xref: /openbmc/linux/net/xfrm/xfrm_input.c (revision dd21bfa4)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * xfrm_input.c
4  *
5  * Changes:
6  * 	YOSHIFUJI Hideaki @USAGI
7  * 		Split up af-specific portion
8  *
9  */
10 
11 #include <linux/bottom_half.h>
12 #include <linux/cache.h>
13 #include <linux/interrupt.h>
14 #include <linux/slab.h>
15 #include <linux/module.h>
16 #include <linux/netdevice.h>
17 #include <linux/percpu.h>
18 #include <net/dst.h>
19 #include <net/ip.h>
20 #include <net/xfrm.h>
21 #include <net/ip_tunnels.h>
22 #include <net/ip6_tunnel.h>
23 
24 #include "xfrm_inout.h"
25 
26 struct xfrm_trans_tasklet {
27 	struct tasklet_struct tasklet;
28 	struct sk_buff_head queue;
29 };
30 
31 struct xfrm_trans_cb {
32 	union {
33 		struct inet_skb_parm	h4;
34 #if IS_ENABLED(CONFIG_IPV6)
35 		struct inet6_skb_parm	h6;
36 #endif
37 	} header;
38 	int (*finish)(struct net *net, struct sock *sk, struct sk_buff *skb);
39 	struct net *net;
40 };
41 
42 #define XFRM_TRANS_SKB_CB(__skb) ((struct xfrm_trans_cb *)&((__skb)->cb[0]))
43 
44 static DEFINE_SPINLOCK(xfrm_input_afinfo_lock);
45 static struct xfrm_input_afinfo const __rcu *xfrm_input_afinfo[2][AF_INET6 + 1];
46 
47 static struct gro_cells gro_cells;
48 static struct net_device xfrm_napi_dev;
49 
50 static DEFINE_PER_CPU(struct xfrm_trans_tasklet, xfrm_trans_tasklet);
51 
52 int xfrm_input_register_afinfo(const struct xfrm_input_afinfo *afinfo)
53 {
54 	int err = 0;
55 
56 	if (WARN_ON(afinfo->family > AF_INET6))
57 		return -EAFNOSUPPORT;
58 
59 	spin_lock_bh(&xfrm_input_afinfo_lock);
60 	if (unlikely(xfrm_input_afinfo[afinfo->is_ipip][afinfo->family]))
61 		err = -EEXIST;
62 	else
63 		rcu_assign_pointer(xfrm_input_afinfo[afinfo->is_ipip][afinfo->family], afinfo);
64 	spin_unlock_bh(&xfrm_input_afinfo_lock);
65 	return err;
66 }
67 EXPORT_SYMBOL(xfrm_input_register_afinfo);
68 
69 int xfrm_input_unregister_afinfo(const struct xfrm_input_afinfo *afinfo)
70 {
71 	int err = 0;
72 
73 	spin_lock_bh(&xfrm_input_afinfo_lock);
74 	if (likely(xfrm_input_afinfo[afinfo->is_ipip][afinfo->family])) {
75 		if (unlikely(xfrm_input_afinfo[afinfo->is_ipip][afinfo->family] != afinfo))
76 			err = -EINVAL;
77 		else
78 			RCU_INIT_POINTER(xfrm_input_afinfo[afinfo->is_ipip][afinfo->family], NULL);
79 	}
80 	spin_unlock_bh(&xfrm_input_afinfo_lock);
81 	synchronize_rcu();
82 	return err;
83 }
84 EXPORT_SYMBOL(xfrm_input_unregister_afinfo);
85 
86 static const struct xfrm_input_afinfo *xfrm_input_get_afinfo(u8 family, bool is_ipip)
87 {
88 	const struct xfrm_input_afinfo *afinfo;
89 
90 	if (WARN_ON_ONCE(family > AF_INET6))
91 		return NULL;
92 
93 	rcu_read_lock();
94 	afinfo = rcu_dereference(xfrm_input_afinfo[is_ipip][family]);
95 	if (unlikely(!afinfo))
96 		rcu_read_unlock();
97 	return afinfo;
98 }
99 
100 static int xfrm_rcv_cb(struct sk_buff *skb, unsigned int family, u8 protocol,
101 		       int err)
102 {
103 	bool is_ipip = (protocol == IPPROTO_IPIP || protocol == IPPROTO_IPV6);
104 	const struct xfrm_input_afinfo *afinfo;
105 	int ret;
106 
107 	afinfo = xfrm_input_get_afinfo(family, is_ipip);
108 	if (!afinfo)
109 		return -EAFNOSUPPORT;
110 
111 	ret = afinfo->callback(skb, protocol, err);
112 	rcu_read_unlock();
113 
114 	return ret;
115 }
116 
117 struct sec_path *secpath_set(struct sk_buff *skb)
118 {
119 	struct sec_path *sp, *tmp = skb_ext_find(skb, SKB_EXT_SEC_PATH);
120 
121 	sp = skb_ext_add(skb, SKB_EXT_SEC_PATH);
122 	if (!sp)
123 		return NULL;
124 
125 	if (tmp) /* reused existing one (was COW'd if needed) */
126 		return sp;
127 
128 	/* allocated new secpath */
129 	memset(sp->ovec, 0, sizeof(sp->ovec));
130 	sp->olen = 0;
131 	sp->len = 0;
132 
133 	return sp;
134 }
135 EXPORT_SYMBOL(secpath_set);
136 
137 /* Fetch spi and seq from ipsec header */
138 
139 int xfrm_parse_spi(struct sk_buff *skb, u8 nexthdr, __be32 *spi, __be32 *seq)
140 {
141 	int offset, offset_seq;
142 	int hlen;
143 
144 	switch (nexthdr) {
145 	case IPPROTO_AH:
146 		hlen = sizeof(struct ip_auth_hdr);
147 		offset = offsetof(struct ip_auth_hdr, spi);
148 		offset_seq = offsetof(struct ip_auth_hdr, seq_no);
149 		break;
150 	case IPPROTO_ESP:
151 		hlen = sizeof(struct ip_esp_hdr);
152 		offset = offsetof(struct ip_esp_hdr, spi);
153 		offset_seq = offsetof(struct ip_esp_hdr, seq_no);
154 		break;
155 	case IPPROTO_COMP:
156 		if (!pskb_may_pull(skb, sizeof(struct ip_comp_hdr)))
157 			return -EINVAL;
158 		*spi = htonl(ntohs(*(__be16 *)(skb_transport_header(skb) + 2)));
159 		*seq = 0;
160 		return 0;
161 	default:
162 		return 1;
163 	}
164 
165 	if (!pskb_may_pull(skb, hlen))
166 		return -EINVAL;
167 
168 	*spi = *(__be32 *)(skb_transport_header(skb) + offset);
169 	*seq = *(__be32 *)(skb_transport_header(skb) + offset_seq);
170 	return 0;
171 }
172 EXPORT_SYMBOL(xfrm_parse_spi);
173 
174 static int xfrm4_remove_beet_encap(struct xfrm_state *x, struct sk_buff *skb)
175 {
176 	struct iphdr *iph;
177 	int optlen = 0;
178 	int err = -EINVAL;
179 
180 	if (unlikely(XFRM_MODE_SKB_CB(skb)->protocol == IPPROTO_BEETPH)) {
181 		struct ip_beet_phdr *ph;
182 		int phlen;
183 
184 		if (!pskb_may_pull(skb, sizeof(*ph)))
185 			goto out;
186 
187 		ph = (struct ip_beet_phdr *)skb->data;
188 
189 		phlen = sizeof(*ph) + ph->padlen;
190 		optlen = ph->hdrlen * 8 + (IPV4_BEET_PHMAXLEN - phlen);
191 		if (optlen < 0 || optlen & 3 || optlen > 250)
192 			goto out;
193 
194 		XFRM_MODE_SKB_CB(skb)->protocol = ph->nexthdr;
195 
196 		if (!pskb_may_pull(skb, phlen))
197 			goto out;
198 		__skb_pull(skb, phlen);
199 	}
200 
201 	skb_push(skb, sizeof(*iph));
202 	skb_reset_network_header(skb);
203 	skb_mac_header_rebuild(skb);
204 
205 	xfrm4_beet_make_header(skb);
206 
207 	iph = ip_hdr(skb);
208 
209 	iph->ihl += optlen / 4;
210 	iph->tot_len = htons(skb->len);
211 	iph->daddr = x->sel.daddr.a4;
212 	iph->saddr = x->sel.saddr.a4;
213 	iph->check = 0;
214 	iph->check = ip_fast_csum(skb_network_header(skb), iph->ihl);
215 	err = 0;
216 out:
217 	return err;
218 }
219 
220 static void ipip_ecn_decapsulate(struct sk_buff *skb)
221 {
222 	struct iphdr *inner_iph = ipip_hdr(skb);
223 
224 	if (INET_ECN_is_ce(XFRM_MODE_SKB_CB(skb)->tos))
225 		IP_ECN_set_ce(inner_iph);
226 }
227 
228 static int xfrm4_remove_tunnel_encap(struct xfrm_state *x, struct sk_buff *skb)
229 {
230 	int err = -EINVAL;
231 
232 	if (XFRM_MODE_SKB_CB(skb)->protocol != IPPROTO_IPIP)
233 		goto out;
234 
235 	if (!pskb_may_pull(skb, sizeof(struct iphdr)))
236 		goto out;
237 
238 	err = skb_unclone(skb, GFP_ATOMIC);
239 	if (err)
240 		goto out;
241 
242 	if (x->props.flags & XFRM_STATE_DECAP_DSCP)
243 		ipv4_copy_dscp(XFRM_MODE_SKB_CB(skb)->tos, ipip_hdr(skb));
244 	if (!(x->props.flags & XFRM_STATE_NOECN))
245 		ipip_ecn_decapsulate(skb);
246 
247 	skb_reset_network_header(skb);
248 	skb_mac_header_rebuild(skb);
249 	if (skb->mac_len)
250 		eth_hdr(skb)->h_proto = skb->protocol;
251 
252 	err = 0;
253 
254 out:
255 	return err;
256 }
257 
258 static void ipip6_ecn_decapsulate(struct sk_buff *skb)
259 {
260 	struct ipv6hdr *inner_iph = ipipv6_hdr(skb);
261 
262 	if (INET_ECN_is_ce(XFRM_MODE_SKB_CB(skb)->tos))
263 		IP6_ECN_set_ce(skb, inner_iph);
264 }
265 
266 static int xfrm6_remove_tunnel_encap(struct xfrm_state *x, struct sk_buff *skb)
267 {
268 	int err = -EINVAL;
269 
270 	if (XFRM_MODE_SKB_CB(skb)->protocol != IPPROTO_IPV6)
271 		goto out;
272 	if (!pskb_may_pull(skb, sizeof(struct ipv6hdr)))
273 		goto out;
274 
275 	err = skb_unclone(skb, GFP_ATOMIC);
276 	if (err)
277 		goto out;
278 
279 	if (x->props.flags & XFRM_STATE_DECAP_DSCP)
280 		ipv6_copy_dscp(ipv6_get_dsfield(ipv6_hdr(skb)),
281 			       ipipv6_hdr(skb));
282 	if (!(x->props.flags & XFRM_STATE_NOECN))
283 		ipip6_ecn_decapsulate(skb);
284 
285 	skb_reset_network_header(skb);
286 	skb_mac_header_rebuild(skb);
287 	if (skb->mac_len)
288 		eth_hdr(skb)->h_proto = skb->protocol;
289 
290 	err = 0;
291 
292 out:
293 	return err;
294 }
295 
296 static int xfrm6_remove_beet_encap(struct xfrm_state *x, struct sk_buff *skb)
297 {
298 	struct ipv6hdr *ip6h;
299 	int size = sizeof(struct ipv6hdr);
300 	int err;
301 
302 	err = skb_cow_head(skb, size + skb->mac_len);
303 	if (err)
304 		goto out;
305 
306 	__skb_push(skb, size);
307 	skb_reset_network_header(skb);
308 	skb_mac_header_rebuild(skb);
309 
310 	xfrm6_beet_make_header(skb);
311 
312 	ip6h = ipv6_hdr(skb);
313 	ip6h->payload_len = htons(skb->len - size);
314 	ip6h->daddr = x->sel.daddr.in6;
315 	ip6h->saddr = x->sel.saddr.in6;
316 	err = 0;
317 out:
318 	return err;
319 }
320 
321 /* Remove encapsulation header.
322  *
323  * The IP header will be moved over the top of the encapsulation
324  * header.
325  *
326  * On entry, the transport header shall point to where the IP header
327  * should be and the network header shall be set to where the IP
328  * header currently is.  skb->data shall point to the start of the
329  * payload.
330  */
331 static int
332 xfrm_inner_mode_encap_remove(struct xfrm_state *x,
333 			     const struct xfrm_mode *inner_mode,
334 			     struct sk_buff *skb)
335 {
336 	switch (inner_mode->encap) {
337 	case XFRM_MODE_BEET:
338 		if (inner_mode->family == AF_INET)
339 			return xfrm4_remove_beet_encap(x, skb);
340 		if (inner_mode->family == AF_INET6)
341 			return xfrm6_remove_beet_encap(x, skb);
342 		break;
343 	case XFRM_MODE_TUNNEL:
344 		if (inner_mode->family == AF_INET)
345 			return xfrm4_remove_tunnel_encap(x, skb);
346 		if (inner_mode->family == AF_INET6)
347 			return xfrm6_remove_tunnel_encap(x, skb);
348 		break;
349 	}
350 
351 	WARN_ON_ONCE(1);
352 	return -EOPNOTSUPP;
353 }
354 
355 static int xfrm_prepare_input(struct xfrm_state *x, struct sk_buff *skb)
356 {
357 	const struct xfrm_mode *inner_mode = &x->inner_mode;
358 
359 	switch (x->outer_mode.family) {
360 	case AF_INET:
361 		xfrm4_extract_header(skb);
362 		break;
363 	case AF_INET6:
364 		xfrm6_extract_header(skb);
365 		break;
366 	default:
367 		WARN_ON_ONCE(1);
368 		return -EAFNOSUPPORT;
369 	}
370 
371 	if (x->sel.family == AF_UNSPEC) {
372 		inner_mode = xfrm_ip2inner_mode(x, XFRM_MODE_SKB_CB(skb)->protocol);
373 		if (!inner_mode)
374 			return -EAFNOSUPPORT;
375 	}
376 
377 	switch (inner_mode->family) {
378 	case AF_INET:
379 		skb->protocol = htons(ETH_P_IP);
380 		break;
381 	case AF_INET6:
382 		skb->protocol = htons(ETH_P_IPV6);
383 		break;
384 	default:
385 		WARN_ON_ONCE(1);
386 		break;
387 	}
388 
389 	return xfrm_inner_mode_encap_remove(x, inner_mode, skb);
390 }
391 
392 /* Remove encapsulation header.
393  *
394  * The IP header will be moved over the top of the encapsulation header.
395  *
396  * On entry, skb_transport_header() shall point to where the IP header
397  * should be and skb_network_header() shall be set to where the IP header
398  * currently is.  skb->data shall point to the start of the payload.
399  */
400 static int xfrm4_transport_input(struct xfrm_state *x, struct sk_buff *skb)
401 {
402 	int ihl = skb->data - skb_transport_header(skb);
403 
404 	if (skb->transport_header != skb->network_header) {
405 		memmove(skb_transport_header(skb),
406 			skb_network_header(skb), ihl);
407 		skb->network_header = skb->transport_header;
408 	}
409 	ip_hdr(skb)->tot_len = htons(skb->len + ihl);
410 	skb_reset_transport_header(skb);
411 	return 0;
412 }
413 
414 static int xfrm6_transport_input(struct xfrm_state *x, struct sk_buff *skb)
415 {
416 #if IS_ENABLED(CONFIG_IPV6)
417 	int ihl = skb->data - skb_transport_header(skb);
418 
419 	if (skb->transport_header != skb->network_header) {
420 		memmove(skb_transport_header(skb),
421 			skb_network_header(skb), ihl);
422 		skb->network_header = skb->transport_header;
423 	}
424 	ipv6_hdr(skb)->payload_len = htons(skb->len + ihl -
425 					   sizeof(struct ipv6hdr));
426 	skb_reset_transport_header(skb);
427 	return 0;
428 #else
429 	WARN_ON_ONCE(1);
430 	return -EAFNOSUPPORT;
431 #endif
432 }
433 
434 static int xfrm_inner_mode_input(struct xfrm_state *x,
435 				 const struct xfrm_mode *inner_mode,
436 				 struct sk_buff *skb)
437 {
438 	switch (inner_mode->encap) {
439 	case XFRM_MODE_BEET:
440 	case XFRM_MODE_TUNNEL:
441 		return xfrm_prepare_input(x, skb);
442 	case XFRM_MODE_TRANSPORT:
443 		if (inner_mode->family == AF_INET)
444 			return xfrm4_transport_input(x, skb);
445 		if (inner_mode->family == AF_INET6)
446 			return xfrm6_transport_input(x, skb);
447 		break;
448 	case XFRM_MODE_ROUTEOPTIMIZATION:
449 		WARN_ON_ONCE(1);
450 		break;
451 	default:
452 		WARN_ON_ONCE(1);
453 		break;
454 	}
455 
456 	return -EOPNOTSUPP;
457 }
458 
459 int xfrm_input(struct sk_buff *skb, int nexthdr, __be32 spi, int encap_type)
460 {
461 	const struct xfrm_state_afinfo *afinfo;
462 	struct net *net = dev_net(skb->dev);
463 	const struct xfrm_mode *inner_mode;
464 	int err;
465 	__be32 seq;
466 	__be32 seq_hi;
467 	struct xfrm_state *x = NULL;
468 	xfrm_address_t *daddr;
469 	u32 mark = skb->mark;
470 	unsigned int family = AF_UNSPEC;
471 	int decaps = 0;
472 	int async = 0;
473 	bool xfrm_gro = false;
474 	bool crypto_done = false;
475 	struct xfrm_offload *xo = xfrm_offload(skb);
476 	struct sec_path *sp;
477 
478 	if (encap_type < 0) {
479 		x = xfrm_input_state(skb);
480 
481 		if (unlikely(x->km.state != XFRM_STATE_VALID)) {
482 			if (x->km.state == XFRM_STATE_ACQ)
483 				XFRM_INC_STATS(net, LINUX_MIB_XFRMACQUIREERROR);
484 			else
485 				XFRM_INC_STATS(net,
486 					       LINUX_MIB_XFRMINSTATEINVALID);
487 
488 			if (encap_type == -1)
489 				dev_put(skb->dev);
490 			goto drop;
491 		}
492 
493 		family = x->outer_mode.family;
494 
495 		/* An encap_type of -1 indicates async resumption. */
496 		if (encap_type == -1) {
497 			async = 1;
498 			seq = XFRM_SKB_CB(skb)->seq.input.low;
499 			goto resume;
500 		}
501 
502 		/* encap_type < -1 indicates a GRO call. */
503 		encap_type = 0;
504 		seq = XFRM_SPI_SKB_CB(skb)->seq;
505 
506 		if (xo && (xo->flags & CRYPTO_DONE)) {
507 			crypto_done = true;
508 			family = XFRM_SPI_SKB_CB(skb)->family;
509 
510 			if (!(xo->status & CRYPTO_SUCCESS)) {
511 				if (xo->status &
512 				    (CRYPTO_TRANSPORT_AH_AUTH_FAILED |
513 				     CRYPTO_TRANSPORT_ESP_AUTH_FAILED |
514 				     CRYPTO_TUNNEL_AH_AUTH_FAILED |
515 				     CRYPTO_TUNNEL_ESP_AUTH_FAILED)) {
516 
517 					xfrm_audit_state_icvfail(x, skb,
518 								 x->type->proto);
519 					x->stats.integrity_failed++;
520 					XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEPROTOERROR);
521 					goto drop;
522 				}
523 
524 				if (xo->status & CRYPTO_INVALID_PROTOCOL) {
525 					XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEPROTOERROR);
526 					goto drop;
527 				}
528 
529 				XFRM_INC_STATS(net, LINUX_MIB_XFRMINBUFFERERROR);
530 				goto drop;
531 			}
532 
533 			if (xfrm_parse_spi(skb, nexthdr, &spi, &seq)) {
534 				XFRM_INC_STATS(net, LINUX_MIB_XFRMINHDRERROR);
535 				goto drop;
536 			}
537 		}
538 
539 		goto lock;
540 	}
541 
542 	family = XFRM_SPI_SKB_CB(skb)->family;
543 
544 	/* if tunnel is present override skb->mark value with tunnel i_key */
545 	switch (family) {
546 	case AF_INET:
547 		if (XFRM_TUNNEL_SKB_CB(skb)->tunnel.ip4)
548 			mark = be32_to_cpu(XFRM_TUNNEL_SKB_CB(skb)->tunnel.ip4->parms.i_key);
549 		break;
550 	case AF_INET6:
551 		if (XFRM_TUNNEL_SKB_CB(skb)->tunnel.ip6)
552 			mark = be32_to_cpu(XFRM_TUNNEL_SKB_CB(skb)->tunnel.ip6->parms.i_key);
553 		break;
554 	}
555 
556 	sp = secpath_set(skb);
557 	if (!sp) {
558 		XFRM_INC_STATS(net, LINUX_MIB_XFRMINERROR);
559 		goto drop;
560 	}
561 
562 	seq = 0;
563 	if (!spi && xfrm_parse_spi(skb, nexthdr, &spi, &seq)) {
564 		secpath_reset(skb);
565 		XFRM_INC_STATS(net, LINUX_MIB_XFRMINHDRERROR);
566 		goto drop;
567 	}
568 
569 	daddr = (xfrm_address_t *)(skb_network_header(skb) +
570 				   XFRM_SPI_SKB_CB(skb)->daddroff);
571 	do {
572 		sp = skb_sec_path(skb);
573 
574 		if (sp->len == XFRM_MAX_DEPTH) {
575 			secpath_reset(skb);
576 			XFRM_INC_STATS(net, LINUX_MIB_XFRMINBUFFERERROR);
577 			goto drop;
578 		}
579 
580 		x = xfrm_state_lookup(net, mark, daddr, spi, nexthdr, family);
581 		if (x == NULL) {
582 			secpath_reset(skb);
583 			XFRM_INC_STATS(net, LINUX_MIB_XFRMINNOSTATES);
584 			xfrm_audit_state_notfound(skb, family, spi, seq);
585 			goto drop;
586 		}
587 
588 		skb->mark = xfrm_smark_get(skb->mark, x);
589 
590 		sp->xvec[sp->len++] = x;
591 
592 		skb_dst_force(skb);
593 		if (!skb_dst(skb)) {
594 			XFRM_INC_STATS(net, LINUX_MIB_XFRMINERROR);
595 			goto drop;
596 		}
597 
598 lock:
599 		spin_lock(&x->lock);
600 
601 		if (unlikely(x->km.state != XFRM_STATE_VALID)) {
602 			if (x->km.state == XFRM_STATE_ACQ)
603 				XFRM_INC_STATS(net, LINUX_MIB_XFRMACQUIREERROR);
604 			else
605 				XFRM_INC_STATS(net,
606 					       LINUX_MIB_XFRMINSTATEINVALID);
607 			goto drop_unlock;
608 		}
609 
610 		if ((x->encap ? x->encap->encap_type : 0) != encap_type) {
611 			XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMISMATCH);
612 			goto drop_unlock;
613 		}
614 
615 		if (xfrm_replay_check(x, skb, seq)) {
616 			XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATESEQERROR);
617 			goto drop_unlock;
618 		}
619 
620 		if (xfrm_state_check_expire(x)) {
621 			XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEEXPIRED);
622 			goto drop_unlock;
623 		}
624 
625 		spin_unlock(&x->lock);
626 
627 		if (xfrm_tunnel_check(skb, x, family)) {
628 			XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMODEERROR);
629 			goto drop;
630 		}
631 
632 		seq_hi = htonl(xfrm_replay_seqhi(x, seq));
633 
634 		XFRM_SKB_CB(skb)->seq.input.low = seq;
635 		XFRM_SKB_CB(skb)->seq.input.hi = seq_hi;
636 
637 		dev_hold(skb->dev);
638 
639 		if (crypto_done)
640 			nexthdr = x->type_offload->input_tail(x, skb);
641 		else
642 			nexthdr = x->type->input(x, skb);
643 
644 		if (nexthdr == -EINPROGRESS)
645 			return 0;
646 resume:
647 		dev_put(skb->dev);
648 
649 		spin_lock(&x->lock);
650 		if (nexthdr < 0) {
651 			if (nexthdr == -EBADMSG) {
652 				xfrm_audit_state_icvfail(x, skb,
653 							 x->type->proto);
654 				x->stats.integrity_failed++;
655 			}
656 			XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEPROTOERROR);
657 			goto drop_unlock;
658 		}
659 
660 		/* only the first xfrm gets the encap type */
661 		encap_type = 0;
662 
663 		if (xfrm_replay_recheck(x, skb, seq)) {
664 			XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATESEQERROR);
665 			goto drop_unlock;
666 		}
667 
668 		xfrm_replay_advance(x, seq);
669 
670 		x->curlft.bytes += skb->len;
671 		x->curlft.packets++;
672 		x->curlft.use_time = ktime_get_real_seconds();
673 
674 		spin_unlock(&x->lock);
675 
676 		XFRM_MODE_SKB_CB(skb)->protocol = nexthdr;
677 
678 		inner_mode = &x->inner_mode;
679 
680 		if (x->sel.family == AF_UNSPEC) {
681 			inner_mode = xfrm_ip2inner_mode(x, XFRM_MODE_SKB_CB(skb)->protocol);
682 			if (inner_mode == NULL) {
683 				XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMODEERROR);
684 				goto drop;
685 			}
686 		}
687 
688 		if (xfrm_inner_mode_input(x, inner_mode, skb)) {
689 			XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMODEERROR);
690 			goto drop;
691 		}
692 
693 		if (x->outer_mode.flags & XFRM_MODE_FLAG_TUNNEL) {
694 			decaps = 1;
695 			break;
696 		}
697 
698 		/*
699 		 * We need the inner address.  However, we only get here for
700 		 * transport mode so the outer address is identical.
701 		 */
702 		daddr = &x->id.daddr;
703 		family = x->outer_mode.family;
704 
705 		err = xfrm_parse_spi(skb, nexthdr, &spi, &seq);
706 		if (err < 0) {
707 			XFRM_INC_STATS(net, LINUX_MIB_XFRMINHDRERROR);
708 			goto drop;
709 		}
710 		crypto_done = false;
711 	} while (!err);
712 
713 	err = xfrm_rcv_cb(skb, family, x->type->proto, 0);
714 	if (err)
715 		goto drop;
716 
717 	nf_reset_ct(skb);
718 
719 	if (decaps) {
720 		sp = skb_sec_path(skb);
721 		if (sp)
722 			sp->olen = 0;
723 		skb_dst_drop(skb);
724 		gro_cells_receive(&gro_cells, skb);
725 		return 0;
726 	} else {
727 		xo = xfrm_offload(skb);
728 		if (xo)
729 			xfrm_gro = xo->flags & XFRM_GRO;
730 
731 		err = -EAFNOSUPPORT;
732 		rcu_read_lock();
733 		afinfo = xfrm_state_afinfo_get_rcu(x->inner_mode.family);
734 		if (likely(afinfo))
735 			err = afinfo->transport_finish(skb, xfrm_gro || async);
736 		rcu_read_unlock();
737 		if (xfrm_gro) {
738 			sp = skb_sec_path(skb);
739 			if (sp)
740 				sp->olen = 0;
741 			skb_dst_drop(skb);
742 			gro_cells_receive(&gro_cells, skb);
743 			return err;
744 		}
745 
746 		return err;
747 	}
748 
749 drop_unlock:
750 	spin_unlock(&x->lock);
751 drop:
752 	xfrm_rcv_cb(skb, family, x && x->type ? x->type->proto : nexthdr, -1);
753 	kfree_skb(skb);
754 	return 0;
755 }
756 EXPORT_SYMBOL(xfrm_input);
757 
758 int xfrm_input_resume(struct sk_buff *skb, int nexthdr)
759 {
760 	return xfrm_input(skb, nexthdr, 0, -1);
761 }
762 EXPORT_SYMBOL(xfrm_input_resume);
763 
764 static void xfrm_trans_reinject(struct tasklet_struct *t)
765 {
766 	struct xfrm_trans_tasklet *trans = from_tasklet(trans, t, tasklet);
767 	struct sk_buff_head queue;
768 	struct sk_buff *skb;
769 
770 	__skb_queue_head_init(&queue);
771 	skb_queue_splice_init(&trans->queue, &queue);
772 
773 	while ((skb = __skb_dequeue(&queue)))
774 		XFRM_TRANS_SKB_CB(skb)->finish(XFRM_TRANS_SKB_CB(skb)->net,
775 					       NULL, skb);
776 }
777 
778 int xfrm_trans_queue_net(struct net *net, struct sk_buff *skb,
779 			 int (*finish)(struct net *, struct sock *,
780 				       struct sk_buff *))
781 {
782 	struct xfrm_trans_tasklet *trans;
783 
784 	trans = this_cpu_ptr(&xfrm_trans_tasklet);
785 
786 	if (skb_queue_len(&trans->queue) >= netdev_max_backlog)
787 		return -ENOBUFS;
788 
789 	BUILD_BUG_ON(sizeof(struct xfrm_trans_cb) > sizeof(skb->cb));
790 
791 	XFRM_TRANS_SKB_CB(skb)->finish = finish;
792 	XFRM_TRANS_SKB_CB(skb)->net = net;
793 	__skb_queue_tail(&trans->queue, skb);
794 	tasklet_schedule(&trans->tasklet);
795 	return 0;
796 }
797 EXPORT_SYMBOL(xfrm_trans_queue_net);
798 
799 int xfrm_trans_queue(struct sk_buff *skb,
800 		     int (*finish)(struct net *, struct sock *,
801 				   struct sk_buff *))
802 {
803 	return xfrm_trans_queue_net(dev_net(skb->dev), skb, finish);
804 }
805 EXPORT_SYMBOL(xfrm_trans_queue);
806 
807 void __init xfrm_input_init(void)
808 {
809 	int err;
810 	int i;
811 
812 	init_dummy_netdev(&xfrm_napi_dev);
813 	err = gro_cells_init(&gro_cells, &xfrm_napi_dev);
814 	if (err)
815 		gro_cells.cells = NULL;
816 
817 	for_each_possible_cpu(i) {
818 		struct xfrm_trans_tasklet *trans;
819 
820 		trans = &per_cpu(xfrm_trans_tasklet, i);
821 		__skb_queue_head_init(&trans->queue);
822 		tasklet_setup(&trans->tasklet, xfrm_trans_reinject);
823 	}
824 }
825