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, (re)allocate arrays.
 650  *
 651  * The new elements are properly initialized.
 652  */
 653 static int osdmap_set_max_osd(struct ceph_osdmap *map, int max)
 654 {
 655         u8 *state;
 656         u32 *weight;
 657         struct ceph_entity_addr *addr;
 658         int i;
 659
 660         state = krealloc(map->osd_state, max*sizeof(*state), GFP_NOFS);
 661         weight = krealloc(map->osd_weight, max*sizeof(*weight), GFP_NOFS);
 662         addr = krealloc(map->osd_addr, max*sizeof(*addr), GFP_NOFS);
 663         if (!state || !weight || !addr) {
 664                 kfree(state);
 665                 kfree(weight);
 666                 kfree(addr);
 667
 668                 return -ENOMEM;
 669         }
 670
 671         for (i = map->max_osd; i < max; i++) {
 672                 state[i] = 0;
 673                 weight[i] = CEPH_OSD_OUT;
 674                 memset(addr + i, 0, sizeof(*addr));
 675         }
 676
 677         map->osd_state = state;
 678         map->osd_weight = weight;
 679         map->osd_addr = addr;
 680
 681         map->max_osd = max;
 682
 683         return 0;
 684 }
 685
 686 static int __decode_pools(void **p, void *end, struct ceph_osdmap *map,
 687                           bool incremental)
 688 {
 689         u32 n;
 690
 691         ceph_decode_32_safe(p, end, n, e_inval);
 692         while (n--) {
 693                 struct ceph_pg_pool_info *pi;
 694                 u64 pool;
 695                 int ret;
 696
 697                 ceph_decode_64_safe(p, end, pool, e_inval);
 698
 699                 pi = __lookup_pg_pool(&map->pg_pools, pool);
 700                 if (!incremental || !pi) {
 701                         pi = kzalloc(sizeof(*pi), GFP_NOFS);
 702                         if (!pi)
 703                                 return -ENOMEM;
 704
 705                         pi->id = pool;
 706
 707                         ret = __insert_pg_pool(&map->pg_pools, pi);
 708                         if (ret) {
 709                                 kfree(pi);
 710                                 return ret;
 711                         }
 712                 }
 713
 714                 ret = decode_pool(p, end, pi);
 715                 if (ret)
 716                         return ret;
 717         }
 718
 719         return 0;
 720
 721 e_inval:
 722         return -EINVAL;
 723 }
 724
 725 static int decode_pools(void **p, void *end, struct ceph_osdmap *map)
 726 {
 727         return __decode_pools(p, end, map, false);
 728 }
 729
 730 static int decode_new_pools(void **p, void *end, struct ceph_osdmap *map)
 731 {
 732         return __decode_pools(p, end, map, true);
 733 }
 734
 735 static int __decode_pg_temp(void **p, void *end, struct ceph_osdmap *map,
 736                             bool incremental)
 737 {
 738         u32 n;
 739
 740         ceph_decode_32_safe(p, end, n, e_inval);
 741         while (n--) {
 742                 struct ceph_pg pgid;
 743                 u32 len, i;
 744                 int ret;
 745
 746                 ret = ceph_decode_pgid(p, end, &pgid);
 747                 if (ret)
 748                         return ret;
 749
 750                 ceph_decode_32_safe(p, end, len, e_inval);
 751
 752                 ret = __remove_pg_mapping(&map->pg_temp, pgid);
 753                 BUG_ON(!incremental && ret != -ENOENT);
 754
 755                 if (!incremental || len > 0) {
 756                         struct ceph_pg_mapping *pg;
 757
 758                         ceph_decode_need(p, end, len*sizeof(u32), e_inval);
 759
 760                         if (len > (UINT_MAX - sizeof(*pg)) / sizeof(u32))
 761                                 return -EINVAL;
 762
 763                         pg = kzalloc(sizeof(*pg) + len*sizeof(u32), GFP_NOFS);
 764                         if (!pg)
 765                                 return -ENOMEM;
 766
 767                         pg->pgid = pgid;
 768                         pg->len = len;
 769                         for (i = 0; i < len; i++)
 770                                 pg->osds[i] = ceph_decode_32(p);
 771
 772                         ret = __insert_pg_mapping(pg, &map->pg_temp);
 773                         if (ret) {
 774                                 kfree(pg);
 775                                 return ret;
 776                         }
 777                 }
 778         }
 779
 780         return 0;
 781
 782 e_inval:
 783         return -EINVAL;
 784 }
 785
 786 static int decode_pg_temp(void **p, void *end, struct ceph_osdmap *map)
 787 {
 788         return __decode_pg_temp(p, end, map, false);
 789 }
 790
 791 static int decode_new_pg_temp(void **p, void *end, struct ceph_osdmap *map)
 792 {
 793         return __decode_pg_temp(p, end, map, true);
 794 }
 795
 796 /*
 797  * decode a full map.
 798  */
 799 static int osdmap_decode(void **p, void *end, struct ceph_osdmap *map)
 800 {
 801         u16 version;
 802         u32 epoch = 0;
 803         void *start = *p;
 804         u32 max;
 805         u32 len, i;
 806         int err;
 807
 808         dout("%s %p to %p len %d\n", __func__, *p, end, (int)(end - *p));
 809
 810         ceph_decode_16_safe(p, end, version, e_inval);
 811         if (version > 6) {
 812                 pr_warning("got unknown v %d > 6 of osdmap\n", version);
 813                 goto e_inval;
 814         }
 815         if (version < 6) {
 816                 pr_warning("got old v %d < 6 of osdmap\n", version);
 817                 goto e_inval;
 818         }
 819
 820         /* fsid, epoch, created, modified */
 821         ceph_decode_need(p, end, sizeof(map->fsid) + sizeof(u32) +
 822                          sizeof(map->created) + sizeof(map->modified), e_inval);
 823         ceph_decode_copy(p, &map->fsid, sizeof(map->fsid));
 824         epoch = map->epoch = ceph_decode_32(p);
 825         ceph_decode_copy(p, &map->created, sizeof(map->created));
 826         ceph_decode_copy(p, &map->modified, sizeof(map->modified));
 827
 828         /* pools */
 829         err = decode_pools(p, end, map);
 830         if (err)
 831                 goto bad;
 832
 833         /* pool_name */
 834         err = decode_pool_names(p, end, map);
 835         if (err)
 836                 goto bad;
 837
 838         ceph_decode_32_safe(p, end, map->pool_max, e_inval);
 839
 840         ceph_decode_32_safe(p, end, map->flags, e_inval);
 841
 842         /* max_osd */
 843         ceph_decode_32_safe(p, end, max, e_inval);
 844
 845         /* (re)alloc osd arrays */
 846         err = osdmap_set_max_osd(map, max);
 847         if (err)
 848                 goto bad;
 849
 850         /* osd_state, osd_weight, osd_addrs->client_addr */
 851         ceph_decode_need(p, end, 3*sizeof(u32) +
 852                          map->max_osd*(1 + sizeof(*map->osd_weight) +
 853                                        sizeof(*map->osd_addr)), e_inval);
 854
 855         if (ceph_decode_32(p) != map->max_osd)
 856                 goto e_inval;
 857
 858         ceph_decode_copy(p, map->osd_state, map->max_osd);
 859
 860         if (ceph_decode_32(p) != map->max_osd)
 861                 goto e_inval;
 862
 863         for (i = 0; i < map->max_osd; i++)
 864                 map->osd_weight[i] = ceph_decode_32(p);
 865
 866         if (ceph_decode_32(p) != map->max_osd)
 867                 goto e_inval;
 868
 869         ceph_decode_copy(p, map->osd_addr, map->max_osd*sizeof(*map->osd_addr));
 870         for (i = 0; i < map->max_osd; i++)
 871                 ceph_decode_addr(&map->osd_addr[i]);
 872
 873         /* pg_temp */
 874         err = decode_pg_temp(p, end, map);
 875         if (err)
 876                 goto bad;
 877
 878         /* crush */
 879         ceph_decode_32_safe(p, end, len, e_inval);
 880         map->crush = crush_decode(*p, min(*p + len, end));
 881         if (IS_ERR(map->crush)) {
 882                 err = PTR_ERR(map->crush);
 883                 map->crush = NULL;
 884                 goto bad;
 885         }
 886         *p += len;
 887
 888         /* ignore the rest */
 889         *p = end;
 890
 891         dout("full osdmap epoch %d max_osd %d\n", map->epoch, map->max_osd);
 892         return 0;
 893
 894 e_inval:
 895         err = -EINVAL;
 896 bad:
 897         pr_err("corrupt full osdmap (%d) epoch %d off %d (%p of %p-%p)\n",
 898                err, epoch, (int)(*p - start), *p, start, end);
 899         print_hex_dump(KERN_DEBUG, "osdmap: ",
 900                        DUMP_PREFIX_OFFSET, 16, 1,
 901                        start, end - start, true);
 902         return err;
 903 }
 904
 905 /*
 906  * Allocate and decode a full map.
 907  */
 908 struct ceph_osdmap *ceph_osdmap_decode(void **p, void *end)
 909 {
 910         struct ceph_osdmap *map;
 911         int ret;
 912
 913         map = kzalloc(sizeof(*map), GFP_NOFS);
 914         if (!map)
 915                 return ERR_PTR(-ENOMEM);
 916
 917         map->pg_temp = RB_ROOT;
 918         mutex_init(&map->crush_scratch_mutex);
 919
 920         ret = osdmap_decode(p, end, map);
 921         if (ret) {
 922                 ceph_osdmap_destroy(map);
 923                 return ERR_PTR(ret);
 924         }
 925
 926         return map;
 927 }
 928
 929 /*
 930  * decode and apply an incremental map update.
 931  */
 932 struct ceph_osdmap *osdmap_apply_incremental(void **p, void *end,
 933                                              struct ceph_osdmap *map,
 934                                              struct ceph_messenger *msgr)
 935 {
 936         struct crush_map *newcrush = NULL;
 937         struct ceph_fsid fsid;
 938         u32 epoch = 0;
 939         struct ceph_timespec modified;
 940         s32 len;
 941         u64 pool;
 942         __s64 new_pool_max;
 943         __s32 new_flags, max;
 944         void *start = *p;
 945         int err;
 946         u16 version;
 947
 948         dout("%s %p to %p len %d\n", __func__, *p, end, (int)(end - *p));
 949
 950         ceph_decode_16_safe(p, end, version, e_inval);
 951         if (version != 6) {
 952                 pr_warning("got unknown v %d != 6 of inc osdmap\n", version);
 953                 goto e_inval;
 954         }
 955
 956         /* fsid, epoch, modified, new_pool_max, new_flags */
 957         ceph_decode_need(p, end, sizeof(fsid) + sizeof(u32) + sizeof(modified) +
 958                          sizeof(u64) + sizeof(u32), e_inval);
 959         ceph_decode_copy(p, &fsid, sizeof(fsid));
 960         epoch = ceph_decode_32(p);
 961         BUG_ON(epoch != map->epoch+1);
 962         ceph_decode_copy(p, &modified, sizeof(modified));
 963         new_pool_max = ceph_decode_64(p);
 964         new_flags = ceph_decode_32(p);
 965
 966         /* full map? */
 967         ceph_decode_32_safe(p, end, len, e_inval);
 968         if (len > 0) {
 969                 dout("apply_incremental full map len %d, %p to %p\n",
 970                      len, *p, end);
 971                 return ceph_osdmap_decode(p, min(*p+len, end));
 972         }
 973
 974         /* new crush? */
 975         ceph_decode_32_safe(p, end, len, e_inval);
 976         if (len > 0) {
 977                 newcrush = crush_decode(*p, min(*p+len, end));
 978                 if (IS_ERR(newcrush)) {
 979                         err = PTR_ERR(newcrush);
 980                         newcrush = NULL;
 981                         goto bad;
 982                 }
 983                 *p += len;
 984         }
 985
 986         /* new flags? */
 987         if (new_flags >= 0)
 988                 map->flags = new_flags;
 989         if (new_pool_max >= 0)
 990                 map->pool_max = new_pool_max;
 991
 992         /* new max? */
 993         ceph_decode_32_safe(p, end, max, e_inval);
 994         if (max >= 0) {
 995                 err = osdmap_set_max_osd(map, max);
 996                 if (err)
 997                         goto bad;
 998         }
 999
1000         map->epoch++;
1001         map->modified = modified;
1002         if (newcrush) {
1003                 if (map->crush)
1004                         crush_destroy(map->crush);
1005                 map->crush = newcrush;
1006                 newcrush = NULL;
1007         }
1008
1009         /* new_pools */
1010         err = decode_new_pools(p, end, map);
1011         if (err)
1012                 goto bad;
1013
1014         /* new_pool_names */
1015         err = decode_pool_names(p, end, map);
1016         if (err)
1017                 goto bad;
1018
1019         /* old_pool */
1020         ceph_decode_32_safe(p, end, len, e_inval);
1021         while (len--) {
1022                 struct ceph_pg_pool_info *pi;
1023
1024                 ceph_decode_64_safe(p, end, pool, e_inval);
1025                 pi = __lookup_pg_pool(&map->pg_pools, pool);
1026                 if (pi)
1027                         __remove_pg_pool(&map->pg_pools, pi);
1028         }
1029
1030         /* new_up */
1031         ceph_decode_32_safe(p, end, len, e_inval);
1032         while (len--) {
1033                 u32 osd;
1034                 struct ceph_entity_addr addr;
1035                 ceph_decode_32_safe(p, end, osd, e_inval);
1036                 ceph_decode_copy_safe(p, end, &addr, sizeof(addr), e_inval);
1037                 ceph_decode_addr(&addr);
1038                 pr_info("osd%d up\n", osd);
1039                 BUG_ON(osd >= map->max_osd);
1040                 map->osd_state[osd] |= CEPH_OSD_UP;
1041                 map->osd_addr[osd] = addr;
1042         }
1043
1044         /* new_state */
1045         ceph_decode_32_safe(p, end, len, e_inval);
1046         while (len--) {
1047                 u32 osd;
1048                 u8 xorstate;
1049                 ceph_decode_32_safe(p, end, osd, e_inval);
1050                 xorstate = **(u8 **)p;
1051                 (*p)++;  /* clean flag */
1052                 if (xorstate == 0)
1053                         xorstate = CEPH_OSD_UP;
1054                 if (xorstate & CEPH_OSD_UP)
1055                         pr_info("osd%d down\n", osd);
1056                 if (osd < map->max_osd)
1057                         map->osd_state[osd] ^= xorstate;
1058         }
1059
1060         /* new_weight */
1061         ceph_decode_32_safe(p, end, len, e_inval);
1062         while (len--) {
1063                 u32 osd, off;
1064                 ceph_decode_need(p, end, sizeof(u32)*2, e_inval);
1065                 osd = ceph_decode_32(p);
1066                 off = ceph_decode_32(p);
1067                 pr_info("osd%d weight 0x%x %s\n", osd, off,
1068                      off == CEPH_OSD_IN ? "(in)" :
1069                      (off == CEPH_OSD_OUT ? "(out)" : ""));
1070                 if (osd < map->max_osd)
1071                         map->osd_weight[osd] = off;
1072         }
1073
1074         /* new_pg_temp */
1075         err = decode_new_pg_temp(p, end, map);
1076         if (err)
1077                 goto bad;
1078
1079         /* ignore the rest */
1080         *p = end;
1081
1082         dout("inc osdmap epoch %d max_osd %d\n", map->epoch, map->max_osd);
1083         return map;
1084
1085 e_inval:
1086         err = -EINVAL;
1087 bad:
1088         pr_err("corrupt inc osdmap (%d) epoch %d off %d (%p of %p-%p)\n",
1089                err, epoch, (int)(*p - start), *p, start, end);
1090         print_hex_dump(KERN_DEBUG, "osdmap: ",
1091                        DUMP_PREFIX_OFFSET, 16, 1,
1092                        start, end - start, true);
1093         if (newcrush)
1094                 crush_destroy(newcrush);
1095         return ERR_PTR(err);
1096 }
1097
1098
1099
1100
1101 /*
1102  * calculate file layout from given offset, length.
1103  * fill in correct oid, logical length, and object extent
1104  * offset, length.
1105  *
1106  * for now, we write only a single su, until we can
1107  * pass a stride back to the caller.
1108  */
1109 int ceph_calc_file_object_mapping(struct ceph_file_layout *layout,
1110                                    u64 off, u64 len,
1111                                    u64 *ono,
1112                                    u64 *oxoff, u64 *oxlen)
1113 {
1114         u32 osize = le32_to_cpu(layout->fl_object_size);
1115         u32 su = le32_to_cpu(layout->fl_stripe_unit);
1116         u32 sc = le32_to_cpu(layout->fl_stripe_count);
1117         u32 bl, stripeno, stripepos, objsetno;
1118         u32 su_per_object;
1119         u64 t, su_offset;
1120
1121         dout("mapping %llu~%llu  osize %u fl_su %u\n", off, len,
1122              osize, su);
1123         if (su == 0 || sc == 0)
1124                 goto invalid;
1125         su_per_object = osize / su;
1126         if (su_per_object == 0)
1127                 goto invalid;
1128         dout("osize %u / su %u = su_per_object %u\n", osize, su,
1129              su_per_object);
1130
1131         if ((su & ~PAGE_MASK) != 0)
1132                 goto invalid;
1133
1134         /* bl = *off / su; */
1135         t = off;
1136         do_div(t, su);
1137         bl = t;
1138         dout("off %llu / su %u = bl %u\n", off, su, bl);
1139
1140         stripeno = bl / sc;
1141         stripepos = bl % sc;
1142         objsetno = stripeno / su_per_object;
1143
1144         *ono = objsetno * sc + stripepos;
1145         dout("objset %u * sc %u = ono %u\n", objsetno, sc, (unsigned int)*ono);
1146
1147         /* *oxoff = *off % layout->fl_stripe_unit;  # offset in su */
1148         t = off;
1149         su_offset = do_div(t, su);
1150         *oxoff = su_offset + (stripeno % su_per_object) * su;
1151
1152         /*
1153          * Calculate the length of the extent being written to the selected
1154          * object. This is the minimum of the full length requested (len) or
1155          * the remainder of the current stripe being written to.
1156          */
1157         *oxlen = min_t(u64, len, su - su_offset);
1158
1159         dout(" obj extent %llu~%llu\n", *oxoff, *oxlen);
1160         return 0;
1161
1162 invalid:
1163         dout(" invalid layout\n");
1164         *ono = 0;
1165         *oxoff = 0;
1166         *oxlen = 0;
1167         return -EINVAL;
1168 }
1169 EXPORT_SYMBOL(ceph_calc_file_object_mapping);
1170
1171 /*
1172  * Calculate mapping of a (oloc, oid) pair to a PG.  Should only be
1173  * called with target's (oloc, oid), since tiering isn't taken into
1174  * account.
1175  */
1176 int ceph_oloc_oid_to_pg(struct ceph_osdmap *osdmap,
1177                         struct ceph_object_locator *oloc,
1178                         struct ceph_object_id *oid,
1179                         struct ceph_pg *pg_out)
1180 {
1181         struct ceph_pg_pool_info *pi;
1182
1183         pi = __lookup_pg_pool(&osdmap->pg_pools, oloc->pool);
1184         if (!pi)
1185                 return -EIO;
1186
1187         pg_out->pool = oloc->pool;
1188         pg_out->seed = ceph_str_hash(pi->object_hash, oid->name,
1189                                      oid->name_len);
1190
1191         dout("%s '%.*s' pgid %llu.%x\n", __func__, oid->name_len, oid->name,
1192              pg_out->pool, pg_out->seed);
1193         return 0;
1194 }
1195 EXPORT_SYMBOL(ceph_oloc_oid_to_pg);
1196
1197 static int do_crush(struct ceph_osdmap *map, int ruleno, int x,
1198                     int *result, int result_max,
1199                     const __u32 *weight, int weight_max)
1200 {
1201         int r;
1202
1203         BUG_ON(result_max > CEPH_PG_MAX_SIZE);
1204
1205         mutex_lock(&map->crush_scratch_mutex);
1206         r = crush_do_rule(map->crush, ruleno, x, result, result_max,
1207                           weight, weight_max, map->crush_scratch_ary);
1208         mutex_unlock(&map->crush_scratch_mutex);
1209
1210         return r;
1211 }
1212
1213 /*
1214  * Calculate raw osd vector for the given pgid.  Return pointer to osd
1215  * array, or NULL on failure.
1216  */
1217 static int *calc_pg_raw(struct ceph_osdmap *osdmap, struct ceph_pg pgid,
1218                         int *osds, int *num)
1219 {
1220         struct ceph_pg_mapping *pg;
1221         struct ceph_pg_pool_info *pool;
1222         int ruleno;
1223         int r;
1224         u32 pps;
1225
1226         pool = __lookup_pg_pool(&osdmap->pg_pools, pgid.pool);
1227         if (!pool)
1228                 return NULL;
1229
1230         /* pg_temp? */
1231         pgid.seed = ceph_stable_mod(pgid.seed, pool->pg_num,
1232                                     pool->pg_num_mask);
1233         pg = __lookup_pg_mapping(&osdmap->pg_temp, pgid);
1234         if (pg) {
1235                 *num = pg->len;
1236                 return pg->osds;
1237         }
1238
1239         /* crush */
1240         ruleno = crush_find_rule(osdmap->crush, pool->crush_ruleset,
1241                                  pool->type, pool->size);
1242         if (ruleno < 0) {
1243                 pr_err("no crush rule pool %lld ruleset %d type %d size %d\n",
1244                        pgid.pool, pool->crush_ruleset, pool->type,
1245                        pool->size);
1246                 return NULL;
1247         }
1248
1249         if (pool->flags & CEPH_POOL_FLAG_HASHPSPOOL) {
1250                 /* hash pool id and seed sothat pool PGs do not overlap */
1251                 pps = crush_hash32_2(CRUSH_HASH_RJENKINS1,
1252                                      ceph_stable_mod(pgid.seed, pool->pgp_num,
1253                                                      pool->pgp_num_mask),
1254                                      pgid.pool);
1255         } else {
1256                 /*
1257                  * legacy ehavior: add ps and pool together.  this is
1258                  * not a great approach because the PGs from each pool
1259                  * will overlap on top of each other: 0.5 == 1.4 ==
1260                  * 2.3 == ...
1261                  */
1262                 pps = ceph_stable_mod(pgid.seed, pool->pgp_num,
1263                                       pool->pgp_num_mask) +
1264                         (unsigned)pgid.pool;
1265         }
1266         r = do_crush(osdmap, ruleno, pps, osds, min_t(int, pool->size, *num),
1267                      osdmap->osd_weight, osdmap->max_osd);
1268         if (r < 0) {
1269                 pr_err("error %d from crush rule: pool %lld ruleset %d type %d"
1270                        " size %d\n", r, pgid.pool, pool->crush_ruleset,
1271                        pool->type, pool->size);
1272                 return NULL;
1273         }
1274         *num = r;
1275         return osds;
1276 }
1277
1278 /*
1279  * Return acting set for given pgid.
1280  */
1281 int ceph_calc_pg_acting(struct ceph_osdmap *osdmap, struct ceph_pg pgid,
1282                         int *acting)
1283 {
1284         int rawosds[CEPH_PG_MAX_SIZE], *osds;
1285         int i, o, num = CEPH_PG_MAX_SIZE;
1286
1287         osds = calc_pg_raw(osdmap, pgid, rawosds, &num);
1288         if (!osds)
1289                 return -1;
1290
1291         /* primary is first up osd */
1292         o = 0;
1293         for (i = 0; i < num; i++)
1294                 if (ceph_osd_is_up(osdmap, osds[i]))
1295                         acting[o++] = osds[i];
1296         return o;
1297 }
1298
1299 /*
1300  * Return primary osd for given pgid, or -1 if none.
1301  */
1302 int ceph_calc_pg_primary(struct ceph_osdmap *osdmap, struct ceph_pg pgid)
1303 {
1304         int rawosds[CEPH_PG_MAX_SIZE], *osds;
1305         int i, num = CEPH_PG_MAX_SIZE;
1306
1307         osds = calc_pg_raw(osdmap, pgid, rawosds, &num);
1308         if (!osds)
1309                 return -1;
1310
1311         /* primary is first up osd */
1312         for (i = 0; i < num; i++)
1313                 if (ceph_osd_is_up(osdmap, osds[i]))
1314                         return osds[i];
1315         return -1;
1316 }
1317 EXPORT_SYMBOL(ceph_calc_pg_primary);