xref: /linux/net/xfrm/xfrm_interface_core.c (revision c532de5a67a70f8533d495f8f2aaa9a0491c3ad0)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  *	XFRM virtual interface
4  *
5  *	Copyright (C) 2018 secunet Security Networks AG
6  *
7  *	Author:
8  *	Steffen Klassert <steffen.klassert@secunet.com>
9  */
10 
11 #include <linux/module.h>
12 #include <linux/capability.h>
13 #include <linux/errno.h>
14 #include <linux/types.h>
15 #include <linux/sockios.h>
16 #include <linux/icmp.h>
17 #include <linux/if.h>
18 #include <linux/in.h>
19 #include <linux/ip.h>
20 #include <linux/net.h>
21 #include <linux/in6.h>
22 #include <linux/netdevice.h>
23 #include <linux/if_link.h>
24 #include <linux/if_arp.h>
25 #include <linux/icmpv6.h>
26 #include <linux/init.h>
27 #include <linux/route.h>
28 #include <linux/rtnetlink.h>
29 #include <linux/netfilter_ipv6.h>
30 #include <linux/slab.h>
31 #include <linux/hash.h>
32 
33 #include <linux/uaccess.h>
34 #include <linux/atomic.h>
35 
36 #include <net/gso.h>
37 #include <net/icmp.h>
38 #include <net/ip.h>
39 #include <net/ipv6.h>
40 #include <net/ip6_route.h>
41 #include <net/ip_tunnels.h>
42 #include <net/addrconf.h>
43 #include <net/xfrm.h>
44 #include <net/net_namespace.h>
45 #include <net/dst_metadata.h>
46 #include <net/netns/generic.h>
47 #include <linux/etherdevice.h>
48 
49 static int xfrmi_dev_init(struct net_device *dev);
50 static void xfrmi_dev_setup(struct net_device *dev);
51 static struct rtnl_link_ops xfrmi_link_ops __read_mostly;
52 static unsigned int xfrmi_net_id __read_mostly;
53 static const struct net_device_ops xfrmi_netdev_ops;
54 
55 #define XFRMI_HASH_BITS	8
56 #define XFRMI_HASH_SIZE	BIT(XFRMI_HASH_BITS)
57 
58 struct xfrmi_net {
59 	/* lists for storing interfaces in use */
60 	struct xfrm_if __rcu *xfrmi[XFRMI_HASH_SIZE];
61 	struct xfrm_if __rcu *collect_md_xfrmi;
62 };
63 
64 static const struct nla_policy xfrm_lwt_policy[LWT_XFRM_MAX + 1] = {
65 	[LWT_XFRM_IF_ID]	= NLA_POLICY_MIN(NLA_U32, 1),
66 	[LWT_XFRM_LINK]		= NLA_POLICY_MIN(NLA_U32, 1),
67 };
68 
69 static void xfrmi_destroy_state(struct lwtunnel_state *lwt)
70 {
71 }
72 
73 static int xfrmi_build_state(struct net *net, struct nlattr *nla,
74 			     unsigned int family, const void *cfg,
75 			     struct lwtunnel_state **ts,
76 			     struct netlink_ext_ack *extack)
77 {
78 	struct nlattr *tb[LWT_XFRM_MAX + 1];
79 	struct lwtunnel_state *new_state;
80 	struct xfrm_md_info *info;
81 	int ret;
82 
83 	ret = nla_parse_nested(tb, LWT_XFRM_MAX, nla, xfrm_lwt_policy, extack);
84 	if (ret < 0)
85 		return ret;
86 
87 	if (!tb[LWT_XFRM_IF_ID]) {
88 		NL_SET_ERR_MSG(extack, "if_id must be set");
89 		return -EINVAL;
90 	}
91 
92 	new_state = lwtunnel_state_alloc(sizeof(*info));
93 	if (!new_state) {
94 		NL_SET_ERR_MSG(extack, "failed to create encap info");
95 		return -ENOMEM;
96 	}
97 
98 	new_state->type = LWTUNNEL_ENCAP_XFRM;
99 
100 	info = lwt_xfrm_info(new_state);
101 
102 	info->if_id = nla_get_u32(tb[LWT_XFRM_IF_ID]);
103 
104 	if (tb[LWT_XFRM_LINK])
105 		info->link = nla_get_u32(tb[LWT_XFRM_LINK]);
106 
107 	*ts = new_state;
108 	return 0;
109 }
110 
111 static int xfrmi_fill_encap_info(struct sk_buff *skb,
112 				 struct lwtunnel_state *lwt)
113 {
114 	struct xfrm_md_info *info = lwt_xfrm_info(lwt);
115 
116 	if (nla_put_u32(skb, LWT_XFRM_IF_ID, info->if_id) ||
117 	    (info->link && nla_put_u32(skb, LWT_XFRM_LINK, info->link)))
118 		return -EMSGSIZE;
119 
120 	return 0;
121 }
122 
123 static int xfrmi_encap_nlsize(struct lwtunnel_state *lwtstate)
124 {
125 	return nla_total_size(sizeof(u32)) + /* LWT_XFRM_IF_ID */
126 		nla_total_size(sizeof(u32)); /* LWT_XFRM_LINK */
127 }
128 
129 static int xfrmi_encap_cmp(struct lwtunnel_state *a, struct lwtunnel_state *b)
130 {
131 	struct xfrm_md_info *a_info = lwt_xfrm_info(a);
132 	struct xfrm_md_info *b_info = lwt_xfrm_info(b);
133 
134 	return memcmp(a_info, b_info, sizeof(*a_info));
135 }
136 
137 static const struct lwtunnel_encap_ops xfrmi_encap_ops = {
138 	.build_state	= xfrmi_build_state,
139 	.destroy_state	= xfrmi_destroy_state,
140 	.fill_encap	= xfrmi_fill_encap_info,
141 	.get_encap_size = xfrmi_encap_nlsize,
142 	.cmp_encap	= xfrmi_encap_cmp,
143 	.owner		= THIS_MODULE,
144 };
145 
146 #define for_each_xfrmi_rcu(start, xi) \
147 	for (xi = rcu_dereference(start); xi; xi = rcu_dereference(xi->next))
148 
149 static u32 xfrmi_hash(u32 if_id)
150 {
151 	return hash_32(if_id, XFRMI_HASH_BITS);
152 }
153 
154 static struct xfrm_if *xfrmi_lookup(struct net *net, struct xfrm_state *x)
155 {
156 	struct xfrmi_net *xfrmn = net_generic(net, xfrmi_net_id);
157 	struct xfrm_if *xi;
158 
159 	for_each_xfrmi_rcu(xfrmn->xfrmi[xfrmi_hash(x->if_id)], xi) {
160 		if (x->if_id == xi->p.if_id &&
161 		    (xi->dev->flags & IFF_UP))
162 			return xi;
163 	}
164 
165 	xi = rcu_dereference(xfrmn->collect_md_xfrmi);
166 	if (xi && (xi->dev->flags & IFF_UP))
167 		return xi;
168 
169 	return NULL;
170 }
171 
172 static bool xfrmi_decode_session(struct sk_buff *skb,
173 				 unsigned short family,
174 				 struct xfrm_if_decode_session_result *res)
175 {
176 	struct net_device *dev;
177 	struct xfrm_if *xi;
178 	int ifindex = 0;
179 
180 	if (!secpath_exists(skb) || !skb->dev)
181 		return false;
182 
183 	switch (family) {
184 	case AF_INET6:
185 		ifindex = inet6_sdif(skb);
186 		break;
187 	case AF_INET:
188 		ifindex = inet_sdif(skb);
189 		break;
190 	}
191 
192 	if (ifindex) {
193 		struct net *net = xs_net(xfrm_input_state(skb));
194 
195 		dev = dev_get_by_index_rcu(net, ifindex);
196 	} else {
197 		dev = skb->dev;
198 	}
199 
200 	if (!dev || !(dev->flags & IFF_UP))
201 		return false;
202 	if (dev->netdev_ops != &xfrmi_netdev_ops)
203 		return false;
204 
205 	xi = netdev_priv(dev);
206 	res->net = xi->net;
207 
208 	if (xi->p.collect_md)
209 		res->if_id = xfrm_input_state(skb)->if_id;
210 	else
211 		res->if_id = xi->p.if_id;
212 	return true;
213 }
214 
215 static void xfrmi_link(struct xfrmi_net *xfrmn, struct xfrm_if *xi)
216 {
217 	struct xfrm_if __rcu **xip = &xfrmn->xfrmi[xfrmi_hash(xi->p.if_id)];
218 
219 	rcu_assign_pointer(xi->next , rtnl_dereference(*xip));
220 	rcu_assign_pointer(*xip, xi);
221 }
222 
223 static void xfrmi_unlink(struct xfrmi_net *xfrmn, struct xfrm_if *xi)
224 {
225 	struct xfrm_if __rcu **xip;
226 	struct xfrm_if *iter;
227 
228 	for (xip = &xfrmn->xfrmi[xfrmi_hash(xi->p.if_id)];
229 	     (iter = rtnl_dereference(*xip)) != NULL;
230 	     xip = &iter->next) {
231 		if (xi == iter) {
232 			rcu_assign_pointer(*xip, xi->next);
233 			break;
234 		}
235 	}
236 }
237 
238 static void xfrmi_dev_free(struct net_device *dev)
239 {
240 	struct xfrm_if *xi = netdev_priv(dev);
241 
242 	gro_cells_destroy(&xi->gro_cells);
243 }
244 
245 static int xfrmi_create(struct net_device *dev)
246 {
247 	struct xfrm_if *xi = netdev_priv(dev);
248 	struct net *net = dev_net(dev);
249 	struct xfrmi_net *xfrmn = net_generic(net, xfrmi_net_id);
250 	int err;
251 
252 	dev->rtnl_link_ops = &xfrmi_link_ops;
253 	err = register_netdevice(dev);
254 	if (err < 0)
255 		goto out;
256 
257 	if (xi->p.collect_md)
258 		rcu_assign_pointer(xfrmn->collect_md_xfrmi, xi);
259 	else
260 		xfrmi_link(xfrmn, xi);
261 
262 	return 0;
263 
264 out:
265 	return err;
266 }
267 
268 static struct xfrm_if *xfrmi_locate(struct net *net, struct xfrm_if_parms *p)
269 {
270 	struct xfrm_if __rcu **xip;
271 	struct xfrm_if *xi;
272 	struct xfrmi_net *xfrmn = net_generic(net, xfrmi_net_id);
273 
274 	for (xip = &xfrmn->xfrmi[xfrmi_hash(p->if_id)];
275 	     (xi = rtnl_dereference(*xip)) != NULL;
276 	     xip = &xi->next)
277 		if (xi->p.if_id == p->if_id)
278 			return xi;
279 
280 	return NULL;
281 }
282 
283 static void xfrmi_dev_uninit(struct net_device *dev)
284 {
285 	struct xfrm_if *xi = netdev_priv(dev);
286 	struct xfrmi_net *xfrmn = net_generic(xi->net, xfrmi_net_id);
287 
288 	if (xi->p.collect_md)
289 		RCU_INIT_POINTER(xfrmn->collect_md_xfrmi, NULL);
290 	else
291 		xfrmi_unlink(xfrmn, xi);
292 }
293 
294 static void xfrmi_scrub_packet(struct sk_buff *skb, bool xnet)
295 {
296 	skb_clear_tstamp(skb);
297 	skb->pkt_type = PACKET_HOST;
298 	skb->skb_iif = 0;
299 	skb->ignore_df = 0;
300 	skb_dst_drop(skb);
301 	nf_reset_ct(skb);
302 	nf_reset_trace(skb);
303 
304 	if (!xnet)
305 		return;
306 
307 	ipvs_reset(skb);
308 	secpath_reset(skb);
309 	skb_orphan(skb);
310 	skb->mark = 0;
311 }
312 
313 static int xfrmi_input(struct sk_buff *skb, int nexthdr, __be32 spi,
314 		       int encap_type, unsigned short family)
315 {
316 	struct sec_path *sp;
317 
318 	sp = skb_sec_path(skb);
319 	if (sp && (sp->len || sp->olen) &&
320 	    !xfrm_policy_check(NULL, XFRM_POLICY_IN, skb, family))
321 		goto discard;
322 
323 	XFRM_SPI_SKB_CB(skb)->family = family;
324 	if (family == AF_INET) {
325 		XFRM_SPI_SKB_CB(skb)->daddroff = offsetof(struct iphdr, daddr);
326 		XFRM_TUNNEL_SKB_CB(skb)->tunnel.ip4 = NULL;
327 	} else {
328 		XFRM_SPI_SKB_CB(skb)->daddroff = offsetof(struct ipv6hdr, daddr);
329 		XFRM_TUNNEL_SKB_CB(skb)->tunnel.ip6 = NULL;
330 	}
331 
332 	return xfrm_input(skb, nexthdr, spi, encap_type);
333 discard:
334 	kfree_skb(skb);
335 	return 0;
336 }
337 
338 static int xfrmi4_rcv(struct sk_buff *skb)
339 {
340 	return xfrmi_input(skb, ip_hdr(skb)->protocol, 0, 0, AF_INET);
341 }
342 
343 static int xfrmi6_rcv(struct sk_buff *skb)
344 {
345 	return xfrmi_input(skb, skb_network_header(skb)[IP6CB(skb)->nhoff],
346 			   0, 0, AF_INET6);
347 }
348 
349 static int xfrmi4_input(struct sk_buff *skb, int nexthdr, __be32 spi, int encap_type)
350 {
351 	return xfrmi_input(skb, nexthdr, spi, encap_type, AF_INET);
352 }
353 
354 static int xfrmi6_input(struct sk_buff *skb, int nexthdr, __be32 spi, int encap_type)
355 {
356 	return xfrmi_input(skb, nexthdr, spi, encap_type, AF_INET6);
357 }
358 
359 static int xfrmi_rcv_cb(struct sk_buff *skb, int err)
360 {
361 	const struct xfrm_mode *inner_mode;
362 	struct net_device *dev;
363 	struct xfrm_state *x;
364 	struct xfrm_if *xi;
365 	bool xnet;
366 	int link;
367 
368 	if (err && !secpath_exists(skb))
369 		return 0;
370 
371 	x = xfrm_input_state(skb);
372 
373 	xi = xfrmi_lookup(xs_net(x), x);
374 	if (!xi)
375 		return 1;
376 
377 	link = skb->dev->ifindex;
378 	dev = xi->dev;
379 	skb->dev = dev;
380 
381 	if (err) {
382 		DEV_STATS_INC(dev, rx_errors);
383 		DEV_STATS_INC(dev, rx_dropped);
384 
385 		return 0;
386 	}
387 
388 	xnet = !net_eq(xi->net, dev_net(skb->dev));
389 
390 	if (xnet) {
391 		inner_mode = &x->inner_mode;
392 
393 		if (x->sel.family == AF_UNSPEC) {
394 			inner_mode = xfrm_ip2inner_mode(x, XFRM_MODE_SKB_CB(skb)->protocol);
395 			if (inner_mode == NULL) {
396 				XFRM_INC_STATS(dev_net(skb->dev),
397 					       LINUX_MIB_XFRMINSTATEMODEERROR);
398 				return -EINVAL;
399 			}
400 		}
401 
402 		if (!xfrm_policy_check(NULL, XFRM_POLICY_IN, skb,
403 				       inner_mode->family))
404 			return -EPERM;
405 	}
406 
407 	xfrmi_scrub_packet(skb, xnet);
408 	if (xi->p.collect_md) {
409 		struct metadata_dst *md_dst;
410 
411 		md_dst = metadata_dst_alloc(0, METADATA_XFRM, GFP_ATOMIC);
412 		if (!md_dst)
413 			return -ENOMEM;
414 
415 		md_dst->u.xfrm_info.if_id = x->if_id;
416 		md_dst->u.xfrm_info.link = link;
417 		skb_dst_set(skb, (struct dst_entry *)md_dst);
418 	}
419 	dev_sw_netstats_rx_add(dev, skb->len);
420 
421 	return 0;
422 }
423 
424 static int
425 xfrmi_xmit2(struct sk_buff *skb, struct net_device *dev, struct flowi *fl)
426 {
427 	struct xfrm_if *xi = netdev_priv(dev);
428 	struct dst_entry *dst = skb_dst(skb);
429 	unsigned int length = skb->len;
430 	struct net_device *tdev;
431 	struct xfrm_state *x;
432 	int err = -1;
433 	u32 if_id;
434 	int mtu;
435 
436 	if (xi->p.collect_md) {
437 		struct xfrm_md_info *md_info = skb_xfrm_md_info(skb);
438 
439 		if (unlikely(!md_info))
440 			return -EINVAL;
441 
442 		if_id = md_info->if_id;
443 		fl->flowi_oif = md_info->link;
444 		if (md_info->dst_orig) {
445 			struct dst_entry *tmp_dst = dst;
446 
447 			dst = md_info->dst_orig;
448 			skb_dst_set(skb, dst);
449 			md_info->dst_orig = NULL;
450 			dst_release(tmp_dst);
451 		}
452 	} else {
453 		if_id = xi->p.if_id;
454 	}
455 
456 	dst_hold(dst);
457 	dst = xfrm_lookup_with_ifid(xi->net, dst, fl, NULL, 0, if_id);
458 	if (IS_ERR(dst)) {
459 		err = PTR_ERR(dst);
460 		dst = NULL;
461 		goto tx_err_link_failure;
462 	}
463 
464 	x = dst->xfrm;
465 	if (!x)
466 		goto tx_err_link_failure;
467 
468 	if (x->if_id != if_id)
469 		goto tx_err_link_failure;
470 
471 	tdev = dst->dev;
472 
473 	if (tdev == dev) {
474 		DEV_STATS_INC(dev, collisions);
475 		net_warn_ratelimited("%s: Local routing loop detected!\n",
476 				     dev->name);
477 		goto tx_err_dst_release;
478 	}
479 
480 	mtu = dst_mtu(dst);
481 	if ((!skb_is_gso(skb) && skb->len > mtu) ||
482 	    (skb_is_gso(skb) && !skb_gso_validate_network_len(skb, mtu))) {
483 		skb_dst_update_pmtu_no_confirm(skb, mtu);
484 
485 		if (skb->protocol == htons(ETH_P_IPV6)) {
486 			if (mtu < IPV6_MIN_MTU)
487 				mtu = IPV6_MIN_MTU;
488 
489 			if (skb->len > 1280)
490 				icmpv6_ndo_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu);
491 			else
492 				goto xmit;
493 		} else {
494 			if (!(ip_hdr(skb)->frag_off & htons(IP_DF)))
495 				goto xmit;
496 			icmp_ndo_send(skb, ICMP_DEST_UNREACH, ICMP_FRAG_NEEDED,
497 				      htonl(mtu));
498 		}
499 
500 		dst_release(dst);
501 		return -EMSGSIZE;
502 	}
503 
504 xmit:
505 	xfrmi_scrub_packet(skb, !net_eq(xi->net, dev_net(dev)));
506 	skb_dst_set(skb, dst);
507 	skb->dev = tdev;
508 
509 	err = dst_output(xi->net, skb->sk, skb);
510 	if (net_xmit_eval(err) == 0) {
511 		dev_sw_netstats_tx_add(dev, 1, length);
512 	} else {
513 		DEV_STATS_INC(dev, tx_errors);
514 		DEV_STATS_INC(dev, tx_aborted_errors);
515 	}
516 
517 	return 0;
518 tx_err_link_failure:
519 	DEV_STATS_INC(dev, tx_carrier_errors);
520 	dst_link_failure(skb);
521 tx_err_dst_release:
522 	dst_release(dst);
523 	return err;
524 }
525 
526 static netdev_tx_t xfrmi_xmit(struct sk_buff *skb, struct net_device *dev)
527 {
528 	struct xfrm_if *xi = netdev_priv(dev);
529 	struct dst_entry *dst = skb_dst(skb);
530 	struct flowi fl;
531 	int ret;
532 
533 	memset(&fl, 0, sizeof(fl));
534 
535 	switch (skb->protocol) {
536 	case htons(ETH_P_IPV6):
537 		memset(IP6CB(skb), 0, sizeof(*IP6CB(skb)));
538 		xfrm_decode_session(dev_net(dev), skb, &fl, AF_INET6);
539 		if (!dst) {
540 			fl.u.ip6.flowi6_oif = dev->ifindex;
541 			fl.u.ip6.flowi6_flags |= FLOWI_FLAG_ANYSRC;
542 			dst = ip6_route_output(dev_net(dev), NULL, &fl.u.ip6);
543 			if (dst->error) {
544 				dst_release(dst);
545 				DEV_STATS_INC(dev, tx_carrier_errors);
546 				goto tx_err;
547 			}
548 			skb_dst_set(skb, dst);
549 		}
550 		break;
551 	case htons(ETH_P_IP):
552 		memset(IPCB(skb), 0, sizeof(*IPCB(skb)));
553 		xfrm_decode_session(dev_net(dev), skb, &fl, AF_INET);
554 		if (!dst) {
555 			struct rtable *rt;
556 
557 			fl.u.ip4.flowi4_oif = dev->ifindex;
558 			fl.u.ip4.flowi4_flags |= FLOWI_FLAG_ANYSRC;
559 			rt = __ip_route_output_key(dev_net(dev), &fl.u.ip4);
560 			if (IS_ERR(rt)) {
561 				DEV_STATS_INC(dev, tx_carrier_errors);
562 				goto tx_err;
563 			}
564 			skb_dst_set(skb, &rt->dst);
565 		}
566 		break;
567 	default:
568 		goto tx_err;
569 	}
570 
571 	fl.flowi_oif = xi->p.link;
572 
573 	ret = xfrmi_xmit2(skb, dev, &fl);
574 	if (ret < 0)
575 		goto tx_err;
576 
577 	return NETDEV_TX_OK;
578 
579 tx_err:
580 	DEV_STATS_INC(dev, tx_errors);
581 	DEV_STATS_INC(dev, tx_dropped);
582 	kfree_skb(skb);
583 	return NETDEV_TX_OK;
584 }
585 
586 static int xfrmi4_err(struct sk_buff *skb, u32 info)
587 {
588 	const struct iphdr *iph = (const struct iphdr *)skb->data;
589 	struct net *net = dev_net(skb->dev);
590 	int protocol = iph->protocol;
591 	struct ip_comp_hdr *ipch;
592 	struct ip_esp_hdr *esph;
593 	struct ip_auth_hdr *ah ;
594 	struct xfrm_state *x;
595 	struct xfrm_if *xi;
596 	__be32 spi;
597 
598 	switch (protocol) {
599 	case IPPROTO_ESP:
600 		esph = (struct ip_esp_hdr *)(skb->data+(iph->ihl<<2));
601 		spi = esph->spi;
602 		break;
603 	case IPPROTO_AH:
604 		ah = (struct ip_auth_hdr *)(skb->data+(iph->ihl<<2));
605 		spi = ah->spi;
606 		break;
607 	case IPPROTO_COMP:
608 		ipch = (struct ip_comp_hdr *)(skb->data+(iph->ihl<<2));
609 		spi = htonl(ntohs(ipch->cpi));
610 		break;
611 	default:
612 		return 0;
613 	}
614 
615 	switch (icmp_hdr(skb)->type) {
616 	case ICMP_DEST_UNREACH:
617 		if (icmp_hdr(skb)->code != ICMP_FRAG_NEEDED)
618 			return 0;
619 		break;
620 	case ICMP_REDIRECT:
621 		break;
622 	default:
623 		return 0;
624 	}
625 
626 	x = xfrm_state_lookup(net, skb->mark, (const xfrm_address_t *)&iph->daddr,
627 			      spi, protocol, AF_INET);
628 	if (!x)
629 		return 0;
630 
631 	xi = xfrmi_lookup(net, x);
632 	if (!xi) {
633 		xfrm_state_put(x);
634 		return -1;
635 	}
636 
637 	if (icmp_hdr(skb)->type == ICMP_DEST_UNREACH)
638 		ipv4_update_pmtu(skb, net, info, 0, protocol);
639 	else
640 		ipv4_redirect(skb, net, 0, protocol);
641 	xfrm_state_put(x);
642 
643 	return 0;
644 }
645 
646 static int xfrmi6_err(struct sk_buff *skb, struct inet6_skb_parm *opt,
647 		    u8 type, u8 code, int offset, __be32 info)
648 {
649 	const struct ipv6hdr *iph = (const struct ipv6hdr *)skb->data;
650 	struct net *net = dev_net(skb->dev);
651 	int protocol = iph->nexthdr;
652 	struct ip_comp_hdr *ipch;
653 	struct ip_esp_hdr *esph;
654 	struct ip_auth_hdr *ah;
655 	struct xfrm_state *x;
656 	struct xfrm_if *xi;
657 	__be32 spi;
658 
659 	switch (protocol) {
660 	case IPPROTO_ESP:
661 		esph = (struct ip_esp_hdr *)(skb->data + offset);
662 		spi = esph->spi;
663 		break;
664 	case IPPROTO_AH:
665 		ah = (struct ip_auth_hdr *)(skb->data + offset);
666 		spi = ah->spi;
667 		break;
668 	case IPPROTO_COMP:
669 		ipch = (struct ip_comp_hdr *)(skb->data + offset);
670 		spi = htonl(ntohs(ipch->cpi));
671 		break;
672 	default:
673 		return 0;
674 	}
675 
676 	if (type != ICMPV6_PKT_TOOBIG &&
677 	    type != NDISC_REDIRECT)
678 		return 0;
679 
680 	x = xfrm_state_lookup(net, skb->mark, (const xfrm_address_t *)&iph->daddr,
681 			      spi, protocol, AF_INET6);
682 	if (!x)
683 		return 0;
684 
685 	xi = xfrmi_lookup(net, x);
686 	if (!xi) {
687 		xfrm_state_put(x);
688 		return -1;
689 	}
690 
691 	if (type == NDISC_REDIRECT)
692 		ip6_redirect(skb, net, skb->dev->ifindex, 0,
693 			     sock_net_uid(net, NULL));
694 	else
695 		ip6_update_pmtu(skb, net, info, 0, 0, sock_net_uid(net, NULL));
696 	xfrm_state_put(x);
697 
698 	return 0;
699 }
700 
701 static int xfrmi_change(struct xfrm_if *xi, const struct xfrm_if_parms *p)
702 {
703 	if (xi->p.link != p->link)
704 		return -EINVAL;
705 
706 	xi->p.if_id = p->if_id;
707 
708 	return 0;
709 }
710 
711 static int xfrmi_update(struct xfrm_if *xi, struct xfrm_if_parms *p)
712 {
713 	struct net *net = xi->net;
714 	struct xfrmi_net *xfrmn = net_generic(net, xfrmi_net_id);
715 	int err;
716 
717 	xfrmi_unlink(xfrmn, xi);
718 	synchronize_net();
719 	err = xfrmi_change(xi, p);
720 	xfrmi_link(xfrmn, xi);
721 	netdev_state_change(xi->dev);
722 	return err;
723 }
724 
725 static int xfrmi_get_iflink(const struct net_device *dev)
726 {
727 	struct xfrm_if *xi = netdev_priv(dev);
728 
729 	return READ_ONCE(xi->p.link);
730 }
731 
732 static const struct net_device_ops xfrmi_netdev_ops = {
733 	.ndo_init	= xfrmi_dev_init,
734 	.ndo_uninit	= xfrmi_dev_uninit,
735 	.ndo_start_xmit = xfrmi_xmit,
736 	.ndo_get_stats64 = dev_get_tstats64,
737 	.ndo_get_iflink = xfrmi_get_iflink,
738 };
739 
740 static void xfrmi_dev_setup(struct net_device *dev)
741 {
742 	dev->netdev_ops 	= &xfrmi_netdev_ops;
743 	dev->header_ops		= &ip_tunnel_header_ops;
744 	dev->type		= ARPHRD_NONE;
745 	dev->mtu		= ETH_DATA_LEN;
746 	dev->min_mtu		= ETH_MIN_MTU;
747 	dev->max_mtu		= IP_MAX_MTU;
748 	dev->flags 		= IFF_NOARP;
749 	dev->needs_free_netdev	= true;
750 	dev->priv_destructor	= xfrmi_dev_free;
751 	dev->pcpu_stat_type	= NETDEV_PCPU_STAT_TSTATS;
752 	netif_keep_dst(dev);
753 
754 	eth_broadcast_addr(dev->broadcast);
755 }
756 
757 #define XFRMI_FEATURES (NETIF_F_SG |		\
758 			NETIF_F_FRAGLIST |	\
759 			NETIF_F_GSO_SOFTWARE |	\
760 			NETIF_F_HW_CSUM)
761 
762 static int xfrmi_dev_init(struct net_device *dev)
763 {
764 	struct xfrm_if *xi = netdev_priv(dev);
765 	struct net_device *phydev = __dev_get_by_index(xi->net, xi->p.link);
766 	int err;
767 
768 	err = gro_cells_init(&xi->gro_cells, dev);
769 	if (err)
770 		return err;
771 
772 	dev->lltx = true;
773 	dev->features |= XFRMI_FEATURES;
774 	dev->hw_features |= XFRMI_FEATURES;
775 
776 	if (phydev) {
777 		dev->needed_headroom = phydev->needed_headroom;
778 		dev->needed_tailroom = phydev->needed_tailroom;
779 
780 		if (is_zero_ether_addr(dev->dev_addr))
781 			eth_hw_addr_inherit(dev, phydev);
782 		if (is_zero_ether_addr(dev->broadcast))
783 			memcpy(dev->broadcast, phydev->broadcast,
784 			       dev->addr_len);
785 	} else {
786 		eth_hw_addr_random(dev);
787 		eth_broadcast_addr(dev->broadcast);
788 	}
789 
790 	return 0;
791 }
792 
793 static int xfrmi_validate(struct nlattr *tb[], struct nlattr *data[],
794 			 struct netlink_ext_ack *extack)
795 {
796 	return 0;
797 }
798 
799 static void xfrmi_netlink_parms(struct nlattr *data[],
800 			       struct xfrm_if_parms *parms)
801 {
802 	memset(parms, 0, sizeof(*parms));
803 
804 	if (!data)
805 		return;
806 
807 	if (data[IFLA_XFRM_LINK])
808 		parms->link = nla_get_u32(data[IFLA_XFRM_LINK]);
809 
810 	if (data[IFLA_XFRM_IF_ID])
811 		parms->if_id = nla_get_u32(data[IFLA_XFRM_IF_ID]);
812 
813 	if (data[IFLA_XFRM_COLLECT_METADATA])
814 		parms->collect_md = true;
815 }
816 
817 static int xfrmi_newlink(struct net *src_net, struct net_device *dev,
818 			struct nlattr *tb[], struct nlattr *data[],
819 			struct netlink_ext_ack *extack)
820 {
821 	struct net *net = dev_net(dev);
822 	struct xfrm_if_parms p = {};
823 	struct xfrm_if *xi;
824 	int err;
825 
826 	xfrmi_netlink_parms(data, &p);
827 	if (p.collect_md) {
828 		struct xfrmi_net *xfrmn = net_generic(net, xfrmi_net_id);
829 
830 		if (p.link || p.if_id) {
831 			NL_SET_ERR_MSG(extack, "link and if_id must be zero");
832 			return -EINVAL;
833 		}
834 
835 		if (rtnl_dereference(xfrmn->collect_md_xfrmi))
836 			return -EEXIST;
837 
838 	} else {
839 		if (!p.if_id) {
840 			NL_SET_ERR_MSG(extack, "if_id must be non zero");
841 			return -EINVAL;
842 		}
843 
844 		xi = xfrmi_locate(net, &p);
845 		if (xi)
846 			return -EEXIST;
847 	}
848 
849 	xi = netdev_priv(dev);
850 	xi->p = p;
851 	xi->net = net;
852 	xi->dev = dev;
853 
854 	err = xfrmi_create(dev);
855 	return err;
856 }
857 
858 static void xfrmi_dellink(struct net_device *dev, struct list_head *head)
859 {
860 	unregister_netdevice_queue(dev, head);
861 }
862 
863 static int xfrmi_changelink(struct net_device *dev, struct nlattr *tb[],
864 			   struct nlattr *data[],
865 			   struct netlink_ext_ack *extack)
866 {
867 	struct xfrm_if *xi = netdev_priv(dev);
868 	struct net *net = xi->net;
869 	struct xfrm_if_parms p = {};
870 
871 	xfrmi_netlink_parms(data, &p);
872 	if (!p.if_id) {
873 		NL_SET_ERR_MSG(extack, "if_id must be non zero");
874 		return -EINVAL;
875 	}
876 
877 	if (p.collect_md) {
878 		NL_SET_ERR_MSG(extack, "collect_md can't be changed");
879 		return -EINVAL;
880 	}
881 
882 	xi = xfrmi_locate(net, &p);
883 	if (!xi) {
884 		xi = netdev_priv(dev);
885 	} else {
886 		if (xi->dev != dev)
887 			return -EEXIST;
888 		if (xi->p.collect_md) {
889 			NL_SET_ERR_MSG(extack,
890 				       "device can't be changed to collect_md");
891 			return -EINVAL;
892 		}
893 	}
894 
895 	return xfrmi_update(xi, &p);
896 }
897 
898 static size_t xfrmi_get_size(const struct net_device *dev)
899 {
900 	return
901 		/* IFLA_XFRM_LINK */
902 		nla_total_size(4) +
903 		/* IFLA_XFRM_IF_ID */
904 		nla_total_size(4) +
905 		/* IFLA_XFRM_COLLECT_METADATA */
906 		nla_total_size(0) +
907 		0;
908 }
909 
910 static int xfrmi_fill_info(struct sk_buff *skb, const struct net_device *dev)
911 {
912 	struct xfrm_if *xi = netdev_priv(dev);
913 	struct xfrm_if_parms *parm = &xi->p;
914 
915 	if (nla_put_u32(skb, IFLA_XFRM_LINK, parm->link) ||
916 	    nla_put_u32(skb, IFLA_XFRM_IF_ID, parm->if_id) ||
917 	    (xi->p.collect_md && nla_put_flag(skb, IFLA_XFRM_COLLECT_METADATA)))
918 		goto nla_put_failure;
919 	return 0;
920 
921 nla_put_failure:
922 	return -EMSGSIZE;
923 }
924 
925 static struct net *xfrmi_get_link_net(const struct net_device *dev)
926 {
927 	struct xfrm_if *xi = netdev_priv(dev);
928 
929 	return READ_ONCE(xi->net);
930 }
931 
932 static const struct nla_policy xfrmi_policy[IFLA_XFRM_MAX + 1] = {
933 	[IFLA_XFRM_UNSPEC]		= { .strict_start_type = IFLA_XFRM_COLLECT_METADATA },
934 	[IFLA_XFRM_LINK]		= { .type = NLA_U32 },
935 	[IFLA_XFRM_IF_ID]		= { .type = NLA_U32 },
936 	[IFLA_XFRM_COLLECT_METADATA]	= { .type = NLA_FLAG },
937 };
938 
939 static struct rtnl_link_ops xfrmi_link_ops __read_mostly = {
940 	.kind		= "xfrm",
941 	.maxtype	= IFLA_XFRM_MAX,
942 	.policy		= xfrmi_policy,
943 	.priv_size	= sizeof(struct xfrm_if),
944 	.setup		= xfrmi_dev_setup,
945 	.validate	= xfrmi_validate,
946 	.newlink	= xfrmi_newlink,
947 	.dellink	= xfrmi_dellink,
948 	.changelink	= xfrmi_changelink,
949 	.get_size	= xfrmi_get_size,
950 	.fill_info	= xfrmi_fill_info,
951 	.get_link_net	= xfrmi_get_link_net,
952 };
953 
954 static void __net_exit xfrmi_exit_batch_rtnl(struct list_head *net_exit_list,
955 					     struct list_head *dev_to_kill)
956 {
957 	struct net *net;
958 
959 	ASSERT_RTNL();
960 	list_for_each_entry(net, net_exit_list, exit_list) {
961 		struct xfrmi_net *xfrmn = net_generic(net, xfrmi_net_id);
962 		struct xfrm_if __rcu **xip;
963 		struct xfrm_if *xi;
964 		int i;
965 
966 		for (i = 0; i < XFRMI_HASH_SIZE; i++) {
967 			for (xip = &xfrmn->xfrmi[i];
968 			     (xi = rtnl_dereference(*xip)) != NULL;
969 			     xip = &xi->next)
970 				unregister_netdevice_queue(xi->dev, dev_to_kill);
971 		}
972 		xi = rtnl_dereference(xfrmn->collect_md_xfrmi);
973 		if (xi)
974 			unregister_netdevice_queue(xi->dev, dev_to_kill);
975 	}
976 }
977 
978 static struct pernet_operations xfrmi_net_ops = {
979 	.exit_batch_rtnl = xfrmi_exit_batch_rtnl,
980 	.id   = &xfrmi_net_id,
981 	.size = sizeof(struct xfrmi_net),
982 };
983 
984 static struct xfrm6_protocol xfrmi_esp6_protocol __read_mostly = {
985 	.handler	=	xfrmi6_rcv,
986 	.input_handler	=	xfrmi6_input,
987 	.cb_handler	=	xfrmi_rcv_cb,
988 	.err_handler	=	xfrmi6_err,
989 	.priority	=	10,
990 };
991 
992 static struct xfrm6_protocol xfrmi_ah6_protocol __read_mostly = {
993 	.handler	=	xfrm6_rcv,
994 	.input_handler	=	xfrm_input,
995 	.cb_handler	=	xfrmi_rcv_cb,
996 	.err_handler	=	xfrmi6_err,
997 	.priority	=	10,
998 };
999 
1000 static struct xfrm6_protocol xfrmi_ipcomp6_protocol __read_mostly = {
1001 	.handler	=	xfrm6_rcv,
1002 	.input_handler	=	xfrm_input,
1003 	.cb_handler	=	xfrmi_rcv_cb,
1004 	.err_handler	=	xfrmi6_err,
1005 	.priority	=	10,
1006 };
1007 
1008 #if IS_REACHABLE(CONFIG_INET6_XFRM_TUNNEL)
1009 static int xfrmi6_rcv_tunnel(struct sk_buff *skb)
1010 {
1011 	const xfrm_address_t *saddr;
1012 	__be32 spi;
1013 
1014 	saddr = (const xfrm_address_t *)&ipv6_hdr(skb)->saddr;
1015 	spi = xfrm6_tunnel_spi_lookup(dev_net(skb->dev), saddr);
1016 
1017 	return xfrm6_rcv_spi(skb, IPPROTO_IPV6, spi, NULL);
1018 }
1019 
1020 static struct xfrm6_tunnel xfrmi_ipv6_handler __read_mostly = {
1021 	.handler	=	xfrmi6_rcv_tunnel,
1022 	.cb_handler	=	xfrmi_rcv_cb,
1023 	.err_handler	=	xfrmi6_err,
1024 	.priority	=	2,
1025 };
1026 
1027 static struct xfrm6_tunnel xfrmi_ip6ip_handler __read_mostly = {
1028 	.handler	=	xfrmi6_rcv_tunnel,
1029 	.cb_handler	=	xfrmi_rcv_cb,
1030 	.err_handler	=	xfrmi6_err,
1031 	.priority	=	2,
1032 };
1033 #endif
1034 
1035 static struct xfrm4_protocol xfrmi_esp4_protocol __read_mostly = {
1036 	.handler	=	xfrmi4_rcv,
1037 	.input_handler	=	xfrmi4_input,
1038 	.cb_handler	=	xfrmi_rcv_cb,
1039 	.err_handler	=	xfrmi4_err,
1040 	.priority	=	10,
1041 };
1042 
1043 static struct xfrm4_protocol xfrmi_ah4_protocol __read_mostly = {
1044 	.handler	=	xfrm4_rcv,
1045 	.input_handler	=	xfrm_input,
1046 	.cb_handler	=	xfrmi_rcv_cb,
1047 	.err_handler	=	xfrmi4_err,
1048 	.priority	=	10,
1049 };
1050 
1051 static struct xfrm4_protocol xfrmi_ipcomp4_protocol __read_mostly = {
1052 	.handler	=	xfrm4_rcv,
1053 	.input_handler	=	xfrm_input,
1054 	.cb_handler	=	xfrmi_rcv_cb,
1055 	.err_handler	=	xfrmi4_err,
1056 	.priority	=	10,
1057 };
1058 
1059 #if IS_REACHABLE(CONFIG_INET_XFRM_TUNNEL)
1060 static int xfrmi4_rcv_tunnel(struct sk_buff *skb)
1061 {
1062 	return xfrm4_rcv_spi(skb, IPPROTO_IPIP, ip_hdr(skb)->saddr);
1063 }
1064 
1065 static struct xfrm_tunnel xfrmi_ipip_handler __read_mostly = {
1066 	.handler	=	xfrmi4_rcv_tunnel,
1067 	.cb_handler	=	xfrmi_rcv_cb,
1068 	.err_handler	=	xfrmi4_err,
1069 	.priority	=	3,
1070 };
1071 
1072 static struct xfrm_tunnel xfrmi_ipip6_handler __read_mostly = {
1073 	.handler	=	xfrmi4_rcv_tunnel,
1074 	.cb_handler	=	xfrmi_rcv_cb,
1075 	.err_handler	=	xfrmi4_err,
1076 	.priority	=	2,
1077 };
1078 #endif
1079 
1080 static int __init xfrmi4_init(void)
1081 {
1082 	int err;
1083 
1084 	err = xfrm4_protocol_register(&xfrmi_esp4_protocol, IPPROTO_ESP);
1085 	if (err < 0)
1086 		goto xfrm_proto_esp_failed;
1087 	err = xfrm4_protocol_register(&xfrmi_ah4_protocol, IPPROTO_AH);
1088 	if (err < 0)
1089 		goto xfrm_proto_ah_failed;
1090 	err = xfrm4_protocol_register(&xfrmi_ipcomp4_protocol, IPPROTO_COMP);
1091 	if (err < 0)
1092 		goto xfrm_proto_comp_failed;
1093 #if IS_REACHABLE(CONFIG_INET_XFRM_TUNNEL)
1094 	err = xfrm4_tunnel_register(&xfrmi_ipip_handler, AF_INET);
1095 	if (err < 0)
1096 		goto xfrm_tunnel_ipip_failed;
1097 	err = xfrm4_tunnel_register(&xfrmi_ipip6_handler, AF_INET6);
1098 	if (err < 0)
1099 		goto xfrm_tunnel_ipip6_failed;
1100 #endif
1101 
1102 	return 0;
1103 
1104 #if IS_REACHABLE(CONFIG_INET_XFRM_TUNNEL)
1105 xfrm_tunnel_ipip6_failed:
1106 	xfrm4_tunnel_deregister(&xfrmi_ipip_handler, AF_INET);
1107 xfrm_tunnel_ipip_failed:
1108 	xfrm4_protocol_deregister(&xfrmi_ipcomp4_protocol, IPPROTO_COMP);
1109 #endif
1110 xfrm_proto_comp_failed:
1111 	xfrm4_protocol_deregister(&xfrmi_ah4_protocol, IPPROTO_AH);
1112 xfrm_proto_ah_failed:
1113 	xfrm4_protocol_deregister(&xfrmi_esp4_protocol, IPPROTO_ESP);
1114 xfrm_proto_esp_failed:
1115 	return err;
1116 }
1117 
1118 static void xfrmi4_fini(void)
1119 {
1120 #if IS_REACHABLE(CONFIG_INET_XFRM_TUNNEL)
1121 	xfrm4_tunnel_deregister(&xfrmi_ipip6_handler, AF_INET6);
1122 	xfrm4_tunnel_deregister(&xfrmi_ipip_handler, AF_INET);
1123 #endif
1124 	xfrm4_protocol_deregister(&xfrmi_ipcomp4_protocol, IPPROTO_COMP);
1125 	xfrm4_protocol_deregister(&xfrmi_ah4_protocol, IPPROTO_AH);
1126 	xfrm4_protocol_deregister(&xfrmi_esp4_protocol, IPPROTO_ESP);
1127 }
1128 
1129 static int __init xfrmi6_init(void)
1130 {
1131 	int err;
1132 
1133 	err = xfrm6_protocol_register(&xfrmi_esp6_protocol, IPPROTO_ESP);
1134 	if (err < 0)
1135 		goto xfrm_proto_esp_failed;
1136 	err = xfrm6_protocol_register(&xfrmi_ah6_protocol, IPPROTO_AH);
1137 	if (err < 0)
1138 		goto xfrm_proto_ah_failed;
1139 	err = xfrm6_protocol_register(&xfrmi_ipcomp6_protocol, IPPROTO_COMP);
1140 	if (err < 0)
1141 		goto xfrm_proto_comp_failed;
1142 #if IS_REACHABLE(CONFIG_INET6_XFRM_TUNNEL)
1143 	err = xfrm6_tunnel_register(&xfrmi_ipv6_handler, AF_INET6);
1144 	if (err < 0)
1145 		goto xfrm_tunnel_ipv6_failed;
1146 	err = xfrm6_tunnel_register(&xfrmi_ip6ip_handler, AF_INET);
1147 	if (err < 0)
1148 		goto xfrm_tunnel_ip6ip_failed;
1149 #endif
1150 
1151 	return 0;
1152 
1153 #if IS_REACHABLE(CONFIG_INET6_XFRM_TUNNEL)
1154 xfrm_tunnel_ip6ip_failed:
1155 	xfrm6_tunnel_deregister(&xfrmi_ipv6_handler, AF_INET6);
1156 xfrm_tunnel_ipv6_failed:
1157 	xfrm6_protocol_deregister(&xfrmi_ipcomp6_protocol, IPPROTO_COMP);
1158 #endif
1159 xfrm_proto_comp_failed:
1160 	xfrm6_protocol_deregister(&xfrmi_ah6_protocol, IPPROTO_AH);
1161 xfrm_proto_ah_failed:
1162 	xfrm6_protocol_deregister(&xfrmi_esp6_protocol, IPPROTO_ESP);
1163 xfrm_proto_esp_failed:
1164 	return err;
1165 }
1166 
1167 static void xfrmi6_fini(void)
1168 {
1169 #if IS_REACHABLE(CONFIG_INET6_XFRM_TUNNEL)
1170 	xfrm6_tunnel_deregister(&xfrmi_ip6ip_handler, AF_INET);
1171 	xfrm6_tunnel_deregister(&xfrmi_ipv6_handler, AF_INET6);
1172 #endif
1173 	xfrm6_protocol_deregister(&xfrmi_ipcomp6_protocol, IPPROTO_COMP);
1174 	xfrm6_protocol_deregister(&xfrmi_ah6_protocol, IPPROTO_AH);
1175 	xfrm6_protocol_deregister(&xfrmi_esp6_protocol, IPPROTO_ESP);
1176 }
1177 
1178 static const struct xfrm_if_cb xfrm_if_cb = {
1179 	.decode_session =	xfrmi_decode_session,
1180 };
1181 
1182 static int __init xfrmi_init(void)
1183 {
1184 	const char *msg;
1185 	int err;
1186 
1187 	pr_info("IPsec XFRM device driver\n");
1188 
1189 	msg = "tunnel device";
1190 	err = register_pernet_device(&xfrmi_net_ops);
1191 	if (err < 0)
1192 		goto pernet_dev_failed;
1193 
1194 	msg = "xfrm4 protocols";
1195 	err = xfrmi4_init();
1196 	if (err < 0)
1197 		goto xfrmi4_failed;
1198 
1199 	msg = "xfrm6 protocols";
1200 	err = xfrmi6_init();
1201 	if (err < 0)
1202 		goto xfrmi6_failed;
1203 
1204 
1205 	msg = "netlink interface";
1206 	err = rtnl_link_register(&xfrmi_link_ops);
1207 	if (err < 0)
1208 		goto rtnl_link_failed;
1209 
1210 	err = register_xfrm_interface_bpf();
1211 	if (err < 0)
1212 		goto kfunc_failed;
1213 
1214 	lwtunnel_encap_add_ops(&xfrmi_encap_ops, LWTUNNEL_ENCAP_XFRM);
1215 
1216 	xfrm_if_register_cb(&xfrm_if_cb);
1217 
1218 	return err;
1219 
1220 kfunc_failed:
1221 	rtnl_link_unregister(&xfrmi_link_ops);
1222 rtnl_link_failed:
1223 	xfrmi6_fini();
1224 xfrmi6_failed:
1225 	xfrmi4_fini();
1226 xfrmi4_failed:
1227 	unregister_pernet_device(&xfrmi_net_ops);
1228 pernet_dev_failed:
1229 	pr_err("xfrmi init: failed to register %s\n", msg);
1230 	return err;
1231 }
1232 
1233 static void __exit xfrmi_fini(void)
1234 {
1235 	xfrm_if_unregister_cb();
1236 	lwtunnel_encap_del_ops(&xfrmi_encap_ops, LWTUNNEL_ENCAP_XFRM);
1237 	rtnl_link_unregister(&xfrmi_link_ops);
1238 	xfrmi4_fini();
1239 	xfrmi6_fini();
1240 	unregister_pernet_device(&xfrmi_net_ops);
1241 }
1242 
1243 module_init(xfrmi_init);
1244 module_exit(xfrmi_fini);
1245 MODULE_LICENSE("GPL");
1246 MODULE_ALIAS_RTNL_LINK("xfrm");
1247 MODULE_ALIAS_NETDEV("xfrm0");
1248 MODULE_AUTHOR("Steffen Klassert");
1249 MODULE_DESCRIPTION("XFRM virtual interface");
1250