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 /*
 384  * rbtree of pg_mapping for handling pg_temp (explicit mapping of pgid
 385  * to a set of osds) and primary_temp (explicit primary setting)
 386  */
 387 static int pgid_cmp(struct ceph_pg l, struct ceph_pg r)
 388 {
 389         if (l.pool < r.pool)
 390                 return -1;
 391         if (l.pool > r.pool)
 392                 return 1;
 393         if (l.seed < r.seed)
 394                 return -1;
 395         if (l.seed > r.seed)
 396                 return 1;
 397         return 0;
 398 }
 399
 400 static int __insert_pg_mapping(struct ceph_pg_mapping *new,
 401                                struct rb_root *root)
 402 {
 403         struct rb_node **p = &root->rb_node;
 404         struct rb_node *parent = NULL;
 405         struct ceph_pg_mapping *pg = NULL;
 406         int c;
 407
 408         dout("__insert_pg_mapping %llx %p\n", *(u64 *)&new->pgid, new);
 409         while (*p) {
 410                 parent = *p;
 411                 pg = rb_entry(parent, struct ceph_pg_mapping, node);
 412                 c = pgid_cmp(new->pgid, pg->pgid);
 413                 if (c < 0)
 414                         p = &(*p)->rb_left;
 415                 else if (c > 0)
 416                         p = &(*p)->rb_right;
 417                 else
 418                         return -EEXIST;
 419         }
 420
 421         rb_link_node(&new->node, parent, p);
 422         rb_insert_color(&new->node, root);
 423         return 0;
 424 }
 425
 426 static struct ceph_pg_mapping *__lookup_pg_mapping(struct rb_root *root,
 427                                                    struct ceph_pg pgid)
 428 {
 429         struct rb_node *n = root->rb_node;
 430         struct ceph_pg_mapping *pg;
 431         int c;
 432
 433         while (n) {
 434                 pg = rb_entry(n, struct ceph_pg_mapping, node);
 435                 c = pgid_cmp(pgid, pg->pgid);
 436                 if (c < 0) {
 437                         n = n->rb_left;
 438                 } else if (c > 0) {
 439                         n = n->rb_right;
 440                 } else {
 441                         dout("__lookup_pg_mapping %lld.%x got %p\n",
 442                              pgid.pool, pgid.seed, pg);
 443                         return pg;
 444                 }
 445         }
 446         return NULL;
 447 }
 448
 449 static int __remove_pg_mapping(struct rb_root *root, struct ceph_pg pgid)
 450 {
 451         struct ceph_pg_mapping *pg = __lookup_pg_mapping(root, pgid);
 452
 453         if (pg) {
 454                 dout("__remove_pg_mapping %lld.%x %p\n", pgid.pool, pgid.seed,
 455                      pg);
 456                 rb_erase(&pg->node, root);
 457                 kfree(pg);
 458                 return 0;
 459         }
 460         dout("__remove_pg_mapping %lld.%x dne\n", pgid.pool, pgid.seed);
 461         return -ENOENT;
 462 }
 463
 464 /*
 465  * rbtree of pg pool info
 466  */
 467 static int __insert_pg_pool(struct rb_root *root, struct ceph_pg_pool_info *new)
 468 {
 469         struct rb_node **p = &root->rb_node;
 470         struct rb_node *parent = NULL;
 471         struct ceph_pg_pool_info *pi = NULL;
 472
 473         while (*p) {
 474                 parent = *p;
 475                 pi = rb_entry(parent, struct ceph_pg_pool_info, node);
 476                 if (new->id < pi->id)
 477                         p = &(*p)->rb_left;
 478                 else if (new->id > pi->id)
 479                         p = &(*p)->rb_right;
 480                 else
 481                         return -EEXIST;
 482         }
 483
 484         rb_link_node(&new->node, parent, p);
 485         rb_insert_color(&new->node, root);
 486         return 0;
 487 }
 488
 489 static struct ceph_pg_pool_info *__lookup_pg_pool(struct rb_root *root, u64 id)
 490 {
 491         struct ceph_pg_pool_info *pi;
 492         struct rb_node *n = root->rb_node;
 493
 494         while (n) {
 495                 pi = rb_entry(n, struct ceph_pg_pool_info, node);
 496                 if (id < pi->id)
 497                         n = n->rb_left;
 498                 else if (id > pi->id)
 499                         n = n->rb_right;
 500                 else
 501                         return pi;
 502         }
 503         return NULL;
 504 }
 505
 506 struct ceph_pg_pool_info *ceph_pg_pool_by_id(struct ceph_osdmap *map, u64 id)
 507 {
 508         return __lookup_pg_pool(&map->pg_pools, id);
 509 }
 510
 511 const char *ceph_pg_pool_name_by_id(struct ceph_osdmap *map, u64 id)
 512 {
 513         struct ceph_pg_pool_info *pi;
 514
 515         if (id == CEPH_NOPOOL)
 516                 return NULL;
 517
 518         if (WARN_ON_ONCE(id > (u64) INT_MAX))
 519                 return NULL;
 520
 521         pi = __lookup_pg_pool(&map->pg_pools, (int) id);
 522
 523         return pi ? pi->name : NULL;
 524 }
 525 EXPORT_SYMBOL(ceph_pg_pool_name_by_id);
 526
 527 int ceph_pg_poolid_by_name(struct ceph_osdmap *map, const char *name)
 528 {
 529         struct rb_node *rbp;
 530
 531         for (rbp = rb_first(&map->pg_pools); rbp; rbp = rb_next(rbp)) {
 532                 struct ceph_pg_pool_info *pi =
 533                         rb_entry(rbp, struct ceph_pg_pool_info, node);
 534                 if (pi->name && strcmp(pi->name, name) == 0)
 535                         return pi->id;
 536         }
 537         return -ENOENT;
 538 }
 539 EXPORT_SYMBOL(ceph_pg_poolid_by_name);
 540
 541 static void __remove_pg_pool(struct rb_root *root, struct ceph_pg_pool_info *pi)
 542 {
 543         rb_erase(&pi->node, root);
 544         kfree(pi->name);
 545         kfree(pi);
 546 }
 547
 548 static int decode_pool(void **p, void *end, struct ceph_pg_pool_info *pi)
 549 {
 550         u8 ev, cv;
 551         unsigned len, num;
 552         void *pool_end;
 553
 554         ceph_decode_need(p, end, 2 + 4, bad);
 555         ev = ceph_decode_8(p);  /* encoding version */
 556         cv = ceph_decode_8(p); /* compat version */
 557         if (ev < 5) {
 558                 pr_warn("got v %d < 5 cv %d of ceph_pg_pool\n", ev, cv);
 559                 return -EINVAL;
 560         }
 561         if (cv > 9) {
 562                 pr_warn("got v %d cv %d > 9 of ceph_pg_pool\n", ev, cv);
 563                 return -EINVAL;
 564         }
 565         len = ceph_decode_32(p);
 566         ceph_decode_need(p, end, len, bad);
 567         pool_end = *p + len;
 568
 569         pi->type = ceph_decode_8(p);
 570         pi->size = ceph_decode_8(p);
 571         pi->crush_ruleset = ceph_decode_8(p);
 572         pi->object_hash = ceph_decode_8(p);
 573
 574         pi->pg_num = ceph_decode_32(p);
 575         pi->pgp_num = ceph_decode_32(p);
 576
 577         *p += 4 + 4;  /* skip lpg* */
 578         *p += 4;      /* skip last_change */
 579         *p += 8 + 4;  /* skip snap_seq, snap_epoch */
 580
 581         /* skip snaps */
 582         num = ceph_decode_32(p);
 583         while (num--) {
 584                 *p += 8;  /* snapid key */
 585                 *p += 1 + 1; /* versions */
 586                 len = ceph_decode_32(p);
 587                 *p += len;
 588         }
 589
 590         /* skip removed_snaps */
 591         num = ceph_decode_32(p);
 592         *p += num * (8 + 8);
 593
 594         *p += 8;  /* skip auid */
 595         pi->flags = ceph_decode_64(p);
 596         *p += 4;  /* skip crash_replay_interval */
 597
 598         if (ev >= 7)
 599                 *p += 1;  /* skip min_size */
 600
 601         if (ev >= 8)
 602                 *p += 8 + 8;  /* skip quota_max_* */
 603
 604         if (ev >= 9) {
 605                 /* skip tiers */
 606                 num = ceph_decode_32(p);
 607                 *p += num * 8;
 608
 609                 *p += 8;  /* skip tier_of */
 610                 *p += 1;  /* skip cache_mode */
 611
 612                 pi->read_tier = ceph_decode_64(p);
 613                 pi->write_tier = ceph_decode_64(p);
 614         } else {
 615                 pi->read_tier = -1;
 616                 pi->write_tier = -1;
 617         }
 618
 619         /* ignore the rest */
 620
 621         *p = pool_end;
 622         calc_pg_masks(pi);
 623         return 0;
 624
 625 bad:
 626         return -EINVAL;
 627 }
 628
 629 static int decode_pool_names(void **p, void *end, struct ceph_osdmap *map)
 630 {
 631         struct ceph_pg_pool_info *pi;
 632         u32 num, len;
 633         u64 pool;
 634
 635         ceph_decode_32_safe(p, end, num, bad);
 636         dout(" %d pool names\n", num);
 637         while (num--) {
 638                 ceph_decode_64_safe(p, end, pool, bad);
 639                 ceph_decode_32_safe(p, end, len, bad);
 640                 dout("  pool %llu len %d\n", pool, len);
 641                 ceph_decode_need(p, end, len, bad);
 642                 pi = __lookup_pg_pool(&map->pg_pools, pool);
 643                 if (pi) {
 644                         char *name = kstrndup(*p, len, GFP_NOFS);
 645
 646                         if (!name)
 647                                 return -ENOMEM;
 648                         kfree(pi->name);
 649                         pi->name = name;
 650                         dout("  name is %s\n", pi->name);
 651                 }
 652                 *p += len;
 653         }
 654         return 0;
 655
 656 bad:
 657         return -EINVAL;
 658 }
 659
 660 /*
 661  * osd map
 662  */
 663 void ceph_osdmap_destroy(struct ceph_osdmap *map)
 664 {
 665         dout("osdmap_destroy %p\n", map);
 666         if (map->crush)
 667                 crush_destroy(map->crush);
 668         while (!RB_EMPTY_ROOT(&map->pg_temp)) {
 669                 struct ceph_pg_mapping *pg =
 670                         rb_entry(rb_first(&map->pg_temp),
 671                                  struct ceph_pg_mapping, node);
 672                 rb_erase(&pg->node, &map->pg_temp);
 673                 kfree(pg);
 674         }
 675         while (!RB_EMPTY_ROOT(&map->primary_temp)) {
 676                 struct ceph_pg_mapping *pg =
 677                         rb_entry(rb_first(&map->primary_temp),
 678                                  struct ceph_pg_mapping, node);
 679                 rb_erase(&pg->node, &map->primary_temp);
 680                 kfree(pg);
 681         }
 682         while (!RB_EMPTY_ROOT(&map->pg_pools)) {
 683                 struct ceph_pg_pool_info *pi =
 684                         rb_entry(rb_first(&map->pg_pools),
 685                                  struct ceph_pg_pool_info, node);
 686                 __remove_pg_pool(&map->pg_pools, pi);
 687         }
 688         kfree(map->osd_state);
 689         kfree(map->osd_weight);
 690         kfree(map->osd_addr);
 691         kfree(map->osd_primary_affinity);
 692         kfree(map);
 693 }
 694
 695 /*
 696  * Adjust max_osd value, (re)allocate arrays.
 697  *
 698  * The new elements are properly initialized.
 699  */
 700 static int osdmap_set_max_osd(struct ceph_osdmap *map, int max)
 701 {
 702         u8 *state;
 703         u32 *weight;
 704         struct ceph_entity_addr *addr;
 705         int i;
 706
 707         state = krealloc(map->osd_state, max*sizeof(*state), GFP_NOFS);
 708         if (!state)
 709                 return -ENOMEM;
 710         map->osd_state = state;
 711
 712         weight = krealloc(map->osd_weight, max*sizeof(*weight), GFP_NOFS);
 713         if (!weight)
 714                 return -ENOMEM;
 715         map->osd_weight = weight;
 716
 717         addr = krealloc(map->osd_addr, max*sizeof(*addr), GFP_NOFS);
 718         if (!addr)
 719                 return -ENOMEM;
 720         map->osd_addr = addr;
 721
 722         for (i = map->max_osd; i < max; i++) {
 723                 map->osd_state[i] = 0;
 724                 map->osd_weight[i] = CEPH_OSD_OUT;
 725                 memset(map->osd_addr + i, 0, sizeof(*map->osd_addr));
 726         }
 727
 728         if (map->osd_primary_affinity) {
 729                 u32 *affinity;
 730
 731                 affinity = krealloc(map->osd_primary_affinity,
 732                                     max*sizeof(*affinity), GFP_NOFS);
 733                 if (!affinity)
 734                         return -ENOMEM;
 735                 map->osd_primary_affinity = affinity;
 736
 737                 for (i = map->max_osd; i < max; i++)
 738                         map->osd_primary_affinity[i] =
 739                             CEPH_OSD_DEFAULT_PRIMARY_AFFINITY;
 740         }
 741
 742         map->max_osd = max;
 743
 744         return 0;
 745 }
 746
 747 #define OSDMAP_WRAPPER_COMPAT_VER       7
 748 #define OSDMAP_CLIENT_DATA_COMPAT_VER   1
 749
 750 /*
 751  * Return 0 or error.  On success, *v is set to 0 for old (v6) osdmaps,
 752  * to struct_v of the client_data section for new (v7 and above)
 753  * osdmaps.
 754  */
 755 static int get_osdmap_client_data_v(void **p, void *end,
 756                                     const char *prefix, u8 *v)
 757 {
 758         u8 struct_v;
 759
 760         ceph_decode_8_safe(p, end, struct_v, e_inval);
 761         if (struct_v >= 7) {
 762                 u8 struct_compat;
 763
 764                 ceph_decode_8_safe(p, end, struct_compat, e_inval);
 765                 if (struct_compat > OSDMAP_WRAPPER_COMPAT_VER) {
 766                         pr_warn("got v %d cv %d > %d of %s ceph_osdmap\n",
 767                                 struct_v, struct_compat,
 768                                 OSDMAP_WRAPPER_COMPAT_VER, prefix);
 769                         return -EINVAL;
 770                 }
 771                 *p += 4; /* ignore wrapper struct_len */
 772
 773                 ceph_decode_8_safe(p, end, struct_v, e_inval);
 774                 ceph_decode_8_safe(p, end, struct_compat, e_inval);
 775                 if (struct_compat > OSDMAP_CLIENT_DATA_COMPAT_VER) {
 776                         pr_warn("got v %d cv %d > %d of %s ceph_osdmap client data\n",
 777                                 struct_v, struct_compat,
 778                                 OSDMAP_CLIENT_DATA_COMPAT_VER, prefix);
 779                         return -EINVAL;
 780                 }
 781                 *p += 4; /* ignore client data struct_len */
 782         } else {
 783                 u16 version;
 784
 785                 *p -= 1;
 786                 ceph_decode_16_safe(p, end, version, e_inval);
 787                 if (version < 6) {
 788                         pr_warn("got v %d < 6 of %s ceph_osdmap\n",
 789                                 version, prefix);
 790                         return -EINVAL;
 791                 }
 792
 793                 /* old osdmap enconding */
 794                 struct_v = 0;
 795         }
 796
 797         *v = struct_v;
 798         return 0;
 799
 800 e_inval:
 801         return -EINVAL;
 802 }
 803
 804 static int __decode_pools(void **p, void *end, struct ceph_osdmap *map,
 805                           bool incremental)
 806 {
 807         u32 n;
 808
 809         ceph_decode_32_safe(p, end, n, e_inval);
 810         while (n--) {
 811                 struct ceph_pg_pool_info *pi;
 812                 u64 pool;
 813                 int ret;
 814
 815                 ceph_decode_64_safe(p, end, pool, e_inval);
 816
 817                 pi = __lookup_pg_pool(&map->pg_pools, pool);
 818                 if (!incremental || !pi) {
 819                         pi = kzalloc(sizeof(*pi), GFP_NOFS);
 820                         if (!pi)
 821                                 return -ENOMEM;
 822
 823                         pi->id = pool;
 824
 825                         ret = __insert_pg_pool(&map->pg_pools, pi);
 826                         if (ret) {
 827                                 kfree(pi);
 828                                 return ret;
 829                         }
 830                 }
 831
 832                 ret = decode_pool(p, end, pi);
 833                 if (ret)
 834                         return ret;
 835         }
 836
 837         return 0;
 838
 839 e_inval:
 840         return -EINVAL;
 841 }
 842
 843 static int decode_pools(void **p, void *end, struct ceph_osdmap *map)
 844 {
 845         return __decode_pools(p, end, map, false);
 846 }
 847
 848 static int decode_new_pools(void **p, void *end, struct ceph_osdmap *map)
 849 {
 850         return __decode_pools(p, end, map, true);
 851 }
 852
 853 static int __decode_pg_temp(void **p, void *end, struct ceph_osdmap *map,
 854                             bool incremental)
 855 {
 856         u32 n;
 857
 858         ceph_decode_32_safe(p, end, n, e_inval);
 859         while (n--) {
 860                 struct ceph_pg pgid;
 861                 u32 len, i;
 862                 int ret;
 863
 864                 ret = ceph_decode_pgid(p, end, &pgid);
 865                 if (ret)
 866                         return ret;
 867
 868                 ceph_decode_32_safe(p, end, len, e_inval);
 869
 870                 ret = __remove_pg_mapping(&map->pg_temp, pgid);
 871                 BUG_ON(!incremental && ret != -ENOENT);
 872
 873                 if (!incremental || len > 0) {
 874                         struct ceph_pg_mapping *pg;
 875
 876                         ceph_decode_need(p, end, len*sizeof(u32), e_inval);
 877
 878                         if (len > (UINT_MAX - sizeof(*pg)) / sizeof(u32))
 879                                 return -EINVAL;
 880
 881                         pg = kzalloc(sizeof(*pg) + len*sizeof(u32), GFP_NOFS);
 882                         if (!pg)
 883                                 return -ENOMEM;
 884
 885                         pg->pgid = pgid;
 886                         pg->pg_temp.len = len;
 887                         for (i = 0; i < len; i++)
 888                                 pg->pg_temp.osds[i] = ceph_decode_32(p);
 889
 890                         ret = __insert_pg_mapping(pg, &map->pg_temp);
 891                         if (ret) {
 892                                 kfree(pg);
 893                                 return ret;
 894                         }
 895                 }
 896         }
 897
 898         return 0;
 899
 900 e_inval:
 901         return -EINVAL;
 902 }
 903
 904 static int decode_pg_temp(void **p, void *end, struct ceph_osdmap *map)
 905 {
 906         return __decode_pg_temp(p, end, map, false);
 907 }
 908
 909 static int decode_new_pg_temp(void **p, void *end, struct ceph_osdmap *map)
 910 {
 911         return __decode_pg_temp(p, end, map, true);
 912 }
 913
 914 static int __decode_primary_temp(void **p, void *end, struct ceph_osdmap *map,
 915                                  bool incremental)
 916 {
 917         u32 n;
 918
 919         ceph_decode_32_safe(p, end, n, e_inval);
 920         while (n--) {
 921                 struct ceph_pg pgid;
 922                 u32 osd;
 923                 int ret;
 924
 925                 ret = ceph_decode_pgid(p, end, &pgid);
 926                 if (ret)
 927                         return ret;
 928
 929                 ceph_decode_32_safe(p, end, osd, e_inval);
 930
 931                 ret = __remove_pg_mapping(&map->primary_temp, pgid);
 932                 BUG_ON(!incremental && ret != -ENOENT);
 933
 934                 if (!incremental || osd != (u32)-1) {
 935                         struct ceph_pg_mapping *pg;
 936
 937                         pg = kzalloc(sizeof(*pg), GFP_NOFS);
 938                         if (!pg)
 939                                 return -ENOMEM;
 940
 941                         pg->pgid = pgid;
 942                         pg->primary_temp.osd = osd;
 943
 944                         ret = __insert_pg_mapping(pg, &map->primary_temp);
 945                         if (ret) {
 946                                 kfree(pg);
 947                                 return ret;
 948                         }
 949                 }
 950         }
 951
 952         return 0;
 953
 954 e_inval:
 955         return -EINVAL;
 956 }
 957
 958 static int decode_primary_temp(void **p, void *end, struct ceph_osdmap *map)
 959 {
 960         return __decode_primary_temp(p, end, map, false);
 961 }
 962
 963 static int decode_new_primary_temp(void **p, void *end,
 964                                    struct ceph_osdmap *map)
 965 {
 966         return __decode_primary_temp(p, end, map, true);
 967 }
 968
 969 u32 ceph_get_primary_affinity(struct ceph_osdmap *map, int osd)
 970 {
 971         BUG_ON(osd >= map->max_osd);
 972
 973         if (!map->osd_primary_affinity)
 974                 return CEPH_OSD_DEFAULT_PRIMARY_AFFINITY;
 975
 976         return map->osd_primary_affinity[osd];
 977 }
 978
 979 static int set_primary_affinity(struct ceph_osdmap *map, int osd, u32 aff)
 980 {
 981         BUG_ON(osd >= map->max_osd);
 982
 983         if (!map->osd_primary_affinity) {
 984                 int i;
 985
 986                 map->osd_primary_affinity = kmalloc(map->max_osd*sizeof(u32),
 987                                                     GFP_NOFS);
 988                 if (!map->osd_primary_affinity)
 989                         return -ENOMEM;
 990
 991                 for (i = 0; i < map->max_osd; i++)
 992                         map->osd_primary_affinity[i] =
 993                             CEPH_OSD_DEFAULT_PRIMARY_AFFINITY;
 994         }
 995
 996         map->osd_primary_affinity[osd] = aff;
 997
 998         return 0;
 999 }
1000
1001 static int decode_primary_affinity(void **p, void *end,
1002                                    struct ceph_osdmap *map)
1003 {
1004         u32 len, i;
1005
1006         ceph_decode_32_safe(p, end, len, e_inval);
1007         if (len == 0) {
1008                 kfree(map->osd_primary_affinity);
1009                 map->osd_primary_affinity = NULL;
1010                 return 0;
1011         }
1012         if (len != map->max_osd)
1013                 goto e_inval;
1014
1015         ceph_decode_need(p, end, map->max_osd*sizeof(u32), e_inval);
1016
1017         for (i = 0; i < map->max_osd; i++) {
1018                 int ret;
1019
1020                 ret = set_primary_affinity(map, i, ceph_decode_32(p));
1021                 if (ret)
1022                         return ret;
1023         }
1024
1025         return 0;
1026
1027 e_inval:
1028         return -EINVAL;
1029 }
1030
1031 static int decode_new_primary_affinity(void **p, void *end,
1032                                        struct ceph_osdmap *map)
1033 {
1034         u32 n;
1035
1036         ceph_decode_32_safe(p, end, n, e_inval);
1037         while (n--) {
1038                 u32 osd, aff;
1039                 int ret;
1040
1041                 ceph_decode_32_safe(p, end, osd, e_inval);
1042                 ceph_decode_32_safe(p, end, aff, e_inval);
1043
1044                 ret = set_primary_affinity(map, osd, aff);
1045                 if (ret)
1046                         return ret;
1047
1048                 pr_info("osd%d primary-affinity 0x%x\n", osd, aff);
1049         }
1050
1051         return 0;
1052
1053 e_inval:
1054         return -EINVAL;
1055 }
1056
1057 /*
1058  * decode a full map.
1059  */
1060 static int osdmap_decode(void **p, void *end, struct ceph_osdmap *map)
1061 {
1062         u8 struct_v;
1063         u32 epoch = 0;
1064         void *start = *p;
1065         u32 max;
1066         u32 len, i;
1067         int err;
1068
1069         dout("%s %p to %p len %d\n", __func__, *p, end, (int)(end - *p));
1070
1071         err = get_osdmap_client_data_v(p, end, "full", &struct_v);
1072         if (err)
1073                 goto bad;
1074
1075         /* fsid, epoch, created, modified */
1076         ceph_decode_need(p, end, sizeof(map->fsid) + sizeof(u32) +
1077                          sizeof(map->created) + sizeof(map->modified), e_inval);
1078         ceph_decode_copy(p, &map->fsid, sizeof(map->fsid));
1079         epoch = map->epoch = ceph_decode_32(p);
1080         ceph_decode_copy(p, &map->created, sizeof(map->created));
1081         ceph_decode_copy(p, &map->modified, sizeof(map->modified));
1082
1083         /* pools */
1084         err = decode_pools(p, end, map);
1085         if (err)
1086                 goto bad;
1087
1088         /* pool_name */
1089         err = decode_pool_names(p, end, map);
1090         if (err)
1091                 goto bad;
1092
1093         ceph_decode_32_safe(p, end, map->pool_max, e_inval);
1094
1095         ceph_decode_32_safe(p, end, map->flags, e_inval);
1096
1097         /* max_osd */
1098         ceph_decode_32_safe(p, end, max, e_inval);
1099
1100         /* (re)alloc osd arrays */
1101         err = osdmap_set_max_osd(map, max);
1102         if (err)
1103                 goto bad;
1104
1105         /* osd_state, osd_weight, osd_addrs->client_addr */
1106         ceph_decode_need(p, end, 3*sizeof(u32) +
1107                          map->max_osd*(1 + sizeof(*map->osd_weight) +
1108                                        sizeof(*map->osd_addr)), e_inval);
1109
1110         if (ceph_decode_32(p) != map->max_osd)
1111                 goto e_inval;
1112
1113         ceph_decode_copy(p, map->osd_state, map->max_osd);
1114
1115         if (ceph_decode_32(p) != map->max_osd)
1116                 goto e_inval;
1117
1118         for (i = 0; i < map->max_osd; i++)
1119                 map->osd_weight[i] = ceph_decode_32(p);
1120
1121         if (ceph_decode_32(p) != map->max_osd)
1122                 goto e_inval;
1123
1124         ceph_decode_copy(p, map->osd_addr, map->max_osd*sizeof(*map->osd_addr));
1125         for (i = 0; i < map->max_osd; i++)
1126                 ceph_decode_addr(&map->osd_addr[i]);
1127
1128         /* pg_temp */
1129         err = decode_pg_temp(p, end, map);
1130         if (err)
1131                 goto bad;
1132
1133         /* primary_temp */
1134         if (struct_v >= 1) {
1135                 err = decode_primary_temp(p, end, map);
1136                 if (err)
1137                         goto bad;
1138         }
1139
1140         /* primary_affinity */
1141         if (struct_v >= 2) {
1142                 err = decode_primary_affinity(p, end, map);
1143                 if (err)
1144                         goto bad;
1145         } else {
1146                 /* XXX can this happen? */
1147                 kfree(map->osd_primary_affinity);
1148                 map->osd_primary_affinity = NULL;
1149         }
1150
1151         /* crush */
1152         ceph_decode_32_safe(p, end, len, e_inval);
1153         map->crush = crush_decode(*p, min(*p + len, end));
1154         if (IS_ERR(map->crush)) {
1155                 err = PTR_ERR(map->crush);
1156                 map->crush = NULL;
1157                 goto bad;
1158         }
1159         *p += len;
1160
1161         /* ignore the rest */
1162         *p = end;
1163
1164         dout("full osdmap epoch %d max_osd %d\n", map->epoch, map->max_osd);
1165         return 0;
1166
1167 e_inval:
1168         err = -EINVAL;
1169 bad:
1170         pr_err("corrupt full osdmap (%d) epoch %d off %d (%p of %p-%p)\n",
1171                err, epoch, (int)(*p - start), *p, start, end);
1172         print_hex_dump(KERN_DEBUG, "osdmap: ",
1173                        DUMP_PREFIX_OFFSET, 16, 1,
1174                        start, end - start, true);
1175         return err;
1176 }
1177
1178 /*
1179  * Allocate and decode a full map.
1180  */
1181 struct ceph_osdmap *ceph_osdmap_decode(void **p, void *end)
1182 {
1183         struct ceph_osdmap *map;
1184         int ret;
1185
1186         map = kzalloc(sizeof(*map), GFP_NOFS);
1187         if (!map)
1188                 return ERR_PTR(-ENOMEM);
1189
1190         map->pg_temp = RB_ROOT;
1191         map->primary_temp = RB_ROOT;
1192         mutex_init(&map->crush_scratch_mutex);
1193
1194         ret = osdmap_decode(p, end, map);
1195         if (ret) {
1196                 ceph_osdmap_destroy(map);
1197                 return ERR_PTR(ret);
1198         }
1199
1200         return map;
1201 }
1202
1203 /*
1204  * decode and apply an incremental map update.
1205  */
1206 struct ceph_osdmap *osdmap_apply_incremental(void **p, void *end,
1207                                              struct ceph_osdmap *map)
1208 {
1209         struct crush_map *newcrush = NULL;
1210         struct ceph_fsid fsid;
1211         u32 epoch = 0;
1212         struct ceph_timespec modified;
1213         s32 len;
1214         u64 pool;
1215         __s64 new_pool_max;
1216         __s32 new_flags, max;
1217         void *start = *p;
1218         int err;
1219         u8 struct_v;
1220
1221         dout("%s %p to %p len %d\n", __func__, *p, end, (int)(end - *p));
1222
1223         err = get_osdmap_client_data_v(p, end, "inc", &struct_v);
1224         if (err)
1225                 goto bad;
1226
1227         /* fsid, epoch, modified, new_pool_max, new_flags */
1228         ceph_decode_need(p, end, sizeof(fsid) + sizeof(u32) + sizeof(modified) +
1229                          sizeof(u64) + sizeof(u32), e_inval);
1230         ceph_decode_copy(p, &fsid, sizeof(fsid));
1231         epoch = ceph_decode_32(p);
1232         BUG_ON(epoch != map->epoch+1);
1233         ceph_decode_copy(p, &modified, sizeof(modified));
1234         new_pool_max = ceph_decode_64(p);
1235         new_flags = ceph_decode_32(p);
1236
1237         /* full map? */
1238         ceph_decode_32_safe(p, end, len, e_inval);
1239         if (len > 0) {
1240                 dout("apply_incremental full map len %d, %p to %p\n",
1241                      len, *p, end);
1242                 return ceph_osdmap_decode(p, min(*p+len, end));
1243         }
1244
1245         /* new crush? */
1246         ceph_decode_32_safe(p, end, len, e_inval);
1247         if (len > 0) {
1248                 newcrush = crush_decode(*p, min(*p+len, end));
1249                 if (IS_ERR(newcrush)) {
1250                         err = PTR_ERR(newcrush);
1251                         newcrush = NULL;
1252                         goto bad;
1253                 }
1254                 *p += len;
1255         }
1256
1257         /* new flags? */
1258         if (new_flags >= 0)
1259                 map->flags = new_flags;
1260         if (new_pool_max >= 0)
1261                 map->pool_max = new_pool_max;
1262
1263         /* new max? */
1264         ceph_decode_32_safe(p, end, max, e_inval);
1265         if (max >= 0) {
1266                 err = osdmap_set_max_osd(map, max);
1267                 if (err)
1268                         goto bad;
1269         }
1270
1271         map->epoch++;
1272         map->modified = modified;
1273         if (newcrush) {
1274                 if (map->crush)
1275                         crush_destroy(map->crush);
1276                 map->crush = newcrush;
1277                 newcrush = NULL;
1278         }
1279
1280         /* new_pools */
1281         err = decode_new_pools(p, end, map);
1282         if (err)
1283                 goto bad;
1284
1285         /* new_pool_names */
1286         err = decode_pool_names(p, end, map);
1287         if (err)
1288                 goto bad;
1289
1290         /* old_pool */
1291         ceph_decode_32_safe(p, end, len, e_inval);
1292         while (len--) {
1293                 struct ceph_pg_pool_info *pi;
1294
1295                 ceph_decode_64_safe(p, end, pool, e_inval);
1296                 pi = __lookup_pg_pool(&map->pg_pools, pool);
1297                 if (pi)
1298                         __remove_pg_pool(&map->pg_pools, pi);
1299         }
1300
1301         /* new_up */
1302         ceph_decode_32_safe(p, end, len, e_inval);
1303         while (len--) {
1304                 u32 osd;
1305                 struct ceph_entity_addr addr;
1306                 ceph_decode_32_safe(p, end, osd, e_inval);
1307                 ceph_decode_copy_safe(p, end, &addr, sizeof(addr), e_inval);
1308                 ceph_decode_addr(&addr);
1309                 pr_info("osd%d up\n", osd);
1310                 BUG_ON(osd >= map->max_osd);
1311                 map->osd_state[osd] |= CEPH_OSD_UP | CEPH_OSD_EXISTS;
1312                 map->osd_addr[osd] = addr;
1313         }
1314
1315         /* new_state */
1316         ceph_decode_32_safe(p, end, len, e_inval);
1317         while (len--) {
1318                 u32 osd;
1319                 u8 xorstate;
1320                 ceph_decode_32_safe(p, end, osd, e_inval);
1321                 xorstate = **(u8 **)p;
1322                 (*p)++;  /* clean flag */
1323                 if (xorstate == 0)
1324                         xorstate = CEPH_OSD_UP;
1325                 if (xorstate & CEPH_OSD_UP)
1326                         pr_info("osd%d down\n", osd);
1327                 if (osd < map->max_osd)
1328                         map->osd_state[osd] ^= xorstate;
1329         }
1330
1331         /* new_weight */
1332         ceph_decode_32_safe(p, end, len, e_inval);
1333         while (len--) {
1334                 u32 osd, off;
1335                 ceph_decode_need(p, end, sizeof(u32)*2, e_inval);
1336                 osd = ceph_decode_32(p);
1337                 off = ceph_decode_32(p);
1338                 pr_info("osd%d weight 0x%x %s\n", osd, off,
1339                      off == CEPH_OSD_IN ? "(in)" :
1340                      (off == CEPH_OSD_OUT ? "(out)" : ""));
1341                 if (osd < map->max_osd)
1342                         map->osd_weight[osd] = off;
1343         }
1344
1345         /* new_pg_temp */
1346         err = decode_new_pg_temp(p, end, map);
1347         if (err)
1348                 goto bad;
1349
1350         /* new_primary_temp */
1351         if (struct_v >= 1) {
1352                 err = decode_new_primary_temp(p, end, map);
1353                 if (err)
1354                         goto bad;
1355         }
1356
1357         /* new_primary_affinity */
1358         if (struct_v >= 2) {
1359                 err = decode_new_primary_affinity(p, end, map);
1360                 if (err)
1361                         goto bad;
1362         }
1363
1364         /* ignore the rest */
1365         *p = end;
1366
1367         dout("inc osdmap epoch %d max_osd %d\n", map->epoch, map->max_osd);
1368         return map;
1369
1370 e_inval:
1371         err = -EINVAL;
1372 bad:
1373         pr_err("corrupt inc osdmap (%d) epoch %d off %d (%p of %p-%p)\n",
1374                err, epoch, (int)(*p - start), *p, start, end);
1375         print_hex_dump(KERN_DEBUG, "osdmap: ",
1376                        DUMP_PREFIX_OFFSET, 16, 1,
1377                        start, end - start, true);
1378         if (newcrush)
1379                 crush_destroy(newcrush);
1380         return ERR_PTR(err);
1381 }
1382
1383 void ceph_oid_copy(struct ceph_object_id *dest,
1384                    const struct ceph_object_id *src)
1385 {
1386         WARN_ON(!ceph_oid_empty(dest));
1387
1388         if (src->name != src->inline_name) {
1389                 /* very rare, see ceph_object_id definition */
1390                 dest->name = kmalloc(src->name_len + 1,
1391                                      GFP_NOIO | __GFP_NOFAIL);
1392         }
1393
1394         memcpy(dest->name, src->name, src->name_len + 1);
1395         dest->name_len = src->name_len;
1396 }
1397 EXPORT_SYMBOL(ceph_oid_copy);
1398
1399 static __printf(2, 0)
1400 int oid_printf_vargs(struct ceph_object_id *oid, const char *fmt, va_list ap)
1401 {
1402         int len;
1403
1404         WARN_ON(!ceph_oid_empty(oid));
1405
1406         len = vsnprintf(oid->inline_name, sizeof(oid->inline_name), fmt, ap);
1407         if (len >= sizeof(oid->inline_name))
1408                 return len;
1409
1410         oid->name_len = len;
1411         return 0;
1412 }
1413
1414 /*
1415  * If oid doesn't fit into inline buffer, BUG.
1416  */
1417 void ceph_oid_printf(struct ceph_object_id *oid, const char *fmt, ...)
1418 {
1419         va_list ap;
1420
1421         va_start(ap, fmt);
1422         BUG_ON(oid_printf_vargs(oid, fmt, ap));
1423         va_end(ap);
1424 }
1425 EXPORT_SYMBOL(ceph_oid_printf);
1426
1427 static __printf(3, 0)
1428 int oid_aprintf_vargs(struct ceph_object_id *oid, gfp_t gfp,
1429                       const char *fmt, va_list ap)
1430 {
1431         va_list aq;
1432         int len;
1433
1434         va_copy(aq, ap);
1435         len = oid_printf_vargs(oid, fmt, aq);
1436         va_end(aq);
1437
1438         if (len) {
1439                 char *external_name;
1440
1441                 external_name = kmalloc(len + 1, gfp);
1442                 if (!external_name)
1443                         return -ENOMEM;
1444
1445                 oid->name = external_name;
1446                 WARN_ON(vsnprintf(oid->name, len + 1, fmt, ap) != len);
1447                 oid->name_len = len;
1448         }
1449
1450         return 0;
1451 }
1452
1453 /*
1454  * If oid doesn't fit into inline buffer, allocate.
1455  */
1456 int ceph_oid_aprintf(struct ceph_object_id *oid, gfp_t gfp,
1457                      const char *fmt, ...)
1458 {
1459         va_list ap;
1460         int ret;
1461
1462         va_start(ap, fmt);
1463         ret = oid_aprintf_vargs(oid, gfp, fmt, ap);
1464         va_end(ap);
1465
1466         return ret;
1467 }
1468 EXPORT_SYMBOL(ceph_oid_aprintf);
1469
1470 void ceph_oid_destroy(struct ceph_object_id *oid)
1471 {
1472         if (oid->name != oid->inline_name)
1473                 kfree(oid->name);
1474 }
1475 EXPORT_SYMBOL(ceph_oid_destroy);
1476
1477 /*
1478  * calculate file layout from given offset, length.
1479  * fill in correct oid, logical length, and object extent
1480  * offset, length.
1481  *
1482  * for now, we write only a single su, until we can
1483  * pass a stride back to the caller.
1484  */
1485 int ceph_calc_file_object_mapping(struct ceph_file_layout *layout,
1486                                    u64 off, u64 len,
1487                                    u64 *ono,
1488                                    u64 *oxoff, u64 *oxlen)
1489 {
1490         u32 osize = le32_to_cpu(layout->fl_object_size);
1491         u32 su = le32_to_cpu(layout->fl_stripe_unit);
1492         u32 sc = le32_to_cpu(layout->fl_stripe_count);
1493         u32 bl, stripeno, stripepos, objsetno;
1494         u32 su_per_object;
1495         u64 t, su_offset;
1496
1497         dout("mapping %llu~%llu  osize %u fl_su %u\n", off, len,
1498              osize, su);
1499         if (su == 0 || sc == 0)
1500                 goto invalid;
1501         su_per_object = osize / su;
1502         if (su_per_object == 0)
1503                 goto invalid;
1504         dout("osize %u / su %u = su_per_object %u\n", osize, su,
1505              su_per_object);
1506
1507         if ((su & ~PAGE_MASK) != 0)
1508                 goto invalid;
1509
1510         /* bl = *off / su; */
1511         t = off;
1512         do_div(t, su);
1513         bl = t;
1514         dout("off %llu / su %u = bl %u\n", off, su, bl);
1515
1516         stripeno = bl / sc;
1517         stripepos = bl % sc;
1518         objsetno = stripeno / su_per_object;
1519
1520         *ono = objsetno * sc + stripepos;
1521         dout("objset %u * sc %u = ono %u\n", objsetno, sc, (unsigned int)*ono);
1522
1523         /* *oxoff = *off % layout->fl_stripe_unit;  # offset in su */
1524         t = off;
1525         su_offset = do_div(t, su);
1526         *oxoff = su_offset + (stripeno % su_per_object) * su;
1527
1528         /*
1529          * Calculate the length of the extent being written to the selected
1530          * object. This is the minimum of the full length requested (len) or
1531          * the remainder of the current stripe being written to.
1532          */
1533         *oxlen = min_t(u64, len, su - su_offset);
1534
1535         dout(" obj extent %llu~%llu\n", *oxoff, *oxlen);
1536         return 0;
1537
1538 invalid:
1539         dout(" invalid layout\n");
1540         *ono = 0;
1541         *oxoff = 0;
1542         *oxlen = 0;
1543         return -EINVAL;
1544 }
1545 EXPORT_SYMBOL(ceph_calc_file_object_mapping);
1546
1547 /*
1548  * Calculate mapping of a (oloc, oid) pair to a PG.  Should only be
1549  * called with target's (oloc, oid), since tiering isn't taken into
1550  * account.
1551  */
1552 int ceph_oloc_oid_to_pg(struct ceph_osdmap *osdmap,
1553                         struct ceph_object_locator *oloc,
1554                         struct ceph_object_id *oid,
1555                         struct ceph_pg *pg_out)
1556 {
1557         struct ceph_pg_pool_info *pi;
1558
1559         pi = __lookup_pg_pool(&osdmap->pg_pools, oloc->pool);
1560         if (!pi)
1561                 return -EIO;
1562
1563         pg_out->pool = oloc->pool;
1564         pg_out->seed = ceph_str_hash(pi->object_hash, oid->name,
1565                                      oid->name_len);
1566
1567         dout("%s %*pE pgid %llu.%x\n", __func__, oid->name_len, oid->name,
1568              pg_out->pool, pg_out->seed);
1569         return 0;
1570 }
1571 EXPORT_SYMBOL(ceph_oloc_oid_to_pg);
1572
1573 static int do_crush(struct ceph_osdmap *map, int ruleno, int x,
1574                     int *result, int result_max,
1575                     const __u32 *weight, int weight_max)
1576 {
1577         int r;
1578
1579         BUG_ON(result_max > CEPH_PG_MAX_SIZE);
1580
1581         mutex_lock(&map->crush_scratch_mutex);
1582         r = crush_do_rule(map->crush, ruleno, x, result, result_max,
1583                           weight, weight_max, map->crush_scratch_ary);
1584         mutex_unlock(&map->crush_scratch_mutex);
1585
1586         return r;
1587 }
1588
1589 /*
1590  * Calculate raw (crush) set for given pgid.
1591  *
1592  * Return raw set length, or error.
1593  */
1594 static int pg_to_raw_osds(struct ceph_osdmap *osdmap,
1595                           struct ceph_pg_pool_info *pool,
1596                           struct ceph_pg pgid, u32 pps, int *osds)
1597 {
1598         int ruleno;
1599         int len;
1600
1601         /* crush */
1602         ruleno = crush_find_rule(osdmap->crush, pool->crush_ruleset,
1603                                  pool->type, pool->size);
1604         if (ruleno < 0) {
1605                 pr_err("no crush rule: pool %lld ruleset %d type %d size %d\n",
1606                        pgid.pool, pool->crush_ruleset, pool->type,
1607                        pool->size);
1608                 return -ENOENT;
1609         }
1610
1611         len = do_crush(osdmap, ruleno, pps, osds,
1612                        min_t(int, pool->size, CEPH_PG_MAX_SIZE),
1613                        osdmap->osd_weight, osdmap->max_osd);
1614         if (len < 0) {
1615                 pr_err("error %d from crush rule %d: pool %lld ruleset %d type %d size %d\n",
1616                        len, ruleno, pgid.pool, pool->crush_ruleset,
1617                        pool->type, pool->size);
1618                 return len;
1619         }
1620
1621         return len;
1622 }
1623
1624 /*
1625  * Given raw set, calculate up set and up primary.
1626  *
1627  * Return up set length.  *primary is set to up primary osd id, or -1
1628  * if up set is empty.
1629  */
1630 static int raw_to_up_osds(struct ceph_osdmap *osdmap,
1631                           struct ceph_pg_pool_info *pool,
1632                           int *osds, int len, int *primary)
1633 {
1634         int up_primary = -1;
1635         int i;
1636
1637         if (ceph_can_shift_osds(pool)) {
1638                 int removed = 0;
1639
1640                 for (i = 0; i < len; i++) {
1641                         if (ceph_osd_is_down(osdmap, osds[i])) {
1642                                 removed++;
1643                                 continue;
1644                         }
1645                         if (removed)
1646                                 osds[i - removed] = osds[i];
1647                 }
1648
1649                 len -= removed;
1650                 if (len > 0)
1651                         up_primary = osds[0];
1652         } else {
1653                 for (i = len - 1; i >= 0; i--) {
1654                         if (ceph_osd_is_down(osdmap, osds[i]))
1655                                 osds[i] = CRUSH_ITEM_NONE;
1656                         else
1657                                 up_primary = osds[i];
1658                 }
1659         }
1660
1661         *primary = up_primary;
1662         return len;
1663 }
1664
1665 static void apply_primary_affinity(struct ceph_osdmap *osdmap, u32 pps,
1666                                    struct ceph_pg_pool_info *pool,
1667                                    int *osds, int len, int *primary)
1668 {
1669         int i;
1670         int pos = -1;
1671
1672         /*
1673          * Do we have any non-default primary_affinity values for these
1674          * osds?
1675          */
1676         if (!osdmap->osd_primary_affinity)
1677                 return;
1678
1679         for (i = 0; i < len; i++) {
1680                 int osd = osds[i];
1681
1682                 if (osd != CRUSH_ITEM_NONE &&
1683                     osdmap->osd_primary_affinity[osd] !=
1684                                         CEPH_OSD_DEFAULT_PRIMARY_AFFINITY) {
1685                         break;
1686                 }
1687         }
1688         if (i == len)
1689                 return;
1690
1691         /*
1692          * Pick the primary.  Feed both the seed (for the pg) and the
1693          * osd into the hash/rng so that a proportional fraction of an
1694          * osd's pgs get rejected as primary.
1695          */
1696         for (i = 0; i < len; i++) {
1697                 int osd = osds[i];
1698                 u32 aff;
1699
1700                 if (osd == CRUSH_ITEM_NONE)
1701                         continue;
1702
1703                 aff = osdmap->osd_primary_affinity[osd];
1704                 if (aff < CEPH_OSD_MAX_PRIMARY_AFFINITY &&
1705                     (crush_hash32_2(CRUSH_HASH_RJENKINS1,
1706                                     pps, osd) >> 16) >= aff) {
1707                         /*
1708                          * We chose not to use this primary.  Note it
1709                          * anyway as a fallback in case we don't pick
1710                          * anyone else, but keep looking.
1711                          */
1712                         if (pos < 0)
1713                                 pos = i;
1714                 } else {
1715                         pos = i;
1716                         break;
1717                 }
1718         }
1719         if (pos < 0)
1720                 return;
1721
1722         *primary = osds[pos];
1723
1724         if (ceph_can_shift_osds(pool) && pos > 0) {
1725                 /* move the new primary to the front */
1726                 for (i = pos; i > 0; i--)
1727                         osds[i] = osds[i - 1];
1728                 osds[0] = *primary;
1729         }
1730 }
1731
1732 /*
1733  * Given up set, apply pg_temp and primary_temp mappings.
1734  *
1735  * Return acting set length.  *primary is set to acting primary osd id,
1736  * or -1 if acting set is empty.
1737  */
1738 static int apply_temps(struct ceph_osdmap *osdmap,
1739                        struct ceph_pg_pool_info *pool, struct ceph_pg pgid,
1740                        int *osds, int len, int *primary)
1741 {
1742         struct ceph_pg_mapping *pg;
1743         int temp_len;
1744         int temp_primary;
1745         int i;
1746
1747         /* raw_pg -> pg */
1748         pgid.seed = ceph_stable_mod(pgid.seed, pool->pg_num,
1749                                     pool->pg_num_mask);
1750
1751         /* pg_temp? */
1752         pg = __lookup_pg_mapping(&osdmap->pg_temp, pgid);
1753         if (pg) {
1754                 temp_len = 0;
1755                 temp_primary = -1;
1756
1757                 for (i = 0; i < pg->pg_temp.len; i++) {
1758                         if (ceph_osd_is_down(osdmap, pg->pg_temp.osds[i])) {
1759                                 if (ceph_can_shift_osds(pool))
1760                                         continue;
1761                                 else
1762                                         osds[temp_len++] = CRUSH_ITEM_NONE;
1763                         } else {
1764                                 osds[temp_len++] = pg->pg_temp.osds[i];
1765                         }
1766                 }
1767
1768                 /* apply pg_temp's primary */
1769                 for (i = 0; i < temp_len; i++) {
1770                         if (osds[i] != CRUSH_ITEM_NONE) {
1771                                 temp_primary = osds[i];
1772                                 break;
1773                         }
1774                 }
1775         } else {
1776                 temp_len = len;
1777                 temp_primary = *primary;
1778         }
1779
1780         /* primary_temp? */
1781         pg = __lookup_pg_mapping(&osdmap->primary_temp, pgid);
1782         if (pg)
1783                 temp_primary = pg->primary_temp.osd;
1784
1785         *primary = temp_primary;
1786         return temp_len;
1787 }
1788
1789 /*
1790  * Calculate acting set for given pgid.
1791  *
1792  * Return acting set length, or error.  *primary is set to acting
1793  * primary osd id, or -1 if acting set is empty or on error.
1794  */
1795 int ceph_calc_pg_acting(struct ceph_osdmap *osdmap, struct ceph_pg pgid,
1796                         int *osds, int *primary)
1797 {
1798         struct ceph_pg_pool_info *pool;
1799         u32 pps;
1800         int len;
1801
1802         pool = __lookup_pg_pool(&osdmap->pg_pools, pgid.pool);
1803         if (!pool) {
1804                 *primary = -1;
1805                 return -ENOENT;
1806         }
1807
1808         if (pool->flags & CEPH_POOL_FLAG_HASHPSPOOL) {
1809                 /* hash pool id and seed so that pool PGs do not overlap */
1810                 pps = crush_hash32_2(CRUSH_HASH_RJENKINS1,
1811                                      ceph_stable_mod(pgid.seed, pool->pgp_num,
1812                                                      pool->pgp_num_mask),
1813                                      pgid.pool);
1814         } else {
1815                 /*
1816                  * legacy behavior: add ps and pool together.  this is
1817                  * not a great approach because the PGs from each pool
1818                  * will overlap on top of each other: 0.5 == 1.4 ==
1819                  * 2.3 == ...
1820                  */
1821                 pps = ceph_stable_mod(pgid.seed, pool->pgp_num,
1822                                       pool->pgp_num_mask) +
1823                         (unsigned)pgid.pool;
1824         }
1825
1826         len = pg_to_raw_osds(osdmap, pool, pgid, pps, osds);
1827         if (len < 0) {
1828                 *primary = -1;
1829                 return len;
1830         }
1831
1832         len = raw_to_up_osds(osdmap, pool, osds, len, primary);
1833
1834         apply_primary_affinity(osdmap, pps, pool, osds, len, primary);
1835
1836         len = apply_temps(osdmap, pool, pgid, osds, len, primary);
1837
1838         return len;
1839 }
1840
1841 /*
1842  * Return primary osd for given pgid, or -1 if none.
1843  */
1844 int ceph_calc_pg_primary(struct ceph_osdmap *osdmap, struct ceph_pg pgid)
1845 {
1846         int osds[CEPH_PG_MAX_SIZE];
1847         int primary;
1848
1849         ceph_calc_pg_acting(osdmap, pgid, osds, &primary);
1850
1851         return primary;
1852 }
1853 EXPORT_SYMBOL(ceph_calc_pg_primary);