]> 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
Merge remote-tracking branch 'mkp-scsi/4.11/scsi-fixes' into fixes
[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
320 ceph_decode_32_safe(p, end, yes, bad);
321 if (!yes) {
322 dout("crush_decode NO rule %d off %x %p to %p\n",
323 i, (int)(*p-start), *p, end);
324 c->rules[i] = NULL;
325 continue;
326 }
327
328 dout("crush_decode rule %d off %x %p to %p\n",
329 i, (int)(*p-start), *p, end);
330
331 /* len */
332 ceph_decode_32_safe(p, end, yes, bad);
333#if BITS_PER_LONG == 32
30dc6381 334 err = -EINVAL;
64486697
XW		 335 if (yes > (ULONG_MAX - sizeof(*r))
336 / sizeof(struct crush_rule_step))
f24e9980
SW		 337 goto bad;
338#endif
339 r = c->rules[i] = kmalloc(sizeof(*r) +
340 yes*sizeof(struct crush_rule_step),
341 GFP_NOFS);
342 if (r == NULL)
343 goto badmem;
344 dout(" rule %d is at %p\n", i, r);
345 r->len = yes;
346 ceph_decode_copy_safe(p, end, &r->mask, 4, bad); /* 4 u8's */
347 ceph_decode_need(p, end, r->len*3*sizeof(u32), bad);
348 for (j = 0; j < r->len; j++) {
c89136ea
SW		 349 r->steps[j].op = ceph_decode_32(p);
350 r->steps[j].arg1 = ceph_decode_32(p);
351 r->steps[j].arg2 = ceph_decode_32(p);
f24e9980
SW		 352 }
353 }
354
355 /* ignore trailing name maps. */
546f04ef
SW		 356 for (num_name_maps = 0; num_name_maps < 3; num_name_maps++) {
357 err = skip_name_map(p, end);
358 if (err < 0)
359 goto done;
360 }
361
362 /* tunables */
363 ceph_decode_need(p, end, 3*sizeof(u32), done);
364 c->choose_local_tries = ceph_decode_32(p);
365 c->choose_local_fallback_tries = ceph_decode_32(p);
366 c->choose_total_tries = ceph_decode_32(p);
b9b519b7 367 dout("crush decode tunable choose_local_tries = %d\n",
546f04ef 368 c->choose_local_tries);
b9b519b7 369 dout("crush decode tunable choose_local_fallback_tries = %d\n",
546f04ef 370 c->choose_local_fallback_tries);
b9b519b7 371 dout("crush decode tunable choose_total_tries = %d\n",
546f04ef 372 c->choose_total_tries);
f24e9980 373
1604f488
JS		 374 ceph_decode_need(p, end, sizeof(u32), done);
375 c->chooseleaf_descend_once = ceph_decode_32(p);
b9b519b7 376 dout("crush decode tunable chooseleaf_descend_once = %d\n",
1604f488
JS		 377 c->chooseleaf_descend_once);
378
f140662f
ID		 379 ceph_decode_need(p, end, sizeof(u8), done);
380 c->chooseleaf_vary_r = ceph_decode_8(p);
b9b519b7 381 dout("crush decode tunable chooseleaf_vary_r = %d\n",
f140662f
ID		 382 c->chooseleaf_vary_r);
383
b9b519b7
ID		 384 /* skip straw_calc_version, allowed_bucket_algs */
385 ceph_decode_need(p, end, sizeof(u8) + sizeof(u32), done);
386 *p += sizeof(u8) + sizeof(u32);
387
388 ceph_decode_need(p, end, sizeof(u8), done);
389 c->chooseleaf_stable = ceph_decode_8(p);
390 dout("crush decode tunable chooseleaf_stable = %d\n",
391 c->chooseleaf_stable);
392
66a0e2d5
ID		 393 crush_finalize(c);
394
546f04ef 395done:
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
685 if (ev >= 15)
686 pi->last_force_request_resend = ceph_decode_32(p);
687 else
688 pi->last_force_request_resend = 0;
689
4f6a7e5e
SW		 690 /* ignore the rest */
691
692 *p = pool_end;
693 calc_pg_masks(pi);
73a7e693
SW		 694 return 0;
695
696bad:
697 return -EINVAL;
efd7576b
SW		 698}
699
0f70c7ee 700static int decode_pool_names(void **p, void *end, struct ceph_osdmap *map)
2844a76a
SW		 701{
702 struct ceph_pg_pool_info *pi;
4f6a7e5e
SW		 703 u32 num, len;
704 u64 pool;
2844a76a
SW		 705
706 ceph_decode_32_safe(p, end, num, bad);
707 dout(" %d pool names\n", num);
708 while (num--) {
4f6a7e5e 709 ceph_decode_64_safe(p, end, pool, bad);
2844a76a 710 ceph_decode_32_safe(p, end, len, bad);
4f6a7e5e 711 dout(" pool %llu len %d\n", pool, len);
ad3b904c 712 ceph_decode_need(p, end, len, bad);
2844a76a
SW		 713 pi = __lookup_pg_pool(&map->pg_pools, pool);
714 if (pi) {
ad3b904c
XW		 715 char *name = kstrndup(*p, len, GFP_NOFS);
716
717 if (!name)
718 return -ENOMEM;
2844a76a 719 kfree(pi->name);
ad3b904c
XW		 720 pi->name = name;
721 dout(" name is %s\n", pi->name);
2844a76a
SW		 722 }
723 *p += len;
724 }
725 return 0;
726
727bad:
728 return -EINVAL;
729}
730
731/*
732 * osd map
733 */
e5253a7b
ID		 734struct ceph_osdmap *ceph_osdmap_alloc(void)
735{
736 struct ceph_osdmap *map;
737
738 map = kzalloc(sizeof(*map), GFP_NOIO);
739 if (!map)
740 return NULL;
741
742 map->pg_pools = RB_ROOT;
743 map->pool_max = -1;
744 map->pg_temp = RB_ROOT;
745 map->primary_temp = RB_ROOT;
743efcff 746 mutex_init(&map->crush_workspace_mutex);
e5253a7b
ID		 747
748 return map;
749}
750
2844a76a
SW		 751void ceph_osdmap_destroy(struct ceph_osdmap *map)
752{
753 dout("osdmap_destroy %p\n", map);
754 if (map->crush)
755 crush_destroy(map->crush);
756 while (!RB_EMPTY_ROOT(&map->pg_temp)) {
757 struct ceph_pg_mapping *pg =
758 rb_entry(rb_first(&map->pg_temp),
759 struct ceph_pg_mapping, node);
760 rb_erase(&pg->node, &map->pg_temp);
761 kfree(pg);
762 }
9686f94c
ID		 763 while (!RB_EMPTY_ROOT(&map->primary_temp)) {
764 struct ceph_pg_mapping *pg =
765 rb_entry(rb_first(&map->primary_temp),
766 struct ceph_pg_mapping, node);
767 rb_erase(&pg->node, &map->primary_temp);
768 kfree(pg);
769 }
2844a76a
SW		 770 while (!RB_EMPTY_ROOT(&map->pg_pools)) {
771 struct ceph_pg_pool_info *pi =
772 rb_entry(rb_first(&map->pg_pools),
773 struct ceph_pg_pool_info, node);
774 __remove_pg_pool(&map->pg_pools, pi);
775 }
776 kfree(map->osd_state);
777 kfree(map->osd_weight);
778 kfree(map->osd_addr);
2cfa34f2 779 kfree(map->osd_primary_affinity);
66a0e2d5 780 kfree(map->crush_workspace);
2844a76a
SW		 781 kfree(map);
782}
783
784/*
4d60351f
ID		 785 * Adjust max_osd value, (re)allocate arrays.
786 *
787 * The new elements are properly initialized.
2844a76a
SW		 788 */
789static int osdmap_set_max_osd(struct ceph_osdmap *map, int max)
790{
791 u8 *state;
2844a76a 792 u32 *weight;
4d60351f
ID		 793 struct ceph_entity_addr *addr;
794 int i;
2844a76a 795
4d60351f 796 state = krealloc(map->osd_state, max*sizeof(*state), GFP_NOFS);
589506f1
LR		 797 if (!state)
798 return -ENOMEM;
799 map->osd_state = state;
800
4d60351f 801 weight = krealloc(map->osd_weight, max*sizeof(*weight), GFP_NOFS);
589506f1
LR		 802 if (!weight)
803 return -ENOMEM;
804 map->osd_weight = weight;
4d60351f 805
589506f1
LR		 806 addr = krealloc(map->osd_addr, max*sizeof(*addr), GFP_NOFS);
807 if (!addr)
2844a76a 808 return -ENOMEM;
589506f1 809 map->osd_addr = addr;
2844a76a 810
4d60351f 811 for (i = map->max_osd; i < max; i++) {
589506f1
LR		 812 map->osd_state[i] = 0;
813 map->osd_weight[i] = CEPH_OSD_OUT;
814 memset(map->osd_addr + i, 0, sizeof(*map->osd_addr));
2844a76a
SW		 815 }
816
2cfa34f2
ID		 817 if (map->osd_primary_affinity) {
818 u32 *affinity;
819
820 affinity = krealloc(map->osd_primary_affinity,
821 max*sizeof(*affinity), GFP_NOFS);
822 if (!affinity)
823 return -ENOMEM;
589506f1 824 map->osd_primary_affinity = affinity;
2cfa34f2
ID		 825
826 for (i = map->max_osd; i < max; i++)
589506f1
LR		 827 map->osd_primary_affinity[i] =
828 CEPH_OSD_DEFAULT_PRIMARY_AFFINITY;
2cfa34f2
ID		 829 }
830
2844a76a 831 map->max_osd = max;
4d60351f 832
2844a76a
SW		 833 return 0;
834}
835
1b6a78b5
ID		 836static int osdmap_set_crush(struct ceph_osdmap *map, struct crush_map *crush)
837{
66a0e2d5 838 void *workspace;
743efcff 839 size_t work_size;
66a0e2d5 840
1b6a78b5
ID		 841 if (IS_ERR(crush))
842 return PTR_ERR(crush);
843
743efcff
ID		 844 work_size = crush_work_size(crush, CEPH_PG_MAX_SIZE);
845 dout("%s work_size %zu bytes\n", __func__, work_size);
846 workspace = kmalloc(work_size, GFP_NOIO);
66a0e2d5
ID		 847 if (!workspace) {
848 crush_destroy(crush);
849 return -ENOMEM;
850 }
851 crush_init_workspace(crush, workspace);
852
1b6a78b5
ID		 853 if (map->crush)
854 crush_destroy(map->crush);
66a0e2d5 855 kfree(map->crush_workspace);
1b6a78b5 856 map->crush = crush;
66a0e2d5 857 map->crush_workspace = workspace;
1b6a78b5
ID		 858 return 0;
859}
860
ec7af972
ID		 861#define OSDMAP_WRAPPER_COMPAT_VER 7
862#define OSDMAP_CLIENT_DATA_COMPAT_VER 1
863
864/*
865 * Return 0 or error. On success, *v is set to 0 for old (v6) osdmaps,
866 * to struct_v of the client_data section for new (v7 and above)
867 * osdmaps.
868 */
869static int get_osdmap_client_data_v(void **p, void *end,
870 const char *prefix, u8 *v)
871{
872 u8 struct_v;
873
874 ceph_decode_8_safe(p, end, struct_v, e_inval);
875 if (struct_v >= 7) {
876 u8 struct_compat;
877
878 ceph_decode_8_safe(p, end, struct_compat, e_inval);
879 if (struct_compat > OSDMAP_WRAPPER_COMPAT_VER) {
b9a67899
JP		 880 pr_warn("got v %d cv %d > %d of %s ceph_osdmap\n",
881 struct_v, struct_compat,
882 OSDMAP_WRAPPER_COMPAT_VER, prefix);
ec7af972
ID		 883 return -EINVAL;
884 }
885 *p += 4; /* ignore wrapper struct_len */
886
887 ceph_decode_8_safe(p, end, struct_v, e_inval);
888 ceph_decode_8_safe(p, end, struct_compat, e_inval);
889 if (struct_compat > OSDMAP_CLIENT_DATA_COMPAT_VER) {
b9a67899
JP		 890 pr_warn("got v %d cv %d > %d of %s ceph_osdmap client data\n",
891 struct_v, struct_compat,
892 OSDMAP_CLIENT_DATA_COMPAT_VER, prefix);
ec7af972
ID		 893 return -EINVAL;
894 }
895 *p += 4; /* ignore client data struct_len */
896 } else {
897 u16 version;
898
899 *p -= 1;
900 ceph_decode_16_safe(p, end, version, e_inval);
901 if (version < 6) {
b9a67899
JP		 902 pr_warn("got v %d < 6 of %s ceph_osdmap\n",
903 version, prefix);
ec7af972
ID		 904 return -EINVAL;
905 }
906
907 /* old osdmap enconding */
908 struct_v = 0;
909 }
910
911 *v = struct_v;
912 return 0;
913
914e_inval:
915 return -EINVAL;
916}
917
433fbdd3
ID		 918static int __decode_pools(void **p, void *end, struct ceph_osdmap *map,
919 bool incremental)
920{
921 u32 n;
922
923 ceph_decode_32_safe(p, end, n, e_inval);
924 while (n--) {
925 struct ceph_pg_pool_info *pi;
926 u64 pool;
927 int ret;
928
929 ceph_decode_64_safe(p, end, pool, e_inval);
930
931 pi = __lookup_pg_pool(&map->pg_pools, pool);
932 if (!incremental || !pi) {
933 pi = kzalloc(sizeof(*pi), GFP_NOFS);
934 if (!pi)
935 return -ENOMEM;
936
937 pi->id = pool;
938
939 ret = __insert_pg_pool(&map->pg_pools, pi);
940 if (ret) {
941 kfree(pi);
942 return ret;
943 }
944 }
945
946 ret = decode_pool(p, end, pi);
947 if (ret)
948 return ret;
949 }
950
951 return 0;
952
953e_inval:
954 return -EINVAL;
955}
956
957static int decode_pools(void **p, void *end, struct ceph_osdmap *map)
958{
959 return __decode_pools(p, end, map, false);
960}
961
962static int decode_new_pools(void **p, void *end, struct ceph_osdmap *map)
963{
964 return __decode_pools(p, end, map, true);
965}
966
10db634e
ID		 967static int __decode_pg_temp(void **p, void *end, struct ceph_osdmap *map,
968 bool incremental)
969{
970 u32 n;
971
972 ceph_decode_32_safe(p, end, n, e_inval);
973 while (n--) {
974 struct ceph_pg pgid;
975 u32 len, i;
976 int ret;
977
978 ret = ceph_decode_pgid(p, end, &pgid);
979 if (ret)
980 return ret;
981
982 ceph_decode_32_safe(p, end, len, e_inval);
983
984 ret = __remove_pg_mapping(&map->pg_temp, pgid);
985 BUG_ON(!incremental && ret != -ENOENT);
986
987 if (!incremental || len > 0) {
988 struct ceph_pg_mapping *pg;
989
990 ceph_decode_need(p, end, len*sizeof(u32), e_inval);
991
992 if (len > (UINT_MAX - sizeof(*pg)) / sizeof(u32))
993 return -EINVAL;
994
995 pg = kzalloc(sizeof(*pg) + len*sizeof(u32), GFP_NOFS);
996 if (!pg)
997 return -ENOMEM;
998
999 pg->pgid = pgid;
35a935d7 1000 pg->pg_temp.len = len;
10db634e 1001 for (i = 0; i < len; i++)
35a935d7 1002 pg->pg_temp.osds[i] = ceph_decode_32(p);
10db634e
ID		 1003
1004 ret = __insert_pg_mapping(pg, &map->pg_temp);
1005 if (ret) {
1006 kfree(pg);
1007 return ret;
1008 }
1009 }
1010 }
1011
1012 return 0;
1013
1014e_inval:
1015 return -EINVAL;
1016}
1017
1018static int decode_pg_temp(void **p, void *end, struct ceph_osdmap *map)
1019{
1020 return __decode_pg_temp(p, end, map, false);
1021}
1022
1023static int decode_new_pg_temp(void **p, void *end, struct ceph_osdmap *map)
1024{
1025 return __decode_pg_temp(p, end, map, true);
1026}
1027
d286de79
ID		 1028static int __decode_primary_temp(void **p, void *end, struct ceph_osdmap *map,
1029 bool incremental)
1030{
1031 u32 n;
1032
1033 ceph_decode_32_safe(p, end, n, e_inval);
1034 while (n--) {
1035 struct ceph_pg pgid;
1036 u32 osd;
1037 int ret;
1038
1039 ret = ceph_decode_pgid(p, end, &pgid);
1040 if (ret)
1041 return ret;
1042
1043 ceph_decode_32_safe(p, end, osd, e_inval);
1044
1045 ret = __remove_pg_mapping(&map->primary_temp, pgid);
1046 BUG_ON(!incremental && ret != -ENOENT);
1047
1048 if (!incremental || osd != (u32)-1) {
1049 struct ceph_pg_mapping *pg;
1050
1051 pg = kzalloc(sizeof(*pg), GFP_NOFS);
1052 if (!pg)
1053 return -ENOMEM;
1054
1055 pg->pgid = pgid;
1056 pg->primary_temp.osd = osd;
1057
1058 ret = __insert_pg_mapping(pg, &map->primary_temp);
1059 if (ret) {
1060 kfree(pg);
1061 return ret;
1062 }
1063 }
1064 }
1065
1066 return 0;
1067
1068e_inval:
1069 return -EINVAL;
1070}
1071
1072static int decode_primary_temp(void **p, void *end, struct ceph_osdmap *map)
1073{
1074 return __decode_primary_temp(p, end, map, false);
1075}
1076
1077static int decode_new_primary_temp(void **p, void *end,
1078 struct ceph_osdmap *map)
1079{
1080 return __decode_primary_temp(p, end, map, true);
1081}
1082
2cfa34f2
ID		 1083u32 ceph_get_primary_affinity(struct ceph_osdmap *map, int osd)
1084{
1085 BUG_ON(osd >= map->max_osd);
1086
1087 if (!map->osd_primary_affinity)
1088 return CEPH_OSD_DEFAULT_PRIMARY_AFFINITY;
1089
1090 return map->osd_primary_affinity[osd];
1091}
1092
1093static int set_primary_affinity(struct ceph_osdmap *map, int osd, u32 aff)
1094{
1095 BUG_ON(osd >= map->max_osd);
1096
1097 if (!map->osd_primary_affinity) {
1098 int i;
1099
1100 map->osd_primary_affinity = kmalloc(map->max_osd*sizeof(u32),
1101 GFP_NOFS);
1102 if (!map->osd_primary_affinity)
1103 return -ENOMEM;
1104
1105 for (i = 0; i < map->max_osd; i++)
1106 map->osd_primary_affinity[i] =
1107 CEPH_OSD_DEFAULT_PRIMARY_AFFINITY;
1108 }
1109
1110 map->osd_primary_affinity[osd] = aff;
1111
1112 return 0;
1113}
1114
63a6993f
ID		 1115static int decode_primary_affinity(void **p, void *end,
1116 struct ceph_osdmap *map)
1117{
1118 u32 len, i;
1119
1120 ceph_decode_32_safe(p, end, len, e_inval);
1121 if (len == 0) {
1122 kfree(map->osd_primary_affinity);
1123 map->osd_primary_affinity = NULL;
1124 return 0;
1125 }
1126 if (len != map->max_osd)
1127 goto e_inval;
1128
1129 ceph_decode_need(p, end, map->max_osd*sizeof(u32), e_inval);
1130
1131 for (i = 0; i < map->max_osd; i++) {
1132 int ret;
1133
1134 ret = set_primary_affinity(map, i, ceph_decode_32(p));
1135 if (ret)
1136 return ret;
1137 }
1138
1139 return 0;
1140
1141e_inval:
1142 return -EINVAL;
1143}
1144
1145static int decode_new_primary_affinity(void **p, void *end,
1146 struct ceph_osdmap *map)
1147{
1148 u32 n;
1149
1150 ceph_decode_32_safe(p, end, n, e_inval);
1151 while (n--) {
1152 u32 osd, aff;
1153 int ret;
1154
1155 ceph_decode_32_safe(p, end, osd, e_inval);
1156 ceph_decode_32_safe(p, end, aff, e_inval);
1157
1158 ret = set_primary_affinity(map, osd, aff);
1159 if (ret)
1160 return ret;
f31da0f3
ID		 1161
1162 pr_info("osd%d primary-affinity 0x%x\n", osd, aff);
63a6993f
ID		 1163 }
1164
1165 return 0;
1166
1167e_inval:
1168 return -EINVAL;
1169}
1170
f24e9980
SW		 1171/*
1172 * decode a full map.
1173 */
a2505d63 1174static int osdmap_decode(void **p, void *end, struct ceph_osdmap *map)
f24e9980 1175{
ec7af972 1176 u8 struct_v;
38a8d560 1177 u32 epoch = 0;
f24e9980 1178 void *start = *p;
3977058c
ID		 1179 u32 max;
1180 u32 len, i;
597b52f6 1181 int err;
f24e9980 1182
a2505d63 1183 dout("%s %p to %p len %d\n", __func__, *p, end, (int)(end - *p));
f24e9980 1184
ec7af972
ID		 1185 err = get_osdmap_client_data_v(p, end, "full", &struct_v);
1186 if (err)
1187 goto bad;
f24e9980 1188
53bbaba9
ID		 1189 /* fsid, epoch, created, modified */
1190 ceph_decode_need(p, end, sizeof(map->fsid) + sizeof(u32) +
1191 sizeof(map->created) + sizeof(map->modified), e_inval);
f24e9980 1192 ceph_decode_copy(p, &map->fsid, sizeof(map->fsid));
38a8d560 1193 epoch = map->epoch = ceph_decode_32(p);
f24e9980
SW		 1194 ceph_decode_copy(p, &map->created, sizeof(map->created));
1195 ceph_decode_copy(p, &map->modified, sizeof(map->modified));
1196
433fbdd3
ID		 1197 /* pools */
1198 err = decode_pools(p, end, map);
1199 if (err)
1200 goto bad;
2844a76a 1201
0f70c7ee
ID		 1202 /* pool_name */
1203 err = decode_pool_names(p, end, map);
597b52f6 1204 if (err)
4f6a7e5e 1205 goto bad;
2844a76a 1206
597b52f6 1207 ceph_decode_32_safe(p, end, map->pool_max, e_inval);
f24e9980 1208
597b52f6 1209 ceph_decode_32_safe(p, end, map->flags, e_inval);
f24e9980 1210
3977058c
ID		 1211 /* max_osd */
1212 ceph_decode_32_safe(p, end, max, e_inval);
f24e9980
SW		 1213
1214 /* (re)alloc osd arrays */
1215 err = osdmap_set_max_osd(map, max);
597b52f6 1216 if (err)
f24e9980 1217 goto bad;
f24e9980 1218
2d88b2e0 1219 /* osd_state, osd_weight, osd_addrs->client_addr */
f24e9980
SW		 1220 ceph_decode_need(p, end, 3*sizeof(u32) +
1221 map->max_osd*(1 + sizeof(*map->osd_weight) +
597b52f6
ID		 1222 sizeof(*map->osd_addr)), e_inval);
1223
2d88b2e0
ID		 1224 if (ceph_decode_32(p) != map->max_osd)
1225 goto e_inval;
1226
f24e9980
SW		 1227 ceph_decode_copy(p, map->osd_state, map->max_osd);
1228
2d88b2e0
ID		 1229 if (ceph_decode_32(p) != map->max_osd)
1230 goto e_inval;
1231
f24e9980 1232 for (i = 0; i < map->max_osd; i++)
c89136ea 1233 map->osd_weight[i] = ceph_decode_32(p);
f24e9980 1234
2d88b2e0
ID		 1235 if (ceph_decode_32(p) != map->max_osd)
1236 goto e_inval;
1237
f24e9980 1238 ceph_decode_copy(p, map->osd_addr, map->max_osd*sizeof(*map->osd_addr));
63f2d211
SW		 1239 for (i = 0; i < map->max_osd; i++)
1240 ceph_decode_addr(&map->osd_addr[i]);
f24e9980
SW		 1241
1242 /* pg_temp */
10db634e
ID		 1243 err = decode_pg_temp(p, end, map);
1244 if (err)
1245 goto bad;
f24e9980 1246
d286de79
ID		 1247 /* primary_temp */
1248 if (struct_v >= 1) {
1249 err = decode_primary_temp(p, end, map);
1250 if (err)
1251 goto bad;
1252 }
1253
63a6993f
ID		 1254 /* primary_affinity */
1255 if (struct_v >= 2) {
1256 err = decode_primary_affinity(p, end, map);
1257 if (err)
1258 goto bad;
1259 } else {
1260 /* XXX can this happen? */
1261 kfree(map->osd_primary_affinity);
1262 map->osd_primary_affinity = NULL;
1263 }
1264
f24e9980 1265 /* crush */
597b52f6 1266 ceph_decode_32_safe(p, end, len, e_inval);
1b6a78b5
ID		 1267 err = osdmap_set_crush(map, crush_decode(*p, min(*p + len, end)));
1268 if (err)
f24e9980 1269 goto bad;
f24e9980 1270
38a8d560 1271 /* ignore the rest */
f24e9980
SW		 1272 *p = end;
1273
38a8d560 1274 dout("full osdmap epoch %d max_osd %d\n", map->epoch, map->max_osd);
a2505d63 1275 return 0;
f24e9980 1276
597b52f6
ID		 1277e_inval:
1278 err = -EINVAL;
f24e9980 1279bad:
38a8d560
ID		 1280 pr_err("corrupt full osdmap (%d) epoch %d off %d (%p of %p-%p)\n",
1281 err, epoch, (int)(*p - start), *p, start, end);
1282 print_hex_dump(KERN_DEBUG, "osdmap: ",
1283 DUMP_PREFIX_OFFSET, 16, 1,
1284 start, end - start, true);
a2505d63
ID		 1285 return err;
1286}
1287
1288/*
1289 * Allocate and decode a full map.
1290 */
1291struct ceph_osdmap *ceph_osdmap_decode(void **p, void *end)
1292{
1293 struct ceph_osdmap *map;
1294 int ret;
1295
e5253a7b 1296 map = ceph_osdmap_alloc();
a2505d63
ID		 1297 if (!map)
1298 return ERR_PTR(-ENOMEM);
1299
a2505d63
ID		 1300 ret = osdmap_decode(p, end, map);
1301 if (ret) {
1302 ceph_osdmap_destroy(map);
1303 return ERR_PTR(ret);
1304 }
1305
1306 return map;
f24e9980
SW		 1307}
1308
930c5328
ID		 1309/*
1310 * Encoding order is (new_up_client, new_state, new_weight). Need to
1311 * apply in the (new_weight, new_state, new_up_client) order, because
1312 * an incremental map may look like e.g.
1313 *
1314 * new_up_client: { osd=6, addr=... } # set osd_state and addr
1315 * new_state: { osd=6, xorstate=EXISTS } # clear osd_state
1316 */
1317static int decode_new_up_state_weight(void **p, void *end,
1318 struct ceph_osdmap *map)
1319{
1320 void *new_up_client;
1321 void *new_state;
1322 void *new_weight_end;
1323 u32 len;
1324
1325 new_up_client = *p;
1326 ceph_decode_32_safe(p, end, len, e_inval);
1327 len *= sizeof(u32) + sizeof(struct ceph_entity_addr);
1328 ceph_decode_need(p, end, len, e_inval);
1329 *p += len;
1330
1331 new_state = *p;
1332 ceph_decode_32_safe(p, end, len, e_inval);
1333 len *= sizeof(u32) + sizeof(u8);
1334 ceph_decode_need(p, end, len, e_inval);
1335 *p += len;
1336
1337 /* new_weight */
1338 ceph_decode_32_safe(p, end, len, e_inval);
1339 while (len--) {
1340 s32 osd;
1341 u32 w;
1342
1343 ceph_decode_need(p, end, 2*sizeof(u32), e_inval);
1344 osd = ceph_decode_32(p);
1345 w = ceph_decode_32(p);
1346 BUG_ON(osd >= map->max_osd);
1347 pr_info("osd%d weight 0x%x %s\n", osd, w,
1348 w == CEPH_OSD_IN ? "(in)" :
1349 (w == CEPH_OSD_OUT ? "(out)" : ""));
1350 map->osd_weight[osd] = w;
1351
1352 /*
1353 * If we are marking in, set the EXISTS, and clear the
1354 * AUTOOUT and NEW bits.
1355 */
1356 if (w) {
1357 map->osd_state[osd] |= CEPH_OSD_EXISTS;
1358 map->osd_state[osd] &= ~(CEPH_OSD_AUTOOUT |
1359 CEPH_OSD_NEW);
1360 }
1361 }
1362 new_weight_end = *p;
1363
1364 /* new_state (up/down) */
1365 *p = new_state;
1366 len = ceph_decode_32(p);
1367 while (len--) {
1368 s32 osd;
1369 u8 xorstate;
1370 int ret;
1371
1372 osd = ceph_decode_32(p);
1373 xorstate = ceph_decode_8(p);
1374 if (xorstate == 0)
1375 xorstate = CEPH_OSD_UP;
1376 BUG_ON(osd >= map->max_osd);
1377 if ((map->osd_state[osd] & CEPH_OSD_UP) &&
1378 (xorstate & CEPH_OSD_UP))
1379 pr_info("osd%d down\n", osd);
1380 if ((map->osd_state[osd] & CEPH_OSD_EXISTS) &&
1381 (xorstate & CEPH_OSD_EXISTS)) {
1382 pr_info("osd%d does not exist\n", osd);
1383 map->osd_weight[osd] = CEPH_OSD_IN;
1384 ret = set_primary_affinity(map, osd,
1385 CEPH_OSD_DEFAULT_PRIMARY_AFFINITY);
1386 if (ret)
1387 return ret;
1388 memset(map->osd_addr + osd, 0, sizeof(*map->osd_addr));
1389 map->osd_state[osd] = 0;
1390 } else {
1391 map->osd_state[osd] ^= xorstate;
1392 }
1393 }
1394
1395 /* new_up_client */
1396 *p = new_up_client;
1397 len = ceph_decode_32(p);
1398 while (len--) {
1399 s32 osd;
1400 struct ceph_entity_addr addr;
1401
1402 osd = ceph_decode_32(p);
1403 ceph_decode_copy(p, &addr, sizeof(addr));
1404 ceph_decode_addr(&addr);
1405 BUG_ON(osd >= map->max_osd);
1406 pr_info("osd%d up\n", osd);
1407 map->osd_state[osd] |= CEPH_OSD_EXISTS | CEPH_OSD_UP;
1408 map->osd_addr[osd] = addr;
1409 }
1410
1411 *p = new_weight_end;
1412 return 0;
1413
1414e_inval:
1415 return -EINVAL;
1416}
1417
f24e9980
SW		 1418/*
1419 * decode and apply an incremental map update.
1420 */
1421struct ceph_osdmap *osdmap_apply_incremental(void **p, void *end,
0c0a8de1 1422 struct ceph_osdmap *map)
f24e9980 1423{
f24e9980
SW		 1424 struct ceph_fsid fsid;
1425 u32 epoch = 0;
1426 struct ceph_timespec modified;
4f6a7e5e
SW		 1427 s32 len;
1428 u64 pool;
1429 __s64 new_pool_max;
1430 __s32 new_flags, max;
f24e9980 1431 void *start = *p;
86f1742b 1432 int err;
ec7af972 1433 u8 struct_v;
f24e9980 1434
38a8d560
ID		 1435 dout("%s %p to %p len %d\n", __func__, *p, end, (int)(end - *p));
1436
ec7af972
ID		 1437 err = get_osdmap_client_data_v(p, end, "inc", &struct_v);
1438 if (err)
1439 goto bad;
f24e9980 1440
53bbaba9
ID		 1441 /* fsid, epoch, modified, new_pool_max, new_flags */
1442 ceph_decode_need(p, end, sizeof(fsid) + sizeof(u32) + sizeof(modified) +
1443 sizeof(u64) + sizeof(u32), e_inval);
f24e9980 1444 ceph_decode_copy(p, &fsid, sizeof(fsid));
c89136ea 1445 epoch = ceph_decode_32(p);
f24e9980
SW		 1446 BUG_ON(epoch != map->epoch+1);
1447 ceph_decode_copy(p, &modified, sizeof(modified));
4f6a7e5e 1448 new_pool_max = ceph_decode_64(p);
c89136ea 1449 new_flags = ceph_decode_32(p);
f24e9980
SW		 1450
1451 /* full map? */
86f1742b 1452 ceph_decode_32_safe(p, end, len, e_inval);
f24e9980
SW		 1453 if (len > 0) {
1454 dout("apply_incremental full map len %d, %p to %p\n",
1455 len, *p, end);
a2505d63 1456 return ceph_osdmap_decode(p, min(*p+len, end));
f24e9980
SW		 1457 }
1458
1459 /* new crush? */
86f1742b 1460 ceph_decode_32_safe(p, end, len, e_inval);
f24e9980 1461 if (len > 0) {
1b6a78b5
ID		 1462 err = osdmap_set_crush(map,
1463 crush_decode(*p, min(*p + len, end)));
1464 if (err)
86f1742b 1465 goto bad;
cebc5be6 1466 *p += len;
f24e9980
SW		 1467 }
1468
1469 /* new flags? */
1470 if (new_flags >= 0)
1471 map->flags = new_flags;
4fc51be8
SW		 1472 if (new_pool_max >= 0)
1473 map->pool_max = new_pool_max;
f24e9980 1474
f24e9980 1475 /* new max? */
53bbaba9 1476 ceph_decode_32_safe(p, end, max, e_inval);
f24e9980
SW		 1477 if (max >= 0) {
1478 err = osdmap_set_max_osd(map, max);
86f1742b 1479 if (err)
f24e9980
SW		 1480 goto bad;
1481 }
1482
1483 map->epoch++;
31456665 1484 map->modified = modified;
f24e9980 1485
433fbdd3
ID		 1486 /* new_pools */
1487 err = decode_new_pools(p, end, map);
1488 if (err)
1489 goto bad;
9464d008 1490
0f70c7ee
ID		 1491 /* new_pool_names */
1492 err = decode_pool_names(p, end, map);
9464d008
ID		 1493 if (err)
1494 goto bad;
f24e9980 1495
4fc51be8 1496 /* old_pool */
86f1742b 1497 ceph_decode_32_safe(p, end, len, e_inval);
4fc51be8
SW		 1498 while (len--) {
1499 struct ceph_pg_pool_info *pi;
1500
86f1742b 1501 ceph_decode_64_safe(p, end, pool, e_inval);
4fc51be8 1502 pi = __lookup_pg_pool(&map->pg_pools, pool);
2844a76a
SW		 1503 if (pi)
1504 __remove_pg_pool(&map->pg_pools, pi);
4fc51be8 1505 }
f24e9980 1506
930c5328
ID		 1507 /* new_up_client, new_state, new_weight */
1508 err = decode_new_up_state_weight(p, end, map);
1509 if (err)
1510 goto bad;
f24e9980
SW		 1511
1512 /* new_pg_temp */
10db634e
ID		 1513 err = decode_new_pg_temp(p, end, map);
1514 if (err)
1515 goto bad;
f24e9980 1516
d286de79
ID		 1517 /* new_primary_temp */
1518 if (struct_v >= 1) {
1519 err = decode_new_primary_temp(p, end, map);
1520 if (err)
1521 goto bad;
1522 }
1523
63a6993f
ID		 1524 /* new_primary_affinity */
1525 if (struct_v >= 2) {
1526 err = decode_new_primary_affinity(p, end, map);
1527 if (err)
1528 goto bad;
1529 }
1530
f24e9980
SW		 1531 /* ignore the rest */
1532 *p = end;
38a8d560
ID		 1533
1534 dout("inc osdmap epoch %d max_osd %d\n", map->epoch, map->max_osd);
f24e9980
SW		 1535 return map;
1536
86f1742b
ID		 1537e_inval:
1538 err = -EINVAL;
f24e9980 1539bad:
38a8d560
ID		 1540 pr_err("corrupt inc osdmap (%d) epoch %d off %d (%p of %p-%p)\n",
1541 err, epoch, (int)(*p - start), *p, start, end);
9ec7cab1
SW		 1542 print_hex_dump(KERN_DEBUG, "osdmap: ",
1543 DUMP_PREFIX_OFFSET, 16, 1,
1544 start, end - start, true);
f24e9980
SW		 1545 return ERR_PTR(err);
1546}
1547
30c156d9
YZ		 1548void ceph_oloc_copy(struct ceph_object_locator *dest,
1549 const struct ceph_object_locator *src)
1550{
1551 WARN_ON(!ceph_oloc_empty(dest));
1552 WARN_ON(dest->pool_ns); /* empty() only covers ->pool */
1553
1554 dest->pool = src->pool;
1555 if (src->pool_ns)
1556 dest->pool_ns = ceph_get_string(src->pool_ns);
1557}
1558EXPORT_SYMBOL(ceph_oloc_copy);
1559
1560void ceph_oloc_destroy(struct ceph_object_locator *oloc)
1561{
1562 ceph_put_string(oloc->pool_ns);
1563}
1564EXPORT_SYMBOL(ceph_oloc_destroy);
1565
d30291b9
ID		 1566void ceph_oid_copy(struct ceph_object_id *dest,
1567 const struct ceph_object_id *src)
1568{
1569 WARN_ON(!ceph_oid_empty(dest));
1570
1571 if (src->name != src->inline_name) {
1572 /* very rare, see ceph_object_id definition */
1573 dest->name = kmalloc(src->name_len + 1,
1574 GFP_NOIO | __GFP_NOFAIL);
1575 }
1576
1577 memcpy(dest->name, src->name, src->name_len + 1);
1578 dest->name_len = src->name_len;
1579}
1580EXPORT_SYMBOL(ceph_oid_copy);
1581
1582static __printf(2, 0)
1583int oid_printf_vargs(struct ceph_object_id *oid, const char *fmt, va_list ap)
1584{
1585 int len;
1586
1587 WARN_ON(!ceph_oid_empty(oid));
1588
1589 len = vsnprintf(oid->inline_name, sizeof(oid->inline_name), fmt, ap);
1590 if (len >= sizeof(oid->inline_name))
1591 return len;
1592
1593 oid->name_len = len;
1594 return 0;
1595}
1596
1597/*
1598 * If oid doesn't fit into inline buffer, BUG.
1599 */
1600void ceph_oid_printf(struct ceph_object_id *oid, const char *fmt, ...)
1601{
1602 va_list ap;
1603
1604 va_start(ap, fmt);
1605 BUG_ON(oid_printf_vargs(oid, fmt, ap));
1606 va_end(ap);
1607}
1608EXPORT_SYMBOL(ceph_oid_printf);
1609
1610static __printf(3, 0)
1611int oid_aprintf_vargs(struct ceph_object_id *oid, gfp_t gfp,
1612 const char *fmt, va_list ap)
1613{
1614 va_list aq;
1615 int len;
1616
1617 va_copy(aq, ap);
1618 len = oid_printf_vargs(oid, fmt, aq);
1619 va_end(aq);
1620
1621 if (len) {
1622 char *external_name;
1623
1624 external_name = kmalloc(len + 1, gfp);
1625 if (!external_name)
1626 return -ENOMEM;
f24e9980 1627
d30291b9
ID		 1628 oid->name = external_name;
1629 WARN_ON(vsnprintf(oid->name, len + 1, fmt, ap) != len);
1630 oid->name_len = len;
1631 }
f24e9980 1632
d30291b9
ID		 1633 return 0;
1634}
1635
1636/*
1637 * If oid doesn't fit into inline buffer, allocate.
1638 */
1639int ceph_oid_aprintf(struct ceph_object_id *oid, gfp_t gfp,
1640 const char *fmt, ...)
1641{
1642 va_list ap;
1643 int ret;
1644
1645 va_start(ap, fmt);
1646 ret = oid_aprintf_vargs(oid, gfp, fmt, ap);
1647 va_end(ap);
1648
1649 return ret;
1650}
1651EXPORT_SYMBOL(ceph_oid_aprintf);
1652
1653void ceph_oid_destroy(struct ceph_object_id *oid)
1654{
1655 if (oid->name != oid->inline_name)
1656 kfree(oid->name);
1657}
1658EXPORT_SYMBOL(ceph_oid_destroy);
f24e9980 1659
63244fa1
ID		 1660/*
1661 * osds only
1662 */
1663static bool __osds_equal(const struct ceph_osds *lhs,
1664 const struct ceph_osds *rhs)
1665{
1666 if (lhs->size == rhs->size &&
1667 !memcmp(lhs->osds, rhs->osds, rhs->size * sizeof(rhs->osds[0])))
1668 return true;
1669
1670 return false;
1671}
1672
1673/*
1674 * osds + primary
1675 */
1676static bool osds_equal(const struct ceph_osds *lhs,
1677 const struct ceph_osds *rhs)
1678{
1679 if (__osds_equal(lhs, rhs) &&
1680 lhs->primary == rhs->primary)
1681 return true;
1682
1683 return false;
1684}
1685
6f3bfd45
ID		 1686static bool osds_valid(const struct ceph_osds *set)
1687{
1688 /* non-empty set */
1689 if (set->size > 0 && set->primary >= 0)
1690 return true;
1691
1692 /* empty can_shift_osds set */
1693 if (!set->size && set->primary == -1)
1694 return true;
1695
1696 /* empty !can_shift_osds set - all NONE */
1697 if (set->size > 0 && set->primary == -1) {
1698 int i;
1699
1700 for (i = 0; i < set->size; i++) {
1701 if (set->osds[i] != CRUSH_ITEM_NONE)
1702 break;
1703 }
1704 if (i == set->size)
1705 return true;
1706 }
1707
1708 return false;
1709}
1710
1711void ceph_osds_copy(struct ceph_osds *dest, const struct ceph_osds *src)
1712{
1713 memcpy(dest->osds, src->osds, src->size * sizeof(src->osds[0]));
1714 dest->size = src->size;
1715 dest->primary = src->primary;
1716}
1717
63244fa1
ID		 1718static bool is_split(const struct ceph_pg *pgid,
1719 u32 old_pg_num,
1720 u32 new_pg_num)
1721{
1722 int old_bits = calc_bits_of(old_pg_num);
1723 int old_mask = (1 << old_bits) - 1;
1724 int n;
1725
1726 WARN_ON(pgid->seed >= old_pg_num);
1727 if (new_pg_num <= old_pg_num)
1728 return false;
1729
1730 for (n = 1; ; n++) {
1731 int next_bit = n << (old_bits - 1);
1732 u32 s = next_bit | pgid->seed;
1733
1734 if (s < old_pg_num || s == pgid->seed)
1735 continue;
1736 if (s >= new_pg_num)
1737 break;
1738
1739 s = ceph_stable_mod(s, old_pg_num, old_mask);
1740 if (s == pgid->seed)
1741 return true;
1742 }
1743
1744 return false;
1745}
1746
1747bool ceph_is_new_interval(const struct ceph_osds *old_acting,
1748 const struct ceph_osds *new_acting,
1749 const struct ceph_osds *old_up,
1750 const struct ceph_osds *new_up,
1751 int old_size,
1752 int new_size,
1753 int old_min_size,
1754 int new_min_size,
1755 u32 old_pg_num,
1756 u32 new_pg_num,
1757 bool old_sort_bitwise,
1758 bool new_sort_bitwise,
1759 const struct ceph_pg *pgid)
1760{
1761 return !osds_equal(old_acting, new_acting) ||
1762 !osds_equal(old_up, new_up) ||
1763 old_size != new_size ||
1764 old_min_size != new_min_size ||
1765 is_split(pgid, old_pg_num, new_pg_num) ||
1766 old_sort_bitwise != new_sort_bitwise;
1767}
1768
1769static int calc_pg_rank(int osd, const struct ceph_osds *acting)
1770{
1771 int i;
1772
1773 for (i = 0; i < acting->size; i++) {
1774 if (acting->osds[i] == osd)
1775 return i;
1776 }
1777
1778 return -1;
1779}
1780
1781static bool primary_changed(const struct ceph_osds *old_acting,
1782 const struct ceph_osds *new_acting)
1783{
1784 if (!old_acting->size && !new_acting->size)
1785 return false; /* both still empty */
1786
1787 if (!old_acting->size ^ !new_acting->size)
1788 return true; /* was empty, now not, or vice versa */
1789
1790 if (old_acting->primary != new_acting->primary)
1791 return true; /* primary changed */
1792
1793 if (calc_pg_rank(old_acting->primary, old_acting) !=
1794 calc_pg_rank(new_acting->primary, new_acting))
1795 return true;
1796
1797 return false; /* same primary (tho replicas may have changed) */
1798}
1799
1800bool ceph_osds_changed(const struct ceph_osds *old_acting,
1801 const struct ceph_osds *new_acting,
1802 bool any_change)
1803{
1804 if (primary_changed(old_acting, new_acting))
1805 return true;
1806
1807 if (any_change && !__osds_equal(old_acting, new_acting))
1808 return true;
1809
1810 return false;
1811}
1812
f24e9980
SW		 1813/*
1814 * calculate file layout from given offset, length.
1815 * fill in correct oid, logical length, and object extent
1816 * offset, length.
1817 *
1818 * for now, we write only a single su, until we can
1819 * pass a stride back to the caller.
1820 */
d63b77f4 1821int ceph_calc_file_object_mapping(struct ceph_file_layout *layout,
e8afad65 1822 u64 off, u64 len,
645a1025 1823 u64 *ono,
f24e9980
SW		 1824 u64 *oxoff, u64 *oxlen)
1825{
7627151e
YZ		 1826 u32 osize = layout->object_size;
1827 u32 su = layout->stripe_unit;
1828 u32 sc = layout->stripe_count;
f24e9980
SW		 1829 u32 bl, stripeno, stripepos, objsetno;
1830 u32 su_per_object;
ff1d1f71 1831 u64 t, su_offset;
f24e9980 1832
e8afad65 1833 dout("mapping %llu~%llu osize %u fl_su %u\n", off, len,
f24e9980 1834 osize, su);
d63b77f4
SW		 1835 if (su == 0 || sc == 0)
1836 goto invalid;
35e054a6 1837 su_per_object = osize / su;
d63b77f4
SW		 1838 if (su_per_object == 0)
1839 goto invalid;
f24e9980
SW		 1840 dout("osize %u / su %u = su_per_object %u\n", osize, su,
1841 su_per_object);
1842
d63b77f4
SW		 1843 if ((su & ~PAGE_MASK) != 0)
1844 goto invalid;
1845
f24e9980
SW		 1846 /* bl = *off / su; */
1847 t = off;
1848 do_div(t, su);
1849 bl = t;
1850 dout("off %llu / su %u = bl %u\n", off, su, bl);
1851
1852 stripeno = bl / sc;
1853 stripepos = bl % sc;
1854 objsetno = stripeno / su_per_object;
1855
645a1025 1856 *ono = objsetno * sc + stripepos;
95c96174 1857 dout("objset %u * sc %u = ono %u\n", objsetno, sc, (unsigned int)*ono);
645a1025
SW		 1858
1859 /* *oxoff = *off % layout->fl_stripe_unit; # offset in su */
f24e9980 1860 t = off;
ff1d1f71
NW		 1861 su_offset = do_div(t, su);
1862 *oxoff = su_offset + (stripeno % su_per_object) * su;
1863
1864 /*
1865 * Calculate the length of the extent being written to the selected
e8afad65 1866 * object. This is the minimum of the full length requested (len) or
ff1d1f71
NW		 1867 * the remainder of the current stripe being written to.
1868 */
e8afad65 1869 *oxlen = min_t(u64, len, su - su_offset);
f24e9980
SW		 1870
1871 dout(" obj extent %llu~%llu\n", *oxoff, *oxlen);
d63b77f4
SW		 1872 return 0;
1873
1874invalid:
1875 dout(" invalid layout\n");
1876 *ono = 0;
1877 *oxoff = 0;
1878 *oxlen = 0;
1879 return -EINVAL;
f24e9980 1880}
3d14c5d2 1881EXPORT_SYMBOL(ceph_calc_file_object_mapping);
f24e9980
SW		 1882
1883/*
d9591f5e
ID		 1884 * Map an object into a PG.
1885 *
1886 * Should only be called with target_oid and target_oloc (as opposed to
1887 * base_oid and base_oloc), since tiering isn't taken into account.
f24e9980 1888 */
d9591f5e
ID		 1889int ceph_object_locator_to_pg(struct ceph_osdmap *osdmap,
1890 struct ceph_object_id *oid,
1891 struct ceph_object_locator *oloc,
1892 struct ceph_pg *raw_pgid)
f24e9980 1893{
7c13cb64 1894 struct ceph_pg_pool_info *pi;
f24e9980 1895
d9591f5e 1896 pi = ceph_pg_pool_by_id(osdmap, oloc->pool);
7c13cb64 1897 if (!pi)
d9591f5e 1898 return -ENOENT;
f24e9980 1899
30c156d9
YZ		 1900 if (!oloc->pool_ns) {
1901 raw_pgid->pool = oloc->pool;
1902 raw_pgid->seed = ceph_str_hash(pi->object_hash, oid->name,
1903 oid->name_len);
1904 dout("%s %s -> raw_pgid %llu.%x\n", __func__, oid->name,
1905 raw_pgid->pool, raw_pgid->seed);
1906 } else {
1907 char stack_buf[256];
1908 char *buf = stack_buf;
1909 int nsl = oloc->pool_ns->len;
1910 size_t total = nsl + 1 + oid->name_len;
1911
1912 if (total > sizeof(stack_buf)) {
1913 buf = kmalloc(total, GFP_NOIO);
1914 if (!buf)
1915 return -ENOMEM;
1916 }
1917 memcpy(buf, oloc->pool_ns->str, nsl);
1918 buf[nsl] = '\037';
1919 memcpy(buf + nsl + 1, oid->name, oid->name_len);
1920 raw_pgid->pool = oloc->pool;
1921 raw_pgid->seed = ceph_str_hash(pi->object_hash, buf, total);
1922 if (buf != stack_buf)
1923 kfree(buf);
1924 dout("%s %s ns %.*s -> raw_pgid %llu.%x\n", __func__,
1925 oid->name, nsl, oloc->pool_ns->str,
1926 raw_pgid->pool, raw_pgid->seed);
1927 }
f24e9980
SW		 1928 return 0;
1929}
d9591f5e 1930EXPORT_SYMBOL(ceph_object_locator_to_pg);
f24e9980 1931
6f3bfd45
ID		 1932/*
1933 * Map a raw PG (full precision ps) into an actual PG.
1934 */
1935static void raw_pg_to_pg(struct ceph_pg_pool_info *pi,
1936 const struct ceph_pg *raw_pgid,
1937 struct ceph_pg *pgid)
1938{
1939 pgid->pool = raw_pgid->pool;
1940 pgid->seed = ceph_stable_mod(raw_pgid->seed, pi->pg_num,
1941 pi->pg_num_mask);
1942}
1943
1944/*
1945 * Map a raw PG (full precision ps) into a placement ps (placement
1946 * seed). Include pool id in that value so that different pools don't
1947 * use the same seeds.
1948 */
1949static u32 raw_pg_to_pps(struct ceph_pg_pool_info *pi,
1950 const struct ceph_pg *raw_pgid)
1951{
1952 if (pi->flags & CEPH_POOL_FLAG_HASHPSPOOL) {
1953 /* hash pool id and seed so that pool PGs do not overlap */
1954 return crush_hash32_2(CRUSH_HASH_RJENKINS1,
1955 ceph_stable_mod(raw_pgid->seed,
1956 pi->pgp_num,
1957 pi->pgp_num_mask),
1958 raw_pgid->pool);
1959 } else {
1960 /*
1961 * legacy behavior: add ps and pool together. this is
1962 * not a great approach because the PGs from each pool
1963 * will overlap on top of each other: 0.5 == 1.4 ==
1964 * 2.3 == ...
1965 */
1966 return ceph_stable_mod(raw_pgid->seed, pi->pgp_num,
1967 pi->pgp_num_mask) +
1968 (unsigned)raw_pgid->pool;
1969 }
1970}
1971
9d521470
ID		 1972static int do_crush(struct ceph_osdmap *map, int ruleno, int x,
1973 int *result, int result_max,
1974 const __u32 *weight, int weight_max)
e8ef19c4 1975{
9d521470
ID		 1976 int r;
1977
1978 BUG_ON(result_max > CEPH_PG_MAX_SIZE);
1979
743efcff 1980 mutex_lock(&map->crush_workspace_mutex);
9d521470 1981 r = crush_do_rule(map->crush, ruleno, x, result, result_max,
743efcff
ID		 1982 weight, weight_max, map->crush_workspace);
1983 mutex_unlock(&map->crush_workspace_mutex);
e8ef19c4 1984
9d521470 1985 return r;
e8ef19c4
ID		 1986}
1987
2bd93d4d 1988/*
6f3bfd45
ID		 1989 * Calculate raw set (CRUSH output) for given PG. The result may
1990 * contain nonexistent OSDs. ->primary is undefined for a raw set.
2bd93d4d 1991 *
6f3bfd45 1992 * Placement seed (CRUSH input) is returned through @ppps.
2bd93d4d 1993 */
6f3bfd45
ID		 1994static void pg_to_raw_osds(struct ceph_osdmap *osdmap,
1995 struct ceph_pg_pool_info *pi,
1996 const struct ceph_pg *raw_pgid,
1997 struct ceph_osds *raw,
1998 u32 *ppps)
2bd93d4d 1999{
6f3bfd45 2000 u32 pps = raw_pg_to_pps(pi, raw_pgid);
2bd93d4d
ID		 2001 int ruleno;
2002 int len;
2003
6f3bfd45
ID		 2004 ceph_osds_init(raw);
2005 if (ppps)
2006 *ppps = pps;
2007
2008 ruleno = crush_find_rule(osdmap->crush, pi->crush_ruleset, pi->type,
2009 pi->size);
2bd93d4d
ID		 2010 if (ruleno < 0) {
2011 pr_err("no crush rule: pool %lld ruleset %d type %d size %d\n",
6f3bfd45
ID		 2012 pi->id, pi->crush_ruleset, pi->type, pi->size);
2013 return;
2bd93d4d
ID		 2014 }
2015
ef9324bb
ID		 2016 if (pi->size > ARRAY_SIZE(raw->osds)) {
2017 pr_err_ratelimited("pool %lld ruleset %d type %d too wide: size %d > %zu\n",
2018 pi->id, pi->crush_ruleset, pi->type, pi->size,
2019 ARRAY_SIZE(raw->osds));
2020 return;
2021 }
2022
2023 len = do_crush(osdmap, ruleno, pps, raw->osds, pi->size,
2bd93d4d
ID		 2024 osdmap->osd_weight, osdmap->max_osd);
2025 if (len < 0) {
2026 pr_err("error %d from crush rule %d: pool %lld ruleset %d type %d size %d\n",
6f3bfd45
ID		 2027 len, ruleno, pi->id, pi->crush_ruleset, pi->type,
2028 pi->size);
2029 return;
2bd93d4d
ID		 2030 }
2031
6f3bfd45 2032 raw->size = len;
2bd93d4d
ID		 2033}
2034
2035/*
6f3bfd45
ID		 2036 * Given raw set, calculate up set and up primary. By definition of an
2037 * up set, the result won't contain nonexistent or down OSDs.
2bd93d4d 2038 *
6f3bfd45
ID		 2039 * This is done in-place - on return @set is the up set. If it's
2040 * empty, ->primary will remain undefined.
2bd93d4d 2041 */
6f3bfd45
ID		 2042static void raw_to_up_osds(struct ceph_osdmap *osdmap,
2043 struct ceph_pg_pool_info *pi,
2044 struct ceph_osds *set)
2bd93d4d 2045{
2bd93d4d
ID		 2046 int i;
2047
6f3bfd45
ID		 2048 /* ->primary is undefined for a raw set */
2049 BUG_ON(set->primary != -1);
2050
2051 if (ceph_can_shift_osds(pi)) {
2bd93d4d
ID		 2052 int removed = 0;
2053
6f3bfd45
ID		 2054 /* shift left */
2055 for (i = 0; i < set->size; i++) {
2056 if (ceph_osd_is_down(osdmap, set->osds[i])) {
2bd93d4d
ID		 2057 removed++;
2058 continue;
2059 }
2060 if (removed)
6f3bfd45 2061 set->osds[i - removed] = set->osds[i];
2bd93d4d 2062 }
6f3bfd45
ID		 2063 set->size -= removed;
2064 if (set->size > 0)
2065 set->primary = set->osds[0];
2bd93d4d 2066 } else {
6f3bfd45
ID		 2067 /* set down/dne devices to NONE */
2068 for (i = set->size - 1; i >= 0; i--) {
2069 if (ceph_osd_is_down(osdmap, set->osds[i]))
2070 set->osds[i] = CRUSH_ITEM_NONE;
2bd93d4d 2071 else
6f3bfd45 2072 set->primary = set->osds[i];
2bd93d4d
ID		 2073 }
2074 }
2bd93d4d
ID		 2075}
2076
6f3bfd45
ID		 2077static void apply_primary_affinity(struct ceph_osdmap *osdmap,
2078 struct ceph_pg_pool_info *pi,
2079 u32 pps,
2080 struct ceph_osds *up)
47ec1f3c
ID		 2081{
2082 int i;
2083 int pos = -1;
2084
2085 /*
2086 * Do we have any non-default primary_affinity values for these
2087 * osds?
2088 */
2089 if (!osdmap->osd_primary_affinity)
2090 return;
2091
6f3bfd45
ID		 2092 for (i = 0; i < up->size; i++) {
2093 int osd = up->osds[i];
92b2e751
ID		 2094
2095 if (osd != CRUSH_ITEM_NONE &&
2096 osdmap->osd_primary_affinity[osd] !=
47ec1f3c
ID		 2097 CEPH_OSD_DEFAULT_PRIMARY_AFFINITY) {
2098 break;
2099 }
2100 }
6f3bfd45 2101 if (i == up->size)
47ec1f3c
ID		 2102 return;
2103
2104 /*
2105 * Pick the primary. Feed both the seed (for the pg) and the
2106 * osd into the hash/rng so that a proportional fraction of an
2107 * osd's pgs get rejected as primary.
2108 */
6f3bfd45
ID		 2109 for (i = 0; i < up->size; i++) {
2110 int osd = up->osds[i];
47ec1f3c
ID		 2111 u32 aff;
2112
47ec1f3c
ID		 2113 if (osd == CRUSH_ITEM_NONE)
2114 continue;
2115
2116 aff = osdmap->osd_primary_affinity[osd];
2117 if (aff < CEPH_OSD_MAX_PRIMARY_AFFINITY &&
2118 (crush_hash32_2(CRUSH_HASH_RJENKINS1,
2119 pps, osd) >> 16) >= aff) {
2120 /*
2121 * We chose not to use this primary. Note it
2122 * anyway as a fallback in case we don't pick
2123 * anyone else, but keep looking.
2124 */
2125 if (pos < 0)
2126 pos = i;
2127 } else {
2128 pos = i;
2129 break;
2130 }
2131 }
2132 if (pos < 0)
2133 return;
2134
6f3bfd45 2135 up->primary = up->osds[pos];
47ec1f3c 2136
6f3bfd45 2137 if (ceph_can_shift_osds(pi) && pos > 0) {
47ec1f3c
ID		 2138 /* move the new primary to the front */
2139 for (i = pos; i > 0; i--)
6f3bfd45
ID		 2140 up->osds[i] = up->osds[i - 1];
2141 up->osds[0] = up->primary;
47ec1f3c
ID		 2142 }
2143}
2144
45966c34 2145/*
6f3bfd45 2146 * Get pg_temp and primary_temp mappings for given PG.
45966c34 2147 *
6f3bfd45
ID		 2148 * Note that a PG may have none, only pg_temp, only primary_temp or
2149 * both pg_temp and primary_temp mappings. This means @temp isn't
2150 * always a valid OSD set on return: in the "only primary_temp" case,
2151 * @temp will have its ->primary >= 0 but ->size == 0.
45966c34 2152 */
6f3bfd45
ID		 2153static void get_temp_osds(struct ceph_osdmap *osdmap,
2154 struct ceph_pg_pool_info *pi,
2155 const struct ceph_pg *raw_pgid,
2156 struct ceph_osds *temp)
45966c34 2157{
6f3bfd45 2158 struct ceph_pg pgid;
45966c34 2159 struct ceph_pg_mapping *pg;
45966c34
ID		 2160 int i;
2161
6f3bfd45
ID		 2162 raw_pg_to_pg(pi, raw_pgid, &pgid);
2163 ceph_osds_init(temp);
45966c34
ID		 2164
2165 /* pg_temp? */
2166 pg = __lookup_pg_mapping(&osdmap->pg_temp, pgid);
2167 if (pg) {
45966c34
ID		 2168 for (i = 0; i < pg->pg_temp.len; i++) {
2169 if (ceph_osd_is_down(osdmap, pg->pg_temp.osds[i])) {
6f3bfd45 2170 if (ceph_can_shift_osds(pi))
45966c34 2171 continue;
6f3bfd45
ID		 2172
2173 temp->osds[temp->size++] = CRUSH_ITEM_NONE;
45966c34 2174 } else {
6f3bfd45 2175 temp->osds[temp->size++] = pg->pg_temp.osds[i];
45966c34
ID		 2176 }
2177 }
2178
2179 /* apply pg_temp's primary */
6f3bfd45
ID		 2180 for (i = 0; i < temp->size; i++) {
2181 if (temp->osds[i] != CRUSH_ITEM_NONE) {
2182 temp->primary = temp->osds[i];
45966c34
ID		 2183 break;
2184 }
2185 }
45966c34
ID		 2186 }
2187
5e8d4d36
ID		 2188 /* primary_temp? */
2189 pg = __lookup_pg_mapping(&osdmap->primary_temp, pgid);
2190 if (pg)
6f3bfd45 2191 temp->primary = pg->primary_temp.osd;
45966c34
ID		 2192}
2193
d85b7056 2194/*
6f3bfd45 2195 * Map a PG to its acting set as well as its up set.
ac972230 2196 *
6f3bfd45
ID		 2197 * Acting set is used for data mapping purposes, while up set can be
2198 * recorded for detecting interval changes and deciding whether to
2199 * resend a request.
d85b7056 2200 */
6f3bfd45
ID		 2201void ceph_pg_to_up_acting_osds(struct ceph_osdmap *osdmap,
2202 const struct ceph_pg *raw_pgid,
2203 struct ceph_osds *up,
2204 struct ceph_osds *acting)
d85b7056 2205{
6f3bfd45 2206 struct ceph_pg_pool_info *pi;
ac972230 2207 u32 pps;
d85b7056 2208
6f3bfd45
ID		 2209 pi = ceph_pg_pool_by_id(osdmap, raw_pgid->pool);
2210 if (!pi) {
2211 ceph_osds_init(up);
2212 ceph_osds_init(acting);
2213 goto out;
8008ab10 2214 }
d85b7056 2215
6f3bfd45
ID		 2216 pg_to_raw_osds(osdmap, pi, raw_pgid, up, &pps);
2217 raw_to_up_osds(osdmap, pi, up);
2218 apply_primary_affinity(osdmap, pi, pps, up);
2219 get_temp_osds(osdmap, pi, raw_pgid, acting);
2220 if (!acting->size) {
2221 memcpy(acting->osds, up->osds, up->size * sizeof(up->osds[0]));
2222 acting->size = up->size;
2223 if (acting->primary == -1)
2224 acting->primary = up->primary;
8008ab10 2225 }
6f3bfd45
ID		 2226out:
2227 WARN_ON(!osds_valid(up) || !osds_valid(acting));
d85b7056
SW		 2228}
2229
f24e9980 2230/*
f81f1633 2231 * Return acting primary for given PG, or -1 if none.
f24e9980 2232 */
f81f1633
ID		 2233int ceph_pg_to_acting_primary(struct ceph_osdmap *osdmap,
2234 const struct ceph_pg *raw_pgid)
f24e9980 2235{
6f3bfd45 2236 struct ceph_osds up, acting;
f24e9980 2237
f81f1633 2238 ceph_pg_to_up_acting_osds(osdmap, raw_pgid, &up, &acting);
6f3bfd45 2239 return acting.primary;
f24e9980 2240}
f81f1633 2241EXPORT_SYMBOL(ceph_pg_to_acting_primary);
Ubuntu Artful kernel mirror