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_conntrack_common.h>
10 #include <linux/netfilter/nf_tables.h>
11 #include <net/ip.h> /* for ipv4 options. */
12 #include <net/netfilter/nf_tables.h>
13 #include <net/netfilter/nf_tables_core.h>
14 #include <net/netfilter/nf_conntrack_core.h>
15 #include <net/netfilter/nf_conntrack_extend.h>
16 #include <net/netfilter/nf_flow_table.h>
17
18 struct nft_flow_offload {
19 struct nft_flowtable *flowtable;
20 };
21
nft_flow_route(const struct nft_pktinfo * pkt,const struct nf_conn * ct,struct nf_flow_route * route,enum ip_conntrack_dir dir)22 static int nft_flow_route(const struct nft_pktinfo *pkt,
23 const struct nf_conn *ct,
24 struct nf_flow_route *route,
25 enum ip_conntrack_dir dir)
26 {
27 struct dst_entry *this_dst = skb_dst(pkt->skb);
28 struct dst_entry *other_dst = NULL;
29 struct flowi fl;
30
31 memset(&fl, 0, sizeof(fl));
32 switch (nft_pf(pkt)) {
33 case NFPROTO_IPV4:
34 fl.u.ip4.daddr = ct->tuplehash[dir].tuple.src.u3.ip;
35 fl.u.ip4.flowi4_oif = nft_in(pkt)->ifindex;
36 break;
37 case NFPROTO_IPV6:
38 fl.u.ip6.daddr = ct->tuplehash[dir].tuple.src.u3.in6;
39 fl.u.ip6.flowi6_oif = nft_in(pkt)->ifindex;
40 break;
41 }
42
43 nf_route(nft_net(pkt), &other_dst, &fl, false, nft_pf(pkt));
44 if (!other_dst)
45 return -ENOENT;
46
47 route->tuple[dir].dst = this_dst;
48 route->tuple[!dir].dst = other_dst;
49
50 return 0;
51 }
52
nft_flow_offload_skip(struct sk_buff * skb,int family)53 static bool nft_flow_offload_skip(struct sk_buff *skb, int family)
54 {
55 if (skb_sec_path(skb))
56 return true;
57
58 if (family == NFPROTO_IPV4) {
59 const struct ip_options *opt;
60
61 opt = &(IPCB(skb)->opt);
62
63 if (unlikely(opt->optlen))
64 return true;
65 }
66
67 return false;
68 }
69
nft_flow_offload_eval(const struct nft_expr * expr,struct nft_regs * regs,const struct nft_pktinfo * pkt)70 static void nft_flow_offload_eval(const struct nft_expr *expr,
71 struct nft_regs *regs,
72 const struct nft_pktinfo *pkt)
73 {
74 struct nft_flow_offload *priv = nft_expr_priv(expr);
75 struct nf_flowtable *flowtable = &priv->flowtable->data;
76 struct tcphdr _tcph, *tcph = NULL;
77 enum ip_conntrack_info ctinfo;
78 struct nf_flow_route route;
79 struct flow_offload *flow;
80 enum ip_conntrack_dir dir;
81 struct nf_conn *ct;
82 int ret;
83
84 if (nft_flow_offload_skip(pkt->skb, nft_pf(pkt)))
85 goto out;
86
87 ct = nf_ct_get(pkt->skb, &ctinfo);
88 if (!ct)
89 goto out;
90
91 switch (ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple.dst.protonum) {
92 case IPPROTO_TCP:
93 tcph = skb_header_pointer(pkt->skb, pkt->xt.thoff,
94 sizeof(_tcph), &_tcph);
95 if (unlikely(!tcph || tcph->fin || tcph->rst))
96 goto out;
97 break;
98 case IPPROTO_UDP:
99 break;
100 default:
101 goto out;
102 }
103
104 if (nf_ct_ext_exist(ct, NF_CT_EXT_HELPER) ||
105 ct->status & IPS_SEQ_ADJUST)
106 goto out;
107
108 if (!nf_ct_is_confirmed(ct))
109 goto out;
110
111 if (test_and_set_bit(IPS_OFFLOAD_BIT, &ct->status))
112 goto out;
113
114 dir = CTINFO2DIR(ctinfo);
115 if (nft_flow_route(pkt, ct, &route, dir) < 0)
116 goto err_flow_route;
117
118 flow = flow_offload_alloc(ct, &route);
119 if (!flow)
120 goto err_flow_alloc;
121
122 if (tcph) {
123 ct->proto.tcp.seen[0].flags |= IP_CT_TCP_FLAG_BE_LIBERAL;
124 ct->proto.tcp.seen[1].flags |= IP_CT_TCP_FLAG_BE_LIBERAL;
125 }
126
127 ret = flow_offload_add(flowtable, flow);
128 if (ret < 0)
129 goto err_flow_add;
130
131 dst_release(route.tuple[!dir].dst);
132 return;
133
134 err_flow_add:
135 flow_offload_free(flow);
136 err_flow_alloc:
137 dst_release(route.tuple[!dir].dst);
138 err_flow_route:
139 clear_bit(IPS_OFFLOAD_BIT, &ct->status);
140 out:
141 regs->verdict.code = NFT_BREAK;
142 }
143
nft_flow_offload_validate(const struct nft_ctx * ctx,const struct nft_expr * expr,const struct nft_data ** data)144 static int nft_flow_offload_validate(const struct nft_ctx *ctx,
145 const struct nft_expr *expr,
146 const struct nft_data **data)
147 {
148 unsigned int hook_mask = (1 << NF_INET_FORWARD);
149
150 return nft_chain_validate_hooks(ctx->chain, hook_mask);
151 }
152
153 static const struct nla_policy nft_flow_offload_policy[NFTA_FLOW_MAX + 1] = {
154 [NFTA_FLOW_TABLE_NAME] = { .type = NLA_STRING,
155 .len = NFT_NAME_MAXLEN - 1 },
156 };
157
nft_flow_offload_init(const struct nft_ctx * ctx,const struct nft_expr * expr,const struct nlattr * const tb[])158 static int nft_flow_offload_init(const struct nft_ctx *ctx,
159 const struct nft_expr *expr,
160 const struct nlattr * const tb[])
161 {
162 struct nft_flow_offload *priv = nft_expr_priv(expr);
163 u8 genmask = nft_genmask_next(ctx->net);
164 struct nft_flowtable *flowtable;
165
166 if (!tb[NFTA_FLOW_TABLE_NAME])
167 return -EINVAL;
168
169 flowtable = nft_flowtable_lookup(ctx->table, tb[NFTA_FLOW_TABLE_NAME],
170 genmask);
171 if (IS_ERR(flowtable))
172 return PTR_ERR(flowtable);
173
174 priv->flowtable = flowtable;
175 flowtable->use++;
176
177 return nf_ct_netns_get(ctx->net, ctx->family);
178 }
179
nft_flow_offload_deactivate(const struct nft_ctx * ctx,const struct nft_expr * expr,enum nft_trans_phase phase)180 static void nft_flow_offload_deactivate(const struct nft_ctx *ctx,
181 const struct nft_expr *expr,
182 enum nft_trans_phase phase)
183 {
184 struct nft_flow_offload *priv = nft_expr_priv(expr);
185
186 nf_tables_deactivate_flowtable(ctx, priv->flowtable, phase);
187 }
188
nft_flow_offload_activate(const struct nft_ctx * ctx,const struct nft_expr * expr)189 static void nft_flow_offload_activate(const struct nft_ctx *ctx,
190 const struct nft_expr *expr)
191 {
192 struct nft_flow_offload *priv = nft_expr_priv(expr);
193
194 priv->flowtable->use++;
195 }
196
nft_flow_offload_destroy(const struct nft_ctx * ctx,const struct nft_expr * expr)197 static void nft_flow_offload_destroy(const struct nft_ctx *ctx,
198 const struct nft_expr *expr)
199 {
200 nf_ct_netns_put(ctx->net, ctx->family);
201 }
202
nft_flow_offload_dump(struct sk_buff * skb,const struct nft_expr * expr)203 static int nft_flow_offload_dump(struct sk_buff *skb, const struct nft_expr *expr)
204 {
205 struct nft_flow_offload *priv = nft_expr_priv(expr);
206
207 if (nla_put_string(skb, NFTA_FLOW_TABLE_NAME, priv->flowtable->name))
208 goto nla_put_failure;
209
210 return 0;
211
212 nla_put_failure:
213 return -1;
214 }
215
216 static struct nft_expr_type nft_flow_offload_type;
217 static const struct nft_expr_ops nft_flow_offload_ops = {
218 .type = &nft_flow_offload_type,
219 .size = NFT_EXPR_SIZE(sizeof(struct nft_flow_offload)),
220 .eval = nft_flow_offload_eval,
221 .init = nft_flow_offload_init,
222 .activate = nft_flow_offload_activate,
223 .deactivate = nft_flow_offload_deactivate,
224 .destroy = nft_flow_offload_destroy,
225 .validate = nft_flow_offload_validate,
226 .dump = nft_flow_offload_dump,
227 };
228
229 static struct nft_expr_type nft_flow_offload_type __read_mostly = {
230 .name = "flow_offload",
231 .ops = &nft_flow_offload_ops,
232 .policy = nft_flow_offload_policy,
233 .maxattr = NFTA_FLOW_MAX,
234 .owner = THIS_MODULE,
235 };
236
flow_offload_netdev_event(struct notifier_block * this,unsigned long event,void * ptr)237 static int flow_offload_netdev_event(struct notifier_block *this,
238 unsigned long event, void *ptr)
239 {
240 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
241
242 if (event != NETDEV_DOWN)
243 return NOTIFY_DONE;
244
245 nf_flow_table_cleanup(dev);
246
247 return NOTIFY_DONE;
248 }
249
250 static struct notifier_block flow_offload_netdev_notifier = {
251 .notifier_call = flow_offload_netdev_event,
252 };
253
nft_flow_offload_module_init(void)254 static int __init nft_flow_offload_module_init(void)
255 {
256 int err;
257
258 err = register_netdevice_notifier(&flow_offload_netdev_notifier);
259 if (err)
260 goto err;
261
262 err = nft_register_expr(&nft_flow_offload_type);
263 if (err < 0)
264 goto register_expr;
265
266 return 0;
267
268 register_expr:
269 unregister_netdevice_notifier(&flow_offload_netdev_notifier);
270 err:
271 return err;
272 }
273
nft_flow_offload_module_exit(void)274 static void __exit nft_flow_offload_module_exit(void)
275 {
276 nft_unregister_expr(&nft_flow_offload_type);
277 unregister_netdevice_notifier(&flow_offload_netdev_notifier);
278 }
279
280 module_init(nft_flow_offload_module_init);
281 module_exit(nft_flow_offload_module_exit);
282
283 MODULE_LICENSE("GPL");
284 MODULE_AUTHOR("Pablo Neira Ayuso <pablo@netfilter.org>");
285 MODULE_ALIAS_NFT_EXPR("flow_offload");
286