projects / linux-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
libceph: kill __{insert,lookup,remove}_pg_mapping()
[linux-block.git] / net / ceph / osdmap.c
CommitLineData
f24e9980 1
3d14c5d2 2#include <linux/ceph/ceph_debug.h>
5a0e3ad6 3
3d14c5d2 4#include <linux/module.h>
5a0e3ad6 5#include <linux/slab.h>
f24e9980
SW		 6#include <asm/div64.h>
7
3d14c5d2
YS		 8#include <linux/ceph/libceph.h>
9#include <linux/ceph/osdmap.h>
10#include <linux/ceph/decode.h>
11#include <linux/crush/hash.h>
12#include <linux/crush/mapper.h>
f24e9980
SW		 13
14char *ceph_osdmap_state_str(char *str, int len, int state)
15{
f24e9980 16 if (!len)
1ec3911d 17 return str;
f24e9980 18
1ec3911d
CD		 19 if ((state & CEPH_OSD_EXISTS) && (state & CEPH_OSD_UP))
20 snprintf(str, len, "exists, up");
21 else if (state & CEPH_OSD_EXISTS)
22 snprintf(str, len, "exists");
23 else if (state & CEPH_OSD_UP)
24 snprintf(str, len, "up");
25 else
f24e9980 26 snprintf(str, len, "doesn't exist");
1ec3911d 27
f24e9980
SW		 28 return str;
29}
30
31/* maps */
32
95c96174 33static int calc_bits_of(unsigned int t)
f24e9980
SW		 34{
35 int b = 0;
36 while (t) {
37 t = t >> 1;
38 b++;
39 }
40 return b;
41}
42
43/*
44 * the foo_mask is the smallest value 2^n-1 that is >= foo.
45 */
46static void calc_pg_masks(struct ceph_pg_pool_info *pi)
47{
4f6a7e5e
SW		 48 pi->pg_num_mask = (1 << calc_bits_of(pi->pg_num-1)) - 1;
49 pi->pgp_num_mask = (1 << calc_bits_of(pi->pgp_num-1)) - 1;
f24e9980
SW		 50}
51
52/*
53 * decode crush map
54 */
55static int crush_decode_uniform_bucket(void **p, void *end,
56 struct crush_bucket_uniform *b)
57{
58 dout("crush_decode_uniform_bucket %p to %p\n", *p, end);
59 ceph_decode_need(p, end, (1+b->h.size) * sizeof(u32), bad);
c89136ea 60 b->item_weight = ceph_decode_32(p);
f24e9980
SW		 61 return 0;
62bad:
63 return -EINVAL;
64}
65
66static int crush_decode_list_bucket(void **p, void *end,
67 struct crush_bucket_list *b)
68{
69 int j;
70 dout("crush_decode_list_bucket %p to %p\n", *p, end);
71 b->item_weights = kcalloc(b->h.size, sizeof(u32), GFP_NOFS);
72 if (b->item_weights == NULL)
73 return -ENOMEM;
74 b->sum_weights = kcalloc(b->h.size, sizeof(u32), GFP_NOFS);
75 if (b->sum_weights == NULL)
76 return -ENOMEM;
77 ceph_decode_need(p, end, 2 * b->h.size * sizeof(u32), bad);
78 for (j = 0; j < b->h.size; j++) {
c89136ea
SW		 79 b->item_weights[j] = ceph_decode_32(p);
80 b->sum_weights[j] = ceph_decode_32(p);
f24e9980
SW		 81 }
82 return 0;
83bad:
84 return -EINVAL;
85}
86
87static int crush_decode_tree_bucket(void **p, void *end,
88 struct crush_bucket_tree *b)
89{
90 int j;
91 dout("crush_decode_tree_bucket %p to %p\n", *p, end);
82cd003a 92 ceph_decode_8_safe(p, end, b->num_nodes, bad);
f24e9980
SW		 93 b->node_weights = kcalloc(b->num_nodes, sizeof(u32), GFP_NOFS);
94 if (b->node_weights == NULL)
95 return -ENOMEM;
96 ceph_decode_need(p, end, b->num_nodes * sizeof(u32), bad);
97 for (j = 0; j < b->num_nodes; j++)
c89136ea 98 b->node_weights[j] = ceph_decode_32(p);
f24e9980
SW		 99 return 0;
100bad:
101 return -EINVAL;
102}
103
104static int crush_decode_straw_bucket(void **p, void *end,
105 struct crush_bucket_straw *b)
106{
107 int j;
108 dout("crush_decode_straw_bucket %p to %p\n", *p, end);
109 b->item_weights = kcalloc(b->h.size, sizeof(u32), GFP_NOFS);
110 if (b->item_weights == NULL)
111 return -ENOMEM;
112 b->straws = kcalloc(b->h.size, sizeof(u32), GFP_NOFS);
113 if (b->straws == NULL)
114 return -ENOMEM;
115 ceph_decode_need(p, end, 2 * b->h.size * sizeof(u32), bad);
116 for (j = 0; j < b->h.size; j++) {
c89136ea
SW		 117 b->item_weights[j] = ceph_decode_32(p);
118 b->straws[j] = ceph_decode_32(p);
f24e9980
SW		 119 }
120 return 0;
121bad:
122 return -EINVAL;
123}
124
958a2765
ID		 125static int crush_decode_straw2_bucket(void **p, void *end,
126 struct crush_bucket_straw2 *b)
127{
128 int j;
129 dout("crush_decode_straw2_bucket %p to %p\n", *p, end);
130 b->item_weights = kcalloc(b->h.size, sizeof(u32), GFP_NOFS);
131 if (b->item_weights == NULL)
132 return -ENOMEM;
133 ceph_decode_need(p, end, b->h.size * sizeof(u32), bad);
134 for (j = 0; j < b->h.size; j++)
135 b->item_weights[j] = ceph_decode_32(p);
136 return 0;
137bad:
138 return -EINVAL;
139}
140
546f04ef
SW		 141static int skip_name_map(void **p, void *end)
142{
143 int len;
144 ceph_decode_32_safe(p, end, len ,bad);
145 while (len--) {
146 int strlen;
147 *p += sizeof(u32);
148 ceph_decode_32_safe(p, end, strlen, bad);
149 *p += strlen;
150}
151 return 0;
152bad:
153 return -EINVAL;
154}
155
66a0e2d5
ID		 156static void crush_finalize(struct crush_map *c)
157{
158 __s32 b;
159
160 /* Space for the array of pointers to per-bucket workspace */
161 c->working_size = sizeof(struct crush_work) +
162 c->max_buckets * sizeof(struct crush_work_bucket *);
163
164 for (b = 0; b < c->max_buckets; b++) {
165 if (!c->buckets[b])
166 continue;
167
168 switch (c->buckets[b]->alg) {
169 default:
170 /*
171 * The base case, permutation variables and
172 * the pointer to the permutation array.
173 */
174 c->working_size += sizeof(struct crush_work_bucket);
175 break;
176 }
177 /* Every bucket has a permutation array. */
178 c->working_size += c->buckets[b]->size * sizeof(__u32);
179 }
180}
181
f24e9980
SW		 182static struct crush_map *crush_decode(void *pbyval, void *end)
183{
184 struct crush_map *c;
185 int err = -EINVAL;
186 int i, j;
187 void **p = &pbyval;
188 void *start = pbyval;
189 u32 magic;
546f04ef 190 u32 num_name_maps;
f24e9980
SW		 191
192 dout("crush_decode %p to %p len %d\n", *p, end, (int)(end - *p));
193
194 c = kzalloc(sizeof(*c), GFP_NOFS);
195 if (c == NULL)
196 return ERR_PTR(-ENOMEM);
197
546f04ef
SW		 198 /* set tunables to default values */
199 c->choose_local_tries = 2;
200 c->choose_local_fallback_tries = 5;
201 c->choose_total_tries = 19;
1604f488 202 c->chooseleaf_descend_once = 0;
546f04ef 203
f24e9980 204 ceph_decode_need(p, end, 4*sizeof(u32), bad);
c89136ea 205 magic = ceph_decode_32(p);
f24e9980
SW		 206 if (magic != CRUSH_MAGIC) {
207 pr_err("crush_decode magic %x != current %x\n",
95c96174 208 (unsigned int)magic, (unsigned int)CRUSH_MAGIC);
f24e9980
SW		 209 goto bad;
210 }
c89136ea
SW		 211 c->max_buckets = ceph_decode_32(p);
212 c->max_rules = ceph_decode_32(p);
213 c->max_devices = ceph_decode_32(p);
f24e9980 214
f24e9980
SW		 215 c->buckets = kcalloc(c->max_buckets, sizeof(*c->buckets), GFP_NOFS);
216 if (c->buckets == NULL)
217 goto badmem;
218 c->rules = kcalloc(c->max_rules, sizeof(*c->rules), GFP_NOFS);
219 if (c->rules == NULL)
220 goto badmem;
221
222 /* buckets */
223 for (i = 0; i < c->max_buckets; i++) {
224 int size = 0;
225 u32 alg;
226 struct crush_bucket *b;
227
228 ceph_decode_32_safe(p, end, alg, bad);
229 if (alg == 0) {
230 c->buckets[i] = NULL;
231 continue;
232 }
233 dout("crush_decode bucket %d off %x %p to %p\n",
234 i, (int)(*p-start), *p, end);
235
236 switch (alg) {
237 case CRUSH_BUCKET_UNIFORM:
238 size = sizeof(struct crush_bucket_uniform);
239 break;
240 case CRUSH_BUCKET_LIST:
241 size = sizeof(struct crush_bucket_list);
242 break;
243 case CRUSH_BUCKET_TREE:
244 size = sizeof(struct crush_bucket_tree);
245 break;
246 case CRUSH_BUCKET_STRAW:
247 size = sizeof(struct crush_bucket_straw);
248 break;
958a2765
ID		 249 case CRUSH_BUCKET_STRAW2:
250 size = sizeof(struct crush_bucket_straw2);
251 break;
f24e9980 252 default:
30dc6381 253 err = -EINVAL;
f24e9980
SW		 254 goto bad;
255 }
256 BUG_ON(size == 0);
257 b = c->buckets[i] = kzalloc(size, GFP_NOFS);
258 if (b == NULL)
259 goto badmem;
260
261 ceph_decode_need(p, end, 4*sizeof(u32), bad);
c89136ea
SW		 262 b->id = ceph_decode_32(p);
263 b->type = ceph_decode_16(p);
fb690390
SW		 264 b->alg = ceph_decode_8(p);
265 b->hash = ceph_decode_8(p);
c89136ea
SW		 266 b->weight = ceph_decode_32(p);
267 b->size = ceph_decode_32(p);
f24e9980
SW		 268
269 dout("crush_decode bucket size %d off %x %p to %p\n",
270 b->size, (int)(*p-start), *p, end);
271
272 b->items = kcalloc(b->size, sizeof(__s32), GFP_NOFS);
273 if (b->items == NULL)
274 goto badmem;
f24e9980
SW		 275
276 ceph_decode_need(p, end, b->size*sizeof(u32), bad);
277 for (j = 0; j < b->size; j++)
c89136ea 278 b->items[j] = ceph_decode_32(p);
f24e9980
SW		 279
280 switch (b->alg) {
281 case CRUSH_BUCKET_UNIFORM:
282 err = crush_decode_uniform_bucket(p, end,
283 (struct crush_bucket_uniform *)b);
284 if (err < 0)
285 goto bad;
286 break;
287 case CRUSH_BUCKET_LIST:
288 err = crush_decode_list_bucket(p, end,
289 (struct crush_bucket_list *)b);
290 if (err < 0)
291 goto bad;
292 break;
293 case CRUSH_BUCKET_TREE:
294 err = crush_decode_tree_bucket(p, end,
295 (struct crush_bucket_tree *)b);
296 if (err < 0)
297 goto bad;
298 break;
299 case CRUSH_BUCKET_STRAW:
300 err = crush_decode_straw_bucket(p, end,
301 (struct crush_bucket_straw *)b);
302 if (err < 0)
303 goto bad;
304 break;
958a2765
ID		 305 case CRUSH_BUCKET_STRAW2:
306 err = crush_decode_straw2_bucket(p, end,
307 (struct crush_bucket_straw2 *)b);
308 if (err < 0)
309 goto bad;
310 break;
f24e9980
SW		 311 }
312 }
313
314 /* rules */
315 dout("rule vec is %p\n", c->rules);
316 for (i = 0; i < c->max_rules; i++) {
317 u32 yes;
318 struct crush_rule *r;
319
293dffaa 320 err = -EINVAL;
f24e9980
SW		 321 ceph_decode_32_safe(p, end, yes, bad);
322 if (!yes) {
323 dout("crush_decode NO rule %d off %x %p to %p\n",
324 i, (int)(*p-start), *p, end);
325 c->rules[i] = NULL;
326 continue;
327 }
328
329 dout("crush_decode rule %d off %x %p to %p\n",
330 i, (int)(*p-start), *p, end);
331
332 /* len */
333 ceph_decode_32_safe(p, end, yes, bad);
334#if BITS_PER_LONG == 32
30dc6381 335 err = -EINVAL;
64486697
XW		 336 if (yes > (ULONG_MAX - sizeof(*r))
337 / sizeof(struct crush_rule_step))
f24e9980
SW		 338 goto bad;
339#endif
340 r = c->rules[i] = kmalloc(sizeof(*r) +
341 yes*sizeof(struct crush_rule_step),
342 GFP_NOFS);
343 if (r == NULL)
344 goto badmem;
345 dout(" rule %d is at %p\n", i, r);
346 r->len = yes;
347 ceph_decode_copy_safe(p, end, &r->mask, 4, bad); /* 4 u8's */
348 ceph_decode_need(p, end, r->len*3*sizeof(u32), bad);
349 for (j = 0; j < r->len; j++) {
c89136ea
SW		 350 r->steps[j].op = ceph_decode_32(p);
351 r->steps[j].arg1 = ceph_decode_32(p);
352 r->steps[j].arg2 = ceph_decode_32(p);
f24e9980
SW		 353 }
354 }
355
356 /* ignore trailing name maps. */
546f04ef
SW		 357 for (num_name_maps = 0; num_name_maps < 3; num_name_maps++) {
358 err = skip_name_map(p, end);
359 if (err < 0)
360 goto done;
361 }
362
363 /* tunables */
364 ceph_decode_need(p, end, 3*sizeof(u32), done);
365 c->choose_local_tries = ceph_decode_32(p);
366 c->choose_local_fallback_tries = ceph_decode_32(p);
367 c->choose_total_tries = ceph_decode_32(p);
b9b519b7 368 dout("crush decode tunable choose_local_tries = %d\n",
546f04ef 369 c->choose_local_tries);
b9b519b7 370 dout("crush decode tunable choose_local_fallback_tries = %d\n",
546f04ef 371 c->choose_local_fallback_tries);
b9b519b7 372 dout("crush decode tunable choose_total_tries = %d\n",
546f04ef 373 c->choose_total_tries);
f24e9980 374
1604f488
JS		 375 ceph_decode_need(p, end, sizeof(u32), done);
376 c->chooseleaf_descend_once = ceph_decode_32(p);
b9b519b7 377 dout("crush decode tunable chooseleaf_descend_once = %d\n",
1604f488
JS		 378 c->chooseleaf_descend_once);
379
f140662f
ID		 380 ceph_decode_need(p, end, sizeof(u8), done);
381 c->chooseleaf_vary_r = ceph_decode_8(p);
b9b519b7 382 dout("crush decode tunable chooseleaf_vary_r = %d\n",
f140662f
ID		 383 c->chooseleaf_vary_r);
384
b9b519b7
ID		 385 /* skip straw_calc_version, allowed_bucket_algs */
386 ceph_decode_need(p, end, sizeof(u8) + sizeof(u32), done);
387 *p += sizeof(u8) + sizeof(u32);
388
389 ceph_decode_need(p, end, sizeof(u8), done);
390 c->chooseleaf_stable = ceph_decode_8(p);
391 dout("crush decode tunable chooseleaf_stable = %d\n",
392 c->chooseleaf_stable);
393
546f04ef 394done:
9afd30db 395 crush_finalize(c);
f24e9980
SW		 396 dout("crush_decode success\n");
397 return c;
398
399badmem:
400 err = -ENOMEM;
401bad:
402 dout("crush_decode fail %d\n", err);
403 crush_destroy(c);
404 return ERR_PTR(err);
405}
406
f984cb76 407int ceph_pg_compare(const struct ceph_pg *lhs, const struct ceph_pg *rhs)
51042122 408{
f984cb76 409 if (lhs->pool < rhs->pool)
5b191d99 410 return -1;
f984cb76 411 if (lhs->pool > rhs->pool)
5b191d99 412 return 1;
f984cb76 413 if (lhs->seed < rhs->seed)
51042122 414 return -1;
f984cb76 415 if (lhs->seed > rhs->seed)
51042122 416 return 1;
f984cb76 417
51042122
SW		 418 return 0;
419}
420
a02a946d
ID		 421int ceph_spg_compare(const struct ceph_spg *lhs, const struct ceph_spg *rhs)
422{
423 int ret;
424
425 ret = ceph_pg_compare(&lhs->pgid, &rhs->pgid);
426 if (ret)
427 return ret;
428
429 if (lhs->shard < rhs->shard)
430 return -1;
431 if (lhs->shard > rhs->shard)
432 return 1;
433
434 return 0;
435}
436
a303bb0e
ID		 437static struct ceph_pg_mapping *alloc_pg_mapping(size_t payload_len)
438{
439 struct ceph_pg_mapping *pg;
440
441 pg = kmalloc(sizeof(*pg) + payload_len, GFP_NOIO);
442 if (!pg)
443 return NULL;
444
445 RB_CLEAR_NODE(&pg->node);
446 return pg;
447}
448
449static void free_pg_mapping(struct ceph_pg_mapping *pg)
450{
451 WARN_ON(!RB_EMPTY_NODE(&pg->node));
452
453 kfree(pg);
454}
455
f984cb76
ID		 456/*
457 * rbtree of pg_mapping for handling pg_temp (explicit mapping of pgid
458 * to a set of osds) and primary_temp (explicit primary setting)
459 */
ab75144b
ID		 460DEFINE_RB_FUNCS2(pg_mapping, struct ceph_pg_mapping, pgid, ceph_pg_compare,
461 RB_BYPTR, const struct ceph_pg *, node)
8adc8b3d 462
4fc51be8
SW		 463/*
464 * rbtree of pg pool info
465 */
466static int __insert_pg_pool(struct rb_root *root, struct ceph_pg_pool_info *new)
467{
468 struct rb_node **p = &root->rb_node;
469 struct rb_node *parent = NULL;
470 struct ceph_pg_pool_info *pi = NULL;
471
472 while (*p) {
473 parent = *p;
474 pi = rb_entry(parent, struct ceph_pg_pool_info, node);
475 if (new->id < pi->id)
476 p = &(*p)->rb_left;
477 else if (new->id > pi->id)
478 p = &(*p)->rb_right;
479 else
480 return -EEXIST;
481 }
482
483 rb_link_node(&new->node, parent, p);
484 rb_insert_color(&new->node, root);
485 return 0;
486}
487
4f6a7e5e 488static struct ceph_pg_pool_info *__lookup_pg_pool(struct rb_root *root, u64 id)
4fc51be8
SW		 489{
490 struct ceph_pg_pool_info *pi;
491 struct rb_node *n = root->rb_node;
492
493 while (n) {
494 pi = rb_entry(n, struct ceph_pg_pool_info, node);
495 if (id < pi->id)
496 n = n->rb_left;
497 else if (id > pi->id)
498 n = n->rb_right;
499 else
500 return pi;
501 }
502 return NULL;
503}
504
ce7f6a27
ID		 505struct ceph_pg_pool_info *ceph_pg_pool_by_id(struct ceph_osdmap *map, u64 id)
506{
507 return __lookup_pg_pool(&map->pg_pools, id);
508}
509
72afc71f
AE		 510const char *ceph_pg_pool_name_by_id(struct ceph_osdmap *map, u64 id)
511{
512 struct ceph_pg_pool_info *pi;
513
514 if (id == CEPH_NOPOOL)
515 return NULL;
516
517 if (WARN_ON_ONCE(id > (u64) INT_MAX))
518 return NULL;
519
520 pi = __lookup_pg_pool(&map->pg_pools, (int) id);
521
522 return pi ? pi->name : NULL;
523}
524EXPORT_SYMBOL(ceph_pg_pool_name_by_id);
525
7669a2c9
YS		 526int ceph_pg_poolid_by_name(struct ceph_osdmap *map, const char *name)
527{
528 struct rb_node *rbp;
529
530 for (rbp = rb_first(&map->pg_pools); rbp; rbp = rb_next(rbp)) {
531 struct ceph_pg_pool_info *pi =
532 rb_entry(rbp, struct ceph_pg_pool_info, node);
533 if (pi->name && strcmp(pi->name, name) == 0)
534 return pi->id;
535 }
536 return -ENOENT;
537}
3d14c5d2 538EXPORT_SYMBOL(ceph_pg_poolid_by_name);
7669a2c9 539
2844a76a
SW		 540static void __remove_pg_pool(struct rb_root *root, struct ceph_pg_pool_info *pi)
541{
542 rb_erase(&pi->node, root);
543 kfree(pi->name);
544 kfree(pi);
545}
546
0f70c7ee 547static int decode_pool(void **p, void *end, struct ceph_pg_pool_info *pi)
efd7576b 548{
4f6a7e5e
SW		 549 u8 ev, cv;
550 unsigned len, num;
551 void *pool_end;
552
553 ceph_decode_need(p, end, 2 + 4, bad);
554 ev = ceph_decode_8(p); /* encoding version */
555 cv = ceph_decode_8(p); /* compat version */
556 if (ev < 5) {
b9a67899 557 pr_warn("got v %d < 5 cv %d of ceph_pg_pool\n", ev, cv);
4f6a7e5e
SW		 558 return -EINVAL;
559 }
17a13e40 560 if (cv > 9) {
b9a67899 561 pr_warn("got v %d cv %d > 9 of ceph_pg_pool\n", ev, cv);
4f6a7e5e
SW		 562 return -EINVAL;
563 }
564 len = ceph_decode_32(p);
565 ceph_decode_need(p, end, len, bad);
566 pool_end = *p + len;
73a7e693 567
4f6a7e5e
SW		 568 pi->type = ceph_decode_8(p);
569 pi->size = ceph_decode_8(p);
570 pi->crush_ruleset = ceph_decode_8(p);
571 pi->object_hash = ceph_decode_8(p);
73a7e693 572
4f6a7e5e
SW		 573 pi->pg_num = ceph_decode_32(p);
574 pi->pgp_num = ceph_decode_32(p);
575
576 *p += 4 + 4; /* skip lpg* */
577 *p += 4; /* skip last_change */
578 *p += 8 + 4; /* skip snap_seq, snap_epoch */
579
580 /* skip snaps */
581 num = ceph_decode_32(p);
582 while (num--) {
583 *p += 8; /* snapid key */
584 *p += 1 + 1; /* versions */
585 len = ceph_decode_32(p);
586 *p += len;
73a7e693
SW		 587 }
588
17a13e40 589 /* skip removed_snaps */
4f6a7e5e
SW		 590 num = ceph_decode_32(p);
591 *p += num * (8 + 8);
592
593 *p += 8; /* skip auid */
594 pi->flags = ceph_decode_64(p);
17a13e40
ID		 595 *p += 4; /* skip crash_replay_interval */
596
597 if (ev >= 7)
04812acf
ID		 598 pi->min_size = ceph_decode_8(p);
599 else
600 pi->min_size = pi->size - pi->size / 2;
17a13e40
ID		 601
602 if (ev >= 8)
603 *p += 8 + 8; /* skip quota_max_* */
604
605 if (ev >= 9) {
606 /* skip tiers */
607 num = ceph_decode_32(p);
608 *p += num * 8;
609
610 *p += 8; /* skip tier_of */
611 *p += 1; /* skip cache_mode */
612
613 pi->read_tier = ceph_decode_64(p);
614 pi->write_tier = ceph_decode_64(p);
615 } else {
616 pi->read_tier = -1;
617 pi->write_tier = -1;
618 }
4f6a7e5e 619
04812acf
ID		 620 if (ev >= 10) {
621 /* skip properties */
622 num = ceph_decode_32(p);
623 while (num--) {
624 len = ceph_decode_32(p);
625 *p += len; /* key */
626 len = ceph_decode_32(p);
627 *p += len; /* val */
628 }
629 }
630
631 if (ev >= 11) {
632 /* skip hit_set_params */
633 *p += 1 + 1; /* versions */
634 len = ceph_decode_32(p);
635 *p += len;
636
637 *p += 4; /* skip hit_set_period */
638 *p += 4; /* skip hit_set_count */
639 }
640
641 if (ev >= 12)
642 *p += 4; /* skip stripe_width */
643
644 if (ev >= 13) {
645 *p += 8; /* skip target_max_bytes */
646 *p += 8; /* skip target_max_objects */
647 *p += 4; /* skip cache_target_dirty_ratio_micro */
648 *p += 4; /* skip cache_target_full_ratio_micro */
649 *p += 4; /* skip cache_min_flush_age */
650 *p += 4; /* skip cache_min_evict_age */
651 }
652
653 if (ev >= 14) {
654 /* skip erasure_code_profile */
655 len = ceph_decode_32(p);
656 *p += len;
657 }
658
8e48cf00
ID		 659 /*
660 * last_force_op_resend_preluminous, will be overridden if the
661 * map was encoded with RESEND_ON_SPLIT
662 */
04812acf
ID		 663 if (ev >= 15)
664 pi->last_force_request_resend = ceph_decode_32(p);
665 else
666 pi->last_force_request_resend = 0;
667
8e48cf00
ID		 668 if (ev >= 16)
669 *p += 4; /* skip min_read_recency_for_promote */
670
671 if (ev >= 17)
672 *p += 8; /* skip expected_num_objects */
673
674 if (ev >= 19)
675 *p += 4; /* skip cache_target_dirty_high_ratio_micro */
676
677 if (ev >= 20)
678 *p += 4; /* skip min_write_recency_for_promote */
679
680 if (ev >= 21)
681 *p += 1; /* skip use_gmt_hitset */
682
683 if (ev >= 22)
684 *p += 1; /* skip fast_read */
685
686 if (ev >= 23) {
687 *p += 4; /* skip hit_set_grade_decay_rate */
688 *p += 4; /* skip hit_set_search_last_n */
689 }
690
691 if (ev >= 24) {
692 /* skip opts */
693 *p += 1 + 1; /* versions */
694 len = ceph_decode_32(p);
695 *p += len;
696 }
697
698 if (ev >= 25)
699 pi->last_force_request_resend = ceph_decode_32(p);
700
4f6a7e5e
SW		 701 /* ignore the rest */
702
703 *p = pool_end;
704 calc_pg_masks(pi);
73a7e693
SW		 705 return 0;
706
707bad:
708 return -EINVAL;
efd7576b
SW		 709}
710
0f70c7ee 711static int decode_pool_names(void **p, void *end, struct ceph_osdmap *map)
2844a76a
SW		 712{
713 struct ceph_pg_pool_info *pi;
4f6a7e5e
SW		 714 u32 num, len;
715 u64 pool;
2844a76a
SW		 716
717 ceph_decode_32_safe(p, end, num, bad);
718 dout(" %d pool names\n", num);
719 while (num--) {
4f6a7e5e 720 ceph_decode_64_safe(p, end, pool, bad);
2844a76a 721 ceph_decode_32_safe(p, end, len, bad);
4f6a7e5e 722 dout(" pool %llu len %d\n", pool, len);
ad3b904c 723 ceph_decode_need(p, end, len, bad);
2844a76a
SW		 724 pi = __lookup_pg_pool(&map->pg_pools, pool);
725 if (pi) {
ad3b904c
XW		 726 char *name = kstrndup(*p, len, GFP_NOFS);
727
728 if (!name)
729 return -ENOMEM;
2844a76a 730 kfree(pi->name);
ad3b904c
XW		 731 pi->name = name;
732 dout(" name is %s\n", pi->name);
2844a76a
SW		 733 }
734 *p += len;
735 }
736 return 0;
737
738bad:
739 return -EINVAL;
740}
741
742/*
743 * osd map
744 */
e5253a7b
ID		 745struct ceph_osdmap *ceph_osdmap_alloc(void)
746{
747 struct ceph_osdmap *map;
748
749 map = kzalloc(sizeof(*map), GFP_NOIO);
750 if (!map)
751 return NULL;
752
753 map->pg_pools = RB_ROOT;
754 map->pool_max = -1;
755 map->pg_temp = RB_ROOT;
756 map->primary_temp = RB_ROOT;
743efcff 757 mutex_init(&map->crush_workspace_mutex);
e5253a7b
ID		 758
759 return map;
760}
761
2844a76a
SW		 762void ceph_osdmap_destroy(struct ceph_osdmap *map)
763{
764 dout("osdmap_destroy %p\n", map);
765 if (map->crush)
766 crush_destroy(map->crush);
767 while (!RB_EMPTY_ROOT(&map->pg_temp)) {
768 struct ceph_pg_mapping *pg =
769 rb_entry(rb_first(&map->pg_temp),
770 struct ceph_pg_mapping, node);
ab75144b
ID		 771 erase_pg_mapping(&map->pg_temp, pg);
772 free_pg_mapping(pg);
2844a76a 773 }
9686f94c
ID		 774 while (!RB_EMPTY_ROOT(&map->primary_temp)) {
775 struct ceph_pg_mapping *pg =
776 rb_entry(rb_first(&map->primary_temp),
777 struct ceph_pg_mapping, node);
ab75144b
ID		 778 erase_pg_mapping(&map->primary_temp, pg);
779 free_pg_mapping(pg);
9686f94c 780 }
2844a76a
SW		 781 while (!RB_EMPTY_ROOT(&map->pg_pools)) {
782 struct ceph_pg_pool_info *pi =
783 rb_entry(rb_first(&map->pg_pools),
784 struct ceph_pg_pool_info, node);
785 __remove_pg_pool(&map->pg_pools, pi);
786 }
787 kfree(map->osd_state);
788 kfree(map->osd_weight);
789 kfree(map->osd_addr);
2cfa34f2 790 kfree(map->osd_primary_affinity);
66a0e2d5 791 kfree(map->crush_workspace);
2844a76a
SW		 792 kfree(map);
793}
794
795/*
4d60351f
ID		 796 * Adjust max_osd value, (re)allocate arrays.
797 *
798 * The new elements are properly initialized.
2844a76a
SW		 799 */
800static int osdmap_set_max_osd(struct ceph_osdmap *map, int max)
801{
802 u8 *state;
2844a76a 803 u32 *weight;
4d60351f
ID		 804 struct ceph_entity_addr *addr;
805 int i;
2844a76a 806
4d60351f 807 state = krealloc(map->osd_state, max*sizeof(*state), GFP_NOFS);
589506f1
LR		 808 if (!state)
809 return -ENOMEM;
810 map->osd_state = state;
811
4d60351f 812 weight = krealloc(map->osd_weight, max*sizeof(*weight), GFP_NOFS);
589506f1
LR		 813 if (!weight)
814 return -ENOMEM;
815 map->osd_weight = weight;
4d60351f 816
589506f1
LR		 817 addr = krealloc(map->osd_addr, max*sizeof(*addr), GFP_NOFS);
818 if (!addr)
2844a76a 819 return -ENOMEM;
589506f1 820 map->osd_addr = addr;
2844a76a 821
4d60351f 822 for (i = map->max_osd; i < max; i++) {
589506f1
LR		 823 map->osd_state[i] = 0;
824 map->osd_weight[i] = CEPH_OSD_OUT;
825 memset(map->osd_addr + i, 0, sizeof(*map->osd_addr));
2844a76a
SW		 826 }
827
2cfa34f2
ID		 828 if (map->osd_primary_affinity) {
829 u32 *affinity;
830
831 affinity = krealloc(map->osd_primary_affinity,
832 max*sizeof(*affinity), GFP_NOFS);
833 if (!affinity)
834 return -ENOMEM;
589506f1 835 map->osd_primary_affinity = affinity;
2cfa34f2
ID		 836
837 for (i = map->max_osd; i < max; i++)
589506f1
LR		 838 map->osd_primary_affinity[i] =
839 CEPH_OSD_DEFAULT_PRIMARY_AFFINITY;
2cfa34f2
ID		 840 }
841
2844a76a 842 map->max_osd = max;
4d60351f 843
2844a76a
SW		 844 return 0;
845}
846
1b6a78b5
ID		 847static int osdmap_set_crush(struct ceph_osdmap *map, struct crush_map *crush)
848{
66a0e2d5 849 void *workspace;
743efcff 850 size_t work_size;
66a0e2d5 851
1b6a78b5
ID		 852 if (IS_ERR(crush))
853 return PTR_ERR(crush);
854
743efcff
ID		 855 work_size = crush_work_size(crush, CEPH_PG_MAX_SIZE);
856 dout("%s work_size %zu bytes\n", __func__, work_size);
857 workspace = kmalloc(work_size, GFP_NOIO);
66a0e2d5
ID		 858 if (!workspace) {
859 crush_destroy(crush);
860 return -ENOMEM;
861 }
862 crush_init_workspace(crush, workspace);
863
1b6a78b5
ID		 864 if (map->crush)
865 crush_destroy(map->crush);
66a0e2d5 866 kfree(map->crush_workspace);
1b6a78b5 867 map->crush = crush;
66a0e2d5 868 map->crush_workspace = workspace;
1b6a78b5
ID		 869 return 0;
870}
871
ec7af972
ID		 872#define OSDMAP_WRAPPER_COMPAT_VER 7
873#define OSDMAP_CLIENT_DATA_COMPAT_VER 1
874
875/*
876 * Return 0 or error. On success, *v is set to 0 for old (v6) osdmaps,
877 * to struct_v of the client_data section for new (v7 and above)
878 * osdmaps.
879 */
880static int get_osdmap_client_data_v(void **p, void *end,
881 const char *prefix, u8 *v)
882{
883 u8 struct_v;
884
885 ceph_decode_8_safe(p, end, struct_v, e_inval);
886 if (struct_v >= 7) {
887 u8 struct_compat;
888
889 ceph_decode_8_safe(p, end, struct_compat, e_inval);
890 if (struct_compat > OSDMAP_WRAPPER_COMPAT_VER) {
b9a67899
JP		 891 pr_warn("got v %d cv %d > %d of %s ceph_osdmap\n",
892 struct_v, struct_compat,
893 OSDMAP_WRAPPER_COMPAT_VER, prefix);
ec7af972
ID		 894 return -EINVAL;
895 }
896 *p += 4; /* ignore wrapper struct_len */
897
898 ceph_decode_8_safe(p, end, struct_v, e_inval);
899 ceph_decode_8_safe(p, end, struct_compat, e_inval);
900 if (struct_compat > OSDMAP_CLIENT_DATA_COMPAT_VER) {
b9a67899
JP		 901 pr_warn("got v %d cv %d > %d of %s ceph_osdmap client data\n",
902 struct_v, struct_compat,
903 OSDMAP_CLIENT_DATA_COMPAT_VER, prefix);
ec7af972
ID		 904 return -EINVAL;
905 }
906 *p += 4; /* ignore client data struct_len */
907 } else {
908 u16 version;
909
910 *p -= 1;
911 ceph_decode_16_safe(p, end, version, e_inval);
912 if (version < 6) {
b9a67899
JP		 913 pr_warn("got v %d < 6 of %s ceph_osdmap\n",
914 version, prefix);
ec7af972
ID		 915 return -EINVAL;
916 }
917
918 /* old osdmap enconding */
919 struct_v = 0;
920 }
921
922 *v = struct_v;
923 return 0;
924
925e_inval:
926 return -EINVAL;
927}
928
433fbdd3
ID		 929static int __decode_pools(void **p, void *end, struct ceph_osdmap *map,
930 bool incremental)
931{
932 u32 n;
933
934 ceph_decode_32_safe(p, end, n, e_inval);
935 while (n--) {
936 struct ceph_pg_pool_info *pi;
937 u64 pool;
938 int ret;
939
940 ceph_decode_64_safe(p, end, pool, e_inval);
941
942 pi = __lookup_pg_pool(&map->pg_pools, pool);
943 if (!incremental || !pi) {
944 pi = kzalloc(sizeof(*pi), GFP_NOFS);
945 if (!pi)
946 return -ENOMEM;
947
948 pi->id = pool;
949
950 ret = __insert_pg_pool(&map->pg_pools, pi);
951 if (ret) {
952 kfree(pi);
953 return ret;
954 }
955 }
956
957 ret = decode_pool(p, end, pi);
958 if (ret)
959 return ret;
960 }
961
962 return 0;
963
964e_inval:
965 return -EINVAL;
966}
967
968static int decode_pools(void **p, void *end, struct ceph_osdmap *map)
969{
970 return __decode_pools(p, end, map, false);
971}
972
973static int decode_new_pools(void **p, void *end, struct ceph_osdmap *map)
974{
975 return __decode_pools(p, end, map, true);
976}
977
a303bb0e
ID		 978typedef struct ceph_pg_mapping *(*decode_mapping_fn_t)(void **, void *, bool);
979
980static int decode_pg_mapping(void **p, void *end, struct rb_root *mapping_root,
981 decode_mapping_fn_t fn, bool incremental)
10db634e
ID		 982{
983 u32 n;
984
a303bb0e
ID		 985 WARN_ON(!incremental && !fn);
986
10db634e
ID		 987 ceph_decode_32_safe(p, end, n, e_inval);
988 while (n--) {
a303bb0e 989 struct ceph_pg_mapping *pg;
10db634e 990 struct ceph_pg pgid;
10db634e
ID		 991 int ret;
992
993 ret = ceph_decode_pgid(p, end, &pgid);
994 if (ret)
995 return ret;
996
ab75144b
ID		 997 pg = lookup_pg_mapping(mapping_root, &pgid);
998 if (pg) {
999 WARN_ON(!incremental);
1000 erase_pg_mapping(mapping_root, pg);
1001 free_pg_mapping(pg);
1002 }
10db634e 1003
a303bb0e
ID		 1004 if (fn) {
1005 pg = fn(p, end, incremental);
1006 if (IS_ERR(pg))
1007 return PTR_ERR(pg);
10db634e 1008
a303bb0e
ID		 1009 if (pg) {
1010 pg->pgid = pgid; /* struct */
ab75144b 1011 insert_pg_mapping(mapping_root, pg);
10db634e
ID		 1012 }
1013 }
1014 }
1015
1016 return 0;
1017
1018e_inval:
1019 return -EINVAL;
1020}
1021
a303bb0e
ID		 1022static struct ceph_pg_mapping *__decode_pg_temp(void **p, void *end,
1023 bool incremental)
10db634e 1024{
a303bb0e
ID		 1025 struct ceph_pg_mapping *pg;
1026 u32 len, i;
10db634e 1027
a303bb0e
ID		 1028 ceph_decode_32_safe(p, end, len, e_inval);
1029 if (len == 0 && incremental)
1030 return NULL; /* new_pg_temp: [] to remove */
1031 if (len > (SIZE_MAX - sizeof(*pg)) / sizeof(u32))
1032 return ERR_PTR(-EINVAL);
10db634e 1033
a303bb0e
ID		 1034 ceph_decode_need(p, end, len * sizeof(u32), e_inval);
1035 pg = alloc_pg_mapping(len * sizeof(u32));
1036 if (!pg)
1037 return ERR_PTR(-ENOMEM);
d286de79 1038
a303bb0e
ID		 1039 pg->pg_temp.len = len;
1040 for (i = 0; i < len; i++)
1041 pg->pg_temp.osds[i] = ceph_decode_32(p);
d286de79 1042
a303bb0e 1043 return pg;
d286de79 1044
a303bb0e
ID		 1045e_inval:
1046 return ERR_PTR(-EINVAL);
1047}
d286de79 1048
a303bb0e
ID		 1049static int decode_pg_temp(void **p, void *end, struct ceph_osdmap *map)
1050{
1051 return decode_pg_mapping(p, end, &map->pg_temp, __decode_pg_temp,
1052 false);
1053}
d286de79 1054
a303bb0e
ID		 1055static int decode_new_pg_temp(void **p, void *end, struct ceph_osdmap *map)
1056{
1057 return decode_pg_mapping(p, end, &map->pg_temp, __decode_pg_temp,
1058 true);
1059}
d286de79 1060
a303bb0e
ID		 1061static struct ceph_pg_mapping *__decode_primary_temp(void **p, void *end,
1062 bool incremental)
1063{
1064 struct ceph_pg_mapping *pg;
1065 u32 osd;
d286de79 1066
a303bb0e
ID		 1067 ceph_decode_32_safe(p, end, osd, e_inval);
1068 if (osd == (u32)-1 && incremental)
1069 return NULL; /* new_primary_temp: -1 to remove */
d286de79 1070
a303bb0e
ID		 1071 pg = alloc_pg_mapping(0);
1072 if (!pg)
1073 return ERR_PTR(-ENOMEM);
d286de79 1074
a303bb0e
ID		 1075 pg->primary_temp.osd = osd;
1076 return pg;
d286de79
ID		 1077
1078e_inval:
a303bb0e 1079 return ERR_PTR(-EINVAL);
d286de79
ID		 1080}
1081
1082static int decode_primary_temp(void **p, void *end, struct ceph_osdmap *map)
1083{
a303bb0e
ID		 1084 return decode_pg_mapping(p, end, &map->primary_temp,
1085 __decode_primary_temp, false);
d286de79
ID		 1086}
1087
1088static int decode_new_primary_temp(void **p, void *end,
1089 struct ceph_osdmap *map)
1090{
a303bb0e
ID		 1091 return decode_pg_mapping(p, end, &map->primary_temp,
1092 __decode_primary_temp, true);
d286de79
ID		 1093}
1094
2cfa34f2
ID		 1095u32 ceph_get_primary_affinity(struct ceph_osdmap *map, int osd)
1096{
1097 BUG_ON(osd >= map->max_osd);
1098
1099 if (!map->osd_primary_affinity)
1100 return CEPH_OSD_DEFAULT_PRIMARY_AFFINITY;
1101
1102 return map->osd_primary_affinity[osd];
1103}
1104
1105static int set_primary_affinity(struct ceph_osdmap *map, int osd, u32 aff)
1106{
1107 BUG_ON(osd >= map->max_osd);
1108
1109 if (!map->osd_primary_affinity) {
1110 int i;
1111
1112 map->osd_primary_affinity = kmalloc(map->max_osd*sizeof(u32),
1113 GFP_NOFS);
1114 if (!map->osd_primary_affinity)
1115 return -ENOMEM;
1116
1117 for (i = 0; i < map->max_osd; i++)
1118 map->osd_primary_affinity[i] =
1119 CEPH_OSD_DEFAULT_PRIMARY_AFFINITY;
1120 }
1121
1122 map->osd_primary_affinity[osd] = aff;
1123
1124 return 0;
1125}
1126
63a6993f
ID		 1127static int decode_primary_affinity(void **p, void *end,
1128 struct ceph_osdmap *map)
1129{
1130 u32 len, i;
1131
1132 ceph_decode_32_safe(p, end, len, e_inval);
1133 if (len == 0) {
1134 kfree(map->osd_primary_affinity);
1135 map->osd_primary_affinity = NULL;
1136 return 0;
1137 }
1138 if (len != map->max_osd)
1139 goto e_inval;
1140
1141 ceph_decode_need(p, end, map->max_osd*sizeof(u32), e_inval);
1142
1143 for (i = 0; i < map->max_osd; i++) {
1144 int ret;
1145
1146 ret = set_primary_affinity(map, i, ceph_decode_32(p));
1147 if (ret)
1148 return ret;
1149 }
1150
1151 return 0;
1152
1153e_inval:
1154 return -EINVAL;
1155}
1156
1157static int decode_new_primary_affinity(void **p, void *end,
1158 struct ceph_osdmap *map)
1159{
1160 u32 n;
1161
1162 ceph_decode_32_safe(p, end, n, e_inval);
1163 while (n--) {
1164 u32 osd, aff;
1165 int ret;
1166
1167 ceph_decode_32_safe(p, end, osd, e_inval);
1168 ceph_decode_32_safe(p, end, aff, e_inval);
1169
1170 ret = set_primary_affinity(map, osd, aff);
1171 if (ret)
1172 return ret;
f31da0f3
ID		 1173
1174 pr_info("osd%d primary-affinity 0x%x\n", osd, aff);
63a6993f
ID		 1175 }
1176
1177 return 0;
1178
1179e_inval:
1180 return -EINVAL;
1181}
1182
f24e9980
SW		 1183/*
1184 * decode a full map.
1185 */
a2505d63 1186static int osdmap_decode(void **p, void *end, struct ceph_osdmap *map)
f24e9980 1187{
ec7af972 1188 u8 struct_v;
38a8d560 1189 u32 epoch = 0;
f24e9980 1190 void *start = *p;
3977058c
ID		 1191 u32 max;
1192 u32 len, i;
597b52f6 1193 int err;
f24e9980 1194
a2505d63 1195 dout("%s %p to %p len %d\n", __func__, *p, end, (int)(end - *p));
f24e9980 1196
ec7af972
ID		 1197 err = get_osdmap_client_data_v(p, end, "full", &struct_v);
1198 if (err)
1199 goto bad;
f24e9980 1200
53bbaba9
ID		 1201 /* fsid, epoch, created, modified */
1202 ceph_decode_need(p, end, sizeof(map->fsid) + sizeof(u32) +
1203 sizeof(map->created) + sizeof(map->modified), e_inval);
f24e9980 1204 ceph_decode_copy(p, &map->fsid, sizeof(map->fsid));
38a8d560 1205 epoch = map->epoch = ceph_decode_32(p);
f24e9980
SW		 1206 ceph_decode_copy(p, &map->created, sizeof(map->created));
1207 ceph_decode_copy(p, &map->modified, sizeof(map->modified));
1208
433fbdd3
ID		 1209 /* pools */
1210 err = decode_pools(p, end, map);
1211 if (err)
1212 goto bad;
2844a76a 1213
0f70c7ee
ID		 1214 /* pool_name */
1215 err = decode_pool_names(p, end, map);
597b52f6 1216 if (err)
4f6a7e5e 1217 goto bad;
2844a76a 1218
597b52f6 1219 ceph_decode_32_safe(p, end, map->pool_max, e_inval);
f24e9980 1220
597b52f6 1221 ceph_decode_32_safe(p, end, map->flags, e_inval);
f24e9980 1222
3977058c
ID		 1223 /* max_osd */
1224 ceph_decode_32_safe(p, end, max, e_inval);
f24e9980
SW		 1225
1226 /* (re)alloc osd arrays */
1227 err = osdmap_set_max_osd(map, max);
597b52f6 1228 if (err)
f24e9980 1229 goto bad;
f24e9980 1230
2d88b2e0 1231 /* osd_state, osd_weight, osd_addrs->client_addr */
f24e9980
SW		 1232 ceph_decode_need(p, end, 3*sizeof(u32) +
1233 map->max_osd*(1 + sizeof(*map->osd_weight) +
597b52f6
ID		 1234 sizeof(*map->osd_addr)), e_inval);
1235
2d88b2e0
ID		 1236 if (ceph_decode_32(p) != map->max_osd)
1237 goto e_inval;
1238
f24e9980
SW		 1239 ceph_decode_copy(p, map->osd_state, map->max_osd);
1240
2d88b2e0
ID		 1241 if (ceph_decode_32(p) != map->max_osd)
1242 goto e_inval;
1243
f24e9980 1244 for (i = 0; i < map->max_osd; i++)
c89136ea 1245 map->osd_weight[i] = ceph_decode_32(p);
f24e9980 1246
2d88b2e0
ID		 1247 if (ceph_decode_32(p) != map->max_osd)
1248 goto e_inval;
1249
f24e9980 1250 ceph_decode_copy(p, map->osd_addr, map->max_osd*sizeof(*map->osd_addr));
63f2d211
SW		 1251 for (i = 0; i < map->max_osd; i++)
1252 ceph_decode_addr(&map->osd_addr[i]);
f24e9980
SW		 1253
1254 /* pg_temp */
10db634e
ID		 1255 err = decode_pg_temp(p, end, map);
1256 if (err)
1257 goto bad;
f24e9980 1258
d286de79
ID		 1259 /* primary_temp */
1260 if (struct_v >= 1) {
1261 err = decode_primary_temp(p, end, map);
1262 if (err)
1263 goto bad;
1264 }
1265
63a6993f
ID		 1266 /* primary_affinity */
1267 if (struct_v >= 2) {
1268 err = decode_primary_affinity(p, end, map);
1269 if (err)
1270 goto bad;
1271 } else {
1272 /* XXX can this happen? */
1273 kfree(map->osd_primary_affinity);
1274 map->osd_primary_affinity = NULL;
1275 }
1276
f24e9980 1277 /* crush */
597b52f6 1278 ceph_decode_32_safe(p, end, len, e_inval);
1b6a78b5
ID		 1279 err = osdmap_set_crush(map, crush_decode(*p, min(*p + len, end)));
1280 if (err)
f24e9980 1281 goto bad;
f24e9980 1282
38a8d560 1283 /* ignore the rest */
f24e9980
SW		 1284 *p = end;
1285
38a8d560 1286 dout("full osdmap epoch %d max_osd %d\n", map->epoch, map->max_osd);
a2505d63 1287 return 0;
f24e9980 1288
597b52f6
ID		 1289e_inval:
1290 err = -EINVAL;
f24e9980 1291bad:
38a8d560
ID		 1292 pr_err("corrupt full osdmap (%d) epoch %d off %d (%p of %p-%p)\n",
1293 err, epoch, (int)(*p - start), *p, start, end);
1294 print_hex_dump(KERN_DEBUG, "osdmap: ",
1295 DUMP_PREFIX_OFFSET, 16, 1,
1296 start, end - start, true);
a2505d63
ID		 1297 return err;
1298}
1299
1300/*
1301 * Allocate and decode a full map.
1302 */
1303struct ceph_osdmap *ceph_osdmap_decode(void **p, void *end)
1304{
1305 struct ceph_osdmap *map;
1306 int ret;
1307
e5253a7b 1308 map = ceph_osdmap_alloc();
a2505d63
ID		 1309 if (!map)
1310 return ERR_PTR(-ENOMEM);
1311
a2505d63
ID		 1312 ret = osdmap_decode(p, end, map);
1313 if (ret) {
1314 ceph_osdmap_destroy(map);
1315 return ERR_PTR(ret);
1316 }
1317
1318 return map;
f24e9980
SW		 1319}
1320
930c5328
ID		 1321/*
1322 * Encoding order is (new_up_client, new_state, new_weight). Need to
1323 * apply in the (new_weight, new_state, new_up_client) order, because
1324 * an incremental map may look like e.g.
1325 *
1326 * new_up_client: { osd=6, addr=... } # set osd_state and addr
1327 * new_state: { osd=6, xorstate=EXISTS } # clear osd_state
1328 */
1329static int decode_new_up_state_weight(void **p, void *end,
1330 struct ceph_osdmap *map)
1331{
1332 void *new_up_client;
1333 void *new_state;
1334 void *new_weight_end;
1335 u32 len;
1336
1337 new_up_client = *p;
1338 ceph_decode_32_safe(p, end, len, e_inval);
1339 len *= sizeof(u32) + sizeof(struct ceph_entity_addr);
1340 ceph_decode_need(p, end, len, e_inval);
1341 *p += len;
1342
1343 new_state = *p;
1344 ceph_decode_32_safe(p, end, len, e_inval);
1345 len *= sizeof(u32) + sizeof(u8);
1346 ceph_decode_need(p, end, len, e_inval);
1347 *p += len;
1348
1349 /* new_weight */
1350 ceph_decode_32_safe(p, end, len, e_inval);
1351 while (len--) {
1352 s32 osd;
1353 u32 w;
1354
1355 ceph_decode_need(p, end, 2*sizeof(u32), e_inval);
1356 osd = ceph_decode_32(p);
1357 w = ceph_decode_32(p);
1358 BUG_ON(osd >= map->max_osd);
1359 pr_info("osd%d weight 0x%x %s\n", osd, w,
1360 w == CEPH_OSD_IN ? "(in)" :
1361 (w == CEPH_OSD_OUT ? "(out)" : ""));
1362 map->osd_weight[osd] = w;
1363
1364 /*
1365 * If we are marking in, set the EXISTS, and clear the
1366 * AUTOOUT and NEW bits.
1367 */
1368 if (w) {
1369 map->osd_state[osd] |= CEPH_OSD_EXISTS;
1370 map->osd_state[osd] &= ~(CEPH_OSD_AUTOOUT |
1371 CEPH_OSD_NEW);
1372 }
1373 }
1374 new_weight_end = *p;
1375
1376 /* new_state (up/down) */
1377 *p = new_state;
1378 len = ceph_decode_32(p);
1379 while (len--) {
1380 s32 osd;
1381 u8 xorstate;
1382 int ret;
1383
1384 osd = ceph_decode_32(p);
1385 xorstate = ceph_decode_8(p);
1386 if (xorstate == 0)
1387 xorstate = CEPH_OSD_UP;
1388 BUG_ON(osd >= map->max_osd);
1389 if ((map->osd_state[osd] & CEPH_OSD_UP) &&
1390 (xorstate & CEPH_OSD_UP))
1391 pr_info("osd%d down\n", osd);
1392 if ((map->osd_state[osd] & CEPH_OSD_EXISTS) &&
1393 (xorstate & CEPH_OSD_EXISTS)) {
1394 pr_info("osd%d does not exist\n", osd);
930c5328
ID		 1395 ret = set_primary_affinity(map, osd,
1396 CEPH_OSD_DEFAULT_PRIMARY_AFFINITY);
1397 if (ret)
1398 return ret;
1399 memset(map->osd_addr + osd, 0, sizeof(*map->osd_addr));
1400 map->osd_state[osd] = 0;
1401 } else {
1402 map->osd_state[osd] ^= xorstate;
1403 }
1404 }
1405
1406 /* new_up_client */
1407 *p = new_up_client;
1408 len = ceph_decode_32(p);
1409 while (len--) {
1410 s32 osd;
1411 struct ceph_entity_addr addr;
1412
1413 osd = ceph_decode_32(p);
1414 ceph_decode_copy(p, &addr, sizeof(addr));
1415 ceph_decode_addr(&addr);
1416 BUG_ON(osd >= map->max_osd);
1417 pr_info("osd%d up\n", osd);
1418 map->osd_state[osd] |= CEPH_OSD_EXISTS | CEPH_OSD_UP;
1419 map->osd_addr[osd] = addr;
1420 }
1421
1422 *p = new_weight_end;
1423 return 0;
1424
1425e_inval:
1426 return -EINVAL;
1427}
1428
f24e9980
SW		 1429/*
1430 * decode and apply an incremental map update.
1431 */
1432struct ceph_osdmap *osdmap_apply_incremental(void **p, void *end,
0c0a8de1 1433 struct ceph_osdmap *map)
f24e9980 1434{
f24e9980
SW		 1435 struct ceph_fsid fsid;
1436 u32 epoch = 0;
1437 struct ceph_timespec modified;
4f6a7e5e
SW		 1438 s32 len;
1439 u64 pool;
1440 __s64 new_pool_max;
1441 __s32 new_flags, max;
f24e9980 1442 void *start = *p;
86f1742b 1443 int err;
ec7af972 1444 u8 struct_v;
f24e9980 1445
38a8d560
ID		 1446 dout("%s %p to %p len %d\n", __func__, *p, end, (int)(end - *p));
1447
ec7af972
ID		 1448 err = get_osdmap_client_data_v(p, end, "inc", &struct_v);
1449 if (err)
1450 goto bad;
f24e9980 1451
53bbaba9
ID		 1452 /* fsid, epoch, modified, new_pool_max, new_flags */
1453 ceph_decode_need(p, end, sizeof(fsid) + sizeof(u32) + sizeof(modified) +
1454 sizeof(u64) + sizeof(u32), e_inval);
f24e9980 1455 ceph_decode_copy(p, &fsid, sizeof(fsid));
c89136ea 1456 epoch = ceph_decode_32(p);
f24e9980
SW		 1457 BUG_ON(epoch != map->epoch+1);
1458 ceph_decode_copy(p, &modified, sizeof(modified));
4f6a7e5e 1459 new_pool_max = ceph_decode_64(p);
c89136ea 1460 new_flags = ceph_decode_32(p);
f24e9980
SW		 1461
1462 /* full map? */
86f1742b 1463 ceph_decode_32_safe(p, end, len, e_inval);
f24e9980
SW		 1464 if (len > 0) {
1465 dout("apply_incremental full map len %d, %p to %p\n",
1466 len, *p, end);
a2505d63 1467 return ceph_osdmap_decode(p, min(*p+len, end));
f24e9980
SW		 1468 }
1469
1470 /* new crush? */
86f1742b 1471 ceph_decode_32_safe(p, end, len, e_inval);
f24e9980 1472 if (len > 0) {
1b6a78b5
ID		 1473 err = osdmap_set_crush(map,
1474 crush_decode(*p, min(*p + len, end)));
1475 if (err)
86f1742b 1476 goto bad;
cebc5be6 1477 *p += len;
f24e9980
SW		 1478 }
1479
1480 /* new flags? */
1481 if (new_flags >= 0)
1482 map->flags = new_flags;
4fc51be8
SW		 1483 if (new_pool_max >= 0)
1484 map->pool_max = new_pool_max;
f24e9980 1485
f24e9980 1486 /* new max? */
53bbaba9 1487 ceph_decode_32_safe(p, end, max, e_inval);
f24e9980
SW		 1488 if (max >= 0) {
1489 err = osdmap_set_max_osd(map, max);
86f1742b 1490 if (err)
f24e9980
SW		 1491 goto bad;
1492 }
1493
1494 map->epoch++;
31456665 1495 map->modified = modified;
f24e9980 1496
433fbdd3
ID		 1497 /* new_pools */
1498 err = decode_new_pools(p, end, map);
1499 if (err)
1500 goto bad;
9464d008 1501
0f70c7ee
ID		 1502 /* new_pool_names */
1503 err = decode_pool_names(p, end, map);
9464d008
ID		 1504 if (err)
1505 goto bad;
f24e9980 1506
4fc51be8 1507 /* old_pool */
86f1742b 1508 ceph_decode_32_safe(p, end, len, e_inval);
4fc51be8
SW		 1509 while (len--) {
1510 struct ceph_pg_pool_info *pi;
1511
86f1742b 1512 ceph_decode_64_safe(p, end, pool, e_inval);
4fc51be8 1513 pi = __lookup_pg_pool(&map->pg_pools, pool);
2844a76a
SW		 1514 if (pi)
1515 __remove_pg_pool(&map->pg_pools, pi);
4fc51be8 1516 }
f24e9980 1517
930c5328
ID		 1518 /* new_up_client, new_state, new_weight */
1519 err = decode_new_up_state_weight(p, end, map);
1520 if (err)
1521 goto bad;
f24e9980
SW		 1522
1523 /* new_pg_temp */
10db634e
ID		 1524 err = decode_new_pg_temp(p, end, map);
1525 if (err)
1526 goto bad;
f24e9980 1527
d286de79
ID		 1528 /* new_primary_temp */
1529 if (struct_v >= 1) {
1530 err = decode_new_primary_temp(p, end, map);
1531 if (err)
1532 goto bad;
1533 }
1534
63a6993f
ID		 1535 /* new_primary_affinity */
1536 if (struct_v >= 2) {
1537 err = decode_new_primary_affinity(p, end, map);
1538 if (err)
1539 goto bad;
1540 }
1541
f24e9980
SW		 1542 /* ignore the rest */
1543 *p = end;
38a8d560
ID		 1544
1545 dout("inc osdmap epoch %d max_osd %d\n", map->epoch, map->max_osd);
f24e9980
SW		 1546 return map;
1547
86f1742b
ID		 1548e_inval:
1549 err = -EINVAL;
f24e9980 1550bad:
38a8d560
ID		 1551 pr_err("corrupt inc osdmap (%d) epoch %d off %d (%p of %p-%p)\n",
1552 err, epoch, (int)(*p - start), *p, start, end);
9ec7cab1
SW		 1553 print_hex_dump(KERN_DEBUG, "osdmap: ",
1554 DUMP_PREFIX_OFFSET, 16, 1,
1555 start, end - start, true);
f24e9980
SW		 1556 return ERR_PTR(err);
1557}
1558
30c156d9
YZ		 1559void ceph_oloc_copy(struct ceph_object_locator *dest,
1560 const struct ceph_object_locator *src)
1561{
ca35ffea 1562 ceph_oloc_destroy(dest);
30c156d9
YZ		 1563
1564 dest->pool = src->pool;
1565 if (src->pool_ns)
1566 dest->pool_ns = ceph_get_string(src->pool_ns);
ca35ffea
ID		 1567 else
1568 dest->pool_ns = NULL;
30c156d9
YZ		 1569}
1570EXPORT_SYMBOL(ceph_oloc_copy);
1571
1572void ceph_oloc_destroy(struct ceph_object_locator *oloc)
1573{
1574 ceph_put_string(oloc->pool_ns);
1575}
1576EXPORT_SYMBOL(ceph_oloc_destroy);
1577
d30291b9
ID		 1578void ceph_oid_copy(struct ceph_object_id *dest,
1579 const struct ceph_object_id *src)
1580{
ca35ffea 1581 ceph_oid_destroy(dest);
d30291b9
ID		 1582
1583 if (src->name != src->inline_name) {
1584 /* very rare, see ceph_object_id definition */
1585 dest->name = kmalloc(src->name_len + 1,
1586 GFP_NOIO | __GFP_NOFAIL);
ca35ffea
ID		 1587 } else {
1588 dest->name = dest->inline_name;
d30291b9 1589 }
d30291b9
ID		 1590 memcpy(dest->name, src->name, src->name_len + 1);
1591 dest->name_len = src->name_len;
1592}
1593EXPORT_SYMBOL(ceph_oid_copy);
1594
1595static __printf(2, 0)
1596int oid_printf_vargs(struct ceph_object_id *oid, const char *fmt, va_list ap)
1597{
1598 int len;
1599
1600 WARN_ON(!ceph_oid_empty(oid));
1601
1602 len = vsnprintf(oid->inline_name, sizeof(oid->inline_name), fmt, ap);
1603 if (len >= sizeof(oid->inline_name))
1604 return len;
1605
1606 oid->name_len = len;
1607 return 0;
1608}
1609
1610/*
1611 * If oid doesn't fit into inline buffer, BUG.
1612 */
1613void ceph_oid_printf(struct ceph_object_id *oid, const char *fmt, ...)
1614{
1615 va_list ap;
1616
1617 va_start(ap, fmt);
1618 BUG_ON(oid_printf_vargs(oid, fmt, ap));
1619 va_end(ap);
1620}
1621EXPORT_SYMBOL(ceph_oid_printf);
1622
1623static __printf(3, 0)
1624int oid_aprintf_vargs(struct ceph_object_id *oid, gfp_t gfp,
1625 const char *fmt, va_list ap)
1626{
1627 va_list aq;
1628 int len;
1629
1630 va_copy(aq, ap);
1631 len = oid_printf_vargs(oid, fmt, aq);
1632 va_end(aq);
1633
1634 if (len) {
1635 char *external_name;
1636
1637 external_name = kmalloc(len + 1, gfp);
1638 if (!external_name)
1639 return -ENOMEM;
f24e9980 1640
d30291b9
ID		 1641 oid->name = external_name;
1642 WARN_ON(vsnprintf(oid->name, len + 1, fmt, ap) != len);
1643 oid->name_len = len;
1644 }
f24e9980 1645
d30291b9
ID		 1646 return 0;
1647}
1648
1649/*
1650 * If oid doesn't fit into inline buffer, allocate.
1651 */
1652int ceph_oid_aprintf(struct ceph_object_id *oid, gfp_t gfp,
1653 const char *fmt, ...)
1654{
1655 va_list ap;
1656 int ret;
1657
1658 va_start(ap, fmt);
1659 ret = oid_aprintf_vargs(oid, gfp, fmt, ap);
1660 va_end(ap);
1661
1662 return ret;
1663}
1664EXPORT_SYMBOL(ceph_oid_aprintf);
1665
1666void ceph_oid_destroy(struct ceph_object_id *oid)
1667{
1668 if (oid->name != oid->inline_name)
1669 kfree(oid->name);
1670}
1671EXPORT_SYMBOL(ceph_oid_destroy);
f24e9980 1672
63244fa1
ID		 1673/*
1674 * osds only
1675 */
1676static bool __osds_equal(const struct ceph_osds *lhs,
1677 const struct ceph_osds *rhs)
1678{
1679 if (lhs->size == rhs->size &&
1680 !memcmp(lhs->osds, rhs->osds, rhs->size * sizeof(rhs->osds[0])))
1681 return true;
1682
1683 return false;
1684}
1685
1686/*
1687 * osds + primary
1688 */
1689static bool osds_equal(const struct ceph_osds *lhs,
1690 const struct ceph_osds *rhs)
1691{
1692 if (__osds_equal(lhs, rhs) &&
1693 lhs->primary == rhs->primary)
1694 return true;
1695
1696 return false;
1697}
1698
6f3bfd45
ID		 1699static bool osds_valid(const struct ceph_osds *set)
1700{
1701 /* non-empty set */
1702 if (set->size > 0 && set->primary >= 0)
1703 return true;
1704
1705 /* empty can_shift_osds set */
1706 if (!set->size && set->primary == -1)
1707 return true;
1708
1709 /* empty !can_shift_osds set - all NONE */
1710 if (set->size > 0 && set->primary == -1) {
1711 int i;
1712
1713 for (i = 0; i < set->size; i++) {
1714 if (set->osds[i] != CRUSH_ITEM_NONE)
1715 break;
1716 }
1717 if (i == set->size)
1718 return true;
1719 }
1720
1721 return false;
1722}
1723
1724void ceph_osds_copy(struct ceph_osds *dest, const struct ceph_osds *src)
1725{
1726 memcpy(dest->osds, src->osds, src->size * sizeof(src->osds[0]));
1727 dest->size = src->size;
1728 dest->primary = src->primary;
1729}
1730
7de030d6
ID		 1731bool ceph_pg_is_split(const struct ceph_pg *pgid, u32 old_pg_num,
1732 u32 new_pg_num)
63244fa1
ID		 1733{
1734 int old_bits = calc_bits_of(old_pg_num);
1735 int old_mask = (1 << old_bits) - 1;
1736 int n;
1737
1738 WARN_ON(pgid->seed >= old_pg_num);
1739 if (new_pg_num <= old_pg_num)
1740 return false;
1741
1742 for (n = 1; ; n++) {
1743 int next_bit = n << (old_bits - 1);
1744 u32 s = next_bit | pgid->seed;
1745
1746 if (s < old_pg_num || s == pgid->seed)
1747 continue;
1748 if (s >= new_pg_num)
1749 break;
1750
1751 s = ceph_stable_mod(s, old_pg_num, old_mask);
1752 if (s == pgid->seed)
1753 return true;
1754 }
1755
1756 return false;
1757}
1758
1759bool ceph_is_new_interval(const struct ceph_osds *old_acting,
1760 const struct ceph_osds *new_acting,
1761 const struct ceph_osds *old_up,
1762 const struct ceph_osds *new_up,
1763 int old_size,
1764 int new_size,
1765 int old_min_size,
1766 int new_min_size,
1767 u32 old_pg_num,
1768 u32 new_pg_num,
1769 bool old_sort_bitwise,
1770 bool new_sort_bitwise,
1771 const struct ceph_pg *pgid)
1772{
1773 return !osds_equal(old_acting, new_acting) ||
1774 !osds_equal(old_up, new_up) ||
1775 old_size != new_size ||
1776 old_min_size != new_min_size ||
7de030d6 1777 ceph_pg_is_split(pgid, old_pg_num, new_pg_num) ||
63244fa1
ID		 1778 old_sort_bitwise != new_sort_bitwise;
1779}
1780
1781static int calc_pg_rank(int osd, const struct ceph_osds *acting)
1782{
1783 int i;
1784
1785 for (i = 0; i < acting->size; i++) {
1786 if (acting->osds[i] == osd)
1787 return i;
1788 }
1789
1790 return -1;
1791}
1792
1793static bool primary_changed(const struct ceph_osds *old_acting,
1794 const struct ceph_osds *new_acting)
1795{
1796 if (!old_acting->size && !new_acting->size)
1797 return false; /* both still empty */
1798
1799 if (!old_acting->size ^ !new_acting->size)
1800 return true; /* was empty, now not, or vice versa */
1801
1802 if (old_acting->primary != new_acting->primary)
1803 return true; /* primary changed */
1804
1805 if (calc_pg_rank(old_acting->primary, old_acting) !=
1806 calc_pg_rank(new_acting->primary, new_acting))
1807 return true;
1808
1809 return false; /* same primary (tho replicas may have changed) */
1810}
1811
1812bool ceph_osds_changed(const struct ceph_osds *old_acting,
1813 const struct ceph_osds *new_acting,
1814 bool any_change)
1815{
1816 if (primary_changed(old_acting, new_acting))
1817 return true;
1818
1819 if (any_change && !__osds_equal(old_acting, new_acting))
1820 return true;
1821
1822 return false;
1823}
1824
f24e9980
SW		 1825/*
1826 * calculate file layout from given offset, length.
1827 * fill in correct oid, logical length, and object extent
1828 * offset, length.
1829 *
1830 * for now, we write only a single su, until we can
1831 * pass a stride back to the caller.
1832 */
d63b77f4 1833int ceph_calc_file_object_mapping(struct ceph_file_layout *layout,
e8afad65 1834 u64 off, u64 len,
645a1025 1835 u64 *ono,
f24e9980
SW		 1836 u64 *oxoff, u64 *oxlen)
1837{
7627151e
YZ		 1838 u32 osize = layout->object_size;
1839 u32 su = layout->stripe_unit;
1840 u32 sc = layout->stripe_count;
f24e9980
SW		 1841 u32 bl, stripeno, stripepos, objsetno;
1842 u32 su_per_object;
ff1d1f71 1843 u64 t, su_offset;
f24e9980 1844
e8afad65 1845 dout("mapping %llu~%llu osize %u fl_su %u\n", off, len,
f24e9980 1846 osize, su);
d63b77f4
SW		 1847 if (su == 0 || sc == 0)
1848 goto invalid;
35e054a6 1849 su_per_object = osize / su;
d63b77f4
SW		 1850 if (su_per_object == 0)
1851 goto invalid;
f24e9980
SW		 1852 dout("osize %u / su %u = su_per_object %u\n", osize, su,
1853 su_per_object);
1854
d63b77f4
SW		 1855 if ((su & ~PAGE_MASK) != 0)
1856 goto invalid;
1857
f24e9980
SW		 1858 /* bl = *off / su; */
1859 t = off;
1860 do_div(t, su);
1861 bl = t;
1862 dout("off %llu / su %u = bl %u\n", off, su, bl);
1863
1864 stripeno = bl / sc;
1865 stripepos = bl % sc;
1866 objsetno = stripeno / su_per_object;
1867
645a1025 1868 *ono = objsetno * sc + stripepos;
95c96174 1869 dout("objset %u * sc %u = ono %u\n", objsetno, sc, (unsigned int)*ono);
645a1025
SW		 1870
1871 /* *oxoff = *off % layout->fl_stripe_unit; # offset in su */
f24e9980 1872 t = off;
ff1d1f71
NW		 1873 su_offset = do_div(t, su);
1874 *oxoff = su_offset + (stripeno % su_per_object) * su;
1875
1876 /*
1877 * Calculate the length of the extent being written to the selected
e8afad65 1878 * object. This is the minimum of the full length requested (len) or
ff1d1f71
NW		 1879 * the remainder of the current stripe being written to.
1880 */
e8afad65 1881 *oxlen = min_t(u64, len, su - su_offset);
f24e9980
SW		 1882
1883 dout(" obj extent %llu~%llu\n", *oxoff, *oxlen);
d63b77f4
SW		 1884 return 0;
1885
1886invalid:
1887 dout(" invalid layout\n");
1888 *ono = 0;
1889 *oxoff = 0;
1890 *oxlen = 0;
1891 return -EINVAL;
f24e9980 1892}
3d14c5d2 1893EXPORT_SYMBOL(ceph_calc_file_object_mapping);
f24e9980
SW		 1894
1895/*
d9591f5e
ID		 1896 * Map an object into a PG.
1897 *
1898 * Should only be called with target_oid and target_oloc (as opposed to
1899 * base_oid and base_oloc), since tiering isn't taken into account.
f24e9980 1900 */
df28152d
ID		 1901int __ceph_object_locator_to_pg(struct ceph_pg_pool_info *pi,
1902 const struct ceph_object_id *oid,
1903 const struct ceph_object_locator *oloc,
1904 struct ceph_pg *raw_pgid)
f24e9980 1905{
df28152d 1906 WARN_ON(pi->id != oloc->pool);
f24e9980 1907
30c156d9
YZ		 1908 if (!oloc->pool_ns) {
1909 raw_pgid->pool = oloc->pool;
1910 raw_pgid->seed = ceph_str_hash(pi->object_hash, oid->name,
1911 oid->name_len);
1912 dout("%s %s -> raw_pgid %llu.%x\n", __func__, oid->name,
1913 raw_pgid->pool, raw_pgid->seed);
1914 } else {
1915 char stack_buf[256];
1916 char *buf = stack_buf;
1917 int nsl = oloc->pool_ns->len;
1918 size_t total = nsl + 1 + oid->name_len;
1919
1920 if (total > sizeof(stack_buf)) {
1921 buf = kmalloc(total, GFP_NOIO);
1922 if (!buf)
1923 return -ENOMEM;
1924 }
1925 memcpy(buf, oloc->pool_ns->str, nsl);
1926 buf[nsl] = '\037';
1927 memcpy(buf + nsl + 1, oid->name, oid->name_len);
1928 raw_pgid->pool = oloc->pool;
1929 raw_pgid->seed = ceph_str_hash(pi->object_hash, buf, total);
1930 if (buf != stack_buf)
1931 kfree(buf);
1932 dout("%s %s ns %.*s -> raw_pgid %llu.%x\n", __func__,
1933 oid->name, nsl, oloc->pool_ns->str,
1934 raw_pgid->pool, raw_pgid->seed);
1935 }
f24e9980
SW		 1936 return 0;
1937}
df28152d
ID		 1938
1939int ceph_object_locator_to_pg(struct ceph_osdmap *osdmap,
1940 const struct ceph_object_id *oid,
1941 const struct ceph_object_locator *oloc,
1942 struct ceph_pg *raw_pgid)
1943{
1944 struct ceph_pg_pool_info *pi;
1945
1946 pi = ceph_pg_pool_by_id(osdmap, oloc->pool);
1947 if (!pi)
1948 return -ENOENT;
1949
1950 return __ceph_object_locator_to_pg(pi, oid, oloc, raw_pgid);
1951}
d9591f5e 1952EXPORT_SYMBOL(ceph_object_locator_to_pg);
f24e9980 1953
6f3bfd45
ID		 1954/*
1955 * Map a raw PG (full precision ps) into an actual PG.
1956 */
1957static void raw_pg_to_pg(struct ceph_pg_pool_info *pi,
1958 const struct ceph_pg *raw_pgid,
1959 struct ceph_pg *pgid)
1960{
1961 pgid->pool = raw_pgid->pool;
1962 pgid->seed = ceph_stable_mod(raw_pgid->seed, pi->pg_num,
1963 pi->pg_num_mask);
1964}
1965
1966/*
1967 * Map a raw PG (full precision ps) into a placement ps (placement
1968 * seed). Include pool id in that value so that different pools don't
1969 * use the same seeds.
1970 */
1971static u32 raw_pg_to_pps(struct ceph_pg_pool_info *pi,
1972 const struct ceph_pg *raw_pgid)
1973{
1974 if (pi->flags & CEPH_POOL_FLAG_HASHPSPOOL) {
1975 /* hash pool id and seed so that pool PGs do not overlap */
1976 return crush_hash32_2(CRUSH_HASH_RJENKINS1,
1977 ceph_stable_mod(raw_pgid->seed,
1978 pi->pgp_num,
1979 pi->pgp_num_mask),
1980 raw_pgid->pool);
1981 } else {
1982 /*
1983 * legacy behavior: add ps and pool together. this is
1984 * not a great approach because the PGs from each pool
1985 * will overlap on top of each other: 0.5 == 1.4 ==
1986 * 2.3 == ...
1987 */
1988 return ceph_stable_mod(raw_pgid->seed, pi->pgp_num,
1989 pi->pgp_num_mask) +
1990 (unsigned)raw_pgid->pool;
1991 }
1992}
1993
9d521470
ID		 1994static int do_crush(struct ceph_osdmap *map, int ruleno, int x,
1995 int *result, int result_max,
1996 const __u32 *weight, int weight_max)
e8ef19c4 1997{
9d521470
ID		 1998 int r;
1999
2000 BUG_ON(result_max > CEPH_PG_MAX_SIZE);
2001
743efcff 2002 mutex_lock(&map->crush_workspace_mutex);
9d521470 2003 r = crush_do_rule(map->crush, ruleno, x, result, result_max,
743efcff
ID		 2004 weight, weight_max, map->crush_workspace);
2005 mutex_unlock(&map->crush_workspace_mutex);
e8ef19c4 2006
9d521470 2007 return r;
e8ef19c4
ID		 2008}
2009
2bd93d4d 2010/*
6f3bfd45
ID		 2011 * Calculate raw set (CRUSH output) for given PG. The result may
2012 * contain nonexistent OSDs. ->primary is undefined for a raw set.
2bd93d4d 2013 *
6f3bfd45 2014 * Placement seed (CRUSH input) is returned through @ppps.
2bd93d4d 2015 */
6f3bfd45
ID		 2016static void pg_to_raw_osds(struct ceph_osdmap *osdmap,
2017 struct ceph_pg_pool_info *pi,
2018 const struct ceph_pg *raw_pgid,
2019 struct ceph_osds *raw,
2020 u32 *ppps)
2bd93d4d 2021{
6f3bfd45 2022 u32 pps = raw_pg_to_pps(pi, raw_pgid);
2bd93d4d
ID		 2023 int ruleno;
2024 int len;
2025
6f3bfd45
ID		 2026 ceph_osds_init(raw);
2027 if (ppps)
2028 *ppps = pps;
2029
2030 ruleno = crush_find_rule(osdmap->crush, pi->crush_ruleset, pi->type,
2031 pi->size);
2bd93d4d
ID		 2032 if (ruleno < 0) {
2033 pr_err("no crush rule: pool %lld ruleset %d type %d size %d\n",
6f3bfd45
ID		 2034 pi->id, pi->crush_ruleset, pi->type, pi->size);
2035 return;
2bd93d4d
ID		 2036 }
2037
ef9324bb
ID		 2038 if (pi->size > ARRAY_SIZE(raw->osds)) {
2039 pr_err_ratelimited("pool %lld ruleset %d type %d too wide: size %d > %zu\n",
2040 pi->id, pi->crush_ruleset, pi->type, pi->size,
2041 ARRAY_SIZE(raw->osds));
2042 return;
2043 }
2044
2045 len = do_crush(osdmap, ruleno, pps, raw->osds, pi->size,
2bd93d4d
ID		 2046 osdmap->osd_weight, osdmap->max_osd);
2047 if (len < 0) {
2048 pr_err("error %d from crush rule %d: pool %lld ruleset %d type %d size %d\n",
6f3bfd45
ID		 2049 len, ruleno, pi->id, pi->crush_ruleset, pi->type,
2050 pi->size);
2051 return;
2bd93d4d
ID		 2052 }
2053
6f3bfd45 2054 raw->size = len;
2bd93d4d
ID		 2055}
2056
2057/*
6f3bfd45
ID		 2058 * Given raw set, calculate up set and up primary. By definition of an
2059 * up set, the result won't contain nonexistent or down OSDs.
2bd93d4d 2060 *
6f3bfd45
ID		 2061 * This is done in-place - on return @set is the up set. If it's
2062 * empty, ->primary will remain undefined.
2bd93d4d 2063 */
6f3bfd45
ID		 2064static void raw_to_up_osds(struct ceph_osdmap *osdmap,
2065 struct ceph_pg_pool_info *pi,
2066 struct ceph_osds *set)
2bd93d4d 2067{
2bd93d4d
ID		 2068 int i;
2069
6f3bfd45
ID		 2070 /* ->primary is undefined for a raw set */
2071 BUG_ON(set->primary != -1);
2072
2073 if (ceph_can_shift_osds(pi)) {
2bd93d4d
ID		 2074 int removed = 0;
2075
6f3bfd45
ID		 2076 /* shift left */
2077 for (i = 0; i < set->size; i++) {
2078 if (ceph_osd_is_down(osdmap, set->osds[i])) {
2bd93d4d
ID		 2079 removed++;
2080 continue;
2081 }
2082 if (removed)
6f3bfd45 2083 set->osds[i - removed] = set->osds[i];
2bd93d4d 2084 }
6f3bfd45
ID		 2085 set->size -= removed;
2086 if (set->size > 0)
2087 set->primary = set->osds[0];
2bd93d4d 2088 } else {
6f3bfd45
ID		 2089 /* set down/dne devices to NONE */
2090 for (i = set->size - 1; i >= 0; i--) {
2091 if (ceph_osd_is_down(osdmap, set->osds[i]))
2092 set->osds[i] = CRUSH_ITEM_NONE;
2bd93d4d 2093 else
6f3bfd45 2094 set->primary = set->osds[i];
2bd93d4d
ID		 2095 }
2096 }
2bd93d4d
ID		 2097}
2098
6f3bfd45
ID		 2099static void apply_primary_affinity(struct ceph_osdmap *osdmap,
2100 struct ceph_pg_pool_info *pi,
2101 u32 pps,
2102 struct ceph_osds *up)
47ec1f3c
ID		 2103{
2104 int i;
2105 int pos = -1;
2106
2107 /*
2108 * Do we have any non-default primary_affinity values for these
2109 * osds?
2110 */
2111 if (!osdmap->osd_primary_affinity)
2112 return;
2113
6f3bfd45
ID		 2114 for (i = 0; i < up->size; i++) {
2115 int osd = up->osds[i];
92b2e751
ID		 2116
2117 if (osd != CRUSH_ITEM_NONE &&
2118 osdmap->osd_primary_affinity[osd] !=
47ec1f3c
ID		 2119 CEPH_OSD_DEFAULT_PRIMARY_AFFINITY) {
2120 break;
2121 }
2122 }
6f3bfd45 2123 if (i == up->size)
47ec1f3c
ID		 2124 return;
2125
2126 /*
2127 * Pick the primary. Feed both the seed (for the pg) and the
2128 * osd into the hash/rng so that a proportional fraction of an
2129 * osd's pgs get rejected as primary.
2130 */
6f3bfd45
ID		 2131 for (i = 0; i < up->size; i++) {
2132 int osd = up->osds[i];
47ec1f3c
ID		 2133 u32 aff;
2134
47ec1f3c
ID		 2135 if (osd == CRUSH_ITEM_NONE)
2136 continue;
2137
2138 aff = osdmap->osd_primary_affinity[osd];
2139 if (aff < CEPH_OSD_MAX_PRIMARY_AFFINITY &&
2140 (crush_hash32_2(CRUSH_HASH_RJENKINS1,
2141 pps, osd) >> 16) >= aff) {
2142 /*
2143 * We chose not to use this primary. Note it
2144 * anyway as a fallback in case we don't pick
2145 * anyone else, but keep looking.
2146 */
2147 if (pos < 0)
2148 pos = i;
2149 } else {
2150 pos = i;
2151 break;
2152 }
2153 }
2154 if (pos < 0)
2155 return;
2156
6f3bfd45 2157 up->primary = up->osds[pos];
47ec1f3c 2158
6f3bfd45 2159 if (ceph_can_shift_osds(pi) && pos > 0) {
47ec1f3c
ID		 2160 /* move the new primary to the front */
2161 for (i = pos; i > 0; i--)
6f3bfd45
ID		 2162 up->osds[i] = up->osds[i - 1];
2163 up->osds[0] = up->primary;
47ec1f3c
ID		 2164 }
2165}
2166
45966c34 2167/*
6f3bfd45 2168 * Get pg_temp and primary_temp mappings for given PG.
45966c34 2169 *
6f3bfd45
ID		 2170 * Note that a PG may have none, only pg_temp, only primary_temp or
2171 * both pg_temp and primary_temp mappings. This means @temp isn't
2172 * always a valid OSD set on return: in the "only primary_temp" case,
2173 * @temp will have its ->primary >= 0 but ->size == 0.
45966c34 2174 */
6f3bfd45
ID		 2175static void get_temp_osds(struct ceph_osdmap *osdmap,
2176 struct ceph_pg_pool_info *pi,
2177 const struct ceph_pg *raw_pgid,
2178 struct ceph_osds *temp)
45966c34 2179{
6f3bfd45 2180 struct ceph_pg pgid;
45966c34 2181 struct ceph_pg_mapping *pg;
45966c34
ID		 2182 int i;
2183
6f3bfd45
ID		 2184 raw_pg_to_pg(pi, raw_pgid, &pgid);
2185 ceph_osds_init(temp);
45966c34
ID		 2186
2187 /* pg_temp? */
ab75144b 2188 pg = lookup_pg_mapping(&osdmap->pg_temp, &pgid);
45966c34 2189 if (pg) {
45966c34
ID		 2190 for (i = 0; i < pg->pg_temp.len; i++) {
2191 if (ceph_osd_is_down(osdmap, pg->pg_temp.osds[i])) {
6f3bfd45 2192 if (ceph_can_shift_osds(pi))
45966c34 2193 continue;
6f3bfd45
ID		 2194
2195 temp->osds[temp->size++] = CRUSH_ITEM_NONE;
45966c34 2196 } else {
6f3bfd45 2197 temp->osds[temp->size++] = pg->pg_temp.osds[i];
45966c34
ID		 2198 }
2199 }
2200
2201 /* apply pg_temp's primary */
6f3bfd45
ID		 2202 for (i = 0; i < temp->size; i++) {
2203 if (temp->osds[i] != CRUSH_ITEM_NONE) {
2204 temp->primary = temp->osds[i];
45966c34
ID		 2205 break;
2206 }
2207 }
45966c34
ID		 2208 }
2209
5e8d4d36 2210 /* primary_temp? */
ab75144b 2211 pg = lookup_pg_mapping(&osdmap->primary_temp, &pgid);
5e8d4d36 2212 if (pg)
6f3bfd45 2213 temp->primary = pg->primary_temp.osd;
45966c34
ID		 2214}
2215
d85b7056 2216/*
6f3bfd45 2217 * Map a PG to its acting set as well as its up set.
ac972230 2218 *
6f3bfd45
ID		 2219 * Acting set is used for data mapping purposes, while up set can be
2220 * recorded for detecting interval changes and deciding whether to
2221 * resend a request.
d85b7056 2222 */
6f3bfd45 2223void ceph_pg_to_up_acting_osds(struct ceph_osdmap *osdmap,
df28152d 2224 struct ceph_pg_pool_info *pi,
6f3bfd45
ID		 2225 const struct ceph_pg *raw_pgid,
2226 struct ceph_osds *up,
2227 struct ceph_osds *acting)
d85b7056 2228{
ac972230 2229 u32 pps;
d85b7056 2230
df28152d 2231 WARN_ON(pi->id != raw_pgid->pool);
d85b7056 2232
6f3bfd45
ID		 2233 pg_to_raw_osds(osdmap, pi, raw_pgid, up, &pps);
2234 raw_to_up_osds(osdmap, pi, up);
2235 apply_primary_affinity(osdmap, pi, pps, up);
2236 get_temp_osds(osdmap, pi, raw_pgid, acting);
2237 if (!acting->size) {
2238 memcpy(acting->osds, up->osds, up->size * sizeof(up->osds[0]));
2239 acting->size = up->size;
2240 if (acting->primary == -1)
2241 acting->primary = up->primary;
8008ab10 2242 }
6f3bfd45 2243 WARN_ON(!osds_valid(up) || !osds_valid(acting));
d85b7056
SW		 2244}
2245
dc98ff72 2246bool ceph_pg_to_primary_shard(struct ceph_osdmap *osdmap,
df28152d 2247 struct ceph_pg_pool_info *pi,
dc98ff72
ID		 2248 const struct ceph_pg *raw_pgid,
2249 struct ceph_spg *spgid)
2250{
dc98ff72
ID		 2251 struct ceph_pg pgid;
2252 struct ceph_osds up, acting;
2253 int i;
2254
df28152d 2255 WARN_ON(pi->id != raw_pgid->pool);
dc98ff72
ID		 2256 raw_pg_to_pg(pi, raw_pgid, &pgid);
2257
2258 if (ceph_can_shift_osds(pi)) {
2259 spgid->pgid = pgid; /* struct */
2260 spgid->shard = CEPH_SPG_NOSHARD;
2261 return true;
2262 }
2263
df28152d 2264 ceph_pg_to_up_acting_osds(osdmap, pi, &pgid, &up, &acting);
dc98ff72
ID		 2265 for (i = 0; i < acting.size; i++) {
2266 if (acting.osds[i] == acting.primary) {
2267 spgid->pgid = pgid; /* struct */
2268 spgid->shard = i;
2269 return true;
2270 }
2271 }
2272
2273 return false;
2274}
2275
f24e9980 2276/*
f81f1633 2277 * Return acting primary for given PG, or -1 if none.
f24e9980 2278 */
f81f1633
ID		 2279int ceph_pg_to_acting_primary(struct ceph_osdmap *osdmap,
2280 const struct ceph_pg *raw_pgid)
f24e9980 2281{
df28152d 2282 struct ceph_pg_pool_info *pi;
6f3bfd45 2283 struct ceph_osds up, acting;
f24e9980 2284
df28152d
ID		 2285 pi = ceph_pg_pool_by_id(osdmap, raw_pgid->pool);
2286 if (!pi)
2287 return -1;
2288
2289 ceph_pg_to_up_acting_osds(osdmap, pi, raw_pgid, &up, &acting);
6f3bfd45 2290 return acting.primary;
f24e9980 2291}
f81f1633 2292EXPORT_SYMBOL(ceph_pg_to_acting_primary);
Linux 6.x block layer and io_uring tree(s)