SAFE public projects git trees. - safe/jmp/linux-2.6/blob - fs/ceph/osdmap.c

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