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 struct crush_name_node {
 142         struct rb_node cn_node;
 143         int cn_id;
 144         char cn_name[];
 145 };
 146
 147 static struct crush_name_node *alloc_crush_name(size_t name_len)
 148 {
 149         struct crush_name_node *cn;
 150
 151         cn = kmalloc(sizeof(*cn) + name_len + 1, GFP_NOIO);
 152         if (!cn)
 153                 return NULL;
 154
 155         RB_CLEAR_NODE(&cn->cn_node);
 156         return cn;
 157 }
 158
 159 static void free_crush_name(struct crush_name_node *cn)
 160 {
 161         WARN_ON(!RB_EMPTY_NODE(&cn->cn_node));
 162
 163         kfree(cn);
 164 }
 165
 166 DEFINE_RB_FUNCS(crush_name, struct crush_name_node, cn_id, cn_node)
 167
 168 static int decode_crush_names(void **p, void *end, struct rb_root *root)
 169 {
 170         u32 n;
 171
 172         ceph_decode_32_safe(p, end, n, e_inval);
 173         while (n--) {
 174                 struct crush_name_node *cn;
 175                 int id;
 176                 u32 name_len;
 177
 178                 ceph_decode_32_safe(p, end, id, e_inval);
 179                 ceph_decode_32_safe(p, end, name_len, e_inval);
 180                 ceph_decode_need(p, end, name_len, e_inval);
 181
 182                 cn = alloc_crush_name(name_len);
 183                 if (!cn)
 184                         return -ENOMEM;
 185
 186                 cn->cn_id = id;
 187                 memcpy(cn->cn_name, *p, name_len);
 188                 cn->cn_name[name_len] = '\0';
 189                 *p += name_len;
 190
 191                 if (!__insert_crush_name(root, cn)) {
 192                         free_crush_name(cn);
 193                         return -EEXIST;
 194                 }
 195         }
 196
 197         return 0;
 198
 199 e_inval:
 200         return -EINVAL;
 201 }
 202
 203 void clear_crush_names(struct rb_root *root)
 204 {
 205         while (!RB_EMPTY_ROOT(root)) {
 206                 struct crush_name_node *cn =
 207                     rb_entry(rb_first(root), struct crush_name_node, cn_node);
 208
 209                 erase_crush_name(root, cn);
 210                 free_crush_name(cn);
 211         }
 212 }
 213
 214 static struct crush_choose_arg_map *alloc_choose_arg_map(void)
 215 {
 216         struct crush_choose_arg_map *arg_map;
 217
 218         arg_map = kzalloc(sizeof(*arg_map), GFP_NOIO);
 219         if (!arg_map)
 220                 return NULL;
 221
 222         RB_CLEAR_NODE(&arg_map->node);
 223         return arg_map;
 224 }
 225
 226 static void free_choose_arg_map(struct crush_choose_arg_map *arg_map)
 227 {
 228         if (arg_map) {
 229                 int i, j;
 230
 231                 WARN_ON(!RB_EMPTY_NODE(&arg_map->node));
 232
 233                 for (i = 0; i < arg_map->size; i++) {
 234                         struct crush_choose_arg *arg = &arg_map->args[i];
 235
 236                         for (j = 0; j < arg->weight_set_size; j++)
 237                                 kfree(arg->weight_set[j].weights);
 238                         kfree(arg->weight_set);
 239                         kfree(arg->ids);
 240                 }
 241                 kfree(arg_map->args);
 242                 kfree(arg_map);
 243         }
 244 }
 245
 246 DEFINE_RB_FUNCS(choose_arg_map, struct crush_choose_arg_map, choose_args_index,
 247                 node);
 248
 249 void clear_choose_args(struct crush_map *c)
 250 {
 251         while (!RB_EMPTY_ROOT(&c->choose_args)) {
 252                 struct crush_choose_arg_map *arg_map =
 253                     rb_entry(rb_first(&c->choose_args),
 254                              struct crush_choose_arg_map, node);
 255
 256                 erase_choose_arg_map(&c->choose_args, arg_map);
 257                 free_choose_arg_map(arg_map);
 258         }
 259 }
 260
 261 static u32 *decode_array_32_alloc(void **p, void *end, u32 *plen)
 262 {
 263         u32 *a = NULL;
 264         u32 len;
 265         int ret;
 266
 267         ceph_decode_32_safe(p, end, len, e_inval);
 268         if (len) {
 269                 u32 i;
 270
 271                 a = kmalloc_array(len, sizeof(u32), GFP_NOIO);
 272                 if (!a) {
 273                         ret = -ENOMEM;
 274                         goto fail;
 275                 }
 276
 277                 ceph_decode_need(p, end, len * sizeof(u32), e_inval);
 278                 for (i = 0; i < len; i++)
 279                         a[i] = ceph_decode_32(p);
 280         }
 281
 282         *plen = len;
 283         return a;
 284
 285 e_inval:
 286         ret = -EINVAL;
 287 fail:
 288         kfree(a);
 289         return ERR_PTR(ret);
 290 }
 291
 292 /*
 293  * Assumes @arg is zero-initialized.
 294  */
 295 static int decode_choose_arg(void **p, void *end, struct crush_choose_arg *arg)
 296 {
 297         int ret;
 298
 299         ceph_decode_32_safe(p, end, arg->weight_set_size, e_inval);
 300         if (arg->weight_set_size) {
 301                 u32 i;
 302
 303                 arg->weight_set = kmalloc_array(arg->weight_set_size,
 304                                                 sizeof(*arg->weight_set),
 305                                                 GFP_NOIO);
 306                 if (!arg->weight_set)
 307                         return -ENOMEM;
 308
 309                 for (i = 0; i < arg->weight_set_size; i++) {
 310                         struct crush_weight_set *w = &arg->weight_set[i];
 311
 312                         w->weights = decode_array_32_alloc(p, end, &w->size);
 313                         if (IS_ERR(w->weights)) {
 314                                 ret = PTR_ERR(w->weights);
 315                                 w->weights = NULL;
 316                                 return ret;
 317                         }
 318                 }
 319         }
 320
 321         arg->ids = decode_array_32_alloc(p, end, &arg->ids_size);
 322         if (IS_ERR(arg->ids)) {
 323                 ret = PTR_ERR(arg->ids);
 324                 arg->ids = NULL;
 325                 return ret;
 326         }
 327
 328         return 0;
 329
 330 e_inval:
 331         return -EINVAL;
 332 }
 333
 334 static int decode_choose_args(void **p, void *end, struct crush_map *c)
 335 {
 336         struct crush_choose_arg_map *arg_map = NULL;
 337         u32 num_choose_arg_maps, num_buckets;
 338         int ret;
 339
 340         ceph_decode_32_safe(p, end, num_choose_arg_maps, e_inval);
 341         while (num_choose_arg_maps--) {
 342                 arg_map = alloc_choose_arg_map();
 343                 if (!arg_map) {
 344                         ret = -ENOMEM;
 345                         goto fail;
 346                 }
 347
 348                 ceph_decode_64_safe(p, end, arg_map->choose_args_index,
 349                                     e_inval);
 350                 arg_map->size = c->max_buckets;
 351                 arg_map->args = kcalloc(arg_map->size, sizeof(*arg_map->args),
 352                                         GFP_NOIO);
 353                 if (!arg_map->args) {
 354                         ret = -ENOMEM;
 355                         goto fail;
 356                 }
 357
 358                 ceph_decode_32_safe(p, end, num_buckets, e_inval);
 359                 while (num_buckets--) {
 360                         struct crush_choose_arg *arg;
 361                         u32 bucket_index;
 362
 363                         ceph_decode_32_safe(p, end, bucket_index, e_inval);
 364                         if (bucket_index >= arg_map->size)
 365                                 goto e_inval;
 366
 367                         arg = &arg_map->args[bucket_index];
 368                         ret = decode_choose_arg(p, end, arg);
 369                         if (ret)
 370                                 goto fail;
 371
 372                         if (arg->ids_size &&
 373                             arg->ids_size != c->buckets[bucket_index]->size)
 374                                 goto e_inval;
 375                 }
 376
 377                 insert_choose_arg_map(&c->choose_args, arg_map);
 378         }
 379
 380         return 0;
 381
 382 e_inval:
 383         ret = -EINVAL;
 384 fail:
 385         free_choose_arg_map(arg_map);
 386         return ret;
 387 }
 388
 389 static void crush_finalize(struct crush_map *c)
 390 {
 391         __s32 b;
 392
 393         /* Space for the array of pointers to per-bucket workspace */
 394         c->working_size = sizeof(struct crush_work) +
 395             c->max_buckets * sizeof(struct crush_work_bucket *);
 396
 397         for (b = 0; b < c->max_buckets; b++) {
 398                 if (!c->buckets[b])
 399                         continue;
 400
 401                 switch (c->buckets[b]->alg) {
 402                 default:
 403                         /*
 404                          * The base case, permutation variables and
 405                          * the pointer to the permutation array.
 406                          */
 407                         c->working_size += sizeof(struct crush_work_bucket);
 408                         break;
 409                 }
 410                 /* Every bucket has a permutation array. */
 411                 c->working_size += c->buckets[b]->size * sizeof(__u32);
 412         }
 413 }
 414
 415 static struct crush_map *crush_decode(void *pbyval, void *end)
 416 {
 417         struct crush_map *c;
 418         int err;
 419         int i, j;
 420         void **p = &pbyval;
 421         void *start = pbyval;
 422         u32 magic;
 423
 424         dout("crush_decode %p to %p len %d\n", *p, end, (int)(end - *p));
 425
 426         c = kzalloc(sizeof(*c), GFP_NOFS);
 427         if (c == NULL)
 428                 return ERR_PTR(-ENOMEM);
 429
 430         c->type_names = RB_ROOT;
 431         c->names = RB_ROOT;
 432         c->choose_args = RB_ROOT;
 433
 434         /* set tunables to default values */
 435         c->choose_local_tries = 2;
 436         c->choose_local_fallback_tries = 5;
 437         c->choose_total_tries = 19;
 438         c->chooseleaf_descend_once = 0;
 439
 440         ceph_decode_need(p, end, 4*sizeof(u32), bad);
 441         magic = ceph_decode_32(p);
 442         if (magic != CRUSH_MAGIC) {
 443                 pr_err("crush_decode magic %x != current %x\n",
 444                        (unsigned int)magic, (unsigned int)CRUSH_MAGIC);
 445                 goto bad;
 446         }
 447         c->max_buckets = ceph_decode_32(p);
 448         c->max_rules = ceph_decode_32(p);
 449         c->max_devices = ceph_decode_32(p);
 450
 451         c->buckets = kcalloc(c->max_buckets, sizeof(*c->buckets), GFP_NOFS);
 452         if (c->buckets == NULL)
 453                 goto badmem;
 454         c->rules = kcalloc(c->max_rules, sizeof(*c->rules), GFP_NOFS);
 455         if (c->rules == NULL)
 456                 goto badmem;
 457
 458         /* buckets */
 459         for (i = 0; i < c->max_buckets; i++) {
 460                 int size = 0;
 461                 u32 alg;
 462                 struct crush_bucket *b;
 463
 464                 ceph_decode_32_safe(p, end, alg, bad);
 465                 if (alg == 0) {
 466                         c->buckets[i] = NULL;
 467                         continue;
 468                 }
 469                 dout("crush_decode bucket %d off %x %p to %p\n",
 470                      i, (int)(*p-start), *p, end);
 471
 472                 switch (alg) {
 473                 case CRUSH_BUCKET_UNIFORM:
 474                         size = sizeof(struct crush_bucket_uniform);
 475                         break;
 476                 case CRUSH_BUCKET_LIST:
 477                         size = sizeof(struct crush_bucket_list);
 478                         break;
 479                 case CRUSH_BUCKET_TREE:
 480                         size = sizeof(struct crush_bucket_tree);
 481                         break;
 482                 case CRUSH_BUCKET_STRAW:
 483                         size = sizeof(struct crush_bucket_straw);
 484                         break;
 485                 case CRUSH_BUCKET_STRAW2:
 486                         size = sizeof(struct crush_bucket_straw2);
 487                         break;
 488                 default:
 489                         goto bad;
 490                 }
 491                 BUG_ON(size == 0);
 492                 b = c->buckets[i] = kzalloc(size, GFP_NOFS);
 493                 if (b == NULL)
 494                         goto badmem;
 495
 496                 ceph_decode_need(p, end, 4*sizeof(u32), bad);
 497                 b->id = ceph_decode_32(p);
 498                 b->type = ceph_decode_16(p);
 499                 b->alg = ceph_decode_8(p);
 500                 b->hash = ceph_decode_8(p);
 501                 b->weight = ceph_decode_32(p);
 502                 b->size = ceph_decode_32(p);
 503
 504                 dout("crush_decode bucket size %d off %x %p to %p\n",
 505                      b->size, (int)(*p-start), *p, end);
 506
 507                 b->items = kcalloc(b->size, sizeof(__s32), GFP_NOFS);
 508                 if (b->items == NULL)
 509                         goto badmem;
 510
 511                 ceph_decode_need(p, end, b->size*sizeof(u32), bad);
 512                 for (j = 0; j < b->size; j++)
 513                         b->items[j] = ceph_decode_32(p);
 514
 515                 switch (b->alg) {
 516                 case CRUSH_BUCKET_UNIFORM:
 517                         err = crush_decode_uniform_bucket(p, end,
 518                                   (struct crush_bucket_uniform *)b);
 519                         if (err < 0)
 520                                 goto fail;
 521                         break;
 522                 case CRUSH_BUCKET_LIST:
 523                         err = crush_decode_list_bucket(p, end,
 524                                (struct crush_bucket_list *)b);
 525                         if (err < 0)
 526                                 goto fail;
 527                         break;
 528                 case CRUSH_BUCKET_TREE:
 529                         err = crush_decode_tree_bucket(p, end,
 530                                 (struct crush_bucket_tree *)b);
 531                         if (err < 0)
 532                                 goto fail;
 533                         break;
 534                 case CRUSH_BUCKET_STRAW:
 535                         err = crush_decode_straw_bucket(p, end,
 536                                 (struct crush_bucket_straw *)b);
 537                         if (err < 0)
 538                                 goto fail;
 539                         break;
 540                 case CRUSH_BUCKET_STRAW2:
 541                         err = crush_decode_straw2_bucket(p, end,
 542                                 (struct crush_bucket_straw2 *)b);
 543                         if (err < 0)
 544                                 goto fail;
 545                         break;
 546                 }
 547         }
 548
 549         /* rules */
 550         dout("rule vec is %p\n", c->rules);
 551         for (i = 0; i < c->max_rules; i++) {
 552                 u32 yes;
 553                 struct crush_rule *r;
 554
 555                 ceph_decode_32_safe(p, end, yes, bad);
 556                 if (!yes) {
 557                         dout("crush_decode NO rule %d off %x %p to %p\n",
 558                              i, (int)(*p-start), *p, end);
 559                         c->rules[i] = NULL;
 560                         continue;
 561                 }
 562
 563                 dout("crush_decode rule %d off %x %p to %p\n",
 564                      i, (int)(*p-start), *p, end);
 565
 566                 /* len */
 567                 ceph_decode_32_safe(p, end, yes, bad);
 568 #if BITS_PER_LONG == 32
 569                 if (yes > (ULONG_MAX - sizeof(*r))
 570                           / sizeof(struct crush_rule_step))
 571                         goto bad;
 572 #endif
 573                 r = kmalloc(struct_size(r, steps, yes), GFP_NOFS);
 574                 c->rules[i] = r;
 575                 if (r == NULL)
 576                         goto badmem;
 577                 dout(" rule %d is at %p\n", i, r);
 578                 r->len = yes;
 579                 ceph_decode_copy_safe(p, end, &r->mask, 4, bad); /* 4 u8's */
 580                 ceph_decode_need(p, end, r->len*3*sizeof(u32), bad);
 581                 for (j = 0; j < r->len; j++) {
 582                         r->steps[j].op = ceph_decode_32(p);
 583                         r->steps[j].arg1 = ceph_decode_32(p);
 584                         r->steps[j].arg2 = ceph_decode_32(p);
 585                 }
 586         }
 587
 588         err = decode_crush_names(p, end, &c->type_names);
 589         if (err)
 590                 goto fail;
 591
 592         err = decode_crush_names(p, end, &c->names);
 593         if (err)
 594                 goto fail;
 595
 596         ceph_decode_skip_map(p, end, 32, string, bad); /* rule_name_map */
 597
 598         /* tunables */
 599         ceph_decode_need(p, end, 3*sizeof(u32), done);
 600         c->choose_local_tries = ceph_decode_32(p);
 601         c->choose_local_fallback_tries =  ceph_decode_32(p);
 602         c->choose_total_tries = ceph_decode_32(p);
 603         dout("crush decode tunable choose_local_tries = %d\n",
 604              c->choose_local_tries);
 605         dout("crush decode tunable choose_local_fallback_tries = %d\n",
 606              c->choose_local_fallback_tries);
 607         dout("crush decode tunable choose_total_tries = %d\n",
 608              c->choose_total_tries);
 609
 610         ceph_decode_need(p, end, sizeof(u32), done);
 611         c->chooseleaf_descend_once = ceph_decode_32(p);
 612         dout("crush decode tunable chooseleaf_descend_once = %d\n",
 613              c->chooseleaf_descend_once);
 614
 615         ceph_decode_need(p, end, sizeof(u8), done);
 616         c->chooseleaf_vary_r = ceph_decode_8(p);
 617         dout("crush decode tunable chooseleaf_vary_r = %d\n",
 618              c->chooseleaf_vary_r);
 619
 620         /* skip straw_calc_version, allowed_bucket_algs */
 621         ceph_decode_need(p, end, sizeof(u8) + sizeof(u32), done);
 622         *p += sizeof(u8) + sizeof(u32);
 623
 624         ceph_decode_need(p, end, sizeof(u8), done);
 625         c->chooseleaf_stable = ceph_decode_8(p);
 626         dout("crush decode tunable chooseleaf_stable = %d\n",
 627              c->chooseleaf_stable);
 628
 629         if (*p != end) {
 630                 /* class_map */
 631                 ceph_decode_skip_map(p, end, 32, 32, bad);
 632                 /* class_name */
 633                 ceph_decode_skip_map(p, end, 32, string, bad);
 634                 /* class_bucket */
 635                 ceph_decode_skip_map_of_map(p, end, 32, 32, 32, bad);
 636         }
 637
 638         if (*p != end) {
 639                 err = decode_choose_args(p, end, c);
 640                 if (err)
 641                         goto fail;
 642         }
 643
 644 done:
 645         crush_finalize(c);
 646         dout("crush_decode success\n");
 647         return c;
 648
 649 badmem:
 650         err = -ENOMEM;
 651 fail:
 652         dout("crush_decode fail %d\n", err);
 653         crush_destroy(c);
 654         return ERR_PTR(err);
 655
 656 bad:
 657         err = -EINVAL;
 658         goto fail;
 659 }
 660
 661 int ceph_pg_compare(const struct ceph_pg *lhs, const struct ceph_pg *rhs)
 662 {
 663         if (lhs->pool < rhs->pool)
 664                 return -1;
 665         if (lhs->pool > rhs->pool)
 666                 return 1;
 667         if (lhs->seed < rhs->seed)
 668                 return -1;
 669         if (lhs->seed > rhs->seed)
 670                 return 1;
 671
 672         return 0;
 673 }
 674
 675 int ceph_spg_compare(const struct ceph_spg *lhs, const struct ceph_spg *rhs)
 676 {
 677         int ret;
 678
 679         ret = ceph_pg_compare(&lhs->pgid, &rhs->pgid);
 680         if (ret)
 681                 return ret;
 682
 683         if (lhs->shard < rhs->shard)
 684                 return -1;
 685         if (lhs->shard > rhs->shard)
 686                 return 1;
 687
 688         return 0;
 689 }
 690
 691 static struct ceph_pg_mapping *alloc_pg_mapping(size_t payload_len)
 692 {
 693         struct ceph_pg_mapping *pg;
 694
 695         pg = kmalloc(sizeof(*pg) + payload_len, GFP_NOIO);
 696         if (!pg)
 697                 return NULL;
 698
 699         RB_CLEAR_NODE(&pg->node);
 700         return pg;
 701 }
 702
 703 static void free_pg_mapping(struct ceph_pg_mapping *pg)
 704 {
 705         WARN_ON(!RB_EMPTY_NODE(&pg->node));
 706
 707         kfree(pg);
 708 }
 709
 710 /*
 711  * rbtree of pg_mapping for handling pg_temp (explicit mapping of pgid
 712  * to a set of osds) and primary_temp (explicit primary setting)
 713  */
 714 DEFINE_RB_FUNCS2(pg_mapping, struct ceph_pg_mapping, pgid, ceph_pg_compare,
 715                  RB_BYPTR, const struct ceph_pg *, node)
 716
 717 /*
 718  * rbtree of pg pool info
 719  */
 720 DEFINE_RB_FUNCS(pg_pool, struct ceph_pg_pool_info, id, node)
 721
 722 struct ceph_pg_pool_info *ceph_pg_pool_by_id(struct ceph_osdmap *map, u64 id)
 723 {
 724         return lookup_pg_pool(&map->pg_pools, id);
 725 }
 726
 727 const char *ceph_pg_pool_name_by_id(struct ceph_osdmap *map, u64 id)
 728 {
 729         struct ceph_pg_pool_info *pi;
 730
 731         if (id == CEPH_NOPOOL)
 732                 return NULL;
 733
 734         if (WARN_ON_ONCE(id > (u64) INT_MAX))
 735                 return NULL;
 736
 737         pi = lookup_pg_pool(&map->pg_pools, id);
 738         return pi ? pi->name : NULL;
 739 }
 740 EXPORT_SYMBOL(ceph_pg_pool_name_by_id);
 741
 742 int ceph_pg_poolid_by_name(struct ceph_osdmap *map, const char *name)
 743 {
 744         struct rb_node *rbp;
 745
 746         for (rbp = rb_first(&map->pg_pools); rbp; rbp = rb_next(rbp)) {
 747                 struct ceph_pg_pool_info *pi =
 748                         rb_entry(rbp, struct ceph_pg_pool_info, node);
 749                 if (pi->name && strcmp(pi->name, name) == 0)
 750                         return pi->id;
 751         }
 752         return -ENOENT;
 753 }
 754 EXPORT_SYMBOL(ceph_pg_poolid_by_name);
 755
 756 u64 ceph_pg_pool_flags(struct ceph_osdmap *map, u64 id)
 757 {
 758         struct ceph_pg_pool_info *pi;
 759
 760         pi = lookup_pg_pool(&map->pg_pools, id);
 761         return pi ? pi->flags : 0;
 762 }
 763 EXPORT_SYMBOL(ceph_pg_pool_flags);
 764
 765 static void __remove_pg_pool(struct rb_root *root, struct ceph_pg_pool_info *pi)
 766 {
 767         erase_pg_pool(root, pi);
 768         kfree(pi->name);
 769         kfree(pi);
 770 }
 771
 772 static int decode_pool(void **p, void *end, struct ceph_pg_pool_info *pi)
 773 {
 774         u8 ev, cv;
 775         unsigned len, num;
 776         void *pool_end;
 777
 778         ceph_decode_need(p, end, 2 + 4, bad);
 779         ev = ceph_decode_8(p);  /* encoding version */
 780         cv = ceph_decode_8(p); /* compat version */
 781         if (ev < 5) {
 782                 pr_warn("got v %d < 5 cv %d of ceph_pg_pool\n", ev, cv);
 783                 return -EINVAL;
 784         }
 785         if (cv > 9) {
 786                 pr_warn("got v %d cv %d > 9 of ceph_pg_pool\n", ev, cv);
 787                 return -EINVAL;
 788         }
 789         len = ceph_decode_32(p);
 790         ceph_decode_need(p, end, len, bad);
 791         pool_end = *p + len;
 792
 793         pi->type = ceph_decode_8(p);
 794         pi->size = ceph_decode_8(p);
 795         pi->crush_ruleset = ceph_decode_8(p);
 796         pi->object_hash = ceph_decode_8(p);
 797
 798         pi->pg_num = ceph_decode_32(p);
 799         pi->pgp_num = ceph_decode_32(p);
 800
 801         *p += 4 + 4;  /* skip lpg* */
 802         *p += 4;      /* skip last_change */
 803         *p += 8 + 4;  /* skip snap_seq, snap_epoch */
 804
 805         /* skip snaps */
 806         num = ceph_decode_32(p);
 807         while (num--) {
 808                 *p += 8;  /* snapid key */
 809                 *p += 1 + 1; /* versions */
 810                 len = ceph_decode_32(p);
 811                 *p += len;
 812         }
 813
 814         /* skip removed_snaps */
 815         num = ceph_decode_32(p);
 816         *p += num * (8 + 8);
 817
 818         *p += 8;  /* skip auid */
 819         pi->flags = ceph_decode_64(p);
 820         *p += 4;  /* skip crash_replay_interval */
 821
 822         if (ev >= 7)
 823                 pi->min_size = ceph_decode_8(p);
 824         else
 825                 pi->min_size = pi->size - pi->size / 2;
 826
 827         if (ev >= 8)
 828                 *p += 8 + 8;  /* skip quota_max_* */
 829
 830         if (ev >= 9) {
 831                 /* skip tiers */
 832                 num = ceph_decode_32(p);
 833                 *p += num * 8;
 834
 835                 *p += 8;  /* skip tier_of */
 836                 *p += 1;  /* skip cache_mode */
 837
 838                 pi->read_tier = ceph_decode_64(p);
 839                 pi->write_tier = ceph_decode_64(p);
 840         } else {
 841                 pi->read_tier = -1;
 842                 pi->write_tier = -1;
 843         }
 844
 845         if (ev >= 10) {
 846                 /* skip properties */
 847                 num = ceph_decode_32(p);
 848                 while (num--) {
 849                         len = ceph_decode_32(p);
 850                         *p += len; /* key */
 851                         len = ceph_decode_32(p);
 852                         *p += len; /* val */
 853                 }
 854         }
 855
 856         if (ev >= 11) {
 857                 /* skip hit_set_params */
 858                 *p += 1 + 1; /* versions */
 859                 len = ceph_decode_32(p);
 860                 *p += len;
 861
 862                 *p += 4; /* skip hit_set_period */
 863                 *p += 4; /* skip hit_set_count */
 864         }
 865
 866         if (ev >= 12)
 867                 *p += 4; /* skip stripe_width */
 868
 869         if (ev >= 13) {
 870                 *p += 8; /* skip target_max_bytes */
 871                 *p += 8; /* skip target_max_objects */
 872                 *p += 4; /* skip cache_target_dirty_ratio_micro */
 873                 *p += 4; /* skip cache_target_full_ratio_micro */
 874                 *p += 4; /* skip cache_min_flush_age */
 875                 *p += 4; /* skip cache_min_evict_age */
 876         }
 877
 878         if (ev >=  14) {
 879                 /* skip erasure_code_profile */
 880                 len = ceph_decode_32(p);
 881                 *p += len;
 882         }
 883
 884         /*
 885          * last_force_op_resend_preluminous, will be overridden if the
 886          * map was encoded with RESEND_ON_SPLIT
 887          */
 888         if (ev >= 15)
 889                 pi->last_force_request_resend = ceph_decode_32(p);
 890         else
 891                 pi->last_force_request_resend = 0;
 892
 893         if (ev >= 16)
 894                 *p += 4; /* skip min_read_recency_for_promote */
 895
 896         if (ev >= 17)
 897                 *p += 8; /* skip expected_num_objects */
 898
 899         if (ev >= 19)
 900                 *p += 4; /* skip cache_target_dirty_high_ratio_micro */
 901
 902         if (ev >= 20)
 903                 *p += 4; /* skip min_write_recency_for_promote */
 904
 905         if (ev >= 21)
 906                 *p += 1; /* skip use_gmt_hitset */
 907
 908         if (ev >= 22)
 909                 *p += 1; /* skip fast_read */
 910
 911         if (ev >= 23) {
 912                 *p += 4; /* skip hit_set_grade_decay_rate */
 913                 *p += 4; /* skip hit_set_search_last_n */
 914         }
 915
 916         if (ev >= 24) {
 917                 /* skip opts */
 918                 *p += 1 + 1; /* versions */
 919                 len = ceph_decode_32(p);
 920                 *p += len;
 921         }
 922
 923         if (ev >= 25)
 924                 pi->last_force_request_resend = ceph_decode_32(p);
 925
 926         /* ignore the rest */
 927
 928         *p = pool_end;
 929         calc_pg_masks(pi);
 930         return 0;
 931
 932 bad:
 933         return -EINVAL;
 934 }
 935
 936 static int decode_pool_names(void **p, void *end, struct ceph_osdmap *map)
 937 {
 938         struct ceph_pg_pool_info *pi;
 939         u32 num, len;
 940         u64 pool;
 941
 942         ceph_decode_32_safe(p, end, num, bad);
 943         dout(" %d pool names\n", num);
 944         while (num--) {
 945                 ceph_decode_64_safe(p, end, pool, bad);
 946                 ceph_decode_32_safe(p, end, len, bad);
 947                 dout("  pool %llu len %d\n", pool, len);
 948                 ceph_decode_need(p, end, len, bad);
 949                 pi = lookup_pg_pool(&map->pg_pools, pool);
 950                 if (pi) {
 951                         char *name = kstrndup(*p, len, GFP_NOFS);
 952
 953                         if (!name)
 954                                 return -ENOMEM;
 955                         kfree(pi->name);
 956                         pi->name = name;
 957                         dout("  name is %s\n", pi->name);
 958                 }
 959                 *p += len;
 960         }
 961         return 0;
 962
 963 bad:
 964         return -EINVAL;
 965 }
 966
 967 /*
 968  * CRUSH workspaces
 969  *
 970  * workspace_manager framework borrowed from fs/btrfs/compression.c.
 971  * Two simplifications: there is only one type of workspace and there
 972  * is always at least one workspace.
 973  */
 974 static struct crush_work *alloc_workspace(const struct crush_map *c)
 975 {
 976         struct crush_work *work;
 977         size_t work_size;
 978
 979         WARN_ON(!c->working_size);
 980         work_size = crush_work_size(c, CEPH_PG_MAX_SIZE);
 981         dout("%s work_size %zu bytes\n", __func__, work_size);
 982
 983         work = ceph_kvmalloc(work_size, GFP_NOIO);
 984         if (!work)
 985                 return NULL;
 986
 987         INIT_LIST_HEAD(&work->item);
 988         crush_init_workspace(c, work);
 989         return work;
 990 }
 991
 992 static void free_workspace(struct crush_work *work)
 993 {
 994         WARN_ON(!list_empty(&work->item));
 995         kvfree(work);
 996 }
 997
 998 static void init_workspace_manager(struct workspace_manager *wsm)
 999 {
1000         INIT_LIST_HEAD(&wsm->idle_ws);
1001         spin_lock_init(&wsm->ws_lock);
1002         atomic_set(&wsm->total_ws, 0);
1003         wsm->free_ws = 0;
1004         init_waitqueue_head(&wsm->ws_wait);
1005 }
1006
1007 static void add_initial_workspace(struct workspace_manager *wsm,
1008                                   struct crush_work *work)
1009 {
1010         WARN_ON(!list_empty(&wsm->idle_ws));
1011
1012         list_add(&work->item, &wsm->idle_ws);
1013         atomic_set(&wsm->total_ws, 1);
1014         wsm->free_ws = 1;
1015 }
1016
1017 static void cleanup_workspace_manager(struct workspace_manager *wsm)
1018 {
1019         struct crush_work *work;
1020
1021         while (!list_empty(&wsm->idle_ws)) {
1022                 work = list_first_entry(&wsm->idle_ws, struct crush_work,
1023                                         item);
1024                 list_del_init(&work->item);
1025                 free_workspace(work);
1026         }
1027         atomic_set(&wsm->total_ws, 0);
1028         wsm->free_ws = 0;
1029 }
1030
1031 /*
1032  * Finds an available workspace or allocates a new one.  If it's not
1033  * possible to allocate a new one, waits until there is one.
1034  */
1035 static struct crush_work *get_workspace(struct workspace_manager *wsm,
1036                                         const struct crush_map *c)
1037 {
1038         struct crush_work *work;
1039         int cpus = num_online_cpus();
1040
1041 again:
1042         spin_lock(&wsm->ws_lock);
1043         if (!list_empty(&wsm->idle_ws)) {
1044                 work = list_first_entry(&wsm->idle_ws, struct crush_work,
1045                                         item);
1046                 list_del_init(&work->item);
1047                 wsm->free_ws--;
1048                 spin_unlock(&wsm->ws_lock);
1049                 return work;
1050
1051         }
1052         if (atomic_read(&wsm->total_ws) > cpus) {
1053                 DEFINE_WAIT(wait);
1054
1055                 spin_unlock(&wsm->ws_lock);
1056                 prepare_to_wait(&wsm->ws_wait, &wait, TASK_UNINTERRUPTIBLE);
1057                 if (atomic_read(&wsm->total_ws) > cpus && !wsm->free_ws)
1058                         schedule();
1059                 finish_wait(&wsm->ws_wait, &wait);
1060                 goto again;
1061         }
1062         atomic_inc(&wsm->total_ws);
1063         spin_unlock(&wsm->ws_lock);
1064
1065         work = alloc_workspace(c);
1066         if (!work) {
1067                 atomic_dec(&wsm->total_ws);
1068                 wake_up(&wsm->ws_wait);
1069
1070                 /*
1071                  * Do not return the error but go back to waiting.  We
1072                  * have the initial workspace and the CRUSH computation
1073                  * time is bounded so we will get it eventually.
1074                  */
1075                 WARN_ON(atomic_read(&wsm->total_ws) < 1);
1076                 goto again;
1077         }
1078         return work;
1079 }
1080
1081 /*
1082  * Puts a workspace back on the list or frees it if we have enough
1083  * idle ones sitting around.
1084  */
1085 static void put_workspace(struct workspace_manager *wsm,
1086                           struct crush_work *work)
1087 {
1088         spin_lock(&wsm->ws_lock);
1089         if (wsm->free_ws <= num_online_cpus()) {
1090                 list_add(&work->item, &wsm->idle_ws);
1091                 wsm->free_ws++;
1092                 spin_unlock(&wsm->ws_lock);
1093                 goto wake;
1094         }
1095         spin_unlock(&wsm->ws_lock);
1096
1097         free_workspace(work);
1098         atomic_dec(&wsm->total_ws);
1099 wake:
1100         if (wq_has_sleeper(&wsm->ws_wait))
1101                 wake_up(&wsm->ws_wait);
1102 }
1103
1104 /*
1105  * osd map
1106  */
1107 struct ceph_osdmap *ceph_osdmap_alloc(void)
1108 {
1109         struct ceph_osdmap *map;
1110
1111         map = kzalloc(sizeof(*map), GFP_NOIO);
1112         if (!map)
1113                 return NULL;
1114
1115         map->pg_pools = RB_ROOT;
1116         map->pool_max = -1;
1117         map->pg_temp = RB_ROOT;
1118         map->primary_temp = RB_ROOT;
1119         map->pg_upmap = RB_ROOT;
1120         map->pg_upmap_items = RB_ROOT;
1121
1122         init_workspace_manager(&map->crush_wsm);
1123
1124         return map;
1125 }
1126
1127 void ceph_osdmap_destroy(struct ceph_osdmap *map)
1128 {
1129         dout("osdmap_destroy %p\n", map);
1130
1131         if (map->crush)
1132                 crush_destroy(map->crush);
1133         cleanup_workspace_manager(&map->crush_wsm);
1134
1135         while (!RB_EMPTY_ROOT(&map->pg_temp)) {
1136                 struct ceph_pg_mapping *pg =
1137                         rb_entry(rb_first(&map->pg_temp),
1138                                  struct ceph_pg_mapping, node);
1139                 erase_pg_mapping(&map->pg_temp, pg);
1140                 free_pg_mapping(pg);
1141         }
1142         while (!RB_EMPTY_ROOT(&map->primary_temp)) {
1143                 struct ceph_pg_mapping *pg =
1144                         rb_entry(rb_first(&map->primary_temp),
1145                                  struct ceph_pg_mapping, node);
1146                 erase_pg_mapping(&map->primary_temp, pg);
1147                 free_pg_mapping(pg);
1148         }
1149         while (!RB_EMPTY_ROOT(&map->pg_upmap)) {
1150                 struct ceph_pg_mapping *pg =
1151                         rb_entry(rb_first(&map->pg_upmap),
1152                                  struct ceph_pg_mapping, node);
1153                 rb_erase(&pg->node, &map->pg_upmap);
1154                 kfree(pg);
1155         }
1156         while (!RB_EMPTY_ROOT(&map->pg_upmap_items)) {
1157                 struct ceph_pg_mapping *pg =
1158                         rb_entry(rb_first(&map->pg_upmap_items),
1159                                  struct ceph_pg_mapping, node);
1160                 rb_erase(&pg->node, &map->pg_upmap_items);
1161                 kfree(pg);
1162         }
1163         while (!RB_EMPTY_ROOT(&map->pg_pools)) {
1164                 struct ceph_pg_pool_info *pi =
1165                         rb_entry(rb_first(&map->pg_pools),
1166                                  struct ceph_pg_pool_info, node);
1167                 __remove_pg_pool(&map->pg_pools, pi);
1168         }
1169         kvfree(map->osd_state);
1170         kvfree(map->osd_weight);
1171         kvfree(map->osd_addr);
1172         kvfree(map->osd_primary_affinity);
1173         kfree(map);
1174 }
1175
1176 /*
1177  * Adjust max_osd value, (re)allocate arrays.
1178  *
1179  * The new elements are properly initialized.
1180  */
1181 static int osdmap_set_max_osd(struct ceph_osdmap *map, u32 max)
1182 {
1183         u32 *state;
1184         u32 *weight;
1185         struct ceph_entity_addr *addr;
1186         u32 to_copy;
1187         int i;
1188
1189         dout("%s old %u new %u\n", __func__, map->max_osd, max);
1190         if (max == map->max_osd)
1191                 return 0;
1192
1193         state = ceph_kvmalloc(array_size(max, sizeof(*state)), GFP_NOFS);
1194         weight = ceph_kvmalloc(array_size(max, sizeof(*weight)), GFP_NOFS);
1195         addr = ceph_kvmalloc(array_size(max, sizeof(*addr)), GFP_NOFS);
1196         if (!state || !weight || !addr) {
1197                 kvfree(state);
1198                 kvfree(weight);
1199                 kvfree(addr);
1200                 return -ENOMEM;
1201         }
1202
1203         to_copy = min(map->max_osd, max);
1204         if (map->osd_state) {
1205                 memcpy(state, map->osd_state, to_copy * sizeof(*state));
1206                 memcpy(weight, map->osd_weight, to_copy * sizeof(*weight));
1207                 memcpy(addr, map->osd_addr, to_copy * sizeof(*addr));
1208                 kvfree(map->osd_state);
1209                 kvfree(map->osd_weight);
1210                 kvfree(map->osd_addr);
1211         }
1212
1213         map->osd_state = state;
1214         map->osd_weight = weight;
1215         map->osd_addr = addr;
1216         for (i = map->max_osd; i < max; i++) {
1217                 map->osd_state[i] = 0;
1218                 map->osd_weight[i] = CEPH_OSD_OUT;
1219                 memset(map->osd_addr + i, 0, sizeof(*map->osd_addr));
1220         }
1221
1222         if (map->osd_primary_affinity) {
1223                 u32 *affinity;
1224
1225                 affinity = ceph_kvmalloc(array_size(max, sizeof(*affinity)),
1226                                          GFP_NOFS);
1227                 if (!affinity)
1228                         return -ENOMEM;
1229
1230                 memcpy(affinity, map->osd_primary_affinity,
1231                        to_copy * sizeof(*affinity));
1232                 kvfree(map->osd_primary_affinity);
1233
1234                 map->osd_primary_affinity = affinity;
1235                 for (i = map->max_osd; i < max; i++)
1236                         map->osd_primary_affinity[i] =
1237                             CEPH_OSD_DEFAULT_PRIMARY_AFFINITY;
1238         }
1239
1240         map->max_osd = max;
1241
1242         return 0;
1243 }
1244
1245 static int osdmap_set_crush(struct ceph_osdmap *map, struct crush_map *crush)
1246 {
1247         struct crush_work *work;
1248
1249         if (IS_ERR(crush))
1250                 return PTR_ERR(crush);
1251
1252         work = alloc_workspace(crush);
1253         if (!work) {
1254                 crush_destroy(crush);
1255                 return -ENOMEM;
1256         }
1257
1258         if (map->crush)
1259                 crush_destroy(map->crush);
1260         cleanup_workspace_manager(&map->crush_wsm);
1261         map->crush = crush;
1262         add_initial_workspace(&map->crush_wsm, work);
1263         return 0;
1264 }
1265
1266 #define OSDMAP_WRAPPER_COMPAT_VER       7
1267 #define OSDMAP_CLIENT_DATA_COMPAT_VER   1
1268
1269 /*
1270  * Return 0 or error.  On success, *v is set to 0 for old (v6) osdmaps,
1271  * to struct_v of the client_data section for new (v7 and above)
1272  * osdmaps.
1273  */
1274 static int get_osdmap_client_data_v(void **p, void *end,
1275                                     const char *prefix, u8 *v)
1276 {
1277         u8 struct_v;
1278
1279         ceph_decode_8_safe(p, end, struct_v, e_inval);
1280         if (struct_v >= 7) {
1281                 u8 struct_compat;
1282
1283                 ceph_decode_8_safe(p, end, struct_compat, e_inval);
1284                 if (struct_compat > OSDMAP_WRAPPER_COMPAT_VER) {
1285                         pr_warn("got v %d cv %d > %d of %s ceph_osdmap\n",
1286                                 struct_v, struct_compat,
1287                                 OSDMAP_WRAPPER_COMPAT_VER, prefix);
1288                         return -EINVAL;
1289                 }
1290                 *p += 4; /* ignore wrapper struct_len */
1291
1292                 ceph_decode_8_safe(p, end, struct_v, e_inval);
1293                 ceph_decode_8_safe(p, end, struct_compat, e_inval);
1294                 if (struct_compat > OSDMAP_CLIENT_DATA_COMPAT_VER) {
1295                         pr_warn("got v %d cv %d > %d of %s ceph_osdmap client data\n",
1296                                 struct_v, struct_compat,
1297                                 OSDMAP_CLIENT_DATA_COMPAT_VER, prefix);
1298                         return -EINVAL;
1299                 }
1300                 *p += 4; /* ignore client data struct_len */
1301         } else {
1302                 u16 version;
1303
1304                 *p -= 1;
1305                 ceph_decode_16_safe(p, end, version, e_inval);
1306                 if (version < 6) {
1307                         pr_warn("got v %d < 6 of %s ceph_osdmap\n",
1308                                 version, prefix);
1309                         return -EINVAL;
1310                 }
1311
1312                 /* old osdmap encoding */
1313                 struct_v = 0;
1314         }
1315
1316         *v = struct_v;
1317         return 0;
1318
1319 e_inval:
1320         return -EINVAL;
1321 }
1322
1323 static int __decode_pools(void **p, void *end, struct ceph_osdmap *map,
1324                           bool incremental)
1325 {
1326         u32 n;
1327
1328         ceph_decode_32_safe(p, end, n, e_inval);
1329         while (n--) {
1330                 struct ceph_pg_pool_info *pi;
1331                 u64 pool;
1332                 int ret;
1333
1334                 ceph_decode_64_safe(p, end, pool, e_inval);
1335
1336                 pi = lookup_pg_pool(&map->pg_pools, pool);
1337                 if (!incremental || !pi) {
1338                         pi = kzalloc(sizeof(*pi), GFP_NOFS);
1339                         if (!pi)
1340                                 return -ENOMEM;
1341
1342                         RB_CLEAR_NODE(&pi->node);
1343                         pi->id = pool;
1344
1345                         if (!__insert_pg_pool(&map->pg_pools, pi)) {
1346                                 kfree(pi);
1347                                 return -EEXIST;
1348                         }
1349                 }
1350
1351                 ret = decode_pool(p, end, pi);
1352                 if (ret)
1353                         return ret;
1354         }
1355
1356         return 0;
1357
1358 e_inval:
1359         return -EINVAL;
1360 }
1361
1362 static int decode_pools(void **p, void *end, struct ceph_osdmap *map)
1363 {
1364         return __decode_pools(p, end, map, false);
1365 }
1366
1367 static int decode_new_pools(void **p, void *end, struct ceph_osdmap *map)
1368 {
1369         return __decode_pools(p, end, map, true);
1370 }
1371
1372 typedef struct ceph_pg_mapping *(*decode_mapping_fn_t)(void **, void *, bool);
1373
1374 static int decode_pg_mapping(void **p, void *end, struct rb_root *mapping_root,
1375                              decode_mapping_fn_t fn, bool incremental)
1376 {
1377         u32 n;
1378
1379         WARN_ON(!incremental && !fn);
1380
1381         ceph_decode_32_safe(p, end, n, e_inval);
1382         while (n--) {
1383                 struct ceph_pg_mapping *pg;
1384                 struct ceph_pg pgid;
1385                 int ret;
1386
1387                 ret = ceph_decode_pgid(p, end, &pgid);
1388                 if (ret)
1389                         return ret;
1390
1391                 pg = lookup_pg_mapping(mapping_root, &pgid);
1392                 if (pg) {
1393                         WARN_ON(!incremental);
1394                         erase_pg_mapping(mapping_root, pg);
1395                         free_pg_mapping(pg);
1396                 }
1397
1398                 if (fn) {
1399                         pg = fn(p, end, incremental);
1400                         if (IS_ERR(pg))
1401                                 return PTR_ERR(pg);
1402
1403                         if (pg) {
1404                                 pg->pgid = pgid; /* struct */
1405                                 insert_pg_mapping(mapping_root, pg);
1406                         }
1407                 }
1408         }
1409
1410         return 0;
1411
1412 e_inval:
1413         return -EINVAL;
1414 }
1415
1416 static struct ceph_pg_mapping *__decode_pg_temp(void **p, void *end,
1417                                                 bool incremental)
1418 {
1419         struct ceph_pg_mapping *pg;
1420         u32 len, i;
1421
1422         ceph_decode_32_safe(p, end, len, e_inval);
1423         if (len == 0 && incremental)
1424                 return NULL;    /* new_pg_temp: [] to remove */
1425         if (len > (SIZE_MAX - sizeof(*pg)) / sizeof(u32))
1426                 return ERR_PTR(-EINVAL);
1427
1428         ceph_decode_need(p, end, len * sizeof(u32), e_inval);
1429         pg = alloc_pg_mapping(len * sizeof(u32));
1430         if (!pg)
1431                 return ERR_PTR(-ENOMEM);
1432
1433         pg->pg_temp.len = len;
1434         for (i = 0; i < len; i++)
1435                 pg->pg_temp.osds[i] = ceph_decode_32(p);
1436
1437         return pg;
1438
1439 e_inval:
1440         return ERR_PTR(-EINVAL);
1441 }
1442
1443 static int decode_pg_temp(void **p, void *end, struct ceph_osdmap *map)
1444 {
1445         return decode_pg_mapping(p, end, &map->pg_temp, __decode_pg_temp,
1446                                  false);
1447 }
1448
1449 static int decode_new_pg_temp(void **p, void *end, struct ceph_osdmap *map)
1450 {
1451         return decode_pg_mapping(p, end, &map->pg_temp, __decode_pg_temp,
1452                                  true);
1453 }
1454
1455 static struct ceph_pg_mapping *__decode_primary_temp(void **p, void *end,
1456                                                      bool incremental)
1457 {
1458         struct ceph_pg_mapping *pg;
1459         u32 osd;
1460
1461         ceph_decode_32_safe(p, end, osd, e_inval);
1462         if (osd == (u32)-1 && incremental)
1463                 return NULL;    /* new_primary_temp: -1 to remove */
1464
1465         pg = alloc_pg_mapping(0);
1466         if (!pg)
1467                 return ERR_PTR(-ENOMEM);
1468
1469         pg->primary_temp.osd = osd;
1470         return pg;
1471
1472 e_inval:
1473         return ERR_PTR(-EINVAL);
1474 }
1475
1476 static int decode_primary_temp(void **p, void *end, struct ceph_osdmap *map)
1477 {
1478         return decode_pg_mapping(p, end, &map->primary_temp,
1479                                  __decode_primary_temp, false);
1480 }
1481
1482 static int decode_new_primary_temp(void **p, void *end,
1483                                    struct ceph_osdmap *map)
1484 {
1485         return decode_pg_mapping(p, end, &map->primary_temp,
1486                                  __decode_primary_temp, true);
1487 }
1488
1489 u32 ceph_get_primary_affinity(struct ceph_osdmap *map, int osd)
1490 {
1491         BUG_ON(osd >= map->max_osd);
1492
1493         if (!map->osd_primary_affinity)
1494                 return CEPH_OSD_DEFAULT_PRIMARY_AFFINITY;
1495
1496         return map->osd_primary_affinity[osd];
1497 }
1498
1499 static int set_primary_affinity(struct ceph_osdmap *map, int osd, u32 aff)
1500 {
1501         BUG_ON(osd >= map->max_osd);
1502
1503         if (!map->osd_primary_affinity) {
1504                 int i;
1505
1506                 map->osd_primary_affinity = ceph_kvmalloc(
1507                     array_size(map->max_osd, sizeof(*map->osd_primary_affinity)),
1508                     GFP_NOFS);
1509                 if (!map->osd_primary_affinity)
1510                         return -ENOMEM;
1511
1512                 for (i = 0; i < map->max_osd; i++)
1513                         map->osd_primary_affinity[i] =
1514                             CEPH_OSD_DEFAULT_PRIMARY_AFFINITY;
1515         }
1516
1517         map->osd_primary_affinity[osd] = aff;
1518
1519         return 0;
1520 }
1521
1522 static int decode_primary_affinity(void **p, void *end,
1523                                    struct ceph_osdmap *map)
1524 {
1525         u32 len, i;
1526
1527         ceph_decode_32_safe(p, end, len, e_inval);
1528         if (len == 0) {
1529                 kvfree(map->osd_primary_affinity);
1530                 map->osd_primary_affinity = NULL;
1531                 return 0;
1532         }
1533         if (len != map->max_osd)
1534                 goto e_inval;
1535
1536         ceph_decode_need(p, end, map->max_osd*sizeof(u32), e_inval);
1537
1538         for (i = 0; i < map->max_osd; i++) {
1539                 int ret;
1540
1541                 ret = set_primary_affinity(map, i, ceph_decode_32(p));
1542                 if (ret)
1543                         return ret;
1544         }
1545
1546         return 0;
1547
1548 e_inval:
1549         return -EINVAL;
1550 }
1551
1552 static int decode_new_primary_affinity(void **p, void *end,
1553                                        struct ceph_osdmap *map)
1554 {
1555         u32 n;
1556
1557         ceph_decode_32_safe(p, end, n, e_inval);
1558         while (n--) {
1559                 u32 osd, aff;
1560                 int ret;
1561
1562                 ceph_decode_32_safe(p, end, osd, e_inval);
1563                 ceph_decode_32_safe(p, end, aff, e_inval);
1564
1565                 ret = set_primary_affinity(map, osd, aff);
1566                 if (ret)
1567                         return ret;
1568
1569                 pr_info("osd%d primary-affinity 0x%x\n", osd, aff);
1570         }
1571
1572         return 0;
1573
1574 e_inval:
1575         return -EINVAL;
1576 }
1577
1578 static struct ceph_pg_mapping *__decode_pg_upmap(void **p, void *end,
1579                                                  bool __unused)
1580 {
1581         return __decode_pg_temp(p, end, false);
1582 }
1583
1584 static int decode_pg_upmap(void **p, void *end, struct ceph_osdmap *map)
1585 {
1586         return decode_pg_mapping(p, end, &map->pg_upmap, __decode_pg_upmap,
1587                                  false);
1588 }
1589
1590 static int decode_new_pg_upmap(void **p, void *end, struct ceph_osdmap *map)
1591 {
1592         return decode_pg_mapping(p, end, &map->pg_upmap, __decode_pg_upmap,
1593                                  true);
1594 }
1595
1596 static int decode_old_pg_upmap(void **p, void *end, struct ceph_osdmap *map)
1597 {
1598         return decode_pg_mapping(p, end, &map->pg_upmap, NULL, true);
1599 }
1600
1601 static struct ceph_pg_mapping *__decode_pg_upmap_items(void **p, void *end,
1602                                                        bool __unused)
1603 {
1604         struct ceph_pg_mapping *pg;
1605         u32 len, i;
1606
1607         ceph_decode_32_safe(p, end, len, e_inval);
1608         if (len > (SIZE_MAX - sizeof(*pg)) / (2 * sizeof(u32)))
1609                 return ERR_PTR(-EINVAL);
1610
1611         ceph_decode_need(p, end, 2 * len * sizeof(u32), e_inval);
1612         pg = alloc_pg_mapping(2 * len * sizeof(u32));
1613         if (!pg)
1614                 return ERR_PTR(-ENOMEM);
1615
1616         pg->pg_upmap_items.len = len;
1617         for (i = 0; i < len; i++) {
1618                 pg->pg_upmap_items.from_to[i][0] = ceph_decode_32(p);
1619                 pg->pg_upmap_items.from_to[i][1] = ceph_decode_32(p);
1620         }
1621
1622         return pg;
1623
1624 e_inval:
1625         return ERR_PTR(-EINVAL);
1626 }
1627
1628 static int decode_pg_upmap_items(void **p, void *end, struct ceph_osdmap *map)
1629 {
1630         return decode_pg_mapping(p, end, &map->pg_upmap_items,
1631                                  __decode_pg_upmap_items, false);
1632 }
1633
1634 static int decode_new_pg_upmap_items(void **p, void *end,
1635                                      struct ceph_osdmap *map)
1636 {
1637         return decode_pg_mapping(p, end, &map->pg_upmap_items,
1638                                  __decode_pg_upmap_items, true);
1639 }
1640
1641 static int decode_old_pg_upmap_items(void **p, void *end,
1642                                      struct ceph_osdmap *map)
1643 {
1644         return decode_pg_mapping(p, end, &map->pg_upmap_items, NULL, true);
1645 }
1646
1647 /*
1648  * decode a full map.
1649  */
1650 static int osdmap_decode(void **p, void *end, bool msgr2,
1651                          struct ceph_osdmap *map)
1652 {
1653         u8 struct_v;
1654         u32 epoch = 0;
1655         void *start = *p;
1656         u32 max;
1657         u32 len, i;
1658         int err;
1659
1660         dout("%s %p to %p len %d\n", __func__, *p, end, (int)(end - *p));
1661
1662         err = get_osdmap_client_data_v(p, end, "full", &struct_v);
1663         if (err)
1664                 goto bad;
1665
1666         /* fsid, epoch, created, modified */
1667         ceph_decode_need(p, end, sizeof(map->fsid) + sizeof(u32) +
1668                          sizeof(map->created) + sizeof(map->modified), e_inval);
1669         ceph_decode_copy(p, &map->fsid, sizeof(map->fsid));
1670         epoch = map->epoch = ceph_decode_32(p);
1671         ceph_decode_copy(p, &map->created, sizeof(map->created));
1672         ceph_decode_copy(p, &map->modified, sizeof(map->modified));
1673
1674         /* pools */
1675         err = decode_pools(p, end, map);
1676         if (err)
1677                 goto bad;
1678
1679         /* pool_name */
1680         err = decode_pool_names(p, end, map);
1681         if (err)
1682                 goto bad;
1683
1684         ceph_decode_32_safe(p, end, map->pool_max, e_inval);
1685
1686         ceph_decode_32_safe(p, end, map->flags, e_inval);
1687
1688         /* max_osd */
1689         ceph_decode_32_safe(p, end, max, e_inval);
1690
1691         /* (re)alloc osd arrays */
1692         err = osdmap_set_max_osd(map, max);
1693         if (err)
1694                 goto bad;
1695
1696         /* osd_state, osd_weight, osd_addrs->client_addr */
1697         ceph_decode_need(p, end, 3*sizeof(u32) +
1698                          map->max_osd*(struct_v >= 5 ? sizeof(u32) :
1699                                                        sizeof(u8)) +
1700                                        sizeof(*map->osd_weight), e_inval);
1701         if (ceph_decode_32(p) != map->max_osd)
1702                 goto e_inval;
1703
1704         if (struct_v >= 5) {
1705                 for (i = 0; i < map->max_osd; i++)
1706                         map->osd_state[i] = ceph_decode_32(p);
1707         } else {
1708                 for (i = 0; i < map->max_osd; i++)
1709                         map->osd_state[i] = ceph_decode_8(p);
1710         }
1711
1712         if (ceph_decode_32(p) != map->max_osd)
1713                 goto e_inval;
1714
1715         for (i = 0; i < map->max_osd; i++)
1716                 map->osd_weight[i] = ceph_decode_32(p);
1717
1718         if (ceph_decode_32(p) != map->max_osd)
1719                 goto e_inval;
1720
1721         for (i = 0; i < map->max_osd; i++) {
1722                 struct ceph_entity_addr *addr = &map->osd_addr[i];
1723
1724                 if (struct_v >= 8)
1725                         err = ceph_decode_entity_addrvec(p, end, msgr2, addr);
1726                 else
1727                         err = ceph_decode_entity_addr(p, end, addr);
1728                 if (err)
1729                         goto bad;
1730
1731                 dout("%s osd%d addr %s\n", __func__, i, ceph_pr_addr(addr));
1732         }
1733
1734         /* pg_temp */
1735         err = decode_pg_temp(p, end, map);
1736         if (err)
1737                 goto bad;
1738
1739         /* primary_temp */
1740         if (struct_v >= 1) {
1741                 err = decode_primary_temp(p, end, map);
1742                 if (err)
1743                         goto bad;
1744         }
1745
1746         /* primary_affinity */
1747         if (struct_v >= 2) {
1748                 err = decode_primary_affinity(p, end, map);
1749                 if (err)
1750                         goto bad;
1751         } else {
1752                 WARN_ON(map->osd_primary_affinity);
1753         }
1754
1755         /* crush */
1756         ceph_decode_32_safe(p, end, len, e_inval);
1757         err = osdmap_set_crush(map, crush_decode(*p, min(*p + len, end)));
1758         if (err)
1759                 goto bad;
1760
1761         *p += len;
1762         if (struct_v >= 3) {
1763                 /* erasure_code_profiles */
1764                 ceph_decode_skip_map_of_map(p, end, string, string, string,
1765                                             e_inval);
1766         }
1767
1768         if (struct_v >= 4) {
1769                 err = decode_pg_upmap(p, end, map);
1770                 if (err)
1771                         goto bad;
1772
1773                 err = decode_pg_upmap_items(p, end, map);
1774                 if (err)
1775                         goto bad;
1776         } else {
1777                 WARN_ON(!RB_EMPTY_ROOT(&map->pg_upmap));
1778                 WARN_ON(!RB_EMPTY_ROOT(&map->pg_upmap_items));
1779         }
1780
1781         /* ignore the rest */
1782         *p = end;
1783
1784         dout("full osdmap epoch %d max_osd %d\n", map->epoch, map->max_osd);
1785         return 0;
1786
1787 e_inval:
1788         err = -EINVAL;
1789 bad:
1790         pr_err("corrupt full osdmap (%d) epoch %d off %d (%p of %p-%p)\n",
1791                err, epoch, (int)(*p - start), *p, start, end);
1792         print_hex_dump(KERN_DEBUG, "osdmap: ",
1793                        DUMP_PREFIX_OFFSET, 16, 1,
1794                        start, end - start, true);
1795         return err;
1796 }
1797
1798 /*
1799  * Allocate and decode a full map.
1800  */
1801 struct ceph_osdmap *ceph_osdmap_decode(void **p, void *end, bool msgr2)
1802 {
1803         struct ceph_osdmap *map;
1804         int ret;
1805
1806         map = ceph_osdmap_alloc();
1807         if (!map)
1808                 return ERR_PTR(-ENOMEM);
1809
1810         ret = osdmap_decode(p, end, msgr2, map);
1811         if (ret) {
1812                 ceph_osdmap_destroy(map);
1813                 return ERR_PTR(ret);
1814         }
1815
1816         return map;
1817 }
1818
1819 /*
1820  * Encoding order is (new_up_client, new_state, new_weight).  Need to
1821  * apply in the (new_weight, new_state, new_up_client) order, because
1822  * an incremental map may look like e.g.
1823  *
1824  *     new_up_client: { osd=6, addr=... } # set osd_state and addr
1825  *     new_state: { osd=6, xorstate=EXISTS } # clear osd_state
1826  */
1827 static int decode_new_up_state_weight(void **p, void *end, u8 struct_v,
1828                                       bool msgr2, struct ceph_osdmap *map)
1829 {
1830         void *new_up_client;
1831         void *new_state;
1832         void *new_weight_end;
1833         u32 len;
1834         int ret;
1835         int i;
1836
1837         new_up_client = *p;
1838         ceph_decode_32_safe(p, end, len, e_inval);
1839         for (i = 0; i < len; ++i) {
1840                 struct ceph_entity_addr addr;
1841
1842                 ceph_decode_skip_32(p, end, e_inval);
1843                 if (struct_v >= 7)
1844                         ret = ceph_decode_entity_addrvec(p, end, msgr2, &addr);
1845                 else
1846                         ret = ceph_decode_entity_addr(p, end, &addr);
1847                 if (ret)
1848                         return ret;
1849         }
1850
1851         new_state = *p;
1852         ceph_decode_32_safe(p, end, len, e_inval);
1853         len *= sizeof(u32) + (struct_v >= 5 ? sizeof(u32) : sizeof(u8));
1854         ceph_decode_need(p, end, len, e_inval);
1855         *p += len;
1856
1857         /* new_weight */
1858         ceph_decode_32_safe(p, end, len, e_inval);
1859         while (len--) {
1860                 s32 osd;
1861                 u32 w;
1862
1863                 ceph_decode_need(p, end, 2*sizeof(u32), e_inval);
1864                 osd = ceph_decode_32(p);
1865                 w = ceph_decode_32(p);
1866                 BUG_ON(osd >= map->max_osd);
1867                 pr_info("osd%d weight 0x%x %s\n", osd, w,
1868                      w == CEPH_OSD_IN ? "(in)" :
1869                      (w == CEPH_OSD_OUT ? "(out)" : ""));
1870                 map->osd_weight[osd] = w;
1871
1872                 /*
1873                  * If we are marking in, set the EXISTS, and clear the
1874                  * AUTOOUT and NEW bits.
1875                  */
1876                 if (w) {
1877                         map->osd_state[osd] |= CEPH_OSD_EXISTS;
1878                         map->osd_state[osd] &= ~(CEPH_OSD_AUTOOUT |
1879                                                  CEPH_OSD_NEW);
1880                 }
1881         }
1882         new_weight_end = *p;
1883
1884         /* new_state (up/down) */
1885         *p = new_state;
1886         len = ceph_decode_32(p);
1887         while (len--) {
1888                 s32 osd;
1889                 u32 xorstate;
1890
1891                 osd = ceph_decode_32(p);
1892                 if (struct_v >= 5)
1893                         xorstate = ceph_decode_32(p);
1894                 else
1895                         xorstate = ceph_decode_8(p);
1896                 if (xorstate == 0)
1897                         xorstate = CEPH_OSD_UP;
1898                 BUG_ON(osd >= map->max_osd);
1899                 if ((map->osd_state[osd] & CEPH_OSD_UP) &&
1900                     (xorstate & CEPH_OSD_UP))
1901                         pr_info("osd%d down\n", osd);
1902                 if ((map->osd_state[osd] & CEPH_OSD_EXISTS) &&
1903                     (xorstate & CEPH_OSD_EXISTS)) {
1904                         pr_info("osd%d does not exist\n", osd);
1905                         ret = set_primary_affinity(map, osd,
1906                                                    CEPH_OSD_DEFAULT_PRIMARY_AFFINITY);
1907                         if (ret)
1908                                 return ret;
1909                         memset(map->osd_addr + osd, 0, sizeof(*map->osd_addr));
1910                         map->osd_state[osd] = 0;
1911                 } else {
1912                         map->osd_state[osd] ^= xorstate;
1913                 }
1914         }
1915
1916         /* new_up_client */
1917         *p = new_up_client;
1918         len = ceph_decode_32(p);
1919         while (len--) {
1920                 s32 osd;
1921                 struct ceph_entity_addr addr;
1922
1923                 osd = ceph_decode_32(p);
1924                 BUG_ON(osd >= map->max_osd);
1925                 if (struct_v >= 7)
1926                         ret = ceph_decode_entity_addrvec(p, end, msgr2, &addr);
1927                 else
1928                         ret = ceph_decode_entity_addr(p, end, &addr);
1929                 if (ret)
1930                         return ret;
1931
1932                 dout("%s osd%d addr %s\n", __func__, osd, ceph_pr_addr(&addr));
1933
1934                 pr_info("osd%d up\n", osd);
1935                 map->osd_state[osd] |= CEPH_OSD_EXISTS | CEPH_OSD_UP;
1936                 map->osd_addr[osd] = addr;
1937         }
1938
1939         *p = new_weight_end;
1940         return 0;
1941
1942 e_inval:
1943         return -EINVAL;
1944 }
1945
1946 /*
1947  * decode and apply an incremental map update.
1948  */
1949 struct ceph_osdmap *osdmap_apply_incremental(void **p, void *end, bool msgr2,
1950                                              struct ceph_osdmap *map)
1951 {
1952         struct ceph_fsid fsid;
1953         u32 epoch = 0;
1954         struct ceph_timespec modified;
1955         s32 len;
1956         u64 pool;
1957         __s64 new_pool_max;
1958         __s32 new_flags, max;
1959         void *start = *p;
1960         int err;
1961         u8 struct_v;
1962
1963         dout("%s %p to %p len %d\n", __func__, *p, end, (int)(end - *p));
1964
1965         err = get_osdmap_client_data_v(p, end, "inc", &struct_v);
1966         if (err)
1967                 goto bad;
1968
1969         /* fsid, epoch, modified, new_pool_max, new_flags */
1970         ceph_decode_need(p, end, sizeof(fsid) + sizeof(u32) + sizeof(modified) +
1971                          sizeof(u64) + sizeof(u32), e_inval);
1972         ceph_decode_copy(p, &fsid, sizeof(fsid));
1973         epoch = ceph_decode_32(p);
1974         BUG_ON(epoch != map->epoch+1);
1975         ceph_decode_copy(p, &modified, sizeof(modified));
1976         new_pool_max = ceph_decode_64(p);
1977         new_flags = ceph_decode_32(p);
1978
1979         /* full map? */
1980         ceph_decode_32_safe(p, end, len, e_inval);
1981         if (len > 0) {
1982                 dout("apply_incremental full map len %d, %p to %p\n",
1983                      len, *p, end);
1984                 return ceph_osdmap_decode(p, min(*p+len, end), msgr2);
1985         }
1986
1987         /* new crush? */
1988         ceph_decode_32_safe(p, end, len, e_inval);
1989         if (len > 0) {
1990                 err = osdmap_set_crush(map,
1991                                        crush_decode(*p, min(*p + len, end)));
1992                 if (err)
1993                         goto bad;
1994                 *p += len;
1995         }
1996
1997         /* new flags? */
1998         if (new_flags >= 0)
1999                 map->flags = new_flags;
2000         if (new_pool_max >= 0)
2001                 map->pool_max = new_pool_max;
2002
2003         /* new max? */
2004         ceph_decode_32_safe(p, end, max, e_inval);
2005         if (max >= 0) {
2006                 err = osdmap_set_max_osd(map, max);
2007                 if (err)
2008                         goto bad;
2009         }
2010
2011         map->epoch++;
2012         map->modified = modified;
2013
2014         /* new_pools */
2015         err = decode_new_pools(p, end, map);
2016         if (err)
2017                 goto bad;
2018
2019         /* new_pool_names */
2020         err = decode_pool_names(p, end, map);
2021         if (err)
2022                 goto bad;
2023
2024         /* old_pool */
2025         ceph_decode_32_safe(p, end, len, e_inval);
2026         while (len--) {
2027                 struct ceph_pg_pool_info *pi;
2028
2029                 ceph_decode_64_safe(p, end, pool, e_inval);
2030                 pi = lookup_pg_pool(&map->pg_pools, pool);
2031                 if (pi)
2032                         __remove_pg_pool(&map->pg_pools, pi);
2033         }
2034
2035         /* new_up_client, new_state, new_weight */
2036         err = decode_new_up_state_weight(p, end, struct_v, msgr2, map);
2037         if (err)
2038                 goto bad;
2039
2040         /* new_pg_temp */
2041         err = decode_new_pg_temp(p, end, map);
2042         if (err)
2043                 goto bad;
2044
2045         /* new_primary_temp */
2046         if (struct_v >= 1) {
2047                 err = decode_new_primary_temp(p, end, map);
2048                 if (err)
2049                         goto bad;
2050         }
2051
2052         /* new_primary_affinity */
2053         if (struct_v >= 2) {
2054                 err = decode_new_primary_affinity(p, end, map);
2055                 if (err)
2056                         goto bad;
2057         }
2058
2059         if (struct_v >= 3) {
2060                 /* new_erasure_code_profiles */
2061                 ceph_decode_skip_map_of_map(p, end, string, string, string,
2062                                             e_inval);
2063                 /* old_erasure_code_profiles */
2064                 ceph_decode_skip_set(p, end, string, e_inval);
2065         }
2066
2067         if (struct_v >= 4) {
2068                 err = decode_new_pg_upmap(p, end, map);
2069                 if (err)
2070                         goto bad;
2071
2072                 err = decode_old_pg_upmap(p, end, map);
2073                 if (err)
2074                         goto bad;
2075
2076                 err = decode_new_pg_upmap_items(p, end, map);
2077                 if (err)
2078                         goto bad;
2079
2080                 err = decode_old_pg_upmap_items(p, end, map);
2081                 if (err)
2082                         goto bad;
2083         }
2084
2085         /* ignore the rest */
2086         *p = end;
2087
2088         dout("inc osdmap epoch %d max_osd %d\n", map->epoch, map->max_osd);
2089         return map;
2090
2091 e_inval:
2092         err = -EINVAL;
2093 bad:
2094         pr_err("corrupt inc osdmap (%d) epoch %d off %d (%p of %p-%p)\n",
2095                err, epoch, (int)(*p - start), *p, start, end);
2096         print_hex_dump(KERN_DEBUG, "osdmap: ",
2097                        DUMP_PREFIX_OFFSET, 16, 1,
2098                        start, end - start, true);
2099         return ERR_PTR(err);
2100 }
2101
2102 void ceph_oloc_copy(struct ceph_object_locator *dest,
2103                     const struct ceph_object_locator *src)
2104 {
2105         ceph_oloc_destroy(dest);
2106
2107         dest->pool = src->pool;
2108         if (src->pool_ns)
2109                 dest->pool_ns = ceph_get_string(src->pool_ns);
2110         else
2111                 dest->pool_ns = NULL;
2112 }
2113 EXPORT_SYMBOL(ceph_oloc_copy);
2114
2115 void ceph_oloc_destroy(struct ceph_object_locator *oloc)
2116 {
2117         ceph_put_string(oloc->pool_ns);
2118 }
2119 EXPORT_SYMBOL(ceph_oloc_destroy);
2120
2121 void ceph_oid_copy(struct ceph_object_id *dest,
2122                    const struct ceph_object_id *src)
2123 {
2124         ceph_oid_destroy(dest);
2125
2126         if (src->name != src->inline_name) {
2127                 /* very rare, see ceph_object_id definition */
2128                 dest->name = kmalloc(src->name_len + 1,
2129                                      GFP_NOIO | __GFP_NOFAIL);
2130         } else {
2131                 dest->name = dest->inline_name;
2132         }
2133         memcpy(dest->name, src->name, src->name_len + 1);
2134         dest->name_len = src->name_len;
2135 }
2136 EXPORT_SYMBOL(ceph_oid_copy);
2137
2138 static __printf(2, 0)
2139 int oid_printf_vargs(struct ceph_object_id *oid, const char *fmt, va_list ap)
2140 {
2141         int len;
2142
2143         WARN_ON(!ceph_oid_empty(oid));
2144
2145         len = vsnprintf(oid->inline_name, sizeof(oid->inline_name), fmt, ap);
2146         if (len >= sizeof(oid->inline_name))
2147                 return len;
2148
2149         oid->name_len = len;
2150         return 0;
2151 }
2152
2153 /*
2154  * If oid doesn't fit into inline buffer, BUG.
2155  */
2156 void ceph_oid_printf(struct ceph_object_id *oid, const char *fmt, ...)
2157 {
2158         va_list ap;
2159
2160         va_start(ap, fmt);
2161         BUG_ON(oid_printf_vargs(oid, fmt, ap));
2162         va_end(ap);
2163 }
2164 EXPORT_SYMBOL(ceph_oid_printf);
2165
2166 static __printf(3, 0)
2167 int oid_aprintf_vargs(struct ceph_object_id *oid, gfp_t gfp,
2168                       const char *fmt, va_list ap)
2169 {
2170         va_list aq;
2171         int len;
2172
2173         va_copy(aq, ap);
2174         len = oid_printf_vargs(oid, fmt, aq);
2175         va_end(aq);
2176
2177         if (len) {
2178                 char *external_name;
2179
2180                 external_name = kmalloc(len + 1, gfp);
2181                 if (!external_name)
2182                         return -ENOMEM;
2183
2184                 oid->name = external_name;
2185                 WARN_ON(vsnprintf(oid->name, len + 1, fmt, ap) != len);
2186                 oid->name_len = len;
2187         }
2188
2189         return 0;
2190 }
2191
2192 /*
2193  * If oid doesn't fit into inline buffer, allocate.
2194  */
2195 int ceph_oid_aprintf(struct ceph_object_id *oid, gfp_t gfp,
2196                      const char *fmt, ...)
2197 {
2198         va_list ap;
2199         int ret;
2200
2201         va_start(ap, fmt);
2202         ret = oid_aprintf_vargs(oid, gfp, fmt, ap);
2203         va_end(ap);
2204
2205         return ret;
2206 }
2207 EXPORT_SYMBOL(ceph_oid_aprintf);
2208
2209 void ceph_oid_destroy(struct ceph_object_id *oid)
2210 {
2211         if (oid->name != oid->inline_name)
2212                 kfree(oid->name);
2213 }
2214 EXPORT_SYMBOL(ceph_oid_destroy);
2215
2216 /*
2217  * osds only
2218  */
2219 static bool __osds_equal(const struct ceph_osds *lhs,
2220                          const struct ceph_osds *rhs)
2221 {
2222         if (lhs->size == rhs->size &&
2223             !memcmp(lhs->osds, rhs->osds, rhs->size * sizeof(rhs->osds[0])))
2224                 return true;
2225
2226         return false;
2227 }
2228
2229 /*
2230  * osds + primary
2231  */
2232 static bool osds_equal(const struct ceph_osds *lhs,
2233                        const struct ceph_osds *rhs)
2234 {
2235         if (__osds_equal(lhs, rhs) &&
2236             lhs->primary == rhs->primary)
2237                 return true;
2238
2239         return false;
2240 }
2241
2242 static bool osds_valid(const struct ceph_osds *set)
2243 {
2244         /* non-empty set */
2245         if (set->size > 0 && set->primary >= 0)
2246                 return true;
2247
2248         /* empty can_shift_osds set */
2249         if (!set->size && set->primary == -1)
2250                 return true;
2251
2252         /* empty !can_shift_osds set - all NONE */
2253         if (set->size > 0 && set->primary == -1) {
2254                 int i;
2255
2256                 for (i = 0; i < set->size; i++) {
2257                         if (set->osds[i] != CRUSH_ITEM_NONE)
2258                                 break;
2259                 }
2260                 if (i == set->size)
2261                         return true;
2262         }
2263
2264         return false;
2265 }
2266
2267 void ceph_osds_copy(struct ceph_osds *dest, const struct ceph_osds *src)
2268 {
2269         memcpy(dest->osds, src->osds, src->size * sizeof(src->osds[0]));
2270         dest->size = src->size;
2271         dest->primary = src->primary;
2272 }
2273
2274 bool ceph_pg_is_split(const struct ceph_pg *pgid, u32 old_pg_num,
2275                       u32 new_pg_num)
2276 {
2277         int old_bits = calc_bits_of(old_pg_num);
2278         int old_mask = (1 << old_bits) - 1;
2279         int n;
2280
2281         WARN_ON(pgid->seed >= old_pg_num);
2282         if (new_pg_num <= old_pg_num)
2283                 return false;
2284
2285         for (n = 1; ; n++) {
2286                 int next_bit = n << (old_bits - 1);
2287                 u32 s = next_bit | pgid->seed;
2288
2289                 if (s < old_pg_num || s == pgid->seed)
2290                         continue;
2291                 if (s >= new_pg_num)
2292                         break;
2293
2294                 s = ceph_stable_mod(s, old_pg_num, old_mask);
2295                 if (s == pgid->seed)
2296                         return true;
2297         }
2298
2299         return false;
2300 }
2301
2302 bool ceph_is_new_interval(const struct ceph_osds *old_acting,
2303                           const struct ceph_osds *new_acting,
2304                           const struct ceph_osds *old_up,
2305                           const struct ceph_osds *new_up,
2306                           int old_size,
2307                           int new_size,
2308                           int old_min_size,
2309                           int new_min_size,
2310                           u32 old_pg_num,
2311                           u32 new_pg_num,
2312                           bool old_sort_bitwise,
2313                           bool new_sort_bitwise,
2314                           bool old_recovery_deletes,
2315                           bool new_recovery_deletes,
2316                           const struct ceph_pg *pgid)
2317 {
2318         return !osds_equal(old_acting, new_acting) ||
2319                !osds_equal(old_up, new_up) ||
2320                old_size != new_size ||
2321                old_min_size != new_min_size ||
2322                ceph_pg_is_split(pgid, old_pg_num, new_pg_num) ||
2323                old_sort_bitwise != new_sort_bitwise ||
2324                old_recovery_deletes != new_recovery_deletes;
2325 }
2326
2327 static int calc_pg_rank(int osd, const struct ceph_osds *acting)
2328 {
2329         int i;
2330
2331         for (i = 0; i < acting->size; i++) {
2332                 if (acting->osds[i] == osd)
2333                         return i;
2334         }
2335
2336         return -1;
2337 }
2338
2339 static bool primary_changed(const struct ceph_osds *old_acting,
2340                             const struct ceph_osds *new_acting)
2341 {
2342         if (!old_acting->size && !new_acting->size)
2343                 return false; /* both still empty */
2344
2345         if (!old_acting->size ^ !new_acting->size)
2346                 return true; /* was empty, now not, or vice versa */
2347
2348         if (old_acting->primary != new_acting->primary)
2349                 return true; /* primary changed */
2350
2351         if (calc_pg_rank(old_acting->primary, old_acting) !=
2352             calc_pg_rank(new_acting->primary, new_acting))
2353                 return true;
2354
2355         return false; /* same primary (tho replicas may have changed) */
2356 }
2357
2358 bool ceph_osds_changed(const struct ceph_osds *old_acting,
2359                        const struct ceph_osds *new_acting,
2360                        bool any_change)
2361 {
2362         if (primary_changed(old_acting, new_acting))
2363                 return true;
2364
2365         if (any_change && !__osds_equal(old_acting, new_acting))
2366                 return true;
2367
2368         return false;
2369 }
2370
2371 /*
2372  * Map an object into a PG.
2373  *
2374  * Should only be called with target_oid and target_oloc (as opposed to
2375  * base_oid and base_oloc), since tiering isn't taken into account.
2376  */
2377 void __ceph_object_locator_to_pg(struct ceph_pg_pool_info *pi,
2378                                  const struct ceph_object_id *oid,
2379                                  const struct ceph_object_locator *oloc,
2380                                  struct ceph_pg *raw_pgid)
2381 {
2382         WARN_ON(pi->id != oloc->pool);
2383
2384         if (!oloc->pool_ns) {
2385                 raw_pgid->pool = oloc->pool;
2386                 raw_pgid->seed = ceph_str_hash(pi->object_hash, oid->name,
2387                                              oid->name_len);
2388                 dout("%s %s -> raw_pgid %llu.%x\n", __func__, oid->name,
2389                      raw_pgid->pool, raw_pgid->seed);
2390         } else {
2391                 char stack_buf[256];
2392                 char *buf = stack_buf;
2393                 int nsl = oloc->pool_ns->len;
2394                 size_t total = nsl + 1 + oid->name_len;
2395
2396                 if (total > sizeof(stack_buf))
2397                         buf = kmalloc(total, GFP_NOIO | __GFP_NOFAIL);
2398                 memcpy(buf, oloc->pool_ns->str, nsl);
2399                 buf[nsl] = '\037';
2400                 memcpy(buf + nsl + 1, oid->name, oid->name_len);
2401                 raw_pgid->pool = oloc->pool;
2402                 raw_pgid->seed = ceph_str_hash(pi->object_hash, buf, total);
2403                 if (buf != stack_buf)
2404                         kfree(buf);
2405                 dout("%s %s ns %.*s -> raw_pgid %llu.%x\n", __func__,
2406                      oid->name, nsl, oloc->pool_ns->str,
2407                      raw_pgid->pool, raw_pgid->seed);
2408         }
2409 }
2410
2411 int ceph_object_locator_to_pg(struct ceph_osdmap *osdmap,
2412                               const struct ceph_object_id *oid,
2413                               const struct ceph_object_locator *oloc,
2414                               struct ceph_pg *raw_pgid)
2415 {
2416         struct ceph_pg_pool_info *pi;
2417
2418         pi = ceph_pg_pool_by_id(osdmap, oloc->pool);
2419         if (!pi)
2420                 return -ENOENT;
2421
2422         __ceph_object_locator_to_pg(pi, oid, oloc, raw_pgid);
2423         return 0;
2424 }
2425 EXPORT_SYMBOL(ceph_object_locator_to_pg);
2426
2427 /*
2428  * Map a raw PG (full precision ps) into an actual PG.
2429  */
2430 static void raw_pg_to_pg(struct ceph_pg_pool_info *pi,
2431                          const struct ceph_pg *raw_pgid,
2432                          struct ceph_pg *pgid)
2433 {
2434         pgid->pool = raw_pgid->pool;
2435         pgid->seed = ceph_stable_mod(raw_pgid->seed, pi->pg_num,
2436                                      pi->pg_num_mask);
2437 }
2438
2439 /*
2440  * Map a raw PG (full precision ps) into a placement ps (placement
2441  * seed).  Include pool id in that value so that different pools don't
2442  * use the same seeds.
2443  */
2444 static u32 raw_pg_to_pps(struct ceph_pg_pool_info *pi,
2445                          const struct ceph_pg *raw_pgid)
2446 {
2447         if (pi->flags & CEPH_POOL_FLAG_HASHPSPOOL) {
2448                 /* hash pool id and seed so that pool PGs do not overlap */
2449                 return crush_hash32_2(CRUSH_HASH_RJENKINS1,
2450                                       ceph_stable_mod(raw_pgid->seed,
2451                                                       pi->pgp_num,
2452                                                       pi->pgp_num_mask),
2453                                       raw_pgid->pool);
2454         } else {
2455                 /*
2456                  * legacy behavior: add ps and pool together.  this is
2457                  * not a great approach because the PGs from each pool
2458                  * will overlap on top of each other: 0.5 == 1.4 ==
2459                  * 2.3 == ...
2460                  */
2461                 return ceph_stable_mod(raw_pgid->seed, pi->pgp_num,
2462                                        pi->pgp_num_mask) +
2463                        (unsigned)raw_pgid->pool;
2464         }
2465 }
2466
2467 /*
2468  * Magic value used for a "default" fallback choose_args, used if the
2469  * crush_choose_arg_map passed to do_crush() does not exist.  If this
2470  * also doesn't exist, fall back to canonical weights.
2471  */
2472 #define CEPH_DEFAULT_CHOOSE_ARGS        -1
2473
2474 static int do_crush(struct ceph_osdmap *map, int ruleno, int x,
2475                     int *result, int result_max,
2476                     const __u32 *weight, int weight_max,
2477                     s64 choose_args_index)
2478 {
2479         struct crush_choose_arg_map *arg_map;
2480         struct crush_work *work;
2481         int r;
2482
2483         BUG_ON(result_max > CEPH_PG_MAX_SIZE);
2484
2485         arg_map = lookup_choose_arg_map(&map->crush->choose_args,
2486                                         choose_args_index);
2487         if (!arg_map)
2488                 arg_map = lookup_choose_arg_map(&map->crush->choose_args,
2489                                                 CEPH_DEFAULT_CHOOSE_ARGS);
2490
2491         work = get_workspace(&map->crush_wsm, map->crush);
2492         r = crush_do_rule(map->crush, ruleno, x, result, result_max,
2493                           weight, weight_max, work,
2494                           arg_map ? arg_map->args : NULL);
2495         put_workspace(&map->crush_wsm, work);
2496         return r;
2497 }
2498
2499 static void remove_nonexistent_osds(struct ceph_osdmap *osdmap,
2500                                     struct ceph_pg_pool_info *pi,
2501                                     struct ceph_osds *set)
2502 {
2503         int i;
2504
2505         if (ceph_can_shift_osds(pi)) {
2506                 int removed = 0;
2507
2508                 /* shift left */
2509                 for (i = 0; i < set->size; i++) {
2510                         if (!ceph_osd_exists(osdmap, set->osds[i])) {
2511                                 removed++;
2512                                 continue;
2513                         }
2514                         if (removed)
2515                                 set->osds[i - removed] = set->osds[i];
2516                 }
2517                 set->size -= removed;
2518         } else {
2519                 /* set dne devices to NONE */
2520                 for (i = 0; i < set->size; i++) {
2521                         if (!ceph_osd_exists(osdmap, set->osds[i]))
2522                                 set->osds[i] = CRUSH_ITEM_NONE;
2523                 }
2524         }
2525 }
2526
2527 /*
2528  * Calculate raw set (CRUSH output) for given PG and filter out
2529  * nonexistent OSDs.  ->primary is undefined for a raw set.
2530  *
2531  * Placement seed (CRUSH input) is returned through @ppps.
2532  */
2533 static void pg_to_raw_osds(struct ceph_osdmap *osdmap,
2534                            struct ceph_pg_pool_info *pi,
2535                            const struct ceph_pg *raw_pgid,
2536                            struct ceph_osds *raw,
2537                            u32 *ppps)
2538 {
2539         u32 pps = raw_pg_to_pps(pi, raw_pgid);
2540         int ruleno;
2541         int len;
2542
2543         ceph_osds_init(raw);
2544         if (ppps)
2545                 *ppps = pps;
2546
2547         ruleno = crush_find_rule(osdmap->crush, pi->crush_ruleset, pi->type,
2548                                  pi->size);
2549         if (ruleno < 0) {
2550                 pr_err("no crush rule: pool %lld ruleset %d type %d size %d\n",
2551                        pi->id, pi->crush_ruleset, pi->type, pi->size);
2552                 return;
2553         }
2554
2555         if (pi->size > ARRAY_SIZE(raw->osds)) {
2556                 pr_err_ratelimited("pool %lld ruleset %d type %d too wide: size %d > %zu\n",
2557                        pi->id, pi->crush_ruleset, pi->type, pi->size,
2558                        ARRAY_SIZE(raw->osds));
2559                 return;
2560         }
2561
2562         len = do_crush(osdmap, ruleno, pps, raw->osds, pi->size,
2563                        osdmap->osd_weight, osdmap->max_osd, pi->id);
2564         if (len < 0) {
2565                 pr_err("error %d from crush rule %d: pool %lld ruleset %d type %d size %d\n",
2566                        len, ruleno, pi->id, pi->crush_ruleset, pi->type,
2567                        pi->size);
2568                 return;
2569         }
2570
2571         raw->size = len;
2572         remove_nonexistent_osds(osdmap, pi, raw);
2573 }
2574
2575 /* apply pg_upmap[_items] mappings */
2576 static void apply_upmap(struct ceph_osdmap *osdmap,
2577                         const struct ceph_pg *pgid,
2578                         struct ceph_osds *raw)
2579 {
2580         struct ceph_pg_mapping *pg;
2581         int i, j;
2582
2583         pg = lookup_pg_mapping(&osdmap->pg_upmap, pgid);
2584         if (pg) {
2585                 /* make sure targets aren't marked out */
2586                 for (i = 0; i < pg->pg_upmap.len; i++) {
2587                         int osd = pg->pg_upmap.osds[i];
2588
2589                         if (osd != CRUSH_ITEM_NONE &&
2590                             osd < osdmap->max_osd &&
2591                             osdmap->osd_weight[osd] == 0) {
2592                                 /* reject/ignore explicit mapping */
2593                                 return;
2594                         }
2595                 }
2596                 for (i = 0; i < pg->pg_upmap.len; i++)
2597                         raw->osds[i] = pg->pg_upmap.osds[i];
2598                 raw->size = pg->pg_upmap.len;
2599                 /* check and apply pg_upmap_items, if any */
2600         }
2601
2602         pg = lookup_pg_mapping(&osdmap->pg_upmap_items, pgid);
2603         if (pg) {
2604                 /*
2605                  * Note: this approach does not allow a bidirectional swap,
2606                  * e.g., [[1,2],[2,1]] applied to [0,1,2] -> [0,2,1].
2607                  */
2608                 for (i = 0; i < pg->pg_upmap_items.len; i++) {
2609                         int from = pg->pg_upmap_items.from_to[i][0];
2610                         int to = pg->pg_upmap_items.from_to[i][1];
2611                         int pos = -1;
2612                         bool exists = false;
2613
2614                         /* make sure replacement doesn't already appear */
2615                         for (j = 0; j < raw->size; j++) {
2616                                 int osd = raw->osds[j];
2617
2618                                 if (osd == to) {
2619                                         exists = true;
2620                                         break;
2621                                 }
2622                                 /* ignore mapping if target is marked out */
2623                                 if (osd == from && pos < 0 &&
2624                                     !(to != CRUSH_ITEM_NONE &&
2625                                       to < osdmap->max_osd &&
2626                                       osdmap->osd_weight[to] == 0)) {
2627                                         pos = j;
2628                                 }
2629                         }
2630                         if (!exists && pos >= 0)
2631                                 raw->osds[pos] = to;
2632                 }
2633         }
2634 }
2635
2636 /*
2637  * Given raw set, calculate up set and up primary.  By definition of an
2638  * up set, the result won't contain nonexistent or down OSDs.
2639  *
2640  * This is done in-place - on return @set is the up set.  If it's
2641  * empty, ->primary will remain undefined.
2642  */
2643 static void raw_to_up_osds(struct ceph_osdmap *osdmap,
2644                            struct ceph_pg_pool_info *pi,
2645                            struct ceph_osds *set)
2646 {
2647         int i;
2648
2649         /* ->primary is undefined for a raw set */
2650         BUG_ON(set->primary != -1);
2651
2652         if (ceph_can_shift_osds(pi)) {
2653                 int removed = 0;
2654
2655                 /* shift left */
2656                 for (i = 0; i < set->size; i++) {
2657                         if (ceph_osd_is_down(osdmap, set->osds[i])) {
2658                                 removed++;
2659                                 continue;
2660                         }
2661                         if (removed)
2662                                 set->osds[i - removed] = set->osds[i];
2663                 }
2664                 set->size -= removed;
2665                 if (set->size > 0)
2666                         set->primary = set->osds[0];
2667         } else {
2668                 /* set down/dne devices to NONE */
2669                 for (i = set->size - 1; i >= 0; i--) {
2670                         if (ceph_osd_is_down(osdmap, set->osds[i]))
2671                                 set->osds[i] = CRUSH_ITEM_NONE;
2672                         else
2673                                 set->primary = set->osds[i];
2674                 }
2675         }
2676 }
2677
2678 static void apply_primary_affinity(struct ceph_osdmap *osdmap,
2679                                    struct ceph_pg_pool_info *pi,
2680                                    u32 pps,
2681                                    struct ceph_osds *up)
2682 {
2683         int i;
2684         int pos = -1;
2685
2686         /*
2687          * Do we have any non-default primary_affinity values for these
2688          * osds?
2689          */
2690         if (!osdmap->osd_primary_affinity)
2691                 return;
2692
2693         for (i = 0; i < up->size; i++) {
2694                 int osd = up->osds[i];
2695
2696                 if (osd != CRUSH_ITEM_NONE &&
2697                     osdmap->osd_primary_affinity[osd] !=
2698                                         CEPH_OSD_DEFAULT_PRIMARY_AFFINITY) {
2699                         break;
2700                 }
2701         }
2702         if (i == up->size)
2703                 return;
2704
2705         /*
2706          * Pick the primary.  Feed both the seed (for the pg) and the
2707          * osd into the hash/rng so that a proportional fraction of an
2708          * osd's pgs get rejected as primary.
2709          */
2710         for (i = 0; i < up->size; i++) {
2711                 int osd = up->osds[i];
2712                 u32 aff;
2713
2714                 if (osd == CRUSH_ITEM_NONE)
2715                         continue;
2716
2717                 aff = osdmap->osd_primary_affinity[osd];
2718                 if (aff < CEPH_OSD_MAX_PRIMARY_AFFINITY &&
2719                     (crush_hash32_2(CRUSH_HASH_RJENKINS1,
2720                                     pps, osd) >> 16) >= aff) {
2721                         /*
2722                          * We chose not to use this primary.  Note it
2723                          * anyway as a fallback in case we don't pick
2724                          * anyone else, but keep looking.
2725                          */
2726                         if (pos < 0)
2727                                 pos = i;
2728                 } else {
2729                         pos = i;
2730                         break;
2731                 }
2732         }
2733         if (pos < 0)
2734                 return;
2735
2736         up->primary = up->osds[pos];
2737
2738         if (ceph_can_shift_osds(pi) && pos > 0) {
2739                 /* move the new primary to the front */
2740                 for (i = pos; i > 0; i--)
2741                         up->osds[i] = up->osds[i - 1];
2742                 up->osds[0] = up->primary;
2743         }
2744 }
2745
2746 /*
2747  * Get pg_temp and primary_temp mappings for given PG.
2748  *
2749  * Note that a PG may have none, only pg_temp, only primary_temp or
2750  * both pg_temp and primary_temp mappings.  This means @temp isn't
2751  * always a valid OSD set on return: in the "only primary_temp" case,
2752  * @temp will have its ->primary >= 0 but ->size == 0.
2753  */
2754 static void get_temp_osds(struct ceph_osdmap *osdmap,
2755                           struct ceph_pg_pool_info *pi,
2756                           const struct ceph_pg *pgid,
2757                           struct ceph_osds *temp)
2758 {
2759         struct ceph_pg_mapping *pg;
2760         int i;
2761
2762         ceph_osds_init(temp);
2763
2764         /* pg_temp? */
2765         pg = lookup_pg_mapping(&osdmap->pg_temp, pgid);
2766         if (pg) {
2767                 for (i = 0; i < pg->pg_temp.len; i++) {
2768                         if (ceph_osd_is_down(osdmap, pg->pg_temp.osds[i])) {
2769                                 if (ceph_can_shift_osds(pi))
2770                                         continue;
2771
2772                                 temp->osds[temp->size++] = CRUSH_ITEM_NONE;
2773                         } else {
2774                                 temp->osds[temp->size++] = pg->pg_temp.osds[i];
2775                         }
2776                 }
2777
2778                 /* apply pg_temp's primary */
2779                 for (i = 0; i < temp->size; i++) {
2780                         if (temp->osds[i] != CRUSH_ITEM_NONE) {
2781                                 temp->primary = temp->osds[i];
2782                                 break;
2783                         }
2784                 }
2785         }
2786
2787         /* primary_temp? */
2788         pg = lookup_pg_mapping(&osdmap->primary_temp, pgid);
2789         if (pg)
2790                 temp->primary = pg->primary_temp.osd;
2791 }
2792
2793 /*
2794  * Map a PG to its acting set as well as its up set.
2795  *
2796  * Acting set is used for data mapping purposes, while up set can be
2797  * recorded for detecting interval changes and deciding whether to
2798  * resend a request.
2799  */
2800 void ceph_pg_to_up_acting_osds(struct ceph_osdmap *osdmap,
2801                                struct ceph_pg_pool_info *pi,
2802                                const struct ceph_pg *raw_pgid,
2803                                struct ceph_osds *up,
2804                                struct ceph_osds *acting)
2805 {
2806         struct ceph_pg pgid;
2807         u32 pps;
2808
2809         WARN_ON(pi->id != raw_pgid->pool);
2810         raw_pg_to_pg(pi, raw_pgid, &pgid);
2811
2812         pg_to_raw_osds(osdmap, pi, raw_pgid, up, &pps);
2813         apply_upmap(osdmap, &pgid, up);
2814         raw_to_up_osds(osdmap, pi, up);
2815         apply_primary_affinity(osdmap, pi, pps, up);
2816         get_temp_osds(osdmap, pi, &pgid, acting);
2817         if (!acting->size) {
2818                 memcpy(acting->osds, up->osds, up->size * sizeof(up->osds[0]));
2819                 acting->size = up->size;
2820                 if (acting->primary == -1)
2821                         acting->primary = up->primary;
2822         }
2823         WARN_ON(!osds_valid(up) || !osds_valid(acting));
2824 }
2825
2826 bool ceph_pg_to_primary_shard(struct ceph_osdmap *osdmap,
2827                               struct ceph_pg_pool_info *pi,
2828                               const struct ceph_pg *raw_pgid,
2829                               struct ceph_spg *spgid)
2830 {
2831         struct ceph_pg pgid;
2832         struct ceph_osds up, acting;
2833         int i;
2834
2835         WARN_ON(pi->id != raw_pgid->pool);
2836         raw_pg_to_pg(pi, raw_pgid, &pgid);
2837
2838         if (ceph_can_shift_osds(pi)) {
2839                 spgid->pgid = pgid; /* struct */
2840                 spgid->shard = CEPH_SPG_NOSHARD;
2841                 return true;
2842         }
2843
2844         ceph_pg_to_up_acting_osds(osdmap, pi, &pgid, &up, &acting);
2845         for (i = 0; i < acting.size; i++) {
2846                 if (acting.osds[i] == acting.primary) {
2847                         spgid->pgid = pgid; /* struct */
2848                         spgid->shard = i;
2849                         return true;
2850                 }
2851         }
2852
2853         return false;
2854 }
2855
2856 /*
2857  * Return acting primary for given PG, or -1 if none.
2858  */
2859 int ceph_pg_to_acting_primary(struct ceph_osdmap *osdmap,
2860                               const struct ceph_pg *raw_pgid)
2861 {
2862         struct ceph_pg_pool_info *pi;
2863         struct ceph_osds up, acting;
2864
2865         pi = ceph_pg_pool_by_id(osdmap, raw_pgid->pool);
2866         if (!pi)
2867                 return -1;
2868
2869         ceph_pg_to_up_acting_osds(osdmap, pi, raw_pgid, &up, &acting);
2870         return acting.primary;
2871 }
2872 EXPORT_SYMBOL(ceph_pg_to_acting_primary);
2873
2874 static struct crush_loc_node *alloc_crush_loc(size_t type_name_len,
2875                                               size_t name_len)
2876 {
2877         struct crush_loc_node *loc;
2878
2879         loc = kmalloc(sizeof(*loc) + type_name_len + name_len + 2, GFP_NOIO);
2880         if (!loc)
2881                 return NULL;
2882
2883         RB_CLEAR_NODE(&loc->cl_node);
2884         return loc;
2885 }
2886
2887 static void free_crush_loc(struct crush_loc_node *loc)
2888 {
2889         WARN_ON(!RB_EMPTY_NODE(&loc->cl_node));
2890
2891         kfree(loc);
2892 }
2893
2894 static int crush_loc_compare(const struct crush_loc *loc1,
2895                              const struct crush_loc *loc2)
2896 {
2897         return strcmp(loc1->cl_type_name, loc2->cl_type_name) ?:
2898                strcmp(loc1->cl_name, loc2->cl_name);
2899 }
2900
2901 DEFINE_RB_FUNCS2(crush_loc, struct crush_loc_node, cl_loc, crush_loc_compare,
2902                  RB_BYPTR, const struct crush_loc *, cl_node)
2903
2904 /*
2905  * Parses a set of <bucket type name>':'<bucket name> pairs separated
2906  * by '|', e.g. "rack:foo1|rack:foo2|datacenter:bar".
2907  *
2908  * Note that @crush_location is modified by strsep().
2909  */
2910 int ceph_parse_crush_location(char *crush_location, struct rb_root *locs)
2911 {
2912         struct crush_loc_node *loc;
2913         const char *type_name, *name, *colon;
2914         size_t type_name_len, name_len;
2915
2916         dout("%s '%s'\n", __func__, crush_location);
2917         while ((type_name = strsep(&crush_location, "|"))) {
2918                 colon = strchr(type_name, ':');
2919                 if (!colon)
2920                         return -EINVAL;
2921
2922                 type_name_len = colon - type_name;
2923                 if (type_name_len == 0)
2924                         return -EINVAL;
2925
2926                 name = colon + 1;
2927                 name_len = strlen(name);
2928                 if (name_len == 0)
2929                         return -EINVAL;
2930
2931                 loc = alloc_crush_loc(type_name_len, name_len);
2932                 if (!loc)
2933                         return -ENOMEM;
2934
2935                 loc->cl_loc.cl_type_name = loc->cl_data;
2936                 memcpy(loc->cl_loc.cl_type_name, type_name, type_name_len);
2937                 loc->cl_loc.cl_type_name[type_name_len] = '\0';
2938
2939                 loc->cl_loc.cl_name = loc->cl_data + type_name_len + 1;
2940                 memcpy(loc->cl_loc.cl_name, name, name_len);
2941                 loc->cl_loc.cl_name[name_len] = '\0';
2942
2943                 if (!__insert_crush_loc(locs, loc)) {
2944                         free_crush_loc(loc);
2945                         return -EEXIST;
2946                 }
2947
2948                 dout("%s type_name '%s' name '%s'\n", __func__,
2949                      loc->cl_loc.cl_type_name, loc->cl_loc.cl_name);
2950         }
2951
2952         return 0;
2953 }
2954
2955 int ceph_compare_crush_locs(struct rb_root *locs1, struct rb_root *locs2)
2956 {
2957         struct rb_node *n1 = rb_first(locs1);
2958         struct rb_node *n2 = rb_first(locs2);
2959         int ret;
2960
2961         for ( ; n1 && n2; n1 = rb_next(n1), n2 = rb_next(n2)) {
2962                 struct crush_loc_node *loc1 =
2963                     rb_entry(n1, struct crush_loc_node, cl_node);
2964                 struct crush_loc_node *loc2 =
2965                     rb_entry(n2, struct crush_loc_node, cl_node);
2966
2967                 ret = crush_loc_compare(&loc1->cl_loc, &loc2->cl_loc);
2968                 if (ret)
2969                         return ret;
2970         }
2971
2972         if (!n1 && n2)
2973                 return -1;
2974         if (n1 && !n2)
2975                 return 1;
2976         return 0;
2977 }
2978
2979 void ceph_clear_crush_locs(struct rb_root *locs)
2980 {
2981         while (!RB_EMPTY_ROOT(locs)) {
2982                 struct crush_loc_node *loc =
2983                     rb_entry(rb_first(locs), struct crush_loc_node, cl_node);
2984
2985                 erase_crush_loc(locs, loc);
2986                 free_crush_loc(loc);
2987         }
2988 }
2989
2990 /*
2991  * [a-zA-Z0-9-_.]+
2992  */
2993 static bool is_valid_crush_name(const char *name)
2994 {
2995         do {
2996                 if (!('a' <= *name && *name <= 'z') &&
2997                     !('A' <= *name && *name <= 'Z') &&
2998                     !('0' <= *name && *name <= '9') &&
2999                     *name != '-' && *name != '_' && *name != '.')
3000                         return false;
3001         } while (*++name != '\0');
3002
3003         return true;
3004 }
3005
3006 /*
3007  * Gets the parent of an item.  Returns its id (<0 because the
3008  * parent is always a bucket), type id (>0 for the same reason,
3009  * via @parent_type_id) and location (via @parent_loc).  If no
3010  * parent, returns 0.
3011  *
3012  * Does a linear search, as there are no parent pointers of any
3013  * kind.  Note that the result is ambiguous for items that occur
3014  * multiple times in the map.
3015  */
3016 static int get_immediate_parent(struct crush_map *c, int id,
3017                                 u16 *parent_type_id,
3018                                 struct crush_loc *parent_loc)
3019 {
3020         struct crush_bucket *b;
3021         struct crush_name_node *type_cn, *cn;
3022         int i, j;
3023
3024         for (i = 0; i < c->max_buckets; i++) {
3025                 b = c->buckets[i];
3026                 if (!b)
3027                         continue;
3028
3029                 /* ignore per-class shadow hierarchy */
3030                 cn = lookup_crush_name(&c->names, b->id);
3031                 if (!cn || !is_valid_crush_name(cn->cn_name))
3032                         continue;
3033
3034                 for (j = 0; j < b->size; j++) {
3035                         if (b->items[j] != id)
3036                                 continue;
3037
3038                         *parent_type_id = b->type;
3039                         type_cn = lookup_crush_name(&c->type_names, b->type);
3040                         parent_loc->cl_type_name = type_cn->cn_name;
3041                         parent_loc->cl_name = cn->cn_name;
3042                         return b->id;
3043                 }
3044         }
3045
3046         return 0;  /* no parent */
3047 }
3048
3049 /*
3050  * Calculates the locality/distance from an item to a client
3051  * location expressed in terms of CRUSH hierarchy as a set of
3052  * (bucket type name, bucket name) pairs.  Specifically, looks
3053  * for the lowest-valued bucket type for which the location of
3054  * @id matches one of the locations in @locs, so for standard
3055  * bucket types (host = 1, rack = 3, datacenter = 8, zone = 9)
3056  * a matching host is closer than a matching rack and a matching
3057  * data center is closer than a matching zone.
3058  *
3059  * Specifying multiple locations (a "multipath" location) such
3060  * as "rack=foo1 rack=foo2 datacenter=bar" is allowed -- @locs
3061  * is a multimap.  The locality will be:
3062  *
3063  * - 3 for OSDs in racks foo1 and foo2
3064  * - 8 for OSDs in data center bar
3065  * - -1 for all other OSDs
3066  *
3067  * The lowest possible bucket type is 1, so the best locality
3068  * for an OSD is 1 (i.e. a matching host).  Locality 0 would be
3069  * the OSD itself.
3070  */
3071 int ceph_get_crush_locality(struct ceph_osdmap *osdmap, int id,
3072                             struct rb_root *locs)
3073 {
3074         struct crush_loc loc;
3075         u16 type_id;
3076
3077         /*
3078          * Instead of repeated get_immediate_parent() calls,
3079          * the location of @id could be obtained with a single
3080          * depth-first traversal.
3081          */
3082         for (;;) {
3083                 id = get_immediate_parent(osdmap->crush, id, &type_id, &loc);
3084                 if (id >= 0)
3085                         return -1;  /* not local */
3086
3087                 if (lookup_crush_loc(locs, &loc))
3088                         return type_id;
3089         }
3090 }