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/jhash.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 #include "datapath.h" 48 49 #define TBL_MIN_BUCKETS 1024 50 #define REHASH_INTERVAL (10 * 60 * HZ) 51 52 static struct kmem_cache *flow_cache; 53 54 static u16 range_n_bytes(const struct sw_flow_key_range *range) 55 { 56 return range->end - range->start; 57 } 58 59 void ovs_flow_mask_key(struct sw_flow_key *dst, const struct sw_flow_key *src, 60 const struct sw_flow_mask *mask) 61 { 62 const long *m = (long *)((u8 *)&mask->key + mask->range.start); 63 const long *s = (long *)((u8 *)src + mask->range.start); 64 long *d = (long *)((u8 *)dst + mask->range.start); 65 int i; 66 67 /* The memory outside of the 'mask->range' are not set since 68 * further operations on 'dst' only uses contents within 69 * 'mask->range'. 70 */ 71 for (i = 0; i < range_n_bytes(&mask->range); i += sizeof(long)) 72 *d++ = *s++ & *m++; 73 } 74 75 struct sw_flow *ovs_flow_alloc(void) 76 { 77 struct sw_flow *flow; 78 79 flow = kmem_cache_alloc(flow_cache, GFP_KERNEL); 80 if (!flow) 81 return ERR_PTR(-ENOMEM); 82 83 spin_lock_init(&flow->lock); 84 flow->sf_acts = NULL; 85 flow->mask = NULL; 86 87 return flow; 88 } 89 90 int ovs_flow_tbl_count(struct flow_table *table) 91 { 92 return table->count; 93 } 94 95 static struct flex_array *alloc_buckets(unsigned int n_buckets) 96 { 97 struct flex_array *buckets; 98 int i, err; 99 100 buckets = flex_array_alloc(sizeof(struct hlist_head), 101 n_buckets, GFP_KERNEL); 102 if (!buckets) 103 return NULL; 104 105 err = flex_array_prealloc(buckets, 0, n_buckets, GFP_KERNEL); 106 if (err) { 107 flex_array_free(buckets); 108 return NULL; 109 } 110 111 for (i = 0; i < n_buckets; i++) 112 INIT_HLIST_HEAD((struct hlist_head *) 113 flex_array_get(buckets, i)); 114 115 return buckets; 116 } 117 118 static void flow_free(struct sw_flow *flow) 119 { 120 kfree((struct sf_flow_acts __force *)flow->sf_acts); 121 kmem_cache_free(flow_cache, flow); 122 } 123 124 static void rcu_free_flow_callback(struct rcu_head *rcu) 125 { 126 struct sw_flow *flow = container_of(rcu, struct sw_flow, rcu); 127 128 flow_free(flow); 129 } 130 131 void ovs_flow_free(struct sw_flow *flow, bool deferred) 132 { 133 if (!flow) 134 return; 135 136 ovs_sw_flow_mask_del_ref(flow->mask, deferred); 137 138 if (deferred) 139 call_rcu(&flow->rcu, rcu_free_flow_callback); 140 else 141 flow_free(flow); 142 } 143 144 static void free_buckets(struct flex_array *buckets) 145 { 146 flex_array_free(buckets); 147 } 148 149 static void __table_instance_destroy(struct table_instance *ti) 150 { 151 int i; 152 153 if (ti->keep_flows) 154 goto skip_flows; 155 156 for (i = 0; i < ti->n_buckets; i++) { 157 struct sw_flow *flow; 158 struct hlist_head *head = flex_array_get(ti->buckets, i); 159 struct hlist_node *n; 160 int ver = ti->node_ver; 161 162 hlist_for_each_entry_safe(flow, n, head, hash_node[ver]) { 163 hlist_del(&flow->hash_node[ver]); 164 ovs_flow_free(flow, false); 165 } 166 } 167 168 skip_flows: 169 free_buckets(ti->buckets); 170 kfree(ti); 171 } 172 173 static struct table_instance *table_instance_alloc(int new_size) 174 { 175 struct table_instance *ti = kmalloc(sizeof(*ti), GFP_KERNEL); 176 177 if (!ti) 178 return NULL; 179 180 ti->buckets = alloc_buckets(new_size); 181 182 if (!ti->buckets) { 183 kfree(ti); 184 return NULL; 185 } 186 ti->n_buckets = new_size; 187 ti->node_ver = 0; 188 ti->keep_flows = false; 189 get_random_bytes(&ti->hash_seed, sizeof(u32)); 190 191 return ti; 192 } 193 194 int ovs_flow_tbl_init(struct flow_table *table) 195 { 196 struct table_instance *ti; 197 198 ti = table_instance_alloc(TBL_MIN_BUCKETS); 199 200 if (!ti) 201 return -ENOMEM; 202 203 rcu_assign_pointer(table->ti, ti); 204 INIT_LIST_HEAD(&table->mask_list); 205 table->last_rehash = jiffies; 206 table->count = 0; 207 return 0; 208 } 209 210 static void flow_tbl_destroy_rcu_cb(struct rcu_head *rcu) 211 { 212 struct table_instance *ti = container_of(rcu, struct table_instance, rcu); 213 214 __table_instance_destroy(ti); 215 } 216 217 static void table_instance_destroy(struct table_instance *ti, bool deferred) 218 { 219 if (!ti) 220 return; 221 222 if (deferred) 223 call_rcu(&ti->rcu, flow_tbl_destroy_rcu_cb); 224 else 225 __table_instance_destroy(ti); 226 } 227 228 void ovs_flow_tbl_destroy(struct flow_table *table, bool deferred) 229 { 230 struct table_instance *ti = ovsl_dereference(table->ti); 231 232 table_instance_destroy(ti, deferred); 233 } 234 235 struct sw_flow *ovs_flow_tbl_dump_next(struct table_instance *ti, 236 u32 *bucket, u32 *last) 237 { 238 struct sw_flow *flow; 239 struct hlist_head *head; 240 int ver; 241 int i; 242 243 ver = ti->node_ver; 244 while (*bucket < ti->n_buckets) { 245 i = 0; 246 head = flex_array_get(ti->buckets, *bucket); 247 hlist_for_each_entry_rcu(flow, head, hash_node[ver]) { 248 if (i < *last) { 249 i++; 250 continue; 251 } 252 *last = i + 1; 253 return flow; 254 } 255 (*bucket)++; 256 *last = 0; 257 } 258 259 return NULL; 260 } 261 262 static struct hlist_head *find_bucket(struct table_instance *ti, u32 hash) 263 { 264 hash = jhash_1word(hash, ti->hash_seed); 265 return flex_array_get(ti->buckets, 266 (hash & (ti->n_buckets - 1))); 267 } 268 269 static void table_instance_insert(struct table_instance *ti, struct sw_flow *flow) 270 { 271 struct hlist_head *head; 272 273 head = find_bucket(ti, flow->hash); 274 hlist_add_head_rcu(&flow->hash_node[ti->node_ver], head); 275 } 276 277 static void flow_table_copy_flows(struct table_instance *old, 278 struct table_instance *new) 279 { 280 int old_ver; 281 int i; 282 283 old_ver = old->node_ver; 284 new->node_ver = !old_ver; 285 286 /* Insert in new table. */ 287 for (i = 0; i < old->n_buckets; i++) { 288 struct sw_flow *flow; 289 struct hlist_head *head; 290 291 head = flex_array_get(old->buckets, i); 292 293 hlist_for_each_entry(flow, head, hash_node[old_ver]) 294 table_instance_insert(new, flow); 295 } 296 297 old->keep_flows = true; 298 } 299 300 static struct table_instance *table_instance_rehash(struct table_instance *ti, 301 int n_buckets) 302 { 303 struct table_instance *new_ti; 304 305 new_ti = table_instance_alloc(n_buckets); 306 if (!new_ti) 307 return ERR_PTR(-ENOMEM); 308 309 flow_table_copy_flows(ti, new_ti); 310 311 return new_ti; 312 } 313 314 int ovs_flow_tbl_flush(struct flow_table *flow_table) 315 { 316 struct table_instance *old_ti; 317 struct table_instance *new_ti; 318 319 old_ti = ovsl_dereference(flow_table->ti); 320 new_ti = table_instance_alloc(TBL_MIN_BUCKETS); 321 if (!new_ti) 322 return -ENOMEM; 323 324 rcu_assign_pointer(flow_table->ti, new_ti); 325 flow_table->last_rehash = jiffies; 326 flow_table->count = 0; 327 328 table_instance_destroy(old_ti, true); 329 return 0; 330 } 331 332 static u32 flow_hash(const struct sw_flow_key *key, int key_start, 333 int key_end) 334 { 335 u32 *hash_key = (u32 *)((u8 *)key + key_start); 336 int hash_u32s = (key_end - key_start) >> 2; 337 338 /* Make sure number of hash bytes are multiple of u32. */ 339 BUILD_BUG_ON(sizeof(long) % sizeof(u32)); 340 341 return jhash2(hash_key, hash_u32s, 0); 342 } 343 344 static int flow_key_start(const struct sw_flow_key *key) 345 { 346 if (key->tun_key.ipv4_dst) 347 return 0; 348 else 349 return rounddown(offsetof(struct sw_flow_key, phy), 350 sizeof(long)); 351 } 352 353 static bool cmp_key(const struct sw_flow_key *key1, 354 const struct sw_flow_key *key2, 355 int key_start, int key_end) 356 { 357 const long *cp1 = (long *)((u8 *)key1 + key_start); 358 const long *cp2 = (long *)((u8 *)key2 + key_start); 359 long diffs = 0; 360 int i; 361 362 for (i = key_start; i < key_end; i += sizeof(long)) 363 diffs |= *cp1++ ^ *cp2++; 364 365 return diffs == 0; 366 } 367 368 static bool flow_cmp_masked_key(const struct sw_flow *flow, 369 const struct sw_flow_key *key, 370 int key_start, int key_end) 371 { 372 return cmp_key(&flow->key, key, key_start, key_end); 373 } 374 375 bool ovs_flow_cmp_unmasked_key(const struct sw_flow *flow, 376 struct sw_flow_match *match) 377 { 378 struct sw_flow_key *key = match->key; 379 int key_start = flow_key_start(key); 380 int key_end = match->range.end; 381 382 return cmp_key(&flow->unmasked_key, key, key_start, key_end); 383 } 384 385 static struct sw_flow *masked_flow_lookup(struct table_instance *ti, 386 const struct sw_flow_key *unmasked, 387 struct sw_flow_mask *mask) 388 { 389 struct sw_flow *flow; 390 struct hlist_head *head; 391 int key_start = mask->range.start; 392 int key_end = mask->range.end; 393 u32 hash; 394 struct sw_flow_key masked_key; 395 396 ovs_flow_mask_key(&masked_key, unmasked, mask); 397 hash = flow_hash(&masked_key, key_start, key_end); 398 head = find_bucket(ti, hash); 399 hlist_for_each_entry_rcu(flow, head, hash_node[ti->node_ver]) { 400 if (flow->mask == mask && 401 flow_cmp_masked_key(flow, &masked_key, 402 key_start, key_end)) 403 return flow; 404 } 405 return NULL; 406 } 407 408 struct sw_flow *ovs_flow_tbl_lookup(struct flow_table *tbl, 409 const struct sw_flow_key *key) 410 { 411 struct table_instance *ti = rcu_dereference(tbl->ti); 412 struct sw_flow_mask *mask; 413 struct sw_flow *flow; 414 415 list_for_each_entry_rcu(mask, &tbl->mask_list, list) { 416 flow = masked_flow_lookup(ti, key, mask); 417 if (flow) /* Found */ 418 return flow; 419 } 420 return NULL; 421 } 422 423 static struct table_instance *table_instance_expand(struct table_instance *ti) 424 { 425 return table_instance_rehash(ti, ti->n_buckets * 2); 426 } 427 428 void ovs_flow_tbl_insert(struct flow_table *table, struct sw_flow *flow) 429 { 430 struct table_instance *ti = NULL; 431 struct table_instance *new_ti = NULL; 432 433 ti = ovsl_dereference(table->ti); 434 435 /* Expand table, if necessary, to make room. */ 436 if (table->count > ti->n_buckets) 437 new_ti = table_instance_expand(ti); 438 else if (time_after(jiffies, table->last_rehash + REHASH_INTERVAL)) 439 new_ti = table_instance_rehash(ti, ti->n_buckets); 440 441 if (new_ti && !IS_ERR(new_ti)) { 442 rcu_assign_pointer(table->ti, new_ti); 443 ovs_flow_tbl_destroy(table, true); 444 ti = ovsl_dereference(table->ti); 445 table->last_rehash = jiffies; 446 } 447 448 flow->hash = flow_hash(&flow->key, flow->mask->range.start, 449 flow->mask->range.end); 450 table_instance_insert(ti, flow); 451 table->count++; 452 } 453 454 void ovs_flow_tbl_remove(struct flow_table *table, struct sw_flow *flow) 455 { 456 struct table_instance *ti = ovsl_dereference(table->ti); 457 458 BUG_ON(table->count == 0); 459 hlist_del_rcu(&flow->hash_node[ti->node_ver]); 460 table->count--; 461 } 462 463 struct sw_flow_mask *ovs_sw_flow_mask_alloc(void) 464 { 465 struct sw_flow_mask *mask; 466 467 mask = kmalloc(sizeof(*mask), GFP_KERNEL); 468 if (mask) 469 mask->ref_count = 0; 470 471 return mask; 472 } 473 474 void ovs_sw_flow_mask_add_ref(struct sw_flow_mask *mask) 475 { 476 mask->ref_count++; 477 } 478 479 static void rcu_free_sw_flow_mask_cb(struct rcu_head *rcu) 480 { 481 struct sw_flow_mask *mask = container_of(rcu, struct sw_flow_mask, rcu); 482 483 kfree(mask); 484 } 485 486 void ovs_sw_flow_mask_del_ref(struct sw_flow_mask *mask, bool deferred) 487 { 488 if (!mask) 489 return; 490 491 BUG_ON(!mask->ref_count); 492 mask->ref_count--; 493 494 if (!mask->ref_count) { 495 list_del_rcu(&mask->list); 496 if (deferred) 497 call_rcu(&mask->rcu, rcu_free_sw_flow_mask_cb); 498 else 499 kfree(mask); 500 } 501 } 502 503 static bool mask_equal(const struct sw_flow_mask *a, 504 const struct sw_flow_mask *b) 505 { 506 u8 *a_ = (u8 *)&a->key + a->range.start; 507 u8 *b_ = (u8 *)&b->key + b->range.start; 508 509 return (a->range.end == b->range.end) 510 && (a->range.start == b->range.start) 511 && (memcmp(a_, b_, range_n_bytes(&a->range)) == 0); 512 } 513 514 struct sw_flow_mask *ovs_sw_flow_mask_find(const struct flow_table *tbl, 515 const struct sw_flow_mask *mask) 516 { 517 struct list_head *ml; 518 519 list_for_each(ml, &tbl->mask_list) { 520 struct sw_flow_mask *m; 521 m = container_of(ml, struct sw_flow_mask, list); 522 if (mask_equal(mask, m)) 523 return m; 524 } 525 526 return NULL; 527 } 528 529 /** 530 * add a new mask into the mask list. 531 * The caller needs to make sure that 'mask' is not the same 532 * as any masks that are already on the list. 533 */ 534 void ovs_sw_flow_mask_insert(struct flow_table *tbl, struct sw_flow_mask *mask) 535 { 536 list_add_rcu(&mask->list, &tbl->mask_list); 537 } 538 539 /* Initializes the flow module. 540 * Returns zero if successful or a negative error code. */ 541 int ovs_flow_init(void) 542 { 543 BUILD_BUG_ON(__alignof__(struct sw_flow_key) % __alignof__(long)); 544 BUILD_BUG_ON(sizeof(struct sw_flow_key) % sizeof(long)); 545 546 flow_cache = kmem_cache_create("sw_flow", sizeof(struct sw_flow), 0, 547 0, NULL); 548 if (flow_cache == NULL) 549 return -ENOMEM; 550 551 return 0; 552 } 553 554 /* Uninitializes the flow module. */ 555 void ovs_flow_exit(void) 556 { 557 kmem_cache_destroy(flow_cache); 558 } 559