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;
69 const struct net_device *hw_outdev;
73 } encap[NF_FLOW_TABLE_ENCAP_MAX];
75 u8 h_source[ETH_ALEN];
77 enum flow_offload_xmit_type xmit_type;
80 static bool nft_is_valid_ether_device(const struct net_device *dev)
82 if (!dev || (dev->flags & IFF_LOOPBACK) || dev->type != ARPHRD_ETHER ||
83 dev->addr_len != ETH_ALEN || !is_valid_ether_addr(dev->dev_addr))
89 static void nft_dev_path_info(const struct net_device_path_stack *stack,
90 struct nft_forward_info *info,
91 unsigned char *ha, struct nf_flowtable *flowtable)
93 const struct net_device_path *path;
96 memcpy(info->h_dest, ha, ETH_ALEN);
98 for (i = 0; i < stack->num_paths; i++) {
99 path = &stack->path[i];
100 switch (path->type) {
101 case DEV_PATH_ETHERNET:
105 info->indev = path->dev;
106 if (is_zero_ether_addr(info->h_source))
107 memcpy(info->h_source, path->dev->dev_addr, ETH_ALEN);
109 if (path->type == DEV_PATH_ETHERNET)
111 if (path->type == DEV_PATH_DSA) {
112 i = stack->num_paths;
116 /* DEV_PATH_VLAN and DEV_PATH_PPPOE */
117 if (info->num_encaps >= NF_FLOW_TABLE_ENCAP_MAX) {
121 info->outdev = path->dev;
122 info->encap[info->num_encaps].id = path->encap.id;
123 info->encap[info->num_encaps].proto = path->encap.proto;
125 if (path->type == DEV_PATH_PPPOE)
126 memcpy(info->h_dest, path->encap.h_dest, ETH_ALEN);
128 case DEV_PATH_BRIDGE:
129 if (is_zero_ether_addr(info->h_source))
130 memcpy(info->h_source, path->dev->dev_addr, ETH_ALEN);
132 switch (path->bridge.vlan_mode) {
133 case DEV_PATH_BR_VLAN_TAG:
134 info->encap[info->num_encaps].id = path->bridge.vlan_id;
135 info->encap[info->num_encaps].proto = path->bridge.vlan_proto;
138 case DEV_PATH_BR_VLAN_UNTAG:
141 case DEV_PATH_BR_VLAN_KEEP:
144 info->xmit_type = FLOW_OFFLOAD_XMIT_DIRECT;
152 info->outdev = info->indev;
154 info->hw_outdev = info->indev;
156 if (nf_flowtable_hw_offload(flowtable) &&
157 nft_is_valid_ether_device(info->indev))
158 info->xmit_type = FLOW_OFFLOAD_XMIT_DIRECT;
161 static bool nft_flowtable_find_dev(const struct net_device *dev,
162 struct nft_flowtable *ft)
164 struct nft_hook *hook;
167 list_for_each_entry_rcu(hook, &ft->hook_list, list) {
168 if (hook->ops.dev != dev)
178 static void nft_dev_forward_path(struct nf_flow_route *route,
179 const struct nf_conn *ct,
180 enum ip_conntrack_dir dir,
181 struct nft_flowtable *ft)
183 const struct dst_entry *dst = route->tuple[dir].dst;
184 struct net_device_path_stack stack;
185 struct nft_forward_info info = {};
186 unsigned char ha[ETH_ALEN];
189 if (nft_dev_fill_forward_path(route, dst, ct, dir, ha, &stack) >= 0)
190 nft_dev_path_info(&stack, &info, ha, &ft->data);
192 if (!info.indev || !nft_flowtable_find_dev(info.indev, ft))
195 route->tuple[!dir].in.ifindex = info.indev->ifindex;
196 for (i = 0; i < info.num_encaps; i++) {
197 route->tuple[!dir].in.encap[i].id = info.encap[i].id;
198 route->tuple[!dir].in.encap[i].proto = info.encap[i].proto;
200 route->tuple[!dir].in.num_encaps = info.num_encaps;
202 if (info.xmit_type == FLOW_OFFLOAD_XMIT_DIRECT) {
203 memcpy(route->tuple[dir].out.h_source, info.h_source, ETH_ALEN);
204 memcpy(route->tuple[dir].out.h_dest, info.h_dest, ETH_ALEN);
205 route->tuple[dir].out.ifindex = info.outdev->ifindex;
206 route->tuple[dir].out.hw_ifindex = info.hw_outdev->ifindex;
207 route->tuple[dir].xmit_type = info.xmit_type;
211 static int nft_flow_route(const struct nft_pktinfo *pkt,
212 const struct nf_conn *ct,
213 struct nf_flow_route *route,
214 enum ip_conntrack_dir dir,
215 struct nft_flowtable *ft)
217 struct dst_entry *this_dst = skb_dst(pkt->skb);
218 struct dst_entry *other_dst = NULL;
221 memset(&fl, 0, sizeof(fl));
222 switch (nft_pf(pkt)) {
224 fl.u.ip4.daddr = ct->tuplehash[dir].tuple.src.u3.ip;
225 fl.u.ip4.flowi4_oif = nft_in(pkt)->ifindex;
228 fl.u.ip6.daddr = ct->tuplehash[dir].tuple.src.u3.in6;
229 fl.u.ip6.flowi6_oif = nft_in(pkt)->ifindex;
233 nf_route(nft_net(pkt), &other_dst, &fl, false, nft_pf(pkt));
237 nft_default_forward_path(route, this_dst, dir);
238 nft_default_forward_path(route, other_dst, !dir);
240 if (route->tuple[dir].xmit_type == FLOW_OFFLOAD_XMIT_NEIGH &&
241 route->tuple[!dir].xmit_type == FLOW_OFFLOAD_XMIT_NEIGH) {
242 nft_dev_forward_path(route, ct, dir, ft);
243 nft_dev_forward_path(route, ct, !dir, ft);
249 static bool nft_flow_offload_skip(struct sk_buff *skb, int family)
251 if (skb_sec_path(skb))
254 if (family == NFPROTO_IPV4) {
255 const struct ip_options *opt;
257 opt = &(IPCB(skb)->opt);
259 if (unlikely(opt->optlen))
266 static void nft_flow_offload_eval(const struct nft_expr *expr,
267 struct nft_regs *regs,
268 const struct nft_pktinfo *pkt)
270 struct nft_flow_offload *priv = nft_expr_priv(expr);
271 struct nf_flowtable *flowtable = &priv->flowtable->data;
272 struct tcphdr _tcph, *tcph = NULL;
273 struct nf_flow_route route = {};
274 enum ip_conntrack_info ctinfo;
275 struct flow_offload *flow;
276 enum ip_conntrack_dir dir;
280 if (nft_flow_offload_skip(pkt->skb, nft_pf(pkt)))
283 ct = nf_ct_get(pkt->skb, &ctinfo);
287 switch (ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple.dst.protonum) {
289 tcph = skb_header_pointer(pkt->skb, pkt->xt.thoff,
290 sizeof(_tcph), &_tcph);
291 if (unlikely(!tcph || tcph->fin || tcph->rst))
300 if (nf_ct_ext_exist(ct, NF_CT_EXT_HELPER) ||
301 ct->status & (IPS_SEQ_ADJUST | IPS_NAT_CLASH))
304 if (!nf_ct_is_confirmed(ct))
307 if (test_and_set_bit(IPS_OFFLOAD_BIT, &ct->status))
310 dir = CTINFO2DIR(ctinfo);
311 if (nft_flow_route(pkt, ct, &route, dir, priv->flowtable) < 0)
314 flow = flow_offload_alloc(ct);
318 if (flow_offload_route_init(flow, &route) < 0)
322 ct->proto.tcp.seen[0].flags |= IP_CT_TCP_FLAG_BE_LIBERAL;
323 ct->proto.tcp.seen[1].flags |= IP_CT_TCP_FLAG_BE_LIBERAL;
326 ret = flow_offload_add(flowtable, flow);
330 dst_release(route.tuple[!dir].dst);
334 flow_offload_free(flow);
336 dst_release(route.tuple[!dir].dst);
338 clear_bit(IPS_OFFLOAD_BIT, &ct->status);
340 regs->verdict.code = NFT_BREAK;
343 static int nft_flow_offload_validate(const struct nft_ctx *ctx,
344 const struct nft_expr *expr,
345 const struct nft_data **data)
347 unsigned int hook_mask = (1 << NF_INET_FORWARD);
349 return nft_chain_validate_hooks(ctx->chain, hook_mask);
352 static const struct nla_policy nft_flow_offload_policy[NFTA_FLOW_MAX + 1] = {
353 [NFTA_FLOW_TABLE_NAME] = { .type = NLA_STRING,
354 .len = NFT_NAME_MAXLEN - 1 },
357 static int nft_flow_offload_init(const struct nft_ctx *ctx,
358 const struct nft_expr *expr,
359 const struct nlattr * const tb[])
361 struct nft_flow_offload *priv = nft_expr_priv(expr);
362 u8 genmask = nft_genmask_next(ctx->net);
363 struct nft_flowtable *flowtable;
365 if (!tb[NFTA_FLOW_TABLE_NAME])
368 flowtable = nft_flowtable_lookup(ctx->table, tb[NFTA_FLOW_TABLE_NAME],
370 if (IS_ERR(flowtable))
371 return PTR_ERR(flowtable);
373 priv->flowtable = flowtable;
376 return nf_ct_netns_get(ctx->net, ctx->family);
379 static void nft_flow_offload_deactivate(const struct nft_ctx *ctx,
380 const struct nft_expr *expr,
381 enum nft_trans_phase phase)
383 struct nft_flow_offload *priv = nft_expr_priv(expr);
385 nf_tables_deactivate_flowtable(ctx, priv->flowtable, phase);
388 static void nft_flow_offload_activate(const struct nft_ctx *ctx,
389 const struct nft_expr *expr)
391 struct nft_flow_offload *priv = nft_expr_priv(expr);
393 priv->flowtable->use++;
396 static void nft_flow_offload_destroy(const struct nft_ctx *ctx,
397 const struct nft_expr *expr)
399 nf_ct_netns_put(ctx->net, ctx->family);
402 static int nft_flow_offload_dump(struct sk_buff *skb, const struct nft_expr *expr)
404 struct nft_flow_offload *priv = nft_expr_priv(expr);
406 if (nla_put_string(skb, NFTA_FLOW_TABLE_NAME, priv->flowtable->name))
407 goto nla_put_failure;
415 static struct nft_expr_type nft_flow_offload_type;
416 static const struct nft_expr_ops nft_flow_offload_ops = {
417 .type = &nft_flow_offload_type,
418 .size = NFT_EXPR_SIZE(sizeof(struct nft_flow_offload)),
419 .eval = nft_flow_offload_eval,
420 .init = nft_flow_offload_init,
421 .activate = nft_flow_offload_activate,
422 .deactivate = nft_flow_offload_deactivate,
423 .destroy = nft_flow_offload_destroy,
424 .validate = nft_flow_offload_validate,
425 .dump = nft_flow_offload_dump,
428 static struct nft_expr_type nft_flow_offload_type __read_mostly = {
429 .name = "flow_offload",
430 .ops = &nft_flow_offload_ops,
431 .policy = nft_flow_offload_policy,
432 .maxattr = NFTA_FLOW_MAX,
433 .owner = THIS_MODULE,
436 static int flow_offload_netdev_event(struct notifier_block *this,
437 unsigned long event, void *ptr)
439 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
441 if (event != NETDEV_DOWN)
444 nf_flow_table_cleanup(dev);
449 static struct notifier_block flow_offload_netdev_notifier = {
450 .notifier_call = flow_offload_netdev_event,
453 static int __init nft_flow_offload_module_init(void)
457 err = register_netdevice_notifier(&flow_offload_netdev_notifier);
461 err = nft_register_expr(&nft_flow_offload_type);
468 unregister_netdevice_notifier(&flow_offload_netdev_notifier);
473 static void __exit nft_flow_offload_module_exit(void)
475 nft_unregister_expr(&nft_flow_offload_type);
476 unregister_netdevice_notifier(&flow_offload_netdev_notifier);
479 module_init(nft_flow_offload_module_init);
480 module_exit(nft_flow_offload_module_exit);
482 MODULE_LICENSE("GPL");
483 MODULE_AUTHOR("Pablo Neira Ayuso <pablo@netfilter.org>");
484 MODULE_ALIAS_NFT_EXPR("flow_offload");
485 MODULE_DESCRIPTION("nftables hardware flow offload module");