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_tuple.h>
16 struct flow_offload_entry {
17 struct flow_offload flow;
19 struct rcu_head rcu_head;
22 static DEFINE_MUTEX(flowtable_lock);
23 static LIST_HEAD(flowtables);
26 flow_offload_fill_dir(struct flow_offload *flow, struct nf_conn *ct,
27 struct nf_flow_route *route,
28 enum flow_offload_tuple_dir dir)
30 struct flow_offload_tuple *ft = &flow->tuplehash[dir].tuple;
31 struct nf_conntrack_tuple *ctt = &ct->tuplehash[dir].tuple;
32 struct dst_entry *other_dst = route->tuple[!dir].dst;
33 struct dst_entry *dst = route->tuple[dir].dst;
37 switch (ctt->src.l3num) {
39 ft->src_v4 = ctt->src.u3.in;
40 ft->dst_v4 = ctt->dst.u3.in;
41 ft->mtu = ip_dst_mtu_maybe_forward(dst, true);
44 ft->src_v6 = ctt->src.u3.in6;
45 ft->dst_v6 = ctt->dst.u3.in6;
46 ft->mtu = ip6_dst_mtu_forward(dst);
50 ft->l3proto = ctt->src.l3num;
51 ft->l4proto = ctt->dst.protonum;
52 ft->src_port = ctt->src.u.tcp.port;
53 ft->dst_port = ctt->dst.u.tcp.port;
55 ft->iifidx = other_dst->dev->ifindex;
56 ft->oifidx = dst->dev->ifindex;
61 flow_offload_alloc(struct nf_conn *ct, struct nf_flow_route *route)
63 struct flow_offload_entry *entry;
64 struct flow_offload *flow;
66 if (unlikely(nf_ct_is_dying(ct) ||
67 !atomic_inc_not_zero(&ct->ct_general.use)))
70 entry = kzalloc(sizeof(*entry), GFP_ATOMIC);
76 if (!dst_hold_safe(route->tuple[FLOW_OFFLOAD_DIR_ORIGINAL].dst))
77 goto err_dst_cache_original;
79 if (!dst_hold_safe(route->tuple[FLOW_OFFLOAD_DIR_REPLY].dst))
80 goto err_dst_cache_reply;
84 flow_offload_fill_dir(flow, ct, route, FLOW_OFFLOAD_DIR_ORIGINAL);
85 flow_offload_fill_dir(flow, ct, route, FLOW_OFFLOAD_DIR_REPLY);
87 if (ct->status & IPS_SRC_NAT)
88 flow->flags |= FLOW_OFFLOAD_SNAT;
89 if (ct->status & IPS_DST_NAT)
90 flow->flags |= FLOW_OFFLOAD_DNAT;
95 dst_release(route->tuple[FLOW_OFFLOAD_DIR_ORIGINAL].dst);
96 err_dst_cache_original:
103 EXPORT_SYMBOL_GPL(flow_offload_alloc);
105 static void flow_offload_fixup_tcp(struct ip_ct_tcp *tcp)
107 tcp->state = TCP_CONNTRACK_ESTABLISHED;
108 tcp->seen[0].td_maxwin = 0;
109 tcp->seen[1].td_maxwin = 0;
112 #define NF_FLOWTABLE_TCP_PICKUP_TIMEOUT (120 * HZ)
113 #define NF_FLOWTABLE_UDP_PICKUP_TIMEOUT (30 * HZ)
115 static void flow_offload_fixup_ct_state(struct nf_conn *ct)
117 const struct nf_conntrack_l4proto *l4proto;
118 unsigned int timeout;
121 l4num = nf_ct_protonum(ct);
122 if (l4num == IPPROTO_TCP)
123 flow_offload_fixup_tcp(&ct->proto.tcp);
125 l4proto = nf_ct_l4proto_find(l4num);
129 if (l4num == IPPROTO_TCP)
130 timeout = NF_FLOWTABLE_TCP_PICKUP_TIMEOUT;
131 else if (l4num == IPPROTO_UDP)
132 timeout = NF_FLOWTABLE_UDP_PICKUP_TIMEOUT;
136 ct->timeout = nfct_time_stamp + timeout;
139 void flow_offload_free(struct flow_offload *flow)
141 struct flow_offload_entry *e;
143 dst_release(flow->tuplehash[FLOW_OFFLOAD_DIR_ORIGINAL].tuple.dst_cache);
144 dst_release(flow->tuplehash[FLOW_OFFLOAD_DIR_REPLY].tuple.dst_cache);
145 e = container_of(flow, struct flow_offload_entry, flow);
146 if (flow->flags & FLOW_OFFLOAD_DYING)
147 nf_ct_delete(e->ct, 0, 0);
149 kfree_rcu(e, rcu_head);
151 EXPORT_SYMBOL_GPL(flow_offload_free);
153 static u32 flow_offload_hash(const void *data, u32 len, u32 seed)
155 const struct flow_offload_tuple *tuple = data;
157 return jhash(tuple, offsetof(struct flow_offload_tuple, dir), seed);
160 static u32 flow_offload_hash_obj(const void *data, u32 len, u32 seed)
162 const struct flow_offload_tuple_rhash *tuplehash = data;
164 return jhash(&tuplehash->tuple, offsetof(struct flow_offload_tuple, dir), seed);
167 static int flow_offload_hash_cmp(struct rhashtable_compare_arg *arg,
170 const struct flow_offload_tuple *tuple = arg->key;
171 const struct flow_offload_tuple_rhash *x = ptr;
173 if (memcmp(&x->tuple, tuple, offsetof(struct flow_offload_tuple, dir)))
179 static const struct rhashtable_params nf_flow_offload_rhash_params = {
180 .head_offset = offsetof(struct flow_offload_tuple_rhash, node),
181 .hashfn = flow_offload_hash,
182 .obj_hashfn = flow_offload_hash_obj,
183 .obj_cmpfn = flow_offload_hash_cmp,
184 .automatic_shrinking = true,
187 int flow_offload_add(struct nf_flowtable *flow_table, struct flow_offload *flow)
191 err = rhashtable_insert_fast(&flow_table->rhashtable,
192 &flow->tuplehash[0].node,
193 nf_flow_offload_rhash_params);
197 err = rhashtable_insert_fast(&flow_table->rhashtable,
198 &flow->tuplehash[1].node,
199 nf_flow_offload_rhash_params);
201 rhashtable_remove_fast(&flow_table->rhashtable,
202 &flow->tuplehash[0].node,
203 nf_flow_offload_rhash_params);
207 flow->timeout = (u32)jiffies;
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;
247 struct flow_offload_entry *e;
250 tuplehash = rhashtable_lookup(&flow_table->rhashtable, tuple,
251 nf_flow_offload_rhash_params);
255 dir = tuplehash->tuple.dir;
256 flow = container_of(tuplehash, struct flow_offload, tuplehash[dir]);
257 if (flow->flags & (FLOW_OFFLOAD_DYING | FLOW_OFFLOAD_TEARDOWN))
260 e = container_of(flow, struct flow_offload_entry, flow);
261 if (unlikely(nf_ct_is_dying(e->ct)))
266 EXPORT_SYMBOL_GPL(flow_offload_lookup);
269 nf_flow_table_iterate(struct nf_flowtable *flow_table,
270 void (*iter)(struct flow_offload *flow, void *data),
273 struct flow_offload_tuple_rhash *tuplehash;
274 struct rhashtable_iter hti;
275 struct flow_offload *flow;
278 rhashtable_walk_enter(&flow_table->rhashtable, &hti);
279 rhashtable_walk_start(&hti);
281 while ((tuplehash = rhashtable_walk_next(&hti))) {
282 if (IS_ERR(tuplehash)) {
283 if (PTR_ERR(tuplehash) != -EAGAIN) {
284 err = PTR_ERR(tuplehash);
289 if (tuplehash->tuple.dir)
292 flow = container_of(tuplehash, struct flow_offload, tuplehash[0]);
296 rhashtable_walk_stop(&hti);
297 rhashtable_walk_exit(&hti);
302 static inline bool nf_flow_has_expired(const struct flow_offload *flow)
304 return (__s32)(flow->timeout - (u32)jiffies) <= 0;
307 static void nf_flow_offload_gc_step(struct flow_offload *flow, void *data)
309 struct nf_flowtable *flow_table = data;
310 struct flow_offload_entry *e;
312 e = container_of(flow, struct flow_offload_entry, flow);
313 if (nf_flow_has_expired(flow) || nf_ct_is_dying(e->ct) ||
314 (flow->flags & (FLOW_OFFLOAD_DYING | FLOW_OFFLOAD_TEARDOWN)))
315 flow_offload_del(flow_table, flow);
318 static void nf_flow_offload_work_gc(struct work_struct *work)
320 struct nf_flowtable *flow_table;
322 flow_table = container_of(work, struct nf_flowtable, gc_work.work);
323 nf_flow_table_iterate(flow_table, nf_flow_offload_gc_step, flow_table);
324 queue_delayed_work(system_power_efficient_wq, &flow_table->gc_work, HZ);
327 static int nf_flow_nat_port_tcp(struct sk_buff *skb, unsigned int thoff,
328 __be16 port, __be16 new_port)
332 if (!pskb_may_pull(skb, thoff + sizeof(*tcph)) ||
333 skb_try_make_writable(skb, thoff + sizeof(*tcph)))
336 tcph = (void *)(skb_network_header(skb) + thoff);
337 inet_proto_csum_replace2(&tcph->check, skb, port, new_port, true);
342 static int nf_flow_nat_port_udp(struct sk_buff *skb, unsigned int thoff,
343 __be16 port, __be16 new_port)
347 if (!pskb_may_pull(skb, thoff + sizeof(*udph)) ||
348 skb_try_make_writable(skb, thoff + sizeof(*udph)))
351 udph = (void *)(skb_network_header(skb) + thoff);
352 if (udph->check || skb->ip_summed == CHECKSUM_PARTIAL) {
353 inet_proto_csum_replace2(&udph->check, skb, port,
356 udph->check = CSUM_MANGLED_0;
362 static int nf_flow_nat_port(struct sk_buff *skb, unsigned int thoff,
363 u8 protocol, __be16 port, __be16 new_port)
367 if (nf_flow_nat_port_tcp(skb, thoff, port, new_port) < 0)
371 if (nf_flow_nat_port_udp(skb, thoff, port, new_port) < 0)
379 int nf_flow_snat_port(const struct flow_offload *flow,
380 struct sk_buff *skb, unsigned int thoff,
381 u8 protocol, enum flow_offload_tuple_dir dir)
383 struct flow_ports *hdr;
384 __be16 port, new_port;
386 if (!pskb_may_pull(skb, thoff + sizeof(*hdr)) ||
387 skb_try_make_writable(skb, thoff + sizeof(*hdr)))
390 hdr = (void *)(skb_network_header(skb) + thoff);
393 case FLOW_OFFLOAD_DIR_ORIGINAL:
395 new_port = flow->tuplehash[FLOW_OFFLOAD_DIR_REPLY].tuple.dst_port;
396 hdr->source = new_port;
398 case FLOW_OFFLOAD_DIR_REPLY:
400 new_port = flow->tuplehash[FLOW_OFFLOAD_DIR_ORIGINAL].tuple.src_port;
401 hdr->dest = new_port;
407 return nf_flow_nat_port(skb, thoff, protocol, port, new_port);
409 EXPORT_SYMBOL_GPL(nf_flow_snat_port);
411 int nf_flow_dnat_port(const struct flow_offload *flow,
412 struct sk_buff *skb, unsigned int thoff,
413 u8 protocol, enum flow_offload_tuple_dir dir)
415 struct flow_ports *hdr;
416 __be16 port, new_port;
418 if (!pskb_may_pull(skb, thoff + sizeof(*hdr)) ||
419 skb_try_make_writable(skb, thoff + sizeof(*hdr)))
422 hdr = (void *)(skb_network_header(skb) + thoff);
425 case FLOW_OFFLOAD_DIR_ORIGINAL:
427 new_port = flow->tuplehash[FLOW_OFFLOAD_DIR_REPLY].tuple.src_port;
428 hdr->dest = new_port;
430 case FLOW_OFFLOAD_DIR_REPLY:
432 new_port = flow->tuplehash[FLOW_OFFLOAD_DIR_ORIGINAL].tuple.dst_port;
433 hdr->source = new_port;
439 return nf_flow_nat_port(skb, thoff, protocol, port, new_port);
441 EXPORT_SYMBOL_GPL(nf_flow_dnat_port);
443 int nf_flow_table_init(struct nf_flowtable *flowtable)
447 INIT_DEFERRABLE_WORK(&flowtable->gc_work, nf_flow_offload_work_gc);
449 err = rhashtable_init(&flowtable->rhashtable,
450 &nf_flow_offload_rhash_params);
454 queue_delayed_work(system_power_efficient_wq,
455 &flowtable->gc_work, HZ);
457 mutex_lock(&flowtable_lock);
458 list_add(&flowtable->list, &flowtables);
459 mutex_unlock(&flowtable_lock);
463 EXPORT_SYMBOL_GPL(nf_flow_table_init);
465 static void nf_flow_table_do_cleanup(struct flow_offload *flow, void *data)
467 struct net_device *dev = data;
468 struct flow_offload_entry *e;
470 e = container_of(flow, struct flow_offload_entry, flow);
473 flow_offload_teardown(flow);
476 if (net_eq(nf_ct_net(e->ct), dev_net(dev)) &&
477 (flow->tuplehash[0].tuple.iifidx == dev->ifindex ||
478 flow->tuplehash[1].tuple.iifidx == dev->ifindex))
479 flow_offload_dead(flow);
482 static void nf_flow_table_iterate_cleanup(struct nf_flowtable *flowtable,
483 struct net_device *dev)
485 nf_flow_table_iterate(flowtable, nf_flow_table_do_cleanup, dev);
486 flush_delayed_work(&flowtable->gc_work);
489 void nf_flow_table_cleanup(struct net_device *dev)
491 struct nf_flowtable *flowtable;
493 mutex_lock(&flowtable_lock);
494 list_for_each_entry(flowtable, &flowtables, list)
495 nf_flow_table_iterate_cleanup(flowtable, dev);
496 mutex_unlock(&flowtable_lock);
498 EXPORT_SYMBOL_GPL(nf_flow_table_cleanup);
500 void nf_flow_table_free(struct nf_flowtable *flow_table)
502 mutex_lock(&flowtable_lock);
503 list_del(&flow_table->list);
504 mutex_unlock(&flowtable_lock);
505 cancel_delayed_work_sync(&flow_table->gc_work);
506 nf_flow_table_iterate(flow_table, nf_flow_table_do_cleanup, NULL);
507 nf_flow_table_iterate(flow_table, nf_flow_offload_gc_step, flow_table);
508 rhashtable_destroy(&flow_table->rhashtable);
510 EXPORT_SYMBOL_GPL(nf_flow_table_free);
512 MODULE_LICENSE("GPL");
513 MODULE_AUTHOR("Pablo Neira Ayuso <pablo@netfilter.org>");