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_32_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 skip_name_map(void **p, void *end)
 126 {
 127         int len;
 128         ceph_decode_32_safe(p, end, len ,bad);
 129         while (len--) {
 130                 int strlen;
 131                 *p += sizeof(u32);
 132                 ceph_decode_32_safe(p, end, strlen, bad);
 133                 *p += strlen;
 134 }
 135         return 0;
 136 bad:
 137         return -EINVAL;
 138 }
 139
 140 static struct crush_map *crush_decode(void *pbyval, void *end)
 141 {
 142         struct crush_map *c;
 143         int err = -EINVAL;
 144         int i, j;
 145         void **p = &pbyval;
 146         void *start = pbyval;
 147         u32 magic;
 148         u32 num_name_maps;
 149
 150         dout("crush_decode %p to %p len %d\n", *p, end, (int)(end - *p));
 151
 152         c = kzalloc(sizeof(*c), GFP_NOFS);
 153         if (c == NULL)
 154                 return ERR_PTR(-ENOMEM);
 155
 156         /* set tunables to default values */
 157         c->choose_local_tries = 2;
 158         c->choose_local_fallback_tries = 5;
 159         c->choose_total_tries = 19;
 160         c->chooseleaf_descend_once = 0;
 161
 162         ceph_decode_need(p, end, 4*sizeof(u32), bad);
 163         magic = ceph_decode_32(p);
 164         if (magic != CRUSH_MAGIC) {
 165                 pr_err("crush_decode magic %x != current %x\n",
 166                        (unsigned int)magic, (unsigned int)CRUSH_MAGIC);
 167                 goto bad;
 168         }
 169         c->max_buckets = ceph_decode_32(p);
 170         c->max_rules = ceph_decode_32(p);
 171         c->max_devices = ceph_decode_32(p);
 172
 173         c->buckets = kcalloc(c->max_buckets, sizeof(*c->buckets), GFP_NOFS);
 174         if (c->buckets == NULL)
 175                 goto badmem;
 176         c->rules = kcalloc(c->max_rules, sizeof(*c->rules), GFP_NOFS);
 177         if (c->rules == NULL)
 178                 goto badmem;
 179
 180         /* buckets */
 181         for (i = 0; i < c->max_buckets; i++) {
 182                 int size = 0;
 183                 u32 alg;
 184                 struct crush_bucket *b;
 185
 186                 ceph_decode_32_safe(p, end, alg, bad);
 187                 if (alg == 0) {
 188                         c->buckets[i] = NULL;
 189                         continue;
 190                 }
 191                 dout("crush_decode bucket %d off %x %p to %p\n",
 192                      i, (int)(*p-start), *p, end);
 193
 194                 switch (alg) {
 195                 case CRUSH_BUCKET_UNIFORM:
 196                         size = sizeof(struct crush_bucket_uniform);
 197                         break;
 198                 case CRUSH_BUCKET_LIST:
 199                         size = sizeof(struct crush_bucket_list);
 200                         break;
 201                 case CRUSH_BUCKET_TREE:
 202                         size = sizeof(struct crush_bucket_tree);
 203                         break;
 204                 case CRUSH_BUCKET_STRAW:
 205                         size = sizeof(struct crush_bucket_straw);
 206                         break;
 207                 default:
 208                         err = -EINVAL;
 209                         goto bad;
 210                 }
 211                 BUG_ON(size == 0);
 212                 b = c->buckets[i] = kzalloc(size, GFP_NOFS);
 213                 if (b == NULL)
 214                         goto badmem;
 215
 216                 ceph_decode_need(p, end, 4*sizeof(u32), bad);
 217                 b->id = ceph_decode_32(p);
 218                 b->type = ceph_decode_16(p);
 219                 b->alg = ceph_decode_8(p);
 220                 b->hash = ceph_decode_8(p);
 221                 b->weight = ceph_decode_32(p);
 222                 b->size = ceph_decode_32(p);
 223
 224                 dout("crush_decode bucket size %d off %x %p to %p\n",
 225                      b->size, (int)(*p-start), *p, end);
 226
 227                 b->items = kcalloc(b->size, sizeof(__s32), GFP_NOFS);
 228                 if (b->items == NULL)
 229                         goto badmem;
 230                 b->perm = kcalloc(b->size, sizeof(u32), GFP_NOFS);
 231                 if (b->perm == NULL)
 232                         goto badmem;
 233                 b->perm_n = 0;
 234
 235                 ceph_decode_need(p, end, b->size*sizeof(u32), bad);
 236                 for (j = 0; j < b->size; j++)
 237                         b->items[j] = ceph_decode_32(p);
 238
 239                 switch (b->alg) {
 240                 case CRUSH_BUCKET_UNIFORM:
 241                         err = crush_decode_uniform_bucket(p, end,
 242                                   (struct crush_bucket_uniform *)b);
 243                         if (err < 0)
 244                                 goto bad;
 245                         break;
 246                 case CRUSH_BUCKET_LIST:
 247                         err = crush_decode_list_bucket(p, end,
 248                                (struct crush_bucket_list *)b);
 249                         if (err < 0)
 250                                 goto bad;
 251                         break;
 252                 case CRUSH_BUCKET_TREE:
 253                         err = crush_decode_tree_bucket(p, end,
 254                                 (struct crush_bucket_tree *)b);
 255                         if (err < 0)
 256                                 goto bad;
 257                         break;
 258                 case CRUSH_BUCKET_STRAW:
 259                         err = crush_decode_straw_bucket(p, end,
 260                                 (struct crush_bucket_straw *)b);
 261                         if (err < 0)
 262                                 goto bad;
 263                         break;
 264                 }
 265         }
 266
 267         /* rules */
 268         dout("rule vec is %p\n", c->rules);
 269         for (i = 0; i < c->max_rules; i++) {
 270                 u32 yes;
 271                 struct crush_rule *r;
 272
 273                 ceph_decode_32_safe(p, end, yes, bad);
 274                 if (!yes) {
 275                         dout("crush_decode NO rule %d off %x %p to %p\n",
 276                              i, (int)(*p-start), *p, end);
 277                         c->rules[i] = NULL;
 278                         continue;
 279                 }
 280
 281                 dout("crush_decode rule %d off %x %p to %p\n",
 282                      i, (int)(*p-start), *p, end);
 283
 284                 /* len */
 285                 ceph_decode_32_safe(p, end, yes, bad);
 286 #if BITS_PER_LONG == 32
 287                 err = -EINVAL;
 288                 if (yes > (ULONG_MAX - sizeof(*r))
 289                           / sizeof(struct crush_rule_step))
 290                         goto bad;
 291 #endif
 292                 r = c->rules[i] = kmalloc(sizeof(*r) +
 293                                           yes*sizeof(struct crush_rule_step),
 294                                           GFP_NOFS);
 295                 if (r == NULL)
 296                         goto badmem;
 297                 dout(" rule %d is at %p\n", i, r);
 298                 r->len = yes;
 299                 ceph_decode_copy_safe(p, end, &r->mask, 4, bad); /* 4 u8's */
 300                 ceph_decode_need(p, end, r->len*3*sizeof(u32), bad);
 301                 for (j = 0; j < r->len; j++) {
 302                         r->steps[j].op = ceph_decode_32(p);
 303                         r->steps[j].arg1 = ceph_decode_32(p);
 304                         r->steps[j].arg2 = ceph_decode_32(p);
 305                 }
 306         }
 307
 308         /* ignore trailing name maps. */
 309         for (num_name_maps = 0; num_name_maps < 3; num_name_maps++) {
 310                 err = skip_name_map(p, end);
 311                 if (err < 0)
 312                         goto done;
 313         }
 314
 315         /* tunables */
 316         ceph_decode_need(p, end, 3*sizeof(u32), done);
 317         c->choose_local_tries = ceph_decode_32(p);
 318         c->choose_local_fallback_tries =  ceph_decode_32(p);
 319         c->choose_total_tries = ceph_decode_32(p);
 320         dout("crush decode tunable choose_local_tries = %d",
 321              c->choose_local_tries);
 322         dout("crush decode tunable choose_local_fallback_tries = %d",
 323              c->choose_local_fallback_tries);
 324         dout("crush decode tunable choose_total_tries = %d",
 325              c->choose_total_tries);
 326
 327         ceph_decode_need(p, end, sizeof(u32), done);
 328         c->chooseleaf_descend_once = ceph_decode_32(p);
 329         dout("crush decode tunable chooseleaf_descend_once = %d",
 330              c->chooseleaf_descend_once);
 331
 332 done:
 333         dout("crush_decode success\n");
 334         return c;
 335
 336 badmem:
 337         err = -ENOMEM;
 338 bad:
 339         dout("crush_decode fail %d\n", err);
 340         crush_destroy(c);
 341         return ERR_PTR(err);
 342 }
 343
 344 /*
 345  * rbtree of pg_mapping for handling pg_temp (explicit mapping of pgid
 346  * to a set of osds)
 347  */
 348 static int pgid_cmp(struct ceph_pg l, struct ceph_pg r)
 349 {
 350         if (l.pool < r.pool)
 351                 return -1;
 352         if (l.pool > r.pool)
 353                 return 1;
 354         if (l.seed < r.seed)
 355                 return -1;
 356         if (l.seed > r.seed)
 357                 return 1;
 358         return 0;
 359 }
 360
 361 static int __insert_pg_mapping(struct ceph_pg_mapping *new,
 362                                struct rb_root *root)
 363 {
 364         struct rb_node **p = &root->rb_node;
 365         struct rb_node *parent = NULL;
 366         struct ceph_pg_mapping *pg = NULL;
 367         int c;
 368
 369         dout("__insert_pg_mapping %llx %p\n", *(u64 *)&new->pgid, new);
 370         while (*p) {
 371                 parent = *p;
 372                 pg = rb_entry(parent, struct ceph_pg_mapping, node);
 373                 c = pgid_cmp(new->pgid, pg->pgid);
 374                 if (c < 0)
 375                         p = &(*p)->rb_left;
 376                 else if (c > 0)
 377                         p = &(*p)->rb_right;
 378                 else
 379                         return -EEXIST;
 380         }
 381
 382         rb_link_node(&new->node, parent, p);
 383         rb_insert_color(&new->node, root);
 384         return 0;
 385 }
 386
 387 static struct ceph_pg_mapping *__lookup_pg_mapping(struct rb_root *root,
 388                                                    struct ceph_pg pgid)
 389 {
 390         struct rb_node *n = root->rb_node;
 391         struct ceph_pg_mapping *pg;
 392         int c;
 393
 394         while (n) {
 395                 pg = rb_entry(n, struct ceph_pg_mapping, node);
 396                 c = pgid_cmp(pgid, pg->pgid);
 397                 if (c < 0) {
 398                         n = n->rb_left;
 399                 } else if (c > 0) {
 400                         n = n->rb_right;
 401                 } else {
 402                         dout("__lookup_pg_mapping %lld.%x got %p\n",
 403                              pgid.pool, pgid.seed, pg);
 404                         return pg;
 405                 }
 406         }
 407         return NULL;
 408 }
 409
 410 static int __remove_pg_mapping(struct rb_root *root, struct ceph_pg pgid)
 411 {
 412         struct ceph_pg_mapping *pg = __lookup_pg_mapping(root, pgid);
 413
 414         if (pg) {
 415                 dout("__remove_pg_mapping %lld.%x %p\n", pgid.pool, pgid.seed,
 416                      pg);
 417                 rb_erase(&pg->node, root);
 418                 kfree(pg);
 419                 return 0;
 420         }
 421         dout("__remove_pg_mapping %lld.%x dne\n", pgid.pool, pgid.seed);
 422         return -ENOENT;
 423 }
 424
 425 /*
 426  * rbtree of pg pool info
 427  */
 428 static int __insert_pg_pool(struct rb_root *root, struct ceph_pg_pool_info *new)
 429 {
 430         struct rb_node **p = &root->rb_node;
 431         struct rb_node *parent = NULL;
 432         struct ceph_pg_pool_info *pi = NULL;
 433
 434         while (*p) {
 435                 parent = *p;
 436                 pi = rb_entry(parent, struct ceph_pg_pool_info, node);
 437                 if (new->id < pi->id)
 438                         p = &(*p)->rb_left;
 439                 else if (new->id > pi->id)
 440                         p = &(*p)->rb_right;
 441                 else
 442                         return -EEXIST;
 443         }
 444
 445         rb_link_node(&new->node, parent, p);
 446         rb_insert_color(&new->node, root);
 447         return 0;
 448 }
 449
 450 static struct ceph_pg_pool_info *__lookup_pg_pool(struct rb_root *root, u64 id)
 451 {
 452         struct ceph_pg_pool_info *pi;
 453         struct rb_node *n = root->rb_node;
 454
 455         while (n) {
 456                 pi = rb_entry(n, struct ceph_pg_pool_info, node);
 457                 if (id < pi->id)
 458                         n = n->rb_left;
 459                 else if (id > pi->id)
 460                         n = n->rb_right;
 461                 else
 462                         return pi;
 463         }
 464         return NULL;
 465 }
 466
 467 struct ceph_pg_pool_info *ceph_pg_pool_by_id(struct ceph_osdmap *map, u64 id)
 468 {
 469         return __lookup_pg_pool(&map->pg_pools, id);
 470 }
 471
 472 const char *ceph_pg_pool_name_by_id(struct ceph_osdmap *map, u64 id)
 473 {
 474         struct ceph_pg_pool_info *pi;
 475
 476         if (id == CEPH_NOPOOL)
 477                 return NULL;
 478
 479         if (WARN_ON_ONCE(id > (u64) INT_MAX))
 480                 return NULL;
 481
 482         pi = __lookup_pg_pool(&map->pg_pools, (int) id);
 483
 484         return pi ? pi->name : NULL;
 485 }
 486 EXPORT_SYMBOL(ceph_pg_pool_name_by_id);
 487
 488 int ceph_pg_poolid_by_name(struct ceph_osdmap *map, const char *name)
 489 {
 490         struct rb_node *rbp;
 491
 492         for (rbp = rb_first(&map->pg_pools); rbp; rbp = rb_next(rbp)) {
 493                 struct ceph_pg_pool_info *pi =
 494                         rb_entry(rbp, struct ceph_pg_pool_info, node);
 495                 if (pi->name && strcmp(pi->name, name) == 0)
 496                         return pi->id;
 497         }
 498         return -ENOENT;
 499 }
 500 EXPORT_SYMBOL(ceph_pg_poolid_by_name);
 501
 502 static void __remove_pg_pool(struct rb_root *root, struct ceph_pg_pool_info *pi)
 503 {
 504         rb_erase(&pi->node, root);
 505         kfree(pi->name);
 506         kfree(pi);
 507 }
 508
 509 static int __decode_pool(void **p, void *end, struct ceph_pg_pool_info *pi)
 510 {
 511         u8 ev, cv;
 512         unsigned len, num;
 513         void *pool_end;
 514
 515         ceph_decode_need(p, end, 2 + 4, bad);
 516         ev = ceph_decode_8(p);  /* encoding version */
 517         cv = ceph_decode_8(p); /* compat version */
 518         if (ev < 5) {
 519                 pr_warning("got v %d < 5 cv %d of ceph_pg_pool\n", ev, cv);
 520                 return -EINVAL;
 521         }
 522         if (cv > 9) {
 523                 pr_warning("got v %d cv %d > 9 of ceph_pg_pool\n", ev, cv);
 524                 return -EINVAL;
 525         }
 526         len = ceph_decode_32(p);
 527         ceph_decode_need(p, end, len, bad);
 528         pool_end = *p + len;
 529
 530         pi->type = ceph_decode_8(p);
 531         pi->size = ceph_decode_8(p);
 532         pi->crush_ruleset = ceph_decode_8(p);
 533         pi->object_hash = ceph_decode_8(p);
 534
 535         pi->pg_num = ceph_decode_32(p);
 536         pi->pgp_num = ceph_decode_32(p);
 537
 538         *p += 4 + 4;  /* skip lpg* */
 539         *p += 4;      /* skip last_change */
 540         *p += 8 + 4;  /* skip snap_seq, snap_epoch */
 541
 542         /* skip snaps */
 543         num = ceph_decode_32(p);
 544         while (num--) {
 545                 *p += 8;  /* snapid key */
 546                 *p += 1 + 1; /* versions */
 547                 len = ceph_decode_32(p);
 548                 *p += len;
 549         }
 550
 551         /* skip removed_snaps */
 552         num = ceph_decode_32(p);
 553         *p += num * (8 + 8);
 554
 555         *p += 8;  /* skip auid */
 556         pi->flags = ceph_decode_64(p);
 557         *p += 4;  /* skip crash_replay_interval */
 558
 559         if (ev >= 7)
 560                 *p += 1;  /* skip min_size */
 561
 562         if (ev >= 8)
 563                 *p += 8 + 8;  /* skip quota_max_* */
 564
 565         if (ev >= 9) {
 566                 /* skip tiers */
 567                 num = ceph_decode_32(p);
 568                 *p += num * 8;
 569
 570                 *p += 8;  /* skip tier_of */
 571                 *p += 1;  /* skip cache_mode */
 572
 573                 pi->read_tier = ceph_decode_64(p);
 574                 pi->write_tier = ceph_decode_64(p);
 575         } else {
 576                 pi->read_tier = -1;
 577                 pi->write_tier = -1;
 578         }
 579
 580         /* ignore the rest */
 581
 582         *p = pool_end;
 583         calc_pg_masks(pi);
 584         return 0;
 585
 586 bad:
 587         return -EINVAL;
 588 }
 589
 590 static int __decode_pool_names(void **p, void *end, struct ceph_osdmap *map)
 591 {
 592         struct ceph_pg_pool_info *pi;
 593         u32 num, len;
 594         u64 pool;
 595
 596         ceph_decode_32_safe(p, end, num, bad);
 597         dout(" %d pool names\n", num);
 598         while (num--) {
 599                 ceph_decode_64_safe(p, end, pool, bad);
 600                 ceph_decode_32_safe(p, end, len, bad);
 601                 dout("  pool %llu len %d\n", pool, len);
 602                 ceph_decode_need(p, end, len, bad);
 603                 pi = __lookup_pg_pool(&map->pg_pools, pool);
 604                 if (pi) {
 605                         char *name = kstrndup(*p, len, GFP_NOFS);
 606
 607                         if (!name)
 608                                 return -ENOMEM;
 609                         kfree(pi->name);
 610                         pi->name = name;
 611                         dout("  name is %s\n", pi->name);
 612                 }
 613                 *p += len;
 614         }
 615         return 0;
 616
 617 bad:
 618         return -EINVAL;
 619 }
 620
 621 /*
 622  * osd map
 623  */
 624 void ceph_osdmap_destroy(struct ceph_osdmap *map)
 625 {
 626         dout("osdmap_destroy %p\n", map);
 627         if (map->crush)
 628                 crush_destroy(map->crush);
 629         while (!RB_EMPTY_ROOT(&map->pg_temp)) {
 630                 struct ceph_pg_mapping *pg =
 631                         rb_entry(rb_first(&map->pg_temp),
 632                                  struct ceph_pg_mapping, node);
 633                 rb_erase(&pg->node, &map->pg_temp);
 634                 kfree(pg);
 635         }
 636         while (!RB_EMPTY_ROOT(&map->pg_pools)) {
 637                 struct ceph_pg_pool_info *pi =
 638                         rb_entry(rb_first(&map->pg_pools),
 639                                  struct ceph_pg_pool_info, node);
 640                 __remove_pg_pool(&map->pg_pools, pi);
 641         }
 642         kfree(map->osd_state);
 643         kfree(map->osd_weight);
 644         kfree(map->osd_addr);
 645         kfree(map);
 646 }
 647
 648 /*
 649  * adjust max osd value.  reallocate arrays.
 650  */
 651 static int osdmap_set_max_osd(struct ceph_osdmap *map, int max)
 652 {
 653         u8 *state;
 654         struct ceph_entity_addr *addr;
 655         u32 *weight;
 656
 657         state = kcalloc(max, sizeof(*state), GFP_NOFS);
 658         addr = kcalloc(max, sizeof(*addr), GFP_NOFS);
 659         weight = kcalloc(max, sizeof(*weight), GFP_NOFS);
 660         if (state == NULL || addr == NULL || weight == NULL) {
 661                 kfree(state);
 662                 kfree(addr);
 663                 kfree(weight);
 664                 return -ENOMEM;
 665         }
 666
 667         /* copy old? */
 668         if (map->osd_state) {
 669                 memcpy(state, map->osd_state, map->max_osd*sizeof(*state));
 670                 memcpy(addr, map->osd_addr, map->max_osd*sizeof(*addr));
 671                 memcpy(weight, map->osd_weight, map->max_osd*sizeof(*weight));
 672                 kfree(map->osd_state);
 673                 kfree(map->osd_addr);
 674                 kfree(map->osd_weight);
 675         }
 676
 677         map->osd_state = state;
 678         map->osd_weight = weight;
 679         map->osd_addr = addr;
 680         map->max_osd = max;
 681         return 0;
 682 }
 683
 684 /*
 685  * decode a full map.
 686  */
 687 static int osdmap_decode(void **p, void *end, struct ceph_osdmap *map)
 688 {
 689         u16 version;
 690         u32 epoch = 0;
 691         void *start = *p;
 692         u32 max;
 693         u32 len, i;
 694         int err;
 695         struct ceph_pg_pool_info *pi;
 696
 697         dout("%s %p to %p len %d\n", __func__, *p, end, (int)(end - *p));
 698
 699         ceph_decode_16_safe(p, end, version, e_inval);
 700         if (version > 6) {
 701                 pr_warning("got unknown v %d > 6 of osdmap\n", version);
 702                 goto e_inval;
 703         }
 704         if (version < 6) {
 705                 pr_warning("got old v %d < 6 of osdmap\n", version);
 706                 goto e_inval;
 707         }
 708
 709         ceph_decode_need(p, end, 2*sizeof(u64)+6*sizeof(u32), e_inval);
 710         ceph_decode_copy(p, &map->fsid, sizeof(map->fsid));
 711         epoch = map->epoch = ceph_decode_32(p);
 712         ceph_decode_copy(p, &map->created, sizeof(map->created));
 713         ceph_decode_copy(p, &map->modified, sizeof(map->modified));
 714
 715         ceph_decode_32_safe(p, end, max, e_inval);
 716         while (max--) {
 717                 ceph_decode_need(p, end, 8 + 2, e_inval);
 718                 pi = kzalloc(sizeof(*pi), GFP_NOFS);
 719                 if (!pi) {
 720                         err = -ENOMEM;
 721                         goto bad;
 722                 }
 723                 pi->id = ceph_decode_64(p);
 724                 err = __decode_pool(p, end, pi);
 725                 if (err < 0) {
 726                         kfree(pi);
 727                         goto bad;
 728                 }
 729                 __insert_pg_pool(&map->pg_pools, pi);
 730         }
 731
 732         err = __decode_pool_names(p, end, map);
 733         if (err)
 734                 goto bad;
 735
 736         ceph_decode_32_safe(p, end, map->pool_max, e_inval);
 737
 738         ceph_decode_32_safe(p, end, map->flags, e_inval);
 739
 740         /* max_osd */
 741         ceph_decode_32_safe(p, end, max, e_inval);
 742
 743         /* (re)alloc osd arrays */
 744         err = osdmap_set_max_osd(map, max);
 745         if (err)
 746                 goto bad;
 747
 748         /* osds */
 749         ceph_decode_need(p, end, 3*sizeof(u32) +
 750                          map->max_osd*(1 + sizeof(*map->osd_weight) +
 751                                        sizeof(*map->osd_addr)), e_inval);
 752
 753         *p += 4; /* skip length field (should match max) */
 754         ceph_decode_copy(p, map->osd_state, map->max_osd);
 755
 756         *p += 4; /* skip length field (should match max) */
 757         for (i = 0; i < map->max_osd; i++)
 758                 map->osd_weight[i] = ceph_decode_32(p);
 759
 760         *p += 4; /* skip length field (should match max) */
 761         ceph_decode_copy(p, map->osd_addr, map->max_osd*sizeof(*map->osd_addr));
 762         for (i = 0; i < map->max_osd; i++)
 763                 ceph_decode_addr(&map->osd_addr[i]);
 764
 765         /* pg_temp */
 766         ceph_decode_32_safe(p, end, len, e_inval);
 767         for (i = 0; i < len; i++) {
 768                 int n, j;
 769                 struct ceph_pg pgid;
 770                 struct ceph_pg_mapping *pg;
 771
 772                 err = ceph_decode_pgid(p, end, &pgid);
 773                 if (err)
 774                         goto bad;
 775                 ceph_decode_need(p, end, sizeof(u32), e_inval);
 776                 n = ceph_decode_32(p);
 777                 if (n > (UINT_MAX - sizeof(*pg)) / sizeof(u32))
 778                         goto e_inval;
 779                 ceph_decode_need(p, end, n * sizeof(u32), e_inval);
 780                 pg = kmalloc(sizeof(*pg) + n*sizeof(u32), GFP_NOFS);
 781                 if (!pg) {
 782                         err = -ENOMEM;
 783                         goto bad;
 784                 }
 785                 pg->pgid = pgid;
 786                 pg->len = n;
 787                 for (j = 0; j < n; j++)
 788                         pg->osds[j] = ceph_decode_32(p);
 789
 790                 err = __insert_pg_mapping(pg, &map->pg_temp);
 791                 if (err)
 792                         goto bad;
 793                 dout(" added pg_temp %lld.%x len %d\n", pgid.pool, pgid.seed,
 794                      len);
 795         }
 796
 797         /* crush */
 798         ceph_decode_32_safe(p, end, len, e_inval);
 799         dout("osdmap_decode crush len %d from off 0x%x\n", len,
 800              (int)(*p - start));
 801         ceph_decode_need(p, end, len, e_inval);
 802         map->crush = crush_decode(*p, end);
 803         *p += len;
 804         if (IS_ERR(map->crush)) {
 805                 err = PTR_ERR(map->crush);
 806                 map->crush = NULL;
 807                 goto bad;
 808         }
 809
 810         /* ignore the rest */
 811         *p = end;
 812
 813         dout("full osdmap epoch %d max_osd %d\n", map->epoch, map->max_osd);
 814         return 0;
 815
 816 e_inval:
 817         err = -EINVAL;
 818 bad:
 819         pr_err("corrupt full osdmap (%d) epoch %d off %d (%p of %p-%p)\n",
 820                err, epoch, (int)(*p - start), *p, start, end);
 821         print_hex_dump(KERN_DEBUG, "osdmap: ",
 822                        DUMP_PREFIX_OFFSET, 16, 1,
 823                        start, end - start, true);
 824         return err;
 825 }
 826
 827 /*
 828  * Allocate and decode a full map.
 829  */
 830 struct ceph_osdmap *ceph_osdmap_decode(void **p, void *end)
 831 {
 832         struct ceph_osdmap *map;
 833         int ret;
 834
 835         map = kzalloc(sizeof(*map), GFP_NOFS);
 836         if (!map)
 837                 return ERR_PTR(-ENOMEM);
 838
 839         map->pg_temp = RB_ROOT;
 840         mutex_init(&map->crush_scratch_mutex);
 841
 842         ret = osdmap_decode(p, end, map);
 843         if (ret) {
 844                 ceph_osdmap_destroy(map);
 845                 return ERR_PTR(ret);
 846         }
 847
 848         return map;
 849 }
 850
 851 /*
 852  * decode and apply an incremental map update.
 853  */
 854 struct ceph_osdmap *osdmap_apply_incremental(void **p, void *end,
 855                                              struct ceph_osdmap *map,
 856                                              struct ceph_messenger *msgr)
 857 {
 858         struct crush_map *newcrush = NULL;
 859         struct ceph_fsid fsid;
 860         u32 epoch = 0;
 861         struct ceph_timespec modified;
 862         s32 len;
 863         u64 pool;
 864         __s64 new_pool_max;
 865         __s32 new_flags, max;
 866         void *start = *p;
 867         int err = -EINVAL;
 868         u16 version;
 869
 870         dout("%s %p to %p len %d\n", __func__, *p, end, (int)(end - *p));
 871
 872         ceph_decode_16_safe(p, end, version, bad);
 873         if (version != 6) {
 874                 pr_warning("got unknown v %d != 6 of inc osdmap\n", version);
 875                 goto bad;
 876         }
 877
 878         ceph_decode_need(p, end, sizeof(fsid)+sizeof(modified)+2*sizeof(u32),
 879                          bad);
 880         ceph_decode_copy(p, &fsid, sizeof(fsid));
 881         epoch = ceph_decode_32(p);
 882         BUG_ON(epoch != map->epoch+1);
 883         ceph_decode_copy(p, &modified, sizeof(modified));
 884         new_pool_max = ceph_decode_64(p);
 885         new_flags = ceph_decode_32(p);
 886
 887         /* full map? */
 888         ceph_decode_32_safe(p, end, len, bad);
 889         if (len > 0) {
 890                 dout("apply_incremental full map len %d, %p to %p\n",
 891                      len, *p, end);
 892                 return ceph_osdmap_decode(p, min(*p+len, end));
 893         }
 894
 895         /* new crush? */
 896         ceph_decode_32_safe(p, end, len, bad);
 897         if (len > 0) {
 898                 dout("apply_incremental new crush map len %d, %p to %p\n",
 899                      len, *p, end);
 900                 newcrush = crush_decode(*p, min(*p+len, end));
 901                 if (IS_ERR(newcrush))
 902                         return ERR_CAST(newcrush);
 903                 *p += len;
 904         }
 905
 906         /* new flags? */
 907         if (new_flags >= 0)
 908                 map->flags = new_flags;
 909         if (new_pool_max >= 0)
 910                 map->pool_max = new_pool_max;
 911
 912         ceph_decode_need(p, end, 5*sizeof(u32), bad);
 913
 914         /* new max? */
 915         max = ceph_decode_32(p);
 916         if (max >= 0) {
 917                 err = osdmap_set_max_osd(map, max);
 918                 if (err < 0)
 919                         goto bad;
 920         }
 921
 922         map->epoch++;
 923         map->modified = modified;
 924         if (newcrush) {
 925                 if (map->crush)
 926                         crush_destroy(map->crush);
 927                 map->crush = newcrush;
 928                 newcrush = NULL;
 929         }
 930
 931         /* new_pool */
 932         ceph_decode_32_safe(p, end, len, bad);
 933         while (len--) {
 934                 struct ceph_pg_pool_info *pi;
 935
 936                 ceph_decode_64_safe(p, end, pool, bad);
 937                 pi = __lookup_pg_pool(&map->pg_pools, pool);
 938                 if (!pi) {
 939                         pi = kzalloc(sizeof(*pi), GFP_NOFS);
 940                         if (!pi) {
 941                                 err = -ENOMEM;
 942                                 goto bad;
 943                         }
 944                         pi->id = pool;
 945                         __insert_pg_pool(&map->pg_pools, pi);
 946                 }
 947                 err = __decode_pool(p, end, pi);
 948                 if (err < 0)
 949                         goto bad;
 950         }
 951         if (version >= 5) {
 952                 err = __decode_pool_names(p, end, map);
 953                 if (err < 0)
 954                         goto bad;
 955         }
 956
 957         /* old_pool */
 958         ceph_decode_32_safe(p, end, len, bad);
 959         while (len--) {
 960                 struct ceph_pg_pool_info *pi;
 961
 962                 ceph_decode_64_safe(p, end, pool, bad);
 963                 pi = __lookup_pg_pool(&map->pg_pools, pool);
 964                 if (pi)
 965                         __remove_pg_pool(&map->pg_pools, pi);
 966         }
 967
 968         /* new_up */
 969         err = -EINVAL;
 970         ceph_decode_32_safe(p, end, len, bad);
 971         while (len--) {
 972                 u32 osd;
 973                 struct ceph_entity_addr addr;
 974                 ceph_decode_32_safe(p, end, osd, bad);
 975                 ceph_decode_copy_safe(p, end, &addr, sizeof(addr), bad);
 976                 ceph_decode_addr(&addr);
 977                 pr_info("osd%d up\n", osd);
 978                 BUG_ON(osd >= map->max_osd);
 979                 map->osd_state[osd] |= CEPH_OSD_UP;
 980                 map->osd_addr[osd] = addr;
 981         }
 982
 983         /* new_state */
 984         ceph_decode_32_safe(p, end, len, bad);
 985         while (len--) {
 986                 u32 osd;
 987                 u8 xorstate;
 988                 ceph_decode_32_safe(p, end, osd, bad);
 989                 xorstate = **(u8 **)p;
 990                 (*p)++;  /* clean flag */
 991                 if (xorstate == 0)
 992                         xorstate = CEPH_OSD_UP;
 993                 if (xorstate & CEPH_OSD_UP)
 994                         pr_info("osd%d down\n", osd);
 995                 if (osd < map->max_osd)
 996                         map->osd_state[osd] ^= xorstate;
 997         }
 998
 999         /* new_weight */
1000         ceph_decode_32_safe(p, end, len, bad);
1001         while (len--) {
1002                 u32 osd, off;
1003                 ceph_decode_need(p, end, sizeof(u32)*2, bad);
1004                 osd = ceph_decode_32(p);
1005                 off = ceph_decode_32(p);
1006                 pr_info("osd%d weight 0x%x %s\n", osd, off,
1007                      off == CEPH_OSD_IN ? "(in)" :
1008                      (off == CEPH_OSD_OUT ? "(out)" : ""));
1009                 if (osd < map->max_osd)
1010                         map->osd_weight[osd] = off;
1011         }
1012
1013         /* new_pg_temp */
1014         ceph_decode_32_safe(p, end, len, bad);
1015         while (len--) {
1016                 struct ceph_pg_mapping *pg;
1017                 int j;
1018                 struct ceph_pg pgid;
1019                 u32 pglen;
1020
1021                 err = ceph_decode_pgid(p, end, &pgid);
1022                 if (err)
1023                         goto bad;
1024                 ceph_decode_need(p, end, sizeof(u32), bad);
1025                 pglen = ceph_decode_32(p);
1026                 if (pglen) {
1027                         ceph_decode_need(p, end, pglen*sizeof(u32), bad);
1028
1029                         /* removing existing (if any) */
1030                         (void) __remove_pg_mapping(&map->pg_temp, pgid);
1031
1032                         /* insert */
1033                         err = -EINVAL;
1034                         if (pglen > (UINT_MAX - sizeof(*pg)) / sizeof(u32))
1035                                 goto bad;
1036                         err = -ENOMEM;
1037                         pg = kmalloc(sizeof(*pg) + sizeof(u32)*pglen, GFP_NOFS);
1038                         if (!pg)
1039                                 goto bad;
1040                         pg->pgid = pgid;
1041                         pg->len = pglen;
1042                         for (j = 0; j < pglen; j++)
1043                                 pg->osds[j] = ceph_decode_32(p);
1044                         err = __insert_pg_mapping(pg, &map->pg_temp);
1045                         if (err) {
1046                                 kfree(pg);
1047                                 goto bad;
1048                         }
1049                         dout(" added pg_temp %lld.%x len %d\n", pgid.pool,
1050                              pgid.seed, pglen);
1051                 } else {
1052                         /* remove */
1053                         __remove_pg_mapping(&map->pg_temp, pgid);
1054                 }
1055         }
1056
1057         /* ignore the rest */
1058         *p = end;
1059
1060         dout("inc osdmap epoch %d max_osd %d\n", map->epoch, map->max_osd);
1061         return map;
1062
1063 bad:
1064         pr_err("corrupt inc osdmap (%d) epoch %d off %d (%p of %p-%p)\n",
1065                err, epoch, (int)(*p - start), *p, start, end);
1066         print_hex_dump(KERN_DEBUG, "osdmap: ",
1067                        DUMP_PREFIX_OFFSET, 16, 1,
1068                        start, end - start, true);
1069         if (newcrush)
1070                 crush_destroy(newcrush);
1071         return ERR_PTR(err);
1072 }
1073
1074
1075
1076
1077 /*
1078  * calculate file layout from given offset, length.
1079  * fill in correct oid, logical length, and object extent
1080  * offset, length.
1081  *
1082  * for now, we write only a single su, until we can
1083  * pass a stride back to the caller.
1084  */
1085 int ceph_calc_file_object_mapping(struct ceph_file_layout *layout,
1086                                    u64 off, u64 len,
1087                                    u64 *ono,
1088                                    u64 *oxoff, u64 *oxlen)
1089 {
1090         u32 osize = le32_to_cpu(layout->fl_object_size);
1091         u32 su = le32_to_cpu(layout->fl_stripe_unit);
1092         u32 sc = le32_to_cpu(layout->fl_stripe_count);
1093         u32 bl, stripeno, stripepos, objsetno;
1094         u32 su_per_object;
1095         u64 t, su_offset;
1096
1097         dout("mapping %llu~%llu  osize %u fl_su %u\n", off, len,
1098              osize, su);
1099         if (su == 0 || sc == 0)
1100                 goto invalid;
1101         su_per_object = osize / su;
1102         if (su_per_object == 0)
1103                 goto invalid;
1104         dout("osize %u / su %u = su_per_object %u\n", osize, su,
1105              su_per_object);
1106
1107         if ((su & ~PAGE_MASK) != 0)
1108                 goto invalid;
1109
1110         /* bl = *off / su; */
1111         t = off;
1112         do_div(t, su);
1113         bl = t;
1114         dout("off %llu / su %u = bl %u\n", off, su, bl);
1115
1116         stripeno = bl / sc;
1117         stripepos = bl % sc;
1118         objsetno = stripeno / su_per_object;
1119
1120         *ono = objsetno * sc + stripepos;
1121         dout("objset %u * sc %u = ono %u\n", objsetno, sc, (unsigned int)*ono);
1122
1123         /* *oxoff = *off % layout->fl_stripe_unit;  # offset in su */
1124         t = off;
1125         su_offset = do_div(t, su);
1126         *oxoff = su_offset + (stripeno % su_per_object) * su;
1127
1128         /*
1129          * Calculate the length of the extent being written to the selected
1130          * object. This is the minimum of the full length requested (len) or
1131          * the remainder of the current stripe being written to.
1132          */
1133         *oxlen = min_t(u64, len, su - su_offset);
1134
1135         dout(" obj extent %llu~%llu\n", *oxoff, *oxlen);
1136         return 0;
1137
1138 invalid:
1139         dout(" invalid layout\n");
1140         *ono = 0;
1141         *oxoff = 0;
1142         *oxlen = 0;
1143         return -EINVAL;
1144 }
1145 EXPORT_SYMBOL(ceph_calc_file_object_mapping);
1146
1147 /*
1148  * Calculate mapping of a (oloc, oid) pair to a PG.  Should only be
1149  * called with target's (oloc, oid), since tiering isn't taken into
1150  * account.
1151  */
1152 int ceph_oloc_oid_to_pg(struct ceph_osdmap *osdmap,
1153                         struct ceph_object_locator *oloc,
1154                         struct ceph_object_id *oid,
1155                         struct ceph_pg *pg_out)
1156 {
1157         struct ceph_pg_pool_info *pi;
1158
1159         pi = __lookup_pg_pool(&osdmap->pg_pools, oloc->pool);
1160         if (!pi)
1161                 return -EIO;
1162
1163         pg_out->pool = oloc->pool;
1164         pg_out->seed = ceph_str_hash(pi->object_hash, oid->name,
1165                                      oid->name_len);
1166
1167         dout("%s '%.*s' pgid %llu.%x\n", __func__, oid->name_len, oid->name,
1168              pg_out->pool, pg_out->seed);
1169         return 0;
1170 }
1171 EXPORT_SYMBOL(ceph_oloc_oid_to_pg);
1172
1173 static int do_crush(struct ceph_osdmap *map, int ruleno, int x,
1174                     int *result, int result_max,
1175                     const __u32 *weight, int weight_max)
1176 {
1177         int r;
1178
1179         BUG_ON(result_max > CEPH_PG_MAX_SIZE);
1180
1181         mutex_lock(&map->crush_scratch_mutex);
1182         r = crush_do_rule(map->crush, ruleno, x, result, result_max,
1183                           weight, weight_max, map->crush_scratch_ary);
1184         mutex_unlock(&map->crush_scratch_mutex);
1185
1186         return r;
1187 }
1188
1189 /*
1190  * Calculate raw osd vector for the given pgid.  Return pointer to osd
1191  * array, or NULL on failure.
1192  */
1193 static int *calc_pg_raw(struct ceph_osdmap *osdmap, struct ceph_pg pgid,
1194                         int *osds, int *num)
1195 {
1196         struct ceph_pg_mapping *pg;
1197         struct ceph_pg_pool_info *pool;
1198         int ruleno;
1199         int r;
1200         u32 pps;
1201
1202         pool = __lookup_pg_pool(&osdmap->pg_pools, pgid.pool);
1203         if (!pool)
1204                 return NULL;
1205
1206         /* pg_temp? */
1207         pgid.seed = ceph_stable_mod(pgid.seed, pool->pg_num,
1208                                     pool->pg_num_mask);
1209         pg = __lookup_pg_mapping(&osdmap->pg_temp, pgid);
1210         if (pg) {
1211                 *num = pg->len;
1212                 return pg->osds;
1213         }
1214
1215         /* crush */
1216         ruleno = crush_find_rule(osdmap->crush, pool->crush_ruleset,
1217                                  pool->type, pool->size);
1218         if (ruleno < 0) {
1219                 pr_err("no crush rule pool %lld ruleset %d type %d size %d\n",
1220                        pgid.pool, pool->crush_ruleset, pool->type,
1221                        pool->size);
1222                 return NULL;
1223         }
1224
1225         if (pool->flags & CEPH_POOL_FLAG_HASHPSPOOL) {
1226                 /* hash pool id and seed sothat pool PGs do not overlap */
1227                 pps = crush_hash32_2(CRUSH_HASH_RJENKINS1,
1228                                      ceph_stable_mod(pgid.seed, pool->pgp_num,
1229                                                      pool->pgp_num_mask),
1230                                      pgid.pool);
1231         } else {
1232                 /*
1233                  * legacy ehavior: add ps and pool together.  this is
1234                  * not a great approach because the PGs from each pool
1235                  * will overlap on top of each other: 0.5 == 1.4 ==
1236                  * 2.3 == ...
1237                  */
1238                 pps = ceph_stable_mod(pgid.seed, pool->pgp_num,
1239                                       pool->pgp_num_mask) +
1240                         (unsigned)pgid.pool;
1241         }
1242         r = do_crush(osdmap, ruleno, pps, osds, min_t(int, pool->size, *num),
1243                      osdmap->osd_weight, osdmap->max_osd);
1244         if (r < 0) {
1245                 pr_err("error %d from crush rule: pool %lld ruleset %d type %d"
1246                        " size %d\n", r, pgid.pool, pool->crush_ruleset,
1247                        pool->type, pool->size);
1248                 return NULL;
1249         }
1250         *num = r;
1251         return osds;
1252 }
1253
1254 /*
1255  * Return acting set for given pgid.
1256  */
1257 int ceph_calc_pg_acting(struct ceph_osdmap *osdmap, struct ceph_pg pgid,
1258                         int *acting)
1259 {
1260         int rawosds[CEPH_PG_MAX_SIZE], *osds;
1261         int i, o, num = CEPH_PG_MAX_SIZE;
1262
1263         osds = calc_pg_raw(osdmap, pgid, rawosds, &num);
1264         if (!osds)
1265                 return -1;
1266
1267         /* primary is first up osd */
1268         o = 0;
1269         for (i = 0; i < num; i++)
1270                 if (ceph_osd_is_up(osdmap, osds[i]))
1271                         acting[o++] = osds[i];
1272         return o;
1273 }
1274
1275 /*
1276  * Return primary osd for given pgid, or -1 if none.
1277  */
1278 int ceph_calc_pg_primary(struct ceph_osdmap *osdmap, struct ceph_pg pgid)
1279 {
1280         int rawosds[CEPH_PG_MAX_SIZE], *osds;
1281         int i, num = CEPH_PG_MAX_SIZE;
1282
1283         osds = calc_pg_raw(osdmap, pgid, rawosds, &num);
1284         if (!osds)
1285                 return -1;
1286
1287         /* primary is first up osd */
1288         for (i = 0; i < num; i++)
1289                 if (ceph_osd_is_up(osdmap, osds[i]))
1290                         return osds[i];
1291         return -1;
1292 }
1293 EXPORT_SYMBOL(ceph_calc_pg_primary);