1 #include <linux/kernel.h>
2 #include <linux/init.h>
3 #include <linux/module.h>
4 #include <linux/netfilter.h>
5 #include <linux/rhashtable.h>
6 #include <linux/netdevice.h>
8 #include <net/ip6_route.h>
9 #include <net/netfilter/nf_tables.h>
10 #include <net/netfilter/nf_flow_table.h>
11 #include <net/netfilter/nf_conntrack.h>
12 #include <net/netfilter/nf_conntrack_core.h>
13 #include <net/netfilter/nf_conntrack_tuple.h>
15 struct flow_offload_entry {
16 struct flow_offload flow;
18 struct rcu_head rcu_head;
21 static DEFINE_MUTEX(flowtable_lock);
22 static LIST_HEAD(flowtables);
25 flow_offload_fill_dir(struct flow_offload *flow, struct nf_conn *ct,
26 struct nf_flow_route *route,
27 enum flow_offload_tuple_dir dir)
29 struct flow_offload_tuple *ft = &flow->tuplehash[dir].tuple;
30 struct nf_conntrack_tuple *ctt = &ct->tuplehash[dir].tuple;
31 struct dst_entry *other_dst = route->tuple[!dir].dst;
32 struct dst_entry *dst = route->tuple[dir].dst;
36 switch (ctt->src.l3num) {
38 ft->src_v4 = ctt->src.u3.in;
39 ft->dst_v4 = ctt->dst.u3.in;
40 ft->mtu = ip_dst_mtu_maybe_forward(dst, true);
43 ft->src_v6 = ctt->src.u3.in6;
44 ft->dst_v6 = ctt->dst.u3.in6;
45 ft->mtu = ip6_dst_mtu_forward(dst);
49 ft->l3proto = ctt->src.l3num;
50 ft->l4proto = ctt->dst.protonum;
51 ft->src_port = ctt->src.u.tcp.port;
52 ft->dst_port = ctt->dst.u.tcp.port;
54 ft->iifidx = other_dst->dev->ifindex;
55 ft->oifidx = dst->dev->ifindex;
60 flow_offload_alloc(struct nf_conn *ct, struct nf_flow_route *route)
62 struct flow_offload_entry *entry;
63 struct flow_offload *flow;
65 if (unlikely(nf_ct_is_dying(ct) ||
66 !atomic_inc_not_zero(&ct->ct_general.use)))
69 entry = kzalloc(sizeof(*entry), GFP_ATOMIC);
75 if (!dst_hold_safe(route->tuple[FLOW_OFFLOAD_DIR_ORIGINAL].dst))
76 goto err_dst_cache_original;
78 if (!dst_hold_safe(route->tuple[FLOW_OFFLOAD_DIR_REPLY].dst))
79 goto err_dst_cache_reply;
83 flow_offload_fill_dir(flow, ct, route, FLOW_OFFLOAD_DIR_ORIGINAL);
84 flow_offload_fill_dir(flow, ct, route, FLOW_OFFLOAD_DIR_REPLY);
86 if (ct->status & IPS_SRC_NAT)
87 flow->flags |= FLOW_OFFLOAD_SNAT;
88 if (ct->status & IPS_DST_NAT)
89 flow->flags |= FLOW_OFFLOAD_DNAT;
94 dst_release(route->tuple[FLOW_OFFLOAD_DIR_ORIGINAL].dst);
95 err_dst_cache_original:
102 EXPORT_SYMBOL_GPL(flow_offload_alloc);
104 static void flow_offload_fixup_tcp(struct ip_ct_tcp *tcp)
106 tcp->state = TCP_CONNTRACK_ESTABLISHED;
107 tcp->seen[0].td_maxwin = 0;
108 tcp->seen[1].td_maxwin = 0;
111 #define NF_FLOWTABLE_TCP_PICKUP_TIMEOUT (120 * HZ)
112 #define NF_FLOWTABLE_UDP_PICKUP_TIMEOUT (30 * HZ)
114 static void flow_offload_fixup_ct_state(struct nf_conn *ct)
116 const struct nf_conntrack_l4proto *l4proto;
117 unsigned int timeout;
120 l4num = nf_ct_protonum(ct);
121 if (l4num == IPPROTO_TCP)
122 flow_offload_fixup_tcp(&ct->proto.tcp);
124 l4proto = __nf_ct_l4proto_find(nf_ct_l3num(ct), l4num);
128 if (l4num == IPPROTO_TCP)
129 timeout = NF_FLOWTABLE_TCP_PICKUP_TIMEOUT;
130 else if (l4num == IPPROTO_UDP)
131 timeout = NF_FLOWTABLE_UDP_PICKUP_TIMEOUT;
135 ct->timeout = nfct_time_stamp + timeout;
138 void flow_offload_free(struct flow_offload *flow)
140 struct flow_offload_entry *e;
142 dst_release(flow->tuplehash[FLOW_OFFLOAD_DIR_ORIGINAL].tuple.dst_cache);
143 dst_release(flow->tuplehash[FLOW_OFFLOAD_DIR_REPLY].tuple.dst_cache);
144 e = container_of(flow, struct flow_offload_entry, flow);
145 if (flow->flags & FLOW_OFFLOAD_DYING)
146 nf_ct_delete(e->ct, 0, 0);
148 kfree_rcu(e, rcu_head);
150 EXPORT_SYMBOL_GPL(flow_offload_free);
152 static u32 flow_offload_hash(const void *data, u32 len, u32 seed)
154 const struct flow_offload_tuple *tuple = data;
156 return jhash(tuple, offsetof(struct flow_offload_tuple, dir), seed);
159 static u32 flow_offload_hash_obj(const void *data, u32 len, u32 seed)
161 const struct flow_offload_tuple_rhash *tuplehash = data;
163 return jhash(&tuplehash->tuple, offsetof(struct flow_offload_tuple, dir), seed);
166 static int flow_offload_hash_cmp(struct rhashtable_compare_arg *arg,
169 const struct flow_offload_tuple *tuple = arg->key;
170 const struct flow_offload_tuple_rhash *x = ptr;
172 if (memcmp(&x->tuple, tuple, offsetof(struct flow_offload_tuple, dir)))
178 static const struct rhashtable_params nf_flow_offload_rhash_params = {
179 .head_offset = offsetof(struct flow_offload_tuple_rhash, node),
180 .hashfn = flow_offload_hash,
181 .obj_hashfn = flow_offload_hash_obj,
182 .obj_cmpfn = flow_offload_hash_cmp,
183 .automatic_shrinking = true,
186 int flow_offload_add(struct nf_flowtable *flow_table, struct flow_offload *flow)
190 flow->timeout = (u32)jiffies + NF_FLOW_TIMEOUT;
192 err = rhashtable_insert_fast(&flow_table->rhashtable,
193 &flow->tuplehash[0].node,
194 nf_flow_offload_rhash_params);
198 err = rhashtable_insert_fast(&flow_table->rhashtable,
199 &flow->tuplehash[1].node,
200 nf_flow_offload_rhash_params);
202 rhashtable_remove_fast(&flow_table->rhashtable,
203 &flow->tuplehash[0].node,
204 nf_flow_offload_rhash_params);
210 EXPORT_SYMBOL_GPL(flow_offload_add);
212 static void flow_offload_del(struct nf_flowtable *flow_table,
213 struct flow_offload *flow)
215 struct flow_offload_entry *e;
217 rhashtable_remove_fast(&flow_table->rhashtable,
218 &flow->tuplehash[FLOW_OFFLOAD_DIR_ORIGINAL].node,
219 nf_flow_offload_rhash_params);
220 rhashtable_remove_fast(&flow_table->rhashtable,
221 &flow->tuplehash[FLOW_OFFLOAD_DIR_REPLY].node,
222 nf_flow_offload_rhash_params);
224 e = container_of(flow, struct flow_offload_entry, flow);
225 clear_bit(IPS_OFFLOAD_BIT, &e->ct->status);
227 flow_offload_free(flow);
230 void flow_offload_teardown(struct flow_offload *flow)
232 struct flow_offload_entry *e;
234 flow->flags |= FLOW_OFFLOAD_TEARDOWN;
236 e = container_of(flow, struct flow_offload_entry, flow);
237 flow_offload_fixup_ct_state(e->ct);
239 EXPORT_SYMBOL_GPL(flow_offload_teardown);
241 struct flow_offload_tuple_rhash *
242 flow_offload_lookup(struct nf_flowtable *flow_table,
243 struct flow_offload_tuple *tuple)
245 struct flow_offload_tuple_rhash *tuplehash;
246 struct flow_offload *flow;
249 tuplehash = rhashtable_lookup_fast(&flow_table->rhashtable, tuple,
250 nf_flow_offload_rhash_params);
254 dir = tuplehash->tuple.dir;
255 flow = container_of(tuplehash, struct flow_offload, tuplehash[dir]);
256 if (flow->flags & (FLOW_OFFLOAD_DYING | FLOW_OFFLOAD_TEARDOWN))
261 EXPORT_SYMBOL_GPL(flow_offload_lookup);
263 int nf_flow_table_iterate(struct nf_flowtable *flow_table,
264 void (*iter)(struct flow_offload *flow, void *data),
267 struct flow_offload_tuple_rhash *tuplehash;
268 struct rhashtable_iter hti;
269 struct flow_offload *flow;
272 err = rhashtable_walk_init(&flow_table->rhashtable, &hti, GFP_KERNEL);
276 rhashtable_walk_start(&hti);
278 while ((tuplehash = rhashtable_walk_next(&hti))) {
279 if (IS_ERR(tuplehash)) {
280 err = PTR_ERR(tuplehash);
286 if (tuplehash->tuple.dir)
289 flow = container_of(tuplehash, struct flow_offload, tuplehash[0]);
294 rhashtable_walk_stop(&hti);
295 rhashtable_walk_exit(&hti);
299 EXPORT_SYMBOL_GPL(nf_flow_table_iterate);
301 static inline bool nf_flow_has_expired(const struct flow_offload *flow)
303 return (__s32)(flow->timeout - (u32)jiffies) <= 0;
306 static int nf_flow_offload_gc_step(struct nf_flowtable *flow_table)
308 struct flow_offload_tuple_rhash *tuplehash;
309 struct rhashtable_iter hti;
310 struct flow_offload *flow;
313 err = rhashtable_walk_init(&flow_table->rhashtable, &hti, GFP_KERNEL);
317 rhashtable_walk_start(&hti);
319 while ((tuplehash = rhashtable_walk_next(&hti))) {
320 if (IS_ERR(tuplehash)) {
321 err = PTR_ERR(tuplehash);
327 if (tuplehash->tuple.dir)
330 flow = container_of(tuplehash, struct flow_offload, tuplehash[0]);
332 if (nf_flow_has_expired(flow) ||
333 (flow->flags & (FLOW_OFFLOAD_DYING |
334 FLOW_OFFLOAD_TEARDOWN)))
335 flow_offload_del(flow_table, flow);
338 rhashtable_walk_stop(&hti);
339 rhashtable_walk_exit(&hti);
344 static void nf_flow_offload_work_gc(struct work_struct *work)
346 struct nf_flowtable *flow_table;
348 flow_table = container_of(work, struct nf_flowtable, gc_work.work);
349 nf_flow_offload_gc_step(flow_table);
350 queue_delayed_work(system_power_efficient_wq, &flow_table->gc_work, HZ);
353 static int nf_flow_nat_port_tcp(struct sk_buff *skb, unsigned int thoff,
354 __be16 port, __be16 new_port)
358 if (!pskb_may_pull(skb, thoff + sizeof(*tcph)) ||
359 skb_try_make_writable(skb, thoff + sizeof(*tcph)))
362 tcph = (void *)(skb_network_header(skb) + thoff);
363 inet_proto_csum_replace2(&tcph->check, skb, port, new_port, false);
368 static int nf_flow_nat_port_udp(struct sk_buff *skb, unsigned int thoff,
369 __be16 port, __be16 new_port)
373 if (!pskb_may_pull(skb, thoff + sizeof(*udph)) ||
374 skb_try_make_writable(skb, thoff + sizeof(*udph)))
377 udph = (void *)(skb_network_header(skb) + thoff);
378 if (udph->check || skb->ip_summed == CHECKSUM_PARTIAL) {
379 inet_proto_csum_replace2(&udph->check, skb, port,
382 udph->check = CSUM_MANGLED_0;
388 static int nf_flow_nat_port(struct sk_buff *skb, unsigned int thoff,
389 u8 protocol, __be16 port, __be16 new_port)
393 if (nf_flow_nat_port_tcp(skb, thoff, port, new_port) < 0)
397 if (nf_flow_nat_port_udp(skb, thoff, port, new_port) < 0)
405 int nf_flow_snat_port(const struct flow_offload *flow,
406 struct sk_buff *skb, unsigned int thoff,
407 u8 protocol, enum flow_offload_tuple_dir dir)
409 struct flow_ports *hdr;
410 __be16 port, new_port;
412 if (!pskb_may_pull(skb, thoff + sizeof(*hdr)) ||
413 skb_try_make_writable(skb, thoff + sizeof(*hdr)))
416 hdr = (void *)(skb_network_header(skb) + thoff);
419 case FLOW_OFFLOAD_DIR_ORIGINAL:
421 new_port = flow->tuplehash[FLOW_OFFLOAD_DIR_REPLY].tuple.dst_port;
422 hdr->source = new_port;
424 case FLOW_OFFLOAD_DIR_REPLY:
426 new_port = flow->tuplehash[FLOW_OFFLOAD_DIR_ORIGINAL].tuple.src_port;
427 hdr->dest = new_port;
433 return nf_flow_nat_port(skb, thoff, protocol, port, new_port);
435 EXPORT_SYMBOL_GPL(nf_flow_snat_port);
437 int nf_flow_dnat_port(const struct flow_offload *flow,
438 struct sk_buff *skb, unsigned int thoff,
439 u8 protocol, enum flow_offload_tuple_dir dir)
441 struct flow_ports *hdr;
442 __be16 port, new_port;
444 if (!pskb_may_pull(skb, thoff + sizeof(*hdr)) ||
445 skb_try_make_writable(skb, thoff + sizeof(*hdr)))
448 hdr = (void *)(skb_network_header(skb) + thoff);
451 case FLOW_OFFLOAD_DIR_ORIGINAL:
453 new_port = flow->tuplehash[FLOW_OFFLOAD_DIR_REPLY].tuple.src_port;
454 hdr->dest = new_port;
456 case FLOW_OFFLOAD_DIR_REPLY:
458 new_port = flow->tuplehash[FLOW_OFFLOAD_DIR_ORIGINAL].tuple.dst_port;
459 hdr->source = new_port;
465 return nf_flow_nat_port(skb, thoff, protocol, port, new_port);
467 EXPORT_SYMBOL_GPL(nf_flow_dnat_port);
469 int nf_flow_table_init(struct nf_flowtable *flowtable)
473 INIT_DEFERRABLE_WORK(&flowtable->gc_work, nf_flow_offload_work_gc);
475 err = rhashtable_init(&flowtable->rhashtable,
476 &nf_flow_offload_rhash_params);
480 queue_delayed_work(system_power_efficient_wq,
481 &flowtable->gc_work, HZ);
483 mutex_lock(&flowtable_lock);
484 list_add(&flowtable->list, &flowtables);
485 mutex_unlock(&flowtable_lock);
489 EXPORT_SYMBOL_GPL(nf_flow_table_init);
491 static void nf_flow_table_do_cleanup(struct flow_offload *flow, void *data)
493 struct net_device *dev = data;
494 struct flow_offload_entry *e;
496 e = container_of(flow, struct flow_offload_entry, flow);
499 flow_offload_teardown(flow);
502 if (net_eq(nf_ct_net(e->ct), dev_net(dev)) &&
503 (flow->tuplehash[0].tuple.iifidx == dev->ifindex ||
504 flow->tuplehash[1].tuple.iifidx == dev->ifindex))
505 flow_offload_dead(flow);
508 static void nf_flow_table_iterate_cleanup(struct nf_flowtable *flowtable,
509 struct net_device *dev)
511 nf_flow_table_iterate(flowtable, nf_flow_table_do_cleanup, dev);
512 flush_delayed_work(&flowtable->gc_work);
515 void nf_flow_table_cleanup(struct net *net, struct net_device *dev)
517 struct nf_flowtable *flowtable;
519 mutex_lock(&flowtable_lock);
520 list_for_each_entry(flowtable, &flowtables, list)
521 nf_flow_table_iterate_cleanup(flowtable, dev);
522 mutex_unlock(&flowtable_lock);
524 EXPORT_SYMBOL_GPL(nf_flow_table_cleanup);
526 void nf_flow_table_free(struct nf_flowtable *flow_table)
528 mutex_lock(&flowtable_lock);
529 list_del(&flow_table->list);
530 mutex_unlock(&flowtable_lock);
531 cancel_delayed_work_sync(&flow_table->gc_work);
532 nf_flow_table_iterate(flow_table, nf_flow_table_do_cleanup, NULL);
533 WARN_ON(!nf_flow_offload_gc_step(flow_table));
534 rhashtable_destroy(&flow_table->rhashtable);
536 EXPORT_SYMBOL_GPL(nf_flow_table_free);
538 MODULE_LICENSE("GPL");
539 MODULE_AUTHOR("Pablo Neira Ayuso <pablo@netfilter.org>");