2 * Copyright (c) 2007-2014 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/hash.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>
47 #define TBL_MIN_BUCKETS 1024
48 #define REHASH_INTERVAL (10 * 60 * HZ)
50 static struct kmem_cache *flow_cache;
51 struct kmem_cache *flow_stats_cache __read_mostly;
53 static u16 range_n_bytes(const struct sw_flow_key_range *range)
55 return range->end - range->start;
58 void ovs_flow_mask_key(struct sw_flow_key *dst, const struct sw_flow_key *src,
59 const struct sw_flow_mask *mask)
61 const long *m = (const long *)((const u8 *)&mask->key +
63 const long *s = (const long *)((const u8 *)src +
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)
79 struct flow_stats *stats;
82 flow = kmem_cache_alloc(flow_cache, GFP_KERNEL);
84 return ERR_PTR(-ENOMEM);
88 flow->stats_last_writer = NUMA_NO_NODE;
90 /* Initialize the default stat node. */
91 stats = kmem_cache_alloc_node(flow_stats_cache,
92 GFP_KERNEL | __GFP_ZERO, 0);
96 spin_lock_init(&stats->lock);
98 RCU_INIT_POINTER(flow->stats[0], stats);
102 RCU_INIT_POINTER(flow->stats[node], NULL);
106 kmem_cache_free(flow_cache, flow);
107 return ERR_PTR(-ENOMEM);
110 int ovs_flow_tbl_count(struct flow_table *table)
115 static struct flex_array *alloc_buckets(unsigned int n_buckets)
117 struct flex_array *buckets;
120 buckets = flex_array_alloc(sizeof(struct hlist_head),
121 n_buckets, GFP_KERNEL);
125 err = flex_array_prealloc(buckets, 0, n_buckets, GFP_KERNEL);
127 flex_array_free(buckets);
131 for (i = 0; i < n_buckets; i++)
132 INIT_HLIST_HEAD((struct hlist_head *)
133 flex_array_get(buckets, i));
138 static void flow_free(struct sw_flow *flow)
142 kfree((struct sw_flow_actions __force *)flow->sf_acts);
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);
150 static void rcu_free_flow_callback(struct rcu_head *rcu)
152 struct sw_flow *flow = container_of(rcu, struct sw_flow, rcu);
157 void ovs_flow_free(struct sw_flow *flow, bool deferred)
163 call_rcu(&flow->rcu, rcu_free_flow_callback);
168 static void free_buckets(struct flex_array *buckets)
170 flex_array_free(buckets);
174 static void __table_instance_destroy(struct table_instance *ti)
176 free_buckets(ti->buckets);
180 static struct table_instance *table_instance_alloc(int new_size)
182 struct table_instance *ti = kmalloc(sizeof(*ti), GFP_KERNEL);
187 ti->buckets = alloc_buckets(new_size);
193 ti->n_buckets = new_size;
195 ti->keep_flows = false;
196 get_random_bytes(&ti->hash_seed, sizeof(u32));
201 int ovs_flow_tbl_init(struct flow_table *table)
203 struct table_instance *ti;
205 ti = table_instance_alloc(TBL_MIN_BUCKETS);
210 rcu_assign_pointer(table->ti, ti);
211 INIT_LIST_HEAD(&table->mask_list);
212 table->last_rehash = jiffies;
217 static void flow_tbl_destroy_rcu_cb(struct rcu_head *rcu)
219 struct table_instance *ti = container_of(rcu, struct table_instance, rcu);
221 __table_instance_destroy(ti);
224 static void table_instance_destroy(struct table_instance *ti, bool deferred)
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;
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);
248 call_rcu(&ti->rcu, flow_tbl_destroy_rcu_cb);
250 __table_instance_destroy(ti);
253 /* No need for locking this function is called from RCU callback or
256 void ovs_flow_tbl_destroy(struct flow_table *table)
258 struct table_instance *ti = rcu_dereference_raw(table->ti);
260 table_instance_destroy(ti, false);
263 struct sw_flow *ovs_flow_tbl_dump_next(struct table_instance *ti,
264 u32 *bucket, u32 *last)
266 struct sw_flow *flow;
267 struct hlist_head *head;
272 while (*bucket < ti->n_buckets) {
274 head = flex_array_get(ti->buckets, *bucket);
275 hlist_for_each_entry_rcu(flow, head, hash_node[ver]) {
290 static struct hlist_head *find_bucket(struct table_instance *ti, u32 hash)
292 hash = jhash_1word(hash, ti->hash_seed);
293 return flex_array_get(ti->buckets,
294 (hash & (ti->n_buckets - 1)));
297 static void table_instance_insert(struct table_instance *ti, struct sw_flow *flow)
299 struct hlist_head *head;
301 head = find_bucket(ti, flow->hash);
302 hlist_add_head_rcu(&flow->hash_node[ti->node_ver], head);
305 static void flow_table_copy_flows(struct table_instance *old,
306 struct table_instance *new)
311 old_ver = old->node_ver;
312 new->node_ver = !old_ver;
314 /* Insert in new table. */
315 for (i = 0; i < old->n_buckets; i++) {
316 struct sw_flow *flow;
317 struct hlist_head *head;
319 head = flex_array_get(old->buckets, i);
321 hlist_for_each_entry(flow, head, hash_node[old_ver])
322 table_instance_insert(new, flow);
325 old->keep_flows = true;
328 static struct table_instance *table_instance_rehash(struct table_instance *ti,
331 struct table_instance *new_ti;
333 new_ti = table_instance_alloc(n_buckets);
337 flow_table_copy_flows(ti, new_ti);
342 int ovs_flow_tbl_flush(struct flow_table *flow_table)
344 struct table_instance *old_ti;
345 struct table_instance *new_ti;
347 old_ti = ovsl_dereference(flow_table->ti);
348 new_ti = table_instance_alloc(TBL_MIN_BUCKETS);
352 rcu_assign_pointer(flow_table->ti, new_ti);
353 flow_table->last_rehash = jiffies;
354 flow_table->count = 0;
356 table_instance_destroy(old_ti, true);
360 static u32 flow_hash(const struct sw_flow_key *key, int key_start,
363 const u32 *hash_key = (const u32 *)((const u8 *)key + key_start);
364 int hash_u32s = (key_end - key_start) >> 2;
366 /* Make sure number of hash bytes are multiple of u32. */
367 BUILD_BUG_ON(sizeof(long) % sizeof(u32));
369 return arch_fast_hash2(hash_key, hash_u32s, 0);
372 static int flow_key_start(const struct sw_flow_key *key)
374 if (key->tun_key.ipv4_dst)
377 return rounddown(offsetof(struct sw_flow_key, phy),
381 static bool cmp_key(const struct sw_flow_key *key1,
382 const struct sw_flow_key *key2,
383 int key_start, int key_end)
385 const long *cp1 = (const long *)((const u8 *)key1 + key_start);
386 const long *cp2 = (const long *)((const u8 *)key2 + key_start);
390 for (i = key_start; i < key_end; i += sizeof(long))
391 diffs |= *cp1++ ^ *cp2++;
396 static bool flow_cmp_masked_key(const struct sw_flow *flow,
397 const struct sw_flow_key *key,
398 int key_start, int key_end)
400 return cmp_key(&flow->key, key, key_start, key_end);
403 bool ovs_flow_cmp_unmasked_key(const struct sw_flow *flow,
404 struct sw_flow_match *match)
406 struct sw_flow_key *key = match->key;
407 int key_start = flow_key_start(key);
408 int key_end = match->range.end;
410 return cmp_key(&flow->unmasked_key, key, key_start, key_end);
413 static struct sw_flow *masked_flow_lookup(struct table_instance *ti,
414 const struct sw_flow_key *unmasked,
415 struct sw_flow_mask *mask)
417 struct sw_flow *flow;
418 struct hlist_head *head;
419 int key_start = mask->range.start;
420 int key_end = mask->range.end;
422 struct sw_flow_key masked_key;
424 ovs_flow_mask_key(&masked_key, unmasked, mask);
425 hash = flow_hash(&masked_key, key_start, key_end);
426 head = find_bucket(ti, hash);
427 hlist_for_each_entry_rcu(flow, head, hash_node[ti->node_ver]) {
428 if (flow->mask == mask && flow->hash == hash &&
429 flow_cmp_masked_key(flow, &masked_key,
436 struct sw_flow *ovs_flow_tbl_lookup_stats(struct flow_table *tbl,
437 const struct sw_flow_key *key,
440 struct table_instance *ti = rcu_dereference_ovsl(tbl->ti);
441 struct sw_flow_mask *mask;
442 struct sw_flow *flow;
445 list_for_each_entry_rcu(mask, &tbl->mask_list, list) {
447 flow = masked_flow_lookup(ti, key, mask);
448 if (flow) /* Found */
454 struct sw_flow *ovs_flow_tbl_lookup(struct flow_table *tbl,
455 const struct sw_flow_key *key)
457 u32 __always_unused n_mask_hit;
459 return ovs_flow_tbl_lookup_stats(tbl, key, &n_mask_hit);
462 struct sw_flow *ovs_flow_tbl_lookup_exact(struct flow_table *tbl,
463 struct sw_flow_match *match)
465 struct table_instance *ti = rcu_dereference_ovsl(tbl->ti);
466 struct sw_flow_mask *mask;
467 struct sw_flow *flow;
469 /* Always called under ovs-mutex. */
470 list_for_each_entry(mask, &tbl->mask_list, list) {
471 flow = masked_flow_lookup(ti, match->key, mask);
472 if (flow && ovs_flow_cmp_unmasked_key(flow, match)) /* Found */
478 int ovs_flow_tbl_num_masks(const struct flow_table *table)
480 struct sw_flow_mask *mask;
483 list_for_each_entry(mask, &table->mask_list, list)
489 static struct table_instance *table_instance_expand(struct table_instance *ti)
491 return table_instance_rehash(ti, ti->n_buckets * 2);
494 /* Remove 'mask' from the mask list, if it is not needed any more. */
495 static void flow_mask_remove(struct flow_table *tbl, struct sw_flow_mask *mask)
498 /* ovs-lock is required to protect mask-refcount and
502 BUG_ON(!mask->ref_count);
505 if (!mask->ref_count) {
506 list_del_rcu(&mask->list);
507 kfree_rcu(mask, rcu);
512 /* Must be called with OVS mutex held. */
513 void ovs_flow_tbl_remove(struct flow_table *table, struct sw_flow *flow)
515 struct table_instance *ti = ovsl_dereference(table->ti);
517 BUG_ON(table->count == 0);
518 hlist_del_rcu(&flow->hash_node[ti->node_ver]);
521 /* RCU delete the mask. 'flow->mask' is not NULLed, as it should be
522 * accessible as long as the RCU read lock is held.
524 flow_mask_remove(table, flow->mask);
527 static struct sw_flow_mask *mask_alloc(void)
529 struct sw_flow_mask *mask;
531 mask = kmalloc(sizeof(*mask), GFP_KERNEL);
538 static bool mask_equal(const struct sw_flow_mask *a,
539 const struct sw_flow_mask *b)
541 const u8 *a_ = (const u8 *)&a->key + a->range.start;
542 const u8 *b_ = (const u8 *)&b->key + b->range.start;
544 return (a->range.end == b->range.end)
545 && (a->range.start == b->range.start)
546 && (memcmp(a_, b_, range_n_bytes(&a->range)) == 0);
549 static struct sw_flow_mask *flow_mask_find(const struct flow_table *tbl,
550 const struct sw_flow_mask *mask)
552 struct list_head *ml;
554 list_for_each(ml, &tbl->mask_list) {
555 struct sw_flow_mask *m;
556 m = container_of(ml, struct sw_flow_mask, list);
557 if (mask_equal(mask, m))
564 /* Add 'mask' into the mask list, if it is not already there. */
565 static int flow_mask_insert(struct flow_table *tbl, struct sw_flow *flow,
566 struct sw_flow_mask *new)
568 struct sw_flow_mask *mask;
569 mask = flow_mask_find(tbl, new);
571 /* Allocate a new mask if none exsits. */
575 mask->key = new->key;
576 mask->range = new->range;
577 list_add_rcu(&mask->list, &tbl->mask_list);
579 BUG_ON(!mask->ref_count);
587 /* Must be called with OVS mutex held. */
588 int ovs_flow_tbl_insert(struct flow_table *table, struct sw_flow *flow,
589 struct sw_flow_mask *mask)
591 struct table_instance *new_ti = NULL;
592 struct table_instance *ti;
595 err = flow_mask_insert(table, flow, mask);
599 flow->hash = flow_hash(&flow->key, flow->mask->range.start,
600 flow->mask->range.end);
601 ti = ovsl_dereference(table->ti);
602 table_instance_insert(ti, flow);
605 /* Expand table, if necessary, to make room. */
606 if (table->count > ti->n_buckets)
607 new_ti = table_instance_expand(ti);
608 else if (time_after(jiffies, table->last_rehash + REHASH_INTERVAL))
609 new_ti = table_instance_rehash(ti, ti->n_buckets);
612 rcu_assign_pointer(table->ti, new_ti);
613 table_instance_destroy(ti, true);
614 table->last_rehash = jiffies;
619 /* Initializes the flow module.
620 * Returns zero if successful or a negative error code. */
621 int ovs_flow_init(void)
623 BUILD_BUG_ON(__alignof__(struct sw_flow_key) % __alignof__(long));
624 BUILD_BUG_ON(sizeof(struct sw_flow_key) % sizeof(long));
626 flow_cache = kmem_cache_create("sw_flow", sizeof(struct sw_flow)
627 + (num_possible_nodes()
628 * sizeof(struct flow_stats *)),
630 if (flow_cache == NULL)
634 = kmem_cache_create("sw_flow_stats", sizeof(struct flow_stats),
635 0, SLAB_HWCACHE_ALIGN, NULL);
636 if (flow_stats_cache == NULL) {
637 kmem_cache_destroy(flow_cache);
645 /* Uninitializes the flow module. */
646 void ovs_flow_exit(void)
648 kmem_cache_destroy(flow_stats_cache);
649 kmem_cache_destroy(flow_cache);