1
2 #include <linux/ceph/ceph_debug.h>
3
4 #include <linux/module.h>
5 #include <linux/slab.h>
6 #include <asm/div64.h>
7
8 #include <linux/ceph/libceph.h>
9 #include <linux/ceph/osdmap.h>
10 #include <linux/ceph/decode.h>
11 #include <linux/crush/hash.h>
12 #include <linux/crush/mapper.h>
13
ceph_osdmap_state_str(char * str,int len,int state)14 char *ceph_osdmap_state_str(char *str, int len, int state)
15 {
16 int flag = 0;
17
18 if (!len)
19 goto done;
20
21 *str = '\0';
22 if (state) {
23 if (state & CEPH_OSD_EXISTS) {
24 snprintf(str, len, "exists");
25 flag = 1;
26 }
27 if (state & CEPH_OSD_UP) {
28 snprintf(str, len, "%s%s%s", str, (flag ? ", " : ""),
29 "up");
30 flag = 1;
31 }
32 } else {
33 snprintf(str, len, "doesn't exist");
34 }
35 done:
36 return str;
37 }
38
39 /* maps */
40
calc_bits_of(unsigned t)41 static int calc_bits_of(unsigned t)
42 {
43 int b = 0;
44 while (t) {
45 t = t >> 1;
46 b++;
47 }
48 return b;
49 }
50
51 /*
52 * the foo_mask is the smallest value 2^n-1 that is >= foo.
53 */
calc_pg_masks(struct ceph_pg_pool_info * pi)54 static void calc_pg_masks(struct ceph_pg_pool_info *pi)
55 {
56 pi->pg_num_mask = (1 << calc_bits_of(le32_to_cpu(pi->v.pg_num)-1)) - 1;
57 pi->pgp_num_mask =
58 (1 << calc_bits_of(le32_to_cpu(pi->v.pgp_num)-1)) - 1;
59 pi->lpg_num_mask =
60 (1 << calc_bits_of(le32_to_cpu(pi->v.lpg_num)-1)) - 1;
61 pi->lpgp_num_mask =
62 (1 << calc_bits_of(le32_to_cpu(pi->v.lpgp_num)-1)) - 1;
63 }
64
65 /*
66 * decode crush map
67 */
crush_decode_uniform_bucket(void ** p,void * end,struct crush_bucket_uniform * b)68 static int crush_decode_uniform_bucket(void **p, void *end,
69 struct crush_bucket_uniform *b)
70 {
71 dout("crush_decode_uniform_bucket %p to %p\n", *p, end);
72 ceph_decode_need(p, end, (1+b->h.size) * sizeof(u32), bad);
73 b->item_weight = ceph_decode_32(p);
74 return 0;
75 bad:
76 return -EINVAL;
77 }
78
crush_decode_list_bucket(void ** p,void * end,struct crush_bucket_list * b)79 static int crush_decode_list_bucket(void **p, void *end,
80 struct crush_bucket_list *b)
81 {
82 int j;
83 dout("crush_decode_list_bucket %p to %p\n", *p, end);
84 b->item_weights = kcalloc(b->h.size, sizeof(u32), GFP_NOFS);
85 if (b->item_weights == NULL)
86 return -ENOMEM;
87 b->sum_weights = kcalloc(b->h.size, sizeof(u32), GFP_NOFS);
88 if (b->sum_weights == NULL)
89 return -ENOMEM;
90 ceph_decode_need(p, end, 2 * b->h.size * sizeof(u32), bad);
91 for (j = 0; j < b->h.size; j++) {
92 b->item_weights[j] = ceph_decode_32(p);
93 b->sum_weights[j] = ceph_decode_32(p);
94 }
95 return 0;
96 bad:
97 return -EINVAL;
98 }
99
crush_decode_tree_bucket(void ** p,void * end,struct crush_bucket_tree * b)100 static int crush_decode_tree_bucket(void **p, void *end,
101 struct crush_bucket_tree *b)
102 {
103 int j;
104 dout("crush_decode_tree_bucket %p to %p\n", *p, end);
105 ceph_decode_32_safe(p, end, b->num_nodes, bad);
106 b->node_weights = kcalloc(b->num_nodes, sizeof(u32), GFP_NOFS);
107 if (b->node_weights == NULL)
108 return -ENOMEM;
109 ceph_decode_need(p, end, b->num_nodes * sizeof(u32), bad);
110 for (j = 0; j < b->num_nodes; j++)
111 b->node_weights[j] = ceph_decode_32(p);
112 return 0;
113 bad:
114 return -EINVAL;
115 }
116
crush_decode_straw_bucket(void ** p,void * end,struct crush_bucket_straw * b)117 static int crush_decode_straw_bucket(void **p, void *end,
118 struct crush_bucket_straw *b)
119 {
120 int j;
121 dout("crush_decode_straw_bucket %p to %p\n", *p, end);
122 b->item_weights = kcalloc(b->h.size, sizeof(u32), GFP_NOFS);
123 if (b->item_weights == NULL)
124 return -ENOMEM;
125 b->straws = kcalloc(b->h.size, sizeof(u32), GFP_NOFS);
126 if (b->straws == NULL)
127 return -ENOMEM;
128 ceph_decode_need(p, end, 2 * b->h.size * sizeof(u32), bad);
129 for (j = 0; j < b->h.size; j++) {
130 b->item_weights[j] = ceph_decode_32(p);
131 b->straws[j] = ceph_decode_32(p);
132 }
133 return 0;
134 bad:
135 return -EINVAL;
136 }
137
crush_decode(void * pbyval,void * end)138 static struct crush_map *crush_decode(void *pbyval, void *end)
139 {
140 struct crush_map *c;
141 int err = -EINVAL;
142 int i, j;
143 void **p = &pbyval;
144 void *start = pbyval;
145 u32 magic;
146
147 dout("crush_decode %p to %p len %d\n", *p, end, (int)(end - *p));
148
149 c = kzalloc(sizeof(*c), GFP_NOFS);
150 if (c == NULL)
151 return ERR_PTR(-ENOMEM);
152
153 ceph_decode_need(p, end, 4*sizeof(u32), bad);
154 magic = ceph_decode_32(p);
155 if (magic != CRUSH_MAGIC) {
156 pr_err("crush_decode magic %x != current %x\n",
157 (unsigned)magic, (unsigned)CRUSH_MAGIC);
158 goto bad;
159 }
160 c->max_buckets = ceph_decode_32(p);
161 c->max_rules = ceph_decode_32(p);
162 c->max_devices = ceph_decode_32(p);
163
164 c->device_parents = kcalloc(c->max_devices, sizeof(u32), GFP_NOFS);
165 if (c->device_parents == NULL)
166 goto badmem;
167 c->bucket_parents = kcalloc(c->max_buckets, sizeof(u32), GFP_NOFS);
168 if (c->bucket_parents == NULL)
169 goto badmem;
170
171 c->buckets = kcalloc(c->max_buckets, sizeof(*c->buckets), GFP_NOFS);
172 if (c->buckets == NULL)
173 goto badmem;
174 c->rules = kcalloc(c->max_rules, sizeof(*c->rules), GFP_NOFS);
175 if (c->rules == NULL)
176 goto badmem;
177
178 /* buckets */
179 for (i = 0; i < c->max_buckets; i++) {
180 int size = 0;
181 u32 alg;
182 struct crush_bucket *b;
183
184 ceph_decode_32_safe(p, end, alg, bad);
185 if (alg == 0) {
186 c->buckets[i] = NULL;
187 continue;
188 }
189 dout("crush_decode bucket %d off %x %p to %p\n",
190 i, (int)(*p-start), *p, end);
191
192 switch (alg) {
193 case CRUSH_BUCKET_UNIFORM:
194 size = sizeof(struct crush_bucket_uniform);
195 break;
196 case CRUSH_BUCKET_LIST:
197 size = sizeof(struct crush_bucket_list);
198 break;
199 case CRUSH_BUCKET_TREE:
200 size = sizeof(struct crush_bucket_tree);
201 break;
202 case CRUSH_BUCKET_STRAW:
203 size = sizeof(struct crush_bucket_straw);
204 break;
205 default:
206 err = -EINVAL;
207 goto bad;
208 }
209 BUG_ON(size == 0);
210 b = c->buckets[i] = kzalloc(size, GFP_NOFS);
211 if (b == NULL)
212 goto badmem;
213
214 ceph_decode_need(p, end, 4*sizeof(u32), bad);
215 b->id = ceph_decode_32(p);
216 b->type = ceph_decode_16(p);
217 b->alg = ceph_decode_8(p);
218 b->hash = ceph_decode_8(p);
219 b->weight = ceph_decode_32(p);
220 b->size = ceph_decode_32(p);
221
222 dout("crush_decode bucket size %d off %x %p to %p\n",
223 b->size, (int)(*p-start), *p, end);
224
225 b->items = kcalloc(b->size, sizeof(__s32), GFP_NOFS);
226 if (b->items == NULL)
227 goto badmem;
228 b->perm = kcalloc(b->size, sizeof(u32), GFP_NOFS);
229 if (b->perm == NULL)
230 goto badmem;
231 b->perm_n = 0;
232
233 ceph_decode_need(p, end, b->size*sizeof(u32), bad);
234 for (j = 0; j < b->size; j++)
235 b->items[j] = ceph_decode_32(p);
236
237 switch (b->alg) {
238 case CRUSH_BUCKET_UNIFORM:
239 err = crush_decode_uniform_bucket(p, end,
240 (struct crush_bucket_uniform *)b);
241 if (err < 0)
242 goto bad;
243 break;
244 case CRUSH_BUCKET_LIST:
245 err = crush_decode_list_bucket(p, end,
246 (struct crush_bucket_list *)b);
247 if (err < 0)
248 goto bad;
249 break;
250 case CRUSH_BUCKET_TREE:
251 err = crush_decode_tree_bucket(p, end,
252 (struct crush_bucket_tree *)b);
253 if (err < 0)
254 goto bad;
255 break;
256 case CRUSH_BUCKET_STRAW:
257 err = crush_decode_straw_bucket(p, end,
258 (struct crush_bucket_straw *)b);
259 if (err < 0)
260 goto bad;
261 break;
262 }
263 }
264
265 /* rules */
266 dout("rule vec is %p\n", c->rules);
267 for (i = 0; i < c->max_rules; i++) {
268 u32 yes;
269 struct crush_rule *r;
270
271 ceph_decode_32_safe(p, end, yes, bad);
272 if (!yes) {
273 dout("crush_decode NO rule %d off %x %p to %p\n",
274 i, (int)(*p-start), *p, end);
275 c->rules[i] = NULL;
276 continue;
277 }
278
279 dout("crush_decode rule %d off %x %p to %p\n",
280 i, (int)(*p-start), *p, end);
281
282 /* len */
283 ceph_decode_32_safe(p, end, yes, bad);
284 #if BITS_PER_LONG == 32
285 err = -EINVAL;
286 if (yes > (ULONG_MAX - sizeof(*r))
287 / sizeof(struct crush_rule_step))
288 goto bad;
289 #endif
290 r = c->rules[i] = kmalloc(sizeof(*r) +
291 yes*sizeof(struct crush_rule_step),
292 GFP_NOFS);
293 if (r == NULL)
294 goto badmem;
295 dout(" rule %d is at %p\n", i, r);
296 r->len = yes;
297 ceph_decode_copy_safe(p, end, &r->mask, 4, bad); /* 4 u8's */
298 ceph_decode_need(p, end, r->len*3*sizeof(u32), bad);
299 for (j = 0; j < r->len; j++) {
300 r->steps[j].op = ceph_decode_32(p);
301 r->steps[j].arg1 = ceph_decode_32(p);
302 r->steps[j].arg2 = ceph_decode_32(p);
303 }
304 }
305
306 /* ignore trailing name maps. */
307
308 dout("crush_decode success\n");
309 return c;
310
311 badmem:
312 err = -ENOMEM;
313 bad:
314 dout("crush_decode fail %d\n", err);
315 crush_destroy(c);
316 return ERR_PTR(err);
317 }
318
319 /*
320 * rbtree of pg_mapping for handling pg_temp (explicit mapping of pgid
321 * to a set of osds)
322 */
pgid_cmp(struct ceph_pg l,struct ceph_pg r)323 static int pgid_cmp(struct ceph_pg l, struct ceph_pg r)
324 {
325 u64 a = *(u64 *)&l;
326 u64 b = *(u64 *)&r;
327
328 if (a < b)
329 return -1;
330 if (a > b)
331 return 1;
332 return 0;
333 }
334
__insert_pg_mapping(struct ceph_pg_mapping * new,struct rb_root * root)335 static int __insert_pg_mapping(struct ceph_pg_mapping *new,
336 struct rb_root *root)
337 {
338 struct rb_node **p = &root->rb_node;
339 struct rb_node *parent = NULL;
340 struct ceph_pg_mapping *pg = NULL;
341 int c;
342
343 dout("__insert_pg_mapping %llx %p\n", *(u64 *)&new->pgid, new);
344 while (*p) {
345 parent = *p;
346 pg = rb_entry(parent, struct ceph_pg_mapping, node);
347 c = pgid_cmp(new->pgid, pg->pgid);
348 if (c < 0)
349 p = &(*p)->rb_left;
350 else if (c > 0)
351 p = &(*p)->rb_right;
352 else
353 return -EEXIST;
354 }
355
356 rb_link_node(&new->node, parent, p);
357 rb_insert_color(&new->node, root);
358 return 0;
359 }
360
__lookup_pg_mapping(struct rb_root * root,struct ceph_pg pgid)361 static struct ceph_pg_mapping *__lookup_pg_mapping(struct rb_root *root,
362 struct ceph_pg pgid)
363 {
364 struct rb_node *n = root->rb_node;
365 struct ceph_pg_mapping *pg;
366 int c;
367
368 while (n) {
369 pg = rb_entry(n, struct ceph_pg_mapping, node);
370 c = pgid_cmp(pgid, pg->pgid);
371 if (c < 0) {
372 n = n->rb_left;
373 } else if (c > 0) {
374 n = n->rb_right;
375 } else {
376 dout("__lookup_pg_mapping %llx got %p\n",
377 *(u64 *)&pgid, pg);
378 return pg;
379 }
380 }
381 return NULL;
382 }
383
__remove_pg_mapping(struct rb_root * root,struct ceph_pg pgid)384 static int __remove_pg_mapping(struct rb_root *root, struct ceph_pg pgid)
385 {
386 struct ceph_pg_mapping *pg = __lookup_pg_mapping(root, pgid);
387
388 if (pg) {
389 dout("__remove_pg_mapping %llx %p\n", *(u64 *)&pgid, pg);
390 rb_erase(&pg->node, root);
391 kfree(pg);
392 return 0;
393 }
394 dout("__remove_pg_mapping %llx dne\n", *(u64 *)&pgid);
395 return -ENOENT;
396 }
397
398 /*
399 * rbtree of pg pool info
400 */
__insert_pg_pool(struct rb_root * root,struct ceph_pg_pool_info * new)401 static int __insert_pg_pool(struct rb_root *root, struct ceph_pg_pool_info *new)
402 {
403 struct rb_node **p = &root->rb_node;
404 struct rb_node *parent = NULL;
405 struct ceph_pg_pool_info *pi = NULL;
406
407 while (*p) {
408 parent = *p;
409 pi = rb_entry(parent, struct ceph_pg_pool_info, node);
410 if (new->id < pi->id)
411 p = &(*p)->rb_left;
412 else if (new->id > pi->id)
413 p = &(*p)->rb_right;
414 else
415 return -EEXIST;
416 }
417
418 rb_link_node(&new->node, parent, p);
419 rb_insert_color(&new->node, root);
420 return 0;
421 }
422
__lookup_pg_pool(struct rb_root * root,int id)423 static struct ceph_pg_pool_info *__lookup_pg_pool(struct rb_root *root, int id)
424 {
425 struct ceph_pg_pool_info *pi;
426 struct rb_node *n = root->rb_node;
427
428 while (n) {
429 pi = rb_entry(n, struct ceph_pg_pool_info, node);
430 if (id < pi->id)
431 n = n->rb_left;
432 else if (id > pi->id)
433 n = n->rb_right;
434 else
435 return pi;
436 }
437 return NULL;
438 }
439
ceph_pg_poolid_by_name(struct ceph_osdmap * map,const char * name)440 int ceph_pg_poolid_by_name(struct ceph_osdmap *map, const char *name)
441 {
442 struct rb_node *rbp;
443
444 for (rbp = rb_first(&map->pg_pools); rbp; rbp = rb_next(rbp)) {
445 struct ceph_pg_pool_info *pi =
446 rb_entry(rbp, struct ceph_pg_pool_info, node);
447 if (pi->name && strcmp(pi->name, name) == 0)
448 return pi->id;
449 }
450 return -ENOENT;
451 }
452 EXPORT_SYMBOL(ceph_pg_poolid_by_name);
453
__remove_pg_pool(struct rb_root * root,struct ceph_pg_pool_info * pi)454 static void __remove_pg_pool(struct rb_root *root, struct ceph_pg_pool_info *pi)
455 {
456 rb_erase(&pi->node, root);
457 kfree(pi->name);
458 kfree(pi);
459 }
460
__decode_pool(void ** p,void * end,struct ceph_pg_pool_info * pi)461 static int __decode_pool(void **p, void *end, struct ceph_pg_pool_info *pi)
462 {
463 unsigned n, m;
464
465 ceph_decode_copy(p, &pi->v, sizeof(pi->v));
466 calc_pg_masks(pi);
467
468 /* num_snaps * snap_info_t */
469 n = le32_to_cpu(pi->v.num_snaps);
470 while (n--) {
471 ceph_decode_need(p, end, sizeof(u64) + 1 + sizeof(u64) +
472 sizeof(struct ceph_timespec), bad);
473 *p += sizeof(u64) + /* key */
474 1 + sizeof(u64) + /* u8, snapid */
475 sizeof(struct ceph_timespec);
476 m = ceph_decode_32(p); /* snap name */
477 *p += m;
478 }
479
480 *p += le32_to_cpu(pi->v.num_removed_snap_intervals) * sizeof(u64) * 2;
481 return 0;
482
483 bad:
484 return -EINVAL;
485 }
486
__decode_pool_names(void ** p,void * end,struct ceph_osdmap * map)487 static int __decode_pool_names(void **p, void *end, struct ceph_osdmap *map)
488 {
489 struct ceph_pg_pool_info *pi;
490 u32 num, len, pool;
491
492 ceph_decode_32_safe(p, end, num, bad);
493 dout(" %d pool names\n", num);
494 while (num--) {
495 ceph_decode_32_safe(p, end, pool, bad);
496 ceph_decode_32_safe(p, end, len, bad);
497 dout(" pool %d len %d\n", pool, len);
498 ceph_decode_need(p, end, len, bad);
499 pi = __lookup_pg_pool(&map->pg_pools, pool);
500 if (pi) {
501 char *name = kstrndup(*p, len, GFP_NOFS);
502
503 if (!name)
504 return -ENOMEM;
505 kfree(pi->name);
506 pi->name = name;
507 dout(" name is %s\n", pi->name);
508 }
509 *p += len;
510 }
511 return 0;
512
513 bad:
514 return -EINVAL;
515 }
516
517 /*
518 * osd map
519 */
ceph_osdmap_destroy(struct ceph_osdmap * map)520 void ceph_osdmap_destroy(struct ceph_osdmap *map)
521 {
522 dout("osdmap_destroy %p\n", map);
523 if (map->crush)
524 crush_destroy(map->crush);
525 while (!RB_EMPTY_ROOT(&map->pg_temp)) {
526 struct ceph_pg_mapping *pg =
527 rb_entry(rb_first(&map->pg_temp),
528 struct ceph_pg_mapping, node);
529 rb_erase(&pg->node, &map->pg_temp);
530 kfree(pg);
531 }
532 while (!RB_EMPTY_ROOT(&map->pg_pools)) {
533 struct ceph_pg_pool_info *pi =
534 rb_entry(rb_first(&map->pg_pools),
535 struct ceph_pg_pool_info, node);
536 __remove_pg_pool(&map->pg_pools, pi);
537 }
538 kfree(map->osd_state);
539 kfree(map->osd_weight);
540 kfree(map->osd_addr);
541 kfree(map);
542 }
543
544 /*
545 * adjust max osd value. reallocate arrays.
546 */
osdmap_set_max_osd(struct ceph_osdmap * map,int max)547 static int osdmap_set_max_osd(struct ceph_osdmap *map, int max)
548 {
549 u8 *state;
550 struct ceph_entity_addr *addr;
551 u32 *weight;
552
553 state = kcalloc(max, sizeof(*state), GFP_NOFS);
554 addr = kcalloc(max, sizeof(*addr), GFP_NOFS);
555 weight = kcalloc(max, sizeof(*weight), GFP_NOFS);
556 if (state == NULL || addr == NULL || weight == NULL) {
557 kfree(state);
558 kfree(addr);
559 kfree(weight);
560 return -ENOMEM;
561 }
562
563 /* copy old? */
564 if (map->osd_state) {
565 memcpy(state, map->osd_state, map->max_osd*sizeof(*state));
566 memcpy(addr, map->osd_addr, map->max_osd*sizeof(*addr));
567 memcpy(weight, map->osd_weight, map->max_osd*sizeof(*weight));
568 kfree(map->osd_state);
569 kfree(map->osd_addr);
570 kfree(map->osd_weight);
571 }
572
573 map->osd_state = state;
574 map->osd_weight = weight;
575 map->osd_addr = addr;
576 map->max_osd = max;
577 return 0;
578 }
579
580 /*
581 * decode a full map.
582 */
osdmap_decode(void ** p,void * end)583 struct ceph_osdmap *osdmap_decode(void **p, void *end)
584 {
585 struct ceph_osdmap *map;
586 u16 version;
587 u32 len, max, i;
588 u8 ev;
589 int err = -EINVAL;
590 void *start = *p;
591 struct ceph_pg_pool_info *pi;
592
593 dout("osdmap_decode %p to %p len %d\n", *p, end, (int)(end - *p));
594
595 map = kzalloc(sizeof(*map), GFP_NOFS);
596 if (map == NULL)
597 return ERR_PTR(-ENOMEM);
598 map->pg_temp = RB_ROOT;
599
600 ceph_decode_16_safe(p, end, version, bad);
601 if (version > CEPH_OSDMAP_VERSION) {
602 pr_warning("got unknown v %d > %d of osdmap\n", version,
603 CEPH_OSDMAP_VERSION);
604 goto bad;
605 }
606
607 ceph_decode_need(p, end, 2*sizeof(u64)+6*sizeof(u32), bad);
608 ceph_decode_copy(p, &map->fsid, sizeof(map->fsid));
609 map->epoch = ceph_decode_32(p);
610 ceph_decode_copy(p, &map->created, sizeof(map->created));
611 ceph_decode_copy(p, &map->modified, sizeof(map->modified));
612
613 ceph_decode_32_safe(p, end, max, bad);
614 while (max--) {
615 ceph_decode_need(p, end, 4 + 1 + sizeof(pi->v), bad);
616 err = -ENOMEM;
617 pi = kzalloc(sizeof(*pi), GFP_NOFS);
618 if (!pi)
619 goto bad;
620 pi->id = ceph_decode_32(p);
621 err = -EINVAL;
622 ev = ceph_decode_8(p); /* encoding version */
623 if (ev > CEPH_PG_POOL_VERSION) {
624 pr_warning("got unknown v %d > %d of ceph_pg_pool\n",
625 ev, CEPH_PG_POOL_VERSION);
626 kfree(pi);
627 goto bad;
628 }
629 err = __decode_pool(p, end, pi);
630 if (err < 0) {
631 kfree(pi);
632 goto bad;
633 }
634 __insert_pg_pool(&map->pg_pools, pi);
635 }
636
637 if (version >= 5) {
638 err = __decode_pool_names(p, end, map);
639 if (err < 0) {
640 dout("fail to decode pool names");
641 goto bad;
642 }
643 }
644
645 ceph_decode_32_safe(p, end, map->pool_max, bad);
646
647 ceph_decode_32_safe(p, end, map->flags, bad);
648
649 max = ceph_decode_32(p);
650
651 /* (re)alloc osd arrays */
652 err = osdmap_set_max_osd(map, max);
653 if (err < 0)
654 goto bad;
655 dout("osdmap_decode max_osd = %d\n", map->max_osd);
656
657 /* osds */
658 err = -EINVAL;
659 ceph_decode_need(p, end, 3*sizeof(u32) +
660 map->max_osd*(1 + sizeof(*map->osd_weight) +
661 sizeof(*map->osd_addr)), bad);
662 *p += 4; /* skip length field (should match max) */
663 ceph_decode_copy(p, map->osd_state, map->max_osd);
664
665 *p += 4; /* skip length field (should match max) */
666 for (i = 0; i < map->max_osd; i++)
667 map->osd_weight[i] = ceph_decode_32(p);
668
669 *p += 4; /* skip length field (should match max) */
670 ceph_decode_copy(p, map->osd_addr, map->max_osd*sizeof(*map->osd_addr));
671 for (i = 0; i < map->max_osd; i++)
672 ceph_decode_addr(&map->osd_addr[i]);
673
674 /* pg_temp */
675 ceph_decode_32_safe(p, end, len, bad);
676 for (i = 0; i < len; i++) {
677 int n, j;
678 struct ceph_pg pgid;
679 struct ceph_pg_mapping *pg;
680
681 ceph_decode_need(p, end, sizeof(u32) + sizeof(u64), bad);
682 ceph_decode_copy(p, &pgid, sizeof(pgid));
683 n = ceph_decode_32(p);
684 err = -EINVAL;
685 if (n > (UINT_MAX - sizeof(*pg)) / sizeof(u32))
686 goto bad;
687 ceph_decode_need(p, end, n * sizeof(u32), bad);
688 err = -ENOMEM;
689 pg = kmalloc(sizeof(*pg) + n*sizeof(u32), GFP_NOFS);
690 if (!pg)
691 goto bad;
692 pg->pgid = pgid;
693 pg->len = n;
694 for (j = 0; j < n; j++)
695 pg->osds[j] = ceph_decode_32(p);
696
697 err = __insert_pg_mapping(pg, &map->pg_temp);
698 if (err)
699 goto bad;
700 dout(" added pg_temp %llx len %d\n", *(u64 *)&pgid, len);
701 }
702
703 /* crush */
704 ceph_decode_32_safe(p, end, len, bad);
705 dout("osdmap_decode crush len %d from off 0x%x\n", len,
706 (int)(*p - start));
707 ceph_decode_need(p, end, len, bad);
708 map->crush = crush_decode(*p, end);
709 *p += len;
710 if (IS_ERR(map->crush)) {
711 err = PTR_ERR(map->crush);
712 map->crush = NULL;
713 goto bad;
714 }
715
716 /* ignore the rest of the map */
717 *p = end;
718
719 dout("osdmap_decode done %p %p\n", *p, end);
720 return map;
721
722 bad:
723 dout("osdmap_decode fail err %d\n", err);
724 ceph_osdmap_destroy(map);
725 return ERR_PTR(err);
726 }
727
728 /*
729 * decode and apply an incremental map update.
730 */
osdmap_apply_incremental(void ** p,void * end,struct ceph_osdmap * map,struct ceph_messenger * msgr)731 struct ceph_osdmap *osdmap_apply_incremental(void **p, void *end,
732 struct ceph_osdmap *map,
733 struct ceph_messenger *msgr)
734 {
735 struct crush_map *newcrush = NULL;
736 struct ceph_fsid fsid;
737 u32 epoch = 0;
738 struct ceph_timespec modified;
739 u32 len, pool;
740 __s32 new_pool_max, new_flags, max;
741 void *start = *p;
742 int err = -EINVAL;
743 u16 version;
744
745 ceph_decode_16_safe(p, end, version, bad);
746 if (version > CEPH_OSDMAP_INC_VERSION) {
747 pr_warning("got unknown v %d > %d of inc osdmap\n", version,
748 CEPH_OSDMAP_INC_VERSION);
749 goto bad;
750 }
751
752 ceph_decode_need(p, end, sizeof(fsid)+sizeof(modified)+2*sizeof(u32),
753 bad);
754 ceph_decode_copy(p, &fsid, sizeof(fsid));
755 epoch = ceph_decode_32(p);
756 BUG_ON(epoch != map->epoch+1);
757 ceph_decode_copy(p, &modified, sizeof(modified));
758 new_pool_max = ceph_decode_32(p);
759 new_flags = ceph_decode_32(p);
760
761 /* full map? */
762 ceph_decode_32_safe(p, end, len, bad);
763 if (len > 0) {
764 dout("apply_incremental full map len %d, %p to %p\n",
765 len, *p, end);
766 return osdmap_decode(p, min(*p+len, end));
767 }
768
769 /* new crush? */
770 ceph_decode_32_safe(p, end, len, bad);
771 if (len > 0) {
772 dout("apply_incremental new crush map len %d, %p to %p\n",
773 len, *p, end);
774 newcrush = crush_decode(*p, min(*p+len, end));
775 if (IS_ERR(newcrush))
776 return ERR_CAST(newcrush);
777 *p += len;
778 }
779
780 /* new flags? */
781 if (new_flags >= 0)
782 map->flags = new_flags;
783 if (new_pool_max >= 0)
784 map->pool_max = new_pool_max;
785
786 ceph_decode_need(p, end, 5*sizeof(u32), bad);
787
788 /* new max? */
789 max = ceph_decode_32(p);
790 if (max >= 0) {
791 err = osdmap_set_max_osd(map, max);
792 if (err < 0)
793 goto bad;
794 }
795
796 map->epoch++;
797 map->modified = modified;
798 if (newcrush) {
799 if (map->crush)
800 crush_destroy(map->crush);
801 map->crush = newcrush;
802 newcrush = NULL;
803 }
804
805 /* new_pool */
806 ceph_decode_32_safe(p, end, len, bad);
807 while (len--) {
808 __u8 ev;
809 struct ceph_pg_pool_info *pi;
810
811 ceph_decode_32_safe(p, end, pool, bad);
812 ceph_decode_need(p, end, 1 + sizeof(pi->v), bad);
813 ev = ceph_decode_8(p); /* encoding version */
814 if (ev > CEPH_PG_POOL_VERSION) {
815 pr_warning("got unknown v %d > %d of ceph_pg_pool\n",
816 ev, CEPH_PG_POOL_VERSION);
817 err = -EINVAL;
818 goto bad;
819 }
820 pi = __lookup_pg_pool(&map->pg_pools, pool);
821 if (!pi) {
822 pi = kzalloc(sizeof(*pi), GFP_NOFS);
823 if (!pi) {
824 err = -ENOMEM;
825 goto bad;
826 }
827 pi->id = pool;
828 __insert_pg_pool(&map->pg_pools, pi);
829 }
830 err = __decode_pool(p, end, pi);
831 if (err < 0)
832 goto bad;
833 }
834 if (version >= 5) {
835 err = __decode_pool_names(p, end, map);
836 if (err < 0)
837 goto bad;
838 }
839
840 /* old_pool */
841 ceph_decode_32_safe(p, end, len, bad);
842 while (len--) {
843 struct ceph_pg_pool_info *pi;
844
845 ceph_decode_32_safe(p, end, pool, bad);
846 pi = __lookup_pg_pool(&map->pg_pools, pool);
847 if (pi)
848 __remove_pg_pool(&map->pg_pools, pi);
849 }
850
851 /* new_up */
852 err = -EINVAL;
853 ceph_decode_32_safe(p, end, len, bad);
854 while (len--) {
855 u32 osd;
856 struct ceph_entity_addr addr;
857 ceph_decode_32_safe(p, end, osd, bad);
858 ceph_decode_copy_safe(p, end, &addr, sizeof(addr), bad);
859 ceph_decode_addr(&addr);
860 pr_info("osd%d up\n", osd);
861 BUG_ON(osd >= map->max_osd);
862 map->osd_state[osd] |= CEPH_OSD_UP;
863 map->osd_addr[osd] = addr;
864 }
865
866 /* new_state */
867 ceph_decode_32_safe(p, end, len, bad);
868 while (len--) {
869 u32 osd;
870 u8 xorstate;
871 ceph_decode_32_safe(p, end, osd, bad);
872 xorstate = **(u8 **)p;
873 (*p)++; /* clean flag */
874 if (xorstate == 0)
875 xorstate = CEPH_OSD_UP;
876 if (xorstate & CEPH_OSD_UP)
877 pr_info("osd%d down\n", osd);
878 if (osd < map->max_osd)
879 map->osd_state[osd] ^= xorstate;
880 }
881
882 /* new_weight */
883 ceph_decode_32_safe(p, end, len, bad);
884 while (len--) {
885 u32 osd, off;
886 ceph_decode_need(p, end, sizeof(u32)*2, bad);
887 osd = ceph_decode_32(p);
888 off = ceph_decode_32(p);
889 pr_info("osd%d weight 0x%x %s\n", osd, off,
890 off == CEPH_OSD_IN ? "(in)" :
891 (off == CEPH_OSD_OUT ? "(out)" : ""));
892 if (osd < map->max_osd)
893 map->osd_weight[osd] = off;
894 }
895
896 /* new_pg_temp */
897 ceph_decode_32_safe(p, end, len, bad);
898 while (len--) {
899 struct ceph_pg_mapping *pg;
900 int j;
901 struct ceph_pg pgid;
902 u32 pglen;
903 ceph_decode_need(p, end, sizeof(u64) + sizeof(u32), bad);
904 ceph_decode_copy(p, &pgid, sizeof(pgid));
905 pglen = ceph_decode_32(p);
906
907 if (pglen) {
908 ceph_decode_need(p, end, pglen*sizeof(u32), bad);
909
910 /* removing existing (if any) */
911 (void) __remove_pg_mapping(&map->pg_temp, pgid);
912
913 /* insert */
914 err = -EINVAL;
915 if (pglen > (UINT_MAX - sizeof(*pg)) / sizeof(u32))
916 goto bad;
917 err = -ENOMEM;
918 pg = kmalloc(sizeof(*pg) + sizeof(u32)*pglen, GFP_NOFS);
919 if (!pg)
920 goto bad;
921 pg->pgid = pgid;
922 pg->len = pglen;
923 for (j = 0; j < pglen; j++)
924 pg->osds[j] = ceph_decode_32(p);
925 err = __insert_pg_mapping(pg, &map->pg_temp);
926 if (err) {
927 kfree(pg);
928 goto bad;
929 }
930 dout(" added pg_temp %llx len %d\n", *(u64 *)&pgid,
931 pglen);
932 } else {
933 /* remove */
934 __remove_pg_mapping(&map->pg_temp, pgid);
935 }
936 }
937
938 /* ignore the rest */
939 *p = end;
940 return map;
941
942 bad:
943 pr_err("corrupt inc osdmap epoch %d off %d (%p of %p-%p)\n",
944 epoch, (int)(*p - start), *p, start, end);
945 print_hex_dump(KERN_DEBUG, "osdmap: ",
946 DUMP_PREFIX_OFFSET, 16, 1,
947 start, end - start, true);
948 if (newcrush)
949 crush_destroy(newcrush);
950 return ERR_PTR(err);
951 }
952
953
954
955
956 /*
957 * calculate file layout from given offset, length.
958 * fill in correct oid, logical length, and object extent
959 * offset, length.
960 *
961 * for now, we write only a single su, until we can
962 * pass a stride back to the caller.
963 */
ceph_calc_file_object_mapping(struct ceph_file_layout * layout,u64 off,u64 * plen,u64 * ono,u64 * oxoff,u64 * oxlen)964 int ceph_calc_file_object_mapping(struct ceph_file_layout *layout,
965 u64 off, u64 *plen,
966 u64 *ono,
967 u64 *oxoff, u64 *oxlen)
968 {
969 u32 osize = le32_to_cpu(layout->fl_object_size);
970 u32 su = le32_to_cpu(layout->fl_stripe_unit);
971 u32 sc = le32_to_cpu(layout->fl_stripe_count);
972 u32 bl, stripeno, stripepos, objsetno;
973 u32 su_per_object;
974 u64 t, su_offset;
975
976 dout("mapping %llu~%llu osize %u fl_su %u\n", off, *plen,
977 osize, su);
978 if (su == 0 || sc == 0)
979 goto invalid;
980 su_per_object = osize / su;
981 if (su_per_object == 0)
982 goto invalid;
983 dout("osize %u / su %u = su_per_object %u\n", osize, su,
984 su_per_object);
985
986 if ((su & ~PAGE_MASK) != 0)
987 goto invalid;
988
989 /* bl = *off / su; */
990 t = off;
991 do_div(t, su);
992 bl = t;
993 dout("off %llu / su %u = bl %u\n", off, su, bl);
994
995 stripeno = bl / sc;
996 stripepos = bl % sc;
997 objsetno = stripeno / su_per_object;
998
999 *ono = objsetno * sc + stripepos;
1000 dout("objset %u * sc %u = ono %u\n", objsetno, sc, (unsigned)*ono);
1001
1002 /* *oxoff = *off % layout->fl_stripe_unit; # offset in su */
1003 t = off;
1004 su_offset = do_div(t, su);
1005 *oxoff = su_offset + (stripeno % su_per_object) * su;
1006
1007 /*
1008 * Calculate the length of the extent being written to the selected
1009 * object. This is the minimum of the full length requested (plen) or
1010 * the remainder of the current stripe being written to.
1011 */
1012 *oxlen = min_t(u64, *plen, su - su_offset);
1013 *plen = *oxlen;
1014
1015 dout(" obj extent %llu~%llu\n", *oxoff, *oxlen);
1016 return 0;
1017
1018 invalid:
1019 dout(" invalid layout\n");
1020 *ono = 0;
1021 *oxoff = 0;
1022 *oxlen = 0;
1023 return -EINVAL;
1024 }
1025 EXPORT_SYMBOL(ceph_calc_file_object_mapping);
1026
1027 /*
1028 * calculate an object layout (i.e. pgid) from an oid,
1029 * file_layout, and osdmap
1030 */
ceph_calc_object_layout(struct ceph_object_layout * ol,const char * oid,struct ceph_file_layout * fl,struct ceph_osdmap * osdmap)1031 int ceph_calc_object_layout(struct ceph_object_layout *ol,
1032 const char *oid,
1033 struct ceph_file_layout *fl,
1034 struct ceph_osdmap *osdmap)
1035 {
1036 unsigned num, num_mask;
1037 struct ceph_pg pgid;
1038 s32 preferred = (s32)le32_to_cpu(fl->fl_pg_preferred);
1039 int poolid = le32_to_cpu(fl->fl_pg_pool);
1040 struct ceph_pg_pool_info *pool;
1041 unsigned ps;
1042
1043 BUG_ON(!osdmap);
1044
1045 pool = __lookup_pg_pool(&osdmap->pg_pools, poolid);
1046 if (!pool)
1047 return -EIO;
1048 ps = ceph_str_hash(pool->v.object_hash, oid, strlen(oid));
1049 if (preferred >= 0) {
1050 ps += preferred;
1051 num = le32_to_cpu(pool->v.lpg_num);
1052 num_mask = pool->lpg_num_mask;
1053 } else {
1054 num = le32_to_cpu(pool->v.pg_num);
1055 num_mask = pool->pg_num_mask;
1056 }
1057
1058 pgid.ps = cpu_to_le16(ps);
1059 pgid.preferred = cpu_to_le16(preferred);
1060 pgid.pool = fl->fl_pg_pool;
1061 if (preferred >= 0)
1062 dout("calc_object_layout '%s' pgid %d.%xp%d\n", oid, poolid, ps,
1063 (int)preferred);
1064 else
1065 dout("calc_object_layout '%s' pgid %d.%x\n", oid, poolid, ps);
1066
1067 ol->ol_pgid = pgid;
1068 ol->ol_stripe_unit = fl->fl_object_stripe_unit;
1069 return 0;
1070 }
1071 EXPORT_SYMBOL(ceph_calc_object_layout);
1072
1073 /*
1074 * Calculate raw osd vector for the given pgid. Return pointer to osd
1075 * array, or NULL on failure.
1076 */
calc_pg_raw(struct ceph_osdmap * osdmap,struct ceph_pg pgid,int * osds,int * num)1077 static int *calc_pg_raw(struct ceph_osdmap *osdmap, struct ceph_pg pgid,
1078 int *osds, int *num)
1079 {
1080 struct ceph_pg_mapping *pg;
1081 struct ceph_pg_pool_info *pool;
1082 int ruleno;
1083 unsigned poolid, ps, pps, t;
1084 int preferred;
1085
1086 poolid = le32_to_cpu(pgid.pool);
1087 ps = le16_to_cpu(pgid.ps);
1088 preferred = (s16)le16_to_cpu(pgid.preferred);
1089
1090 pool = __lookup_pg_pool(&osdmap->pg_pools, poolid);
1091 if (!pool)
1092 return NULL;
1093
1094 /* pg_temp? */
1095 if (preferred >= 0)
1096 t = ceph_stable_mod(ps, le32_to_cpu(pool->v.lpg_num),
1097 pool->lpgp_num_mask);
1098 else
1099 t = ceph_stable_mod(ps, le32_to_cpu(pool->v.pg_num),
1100 pool->pgp_num_mask);
1101 pgid.ps = cpu_to_le16(t);
1102 pg = __lookup_pg_mapping(&osdmap->pg_temp, pgid);
1103 if (pg) {
1104 *num = pg->len;
1105 return pg->osds;
1106 }
1107
1108 /* crush */
1109 ruleno = crush_find_rule(osdmap->crush, pool->v.crush_ruleset,
1110 pool->v.type, pool->v.size);
1111 if (ruleno < 0) {
1112 pr_err("no crush rule pool %d ruleset %d type %d size %d\n",
1113 poolid, pool->v.crush_ruleset, pool->v.type,
1114 pool->v.size);
1115 return NULL;
1116 }
1117
1118 /* don't forcefeed bad device ids to crush */
1119 if (preferred >= osdmap->max_osd ||
1120 preferred >= osdmap->crush->max_devices)
1121 preferred = -1;
1122
1123 if (preferred >= 0)
1124 pps = ceph_stable_mod(ps,
1125 le32_to_cpu(pool->v.lpgp_num),
1126 pool->lpgp_num_mask);
1127 else
1128 pps = ceph_stable_mod(ps,
1129 le32_to_cpu(pool->v.pgp_num),
1130 pool->pgp_num_mask);
1131 pps += poolid;
1132 *num = crush_do_rule(osdmap->crush, ruleno, pps, osds,
1133 min_t(int, pool->v.size, *num),
1134 preferred, osdmap->osd_weight);
1135 return osds;
1136 }
1137
1138 /*
1139 * Return acting set for given pgid.
1140 */
ceph_calc_pg_acting(struct ceph_osdmap * osdmap,struct ceph_pg pgid,int * acting)1141 int ceph_calc_pg_acting(struct ceph_osdmap *osdmap, struct ceph_pg pgid,
1142 int *acting)
1143 {
1144 int rawosds[CEPH_PG_MAX_SIZE], *osds;
1145 int i, o, num = CEPH_PG_MAX_SIZE;
1146
1147 osds = calc_pg_raw(osdmap, pgid, rawosds, &num);
1148 if (!osds)
1149 return -1;
1150
1151 /* primary is first up osd */
1152 o = 0;
1153 for (i = 0; i < num; i++)
1154 if (ceph_osd_is_up(osdmap, osds[i]))
1155 acting[o++] = osds[i];
1156 return o;
1157 }
1158
1159 /*
1160 * Return primary osd for given pgid, or -1 if none.
1161 */
ceph_calc_pg_primary(struct ceph_osdmap * osdmap,struct ceph_pg pgid)1162 int ceph_calc_pg_primary(struct ceph_osdmap *osdmap, struct ceph_pg pgid)
1163 {
1164 int rawosds[CEPH_PG_MAX_SIZE], *osds;
1165 int i, num = CEPH_PG_MAX_SIZE;
1166
1167 osds = calc_pg_raw(osdmap, pgid, rawosds, &num);
1168 if (!osds)
1169 return -1;
1170
1171 /* primary is first up osd */
1172 for (i = 0; i < num; i++)
1173 if (ceph_osd_is_up(osdmap, osds[i]))
1174 return osds[i];
1175 return -1;
1176 }
1177 EXPORT_SYMBOL(ceph_calc_pg_primary);
1178