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         kvfree(map->osd_state);
 977         kvfree(map->osd_weight);
 978         kvfree(map->osd_addr);
 979         kvfree(map->osd_primary_affinity);
 980         kvfree(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, u32 max)
 990 {
 991         u32 *state;
 992         u32 *weight;
 993         struct ceph_entity_addr *addr;
 994         u32 to_copy;
 995         int i;
 996
 997         dout("%s old %u new %u\n", __func__, map->max_osd, max);
 998         if (max == map->max_osd)
 999                 return 0;
1000
1001         state = ceph_kvmalloc(array_size(max, sizeof(*state)), GFP_NOFS);
1002         weight = ceph_kvmalloc(array_size(max, sizeof(*weight)), GFP_NOFS);
1003         addr = ceph_kvmalloc(array_size(max, sizeof(*addr)), GFP_NOFS);
1004         if (!state || !weight || !addr) {
1005                 kvfree(state);
1006                 kvfree(weight);
1007                 kvfree(addr);
1008                 return -ENOMEM;
1009         }
1010
1011         to_copy = min(map->max_osd, max);
1012         if (map->osd_state) {
1013                 memcpy(state, map->osd_state, to_copy * sizeof(*state));
1014                 memcpy(weight, map->osd_weight, to_copy * sizeof(*weight));
1015                 memcpy(addr, map->osd_addr, to_copy * sizeof(*addr));
1016                 kvfree(map->osd_state);
1017                 kvfree(map->osd_weight);
1018                 kvfree(map->osd_addr);
1019         }
1020
1021         map->osd_state = state;
1022         map->osd_weight = weight;
1023         map->osd_addr = addr;
1024         for (i = map->max_osd; i < max; i++) {
1025                 map->osd_state[i] = 0;
1026                 map->osd_weight[i] = CEPH_OSD_OUT;
1027                 memset(map->osd_addr + i, 0, sizeof(*map->osd_addr));
1028         }
1029
1030         if (map->osd_primary_affinity) {
1031                 u32 *affinity;
1032
1033                 affinity = ceph_kvmalloc(array_size(max, sizeof(*affinity)),
1034                                          GFP_NOFS);
1035                 if (!affinity)
1036                         return -ENOMEM;
1037
1038                 memcpy(affinity, map->osd_primary_affinity,
1039                        to_copy * sizeof(*affinity));
1040                 kvfree(map->osd_primary_affinity);
1041
1042                 map->osd_primary_affinity = affinity;
1043                 for (i = map->max_osd; i < max; i++)
1044                         map->osd_primary_affinity[i] =
1045                             CEPH_OSD_DEFAULT_PRIMARY_AFFINITY;
1046         }
1047
1048         map->max_osd = max;
1049
1050         return 0;
1051 }
1052
1053 static int osdmap_set_crush(struct ceph_osdmap *map, struct crush_map *crush)
1054 {
1055         void *workspace;
1056         size_t work_size;
1057
1058         if (IS_ERR(crush))
1059                 return PTR_ERR(crush);
1060
1061         work_size = crush_work_size(crush, CEPH_PG_MAX_SIZE);
1062         dout("%s work_size %zu bytes\n", __func__, work_size);
1063         workspace = ceph_kvmalloc(work_size, GFP_NOIO);
1064         if (!workspace) {
1065                 crush_destroy(crush);
1066                 return -ENOMEM;
1067         }
1068         crush_init_workspace(crush, workspace);
1069
1070         if (map->crush)
1071                 crush_destroy(map->crush);
1072         kvfree(map->crush_workspace);
1073         map->crush = crush;
1074         map->crush_workspace = workspace;
1075         return 0;
1076 }
1077
1078 #define OSDMAP_WRAPPER_COMPAT_VER       7
1079 #define OSDMAP_CLIENT_DATA_COMPAT_VER   1
1080
1081 /*
1082  * Return 0 or error.  On success, *v is set to 0 for old (v6) osdmaps,
1083  * to struct_v of the client_data section for new (v7 and above)
1084  * osdmaps.
1085  */
1086 static int get_osdmap_client_data_v(void **p, void *end,
1087                                     const char *prefix, u8 *v)
1088 {
1089         u8 struct_v;
1090
1091         ceph_decode_8_safe(p, end, struct_v, e_inval);
1092         if (struct_v >= 7) {
1093                 u8 struct_compat;
1094
1095                 ceph_decode_8_safe(p, end, struct_compat, e_inval);
1096                 if (struct_compat > OSDMAP_WRAPPER_COMPAT_VER) {
1097                         pr_warn("got v %d cv %d > %d of %s ceph_osdmap\n",
1098                                 struct_v, struct_compat,
1099                                 OSDMAP_WRAPPER_COMPAT_VER, prefix);
1100                         return -EINVAL;
1101                 }
1102                 *p += 4; /* ignore wrapper struct_len */
1103
1104                 ceph_decode_8_safe(p, end, struct_v, e_inval);
1105                 ceph_decode_8_safe(p, end, struct_compat, e_inval);
1106                 if (struct_compat > OSDMAP_CLIENT_DATA_COMPAT_VER) {
1107                         pr_warn("got v %d cv %d > %d of %s ceph_osdmap client data\n",
1108                                 struct_v, struct_compat,
1109                                 OSDMAP_CLIENT_DATA_COMPAT_VER, prefix);
1110                         return -EINVAL;
1111                 }
1112                 *p += 4; /* ignore client data struct_len */
1113         } else {
1114                 u16 version;
1115
1116                 *p -= 1;
1117                 ceph_decode_16_safe(p, end, version, e_inval);
1118                 if (version < 6) {
1119                         pr_warn("got v %d < 6 of %s ceph_osdmap\n",
1120                                 version, prefix);
1121                         return -EINVAL;
1122                 }
1123
1124                 /* old osdmap enconding */
1125                 struct_v = 0;
1126         }
1127
1128         *v = struct_v;
1129         return 0;
1130
1131 e_inval:
1132         return -EINVAL;
1133 }
1134
1135 static int __decode_pools(void **p, void *end, struct ceph_osdmap *map,
1136                           bool incremental)
1137 {
1138         u32 n;
1139
1140         ceph_decode_32_safe(p, end, n, e_inval);
1141         while (n--) {
1142                 struct ceph_pg_pool_info *pi;
1143                 u64 pool;
1144                 int ret;
1145
1146                 ceph_decode_64_safe(p, end, pool, e_inval);
1147
1148                 pi = __lookup_pg_pool(&map->pg_pools, pool);
1149                 if (!incremental || !pi) {
1150                         pi = kzalloc(sizeof(*pi), GFP_NOFS);
1151                         if (!pi)
1152                                 return -ENOMEM;
1153
1154                         pi->id = pool;
1155
1156                         ret = __insert_pg_pool(&map->pg_pools, pi);
1157                         if (ret) {
1158                                 kfree(pi);
1159                                 return ret;
1160                         }
1161                 }
1162
1163                 ret = decode_pool(p, end, pi);
1164                 if (ret)
1165                         return ret;
1166         }
1167
1168         return 0;
1169
1170 e_inval:
1171         return -EINVAL;
1172 }
1173
1174 static int decode_pools(void **p, void *end, struct ceph_osdmap *map)
1175 {
1176         return __decode_pools(p, end, map, false);
1177 }
1178
1179 static int decode_new_pools(void **p, void *end, struct ceph_osdmap *map)
1180 {
1181         return __decode_pools(p, end, map, true);
1182 }
1183
1184 typedef struct ceph_pg_mapping *(*decode_mapping_fn_t)(void **, void *, bool);
1185
1186 static int decode_pg_mapping(void **p, void *end, struct rb_root *mapping_root,
1187                              decode_mapping_fn_t fn, bool incremental)
1188 {
1189         u32 n;
1190
1191         WARN_ON(!incremental && !fn);
1192
1193         ceph_decode_32_safe(p, end, n, e_inval);
1194         while (n--) {
1195                 struct ceph_pg_mapping *pg;
1196                 struct ceph_pg pgid;
1197                 int ret;
1198
1199                 ret = ceph_decode_pgid(p, end, &pgid);
1200                 if (ret)
1201                         return ret;
1202
1203                 pg = lookup_pg_mapping(mapping_root, &pgid);
1204                 if (pg) {
1205                         WARN_ON(!incremental);
1206                         erase_pg_mapping(mapping_root, pg);
1207                         free_pg_mapping(pg);
1208                 }
1209
1210                 if (fn) {
1211                         pg = fn(p, end, incremental);
1212                         if (IS_ERR(pg))
1213                                 return PTR_ERR(pg);
1214
1215                         if (pg) {
1216                                 pg->pgid = pgid; /* struct */
1217                                 insert_pg_mapping(mapping_root, pg);
1218                         }
1219                 }
1220         }
1221
1222         return 0;
1223
1224 e_inval:
1225         return -EINVAL;
1226 }
1227
1228 static struct ceph_pg_mapping *__decode_pg_temp(void **p, void *end,
1229                                                 bool incremental)
1230 {
1231         struct ceph_pg_mapping *pg;
1232         u32 len, i;
1233
1234         ceph_decode_32_safe(p, end, len, e_inval);
1235         if (len == 0 && incremental)
1236                 return NULL;    /* new_pg_temp: [] to remove */
1237         if (len > (SIZE_MAX - sizeof(*pg)) / sizeof(u32))
1238                 return ERR_PTR(-EINVAL);
1239
1240         ceph_decode_need(p, end, len * sizeof(u32), e_inval);
1241         pg = alloc_pg_mapping(len * sizeof(u32));
1242         if (!pg)
1243                 return ERR_PTR(-ENOMEM);
1244
1245         pg->pg_temp.len = len;
1246         for (i = 0; i < len; i++)
1247                 pg->pg_temp.osds[i] = ceph_decode_32(p);
1248
1249         return pg;
1250
1251 e_inval:
1252         return ERR_PTR(-EINVAL);
1253 }
1254
1255 static int decode_pg_temp(void **p, void *end, struct ceph_osdmap *map)
1256 {
1257         return decode_pg_mapping(p, end, &map->pg_temp, __decode_pg_temp,
1258                                  false);
1259 }
1260
1261 static int decode_new_pg_temp(void **p, void *end, struct ceph_osdmap *map)
1262 {
1263         return decode_pg_mapping(p, end, &map->pg_temp, __decode_pg_temp,
1264                                  true);
1265 }
1266
1267 static struct ceph_pg_mapping *__decode_primary_temp(void **p, void *end,
1268                                                      bool incremental)
1269 {
1270         struct ceph_pg_mapping *pg;
1271         u32 osd;
1272
1273         ceph_decode_32_safe(p, end, osd, e_inval);
1274         if (osd == (u32)-1 && incremental)
1275                 return NULL;    /* new_primary_temp: -1 to remove */
1276
1277         pg = alloc_pg_mapping(0);
1278         if (!pg)
1279                 return ERR_PTR(-ENOMEM);
1280
1281         pg->primary_temp.osd = osd;
1282         return pg;
1283
1284 e_inval:
1285         return ERR_PTR(-EINVAL);
1286 }
1287
1288 static int decode_primary_temp(void **p, void *end, struct ceph_osdmap *map)
1289 {
1290         return decode_pg_mapping(p, end, &map->primary_temp,
1291                                  __decode_primary_temp, false);
1292 }
1293
1294 static int decode_new_primary_temp(void **p, void *end,
1295                                    struct ceph_osdmap *map)
1296 {
1297         return decode_pg_mapping(p, end, &map->primary_temp,
1298                                  __decode_primary_temp, true);
1299 }
1300
1301 u32 ceph_get_primary_affinity(struct ceph_osdmap *map, int osd)
1302 {
1303         BUG_ON(osd >= map->max_osd);
1304
1305         if (!map->osd_primary_affinity)
1306                 return CEPH_OSD_DEFAULT_PRIMARY_AFFINITY;
1307
1308         return map->osd_primary_affinity[osd];
1309 }
1310
1311 static int set_primary_affinity(struct ceph_osdmap *map, int osd, u32 aff)
1312 {
1313         BUG_ON(osd >= map->max_osd);
1314
1315         if (!map->osd_primary_affinity) {
1316                 int i;
1317
1318                 map->osd_primary_affinity = ceph_kvmalloc(
1319                     array_size(map->max_osd, sizeof(*map->osd_primary_affinity)),
1320                     GFP_NOFS);
1321                 if (!map->osd_primary_affinity)
1322                         return -ENOMEM;
1323
1324                 for (i = 0; i < map->max_osd; i++)
1325                         map->osd_primary_affinity[i] =
1326                             CEPH_OSD_DEFAULT_PRIMARY_AFFINITY;
1327         }
1328
1329         map->osd_primary_affinity[osd] = aff;
1330
1331         return 0;
1332 }
1333
1334 static int decode_primary_affinity(void **p, void *end,
1335                                    struct ceph_osdmap *map)
1336 {
1337         u32 len, i;
1338
1339         ceph_decode_32_safe(p, end, len, e_inval);
1340         if (len == 0) {
1341                 kvfree(map->osd_primary_affinity);
1342                 map->osd_primary_affinity = NULL;
1343                 return 0;
1344         }
1345         if (len != map->max_osd)
1346                 goto e_inval;
1347
1348         ceph_decode_need(p, end, map->max_osd*sizeof(u32), e_inval);
1349
1350         for (i = 0; i < map->max_osd; i++) {
1351                 int ret;
1352
1353                 ret = set_primary_affinity(map, i, ceph_decode_32(p));
1354                 if (ret)
1355                         return ret;
1356         }
1357
1358         return 0;
1359
1360 e_inval:
1361         return -EINVAL;
1362 }
1363
1364 static int decode_new_primary_affinity(void **p, void *end,
1365                                        struct ceph_osdmap *map)
1366 {
1367         u32 n;
1368
1369         ceph_decode_32_safe(p, end, n, e_inval);
1370         while (n--) {
1371                 u32 osd, aff;
1372                 int ret;
1373
1374                 ceph_decode_32_safe(p, end, osd, e_inval);
1375                 ceph_decode_32_safe(p, end, aff, e_inval);
1376
1377                 ret = set_primary_affinity(map, osd, aff);
1378                 if (ret)
1379                         return ret;
1380
1381                 pr_info("osd%d primary-affinity 0x%x\n", osd, aff);
1382         }
1383
1384         return 0;
1385
1386 e_inval:
1387         return -EINVAL;
1388 }
1389
1390 static struct ceph_pg_mapping *__decode_pg_upmap(void **p, void *end,
1391                                                  bool __unused)
1392 {
1393         return __decode_pg_temp(p, end, false);
1394 }
1395
1396 static int decode_pg_upmap(void **p, void *end, struct ceph_osdmap *map)
1397 {
1398         return decode_pg_mapping(p, end, &map->pg_upmap, __decode_pg_upmap,
1399                                  false);
1400 }
1401
1402 static int decode_new_pg_upmap(void **p, void *end, struct ceph_osdmap *map)
1403 {
1404         return decode_pg_mapping(p, end, &map->pg_upmap, __decode_pg_upmap,
1405                                  true);
1406 }
1407
1408 static int decode_old_pg_upmap(void **p, void *end, struct ceph_osdmap *map)
1409 {
1410         return decode_pg_mapping(p, end, &map->pg_upmap, NULL, true);
1411 }
1412
1413 static struct ceph_pg_mapping *__decode_pg_upmap_items(void **p, void *end,
1414                                                        bool __unused)
1415 {
1416         struct ceph_pg_mapping *pg;
1417         u32 len, i;
1418
1419         ceph_decode_32_safe(p, end, len, e_inval);
1420         if (len > (SIZE_MAX - sizeof(*pg)) / (2 * sizeof(u32)))
1421                 return ERR_PTR(-EINVAL);
1422
1423         ceph_decode_need(p, end, 2 * len * sizeof(u32), e_inval);
1424         pg = alloc_pg_mapping(2 * len * sizeof(u32));
1425         if (!pg)
1426                 return ERR_PTR(-ENOMEM);
1427
1428         pg->pg_upmap_items.len = len;
1429         for (i = 0; i < len; i++) {
1430                 pg->pg_upmap_items.from_to[i][0] = ceph_decode_32(p);
1431                 pg->pg_upmap_items.from_to[i][1] = ceph_decode_32(p);
1432         }
1433
1434         return pg;
1435
1436 e_inval:
1437         return ERR_PTR(-EINVAL);
1438 }
1439
1440 static int decode_pg_upmap_items(void **p, void *end, struct ceph_osdmap *map)
1441 {
1442         return decode_pg_mapping(p, end, &map->pg_upmap_items,
1443                                  __decode_pg_upmap_items, false);
1444 }
1445
1446 static int decode_new_pg_upmap_items(void **p, void *end,
1447                                      struct ceph_osdmap *map)
1448 {
1449         return decode_pg_mapping(p, end, &map->pg_upmap_items,
1450                                  __decode_pg_upmap_items, true);
1451 }
1452
1453 static int decode_old_pg_upmap_items(void **p, void *end,
1454                                      struct ceph_osdmap *map)
1455 {
1456         return decode_pg_mapping(p, end, &map->pg_upmap_items, NULL, true);
1457 }
1458
1459 /*
1460  * decode a full map.
1461  */
1462 static int osdmap_decode(void **p, void *end, struct ceph_osdmap *map)
1463 {
1464         u8 struct_v;
1465         u32 epoch = 0;
1466         void *start = *p;
1467         u32 max;
1468         u32 len, i;
1469         int err;
1470
1471         dout("%s %p to %p len %d\n", __func__, *p, end, (int)(end - *p));
1472
1473         err = get_osdmap_client_data_v(p, end, "full", &struct_v);
1474         if (err)
1475                 goto bad;
1476
1477         /* fsid, epoch, created, modified */
1478         ceph_decode_need(p, end, sizeof(map->fsid) + sizeof(u32) +
1479                          sizeof(map->created) + sizeof(map->modified), e_inval);
1480         ceph_decode_copy(p, &map->fsid, sizeof(map->fsid));
1481         epoch = map->epoch = ceph_decode_32(p);
1482         ceph_decode_copy(p, &map->created, sizeof(map->created));
1483         ceph_decode_copy(p, &map->modified, sizeof(map->modified));
1484
1485         /* pools */
1486         err = decode_pools(p, end, map);
1487         if (err)
1488                 goto bad;
1489
1490         /* pool_name */
1491         err = decode_pool_names(p, end, map);
1492         if (err)
1493                 goto bad;
1494
1495         ceph_decode_32_safe(p, end, map->pool_max, e_inval);
1496
1497         ceph_decode_32_safe(p, end, map->flags, e_inval);
1498
1499         /* max_osd */
1500         ceph_decode_32_safe(p, end, max, e_inval);
1501
1502         /* (re)alloc osd arrays */
1503         err = osdmap_set_max_osd(map, max);
1504         if (err)
1505                 goto bad;
1506
1507         /* osd_state, osd_weight, osd_addrs->client_addr */
1508         ceph_decode_need(p, end, 3*sizeof(u32) +
1509                          map->max_osd*(struct_v >= 5 ? sizeof(u32) :
1510                                                        sizeof(u8)) +
1511                                        sizeof(*map->osd_weight), e_inval);
1512         if (ceph_decode_32(p) != map->max_osd)
1513                 goto e_inval;
1514
1515         if (struct_v >= 5) {
1516                 for (i = 0; i < map->max_osd; i++)
1517                         map->osd_state[i] = ceph_decode_32(p);
1518         } else {
1519                 for (i = 0; i < map->max_osd; i++)
1520                         map->osd_state[i] = ceph_decode_8(p);
1521         }
1522
1523         if (ceph_decode_32(p) != map->max_osd)
1524                 goto e_inval;
1525
1526         for (i = 0; i < map->max_osd; i++)
1527                 map->osd_weight[i] = ceph_decode_32(p);
1528
1529         if (ceph_decode_32(p) != map->max_osd)
1530                 goto e_inval;
1531
1532         for (i = 0; i < map->max_osd; i++) {
1533                 err = ceph_decode_entity_addr(p, end, &map->osd_addr[i]);
1534                 if (err)
1535                         goto bad;
1536         }
1537
1538         /* pg_temp */
1539         err = decode_pg_temp(p, end, map);
1540         if (err)
1541                 goto bad;
1542
1543         /* primary_temp */
1544         if (struct_v >= 1) {
1545                 err = decode_primary_temp(p, end, map);
1546                 if (err)
1547                         goto bad;
1548         }
1549
1550         /* primary_affinity */
1551         if (struct_v >= 2) {
1552                 err = decode_primary_affinity(p, end, map);
1553                 if (err)
1554                         goto bad;
1555         } else {
1556                 WARN_ON(map->osd_primary_affinity);
1557         }
1558
1559         /* crush */
1560         ceph_decode_32_safe(p, end, len, e_inval);
1561         err = osdmap_set_crush(map, crush_decode(*p, min(*p + len, end)));
1562         if (err)
1563                 goto bad;
1564
1565         *p += len;
1566         if (struct_v >= 3) {
1567                 /* erasure_code_profiles */
1568                 ceph_decode_skip_map_of_map(p, end, string, string, string,
1569                                             e_inval);
1570         }
1571
1572         if (struct_v >= 4) {
1573                 err = decode_pg_upmap(p, end, map);
1574                 if (err)
1575                         goto bad;
1576
1577                 err = decode_pg_upmap_items(p, end, map);
1578                 if (err)
1579                         goto bad;
1580         } else {
1581                 WARN_ON(!RB_EMPTY_ROOT(&map->pg_upmap));
1582                 WARN_ON(!RB_EMPTY_ROOT(&map->pg_upmap_items));
1583         }
1584
1585         /* ignore the rest */
1586         *p = end;
1587
1588         dout("full osdmap epoch %d max_osd %d\n", map->epoch, map->max_osd);
1589         return 0;
1590
1591 e_inval:
1592         err = -EINVAL;
1593 bad:
1594         pr_err("corrupt full osdmap (%d) epoch %d off %d (%p of %p-%p)\n",
1595                err, epoch, (int)(*p - start), *p, start, end);
1596         print_hex_dump(KERN_DEBUG, "osdmap: ",
1597                        DUMP_PREFIX_OFFSET, 16, 1,
1598                        start, end - start, true);
1599         return err;
1600 }
1601
1602 /*
1603  * Allocate and decode a full map.
1604  */
1605 struct ceph_osdmap *ceph_osdmap_decode(void **p, void *end)
1606 {
1607         struct ceph_osdmap *map;
1608         int ret;
1609
1610         map = ceph_osdmap_alloc();
1611         if (!map)
1612                 return ERR_PTR(-ENOMEM);
1613
1614         ret = osdmap_decode(p, end, map);
1615         if (ret) {
1616                 ceph_osdmap_destroy(map);
1617                 return ERR_PTR(ret);
1618         }
1619
1620         return map;
1621 }
1622
1623 /*
1624  * Encoding order is (new_up_client, new_state, new_weight).  Need to
1625  * apply in the (new_weight, new_state, new_up_client) order, because
1626  * an incremental map may look like e.g.
1627  *
1628  *     new_up_client: { osd=6, addr=... } # set osd_state and addr
1629  *     new_state: { osd=6, xorstate=EXISTS } # clear osd_state
1630  */
1631 static int decode_new_up_state_weight(void **p, void *end, u8 struct_v,
1632                                       struct ceph_osdmap *map)
1633 {
1634         void *new_up_client;
1635         void *new_state;
1636         void *new_weight_end;
1637         u32 len;
1638         int i;
1639
1640         new_up_client = *p;
1641         ceph_decode_32_safe(p, end, len, e_inval);
1642         for (i = 0; i < len; ++i) {
1643                 struct ceph_entity_addr addr;
1644
1645                 ceph_decode_skip_32(p, end, e_inval);
1646                 if (ceph_decode_entity_addr(p, end, &addr))
1647                         goto e_inval;
1648         }
1649
1650         new_state = *p;
1651         ceph_decode_32_safe(p, end, len, e_inval);
1652         len *= sizeof(u32) + (struct_v >= 5 ? sizeof(u32) : sizeof(u8));
1653         ceph_decode_need(p, end, len, e_inval);
1654         *p += len;
1655
1656         /* new_weight */
1657         ceph_decode_32_safe(p, end, len, e_inval);
1658         while (len--) {
1659                 s32 osd;
1660                 u32 w;
1661
1662                 ceph_decode_need(p, end, 2*sizeof(u32), e_inval);
1663                 osd = ceph_decode_32(p);
1664                 w = ceph_decode_32(p);
1665                 BUG_ON(osd >= map->max_osd);
1666                 pr_info("osd%d weight 0x%x %s\n", osd, w,
1667                      w == CEPH_OSD_IN ? "(in)" :
1668                      (w == CEPH_OSD_OUT ? "(out)" : ""));
1669                 map->osd_weight[osd] = w;
1670
1671                 /*
1672                  * If we are marking in, set the EXISTS, and clear the
1673                  * AUTOOUT and NEW bits.
1674                  */
1675                 if (w) {
1676                         map->osd_state[osd] |= CEPH_OSD_EXISTS;
1677                         map->osd_state[osd] &= ~(CEPH_OSD_AUTOOUT |
1678                                                  CEPH_OSD_NEW);
1679                 }
1680         }
1681         new_weight_end = *p;
1682
1683         /* new_state (up/down) */
1684         *p = new_state;
1685         len = ceph_decode_32(p);
1686         while (len--) {
1687                 s32 osd;
1688                 u32 xorstate;
1689                 int ret;
1690
1691                 osd = ceph_decode_32(p);
1692                 if (struct_v >= 5)
1693                         xorstate = ceph_decode_32(p);
1694                 else
1695                         xorstate = ceph_decode_8(p);
1696                 if (xorstate == 0)
1697                         xorstate = CEPH_OSD_UP;
1698                 BUG_ON(osd >= map->max_osd);
1699                 if ((map->osd_state[osd] & CEPH_OSD_UP) &&
1700                     (xorstate & CEPH_OSD_UP))
1701                         pr_info("osd%d down\n", osd);
1702                 if ((map->osd_state[osd] & CEPH_OSD_EXISTS) &&
1703                     (xorstate & CEPH_OSD_EXISTS)) {
1704                         pr_info("osd%d does not exist\n", osd);
1705                         ret = set_primary_affinity(map, osd,
1706                                                    CEPH_OSD_DEFAULT_PRIMARY_AFFINITY);
1707                         if (ret)
1708                                 return ret;
1709                         memset(map->osd_addr + osd, 0, sizeof(*map->osd_addr));
1710                         map->osd_state[osd] = 0;
1711                 } else {
1712                         map->osd_state[osd] ^= xorstate;
1713                 }
1714         }
1715
1716         /* new_up_client */
1717         *p = new_up_client;
1718         len = ceph_decode_32(p);
1719         while (len--) {
1720                 s32 osd;
1721                 struct ceph_entity_addr addr;
1722
1723                 osd = ceph_decode_32(p);
1724                 BUG_ON(osd >= map->max_osd);
1725                 if (ceph_decode_entity_addr(p, end, &addr))
1726                         goto e_inval;
1727                 pr_info("osd%d up\n", osd);
1728                 map->osd_state[osd] |= CEPH_OSD_EXISTS | CEPH_OSD_UP;
1729                 map->osd_addr[osd] = addr;
1730         }
1731
1732         *p = new_weight_end;
1733         return 0;
1734
1735 e_inval:
1736         return -EINVAL;
1737 }
1738
1739 /*
1740  * decode and apply an incremental map update.
1741  */
1742 struct ceph_osdmap *osdmap_apply_incremental(void **p, void *end,
1743                                              struct ceph_osdmap *map)
1744 {
1745         struct ceph_fsid fsid;
1746         u32 epoch = 0;
1747         struct ceph_timespec modified;
1748         s32 len;
1749         u64 pool;
1750         __s64 new_pool_max;
1751         __s32 new_flags, max;
1752         void *start = *p;
1753         int err;
1754         u8 struct_v;
1755
1756         dout("%s %p to %p len %d\n", __func__, *p, end, (int)(end - *p));
1757
1758         err = get_osdmap_client_data_v(p, end, "inc", &struct_v);
1759         if (err)
1760                 goto bad;
1761
1762         /* fsid, epoch, modified, new_pool_max, new_flags */
1763         ceph_decode_need(p, end, sizeof(fsid) + sizeof(u32) + sizeof(modified) +
1764                          sizeof(u64) + sizeof(u32), e_inval);
1765         ceph_decode_copy(p, &fsid, sizeof(fsid));
1766         epoch = ceph_decode_32(p);
1767         BUG_ON(epoch != map->epoch+1);
1768         ceph_decode_copy(p, &modified, sizeof(modified));
1769         new_pool_max = ceph_decode_64(p);
1770         new_flags = ceph_decode_32(p);
1771
1772         /* full map? */
1773         ceph_decode_32_safe(p, end, len, e_inval);
1774         if (len > 0) {
1775                 dout("apply_incremental full map len %d, %p to %p\n",
1776                      len, *p, end);
1777                 return ceph_osdmap_decode(p, min(*p+len, end));
1778         }
1779
1780         /* new crush? */
1781         ceph_decode_32_safe(p, end, len, e_inval);
1782         if (len > 0) {
1783                 err = osdmap_set_crush(map,
1784                                        crush_decode(*p, min(*p + len, end)));
1785                 if (err)
1786                         goto bad;
1787                 *p += len;
1788         }
1789
1790         /* new flags? */
1791         if (new_flags >= 0)
1792                 map->flags = new_flags;
1793         if (new_pool_max >= 0)
1794                 map->pool_max = new_pool_max;
1795
1796         /* new max? */
1797         ceph_decode_32_safe(p, end, max, e_inval);
1798         if (max >= 0) {
1799                 err = osdmap_set_max_osd(map, max);
1800                 if (err)
1801                         goto bad;
1802         }
1803
1804         map->epoch++;
1805         map->modified = modified;
1806
1807         /* new_pools */
1808         err = decode_new_pools(p, end, map);
1809         if (err)
1810                 goto bad;
1811
1812         /* new_pool_names */
1813         err = decode_pool_names(p, end, map);
1814         if (err)
1815                 goto bad;
1816
1817         /* old_pool */
1818         ceph_decode_32_safe(p, end, len, e_inval);
1819         while (len--) {
1820                 struct ceph_pg_pool_info *pi;
1821
1822                 ceph_decode_64_safe(p, end, pool, e_inval);
1823                 pi = __lookup_pg_pool(&map->pg_pools, pool);
1824                 if (pi)
1825                         __remove_pg_pool(&map->pg_pools, pi);
1826         }
1827
1828         /* new_up_client, new_state, new_weight */
1829         err = decode_new_up_state_weight(p, end, struct_v, map);
1830         if (err)
1831                 goto bad;
1832
1833         /* new_pg_temp */
1834         err = decode_new_pg_temp(p, end, map);
1835         if (err)
1836                 goto bad;
1837
1838         /* new_primary_temp */
1839         if (struct_v >= 1) {
1840                 err = decode_new_primary_temp(p, end, map);
1841                 if (err)
1842                         goto bad;
1843         }
1844
1845         /* new_primary_affinity */
1846         if (struct_v >= 2) {
1847                 err = decode_new_primary_affinity(p, end, map);
1848                 if (err)
1849                         goto bad;
1850         }
1851
1852         if (struct_v >= 3) {
1853                 /* new_erasure_code_profiles */
1854                 ceph_decode_skip_map_of_map(p, end, string, string, string,
1855                                             e_inval);
1856                 /* old_erasure_code_profiles */
1857                 ceph_decode_skip_set(p, end, string, e_inval);
1858         }
1859
1860         if (struct_v >= 4) {
1861                 err = decode_new_pg_upmap(p, end, map);
1862                 if (err)
1863                         goto bad;
1864
1865                 err = decode_old_pg_upmap(p, end, map);
1866                 if (err)
1867                         goto bad;
1868
1869                 err = decode_new_pg_upmap_items(p, end, map);
1870                 if (err)
1871                         goto bad;
1872
1873                 err = decode_old_pg_upmap_items(p, end, map);
1874                 if (err)
1875                         goto bad;
1876         }
1877
1878         /* ignore the rest */
1879         *p = end;
1880
1881         dout("inc osdmap epoch %d max_osd %d\n", map->epoch, map->max_osd);
1882         return map;
1883
1884 e_inval:
1885         err = -EINVAL;
1886 bad:
1887         pr_err("corrupt inc osdmap (%d) epoch %d off %d (%p of %p-%p)\n",
1888                err, epoch, (int)(*p - start), *p, start, end);
1889         print_hex_dump(KERN_DEBUG, "osdmap: ",
1890                        DUMP_PREFIX_OFFSET, 16, 1,
1891                        start, end - start, true);
1892         return ERR_PTR(err);
1893 }
1894
1895 void ceph_oloc_copy(struct ceph_object_locator *dest,
1896                     const struct ceph_object_locator *src)
1897 {
1898         ceph_oloc_destroy(dest);
1899
1900         dest->pool = src->pool;
1901         if (src->pool_ns)
1902                 dest->pool_ns = ceph_get_string(src->pool_ns);
1903         else
1904                 dest->pool_ns = NULL;
1905 }
1906 EXPORT_SYMBOL(ceph_oloc_copy);
1907
1908 void ceph_oloc_destroy(struct ceph_object_locator *oloc)
1909 {
1910         ceph_put_string(oloc->pool_ns);
1911 }
1912 EXPORT_SYMBOL(ceph_oloc_destroy);
1913
1914 void ceph_oid_copy(struct ceph_object_id *dest,
1915                    const struct ceph_object_id *src)
1916 {
1917         ceph_oid_destroy(dest);
1918
1919         if (src->name != src->inline_name) {
1920                 /* very rare, see ceph_object_id definition */
1921                 dest->name = kmalloc(src->name_len + 1,
1922                                      GFP_NOIO | __GFP_NOFAIL);
1923         } else {
1924                 dest->name = dest->inline_name;
1925         }
1926         memcpy(dest->name, src->name, src->name_len + 1);
1927         dest->name_len = src->name_len;
1928 }
1929 EXPORT_SYMBOL(ceph_oid_copy);
1930
1931 static __printf(2, 0)
1932 int oid_printf_vargs(struct ceph_object_id *oid, const char *fmt, va_list ap)
1933 {
1934         int len;
1935
1936         WARN_ON(!ceph_oid_empty(oid));
1937
1938         len = vsnprintf(oid->inline_name, sizeof(oid->inline_name), fmt, ap);
1939         if (len >= sizeof(oid->inline_name))
1940                 return len;
1941
1942         oid->name_len = len;
1943         return 0;
1944 }
1945
1946 /*
1947  * If oid doesn't fit into inline buffer, BUG.
1948  */
1949 void ceph_oid_printf(struct ceph_object_id *oid, const char *fmt, ...)
1950 {
1951         va_list ap;
1952
1953         va_start(ap, fmt);
1954         BUG_ON(oid_printf_vargs(oid, fmt, ap));
1955         va_end(ap);
1956 }
1957 EXPORT_SYMBOL(ceph_oid_printf);
1958
1959 static __printf(3, 0)
1960 int oid_aprintf_vargs(struct ceph_object_id *oid, gfp_t gfp,
1961                       const char *fmt, va_list ap)
1962 {
1963         va_list aq;
1964         int len;
1965
1966         va_copy(aq, ap);
1967         len = oid_printf_vargs(oid, fmt, aq);
1968         va_end(aq);
1969
1970         if (len) {
1971                 char *external_name;
1972
1973                 external_name = kmalloc(len + 1, gfp);
1974                 if (!external_name)
1975                         return -ENOMEM;
1976
1977                 oid->name = external_name;
1978                 WARN_ON(vsnprintf(oid->name, len + 1, fmt, ap) != len);
1979                 oid->name_len = len;
1980         }
1981
1982         return 0;
1983 }
1984
1985 /*
1986  * If oid doesn't fit into inline buffer, allocate.
1987  */
1988 int ceph_oid_aprintf(struct ceph_object_id *oid, gfp_t gfp,
1989                      const char *fmt, ...)
1990 {
1991         va_list ap;
1992         int ret;
1993
1994         va_start(ap, fmt);
1995         ret = oid_aprintf_vargs(oid, gfp, fmt, ap);
1996         va_end(ap);
1997
1998         return ret;
1999 }
2000 EXPORT_SYMBOL(ceph_oid_aprintf);
2001
2002 void ceph_oid_destroy(struct ceph_object_id *oid)
2003 {
2004         if (oid->name != oid->inline_name)
2005                 kfree(oid->name);
2006 }
2007 EXPORT_SYMBOL(ceph_oid_destroy);
2008
2009 /*
2010  * osds only
2011  */
2012 static bool __osds_equal(const struct ceph_osds *lhs,
2013                          const struct ceph_osds *rhs)
2014 {
2015         if (lhs->size == rhs->size &&
2016             !memcmp(lhs->osds, rhs->osds, rhs->size * sizeof(rhs->osds[0])))
2017                 return true;
2018
2019         return false;
2020 }
2021
2022 /*
2023  * osds + primary
2024  */
2025 static bool osds_equal(const struct ceph_osds *lhs,
2026                        const struct ceph_osds *rhs)
2027 {
2028         if (__osds_equal(lhs, rhs) &&
2029             lhs->primary == rhs->primary)
2030                 return true;
2031
2032         return false;
2033 }
2034
2035 static bool osds_valid(const struct ceph_osds *set)
2036 {
2037         /* non-empty set */
2038         if (set->size > 0 && set->primary >= 0)
2039                 return true;
2040
2041         /* empty can_shift_osds set */
2042         if (!set->size && set->primary == -1)
2043                 return true;
2044
2045         /* empty !can_shift_osds set - all NONE */
2046         if (set->size > 0 && set->primary == -1) {
2047                 int i;
2048
2049                 for (i = 0; i < set->size; i++) {
2050                         if (set->osds[i] != CRUSH_ITEM_NONE)
2051                                 break;
2052                 }
2053                 if (i == set->size)
2054                         return true;
2055         }
2056
2057         return false;
2058 }
2059
2060 void ceph_osds_copy(struct ceph_osds *dest, const struct ceph_osds *src)
2061 {
2062         memcpy(dest->osds, src->osds, src->size * sizeof(src->osds[0]));
2063         dest->size = src->size;
2064         dest->primary = src->primary;
2065 }
2066
2067 bool ceph_pg_is_split(const struct ceph_pg *pgid, u32 old_pg_num,
2068                       u32 new_pg_num)
2069 {
2070         int old_bits = calc_bits_of(old_pg_num);
2071         int old_mask = (1 << old_bits) - 1;
2072         int n;
2073
2074         WARN_ON(pgid->seed >= old_pg_num);
2075         if (new_pg_num <= old_pg_num)
2076                 return false;
2077
2078         for (n = 1; ; n++) {
2079                 int next_bit = n << (old_bits - 1);
2080                 u32 s = next_bit | pgid->seed;
2081
2082                 if (s < old_pg_num || s == pgid->seed)
2083                         continue;
2084                 if (s >= new_pg_num)
2085                         break;
2086
2087                 s = ceph_stable_mod(s, old_pg_num, old_mask);
2088                 if (s == pgid->seed)
2089                         return true;
2090         }
2091
2092         return false;
2093 }
2094
2095 bool ceph_is_new_interval(const struct ceph_osds *old_acting,
2096                           const struct ceph_osds *new_acting,
2097                           const struct ceph_osds *old_up,
2098                           const struct ceph_osds *new_up,
2099                           int old_size,
2100                           int new_size,
2101                           int old_min_size,
2102                           int new_min_size,
2103                           u32 old_pg_num,
2104                           u32 new_pg_num,
2105                           bool old_sort_bitwise,
2106                           bool new_sort_bitwise,
2107                           bool old_recovery_deletes,
2108                           bool new_recovery_deletes,
2109                           const struct ceph_pg *pgid)
2110 {
2111         return !osds_equal(old_acting, new_acting) ||
2112                !osds_equal(old_up, new_up) ||
2113                old_size != new_size ||
2114                old_min_size != new_min_size ||
2115                ceph_pg_is_split(pgid, old_pg_num, new_pg_num) ||
2116                old_sort_bitwise != new_sort_bitwise ||
2117                old_recovery_deletes != new_recovery_deletes;
2118 }
2119
2120 static int calc_pg_rank(int osd, const struct ceph_osds *acting)
2121 {
2122         int i;
2123
2124         for (i = 0; i < acting->size; i++) {
2125                 if (acting->osds[i] == osd)
2126                         return i;
2127         }
2128
2129         return -1;
2130 }
2131
2132 static bool primary_changed(const struct ceph_osds *old_acting,
2133                             const struct ceph_osds *new_acting)
2134 {
2135         if (!old_acting->size && !new_acting->size)
2136                 return false; /* both still empty */
2137
2138         if (!old_acting->size ^ !new_acting->size)
2139                 return true; /* was empty, now not, or vice versa */
2140
2141         if (old_acting->primary != new_acting->primary)
2142                 return true; /* primary changed */
2143
2144         if (calc_pg_rank(old_acting->primary, old_acting) !=
2145             calc_pg_rank(new_acting->primary, new_acting))
2146                 return true;
2147
2148         return false; /* same primary (tho replicas may have changed) */
2149 }
2150
2151 bool ceph_osds_changed(const struct ceph_osds *old_acting,
2152                        const struct ceph_osds *new_acting,
2153                        bool any_change)
2154 {
2155         if (primary_changed(old_acting, new_acting))
2156                 return true;
2157
2158         if (any_change && !__osds_equal(old_acting, new_acting))
2159                 return true;
2160
2161         return false;
2162 }
2163
2164 /*
2165  * Map an object into a PG.
2166  *
2167  * Should only be called with target_oid and target_oloc (as opposed to
2168  * base_oid and base_oloc), since tiering isn't taken into account.
2169  */
2170 void __ceph_object_locator_to_pg(struct ceph_pg_pool_info *pi,
2171                                  const struct ceph_object_id *oid,
2172                                  const struct ceph_object_locator *oloc,
2173                                  struct ceph_pg *raw_pgid)
2174 {
2175         WARN_ON(pi->id != oloc->pool);
2176
2177         if (!oloc->pool_ns) {
2178                 raw_pgid->pool = oloc->pool;
2179                 raw_pgid->seed = ceph_str_hash(pi->object_hash, oid->name,
2180                                              oid->name_len);
2181                 dout("%s %s -> raw_pgid %llu.%x\n", __func__, oid->name,
2182                      raw_pgid->pool, raw_pgid->seed);
2183         } else {
2184                 char stack_buf[256];
2185                 char *buf = stack_buf;
2186                 int nsl = oloc->pool_ns->len;
2187                 size_t total = nsl + 1 + oid->name_len;
2188
2189                 if (total > sizeof(stack_buf))
2190                         buf = kmalloc(total, GFP_NOIO | __GFP_NOFAIL);
2191                 memcpy(buf, oloc->pool_ns->str, nsl);
2192                 buf[nsl] = '\037';
2193                 memcpy(buf + nsl + 1, oid->name, oid->name_len);
2194                 raw_pgid->pool = oloc->pool;
2195                 raw_pgid->seed = ceph_str_hash(pi->object_hash, buf, total);
2196                 if (buf != stack_buf)
2197                         kfree(buf);
2198                 dout("%s %s ns %.*s -> raw_pgid %llu.%x\n", __func__,
2199                      oid->name, nsl, oloc->pool_ns->str,
2200                      raw_pgid->pool, raw_pgid->seed);
2201         }
2202 }
2203
2204 int ceph_object_locator_to_pg(struct ceph_osdmap *osdmap,
2205                               const struct ceph_object_id *oid,
2206                               const struct ceph_object_locator *oloc,
2207                               struct ceph_pg *raw_pgid)
2208 {
2209         struct ceph_pg_pool_info *pi;
2210
2211         pi = ceph_pg_pool_by_id(osdmap, oloc->pool);
2212         if (!pi)
2213                 return -ENOENT;
2214
2215         __ceph_object_locator_to_pg(pi, oid, oloc, raw_pgid);
2216         return 0;
2217 }
2218 EXPORT_SYMBOL(ceph_object_locator_to_pg);
2219
2220 /*
2221  * Map a raw PG (full precision ps) into an actual PG.
2222  */
2223 static void raw_pg_to_pg(struct ceph_pg_pool_info *pi,
2224                          const struct ceph_pg *raw_pgid,
2225                          struct ceph_pg *pgid)
2226 {
2227         pgid->pool = raw_pgid->pool;
2228         pgid->seed = ceph_stable_mod(raw_pgid->seed, pi->pg_num,
2229                                      pi->pg_num_mask);
2230 }
2231
2232 /*
2233  * Map a raw PG (full precision ps) into a placement ps (placement
2234  * seed).  Include pool id in that value so that different pools don't
2235  * use the same seeds.
2236  */
2237 static u32 raw_pg_to_pps(struct ceph_pg_pool_info *pi,
2238                          const struct ceph_pg *raw_pgid)
2239 {
2240         if (pi->flags & CEPH_POOL_FLAG_HASHPSPOOL) {
2241                 /* hash pool id and seed so that pool PGs do not overlap */
2242                 return crush_hash32_2(CRUSH_HASH_RJENKINS1,
2243                                       ceph_stable_mod(raw_pgid->seed,
2244                                                       pi->pgp_num,
2245                                                       pi->pgp_num_mask),
2246                                       raw_pgid->pool);
2247         } else {
2248                 /*
2249                  * legacy behavior: add ps and pool together.  this is
2250                  * not a great approach because the PGs from each pool
2251                  * will overlap on top of each other: 0.5 == 1.4 ==
2252                  * 2.3 == ...
2253                  */
2254                 return ceph_stable_mod(raw_pgid->seed, pi->pgp_num,
2255                                        pi->pgp_num_mask) +
2256                        (unsigned)raw_pgid->pool;
2257         }
2258 }
2259
2260 /*
2261  * Magic value used for a "default" fallback choose_args, used if the
2262  * crush_choose_arg_map passed to do_crush() does not exist.  If this
2263  * also doesn't exist, fall back to canonical weights.
2264  */
2265 #define CEPH_DEFAULT_CHOOSE_ARGS        -1
2266
2267 static int do_crush(struct ceph_osdmap *map, int ruleno, int x,
2268                     int *result, int result_max,
2269                     const __u32 *weight, int weight_max,
2270                     s64 choose_args_index)
2271 {
2272         struct crush_choose_arg_map *arg_map;
2273         int r;
2274
2275         BUG_ON(result_max > CEPH_PG_MAX_SIZE);
2276
2277         arg_map = lookup_choose_arg_map(&map->crush->choose_args,
2278                                         choose_args_index);
2279         if (!arg_map)
2280                 arg_map = lookup_choose_arg_map(&map->crush->choose_args,
2281                                                 CEPH_DEFAULT_CHOOSE_ARGS);
2282
2283         mutex_lock(&map->crush_workspace_mutex);
2284         r = crush_do_rule(map->crush, ruleno, x, result, result_max,
2285                           weight, weight_max, map->crush_workspace,
2286                           arg_map ? arg_map->args : NULL);
2287         mutex_unlock(&map->crush_workspace_mutex);
2288
2289         return r;
2290 }
2291
2292 static void remove_nonexistent_osds(struct ceph_osdmap *osdmap,
2293                                     struct ceph_pg_pool_info *pi,
2294                                     struct ceph_osds *set)
2295 {
2296         int i;
2297
2298         if (ceph_can_shift_osds(pi)) {
2299                 int removed = 0;
2300
2301                 /* shift left */
2302                 for (i = 0; i < set->size; i++) {
2303                         if (!ceph_osd_exists(osdmap, set->osds[i])) {
2304                                 removed++;
2305                                 continue;
2306                         }
2307                         if (removed)
2308                                 set->osds[i - removed] = set->osds[i];
2309                 }
2310                 set->size -= removed;
2311         } else {
2312                 /* set dne devices to NONE */
2313                 for (i = 0; i < set->size; i++) {
2314                         if (!ceph_osd_exists(osdmap, set->osds[i]))
2315                                 set->osds[i] = CRUSH_ITEM_NONE;
2316                 }
2317         }
2318 }
2319
2320 /*
2321  * Calculate raw set (CRUSH output) for given PG and filter out
2322  * nonexistent OSDs.  ->primary is undefined for a raw set.
2323  *
2324  * Placement seed (CRUSH input) is returned through @ppps.
2325  */
2326 static void pg_to_raw_osds(struct ceph_osdmap *osdmap,
2327                            struct ceph_pg_pool_info *pi,
2328                            const struct ceph_pg *raw_pgid,
2329                            struct ceph_osds *raw,
2330                            u32 *ppps)
2331 {
2332         u32 pps = raw_pg_to_pps(pi, raw_pgid);
2333         int ruleno;
2334         int len;
2335
2336         ceph_osds_init(raw);
2337         if (ppps)
2338                 *ppps = pps;
2339
2340         ruleno = crush_find_rule(osdmap->crush, pi->crush_ruleset, pi->type,
2341                                  pi->size);
2342         if (ruleno < 0) {
2343                 pr_err("no crush rule: pool %lld ruleset %d type %d size %d\n",
2344                        pi->id, pi->crush_ruleset, pi->type, pi->size);
2345                 return;
2346         }
2347
2348         if (pi->size > ARRAY_SIZE(raw->osds)) {
2349                 pr_err_ratelimited("pool %lld ruleset %d type %d too wide: size %d > %zu\n",
2350                        pi->id, pi->crush_ruleset, pi->type, pi->size,
2351                        ARRAY_SIZE(raw->osds));
2352                 return;
2353         }
2354
2355         len = do_crush(osdmap, ruleno, pps, raw->osds, pi->size,
2356                        osdmap->osd_weight, osdmap->max_osd, pi->id);
2357         if (len < 0) {
2358                 pr_err("error %d from crush rule %d: pool %lld ruleset %d type %d size %d\n",
2359                        len, ruleno, pi->id, pi->crush_ruleset, pi->type,
2360                        pi->size);
2361                 return;
2362         }
2363
2364         raw->size = len;
2365         remove_nonexistent_osds(osdmap, pi, raw);
2366 }
2367
2368 /* apply pg_upmap[_items] mappings */
2369 static void apply_upmap(struct ceph_osdmap *osdmap,
2370                         const struct ceph_pg *pgid,
2371                         struct ceph_osds *raw)
2372 {
2373         struct ceph_pg_mapping *pg;
2374         int i, j;
2375
2376         pg = lookup_pg_mapping(&osdmap->pg_upmap, pgid);
2377         if (pg) {
2378                 /* make sure targets aren't marked out */
2379                 for (i = 0; i < pg->pg_upmap.len; i++) {
2380                         int osd = pg->pg_upmap.osds[i];
2381
2382                         if (osd != CRUSH_ITEM_NONE &&
2383                             osd < osdmap->max_osd &&
2384                             osdmap->osd_weight[osd] == 0) {
2385                                 /* reject/ignore explicit mapping */
2386                                 return;
2387                         }
2388                 }
2389                 for (i = 0; i < pg->pg_upmap.len; i++)
2390                         raw->osds[i] = pg->pg_upmap.osds[i];
2391                 raw->size = pg->pg_upmap.len;
2392                 /* check and apply pg_upmap_items, if any */
2393         }
2394
2395         pg = lookup_pg_mapping(&osdmap->pg_upmap_items, pgid);
2396         if (pg) {
2397                 /*
2398                  * Note: this approach does not allow a bidirectional swap,
2399                  * e.g., [[1,2],[2,1]] applied to [0,1,2] -> [0,2,1].
2400                  */
2401                 for (i = 0; i < pg->pg_upmap_items.len; i++) {
2402                         int from = pg->pg_upmap_items.from_to[i][0];
2403                         int to = pg->pg_upmap_items.from_to[i][1];
2404                         int pos = -1;
2405                         bool exists = false;
2406
2407                         /* make sure replacement doesn't already appear */
2408                         for (j = 0; j < raw->size; j++) {
2409                                 int osd = raw->osds[j];
2410
2411                                 if (osd == to) {
2412                                         exists = true;
2413                                         break;
2414                                 }
2415                                 /* ignore mapping if target is marked out */
2416                                 if (osd == from && pos < 0 &&
2417                                     !(to != CRUSH_ITEM_NONE &&
2418                                       to < osdmap->max_osd &&
2419                                       osdmap->osd_weight[to] == 0)) {
2420                                         pos = j;
2421                                 }
2422                         }
2423                         if (!exists && pos >= 0)
2424                                 raw->osds[pos] = to;
2425                 }
2426         }
2427 }
2428
2429 /*
2430  * Given raw set, calculate up set and up primary.  By definition of an
2431  * up set, the result won't contain nonexistent or down OSDs.
2432  *
2433  * This is done in-place - on return @set is the up set.  If it's
2434  * empty, ->primary will remain undefined.
2435  */
2436 static void raw_to_up_osds(struct ceph_osdmap *osdmap,
2437                            struct ceph_pg_pool_info *pi,
2438                            struct ceph_osds *set)
2439 {
2440         int i;
2441
2442         /* ->primary is undefined for a raw set */
2443         BUG_ON(set->primary != -1);
2444
2445         if (ceph_can_shift_osds(pi)) {
2446                 int removed = 0;
2447
2448                 /* shift left */
2449                 for (i = 0; i < set->size; i++) {
2450                         if (ceph_osd_is_down(osdmap, set->osds[i])) {
2451                                 removed++;
2452                                 continue;
2453                         }
2454                         if (removed)
2455                                 set->osds[i - removed] = set->osds[i];
2456                 }
2457                 set->size -= removed;
2458                 if (set->size > 0)
2459                         set->primary = set->osds[0];
2460         } else {
2461                 /* set down/dne devices to NONE */
2462                 for (i = set->size - 1; i >= 0; i--) {
2463                         if (ceph_osd_is_down(osdmap, set->osds[i]))
2464                                 set->osds[i] = CRUSH_ITEM_NONE;
2465                         else
2466                                 set->primary = set->osds[i];
2467                 }
2468         }
2469 }
2470
2471 static void apply_primary_affinity(struct ceph_osdmap *osdmap,
2472                                    struct ceph_pg_pool_info *pi,
2473                                    u32 pps,
2474                                    struct ceph_osds *up)
2475 {
2476         int i;
2477         int pos = -1;
2478
2479         /*
2480          * Do we have any non-default primary_affinity values for these
2481          * osds?
2482          */
2483         if (!osdmap->osd_primary_affinity)
2484                 return;
2485
2486         for (i = 0; i < up->size; i++) {
2487                 int osd = up->osds[i];
2488
2489                 if (osd != CRUSH_ITEM_NONE &&
2490                     osdmap->osd_primary_affinity[osd] !=
2491                                         CEPH_OSD_DEFAULT_PRIMARY_AFFINITY) {
2492                         break;
2493                 }
2494         }
2495         if (i == up->size)
2496                 return;
2497
2498         /*
2499          * Pick the primary.  Feed both the seed (for the pg) and the
2500          * osd into the hash/rng so that a proportional fraction of an
2501          * osd's pgs get rejected as primary.
2502          */
2503         for (i = 0; i < up->size; i++) {
2504                 int osd = up->osds[i];
2505                 u32 aff;
2506
2507                 if (osd == CRUSH_ITEM_NONE)
2508                         continue;
2509
2510                 aff = osdmap->osd_primary_affinity[osd];
2511                 if (aff < CEPH_OSD_MAX_PRIMARY_AFFINITY &&
2512                     (crush_hash32_2(CRUSH_HASH_RJENKINS1,
2513                                     pps, osd) >> 16) >= aff) {
2514                         /*
2515                          * We chose not to use this primary.  Note it
2516                          * anyway as a fallback in case we don't pick
2517                          * anyone else, but keep looking.
2518                          */
2519                         if (pos < 0)
2520                                 pos = i;
2521                 } else {
2522                         pos = i;
2523                         break;
2524                 }
2525         }
2526         if (pos < 0)
2527                 return;
2528
2529         up->primary = up->osds[pos];
2530
2531         if (ceph_can_shift_osds(pi) && pos > 0) {
2532                 /* move the new primary to the front */
2533                 for (i = pos; i > 0; i--)
2534                         up->osds[i] = up->osds[i - 1];
2535                 up->osds[0] = up->primary;
2536         }
2537 }
2538
2539 /*
2540  * Get pg_temp and primary_temp mappings for given PG.
2541  *
2542  * Note that a PG may have none, only pg_temp, only primary_temp or
2543  * both pg_temp and primary_temp mappings.  This means @temp isn't
2544  * always a valid OSD set on return: in the "only primary_temp" case,
2545  * @temp will have its ->primary >= 0 but ->size == 0.
2546  */
2547 static void get_temp_osds(struct ceph_osdmap *osdmap,
2548                           struct ceph_pg_pool_info *pi,
2549                           const struct ceph_pg *pgid,
2550                           struct ceph_osds *temp)
2551 {
2552         struct ceph_pg_mapping *pg;
2553         int i;
2554
2555         ceph_osds_init(temp);
2556
2557         /* pg_temp? */
2558         pg = lookup_pg_mapping(&osdmap->pg_temp, pgid);
2559         if (pg) {
2560                 for (i = 0; i < pg->pg_temp.len; i++) {
2561                         if (ceph_osd_is_down(osdmap, pg->pg_temp.osds[i])) {
2562                                 if (ceph_can_shift_osds(pi))
2563                                         continue;
2564
2565                                 temp->osds[temp->size++] = CRUSH_ITEM_NONE;
2566                         } else {
2567                                 temp->osds[temp->size++] = pg->pg_temp.osds[i];
2568                         }
2569                 }
2570
2571                 /* apply pg_temp's primary */
2572                 for (i = 0; i < temp->size; i++) {
2573                         if (temp->osds[i] != CRUSH_ITEM_NONE) {
2574                                 temp->primary = temp->osds[i];
2575                                 break;
2576                         }
2577                 }
2578         }
2579
2580         /* primary_temp? */
2581         pg = lookup_pg_mapping(&osdmap->primary_temp, pgid);
2582         if (pg)
2583                 temp->primary = pg->primary_temp.osd;
2584 }
2585
2586 /*
2587  * Map a PG to its acting set as well as its up set.
2588  *
2589  * Acting set is used for data mapping purposes, while up set can be
2590  * recorded for detecting interval changes and deciding whether to
2591  * resend a request.
2592  */
2593 void ceph_pg_to_up_acting_osds(struct ceph_osdmap *osdmap,
2594                                struct ceph_pg_pool_info *pi,
2595                                const struct ceph_pg *raw_pgid,
2596                                struct ceph_osds *up,
2597                                struct ceph_osds *acting)
2598 {
2599         struct ceph_pg pgid;
2600         u32 pps;
2601
2602         WARN_ON(pi->id != raw_pgid->pool);
2603         raw_pg_to_pg(pi, raw_pgid, &pgid);
2604
2605         pg_to_raw_osds(osdmap, pi, raw_pgid, up, &pps);
2606         apply_upmap(osdmap, &pgid, up);
2607         raw_to_up_osds(osdmap, pi, up);
2608         apply_primary_affinity(osdmap, pi, pps, up);
2609         get_temp_osds(osdmap, pi, &pgid, acting);
2610         if (!acting->size) {
2611                 memcpy(acting->osds, up->osds, up->size * sizeof(up->osds[0]));
2612                 acting->size = up->size;
2613                 if (acting->primary == -1)
2614                         acting->primary = up->primary;
2615         }
2616         WARN_ON(!osds_valid(up) || !osds_valid(acting));
2617 }
2618
2619 bool ceph_pg_to_primary_shard(struct ceph_osdmap *osdmap,
2620                               struct ceph_pg_pool_info *pi,
2621                               const struct ceph_pg *raw_pgid,
2622                               struct ceph_spg *spgid)
2623 {
2624         struct ceph_pg pgid;
2625         struct ceph_osds up, acting;
2626         int i;
2627
2628         WARN_ON(pi->id != raw_pgid->pool);
2629         raw_pg_to_pg(pi, raw_pgid, &pgid);
2630
2631         if (ceph_can_shift_osds(pi)) {
2632                 spgid->pgid = pgid; /* struct */
2633                 spgid->shard = CEPH_SPG_NOSHARD;
2634                 return true;
2635         }
2636
2637         ceph_pg_to_up_acting_osds(osdmap, pi, &pgid, &up, &acting);
2638         for (i = 0; i < acting.size; i++) {
2639                 if (acting.osds[i] == acting.primary) {
2640                         spgid->pgid = pgid; /* struct */
2641                         spgid->shard = i;
2642                         return true;
2643                 }
2644         }
2645
2646         return false;
2647 }
2648
2649 /*
2650  * Return acting primary for given PG, or -1 if none.
2651  */
2652 int ceph_pg_to_acting_primary(struct ceph_osdmap *osdmap,
2653                               const struct ceph_pg *raw_pgid)
2654 {
2655         struct ceph_pg_pool_info *pi;
2656         struct ceph_osds up, acting;
2657
2658         pi = ceph_pg_pool_by_id(osdmap, raw_pgid->pool);
2659         if (!pi)
2660                 return -1;
2661
2662         ceph_pg_to_up_acting_osds(osdmap, pi, raw_pgid, &up, &acting);
2663         return acting.primary;
2664 }
2665 EXPORT_SYMBOL(ceph_pg_to_acting_primary);