]> git.proxmox.com Git - mirror_ubuntu-artful-kernel.git/blame - net/ceph/osdmap.c
projects / mirror_ubuntu-artful-kernel.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
libceph: use target pi for calc_target() calculations
[mirror_ubuntu-artful-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
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
f984cb76
ID		 421/*
422 * rbtree of pg_mapping for handling pg_temp (explicit mapping of pgid
423 * to a set of osds) and primary_temp (explicit primary setting)
424 */
991abb6e
SW		 425static int __insert_pg_mapping(struct ceph_pg_mapping *new,
426 struct rb_root *root)
f24e9980
SW		 427{
428 struct rb_node **p = &root->rb_node;
429 struct rb_node *parent = NULL;
430 struct ceph_pg_mapping *pg = NULL;
51042122 431 int c;
f24e9980 432
8adc8b3d 433 dout("__insert_pg_mapping %llx %p\n", *(u64 *)&new->pgid, new);
f24e9980
SW		 434 while (*p) {
435 parent = *p;
436 pg = rb_entry(parent, struct ceph_pg_mapping, node);
f984cb76 437 c = ceph_pg_compare(&new->pgid, &pg->pgid);
51042122 438 if (c < 0)
f24e9980 439 p = &(*p)->rb_left;
51042122 440 else if (c > 0)
f24e9980
SW		 441 p = &(*p)->rb_right;
442 else
991abb6e 443 return -EEXIST;
f24e9980
SW		 444 }
445
446 rb_link_node(&new->node, parent, p);
447 rb_insert_color(&new->node, root);
991abb6e 448 return 0;
f24e9980
SW		 449}
450
9794b146 451static struct ceph_pg_mapping *__lookup_pg_mapping(struct rb_root *root,
5b191d99 452 struct ceph_pg pgid)
9794b146
SW		 453{
454 struct rb_node *n = root->rb_node;
455 struct ceph_pg_mapping *pg;
456 int c;
457
458 while (n) {
459 pg = rb_entry(n, struct ceph_pg_mapping, node);
f984cb76 460 c = ceph_pg_compare(&pgid, &pg->pgid);
8adc8b3d 461 if (c < 0) {
9794b146 462 n = n->rb_left;
8adc8b3d 463 } else if (c > 0) {
9794b146 464 n = n->rb_right;
8adc8b3d 465 } else {
5b191d99
SW		 466 dout("__lookup_pg_mapping %lld.%x got %p\n",
467 pgid.pool, pgid.seed, pg);
9794b146 468 return pg;
8adc8b3d 469 }
9794b146
SW		 470 }
471 return NULL;
472}
473
5b191d99 474static int __remove_pg_mapping(struct rb_root *root, struct ceph_pg pgid)
8adc8b3d
SW		 475{
476 struct ceph_pg_mapping *pg = __lookup_pg_mapping(root, pgid);
477
478 if (pg) {
5b191d99
SW		 479 dout("__remove_pg_mapping %lld.%x %p\n", pgid.pool, pgid.seed,
480 pg);
8adc8b3d
SW		 481 rb_erase(&pg->node, root);
482 kfree(pg);
483 return 0;
484 }
5b191d99 485 dout("__remove_pg_mapping %lld.%x dne\n", pgid.pool, pgid.seed);
8adc8b3d
SW		 486 return -ENOENT;
487}
488
4fc51be8
SW		 489/*
490 * rbtree of pg pool info
491 */
492static int __insert_pg_pool(struct rb_root *root, struct ceph_pg_pool_info *new)
493{
494 struct rb_node **p = &root->rb_node;
495 struct rb_node *parent = NULL;
496 struct ceph_pg_pool_info *pi = NULL;
497
498 while (*p) {
499 parent = *p;
500 pi = rb_entry(parent, struct ceph_pg_pool_info, node);
501 if (new->id < pi->id)
502 p = &(*p)->rb_left;
503 else if (new->id > pi->id)
504 p = &(*p)->rb_right;
505 else
506 return -EEXIST;
507 }
508
509 rb_link_node(&new->node, parent, p);
510 rb_insert_color(&new->node, root);
511 return 0;
512}
513
4f6a7e5e 514static struct ceph_pg_pool_info *__lookup_pg_pool(struct rb_root *root, u64 id)
4fc51be8
SW		 515{
516 struct ceph_pg_pool_info *pi;
517 struct rb_node *n = root->rb_node;
518
519 while (n) {
520 pi = rb_entry(n, struct ceph_pg_pool_info, node);
521 if (id < pi->id)
522 n = n->rb_left;
523 else if (id > pi->id)
524 n = n->rb_right;
525 else
526 return pi;
527 }
528 return NULL;
529}
530
ce7f6a27
ID		 531struct ceph_pg_pool_info *ceph_pg_pool_by_id(struct ceph_osdmap *map, u64 id)
532{
533 return __lookup_pg_pool(&map->pg_pools, id);
534}
535
72afc71f
AE		 536const char *ceph_pg_pool_name_by_id(struct ceph_osdmap *map, u64 id)
537{
538 struct ceph_pg_pool_info *pi;
539
540 if (id == CEPH_NOPOOL)
541 return NULL;
542
543 if (WARN_ON_ONCE(id > (u64) INT_MAX))
544 return NULL;
545
546 pi = __lookup_pg_pool(&map->pg_pools, (int) id);
547
548 return pi ? pi->name : NULL;
549}
550EXPORT_SYMBOL(ceph_pg_pool_name_by_id);
551
7669a2c9
YS		 552int ceph_pg_poolid_by_name(struct ceph_osdmap *map, const char *name)
553{
554 struct rb_node *rbp;
555
556 for (rbp = rb_first(&map->pg_pools); rbp; rbp = rb_next(rbp)) {
557 struct ceph_pg_pool_info *pi =
558 rb_entry(rbp, struct ceph_pg_pool_info, node);
559 if (pi->name && strcmp(pi->name, name) == 0)
560 return pi->id;
561 }
562 return -ENOENT;
563}
3d14c5d2 564EXPORT_SYMBOL(ceph_pg_poolid_by_name);
7669a2c9 565
2844a76a
SW		 566static void __remove_pg_pool(struct rb_root *root, struct ceph_pg_pool_info *pi)
567{
568 rb_erase(&pi->node, root);
569 kfree(pi->name);
570 kfree(pi);
571}
572
0f70c7ee 573static int decode_pool(void **p, void *end, struct ceph_pg_pool_info *pi)
efd7576b 574{
4f6a7e5e
SW		 575 u8 ev, cv;
576 unsigned len, num;
577 void *pool_end;
578
579 ceph_decode_need(p, end, 2 + 4, bad);
580 ev = ceph_decode_8(p); /* encoding version */
581 cv = ceph_decode_8(p); /* compat version */
582 if (ev < 5) {
b9a67899 583 pr_warn("got v %d < 5 cv %d of ceph_pg_pool\n", ev, cv);
4f6a7e5e
SW		 584 return -EINVAL;
585 }
17a13e40 586 if (cv > 9) {
b9a67899 587 pr_warn("got v %d cv %d > 9 of ceph_pg_pool\n", ev, cv);
4f6a7e5e
SW		 588 return -EINVAL;
589 }
590 len = ceph_decode_32(p);
591 ceph_decode_need(p, end, len, bad);
592 pool_end = *p + len;
73a7e693 593
4f6a7e5e
SW		 594 pi->type = ceph_decode_8(p);
595 pi->size = ceph_decode_8(p);
596 pi->crush_ruleset = ceph_decode_8(p);
597 pi->object_hash = ceph_decode_8(p);
73a7e693 598
4f6a7e5e
SW		 599 pi->pg_num = ceph_decode_32(p);
600 pi->pgp_num = ceph_decode_32(p);
601
602 *p += 4 + 4; /* skip lpg* */
603 *p += 4; /* skip last_change */
604 *p += 8 + 4; /* skip snap_seq, snap_epoch */
605
606 /* skip snaps */
607 num = ceph_decode_32(p);
608 while (num--) {
609 *p += 8; /* snapid key */
610 *p += 1 + 1; /* versions */
611 len = ceph_decode_32(p);
612 *p += len;
73a7e693
SW		 613 }
614
17a13e40 615 /* skip removed_snaps */
4f6a7e5e
SW		 616 num = ceph_decode_32(p);
617 *p += num * (8 + 8);
618
619 *p += 8; /* skip auid */
620 pi->flags = ceph_decode_64(p);
17a13e40
ID		 621 *p += 4; /* skip crash_replay_interval */
622
623 if (ev >= 7)
04812acf
ID		 624 pi->min_size = ceph_decode_8(p);
625 else
626 pi->min_size = pi->size - pi->size / 2;
17a13e40
ID		 627
628 if (ev >= 8)
629 *p += 8 + 8; /* skip quota_max_* */
630
631 if (ev >= 9) {
632 /* skip tiers */
633 num = ceph_decode_32(p);
634 *p += num * 8;
635
636 *p += 8; /* skip tier_of */
637 *p += 1; /* skip cache_mode */
638
639 pi->read_tier = ceph_decode_64(p);
640 pi->write_tier = ceph_decode_64(p);
641 } else {
642 pi->read_tier = -1;
643 pi->write_tier = -1;
644 }
4f6a7e5e 645
04812acf
ID		 646 if (ev >= 10) {
647 /* skip properties */
648 num = ceph_decode_32(p);
649 while (num--) {
650 len = ceph_decode_32(p);
651 *p += len; /* key */
652 len = ceph_decode_32(p);
653 *p += len; /* val */
654 }
655 }
656
657 if (ev >= 11) {
658 /* skip hit_set_params */
659 *p += 1 + 1; /* versions */
660 len = ceph_decode_32(p);
661 *p += len;
662
663 *p += 4; /* skip hit_set_period */
664 *p += 4; /* skip hit_set_count */
665 }
666
667 if (ev >= 12)
668 *p += 4; /* skip stripe_width */
669
670 if (ev >= 13) {
671 *p += 8; /* skip target_max_bytes */
672 *p += 8; /* skip target_max_objects */
673 *p += 4; /* skip cache_target_dirty_ratio_micro */
674 *p += 4; /* skip cache_target_full_ratio_micro */
675 *p += 4; /* skip cache_min_flush_age */
676 *p += 4; /* skip cache_min_evict_age */
677 }
678
679 if (ev >= 14) {
680 /* skip erasure_code_profile */
681 len = ceph_decode_32(p);
682 *p += len;
683 }
684
8e48cf00
ID		 685 /*
686 * last_force_op_resend_preluminous, will be overridden if the
687 * map was encoded with RESEND_ON_SPLIT
688 */
04812acf
ID		 689 if (ev >= 15)
690 pi->last_force_request_resend = ceph_decode_32(p);
691 else
692 pi->last_force_request_resend = 0;
693
8e48cf00
ID		 694 if (ev >= 16)
695 *p += 4; /* skip min_read_recency_for_promote */
696
697 if (ev >= 17)
698 *p += 8; /* skip expected_num_objects */
699
700 if (ev >= 19)
701 *p += 4; /* skip cache_target_dirty_high_ratio_micro */
702
703 if (ev >= 20)
704 *p += 4; /* skip min_write_recency_for_promote */
705
706 if (ev >= 21)
707 *p += 1; /* skip use_gmt_hitset */
708
709 if (ev >= 22)
710 *p += 1; /* skip fast_read */
711
712 if (ev >= 23) {
713 *p += 4; /* skip hit_set_grade_decay_rate */
714 *p += 4; /* skip hit_set_search_last_n */
715 }
716
717 if (ev >= 24) {
718 /* skip opts */
719 *p += 1 + 1; /* versions */
720 len = ceph_decode_32(p);
721 *p += len;
722 }
723
724 if (ev >= 25)
725 pi->last_force_request_resend = ceph_decode_32(p);
726
4f6a7e5e
SW		 727 /* ignore the rest */
728
729 *p = pool_end;
730 calc_pg_masks(pi);
73a7e693
SW		 731 return 0;
732
733bad:
734 return -EINVAL;
efd7576b
SW		 735}
736
0f70c7ee 737static int decode_pool_names(void **p, void *end, struct ceph_osdmap *map)
2844a76a
SW		 738{
739 struct ceph_pg_pool_info *pi;
4f6a7e5e
SW		 740 u32 num, len;
741 u64 pool;
2844a76a
SW		 742
743 ceph_decode_32_safe(p, end, num, bad);
744 dout(" %d pool names\n", num);
745 while (num--) {
4f6a7e5e 746 ceph_decode_64_safe(p, end, pool, bad);
2844a76a 747 ceph_decode_32_safe(p, end, len, bad);
4f6a7e5e 748 dout(" pool %llu len %d\n", pool, len);
ad3b904c 749 ceph_decode_need(p, end, len, bad);
2844a76a
SW		 750 pi = __lookup_pg_pool(&map->pg_pools, pool);
751 if (pi) {
ad3b904c
XW		 752 char *name = kstrndup(*p, len, GFP_NOFS);
753
754 if (!name)
755 return -ENOMEM;
2844a76a 756 kfree(pi->name);
ad3b904c
XW		 757 pi->name = name;
758 dout(" name is %s\n", pi->name);
2844a76a
SW		 759 }
760 *p += len;
761 }
762 return 0;
763
764bad:
765 return -EINVAL;
766}
767
768/*
769 * osd map
770 */
e5253a7b
ID		 771struct ceph_osdmap *ceph_osdmap_alloc(void)
772{
773 struct ceph_osdmap *map;
774
775 map = kzalloc(sizeof(*map), GFP_NOIO);
776 if (!map)
777 return NULL;
778
779 map->pg_pools = RB_ROOT;
780 map->pool_max = -1;
781 map->pg_temp = RB_ROOT;
782 map->primary_temp = RB_ROOT;
743efcff 783 mutex_init(&map->crush_workspace_mutex);
e5253a7b
ID		 784
785 return map;
786}
787
2844a76a
SW		 788void ceph_osdmap_destroy(struct ceph_osdmap *map)
789{
790 dout("osdmap_destroy %p\n", map);
791 if (map->crush)
792 crush_destroy(map->crush);
793 while (!RB_EMPTY_ROOT(&map->pg_temp)) {
794 struct ceph_pg_mapping *pg =
795 rb_entry(rb_first(&map->pg_temp),
796 struct ceph_pg_mapping, node);
797 rb_erase(&pg->node, &map->pg_temp);
798 kfree(pg);
799 }
9686f94c
ID		 800 while (!RB_EMPTY_ROOT(&map->primary_temp)) {
801 struct ceph_pg_mapping *pg =
802 rb_entry(rb_first(&map->primary_temp),
803 struct ceph_pg_mapping, node);
804 rb_erase(&pg->node, &map->primary_temp);
805 kfree(pg);
806 }
2844a76a
SW		 807 while (!RB_EMPTY_ROOT(&map->pg_pools)) {
808 struct ceph_pg_pool_info *pi =
809 rb_entry(rb_first(&map->pg_pools),
810 struct ceph_pg_pool_info, node);
811 __remove_pg_pool(&map->pg_pools, pi);
812 }
813 kfree(map->osd_state);
814 kfree(map->osd_weight);
815 kfree(map->osd_addr);
2cfa34f2 816 kfree(map->osd_primary_affinity);
66a0e2d5 817 kfree(map->crush_workspace);
2844a76a
SW		 818 kfree(map);
819}
820
821/*
4d60351f
ID		 822 * Adjust max_osd value, (re)allocate arrays.
823 *
824 * The new elements are properly initialized.
2844a76a
SW		 825 */
826static int osdmap_set_max_osd(struct ceph_osdmap *map, int max)
827{
828 u8 *state;
2844a76a 829 u32 *weight;
4d60351f
ID		 830 struct ceph_entity_addr *addr;
831 int i;
2844a76a 832
4d60351f 833 state = krealloc(map->osd_state, max*sizeof(*state), GFP_NOFS);
589506f1
LR		 834 if (!state)
835 return -ENOMEM;
836 map->osd_state = state;
837
4d60351f 838 weight = krealloc(map->osd_weight, max*sizeof(*weight), GFP_NOFS);
589506f1
LR		 839 if (!weight)
840 return -ENOMEM;
841 map->osd_weight = weight;
4d60351f 842
589506f1
LR		 843 addr = krealloc(map->osd_addr, max*sizeof(*addr), GFP_NOFS);
844 if (!addr)
2844a76a 845 return -ENOMEM;
589506f1 846 map->osd_addr = addr;
2844a76a 847
4d60351f 848 for (i = map->max_osd; i < max; i++) {
589506f1
LR		 849 map->osd_state[i] = 0;
850 map->osd_weight[i] = CEPH_OSD_OUT;
851 memset(map->osd_addr + i, 0, sizeof(*map->osd_addr));
2844a76a
SW		 852 }
853
2cfa34f2
ID		 854 if (map->osd_primary_affinity) {
855 u32 *affinity;
856
857 affinity = krealloc(map->osd_primary_affinity,
858 max*sizeof(*affinity), GFP_NOFS);
859 if (!affinity)
860 return -ENOMEM;
589506f1 861 map->osd_primary_affinity = affinity;
2cfa34f2
ID		 862
863 for (i = map->max_osd; i < max; i++)
589506f1
LR		 864 map->osd_primary_affinity[i] =
865 CEPH_OSD_DEFAULT_PRIMARY_AFFINITY;
2cfa34f2
ID		 866 }
867
2844a76a 868 map->max_osd = max;
4d60351f 869
2844a76a
SW		 870 return 0;
871}
872
1b6a78b5
ID		 873static int osdmap_set_crush(struct ceph_osdmap *map, struct crush_map *crush)
874{
66a0e2d5 875 void *workspace;
743efcff 876 size_t work_size;
66a0e2d5 877
1b6a78b5
ID		 878 if (IS_ERR(crush))
879 return PTR_ERR(crush);
880
743efcff
ID		 881 work_size = crush_work_size(crush, CEPH_PG_MAX_SIZE);
882 dout("%s work_size %zu bytes\n", __func__, work_size);
883 workspace = kmalloc(work_size, GFP_NOIO);
66a0e2d5
ID		 884 if (!workspace) {
885 crush_destroy(crush);
886 return -ENOMEM;
887 }
888 crush_init_workspace(crush, workspace);
889
1b6a78b5
ID		 890 if (map->crush)
891 crush_destroy(map->crush);
66a0e2d5 892 kfree(map->crush_workspace);
1b6a78b5 893 map->crush = crush;
66a0e2d5 894 map->crush_workspace = workspace;
1b6a78b5
ID		 895 return 0;
896}
897
ec7af972
ID		 898#define OSDMAP_WRAPPER_COMPAT_VER 7
899#define OSDMAP_CLIENT_DATA_COMPAT_VER 1
900
901/*
902 * Return 0 or error. On success, *v is set to 0 for old (v6) osdmaps,
903 * to struct_v of the client_data section for new (v7 and above)
904 * osdmaps.
905 */
906static int get_osdmap_client_data_v(void **p, void *end,
907 const char *prefix, u8 *v)
908{
909 u8 struct_v;
910
911 ceph_decode_8_safe(p, end, struct_v, e_inval);
912 if (struct_v >= 7) {
913 u8 struct_compat;
914
915 ceph_decode_8_safe(p, end, struct_compat, e_inval);
916 if (struct_compat > OSDMAP_WRAPPER_COMPAT_VER) {
b9a67899
JP		 917 pr_warn("got v %d cv %d > %d of %s ceph_osdmap\n",
918 struct_v, struct_compat,
919 OSDMAP_WRAPPER_COMPAT_VER, prefix);
ec7af972
ID		 920 return -EINVAL;
921 }
922 *p += 4; /* ignore wrapper struct_len */
923
924 ceph_decode_8_safe(p, end, struct_v, e_inval);
925 ceph_decode_8_safe(p, end, struct_compat, e_inval);
926 if (struct_compat > OSDMAP_CLIENT_DATA_COMPAT_VER) {
b9a67899
JP		 927 pr_warn("got v %d cv %d > %d of %s ceph_osdmap client data\n",
928 struct_v, struct_compat,
929 OSDMAP_CLIENT_DATA_COMPAT_VER, prefix);
ec7af972
ID		 930 return -EINVAL;
931 }
932 *p += 4; /* ignore client data struct_len */
933 } else {
934 u16 version;
935
936 *p -= 1;
937 ceph_decode_16_safe(p, end, version, e_inval);
938 if (version < 6) {
b9a67899
JP		 939 pr_warn("got v %d < 6 of %s ceph_osdmap\n",
940 version, prefix);
ec7af972
ID		 941 return -EINVAL;
942 }
943
944 /* old osdmap enconding */
945 struct_v = 0;
946 }
947
948 *v = struct_v;
949 return 0;
950
951e_inval:
952 return -EINVAL;
953}
954
433fbdd3
ID		 955static int __decode_pools(void **p, void *end, struct ceph_osdmap *map,
956 bool incremental)
957{
958 u32 n;
959
960 ceph_decode_32_safe(p, end, n, e_inval);
961 while (n--) {
962 struct ceph_pg_pool_info *pi;
963 u64 pool;
964 int ret;
965
966 ceph_decode_64_safe(p, end, pool, e_inval);
967
968 pi = __lookup_pg_pool(&map->pg_pools, pool);
969 if (!incremental || !pi) {
970 pi = kzalloc(sizeof(*pi), GFP_NOFS);
971 if (!pi)
972 return -ENOMEM;
973
974 pi->id = pool;
975
976 ret = __insert_pg_pool(&map->pg_pools, pi);
977 if (ret) {
978 kfree(pi);
979 return ret;
980 }
981 }
982
983 ret = decode_pool(p, end, pi);
984 if (ret)
985 return ret;
986 }
987
988 return 0;
989
990e_inval:
991 return -EINVAL;
992}
993
994static int decode_pools(void **p, void *end, struct ceph_osdmap *map)
995{
996 return __decode_pools(p, end, map, false);
997}
998
999static int decode_new_pools(void **p, void *end, struct ceph_osdmap *map)
1000{
1001 return __decode_pools(p, end, map, true);
1002}
1003
10db634e
ID		 1004static int __decode_pg_temp(void **p, void *end, struct ceph_osdmap *map,
1005 bool incremental)
1006{
1007 u32 n;
1008
1009 ceph_decode_32_safe(p, end, n, e_inval);
1010 while (n--) {
1011 struct ceph_pg pgid;
1012 u32 len, i;
1013 int ret;
1014
1015 ret = ceph_decode_pgid(p, end, &pgid);
1016 if (ret)
1017 return ret;
1018
1019 ceph_decode_32_safe(p, end, len, e_inval);
1020
1021 ret = __remove_pg_mapping(&map->pg_temp, pgid);
1022 BUG_ON(!incremental && ret != -ENOENT);
1023
1024 if (!incremental || len > 0) {
1025 struct ceph_pg_mapping *pg;
1026
1027 ceph_decode_need(p, end, len*sizeof(u32), e_inval);
1028
1029 if (len > (UINT_MAX - sizeof(*pg)) / sizeof(u32))
1030 return -EINVAL;
1031
1032 pg = kzalloc(sizeof(*pg) + len*sizeof(u32), GFP_NOFS);
1033 if (!pg)
1034 return -ENOMEM;
1035
1036 pg->pgid = pgid;
35a935d7 1037 pg->pg_temp.len = len;
10db634e 1038 for (i = 0; i < len; i++)
35a935d7 1039 pg->pg_temp.osds[i] = ceph_decode_32(p);
10db634e
ID		 1040
1041 ret = __insert_pg_mapping(pg, &map->pg_temp);
1042 if (ret) {
1043 kfree(pg);
1044 return ret;
1045 }
1046 }
1047 }
1048
1049 return 0;
1050
1051e_inval:
1052 return -EINVAL;
1053}
1054
1055static int decode_pg_temp(void **p, void *end, struct ceph_osdmap *map)
1056{
1057 return __decode_pg_temp(p, end, map, false);
1058}
1059
1060static int decode_new_pg_temp(void **p, void *end, struct ceph_osdmap *map)
1061{
1062 return __decode_pg_temp(p, end, map, true);
1063}
1064
d286de79
ID		 1065static int __decode_primary_temp(void **p, void *end, struct ceph_osdmap *map,
1066 bool incremental)
1067{
1068 u32 n;
1069
1070 ceph_decode_32_safe(p, end, n, e_inval);
1071 while (n--) {
1072 struct ceph_pg pgid;
1073 u32 osd;
1074 int ret;
1075
1076 ret = ceph_decode_pgid(p, end, &pgid);
1077 if (ret)
1078 return ret;
1079
1080 ceph_decode_32_safe(p, end, osd, e_inval);
1081
1082 ret = __remove_pg_mapping(&map->primary_temp, pgid);
1083 BUG_ON(!incremental && ret != -ENOENT);
1084
1085 if (!incremental || osd != (u32)-1) {
1086 struct ceph_pg_mapping *pg;
1087
1088 pg = kzalloc(sizeof(*pg), GFP_NOFS);
1089 if (!pg)
1090 return -ENOMEM;
1091
1092 pg->pgid = pgid;
1093 pg->primary_temp.osd = osd;
1094
1095 ret = __insert_pg_mapping(pg, &map->primary_temp);
1096 if (ret) {
1097 kfree(pg);
1098 return ret;
1099 }
1100 }
1101 }
1102
1103 return 0;
1104
1105e_inval:
1106 return -EINVAL;
1107}
1108
1109static int decode_primary_temp(void **p, void *end, struct ceph_osdmap *map)
1110{
1111 return __decode_primary_temp(p, end, map, false);
1112}
1113
1114static int decode_new_primary_temp(void **p, void *end,
1115 struct ceph_osdmap *map)
1116{
1117 return __decode_primary_temp(p, end, map, true);
1118}
1119
2cfa34f2
ID		 1120u32 ceph_get_primary_affinity(struct ceph_osdmap *map, int osd)
1121{
1122 BUG_ON(osd >= map->max_osd);
1123
1124 if (!map->osd_primary_affinity)
1125 return CEPH_OSD_DEFAULT_PRIMARY_AFFINITY;
1126
1127 return map->osd_primary_affinity[osd];
1128}
1129
1130static int set_primary_affinity(struct ceph_osdmap *map, int osd, u32 aff)
1131{
1132 BUG_ON(osd >= map->max_osd);
1133
1134 if (!map->osd_primary_affinity) {
1135 int i;
1136
1137 map->osd_primary_affinity = kmalloc(map->max_osd*sizeof(u32),
1138 GFP_NOFS);
1139 if (!map->osd_primary_affinity)
1140 return -ENOMEM;
1141
1142 for (i = 0; i < map->max_osd; i++)
1143 map->osd_primary_affinity[i] =
1144 CEPH_OSD_DEFAULT_PRIMARY_AFFINITY;
1145 }
1146
1147 map->osd_primary_affinity[osd] = aff;
1148
1149 return 0;
1150}
1151
63a6993f
ID		 1152static int decode_primary_affinity(void **p, void *end,
1153 struct ceph_osdmap *map)
1154{
1155 u32 len, i;
1156
1157 ceph_decode_32_safe(p, end, len, e_inval);
1158 if (len == 0) {
1159 kfree(map->osd_primary_affinity);
1160 map->osd_primary_affinity = NULL;
1161 return 0;
1162 }
1163 if (len != map->max_osd)
1164 goto e_inval;
1165
1166 ceph_decode_need(p, end, map->max_osd*sizeof(u32), e_inval);
1167
1168 for (i = 0; i < map->max_osd; i++) {
1169 int ret;
1170
1171 ret = set_primary_affinity(map, i, ceph_decode_32(p));
1172 if (ret)
1173 return ret;
1174 }
1175
1176 return 0;
1177
1178e_inval:
1179 return -EINVAL;
1180}
1181
1182static int decode_new_primary_affinity(void **p, void *end,
1183 struct ceph_osdmap *map)
1184{
1185 u32 n;
1186
1187 ceph_decode_32_safe(p, end, n, e_inval);
1188 while (n--) {
1189 u32 osd, aff;
1190 int ret;
1191
1192 ceph_decode_32_safe(p, end, osd, e_inval);
1193 ceph_decode_32_safe(p, end, aff, e_inval);
1194
1195 ret = set_primary_affinity(map, osd, aff);
1196 if (ret)
1197 return ret;
f31da0f3
ID		 1198
1199 pr_info("osd%d primary-affinity 0x%x\n", osd, aff);
63a6993f
ID		 1200 }
1201
1202 return 0;
1203
1204e_inval:
1205 return -EINVAL;
1206}
1207
f24e9980
SW		 1208/*
1209 * decode a full map.
1210 */
a2505d63 1211static int osdmap_decode(void **p, void *end, struct ceph_osdmap *map)
f24e9980 1212{
ec7af972 1213 u8 struct_v;
38a8d560 1214 u32 epoch = 0;
f24e9980 1215 void *start = *p;
3977058c
ID		 1216 u32 max;
1217 u32 len, i;
597b52f6 1218 int err;
f24e9980 1219
a2505d63 1220 dout("%s %p to %p len %d\n", __func__, *p, end, (int)(end - *p));
f24e9980 1221
ec7af972
ID		 1222 err = get_osdmap_client_data_v(p, end, "full", &struct_v);
1223 if (err)
1224 goto bad;
f24e9980 1225
53bbaba9
ID		 1226 /* fsid, epoch, created, modified */
1227 ceph_decode_need(p, end, sizeof(map->fsid) + sizeof(u32) +
1228 sizeof(map->created) + sizeof(map->modified), e_inval);
f24e9980 1229 ceph_decode_copy(p, &map->fsid, sizeof(map->fsid));
38a8d560 1230 epoch = map->epoch = ceph_decode_32(p);
f24e9980
SW		 1231 ceph_decode_copy(p, &map->created, sizeof(map->created));
1232 ceph_decode_copy(p, &map->modified, sizeof(map->modified));
1233
433fbdd3
ID		 1234 /* pools */
1235 err = decode_pools(p, end, map);
1236 if (err)
1237 goto bad;
2844a76a 1238
0f70c7ee
ID		 1239 /* pool_name */
1240 err = decode_pool_names(p, end, map);
597b52f6 1241 if (err)
4f6a7e5e 1242 goto bad;
2844a76a 1243
597b52f6 1244 ceph_decode_32_safe(p, end, map->pool_max, e_inval);
f24e9980 1245
597b52f6 1246 ceph_decode_32_safe(p, end, map->flags, e_inval);
f24e9980 1247
3977058c
ID		 1248 /* max_osd */
1249 ceph_decode_32_safe(p, end, max, e_inval);
f24e9980
SW		 1250
1251 /* (re)alloc osd arrays */
1252 err = osdmap_set_max_osd(map, max);
597b52f6 1253 if (err)
f24e9980 1254 goto bad;
f24e9980 1255
2d88b2e0 1256 /* osd_state, osd_weight, osd_addrs->client_addr */
f24e9980
SW		 1257 ceph_decode_need(p, end, 3*sizeof(u32) +
1258 map->max_osd*(1 + sizeof(*map->osd_weight) +
597b52f6
ID		 1259 sizeof(*map->osd_addr)), e_inval);
1260
2d88b2e0
ID		 1261 if (ceph_decode_32(p) != map->max_osd)
1262 goto e_inval;
1263
f24e9980
SW		 1264 ceph_decode_copy(p, map->osd_state, map->max_osd);
1265
2d88b2e0
ID		 1266 if (ceph_decode_32(p) != map->max_osd)
1267 goto e_inval;
1268
f24e9980 1269 for (i = 0; i < map->max_osd; i++)
c89136ea 1270 map->osd_weight[i] = ceph_decode_32(p);
f24e9980 1271
2d88b2e0
ID		 1272 if (ceph_decode_32(p) != map->max_osd)
1273 goto e_inval;
1274
f24e9980 1275 ceph_decode_copy(p, map->osd_addr, map->max_osd*sizeof(*map->osd_addr));
63f2d211
SW		 1276 for (i = 0; i < map->max_osd; i++)
1277 ceph_decode_addr(&map->osd_addr[i]);
f24e9980
SW		 1278
1279 /* pg_temp */
10db634e
ID		 1280 err = decode_pg_temp(p, end, map);
1281 if (err)
1282 goto bad;
f24e9980 1283
d286de79
ID		 1284 /* primary_temp */
1285 if (struct_v >= 1) {
1286 err = decode_primary_temp(p, end, map);
1287 if (err)
1288 goto bad;
1289 }
1290
63a6993f
ID		 1291 /* primary_affinity */
1292 if (struct_v >= 2) {
1293 err = decode_primary_affinity(p, end, map);
1294 if (err)
1295 goto bad;
1296 } else {
1297 /* XXX can this happen? */
1298 kfree(map->osd_primary_affinity);
1299 map->osd_primary_affinity = NULL;
1300 }
1301
f24e9980 1302 /* crush */
597b52f6 1303 ceph_decode_32_safe(p, end, len, e_inval);
1b6a78b5
ID		 1304 err = osdmap_set_crush(map, crush_decode(*p, min(*p + len, end)));
1305 if (err)
f24e9980 1306 goto bad;
f24e9980 1307
38a8d560 1308 /* ignore the rest */
f24e9980
SW		 1309 *p = end;
1310
38a8d560 1311 dout("full osdmap epoch %d max_osd %d\n", map->epoch, map->max_osd);
a2505d63 1312 return 0;
f24e9980 1313
597b52f6
ID		 1314e_inval:
1315 err = -EINVAL;
f24e9980 1316bad:
38a8d560
ID		 1317 pr_err("corrupt full osdmap (%d) epoch %d off %d (%p of %p-%p)\n",
1318 err, epoch, (int)(*p - start), *p, start, end);
1319 print_hex_dump(KERN_DEBUG, "osdmap: ",
1320 DUMP_PREFIX_OFFSET, 16, 1,
1321 start, end - start, true);
a2505d63
ID		 1322 return err;
1323}
1324
1325/*
1326 * Allocate and decode a full map.
1327 */
1328struct ceph_osdmap *ceph_osdmap_decode(void **p, void *end)
1329{
1330 struct ceph_osdmap *map;
1331 int ret;
1332
e5253a7b 1333 map = ceph_osdmap_alloc();
a2505d63
ID		 1334 if (!map)
1335 return ERR_PTR(-ENOMEM);
1336
a2505d63
ID		 1337 ret = osdmap_decode(p, end, map);
1338 if (ret) {
1339 ceph_osdmap_destroy(map);
1340 return ERR_PTR(ret);
1341 }
1342
1343 return map;
f24e9980
SW		 1344}
1345
930c5328
ID		 1346/*
1347 * Encoding order is (new_up_client, new_state, new_weight). Need to
1348 * apply in the (new_weight, new_state, new_up_client) order, because
1349 * an incremental map may look like e.g.
1350 *
1351 * new_up_client: { osd=6, addr=... } # set osd_state and addr
1352 * new_state: { osd=6, xorstate=EXISTS } # clear osd_state
1353 */
1354static int decode_new_up_state_weight(void **p, void *end,
1355 struct ceph_osdmap *map)
1356{
1357 void *new_up_client;
1358 void *new_state;
1359 void *new_weight_end;
1360 u32 len;
1361
1362 new_up_client = *p;
1363 ceph_decode_32_safe(p, end, len, e_inval);
1364 len *= sizeof(u32) + sizeof(struct ceph_entity_addr);
1365 ceph_decode_need(p, end, len, e_inval);
1366 *p += len;
1367
1368 new_state = *p;
1369 ceph_decode_32_safe(p, end, len, e_inval);
1370 len *= sizeof(u32) + sizeof(u8);
1371 ceph_decode_need(p, end, len, e_inval);
1372 *p += len;
1373
1374 /* new_weight */
1375 ceph_decode_32_safe(p, end, len, e_inval);
1376 while (len--) {
1377 s32 osd;
1378 u32 w;
1379
1380 ceph_decode_need(p, end, 2*sizeof(u32), e_inval);
1381 osd = ceph_decode_32(p);
1382 w = ceph_decode_32(p);
1383 BUG_ON(osd >= map->max_osd);
1384 pr_info("osd%d weight 0x%x %s\n", osd, w,
1385 w == CEPH_OSD_IN ? "(in)" :
1386 (w == CEPH_OSD_OUT ? "(out)" : ""));
1387 map->osd_weight[osd] = w;
1388
1389 /*
1390 * If we are marking in, set the EXISTS, and clear the
1391 * AUTOOUT and NEW bits.
1392 */
1393 if (w) {
1394 map->osd_state[osd] |= CEPH_OSD_EXISTS;
1395 map->osd_state[osd] &= ~(CEPH_OSD_AUTOOUT |
1396 CEPH_OSD_NEW);
1397 }
1398 }
1399 new_weight_end = *p;
1400
1401 /* new_state (up/down) */
1402 *p = new_state;
1403 len = ceph_decode_32(p);
1404 while (len--) {
1405 s32 osd;
1406 u8 xorstate;
1407 int ret;
1408
1409 osd = ceph_decode_32(p);
1410 xorstate = ceph_decode_8(p);
1411 if (xorstate == 0)
1412 xorstate = CEPH_OSD_UP;
1413 BUG_ON(osd >= map->max_osd);
1414 if ((map->osd_state[osd] & CEPH_OSD_UP) &&
1415 (xorstate & CEPH_OSD_UP))
1416 pr_info("osd%d down\n", osd);
1417 if ((map->osd_state[osd] & CEPH_OSD_EXISTS) &&
1418 (xorstate & CEPH_OSD_EXISTS)) {
1419 pr_info("osd%d does not exist\n", osd);
930c5328
ID		 1420 ret = set_primary_affinity(map, osd,
1421 CEPH_OSD_DEFAULT_PRIMARY_AFFINITY);
1422 if (ret)
1423 return ret;
1424 memset(map->osd_addr + osd, 0, sizeof(*map->osd_addr));
1425 map->osd_state[osd] = 0;
1426 } else {
1427 map->osd_state[osd] ^= xorstate;
1428 }
1429 }
1430
1431 /* new_up_client */
1432 *p = new_up_client;
1433 len = ceph_decode_32(p);
1434 while (len--) {
1435 s32 osd;
1436 struct ceph_entity_addr addr;
1437
1438 osd = ceph_decode_32(p);
1439 ceph_decode_copy(p, &addr, sizeof(addr));
1440 ceph_decode_addr(&addr);
1441 BUG_ON(osd >= map->max_osd);
1442 pr_info("osd%d up\n", osd);
1443 map->osd_state[osd] |= CEPH_OSD_EXISTS | CEPH_OSD_UP;
1444 map->osd_addr[osd] = addr;
1445 }
1446
1447 *p = new_weight_end;
1448 return 0;
1449
1450e_inval:
1451 return -EINVAL;
1452}
1453
f24e9980
SW		 1454/*
1455 * decode and apply an incremental map update.
1456 */
1457struct ceph_osdmap *osdmap_apply_incremental(void **p, void *end,
0c0a8de1 1458 struct ceph_osdmap *map)
f24e9980 1459{
f24e9980
SW		 1460 struct ceph_fsid fsid;
1461 u32 epoch = 0;
1462 struct ceph_timespec modified;
4f6a7e5e
SW		 1463 s32 len;
1464 u64 pool;
1465 __s64 new_pool_max;
1466 __s32 new_flags, max;
f24e9980 1467 void *start = *p;
86f1742b 1468 int err;
ec7af972 1469 u8 struct_v;
f24e9980 1470
38a8d560
ID		 1471 dout("%s %p to %p len %d\n", __func__, *p, end, (int)(end - *p));
1472
ec7af972
ID		 1473 err = get_osdmap_client_data_v(p, end, "inc", &struct_v);
1474 if (err)
1475 goto bad;
f24e9980 1476
53bbaba9
ID		 1477 /* fsid, epoch, modified, new_pool_max, new_flags */
1478 ceph_decode_need(p, end, sizeof(fsid) + sizeof(u32) + sizeof(modified) +
1479 sizeof(u64) + sizeof(u32), e_inval);
f24e9980 1480 ceph_decode_copy(p, &fsid, sizeof(fsid));
c89136ea 1481 epoch = ceph_decode_32(p);
f24e9980
SW		 1482 BUG_ON(epoch != map->epoch+1);
1483 ceph_decode_copy(p, &modified, sizeof(modified));
4f6a7e5e 1484 new_pool_max = ceph_decode_64(p);
c89136ea 1485 new_flags = ceph_decode_32(p);
f24e9980
SW		 1486
1487 /* full map? */
86f1742b 1488 ceph_decode_32_safe(p, end, len, e_inval);
f24e9980
SW		 1489 if (len > 0) {
1490 dout("apply_incremental full map len %d, %p to %p\n",
1491 len, *p, end);
a2505d63 1492 return ceph_osdmap_decode(p, min(*p+len, end));
f24e9980
SW		 1493 }
1494
1495 /* new crush? */
86f1742b 1496 ceph_decode_32_safe(p, end, len, e_inval);
f24e9980 1497 if (len > 0) {
1b6a78b5
ID		 1498 err = osdmap_set_crush(map,
1499 crush_decode(*p, min(*p + len, end)));
1500 if (err)
86f1742b 1501 goto bad;
cebc5be6 1502 *p += len;
f24e9980
SW		 1503 }
1504
1505 /* new flags? */
1506 if (new_flags >= 0)
1507 map->flags = new_flags;
4fc51be8
SW		 1508 if (new_pool_max >= 0)
1509 map->pool_max = new_pool_max;
f24e9980 1510
f24e9980 1511 /* new max? */
53bbaba9 1512 ceph_decode_32_safe(p, end, max, e_inval);
f24e9980
SW		 1513 if (max >= 0) {
1514 err = osdmap_set_max_osd(map, max);
86f1742b 1515 if (err)
f24e9980
SW		 1516 goto bad;
1517 }
1518
1519 map->epoch++;
31456665 1520 map->modified = modified;
f24e9980 1521
433fbdd3
ID		 1522 /* new_pools */
1523 err = decode_new_pools(p, end, map);
1524 if (err)
1525 goto bad;
9464d008 1526
0f70c7ee
ID		 1527 /* new_pool_names */
1528 err = decode_pool_names(p, end, map);
9464d008
ID		 1529 if (err)
1530 goto bad;
f24e9980 1531
4fc51be8 1532 /* old_pool */
86f1742b 1533 ceph_decode_32_safe(p, end, len, e_inval);
4fc51be8
SW		 1534 while (len--) {
1535 struct ceph_pg_pool_info *pi;
1536
86f1742b 1537 ceph_decode_64_safe(p, end, pool, e_inval);
4fc51be8 1538 pi = __lookup_pg_pool(&map->pg_pools, pool);
2844a76a
SW		 1539 if (pi)
1540 __remove_pg_pool(&map->pg_pools, pi);
4fc51be8 1541 }
f24e9980 1542
930c5328
ID		 1543 /* new_up_client, new_state, new_weight */
1544 err = decode_new_up_state_weight(p, end, map);
1545 if (err)
1546 goto bad;
f24e9980
SW		 1547
1548 /* new_pg_temp */
10db634e
ID		 1549 err = decode_new_pg_temp(p, end, map);
1550 if (err)
1551 goto bad;
f24e9980 1552
d286de79
ID		 1553 /* new_primary_temp */
1554 if (struct_v >= 1) {
1555 err = decode_new_primary_temp(p, end, map);
1556 if (err)
1557 goto bad;
1558 }
1559
63a6993f
ID		 1560 /* new_primary_affinity */
1561 if (struct_v >= 2) {
1562 err = decode_new_primary_affinity(p, end, map);
1563 if (err)
1564 goto bad;
1565 }
1566
f24e9980
SW		 1567 /* ignore the rest */
1568 *p = end;
38a8d560
ID		 1569
1570 dout("inc osdmap epoch %d max_osd %d\n", map->epoch, map->max_osd);
f24e9980
SW		 1571 return map;
1572
86f1742b
ID		 1573e_inval:
1574 err = -EINVAL;
f24e9980 1575bad:
38a8d560
ID		 1576 pr_err("corrupt inc osdmap (%d) epoch %d off %d (%p of %p-%p)\n",
1577 err, epoch, (int)(*p - start), *p, start, end);
9ec7cab1
SW		 1578 print_hex_dump(KERN_DEBUG, "osdmap: ",
1579 DUMP_PREFIX_OFFSET, 16, 1,
1580 start, end - start, true);
f24e9980
SW		 1581 return ERR_PTR(err);
1582}
1583
30c156d9
YZ		 1584void ceph_oloc_copy(struct ceph_object_locator *dest,
1585 const struct ceph_object_locator *src)
1586{
ca35ffea 1587 ceph_oloc_destroy(dest);
30c156d9
YZ		 1588
1589 dest->pool = src->pool;
1590 if (src->pool_ns)
1591 dest->pool_ns = ceph_get_string(src->pool_ns);
ca35ffea
ID		 1592 else
1593 dest->pool_ns = NULL;
30c156d9
YZ		 1594}
1595EXPORT_SYMBOL(ceph_oloc_copy);
1596
1597void ceph_oloc_destroy(struct ceph_object_locator *oloc)
1598{
1599 ceph_put_string(oloc->pool_ns);
1600}
1601EXPORT_SYMBOL(ceph_oloc_destroy);
1602
d30291b9
ID		 1603void ceph_oid_copy(struct ceph_object_id *dest,
1604 const struct ceph_object_id *src)
1605{
ca35ffea 1606 ceph_oid_destroy(dest);
d30291b9
ID		 1607
1608 if (src->name != src->inline_name) {
1609 /* very rare, see ceph_object_id definition */
1610 dest->name = kmalloc(src->name_len + 1,
1611 GFP_NOIO | __GFP_NOFAIL);
ca35ffea
ID		 1612 } else {
1613 dest->name = dest->inline_name;
d30291b9 1614 }
d30291b9
ID		 1615 memcpy(dest->name, src->name, src->name_len + 1);
1616 dest->name_len = src->name_len;
1617}
1618EXPORT_SYMBOL(ceph_oid_copy);
1619
1620static __printf(2, 0)
1621int oid_printf_vargs(struct ceph_object_id *oid, const char *fmt, va_list ap)
1622{
1623 int len;
1624
1625 WARN_ON(!ceph_oid_empty(oid));
1626
1627 len = vsnprintf(oid->inline_name, sizeof(oid->inline_name), fmt, ap);
1628 if (len >= sizeof(oid->inline_name))
1629 return len;
1630
1631 oid->name_len = len;
1632 return 0;
1633}
1634
1635/*
1636 * If oid doesn't fit into inline buffer, BUG.
1637 */
1638void ceph_oid_printf(struct ceph_object_id *oid, const char *fmt, ...)
1639{
1640 va_list ap;
1641
1642 va_start(ap, fmt);
1643 BUG_ON(oid_printf_vargs(oid, fmt, ap));
1644 va_end(ap);
1645}
1646EXPORT_SYMBOL(ceph_oid_printf);
1647
1648static __printf(3, 0)
1649int oid_aprintf_vargs(struct ceph_object_id *oid, gfp_t gfp,
1650 const char *fmt, va_list ap)
1651{
1652 va_list aq;
1653 int len;
1654
1655 va_copy(aq, ap);
1656 len = oid_printf_vargs(oid, fmt, aq);
1657 va_end(aq);
1658
1659 if (len) {
1660 char *external_name;
1661
1662 external_name = kmalloc(len + 1, gfp);
1663 if (!external_name)
1664 return -ENOMEM;
f24e9980 1665
d30291b9
ID		 1666 oid->name = external_name;
1667 WARN_ON(vsnprintf(oid->name, len + 1, fmt, ap) != len);
1668 oid->name_len = len;
1669 }
f24e9980 1670
d30291b9
ID		 1671 return 0;
1672}
1673
1674/*
1675 * If oid doesn't fit into inline buffer, allocate.
1676 */
1677int ceph_oid_aprintf(struct ceph_object_id *oid, gfp_t gfp,
1678 const char *fmt, ...)
1679{
1680 va_list ap;
1681 int ret;
1682
1683 va_start(ap, fmt);
1684 ret = oid_aprintf_vargs(oid, gfp, fmt, ap);
1685 va_end(ap);
1686
1687 return ret;
1688}
1689EXPORT_SYMBOL(ceph_oid_aprintf);
1690
1691void ceph_oid_destroy(struct ceph_object_id *oid)
1692{
1693 if (oid->name != oid->inline_name)
1694 kfree(oid->name);
1695}
1696EXPORT_SYMBOL(ceph_oid_destroy);
f24e9980 1697
63244fa1
ID		 1698/*
1699 * osds only
1700 */
1701static bool __osds_equal(const struct ceph_osds *lhs,
1702 const struct ceph_osds *rhs)
1703{
1704 if (lhs->size == rhs->size &&
1705 !memcmp(lhs->osds, rhs->osds, rhs->size * sizeof(rhs->osds[0])))
1706 return true;
1707
1708 return false;
1709}
1710
1711/*
1712 * osds + primary
1713 */
1714static bool osds_equal(const struct ceph_osds *lhs,
1715 const struct ceph_osds *rhs)
1716{
1717 if (__osds_equal(lhs, rhs) &&
1718 lhs->primary == rhs->primary)
1719 return true;
1720
1721 return false;
1722}
1723
6f3bfd45
ID		 1724static bool osds_valid(const struct ceph_osds *set)
1725{
1726 /* non-empty set */
1727 if (set->size > 0 && set->primary >= 0)
1728 return true;
1729
1730 /* empty can_shift_osds set */
1731 if (!set->size && set->primary == -1)
1732 return true;
1733
1734 /* empty !can_shift_osds set - all NONE */
1735 if (set->size > 0 && set->primary == -1) {
1736 int i;
1737
1738 for (i = 0; i < set->size; i++) {
1739 if (set->osds[i] != CRUSH_ITEM_NONE)
1740 break;
1741 }
1742 if (i == set->size)
1743 return true;
1744 }
1745
1746 return false;
1747}
1748
1749void ceph_osds_copy(struct ceph_osds *dest, const struct ceph_osds *src)
1750{
1751 memcpy(dest->osds, src->osds, src->size * sizeof(src->osds[0]));
1752 dest->size = src->size;
1753 dest->primary = src->primary;
1754}
1755
7de030d6
ID		 1756bool ceph_pg_is_split(const struct ceph_pg *pgid, u32 old_pg_num,
1757 u32 new_pg_num)
63244fa1
ID		 1758{
1759 int old_bits = calc_bits_of(old_pg_num);
1760 int old_mask = (1 << old_bits) - 1;
1761 int n;
1762
1763 WARN_ON(pgid->seed >= old_pg_num);
1764 if (new_pg_num <= old_pg_num)
1765 return false;
1766
1767 for (n = 1; ; n++) {
1768 int next_bit = n << (old_bits - 1);
1769 u32 s = next_bit | pgid->seed;
1770
1771 if (s < old_pg_num || s == pgid->seed)
1772 continue;
1773 if (s >= new_pg_num)
1774 break;
1775
1776 s = ceph_stable_mod(s, old_pg_num, old_mask);
1777 if (s == pgid->seed)
1778 return true;
1779 }
1780
1781 return false;
1782}
1783
1784bool ceph_is_new_interval(const struct ceph_osds *old_acting,
1785 const struct ceph_osds *new_acting,
1786 const struct ceph_osds *old_up,
1787 const struct ceph_osds *new_up,
1788 int old_size,
1789 int new_size,
1790 int old_min_size,
1791 int new_min_size,
1792 u32 old_pg_num,
1793 u32 new_pg_num,
1794 bool old_sort_bitwise,
1795 bool new_sort_bitwise,
1796 const struct ceph_pg *pgid)
1797{
1798 return !osds_equal(old_acting, new_acting) ||
1799 !osds_equal(old_up, new_up) ||
1800 old_size != new_size ||
1801 old_min_size != new_min_size ||
7de030d6 1802 ceph_pg_is_split(pgid, old_pg_num, new_pg_num) ||
63244fa1
ID		 1803 old_sort_bitwise != new_sort_bitwise;
1804}
1805
1806static int calc_pg_rank(int osd, const struct ceph_osds *acting)
1807{
1808 int i;
1809
1810 for (i = 0; i < acting->size; i++) {
1811 if (acting->osds[i] == osd)
1812 return i;
1813 }
1814
1815 return -1;
1816}
1817
1818static bool primary_changed(const struct ceph_osds *old_acting,
1819 const struct ceph_osds *new_acting)
1820{
1821 if (!old_acting->size && !new_acting->size)
1822 return false; /* both still empty */
1823
1824 if (!old_acting->size ^ !new_acting->size)
1825 return true; /* was empty, now not, or vice versa */
1826
1827 if (old_acting->primary != new_acting->primary)
1828 return true; /* primary changed */
1829
1830 if (calc_pg_rank(old_acting->primary, old_acting) !=
1831 calc_pg_rank(new_acting->primary, new_acting))
1832 return true;
1833
1834 return false; /* same primary (tho replicas may have changed) */
1835}
1836
1837bool ceph_osds_changed(const struct ceph_osds *old_acting,
1838 const struct ceph_osds *new_acting,
1839 bool any_change)
1840{
1841 if (primary_changed(old_acting, new_acting))
1842 return true;
1843
1844 if (any_change && !__osds_equal(old_acting, new_acting))
1845 return true;
1846
1847 return false;
1848}
1849
f24e9980
SW		 1850/*
1851 * calculate file layout from given offset, length.
1852 * fill in correct oid, logical length, and object extent
1853 * offset, length.
1854 *
1855 * for now, we write only a single su, until we can
1856 * pass a stride back to the caller.
1857 */
d63b77f4 1858int ceph_calc_file_object_mapping(struct ceph_file_layout *layout,
e8afad65 1859 u64 off, u64 len,
645a1025 1860 u64 *ono,
f24e9980
SW		 1861 u64 *oxoff, u64 *oxlen)
1862{
7627151e
YZ		 1863 u32 osize = layout->object_size;
1864 u32 su = layout->stripe_unit;
1865 u32 sc = layout->stripe_count;
f24e9980
SW		 1866 u32 bl, stripeno, stripepos, objsetno;
1867 u32 su_per_object;
ff1d1f71 1868 u64 t, su_offset;
f24e9980 1869
e8afad65 1870 dout("mapping %llu~%llu osize %u fl_su %u\n", off, len,
f24e9980 1871 osize, su);
d63b77f4
SW		 1872 if (su == 0 || sc == 0)
1873 goto invalid;
35e054a6 1874 su_per_object = osize / su;
d63b77f4
SW		 1875 if (su_per_object == 0)
1876 goto invalid;
f24e9980
SW		 1877 dout("osize %u / su %u = su_per_object %u\n", osize, su,
1878 su_per_object);
1879
d63b77f4
SW		 1880 if ((su & ~PAGE_MASK) != 0)
1881 goto invalid;
1882
f24e9980
SW		 1883 /* bl = *off / su; */
1884 t = off;
1885 do_div(t, su);
1886 bl = t;
1887 dout("off %llu / su %u = bl %u\n", off, su, bl);
1888
1889 stripeno = bl / sc;
1890 stripepos = bl % sc;
1891 objsetno = stripeno / su_per_object;
1892
645a1025 1893 *ono = objsetno * sc + stripepos;
95c96174 1894 dout("objset %u * sc %u = ono %u\n", objsetno, sc, (unsigned int)*ono);
645a1025
SW		 1895
1896 /* *oxoff = *off % layout->fl_stripe_unit; # offset in su */
f24e9980 1897 t = off;
ff1d1f71
NW		 1898 su_offset = do_div(t, su);
1899 *oxoff = su_offset + (stripeno % su_per_object) * su;
1900
1901 /*
1902 * Calculate the length of the extent being written to the selected
e8afad65 1903 * object. This is the minimum of the full length requested (len) or
ff1d1f71
NW		 1904 * the remainder of the current stripe being written to.
1905 */
e8afad65 1906 *oxlen = min_t(u64, len, su - su_offset);
f24e9980
SW		 1907
1908 dout(" obj extent %llu~%llu\n", *oxoff, *oxlen);
d63b77f4
SW		 1909 return 0;
1910
1911invalid:
1912 dout(" invalid layout\n");
1913 *ono = 0;
1914 *oxoff = 0;
1915 *oxlen = 0;
1916 return -EINVAL;
f24e9980 1917}
3d14c5d2 1918EXPORT_SYMBOL(ceph_calc_file_object_mapping);
f24e9980
SW		 1919
1920/*
d9591f5e
ID		 1921 * Map an object into a PG.
1922 *
1923 * Should only be called with target_oid and target_oloc (as opposed to
1924 * base_oid and base_oloc), since tiering isn't taken into account.
f24e9980 1925 */
d9591f5e
ID		 1926int ceph_object_locator_to_pg(struct ceph_osdmap *osdmap,
1927 struct ceph_object_id *oid,
1928 struct ceph_object_locator *oloc,
1929 struct ceph_pg *raw_pgid)
f24e9980 1930{
7c13cb64 1931 struct ceph_pg_pool_info *pi;
f24e9980 1932
d9591f5e 1933 pi = ceph_pg_pool_by_id(osdmap, oloc->pool);
7c13cb64 1934 if (!pi)
d9591f5e 1935 return -ENOENT;
f24e9980 1936
30c156d9
YZ		 1937 if (!oloc->pool_ns) {
1938 raw_pgid->pool = oloc->pool;
1939 raw_pgid->seed = ceph_str_hash(pi->object_hash, oid->name,
1940 oid->name_len);
1941 dout("%s %s -> raw_pgid %llu.%x\n", __func__, oid->name,
1942 raw_pgid->pool, raw_pgid->seed);
1943 } else {
1944 char stack_buf[256];
1945 char *buf = stack_buf;
1946 int nsl = oloc->pool_ns->len;
1947 size_t total = nsl + 1 + oid->name_len;
1948
1949 if (total > sizeof(stack_buf)) {
1950 buf = kmalloc(total, GFP_NOIO);
1951 if (!buf)
1952 return -ENOMEM;
1953 }
1954 memcpy(buf, oloc->pool_ns->str, nsl);
1955 buf[nsl] = '\037';
1956 memcpy(buf + nsl + 1, oid->name, oid->name_len);
1957 raw_pgid->pool = oloc->pool;
1958 raw_pgid->seed = ceph_str_hash(pi->object_hash, buf, total);
1959 if (buf != stack_buf)
1960 kfree(buf);
1961 dout("%s %s ns %.*s -> raw_pgid %llu.%x\n", __func__,
1962 oid->name, nsl, oloc->pool_ns->str,
1963 raw_pgid->pool, raw_pgid->seed);
1964 }
f24e9980
SW		 1965 return 0;
1966}
d9591f5e 1967EXPORT_SYMBOL(ceph_object_locator_to_pg);
f24e9980 1968
6f3bfd45
ID		 1969/*
1970 * Map a raw PG (full precision ps) into an actual PG.
1971 */
1972static void raw_pg_to_pg(struct ceph_pg_pool_info *pi,
1973 const struct ceph_pg *raw_pgid,
1974 struct ceph_pg *pgid)
1975{
1976 pgid->pool = raw_pgid->pool;
1977 pgid->seed = ceph_stable_mod(raw_pgid->seed, pi->pg_num,
1978 pi->pg_num_mask);
1979}
1980
1981/*
1982 * Map a raw PG (full precision ps) into a placement ps (placement
1983 * seed). Include pool id in that value so that different pools don't
1984 * use the same seeds.
1985 */
1986static u32 raw_pg_to_pps(struct ceph_pg_pool_info *pi,
1987 const struct ceph_pg *raw_pgid)
1988{
1989 if (pi->flags & CEPH_POOL_FLAG_HASHPSPOOL) {
1990 /* hash pool id and seed so that pool PGs do not overlap */
1991 return crush_hash32_2(CRUSH_HASH_RJENKINS1,
1992 ceph_stable_mod(raw_pgid->seed,
1993 pi->pgp_num,
1994 pi->pgp_num_mask),
1995 raw_pgid->pool);
1996 } else {
1997 /*
1998 * legacy behavior: add ps and pool together. this is
1999 * not a great approach because the PGs from each pool
2000 * will overlap on top of each other: 0.5 == 1.4 ==
2001 * 2.3 == ...
2002 */
2003 return ceph_stable_mod(raw_pgid->seed, pi->pgp_num,
2004 pi->pgp_num_mask) +
2005 (unsigned)raw_pgid->pool;
2006 }
2007}
2008
9d521470
ID		 2009static int do_crush(struct ceph_osdmap *map, int ruleno, int x,
2010 int *result, int result_max,
2011 const __u32 *weight, int weight_max)
e8ef19c4 2012{
9d521470
ID		 2013 int r;
2014
2015 BUG_ON(result_max > CEPH_PG_MAX_SIZE);
2016
743efcff 2017 mutex_lock(&map->crush_workspace_mutex);
9d521470 2018 r = crush_do_rule(map->crush, ruleno, x, result, result_max,
743efcff
ID		 2019 weight, weight_max, map->crush_workspace);
2020 mutex_unlock(&map->crush_workspace_mutex);
e8ef19c4 2021
9d521470 2022 return r;
e8ef19c4
ID		 2023}
2024
2bd93d4d 2025/*
6f3bfd45
ID		 2026 * Calculate raw set (CRUSH output) for given PG. The result may
2027 * contain nonexistent OSDs. ->primary is undefined for a raw set.
2bd93d4d 2028 *
6f3bfd45 2029 * Placement seed (CRUSH input) is returned through @ppps.
2bd93d4d 2030 */
6f3bfd45
ID		 2031static void pg_to_raw_osds(struct ceph_osdmap *osdmap,
2032 struct ceph_pg_pool_info *pi,
2033 const struct ceph_pg *raw_pgid,
2034 struct ceph_osds *raw,
2035 u32 *ppps)
2bd93d4d 2036{
6f3bfd45 2037 u32 pps = raw_pg_to_pps(pi, raw_pgid);
2bd93d4d
ID		 2038 int ruleno;
2039 int len;
2040
6f3bfd45
ID		 2041 ceph_osds_init(raw);
2042 if (ppps)
2043 *ppps = pps;
2044
2045 ruleno = crush_find_rule(osdmap->crush, pi->crush_ruleset, pi->type,
2046 pi->size);
2bd93d4d
ID		 2047 if (ruleno < 0) {
2048 pr_err("no crush rule: pool %lld ruleset %d type %d size %d\n",
6f3bfd45
ID		 2049 pi->id, pi->crush_ruleset, pi->type, pi->size);
2050 return;
2bd93d4d
ID		 2051 }
2052
ef9324bb
ID		 2053 if (pi->size > ARRAY_SIZE(raw->osds)) {
2054 pr_err_ratelimited("pool %lld ruleset %d type %d too wide: size %d > %zu\n",
2055 pi->id, pi->crush_ruleset, pi->type, pi->size,
2056 ARRAY_SIZE(raw->osds));
2057 return;
2058 }
2059
2060 len = do_crush(osdmap, ruleno, pps, raw->osds, pi->size,
2bd93d4d
ID		 2061 osdmap->osd_weight, osdmap->max_osd);
2062 if (len < 0) {
2063 pr_err("error %d from crush rule %d: pool %lld ruleset %d type %d size %d\n",
6f3bfd45
ID		 2064 len, ruleno, pi->id, pi->crush_ruleset, pi->type,
2065 pi->size);
2066 return;
2bd93d4d
ID		 2067 }
2068
6f3bfd45 2069 raw->size = len;
2bd93d4d
ID		 2070}
2071
2072/*
6f3bfd45
ID		 2073 * Given raw set, calculate up set and up primary. By definition of an
2074 * up set, the result won't contain nonexistent or down OSDs.
2bd93d4d 2075 *
6f3bfd45
ID		 2076 * This is done in-place - on return @set is the up set. If it's
2077 * empty, ->primary will remain undefined.
2bd93d4d 2078 */
6f3bfd45
ID		 2079static void raw_to_up_osds(struct ceph_osdmap *osdmap,
2080 struct ceph_pg_pool_info *pi,
2081 struct ceph_osds *set)
2bd93d4d 2082{
2bd93d4d
ID		 2083 int i;
2084
6f3bfd45
ID		 2085 /* ->primary is undefined for a raw set */
2086 BUG_ON(set->primary != -1);
2087
2088 if (ceph_can_shift_osds(pi)) {
2bd93d4d
ID		 2089 int removed = 0;
2090
6f3bfd45
ID		 2091 /* shift left */
2092 for (i = 0; i < set->size; i++) {
2093 if (ceph_osd_is_down(osdmap, set->osds[i])) {
2bd93d4d
ID		 2094 removed++;
2095 continue;
2096 }
2097 if (removed)
6f3bfd45 2098 set->osds[i - removed] = set->osds[i];
2bd93d4d 2099 }
6f3bfd45
ID		 2100 set->size -= removed;
2101 if (set->size > 0)
2102 set->primary = set->osds[0];
2bd93d4d 2103 } else {
6f3bfd45
ID		 2104 /* set down/dne devices to NONE */
2105 for (i = set->size - 1; i >= 0; i--) {
2106 if (ceph_osd_is_down(osdmap, set->osds[i]))
2107 set->osds[i] = CRUSH_ITEM_NONE;
2bd93d4d 2108 else
6f3bfd45 2109 set->primary = set->osds[i];
2bd93d4d
ID		 2110 }
2111 }
2bd93d4d
ID		 2112}
2113
6f3bfd45
ID		 2114static void apply_primary_affinity(struct ceph_osdmap *osdmap,
2115 struct ceph_pg_pool_info *pi,
2116 u32 pps,
2117 struct ceph_osds *up)
47ec1f3c
ID		 2118{
2119 int i;
2120 int pos = -1;
2121
2122 /*
2123 * Do we have any non-default primary_affinity values for these
2124 * osds?
2125 */
2126 if (!osdmap->osd_primary_affinity)
2127 return;
2128
6f3bfd45
ID		 2129 for (i = 0; i < up->size; i++) {
2130 int osd = up->osds[i];
92b2e751
ID		 2131
2132 if (osd != CRUSH_ITEM_NONE &&
2133 osdmap->osd_primary_affinity[osd] !=
47ec1f3c
ID		 2134 CEPH_OSD_DEFAULT_PRIMARY_AFFINITY) {
2135 break;
2136 }
2137 }
6f3bfd45 2138 if (i == up->size)
47ec1f3c
ID		 2139 return;
2140
2141 /*
2142 * Pick the primary. Feed both the seed (for the pg) and the
2143 * osd into the hash/rng so that a proportional fraction of an
2144 * osd's pgs get rejected as primary.
2145 */
6f3bfd45
ID		 2146 for (i = 0; i < up->size; i++) {
2147 int osd = up->osds[i];
47ec1f3c
ID		 2148 u32 aff;
2149
47ec1f3c
ID		 2150 if (osd == CRUSH_ITEM_NONE)
2151 continue;
2152
2153 aff = osdmap->osd_primary_affinity[osd];
2154 if (aff < CEPH_OSD_MAX_PRIMARY_AFFINITY &&
2155 (crush_hash32_2(CRUSH_HASH_RJENKINS1,
2156 pps, osd) >> 16) >= aff) {
2157 /*
2158 * We chose not to use this primary. Note it
2159 * anyway as a fallback in case we don't pick
2160 * anyone else, but keep looking.
2161 */
2162 if (pos < 0)
2163 pos = i;
2164 } else {
2165 pos = i;
2166 break;
2167 }
2168 }
2169 if (pos < 0)
2170 return;
2171
6f3bfd45 2172 up->primary = up->osds[pos];
47ec1f3c 2173
6f3bfd45 2174 if (ceph_can_shift_osds(pi) && pos > 0) {
47ec1f3c
ID		 2175 /* move the new primary to the front */
2176 for (i = pos; i > 0; i--)
6f3bfd45
ID		 2177 up->osds[i] = up->osds[i - 1];
2178 up->osds[0] = up->primary;
47ec1f3c
ID		 2179 }
2180}
2181
45966c34 2182/*
6f3bfd45 2183 * Get pg_temp and primary_temp mappings for given PG.
45966c34 2184 *
6f3bfd45
ID		 2185 * Note that a PG may have none, only pg_temp, only primary_temp or
2186 * both pg_temp and primary_temp mappings. This means @temp isn't
2187 * always a valid OSD set on return: in the "only primary_temp" case,
2188 * @temp will have its ->primary >= 0 but ->size == 0.
45966c34 2189 */
6f3bfd45
ID		 2190static void get_temp_osds(struct ceph_osdmap *osdmap,
2191 struct ceph_pg_pool_info *pi,
2192 const struct ceph_pg *raw_pgid,
2193 struct ceph_osds *temp)
45966c34 2194{
6f3bfd45 2195 struct ceph_pg pgid;
45966c34 2196 struct ceph_pg_mapping *pg;
45966c34
ID		 2197 int i;
2198
6f3bfd45
ID		 2199 raw_pg_to_pg(pi, raw_pgid, &pgid);
2200 ceph_osds_init(temp);
45966c34
ID		 2201
2202 /* pg_temp? */
2203 pg = __lookup_pg_mapping(&osdmap->pg_temp, pgid);
2204 if (pg) {
45966c34
ID		 2205 for (i = 0; i < pg->pg_temp.len; i++) {
2206 if (ceph_osd_is_down(osdmap, pg->pg_temp.osds[i])) {
6f3bfd45 2207 if (ceph_can_shift_osds(pi))
45966c34 2208 continue;
6f3bfd45
ID		 2209
2210 temp->osds[temp->size++] = CRUSH_ITEM_NONE;
45966c34 2211 } else {
6f3bfd45 2212 temp->osds[temp->size++] = pg->pg_temp.osds[i];
45966c34
ID		 2213 }
2214 }
2215
2216 /* apply pg_temp's primary */
6f3bfd45
ID		 2217 for (i = 0; i < temp->size; i++) {
2218 if (temp->osds[i] != CRUSH_ITEM_NONE) {
2219 temp->primary = temp->osds[i];
45966c34
ID		 2220 break;
2221 }
2222 }
45966c34
ID		 2223 }
2224
5e8d4d36
ID		 2225 /* primary_temp? */
2226 pg = __lookup_pg_mapping(&osdmap->primary_temp, pgid);
2227 if (pg)
6f3bfd45 2228 temp->primary = pg->primary_temp.osd;
45966c34
ID		 2229}
2230
d85b7056 2231/*
6f3bfd45 2232 * Map a PG to its acting set as well as its up set.
ac972230 2233 *
6f3bfd45
ID		 2234 * Acting set is used for data mapping purposes, while up set can be
2235 * recorded for detecting interval changes and deciding whether to
2236 * resend a request.
d85b7056 2237 */
6f3bfd45
ID		 2238void ceph_pg_to_up_acting_osds(struct ceph_osdmap *osdmap,
2239 const struct ceph_pg *raw_pgid,
2240 struct ceph_osds *up,
2241 struct ceph_osds *acting)
d85b7056 2242{
6f3bfd45 2243 struct ceph_pg_pool_info *pi;
ac972230 2244 u32 pps;
d85b7056 2245
6f3bfd45
ID		 2246 pi = ceph_pg_pool_by_id(osdmap, raw_pgid->pool);
2247 if (!pi) {
2248 ceph_osds_init(up);
2249 ceph_osds_init(acting);
2250 goto out;
8008ab10 2251 }
d85b7056 2252
6f3bfd45
ID		 2253 pg_to_raw_osds(osdmap, pi, raw_pgid, up, &pps);
2254 raw_to_up_osds(osdmap, pi, up);
2255 apply_primary_affinity(osdmap, pi, pps, up);
2256 get_temp_osds(osdmap, pi, raw_pgid, acting);
2257 if (!acting->size) {
2258 memcpy(acting->osds, up->osds, up->size * sizeof(up->osds[0]));
2259 acting->size = up->size;
2260 if (acting->primary == -1)
2261 acting->primary = up->primary;
8008ab10 2262 }
6f3bfd45
ID		 2263out:
2264 WARN_ON(!osds_valid(up) || !osds_valid(acting));
d85b7056
SW		 2265}
2266
dc98ff72
ID		 2267bool ceph_pg_to_primary_shard(struct ceph_osdmap *osdmap,
2268 const struct ceph_pg *raw_pgid,
2269 struct ceph_spg *spgid)
2270{
2271 struct ceph_pg_pool_info *pi;
2272 struct ceph_pg pgid;
2273 struct ceph_osds up, acting;
2274 int i;
2275
2276 pi = ceph_pg_pool_by_id(osdmap, raw_pgid->pool);
2277 if (!pi)
2278 return false;
2279
2280 raw_pg_to_pg(pi, raw_pgid, &pgid);
2281
2282 if (ceph_can_shift_osds(pi)) {
2283 spgid->pgid = pgid; /* struct */
2284 spgid->shard = CEPH_SPG_NOSHARD;
2285 return true;
2286 }
2287
2288 ceph_pg_to_up_acting_osds(osdmap, &pgid, &up, &acting);
2289 for (i = 0; i < acting.size; i++) {
2290 if (acting.osds[i] == acting.primary) {
2291 spgid->pgid = pgid; /* struct */
2292 spgid->shard = i;
2293 return true;
2294 }
2295 }
2296
2297 return false;
2298}
2299
f24e9980 2300/*
f81f1633 2301 * Return acting primary for given PG, or -1 if none.
f24e9980 2302 */
f81f1633
ID		 2303int ceph_pg_to_acting_primary(struct ceph_osdmap *osdmap,
2304 const struct ceph_pg *raw_pgid)
f24e9980 2305{
6f3bfd45 2306 struct ceph_osds up, acting;
f24e9980 2307
f81f1633 2308 ceph_pg_to_up_acting_osds(osdmap, raw_pgid, &up, &acting);
6f3bfd45 2309 return acting.primary;
f24e9980 2310}
f81f1633 2311EXPORT_SYMBOL(ceph_pg_to_acting_primary);
Ubuntu Artful kernel mirror