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