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