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         if (l.pool < r.pool)
 356                 return -1;
 357         if (l.pool > r.pool)
 358                 return 1;
 359         if (l.seed < r.seed)
 360                 return -1;
 361         if (l.seed > r.seed)
 362                 return 1;
 363         return 0;
 364 }
 365
 366 static int __insert_pg_mapping(struct ceph_pg_mapping *new,
 367                                struct rb_root *root)
 368 {
 369         struct rb_node **p = &root->rb_node;
 370         struct rb_node *parent = NULL;
 371         struct ceph_pg_mapping *pg = NULL;
 372         int c;
 373
 374         dout("__insert_pg_mapping %llx %p\n", *(u64 *)&new->pgid, new);
 375         while (*p) {
 376                 parent = *p;
 377                 pg = rb_entry(parent, struct ceph_pg_mapping, node);
 378                 c = pgid_cmp(new->pgid, pg->pgid);
 379                 if (c < 0)
 380                         p = &(*p)->rb_left;
 381                 else if (c > 0)
 382                         p = &(*p)->rb_right;
 383                 else
 384                         return -EEXIST;
 385         }
 386
 387         rb_link_node(&new->node, parent, p);
 388         rb_insert_color(&new->node, root);
 389         return 0;
 390 }
 391
 392 static struct ceph_pg_mapping *__lookup_pg_mapping(struct rb_root *root,
 393                                                    struct ceph_pg pgid)
 394 {
 395         struct rb_node *n = root->rb_node;
 396         struct ceph_pg_mapping *pg;
 397         int c;
 398
 399         while (n) {
 400                 pg = rb_entry(n, struct ceph_pg_mapping, node);
 401                 c = pgid_cmp(pgid, pg->pgid);
 402                 if (c < 0) {
 403                         n = n->rb_left;
 404                 } else if (c > 0) {
 405                         n = n->rb_right;
 406                 } else {
 407                         dout("__lookup_pg_mapping %lld.%x got %p\n",
 408                              pgid.pool, pgid.seed, pg);
 409                         return pg;
 410                 }
 411         }
 412         return NULL;
 413 }
 414
 415 static int __remove_pg_mapping(struct rb_root *root, struct ceph_pg pgid)
 416 {
 417         struct ceph_pg_mapping *pg = __lookup_pg_mapping(root, pgid);
 418
 419         if (pg) {
 420                 dout("__remove_pg_mapping %lld.%x %p\n", pgid.pool, pgid.seed,
 421                      pg);
 422                 rb_erase(&pg->node, root);
 423                 kfree(pg);
 424                 return 0;
 425         }
 426         dout("__remove_pg_mapping %lld.%x dne\n", pgid.pool, pgid.seed);
 427         return -ENOENT;
 428 }
 429
 430 /*
 431  * rbtree of pg pool info
 432  */
 433 static int __insert_pg_pool(struct rb_root *root, struct ceph_pg_pool_info *new)
 434 {
 435         struct rb_node **p = &root->rb_node;
 436         struct rb_node *parent = NULL;
 437         struct ceph_pg_pool_info *pi = NULL;
 438
 439         while (*p) {
 440                 parent = *p;
 441                 pi = rb_entry(parent, struct ceph_pg_pool_info, node);
 442                 if (new->id < pi->id)
 443                         p = &(*p)->rb_left;
 444                 else if (new->id > pi->id)
 445                         p = &(*p)->rb_right;
 446                 else
 447                         return -EEXIST;
 448         }
 449
 450         rb_link_node(&new->node, parent, p);
 451         rb_insert_color(&new->node, root);
 452         return 0;
 453 }
 454
 455 static struct ceph_pg_pool_info *__lookup_pg_pool(struct rb_root *root, int id)
 456 {
 457         struct ceph_pg_pool_info *pi;
 458         struct rb_node *n = root->rb_node;
 459
 460         while (n) {
 461                 pi = rb_entry(n, struct ceph_pg_pool_info, node);
 462                 if (id < pi->id)
 463                         n = n->rb_left;
 464                 else if (id > pi->id)
 465                         n = n->rb_right;
 466                 else
 467                         return pi;
 468         }
 469         return NULL;
 470 }
 471
 472 const char *ceph_pg_pool_name_by_id(struct ceph_osdmap *map, u64 id)
 473 {
 474         struct ceph_pg_pool_info *pi;
 475
 476         if (id == CEPH_NOPOOL)
 477                 return NULL;
 478
 479         if (WARN_ON_ONCE(id > (u64) INT_MAX))
 480                 return NULL;
 481
 482         pi = __lookup_pg_pool(&map->pg_pools, (int) id);
 483
 484         return pi ? pi->name : NULL;
 485 }
 486 EXPORT_SYMBOL(ceph_pg_pool_name_by_id);
 487
 488 int ceph_pg_poolid_by_name(struct ceph_osdmap *map, const char *name)
 489 {
 490         struct rb_node *rbp;
 491
 492         for (rbp = rb_first(&map->pg_pools); rbp; rbp = rb_next(rbp)) {
 493                 struct ceph_pg_pool_info *pi =
 494                         rb_entry(rbp, struct ceph_pg_pool_info, node);
 495                 if (pi->name && strcmp(pi->name, name) == 0)
 496                         return pi->id;
 497         }
 498         return -ENOENT;
 499 }
 500 EXPORT_SYMBOL(ceph_pg_poolid_by_name);
 501
 502 static void __remove_pg_pool(struct rb_root *root, struct ceph_pg_pool_info *pi)
 503 {
 504         rb_erase(&pi->node, root);
 505         kfree(pi->name);
 506         kfree(pi);
 507 }
 508
 509 static int __decode_pool(void **p, void *end, struct ceph_pg_pool_info *pi)
 510 {
 511         unsigned int n, m;
 512
 513         ceph_decode_copy(p, &pi->v, sizeof(pi->v));
 514         calc_pg_masks(pi);
 515
 516         /* num_snaps * snap_info_t */
 517         n = le32_to_cpu(pi->v.num_snaps);
 518         while (n--) {
 519                 ceph_decode_need(p, end, sizeof(u64) + 1 + sizeof(u64) +
 520                                  sizeof(struct ceph_timespec), bad);
 521                 *p += sizeof(u64) +       /* key */
 522                         1 + sizeof(u64) + /* u8, snapid */
 523                         sizeof(struct ceph_timespec);
 524                 m = ceph_decode_32(p);    /* snap name */
 525                 *p += m;
 526         }
 527
 528         *p += le32_to_cpu(pi->v.num_removed_snap_intervals) * sizeof(u64) * 2;
 529         return 0;
 530
 531 bad:
 532         return -EINVAL;
 533 }
 534
 535 static int __decode_pool_names(void **p, void *end, struct ceph_osdmap *map)
 536 {
 537         struct ceph_pg_pool_info *pi;
 538         u32 num, len, pool;
 539
 540         ceph_decode_32_safe(p, end, num, bad);
 541         dout(" %d pool names\n", num);
 542         while (num--) {
 543                 ceph_decode_32_safe(p, end, pool, bad);
 544                 ceph_decode_32_safe(p, end, len, bad);
 545                 dout("  pool %d len %d\n", pool, len);
 546                 ceph_decode_need(p, end, len, bad);
 547                 pi = __lookup_pg_pool(&map->pg_pools, pool);
 548                 if (pi) {
 549                         char *name = kstrndup(*p, len, GFP_NOFS);
 550
 551                         if (!name)
 552                                 return -ENOMEM;
 553                         kfree(pi->name);
 554                         pi->name = name;
 555                         dout("  name is %s\n", pi->name);
 556                 }
 557                 *p += len;
 558         }
 559         return 0;
 560
 561 bad:
 562         return -EINVAL;
 563 }
 564
 565 /*
 566  * osd map
 567  */
 568 void ceph_osdmap_destroy(struct ceph_osdmap *map)
 569 {
 570         dout("osdmap_destroy %p\n", map);
 571         if (map->crush)
 572                 crush_destroy(map->crush);
 573         while (!RB_EMPTY_ROOT(&map->pg_temp)) {
 574                 struct ceph_pg_mapping *pg =
 575                         rb_entry(rb_first(&map->pg_temp),
 576                                  struct ceph_pg_mapping, node);
 577                 rb_erase(&pg->node, &map->pg_temp);
 578                 kfree(pg);
 579         }
 580         while (!RB_EMPTY_ROOT(&map->pg_pools)) {
 581                 struct ceph_pg_pool_info *pi =
 582                         rb_entry(rb_first(&map->pg_pools),
 583                                  struct ceph_pg_pool_info, node);
 584                 __remove_pg_pool(&map->pg_pools, pi);
 585         }
 586         kfree(map->osd_state);
 587         kfree(map->osd_weight);
 588         kfree(map->osd_addr);
 589         kfree(map);
 590 }
 591
 592 /*
 593  * adjust max osd value.  reallocate arrays.
 594  */
 595 static int osdmap_set_max_osd(struct ceph_osdmap *map, int max)
 596 {
 597         u8 *state;
 598         struct ceph_entity_addr *addr;
 599         u32 *weight;
 600
 601         state = kcalloc(max, sizeof(*state), GFP_NOFS);
 602         addr = kcalloc(max, sizeof(*addr), GFP_NOFS);
 603         weight = kcalloc(max, sizeof(*weight), GFP_NOFS);
 604         if (state == NULL || addr == NULL || weight == NULL) {
 605                 kfree(state);
 606                 kfree(addr);
 607                 kfree(weight);
 608                 return -ENOMEM;
 609         }
 610
 611         /* copy old? */
 612         if (map->osd_state) {
 613                 memcpy(state, map->osd_state, map->max_osd*sizeof(*state));
 614                 memcpy(addr, map->osd_addr, map->max_osd*sizeof(*addr));
 615                 memcpy(weight, map->osd_weight, map->max_osd*sizeof(*weight));
 616                 kfree(map->osd_state);
 617                 kfree(map->osd_addr);
 618                 kfree(map->osd_weight);
 619         }
 620
 621         map->osd_state = state;
 622         map->osd_weight = weight;
 623         map->osd_addr = addr;
 624         map->max_osd = max;
 625         return 0;
 626 }
 627
 628 /*
 629  * decode a full map.
 630  */
 631 struct ceph_osdmap *osdmap_decode(void **p, void *end)
 632 {
 633         struct ceph_osdmap *map;
 634         u16 version;
 635         u32 len, max, i;
 636         u8 ev;
 637         int err = -EINVAL;
 638         void *start = *p;
 639         struct ceph_pg_pool_info *pi;
 640
 641         dout("osdmap_decode %p to %p len %d\n", *p, end, (int)(end - *p));
 642
 643         map = kzalloc(sizeof(*map), GFP_NOFS);
 644         if (map == NULL)
 645                 return ERR_PTR(-ENOMEM);
 646         map->pg_temp = RB_ROOT;
 647
 648         ceph_decode_16_safe(p, end, version, bad);
 649         if (version > CEPH_OSDMAP_VERSION) {
 650                 pr_warning("got unknown v %d > %d of osdmap\n", version,
 651                            CEPH_OSDMAP_VERSION);
 652                 goto bad;
 653         }
 654
 655         ceph_decode_need(p, end, 2*sizeof(u64)+6*sizeof(u32), bad);
 656         ceph_decode_copy(p, &map->fsid, sizeof(map->fsid));
 657         map->epoch = ceph_decode_32(p);
 658         ceph_decode_copy(p, &map->created, sizeof(map->created));
 659         ceph_decode_copy(p, &map->modified, sizeof(map->modified));
 660
 661         ceph_decode_32_safe(p, end, max, bad);
 662         while (max--) {
 663                 ceph_decode_need(p, end, 4 + 1 + sizeof(pi->v), bad);
 664                 err = -ENOMEM;
 665                 pi = kzalloc(sizeof(*pi), GFP_NOFS);
 666                 if (!pi)
 667                         goto bad;
 668                 pi->id = ceph_decode_32(p);
 669                 err = -EINVAL;
 670                 ev = ceph_decode_8(p); /* encoding version */
 671                 if (ev > CEPH_PG_POOL_VERSION) {
 672                         pr_warning("got unknown v %d > %d of ceph_pg_pool\n",
 673                                    ev, CEPH_PG_POOL_VERSION);
 674                         kfree(pi);
 675                         goto bad;
 676                 }
 677                 err = __decode_pool(p, end, pi);
 678                 if (err < 0) {
 679                         kfree(pi);
 680                         goto bad;
 681                 }
 682                 __insert_pg_pool(&map->pg_pools, pi);
 683         }
 684
 685         if (version >= 5) {
 686                 err = __decode_pool_names(p, end, map);
 687                 if (err < 0) {
 688                         dout("fail to decode pool names");
 689                         goto bad;
 690                 }
 691         }
 692
 693         ceph_decode_32_safe(p, end, map->pool_max, bad);
 694
 695         ceph_decode_32_safe(p, end, map->flags, bad);
 696
 697         max = ceph_decode_32(p);
 698
 699         /* (re)alloc osd arrays */
 700         err = osdmap_set_max_osd(map, max);
 701         if (err < 0)
 702                 goto bad;
 703         dout("osdmap_decode max_osd = %d\n", map->max_osd);
 704
 705         /* osds */
 706         err = -EINVAL;
 707         ceph_decode_need(p, end, 3*sizeof(u32) +
 708                          map->max_osd*(1 + sizeof(*map->osd_weight) +
 709                                        sizeof(*map->osd_addr)), bad);
 710         *p += 4; /* skip length field (should match max) */
 711         ceph_decode_copy(p, map->osd_state, map->max_osd);
 712
 713         *p += 4; /* skip length field (should match max) */
 714         for (i = 0; i < map->max_osd; i++)
 715                 map->osd_weight[i] = ceph_decode_32(p);
 716
 717         *p += 4; /* skip length field (should match max) */
 718         ceph_decode_copy(p, map->osd_addr, map->max_osd*sizeof(*map->osd_addr));
 719         for (i = 0; i < map->max_osd; i++)
 720                 ceph_decode_addr(&map->osd_addr[i]);
 721
 722         /* pg_temp */
 723         ceph_decode_32_safe(p, end, len, bad);
 724         for (i = 0; i < len; i++) {
 725                 int n, j;
 726                 struct ceph_pg pgid;
 727                 struct ceph_pg_v1 pgid_v1;
 728                 struct ceph_pg_mapping *pg;
 729
 730                 ceph_decode_need(p, end, sizeof(u32) + sizeof(u64), bad);
 731                 ceph_decode_copy(p, &pgid_v1, sizeof(pgid_v1));
 732                 pgid.pool = le32_to_cpu(pgid_v1.pool);
 733                 pgid.seed = le16_to_cpu(pgid_v1.ps);
 734                 n = ceph_decode_32(p);
 735                 err = -EINVAL;
 736                 if (n > (UINT_MAX - sizeof(*pg)) / sizeof(u32))
 737                         goto bad;
 738                 ceph_decode_need(p, end, n * sizeof(u32), bad);
 739                 err = -ENOMEM;
 740                 pg = kmalloc(sizeof(*pg) + n*sizeof(u32), GFP_NOFS);
 741                 if (!pg)
 742                         goto bad;
 743                 pg->pgid = pgid;
 744                 pg->len = n;
 745                 for (j = 0; j < n; j++)
 746                         pg->osds[j] = ceph_decode_32(p);
 747
 748                 err = __insert_pg_mapping(pg, &map->pg_temp);
 749                 if (err)
 750                         goto bad;
 751                 dout(" added pg_temp %lld.%x len %d\n", pgid.pool, pgid.seed,
 752                      len);
 753         }
 754
 755         /* crush */
 756         ceph_decode_32_safe(p, end, len, bad);
 757         dout("osdmap_decode crush len %d from off 0x%x\n", len,
 758              (int)(*p - start));
 759         ceph_decode_need(p, end, len, bad);
 760         map->crush = crush_decode(*p, end);
 761         *p += len;
 762         if (IS_ERR(map->crush)) {
 763                 err = PTR_ERR(map->crush);
 764                 map->crush = NULL;
 765                 goto bad;
 766         }
 767
 768         /* ignore the rest of the map */
 769         *p = end;
 770
 771         dout("osdmap_decode done %p %p\n", *p, end);
 772         return map;
 773
 774 bad:
 775         dout("osdmap_decode fail err %d\n", err);
 776         ceph_osdmap_destroy(map);
 777         return ERR_PTR(err);
 778 }
 779
 780 /*
 781  * decode and apply an incremental map update.
 782  */
 783 struct ceph_osdmap *osdmap_apply_incremental(void **p, void *end,
 784                                              struct ceph_osdmap *map,
 785                                              struct ceph_messenger *msgr)
 786 {
 787         struct crush_map *newcrush = NULL;
 788         struct ceph_fsid fsid;
 789         u32 epoch = 0;
 790         struct ceph_timespec modified;
 791         u32 len, pool;
 792         __s32 new_pool_max, new_flags, max;
 793         void *start = *p;
 794         int err = -EINVAL;
 795         u16 version;
 796
 797         ceph_decode_16_safe(p, end, version, bad);
 798         if (version > CEPH_OSDMAP_INC_VERSION) {
 799                 pr_warning("got unknown v %d > %d of inc osdmap\n", version,
 800                            CEPH_OSDMAP_INC_VERSION);
 801                 goto bad;
 802         }
 803
 804         ceph_decode_need(p, end, sizeof(fsid)+sizeof(modified)+2*sizeof(u32),
 805                          bad);
 806         ceph_decode_copy(p, &fsid, sizeof(fsid));
 807         epoch = ceph_decode_32(p);
 808         BUG_ON(epoch != map->epoch+1);
 809         ceph_decode_copy(p, &modified, sizeof(modified));
 810         new_pool_max = ceph_decode_32(p);
 811         new_flags = ceph_decode_32(p);
 812
 813         /* full map? */
 814         ceph_decode_32_safe(p, end, len, bad);
 815         if (len > 0) {
 816                 dout("apply_incremental full map len %d, %p to %p\n",
 817                      len, *p, end);
 818                 return osdmap_decode(p, min(*p+len, end));
 819         }
 820
 821         /* new crush? */
 822         ceph_decode_32_safe(p, end, len, bad);
 823         if (len > 0) {
 824                 dout("apply_incremental new crush map len %d, %p to %p\n",
 825                      len, *p, end);
 826                 newcrush = crush_decode(*p, min(*p+len, end));
 827                 if (IS_ERR(newcrush))
 828                         return ERR_CAST(newcrush);
 829                 *p += len;
 830         }
 831
 832         /* new flags? */
 833         if (new_flags >= 0)
 834                 map->flags = new_flags;
 835         if (new_pool_max >= 0)
 836                 map->pool_max = new_pool_max;
 837
 838         ceph_decode_need(p, end, 5*sizeof(u32), bad);
 839
 840         /* new max? */
 841         max = ceph_decode_32(p);
 842         if (max >= 0) {
 843                 err = osdmap_set_max_osd(map, max);
 844                 if (err < 0)
 845                         goto bad;
 846         }
 847
 848         map->epoch++;
 849         map->modified = modified;
 850         if (newcrush) {
 851                 if (map->crush)
 852                         crush_destroy(map->crush);
 853                 map->crush = newcrush;
 854                 newcrush = NULL;
 855         }
 856
 857         /* new_pool */
 858         ceph_decode_32_safe(p, end, len, bad);
 859         while (len--) {
 860                 __u8 ev;
 861                 struct ceph_pg_pool_info *pi;
 862
 863                 ceph_decode_32_safe(p, end, pool, bad);
 864                 ceph_decode_need(p, end, 1 + sizeof(pi->v), bad);
 865                 ev = ceph_decode_8(p);  /* encoding version */
 866                 if (ev > CEPH_PG_POOL_VERSION) {
 867                         pr_warning("got unknown v %d > %d of ceph_pg_pool\n",
 868                                    ev, CEPH_PG_POOL_VERSION);
 869                         err = -EINVAL;
 870                         goto bad;
 871                 }
 872                 pi = __lookup_pg_pool(&map->pg_pools, pool);
 873                 if (!pi) {
 874                         pi = kzalloc(sizeof(*pi), GFP_NOFS);
 875                         if (!pi) {
 876                                 err = -ENOMEM;
 877                                 goto bad;
 878                         }
 879                         pi->id = pool;
 880                         __insert_pg_pool(&map->pg_pools, pi);
 881                 }
 882                 err = __decode_pool(p, end, pi);
 883                 if (err < 0)
 884                         goto bad;
 885         }
 886         if (version >= 5) {
 887                 err = __decode_pool_names(p, end, map);
 888                 if (err < 0)
 889                         goto bad;
 890         }
 891
 892         /* old_pool */
 893         ceph_decode_32_safe(p, end, len, bad);
 894         while (len--) {
 895                 struct ceph_pg_pool_info *pi;
 896
 897                 ceph_decode_32_safe(p, end, pool, bad);
 898                 pi = __lookup_pg_pool(&map->pg_pools, pool);
 899                 if (pi)
 900                         __remove_pg_pool(&map->pg_pools, pi);
 901         }
 902
 903         /* new_up */
 904         err = -EINVAL;
 905         ceph_decode_32_safe(p, end, len, bad);
 906         while (len--) {
 907                 u32 osd;
 908                 struct ceph_entity_addr addr;
 909                 ceph_decode_32_safe(p, end, osd, bad);
 910                 ceph_decode_copy_safe(p, end, &addr, sizeof(addr), bad);
 911                 ceph_decode_addr(&addr);
 912                 pr_info("osd%d up\n", osd);
 913                 BUG_ON(osd >= map->max_osd);
 914                 map->osd_state[osd] |= CEPH_OSD_UP;
 915                 map->osd_addr[osd] = addr;
 916         }
 917
 918         /* new_state */
 919         ceph_decode_32_safe(p, end, len, bad);
 920         while (len--) {
 921                 u32 osd;
 922                 u8 xorstate;
 923                 ceph_decode_32_safe(p, end, osd, bad);
 924                 xorstate = **(u8 **)p;
 925                 (*p)++;  /* clean flag */
 926                 if (xorstate == 0)
 927                         xorstate = CEPH_OSD_UP;
 928                 if (xorstate & CEPH_OSD_UP)
 929                         pr_info("osd%d down\n", osd);
 930                 if (osd < map->max_osd)
 931                         map->osd_state[osd] ^= xorstate;
 932         }
 933
 934         /* new_weight */
 935         ceph_decode_32_safe(p, end, len, bad);
 936         while (len--) {
 937                 u32 osd, off;
 938                 ceph_decode_need(p, end, sizeof(u32)*2, bad);
 939                 osd = ceph_decode_32(p);
 940                 off = ceph_decode_32(p);
 941                 pr_info("osd%d weight 0x%x %s\n", osd, off,
 942                      off == CEPH_OSD_IN ? "(in)" :
 943                      (off == CEPH_OSD_OUT ? "(out)" : ""));
 944                 if (osd < map->max_osd)
 945                         map->osd_weight[osd] = off;
 946         }
 947
 948         /* new_pg_temp */
 949         ceph_decode_32_safe(p, end, len, bad);
 950         while (len--) {
 951                 struct ceph_pg_mapping *pg;
 952                 int j;
 953                 struct ceph_pg_v1 pgid_v1;
 954                 struct ceph_pg pgid;
 955                 u32 pglen;
 956                 ceph_decode_need(p, end, sizeof(u64) + sizeof(u32), bad);
 957                 ceph_decode_copy(p, &pgid_v1, sizeof(pgid_v1));
 958                 pgid.pool = le32_to_cpu(pgid_v1.pool);
 959                 pgid.seed = le16_to_cpu(pgid_v1.ps);
 960                 pglen = ceph_decode_32(p);
 961
 962                 if (pglen) {
 963                         ceph_decode_need(p, end, pglen*sizeof(u32), bad);
 964
 965                         /* removing existing (if any) */
 966                         (void) __remove_pg_mapping(&map->pg_temp, pgid);
 967
 968                         /* insert */
 969                         err = -EINVAL;
 970                         if (pglen > (UINT_MAX - sizeof(*pg)) / sizeof(u32))
 971                                 goto bad;
 972                         err = -ENOMEM;
 973                         pg = kmalloc(sizeof(*pg) + sizeof(u32)*pglen, GFP_NOFS);
 974                         if (!pg)
 975                                 goto bad;
 976                         pg->pgid = pgid;
 977                         pg->len = pglen;
 978                         for (j = 0; j < pglen; j++)
 979                                 pg->osds[j] = ceph_decode_32(p);
 980                         err = __insert_pg_mapping(pg, &map->pg_temp);
 981                         if (err) {
 982                                 kfree(pg);
 983                                 goto bad;
 984                         }
 985                         dout(" added pg_temp %lld.%x len %d\n", pgid.pool,
 986                              pgid.seed, pglen);
 987                 } else {
 988                         /* remove */
 989                         __remove_pg_mapping(&map->pg_temp, pgid);
 990                 }
 991         }
 992
 993         /* ignore the rest */
 994         *p = end;
 995         return map;
 996
 997 bad:
 998         pr_err("corrupt inc osdmap epoch %d off %d (%p of %p-%p)\n",
 999                epoch, (int)(*p - start), *p, start, end);
1000         print_hex_dump(KERN_DEBUG, "osdmap: ",
1001                        DUMP_PREFIX_OFFSET, 16, 1,
1002                        start, end - start, true);
1003         if (newcrush)
1004                 crush_destroy(newcrush);
1005         return ERR_PTR(err);
1006 }
1007
1008
1009
1010
1011 /*
1012  * calculate file layout from given offset, length.
1013  * fill in correct oid, logical length, and object extent
1014  * offset, length.
1015  *
1016  * for now, we write only a single su, until we can
1017  * pass a stride back to the caller.
1018  */
1019 int ceph_calc_file_object_mapping(struct ceph_file_layout *layout,
1020                                    u64 off, u64 len,
1021                                    u64 *ono,
1022                                    u64 *oxoff, u64 *oxlen)
1023 {
1024         u32 osize = le32_to_cpu(layout->fl_object_size);
1025         u32 su = le32_to_cpu(layout->fl_stripe_unit);
1026         u32 sc = le32_to_cpu(layout->fl_stripe_count);
1027         u32 bl, stripeno, stripepos, objsetno;
1028         u32 su_per_object;
1029         u64 t, su_offset;
1030
1031         dout("mapping %llu~%llu  osize %u fl_su %u\n", off, len,
1032              osize, su);
1033         if (su == 0 || sc == 0)
1034                 goto invalid;
1035         su_per_object = osize / su;
1036         if (su_per_object == 0)
1037                 goto invalid;
1038         dout("osize %u / su %u = su_per_object %u\n", osize, su,
1039              su_per_object);
1040
1041         if ((su & ~PAGE_MASK) != 0)
1042                 goto invalid;
1043
1044         /* bl = *off / su; */
1045         t = off;
1046         do_div(t, su);
1047         bl = t;
1048         dout("off %llu / su %u = bl %u\n", off, su, bl);
1049
1050         stripeno = bl / sc;
1051         stripepos = bl % sc;
1052         objsetno = stripeno / su_per_object;
1053
1054         *ono = objsetno * sc + stripepos;
1055         dout("objset %u * sc %u = ono %u\n", objsetno, sc, (unsigned int)*ono);
1056
1057         /* *oxoff = *off % layout->fl_stripe_unit;  # offset in su */
1058         t = off;
1059         su_offset = do_div(t, su);
1060         *oxoff = su_offset + (stripeno % su_per_object) * su;
1061
1062         /*
1063          * Calculate the length of the extent being written to the selected
1064          * object. This is the minimum of the full length requested (len) or
1065          * the remainder of the current stripe being written to.
1066          */
1067         *oxlen = min_t(u64, len, su - su_offset);
1068
1069         dout(" obj extent %llu~%llu\n", *oxoff, *oxlen);
1070         return 0;
1071
1072 invalid:
1073         dout(" invalid layout\n");
1074         *ono = 0;
1075         *oxoff = 0;
1076         *oxlen = 0;
1077         return -EINVAL;
1078 }
1079 EXPORT_SYMBOL(ceph_calc_file_object_mapping);
1080
1081 /*
1082  * calculate an object layout (i.e. pgid) from an oid,
1083  * file_layout, and osdmap
1084  */
1085 int ceph_calc_object_layout(struct ceph_object_layout *ol,
1086                             const char *oid,
1087                             struct ceph_file_layout *fl,
1088                             struct ceph_osdmap *osdmap)
1089 {
1090         unsigned int num, num_mask;
1091         struct ceph_pg pgid;
1092         struct ceph_pg_pool_info *pool;
1093
1094         BUG_ON(!osdmap);
1095
1096         pgid.pool = le32_to_cpu(fl->fl_pg_pool);
1097         pool = __lookup_pg_pool(&osdmap->pg_pools, pgid.pool);
1098         if (!pool)
1099                 return -EIO;
1100         pgid.seed = ceph_str_hash(pool->v.object_hash, oid, strlen(oid));
1101         num = le32_to_cpu(pool->v.pg_num);
1102         num_mask = pool->pg_num_mask;
1103
1104         dout("calc_object_layout '%s' pgid %lld.%x\n", oid, pgid.pool,
1105              pgid.seed);
1106
1107         ol->ol_pgid.ps = cpu_to_le16(pgid.seed);
1108         ol->ol_pgid.pool = fl->fl_pg_pool;
1109         ol->ol_pgid.preferred = cpu_to_le16(-1);
1110         ol->ol_stripe_unit = fl->fl_object_stripe_unit;
1111         return 0;
1112 }
1113 EXPORT_SYMBOL(ceph_calc_object_layout);
1114
1115 /*
1116  * Calculate raw osd vector for the given pgid.  Return pointer to osd
1117  * array, or NULL on failure.
1118  */
1119 static int *calc_pg_raw(struct ceph_osdmap *osdmap, struct ceph_pg pgid,
1120                         int *osds, int *num)
1121 {
1122         struct ceph_pg_mapping *pg;
1123         struct ceph_pg_pool_info *pool;
1124         int ruleno;
1125         unsigned int poolid, ps, pps, t, r;
1126
1127         poolid = pgid.pool;
1128         ps = pgid.seed;
1129
1130         pool = __lookup_pg_pool(&osdmap->pg_pools, poolid);
1131         if (!pool)
1132                 return NULL;
1133
1134         /* pg_temp? */
1135         t = ceph_stable_mod(ps, le32_to_cpu(pool->v.pg_num),
1136                             pool->pgp_num_mask);
1137         pgid.seed = t;
1138         pg = __lookup_pg_mapping(&osdmap->pg_temp, pgid);
1139         if (pg) {
1140                 *num = pg->len;
1141                 return pg->osds;
1142         }
1143
1144         /* crush */
1145         ruleno = crush_find_rule(osdmap->crush, pool->v.crush_ruleset,
1146                                  pool->v.type, pool->v.size);
1147         if (ruleno < 0) {
1148                 pr_err("no crush rule pool %d ruleset %d type %d size %d\n",
1149                        poolid, pool->v.crush_ruleset, pool->v.type,
1150                        pool->v.size);
1151                 return NULL;
1152         }
1153
1154         pps = ceph_stable_mod(ps,
1155                               le32_to_cpu(pool->v.pgp_num),
1156                               pool->pgp_num_mask);
1157         pps += poolid;
1158         r = crush_do_rule(osdmap->crush, ruleno, pps, osds,
1159                           min_t(int, pool->v.size, *num),
1160                           osdmap->osd_weight);
1161         if (r < 0) {
1162                 pr_err("error %d from crush rule: pool %d ruleset %d type %d"
1163                        " size %d\n", r, poolid, pool->v.crush_ruleset,
1164                        pool->v.type, pool->v.size);
1165                 return NULL;
1166         }
1167         *num = r;
1168         return osds;
1169 }
1170
1171 /*
1172  * Return acting set for given pgid.
1173  */
1174 int ceph_calc_pg_acting(struct ceph_osdmap *osdmap, struct ceph_pg pgid,
1175                         int *acting)
1176 {
1177         int rawosds[CEPH_PG_MAX_SIZE], *osds;
1178         int i, o, num = CEPH_PG_MAX_SIZE;
1179
1180         osds = calc_pg_raw(osdmap, pgid, rawosds, &num);
1181         if (!osds)
1182                 return -1;
1183
1184         /* primary is first up osd */
1185         o = 0;
1186         for (i = 0; i < num; i++)
1187                 if (ceph_osd_is_up(osdmap, osds[i]))
1188                         acting[o++] = osds[i];
1189         return o;
1190 }
1191
1192 /*
1193  * Return primary osd for given pgid, or -1 if none.
1194  */
1195 int ceph_calc_pg_primary(struct ceph_osdmap *osdmap, struct ceph_pg pgid)
1196 {
1197         int rawosds[CEPH_PG_MAX_SIZE], *osds;
1198         int i, num = CEPH_PG_MAX_SIZE;
1199
1200         osds = calc_pg_raw(osdmap, pgid, rawosds, &num);
1201         if (!osds)
1202                 return -1;
1203
1204         /* primary is first up osd */
1205         for (i = 0; i < num; i++)
1206                 if (ceph_osd_is_up(osdmap, osds[i]))
1207                         return osds[i];
1208         return -1;
1209 }
1210 EXPORT_SYMBOL(ceph_calc_pg_primary);