net/ceph/osdmap.c

   1 // SPDX-License-Identifier: GPL-2.0
   2
   3 #include <linux/ceph/ceph_debug.h>
   4
   5 #include <linux/module.h>
   6 #include <linux/slab.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, u32 state)
  15 {
  16         if (!len)
  17                 return str;
  18
  19         if ((state & CEPH_OSD_EXISTS) && (state & CEPH_OSD_UP))
  20                 snprintf(str, len, "exists, up");
  21         else if (state & CEPH_OSD_EXISTS)
  22                 snprintf(str, len, "exists");
  23         else if (state & CEPH_OSD_UP)
  24                 snprintf(str, len, "up");
  25         else
  26                 snprintf(str, len, "doesn't exist");
  27
  28         return str;
  29 }
  30
  31 /* maps */
  32
  33 static int calc_bits_of(unsigned int t)
  34 {
  35         int b = 0;
  36         while (t) {
  37                 t = t >> 1;
  38                 b++;
  39         }
  40         return b;
  41 }
  42
  43 /*
  44  * the foo_mask is the smallest value 2^n-1 that is >= foo.
  45  */
  46 static void calc_pg_masks(struct ceph_pg_pool_info *pi)
  47 {
  48         pi->pg_num_mask = (1 << calc_bits_of(pi->pg_num-1)) - 1;
  49         pi->pgp_num_mask = (1 << calc_bits_of(pi->pgp_num-1)) - 1;
  50 }
  51
  52 /*
  53  * decode crush map
  54  */
  55 static int crush_decode_uniform_bucket(void **p, void *end,
  56                                        struct crush_bucket_uniform *b)
  57 {
  58         dout("crush_decode_uniform_bucket %p to %p\n", *p, end);
  59         ceph_decode_need(p, end, (1+b->h.size) * sizeof(u32), bad);
  60         b->item_weight = ceph_decode_32(p);
  61         return 0;
  62 bad:
  63         return -EINVAL;
  64 }
  65
  66 static int crush_decode_list_bucket(void **p, void *end,
  67                                     struct crush_bucket_list *b)
  68 {
  69         int j;
  70         dout("crush_decode_list_bucket %p to %p\n", *p, end);
  71         b->item_weights = kcalloc(b->h.size, sizeof(u32), GFP_NOFS);
  72         if (b->item_weights == NULL)
  73                 return -ENOMEM;
  74         b->sum_weights = kcalloc(b->h.size, sizeof(u32), GFP_NOFS);
  75         if (b->sum_weights == NULL)
  76                 return -ENOMEM;
  77         ceph_decode_need(p, end, 2 * b->h.size * sizeof(u32), bad);
  78         for (j = 0; j < b->h.size; j++) {
  79                 b->item_weights[j] = ceph_decode_32(p);
  80                 b->sum_weights[j] = ceph_decode_32(p);
  81         }
  82         return 0;
  83 bad:
  84         return -EINVAL;
  85 }
  86
  87 static int crush_decode_tree_bucket(void **p, void *end,
  88                                     struct crush_bucket_tree *b)
  89 {
  90         int j;
  91         dout("crush_decode_tree_bucket %p to %p\n", *p, end);
  92         ceph_decode_8_safe(p, end, b->num_nodes, bad);
  93         b->node_weights = kcalloc(b->num_nodes, sizeof(u32), GFP_NOFS);
  94         if (b->node_weights == NULL)
  95                 return -ENOMEM;
  96         ceph_decode_need(p, end, b->num_nodes * sizeof(u32), bad);
  97         for (j = 0; j < b->num_nodes; j++)
  98                 b->node_weights[j] = ceph_decode_32(p);
  99         return 0;
 100 bad:
 101         return -EINVAL;
 102 }
 103
 104 static int crush_decode_straw_bucket(void **p, void *end,
 105                                      struct crush_bucket_straw *b)
 106 {
 107         int j;
 108         dout("crush_decode_straw_bucket %p to %p\n", *p, end);
 109         b->item_weights = kcalloc(b->h.size, sizeof(u32), GFP_NOFS);
 110         if (b->item_weights == NULL)
 111                 return -ENOMEM;
 112         b->straws = kcalloc(b->h.size, sizeof(u32), GFP_NOFS);
 113         if (b->straws == NULL)
 114                 return -ENOMEM;
 115         ceph_decode_need(p, end, 2 * b->h.size * sizeof(u32), bad);
 116         for (j = 0; j < b->h.size; j++) {
 117                 b->item_weights[j] = ceph_decode_32(p);
 118                 b->straws[j] = ceph_decode_32(p);
 119         }
 120         return 0;
 121 bad:
 122         return -EINVAL;
 123 }
 124
 125 static int crush_decode_straw2_bucket(void **p, void *end,
 126                                       struct crush_bucket_straw2 *b)
 127 {
 128         int j;
 129         dout("crush_decode_straw2_bucket %p to %p\n", *p, end);
 130         b->item_weights = kcalloc(b->h.size, sizeof(u32), GFP_NOFS);
 131         if (b->item_weights == NULL)
 132                 return -ENOMEM;
 133         ceph_decode_need(p, end, b->h.size * sizeof(u32), bad);
 134         for (j = 0; j < b->h.size; j++)
 135                 b->item_weights[j] = ceph_decode_32(p);
 136         return 0;
 137 bad:
 138         return -EINVAL;
 139 }
 140
 141 static struct crush_choose_arg_map *alloc_choose_arg_map(void)
 142 {
 143         struct crush_choose_arg_map *arg_map;
 144
 145         arg_map = kzalloc(sizeof(*arg_map), GFP_NOIO);
 146         if (!arg_map)
 147                 return NULL;
 148
 149         RB_CLEAR_NODE(&arg_map->node);
 150         return arg_map;
 151 }
 152
 153 static void free_choose_arg_map(struct crush_choose_arg_map *arg_map)
 154 {
 155         if (arg_map) {
 156                 int i, j;
 157
 158                 WARN_ON(!RB_EMPTY_NODE(&arg_map->node));
 159
 160                 for (i = 0; i < arg_map->size; i++) {
 161                         struct crush_choose_arg *arg = &arg_map->args[i];
 162
 163                         for (j = 0; j < arg->weight_set_size; j++)
 164                                 kfree(arg->weight_set[j].weights);
 165                         kfree(arg->weight_set);
 166                         kfree(arg->ids);
 167                 }
 168                 kfree(arg_map->args);
 169                 kfree(arg_map);
 170         }
 171 }
 172
 173 DEFINE_RB_FUNCS(choose_arg_map, struct crush_choose_arg_map, choose_args_index,
 174                 node);
 175
 176 void clear_choose_args(struct crush_map *c)
 177 {
 178         while (!RB_EMPTY_ROOT(&c->choose_args)) {
 179                 struct crush_choose_arg_map *arg_map =
 180                     rb_entry(rb_first(&c->choose_args),
 181                              struct crush_choose_arg_map, node);
 182
 183                 erase_choose_arg_map(&c->choose_args, arg_map);
 184                 free_choose_arg_map(arg_map);
 185         }
 186 }
 187
 188 static u32 *decode_array_32_alloc(void **p, void *end, u32 *plen)
 189 {
 190         u32 *a = NULL;
 191         u32 len;
 192         int ret;
 193
 194         ceph_decode_32_safe(p, end, len, e_inval);
 195         if (len) {
 196                 u32 i;
 197
 198                 a = kmalloc_array(len, sizeof(u32), GFP_NOIO);
 199                 if (!a) {
 200                         ret = -ENOMEM;
 201                         goto fail;
 202                 }
 203
 204                 ceph_decode_need(p, end, len * sizeof(u32), e_inval);
 205                 for (i = 0; i < len; i++)
 206                         a[i] = ceph_decode_32(p);
 207         }
 208
 209         *plen = len;
 210         return a;
 211
 212 e_inval:
 213         ret = -EINVAL;
 214 fail:
 215         kfree(a);
 216         return ERR_PTR(ret);
 217 }
 218
 219 /*
 220  * Assumes @arg is zero-initialized.
 221  */
 222 static int decode_choose_arg(void **p, void *end, struct crush_choose_arg *arg)
 223 {
 224         int ret;
 225
 226         ceph_decode_32_safe(p, end, arg->weight_set_size, e_inval);
 227         if (arg->weight_set_size) {
 228                 u32 i;
 229
 230                 arg->weight_set = kmalloc_array(arg->weight_set_size,
 231                                                 sizeof(*arg->weight_set),
 232                                                 GFP_NOIO);
 233                 if (!arg->weight_set)
 234                         return -ENOMEM;
 235
 236                 for (i = 0; i < arg->weight_set_size; i++) {
 237                         struct crush_weight_set *w = &arg->weight_set[i];
 238
 239                         w->weights = decode_array_32_alloc(p, end, &w->size);
 240                         if (IS_ERR(w->weights)) {
 241                                 ret = PTR_ERR(w->weights);
 242                                 w->weights = NULL;
 243                                 return ret;
 244                         }
 245                 }
 246         }
 247
 248         arg->ids = decode_array_32_alloc(p, end, &arg->ids_size);
 249         if (IS_ERR(arg->ids)) {
 250                 ret = PTR_ERR(arg->ids);
 251                 arg->ids = NULL;
 252                 return ret;
 253         }
 254
 255         return 0;
 256
 257 e_inval:
 258         return -EINVAL;
 259 }
 260
 261 static int decode_choose_args(void **p, void *end, struct crush_map *c)
 262 {
 263         struct crush_choose_arg_map *arg_map = NULL;
 264         u32 num_choose_arg_maps, num_buckets;
 265         int ret;
 266
 267         ceph_decode_32_safe(p, end, num_choose_arg_maps, e_inval);
 268         while (num_choose_arg_maps--) {
 269                 arg_map = alloc_choose_arg_map();
 270                 if (!arg_map) {
 271                         ret = -ENOMEM;
 272                         goto fail;
 273                 }
 274
 275                 ceph_decode_64_safe(p, end, arg_map->choose_args_index,
 276                                     e_inval);
 277                 arg_map->size = c->max_buckets;
 278                 arg_map->args = kcalloc(arg_map->size, sizeof(*arg_map->args),
 279                                         GFP_NOIO);
 280                 if (!arg_map->args) {
 281                         ret = -ENOMEM;
 282                         goto fail;
 283                 }
 284
 285                 ceph_decode_32_safe(p, end, num_buckets, e_inval);
 286                 while (num_buckets--) {
 287                         struct crush_choose_arg *arg;
 288                         u32 bucket_index;
 289
 290                         ceph_decode_32_safe(p, end, bucket_index, e_inval);
 291                         if (bucket_index >= arg_map->size)
 292                                 goto e_inval;
 293
 294                         arg = &arg_map->args[bucket_index];
 295                         ret = decode_choose_arg(p, end, arg);
 296                         if (ret)
 297                                 goto fail;
 298
 299                         if (arg->ids_size &&
 300                             arg->ids_size != c->buckets[bucket_index]->size)
 301                                 goto e_inval;
 302                 }
 303
 304                 insert_choose_arg_map(&c->choose_args, arg_map);
 305         }
 306
 307         return 0;
 308
 309 e_inval:
 310         ret = -EINVAL;
 311 fail:
 312         free_choose_arg_map(arg_map);
 313         return ret;
 314 }
 315
 316 static void crush_finalize(struct crush_map *c)
 317 {
 318         __s32 b;
 319
 320         /* Space for the array of pointers to per-bucket workspace */
 321         c->working_size = sizeof(struct crush_work) +
 322             c->max_buckets * sizeof(struct crush_work_bucket *);
 323
 324         for (b = 0; b < c->max_buckets; b++) {
 325                 if (!c->buckets[b])
 326                         continue;
 327
 328                 switch (c->buckets[b]->alg) {
 329                 default:
 330                         /*
 331                          * The base case, permutation variables and
 332                          * the pointer to the permutation array.
 333                          */
 334                         c->working_size += sizeof(struct crush_work_bucket);
 335                         break;
 336                 }
 337                 /* Every bucket has a permutation array. */
 338                 c->working_size += c->buckets[b]->size * sizeof(__u32);
 339         }
 340 }
 341
 342 static struct crush_map *crush_decode(void *pbyval, void *end)
 343 {
 344         struct crush_map *c;
 345         int err;
 346         int i, j;
 347         void **p = &pbyval;
 348         void *start = pbyval;
 349         u32 magic;
 350
 351         dout("crush_decode %p to %p len %d\n", *p, end, (int)(end - *p));
 352
 353         c = kzalloc(sizeof(*c), GFP_NOFS);
 354         if (c == NULL)
 355                 return ERR_PTR(-ENOMEM);
 356
 357         c->choose_args = RB_ROOT;
 358
 359         /* set tunables to default values */
 360         c->choose_local_tries = 2;
 361         c->choose_local_fallback_tries = 5;
 362         c->choose_total_tries = 19;
 363         c->chooseleaf_descend_once = 0;
 364
 365         ceph_decode_need(p, end, 4*sizeof(u32), bad);
 366         magic = ceph_decode_32(p);
 367         if (magic != CRUSH_MAGIC) {
 368                 pr_err("crush_decode magic %x != current %x\n",
 369                        (unsigned int)magic, (unsigned int)CRUSH_MAGIC);
 370                 goto bad;
 371         }
 372         c->max_buckets = ceph_decode_32(p);
 373         c->max_rules = ceph_decode_32(p);
 374         c->max_devices = ceph_decode_32(p);
 375
 376         c->buckets = kcalloc(c->max_buckets, sizeof(*c->buckets), GFP_NOFS);
 377         if (c->buckets == NULL)
 378                 goto badmem;
 379         c->rules = kcalloc(c->max_rules, sizeof(*c->rules), GFP_NOFS);
 380         if (c->rules == NULL)
 381                 goto badmem;
 382
 383         /* buckets */
 384         for (i = 0; i < c->max_buckets; i++) {
 385                 int size = 0;
 386                 u32 alg;
 387                 struct crush_bucket *b;
 388
 389                 ceph_decode_32_safe(p, end, alg, bad);
 390                 if (alg == 0) {
 391                         c->buckets[i] = NULL;
 392                         continue;
 393                 }
 394                 dout("crush_decode bucket %d off %x %p to %p\n",
 395                      i, (int)(*p-start), *p, end);
 396
 397                 switch (alg) {
 398                 case CRUSH_BUCKET_UNIFORM:
 399                         size = sizeof(struct crush_bucket_uniform);
 400                         break;
 401                 case CRUSH_BUCKET_LIST:
 402                         size = sizeof(struct crush_bucket_list);
 403                         break;
 404                 case CRUSH_BUCKET_TREE:
 405                         size = sizeof(struct crush_bucket_tree);
 406                         break;
 407                 case CRUSH_BUCKET_STRAW:
 408                         size = sizeof(struct crush_bucket_straw);
 409                         break;
 410                 case CRUSH_BUCKET_STRAW2:
 411                         size = sizeof(struct crush_bucket_straw2);
 412                         break;
 413                 default:
 414                         goto bad;
 415                 }
 416                 BUG_ON(size == 0);
 417                 b = c->buckets[i] = kzalloc(size, GFP_NOFS);
 418                 if (b == NULL)
 419                         goto badmem;
 420
 421                 ceph_decode_need(p, end, 4*sizeof(u32), bad);
 422                 b->id = ceph_decode_32(p);
 423                 b->type = ceph_decode_16(p);
 424                 b->alg = ceph_decode_8(p);
 425                 b->hash = ceph_decode_8(p);
 426                 b->weight = ceph_decode_32(p);
 427                 b->size = ceph_decode_32(p);
 428
 429                 dout("crush_decode bucket size %d off %x %p to %p\n",
 430                      b->size, (int)(*p-start), *p, end);
 431
 432                 b->items = kcalloc(b->size, sizeof(__s32), GFP_NOFS);
 433                 if (b->items == NULL)
 434                         goto badmem;
 435
 436                 ceph_decode_need(p, end, b->size*sizeof(u32), bad);
 437                 for (j = 0; j < b->size; j++)
 438                         b->items[j] = ceph_decode_32(p);
 439
 440                 switch (b->alg) {
 441                 case CRUSH_BUCKET_UNIFORM:
 442                         err = crush_decode_uniform_bucket(p, end,
 443                                   (struct crush_bucket_uniform *)b);
 444                         if (err < 0)
 445                                 goto fail;
 446                         break;
 447                 case CRUSH_BUCKET_LIST:
 448                         err = crush_decode_list_bucket(p, end,
 449                                (struct crush_bucket_list *)b);
 450                         if (err < 0)
 451                                 goto fail;
 452                         break;
 453                 case CRUSH_BUCKET_TREE:
 454                         err = crush_decode_tree_bucket(p, end,
 455                                 (struct crush_bucket_tree *)b);
 456                         if (err < 0)
 457                                 goto fail;
 458                         break;
 459                 case CRUSH_BUCKET_STRAW:
 460                         err = crush_decode_straw_bucket(p, end,
 461                                 (struct crush_bucket_straw *)b);
 462                         if (err < 0)
 463                                 goto fail;
 464                         break;
 465                 case CRUSH_BUCKET_STRAW2:
 466                         err = crush_decode_straw2_bucket(p, end,
 467                                 (struct crush_bucket_straw2 *)b);
 468                         if (err < 0)
 469                                 goto fail;
 470                         break;
 471                 }
 472         }
 473
 474         /* rules */
 475         dout("rule vec is %p\n", c->rules);
 476         for (i = 0; i < c->max_rules; i++) {
 477                 u32 yes;
 478                 struct crush_rule *r;
 479
 480                 ceph_decode_32_safe(p, end, yes, bad);
 481                 if (!yes) {
 482                         dout("crush_decode NO rule %d off %x %p to %p\n",
 483                              i, (int)(*p-start), *p, end);
 484                         c->rules[i] = NULL;
 485                         continue;
 486                 }
 487
 488                 dout("crush_decode rule %d off %x %p to %p\n",
 489                      i, (int)(*p-start), *p, end);
 490
 491                 /* len */
 492                 ceph_decode_32_safe(p, end, yes, bad);
 493 #if BITS_PER_LONG == 32
 494                 if (yes > (ULONG_MAX - sizeof(*r))
 495                           / sizeof(struct crush_rule_step))
 496                         goto bad;
 497 #endif
 498                 r = kmalloc(struct_size(r, steps, yes), GFP_NOFS);
 499                 c->rules[i] = r;
 500                 if (r == NULL)
 501                         goto badmem;
 502                 dout(" rule %d is at %p\n", i, r);
 503                 r->len = yes;
 504                 ceph_decode_copy_safe(p, end, &r->mask, 4, bad); /* 4 u8's */
 505                 ceph_decode_need(p, end, r->len*3*sizeof(u32), bad);
 506                 for (j = 0; j < r->len; j++) {
 507                         r->steps[j].op = ceph_decode_32(p);
 508                         r->steps[j].arg1 = ceph_decode_32(p);
 509                         r->steps[j].arg2 = ceph_decode_32(p);
 510                 }
 511         }
 512
 513         ceph_decode_skip_map(p, end, 32, string, bad); /* type_map */
 514         ceph_decode_skip_map(p, end, 32, string, bad); /* name_map */
 515         ceph_decode_skip_map(p, end, 32, string, bad); /* rule_name_map */
 516
 517         /* tunables */
 518         ceph_decode_need(p, end, 3*sizeof(u32), done);
 519         c->choose_local_tries = ceph_decode_32(p);
 520         c->choose_local_fallback_tries =  ceph_decode_32(p);
 521         c->choose_total_tries = ceph_decode_32(p);
 522         dout("crush decode tunable choose_local_tries = %d\n",
 523              c->choose_local_tries);
 524         dout("crush decode tunable choose_local_fallback_tries = %d\n",
 525              c->choose_local_fallback_tries);
 526         dout("crush decode tunable choose_total_tries = %d\n",
 527              c->choose_total_tries);
 528
 529         ceph_decode_need(p, end, sizeof(u32), done);
 530         c->chooseleaf_descend_once = ceph_decode_32(p);
 531         dout("crush decode tunable chooseleaf_descend_once = %d\n",
 532              c->chooseleaf_descend_once);
 533
 534         ceph_decode_need(p, end, sizeof(u8), done);
 535         c->chooseleaf_vary_r = ceph_decode_8(p);
 536         dout("crush decode tunable chooseleaf_vary_r = %d\n",
 537              c->chooseleaf_vary_r);
 538
 539         /* skip straw_calc_version, allowed_bucket_algs */
 540         ceph_decode_need(p, end, sizeof(u8) + sizeof(u32), done);
 541         *p += sizeof(u8) + sizeof(u32);
 542
 543         ceph_decode_need(p, end, sizeof(u8), done);
 544         c->chooseleaf_stable = ceph_decode_8(p);
 545         dout("crush decode tunable chooseleaf_stable = %d\n",
 546              c->chooseleaf_stable);
 547
 548         if (*p != end) {
 549                 /* class_map */
 550                 ceph_decode_skip_map(p, end, 32, 32, bad);
 551                 /* class_name */
 552                 ceph_decode_skip_map(p, end, 32, string, bad);
 553                 /* class_bucket */
 554                 ceph_decode_skip_map_of_map(p, end, 32, 32, 32, bad);
 555         }
 556
 557         if (*p != end) {
 558                 err = decode_choose_args(p, end, c);
 559                 if (err)
 560                         goto fail;
 561         }
 562
 563 done:
 564         crush_finalize(c);
 565         dout("crush_decode success\n");
 566         return c;
 567
 568 badmem:
 569         err = -ENOMEM;
 570 fail:
 571         dout("crush_decode fail %d\n", err);
 572         crush_destroy(c);
 573         return ERR_PTR(err);
 574
 575 bad:
 576         err = -EINVAL;
 577         goto fail;
 578 }
 579
 580 int ceph_pg_compare(const struct ceph_pg *lhs, const struct ceph_pg *rhs)
 581 {
 582         if (lhs->pool < rhs->pool)
 583                 return -1;
 584         if (lhs->pool > rhs->pool)
 585                 return 1;
 586         if (lhs->seed < rhs->seed)
 587                 return -1;
 588         if (lhs->seed > rhs->seed)
 589                 return 1;
 590
 591         return 0;
 592 }
 593
 594 int ceph_spg_compare(const struct ceph_spg *lhs, const struct ceph_spg *rhs)
 595 {
 596         int ret;
 597
 598         ret = ceph_pg_compare(&lhs->pgid, &rhs->pgid);
 599         if (ret)
 600                 return ret;
 601
 602         if (lhs->shard < rhs->shard)
 603                 return -1;
 604         if (lhs->shard > rhs->shard)
 605                 return 1;
 606
 607         return 0;
 608 }
 609
 610 static struct ceph_pg_mapping *alloc_pg_mapping(size_t payload_len)
 611 {
 612         struct ceph_pg_mapping *pg;
 613
 614         pg = kmalloc(sizeof(*pg) + payload_len, GFP_NOIO);
 615         if (!pg)
 616                 return NULL;
 617
 618         RB_CLEAR_NODE(&pg->node);
 619         return pg;
 620 }
 621
 622 static void free_pg_mapping(struct ceph_pg_mapping *pg)
 623 {
 624         WARN_ON(!RB_EMPTY_NODE(&pg->node));
 625
 626         kfree(pg);
 627 }
 628
 629 /*
 630  * rbtree of pg_mapping for handling pg_temp (explicit mapping of pgid
 631  * to a set of osds) and primary_temp (explicit primary setting)
 632  */
 633 DEFINE_RB_FUNCS2(pg_mapping, struct ceph_pg_mapping, pgid, ceph_pg_compare,
 634                  RB_BYPTR, const struct ceph_pg *, node)
 635
 636 /*
 637  * rbtree of pg pool info
 638  */
 639 static int __insert_pg_pool(struct rb_root *root, struct ceph_pg_pool_info *new)
 640 {
 641         struct rb_node **p = &root->rb_node;
 642         struct rb_node *parent = NULL;
 643         struct ceph_pg_pool_info *pi = NULL;
 644
 645         while (*p) {
 646                 parent = *p;
 647                 pi = rb_entry(parent, struct ceph_pg_pool_info, node);
 648                 if (new->id < pi->id)
 649                         p = &(*p)->rb_left;
 650                 else if (new->id > pi->id)
 651                         p = &(*p)->rb_right;
 652                 else
 653                         return -EEXIST;
 654         }
 655
 656         rb_link_node(&new->node, parent, p);
 657         rb_insert_color(&new->node, root);
 658         return 0;
 659 }
 660
 661 static struct ceph_pg_pool_info *__lookup_pg_pool(struct rb_root *root, u64 id)
 662 {
 663         struct ceph_pg_pool_info *pi;
 664         struct rb_node *n = root->rb_node;
 665
 666         while (n) {
 667                 pi = rb_entry(n, struct ceph_pg_pool_info, node);
 668                 if (id < pi->id)
 669                         n = n->rb_left;
 670                 else if (id > pi->id)
 671                         n = n->rb_right;
 672                 else
 673                         return pi;
 674         }
 675         return NULL;
 676 }
 677
 678 struct ceph_pg_pool_info *ceph_pg_pool_by_id(struct ceph_osdmap *map, u64 id)
 679 {
 680         return __lookup_pg_pool(&map->pg_pools, id);
 681 }
 682
 683 const char *ceph_pg_pool_name_by_id(struct ceph_osdmap *map, u64 id)
 684 {
 685         struct ceph_pg_pool_info *pi;
 686
 687         if (id == CEPH_NOPOOL)
 688                 return NULL;
 689
 690         if (WARN_ON_ONCE(id > (u64) INT_MAX))
 691                 return NULL;
 692
 693         pi = __lookup_pg_pool(&map->pg_pools, (int) id);
 694
 695         return pi ? pi->name : NULL;
 696 }
 697 EXPORT_SYMBOL(ceph_pg_pool_name_by_id);
 698
 699 int ceph_pg_poolid_by_name(struct ceph_osdmap *map, const char *name)
 700 {
 701         struct rb_node *rbp;
 702
 703         for (rbp = rb_first(&map->pg_pools); rbp; rbp = rb_next(rbp)) {
 704                 struct ceph_pg_pool_info *pi =
 705                         rb_entry(rbp, struct ceph_pg_pool_info, node);
 706                 if (pi->name && strcmp(pi->name, name) == 0)
 707                         return pi->id;
 708         }
 709         return -ENOENT;
 710 }
 711 EXPORT_SYMBOL(ceph_pg_poolid_by_name);
 712
 713 static void __remove_pg_pool(struct rb_root *root, struct ceph_pg_pool_info *pi)
 714 {
 715         rb_erase(&pi->node, root);
 716         kfree(pi->name);
 717         kfree(pi);
 718 }
 719
 720 static int decode_pool(void **p, void *end, struct ceph_pg_pool_info *pi)
 721 {
 722         u8 ev, cv;
 723         unsigned len, num;
 724         void *pool_end;
 725
 726         ceph_decode_need(p, end, 2 + 4, bad);
 727         ev = ceph_decode_8(p);  /* encoding version */
 728         cv = ceph_decode_8(p); /* compat version */
 729         if (ev < 5) {
 730                 pr_warn("got v %d < 5 cv %d of ceph_pg_pool\n", ev, cv);
 731                 return -EINVAL;
 732         }
 733         if (cv > 9) {
 734                 pr_warn("got v %d cv %d > 9 of ceph_pg_pool\n", ev, cv);
 735                 return -EINVAL;
 736         }
 737         len = ceph_decode_32(p);
 738         ceph_decode_need(p, end, len, bad);
 739         pool_end = *p + len;
 740
 741         pi->type = ceph_decode_8(p);
 742         pi->size = ceph_decode_8(p);
 743         pi->crush_ruleset = ceph_decode_8(p);
 744         pi->object_hash = ceph_decode_8(p);
 745
 746         pi->pg_num = ceph_decode_32(p);
 747         pi->pgp_num = ceph_decode_32(p);
 748
 749         *p += 4 + 4;  /* skip lpg* */
 750         *p += 4;      /* skip last_change */
 751         *p += 8 + 4;  /* skip snap_seq, snap_epoch */
 752
 753         /* skip snaps */
 754         num = ceph_decode_32(p);
 755         while (num--) {
 756                 *p += 8;  /* snapid key */
 757                 *p += 1 + 1; /* versions */
 758                 len = ceph_decode_32(p);
 759                 *p += len;
 760         }
 761
 762         /* skip removed_snaps */
 763         num = ceph_decode_32(p);
 764         *p += num * (8 + 8);
 765
 766         *p += 8;  /* skip auid */
 767         pi->flags = ceph_decode_64(p);
 768         *p += 4;  /* skip crash_replay_interval */
 769
 770         if (ev >= 7)
 771                 pi->min_size = ceph_decode_8(p);
 772         else
 773                 pi->min_size = pi->size - pi->size / 2;
 774
 775         if (ev >= 8)
 776                 *p += 8 + 8;  /* skip quota_max_* */
 777
 778         if (ev >= 9) {
 779                 /* skip tiers */
 780                 num = ceph_decode_32(p);
 781                 *p += num * 8;
 782
 783                 *p += 8;  /* skip tier_of */
 784                 *p += 1;  /* skip cache_mode */
 785
 786                 pi->read_tier = ceph_decode_64(p);
 787                 pi->write_tier = ceph_decode_64(p);
 788         } else {
 789                 pi->read_tier = -1;
 790                 pi->write_tier = -1;
 791         }
 792
 793         if (ev >= 10) {
 794                 /* skip properties */
 795                 num = ceph_decode_32(p);
 796                 while (num--) {
 797                         len = ceph_decode_32(p);
 798                         *p += len; /* key */
 799                         len = ceph_decode_32(p);
 800                         *p += len; /* val */
 801                 }
 802         }
 803
 804         if (ev >= 11) {
 805                 /* skip hit_set_params */
 806                 *p += 1 + 1; /* versions */
 807                 len = ceph_decode_32(p);
 808                 *p += len;
 809
 810                 *p += 4; /* skip hit_set_period */
 811                 *p += 4; /* skip hit_set_count */
 812         }
 813
 814         if (ev >= 12)
 815                 *p += 4; /* skip stripe_width */
 816
 817         if (ev >= 13) {
 818                 *p += 8; /* skip target_max_bytes */
 819                 *p += 8; /* skip target_max_objects */
 820                 *p += 4; /* skip cache_target_dirty_ratio_micro */
 821                 *p += 4; /* skip cache_target_full_ratio_micro */
 822                 *p += 4; /* skip cache_min_flush_age */
 823                 *p += 4; /* skip cache_min_evict_age */
 824         }
 825
 826         if (ev >=  14) {
 827                 /* skip erasure_code_profile */
 828                 len = ceph_decode_32(p);
 829                 *p += len;
 830         }
 831
 832         /*
 833          * last_force_op_resend_preluminous, will be overridden if the
 834          * map was encoded with RESEND_ON_SPLIT
 835          */
 836         if (ev >= 15)
 837                 pi->last_force_request_resend = ceph_decode_32(p);
 838         else
 839                 pi->last_force_request_resend = 0;
 840
 841         if (ev >= 16)
 842                 *p += 4; /* skip min_read_recency_for_promote */
 843
 844         if (ev >= 17)
 845                 *p += 8; /* skip expected_num_objects */
 846
 847         if (ev >= 19)
 848                 *p += 4; /* skip cache_target_dirty_high_ratio_micro */
 849
 850         if (ev >= 20)
 851                 *p += 4; /* skip min_write_recency_for_promote */
 852
 853         if (ev >= 21)
 854                 *p += 1; /* skip use_gmt_hitset */
 855
 856         if (ev >= 22)
 857                 *p += 1; /* skip fast_read */
 858
 859         if (ev >= 23) {
 860                 *p += 4; /* skip hit_set_grade_decay_rate */
 861                 *p += 4; /* skip hit_set_search_last_n */
 862         }
 863
 864         if (ev >= 24) {
 865                 /* skip opts */
 866                 *p += 1 + 1; /* versions */
 867                 len = ceph_decode_32(p);
 868                 *p += len;
 869         }
 870
 871         if (ev >= 25)
 872                 pi->last_force_request_resend = ceph_decode_32(p);
 873
 874         /* ignore the rest */
 875
 876         *p = pool_end;
 877         calc_pg_masks(pi);
 878         return 0;
 879
 880 bad:
 881         return -EINVAL;
 882 }
 883
 884 static int decode_pool_names(void **p, void *end, struct ceph_osdmap *map)
 885 {
 886         struct ceph_pg_pool_info *pi;
 887         u32 num, len;
 888         u64 pool;
 889
 890         ceph_decode_32_safe(p, end, num, bad);
 891         dout(" %d pool names\n", num);
 892         while (num--) {
 893                 ceph_decode_64_safe(p, end, pool, bad);
 894                 ceph_decode_32_safe(p, end, len, bad);
 895                 dout("  pool %llu len %d\n", pool, len);
 896                 ceph_decode_need(p, end, len, bad);
 897                 pi = __lookup_pg_pool(&map->pg_pools, pool);
 898                 if (pi) {
 899                         char *name = kstrndup(*p, len, GFP_NOFS);
 900
 901                         if (!name)
 902                                 return -ENOMEM;
 903                         kfree(pi->name);
 904                         pi->name = name;
 905                         dout("  name is %s\n", pi->name);
 906                 }
 907                 *p += len;
 908         }
 909         return 0;
 910
 911 bad:
 912         return -EINVAL;
 913 }
 914
 915 /*
 916  * osd map
 917  */
 918 struct ceph_osdmap *ceph_osdmap_alloc(void)
 919 {
 920         struct ceph_osdmap *map;
 921
 922         map = kzalloc(sizeof(*map), GFP_NOIO);
 923         if (!map)
 924                 return NULL;
 925
 926         map->pg_pools = RB_ROOT;
 927         map->pool_max = -1;
 928         map->pg_temp = RB_ROOT;
 929         map->primary_temp = RB_ROOT;
 930         map->pg_upmap = RB_ROOT;
 931         map->pg_upmap_items = RB_ROOT;
 932         mutex_init(&map->crush_workspace_mutex);
 933
 934         return map;
 935 }
 936
 937 void ceph_osdmap_destroy(struct ceph_osdmap *map)
 938 {
 939         dout("osdmap_destroy %p\n", map);
 940         if (map->crush)
 941                 crush_destroy(map->crush);
 942         while (!RB_EMPTY_ROOT(&map->pg_temp)) {
 943                 struct ceph_pg_mapping *pg =
 944                         rb_entry(rb_first(&map->pg_temp),
 945                                  struct ceph_pg_mapping, node);
 946                 erase_pg_mapping(&map->pg_temp, pg);
 947                 free_pg_mapping(pg);
 948         }
 949         while (!RB_EMPTY_ROOT(&map->primary_temp)) {
 950                 struct ceph_pg_mapping *pg =
 951                         rb_entry(rb_first(&map->primary_temp),
 952                                  struct ceph_pg_mapping, node);
 953                 erase_pg_mapping(&map->primary_temp, pg);
 954                 free_pg_mapping(pg);
 955         }
 956         while (!RB_EMPTY_ROOT(&map->pg_upmap)) {
 957                 struct ceph_pg_mapping *pg =
 958                         rb_entry(rb_first(&map->pg_upmap),
 959                                  struct ceph_pg_mapping, node);
 960                 rb_erase(&pg->node, &map->pg_upmap);
 961                 kfree(pg);
 962         }
 963         while (!RB_EMPTY_ROOT(&map->pg_upmap_items)) {
 964                 struct ceph_pg_mapping *pg =
 965                         rb_entry(rb_first(&map->pg_upmap_items),
 966                                  struct ceph_pg_mapping, node);
 967                 rb_erase(&pg->node, &map->pg_upmap_items);
 968                 kfree(pg);
 969         }
 970         while (!RB_EMPTY_ROOT(&map->pg_pools)) {
 971                 struct ceph_pg_pool_info *pi =
 972                         rb_entry(rb_first(&map->pg_pools),
 973                                  struct ceph_pg_pool_info, node);
 974                 __remove_pg_pool(&map->pg_pools, pi);
 975         }
 976         kfree(map->osd_state);
 977         kfree(map->osd_weight);
 978         kfree(map->osd_addr);
 979         kfree(map->osd_primary_affinity);
 980         kfree(map->crush_workspace);
 981         kfree(map);
 982 }
 983
 984 /*
 985  * Adjust max_osd value, (re)allocate arrays.
 986  *
 987  * The new elements are properly initialized.
 988  */
 989 static int osdmap_set_max_osd(struct ceph_osdmap *map, int max)
 990 {
 991         u32 *state;
 992         u32 *weight;
 993         struct ceph_entity_addr *addr;
 994         int i;
 995
 996         state = krealloc(map->osd_state, max*sizeof(*state), GFP_NOFS);
 997         if (!state)
 998                 return -ENOMEM;
 999         map->osd_state = state;
1000
1001         weight = krealloc(map->osd_weight, max*sizeof(*weight), GFP_NOFS);
1002         if (!weight)
1003                 return -ENOMEM;
1004         map->osd_weight = weight;
1005
1006         addr = krealloc(map->osd_addr, max*sizeof(*addr), GFP_NOFS);
1007         if (!addr)
1008                 return -ENOMEM;
1009         map->osd_addr = addr;
1010
1011         for (i = map->max_osd; i < max; i++) {
1012                 map->osd_state[i] = 0;
1013                 map->osd_weight[i] = CEPH_OSD_OUT;
1014                 memset(map->osd_addr + i, 0, sizeof(*map->osd_addr));
1015         }
1016
1017         if (map->osd_primary_affinity) {
1018                 u32 *affinity;
1019
1020                 affinity = krealloc(map->osd_primary_affinity,
1021                                     max*sizeof(*affinity), GFP_NOFS);
1022                 if (!affinity)
1023                         return -ENOMEM;
1024                 map->osd_primary_affinity = affinity;
1025
1026                 for (i = map->max_osd; i < max; i++)
1027                         map->osd_primary_affinity[i] =
1028                             CEPH_OSD_DEFAULT_PRIMARY_AFFINITY;
1029         }
1030
1031         map->max_osd = max;
1032
1033         return 0;
1034 }
1035
1036 static int osdmap_set_crush(struct ceph_osdmap *map, struct crush_map *crush)
1037 {
1038         void *workspace;
1039         size_t work_size;
1040
1041         if (IS_ERR(crush))
1042                 return PTR_ERR(crush);
1043
1044         work_size = crush_work_size(crush, CEPH_PG_MAX_SIZE);
1045         dout("%s work_size %zu bytes\n", __func__, work_size);
1046         workspace = kmalloc(work_size, GFP_NOIO);
1047         if (!workspace) {
1048                 crush_destroy(crush);
1049                 return -ENOMEM;
1050         }
1051         crush_init_workspace(crush, workspace);
1052
1053         if (map->crush)
1054                 crush_destroy(map->crush);
1055         kfree(map->crush_workspace);
1056         map->crush = crush;
1057         map->crush_workspace = workspace;
1058         return 0;
1059 }
1060
1061 #define OSDMAP_WRAPPER_COMPAT_VER       7
1062 #define OSDMAP_CLIENT_DATA_COMPAT_VER   1
1063
1064 /*
1065  * Return 0 or error.  On success, *v is set to 0 for old (v6) osdmaps,
1066  * to struct_v of the client_data section for new (v7 and above)
1067  * osdmaps.
1068  */
1069 static int get_osdmap_client_data_v(void **p, void *end,
1070                                     const char *prefix, u8 *v)
1071 {
1072         u8 struct_v;
1073
1074         ceph_decode_8_safe(p, end, struct_v, e_inval);
1075         if (struct_v >= 7) {
1076                 u8 struct_compat;
1077
1078                 ceph_decode_8_safe(p, end, struct_compat, e_inval);
1079                 if (struct_compat > OSDMAP_WRAPPER_COMPAT_VER) {
1080                         pr_warn("got v %d cv %d > %d of %s ceph_osdmap\n",
1081                                 struct_v, struct_compat,
1082                                 OSDMAP_WRAPPER_COMPAT_VER, prefix);
1083                         return -EINVAL;
1084                 }
1085                 *p += 4; /* ignore wrapper struct_len */
1086
1087                 ceph_decode_8_safe(p, end, struct_v, e_inval);
1088                 ceph_decode_8_safe(p, end, struct_compat, e_inval);
1089                 if (struct_compat > OSDMAP_CLIENT_DATA_COMPAT_VER) {
1090                         pr_warn("got v %d cv %d > %d of %s ceph_osdmap client data\n",
1091                                 struct_v, struct_compat,
1092                                 OSDMAP_CLIENT_DATA_COMPAT_VER, prefix);
1093                         return -EINVAL;
1094                 }
1095                 *p += 4; /* ignore client data struct_len */
1096         } else {
1097                 u16 version;
1098
1099                 *p -= 1;
1100                 ceph_decode_16_safe(p, end, version, e_inval);
1101                 if (version < 6) {
1102                         pr_warn("got v %d < 6 of %s ceph_osdmap\n",
1103                                 version, prefix);
1104                         return -EINVAL;
1105                 }
1106
1107                 /* old osdmap enconding */
1108                 struct_v = 0;
1109         }
1110
1111         *v = struct_v;
1112         return 0;
1113
1114 e_inval:
1115         return -EINVAL;
1116 }
1117
1118 static int __decode_pools(void **p, void *end, struct ceph_osdmap *map,
1119                           bool incremental)
1120 {
1121         u32 n;
1122
1123         ceph_decode_32_safe(p, end, n, e_inval);
1124         while (n--) {
1125                 struct ceph_pg_pool_info *pi;
1126                 u64 pool;
1127                 int ret;
1128
1129                 ceph_decode_64_safe(p, end, pool, e_inval);
1130
1131                 pi = __lookup_pg_pool(&map->pg_pools, pool);
1132                 if (!incremental || !pi) {
1133                         pi = kzalloc(sizeof(*pi), GFP_NOFS);
1134                         if (!pi)
1135                                 return -ENOMEM;
1136
1137                         pi->id = pool;
1138
1139                         ret = __insert_pg_pool(&map->pg_pools, pi);
1140                         if (ret) {
1141                                 kfree(pi);
1142                                 return ret;
1143                         }
1144                 }
1145
1146                 ret = decode_pool(p, end, pi);
1147                 if (ret)
1148                         return ret;
1149         }
1150
1151         return 0;
1152
1153 e_inval:
1154         return -EINVAL;
1155 }
1156
1157 static int decode_pools(void **p, void *end, struct ceph_osdmap *map)
1158 {
1159         return __decode_pools(p, end, map, false);
1160 }
1161
1162 static int decode_new_pools(void **p, void *end, struct ceph_osdmap *map)
1163 {
1164         return __decode_pools(p, end, map, true);
1165 }
1166
1167 typedef struct ceph_pg_mapping *(*decode_mapping_fn_t)(void **, void *, bool);
1168
1169 static int decode_pg_mapping(void **p, void *end, struct rb_root *mapping_root,
1170                              decode_mapping_fn_t fn, bool incremental)
1171 {
1172         u32 n;
1173
1174         WARN_ON(!incremental && !fn);
1175
1176         ceph_decode_32_safe(p, end, n, e_inval);
1177         while (n--) {
1178                 struct ceph_pg_mapping *pg;
1179                 struct ceph_pg pgid;
1180                 int ret;
1181
1182                 ret = ceph_decode_pgid(p, end, &pgid);
1183                 if (ret)
1184                         return ret;
1185
1186                 pg = lookup_pg_mapping(mapping_root, &pgid);
1187                 if (pg) {
1188                         WARN_ON(!incremental);
1189                         erase_pg_mapping(mapping_root, pg);
1190                         free_pg_mapping(pg);
1191                 }
1192
1193                 if (fn) {
1194                         pg = fn(p, end, incremental);
1195                         if (IS_ERR(pg))
1196                                 return PTR_ERR(pg);
1197
1198                         if (pg) {
1199                                 pg->pgid = pgid; /* struct */
1200                                 insert_pg_mapping(mapping_root, pg);
1201                         }
1202                 }
1203         }
1204
1205         return 0;
1206
1207 e_inval:
1208         return -EINVAL;
1209 }
1210
1211 static struct ceph_pg_mapping *__decode_pg_temp(void **p, void *end,
1212                                                 bool incremental)
1213 {
1214         struct ceph_pg_mapping *pg;
1215         u32 len, i;
1216
1217         ceph_decode_32_safe(p, end, len, e_inval);
1218         if (len == 0 && incremental)
1219                 return NULL;    /* new_pg_temp: [] to remove */
1220         if (len > (SIZE_MAX - sizeof(*pg)) / sizeof(u32))
1221                 return ERR_PTR(-EINVAL);
1222
1223         ceph_decode_need(p, end, len * sizeof(u32), e_inval);
1224         pg = alloc_pg_mapping(len * sizeof(u32));
1225         if (!pg)
1226                 return ERR_PTR(-ENOMEM);
1227
1228         pg->pg_temp.len = len;
1229         for (i = 0; i < len; i++)
1230                 pg->pg_temp.osds[i] = ceph_decode_32(p);
1231
1232         return pg;
1233
1234 e_inval:
1235         return ERR_PTR(-EINVAL);
1236 }
1237
1238 static int decode_pg_temp(void **p, void *end, struct ceph_osdmap *map)
1239 {
1240         return decode_pg_mapping(p, end, &map->pg_temp, __decode_pg_temp,
1241                                  false);
1242 }
1243
1244 static int decode_new_pg_temp(void **p, void *end, struct ceph_osdmap *map)
1245 {
1246         return decode_pg_mapping(p, end, &map->pg_temp, __decode_pg_temp,
1247                                  true);
1248 }
1249
1250 static struct ceph_pg_mapping *__decode_primary_temp(void **p, void *end,
1251                                                      bool incremental)
1252 {
1253         struct ceph_pg_mapping *pg;
1254         u32 osd;
1255
1256         ceph_decode_32_safe(p, end, osd, e_inval);
1257         if (osd == (u32)-1 && incremental)
1258                 return NULL;    /* new_primary_temp: -1 to remove */
1259
1260         pg = alloc_pg_mapping(0);
1261         if (!pg)
1262                 return ERR_PTR(-ENOMEM);
1263
1264         pg->primary_temp.osd = osd;
1265         return pg;
1266
1267 e_inval:
1268         return ERR_PTR(-EINVAL);
1269 }
1270
1271 static int decode_primary_temp(void **p, void *end, struct ceph_osdmap *map)
1272 {
1273         return decode_pg_mapping(p, end, &map->primary_temp,
1274                                  __decode_primary_temp, false);
1275 }
1276
1277 static int decode_new_primary_temp(void **p, void *end,
1278                                    struct ceph_osdmap *map)
1279 {
1280         return decode_pg_mapping(p, end, &map->primary_temp,
1281                                  __decode_primary_temp, true);
1282 }
1283
1284 u32 ceph_get_primary_affinity(struct ceph_osdmap *map, int osd)
1285 {
1286         BUG_ON(osd >= map->max_osd);
1287
1288         if (!map->osd_primary_affinity)
1289                 return CEPH_OSD_DEFAULT_PRIMARY_AFFINITY;
1290
1291         return map->osd_primary_affinity[osd];
1292 }
1293
1294 static int set_primary_affinity(struct ceph_osdmap *map, int osd, u32 aff)
1295 {
1296         BUG_ON(osd >= map->max_osd);
1297
1298         if (!map->osd_primary_affinity) {
1299                 int i;
1300
1301                 map->osd_primary_affinity = kmalloc_array(map->max_osd,
1302                                                           sizeof(u32),
1303                                                           GFP_NOFS);
1304                 if (!map->osd_primary_affinity)
1305                         return -ENOMEM;
1306
1307                 for (i = 0; i < map->max_osd; i++)
1308                         map->osd_primary_affinity[i] =
1309                             CEPH_OSD_DEFAULT_PRIMARY_AFFINITY;
1310         }
1311
1312         map->osd_primary_affinity[osd] = aff;
1313
1314         return 0;
1315 }
1316
1317 static int decode_primary_affinity(void **p, void *end,
1318                                    struct ceph_osdmap *map)
1319 {
1320         u32 len, i;
1321
1322         ceph_decode_32_safe(p, end, len, e_inval);
1323         if (len == 0) {
1324                 kfree(map->osd_primary_affinity);
1325                 map->osd_primary_affinity = NULL;
1326                 return 0;
1327         }
1328         if (len != map->max_osd)
1329                 goto e_inval;
1330
1331         ceph_decode_need(p, end, map->max_osd*sizeof(u32), e_inval);
1332
1333         for (i = 0; i < map->max_osd; i++) {
1334                 int ret;
1335
1336                 ret = set_primary_affinity(map, i, ceph_decode_32(p));
1337                 if (ret)
1338                         return ret;
1339         }
1340
1341         return 0;
1342
1343 e_inval:
1344         return -EINVAL;
1345 }
1346
1347 static int decode_new_primary_affinity(void **p, void *end,
1348                                        struct ceph_osdmap *map)
1349 {
1350         u32 n;
1351
1352         ceph_decode_32_safe(p, end, n, e_inval);
1353         while (n--) {
1354                 u32 osd, aff;
1355                 int ret;
1356
1357                 ceph_decode_32_safe(p, end, osd, e_inval);
1358                 ceph_decode_32_safe(p, end, aff, e_inval);
1359
1360                 ret = set_primary_affinity(map, osd, aff);
1361                 if (ret)
1362                         return ret;
1363
1364                 pr_info("osd%d primary-affinity 0x%x\n", osd, aff);
1365         }
1366
1367         return 0;
1368
1369 e_inval:
1370         return -EINVAL;
1371 }
1372
1373 static struct ceph_pg_mapping *__decode_pg_upmap(void **p, void *end,
1374                                                  bool __unused)
1375 {
1376         return __decode_pg_temp(p, end, false);
1377 }
1378
1379 static int decode_pg_upmap(void **p, void *end, struct ceph_osdmap *map)
1380 {
1381         return decode_pg_mapping(p, end, &map->pg_upmap, __decode_pg_upmap,
1382                                  false);
1383 }
1384
1385 static int decode_new_pg_upmap(void **p, void *end, struct ceph_osdmap *map)
1386 {
1387         return decode_pg_mapping(p, end, &map->pg_upmap, __decode_pg_upmap,
1388                                  true);
1389 }
1390
1391 static int decode_old_pg_upmap(void **p, void *end, struct ceph_osdmap *map)
1392 {
1393         return decode_pg_mapping(p, end, &map->pg_upmap, NULL, true);
1394 }
1395
1396 static struct ceph_pg_mapping *__decode_pg_upmap_items(void **p, void *end,
1397                                                        bool __unused)
1398 {
1399         struct ceph_pg_mapping *pg;
1400         u32 len, i;
1401
1402         ceph_decode_32_safe(p, end, len, e_inval);
1403         if (len > (SIZE_MAX - sizeof(*pg)) / (2 * sizeof(u32)))
1404                 return ERR_PTR(-EINVAL);
1405
1406         ceph_decode_need(p, end, 2 * len * sizeof(u32), e_inval);
1407         pg = alloc_pg_mapping(2 * len * sizeof(u32));
1408         if (!pg)
1409                 return ERR_PTR(-ENOMEM);
1410
1411         pg->pg_upmap_items.len = len;
1412         for (i = 0; i < len; i++) {
1413                 pg->pg_upmap_items.from_to[i][0] = ceph_decode_32(p);
1414                 pg->pg_upmap_items.from_to[i][1] = ceph_decode_32(p);
1415         }
1416
1417         return pg;
1418
1419 e_inval:
1420         return ERR_PTR(-EINVAL);
1421 }
1422
1423 static int decode_pg_upmap_items(void **p, void *end, struct ceph_osdmap *map)
1424 {
1425         return decode_pg_mapping(p, end, &map->pg_upmap_items,
1426                                  __decode_pg_upmap_items, false);
1427 }
1428
1429 static int decode_new_pg_upmap_items(void **p, void *end,
1430                                      struct ceph_osdmap *map)
1431 {
1432         return decode_pg_mapping(p, end, &map->pg_upmap_items,
1433                                  __decode_pg_upmap_items, true);
1434 }
1435
1436 static int decode_old_pg_upmap_items(void **p, void *end,
1437                                      struct ceph_osdmap *map)
1438 {
1439         return decode_pg_mapping(p, end, &map->pg_upmap_items, NULL, true);
1440 }
1441
1442 /*
1443  * decode a full map.
1444  */
1445 static int osdmap_decode(void **p, void *end, struct ceph_osdmap *map)
1446 {
1447         u8 struct_v;
1448         u32 epoch = 0;
1449         void *start = *p;
1450         u32 max;
1451         u32 len, i;
1452         int err;
1453
1454         dout("%s %p to %p len %d\n", __func__, *p, end, (int)(end - *p));
1455
1456         err = get_osdmap_client_data_v(p, end, "full", &struct_v);
1457         if (err)
1458                 goto bad;
1459
1460         /* fsid, epoch, created, modified */
1461         ceph_decode_need(p, end, sizeof(map->fsid) + sizeof(u32) +
1462                          sizeof(map->created) + sizeof(map->modified), e_inval);
1463         ceph_decode_copy(p, &map->fsid, sizeof(map->fsid));
1464         epoch = map->epoch = ceph_decode_32(p);
1465         ceph_decode_copy(p, &map->created, sizeof(map->created));
1466         ceph_decode_copy(p, &map->modified, sizeof(map->modified));
1467
1468         /* pools */
1469         err = decode_pools(p, end, map);
1470         if (err)
1471                 goto bad;
1472
1473         /* pool_name */
1474         err = decode_pool_names(p, end, map);
1475         if (err)
1476                 goto bad;
1477
1478         ceph_decode_32_safe(p, end, map->pool_max, e_inval);
1479
1480         ceph_decode_32_safe(p, end, map->flags, e_inval);
1481
1482         /* max_osd */
1483         ceph_decode_32_safe(p, end, max, e_inval);
1484
1485         /* (re)alloc osd arrays */
1486         err = osdmap_set_max_osd(map, max);
1487         if (err)
1488                 goto bad;
1489
1490         /* osd_state, osd_weight, osd_addrs->client_addr */
1491         ceph_decode_need(p, end, 3*sizeof(u32) +
1492                          map->max_osd*(struct_v >= 5 ? sizeof(u32) :
1493                                                        sizeof(u8)) +
1494                                        sizeof(*map->osd_weight), e_inval);
1495         if (ceph_decode_32(p) != map->max_osd)
1496                 goto e_inval;
1497
1498         if (struct_v >= 5) {
1499                 for (i = 0; i < map->max_osd; i++)
1500                         map->osd_state[i] = ceph_decode_32(p);
1501         } else {
1502                 for (i = 0; i < map->max_osd; i++)
1503                         map->osd_state[i] = ceph_decode_8(p);
1504         }
1505
1506         if (ceph_decode_32(p) != map->max_osd)
1507                 goto e_inval;
1508
1509         for (i = 0; i < map->max_osd; i++)
1510                 map->osd_weight[i] = ceph_decode_32(p);
1511
1512         if (ceph_decode_32(p) != map->max_osd)
1513                 goto e_inval;
1514
1515         for (i = 0; i < map->max_osd; i++) {
1516                 err = ceph_decode_entity_addr(p, end, &map->osd_addr[i]);
1517                 if (err)
1518                         goto bad;
1519         }
1520
1521         /* pg_temp */
1522         err = decode_pg_temp(p, end, map);
1523         if (err)
1524                 goto bad;
1525
1526         /* primary_temp */
1527         if (struct_v >= 1) {
1528                 err = decode_primary_temp(p, end, map);
1529                 if (err)
1530                         goto bad;
1531         }
1532
1533         /* primary_affinity */
1534         if (struct_v >= 2) {
1535                 err = decode_primary_affinity(p, end, map);
1536                 if (err)
1537                         goto bad;
1538         } else {
1539                 WARN_ON(map->osd_primary_affinity);
1540         }
1541
1542         /* crush */
1543         ceph_decode_32_safe(p, end, len, e_inval);
1544         err = osdmap_set_crush(map, crush_decode(*p, min(*p + len, end)));
1545         if (err)
1546                 goto bad;
1547
1548         *p += len;
1549         if (struct_v >= 3) {
1550                 /* erasure_code_profiles */
1551                 ceph_decode_skip_map_of_map(p, end, string, string, string,
1552                                             e_inval);
1553         }
1554
1555         if (struct_v >= 4) {
1556                 err = decode_pg_upmap(p, end, map);
1557                 if (err)
1558                         goto bad;
1559
1560                 err = decode_pg_upmap_items(p, end, map);
1561                 if (err)
1562                         goto bad;
1563         } else {
1564                 WARN_ON(!RB_EMPTY_ROOT(&map->pg_upmap));
1565                 WARN_ON(!RB_EMPTY_ROOT(&map->pg_upmap_items));
1566         }
1567
1568         /* ignore the rest */
1569         *p = end;
1570
1571         dout("full osdmap epoch %d max_osd %d\n", map->epoch, map->max_osd);
1572         return 0;
1573
1574 e_inval:
1575         err = -EINVAL;
1576 bad:
1577         pr_err("corrupt full osdmap (%d) epoch %d off %d (%p of %p-%p)\n",
1578                err, epoch, (int)(*p - start), *p, start, end);
1579         print_hex_dump(KERN_DEBUG, "osdmap: ",
1580                        DUMP_PREFIX_OFFSET, 16, 1,
1581                        start, end - start, true);
1582         return err;
1583 }
1584
1585 /*
1586  * Allocate and decode a full map.
1587  */
1588 struct ceph_osdmap *ceph_osdmap_decode(void **p, void *end)
1589 {
1590         struct ceph_osdmap *map;
1591         int ret;
1592
1593         map = ceph_osdmap_alloc();
1594         if (!map)
1595                 return ERR_PTR(-ENOMEM);
1596
1597         ret = osdmap_decode(p, end, map);
1598         if (ret) {
1599                 ceph_osdmap_destroy(map);
1600                 return ERR_PTR(ret);
1601         }
1602
1603         return map;
1604 }
1605
1606 /*
1607  * Encoding order is (new_up_client, new_state, new_weight).  Need to
1608  * apply in the (new_weight, new_state, new_up_client) order, because
1609  * an incremental map may look like e.g.
1610  *
1611  *     new_up_client: { osd=6, addr=... } # set osd_state and addr
1612  *     new_state: { osd=6, xorstate=EXISTS } # clear osd_state
1613  */
1614 static int decode_new_up_state_weight(void **p, void *end, u8 struct_v,
1615                                       struct ceph_osdmap *map)
1616 {
1617         void *new_up_client;
1618         void *new_state;
1619         void *new_weight_end;
1620         u32 len;
1621         int i;
1622
1623         new_up_client = *p;
1624         ceph_decode_32_safe(p, end, len, e_inval);
1625         for (i = 0; i < len; ++i) {
1626                 struct ceph_entity_addr addr;
1627
1628                 ceph_decode_skip_32(p, end, e_inval);
1629                 if (ceph_decode_entity_addr(p, end, &addr))
1630                         goto e_inval;
1631         }
1632
1633         new_state = *p;
1634         ceph_decode_32_safe(p, end, len, e_inval);
1635         len *= sizeof(u32) + (struct_v >= 5 ? sizeof(u32) : sizeof(u8));
1636         ceph_decode_need(p, end, len, e_inval);
1637         *p += len;
1638
1639         /* new_weight */
1640         ceph_decode_32_safe(p, end, len, e_inval);
1641         while (len--) {
1642                 s32 osd;
1643                 u32 w;
1644
1645                 ceph_decode_need(p, end, 2*sizeof(u32), e_inval);
1646                 osd = ceph_decode_32(p);
1647                 w = ceph_decode_32(p);
1648                 BUG_ON(osd >= map->max_osd);
1649                 pr_info("osd%d weight 0x%x %s\n", osd, w,
1650                      w == CEPH_OSD_IN ? "(in)" :
1651                      (w == CEPH_OSD_OUT ? "(out)" : ""));
1652                 map->osd_weight[osd] = w;
1653
1654                 /*
1655                  * If we are marking in, set the EXISTS, and clear the
1656                  * AUTOOUT and NEW bits.
1657                  */
1658                 if (w) {
1659                         map->osd_state[osd] |= CEPH_OSD_EXISTS;
1660                         map->osd_state[osd] &= ~(CEPH_OSD_AUTOOUT |
1661                                                  CEPH_OSD_NEW);
1662                 }
1663         }
1664         new_weight_end = *p;
1665
1666         /* new_state (up/down) */
1667         *p = new_state;
1668         len = ceph_decode_32(p);
1669         while (len--) {
1670                 s32 osd;
1671                 u32 xorstate;
1672                 int ret;
1673
1674                 osd = ceph_decode_32(p);
1675                 if (struct_v >= 5)
1676                         xorstate = ceph_decode_32(p);
1677                 else
1678                         xorstate = ceph_decode_8(p);
1679                 if (xorstate == 0)
1680                         xorstate = CEPH_OSD_UP;
1681                 BUG_ON(osd >= map->max_osd);
1682                 if ((map->osd_state[osd] & CEPH_OSD_UP) &&
1683                     (xorstate & CEPH_OSD_UP))
1684                         pr_info("osd%d down\n", osd);
1685                 if ((map->osd_state[osd] & CEPH_OSD_EXISTS) &&
1686                     (xorstate & CEPH_OSD_EXISTS)) {
1687                         pr_info("osd%d does not exist\n", osd);
1688                         ret = set_primary_affinity(map, osd,
1689                                                    CEPH_OSD_DEFAULT_PRIMARY_AFFINITY);
1690                         if (ret)
1691                                 return ret;
1692                         memset(map->osd_addr + osd, 0, sizeof(*map->osd_addr));
1693                         map->osd_state[osd] = 0;
1694                 } else {
1695                         map->osd_state[osd] ^= xorstate;
1696                 }
1697         }
1698
1699         /* new_up_client */
1700         *p = new_up_client;
1701         len = ceph_decode_32(p);
1702         while (len--) {
1703                 s32 osd;
1704                 struct ceph_entity_addr addr;
1705
1706                 osd = ceph_decode_32(p);
1707                 BUG_ON(osd >= map->max_osd);
1708                 if (ceph_decode_entity_addr(p, end, &addr))
1709                         goto e_inval;
1710                 pr_info("osd%d up\n", osd);
1711                 map->osd_state[osd] |= CEPH_OSD_EXISTS | CEPH_OSD_UP;
1712                 map->osd_addr[osd] = addr;
1713         }
1714
1715         *p = new_weight_end;
1716         return 0;
1717
1718 e_inval:
1719         return -EINVAL;
1720 }
1721
1722 /*
1723  * decode and apply an incremental map update.
1724  */
1725 struct ceph_osdmap *osdmap_apply_incremental(void **p, void *end,
1726                                              struct ceph_osdmap *map)
1727 {
1728         struct ceph_fsid fsid;
1729         u32 epoch = 0;
1730         struct ceph_timespec modified;
1731         s32 len;
1732         u64 pool;
1733         __s64 new_pool_max;
1734         __s32 new_flags, max;
1735         void *start = *p;
1736         int err;
1737         u8 struct_v;
1738
1739         dout("%s %p to %p len %d\n", __func__, *p, end, (int)(end - *p));
1740
1741         err = get_osdmap_client_data_v(p, end, "inc", &struct_v);
1742         if (err)
1743                 goto bad;
1744
1745         /* fsid, epoch, modified, new_pool_max, new_flags */
1746         ceph_decode_need(p, end, sizeof(fsid) + sizeof(u32) + sizeof(modified) +
1747                          sizeof(u64) + sizeof(u32), e_inval);
1748         ceph_decode_copy(p, &fsid, sizeof(fsid));
1749         epoch = ceph_decode_32(p);
1750         BUG_ON(epoch != map->epoch+1);
1751         ceph_decode_copy(p, &modified, sizeof(modified));
1752         new_pool_max = ceph_decode_64(p);
1753         new_flags = ceph_decode_32(p);
1754
1755         /* full map? */
1756         ceph_decode_32_safe(p, end, len, e_inval);
1757         if (len > 0) {
1758                 dout("apply_incremental full map len %d, %p to %p\n",
1759                      len, *p, end);
1760                 return ceph_osdmap_decode(p, min(*p+len, end));
1761         }
1762
1763         /* new crush? */
1764         ceph_decode_32_safe(p, end, len, e_inval);
1765         if (len > 0) {
1766                 err = osdmap_set_crush(map,
1767                                        crush_decode(*p, min(*p + len, end)));
1768                 if (err)
1769                         goto bad;
1770                 *p += len;
1771         }
1772
1773         /* new flags? */
1774         if (new_flags >= 0)
1775                 map->flags = new_flags;
1776         if (new_pool_max >= 0)
1777                 map->pool_max = new_pool_max;
1778
1779         /* new max? */
1780         ceph_decode_32_safe(p, end, max, e_inval);
1781         if (max >= 0) {
1782                 err = osdmap_set_max_osd(map, max);
1783                 if (err)
1784                         goto bad;
1785         }
1786
1787         map->epoch++;
1788         map->modified = modified;
1789
1790         /* new_pools */
1791         err = decode_new_pools(p, end, map);
1792         if (err)
1793                 goto bad;
1794
1795         /* new_pool_names */
1796         err = decode_pool_names(p, end, map);
1797         if (err)
1798                 goto bad;
1799
1800         /* old_pool */
1801         ceph_decode_32_safe(p, end, len, e_inval);
1802         while (len--) {
1803                 struct ceph_pg_pool_info *pi;
1804
1805                 ceph_decode_64_safe(p, end, pool, e_inval);
1806                 pi = __lookup_pg_pool(&map->pg_pools, pool);
1807                 if (pi)
1808                         __remove_pg_pool(&map->pg_pools, pi);
1809         }
1810
1811         /* new_up_client, new_state, new_weight */
1812         err = decode_new_up_state_weight(p, end, struct_v, map);
1813         if (err)
1814                 goto bad;
1815
1816         /* new_pg_temp */
1817         err = decode_new_pg_temp(p, end, map);
1818         if (err)
1819                 goto bad;
1820
1821         /* new_primary_temp */
1822         if (struct_v >= 1) {
1823                 err = decode_new_primary_temp(p, end, map);
1824                 if (err)
1825                         goto bad;
1826         }
1827
1828         /* new_primary_affinity */
1829         if (struct_v >= 2) {
1830                 err = decode_new_primary_affinity(p, end, map);
1831                 if (err)
1832                         goto bad;
1833         }
1834
1835         if (struct_v >= 3) {
1836                 /* new_erasure_code_profiles */
1837                 ceph_decode_skip_map_of_map(p, end, string, string, string,
1838                                             e_inval);
1839                 /* old_erasure_code_profiles */
1840                 ceph_decode_skip_set(p, end, string, e_inval);
1841         }
1842
1843         if (struct_v >= 4) {
1844                 err = decode_new_pg_upmap(p, end, map);
1845                 if (err)
1846                         goto bad;
1847
1848                 err = decode_old_pg_upmap(p, end, map);
1849                 if (err)
1850                         goto bad;
1851
1852                 err = decode_new_pg_upmap_items(p, end, map);
1853                 if (err)
1854                         goto bad;
1855
1856                 err = decode_old_pg_upmap_items(p, end, map);
1857                 if (err)
1858                         goto bad;
1859         }
1860
1861         /* ignore the rest */
1862         *p = end;
1863
1864         dout("inc osdmap epoch %d max_osd %d\n", map->epoch, map->max_osd);
1865         return map;
1866
1867 e_inval:
1868         err = -EINVAL;
1869 bad:
1870         pr_err("corrupt inc osdmap (%d) epoch %d off %d (%p of %p-%p)\n",
1871                err, epoch, (int)(*p - start), *p, start, end);
1872         print_hex_dump(KERN_DEBUG, "osdmap: ",
1873                        DUMP_PREFIX_OFFSET, 16, 1,
1874                        start, end - start, true);
1875         return ERR_PTR(err);
1876 }
1877
1878 void ceph_oloc_copy(struct ceph_object_locator *dest,
1879                     const struct ceph_object_locator *src)
1880 {
1881         ceph_oloc_destroy(dest);
1882
1883         dest->pool = src->pool;
1884         if (src->pool_ns)
1885                 dest->pool_ns = ceph_get_string(src->pool_ns);
1886         else
1887                 dest->pool_ns = NULL;
1888 }
1889 EXPORT_SYMBOL(ceph_oloc_copy);
1890
1891 void ceph_oloc_destroy(struct ceph_object_locator *oloc)
1892 {
1893         ceph_put_string(oloc->pool_ns);
1894 }
1895 EXPORT_SYMBOL(ceph_oloc_destroy);
1896
1897 void ceph_oid_copy(struct ceph_object_id *dest,
1898                    const struct ceph_object_id *src)
1899 {
1900         ceph_oid_destroy(dest);
1901
1902         if (src->name != src->inline_name) {
1903                 /* very rare, see ceph_object_id definition */
1904                 dest->name = kmalloc(src->name_len + 1,
1905                                      GFP_NOIO | __GFP_NOFAIL);
1906         } else {
1907                 dest->name = dest->inline_name;
1908         }
1909         memcpy(dest->name, src->name, src->name_len + 1);
1910         dest->name_len = src->name_len;
1911 }
1912 EXPORT_SYMBOL(ceph_oid_copy);
1913
1914 static __printf(2, 0)
1915 int oid_printf_vargs(struct ceph_object_id *oid, const char *fmt, va_list ap)
1916 {
1917         int len;
1918
1919         WARN_ON(!ceph_oid_empty(oid));
1920
1921         len = vsnprintf(oid->inline_name, sizeof(oid->inline_name), fmt, ap);
1922         if (len >= sizeof(oid->inline_name))
1923                 return len;
1924
1925         oid->name_len = len;
1926         return 0;
1927 }
1928
1929 /*
1930  * If oid doesn't fit into inline buffer, BUG.
1931  */
1932 void ceph_oid_printf(struct ceph_object_id *oid, const char *fmt, ...)
1933 {
1934         va_list ap;
1935
1936         va_start(ap, fmt);
1937         BUG_ON(oid_printf_vargs(oid, fmt, ap));
1938         va_end(ap);
1939 }
1940 EXPORT_SYMBOL(ceph_oid_printf);
1941
1942 static __printf(3, 0)
1943 int oid_aprintf_vargs(struct ceph_object_id *oid, gfp_t gfp,
1944                       const char *fmt, va_list ap)
1945 {
1946         va_list aq;
1947         int len;
1948
1949         va_copy(aq, ap);
1950         len = oid_printf_vargs(oid, fmt, aq);
1951         va_end(aq);
1952
1953         if (len) {
1954                 char *external_name;
1955
1956                 external_name = kmalloc(len + 1, gfp);
1957                 if (!external_name)
1958                         return -ENOMEM;
1959
1960                 oid->name = external_name;
1961                 WARN_ON(vsnprintf(oid->name, len + 1, fmt, ap) != len);
1962                 oid->name_len = len;
1963         }
1964
1965         return 0;
1966 }
1967
1968 /*
1969  * If oid doesn't fit into inline buffer, allocate.
1970  */
1971 int ceph_oid_aprintf(struct ceph_object_id *oid, gfp_t gfp,
1972                      const char *fmt, ...)
1973 {
1974         va_list ap;
1975         int ret;
1976
1977         va_start(ap, fmt);
1978         ret = oid_aprintf_vargs(oid, gfp, fmt, ap);
1979         va_end(ap);
1980
1981         return ret;
1982 }
1983 EXPORT_SYMBOL(ceph_oid_aprintf);
1984
1985 void ceph_oid_destroy(struct ceph_object_id *oid)
1986 {
1987         if (oid->name != oid->inline_name)
1988                 kfree(oid->name);
1989 }
1990 EXPORT_SYMBOL(ceph_oid_destroy);
1991
1992 /*
1993  * osds only
1994  */
1995 static bool __osds_equal(const struct ceph_osds *lhs,
1996                          const struct ceph_osds *rhs)
1997 {
1998         if (lhs->size == rhs->size &&
1999             !memcmp(lhs->osds, rhs->osds, rhs->size * sizeof(rhs->osds[0])))
2000                 return true;
2001
2002         return false;
2003 }
2004
2005 /*
2006  * osds + primary
2007  */
2008 static bool osds_equal(const struct ceph_osds *lhs,
2009                        const struct ceph_osds *rhs)
2010 {
2011         if (__osds_equal(lhs, rhs) &&
2012             lhs->primary == rhs->primary)
2013                 return true;
2014
2015         return false;
2016 }
2017
2018 static bool osds_valid(const struct ceph_osds *set)
2019 {
2020         /* non-empty set */
2021         if (set->size > 0 && set->primary >= 0)
2022                 return true;
2023
2024         /* empty can_shift_osds set */
2025         if (!set->size && set->primary == -1)
2026                 return true;
2027
2028         /* empty !can_shift_osds set - all NONE */
2029         if (set->size > 0 && set->primary == -1) {
2030                 int i;
2031
2032                 for (i = 0; i < set->size; i++) {
2033                         if (set->osds[i] != CRUSH_ITEM_NONE)
2034                                 break;
2035                 }
2036                 if (i == set->size)
2037                         return true;
2038         }
2039
2040         return false;
2041 }
2042
2043 void ceph_osds_copy(struct ceph_osds *dest, const struct ceph_osds *src)
2044 {
2045         memcpy(dest->osds, src->osds, src->size * sizeof(src->osds[0]));
2046         dest->size = src->size;
2047         dest->primary = src->primary;
2048 }
2049
2050 bool ceph_pg_is_split(const struct ceph_pg *pgid, u32 old_pg_num,
2051                       u32 new_pg_num)
2052 {
2053         int old_bits = calc_bits_of(old_pg_num);
2054         int old_mask = (1 << old_bits) - 1;
2055         int n;
2056
2057         WARN_ON(pgid->seed >= old_pg_num);
2058         if (new_pg_num <= old_pg_num)
2059                 return false;
2060
2061         for (n = 1; ; n++) {
2062                 int next_bit = n << (old_bits - 1);
2063                 u32 s = next_bit | pgid->seed;
2064
2065                 if (s < old_pg_num || s == pgid->seed)
2066                         continue;
2067                 if (s >= new_pg_num)
2068                         break;
2069
2070                 s = ceph_stable_mod(s, old_pg_num, old_mask);
2071                 if (s == pgid->seed)
2072                         return true;
2073         }
2074
2075         return false;
2076 }
2077
2078 bool ceph_is_new_interval(const struct ceph_osds *old_acting,
2079                           const struct ceph_osds *new_acting,
2080                           const struct ceph_osds *old_up,
2081                           const struct ceph_osds *new_up,
2082                           int old_size,
2083                           int new_size,
2084                           int old_min_size,
2085                           int new_min_size,
2086                           u32 old_pg_num,
2087                           u32 new_pg_num,
2088                           bool old_sort_bitwise,
2089                           bool new_sort_bitwise,
2090                           bool old_recovery_deletes,
2091                           bool new_recovery_deletes,
2092                           const struct ceph_pg *pgid)
2093 {
2094         return !osds_equal(old_acting, new_acting) ||
2095                !osds_equal(old_up, new_up) ||
2096                old_size != new_size ||
2097                old_min_size != new_min_size ||
2098                ceph_pg_is_split(pgid, old_pg_num, new_pg_num) ||
2099                old_sort_bitwise != new_sort_bitwise ||
2100                old_recovery_deletes != new_recovery_deletes;
2101 }
2102
2103 static int calc_pg_rank(int osd, const struct ceph_osds *acting)
2104 {
2105         int i;
2106
2107         for (i = 0; i < acting->size; i++) {
2108                 if (acting->osds[i] == osd)
2109                         return i;
2110         }
2111
2112         return -1;
2113 }
2114
2115 static bool primary_changed(const struct ceph_osds *old_acting,
2116                             const struct ceph_osds *new_acting)
2117 {
2118         if (!old_acting->size && !new_acting->size)
2119                 return false; /* both still empty */
2120
2121         if (!old_acting->size ^ !new_acting->size)
2122                 return true; /* was empty, now not, or vice versa */
2123
2124         if (old_acting->primary != new_acting->primary)
2125                 return true; /* primary changed */
2126
2127         if (calc_pg_rank(old_acting->primary, old_acting) !=
2128             calc_pg_rank(new_acting->primary, new_acting))
2129                 return true;
2130
2131         return false; /* same primary (tho replicas may have changed) */
2132 }
2133
2134 bool ceph_osds_changed(const struct ceph_osds *old_acting,
2135                        const struct ceph_osds *new_acting,
2136                        bool any_change)
2137 {
2138         if (primary_changed(old_acting, new_acting))
2139                 return true;
2140
2141         if (any_change && !__osds_equal(old_acting, new_acting))
2142                 return true;
2143
2144         return false;
2145 }
2146
2147 /*
2148  * Map an object into a PG.
2149  *
2150  * Should only be called with target_oid and target_oloc (as opposed to
2151  * base_oid and base_oloc), since tiering isn't taken into account.
2152  */
2153 void __ceph_object_locator_to_pg(struct ceph_pg_pool_info *pi,
2154                                  const struct ceph_object_id *oid,
2155                                  const struct ceph_object_locator *oloc,
2156                                  struct ceph_pg *raw_pgid)
2157 {
2158         WARN_ON(pi->id != oloc->pool);
2159
2160         if (!oloc->pool_ns) {
2161                 raw_pgid->pool = oloc->pool;
2162                 raw_pgid->seed = ceph_str_hash(pi->object_hash, oid->name,
2163                                              oid->name_len);
2164                 dout("%s %s -> raw_pgid %llu.%x\n", __func__, oid->name,
2165                      raw_pgid->pool, raw_pgid->seed);
2166         } else {
2167                 char stack_buf[256];
2168                 char *buf = stack_buf;
2169                 int nsl = oloc->pool_ns->len;
2170                 size_t total = nsl + 1 + oid->name_len;
2171
2172                 if (total > sizeof(stack_buf))
2173                         buf = kmalloc(total, GFP_NOIO | __GFP_NOFAIL);
2174                 memcpy(buf, oloc->pool_ns->str, nsl);
2175                 buf[nsl] = '\037';
2176                 memcpy(buf + nsl + 1, oid->name, oid->name_len);
2177                 raw_pgid->pool = oloc->pool;
2178                 raw_pgid->seed = ceph_str_hash(pi->object_hash, buf, total);
2179                 if (buf != stack_buf)
2180                         kfree(buf);
2181                 dout("%s %s ns %.*s -> raw_pgid %llu.%x\n", __func__,
2182                      oid->name, nsl, oloc->pool_ns->str,
2183                      raw_pgid->pool, raw_pgid->seed);
2184         }
2185 }
2186
2187 int ceph_object_locator_to_pg(struct ceph_osdmap *osdmap,
2188                               const struct ceph_object_id *oid,
2189                               const struct ceph_object_locator *oloc,
2190                               struct ceph_pg *raw_pgid)
2191 {
2192         struct ceph_pg_pool_info *pi;
2193
2194         pi = ceph_pg_pool_by_id(osdmap, oloc->pool);
2195         if (!pi)
2196                 return -ENOENT;
2197
2198         __ceph_object_locator_to_pg(pi, oid, oloc, raw_pgid);
2199         return 0;
2200 }
2201 EXPORT_SYMBOL(ceph_object_locator_to_pg);
2202
2203 /*
2204  * Map a raw PG (full precision ps) into an actual PG.
2205  */
2206 static void raw_pg_to_pg(struct ceph_pg_pool_info *pi,
2207                          const struct ceph_pg *raw_pgid,
2208                          struct ceph_pg *pgid)
2209 {
2210         pgid->pool = raw_pgid->pool;
2211         pgid->seed = ceph_stable_mod(raw_pgid->seed, pi->pg_num,
2212                                      pi->pg_num_mask);
2213 }
2214
2215 /*
2216  * Map a raw PG (full precision ps) into a placement ps (placement
2217  * seed).  Include pool id in that value so that different pools don't
2218  * use the same seeds.
2219  */
2220 static u32 raw_pg_to_pps(struct ceph_pg_pool_info *pi,
2221                          const struct ceph_pg *raw_pgid)
2222 {
2223         if (pi->flags & CEPH_POOL_FLAG_HASHPSPOOL) {
2224                 /* hash pool id and seed so that pool PGs do not overlap */
2225                 return crush_hash32_2(CRUSH_HASH_RJENKINS1,
2226                                       ceph_stable_mod(raw_pgid->seed,
2227                                                       pi->pgp_num,
2228                                                       pi->pgp_num_mask),
2229                                       raw_pgid->pool);
2230         } else {
2231                 /*
2232                  * legacy behavior: add ps and pool together.  this is
2233                  * not a great approach because the PGs from each pool
2234                  * will overlap on top of each other: 0.5 == 1.4 ==
2235                  * 2.3 == ...
2236                  */
2237                 return ceph_stable_mod(raw_pgid->seed, pi->pgp_num,
2238                                        pi->pgp_num_mask) +
2239                        (unsigned)raw_pgid->pool;
2240         }
2241 }
2242
2243 /*
2244  * Magic value used for a "default" fallback choose_args, used if the
2245  * crush_choose_arg_map passed to do_crush() does not exist.  If this
2246  * also doesn't exist, fall back to canonical weights.
2247  */
2248 #define CEPH_DEFAULT_CHOOSE_ARGS        -1
2249
2250 static int do_crush(struct ceph_osdmap *map, int ruleno, int x,
2251                     int *result, int result_max,
2252                     const __u32 *weight, int weight_max,
2253                     s64 choose_args_index)
2254 {
2255         struct crush_choose_arg_map *arg_map;
2256         int r;
2257
2258         BUG_ON(result_max > CEPH_PG_MAX_SIZE);
2259
2260         arg_map = lookup_choose_arg_map(&map->crush->choose_args,
2261                                         choose_args_index);
2262         if (!arg_map)
2263                 arg_map = lookup_choose_arg_map(&map->crush->choose_args,
2264                                                 CEPH_DEFAULT_CHOOSE_ARGS);
2265
2266         mutex_lock(&map->crush_workspace_mutex);
2267         r = crush_do_rule(map->crush, ruleno, x, result, result_max,
2268                           weight, weight_max, map->crush_workspace,
2269                           arg_map ? arg_map->args : NULL);
2270         mutex_unlock(&map->crush_workspace_mutex);
2271
2272         return r;
2273 }
2274
2275 static void remove_nonexistent_osds(struct ceph_osdmap *osdmap,
2276                                     struct ceph_pg_pool_info *pi,
2277                                     struct ceph_osds *set)
2278 {
2279         int i;
2280
2281         if (ceph_can_shift_osds(pi)) {
2282                 int removed = 0;
2283
2284                 /* shift left */
2285                 for (i = 0; i < set->size; i++) {
2286                         if (!ceph_osd_exists(osdmap, set->osds[i])) {
2287                                 removed++;
2288                                 continue;
2289                         }
2290                         if (removed)
2291                                 set->osds[i - removed] = set->osds[i];
2292                 }
2293                 set->size -= removed;
2294         } else {
2295                 /* set dne devices to NONE */
2296                 for (i = 0; i < set->size; i++) {
2297                         if (!ceph_osd_exists(osdmap, set->osds[i]))
2298                                 set->osds[i] = CRUSH_ITEM_NONE;
2299                 }
2300         }
2301 }
2302
2303 /*
2304  * Calculate raw set (CRUSH output) for given PG and filter out
2305  * nonexistent OSDs.  ->primary is undefined for a raw set.
2306  *
2307  * Placement seed (CRUSH input) is returned through @ppps.
2308  */
2309 static void pg_to_raw_osds(struct ceph_osdmap *osdmap,
2310                            struct ceph_pg_pool_info *pi,
2311                            const struct ceph_pg *raw_pgid,
2312                            struct ceph_osds *raw,
2313                            u32 *ppps)
2314 {
2315         u32 pps = raw_pg_to_pps(pi, raw_pgid);
2316         int ruleno;
2317         int len;
2318
2319         ceph_osds_init(raw);
2320         if (ppps)
2321                 *ppps = pps;
2322
2323         ruleno = crush_find_rule(osdmap->crush, pi->crush_ruleset, pi->type,
2324                                  pi->size);
2325         if (ruleno < 0) {
2326                 pr_err("no crush rule: pool %lld ruleset %d type %d size %d\n",
2327                        pi->id, pi->crush_ruleset, pi->type, pi->size);
2328                 return;
2329         }
2330
2331         if (pi->size > ARRAY_SIZE(raw->osds)) {
2332                 pr_err_ratelimited("pool %lld ruleset %d type %d too wide: size %d > %zu\n",
2333                        pi->id, pi->crush_ruleset, pi->type, pi->size,
2334                        ARRAY_SIZE(raw->osds));
2335                 return;
2336         }
2337
2338         len = do_crush(osdmap, ruleno, pps, raw->osds, pi->size,
2339                        osdmap->osd_weight, osdmap->max_osd, pi->id);
2340         if (len < 0) {
2341                 pr_err("error %d from crush rule %d: pool %lld ruleset %d type %d size %d\n",
2342                        len, ruleno, pi->id, pi->crush_ruleset, pi->type,
2343                        pi->size);
2344                 return;
2345         }
2346
2347         raw->size = len;
2348         remove_nonexistent_osds(osdmap, pi, raw);
2349 }
2350
2351 /* apply pg_upmap[_items] mappings */
2352 static void apply_upmap(struct ceph_osdmap *osdmap,
2353                         const struct ceph_pg *pgid,
2354                         struct ceph_osds *raw)
2355 {
2356         struct ceph_pg_mapping *pg;
2357         int i, j;
2358
2359         pg = lookup_pg_mapping(&osdmap->pg_upmap, pgid);
2360         if (pg) {
2361                 /* make sure targets aren't marked out */
2362                 for (i = 0; i < pg->pg_upmap.len; i++) {
2363                         int osd = pg->pg_upmap.osds[i];
2364
2365                         if (osd != CRUSH_ITEM_NONE &&
2366                             osd < osdmap->max_osd &&
2367                             osdmap->osd_weight[osd] == 0) {
2368                                 /* reject/ignore explicit mapping */
2369                                 return;
2370                         }
2371                 }
2372                 for (i = 0; i < pg->pg_upmap.len; i++)
2373                         raw->osds[i] = pg->pg_upmap.osds[i];
2374                 raw->size = pg->pg_upmap.len;
2375                 /* check and apply pg_upmap_items, if any */
2376         }
2377
2378         pg = lookup_pg_mapping(&osdmap->pg_upmap_items, pgid);
2379         if (pg) {
2380                 /*
2381                  * Note: this approach does not allow a bidirectional swap,
2382                  * e.g., [[1,2],[2,1]] applied to [0,1,2] -> [0,2,1].
2383                  */
2384                 for (i = 0; i < pg->pg_upmap_items.len; i++) {
2385                         int from = pg->pg_upmap_items.from_to[i][0];
2386                         int to = pg->pg_upmap_items.from_to[i][1];
2387                         int pos = -1;
2388                         bool exists = false;
2389
2390                         /* make sure replacement doesn't already appear */
2391                         for (j = 0; j < raw->size; j++) {
2392                                 int osd = raw->osds[j];
2393
2394                                 if (osd == to) {
2395                                         exists = true;
2396                                         break;
2397                                 }
2398                                 /* ignore mapping if target is marked out */
2399                                 if (osd == from && pos < 0 &&
2400                                     !(to != CRUSH_ITEM_NONE &&
2401                                       to < osdmap->max_osd &&
2402                                       osdmap->osd_weight[to] == 0)) {
2403                                         pos = j;
2404                                 }
2405                         }
2406                         if (!exists && pos >= 0)
2407                                 raw->osds[pos] = to;
2408                 }
2409         }
2410 }
2411
2412 /*
2413  * Given raw set, calculate up set and up primary.  By definition of an
2414  * up set, the result won't contain nonexistent or down OSDs.
2415  *
2416  * This is done in-place - on return @set is the up set.  If it's
2417  * empty, ->primary will remain undefined.
2418  */
2419 static void raw_to_up_osds(struct ceph_osdmap *osdmap,
2420                            struct ceph_pg_pool_info *pi,
2421                            struct ceph_osds *set)
2422 {
2423         int i;
2424
2425         /* ->primary is undefined for a raw set */
2426         BUG_ON(set->primary != -1);
2427
2428         if (ceph_can_shift_osds(pi)) {
2429                 int removed = 0;
2430
2431                 /* shift left */
2432                 for (i = 0; i < set->size; i++) {
2433                         if (ceph_osd_is_down(osdmap, set->osds[i])) {
2434                                 removed++;
2435                                 continue;
2436                         }
2437                         if (removed)
2438                                 set->osds[i - removed] = set->osds[i];
2439                 }
2440                 set->size -= removed;
2441                 if (set->size > 0)
2442                         set->primary = set->osds[0];
2443         } else {
2444                 /* set down/dne devices to NONE */
2445                 for (i = set->size - 1; i >= 0; i--) {
2446                         if (ceph_osd_is_down(osdmap, set->osds[i]))
2447                                 set->osds[i] = CRUSH_ITEM_NONE;
2448                         else
2449                                 set->primary = set->osds[i];
2450                 }
2451         }
2452 }
2453
2454 static void apply_primary_affinity(struct ceph_osdmap *osdmap,
2455                                    struct ceph_pg_pool_info *pi,
2456                                    u32 pps,
2457                                    struct ceph_osds *up)
2458 {
2459         int i;
2460         int pos = -1;
2461
2462         /*
2463          * Do we have any non-default primary_affinity values for these
2464          * osds?
2465          */
2466         if (!osdmap->osd_primary_affinity)
2467                 return;
2468
2469         for (i = 0; i < up->size; i++) {
2470                 int osd = up->osds[i];
2471
2472                 if (osd != CRUSH_ITEM_NONE &&
2473                     osdmap->osd_primary_affinity[osd] !=
2474                                         CEPH_OSD_DEFAULT_PRIMARY_AFFINITY) {
2475                         break;
2476                 }
2477         }
2478         if (i == up->size)
2479                 return;
2480
2481         /*
2482          * Pick the primary.  Feed both the seed (for the pg) and the
2483          * osd into the hash/rng so that a proportional fraction of an
2484          * osd's pgs get rejected as primary.
2485          */
2486         for (i = 0; i < up->size; i++) {
2487                 int osd = up->osds[i];
2488                 u32 aff;
2489
2490                 if (osd == CRUSH_ITEM_NONE)
2491                         continue;
2492
2493                 aff = osdmap->osd_primary_affinity[osd];
2494                 if (aff < CEPH_OSD_MAX_PRIMARY_AFFINITY &&
2495                     (crush_hash32_2(CRUSH_HASH_RJENKINS1,
2496                                     pps, osd) >> 16) >= aff) {
2497                         /*
2498                          * We chose not to use this primary.  Note it
2499                          * anyway as a fallback in case we don't pick
2500                          * anyone else, but keep looking.
2501                          */
2502                         if (pos < 0)
2503                                 pos = i;
2504                 } else {
2505                         pos = i;
2506                         break;
2507                 }
2508         }
2509         if (pos < 0)
2510                 return;
2511
2512         up->primary = up->osds[pos];
2513
2514         if (ceph_can_shift_osds(pi) && pos > 0) {
2515                 /* move the new primary to the front */
2516                 for (i = pos; i > 0; i--)
2517                         up->osds[i] = up->osds[i - 1];
2518                 up->osds[0] = up->primary;
2519         }
2520 }
2521
2522 /*
2523  * Get pg_temp and primary_temp mappings for given PG.
2524  *
2525  * Note that a PG may have none, only pg_temp, only primary_temp or
2526  * both pg_temp and primary_temp mappings.  This means @temp isn't
2527  * always a valid OSD set on return: in the "only primary_temp" case,
2528  * @temp will have its ->primary >= 0 but ->size == 0.
2529  */
2530 static void get_temp_osds(struct ceph_osdmap *osdmap,
2531                           struct ceph_pg_pool_info *pi,
2532                           const struct ceph_pg *pgid,
2533                           struct ceph_osds *temp)
2534 {
2535         struct ceph_pg_mapping *pg;
2536         int i;
2537
2538         ceph_osds_init(temp);
2539
2540         /* pg_temp? */
2541         pg = lookup_pg_mapping(&osdmap->pg_temp, pgid);
2542         if (pg) {
2543                 for (i = 0; i < pg->pg_temp.len; i++) {
2544                         if (ceph_osd_is_down(osdmap, pg->pg_temp.osds[i])) {
2545                                 if (ceph_can_shift_osds(pi))
2546                                         continue;
2547
2548                                 temp->osds[temp->size++] = CRUSH_ITEM_NONE;
2549                         } else {
2550                                 temp->osds[temp->size++] = pg->pg_temp.osds[i];
2551                         }
2552                 }
2553
2554                 /* apply pg_temp's primary */
2555                 for (i = 0; i < temp->size; i++) {
2556                         if (temp->osds[i] != CRUSH_ITEM_NONE) {
2557                                 temp->primary = temp->osds[i];
2558                                 break;
2559                         }
2560                 }
2561         }
2562
2563         /* primary_temp? */
2564         pg = lookup_pg_mapping(&osdmap->primary_temp, pgid);
2565         if (pg)
2566                 temp->primary = pg->primary_temp.osd;
2567 }
2568
2569 /*
2570  * Map a PG to its acting set as well as its up set.
2571  *
2572  * Acting set is used for data mapping purposes, while up set can be
2573  * recorded for detecting interval changes and deciding whether to
2574  * resend a request.
2575  */
2576 void ceph_pg_to_up_acting_osds(struct ceph_osdmap *osdmap,
2577                                struct ceph_pg_pool_info *pi,
2578                                const struct ceph_pg *raw_pgid,
2579                                struct ceph_osds *up,
2580                                struct ceph_osds *acting)
2581 {
2582         struct ceph_pg pgid;
2583         u32 pps;
2584
2585         WARN_ON(pi->id != raw_pgid->pool);
2586         raw_pg_to_pg(pi, raw_pgid, &pgid);
2587
2588         pg_to_raw_osds(osdmap, pi, raw_pgid, up, &pps);
2589         apply_upmap(osdmap, &pgid, up);
2590         raw_to_up_osds(osdmap, pi, up);
2591         apply_primary_affinity(osdmap, pi, pps, up);
2592         get_temp_osds(osdmap, pi, &pgid, acting);
2593         if (!acting->size) {
2594                 memcpy(acting->osds, up->osds, up->size * sizeof(up->osds[0]));
2595                 acting->size = up->size;
2596                 if (acting->primary == -1)
2597                         acting->primary = up->primary;
2598         }
2599         WARN_ON(!osds_valid(up) || !osds_valid(acting));
2600 }
2601
2602 bool ceph_pg_to_primary_shard(struct ceph_osdmap *osdmap,
2603                               struct ceph_pg_pool_info *pi,
2604                               const struct ceph_pg *raw_pgid,
2605                               struct ceph_spg *spgid)
2606 {
2607         struct ceph_pg pgid;
2608         struct ceph_osds up, acting;
2609         int i;
2610
2611         WARN_ON(pi->id != raw_pgid->pool);
2612         raw_pg_to_pg(pi, raw_pgid, &pgid);
2613
2614         if (ceph_can_shift_osds(pi)) {
2615                 spgid->pgid = pgid; /* struct */
2616                 spgid->shard = CEPH_SPG_NOSHARD;
2617                 return true;
2618         }
2619
2620         ceph_pg_to_up_acting_osds(osdmap, pi, &pgid, &up, &acting);
2621         for (i = 0; i < acting.size; i++) {
2622                 if (acting.osds[i] == acting.primary) {
2623                         spgid->pgid = pgid; /* struct */
2624                         spgid->shard = i;
2625                         return true;
2626                 }
2627         }
2628
2629         return false;
2630 }
2631
2632 /*
2633  * Return acting primary for given PG, or -1 if none.
2634  */
2635 int ceph_pg_to_acting_primary(struct ceph_osdmap *osdmap,
2636                               const struct ceph_pg *raw_pgid)
2637 {
2638         struct ceph_pg_pool_info *pi;
2639         struct ceph_osds up, acting;
2640
2641         pi = ceph_pg_pool_by_id(osdmap, raw_pgid->pool);
2642         if (!pi)
2643                 return -1;
2644
2645         ceph_pg_to_up_acting_osds(osdmap, pi, raw_pgid, &up, &acting);
2646         return acting.primary;
2647 }
2648 EXPORT_SYMBOL(ceph_pg_to_acting_primary);