net/ceph/osdmap.c

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