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 	free_percpu(dev->tstats);
244 }
245 
246 static int xfrmi_create(struct net_device *dev)
247 {
248 	struct xfrm_if *xi = netdev_priv(dev);
249 	struct net *net = dev_net(dev);
250 	struct xfrmi_net *xfrmn = net_generic(net, xfrmi_net_id);
251 	int err;
252 
253 	dev->rtnl_link_ops = &xfrmi_link_ops;
254 	err = register_netdevice(dev);
255 	if (err < 0)
256 		goto out;
257 
258 	if (xi->p.collect_md)
259 		rcu_assign_pointer(xfrmn->collect_md_xfrmi, xi);
260 	else
261 		xfrmi_link(xfrmn, xi);
262 
263 	return 0;
264 
265 out:
266 	return err;
267 }
268 
269 static struct xfrm_if *xfrmi_locate(struct net *net, struct xfrm_if_parms *p)
270 {
271 	struct xfrm_if __rcu **xip;
272 	struct xfrm_if *xi;
273 	struct xfrmi_net *xfrmn = net_generic(net, xfrmi_net_id);
274 
275 	for (xip = &xfrmn->xfrmi[xfrmi_hash(p->if_id)];
276 	     (xi = rtnl_dereference(*xip)) != NULL;
277 	     xip = &xi->next)
278 		if (xi->p.if_id == p->if_id)
279 			return xi;
280 
281 	return NULL;
282 }
283 
284 static void xfrmi_dev_uninit(struct net_device *dev)
285 {
286 	struct xfrm_if *xi = netdev_priv(dev);
287 	struct xfrmi_net *xfrmn = net_generic(xi->net, xfrmi_net_id);
288 
289 	if (xi->p.collect_md)
290 		RCU_INIT_POINTER(xfrmn->collect_md_xfrmi, NULL);
291 	else
292 		xfrmi_unlink(xfrmn, xi);
293 }
294 
295 static void xfrmi_scrub_packet(struct sk_buff *skb, bool xnet)
296 {
297 	skb_clear_tstamp(skb);
298 	skb->pkt_type = PACKET_HOST;
299 	skb->skb_iif = 0;
300 	skb->ignore_df = 0;
301 	skb_dst_drop(skb);
302 	nf_reset_ct(skb);
303 	nf_reset_trace(skb);
304 
305 	if (!xnet)
306 		return;
307 
308 	ipvs_reset(skb);
309 	secpath_reset(skb);
310 	skb_orphan(skb);
311 	skb->mark = 0;
312 }
313 
314 static int xfrmi_input(struct sk_buff *skb, int nexthdr, __be32 spi,
315 		       int encap_type, unsigned short family)
316 {
317 	struct sec_path *sp;
318 
319 	sp = skb_sec_path(skb);
320 	if (sp && (sp->len || sp->olen) &&
321 	    !xfrm_policy_check(NULL, XFRM_POLICY_IN, skb, family))
322 		goto discard;
323 
324 	XFRM_SPI_SKB_CB(skb)->family = family;
325 	if (family == AF_INET) {
326 		XFRM_SPI_SKB_CB(skb)->daddroff = offsetof(struct iphdr, daddr);
327 		XFRM_TUNNEL_SKB_CB(skb)->tunnel.ip4 = NULL;
328 	} else {
329 		XFRM_SPI_SKB_CB(skb)->daddroff = offsetof(struct ipv6hdr, daddr);
330 		XFRM_TUNNEL_SKB_CB(skb)->tunnel.ip6 = NULL;
331 	}
332 
333 	return xfrm_input(skb, nexthdr, spi, encap_type);
334 discard:
335 	kfree_skb(skb);
336 	return 0;
337 }
338 
339 static int xfrmi4_rcv(struct sk_buff *skb)
340 {
341 	return xfrmi_input(skb, ip_hdr(skb)->protocol, 0, 0, AF_INET);
342 }
343 
344 static int xfrmi6_rcv(struct sk_buff *skb)
345 {
346 	return xfrmi_input(skb, skb_network_header(skb)[IP6CB(skb)->nhoff],
347 			   0, 0, AF_INET6);
348 }
349 
350 static int xfrmi4_input(struct sk_buff *skb, int nexthdr, __be32 spi, int encap_type)
351 {
352 	return xfrmi_input(skb, nexthdr, spi, encap_type, AF_INET);
353 }
354 
355 static int xfrmi6_input(struct sk_buff *skb, int nexthdr, __be32 spi, int encap_type)
356 {
357 	return xfrmi_input(skb, nexthdr, spi, encap_type, AF_INET6);
358 }
359 
360 static int xfrmi_rcv_cb(struct sk_buff *skb, int err)
361 {
362 	const struct xfrm_mode *inner_mode;
363 	struct net_device *dev;
364 	struct xfrm_state *x;
365 	struct xfrm_if *xi;
366 	bool xnet;
367 	int link;
368 
369 	if (err && !secpath_exists(skb))
370 		return 0;
371 
372 	x = xfrm_input_state(skb);
373 
374 	xi = xfrmi_lookup(xs_net(x), x);
375 	if (!xi)
376 		return 1;
377 
378 	link = skb->dev->ifindex;
379 	dev = xi->dev;
380 	skb->dev = dev;
381 
382 	if (err) {
383 		dev->stats.rx_errors++;
384 		dev->stats.rx_dropped++;
385 
386 		return 0;
387 	}
388 
389 	xnet = !net_eq(xi->net, dev_net(skb->dev));
390 
391 	if (xnet) {
392 		inner_mode = &x->inner_mode;
393 
394 		if (x->sel.family == AF_UNSPEC) {
395 			inner_mode = xfrm_ip2inner_mode(x, XFRM_MODE_SKB_CB(skb)->protocol);
396 			if (inner_mode == NULL) {
397 				XFRM_INC_STATS(dev_net(skb->dev),
398 					       LINUX_MIB_XFRMINSTATEMODEERROR);
399 				return -EINVAL;
400 			}
401 		}
402 
403 		if (!xfrm_policy_check(NULL, XFRM_POLICY_IN, skb,
404 				       inner_mode->family))
405 			return -EPERM;
406 	}
407 
408 	xfrmi_scrub_packet(skb, xnet);
409 	if (xi->p.collect_md) {
410 		struct metadata_dst *md_dst;
411 
412 		md_dst = metadata_dst_alloc(0, METADATA_XFRM, GFP_ATOMIC);
413 		if (!md_dst)
414 			return -ENOMEM;
415 
416 		md_dst->u.xfrm_info.if_id = x->if_id;
417 		md_dst->u.xfrm_info.link = link;
418 		skb_dst_set(skb, (struct dst_entry *)md_dst);
419 	}
420 	dev_sw_netstats_rx_add(dev, skb->len);
421 
422 	return 0;
423 }
424 
425 static int
426 xfrmi_xmit2(struct sk_buff *skb, struct net_device *dev, struct flowi *fl)
427 {
428 	struct xfrm_if *xi = netdev_priv(dev);
429 	struct net_device_stats *stats = &xi->dev->stats;
430 	struct dst_entry *dst = skb_dst(skb);
431 	unsigned int length = skb->len;
432 	struct net_device *tdev;
433 	struct xfrm_state *x;
434 	int err = -1;
435 	u32 if_id;
436 	int mtu;
437 
438 	if (xi->p.collect_md) {
439 		struct xfrm_md_info *md_info = skb_xfrm_md_info(skb);
440 
441 		if (unlikely(!md_info))
442 			return -EINVAL;
443 
444 		if_id = md_info->if_id;
445 		fl->flowi_oif = md_info->link;
446 		if (md_info->dst_orig) {
447 			struct dst_entry *tmp_dst = dst;
448 
449 			dst = md_info->dst_orig;
450 			skb_dst_set(skb, dst);
451 			md_info->dst_orig = NULL;
452 			dst_release(tmp_dst);
453 		}
454 	} else {
455 		if_id = xi->p.if_id;
456 	}
457 
458 	dst_hold(dst);
459 	dst = xfrm_lookup_with_ifid(xi->net, dst, fl, NULL, 0, if_id);
460 	if (IS_ERR(dst)) {
461 		err = PTR_ERR(dst);
462 		dst = NULL;
463 		goto tx_err_link_failure;
464 	}
465 
466 	x = dst->xfrm;
467 	if (!x)
468 		goto tx_err_link_failure;
469 
470 	if (x->if_id != if_id)
471 		goto tx_err_link_failure;
472 
473 	tdev = dst->dev;
474 
475 	if (tdev == dev) {
476 		stats->collisions++;
477 		net_warn_ratelimited("%s: Local routing loop detected!\n",
478 				     dev->name);
479 		goto tx_err_dst_release;
480 	}
481 
482 	mtu = dst_mtu(dst);
483 	if ((!skb_is_gso(skb) && skb->len > mtu) ||
484 	    (skb_is_gso(skb) && !skb_gso_validate_network_len(skb, mtu))) {
485 		skb_dst_update_pmtu_no_confirm(skb, mtu);
486 
487 		if (skb->protocol == htons(ETH_P_IPV6)) {
488 			if (mtu < IPV6_MIN_MTU)
489 				mtu = IPV6_MIN_MTU;
490 
491 			if (skb->len > 1280)
492 				icmpv6_ndo_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu);
493 			else
494 				goto xmit;
495 		} else {
496 			if (!(ip_hdr(skb)->frag_off & htons(IP_DF)))
497 				goto xmit;
498 			icmp_ndo_send(skb, ICMP_DEST_UNREACH, ICMP_FRAG_NEEDED,
499 				      htonl(mtu));
500 		}
501 
502 		dst_release(dst);
503 		return -EMSGSIZE;
504 	}
505 
506 xmit:
507 	xfrmi_scrub_packet(skb, !net_eq(xi->net, dev_net(dev)));
508 	skb_dst_set(skb, dst);
509 	skb->dev = tdev;
510 
511 	err = dst_output(xi->net, skb->sk, skb);
512 	if (net_xmit_eval(err) == 0) {
513 		dev_sw_netstats_tx_add(dev, 1, length);
514 	} else {
515 		stats->tx_errors++;
516 		stats->tx_aborted_errors++;
517 	}
518 
519 	return 0;
520 tx_err_link_failure:
521 	stats->tx_carrier_errors++;
522 	dst_link_failure(skb);
523 tx_err_dst_release:
524 	dst_release(dst);
525 	return err;
526 }
527 
528 static netdev_tx_t xfrmi_xmit(struct sk_buff *skb, struct net_device *dev)
529 {
530 	struct xfrm_if *xi = netdev_priv(dev);
531 	struct net_device_stats *stats = &xi->dev->stats;
532 	struct dst_entry *dst = skb_dst(skb);
533 	struct flowi fl;
534 	int ret;
535 
536 	memset(&fl, 0, sizeof(fl));
537 
538 	switch (skb->protocol) {
539 	case htons(ETH_P_IPV6):
540 		xfrm_decode_session(skb, &fl, AF_INET6);
541 		memset(IP6CB(skb), 0, sizeof(*IP6CB(skb)));
542 		if (!dst) {
543 			fl.u.ip6.flowi6_oif = dev->ifindex;
544 			fl.u.ip6.flowi6_flags |= FLOWI_FLAG_ANYSRC;
545 			dst = ip6_route_output(dev_net(dev), NULL, &fl.u.ip6);
546 			if (dst->error) {
547 				dst_release(dst);
548 				stats->tx_carrier_errors++;
549 				goto tx_err;
550 			}
551 			skb_dst_set(skb, dst);
552 		}
553 		break;
554 	case htons(ETH_P_IP):
555 		xfrm_decode_session(skb, &fl, AF_INET);
556 		memset(IPCB(skb), 0, sizeof(*IPCB(skb)));
557 		if (!dst) {
558 			struct rtable *rt;
559 
560 			fl.u.ip4.flowi4_oif = dev->ifindex;
561 			fl.u.ip4.flowi4_flags |= FLOWI_FLAG_ANYSRC;
562 			rt = __ip_route_output_key(dev_net(dev), &fl.u.ip4);
563 			if (IS_ERR(rt)) {
564 				stats->tx_carrier_errors++;
565 				goto tx_err;
566 			}
567 			skb_dst_set(skb, &rt->dst);
568 		}
569 		break;
570 	default:
571 		goto tx_err;
572 	}
573 
574 	fl.flowi_oif = xi->p.link;
575 
576 	ret = xfrmi_xmit2(skb, dev, &fl);
577 	if (ret < 0)
578 		goto tx_err;
579 
580 	return NETDEV_TX_OK;
581 
582 tx_err:
583 	stats->tx_errors++;
584 	stats->tx_dropped++;
585 	kfree_skb(skb);
586 	return NETDEV_TX_OK;
587 }
588 
589 static int xfrmi4_err(struct sk_buff *skb, u32 info)
590 {
591 	const struct iphdr *iph = (const struct iphdr *)skb->data;
592 	struct net *net = dev_net(skb->dev);
593 	int protocol = iph->protocol;
594 	struct ip_comp_hdr *ipch;
595 	struct ip_esp_hdr *esph;
596 	struct ip_auth_hdr *ah ;
597 	struct xfrm_state *x;
598 	struct xfrm_if *xi;
599 	__be32 spi;
600 
601 	switch (protocol) {
602 	case IPPROTO_ESP:
603 		esph = (struct ip_esp_hdr *)(skb->data+(iph->ihl<<2));
604 		spi = esph->spi;
605 		break;
606 	case IPPROTO_AH:
607 		ah = (struct ip_auth_hdr *)(skb->data+(iph->ihl<<2));
608 		spi = ah->spi;
609 		break;
610 	case IPPROTO_COMP:
611 		ipch = (struct ip_comp_hdr *)(skb->data+(iph->ihl<<2));
612 		spi = htonl(ntohs(ipch->cpi));
613 		break;
614 	default:
615 		return 0;
616 	}
617 
618 	switch (icmp_hdr(skb)->type) {
619 	case ICMP_DEST_UNREACH:
620 		if (icmp_hdr(skb)->code != ICMP_FRAG_NEEDED)
621 			return 0;
622 		break;
623 	case ICMP_REDIRECT:
624 		break;
625 	default:
626 		return 0;
627 	}
628 
629 	x = xfrm_state_lookup(net, skb->mark, (const xfrm_address_t *)&iph->daddr,
630 			      spi, protocol, AF_INET);
631 	if (!x)
632 		return 0;
633 
634 	xi = xfrmi_lookup(net, x);
635 	if (!xi) {
636 		xfrm_state_put(x);
637 		return -1;
638 	}
639 
640 	if (icmp_hdr(skb)->type == ICMP_DEST_UNREACH)
641 		ipv4_update_pmtu(skb, net, info, 0, protocol);
642 	else
643 		ipv4_redirect(skb, net, 0, protocol);
644 	xfrm_state_put(x);
645 
646 	return 0;
647 }
648 
649 static int xfrmi6_err(struct sk_buff *skb, struct inet6_skb_parm *opt,
650 		    u8 type, u8 code, int offset, __be32 info)
651 {
652 	const struct ipv6hdr *iph = (const struct ipv6hdr *)skb->data;
653 	struct net *net = dev_net(skb->dev);
654 	int protocol = iph->nexthdr;
655 	struct ip_comp_hdr *ipch;
656 	struct ip_esp_hdr *esph;
657 	struct ip_auth_hdr *ah;
658 	struct xfrm_state *x;
659 	struct xfrm_if *xi;
660 	__be32 spi;
661 
662 	switch (protocol) {
663 	case IPPROTO_ESP:
664 		esph = (struct ip_esp_hdr *)(skb->data + offset);
665 		spi = esph->spi;
666 		break;
667 	case IPPROTO_AH:
668 		ah = (struct ip_auth_hdr *)(skb->data + offset);
669 		spi = ah->spi;
670 		break;
671 	case IPPROTO_COMP:
672 		ipch = (struct ip_comp_hdr *)(skb->data + offset);
673 		spi = htonl(ntohs(ipch->cpi));
674 		break;
675 	default:
676 		return 0;
677 	}
678 
679 	if (type != ICMPV6_PKT_TOOBIG &&
680 	    type != NDISC_REDIRECT)
681 		return 0;
682 
683 	x = xfrm_state_lookup(net, skb->mark, (const xfrm_address_t *)&iph->daddr,
684 			      spi, protocol, AF_INET6);
685 	if (!x)
686 		return 0;
687 
688 	xi = xfrmi_lookup(net, x);
689 	if (!xi) {
690 		xfrm_state_put(x);
691 		return -1;
692 	}
693 
694 	if (type == NDISC_REDIRECT)
695 		ip6_redirect(skb, net, skb->dev->ifindex, 0,
696 			     sock_net_uid(net, NULL));
697 	else
698 		ip6_update_pmtu(skb, net, info, 0, 0, sock_net_uid(net, NULL));
699 	xfrm_state_put(x);
700 
701 	return 0;
702 }
703 
704 static int xfrmi_change(struct xfrm_if *xi, const struct xfrm_if_parms *p)
705 {
706 	if (xi->p.link != p->link)
707 		return -EINVAL;
708 
709 	xi->p.if_id = p->if_id;
710 
711 	return 0;
712 }
713 
714 static int xfrmi_update(struct xfrm_if *xi, struct xfrm_if_parms *p)
715 {
716 	struct net *net = xi->net;
717 	struct xfrmi_net *xfrmn = net_generic(net, xfrmi_net_id);
718 	int err;
719 
720 	xfrmi_unlink(xfrmn, xi);
721 	synchronize_net();
722 	err = xfrmi_change(xi, p);
723 	xfrmi_link(xfrmn, xi);
724 	netdev_state_change(xi->dev);
725 	return err;
726 }
727 
728 static int xfrmi_get_iflink(const struct net_device *dev)
729 {
730 	struct xfrm_if *xi = netdev_priv(dev);
731 
732 	return xi->p.link;
733 }
734 
735 static const struct net_device_ops xfrmi_netdev_ops = {
736 	.ndo_init	= xfrmi_dev_init,
737 	.ndo_uninit	= xfrmi_dev_uninit,
738 	.ndo_start_xmit = xfrmi_xmit,
739 	.ndo_get_stats64 = dev_get_tstats64,
740 	.ndo_get_iflink = xfrmi_get_iflink,
741 };
742 
743 static void xfrmi_dev_setup(struct net_device *dev)
744 {
745 	dev->netdev_ops 	= &xfrmi_netdev_ops;
746 	dev->header_ops		= &ip_tunnel_header_ops;
747 	dev->type		= ARPHRD_NONE;
748 	dev->mtu		= ETH_DATA_LEN;
749 	dev->min_mtu		= ETH_MIN_MTU;
750 	dev->max_mtu		= IP_MAX_MTU;
751 	dev->flags 		= IFF_NOARP;
752 	dev->needs_free_netdev	= true;
753 	dev->priv_destructor	= xfrmi_dev_free;
754 	netif_keep_dst(dev);
755 
756 	eth_broadcast_addr(dev->broadcast);
757 }
758 
759 #define XFRMI_FEATURES (NETIF_F_SG |		\
760 			NETIF_F_FRAGLIST |	\
761 			NETIF_F_GSO_SOFTWARE |	\
762 			NETIF_F_HW_CSUM)
763 
764 static int xfrmi_dev_init(struct net_device *dev)
765 {
766 	struct xfrm_if *xi = netdev_priv(dev);
767 	struct net_device *phydev = __dev_get_by_index(xi->net, xi->p.link);
768 	int err;
769 
770 	dev->tstats = netdev_alloc_pcpu_stats(struct pcpu_sw_netstats);
771 	if (!dev->tstats)
772 		return -ENOMEM;
773 
774 	err = gro_cells_init(&xi->gro_cells, dev);
775 	if (err) {
776 		free_percpu(dev->tstats);
777 		return err;
778 	}
779 
780 	dev->features |= NETIF_F_LLTX;
781 	dev->features |= XFRMI_FEATURES;
782 	dev->hw_features |= XFRMI_FEATURES;
783 
784 	if (phydev) {
785 		dev->needed_headroom = phydev->needed_headroom;
786 		dev->needed_tailroom = phydev->needed_tailroom;
787 
788 		if (is_zero_ether_addr(dev->dev_addr))
789 			eth_hw_addr_inherit(dev, phydev);
790 		if (is_zero_ether_addr(dev->broadcast))
791 			memcpy(dev->broadcast, phydev->broadcast,
792 			       dev->addr_len);
793 	} else {
794 		eth_hw_addr_random(dev);
795 		eth_broadcast_addr(dev->broadcast);
796 	}
797 
798 	return 0;
799 }
800 
801 static int xfrmi_validate(struct nlattr *tb[], struct nlattr *data[],
802 			 struct netlink_ext_ack *extack)
803 {
804 	return 0;
805 }
806 
807 static void xfrmi_netlink_parms(struct nlattr *data[],
808 			       struct xfrm_if_parms *parms)
809 {
810 	memset(parms, 0, sizeof(*parms));
811 
812 	if (!data)
813 		return;
814 
815 	if (data[IFLA_XFRM_LINK])
816 		parms->link = nla_get_u32(data[IFLA_XFRM_LINK]);
817 
818 	if (data[IFLA_XFRM_IF_ID])
819 		parms->if_id = nla_get_u32(data[IFLA_XFRM_IF_ID]);
820 
821 	if (data[IFLA_XFRM_COLLECT_METADATA])
822 		parms->collect_md = true;
823 }
824 
825 static int xfrmi_newlink(struct net *src_net, struct net_device *dev,
826 			struct nlattr *tb[], struct nlattr *data[],
827 			struct netlink_ext_ack *extack)
828 {
829 	struct net *net = dev_net(dev);
830 	struct xfrm_if_parms p = {};
831 	struct xfrm_if *xi;
832 	int err;
833 
834 	xfrmi_netlink_parms(data, &p);
835 	if (p.collect_md) {
836 		struct xfrmi_net *xfrmn = net_generic(net, xfrmi_net_id);
837 
838 		if (p.link || p.if_id) {
839 			NL_SET_ERR_MSG(extack, "link and if_id must be zero");
840 			return -EINVAL;
841 		}
842 
843 		if (rtnl_dereference(xfrmn->collect_md_xfrmi))
844 			return -EEXIST;
845 
846 	} else {
847 		if (!p.if_id) {
848 			NL_SET_ERR_MSG(extack, "if_id must be non zero");
849 			return -EINVAL;
850 		}
851 
852 		xi = xfrmi_locate(net, &p);
853 		if (xi)
854 			return -EEXIST;
855 	}
856 
857 	xi = netdev_priv(dev);
858 	xi->p = p;
859 	xi->net = net;
860 	xi->dev = dev;
861 
862 	err = xfrmi_create(dev);
863 	return err;
864 }
865 
866 static void xfrmi_dellink(struct net_device *dev, struct list_head *head)
867 {
868 	unregister_netdevice_queue(dev, head);
869 }
870 
871 static int xfrmi_changelink(struct net_device *dev, struct nlattr *tb[],
872 			   struct nlattr *data[],
873 			   struct netlink_ext_ack *extack)
874 {
875 	struct xfrm_if *xi = netdev_priv(dev);
876 	struct net *net = xi->net;
877 	struct xfrm_if_parms p = {};
878 
879 	xfrmi_netlink_parms(data, &p);
880 	if (!p.if_id) {
881 		NL_SET_ERR_MSG(extack, "if_id must be non zero");
882 		return -EINVAL;
883 	}
884 
885 	if (p.collect_md) {
886 		NL_SET_ERR_MSG(extack, "collect_md can't be changed");
887 		return -EINVAL;
888 	}
889 
890 	xi = xfrmi_locate(net, &p);
891 	if (!xi) {
892 		xi = netdev_priv(dev);
893 	} else {
894 		if (xi->dev != dev)
895 			return -EEXIST;
896 		if (xi->p.collect_md) {
897 			NL_SET_ERR_MSG(extack,
898 				       "device can't be changed to collect_md");
899 			return -EINVAL;
900 		}
901 	}
902 
903 	return xfrmi_update(xi, &p);
904 }
905 
906 static size_t xfrmi_get_size(const struct net_device *dev)
907 {
908 	return
909 		/* IFLA_XFRM_LINK */
910 		nla_total_size(4) +
911 		/* IFLA_XFRM_IF_ID */
912 		nla_total_size(4) +
913 		/* IFLA_XFRM_COLLECT_METADATA */
914 		nla_total_size(0) +
915 		0;
916 }
917 
918 static int xfrmi_fill_info(struct sk_buff *skb, const struct net_device *dev)
919 {
920 	struct xfrm_if *xi = netdev_priv(dev);
921 	struct xfrm_if_parms *parm = &xi->p;
922 
923 	if (nla_put_u32(skb, IFLA_XFRM_LINK, parm->link) ||
924 	    nla_put_u32(skb, IFLA_XFRM_IF_ID, parm->if_id) ||
925 	    (xi->p.collect_md && nla_put_flag(skb, IFLA_XFRM_COLLECT_METADATA)))
926 		goto nla_put_failure;
927 	return 0;
928 
929 nla_put_failure:
930 	return -EMSGSIZE;
931 }
932 
933 static struct net *xfrmi_get_link_net(const struct net_device *dev)
934 {
935 	struct xfrm_if *xi = netdev_priv(dev);
936 
937 	return xi->net;
938 }
939 
940 static const struct nla_policy xfrmi_policy[IFLA_XFRM_MAX + 1] = {
941 	[IFLA_XFRM_UNSPEC]		= { .strict_start_type = IFLA_XFRM_COLLECT_METADATA },
942 	[IFLA_XFRM_LINK]		= { .type = NLA_U32 },
943 	[IFLA_XFRM_IF_ID]		= { .type = NLA_U32 },
944 	[IFLA_XFRM_COLLECT_METADATA]	= { .type = NLA_FLAG },
945 };
946 
947 static struct rtnl_link_ops xfrmi_link_ops __read_mostly = {
948 	.kind		= "xfrm",
949 	.maxtype	= IFLA_XFRM_MAX,
950 	.policy		= xfrmi_policy,
951 	.priv_size	= sizeof(struct xfrm_if),
952 	.setup		= xfrmi_dev_setup,
953 	.validate	= xfrmi_validate,
954 	.newlink	= xfrmi_newlink,
955 	.dellink	= xfrmi_dellink,
956 	.changelink	= xfrmi_changelink,
957 	.get_size	= xfrmi_get_size,
958 	.fill_info	= xfrmi_fill_info,
959 	.get_link_net	= xfrmi_get_link_net,
960 };
961 
962 static void __net_exit xfrmi_exit_batch_net(struct list_head *net_exit_list)
963 {
964 	struct net *net;
965 	LIST_HEAD(list);
966 
967 	rtnl_lock();
968 	list_for_each_entry(net, net_exit_list, exit_list) {
969 		struct xfrmi_net *xfrmn = net_generic(net, xfrmi_net_id);
970 		struct xfrm_if __rcu **xip;
971 		struct xfrm_if *xi;
972 		int i;
973 
974 		for (i = 0; i < XFRMI_HASH_SIZE; i++) {
975 			for (xip = &xfrmn->xfrmi[i];
976 			     (xi = rtnl_dereference(*xip)) != NULL;
977 			     xip = &xi->next)
978 				unregister_netdevice_queue(xi->dev, &list);
979 		}
980 		xi = rtnl_dereference(xfrmn->collect_md_xfrmi);
981 		if (xi)
982 			unregister_netdevice_queue(xi->dev, &list);
983 	}
984 	unregister_netdevice_many(&list);
985 	rtnl_unlock();
986 }
987 
988 static struct pernet_operations xfrmi_net_ops = {
989 	.exit_batch = xfrmi_exit_batch_net,
990 	.id   = &xfrmi_net_id,
991 	.size = sizeof(struct xfrmi_net),
992 };
993 
994 static struct xfrm6_protocol xfrmi_esp6_protocol __read_mostly = {
995 	.handler	=	xfrmi6_rcv,
996 	.input_handler	=	xfrmi6_input,
997 	.cb_handler	=	xfrmi_rcv_cb,
998 	.err_handler	=	xfrmi6_err,
999 	.priority	=	10,
1000 };
1001 
1002 static struct xfrm6_protocol xfrmi_ah6_protocol __read_mostly = {
1003 	.handler	=	xfrm6_rcv,
1004 	.input_handler	=	xfrm_input,
1005 	.cb_handler	=	xfrmi_rcv_cb,
1006 	.err_handler	=	xfrmi6_err,
1007 	.priority	=	10,
1008 };
1009 
1010 static struct xfrm6_protocol xfrmi_ipcomp6_protocol __read_mostly = {
1011 	.handler	=	xfrm6_rcv,
1012 	.input_handler	=	xfrm_input,
1013 	.cb_handler	=	xfrmi_rcv_cb,
1014 	.err_handler	=	xfrmi6_err,
1015 	.priority	=	10,
1016 };
1017 
1018 #if IS_REACHABLE(CONFIG_INET6_XFRM_TUNNEL)
1019 static int xfrmi6_rcv_tunnel(struct sk_buff *skb)
1020 {
1021 	const xfrm_address_t *saddr;
1022 	__be32 spi;
1023 
1024 	saddr = (const xfrm_address_t *)&ipv6_hdr(skb)->saddr;
1025 	spi = xfrm6_tunnel_spi_lookup(dev_net(skb->dev), saddr);
1026 
1027 	return xfrm6_rcv_spi(skb, IPPROTO_IPV6, spi, NULL);
1028 }
1029 
1030 static struct xfrm6_tunnel xfrmi_ipv6_handler __read_mostly = {
1031 	.handler	=	xfrmi6_rcv_tunnel,
1032 	.cb_handler	=	xfrmi_rcv_cb,
1033 	.err_handler	=	xfrmi6_err,
1034 	.priority	=	2,
1035 };
1036 
1037 static struct xfrm6_tunnel xfrmi_ip6ip_handler __read_mostly = {
1038 	.handler	=	xfrmi6_rcv_tunnel,
1039 	.cb_handler	=	xfrmi_rcv_cb,
1040 	.err_handler	=	xfrmi6_err,
1041 	.priority	=	2,
1042 };
1043 #endif
1044 
1045 static struct xfrm4_protocol xfrmi_esp4_protocol __read_mostly = {
1046 	.handler	=	xfrmi4_rcv,
1047 	.input_handler	=	xfrmi4_input,
1048 	.cb_handler	=	xfrmi_rcv_cb,
1049 	.err_handler	=	xfrmi4_err,
1050 	.priority	=	10,
1051 };
1052 
1053 static struct xfrm4_protocol xfrmi_ah4_protocol __read_mostly = {
1054 	.handler	=	xfrm4_rcv,
1055 	.input_handler	=	xfrm_input,
1056 	.cb_handler	=	xfrmi_rcv_cb,
1057 	.err_handler	=	xfrmi4_err,
1058 	.priority	=	10,
1059 };
1060 
1061 static struct xfrm4_protocol xfrmi_ipcomp4_protocol __read_mostly = {
1062 	.handler	=	xfrm4_rcv,
1063 	.input_handler	=	xfrm_input,
1064 	.cb_handler	=	xfrmi_rcv_cb,
1065 	.err_handler	=	xfrmi4_err,
1066 	.priority	=	10,
1067 };
1068 
1069 #if IS_REACHABLE(CONFIG_INET_XFRM_TUNNEL)
1070 static int xfrmi4_rcv_tunnel(struct sk_buff *skb)
1071 {
1072 	return xfrm4_rcv_spi(skb, IPPROTO_IPIP, ip_hdr(skb)->saddr);
1073 }
1074 
1075 static struct xfrm_tunnel xfrmi_ipip_handler __read_mostly = {
1076 	.handler	=	xfrmi4_rcv_tunnel,
1077 	.cb_handler	=	xfrmi_rcv_cb,
1078 	.err_handler	=	xfrmi4_err,
1079 	.priority	=	3,
1080 };
1081 
1082 static struct xfrm_tunnel xfrmi_ipip6_handler __read_mostly = {
1083 	.handler	=	xfrmi4_rcv_tunnel,
1084 	.cb_handler	=	xfrmi_rcv_cb,
1085 	.err_handler	=	xfrmi4_err,
1086 	.priority	=	2,
1087 };
1088 #endif
1089 
1090 static int __init xfrmi4_init(void)
1091 {
1092 	int err;
1093 
1094 	err = xfrm4_protocol_register(&xfrmi_esp4_protocol, IPPROTO_ESP);
1095 	if (err < 0)
1096 		goto xfrm_proto_esp_failed;
1097 	err = xfrm4_protocol_register(&xfrmi_ah4_protocol, IPPROTO_AH);
1098 	if (err < 0)
1099 		goto xfrm_proto_ah_failed;
1100 	err = xfrm4_protocol_register(&xfrmi_ipcomp4_protocol, IPPROTO_COMP);
1101 	if (err < 0)
1102 		goto xfrm_proto_comp_failed;
1103 #if IS_REACHABLE(CONFIG_INET_XFRM_TUNNEL)
1104 	err = xfrm4_tunnel_register(&xfrmi_ipip_handler, AF_INET);
1105 	if (err < 0)
1106 		goto xfrm_tunnel_ipip_failed;
1107 	err = xfrm4_tunnel_register(&xfrmi_ipip6_handler, AF_INET6);
1108 	if (err < 0)
1109 		goto xfrm_tunnel_ipip6_failed;
1110 #endif
1111 
1112 	return 0;
1113 
1114 #if IS_REACHABLE(CONFIG_INET_XFRM_TUNNEL)
1115 xfrm_tunnel_ipip6_failed:
1116 	xfrm4_tunnel_deregister(&xfrmi_ipip_handler, AF_INET);
1117 xfrm_tunnel_ipip_failed:
1118 	xfrm4_protocol_deregister(&xfrmi_ipcomp4_protocol, IPPROTO_COMP);
1119 #endif
1120 xfrm_proto_comp_failed:
1121 	xfrm4_protocol_deregister(&xfrmi_ah4_protocol, IPPROTO_AH);
1122 xfrm_proto_ah_failed:
1123 	xfrm4_protocol_deregister(&xfrmi_esp4_protocol, IPPROTO_ESP);
1124 xfrm_proto_esp_failed:
1125 	return err;
1126 }
1127 
1128 static void xfrmi4_fini(void)
1129 {
1130 #if IS_REACHABLE(CONFIG_INET_XFRM_TUNNEL)
1131 	xfrm4_tunnel_deregister(&xfrmi_ipip6_handler, AF_INET6);
1132 	xfrm4_tunnel_deregister(&xfrmi_ipip_handler, AF_INET);
1133 #endif
1134 	xfrm4_protocol_deregister(&xfrmi_ipcomp4_protocol, IPPROTO_COMP);
1135 	xfrm4_protocol_deregister(&xfrmi_ah4_protocol, IPPROTO_AH);
1136 	xfrm4_protocol_deregister(&xfrmi_esp4_protocol, IPPROTO_ESP);
1137 }
1138 
1139 static int __init xfrmi6_init(void)
1140 {
1141 	int err;
1142 
1143 	err = xfrm6_protocol_register(&xfrmi_esp6_protocol, IPPROTO_ESP);
1144 	if (err < 0)
1145 		goto xfrm_proto_esp_failed;
1146 	err = xfrm6_protocol_register(&xfrmi_ah6_protocol, IPPROTO_AH);
1147 	if (err < 0)
1148 		goto xfrm_proto_ah_failed;
1149 	err = xfrm6_protocol_register(&xfrmi_ipcomp6_protocol, IPPROTO_COMP);
1150 	if (err < 0)
1151 		goto xfrm_proto_comp_failed;
1152 #if IS_REACHABLE(CONFIG_INET6_XFRM_TUNNEL)
1153 	err = xfrm6_tunnel_register(&xfrmi_ipv6_handler, AF_INET6);
1154 	if (err < 0)
1155 		goto xfrm_tunnel_ipv6_failed;
1156 	err = xfrm6_tunnel_register(&xfrmi_ip6ip_handler, AF_INET);
1157 	if (err < 0)
1158 		goto xfrm_tunnel_ip6ip_failed;
1159 #endif
1160 
1161 	return 0;
1162 
1163 #if IS_REACHABLE(CONFIG_INET6_XFRM_TUNNEL)
1164 xfrm_tunnel_ip6ip_failed:
1165 	xfrm6_tunnel_deregister(&xfrmi_ipv6_handler, AF_INET6);
1166 xfrm_tunnel_ipv6_failed:
1167 	xfrm6_protocol_deregister(&xfrmi_ipcomp6_protocol, IPPROTO_COMP);
1168 #endif
1169 xfrm_proto_comp_failed:
1170 	xfrm6_protocol_deregister(&xfrmi_ah6_protocol, IPPROTO_AH);
1171 xfrm_proto_ah_failed:
1172 	xfrm6_protocol_deregister(&xfrmi_esp6_protocol, IPPROTO_ESP);
1173 xfrm_proto_esp_failed:
1174 	return err;
1175 }
1176 
1177 static void xfrmi6_fini(void)
1178 {
1179 #if IS_REACHABLE(CONFIG_INET6_XFRM_TUNNEL)
1180 	xfrm6_tunnel_deregister(&xfrmi_ip6ip_handler, AF_INET);
1181 	xfrm6_tunnel_deregister(&xfrmi_ipv6_handler, AF_INET6);
1182 #endif
1183 	xfrm6_protocol_deregister(&xfrmi_ipcomp6_protocol, IPPROTO_COMP);
1184 	xfrm6_protocol_deregister(&xfrmi_ah6_protocol, IPPROTO_AH);
1185 	xfrm6_protocol_deregister(&xfrmi_esp6_protocol, IPPROTO_ESP);
1186 }
1187 
1188 static const struct xfrm_if_cb xfrm_if_cb = {
1189 	.decode_session =	xfrmi_decode_session,
1190 };
1191 
1192 static int __init xfrmi_init(void)
1193 {
1194 	const char *msg;
1195 	int err;
1196 
1197 	pr_info("IPsec XFRM device driver\n");
1198 
1199 	msg = "tunnel device";
1200 	err = register_pernet_device(&xfrmi_net_ops);
1201 	if (err < 0)
1202 		goto pernet_dev_failed;
1203 
1204 	msg = "xfrm4 protocols";
1205 	err = xfrmi4_init();
1206 	if (err < 0)
1207 		goto xfrmi4_failed;
1208 
1209 	msg = "xfrm6 protocols";
1210 	err = xfrmi6_init();
1211 	if (err < 0)
1212 		goto xfrmi6_failed;
1213 
1214 
1215 	msg = "netlink interface";
1216 	err = rtnl_link_register(&xfrmi_link_ops);
1217 	if (err < 0)
1218 		goto rtnl_link_failed;
1219 
1220 	err = register_xfrm_interface_bpf();
1221 	if (err < 0)
1222 		goto kfunc_failed;
1223 
1224 	lwtunnel_encap_add_ops(&xfrmi_encap_ops, LWTUNNEL_ENCAP_XFRM);
1225 
1226 	xfrm_if_register_cb(&xfrm_if_cb);
1227 
1228 	return err;
1229 
1230 kfunc_failed:
1231 	rtnl_link_unregister(&xfrmi_link_ops);
1232 rtnl_link_failed:
1233 	xfrmi6_fini();
1234 xfrmi6_failed:
1235 	xfrmi4_fini();
1236 xfrmi4_failed:
1237 	unregister_pernet_device(&xfrmi_net_ops);
1238 pernet_dev_failed:
1239 	pr_err("xfrmi init: failed to register %s\n", msg);
1240 	return err;
1241 }
1242 
1243 static void __exit xfrmi_fini(void)
1244 {
1245 	xfrm_if_unregister_cb();
1246 	lwtunnel_encap_del_ops(&xfrmi_encap_ops, LWTUNNEL_ENCAP_XFRM);
1247 	rtnl_link_unregister(&xfrmi_link_ops);
1248 	xfrmi4_fini();
1249 	xfrmi6_fini();
1250 	unregister_pernet_device(&xfrmi_net_ops);
1251 }
1252 
1253 module_init(xfrmi_init);
1254 module_exit(xfrmi_fini);
1255 MODULE_LICENSE("GPL");
1256 MODULE_ALIAS_RTNL_LINK("xfrm");
1257 MODULE_ALIAS_NETDEV("xfrm0");
1258 MODULE_AUTHOR("Steffen Klassert");
1259 MODULE_DESCRIPTION("XFRM virtual interface");
1260