projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
libceph: don't pass pgid by value
[linux-2.6-block.git] / net / ceph / osdmap.c
CommitLineData
f24e9980 1
3d14c5d2 2#include <linux/ceph/ceph_debug.h>
5a0e3ad6 3
3d14c5d2 4#include <linux/module.h>
5a0e3ad6 5#include <linux/slab.h>
f24e9980
SW		 6#include <asm/div64.h>
7
3d14c5d2
YS		 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>
f24e9980
SW		 13
14char *ceph_osdmap_state_str(char *str, int len, int state)
15{
f24e9980 16 if (!len)
1ec3911d 17 return str;
f24e9980 18
1ec3911d
CD		 19 if ((state & CEPH_OSD_EXISTS) && (state & CEPH_OSD_UP))
20 snprintf(str, len, "exists, up");
21 else if (state & CEPH_OSD_EXISTS)
22 snprintf(str, len, "exists");
23 else if (state & CEPH_OSD_UP)
24 snprintf(str, len, "up");
25 else
f24e9980 26 snprintf(str, len, "doesn't exist");
1ec3911d 27
f24e9980
SW		 28 return str;
29}
30
31/* maps */
32
95c96174 33static int calc_bits_of(unsigned int t)
f24e9980
SW		 34{
35 int b = 0;
36 while (t) {
37 t = t >> 1;
38 b++;
39 }
40 return b;
41}
42
43/*
44 * the foo_mask is the smallest value 2^n-1 that is >= foo.
45 */
46static void calc_pg_masks(struct ceph_pg_pool_info *pi)
47{
4f6a7e5e
SW		 48 pi->pg_num_mask = (1 << calc_bits_of(pi->pg_num-1)) - 1;
49 pi->pgp_num_mask = (1 << calc_bits_of(pi->pgp_num-1)) - 1;
f24e9980
SW		 50}
51
52/*
53 * decode crush map
54 */
55static int crush_decode_uniform_bucket(void **p, void *end,
56 struct crush_bucket_uniform *b)
57{
58 dout("crush_decode_uniform_bucket %p to %p\n", *p, end);
59 ceph_decode_need(p, end, (1+b->h.size) * sizeof(u32), bad);
c89136ea 60 b->item_weight = ceph_decode_32(p);
f24e9980
SW		 61 return 0;
62bad:
63 return -EINVAL;
64}
65
66static int crush_decode_list_bucket(void **p, void *end,
67 struct crush_bucket_list *b)
68{
69 int j;
70 dout("crush_decode_list_bucket %p to %p\n", *p, end);
71 b->item_weights = kcalloc(b->h.size, sizeof(u32), GFP_NOFS);
72 if (b->item_weights == NULL)
73 return -ENOMEM;
74 b->sum_weights = kcalloc(b->h.size, sizeof(u32), GFP_NOFS);
75 if (b->sum_weights == NULL)
76 return -ENOMEM;
77 ceph_decode_need(p, end, 2 * b->h.size * sizeof(u32), bad);
78 for (j = 0; j < b->h.size; j++) {
c89136ea
SW		 79 b->item_weights[j] = ceph_decode_32(p);
80 b->sum_weights[j] = ceph_decode_32(p);
f24e9980
SW		 81 }
82 return 0;
83bad:
84 return -EINVAL;
85}
86
87static int crush_decode_tree_bucket(void **p, void *end,
88 struct crush_bucket_tree *b)
89{
90 int j;
91 dout("crush_decode_tree_bucket %p to %p\n", *p, end);
82cd003a 92 ceph_decode_8_safe(p, end, b->num_nodes, bad);
f24e9980
SW		 93 b->node_weights = kcalloc(b->num_nodes, sizeof(u32), GFP_NOFS);
94 if (b->node_weights == NULL)
95 return -ENOMEM;
96 ceph_decode_need(p, end, b->num_nodes * sizeof(u32), bad);
97 for (j = 0; j < b->num_nodes; j++)
c89136ea 98 b->node_weights[j] = ceph_decode_32(p);
f24e9980
SW		 99 return 0;
100bad:
101 return -EINVAL;
102}
103
104static int crush_decode_straw_bucket(void **p, void *end,
105 struct crush_bucket_straw *b)
106{
107 int j;
108 dout("crush_decode_straw_bucket %p to %p\n", *p, end);
109 b->item_weights = kcalloc(b->h.size, sizeof(u32), GFP_NOFS);
110 if (b->item_weights == NULL)
111 return -ENOMEM;
112 b->straws = kcalloc(b->h.size, sizeof(u32), GFP_NOFS);
113 if (b->straws == NULL)
114 return -ENOMEM;
115 ceph_decode_need(p, end, 2 * b->h.size * sizeof(u32), bad);
116 for (j = 0; j < b->h.size; j++) {
c89136ea
SW		 117 b->item_weights[j] = ceph_decode_32(p);
118 b->straws[j] = ceph_decode_32(p);
f24e9980
SW		 119 }
120 return 0;
121bad:
122 return -EINVAL;
123}
124
958a2765
ID		 125static int crush_decode_straw2_bucket(void **p, void *end,
126 struct crush_bucket_straw2 *b)
127{
128 int j;
129 dout("crush_decode_straw2_bucket %p to %p\n", *p, end);
130 b->item_weights = kcalloc(b->h.size, sizeof(u32), GFP_NOFS);
131 if (b->item_weights == NULL)
132 return -ENOMEM;
133 ceph_decode_need(p, end, b->h.size * sizeof(u32), bad);
134 for (j = 0; j < b->h.size; j++)
135 b->item_weights[j] = ceph_decode_32(p);
136 return 0;
137bad:
138 return -EINVAL;
139}
140
546f04ef
SW		 141static int skip_name_map(void **p, void *end)
142{
143 int len;
144 ceph_decode_32_safe(p, end, len ,bad);
145 while (len--) {
146 int strlen;
147 *p += sizeof(u32);
148 ceph_decode_32_safe(p, end, strlen, bad);
149 *p += strlen;
150}
151 return 0;
152bad:
153 return -EINVAL;
154}
155
66a0e2d5
ID		 156static void crush_finalize(struct crush_map *c)
157{
158 __s32 b;
159
160 /* Space for the array of pointers to per-bucket workspace */
161 c->working_size = sizeof(struct crush_work) +
162 c->max_buckets * sizeof(struct crush_work_bucket *);
163
164 for (b = 0; b < c->max_buckets; b++) {
165 if (!c->buckets[b])
166 continue;
167
168 switch (c->buckets[b]->alg) {
169 default:
170 /*
171 * The base case, permutation variables and
172 * the pointer to the permutation array.
173 */
174 c->working_size += sizeof(struct crush_work_bucket);
175 break;
176 }
177 /* Every bucket has a permutation array. */
178 c->working_size += c->buckets[b]->size * sizeof(__u32);
179 }
180}
181
f24e9980
SW		 182static struct crush_map *crush_decode(void *pbyval, void *end)
183{
184 struct crush_map *c;
185 int err = -EINVAL;
186 int i, j;
187 void **p = &pbyval;
188 void *start = pbyval;
189 u32 magic;
546f04ef 190 u32 num_name_maps;
f24e9980
SW		 191
192 dout("crush_decode %p to %p len %d\n", *p, end, (int)(end - *p));
193
194 c = kzalloc(sizeof(*c), GFP_NOFS);
195 if (c == NULL)
196 return ERR_PTR(-ENOMEM);
197
546f04ef
SW		 198 /* set tunables to default values */
199 c->choose_local_tries = 2;
200 c->choose_local_fallback_tries = 5;
201 c->choose_total_tries = 19;
1604f488 202 c->chooseleaf_descend_once = 0;
546f04ef 203
f24e9980 204 ceph_decode_need(p, end, 4*sizeof(u32), bad);
c89136ea 205 magic = ceph_decode_32(p);
f24e9980
SW		 206 if (magic != CRUSH_MAGIC) {
207 pr_err("crush_decode magic %x != current %x\n",
95c96174 208 (unsigned int)magic, (unsigned int)CRUSH_MAGIC);
f24e9980
SW		 209 goto bad;
210 }
c89136ea
SW		 211 c->max_buckets = ceph_decode_32(p);
212 c->max_rules = ceph_decode_32(p);
213 c->max_devices = ceph_decode_32(p);
f24e9980 214
f24e9980
SW		 215 c->buckets = kcalloc(c->max_buckets, sizeof(*c->buckets), GFP_NOFS);
216 if (c->buckets == NULL)
217 goto badmem;
218 c->rules = kcalloc(c->max_rules, sizeof(*c->rules), GFP_NOFS);
219 if (c->rules == NULL)
220 goto badmem;
221
222 /* buckets */
223 for (i = 0; i < c->max_buckets; i++) {
224 int size = 0;
225 u32 alg;
226 struct crush_bucket *b;
227
228 ceph_decode_32_safe(p, end, alg, bad);
229 if (alg == 0) {
230 c->buckets[i] = NULL;
231 continue;
232 }
233 dout("crush_decode bucket %d off %x %p to %p\n",
234 i, (int)(*p-start), *p, end);
235
236 switch (alg) {
237 case CRUSH_BUCKET_UNIFORM:
238 size = sizeof(struct crush_bucket_uniform);
239 break;
240 case CRUSH_BUCKET_LIST:
241 size = sizeof(struct crush_bucket_list);
242 break;
243 case CRUSH_BUCKET_TREE:
244 size = sizeof(struct crush_bucket_tree);
245 break;
246 case CRUSH_BUCKET_STRAW:
247 size = sizeof(struct crush_bucket_straw);
248 break;
958a2765
ID		 249 case CRUSH_BUCKET_STRAW2:
250 size = sizeof(struct crush_bucket_straw2);
251 break;
f24e9980 252 default:
30dc6381 253 err = -EINVAL;
f24e9980
SW		 254 goto bad;
255 }
256 BUG_ON(size == 0);
257 b = c->buckets[i] = kzalloc(size, GFP_NOFS);
258 if (b == NULL)
259 goto badmem;
260
261 ceph_decode_need(p, end, 4*sizeof(u32), bad);
c89136ea
SW		 262 b->id = ceph_decode_32(p);
263 b->type = ceph_decode_16(p);
fb690390
SW		 264 b->alg = ceph_decode_8(p);
265 b->hash = ceph_decode_8(p);
c89136ea
SW		 266 b->weight = ceph_decode_32(p);
267 b->size = ceph_decode_32(p);
f24e9980
SW		 268
269 dout("crush_decode bucket size %d off %x %p to %p\n",
270 b->size, (int)(*p-start), *p, end);
271
272 b->items = kcalloc(b->size, sizeof(__s32), GFP_NOFS);
273 if (b->items == NULL)
274 goto badmem;
f24e9980
SW		 275
276 ceph_decode_need(p, end, b->size*sizeof(u32), bad);
277 for (j = 0; j < b->size; j++)
c89136ea 278 b->items[j] = ceph_decode_32(p);
f24e9980
SW		 279
280 switch (b->alg) {
281 case CRUSH_BUCKET_UNIFORM:
282 err = crush_decode_uniform_bucket(p, end,
283 (struct crush_bucket_uniform *)b);
284 if (err < 0)
285 goto bad;
286 break;
287 case CRUSH_BUCKET_LIST:
288 err = crush_decode_list_bucket(p, end,
289 (struct crush_bucket_list *)b);
290 if (err < 0)
291 goto bad;
292 break;
293 case CRUSH_BUCKET_TREE:
294 err = crush_decode_tree_bucket(p, end,
295 (struct crush_bucket_tree *)b);
296 if (err < 0)
297 goto bad;
298 break;
299 case CRUSH_BUCKET_STRAW:
300 err = crush_decode_straw_bucket(p, end,
301 (struct crush_bucket_straw *)b);
302 if (err < 0)
303 goto bad;
304 break;
958a2765
ID		 305 case CRUSH_BUCKET_STRAW2:
306 err = crush_decode_straw2_bucket(p, end,
307 (struct crush_bucket_straw2 *)b);
308 if (err < 0)
309 goto bad;
310 break;
f24e9980
SW		 311 }
312 }
313
314 /* rules */
315 dout("rule vec is %p\n", c->rules);
316 for (i = 0; i < c->max_rules; i++) {
317 u32 yes;
318 struct crush_rule *r;
319
293dffaa 320 err = -EINVAL;
f24e9980
SW		 321 ceph_decode_32_safe(p, end, yes, bad);
322 if (!yes) {
323 dout("crush_decode NO rule %d off %x %p to %p\n",
324 i, (int)(*p-start), *p, end);
325 c->rules[i] = NULL;
326 continue;
327 }
328
329 dout("crush_decode rule %d off %x %p to %p\n",
330 i, (int)(*p-start), *p, end);
331
332 /* len */
333 ceph_decode_32_safe(p, end, yes, bad);
334#if BITS_PER_LONG == 32
30dc6381 335 err = -EINVAL;
64486697
XW		 336 if (yes > (ULONG_MAX - sizeof(*r))
337 / sizeof(struct crush_rule_step))
f24e9980
SW		 338 goto bad;
339#endif
340 r = c->rules[i] = kmalloc(sizeof(*r) +
341 yes*sizeof(struct crush_rule_step),
342 GFP_NOFS);
343 if (r == NULL)
344 goto badmem;
345 dout(" rule %d is at %p\n", i, r);
346 r->len = yes;
347 ceph_decode_copy_safe(p, end, &r->mask, 4, bad); /* 4 u8's */
348 ceph_decode_need(p, end, r->len*3*sizeof(u32), bad);
349 for (j = 0; j < r->len; j++) {
c89136ea
SW		 350 r->steps[j].op = ceph_decode_32(p);
351 r->steps[j].arg1 = ceph_decode_32(p);
352 r->steps[j].arg2 = ceph_decode_32(p);
f24e9980
SW		 353 }
354 }
355
356 /* ignore trailing name maps. */
546f04ef
SW		 357 for (num_name_maps = 0; num_name_maps < 3; num_name_maps++) {
358 err = skip_name_map(p, end);
359 if (err < 0)
360 goto done;
361 }
362
363 /* tunables */
364 ceph_decode_need(p, end, 3*sizeof(u32), done);
365 c->choose_local_tries = ceph_decode_32(p);
366 c->choose_local_fallback_tries = ceph_decode_32(p);
367 c->choose_total_tries = ceph_decode_32(p);
b9b519b7 368 dout("crush decode tunable choose_local_tries = %d\n",
546f04ef 369 c->choose_local_tries);
b9b519b7 370 dout("crush decode tunable choose_local_fallback_tries = %d\n",
546f04ef 371 c->choose_local_fallback_tries);
b9b519b7 372 dout("crush decode tunable choose_total_tries = %d\n",
546f04ef 373 c->choose_total_tries);
f24e9980 374
1604f488
JS		 375 ceph_decode_need(p, end, sizeof(u32), done);
376 c->chooseleaf_descend_once = ceph_decode_32(p);
b9b519b7 377 dout("crush decode tunable chooseleaf_descend_once = %d\n",
1604f488
JS		 378 c->chooseleaf_descend_once);
379
f140662f
ID		 380 ceph_decode_need(p, end, sizeof(u8), done);
381 c->chooseleaf_vary_r = ceph_decode_8(p);
b9b519b7 382 dout("crush decode tunable chooseleaf_vary_r = %d\n",
f140662f
ID		 383 c->chooseleaf_vary_r);
384
b9b519b7
ID		 385 /* skip straw_calc_version, allowed_bucket_algs */
386 ceph_decode_need(p, end, sizeof(u8) + sizeof(u32), done);
387 *p += sizeof(u8) + sizeof(u32);
388
389 ceph_decode_need(p, end, sizeof(u8), done);
390 c->chooseleaf_stable = ceph_decode_8(p);
391 dout("crush decode tunable chooseleaf_stable = %d\n",
392 c->chooseleaf_stable);
393
546f04ef 394done:
9afd30db 395 crush_finalize(c);
f24e9980
SW		 396 dout("crush_decode success\n");
397 return c;
398
399badmem:
400 err = -ENOMEM;
401bad:
402 dout("crush_decode fail %d\n", err);
403 crush_destroy(c);
404 return ERR_PTR(err);
405}
406
f984cb76 407int ceph_pg_compare(const struct ceph_pg *lhs, const struct ceph_pg *rhs)
51042122 408{
f984cb76 409 if (lhs->pool < rhs->pool)
5b191d99 410 return -1;
f984cb76 411 if (lhs->pool > rhs->pool)
5b191d99 412 return 1;
f984cb76 413 if (lhs->seed < rhs->seed)
51042122 414 return -1;
f984cb76 415 if (lhs->seed > rhs->seed)
51042122 416 return 1;
f984cb76 417
51042122
SW		 418 return 0;
419}
420
a02a946d
ID		 421int ceph_spg_compare(const struct ceph_spg *lhs, const struct ceph_spg *rhs)
422{
423 int ret;
424
425 ret = ceph_pg_compare(&lhs->pgid, &rhs->pgid);
426 if (ret)
427 return ret;
428
429 if (lhs->shard < rhs->shard)
430 return -1;
431 if (lhs->shard > rhs->shard)
432 return 1;
433
434 return 0;
435}
436
f984cb76
ID		 437/*
438 * rbtree of pg_mapping for handling pg_temp (explicit mapping of pgid
439 * to a set of osds) and primary_temp (explicit primary setting)
440 */
991abb6e
SW		 441static int __insert_pg_mapping(struct ceph_pg_mapping *new,
442 struct rb_root *root)
f24e9980
SW		 443{
444 struct rb_node **p = &root->rb_node;
445 struct rb_node *parent = NULL;
446 struct ceph_pg_mapping *pg = NULL;
51042122 447 int c;
f24e9980 448
8adc8b3d 449 dout("__insert_pg_mapping %llx %p\n", *(u64 *)&new->pgid, new);
f24e9980
SW		 450 while (*p) {
451 parent = *p;
452 pg = rb_entry(parent, struct ceph_pg_mapping, node);
f984cb76 453 c = ceph_pg_compare(&new->pgid, &pg->pgid);
51042122 454 if (c < 0)
f24e9980 455 p = &(*p)->rb_left;
51042122 456 else if (c > 0)
f24e9980
SW		 457 p = &(*p)->rb_right;
458 else
991abb6e 459 return -EEXIST;
f24e9980
SW		 460 }
461
462 rb_link_node(&new->node, parent, p);
463 rb_insert_color(&new->node, root);
991abb6e 464 return 0;
f24e9980
SW		 465}
466
9794b146 467static struct ceph_pg_mapping *__lookup_pg_mapping(struct rb_root *root,
33333d10 468 const struct ceph_pg *pgid)
9794b146
SW		 469{
470 struct rb_node *n = root->rb_node;
471 struct ceph_pg_mapping *pg;
472 int c;
473
474 while (n) {
475 pg = rb_entry(n, struct ceph_pg_mapping, node);
33333d10 476 c = ceph_pg_compare(pgid, &pg->pgid);
8adc8b3d 477 if (c < 0) {
9794b146 478 n = n->rb_left;
8adc8b3d 479 } else if (c > 0) {
9794b146 480 n = n->rb_right;
8adc8b3d 481 } else {
5b191d99 482 dout("__lookup_pg_mapping %lld.%x got %p\n",
33333d10 483 pgid->pool, pgid->seed, pg);
9794b146 484 return pg;
8adc8b3d 485 }
9794b146
SW		 486 }
487 return NULL;
488}
489
33333d10 490static int __remove_pg_mapping(struct rb_root *root, const struct ceph_pg *pgid)
8adc8b3d
SW		 491{
492 struct ceph_pg_mapping *pg = __lookup_pg_mapping(root, pgid);
493
494 if (pg) {
33333d10 495 dout("__remove_pg_mapping %lld.%x %p\n", pgid->pool, pgid->seed,
5b191d99 496 pg);
8adc8b3d
SW		 497 rb_erase(&pg->node, root);
498 kfree(pg);
499 return 0;
500 }
33333d10 501 dout("__remove_pg_mapping %lld.%x dne\n", pgid->pool, pgid->seed);
8adc8b3d
SW		 502 return -ENOENT;
503}
504
4fc51be8
SW		 505/*
506 * rbtree of pg pool info
507 */
508static int __insert_pg_pool(struct rb_root *root, struct ceph_pg_pool_info *new)
509{
510 struct rb_node **p = &root->rb_node;
511 struct rb_node *parent = NULL;
512 struct ceph_pg_pool_info *pi = NULL;
513
514 while (*p) {
515 parent = *p;
516 pi = rb_entry(parent, struct ceph_pg_pool_info, node);
517 if (new->id < pi->id)
518 p = &(*p)->rb_left;
519 else if (new->id > pi->id)
520 p = &(*p)->rb_right;
521 else
522 return -EEXIST;
523 }
524
525 rb_link_node(&new->node, parent, p);
526 rb_insert_color(&new->node, root);
527 return 0;
528}
529
4f6a7e5e 530static struct ceph_pg_pool_info *__lookup_pg_pool(struct rb_root *root, u64 id)
4fc51be8
SW		 531{
532 struct ceph_pg_pool_info *pi;
533 struct rb_node *n = root->rb_node;
534
535 while (n) {
536 pi = rb_entry(n, struct ceph_pg_pool_info, node);
537 if (id < pi->id)
538 n = n->rb_left;
539 else if (id > pi->id)
540 n = n->rb_right;
541 else
542 return pi;
543 }
544 return NULL;
545}
546
ce7f6a27
ID		 547struct ceph_pg_pool_info *ceph_pg_pool_by_id(struct ceph_osdmap *map, u64 id)
548{
549 return __lookup_pg_pool(&map->pg_pools, id);
550}
551
72afc71f
AE		 552const char *ceph_pg_pool_name_by_id(struct ceph_osdmap *map, u64 id)
553{
554 struct ceph_pg_pool_info *pi;
555
556 if (id == CEPH_NOPOOL)
557 return NULL;
558
559 if (WARN_ON_ONCE(id > (u64) INT_MAX))
560 return NULL;
561
562 pi = __lookup_pg_pool(&map->pg_pools, (int) id);
563
564 return pi ? pi->name : NULL;
565}
566EXPORT_SYMBOL(ceph_pg_pool_name_by_id);
567
7669a2c9
YS		 568int ceph_pg_poolid_by_name(struct ceph_osdmap *map, const char *name)
569{
570 struct rb_node *rbp;
571
572 for (rbp = rb_first(&map->pg_pools); rbp; rbp = rb_next(rbp)) {
573 struct ceph_pg_pool_info *pi =
574 rb_entry(rbp, struct ceph_pg_pool_info, node);
575 if (pi->name && strcmp(pi->name, name) == 0)
576 return pi->id;
577 }
578 return -ENOENT;
579}
3d14c5d2 580EXPORT_SYMBOL(ceph_pg_poolid_by_name);
7669a2c9 581
2844a76a
SW		 582static void __remove_pg_pool(struct rb_root *root, struct ceph_pg_pool_info *pi)
583{
584 rb_erase(&pi->node, root);
585 kfree(pi->name);
586 kfree(pi);
587}
588
0f70c7ee 589static int decode_pool(void **p, void *end, struct ceph_pg_pool_info *pi)
efd7576b 590{
4f6a7e5e
SW		 591 u8 ev, cv;
592 unsigned len, num;
593 void *pool_end;
594
595 ceph_decode_need(p, end, 2 + 4, bad);
596 ev = ceph_decode_8(p); /* encoding version */
597 cv = ceph_decode_8(p); /* compat version */
598 if (ev < 5) {
b9a67899 599 pr_warn("got v %d < 5 cv %d of ceph_pg_pool\n", ev, cv);
4f6a7e5e
SW		 600 return -EINVAL;
601 }
17a13e40 602 if (cv > 9) {
b9a67899 603 pr_warn("got v %d cv %d > 9 of ceph_pg_pool\n", ev, cv);
4f6a7e5e
SW		 604 return -EINVAL;
605 }
606 len = ceph_decode_32(p);
607 ceph_decode_need(p, end, len, bad);
608 pool_end = *p + len;
73a7e693 609
4f6a7e5e
SW		 610 pi->type = ceph_decode_8(p);
611 pi->size = ceph_decode_8(p);
612 pi->crush_ruleset = ceph_decode_8(p);
613 pi->object_hash = ceph_decode_8(p);
73a7e693 614
4f6a7e5e
SW		 615 pi->pg_num = ceph_decode_32(p);
616 pi->pgp_num = ceph_decode_32(p);
617
618 *p += 4 + 4; /* skip lpg* */
619 *p += 4; /* skip last_change */
620 *p += 8 + 4; /* skip snap_seq, snap_epoch */
621
622 /* skip snaps */
623 num = ceph_decode_32(p);
624 while (num--) {
625 *p += 8; /* snapid key */
626 *p += 1 + 1; /* versions */
627 len = ceph_decode_32(p);
628 *p += len;
73a7e693
SW		 629 }
630
17a13e40 631 /* skip removed_snaps */
4f6a7e5e
SW		 632 num = ceph_decode_32(p);
633 *p += num * (8 + 8);
634
635 *p += 8; /* skip auid */
636 pi->flags = ceph_decode_64(p);
17a13e40
ID		 637 *p += 4; /* skip crash_replay_interval */
638
639 if (ev >= 7)
04812acf
ID		 640 pi->min_size = ceph_decode_8(p);
641 else
642 pi->min_size = pi->size - pi->size / 2;
17a13e40
ID		 643
644 if (ev >= 8)
645 *p += 8 + 8; /* skip quota_max_* */
646
647 if (ev >= 9) {
648 /* skip tiers */
649 num = ceph_decode_32(p);
650 *p += num * 8;
651
652 *p += 8; /* skip tier_of */
653 *p += 1; /* skip cache_mode */
654
655 pi->read_tier = ceph_decode_64(p);
656 pi->write_tier = ceph_decode_64(p);
657 } else {
658 pi->read_tier = -1;
659 pi->write_tier = -1;
660 }
4f6a7e5e 661
04812acf
ID		 662 if (ev >= 10) {
663 /* skip properties */
664 num = ceph_decode_32(p);
665 while (num--) {
666 len = ceph_decode_32(p);
667 *p += len; /* key */
668 len = ceph_decode_32(p);
669 *p += len; /* val */
670 }
671 }
672
673 if (ev >= 11) {
674 /* skip hit_set_params */
675 *p += 1 + 1; /* versions */
676 len = ceph_decode_32(p);
677 *p += len;
678
679 *p += 4; /* skip hit_set_period */
680 *p += 4; /* skip hit_set_count */
681 }
682
683 if (ev >= 12)
684 *p += 4; /* skip stripe_width */
685
686 if (ev >= 13) {
687 *p += 8; /* skip target_max_bytes */
688 *p += 8; /* skip target_max_objects */
689 *p += 4; /* skip cache_target_dirty_ratio_micro */
690 *p += 4; /* skip cache_target_full_ratio_micro */
691 *p += 4; /* skip cache_min_flush_age */
692 *p += 4; /* skip cache_min_evict_age */
693 }
694
695 if (ev >= 14) {
696 /* skip erasure_code_profile */
697 len = ceph_decode_32(p);
698 *p += len;
699 }
700
8e48cf00
ID		 701 /*
702 * last_force_op_resend_preluminous, will be overridden if the
703 * map was encoded with RESEND_ON_SPLIT
704 */
04812acf
ID		 705 if (ev >= 15)
706 pi->last_force_request_resend = ceph_decode_32(p);
707 else
708 pi->last_force_request_resend = 0;
709
8e48cf00
ID		 710 if (ev >= 16)
711 *p += 4; /* skip min_read_recency_for_promote */
712
713 if (ev >= 17)
714 *p += 8; /* skip expected_num_objects */
715
716 if (ev >= 19)
717 *p += 4; /* skip cache_target_dirty_high_ratio_micro */
718
719 if (ev >= 20)
720 *p += 4; /* skip min_write_recency_for_promote */
721
722 if (ev >= 21)
723 *p += 1; /* skip use_gmt_hitset */
724
725 if (ev >= 22)
726 *p += 1; /* skip fast_read */
727
728 if (ev >= 23) {
729 *p += 4; /* skip hit_set_grade_decay_rate */
730 *p += 4; /* skip hit_set_search_last_n */
731 }
732
733 if (ev >= 24) {
734 /* skip opts */
735 *p += 1 + 1; /* versions */
736 len = ceph_decode_32(p);
737 *p += len;
738 }
739
740 if (ev >= 25)
741 pi->last_force_request_resend = ceph_decode_32(p);
742
4f6a7e5e
SW		 743 /* ignore the rest */
744
745 *p = pool_end;
746 calc_pg_masks(pi);
73a7e693
SW		 747 return 0;
748
749bad:
750 return -EINVAL;
efd7576b
SW		 751}
752
0f70c7ee 753static int decode_pool_names(void **p, void *end, struct ceph_osdmap *map)
2844a76a
SW		 754{
755 struct ceph_pg_pool_info *pi;
4f6a7e5e
SW		 756 u32 num, len;
757 u64 pool;
2844a76a
SW		 758
759 ceph_decode_32_safe(p, end, num, bad);
760 dout(" %d pool names\n", num);
761 while (num--) {
4f6a7e5e 762 ceph_decode_64_safe(p, end, pool, bad);
2844a76a 763 ceph_decode_32_safe(p, end, len, bad);
4f6a7e5e 764 dout(" pool %llu len %d\n", pool, len);
ad3b904c 765 ceph_decode_need(p, end, len, bad);
2844a76a
SW		 766 pi = __lookup_pg_pool(&map->pg_pools, pool);
767 if (pi) {
ad3b904c
XW		 768 char *name = kstrndup(*p, len, GFP_NOFS);
769
770 if (!name)
771 return -ENOMEM;
2844a76a 772 kfree(pi->name);
ad3b904c
XW		 773 pi->name = name;
774 dout(" name is %s\n", pi->name);
2844a76a
SW		 775 }
776 *p += len;
777 }
778 return 0;
779
780bad:
781 return -EINVAL;
782}
783
784/*
785 * osd map
786 */
e5253a7b
ID		 787struct ceph_osdmap *ceph_osdmap_alloc(void)
788{
789 struct ceph_osdmap *map;
790
791 map = kzalloc(sizeof(*map), GFP_NOIO);
792 if (!map)
793 return NULL;
794
795 map->pg_pools = RB_ROOT;
796 map->pool_max = -1;
797 map->pg_temp = RB_ROOT;
798 map->primary_temp = RB_ROOT;
743efcff 799 mutex_init(&map->crush_workspace_mutex);
e5253a7b
ID		 800
801 return map;
802}
803
2844a76a
SW		 804void ceph_osdmap_destroy(struct ceph_osdmap *map)
805{
806 dout("osdmap_destroy %p\n", map);
807 if (map->crush)
808 crush_destroy(map->crush);
809 while (!RB_EMPTY_ROOT(&map->pg_temp)) {
810 struct ceph_pg_mapping *pg =
811 rb_entry(rb_first(&map->pg_temp),
812 struct ceph_pg_mapping, node);
813 rb_erase(&pg->node, &map->pg_temp);
814 kfree(pg);
815 }
9686f94c
ID		 816 while (!RB_EMPTY_ROOT(&map->primary_temp)) {
817 struct ceph_pg_mapping *pg =
818 rb_entry(rb_first(&map->primary_temp),
819 struct ceph_pg_mapping, node);
820 rb_erase(&pg->node, &map->primary_temp);
821 kfree(pg);
822 }
2844a76a
SW		 823 while (!RB_EMPTY_ROOT(&map->pg_pools)) {
824 struct ceph_pg_pool_info *pi =
825 rb_entry(rb_first(&map->pg_pools),
826 struct ceph_pg_pool_info, node);
827 __remove_pg_pool(&map->pg_pools, pi);
828 }
829 kfree(map->osd_state);
830 kfree(map->osd_weight);
831 kfree(map->osd_addr);
2cfa34f2 832 kfree(map->osd_primary_affinity);
66a0e2d5 833 kfree(map->crush_workspace);
2844a76a
SW		 834 kfree(map);
835}
836
837/*
4d60351f
ID		 838 * Adjust max_osd value, (re)allocate arrays.
839 *
840 * The new elements are properly initialized.
2844a76a
SW		 841 */
842static int osdmap_set_max_osd(struct ceph_osdmap *map, int max)
843{
844 u8 *state;
2844a76a 845 u32 *weight;
4d60351f
ID		 846 struct ceph_entity_addr *addr;
847 int i;
2844a76a 848
4d60351f 849 state = krealloc(map->osd_state, max*sizeof(*state), GFP_NOFS);
589506f1
LR		 850 if (!state)
851 return -ENOMEM;
852 map->osd_state = state;
853
4d60351f 854 weight = krealloc(map->osd_weight, max*sizeof(*weight), GFP_NOFS);
589506f1
LR		 855 if (!weight)
856 return -ENOMEM;
857 map->osd_weight = weight;
4d60351f 858
589506f1
LR		 859 addr = krealloc(map->osd_addr, max*sizeof(*addr), GFP_NOFS);
860 if (!addr)
2844a76a 861 return -ENOMEM;
589506f1 862 map->osd_addr = addr;
2844a76a 863
4d60351f 864 for (i = map->max_osd; i < max; i++) {
589506f1
LR		 865 map->osd_state[i] = 0;
866 map->osd_weight[i] = CEPH_OSD_OUT;
867 memset(map->osd_addr + i, 0, sizeof(*map->osd_addr));
2844a76a
SW		 868 }
869
2cfa34f2
ID		 870 if (map->osd_primary_affinity) {
871 u32 *affinity;
872
873 affinity = krealloc(map->osd_primary_affinity,
874 max*sizeof(*affinity), GFP_NOFS);
875 if (!affinity)
876 return -ENOMEM;
589506f1 877 map->osd_primary_affinity = affinity;
2cfa34f2
ID		 878
879 for (i = map->max_osd; i < max; i++)
589506f1
LR		 880 map->osd_primary_affinity[i] =
881 CEPH_OSD_DEFAULT_PRIMARY_AFFINITY;
2cfa34f2
ID		 882 }
883
2844a76a 884 map->max_osd = max;
4d60351f 885
2844a76a
SW		 886 return 0;
887}
888
1b6a78b5
ID		 889static int osdmap_set_crush(struct ceph_osdmap *map, struct crush_map *crush)
890{
66a0e2d5 891 void *workspace;
743efcff 892 size_t work_size;
66a0e2d5 893
1b6a78b5
ID		 894 if (IS_ERR(crush))
895 return PTR_ERR(crush);
896
743efcff
ID		 897 work_size = crush_work_size(crush, CEPH_PG_MAX_SIZE);
898 dout("%s work_size %zu bytes\n", __func__, work_size);
899 workspace = kmalloc(work_size, GFP_NOIO);
66a0e2d5
ID		 900 if (!workspace) {
901 crush_destroy(crush);
902 return -ENOMEM;
903 }
904 crush_init_workspace(crush, workspace);
905
1b6a78b5
ID		 906 if (map->crush)
907 crush_destroy(map->crush);
66a0e2d5 908 kfree(map->crush_workspace);
1b6a78b5 909 map->crush = crush;
66a0e2d5 910 map->crush_workspace = workspace;
1b6a78b5
ID		 911 return 0;
912}
913
ec7af972
ID		 914#define OSDMAP_WRAPPER_COMPAT_VER 7
915#define OSDMAP_CLIENT_DATA_COMPAT_VER 1
916
917/*
918 * Return 0 or error. On success, *v is set to 0 for old (v6) osdmaps,
919 * to struct_v of the client_data section for new (v7 and above)
920 * osdmaps.
921 */
922static int get_osdmap_client_data_v(void **p, void *end,
923 const char *prefix, u8 *v)
924{
925 u8 struct_v;
926
927 ceph_decode_8_safe(p, end, struct_v, e_inval);
928 if (struct_v >= 7) {
929 u8 struct_compat;
930
931 ceph_decode_8_safe(p, end, struct_compat, e_inval);
932 if (struct_compat > OSDMAP_WRAPPER_COMPAT_VER) {
b9a67899
JP		 933 pr_warn("got v %d cv %d > %d of %s ceph_osdmap\n",
934 struct_v, struct_compat,
935 OSDMAP_WRAPPER_COMPAT_VER, prefix);
ec7af972
ID		 936 return -EINVAL;
937 }
938 *p += 4; /* ignore wrapper struct_len */
939
940 ceph_decode_8_safe(p, end, struct_v, e_inval);
941 ceph_decode_8_safe(p, end, struct_compat, e_inval);
942 if (struct_compat > OSDMAP_CLIENT_DATA_COMPAT_VER) {
b9a67899
JP		 943 pr_warn("got v %d cv %d > %d of %s ceph_osdmap client data\n",
944 struct_v, struct_compat,
945 OSDMAP_CLIENT_DATA_COMPAT_VER, prefix);
ec7af972
ID		 946 return -EINVAL;
947 }
948 *p += 4; /* ignore client data struct_len */
949 } else {
950 u16 version;
951
952 *p -= 1;
953 ceph_decode_16_safe(p, end, version, e_inval);
954 if (version < 6) {
b9a67899
JP		 955 pr_warn("got v %d < 6 of %s ceph_osdmap\n",
956 version, prefix);
ec7af972
ID		 957 return -EINVAL;
958 }
959
960 /* old osdmap enconding */
961 struct_v = 0;
962 }
963
964 *v = struct_v;
965 return 0;
966
967e_inval:
968 return -EINVAL;
969}
970
433fbdd3
ID		 971static int __decode_pools(void **p, void *end, struct ceph_osdmap *map,
972 bool incremental)
973{
974 u32 n;
975
976 ceph_decode_32_safe(p, end, n, e_inval);
977 while (n--) {
978 struct ceph_pg_pool_info *pi;
979 u64 pool;
980 int ret;
981
982 ceph_decode_64_safe(p, end, pool, e_inval);
983
984 pi = __lookup_pg_pool(&map->pg_pools, pool);
985 if (!incremental || !pi) {
986 pi = kzalloc(sizeof(*pi), GFP_NOFS);
987 if (!pi)
988 return -ENOMEM;
989
990 pi->id = pool;
991
992 ret = __insert_pg_pool(&map->pg_pools, pi);
993 if (ret) {
994 kfree(pi);
995 return ret;
996 }
997 }
998
999 ret = decode_pool(p, end, pi);
1000 if (ret)
1001 return ret;
1002 }
1003
1004 return 0;
1005
1006e_inval:
1007 return -EINVAL;
1008}
1009
1010static int decode_pools(void **p, void *end, struct ceph_osdmap *map)
1011{
1012 return __decode_pools(p, end, map, false);
1013}
1014
1015static int decode_new_pools(void **p, void *end, struct ceph_osdmap *map)
1016{
1017 return __decode_pools(p, end, map, true);
1018}
1019
10db634e
ID		 1020static int __decode_pg_temp(void **p, void *end, struct ceph_osdmap *map,
1021 bool incremental)
1022{
1023 u32 n;
1024
1025 ceph_decode_32_safe(p, end, n, e_inval);
1026 while (n--) {
1027 struct ceph_pg pgid;
1028 u32 len, i;
1029 int ret;
1030
1031 ret = ceph_decode_pgid(p, end, &pgid);
1032 if (ret)
1033 return ret;
1034
1035 ceph_decode_32_safe(p, end, len, e_inval);
1036
33333d10 1037 ret = __remove_pg_mapping(&map->pg_temp, &pgid);
10db634e
ID		 1038 BUG_ON(!incremental && ret != -ENOENT);
1039
1040 if (!incremental || len > 0) {
1041 struct ceph_pg_mapping *pg;
1042
1043 ceph_decode_need(p, end, len*sizeof(u32), e_inval);
1044
1045 if (len > (UINT_MAX - sizeof(*pg)) / sizeof(u32))
1046 return -EINVAL;
1047
1048 pg = kzalloc(sizeof(*pg) + len*sizeof(u32), GFP_NOFS);
1049 if (!pg)
1050 return -ENOMEM;
1051
1052 pg->pgid = pgid;
35a935d7 1053 pg->pg_temp.len = len;
10db634e 1054 for (i = 0; i < len; i++)
35a935d7 1055 pg->pg_temp.osds[i] = ceph_decode_32(p);
10db634e
ID		 1056
1057 ret = __insert_pg_mapping(pg, &map->pg_temp);
1058 if (ret) {
1059 kfree(pg);
1060 return ret;
1061 }
1062 }
1063 }
1064
1065 return 0;
1066
1067e_inval:
1068 return -EINVAL;
1069}
1070
1071static int decode_pg_temp(void **p, void *end, struct ceph_osdmap *map)
1072{
1073 return __decode_pg_temp(p, end, map, false);
1074}
1075
1076static int decode_new_pg_temp(void **p, void *end, struct ceph_osdmap *map)
1077{
1078 return __decode_pg_temp(p, end, map, true);
1079}
1080
d286de79
ID		 1081static int __decode_primary_temp(void **p, void *end, struct ceph_osdmap *map,
1082 bool incremental)
1083{
1084 u32 n;
1085
1086 ceph_decode_32_safe(p, end, n, e_inval);
1087 while (n--) {
1088 struct ceph_pg pgid;
1089 u32 osd;
1090 int ret;
1091
1092 ret = ceph_decode_pgid(p, end, &pgid);
1093 if (ret)
1094 return ret;
1095
1096 ceph_decode_32_safe(p, end, osd, e_inval);
1097
33333d10 1098 ret = __remove_pg_mapping(&map->primary_temp, &pgid);
d286de79
ID		 1099 BUG_ON(!incremental && ret != -ENOENT);
1100
1101 if (!incremental || osd != (u32)-1) {
1102 struct ceph_pg_mapping *pg;
1103
1104 pg = kzalloc(sizeof(*pg), GFP_NOFS);
1105 if (!pg)
1106 return -ENOMEM;
1107
1108 pg->pgid = pgid;
1109 pg->primary_temp.osd = osd;
1110
1111 ret = __insert_pg_mapping(pg, &map->primary_temp);
1112 if (ret) {
1113 kfree(pg);
1114 return ret;
1115 }
1116 }
1117 }
1118
1119 return 0;
1120
1121e_inval:
1122 return -EINVAL;
1123}
1124
1125static int decode_primary_temp(void **p, void *end, struct ceph_osdmap *map)
1126{
1127 return __decode_primary_temp(p, end, map, false);
1128}
1129
1130static int decode_new_primary_temp(void **p, void *end,
1131 struct ceph_osdmap *map)
1132{
1133 return __decode_primary_temp(p, end, map, true);
1134}
1135
2cfa34f2
ID		 1136u32 ceph_get_primary_affinity(struct ceph_osdmap *map, int osd)
1137{
1138 BUG_ON(osd >= map->max_osd);
1139
1140 if (!map->osd_primary_affinity)
1141 return CEPH_OSD_DEFAULT_PRIMARY_AFFINITY;
1142
1143 return map->osd_primary_affinity[osd];
1144}
1145
1146static int set_primary_affinity(struct ceph_osdmap *map, int osd, u32 aff)
1147{
1148 BUG_ON(osd >= map->max_osd);
1149
1150 if (!map->osd_primary_affinity) {
1151 int i;
1152
1153 map->osd_primary_affinity = kmalloc(map->max_osd*sizeof(u32),
1154 GFP_NOFS);
1155 if (!map->osd_primary_affinity)
1156 return -ENOMEM;
1157
1158 for (i = 0; i < map->max_osd; i++)
1159 map->osd_primary_affinity[i] =
1160 CEPH_OSD_DEFAULT_PRIMARY_AFFINITY;
1161 }
1162
1163 map->osd_primary_affinity[osd] = aff;
1164
1165 return 0;
1166}
1167
63a6993f
ID		 1168static int decode_primary_affinity(void **p, void *end,
1169 struct ceph_osdmap *map)
1170{
1171 u32 len, i;
1172
1173 ceph_decode_32_safe(p, end, len, e_inval);
1174 if (len == 0) {
1175 kfree(map->osd_primary_affinity);
1176 map->osd_primary_affinity = NULL;
1177 return 0;
1178 }
1179 if (len != map->max_osd)
1180 goto e_inval;
1181
1182 ceph_decode_need(p, end, map->max_osd*sizeof(u32), e_inval);
1183
1184 for (i = 0; i < map->max_osd; i++) {
1185 int ret;
1186
1187 ret = set_primary_affinity(map, i, ceph_decode_32(p));
1188 if (ret)
1189 return ret;
1190 }
1191
1192 return 0;
1193
1194e_inval:
1195 return -EINVAL;
1196}
1197
1198static int decode_new_primary_affinity(void **p, void *end,
1199 struct ceph_osdmap *map)
1200{
1201 u32 n;
1202
1203 ceph_decode_32_safe(p, end, n, e_inval);
1204 while (n--) {
1205 u32 osd, aff;
1206 int ret;
1207
1208 ceph_decode_32_safe(p, end, osd, e_inval);
1209 ceph_decode_32_safe(p, end, aff, e_inval);
1210
1211 ret = set_primary_affinity(map, osd, aff);
1212 if (ret)
1213 return ret;
f31da0f3
ID		 1214
1215 pr_info("osd%d primary-affinity 0x%x\n", osd, aff);
63a6993f
ID		 1216 }
1217
1218 return 0;
1219
1220e_inval:
1221 return -EINVAL;
1222}
1223
f24e9980
SW		 1224/*
1225 * decode a full map.
1226 */
a2505d63 1227static int osdmap_decode(void **p, void *end, struct ceph_osdmap *map)
f24e9980 1228{
ec7af972 1229 u8 struct_v;
38a8d560 1230 u32 epoch = 0;
f24e9980 1231 void *start = *p;
3977058c
ID		 1232 u32 max;
1233 u32 len, i;
597b52f6 1234 int err;
f24e9980 1235
a2505d63 1236 dout("%s %p to %p len %d\n", __func__, *p, end, (int)(end - *p));
f24e9980 1237
ec7af972
ID		 1238 err = get_osdmap_client_data_v(p, end, "full", &struct_v);
1239 if (err)
1240 goto bad;
f24e9980 1241
53bbaba9
ID		 1242 /* fsid, epoch, created, modified */
1243 ceph_decode_need(p, end, sizeof(map->fsid) + sizeof(u32) +
1244 sizeof(map->created) + sizeof(map->modified), e_inval);
f24e9980 1245 ceph_decode_copy(p, &map->fsid, sizeof(map->fsid));
38a8d560 1246 epoch = map->epoch = ceph_decode_32(p);
f24e9980
SW		 1247 ceph_decode_copy(p, &map->created, sizeof(map->created));
1248 ceph_decode_copy(p, &map->modified, sizeof(map->modified));
1249
433fbdd3
ID		 1250 /* pools */
1251 err = decode_pools(p, end, map);
1252 if (err)
1253 goto bad;
2844a76a 1254
0f70c7ee
ID		 1255 /* pool_name */
1256 err = decode_pool_names(p, end, map);
597b52f6 1257 if (err)
4f6a7e5e 1258 goto bad;
2844a76a 1259
597b52f6 1260 ceph_decode_32_safe(p, end, map->pool_max, e_inval);
f24e9980 1261
597b52f6 1262 ceph_decode_32_safe(p, end, map->flags, e_inval);
f24e9980 1263
3977058c
ID		 1264 /* max_osd */
1265 ceph_decode_32_safe(p, end, max, e_inval);
f24e9980
SW		 1266
1267 /* (re)alloc osd arrays */
1268 err = osdmap_set_max_osd(map, max);
597b52f6 1269 if (err)
f24e9980 1270 goto bad;
f24e9980 1271
2d88b2e0 1272 /* osd_state, osd_weight, osd_addrs->client_addr */
f24e9980
SW		 1273 ceph_decode_need(p, end, 3*sizeof(u32) +
1274 map->max_osd*(1 + sizeof(*map->osd_weight) +
597b52f6
ID		 1275 sizeof(*map->osd_addr)), e_inval);
1276
2d88b2e0
ID		 1277 if (ceph_decode_32(p) != map->max_osd)
1278 goto e_inval;
1279
f24e9980
SW		 1280 ceph_decode_copy(p, map->osd_state, map->max_osd);
1281
2d88b2e0
ID		 1282 if (ceph_decode_32(p) != map->max_osd)
1283 goto e_inval;
1284
f24e9980 1285 for (i = 0; i < map->max_osd; i++)
c89136ea 1286 map->osd_weight[i] = ceph_decode_32(p);
f24e9980 1287
2d88b2e0
ID		 1288 if (ceph_decode_32(p) != map->max_osd)
1289 goto e_inval;
1290
f24e9980 1291 ceph_decode_copy(p, map->osd_addr, map->max_osd*sizeof(*map->osd_addr));
63f2d211
SW		 1292 for (i = 0; i < map->max_osd; i++)
1293 ceph_decode_addr(&map->osd_addr[i]);
f24e9980
SW		 1294
1295 /* pg_temp */
10db634e
ID		 1296 err = decode_pg_temp(p, end, map);
1297 if (err)
1298 goto bad;
f24e9980 1299
d286de79
ID		 1300 /* primary_temp */
1301 if (struct_v >= 1) {
1302 err = decode_primary_temp(p, end, map);
1303 if (err)
1304 goto bad;
1305 }
1306
63a6993f
ID		 1307 /* primary_affinity */
1308 if (struct_v >= 2) {
1309 err = decode_primary_affinity(p, end, map);
1310 if (err)
1311 goto bad;
1312 } else {
1313 /* XXX can this happen? */
1314 kfree(map->osd_primary_affinity);
1315 map->osd_primary_affinity = NULL;
1316 }
1317
f24e9980 1318 /* crush */
597b52f6 1319 ceph_decode_32_safe(p, end, len, e_inval);
1b6a78b5
ID		 1320 err = osdmap_set_crush(map, crush_decode(*p, min(*p + len, end)));
1321 if (err)
f24e9980 1322 goto bad;
f24e9980 1323
38a8d560 1324 /* ignore the rest */
f24e9980
SW		 1325 *p = end;
1326
38a8d560 1327 dout("full osdmap epoch %d max_osd %d\n", map->epoch, map->max_osd);
a2505d63 1328 return 0;
f24e9980 1329
597b52f6
ID		 1330e_inval:
1331 err = -EINVAL;
f24e9980 1332bad:
38a8d560
ID		 1333 pr_err("corrupt full osdmap (%d) epoch %d off %d (%p of %p-%p)\n",
1334 err, epoch, (int)(*p - start), *p, start, end);
1335 print_hex_dump(KERN_DEBUG, "osdmap: ",
1336 DUMP_PREFIX_OFFSET, 16, 1,
1337 start, end - start, true);
a2505d63
ID		 1338 return err;
1339}
1340
1341/*
1342 * Allocate and decode a full map.
1343 */
1344struct ceph_osdmap *ceph_osdmap_decode(void **p, void *end)
1345{
1346 struct ceph_osdmap *map;
1347 int ret;
1348
e5253a7b 1349 map = ceph_osdmap_alloc();
a2505d63
ID		 1350 if (!map)
1351 return ERR_PTR(-ENOMEM);
1352
a2505d63
ID		 1353 ret = osdmap_decode(p, end, map);
1354 if (ret) {
1355 ceph_osdmap_destroy(map);
1356 return ERR_PTR(ret);
1357 }
1358
1359 return map;
f24e9980
SW		 1360}
1361
930c5328
ID		 1362/*
1363 * Encoding order is (new_up_client, new_state, new_weight). Need to
1364 * apply in the (new_weight, new_state, new_up_client) order, because
1365 * an incremental map may look like e.g.
1366 *
1367 * new_up_client: { osd=6, addr=... } # set osd_state and addr
1368 * new_state: { osd=6, xorstate=EXISTS } # clear osd_state
1369 */
1370static int decode_new_up_state_weight(void **p, void *end,
1371 struct ceph_osdmap *map)
1372{
1373 void *new_up_client;
1374 void *new_state;
1375 void *new_weight_end;
1376 u32 len;
1377
1378 new_up_client = *p;
1379 ceph_decode_32_safe(p, end, len, e_inval);
1380 len *= sizeof(u32) + sizeof(struct ceph_entity_addr);
1381 ceph_decode_need(p, end, len, e_inval);
1382 *p += len;
1383
1384 new_state = *p;
1385 ceph_decode_32_safe(p, end, len, e_inval);
1386 len *= sizeof(u32) + sizeof(u8);
1387 ceph_decode_need(p, end, len, e_inval);
1388 *p += len;
1389
1390 /* new_weight */
1391 ceph_decode_32_safe(p, end, len, e_inval);
1392 while (len--) {
1393 s32 osd;
1394 u32 w;
1395
1396 ceph_decode_need(p, end, 2*sizeof(u32), e_inval);
1397 osd = ceph_decode_32(p);
1398 w = ceph_decode_32(p);
1399 BUG_ON(osd >= map->max_osd);
1400 pr_info("osd%d weight 0x%x %s\n", osd, w,
1401 w == CEPH_OSD_IN ? "(in)" :
1402 (w == CEPH_OSD_OUT ? "(out)" : ""));
1403 map->osd_weight[osd] = w;
1404
1405 /*
1406 * If we are marking in, set the EXISTS, and clear the
1407 * AUTOOUT and NEW bits.
1408 */
1409 if (w) {
1410 map->osd_state[osd] |= CEPH_OSD_EXISTS;
1411 map->osd_state[osd] &= ~(CEPH_OSD_AUTOOUT |
1412 CEPH_OSD_NEW);
1413 }
1414 }
1415 new_weight_end = *p;
1416
1417 /* new_state (up/down) */
1418 *p = new_state;
1419 len = ceph_decode_32(p);
1420 while (len--) {
1421 s32 osd;
1422 u8 xorstate;
1423 int ret;
1424
1425 osd = ceph_decode_32(p);
1426 xorstate = ceph_decode_8(p);
1427 if (xorstate == 0)
1428 xorstate = CEPH_OSD_UP;
1429 BUG_ON(osd >= map->max_osd);
1430 if ((map->osd_state[osd] & CEPH_OSD_UP) &&
1431 (xorstate & CEPH_OSD_UP))
1432 pr_info("osd%d down\n", osd);
1433 if ((map->osd_state[osd] & CEPH_OSD_EXISTS) &&
1434 (xorstate & CEPH_OSD_EXISTS)) {
1435 pr_info("osd%d does not exist\n", osd);
930c5328
ID		 1436 ret = set_primary_affinity(map, osd,
1437 CEPH_OSD_DEFAULT_PRIMARY_AFFINITY);
1438 if (ret)
1439 return ret;
1440 memset(map->osd_addr + osd, 0, sizeof(*map->osd_addr));
1441 map->osd_state[osd] = 0;
1442 } else {
1443 map->osd_state[osd] ^= xorstate;
1444 }
1445 }
1446
1447 /* new_up_client */
1448 *p = new_up_client;
1449 len = ceph_decode_32(p);
1450 while (len--) {
1451 s32 osd;
1452 struct ceph_entity_addr addr;
1453
1454 osd = ceph_decode_32(p);
1455 ceph_decode_copy(p, &addr, sizeof(addr));
1456 ceph_decode_addr(&addr);
1457 BUG_ON(osd >= map->max_osd);
1458 pr_info("osd%d up\n", osd);
1459 map->osd_state[osd] |= CEPH_OSD_EXISTS | CEPH_OSD_UP;
1460 map->osd_addr[osd] = addr;
1461 }
1462
1463 *p = new_weight_end;
1464 return 0;
1465
1466e_inval:
1467 return -EINVAL;
1468}
1469
f24e9980
SW		 1470/*
1471 * decode and apply an incremental map update.
1472 */
1473struct ceph_osdmap *osdmap_apply_incremental(void **p, void *end,
0c0a8de1 1474 struct ceph_osdmap *map)
f24e9980 1475{
f24e9980
SW		 1476 struct ceph_fsid fsid;
1477 u32 epoch = 0;
1478 struct ceph_timespec modified;
4f6a7e5e
SW		 1479 s32 len;
1480 u64 pool;
1481 __s64 new_pool_max;
1482 __s32 new_flags, max;
f24e9980 1483 void *start = *p;
86f1742b 1484 int err;
ec7af972 1485 u8 struct_v;
f24e9980 1486
38a8d560
ID		 1487 dout("%s %p to %p len %d\n", __func__, *p, end, (int)(end - *p));
1488
ec7af972
ID		 1489 err = get_osdmap_client_data_v(p, end, "inc", &struct_v);
1490 if (err)
1491 goto bad;
f24e9980 1492
53bbaba9
ID		 1493 /* fsid, epoch, modified, new_pool_max, new_flags */
1494 ceph_decode_need(p, end, sizeof(fsid) + sizeof(u32) + sizeof(modified) +
1495 sizeof(u64) + sizeof(u32), e_inval);
f24e9980 1496 ceph_decode_copy(p, &fsid, sizeof(fsid));
c89136ea 1497 epoch = ceph_decode_32(p);
f24e9980
SW		 1498 BUG_ON(epoch != map->epoch+1);
1499 ceph_decode_copy(p, &modified, sizeof(modified));
4f6a7e5e 1500 new_pool_max = ceph_decode_64(p);
c89136ea 1501 new_flags = ceph_decode_32(p);
f24e9980
SW		 1502
1503 /* full map? */
86f1742b 1504 ceph_decode_32_safe(p, end, len, e_inval);
f24e9980
SW		 1505 if (len > 0) {
1506 dout("apply_incremental full map len %d, %p to %p\n",
1507 len, *p, end);
a2505d63 1508 return ceph_osdmap_decode(p, min(*p+len, end));
f24e9980
SW		 1509 }
1510
1511 /* new crush? */
86f1742b 1512 ceph_decode_32_safe(p, end, len, e_inval);
f24e9980 1513 if (len > 0) {
1b6a78b5
ID		 1514 err = osdmap_set_crush(map,
1515 crush_decode(*p, min(*p + len, end)));
1516 if (err)
86f1742b 1517 goto bad;
cebc5be6 1518 *p += len;
f24e9980
SW		 1519 }
1520
1521 /* new flags? */
1522 if (new_flags >= 0)
1523 map->flags = new_flags;
4fc51be8
SW		 1524 if (new_pool_max >= 0)
1525 map->pool_max = new_pool_max;
f24e9980 1526
f24e9980 1527 /* new max? */
53bbaba9 1528 ceph_decode_32_safe(p, end, max, e_inval);
f24e9980
SW		 1529 if (max >= 0) {
1530 err = osdmap_set_max_osd(map, max);
86f1742b 1531 if (err)
f24e9980
SW		 1532 goto bad;
1533 }
1534
1535 map->epoch++;
31456665 1536 map->modified = modified;
f24e9980 1537
433fbdd3
ID		 1538 /* new_pools */
1539 err = decode_new_pools(p, end, map);
1540 if (err)
1541 goto bad;
9464d008 1542
0f70c7ee
ID		 1543 /* new_pool_names */
1544 err = decode_pool_names(p, end, map);
9464d008
ID		 1545 if (err)
1546 goto bad;
f24e9980 1547
4fc51be8 1548 /* old_pool */
86f1742b 1549 ceph_decode_32_safe(p, end, len, e_inval);
4fc51be8
SW		 1550 while (len--) {
1551 struct ceph_pg_pool_info *pi;
1552
86f1742b 1553 ceph_decode_64_safe(p, end, pool, e_inval);
4fc51be8 1554 pi = __lookup_pg_pool(&map->pg_pools, pool);
2844a76a
SW		 1555 if (pi)
1556 __remove_pg_pool(&map->pg_pools, pi);
4fc51be8 1557 }
f24e9980 1558
930c5328
ID		 1559 /* new_up_client, new_state, new_weight */
1560 err = decode_new_up_state_weight(p, end, map);
1561 if (err)
1562 goto bad;
f24e9980
SW		 1563
1564 /* new_pg_temp */
10db634e
ID		 1565 err = decode_new_pg_temp(p, end, map);
1566 if (err)
1567 goto bad;
f24e9980 1568
d286de79
ID		 1569 /* new_primary_temp */
1570 if (struct_v >= 1) {
1571 err = decode_new_primary_temp(p, end, map);
1572 if (err)
1573 goto bad;
1574 }
1575
63a6993f
ID		 1576 /* new_primary_affinity */
1577 if (struct_v >= 2) {
1578 err = decode_new_primary_affinity(p, end, map);
1579 if (err)
1580 goto bad;
1581 }
1582
f24e9980
SW		 1583 /* ignore the rest */
1584 *p = end;
38a8d560
ID		 1585
1586 dout("inc osdmap epoch %d max_osd %d\n", map->epoch, map->max_osd);
f24e9980
SW		 1587 return map;
1588
86f1742b
ID		 1589e_inval:
1590 err = -EINVAL;
f24e9980 1591bad:
38a8d560
ID		 1592 pr_err("corrupt inc osdmap (%d) epoch %d off %d (%p of %p-%p)\n",
1593 err, epoch, (int)(*p - start), *p, start, end);
9ec7cab1
SW		 1594 print_hex_dump(KERN_DEBUG, "osdmap: ",
1595 DUMP_PREFIX_OFFSET, 16, 1,
1596 start, end - start, true);
f24e9980
SW		 1597 return ERR_PTR(err);
1598}
1599
30c156d9
YZ		 1600void ceph_oloc_copy(struct ceph_object_locator *dest,
1601 const struct ceph_object_locator *src)
1602{
ca35ffea 1603 ceph_oloc_destroy(dest);
30c156d9
YZ		 1604
1605 dest->pool = src->pool;
1606 if (src->pool_ns)
1607 dest->pool_ns = ceph_get_string(src->pool_ns);
ca35ffea
ID		 1608 else
1609 dest->pool_ns = NULL;
30c156d9
YZ		 1610}
1611EXPORT_SYMBOL(ceph_oloc_copy);
1612
1613void ceph_oloc_destroy(struct ceph_object_locator *oloc)
1614{
1615 ceph_put_string(oloc->pool_ns);
1616}
1617EXPORT_SYMBOL(ceph_oloc_destroy);
1618
d30291b9
ID		 1619void ceph_oid_copy(struct ceph_object_id *dest,
1620 const struct ceph_object_id *src)
1621{
ca35ffea 1622 ceph_oid_destroy(dest);
d30291b9
ID		 1623
1624 if (src->name != src->inline_name) {
1625 /* very rare, see ceph_object_id definition */
1626 dest->name = kmalloc(src->name_len + 1,
1627 GFP_NOIO | __GFP_NOFAIL);
ca35ffea
ID		 1628 } else {
1629 dest->name = dest->inline_name;
d30291b9 1630 }
d30291b9
ID		 1631 memcpy(dest->name, src->name, src->name_len + 1);
1632 dest->name_len = src->name_len;
1633}
1634EXPORT_SYMBOL(ceph_oid_copy);
1635
1636static __printf(2, 0)
1637int oid_printf_vargs(struct ceph_object_id *oid, const char *fmt, va_list ap)
1638{
1639 int len;
1640
1641 WARN_ON(!ceph_oid_empty(oid));
1642
1643 len = vsnprintf(oid->inline_name, sizeof(oid->inline_name), fmt, ap);
1644 if (len >= sizeof(oid->inline_name))
1645 return len;
1646
1647 oid->name_len = len;
1648 return 0;
1649}
1650
1651/*
1652 * If oid doesn't fit into inline buffer, BUG.
1653 */
1654void ceph_oid_printf(struct ceph_object_id *oid, const char *fmt, ...)
1655{
1656 va_list ap;
1657
1658 va_start(ap, fmt);
1659 BUG_ON(oid_printf_vargs(oid, fmt, ap));
1660 va_end(ap);
1661}
1662EXPORT_SYMBOL(ceph_oid_printf);
1663
1664static __printf(3, 0)
1665int oid_aprintf_vargs(struct ceph_object_id *oid, gfp_t gfp,
1666 const char *fmt, va_list ap)
1667{
1668 va_list aq;
1669 int len;
1670
1671 va_copy(aq, ap);
1672 len = oid_printf_vargs(oid, fmt, aq);
1673 va_end(aq);
1674
1675 if (len) {
1676 char *external_name;
1677
1678 external_name = kmalloc(len + 1, gfp);
1679 if (!external_name)
1680 return -ENOMEM;
f24e9980 1681
d30291b9
ID		 1682 oid->name = external_name;
1683 WARN_ON(vsnprintf(oid->name, len + 1, fmt, ap) != len);
1684 oid->name_len = len;
1685 }
f24e9980 1686
d30291b9
ID		 1687 return 0;
1688}
1689
1690/*
1691 * If oid doesn't fit into inline buffer, allocate.
1692 */
1693int ceph_oid_aprintf(struct ceph_object_id *oid, gfp_t gfp,
1694 const char *fmt, ...)
1695{
1696 va_list ap;
1697 int ret;
1698
1699 va_start(ap, fmt);
1700 ret = oid_aprintf_vargs(oid, gfp, fmt, ap);
1701 va_end(ap);
1702
1703 return ret;
1704}
1705EXPORT_SYMBOL(ceph_oid_aprintf);
1706
1707void ceph_oid_destroy(struct ceph_object_id *oid)
1708{
1709 if (oid->name != oid->inline_name)
1710 kfree(oid->name);
1711}
1712EXPORT_SYMBOL(ceph_oid_destroy);
f24e9980 1713
63244fa1
ID		 1714/*
1715 * osds only
1716 */
1717static bool __osds_equal(const struct ceph_osds *lhs,
1718 const struct ceph_osds *rhs)
1719{
1720 if (lhs->size == rhs->size &&
1721 !memcmp(lhs->osds, rhs->osds, rhs->size * sizeof(rhs->osds[0])))
1722 return true;
1723
1724 return false;
1725}
1726
1727/*
1728 * osds + primary
1729 */
1730static bool osds_equal(const struct ceph_osds *lhs,
1731 const struct ceph_osds *rhs)
1732{
1733 if (__osds_equal(lhs, rhs) &&
1734 lhs->primary == rhs->primary)
1735 return true;
1736
1737 return false;
1738}
1739
6f3bfd45
ID		 1740static bool osds_valid(const struct ceph_osds *set)
1741{
1742 /* non-empty set */
1743 if (set->size > 0 && set->primary >= 0)
1744 return true;
1745
1746 /* empty can_shift_osds set */
1747 if (!set->size && set->primary == -1)
1748 return true;
1749
1750 /* empty !can_shift_osds set - all NONE */
1751 if (set->size > 0 && set->primary == -1) {
1752 int i;
1753
1754 for (i = 0; i < set->size; i++) {
1755 if (set->osds[i] != CRUSH_ITEM_NONE)
1756 break;
1757 }
1758 if (i == set->size)
1759 return true;
1760 }
1761
1762 return false;
1763}
1764
1765void ceph_osds_copy(struct ceph_osds *dest, const struct ceph_osds *src)
1766{
1767 memcpy(dest->osds, src->osds, src->size * sizeof(src->osds[0]));
1768 dest->size = src->size;
1769 dest->primary = src->primary;
1770}
1771
7de030d6
ID		 1772bool ceph_pg_is_split(const struct ceph_pg *pgid, u32 old_pg_num,
1773 u32 new_pg_num)
63244fa1
ID		 1774{
1775 int old_bits = calc_bits_of(old_pg_num);
1776 int old_mask = (1 << old_bits) - 1;
1777 int n;
1778
1779 WARN_ON(pgid->seed >= old_pg_num);
1780 if (new_pg_num <= old_pg_num)
1781 return false;
1782
1783 for (n = 1; ; n++) {
1784 int next_bit = n << (old_bits - 1);
1785 u32 s = next_bit | pgid->seed;
1786
1787 if (s < old_pg_num || s == pgid->seed)
1788 continue;
1789 if (s >= new_pg_num)
1790 break;
1791
1792 s = ceph_stable_mod(s, old_pg_num, old_mask);
1793 if (s == pgid->seed)
1794 return true;
1795 }
1796
1797 return false;
1798}
1799
1800bool ceph_is_new_interval(const struct ceph_osds *old_acting,
1801 const struct ceph_osds *new_acting,
1802 const struct ceph_osds *old_up,
1803 const struct ceph_osds *new_up,
1804 int old_size,
1805 int new_size,
1806 int old_min_size,
1807 int new_min_size,
1808 u32 old_pg_num,
1809 u32 new_pg_num,
1810 bool old_sort_bitwise,
1811 bool new_sort_bitwise,
1812 const struct ceph_pg *pgid)
1813{
1814 return !osds_equal(old_acting, new_acting) ||
1815 !osds_equal(old_up, new_up) ||
1816 old_size != new_size ||
1817 old_min_size != new_min_size ||
7de030d6 1818 ceph_pg_is_split(pgid, old_pg_num, new_pg_num) ||
63244fa1
ID		 1819 old_sort_bitwise != new_sort_bitwise;
1820}
1821
1822static int calc_pg_rank(int osd, const struct ceph_osds *acting)
1823{
1824 int i;
1825
1826 for (i = 0; i < acting->size; i++) {
1827 if (acting->osds[i] == osd)
1828 return i;
1829 }
1830
1831 return -1;
1832}
1833
1834static bool primary_changed(const struct ceph_osds *old_acting,
1835 const struct ceph_osds *new_acting)
1836{
1837 if (!old_acting->size && !new_acting->size)
1838 return false; /* both still empty */
1839
1840 if (!old_acting->size ^ !new_acting->size)
1841 return true; /* was empty, now not, or vice versa */
1842
1843 if (old_acting->primary != new_acting->primary)
1844 return true; /* primary changed */
1845
1846 if (calc_pg_rank(old_acting->primary, old_acting) !=
1847 calc_pg_rank(new_acting->primary, new_acting))
1848 return true;
1849
1850 return false; /* same primary (tho replicas may have changed) */
1851}
1852
1853bool ceph_osds_changed(const struct ceph_osds *old_acting,
1854 const struct ceph_osds *new_acting,
1855 bool any_change)
1856{
1857 if (primary_changed(old_acting, new_acting))
1858 return true;
1859
1860 if (any_change && !__osds_equal(old_acting, new_acting))
1861 return true;
1862
1863 return false;
1864}
1865
f24e9980
SW		 1866/*
1867 * calculate file layout from given offset, length.
1868 * fill in correct oid, logical length, and object extent
1869 * offset, length.
1870 *
1871 * for now, we write only a single su, until we can
1872 * pass a stride back to the caller.
1873 */
d63b77f4 1874int ceph_calc_file_object_mapping(struct ceph_file_layout *layout,
e8afad65 1875 u64 off, u64 len,
645a1025 1876 u64 *ono,
f24e9980
SW		 1877 u64 *oxoff, u64 *oxlen)
1878{
7627151e
YZ		 1879 u32 osize = layout->object_size;
1880 u32 su = layout->stripe_unit;
1881 u32 sc = layout->stripe_count;
f24e9980
SW		 1882 u32 bl, stripeno, stripepos, objsetno;
1883 u32 su_per_object;
ff1d1f71 1884 u64 t, su_offset;
f24e9980 1885
e8afad65 1886 dout("mapping %llu~%llu osize %u fl_su %u\n", off, len,
f24e9980 1887 osize, su);
d63b77f4
SW		 1888 if (su == 0 || sc == 0)
1889 goto invalid;
35e054a6 1890 su_per_object = osize / su;
d63b77f4
SW		 1891 if (su_per_object == 0)
1892 goto invalid;
f24e9980
SW		 1893 dout("osize %u / su %u = su_per_object %u\n", osize, su,
1894 su_per_object);
1895
d63b77f4
SW		 1896 if ((su & ~PAGE_MASK) != 0)
1897 goto invalid;
1898
f24e9980
SW		 1899 /* bl = *off / su; */
1900 t = off;
1901 do_div(t, su);
1902 bl = t;
1903 dout("off %llu / su %u = bl %u\n", off, su, bl);
1904
1905 stripeno = bl / sc;
1906 stripepos = bl % sc;
1907 objsetno = stripeno / su_per_object;
1908
645a1025 1909 *ono = objsetno * sc + stripepos;
95c96174 1910 dout("objset %u * sc %u = ono %u\n", objsetno, sc, (unsigned int)*ono);
645a1025
SW		 1911
1912 /* *oxoff = *off % layout->fl_stripe_unit; # offset in su */
f24e9980 1913 t = off;
ff1d1f71
NW		 1914 su_offset = do_div(t, su);
1915 *oxoff = su_offset + (stripeno % su_per_object) * su;
1916
1917 /*
1918 * Calculate the length of the extent being written to the selected
e8afad65 1919 * object. This is the minimum of the full length requested (len) or
ff1d1f71
NW		 1920 * the remainder of the current stripe being written to.
1921 */
e8afad65 1922 *oxlen = min_t(u64, len, su - su_offset);
f24e9980
SW		 1923
1924 dout(" obj extent %llu~%llu\n", *oxoff, *oxlen);
d63b77f4
SW		 1925 return 0;
1926
1927invalid:
1928 dout(" invalid layout\n");
1929 *ono = 0;
1930 *oxoff = 0;
1931 *oxlen = 0;
1932 return -EINVAL;
f24e9980 1933}
3d14c5d2 1934EXPORT_SYMBOL(ceph_calc_file_object_mapping);
f24e9980
SW		 1935
1936/*
d9591f5e
ID		 1937 * Map an object into a PG.
1938 *
1939 * Should only be called with target_oid and target_oloc (as opposed to
1940 * base_oid and base_oloc), since tiering isn't taken into account.
f24e9980 1941 */
df28152d
ID		 1942int __ceph_object_locator_to_pg(struct ceph_pg_pool_info *pi,
1943 const struct ceph_object_id *oid,
1944 const struct ceph_object_locator *oloc,
1945 struct ceph_pg *raw_pgid)
f24e9980 1946{
df28152d 1947 WARN_ON(pi->id != oloc->pool);
f24e9980 1948
30c156d9
YZ		 1949 if (!oloc->pool_ns) {
1950 raw_pgid->pool = oloc->pool;
1951 raw_pgid->seed = ceph_str_hash(pi->object_hash, oid->name,
1952 oid->name_len);
1953 dout("%s %s -> raw_pgid %llu.%x\n", __func__, oid->name,
1954 raw_pgid->pool, raw_pgid->seed);
1955 } else {
1956 char stack_buf[256];
1957 char *buf = stack_buf;
1958 int nsl = oloc->pool_ns->len;
1959 size_t total = nsl + 1 + oid->name_len;
1960
1961 if (total > sizeof(stack_buf)) {
1962 buf = kmalloc(total, GFP_NOIO);
1963 if (!buf)
1964 return -ENOMEM;
1965 }
1966 memcpy(buf, oloc->pool_ns->str, nsl);
1967 buf[nsl] = '\037';
1968 memcpy(buf + nsl + 1, oid->name, oid->name_len);
1969 raw_pgid->pool = oloc->pool;
1970 raw_pgid->seed = ceph_str_hash(pi->object_hash, buf, total);
1971 if (buf != stack_buf)
1972 kfree(buf);
1973 dout("%s %s ns %.*s -> raw_pgid %llu.%x\n", __func__,
1974 oid->name, nsl, oloc->pool_ns->str,
1975 raw_pgid->pool, raw_pgid->seed);
1976 }
f24e9980
SW		 1977 return 0;
1978}
df28152d
ID		 1979
1980int ceph_object_locator_to_pg(struct ceph_osdmap *osdmap,
1981 const struct ceph_object_id *oid,
1982 const struct ceph_object_locator *oloc,
1983 struct ceph_pg *raw_pgid)
1984{
1985 struct ceph_pg_pool_info *pi;
1986
1987 pi = ceph_pg_pool_by_id(osdmap, oloc->pool);
1988 if (!pi)
1989 return -ENOENT;
1990
1991 return __ceph_object_locator_to_pg(pi, oid, oloc, raw_pgid);
1992}
d9591f5e 1993EXPORT_SYMBOL(ceph_object_locator_to_pg);
f24e9980 1994
6f3bfd45
ID		 1995/*
1996 * Map a raw PG (full precision ps) into an actual PG.
1997 */
1998static void raw_pg_to_pg(struct ceph_pg_pool_info *pi,
1999 const struct ceph_pg *raw_pgid,
2000 struct ceph_pg *pgid)
2001{
2002 pgid->pool = raw_pgid->pool;
2003 pgid->seed = ceph_stable_mod(raw_pgid->seed, pi->pg_num,
2004 pi->pg_num_mask);
2005}
2006
2007/*
2008 * Map a raw PG (full precision ps) into a placement ps (placement
2009 * seed). Include pool id in that value so that different pools don't
2010 * use the same seeds.
2011 */
2012static u32 raw_pg_to_pps(struct ceph_pg_pool_info *pi,
2013 const struct ceph_pg *raw_pgid)
2014{
2015 if (pi->flags & CEPH_POOL_FLAG_HASHPSPOOL) {
2016 /* hash pool id and seed so that pool PGs do not overlap */
2017 return crush_hash32_2(CRUSH_HASH_RJENKINS1,
2018 ceph_stable_mod(raw_pgid->seed,
2019 pi->pgp_num,
2020 pi->pgp_num_mask),
2021 raw_pgid->pool);
2022 } else {
2023 /*
2024 * legacy behavior: add ps and pool together. this is
2025 * not a great approach because the PGs from each pool
2026 * will overlap on top of each other: 0.5 == 1.4 ==
2027 * 2.3 == ...
2028 */
2029 return ceph_stable_mod(raw_pgid->seed, pi->pgp_num,
2030 pi->pgp_num_mask) +
2031 (unsigned)raw_pgid->pool;
2032 }
2033}
2034
9d521470
ID		 2035static int do_crush(struct ceph_osdmap *map, int ruleno, int x,
2036 int *result, int result_max,
2037 const __u32 *weight, int weight_max)
e8ef19c4 2038{
9d521470
ID		 2039 int r;
2040
2041 BUG_ON(result_max > CEPH_PG_MAX_SIZE);
2042
743efcff 2043 mutex_lock(&map->crush_workspace_mutex);
9d521470 2044 r = crush_do_rule(map->crush, ruleno, x, result, result_max,
743efcff
ID		 2045 weight, weight_max, map->crush_workspace);
2046 mutex_unlock(&map->crush_workspace_mutex);
e8ef19c4 2047
9d521470 2048 return r;
e8ef19c4
ID		 2049}
2050
2bd93d4d 2051/*
6f3bfd45
ID		 2052 * Calculate raw set (CRUSH output) for given PG. The result may
2053 * contain nonexistent OSDs. ->primary is undefined for a raw set.
2bd93d4d 2054 *
6f3bfd45 2055 * Placement seed (CRUSH input) is returned through @ppps.
2bd93d4d 2056 */
6f3bfd45
ID		 2057static void pg_to_raw_osds(struct ceph_osdmap *osdmap,
2058 struct ceph_pg_pool_info *pi,
2059 const struct ceph_pg *raw_pgid,
2060 struct ceph_osds *raw,
2061 u32 *ppps)
2bd93d4d 2062{
6f3bfd45 2063 u32 pps = raw_pg_to_pps(pi, raw_pgid);
2bd93d4d
ID		 2064 int ruleno;
2065 int len;
2066
6f3bfd45
ID		 2067 ceph_osds_init(raw);
2068 if (ppps)
2069 *ppps = pps;
2070
2071 ruleno = crush_find_rule(osdmap->crush, pi->crush_ruleset, pi->type,
2072 pi->size);
2bd93d4d
ID		 2073 if (ruleno < 0) {
2074 pr_err("no crush rule: pool %lld ruleset %d type %d size %d\n",
6f3bfd45
ID		 2075 pi->id, pi->crush_ruleset, pi->type, pi->size);
2076 return;
2bd93d4d
ID		 2077 }
2078
ef9324bb
ID		 2079 if (pi->size > ARRAY_SIZE(raw->osds)) {
2080 pr_err_ratelimited("pool %lld ruleset %d type %d too wide: size %d > %zu\n",
2081 pi->id, pi->crush_ruleset, pi->type, pi->size,
2082 ARRAY_SIZE(raw->osds));
2083 return;
2084 }
2085
2086 len = do_crush(osdmap, ruleno, pps, raw->osds, pi->size,
2bd93d4d
ID		 2087 osdmap->osd_weight, osdmap->max_osd);
2088 if (len < 0) {
2089 pr_err("error %d from crush rule %d: pool %lld ruleset %d type %d size %d\n",
6f3bfd45
ID		 2090 len, ruleno, pi->id, pi->crush_ruleset, pi->type,
2091 pi->size);
2092 return;
2bd93d4d
ID		 2093 }
2094
6f3bfd45 2095 raw->size = len;
2bd93d4d
ID		 2096}
2097
2098/*
6f3bfd45
ID		 2099 * Given raw set, calculate up set and up primary. By definition of an
2100 * up set, the result won't contain nonexistent or down OSDs.
2bd93d4d 2101 *
6f3bfd45
ID		 2102 * This is done in-place - on return @set is the up set. If it's
2103 * empty, ->primary will remain undefined.
2bd93d4d 2104 */
6f3bfd45
ID		 2105static void raw_to_up_osds(struct ceph_osdmap *osdmap,
2106 struct ceph_pg_pool_info *pi,
2107 struct ceph_osds *set)
2bd93d4d 2108{
2bd93d4d
ID		 2109 int i;
2110
6f3bfd45
ID		 2111 /* ->primary is undefined for a raw set */
2112 BUG_ON(set->primary != -1);
2113
2114 if (ceph_can_shift_osds(pi)) {
2bd93d4d
ID		 2115 int removed = 0;
2116
6f3bfd45
ID		 2117 /* shift left */
2118 for (i = 0; i < set->size; i++) {
2119 if (ceph_osd_is_down(osdmap, set->osds[i])) {
2bd93d4d
ID		 2120 removed++;
2121 continue;
2122 }
2123 if (removed)
6f3bfd45 2124 set->osds[i - removed] = set->osds[i];
2bd93d4d 2125 }
6f3bfd45
ID		 2126 set->size -= removed;
2127 if (set->size > 0)
2128 set->primary = set->osds[0];
2bd93d4d 2129 } else {
6f3bfd45
ID		 2130 /* set down/dne devices to NONE */
2131 for (i = set->size - 1; i >= 0; i--) {
2132 if (ceph_osd_is_down(osdmap, set->osds[i]))
2133 set->osds[i] = CRUSH_ITEM_NONE;
2bd93d4d 2134 else
6f3bfd45 2135 set->primary = set->osds[i];
2bd93d4d
ID		 2136 }
2137 }
2bd93d4d
ID		 2138}
2139
6f3bfd45
ID		 2140static void apply_primary_affinity(struct ceph_osdmap *osdmap,
2141 struct ceph_pg_pool_info *pi,
2142 u32 pps,
2143 struct ceph_osds *up)
47ec1f3c
ID		 2144{
2145 int i;
2146 int pos = -1;
2147
2148 /*
2149 * Do we have any non-default primary_affinity values for these
2150 * osds?
2151 */
2152 if (!osdmap->osd_primary_affinity)
2153 return;
2154
6f3bfd45
ID		 2155 for (i = 0; i < up->size; i++) {
2156 int osd = up->osds[i];
92b2e751
ID		 2157
2158 if (osd != CRUSH_ITEM_NONE &&
2159 osdmap->osd_primary_affinity[osd] !=
47ec1f3c
ID		 2160 CEPH_OSD_DEFAULT_PRIMARY_AFFINITY) {
2161 break;
2162 }
2163 }
6f3bfd45 2164 if (i == up->size)
47ec1f3c
ID		 2165 return;
2166
2167 /*
2168 * Pick the primary. Feed both the seed (for the pg) and the
2169 * osd into the hash/rng so that a proportional fraction of an
2170 * osd's pgs get rejected as primary.
2171 */
6f3bfd45
ID		 2172 for (i = 0; i < up->size; i++) {
2173 int osd = up->osds[i];
47ec1f3c
ID		 2174 u32 aff;
2175
47ec1f3c
ID		 2176 if (osd == CRUSH_ITEM_NONE)
2177 continue;
2178
2179 aff = osdmap->osd_primary_affinity[osd];
2180 if (aff < CEPH_OSD_MAX_PRIMARY_AFFINITY &&
2181 (crush_hash32_2(CRUSH_HASH_RJENKINS1,
2182 pps, osd) >> 16) >= aff) {
2183 /*
2184 * We chose not to use this primary. Note it
2185 * anyway as a fallback in case we don't pick
2186 * anyone else, but keep looking.
2187 */
2188 if (pos < 0)
2189 pos = i;
2190 } else {
2191 pos = i;
2192 break;
2193 }
2194 }
2195 if (pos < 0)
2196 return;
2197
6f3bfd45 2198 up->primary = up->osds[pos];
47ec1f3c 2199
6f3bfd45 2200 if (ceph_can_shift_osds(pi) && pos > 0) {
47ec1f3c
ID		 2201 /* move the new primary to the front */
2202 for (i = pos; i > 0; i--)
6f3bfd45
ID		 2203 up->osds[i] = up->osds[i - 1];
2204 up->osds[0] = up->primary;
47ec1f3c
ID		 2205 }
2206}
2207
45966c34 2208/*
6f3bfd45 2209 * Get pg_temp and primary_temp mappings for given PG.
45966c34 2210 *
6f3bfd45
ID		 2211 * Note that a PG may have none, only pg_temp, only primary_temp or
2212 * both pg_temp and primary_temp mappings. This means @temp isn't
2213 * always a valid OSD set on return: in the "only primary_temp" case,
2214 * @temp will have its ->primary >= 0 but ->size == 0.
45966c34 2215 */
6f3bfd45
ID		 2216static void get_temp_osds(struct ceph_osdmap *osdmap,
2217 struct ceph_pg_pool_info *pi,
2218 const struct ceph_pg *raw_pgid,
2219 struct ceph_osds *temp)
45966c34 2220{
6f3bfd45 2221 struct ceph_pg pgid;
45966c34 2222 struct ceph_pg_mapping *pg;
45966c34
ID		 2223 int i;
2224
6f3bfd45
ID		 2225 raw_pg_to_pg(pi, raw_pgid, &pgid);
2226 ceph_osds_init(temp);
45966c34
ID		 2227
2228 /* pg_temp? */
33333d10 2229 pg = __lookup_pg_mapping(&osdmap->pg_temp, &pgid);
45966c34 2230 if (pg) {
45966c34
ID		 2231 for (i = 0; i < pg->pg_temp.len; i++) {
2232 if (ceph_osd_is_down(osdmap, pg->pg_temp.osds[i])) {
6f3bfd45 2233 if (ceph_can_shift_osds(pi))
45966c34 2234 continue;
6f3bfd45
ID		 2235
2236 temp->osds[temp->size++] = CRUSH_ITEM_NONE;
45966c34 2237 } else {
6f3bfd45 2238 temp->osds[temp->size++] = pg->pg_temp.osds[i];
45966c34
ID		 2239 }
2240 }
2241
2242 /* apply pg_temp's primary */
6f3bfd45
ID		 2243 for (i = 0; i < temp->size; i++) {
2244 if (temp->osds[i] != CRUSH_ITEM_NONE) {
2245 temp->primary = temp->osds[i];
45966c34
ID		 2246 break;
2247 }
2248 }
45966c34
ID		 2249 }
2250
5e8d4d36 2251 /* primary_temp? */
33333d10 2252 pg = __lookup_pg_mapping(&osdmap->primary_temp, &pgid);
5e8d4d36 2253 if (pg)
6f3bfd45 2254 temp->primary = pg->primary_temp.osd;
45966c34
ID		 2255}
2256
d85b7056 2257/*
6f3bfd45 2258 * Map a PG to its acting set as well as its up set.
ac972230 2259 *
6f3bfd45
ID		 2260 * Acting set is used for data mapping purposes, while up set can be
2261 * recorded for detecting interval changes and deciding whether to
2262 * resend a request.
d85b7056 2263 */
6f3bfd45 2264void ceph_pg_to_up_acting_osds(struct ceph_osdmap *osdmap,
df28152d 2265 struct ceph_pg_pool_info *pi,
6f3bfd45
ID		 2266 const struct ceph_pg *raw_pgid,
2267 struct ceph_osds *up,
2268 struct ceph_osds *acting)
d85b7056 2269{
ac972230 2270 u32 pps;
d85b7056 2271
df28152d 2272 WARN_ON(pi->id != raw_pgid->pool);
d85b7056 2273
6f3bfd45
ID		 2274 pg_to_raw_osds(osdmap, pi, raw_pgid, up, &pps);
2275 raw_to_up_osds(osdmap, pi, up);
2276 apply_primary_affinity(osdmap, pi, pps, up);
2277 get_temp_osds(osdmap, pi, raw_pgid, acting);
2278 if (!acting->size) {
2279 memcpy(acting->osds, up->osds, up->size * sizeof(up->osds[0]));
2280 acting->size = up->size;
2281 if (acting->primary == -1)
2282 acting->primary = up->primary;
8008ab10 2283 }
6f3bfd45 2284 WARN_ON(!osds_valid(up) || !osds_valid(acting));
d85b7056
SW		 2285}
2286
dc98ff72 2287bool ceph_pg_to_primary_shard(struct ceph_osdmap *osdmap,
df28152d 2288 struct ceph_pg_pool_info *pi,
dc98ff72
ID		 2289 const struct ceph_pg *raw_pgid,
2290 struct ceph_spg *spgid)
2291{
dc98ff72
ID		 2292 struct ceph_pg pgid;
2293 struct ceph_osds up, acting;
2294 int i;
2295
df28152d 2296 WARN_ON(pi->id != raw_pgid->pool);
dc98ff72
ID		 2297 raw_pg_to_pg(pi, raw_pgid, &pgid);
2298
2299 if (ceph_can_shift_osds(pi)) {
2300 spgid->pgid = pgid; /* struct */
2301 spgid->shard = CEPH_SPG_NOSHARD;
2302 return true;
2303 }
2304
df28152d 2305 ceph_pg_to_up_acting_osds(osdmap, pi, &pgid, &up, &acting);
dc98ff72
ID		 2306 for (i = 0; i < acting.size; i++) {
2307 if (acting.osds[i] == acting.primary) {
2308 spgid->pgid = pgid; /* struct */
2309 spgid->shard = i;
2310 return true;
2311 }
2312 }
2313
2314 return false;
2315}
2316
f24e9980 2317/*
f81f1633 2318 * Return acting primary for given PG, or -1 if none.
f24e9980 2319 */
f81f1633
ID		 2320int ceph_pg_to_acting_primary(struct ceph_osdmap *osdmap,
2321 const struct ceph_pg *raw_pgid)
f24e9980 2322{
df28152d 2323 struct ceph_pg_pool_info *pi;
6f3bfd45 2324 struct ceph_osds up, acting;
f24e9980 2325
df28152d
ID		 2326 pi = ceph_pg_pool_by_id(osdmap, raw_pgid->pool);
2327 if (!pi)
2328 return -1;
2329
2330 ceph_pg_to_up_acting_osds(osdmap, pi, raw_pgid, &up, &acting);
6f3bfd45 2331 return acting.primary;
f24e9980 2332}
f81f1633 2333EXPORT_SYMBOL(ceph_pg_to_acting_primary);
Linux 6.x block layer and io_uring tree(s)