1 // SPDX-License-Identifier: GPL-2.0-only
2 #include <linux/kernel.h>
3 #include <linux/module.h>
4 #include <linux/init.h>
5 #include <linux/netlink.h>
6 #include <linux/netfilter.h>
7 #include <linux/workqueue.h>
8 #include <linux/spinlock.h>
9 #include <linux/netfilter/nf_tables.h>
10 #include <net/ip.h> /* for ipv4 options. */
11 #include <net/netfilter/nf_tables.h>
12 #include <net/netfilter/nf_tables_core.h>
13 #include <net/netfilter/nf_conntrack_core.h>
14 #include <linux/netfilter/nf_conntrack_common.h>
15 #include <net/netfilter/nf_flow_table.h>
17 struct nft_flow_offload {
18 struct nft_flowtable *flowtable;
21 static int nft_flow_route(const struct nft_pktinfo *pkt,
22 const struct nf_conn *ct,
23 struct nf_flow_route *route,
24 enum ip_conntrack_dir dir)
26 struct dst_entry *this_dst = skb_dst(pkt->skb);
27 struct dst_entry *other_dst = NULL;
30 memset(&fl, 0, sizeof(fl));
31 switch (nft_pf(pkt)) {
33 fl.u.ip4.daddr = ct->tuplehash[dir].tuple.src.u3.ip;
34 fl.u.ip4.flowi4_oif = nft_in(pkt)->ifindex;
37 fl.u.ip6.daddr = ct->tuplehash[dir].tuple.src.u3.in6;
38 fl.u.ip6.flowi6_oif = nft_in(pkt)->ifindex;
42 nf_route(nft_net(pkt), &other_dst, &fl, false, nft_pf(pkt));
46 route->tuple[dir].dst = this_dst;
47 route->tuple[!dir].dst = other_dst;
52 static bool nft_flow_offload_skip(struct sk_buff *skb, int family)
54 if (skb_sec_path(skb))
57 if (family == NFPROTO_IPV4) {
58 const struct ip_options *opt;
60 opt = &(IPCB(skb)->opt);
62 if (unlikely(opt->optlen))
69 static void nft_flow_offload_eval(const struct nft_expr *expr,
70 struct nft_regs *regs,
71 const struct nft_pktinfo *pkt)
73 struct nft_flow_offload *priv = nft_expr_priv(expr);
74 struct nf_flowtable *flowtable = &priv->flowtable->data;
75 struct tcphdr _tcph, *tcph = NULL;
76 enum ip_conntrack_info ctinfo;
77 struct nf_flow_route route;
78 struct flow_offload *flow;
79 enum ip_conntrack_dir dir;
83 if (nft_flow_offload_skip(pkt->skb, nft_pf(pkt)))
86 ct = nf_ct_get(pkt->skb, &ctinfo);
90 switch (ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple.dst.protonum) {
92 tcph = skb_header_pointer(pkt->skb, pkt->xt.thoff,
93 sizeof(_tcph), &_tcph);
94 if (unlikely(!tcph || tcph->fin || tcph->rst))
103 if (nf_ct_ext_exist(ct, NF_CT_EXT_HELPER) ||
104 ct->status & IPS_SEQ_ADJUST)
107 if (!nf_ct_is_confirmed(ct))
110 if (test_and_set_bit(IPS_OFFLOAD_BIT, &ct->status))
113 dir = CTINFO2DIR(ctinfo);
114 if (nft_flow_route(pkt, ct, &route, dir) < 0)
117 flow = flow_offload_alloc(ct, &route);
122 ct->proto.tcp.seen[0].flags |= IP_CT_TCP_FLAG_BE_LIBERAL;
123 ct->proto.tcp.seen[1].flags |= IP_CT_TCP_FLAG_BE_LIBERAL;
126 ret = flow_offload_add(flowtable, flow);
130 dst_release(route.tuple[!dir].dst);
134 flow_offload_free(flow);
136 dst_release(route.tuple[!dir].dst);
138 clear_bit(IPS_OFFLOAD_BIT, &ct->status);
140 regs->verdict.code = NFT_BREAK;
143 static int nft_flow_offload_validate(const struct nft_ctx *ctx,
144 const struct nft_expr *expr,
145 const struct nft_data **data)
147 unsigned int hook_mask = (1 << NF_INET_FORWARD);
149 return nft_chain_validate_hooks(ctx->chain, hook_mask);
152 static const struct nla_policy nft_flow_offload_policy[NFTA_FLOW_MAX + 1] = {
153 [NFTA_FLOW_TABLE_NAME] = { .type = NLA_STRING,
154 .len = NFT_NAME_MAXLEN - 1 },
157 static int nft_flow_offload_init(const struct nft_ctx *ctx,
158 const struct nft_expr *expr,
159 const struct nlattr * const tb[])
161 struct nft_flow_offload *priv = nft_expr_priv(expr);
162 u8 genmask = nft_genmask_next(ctx->net);
163 struct nft_flowtable *flowtable;
165 if (!tb[NFTA_FLOW_TABLE_NAME])
168 flowtable = nft_flowtable_lookup(ctx->table, tb[NFTA_FLOW_TABLE_NAME],
170 if (IS_ERR(flowtable))
171 return PTR_ERR(flowtable);
173 priv->flowtable = flowtable;
176 return nf_ct_netns_get(ctx->net, ctx->family);
179 static void nft_flow_offload_destroy(const struct nft_ctx *ctx,
180 const struct nft_expr *expr)
182 struct nft_flow_offload *priv = nft_expr_priv(expr);
184 priv->flowtable->use--;
185 nf_ct_netns_put(ctx->net, ctx->family);
188 static int nft_flow_offload_dump(struct sk_buff *skb, const struct nft_expr *expr)
190 struct nft_flow_offload *priv = nft_expr_priv(expr);
192 if (nla_put_string(skb, NFTA_FLOW_TABLE_NAME, priv->flowtable->name))
193 goto nla_put_failure;
201 static struct nft_expr_type nft_flow_offload_type;
202 static const struct nft_expr_ops nft_flow_offload_ops = {
203 .type = &nft_flow_offload_type,
204 .size = NFT_EXPR_SIZE(sizeof(struct nft_flow_offload)),
205 .eval = nft_flow_offload_eval,
206 .init = nft_flow_offload_init,
207 .destroy = nft_flow_offload_destroy,
208 .validate = nft_flow_offload_validate,
209 .dump = nft_flow_offload_dump,
212 static struct nft_expr_type nft_flow_offload_type __read_mostly = {
213 .name = "flow_offload",
214 .ops = &nft_flow_offload_ops,
215 .policy = nft_flow_offload_policy,
216 .maxattr = NFTA_FLOW_MAX,
217 .owner = THIS_MODULE,
220 static int flow_offload_netdev_event(struct notifier_block *this,
221 unsigned long event, void *ptr)
223 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
225 if (event != NETDEV_DOWN)
228 nf_flow_table_cleanup(dev);
233 static struct notifier_block flow_offload_netdev_notifier = {
234 .notifier_call = flow_offload_netdev_event,
237 static int __init nft_flow_offload_module_init(void)
241 err = register_netdevice_notifier(&flow_offload_netdev_notifier);
245 err = nft_register_expr(&nft_flow_offload_type);
252 unregister_netdevice_notifier(&flow_offload_netdev_notifier);
257 static void __exit nft_flow_offload_module_exit(void)
259 nft_unregister_expr(&nft_flow_offload_type);
260 unregister_netdevice_notifier(&flow_offload_netdev_notifier);
263 module_init(nft_flow_offload_module_init);
264 module_exit(nft_flow_offload_module_exit);
266 MODULE_LICENSE("GPL");
267 MODULE_AUTHOR("Pablo Neira Ayuso <pablo@netfilter.org>");
268 MODULE_ALIAS_NFT_EXPR("flow_offload");