xref: /openbmc/linux/net/xfrm/xfrm_input.c (revision 726ccdba)
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 (!pskb_may_pull(skb, sizeof(struct iphdr)))
235 		goto out;
236 
237 	err = skb_unclone(skb, GFP_ATOMIC);
238 	if (err)
239 		goto out;
240 
241 	if (x->props.flags & XFRM_STATE_DECAP_DSCP)
242 		ipv4_copy_dscp(XFRM_MODE_SKB_CB(skb)->tos, ipip_hdr(skb));
243 	if (!(x->props.flags & XFRM_STATE_NOECN))
244 		ipip_ecn_decapsulate(skb);
245 
246 	skb_reset_network_header(skb);
247 	skb_mac_header_rebuild(skb);
248 	if (skb->mac_len)
249 		eth_hdr(skb)->h_proto = skb->protocol;
250 
251 	err = 0;
252 
253 out:
254 	return err;
255 }
256 
257 static void ipip6_ecn_decapsulate(struct sk_buff *skb)
258 {
259 	struct ipv6hdr *inner_iph = ipipv6_hdr(skb);
260 
261 	if (INET_ECN_is_ce(XFRM_MODE_SKB_CB(skb)->tos))
262 		IP6_ECN_set_ce(skb, inner_iph);
263 }
264 
265 static int xfrm6_remove_tunnel_encap(struct xfrm_state *x, struct sk_buff *skb)
266 {
267 	int err = -EINVAL;
268 
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(XFRM_MODE_SKB_CB(skb)->tos, ipipv6_hdr(skb));
278 	if (!(x->props.flags & XFRM_STATE_NOECN))
279 		ipip6_ecn_decapsulate(skb);
280 
281 	skb_reset_network_header(skb);
282 	skb_mac_header_rebuild(skb);
283 	if (skb->mac_len)
284 		eth_hdr(skb)->h_proto = skb->protocol;
285 
286 	err = 0;
287 
288 out:
289 	return err;
290 }
291 
292 static int xfrm6_remove_beet_encap(struct xfrm_state *x, struct sk_buff *skb)
293 {
294 	struct ipv6hdr *ip6h;
295 	int size = sizeof(struct ipv6hdr);
296 	int err;
297 
298 	err = skb_cow_head(skb, size + skb->mac_len);
299 	if (err)
300 		goto out;
301 
302 	__skb_push(skb, size);
303 	skb_reset_network_header(skb);
304 	skb_mac_header_rebuild(skb);
305 
306 	xfrm6_beet_make_header(skb);
307 
308 	ip6h = ipv6_hdr(skb);
309 	ip6h->payload_len = htons(skb->len - size);
310 	ip6h->daddr = x->sel.daddr.in6;
311 	ip6h->saddr = x->sel.saddr.in6;
312 	err = 0;
313 out:
314 	return err;
315 }
316 
317 /* Remove encapsulation header.
318  *
319  * The IP header will be moved over the top of the encapsulation
320  * header.
321  *
322  * On entry, the transport header shall point to where the IP header
323  * should be and the network header shall be set to where the IP
324  * header currently is.  skb->data shall point to the start of the
325  * payload.
326  */
327 static int
328 xfrm_inner_mode_encap_remove(struct xfrm_state *x,
329 			     struct sk_buff *skb)
330 {
331 	switch (x->props.mode) {
332 	case XFRM_MODE_BEET:
333 		switch (XFRM_MODE_SKB_CB(skb)->protocol) {
334 		case IPPROTO_IPIP:
335 		case IPPROTO_BEETPH:
336 			return xfrm4_remove_beet_encap(x, skb);
337 		case IPPROTO_IPV6:
338 			return xfrm6_remove_beet_encap(x, skb);
339 		}
340 		break;
341 	case XFRM_MODE_TUNNEL:
342 		switch (XFRM_MODE_SKB_CB(skb)->protocol) {
343 		case IPPROTO_IPIP:
344 			return xfrm4_remove_tunnel_encap(x, skb);
345 		case IPPROTO_IPV6:
346 			return xfrm6_remove_tunnel_encap(x, skb);
347 		break;
348 		}
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 	switch (x->props.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 	switch (XFRM_MODE_SKB_CB(skb)->protocol) {
370 	case IPPROTO_IPIP:
371 	case IPPROTO_BEETPH:
372 		skb->protocol = htons(ETH_P_IP);
373 		break;
374 	case IPPROTO_IPV6:
375 		skb->protocol = htons(ETH_P_IPV6);
376 		break;
377 	default:
378 		WARN_ON_ONCE(1);
379 		break;
380 	}
381 
382 	return xfrm_inner_mode_encap_remove(x, skb);
383 }
384 
385 /* Remove encapsulation header.
386  *
387  * The IP header will be moved over the top of the encapsulation header.
388  *
389  * On entry, skb_transport_header() shall point to where the IP header
390  * should be and skb_network_header() shall be set to where the IP header
391  * currently is.  skb->data shall point to the start of the payload.
392  */
393 static int xfrm4_transport_input(struct xfrm_state *x, struct sk_buff *skb)
394 {
395 	int ihl = skb->data - skb_transport_header(skb);
396 
397 	if (skb->transport_header != skb->network_header) {
398 		memmove(skb_transport_header(skb),
399 			skb_network_header(skb), ihl);
400 		skb->network_header = skb->transport_header;
401 	}
402 	ip_hdr(skb)->tot_len = htons(skb->len + ihl);
403 	skb_reset_transport_header(skb);
404 	return 0;
405 }
406 
407 static int xfrm6_transport_input(struct xfrm_state *x, struct sk_buff *skb)
408 {
409 #if IS_ENABLED(CONFIG_IPV6)
410 	int ihl = skb->data - skb_transport_header(skb);
411 
412 	if (skb->transport_header != skb->network_header) {
413 		memmove(skb_transport_header(skb),
414 			skb_network_header(skb), ihl);
415 		skb->network_header = skb->transport_header;
416 	}
417 	ipv6_hdr(skb)->payload_len = htons(skb->len + ihl -
418 					   sizeof(struct ipv6hdr));
419 	skb_reset_transport_header(skb);
420 	return 0;
421 #else
422 	WARN_ON_ONCE(1);
423 	return -EAFNOSUPPORT;
424 #endif
425 }
426 
427 static int xfrm_inner_mode_input(struct xfrm_state *x,
428 				 struct sk_buff *skb)
429 {
430 	switch (x->props.mode) {
431 	case XFRM_MODE_BEET:
432 	case XFRM_MODE_TUNNEL:
433 		return xfrm_prepare_input(x, skb);
434 	case XFRM_MODE_TRANSPORT:
435 		if (x->props.family == AF_INET)
436 			return xfrm4_transport_input(x, skb);
437 		if (x->props.family == AF_INET6)
438 			return xfrm6_transport_input(x, skb);
439 		break;
440 	case XFRM_MODE_ROUTEOPTIMIZATION:
441 		WARN_ON_ONCE(1);
442 		break;
443 	default:
444 		WARN_ON_ONCE(1);
445 		break;
446 	}
447 
448 	return -EOPNOTSUPP;
449 }
450 
451 int xfrm_input(struct sk_buff *skb, int nexthdr, __be32 spi, int encap_type)
452 {
453 	const struct xfrm_state_afinfo *afinfo;
454 	struct net *net = dev_net(skb->dev);
455 	int err;
456 	__be32 seq;
457 	__be32 seq_hi;
458 	struct xfrm_state *x = NULL;
459 	xfrm_address_t *daddr;
460 	u32 mark = skb->mark;
461 	unsigned int family = AF_UNSPEC;
462 	int decaps = 0;
463 	int async = 0;
464 	bool xfrm_gro = false;
465 	bool crypto_done = false;
466 	struct xfrm_offload *xo = xfrm_offload(skb);
467 	struct sec_path *sp;
468 
469 	if (encap_type < 0) {
470 		x = xfrm_input_state(skb);
471 
472 		if (unlikely(x->km.state != XFRM_STATE_VALID)) {
473 			if (x->km.state == XFRM_STATE_ACQ)
474 				XFRM_INC_STATS(net, LINUX_MIB_XFRMACQUIREERROR);
475 			else
476 				XFRM_INC_STATS(net,
477 					       LINUX_MIB_XFRMINSTATEINVALID);
478 
479 			if (encap_type == -1)
480 				dev_put(skb->dev);
481 			goto drop;
482 		}
483 
484 		family = x->props.family;
485 
486 		/* An encap_type of -1 indicates async resumption. */
487 		if (encap_type == -1) {
488 			async = 1;
489 			seq = XFRM_SKB_CB(skb)->seq.input.low;
490 			goto resume;
491 		}
492 
493 		/* encap_type < -1 indicates a GRO call. */
494 		encap_type = 0;
495 		seq = XFRM_SPI_SKB_CB(skb)->seq;
496 
497 		if (xo && (xo->flags & CRYPTO_DONE)) {
498 			crypto_done = true;
499 			family = XFRM_SPI_SKB_CB(skb)->family;
500 
501 			if (!(xo->status & CRYPTO_SUCCESS)) {
502 				if (xo->status &
503 				    (CRYPTO_TRANSPORT_AH_AUTH_FAILED |
504 				     CRYPTO_TRANSPORT_ESP_AUTH_FAILED |
505 				     CRYPTO_TUNNEL_AH_AUTH_FAILED |
506 				     CRYPTO_TUNNEL_ESP_AUTH_FAILED)) {
507 
508 					xfrm_audit_state_icvfail(x, skb,
509 								 x->type->proto);
510 					x->stats.integrity_failed++;
511 					XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEPROTOERROR);
512 					goto drop;
513 				}
514 
515 				if (xo->status & CRYPTO_INVALID_PROTOCOL) {
516 					XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEPROTOERROR);
517 					goto drop;
518 				}
519 
520 				XFRM_INC_STATS(net, LINUX_MIB_XFRMINBUFFERERROR);
521 				goto drop;
522 			}
523 
524 			if (xfrm_parse_spi(skb, nexthdr, &spi, &seq)) {
525 				XFRM_INC_STATS(net, LINUX_MIB_XFRMINHDRERROR);
526 				goto drop;
527 			}
528 		}
529 
530 		goto lock;
531 	}
532 
533 	family = XFRM_SPI_SKB_CB(skb)->family;
534 
535 	/* if tunnel is present override skb->mark value with tunnel i_key */
536 	switch (family) {
537 	case AF_INET:
538 		if (XFRM_TUNNEL_SKB_CB(skb)->tunnel.ip4)
539 			mark = be32_to_cpu(XFRM_TUNNEL_SKB_CB(skb)->tunnel.ip4->parms.i_key);
540 		break;
541 	case AF_INET6:
542 		if (XFRM_TUNNEL_SKB_CB(skb)->tunnel.ip6)
543 			mark = be32_to_cpu(XFRM_TUNNEL_SKB_CB(skb)->tunnel.ip6->parms.i_key);
544 		break;
545 	}
546 
547 	sp = secpath_set(skb);
548 	if (!sp) {
549 		XFRM_INC_STATS(net, LINUX_MIB_XFRMINERROR);
550 		goto drop;
551 	}
552 
553 	seq = 0;
554 	if (!spi && xfrm_parse_spi(skb, nexthdr, &spi, &seq)) {
555 		secpath_reset(skb);
556 		XFRM_INC_STATS(net, LINUX_MIB_XFRMINHDRERROR);
557 		goto drop;
558 	}
559 
560 	daddr = (xfrm_address_t *)(skb_network_header(skb) +
561 				   XFRM_SPI_SKB_CB(skb)->daddroff);
562 	do {
563 		sp = skb_sec_path(skb);
564 
565 		if (sp->len == XFRM_MAX_DEPTH) {
566 			secpath_reset(skb);
567 			XFRM_INC_STATS(net, LINUX_MIB_XFRMINBUFFERERROR);
568 			goto drop;
569 		}
570 
571 		x = xfrm_state_lookup(net, mark, daddr, spi, nexthdr, family);
572 		if (x == NULL) {
573 			secpath_reset(skb);
574 			XFRM_INC_STATS(net, LINUX_MIB_XFRMINNOSTATES);
575 			xfrm_audit_state_notfound(skb, family, spi, seq);
576 			goto drop;
577 		}
578 
579 		skb->mark = xfrm_smark_get(skb->mark, x);
580 
581 		sp->xvec[sp->len++] = x;
582 
583 		skb_dst_force(skb);
584 		if (!skb_dst(skb)) {
585 			XFRM_INC_STATS(net, LINUX_MIB_XFRMINERROR);
586 			goto drop;
587 		}
588 
589 lock:
590 		spin_lock(&x->lock);
591 
592 		if (unlikely(x->km.state != XFRM_STATE_VALID)) {
593 			if (x->km.state == XFRM_STATE_ACQ)
594 				XFRM_INC_STATS(net, LINUX_MIB_XFRMACQUIREERROR);
595 			else
596 				XFRM_INC_STATS(net,
597 					       LINUX_MIB_XFRMINSTATEINVALID);
598 			goto drop_unlock;
599 		}
600 
601 		if ((x->encap ? x->encap->encap_type : 0) != encap_type) {
602 			XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMISMATCH);
603 			goto drop_unlock;
604 		}
605 
606 		if (xfrm_replay_check(x, skb, seq)) {
607 			XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATESEQERROR);
608 			goto drop_unlock;
609 		}
610 
611 		if (xfrm_state_check_expire(x)) {
612 			XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEEXPIRED);
613 			goto drop_unlock;
614 		}
615 
616 		spin_unlock(&x->lock);
617 
618 		if (xfrm_tunnel_check(skb, x, family)) {
619 			XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMODEERROR);
620 			goto drop;
621 		}
622 
623 		seq_hi = htonl(xfrm_replay_seqhi(x, seq));
624 
625 		XFRM_SKB_CB(skb)->seq.input.low = seq;
626 		XFRM_SKB_CB(skb)->seq.input.hi = seq_hi;
627 
628 		dev_hold(skb->dev);
629 
630 		if (crypto_done)
631 			nexthdr = x->type_offload->input_tail(x, skb);
632 		else
633 			nexthdr = x->type->input(x, skb);
634 
635 		if (nexthdr == -EINPROGRESS)
636 			return 0;
637 resume:
638 		dev_put(skb->dev);
639 
640 		spin_lock(&x->lock);
641 		if (nexthdr < 0) {
642 			if (nexthdr == -EBADMSG) {
643 				xfrm_audit_state_icvfail(x, skb,
644 							 x->type->proto);
645 				x->stats.integrity_failed++;
646 			}
647 			XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEPROTOERROR);
648 			goto drop_unlock;
649 		}
650 
651 		/* only the first xfrm gets the encap type */
652 		encap_type = 0;
653 
654 		if (xfrm_replay_recheck(x, skb, seq)) {
655 			XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATESEQERROR);
656 			goto drop_unlock;
657 		}
658 
659 		xfrm_replay_advance(x, seq);
660 
661 		x->curlft.bytes += skb->len;
662 		x->curlft.packets++;
663 		x->lastused = ktime_get_real_seconds();
664 
665 		spin_unlock(&x->lock);
666 
667 		XFRM_MODE_SKB_CB(skb)->protocol = nexthdr;
668 
669 		if (xfrm_inner_mode_input(x, skb)) {
670 			XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMODEERROR);
671 			goto drop;
672 		}
673 
674 		if (x->outer_mode.flags & XFRM_MODE_FLAG_TUNNEL) {
675 			decaps = 1;
676 			break;
677 		}
678 
679 		/*
680 		 * We need the inner address.  However, we only get here for
681 		 * transport mode so the outer address is identical.
682 		 */
683 		daddr = &x->id.daddr;
684 		family = x->props.family;
685 
686 		err = xfrm_parse_spi(skb, nexthdr, &spi, &seq);
687 		if (err < 0) {
688 			XFRM_INC_STATS(net, LINUX_MIB_XFRMINHDRERROR);
689 			goto drop;
690 		}
691 		crypto_done = false;
692 	} while (!err);
693 
694 	err = xfrm_rcv_cb(skb, family, x->type->proto, 0);
695 	if (err)
696 		goto drop;
697 
698 	nf_reset_ct(skb);
699 
700 	if (decaps) {
701 		sp = skb_sec_path(skb);
702 		if (sp)
703 			sp->olen = 0;
704 		if (skb_valid_dst(skb))
705 			skb_dst_drop(skb);
706 		gro_cells_receive(&gro_cells, skb);
707 		return 0;
708 	} else {
709 		xo = xfrm_offload(skb);
710 		if (xo)
711 			xfrm_gro = xo->flags & XFRM_GRO;
712 
713 		err = -EAFNOSUPPORT;
714 		rcu_read_lock();
715 		afinfo = xfrm_state_afinfo_get_rcu(x->props.family);
716 		if (likely(afinfo))
717 			err = afinfo->transport_finish(skb, xfrm_gro || async);
718 		rcu_read_unlock();
719 		if (xfrm_gro) {
720 			sp = skb_sec_path(skb);
721 			if (sp)
722 				sp->olen = 0;
723 			if (skb_valid_dst(skb))
724 				skb_dst_drop(skb);
725 			gro_cells_receive(&gro_cells, skb);
726 			return err;
727 		}
728 
729 		return err;
730 	}
731 
732 drop_unlock:
733 	spin_unlock(&x->lock);
734 drop:
735 	xfrm_rcv_cb(skb, family, x && x->type ? x->type->proto : nexthdr, -1);
736 	kfree_skb(skb);
737 	return 0;
738 }
739 EXPORT_SYMBOL(xfrm_input);
740 
741 int xfrm_input_resume(struct sk_buff *skb, int nexthdr)
742 {
743 	return xfrm_input(skb, nexthdr, 0, -1);
744 }
745 EXPORT_SYMBOL(xfrm_input_resume);
746 
747 static void xfrm_trans_reinject(struct work_struct *work)
748 {
749 	struct xfrm_trans_tasklet *trans = container_of(work, struct xfrm_trans_tasklet, work);
750 	struct sk_buff_head queue;
751 	struct sk_buff *skb;
752 
753 	__skb_queue_head_init(&queue);
754 	spin_lock_bh(&trans->queue_lock);
755 	skb_queue_splice_init(&trans->queue, &queue);
756 	spin_unlock_bh(&trans->queue_lock);
757 
758 	local_bh_disable();
759 	while ((skb = __skb_dequeue(&queue)))
760 		XFRM_TRANS_SKB_CB(skb)->finish(XFRM_TRANS_SKB_CB(skb)->net,
761 					       NULL, skb);
762 	local_bh_enable();
763 }
764 
765 int xfrm_trans_queue_net(struct net *net, struct sk_buff *skb,
766 			 int (*finish)(struct net *, struct sock *,
767 				       struct sk_buff *))
768 {
769 	struct xfrm_trans_tasklet *trans;
770 
771 	trans = this_cpu_ptr(&xfrm_trans_tasklet);
772 
773 	if (skb_queue_len(&trans->queue) >= READ_ONCE(netdev_max_backlog))
774 		return -ENOBUFS;
775 
776 	BUILD_BUG_ON(sizeof(struct xfrm_trans_cb) > sizeof(skb->cb));
777 
778 	XFRM_TRANS_SKB_CB(skb)->finish = finish;
779 	XFRM_TRANS_SKB_CB(skb)->net = net;
780 	spin_lock_bh(&trans->queue_lock);
781 	__skb_queue_tail(&trans->queue, skb);
782 	spin_unlock_bh(&trans->queue_lock);
783 	schedule_work(&trans->work);
784 	return 0;
785 }
786 EXPORT_SYMBOL(xfrm_trans_queue_net);
787 
788 int xfrm_trans_queue(struct sk_buff *skb,
789 		     int (*finish)(struct net *, struct sock *,
790 				   struct sk_buff *))
791 {
792 	return xfrm_trans_queue_net(dev_net(skb->dev), skb, finish);
793 }
794 EXPORT_SYMBOL(xfrm_trans_queue);
795 
796 void __init xfrm_input_init(void)
797 {
798 	int err;
799 	int i;
800 
801 	init_dummy_netdev(&xfrm_napi_dev);
802 	err = gro_cells_init(&gro_cells, &xfrm_napi_dev);
803 	if (err)
804 		gro_cells.cells = NULL;
805 
806 	for_each_possible_cpu(i) {
807 		struct xfrm_trans_tasklet *trans;
808 
809 		trans = &per_cpu(xfrm_trans_tasklet, i);
810 		spin_lock_init(&trans->queue_lock);
811 		__skb_queue_head_init(&trans->queue);
812 		INIT_WORK(&trans->work, xfrm_trans_reinject);
813 	}
814 }
815