net/openvswitch/flow_table.c

   1 /*
   2  * Copyright (c) 2007-2014 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 struct kmem_cache *flow_stats_cache __read_mostly;
  52
  53 static u16 range_n_bytes(const struct sw_flow_key_range *range)
  54 {
  55         return range->end - range->start;
  56 }
  57
  58 void ovs_flow_mask_key(struct sw_flow_key *dst, const struct sw_flow_key *src,
  59                        const struct sw_flow_mask *mask)
  60 {
  61         const long *m = (const long *)((const u8 *)&mask->key +
  62                                 mask->range.start);
  63         const long *s = (const long *)((const u8 *)src +
  64                                 mask->range.start);
  65         long *d = (long *)((u8 *)dst + mask->range.start);
  66         int i;
  67
  68         /* The memory outside of the 'mask->range' are not set since
  69          * further operations on 'dst' only uses contents within
  70          * 'mask->range'.
  71          */
  72         for (i = 0; i < range_n_bytes(&mask->range); i += sizeof(long))
  73                 *d++ = *s++ & *m++;
  74 }
  75
  76 struct sw_flow *ovs_flow_alloc(void)
  77 {
  78         struct sw_flow *flow;
  79         struct flow_stats *stats;
  80         int node;
  81
  82         flow = kmem_cache_alloc(flow_cache, GFP_KERNEL);
  83         if (!flow)
  84                 return ERR_PTR(-ENOMEM);
  85
  86         flow->sf_acts = NULL;
  87         flow->mask = NULL;
  88         flow->stats_last_writer = NUMA_NO_NODE;
  89
  90         /* Initialize the default stat node. */
  91         stats = kmem_cache_alloc_node(flow_stats_cache,
  92                                       GFP_KERNEL | __GFP_ZERO, 0);
  93         if (!stats)
  94                 goto err;
  95
  96         spin_lock_init(&stats->lock);
  97
  98         RCU_INIT_POINTER(flow->stats[0], stats);
  99
 100         for_each_node(node)
 101                 if (node != 0)
 102                         RCU_INIT_POINTER(flow->stats[node], NULL);
 103
 104         return flow;
 105 err:
 106         kmem_cache_free(flow_cache, flow);
 107         return ERR_PTR(-ENOMEM);
 108 }
 109
 110 int ovs_flow_tbl_count(const struct flow_table *table)
 111 {
 112         return table->count;
 113 }
 114
 115 static struct flex_array *alloc_buckets(unsigned int n_buckets)
 116 {
 117         struct flex_array *buckets;
 118         int i, err;
 119
 120         buckets = flex_array_alloc(sizeof(struct hlist_head),
 121                                    n_buckets, GFP_KERNEL);
 122         if (!buckets)
 123                 return NULL;
 124
 125         err = flex_array_prealloc(buckets, 0, n_buckets, GFP_KERNEL);
 126         if (err) {
 127                 flex_array_free(buckets);
 128                 return NULL;
 129         }
 130
 131         for (i = 0; i < n_buckets; i++)
 132                 INIT_HLIST_HEAD((struct hlist_head *)
 133                                         flex_array_get(buckets, i));
 134
 135         return buckets;
 136 }
 137
 138 static void flow_free(struct sw_flow *flow)
 139 {
 140         int node;
 141
 142         kfree((struct sw_flow_actions __force *)flow->sf_acts);
 143         for_each_node(node)
 144                 if (flow->stats[node])
 145                         kmem_cache_free(flow_stats_cache,
 146                                         (struct flow_stats __force *)flow->stats[node]);
 147         kmem_cache_free(flow_cache, flow);
 148 }
 149
 150 static void rcu_free_flow_callback(struct rcu_head *rcu)
 151 {
 152         struct sw_flow *flow = container_of(rcu, struct sw_flow, rcu);
 153
 154         flow_free(flow);
 155 }
 156
 157 void ovs_flow_free(struct sw_flow *flow, bool deferred)
 158 {
 159         if (!flow)
 160                 return;
 161
 162         if (deferred)
 163                 call_rcu(&flow->rcu, rcu_free_flow_callback);
 164         else
 165                 flow_free(flow);
 166 }
 167
 168 static void free_buckets(struct flex_array *buckets)
 169 {
 170         flex_array_free(buckets);
 171 }
 172
 173
 174 static void __table_instance_destroy(struct table_instance *ti)
 175 {
 176         free_buckets(ti->buckets);
 177         kfree(ti);
 178 }
 179
 180 static struct table_instance *table_instance_alloc(int new_size)
 181 {
 182         struct table_instance *ti = kmalloc(sizeof(*ti), GFP_KERNEL);
 183
 184         if (!ti)
 185                 return NULL;
 186
 187         ti->buckets = alloc_buckets(new_size);
 188
 189         if (!ti->buckets) {
 190                 kfree(ti);
 191                 return NULL;
 192         }
 193         ti->n_buckets = new_size;
 194         ti->node_ver = 0;
 195         ti->keep_flows = false;
 196         get_random_bytes(&ti->hash_seed, sizeof(u32));
 197
 198         return ti;
 199 }
 200
 201 int ovs_flow_tbl_init(struct flow_table *table)
 202 {
 203         struct table_instance *ti;
 204
 205         ti = table_instance_alloc(TBL_MIN_BUCKETS);
 206
 207         if (!ti)
 208                 return -ENOMEM;
 209
 210         rcu_assign_pointer(table->ti, ti);
 211         INIT_LIST_HEAD(&table->mask_list);
 212         table->last_rehash = jiffies;
 213         table->count = 0;
 214         return 0;
 215 }
 216
 217 static void flow_tbl_destroy_rcu_cb(struct rcu_head *rcu)
 218 {
 219         struct table_instance *ti = container_of(rcu, struct table_instance, rcu);
 220
 221         __table_instance_destroy(ti);
 222 }
 223
 224 static void table_instance_destroy(struct table_instance *ti, bool deferred)
 225 {
 226         int i;
 227
 228         if (!ti)
 229                 return;
 230
 231         if (ti->keep_flows)
 232                 goto skip_flows;
 233
 234         for (i = 0; i < ti->n_buckets; i++) {
 235                 struct sw_flow *flow;
 236                 struct hlist_head *head = flex_array_get(ti->buckets, i);
 237                 struct hlist_node *n;
 238                 int ver = ti->node_ver;
 239
 240                 hlist_for_each_entry_safe(flow, n, head, hash_node[ver]) {
 241                         hlist_del_rcu(&flow->hash_node[ver]);
 242                         ovs_flow_free(flow, deferred);
 243                 }
 244         }
 245
 246 skip_flows:
 247         if (deferred)
 248                 call_rcu(&ti->rcu, flow_tbl_destroy_rcu_cb);
 249         else
 250                 __table_instance_destroy(ti);
 251 }
 252
 253 /* No need for locking this function is called from RCU callback or
 254  * error path.
 255  */
 256 void ovs_flow_tbl_destroy(struct flow_table *table)
 257 {
 258         struct table_instance *ti = rcu_dereference_raw(table->ti);
 259
 260         table_instance_destroy(ti, false);
 261 }
 262
 263 struct sw_flow *ovs_flow_tbl_dump_next(struct table_instance *ti,
 264                                        u32 *bucket, u32 *last)
 265 {
 266         struct sw_flow *flow;
 267         struct hlist_head *head;
 268         int ver;
 269         int i;
 270
 271         ver = ti->node_ver;
 272         while (*bucket < ti->n_buckets) {
 273                 i = 0;
 274                 head = flex_array_get(ti->buckets, *bucket);
 275                 hlist_for_each_entry_rcu(flow, head, hash_node[ver]) {
 276                         if (i < *last) {
 277                                 i++;
 278                                 continue;
 279                         }
 280                         *last = i + 1;
 281                         return flow;
 282                 }
 283                 (*bucket)++;
 284                 *last = 0;
 285         }
 286
 287         return NULL;
 288 }
 289
 290 static struct hlist_head *find_bucket(struct table_instance *ti, u32 hash)
 291 {
 292         hash = jhash_1word(hash, ti->hash_seed);
 293         return flex_array_get(ti->buckets,
 294                                 (hash & (ti->n_buckets - 1)));
 295 }
 296
 297 static void table_instance_insert(struct table_instance *ti, struct sw_flow *flow)
 298 {
 299         struct hlist_head *head;
 300
 301         head = find_bucket(ti, flow->hash);
 302         hlist_add_head_rcu(&flow->hash_node[ti->node_ver], head);
 303 }
 304
 305 static void flow_table_copy_flows(struct table_instance *old,
 306                                   struct table_instance *new)
 307 {
 308         int old_ver;
 309         int i;
 310
 311         old_ver = old->node_ver;
 312         new->node_ver = !old_ver;
 313
 314         /* Insert in new table. */
 315         for (i = 0; i < old->n_buckets; i++) {
 316                 struct sw_flow *flow;
 317                 struct hlist_head *head;
 318
 319                 head = flex_array_get(old->buckets, i);
 320
 321                 hlist_for_each_entry(flow, head, hash_node[old_ver])
 322                         table_instance_insert(new, flow);
 323         }
 324
 325         old->keep_flows = true;
 326 }
 327
 328 static struct table_instance *table_instance_rehash(struct table_instance *ti,
 329                                             int n_buckets)
 330 {
 331         struct table_instance *new_ti;
 332
 333         new_ti = table_instance_alloc(n_buckets);
 334         if (!new_ti)
 335                 return NULL;
 336
 337         flow_table_copy_flows(ti, new_ti);
 338
 339         return new_ti;
 340 }
 341
 342 int ovs_flow_tbl_flush(struct flow_table *flow_table)
 343 {
 344         struct table_instance *old_ti;
 345         struct table_instance *new_ti;
 346
 347         old_ti = ovsl_dereference(flow_table->ti);
 348         new_ti = table_instance_alloc(TBL_MIN_BUCKETS);
 349         if (!new_ti)
 350                 return -ENOMEM;
 351
 352         rcu_assign_pointer(flow_table->ti, new_ti);
 353         flow_table->last_rehash = jiffies;
 354         flow_table->count = 0;
 355
 356         table_instance_destroy(old_ti, true);
 357         return 0;
 358 }
 359
 360 static u32 flow_hash(const struct sw_flow_key *key, int key_start,
 361                      int key_end)
 362 {
 363         const u32 *hash_key = (const u32 *)((const u8 *)key + key_start);
 364         int hash_u32s = (key_end - key_start) >> 2;
 365
 366         /* Make sure number of hash bytes are multiple of u32. */
 367         BUILD_BUG_ON(sizeof(long) % sizeof(u32));
 368
 369         return arch_fast_hash2(hash_key, hash_u32s, 0);
 370 }
 371
 372 static int flow_key_start(const struct sw_flow_key *key)
 373 {
 374         if (key->tun_key.ipv4_dst)
 375                 return 0;
 376         else
 377                 return rounddown(offsetof(struct sw_flow_key, phy),
 378                                           sizeof(long));
 379 }
 380
 381 static bool cmp_key(const struct sw_flow_key *key1,
 382                     const struct sw_flow_key *key2,
 383                     int key_start, int key_end)
 384 {
 385         const long *cp1 = (const long *)((const u8 *)key1 + key_start);
 386         const long *cp2 = (const long *)((const u8 *)key2 + key_start);
 387         long diffs = 0;
 388         int i;
 389
 390         for (i = key_start; i < key_end;  i += sizeof(long))
 391                 diffs |= *cp1++ ^ *cp2++;
 392
 393         return diffs == 0;
 394 }
 395
 396 static bool flow_cmp_masked_key(const struct sw_flow *flow,
 397                                 const struct sw_flow_key *key,
 398                                 int key_start, int key_end)
 399 {
 400         return cmp_key(&flow->key, key, key_start, key_end);
 401 }
 402
 403 bool ovs_flow_cmp_unmasked_key(const struct sw_flow *flow,
 404                                const struct sw_flow_match *match)
 405 {
 406         struct sw_flow_key *key = match->key;
 407         int key_start = flow_key_start(key);
 408         int key_end = match->range.end;
 409
 410         return cmp_key(&flow->unmasked_key, key, key_start, key_end);
 411 }
 412
 413 static struct sw_flow *masked_flow_lookup(struct table_instance *ti,
 414                                           const struct sw_flow_key *unmasked,
 415                                           const struct sw_flow_mask *mask)
 416 {
 417         struct sw_flow *flow;
 418         struct hlist_head *head;
 419         int key_start = mask->range.start;
 420         int key_end = mask->range.end;
 421         u32 hash;
 422         struct sw_flow_key masked_key;
 423
 424         ovs_flow_mask_key(&masked_key, unmasked, mask);
 425         hash = flow_hash(&masked_key, key_start, key_end);
 426         head = find_bucket(ti, hash);
 427         hlist_for_each_entry_rcu(flow, head, hash_node[ti->node_ver]) {
 428                 if (flow->mask == mask && flow->hash == hash &&
 429                     flow_cmp_masked_key(flow, &masked_key,
 430                                           key_start, key_end))
 431                         return flow;
 432         }
 433         return NULL;
 434 }
 435
 436 struct sw_flow *ovs_flow_tbl_lookup_stats(struct flow_table *tbl,
 437                                     const struct sw_flow_key *key,
 438                                     u32 *n_mask_hit)
 439 {
 440         struct table_instance *ti = rcu_dereference_ovsl(tbl->ti);
 441         struct sw_flow_mask *mask;
 442         struct sw_flow *flow;
 443
 444         *n_mask_hit = 0;
 445         list_for_each_entry_rcu(mask, &tbl->mask_list, list) {
 446                 (*n_mask_hit)++;
 447                 flow = masked_flow_lookup(ti, key, mask);
 448                 if (flow)  /* Found */
 449                         return flow;
 450         }
 451         return NULL;
 452 }
 453
 454 struct sw_flow *ovs_flow_tbl_lookup(struct flow_table *tbl,
 455                                     const struct sw_flow_key *key)
 456 {
 457         u32 __always_unused n_mask_hit;
 458
 459         return ovs_flow_tbl_lookup_stats(tbl, key, &n_mask_hit);
 460 }
 461
 462 struct sw_flow *ovs_flow_tbl_lookup_exact(struct flow_table *tbl,
 463                                           const struct sw_flow_match *match)
 464 {
 465         struct table_instance *ti = rcu_dereference_ovsl(tbl->ti);
 466         struct sw_flow_mask *mask;
 467         struct sw_flow *flow;
 468
 469         /* Always called under ovs-mutex. */
 470         list_for_each_entry(mask, &tbl->mask_list, list) {
 471                 flow = masked_flow_lookup(ti, match->key, mask);
 472                 if (flow && ovs_flow_cmp_unmasked_key(flow, match))  /* Found */
 473                         return flow;
 474         }
 475         return NULL;
 476 }
 477
 478 int ovs_flow_tbl_num_masks(const struct flow_table *table)
 479 {
 480         struct sw_flow_mask *mask;
 481         int num = 0;
 482
 483         list_for_each_entry(mask, &table->mask_list, list)
 484                 num++;
 485
 486         return num;
 487 }
 488
 489 static struct table_instance *table_instance_expand(struct table_instance *ti)
 490 {
 491         return table_instance_rehash(ti, ti->n_buckets * 2);
 492 }
 493
 494 /* Remove 'mask' from the mask list, if it is not needed any more. */
 495 static void flow_mask_remove(struct flow_table *tbl, struct sw_flow_mask *mask)
 496 {
 497         if (mask) {
 498                 /* ovs-lock is required to protect mask-refcount and
 499                  * mask list.
 500                  */
 501                 ASSERT_OVSL();
 502                 BUG_ON(!mask->ref_count);
 503                 mask->ref_count--;
 504
 505                 if (!mask->ref_count) {
 506                         list_del_rcu(&mask->list);
 507                         kfree_rcu(mask, rcu);
 508                 }
 509         }
 510 }
 511
 512 /* Must be called with OVS mutex held. */
 513 void ovs_flow_tbl_remove(struct flow_table *table, struct sw_flow *flow)
 514 {
 515         struct table_instance *ti = ovsl_dereference(table->ti);
 516
 517         BUG_ON(table->count == 0);
 518         hlist_del_rcu(&flow->hash_node[ti->node_ver]);
 519         table->count--;
 520
 521         /* RCU delete the mask. 'flow->mask' is not NULLed, as it should be
 522          * accessible as long as the RCU read lock is held.
 523          */
 524         flow_mask_remove(table, flow->mask);
 525 }
 526
 527 static struct sw_flow_mask *mask_alloc(void)
 528 {
 529         struct sw_flow_mask *mask;
 530
 531         mask = kmalloc(sizeof(*mask), GFP_KERNEL);
 532         if (mask)
 533                 mask->ref_count = 1;
 534
 535         return mask;
 536 }
 537
 538 static bool mask_equal(const struct sw_flow_mask *a,
 539                        const struct sw_flow_mask *b)
 540 {
 541         const u8 *a_ = (const u8 *)&a->key + a->range.start;
 542         const u8 *b_ = (const u8 *)&b->key + b->range.start;
 543
 544         return  (a->range.end == b->range.end)
 545                 && (a->range.start == b->range.start)
 546                 && (memcmp(a_, b_, range_n_bytes(&a->range)) == 0);
 547 }
 548
 549 static struct sw_flow_mask *flow_mask_find(const struct flow_table *tbl,
 550                                            const struct sw_flow_mask *mask)
 551 {
 552         struct list_head *ml;
 553
 554         list_for_each(ml, &tbl->mask_list) {
 555                 struct sw_flow_mask *m;
 556                 m = container_of(ml, struct sw_flow_mask, list);
 557                 if (mask_equal(mask, m))
 558                         return m;
 559         }
 560
 561         return NULL;
 562 }
 563
 564 /* Add 'mask' into the mask list, if it is not already there. */
 565 static int flow_mask_insert(struct flow_table *tbl, struct sw_flow *flow,
 566                             const struct sw_flow_mask *new)
 567 {
 568         struct sw_flow_mask *mask;
 569         mask = flow_mask_find(tbl, new);
 570         if (!mask) {
 571                 /* Allocate a new mask if none exsits. */
 572                 mask = mask_alloc();
 573                 if (!mask)
 574                         return -ENOMEM;
 575                 mask->key = new->key;
 576                 mask->range = new->range;
 577                 list_add_rcu(&mask->list, &tbl->mask_list);
 578         } else {
 579                 BUG_ON(!mask->ref_count);
 580                 mask->ref_count++;
 581         }
 582
 583         flow->mask = mask;
 584         return 0;
 585 }
 586
 587 /* Must be called with OVS mutex held. */
 588 int ovs_flow_tbl_insert(struct flow_table *table, struct sw_flow *flow,
 589                         const struct sw_flow_mask *mask)
 590 {
 591         struct table_instance *new_ti = NULL;
 592         struct table_instance *ti;
 593         int err;
 594
 595         err = flow_mask_insert(table, flow, mask);
 596         if (err)
 597                 return err;
 598
 599         flow->hash = flow_hash(&flow->key, flow->mask->range.start,
 600                         flow->mask->range.end);
 601         ti = ovsl_dereference(table->ti);
 602         table_instance_insert(ti, flow);
 603         table->count++;
 604
 605         /* Expand table, if necessary, to make room. */
 606         if (table->count > ti->n_buckets)
 607                 new_ti = table_instance_expand(ti);
 608         else if (time_after(jiffies, table->last_rehash + REHASH_INTERVAL))
 609                 new_ti = table_instance_rehash(ti, ti->n_buckets);
 610
 611         if (new_ti) {
 612                 rcu_assign_pointer(table->ti, new_ti);
 613                 table_instance_destroy(ti, true);
 614                 table->last_rehash = jiffies;
 615         }
 616         return 0;
 617 }
 618
 619 /* Initializes the flow module.
 620  * Returns zero if successful or a negative error code. */
 621 int ovs_flow_init(void)
 622 {
 623         BUILD_BUG_ON(__alignof__(struct sw_flow_key) % __alignof__(long));
 624         BUILD_BUG_ON(sizeof(struct sw_flow_key) % sizeof(long));
 625
 626         flow_cache = kmem_cache_create("sw_flow", sizeof(struct sw_flow)
 627                                        + (num_possible_nodes()
 628                                           * sizeof(struct flow_stats *)),
 629                                        0, 0, NULL);
 630         if (flow_cache == NULL)
 631                 return -ENOMEM;
 632
 633         flow_stats_cache
 634                 = kmem_cache_create("sw_flow_stats", sizeof(struct flow_stats),
 635                                     0, SLAB_HWCACHE_ALIGN, NULL);
 636         if (flow_stats_cache == NULL) {
 637                 kmem_cache_destroy(flow_cache);
 638                 flow_cache = NULL;
 639                 return -ENOMEM;
 640         }
 641
 642         return 0;
 643 }
 644
 645 /* Uninitializes the flow module. */
 646 void ovs_flow_exit(void)
 647 {
 648         kmem_cache_destroy(flow_stats_cache);
 649         kmem_cache_destroy(flow_cache);
 650 }