xref: /openbmc/linux/net/xfrm/xfrm_input.c (revision 74be2d3b)
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[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 >= ARRAY_SIZE(xfrm_input_afinfo)))
57 		return -EAFNOSUPPORT;
58 
59 	spin_lock_bh(&xfrm_input_afinfo_lock);
60 	if (unlikely(xfrm_input_afinfo[afinfo->family] != NULL))
61 		err = -EEXIST;
62 	else
63 		rcu_assign_pointer(xfrm_input_afinfo[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->family] != NULL)) {
75 		if (unlikely(xfrm_input_afinfo[afinfo->family] != afinfo))
76 			err = -EINVAL;
77 		else
78 			RCU_INIT_POINTER(xfrm_input_afinfo[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(unsigned int family)
87 {
88 	const struct xfrm_input_afinfo *afinfo;
89 
90 	if (WARN_ON_ONCE(family >= ARRAY_SIZE(xfrm_input_afinfo)))
91 		return NULL;
92 
93 	rcu_read_lock();
94 	afinfo = rcu_dereference(xfrm_input_afinfo[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 	int ret;
104 	const struct xfrm_input_afinfo *afinfo = xfrm_input_get_afinfo(family);
105 
106 	if (!afinfo)
107 		return -EAFNOSUPPORT;
108 
109 	ret = afinfo->callback(skb, protocol, err);
110 	rcu_read_unlock();
111 
112 	return ret;
113 }
114 
115 struct sec_path *secpath_set(struct sk_buff *skb)
116 {
117 	struct sec_path *sp, *tmp = skb_ext_find(skb, SKB_EXT_SEC_PATH);
118 
119 	sp = skb_ext_add(skb, SKB_EXT_SEC_PATH);
120 	if (!sp)
121 		return NULL;
122 
123 	if (tmp) /* reused existing one (was COW'd if needed) */
124 		return sp;
125 
126 	/* allocated new secpath */
127 	memset(sp->ovec, 0, sizeof(sp->ovec));
128 	sp->olen = 0;
129 	sp->len = 0;
130 
131 	return sp;
132 }
133 EXPORT_SYMBOL(secpath_set);
134 
135 /* Fetch spi and seq from ipsec header */
136 
137 int xfrm_parse_spi(struct sk_buff *skb, u8 nexthdr, __be32 *spi, __be32 *seq)
138 {
139 	int offset, offset_seq;
140 	int hlen;
141 
142 	switch (nexthdr) {
143 	case IPPROTO_AH:
144 		hlen = sizeof(struct ip_auth_hdr);
145 		offset = offsetof(struct ip_auth_hdr, spi);
146 		offset_seq = offsetof(struct ip_auth_hdr, seq_no);
147 		break;
148 	case IPPROTO_ESP:
149 		hlen = sizeof(struct ip_esp_hdr);
150 		offset = offsetof(struct ip_esp_hdr, spi);
151 		offset_seq = offsetof(struct ip_esp_hdr, seq_no);
152 		break;
153 	case IPPROTO_COMP:
154 		if (!pskb_may_pull(skb, sizeof(struct ip_comp_hdr)))
155 			return -EINVAL;
156 		*spi = htonl(ntohs(*(__be16 *)(skb_transport_header(skb) + 2)));
157 		*seq = 0;
158 		return 0;
159 	default:
160 		return 1;
161 	}
162 
163 	if (!pskb_may_pull(skb, hlen))
164 		return -EINVAL;
165 
166 	*spi = *(__be32 *)(skb_transport_header(skb) + offset);
167 	*seq = *(__be32 *)(skb_transport_header(skb) + offset_seq);
168 	return 0;
169 }
170 EXPORT_SYMBOL(xfrm_parse_spi);
171 
172 static int xfrm4_remove_beet_encap(struct xfrm_state *x, struct sk_buff *skb)
173 {
174 	struct iphdr *iph;
175 	int optlen = 0;
176 	int err = -EINVAL;
177 
178 	if (unlikely(XFRM_MODE_SKB_CB(skb)->protocol == IPPROTO_BEETPH)) {
179 		struct ip_beet_phdr *ph;
180 		int phlen;
181 
182 		if (!pskb_may_pull(skb, sizeof(*ph)))
183 			goto out;
184 
185 		ph = (struct ip_beet_phdr *)skb->data;
186 
187 		phlen = sizeof(*ph) + ph->padlen;
188 		optlen = ph->hdrlen * 8 + (IPV4_BEET_PHMAXLEN - phlen);
189 		if (optlen < 0 || optlen & 3 || optlen > 250)
190 			goto out;
191 
192 		XFRM_MODE_SKB_CB(skb)->protocol = ph->nexthdr;
193 
194 		if (!pskb_may_pull(skb, phlen))
195 			goto out;
196 		__skb_pull(skb, phlen);
197 	}
198 
199 	skb_push(skb, sizeof(*iph));
200 	skb_reset_network_header(skb);
201 	skb_mac_header_rebuild(skb);
202 
203 	xfrm4_beet_make_header(skb);
204 
205 	iph = ip_hdr(skb);
206 
207 	iph->ihl += optlen / 4;
208 	iph->tot_len = htons(skb->len);
209 	iph->daddr = x->sel.daddr.a4;
210 	iph->saddr = x->sel.saddr.a4;
211 	iph->check = 0;
212 	iph->check = ip_fast_csum(skb_network_header(skb), iph->ihl);
213 	err = 0;
214 out:
215 	return err;
216 }
217 
218 static void ipip_ecn_decapsulate(struct sk_buff *skb)
219 {
220 	struct iphdr *inner_iph = ipip_hdr(skb);
221 
222 	if (INET_ECN_is_ce(XFRM_MODE_SKB_CB(skb)->tos))
223 		IP_ECN_set_ce(inner_iph);
224 }
225 
226 static int xfrm4_remove_tunnel_encap(struct xfrm_state *x, struct sk_buff *skb)
227 {
228 	int err = -EINVAL;
229 
230 	if (XFRM_MODE_SKB_CB(skb)->protocol != IPPROTO_IPIP)
231 		goto out;
232 
233 	if (!pskb_may_pull(skb, sizeof(struct iphdr)))
234 		goto out;
235 
236 	err = skb_unclone(skb, GFP_ATOMIC);
237 	if (err)
238 		goto out;
239 
240 	if (x->props.flags & XFRM_STATE_DECAP_DSCP)
241 		ipv4_copy_dscp(XFRM_MODE_SKB_CB(skb)->tos, ipip_hdr(skb));
242 	if (!(x->props.flags & XFRM_STATE_NOECN))
243 		ipip_ecn_decapsulate(skb);
244 
245 	skb_reset_network_header(skb);
246 	skb_mac_header_rebuild(skb);
247 	if (skb->mac_len)
248 		eth_hdr(skb)->h_proto = skb->protocol;
249 
250 	err = 0;
251 
252 out:
253 	return err;
254 }
255 
256 static void ipip6_ecn_decapsulate(struct sk_buff *skb)
257 {
258 	struct ipv6hdr *inner_iph = ipipv6_hdr(skb);
259 
260 	if (INET_ECN_is_ce(XFRM_MODE_SKB_CB(skb)->tos))
261 		IP6_ECN_set_ce(skb, inner_iph);
262 }
263 
264 static int xfrm6_remove_tunnel_encap(struct xfrm_state *x, struct sk_buff *skb)
265 {
266 	int err = -EINVAL;
267 
268 	if (XFRM_MODE_SKB_CB(skb)->protocol != IPPROTO_IPV6)
269 		goto out;
270 	if (!pskb_may_pull(skb, sizeof(struct ipv6hdr)))
271 		goto out;
272 
273 	err = skb_unclone(skb, GFP_ATOMIC);
274 	if (err)
275 		goto out;
276 
277 	if (x->props.flags & XFRM_STATE_DECAP_DSCP)
278 		ipv6_copy_dscp(ipv6_get_dsfield(ipv6_hdr(skb)),
279 			       ipipv6_hdr(skb));
280 	if (!(x->props.flags & XFRM_STATE_NOECN))
281 		ipip6_ecn_decapsulate(skb);
282 
283 	skb_reset_network_header(skb);
284 	skb_mac_header_rebuild(skb);
285 	if (skb->mac_len)
286 		eth_hdr(skb)->h_proto = skb->protocol;
287 
288 	err = 0;
289 
290 out:
291 	return err;
292 }
293 
294 static int xfrm6_remove_beet_encap(struct xfrm_state *x, struct sk_buff *skb)
295 {
296 	struct ipv6hdr *ip6h;
297 	int size = sizeof(struct ipv6hdr);
298 	int err;
299 
300 	err = skb_cow_head(skb, size + skb->mac_len);
301 	if (err)
302 		goto out;
303 
304 	__skb_push(skb, size);
305 	skb_reset_network_header(skb);
306 	skb_mac_header_rebuild(skb);
307 
308 	xfrm6_beet_make_header(skb);
309 
310 	ip6h = ipv6_hdr(skb);
311 	ip6h->payload_len = htons(skb->len - size);
312 	ip6h->daddr = x->sel.daddr.in6;
313 	ip6h->saddr = x->sel.saddr.in6;
314 	err = 0;
315 out:
316 	return err;
317 }
318 
319 /* Remove encapsulation header.
320  *
321  * The IP header will be moved over the top of the encapsulation
322  * header.
323  *
324  * On entry, the transport header shall point to where the IP header
325  * should be and the network header shall be set to where the IP
326  * header currently is.  skb->data shall point to the start of the
327  * payload.
328  */
329 static int
330 xfrm_inner_mode_encap_remove(struct xfrm_state *x,
331 			     const struct xfrm_mode *inner_mode,
332 			     struct sk_buff *skb)
333 {
334 	switch (inner_mode->encap) {
335 	case XFRM_MODE_BEET:
336 		if (inner_mode->family == AF_INET)
337 			return xfrm4_remove_beet_encap(x, skb);
338 		if (inner_mode->family == AF_INET6)
339 			return xfrm6_remove_beet_encap(x, skb);
340 		break;
341 	case XFRM_MODE_TUNNEL:
342 		if (inner_mode->family == AF_INET)
343 			return xfrm4_remove_tunnel_encap(x, skb);
344 		if (inner_mode->family == AF_INET6)
345 			return xfrm6_remove_tunnel_encap(x, skb);
346 		break;
347 	}
348 
349 	WARN_ON_ONCE(1);
350 	return -EOPNOTSUPP;
351 }
352 
353 static int xfrm_prepare_input(struct xfrm_state *x, struct sk_buff *skb)
354 {
355 	const struct xfrm_mode *inner_mode = &x->inner_mode;
356 
357 	switch (x->outer_mode.family) {
358 	case AF_INET:
359 		xfrm4_extract_header(skb);
360 		break;
361 	case AF_INET6:
362 		xfrm6_extract_header(skb);
363 		break;
364 	default:
365 		WARN_ON_ONCE(1);
366 		return -EAFNOSUPPORT;
367 	}
368 
369 	if (x->sel.family == AF_UNSPEC) {
370 		inner_mode = xfrm_ip2inner_mode(x, XFRM_MODE_SKB_CB(skb)->protocol);
371 		if (!inner_mode)
372 			return -EAFNOSUPPORT;
373 	}
374 
375 	switch (inner_mode->family) {
376 	case AF_INET:
377 		skb->protocol = htons(ETH_P_IP);
378 		break;
379 	case AF_INET6:
380 		skb->protocol = htons(ETH_P_IPV6);
381 		break;
382 	default:
383 		WARN_ON_ONCE(1);
384 		break;
385 	}
386 
387 	return xfrm_inner_mode_encap_remove(x, inner_mode, skb);
388 }
389 
390 /* Remove encapsulation header.
391  *
392  * The IP header will be moved over the top of the encapsulation header.
393  *
394  * On entry, skb_transport_header() shall point to where the IP header
395  * should be and skb_network_header() shall be set to where the IP header
396  * currently is.  skb->data shall point to the start of the payload.
397  */
398 static int xfrm4_transport_input(struct xfrm_state *x, struct sk_buff *skb)
399 {
400 	int ihl = skb->data - skb_transport_header(skb);
401 
402 	if (skb->transport_header != skb->network_header) {
403 		memmove(skb_transport_header(skb),
404 			skb_network_header(skb), ihl);
405 		skb->network_header = skb->transport_header;
406 	}
407 	ip_hdr(skb)->tot_len = htons(skb->len + ihl);
408 	skb_reset_transport_header(skb);
409 	return 0;
410 }
411 
412 static int xfrm6_transport_input(struct xfrm_state *x, struct sk_buff *skb)
413 {
414 #if IS_ENABLED(CONFIG_IPV6)
415 	int ihl = skb->data - skb_transport_header(skb);
416 
417 	if (skb->transport_header != skb->network_header) {
418 		memmove(skb_transport_header(skb),
419 			skb_network_header(skb), ihl);
420 		skb->network_header = skb->transport_header;
421 	}
422 	ipv6_hdr(skb)->payload_len = htons(skb->len + ihl -
423 					   sizeof(struct ipv6hdr));
424 	skb_reset_transport_header(skb);
425 	return 0;
426 #else
427 	WARN_ON_ONCE(1);
428 	return -EAFNOSUPPORT;
429 #endif
430 }
431 
432 static int xfrm_inner_mode_input(struct xfrm_state *x,
433 				 const struct xfrm_mode *inner_mode,
434 				 struct sk_buff *skb)
435 {
436 	switch (inner_mode->encap) {
437 	case XFRM_MODE_BEET:
438 	case XFRM_MODE_TUNNEL:
439 		return xfrm_prepare_input(x, skb);
440 	case XFRM_MODE_TRANSPORT:
441 		if (inner_mode->family == AF_INET)
442 			return xfrm4_transport_input(x, skb);
443 		if (inner_mode->family == AF_INET6)
444 			return xfrm6_transport_input(x, skb);
445 		break;
446 	case XFRM_MODE_ROUTEOPTIMIZATION:
447 		WARN_ON_ONCE(1);
448 		break;
449 	default:
450 		WARN_ON_ONCE(1);
451 		break;
452 	}
453 
454 	return -EOPNOTSUPP;
455 }
456 
457 int xfrm_input(struct sk_buff *skb, int nexthdr, __be32 spi, int encap_type)
458 {
459 	const struct xfrm_state_afinfo *afinfo;
460 	struct net *net = dev_net(skb->dev);
461 	const struct xfrm_mode *inner_mode;
462 	int err;
463 	__be32 seq;
464 	__be32 seq_hi;
465 	struct xfrm_state *x = NULL;
466 	xfrm_address_t *daddr;
467 	u32 mark = skb->mark;
468 	unsigned int family = AF_UNSPEC;
469 	int decaps = 0;
470 	int async = 0;
471 	bool xfrm_gro = false;
472 	bool crypto_done = false;
473 	struct xfrm_offload *xo = xfrm_offload(skb);
474 	struct sec_path *sp;
475 
476 	if (encap_type < 0) {
477 		x = xfrm_input_state(skb);
478 
479 		if (unlikely(x->km.state != XFRM_STATE_VALID)) {
480 			if (x->km.state == XFRM_STATE_ACQ)
481 				XFRM_INC_STATS(net, LINUX_MIB_XFRMACQUIREERROR);
482 			else
483 				XFRM_INC_STATS(net,
484 					       LINUX_MIB_XFRMINSTATEINVALID);
485 
486 			if (encap_type == -1)
487 				dev_put(skb->dev);
488 			goto drop;
489 		}
490 
491 		family = x->outer_mode.family;
492 
493 		/* An encap_type of -1 indicates async resumption. */
494 		if (encap_type == -1) {
495 			async = 1;
496 			seq = XFRM_SKB_CB(skb)->seq.input.low;
497 			goto resume;
498 		}
499 
500 		/* encap_type < -1 indicates a GRO call. */
501 		encap_type = 0;
502 		seq = XFRM_SPI_SKB_CB(skb)->seq;
503 
504 		if (xo && (xo->flags & CRYPTO_DONE)) {
505 			crypto_done = true;
506 			family = XFRM_SPI_SKB_CB(skb)->family;
507 
508 			if (!(xo->status & CRYPTO_SUCCESS)) {
509 				if (xo->status &
510 				    (CRYPTO_TRANSPORT_AH_AUTH_FAILED |
511 				     CRYPTO_TRANSPORT_ESP_AUTH_FAILED |
512 				     CRYPTO_TUNNEL_AH_AUTH_FAILED |
513 				     CRYPTO_TUNNEL_ESP_AUTH_FAILED)) {
514 
515 					xfrm_audit_state_icvfail(x, skb,
516 								 x->type->proto);
517 					x->stats.integrity_failed++;
518 					XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEPROTOERROR);
519 					goto drop;
520 				}
521 
522 				if (xo->status & CRYPTO_INVALID_PROTOCOL) {
523 					XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEPROTOERROR);
524 					goto drop;
525 				}
526 
527 				XFRM_INC_STATS(net, LINUX_MIB_XFRMINBUFFERERROR);
528 				goto drop;
529 			}
530 
531 			if ((err = xfrm_parse_spi(skb, nexthdr, &spi, &seq)) != 0) {
532 				XFRM_INC_STATS(net, LINUX_MIB_XFRMINHDRERROR);
533 				goto drop;
534 			}
535 		}
536 
537 		goto lock;
538 	}
539 
540 	family = XFRM_SPI_SKB_CB(skb)->family;
541 
542 	/* if tunnel is present override skb->mark value with tunnel i_key */
543 	switch (family) {
544 	case AF_INET:
545 		if (XFRM_TUNNEL_SKB_CB(skb)->tunnel.ip4)
546 			mark = be32_to_cpu(XFRM_TUNNEL_SKB_CB(skb)->tunnel.ip4->parms.i_key);
547 		break;
548 	case AF_INET6:
549 		if (XFRM_TUNNEL_SKB_CB(skb)->tunnel.ip6)
550 			mark = be32_to_cpu(XFRM_TUNNEL_SKB_CB(skb)->tunnel.ip6->parms.i_key);
551 		break;
552 	}
553 
554 	sp = secpath_set(skb);
555 	if (!sp) {
556 		XFRM_INC_STATS(net, LINUX_MIB_XFRMINERROR);
557 		goto drop;
558 	}
559 
560 	seq = 0;
561 	if (!spi && (err = xfrm_parse_spi(skb, nexthdr, &spi, &seq)) != 0) {
562 		secpath_reset(skb);
563 		XFRM_INC_STATS(net, LINUX_MIB_XFRMINHDRERROR);
564 		goto drop;
565 	}
566 
567 	daddr = (xfrm_address_t *)(skb_network_header(skb) +
568 				   XFRM_SPI_SKB_CB(skb)->daddroff);
569 	do {
570 		sp = skb_sec_path(skb);
571 
572 		if (sp->len == XFRM_MAX_DEPTH) {
573 			secpath_reset(skb);
574 			XFRM_INC_STATS(net, LINUX_MIB_XFRMINBUFFERERROR);
575 			goto drop;
576 		}
577 
578 		x = xfrm_state_lookup(net, mark, daddr, spi, nexthdr, family);
579 		if (x == NULL) {
580 			secpath_reset(skb);
581 			XFRM_INC_STATS(net, LINUX_MIB_XFRMINNOSTATES);
582 			xfrm_audit_state_notfound(skb, family, spi, seq);
583 			goto drop;
584 		}
585 
586 		skb->mark = xfrm_smark_get(skb->mark, x);
587 
588 		sp->xvec[sp->len++] = x;
589 
590 		skb_dst_force(skb);
591 		if (!skb_dst(skb)) {
592 			XFRM_INC_STATS(net, LINUX_MIB_XFRMINERROR);
593 			goto drop;
594 		}
595 
596 lock:
597 		spin_lock(&x->lock);
598 
599 		if (unlikely(x->km.state != XFRM_STATE_VALID)) {
600 			if (x->km.state == XFRM_STATE_ACQ)
601 				XFRM_INC_STATS(net, LINUX_MIB_XFRMACQUIREERROR);
602 			else
603 				XFRM_INC_STATS(net,
604 					       LINUX_MIB_XFRMINSTATEINVALID);
605 			goto drop_unlock;
606 		}
607 
608 		if ((x->encap ? x->encap->encap_type : 0) != encap_type) {
609 			XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMISMATCH);
610 			goto drop_unlock;
611 		}
612 
613 		if (x->repl->check(x, skb, seq)) {
614 			XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATESEQERROR);
615 			goto drop_unlock;
616 		}
617 
618 		if (xfrm_state_check_expire(x)) {
619 			XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEEXPIRED);
620 			goto drop_unlock;
621 		}
622 
623 		spin_unlock(&x->lock);
624 
625 		if (xfrm_tunnel_check(skb, x, family)) {
626 			XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMODEERROR);
627 			goto drop;
628 		}
629 
630 		seq_hi = htonl(xfrm_replay_seqhi(x, seq));
631 
632 		XFRM_SKB_CB(skb)->seq.input.low = seq;
633 		XFRM_SKB_CB(skb)->seq.input.hi = seq_hi;
634 
635 		dev_hold(skb->dev);
636 
637 		if (crypto_done)
638 			nexthdr = x->type_offload->input_tail(x, skb);
639 		else
640 			nexthdr = x->type->input(x, skb);
641 
642 		if (nexthdr == -EINPROGRESS)
643 			return 0;
644 resume:
645 		dev_put(skb->dev);
646 
647 		spin_lock(&x->lock);
648 		if (nexthdr < 0) {
649 			if (nexthdr == -EBADMSG) {
650 				xfrm_audit_state_icvfail(x, skb,
651 							 x->type->proto);
652 				x->stats.integrity_failed++;
653 			}
654 			XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEPROTOERROR);
655 			goto drop_unlock;
656 		}
657 
658 		/* only the first xfrm gets the encap type */
659 		encap_type = 0;
660 
661 		if (async && x->repl->recheck(x, skb, seq)) {
662 			XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATESEQERROR);
663 			goto drop_unlock;
664 		}
665 
666 		x->repl->advance(x, seq);
667 
668 		x->curlft.bytes += skb->len;
669 		x->curlft.packets++;
670 
671 		spin_unlock(&x->lock);
672 
673 		XFRM_MODE_SKB_CB(skb)->protocol = nexthdr;
674 
675 		inner_mode = &x->inner_mode;
676 
677 		if (x->sel.family == AF_UNSPEC) {
678 			inner_mode = xfrm_ip2inner_mode(x, XFRM_MODE_SKB_CB(skb)->protocol);
679 			if (inner_mode == NULL) {
680 				XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMODEERROR);
681 				goto drop;
682 			}
683 		}
684 
685 		if (xfrm_inner_mode_input(x, inner_mode, skb)) {
686 			XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMODEERROR);
687 			goto drop;
688 		}
689 
690 		if (x->outer_mode.flags & XFRM_MODE_FLAG_TUNNEL) {
691 			decaps = 1;
692 			break;
693 		}
694 
695 		/*
696 		 * We need the inner address.  However, we only get here for
697 		 * transport mode so the outer address is identical.
698 		 */
699 		daddr = &x->id.daddr;
700 		family = x->outer_mode.family;
701 
702 		err = xfrm_parse_spi(skb, nexthdr, &spi, &seq);
703 		if (err < 0) {
704 			XFRM_INC_STATS(net, LINUX_MIB_XFRMINHDRERROR);
705 			goto drop;
706 		}
707 		crypto_done = false;
708 	} while (!err);
709 
710 	err = xfrm_rcv_cb(skb, family, x->type->proto, 0);
711 	if (err)
712 		goto drop;
713 
714 	nf_reset_ct(skb);
715 
716 	if (decaps) {
717 		sp = skb_sec_path(skb);
718 		if (sp)
719 			sp->olen = 0;
720 		skb_dst_drop(skb);
721 		gro_cells_receive(&gro_cells, skb);
722 		return 0;
723 	} else {
724 		xo = xfrm_offload(skb);
725 		if (xo)
726 			xfrm_gro = xo->flags & XFRM_GRO;
727 
728 		err = -EAFNOSUPPORT;
729 		rcu_read_lock();
730 		afinfo = xfrm_state_afinfo_get_rcu(x->inner_mode.family);
731 		if (likely(afinfo))
732 			err = afinfo->transport_finish(skb, xfrm_gro || async);
733 		rcu_read_unlock();
734 		if (xfrm_gro) {
735 			sp = skb_sec_path(skb);
736 			if (sp)
737 				sp->olen = 0;
738 			skb_dst_drop(skb);
739 			gro_cells_receive(&gro_cells, skb);
740 			return err;
741 		}
742 
743 		return err;
744 	}
745 
746 drop_unlock:
747 	spin_unlock(&x->lock);
748 drop:
749 	xfrm_rcv_cb(skb, family, x && x->type ? x->type->proto : nexthdr, -1);
750 	kfree_skb(skb);
751 	return 0;
752 }
753 EXPORT_SYMBOL(xfrm_input);
754 
755 int xfrm_input_resume(struct sk_buff *skb, int nexthdr)
756 {
757 	return xfrm_input(skb, nexthdr, 0, -1);
758 }
759 EXPORT_SYMBOL(xfrm_input_resume);
760 
761 static void xfrm_trans_reinject(unsigned long data)
762 {
763 	struct xfrm_trans_tasklet *trans = (void *)data;
764 	struct sk_buff_head queue;
765 	struct sk_buff *skb;
766 
767 	__skb_queue_head_init(&queue);
768 	skb_queue_splice_init(&trans->queue, &queue);
769 
770 	while ((skb = __skb_dequeue(&queue)))
771 		XFRM_TRANS_SKB_CB(skb)->finish(XFRM_TRANS_SKB_CB(skb)->net,
772 					       NULL, skb);
773 }
774 
775 int xfrm_trans_queue_net(struct net *net, struct sk_buff *skb,
776 			 int (*finish)(struct net *, struct sock *,
777 				       struct sk_buff *))
778 {
779 	struct xfrm_trans_tasklet *trans;
780 
781 	trans = this_cpu_ptr(&xfrm_trans_tasklet);
782 
783 	if (skb_queue_len(&trans->queue) >= netdev_max_backlog)
784 		return -ENOBUFS;
785 
786 	BUILD_BUG_ON(sizeof(struct xfrm_trans_cb) > sizeof(skb->cb));
787 
788 	XFRM_TRANS_SKB_CB(skb)->finish = finish;
789 	XFRM_TRANS_SKB_CB(skb)->net = net;
790 	__skb_queue_tail(&trans->queue, skb);
791 	tasklet_schedule(&trans->tasklet);
792 	return 0;
793 }
794 EXPORT_SYMBOL(xfrm_trans_queue_net);
795 
796 int xfrm_trans_queue(struct sk_buff *skb,
797 		     int (*finish)(struct net *, struct sock *,
798 				   struct sk_buff *))
799 {
800 	return xfrm_trans_queue_net(dev_net(skb->dev), skb, finish);
801 }
802 EXPORT_SYMBOL(xfrm_trans_queue);
803 
804 void __init xfrm_input_init(void)
805 {
806 	int err;
807 	int i;
808 
809 	init_dummy_netdev(&xfrm_napi_dev);
810 	err = gro_cells_init(&gro_cells, &xfrm_napi_dev);
811 	if (err)
812 		gro_cells.cells = NULL;
813 
814 	for_each_possible_cpu(i) {
815 		struct xfrm_trans_tasklet *trans;
816 
817 		trans = &per_cpu(xfrm_trans_tasklet, i);
818 		__skb_queue_head_init(&trans->queue);
819 		tasklet_init(&trans->tasklet, xfrm_trans_reinject,
820 			     (unsigned long)trans);
821 	}
822 }
823