xref: /openbmc/linux/net/xfrm/xfrm_input.c (revision f8e17c17)
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 	const struct xfrm_state_afinfo *afinfo;
357 	int err = -EAFNOSUPPORT;
358 
359 	rcu_read_lock();
360 	afinfo = xfrm_state_afinfo_get_rcu(x->outer_mode.family);
361 	if (likely(afinfo))
362 		err = afinfo->extract_input(x, skb);
363 	rcu_read_unlock();
364 
365 	if (err)
366 		return err;
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 			return -EAFNOSUPPORT;
372 	}
373 
374 	switch (inner_mode->family) {
375 	case AF_INET:
376 		skb->protocol = htons(ETH_P_IP);
377 		break;
378 	case AF_INET6:
379 		skb->protocol = htons(ETH_P_IPV6);
380 		break;
381 	default:
382 		WARN_ON_ONCE(1);
383 		break;
384 	}
385 
386 	return xfrm_inner_mode_encap_remove(x, inner_mode, skb);
387 }
388 
389 /* Remove encapsulation header.
390  *
391  * The IP header will be moved over the top of the encapsulation header.
392  *
393  * On entry, skb_transport_header() shall point to where the IP header
394  * should be and skb_network_header() shall be set to where the IP header
395  * currently is.  skb->data shall point to the start of the payload.
396  */
397 static int xfrm4_transport_input(struct xfrm_state *x, struct sk_buff *skb)
398 {
399 	int ihl = skb->data - skb_transport_header(skb);
400 
401 	if (skb->transport_header != skb->network_header) {
402 		memmove(skb_transport_header(skb),
403 			skb_network_header(skb), ihl);
404 		skb->network_header = skb->transport_header;
405 	}
406 	ip_hdr(skb)->tot_len = htons(skb->len + ihl);
407 	skb_reset_transport_header(skb);
408 	return 0;
409 }
410 
411 static int xfrm6_transport_input(struct xfrm_state *x, struct sk_buff *skb)
412 {
413 #if IS_ENABLED(CONFIG_IPV6)
414 	int ihl = skb->data - skb_transport_header(skb);
415 
416 	if (skb->transport_header != skb->network_header) {
417 		memmove(skb_transport_header(skb),
418 			skb_network_header(skb), ihl);
419 		skb->network_header = skb->transport_header;
420 	}
421 	ipv6_hdr(skb)->payload_len = htons(skb->len + ihl -
422 					   sizeof(struct ipv6hdr));
423 	skb_reset_transport_header(skb);
424 	return 0;
425 #else
426 	WARN_ON_ONCE(1);
427 	return -EAFNOSUPPORT;
428 #endif
429 }
430 
431 static int xfrm_inner_mode_input(struct xfrm_state *x,
432 				 const struct xfrm_mode *inner_mode,
433 				 struct sk_buff *skb)
434 {
435 	switch (inner_mode->encap) {
436 	case XFRM_MODE_BEET:
437 	case XFRM_MODE_TUNNEL:
438 		return xfrm_prepare_input(x, skb);
439 	case XFRM_MODE_TRANSPORT:
440 		if (inner_mode->family == AF_INET)
441 			return xfrm4_transport_input(x, skb);
442 		if (inner_mode->family == AF_INET6)
443 			return xfrm6_transport_input(x, skb);
444 		break;
445 	case XFRM_MODE_ROUTEOPTIMIZATION:
446 		WARN_ON_ONCE(1);
447 		break;
448 	default:
449 		WARN_ON_ONCE(1);
450 		break;
451 	}
452 
453 	return -EOPNOTSUPP;
454 }
455 
456 int xfrm_input(struct sk_buff *skb, int nexthdr, __be32 spi, int encap_type)
457 {
458 	const struct xfrm_state_afinfo *afinfo;
459 	struct net *net = dev_net(skb->dev);
460 	const struct xfrm_mode *inner_mode;
461 	int err;
462 	__be32 seq;
463 	__be32 seq_hi;
464 	struct xfrm_state *x = NULL;
465 	xfrm_address_t *daddr;
466 	u32 mark = skb->mark;
467 	unsigned int family = AF_UNSPEC;
468 	int decaps = 0;
469 	int async = 0;
470 	bool xfrm_gro = false;
471 	bool crypto_done = false;
472 	struct xfrm_offload *xo = xfrm_offload(skb);
473 	struct sec_path *sp;
474 
475 	if (encap_type < 0) {
476 		x = xfrm_input_state(skb);
477 
478 		if (unlikely(x->km.state != XFRM_STATE_VALID)) {
479 			if (x->km.state == XFRM_STATE_ACQ)
480 				XFRM_INC_STATS(net, LINUX_MIB_XFRMACQUIREERROR);
481 			else
482 				XFRM_INC_STATS(net,
483 					       LINUX_MIB_XFRMINSTATEINVALID);
484 
485 			if (encap_type == -1)
486 				dev_put(skb->dev);
487 			goto drop;
488 		}
489 
490 		family = x->outer_mode.family;
491 
492 		/* An encap_type of -1 indicates async resumption. */
493 		if (encap_type == -1) {
494 			async = 1;
495 			seq = XFRM_SKB_CB(skb)->seq.input.low;
496 			goto resume;
497 		}
498 
499 		/* encap_type < -1 indicates a GRO call. */
500 		encap_type = 0;
501 		seq = XFRM_SPI_SKB_CB(skb)->seq;
502 
503 		if (xo && (xo->flags & CRYPTO_DONE)) {
504 			crypto_done = true;
505 			family = XFRM_SPI_SKB_CB(skb)->family;
506 
507 			if (!(xo->status & CRYPTO_SUCCESS)) {
508 				if (xo->status &
509 				    (CRYPTO_TRANSPORT_AH_AUTH_FAILED |
510 				     CRYPTO_TRANSPORT_ESP_AUTH_FAILED |
511 				     CRYPTO_TUNNEL_AH_AUTH_FAILED |
512 				     CRYPTO_TUNNEL_ESP_AUTH_FAILED)) {
513 
514 					xfrm_audit_state_icvfail(x, skb,
515 								 x->type->proto);
516 					x->stats.integrity_failed++;
517 					XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEPROTOERROR);
518 					goto drop;
519 				}
520 
521 				if (xo->status & CRYPTO_INVALID_PROTOCOL) {
522 					XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEPROTOERROR);
523 					goto drop;
524 				}
525 
526 				XFRM_INC_STATS(net, LINUX_MIB_XFRMINBUFFERERROR);
527 				goto drop;
528 			}
529 
530 			if ((err = xfrm_parse_spi(skb, nexthdr, &spi, &seq)) != 0) {
531 				XFRM_INC_STATS(net, LINUX_MIB_XFRMINHDRERROR);
532 				goto drop;
533 			}
534 		}
535 
536 		goto lock;
537 	}
538 
539 	family = XFRM_SPI_SKB_CB(skb)->family;
540 
541 	/* if tunnel is present override skb->mark value with tunnel i_key */
542 	switch (family) {
543 	case AF_INET:
544 		if (XFRM_TUNNEL_SKB_CB(skb)->tunnel.ip4)
545 			mark = be32_to_cpu(XFRM_TUNNEL_SKB_CB(skb)->tunnel.ip4->parms.i_key);
546 		break;
547 	case AF_INET6:
548 		if (XFRM_TUNNEL_SKB_CB(skb)->tunnel.ip6)
549 			mark = be32_to_cpu(XFRM_TUNNEL_SKB_CB(skb)->tunnel.ip6->parms.i_key);
550 		break;
551 	}
552 
553 	sp = secpath_set(skb);
554 	if (!sp) {
555 		XFRM_INC_STATS(net, LINUX_MIB_XFRMINERROR);
556 		goto drop;
557 	}
558 
559 	seq = 0;
560 	if (!spi && (err = xfrm_parse_spi(skb, nexthdr, &spi, &seq)) != 0) {
561 		secpath_reset(skb);
562 		XFRM_INC_STATS(net, LINUX_MIB_XFRMINHDRERROR);
563 		goto drop;
564 	}
565 
566 	daddr = (xfrm_address_t *)(skb_network_header(skb) +
567 				   XFRM_SPI_SKB_CB(skb)->daddroff);
568 	do {
569 		sp = skb_sec_path(skb);
570 
571 		if (sp->len == XFRM_MAX_DEPTH) {
572 			secpath_reset(skb);
573 			XFRM_INC_STATS(net, LINUX_MIB_XFRMINBUFFERERROR);
574 			goto drop;
575 		}
576 
577 		x = xfrm_state_lookup(net, mark, daddr, spi, nexthdr, family);
578 		if (x == NULL) {
579 			secpath_reset(skb);
580 			XFRM_INC_STATS(net, LINUX_MIB_XFRMINNOSTATES);
581 			xfrm_audit_state_notfound(skb, family, spi, seq);
582 			goto drop;
583 		}
584 
585 		skb->mark = xfrm_smark_get(skb->mark, x);
586 
587 		sp->xvec[sp->len++] = x;
588 
589 		skb_dst_force(skb);
590 		if (!skb_dst(skb)) {
591 			XFRM_INC_STATS(net, LINUX_MIB_XFRMINERROR);
592 			goto drop;
593 		}
594 
595 lock:
596 		spin_lock(&x->lock);
597 
598 		if (unlikely(x->km.state != XFRM_STATE_VALID)) {
599 			if (x->km.state == XFRM_STATE_ACQ)
600 				XFRM_INC_STATS(net, LINUX_MIB_XFRMACQUIREERROR);
601 			else
602 				XFRM_INC_STATS(net,
603 					       LINUX_MIB_XFRMINSTATEINVALID);
604 			goto drop_unlock;
605 		}
606 
607 		if ((x->encap ? x->encap->encap_type : 0) != encap_type) {
608 			XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMISMATCH);
609 			goto drop_unlock;
610 		}
611 
612 		if (x->repl->check(x, skb, seq)) {
613 			XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATESEQERROR);
614 			goto drop_unlock;
615 		}
616 
617 		if (xfrm_state_check_expire(x)) {
618 			XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEEXPIRED);
619 			goto drop_unlock;
620 		}
621 
622 		spin_unlock(&x->lock);
623 
624 		if (xfrm_tunnel_check(skb, x, family)) {
625 			XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMODEERROR);
626 			goto drop;
627 		}
628 
629 		seq_hi = htonl(xfrm_replay_seqhi(x, seq));
630 
631 		XFRM_SKB_CB(skb)->seq.input.low = seq;
632 		XFRM_SKB_CB(skb)->seq.input.hi = seq_hi;
633 
634 		dev_hold(skb->dev);
635 
636 		if (crypto_done)
637 			nexthdr = x->type_offload->input_tail(x, skb);
638 		else
639 			nexthdr = x->type->input(x, skb);
640 
641 		if (nexthdr == -EINPROGRESS)
642 			return 0;
643 resume:
644 		dev_put(skb->dev);
645 
646 		spin_lock(&x->lock);
647 		if (nexthdr <= 0) {
648 			if (nexthdr == -EBADMSG) {
649 				xfrm_audit_state_icvfail(x, skb,
650 							 x->type->proto);
651 				x->stats.integrity_failed++;
652 			}
653 			XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEPROTOERROR);
654 			goto drop_unlock;
655 		}
656 
657 		/* only the first xfrm gets the encap type */
658 		encap_type = 0;
659 
660 		if (async && x->repl->recheck(x, skb, seq)) {
661 			XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATESEQERROR);
662 			goto drop_unlock;
663 		}
664 
665 		x->repl->advance(x, seq);
666 
667 		x->curlft.bytes += skb->len;
668 		x->curlft.packets++;
669 
670 		spin_unlock(&x->lock);
671 
672 		XFRM_MODE_SKB_CB(skb)->protocol = nexthdr;
673 
674 		inner_mode = &x->inner_mode;
675 
676 		if (x->sel.family == AF_UNSPEC) {
677 			inner_mode = xfrm_ip2inner_mode(x, XFRM_MODE_SKB_CB(skb)->protocol);
678 			if (inner_mode == NULL) {
679 				XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMODEERROR);
680 				goto drop;
681 			}
682 		}
683 
684 		if (xfrm_inner_mode_input(x, inner_mode, skb)) {
685 			XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMODEERROR);
686 			goto drop;
687 		}
688 
689 		if (x->outer_mode.flags & XFRM_MODE_FLAG_TUNNEL) {
690 			decaps = 1;
691 			break;
692 		}
693 
694 		/*
695 		 * We need the inner address.  However, we only get here for
696 		 * transport mode so the outer address is identical.
697 		 */
698 		daddr = &x->id.daddr;
699 		family = x->outer_mode.family;
700 
701 		err = xfrm_parse_spi(skb, nexthdr, &spi, &seq);
702 		if (err < 0) {
703 			XFRM_INC_STATS(net, LINUX_MIB_XFRMINHDRERROR);
704 			goto drop;
705 		}
706 		crypto_done = false;
707 	} while (!err);
708 
709 	err = xfrm_rcv_cb(skb, family, x->type->proto, 0);
710 	if (err)
711 		goto drop;
712 
713 	nf_reset_ct(skb);
714 
715 	if (decaps) {
716 		sp = skb_sec_path(skb);
717 		if (sp)
718 			sp->olen = 0;
719 		skb_dst_drop(skb);
720 		gro_cells_receive(&gro_cells, skb);
721 		return 0;
722 	} else {
723 		xo = xfrm_offload(skb);
724 		if (xo)
725 			xfrm_gro = xo->flags & XFRM_GRO;
726 
727 		err = -EAFNOSUPPORT;
728 		rcu_read_lock();
729 		afinfo = xfrm_state_afinfo_get_rcu(x->inner_mode.family);
730 		if (likely(afinfo))
731 			err = afinfo->transport_finish(skb, xfrm_gro || async);
732 		rcu_read_unlock();
733 		if (xfrm_gro) {
734 			sp = skb_sec_path(skb);
735 			if (sp)
736 				sp->olen = 0;
737 			skb_dst_drop(skb);
738 			gro_cells_receive(&gro_cells, skb);
739 			return err;
740 		}
741 
742 		return err;
743 	}
744 
745 drop_unlock:
746 	spin_unlock(&x->lock);
747 drop:
748 	xfrm_rcv_cb(skb, family, x && x->type ? x->type->proto : nexthdr, -1);
749 	kfree_skb(skb);
750 	return 0;
751 }
752 EXPORT_SYMBOL(xfrm_input);
753 
754 int xfrm_input_resume(struct sk_buff *skb, int nexthdr)
755 {
756 	return xfrm_input(skb, nexthdr, 0, -1);
757 }
758 EXPORT_SYMBOL(xfrm_input_resume);
759 
760 static void xfrm_trans_reinject(unsigned long data)
761 {
762 	struct xfrm_trans_tasklet *trans = (void *)data;
763 	struct sk_buff_head queue;
764 	struct sk_buff *skb;
765 
766 	__skb_queue_head_init(&queue);
767 	skb_queue_splice_init(&trans->queue, &queue);
768 
769 	while ((skb = __skb_dequeue(&queue)))
770 		XFRM_TRANS_SKB_CB(skb)->finish(XFRM_TRANS_SKB_CB(skb)->net,
771 					       NULL, skb);
772 }
773 
774 int xfrm_trans_queue_net(struct net *net, struct sk_buff *skb,
775 			 int (*finish)(struct net *, struct sock *,
776 				       struct sk_buff *))
777 {
778 	struct xfrm_trans_tasklet *trans;
779 
780 	trans = this_cpu_ptr(&xfrm_trans_tasklet);
781 
782 	if (skb_queue_len(&trans->queue) >= netdev_max_backlog)
783 		return -ENOBUFS;
784 
785 	BUILD_BUG_ON(sizeof(struct xfrm_trans_cb) > sizeof(skb->cb));
786 
787 	XFRM_TRANS_SKB_CB(skb)->finish = finish;
788 	XFRM_TRANS_SKB_CB(skb)->net = net;
789 	__skb_queue_tail(&trans->queue, skb);
790 	tasklet_schedule(&trans->tasklet);
791 	return 0;
792 }
793 EXPORT_SYMBOL(xfrm_trans_queue_net);
794 
795 int xfrm_trans_queue(struct sk_buff *skb,
796 		     int (*finish)(struct net *, struct sock *,
797 				   struct sk_buff *))
798 {
799 	return xfrm_trans_queue_net(dev_net(skb->dev), skb, finish);
800 }
801 EXPORT_SYMBOL(xfrm_trans_queue);
802 
803 void __init xfrm_input_init(void)
804 {
805 	int err;
806 	int i;
807 
808 	init_dummy_netdev(&xfrm_napi_dev);
809 	err = gro_cells_init(&gro_cells, &xfrm_napi_dev);
810 	if (err)
811 		gro_cells.cells = NULL;
812 
813 	for_each_possible_cpu(i) {
814 		struct xfrm_trans_tasklet *trans;
815 
816 		trans = &per_cpu(xfrm_trans_tasklet, i);
817 		__skb_queue_head_init(&trans->queue);
818 		tasklet_init(&trans->tasklet, xfrm_trans_reinject,
819 			     (unsigned long)trans);
820 	}
821 }
822