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/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;
52 static u16 range_n_bytes(const struct sw_flow_key_range *range)
54 return range->end - range->start;
57 void ovs_flow_mask_key(struct sw_flow_key *dst, const struct sw_flow_key *src,
58 const struct sw_flow_mask *mask)
60 const long *m = (const long *)((const u8 *)&mask->key +
62 const long *s = (const long *)((const u8 *)src +
64 long *d = (long *)((u8 *)dst + mask->range.start);
67 /* The memory outside of the 'mask->range' are not set since
68 * further operations on 'dst' only uses contents within
71 for (i = 0; i < range_n_bytes(&mask->range); i += sizeof(long))
75 struct sw_flow *ovs_flow_alloc(bool percpu_stats)
80 flow = kmem_cache_alloc(flow_cache, GFP_KERNEL);
82 return ERR_PTR(-ENOMEM);
87 flow->stats.is_percpu = percpu_stats;
90 flow->stats.stat = kzalloc(sizeof(*flow->stats.stat), GFP_KERNEL);
91 if (!flow->stats.stat)
94 spin_lock_init(&flow->stats.stat->lock);
96 flow->stats.cpu_stats = alloc_percpu(struct flow_stats);
97 if (!flow->stats.cpu_stats)
100 for_each_possible_cpu(cpu) {
101 struct flow_stats *cpu_stats;
103 cpu_stats = per_cpu_ptr(flow->stats.cpu_stats, cpu);
104 spin_lock_init(&cpu_stats->lock);
109 kmem_cache_free(flow_cache, flow);
110 return ERR_PTR(-ENOMEM);
113 int ovs_flow_tbl_count(struct flow_table *table)
118 static struct flex_array *alloc_buckets(unsigned int n_buckets)
120 struct flex_array *buckets;
123 buckets = flex_array_alloc(sizeof(struct hlist_head),
124 n_buckets, GFP_KERNEL);
128 err = flex_array_prealloc(buckets, 0, n_buckets, GFP_KERNEL);
130 flex_array_free(buckets);
134 for (i = 0; i < n_buckets; i++)
135 INIT_HLIST_HEAD((struct hlist_head *)
136 flex_array_get(buckets, i));
141 static void flow_free(struct sw_flow *flow)
143 kfree((struct sf_flow_acts __force *)flow->sf_acts);
144 if (flow->stats.is_percpu)
145 free_percpu(flow->stats.cpu_stats);
147 kfree(flow->stats.stat);
148 kmem_cache_free(flow_cache, flow);
151 static void rcu_free_flow_callback(struct rcu_head *rcu)
153 struct sw_flow *flow = container_of(rcu, struct sw_flow, rcu);
158 void ovs_flow_free(struct sw_flow *flow, bool deferred)
164 struct sw_flow_mask *mask = flow->mask;
166 /* ovs-lock is required to protect mask-refcount and
170 BUG_ON(!mask->ref_count);
173 if (!mask->ref_count) {
174 list_del_rcu(&mask->list);
176 kfree_rcu(mask, rcu);
183 call_rcu(&flow->rcu, rcu_free_flow_callback);
188 static void free_buckets(struct flex_array *buckets)
190 flex_array_free(buckets);
194 static void __table_instance_destroy(struct table_instance *ti)
196 free_buckets(ti->buckets);
200 static struct table_instance *table_instance_alloc(int new_size)
202 struct table_instance *ti = kmalloc(sizeof(*ti), GFP_KERNEL);
207 ti->buckets = alloc_buckets(new_size);
213 ti->n_buckets = new_size;
215 ti->keep_flows = false;
216 get_random_bytes(&ti->hash_seed, sizeof(u32));
221 int ovs_flow_tbl_init(struct flow_table *table)
223 struct table_instance *ti;
225 ti = table_instance_alloc(TBL_MIN_BUCKETS);
230 rcu_assign_pointer(table->ti, ti);
231 INIT_LIST_HEAD(&table->mask_list);
232 table->last_rehash = jiffies;
237 static void flow_tbl_destroy_rcu_cb(struct rcu_head *rcu)
239 struct table_instance *ti = container_of(rcu, struct table_instance, rcu);
241 __table_instance_destroy(ti);
244 static void table_instance_destroy(struct table_instance *ti, bool deferred)
254 for (i = 0; i < ti->n_buckets; i++) {
255 struct sw_flow *flow;
256 struct hlist_head *head = flex_array_get(ti->buckets, i);
257 struct hlist_node *n;
258 int ver = ti->node_ver;
260 hlist_for_each_entry_safe(flow, n, head, hash_node[ver]) {
261 hlist_del_rcu(&flow->hash_node[ver]);
262 ovs_flow_free(flow, deferred);
268 call_rcu(&ti->rcu, flow_tbl_destroy_rcu_cb);
270 __table_instance_destroy(ti);
273 void ovs_flow_tbl_destroy(struct flow_table *table, bool deferred)
275 struct table_instance *ti = ovsl_dereference(table->ti);
277 table_instance_destroy(ti, deferred);
280 struct sw_flow *ovs_flow_tbl_dump_next(struct table_instance *ti,
281 u32 *bucket, u32 *last)
283 struct sw_flow *flow;
284 struct hlist_head *head;
289 while (*bucket < ti->n_buckets) {
291 head = flex_array_get(ti->buckets, *bucket);
292 hlist_for_each_entry_rcu(flow, head, hash_node[ver]) {
307 static struct hlist_head *find_bucket(struct table_instance *ti, u32 hash)
309 hash = jhash_1word(hash, ti->hash_seed);
310 return flex_array_get(ti->buckets,
311 (hash & (ti->n_buckets - 1)));
314 static void table_instance_insert(struct table_instance *ti, struct sw_flow *flow)
316 struct hlist_head *head;
318 head = find_bucket(ti, flow->hash);
319 hlist_add_head_rcu(&flow->hash_node[ti->node_ver], head);
322 static void flow_table_copy_flows(struct table_instance *old,
323 struct table_instance *new)
328 old_ver = old->node_ver;
329 new->node_ver = !old_ver;
331 /* Insert in new table. */
332 for (i = 0; i < old->n_buckets; i++) {
333 struct sw_flow *flow;
334 struct hlist_head *head;
336 head = flex_array_get(old->buckets, i);
338 hlist_for_each_entry(flow, head, hash_node[old_ver])
339 table_instance_insert(new, flow);
342 old->keep_flows = true;
345 static struct table_instance *table_instance_rehash(struct table_instance *ti,
348 struct table_instance *new_ti;
350 new_ti = table_instance_alloc(n_buckets);
354 flow_table_copy_flows(ti, new_ti);
359 int ovs_flow_tbl_flush(struct flow_table *flow_table)
361 struct table_instance *old_ti;
362 struct table_instance *new_ti;
364 old_ti = ovsl_dereference(flow_table->ti);
365 new_ti = table_instance_alloc(TBL_MIN_BUCKETS);
369 rcu_assign_pointer(flow_table->ti, new_ti);
370 flow_table->last_rehash = jiffies;
371 flow_table->count = 0;
373 table_instance_destroy(old_ti, true);
377 static u32 flow_hash(const struct sw_flow_key *key, int key_start,
380 const u32 *hash_key = (const u32 *)((const u8 *)key + key_start);
381 int hash_u32s = (key_end - key_start) >> 2;
383 /* Make sure number of hash bytes are multiple of u32. */
384 BUILD_BUG_ON(sizeof(long) % sizeof(u32));
386 return arch_fast_hash2(hash_key, hash_u32s, 0);
389 static int flow_key_start(const struct sw_flow_key *key)
391 if (key->tun_key.ipv4_dst)
394 return rounddown(offsetof(struct sw_flow_key, phy),
398 static bool cmp_key(const struct sw_flow_key *key1,
399 const struct sw_flow_key *key2,
400 int key_start, int key_end)
402 const long *cp1 = (const long *)((const u8 *)key1 + key_start);
403 const long *cp2 = (const long *)((const u8 *)key2 + key_start);
407 for (i = key_start; i < key_end; i += sizeof(long))
408 diffs |= *cp1++ ^ *cp2++;
413 static bool flow_cmp_masked_key(const struct sw_flow *flow,
414 const struct sw_flow_key *key,
415 int key_start, int key_end)
417 return cmp_key(&flow->key, key, key_start, key_end);
420 bool ovs_flow_cmp_unmasked_key(const struct sw_flow *flow,
421 struct sw_flow_match *match)
423 struct sw_flow_key *key = match->key;
424 int key_start = flow_key_start(key);
425 int key_end = match->range.end;
427 return cmp_key(&flow->unmasked_key, key, key_start, key_end);
430 static struct sw_flow *masked_flow_lookup(struct table_instance *ti,
431 const struct sw_flow_key *unmasked,
432 struct sw_flow_mask *mask)
434 struct sw_flow *flow;
435 struct hlist_head *head;
436 int key_start = mask->range.start;
437 int key_end = mask->range.end;
439 struct sw_flow_key masked_key;
441 ovs_flow_mask_key(&masked_key, unmasked, mask);
442 hash = flow_hash(&masked_key, key_start, key_end);
443 head = find_bucket(ti, hash);
444 hlist_for_each_entry_rcu(flow, head, hash_node[ti->node_ver]) {
445 if (flow->mask == mask && flow->hash == hash &&
446 flow_cmp_masked_key(flow, &masked_key,
453 struct sw_flow *ovs_flow_tbl_lookup_stats(struct flow_table *tbl,
454 const struct sw_flow_key *key,
457 struct table_instance *ti = rcu_dereference_ovsl(tbl->ti);
458 struct sw_flow_mask *mask;
459 struct sw_flow *flow;
462 list_for_each_entry_rcu(mask, &tbl->mask_list, list) {
464 flow = masked_flow_lookup(ti, key, mask);
465 if (flow) /* Found */
471 struct sw_flow *ovs_flow_tbl_lookup(struct flow_table *tbl,
472 const struct sw_flow_key *key)
474 u32 __always_unused n_mask_hit;
476 return ovs_flow_tbl_lookup_stats(tbl, key, &n_mask_hit);
479 int ovs_flow_tbl_num_masks(const struct flow_table *table)
481 struct sw_flow_mask *mask;
484 list_for_each_entry(mask, &table->mask_list, list)
490 static struct table_instance *table_instance_expand(struct table_instance *ti)
492 return table_instance_rehash(ti, ti->n_buckets * 2);
495 void ovs_flow_tbl_remove(struct flow_table *table, struct sw_flow *flow)
497 struct table_instance *ti = ovsl_dereference(table->ti);
499 BUG_ON(table->count == 0);
500 hlist_del_rcu(&flow->hash_node[ti->node_ver]);
504 static struct sw_flow_mask *mask_alloc(void)
506 struct sw_flow_mask *mask;
508 mask = kmalloc(sizeof(*mask), GFP_KERNEL);
515 static bool mask_equal(const struct sw_flow_mask *a,
516 const struct sw_flow_mask *b)
518 const u8 *a_ = (const u8 *)&a->key + a->range.start;
519 const u8 *b_ = (const u8 *)&b->key + b->range.start;
521 return (a->range.end == b->range.end)
522 && (a->range.start == b->range.start)
523 && (memcmp(a_, b_, range_n_bytes(&a->range)) == 0);
526 static struct sw_flow_mask *flow_mask_find(const struct flow_table *tbl,
527 const struct sw_flow_mask *mask)
529 struct list_head *ml;
531 list_for_each(ml, &tbl->mask_list) {
532 struct sw_flow_mask *m;
533 m = container_of(ml, struct sw_flow_mask, list);
534 if (mask_equal(mask, m))
541 /* Add 'mask' into the mask list, if it is not already there. */
542 static int flow_mask_insert(struct flow_table *tbl, struct sw_flow *flow,
543 struct sw_flow_mask *new)
545 struct sw_flow_mask *mask;
546 mask = flow_mask_find(tbl, new);
548 /* Allocate a new mask if none exsits. */
552 mask->key = new->key;
553 mask->range = new->range;
554 list_add_rcu(&mask->list, &tbl->mask_list);
556 BUG_ON(!mask->ref_count);
564 int ovs_flow_tbl_insert(struct flow_table *table, struct sw_flow *flow,
565 struct sw_flow_mask *mask)
567 struct table_instance *new_ti = NULL;
568 struct table_instance *ti;
571 err = flow_mask_insert(table, flow, mask);
575 flow->hash = flow_hash(&flow->key, flow->mask->range.start,
576 flow->mask->range.end);
577 ti = ovsl_dereference(table->ti);
578 table_instance_insert(ti, flow);
581 /* Expand table, if necessary, to make room. */
582 if (table->count > ti->n_buckets)
583 new_ti = table_instance_expand(ti);
584 else if (time_after(jiffies, table->last_rehash + REHASH_INTERVAL))
585 new_ti = table_instance_rehash(ti, ti->n_buckets);
588 rcu_assign_pointer(table->ti, new_ti);
589 table_instance_destroy(ti, true);
590 table->last_rehash = jiffies;
595 /* Initializes the flow module.
596 * Returns zero if successful or a negative error code. */
597 int ovs_flow_init(void)
599 BUILD_BUG_ON(__alignof__(struct sw_flow_key) % __alignof__(long));
600 BUILD_BUG_ON(sizeof(struct sw_flow_key) % sizeof(long));
602 flow_cache = kmem_cache_create("sw_flow", sizeof(struct sw_flow), 0,
604 if (flow_cache == NULL)
610 /* Uninitializes the flow module. */
611 void ovs_flow_exit(void)
613 kmem_cache_destroy(flow_cache);