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                 err = -EINVAL;
 299                 ceph_decode_32_safe(p, end, yes, bad);
 300                 if (!yes) {
 301                         dout("crush_decode NO rule %d off %x %p to %p\n",
 302                              i, (int)(*p-start), *p, end);
 303                         c->rules[i] = NULL;
 304                         continue;
 305                 }
 306
 307                 dout("crush_decode rule %d off %x %p to %p\n",
 308                      i, (int)(*p-start), *p, end);
 309
 310                 /* len */
 311                 ceph_decode_32_safe(p, end, yes, bad);
 312 #if BITS_PER_LONG == 32
 313                 err = -EINVAL;
 314                 if (yes > (ULONG_MAX - sizeof(*r))
 315                           / sizeof(struct crush_rule_step))
 316                         goto bad;
 317 #endif
 318                 r = c->rules[i] = kmalloc(sizeof(*r) +
 319                                           yes*sizeof(struct crush_rule_step),
 320                                           GFP_NOFS);
 321                 if (r == NULL)
 322                         goto badmem;
 323                 dout(" rule %d is at %p\n", i, r);
 324                 r->len = yes;
 325                 ceph_decode_copy_safe(p, end, &r->mask, 4, bad); /* 4 u8's */
 326                 ceph_decode_need(p, end, r->len*3*sizeof(u32), bad);
 327                 for (j = 0; j < r->len; j++) {
 328                         r->steps[j].op = ceph_decode_32(p);
 329                         r->steps[j].arg1 = ceph_decode_32(p);
 330                         r->steps[j].arg2 = ceph_decode_32(p);
 331                 }
 332         }
 333
 334         /* ignore trailing name maps. */
 335         for (num_name_maps = 0; num_name_maps < 3; num_name_maps++) {
 336                 err = skip_name_map(p, end);
 337                 if (err < 0)
 338                         goto done;
 339         }
 340
 341         /* tunables */
 342         ceph_decode_need(p, end, 3*sizeof(u32), done);
 343         c->choose_local_tries = ceph_decode_32(p);
 344         c->choose_local_fallback_tries =  ceph_decode_32(p);
 345         c->choose_total_tries = ceph_decode_32(p);
 346         dout("crush decode tunable choose_local_tries = %d",
 347              c->choose_local_tries);
 348         dout("crush decode tunable choose_local_fallback_tries = %d",
 349              c->choose_local_fallback_tries);
 350         dout("crush decode tunable choose_total_tries = %d",
 351              c->choose_total_tries);
 352
 353         ceph_decode_need(p, end, sizeof(u32), done);
 354         c->chooseleaf_descend_once = ceph_decode_32(p);
 355         dout("crush decode tunable chooseleaf_descend_once = %d",
 356              c->chooseleaf_descend_once);
 357
 358         ceph_decode_need(p, end, sizeof(u8), done);
 359         c->chooseleaf_vary_r = ceph_decode_8(p);
 360         dout("crush decode tunable chooseleaf_vary_r = %d",
 361              c->chooseleaf_vary_r);
 362
 363 done:
 364         dout("crush_decode success\n");
 365         return c;
 366
 367 badmem:
 368         err = -ENOMEM;
 369 bad:
 370         dout("crush_decode fail %d\n", err);
 371         crush_destroy(c);
 372         return ERR_PTR(err);
 373 }
 374
 375 /*
 376  * rbtree of pg_mapping for handling pg_temp (explicit mapping of pgid
 377  * to a set of osds) and primary_temp (explicit primary setting)
 378  */
 379 static int pgid_cmp(struct ceph_pg l, struct ceph_pg r)
 380 {
 381         if (l.pool < r.pool)
 382                 return -1;
 383         if (l.pool > r.pool)
 384                 return 1;
 385         if (l.seed < r.seed)
 386                 return -1;
 387         if (l.seed > r.seed)
 388                 return 1;
 389         return 0;
 390 }
 391
 392 static int __insert_pg_mapping(struct ceph_pg_mapping *new,
 393                                struct rb_root *root)
 394 {
 395         struct rb_node **p = &root->rb_node;
 396         struct rb_node *parent = NULL;
 397         struct ceph_pg_mapping *pg = NULL;
 398         int c;
 399
 400         dout("__insert_pg_mapping %llx %p\n", *(u64 *)&new->pgid, new);
 401         while (*p) {
 402                 parent = *p;
 403                 pg = rb_entry(parent, struct ceph_pg_mapping, node);
 404                 c = pgid_cmp(new->pgid, pg->pgid);
 405                 if (c < 0)
 406                         p = &(*p)->rb_left;
 407                 else if (c > 0)
 408                         p = &(*p)->rb_right;
 409                 else
 410                         return -EEXIST;
 411         }
 412
 413         rb_link_node(&new->node, parent, p);
 414         rb_insert_color(&new->node, root);
 415         return 0;
 416 }
 417
 418 static struct ceph_pg_mapping *__lookup_pg_mapping(struct rb_root *root,
 419                                                    struct ceph_pg pgid)
 420 {
 421         struct rb_node *n = root->rb_node;
 422         struct ceph_pg_mapping *pg;
 423         int c;
 424
 425         while (n) {
 426                 pg = rb_entry(n, struct ceph_pg_mapping, node);
 427                 c = pgid_cmp(pgid, pg->pgid);
 428                 if (c < 0) {
 429                         n = n->rb_left;
 430                 } else if (c > 0) {
 431                         n = n->rb_right;
 432                 } else {
 433                         dout("__lookup_pg_mapping %lld.%x got %p\n",
 434                              pgid.pool, pgid.seed, pg);
 435                         return pg;
 436                 }
 437         }
 438         return NULL;
 439 }
 440
 441 static int __remove_pg_mapping(struct rb_root *root, struct ceph_pg pgid)
 442 {
 443         struct ceph_pg_mapping *pg = __lookup_pg_mapping(root, pgid);
 444
 445         if (pg) {
 446                 dout("__remove_pg_mapping %lld.%x %p\n", pgid.pool, pgid.seed,
 447                      pg);
 448                 rb_erase(&pg->node, root);
 449                 kfree(pg);
 450                 return 0;
 451         }
 452         dout("__remove_pg_mapping %lld.%x dne\n", pgid.pool, pgid.seed);
 453         return -ENOENT;
 454 }
 455
 456 /*
 457  * rbtree of pg pool info
 458  */
 459 static int __insert_pg_pool(struct rb_root *root, struct ceph_pg_pool_info *new)
 460 {
 461         struct rb_node **p = &root->rb_node;
 462         struct rb_node *parent = NULL;
 463         struct ceph_pg_pool_info *pi = NULL;
 464
 465         while (*p) {
 466                 parent = *p;
 467                 pi = rb_entry(parent, struct ceph_pg_pool_info, node);
 468                 if (new->id < pi->id)
 469                         p = &(*p)->rb_left;
 470                 else if (new->id > pi->id)
 471                         p = &(*p)->rb_right;
 472                 else
 473                         return -EEXIST;
 474         }
 475
 476         rb_link_node(&new->node, parent, p);
 477         rb_insert_color(&new->node, root);
 478         return 0;
 479 }
 480
 481 static struct ceph_pg_pool_info *__lookup_pg_pool(struct rb_root *root, u64 id)
 482 {
 483         struct ceph_pg_pool_info *pi;
 484         struct rb_node *n = root->rb_node;
 485
 486         while (n) {
 487                 pi = rb_entry(n, struct ceph_pg_pool_info, node);
 488                 if (id < pi->id)
 489                         n = n->rb_left;
 490                 else if (id > pi->id)
 491                         n = n->rb_right;
 492                 else
 493                         return pi;
 494         }
 495         return NULL;
 496 }
 497
 498 struct ceph_pg_pool_info *ceph_pg_pool_by_id(struct ceph_osdmap *map, u64 id)
 499 {
 500         return __lookup_pg_pool(&map->pg_pools, id);
 501 }
 502
 503 const char *ceph_pg_pool_name_by_id(struct ceph_osdmap *map, u64 id)
 504 {
 505         struct ceph_pg_pool_info *pi;
 506
 507         if (id == CEPH_NOPOOL)
 508                 return NULL;
 509
 510         if (WARN_ON_ONCE(id > (u64) INT_MAX))
 511                 return NULL;
 512
 513         pi = __lookup_pg_pool(&map->pg_pools, (int) id);
 514
 515         return pi ? pi->name : NULL;
 516 }
 517 EXPORT_SYMBOL(ceph_pg_pool_name_by_id);
 518
 519 int ceph_pg_poolid_by_name(struct ceph_osdmap *map, const char *name)
 520 {
 521         struct rb_node *rbp;
 522
 523         for (rbp = rb_first(&map->pg_pools); rbp; rbp = rb_next(rbp)) {
 524                 struct ceph_pg_pool_info *pi =
 525                         rb_entry(rbp, struct ceph_pg_pool_info, node);
 526                 if (pi->name && strcmp(pi->name, name) == 0)
 527                         return pi->id;
 528         }
 529         return -ENOENT;
 530 }
 531 EXPORT_SYMBOL(ceph_pg_poolid_by_name);
 532
 533 static void __remove_pg_pool(struct rb_root *root, struct ceph_pg_pool_info *pi)
 534 {
 535         rb_erase(&pi->node, root);
 536         kfree(pi->name);
 537         kfree(pi);
 538 }
 539
 540 static int decode_pool(void **p, void *end, struct ceph_pg_pool_info *pi)
 541 {
 542         u8 ev, cv;
 543         unsigned len, num;
 544         void *pool_end;
 545
 546         ceph_decode_need(p, end, 2 + 4, bad);
 547         ev = ceph_decode_8(p);  /* encoding version */
 548         cv = ceph_decode_8(p); /* compat version */
 549         if (ev < 5) {
 550                 pr_warn("got v %d < 5 cv %d of ceph_pg_pool\n", ev, cv);
 551                 return -EINVAL;
 552         }
 553         if (cv > 9) {
 554                 pr_warn("got v %d cv %d > 9 of ceph_pg_pool\n", ev, cv);
 555                 return -EINVAL;
 556         }
 557         len = ceph_decode_32(p);
 558         ceph_decode_need(p, end, len, bad);
 559         pool_end = *p + len;
 560
 561         pi->type = ceph_decode_8(p);
 562         pi->size = ceph_decode_8(p);
 563         pi->crush_ruleset = ceph_decode_8(p);
 564         pi->object_hash = ceph_decode_8(p);
 565
 566         pi->pg_num = ceph_decode_32(p);
 567         pi->pgp_num = ceph_decode_32(p);
 568
 569         *p += 4 + 4;  /* skip lpg* */
 570         *p += 4;      /* skip last_change */
 571         *p += 8 + 4;  /* skip snap_seq, snap_epoch */
 572
 573         /* skip snaps */
 574         num = ceph_decode_32(p);
 575         while (num--) {
 576                 *p += 8;  /* snapid key */
 577                 *p += 1 + 1; /* versions */
 578                 len = ceph_decode_32(p);
 579                 *p += len;
 580         }
 581
 582         /* skip removed_snaps */
 583         num = ceph_decode_32(p);
 584         *p += num * (8 + 8);
 585
 586         *p += 8;  /* skip auid */
 587         pi->flags = ceph_decode_64(p);
 588         *p += 4;  /* skip crash_replay_interval */
 589
 590         if (ev >= 7)
 591                 *p += 1;  /* skip min_size */
 592
 593         if (ev >= 8)
 594                 *p += 8 + 8;  /* skip quota_max_* */
 595
 596         if (ev >= 9) {
 597                 /* skip tiers */
 598                 num = ceph_decode_32(p);
 599                 *p += num * 8;
 600
 601                 *p += 8;  /* skip tier_of */
 602                 *p += 1;  /* skip cache_mode */
 603
 604                 pi->read_tier = ceph_decode_64(p);
 605                 pi->write_tier = ceph_decode_64(p);
 606         } else {
 607                 pi->read_tier = -1;
 608                 pi->write_tier = -1;
 609         }
 610
 611         /* ignore the rest */
 612
 613         *p = pool_end;
 614         calc_pg_masks(pi);
 615         return 0;
 616
 617 bad:
 618         return -EINVAL;
 619 }
 620
 621 static int decode_pool_names(void **p, void *end, struct ceph_osdmap *map)
 622 {
 623         struct ceph_pg_pool_info *pi;
 624         u32 num, len;
 625         u64 pool;
 626
 627         ceph_decode_32_safe(p, end, num, bad);
 628         dout(" %d pool names\n", num);
 629         while (num--) {
 630                 ceph_decode_64_safe(p, end, pool, bad);
 631                 ceph_decode_32_safe(p, end, len, bad);
 632                 dout("  pool %llu len %d\n", pool, len);
 633                 ceph_decode_need(p, end, len, bad);
 634                 pi = __lookup_pg_pool(&map->pg_pools, pool);
 635                 if (pi) {
 636                         char *name = kstrndup(*p, len, GFP_NOFS);
 637
 638                         if (!name)
 639                                 return -ENOMEM;
 640                         kfree(pi->name);
 641                         pi->name = name;
 642                         dout("  name is %s\n", pi->name);
 643                 }
 644                 *p += len;
 645         }
 646         return 0;
 647
 648 bad:
 649         return -EINVAL;
 650 }
 651
 652 /*
 653  * osd map
 654  */
 655 void ceph_osdmap_destroy(struct ceph_osdmap *map)
 656 {
 657         dout("osdmap_destroy %p\n", map);
 658         if (map->crush)
 659                 crush_destroy(map->crush);
 660         while (!RB_EMPTY_ROOT(&map->pg_temp)) {
 661                 struct ceph_pg_mapping *pg =
 662                         rb_entry(rb_first(&map->pg_temp),
 663                                  struct ceph_pg_mapping, node);
 664                 rb_erase(&pg->node, &map->pg_temp);
 665                 kfree(pg);
 666         }
 667         while (!RB_EMPTY_ROOT(&map->primary_temp)) {
 668                 struct ceph_pg_mapping *pg =
 669                         rb_entry(rb_first(&map->primary_temp),
 670                                  struct ceph_pg_mapping, node);
 671                 rb_erase(&pg->node, &map->primary_temp);
 672                 kfree(pg);
 673         }
 674         while (!RB_EMPTY_ROOT(&map->pg_pools)) {
 675                 struct ceph_pg_pool_info *pi =
 676                         rb_entry(rb_first(&map->pg_pools),
 677                                  struct ceph_pg_pool_info, node);
 678                 __remove_pg_pool(&map->pg_pools, pi);
 679         }
 680         kfree(map->osd_state);
 681         kfree(map->osd_weight);
 682         kfree(map->osd_addr);
 683         kfree(map->osd_primary_affinity);
 684         kfree(map);
 685 }
 686
 687 /*
 688  * Adjust max_osd value, (re)allocate arrays.
 689  *
 690  * The new elements are properly initialized.
 691  */
 692 static int osdmap_set_max_osd(struct ceph_osdmap *map, int max)
 693 {
 694         u8 *state;
 695         u32 *weight;
 696         struct ceph_entity_addr *addr;
 697         int i;
 698
 699         state = krealloc(map->osd_state, max*sizeof(*state), GFP_NOFS);
 700         if (!state)
 701                 return -ENOMEM;
 702         map->osd_state = state;
 703
 704         weight = krealloc(map->osd_weight, max*sizeof(*weight), GFP_NOFS);
 705         if (!weight)
 706                 return -ENOMEM;
 707         map->osd_weight = weight;
 708
 709         addr = krealloc(map->osd_addr, max*sizeof(*addr), GFP_NOFS);
 710         if (!addr)
 711                 return -ENOMEM;
 712         map->osd_addr = addr;
 713
 714         for (i = map->max_osd; i < max; i++) {
 715                 map->osd_state[i] = 0;
 716                 map->osd_weight[i] = CEPH_OSD_OUT;
 717                 memset(map->osd_addr + i, 0, sizeof(*map->osd_addr));
 718         }
 719
 720         if (map->osd_primary_affinity) {
 721                 u32 *affinity;
 722
 723                 affinity = krealloc(map->osd_primary_affinity,
 724                                     max*sizeof(*affinity), GFP_NOFS);
 725                 if (!affinity)
 726                         return -ENOMEM;
 727                 map->osd_primary_affinity = affinity;
 728
 729                 for (i = map->max_osd; i < max; i++)
 730                         map->osd_primary_affinity[i] =
 731                             CEPH_OSD_DEFAULT_PRIMARY_AFFINITY;
 732         }
 733
 734         map->max_osd = max;
 735
 736         return 0;
 737 }
 738
 739 #define OSDMAP_WRAPPER_COMPAT_VER       7
 740 #define OSDMAP_CLIENT_DATA_COMPAT_VER   1
 741
 742 /*
 743  * Return 0 or error.  On success, *v is set to 0 for old (v6) osdmaps,
 744  * to struct_v of the client_data section for new (v7 and above)
 745  * osdmaps.
 746  */
 747 static int get_osdmap_client_data_v(void **p, void *end,
 748                                     const char *prefix, u8 *v)
 749 {
 750         u8 struct_v;
 751
 752         ceph_decode_8_safe(p, end, struct_v, e_inval);
 753         if (struct_v >= 7) {
 754                 u8 struct_compat;
 755
 756                 ceph_decode_8_safe(p, end, struct_compat, e_inval);
 757                 if (struct_compat > OSDMAP_WRAPPER_COMPAT_VER) {
 758                         pr_warn("got v %d cv %d > %d of %s ceph_osdmap\n",
 759                                 struct_v, struct_compat,
 760                                 OSDMAP_WRAPPER_COMPAT_VER, prefix);
 761                         return -EINVAL;
 762                 }
 763                 *p += 4; /* ignore wrapper struct_len */
 764
 765                 ceph_decode_8_safe(p, end, struct_v, e_inval);
 766                 ceph_decode_8_safe(p, end, struct_compat, e_inval);
 767                 if (struct_compat > OSDMAP_CLIENT_DATA_COMPAT_VER) {
 768                         pr_warn("got v %d cv %d > %d of %s ceph_osdmap client data\n",
 769                                 struct_v, struct_compat,
 770                                 OSDMAP_CLIENT_DATA_COMPAT_VER, prefix);
 771                         return -EINVAL;
 772                 }
 773                 *p += 4; /* ignore client data struct_len */
 774         } else {
 775                 u16 version;
 776
 777                 *p -= 1;
 778                 ceph_decode_16_safe(p, end, version, e_inval);
 779                 if (version < 6) {
 780                         pr_warn("got v %d < 6 of %s ceph_osdmap\n",
 781                                 version, prefix);
 782                         return -EINVAL;
 783                 }
 784
 785                 /* old osdmap enconding */
 786                 struct_v = 0;
 787         }
 788
 789         *v = struct_v;
 790         return 0;
 791
 792 e_inval:
 793         return -EINVAL;
 794 }
 795
 796 static int __decode_pools(void **p, void *end, struct ceph_osdmap *map,
 797                           bool incremental)
 798 {
 799         u32 n;
 800
 801         ceph_decode_32_safe(p, end, n, e_inval);
 802         while (n--) {
 803                 struct ceph_pg_pool_info *pi;
 804                 u64 pool;
 805                 int ret;
 806
 807                 ceph_decode_64_safe(p, end, pool, e_inval);
 808
 809                 pi = __lookup_pg_pool(&map->pg_pools, pool);
 810                 if (!incremental || !pi) {
 811                         pi = kzalloc(sizeof(*pi), GFP_NOFS);
 812                         if (!pi)
 813                                 return -ENOMEM;
 814
 815                         pi->id = pool;
 816
 817                         ret = __insert_pg_pool(&map->pg_pools, pi);
 818                         if (ret) {
 819                                 kfree(pi);
 820                                 return ret;
 821                         }
 822                 }
 823
 824                 ret = decode_pool(p, end, pi);
 825                 if (ret)
 826                         return ret;
 827         }
 828
 829         return 0;
 830
 831 e_inval:
 832         return -EINVAL;
 833 }
 834
 835 static int decode_pools(void **p, void *end, struct ceph_osdmap *map)
 836 {
 837         return __decode_pools(p, end, map, false);
 838 }
 839
 840 static int decode_new_pools(void **p, void *end, struct ceph_osdmap *map)
 841 {
 842         return __decode_pools(p, end, map, true);
 843 }
 844
 845 static int __decode_pg_temp(void **p, void *end, struct ceph_osdmap *map,
 846                             bool incremental)
 847 {
 848         u32 n;
 849
 850         ceph_decode_32_safe(p, end, n, e_inval);
 851         while (n--) {
 852                 struct ceph_pg pgid;
 853                 u32 len, i;
 854                 int ret;
 855
 856                 ret = ceph_decode_pgid(p, end, &pgid);
 857                 if (ret)
 858                         return ret;
 859
 860                 ceph_decode_32_safe(p, end, len, e_inval);
 861
 862                 ret = __remove_pg_mapping(&map->pg_temp, pgid);
 863                 BUG_ON(!incremental && ret != -ENOENT);
 864
 865                 if (!incremental || len > 0) {
 866                         struct ceph_pg_mapping *pg;
 867
 868                         ceph_decode_need(p, end, len*sizeof(u32), e_inval);
 869
 870                         if (len > (UINT_MAX - sizeof(*pg)) / sizeof(u32))
 871                                 return -EINVAL;
 872
 873                         pg = kzalloc(sizeof(*pg) + len*sizeof(u32), GFP_NOFS);
 874                         if (!pg)
 875                                 return -ENOMEM;
 876
 877                         pg->pgid = pgid;
 878                         pg->pg_temp.len = len;
 879                         for (i = 0; i < len; i++)
 880                                 pg->pg_temp.osds[i] = ceph_decode_32(p);
 881
 882                         ret = __insert_pg_mapping(pg, &map->pg_temp);
 883                         if (ret) {
 884                                 kfree(pg);
 885                                 return ret;
 886                         }
 887                 }
 888         }
 889
 890         return 0;
 891
 892 e_inval:
 893         return -EINVAL;
 894 }
 895
 896 static int decode_pg_temp(void **p, void *end, struct ceph_osdmap *map)
 897 {
 898         return __decode_pg_temp(p, end, map, false);
 899 }
 900
 901 static int decode_new_pg_temp(void **p, void *end, struct ceph_osdmap *map)
 902 {
 903         return __decode_pg_temp(p, end, map, true);
 904 }
 905
 906 static int __decode_primary_temp(void **p, void *end, struct ceph_osdmap *map,
 907                                  bool incremental)
 908 {
 909         u32 n;
 910
 911         ceph_decode_32_safe(p, end, n, e_inval);
 912         while (n--) {
 913                 struct ceph_pg pgid;
 914                 u32 osd;
 915                 int ret;
 916
 917                 ret = ceph_decode_pgid(p, end, &pgid);
 918                 if (ret)
 919                         return ret;
 920
 921                 ceph_decode_32_safe(p, end, osd, e_inval);
 922
 923                 ret = __remove_pg_mapping(&map->primary_temp, pgid);
 924                 BUG_ON(!incremental && ret != -ENOENT);
 925
 926                 if (!incremental || osd != (u32)-1) {
 927                         struct ceph_pg_mapping *pg;
 928
 929                         pg = kzalloc(sizeof(*pg), GFP_NOFS);
 930                         if (!pg)
 931                                 return -ENOMEM;
 932
 933                         pg->pgid = pgid;
 934                         pg->primary_temp.osd = osd;
 935
 936                         ret = __insert_pg_mapping(pg, &map->primary_temp);
 937                         if (ret) {
 938                                 kfree(pg);
 939                                 return ret;
 940                         }
 941                 }
 942         }
 943
 944         return 0;
 945
 946 e_inval:
 947         return -EINVAL;
 948 }
 949
 950 static int decode_primary_temp(void **p, void *end, struct ceph_osdmap *map)
 951 {
 952         return __decode_primary_temp(p, end, map, false);
 953 }
 954
 955 static int decode_new_primary_temp(void **p, void *end,
 956                                    struct ceph_osdmap *map)
 957 {
 958         return __decode_primary_temp(p, end, map, true);
 959 }
 960
 961 u32 ceph_get_primary_affinity(struct ceph_osdmap *map, int osd)
 962 {
 963         BUG_ON(osd >= map->max_osd);
 964
 965         if (!map->osd_primary_affinity)
 966                 return CEPH_OSD_DEFAULT_PRIMARY_AFFINITY;
 967
 968         return map->osd_primary_affinity[osd];
 969 }
 970
 971 static int set_primary_affinity(struct ceph_osdmap *map, int osd, u32 aff)
 972 {
 973         BUG_ON(osd >= map->max_osd);
 974
 975         if (!map->osd_primary_affinity) {
 976                 int i;
 977
 978                 map->osd_primary_affinity = kmalloc(map->max_osd*sizeof(u32),
 979                                                     GFP_NOFS);
 980                 if (!map->osd_primary_affinity)
 981                         return -ENOMEM;
 982
 983                 for (i = 0; i < map->max_osd; i++)
 984                         map->osd_primary_affinity[i] =
 985                             CEPH_OSD_DEFAULT_PRIMARY_AFFINITY;
 986         }
 987
 988         map->osd_primary_affinity[osd] = aff;
 989
 990         return 0;
 991 }
 992
 993 static int decode_primary_affinity(void **p, void *end,
 994                                    struct ceph_osdmap *map)
 995 {
 996         u32 len, i;
 997
 998         ceph_decode_32_safe(p, end, len, e_inval);
 999         if (len == 0) {
1000                 kfree(map->osd_primary_affinity);
1001                 map->osd_primary_affinity = NULL;
1002                 return 0;
1003         }
1004         if (len != map->max_osd)
1005                 goto e_inval;
1006
1007         ceph_decode_need(p, end, map->max_osd*sizeof(u32), e_inval);
1008
1009         for (i = 0; i < map->max_osd; i++) {
1010                 int ret;
1011
1012                 ret = set_primary_affinity(map, i, ceph_decode_32(p));
1013                 if (ret)
1014                         return ret;
1015         }
1016
1017         return 0;
1018
1019 e_inval:
1020         return -EINVAL;
1021 }
1022
1023 static int decode_new_primary_affinity(void **p, void *end,
1024                                        struct ceph_osdmap *map)
1025 {
1026         u32 n;
1027
1028         ceph_decode_32_safe(p, end, n, e_inval);
1029         while (n--) {
1030                 u32 osd, aff;
1031                 int ret;
1032
1033                 ceph_decode_32_safe(p, end, osd, e_inval);
1034                 ceph_decode_32_safe(p, end, aff, e_inval);
1035
1036                 ret = set_primary_affinity(map, osd, aff);
1037                 if (ret)
1038                         return ret;
1039
1040                 pr_info("osd%d primary-affinity 0x%x\n", osd, aff);
1041         }
1042
1043         return 0;
1044
1045 e_inval:
1046         return -EINVAL;
1047 }
1048
1049 /*
1050  * decode a full map.
1051  */
1052 static int osdmap_decode(void **p, void *end, struct ceph_osdmap *map)
1053 {
1054         u8 struct_v;
1055         u32 epoch = 0;
1056         void *start = *p;
1057         u32 max;
1058         u32 len, i;
1059         int err;
1060
1061         dout("%s %p to %p len %d\n", __func__, *p, end, (int)(end - *p));
1062
1063         err = get_osdmap_client_data_v(p, end, "full", &struct_v);
1064         if (err)
1065                 goto bad;
1066
1067         /* fsid, epoch, created, modified */
1068         ceph_decode_need(p, end, sizeof(map->fsid) + sizeof(u32) +
1069                          sizeof(map->created) + sizeof(map->modified), e_inval);
1070         ceph_decode_copy(p, &map->fsid, sizeof(map->fsid));
1071         epoch = map->epoch = ceph_decode_32(p);
1072         ceph_decode_copy(p, &map->created, sizeof(map->created));
1073         ceph_decode_copy(p, &map->modified, sizeof(map->modified));
1074
1075         /* pools */
1076         err = decode_pools(p, end, map);
1077         if (err)
1078                 goto bad;
1079
1080         /* pool_name */
1081         err = decode_pool_names(p, end, map);
1082         if (err)
1083                 goto bad;
1084
1085         ceph_decode_32_safe(p, end, map->pool_max, e_inval);
1086
1087         ceph_decode_32_safe(p, end, map->flags, e_inval);
1088
1089         /* max_osd */
1090         ceph_decode_32_safe(p, end, max, e_inval);
1091
1092         /* (re)alloc osd arrays */
1093         err = osdmap_set_max_osd(map, max);
1094         if (err)
1095                 goto bad;
1096
1097         /* osd_state, osd_weight, osd_addrs->client_addr */
1098         ceph_decode_need(p, end, 3*sizeof(u32) +
1099                          map->max_osd*(1 + sizeof(*map->osd_weight) +
1100                                        sizeof(*map->osd_addr)), e_inval);
1101
1102         if (ceph_decode_32(p) != map->max_osd)
1103                 goto e_inval;
1104
1105         ceph_decode_copy(p, map->osd_state, map->max_osd);
1106
1107         if (ceph_decode_32(p) != map->max_osd)
1108                 goto e_inval;
1109
1110         for (i = 0; i < map->max_osd; i++)
1111                 map->osd_weight[i] = ceph_decode_32(p);
1112
1113         if (ceph_decode_32(p) != map->max_osd)
1114                 goto e_inval;
1115
1116         ceph_decode_copy(p, map->osd_addr, map->max_osd*sizeof(*map->osd_addr));
1117         for (i = 0; i < map->max_osd; i++)
1118                 ceph_decode_addr(&map->osd_addr[i]);
1119
1120         /* pg_temp */
1121         err = decode_pg_temp(p, end, map);
1122         if (err)
1123                 goto bad;
1124
1125         /* primary_temp */
1126         if (struct_v >= 1) {
1127                 err = decode_primary_temp(p, end, map);
1128                 if (err)
1129                         goto bad;
1130         }
1131
1132         /* primary_affinity */
1133         if (struct_v >= 2) {
1134                 err = decode_primary_affinity(p, end, map);
1135                 if (err)
1136                         goto bad;
1137         } else {
1138                 /* XXX can this happen? */
1139                 kfree(map->osd_primary_affinity);
1140                 map->osd_primary_affinity = NULL;
1141         }
1142
1143         /* crush */
1144         ceph_decode_32_safe(p, end, len, e_inval);
1145         map->crush = crush_decode(*p, min(*p + len, end));
1146         if (IS_ERR(map->crush)) {
1147                 err = PTR_ERR(map->crush);
1148                 map->crush = NULL;
1149                 goto bad;
1150         }
1151         *p += len;
1152
1153         /* ignore the rest */
1154         *p = end;
1155
1156         dout("full osdmap epoch %d max_osd %d\n", map->epoch, map->max_osd);
1157         return 0;
1158
1159 e_inval:
1160         err = -EINVAL;
1161 bad:
1162         pr_err("corrupt full osdmap (%d) epoch %d off %d (%p of %p-%p)\n",
1163                err, epoch, (int)(*p - start), *p, start, end);
1164         print_hex_dump(KERN_DEBUG, "osdmap: ",
1165                        DUMP_PREFIX_OFFSET, 16, 1,
1166                        start, end - start, true);
1167         return err;
1168 }
1169
1170 /*
1171  * Allocate and decode a full map.
1172  */
1173 struct ceph_osdmap *ceph_osdmap_decode(void **p, void *end)
1174 {
1175         struct ceph_osdmap *map;
1176         int ret;
1177
1178         map = kzalloc(sizeof(*map), GFP_NOFS);
1179         if (!map)
1180                 return ERR_PTR(-ENOMEM);
1181
1182         map->pg_temp = RB_ROOT;
1183         map->primary_temp = RB_ROOT;
1184         mutex_init(&map->crush_scratch_mutex);
1185
1186         ret = osdmap_decode(p, end, map);
1187         if (ret) {
1188                 ceph_osdmap_destroy(map);
1189                 return ERR_PTR(ret);
1190         }
1191
1192         return map;
1193 }
1194
1195 /*
1196  * Encoding order is (new_up_client, new_state, new_weight).  Need to
1197  * apply in the (new_weight, new_state, new_up_client) order, because
1198  * an incremental map may look like e.g.
1199  *
1200  *     new_up_client: { osd=6, addr=... } # set osd_state and addr
1201  *     new_state: { osd=6, xorstate=EXISTS } # clear osd_state
1202  */
1203 static int decode_new_up_state_weight(void **p, void *end,
1204                                       struct ceph_osdmap *map)
1205 {
1206         void *new_up_client;
1207         void *new_state;
1208         void *new_weight_end;
1209         u32 len;
1210
1211         new_up_client = *p;
1212         ceph_decode_32_safe(p, end, len, e_inval);
1213         len *= sizeof(u32) + sizeof(struct ceph_entity_addr);
1214         ceph_decode_need(p, end, len, e_inval);
1215         *p += len;
1216
1217         new_state = *p;
1218         ceph_decode_32_safe(p, end, len, e_inval);
1219         len *= sizeof(u32) + sizeof(u8);
1220         ceph_decode_need(p, end, len, e_inval);
1221         *p += len;
1222
1223         /* new_weight */
1224         ceph_decode_32_safe(p, end, len, e_inval);
1225         while (len--) {
1226                 s32 osd;
1227                 u32 w;
1228
1229                 ceph_decode_need(p, end, 2*sizeof(u32), e_inval);
1230                 osd = ceph_decode_32(p);
1231                 w = ceph_decode_32(p);
1232                 BUG_ON(osd >= map->max_osd);
1233                 pr_info("osd%d weight 0x%x %s\n", osd, w,
1234                      w == CEPH_OSD_IN ? "(in)" :
1235                      (w == CEPH_OSD_OUT ? "(out)" : ""));
1236                 map->osd_weight[osd] = w;
1237
1238                 /*
1239                  * If we are marking in, set the EXISTS, and clear the
1240                  * AUTOOUT and NEW bits.
1241                  */
1242                 if (w) {
1243                         map->osd_state[osd] |= CEPH_OSD_EXISTS;
1244                         map->osd_state[osd] &= ~(CEPH_OSD_AUTOOUT |
1245                                                  CEPH_OSD_NEW);
1246                 }
1247         }
1248         new_weight_end = *p;
1249
1250         /* new_state (up/down) */
1251         *p = new_state;
1252         len = ceph_decode_32(p);
1253         while (len--) {
1254                 s32 osd;
1255                 u8 xorstate;
1256                 int ret;
1257
1258                 osd = ceph_decode_32(p);
1259                 xorstate = ceph_decode_8(p);
1260                 if (xorstate == 0)
1261                         xorstate = CEPH_OSD_UP;
1262                 BUG_ON(osd >= map->max_osd);
1263                 if ((map->osd_state[osd] & CEPH_OSD_UP) &&
1264                     (xorstate & CEPH_OSD_UP))
1265                         pr_info("osd%d down\n", osd);
1266                 if ((map->osd_state[osd] & CEPH_OSD_EXISTS) &&
1267                     (xorstate & CEPH_OSD_EXISTS)) {
1268                         pr_info("osd%d does not exist\n", osd);
1269                         ret = set_primary_affinity(map, osd,
1270                                                    CEPH_OSD_DEFAULT_PRIMARY_AFFINITY);
1271                         if (ret)
1272                                 return ret;
1273                         memset(map->osd_addr + osd, 0, sizeof(*map->osd_addr));
1274                         map->osd_state[osd] = 0;
1275                 } else {
1276                         map->osd_state[osd] ^= xorstate;
1277                 }
1278         }
1279
1280         /* new_up_client */
1281         *p = new_up_client;
1282         len = ceph_decode_32(p);
1283         while (len--) {
1284                 s32 osd;
1285                 struct ceph_entity_addr addr;
1286
1287                 osd = ceph_decode_32(p);
1288                 ceph_decode_copy(p, &addr, sizeof(addr));
1289                 ceph_decode_addr(&addr);
1290                 BUG_ON(osd >= map->max_osd);
1291                 pr_info("osd%d up\n", osd);
1292                 map->osd_state[osd] |= CEPH_OSD_EXISTS | CEPH_OSD_UP;
1293                 map->osd_addr[osd] = addr;
1294         }
1295
1296         *p = new_weight_end;
1297         return 0;
1298
1299 e_inval:
1300         return -EINVAL;
1301 }
1302
1303 /*
1304  * decode and apply an incremental map update.
1305  */
1306 struct ceph_osdmap *osdmap_apply_incremental(void **p, void *end,
1307                                              struct ceph_osdmap *map,
1308                                              struct ceph_messenger *msgr)
1309 {
1310         struct crush_map *newcrush = NULL;
1311         struct ceph_fsid fsid;
1312         u32 epoch = 0;
1313         struct ceph_timespec modified;
1314         s32 len;
1315         u64 pool;
1316         __s64 new_pool_max;
1317         __s32 new_flags, max;
1318         void *start = *p;
1319         int err;
1320         u8 struct_v;
1321
1322         dout("%s %p to %p len %d\n", __func__, *p, end, (int)(end - *p));
1323
1324         err = get_osdmap_client_data_v(p, end, "inc", &struct_v);
1325         if (err)
1326                 goto bad;
1327
1328         /* fsid, epoch, modified, new_pool_max, new_flags */
1329         ceph_decode_need(p, end, sizeof(fsid) + sizeof(u32) + sizeof(modified) +
1330                          sizeof(u64) + sizeof(u32), e_inval);
1331         ceph_decode_copy(p, &fsid, sizeof(fsid));
1332         epoch = ceph_decode_32(p);
1333         BUG_ON(epoch != map->epoch+1);
1334         ceph_decode_copy(p, &modified, sizeof(modified));
1335         new_pool_max = ceph_decode_64(p);
1336         new_flags = ceph_decode_32(p);
1337
1338         /* full map? */
1339         ceph_decode_32_safe(p, end, len, e_inval);
1340         if (len > 0) {
1341                 dout("apply_incremental full map len %d, %p to %p\n",
1342                      len, *p, end);
1343                 return ceph_osdmap_decode(p, min(*p+len, end));
1344         }
1345
1346         /* new crush? */
1347         ceph_decode_32_safe(p, end, len, e_inval);
1348         if (len > 0) {
1349                 newcrush = crush_decode(*p, min(*p+len, end));
1350                 if (IS_ERR(newcrush)) {
1351                         err = PTR_ERR(newcrush);
1352                         newcrush = NULL;
1353                         goto bad;
1354                 }
1355                 *p += len;
1356         }
1357
1358         /* new flags? */
1359         if (new_flags >= 0)
1360                 map->flags = new_flags;
1361         if (new_pool_max >= 0)
1362                 map->pool_max = new_pool_max;
1363
1364         /* new max? */
1365         ceph_decode_32_safe(p, end, max, e_inval);
1366         if (max >= 0) {
1367                 err = osdmap_set_max_osd(map, max);
1368                 if (err)
1369                         goto bad;
1370         }
1371
1372         map->epoch++;
1373         map->modified = modified;
1374         if (newcrush) {
1375                 if (map->crush)
1376                         crush_destroy(map->crush);
1377                 map->crush = newcrush;
1378                 newcrush = NULL;
1379         }
1380
1381         /* new_pools */
1382         err = decode_new_pools(p, end, map);
1383         if (err)
1384                 goto bad;
1385
1386         /* new_pool_names */
1387         err = decode_pool_names(p, end, map);
1388         if (err)
1389                 goto bad;
1390
1391         /* old_pool */
1392         ceph_decode_32_safe(p, end, len, e_inval);
1393         while (len--) {
1394                 struct ceph_pg_pool_info *pi;
1395
1396                 ceph_decode_64_safe(p, end, pool, e_inval);
1397                 pi = __lookup_pg_pool(&map->pg_pools, pool);
1398                 if (pi)
1399                         __remove_pg_pool(&map->pg_pools, pi);
1400         }
1401
1402         /* new_up_client, new_state, new_weight */
1403         err = decode_new_up_state_weight(p, end, map);
1404         if (err)
1405                 goto bad;
1406
1407         /* new_pg_temp */
1408         err = decode_new_pg_temp(p, end, map);
1409         if (err)
1410                 goto bad;
1411
1412         /* new_primary_temp */
1413         if (struct_v >= 1) {
1414                 err = decode_new_primary_temp(p, end, map);
1415                 if (err)
1416                         goto bad;
1417         }
1418
1419         /* new_primary_affinity */
1420         if (struct_v >= 2) {
1421                 err = decode_new_primary_affinity(p, end, map);
1422                 if (err)
1423                         goto bad;
1424         }
1425
1426         /* ignore the rest */
1427         *p = end;
1428
1429         dout("inc osdmap epoch %d max_osd %d\n", map->epoch, map->max_osd);
1430         return map;
1431
1432 e_inval:
1433         err = -EINVAL;
1434 bad:
1435         pr_err("corrupt inc osdmap (%d) epoch %d off %d (%p of %p-%p)\n",
1436                err, epoch, (int)(*p - start), *p, start, end);
1437         print_hex_dump(KERN_DEBUG, "osdmap: ",
1438                        DUMP_PREFIX_OFFSET, 16, 1,
1439                        start, end - start, true);
1440         if (newcrush)
1441                 crush_destroy(newcrush);
1442         return ERR_PTR(err);
1443 }
1444
1445
1446
1447
1448 /*
1449  * calculate file layout from given offset, length.
1450  * fill in correct oid, logical length, and object extent
1451  * offset, length.
1452  *
1453  * for now, we write only a single su, until we can
1454  * pass a stride back to the caller.
1455  */
1456 int ceph_calc_file_object_mapping(struct ceph_file_layout *layout,
1457                                    u64 off, u64 len,
1458                                    u64 *ono,
1459                                    u64 *oxoff, u64 *oxlen)
1460 {
1461         u32 osize = le32_to_cpu(layout->fl_object_size);
1462         u32 su = le32_to_cpu(layout->fl_stripe_unit);
1463         u32 sc = le32_to_cpu(layout->fl_stripe_count);
1464         u32 bl, stripeno, stripepos, objsetno;
1465         u32 su_per_object;
1466         u64 t, su_offset;
1467
1468         dout("mapping %llu~%llu  osize %u fl_su %u\n", off, len,
1469              osize, su);
1470         if (su == 0 || sc == 0)
1471                 goto invalid;
1472         su_per_object = osize / su;
1473         if (su_per_object == 0)
1474                 goto invalid;
1475         dout("osize %u / su %u = su_per_object %u\n", osize, su,
1476              su_per_object);
1477
1478         if ((su & ~PAGE_MASK) != 0)
1479                 goto invalid;
1480
1481         /* bl = *off / su; */
1482         t = off;
1483         do_div(t, su);
1484         bl = t;
1485         dout("off %llu / su %u = bl %u\n", off, su, bl);
1486
1487         stripeno = bl / sc;
1488         stripepos = bl % sc;
1489         objsetno = stripeno / su_per_object;
1490
1491         *ono = objsetno * sc + stripepos;
1492         dout("objset %u * sc %u = ono %u\n", objsetno, sc, (unsigned int)*ono);
1493
1494         /* *oxoff = *off % layout->fl_stripe_unit;  # offset in su */
1495         t = off;
1496         su_offset = do_div(t, su);
1497         *oxoff = su_offset + (stripeno % su_per_object) * su;
1498
1499         /*
1500          * Calculate the length of the extent being written to the selected
1501          * object. This is the minimum of the full length requested (len) or
1502          * the remainder of the current stripe being written to.
1503          */
1504         *oxlen = min_t(u64, len, su - su_offset);
1505
1506         dout(" obj extent %llu~%llu\n", *oxoff, *oxlen);
1507         return 0;
1508
1509 invalid:
1510         dout(" invalid layout\n");
1511         *ono = 0;
1512         *oxoff = 0;
1513         *oxlen = 0;
1514         return -EINVAL;
1515 }
1516 EXPORT_SYMBOL(ceph_calc_file_object_mapping);
1517
1518 /*
1519  * Calculate mapping of a (oloc, oid) pair to a PG.  Should only be
1520  * called with target's (oloc, oid), since tiering isn't taken into
1521  * account.
1522  */
1523 int ceph_oloc_oid_to_pg(struct ceph_osdmap *osdmap,
1524                         struct ceph_object_locator *oloc,
1525                         struct ceph_object_id *oid,
1526                         struct ceph_pg *pg_out)
1527 {
1528         struct ceph_pg_pool_info *pi;
1529
1530         pi = __lookup_pg_pool(&osdmap->pg_pools, oloc->pool);
1531         if (!pi)
1532                 return -EIO;
1533
1534         pg_out->pool = oloc->pool;
1535         pg_out->seed = ceph_str_hash(pi->object_hash, oid->name,
1536                                      oid->name_len);
1537
1538         dout("%s '%.*s' pgid %llu.%x\n", __func__, oid->name_len, oid->name,
1539              pg_out->pool, pg_out->seed);
1540         return 0;
1541 }
1542 EXPORT_SYMBOL(ceph_oloc_oid_to_pg);
1543
1544 static int do_crush(struct ceph_osdmap *map, int ruleno, int x,
1545                     int *result, int result_max,
1546                     const __u32 *weight, int weight_max)
1547 {
1548         int r;
1549
1550         BUG_ON(result_max > CEPH_PG_MAX_SIZE);
1551
1552         mutex_lock(&map->crush_scratch_mutex);
1553         r = crush_do_rule(map->crush, ruleno, x, result, result_max,
1554                           weight, weight_max, map->crush_scratch_ary);
1555         mutex_unlock(&map->crush_scratch_mutex);
1556
1557         return r;
1558 }
1559
1560 /*
1561  * Calculate raw (crush) set for given pgid.
1562  *
1563  * Return raw set length, or error.
1564  */
1565 static int pg_to_raw_osds(struct ceph_osdmap *osdmap,
1566                           struct ceph_pg_pool_info *pool,
1567                           struct ceph_pg pgid, u32 pps, int *osds)
1568 {
1569         int ruleno;
1570         int len;
1571
1572         /* crush */
1573         ruleno = crush_find_rule(osdmap->crush, pool->crush_ruleset,
1574                                  pool->type, pool->size);
1575         if (ruleno < 0) {
1576                 pr_err("no crush rule: pool %lld ruleset %d type %d size %d\n",
1577                        pgid.pool, pool->crush_ruleset, pool->type,
1578                        pool->size);
1579                 return -ENOENT;
1580         }
1581
1582         len = do_crush(osdmap, ruleno, pps, osds,
1583                        min_t(int, pool->size, CEPH_PG_MAX_SIZE),
1584                        osdmap->osd_weight, osdmap->max_osd);
1585         if (len < 0) {
1586                 pr_err("error %d from crush rule %d: pool %lld ruleset %d type %d size %d\n",
1587                        len, ruleno, pgid.pool, pool->crush_ruleset,
1588                        pool->type, pool->size);
1589                 return len;
1590         }
1591
1592         return len;
1593 }
1594
1595 /*
1596  * Given raw set, calculate up set and up primary.
1597  *
1598  * Return up set length.  *primary is set to up primary osd id, or -1
1599  * if up set is empty.
1600  */
1601 static int raw_to_up_osds(struct ceph_osdmap *osdmap,
1602                           struct ceph_pg_pool_info *pool,
1603                           int *osds, int len, int *primary)
1604 {
1605         int up_primary = -1;
1606         int i;
1607
1608         if (ceph_can_shift_osds(pool)) {
1609                 int removed = 0;
1610
1611                 for (i = 0; i < len; i++) {
1612                         if (ceph_osd_is_down(osdmap, osds[i])) {
1613                                 removed++;
1614                                 continue;
1615                         }
1616                         if (removed)
1617                                 osds[i - removed] = osds[i];
1618                 }
1619
1620                 len -= removed;
1621                 if (len > 0)
1622                         up_primary = osds[0];
1623         } else {
1624                 for (i = len - 1; i >= 0; i--) {
1625                         if (ceph_osd_is_down(osdmap, osds[i]))
1626                                 osds[i] = CRUSH_ITEM_NONE;
1627                         else
1628                                 up_primary = osds[i];
1629                 }
1630         }
1631
1632         *primary = up_primary;
1633         return len;
1634 }
1635
1636 static void apply_primary_affinity(struct ceph_osdmap *osdmap, u32 pps,
1637                                    struct ceph_pg_pool_info *pool,
1638                                    int *osds, int len, int *primary)
1639 {
1640         int i;
1641         int pos = -1;
1642
1643         /*
1644          * Do we have any non-default primary_affinity values for these
1645          * osds?
1646          */
1647         if (!osdmap->osd_primary_affinity)
1648                 return;
1649
1650         for (i = 0; i < len; i++) {
1651                 int osd = osds[i];
1652
1653                 if (osd != CRUSH_ITEM_NONE &&
1654                     osdmap->osd_primary_affinity[osd] !=
1655                                         CEPH_OSD_DEFAULT_PRIMARY_AFFINITY) {
1656                         break;
1657                 }
1658         }
1659         if (i == len)
1660                 return;
1661
1662         /*
1663          * Pick the primary.  Feed both the seed (for the pg) and the
1664          * osd into the hash/rng so that a proportional fraction of an
1665          * osd's pgs get rejected as primary.
1666          */
1667         for (i = 0; i < len; i++) {
1668                 int osd = osds[i];
1669                 u32 aff;
1670
1671                 if (osd == CRUSH_ITEM_NONE)
1672                         continue;
1673
1674                 aff = osdmap->osd_primary_affinity[osd];
1675                 if (aff < CEPH_OSD_MAX_PRIMARY_AFFINITY &&
1676                     (crush_hash32_2(CRUSH_HASH_RJENKINS1,
1677                                     pps, osd) >> 16) >= aff) {
1678                         /*
1679                          * We chose not to use this primary.  Note it
1680                          * anyway as a fallback in case we don't pick
1681                          * anyone else, but keep looking.
1682                          */
1683                         if (pos < 0)
1684                                 pos = i;
1685                 } else {
1686                         pos = i;
1687                         break;
1688                 }
1689         }
1690         if (pos < 0)
1691                 return;
1692
1693         *primary = osds[pos];
1694
1695         if (ceph_can_shift_osds(pool) && pos > 0) {
1696                 /* move the new primary to the front */
1697                 for (i = pos; i > 0; i--)
1698                         osds[i] = osds[i - 1];
1699                 osds[0] = *primary;
1700         }
1701 }
1702
1703 /*
1704  * Given up set, apply pg_temp and primary_temp mappings.
1705  *
1706  * Return acting set length.  *primary is set to acting primary osd id,
1707  * or -1 if acting set is empty.
1708  */
1709 static int apply_temps(struct ceph_osdmap *osdmap,
1710                        struct ceph_pg_pool_info *pool, struct ceph_pg pgid,
1711                        int *osds, int len, int *primary)
1712 {
1713         struct ceph_pg_mapping *pg;
1714         int temp_len;
1715         int temp_primary;
1716         int i;
1717
1718         /* raw_pg -> pg */
1719         pgid.seed = ceph_stable_mod(pgid.seed, pool->pg_num,
1720                                     pool->pg_num_mask);
1721
1722         /* pg_temp? */
1723         pg = __lookup_pg_mapping(&osdmap->pg_temp, pgid);
1724         if (pg) {
1725                 temp_len = 0;
1726                 temp_primary = -1;
1727
1728                 for (i = 0; i < pg->pg_temp.len; i++) {
1729                         if (ceph_osd_is_down(osdmap, pg->pg_temp.osds[i])) {
1730                                 if (ceph_can_shift_osds(pool))
1731                                         continue;
1732                                 else
1733                                         osds[temp_len++] = CRUSH_ITEM_NONE;
1734                         } else {
1735                                 osds[temp_len++] = pg->pg_temp.osds[i];
1736                         }
1737                 }
1738
1739                 /* apply pg_temp's primary */
1740                 for (i = 0; i < temp_len; i++) {
1741                         if (osds[i] != CRUSH_ITEM_NONE) {
1742                                 temp_primary = osds[i];
1743                                 break;
1744                         }
1745                 }
1746         } else {
1747                 temp_len = len;
1748                 temp_primary = *primary;
1749         }
1750
1751         /* primary_temp? */
1752         pg = __lookup_pg_mapping(&osdmap->primary_temp, pgid);
1753         if (pg)
1754                 temp_primary = pg->primary_temp.osd;
1755
1756         *primary = temp_primary;
1757         return temp_len;
1758 }
1759
1760 /*
1761  * Calculate acting set for given pgid.
1762  *
1763  * Return acting set length, or error.  *primary is set to acting
1764  * primary osd id, or -1 if acting set is empty or on error.
1765  */
1766 int ceph_calc_pg_acting(struct ceph_osdmap *osdmap, struct ceph_pg pgid,
1767                         int *osds, int *primary)
1768 {
1769         struct ceph_pg_pool_info *pool;
1770         u32 pps;
1771         int len;
1772
1773         pool = __lookup_pg_pool(&osdmap->pg_pools, pgid.pool);
1774         if (!pool) {
1775                 *primary = -1;
1776                 return -ENOENT;
1777         }
1778
1779         if (pool->flags & CEPH_POOL_FLAG_HASHPSPOOL) {
1780                 /* hash pool id and seed so that pool PGs do not overlap */
1781                 pps = crush_hash32_2(CRUSH_HASH_RJENKINS1,
1782                                      ceph_stable_mod(pgid.seed, pool->pgp_num,
1783                                                      pool->pgp_num_mask),
1784                                      pgid.pool);
1785         } else {
1786                 /*
1787                  * legacy behavior: add ps and pool together.  this is
1788                  * not a great approach because the PGs from each pool
1789                  * will overlap on top of each other: 0.5 == 1.4 ==
1790                  * 2.3 == ...
1791                  */
1792                 pps = ceph_stable_mod(pgid.seed, pool->pgp_num,
1793                                       pool->pgp_num_mask) +
1794                         (unsigned)pgid.pool;
1795         }
1796
1797         len = pg_to_raw_osds(osdmap, pool, pgid, pps, osds);
1798         if (len < 0) {
1799                 *primary = -1;
1800                 return len;
1801         }
1802
1803         len = raw_to_up_osds(osdmap, pool, osds, len, primary);
1804
1805         apply_primary_affinity(osdmap, pps, pool, osds, len, primary);
1806
1807         len = apply_temps(osdmap, pool, pgid, osds, len, primary);
1808
1809         return len;
1810 }
1811
1812 /*
1813  * Return primary osd for given pgid, or -1 if none.
1814  */
1815 int ceph_calc_pg_primary(struct ceph_osdmap *osdmap, struct ceph_pg pgid)
1816 {
1817         int osds[CEPH_PG_MAX_SIZE];
1818         int primary;
1819
1820         ceph_calc_pg_acting(osdmap, pgid, osds, &primary);
1821
1822         return primary;
1823 }
1824 EXPORT_SYMBOL(ceph_calc_pg_primary);