1 /* 2 * Copyright (c) 2007-2013 Nicira, Inc. 3 * 4 * This program is free software; you can redistribute it and/or 5 * modify it under the terms of version 2 of the GNU General Public 6 * License as published by the Free Software Foundation. 7 * 8 * This program is distributed in the hope that it will be useful, but 9 * WITHOUT ANY WARRANTY; without even the implied warranty of 10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU 11 * General Public License for more details. 12 * 13 * You should have received a copy of the GNU General Public License 14 * along with this program; if not, write to the Free Software 15 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 16 * 02110-1301, USA 17 */ 18 19 #include "flow.h" 20 #include "datapath.h" 21 #include <linux/uaccess.h> 22 #include <linux/netdevice.h> 23 #include <linux/etherdevice.h> 24 #include <linux/if_ether.h> 25 #include <linux/if_vlan.h> 26 #include <net/llc_pdu.h> 27 #include <linux/kernel.h> 28 #include <linux/hash.h> 29 #include <linux/jiffies.h> 30 #include <linux/llc.h> 31 #include <linux/module.h> 32 #include <linux/in.h> 33 #include <linux/rcupdate.h> 34 #include <linux/if_arp.h> 35 #include <linux/ip.h> 36 #include <linux/ipv6.h> 37 #include <linux/sctp.h> 38 #include <linux/tcp.h> 39 #include <linux/udp.h> 40 #include <linux/icmp.h> 41 #include <linux/icmpv6.h> 42 #include <linux/rculist.h> 43 #include <net/ip.h> 44 #include <net/ipv6.h> 45 #include <net/ndisc.h> 46 47 #define TBL_MIN_BUCKETS 1024 48 #define REHASH_INTERVAL (10 * 60 * HZ) 49 50 static struct kmem_cache *flow_cache; 51 52 static u16 range_n_bytes(const struct sw_flow_key_range *range) 53 { 54 return range->end - range->start; 55 } 56 57 void ovs_flow_mask_key(struct sw_flow_key *dst, const struct sw_flow_key *src, 58 const struct sw_flow_mask *mask) 59 { 60 const long *m = (long *)((u8 *)&mask->key + mask->range.start); 61 const long *s = (long *)((u8 *)src + mask->range.start); 62 long *d = (long *)((u8 *)dst + mask->range.start); 63 int i; 64 65 /* The memory outside of the 'mask->range' are not set since 66 * further operations on 'dst' only uses contents within 67 * 'mask->range'. 68 */ 69 for (i = 0; i < range_n_bytes(&mask->range); i += sizeof(long)) 70 *d++ = *s++ & *m++; 71 } 72 73 struct sw_flow *ovs_flow_alloc(bool percpu_stats) 74 { 75 struct sw_flow *flow; 76 int cpu; 77 78 flow = kmem_cache_alloc(flow_cache, GFP_KERNEL); 79 if (!flow) 80 return ERR_PTR(-ENOMEM); 81 82 flow->sf_acts = NULL; 83 flow->mask = NULL; 84 85 flow->stats.is_percpu = percpu_stats; 86 87 if (!percpu_stats) { 88 flow->stats.stat = kzalloc(sizeof(*flow->stats.stat), GFP_KERNEL); 89 if (!flow->stats.stat) 90 goto err; 91 92 spin_lock_init(&flow->stats.stat->lock); 93 } else { 94 flow->stats.cpu_stats = alloc_percpu(struct flow_stats); 95 if (!flow->stats.cpu_stats) 96 goto err; 97 98 for_each_possible_cpu(cpu) { 99 struct flow_stats *cpu_stats; 100 101 cpu_stats = per_cpu_ptr(flow->stats.cpu_stats, cpu); 102 spin_lock_init(&cpu_stats->lock); 103 } 104 } 105 return flow; 106 err: 107 kmem_cache_free(flow_cache, flow); 108 return ERR_PTR(-ENOMEM); 109 } 110 111 int ovs_flow_tbl_count(struct flow_table *table) 112 { 113 return table->count; 114 } 115 116 static struct flex_array *alloc_buckets(unsigned int n_buckets) 117 { 118 struct flex_array *buckets; 119 int i, err; 120 121 buckets = flex_array_alloc(sizeof(struct hlist_head), 122 n_buckets, GFP_KERNEL); 123 if (!buckets) 124 return NULL; 125 126 err = flex_array_prealloc(buckets, 0, n_buckets, GFP_KERNEL); 127 if (err) { 128 flex_array_free(buckets); 129 return NULL; 130 } 131 132 for (i = 0; i < n_buckets; i++) 133 INIT_HLIST_HEAD((struct hlist_head *) 134 flex_array_get(buckets, i)); 135 136 return buckets; 137 } 138 139 static void flow_free(struct sw_flow *flow) 140 { 141 kfree((struct sf_flow_acts __force *)flow->sf_acts); 142 if (flow->stats.is_percpu) 143 free_percpu(flow->stats.cpu_stats); 144 else 145 kfree(flow->stats.stat); 146 kmem_cache_free(flow_cache, flow); 147 } 148 149 static void rcu_free_flow_callback(struct rcu_head *rcu) 150 { 151 struct sw_flow *flow = container_of(rcu, struct sw_flow, rcu); 152 153 flow_free(flow); 154 } 155 156 void ovs_flow_free(struct sw_flow *flow, bool deferred) 157 { 158 if (!flow) 159 return; 160 161 if (flow->mask) { 162 struct sw_flow_mask *mask = flow->mask; 163 164 /* ovs-lock is required to protect mask-refcount and 165 * mask list. 166 */ 167 ASSERT_OVSL(); 168 BUG_ON(!mask->ref_count); 169 mask->ref_count--; 170 171 if (!mask->ref_count) { 172 list_del_rcu(&mask->list); 173 if (deferred) 174 kfree_rcu(mask, rcu); 175 else 176 kfree(mask); 177 } 178 } 179 180 if (deferred) 181 call_rcu(&flow->rcu, rcu_free_flow_callback); 182 else 183 flow_free(flow); 184 } 185 186 static void free_buckets(struct flex_array *buckets) 187 { 188 flex_array_free(buckets); 189 } 190 191 192 static void __table_instance_destroy(struct table_instance *ti) 193 { 194 free_buckets(ti->buckets); 195 kfree(ti); 196 } 197 198 static struct table_instance *table_instance_alloc(int new_size) 199 { 200 struct table_instance *ti = kmalloc(sizeof(*ti), GFP_KERNEL); 201 202 if (!ti) 203 return NULL; 204 205 ti->buckets = alloc_buckets(new_size); 206 207 if (!ti->buckets) { 208 kfree(ti); 209 return NULL; 210 } 211 ti->n_buckets = new_size; 212 ti->node_ver = 0; 213 ti->keep_flows = false; 214 get_random_bytes(&ti->hash_seed, sizeof(u32)); 215 216 return ti; 217 } 218 219 int ovs_flow_tbl_init(struct flow_table *table) 220 { 221 struct table_instance *ti; 222 223 ti = table_instance_alloc(TBL_MIN_BUCKETS); 224 225 if (!ti) 226 return -ENOMEM; 227 228 rcu_assign_pointer(table->ti, ti); 229 INIT_LIST_HEAD(&table->mask_list); 230 table->last_rehash = jiffies; 231 table->count = 0; 232 return 0; 233 } 234 235 static void flow_tbl_destroy_rcu_cb(struct rcu_head *rcu) 236 { 237 struct table_instance *ti = container_of(rcu, struct table_instance, rcu); 238 239 __table_instance_destroy(ti); 240 } 241 242 static void table_instance_destroy(struct table_instance *ti, bool deferred) 243 { 244 int i; 245 246 if (!ti) 247 return; 248 249 if (ti->keep_flows) 250 goto skip_flows; 251 252 for (i = 0; i < ti->n_buckets; i++) { 253 struct sw_flow *flow; 254 struct hlist_head *head = flex_array_get(ti->buckets, i); 255 struct hlist_node *n; 256 int ver = ti->node_ver; 257 258 hlist_for_each_entry_safe(flow, n, head, hash_node[ver]) { 259 hlist_del_rcu(&flow->hash_node[ver]); 260 ovs_flow_free(flow, deferred); 261 } 262 } 263 264 skip_flows: 265 if (deferred) 266 call_rcu(&ti->rcu, flow_tbl_destroy_rcu_cb); 267 else 268 __table_instance_destroy(ti); 269 } 270 271 void ovs_flow_tbl_destroy(struct flow_table *table, bool deferred) 272 { 273 struct table_instance *ti = ovsl_dereference(table->ti); 274 275 table_instance_destroy(ti, deferred); 276 } 277 278 struct sw_flow *ovs_flow_tbl_dump_next(struct table_instance *ti, 279 u32 *bucket, u32 *last) 280 { 281 struct sw_flow *flow; 282 struct hlist_head *head; 283 int ver; 284 int i; 285 286 ver = ti->node_ver; 287 while (*bucket < ti->n_buckets) { 288 i = 0; 289 head = flex_array_get(ti->buckets, *bucket); 290 hlist_for_each_entry_rcu(flow, head, hash_node[ver]) { 291 if (i < *last) { 292 i++; 293 continue; 294 } 295 *last = i + 1; 296 return flow; 297 } 298 (*bucket)++; 299 *last = 0; 300 } 301 302 return NULL; 303 } 304 305 static struct hlist_head *find_bucket(struct table_instance *ti, u32 hash) 306 { 307 hash = jhash_1word(hash, ti->hash_seed); 308 return flex_array_get(ti->buckets, 309 (hash & (ti->n_buckets - 1))); 310 } 311 312 static void table_instance_insert(struct table_instance *ti, struct sw_flow *flow) 313 { 314 struct hlist_head *head; 315 316 head = find_bucket(ti, flow->hash); 317 hlist_add_head_rcu(&flow->hash_node[ti->node_ver], head); 318 } 319 320 static void flow_table_copy_flows(struct table_instance *old, 321 struct table_instance *new) 322 { 323 int old_ver; 324 int i; 325 326 old_ver = old->node_ver; 327 new->node_ver = !old_ver; 328 329 /* Insert in new table. */ 330 for (i = 0; i < old->n_buckets; i++) { 331 struct sw_flow *flow; 332 struct hlist_head *head; 333 334 head = flex_array_get(old->buckets, i); 335 336 hlist_for_each_entry(flow, head, hash_node[old_ver]) 337 table_instance_insert(new, flow); 338 } 339 340 old->keep_flows = true; 341 } 342 343 static struct table_instance *table_instance_rehash(struct table_instance *ti, 344 int n_buckets) 345 { 346 struct table_instance *new_ti; 347 348 new_ti = table_instance_alloc(n_buckets); 349 if (!new_ti) 350 return NULL; 351 352 flow_table_copy_flows(ti, new_ti); 353 354 return new_ti; 355 } 356 357 int ovs_flow_tbl_flush(struct flow_table *flow_table) 358 { 359 struct table_instance *old_ti; 360 struct table_instance *new_ti; 361 362 old_ti = ovsl_dereference(flow_table->ti); 363 new_ti = table_instance_alloc(TBL_MIN_BUCKETS); 364 if (!new_ti) 365 return -ENOMEM; 366 367 rcu_assign_pointer(flow_table->ti, new_ti); 368 flow_table->last_rehash = jiffies; 369 flow_table->count = 0; 370 371 table_instance_destroy(old_ti, true); 372 return 0; 373 } 374 375 static u32 flow_hash(const struct sw_flow_key *key, int key_start, 376 int key_end) 377 { 378 u32 *hash_key = (u32 *)((u8 *)key + key_start); 379 int hash_u32s = (key_end - key_start) >> 2; 380 381 /* Make sure number of hash bytes are multiple of u32. */ 382 BUILD_BUG_ON(sizeof(long) % sizeof(u32)); 383 384 return arch_fast_hash2(hash_key, hash_u32s, 0); 385 } 386 387 static int flow_key_start(const struct sw_flow_key *key) 388 { 389 if (key->tun_key.ipv4_dst) 390 return 0; 391 else 392 return rounddown(offsetof(struct sw_flow_key, phy), 393 sizeof(long)); 394 } 395 396 static bool cmp_key(const struct sw_flow_key *key1, 397 const struct sw_flow_key *key2, 398 int key_start, int key_end) 399 { 400 const long *cp1 = (long *)((u8 *)key1 + key_start); 401 const long *cp2 = (long *)((u8 *)key2 + key_start); 402 long diffs = 0; 403 int i; 404 405 for (i = key_start; i < key_end; i += sizeof(long)) 406 diffs |= *cp1++ ^ *cp2++; 407 408 return diffs == 0; 409 } 410 411 static bool flow_cmp_masked_key(const struct sw_flow *flow, 412 const struct sw_flow_key *key, 413 int key_start, int key_end) 414 { 415 return cmp_key(&flow->key, key, key_start, key_end); 416 } 417 418 bool ovs_flow_cmp_unmasked_key(const struct sw_flow *flow, 419 struct sw_flow_match *match) 420 { 421 struct sw_flow_key *key = match->key; 422 int key_start = flow_key_start(key); 423 int key_end = match->range.end; 424 425 return cmp_key(&flow->unmasked_key, key, key_start, key_end); 426 } 427 428 static struct sw_flow *masked_flow_lookup(struct table_instance *ti, 429 const struct sw_flow_key *unmasked, 430 struct sw_flow_mask *mask) 431 { 432 struct sw_flow *flow; 433 struct hlist_head *head; 434 int key_start = mask->range.start; 435 int key_end = mask->range.end; 436 u32 hash; 437 struct sw_flow_key masked_key; 438 439 ovs_flow_mask_key(&masked_key, unmasked, mask); 440 hash = flow_hash(&masked_key, key_start, key_end); 441 head = find_bucket(ti, hash); 442 hlist_for_each_entry_rcu(flow, head, hash_node[ti->node_ver]) { 443 if (flow->mask == mask && flow->hash == hash && 444 flow_cmp_masked_key(flow, &masked_key, 445 key_start, key_end)) 446 return flow; 447 } 448 return NULL; 449 } 450 451 struct sw_flow *ovs_flow_tbl_lookup_stats(struct flow_table *tbl, 452 const struct sw_flow_key *key, 453 u32 *n_mask_hit) 454 { 455 struct table_instance *ti = rcu_dereference_ovsl(tbl->ti); 456 struct sw_flow_mask *mask; 457 struct sw_flow *flow; 458 459 *n_mask_hit = 0; 460 list_for_each_entry_rcu(mask, &tbl->mask_list, list) { 461 (*n_mask_hit)++; 462 flow = masked_flow_lookup(ti, key, mask); 463 if (flow) /* Found */ 464 return flow; 465 } 466 return NULL; 467 } 468 469 struct sw_flow *ovs_flow_tbl_lookup(struct flow_table *tbl, 470 const struct sw_flow_key *key) 471 { 472 u32 __always_unused n_mask_hit; 473 474 return ovs_flow_tbl_lookup_stats(tbl, key, &n_mask_hit); 475 } 476 477 int ovs_flow_tbl_num_masks(const struct flow_table *table) 478 { 479 struct sw_flow_mask *mask; 480 int num = 0; 481 482 list_for_each_entry(mask, &table->mask_list, list) 483 num++; 484 485 return num; 486 } 487 488 static struct table_instance *table_instance_expand(struct table_instance *ti) 489 { 490 return table_instance_rehash(ti, ti->n_buckets * 2); 491 } 492 493 void ovs_flow_tbl_remove(struct flow_table *table, struct sw_flow *flow) 494 { 495 struct table_instance *ti = ovsl_dereference(table->ti); 496 497 BUG_ON(table->count == 0); 498 hlist_del_rcu(&flow->hash_node[ti->node_ver]); 499 table->count--; 500 } 501 502 static struct sw_flow_mask *mask_alloc(void) 503 { 504 struct sw_flow_mask *mask; 505 506 mask = kmalloc(sizeof(*mask), GFP_KERNEL); 507 if (mask) 508 mask->ref_count = 1; 509 510 return mask; 511 } 512 513 static bool mask_equal(const struct sw_flow_mask *a, 514 const struct sw_flow_mask *b) 515 { 516 u8 *a_ = (u8 *)&a->key + a->range.start; 517 u8 *b_ = (u8 *)&b->key + b->range.start; 518 519 return (a->range.end == b->range.end) 520 && (a->range.start == b->range.start) 521 && (memcmp(a_, b_, range_n_bytes(&a->range)) == 0); 522 } 523 524 static struct sw_flow_mask *flow_mask_find(const struct flow_table *tbl, 525 const struct sw_flow_mask *mask) 526 { 527 struct list_head *ml; 528 529 list_for_each(ml, &tbl->mask_list) { 530 struct sw_flow_mask *m; 531 m = container_of(ml, struct sw_flow_mask, list); 532 if (mask_equal(mask, m)) 533 return m; 534 } 535 536 return NULL; 537 } 538 539 /* Add 'mask' into the mask list, if it is not already there. */ 540 static int flow_mask_insert(struct flow_table *tbl, struct sw_flow *flow, 541 struct sw_flow_mask *new) 542 { 543 struct sw_flow_mask *mask; 544 mask = flow_mask_find(tbl, new); 545 if (!mask) { 546 /* Allocate a new mask if none exsits. */ 547 mask = mask_alloc(); 548 if (!mask) 549 return -ENOMEM; 550 mask->key = new->key; 551 mask->range = new->range; 552 list_add_rcu(&mask->list, &tbl->mask_list); 553 } else { 554 BUG_ON(!mask->ref_count); 555 mask->ref_count++; 556 } 557 558 flow->mask = mask; 559 return 0; 560 } 561 562 int ovs_flow_tbl_insert(struct flow_table *table, struct sw_flow *flow, 563 struct sw_flow_mask *mask) 564 { 565 struct table_instance *new_ti = NULL; 566 struct table_instance *ti; 567 int err; 568 569 err = flow_mask_insert(table, flow, mask); 570 if (err) 571 return err; 572 573 flow->hash = flow_hash(&flow->key, flow->mask->range.start, 574 flow->mask->range.end); 575 ti = ovsl_dereference(table->ti); 576 table_instance_insert(ti, flow); 577 table->count++; 578 579 /* Expand table, if necessary, to make room. */ 580 if (table->count > ti->n_buckets) 581 new_ti = table_instance_expand(ti); 582 else if (time_after(jiffies, table->last_rehash + REHASH_INTERVAL)) 583 new_ti = table_instance_rehash(ti, ti->n_buckets); 584 585 if (new_ti) { 586 rcu_assign_pointer(table->ti, new_ti); 587 table_instance_destroy(ti, true); 588 table->last_rehash = jiffies; 589 } 590 return 0; 591 } 592 593 /* Initializes the flow module. 594 * Returns zero if successful or a negative error code. */ 595 int ovs_flow_init(void) 596 { 597 BUILD_BUG_ON(__alignof__(struct sw_flow_key) % __alignof__(long)); 598 BUILD_BUG_ON(sizeof(struct sw_flow_key) % sizeof(long)); 599 600 flow_cache = kmem_cache_create("sw_flow", sizeof(struct sw_flow), 0, 601 0, NULL); 602 if (flow_cache == NULL) 603 return -ENOMEM; 604 605 return 0; 606 } 607 608 /* Uninitializes the flow module. */ 609 void ovs_flow_exit(void) 610 { 611 kmem_cache_destroy(flow_cache); 612 } 613