net/ceph/osdmap.c

   1
   2 #include <linux/ceph/ceph_debug.h>
   3
   4 #include <linux/module.h>
   5 #include <linux/slab.h>
   6 #include <asm/div64.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, int 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(le32_to_cpu(pi->v.pg_num)-1)) - 1;
  49         pi->pgp_num_mask =
  50                 (1 << calc_bits_of(le32_to_cpu(pi->v.pgp_num)-1)) - 1;
  51         pi->lpg_num_mask =
  52                 (1 << calc_bits_of(le32_to_cpu(pi->v.lpg_num)-1)) - 1;
  53         pi->lpgp_num_mask =
  54                 (1 << calc_bits_of(le32_to_cpu(pi->v.lpgp_num)-1)) - 1;
  55 }
  56
  57 /*
  58  * decode crush map
  59  */
  60 static int crush_decode_uniform_bucket(void **p, void *end,
  61                                        struct crush_bucket_uniform *b)
  62 {
  63         dout("crush_decode_uniform_bucket %p to %p\n", *p, end);
  64         ceph_decode_need(p, end, (1+b->h.size) * sizeof(u32), bad);
  65         b->item_weight = ceph_decode_32(p);
  66         return 0;
  67 bad:
  68         return -EINVAL;
  69 }
  70
  71 static int crush_decode_list_bucket(void **p, void *end,
  72                                     struct crush_bucket_list *b)
  73 {
  74         int j;
  75         dout("crush_decode_list_bucket %p to %p\n", *p, end);
  76         b->item_weights = kcalloc(b->h.size, sizeof(u32), GFP_NOFS);
  77         if (b->item_weights == NULL)
  78                 return -ENOMEM;
  79         b->sum_weights = kcalloc(b->h.size, sizeof(u32), GFP_NOFS);
  80         if (b->sum_weights == NULL)
  81                 return -ENOMEM;
  82         ceph_decode_need(p, end, 2 * b->h.size * sizeof(u32), bad);
  83         for (j = 0; j < b->h.size; j++) {
  84                 b->item_weights[j] = ceph_decode_32(p);
  85                 b->sum_weights[j] = ceph_decode_32(p);
  86         }
  87         return 0;
  88 bad:
  89         return -EINVAL;
  90 }
  91
  92 static int crush_decode_tree_bucket(void **p, void *end,
  93                                     struct crush_bucket_tree *b)
  94 {
  95         int j;
  96         dout("crush_decode_tree_bucket %p to %p\n", *p, end);
  97         ceph_decode_32_safe(p, end, b->num_nodes, bad);
  98         b->node_weights = kcalloc(b->num_nodes, sizeof(u32), GFP_NOFS);
  99         if (b->node_weights == NULL)
 100                 return -ENOMEM;
 101         ceph_decode_need(p, end, b->num_nodes * sizeof(u32), bad);
 102         for (j = 0; j < b->num_nodes; j++)
 103                 b->node_weights[j] = ceph_decode_32(p);
 104         return 0;
 105 bad:
 106         return -EINVAL;
 107 }
 108
 109 static int crush_decode_straw_bucket(void **p, void *end,
 110                                      struct crush_bucket_straw *b)
 111 {
 112         int j;
 113         dout("crush_decode_straw_bucket %p to %p\n", *p, end);
 114         b->item_weights = kcalloc(b->h.size, sizeof(u32), GFP_NOFS);
 115         if (b->item_weights == NULL)
 116                 return -ENOMEM;
 117         b->straws = kcalloc(b->h.size, sizeof(u32), GFP_NOFS);
 118         if (b->straws == NULL)
 119                 return -ENOMEM;
 120         ceph_decode_need(p, end, 2 * b->h.size * sizeof(u32), bad);
 121         for (j = 0; j < b->h.size; j++) {
 122                 b->item_weights[j] = ceph_decode_32(p);
 123                 b->straws[j] = ceph_decode_32(p);
 124         }
 125         return 0;
 126 bad:
 127         return -EINVAL;
 128 }
 129
 130 static int skip_name_map(void **p, void *end)
 131 {
 132         int len;
 133         ceph_decode_32_safe(p, end, len ,bad);
 134         while (len--) {
 135                 int strlen;
 136                 *p += sizeof(u32);
 137                 ceph_decode_32_safe(p, end, strlen, bad);
 138                 *p += strlen;
 139 }
 140         return 0;
 141 bad:
 142         return -EINVAL;
 143 }
 144
 145 static struct crush_map *crush_decode(void *pbyval, void *end)
 146 {
 147         struct crush_map *c;
 148         int err = -EINVAL;
 149         int i, j;
 150         void **p = &pbyval;
 151         void *start = pbyval;
 152         u32 magic;
 153         u32 num_name_maps;
 154
 155         dout("crush_decode %p to %p len %d\n", *p, end, (int)(end - *p));
 156
 157         c = kzalloc(sizeof(*c), GFP_NOFS);
 158         if (c == NULL)
 159                 return ERR_PTR(-ENOMEM);
 160
 161         /* set tunables to default values */
 162         c->choose_local_tries = 2;
 163         c->choose_local_fallback_tries = 5;
 164         c->choose_total_tries = 19;
 165         c->chooseleaf_descend_once = 0;
 166
 167         ceph_decode_need(p, end, 4*sizeof(u32), bad);
 168         magic = ceph_decode_32(p);
 169         if (magic != CRUSH_MAGIC) {
 170                 pr_err("crush_decode magic %x != current %x\n",
 171                        (unsigned int)magic, (unsigned int)CRUSH_MAGIC);
 172                 goto bad;
 173         }
 174         c->max_buckets = ceph_decode_32(p);
 175         c->max_rules = ceph_decode_32(p);
 176         c->max_devices = ceph_decode_32(p);
 177
 178         c->buckets = kcalloc(c->max_buckets, sizeof(*c->buckets), GFP_NOFS);
 179         if (c->buckets == NULL)
 180                 goto badmem;
 181         c->rules = kcalloc(c->max_rules, sizeof(*c->rules), GFP_NOFS);
 182         if (c->rules == NULL)
 183                 goto badmem;
 184
 185         /* buckets */
 186         for (i = 0; i < c->max_buckets; i++) {
 187                 int size = 0;
 188                 u32 alg;
 189                 struct crush_bucket *b;
 190
 191                 ceph_decode_32_safe(p, end, alg, bad);
 192                 if (alg == 0) {
 193                         c->buckets[i] = NULL;
 194                         continue;
 195                 }
 196                 dout("crush_decode bucket %d off %x %p to %p\n",
 197                      i, (int)(*p-start), *p, end);
 198
 199                 switch (alg) {
 200                 case CRUSH_BUCKET_UNIFORM:
 201                         size = sizeof(struct crush_bucket_uniform);
 202                         break;
 203                 case CRUSH_BUCKET_LIST:
 204                         size = sizeof(struct crush_bucket_list);
 205                         break;
 206                 case CRUSH_BUCKET_TREE:
 207                         size = sizeof(struct crush_bucket_tree);
 208                         break;
 209                 case CRUSH_BUCKET_STRAW:
 210                         size = sizeof(struct crush_bucket_straw);
 211                         break;
 212                 default:
 213                         err = -EINVAL;
 214                         goto bad;
 215                 }
 216                 BUG_ON(size == 0);
 217                 b = c->buckets[i] = kzalloc(size, GFP_NOFS);
 218                 if (b == NULL)
 219                         goto badmem;
 220
 221                 ceph_decode_need(p, end, 4*sizeof(u32), bad);
 222                 b->id = ceph_decode_32(p);
 223                 b->type = ceph_decode_16(p);
 224                 b->alg = ceph_decode_8(p);
 225                 b->hash = ceph_decode_8(p);
 226                 b->weight = ceph_decode_32(p);
 227                 b->size = ceph_decode_32(p);
 228
 229                 dout("crush_decode bucket size %d off %x %p to %p\n",
 230                      b->size, (int)(*p-start), *p, end);
 231
 232                 b->items = kcalloc(b->size, sizeof(__s32), GFP_NOFS);
 233                 if (b->items == NULL)
 234                         goto badmem;
 235                 b->perm = kcalloc(b->size, sizeof(u32), GFP_NOFS);
 236                 if (b->perm == NULL)
 237                         goto badmem;
 238                 b->perm_n = 0;
 239
 240                 ceph_decode_need(p, end, b->size*sizeof(u32), bad);
 241                 for (j = 0; j < b->size; j++)
 242                         b->items[j] = ceph_decode_32(p);
 243
 244                 switch (b->alg) {
 245                 case CRUSH_BUCKET_UNIFORM:
 246                         err = crush_decode_uniform_bucket(p, end,
 247                                   (struct crush_bucket_uniform *)b);
 248                         if (err < 0)
 249                                 goto bad;
 250                         break;
 251                 case CRUSH_BUCKET_LIST:
 252                         err = crush_decode_list_bucket(p, end,
 253                                (struct crush_bucket_list *)b);
 254                         if (err < 0)
 255                                 goto bad;
 256                         break;
 257                 case CRUSH_BUCKET_TREE:
 258                         err = crush_decode_tree_bucket(p, end,
 259                                 (struct crush_bucket_tree *)b);
 260                         if (err < 0)
 261                                 goto bad;
 262                         break;
 263                 case CRUSH_BUCKET_STRAW:
 264                         err = crush_decode_straw_bucket(p, end,
 265                                 (struct crush_bucket_straw *)b);
 266                         if (err < 0)
 267                                 goto bad;
 268                         break;
 269                 }
 270         }
 271
 272         /* rules */
 273         dout("rule vec is %p\n", c->rules);
 274         for (i = 0; i < c->max_rules; i++) {
 275                 u32 yes;
 276                 struct crush_rule *r;
 277
 278                 ceph_decode_32_safe(p, end, yes, bad);
 279                 if (!yes) {
 280                         dout("crush_decode NO rule %d off %x %p to %p\n",
 281                              i, (int)(*p-start), *p, end);
 282                         c->rules[i] = NULL;
 283                         continue;
 284                 }
 285
 286                 dout("crush_decode rule %d off %x %p to %p\n",
 287                      i, (int)(*p-start), *p, end);
 288
 289                 /* len */
 290                 ceph_decode_32_safe(p, end, yes, bad);
 291 #if BITS_PER_LONG == 32
 292                 err = -EINVAL;
 293                 if (yes > (ULONG_MAX - sizeof(*r))
 294                           / sizeof(struct crush_rule_step))
 295                         goto bad;
 296 #endif
 297                 r = c->rules[i] = kmalloc(sizeof(*r) +
 298                                           yes*sizeof(struct crush_rule_step),
 299                                           GFP_NOFS);
 300                 if (r == NULL)
 301                         goto badmem;
 302                 dout(" rule %d is at %p\n", i, r);
 303                 r->len = yes;
 304                 ceph_decode_copy_safe(p, end, &r->mask, 4, bad); /* 4 u8's */
 305                 ceph_decode_need(p, end, r->len*3*sizeof(u32), bad);
 306                 for (j = 0; j < r->len; j++) {
 307                         r->steps[j].op = ceph_decode_32(p);
 308                         r->steps[j].arg1 = ceph_decode_32(p);
 309                         r->steps[j].arg2 = ceph_decode_32(p);
 310                 }
 311         }
 312
 313         /* ignore trailing name maps. */
 314         for (num_name_maps = 0; num_name_maps < 3; num_name_maps++) {
 315                 err = skip_name_map(p, end);
 316                 if (err < 0)
 317                         goto done;
 318         }
 319
 320         /* tunables */
 321         ceph_decode_need(p, end, 3*sizeof(u32), done);
 322         c->choose_local_tries = ceph_decode_32(p);
 323         c->choose_local_fallback_tries =  ceph_decode_32(p);
 324         c->choose_total_tries = ceph_decode_32(p);
 325         dout("crush decode tunable choose_local_tries = %d",
 326              c->choose_local_tries);
 327         dout("crush decode tunable choose_local_fallback_tries = %d",
 328              c->choose_local_fallback_tries);
 329         dout("crush decode tunable choose_total_tries = %d",
 330              c->choose_total_tries);
 331
 332         ceph_decode_need(p, end, sizeof(u32), done);
 333         c->chooseleaf_descend_once = ceph_decode_32(p);
 334         dout("crush decode tunable chooseleaf_descend_once = %d",
 335              c->chooseleaf_descend_once);
 336
 337 done:
 338         dout("crush_decode success\n");
 339         return c;
 340
 341 badmem:
 342         err = -ENOMEM;
 343 bad:
 344         dout("crush_decode fail %d\n", err);
 345         crush_destroy(c);
 346         return ERR_PTR(err);
 347 }
 348
 349 /*
 350  * rbtree of pg_mapping for handling pg_temp (explicit mapping of pgid
 351  * to a set of osds)
 352  */
 353 static int pgid_cmp(struct ceph_pg l, struct ceph_pg r)
 354 {
 355         u64 a = *(u64 *)&l;
 356         u64 b = *(u64 *)&r;
 357
 358         if (a < b)
 359                 return -1;
 360         if (a > b)
 361                 return 1;
 362         return 0;
 363 }
 364
 365 static int __insert_pg_mapping(struct ceph_pg_mapping *new,
 366                                struct rb_root *root)
 367 {
 368         struct rb_node **p = &root->rb_node;
 369         struct rb_node *parent = NULL;
 370         struct ceph_pg_mapping *pg = NULL;
 371         int c;
 372
 373         dout("__insert_pg_mapping %llx %p\n", *(u64 *)&new->pgid, new);
 374         while (*p) {
 375                 parent = *p;
 376                 pg = rb_entry(parent, struct ceph_pg_mapping, node);
 377                 c = pgid_cmp(new->pgid, pg->pgid);
 378                 if (c < 0)
 379                         p = &(*p)->rb_left;
 380                 else if (c > 0)
 381                         p = &(*p)->rb_right;
 382                 else
 383                         return -EEXIST;
 384         }
 385
 386         rb_link_node(&new->node, parent, p);
 387         rb_insert_color(&new->node, root);
 388         return 0;
 389 }
 390
 391 static struct ceph_pg_mapping *__lookup_pg_mapping(struct rb_root *root,
 392                                                    struct ceph_pg pgid)
 393 {
 394         struct rb_node *n = root->rb_node;
 395         struct ceph_pg_mapping *pg;
 396         int c;
 397
 398         while (n) {
 399                 pg = rb_entry(n, struct ceph_pg_mapping, node);
 400                 c = pgid_cmp(pgid, pg->pgid);
 401                 if (c < 0) {
 402                         n = n->rb_left;
 403                 } else if (c > 0) {
 404                         n = n->rb_right;
 405                 } else {
 406                         dout("__lookup_pg_mapping %llx got %p\n",
 407                              *(u64 *)&pgid, pg);
 408                         return pg;
 409                 }
 410         }
 411         return NULL;
 412 }
 413
 414 static int __remove_pg_mapping(struct rb_root *root, struct ceph_pg pgid)
 415 {
 416         struct ceph_pg_mapping *pg = __lookup_pg_mapping(root, pgid);
 417
 418         if (pg) {
 419                 dout("__remove_pg_mapping %llx %p\n", *(u64 *)&pgid, pg);
 420                 rb_erase(&pg->node, root);
 421                 kfree(pg);
 422                 return 0;
 423         }
 424         dout("__remove_pg_mapping %llx dne\n", *(u64 *)&pgid);
 425         return -ENOENT;
 426 }
 427
 428 /*
 429  * rbtree of pg pool info
 430  */
 431 static int __insert_pg_pool(struct rb_root *root, struct ceph_pg_pool_info *new)
 432 {
 433         struct rb_node **p = &root->rb_node;
 434         struct rb_node *parent = NULL;
 435         struct ceph_pg_pool_info *pi = NULL;
 436
 437         while (*p) {
 438                 parent = *p;
 439                 pi = rb_entry(parent, struct ceph_pg_pool_info, node);
 440                 if (new->id < pi->id)
 441                         p = &(*p)->rb_left;
 442                 else if (new->id > pi->id)
 443                         p = &(*p)->rb_right;
 444                 else
 445                         return -EEXIST;
 446         }
 447
 448         rb_link_node(&new->node, parent, p);
 449         rb_insert_color(&new->node, root);
 450         return 0;
 451 }
 452
 453 static struct ceph_pg_pool_info *__lookup_pg_pool(struct rb_root *root, int id)
 454 {
 455         struct ceph_pg_pool_info *pi;
 456         struct rb_node *n = root->rb_node;
 457
 458         while (n) {
 459                 pi = rb_entry(n, struct ceph_pg_pool_info, node);
 460                 if (id < pi->id)
 461                         n = n->rb_left;
 462                 else if (id > pi->id)
 463                         n = n->rb_right;
 464                 else
 465                         return pi;
 466         }
 467         return NULL;
 468 }
 469
 470 const char *ceph_pg_pool_name_by_id(struct ceph_osdmap *map, u64 id)
 471 {
 472         struct ceph_pg_pool_info *pi;
 473
 474         if (id == CEPH_NOPOOL)
 475                 return NULL;
 476
 477         if (WARN_ON_ONCE(id > (u64) INT_MAX))
 478                 return NULL;
 479
 480         pi = __lookup_pg_pool(&map->pg_pools, (int) id);
 481
 482         return pi ? pi->name : NULL;
 483 }
 484 EXPORT_SYMBOL(ceph_pg_pool_name_by_id);
 485
 486 int ceph_pg_poolid_by_name(struct ceph_osdmap *map, const char *name)
 487 {
 488         struct rb_node *rbp;
 489
 490         for (rbp = rb_first(&map->pg_pools); rbp; rbp = rb_next(rbp)) {
 491                 struct ceph_pg_pool_info *pi =
 492                         rb_entry(rbp, struct ceph_pg_pool_info, node);
 493                 if (pi->name && strcmp(pi->name, name) == 0)
 494                         return pi->id;
 495         }
 496         return -ENOENT;
 497 }
 498 EXPORT_SYMBOL(ceph_pg_poolid_by_name);
 499
 500 static void __remove_pg_pool(struct rb_root *root, struct ceph_pg_pool_info *pi)
 501 {
 502         rb_erase(&pi->node, root);
 503         kfree(pi->name);
 504         kfree(pi);
 505 }
 506
 507 static int __decode_pool(void **p, void *end, struct ceph_pg_pool_info *pi)
 508 {
 509         unsigned int n, m;
 510
 511         ceph_decode_copy(p, &pi->v, sizeof(pi->v));
 512         calc_pg_masks(pi);
 513
 514         /* num_snaps * snap_info_t */
 515         n = le32_to_cpu(pi->v.num_snaps);
 516         while (n--) {
 517                 ceph_decode_need(p, end, sizeof(u64) + 1 + sizeof(u64) +
 518                                  sizeof(struct ceph_timespec), bad);
 519                 *p += sizeof(u64) +       /* key */
 520                         1 + sizeof(u64) + /* u8, snapid */
 521                         sizeof(struct ceph_timespec);
 522                 m = ceph_decode_32(p);    /* snap name */
 523                 *p += m;
 524         }
 525
 526         *p += le32_to_cpu(pi->v.num_removed_snap_intervals) * sizeof(u64) * 2;
 527         return 0;
 528
 529 bad:
 530         return -EINVAL;
 531 }
 532
 533 static int __decode_pool_names(void **p, void *end, struct ceph_osdmap *map)
 534 {
 535         struct ceph_pg_pool_info *pi;
 536         u32 num, len, pool;
 537
 538         ceph_decode_32_safe(p, end, num, bad);
 539         dout(" %d pool names\n", num);
 540         while (num--) {
 541                 ceph_decode_32_safe(p, end, pool, bad);
 542                 ceph_decode_32_safe(p, end, len, bad);
 543                 dout("  pool %d len %d\n", pool, len);
 544                 ceph_decode_need(p, end, len, bad);
 545                 pi = __lookup_pg_pool(&map->pg_pools, pool);
 546                 if (pi) {
 547                         char *name = kstrndup(*p, len, GFP_NOFS);
 548
 549                         if (!name)
 550                                 return -ENOMEM;
 551                         kfree(pi->name);
 552                         pi->name = name;
 553                         dout("  name is %s\n", pi->name);
 554                 }
 555                 *p += len;
 556         }
 557         return 0;
 558
 559 bad:
 560         return -EINVAL;
 561 }
 562
 563 /*
 564  * osd map
 565  */
 566 void ceph_osdmap_destroy(struct ceph_osdmap *map)
 567 {
 568         dout("osdmap_destroy %p\n", map);
 569         if (map->crush)
 570                 crush_destroy(map->crush);
 571         while (!RB_EMPTY_ROOT(&map->pg_temp)) {
 572                 struct ceph_pg_mapping *pg =
 573                         rb_entry(rb_first(&map->pg_temp),
 574                                  struct ceph_pg_mapping, node);
 575                 rb_erase(&pg->node, &map->pg_temp);
 576                 kfree(pg);
 577         }
 578         while (!RB_EMPTY_ROOT(&map->pg_pools)) {
 579                 struct ceph_pg_pool_info *pi =
 580                         rb_entry(rb_first(&map->pg_pools),
 581                                  struct ceph_pg_pool_info, node);
 582                 __remove_pg_pool(&map->pg_pools, pi);
 583         }
 584         kfree(map->osd_state);
 585         kfree(map->osd_weight);
 586         kfree(map->osd_addr);
 587         kfree(map);
 588 }
 589
 590 /*
 591  * adjust max osd value.  reallocate arrays.
 592  */
 593 static int osdmap_set_max_osd(struct ceph_osdmap *map, int max)
 594 {
 595         u8 *state;
 596         struct ceph_entity_addr *addr;
 597         u32 *weight;
 598
 599         state = kcalloc(max, sizeof(*state), GFP_NOFS);
 600         addr = kcalloc(max, sizeof(*addr), GFP_NOFS);
 601         weight = kcalloc(max, sizeof(*weight), GFP_NOFS);
 602         if (state == NULL || addr == NULL || weight == NULL) {
 603                 kfree(state);
 604                 kfree(addr);
 605                 kfree(weight);
 606                 return -ENOMEM;
 607         }
 608
 609         /* copy old? */
 610         if (map->osd_state) {
 611                 memcpy(state, map->osd_state, map->max_osd*sizeof(*state));
 612                 memcpy(addr, map->osd_addr, map->max_osd*sizeof(*addr));
 613                 memcpy(weight, map->osd_weight, map->max_osd*sizeof(*weight));
 614                 kfree(map->osd_state);
 615                 kfree(map->osd_addr);
 616                 kfree(map->osd_weight);
 617         }
 618
 619         map->osd_state = state;
 620         map->osd_weight = weight;
 621         map->osd_addr = addr;
 622         map->max_osd = max;
 623         return 0;
 624 }
 625
 626 /*
 627  * decode a full map.
 628  */
 629 struct ceph_osdmap *osdmap_decode(void **p, void *end)
 630 {
 631         struct ceph_osdmap *map;
 632         u16 version;
 633         u32 len, max, i;
 634         u8 ev;
 635         int err = -EINVAL;
 636         void *start = *p;
 637         struct ceph_pg_pool_info *pi;
 638
 639         dout("osdmap_decode %p to %p len %d\n", *p, end, (int)(end - *p));
 640
 641         map = kzalloc(sizeof(*map), GFP_NOFS);
 642         if (map == NULL)
 643                 return ERR_PTR(-ENOMEM);
 644         map->pg_temp = RB_ROOT;
 645
 646         ceph_decode_16_safe(p, end, version, bad);
 647         if (version > CEPH_OSDMAP_VERSION) {
 648                 pr_warning("got unknown v %d > %d of osdmap\n", version,
 649                            CEPH_OSDMAP_VERSION);
 650                 goto bad;
 651         }
 652
 653         ceph_decode_need(p, end, 2*sizeof(u64)+6*sizeof(u32), bad);
 654         ceph_decode_copy(p, &map->fsid, sizeof(map->fsid));
 655         map->epoch = ceph_decode_32(p);
 656         ceph_decode_copy(p, &map->created, sizeof(map->created));
 657         ceph_decode_copy(p, &map->modified, sizeof(map->modified));
 658
 659         ceph_decode_32_safe(p, end, max, bad);
 660         while (max--) {
 661                 ceph_decode_need(p, end, 4 + 1 + sizeof(pi->v), bad);
 662                 err = -ENOMEM;
 663                 pi = kzalloc(sizeof(*pi), GFP_NOFS);
 664                 if (!pi)
 665                         goto bad;
 666                 pi->id = ceph_decode_32(p);
 667                 err = -EINVAL;
 668                 ev = ceph_decode_8(p); /* encoding version */
 669                 if (ev > CEPH_PG_POOL_VERSION) {
 670                         pr_warning("got unknown v %d > %d of ceph_pg_pool\n",
 671                                    ev, CEPH_PG_POOL_VERSION);
 672                         kfree(pi);
 673                         goto bad;
 674                 }
 675                 err = __decode_pool(p, end, pi);
 676                 if (err < 0) {
 677                         kfree(pi);
 678                         goto bad;
 679                 }
 680                 __insert_pg_pool(&map->pg_pools, pi);
 681         }
 682
 683         if (version >= 5) {
 684                 err = __decode_pool_names(p, end, map);
 685                 if (err < 0) {
 686                         dout("fail to decode pool names");
 687                         goto bad;
 688                 }
 689         }
 690
 691         ceph_decode_32_safe(p, end, map->pool_max, bad);
 692
 693         ceph_decode_32_safe(p, end, map->flags, bad);
 694
 695         max = ceph_decode_32(p);
 696
 697         /* (re)alloc osd arrays */
 698         err = osdmap_set_max_osd(map, max);
 699         if (err < 0)
 700                 goto bad;
 701         dout("osdmap_decode max_osd = %d\n", map->max_osd);
 702
 703         /* osds */
 704         err = -EINVAL;
 705         ceph_decode_need(p, end, 3*sizeof(u32) +
 706                          map->max_osd*(1 + sizeof(*map->osd_weight) +
 707                                        sizeof(*map->osd_addr)), bad);
 708         *p += 4; /* skip length field (should match max) */
 709         ceph_decode_copy(p, map->osd_state, map->max_osd);
 710
 711         *p += 4; /* skip length field (should match max) */
 712         for (i = 0; i < map->max_osd; i++)
 713                 map->osd_weight[i] = ceph_decode_32(p);
 714
 715         *p += 4; /* skip length field (should match max) */
 716         ceph_decode_copy(p, map->osd_addr, map->max_osd*sizeof(*map->osd_addr));
 717         for (i = 0; i < map->max_osd; i++)
 718                 ceph_decode_addr(&map->osd_addr[i]);
 719
 720         /* pg_temp */
 721         ceph_decode_32_safe(p, end, len, bad);
 722         for (i = 0; i < len; i++) {
 723                 int n, j;
 724                 struct ceph_pg pgid;
 725                 struct ceph_pg_mapping *pg;
 726
 727                 ceph_decode_need(p, end, sizeof(u32) + sizeof(u64), bad);
 728                 ceph_decode_copy(p, &pgid, sizeof(pgid));
 729                 n = ceph_decode_32(p);
 730                 err = -EINVAL;
 731                 if (n > (UINT_MAX - sizeof(*pg)) / sizeof(u32))
 732                         goto bad;
 733                 ceph_decode_need(p, end, n * sizeof(u32), bad);
 734                 err = -ENOMEM;
 735                 pg = kmalloc(sizeof(*pg) + n*sizeof(u32), GFP_NOFS);
 736                 if (!pg)
 737                         goto bad;
 738                 pg->pgid = pgid;
 739                 pg->len = n;
 740                 for (j = 0; j < n; j++)
 741                         pg->osds[j] = ceph_decode_32(p);
 742
 743                 err = __insert_pg_mapping(pg, &map->pg_temp);
 744                 if (err)
 745                         goto bad;
 746                 dout(" added pg_temp %llx len %d\n", *(u64 *)&pgid, len);
 747         }
 748
 749         /* crush */
 750         ceph_decode_32_safe(p, end, len, bad);
 751         dout("osdmap_decode crush len %d from off 0x%x\n", len,
 752              (int)(*p - start));
 753         ceph_decode_need(p, end, len, bad);
 754         map->crush = crush_decode(*p, end);
 755         *p += len;
 756         if (IS_ERR(map->crush)) {
 757                 err = PTR_ERR(map->crush);
 758                 map->crush = NULL;
 759                 goto bad;
 760         }
 761
 762         /* ignore the rest of the map */
 763         *p = end;
 764
 765         dout("osdmap_decode done %p %p\n", *p, end);
 766         return map;
 767
 768 bad:
 769         dout("osdmap_decode fail err %d\n", err);
 770         ceph_osdmap_destroy(map);
 771         return ERR_PTR(err);
 772 }
 773
 774 /*
 775  * decode and apply an incremental map update.
 776  */
 777 struct ceph_osdmap *osdmap_apply_incremental(void **p, void *end,
 778                                              struct ceph_osdmap *map,
 779                                              struct ceph_messenger *msgr)
 780 {
 781         struct crush_map *newcrush = NULL;
 782         struct ceph_fsid fsid;
 783         u32 epoch = 0;
 784         struct ceph_timespec modified;
 785         u32 len, pool;
 786         __s32 new_pool_max, new_flags, max;
 787         void *start = *p;
 788         int err = -EINVAL;
 789         u16 version;
 790
 791         ceph_decode_16_safe(p, end, version, bad);
 792         if (version > CEPH_OSDMAP_INC_VERSION) {
 793                 pr_warning("got unknown v %d > %d of inc osdmap\n", version,
 794                            CEPH_OSDMAP_INC_VERSION);
 795                 goto bad;
 796         }
 797
 798         ceph_decode_need(p, end, sizeof(fsid)+sizeof(modified)+2*sizeof(u32),
 799                          bad);
 800         ceph_decode_copy(p, &fsid, sizeof(fsid));
 801         epoch = ceph_decode_32(p);
 802         BUG_ON(epoch != map->epoch+1);
 803         ceph_decode_copy(p, &modified, sizeof(modified));
 804         new_pool_max = ceph_decode_32(p);
 805         new_flags = ceph_decode_32(p);
 806
 807         /* full map? */
 808         ceph_decode_32_safe(p, end, len, bad);
 809         if (len > 0) {
 810                 dout("apply_incremental full map len %d, %p to %p\n",
 811                      len, *p, end);
 812                 return osdmap_decode(p, min(*p+len, end));
 813         }
 814
 815         /* new crush? */
 816         ceph_decode_32_safe(p, end, len, bad);
 817         if (len > 0) {
 818                 dout("apply_incremental new crush map len %d, %p to %p\n",
 819                      len, *p, end);
 820                 newcrush = crush_decode(*p, min(*p+len, end));
 821                 if (IS_ERR(newcrush))
 822                         return ERR_CAST(newcrush);
 823                 *p += len;
 824         }
 825
 826         /* new flags? */
 827         if (new_flags >= 0)
 828                 map->flags = new_flags;
 829         if (new_pool_max >= 0)
 830                 map->pool_max = new_pool_max;
 831
 832         ceph_decode_need(p, end, 5*sizeof(u32), bad);
 833
 834         /* new max? */
 835         max = ceph_decode_32(p);
 836         if (max >= 0) {
 837                 err = osdmap_set_max_osd(map, max);
 838                 if (err < 0)
 839                         goto bad;
 840         }
 841
 842         map->epoch++;
 843         map->modified = modified;
 844         if (newcrush) {
 845                 if (map->crush)
 846                         crush_destroy(map->crush);
 847                 map->crush = newcrush;
 848                 newcrush = NULL;
 849         }
 850
 851         /* new_pool */
 852         ceph_decode_32_safe(p, end, len, bad);
 853         while (len--) {
 854                 __u8 ev;
 855                 struct ceph_pg_pool_info *pi;
 856
 857                 ceph_decode_32_safe(p, end, pool, bad);
 858                 ceph_decode_need(p, end, 1 + sizeof(pi->v), bad);
 859                 ev = ceph_decode_8(p);  /* encoding version */
 860                 if (ev > CEPH_PG_POOL_VERSION) {
 861                         pr_warning("got unknown v %d > %d of ceph_pg_pool\n",
 862                                    ev, CEPH_PG_POOL_VERSION);
 863                         err = -EINVAL;
 864                         goto bad;
 865                 }
 866                 pi = __lookup_pg_pool(&map->pg_pools, pool);
 867                 if (!pi) {
 868                         pi = kzalloc(sizeof(*pi), GFP_NOFS);
 869                         if (!pi) {
 870                                 err = -ENOMEM;
 871                                 goto bad;
 872                         }
 873                         pi->id = pool;
 874                         __insert_pg_pool(&map->pg_pools, pi);
 875                 }
 876                 err = __decode_pool(p, end, pi);
 877                 if (err < 0)
 878                         goto bad;
 879         }
 880         if (version >= 5) {
 881                 err = __decode_pool_names(p, end, map);
 882                 if (err < 0)
 883                         goto bad;
 884         }
 885
 886         /* old_pool */
 887         ceph_decode_32_safe(p, end, len, bad);
 888         while (len--) {
 889                 struct ceph_pg_pool_info *pi;
 890
 891                 ceph_decode_32_safe(p, end, pool, bad);
 892                 pi = __lookup_pg_pool(&map->pg_pools, pool);
 893                 if (pi)
 894                         __remove_pg_pool(&map->pg_pools, pi);
 895         }
 896
 897         /* new_up */
 898         err = -EINVAL;
 899         ceph_decode_32_safe(p, end, len, bad);
 900         while (len--) {
 901                 u32 osd;
 902                 struct ceph_entity_addr addr;
 903                 ceph_decode_32_safe(p, end, osd, bad);
 904                 ceph_decode_copy_safe(p, end, &addr, sizeof(addr), bad);
 905                 ceph_decode_addr(&addr);
 906                 pr_info("osd%d up\n", osd);
 907                 BUG_ON(osd >= map->max_osd);
 908                 map->osd_state[osd] |= CEPH_OSD_UP;
 909                 map->osd_addr[osd] = addr;
 910         }
 911
 912         /* new_state */
 913         ceph_decode_32_safe(p, end, len, bad);
 914         while (len--) {
 915                 u32 osd;
 916                 u8 xorstate;
 917                 ceph_decode_32_safe(p, end, osd, bad);
 918                 xorstate = **(u8 **)p;
 919                 (*p)++;  /* clean flag */
 920                 if (xorstate == 0)
 921                         xorstate = CEPH_OSD_UP;
 922                 if (xorstate & CEPH_OSD_UP)
 923                         pr_info("osd%d down\n", osd);
 924                 if (osd < map->max_osd)
 925                         map->osd_state[osd] ^= xorstate;
 926         }
 927
 928         /* new_weight */
 929         ceph_decode_32_safe(p, end, len, bad);
 930         while (len--) {
 931                 u32 osd, off;
 932                 ceph_decode_need(p, end, sizeof(u32)*2, bad);
 933                 osd = ceph_decode_32(p);
 934                 off = ceph_decode_32(p);
 935                 pr_info("osd%d weight 0x%x %s\n", osd, off,
 936                      off == CEPH_OSD_IN ? "(in)" :
 937                      (off == CEPH_OSD_OUT ? "(out)" : ""));
 938                 if (osd < map->max_osd)
 939                         map->osd_weight[osd] = off;
 940         }
 941
 942         /* new_pg_temp */
 943         ceph_decode_32_safe(p, end, len, bad);
 944         while (len--) {
 945                 struct ceph_pg_mapping *pg;
 946                 int j;
 947                 struct ceph_pg pgid;
 948                 u32 pglen;
 949                 ceph_decode_need(p, end, sizeof(u64) + sizeof(u32), bad);
 950                 ceph_decode_copy(p, &pgid, sizeof(pgid));
 951                 pglen = ceph_decode_32(p);
 952
 953                 if (pglen) {
 954                         ceph_decode_need(p, end, pglen*sizeof(u32), bad);
 955
 956                         /* removing existing (if any) */
 957                         (void) __remove_pg_mapping(&map->pg_temp, pgid);
 958
 959                         /* insert */
 960                         err = -EINVAL;
 961                         if (pglen > (UINT_MAX - sizeof(*pg)) / sizeof(u32))
 962                                 goto bad;
 963                         err = -ENOMEM;
 964                         pg = kmalloc(sizeof(*pg) + sizeof(u32)*pglen, GFP_NOFS);
 965                         if (!pg)
 966                                 goto bad;
 967                         pg->pgid = pgid;
 968                         pg->len = pglen;
 969                         for (j = 0; j < pglen; j++)
 970                                 pg->osds[j] = ceph_decode_32(p);
 971                         err = __insert_pg_mapping(pg, &map->pg_temp);
 972                         if (err) {
 973                                 kfree(pg);
 974                                 goto bad;
 975                         }
 976                         dout(" added pg_temp %llx len %d\n", *(u64 *)&pgid,
 977                              pglen);
 978                 } else {
 979                         /* remove */
 980                         __remove_pg_mapping(&map->pg_temp, pgid);
 981                 }
 982         }
 983
 984         /* ignore the rest */
 985         *p = end;
 986         return map;
 987
 988 bad:
 989         pr_err("corrupt inc osdmap epoch %d off %d (%p of %p-%p)\n",
 990                epoch, (int)(*p - start), *p, start, end);
 991         print_hex_dump(KERN_DEBUG, "osdmap: ",
 992                        DUMP_PREFIX_OFFSET, 16, 1,
 993                        start, end - start, true);
 994         if (newcrush)
 995                 crush_destroy(newcrush);
 996         return ERR_PTR(err);
 997 }
 998
 999
1000
1001
1002 /*
1003  * calculate file layout from given offset, length.
1004  * fill in correct oid, logical length, and object extent
1005  * offset, length.
1006  *
1007  * for now, we write only a single su, until we can
1008  * pass a stride back to the caller.
1009  */
1010 int ceph_calc_file_object_mapping(struct ceph_file_layout *layout,
1011                                    u64 off, u64 len,
1012                                    u64 *ono,
1013                                    u64 *oxoff, u64 *oxlen)
1014 {
1015         u32 osize = le32_to_cpu(layout->fl_object_size);
1016         u32 su = le32_to_cpu(layout->fl_stripe_unit);
1017         u32 sc = le32_to_cpu(layout->fl_stripe_count);
1018         u32 bl, stripeno, stripepos, objsetno;
1019         u32 su_per_object;
1020         u64 t, su_offset;
1021
1022         dout("mapping %llu~%llu  osize %u fl_su %u\n", off, len,
1023              osize, su);
1024         if (su == 0 || sc == 0)
1025                 goto invalid;
1026         su_per_object = osize / su;
1027         if (su_per_object == 0)
1028                 goto invalid;
1029         dout("osize %u / su %u = su_per_object %u\n", osize, su,
1030              su_per_object);
1031
1032         if ((su & ~PAGE_MASK) != 0)
1033                 goto invalid;
1034
1035         /* bl = *off / su; */
1036         t = off;
1037         do_div(t, su);
1038         bl = t;
1039         dout("off %llu / su %u = bl %u\n", off, su, bl);
1040
1041         stripeno = bl / sc;
1042         stripepos = bl % sc;
1043         objsetno = stripeno / su_per_object;
1044
1045         *ono = objsetno * sc + stripepos;
1046         dout("objset %u * sc %u = ono %u\n", objsetno, sc, (unsigned int)*ono);
1047
1048         /* *oxoff = *off % layout->fl_stripe_unit;  # offset in su */
1049         t = off;
1050         su_offset = do_div(t, su);
1051         *oxoff = su_offset + (stripeno % su_per_object) * su;
1052
1053         /*
1054          * Calculate the length of the extent being written to the selected
1055          * object. This is the minimum of the full length requested (len) or
1056          * the remainder of the current stripe being written to.
1057          */
1058         *oxlen = min_t(u64, len, su - su_offset);
1059
1060         dout(" obj extent %llu~%llu\n", *oxoff, *oxlen);
1061         return 0;
1062
1063 invalid:
1064         dout(" invalid layout\n");
1065         *ono = 0;
1066         *oxoff = 0;
1067         *oxlen = 0;
1068         return -EINVAL;
1069 }
1070 EXPORT_SYMBOL(ceph_calc_file_object_mapping);
1071
1072 /*
1073  * calculate an object layout (i.e. pgid) from an oid,
1074  * file_layout, and osdmap
1075  */
1076 int ceph_calc_object_layout(struct ceph_object_layout *ol,
1077                             const char *oid,
1078                             struct ceph_file_layout *fl,
1079                             struct ceph_osdmap *osdmap)
1080 {
1081         unsigned int num, num_mask;
1082         struct ceph_pg pgid;
1083         int poolid = le32_to_cpu(fl->fl_pg_pool);
1084         struct ceph_pg_pool_info *pool;
1085         unsigned int ps;
1086
1087         BUG_ON(!osdmap);
1088
1089         pool = __lookup_pg_pool(&osdmap->pg_pools, poolid);
1090         if (!pool)
1091                 return -EIO;
1092         ps = ceph_str_hash(pool->v.object_hash, oid, strlen(oid));
1093         num = le32_to_cpu(pool->v.pg_num);
1094         num_mask = pool->pg_num_mask;
1095
1096         pgid.ps = cpu_to_le16(ps);
1097         pgid.preferred = cpu_to_le16(-1);
1098         pgid.pool = fl->fl_pg_pool;
1099         dout("calc_object_layout '%s' pgid %d.%x\n", oid, poolid, ps);
1100
1101         ol->ol_pgid = pgid;
1102         ol->ol_stripe_unit = fl->fl_object_stripe_unit;
1103         return 0;
1104 }
1105 EXPORT_SYMBOL(ceph_calc_object_layout);
1106
1107 /*
1108  * Calculate raw osd vector for the given pgid.  Return pointer to osd
1109  * array, or NULL on failure.
1110  */
1111 static int *calc_pg_raw(struct ceph_osdmap *osdmap, struct ceph_pg pgid,
1112                         int *osds, int *num)
1113 {
1114         struct ceph_pg_mapping *pg;
1115         struct ceph_pg_pool_info *pool;
1116         int ruleno;
1117         unsigned int poolid, ps, pps, t, r;
1118
1119         poolid = le32_to_cpu(pgid.pool);
1120         ps = le16_to_cpu(pgid.ps);
1121
1122         pool = __lookup_pg_pool(&osdmap->pg_pools, poolid);
1123         if (!pool)
1124                 return NULL;
1125
1126         /* pg_temp? */
1127         t = ceph_stable_mod(ps, le32_to_cpu(pool->v.pg_num),
1128                             pool->pgp_num_mask);
1129         pgid.ps = cpu_to_le16(t);
1130         pg = __lookup_pg_mapping(&osdmap->pg_temp, pgid);
1131         if (pg) {
1132                 *num = pg->len;
1133                 return pg->osds;
1134         }
1135
1136         /* crush */
1137         ruleno = crush_find_rule(osdmap->crush, pool->v.crush_ruleset,
1138                                  pool->v.type, pool->v.size);
1139         if (ruleno < 0) {
1140                 pr_err("no crush rule pool %d ruleset %d type %d size %d\n",
1141                        poolid, pool->v.crush_ruleset, pool->v.type,
1142                        pool->v.size);
1143                 return NULL;
1144         }
1145
1146         pps = ceph_stable_mod(ps,
1147                               le32_to_cpu(pool->v.pgp_num),
1148                               pool->pgp_num_mask);
1149         pps += poolid;
1150         r = crush_do_rule(osdmap->crush, ruleno, pps, osds,
1151                           min_t(int, pool->v.size, *num),
1152                           osdmap->osd_weight);
1153         if (r < 0) {
1154                 pr_err("error %d from crush rule: pool %d ruleset %d type %d"
1155                        " size %d\n", r, poolid, pool->v.crush_ruleset,
1156                        pool->v.type, pool->v.size);
1157                 return NULL;
1158         }
1159         *num = r;
1160         return osds;
1161 }
1162
1163 /*
1164  * Return acting set for given pgid.
1165  */
1166 int ceph_calc_pg_acting(struct ceph_osdmap *osdmap, struct ceph_pg pgid,
1167                         int *acting)
1168 {
1169         int rawosds[CEPH_PG_MAX_SIZE], *osds;
1170         int i, o, num = CEPH_PG_MAX_SIZE;
1171
1172         osds = calc_pg_raw(osdmap, pgid, rawosds, &num);
1173         if (!osds)
1174                 return -1;
1175
1176         /* primary is first up osd */
1177         o = 0;
1178         for (i = 0; i < num; i++)
1179                 if (ceph_osd_is_up(osdmap, osds[i]))
1180                         acting[o++] = osds[i];
1181         return o;
1182 }
1183
1184 /*
1185  * Return primary osd for given pgid, or -1 if none.
1186  */
1187 int ceph_calc_pg_primary(struct ceph_osdmap *osdmap, struct ceph_pg pgid)
1188 {
1189         int rawosds[CEPH_PG_MAX_SIZE], *osds;
1190         int i, num = CEPH_PG_MAX_SIZE;
1191
1192         osds = calc_pg_raw(osdmap, pgid, rawosds, &num);
1193         if (!osds)
1194                 return -1;
1195
1196         /* primary is first up osd */
1197         for (i = 0; i < num; i++)
1198                 if (ceph_osd_is_up(osdmap, osds[i]))
1199                         return osds[i];
1200         return -1;
1201 }
1202 EXPORT_SYMBOL(ceph_calc_pg_primary);