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>
49 #define TBL_MIN_BUCKETS 1024
50 #define REHASH_INTERVAL (10 * 60 * HZ)
52 static struct kmem_cache *flow_cache;
54 static u16 range_n_bytes(const struct sw_flow_key_range *range)
56 return range->end - range->start;
59 void ovs_flow_mask_key(struct sw_flow_key *dst, const struct sw_flow_key *src,
60 const struct sw_flow_mask *mask)
62 const long *m = (long *)((u8 *)&mask->key + mask->range.start);
63 const long *s = (long *)((u8 *)src + mask->range.start);
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(void)
79 flow = kmem_cache_alloc(flow_cache, GFP_KERNEL);
81 return ERR_PTR(-ENOMEM);
83 spin_lock_init(&flow->lock);
90 int ovs_flow_tbl_count(struct flow_table *table)
95 static struct flex_array *alloc_buckets(unsigned int n_buckets)
97 struct flex_array *buckets;
100 buckets = flex_array_alloc(sizeof(struct hlist_head),
101 n_buckets, GFP_KERNEL);
105 err = flex_array_prealloc(buckets, 0, n_buckets, GFP_KERNEL);
107 flex_array_free(buckets);
111 for (i = 0; i < n_buckets; i++)
112 INIT_HLIST_HEAD((struct hlist_head *)
113 flex_array_get(buckets, i));
118 static void flow_free(struct sw_flow *flow)
120 kfree((struct sf_flow_acts __force *)flow->sf_acts);
121 kmem_cache_free(flow_cache, flow);
124 static void rcu_free_flow_callback(struct rcu_head *rcu)
126 struct sw_flow *flow = container_of(rcu, struct sw_flow, rcu);
131 static void rcu_free_sw_flow_mask_cb(struct rcu_head *rcu)
133 struct sw_flow_mask *mask = container_of(rcu, struct sw_flow_mask, rcu);
138 static void flow_mask_del_ref(struct sw_flow_mask *mask, bool deferred)
143 BUG_ON(!mask->ref_count);
146 if (!mask->ref_count) {
147 list_del_rcu(&mask->list);
149 call_rcu(&mask->rcu, rcu_free_sw_flow_mask_cb);
155 void ovs_flow_free(struct sw_flow *flow, bool deferred)
160 flow_mask_del_ref(flow->mask, deferred);
163 call_rcu(&flow->rcu, rcu_free_flow_callback);
168 static void free_buckets(struct flex_array *buckets)
170 flex_array_free(buckets);
173 static void __table_instance_destroy(struct table_instance *ti)
180 for (i = 0; i < ti->n_buckets; i++) {
181 struct sw_flow *flow;
182 struct hlist_head *head = flex_array_get(ti->buckets, i);
183 struct hlist_node *n;
184 int ver = ti->node_ver;
186 hlist_for_each_entry_safe(flow, n, head, hash_node[ver]) {
187 hlist_del(&flow->hash_node[ver]);
188 ovs_flow_free(flow, false);
193 free_buckets(ti->buckets);
197 static struct table_instance *table_instance_alloc(int new_size)
199 struct table_instance *ti = kmalloc(sizeof(*ti), GFP_KERNEL);
204 ti->buckets = alloc_buckets(new_size);
210 ti->n_buckets = new_size;
212 ti->keep_flows = false;
213 get_random_bytes(&ti->hash_seed, sizeof(u32));
218 int ovs_flow_tbl_init(struct flow_table *table)
220 struct table_instance *ti;
222 ti = table_instance_alloc(TBL_MIN_BUCKETS);
227 rcu_assign_pointer(table->ti, ti);
228 INIT_LIST_HEAD(&table->mask_list);
229 table->last_rehash = jiffies;
234 static void flow_tbl_destroy_rcu_cb(struct rcu_head *rcu)
236 struct table_instance *ti = container_of(rcu, struct table_instance, rcu);
238 __table_instance_destroy(ti);
241 static void table_instance_destroy(struct table_instance *ti, bool deferred)
247 call_rcu(&ti->rcu, flow_tbl_destroy_rcu_cb);
249 __table_instance_destroy(ti);
252 void ovs_flow_tbl_destroy(struct flow_table *table)
254 struct table_instance *ti = ovsl_dereference(table->ti);
256 table_instance_destroy(ti, false);
259 struct sw_flow *ovs_flow_tbl_dump_next(struct table_instance *ti,
260 u32 *bucket, u32 *last)
262 struct sw_flow *flow;
263 struct hlist_head *head;
268 while (*bucket < ti->n_buckets) {
270 head = flex_array_get(ti->buckets, *bucket);
271 hlist_for_each_entry_rcu(flow, head, hash_node[ver]) {
286 static struct hlist_head *find_bucket(struct table_instance *ti, u32 hash)
288 hash = jhash_1word(hash, ti->hash_seed);
289 return flex_array_get(ti->buckets,
290 (hash & (ti->n_buckets - 1)));
293 static void table_instance_insert(struct table_instance *ti, struct sw_flow *flow)
295 struct hlist_head *head;
297 head = find_bucket(ti, flow->hash);
298 hlist_add_head_rcu(&flow->hash_node[ti->node_ver], head);
301 static void flow_table_copy_flows(struct table_instance *old,
302 struct table_instance *new)
307 old_ver = old->node_ver;
308 new->node_ver = !old_ver;
310 /* Insert in new table. */
311 for (i = 0; i < old->n_buckets; i++) {
312 struct sw_flow *flow;
313 struct hlist_head *head;
315 head = flex_array_get(old->buckets, i);
317 hlist_for_each_entry(flow, head, hash_node[old_ver])
318 table_instance_insert(new, flow);
321 old->keep_flows = true;
324 static struct table_instance *table_instance_rehash(struct table_instance *ti,
327 struct table_instance *new_ti;
329 new_ti = table_instance_alloc(n_buckets);
333 flow_table_copy_flows(ti, new_ti);
338 int ovs_flow_tbl_flush(struct flow_table *flow_table)
340 struct table_instance *old_ti;
341 struct table_instance *new_ti;
343 old_ti = ovsl_dereference(flow_table->ti);
344 new_ti = table_instance_alloc(TBL_MIN_BUCKETS);
348 rcu_assign_pointer(flow_table->ti, new_ti);
349 flow_table->last_rehash = jiffies;
350 flow_table->count = 0;
352 table_instance_destroy(old_ti, true);
356 static u32 flow_hash(const struct sw_flow_key *key, int key_start,
359 u32 *hash_key = (u32 *)((u8 *)key + key_start);
360 int hash_u32s = (key_end - key_start) >> 2;
362 /* Make sure number of hash bytes are multiple of u32. */
363 BUILD_BUG_ON(sizeof(long) % sizeof(u32));
365 return jhash2(hash_key, hash_u32s, 0);
368 static int flow_key_start(const struct sw_flow_key *key)
370 if (key->tun_key.ipv4_dst)
373 return rounddown(offsetof(struct sw_flow_key, phy),
377 static bool cmp_key(const struct sw_flow_key *key1,
378 const struct sw_flow_key *key2,
379 int key_start, int key_end)
381 const long *cp1 = (long *)((u8 *)key1 + key_start);
382 const long *cp2 = (long *)((u8 *)key2 + key_start);
386 for (i = key_start; i < key_end; i += sizeof(long))
387 diffs |= *cp1++ ^ *cp2++;
392 static bool flow_cmp_masked_key(const struct sw_flow *flow,
393 const struct sw_flow_key *key,
394 int key_start, int key_end)
396 return cmp_key(&flow->key, key, key_start, key_end);
399 bool ovs_flow_cmp_unmasked_key(const struct sw_flow *flow,
400 struct sw_flow_match *match)
402 struct sw_flow_key *key = match->key;
403 int key_start = flow_key_start(key);
404 int key_end = match->range.end;
406 return cmp_key(&flow->unmasked_key, key, key_start, key_end);
409 static struct sw_flow *masked_flow_lookup(struct table_instance *ti,
410 const struct sw_flow_key *unmasked,
411 struct sw_flow_mask *mask)
413 struct sw_flow *flow;
414 struct hlist_head *head;
415 int key_start = mask->range.start;
416 int key_end = mask->range.end;
418 struct sw_flow_key masked_key;
420 ovs_flow_mask_key(&masked_key, unmasked, mask);
421 hash = flow_hash(&masked_key, key_start, key_end);
422 head = find_bucket(ti, hash);
423 hlist_for_each_entry_rcu(flow, head, hash_node[ti->node_ver]) {
424 if (flow->mask == mask && flow->hash == hash &&
425 flow_cmp_masked_key(flow, &masked_key,
432 struct sw_flow *ovs_flow_tbl_lookup(struct flow_table *tbl,
433 const struct sw_flow_key *key,
436 struct table_instance *ti = rcu_dereference(tbl->ti);
437 struct sw_flow_mask *mask;
438 struct sw_flow *flow;
441 list_for_each_entry_rcu(mask, &tbl->mask_list, list) {
443 flow = masked_flow_lookup(ti, key, mask);
444 if (flow) /* Found */
450 int ovs_flow_tbl_num_masks(const struct flow_table *table)
452 struct sw_flow_mask *mask;
455 list_for_each_entry(mask, &table->mask_list, list)
461 static struct table_instance *table_instance_expand(struct table_instance *ti)
463 return table_instance_rehash(ti, ti->n_buckets * 2);
466 void ovs_flow_tbl_remove(struct flow_table *table, struct sw_flow *flow)
468 struct table_instance *ti = ovsl_dereference(table->ti);
470 BUG_ON(table->count == 0);
471 hlist_del_rcu(&flow->hash_node[ti->node_ver]);
475 static struct sw_flow_mask *mask_alloc(void)
477 struct sw_flow_mask *mask;
479 mask = kmalloc(sizeof(*mask), GFP_KERNEL);
486 static void mask_add_ref(struct sw_flow_mask *mask)
491 static bool mask_equal(const struct sw_flow_mask *a,
492 const struct sw_flow_mask *b)
494 u8 *a_ = (u8 *)&a->key + a->range.start;
495 u8 *b_ = (u8 *)&b->key + b->range.start;
497 return (a->range.end == b->range.end)
498 && (a->range.start == b->range.start)
499 && (memcmp(a_, b_, range_n_bytes(&a->range)) == 0);
502 static struct sw_flow_mask *flow_mask_find(const struct flow_table *tbl,
503 const struct sw_flow_mask *mask)
505 struct list_head *ml;
507 list_for_each(ml, &tbl->mask_list) {
508 struct sw_flow_mask *m;
509 m = container_of(ml, struct sw_flow_mask, list);
510 if (mask_equal(mask, m))
518 * add a new mask into the mask list.
519 * The caller needs to make sure that 'mask' is not the same
520 * as any masks that are already on the list.
522 static int flow_mask_insert(struct flow_table *tbl, struct sw_flow *flow,
523 struct sw_flow_mask *new)
525 struct sw_flow_mask *mask;
526 mask = flow_mask_find(tbl, new);
528 /* Allocate a new mask if none exsits. */
532 mask->key = new->key;
533 mask->range = new->range;
534 list_add_rcu(&mask->list, &tbl->mask_list);
542 int ovs_flow_tbl_insert(struct flow_table *table, struct sw_flow *flow,
543 struct sw_flow_mask *mask)
545 struct table_instance *new_ti = NULL;
546 struct table_instance *ti;
549 err = flow_mask_insert(table, flow, mask);
553 flow->hash = flow_hash(&flow->key, flow->mask->range.start,
554 flow->mask->range.end);
555 ti = ovsl_dereference(table->ti);
556 table_instance_insert(ti, flow);
559 /* Expand table, if necessary, to make room. */
560 if (table->count > ti->n_buckets)
561 new_ti = table_instance_expand(ti);
562 else if (time_after(jiffies, table->last_rehash + REHASH_INTERVAL))
563 new_ti = table_instance_rehash(ti, ti->n_buckets);
566 rcu_assign_pointer(table->ti, new_ti);
567 table_instance_destroy(ti, true);
568 table->last_rehash = jiffies;
573 /* Initializes the flow module.
574 * Returns zero if successful or a negative error code. */
575 int ovs_flow_init(void)
577 BUILD_BUG_ON(__alignof__(struct sw_flow_key) % __alignof__(long));
578 BUILD_BUG_ON(sizeof(struct sw_flow_key) % sizeof(long));
580 flow_cache = kmem_cache_create("sw_flow", sizeof(struct sw_flow), 0,
582 if (flow_cache == NULL)
588 /* Uninitializes the flow module. */
589 void ovs_flow_exit(void)
591 kmem_cache_destroy(flow_cache);