]> git.proxmox.com Git - mirror_ubuntu-zesty-kernel.git/blame - net/ceph/osdmap.c
projects / mirror_ubuntu-zesty-kernel.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
KVM: arm64: vgic-v3: Log which GICv3 system registers are trapped
[mirror_ubuntu-zesty-kernel.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
f24e9980
SW		 156static struct crush_map *crush_decode(void *pbyval, void *end)
157{
158 struct crush_map *c;
159 int err = -EINVAL;
160 int i, j;
161 void **p = &pbyval;
162 void *start = pbyval;
163 u32 magic;
546f04ef 164 u32 num_name_maps;
f24e9980
SW		 165
166 dout("crush_decode %p to %p len %d\n", *p, end, (int)(end - *p));
167
168 c = kzalloc(sizeof(*c), GFP_NOFS);
169 if (c == NULL)
170 return ERR_PTR(-ENOMEM);
171
546f04ef
SW		 172 /* set tunables to default values */
173 c->choose_local_tries = 2;
174 c->choose_local_fallback_tries = 5;
175 c->choose_total_tries = 19;
1604f488 176 c->chooseleaf_descend_once = 0;
546f04ef 177
f24e9980 178 ceph_decode_need(p, end, 4*sizeof(u32), bad);
c89136ea 179 magic = ceph_decode_32(p);
f24e9980
SW		 180 if (magic != CRUSH_MAGIC) {
181 pr_err("crush_decode magic %x != current %x\n",
95c96174 182 (unsigned int)magic, (unsigned int)CRUSH_MAGIC);
f24e9980
SW		 183 goto bad;
184 }
c89136ea
SW		 185 c->max_buckets = ceph_decode_32(p);
186 c->max_rules = ceph_decode_32(p);
187 c->max_devices = ceph_decode_32(p);
f24e9980 188
f24e9980
SW		 189 c->buckets = kcalloc(c->max_buckets, sizeof(*c->buckets), GFP_NOFS);
190 if (c->buckets == NULL)
191 goto badmem;
192 c->rules = kcalloc(c->max_rules, sizeof(*c->rules), GFP_NOFS);
193 if (c->rules == NULL)
194 goto badmem;
195
196 /* buckets */
197 for (i = 0; i < c->max_buckets; i++) {
198 int size = 0;
199 u32 alg;
200 struct crush_bucket *b;
201
202 ceph_decode_32_safe(p, end, alg, bad);
203 if (alg == 0) {
204 c->buckets[i] = NULL;
205 continue;
206 }
207 dout("crush_decode bucket %d off %x %p to %p\n",
208 i, (int)(*p-start), *p, end);
209
210 switch (alg) {
211 case CRUSH_BUCKET_UNIFORM:
212 size = sizeof(struct crush_bucket_uniform);
213 break;
214 case CRUSH_BUCKET_LIST:
215 size = sizeof(struct crush_bucket_list);
216 break;
217 case CRUSH_BUCKET_TREE:
218 size = sizeof(struct crush_bucket_tree);
219 break;
220 case CRUSH_BUCKET_STRAW:
221 size = sizeof(struct crush_bucket_straw);
222 break;
958a2765
ID		 223 case CRUSH_BUCKET_STRAW2:
224 size = sizeof(struct crush_bucket_straw2);
225 break;
f24e9980 226 default:
30dc6381 227 err = -EINVAL;
f24e9980
SW		 228 goto bad;
229 }
230 BUG_ON(size == 0);
231 b = c->buckets[i] = kzalloc(size, GFP_NOFS);
232 if (b == NULL)
233 goto badmem;
234
235 ceph_decode_need(p, end, 4*sizeof(u32), bad);
c89136ea
SW		 236 b->id = ceph_decode_32(p);
237 b->type = ceph_decode_16(p);
fb690390
SW		 238 b->alg = ceph_decode_8(p);
239 b->hash = ceph_decode_8(p);
c89136ea
SW		 240 b->weight = ceph_decode_32(p);
241 b->size = ceph_decode_32(p);
f24e9980
SW		 242
243 dout("crush_decode bucket size %d off %x %p to %p\n",
244 b->size, (int)(*p-start), *p, end);
245
246 b->items = kcalloc(b->size, sizeof(__s32), GFP_NOFS);
247 if (b->items == NULL)
248 goto badmem;
249 b->perm = kcalloc(b->size, sizeof(u32), GFP_NOFS);
250 if (b->perm == NULL)
251 goto badmem;
252 b->perm_n = 0;
253
254 ceph_decode_need(p, end, b->size*sizeof(u32), bad);
255 for (j = 0; j < b->size; j++)
c89136ea 256 b->items[j] = ceph_decode_32(p);
f24e9980
SW		 257
258 switch (b->alg) {
259 case CRUSH_BUCKET_UNIFORM:
260 err = crush_decode_uniform_bucket(p, end,
261 (struct crush_bucket_uniform *)b);
262 if (err < 0)
263 goto bad;
264 break;
265 case CRUSH_BUCKET_LIST:
266 err = crush_decode_list_bucket(p, end,
267 (struct crush_bucket_list *)b);
268 if (err < 0)
269 goto bad;
270 break;
271 case CRUSH_BUCKET_TREE:
272 err = crush_decode_tree_bucket(p, end,
273 (struct crush_bucket_tree *)b);
274 if (err < 0)
275 goto bad;
276 break;
277 case CRUSH_BUCKET_STRAW:
278 err = crush_decode_straw_bucket(p, end,
279 (struct crush_bucket_straw *)b);
280 if (err < 0)
281 goto bad;
282 break;
958a2765
ID		 283 case CRUSH_BUCKET_STRAW2:
284 err = crush_decode_straw2_bucket(p, end,
285 (struct crush_bucket_straw2 *)b);
286 if (err < 0)
287 goto bad;
288 break;
f24e9980
SW		 289 }
290 }
291
292 /* rules */
293 dout("rule vec is %p\n", c->rules);
294 for (i = 0; i < c->max_rules; i++) {
295 u32 yes;
296 struct crush_rule *r;
297
298 ceph_decode_32_safe(p, end, yes, bad);
299 if (!yes) {
300 dout("crush_decode NO rule %d off %x %p to %p\n",
301 i, (int)(*p-start), *p, end);
302 c->rules[i] = NULL;
303 continue;
304 }
305
306 dout("crush_decode rule %d off %x %p to %p\n",
307 i, (int)(*p-start), *p, end);
308
309 /* len */
310 ceph_decode_32_safe(p, end, yes, bad);
311#if BITS_PER_LONG == 32
30dc6381 312 err = -EINVAL;
64486697
XW		 313 if (yes > (ULONG_MAX - sizeof(*r))
314 / sizeof(struct crush_rule_step))
f24e9980
SW		 315 goto bad;
316#endif
317 r = c->rules[i] = kmalloc(sizeof(*r) +
318 yes*sizeof(struct crush_rule_step),
319 GFP_NOFS);
320 if (r == NULL)
321 goto badmem;
322 dout(" rule %d is at %p\n", i, r);
323 r->len = yes;
324 ceph_decode_copy_safe(p, end, &r->mask, 4, bad); /* 4 u8's */
325 ceph_decode_need(p, end, r->len*3*sizeof(u32), bad);
326 for (j = 0; j < r->len; j++) {
c89136ea
SW		 327 r->steps[j].op = ceph_decode_32(p);
328 r->steps[j].arg1 = ceph_decode_32(p);
329 r->steps[j].arg2 = ceph_decode_32(p);
f24e9980
SW		 330 }
331 }
332
333 /* ignore trailing name maps. */
546f04ef
SW		 334 for (num_name_maps = 0; num_name_maps < 3; num_name_maps++) {
335 err = skip_name_map(p, end);
336 if (err < 0)
337 goto done;
338 }
339
340 /* tunables */
341 ceph_decode_need(p, end, 3*sizeof(u32), done);
342 c->choose_local_tries = ceph_decode_32(p);
343 c->choose_local_fallback_tries = ceph_decode_32(p);
344 c->choose_total_tries = ceph_decode_32(p);
b9b519b7 345 dout("crush decode tunable choose_local_tries = %d\n",
546f04ef 346 c->choose_local_tries);
b9b519b7 347 dout("crush decode tunable choose_local_fallback_tries = %d\n",
546f04ef 348 c->choose_local_fallback_tries);
b9b519b7 349 dout("crush decode tunable choose_total_tries = %d\n",
546f04ef 350 c->choose_total_tries);
f24e9980 351
1604f488
JS		 352 ceph_decode_need(p, end, sizeof(u32), done);
353 c->chooseleaf_descend_once = ceph_decode_32(p);
b9b519b7 354 dout("crush decode tunable chooseleaf_descend_once = %d\n",
1604f488
JS		 355 c->chooseleaf_descend_once);
356
f140662f
ID		 357 ceph_decode_need(p, end, sizeof(u8), done);
358 c->chooseleaf_vary_r = ceph_decode_8(p);
b9b519b7 359 dout("crush decode tunable chooseleaf_vary_r = %d\n",
f140662f
ID		 360 c->chooseleaf_vary_r);
361
b9b519b7
ID		 362 /* skip straw_calc_version, allowed_bucket_algs */
363 ceph_decode_need(p, end, sizeof(u8) + sizeof(u32), done);
364 *p += sizeof(u8) + sizeof(u32);
365
366 ceph_decode_need(p, end, sizeof(u8), done);
367 c->chooseleaf_stable = ceph_decode_8(p);
368 dout("crush decode tunable chooseleaf_stable = %d\n",
369 c->chooseleaf_stable);
370
546f04ef 371done:
f24e9980
SW		 372 dout("crush_decode success\n");
373 return c;
374
375badmem:
376 err = -ENOMEM;
377bad:
378 dout("crush_decode fail %d\n", err);
379 crush_destroy(c);
380 return ERR_PTR(err);
381}
382
f984cb76 383int ceph_pg_compare(const struct ceph_pg *lhs, const struct ceph_pg *rhs)
51042122 384{
f984cb76 385 if (lhs->pool < rhs->pool)
5b191d99 386 return -1;
f984cb76 387 if (lhs->pool > rhs->pool)
5b191d99 388 return 1;
f984cb76 389 if (lhs->seed < rhs->seed)
51042122 390 return -1;
f984cb76 391 if (lhs->seed > rhs->seed)
51042122 392 return 1;
f984cb76 393
51042122
SW		 394 return 0;
395}
396
f984cb76
ID		 397/*
398 * rbtree of pg_mapping for handling pg_temp (explicit mapping of pgid
399 * to a set of osds) and primary_temp (explicit primary setting)
400 */
991abb6e
SW		 401static int __insert_pg_mapping(struct ceph_pg_mapping *new,
402 struct rb_root *root)
f24e9980
SW		 403{
404 struct rb_node **p = &root->rb_node;
405 struct rb_node *parent = NULL;
406 struct ceph_pg_mapping *pg = NULL;
51042122 407 int c;
f24e9980 408
8adc8b3d 409 dout("__insert_pg_mapping %llx %p\n", *(u64 *)&new->pgid, new);
f24e9980
SW		 410 while (*p) {
411 parent = *p;
412 pg = rb_entry(parent, struct ceph_pg_mapping, node);
f984cb76 413 c = ceph_pg_compare(&new->pgid, &pg->pgid);
51042122 414 if (c < 0)
f24e9980 415 p = &(*p)->rb_left;
51042122 416 else if (c > 0)
f24e9980
SW		 417 p = &(*p)->rb_right;
418 else
991abb6e 419 return -EEXIST;
f24e9980
SW		 420 }
421
422 rb_link_node(&new->node, parent, p);
423 rb_insert_color(&new->node, root);
991abb6e 424 return 0;
f24e9980
SW		 425}
426
9794b146 427static struct ceph_pg_mapping *__lookup_pg_mapping(struct rb_root *root,
5b191d99 428 struct ceph_pg pgid)
9794b146
SW		 429{
430 struct rb_node *n = root->rb_node;
431 struct ceph_pg_mapping *pg;
432 int c;
433
434 while (n) {
435 pg = rb_entry(n, struct ceph_pg_mapping, node);
f984cb76 436 c = ceph_pg_compare(&pgid, &pg->pgid);
8adc8b3d 437 if (c < 0) {
9794b146 438 n = n->rb_left;
8adc8b3d 439 } else if (c > 0) {
9794b146 440 n = n->rb_right;
8adc8b3d 441 } else {
5b191d99
SW		 442 dout("__lookup_pg_mapping %lld.%x got %p\n",
443 pgid.pool, pgid.seed, pg);
9794b146 444 return pg;
8adc8b3d 445 }
9794b146
SW		 446 }
447 return NULL;
448}
449
5b191d99 450static int __remove_pg_mapping(struct rb_root *root, struct ceph_pg pgid)
8adc8b3d
SW		 451{
452 struct ceph_pg_mapping *pg = __lookup_pg_mapping(root, pgid);
453
454 if (pg) {
5b191d99
SW		 455 dout("__remove_pg_mapping %lld.%x %p\n", pgid.pool, pgid.seed,
456 pg);
8adc8b3d
SW		 457 rb_erase(&pg->node, root);
458 kfree(pg);
459 return 0;
460 }
5b191d99 461 dout("__remove_pg_mapping %lld.%x dne\n", pgid.pool, pgid.seed);
8adc8b3d
SW		 462 return -ENOENT;
463}
464
4fc51be8
SW		 465/*
466 * rbtree of pg pool info
467 */
468static int __insert_pg_pool(struct rb_root *root, struct ceph_pg_pool_info *new)
469{
470 struct rb_node **p = &root->rb_node;
471 struct rb_node *parent = NULL;
472 struct ceph_pg_pool_info *pi = NULL;
473
474 while (*p) {
475 parent = *p;
476 pi = rb_entry(parent, struct ceph_pg_pool_info, node);
477 if (new->id < pi->id)
478 p = &(*p)->rb_left;
479 else if (new->id > pi->id)
480 p = &(*p)->rb_right;
481 else
482 return -EEXIST;
483 }
484
485 rb_link_node(&new->node, parent, p);
486 rb_insert_color(&new->node, root);
487 return 0;
488}
489
4f6a7e5e 490static struct ceph_pg_pool_info *__lookup_pg_pool(struct rb_root *root, u64 id)
4fc51be8
SW		 491{
492 struct ceph_pg_pool_info *pi;
493 struct rb_node *n = root->rb_node;
494
495 while (n) {
496 pi = rb_entry(n, struct ceph_pg_pool_info, node);
497 if (id < pi->id)
498 n = n->rb_left;
499 else if (id > pi->id)
500 n = n->rb_right;
501 else
502 return pi;
503 }
504 return NULL;
505}
506
ce7f6a27
ID		 507struct ceph_pg_pool_info *ceph_pg_pool_by_id(struct ceph_osdmap *map, u64 id)
508{
509 return __lookup_pg_pool(&map->pg_pools, id);
510}
511
72afc71f
AE		 512const char *ceph_pg_pool_name_by_id(struct ceph_osdmap *map, u64 id)
513{
514 struct ceph_pg_pool_info *pi;
515
516 if (id == CEPH_NOPOOL)
517 return NULL;
518
519 if (WARN_ON_ONCE(id > (u64) INT_MAX))
520 return NULL;
521
522 pi = __lookup_pg_pool(&map->pg_pools, (int) id);
523
524 return pi ? pi->name : NULL;
525}
526EXPORT_SYMBOL(ceph_pg_pool_name_by_id);
527
7669a2c9
YS		 528int ceph_pg_poolid_by_name(struct ceph_osdmap *map, const char *name)
529{
530 struct rb_node *rbp;
531
532 for (rbp = rb_first(&map->pg_pools); rbp; rbp = rb_next(rbp)) {
533 struct ceph_pg_pool_info *pi =
534 rb_entry(rbp, struct ceph_pg_pool_info, node);
535 if (pi->name && strcmp(pi->name, name) == 0)
536 return pi->id;
537 }
538 return -ENOENT;
539}
3d14c5d2 540EXPORT_SYMBOL(ceph_pg_poolid_by_name);
7669a2c9 541
2844a76a
SW		 542static void __remove_pg_pool(struct rb_root *root, struct ceph_pg_pool_info *pi)
543{
544 rb_erase(&pi->node, root);
545 kfree(pi->name);
546 kfree(pi);
547}
548
0f70c7ee 549static int decode_pool(void **p, void *end, struct ceph_pg_pool_info *pi)
efd7576b 550{
4f6a7e5e
SW		 551 u8 ev, cv;
552 unsigned len, num;
553 void *pool_end;
554
555 ceph_decode_need(p, end, 2 + 4, bad);
556 ev = ceph_decode_8(p); /* encoding version */
557 cv = ceph_decode_8(p); /* compat version */
558 if (ev < 5) {
b9a67899 559 pr_warn("got v %d < 5 cv %d of ceph_pg_pool\n", ev, cv);
4f6a7e5e
SW		 560 return -EINVAL;
561 }
17a13e40 562 if (cv > 9) {
b9a67899 563 pr_warn("got v %d cv %d > 9 of ceph_pg_pool\n", ev, cv);
4f6a7e5e
SW		 564 return -EINVAL;
565 }
566 len = ceph_decode_32(p);
567 ceph_decode_need(p, end, len, bad);
568 pool_end = *p + len;
73a7e693 569
4f6a7e5e
SW		 570 pi->type = ceph_decode_8(p);
571 pi->size = ceph_decode_8(p);
572 pi->crush_ruleset = ceph_decode_8(p);
573 pi->object_hash = ceph_decode_8(p);
73a7e693 574
4f6a7e5e
SW		 575 pi->pg_num = ceph_decode_32(p);
576 pi->pgp_num = ceph_decode_32(p);
577
578 *p += 4 + 4; /* skip lpg* */
579 *p += 4; /* skip last_change */
580 *p += 8 + 4; /* skip snap_seq, snap_epoch */
581
582 /* skip snaps */
583 num = ceph_decode_32(p);
584 while (num--) {
585 *p += 8; /* snapid key */
586 *p += 1 + 1; /* versions */
587 len = ceph_decode_32(p);
588 *p += len;
73a7e693
SW		 589 }
590
17a13e40 591 /* skip removed_snaps */
4f6a7e5e
SW		 592 num = ceph_decode_32(p);
593 *p += num * (8 + 8);
594
595 *p += 8; /* skip auid */
596 pi->flags = ceph_decode_64(p);
17a13e40
ID		 597 *p += 4; /* skip crash_replay_interval */
598
599 if (ev >= 7)
04812acf
ID		 600 pi->min_size = ceph_decode_8(p);
601 else
602 pi->min_size = pi->size - pi->size / 2;
17a13e40
ID		 603
604 if (ev >= 8)
605 *p += 8 + 8; /* skip quota_max_* */
606
607 if (ev >= 9) {
608 /* skip tiers */
609 num = ceph_decode_32(p);
610 *p += num * 8;
611
612 *p += 8; /* skip tier_of */
613 *p += 1; /* skip cache_mode */
614
615 pi->read_tier = ceph_decode_64(p);
616 pi->write_tier = ceph_decode_64(p);
617 } else {
618 pi->read_tier = -1;
619 pi->write_tier = -1;
620 }
4f6a7e5e 621
04812acf
ID		 622 if (ev >= 10) {
623 /* skip properties */
624 num = ceph_decode_32(p);
625 while (num--) {
626 len = ceph_decode_32(p);
627 *p += len; /* key */
628 len = ceph_decode_32(p);
629 *p += len; /* val */
630 }
631 }
632
633 if (ev >= 11) {
634 /* skip hit_set_params */
635 *p += 1 + 1; /* versions */
636 len = ceph_decode_32(p);
637 *p += len;
638
639 *p += 4; /* skip hit_set_period */
640 *p += 4; /* skip hit_set_count */
641 }
642
643 if (ev >= 12)
644 *p += 4; /* skip stripe_width */
645
646 if (ev >= 13) {
647 *p += 8; /* skip target_max_bytes */
648 *p += 8; /* skip target_max_objects */
649 *p += 4; /* skip cache_target_dirty_ratio_micro */
650 *p += 4; /* skip cache_target_full_ratio_micro */
651 *p += 4; /* skip cache_min_flush_age */
652 *p += 4; /* skip cache_min_evict_age */
653 }
654
655 if (ev >= 14) {
656 /* skip erasure_code_profile */
657 len = ceph_decode_32(p);
658 *p += len;
659 }
660
661 if (ev >= 15)
662 pi->last_force_request_resend = ceph_decode_32(p);
663 else
664 pi->last_force_request_resend = 0;
665
4f6a7e5e
SW		 666 /* ignore the rest */
667
668 *p = pool_end;
669 calc_pg_masks(pi);
73a7e693
SW		 670 return 0;
671
672bad:
673 return -EINVAL;
efd7576b
SW		 674}
675
0f70c7ee 676static int decode_pool_names(void **p, void *end, struct ceph_osdmap *map)
2844a76a
SW		 677{
678 struct ceph_pg_pool_info *pi;
4f6a7e5e
SW		 679 u32 num, len;
680 u64 pool;
2844a76a
SW		 681
682 ceph_decode_32_safe(p, end, num, bad);
683 dout(" %d pool names\n", num);
684 while (num--) {
4f6a7e5e 685 ceph_decode_64_safe(p, end, pool, bad);
2844a76a 686 ceph_decode_32_safe(p, end, len, bad);
4f6a7e5e 687 dout(" pool %llu len %d\n", pool, len);
ad3b904c 688 ceph_decode_need(p, end, len, bad);
2844a76a
SW		 689 pi = __lookup_pg_pool(&map->pg_pools, pool);
690 if (pi) {
ad3b904c
XW		 691 char *name = kstrndup(*p, len, GFP_NOFS);
692
693 if (!name)
694 return -ENOMEM;
2844a76a 695 kfree(pi->name);
ad3b904c
XW		 696 pi->name = name;
697 dout(" name is %s\n", pi->name);
2844a76a
SW		 698 }
699 *p += len;
700 }
701 return 0;
702
703bad:
704 return -EINVAL;
705}
706
707/*
708 * osd map
709 */
e5253a7b
ID		 710struct ceph_osdmap *ceph_osdmap_alloc(void)
711{
712 struct ceph_osdmap *map;
713
714 map = kzalloc(sizeof(*map), GFP_NOIO);
715 if (!map)
716 return NULL;
717
718 map->pg_pools = RB_ROOT;
719 map->pool_max = -1;
720 map->pg_temp = RB_ROOT;
721 map->primary_temp = RB_ROOT;
722 mutex_init(&map->crush_scratch_mutex);
723
724 return map;
725}
726
2844a76a
SW		 727void ceph_osdmap_destroy(struct ceph_osdmap *map)
728{
729 dout("osdmap_destroy %p\n", map);
730 if (map->crush)
731 crush_destroy(map->crush);
732 while (!RB_EMPTY_ROOT(&map->pg_temp)) {
733 struct ceph_pg_mapping *pg =
734 rb_entry(rb_first(&map->pg_temp),
735 struct ceph_pg_mapping, node);
736 rb_erase(&pg->node, &map->pg_temp);
737 kfree(pg);
738 }
9686f94c
ID		 739 while (!RB_EMPTY_ROOT(&map->primary_temp)) {
740 struct ceph_pg_mapping *pg =
741 rb_entry(rb_first(&map->primary_temp),
742 struct ceph_pg_mapping, node);
743 rb_erase(&pg->node, &map->primary_temp);
744 kfree(pg);
745 }
2844a76a
SW		 746 while (!RB_EMPTY_ROOT(&map->pg_pools)) {
747 struct ceph_pg_pool_info *pi =
748 rb_entry(rb_first(&map->pg_pools),
749 struct ceph_pg_pool_info, node);
750 __remove_pg_pool(&map->pg_pools, pi);
751 }
752 kfree(map->osd_state);
753 kfree(map->osd_weight);
754 kfree(map->osd_addr);
2cfa34f2 755 kfree(map->osd_primary_affinity);
2844a76a
SW		 756 kfree(map);
757}
758
759/*
4d60351f
ID		 760 * Adjust max_osd value, (re)allocate arrays.
761 *
762 * The new elements are properly initialized.
2844a76a
SW		 763 */
764static int osdmap_set_max_osd(struct ceph_osdmap *map, int max)
765{
766 u8 *state;
2844a76a 767 u32 *weight;
4d60351f
ID		 768 struct ceph_entity_addr *addr;
769 int i;
2844a76a 770
4d60351f 771 state = krealloc(map->osd_state, max*sizeof(*state), GFP_NOFS);
589506f1
LR		 772 if (!state)
773 return -ENOMEM;
774 map->osd_state = state;
775
4d60351f 776 weight = krealloc(map->osd_weight, max*sizeof(*weight), GFP_NOFS);
589506f1
LR		 777 if (!weight)
778 return -ENOMEM;
779 map->osd_weight = weight;
4d60351f 780
589506f1
LR		 781 addr = krealloc(map->osd_addr, max*sizeof(*addr), GFP_NOFS);
782 if (!addr)
2844a76a 783 return -ENOMEM;
589506f1 784 map->osd_addr = addr;
2844a76a 785
4d60351f 786 for (i = map->max_osd; i < max; i++) {
589506f1
LR		 787 map->osd_state[i] = 0;
788 map->osd_weight[i] = CEPH_OSD_OUT;
789 memset(map->osd_addr + i, 0, sizeof(*map->osd_addr));
2844a76a
SW		 790 }
791
2cfa34f2
ID		 792 if (map->osd_primary_affinity) {
793 u32 *affinity;
794
795 affinity = krealloc(map->osd_primary_affinity,
796 max*sizeof(*affinity), GFP_NOFS);
797 if (!affinity)
798 return -ENOMEM;
589506f1 799 map->osd_primary_affinity = affinity;
2cfa34f2
ID		 800
801 for (i = map->max_osd; i < max; i++)
589506f1
LR		 802 map->osd_primary_affinity[i] =
803 CEPH_OSD_DEFAULT_PRIMARY_AFFINITY;
2cfa34f2
ID		 804 }
805
2844a76a 806 map->max_osd = max;
4d60351f 807
2844a76a
SW		 808 return 0;
809}
810
ec7af972
ID		 811#define OSDMAP_WRAPPER_COMPAT_VER 7
812#define OSDMAP_CLIENT_DATA_COMPAT_VER 1
813
814/*
815 * Return 0 or error. On success, *v is set to 0 for old (v6) osdmaps,
816 * to struct_v of the client_data section for new (v7 and above)
817 * osdmaps.
818 */
819static int get_osdmap_client_data_v(void **p, void *end,
820 const char *prefix, u8 *v)
821{
822 u8 struct_v;
823
824 ceph_decode_8_safe(p, end, struct_v, e_inval);
825 if (struct_v >= 7) {
826 u8 struct_compat;
827
828 ceph_decode_8_safe(p, end, struct_compat, e_inval);
829 if (struct_compat > OSDMAP_WRAPPER_COMPAT_VER) {
b9a67899
JP		 830 pr_warn("got v %d cv %d > %d of %s ceph_osdmap\n",
831 struct_v, struct_compat,
832 OSDMAP_WRAPPER_COMPAT_VER, prefix);
ec7af972
ID		 833 return -EINVAL;
834 }
835 *p += 4; /* ignore wrapper struct_len */
836
837 ceph_decode_8_safe(p, end, struct_v, e_inval);
838 ceph_decode_8_safe(p, end, struct_compat, e_inval);
839 if (struct_compat > OSDMAP_CLIENT_DATA_COMPAT_VER) {
b9a67899
JP		 840 pr_warn("got v %d cv %d > %d of %s ceph_osdmap client data\n",
841 struct_v, struct_compat,
842 OSDMAP_CLIENT_DATA_COMPAT_VER, prefix);
ec7af972
ID		 843 return -EINVAL;
844 }
845 *p += 4; /* ignore client data struct_len */
846 } else {
847 u16 version;
848
849 *p -= 1;
850 ceph_decode_16_safe(p, end, version, e_inval);
851 if (version < 6) {
b9a67899
JP		 852 pr_warn("got v %d < 6 of %s ceph_osdmap\n",
853 version, prefix);
ec7af972
ID		 854 return -EINVAL;
855 }
856
857 /* old osdmap enconding */
858 struct_v = 0;
859 }
860
861 *v = struct_v;
862 return 0;
863
864e_inval:
865 return -EINVAL;
866}
867
433fbdd3
ID		 868static int __decode_pools(void **p, void *end, struct ceph_osdmap *map,
869 bool incremental)
870{
871 u32 n;
872
873 ceph_decode_32_safe(p, end, n, e_inval);
874 while (n--) {
875 struct ceph_pg_pool_info *pi;
876 u64 pool;
877 int ret;
878
879 ceph_decode_64_safe(p, end, pool, e_inval);
880
881 pi = __lookup_pg_pool(&map->pg_pools, pool);
882 if (!incremental || !pi) {
883 pi = kzalloc(sizeof(*pi), GFP_NOFS);
884 if (!pi)
885 return -ENOMEM;
886
887 pi->id = pool;
888
889 ret = __insert_pg_pool(&map->pg_pools, pi);
890 if (ret) {
891 kfree(pi);
892 return ret;
893 }
894 }
895
896 ret = decode_pool(p, end, pi);
897 if (ret)
898 return ret;
899 }
900
901 return 0;
902
903e_inval:
904 return -EINVAL;
905}
906
907static int decode_pools(void **p, void *end, struct ceph_osdmap *map)
908{
909 return __decode_pools(p, end, map, false);
910}
911
912static int decode_new_pools(void **p, void *end, struct ceph_osdmap *map)
913{
914 return __decode_pools(p, end, map, true);
915}
916
10db634e
ID		 917static int __decode_pg_temp(void **p, void *end, struct ceph_osdmap *map,
918 bool incremental)
919{
920 u32 n;
921
922 ceph_decode_32_safe(p, end, n, e_inval);
923 while (n--) {
924 struct ceph_pg pgid;
925 u32 len, i;
926 int ret;
927
928 ret = ceph_decode_pgid(p, end, &pgid);
929 if (ret)
930 return ret;
931
932 ceph_decode_32_safe(p, end, len, e_inval);
933
934 ret = __remove_pg_mapping(&map->pg_temp, pgid);
935 BUG_ON(!incremental && ret != -ENOENT);
936
937 if (!incremental || len > 0) {
938 struct ceph_pg_mapping *pg;
939
940 ceph_decode_need(p, end, len*sizeof(u32), e_inval);
941
942 if (len > (UINT_MAX - sizeof(*pg)) / sizeof(u32))
943 return -EINVAL;
944
945 pg = kzalloc(sizeof(*pg) + len*sizeof(u32), GFP_NOFS);
946 if (!pg)
947 return -ENOMEM;
948
949 pg->pgid = pgid;
35a935d7 950 pg->pg_temp.len = len;
10db634e 951 for (i = 0; i < len; i++)
35a935d7 952 pg->pg_temp.osds[i] = ceph_decode_32(p);
10db634e
ID		 953
954 ret = __insert_pg_mapping(pg, &map->pg_temp);
955 if (ret) {
956 kfree(pg);
957 return ret;
958 }
959 }
960 }
961
962 return 0;
963
964e_inval:
965 return -EINVAL;
966}
967
968static int decode_pg_temp(void **p, void *end, struct ceph_osdmap *map)
969{
970 return __decode_pg_temp(p, end, map, false);
971}
972
973static int decode_new_pg_temp(void **p, void *end, struct ceph_osdmap *map)
974{
975 return __decode_pg_temp(p, end, map, true);
976}
977
d286de79
ID		 978static int __decode_primary_temp(void **p, void *end, struct ceph_osdmap *map,
979 bool incremental)
980{
981 u32 n;
982
983 ceph_decode_32_safe(p, end, n, e_inval);
984 while (n--) {
985 struct ceph_pg pgid;
986 u32 osd;
987 int ret;
988
989 ret = ceph_decode_pgid(p, end, &pgid);
990 if (ret)
991 return ret;
992
993 ceph_decode_32_safe(p, end, osd, e_inval);
994
995 ret = __remove_pg_mapping(&map->primary_temp, pgid);
996 BUG_ON(!incremental && ret != -ENOENT);
997
998 if (!incremental || osd != (u32)-1) {
999 struct ceph_pg_mapping *pg;
1000
1001 pg = kzalloc(sizeof(*pg), GFP_NOFS);
1002 if (!pg)
1003 return -ENOMEM;
1004
1005 pg->pgid = pgid;
1006 pg->primary_temp.osd = osd;
1007
1008 ret = __insert_pg_mapping(pg, &map->primary_temp);
1009 if (ret) {
1010 kfree(pg);
1011 return ret;
1012 }
1013 }
1014 }
1015
1016 return 0;
1017
1018e_inval:
1019 return -EINVAL;
1020}
1021
1022static int decode_primary_temp(void **p, void *end, struct ceph_osdmap *map)
1023{
1024 return __decode_primary_temp(p, end, map, false);
1025}
1026
1027static int decode_new_primary_temp(void **p, void *end,
1028 struct ceph_osdmap *map)
1029{
1030 return __decode_primary_temp(p, end, map, true);
1031}
1032
2cfa34f2
ID		 1033u32 ceph_get_primary_affinity(struct ceph_osdmap *map, int osd)
1034{
1035 BUG_ON(osd >= map->max_osd);
1036
1037 if (!map->osd_primary_affinity)
1038 return CEPH_OSD_DEFAULT_PRIMARY_AFFINITY;
1039
1040 return map->osd_primary_affinity[osd];
1041}
1042
1043static int set_primary_affinity(struct ceph_osdmap *map, int osd, u32 aff)
1044{
1045 BUG_ON(osd >= map->max_osd);
1046
1047 if (!map->osd_primary_affinity) {
1048 int i;
1049
1050 map->osd_primary_affinity = kmalloc(map->max_osd*sizeof(u32),
1051 GFP_NOFS);
1052 if (!map->osd_primary_affinity)
1053 return -ENOMEM;
1054
1055 for (i = 0; i < map->max_osd; i++)
1056 map->osd_primary_affinity[i] =
1057 CEPH_OSD_DEFAULT_PRIMARY_AFFINITY;
1058 }
1059
1060 map->osd_primary_affinity[osd] = aff;
1061
1062 return 0;
1063}
1064
63a6993f
ID		 1065static int decode_primary_affinity(void **p, void *end,
1066 struct ceph_osdmap *map)
1067{
1068 u32 len, i;
1069
1070 ceph_decode_32_safe(p, end, len, e_inval);
1071 if (len == 0) {
1072 kfree(map->osd_primary_affinity);
1073 map->osd_primary_affinity = NULL;
1074 return 0;
1075 }
1076 if (len != map->max_osd)
1077 goto e_inval;
1078
1079 ceph_decode_need(p, end, map->max_osd*sizeof(u32), e_inval);
1080
1081 for (i = 0; i < map->max_osd; i++) {
1082 int ret;
1083
1084 ret = set_primary_affinity(map, i, ceph_decode_32(p));
1085 if (ret)
1086 return ret;
1087 }
1088
1089 return 0;
1090
1091e_inval:
1092 return -EINVAL;
1093}
1094
1095static int decode_new_primary_affinity(void **p, void *end,
1096 struct ceph_osdmap *map)
1097{
1098 u32 n;
1099
1100 ceph_decode_32_safe(p, end, n, e_inval);
1101 while (n--) {
1102 u32 osd, aff;
1103 int ret;
1104
1105 ceph_decode_32_safe(p, end, osd, e_inval);
1106 ceph_decode_32_safe(p, end, aff, e_inval);
1107
1108 ret = set_primary_affinity(map, osd, aff);
1109 if (ret)
1110 return ret;
f31da0f3
ID		 1111
1112 pr_info("osd%d primary-affinity 0x%x\n", osd, aff);
63a6993f
ID		 1113 }
1114
1115 return 0;
1116
1117e_inval:
1118 return -EINVAL;
1119}
1120
f24e9980
SW		 1121/*
1122 * decode a full map.
1123 */
a2505d63 1124static int osdmap_decode(void **p, void *end, struct ceph_osdmap *map)
f24e9980 1125{
ec7af972 1126 u8 struct_v;
38a8d560 1127 u32 epoch = 0;
f24e9980 1128 void *start = *p;
3977058c
ID		 1129 u32 max;
1130 u32 len, i;
597b52f6 1131 int err;
f24e9980 1132
a2505d63 1133 dout("%s %p to %p len %d\n", __func__, *p, end, (int)(end - *p));
f24e9980 1134
ec7af972
ID		 1135 err = get_osdmap_client_data_v(p, end, "full", &struct_v);
1136 if (err)
1137 goto bad;
f24e9980 1138
53bbaba9
ID		 1139 /* fsid, epoch, created, modified */
1140 ceph_decode_need(p, end, sizeof(map->fsid) + sizeof(u32) +
1141 sizeof(map->created) + sizeof(map->modified), e_inval);
f24e9980 1142 ceph_decode_copy(p, &map->fsid, sizeof(map->fsid));
38a8d560 1143 epoch = map->epoch = ceph_decode_32(p);
f24e9980
SW		 1144 ceph_decode_copy(p, &map->created, sizeof(map->created));
1145 ceph_decode_copy(p, &map->modified, sizeof(map->modified));
1146
433fbdd3
ID		 1147 /* pools */
1148 err = decode_pools(p, end, map);
1149 if (err)
1150 goto bad;
2844a76a 1151
0f70c7ee
ID		 1152 /* pool_name */
1153 err = decode_pool_names(p, end, map);
597b52f6 1154 if (err)
4f6a7e5e 1155 goto bad;
2844a76a 1156
597b52f6 1157 ceph_decode_32_safe(p, end, map->pool_max, e_inval);
f24e9980 1158
597b52f6 1159 ceph_decode_32_safe(p, end, map->flags, e_inval);
f24e9980 1160
3977058c
ID		 1161 /* max_osd */
1162 ceph_decode_32_safe(p, end, max, e_inval);
f24e9980
SW		 1163
1164 /* (re)alloc osd arrays */
1165 err = osdmap_set_max_osd(map, max);
597b52f6 1166 if (err)
f24e9980 1167 goto bad;
f24e9980 1168
2d88b2e0 1169 /* osd_state, osd_weight, osd_addrs->client_addr */
f24e9980
SW		 1170 ceph_decode_need(p, end, 3*sizeof(u32) +
1171 map->max_osd*(1 + sizeof(*map->osd_weight) +
597b52f6
ID		 1172 sizeof(*map->osd_addr)), e_inval);
1173
2d88b2e0
ID		 1174 if (ceph_decode_32(p) != map->max_osd)
1175 goto e_inval;
1176
f24e9980
SW		 1177 ceph_decode_copy(p, map->osd_state, map->max_osd);
1178
2d88b2e0
ID		 1179 if (ceph_decode_32(p) != map->max_osd)
1180 goto e_inval;
1181
f24e9980 1182 for (i = 0; i < map->max_osd; i++)
c89136ea 1183 map->osd_weight[i] = ceph_decode_32(p);
f24e9980 1184
2d88b2e0
ID		 1185 if (ceph_decode_32(p) != map->max_osd)
1186 goto e_inval;
1187
f24e9980 1188 ceph_decode_copy(p, map->osd_addr, map->max_osd*sizeof(*map->osd_addr));
63f2d211
SW		 1189 for (i = 0; i < map->max_osd; i++)
1190 ceph_decode_addr(&map->osd_addr[i]);
f24e9980
SW		 1191
1192 /* pg_temp */
10db634e
ID		 1193 err = decode_pg_temp(p, end, map);
1194 if (err)
1195 goto bad;
f24e9980 1196
d286de79
ID		 1197 /* primary_temp */
1198 if (struct_v >= 1) {
1199 err = decode_primary_temp(p, end, map);
1200 if (err)
1201 goto bad;
1202 }
1203
63a6993f
ID		 1204 /* primary_affinity */
1205 if (struct_v >= 2) {
1206 err = decode_primary_affinity(p, end, map);
1207 if (err)
1208 goto bad;
1209 } else {
1210 /* XXX can this happen? */
1211 kfree(map->osd_primary_affinity);
1212 map->osd_primary_affinity = NULL;
1213 }
1214
f24e9980 1215 /* crush */
597b52f6 1216 ceph_decode_32_safe(p, end, len, e_inval);
9902e682 1217 map->crush = crush_decode(*p, min(*p + len, end));
f24e9980
SW		 1218 if (IS_ERR(map->crush)) {
1219 err = PTR_ERR(map->crush);
1220 map->crush = NULL;
1221 goto bad;
1222 }
9902e682 1223 *p += len;
f24e9980 1224
38a8d560 1225 /* ignore the rest */
f24e9980
SW		 1226 *p = end;
1227
38a8d560 1228 dout("full osdmap epoch %d max_osd %d\n", map->epoch, map->max_osd);
a2505d63 1229 return 0;
f24e9980 1230
597b52f6
ID		 1231e_inval:
1232 err = -EINVAL;
f24e9980 1233bad:
38a8d560
ID		 1234 pr_err("corrupt full osdmap (%d) epoch %d off %d (%p of %p-%p)\n",
1235 err, epoch, (int)(*p - start), *p, start, end);
1236 print_hex_dump(KERN_DEBUG, "osdmap: ",
1237 DUMP_PREFIX_OFFSET, 16, 1,
1238 start, end - start, true);
a2505d63
ID		 1239 return err;
1240}
1241
1242/*
1243 * Allocate and decode a full map.
1244 */
1245struct ceph_osdmap *ceph_osdmap_decode(void **p, void *end)
1246{
1247 struct ceph_osdmap *map;
1248 int ret;
1249
e5253a7b 1250 map = ceph_osdmap_alloc();
a2505d63
ID		 1251 if (!map)
1252 return ERR_PTR(-ENOMEM);
1253
a2505d63
ID		 1254 ret = osdmap_decode(p, end, map);
1255 if (ret) {
1256 ceph_osdmap_destroy(map);
1257 return ERR_PTR(ret);
1258 }
1259
1260 return map;
f24e9980
SW		 1261}
1262
930c5328
ID		 1263/*
1264 * Encoding order is (new_up_client, new_state, new_weight). Need to
1265 * apply in the (new_weight, new_state, new_up_client) order, because
1266 * an incremental map may look like e.g.
1267 *
1268 * new_up_client: { osd=6, addr=... } # set osd_state and addr
1269 * new_state: { osd=6, xorstate=EXISTS } # clear osd_state
1270 */
1271static int decode_new_up_state_weight(void **p, void *end,
1272 struct ceph_osdmap *map)
1273{
1274 void *new_up_client;
1275 void *new_state;
1276 void *new_weight_end;
1277 u32 len;
1278
1279 new_up_client = *p;
1280 ceph_decode_32_safe(p, end, len, e_inval);
1281 len *= sizeof(u32) + sizeof(struct ceph_entity_addr);
1282 ceph_decode_need(p, end, len, e_inval);
1283 *p += len;
1284
1285 new_state = *p;
1286 ceph_decode_32_safe(p, end, len, e_inval);
1287 len *= sizeof(u32) + sizeof(u8);
1288 ceph_decode_need(p, end, len, e_inval);
1289 *p += len;
1290
1291 /* new_weight */
1292 ceph_decode_32_safe(p, end, len, e_inval);
1293 while (len--) {
1294 s32 osd;
1295 u32 w;
1296
1297 ceph_decode_need(p, end, 2*sizeof(u32), e_inval);
1298 osd = ceph_decode_32(p);
1299 w = ceph_decode_32(p);
1300 BUG_ON(osd >= map->max_osd);
1301 pr_info("osd%d weight 0x%x %s\n", osd, w,
1302 w == CEPH_OSD_IN ? "(in)" :
1303 (w == CEPH_OSD_OUT ? "(out)" : ""));
1304 map->osd_weight[osd] = w;
1305
1306 /*
1307 * If we are marking in, set the EXISTS, and clear the
1308 * AUTOOUT and NEW bits.
1309 */
1310 if (w) {
1311 map->osd_state[osd] |= CEPH_OSD_EXISTS;
1312 map->osd_state[osd] &= ~(CEPH_OSD_AUTOOUT |
1313 CEPH_OSD_NEW);
1314 }
1315 }
1316 new_weight_end = *p;
1317
1318 /* new_state (up/down) */
1319 *p = new_state;
1320 len = ceph_decode_32(p);
1321 while (len--) {
1322 s32 osd;
1323 u8 xorstate;
1324 int ret;
1325
1326 osd = ceph_decode_32(p);
1327 xorstate = ceph_decode_8(p);
1328 if (xorstate == 0)
1329 xorstate = CEPH_OSD_UP;
1330 BUG_ON(osd >= map->max_osd);
1331 if ((map->osd_state[osd] & CEPH_OSD_UP) &&
1332 (xorstate & CEPH_OSD_UP))
1333 pr_info("osd%d down\n", osd);
1334 if ((map->osd_state[osd] & CEPH_OSD_EXISTS) &&
1335 (xorstate & CEPH_OSD_EXISTS)) {
1336 pr_info("osd%d does not exist\n", osd);
930c5328
ID		 1337 ret = set_primary_affinity(map, osd,
1338 CEPH_OSD_DEFAULT_PRIMARY_AFFINITY);
1339 if (ret)
1340 return ret;
1341 memset(map->osd_addr + osd, 0, sizeof(*map->osd_addr));
1342 map->osd_state[osd] = 0;
1343 } else {
1344 map->osd_state[osd] ^= xorstate;
1345 }
1346 }
1347
1348 /* new_up_client */
1349 *p = new_up_client;
1350 len = ceph_decode_32(p);
1351 while (len--) {
1352 s32 osd;
1353 struct ceph_entity_addr addr;
1354
1355 osd = ceph_decode_32(p);
1356 ceph_decode_copy(p, &addr, sizeof(addr));
1357 ceph_decode_addr(&addr);
1358 BUG_ON(osd >= map->max_osd);
1359 pr_info("osd%d up\n", osd);
1360 map->osd_state[osd] |= CEPH_OSD_EXISTS | CEPH_OSD_UP;
1361 map->osd_addr[osd] = addr;
1362 }
1363
1364 *p = new_weight_end;
1365 return 0;
1366
1367e_inval:
1368 return -EINVAL;
1369}
1370
f24e9980
SW		 1371/*
1372 * decode and apply an incremental map update.
1373 */
1374struct ceph_osdmap *osdmap_apply_incremental(void **p, void *end,
0c0a8de1 1375 struct ceph_osdmap *map)
f24e9980 1376{
f24e9980
SW		 1377 struct crush_map *newcrush = NULL;
1378 struct ceph_fsid fsid;
1379 u32 epoch = 0;
1380 struct ceph_timespec modified;
4f6a7e5e
SW		 1381 s32 len;
1382 u64 pool;
1383 __s64 new_pool_max;
1384 __s32 new_flags, max;
f24e9980 1385 void *start = *p;
86f1742b 1386 int err;
ec7af972 1387 u8 struct_v;
f24e9980 1388
38a8d560
ID		 1389 dout("%s %p to %p len %d\n", __func__, *p, end, (int)(end - *p));
1390
ec7af972
ID		 1391 err = get_osdmap_client_data_v(p, end, "inc", &struct_v);
1392 if (err)
1393 goto bad;
f24e9980 1394
53bbaba9
ID		 1395 /* fsid, epoch, modified, new_pool_max, new_flags */
1396 ceph_decode_need(p, end, sizeof(fsid) + sizeof(u32) + sizeof(modified) +
1397 sizeof(u64) + sizeof(u32), e_inval);
f24e9980 1398 ceph_decode_copy(p, &fsid, sizeof(fsid));
c89136ea 1399 epoch = ceph_decode_32(p);
f24e9980
SW		 1400 BUG_ON(epoch != map->epoch+1);
1401 ceph_decode_copy(p, &modified, sizeof(modified));
4f6a7e5e 1402 new_pool_max = ceph_decode_64(p);
c89136ea 1403 new_flags = ceph_decode_32(p);
f24e9980
SW		 1404
1405 /* full map? */
86f1742b 1406 ceph_decode_32_safe(p, end, len, e_inval);
f24e9980
SW		 1407 if (len > 0) {
1408 dout("apply_incremental full map len %d, %p to %p\n",
1409 len, *p, end);
a2505d63 1410 return ceph_osdmap_decode(p, min(*p+len, end));
f24e9980
SW		 1411 }
1412
1413 /* new crush? */
86f1742b 1414 ceph_decode_32_safe(p, end, len, e_inval);
f24e9980 1415 if (len > 0) {
f24e9980 1416 newcrush = crush_decode(*p, min(*p+len, end));
86f1742b
ID		 1417 if (IS_ERR(newcrush)) {
1418 err = PTR_ERR(newcrush);
1419 newcrush = NULL;
1420 goto bad;
1421 }
cebc5be6 1422 *p += len;
f24e9980
SW		 1423 }
1424
1425 /* new flags? */
1426 if (new_flags >= 0)
1427 map->flags = new_flags;
4fc51be8
SW		 1428 if (new_pool_max >= 0)
1429 map->pool_max = new_pool_max;
f24e9980 1430
f24e9980 1431 /* new max? */
53bbaba9 1432 ceph_decode_32_safe(p, end, max, e_inval);
f24e9980
SW		 1433 if (max >= 0) {
1434 err = osdmap_set_max_osd(map, max);
86f1742b 1435 if (err)
f24e9980
SW		 1436 goto bad;
1437 }
1438
1439 map->epoch++;
31456665 1440 map->modified = modified;
f24e9980
SW		 1441 if (newcrush) {
1442 if (map->crush)
1443 crush_destroy(map->crush);
1444 map->crush = newcrush;
1445 newcrush = NULL;
1446 }
1447
433fbdd3
ID		 1448 /* new_pools */
1449 err = decode_new_pools(p, end, map);
1450 if (err)
1451 goto bad;
9464d008 1452
0f70c7ee
ID		 1453 /* new_pool_names */
1454 err = decode_pool_names(p, end, map);
9464d008
ID		 1455 if (err)
1456 goto bad;
f24e9980 1457
4fc51be8 1458 /* old_pool */
86f1742b 1459 ceph_decode_32_safe(p, end, len, e_inval);
4fc51be8
SW		 1460 while (len--) {
1461 struct ceph_pg_pool_info *pi;
1462
86f1742b 1463 ceph_decode_64_safe(p, end, pool, e_inval);
4fc51be8 1464 pi = __lookup_pg_pool(&map->pg_pools, pool);
2844a76a
SW		 1465 if (pi)
1466 __remove_pg_pool(&map->pg_pools, pi);
4fc51be8 1467 }
f24e9980 1468
930c5328
ID		 1469 /* new_up_client, new_state, new_weight */
1470 err = decode_new_up_state_weight(p, end, map);
1471 if (err)
1472 goto bad;
f24e9980
SW		 1473
1474 /* new_pg_temp */
10db634e
ID		 1475 err = decode_new_pg_temp(p, end, map);
1476 if (err)
1477 goto bad;
f24e9980 1478
d286de79
ID		 1479 /* new_primary_temp */
1480 if (struct_v >= 1) {
1481 err = decode_new_primary_temp(p, end, map);
1482 if (err)
1483 goto bad;
1484 }
1485
63a6993f
ID		 1486 /* new_primary_affinity */
1487 if (struct_v >= 2) {
1488 err = decode_new_primary_affinity(p, end, map);
1489 if (err)
1490 goto bad;
1491 }
1492
f24e9980
SW		 1493 /* ignore the rest */
1494 *p = end;
38a8d560
ID		 1495
1496 dout("inc osdmap epoch %d max_osd %d\n", map->epoch, map->max_osd);
f24e9980
SW		 1497 return map;
1498
86f1742b
ID		 1499e_inval:
1500 err = -EINVAL;
f24e9980 1501bad:
38a8d560
ID		 1502 pr_err("corrupt inc osdmap (%d) epoch %d off %d (%p of %p-%p)\n",
1503 err, epoch, (int)(*p - start), *p, start, end);
9ec7cab1
SW		 1504 print_hex_dump(KERN_DEBUG, "osdmap: ",
1505 DUMP_PREFIX_OFFSET, 16, 1,
1506 start, end - start, true);
f24e9980
SW		 1507 if (newcrush)
1508 crush_destroy(newcrush);
1509 return ERR_PTR(err);
1510}
1511
30c156d9
YZ		 1512void ceph_oloc_copy(struct ceph_object_locator *dest,
1513 const struct ceph_object_locator *src)
1514{
1515 WARN_ON(!ceph_oloc_empty(dest));
1516 WARN_ON(dest->pool_ns); /* empty() only covers ->pool */
1517
1518 dest->pool = src->pool;
1519 if (src->pool_ns)
1520 dest->pool_ns = ceph_get_string(src->pool_ns);
1521}
1522EXPORT_SYMBOL(ceph_oloc_copy);
1523
1524void ceph_oloc_destroy(struct ceph_object_locator *oloc)
1525{
1526 ceph_put_string(oloc->pool_ns);
1527}
1528EXPORT_SYMBOL(ceph_oloc_destroy);
1529
d30291b9
ID		 1530void ceph_oid_copy(struct ceph_object_id *dest,
1531 const struct ceph_object_id *src)
1532{
1533 WARN_ON(!ceph_oid_empty(dest));
1534
1535 if (src->name != src->inline_name) {
1536 /* very rare, see ceph_object_id definition */
1537 dest->name = kmalloc(src->name_len + 1,
1538 GFP_NOIO | __GFP_NOFAIL);
1539 }
1540
1541 memcpy(dest->name, src->name, src->name_len + 1);
1542 dest->name_len = src->name_len;
1543}
1544EXPORT_SYMBOL(ceph_oid_copy);
1545
1546static __printf(2, 0)
1547int oid_printf_vargs(struct ceph_object_id *oid, const char *fmt, va_list ap)
1548{
1549 int len;
1550
1551 WARN_ON(!ceph_oid_empty(oid));
1552
1553 len = vsnprintf(oid->inline_name, sizeof(oid->inline_name), fmt, ap);
1554 if (len >= sizeof(oid->inline_name))
1555 return len;
1556
1557 oid->name_len = len;
1558 return 0;
1559}
1560
1561/*
1562 * If oid doesn't fit into inline buffer, BUG.
1563 */
1564void ceph_oid_printf(struct ceph_object_id *oid, const char *fmt, ...)
1565{
1566 va_list ap;
1567
1568 va_start(ap, fmt);
1569 BUG_ON(oid_printf_vargs(oid, fmt, ap));
1570 va_end(ap);
1571}
1572EXPORT_SYMBOL(ceph_oid_printf);
1573
1574static __printf(3, 0)
1575int oid_aprintf_vargs(struct ceph_object_id *oid, gfp_t gfp,
1576 const char *fmt, va_list ap)
1577{
1578 va_list aq;
1579 int len;
1580
1581 va_copy(aq, ap);
1582 len = oid_printf_vargs(oid, fmt, aq);
1583 va_end(aq);
1584
1585 if (len) {
1586 char *external_name;
1587
1588 external_name = kmalloc(len + 1, gfp);
1589 if (!external_name)
1590 return -ENOMEM;
f24e9980 1591
d30291b9
ID		 1592 oid->name = external_name;
1593 WARN_ON(vsnprintf(oid->name, len + 1, fmt, ap) != len);
1594 oid->name_len = len;
1595 }
f24e9980 1596
d30291b9
ID		 1597 return 0;
1598}
1599
1600/*
1601 * If oid doesn't fit into inline buffer, allocate.
1602 */
1603int ceph_oid_aprintf(struct ceph_object_id *oid, gfp_t gfp,
1604 const char *fmt, ...)
1605{
1606 va_list ap;
1607 int ret;
1608
1609 va_start(ap, fmt);
1610 ret = oid_aprintf_vargs(oid, gfp, fmt, ap);
1611 va_end(ap);
1612
1613 return ret;
1614}
1615EXPORT_SYMBOL(ceph_oid_aprintf);
1616
1617void ceph_oid_destroy(struct ceph_object_id *oid)
1618{
1619 if (oid->name != oid->inline_name)
1620 kfree(oid->name);
1621}
1622EXPORT_SYMBOL(ceph_oid_destroy);
f24e9980 1623
63244fa1
ID		 1624/*
1625 * osds only
1626 */
1627static bool __osds_equal(const struct ceph_osds *lhs,
1628 const struct ceph_osds *rhs)
1629{
1630 if (lhs->size == rhs->size &&
1631 !memcmp(lhs->osds, rhs->osds, rhs->size * sizeof(rhs->osds[0])))
1632 return true;
1633
1634 return false;
1635}
1636
1637/*
1638 * osds + primary
1639 */
1640static bool osds_equal(const struct ceph_osds *lhs,
1641 const struct ceph_osds *rhs)
1642{
1643 if (__osds_equal(lhs, rhs) &&
1644 lhs->primary == rhs->primary)
1645 return true;
1646
1647 return false;
1648}
1649
6f3bfd45
ID		 1650static bool osds_valid(const struct ceph_osds *set)
1651{
1652 /* non-empty set */
1653 if (set->size > 0 && set->primary >= 0)
1654 return true;
1655
1656 /* empty can_shift_osds set */
1657 if (!set->size && set->primary == -1)
1658 return true;
1659
1660 /* empty !can_shift_osds set - all NONE */
1661 if (set->size > 0 && set->primary == -1) {
1662 int i;
1663
1664 for (i = 0; i < set->size; i++) {
1665 if (set->osds[i] != CRUSH_ITEM_NONE)
1666 break;
1667 }
1668 if (i == set->size)
1669 return true;
1670 }
1671
1672 return false;
1673}
1674
1675void ceph_osds_copy(struct ceph_osds *dest, const struct ceph_osds *src)
1676{
1677 memcpy(dest->osds, src->osds, src->size * sizeof(src->osds[0]));
1678 dest->size = src->size;
1679 dest->primary = src->primary;
1680}
1681
63244fa1
ID		 1682static bool is_split(const struct ceph_pg *pgid,
1683 u32 old_pg_num,
1684 u32 new_pg_num)
1685{
1686 int old_bits = calc_bits_of(old_pg_num);
1687 int old_mask = (1 << old_bits) - 1;
1688 int n;
1689
1690 WARN_ON(pgid->seed >= old_pg_num);
1691 if (new_pg_num <= old_pg_num)
1692 return false;
1693
1694 for (n = 1; ; n++) {
1695 int next_bit = n << (old_bits - 1);
1696 u32 s = next_bit | pgid->seed;
1697
1698 if (s < old_pg_num || s == pgid->seed)
1699 continue;
1700 if (s >= new_pg_num)
1701 break;
1702
1703 s = ceph_stable_mod(s, old_pg_num, old_mask);
1704 if (s == pgid->seed)
1705 return true;
1706 }
1707
1708 return false;
1709}
1710
1711bool ceph_is_new_interval(const struct ceph_osds *old_acting,
1712 const struct ceph_osds *new_acting,
1713 const struct ceph_osds *old_up,
1714 const struct ceph_osds *new_up,
1715 int old_size,
1716 int new_size,
1717 int old_min_size,
1718 int new_min_size,
1719 u32 old_pg_num,
1720 u32 new_pg_num,
1721 bool old_sort_bitwise,
1722 bool new_sort_bitwise,
1723 const struct ceph_pg *pgid)
1724{
1725 return !osds_equal(old_acting, new_acting) ||
1726 !osds_equal(old_up, new_up) ||
1727 old_size != new_size ||
1728 old_min_size != new_min_size ||
1729 is_split(pgid, old_pg_num, new_pg_num) ||
1730 old_sort_bitwise != new_sort_bitwise;
1731}
1732
1733static int calc_pg_rank(int osd, const struct ceph_osds *acting)
1734{
1735 int i;
1736
1737 for (i = 0; i < acting->size; i++) {
1738 if (acting->osds[i] == osd)
1739 return i;
1740 }
1741
1742 return -1;
1743}
1744
1745static bool primary_changed(const struct ceph_osds *old_acting,
1746 const struct ceph_osds *new_acting)
1747{
1748 if (!old_acting->size && !new_acting->size)
1749 return false; /* both still empty */
1750
1751 if (!old_acting->size ^ !new_acting->size)
1752 return true; /* was empty, now not, or vice versa */
1753
1754 if (old_acting->primary != new_acting->primary)
1755 return true; /* primary changed */
1756
1757 if (calc_pg_rank(old_acting->primary, old_acting) !=
1758 calc_pg_rank(new_acting->primary, new_acting))
1759 return true;
1760
1761 return false; /* same primary (tho replicas may have changed) */
1762}
1763
1764bool ceph_osds_changed(const struct ceph_osds *old_acting,
1765 const struct ceph_osds *new_acting,
1766 bool any_change)
1767{
1768 if (primary_changed(old_acting, new_acting))
1769 return true;
1770
1771 if (any_change && !__osds_equal(old_acting, new_acting))
1772 return true;
1773
1774 return false;
1775}
1776
f24e9980
SW		 1777/*
1778 * calculate file layout from given offset, length.
1779 * fill in correct oid, logical length, and object extent
1780 * offset, length.
1781 *
1782 * for now, we write only a single su, until we can
1783 * pass a stride back to the caller.
1784 */
d63b77f4 1785int ceph_calc_file_object_mapping(struct ceph_file_layout *layout,
e8afad65 1786 u64 off, u64 len,
645a1025 1787 u64 *ono,
f24e9980
SW		 1788 u64 *oxoff, u64 *oxlen)
1789{
7627151e
YZ		 1790 u32 osize = layout->object_size;
1791 u32 su = layout->stripe_unit;
1792 u32 sc = layout->stripe_count;
f24e9980
SW		 1793 u32 bl, stripeno, stripepos, objsetno;
1794 u32 su_per_object;
ff1d1f71 1795 u64 t, su_offset;
f24e9980 1796
e8afad65 1797 dout("mapping %llu~%llu osize %u fl_su %u\n", off, len,
f24e9980 1798 osize, su);
d63b77f4
SW		 1799 if (su == 0 || sc == 0)
1800 goto invalid;
35e054a6 1801 su_per_object = osize / su;
d63b77f4
SW		 1802 if (su_per_object == 0)
1803 goto invalid;
f24e9980
SW		 1804 dout("osize %u / su %u = su_per_object %u\n", osize, su,
1805 su_per_object);
1806
d63b77f4
SW		 1807 if ((su & ~PAGE_MASK) != 0)
1808 goto invalid;
1809
f24e9980
SW		 1810 /* bl = *off / su; */
1811 t = off;
1812 do_div(t, su);
1813 bl = t;
1814 dout("off %llu / su %u = bl %u\n", off, su, bl);
1815
1816 stripeno = bl / sc;
1817 stripepos = bl % sc;
1818 objsetno = stripeno / su_per_object;
1819
645a1025 1820 *ono = objsetno * sc + stripepos;
95c96174 1821 dout("objset %u * sc %u = ono %u\n", objsetno, sc, (unsigned int)*ono);
645a1025
SW		 1822
1823 /* *oxoff = *off % layout->fl_stripe_unit; # offset in su */
f24e9980 1824 t = off;
ff1d1f71
NW		 1825 su_offset = do_div(t, su);
1826 *oxoff = su_offset + (stripeno % su_per_object) * su;
1827
1828 /*
1829 * Calculate the length of the extent being written to the selected
e8afad65 1830 * object. This is the minimum of the full length requested (len) or
ff1d1f71
NW		 1831 * the remainder of the current stripe being written to.
1832 */
e8afad65 1833 *oxlen = min_t(u64, len, su - su_offset);
f24e9980
SW		 1834
1835 dout(" obj extent %llu~%llu\n", *oxoff, *oxlen);
d63b77f4
SW		 1836 return 0;
1837
1838invalid:
1839 dout(" invalid layout\n");
1840 *ono = 0;
1841 *oxoff = 0;
1842 *oxlen = 0;
1843 return -EINVAL;
f24e9980 1844}
3d14c5d2 1845EXPORT_SYMBOL(ceph_calc_file_object_mapping);
f24e9980
SW		 1846
1847/*
d9591f5e
ID		 1848 * Map an object into a PG.
1849 *
1850 * Should only be called with target_oid and target_oloc (as opposed to
1851 * base_oid and base_oloc), since tiering isn't taken into account.
f24e9980 1852 */
d9591f5e
ID		 1853int ceph_object_locator_to_pg(struct ceph_osdmap *osdmap,
1854 struct ceph_object_id *oid,
1855 struct ceph_object_locator *oloc,
1856 struct ceph_pg *raw_pgid)
f24e9980 1857{
7c13cb64 1858 struct ceph_pg_pool_info *pi;
f24e9980 1859
d9591f5e 1860 pi = ceph_pg_pool_by_id(osdmap, oloc->pool);
7c13cb64 1861 if (!pi)
d9591f5e 1862 return -ENOENT;
f24e9980 1863
30c156d9
YZ		 1864 if (!oloc->pool_ns) {
1865 raw_pgid->pool = oloc->pool;
1866 raw_pgid->seed = ceph_str_hash(pi->object_hash, oid->name,
1867 oid->name_len);
1868 dout("%s %s -> raw_pgid %llu.%x\n", __func__, oid->name,
1869 raw_pgid->pool, raw_pgid->seed);
1870 } else {
1871 char stack_buf[256];
1872 char *buf = stack_buf;
1873 int nsl = oloc->pool_ns->len;
1874 size_t total = nsl + 1 + oid->name_len;
1875
1876 if (total > sizeof(stack_buf)) {
1877 buf = kmalloc(total, GFP_NOIO);
1878 if (!buf)
1879 return -ENOMEM;
1880 }
1881 memcpy(buf, oloc->pool_ns->str, nsl);
1882 buf[nsl] = '\037';
1883 memcpy(buf + nsl + 1, oid->name, oid->name_len);
1884 raw_pgid->pool = oloc->pool;
1885 raw_pgid->seed = ceph_str_hash(pi->object_hash, buf, total);
1886 if (buf != stack_buf)
1887 kfree(buf);
1888 dout("%s %s ns %.*s -> raw_pgid %llu.%x\n", __func__,
1889 oid->name, nsl, oloc->pool_ns->str,
1890 raw_pgid->pool, raw_pgid->seed);
1891 }
f24e9980
SW		 1892 return 0;
1893}
d9591f5e 1894EXPORT_SYMBOL(ceph_object_locator_to_pg);
f24e9980 1895
6f3bfd45
ID		 1896/*
1897 * Map a raw PG (full precision ps) into an actual PG.
1898 */
1899static void raw_pg_to_pg(struct ceph_pg_pool_info *pi,
1900 const struct ceph_pg *raw_pgid,
1901 struct ceph_pg *pgid)
1902{
1903 pgid->pool = raw_pgid->pool;
1904 pgid->seed = ceph_stable_mod(raw_pgid->seed, pi->pg_num,
1905 pi->pg_num_mask);
1906}
1907
1908/*
1909 * Map a raw PG (full precision ps) into a placement ps (placement
1910 * seed). Include pool id in that value so that different pools don't
1911 * use the same seeds.
1912 */
1913static u32 raw_pg_to_pps(struct ceph_pg_pool_info *pi,
1914 const struct ceph_pg *raw_pgid)
1915{
1916 if (pi->flags & CEPH_POOL_FLAG_HASHPSPOOL) {
1917 /* hash pool id and seed so that pool PGs do not overlap */
1918 return crush_hash32_2(CRUSH_HASH_RJENKINS1,
1919 ceph_stable_mod(raw_pgid->seed,
1920 pi->pgp_num,
1921 pi->pgp_num_mask),
1922 raw_pgid->pool);
1923 } else {
1924 /*
1925 * legacy behavior: add ps and pool together. this is
1926 * not a great approach because the PGs from each pool
1927 * will overlap on top of each other: 0.5 == 1.4 ==
1928 * 2.3 == ...
1929 */
1930 return ceph_stable_mod(raw_pgid->seed, pi->pgp_num,
1931 pi->pgp_num_mask) +
1932 (unsigned)raw_pgid->pool;
1933 }
1934}
1935
9d521470
ID		 1936static int do_crush(struct ceph_osdmap *map, int ruleno, int x,
1937 int *result, int result_max,
1938 const __u32 *weight, int weight_max)
e8ef19c4 1939{
9d521470
ID		 1940 int r;
1941
1942 BUG_ON(result_max > CEPH_PG_MAX_SIZE);
1943
1944 mutex_lock(&map->crush_scratch_mutex);
1945 r = crush_do_rule(map->crush, ruleno, x, result, result_max,
1946 weight, weight_max, map->crush_scratch_ary);
1947 mutex_unlock(&map->crush_scratch_mutex);
e8ef19c4 1948
9d521470 1949 return r;
e8ef19c4
ID		 1950}
1951
2bd93d4d 1952/*
6f3bfd45
ID		 1953 * Calculate raw set (CRUSH output) for given PG. The result may
1954 * contain nonexistent OSDs. ->primary is undefined for a raw set.
2bd93d4d 1955 *
6f3bfd45 1956 * Placement seed (CRUSH input) is returned through @ppps.
2bd93d4d 1957 */
6f3bfd45
ID		 1958static void pg_to_raw_osds(struct ceph_osdmap *osdmap,
1959 struct ceph_pg_pool_info *pi,
1960 const struct ceph_pg *raw_pgid,
1961 struct ceph_osds *raw,
1962 u32 *ppps)
2bd93d4d 1963{
6f3bfd45 1964 u32 pps = raw_pg_to_pps(pi, raw_pgid);
2bd93d4d
ID		 1965 int ruleno;
1966 int len;
1967
6f3bfd45
ID		 1968 ceph_osds_init(raw);
1969 if (ppps)
1970 *ppps = pps;
1971
1972 ruleno = crush_find_rule(osdmap->crush, pi->crush_ruleset, pi->type,
1973 pi->size);
2bd93d4d
ID		 1974 if (ruleno < 0) {
1975 pr_err("no crush rule: pool %lld ruleset %d type %d size %d\n",
6f3bfd45
ID		 1976 pi->id, pi->crush_ruleset, pi->type, pi->size);
1977 return;
2bd93d4d
ID		 1978 }
1979
6f3bfd45
ID		 1980 len = do_crush(osdmap, ruleno, pps, raw->osds,
1981 min_t(int, pi->size, ARRAY_SIZE(raw->osds)),
2bd93d4d
ID		 1982 osdmap->osd_weight, osdmap->max_osd);
1983 if (len < 0) {
1984 pr_err("error %d from crush rule %d: pool %lld ruleset %d type %d size %d\n",
6f3bfd45
ID		 1985 len, ruleno, pi->id, pi->crush_ruleset, pi->type,
1986 pi->size);
1987 return;
2bd93d4d
ID		 1988 }
1989
6f3bfd45 1990 raw->size = len;
2bd93d4d
ID		 1991}
1992
1993/*
6f3bfd45
ID		 1994 * Given raw set, calculate up set and up primary. By definition of an
1995 * up set, the result won't contain nonexistent or down OSDs.
2bd93d4d 1996 *
6f3bfd45
ID		 1997 * This is done in-place - on return @set is the up set. If it's
1998 * empty, ->primary will remain undefined.
2bd93d4d 1999 */
6f3bfd45
ID		 2000static void raw_to_up_osds(struct ceph_osdmap *osdmap,
2001 struct ceph_pg_pool_info *pi,
2002 struct ceph_osds *set)
2bd93d4d 2003{
2bd93d4d
ID		 2004 int i;
2005
6f3bfd45
ID		 2006 /* ->primary is undefined for a raw set */
2007 BUG_ON(set->primary != -1);
2008
2009 if (ceph_can_shift_osds(pi)) {
2bd93d4d
ID		 2010 int removed = 0;
2011
6f3bfd45
ID		 2012 /* shift left */
2013 for (i = 0; i < set->size; i++) {
2014 if (ceph_osd_is_down(osdmap, set->osds[i])) {
2bd93d4d
ID		 2015 removed++;
2016 continue;
2017 }
2018 if (removed)
6f3bfd45 2019 set->osds[i - removed] = set->osds[i];
2bd93d4d 2020 }
6f3bfd45
ID		 2021 set->size -= removed;
2022 if (set->size > 0)
2023 set->primary = set->osds[0];
2bd93d4d 2024 } else {
6f3bfd45
ID		 2025 /* set down/dne devices to NONE */
2026 for (i = set->size - 1; i >= 0; i--) {
2027 if (ceph_osd_is_down(osdmap, set->osds[i]))
2028 set->osds[i] = CRUSH_ITEM_NONE;
2bd93d4d 2029 else
6f3bfd45 2030 set->primary = set->osds[i];
2bd93d4d
ID		 2031 }
2032 }
2bd93d4d
ID		 2033}
2034
6f3bfd45
ID		 2035static void apply_primary_affinity(struct ceph_osdmap *osdmap,
2036 struct ceph_pg_pool_info *pi,
2037 u32 pps,
2038 struct ceph_osds *up)
47ec1f3c
ID		 2039{
2040 int i;
2041 int pos = -1;
2042
2043 /*
2044 * Do we have any non-default primary_affinity values for these
2045 * osds?
2046 */
2047 if (!osdmap->osd_primary_affinity)
2048 return;
2049
6f3bfd45
ID		 2050 for (i = 0; i < up->size; i++) {
2051 int osd = up->osds[i];
92b2e751
ID		 2052
2053 if (osd != CRUSH_ITEM_NONE &&
2054 osdmap->osd_primary_affinity[osd] !=
47ec1f3c
ID		 2055 CEPH_OSD_DEFAULT_PRIMARY_AFFINITY) {
2056 break;
2057 }
2058 }
6f3bfd45 2059 if (i == up->size)
47ec1f3c
ID		 2060 return;
2061
2062 /*
2063 * Pick the primary. Feed both the seed (for the pg) and the
2064 * osd into the hash/rng so that a proportional fraction of an
2065 * osd's pgs get rejected as primary.
2066 */
6f3bfd45
ID		 2067 for (i = 0; i < up->size; i++) {
2068 int osd = up->osds[i];
47ec1f3c
ID		 2069 u32 aff;
2070
47ec1f3c
ID		 2071 if (osd == CRUSH_ITEM_NONE)
2072 continue;
2073
2074 aff = osdmap->osd_primary_affinity[osd];
2075 if (aff < CEPH_OSD_MAX_PRIMARY_AFFINITY &&
2076 (crush_hash32_2(CRUSH_HASH_RJENKINS1,
2077 pps, osd) >> 16) >= aff) {
2078 /*
2079 * We chose not to use this primary. Note it
2080 * anyway as a fallback in case we don't pick
2081 * anyone else, but keep looking.
2082 */
2083 if (pos < 0)
2084 pos = i;
2085 } else {
2086 pos = i;
2087 break;
2088 }
2089 }
2090 if (pos < 0)
2091 return;
2092
6f3bfd45 2093 up->primary = up->osds[pos];
47ec1f3c 2094
6f3bfd45 2095 if (ceph_can_shift_osds(pi) && pos > 0) {
47ec1f3c
ID		 2096 /* move the new primary to the front */
2097 for (i = pos; i > 0; i--)
6f3bfd45
ID		 2098 up->osds[i] = up->osds[i - 1];
2099 up->osds[0] = up->primary;
47ec1f3c
ID		 2100 }
2101}
2102
45966c34 2103/*
6f3bfd45 2104 * Get pg_temp and primary_temp mappings for given PG.
45966c34 2105 *
6f3bfd45
ID		 2106 * Note that a PG may have none, only pg_temp, only primary_temp or
2107 * both pg_temp and primary_temp mappings. This means @temp isn't
2108 * always a valid OSD set on return: in the "only primary_temp" case,
2109 * @temp will have its ->primary >= 0 but ->size == 0.
45966c34 2110 */
6f3bfd45
ID		 2111static void get_temp_osds(struct ceph_osdmap *osdmap,
2112 struct ceph_pg_pool_info *pi,
2113 const struct ceph_pg *raw_pgid,
2114 struct ceph_osds *temp)
45966c34 2115{
6f3bfd45 2116 struct ceph_pg pgid;
45966c34 2117 struct ceph_pg_mapping *pg;
45966c34
ID		 2118 int i;
2119
6f3bfd45
ID		 2120 raw_pg_to_pg(pi, raw_pgid, &pgid);
2121 ceph_osds_init(temp);
45966c34
ID		 2122
2123 /* pg_temp? */
2124 pg = __lookup_pg_mapping(&osdmap->pg_temp, pgid);
2125 if (pg) {
45966c34
ID		 2126 for (i = 0; i < pg->pg_temp.len; i++) {
2127 if (ceph_osd_is_down(osdmap, pg->pg_temp.osds[i])) {
6f3bfd45 2128 if (ceph_can_shift_osds(pi))
45966c34 2129 continue;
6f3bfd45
ID		 2130
2131 temp->osds[temp->size++] = CRUSH_ITEM_NONE;
45966c34 2132 } else {
6f3bfd45 2133 temp->osds[temp->size++] = pg->pg_temp.osds[i];
45966c34
ID		 2134 }
2135 }
2136
2137 /* apply pg_temp's primary */
6f3bfd45
ID		 2138 for (i = 0; i < temp->size; i++) {
2139 if (temp->osds[i] != CRUSH_ITEM_NONE) {
2140 temp->primary = temp->osds[i];
45966c34
ID		 2141 break;
2142 }
2143 }
45966c34
ID		 2144 }
2145
5e8d4d36
ID		 2146 /* primary_temp? */
2147 pg = __lookup_pg_mapping(&osdmap->primary_temp, pgid);
2148 if (pg)
6f3bfd45 2149 temp->primary = pg->primary_temp.osd;
45966c34
ID		 2150}
2151
d85b7056 2152/*
6f3bfd45 2153 * Map a PG to its acting set as well as its up set.
ac972230 2154 *
6f3bfd45
ID		 2155 * Acting set is used for data mapping purposes, while up set can be
2156 * recorded for detecting interval changes and deciding whether to
2157 * resend a request.
d85b7056 2158 */
6f3bfd45
ID		 2159void ceph_pg_to_up_acting_osds(struct ceph_osdmap *osdmap,
2160 const struct ceph_pg *raw_pgid,
2161 struct ceph_osds *up,
2162 struct ceph_osds *acting)
d85b7056 2163{
6f3bfd45 2164 struct ceph_pg_pool_info *pi;
ac972230 2165 u32 pps;
d85b7056 2166
6f3bfd45
ID		 2167 pi = ceph_pg_pool_by_id(osdmap, raw_pgid->pool);
2168 if (!pi) {
2169 ceph_osds_init(up);
2170 ceph_osds_init(acting);
2171 goto out;
8008ab10 2172 }
d85b7056 2173
6f3bfd45
ID		 2174 pg_to_raw_osds(osdmap, pi, raw_pgid, up, &pps);
2175 raw_to_up_osds(osdmap, pi, up);
2176 apply_primary_affinity(osdmap, pi, pps, up);
2177 get_temp_osds(osdmap, pi, raw_pgid, acting);
2178 if (!acting->size) {
2179 memcpy(acting->osds, up->osds, up->size * sizeof(up->osds[0]));
2180 acting->size = up->size;
2181 if (acting->primary == -1)
2182 acting->primary = up->primary;
8008ab10 2183 }
6f3bfd45
ID		 2184out:
2185 WARN_ON(!osds_valid(up) || !osds_valid(acting));
d85b7056
SW		 2186}
2187
f24e9980 2188/*
f81f1633 2189 * Return acting primary for given PG, or -1 if none.
f24e9980 2190 */
f81f1633
ID		 2191int ceph_pg_to_acting_primary(struct ceph_osdmap *osdmap,
2192 const struct ceph_pg *raw_pgid)
f24e9980 2193{
6f3bfd45 2194 struct ceph_osds up, acting;
f24e9980 2195
f81f1633 2196 ceph_pg_to_up_acting_osds(osdmap, raw_pgid, &up, &acting);
6f3bfd45 2197 return acting.primary;
f24e9980 2198}
f81f1633 2199EXPORT_SYMBOL(ceph_pg_to_acting_primary);
ubuntu-zesty-kernel mirror