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 u64 ceph_pg_pool_flags(struct ceph_osdmap *map, u64 id)
 715 {
 716         struct ceph_pg_pool_info *pi;
 717
 718         pi = __lookup_pg_pool(&map->pg_pools, id);
 719         return pi ? pi->flags : 0;
 720 }
 721 EXPORT_SYMBOL(ceph_pg_pool_flags);
 722
 723 static void __remove_pg_pool(struct rb_root *root, struct ceph_pg_pool_info *pi)
 724 {
 725         rb_erase(&pi->node, root);
 726         kfree(pi->name);
 727         kfree(pi);
 728 }
 729
 730 static int decode_pool(void **p, void *end, struct ceph_pg_pool_info *pi)
 731 {
 732         u8 ev, cv;
 733         unsigned len, num;
 734         void *pool_end;
 735
 736         ceph_decode_need(p, end, 2 + 4, bad);
 737         ev = ceph_decode_8(p);  /* encoding version */
 738         cv = ceph_decode_8(p); /* compat version */
 739         if (ev < 5) {
 740                 pr_warn("got v %d < 5 cv %d of ceph_pg_pool\n", ev, cv);
 741                 return -EINVAL;
 742         }
 743         if (cv > 9) {
 744                 pr_warn("got v %d cv %d > 9 of ceph_pg_pool\n", ev, cv);
 745                 return -EINVAL;
 746         }
 747         len = ceph_decode_32(p);
 748         ceph_decode_need(p, end, len, bad);
 749         pool_end = *p + len;
 750
 751         pi->type = ceph_decode_8(p);
 752         pi->size = ceph_decode_8(p);
 753         pi->crush_ruleset = ceph_decode_8(p);
 754         pi->object_hash = ceph_decode_8(p);
 755
 756         pi->pg_num = ceph_decode_32(p);
 757         pi->pgp_num = ceph_decode_32(p);
 758
 759         *p += 4 + 4;  /* skip lpg* */
 760         *p += 4;      /* skip last_change */
 761         *p += 8 + 4;  /* skip snap_seq, snap_epoch */
 762
 763         /* skip snaps */
 764         num = ceph_decode_32(p);
 765         while (num--) {
 766                 *p += 8;  /* snapid key */
 767                 *p += 1 + 1; /* versions */
 768                 len = ceph_decode_32(p);
 769                 *p += len;
 770         }
 771
 772         /* skip removed_snaps */
 773         num = ceph_decode_32(p);
 774         *p += num * (8 + 8);
 775
 776         *p += 8;  /* skip auid */
 777         pi->flags = ceph_decode_64(p);
 778         *p += 4;  /* skip crash_replay_interval */
 779
 780         if (ev >= 7)
 781                 pi->min_size = ceph_decode_8(p);
 782         else
 783                 pi->min_size = pi->size - pi->size / 2;
 784
 785         if (ev >= 8)
 786                 *p += 8 + 8;  /* skip quota_max_* */
 787
 788         if (ev >= 9) {
 789                 /* skip tiers */
 790                 num = ceph_decode_32(p);
 791                 *p += num * 8;
 792
 793                 *p += 8;  /* skip tier_of */
 794                 *p += 1;  /* skip cache_mode */
 795
 796                 pi->read_tier = ceph_decode_64(p);
 797                 pi->write_tier = ceph_decode_64(p);
 798         } else {
 799                 pi->read_tier = -1;
 800                 pi->write_tier = -1;
 801         }
 802
 803         if (ev >= 10) {
 804                 /* skip properties */
 805                 num = ceph_decode_32(p);
 806                 while (num--) {
 807                         len = ceph_decode_32(p);
 808                         *p += len; /* key */
 809                         len = ceph_decode_32(p);
 810                         *p += len; /* val */
 811                 }
 812         }
 813
 814         if (ev >= 11) {
 815                 /* skip hit_set_params */
 816                 *p += 1 + 1; /* versions */
 817                 len = ceph_decode_32(p);
 818                 *p += len;
 819
 820                 *p += 4; /* skip hit_set_period */
 821                 *p += 4; /* skip hit_set_count */
 822         }
 823
 824         if (ev >= 12)
 825                 *p += 4; /* skip stripe_width */
 826
 827         if (ev >= 13) {
 828                 *p += 8; /* skip target_max_bytes */
 829                 *p += 8; /* skip target_max_objects */
 830                 *p += 4; /* skip cache_target_dirty_ratio_micro */
 831                 *p += 4; /* skip cache_target_full_ratio_micro */
 832                 *p += 4; /* skip cache_min_flush_age */
 833                 *p += 4; /* skip cache_min_evict_age */
 834         }
 835
 836         if (ev >=  14) {
 837                 /* skip erasure_code_profile */
 838                 len = ceph_decode_32(p);
 839                 *p += len;
 840         }
 841
 842         /*
 843          * last_force_op_resend_preluminous, will be overridden if the
 844          * map was encoded with RESEND_ON_SPLIT
 845          */
 846         if (ev >= 15)
 847                 pi->last_force_request_resend = ceph_decode_32(p);
 848         else
 849                 pi->last_force_request_resend = 0;
 850
 851         if (ev >= 16)
 852                 *p += 4; /* skip min_read_recency_for_promote */
 853
 854         if (ev >= 17)
 855                 *p += 8; /* skip expected_num_objects */
 856
 857         if (ev >= 19)
 858                 *p += 4; /* skip cache_target_dirty_high_ratio_micro */
 859
 860         if (ev >= 20)
 861                 *p += 4; /* skip min_write_recency_for_promote */
 862
 863         if (ev >= 21)
 864                 *p += 1; /* skip use_gmt_hitset */
 865
 866         if (ev >= 22)
 867                 *p += 1; /* skip fast_read */
 868
 869         if (ev >= 23) {
 870                 *p += 4; /* skip hit_set_grade_decay_rate */
 871                 *p += 4; /* skip hit_set_search_last_n */
 872         }
 873
 874         if (ev >= 24) {
 875                 /* skip opts */
 876                 *p += 1 + 1; /* versions */
 877                 len = ceph_decode_32(p);
 878                 *p += len;
 879         }
 880
 881         if (ev >= 25)
 882                 pi->last_force_request_resend = ceph_decode_32(p);
 883
 884         /* ignore the rest */
 885
 886         *p = pool_end;
 887         calc_pg_masks(pi);
 888         return 0;
 889
 890 bad:
 891         return -EINVAL;
 892 }
 893
 894 static int decode_pool_names(void **p, void *end, struct ceph_osdmap *map)
 895 {
 896         struct ceph_pg_pool_info *pi;
 897         u32 num, len;
 898         u64 pool;
 899
 900         ceph_decode_32_safe(p, end, num, bad);
 901         dout(" %d pool names\n", num);
 902         while (num--) {
 903                 ceph_decode_64_safe(p, end, pool, bad);
 904                 ceph_decode_32_safe(p, end, len, bad);
 905                 dout("  pool %llu len %d\n", pool, len);
 906                 ceph_decode_need(p, end, len, bad);
 907                 pi = __lookup_pg_pool(&map->pg_pools, pool);
 908                 if (pi) {
 909                         char *name = kstrndup(*p, len, GFP_NOFS);
 910
 911                         if (!name)
 912                                 return -ENOMEM;
 913                         kfree(pi->name);
 914                         pi->name = name;
 915                         dout("  name is %s\n", pi->name);
 916                 }
 917                 *p += len;
 918         }
 919         return 0;
 920
 921 bad:
 922         return -EINVAL;
 923 }
 924
 925 /*
 926  * osd map
 927  */
 928 struct ceph_osdmap *ceph_osdmap_alloc(void)
 929 {
 930         struct ceph_osdmap *map;
 931
 932         map = kzalloc(sizeof(*map), GFP_NOIO);
 933         if (!map)
 934                 return NULL;
 935
 936         map->pg_pools = RB_ROOT;
 937         map->pool_max = -1;
 938         map->pg_temp = RB_ROOT;
 939         map->primary_temp = RB_ROOT;
 940         map->pg_upmap = RB_ROOT;
 941         map->pg_upmap_items = RB_ROOT;
 942         mutex_init(&map->crush_workspace_mutex);
 943
 944         return map;
 945 }
 946
 947 void ceph_osdmap_destroy(struct ceph_osdmap *map)
 948 {
 949         dout("osdmap_destroy %p\n", map);
 950         if (map->crush)
 951                 crush_destroy(map->crush);
 952         while (!RB_EMPTY_ROOT(&map->pg_temp)) {
 953                 struct ceph_pg_mapping *pg =
 954                         rb_entry(rb_first(&map->pg_temp),
 955                                  struct ceph_pg_mapping, node);
 956                 erase_pg_mapping(&map->pg_temp, pg);
 957                 free_pg_mapping(pg);
 958         }
 959         while (!RB_EMPTY_ROOT(&map->primary_temp)) {
 960                 struct ceph_pg_mapping *pg =
 961                         rb_entry(rb_first(&map->primary_temp),
 962                                  struct ceph_pg_mapping, node);
 963                 erase_pg_mapping(&map->primary_temp, pg);
 964                 free_pg_mapping(pg);
 965         }
 966         while (!RB_EMPTY_ROOT(&map->pg_upmap)) {
 967                 struct ceph_pg_mapping *pg =
 968                         rb_entry(rb_first(&map->pg_upmap),
 969                                  struct ceph_pg_mapping, node);
 970                 rb_erase(&pg->node, &map->pg_upmap);
 971                 kfree(pg);
 972         }
 973         while (!RB_EMPTY_ROOT(&map->pg_upmap_items)) {
 974                 struct ceph_pg_mapping *pg =
 975                         rb_entry(rb_first(&map->pg_upmap_items),
 976                                  struct ceph_pg_mapping, node);
 977                 rb_erase(&pg->node, &map->pg_upmap_items);
 978                 kfree(pg);
 979         }
 980         while (!RB_EMPTY_ROOT(&map->pg_pools)) {
 981                 struct ceph_pg_pool_info *pi =
 982                         rb_entry(rb_first(&map->pg_pools),
 983                                  struct ceph_pg_pool_info, node);
 984                 __remove_pg_pool(&map->pg_pools, pi);
 985         }
 986         kfree(map->osd_state);
 987         kfree(map->osd_weight);
 988         kfree(map->osd_addr);
 989         kfree(map->osd_primary_affinity);
 990         kfree(map->crush_workspace);
 991         kfree(map);
 992 }
 993
 994 /*
 995  * Adjust max_osd value, (re)allocate arrays.
 996  *
 997  * The new elements are properly initialized.
 998  */
 999 static int osdmap_set_max_osd(struct ceph_osdmap *map, int max)
1000 {
1001         u32 *state;
1002         u32 *weight;
1003         struct ceph_entity_addr *addr;
1004         int i;
1005
1006         state = krealloc(map->osd_state, max*sizeof(*state), GFP_NOFS);
1007         if (!state)
1008                 return -ENOMEM;
1009         map->osd_state = state;
1010
1011         weight = krealloc(map->osd_weight, max*sizeof(*weight), GFP_NOFS);
1012         if (!weight)
1013                 return -ENOMEM;
1014         map->osd_weight = weight;
1015
1016         addr = krealloc(map->osd_addr, max*sizeof(*addr), GFP_NOFS);
1017         if (!addr)
1018                 return -ENOMEM;
1019         map->osd_addr = addr;
1020
1021         for (i = map->max_osd; i < max; i++) {
1022                 map->osd_state[i] = 0;
1023                 map->osd_weight[i] = CEPH_OSD_OUT;
1024                 memset(map->osd_addr + i, 0, sizeof(*map->osd_addr));
1025         }
1026
1027         if (map->osd_primary_affinity) {
1028                 u32 *affinity;
1029
1030                 affinity = krealloc(map->osd_primary_affinity,
1031                                     max*sizeof(*affinity), GFP_NOFS);
1032                 if (!affinity)
1033                         return -ENOMEM;
1034                 map->osd_primary_affinity = affinity;
1035
1036                 for (i = map->max_osd; i < max; i++)
1037                         map->osd_primary_affinity[i] =
1038                             CEPH_OSD_DEFAULT_PRIMARY_AFFINITY;
1039         }
1040
1041         map->max_osd = max;
1042
1043         return 0;
1044 }
1045
1046 static int osdmap_set_crush(struct ceph_osdmap *map, struct crush_map *crush)
1047 {
1048         void *workspace;
1049         size_t work_size;
1050
1051         if (IS_ERR(crush))
1052                 return PTR_ERR(crush);
1053
1054         work_size = crush_work_size(crush, CEPH_PG_MAX_SIZE);
1055         dout("%s work_size %zu bytes\n", __func__, work_size);
1056         workspace = kmalloc(work_size, GFP_NOIO);
1057         if (!workspace) {
1058                 crush_destroy(crush);
1059                 return -ENOMEM;
1060         }
1061         crush_init_workspace(crush, workspace);
1062
1063         if (map->crush)
1064                 crush_destroy(map->crush);
1065         kfree(map->crush_workspace);
1066         map->crush = crush;
1067         map->crush_workspace = workspace;
1068         return 0;
1069 }
1070
1071 #define OSDMAP_WRAPPER_COMPAT_VER       7
1072 #define OSDMAP_CLIENT_DATA_COMPAT_VER   1
1073
1074 /*
1075  * Return 0 or error.  On success, *v is set to 0 for old (v6) osdmaps,
1076  * to struct_v of the client_data section for new (v7 and above)
1077  * osdmaps.
1078  */
1079 static int get_osdmap_client_data_v(void **p, void *end,
1080                                     const char *prefix, u8 *v)
1081 {
1082         u8 struct_v;
1083
1084         ceph_decode_8_safe(p, end, struct_v, e_inval);
1085         if (struct_v >= 7) {
1086                 u8 struct_compat;
1087
1088                 ceph_decode_8_safe(p, end, struct_compat, e_inval);
1089                 if (struct_compat > OSDMAP_WRAPPER_COMPAT_VER) {
1090                         pr_warn("got v %d cv %d > %d of %s ceph_osdmap\n",
1091                                 struct_v, struct_compat,
1092                                 OSDMAP_WRAPPER_COMPAT_VER, prefix);
1093                         return -EINVAL;
1094                 }
1095                 *p += 4; /* ignore wrapper struct_len */
1096
1097                 ceph_decode_8_safe(p, end, struct_v, e_inval);
1098                 ceph_decode_8_safe(p, end, struct_compat, e_inval);
1099                 if (struct_compat > OSDMAP_CLIENT_DATA_COMPAT_VER) {
1100                         pr_warn("got v %d cv %d > %d of %s ceph_osdmap client data\n",
1101                                 struct_v, struct_compat,
1102                                 OSDMAP_CLIENT_DATA_COMPAT_VER, prefix);
1103                         return -EINVAL;
1104                 }
1105                 *p += 4; /* ignore client data struct_len */
1106         } else {
1107                 u16 version;
1108
1109                 *p -= 1;
1110                 ceph_decode_16_safe(p, end, version, e_inval);
1111                 if (version < 6) {
1112                         pr_warn("got v %d < 6 of %s ceph_osdmap\n",
1113                                 version, prefix);
1114                         return -EINVAL;
1115                 }
1116
1117                 /* old osdmap enconding */
1118                 struct_v = 0;
1119         }
1120
1121         *v = struct_v;
1122         return 0;
1123
1124 e_inval:
1125         return -EINVAL;
1126 }
1127
1128 static int __decode_pools(void **p, void *end, struct ceph_osdmap *map,
1129                           bool incremental)
1130 {
1131         u32 n;
1132
1133         ceph_decode_32_safe(p, end, n, e_inval);
1134         while (n--) {
1135                 struct ceph_pg_pool_info *pi;
1136                 u64 pool;
1137                 int ret;
1138
1139                 ceph_decode_64_safe(p, end, pool, e_inval);
1140
1141                 pi = __lookup_pg_pool(&map->pg_pools, pool);
1142                 if (!incremental || !pi) {
1143                         pi = kzalloc(sizeof(*pi), GFP_NOFS);
1144                         if (!pi)
1145                                 return -ENOMEM;
1146
1147                         pi->id = pool;
1148
1149                         ret = __insert_pg_pool(&map->pg_pools, pi);
1150                         if (ret) {
1151                                 kfree(pi);
1152                                 return ret;
1153                         }
1154                 }
1155
1156                 ret = decode_pool(p, end, pi);
1157                 if (ret)
1158                         return ret;
1159         }
1160
1161         return 0;
1162
1163 e_inval:
1164         return -EINVAL;
1165 }
1166
1167 static int decode_pools(void **p, void *end, struct ceph_osdmap *map)
1168 {
1169         return __decode_pools(p, end, map, false);
1170 }
1171
1172 static int decode_new_pools(void **p, void *end, struct ceph_osdmap *map)
1173 {
1174         return __decode_pools(p, end, map, true);
1175 }
1176
1177 typedef struct ceph_pg_mapping *(*decode_mapping_fn_t)(void **, void *, bool);
1178
1179 static int decode_pg_mapping(void **p, void *end, struct rb_root *mapping_root,
1180                              decode_mapping_fn_t fn, bool incremental)
1181 {
1182         u32 n;
1183
1184         WARN_ON(!incremental && !fn);
1185
1186         ceph_decode_32_safe(p, end, n, e_inval);
1187         while (n--) {
1188                 struct ceph_pg_mapping *pg;
1189                 struct ceph_pg pgid;
1190                 int ret;
1191
1192                 ret = ceph_decode_pgid(p, end, &pgid);
1193                 if (ret)
1194                         return ret;
1195
1196                 pg = lookup_pg_mapping(mapping_root, &pgid);
1197                 if (pg) {
1198                         WARN_ON(!incremental);
1199                         erase_pg_mapping(mapping_root, pg);
1200                         free_pg_mapping(pg);
1201                 }
1202
1203                 if (fn) {
1204                         pg = fn(p, end, incremental);
1205                         if (IS_ERR(pg))
1206                                 return PTR_ERR(pg);
1207
1208                         if (pg) {
1209                                 pg->pgid = pgid; /* struct */
1210                                 insert_pg_mapping(mapping_root, pg);
1211                         }
1212                 }
1213         }
1214
1215         return 0;
1216
1217 e_inval:
1218         return -EINVAL;
1219 }
1220
1221 static struct ceph_pg_mapping *__decode_pg_temp(void **p, void *end,
1222                                                 bool incremental)
1223 {
1224         struct ceph_pg_mapping *pg;
1225         u32 len, i;
1226
1227         ceph_decode_32_safe(p, end, len, e_inval);
1228         if (len == 0 && incremental)
1229                 return NULL;    /* new_pg_temp: [] to remove */
1230         if (len > (SIZE_MAX - sizeof(*pg)) / sizeof(u32))
1231                 return ERR_PTR(-EINVAL);
1232
1233         ceph_decode_need(p, end, len * sizeof(u32), e_inval);
1234         pg = alloc_pg_mapping(len * sizeof(u32));
1235         if (!pg)
1236                 return ERR_PTR(-ENOMEM);
1237
1238         pg->pg_temp.len = len;
1239         for (i = 0; i < len; i++)
1240                 pg->pg_temp.osds[i] = ceph_decode_32(p);
1241
1242         return pg;
1243
1244 e_inval:
1245         return ERR_PTR(-EINVAL);
1246 }
1247
1248 static int decode_pg_temp(void **p, void *end, struct ceph_osdmap *map)
1249 {
1250         return decode_pg_mapping(p, end, &map->pg_temp, __decode_pg_temp,
1251                                  false);
1252 }
1253
1254 static int decode_new_pg_temp(void **p, void *end, struct ceph_osdmap *map)
1255 {
1256         return decode_pg_mapping(p, end, &map->pg_temp, __decode_pg_temp,
1257                                  true);
1258 }
1259
1260 static struct ceph_pg_mapping *__decode_primary_temp(void **p, void *end,
1261                                                      bool incremental)
1262 {
1263         struct ceph_pg_mapping *pg;
1264         u32 osd;
1265
1266         ceph_decode_32_safe(p, end, osd, e_inval);
1267         if (osd == (u32)-1 && incremental)
1268                 return NULL;    /* new_primary_temp: -1 to remove */
1269
1270         pg = alloc_pg_mapping(0);
1271         if (!pg)
1272                 return ERR_PTR(-ENOMEM);
1273
1274         pg->primary_temp.osd = osd;
1275         return pg;
1276
1277 e_inval:
1278         return ERR_PTR(-EINVAL);
1279 }
1280
1281 static int decode_primary_temp(void **p, void *end, struct ceph_osdmap *map)
1282 {
1283         return decode_pg_mapping(p, end, &map->primary_temp,
1284                                  __decode_primary_temp, false);
1285 }
1286
1287 static int decode_new_primary_temp(void **p, void *end,
1288                                    struct ceph_osdmap *map)
1289 {
1290         return decode_pg_mapping(p, end, &map->primary_temp,
1291                                  __decode_primary_temp, true);
1292 }
1293
1294 u32 ceph_get_primary_affinity(struct ceph_osdmap *map, int osd)
1295 {
1296         BUG_ON(osd >= map->max_osd);
1297
1298         if (!map->osd_primary_affinity)
1299                 return CEPH_OSD_DEFAULT_PRIMARY_AFFINITY;
1300
1301         return map->osd_primary_affinity[osd];
1302 }
1303
1304 static int set_primary_affinity(struct ceph_osdmap *map, int osd, u32 aff)
1305 {
1306         BUG_ON(osd >= map->max_osd);
1307
1308         if (!map->osd_primary_affinity) {
1309                 int i;
1310
1311                 map->osd_primary_affinity = kmalloc_array(map->max_osd,
1312                                                           sizeof(u32),
1313                                                           GFP_NOFS);
1314                 if (!map->osd_primary_affinity)
1315                         return -ENOMEM;
1316
1317                 for (i = 0; i < map->max_osd; i++)
1318                         map->osd_primary_affinity[i] =
1319                             CEPH_OSD_DEFAULT_PRIMARY_AFFINITY;
1320         }
1321
1322         map->osd_primary_affinity[osd] = aff;
1323
1324         return 0;
1325 }
1326
1327 static int decode_primary_affinity(void **p, void *end,
1328                                    struct ceph_osdmap *map)
1329 {
1330         u32 len, i;
1331
1332         ceph_decode_32_safe(p, end, len, e_inval);
1333         if (len == 0) {
1334                 kfree(map->osd_primary_affinity);
1335                 map->osd_primary_affinity = NULL;
1336                 return 0;
1337         }
1338         if (len != map->max_osd)
1339                 goto e_inval;
1340
1341         ceph_decode_need(p, end, map->max_osd*sizeof(u32), e_inval);
1342
1343         for (i = 0; i < map->max_osd; i++) {
1344                 int ret;
1345
1346                 ret = set_primary_affinity(map, i, ceph_decode_32(p));
1347                 if (ret)
1348                         return ret;
1349         }
1350
1351         return 0;
1352
1353 e_inval:
1354         return -EINVAL;
1355 }
1356
1357 static int decode_new_primary_affinity(void **p, void *end,
1358                                        struct ceph_osdmap *map)
1359 {
1360         u32 n;
1361
1362         ceph_decode_32_safe(p, end, n, e_inval);
1363         while (n--) {
1364                 u32 osd, aff;
1365                 int ret;
1366
1367                 ceph_decode_32_safe(p, end, osd, e_inval);
1368                 ceph_decode_32_safe(p, end, aff, e_inval);
1369
1370                 ret = set_primary_affinity(map, osd, aff);
1371                 if (ret)
1372                         return ret;
1373
1374                 pr_info("osd%d primary-affinity 0x%x\n", osd, aff);
1375         }
1376
1377         return 0;
1378
1379 e_inval:
1380         return -EINVAL;
1381 }
1382
1383 static struct ceph_pg_mapping *__decode_pg_upmap(void **p, void *end,
1384                                                  bool __unused)
1385 {
1386         return __decode_pg_temp(p, end, false);
1387 }
1388
1389 static int decode_pg_upmap(void **p, void *end, struct ceph_osdmap *map)
1390 {
1391         return decode_pg_mapping(p, end, &map->pg_upmap, __decode_pg_upmap,
1392                                  false);
1393 }
1394
1395 static int decode_new_pg_upmap(void **p, void *end, struct ceph_osdmap *map)
1396 {
1397         return decode_pg_mapping(p, end, &map->pg_upmap, __decode_pg_upmap,
1398                                  true);
1399 }
1400
1401 static int decode_old_pg_upmap(void **p, void *end, struct ceph_osdmap *map)
1402 {
1403         return decode_pg_mapping(p, end, &map->pg_upmap, NULL, true);
1404 }
1405
1406 static struct ceph_pg_mapping *__decode_pg_upmap_items(void **p, void *end,
1407                                                        bool __unused)
1408 {
1409         struct ceph_pg_mapping *pg;
1410         u32 len, i;
1411
1412         ceph_decode_32_safe(p, end, len, e_inval);
1413         if (len > (SIZE_MAX - sizeof(*pg)) / (2 * sizeof(u32)))
1414                 return ERR_PTR(-EINVAL);
1415
1416         ceph_decode_need(p, end, 2 * len * sizeof(u32), e_inval);
1417         pg = alloc_pg_mapping(2 * len * sizeof(u32));
1418         if (!pg)
1419                 return ERR_PTR(-ENOMEM);
1420
1421         pg->pg_upmap_items.len = len;
1422         for (i = 0; i < len; i++) {
1423                 pg->pg_upmap_items.from_to[i][0] = ceph_decode_32(p);
1424                 pg->pg_upmap_items.from_to[i][1] = ceph_decode_32(p);
1425         }
1426
1427         return pg;
1428
1429 e_inval:
1430         return ERR_PTR(-EINVAL);
1431 }
1432
1433 static int decode_pg_upmap_items(void **p, void *end, struct ceph_osdmap *map)
1434 {
1435         return decode_pg_mapping(p, end, &map->pg_upmap_items,
1436                                  __decode_pg_upmap_items, false);
1437 }
1438
1439 static int decode_new_pg_upmap_items(void **p, void *end,
1440                                      struct ceph_osdmap *map)
1441 {
1442         return decode_pg_mapping(p, end, &map->pg_upmap_items,
1443                                  __decode_pg_upmap_items, true);
1444 }
1445
1446 static int decode_old_pg_upmap_items(void **p, void *end,
1447                                      struct ceph_osdmap *map)
1448 {
1449         return decode_pg_mapping(p, end, &map->pg_upmap_items, NULL, true);
1450 }
1451
1452 /*
1453  * decode a full map.
1454  */
1455 static int osdmap_decode(void **p, void *end, struct ceph_osdmap *map)
1456 {
1457         u8 struct_v;
1458         u32 epoch = 0;
1459         void *start = *p;
1460         u32 max;
1461         u32 len, i;
1462         int err;
1463
1464         dout("%s %p to %p len %d\n", __func__, *p, end, (int)(end - *p));
1465
1466         err = get_osdmap_client_data_v(p, end, "full", &struct_v);
1467         if (err)
1468                 goto bad;
1469
1470         /* fsid, epoch, created, modified */
1471         ceph_decode_need(p, end, sizeof(map->fsid) + sizeof(u32) +
1472                          sizeof(map->created) + sizeof(map->modified), e_inval);
1473         ceph_decode_copy(p, &map->fsid, sizeof(map->fsid));
1474         epoch = map->epoch = ceph_decode_32(p);
1475         ceph_decode_copy(p, &map->created, sizeof(map->created));
1476         ceph_decode_copy(p, &map->modified, sizeof(map->modified));
1477
1478         /* pools */
1479         err = decode_pools(p, end, map);
1480         if (err)
1481                 goto bad;
1482
1483         /* pool_name */
1484         err = decode_pool_names(p, end, map);
1485         if (err)
1486                 goto bad;
1487
1488         ceph_decode_32_safe(p, end, map->pool_max, e_inval);
1489
1490         ceph_decode_32_safe(p, end, map->flags, e_inval);
1491
1492         /* max_osd */
1493         ceph_decode_32_safe(p, end, max, e_inval);
1494
1495         /* (re)alloc osd arrays */
1496         err = osdmap_set_max_osd(map, max);
1497         if (err)
1498                 goto bad;
1499
1500         /* osd_state, osd_weight, osd_addrs->client_addr */
1501         ceph_decode_need(p, end, 3*sizeof(u32) +
1502                          map->max_osd*((struct_v >= 5 ? sizeof(u32) :
1503                                                         sizeof(u8)) +
1504                                        sizeof(*map->osd_weight) +
1505                                        sizeof(*map->osd_addr)), e_inval);
1506
1507         if (ceph_decode_32(p) != map->max_osd)
1508                 goto e_inval;
1509
1510         if (struct_v >= 5) {
1511                 for (i = 0; i < map->max_osd; i++)
1512                         map->osd_state[i] = ceph_decode_32(p);
1513         } else {
1514                 for (i = 0; i < map->max_osd; i++)
1515                         map->osd_state[i] = ceph_decode_8(p);
1516         }
1517
1518         if (ceph_decode_32(p) != map->max_osd)
1519                 goto e_inval;
1520
1521         for (i = 0; i < map->max_osd; i++)
1522                 map->osd_weight[i] = ceph_decode_32(p);
1523
1524         if (ceph_decode_32(p) != map->max_osd)
1525                 goto e_inval;
1526
1527         ceph_decode_copy(p, map->osd_addr, map->max_osd*sizeof(*map->osd_addr));
1528         for (i = 0; i < map->max_osd; i++)
1529                 ceph_decode_addr(&map->osd_addr[i]);
1530
1531         /* pg_temp */
1532         err = decode_pg_temp(p, end, map);
1533         if (err)
1534                 goto bad;
1535
1536         /* primary_temp */
1537         if (struct_v >= 1) {
1538                 err = decode_primary_temp(p, end, map);
1539                 if (err)
1540                         goto bad;
1541         }
1542
1543         /* primary_affinity */
1544         if (struct_v >= 2) {
1545                 err = decode_primary_affinity(p, end, map);
1546                 if (err)
1547                         goto bad;
1548         } else {
1549                 WARN_ON(map->osd_primary_affinity);
1550         }
1551
1552         /* crush */
1553         ceph_decode_32_safe(p, end, len, e_inval);
1554         err = osdmap_set_crush(map, crush_decode(*p, min(*p + len, end)));
1555         if (err)
1556                 goto bad;
1557
1558         *p += len;
1559         if (struct_v >= 3) {
1560                 /* erasure_code_profiles */
1561                 ceph_decode_skip_map_of_map(p, end, string, string, string,
1562                                             e_inval);
1563         }
1564
1565         if (struct_v >= 4) {
1566                 err = decode_pg_upmap(p, end, map);
1567                 if (err)
1568                         goto bad;
1569
1570                 err = decode_pg_upmap_items(p, end, map);
1571                 if (err)
1572                         goto bad;
1573         } else {
1574                 WARN_ON(!RB_EMPTY_ROOT(&map->pg_upmap));
1575                 WARN_ON(!RB_EMPTY_ROOT(&map->pg_upmap_items));
1576         }
1577
1578         /* ignore the rest */
1579         *p = end;
1580
1581         dout("full osdmap epoch %d max_osd %d\n", map->epoch, map->max_osd);
1582         return 0;
1583
1584 e_inval:
1585         err = -EINVAL;
1586 bad:
1587         pr_err("corrupt full osdmap (%d) epoch %d off %d (%p of %p-%p)\n",
1588                err, epoch, (int)(*p - start), *p, start, end);
1589         print_hex_dump(KERN_DEBUG, "osdmap: ",
1590                        DUMP_PREFIX_OFFSET, 16, 1,
1591                        start, end - start, true);
1592         return err;
1593 }
1594
1595 /*
1596  * Allocate and decode a full map.
1597  */
1598 struct ceph_osdmap *ceph_osdmap_decode(void **p, void *end)
1599 {
1600         struct ceph_osdmap *map;
1601         int ret;
1602
1603         map = ceph_osdmap_alloc();
1604         if (!map)
1605                 return ERR_PTR(-ENOMEM);
1606
1607         ret = osdmap_decode(p, end, map);
1608         if (ret) {
1609                 ceph_osdmap_destroy(map);
1610                 return ERR_PTR(ret);
1611         }
1612
1613         return map;
1614 }
1615
1616 /*
1617  * Encoding order is (new_up_client, new_state, new_weight).  Need to
1618  * apply in the (new_weight, new_state, new_up_client) order, because
1619  * an incremental map may look like e.g.
1620  *
1621  *     new_up_client: { osd=6, addr=... } # set osd_state and addr
1622  *     new_state: { osd=6, xorstate=EXISTS } # clear osd_state
1623  */
1624 static int decode_new_up_state_weight(void **p, void *end, u8 struct_v,
1625                                       struct ceph_osdmap *map)
1626 {
1627         void *new_up_client;
1628         void *new_state;
1629         void *new_weight_end;
1630         u32 len;
1631
1632         new_up_client = *p;
1633         ceph_decode_32_safe(p, end, len, e_inval);
1634         len *= sizeof(u32) + sizeof(struct ceph_entity_addr);
1635         ceph_decode_need(p, end, len, e_inval);
1636         *p += len;
1637
1638         new_state = *p;
1639         ceph_decode_32_safe(p, end, len, e_inval);
1640         len *= sizeof(u32) + (struct_v >= 5 ? sizeof(u32) : sizeof(u8));
1641         ceph_decode_need(p, end, len, e_inval);
1642         *p += len;
1643
1644         /* new_weight */
1645         ceph_decode_32_safe(p, end, len, e_inval);
1646         while (len--) {
1647                 s32 osd;
1648                 u32 w;
1649
1650                 ceph_decode_need(p, end, 2*sizeof(u32), e_inval);
1651                 osd = ceph_decode_32(p);
1652                 w = ceph_decode_32(p);
1653                 BUG_ON(osd >= map->max_osd);
1654                 pr_info("osd%d weight 0x%x %s\n", osd, w,
1655                      w == CEPH_OSD_IN ? "(in)" :
1656                      (w == CEPH_OSD_OUT ? "(out)" : ""));
1657                 map->osd_weight[osd] = w;
1658
1659                 /*
1660                  * If we are marking in, set the EXISTS, and clear the
1661                  * AUTOOUT and NEW bits.
1662                  */
1663                 if (w) {
1664                         map->osd_state[osd] |= CEPH_OSD_EXISTS;
1665                         map->osd_state[osd] &= ~(CEPH_OSD_AUTOOUT |
1666                                                  CEPH_OSD_NEW);
1667                 }
1668         }
1669         new_weight_end = *p;
1670
1671         /* new_state (up/down) */
1672         *p = new_state;
1673         len = ceph_decode_32(p);
1674         while (len--) {
1675                 s32 osd;
1676                 u32 xorstate;
1677                 int ret;
1678
1679                 osd = ceph_decode_32(p);
1680                 if (struct_v >= 5)
1681                         xorstate = ceph_decode_32(p);
1682                 else
1683                         xorstate = ceph_decode_8(p);
1684                 if (xorstate == 0)
1685                         xorstate = CEPH_OSD_UP;
1686                 BUG_ON(osd >= map->max_osd);
1687                 if ((map->osd_state[osd] & CEPH_OSD_UP) &&
1688                     (xorstate & CEPH_OSD_UP))
1689                         pr_info("osd%d down\n", osd);
1690                 if ((map->osd_state[osd] & CEPH_OSD_EXISTS) &&
1691                     (xorstate & CEPH_OSD_EXISTS)) {
1692                         pr_info("osd%d does not exist\n", osd);
1693                         ret = set_primary_affinity(map, osd,
1694                                                    CEPH_OSD_DEFAULT_PRIMARY_AFFINITY);
1695                         if (ret)
1696                                 return ret;
1697                         memset(map->osd_addr + osd, 0, sizeof(*map->osd_addr));
1698                         map->osd_state[osd] = 0;
1699                 } else {
1700                         map->osd_state[osd] ^= xorstate;
1701                 }
1702         }
1703
1704         /* new_up_client */
1705         *p = new_up_client;
1706         len = ceph_decode_32(p);
1707         while (len--) {
1708                 s32 osd;
1709                 struct ceph_entity_addr addr;
1710
1711                 osd = ceph_decode_32(p);
1712                 ceph_decode_copy(p, &addr, sizeof(addr));
1713                 ceph_decode_addr(&addr);
1714                 BUG_ON(osd >= map->max_osd);
1715                 pr_info("osd%d up\n", osd);
1716                 map->osd_state[osd] |= CEPH_OSD_EXISTS | CEPH_OSD_UP;
1717                 map->osd_addr[osd] = addr;
1718         }
1719
1720         *p = new_weight_end;
1721         return 0;
1722
1723 e_inval:
1724         return -EINVAL;
1725 }
1726
1727 /*
1728  * decode and apply an incremental map update.
1729  */
1730 struct ceph_osdmap *osdmap_apply_incremental(void **p, void *end,
1731                                              struct ceph_osdmap *map)
1732 {
1733         struct ceph_fsid fsid;
1734         u32 epoch = 0;
1735         struct ceph_timespec modified;
1736         s32 len;
1737         u64 pool;
1738         __s64 new_pool_max;
1739         __s32 new_flags, max;
1740         void *start = *p;
1741         int err;
1742         u8 struct_v;
1743
1744         dout("%s %p to %p len %d\n", __func__, *p, end, (int)(end - *p));
1745
1746         err = get_osdmap_client_data_v(p, end, "inc", &struct_v);
1747         if (err)
1748                 goto bad;
1749
1750         /* fsid, epoch, modified, new_pool_max, new_flags */
1751         ceph_decode_need(p, end, sizeof(fsid) + sizeof(u32) + sizeof(modified) +
1752                          sizeof(u64) + sizeof(u32), e_inval);
1753         ceph_decode_copy(p, &fsid, sizeof(fsid));
1754         epoch = ceph_decode_32(p);
1755         BUG_ON(epoch != map->epoch+1);
1756         ceph_decode_copy(p, &modified, sizeof(modified));
1757         new_pool_max = ceph_decode_64(p);
1758         new_flags = ceph_decode_32(p);
1759
1760         /* full map? */
1761         ceph_decode_32_safe(p, end, len, e_inval);
1762         if (len > 0) {
1763                 dout("apply_incremental full map len %d, %p to %p\n",
1764                      len, *p, end);
1765                 return ceph_osdmap_decode(p, min(*p+len, end));
1766         }
1767
1768         /* new crush? */
1769         ceph_decode_32_safe(p, end, len, e_inval);
1770         if (len > 0) {
1771                 err = osdmap_set_crush(map,
1772                                        crush_decode(*p, min(*p + len, end)));
1773                 if (err)
1774                         goto bad;
1775                 *p += len;
1776         }
1777
1778         /* new flags? */
1779         if (new_flags >= 0)
1780                 map->flags = new_flags;
1781         if (new_pool_max >= 0)
1782                 map->pool_max = new_pool_max;
1783
1784         /* new max? */
1785         ceph_decode_32_safe(p, end, max, e_inval);
1786         if (max >= 0) {
1787                 err = osdmap_set_max_osd(map, max);
1788                 if (err)
1789                         goto bad;
1790         }
1791
1792         map->epoch++;
1793         map->modified = modified;
1794
1795         /* new_pools */
1796         err = decode_new_pools(p, end, map);
1797         if (err)
1798                 goto bad;
1799
1800         /* new_pool_names */
1801         err = decode_pool_names(p, end, map);
1802         if (err)
1803                 goto bad;
1804
1805         /* old_pool */
1806         ceph_decode_32_safe(p, end, len, e_inval);
1807         while (len--) {
1808                 struct ceph_pg_pool_info *pi;
1809
1810                 ceph_decode_64_safe(p, end, pool, e_inval);
1811                 pi = __lookup_pg_pool(&map->pg_pools, pool);
1812                 if (pi)
1813                         __remove_pg_pool(&map->pg_pools, pi);
1814         }
1815
1816         /* new_up_client, new_state, new_weight */
1817         err = decode_new_up_state_weight(p, end, struct_v, map);
1818         if (err)
1819                 goto bad;
1820
1821         /* new_pg_temp */
1822         err = decode_new_pg_temp(p, end, map);
1823         if (err)
1824                 goto bad;
1825
1826         /* new_primary_temp */
1827         if (struct_v >= 1) {
1828                 err = decode_new_primary_temp(p, end, map);
1829                 if (err)
1830                         goto bad;
1831         }
1832
1833         /* new_primary_affinity */
1834         if (struct_v >= 2) {
1835                 err = decode_new_primary_affinity(p, end, map);
1836                 if (err)
1837                         goto bad;
1838         }
1839
1840         if (struct_v >= 3) {
1841                 /* new_erasure_code_profiles */
1842                 ceph_decode_skip_map_of_map(p, end, string, string, string,
1843                                             e_inval);
1844                 /* old_erasure_code_profiles */
1845                 ceph_decode_skip_set(p, end, string, e_inval);
1846         }
1847
1848         if (struct_v >= 4) {
1849                 err = decode_new_pg_upmap(p, end, map);
1850                 if (err)
1851                         goto bad;
1852
1853                 err = decode_old_pg_upmap(p, end, map);
1854                 if (err)
1855                         goto bad;
1856
1857                 err = decode_new_pg_upmap_items(p, end, map);
1858                 if (err)
1859                         goto bad;
1860
1861                 err = decode_old_pg_upmap_items(p, end, map);
1862                 if (err)
1863                         goto bad;
1864         }
1865
1866         /* ignore the rest */
1867         *p = end;
1868
1869         dout("inc osdmap epoch %d max_osd %d\n", map->epoch, map->max_osd);
1870         return map;
1871
1872 e_inval:
1873         err = -EINVAL;
1874 bad:
1875         pr_err("corrupt inc osdmap (%d) epoch %d off %d (%p of %p-%p)\n",
1876                err, epoch, (int)(*p - start), *p, start, end);
1877         print_hex_dump(KERN_DEBUG, "osdmap: ",
1878                        DUMP_PREFIX_OFFSET, 16, 1,
1879                        start, end - start, true);
1880         return ERR_PTR(err);
1881 }
1882
1883 void ceph_oloc_copy(struct ceph_object_locator *dest,
1884                     const struct ceph_object_locator *src)
1885 {
1886         ceph_oloc_destroy(dest);
1887
1888         dest->pool = src->pool;
1889         if (src->pool_ns)
1890                 dest->pool_ns = ceph_get_string(src->pool_ns);
1891         else
1892                 dest->pool_ns = NULL;
1893 }
1894 EXPORT_SYMBOL(ceph_oloc_copy);
1895
1896 void ceph_oloc_destroy(struct ceph_object_locator *oloc)
1897 {
1898         ceph_put_string(oloc->pool_ns);
1899 }
1900 EXPORT_SYMBOL(ceph_oloc_destroy);
1901
1902 void ceph_oid_copy(struct ceph_object_id *dest,
1903                    const struct ceph_object_id *src)
1904 {
1905         ceph_oid_destroy(dest);
1906
1907         if (src->name != src->inline_name) {
1908                 /* very rare, see ceph_object_id definition */
1909                 dest->name = kmalloc(src->name_len + 1,
1910                                      GFP_NOIO | __GFP_NOFAIL);
1911         } else {
1912                 dest->name = dest->inline_name;
1913         }
1914         memcpy(dest->name, src->name, src->name_len + 1);
1915         dest->name_len = src->name_len;
1916 }
1917 EXPORT_SYMBOL(ceph_oid_copy);
1918
1919 static __printf(2, 0)
1920 int oid_printf_vargs(struct ceph_object_id *oid, const char *fmt, va_list ap)
1921 {
1922         int len;
1923
1924         WARN_ON(!ceph_oid_empty(oid));
1925
1926         len = vsnprintf(oid->inline_name, sizeof(oid->inline_name), fmt, ap);
1927         if (len >= sizeof(oid->inline_name))
1928                 return len;
1929
1930         oid->name_len = len;
1931         return 0;
1932 }
1933
1934 /*
1935  * If oid doesn't fit into inline buffer, BUG.
1936  */
1937 void ceph_oid_printf(struct ceph_object_id *oid, const char *fmt, ...)
1938 {
1939         va_list ap;
1940
1941         va_start(ap, fmt);
1942         BUG_ON(oid_printf_vargs(oid, fmt, ap));
1943         va_end(ap);
1944 }
1945 EXPORT_SYMBOL(ceph_oid_printf);
1946
1947 static __printf(3, 0)
1948 int oid_aprintf_vargs(struct ceph_object_id *oid, gfp_t gfp,
1949                       const char *fmt, va_list ap)
1950 {
1951         va_list aq;
1952         int len;
1953
1954         va_copy(aq, ap);
1955         len = oid_printf_vargs(oid, fmt, aq);
1956         va_end(aq);
1957
1958         if (len) {
1959                 char *external_name;
1960
1961                 external_name = kmalloc(len + 1, gfp);
1962                 if (!external_name)
1963                         return -ENOMEM;
1964
1965                 oid->name = external_name;
1966                 WARN_ON(vsnprintf(oid->name, len + 1, fmt, ap) != len);
1967                 oid->name_len = len;
1968         }
1969
1970         return 0;
1971 }
1972
1973 /*
1974  * If oid doesn't fit into inline buffer, allocate.
1975  */
1976 int ceph_oid_aprintf(struct ceph_object_id *oid, gfp_t gfp,
1977                      const char *fmt, ...)
1978 {
1979         va_list ap;
1980         int ret;
1981
1982         va_start(ap, fmt);
1983         ret = oid_aprintf_vargs(oid, gfp, fmt, ap);
1984         va_end(ap);
1985
1986         return ret;
1987 }
1988 EXPORT_SYMBOL(ceph_oid_aprintf);
1989
1990 void ceph_oid_destroy(struct ceph_object_id *oid)
1991 {
1992         if (oid->name != oid->inline_name)
1993                 kfree(oid->name);
1994 }
1995 EXPORT_SYMBOL(ceph_oid_destroy);
1996
1997 /*
1998  * osds only
1999  */
2000 static bool __osds_equal(const struct ceph_osds *lhs,
2001                          const struct ceph_osds *rhs)
2002 {
2003         if (lhs->size == rhs->size &&
2004             !memcmp(lhs->osds, rhs->osds, rhs->size * sizeof(rhs->osds[0])))
2005                 return true;
2006
2007         return false;
2008 }
2009
2010 /*
2011  * osds + primary
2012  */
2013 static bool osds_equal(const struct ceph_osds *lhs,
2014                        const struct ceph_osds *rhs)
2015 {
2016         if (__osds_equal(lhs, rhs) &&
2017             lhs->primary == rhs->primary)
2018                 return true;
2019
2020         return false;
2021 }
2022
2023 static bool osds_valid(const struct ceph_osds *set)
2024 {
2025         /* non-empty set */
2026         if (set->size > 0 && set->primary >= 0)
2027                 return true;
2028
2029         /* empty can_shift_osds set */
2030         if (!set->size && set->primary == -1)
2031                 return true;
2032
2033         /* empty !can_shift_osds set - all NONE */
2034         if (set->size > 0 && set->primary == -1) {
2035                 int i;
2036
2037                 for (i = 0; i < set->size; i++) {
2038                         if (set->osds[i] != CRUSH_ITEM_NONE)
2039                                 break;
2040                 }
2041                 if (i == set->size)
2042                         return true;
2043         }
2044
2045         return false;
2046 }
2047
2048 void ceph_osds_copy(struct ceph_osds *dest, const struct ceph_osds *src)
2049 {
2050         memcpy(dest->osds, src->osds, src->size * sizeof(src->osds[0]));
2051         dest->size = src->size;
2052         dest->primary = src->primary;
2053 }
2054
2055 bool ceph_pg_is_split(const struct ceph_pg *pgid, u32 old_pg_num,
2056                       u32 new_pg_num)
2057 {
2058         int old_bits = calc_bits_of(old_pg_num);
2059         int old_mask = (1 << old_bits) - 1;
2060         int n;
2061
2062         WARN_ON(pgid->seed >= old_pg_num);
2063         if (new_pg_num <= old_pg_num)
2064                 return false;
2065
2066         for (n = 1; ; n++) {
2067                 int next_bit = n << (old_bits - 1);
2068                 u32 s = next_bit | pgid->seed;
2069
2070                 if (s < old_pg_num || s == pgid->seed)
2071                         continue;
2072                 if (s >= new_pg_num)
2073                         break;
2074
2075                 s = ceph_stable_mod(s, old_pg_num, old_mask);
2076                 if (s == pgid->seed)
2077                         return true;
2078         }
2079
2080         return false;
2081 }
2082
2083 bool ceph_is_new_interval(const struct ceph_osds *old_acting,
2084                           const struct ceph_osds *new_acting,
2085                           const struct ceph_osds *old_up,
2086                           const struct ceph_osds *new_up,
2087                           int old_size,
2088                           int new_size,
2089                           int old_min_size,
2090                           int new_min_size,
2091                           u32 old_pg_num,
2092                           u32 new_pg_num,
2093                           bool old_sort_bitwise,
2094                           bool new_sort_bitwise,
2095                           bool old_recovery_deletes,
2096                           bool new_recovery_deletes,
2097                           const struct ceph_pg *pgid)
2098 {
2099         return !osds_equal(old_acting, new_acting) ||
2100                !osds_equal(old_up, new_up) ||
2101                old_size != new_size ||
2102                old_min_size != new_min_size ||
2103                ceph_pg_is_split(pgid, old_pg_num, new_pg_num) ||
2104                old_sort_bitwise != new_sort_bitwise ||
2105                old_recovery_deletes != new_recovery_deletes;
2106 }
2107
2108 static int calc_pg_rank(int osd, const struct ceph_osds *acting)
2109 {
2110         int i;
2111
2112         for (i = 0; i < acting->size; i++) {
2113                 if (acting->osds[i] == osd)
2114                         return i;
2115         }
2116
2117         return -1;
2118 }
2119
2120 static bool primary_changed(const struct ceph_osds *old_acting,
2121                             const struct ceph_osds *new_acting)
2122 {
2123         if (!old_acting->size && !new_acting->size)
2124                 return false; /* both still empty */
2125
2126         if (!old_acting->size ^ !new_acting->size)
2127                 return true; /* was empty, now not, or vice versa */
2128
2129         if (old_acting->primary != new_acting->primary)
2130                 return true; /* primary changed */
2131
2132         if (calc_pg_rank(old_acting->primary, old_acting) !=
2133             calc_pg_rank(new_acting->primary, new_acting))
2134                 return true;
2135
2136         return false; /* same primary (tho replicas may have changed) */
2137 }
2138
2139 bool ceph_osds_changed(const struct ceph_osds *old_acting,
2140                        const struct ceph_osds *new_acting,
2141                        bool any_change)
2142 {
2143         if (primary_changed(old_acting, new_acting))
2144                 return true;
2145
2146         if (any_change && !__osds_equal(old_acting, new_acting))
2147                 return true;
2148
2149         return false;
2150 }
2151
2152 /*
2153  * Map an object into a PG.
2154  *
2155  * Should only be called with target_oid and target_oloc (as opposed to
2156  * base_oid and base_oloc), since tiering isn't taken into account.
2157  */
2158 void __ceph_object_locator_to_pg(struct ceph_pg_pool_info *pi,
2159                                  const struct ceph_object_id *oid,
2160                                  const struct ceph_object_locator *oloc,
2161                                  struct ceph_pg *raw_pgid)
2162 {
2163         WARN_ON(pi->id != oloc->pool);
2164
2165         if (!oloc->pool_ns) {
2166                 raw_pgid->pool = oloc->pool;
2167                 raw_pgid->seed = ceph_str_hash(pi->object_hash, oid->name,
2168                                              oid->name_len);
2169                 dout("%s %s -> raw_pgid %llu.%x\n", __func__, oid->name,
2170                      raw_pgid->pool, raw_pgid->seed);
2171         } else {
2172                 char stack_buf[256];
2173                 char *buf = stack_buf;
2174                 int nsl = oloc->pool_ns->len;
2175                 size_t total = nsl + 1 + oid->name_len;
2176
2177                 if (total > sizeof(stack_buf))
2178                         buf = kmalloc(total, GFP_NOIO | __GFP_NOFAIL);
2179                 memcpy(buf, oloc->pool_ns->str, nsl);
2180                 buf[nsl] = '\037';
2181                 memcpy(buf + nsl + 1, oid->name, oid->name_len);
2182                 raw_pgid->pool = oloc->pool;
2183                 raw_pgid->seed = ceph_str_hash(pi->object_hash, buf, total);
2184                 if (buf != stack_buf)
2185                         kfree(buf);
2186                 dout("%s %s ns %.*s -> raw_pgid %llu.%x\n", __func__,
2187                      oid->name, nsl, oloc->pool_ns->str,
2188                      raw_pgid->pool, raw_pgid->seed);
2189         }
2190 }
2191
2192 int ceph_object_locator_to_pg(struct ceph_osdmap *osdmap,
2193                               const struct ceph_object_id *oid,
2194                               const struct ceph_object_locator *oloc,
2195                               struct ceph_pg *raw_pgid)
2196 {
2197         struct ceph_pg_pool_info *pi;
2198
2199         pi = ceph_pg_pool_by_id(osdmap, oloc->pool);
2200         if (!pi)
2201                 return -ENOENT;
2202
2203         __ceph_object_locator_to_pg(pi, oid, oloc, raw_pgid);
2204         return 0;
2205 }
2206 EXPORT_SYMBOL(ceph_object_locator_to_pg);
2207
2208 /*
2209  * Map a raw PG (full precision ps) into an actual PG.
2210  */
2211 static void raw_pg_to_pg(struct ceph_pg_pool_info *pi,
2212                          const struct ceph_pg *raw_pgid,
2213                          struct ceph_pg *pgid)
2214 {
2215         pgid->pool = raw_pgid->pool;
2216         pgid->seed = ceph_stable_mod(raw_pgid->seed, pi->pg_num,
2217                                      pi->pg_num_mask);
2218 }
2219
2220 /*
2221  * Map a raw PG (full precision ps) into a placement ps (placement
2222  * seed).  Include pool id in that value so that different pools don't
2223  * use the same seeds.
2224  */
2225 static u32 raw_pg_to_pps(struct ceph_pg_pool_info *pi,
2226                          const struct ceph_pg *raw_pgid)
2227 {
2228         if (pi->flags & CEPH_POOL_FLAG_HASHPSPOOL) {
2229                 /* hash pool id and seed so that pool PGs do not overlap */
2230                 return crush_hash32_2(CRUSH_HASH_RJENKINS1,
2231                                       ceph_stable_mod(raw_pgid->seed,
2232                                                       pi->pgp_num,
2233                                                       pi->pgp_num_mask),
2234                                       raw_pgid->pool);
2235         } else {
2236                 /*
2237                  * legacy behavior: add ps and pool together.  this is
2238                  * not a great approach because the PGs from each pool
2239                  * will overlap on top of each other: 0.5 == 1.4 ==
2240                  * 2.3 == ...
2241                  */
2242                 return ceph_stable_mod(raw_pgid->seed, pi->pgp_num,
2243                                        pi->pgp_num_mask) +
2244                        (unsigned)raw_pgid->pool;
2245         }
2246 }
2247
2248 /*
2249  * Magic value used for a "default" fallback choose_args, used if the
2250  * crush_choose_arg_map passed to do_crush() does not exist.  If this
2251  * also doesn't exist, fall back to canonical weights.
2252  */
2253 #define CEPH_DEFAULT_CHOOSE_ARGS        -1
2254
2255 static int do_crush(struct ceph_osdmap *map, int ruleno, int x,
2256                     int *result, int result_max,
2257                     const __u32 *weight, int weight_max,
2258                     s64 choose_args_index)
2259 {
2260         struct crush_choose_arg_map *arg_map;
2261         int r;
2262
2263         BUG_ON(result_max > CEPH_PG_MAX_SIZE);
2264
2265         arg_map = lookup_choose_arg_map(&map->crush->choose_args,
2266                                         choose_args_index);
2267         if (!arg_map)
2268                 arg_map = lookup_choose_arg_map(&map->crush->choose_args,
2269                                                 CEPH_DEFAULT_CHOOSE_ARGS);
2270
2271         mutex_lock(&map->crush_workspace_mutex);
2272         r = crush_do_rule(map->crush, ruleno, x, result, result_max,
2273                           weight, weight_max, map->crush_workspace,
2274                           arg_map ? arg_map->args : NULL);
2275         mutex_unlock(&map->crush_workspace_mutex);
2276
2277         return r;
2278 }
2279
2280 static void remove_nonexistent_osds(struct ceph_osdmap *osdmap,
2281                                     struct ceph_pg_pool_info *pi,
2282                                     struct ceph_osds *set)
2283 {
2284         int i;
2285
2286         if (ceph_can_shift_osds(pi)) {
2287                 int removed = 0;
2288
2289                 /* shift left */
2290                 for (i = 0; i < set->size; i++) {
2291                         if (!ceph_osd_exists(osdmap, set->osds[i])) {
2292                                 removed++;
2293                                 continue;
2294                         }
2295                         if (removed)
2296                                 set->osds[i - removed] = set->osds[i];
2297                 }
2298                 set->size -= removed;
2299         } else {
2300                 /* set dne devices to NONE */
2301                 for (i = 0; i < set->size; i++) {
2302                         if (!ceph_osd_exists(osdmap, set->osds[i]))
2303                                 set->osds[i] = CRUSH_ITEM_NONE;
2304                 }
2305         }
2306 }
2307
2308 /*
2309  * Calculate raw set (CRUSH output) for given PG and filter out
2310  * nonexistent OSDs.  ->primary is undefined for a raw set.
2311  *
2312  * Placement seed (CRUSH input) is returned through @ppps.
2313  */
2314 static void pg_to_raw_osds(struct ceph_osdmap *osdmap,
2315                            struct ceph_pg_pool_info *pi,
2316                            const struct ceph_pg *raw_pgid,
2317                            struct ceph_osds *raw,
2318                            u32 *ppps)
2319 {
2320         u32 pps = raw_pg_to_pps(pi, raw_pgid);
2321         int ruleno;
2322         int len;
2323
2324         ceph_osds_init(raw);
2325         if (ppps)
2326                 *ppps = pps;
2327
2328         ruleno = crush_find_rule(osdmap->crush, pi->crush_ruleset, pi->type,
2329                                  pi->size);
2330         if (ruleno < 0) {
2331                 pr_err("no crush rule: pool %lld ruleset %d type %d size %d\n",
2332                        pi->id, pi->crush_ruleset, pi->type, pi->size);
2333                 return;
2334         }
2335
2336         if (pi->size > ARRAY_SIZE(raw->osds)) {
2337                 pr_err_ratelimited("pool %lld ruleset %d type %d too wide: size %d > %zu\n",
2338                        pi->id, pi->crush_ruleset, pi->type, pi->size,
2339                        ARRAY_SIZE(raw->osds));
2340                 return;
2341         }
2342
2343         len = do_crush(osdmap, ruleno, pps, raw->osds, pi->size,
2344                        osdmap->osd_weight, osdmap->max_osd, pi->id);
2345         if (len < 0) {
2346                 pr_err("error %d from crush rule %d: pool %lld ruleset %d type %d size %d\n",
2347                        len, ruleno, pi->id, pi->crush_ruleset, pi->type,
2348                        pi->size);
2349                 return;
2350         }
2351
2352         raw->size = len;
2353         remove_nonexistent_osds(osdmap, pi, raw);
2354 }
2355
2356 /* apply pg_upmap[_items] mappings */
2357 static void apply_upmap(struct ceph_osdmap *osdmap,
2358                         const struct ceph_pg *pgid,
2359                         struct ceph_osds *raw)
2360 {
2361         struct ceph_pg_mapping *pg;
2362         int i, j;
2363
2364         pg = lookup_pg_mapping(&osdmap->pg_upmap, pgid);
2365         if (pg) {
2366                 /* make sure targets aren't marked out */
2367                 for (i = 0; i < pg->pg_upmap.len; i++) {
2368                         int osd = pg->pg_upmap.osds[i];
2369
2370                         if (osd != CRUSH_ITEM_NONE &&
2371                             osd < osdmap->max_osd &&
2372                             osdmap->osd_weight[osd] == 0) {
2373                                 /* reject/ignore explicit mapping */
2374                                 return;
2375                         }
2376                 }
2377                 for (i = 0; i < pg->pg_upmap.len; i++)
2378                         raw->osds[i] = pg->pg_upmap.osds[i];
2379                 raw->size = pg->pg_upmap.len;
2380                 /* check and apply pg_upmap_items, if any */
2381         }
2382
2383         pg = lookup_pg_mapping(&osdmap->pg_upmap_items, pgid);
2384         if (pg) {
2385                 /*
2386                  * Note: this approach does not allow a bidirectional swap,
2387                  * e.g., [[1,2],[2,1]] applied to [0,1,2] -> [0,2,1].
2388                  */
2389                 for (i = 0; i < pg->pg_upmap_items.len; i++) {
2390                         int from = pg->pg_upmap_items.from_to[i][0];
2391                         int to = pg->pg_upmap_items.from_to[i][1];
2392                         int pos = -1;
2393                         bool exists = false;
2394
2395                         /* make sure replacement doesn't already appear */
2396                         for (j = 0; j < raw->size; j++) {
2397                                 int osd = raw->osds[j];
2398
2399                                 if (osd == to) {
2400                                         exists = true;
2401                                         break;
2402                                 }
2403                                 /* ignore mapping if target is marked out */
2404                                 if (osd == from && pos < 0 &&
2405                                     !(to != CRUSH_ITEM_NONE &&
2406                                       to < osdmap->max_osd &&
2407                                       osdmap->osd_weight[to] == 0)) {
2408                                         pos = j;
2409                                 }
2410                         }
2411                         if (!exists && pos >= 0)
2412                                 raw->osds[pos] = to;
2413                 }
2414         }
2415 }
2416
2417 /*
2418  * Given raw set, calculate up set and up primary.  By definition of an
2419  * up set, the result won't contain nonexistent or down OSDs.
2420  *
2421  * This is done in-place - on return @set is the up set.  If it's
2422  * empty, ->primary will remain undefined.
2423  */
2424 static void raw_to_up_osds(struct ceph_osdmap *osdmap,
2425                            struct ceph_pg_pool_info *pi,
2426                            struct ceph_osds *set)
2427 {
2428         int i;
2429
2430         /* ->primary is undefined for a raw set */
2431         BUG_ON(set->primary != -1);
2432
2433         if (ceph_can_shift_osds(pi)) {
2434                 int removed = 0;
2435
2436                 /* shift left */
2437                 for (i = 0; i < set->size; i++) {
2438                         if (ceph_osd_is_down(osdmap, set->osds[i])) {
2439                                 removed++;
2440                                 continue;
2441                         }
2442                         if (removed)
2443                                 set->osds[i - removed] = set->osds[i];
2444                 }
2445                 set->size -= removed;
2446                 if (set->size > 0)
2447                         set->primary = set->osds[0];
2448         } else {
2449                 /* set down/dne devices to NONE */
2450                 for (i = set->size - 1; i >= 0; i--) {
2451                         if (ceph_osd_is_down(osdmap, set->osds[i]))
2452                                 set->osds[i] = CRUSH_ITEM_NONE;
2453                         else
2454                                 set->primary = set->osds[i];
2455                 }
2456         }
2457 }
2458
2459 static void apply_primary_affinity(struct ceph_osdmap *osdmap,
2460                                    struct ceph_pg_pool_info *pi,
2461                                    u32 pps,
2462                                    struct ceph_osds *up)
2463 {
2464         int i;
2465         int pos = -1;
2466
2467         /*
2468          * Do we have any non-default primary_affinity values for these
2469          * osds?
2470          */
2471         if (!osdmap->osd_primary_affinity)
2472                 return;
2473
2474         for (i = 0; i < up->size; i++) {
2475                 int osd = up->osds[i];
2476
2477                 if (osd != CRUSH_ITEM_NONE &&
2478                     osdmap->osd_primary_affinity[osd] !=
2479                                         CEPH_OSD_DEFAULT_PRIMARY_AFFINITY) {
2480                         break;
2481                 }
2482         }
2483         if (i == up->size)
2484                 return;
2485
2486         /*
2487          * Pick the primary.  Feed both the seed (for the pg) and the
2488          * osd into the hash/rng so that a proportional fraction of an
2489          * osd's pgs get rejected as primary.
2490          */
2491         for (i = 0; i < up->size; i++) {
2492                 int osd = up->osds[i];
2493                 u32 aff;
2494
2495                 if (osd == CRUSH_ITEM_NONE)
2496                         continue;
2497
2498                 aff = osdmap->osd_primary_affinity[osd];
2499                 if (aff < CEPH_OSD_MAX_PRIMARY_AFFINITY &&
2500                     (crush_hash32_2(CRUSH_HASH_RJENKINS1,
2501                                     pps, osd) >> 16) >= aff) {
2502                         /*
2503                          * We chose not to use this primary.  Note it
2504                          * anyway as a fallback in case we don't pick
2505                          * anyone else, but keep looking.
2506                          */
2507                         if (pos < 0)
2508                                 pos = i;
2509                 } else {
2510                         pos = i;
2511                         break;
2512                 }
2513         }
2514         if (pos < 0)
2515                 return;
2516
2517         up->primary = up->osds[pos];
2518
2519         if (ceph_can_shift_osds(pi) && pos > 0) {
2520                 /* move the new primary to the front */
2521                 for (i = pos; i > 0; i--)
2522                         up->osds[i] = up->osds[i - 1];
2523                 up->osds[0] = up->primary;
2524         }
2525 }
2526
2527 /*
2528  * Get pg_temp and primary_temp mappings for given PG.
2529  *
2530  * Note that a PG may have none, only pg_temp, only primary_temp or
2531  * both pg_temp and primary_temp mappings.  This means @temp isn't
2532  * always a valid OSD set on return: in the "only primary_temp" case,
2533  * @temp will have its ->primary >= 0 but ->size == 0.
2534  */
2535 static void get_temp_osds(struct ceph_osdmap *osdmap,
2536                           struct ceph_pg_pool_info *pi,
2537                           const struct ceph_pg *pgid,
2538                           struct ceph_osds *temp)
2539 {
2540         struct ceph_pg_mapping *pg;
2541         int i;
2542
2543         ceph_osds_init(temp);
2544
2545         /* pg_temp? */
2546         pg = lookup_pg_mapping(&osdmap->pg_temp, pgid);
2547         if (pg) {
2548                 for (i = 0; i < pg->pg_temp.len; i++) {
2549                         if (ceph_osd_is_down(osdmap, pg->pg_temp.osds[i])) {
2550                                 if (ceph_can_shift_osds(pi))
2551                                         continue;
2552
2553                                 temp->osds[temp->size++] = CRUSH_ITEM_NONE;
2554                         } else {
2555                                 temp->osds[temp->size++] = pg->pg_temp.osds[i];
2556                         }
2557                 }
2558
2559                 /* apply pg_temp's primary */
2560                 for (i = 0; i < temp->size; i++) {
2561                         if (temp->osds[i] != CRUSH_ITEM_NONE) {
2562                                 temp->primary = temp->osds[i];
2563                                 break;
2564                         }
2565                 }
2566         }
2567
2568         /* primary_temp? */
2569         pg = lookup_pg_mapping(&osdmap->primary_temp, pgid);
2570         if (pg)
2571                 temp->primary = pg->primary_temp.osd;
2572 }
2573
2574 /*
2575  * Map a PG to its acting set as well as its up set.
2576  *
2577  * Acting set is used for data mapping purposes, while up set can be
2578  * recorded for detecting interval changes and deciding whether to
2579  * resend a request.
2580  */
2581 void ceph_pg_to_up_acting_osds(struct ceph_osdmap *osdmap,
2582                                struct ceph_pg_pool_info *pi,
2583                                const struct ceph_pg *raw_pgid,
2584                                struct ceph_osds *up,
2585                                struct ceph_osds *acting)
2586 {
2587         struct ceph_pg pgid;
2588         u32 pps;
2589
2590         WARN_ON(pi->id != raw_pgid->pool);
2591         raw_pg_to_pg(pi, raw_pgid, &pgid);
2592
2593         pg_to_raw_osds(osdmap, pi, raw_pgid, up, &pps);
2594         apply_upmap(osdmap, &pgid, up);
2595         raw_to_up_osds(osdmap, pi, up);
2596         apply_primary_affinity(osdmap, pi, pps, up);
2597         get_temp_osds(osdmap, pi, &pgid, acting);
2598         if (!acting->size) {
2599                 memcpy(acting->osds, up->osds, up->size * sizeof(up->osds[0]));
2600                 acting->size = up->size;
2601                 if (acting->primary == -1)
2602                         acting->primary = up->primary;
2603         }
2604         WARN_ON(!osds_valid(up) || !osds_valid(acting));
2605 }
2606
2607 bool ceph_pg_to_primary_shard(struct ceph_osdmap *osdmap,
2608                               struct ceph_pg_pool_info *pi,
2609                               const struct ceph_pg *raw_pgid,
2610                               struct ceph_spg *spgid)
2611 {
2612         struct ceph_pg pgid;
2613         struct ceph_osds up, acting;
2614         int i;
2615
2616         WARN_ON(pi->id != raw_pgid->pool);
2617         raw_pg_to_pg(pi, raw_pgid, &pgid);
2618
2619         if (ceph_can_shift_osds(pi)) {
2620                 spgid->pgid = pgid; /* struct */
2621                 spgid->shard = CEPH_SPG_NOSHARD;
2622                 return true;
2623         }
2624
2625         ceph_pg_to_up_acting_osds(osdmap, pi, &pgid, &up, &acting);
2626         for (i = 0; i < acting.size; i++) {
2627                 if (acting.osds[i] == acting.primary) {
2628                         spgid->pgid = pgid; /* struct */
2629                         spgid->shard = i;
2630                         return true;
2631                 }
2632         }
2633
2634         return false;
2635 }
2636
2637 /*
2638  * Return acting primary for given PG, or -1 if none.
2639  */
2640 int ceph_pg_to_acting_primary(struct ceph_osdmap *osdmap,
2641                               const struct ceph_pg *raw_pgid)
2642 {
2643         struct ceph_pg_pool_info *pi;
2644         struct ceph_osds up, acting;
2645
2646         pi = ceph_pg_pool_by_id(osdmap, raw_pgid->pool);
2647         if (!pi)
2648                 return -1;
2649
2650         ceph_pg_to_up_acting_osds(osdmap, pi, raw_pgid, &up, &acting);
2651         return acting.primary;
2652 }
2653 EXPORT_SYMBOL(ceph_pg_to_acting_primary);