, 1,
782 start, end - start, true);
f24e9980
SW
783 if (newcrush)
784 crush_destroy(newcrush);
785 return ERR_PTR(err);
786}
787
788
789
790
791/*
792 * calculate file layout from given offset, length.
793 * fill in correct oid, logical length, and object extent
794 * offset, length.
795 *
796 * for now, we write only a single su, until we can
797 * pass a stride back to the caller.
798 */
799void ceph_calc_file_object_mapping(struct ceph_file_layout *layout,
800 u64 off, u64 *plen,
645a1025 801 u64 *ono,
f24e9980
SW
802 u64 *oxoff, u64 *oxlen)
803{
804 u32 osize = le32_to_cpu(layout->fl_object_size);
805 u32 su = le32_to_cpu(layout->fl_stripe_unit);
806 u32 sc = le32_to_cpu(layout->fl_stripe_count);
807 u32 bl, stripeno, stripepos, objsetno;
808 u32 su_per_object;
ff1d1f71 809 u64 t, su_offset;
f24e9980
SW
810
811 dout("mapping %llu~%llu osize %u fl_su %u\n", off, *plen,
812 osize, su);
35e054a6 813 su_per_object = osize / su;
f24e9980
SW
814 dout("osize %u / su %u = su_per_object %u\n", osize, su,
815 su_per_object);
816
817 BUG_ON((su & ~PAGE_MASK) != 0);
818 /* bl = *off / su; */
819 t = off;
820 do_div(t, su);
821 bl = t;
822 dout("off %llu / su %u = bl %u\n", off, su, bl);
823
824 stripeno = bl / sc;
825 stripepos = bl % sc;
826 objsetno = stripeno / su_per_object;
827
645a1025
SW
828 *ono = objsetno * sc + stripepos;
829 dout("objset %u * sc %u = ono %u\n", objsetno, sc, (unsigned)*ono);
830
831 /* *oxoff = *off % layout->fl_stripe_unit; # offset in su */
f24e9980 832 t = off;
ff1d1f71
NW
833 su_offset = do_div(t, su);
834 *oxoff = su_offset + (stripeno % su_per_object) * su;
835
836 /*
837 * Calculate the length of the extent being written to the selected
838 * object. This is the minimum of the full length requested (plen) or
839 * the remainder of the current stripe being written to.
840 */
841 *oxlen = min_t(u64, *plen, su - su_offset);
f24e9980
SW
842 *plen = *oxlen;
843
844 dout(" obj extent %llu~%llu\n", *oxoff, *oxlen);
845}
846
847/*
848 * calculate an object layout (i.e. pgid) from an oid,
849 * file_layout, and osdmap
850 */
851int ceph_calc_object_layout(struct ceph_object_layout *ol,
852 const char *oid,
853 struct ceph_file_layout *fl,
854 struct ceph_osdmap *osdmap)
855{
856 unsigned num, num_mask;
51042122 857 struct ceph_pg pgid;
f24e9980
SW
858 s32 preferred = (s32)le32_to_cpu(fl->fl_pg_preferred);
859 int poolid = le32_to_cpu(fl->fl_pg_pool);
860 struct ceph_pg_pool_info *pool;
51042122 861 unsigned ps;
f24e9980 862
30dc6381 863 BUG_ON(!osdmap);
f24e9980
SW
864 if (poolid >= osdmap->num_pools)
865 return -EIO;
f24e9980 866
51042122 867 pool = &osdmap->pg_pool[poolid];
1654dd0c 868 ps = ceph_str_hash(pool->v.object_hash, oid, strlen(oid));
f24e9980 869 if (preferred >= 0) {
51042122 870 ps += preferred;
f24e9980
SW
871 num = le32_to_cpu(pool->v.lpg_num);
872 num_mask = pool->lpg_num_mask;
873 } else {
874 num = le32_to_cpu(pool->v.pg_num);
875 num_mask = pool->pg_num_mask;
876 }
877
51042122
SW
878 pgid.ps = cpu_to_le16(ps);
879 pgid.preferred = cpu_to_le16(preferred);
880 pgid.pool = fl->fl_pg_pool;
f24e9980 881 if (preferred >= 0)
51042122
SW
882 dout("calc_object_layout '%s' pgid %d.%xp%d\n", oid, poolid, ps,
883 (int)preferred);
f24e9980 884 else
51042122 885 dout("calc_object_layout '%s' pgid %d.%x\n", oid, poolid, ps);
f24e9980 886
51042122 887 ol->ol_pgid = pgid;
f24e9980 888 ol->ol_stripe_unit = fl->fl_object_stripe_unit;
f24e9980
SW
889 return 0;
890}
891
892/*
893 * Calculate raw osd vector for the given pgid. Return pointer to osd
894 * array, or NULL on failure.
895 */
51042122 896static int *calc_pg_raw(struct ceph_osdmap *osdmap, struct ceph_pg pgid,
f24e9980
SW
897 int *osds, int *num)
898{
f24e9980
SW
899 struct ceph_pg_mapping *pg;
900 struct ceph_pg_pool_info *pool;
901 int ruleno;
51042122
SW
902 unsigned poolid, ps, pps;
903 int preferred;
f24e9980
SW
904
905 /* pg_temp? */
9794b146
SW
906 pg = __lookup_pg_mapping(&osdmap->pg_temp, pgid);
907 if (pg) {
908 *num = pg->len;
909 return pg->osds;
f24e9980
SW
910 }
911
912 /* crush */
51042122
SW
913 poolid = le32_to_cpu(pgid.pool);
914 ps = le16_to_cpu(pgid.ps);
915 preferred = (s16)le16_to_cpu(pgid.preferred);
916
767ea5c3
SW
917 /* don't forcefeed bad device ids to crush */
918 if (preferred >= osdmap->max_osd ||
919 preferred >= osdmap->crush->max_devices)
920 preferred = -1;
921
51042122 922 if (poolid >= osdmap->num_pools)
f24e9980 923 return NULL;
51042122 924 pool = &osdmap->pg_pool[poolid];
f24e9980
SW
925 ruleno = crush_find_rule(osdmap->crush, pool->v.crush_ruleset,
926 pool->v.type, pool->v.size);
927 if (ruleno < 0) {
928 pr_err("no crush rule pool %d type %d size %d\n",
51042122 929 poolid, pool->v.type, pool->v.size);
f24e9980
SW
930 return NULL;
931 }
932
51042122
SW
933 if (preferred >= 0)
934 pps = ceph_stable_mod(ps,
f24e9980
SW
935 le32_to_cpu(pool->v.lpgp_num),
936 pool->lpgp_num_mask);
937 else
51042122 938 pps = ceph_stable_mod(ps,
f24e9980
SW
939 le32_to_cpu(pool->v.pgp_num),
940 pool->pgp_num_mask);
51042122 941 pps += poolid;
f24e9980
SW
942 *num = crush_do_rule(osdmap->crush, ruleno, pps, osds,
943 min_t(int, pool->v.size, *num),
51042122 944 preferred, osdmap->osd_weight);
f24e9980
SW
945 return osds;
946}
947
948/*
949 * Return primary osd for given pgid, or -1 if none.
950 */
51042122 951int ceph_calc_pg_primary(struct ceph_osdmap *osdmap, struct ceph_pg pgid)
f24e9980
SW
952{
953 int rawosds[10], *osds;
954 int i, num = ARRAY_SIZE(rawosds);
955
956 osds = calc_pg_raw(osdmap, pgid, rawosds, &num);
957 if (!osds)
958 return -1;
959
960 /* primary is first up osd */
961 for (i = 0; i < num; i++)
962 if (ceph_osd_is_up(osdmap, osds[i])) {
963 return osds[i];
964 break;
965 }
966 return -1;
967}