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>
12 #include <net/inet_dscp.h>
13 #include <net/netfilter/nf_tables.h>
14 #include <net/netfilter/nf_tables_core.h>
15 #include <net/netfilter/nf_conntrack_core.h>
16 #include <net/netfilter/nf_conntrack_extend.h>
17 #include <net/netfilter/nf_flow_table.h>
18 
19 struct nft_flow_offload {
20 	struct nft_flowtable	*flowtable;
21 };
22 
23 static enum flow_offload_xmit_type nft_xmit_type(struct dst_entry *dst)
24 {
25 	if (dst_xfrm(dst))
26 		return FLOW_OFFLOAD_XMIT_XFRM;
27 
28 	return FLOW_OFFLOAD_XMIT_NEIGH;
29 }
30 
31 static void nft_default_forward_path(struct nf_flow_route *route,
32 				     struct dst_entry *dst_cache,
33 				     enum ip_conntrack_dir dir)
34 {
35 	route->tuple[!dir].in.ifindex	= dst_cache->dev->ifindex;
36 	route->tuple[dir].dst		= dst_cache;
37 	route->tuple[dir].xmit_type	= nft_xmit_type(dst_cache);
38 }
39 
40 static bool nft_is_valid_ether_device(const struct net_device *dev)
41 {
42 	if (!dev || (dev->flags & IFF_LOOPBACK) || dev->type != ARPHRD_ETHER ||
43 	    dev->addr_len != ETH_ALEN || !is_valid_ether_addr(dev->dev_addr))
44 		return false;
45 
46 	return true;
47 }
48 
49 static int nft_dev_fill_forward_path(const struct nf_flow_route *route,
50 				     const struct dst_entry *dst_cache,
51 				     const struct nf_conn *ct,
52 				     enum ip_conntrack_dir dir, u8 *ha,
53 				     struct net_device_path_stack *stack)
54 {
55 	const void *daddr = &ct->tuplehash[!dir].tuple.src.u3;
56 	struct net_device *dev = dst_cache->dev;
57 	struct neighbour *n;
58 	u8 nud_state;
59 
60 	if (!nft_is_valid_ether_device(dev))
61 		goto out;
62 
63 	n = dst_neigh_lookup(dst_cache, daddr);
64 	if (!n)
65 		return -1;
66 
67 	read_lock_bh(&n->lock);
68 	nud_state = n->nud_state;
69 	ether_addr_copy(ha, n->ha);
70 	read_unlock_bh(&n->lock);
71 	neigh_release(n);
72 
73 	if (!(nud_state & NUD_VALID))
74 		return -1;
75 
76 out:
77 	return dev_fill_forward_path(dev, ha, stack);
78 }
79 
80 struct nft_forward_info {
81 	const struct net_device *indev;
82 	const struct net_device *outdev;
83 	const struct net_device *hw_outdev;
84 	struct id {
85 		__u16	id;
86 		__be16	proto;
87 	} encap[NF_FLOW_TABLE_ENCAP_MAX];
88 	u8 num_encaps;
89 	u8 ingress_vlans;
90 	u8 h_source[ETH_ALEN];
91 	u8 h_dest[ETH_ALEN];
92 	enum flow_offload_xmit_type xmit_type;
93 };
94 
95 static void nft_dev_path_info(const struct net_device_path_stack *stack,
96 			      struct nft_forward_info *info,
97 			      unsigned char *ha, struct nf_flowtable *flowtable)
98 {
99 	const struct net_device_path *path;
100 	int i;
101 
102 	memcpy(info->h_dest, ha, ETH_ALEN);
103 
104 	for (i = 0; i < stack->num_paths; i++) {
105 		path = &stack->path[i];
106 		switch (path->type) {
107 		case DEV_PATH_ETHERNET:
108 		case DEV_PATH_DSA:
109 		case DEV_PATH_VLAN:
110 		case DEV_PATH_PPPOE:
111 			info->indev = path->dev;
112 			if (is_zero_ether_addr(info->h_source))
113 				memcpy(info->h_source, path->dev->dev_addr, ETH_ALEN);
114 
115 			if (path->type == DEV_PATH_ETHERNET)
116 				break;
117 			if (path->type == DEV_PATH_DSA) {
118 				i = stack->num_paths;
119 				break;
120 			}
121 
122 			/* DEV_PATH_VLAN and DEV_PATH_PPPOE */
123 			if (info->num_encaps >= NF_FLOW_TABLE_ENCAP_MAX) {
124 				info->indev = NULL;
125 				break;
126 			}
127 			if (!info->outdev)
128 				info->outdev = path->dev;
129 			info->encap[info->num_encaps].id = path->encap.id;
130 			info->encap[info->num_encaps].proto = path->encap.proto;
131 			info->num_encaps++;
132 			if (path->type == DEV_PATH_PPPOE)
133 				memcpy(info->h_dest, path->encap.h_dest, ETH_ALEN);
134 			break;
135 		case DEV_PATH_BRIDGE:
136 			if (is_zero_ether_addr(info->h_source))
137 				memcpy(info->h_source, path->dev->dev_addr, ETH_ALEN);
138 
139 			switch (path->bridge.vlan_mode) {
140 			case DEV_PATH_BR_VLAN_UNTAG_HW:
141 				info->ingress_vlans |= BIT(info->num_encaps - 1);
142 				break;
143 			case DEV_PATH_BR_VLAN_TAG:
144 				info->encap[info->num_encaps].id = path->bridge.vlan_id;
145 				info->encap[info->num_encaps].proto = path->bridge.vlan_proto;
146 				info->num_encaps++;
147 				break;
148 			case DEV_PATH_BR_VLAN_UNTAG:
149 				info->num_encaps--;
150 				break;
151 			case DEV_PATH_BR_VLAN_KEEP:
152 				break;
153 			}
154 			info->xmit_type = FLOW_OFFLOAD_XMIT_DIRECT;
155 			break;
156 		default:
157 			info->indev = NULL;
158 			break;
159 		}
160 	}
161 	if (!info->outdev)
162 		info->outdev = info->indev;
163 
164 	info->hw_outdev = info->indev;
165 
166 	if (nf_flowtable_hw_offload(flowtable) &&
167 	    nft_is_valid_ether_device(info->indev))
168 		info->xmit_type = FLOW_OFFLOAD_XMIT_DIRECT;
169 }
170 
171 static bool nft_flowtable_find_dev(const struct net_device *dev,
172 				   struct nft_flowtable *ft)
173 {
174 	struct nft_hook *hook;
175 	bool found = false;
176 
177 	list_for_each_entry_rcu(hook, &ft->hook_list, list) {
178 		if (!nft_hook_find_ops_rcu(hook, dev))
179 			continue;
180 
181 		found = true;
182 		break;
183 	}
184 
185 	return found;
186 }
187 
188 static void nft_dev_forward_path(struct nf_flow_route *route,
189 				 const struct nf_conn *ct,
190 				 enum ip_conntrack_dir dir,
191 				 struct nft_flowtable *ft)
192 {
193 	const struct dst_entry *dst = route->tuple[dir].dst;
194 	struct net_device_path_stack stack;
195 	struct nft_forward_info info = {};
196 	unsigned char ha[ETH_ALEN];
197 	int i;
198 
199 	if (nft_dev_fill_forward_path(route, dst, ct, dir, ha, &stack) >= 0)
200 		nft_dev_path_info(&stack, &info, ha, &ft->data);
201 
202 	if (!info.indev || !nft_flowtable_find_dev(info.indev, ft))
203 		return;
204 
205 	route->tuple[!dir].in.ifindex = info.indev->ifindex;
206 	for (i = 0; i < info.num_encaps; i++) {
207 		route->tuple[!dir].in.encap[i].id = info.encap[i].id;
208 		route->tuple[!dir].in.encap[i].proto = info.encap[i].proto;
209 	}
210 	route->tuple[!dir].in.num_encaps = info.num_encaps;
211 	route->tuple[!dir].in.ingress_vlans = info.ingress_vlans;
212 
213 	if (info.xmit_type == FLOW_OFFLOAD_XMIT_DIRECT) {
214 		memcpy(route->tuple[dir].out.h_source, info.h_source, ETH_ALEN);
215 		memcpy(route->tuple[dir].out.h_dest, info.h_dest, ETH_ALEN);
216 		route->tuple[dir].out.ifindex = info.outdev->ifindex;
217 		route->tuple[dir].out.hw_ifindex = info.hw_outdev->ifindex;
218 		route->tuple[dir].xmit_type = info.xmit_type;
219 	}
220 }
221 
222 static int nft_flow_route(const struct nft_pktinfo *pkt,
223 			  const struct nf_conn *ct,
224 			  struct nf_flow_route *route,
225 			  enum ip_conntrack_dir dir,
226 			  struct nft_flowtable *ft)
227 {
228 	struct dst_entry *this_dst = skb_dst(pkt->skb);
229 	struct dst_entry *other_dst = NULL;
230 	struct flowi fl;
231 
232 	memset(&fl, 0, sizeof(fl));
233 	switch (nft_pf(pkt)) {
234 	case NFPROTO_IPV4:
235 		fl.u.ip4.daddr = ct->tuplehash[dir].tuple.src.u3.ip;
236 		fl.u.ip4.saddr = ct->tuplehash[!dir].tuple.src.u3.ip;
237 		fl.u.ip4.flowi4_oif = nft_in(pkt)->ifindex;
238 		fl.u.ip4.flowi4_iif = this_dst->dev->ifindex;
239 		fl.u.ip4.flowi4_tos = inet_dscp_to_dsfield(ip4h_dscp(ip_hdr(pkt->skb)));
240 		fl.u.ip4.flowi4_mark = pkt->skb->mark;
241 		fl.u.ip4.flowi4_flags = FLOWI_FLAG_ANYSRC;
242 		break;
243 	case NFPROTO_IPV6:
244 		fl.u.ip6.daddr = ct->tuplehash[dir].tuple.src.u3.in6;
245 		fl.u.ip6.saddr = ct->tuplehash[!dir].tuple.src.u3.in6;
246 		fl.u.ip6.flowi6_oif = nft_in(pkt)->ifindex;
247 		fl.u.ip6.flowi6_iif = this_dst->dev->ifindex;
248 		fl.u.ip6.flowlabel = ip6_flowinfo(ipv6_hdr(pkt->skb));
249 		fl.u.ip6.flowi6_mark = pkt->skb->mark;
250 		fl.u.ip6.flowi6_flags = FLOWI_FLAG_ANYSRC;
251 		break;
252 	}
253 
254 	if (!dst_hold_safe(this_dst))
255 		return -ENOENT;
256 
257 	nf_route(nft_net(pkt), &other_dst, &fl, false, nft_pf(pkt));
258 	if (!other_dst) {
259 		dst_release(this_dst);
260 		return -ENOENT;
261 	}
262 
263 	nft_default_forward_path(route, this_dst, dir);
264 	nft_default_forward_path(route, other_dst, !dir);
265 
266 	if (route->tuple[dir].xmit_type	== FLOW_OFFLOAD_XMIT_NEIGH &&
267 	    route->tuple[!dir].xmit_type == FLOW_OFFLOAD_XMIT_NEIGH) {
268 		nft_dev_forward_path(route, ct, dir, ft);
269 		nft_dev_forward_path(route, ct, !dir, ft);
270 	}
271 
272 	return 0;
273 }
274 
275 static bool nft_flow_offload_skip(struct sk_buff *skb, int family)
276 {
277 	if (skb_sec_path(skb))
278 		return true;
279 
280 	if (family == NFPROTO_IPV4) {
281 		const struct ip_options *opt;
282 
283 		opt = &(IPCB(skb)->opt);
284 
285 		if (unlikely(opt->optlen))
286 			return true;
287 	}
288 
289 	return false;
290 }
291 
292 static void flow_offload_ct_tcp(struct nf_conn *ct)
293 {
294 	/* conntrack will not see all packets, disable tcp window validation. */
295 	spin_lock_bh(&ct->lock);
296 	ct->proto.tcp.seen[0].flags |= IP_CT_TCP_FLAG_BE_LIBERAL;
297 	ct->proto.tcp.seen[1].flags |= IP_CT_TCP_FLAG_BE_LIBERAL;
298 	spin_unlock_bh(&ct->lock);
299 }
300 
301 static void nft_flow_offload_eval(const struct nft_expr *expr,
302 				  struct nft_regs *regs,
303 				  const struct nft_pktinfo *pkt)
304 {
305 	struct nft_flow_offload *priv = nft_expr_priv(expr);
306 	struct nf_flowtable *flowtable = &priv->flowtable->data;
307 	struct tcphdr _tcph, *tcph = NULL;
308 	struct nf_flow_route route = {};
309 	enum ip_conntrack_info ctinfo;
310 	struct flow_offload *flow;
311 	enum ip_conntrack_dir dir;
312 	struct nf_conn *ct;
313 	int ret;
314 
315 	if (nft_flow_offload_skip(pkt->skb, nft_pf(pkt)))
316 		goto out;
317 
318 	ct = nf_ct_get(pkt->skb, &ctinfo);
319 	if (!ct)
320 		goto out;
321 
322 	switch (ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple.dst.protonum) {
323 	case IPPROTO_TCP:
324 		tcph = skb_header_pointer(pkt->skb, nft_thoff(pkt),
325 					  sizeof(_tcph), &_tcph);
326 		if (unlikely(!tcph || tcph->fin || tcph->rst ||
327 			     !nf_conntrack_tcp_established(ct)))
328 			goto out;
329 		break;
330 	case IPPROTO_UDP:
331 		break;
332 #ifdef CONFIG_NF_CT_PROTO_GRE
333 	case IPPROTO_GRE: {
334 		struct nf_conntrack_tuple *tuple;
335 
336 		if (ct->status & IPS_NAT_MASK)
337 			goto out;
338 		tuple = &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple;
339 		/* No support for GRE v1 */
340 		if (tuple->src.u.gre.key || tuple->dst.u.gre.key)
341 			goto out;
342 		break;
343 	}
344 #endif
345 	default:
346 		goto out;
347 	}
348 
349 	if (nf_ct_ext_exist(ct, NF_CT_EXT_HELPER) ||
350 	    ct->status & (IPS_SEQ_ADJUST | IPS_NAT_CLASH))
351 		goto out;
352 
353 	if (!nf_ct_is_confirmed(ct))
354 		goto out;
355 
356 	if (test_and_set_bit(IPS_OFFLOAD_BIT, &ct->status))
357 		goto out;
358 
359 	dir = CTINFO2DIR(ctinfo);
360 	if (nft_flow_route(pkt, ct, &route, dir, priv->flowtable) < 0)
361 		goto err_flow_route;
362 
363 	flow = flow_offload_alloc(ct);
364 	if (!flow)
365 		goto err_flow_alloc;
366 
367 	flow_offload_route_init(flow, &route);
368 	if (tcph)
369 		flow_offload_ct_tcp(ct);
370 
371 	__set_bit(NF_FLOW_HW_BIDIRECTIONAL, &flow->flags);
372 	ret = flow_offload_add(flowtable, flow);
373 	if (ret < 0)
374 		goto err_flow_add;
375 
376 	return;
377 
378 err_flow_add:
379 	flow_offload_free(flow);
380 err_flow_alloc:
381 	dst_release(route.tuple[dir].dst);
382 	dst_release(route.tuple[!dir].dst);
383 err_flow_route:
384 	clear_bit(IPS_OFFLOAD_BIT, &ct->status);
385 out:
386 	regs->verdict.code = NFT_BREAK;
387 }
388 
389 static int nft_flow_offload_validate(const struct nft_ctx *ctx,
390 				     const struct nft_expr *expr)
391 {
392 	unsigned int hook_mask = (1 << NF_INET_FORWARD);
393 
394 	if (ctx->family != NFPROTO_IPV4 &&
395 	    ctx->family != NFPROTO_IPV6 &&
396 	    ctx->family != NFPROTO_INET)
397 		return -EOPNOTSUPP;
398 
399 	return nft_chain_validate_hooks(ctx->chain, hook_mask);
400 }
401 
402 static const struct nla_policy nft_flow_offload_policy[NFTA_FLOW_MAX + 1] = {
403 	[NFTA_FLOW_TABLE_NAME]	= { .type = NLA_STRING,
404 				    .len = NFT_NAME_MAXLEN - 1 },
405 };
406 
407 static int nft_flow_offload_init(const struct nft_ctx *ctx,
408 				 const struct nft_expr *expr,
409 				 const struct nlattr * const tb[])
410 {
411 	struct nft_flow_offload *priv = nft_expr_priv(expr);
412 	u8 genmask = nft_genmask_next(ctx->net);
413 	struct nft_flowtable *flowtable;
414 
415 	if (!tb[NFTA_FLOW_TABLE_NAME])
416 		return -EINVAL;
417 
418 	flowtable = nft_flowtable_lookup(ctx->net, ctx->table,
419 					 tb[NFTA_FLOW_TABLE_NAME], genmask);
420 	if (IS_ERR(flowtable))
421 		return PTR_ERR(flowtable);
422 
423 	if (!nft_use_inc(&flowtable->use))
424 		return -EMFILE;
425 
426 	priv->flowtable = flowtable;
427 
428 	return nf_ct_netns_get(ctx->net, ctx->family);
429 }
430 
431 static void nft_flow_offload_deactivate(const struct nft_ctx *ctx,
432 					const struct nft_expr *expr,
433 					enum nft_trans_phase phase)
434 {
435 	struct nft_flow_offload *priv = nft_expr_priv(expr);
436 
437 	nf_tables_deactivate_flowtable(ctx, priv->flowtable, phase);
438 }
439 
440 static void nft_flow_offload_activate(const struct nft_ctx *ctx,
441 				      const struct nft_expr *expr)
442 {
443 	struct nft_flow_offload *priv = nft_expr_priv(expr);
444 
445 	nft_use_inc_restore(&priv->flowtable->use);
446 }
447 
448 static void nft_flow_offload_destroy(const struct nft_ctx *ctx,
449 				     const struct nft_expr *expr)
450 {
451 	nf_ct_netns_put(ctx->net, ctx->family);
452 }
453 
454 static int nft_flow_offload_dump(struct sk_buff *skb,
455 				 const struct nft_expr *expr, bool reset)
456 {
457 	struct nft_flow_offload *priv = nft_expr_priv(expr);
458 
459 	if (nla_put_string(skb, NFTA_FLOW_TABLE_NAME, priv->flowtable->name))
460 		goto nla_put_failure;
461 
462 	return 0;
463 
464 nla_put_failure:
465 	return -1;
466 }
467 
468 static struct nft_expr_type nft_flow_offload_type;
469 static const struct nft_expr_ops nft_flow_offload_ops = {
470 	.type		= &nft_flow_offload_type,
471 	.size		= NFT_EXPR_SIZE(sizeof(struct nft_flow_offload)),
472 	.eval		= nft_flow_offload_eval,
473 	.init		= nft_flow_offload_init,
474 	.activate	= nft_flow_offload_activate,
475 	.deactivate	= nft_flow_offload_deactivate,
476 	.destroy	= nft_flow_offload_destroy,
477 	.validate	= nft_flow_offload_validate,
478 	.dump		= nft_flow_offload_dump,
479 	.reduce		= NFT_REDUCE_READONLY,
480 };
481 
482 static struct nft_expr_type nft_flow_offload_type __read_mostly = {
483 	.name		= "flow_offload",
484 	.ops		= &nft_flow_offload_ops,
485 	.policy		= nft_flow_offload_policy,
486 	.maxattr	= NFTA_FLOW_MAX,
487 	.owner		= THIS_MODULE,
488 };
489 
490 static int flow_offload_netdev_event(struct notifier_block *this,
491 				     unsigned long event, void *ptr)
492 {
493 	struct net_device *dev = netdev_notifier_info_to_dev(ptr);
494 
495 	if (event != NETDEV_DOWN)
496 		return NOTIFY_DONE;
497 
498 	nf_flow_table_cleanup(dev);
499 
500 	return NOTIFY_DONE;
501 }
502 
503 static struct notifier_block flow_offload_netdev_notifier = {
504 	.notifier_call	= flow_offload_netdev_event,
505 };
506 
507 static int __init nft_flow_offload_module_init(void)
508 {
509 	int err;
510 
511 	err = register_netdevice_notifier(&flow_offload_netdev_notifier);
512 	if (err)
513 		goto err;
514 
515 	err = nft_register_expr(&nft_flow_offload_type);
516 	if (err < 0)
517 		goto register_expr;
518 
519 	return 0;
520 
521 register_expr:
522 	unregister_netdevice_notifier(&flow_offload_netdev_notifier);
523 err:
524 	return err;
525 }
526 
527 static void __exit nft_flow_offload_module_exit(void)
528 {
529 	nft_unregister_expr(&nft_flow_offload_type);
530 	unregister_netdevice_notifier(&flow_offload_netdev_notifier);
531 }
532 
533 module_init(nft_flow_offload_module_init);
534 module_exit(nft_flow_offload_module_exit);
535 
536 MODULE_LICENSE("GPL");
537 MODULE_AUTHOR("Pablo Neira Ayuso <pablo@netfilter.org>");
538 MODULE_ALIAS_NFT_EXPR("flow_offload");
539 MODULE_DESCRIPTION("nftables hardware flow offload module");
540