xref: /openbmc/linux/net/xfrm/xfrm_input.c (revision ae213c44)
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 };
40 
41 #define XFRM_TRANS_SKB_CB(__skb) ((struct xfrm_trans_cb *)&((__skb)->cb[0]))
42 
43 static DEFINE_SPINLOCK(xfrm_input_afinfo_lock);
44 static struct xfrm_input_afinfo const __rcu *xfrm_input_afinfo[AF_INET6 + 1];
45 
46 static struct gro_cells gro_cells;
47 static struct net_device xfrm_napi_dev;
48 
49 static DEFINE_PER_CPU(struct xfrm_trans_tasklet, xfrm_trans_tasklet);
50 
51 int xfrm_input_register_afinfo(const struct xfrm_input_afinfo *afinfo)
52 {
53 	int err = 0;
54 
55 	if (WARN_ON(afinfo->family >= ARRAY_SIZE(xfrm_input_afinfo)))
56 		return -EAFNOSUPPORT;
57 
58 	spin_lock_bh(&xfrm_input_afinfo_lock);
59 	if (unlikely(xfrm_input_afinfo[afinfo->family] != NULL))
60 		err = -EEXIST;
61 	else
62 		rcu_assign_pointer(xfrm_input_afinfo[afinfo->family], afinfo);
63 	spin_unlock_bh(&xfrm_input_afinfo_lock);
64 	return err;
65 }
66 EXPORT_SYMBOL(xfrm_input_register_afinfo);
67 
68 int xfrm_input_unregister_afinfo(const struct xfrm_input_afinfo *afinfo)
69 {
70 	int err = 0;
71 
72 	spin_lock_bh(&xfrm_input_afinfo_lock);
73 	if (likely(xfrm_input_afinfo[afinfo->family] != NULL)) {
74 		if (unlikely(xfrm_input_afinfo[afinfo->family] != afinfo))
75 			err = -EINVAL;
76 		else
77 			RCU_INIT_POINTER(xfrm_input_afinfo[afinfo->family], NULL);
78 	}
79 	spin_unlock_bh(&xfrm_input_afinfo_lock);
80 	synchronize_rcu();
81 	return err;
82 }
83 EXPORT_SYMBOL(xfrm_input_unregister_afinfo);
84 
85 static const struct xfrm_input_afinfo *xfrm_input_get_afinfo(unsigned int family)
86 {
87 	const struct xfrm_input_afinfo *afinfo;
88 
89 	if (WARN_ON_ONCE(family >= ARRAY_SIZE(xfrm_input_afinfo)))
90 		return NULL;
91 
92 	rcu_read_lock();
93 	afinfo = rcu_dereference(xfrm_input_afinfo[family]);
94 	if (unlikely(!afinfo))
95 		rcu_read_unlock();
96 	return afinfo;
97 }
98 
99 static int xfrm_rcv_cb(struct sk_buff *skb, unsigned int family, u8 protocol,
100 		       int err)
101 {
102 	int ret;
103 	const struct xfrm_input_afinfo *afinfo = xfrm_input_get_afinfo(family);
104 
105 	if (!afinfo)
106 		return -EAFNOSUPPORT;
107 
108 	ret = afinfo->callback(skb, protocol, err);
109 	rcu_read_unlock();
110 
111 	return ret;
112 }
113 
114 struct sec_path *secpath_set(struct sk_buff *skb)
115 {
116 	struct sec_path *sp, *tmp = skb_ext_find(skb, SKB_EXT_SEC_PATH);
117 
118 	sp = skb_ext_add(skb, SKB_EXT_SEC_PATH);
119 	if (!sp)
120 		return NULL;
121 
122 	if (tmp) /* reused existing one (was COW'd if needed) */
123 		return sp;
124 
125 	/* allocated new secpath */
126 	memset(sp->ovec, 0, sizeof(sp->ovec));
127 	sp->olen = 0;
128 	sp->len = 0;
129 
130 	return sp;
131 }
132 EXPORT_SYMBOL(secpath_set);
133 
134 /* Fetch spi and seq from ipsec header */
135 
136 int xfrm_parse_spi(struct sk_buff *skb, u8 nexthdr, __be32 *spi, __be32 *seq)
137 {
138 	int offset, offset_seq;
139 	int hlen;
140 
141 	switch (nexthdr) {
142 	case IPPROTO_AH:
143 		hlen = sizeof(struct ip_auth_hdr);
144 		offset = offsetof(struct ip_auth_hdr, spi);
145 		offset_seq = offsetof(struct ip_auth_hdr, seq_no);
146 		break;
147 	case IPPROTO_ESP:
148 		hlen = sizeof(struct ip_esp_hdr);
149 		offset = offsetof(struct ip_esp_hdr, spi);
150 		offset_seq = offsetof(struct ip_esp_hdr, seq_no);
151 		break;
152 	case IPPROTO_COMP:
153 		if (!pskb_may_pull(skb, sizeof(struct ip_comp_hdr)))
154 			return -EINVAL;
155 		*spi = htonl(ntohs(*(__be16 *)(skb_transport_header(skb) + 2)));
156 		*seq = 0;
157 		return 0;
158 	default:
159 		return 1;
160 	}
161 
162 	if (!pskb_may_pull(skb, hlen))
163 		return -EINVAL;
164 
165 	*spi = *(__be32 *)(skb_transport_header(skb) + offset);
166 	*seq = *(__be32 *)(skb_transport_header(skb) + offset_seq);
167 	return 0;
168 }
169 EXPORT_SYMBOL(xfrm_parse_spi);
170 
171 static int xfrm4_remove_beet_encap(struct xfrm_state *x, struct sk_buff *skb)
172 {
173 	struct iphdr *iph;
174 	int optlen = 0;
175 	int err = -EINVAL;
176 
177 	if (unlikely(XFRM_MODE_SKB_CB(skb)->protocol == IPPROTO_BEETPH)) {
178 		struct ip_beet_phdr *ph;
179 		int phlen;
180 
181 		if (!pskb_may_pull(skb, sizeof(*ph)))
182 			goto out;
183 
184 		ph = (struct ip_beet_phdr *)skb->data;
185 
186 		phlen = sizeof(*ph) + ph->padlen;
187 		optlen = ph->hdrlen * 8 + (IPV4_BEET_PHMAXLEN - phlen);
188 		if (optlen < 0 || optlen & 3 || optlen > 250)
189 			goto out;
190 
191 		XFRM_MODE_SKB_CB(skb)->protocol = ph->nexthdr;
192 
193 		if (!pskb_may_pull(skb, phlen))
194 			goto out;
195 		__skb_pull(skb, phlen);
196 	}
197 
198 	skb_push(skb, sizeof(*iph));
199 	skb_reset_network_header(skb);
200 	skb_mac_header_rebuild(skb);
201 
202 	xfrm4_beet_make_header(skb);
203 
204 	iph = ip_hdr(skb);
205 
206 	iph->ihl += optlen / 4;
207 	iph->tot_len = htons(skb->len);
208 	iph->daddr = x->sel.daddr.a4;
209 	iph->saddr = x->sel.saddr.a4;
210 	iph->check = 0;
211 	iph->check = ip_fast_csum(skb_network_header(skb), iph->ihl);
212 	err = 0;
213 out:
214 	return err;
215 }
216 
217 static void ipip_ecn_decapsulate(struct sk_buff *skb)
218 {
219 	struct iphdr *inner_iph = ipip_hdr(skb);
220 
221 	if (INET_ECN_is_ce(XFRM_MODE_SKB_CB(skb)->tos))
222 		IP_ECN_set_ce(inner_iph);
223 }
224 
225 static int xfrm4_remove_tunnel_encap(struct xfrm_state *x, struct sk_buff *skb)
226 {
227 	int err = -EINVAL;
228 
229 	if (XFRM_MODE_SKB_CB(skb)->protocol != IPPROTO_IPIP)
230 		goto out;
231 
232 	if (!pskb_may_pull(skb, sizeof(struct iphdr)))
233 		goto out;
234 
235 	err = skb_unclone(skb, GFP_ATOMIC);
236 	if (err)
237 		goto out;
238 
239 	if (x->props.flags & XFRM_STATE_DECAP_DSCP)
240 		ipv4_copy_dscp(XFRM_MODE_SKB_CB(skb)->tos, ipip_hdr(skb));
241 	if (!(x->props.flags & XFRM_STATE_NOECN))
242 		ipip_ecn_decapsulate(skb);
243 
244 	skb_reset_network_header(skb);
245 	skb_mac_header_rebuild(skb);
246 	if (skb->mac_len)
247 		eth_hdr(skb)->h_proto = skb->protocol;
248 
249 	err = 0;
250 
251 out:
252 	return err;
253 }
254 
255 static void ipip6_ecn_decapsulate(struct sk_buff *skb)
256 {
257 	struct ipv6hdr *inner_iph = ipipv6_hdr(skb);
258 
259 	if (INET_ECN_is_ce(XFRM_MODE_SKB_CB(skb)->tos))
260 		IP6_ECN_set_ce(skb, inner_iph);
261 }
262 
263 static int xfrm6_remove_tunnel_encap(struct xfrm_state *x, struct sk_buff *skb)
264 {
265 	int err = -EINVAL;
266 
267 	if (XFRM_MODE_SKB_CB(skb)->protocol != IPPROTO_IPV6)
268 		goto out;
269 	if (!pskb_may_pull(skb, sizeof(struct ipv6hdr)))
270 		goto out;
271 
272 	err = skb_unclone(skb, GFP_ATOMIC);
273 	if (err)
274 		goto out;
275 
276 	if (x->props.flags & XFRM_STATE_DECAP_DSCP)
277 		ipv6_copy_dscp(ipv6_get_dsfield(ipv6_hdr(skb)),
278 			       ipipv6_hdr(skb));
279 	if (!(x->props.flags & XFRM_STATE_NOECN))
280 		ipip6_ecn_decapsulate(skb);
281 
282 	skb_reset_network_header(skb);
283 	skb_mac_header_rebuild(skb);
284 	if (skb->mac_len)
285 		eth_hdr(skb)->h_proto = skb->protocol;
286 
287 	err = 0;
288 
289 out:
290 	return err;
291 }
292 
293 static int xfrm6_remove_beet_encap(struct xfrm_state *x, struct sk_buff *skb)
294 {
295 	struct ipv6hdr *ip6h;
296 	int size = sizeof(struct ipv6hdr);
297 	int err;
298 
299 	err = skb_cow_head(skb, size + skb->mac_len);
300 	if (err)
301 		goto out;
302 
303 	__skb_push(skb, size);
304 	skb_reset_network_header(skb);
305 	skb_mac_header_rebuild(skb);
306 
307 	xfrm6_beet_make_header(skb);
308 
309 	ip6h = ipv6_hdr(skb);
310 	ip6h->payload_len = htons(skb->len - size);
311 	ip6h->daddr = x->sel.daddr.in6;
312 	ip6h->saddr = x->sel.saddr.in6;
313 	err = 0;
314 out:
315 	return err;
316 }
317 
318 /* Remove encapsulation header.
319  *
320  * The IP header will be moved over the top of the encapsulation
321  * header.
322  *
323  * On entry, the transport header shall point to where the IP header
324  * should be and the network header shall be set to where the IP
325  * header currently is.  skb->data shall point to the start of the
326  * payload.
327  */
328 static int
329 xfrm_inner_mode_encap_remove(struct xfrm_state *x,
330 			     const struct xfrm_mode *inner_mode,
331 			     struct sk_buff *skb)
332 {
333 	switch (inner_mode->encap) {
334 	case XFRM_MODE_BEET:
335 		if (inner_mode->family == AF_INET)
336 			return xfrm4_remove_beet_encap(x, skb);
337 		if (inner_mode->family == AF_INET6)
338 			return xfrm6_remove_beet_encap(x, skb);
339 		break;
340 	case XFRM_MODE_TUNNEL:
341 		if (inner_mode->family == AF_INET)
342 			return xfrm4_remove_tunnel_encap(x, skb);
343 		if (inner_mode->family == AF_INET6)
344 			return xfrm6_remove_tunnel_encap(x, skb);
345 		break;
346 	}
347 
348 	WARN_ON_ONCE(1);
349 	return -EOPNOTSUPP;
350 }
351 
352 static int xfrm_prepare_input(struct xfrm_state *x, struct sk_buff *skb)
353 {
354 	const struct xfrm_mode *inner_mode = &x->inner_mode;
355 	const struct xfrm_state_afinfo *afinfo;
356 	int err = -EAFNOSUPPORT;
357 
358 	rcu_read_lock();
359 	afinfo = xfrm_state_afinfo_get_rcu(x->outer_mode.family);
360 	if (likely(afinfo))
361 		err = afinfo->extract_input(x, skb);
362 
363 	if (err) {
364 		rcu_read_unlock();
365 		return err;
366 	}
367 
368 	if (x->sel.family == AF_UNSPEC) {
369 		inner_mode = xfrm_ip2inner_mode(x, XFRM_MODE_SKB_CB(skb)->protocol);
370 		if (!inner_mode) {
371 			rcu_read_unlock();
372 			return -EAFNOSUPPORT;
373 		}
374 	}
375 
376 	afinfo = xfrm_state_afinfo_get_rcu(inner_mode->family);
377 	if (unlikely(!afinfo)) {
378 		rcu_read_unlock();
379 		return -EAFNOSUPPORT;
380 	}
381 
382 	skb->protocol = afinfo->eth_proto;
383 	rcu_read_unlock();
384 	return xfrm_inner_mode_encap_remove(x, inner_mode, skb);
385 }
386 
387 /* Remove encapsulation header.
388  *
389  * The IP header will be moved over the top of the encapsulation header.
390  *
391  * On entry, skb_transport_header() shall point to where the IP header
392  * should be and skb_network_header() shall be set to where the IP header
393  * currently is.  skb->data shall point to the start of the payload.
394  */
395 static int xfrm4_transport_input(struct xfrm_state *x, struct sk_buff *skb)
396 {
397 	int ihl = skb->data - skb_transport_header(skb);
398 
399 	if (skb->transport_header != skb->network_header) {
400 		memmove(skb_transport_header(skb),
401 			skb_network_header(skb), ihl);
402 		skb->network_header = skb->transport_header;
403 	}
404 	ip_hdr(skb)->tot_len = htons(skb->len + ihl);
405 	skb_reset_transport_header(skb);
406 	return 0;
407 }
408 
409 static int xfrm6_transport_input(struct xfrm_state *x, struct sk_buff *skb)
410 {
411 #if IS_ENABLED(CONFIG_IPV6)
412 	int ihl = skb->data - skb_transport_header(skb);
413 
414 	if (skb->transport_header != skb->network_header) {
415 		memmove(skb_transport_header(skb),
416 			skb_network_header(skb), ihl);
417 		skb->network_header = skb->transport_header;
418 	}
419 	ipv6_hdr(skb)->payload_len = htons(skb->len + ihl -
420 					   sizeof(struct ipv6hdr));
421 	skb_reset_transport_header(skb);
422 	return 0;
423 #else
424 	WARN_ON_ONCE(1);
425 	return -EAFNOSUPPORT;
426 #endif
427 }
428 
429 static int xfrm_inner_mode_input(struct xfrm_state *x,
430 				 const struct xfrm_mode *inner_mode,
431 				 struct sk_buff *skb)
432 {
433 	switch (inner_mode->encap) {
434 	case XFRM_MODE_BEET:
435 	case XFRM_MODE_TUNNEL:
436 		return xfrm_prepare_input(x, skb);
437 	case XFRM_MODE_TRANSPORT:
438 		if (inner_mode->family == AF_INET)
439 			return xfrm4_transport_input(x, skb);
440 		if (inner_mode->family == AF_INET6)
441 			return xfrm6_transport_input(x, skb);
442 		break;
443 	case XFRM_MODE_ROUTEOPTIMIZATION:
444 		WARN_ON_ONCE(1);
445 		break;
446 	default:
447 		WARN_ON_ONCE(1);
448 		break;
449 	}
450 
451 	return -EOPNOTSUPP;
452 }
453 
454 int xfrm_input(struct sk_buff *skb, int nexthdr, __be32 spi, int encap_type)
455 {
456 	const struct xfrm_state_afinfo *afinfo;
457 	struct net *net = dev_net(skb->dev);
458 	const struct xfrm_mode *inner_mode;
459 	int err;
460 	__be32 seq;
461 	__be32 seq_hi;
462 	struct xfrm_state *x = NULL;
463 	xfrm_address_t *daddr;
464 	u32 mark = skb->mark;
465 	unsigned int family = AF_UNSPEC;
466 	int decaps = 0;
467 	int async = 0;
468 	bool xfrm_gro = false;
469 	bool crypto_done = false;
470 	struct xfrm_offload *xo = xfrm_offload(skb);
471 	struct sec_path *sp;
472 
473 	if (encap_type < 0) {
474 		x = xfrm_input_state(skb);
475 
476 		if (unlikely(x->km.state != XFRM_STATE_VALID)) {
477 			if (x->km.state == XFRM_STATE_ACQ)
478 				XFRM_INC_STATS(net, LINUX_MIB_XFRMACQUIREERROR);
479 			else
480 				XFRM_INC_STATS(net,
481 					       LINUX_MIB_XFRMINSTATEINVALID);
482 			goto drop;
483 		}
484 
485 		family = x->outer_mode.family;
486 
487 		/* An encap_type of -1 indicates async resumption. */
488 		if (encap_type == -1) {
489 			async = 1;
490 			seq = XFRM_SKB_CB(skb)->seq.input.low;
491 			goto resume;
492 		}
493 
494 		/* encap_type < -1 indicates a GRO call. */
495 		encap_type = 0;
496 		seq = XFRM_SPI_SKB_CB(skb)->seq;
497 
498 		if (xo && (xo->flags & CRYPTO_DONE)) {
499 			crypto_done = true;
500 			family = XFRM_SPI_SKB_CB(skb)->family;
501 
502 			if (!(xo->status & CRYPTO_SUCCESS)) {
503 				if (xo->status &
504 				    (CRYPTO_TRANSPORT_AH_AUTH_FAILED |
505 				     CRYPTO_TRANSPORT_ESP_AUTH_FAILED |
506 				     CRYPTO_TUNNEL_AH_AUTH_FAILED |
507 				     CRYPTO_TUNNEL_ESP_AUTH_FAILED)) {
508 
509 					xfrm_audit_state_icvfail(x, skb,
510 								 x->type->proto);
511 					x->stats.integrity_failed++;
512 					XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEPROTOERROR);
513 					goto drop;
514 				}
515 
516 				if (xo->status & CRYPTO_INVALID_PROTOCOL) {
517 					XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEPROTOERROR);
518 					goto drop;
519 				}
520 
521 				XFRM_INC_STATS(net, LINUX_MIB_XFRMINBUFFERERROR);
522 				goto drop;
523 			}
524 
525 			if ((err = xfrm_parse_spi(skb, nexthdr, &spi, &seq)) != 0) {
526 				XFRM_INC_STATS(net, LINUX_MIB_XFRMINHDRERROR);
527 				goto drop;
528 			}
529 		}
530 
531 		goto lock;
532 	}
533 
534 	family = XFRM_SPI_SKB_CB(skb)->family;
535 
536 	/* if tunnel is present override skb->mark value with tunnel i_key */
537 	switch (family) {
538 	case AF_INET:
539 		if (XFRM_TUNNEL_SKB_CB(skb)->tunnel.ip4)
540 			mark = be32_to_cpu(XFRM_TUNNEL_SKB_CB(skb)->tunnel.ip4->parms.i_key);
541 		break;
542 	case AF_INET6:
543 		if (XFRM_TUNNEL_SKB_CB(skb)->tunnel.ip6)
544 			mark = be32_to_cpu(XFRM_TUNNEL_SKB_CB(skb)->tunnel.ip6->parms.i_key);
545 		break;
546 	}
547 
548 	sp = secpath_set(skb);
549 	if (!sp) {
550 		XFRM_INC_STATS(net, LINUX_MIB_XFRMINERROR);
551 		goto drop;
552 	}
553 
554 	seq = 0;
555 	if (!spi && (err = xfrm_parse_spi(skb, nexthdr, &spi, &seq)) != 0) {
556 		secpath_reset(skb);
557 		XFRM_INC_STATS(net, LINUX_MIB_XFRMINHDRERROR);
558 		goto drop;
559 	}
560 
561 	daddr = (xfrm_address_t *)(skb_network_header(skb) +
562 				   XFRM_SPI_SKB_CB(skb)->daddroff);
563 	do {
564 		sp = skb_sec_path(skb);
565 
566 		if (sp->len == XFRM_MAX_DEPTH) {
567 			secpath_reset(skb);
568 			XFRM_INC_STATS(net, LINUX_MIB_XFRMINBUFFERERROR);
569 			goto drop;
570 		}
571 
572 		x = xfrm_state_lookup(net, mark, daddr, spi, nexthdr, family);
573 		if (x == NULL) {
574 			secpath_reset(skb);
575 			XFRM_INC_STATS(net, LINUX_MIB_XFRMINNOSTATES);
576 			xfrm_audit_state_notfound(skb, family, spi, seq);
577 			goto drop;
578 		}
579 
580 		skb->mark = xfrm_smark_get(skb->mark, x);
581 
582 		sp->xvec[sp->len++] = x;
583 
584 		skb_dst_force(skb);
585 		if (!skb_dst(skb)) {
586 			XFRM_INC_STATS(net, LINUX_MIB_XFRMINERROR);
587 			goto drop;
588 		}
589 
590 lock:
591 		spin_lock(&x->lock);
592 
593 		if (unlikely(x->km.state != XFRM_STATE_VALID)) {
594 			if (x->km.state == XFRM_STATE_ACQ)
595 				XFRM_INC_STATS(net, LINUX_MIB_XFRMACQUIREERROR);
596 			else
597 				XFRM_INC_STATS(net,
598 					       LINUX_MIB_XFRMINSTATEINVALID);
599 			goto drop_unlock;
600 		}
601 
602 		if ((x->encap ? x->encap->encap_type : 0) != encap_type) {
603 			XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMISMATCH);
604 			goto drop_unlock;
605 		}
606 
607 		if (x->repl->check(x, skb, seq)) {
608 			XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATESEQERROR);
609 			goto drop_unlock;
610 		}
611 
612 		if (xfrm_state_check_expire(x)) {
613 			XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEEXPIRED);
614 			goto drop_unlock;
615 		}
616 
617 		spin_unlock(&x->lock);
618 
619 		if (xfrm_tunnel_check(skb, x, family)) {
620 			XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMODEERROR);
621 			goto drop;
622 		}
623 
624 		seq_hi = htonl(xfrm_replay_seqhi(x, seq));
625 
626 		XFRM_SKB_CB(skb)->seq.input.low = seq;
627 		XFRM_SKB_CB(skb)->seq.input.hi = seq_hi;
628 
629 		dev_hold(skb->dev);
630 
631 		if (crypto_done)
632 			nexthdr = x->type_offload->input_tail(x, skb);
633 		else
634 			nexthdr = x->type->input(x, skb);
635 
636 		if (nexthdr == -EINPROGRESS)
637 			return 0;
638 resume:
639 		dev_put(skb->dev);
640 
641 		spin_lock(&x->lock);
642 		if (nexthdr <= 0) {
643 			if (nexthdr == -EBADMSG) {
644 				xfrm_audit_state_icvfail(x, skb,
645 							 x->type->proto);
646 				x->stats.integrity_failed++;
647 			}
648 			XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEPROTOERROR);
649 			goto drop_unlock;
650 		}
651 
652 		/* only the first xfrm gets the encap type */
653 		encap_type = 0;
654 
655 		if (async && x->repl->recheck(x, skb, seq)) {
656 			XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATESEQERROR);
657 			goto drop_unlock;
658 		}
659 
660 		x->repl->advance(x, seq);
661 
662 		x->curlft.bytes += skb->len;
663 		x->curlft.packets++;
664 
665 		spin_unlock(&x->lock);
666 
667 		XFRM_MODE_SKB_CB(skb)->protocol = nexthdr;
668 
669 		inner_mode = &x->inner_mode;
670 
671 		if (x->sel.family == AF_UNSPEC) {
672 			inner_mode = xfrm_ip2inner_mode(x, XFRM_MODE_SKB_CB(skb)->protocol);
673 			if (inner_mode == NULL) {
674 				XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMODEERROR);
675 				goto drop;
676 			}
677 		}
678 
679 		if (xfrm_inner_mode_input(x, inner_mode, skb)) {
680 			XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMODEERROR);
681 			goto drop;
682 		}
683 
684 		if (x->outer_mode.flags & XFRM_MODE_FLAG_TUNNEL) {
685 			decaps = 1;
686 			break;
687 		}
688 
689 		/*
690 		 * We need the inner address.  However, we only get here for
691 		 * transport mode so the outer address is identical.
692 		 */
693 		daddr = &x->id.daddr;
694 		family = x->outer_mode.family;
695 
696 		err = xfrm_parse_spi(skb, nexthdr, &spi, &seq);
697 		if (err < 0) {
698 			XFRM_INC_STATS(net, LINUX_MIB_XFRMINHDRERROR);
699 			goto drop;
700 		}
701 		crypto_done = false;
702 	} while (!err);
703 
704 	err = xfrm_rcv_cb(skb, family, x->type->proto, 0);
705 	if (err)
706 		goto drop;
707 
708 	nf_reset(skb);
709 
710 	if (decaps) {
711 		sp = skb_sec_path(skb);
712 		if (sp)
713 			sp->olen = 0;
714 		skb_dst_drop(skb);
715 		gro_cells_receive(&gro_cells, skb);
716 		return 0;
717 	} else {
718 		xo = xfrm_offload(skb);
719 		if (xo)
720 			xfrm_gro = xo->flags & XFRM_GRO;
721 
722 		err = -EAFNOSUPPORT;
723 		rcu_read_lock();
724 		afinfo = xfrm_state_afinfo_get_rcu(x->inner_mode.family);
725 		if (likely(afinfo))
726 			err = afinfo->transport_finish(skb, xfrm_gro || async);
727 		rcu_read_unlock();
728 		if (xfrm_gro) {
729 			sp = skb_sec_path(skb);
730 			if (sp)
731 				sp->olen = 0;
732 			skb_dst_drop(skb);
733 			gro_cells_receive(&gro_cells, skb);
734 			return err;
735 		}
736 
737 		return err;
738 	}
739 
740 drop_unlock:
741 	spin_unlock(&x->lock);
742 drop:
743 	xfrm_rcv_cb(skb, family, x && x->type ? x->type->proto : nexthdr, -1);
744 	kfree_skb(skb);
745 	return 0;
746 }
747 EXPORT_SYMBOL(xfrm_input);
748 
749 int xfrm_input_resume(struct sk_buff *skb, int nexthdr)
750 {
751 	return xfrm_input(skb, nexthdr, 0, -1);
752 }
753 EXPORT_SYMBOL(xfrm_input_resume);
754 
755 static void xfrm_trans_reinject(unsigned long data)
756 {
757 	struct xfrm_trans_tasklet *trans = (void *)data;
758 	struct sk_buff_head queue;
759 	struct sk_buff *skb;
760 
761 	__skb_queue_head_init(&queue);
762 	skb_queue_splice_init(&trans->queue, &queue);
763 
764 	while ((skb = __skb_dequeue(&queue)))
765 		XFRM_TRANS_SKB_CB(skb)->finish(dev_net(skb->dev), NULL, skb);
766 }
767 
768 int xfrm_trans_queue(struct sk_buff *skb,
769 		     int (*finish)(struct net *, struct sock *,
770 				   struct sk_buff *))
771 {
772 	struct xfrm_trans_tasklet *trans;
773 
774 	trans = this_cpu_ptr(&xfrm_trans_tasklet);
775 
776 	if (skb_queue_len(&trans->queue) >= netdev_max_backlog)
777 		return -ENOBUFS;
778 
779 	XFRM_TRANS_SKB_CB(skb)->finish = finish;
780 	__skb_queue_tail(&trans->queue, skb);
781 	tasklet_schedule(&trans->tasklet);
782 	return 0;
783 }
784 EXPORT_SYMBOL(xfrm_trans_queue);
785 
786 void __init xfrm_input_init(void)
787 {
788 	int err;
789 	int i;
790 
791 	init_dummy_netdev(&xfrm_napi_dev);
792 	err = gro_cells_init(&gro_cells, &xfrm_napi_dev);
793 	if (err)
794 		gro_cells.cells = NULL;
795 
796 	for_each_possible_cpu(i) {
797 		struct xfrm_trans_tasklet *trans;
798 
799 		trans = &per_cpu(xfrm_trans_tasklet, i);
800 		__skb_queue_head_init(&trans->queue);
801 		tasklet_init(&trans->tasklet, xfrm_trans_reinject,
802 			     (unsigned long)trans);
803 	}
804 }
805