net/openvswitch/flow_table.c

   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 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(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 void ovs_flow_tbl_destroy(struct flow_table *table, bool deferred)
 254 {
 255         struct table_instance *ti = ovsl_dereference(table->ti);
 256
 257         table_instance_destroy(ti, deferred);
 258 }
 259
 260 struct sw_flow *ovs_flow_tbl_dump_next(struct table_instance *ti,
 261                                        u32 *bucket, u32 *last)
 262 {
 263         struct sw_flow *flow;
 264         struct hlist_head *head;
 265         int ver;
 266         int i;
 267
 268         ver = ti->node_ver;
 269         while (*bucket < ti->n_buckets) {
 270                 i = 0;
 271                 head = flex_array_get(ti->buckets, *bucket);
 272                 hlist_for_each_entry_rcu(flow, head, hash_node[ver]) {
 273                         if (i < *last) {
 274                                 i++;
 275                                 continue;
 276                         }
 277                         *last = i + 1;
 278                         return flow;
 279                 }
 280                 (*bucket)++;
 281                 *last = 0;
 282         }
 283
 284         return NULL;
 285 }
 286
 287 static struct hlist_head *find_bucket(struct table_instance *ti, u32 hash)
 288 {
 289         hash = jhash_1word(hash, ti->hash_seed);
 290         return flex_array_get(ti->buckets,
 291                                 (hash & (ti->n_buckets - 1)));
 292 }
 293
 294 static void table_instance_insert(struct table_instance *ti, struct sw_flow *flow)
 295 {
 296         struct hlist_head *head;
 297
 298         head = find_bucket(ti, flow->hash);
 299         hlist_add_head_rcu(&flow->hash_node[ti->node_ver], head);
 300 }
 301
 302 static void flow_table_copy_flows(struct table_instance *old,
 303                                   struct table_instance *new)
 304 {
 305         int old_ver;
 306         int i;
 307
 308         old_ver = old->node_ver;
 309         new->node_ver = !old_ver;
 310
 311         /* Insert in new table. */
 312         for (i = 0; i < old->n_buckets; i++) {
 313                 struct sw_flow *flow;
 314                 struct hlist_head *head;
 315
 316                 head = flex_array_get(old->buckets, i);
 317
 318                 hlist_for_each_entry(flow, head, hash_node[old_ver])
 319                         table_instance_insert(new, flow);
 320         }
 321
 322         old->keep_flows = true;
 323 }
 324
 325 static struct table_instance *table_instance_rehash(struct table_instance *ti,
 326                                             int n_buckets)
 327 {
 328         struct table_instance *new_ti;
 329
 330         new_ti = table_instance_alloc(n_buckets);
 331         if (!new_ti)
 332                 return NULL;
 333
 334         flow_table_copy_flows(ti, new_ti);
 335
 336         return new_ti;
 337 }
 338
 339 int ovs_flow_tbl_flush(struct flow_table *flow_table)
 340 {
 341         struct table_instance *old_ti;
 342         struct table_instance *new_ti;
 343
 344         old_ti = ovsl_dereference(flow_table->ti);
 345         new_ti = table_instance_alloc(TBL_MIN_BUCKETS);
 346         if (!new_ti)
 347                 return -ENOMEM;
 348
 349         rcu_assign_pointer(flow_table->ti, new_ti);
 350         flow_table->last_rehash = jiffies;
 351         flow_table->count = 0;
 352
 353         table_instance_destroy(old_ti, true);
 354         return 0;
 355 }
 356
 357 static u32 flow_hash(const struct sw_flow_key *key, int key_start,
 358                      int key_end)
 359 {
 360         const u32 *hash_key = (const u32 *)((const u8 *)key + key_start);
 361         int hash_u32s = (key_end - key_start) >> 2;
 362
 363         /* Make sure number of hash bytes are multiple of u32. */
 364         BUILD_BUG_ON(sizeof(long) % sizeof(u32));
 365
 366         return arch_fast_hash2(hash_key, hash_u32s, 0);
 367 }
 368
 369 static int flow_key_start(const struct sw_flow_key *key)
 370 {
 371         if (key->tun_key.ipv4_dst)
 372                 return 0;
 373         else
 374                 return rounddown(offsetof(struct sw_flow_key, phy),
 375                                           sizeof(long));
 376 }
 377
 378 static bool cmp_key(const struct sw_flow_key *key1,
 379                     const struct sw_flow_key *key2,
 380                     int key_start, int key_end)
 381 {
 382         const long *cp1 = (const long *)((const u8 *)key1 + key_start);
 383         const long *cp2 = (const long *)((const u8 *)key2 + key_start);
 384         long diffs = 0;
 385         int i;
 386
 387         for (i = key_start; i < key_end;  i += sizeof(long))
 388                 diffs |= *cp1++ ^ *cp2++;
 389
 390         return diffs == 0;
 391 }
 392
 393 static bool flow_cmp_masked_key(const struct sw_flow *flow,
 394                                 const struct sw_flow_key *key,
 395                                 int key_start, int key_end)
 396 {
 397         return cmp_key(&flow->key, key, key_start, key_end);
 398 }
 399
 400 bool ovs_flow_cmp_unmasked_key(const struct sw_flow *flow,
 401                                struct sw_flow_match *match)
 402 {
 403         struct sw_flow_key *key = match->key;
 404         int key_start = flow_key_start(key);
 405         int key_end = match->range.end;
 406
 407         return cmp_key(&flow->unmasked_key, key, key_start, key_end);
 408 }
 409
 410 static struct sw_flow *masked_flow_lookup(struct table_instance *ti,
 411                                           const struct sw_flow_key *unmasked,
 412                                           struct sw_flow_mask *mask)
 413 {
 414         struct sw_flow *flow;
 415         struct hlist_head *head;
 416         int key_start = mask->range.start;
 417         int key_end = mask->range.end;
 418         u32 hash;
 419         struct sw_flow_key masked_key;
 420
 421         ovs_flow_mask_key(&masked_key, unmasked, mask);
 422         hash = flow_hash(&masked_key, key_start, key_end);
 423         head = find_bucket(ti, hash);
 424         hlist_for_each_entry_rcu(flow, head, hash_node[ti->node_ver]) {
 425                 if (flow->mask == mask && flow->hash == hash &&
 426                     flow_cmp_masked_key(flow, &masked_key,
 427                                           key_start, key_end))
 428                         return flow;
 429         }
 430         return NULL;
 431 }
 432
 433 struct sw_flow *ovs_flow_tbl_lookup_stats(struct flow_table *tbl,
 434                                     const struct sw_flow_key *key,
 435                                     u32 *n_mask_hit)
 436 {
 437         struct table_instance *ti = rcu_dereference_ovsl(tbl->ti);
 438         struct sw_flow_mask *mask;
 439         struct sw_flow *flow;
 440
 441         *n_mask_hit = 0;
 442         list_for_each_entry_rcu(mask, &tbl->mask_list, list) {
 443                 (*n_mask_hit)++;
 444                 flow = masked_flow_lookup(ti, key, mask);
 445                 if (flow)  /* Found */
 446                         return flow;
 447         }
 448         return NULL;
 449 }
 450
 451 struct sw_flow *ovs_flow_tbl_lookup(struct flow_table *tbl,
 452                                     const struct sw_flow_key *key)
 453 {
 454         u32 __always_unused n_mask_hit;
 455
 456         return ovs_flow_tbl_lookup_stats(tbl, key, &n_mask_hit);
 457 }
 458
 459 struct sw_flow *ovs_flow_tbl_lookup_exact(struct flow_table *tbl,
 460                                           struct sw_flow_match *match)
 461 {
 462         struct table_instance *ti = rcu_dereference_ovsl(tbl->ti);
 463         struct sw_flow_mask *mask;
 464         struct sw_flow *flow;
 465
 466         /* Always called under ovs-mutex. */
 467         list_for_each_entry(mask, &tbl->mask_list, list) {
 468                 flow = masked_flow_lookup(ti, match->key, mask);
 469                 if (flow && ovs_flow_cmp_unmasked_key(flow, match))  /* Found */
 470                         return flow;
 471         }
 472         return NULL;
 473 }
 474
 475 int ovs_flow_tbl_num_masks(const struct flow_table *table)
 476 {
 477         struct sw_flow_mask *mask;
 478         int num = 0;
 479
 480         list_for_each_entry(mask, &table->mask_list, list)
 481                 num++;
 482
 483         return num;
 484 }
 485
 486 static struct table_instance *table_instance_expand(struct table_instance *ti)
 487 {
 488         return table_instance_rehash(ti, ti->n_buckets * 2);
 489 }
 490
 491 /* Remove 'mask' from the mask list, if it is not needed any more. */
 492 static void flow_mask_remove(struct flow_table *tbl, struct sw_flow_mask *mask)
 493 {
 494         if (mask) {
 495                 /* ovs-lock is required to protect mask-refcount and
 496                  * mask list.
 497                  */
 498                 ASSERT_OVSL();
 499                 BUG_ON(!mask->ref_count);
 500                 mask->ref_count--;
 501
 502                 if (!mask->ref_count) {
 503                         list_del_rcu(&mask->list);
 504                         kfree_rcu(mask, rcu);
 505                 }
 506         }
 507 }
 508
 509 /* Must be called with OVS mutex held. */
 510 void ovs_flow_tbl_remove(struct flow_table *table, struct sw_flow *flow)
 511 {
 512         struct table_instance *ti = ovsl_dereference(table->ti);
 513
 514         BUG_ON(table->count == 0);
 515         hlist_del_rcu(&flow->hash_node[ti->node_ver]);
 516         table->count--;
 517
 518         /* RCU delete the mask. 'flow->mask' is not NULLed, as it should be
 519          * accessible as long as the RCU read lock is held.
 520          */
 521         flow_mask_remove(table, flow->mask);
 522 }
 523
 524 static struct sw_flow_mask *mask_alloc(void)
 525 {
 526         struct sw_flow_mask *mask;
 527
 528         mask = kmalloc(sizeof(*mask), GFP_KERNEL);
 529         if (mask)
 530                 mask->ref_count = 1;
 531
 532         return mask;
 533 }
 534
 535 static bool mask_equal(const struct sw_flow_mask *a,
 536                        const struct sw_flow_mask *b)
 537 {
 538         const u8 *a_ = (const u8 *)&a->key + a->range.start;
 539         const u8 *b_ = (const u8 *)&b->key + b->range.start;
 540
 541         return  (a->range.end == b->range.end)
 542                 && (a->range.start == b->range.start)
 543                 && (memcmp(a_, b_, range_n_bytes(&a->range)) == 0);
 544 }
 545
 546 static struct sw_flow_mask *flow_mask_find(const struct flow_table *tbl,
 547                                            const struct sw_flow_mask *mask)
 548 {
 549         struct list_head *ml;
 550
 551         list_for_each(ml, &tbl->mask_list) {
 552                 struct sw_flow_mask *m;
 553                 m = container_of(ml, struct sw_flow_mask, list);
 554                 if (mask_equal(mask, m))
 555                         return m;
 556         }
 557
 558         return NULL;
 559 }
 560
 561 /* Add 'mask' into the mask list, if it is not already there. */
 562 static int flow_mask_insert(struct flow_table *tbl, struct sw_flow *flow,
 563                             struct sw_flow_mask *new)
 564 {
 565         struct sw_flow_mask *mask;
 566         mask = flow_mask_find(tbl, new);
 567         if (!mask) {
 568                 /* Allocate a new mask if none exsits. */
 569                 mask = mask_alloc();
 570                 if (!mask)
 571                         return -ENOMEM;
 572                 mask->key = new->key;
 573                 mask->range = new->range;
 574                 list_add_rcu(&mask->list, &tbl->mask_list);
 575         } else {
 576                 BUG_ON(!mask->ref_count);
 577                 mask->ref_count++;
 578         }
 579
 580         flow->mask = mask;
 581         return 0;
 582 }
 583
 584 /* Must be called with OVS mutex held. */
 585 int ovs_flow_tbl_insert(struct flow_table *table, struct sw_flow *flow,
 586                         struct sw_flow_mask *mask)
 587 {
 588         struct table_instance *new_ti = NULL;
 589         struct table_instance *ti;
 590         int err;
 591
 592         err = flow_mask_insert(table, flow, mask);
 593         if (err)
 594                 return err;
 595
 596         flow->hash = flow_hash(&flow->key, flow->mask->range.start,
 597                         flow->mask->range.end);
 598         ti = ovsl_dereference(table->ti);
 599         table_instance_insert(ti, flow);
 600         table->count++;
 601
 602         /* Expand table, if necessary, to make room. */
 603         if (table->count > ti->n_buckets)
 604                 new_ti = table_instance_expand(ti);
 605         else if (time_after(jiffies, table->last_rehash + REHASH_INTERVAL))
 606                 new_ti = table_instance_rehash(ti, ti->n_buckets);
 607
 608         if (new_ti) {
 609                 rcu_assign_pointer(table->ti, new_ti);
 610                 table_instance_destroy(ti, true);
 611                 table->last_rehash = jiffies;
 612         }
 613         return 0;
 614 }
 615
 616 /* Initializes the flow module.
 617  * Returns zero if successful or a negative error code. */
 618 int ovs_flow_init(void)
 619 {
 620         BUILD_BUG_ON(__alignof__(struct sw_flow_key) % __alignof__(long));
 621         BUILD_BUG_ON(sizeof(struct sw_flow_key) % sizeof(long));
 622
 623         flow_cache = kmem_cache_create("sw_flow", sizeof(struct sw_flow)
 624                                        + (num_possible_nodes()
 625                                           * sizeof(struct flow_stats *)),
 626                                        0, 0, NULL);
 627         if (flow_cache == NULL)
 628                 return -ENOMEM;
 629
 630         flow_stats_cache
 631                 = kmem_cache_create("sw_flow_stats", sizeof(struct flow_stats),
 632                                     0, SLAB_HWCACHE_ALIGN, NULL);
 633         if (flow_stats_cache == NULL) {
 634                 kmem_cache_destroy(flow_cache);
 635                 flow_cache = NULL;
 636                 return -ENOMEM;
 637         }
 638
 639         return 0;
 640 }
 641
 642 /* Uninitializes the flow module. */
 643 void ovs_flow_exit(void)
 644 {
 645         kmem_cache_destroy(flow_stats_cache);
 646         kmem_cache_destroy(flow_cache);
 647 }