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) and primary_temp (explicit primary setting)
 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->primary_temp)) {
 637                 struct ceph_pg_mapping *pg =
 638                         rb_entry(rb_first(&map->primary_temp),
 639                                  struct ceph_pg_mapping, node);
 640                 rb_erase(&pg->node, &map->primary_temp);
 641                 kfree(pg);
 642         }
 643         while (!RB_EMPTY_ROOT(&map->pg_pools)) {
 644                 struct ceph_pg_pool_info *pi =
 645                         rb_entry(rb_first(&map->pg_pools),
 646                                  struct ceph_pg_pool_info, node);
 647                 __remove_pg_pool(&map->pg_pools, pi);
 648         }
 649         kfree(map->osd_state);
 650         kfree(map->osd_weight);
 651         kfree(map->osd_addr);
 652         kfree(map->osd_primary_affinity);
 653         kfree(map);
 654 }
 655
 656 /*
 657  * Adjust max_osd value, (re)allocate arrays.
 658  *
 659  * The new elements are properly initialized.
 660  */
 661 static int osdmap_set_max_osd(struct ceph_osdmap *map, int max)
 662 {
 663         u8 *state;
 664         u32 *weight;
 665         struct ceph_entity_addr *addr;
 666         int i;
 667
 668         state = krealloc(map->osd_state, max*sizeof(*state), GFP_NOFS);
 669         weight = krealloc(map->osd_weight, max*sizeof(*weight), GFP_NOFS);
 670         addr = krealloc(map->osd_addr, max*sizeof(*addr), GFP_NOFS);
 671         if (!state || !weight || !addr) {
 672                 kfree(state);
 673                 kfree(weight);
 674                 kfree(addr);
 675
 676                 return -ENOMEM;
 677         }
 678
 679         for (i = map->max_osd; i < max; i++) {
 680                 state[i] = 0;
 681                 weight[i] = CEPH_OSD_OUT;
 682                 memset(addr + i, 0, sizeof(*addr));
 683         }
 684
 685         map->osd_state = state;
 686         map->osd_weight = weight;
 687         map->osd_addr = addr;
 688
 689         if (map->osd_primary_affinity) {
 690                 u32 *affinity;
 691
 692                 affinity = krealloc(map->osd_primary_affinity,
 693                                     max*sizeof(*affinity), GFP_NOFS);
 694                 if (!affinity)
 695                         return -ENOMEM;
 696
 697                 for (i = map->max_osd; i < max; i++)
 698                         affinity[i] = CEPH_OSD_DEFAULT_PRIMARY_AFFINITY;
 699
 700                 map->osd_primary_affinity = affinity;
 701         }
 702
 703         map->max_osd = max;
 704
 705         return 0;
 706 }
 707
 708 #define OSDMAP_WRAPPER_COMPAT_VER       7
 709 #define OSDMAP_CLIENT_DATA_COMPAT_VER   1
 710
 711 /*
 712  * Return 0 or error.  On success, *v is set to 0 for old (v6) osdmaps,
 713  * to struct_v of the client_data section for new (v7 and above)
 714  * osdmaps.
 715  */
 716 static int get_osdmap_client_data_v(void **p, void *end,
 717                                     const char *prefix, u8 *v)
 718 {
 719         u8 struct_v;
 720
 721         ceph_decode_8_safe(p, end, struct_v, e_inval);
 722         if (struct_v >= 7) {
 723                 u8 struct_compat;
 724
 725                 ceph_decode_8_safe(p, end, struct_compat, e_inval);
 726                 if (struct_compat > OSDMAP_WRAPPER_COMPAT_VER) {
 727                         pr_warning("got v %d cv %d > %d of %s ceph_osdmap\n",
 728                                    struct_v, struct_compat,
 729                                    OSDMAP_WRAPPER_COMPAT_VER, prefix);
 730                         return -EINVAL;
 731                 }
 732                 *p += 4; /* ignore wrapper struct_len */
 733
 734                 ceph_decode_8_safe(p, end, struct_v, e_inval);
 735                 ceph_decode_8_safe(p, end, struct_compat, e_inval);
 736                 if (struct_compat > OSDMAP_CLIENT_DATA_COMPAT_VER) {
 737                         pr_warning("got v %d cv %d > %d of %s ceph_osdmap client data\n",
 738                                    struct_v, struct_compat,
 739                                    OSDMAP_CLIENT_DATA_COMPAT_VER, prefix);
 740                         return -EINVAL;
 741                 }
 742                 *p += 4; /* ignore client data struct_len */
 743         } else {
 744                 u16 version;
 745
 746                 *p -= 1;
 747                 ceph_decode_16_safe(p, end, version, e_inval);
 748                 if (version < 6) {
 749                         pr_warning("got v %d < 6 of %s ceph_osdmap\n", version,
 750                                    prefix);
 751                         return -EINVAL;
 752                 }
 753
 754                 /* old osdmap enconding */
 755                 struct_v = 0;
 756         }
 757
 758         *v = struct_v;
 759         return 0;
 760
 761 e_inval:
 762         return -EINVAL;
 763 }
 764
 765 static int __decode_pools(void **p, void *end, struct ceph_osdmap *map,
 766                           bool incremental)
 767 {
 768         u32 n;
 769
 770         ceph_decode_32_safe(p, end, n, e_inval);
 771         while (n--) {
 772                 struct ceph_pg_pool_info *pi;
 773                 u64 pool;
 774                 int ret;
 775
 776                 ceph_decode_64_safe(p, end, pool, e_inval);
 777
 778                 pi = __lookup_pg_pool(&map->pg_pools, pool);
 779                 if (!incremental || !pi) {
 780                         pi = kzalloc(sizeof(*pi), GFP_NOFS);
 781                         if (!pi)
 782                                 return -ENOMEM;
 783
 784                         pi->id = pool;
 785
 786                         ret = __insert_pg_pool(&map->pg_pools, pi);
 787                         if (ret) {
 788                                 kfree(pi);
 789                                 return ret;
 790                         }
 791                 }
 792
 793                 ret = decode_pool(p, end, pi);
 794                 if (ret)
 795                         return ret;
 796         }
 797
 798         return 0;
 799
 800 e_inval:
 801         return -EINVAL;
 802 }
 803
 804 static int decode_pools(void **p, void *end, struct ceph_osdmap *map)
 805 {
 806         return __decode_pools(p, end, map, false);
 807 }
 808
 809 static int decode_new_pools(void **p, void *end, struct ceph_osdmap *map)
 810 {
 811         return __decode_pools(p, end, map, true);
 812 }
 813
 814 static int __decode_pg_temp(void **p, void *end, struct ceph_osdmap *map,
 815                             bool incremental)
 816 {
 817         u32 n;
 818
 819         ceph_decode_32_safe(p, end, n, e_inval);
 820         while (n--) {
 821                 struct ceph_pg pgid;
 822                 u32 len, i;
 823                 int ret;
 824
 825                 ret = ceph_decode_pgid(p, end, &pgid);
 826                 if (ret)
 827                         return ret;
 828
 829                 ceph_decode_32_safe(p, end, len, e_inval);
 830
 831                 ret = __remove_pg_mapping(&map->pg_temp, pgid);
 832                 BUG_ON(!incremental && ret != -ENOENT);
 833
 834                 if (!incremental || len > 0) {
 835                         struct ceph_pg_mapping *pg;
 836
 837                         ceph_decode_need(p, end, len*sizeof(u32), e_inval);
 838
 839                         if (len > (UINT_MAX - sizeof(*pg)) / sizeof(u32))
 840                                 return -EINVAL;
 841
 842                         pg = kzalloc(sizeof(*pg) + len*sizeof(u32), GFP_NOFS);
 843                         if (!pg)
 844                                 return -ENOMEM;
 845
 846                         pg->pgid = pgid;
 847                         pg->pg_temp.len = len;
 848                         for (i = 0; i < len; i++)
 849                                 pg->pg_temp.osds[i] = ceph_decode_32(p);
 850
 851                         ret = __insert_pg_mapping(pg, &map->pg_temp);
 852                         if (ret) {
 853                                 kfree(pg);
 854                                 return ret;
 855                         }
 856                 }
 857         }
 858
 859         return 0;
 860
 861 e_inval:
 862         return -EINVAL;
 863 }
 864
 865 static int decode_pg_temp(void **p, void *end, struct ceph_osdmap *map)
 866 {
 867         return __decode_pg_temp(p, end, map, false);
 868 }
 869
 870 static int decode_new_pg_temp(void **p, void *end, struct ceph_osdmap *map)
 871 {
 872         return __decode_pg_temp(p, end, map, true);
 873 }
 874
 875 static int __decode_primary_temp(void **p, void *end, struct ceph_osdmap *map,
 876                                  bool incremental)
 877 {
 878         u32 n;
 879
 880         ceph_decode_32_safe(p, end, n, e_inval);
 881         while (n--) {
 882                 struct ceph_pg pgid;
 883                 u32 osd;
 884                 int ret;
 885
 886                 ret = ceph_decode_pgid(p, end, &pgid);
 887                 if (ret)
 888                         return ret;
 889
 890                 ceph_decode_32_safe(p, end, osd, e_inval);
 891
 892                 ret = __remove_pg_mapping(&map->primary_temp, pgid);
 893                 BUG_ON(!incremental && ret != -ENOENT);
 894
 895                 if (!incremental || osd != (u32)-1) {
 896                         struct ceph_pg_mapping *pg;
 897
 898                         pg = kzalloc(sizeof(*pg), GFP_NOFS);
 899                         if (!pg)
 900                                 return -ENOMEM;
 901
 902                         pg->pgid = pgid;
 903                         pg->primary_temp.osd = osd;
 904
 905                         ret = __insert_pg_mapping(pg, &map->primary_temp);
 906                         if (ret) {
 907                                 kfree(pg);
 908                                 return ret;
 909                         }
 910                 }
 911         }
 912
 913         return 0;
 914
 915 e_inval:
 916         return -EINVAL;
 917 }
 918
 919 static int decode_primary_temp(void **p, void *end, struct ceph_osdmap *map)
 920 {
 921         return __decode_primary_temp(p, end, map, false);
 922 }
 923
 924 static int decode_new_primary_temp(void **p, void *end,
 925                                    struct ceph_osdmap *map)
 926 {
 927         return __decode_primary_temp(p, end, map, true);
 928 }
 929
 930 u32 ceph_get_primary_affinity(struct ceph_osdmap *map, int osd)
 931 {
 932         BUG_ON(osd >= map->max_osd);
 933
 934         if (!map->osd_primary_affinity)
 935                 return CEPH_OSD_DEFAULT_PRIMARY_AFFINITY;
 936
 937         return map->osd_primary_affinity[osd];
 938 }
 939
 940 static int set_primary_affinity(struct ceph_osdmap *map, int osd, u32 aff)
 941 {
 942         BUG_ON(osd >= map->max_osd);
 943
 944         if (!map->osd_primary_affinity) {
 945                 int i;
 946
 947                 map->osd_primary_affinity = kmalloc(map->max_osd*sizeof(u32),
 948                                                     GFP_NOFS);
 949                 if (!map->osd_primary_affinity)
 950                         return -ENOMEM;
 951
 952                 for (i = 0; i < map->max_osd; i++)
 953                         map->osd_primary_affinity[i] =
 954                             CEPH_OSD_DEFAULT_PRIMARY_AFFINITY;
 955         }
 956
 957         map->osd_primary_affinity[osd] = aff;
 958
 959         return 0;
 960 }
 961
 962 static int decode_primary_affinity(void **p, void *end,
 963                                    struct ceph_osdmap *map)
 964 {
 965         u32 len, i;
 966
 967         ceph_decode_32_safe(p, end, len, e_inval);
 968         if (len == 0) {
 969                 kfree(map->osd_primary_affinity);
 970                 map->osd_primary_affinity = NULL;
 971                 return 0;
 972         }
 973         if (len != map->max_osd)
 974                 goto e_inval;
 975
 976         ceph_decode_need(p, end, map->max_osd*sizeof(u32), e_inval);
 977
 978         for (i = 0; i < map->max_osd; i++) {
 979                 int ret;
 980
 981                 ret = set_primary_affinity(map, i, ceph_decode_32(p));
 982                 if (ret)
 983                         return ret;
 984         }
 985
 986         return 0;
 987
 988 e_inval:
 989         return -EINVAL;
 990 }
 991
 992 static int decode_new_primary_affinity(void **p, void *end,
 993                                        struct ceph_osdmap *map)
 994 {
 995         u32 n;
 996
 997         ceph_decode_32_safe(p, end, n, e_inval);
 998         while (n--) {
 999                 u32 osd, aff;
1000                 int ret;
1001
1002                 ceph_decode_32_safe(p, end, osd, e_inval);
1003                 ceph_decode_32_safe(p, end, aff, e_inval);
1004
1005                 ret = set_primary_affinity(map, osd, aff);
1006                 if (ret)
1007                         return ret;
1008
1009                 pr_info("osd%d primary-affinity 0x%x\n", osd, aff);
1010         }
1011
1012         return 0;
1013
1014 e_inval:
1015         return -EINVAL;
1016 }
1017
1018 /*
1019  * decode a full map.
1020  */
1021 static int osdmap_decode(void **p, void *end, struct ceph_osdmap *map)
1022 {
1023         u8 struct_v;
1024         u32 epoch = 0;
1025         void *start = *p;
1026         u32 max;
1027         u32 len, i;
1028         int err;
1029
1030         dout("%s %p to %p len %d\n", __func__, *p, end, (int)(end - *p));
1031
1032         err = get_osdmap_client_data_v(p, end, "full", &struct_v);
1033         if (err)
1034                 goto bad;
1035
1036         /* fsid, epoch, created, modified */
1037         ceph_decode_need(p, end, sizeof(map->fsid) + sizeof(u32) +
1038                          sizeof(map->created) + sizeof(map->modified), e_inval);
1039         ceph_decode_copy(p, &map->fsid, sizeof(map->fsid));
1040         epoch = map->epoch = ceph_decode_32(p);
1041         ceph_decode_copy(p, &map->created, sizeof(map->created));
1042         ceph_decode_copy(p, &map->modified, sizeof(map->modified));
1043
1044         /* pools */
1045         err = decode_pools(p, end, map);
1046         if (err)
1047                 goto bad;
1048
1049         /* pool_name */
1050         err = decode_pool_names(p, end, map);
1051         if (err)
1052                 goto bad;
1053
1054         ceph_decode_32_safe(p, end, map->pool_max, e_inval);
1055
1056         ceph_decode_32_safe(p, end, map->flags, e_inval);
1057
1058         /* max_osd */
1059         ceph_decode_32_safe(p, end, max, e_inval);
1060
1061         /* (re)alloc osd arrays */
1062         err = osdmap_set_max_osd(map, max);
1063         if (err)
1064                 goto bad;
1065
1066         /* osd_state, osd_weight, osd_addrs->client_addr */
1067         ceph_decode_need(p, end, 3*sizeof(u32) +
1068                          map->max_osd*(1 + sizeof(*map->osd_weight) +
1069                                        sizeof(*map->osd_addr)), e_inval);
1070
1071         if (ceph_decode_32(p) != map->max_osd)
1072                 goto e_inval;
1073
1074         ceph_decode_copy(p, map->osd_state, map->max_osd);
1075
1076         if (ceph_decode_32(p) != map->max_osd)
1077                 goto e_inval;
1078
1079         for (i = 0; i < map->max_osd; i++)
1080                 map->osd_weight[i] = ceph_decode_32(p);
1081
1082         if (ceph_decode_32(p) != map->max_osd)
1083                 goto e_inval;
1084
1085         ceph_decode_copy(p, map->osd_addr, map->max_osd*sizeof(*map->osd_addr));
1086         for (i = 0; i < map->max_osd; i++)
1087                 ceph_decode_addr(&map->osd_addr[i]);
1088
1089         /* pg_temp */
1090         err = decode_pg_temp(p, end, map);
1091         if (err)
1092                 goto bad;
1093
1094         /* primary_temp */
1095         if (struct_v >= 1) {
1096                 err = decode_primary_temp(p, end, map);
1097                 if (err)
1098                         goto bad;
1099         }
1100
1101         /* primary_affinity */
1102         if (struct_v >= 2) {
1103                 err = decode_primary_affinity(p, end, map);
1104                 if (err)
1105                         goto bad;
1106         } else {
1107                 /* XXX can this happen? */
1108                 kfree(map->osd_primary_affinity);
1109                 map->osd_primary_affinity = NULL;
1110         }
1111
1112         /* crush */
1113         ceph_decode_32_safe(p, end, len, e_inval);
1114         map->crush = crush_decode(*p, min(*p + len, end));
1115         if (IS_ERR(map->crush)) {
1116                 err = PTR_ERR(map->crush);
1117                 map->crush = NULL;
1118                 goto bad;
1119         }
1120         *p += len;
1121
1122         /* ignore the rest */
1123         *p = end;
1124
1125         dout("full osdmap epoch %d max_osd %d\n", map->epoch, map->max_osd);
1126         return 0;
1127
1128 e_inval:
1129         err = -EINVAL;
1130 bad:
1131         pr_err("corrupt full osdmap (%d) epoch %d off %d (%p of %p-%p)\n",
1132                err, epoch, (int)(*p - start), *p, start, end);
1133         print_hex_dump(KERN_DEBUG, "osdmap: ",
1134                        DUMP_PREFIX_OFFSET, 16, 1,
1135                        start, end - start, true);
1136         return err;
1137 }
1138
1139 /*
1140  * Allocate and decode a full map.
1141  */
1142 struct ceph_osdmap *ceph_osdmap_decode(void **p, void *end)
1143 {
1144         struct ceph_osdmap *map;
1145         int ret;
1146
1147         map = kzalloc(sizeof(*map), GFP_NOFS);
1148         if (!map)
1149                 return ERR_PTR(-ENOMEM);
1150
1151         map->pg_temp = RB_ROOT;
1152         map->primary_temp = RB_ROOT;
1153         mutex_init(&map->crush_scratch_mutex);
1154
1155         ret = osdmap_decode(p, end, map);
1156         if (ret) {
1157                 ceph_osdmap_destroy(map);
1158                 return ERR_PTR(ret);
1159         }
1160
1161         return map;
1162 }
1163
1164 /*
1165  * decode and apply an incremental map update.
1166  */
1167 struct ceph_osdmap *osdmap_apply_incremental(void **p, void *end,
1168                                              struct ceph_osdmap *map,
1169                                              struct ceph_messenger *msgr)
1170 {
1171         struct crush_map *newcrush = NULL;
1172         struct ceph_fsid fsid;
1173         u32 epoch = 0;
1174         struct ceph_timespec modified;
1175         s32 len;
1176         u64 pool;
1177         __s64 new_pool_max;
1178         __s32 new_flags, max;
1179         void *start = *p;
1180         int err;
1181         u8 struct_v;
1182
1183         dout("%s %p to %p len %d\n", __func__, *p, end, (int)(end - *p));
1184
1185         err = get_osdmap_client_data_v(p, end, "inc", &struct_v);
1186         if (err)
1187                 goto bad;
1188
1189         /* fsid, epoch, modified, new_pool_max, new_flags */
1190         ceph_decode_need(p, end, sizeof(fsid) + sizeof(u32) + sizeof(modified) +
1191                          sizeof(u64) + sizeof(u32), e_inval);
1192         ceph_decode_copy(p, &fsid, sizeof(fsid));
1193         epoch = ceph_decode_32(p);
1194         BUG_ON(epoch != map->epoch+1);
1195         ceph_decode_copy(p, &modified, sizeof(modified));
1196         new_pool_max = ceph_decode_64(p);
1197         new_flags = ceph_decode_32(p);
1198
1199         /* full map? */
1200         ceph_decode_32_safe(p, end, len, e_inval);
1201         if (len > 0) {
1202                 dout("apply_incremental full map len %d, %p to %p\n",
1203                      len, *p, end);
1204                 return ceph_osdmap_decode(p, min(*p+len, end));
1205         }
1206
1207         /* new crush? */
1208         ceph_decode_32_safe(p, end, len, e_inval);
1209         if (len > 0) {
1210                 newcrush = crush_decode(*p, min(*p+len, end));
1211                 if (IS_ERR(newcrush)) {
1212                         err = PTR_ERR(newcrush);
1213                         newcrush = NULL;
1214                         goto bad;
1215                 }
1216                 *p += len;
1217         }
1218
1219         /* new flags? */
1220         if (new_flags >= 0)
1221                 map->flags = new_flags;
1222         if (new_pool_max >= 0)
1223                 map->pool_max = new_pool_max;
1224
1225         /* new max? */
1226         ceph_decode_32_safe(p, end, max, e_inval);
1227         if (max >= 0) {
1228                 err = osdmap_set_max_osd(map, max);
1229                 if (err)
1230                         goto bad;
1231         }
1232
1233         map->epoch++;
1234         map->modified = modified;
1235         if (newcrush) {
1236                 if (map->crush)
1237                         crush_destroy(map->crush);
1238                 map->crush = newcrush;
1239                 newcrush = NULL;
1240         }
1241
1242         /* new_pools */
1243         err = decode_new_pools(p, end, map);
1244         if (err)
1245                 goto bad;
1246
1247         /* new_pool_names */
1248         err = decode_pool_names(p, end, map);
1249         if (err)
1250                 goto bad;
1251
1252         /* old_pool */
1253         ceph_decode_32_safe(p, end, len, e_inval);
1254         while (len--) {
1255                 struct ceph_pg_pool_info *pi;
1256
1257                 ceph_decode_64_safe(p, end, pool, e_inval);
1258                 pi = __lookup_pg_pool(&map->pg_pools, pool);
1259                 if (pi)
1260                         __remove_pg_pool(&map->pg_pools, pi);
1261         }
1262
1263         /* new_up */
1264         ceph_decode_32_safe(p, end, len, e_inval);
1265         while (len--) {
1266                 u32 osd;
1267                 struct ceph_entity_addr addr;
1268                 ceph_decode_32_safe(p, end, osd, e_inval);
1269                 ceph_decode_copy_safe(p, end, &addr, sizeof(addr), e_inval);
1270                 ceph_decode_addr(&addr);
1271                 pr_info("osd%d up\n", osd);
1272                 BUG_ON(osd >= map->max_osd);
1273                 map->osd_state[osd] |= CEPH_OSD_UP;
1274                 map->osd_addr[osd] = addr;
1275         }
1276
1277         /* new_state */
1278         ceph_decode_32_safe(p, end, len, e_inval);
1279         while (len--) {
1280                 u32 osd;
1281                 u8 xorstate;
1282                 ceph_decode_32_safe(p, end, osd, e_inval);
1283                 xorstate = **(u8 **)p;
1284                 (*p)++;  /* clean flag */
1285                 if (xorstate == 0)
1286                         xorstate = CEPH_OSD_UP;
1287                 if (xorstate & CEPH_OSD_UP)
1288                         pr_info("osd%d down\n", osd);
1289                 if (osd < map->max_osd)
1290                         map->osd_state[osd] ^= xorstate;
1291         }
1292
1293         /* new_weight */
1294         ceph_decode_32_safe(p, end, len, e_inval);
1295         while (len--) {
1296                 u32 osd, off;
1297                 ceph_decode_need(p, end, sizeof(u32)*2, e_inval);
1298                 osd = ceph_decode_32(p);
1299                 off = ceph_decode_32(p);
1300                 pr_info("osd%d weight 0x%x %s\n", osd, off,
1301                      off == CEPH_OSD_IN ? "(in)" :
1302                      (off == CEPH_OSD_OUT ? "(out)" : ""));
1303                 if (osd < map->max_osd)
1304                         map->osd_weight[osd] = off;
1305         }
1306
1307         /* new_pg_temp */
1308         err = decode_new_pg_temp(p, end, map);
1309         if (err)
1310                 goto bad;
1311
1312         /* new_primary_temp */
1313         if (struct_v >= 1) {
1314                 err = decode_new_primary_temp(p, end, map);
1315                 if (err)
1316                         goto bad;
1317         }
1318
1319         /* new_primary_affinity */
1320         if (struct_v >= 2) {
1321                 err = decode_new_primary_affinity(p, end, map);
1322                 if (err)
1323                         goto bad;
1324         }
1325
1326         /* ignore the rest */
1327         *p = end;
1328
1329         dout("inc osdmap epoch %d max_osd %d\n", map->epoch, map->max_osd);
1330         return map;
1331
1332 e_inval:
1333         err = -EINVAL;
1334 bad:
1335         pr_err("corrupt inc osdmap (%d) epoch %d off %d (%p of %p-%p)\n",
1336                err, epoch, (int)(*p - start), *p, start, end);
1337         print_hex_dump(KERN_DEBUG, "osdmap: ",
1338                        DUMP_PREFIX_OFFSET, 16, 1,
1339                        start, end - start, true);
1340         if (newcrush)
1341                 crush_destroy(newcrush);
1342         return ERR_PTR(err);
1343 }
1344
1345
1346
1347
1348 /*
1349  * calculate file layout from given offset, length.
1350  * fill in correct oid, logical length, and object extent
1351  * offset, length.
1352  *
1353  * for now, we write only a single su, until we can
1354  * pass a stride back to the caller.
1355  */
1356 int ceph_calc_file_object_mapping(struct ceph_file_layout *layout,
1357                                    u64 off, u64 len,
1358                                    u64 *ono,
1359                                    u64 *oxoff, u64 *oxlen)
1360 {
1361         u32 osize = le32_to_cpu(layout->fl_object_size);
1362         u32 su = le32_to_cpu(layout->fl_stripe_unit);
1363         u32 sc = le32_to_cpu(layout->fl_stripe_count);
1364         u32 bl, stripeno, stripepos, objsetno;
1365         u32 su_per_object;
1366         u64 t, su_offset;
1367
1368         dout("mapping %llu~%llu  osize %u fl_su %u\n", off, len,
1369              osize, su);
1370         if (su == 0 || sc == 0)
1371                 goto invalid;
1372         su_per_object = osize / su;
1373         if (su_per_object == 0)
1374                 goto invalid;
1375         dout("osize %u / su %u = su_per_object %u\n", osize, su,
1376              su_per_object);
1377
1378         if ((su & ~PAGE_MASK) != 0)
1379                 goto invalid;
1380
1381         /* bl = *off / su; */
1382         t = off;
1383         do_div(t, su);
1384         bl = t;
1385         dout("off %llu / su %u = bl %u\n", off, su, bl);
1386
1387         stripeno = bl / sc;
1388         stripepos = bl % sc;
1389         objsetno = stripeno / su_per_object;
1390
1391         *ono = objsetno * sc + stripepos;
1392         dout("objset %u * sc %u = ono %u\n", objsetno, sc, (unsigned int)*ono);
1393
1394         /* *oxoff = *off % layout->fl_stripe_unit;  # offset in su */
1395         t = off;
1396         su_offset = do_div(t, su);
1397         *oxoff = su_offset + (stripeno % su_per_object) * su;
1398
1399         /*
1400          * Calculate the length of the extent being written to the selected
1401          * object. This is the minimum of the full length requested (len) or
1402          * the remainder of the current stripe being written to.
1403          */
1404         *oxlen = min_t(u64, len, su - su_offset);
1405
1406         dout(" obj extent %llu~%llu\n", *oxoff, *oxlen);
1407         return 0;
1408
1409 invalid:
1410         dout(" invalid layout\n");
1411         *ono = 0;
1412         *oxoff = 0;
1413         *oxlen = 0;
1414         return -EINVAL;
1415 }
1416 EXPORT_SYMBOL(ceph_calc_file_object_mapping);
1417
1418 /*
1419  * Calculate mapping of a (oloc, oid) pair to a PG.  Should only be
1420  * called with target's (oloc, oid), since tiering isn't taken into
1421  * account.
1422  */
1423 int ceph_oloc_oid_to_pg(struct ceph_osdmap *osdmap,
1424                         struct ceph_object_locator *oloc,
1425                         struct ceph_object_id *oid,
1426                         struct ceph_pg *pg_out)
1427 {
1428         struct ceph_pg_pool_info *pi;
1429
1430         pi = __lookup_pg_pool(&osdmap->pg_pools, oloc->pool);
1431         if (!pi)
1432                 return -EIO;
1433
1434         pg_out->pool = oloc->pool;
1435         pg_out->seed = ceph_str_hash(pi->object_hash, oid->name,
1436                                      oid->name_len);
1437
1438         dout("%s '%.*s' pgid %llu.%x\n", __func__, oid->name_len, oid->name,
1439              pg_out->pool, pg_out->seed);
1440         return 0;
1441 }
1442 EXPORT_SYMBOL(ceph_oloc_oid_to_pg);
1443
1444 static int do_crush(struct ceph_osdmap *map, int ruleno, int x,
1445                     int *result, int result_max,
1446                     const __u32 *weight, int weight_max)
1447 {
1448         int r;
1449
1450         BUG_ON(result_max > CEPH_PG_MAX_SIZE);
1451
1452         mutex_lock(&map->crush_scratch_mutex);
1453         r = crush_do_rule(map->crush, ruleno, x, result, result_max,
1454                           weight, weight_max, map->crush_scratch_ary);
1455         mutex_unlock(&map->crush_scratch_mutex);
1456
1457         return r;
1458 }
1459
1460 /*
1461  * Calculate raw (crush) set for given pgid.
1462  *
1463  * Return raw set length, or error.
1464  */
1465 static int pg_to_raw_osds(struct ceph_osdmap *osdmap,
1466                           struct ceph_pg_pool_info *pool,
1467                           struct ceph_pg pgid, u32 pps, int *osds)
1468 {
1469         int ruleno;
1470         int len;
1471
1472         /* crush */
1473         ruleno = crush_find_rule(osdmap->crush, pool->crush_ruleset,
1474                                  pool->type, pool->size);
1475         if (ruleno < 0) {
1476                 pr_err("no crush rule: pool %lld ruleset %d type %d size %d\n",
1477                        pgid.pool, pool->crush_ruleset, pool->type,
1478                        pool->size);
1479                 return -ENOENT;
1480         }
1481
1482         len = do_crush(osdmap, ruleno, pps, osds,
1483                        min_t(int, pool->size, CEPH_PG_MAX_SIZE),
1484                        osdmap->osd_weight, osdmap->max_osd);
1485         if (len < 0) {
1486                 pr_err("error %d from crush rule %d: pool %lld ruleset %d type %d size %d\n",
1487                        len, ruleno, pgid.pool, pool->crush_ruleset,
1488                        pool->type, pool->size);
1489                 return len;
1490         }
1491
1492         return len;
1493 }
1494
1495 /*
1496  * Given raw set, calculate up set and up primary.
1497  *
1498  * Return up set length.  *primary is set to up primary osd id, or -1
1499  * if up set is empty.
1500  */
1501 static int raw_to_up_osds(struct ceph_osdmap *osdmap,
1502                           struct ceph_pg_pool_info *pool,
1503                           int *osds, int len, int *primary)
1504 {
1505         int up_primary = -1;
1506         int i;
1507
1508         if (ceph_can_shift_osds(pool)) {
1509                 int removed = 0;
1510
1511                 for (i = 0; i < len; i++) {
1512                         if (ceph_osd_is_down(osdmap, osds[i])) {
1513                                 removed++;
1514                                 continue;
1515                         }
1516                         if (removed)
1517                                 osds[i - removed] = osds[i];
1518                 }
1519
1520                 len -= removed;
1521                 if (len > 0)
1522                         up_primary = osds[0];
1523         } else {
1524                 for (i = len - 1; i >= 0; i--) {
1525                         if (ceph_osd_is_down(osdmap, osds[i]))
1526                                 osds[i] = CRUSH_ITEM_NONE;
1527                         else
1528                                 up_primary = osds[i];
1529                 }
1530         }
1531
1532         *primary = up_primary;
1533         return len;
1534 }
1535
1536 static void apply_primary_affinity(struct ceph_osdmap *osdmap, u32 pps,
1537                                    struct ceph_pg_pool_info *pool,
1538                                    int *osds, int len, int *primary)
1539 {
1540         int i;
1541         int pos = -1;
1542
1543         /*
1544          * Do we have any non-default primary_affinity values for these
1545          * osds?
1546          */
1547         if (!osdmap->osd_primary_affinity)
1548                 return;
1549
1550         for (i = 0; i < len; i++) {
1551                 if (osds[i] != CRUSH_ITEM_NONE &&
1552                     osdmap->osd_primary_affinity[i] !=
1553                                         CEPH_OSD_DEFAULT_PRIMARY_AFFINITY) {
1554                         break;
1555                 }
1556         }
1557         if (i == len)
1558                 return;
1559
1560         /*
1561          * Pick the primary.  Feed both the seed (for the pg) and the
1562          * osd into the hash/rng so that a proportional fraction of an
1563          * osd's pgs get rejected as primary.
1564          */
1565         for (i = 0; i < len; i++) {
1566                 int osd;
1567                 u32 aff;
1568
1569                 osd = osds[i];
1570                 if (osd == CRUSH_ITEM_NONE)
1571                         continue;
1572
1573                 aff = osdmap->osd_primary_affinity[osd];
1574                 if (aff < CEPH_OSD_MAX_PRIMARY_AFFINITY &&
1575                     (crush_hash32_2(CRUSH_HASH_RJENKINS1,
1576                                     pps, osd) >> 16) >= aff) {
1577                         /*
1578                          * We chose not to use this primary.  Note it
1579                          * anyway as a fallback in case we don't pick
1580                          * anyone else, but keep looking.
1581                          */
1582                         if (pos < 0)
1583                                 pos = i;
1584                 } else {
1585                         pos = i;
1586                         break;
1587                 }
1588         }
1589         if (pos < 0)
1590                 return;
1591
1592         *primary = osds[pos];
1593
1594         if (ceph_can_shift_osds(pool) && pos > 0) {
1595                 /* move the new primary to the front */
1596                 for (i = pos; i > 0; i--)
1597                         osds[i] = osds[i - 1];
1598                 osds[0] = *primary;
1599         }
1600 }
1601
1602 /*
1603  * Given up set, apply pg_temp and primary_temp mappings.
1604  *
1605  * Return acting set length.  *primary is set to acting primary osd id,
1606  * or -1 if acting set is empty.
1607  */
1608 static int apply_temps(struct ceph_osdmap *osdmap,
1609                        struct ceph_pg_pool_info *pool, struct ceph_pg pgid,
1610                        int *osds, int len, int *primary)
1611 {
1612         struct ceph_pg_mapping *pg;
1613         int temp_len;
1614         int temp_primary;
1615         int i;
1616
1617         /* raw_pg -> pg */
1618         pgid.seed = ceph_stable_mod(pgid.seed, pool->pg_num,
1619                                     pool->pg_num_mask);
1620
1621         /* pg_temp? */
1622         pg = __lookup_pg_mapping(&osdmap->pg_temp, pgid);
1623         if (pg) {
1624                 temp_len = 0;
1625                 temp_primary = -1;
1626
1627                 for (i = 0; i < pg->pg_temp.len; i++) {
1628                         if (ceph_osd_is_down(osdmap, pg->pg_temp.osds[i])) {
1629                                 if (ceph_can_shift_osds(pool))
1630                                         continue;
1631                                 else
1632                                         osds[temp_len++] = CRUSH_ITEM_NONE;
1633                         } else {
1634                                 osds[temp_len++] = pg->pg_temp.osds[i];
1635                         }
1636                 }
1637
1638                 /* apply pg_temp's primary */
1639                 for (i = 0; i < temp_len; i++) {
1640                         if (osds[i] != CRUSH_ITEM_NONE) {
1641                                 temp_primary = osds[i];
1642                                 break;
1643                         }
1644                 }
1645         } else {
1646                 temp_len = len;
1647                 temp_primary = *primary;
1648         }
1649
1650         /* primary_temp? */
1651         pg = __lookup_pg_mapping(&osdmap->primary_temp, pgid);
1652         if (pg)
1653                 temp_primary = pg->primary_temp.osd;
1654
1655         *primary = temp_primary;
1656         return temp_len;
1657 }
1658
1659 /*
1660  * Calculate acting set for given pgid.
1661  *
1662  * Return acting set length, or error.  *primary is set to acting
1663  * primary osd id, or -1 if acting set is empty or on error.
1664  */
1665 int ceph_calc_pg_acting(struct ceph_osdmap *osdmap, struct ceph_pg pgid,
1666                         int *osds, int *primary)
1667 {
1668         struct ceph_pg_pool_info *pool;
1669         u32 pps;
1670         int len;
1671
1672         pool = __lookup_pg_pool(&osdmap->pg_pools, pgid.pool);
1673         if (!pool) {
1674                 *primary = -1;
1675                 return -ENOENT;
1676         }
1677
1678         if (pool->flags & CEPH_POOL_FLAG_HASHPSPOOL) {
1679                 /* hash pool id and seed so that pool PGs do not overlap */
1680                 pps = crush_hash32_2(CRUSH_HASH_RJENKINS1,
1681                                      ceph_stable_mod(pgid.seed, pool->pgp_num,
1682                                                      pool->pgp_num_mask),
1683                                      pgid.pool);
1684         } else {
1685                 /*
1686                  * legacy behavior: add ps and pool together.  this is
1687                  * not a great approach because the PGs from each pool
1688                  * will overlap on top of each other: 0.5 == 1.4 ==
1689                  * 2.3 == ...
1690                  */
1691                 pps = ceph_stable_mod(pgid.seed, pool->pgp_num,
1692                                       pool->pgp_num_mask) +
1693                         (unsigned)pgid.pool;
1694         }
1695
1696         len = pg_to_raw_osds(osdmap, pool, pgid, pps, osds);
1697         if (len < 0) {
1698                 *primary = -1;
1699                 return len;
1700         }
1701
1702         len = raw_to_up_osds(osdmap, pool, osds, len, primary);
1703
1704         apply_primary_affinity(osdmap, pps, pool, osds, len, primary);
1705
1706         len = apply_temps(osdmap, pool, pgid, osds, len, primary);
1707
1708         return len;
1709 }
1710
1711 /*
1712  * Return primary osd for given pgid, or -1 if none.
1713  */
1714 int ceph_calc_pg_primary(struct ceph_osdmap *osdmap, struct ceph_pg pgid)
1715 {
1716         int osds[CEPH_PG_MAX_SIZE];
1717         int primary;
1718
1719         ceph_calc_pg_acting(osdmap, pgid, osds, &primary);
1720
1721         return primary;
1722 }
1723 EXPORT_SYMBOL(ceph_calc_pg_primary);