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>
18 struct nft_flow_offload {
19 struct nft_flowtable *flowtable;
22 static enum flow_offload_xmit_type nft_xmit_type(struct dst_entry *dst)
25 return FLOW_OFFLOAD_XMIT_XFRM;
27 return FLOW_OFFLOAD_XMIT_NEIGH;
30 static void nft_default_forward_path(struct nf_flow_route *route,
31 struct dst_entry *dst_cache,
32 enum ip_conntrack_dir dir)
34 route->tuple[!dir].in.ifindex = dst_cache->dev->ifindex;
35 route->tuple[dir].dst = dst_cache;
36 route->tuple[dir].xmit_type = nft_xmit_type(dst_cache);
39 static int nft_dev_fill_forward_path(const struct nf_flow_route *route,
40 const struct dst_entry *dst_cache,
41 const struct nf_conn *ct,
42 enum ip_conntrack_dir dir, u8 *ha,
43 struct net_device_path_stack *stack)
45 const void *daddr = &ct->tuplehash[!dir].tuple.src.u3;
46 struct net_device *dev = dst_cache->dev;
50 n = dst_neigh_lookup(dst_cache, daddr);
54 read_lock_bh(&n->lock);
55 nud_state = n->nud_state;
56 ether_addr_copy(ha, n->ha);
57 read_unlock_bh(&n->lock);
60 if (!(nud_state & NUD_VALID))
63 return dev_fill_forward_path(dev, ha, stack);
66 struct nft_forward_info {
67 const struct net_device *indev;
68 const struct net_device *outdev;
72 } encap[NF_FLOW_TABLE_ENCAP_MAX];
74 u8 h_source[ETH_ALEN];
76 enum flow_offload_xmit_type xmit_type;
79 static void nft_dev_path_info(const struct net_device_path_stack *stack,
80 struct nft_forward_info *info,
83 const struct net_device_path *path;
86 memcpy(info->h_dest, ha, ETH_ALEN);
88 for (i = 0; i < stack->num_paths; i++) {
89 path = &stack->path[i];
91 case DEV_PATH_ETHERNET:
93 info->indev = path->dev;
94 if (is_zero_ether_addr(info->h_source))
95 memcpy(info->h_source, path->dev->dev_addr, ETH_ALEN);
97 if (path->type == DEV_PATH_ETHERNET)
101 if (info->num_encaps >= NF_FLOW_TABLE_ENCAP_MAX) {
105 info->outdev = path->dev;
106 info->encap[info->num_encaps].id = path->encap.id;
107 info->encap[info->num_encaps].proto = path->encap.proto;
110 case DEV_PATH_BRIDGE:
111 if (is_zero_ether_addr(info->h_source))
112 memcpy(info->h_source, path->dev->dev_addr, ETH_ALEN);
114 info->xmit_type = FLOW_OFFLOAD_XMIT_DIRECT;
122 info->outdev = info->indev;
125 static bool nft_flowtable_find_dev(const struct net_device *dev,
126 struct nft_flowtable *ft)
128 struct nft_hook *hook;
131 list_for_each_entry_rcu(hook, &ft->hook_list, list) {
132 if (hook->ops.dev != dev)
142 static void nft_dev_forward_path(struct nf_flow_route *route,
143 const struct nf_conn *ct,
144 enum ip_conntrack_dir dir,
145 struct nft_flowtable *ft)
147 const struct dst_entry *dst = route->tuple[dir].dst;
148 struct net_device_path_stack stack;
149 struct nft_forward_info info = {};
150 unsigned char ha[ETH_ALEN];
153 if (nft_dev_fill_forward_path(route, dst, ct, dir, ha, &stack) >= 0)
154 nft_dev_path_info(&stack, &info, ha);
156 if (!info.indev || !nft_flowtable_find_dev(info.indev, ft))
159 route->tuple[!dir].in.ifindex = info.indev->ifindex;
160 for (i = 0; i < info.num_encaps; i++) {
161 route->tuple[!dir].in.encap[i].id = info.encap[i].id;
162 route->tuple[!dir].in.encap[i].proto = info.encap[i].proto;
164 route->tuple[!dir].in.num_encaps = info.num_encaps;
166 if (info.xmit_type == FLOW_OFFLOAD_XMIT_DIRECT) {
167 memcpy(route->tuple[dir].out.h_source, info.h_source, ETH_ALEN);
168 memcpy(route->tuple[dir].out.h_dest, info.h_dest, ETH_ALEN);
169 route->tuple[dir].out.ifindex = info.outdev->ifindex;
170 route->tuple[dir].xmit_type = info.xmit_type;
174 static int nft_flow_route(const struct nft_pktinfo *pkt,
175 const struct nf_conn *ct,
176 struct nf_flow_route *route,
177 enum ip_conntrack_dir dir,
178 struct nft_flowtable *ft)
180 struct dst_entry *this_dst = skb_dst(pkt->skb);
181 struct dst_entry *other_dst = NULL;
184 memset(&fl, 0, sizeof(fl));
185 switch (nft_pf(pkt)) {
187 fl.u.ip4.daddr = ct->tuplehash[dir].tuple.src.u3.ip;
188 fl.u.ip4.flowi4_oif = nft_in(pkt)->ifindex;
191 fl.u.ip6.daddr = ct->tuplehash[dir].tuple.src.u3.in6;
192 fl.u.ip6.flowi6_oif = nft_in(pkt)->ifindex;
196 nf_route(nft_net(pkt), &other_dst, &fl, false, nft_pf(pkt));
200 nft_default_forward_path(route, this_dst, dir);
201 nft_default_forward_path(route, other_dst, !dir);
203 if (route->tuple[dir].xmit_type == FLOW_OFFLOAD_XMIT_NEIGH &&
204 route->tuple[!dir].xmit_type == FLOW_OFFLOAD_XMIT_NEIGH) {
205 nft_dev_forward_path(route, ct, dir, ft);
206 nft_dev_forward_path(route, ct, !dir, ft);
212 static bool nft_flow_offload_skip(struct sk_buff *skb, int family)
214 if (skb_sec_path(skb))
217 if (family == NFPROTO_IPV4) {
218 const struct ip_options *opt;
220 opt = &(IPCB(skb)->opt);
222 if (unlikely(opt->optlen))
229 static void nft_flow_offload_eval(const struct nft_expr *expr,
230 struct nft_regs *regs,
231 const struct nft_pktinfo *pkt)
233 struct nft_flow_offload *priv = nft_expr_priv(expr);
234 struct nf_flowtable *flowtable = &priv->flowtable->data;
235 struct tcphdr _tcph, *tcph = NULL;
236 struct nf_flow_route route = {};
237 enum ip_conntrack_info ctinfo;
238 struct flow_offload *flow;
239 enum ip_conntrack_dir dir;
243 if (nft_flow_offload_skip(pkt->skb, nft_pf(pkt)))
246 ct = nf_ct_get(pkt->skb, &ctinfo);
250 switch (ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple.dst.protonum) {
252 tcph = skb_header_pointer(pkt->skb, pkt->xt.thoff,
253 sizeof(_tcph), &_tcph);
254 if (unlikely(!tcph || tcph->fin || tcph->rst))
263 if (nf_ct_ext_exist(ct, NF_CT_EXT_HELPER) ||
264 ct->status & (IPS_SEQ_ADJUST | IPS_NAT_CLASH))
267 if (!nf_ct_is_confirmed(ct))
270 if (test_and_set_bit(IPS_OFFLOAD_BIT, &ct->status))
273 dir = CTINFO2DIR(ctinfo);
274 if (nft_flow_route(pkt, ct, &route, dir, priv->flowtable) < 0)
277 flow = flow_offload_alloc(ct);
281 if (flow_offload_route_init(flow, &route) < 0)
285 ct->proto.tcp.seen[0].flags |= IP_CT_TCP_FLAG_BE_LIBERAL;
286 ct->proto.tcp.seen[1].flags |= IP_CT_TCP_FLAG_BE_LIBERAL;
289 ret = flow_offload_add(flowtable, flow);
293 dst_release(route.tuple[!dir].dst);
297 flow_offload_free(flow);
299 dst_release(route.tuple[!dir].dst);
301 clear_bit(IPS_OFFLOAD_BIT, &ct->status);
303 regs->verdict.code = NFT_BREAK;
306 static int nft_flow_offload_validate(const struct nft_ctx *ctx,
307 const struct nft_expr *expr,
308 const struct nft_data **data)
310 unsigned int hook_mask = (1 << NF_INET_FORWARD);
312 return nft_chain_validate_hooks(ctx->chain, hook_mask);
315 static const struct nla_policy nft_flow_offload_policy[NFTA_FLOW_MAX + 1] = {
316 [NFTA_FLOW_TABLE_NAME] = { .type = NLA_STRING,
317 .len = NFT_NAME_MAXLEN - 1 },
320 static int nft_flow_offload_init(const struct nft_ctx *ctx,
321 const struct nft_expr *expr,
322 const struct nlattr * const tb[])
324 struct nft_flow_offload *priv = nft_expr_priv(expr);
325 u8 genmask = nft_genmask_next(ctx->net);
326 struct nft_flowtable *flowtable;
328 if (!tb[NFTA_FLOW_TABLE_NAME])
331 flowtable = nft_flowtable_lookup(ctx->table, tb[NFTA_FLOW_TABLE_NAME],
333 if (IS_ERR(flowtable))
334 return PTR_ERR(flowtable);
336 priv->flowtable = flowtable;
339 return nf_ct_netns_get(ctx->net, ctx->family);
342 static void nft_flow_offload_deactivate(const struct nft_ctx *ctx,
343 const struct nft_expr *expr,
344 enum nft_trans_phase phase)
346 struct nft_flow_offload *priv = nft_expr_priv(expr);
348 nf_tables_deactivate_flowtable(ctx, priv->flowtable, phase);
351 static void nft_flow_offload_activate(const struct nft_ctx *ctx,
352 const struct nft_expr *expr)
354 struct nft_flow_offload *priv = nft_expr_priv(expr);
356 priv->flowtable->use++;
359 static void nft_flow_offload_destroy(const struct nft_ctx *ctx,
360 const struct nft_expr *expr)
362 nf_ct_netns_put(ctx->net, ctx->family);
365 static int nft_flow_offload_dump(struct sk_buff *skb, const struct nft_expr *expr)
367 struct nft_flow_offload *priv = nft_expr_priv(expr);
369 if (nla_put_string(skb, NFTA_FLOW_TABLE_NAME, priv->flowtable->name))
370 goto nla_put_failure;
378 static struct nft_expr_type nft_flow_offload_type;
379 static const struct nft_expr_ops nft_flow_offload_ops = {
380 .type = &nft_flow_offload_type,
381 .size = NFT_EXPR_SIZE(sizeof(struct nft_flow_offload)),
382 .eval = nft_flow_offload_eval,
383 .init = nft_flow_offload_init,
384 .activate = nft_flow_offload_activate,
385 .deactivate = nft_flow_offload_deactivate,
386 .destroy = nft_flow_offload_destroy,
387 .validate = nft_flow_offload_validate,
388 .dump = nft_flow_offload_dump,
391 static struct nft_expr_type nft_flow_offload_type __read_mostly = {
392 .name = "flow_offload",
393 .ops = &nft_flow_offload_ops,
394 .policy = nft_flow_offload_policy,
395 .maxattr = NFTA_FLOW_MAX,
396 .owner = THIS_MODULE,
399 static int flow_offload_netdev_event(struct notifier_block *this,
400 unsigned long event, void *ptr)
402 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
404 if (event != NETDEV_DOWN)
407 nf_flow_table_cleanup(dev);
412 static struct notifier_block flow_offload_netdev_notifier = {
413 .notifier_call = flow_offload_netdev_event,
416 static int __init nft_flow_offload_module_init(void)
420 err = register_netdevice_notifier(&flow_offload_netdev_notifier);
424 err = nft_register_expr(&nft_flow_offload_type);
431 unregister_netdevice_notifier(&flow_offload_netdev_notifier);
436 static void __exit nft_flow_offload_module_exit(void)
438 nft_unregister_expr(&nft_flow_offload_type);
439 unregister_netdevice_notifier(&flow_offload_netdev_notifier);
442 module_init(nft_flow_offload_module_init);
443 module_exit(nft_flow_offload_module_exit);
445 MODULE_LICENSE("GPL");
446 MODULE_AUTHOR("Pablo Neira Ayuso <pablo@netfilter.org>");
447 MODULE_ALIAS_NFT_EXPR("flow_offload");
448 MODULE_DESCRIPTION("nftables hardware flow offload module");