xref: /openbmc/linux/net/xfrm/xfrm_input.c (revision a72b9869)
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 #include <net/dst_metadata.h>
24 
25 #include "xfrm_inout.h"
26 
27 struct xfrm_trans_tasklet {
28 	struct work_struct work;
29 	spinlock_t queue_lock;
30 	struct sk_buff_head queue;
31 };
32 
33 struct xfrm_trans_cb {
34 	union {
35 		struct inet_skb_parm	h4;
36 #if IS_ENABLED(CONFIG_IPV6)
37 		struct inet6_skb_parm	h6;
38 #endif
39 	} header;
40 	int (*finish)(struct net *net, struct sock *sk, struct sk_buff *skb);
41 	struct net *net;
42 };
43 
44 #define XFRM_TRANS_SKB_CB(__skb) ((struct xfrm_trans_cb *)&((__skb)->cb[0]))
45 
46 static DEFINE_SPINLOCK(xfrm_input_afinfo_lock);
47 static struct xfrm_input_afinfo const __rcu *xfrm_input_afinfo[2][AF_INET6 + 1];
48 
49 static struct gro_cells gro_cells;
50 static struct net_device xfrm_napi_dev;
51 
52 static DEFINE_PER_CPU(struct xfrm_trans_tasklet, xfrm_trans_tasklet);
53 
54 int xfrm_input_register_afinfo(const struct xfrm_input_afinfo *afinfo)
55 {
56 	int err = 0;
57 
58 	if (WARN_ON(afinfo->family > AF_INET6))
59 		return -EAFNOSUPPORT;
60 
61 	spin_lock_bh(&xfrm_input_afinfo_lock);
62 	if (unlikely(xfrm_input_afinfo[afinfo->is_ipip][afinfo->family]))
63 		err = -EEXIST;
64 	else
65 		rcu_assign_pointer(xfrm_input_afinfo[afinfo->is_ipip][afinfo->family], afinfo);
66 	spin_unlock_bh(&xfrm_input_afinfo_lock);
67 	return err;
68 }
69 EXPORT_SYMBOL(xfrm_input_register_afinfo);
70 
71 int xfrm_input_unregister_afinfo(const struct xfrm_input_afinfo *afinfo)
72 {
73 	int err = 0;
74 
75 	spin_lock_bh(&xfrm_input_afinfo_lock);
76 	if (likely(xfrm_input_afinfo[afinfo->is_ipip][afinfo->family])) {
77 		if (unlikely(xfrm_input_afinfo[afinfo->is_ipip][afinfo->family] != afinfo))
78 			err = -EINVAL;
79 		else
80 			RCU_INIT_POINTER(xfrm_input_afinfo[afinfo->is_ipip][afinfo->family], NULL);
81 	}
82 	spin_unlock_bh(&xfrm_input_afinfo_lock);
83 	synchronize_rcu();
84 	return err;
85 }
86 EXPORT_SYMBOL(xfrm_input_unregister_afinfo);
87 
88 static const struct xfrm_input_afinfo *xfrm_input_get_afinfo(u8 family, bool is_ipip)
89 {
90 	const struct xfrm_input_afinfo *afinfo;
91 
92 	if (WARN_ON_ONCE(family > AF_INET6))
93 		return NULL;
94 
95 	rcu_read_lock();
96 	afinfo = rcu_dereference(xfrm_input_afinfo[is_ipip][family]);
97 	if (unlikely(!afinfo))
98 		rcu_read_unlock();
99 	return afinfo;
100 }
101 
102 static int xfrm_rcv_cb(struct sk_buff *skb, unsigned int family, u8 protocol,
103 		       int err)
104 {
105 	bool is_ipip = (protocol == IPPROTO_IPIP || protocol == IPPROTO_IPV6);
106 	const struct xfrm_input_afinfo *afinfo;
107 	int ret;
108 
109 	afinfo = xfrm_input_get_afinfo(family, is_ipip);
110 	if (!afinfo)
111 		return -EAFNOSUPPORT;
112 
113 	ret = afinfo->callback(skb, protocol, err);
114 	rcu_read_unlock();
115 
116 	return ret;
117 }
118 
119 struct sec_path *secpath_set(struct sk_buff *skb)
120 {
121 	struct sec_path *sp, *tmp = skb_ext_find(skb, SKB_EXT_SEC_PATH);
122 
123 	sp = skb_ext_add(skb, SKB_EXT_SEC_PATH);
124 	if (!sp)
125 		return NULL;
126 
127 	if (tmp) /* reused existing one (was COW'd if needed) */
128 		return sp;
129 
130 	/* allocated new secpath */
131 	memset(sp->ovec, 0, sizeof(sp->ovec));
132 	sp->olen = 0;
133 	sp->len = 0;
134 
135 	return sp;
136 }
137 EXPORT_SYMBOL(secpath_set);
138 
139 /* Fetch spi and seq from ipsec header */
140 
141 int xfrm_parse_spi(struct sk_buff *skb, u8 nexthdr, __be32 *spi, __be32 *seq)
142 {
143 	int offset, offset_seq;
144 	int hlen;
145 
146 	switch (nexthdr) {
147 	case IPPROTO_AH:
148 		hlen = sizeof(struct ip_auth_hdr);
149 		offset = offsetof(struct ip_auth_hdr, spi);
150 		offset_seq = offsetof(struct ip_auth_hdr, seq_no);
151 		break;
152 	case IPPROTO_ESP:
153 		hlen = sizeof(struct ip_esp_hdr);
154 		offset = offsetof(struct ip_esp_hdr, spi);
155 		offset_seq = offsetof(struct ip_esp_hdr, seq_no);
156 		break;
157 	case IPPROTO_COMP:
158 		if (!pskb_may_pull(skb, sizeof(struct ip_comp_hdr)))
159 			return -EINVAL;
160 		*spi = htonl(ntohs(*(__be16 *)(skb_transport_header(skb) + 2)));
161 		*seq = 0;
162 		return 0;
163 	default:
164 		return 1;
165 	}
166 
167 	if (!pskb_may_pull(skb, hlen))
168 		return -EINVAL;
169 
170 	*spi = *(__be32 *)(skb_transport_header(skb) + offset);
171 	*seq = *(__be32 *)(skb_transport_header(skb) + offset_seq);
172 	return 0;
173 }
174 EXPORT_SYMBOL(xfrm_parse_spi);
175 
176 static int xfrm4_remove_beet_encap(struct xfrm_state *x, struct sk_buff *skb)
177 {
178 	struct iphdr *iph;
179 	int optlen = 0;
180 	int err = -EINVAL;
181 
182 	if (unlikely(XFRM_MODE_SKB_CB(skb)->protocol == IPPROTO_BEETPH)) {
183 		struct ip_beet_phdr *ph;
184 		int phlen;
185 
186 		if (!pskb_may_pull(skb, sizeof(*ph)))
187 			goto out;
188 
189 		ph = (struct ip_beet_phdr *)skb->data;
190 
191 		phlen = sizeof(*ph) + ph->padlen;
192 		optlen = ph->hdrlen * 8 + (IPV4_BEET_PHMAXLEN - phlen);
193 		if (optlen < 0 || optlen & 3 || optlen > 250)
194 			goto out;
195 
196 		XFRM_MODE_SKB_CB(skb)->protocol = ph->nexthdr;
197 
198 		if (!pskb_may_pull(skb, phlen))
199 			goto out;
200 		__skb_pull(skb, phlen);
201 	}
202 
203 	skb_push(skb, sizeof(*iph));
204 	skb_reset_network_header(skb);
205 	skb_mac_header_rebuild(skb);
206 
207 	xfrm4_beet_make_header(skb);
208 
209 	iph = ip_hdr(skb);
210 
211 	iph->ihl += optlen / 4;
212 	iph->tot_len = htons(skb->len);
213 	iph->daddr = x->sel.daddr.a4;
214 	iph->saddr = x->sel.saddr.a4;
215 	iph->check = 0;
216 	iph->check = ip_fast_csum(skb_network_header(skb), iph->ihl);
217 	err = 0;
218 out:
219 	return err;
220 }
221 
222 static void ipip_ecn_decapsulate(struct sk_buff *skb)
223 {
224 	struct iphdr *inner_iph = ipip_hdr(skb);
225 
226 	if (INET_ECN_is_ce(XFRM_MODE_SKB_CB(skb)->tos))
227 		IP_ECN_set_ce(inner_iph);
228 }
229 
230 static int xfrm4_remove_tunnel_encap(struct xfrm_state *x, struct sk_buff *skb)
231 {
232 	int err = -EINVAL;
233 
234 	if (XFRM_MODE_SKB_CB(skb)->protocol != IPPROTO_IPIP)
235 		goto out;
236 
237 	if (!pskb_may_pull(skb, sizeof(struct iphdr)))
238 		goto out;
239 
240 	err = skb_unclone(skb, GFP_ATOMIC);
241 	if (err)
242 		goto out;
243 
244 	if (x->props.flags & XFRM_STATE_DECAP_DSCP)
245 		ipv4_copy_dscp(XFRM_MODE_SKB_CB(skb)->tos, ipip_hdr(skb));
246 	if (!(x->props.flags & XFRM_STATE_NOECN))
247 		ipip_ecn_decapsulate(skb);
248 
249 	skb_reset_network_header(skb);
250 	skb_mac_header_rebuild(skb);
251 	if (skb->mac_len)
252 		eth_hdr(skb)->h_proto = skb->protocol;
253 
254 	err = 0;
255 
256 out:
257 	return err;
258 }
259 
260 static void ipip6_ecn_decapsulate(struct sk_buff *skb)
261 {
262 	struct ipv6hdr *inner_iph = ipipv6_hdr(skb);
263 
264 	if (INET_ECN_is_ce(XFRM_MODE_SKB_CB(skb)->tos))
265 		IP6_ECN_set_ce(skb, inner_iph);
266 }
267 
268 static int xfrm6_remove_tunnel_encap(struct xfrm_state *x, struct sk_buff *skb)
269 {
270 	int err = -EINVAL;
271 
272 	if (XFRM_MODE_SKB_CB(skb)->protocol != IPPROTO_IPV6)
273 		goto out;
274 	if (!pskb_may_pull(skb, sizeof(struct ipv6hdr)))
275 		goto out;
276 
277 	err = skb_unclone(skb, GFP_ATOMIC);
278 	if (err)
279 		goto out;
280 
281 	if (x->props.flags & XFRM_STATE_DECAP_DSCP)
282 		ipv6_copy_dscp(XFRM_MODE_SKB_CB(skb)->tos, ipipv6_hdr(skb));
283 	if (!(x->props.flags & XFRM_STATE_NOECN))
284 		ipip6_ecn_decapsulate(skb);
285 
286 	skb_reset_network_header(skb);
287 	skb_mac_header_rebuild(skb);
288 	if (skb->mac_len)
289 		eth_hdr(skb)->h_proto = skb->protocol;
290 
291 	err = 0;
292 
293 out:
294 	return err;
295 }
296 
297 static int xfrm6_remove_beet_encap(struct xfrm_state *x, struct sk_buff *skb)
298 {
299 	struct ipv6hdr *ip6h;
300 	int size = sizeof(struct ipv6hdr);
301 	int err;
302 
303 	err = skb_cow_head(skb, size + skb->mac_len);
304 	if (err)
305 		goto out;
306 
307 	__skb_push(skb, size);
308 	skb_reset_network_header(skb);
309 	skb_mac_header_rebuild(skb);
310 
311 	xfrm6_beet_make_header(skb);
312 
313 	ip6h = ipv6_hdr(skb);
314 	ip6h->payload_len = htons(skb->len - size);
315 	ip6h->daddr = x->sel.daddr.in6;
316 	ip6h->saddr = x->sel.saddr.in6;
317 	err = 0;
318 out:
319 	return err;
320 }
321 
322 /* Remove encapsulation header.
323  *
324  * The IP header will be moved over the top of the encapsulation
325  * header.
326  *
327  * On entry, the transport header shall point to where the IP header
328  * should be and the network header shall be set to where the IP
329  * header currently is.  skb->data shall point to the start of the
330  * payload.
331  */
332 static int
333 xfrm_inner_mode_encap_remove(struct xfrm_state *x,
334 			     const struct xfrm_mode *inner_mode,
335 			     struct sk_buff *skb)
336 {
337 	switch (inner_mode->encap) {
338 	case XFRM_MODE_BEET:
339 		if (inner_mode->family == AF_INET)
340 			return xfrm4_remove_beet_encap(x, skb);
341 		if (inner_mode->family == AF_INET6)
342 			return xfrm6_remove_beet_encap(x, skb);
343 		break;
344 	case XFRM_MODE_TUNNEL:
345 		if (inner_mode->family == AF_INET)
346 			return xfrm4_remove_tunnel_encap(x, skb);
347 		if (inner_mode->family == AF_INET6)
348 			return xfrm6_remove_tunnel_encap(x, skb);
349 		break;
350 	}
351 
352 	WARN_ON_ONCE(1);
353 	return -EOPNOTSUPP;
354 }
355 
356 static int xfrm_prepare_input(struct xfrm_state *x, struct sk_buff *skb)
357 {
358 	const struct xfrm_mode *inner_mode = &x->inner_mode;
359 
360 	switch (x->outer_mode.family) {
361 	case AF_INET:
362 		xfrm4_extract_header(skb);
363 		break;
364 	case AF_INET6:
365 		xfrm6_extract_header(skb);
366 		break;
367 	default:
368 		WARN_ON_ONCE(1);
369 		return -EAFNOSUPPORT;
370 	}
371 
372 	if (x->sel.family == AF_UNSPEC) {
373 		inner_mode = xfrm_ip2inner_mode(x, XFRM_MODE_SKB_CB(skb)->protocol);
374 		if (!inner_mode)
375 			return -EAFNOSUPPORT;
376 	}
377 
378 	switch (inner_mode->family) {
379 	case AF_INET:
380 		skb->protocol = htons(ETH_P_IP);
381 		break;
382 	case AF_INET6:
383 		skb->protocol = htons(ETH_P_IPV6);
384 		break;
385 	default:
386 		WARN_ON_ONCE(1);
387 		break;
388 	}
389 
390 	return xfrm_inner_mode_encap_remove(x, inner_mode, skb);
391 }
392 
393 /* Remove encapsulation header.
394  *
395  * The IP header will be moved over the top of the encapsulation header.
396  *
397  * On entry, skb_transport_header() shall point to where the IP header
398  * should be and skb_network_header() shall be set to where the IP header
399  * currently is.  skb->data shall point to the start of the payload.
400  */
401 static int xfrm4_transport_input(struct xfrm_state *x, struct sk_buff *skb)
402 {
403 	int ihl = skb->data - skb_transport_header(skb);
404 
405 	if (skb->transport_header != skb->network_header) {
406 		memmove(skb_transport_header(skb),
407 			skb_network_header(skb), ihl);
408 		skb->network_header = skb->transport_header;
409 	}
410 	ip_hdr(skb)->tot_len = htons(skb->len + ihl);
411 	skb_reset_transport_header(skb);
412 	return 0;
413 }
414 
415 static int xfrm6_transport_input(struct xfrm_state *x, struct sk_buff *skb)
416 {
417 #if IS_ENABLED(CONFIG_IPV6)
418 	int ihl = skb->data - skb_transport_header(skb);
419 
420 	if (skb->transport_header != skb->network_header) {
421 		memmove(skb_transport_header(skb),
422 			skb_network_header(skb), ihl);
423 		skb->network_header = skb->transport_header;
424 	}
425 	ipv6_hdr(skb)->payload_len = htons(skb->len + ihl -
426 					   sizeof(struct ipv6hdr));
427 	skb_reset_transport_header(skb);
428 	return 0;
429 #else
430 	WARN_ON_ONCE(1);
431 	return -EAFNOSUPPORT;
432 #endif
433 }
434 
435 static int xfrm_inner_mode_input(struct xfrm_state *x,
436 				 const struct xfrm_mode *inner_mode,
437 				 struct sk_buff *skb)
438 {
439 	switch (inner_mode->encap) {
440 	case XFRM_MODE_BEET:
441 	case XFRM_MODE_TUNNEL:
442 		return xfrm_prepare_input(x, skb);
443 	case XFRM_MODE_TRANSPORT:
444 		if (inner_mode->family == AF_INET)
445 			return xfrm4_transport_input(x, skb);
446 		if (inner_mode->family == AF_INET6)
447 			return xfrm6_transport_input(x, skb);
448 		break;
449 	case XFRM_MODE_ROUTEOPTIMIZATION:
450 		WARN_ON_ONCE(1);
451 		break;
452 	default:
453 		WARN_ON_ONCE(1);
454 		break;
455 	}
456 
457 	return -EOPNOTSUPP;
458 }
459 
460 int xfrm_input(struct sk_buff *skb, int nexthdr, __be32 spi, int encap_type)
461 {
462 	const struct xfrm_state_afinfo *afinfo;
463 	struct net *net = dev_net(skb->dev);
464 	const struct xfrm_mode *inner_mode;
465 	int err;
466 	__be32 seq;
467 	__be32 seq_hi;
468 	struct xfrm_state *x = NULL;
469 	xfrm_address_t *daddr;
470 	u32 mark = skb->mark;
471 	unsigned int family = AF_UNSPEC;
472 	int decaps = 0;
473 	int async = 0;
474 	bool xfrm_gro = false;
475 	bool crypto_done = false;
476 	struct xfrm_offload *xo = xfrm_offload(skb);
477 	struct sec_path *sp;
478 
479 	if (encap_type < 0) {
480 		x = xfrm_input_state(skb);
481 
482 		if (unlikely(x->km.state != XFRM_STATE_VALID)) {
483 			if (x->km.state == XFRM_STATE_ACQ)
484 				XFRM_INC_STATS(net, LINUX_MIB_XFRMACQUIREERROR);
485 			else
486 				XFRM_INC_STATS(net,
487 					       LINUX_MIB_XFRMINSTATEINVALID);
488 
489 			if (encap_type == -1)
490 				dev_put(skb->dev);
491 			goto drop;
492 		}
493 
494 		family = x->outer_mode.family;
495 
496 		/* An encap_type of -1 indicates async resumption. */
497 		if (encap_type == -1) {
498 			async = 1;
499 			seq = XFRM_SKB_CB(skb)->seq.input.low;
500 			goto resume;
501 		}
502 
503 		/* encap_type < -1 indicates a GRO call. */
504 		encap_type = 0;
505 		seq = XFRM_SPI_SKB_CB(skb)->seq;
506 
507 		if (xo && (xo->flags & CRYPTO_DONE)) {
508 			crypto_done = true;
509 			family = XFRM_SPI_SKB_CB(skb)->family;
510 
511 			if (!(xo->status & CRYPTO_SUCCESS)) {
512 				if (xo->status &
513 				    (CRYPTO_TRANSPORT_AH_AUTH_FAILED |
514 				     CRYPTO_TRANSPORT_ESP_AUTH_FAILED |
515 				     CRYPTO_TUNNEL_AH_AUTH_FAILED |
516 				     CRYPTO_TUNNEL_ESP_AUTH_FAILED)) {
517 
518 					xfrm_audit_state_icvfail(x, skb,
519 								 x->type->proto);
520 					x->stats.integrity_failed++;
521 					XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEPROTOERROR);
522 					goto drop;
523 				}
524 
525 				if (xo->status & CRYPTO_INVALID_PROTOCOL) {
526 					XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEPROTOERROR);
527 					goto drop;
528 				}
529 
530 				XFRM_INC_STATS(net, LINUX_MIB_XFRMINBUFFERERROR);
531 				goto drop;
532 			}
533 
534 			if (xfrm_parse_spi(skb, nexthdr, &spi, &seq)) {
535 				XFRM_INC_STATS(net, LINUX_MIB_XFRMINHDRERROR);
536 				goto drop;
537 			}
538 		}
539 
540 		goto lock;
541 	}
542 
543 	family = XFRM_SPI_SKB_CB(skb)->family;
544 
545 	/* if tunnel is present override skb->mark value with tunnel i_key */
546 	switch (family) {
547 	case AF_INET:
548 		if (XFRM_TUNNEL_SKB_CB(skb)->tunnel.ip4)
549 			mark = be32_to_cpu(XFRM_TUNNEL_SKB_CB(skb)->tunnel.ip4->parms.i_key);
550 		break;
551 	case AF_INET6:
552 		if (XFRM_TUNNEL_SKB_CB(skb)->tunnel.ip6)
553 			mark = be32_to_cpu(XFRM_TUNNEL_SKB_CB(skb)->tunnel.ip6->parms.i_key);
554 		break;
555 	}
556 
557 	sp = secpath_set(skb);
558 	if (!sp) {
559 		XFRM_INC_STATS(net, LINUX_MIB_XFRMINERROR);
560 		goto drop;
561 	}
562 
563 	seq = 0;
564 	if (!spi && xfrm_parse_spi(skb, nexthdr, &spi, &seq)) {
565 		secpath_reset(skb);
566 		XFRM_INC_STATS(net, LINUX_MIB_XFRMINHDRERROR);
567 		goto drop;
568 	}
569 
570 	daddr = (xfrm_address_t *)(skb_network_header(skb) +
571 				   XFRM_SPI_SKB_CB(skb)->daddroff);
572 	do {
573 		sp = skb_sec_path(skb);
574 
575 		if (sp->len == XFRM_MAX_DEPTH) {
576 			secpath_reset(skb);
577 			XFRM_INC_STATS(net, LINUX_MIB_XFRMINBUFFERERROR);
578 			goto drop;
579 		}
580 
581 		x = xfrm_state_lookup(net, mark, daddr, spi, nexthdr, family);
582 		if (x == NULL) {
583 			secpath_reset(skb);
584 			XFRM_INC_STATS(net, LINUX_MIB_XFRMINNOSTATES);
585 			xfrm_audit_state_notfound(skb, family, spi, seq);
586 			goto drop;
587 		}
588 
589 		skb->mark = xfrm_smark_get(skb->mark, x);
590 
591 		sp->xvec[sp->len++] = x;
592 
593 		skb_dst_force(skb);
594 		if (!skb_dst(skb)) {
595 			XFRM_INC_STATS(net, LINUX_MIB_XFRMINERROR);
596 			goto drop;
597 		}
598 
599 lock:
600 		spin_lock(&x->lock);
601 
602 		if (unlikely(x->km.state != XFRM_STATE_VALID)) {
603 			if (x->km.state == XFRM_STATE_ACQ)
604 				XFRM_INC_STATS(net, LINUX_MIB_XFRMACQUIREERROR);
605 			else
606 				XFRM_INC_STATS(net,
607 					       LINUX_MIB_XFRMINSTATEINVALID);
608 			goto drop_unlock;
609 		}
610 
611 		if ((x->encap ? x->encap->encap_type : 0) != encap_type) {
612 			XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMISMATCH);
613 			goto drop_unlock;
614 		}
615 
616 		if (xfrm_replay_check(x, skb, seq)) {
617 			XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATESEQERROR);
618 			goto drop_unlock;
619 		}
620 
621 		if (xfrm_state_check_expire(x)) {
622 			XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEEXPIRED);
623 			goto drop_unlock;
624 		}
625 
626 		spin_unlock(&x->lock);
627 
628 		if (xfrm_tunnel_check(skb, x, family)) {
629 			XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMODEERROR);
630 			goto drop;
631 		}
632 
633 		seq_hi = htonl(xfrm_replay_seqhi(x, seq));
634 
635 		XFRM_SKB_CB(skb)->seq.input.low = seq;
636 		XFRM_SKB_CB(skb)->seq.input.hi = seq_hi;
637 
638 		dev_hold(skb->dev);
639 
640 		if (crypto_done)
641 			nexthdr = x->type_offload->input_tail(x, skb);
642 		else
643 			nexthdr = x->type->input(x, skb);
644 
645 		if (nexthdr == -EINPROGRESS)
646 			return 0;
647 resume:
648 		dev_put(skb->dev);
649 
650 		spin_lock(&x->lock);
651 		if (nexthdr < 0) {
652 			if (nexthdr == -EBADMSG) {
653 				xfrm_audit_state_icvfail(x, skb,
654 							 x->type->proto);
655 				x->stats.integrity_failed++;
656 			}
657 			XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEPROTOERROR);
658 			goto drop_unlock;
659 		}
660 
661 		/* only the first xfrm gets the encap type */
662 		encap_type = 0;
663 
664 		if (xfrm_replay_recheck(x, skb, seq)) {
665 			XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATESEQERROR);
666 			goto drop_unlock;
667 		}
668 
669 		xfrm_replay_advance(x, seq);
670 
671 		x->curlft.bytes += skb->len;
672 		x->curlft.packets++;
673 		x->lastused = ktime_get_real_seconds();
674 
675 		spin_unlock(&x->lock);
676 
677 		XFRM_MODE_SKB_CB(skb)->protocol = nexthdr;
678 
679 		inner_mode = &x->inner_mode;
680 
681 		if (x->sel.family == AF_UNSPEC) {
682 			inner_mode = xfrm_ip2inner_mode(x, XFRM_MODE_SKB_CB(skb)->protocol);
683 			if (inner_mode == NULL) {
684 				XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMODEERROR);
685 				goto drop;
686 			}
687 		}
688 
689 		if (xfrm_inner_mode_input(x, inner_mode, skb)) {
690 			XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMODEERROR);
691 			goto drop;
692 		}
693 
694 		if (x->outer_mode.flags & XFRM_MODE_FLAG_TUNNEL) {
695 			decaps = 1;
696 			break;
697 		}
698 
699 		/*
700 		 * We need the inner address.  However, we only get here for
701 		 * transport mode so the outer address is identical.
702 		 */
703 		daddr = &x->id.daddr;
704 		family = x->outer_mode.family;
705 
706 		err = xfrm_parse_spi(skb, nexthdr, &spi, &seq);
707 		if (err < 0) {
708 			XFRM_INC_STATS(net, LINUX_MIB_XFRMINHDRERROR);
709 			goto drop;
710 		}
711 		crypto_done = false;
712 	} while (!err);
713 
714 	err = xfrm_rcv_cb(skb, family, x->type->proto, 0);
715 	if (err)
716 		goto drop;
717 
718 	nf_reset_ct(skb);
719 
720 	if (decaps) {
721 		sp = skb_sec_path(skb);
722 		if (sp)
723 			sp->olen = 0;
724 		if (skb_valid_dst(skb))
725 			skb_dst_drop(skb);
726 		gro_cells_receive(&gro_cells, skb);
727 		return 0;
728 	} else {
729 		xo = xfrm_offload(skb);
730 		if (xo)
731 			xfrm_gro = xo->flags & XFRM_GRO;
732 
733 		err = -EAFNOSUPPORT;
734 		rcu_read_lock();
735 		afinfo = xfrm_state_afinfo_get_rcu(x->inner_mode.family);
736 		if (likely(afinfo))
737 			err = afinfo->transport_finish(skb, xfrm_gro || async);
738 		rcu_read_unlock();
739 		if (xfrm_gro) {
740 			sp = skb_sec_path(skb);
741 			if (sp)
742 				sp->olen = 0;
743 			if (skb_valid_dst(skb))
744 				skb_dst_drop(skb);
745 			gro_cells_receive(&gro_cells, skb);
746 			return err;
747 		}
748 
749 		return err;
750 	}
751 
752 drop_unlock:
753 	spin_unlock(&x->lock);
754 drop:
755 	xfrm_rcv_cb(skb, family, x && x->type ? x->type->proto : nexthdr, -1);
756 	kfree_skb(skb);
757 	return 0;
758 }
759 EXPORT_SYMBOL(xfrm_input);
760 
761 int xfrm_input_resume(struct sk_buff *skb, int nexthdr)
762 {
763 	return xfrm_input(skb, nexthdr, 0, -1);
764 }
765 EXPORT_SYMBOL(xfrm_input_resume);
766 
767 static void xfrm_trans_reinject(struct work_struct *work)
768 {
769 	struct xfrm_trans_tasklet *trans = container_of(work, struct xfrm_trans_tasklet, work);
770 	struct sk_buff_head queue;
771 	struct sk_buff *skb;
772 
773 	__skb_queue_head_init(&queue);
774 	spin_lock_bh(&trans->queue_lock);
775 	skb_queue_splice_init(&trans->queue, &queue);
776 	spin_unlock_bh(&trans->queue_lock);
777 
778 	local_bh_disable();
779 	while ((skb = __skb_dequeue(&queue)))
780 		XFRM_TRANS_SKB_CB(skb)->finish(XFRM_TRANS_SKB_CB(skb)->net,
781 					       NULL, skb);
782 	local_bh_enable();
783 }
784 
785 int xfrm_trans_queue_net(struct net *net, struct sk_buff *skb,
786 			 int (*finish)(struct net *, struct sock *,
787 				       struct sk_buff *))
788 {
789 	struct xfrm_trans_tasklet *trans;
790 
791 	trans = this_cpu_ptr(&xfrm_trans_tasklet);
792 
793 	if (skb_queue_len(&trans->queue) >= READ_ONCE(netdev_max_backlog))
794 		return -ENOBUFS;
795 
796 	BUILD_BUG_ON(sizeof(struct xfrm_trans_cb) > sizeof(skb->cb));
797 
798 	XFRM_TRANS_SKB_CB(skb)->finish = finish;
799 	XFRM_TRANS_SKB_CB(skb)->net = net;
800 	spin_lock_bh(&trans->queue_lock);
801 	__skb_queue_tail(&trans->queue, skb);
802 	spin_unlock_bh(&trans->queue_lock);
803 	schedule_work(&trans->work);
804 	return 0;
805 }
806 EXPORT_SYMBOL(xfrm_trans_queue_net);
807 
808 int xfrm_trans_queue(struct sk_buff *skb,
809 		     int (*finish)(struct net *, struct sock *,
810 				   struct sk_buff *))
811 {
812 	return xfrm_trans_queue_net(dev_net(skb->dev), skb, finish);
813 }
814 EXPORT_SYMBOL(xfrm_trans_queue);
815 
816 void __init xfrm_input_init(void)
817 {
818 	int err;
819 	int i;
820 
821 	init_dummy_netdev(&xfrm_napi_dev);
822 	err = gro_cells_init(&gro_cells, &xfrm_napi_dev);
823 	if (err)
824 		gro_cells.cells = NULL;
825 
826 	for_each_possible_cpu(i) {
827 		struct xfrm_trans_tasklet *trans;
828 
829 		trans = &per_cpu(xfrm_trans_tasklet, i);
830 		spin_lock_init(&trans->queue_lock);
831 		__skb_queue_head_init(&trans->queue);
832 		INIT_WORK(&trans->work, xfrm_trans_reinject);
833 	}
834 }
835