fs/ceph/osdmap.c

   1
   2 #include <asm/div64.h>
   3
   4 #include "super.h"
   5 #include "osdmap.h"
   6 #include "crush/hash.h"
   7 #include "crush/mapper.h"
   8 #include "decode.h"
   9 #include "ceph_debug.h"
  10
  11 char *ceph_osdmap_state_str(char *str, int len, int state)
  12 {
  13         int flag = 0;
  14
  15         if (!len)
  16                 goto done;
  17
  18         *str = '\0';
  19         if (state) {
  20                 if (state & CEPH_OSD_EXISTS) {
  21                         snprintf(str, len, "exists");
  22                         flag = 1;
  23                 }
  24                 if (state & CEPH_OSD_UP) {
  25                         snprintf(str, len, "%s%s%s", str, (flag ? ", " : ""),
  26                                  "up");
  27                         flag = 1;
  28                 }
  29         } else {
  30                 snprintf(str, len, "doesn't exist");
  31         }
  32 done:
  33         return str;
  34 }
  35
  36 /* maps */
  37
  38 static int calc_bits_of(unsigned t)
  39 {
  40         int b = 0;
  41         while (t) {
  42                 t = t >> 1;
  43                 b++;
  44         }
  45         return b;
  46 }
  47
  48 /*
  49  * the foo_mask is the smallest value 2^n-1 that is >= foo.
  50  */
  51 static void calc_pg_masks(struct ceph_pg_pool_info *pi)
  52 {
  53         pi->pg_num_mask = (1 << calc_bits_of(le32_to_cpu(pi->v.pg_num)-1)) - 1;
  54         pi->pgp_num_mask =
  55                 (1 << calc_bits_of(le32_to_cpu(pi->v.pgp_num)-1)) - 1;
  56         pi->lpg_num_mask =
  57                 (1 << calc_bits_of(le32_to_cpu(pi->v.lpg_num)-1)) - 1;
  58         pi->lpgp_num_mask =
  59                 (1 << calc_bits_of(le32_to_cpu(pi->v.lpgp_num)-1)) - 1;
  60 }
  61
  62 /*
  63  * decode crush map
  64  */
  65 static int crush_decode_uniform_bucket(void **p, void *end,
  66                                        struct crush_bucket_uniform *b)
  67 {
  68         dout("crush_decode_uniform_bucket %p to %p\n", *p, end);
  69         ceph_decode_need(p, end, (1+b->h.size) * sizeof(u32), bad);
  70         b->item_weight = ceph_decode_32(p);
  71         return 0;
  72 bad:
  73         return -EINVAL;
  74 }
  75
  76 static int crush_decode_list_bucket(void **p, void *end,
  77                                     struct crush_bucket_list *b)
  78 {
  79         int j;
  80         dout("crush_decode_list_bucket %p to %p\n", *p, end);
  81         b->item_weights = kcalloc(b->h.size, sizeof(u32), GFP_NOFS);
  82         if (b->item_weights == NULL)
  83                 return -ENOMEM;
  84         b->sum_weights = kcalloc(b->h.size, sizeof(u32), GFP_NOFS);
  85         if (b->sum_weights == NULL)
  86                 return -ENOMEM;
  87         ceph_decode_need(p, end, 2 * b->h.size * sizeof(u32), bad);
  88         for (j = 0; j < b->h.size; j++) {
  89                 b->item_weights[j] = ceph_decode_32(p);
  90                 b->sum_weights[j] = ceph_decode_32(p);
  91         }
  92         return 0;
  93 bad:
  94         return -EINVAL;
  95 }
  96
  97 static int crush_decode_tree_bucket(void **p, void *end,
  98                                     struct crush_bucket_tree *b)
  99 {
 100         int j;
 101         dout("crush_decode_tree_bucket %p to %p\n", *p, end);
 102         ceph_decode_32_safe(p, end, b->num_nodes, bad);
 103         b->node_weights = kcalloc(b->num_nodes, sizeof(u32), GFP_NOFS);
 104         if (b->node_weights == NULL)
 105                 return -ENOMEM;
 106         ceph_decode_need(p, end, b->num_nodes * sizeof(u32), bad);
 107         for (j = 0; j < b->num_nodes; j++)
 108                 b->node_weights[j] = ceph_decode_32(p);
 109         return 0;
 110 bad:
 111         return -EINVAL;
 112 }
 113
 114 static int crush_decode_straw_bucket(void **p, void *end,
 115                                      struct crush_bucket_straw *b)
 116 {
 117         int j;
 118         dout("crush_decode_straw_bucket %p to %p\n", *p, end);
 119         b->item_weights = kcalloc(b->h.size, sizeof(u32), GFP_NOFS);
 120         if (b->item_weights == NULL)
 121                 return -ENOMEM;
 122         b->straws = kcalloc(b->h.size, sizeof(u32), GFP_NOFS);
 123         if (b->straws == NULL)
 124                 return -ENOMEM;
 125         ceph_decode_need(p, end, 2 * b->h.size * sizeof(u32), bad);
 126         for (j = 0; j < b->h.size; j++) {
 127                 b->item_weights[j] = ceph_decode_32(p);
 128                 b->straws[j] = ceph_decode_32(p);
 129         }
 130         return 0;
 131 bad:
 132         return -EINVAL;
 133 }
 134
 135 static struct crush_map *crush_decode(void *pbyval, void *end)
 136 {
 137         struct crush_map *c;
 138         int err = -EINVAL;
 139         int i, j;
 140         void **p = &pbyval;
 141         void *start = pbyval;
 142         u32 magic;
 143
 144         dout("crush_decode %p to %p len %d\n", *p, end, (int)(end - *p));
 145
 146         c = kzalloc(sizeof(*c), GFP_NOFS);
 147         if (c == NULL)
 148                 return ERR_PTR(-ENOMEM);
 149
 150         ceph_decode_need(p, end, 4*sizeof(u32), bad);
 151         magic = ceph_decode_32(p);
 152         if (magic != CRUSH_MAGIC) {
 153                 pr_err("crush_decode magic %x != current %x\n",
 154                        (unsigned)magic, (unsigned)CRUSH_MAGIC);
 155                 goto bad;
 156         }
 157         c->max_buckets = ceph_decode_32(p);
 158         c->max_rules = ceph_decode_32(p);
 159         c->max_devices = ceph_decode_32(p);
 160
 161         c->device_parents = kcalloc(c->max_devices, sizeof(u32), GFP_NOFS);
 162         if (c->device_parents == NULL)
 163                 goto badmem;
 164         c->bucket_parents = kcalloc(c->max_buckets, sizeof(u32), GFP_NOFS);
 165         if (c->bucket_parents == NULL)
 166                 goto badmem;
 167
 168         c->buckets = kcalloc(c->max_buckets, sizeof(*c->buckets), GFP_NOFS);
 169         if (c->buckets == NULL)
 170                 goto badmem;
 171         c->rules = kcalloc(c->max_rules, sizeof(*c->rules), GFP_NOFS);
 172         if (c->rules == NULL)
 173                 goto badmem;
 174
 175         /* buckets */
 176         for (i = 0; i < c->max_buckets; i++) {
 177                 int size = 0;
 178                 u32 alg;
 179                 struct crush_bucket *b;
 180
 181                 ceph_decode_32_safe(p, end, alg, bad);
 182                 if (alg == 0) {
 183                         c->buckets[i] = NULL;
 184                         continue;
 185                 }
 186                 dout("crush_decode bucket %d off %x %p to %p\n",
 187                      i, (int)(*p-start), *p, end);
 188
 189                 switch (alg) {
 190                 case CRUSH_BUCKET_UNIFORM:
 191                         size = sizeof(struct crush_bucket_uniform);
 192                         break;
 193                 case CRUSH_BUCKET_LIST:
 194                         size = sizeof(struct crush_bucket_list);
 195                         break;
 196                 case CRUSH_BUCKET_TREE:
 197                         size = sizeof(struct crush_bucket_tree);
 198                         break;
 199                 case CRUSH_BUCKET_STRAW:
 200                         size = sizeof(struct crush_bucket_straw);
 201                         break;
 202                 default:
 203                         err = -EINVAL;
 204                         goto bad;
 205                 }
 206                 BUG_ON(size == 0);
 207                 b = c->buckets[i] = kzalloc(size, GFP_NOFS);
 208                 if (b == NULL)
 209                         goto badmem;
 210
 211                 ceph_decode_need(p, end, 4*sizeof(u32), bad);
 212                 b->id = ceph_decode_32(p);
 213                 b->type = ceph_decode_16(p);
 214                 b->alg = ceph_decode_8(p);
 215                 b->hash = ceph_decode_8(p);
 216                 b->weight = ceph_decode_32(p);
 217                 b->size = ceph_decode_32(p);
 218
 219                 dout("crush_decode bucket size %d off %x %p to %p\n",
 220                      b->size, (int)(*p-start), *p, end);
 221
 222                 b->items = kcalloc(b->size, sizeof(__s32), GFP_NOFS);
 223                 if (b->items == NULL)
 224                         goto badmem;
 225                 b->perm = kcalloc(b->size, sizeof(u32), GFP_NOFS);
 226                 if (b->perm == NULL)
 227                         goto badmem;
 228                 b->perm_n = 0;
 229
 230                 ceph_decode_need(p, end, b->size*sizeof(u32), bad);
 231                 for (j = 0; j < b->size; j++)
 232                         b->items[j] = ceph_decode_32(p);
 233
 234                 switch (b->alg) {
 235                 case CRUSH_BUCKET_UNIFORM:
 236                         err = crush_decode_uniform_bucket(p, end,
 237                                   (struct crush_bucket_uniform *)b);
 238                         if (err < 0)
 239                                 goto bad;
 240                         break;
 241                 case CRUSH_BUCKET_LIST:
 242                         err = crush_decode_list_bucket(p, end,
 243                                (struct crush_bucket_list *)b);
 244                         if (err < 0)
 245                                 goto bad;
 246                         break;
 247                 case CRUSH_BUCKET_TREE:
 248                         err = crush_decode_tree_bucket(p, end,
 249                                 (struct crush_bucket_tree *)b);
 250                         if (err < 0)
 251                                 goto bad;
 252                         break;
 253                 case CRUSH_BUCKET_STRAW:
 254                         err = crush_decode_straw_bucket(p, end,
 255                                 (struct crush_bucket_straw *)b);
 256                         if (err < 0)
 257                                 goto bad;
 258                         break;
 259                 }
 260         }
 261
 262         /* rules */
 263         dout("rule vec is %p\n", c->rules);
 264         for (i = 0; i < c->max_rules; i++) {
 265                 u32 yes;
 266                 struct crush_rule *r;
 267
 268                 ceph_decode_32_safe(p, end, yes, bad);
 269                 if (!yes) {
 270                         dout("crush_decode NO rule %d off %x %p to %p\n",
 271                              i, (int)(*p-start), *p, end);
 272                         c->rules[i] = NULL;
 273                         continue;
 274                 }
 275
 276                 dout("crush_decode rule %d off %x %p to %p\n",
 277                      i, (int)(*p-start), *p, end);
 278
 279                 /* len */
 280                 ceph_decode_32_safe(p, end, yes, bad);
 281 #if BITS_PER_LONG == 32
 282                 err = -EINVAL;
 283                 if (yes > ULONG_MAX / sizeof(struct crush_rule_step))
 284                         goto bad;
 285 #endif
 286                 r = c->rules[i] = kmalloc(sizeof(*r) +
 287                                           yes*sizeof(struct crush_rule_step),
 288                                           GFP_NOFS);
 289                 if (r == NULL)
 290                         goto badmem;
 291                 dout(" rule %d is at %p\n", i, r);
 292                 r->len = yes;
 293                 ceph_decode_copy_safe(p, end, &r->mask, 4, bad); /* 4 u8's */
 294                 ceph_decode_need(p, end, r->len*3*sizeof(u32), bad);
 295                 for (j = 0; j < r->len; j++) {
 296                         r->steps[j].op = ceph_decode_32(p);
 297                         r->steps[j].arg1 = ceph_decode_32(p);
 298                         r->steps[j].arg2 = ceph_decode_32(p);
 299                 }
 300         }
 301
 302         /* ignore trailing name maps. */
 303
 304         dout("crush_decode success\n");
 305         return c;
 306
 307 badmem:
 308         err = -ENOMEM;
 309 bad:
 310         dout("crush_decode fail %d\n", err);
 311         crush_destroy(c);
 312         return ERR_PTR(err);
 313 }
 314
 315 /*
 316  * rbtree of pg_mapping for handling pg_temp (explicit mapping of pgid
 317  * to a set of osds)
 318  */
 319 static int pgid_cmp(struct ceph_pg l, struct ceph_pg r)
 320 {
 321         u64 a = *(u64 *)&l;
 322         u64 b = *(u64 *)&r;
 323
 324         if (a < b)
 325                 return -1;
 326         if (a > b)
 327                 return 1;
 328         return 0;
 329 }
 330
 331 static int __insert_pg_mapping(struct ceph_pg_mapping *new,
 332                                struct rb_root *root)
 333 {
 334         struct rb_node **p = &root->rb_node;
 335         struct rb_node *parent = NULL;
 336         struct ceph_pg_mapping *pg = NULL;
 337         int c;
 338
 339         while (*p) {
 340                 parent = *p;
 341                 pg = rb_entry(parent, struct ceph_pg_mapping, node);
 342                 c = pgid_cmp(new->pgid, pg->pgid);
 343                 if (c < 0)
 344                         p = &(*p)->rb_left;
 345                 else if (c > 0)
 346                         p = &(*p)->rb_right;
 347                 else
 348                         return -EEXIST;
 349         }
 350
 351         rb_link_node(&new->node, parent, p);
 352         rb_insert_color(&new->node, root);
 353         return 0;
 354 }
 355
 356 static struct ceph_pg_mapping *__lookup_pg_mapping(struct rb_root *root,
 357                                                    struct ceph_pg pgid)
 358 {
 359         struct rb_node *n = root->rb_node;
 360         struct ceph_pg_mapping *pg;
 361         int c;
 362
 363         while (n) {
 364                 pg = rb_entry(n, struct ceph_pg_mapping, node);
 365                 c = pgid_cmp(pgid, pg->pgid);
 366                 if (c < 0)
 367                         n = n->rb_left;
 368                 else if (c > 0)
 369                         n = n->rb_right;
 370                 else
 371                         return pg;
 372         }
 373         return NULL;
 374 }
 375
 376 /*
 377  * rbtree of pg pool info
 378  */
 379 static int __insert_pg_pool(struct rb_root *root, struct ceph_pg_pool_info *new)
 380 {
 381         struct rb_node **p = &root->rb_node;
 382         struct rb_node *parent = NULL;
 383         struct ceph_pg_pool_info *pi = NULL;
 384
 385         while (*p) {
 386                 parent = *p;
 387                 pi = rb_entry(parent, struct ceph_pg_pool_info, node);
 388                 if (new->id < pi->id)
 389                         p = &(*p)->rb_left;
 390                 else if (new->id > pi->id)
 391                         p = &(*p)->rb_right;
 392                 else
 393                         return -EEXIST;
 394         }
 395
 396         rb_link_node(&new->node, parent, p);
 397         rb_insert_color(&new->node, root);
 398         return 0;
 399 }
 400
 401 static struct ceph_pg_pool_info *__lookup_pg_pool(struct rb_root *root, int id)
 402 {
 403         struct ceph_pg_pool_info *pi;
 404         struct rb_node *n = root->rb_node;
 405
 406         while (n) {
 407                 pi = rb_entry(n, struct ceph_pg_pool_info, node);
 408                 if (id < pi->id)
 409                         n = n->rb_left;
 410                 else if (id > pi->id)
 411                         n = n->rb_right;
 412                 else
 413                         return pi;
 414         }
 415         return NULL;
 416 }
 417
 418 static void __remove_pg_pool(struct rb_root *root, struct ceph_pg_pool_info *pi)
 419 {
 420         rb_erase(&pi->node, root);
 421         kfree(pi->name);
 422         kfree(pi);
 423 }
 424
 425 void __decode_pool(void **p, struct ceph_pg_pool_info *pi)
 426 {
 427         ceph_decode_copy(p, &pi->v, sizeof(pi->v));
 428         calc_pg_masks(pi);
 429         *p += le32_to_cpu(pi->v.num_snaps) * sizeof(u64);
 430         *p += le32_to_cpu(pi->v.num_removed_snap_intervals) * sizeof(u64) * 2;
 431 }
 432
 433 static int __decode_pool_names(void **p, void *end, struct ceph_osdmap *map)
 434 {
 435         struct ceph_pg_pool_info *pi;
 436         u32 num, len, pool;
 437
 438         ceph_decode_32_safe(p, end, num, bad);
 439         dout(" %d pool names\n", num);
 440         while (num--) {
 441                 ceph_decode_32_safe(p, end, pool, bad);
 442                 ceph_decode_32_safe(p, end, len, bad);
 443                 dout("  pool %d len %d\n", pool, len);
 444                 pi = __lookup_pg_pool(&map->pg_pools, pool);
 445                 if (pi) {
 446                         kfree(pi->name);
 447                         pi->name = kmalloc(len + 1, GFP_NOFS);
 448                         if (pi->name) {
 449                                 memcpy(pi->name, *p, len);
 450                                 pi->name[len] = '\0';
 451                                 dout("  name is %s\n", pi->name);
 452                         }
 453                 }
 454                 *p += len;
 455         }
 456         return 0;
 457
 458 bad:
 459         return -EINVAL;
 460 }
 461
 462 /*
 463  * osd map
 464  */
 465 void ceph_osdmap_destroy(struct ceph_osdmap *map)
 466 {
 467         dout("osdmap_destroy %p\n", map);
 468         if (map->crush)
 469                 crush_destroy(map->crush);
 470         while (!RB_EMPTY_ROOT(&map->pg_temp)) {
 471                 struct ceph_pg_mapping *pg =
 472                         rb_entry(rb_first(&map->pg_temp),
 473                                  struct ceph_pg_mapping, node);
 474                 rb_erase(&pg->node, &map->pg_temp);
 475                 kfree(pg);
 476         }
 477         while (!RB_EMPTY_ROOT(&map->pg_pools)) {
 478                 struct ceph_pg_pool_info *pi =
 479                         rb_entry(rb_first(&map->pg_pools),
 480                                  struct ceph_pg_pool_info, node);
 481                 __remove_pg_pool(&map->pg_pools, pi);
 482         }
 483         kfree(map->osd_state);
 484         kfree(map->osd_weight);
 485         kfree(map->osd_addr);
 486         kfree(map);
 487 }
 488
 489 /*
 490  * adjust max osd value.  reallocate arrays.
 491  */
 492 static int osdmap_set_max_osd(struct ceph_osdmap *map, int max)
 493 {
 494         u8 *state;
 495         struct ceph_entity_addr *addr;
 496         u32 *weight;
 497
 498         state = kcalloc(max, sizeof(*state), GFP_NOFS);
 499         addr = kcalloc(max, sizeof(*addr), GFP_NOFS);
 500         weight = kcalloc(max, sizeof(*weight), GFP_NOFS);
 501         if (state == NULL || addr == NULL || weight == NULL) {
 502                 kfree(state);
 503                 kfree(addr);
 504                 kfree(weight);
 505                 return -ENOMEM;
 506         }
 507
 508         /* copy old? */
 509         if (map->osd_state) {
 510                 memcpy(state, map->osd_state, map->max_osd*sizeof(*state));
 511                 memcpy(addr, map->osd_addr, map->max_osd*sizeof(*addr));
 512                 memcpy(weight, map->osd_weight, map->max_osd*sizeof(*weight));
 513                 kfree(map->osd_state);
 514                 kfree(map->osd_addr);
 515                 kfree(map->osd_weight);
 516         }
 517
 518         map->osd_state = state;
 519         map->osd_weight = weight;
 520         map->osd_addr = addr;
 521         map->max_osd = max;
 522         return 0;
 523 }
 524
 525 /*
 526  * decode a full map.
 527  */
 528 struct ceph_osdmap *osdmap_decode(void **p, void *end)
 529 {
 530         struct ceph_osdmap *map;
 531         u16 version;
 532         u32 len, max, i;
 533         u8 ev;
 534         int err = -EINVAL;
 535         void *start = *p;
 536         struct ceph_pg_pool_info *pi;
 537
 538         dout("osdmap_decode %p to %p len %d\n", *p, end, (int)(end - *p));
 539
 540         map = kzalloc(sizeof(*map), GFP_NOFS);
 541         if (map == NULL)
 542                 return ERR_PTR(-ENOMEM);
 543         map->pg_temp = RB_ROOT;
 544
 545         ceph_decode_16_safe(p, end, version, bad);
 546         if (version > CEPH_OSDMAP_VERSION) {
 547                 pr_warning("got unknown v %d > %d of osdmap\n", version,
 548                            CEPH_OSDMAP_VERSION);
 549                 goto bad;
 550         }
 551
 552         ceph_decode_need(p, end, 2*sizeof(u64)+6*sizeof(u32), bad);
 553         ceph_decode_copy(p, &map->fsid, sizeof(map->fsid));
 554         map->epoch = ceph_decode_32(p);
 555         ceph_decode_copy(p, &map->created, sizeof(map->created));
 556         ceph_decode_copy(p, &map->modified, sizeof(map->modified));
 557
 558         ceph_decode_32_safe(p, end, max, bad);
 559         while (max--) {
 560                 ceph_decode_need(p, end, 4 + 1 + sizeof(pi->v), bad);
 561                 pi = kzalloc(sizeof(*pi), GFP_NOFS);
 562                 if (!pi)
 563                         goto bad;
 564                 pi->id = ceph_decode_32(p);
 565                 ev = ceph_decode_8(p); /* encoding version */
 566                 if (ev > CEPH_PG_POOL_VERSION) {
 567                         pr_warning("got unknown v %d > %d of ceph_pg_pool\n",
 568                                    ev, CEPH_PG_POOL_VERSION);
 569                         goto bad;
 570                 }
 571                 __decode_pool(p, pi);
 572                 __insert_pg_pool(&map->pg_pools, pi);
 573         }
 574
 575         if (version >= 5 && __decode_pool_names(p, end, map) < 0)
 576                 goto bad;
 577
 578         ceph_decode_32_safe(p, end, map->pool_max, bad);
 579
 580         ceph_decode_32_safe(p, end, map->flags, bad);
 581
 582         max = ceph_decode_32(p);
 583
 584         /* (re)alloc osd arrays */
 585         err = osdmap_set_max_osd(map, max);
 586         if (err < 0)
 587                 goto bad;
 588         dout("osdmap_decode max_osd = %d\n", map->max_osd);
 589
 590         /* osds */
 591         err = -EINVAL;
 592         ceph_decode_need(p, end, 3*sizeof(u32) +
 593                          map->max_osd*(1 + sizeof(*map->osd_weight) +
 594                                        sizeof(*map->osd_addr)), bad);
 595         *p += 4; /* skip length field (should match max) */
 596         ceph_decode_copy(p, map->osd_state, map->max_osd);
 597
 598         *p += 4; /* skip length field (should match max) */
 599         for (i = 0; i < map->max_osd; i++)
 600                 map->osd_weight[i] = ceph_decode_32(p);
 601
 602         *p += 4; /* skip length field (should match max) */
 603         ceph_decode_copy(p, map->osd_addr, map->max_osd*sizeof(*map->osd_addr));
 604         for (i = 0; i < map->max_osd; i++)
 605                 ceph_decode_addr(&map->osd_addr[i]);
 606
 607         /* pg_temp */
 608         ceph_decode_32_safe(p, end, len, bad);
 609         for (i = 0; i < len; i++) {
 610                 int n, j;
 611                 struct ceph_pg pgid;
 612                 struct ceph_pg_mapping *pg;
 613
 614                 ceph_decode_need(p, end, sizeof(u32) + sizeof(u64), bad);
 615                 ceph_decode_copy(p, &pgid, sizeof(pgid));
 616                 n = ceph_decode_32(p);
 617                 ceph_decode_need(p, end, n * sizeof(u32), bad);
 618                 err = -ENOMEM;
 619                 pg = kmalloc(sizeof(*pg) + n*sizeof(u32), GFP_NOFS);
 620                 if (!pg)
 621                         goto bad;
 622                 pg->pgid = pgid;
 623                 pg->len = n;
 624                 for (j = 0; j < n; j++)
 625                         pg->osds[j] = ceph_decode_32(p);
 626
 627                 err = __insert_pg_mapping(pg, &map->pg_temp);
 628                 if (err)
 629                         goto bad;
 630                 dout(" added pg_temp %llx len %d\n", *(u64 *)&pgid, len);
 631         }
 632
 633         /* crush */
 634         ceph_decode_32_safe(p, end, len, bad);
 635         dout("osdmap_decode crush len %d from off 0x%x\n", len,
 636              (int)(*p - start));
 637         ceph_decode_need(p, end, len, bad);
 638         map->crush = crush_decode(*p, end);
 639         *p += len;
 640         if (IS_ERR(map->crush)) {
 641                 err = PTR_ERR(map->crush);
 642                 map->crush = NULL;
 643                 goto bad;
 644         }
 645
 646         /* ignore the rest of the map */
 647         *p = end;
 648
 649         dout("osdmap_decode done %p %p\n", *p, end);
 650         return map;
 651
 652 bad:
 653         dout("osdmap_decode fail\n");
 654         ceph_osdmap_destroy(map);
 655         return ERR_PTR(err);
 656 }
 657
 658 /*
 659  * decode and apply an incremental map update.
 660  */
 661 struct ceph_osdmap *osdmap_apply_incremental(void **p, void *end,
 662                                              struct ceph_osdmap *map,
 663                                              struct ceph_messenger *msgr)
 664 {
 665         struct crush_map *newcrush = NULL;
 666         struct ceph_fsid fsid;
 667         u32 epoch = 0;
 668         struct ceph_timespec modified;
 669         u32 len, pool;
 670         __s32 new_pool_max, new_flags, max;
 671         void *start = *p;
 672         int err = -EINVAL;
 673         u16 version;
 674         struct rb_node *rbp;
 675
 676         ceph_decode_16_safe(p, end, version, bad);
 677         if (version > CEPH_OSDMAP_INC_VERSION) {
 678                 pr_warning("got unknown v %d > %d of inc osdmap\n", version,
 679                            CEPH_OSDMAP_INC_VERSION);
 680                 goto bad;
 681         }
 682
 683         ceph_decode_need(p, end, sizeof(fsid)+sizeof(modified)+2*sizeof(u32),
 684                          bad);
 685         ceph_decode_copy(p, &fsid, sizeof(fsid));
 686         epoch = ceph_decode_32(p);
 687         BUG_ON(epoch != map->epoch+1);
 688         ceph_decode_copy(p, &modified, sizeof(modified));
 689         new_pool_max = ceph_decode_32(p);
 690         new_flags = ceph_decode_32(p);
 691
 692         /* full map? */
 693         ceph_decode_32_safe(p, end, len, bad);
 694         if (len > 0) {
 695                 dout("apply_incremental full map len %d, %p to %p\n",
 696                      len, *p, end);
 697                 return osdmap_decode(p, min(*p+len, end));
 698         }
 699
 700         /* new crush? */
 701         ceph_decode_32_safe(p, end, len, bad);
 702         if (len > 0) {
 703                 dout("apply_incremental new crush map len %d, %p to %p\n",
 704                      len, *p, end);
 705                 newcrush = crush_decode(*p, min(*p+len, end));
 706                 if (IS_ERR(newcrush))
 707                         return ERR_PTR(PTR_ERR(newcrush));
 708         }
 709
 710         /* new flags? */
 711         if (new_flags >= 0)
 712                 map->flags = new_flags;
 713         if (new_pool_max >= 0)
 714                 map->pool_max = new_pool_max;
 715
 716         ceph_decode_need(p, end, 5*sizeof(u32), bad);
 717
 718         /* new max? */
 719         max = ceph_decode_32(p);
 720         if (max >= 0) {
 721                 err = osdmap_set_max_osd(map, max);
 722                 if (err < 0)
 723                         goto bad;
 724         }
 725
 726         map->epoch++;
 727         map->modified = map->modified;
 728         if (newcrush) {
 729                 if (map->crush)
 730                         crush_destroy(map->crush);
 731                 map->crush = newcrush;
 732                 newcrush = NULL;
 733         }
 734
 735         /* new_pool */
 736         ceph_decode_32_safe(p, end, len, bad);
 737         while (len--) {
 738                 __u8 ev;
 739                 struct ceph_pg_pool_info *pi;
 740
 741                 ceph_decode_32_safe(p, end, pool, bad);
 742                 ceph_decode_need(p, end, 1 + sizeof(pi->v), bad);
 743                 ev = ceph_decode_8(p);  /* encoding version */
 744                 if (ev > CEPH_PG_POOL_VERSION) {
 745                         pr_warning("got unknown v %d > %d of ceph_pg_pool\n",
 746                                    ev, CEPH_PG_POOL_VERSION);
 747                         goto bad;
 748                 }
 749                 pi = __lookup_pg_pool(&map->pg_pools, pool);
 750                 if (!pi) {
 751                         pi = kzalloc(sizeof(*pi), GFP_NOFS);
 752                         if (!pi) {
 753                                 err = -ENOMEM;
 754                                 goto bad;
 755                         }
 756                         pi->id = pool;
 757                         __insert_pg_pool(&map->pg_pools, pi);
 758                 }
 759                 __decode_pool(p, pi);
 760         }
 761         if (version >= 5 && __decode_pool_names(p, end, map) < 0)
 762                 goto bad;
 763
 764         /* old_pool */
 765         ceph_decode_32_safe(p, end, len, bad);
 766         while (len--) {
 767                 struct ceph_pg_pool_info *pi;
 768
 769                 ceph_decode_32_safe(p, end, pool, bad);
 770                 pi = __lookup_pg_pool(&map->pg_pools, pool);
 771                 if (pi)
 772                         __remove_pg_pool(&map->pg_pools, pi);
 773         }
 774
 775         /* new_up */
 776         err = -EINVAL;
 777         ceph_decode_32_safe(p, end, len, bad);
 778         while (len--) {
 779                 u32 osd;
 780                 struct ceph_entity_addr addr;
 781                 ceph_decode_32_safe(p, end, osd, bad);
 782                 ceph_decode_copy_safe(p, end, &addr, sizeof(addr), bad);
 783                 ceph_decode_addr(&addr);
 784                 pr_info("osd%d up\n", osd);
 785                 BUG_ON(osd >= map->max_osd);
 786                 map->osd_state[osd] |= CEPH_OSD_UP;
 787                 map->osd_addr[osd] = addr;
 788         }
 789
 790         /* new_down */
 791         ceph_decode_32_safe(p, end, len, bad);
 792         while (len--) {
 793                 u32 osd;
 794                 ceph_decode_32_safe(p, end, osd, bad);
 795                 (*p)++;  /* clean flag */
 796                 pr_info("osd%d down\n", osd);
 797                 if (osd < map->max_osd)
 798                         map->osd_state[osd] &= ~CEPH_OSD_UP;
 799         }
 800
 801         /* new_weight */
 802         ceph_decode_32_safe(p, end, len, bad);
 803         while (len--) {
 804                 u32 osd, off;
 805                 ceph_decode_need(p, end, sizeof(u32)*2, bad);
 806                 osd = ceph_decode_32(p);
 807                 off = ceph_decode_32(p);
 808                 pr_info("osd%d weight 0x%x %s\n", osd, off,
 809                      off == CEPH_OSD_IN ? "(in)" :
 810                      (off == CEPH_OSD_OUT ? "(out)" : ""));
 811                 if (osd < map->max_osd)
 812                         map->osd_weight[osd] = off;
 813         }
 814
 815         /* new_pg_temp */
 816         rbp = rb_first(&map->pg_temp);
 817         ceph_decode_32_safe(p, end, len, bad);
 818         while (len--) {
 819                 struct ceph_pg_mapping *pg;
 820                 int j;
 821                 struct ceph_pg pgid;
 822                 u32 pglen;
 823                 ceph_decode_need(p, end, sizeof(u64) + sizeof(u32), bad);
 824                 ceph_decode_copy(p, &pgid, sizeof(pgid));
 825                 pglen = ceph_decode_32(p);
 826
 827                 /* remove any? */
 828                 while (rbp && pgid_cmp(rb_entry(rbp, struct ceph_pg_mapping,
 829                                                 node)->pgid, pgid) <= 0) {
 830                         struct rb_node *cur = rbp;
 831                         rbp = rb_next(rbp);
 832                         dout(" removed pg_temp %llx\n",
 833                              *(u64 *)&rb_entry(cur, struct ceph_pg_mapping,
 834                                                node)->pgid);
 835                         rb_erase(cur, &map->pg_temp);
 836                 }
 837
 838                 if (pglen) {
 839                         /* insert */
 840                         ceph_decode_need(p, end, pglen*sizeof(u32), bad);
 841                         pg = kmalloc(sizeof(*pg) + sizeof(u32)*pglen, GFP_NOFS);
 842                         if (!pg) {
 843                                 err = -ENOMEM;
 844                                 goto bad;
 845                         }
 846                         pg->pgid = pgid;
 847                         pg->len = pglen;
 848                         for (j = 0; j < pglen; j++)
 849                                 pg->osds[j] = ceph_decode_32(p);
 850                         err = __insert_pg_mapping(pg, &map->pg_temp);
 851                         if (err)
 852                                 goto bad;
 853                         dout(" added pg_temp %llx len %d\n", *(u64 *)&pgid,
 854                              pglen);
 855                 }
 856         }
 857         while (rbp) {
 858                 struct rb_node *cur = rbp;
 859                 rbp = rb_next(rbp);
 860                 dout(" removed pg_temp %llx\n",
 861                      *(u64 *)&rb_entry(cur, struct ceph_pg_mapping,
 862                                        node)->pgid);
 863                 rb_erase(cur, &map->pg_temp);
 864         }
 865
 866         /* ignore the rest */
 867         *p = end;
 868         return map;
 869
 870 bad:
 871         pr_err("corrupt inc osdmap epoch %d off %d (%p of %p-%p)\n",
 872                epoch, (int)(*p - start), *p, start, end);
 873         print_hex_dump(KERN_DEBUG, "osdmap: ",
 874                        DUMP_PREFIX_OFFSET, 16, 1,
 875                        start, end - start, true);
 876         if (newcrush)
 877                 crush_destroy(newcrush);
 878         return ERR_PTR(err);
 879 }
 880
 881
 882
 883
 884 /*
 885  * calculate file layout from given offset, length.
 886  * fill in correct oid, logical length, and object extent
 887  * offset, length.
 888  *
 889  * for now, we write only a single su, until we can
 890  * pass a stride back to the caller.
 891  */
 892 void ceph_calc_file_object_mapping(struct ceph_file_layout *layout,
 893                                    u64 off, u64 *plen,
 894                                    u64 *ono,
 895                                    u64 *oxoff, u64 *oxlen)
 896 {
 897         u32 osize = le32_to_cpu(layout->fl_object_size);
 898         u32 su = le32_to_cpu(layout->fl_stripe_unit);
 899         u32 sc = le32_to_cpu(layout->fl_stripe_count);
 900         u32 bl, stripeno, stripepos, objsetno;
 901         u32 su_per_object;
 902         u64 t, su_offset;
 903
 904         dout("mapping %llu~%llu  osize %u fl_su %u\n", off, *plen,
 905              osize, su);
 906         su_per_object = osize / su;
 907         dout("osize %u / su %u = su_per_object %u\n", osize, su,
 908              su_per_object);
 909
 910         BUG_ON((su & ~PAGE_MASK) != 0);
 911         /* bl = *off / su; */
 912         t = off;
 913         do_div(t, su);
 914         bl = t;
 915         dout("off %llu / su %u = bl %u\n", off, su, bl);
 916
 917         stripeno = bl / sc;
 918         stripepos = bl % sc;
 919         objsetno = stripeno / su_per_object;
 920
 921         *ono = objsetno * sc + stripepos;
 922         dout("objset %u * sc %u = ono %u\n", objsetno, sc, (unsigned)*ono);
 923
 924         /* *oxoff = *off % layout->fl_stripe_unit;  # offset in su */
 925         t = off;
 926         su_offset = do_div(t, su);
 927         *oxoff = su_offset + (stripeno % su_per_object) * su;
 928
 929         /*
 930          * Calculate the length of the extent being written to the selected
 931          * object. This is the minimum of the full length requested (plen) or
 932          * the remainder of the current stripe being written to.
 933          */
 934         *oxlen = min_t(u64, *plen, su - su_offset);
 935         *plen = *oxlen;
 936
 937         dout(" obj extent %llu~%llu\n", *oxoff, *oxlen);
 938 }
 939
 940 /*
 941  * calculate an object layout (i.e. pgid) from an oid,
 942  * file_layout, and osdmap
 943  */
 944 int ceph_calc_object_layout(struct ceph_object_layout *ol,
 945                             const char *oid,
 946                             struct ceph_file_layout *fl,
 947                             struct ceph_osdmap *osdmap)
 948 {
 949         unsigned num, num_mask;
 950         struct ceph_pg pgid;
 951         s32 preferred = (s32)le32_to_cpu(fl->fl_pg_preferred);
 952         int poolid = le32_to_cpu(fl->fl_pg_pool);
 953         struct ceph_pg_pool_info *pool;
 954         unsigned ps;
 955
 956         BUG_ON(!osdmap);
 957
 958         pool = __lookup_pg_pool(&osdmap->pg_pools, poolid);
 959         if (!pool)
 960                 return -EIO;
 961         ps = ceph_str_hash(pool->v.object_hash, oid, strlen(oid));
 962         if (preferred >= 0) {
 963                 ps += preferred;
 964                 num = le32_to_cpu(pool->v.lpg_num);
 965                 num_mask = pool->lpg_num_mask;
 966         } else {
 967                 num = le32_to_cpu(pool->v.pg_num);
 968                 num_mask = pool->pg_num_mask;
 969         }
 970
 971         pgid.ps = cpu_to_le16(ps);
 972         pgid.preferred = cpu_to_le16(preferred);
 973         pgid.pool = fl->fl_pg_pool;
 974         if (preferred >= 0)
 975                 dout("calc_object_layout '%s' pgid %d.%xp%d\n", oid, poolid, ps,
 976                      (int)preferred);
 977         else
 978                 dout("calc_object_layout '%s' pgid %d.%x\n", oid, poolid, ps);
 979
 980         ol->ol_pgid = pgid;
 981         ol->ol_stripe_unit = fl->fl_object_stripe_unit;
 982         return 0;
 983 }
 984
 985 /*
 986  * Calculate raw osd vector for the given pgid.  Return pointer to osd
 987  * array, or NULL on failure.
 988  */
 989 static int *calc_pg_raw(struct ceph_osdmap *osdmap, struct ceph_pg pgid,
 990                         int *osds, int *num)
 991 {
 992         struct ceph_pg_mapping *pg;
 993         struct ceph_pg_pool_info *pool;
 994         int ruleno;
 995         unsigned poolid, ps, pps;
 996         int preferred;
 997
 998         /* pg_temp? */
 999         pg = __lookup_pg_mapping(&osdmap->pg_temp, pgid);
1000         if (pg) {
1001                 *num = pg->len;
1002                 return pg->osds;
1003         }
1004
1005         /* crush */
1006         poolid = le32_to_cpu(pgid.pool);
1007         ps = le16_to_cpu(pgid.ps);
1008         preferred = (s16)le16_to_cpu(pgid.preferred);
1009
1010         /* don't forcefeed bad device ids to crush */
1011         if (preferred >= osdmap->max_osd ||
1012             preferred >= osdmap->crush->max_devices)
1013                 preferred = -1;
1014
1015         pool = __lookup_pg_pool(&osdmap->pg_pools, poolid);
1016         if (!pool)
1017                 return NULL;
1018         ruleno = crush_find_rule(osdmap->crush, pool->v.crush_ruleset,
1019                                  pool->v.type, pool->v.size);
1020         if (ruleno < 0) {
1021                 pr_err("no crush rule pool %d type %d size %d\n",
1022                        poolid, pool->v.type, pool->v.size);
1023                 return NULL;
1024         }
1025
1026         if (preferred >= 0)
1027                 pps = ceph_stable_mod(ps,
1028                                       le32_to_cpu(pool->v.lpgp_num),
1029                                       pool->lpgp_num_mask);
1030         else
1031                 pps = ceph_stable_mod(ps,
1032                                       le32_to_cpu(pool->v.pgp_num),
1033                                       pool->pgp_num_mask);
1034         pps += poolid;
1035         *num = crush_do_rule(osdmap->crush, ruleno, pps, osds,
1036                              min_t(int, pool->v.size, *num),
1037                              preferred, osdmap->osd_weight);
1038         return osds;
1039 }
1040
1041 /*
1042  * Return primary osd for given pgid, or -1 if none.
1043  */
1044 int ceph_calc_pg_primary(struct ceph_osdmap *osdmap, struct ceph_pg pgid)
1045 {
1046         int rawosds[10], *osds;
1047         int i, num = ARRAY_SIZE(rawosds);
1048
1049         osds = calc_pg_raw(osdmap, pgid, rawosds, &num);
1050         if (!osds)
1051                 return -1;
1052
1053         /* primary is first up osd */
1054         for (i = 0; i < num; i++)
1055                 if (ceph_osd_is_up(osdmap, osds[i])) {
1056                         return osds[i];
1057                         break;
1058                 }
1059         return -1;
1060 }