Sage Weil | f24e998 | 2009-10-06 11:31:10 -0700 | [diff] [blame] | 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 | */ |
Sage Weil | 991abb6 | 2009-10-08 22:22:37 -0700 | [diff] [blame^] | 369 | static int __insert_pg_mapping(struct ceph_pg_mapping *new, |
| 370 | struct rb_root *root) |
Sage Weil | f24e998 | 2009-10-06 11:31:10 -0700 | [diff] [blame] | 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 |
Sage Weil | 991abb6 | 2009-10-08 22:22:37 -0700 | [diff] [blame^] | 384 | return -EEXIST; |
Sage Weil | f24e998 | 2009-10-06 11:31:10 -0700 | [diff] [blame] | 385 | } |
| 386 | |
| 387 | rb_link_node(&new->node, parent, p); |
| 388 | rb_insert_color(&new->node, root); |
Sage Weil | 991abb6 | 2009-10-08 22:22:37 -0700 | [diff] [blame^] | 389 | return 0; |
Sage Weil | f24e998 | 2009-10-06 11:31:10 -0700 | [diff] [blame] | 390 | } |
| 391 | |
| 392 | /* |
| 393 | * decode a full map. |
| 394 | */ |
| 395 | struct ceph_osdmap *osdmap_decode(void **p, void *end) |
| 396 | { |
| 397 | struct ceph_osdmap *map; |
| 398 | u16 version; |
| 399 | u32 len, max, i; |
| 400 | int err = -EINVAL; |
| 401 | void *start = *p; |
| 402 | |
| 403 | dout("osdmap_decode %p to %p len %d\n", *p, end, (int)(end - *p)); |
| 404 | |
| 405 | map = kzalloc(sizeof(*map), GFP_NOFS); |
| 406 | if (map == NULL) |
| 407 | return ERR_PTR(-ENOMEM); |
| 408 | map->pg_temp = RB_ROOT; |
| 409 | |
| 410 | ceph_decode_16_safe(p, end, version, bad); |
| 411 | |
| 412 | ceph_decode_need(p, end, 2*sizeof(u64)+6*sizeof(u32), bad); |
| 413 | ceph_decode_copy(p, &map->fsid, sizeof(map->fsid)); |
| 414 | ceph_decode_32(p, map->epoch); |
| 415 | ceph_decode_copy(p, &map->created, sizeof(map->created)); |
| 416 | ceph_decode_copy(p, &map->modified, sizeof(map->modified)); |
| 417 | |
| 418 | ceph_decode_32(p, map->num_pools); |
| 419 | map->pg_pool = kcalloc(map->num_pools, sizeof(*map->pg_pool), |
| 420 | GFP_NOFS); |
| 421 | if (!map->pg_pool) { |
| 422 | err = -ENOMEM; |
| 423 | goto bad; |
| 424 | } |
| 425 | ceph_decode_32_safe(p, end, max, bad); |
| 426 | while (max--) { |
| 427 | ceph_decode_need(p, end, 4+sizeof(map->pg_pool->v), bad); |
| 428 | ceph_decode_32(p, i); |
| 429 | if (i >= map->num_pools) |
| 430 | goto bad; |
| 431 | ceph_decode_copy(p, &map->pg_pool[i].v, |
| 432 | sizeof(map->pg_pool->v)); |
| 433 | calc_pg_masks(&map->pg_pool[i]); |
| 434 | p += le32_to_cpu(map->pg_pool[i].v.num_snaps) * sizeof(u64); |
| 435 | p += le32_to_cpu(map->pg_pool[i].v.num_removed_snap_intervals) |
| 436 | * sizeof(u64) * 2; |
| 437 | } |
| 438 | |
| 439 | ceph_decode_32_safe(p, end, map->flags, bad); |
| 440 | |
| 441 | ceph_decode_32(p, max); |
| 442 | |
| 443 | /* (re)alloc osd arrays */ |
| 444 | err = osdmap_set_max_osd(map, max); |
| 445 | if (err < 0) |
| 446 | goto bad; |
| 447 | dout("osdmap_decode max_osd = %d\n", map->max_osd); |
| 448 | |
| 449 | /* osds */ |
| 450 | err = -EINVAL; |
| 451 | ceph_decode_need(p, end, 3*sizeof(u32) + |
| 452 | map->max_osd*(1 + sizeof(*map->osd_weight) + |
| 453 | sizeof(*map->osd_addr)), bad); |
| 454 | *p += 4; /* skip length field (should match max) */ |
| 455 | ceph_decode_copy(p, map->osd_state, map->max_osd); |
| 456 | |
| 457 | *p += 4; /* skip length field (should match max) */ |
| 458 | for (i = 0; i < map->max_osd; i++) |
| 459 | ceph_decode_32(p, map->osd_weight[i]); |
| 460 | |
| 461 | *p += 4; /* skip length field (should match max) */ |
| 462 | ceph_decode_copy(p, map->osd_addr, map->max_osd*sizeof(*map->osd_addr)); |
| 463 | |
| 464 | /* pg_temp */ |
| 465 | ceph_decode_32_safe(p, end, len, bad); |
| 466 | for (i = 0; i < len; i++) { |
| 467 | int n, j; |
| 468 | u64 pgid; |
| 469 | struct ceph_pg_mapping *pg; |
| 470 | |
| 471 | ceph_decode_need(p, end, sizeof(u32) + sizeof(u64), bad); |
| 472 | ceph_decode_64(p, pgid); |
| 473 | ceph_decode_32(p, n); |
| 474 | ceph_decode_need(p, end, n * sizeof(u32), bad); |
| 475 | pg = kmalloc(sizeof(*pg) + n*sizeof(u32), GFP_NOFS); |
| 476 | if (!pg) { |
| 477 | err = -ENOMEM; |
| 478 | goto bad; |
| 479 | } |
| 480 | pg->pgid = pgid; |
| 481 | pg->len = n; |
| 482 | for (j = 0; j < n; j++) |
| 483 | ceph_decode_32(p, pg->osds[j]); |
| 484 | |
Sage Weil | 991abb6 | 2009-10-08 22:22:37 -0700 | [diff] [blame^] | 485 | err = __insert_pg_mapping(pg, &map->pg_temp); |
| 486 | if (err) |
| 487 | goto bad; |
Sage Weil | f24e998 | 2009-10-06 11:31:10 -0700 | [diff] [blame] | 488 | dout(" added pg_temp %llx len %d\n", pgid, len); |
| 489 | } |
| 490 | |
| 491 | /* crush */ |
| 492 | ceph_decode_32_safe(p, end, len, bad); |
| 493 | dout("osdmap_decode crush len %d from off 0x%x\n", len, |
| 494 | (int)(*p - start)); |
| 495 | ceph_decode_need(p, end, len, bad); |
| 496 | map->crush = crush_decode(*p, end); |
| 497 | *p += len; |
| 498 | if (IS_ERR(map->crush)) { |
| 499 | err = PTR_ERR(map->crush); |
| 500 | map->crush = NULL; |
| 501 | goto bad; |
| 502 | } |
| 503 | |
| 504 | /* ignore the rest of the map */ |
| 505 | *p = end; |
| 506 | |
| 507 | dout("osdmap_decode done %p %p\n", *p, end); |
| 508 | return map; |
| 509 | |
| 510 | bad: |
| 511 | dout("osdmap_decode fail\n"); |
| 512 | ceph_osdmap_destroy(map); |
| 513 | return ERR_PTR(err); |
| 514 | } |
| 515 | |
| 516 | /* |
| 517 | * decode and apply an incremental map update. |
| 518 | */ |
| 519 | struct ceph_osdmap *osdmap_apply_incremental(void **p, void *end, |
| 520 | struct ceph_osdmap *map, |
| 521 | struct ceph_messenger *msgr) |
| 522 | { |
| 523 | struct ceph_osdmap *newmap = map; |
| 524 | struct crush_map *newcrush = NULL; |
| 525 | struct ceph_fsid fsid; |
| 526 | u32 epoch = 0; |
| 527 | struct ceph_timespec modified; |
| 528 | u32 len, pool; |
| 529 | __s32 new_flags, max; |
| 530 | void *start = *p; |
| 531 | int err = -EINVAL; |
| 532 | u16 version; |
| 533 | struct rb_node *rbp; |
| 534 | |
| 535 | ceph_decode_16_safe(p, end, version, bad); |
| 536 | |
| 537 | ceph_decode_need(p, end, sizeof(fsid)+sizeof(modified)+2*sizeof(u32), |
| 538 | bad); |
| 539 | ceph_decode_copy(p, &fsid, sizeof(fsid)); |
| 540 | ceph_decode_32(p, epoch); |
| 541 | BUG_ON(epoch != map->epoch+1); |
| 542 | ceph_decode_copy(p, &modified, sizeof(modified)); |
| 543 | ceph_decode_32(p, new_flags); |
| 544 | |
| 545 | /* full map? */ |
| 546 | ceph_decode_32_safe(p, end, len, bad); |
| 547 | if (len > 0) { |
| 548 | dout("apply_incremental full map len %d, %p to %p\n", |
| 549 | len, *p, end); |
| 550 | newmap = osdmap_decode(p, min(*p+len, end)); |
| 551 | return newmap; /* error or not */ |
| 552 | } |
| 553 | |
| 554 | /* new crush? */ |
| 555 | ceph_decode_32_safe(p, end, len, bad); |
| 556 | if (len > 0) { |
| 557 | dout("apply_incremental new crush map len %d, %p to %p\n", |
| 558 | len, *p, end); |
| 559 | newcrush = crush_decode(*p, min(*p+len, end)); |
| 560 | if (IS_ERR(newcrush)) |
| 561 | return ERR_PTR(PTR_ERR(newcrush)); |
| 562 | } |
| 563 | |
| 564 | /* new flags? */ |
| 565 | if (new_flags >= 0) |
| 566 | map->flags = new_flags; |
| 567 | |
| 568 | ceph_decode_need(p, end, 5*sizeof(u32), bad); |
| 569 | |
| 570 | /* new max? */ |
| 571 | ceph_decode_32(p, max); |
| 572 | if (max >= 0) { |
| 573 | err = osdmap_set_max_osd(map, max); |
| 574 | if (err < 0) |
| 575 | goto bad; |
| 576 | } |
| 577 | |
| 578 | map->epoch++; |
| 579 | map->modified = map->modified; |
| 580 | if (newcrush) { |
| 581 | if (map->crush) |
| 582 | crush_destroy(map->crush); |
| 583 | map->crush = newcrush; |
| 584 | newcrush = NULL; |
| 585 | } |
| 586 | |
| 587 | /* new_pool */ |
| 588 | ceph_decode_32_safe(p, end, len, bad); |
| 589 | while (len--) { |
| 590 | ceph_decode_32_safe(p, end, pool, bad); |
| 591 | if (pool >= map->num_pools) { |
| 592 | void *pg_pool = kcalloc(pool + 1, |
| 593 | sizeof(*map->pg_pool), |
| 594 | GFP_NOFS); |
| 595 | if (!pg_pool) { |
| 596 | err = -ENOMEM; |
| 597 | goto bad; |
| 598 | } |
| 599 | memcpy(pg_pool, map->pg_pool, |
| 600 | map->num_pools * sizeof(*map->pg_pool)); |
| 601 | kfree(map->pg_pool); |
| 602 | map->pg_pool = pg_pool; |
| 603 | map->num_pools = pool+1; |
| 604 | } |
| 605 | ceph_decode_copy(p, &map->pg_pool[pool].v, |
| 606 | sizeof(map->pg_pool->v)); |
| 607 | calc_pg_masks(&map->pg_pool[pool]); |
| 608 | } |
| 609 | |
| 610 | /* old_pool (ignore) */ |
| 611 | ceph_decode_32_safe(p, end, len, bad); |
| 612 | *p += len * sizeof(u32); |
| 613 | |
| 614 | /* new_up */ |
| 615 | err = -EINVAL; |
| 616 | ceph_decode_32_safe(p, end, len, bad); |
| 617 | while (len--) { |
| 618 | u32 osd; |
| 619 | struct ceph_entity_addr addr; |
| 620 | ceph_decode_32_safe(p, end, osd, bad); |
| 621 | ceph_decode_copy_safe(p, end, &addr, sizeof(addr), bad); |
| 622 | pr_info("osd%d up\n", osd); |
| 623 | BUG_ON(osd >= map->max_osd); |
| 624 | map->osd_state[osd] |= CEPH_OSD_UP; |
| 625 | map->osd_addr[osd] = addr; |
| 626 | } |
| 627 | |
| 628 | /* new_down */ |
| 629 | ceph_decode_32_safe(p, end, len, bad); |
| 630 | while (len--) { |
| 631 | u32 osd; |
| 632 | ceph_decode_32_safe(p, end, osd, bad); |
| 633 | (*p)++; /* clean flag */ |
| 634 | pr_info("ceph osd%d down\n", osd); |
| 635 | if (osd < map->max_osd) |
| 636 | map->osd_state[osd] &= ~CEPH_OSD_UP; |
| 637 | } |
| 638 | |
| 639 | /* new_weight */ |
| 640 | ceph_decode_32_safe(p, end, len, bad); |
| 641 | while (len--) { |
| 642 | u32 osd, off; |
| 643 | ceph_decode_need(p, end, sizeof(u32)*2, bad); |
| 644 | ceph_decode_32(p, osd); |
| 645 | ceph_decode_32(p, off); |
| 646 | pr_info("osd%d weight 0x%x %s\n", osd, off, |
| 647 | off == CEPH_OSD_IN ? "(in)" : |
| 648 | (off == CEPH_OSD_OUT ? "(out)" : "")); |
| 649 | if (osd < map->max_osd) |
| 650 | map->osd_weight[osd] = off; |
| 651 | } |
| 652 | |
| 653 | /* new_pg_temp */ |
| 654 | rbp = rb_first(&map->pg_temp); |
| 655 | ceph_decode_32_safe(p, end, len, bad); |
| 656 | while (len--) { |
| 657 | struct ceph_pg_mapping *pg; |
| 658 | int j; |
| 659 | u64 pgid; |
| 660 | u32 pglen; |
| 661 | ceph_decode_need(p, end, sizeof(u64) + sizeof(u32), bad); |
| 662 | ceph_decode_64(p, pgid); |
| 663 | ceph_decode_32(p, pglen); |
| 664 | |
| 665 | /* remove any? */ |
| 666 | while (rbp && rb_entry(rbp, struct ceph_pg_mapping, |
| 667 | node)->pgid <= pgid) { |
| 668 | struct rb_node *cur = rbp; |
| 669 | rbp = rb_next(rbp); |
| 670 | dout(" removed pg_temp %llx\n", |
| 671 | rb_entry(cur, struct ceph_pg_mapping, node)->pgid); |
| 672 | rb_erase(cur, &map->pg_temp); |
| 673 | } |
| 674 | |
| 675 | if (pglen) { |
| 676 | /* insert */ |
| 677 | ceph_decode_need(p, end, pglen*sizeof(u32), bad); |
| 678 | pg = kmalloc(sizeof(*pg) + sizeof(u32)*pglen, GFP_NOFS); |
| 679 | if (!pg) { |
| 680 | err = -ENOMEM; |
| 681 | goto bad; |
| 682 | } |
| 683 | pg->pgid = pgid; |
| 684 | pg->len = pglen; |
| 685 | for (j = 0; j < len; j++) |
| 686 | ceph_decode_32(p, pg->osds[j]); |
Sage Weil | 991abb6 | 2009-10-08 22:22:37 -0700 | [diff] [blame^] | 687 | err = __insert_pg_mapping(pg, &map->pg_temp); |
| 688 | if (err) |
| 689 | goto bad; |
Sage Weil | f24e998 | 2009-10-06 11:31:10 -0700 | [diff] [blame] | 690 | dout(" added pg_temp %llx len %d\n", pgid, pglen); |
| 691 | } |
| 692 | } |
| 693 | while (rbp) { |
| 694 | struct rb_node *cur = rbp; |
| 695 | rbp = rb_next(rbp); |
| 696 | dout(" removed pg_temp %llx\n", |
| 697 | rb_entry(cur, struct ceph_pg_mapping, node)->pgid); |
| 698 | rb_erase(cur, &map->pg_temp); |
| 699 | } |
| 700 | |
| 701 | /* ignore the rest */ |
| 702 | *p = end; |
| 703 | return map; |
| 704 | |
| 705 | bad: |
| 706 | pr_err("corrupt inc osdmap epoch %d off %d (%p of %p-%p)\n", |
| 707 | epoch, (int)(*p - start), *p, start, end); |
| 708 | if (newcrush) |
| 709 | crush_destroy(newcrush); |
| 710 | return ERR_PTR(err); |
| 711 | } |
| 712 | |
| 713 | |
| 714 | |
| 715 | |
| 716 | /* |
| 717 | * calculate file layout from given offset, length. |
| 718 | * fill in correct oid, logical length, and object extent |
| 719 | * offset, length. |
| 720 | * |
| 721 | * for now, we write only a single su, until we can |
| 722 | * pass a stride back to the caller. |
| 723 | */ |
| 724 | void ceph_calc_file_object_mapping(struct ceph_file_layout *layout, |
| 725 | u64 off, u64 *plen, |
| 726 | u64 *bno, |
| 727 | u64 *oxoff, u64 *oxlen) |
| 728 | { |
| 729 | u32 osize = le32_to_cpu(layout->fl_object_size); |
| 730 | u32 su = le32_to_cpu(layout->fl_stripe_unit); |
| 731 | u32 sc = le32_to_cpu(layout->fl_stripe_count); |
| 732 | u32 bl, stripeno, stripepos, objsetno; |
| 733 | u32 su_per_object; |
| 734 | u64 t; |
| 735 | |
| 736 | dout("mapping %llu~%llu osize %u fl_su %u\n", off, *plen, |
| 737 | osize, su); |
| 738 | su_per_object = osize / le32_to_cpu(layout->fl_stripe_unit); |
| 739 | dout("osize %u / su %u = su_per_object %u\n", osize, su, |
| 740 | su_per_object); |
| 741 | |
| 742 | BUG_ON((su & ~PAGE_MASK) != 0); |
| 743 | /* bl = *off / su; */ |
| 744 | t = off; |
| 745 | do_div(t, su); |
| 746 | bl = t; |
| 747 | dout("off %llu / su %u = bl %u\n", off, su, bl); |
| 748 | |
| 749 | stripeno = bl / sc; |
| 750 | stripepos = bl % sc; |
| 751 | objsetno = stripeno / su_per_object; |
| 752 | |
| 753 | *bno = objsetno * sc + stripepos; |
| 754 | dout("objset %u * sc %u = bno %u\n", objsetno, sc, (unsigned)*bno); |
| 755 | /* *oxoff = *off / layout->fl_stripe_unit; */ |
| 756 | t = off; |
| 757 | *oxoff = do_div(t, su); |
| 758 | *oxlen = min_t(u64, *plen, su - *oxoff); |
| 759 | *plen = *oxlen; |
| 760 | |
| 761 | dout(" obj extent %llu~%llu\n", *oxoff, *oxlen); |
| 762 | } |
| 763 | |
| 764 | /* |
| 765 | * calculate an object layout (i.e. pgid) from an oid, |
| 766 | * file_layout, and osdmap |
| 767 | */ |
| 768 | int ceph_calc_object_layout(struct ceph_object_layout *ol, |
| 769 | const char *oid, |
| 770 | struct ceph_file_layout *fl, |
| 771 | struct ceph_osdmap *osdmap) |
| 772 | { |
| 773 | unsigned num, num_mask; |
| 774 | union ceph_pg pgid; |
| 775 | s32 preferred = (s32)le32_to_cpu(fl->fl_pg_preferred); |
| 776 | int poolid = le32_to_cpu(fl->fl_pg_pool); |
| 777 | struct ceph_pg_pool_info *pool; |
| 778 | |
| 779 | if (poolid >= osdmap->num_pools) |
| 780 | return -EIO; |
| 781 | pool = &osdmap->pg_pool[poolid]; |
| 782 | |
| 783 | if (preferred >= 0) { |
| 784 | num = le32_to_cpu(pool->v.lpg_num); |
| 785 | num_mask = pool->lpg_num_mask; |
| 786 | } else { |
| 787 | num = le32_to_cpu(pool->v.pg_num); |
| 788 | num_mask = pool->pg_num_mask; |
| 789 | } |
| 790 | |
| 791 | pgid.pg64 = 0; /* start with it zeroed out */ |
| 792 | pgid.pg.ps = ceph_full_name_hash(oid, strlen(oid)); |
| 793 | pgid.pg.preferred = preferred; |
| 794 | pgid.pg.pool = le32_to_cpu(fl->fl_pg_pool); |
| 795 | if (preferred >= 0) |
| 796 | dout("calc_object_layout '%s' pgid %d.%xp%d (%llx)\n", oid, |
| 797 | pgid.pg.pool, pgid.pg.ps, (int)preferred, pgid.pg64); |
| 798 | else |
| 799 | dout("calc_object_layout '%s' pgid %d.%x (%llx)\n", oid, |
| 800 | pgid.pg.pool, pgid.pg.ps, pgid.pg64); |
| 801 | |
| 802 | ol->ol_pgid = cpu_to_le64(pgid.pg64); |
| 803 | ol->ol_stripe_unit = fl->fl_object_stripe_unit; |
| 804 | |
| 805 | return 0; |
| 806 | } |
| 807 | |
| 808 | /* |
| 809 | * Calculate raw osd vector for the given pgid. Return pointer to osd |
| 810 | * array, or NULL on failure. |
| 811 | */ |
| 812 | static int *calc_pg_raw(struct ceph_osdmap *osdmap, union ceph_pg pgid, |
| 813 | int *osds, int *num) |
| 814 | { |
| 815 | struct rb_node *n = osdmap->pg_temp.rb_node; |
| 816 | struct ceph_pg_mapping *pg; |
| 817 | struct ceph_pg_pool_info *pool; |
| 818 | int ruleno; |
| 819 | unsigned pps; /* placement ps */ |
| 820 | |
| 821 | /* pg_temp? */ |
| 822 | while (n) { |
| 823 | pg = rb_entry(n, struct ceph_pg_mapping, node); |
| 824 | if (pgid.pg64 < pg->pgid) |
| 825 | n = n->rb_left; |
| 826 | else if (pgid.pg64 > pg->pgid) |
| 827 | n = n->rb_right; |
| 828 | else { |
| 829 | *num = pg->len; |
| 830 | return pg->osds; |
| 831 | } |
| 832 | } |
| 833 | |
| 834 | /* crush */ |
| 835 | if (pgid.pg.pool >= osdmap->num_pools) |
| 836 | return NULL; |
| 837 | pool = &osdmap->pg_pool[pgid.pg.pool]; |
| 838 | ruleno = crush_find_rule(osdmap->crush, pool->v.crush_ruleset, |
| 839 | pool->v.type, pool->v.size); |
| 840 | if (ruleno < 0) { |
| 841 | pr_err("no crush rule pool %d type %d size %d\n", |
| 842 | pgid.pg.pool, pool->v.type, pool->v.size); |
| 843 | return NULL; |
| 844 | } |
| 845 | |
| 846 | if (pgid.pg.preferred >= 0) |
| 847 | pps = ceph_stable_mod(pgid.pg.ps, |
| 848 | le32_to_cpu(pool->v.lpgp_num), |
| 849 | pool->lpgp_num_mask); |
| 850 | else |
| 851 | pps = ceph_stable_mod(pgid.pg.ps, |
| 852 | le32_to_cpu(pool->v.pgp_num), |
| 853 | pool->pgp_num_mask); |
| 854 | pps += pgid.pg.pool; |
| 855 | *num = crush_do_rule(osdmap->crush, ruleno, pps, osds, |
| 856 | min_t(int, pool->v.size, *num), |
| 857 | pgid.pg.preferred, osdmap->osd_weight); |
| 858 | return osds; |
| 859 | } |
| 860 | |
| 861 | /* |
| 862 | * Return primary osd for given pgid, or -1 if none. |
| 863 | */ |
| 864 | int ceph_calc_pg_primary(struct ceph_osdmap *osdmap, union ceph_pg pgid) |
| 865 | { |
| 866 | int rawosds[10], *osds; |
| 867 | int i, num = ARRAY_SIZE(rawosds); |
| 868 | |
| 869 | osds = calc_pg_raw(osdmap, pgid, rawosds, &num); |
| 870 | if (!osds) |
| 871 | return -1; |
| 872 | |
| 873 | /* primary is first up osd */ |
| 874 | for (i = 0; i < num; i++) |
| 875 | if (ceph_osd_is_up(osdmap, osds[i])) { |
| 876 | return osds[i]; |
| 877 | break; |
| 878 | } |
| 879 | return -1; |
| 880 | } |