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         ceph_decode_32(p, b->item_weight);
  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                 ceph_decode_32(p, b->item_weights[j]);
  90                 ceph_decode_32(p, b->sum_weights[j]);
  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                 ceph_decode_32(p, b->node_weights[j]);
 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                 ceph_decode_32(p, b->item_weights[j]);
 128                 ceph_decode_32(p, b->straws[j]);
 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         ceph_decode_32(p, magic);
 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         ceph_decode_32(p, c->max_buckets);
 158         ceph_decode_32(p, c->max_rules);
 159         ceph_decode_32(p, c->max_devices);
 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                         goto bad;
 204                 }
 205                 BUG_ON(size == 0);
 206                 b = c->buckets[i] = kzalloc(size, GFP_NOFS);
 207                 if (b == NULL)
 208                         goto badmem;
 209
 210                 ceph_decode_need(p, end, 4*sizeof(u32), bad);
 211                 ceph_decode_32(p, b->id);
 212                 ceph_decode_16(p, b->type);
 213                 ceph_decode_16(p, b->alg);
 214                 ceph_decode_32(p, b->weight);
 215                 ceph_decode_32(p, b->size);
 216
 217                 dout("crush_decode bucket size %d off %x %p to %p\n",
 218                      b->size, (int)(*p-start), *p, end);
 219
 220                 b->items = kcalloc(b->size, sizeof(__s32), GFP_NOFS);
 221                 if (b->items == NULL)
 222                         goto badmem;
 223                 b->perm = kcalloc(b->size, sizeof(u32), GFP_NOFS);
 224                 if (b->perm == NULL)
 225                         goto badmem;
 226                 b->perm_n = 0;
 227
 228                 ceph_decode_need(p, end, b->size*sizeof(u32), bad);
 229                 for (j = 0; j < b->size; j++)
 230                         ceph_decode_32(p, b->items[j]);
 231
 232                 switch (b->alg) {
 233                 case CRUSH_BUCKET_UNIFORM:
 234                         err = crush_decode_uniform_bucket(p, end,
 235                                   (struct crush_bucket_uniform *)b);
 236                         if (err < 0)
 237                                 goto bad;
 238                         break;
 239                 case CRUSH_BUCKET_LIST:
 240                         err = crush_decode_list_bucket(p, end,
 241                                (struct crush_bucket_list *)b);
 242                         if (err < 0)
 243                                 goto bad;
 244                         break;
 245                 case CRUSH_BUCKET_TREE:
 246                         err = crush_decode_tree_bucket(p, end,
 247                                 (struct crush_bucket_tree *)b);
 248                         if (err < 0)
 249                                 goto bad;
 250                         break;
 251                 case CRUSH_BUCKET_STRAW:
 252                         err = crush_decode_straw_bucket(p, end,
 253                                 (struct crush_bucket_straw *)b);
 254                         if (err < 0)
 255                                 goto bad;
 256                         break;
 257                 }
 258         }
 259
 260         /* rules */
 261         dout("rule vec is %p\n", c->rules);
 262         for (i = 0; i < c->max_rules; i++) {
 263                 u32 yes;
 264                 struct crush_rule *r;
 265
 266                 ceph_decode_32_safe(p, end, yes, bad);
 267                 if (!yes) {
 268                         dout("crush_decode NO rule %d off %x %p to %p\n",
 269                              i, (int)(*p-start), *p, end);
 270                         c->rules[i] = NULL;
 271                         continue;
 272                 }
 273
 274                 dout("crush_decode rule %d off %x %p to %p\n",
 275                      i, (int)(*p-start), *p, end);
 276
 277                 /* len */
 278                 ceph_decode_32_safe(p, end, yes, bad);
 279 #if BITS_PER_LONG == 32
 280                 if (yes > ULONG_MAX / sizeof(struct crush_rule_step))
 281                         goto bad;
 282 #endif
 283                 r = c->rules[i] = kmalloc(sizeof(*r) +
 284                                           yes*sizeof(struct crush_rule_step),
 285                                           GFP_NOFS);
 286                 if (r == NULL)
 287                         goto badmem;
 288                 dout(" rule %d is at %p\n", i, r);
 289                 r->len = yes;
 290                 ceph_decode_copy_safe(p, end, &r->mask, 4, bad); /* 4 u8's */
 291                 ceph_decode_need(p, end, r->len*3*sizeof(u32), bad);
 292                 for (j = 0; j < r->len; j++) {
 293                         ceph_decode_32(p, r->steps[j].op);
 294                         ceph_decode_32(p, r->steps[j].arg1);
 295                         ceph_decode_32(p, r->steps[j].arg2);
 296                 }
 297         }
 298
 299         /* ignore trailing name maps. */
 300
 301         dout("crush_decode success\n");
 302         return c;
 303
 304 badmem:
 305         err = -ENOMEM;
 306 bad:
 307         dout("crush_decode fail %d\n", err);
 308         crush_destroy(c);
 309         return ERR_PTR(err);
 310 }
 311
 312
 313 /*
 314  * osd map
 315  */
 316 void ceph_osdmap_destroy(struct ceph_osdmap *map)
 317 {
 318         dout("osdmap_destroy %p\n", map);
 319         if (map->crush)
 320                 crush_destroy(map->crush);
 321         while (!RB_EMPTY_ROOT(&map->pg_temp))
 322                 rb_erase(rb_first(&map->pg_temp), &map->pg_temp);
 323         kfree(map->osd_state);
 324         kfree(map->osd_weight);
 325         kfree(map->pg_pool);
 326         kfree(map->osd_addr);
 327         kfree(map);
 328 }
 329
 330 /*
 331  * adjust max osd value.  reallocate arrays.
 332  */
 333 static int osdmap_set_max_osd(struct ceph_osdmap *map, int max)
 334 {
 335         u8 *state;
 336         struct ceph_entity_addr *addr;
 337         u32 *weight;
 338
 339         state = kcalloc(max, sizeof(*state), GFP_NOFS);
 340         addr = kcalloc(max, sizeof(*addr), GFP_NOFS);
 341         weight = kcalloc(max, sizeof(*weight), GFP_NOFS);
 342         if (state == NULL || addr == NULL || weight == NULL) {
 343                 kfree(state);
 344                 kfree(addr);
 345                 kfree(weight);
 346                 return -ENOMEM;
 347         }
 348
 349         /* copy old? */
 350         if (map->osd_state) {
 351                 memcpy(state, map->osd_state, map->max_osd*sizeof(*state));
 352                 memcpy(addr, map->osd_addr, map->max_osd*sizeof(*addr));
 353                 memcpy(weight, map->osd_weight, map->max_osd*sizeof(*weight));
 354                 kfree(map->osd_state);
 355                 kfree(map->osd_addr);
 356                 kfree(map->osd_weight);
 357         }
 358
 359         map->osd_state = state;
 360         map->osd_weight = weight;
 361         map->osd_addr = addr;
 362         map->max_osd = max;
 363         return 0;
 364 }
 365
 366 /*
 367  * Insert a new pg_temp mapping
 368  */
 369 static void __insert_pg_mapping(struct ceph_pg_mapping *new,
 370                                 struct rb_root *root)
 371 {
 372         struct rb_node **p = &root->rb_node;
 373         struct rb_node *parent = NULL;
 374         struct ceph_pg_mapping *pg = NULL;
 375
 376         while (*p) {
 377                 parent = *p;
 378                 pg = rb_entry(parent, struct ceph_pg_mapping, node);
 379                 if (new->pgid < pg->pgid)
 380                         p = &(*p)->rb_left;
 381                 else if (new->pgid > pg->pgid)
 382                         p = &(*p)->rb_right;
 383                 else
 384                         BUG();
 385         }
 386
 387         rb_link_node(&new->node, parent, p);
 388         rb_insert_color(&new->node, root);
 389 }
 390
 391 /*
 392  * decode a full map.
 393  */
 394 struct ceph_osdmap *osdmap_decode(void **p, void *end)
 395 {
 396         struct ceph_osdmap *map;
 397         u16 version;
 398         u32 len, max, i;
 399         int err = -EINVAL;
 400         void *start = *p;
 401
 402         dout("osdmap_decode %p to %p len %d\n", *p, end, (int)(end - *p));
 403
 404         map = kzalloc(sizeof(*map), GFP_NOFS);
 405         if (map == NULL)
 406                 return ERR_PTR(-ENOMEM);
 407         map->pg_temp = RB_ROOT;
 408
 409         ceph_decode_16_safe(p, end, version, bad);
 410
 411         ceph_decode_need(p, end, 2*sizeof(u64)+6*sizeof(u32), bad);
 412         ceph_decode_copy(p, &map->fsid, sizeof(map->fsid));
 413         ceph_decode_32(p, map->epoch);
 414         ceph_decode_copy(p, &map->created, sizeof(map->created));
 415         ceph_decode_copy(p, &map->modified, sizeof(map->modified));
 416
 417         ceph_decode_32(p, map->num_pools);
 418         map->pg_pool = kcalloc(map->num_pools, sizeof(*map->pg_pool),
 419                                GFP_NOFS);
 420         if (!map->pg_pool) {
 421                 err = -ENOMEM;
 422                 goto bad;
 423         }
 424         ceph_decode_32_safe(p, end, max, bad);
 425         while (max--) {
 426                 ceph_decode_need(p, end, 4+sizeof(map->pg_pool->v), bad);
 427                 ceph_decode_32(p, i);
 428                 if (i >= map->num_pools)
 429                         goto bad;
 430                 ceph_decode_copy(p, &map->pg_pool[i].v,
 431                                  sizeof(map->pg_pool->v));
 432                 calc_pg_masks(&map->pg_pool[i]);
 433                 p += le32_to_cpu(map->pg_pool[i].v.num_snaps) * sizeof(u64);
 434                 p += le32_to_cpu(map->pg_pool[i].v.num_removed_snap_intervals)
 435                         * sizeof(u64) * 2;
 436         }
 437
 438         ceph_decode_32_safe(p, end, map->flags, bad);
 439
 440         ceph_decode_32(p, max);
 441
 442         /* (re)alloc osd arrays */
 443         err = osdmap_set_max_osd(map, max);
 444         if (err < 0)
 445                 goto bad;
 446         dout("osdmap_decode max_osd = %d\n", map->max_osd);
 447
 448         /* osds */
 449         err = -EINVAL;
 450         ceph_decode_need(p, end, 3*sizeof(u32) +
 451                          map->max_osd*(1 + sizeof(*map->osd_weight) +
 452                                        sizeof(*map->osd_addr)), bad);
 453         *p += 4; /* skip length field (should match max) */
 454         ceph_decode_copy(p, map->osd_state, map->max_osd);
 455
 456         *p += 4; /* skip length field (should match max) */
 457         for (i = 0; i < map->max_osd; i++)
 458                 ceph_decode_32(p, map->osd_weight[i]);
 459
 460         *p += 4; /* skip length field (should match max) */
 461         ceph_decode_copy(p, map->osd_addr, map->max_osd*sizeof(*map->osd_addr));
 462
 463         /* pg_temp */
 464         ceph_decode_32_safe(p, end, len, bad);
 465         for (i = 0; i < len; i++) {
 466                 int n, j;
 467                 u64 pgid;
 468                 struct ceph_pg_mapping *pg;
 469
 470                 ceph_decode_need(p, end, sizeof(u32) + sizeof(u64), bad);
 471                 ceph_decode_64(p, pgid);
 472                 ceph_decode_32(p, n);
 473                 ceph_decode_need(p, end, n * sizeof(u32), bad);
 474                 pg = kmalloc(sizeof(*pg) + n*sizeof(u32), GFP_NOFS);
 475                 if (!pg) {
 476                         err = -ENOMEM;
 477                         goto bad;
 478                 }
 479                 pg->pgid = pgid;
 480                 pg->len = n;
 481                 for (j = 0; j < n; j++)
 482                         ceph_decode_32(p, pg->osds[j]);
 483
 484                 __insert_pg_mapping(pg, &map->pg_temp);
 485                 dout(" added pg_temp %llx len %d\n", pgid, len);
 486         }
 487
 488         /* crush */
 489         ceph_decode_32_safe(p, end, len, bad);
 490         dout("osdmap_decode crush len %d from off 0x%x\n", len,
 491              (int)(*p - start));
 492         ceph_decode_need(p, end, len, bad);
 493         map->crush = crush_decode(*p, end);
 494         *p += len;
 495         if (IS_ERR(map->crush)) {
 496                 err = PTR_ERR(map->crush);
 497                 map->crush = NULL;
 498                 goto bad;
 499         }
 500
 501         /* ignore the rest of the map */
 502         *p = end;
 503
 504         dout("osdmap_decode done %p %p\n", *p, end);
 505         return map;
 506
 507 bad:
 508         dout("osdmap_decode fail\n");
 509         ceph_osdmap_destroy(map);
 510         return ERR_PTR(err);
 511 }
 512
 513 /*
 514  * decode and apply an incremental map update.
 515  */
 516 struct ceph_osdmap *osdmap_apply_incremental(void **p, void *end,
 517                                              struct ceph_osdmap *map,
 518                                              struct ceph_messenger *msgr)
 519 {
 520         struct ceph_osdmap *newmap = map;
 521         struct crush_map *newcrush = NULL;
 522         struct ceph_fsid fsid;
 523         u32 epoch = 0;
 524         struct ceph_timespec modified;
 525         u32 len, pool;
 526         __s32 new_flags, max;
 527         void *start = *p;
 528         int err = -EINVAL;
 529         u16 version;
 530         struct rb_node *rbp;
 531
 532         ceph_decode_16_safe(p, end, version, bad);
 533
 534         ceph_decode_need(p, end, sizeof(fsid)+sizeof(modified)+2*sizeof(u32),
 535                          bad);
 536         ceph_decode_copy(p, &fsid, sizeof(fsid));
 537         ceph_decode_32(p, epoch);
 538         BUG_ON(epoch != map->epoch+1);
 539         ceph_decode_copy(p, &modified, sizeof(modified));
 540         ceph_decode_32(p, new_flags);
 541
 542         /* full map? */
 543         ceph_decode_32_safe(p, end, len, bad);
 544         if (len > 0) {
 545                 dout("apply_incremental full map len %d, %p to %p\n",
 546                      len, *p, end);
 547                 newmap = osdmap_decode(p, min(*p+len, end));
 548                 return newmap;  /* error or not */
 549         }
 550
 551         /* new crush? */
 552         ceph_decode_32_safe(p, end, len, bad);
 553         if (len > 0) {
 554                 dout("apply_incremental new crush map len %d, %p to %p\n",
 555                      len, *p, end);
 556                 newcrush = crush_decode(*p, min(*p+len, end));
 557                 if (IS_ERR(newcrush))
 558                         return ERR_PTR(PTR_ERR(newcrush));
 559         }
 560
 561         /* new flags? */
 562         if (new_flags >= 0)
 563                 map->flags = new_flags;
 564
 565         ceph_decode_need(p, end, 5*sizeof(u32), bad);
 566
 567         /* new max? */
 568         ceph_decode_32(p, max);
 569         if (max >= 0) {
 570                 err = osdmap_set_max_osd(map, max);
 571                 if (err < 0)
 572                         goto bad;
 573         }
 574
 575         map->epoch++;
 576         map->modified = map->modified;
 577         if (newcrush) {
 578                 if (map->crush)
 579                         crush_destroy(map->crush);
 580                 map->crush = newcrush;
 581                 newcrush = NULL;
 582         }
 583
 584         /* new_pool */
 585         ceph_decode_32_safe(p, end, len, bad);
 586         while (len--) {
 587                 ceph_decode_32_safe(p, end, pool, bad);
 588                 if (pool >= map->num_pools) {
 589                         void *pg_pool = kcalloc(pool + 1,
 590                                                 sizeof(*map->pg_pool),
 591                                                 GFP_NOFS);
 592                         if (!pg_pool) {
 593                                 err = -ENOMEM;
 594                                 goto bad;
 595                         }
 596                         memcpy(pg_pool, map->pg_pool,
 597                                map->num_pools * sizeof(*map->pg_pool));
 598                         kfree(map->pg_pool);
 599                         map->pg_pool = pg_pool;
 600                         map->num_pools = pool+1;
 601                 }
 602                 ceph_decode_copy(p, &map->pg_pool[pool].v,
 603                                  sizeof(map->pg_pool->v));
 604                 calc_pg_masks(&map->pg_pool[pool]);
 605         }
 606
 607         /* old_pool (ignore) */
 608         ceph_decode_32_safe(p, end, len, bad);
 609         *p += len * sizeof(u32);
 610
 611         /* new_up */
 612         err = -EINVAL;
 613         ceph_decode_32_safe(p, end, len, bad);
 614         while (len--) {
 615                 u32 osd;
 616                 struct ceph_entity_addr addr;
 617                 ceph_decode_32_safe(p, end, osd, bad);
 618                 ceph_decode_copy_safe(p, end, &addr, sizeof(addr), bad);
 619                 pr_info("osd%d up\n", osd);
 620                 BUG_ON(osd >= map->max_osd);
 621                 map->osd_state[osd] |= CEPH_OSD_UP;
 622                 map->osd_addr[osd] = addr;
 623         }
 624
 625         /* new_down */
 626         ceph_decode_32_safe(p, end, len, bad);
 627         while (len--) {
 628                 u32 osd;
 629                 ceph_decode_32_safe(p, end, osd, bad);
 630                 (*p)++;  /* clean flag */
 631                 pr_info("ceph osd%d down\n", osd);
 632                 if (osd < map->max_osd)
 633                         map->osd_state[osd] &= ~CEPH_OSD_UP;
 634         }
 635
 636         /* new_weight */
 637         ceph_decode_32_safe(p, end, len, bad);
 638         while (len--) {
 639                 u32 osd, off;
 640                 ceph_decode_need(p, end, sizeof(u32)*2, bad);
 641                 ceph_decode_32(p, osd);
 642                 ceph_decode_32(p, off);
 643                 pr_info("osd%d weight 0x%x %s\n", osd, off,
 644                      off == CEPH_OSD_IN ? "(in)" :
 645                      (off == CEPH_OSD_OUT ? "(out)" : ""));
 646                 if (osd < map->max_osd)
 647                         map->osd_weight[osd] = off;
 648         }
 649
 650         /* new_pg_temp */
 651         rbp = rb_first(&map->pg_temp);
 652         ceph_decode_32_safe(p, end, len, bad);
 653         while (len--) {
 654                 struct ceph_pg_mapping *pg;
 655                 int j;
 656                 u64 pgid;
 657                 u32 pglen;
 658                 ceph_decode_need(p, end, sizeof(u64) + sizeof(u32), bad);
 659                 ceph_decode_64(p, pgid);
 660                 ceph_decode_32(p, pglen);
 661
 662                 /* remove any? */
 663                 while (rbp && rb_entry(rbp, struct ceph_pg_mapping,
 664                                        node)->pgid <= pgid) {
 665                         struct rb_node *cur = rbp;
 666                         rbp = rb_next(rbp);
 667                         dout(" removed pg_temp %llx\n",
 668                              rb_entry(cur, struct ceph_pg_mapping, node)->pgid);
 669                         rb_erase(cur, &map->pg_temp);
 670                 }
 671
 672                 if (pglen) {
 673                         /* insert */
 674                         ceph_decode_need(p, end, pglen*sizeof(u32), bad);
 675                         pg = kmalloc(sizeof(*pg) + sizeof(u32)*pglen, GFP_NOFS);
 676                         if (!pg) {
 677                                 err = -ENOMEM;
 678                                 goto bad;
 679                         }
 680                         pg->pgid = pgid;
 681                         pg->len = pglen;
 682                         for (j = 0; j < len; j++)
 683                                 ceph_decode_32(p, pg->osds[j]);
 684                         __insert_pg_mapping(pg, &map->pg_temp);
 685                         dout(" added pg_temp %llx len %d\n", pgid, pglen);
 686                 }
 687         }
 688         while (rbp) {
 689                 struct rb_node *cur = rbp;
 690                 rbp = rb_next(rbp);
 691                 dout(" removed pg_temp %llx\n",
 692                      rb_entry(cur, struct ceph_pg_mapping, node)->pgid);
 693                 rb_erase(cur, &map->pg_temp);
 694         }
 695
 696         /* ignore the rest */
 697         *p = end;
 698         return map;
 699
 700 bad:
 701         pr_err("corrupt inc osdmap epoch %d off %d (%p of %p-%p)\n",
 702                epoch, (int)(*p - start), *p, start, end);
 703         if (newcrush)
 704                 crush_destroy(newcrush);
 705         return ERR_PTR(err);
 706 }
 707
 708
 709
 710
 711 /*
 712  * calculate file layout from given offset, length.
 713  * fill in correct oid, logical length, and object extent
 714  * offset, length.
 715  *
 716  * for now, we write only a single su, until we can
 717  * pass a stride back to the caller.
 718  */
 719 void ceph_calc_file_object_mapping(struct ceph_file_layout *layout,
 720                                    u64 off, u64 *plen,
 721                                    u64 *bno,
 722                                    u64 *oxoff, u64 *oxlen)
 723 {
 724         u32 osize = le32_to_cpu(layout->fl_object_size);
 725         u32 su = le32_to_cpu(layout->fl_stripe_unit);
 726         u32 sc = le32_to_cpu(layout->fl_stripe_count);
 727         u32 bl, stripeno, stripepos, objsetno;
 728         u32 su_per_object;
 729         u64 t;
 730
 731         dout("mapping %llu~%llu  osize %u fl_su %u\n", off, *plen,
 732              osize, su);
 733         su_per_object = osize / le32_to_cpu(layout->fl_stripe_unit);
 734         dout("osize %u / su %u = su_per_object %u\n", osize, su,
 735              su_per_object);
 736
 737         BUG_ON((su & ~PAGE_MASK) != 0);
 738         /* bl = *off / su; */
 739         t = off;
 740         do_div(t, su);
 741         bl = t;
 742         dout("off %llu / su %u = bl %u\n", off, su, bl);
 743
 744         stripeno = bl / sc;
 745         stripepos = bl % sc;
 746         objsetno = stripeno / su_per_object;
 747
 748         *bno = objsetno * sc + stripepos;
 749         dout("objset %u * sc %u = bno %u\n", objsetno, sc, (unsigned)*bno);
 750         /* *oxoff = *off / layout->fl_stripe_unit; */
 751         t = off;
 752         *oxoff = do_div(t, su);
 753         *oxlen = min_t(u64, *plen, su - *oxoff);
 754         *plen = *oxlen;
 755
 756         dout(" obj extent %llu~%llu\n", *oxoff, *oxlen);
 757 }
 758
 759 /*
 760  * calculate an object layout (i.e. pgid) from an oid,
 761  * file_layout, and osdmap
 762  */
 763 int ceph_calc_object_layout(struct ceph_object_layout *ol,
 764                             const char *oid,
 765                             struct ceph_file_layout *fl,
 766                             struct ceph_osdmap *osdmap)
 767 {
 768         unsigned num, num_mask;
 769         union ceph_pg pgid;
 770         s32 preferred = (s32)le32_to_cpu(fl->fl_pg_preferred);
 771         int poolid = le32_to_cpu(fl->fl_pg_pool);
 772         struct ceph_pg_pool_info *pool;
 773
 774         if (poolid >= osdmap->num_pools)
 775                 return -EIO;
 776         pool = &osdmap->pg_pool[poolid];
 777
 778         if (preferred >= 0) {
 779                 num = le32_to_cpu(pool->v.lpg_num);
 780                 num_mask = pool->lpg_num_mask;
 781         } else {
 782                 num = le32_to_cpu(pool->v.pg_num);
 783                 num_mask = pool->pg_num_mask;
 784         }
 785
 786         pgid.pg64 = 0;   /* start with it zeroed out */
 787         pgid.pg.ps = ceph_full_name_hash(oid, strlen(oid));
 788         pgid.pg.preferred = preferred;
 789         pgid.pg.pool = le32_to_cpu(fl->fl_pg_pool);
 790         if (preferred >= 0)
 791                 dout("calc_object_layout '%s' pgid %d.%xp%d (%llx)\n", oid,
 792                      pgid.pg.pool, pgid.pg.ps, (int)preferred, pgid.pg64);
 793         else
 794                 dout("calc_object_layout '%s' pgid %d.%x (%llx)\n", oid,
 795                      pgid.pg.pool, pgid.pg.ps, pgid.pg64);
 796
 797         ol->ol_pgid = cpu_to_le64(pgid.pg64);
 798         ol->ol_stripe_unit = fl->fl_object_stripe_unit;
 799
 800         return 0;
 801 }
 802
 803 /*
 804  * Calculate raw osd vector for the given pgid.  Return pointer to osd
 805  * array, or NULL on failure.
 806  */
 807 static int *calc_pg_raw(struct ceph_osdmap *osdmap, union ceph_pg pgid,
 808                         int *osds, int *num)
 809 {
 810         struct rb_node *n = osdmap->pg_temp.rb_node;
 811         struct ceph_pg_mapping *pg;
 812         struct ceph_pg_pool_info *pool;
 813         int ruleno;
 814         unsigned pps; /* placement ps */
 815
 816         /* pg_temp? */
 817         while (n) {
 818                 pg = rb_entry(n, struct ceph_pg_mapping, node);
 819                 if (pgid.pg64 < pg->pgid)
 820                         n = n->rb_left;
 821                 else if (pgid.pg64 > pg->pgid)
 822                         n = n->rb_right;
 823                 else {
 824                         *num = pg->len;
 825                         return pg->osds;
 826                 }
 827         }
 828
 829         /* crush */
 830         if (pgid.pg.pool >= osdmap->num_pools)
 831                 return NULL;
 832         pool = &osdmap->pg_pool[pgid.pg.pool];
 833         ruleno = crush_find_rule(osdmap->crush, pool->v.crush_ruleset,
 834                                  pool->v.type, pool->v.size);
 835         if (ruleno < 0) {
 836                 pr_err("no crush rule pool %d type %d size %d\n",
 837                        pgid.pg.pool, pool->v.type, pool->v.size);
 838                 return NULL;
 839         }
 840
 841         if (pgid.pg.preferred >= 0)
 842                 pps = ceph_stable_mod(pgid.pg.ps,
 843                                       le32_to_cpu(pool->v.lpgp_num),
 844                                       pool->lpgp_num_mask);
 845         else
 846                 pps = ceph_stable_mod(pgid.pg.ps,
 847                                       le32_to_cpu(pool->v.pgp_num),
 848                                       pool->pgp_num_mask);
 849         pps += pgid.pg.pool;
 850         *num = crush_do_rule(osdmap->crush, ruleno, pps, osds,
 851                              min_t(int, pool->v.size, *num),
 852                              pgid.pg.preferred, osdmap->osd_weight);
 853         return osds;
 854 }
 855
 856 /*
 857  * Return primary osd for given pgid, or -1 if none.
 858  */
 859 int ceph_calc_pg_primary(struct ceph_osdmap *osdmap, union ceph_pg pgid)
 860 {
 861         int rawosds[10], *osds;
 862         int i, num = ARRAY_SIZE(rawosds);
 863
 864         osds = calc_pg_raw(osdmap, pgid, rawosds, &num);
 865         if (!osds)
 866                 return -1;
 867
 868         /* primary is first up osd */
 869         for (i = 0; i < num; i++)
 870                 if (ceph_osd_is_up(osdmap, osds[i])) {
 871                         return osds[i];
 872                         break;
 873                 }
 874         return -1;
 875 }