lib/mac-learning.c

   1 /*
   2  * Copyright (c) 2008, 2009, 2010, 2011, 2012 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 "tag.h"
  29 #include "timeval.h"
  30 #include "unaligned.h"
  31 #include "util.h"
  32 #include "vlan-bitmap.h"
  33 #include "vlog.h"
  34
  35 VLOG_DEFINE_THIS_MODULE(mac_learning);
  36
  37 COVERAGE_DEFINE(mac_learning_learned);
  38 COVERAGE_DEFINE(mac_learning_expired);
  39
  40 /* Returns the number of seconds since 'e' (within 'ml') was last learned. */
  41 int
  42 mac_entry_age(const struct mac_learning *ml, const struct mac_entry *e)
  43 {
  44     time_t remaining = e->expires - time_now();
  45     return ml->idle_time - remaining;
  46 }
  47
  48 static uint32_t
  49 mac_table_hash(const struct mac_learning *ml, const uint8_t mac[ETH_ADDR_LEN],
  50                uint16_t vlan)
  51 {
  52     unsigned int mac1 = get_unaligned_u32(ALIGNED_CAST(uint32_t *, mac));
  53     unsigned int mac2 = get_unaligned_u16(ALIGNED_CAST(uint16_t *, mac + 4));
  54     return hash_3words(mac1, mac2 | (vlan << 16), ml->secret);
  55 }
  56
  57 static struct mac_entry *
  58 mac_entry_from_lru_node(struct list *list)
  59 {
  60     return CONTAINER_OF(list, struct mac_entry, lru_node);
  61 }
  62
  63 /* Returns a tag that represents that 'mac' is on an unknown port in 'vlan'.
  64  * (When we learn where 'mac' is in 'vlan', this allows flows that were
  65  * flooded to be revalidated.) */
  66 static tag_type
  67 make_unknown_mac_tag(const struct mac_learning *ml,
  68                      const uint8_t mac[ETH_ADDR_LEN], uint16_t vlan)
  69 {
  70     return tag_create_deterministic(mac_table_hash(ml, mac, vlan));
  71 }
  72
  73 static struct mac_entry *
  74 mac_entry_lookup(const struct mac_learning *ml,
  75                  const uint8_t mac[ETH_ADDR_LEN], uint16_t vlan)
  76 {
  77     struct mac_entry *e;
  78
  79     HMAP_FOR_EACH_WITH_HASH (e, hmap_node, mac_table_hash(ml, mac, vlan),
  80                              &ml->table) {
  81         if (e->vlan == vlan && eth_addr_equals(e->mac, mac)) {
  82             return e;
  83         }
  84     }
  85     return NULL;
  86 }
  87
  88 /* If the LRU list is not empty, stores the least-recently-used entry in '*e'
  89  * and returns true.  Otherwise, if the LRU list is empty, stores NULL in '*e'
  90  * and return false. */
  91 static bool
  92 get_lru(struct mac_learning *ml, struct mac_entry **e)
  93     OVS_REQ_RDLOCK(ml->rwlock)
  94 {
  95     if (!list_is_empty(&ml->lrus)) {
  96         *e = mac_entry_from_lru_node(ml->lrus.next);
  97         return true;
  98     } else {
  99         *e = NULL;
 100         return false;
 101     }
 102 }
 103
 104 static unsigned int
 105 normalize_idle_time(unsigned int idle_time)
 106 {
 107     return (idle_time < 15 ? 15
 108             : idle_time > 3600 ? 3600
 109             : idle_time);
 110 }
 111
 112 /* Creates and returns a new MAC learning table with an initial MAC aging
 113  * timeout of 'idle_time' seconds and an initial maximum of MAC_DEFAULT_MAX
 114  * entries. */
 115 struct mac_learning *
 116 mac_learning_create(unsigned int idle_time)
 117 {
 118     struct mac_learning *ml;
 119
 120     ml = xmalloc(sizeof *ml);
 121     list_init(&ml->lrus);
 122     hmap_init(&ml->table);
 123     ml->secret = random_uint32();
 124     ml->flood_vlans = NULL;
 125     ml->idle_time = normalize_idle_time(idle_time);
 126     ml->max_entries = MAC_DEFAULT_MAX;
 127     tag_set_init(&ml->tags);
 128     atomic_init(&ml->ref_cnt, 1);
 129     ovs_rwlock_init(&ml->rwlock);
 130     return ml;
 131 }
 132
 133 struct mac_learning *
 134 mac_learning_ref(const struct mac_learning *ml_)
 135 {
 136     struct mac_learning *ml = CONST_CAST(struct mac_learning *, ml_);
 137     if (ml) {
 138         int orig;
 139         atomic_add(&ml->ref_cnt, 1, &orig);
 140         ovs_assert(orig > 0);
 141     }
 142     return ml;
 143 }
 144
 145 /* Unreferences (and possibly destroys) MAC learning table 'ml'. */
 146 void
 147 mac_learning_unref(struct mac_learning *ml)
 148 {
 149     int orig;
 150
 151     if (!ml) {
 152         return;
 153     }
 154
 155     atomic_sub(&ml->ref_cnt, 1, &orig);
 156     ovs_assert(orig > 0);
 157     if (orig == 1) {
 158         struct mac_entry *e, *next;
 159
 160         HMAP_FOR_EACH_SAFE (e, next, hmap_node, &ml->table) {
 161             hmap_remove(&ml->table, &e->hmap_node);
 162             free(e);
 163         }
 164         hmap_destroy(&ml->table);
 165
 166         bitmap_free(ml->flood_vlans);
 167         ovs_rwlock_destroy(&ml->rwlock);
 168         free(ml);
 169     }
 170 }
 171
 172 /* Provides a bitmap of VLANs which have learning disabled, that is, VLANs on
 173  * which all packets are flooded.  Returns true if the set has changed from the
 174  * previous value. */
 175 bool
 176 mac_learning_set_flood_vlans(struct mac_learning *ml,
 177                              const unsigned long *bitmap)
 178 {
 179     if (vlan_bitmap_equal(ml->flood_vlans, bitmap)) {
 180         return false;
 181     } else {
 182         bitmap_free(ml->flood_vlans);
 183         ml->flood_vlans = vlan_bitmap_clone(bitmap);
 184         return true;
 185     }
 186 }
 187
 188 /* Changes the MAC aging timeout of 'ml' to 'idle_time' seconds. */
 189 void
 190 mac_learning_set_idle_time(struct mac_learning *ml, unsigned int idle_time)
 191 {
 192     idle_time = normalize_idle_time(idle_time);
 193     if (idle_time != ml->idle_time) {
 194         struct mac_entry *e;
 195         int delta;
 196
 197         delta = (int) idle_time - (int) ml->idle_time;
 198         LIST_FOR_EACH (e, lru_node, &ml->lrus) {
 199             e->expires += delta;
 200         }
 201         ml->idle_time = idle_time;
 202     }
 203 }
 204
 205 /* Sets the maximum number of entries in 'ml' to 'max_entries', adjusting it
 206  * to be within a reasonable range. */
 207 void
 208 mac_learning_set_max_entries(struct mac_learning *ml, size_t max_entries)
 209 {
 210     ml->max_entries = (max_entries < 10 ? 10
 211                        : max_entries > 1000 * 1000 ? 1000 * 1000
 212                        : max_entries);
 213 }
 214
 215 static bool
 216 is_learning_vlan(const struct mac_learning *ml, uint16_t vlan)
 217 {
 218     return !ml->flood_vlans || !bitmap_is_set(ml->flood_vlans, vlan);
 219 }
 220
 221 /* Returns true if 'src_mac' may be learned on 'vlan' for 'ml'.
 222  * Returns false if 'ml' is NULL, if src_mac is not valid for learning, or if
 223  * 'vlan' is configured on 'ml' to flood all packets. */
 224 bool
 225 mac_learning_may_learn(const struct mac_learning *ml,
 226                        const uint8_t src_mac[ETH_ADDR_LEN], uint16_t vlan)
 227 {
 228     return ml && is_learning_vlan(ml, vlan) && !eth_addr_is_multicast(src_mac);
 229 }
 230
 231 /* Searches 'ml' for and returns a MAC learning entry for 'src_mac' in 'vlan',
 232  * inserting a new entry if necessary.  The caller must have already verified,
 233  * by calling mac_learning_may_learn(), that 'src_mac' and 'vlan' are
 234  * learnable.
 235  *
 236  * If the returned MAC entry is new (as may be determined by calling
 237  * mac_entry_is_new()), then the caller must pass the new entry to
 238  * mac_learning_changed().  The caller must also initialize the new entry's
 239  * 'port' member.  Otherwise calling those functions is at the caller's
 240  * discretion. */
 241 struct mac_entry *
 242 mac_learning_insert(struct mac_learning *ml,
 243                     const uint8_t src_mac[ETH_ADDR_LEN], uint16_t vlan)
 244 {
 245     struct mac_entry *e;
 246
 247     e = mac_entry_lookup(ml, src_mac, vlan);
 248     if (!e) {
 249         uint32_t hash = mac_table_hash(ml, src_mac, vlan);
 250
 251         if (hmap_count(&ml->table) >= ml->max_entries) {
 252             get_lru(ml, &e);
 253             mac_learning_expire(ml, e);
 254         }
 255
 256         e = xmalloc(sizeof *e);
 257         hmap_insert(&ml->table, &e->hmap_node, hash);
 258         memcpy(e->mac, src_mac, ETH_ADDR_LEN);
 259         e->vlan = vlan;
 260         e->tag = 0;
 261         e->grat_arp_lock = TIME_MIN;
 262     } else {
 263         list_remove(&e->lru_node);
 264     }
 265
 266     /* Mark 'e' as recently used. */
 267     list_push_back(&ml->lrus, &e->lru_node);
 268     e->expires = time_now() + ml->idle_time;
 269
 270     return e;
 271 }
 272
 273 /* Changes 'e''s tag to a new, randomly selected one.  Causes
 274  * mac_learning_run() to flag for revalidation the tag that would have been
 275  * previously used for this entry's MAC and VLAN (either before 'e' was
 276  * inserted, if it is new, or otherwise before its port was updated.)
 277  *
 278  * The client should call this function after obtaining a MAC learning entry
 279  * from mac_learning_insert(), if the entry is either new or if its learned
 280  * port has changed. */
 281 void
 282 mac_learning_changed(struct mac_learning *ml, struct mac_entry *e)
 283 {
 284     tag_type tag = e->tag ? e->tag : make_unknown_mac_tag(ml, e->mac, e->vlan);
 285
 286     COVERAGE_INC(mac_learning_learned);
 287
 288     e->tag = tag_create_random();
 289     tag_set_add(&ml->tags, tag);
 290 }
 291
 292 /* Looks up MAC 'dst' for VLAN 'vlan' in 'ml' and returns the associated MAC
 293  * learning entry, if any.  If 'tag' is nonnull, then the tag that associates
 294  * 'dst' and 'vlan' with its currently learned port will be OR'd into
 295  * '*tag'. */
 296 struct mac_entry *
 297 mac_learning_lookup(const struct mac_learning *ml,
 298                     const uint8_t dst[ETH_ADDR_LEN], uint16_t vlan,
 299                     tag_type *tag)
 300 {
 301     if (eth_addr_is_multicast(dst)) {
 302         /* No tag because the treatment of multicast destinations never
 303          * changes. */
 304         return NULL;
 305     } else if (!is_learning_vlan(ml, vlan)) {
 306         /* We don't tag this property.  The set of learning VLANs changes so
 307          * rarely that we revalidate every flow when it changes. */
 308         return NULL;
 309     } else {
 310         struct mac_entry *e = mac_entry_lookup(ml, dst, vlan);
 311
 312         ovs_assert(e == NULL || e->tag != 0);
 313         if (tag) {
 314             /* Tag either the learned port or the lack thereof. */
 315             *tag |= e ? e->tag : make_unknown_mac_tag(ml, dst, vlan);
 316         }
 317         return e;
 318     }
 319 }
 320
 321 /* Expires 'e' from the 'ml' hash table. */
 322 void
 323 mac_learning_expire(struct mac_learning *ml, struct mac_entry *e)
 324 {
 325     hmap_remove(&ml->table, &e->hmap_node);
 326     list_remove(&e->lru_node);
 327     free(e);
 328 }
 329
 330 /* Expires all the mac-learning entries in 'ml'.  If not NULL, the tags in 'ml'
 331  * are added to 'tags'.  Otherwise the tags in 'ml' are discarded.  The client
 332  * is responsible for revalidating any flows that depend on 'ml', if
 333  * necessary. */
 334 void
 335 mac_learning_flush(struct mac_learning *ml, struct tag_set *tags)
 336 {
 337     struct mac_entry *e;
 338     while (get_lru(ml, &e)){
 339         if (tags) {
 340             tag_set_add(tags, e->tag);
 341         }
 342         mac_learning_expire(ml, e);
 343     }
 344     hmap_shrink(&ml->table);
 345 }
 346
 347 void
 348 mac_learning_run(struct mac_learning *ml, struct tag_set *set)
 349 {
 350     struct mac_entry *e;
 351
 352     if (set) {
 353         tag_set_union(set, &ml->tags);
 354     }
 355     tag_set_init(&ml->tags);
 356
 357     while (get_lru(ml, &e)
 358            && (hmap_count(&ml->table) > ml->max_entries
 359                || time_now() >= e->expires)) {
 360         COVERAGE_INC(mac_learning_expired);
 361         if (set) {
 362             tag_set_add(set, e->tag);
 363         }
 364         mac_learning_expire(ml, e);
 365     }
 366 }
 367
 368 void
 369 mac_learning_wait(struct mac_learning *ml)
 370 {
 371     if (hmap_count(&ml->table) > ml->max_entries
 372         || !tag_set_is_empty(&ml->tags)) {
 373         poll_immediate_wake();
 374     } else if (!list_is_empty(&ml->lrus)) {
 375         struct mac_entry *e = mac_entry_from_lru_node(ml->lrus.next);
 376         poll_timer_wait_until(e->expires * 1000LL);
 377     }
 378 }