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 int flow_offload_fill_route(struct flow_offload *flow,
91 const struct nf_flow_route *route,
92 enum flow_offload_tuple_dir dir)
94 struct flow_offload_tuple *flow_tuple = &flow->tuplehash[dir].tuple;
95 struct dst_entry *dst = route->tuple[dir].dst;
98 switch (flow_tuple->l3proto) {
100 flow_tuple->mtu = ip_dst_mtu_maybe_forward(dst, true);
103 flow_tuple->mtu = ip6_dst_mtu_maybe_forward(dst, true);
107 flow_tuple->iifidx = route->tuple[dir].in.ifindex;
108 for (i = route->tuple[dir].in.num_encaps - 1; i >= 0; i--) {
109 flow_tuple->encap[j].id = route->tuple[dir].in.encap[i].id;
110 flow_tuple->encap[j].proto = route->tuple[dir].in.encap[i].proto;
111 if (route->tuple[dir].in.ingress_vlans & BIT(i))
112 flow_tuple->in_vlan_ingress |= BIT(j);
115 flow_tuple->encap_num = route->tuple[dir].in.num_encaps;
117 switch (route->tuple[dir].xmit_type) {
118 case FLOW_OFFLOAD_XMIT_DIRECT:
119 memcpy(flow_tuple->out.h_dest, route->tuple[dir].out.h_dest,
121 memcpy(flow_tuple->out.h_source, route->tuple[dir].out.h_source,
123 flow_tuple->out.ifidx = route->tuple[dir].out.ifindex;
124 flow_tuple->out.hw_ifidx = route->tuple[dir].out.hw_ifindex;
126 case FLOW_OFFLOAD_XMIT_XFRM:
127 case FLOW_OFFLOAD_XMIT_NEIGH:
128 flow_tuple->dst_cache = dst;
129 flow_tuple->dst_cookie = flow_offload_dst_cookie(flow_tuple);
135 flow_tuple->xmit_type = route->tuple[dir].xmit_type;
140 static void nft_flow_dst_release(struct flow_offload *flow,
141 enum flow_offload_tuple_dir dir)
143 if (flow->tuplehash[dir].tuple.xmit_type == FLOW_OFFLOAD_XMIT_NEIGH ||
144 flow->tuplehash[dir].tuple.xmit_type == FLOW_OFFLOAD_XMIT_XFRM)
145 dst_release(flow->tuplehash[dir].tuple.dst_cache);
148 void flow_offload_route_init(struct flow_offload *flow,
149 const struct nf_flow_route *route)
151 flow_offload_fill_route(flow, route, FLOW_OFFLOAD_DIR_ORIGINAL);
152 flow_offload_fill_route(flow, route, FLOW_OFFLOAD_DIR_REPLY);
153 flow->type = NF_FLOW_OFFLOAD_ROUTE;
155 EXPORT_SYMBOL_GPL(flow_offload_route_init);
157 static void flow_offload_fixup_tcp(struct ip_ct_tcp *tcp)
159 tcp->seen[0].td_maxwin = 0;
160 tcp->seen[1].td_maxwin = 0;
163 static void flow_offload_fixup_ct(struct nf_conn *ct)
165 struct net *net = nf_ct_net(ct);
166 int l4num = nf_ct_protonum(ct);
169 if (l4num == IPPROTO_TCP) {
170 struct nf_tcp_net *tn = nf_tcp_pernet(net);
172 flow_offload_fixup_tcp(&ct->proto.tcp);
174 timeout = tn->timeouts[ct->proto.tcp.state];
175 timeout -= tn->offload_timeout;
176 } else if (l4num == IPPROTO_UDP) {
177 struct nf_udp_net *tn = nf_udp_pernet(net);
178 enum udp_conntrack state =
179 test_bit(IPS_SEEN_REPLY_BIT, &ct->status) ?
180 UDP_CT_REPLIED : UDP_CT_UNREPLIED;
182 timeout = tn->timeouts[state];
183 timeout -= tn->offload_timeout;
191 if (nf_flow_timeout_delta(READ_ONCE(ct->timeout)) > (__s32)timeout)
192 WRITE_ONCE(ct->timeout, nfct_time_stamp + timeout);
195 static void flow_offload_route_release(struct flow_offload *flow)
197 nft_flow_dst_release(flow, FLOW_OFFLOAD_DIR_ORIGINAL);
198 nft_flow_dst_release(flow, FLOW_OFFLOAD_DIR_REPLY);
201 void flow_offload_free(struct flow_offload *flow)
203 switch (flow->type) {
204 case NF_FLOW_OFFLOAD_ROUTE:
205 flow_offload_route_release(flow);
211 kfree_rcu(flow, rcu_head);
213 EXPORT_SYMBOL_GPL(flow_offload_free);
215 static u32 flow_offload_hash(const void *data, u32 len, u32 seed)
217 const struct flow_offload_tuple *tuple = data;
219 return jhash(tuple, offsetof(struct flow_offload_tuple, __hash), seed);
222 static u32 flow_offload_hash_obj(const void *data, u32 len, u32 seed)
224 const struct flow_offload_tuple_rhash *tuplehash = data;
226 return jhash(&tuplehash->tuple, offsetof(struct flow_offload_tuple, __hash), seed);
229 static int flow_offload_hash_cmp(struct rhashtable_compare_arg *arg,
232 const struct flow_offload_tuple *tuple = arg->key;
233 const struct flow_offload_tuple_rhash *x = ptr;
235 if (memcmp(&x->tuple, tuple, offsetof(struct flow_offload_tuple, __hash)))
241 static const struct rhashtable_params nf_flow_offload_rhash_params = {
242 .head_offset = offsetof(struct flow_offload_tuple_rhash, node),
243 .hashfn = flow_offload_hash,
244 .obj_hashfn = flow_offload_hash_obj,
245 .obj_cmpfn = flow_offload_hash_cmp,
246 .automatic_shrinking = true,
249 unsigned long flow_offload_get_timeout(struct flow_offload *flow)
251 unsigned long timeout = NF_FLOW_TIMEOUT;
252 struct net *net = nf_ct_net(flow->ct);
253 int l4num = nf_ct_protonum(flow->ct);
255 if (l4num == IPPROTO_TCP) {
256 struct nf_tcp_net *tn = nf_tcp_pernet(net);
258 timeout = tn->offload_timeout;
259 } else if (l4num == IPPROTO_UDP) {
260 struct nf_udp_net *tn = nf_udp_pernet(net);
262 timeout = tn->offload_timeout;
268 int flow_offload_add(struct nf_flowtable *flow_table, struct flow_offload *flow)
272 flow->timeout = nf_flowtable_time_stamp + flow_offload_get_timeout(flow);
274 err = rhashtable_insert_fast(&flow_table->rhashtable,
275 &flow->tuplehash[0].node,
276 nf_flow_offload_rhash_params);
280 err = rhashtable_insert_fast(&flow_table->rhashtable,
281 &flow->tuplehash[1].node,
282 nf_flow_offload_rhash_params);
284 rhashtable_remove_fast(&flow_table->rhashtable,
285 &flow->tuplehash[0].node,
286 nf_flow_offload_rhash_params);
290 nf_ct_offload_timeout(flow->ct);
292 if (nf_flowtable_hw_offload(flow_table)) {
293 __set_bit(NF_FLOW_HW, &flow->flags);
294 nf_flow_offload_add(flow_table, flow);
299 EXPORT_SYMBOL_GPL(flow_offload_add);
301 void flow_offload_refresh(struct nf_flowtable *flow_table,
302 struct flow_offload *flow, bool force)
306 timeout = nf_flowtable_time_stamp + flow_offload_get_timeout(flow);
307 if (force || timeout - READ_ONCE(flow->timeout) > HZ)
308 WRITE_ONCE(flow->timeout, timeout);
312 if (likely(!nf_flowtable_hw_offload(flow_table)))
315 nf_flow_offload_add(flow_table, flow);
317 EXPORT_SYMBOL_GPL(flow_offload_refresh);
319 static inline bool nf_flow_has_expired(const struct flow_offload *flow)
321 return nf_flow_timeout_delta(flow->timeout) <= 0;
324 static void flow_offload_del(struct nf_flowtable *flow_table,
325 struct flow_offload *flow)
327 rhashtable_remove_fast(&flow_table->rhashtable,
328 &flow->tuplehash[FLOW_OFFLOAD_DIR_ORIGINAL].node,
329 nf_flow_offload_rhash_params);
330 rhashtable_remove_fast(&flow_table->rhashtable,
331 &flow->tuplehash[FLOW_OFFLOAD_DIR_REPLY].node,
332 nf_flow_offload_rhash_params);
333 flow_offload_free(flow);
336 void flow_offload_teardown(struct flow_offload *flow)
338 clear_bit(IPS_OFFLOAD_BIT, &flow->ct->status);
339 set_bit(NF_FLOW_TEARDOWN, &flow->flags);
340 flow_offload_fixup_ct(flow->ct);
342 EXPORT_SYMBOL_GPL(flow_offload_teardown);
344 struct flow_offload_tuple_rhash *
345 flow_offload_lookup(struct nf_flowtable *flow_table,
346 struct flow_offload_tuple *tuple)
348 struct flow_offload_tuple_rhash *tuplehash;
349 struct flow_offload *flow;
352 tuplehash = rhashtable_lookup(&flow_table->rhashtable, tuple,
353 nf_flow_offload_rhash_params);
357 dir = tuplehash->tuple.dir;
358 flow = container_of(tuplehash, struct flow_offload, tuplehash[dir]);
359 if (test_bit(NF_FLOW_TEARDOWN, &flow->flags))
362 if (unlikely(nf_ct_is_dying(flow->ct)))
367 EXPORT_SYMBOL_GPL(flow_offload_lookup);
370 nf_flow_table_iterate(struct nf_flowtable *flow_table,
371 void (*iter)(struct nf_flowtable *flowtable,
372 struct flow_offload *flow, void *data),
375 struct flow_offload_tuple_rhash *tuplehash;
376 struct rhashtable_iter hti;
377 struct flow_offload *flow;
380 rhashtable_walk_enter(&flow_table->rhashtable, &hti);
381 rhashtable_walk_start(&hti);
383 while ((tuplehash = rhashtable_walk_next(&hti))) {
384 if (IS_ERR(tuplehash)) {
385 if (PTR_ERR(tuplehash) != -EAGAIN) {
386 err = PTR_ERR(tuplehash);
391 if (tuplehash->tuple.dir)
394 flow = container_of(tuplehash, struct flow_offload, tuplehash[0]);
396 iter(flow_table, flow, data);
398 rhashtable_walk_stop(&hti);
399 rhashtable_walk_exit(&hti);
404 static bool nf_flow_custom_gc(struct nf_flowtable *flow_table,
405 const struct flow_offload *flow)
407 return flow_table->type->gc && flow_table->type->gc(flow);
410 static void nf_flow_offload_gc_step(struct nf_flowtable *flow_table,
411 struct flow_offload *flow, void *data)
413 if (nf_flow_has_expired(flow) ||
414 nf_ct_is_dying(flow->ct) ||
415 nf_flow_custom_gc(flow_table, flow))
416 flow_offload_teardown(flow);
418 if (test_bit(NF_FLOW_TEARDOWN, &flow->flags)) {
419 if (test_bit(NF_FLOW_HW, &flow->flags)) {
420 if (!test_bit(NF_FLOW_HW_DYING, &flow->flags))
421 nf_flow_offload_del(flow_table, flow);
422 else if (test_bit(NF_FLOW_HW_DEAD, &flow->flags))
423 flow_offload_del(flow_table, flow);
425 flow_offload_del(flow_table, flow);
427 } else if (test_bit(NF_FLOW_HW, &flow->flags)) {
428 nf_flow_offload_stats(flow_table, flow);
432 void nf_flow_table_gc_run(struct nf_flowtable *flow_table)
434 nf_flow_table_iterate(flow_table, nf_flow_offload_gc_step, NULL);
437 static void nf_flow_offload_work_gc(struct work_struct *work)
439 struct nf_flowtable *flow_table;
441 flow_table = container_of(work, struct nf_flowtable, gc_work.work);
442 nf_flow_table_gc_run(flow_table);
443 queue_delayed_work(system_power_efficient_wq, &flow_table->gc_work, HZ);
446 static void nf_flow_nat_port_tcp(struct sk_buff *skb, unsigned int thoff,
447 __be16 port, __be16 new_port)
451 tcph = (void *)(skb_network_header(skb) + thoff);
452 inet_proto_csum_replace2(&tcph->check, skb, port, new_port, false);
455 static void nf_flow_nat_port_udp(struct sk_buff *skb, unsigned int thoff,
456 __be16 port, __be16 new_port)
460 udph = (void *)(skb_network_header(skb) + thoff);
461 if (udph->check || skb->ip_summed == CHECKSUM_PARTIAL) {
462 inet_proto_csum_replace2(&udph->check, skb, port,
465 udph->check = CSUM_MANGLED_0;
469 static void nf_flow_nat_port(struct sk_buff *skb, unsigned int thoff,
470 u8 protocol, __be16 port, __be16 new_port)
474 nf_flow_nat_port_tcp(skb, thoff, port, new_port);
477 nf_flow_nat_port_udp(skb, thoff, port, new_port);
482 void nf_flow_snat_port(const struct flow_offload *flow,
483 struct sk_buff *skb, unsigned int thoff,
484 u8 protocol, enum flow_offload_tuple_dir dir)
486 struct flow_ports *hdr;
487 __be16 port, new_port;
489 hdr = (void *)(skb_network_header(skb) + thoff);
492 case FLOW_OFFLOAD_DIR_ORIGINAL:
494 new_port = flow->tuplehash[FLOW_OFFLOAD_DIR_REPLY].tuple.dst_port;
495 hdr->source = new_port;
497 case FLOW_OFFLOAD_DIR_REPLY:
499 new_port = flow->tuplehash[FLOW_OFFLOAD_DIR_ORIGINAL].tuple.src_port;
500 hdr->dest = new_port;
504 nf_flow_nat_port(skb, thoff, protocol, port, new_port);
506 EXPORT_SYMBOL_GPL(nf_flow_snat_port);
508 void nf_flow_dnat_port(const struct flow_offload *flow, struct sk_buff *skb,
509 unsigned int thoff, u8 protocol,
510 enum flow_offload_tuple_dir dir)
512 struct flow_ports *hdr;
513 __be16 port, new_port;
515 hdr = (void *)(skb_network_header(skb) + thoff);
518 case FLOW_OFFLOAD_DIR_ORIGINAL:
520 new_port = flow->tuplehash[FLOW_OFFLOAD_DIR_REPLY].tuple.src_port;
521 hdr->dest = new_port;
523 case FLOW_OFFLOAD_DIR_REPLY:
525 new_port = flow->tuplehash[FLOW_OFFLOAD_DIR_ORIGINAL].tuple.dst_port;
526 hdr->source = new_port;
530 nf_flow_nat_port(skb, thoff, protocol, port, new_port);
532 EXPORT_SYMBOL_GPL(nf_flow_dnat_port);
534 int nf_flow_table_init(struct nf_flowtable *flowtable)
538 INIT_DELAYED_WORK(&flowtable->gc_work, nf_flow_offload_work_gc);
539 flow_block_init(&flowtable->flow_block);
540 init_rwsem(&flowtable->flow_block_lock);
542 err = rhashtable_init(&flowtable->rhashtable,
543 &nf_flow_offload_rhash_params);
547 queue_delayed_work(system_power_efficient_wq,
548 &flowtable->gc_work, HZ);
550 mutex_lock(&flowtable_lock);
551 list_add(&flowtable->list, &flowtables);
552 mutex_unlock(&flowtable_lock);
556 EXPORT_SYMBOL_GPL(nf_flow_table_init);
558 static void nf_flow_table_do_cleanup(struct nf_flowtable *flow_table,
559 struct flow_offload *flow, void *data)
561 struct net_device *dev = data;
564 flow_offload_teardown(flow);
568 if (net_eq(nf_ct_net(flow->ct), dev_net(dev)) &&
569 (flow->tuplehash[0].tuple.iifidx == dev->ifindex ||
570 flow->tuplehash[1].tuple.iifidx == dev->ifindex))
571 flow_offload_teardown(flow);
574 void nf_flow_table_gc_cleanup(struct nf_flowtable *flowtable,
575 struct net_device *dev)
577 nf_flow_table_iterate(flowtable, nf_flow_table_do_cleanup, dev);
578 flush_delayed_work(&flowtable->gc_work);
579 nf_flow_table_offload_flush(flowtable);
582 void nf_flow_table_cleanup(struct net_device *dev)
584 struct nf_flowtable *flowtable;
586 mutex_lock(&flowtable_lock);
587 list_for_each_entry(flowtable, &flowtables, list)
588 nf_flow_table_gc_cleanup(flowtable, dev);
589 mutex_unlock(&flowtable_lock);
591 EXPORT_SYMBOL_GPL(nf_flow_table_cleanup);
593 void nf_flow_table_free(struct nf_flowtable *flow_table)
595 mutex_lock(&flowtable_lock);
596 list_del(&flow_table->list);
597 mutex_unlock(&flowtable_lock);
599 cancel_delayed_work_sync(&flow_table->gc_work);
600 nf_flow_table_offload_flush(flow_table);
601 /* ... no more pending work after this stage ... */
602 nf_flow_table_iterate(flow_table, nf_flow_table_do_cleanup, NULL);
603 nf_flow_table_gc_run(flow_table);
604 nf_flow_table_offload_flush_cleanup(flow_table);
605 rhashtable_destroy(&flow_table->rhashtable);
607 EXPORT_SYMBOL_GPL(nf_flow_table_free);
609 static int nf_flow_table_init_net(struct net *net)
611 net->ft.stat = alloc_percpu(struct nf_flow_table_stat);
612 return net->ft.stat ? 0 : -ENOMEM;
615 static void nf_flow_table_fini_net(struct net *net)
617 free_percpu(net->ft.stat);
620 static int nf_flow_table_pernet_init(struct net *net)
624 ret = nf_flow_table_init_net(net);
628 ret = nf_flow_table_init_proc(net);
635 nf_flow_table_fini_net(net);
639 static void nf_flow_table_pernet_exit(struct list_head *net_exit_list)
643 list_for_each_entry(net, net_exit_list, exit_list) {
644 nf_flow_table_fini_proc(net);
645 nf_flow_table_fini_net(net);
649 static struct pernet_operations nf_flow_table_net_ops = {
650 .init = nf_flow_table_pernet_init,
651 .exit_batch = nf_flow_table_pernet_exit,
654 static int __init nf_flow_table_module_init(void)
658 ret = register_pernet_subsys(&nf_flow_table_net_ops);
662 ret = nf_flow_table_offload_init();
669 unregister_pernet_subsys(&nf_flow_table_net_ops);
673 static void __exit nf_flow_table_module_exit(void)
675 nf_flow_table_offload_exit();
676 unregister_pernet_subsys(&nf_flow_table_net_ops);
679 module_init(nf_flow_table_module_init);
680 module_exit(nf_flow_table_module_exit);
682 MODULE_LICENSE("GPL");
683 MODULE_AUTHOR("Pablo Neira Ayuso <pablo@netfilter.org>");
684 MODULE_DESCRIPTION("Netfilter flow table module");