lib/mac-learning.c

   1 /*
   2  * Copyright (c) 2008, 2009, 2010, 2011, 2012, 2013 Nicira, Inc.
   3  *
   4  * Licensed under the Apache License, Version 2.0 (the "License");
   5  * you may not use this file except in compliance with the License.
   6  * You may obtain a copy of the License at:
   7  *
   8  *     http://www.apache.org/licenses/LICENSE-2.0
   9  *
  10  * Unless required by applicable law or agreed to in writing, software
  11  * distributed under the License is distributed on an "AS IS" BASIS,
  12  * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
  13  * See the License for the specific language governing permissions and
  14  * limitations under the License.
  15  */
  16
  17 #include <config.h>
  18 #include "mac-learning.h"
  19
  20 #include <inttypes.h>
  21 #include <stdlib.h>
  22
  23 #include "bitmap.h"
  24 #include "coverage.h"
  25 #include "hash.h"
  26 #include "list.h"
  27 #include "poll-loop.h"
  28 #include "timeval.h"
  29 #include "unaligned.h"
  30 #include "util.h"
  31 #include "vlan-bitmap.h"
  32
  33 COVERAGE_DEFINE(mac_learning_learned);
  34 COVERAGE_DEFINE(mac_learning_expired);
  35
  36 /* Returns the number of seconds since 'e' (within 'ml') was last learned. */
  37 int
  38 mac_entry_age(const struct mac_learning *ml, const struct mac_entry *e)
  39 {
  40     time_t remaining = e->expires - time_now();
  41     return ml->idle_time - remaining;
  42 }
  43
  44 static uint32_t
  45 mac_table_hash(const struct mac_learning *ml, const uint8_t mac[ETH_ADDR_LEN],
  46                uint16_t vlan)
  47 {
  48     unsigned int mac1 = get_unaligned_u32(ALIGNED_CAST(uint32_t *, mac));
  49     unsigned int mac2 = get_unaligned_u16(ALIGNED_CAST(uint16_t *, mac + 4));
  50     return hash_3words(mac1, mac2 | (vlan << 16), ml->secret);
  51 }
  52
  53 static struct mac_entry *
  54 mac_entry_from_lru_node(struct list *list)
  55 {
  56     return CONTAINER_OF(list, struct mac_entry, lru_node);
  57 }
  58
  59 static struct mac_entry *
  60 mac_entry_lookup(const struct mac_learning *ml,
  61                  const uint8_t mac[ETH_ADDR_LEN], uint16_t vlan)
  62 {
  63     struct mac_entry *e;
  64
  65     HMAP_FOR_EACH_WITH_HASH (e, hmap_node, mac_table_hash(ml, mac, vlan),
  66                              &ml->table) {
  67         if (e->vlan == vlan && eth_addr_equals(e->mac, mac)) {
  68             return e;
  69         }
  70     }
  71     return NULL;
  72 }
  73
  74 /* If the LRU list is not empty, stores the least-recently-used entry in '*e'
  75  * and returns true.  Otherwise, if the LRU list is empty, stores NULL in '*e'
  76  * and return false. */
  77 static bool
  78 get_lru(struct mac_learning *ml, struct mac_entry **e)
  79     OVS_REQ_RDLOCK(ml->rwlock)
  80 {
  81     if (!list_is_empty(&ml->lrus)) {
  82         *e = mac_entry_from_lru_node(ml->lrus.next);
  83         return true;
  84     } else {
  85         *e = NULL;
  86         return false;
  87     }
  88 }
  89
  90 static unsigned int
  91 normalize_idle_time(unsigned int idle_time)
  92 {
  93     return (idle_time < 15 ? 15
  94             : idle_time > 3600 ? 3600
  95             : idle_time);
  96 }
  97
  98 /* Creates and returns a new MAC learning table with an initial MAC aging
  99  * timeout of 'idle_time' seconds and an initial maximum of MAC_DEFAULT_MAX
 100  * entries. */
 101 struct mac_learning *
 102 mac_learning_create(unsigned int idle_time)
 103 {
 104     struct mac_learning *ml;
 105
 106     ml = xmalloc(sizeof *ml);
 107     list_init(&ml->lrus);
 108     hmap_init(&ml->table);
 109     ml->secret = random_uint32();
 110     ml->flood_vlans = NULL;
 111     ml->idle_time = normalize_idle_time(idle_time);
 112     ml->max_entries = MAC_DEFAULT_MAX;
 113     ml->need_revalidate = false;
 114     atomic_init(&ml->ref_cnt, 1);
 115     ovs_rwlock_init(&ml->rwlock);
 116     return ml;
 117 }
 118
 119 struct mac_learning *
 120 mac_learning_ref(const struct mac_learning *ml_)
 121 {
 122     struct mac_learning *ml = CONST_CAST(struct mac_learning *, ml_);
 123     if (ml) {
 124         int orig;
 125         atomic_add(&ml->ref_cnt, 1, &orig);
 126         ovs_assert(orig > 0);
 127     }
 128     return ml;
 129 }
 130
 131 /* Unreferences (and possibly destroys) MAC learning table 'ml'. */
 132 void
 133 mac_learning_unref(struct mac_learning *ml)
 134 {
 135     int orig;
 136
 137     if (!ml) {
 138         return;
 139     }
 140
 141     atomic_sub(&ml->ref_cnt, 1, &orig);
 142     ovs_assert(orig > 0);
 143     if (orig == 1) {
 144         struct mac_entry *e, *next;
 145
 146         HMAP_FOR_EACH_SAFE (e, next, hmap_node, &ml->table) {
 147             hmap_remove(&ml->table, &e->hmap_node);
 148             free(e);
 149         }
 150         hmap_destroy(&ml->table);
 151
 152         bitmap_free(ml->flood_vlans);
 153         ovs_rwlock_destroy(&ml->rwlock);
 154         atomic_destroy(&ml->ref_cnt);
 155         free(ml);
 156     }
 157 }
 158
 159 /* Provides a bitmap of VLANs which have learning disabled, that is, VLANs on
 160  * which all packets are flooded.  Returns true if the set has changed from the
 161  * previous value. */
 162 bool
 163 mac_learning_set_flood_vlans(struct mac_learning *ml,
 164                              const unsigned long *bitmap)
 165 {
 166     if (vlan_bitmap_equal(ml->flood_vlans, bitmap)) {
 167         return false;
 168     } else {
 169         bitmap_free(ml->flood_vlans);
 170         ml->flood_vlans = vlan_bitmap_clone(bitmap);
 171         return true;
 172     }
 173 }
 174
 175 /* Changes the MAC aging timeout of 'ml' to 'idle_time' seconds. */
 176 void
 177 mac_learning_set_idle_time(struct mac_learning *ml, unsigned int idle_time)
 178 {
 179     idle_time = normalize_idle_time(idle_time);
 180     if (idle_time != ml->idle_time) {
 181         struct mac_entry *e;
 182         int delta;
 183
 184         delta = (int) idle_time - (int) ml->idle_time;
 185         LIST_FOR_EACH (e, lru_node, &ml->lrus) {
 186             e->expires += delta;
 187         }
 188         ml->idle_time = idle_time;
 189     }
 190 }
 191
 192 /* Sets the maximum number of entries in 'ml' to 'max_entries', adjusting it
 193  * to be within a reasonable range. */
 194 void
 195 mac_learning_set_max_entries(struct mac_learning *ml, size_t max_entries)
 196 {
 197     ml->max_entries = (max_entries < 10 ? 10
 198                        : max_entries > 1000 * 1000 ? 1000 * 1000
 199                        : max_entries);
 200 }
 201
 202 static bool
 203 is_learning_vlan(const struct mac_learning *ml, uint16_t vlan)
 204 {
 205     return !ml->flood_vlans || !bitmap_is_set(ml->flood_vlans, vlan);
 206 }
 207
 208 /* Returns true if 'src_mac' may be learned on 'vlan' for 'ml'.
 209  * Returns false if 'ml' is NULL, if src_mac is not valid for learning, or if
 210  * 'vlan' is configured on 'ml' to flood all packets. */
 211 bool
 212 mac_learning_may_learn(const struct mac_learning *ml,
 213                        const uint8_t src_mac[ETH_ADDR_LEN], uint16_t vlan)
 214 {
 215     return ml && is_learning_vlan(ml, vlan) && !eth_addr_is_multicast(src_mac);
 216 }
 217
 218 /* Searches 'ml' for and returns a MAC learning entry for 'src_mac' in 'vlan',
 219  * inserting a new entry if necessary.  The caller must have already verified,
 220  * by calling mac_learning_may_learn(), that 'src_mac' and 'vlan' are
 221  * learnable.
 222  *
 223  * If the returned MAC entry is new (as may be determined by calling
 224  * mac_entry_is_new()), then the caller must pass the new entry to
 225  * mac_learning_changed().  The caller must also initialize the new entry's
 226  * 'port' member.  Otherwise calling those functions is at the caller's
 227  * discretion. */
 228 struct mac_entry *
 229 mac_learning_insert(struct mac_learning *ml,
 230                     const uint8_t src_mac[ETH_ADDR_LEN], uint16_t vlan)
 231 {
 232     struct mac_entry *e;
 233
 234     e = mac_entry_lookup(ml, src_mac, vlan);
 235     if (!e) {
 236         uint32_t hash = mac_table_hash(ml, src_mac, vlan);
 237
 238         if (hmap_count(&ml->table) >= ml->max_entries) {
 239             get_lru(ml, &e);
 240             mac_learning_expire(ml, e);
 241         }
 242
 243         e = xmalloc(sizeof *e);
 244         hmap_insert(&ml->table, &e->hmap_node, hash);
 245         memcpy(e->mac, src_mac, ETH_ADDR_LEN);
 246         e->vlan = vlan;
 247         e->grat_arp_lock = TIME_MIN;
 248         e->port.p = NULL;
 249     } else {
 250         list_remove(&e->lru_node);
 251     }
 252
 253     /* Mark 'e' as recently used. */
 254     list_push_back(&ml->lrus, &e->lru_node);
 255     e->expires = time_now() + ml->idle_time;
 256
 257     return e;
 258 }
 259
 260 /* Changes 'e''s tag to a new, randomly selected one.  Causes
 261  * mac_learning_run() to flag for revalidation the tag that would have been
 262  * previously used for this entry's MAC and VLAN (either before 'e' was
 263  * inserted, if it is new, or otherwise before its port was updated.)
 264  *
 265  * The client should call this function after obtaining a MAC learning entry
 266  * from mac_learning_insert(), if the entry is either new or if its learned
 267  * port has changed. */
 268 void
 269 mac_learning_changed(struct mac_learning *ml)
 270 {
 271     COVERAGE_INC(mac_learning_learned);
 272     ml->need_revalidate = true;
 273 }
 274
 275 /* Looks up MAC 'dst' for VLAN 'vlan' in 'ml' and returns the associated MAC
 276  * learning entry, if any.  If 'tag' is nonnull, then the tag that associates
 277  * 'dst' and 'vlan' with its currently learned port will be OR'd into
 278  * '*tag'. */
 279 struct mac_entry *
 280 mac_learning_lookup(const struct mac_learning *ml,
 281                     const uint8_t dst[ETH_ADDR_LEN], uint16_t vlan)
 282 {
 283     if (eth_addr_is_multicast(dst)) {
 284         /* No tag because the treatment of multicast destinations never
 285          * changes. */
 286         return NULL;
 287     } else if (!is_learning_vlan(ml, vlan)) {
 288         /* We don't tag this property.  The set of learning VLANs changes so
 289          * rarely that we revalidate every flow when it changes. */
 290         return NULL;
 291     } else {
 292         struct mac_entry *e = mac_entry_lookup(ml, dst, vlan);
 293
 294         ovs_assert(e == NULL || e->port.p != NULL);
 295         return e;
 296     }
 297 }
 298
 299 /* Expires 'e' from the 'ml' hash table. */
 300 void
 301 mac_learning_expire(struct mac_learning *ml, struct mac_entry *e)
 302 {
 303     hmap_remove(&ml->table, &e->hmap_node);
 304     list_remove(&e->lru_node);
 305     free(e);
 306 }
 307
 308 /* Expires all the mac-learning entries in 'ml'.  If not NULL, the tags in 'ml'
 309  * are added to 'tags'.  Otherwise the tags in 'ml' are discarded.  The client
 310  * is responsible for revalidating any flows that depend on 'ml', if
 311  * necessary. */
 312 void
 313 mac_learning_flush(struct mac_learning *ml)
 314 {
 315     struct mac_entry *e;
 316     while (get_lru(ml, &e)){
 317         ml->need_revalidate = true;
 318         mac_learning_expire(ml, e);
 319     }
 320     hmap_shrink(&ml->table);
 321 }
 322
 323 /* Does periodic work required by 'ml'.  Returns true if something changed that
 324  * may require flow revalidation. */
 325 bool
 326 mac_learning_run(struct mac_learning *ml)
 327 {
 328     bool need_revalidate;
 329     struct mac_entry *e;
 330
 331     while (get_lru(ml, &e)
 332            && (hmap_count(&ml->table) > ml->max_entries
 333                || time_now() >= e->expires)) {
 334         COVERAGE_INC(mac_learning_expired);
 335         ml->need_revalidate = true;
 336         mac_learning_expire(ml, e);
 337     }
 338
 339     need_revalidate = ml->need_revalidate;
 340     ml->need_revalidate = false;
 341     return need_revalidate;
 342 }
 343
 344 void
 345 mac_learning_wait(struct mac_learning *ml)
 346 {
 347     if (hmap_count(&ml->table) > ml->max_entries
 348         || ml->need_revalidate) {
 349         poll_immediate_wake();
 350     } else if (!list_is_empty(&ml->lrus)) {
 351         struct mac_entry *e = mac_entry_from_lru_node(ml->lrus.next);
 352         poll_timer_wait_until(e->expires * 1000LL);
 353     }
 354 }