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;
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(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(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 err = rhashtable_insert_fast(&flow_table->rhashtable,
191 &flow->tuplehash[0].node,
192 nf_flow_offload_rhash_params);
196 err = rhashtable_insert_fast(&flow_table->rhashtable,
197 &flow->tuplehash[1].node,
198 nf_flow_offload_rhash_params);
200 rhashtable_remove_fast(&flow_table->rhashtable,
201 &flow->tuplehash[0].node,
202 nf_flow_offload_rhash_params);
206 flow->timeout = (u32)jiffies;
209 EXPORT_SYMBOL_GPL(flow_offload_add);
211 static inline bool nf_flow_has_expired(const struct flow_offload *flow)
213 return (__s32)(flow->timeout - (u32)jiffies) <= 0;
216 static void flow_offload_del(struct nf_flowtable *flow_table,
217 struct flow_offload *flow)
219 struct flow_offload_entry *e;
221 rhashtable_remove_fast(&flow_table->rhashtable,
222 &flow->tuplehash[FLOW_OFFLOAD_DIR_ORIGINAL].node,
223 nf_flow_offload_rhash_params);
224 rhashtable_remove_fast(&flow_table->rhashtable,
225 &flow->tuplehash[FLOW_OFFLOAD_DIR_REPLY].node,
226 nf_flow_offload_rhash_params);
228 e = container_of(flow, struct flow_offload_entry, flow);
229 clear_bit(IPS_OFFLOAD_BIT, &e->ct->status);
231 if (nf_flow_has_expired(flow))
232 flow_offload_fixup_ct(e->ct);
234 flow_offload_free(flow);
237 void flow_offload_teardown(struct flow_offload *flow)
239 struct flow_offload_entry *e;
241 flow->flags |= FLOW_OFFLOAD_TEARDOWN;
243 e = container_of(flow, struct flow_offload_entry, flow);
244 flow_offload_fixup_ct(e->ct);
246 EXPORT_SYMBOL_GPL(flow_offload_teardown);
248 struct flow_offload_tuple_rhash *
249 flow_offload_lookup(struct nf_flowtable *flow_table,
250 struct flow_offload_tuple *tuple)
252 struct flow_offload_tuple_rhash *tuplehash;
253 struct flow_offload *flow;
254 struct flow_offload_entry *e;
257 tuplehash = rhashtable_lookup(&flow_table->rhashtable, tuple,
258 nf_flow_offload_rhash_params);
262 dir = tuplehash->tuple.dir;
263 flow = container_of(tuplehash, struct flow_offload, tuplehash[dir]);
264 if (flow->flags & (FLOW_OFFLOAD_DYING | FLOW_OFFLOAD_TEARDOWN))
267 e = container_of(flow, struct flow_offload_entry, flow);
268 if (unlikely(nf_ct_is_dying(e->ct)))
273 EXPORT_SYMBOL_GPL(flow_offload_lookup);
276 nf_flow_table_iterate(struct nf_flowtable *flow_table,
277 void (*iter)(struct flow_offload *flow, void *data),
280 struct flow_offload_tuple_rhash *tuplehash;
281 struct rhashtable_iter hti;
282 struct flow_offload *flow;
285 rhashtable_walk_enter(&flow_table->rhashtable, &hti);
286 rhashtable_walk_start(&hti);
288 while ((tuplehash = rhashtable_walk_next(&hti))) {
289 if (IS_ERR(tuplehash)) {
290 if (PTR_ERR(tuplehash) != -EAGAIN) {
291 err = PTR_ERR(tuplehash);
296 if (tuplehash->tuple.dir)
299 flow = container_of(tuplehash, struct flow_offload, tuplehash[0]);
303 rhashtable_walk_stop(&hti);
304 rhashtable_walk_exit(&hti);
309 static void nf_flow_offload_gc_step(struct flow_offload *flow, void *data)
311 struct nf_flowtable *flow_table = data;
312 struct flow_offload_entry *e;
314 e = container_of(flow, struct flow_offload_entry, flow);
315 if (nf_flow_has_expired(flow) || nf_ct_is_dying(e->ct) ||
316 (flow->flags & (FLOW_OFFLOAD_DYING | FLOW_OFFLOAD_TEARDOWN)))
317 flow_offload_del(flow_table, flow);
320 static void nf_flow_offload_work_gc(struct work_struct *work)
322 struct nf_flowtable *flow_table;
324 flow_table = container_of(work, struct nf_flowtable, gc_work.work);
325 nf_flow_table_iterate(flow_table, nf_flow_offload_gc_step, flow_table);
326 queue_delayed_work(system_power_efficient_wq, &flow_table->gc_work, HZ);
329 static int nf_flow_nat_port_tcp(struct sk_buff *skb, unsigned int thoff,
330 __be16 port, __be16 new_port)
334 if (!pskb_may_pull(skb, thoff + sizeof(*tcph)) ||
335 skb_try_make_writable(skb, thoff + sizeof(*tcph)))
338 tcph = (void *)(skb_network_header(skb) + thoff);
339 inet_proto_csum_replace2(&tcph->check, skb, port, new_port, true);
344 static int nf_flow_nat_port_udp(struct sk_buff *skb, unsigned int thoff,
345 __be16 port, __be16 new_port)
349 if (!pskb_may_pull(skb, thoff + sizeof(*udph)) ||
350 skb_try_make_writable(skb, thoff + sizeof(*udph)))
353 udph = (void *)(skb_network_header(skb) + thoff);
354 if (udph->check || skb->ip_summed == CHECKSUM_PARTIAL) {
355 inet_proto_csum_replace2(&udph->check, skb, port,
358 udph->check = CSUM_MANGLED_0;
364 static int nf_flow_nat_port(struct sk_buff *skb, unsigned int thoff,
365 u8 protocol, __be16 port, __be16 new_port)
369 if (nf_flow_nat_port_tcp(skb, thoff, port, new_port) < 0)
373 if (nf_flow_nat_port_udp(skb, thoff, port, new_port) < 0)
381 int nf_flow_snat_port(const struct flow_offload *flow,
382 struct sk_buff *skb, unsigned int thoff,
383 u8 protocol, enum flow_offload_tuple_dir dir)
385 struct flow_ports *hdr;
386 __be16 port, new_port;
388 if (!pskb_may_pull(skb, thoff + sizeof(*hdr)) ||
389 skb_try_make_writable(skb, thoff + sizeof(*hdr)))
392 hdr = (void *)(skb_network_header(skb) + thoff);
395 case FLOW_OFFLOAD_DIR_ORIGINAL:
397 new_port = flow->tuplehash[FLOW_OFFLOAD_DIR_REPLY].tuple.dst_port;
398 hdr->source = new_port;
400 case FLOW_OFFLOAD_DIR_REPLY:
402 new_port = flow->tuplehash[FLOW_OFFLOAD_DIR_ORIGINAL].tuple.src_port;
403 hdr->dest = new_port;
409 return nf_flow_nat_port(skb, thoff, protocol, port, new_port);
411 EXPORT_SYMBOL_GPL(nf_flow_snat_port);
413 int nf_flow_dnat_port(const struct flow_offload *flow,
414 struct sk_buff *skb, unsigned int thoff,
415 u8 protocol, enum flow_offload_tuple_dir dir)
417 struct flow_ports *hdr;
418 __be16 port, new_port;
420 if (!pskb_may_pull(skb, thoff + sizeof(*hdr)) ||
421 skb_try_make_writable(skb, thoff + sizeof(*hdr)))
424 hdr = (void *)(skb_network_header(skb) + thoff);
427 case FLOW_OFFLOAD_DIR_ORIGINAL:
429 new_port = flow->tuplehash[FLOW_OFFLOAD_DIR_REPLY].tuple.src_port;
430 hdr->dest = new_port;
432 case FLOW_OFFLOAD_DIR_REPLY:
434 new_port = flow->tuplehash[FLOW_OFFLOAD_DIR_ORIGINAL].tuple.dst_port;
435 hdr->source = new_port;
441 return nf_flow_nat_port(skb, thoff, protocol, port, new_port);
443 EXPORT_SYMBOL_GPL(nf_flow_dnat_port);
445 int nf_flow_table_init(struct nf_flowtable *flowtable)
449 INIT_DEFERRABLE_WORK(&flowtable->gc_work, nf_flow_offload_work_gc);
451 err = rhashtable_init(&flowtable->rhashtable,
452 &nf_flow_offload_rhash_params);
456 queue_delayed_work(system_power_efficient_wq,
457 &flowtable->gc_work, HZ);
459 mutex_lock(&flowtable_lock);
460 list_add(&flowtable->list, &flowtables);
461 mutex_unlock(&flowtable_lock);
465 EXPORT_SYMBOL_GPL(nf_flow_table_init);
467 static void nf_flow_table_do_cleanup(struct flow_offload *flow, void *data)
469 struct net_device *dev = data;
470 struct flow_offload_entry *e;
472 e = container_of(flow, struct flow_offload_entry, flow);
475 flow_offload_teardown(flow);
478 if (net_eq(nf_ct_net(e->ct), dev_net(dev)) &&
479 (flow->tuplehash[0].tuple.iifidx == dev->ifindex ||
480 flow->tuplehash[1].tuple.iifidx == dev->ifindex))
481 flow_offload_dead(flow);
484 static void nf_flow_table_iterate_cleanup(struct nf_flowtable *flowtable,
485 struct net_device *dev)
487 nf_flow_table_iterate(flowtable, nf_flow_table_do_cleanup, dev);
488 flush_delayed_work(&flowtable->gc_work);
491 void nf_flow_table_cleanup(struct net_device *dev)
493 struct nf_flowtable *flowtable;
495 mutex_lock(&flowtable_lock);
496 list_for_each_entry(flowtable, &flowtables, list)
497 nf_flow_table_iterate_cleanup(flowtable, dev);
498 mutex_unlock(&flowtable_lock);
500 EXPORT_SYMBOL_GPL(nf_flow_table_cleanup);
502 void nf_flow_table_free(struct nf_flowtable *flow_table)
504 mutex_lock(&flowtable_lock);
505 list_del(&flow_table->list);
506 mutex_unlock(&flowtable_lock);
507 cancel_delayed_work_sync(&flow_table->gc_work);
508 nf_flow_table_iterate(flow_table, nf_flow_table_do_cleanup, NULL);
509 nf_flow_table_iterate(flow_table, nf_flow_offload_gc_step, flow_table);
510 rhashtable_destroy(&flow_table->rhashtable);
512 EXPORT_SYMBOL_GPL(nf_flow_table_free);
514 MODULE_LICENSE("GPL");
515 MODULE_AUTHOR("Pablo Neira Ayuso <pablo@netfilter.org>");