2 * Copyright (c) 2007-2013 Nicira, Inc.
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.
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.
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
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/jhash.h>
29 #include <linux/jiffies.h>
30 #include <linux/llc.h>
31 #include <linux/module.h>
33 #include <linux/rcupdate.h>
34 #include <linux/if_arp.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>
45 #include <net/ndisc.h>
50 #define TBL_MIN_BUCKETS 1024
51 #define REHASH_INTERVAL (10 * 60 * HZ)
53 static struct kmem_cache *flow_cache;
55 static u16 range_n_bytes(const struct sw_flow_key_range *range)
57 return range->end - range->start;
60 void ovs_flow_mask_key(struct sw_flow_key *dst, const struct sw_flow_key *src,
61 const struct sw_flow_mask *mask)
63 const long *m = (long *)((u8 *)&mask->key + mask->range.start);
64 const long *s = (long *)((u8 *)src + mask->range.start);
65 long *d = (long *)((u8 *)dst + mask->range.start);
68 /* The memory outside of the 'mask->range' are not set since
69 * further operations on 'dst' only uses contents within
72 for (i = 0; i < range_n_bytes(&mask->range); i += sizeof(long))
76 struct sw_flow *ovs_flow_alloc(void)
81 flow = kmem_cache_alloc(flow_cache, GFP_KERNEL);
83 return ERR_PTR(-ENOMEM);
88 memset(flow->stats, 0, num_possible_cpus() * sizeof(struct sw_flow_stats));
89 for_each_possible_cpu(cpu)
90 spin_lock_init(&flow->stats[cpu].lock);
95 int ovs_flow_tbl_count(struct flow_table *table)
100 static struct flex_array *alloc_buckets(unsigned int n_buckets)
102 struct flex_array *buckets;
105 buckets = flex_array_alloc(sizeof(struct hlist_head),
106 n_buckets, GFP_KERNEL);
110 err = flex_array_prealloc(buckets, 0, n_buckets, GFP_KERNEL);
112 flex_array_free(buckets);
116 for (i = 0; i < n_buckets; i++)
117 INIT_HLIST_HEAD((struct hlist_head *)
118 flex_array_get(buckets, i));
123 static void flow_free(struct sw_flow *flow)
125 kfree((struct sf_flow_acts __force *)flow->sf_acts);
126 kmem_cache_free(flow_cache, flow);
129 static void rcu_free_flow_callback(struct rcu_head *rcu)
131 struct sw_flow *flow = container_of(rcu, struct sw_flow, rcu);
136 static void rcu_free_sw_flow_mask_cb(struct rcu_head *rcu)
138 struct sw_flow_mask *mask = container_of(rcu, struct sw_flow_mask, rcu);
143 static void flow_mask_del_ref(struct sw_flow_mask *mask, bool deferred)
148 BUG_ON(!mask->ref_count);
151 if (!mask->ref_count) {
152 list_del_rcu(&mask->list);
154 call_rcu(&mask->rcu, rcu_free_sw_flow_mask_cb);
160 void ovs_flow_free(struct sw_flow *flow, bool deferred)
165 flow_mask_del_ref(flow->mask, deferred);
168 call_rcu(&flow->rcu, rcu_free_flow_callback);
173 static void free_buckets(struct flex_array *buckets)
175 flex_array_free(buckets);
178 static void __table_instance_destroy(struct table_instance *ti)
185 for (i = 0; i < ti->n_buckets; i++) {
186 struct sw_flow *flow;
187 struct hlist_head *head = flex_array_get(ti->buckets, i);
188 struct hlist_node *n;
189 int ver = ti->node_ver;
191 hlist_for_each_entry_safe(flow, n, head, hash_node[ver]) {
192 hlist_del(&flow->hash_node[ver]);
193 ovs_flow_free(flow, false);
198 free_buckets(ti->buckets);
202 static struct table_instance *table_instance_alloc(int new_size)
204 struct table_instance *ti = kmalloc(sizeof(*ti), GFP_KERNEL);
209 ti->buckets = alloc_buckets(new_size);
215 ti->n_buckets = new_size;
217 ti->keep_flows = false;
218 get_random_bytes(&ti->hash_seed, sizeof(u32));
223 int ovs_flow_tbl_init(struct flow_table *table)
225 struct table_instance *ti;
227 ti = table_instance_alloc(TBL_MIN_BUCKETS);
232 rcu_assign_pointer(table->ti, ti);
233 INIT_LIST_HEAD(&table->mask_list);
234 table->last_rehash = jiffies;
239 static void flow_tbl_destroy_rcu_cb(struct rcu_head *rcu)
241 struct table_instance *ti = container_of(rcu, struct table_instance, rcu);
243 __table_instance_destroy(ti);
246 static void table_instance_destroy(struct table_instance *ti, bool deferred)
252 call_rcu(&ti->rcu, flow_tbl_destroy_rcu_cb);
254 __table_instance_destroy(ti);
257 void ovs_flow_tbl_destroy(struct flow_table *table)
259 struct table_instance *ti = ovsl_dereference(table->ti);
261 table_instance_destroy(ti, false);
264 struct sw_flow *ovs_flow_tbl_dump_next(struct table_instance *ti,
265 u32 *bucket, u32 *last)
267 struct sw_flow *flow;
268 struct hlist_head *head;
273 while (*bucket < ti->n_buckets) {
275 head = flex_array_get(ti->buckets, *bucket);
276 hlist_for_each_entry_rcu(flow, head, hash_node[ver]) {
291 static struct hlist_head *find_bucket(struct table_instance *ti, u32 hash)
293 hash = jhash_1word(hash, ti->hash_seed);
294 return flex_array_get(ti->buckets,
295 (hash & (ti->n_buckets - 1)));
298 static void table_instance_insert(struct table_instance *ti, struct sw_flow *flow)
300 struct hlist_head *head;
302 head = find_bucket(ti, flow->hash);
303 hlist_add_head_rcu(&flow->hash_node[ti->node_ver], head);
306 static void flow_table_copy_flows(struct table_instance *old,
307 struct table_instance *new)
312 old_ver = old->node_ver;
313 new->node_ver = !old_ver;
315 /* Insert in new table. */
316 for (i = 0; i < old->n_buckets; i++) {
317 struct sw_flow *flow;
318 struct hlist_head *head;
320 head = flex_array_get(old->buckets, i);
322 hlist_for_each_entry(flow, head, hash_node[old_ver])
323 table_instance_insert(new, flow);
326 old->keep_flows = true;
329 static struct table_instance *table_instance_rehash(struct table_instance *ti,
332 struct table_instance *new_ti;
334 new_ti = table_instance_alloc(n_buckets);
338 flow_table_copy_flows(ti, new_ti);
343 int ovs_flow_tbl_flush(struct flow_table *flow_table)
345 struct table_instance *old_ti;
346 struct table_instance *new_ti;
348 old_ti = ovsl_dereference(flow_table->ti);
349 new_ti = table_instance_alloc(TBL_MIN_BUCKETS);
353 rcu_assign_pointer(flow_table->ti, new_ti);
354 flow_table->last_rehash = jiffies;
355 flow_table->count = 0;
357 table_instance_destroy(old_ti, true);
361 static u32 flow_hash(const struct sw_flow_key *key, int key_start,
364 u32 *hash_key = (u32 *)((u8 *)key + key_start);
365 int hash_u32s = (key_end - key_start) >> 2;
367 /* Make sure number of hash bytes are multiple of u32. */
368 BUILD_BUG_ON(sizeof(long) % sizeof(u32));
370 return jhash2(hash_key, hash_u32s, 0);
373 static int flow_key_start(const struct sw_flow_key *key)
375 if (key->tun_key.ipv4_dst)
378 return rounddown(offsetof(struct sw_flow_key, phy),
382 static bool cmp_key(const struct sw_flow_key *key1,
383 const struct sw_flow_key *key2,
384 int key_start, int key_end)
386 const long *cp1 = (long *)((u8 *)key1 + key_start);
387 const long *cp2 = (long *)((u8 *)key2 + key_start);
391 for (i = key_start; i < key_end; i += sizeof(long))
392 diffs |= *cp1++ ^ *cp2++;
397 static bool flow_cmp_masked_key(const struct sw_flow *flow,
398 const struct sw_flow_key *key,
399 int key_start, int key_end)
401 return cmp_key(&flow->key, key, key_start, key_end);
404 bool ovs_flow_cmp_unmasked_key(const struct sw_flow *flow,
405 struct sw_flow_match *match)
407 struct sw_flow_key *key = match->key;
408 int key_start = flow_key_start(key);
409 int key_end = match->range.end;
411 return cmp_key(&flow->unmasked_key, key, key_start, key_end);
414 static struct sw_flow *masked_flow_lookup(struct table_instance *ti,
415 const struct sw_flow_key *unmasked,
416 struct sw_flow_mask *mask)
418 struct sw_flow *flow;
419 struct hlist_head *head;
420 int key_start = mask->range.start;
421 int key_end = mask->range.end;
423 struct sw_flow_key masked_key;
425 ovs_flow_mask_key(&masked_key, unmasked, mask);
426 hash = flow_hash(&masked_key, key_start, key_end);
427 head = find_bucket(ti, hash);
428 hlist_for_each_entry_rcu(flow, head, hash_node[ti->node_ver]) {
429 if (flow->mask == mask && flow->hash == hash &&
430 flow_cmp_masked_key(flow, &masked_key,
437 struct sw_flow *ovs_flow_tbl_lookup(struct flow_table *tbl,
438 const struct sw_flow_key *key,
441 struct table_instance *ti = rcu_dereference(tbl->ti);
442 struct sw_flow_mask *mask;
443 struct sw_flow *flow;
446 list_for_each_entry_rcu(mask, &tbl->mask_list, list) {
448 flow = masked_flow_lookup(ti, key, mask);
449 if (flow) /* Found */
455 int ovs_flow_tbl_num_masks(const struct flow_table *table)
457 struct sw_flow_mask *mask;
460 list_for_each_entry(mask, &table->mask_list, list)
466 static struct table_instance *table_instance_expand(struct table_instance *ti)
468 return table_instance_rehash(ti, ti->n_buckets * 2);
471 void ovs_flow_tbl_remove(struct flow_table *table, struct sw_flow *flow)
473 struct table_instance *ti = ovsl_dereference(table->ti);
475 BUG_ON(table->count == 0);
476 hlist_del_rcu(&flow->hash_node[ti->node_ver]);
480 static struct sw_flow_mask *mask_alloc(void)
482 struct sw_flow_mask *mask;
484 mask = kmalloc(sizeof(*mask), GFP_KERNEL);
491 static void mask_add_ref(struct sw_flow_mask *mask)
496 static bool mask_equal(const struct sw_flow_mask *a,
497 const struct sw_flow_mask *b)
499 u8 *a_ = (u8 *)&a->key + a->range.start;
500 u8 *b_ = (u8 *)&b->key + b->range.start;
502 return (a->range.end == b->range.end)
503 && (a->range.start == b->range.start)
504 && (memcmp(a_, b_, range_n_bytes(&a->range)) == 0);
507 static struct sw_flow_mask *flow_mask_find(const struct flow_table *tbl,
508 const struct sw_flow_mask *mask)
510 struct list_head *ml;
512 list_for_each(ml, &tbl->mask_list) {
513 struct sw_flow_mask *m;
514 m = container_of(ml, struct sw_flow_mask, list);
515 if (mask_equal(mask, m))
523 * add a new mask into the mask list.
524 * The caller needs to make sure that 'mask' is not the same
525 * as any masks that are already on the list.
527 static int flow_mask_insert(struct flow_table *tbl, struct sw_flow *flow,
528 struct sw_flow_mask *new)
530 struct sw_flow_mask *mask;
531 mask = flow_mask_find(tbl, new);
533 /* Allocate a new mask if none exsits. */
537 mask->key = new->key;
538 mask->range = new->range;
539 list_add_rcu(&mask->list, &tbl->mask_list);
547 int ovs_flow_tbl_insert(struct flow_table *table, struct sw_flow *flow,
548 struct sw_flow_mask *mask)
550 struct table_instance *new_ti = NULL;
551 struct table_instance *ti;
554 err = flow_mask_insert(table, flow, mask);
558 flow->hash = flow_hash(&flow->key, flow->mask->range.start,
559 flow->mask->range.end);
560 ti = ovsl_dereference(table->ti);
561 table_instance_insert(ti, flow);
564 /* Expand table, if necessary, to make room. */
565 if (table->count > ti->n_buckets)
566 new_ti = table_instance_expand(ti);
567 else if (time_after(jiffies, table->last_rehash + REHASH_INTERVAL))
568 new_ti = table_instance_rehash(ti, ti->n_buckets);
571 rcu_assign_pointer(table->ti, new_ti);
572 table_instance_destroy(ti, true);
573 table->last_rehash = jiffies;
578 /* Initializes the flow module.
579 * Returns zero if successful or a negative error code. */
580 int ovs_flow_init(void)
584 BUILD_BUG_ON(__alignof__(struct sw_flow_key) % __alignof__(long));
585 BUILD_BUG_ON(sizeof(struct sw_flow_key) % sizeof(long));
587 flow_size = sizeof(struct sw_flow) +
588 (num_possible_cpus() * sizeof(struct sw_flow_stats));
590 flow_cache = kmem_cache_create("sw_flow", flow_size, 0, 0, NULL);
591 if (flow_cache == NULL)
597 /* Uninitializes the flow module. */
598 void ovs_flow_exit(void)
600 kmem_cache_destroy(flow_cache);