1 // SPDX-License-Identifier: GPL-2.0-only
2 #include <linux/kernel.h>
3 #include <linux/init.h>
4 #include <linux/module.h>
5 #include <linux/netfilter.h>
6 #include <linux/rhashtable.h>
7 #include <linux/netdevice.h>
9 #include <net/ip6_route.h>
10 #include <net/netfilter/nf_tables.h>
11 #include <net/netfilter/nf_flow_table.h>
12 #include <net/netfilter/nf_conntrack.h>
13 #include <net/netfilter/nf_conntrack_core.h>
14 #include <net/netfilter/nf_conntrack_l4proto.h>
15 #include <net/netfilter/nf_conntrack_tuple.h>
17 static DEFINE_MUTEX(flowtable_lock);
18 static LIST_HEAD(flowtables);
21 flow_offload_fill_dir(struct flow_offload *flow,
22 enum flow_offload_tuple_dir dir)
24 struct flow_offload_tuple *ft = &flow->tuplehash[dir].tuple;
25 struct nf_conntrack_tuple *ctt = &flow->ct->tuplehash[dir].tuple;
29 switch (ctt->src.l3num) {
31 ft->src_v4 = ctt->src.u3.in;
32 ft->dst_v4 = ctt->dst.u3.in;
35 ft->src_v6 = ctt->src.u3.in6;
36 ft->dst_v6 = ctt->dst.u3.in6;
40 ft->l3proto = ctt->src.l3num;
41 ft->l4proto = ctt->dst.protonum;
43 switch (ctt->dst.protonum) {
46 ft->src_port = ctt->src.u.tcp.port;
47 ft->dst_port = ctt->dst.u.tcp.port;
52 struct flow_offload *flow_offload_alloc(struct nf_conn *ct)
54 struct flow_offload *flow;
56 if (unlikely(nf_ct_is_dying(ct)))
59 flow = kzalloc(sizeof(*flow), GFP_ATOMIC);
63 refcount_inc(&ct->ct_general.use);
66 flow_offload_fill_dir(flow, FLOW_OFFLOAD_DIR_ORIGINAL);
67 flow_offload_fill_dir(flow, FLOW_OFFLOAD_DIR_REPLY);
69 if (ct->status & IPS_SRC_NAT)
70 __set_bit(NF_FLOW_SNAT, &flow->flags);
71 if (ct->status & IPS_DST_NAT)
72 __set_bit(NF_FLOW_DNAT, &flow->flags);
76 EXPORT_SYMBOL_GPL(flow_offload_alloc);
78 static u32 flow_offload_dst_cookie(struct flow_offload_tuple *flow_tuple)
80 const struct rt6_info *rt;
82 if (flow_tuple->l3proto == NFPROTO_IPV6) {
83 rt = (const struct rt6_info *)flow_tuple->dst_cache;
84 return rt6_get_cookie(rt);
90 static struct dst_entry *nft_route_dst_fetch(struct nf_flow_route *route,
91 enum flow_offload_tuple_dir dir)
93 struct dst_entry *dst = route->tuple[dir].dst;
95 route->tuple[dir].dst = NULL;
100 static int flow_offload_fill_route(struct flow_offload *flow,
101 struct nf_flow_route *route,
102 enum flow_offload_tuple_dir dir)
104 struct flow_offload_tuple *flow_tuple = &flow->tuplehash[dir].tuple;
105 struct dst_entry *dst = nft_route_dst_fetch(route, dir);
108 switch (flow_tuple->l3proto) {
110 flow_tuple->mtu = ip_dst_mtu_maybe_forward(dst, true);
113 flow_tuple->mtu = ip6_dst_mtu_maybe_forward(dst, true);
117 flow_tuple->iifidx = route->tuple[dir].in.ifindex;
118 for (i = route->tuple[dir].in.num_encaps - 1; i >= 0; i--) {
119 flow_tuple->encap[j].id = route->tuple[dir].in.encap[i].id;
120 flow_tuple->encap[j].proto = route->tuple[dir].in.encap[i].proto;
121 if (route->tuple[dir].in.ingress_vlans & BIT(i))
122 flow_tuple->in_vlan_ingress |= BIT(j);
125 flow_tuple->encap_num = route->tuple[dir].in.num_encaps;
127 switch (route->tuple[dir].xmit_type) {
128 case FLOW_OFFLOAD_XMIT_DIRECT:
129 memcpy(flow_tuple->out.h_dest, route->tuple[dir].out.h_dest,
131 memcpy(flow_tuple->out.h_source, route->tuple[dir].out.h_source,
133 flow_tuple->out.ifidx = route->tuple[dir].out.ifindex;
134 flow_tuple->out.hw_ifidx = route->tuple[dir].out.hw_ifindex;
137 case FLOW_OFFLOAD_XMIT_XFRM:
138 case FLOW_OFFLOAD_XMIT_NEIGH:
139 flow_tuple->dst_cache = dst;
140 flow_tuple->dst_cookie = flow_offload_dst_cookie(flow_tuple);
146 flow_tuple->xmit_type = route->tuple[dir].xmit_type;
151 static void nft_flow_dst_release(struct flow_offload *flow,
152 enum flow_offload_tuple_dir dir)
154 if (flow->tuplehash[dir].tuple.xmit_type == FLOW_OFFLOAD_XMIT_NEIGH ||
155 flow->tuplehash[dir].tuple.xmit_type == FLOW_OFFLOAD_XMIT_XFRM)
156 dst_release(flow->tuplehash[dir].tuple.dst_cache);
159 void flow_offload_route_init(struct flow_offload *flow,
160 struct nf_flow_route *route)
162 flow_offload_fill_route(flow, route, FLOW_OFFLOAD_DIR_ORIGINAL);
163 flow_offload_fill_route(flow, route, FLOW_OFFLOAD_DIR_REPLY);
164 flow->type = NF_FLOW_OFFLOAD_ROUTE;
166 EXPORT_SYMBOL_GPL(flow_offload_route_init);
168 static void flow_offload_fixup_tcp(struct ip_ct_tcp *tcp)
170 tcp->seen[0].td_maxwin = 0;
171 tcp->seen[1].td_maxwin = 0;
174 static void flow_offload_fixup_ct(struct nf_conn *ct)
176 struct net *net = nf_ct_net(ct);
177 int l4num = nf_ct_protonum(ct);
180 if (l4num == IPPROTO_TCP) {
181 struct nf_tcp_net *tn = nf_tcp_pernet(net);
183 flow_offload_fixup_tcp(&ct->proto.tcp);
185 timeout = tn->timeouts[ct->proto.tcp.state];
186 timeout -= tn->offload_timeout;
187 } else if (l4num == IPPROTO_UDP) {
188 struct nf_udp_net *tn = nf_udp_pernet(net);
189 enum udp_conntrack state =
190 test_bit(IPS_SEEN_REPLY_BIT, &ct->status) ?
191 UDP_CT_REPLIED : UDP_CT_UNREPLIED;
193 timeout = tn->timeouts[state];
194 timeout -= tn->offload_timeout;
202 if (nf_flow_timeout_delta(READ_ONCE(ct->timeout)) > (__s32)timeout)
203 WRITE_ONCE(ct->timeout, nfct_time_stamp + timeout);
206 static void flow_offload_route_release(struct flow_offload *flow)
208 nft_flow_dst_release(flow, FLOW_OFFLOAD_DIR_ORIGINAL);
209 nft_flow_dst_release(flow, FLOW_OFFLOAD_DIR_REPLY);
212 void flow_offload_free(struct flow_offload *flow)
214 switch (flow->type) {
215 case NF_FLOW_OFFLOAD_ROUTE:
216 flow_offload_route_release(flow);
222 kfree_rcu(flow, rcu_head);
224 EXPORT_SYMBOL_GPL(flow_offload_free);
226 static u32 flow_offload_hash(const void *data, u32 len, u32 seed)
228 const struct flow_offload_tuple *tuple = data;
230 return jhash(tuple, offsetof(struct flow_offload_tuple, __hash), seed);
233 static u32 flow_offload_hash_obj(const void *data, u32 len, u32 seed)
235 const struct flow_offload_tuple_rhash *tuplehash = data;
237 return jhash(&tuplehash->tuple, offsetof(struct flow_offload_tuple, __hash), seed);
240 static int flow_offload_hash_cmp(struct rhashtable_compare_arg *arg,
243 const struct flow_offload_tuple *tuple = arg->key;
244 const struct flow_offload_tuple_rhash *x = ptr;
246 if (memcmp(&x->tuple, tuple, offsetof(struct flow_offload_tuple, __hash)))
252 static const struct rhashtable_params nf_flow_offload_rhash_params = {
253 .head_offset = offsetof(struct flow_offload_tuple_rhash, node),
254 .hashfn = flow_offload_hash,
255 .obj_hashfn = flow_offload_hash_obj,
256 .obj_cmpfn = flow_offload_hash_cmp,
257 .automatic_shrinking = true,
260 unsigned long flow_offload_get_timeout(struct flow_offload *flow)
262 unsigned long timeout = NF_FLOW_TIMEOUT;
263 struct net *net = nf_ct_net(flow->ct);
264 int l4num = nf_ct_protonum(flow->ct);
266 if (l4num == IPPROTO_TCP) {
267 struct nf_tcp_net *tn = nf_tcp_pernet(net);
269 timeout = tn->offload_timeout;
270 } else if (l4num == IPPROTO_UDP) {
271 struct nf_udp_net *tn = nf_udp_pernet(net);
273 timeout = tn->offload_timeout;
279 int flow_offload_add(struct nf_flowtable *flow_table, struct flow_offload *flow)
283 flow->timeout = nf_flowtable_time_stamp + flow_offload_get_timeout(flow);
285 err = rhashtable_insert_fast(&flow_table->rhashtable,
286 &flow->tuplehash[0].node,
287 nf_flow_offload_rhash_params);
291 err = rhashtable_insert_fast(&flow_table->rhashtable,
292 &flow->tuplehash[1].node,
293 nf_flow_offload_rhash_params);
295 rhashtable_remove_fast(&flow_table->rhashtable,
296 &flow->tuplehash[0].node,
297 nf_flow_offload_rhash_params);
301 nf_ct_offload_timeout(flow->ct);
303 if (nf_flowtable_hw_offload(flow_table)) {
304 __set_bit(NF_FLOW_HW, &flow->flags);
305 nf_flow_offload_add(flow_table, flow);
310 EXPORT_SYMBOL_GPL(flow_offload_add);
312 void flow_offload_refresh(struct nf_flowtable *flow_table,
313 struct flow_offload *flow, bool force)
317 timeout = nf_flowtable_time_stamp + flow_offload_get_timeout(flow);
318 if (force || timeout - READ_ONCE(flow->timeout) > HZ)
319 WRITE_ONCE(flow->timeout, timeout);
323 if (likely(!nf_flowtable_hw_offload(flow_table)))
326 nf_flow_offload_add(flow_table, flow);
328 EXPORT_SYMBOL_GPL(flow_offload_refresh);
330 static inline bool nf_flow_has_expired(const struct flow_offload *flow)
332 return nf_flow_timeout_delta(flow->timeout) <= 0;
335 static void flow_offload_del(struct nf_flowtable *flow_table,
336 struct flow_offload *flow)
338 rhashtable_remove_fast(&flow_table->rhashtable,
339 &flow->tuplehash[FLOW_OFFLOAD_DIR_ORIGINAL].node,
340 nf_flow_offload_rhash_params);
341 rhashtable_remove_fast(&flow_table->rhashtable,
342 &flow->tuplehash[FLOW_OFFLOAD_DIR_REPLY].node,
343 nf_flow_offload_rhash_params);
344 flow_offload_free(flow);
347 void flow_offload_teardown(struct flow_offload *flow)
349 clear_bit(IPS_OFFLOAD_BIT, &flow->ct->status);
350 set_bit(NF_FLOW_TEARDOWN, &flow->flags);
351 flow_offload_fixup_ct(flow->ct);
353 EXPORT_SYMBOL_GPL(flow_offload_teardown);
355 struct flow_offload_tuple_rhash *
356 flow_offload_lookup(struct nf_flowtable *flow_table,
357 struct flow_offload_tuple *tuple)
359 struct flow_offload_tuple_rhash *tuplehash;
360 struct flow_offload *flow;
363 tuplehash = rhashtable_lookup(&flow_table->rhashtable, tuple,
364 nf_flow_offload_rhash_params);
368 dir = tuplehash->tuple.dir;
369 flow = container_of(tuplehash, struct flow_offload, tuplehash[dir]);
370 if (test_bit(NF_FLOW_TEARDOWN, &flow->flags))
373 if (unlikely(nf_ct_is_dying(flow->ct)))
378 EXPORT_SYMBOL_GPL(flow_offload_lookup);
381 nf_flow_table_iterate(struct nf_flowtable *flow_table,
382 void (*iter)(struct nf_flowtable *flowtable,
383 struct flow_offload *flow, void *data),
386 struct flow_offload_tuple_rhash *tuplehash;
387 struct rhashtable_iter hti;
388 struct flow_offload *flow;
391 rhashtable_walk_enter(&flow_table->rhashtable, &hti);
392 rhashtable_walk_start(&hti);
394 while ((tuplehash = rhashtable_walk_next(&hti))) {
395 if (IS_ERR(tuplehash)) {
396 if (PTR_ERR(tuplehash) != -EAGAIN) {
397 err = PTR_ERR(tuplehash);
402 if (tuplehash->tuple.dir)
405 flow = container_of(tuplehash, struct flow_offload, tuplehash[0]);
407 iter(flow_table, flow, data);
409 rhashtable_walk_stop(&hti);
410 rhashtable_walk_exit(&hti);
415 static bool nf_flow_custom_gc(struct nf_flowtable *flow_table,
416 const struct flow_offload *flow)
418 return flow_table->type->gc && flow_table->type->gc(flow);
421 static void nf_flow_offload_gc_step(struct nf_flowtable *flow_table,
422 struct flow_offload *flow, void *data)
424 if (nf_flow_has_expired(flow) ||
425 nf_ct_is_dying(flow->ct) ||
426 nf_flow_custom_gc(flow_table, flow))
427 flow_offload_teardown(flow);
429 if (test_bit(NF_FLOW_TEARDOWN, &flow->flags)) {
430 if (test_bit(NF_FLOW_HW, &flow->flags)) {
431 if (!test_bit(NF_FLOW_HW_DYING, &flow->flags))
432 nf_flow_offload_del(flow_table, flow);
433 else if (test_bit(NF_FLOW_HW_DEAD, &flow->flags))
434 flow_offload_del(flow_table, flow);
436 flow_offload_del(flow_table, flow);
438 } else if (test_bit(NF_FLOW_HW, &flow->flags)) {
439 nf_flow_offload_stats(flow_table, flow);
443 void nf_flow_table_gc_run(struct nf_flowtable *flow_table)
445 nf_flow_table_iterate(flow_table, nf_flow_offload_gc_step, NULL);
448 static void nf_flow_offload_work_gc(struct work_struct *work)
450 struct nf_flowtable *flow_table;
452 flow_table = container_of(work, struct nf_flowtable, gc_work.work);
453 nf_flow_table_gc_run(flow_table);
454 queue_delayed_work(system_power_efficient_wq, &flow_table->gc_work, HZ);
457 static void nf_flow_nat_port_tcp(struct sk_buff *skb, unsigned int thoff,
458 __be16 port, __be16 new_port)
462 tcph = (void *)(skb_network_header(skb) + thoff);
463 inet_proto_csum_replace2(&tcph->check, skb, port, new_port, false);
466 static void nf_flow_nat_port_udp(struct sk_buff *skb, unsigned int thoff,
467 __be16 port, __be16 new_port)
471 udph = (void *)(skb_network_header(skb) + thoff);
472 if (udph->check || skb->ip_summed == CHECKSUM_PARTIAL) {
473 inet_proto_csum_replace2(&udph->check, skb, port,
476 udph->check = CSUM_MANGLED_0;
480 static void nf_flow_nat_port(struct sk_buff *skb, unsigned int thoff,
481 u8 protocol, __be16 port, __be16 new_port)
485 nf_flow_nat_port_tcp(skb, thoff, port, new_port);
488 nf_flow_nat_port_udp(skb, thoff, port, new_port);
493 void nf_flow_snat_port(const struct flow_offload *flow,
494 struct sk_buff *skb, unsigned int thoff,
495 u8 protocol, enum flow_offload_tuple_dir dir)
497 struct flow_ports *hdr;
498 __be16 port, new_port;
500 hdr = (void *)(skb_network_header(skb) + thoff);
503 case FLOW_OFFLOAD_DIR_ORIGINAL:
505 new_port = flow->tuplehash[FLOW_OFFLOAD_DIR_REPLY].tuple.dst_port;
506 hdr->source = new_port;
508 case FLOW_OFFLOAD_DIR_REPLY:
510 new_port = flow->tuplehash[FLOW_OFFLOAD_DIR_ORIGINAL].tuple.src_port;
511 hdr->dest = new_port;
515 nf_flow_nat_port(skb, thoff, protocol, port, new_port);
517 EXPORT_SYMBOL_GPL(nf_flow_snat_port);
519 void nf_flow_dnat_port(const struct flow_offload *flow, struct sk_buff *skb,
520 unsigned int thoff, u8 protocol,
521 enum flow_offload_tuple_dir dir)
523 struct flow_ports *hdr;
524 __be16 port, new_port;
526 hdr = (void *)(skb_network_header(skb) + thoff);
529 case FLOW_OFFLOAD_DIR_ORIGINAL:
531 new_port = flow->tuplehash[FLOW_OFFLOAD_DIR_REPLY].tuple.src_port;
532 hdr->dest = new_port;
534 case FLOW_OFFLOAD_DIR_REPLY:
536 new_port = flow->tuplehash[FLOW_OFFLOAD_DIR_ORIGINAL].tuple.dst_port;
537 hdr->source = new_port;
541 nf_flow_nat_port(skb, thoff, protocol, port, new_port);
543 EXPORT_SYMBOL_GPL(nf_flow_dnat_port);
545 int nf_flow_table_init(struct nf_flowtable *flowtable)
549 INIT_DELAYED_WORK(&flowtable->gc_work, nf_flow_offload_work_gc);
550 flow_block_init(&flowtable->flow_block);
551 init_rwsem(&flowtable->flow_block_lock);
553 err = rhashtable_init(&flowtable->rhashtable,
554 &nf_flow_offload_rhash_params);
558 queue_delayed_work(system_power_efficient_wq,
559 &flowtable->gc_work, HZ);
561 mutex_lock(&flowtable_lock);
562 list_add(&flowtable->list, &flowtables);
563 mutex_unlock(&flowtable_lock);
567 EXPORT_SYMBOL_GPL(nf_flow_table_init);
569 static void nf_flow_table_do_cleanup(struct nf_flowtable *flow_table,
570 struct flow_offload *flow, void *data)
572 struct net_device *dev = data;
575 flow_offload_teardown(flow);
579 if (net_eq(nf_ct_net(flow->ct), dev_net(dev)) &&
580 (flow->tuplehash[0].tuple.iifidx == dev->ifindex ||
581 flow->tuplehash[1].tuple.iifidx == dev->ifindex))
582 flow_offload_teardown(flow);
585 void nf_flow_table_gc_cleanup(struct nf_flowtable *flowtable,
586 struct net_device *dev)
588 nf_flow_table_iterate(flowtable, nf_flow_table_do_cleanup, dev);
589 flush_delayed_work(&flowtable->gc_work);
590 nf_flow_table_offload_flush(flowtable);
593 void nf_flow_table_cleanup(struct net_device *dev)
595 struct nf_flowtable *flowtable;
597 mutex_lock(&flowtable_lock);
598 list_for_each_entry(flowtable, &flowtables, list)
599 nf_flow_table_gc_cleanup(flowtable, dev);
600 mutex_unlock(&flowtable_lock);
602 EXPORT_SYMBOL_GPL(nf_flow_table_cleanup);
604 void nf_flow_table_free(struct nf_flowtable *flow_table)
606 mutex_lock(&flowtable_lock);
607 list_del(&flow_table->list);
608 mutex_unlock(&flowtable_lock);
610 cancel_delayed_work_sync(&flow_table->gc_work);
611 nf_flow_table_offload_flush(flow_table);
612 /* ... no more pending work after this stage ... */
613 nf_flow_table_iterate(flow_table, nf_flow_table_do_cleanup, NULL);
614 nf_flow_table_gc_run(flow_table);
615 nf_flow_table_offload_flush_cleanup(flow_table);
616 rhashtable_destroy(&flow_table->rhashtable);
618 EXPORT_SYMBOL_GPL(nf_flow_table_free);
620 static int nf_flow_table_init_net(struct net *net)
622 net->ft.stat = alloc_percpu(struct nf_flow_table_stat);
623 return net->ft.stat ? 0 : -ENOMEM;
626 static void nf_flow_table_fini_net(struct net *net)
628 free_percpu(net->ft.stat);
631 static int nf_flow_table_pernet_init(struct net *net)
635 ret = nf_flow_table_init_net(net);
639 ret = nf_flow_table_init_proc(net);
646 nf_flow_table_fini_net(net);
650 static void nf_flow_table_pernet_exit(struct list_head *net_exit_list)
654 list_for_each_entry(net, net_exit_list, exit_list) {
655 nf_flow_table_fini_proc(net);
656 nf_flow_table_fini_net(net);
660 static struct pernet_operations nf_flow_table_net_ops = {
661 .init = nf_flow_table_pernet_init,
662 .exit_batch = nf_flow_table_pernet_exit,
665 static int __init nf_flow_table_module_init(void)
669 ret = register_pernet_subsys(&nf_flow_table_net_ops);
673 ret = nf_flow_table_offload_init();
680 unregister_pernet_subsys(&nf_flow_table_net_ops);
684 static void __exit nf_flow_table_module_exit(void)
686 nf_flow_table_offload_exit();
687 unregister_pernet_subsys(&nf_flow_table_net_ops);
690 module_init(nf_flow_table_module_init);
691 module_exit(nf_flow_table_module_exit);
693 MODULE_LICENSE("GPL");
694 MODULE_AUTHOR("Pablo Neira Ayuso <pablo@netfilter.org>");
695 MODULE_DESCRIPTION("Netfilter flow table module");