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  * rbtree of pg_mapping for handling pg_temp (explicit mapping of pgid
 319  * to a set of osds)
 320  */
 321 static int pgid_cmp(struct ceph_pg l, struct ceph_pg r)
 322 {
 323         u64 a = *(u64 *)&l;
 324         u64 b = *(u64 *)&r;
 325
 326         if (a < b)
 327                 return -1;
 328         if (a > b)
 329                 return 1;
 330         return 0;
 331 }
 332
 333 static int __insert_pg_mapping(struct ceph_pg_mapping *new,
 334                                struct rb_root *root)
 335 {
 336         struct rb_node **p = &root->rb_node;
 337         struct rb_node *parent = NULL;
 338         struct ceph_pg_mapping *pg = NULL;
 339         int c;
 340
 341         while (*p) {
 342                 parent = *p;
 343                 pg = rb_entry(parent, struct ceph_pg_mapping, node);
 344                 c = pgid_cmp(new->pgid, pg->pgid);
 345                 if (c < 0)
 346                         p = &(*p)->rb_left;
 347                 else if (c > 0)
 348                         p = &(*p)->rb_right;
 349                 else
 350                         return -EEXIST;
 351         }
 352
 353         rb_link_node(&new->node, parent, p);
 354         rb_insert_color(&new->node, root);
 355         return 0;
 356 }
 357
 358 static struct ceph_pg_mapping *__lookup_pg_mapping(struct rb_root *root,
 359                                                    struct ceph_pg pgid)
 360 {
 361         struct rb_node *n = root->rb_node;
 362         struct ceph_pg_mapping *pg;
 363         int c;
 364
 365         while (n) {
 366                 pg = rb_entry(n, struct ceph_pg_mapping, node);
 367                 c = pgid_cmp(pgid, pg->pgid);
 368                 if (c < 0)
 369                         n = n->rb_left;
 370                 else if (c > 0)
 371                         n = n->rb_right;
 372                 else
 373                         return pg;
 374         }
 375         return NULL;
 376 }
 377
 378 /*
 379  * rbtree of pg pool info
 380  */
 381 static int __insert_pg_pool(struct rb_root *root, struct ceph_pg_pool_info *new)
 382 {
 383         struct rb_node **p = &root->rb_node;
 384         struct rb_node *parent = NULL;
 385         struct ceph_pg_pool_info *pi = NULL;
 386
 387         while (*p) {
 388                 parent = *p;
 389                 pi = rb_entry(parent, struct ceph_pg_pool_info, node);
 390                 if (new->id < pi->id)
 391                         p = &(*p)->rb_left;
 392                 else if (new->id > pi->id)
 393                         p = &(*p)->rb_right;
 394                 else
 395                         return -EEXIST;
 396         }
 397
 398         rb_link_node(&new->node, parent, p);
 399         rb_insert_color(&new->node, root);
 400         return 0;
 401 }
 402
 403 static struct ceph_pg_pool_info *__lookup_pg_pool(struct rb_root *root, int id)
 404 {
 405         struct ceph_pg_pool_info *pi;
 406         struct rb_node *n = root->rb_node;
 407
 408         while (n) {
 409                 pi = rb_entry(n, struct ceph_pg_pool_info, node);
 410                 if (id < pi->id)
 411                         n = n->rb_left;
 412                 else if (id > pi->id)
 413                         n = n->rb_right;
 414                 else
 415                         return pi;
 416         }
 417         return NULL;
 418 }
 419
 420 static void __remove_pg_pool(struct rb_root *root, struct ceph_pg_pool_info *pi)
 421 {
 422         rb_erase(&pi->node, root);
 423         kfree(pi->name);
 424         kfree(pi);
 425 }
 426
 427 static int __decode_pool(void **p, void *end, struct ceph_pg_pool_info *pi)
 428 {
 429         unsigned n, m;
 430
 431         ceph_decode_copy(p, &pi->v, sizeof(pi->v));
 432         calc_pg_masks(pi);
 433
 434         /* num_snaps * snap_info_t */
 435         n = le32_to_cpu(pi->v.num_snaps);
 436         while (n--) {
 437                 ceph_decode_need(p, end, sizeof(u64) + 1 + sizeof(u64) +
 438                                  sizeof(struct ceph_timespec), bad);
 439                 *p += sizeof(u64) +       /* key */
 440                         1 + sizeof(u64) + /* u8, snapid */
 441                         sizeof(struct ceph_timespec);
 442                 m = ceph_decode_32(p);    /* snap name */
 443                 *p += m;
 444         }
 445
 446         *p += le32_to_cpu(pi->v.num_removed_snap_intervals) * sizeof(u64) * 2;
 447         return 0;
 448
 449 bad:
 450         return -EINVAL;
 451 }
 452
 453 static int __decode_pool_names(void **p, void *end, struct ceph_osdmap *map)
 454 {
 455         struct ceph_pg_pool_info *pi;
 456         u32 num, len, pool;
 457
 458         ceph_decode_32_safe(p, end, num, bad);
 459         dout(" %d pool names\n", num);
 460         while (num--) {
 461                 ceph_decode_32_safe(p, end, pool, bad);
 462                 ceph_decode_32_safe(p, end, len, bad);
 463                 dout("  pool %d len %d\n", pool, len);
 464                 pi = __lookup_pg_pool(&map->pg_pools, pool);
 465                 if (pi) {
 466                         kfree(pi->name);
 467                         pi->name = kmalloc(len + 1, GFP_NOFS);
 468                         if (pi->name) {
 469                                 memcpy(pi->name, *p, len);
 470                                 pi->name[len] = '\0';
 471                                 dout("  name is %s\n", pi->name);
 472                         }
 473                 }
 474                 *p += len;
 475         }
 476         return 0;
 477
 478 bad:
 479         return -EINVAL;
 480 }
 481
 482 /*
 483  * osd map
 484  */
 485 void ceph_osdmap_destroy(struct ceph_osdmap *map)
 486 {
 487         dout("osdmap_destroy %p\n", map);
 488         if (map->crush)
 489                 crush_destroy(map->crush);
 490         while (!RB_EMPTY_ROOT(&map->pg_temp)) {
 491                 struct ceph_pg_mapping *pg =
 492                         rb_entry(rb_first(&map->pg_temp),
 493                                  struct ceph_pg_mapping, node);
 494                 rb_erase(&pg->node, &map->pg_temp);
 495                 kfree(pg);
 496         }
 497         while (!RB_EMPTY_ROOT(&map->pg_pools)) {
 498                 struct ceph_pg_pool_info *pi =
 499                         rb_entry(rb_first(&map->pg_pools),
 500                                  struct ceph_pg_pool_info, node);
 501                 __remove_pg_pool(&map->pg_pools, pi);
 502         }
 503         kfree(map->osd_state);
 504         kfree(map->osd_weight);
 505         kfree(map->osd_addr);
 506         kfree(map);
 507 }
 508
 509 /*
 510  * adjust max osd value.  reallocate arrays.
 511  */
 512 static int osdmap_set_max_osd(struct ceph_osdmap *map, int max)
 513 {
 514         u8 *state;
 515         struct ceph_entity_addr *addr;
 516         u32 *weight;
 517
 518         state = kcalloc(max, sizeof(*state), GFP_NOFS);
 519         addr = kcalloc(max, sizeof(*addr), GFP_NOFS);
 520         weight = kcalloc(max, sizeof(*weight), GFP_NOFS);
 521         if (state == NULL || addr == NULL || weight == NULL) {
 522                 kfree(state);
 523                 kfree(addr);
 524                 kfree(weight);
 525                 return -ENOMEM;
 526         }
 527
 528         /* copy old? */
 529         if (map->osd_state) {
 530                 memcpy(state, map->osd_state, map->max_osd*sizeof(*state));
 531                 memcpy(addr, map->osd_addr, map->max_osd*sizeof(*addr));
 532                 memcpy(weight, map->osd_weight, map->max_osd*sizeof(*weight));
 533                 kfree(map->osd_state);
 534                 kfree(map->osd_addr);
 535                 kfree(map->osd_weight);
 536         }
 537
 538         map->osd_state = state;
 539         map->osd_weight = weight;
 540         map->osd_addr = addr;
 541         map->max_osd = max;
 542         return 0;
 543 }
 544
 545 /*
 546  * decode a full map.
 547  */
 548 struct ceph_osdmap *osdmap_decode(void **p, void *end)
 549 {
 550         struct ceph_osdmap *map;
 551         u16 version;
 552         u32 len, max, i;
 553         u8 ev;
 554         int err = -EINVAL;
 555         void *start = *p;
 556         struct ceph_pg_pool_info *pi;
 557
 558         dout("osdmap_decode %p to %p len %d\n", *p, end, (int)(end - *p));
 559
 560         map = kzalloc(sizeof(*map), GFP_NOFS);
 561         if (map == NULL)
 562                 return ERR_PTR(-ENOMEM);
 563         map->pg_temp = RB_ROOT;
 564
 565         ceph_decode_16_safe(p, end, version, bad);
 566         if (version > CEPH_OSDMAP_VERSION) {
 567                 pr_warning("got unknown v %d > %d of osdmap\n", version,
 568                            CEPH_OSDMAP_VERSION);
 569                 goto bad;
 570         }
 571
 572         ceph_decode_need(p, end, 2*sizeof(u64)+6*sizeof(u32), bad);
 573         ceph_decode_copy(p, &map->fsid, sizeof(map->fsid));
 574         map->epoch = ceph_decode_32(p);
 575         ceph_decode_copy(p, &map->created, sizeof(map->created));
 576         ceph_decode_copy(p, &map->modified, sizeof(map->modified));
 577
 578         ceph_decode_32_safe(p, end, max, bad);
 579         while (max--) {
 580                 ceph_decode_need(p, end, 4 + 1 + sizeof(pi->v), bad);
 581                 pi = kzalloc(sizeof(*pi), GFP_NOFS);
 582                 if (!pi)
 583                         goto bad;
 584                 pi->id = ceph_decode_32(p);
 585                 ev = ceph_decode_8(p); /* encoding version */
 586                 if (ev > CEPH_PG_POOL_VERSION) {
 587                         pr_warning("got unknown v %d > %d of ceph_pg_pool\n",
 588                                    ev, CEPH_PG_POOL_VERSION);
 589                         kfree(pi);
 590                         goto bad;
 591                 }
 592                 err = __decode_pool(p, end, pi);
 593                 if (err < 0)
 594                         goto bad;
 595                 __insert_pg_pool(&map->pg_pools, pi);
 596         }
 597
 598         if (version >= 5 && __decode_pool_names(p, end, map) < 0)
 599                 goto bad;
 600
 601         ceph_decode_32_safe(p, end, map->pool_max, bad);
 602
 603         ceph_decode_32_safe(p, end, map->flags, bad);
 604
 605         max = ceph_decode_32(p);
 606
 607         /* (re)alloc osd arrays */
 608         err = osdmap_set_max_osd(map, max);
 609         if (err < 0)
 610                 goto bad;
 611         dout("osdmap_decode max_osd = %d\n", map->max_osd);
 612
 613         /* osds */
 614         err = -EINVAL;
 615         ceph_decode_need(p, end, 3*sizeof(u32) +
 616                          map->max_osd*(1 + sizeof(*map->osd_weight) +
 617                                        sizeof(*map->osd_addr)), bad);
 618         *p += 4; /* skip length field (should match max) */
 619         ceph_decode_copy(p, map->osd_state, map->max_osd);
 620
 621         *p += 4; /* skip length field (should match max) */
 622         for (i = 0; i < map->max_osd; i++)
 623                 map->osd_weight[i] = ceph_decode_32(p);
 624
 625         *p += 4; /* skip length field (should match max) */
 626         ceph_decode_copy(p, map->osd_addr, map->max_osd*sizeof(*map->osd_addr));
 627         for (i = 0; i < map->max_osd; i++)
 628                 ceph_decode_addr(&map->osd_addr[i]);
 629
 630         /* pg_temp */
 631         ceph_decode_32_safe(p, end, len, bad);
 632         for (i = 0; i < len; i++) {
 633                 int n, j;
 634                 struct ceph_pg pgid;
 635                 struct ceph_pg_mapping *pg;
 636
 637                 ceph_decode_need(p, end, sizeof(u32) + sizeof(u64), bad);
 638                 ceph_decode_copy(p, &pgid, sizeof(pgid));
 639                 n = ceph_decode_32(p);
 640                 ceph_decode_need(p, end, n * sizeof(u32), bad);
 641                 err = -ENOMEM;
 642                 pg = kmalloc(sizeof(*pg) + n*sizeof(u32), GFP_NOFS);
 643                 if (!pg)
 644                         goto bad;
 645                 pg->pgid = pgid;
 646                 pg->len = n;
 647                 for (j = 0; j < n; j++)
 648                         pg->osds[j] = ceph_decode_32(p);
 649
 650                 err = __insert_pg_mapping(pg, &map->pg_temp);
 651                 if (err)
 652                         goto bad;
 653                 dout(" added pg_temp %llx len %d\n", *(u64 *)&pgid, len);
 654         }
 655
 656         /* crush */
 657         ceph_decode_32_safe(p, end, len, bad);
 658         dout("osdmap_decode crush len %d from off 0x%x\n", len,
 659              (int)(*p - start));
 660         ceph_decode_need(p, end, len, bad);
 661         map->crush = crush_decode(*p, end);
 662         *p += len;
 663         if (IS_ERR(map->crush)) {
 664                 err = PTR_ERR(map->crush);
 665                 map->crush = NULL;
 666                 goto bad;
 667         }
 668
 669         /* ignore the rest of the map */
 670         *p = end;
 671
 672         dout("osdmap_decode done %p %p\n", *p, end);
 673         return map;
 674
 675 bad:
 676         dout("osdmap_decode fail\n");
 677         ceph_osdmap_destroy(map);
 678         return ERR_PTR(err);
 679 }
 680
 681 /*
 682  * decode and apply an incremental map update.
 683  */
 684 struct ceph_osdmap *osdmap_apply_incremental(void **p, void *end,
 685                                              struct ceph_osdmap *map,
 686                                              struct ceph_messenger *msgr)
 687 {
 688         struct crush_map *newcrush = NULL;
 689         struct ceph_fsid fsid;
 690         u32 epoch = 0;
 691         struct ceph_timespec modified;
 692         u32 len, pool;
 693         __s32 new_pool_max, new_flags, max;
 694         void *start = *p;
 695         int err = -EINVAL;
 696         u16 version;
 697         struct rb_node *rbp;
 698
 699         ceph_decode_16_safe(p, end, version, bad);
 700         if (version > CEPH_OSDMAP_INC_VERSION) {
 701                 pr_warning("got unknown v %d > %d of inc osdmap\n", version,
 702                            CEPH_OSDMAP_INC_VERSION);
 703                 goto bad;
 704         }
 705
 706         ceph_decode_need(p, end, sizeof(fsid)+sizeof(modified)+2*sizeof(u32),
 707                          bad);
 708         ceph_decode_copy(p, &fsid, sizeof(fsid));
 709         epoch = ceph_decode_32(p);
 710         BUG_ON(epoch != map->epoch+1);
 711         ceph_decode_copy(p, &modified, sizeof(modified));
 712         new_pool_max = ceph_decode_32(p);
 713         new_flags = ceph_decode_32(p);
 714
 715         /* full map? */
 716         ceph_decode_32_safe(p, end, len, bad);
 717         if (len > 0) {
 718                 dout("apply_incremental full map len %d, %p to %p\n",
 719                      len, *p, end);
 720                 return osdmap_decode(p, min(*p+len, end));
 721         }
 722
 723         /* new crush? */
 724         ceph_decode_32_safe(p, end, len, bad);
 725         if (len > 0) {
 726                 dout("apply_incremental new crush map len %d, %p to %p\n",
 727                      len, *p, end);
 728                 newcrush = crush_decode(*p, min(*p+len, end));
 729                 if (IS_ERR(newcrush))
 730                         return ERR_CAST(newcrush);
 731                 *p += len;
 732         }
 733
 734         /* new flags? */
 735         if (new_flags >= 0)
 736                 map->flags = new_flags;
 737         if (new_pool_max >= 0)
 738                 map->pool_max = new_pool_max;
 739
 740         ceph_decode_need(p, end, 5*sizeof(u32), bad);
 741
 742         /* new max? */
 743         max = ceph_decode_32(p);
 744         if (max >= 0) {
 745                 err = osdmap_set_max_osd(map, max);
 746                 if (err < 0)
 747                         goto bad;
 748         }
 749
 750         map->epoch++;
 751         map->modified = map->modified;
 752         if (newcrush) {
 753                 if (map->crush)
 754                         crush_destroy(map->crush);
 755                 map->crush = newcrush;
 756                 newcrush = NULL;
 757         }
 758
 759         /* new_pool */
 760         ceph_decode_32_safe(p, end, len, bad);
 761         while (len--) {
 762                 __u8 ev;
 763                 struct ceph_pg_pool_info *pi;
 764
 765                 ceph_decode_32_safe(p, end, pool, bad);
 766                 ceph_decode_need(p, end, 1 + sizeof(pi->v), bad);
 767                 ev = ceph_decode_8(p);  /* encoding version */
 768                 if (ev > CEPH_PG_POOL_VERSION) {
 769                         pr_warning("got unknown v %d > %d of ceph_pg_pool\n",
 770                                    ev, CEPH_PG_POOL_VERSION);
 771                         goto bad;
 772                 }
 773                 pi = __lookup_pg_pool(&map->pg_pools, pool);
 774                 if (!pi) {
 775                         pi = kzalloc(sizeof(*pi), GFP_NOFS);
 776                         if (!pi) {
 777                                 err = -ENOMEM;
 778                                 goto bad;
 779                         }
 780                         pi->id = pool;
 781                         __insert_pg_pool(&map->pg_pools, pi);
 782                 }
 783                 err = __decode_pool(p, end, pi);
 784                 if (err < 0)
 785                         goto bad;
 786         }
 787         if (version >= 5 && __decode_pool_names(p, end, map) < 0)
 788                 goto bad;
 789
 790         /* old_pool */
 791         ceph_decode_32_safe(p, end, len, bad);
 792         while (len--) {
 793                 struct ceph_pg_pool_info *pi;
 794
 795                 ceph_decode_32_safe(p, end, pool, bad);
 796                 pi = __lookup_pg_pool(&map->pg_pools, pool);
 797                 if (pi)
 798                         __remove_pg_pool(&map->pg_pools, pi);
 799         }
 800
 801         /* new_up */
 802         err = -EINVAL;
 803         ceph_decode_32_safe(p, end, len, bad);
 804         while (len--) {
 805                 u32 osd;
 806                 struct ceph_entity_addr addr;
 807                 ceph_decode_32_safe(p, end, osd, bad);
 808                 ceph_decode_copy_safe(p, end, &addr, sizeof(addr), bad);
 809                 ceph_decode_addr(&addr);
 810                 pr_info("osd%d up\n", osd);
 811                 BUG_ON(osd >= map->max_osd);
 812                 map->osd_state[osd] |= CEPH_OSD_UP;
 813                 map->osd_addr[osd] = addr;
 814         }
 815
 816         /* new_down */
 817         ceph_decode_32_safe(p, end, len, bad);
 818         while (len--) {
 819                 u32 osd;
 820                 ceph_decode_32_safe(p, end, osd, bad);
 821                 (*p)++;  /* clean flag */
 822                 pr_info("osd%d down\n", osd);
 823                 if (osd < map->max_osd)
 824                         map->osd_state[osd] &= ~CEPH_OSD_UP;
 825         }
 826
 827         /* new_weight */
 828         ceph_decode_32_safe(p, end, len, bad);
 829         while (len--) {
 830                 u32 osd, off;
 831                 ceph_decode_need(p, end, sizeof(u32)*2, bad);
 832                 osd = ceph_decode_32(p);
 833                 off = ceph_decode_32(p);
 834                 pr_info("osd%d weight 0x%x %s\n", osd, off,
 835                      off == CEPH_OSD_IN ? "(in)" :
 836                      (off == CEPH_OSD_OUT ? "(out)" : ""));
 837                 if (osd < map->max_osd)
 838                         map->osd_weight[osd] = off;
 839         }
 840
 841         /* new_pg_temp */
 842         rbp = rb_first(&map->pg_temp);
 843         ceph_decode_32_safe(p, end, len, bad);
 844         while (len--) {
 845                 struct ceph_pg_mapping *pg;
 846                 int j;
 847                 struct ceph_pg pgid;
 848                 u32 pglen;
 849                 ceph_decode_need(p, end, sizeof(u64) + sizeof(u32), bad);
 850                 ceph_decode_copy(p, &pgid, sizeof(pgid));
 851                 pglen = ceph_decode_32(p);
 852
 853                 /* remove any? */
 854                 while (rbp && pgid_cmp(rb_entry(rbp, struct ceph_pg_mapping,
 855                                                 node)->pgid, pgid) <= 0) {
 856                         struct ceph_pg_mapping *cur =
 857                                 rb_entry(rbp, struct ceph_pg_mapping, node);
 858
 859                         rbp = rb_next(rbp);
 860                         dout(" removed pg_temp %llx\n", *(u64 *)&cur->pgid);
 861                         rb_erase(&cur->node, &map->pg_temp);
 862                         kfree(cur);
 863                 }
 864
 865                 if (pglen) {
 866                         /* insert */
 867                         ceph_decode_need(p, end, pglen*sizeof(u32), bad);
 868                         pg = kmalloc(sizeof(*pg) + sizeof(u32)*pglen, GFP_NOFS);
 869                         if (!pg) {
 870                                 err = -ENOMEM;
 871                                 goto bad;
 872                         }
 873                         pg->pgid = pgid;
 874                         pg->len = pglen;
 875                         for (j = 0; j < pglen; j++)
 876                                 pg->osds[j] = ceph_decode_32(p);
 877                         err = __insert_pg_mapping(pg, &map->pg_temp);
 878                         if (err) {
 879                                 kfree(pg);
 880                                 goto bad;
 881                         }
 882                         dout(" added pg_temp %llx len %d\n", *(u64 *)&pgid,
 883                              pglen);
 884                 }
 885         }
 886         while (rbp) {
 887                 struct ceph_pg_mapping *cur =
 888                         rb_entry(rbp, struct ceph_pg_mapping, node);
 889
 890                 rbp = rb_next(rbp);
 891                 dout(" removed pg_temp %llx\n", *(u64 *)&cur->pgid);
 892                 rb_erase(&cur->node, &map->pg_temp);
 893                 kfree(cur);
 894         }
 895
 896         /* ignore the rest */
 897         *p = end;
 898         return map;
 899
 900 bad:
 901         pr_err("corrupt inc osdmap epoch %d off %d (%p of %p-%p)\n",
 902                epoch, (int)(*p - start), *p, start, end);
 903         print_hex_dump(KERN_DEBUG, "osdmap: ",
 904                        DUMP_PREFIX_OFFSET, 16, 1,
 905                        start, end - start, true);
 906         if (newcrush)
 907                 crush_destroy(newcrush);
 908         return ERR_PTR(err);
 909 }
 910
 911
 912
 913
 914 /*
 915  * calculate file layout from given offset, length.
 916  * fill in correct oid, logical length, and object extent
 917  * offset, length.
 918  *
 919  * for now, we write only a single su, until we can
 920  * pass a stride back to the caller.
 921  */
 922 void ceph_calc_file_object_mapping(struct ceph_file_layout *layout,
 923                                    u64 off, u64 *plen,
 924                                    u64 *ono,
 925                                    u64 *oxoff, u64 *oxlen)
 926 {
 927         u32 osize = le32_to_cpu(layout->fl_object_size);
 928         u32 su = le32_to_cpu(layout->fl_stripe_unit);
 929         u32 sc = le32_to_cpu(layout->fl_stripe_count);
 930         u32 bl, stripeno, stripepos, objsetno;
 931         u32 su_per_object;
 932         u64 t, su_offset;
 933
 934         dout("mapping %llu~%llu  osize %u fl_su %u\n", off, *plen,
 935              osize, su);
 936         su_per_object = osize / su;
 937         dout("osize %u / su %u = su_per_object %u\n", osize, su,
 938              su_per_object);
 939
 940         BUG_ON((su & ~PAGE_MASK) != 0);
 941         /* bl = *off / su; */
 942         t = off;
 943         do_div(t, su);
 944         bl = t;
 945         dout("off %llu / su %u = bl %u\n", off, su, bl);
 946
 947         stripeno = bl / sc;
 948         stripepos = bl % sc;
 949         objsetno = stripeno / su_per_object;
 950
 951         *ono = objsetno * sc + stripepos;
 952         dout("objset %u * sc %u = ono %u\n", objsetno, sc, (unsigned)*ono);
 953
 954         /* *oxoff = *off % layout->fl_stripe_unit;  # offset in su */
 955         t = off;
 956         su_offset = do_div(t, su);
 957         *oxoff = su_offset + (stripeno % su_per_object) * su;
 958
 959         /*
 960          * Calculate the length of the extent being written to the selected
 961          * object. This is the minimum of the full length requested (plen) or
 962          * the remainder of the current stripe being written to.
 963          */
 964         *oxlen = min_t(u64, *plen, su - su_offset);
 965         *plen = *oxlen;
 966
 967         dout(" obj extent %llu~%llu\n", *oxoff, *oxlen);
 968 }
 969
 970 /*
 971  * calculate an object layout (i.e. pgid) from an oid,
 972  * file_layout, and osdmap
 973  */
 974 int ceph_calc_object_layout(struct ceph_object_layout *ol,
 975                             const char *oid,
 976                             struct ceph_file_layout *fl,
 977                             struct ceph_osdmap *osdmap)
 978 {
 979         unsigned num, num_mask;
 980         struct ceph_pg pgid;
 981         s32 preferred = (s32)le32_to_cpu(fl->fl_pg_preferred);
 982         int poolid = le32_to_cpu(fl->fl_pg_pool);
 983         struct ceph_pg_pool_info *pool;
 984         unsigned ps;
 985
 986         BUG_ON(!osdmap);
 987
 988         pool = __lookup_pg_pool(&osdmap->pg_pools, poolid);
 989         if (!pool)
 990                 return -EIO;
 991         ps = ceph_str_hash(pool->v.object_hash, oid, strlen(oid));
 992         if (preferred >= 0) {
 993                 ps += preferred;
 994                 num = le32_to_cpu(pool->v.lpg_num);
 995                 num_mask = pool->lpg_num_mask;
 996         } else {
 997                 num = le32_to_cpu(pool->v.pg_num);
 998                 num_mask = pool->pg_num_mask;
 999         }
1000
1001         pgid.ps = cpu_to_le16(ps);
1002         pgid.preferred = cpu_to_le16(preferred);
1003         pgid.pool = fl->fl_pg_pool;
1004         if (preferred >= 0)
1005                 dout("calc_object_layout '%s' pgid %d.%xp%d\n", oid, poolid, ps,
1006                      (int)preferred);
1007         else
1008                 dout("calc_object_layout '%s' pgid %d.%x\n", oid, poolid, ps);
1009
1010         ol->ol_pgid = pgid;
1011         ol->ol_stripe_unit = fl->fl_object_stripe_unit;
1012         return 0;
1013 }
1014
1015 /*
1016  * Calculate raw osd vector for the given pgid.  Return pointer to osd
1017  * array, or NULL on failure.
1018  */
1019 static int *calc_pg_raw(struct ceph_osdmap *osdmap, struct ceph_pg pgid,
1020                         int *osds, int *num)
1021 {
1022         struct ceph_pg_mapping *pg;
1023         struct ceph_pg_pool_info *pool;
1024         int ruleno;
1025         unsigned poolid, ps, pps;
1026         int preferred;
1027
1028         /* pg_temp? */
1029         pg = __lookup_pg_mapping(&osdmap->pg_temp, pgid);
1030         if (pg) {
1031                 *num = pg->len;
1032                 return pg->osds;
1033         }
1034
1035         /* crush */
1036         poolid = le32_to_cpu(pgid.pool);
1037         ps = le16_to_cpu(pgid.ps);
1038         preferred = (s16)le16_to_cpu(pgid.preferred);
1039
1040         /* don't forcefeed bad device ids to crush */
1041         if (preferred >= osdmap->max_osd ||
1042             preferred >= osdmap->crush->max_devices)
1043                 preferred = -1;
1044
1045         pool = __lookup_pg_pool(&osdmap->pg_pools, poolid);
1046         if (!pool)
1047                 return NULL;
1048         ruleno = crush_find_rule(osdmap->crush, pool->v.crush_ruleset,
1049                                  pool->v.type, pool->v.size);
1050         if (ruleno < 0) {
1051                 pr_err("no crush rule pool %d ruleset %d type %d size %d\n",
1052                        poolid, pool->v.crush_ruleset, pool->v.type,
1053                        pool->v.size);
1054                 return NULL;
1055         }
1056
1057         if (preferred >= 0)
1058                 pps = ceph_stable_mod(ps,
1059                                       le32_to_cpu(pool->v.lpgp_num),
1060                                       pool->lpgp_num_mask);
1061         else
1062                 pps = ceph_stable_mod(ps,
1063                                       le32_to_cpu(pool->v.pgp_num),
1064                                       pool->pgp_num_mask);
1065         pps += poolid;
1066         *num = crush_do_rule(osdmap->crush, ruleno, pps, osds,
1067                              min_t(int, pool->v.size, *num),
1068                              preferred, osdmap->osd_weight);
1069         return osds;
1070 }
1071
1072 /*
1073  * Return acting set for given pgid.
1074  */
1075 int ceph_calc_pg_acting(struct ceph_osdmap *osdmap, struct ceph_pg pgid,
1076                         int *acting)
1077 {
1078         int rawosds[CEPH_PG_MAX_SIZE], *osds;
1079         int i, o, num = CEPH_PG_MAX_SIZE;
1080
1081         osds = calc_pg_raw(osdmap, pgid, rawosds, &num);
1082         if (!osds)
1083                 return -1;
1084
1085         /* primary is first up osd */
1086         o = 0;
1087         for (i = 0; i < num; i++)
1088                 if (ceph_osd_is_up(osdmap, osds[i]))
1089                         acting[o++] = osds[i];
1090         return o;
1091 }
1092
1093 /*
1094  * Return primary osd for given pgid, or -1 if none.
1095  */
1096 int ceph_calc_pg_primary(struct ceph_osdmap *osdmap, struct ceph_pg pgid)
1097 {
1098         int rawosds[CEPH_PG_MAX_SIZE], *osds;
1099         int i, num = CEPH_PG_MAX_SIZE;
1100
1101         osds = calc_pg_raw(osdmap, pgid, rawosds, &num);
1102         if (!osds)
1103                 return -1;
1104
1105         /* primary is first up osd */
1106         for (i = 0; i < num; i++)
1107                 if (ceph_osd_is_up(osdmap, osds[i]))
1108                         return osds[i];
1109         return -1;
1110 }