xref: /linux/net/xfrm/xfrm_input.c (revision ae22a94997b8a03dcb3c922857c203246711f9d4)
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 #include <net/hotdata.h>
25 
26 #include "xfrm_inout.h"
27 
28 struct xfrm_trans_tasklet {
29 	struct work_struct work;
30 	spinlock_t queue_lock;
31 	struct sk_buff_head queue;
32 };
33 
34 struct xfrm_trans_cb {
35 	union {
36 		struct inet_skb_parm	h4;
37 #if IS_ENABLED(CONFIG_IPV6)
38 		struct inet6_skb_parm	h6;
39 #endif
40 	} header;
41 	int (*finish)(struct net *net, struct sock *sk, struct sk_buff *skb);
42 	struct net *net;
43 };
44 
45 #define XFRM_TRANS_SKB_CB(__skb) ((struct xfrm_trans_cb *)&((__skb)->cb[0]))
46 
47 static DEFINE_SPINLOCK(xfrm_input_afinfo_lock);
48 static struct xfrm_input_afinfo const __rcu *xfrm_input_afinfo[2][AF_INET6 + 1];
49 
50 static struct gro_cells gro_cells;
51 static struct net_device xfrm_napi_dev;
52 
53 static DEFINE_PER_CPU(struct xfrm_trans_tasklet, xfrm_trans_tasklet);
54 
55 int xfrm_input_register_afinfo(const struct xfrm_input_afinfo *afinfo)
56 {
57 	int err = 0;
58 
59 	if (WARN_ON(afinfo->family > AF_INET6))
60 		return -EAFNOSUPPORT;
61 
62 	spin_lock_bh(&xfrm_input_afinfo_lock);
63 	if (unlikely(xfrm_input_afinfo[afinfo->is_ipip][afinfo->family]))
64 		err = -EEXIST;
65 	else
66 		rcu_assign_pointer(xfrm_input_afinfo[afinfo->is_ipip][afinfo->family], afinfo);
67 	spin_unlock_bh(&xfrm_input_afinfo_lock);
68 	return err;
69 }
70 EXPORT_SYMBOL(xfrm_input_register_afinfo);
71 
72 int xfrm_input_unregister_afinfo(const struct xfrm_input_afinfo *afinfo)
73 {
74 	int err = 0;
75 
76 	spin_lock_bh(&xfrm_input_afinfo_lock);
77 	if (likely(xfrm_input_afinfo[afinfo->is_ipip][afinfo->family])) {
78 		if (unlikely(xfrm_input_afinfo[afinfo->is_ipip][afinfo->family] != afinfo))
79 			err = -EINVAL;
80 		else
81 			RCU_INIT_POINTER(xfrm_input_afinfo[afinfo->is_ipip][afinfo->family], NULL);
82 	}
83 	spin_unlock_bh(&xfrm_input_afinfo_lock);
84 	synchronize_rcu();
85 	return err;
86 }
87 EXPORT_SYMBOL(xfrm_input_unregister_afinfo);
88 
89 static const struct xfrm_input_afinfo *xfrm_input_get_afinfo(u8 family, bool is_ipip)
90 {
91 	const struct xfrm_input_afinfo *afinfo;
92 
93 	if (WARN_ON_ONCE(family > AF_INET6))
94 		return NULL;
95 
96 	rcu_read_lock();
97 	afinfo = rcu_dereference(xfrm_input_afinfo[is_ipip][family]);
98 	if (unlikely(!afinfo))
99 		rcu_read_unlock();
100 	return afinfo;
101 }
102 
103 static int xfrm_rcv_cb(struct sk_buff *skb, unsigned int family, u8 protocol,
104 		       int err)
105 {
106 	bool is_ipip = (protocol == IPPROTO_IPIP || protocol == IPPROTO_IPV6);
107 	const struct xfrm_input_afinfo *afinfo;
108 	int ret;
109 
110 	afinfo = xfrm_input_get_afinfo(family, is_ipip);
111 	if (!afinfo)
112 		return -EAFNOSUPPORT;
113 
114 	ret = afinfo->callback(skb, protocol, err);
115 	rcu_read_unlock();
116 
117 	return ret;
118 }
119 
120 struct sec_path *secpath_set(struct sk_buff *skb)
121 {
122 	struct sec_path *sp, *tmp = skb_ext_find(skb, SKB_EXT_SEC_PATH);
123 
124 	sp = skb_ext_add(skb, SKB_EXT_SEC_PATH);
125 	if (!sp)
126 		return NULL;
127 
128 	if (tmp) /* reused existing one (was COW'd if needed) */
129 		return sp;
130 
131 	/* allocated new secpath */
132 	memset(sp->ovec, 0, sizeof(sp->ovec));
133 	sp->olen = 0;
134 	sp->len = 0;
135 	sp->verified_cnt = 0;
136 
137 	return sp;
138 }
139 EXPORT_SYMBOL(secpath_set);
140 
141 /* Fetch spi and seq from ipsec header */
142 
143 int xfrm_parse_spi(struct sk_buff *skb, u8 nexthdr, __be32 *spi, __be32 *seq)
144 {
145 	int offset, offset_seq;
146 	int hlen;
147 
148 	switch (nexthdr) {
149 	case IPPROTO_AH:
150 		hlen = sizeof(struct ip_auth_hdr);
151 		offset = offsetof(struct ip_auth_hdr, spi);
152 		offset_seq = offsetof(struct ip_auth_hdr, seq_no);
153 		break;
154 	case IPPROTO_ESP:
155 		hlen = sizeof(struct ip_esp_hdr);
156 		offset = offsetof(struct ip_esp_hdr, spi);
157 		offset_seq = offsetof(struct ip_esp_hdr, seq_no);
158 		break;
159 	case IPPROTO_COMP:
160 		if (!pskb_may_pull(skb, sizeof(struct ip_comp_hdr)))
161 			return -EINVAL;
162 		*spi = htonl(ntohs(*(__be16 *)(skb_transport_header(skb) + 2)));
163 		*seq = 0;
164 		return 0;
165 	default:
166 		return 1;
167 	}
168 
169 	if (!pskb_may_pull(skb, hlen))
170 		return -EINVAL;
171 
172 	*spi = *(__be32 *)(skb_transport_header(skb) + offset);
173 	*seq = *(__be32 *)(skb_transport_header(skb) + offset_seq);
174 	return 0;
175 }
176 EXPORT_SYMBOL(xfrm_parse_spi);
177 
178 static int xfrm4_remove_beet_encap(struct xfrm_state *x, struct sk_buff *skb)
179 {
180 	struct iphdr *iph;
181 	int optlen = 0;
182 	int err = -EINVAL;
183 
184 	skb->protocol = htons(ETH_P_IP);
185 
186 	if (unlikely(XFRM_MODE_SKB_CB(skb)->protocol == IPPROTO_BEETPH)) {
187 		struct ip_beet_phdr *ph;
188 		int phlen;
189 
190 		if (!pskb_may_pull(skb, sizeof(*ph)))
191 			goto out;
192 
193 		ph = (struct ip_beet_phdr *)skb->data;
194 
195 		phlen = sizeof(*ph) + ph->padlen;
196 		optlen = ph->hdrlen * 8 + (IPV4_BEET_PHMAXLEN - phlen);
197 		if (optlen < 0 || optlen & 3 || optlen > 250)
198 			goto out;
199 
200 		XFRM_MODE_SKB_CB(skb)->protocol = ph->nexthdr;
201 
202 		if (!pskb_may_pull(skb, phlen))
203 			goto out;
204 		__skb_pull(skb, phlen);
205 	}
206 
207 	skb_push(skb, sizeof(*iph));
208 	skb_reset_network_header(skb);
209 	skb_mac_header_rebuild(skb);
210 
211 	xfrm4_beet_make_header(skb);
212 
213 	iph = ip_hdr(skb);
214 
215 	iph->ihl += optlen / 4;
216 	iph->tot_len = htons(skb->len);
217 	iph->daddr = x->sel.daddr.a4;
218 	iph->saddr = x->sel.saddr.a4;
219 	iph->check = 0;
220 	iph->check = ip_fast_csum(skb_network_header(skb), iph->ihl);
221 	err = 0;
222 out:
223 	return err;
224 }
225 
226 static void ipip_ecn_decapsulate(struct sk_buff *skb)
227 {
228 	struct iphdr *inner_iph = ipip_hdr(skb);
229 
230 	if (INET_ECN_is_ce(XFRM_MODE_SKB_CB(skb)->tos))
231 		IP_ECN_set_ce(inner_iph);
232 }
233 
234 static int xfrm4_remove_tunnel_encap(struct xfrm_state *x, struct sk_buff *skb)
235 {
236 	int err = -EINVAL;
237 
238 	skb->protocol = htons(ETH_P_IP);
239 
240 	if (!pskb_may_pull(skb, sizeof(struct iphdr)))
241 		goto out;
242 
243 	err = skb_unclone(skb, GFP_ATOMIC);
244 	if (err)
245 		goto out;
246 
247 	if (x->props.flags & XFRM_STATE_DECAP_DSCP)
248 		ipv4_copy_dscp(XFRM_MODE_SKB_CB(skb)->tos, ipip_hdr(skb));
249 	if (!(x->props.flags & XFRM_STATE_NOECN))
250 		ipip_ecn_decapsulate(skb);
251 
252 	skb_reset_network_header(skb);
253 	skb_mac_header_rebuild(skb);
254 	if (skb->mac_len)
255 		eth_hdr(skb)->h_proto = skb->protocol;
256 
257 	err = 0;
258 
259 out:
260 	return err;
261 }
262 
263 static void ipip6_ecn_decapsulate(struct sk_buff *skb)
264 {
265 	struct ipv6hdr *inner_iph = ipipv6_hdr(skb);
266 
267 	if (INET_ECN_is_ce(XFRM_MODE_SKB_CB(skb)->tos))
268 		IP6_ECN_set_ce(skb, inner_iph);
269 }
270 
271 static int xfrm6_remove_tunnel_encap(struct xfrm_state *x, struct sk_buff *skb)
272 {
273 	int err = -EINVAL;
274 
275 	skb->protocol = htons(ETH_P_IPV6);
276 
277 	if (!pskb_may_pull(skb, sizeof(struct ipv6hdr)))
278 		goto out;
279 
280 	err = skb_unclone(skb, GFP_ATOMIC);
281 	if (err)
282 		goto out;
283 
284 	if (x->props.flags & XFRM_STATE_DECAP_DSCP)
285 		ipv6_copy_dscp(XFRM_MODE_SKB_CB(skb)->tos, ipipv6_hdr(skb));
286 	if (!(x->props.flags & XFRM_STATE_NOECN))
287 		ipip6_ecn_decapsulate(skb);
288 
289 	skb_reset_network_header(skb);
290 	skb_mac_header_rebuild(skb);
291 	if (skb->mac_len)
292 		eth_hdr(skb)->h_proto = skb->protocol;
293 
294 	err = 0;
295 
296 out:
297 	return err;
298 }
299 
300 static int xfrm6_remove_beet_encap(struct xfrm_state *x, struct sk_buff *skb)
301 {
302 	struct ipv6hdr *ip6h;
303 	int size = sizeof(struct ipv6hdr);
304 	int err;
305 
306 	skb->protocol = htons(ETH_P_IPV6);
307 
308 	err = skb_cow_head(skb, size + skb->mac_len);
309 	if (err)
310 		goto out;
311 
312 	__skb_push(skb, size);
313 	skb_reset_network_header(skb);
314 	skb_mac_header_rebuild(skb);
315 
316 	xfrm6_beet_make_header(skb);
317 
318 	ip6h = ipv6_hdr(skb);
319 	ip6h->payload_len = htons(skb->len - size);
320 	ip6h->daddr = x->sel.daddr.in6;
321 	ip6h->saddr = x->sel.saddr.in6;
322 	err = 0;
323 out:
324 	return err;
325 }
326 
327 /* Remove encapsulation header.
328  *
329  * The IP header will be moved over the top of the encapsulation
330  * header.
331  *
332  * On entry, the transport header shall point to where the IP header
333  * should be and the network header shall be set to where the IP
334  * header currently is.  skb->data shall point to the start of the
335  * payload.
336  */
337 static int
338 xfrm_inner_mode_encap_remove(struct xfrm_state *x,
339 			     struct sk_buff *skb)
340 {
341 	switch (x->props.mode) {
342 	case XFRM_MODE_BEET:
343 		switch (x->sel.family) {
344 		case AF_INET:
345 			return xfrm4_remove_beet_encap(x, skb);
346 		case AF_INET6:
347 			return xfrm6_remove_beet_encap(x, skb);
348 		}
349 		break;
350 	case XFRM_MODE_TUNNEL:
351 		switch (XFRM_MODE_SKB_CB(skb)->protocol) {
352 		case IPPROTO_IPIP:
353 			return xfrm4_remove_tunnel_encap(x, skb);
354 		case IPPROTO_IPV6:
355 			return xfrm6_remove_tunnel_encap(x, skb);
356 		break;
357 		}
358 		return -EINVAL;
359 	}
360 
361 	WARN_ON_ONCE(1);
362 	return -EOPNOTSUPP;
363 }
364 
365 static int xfrm_prepare_input(struct xfrm_state *x, struct sk_buff *skb)
366 {
367 	switch (x->props.family) {
368 	case AF_INET:
369 		xfrm4_extract_header(skb);
370 		break;
371 	case AF_INET6:
372 		xfrm6_extract_header(skb);
373 		break;
374 	default:
375 		WARN_ON_ONCE(1);
376 		return -EAFNOSUPPORT;
377 	}
378 
379 	return xfrm_inner_mode_encap_remove(x, skb);
380 }
381 
382 /* Remove encapsulation header.
383  *
384  * The IP header will be moved over the top of the encapsulation header.
385  *
386  * On entry, skb_transport_header() shall point to where the IP header
387  * should be and skb_network_header() shall be set to where the IP header
388  * currently is.  skb->data shall point to the start of the payload.
389  */
390 static int xfrm4_transport_input(struct xfrm_state *x, struct sk_buff *skb)
391 {
392 	int ihl = skb->data - skb_transport_header(skb);
393 
394 	if (skb->transport_header != skb->network_header) {
395 		memmove(skb_transport_header(skb),
396 			skb_network_header(skb), ihl);
397 		skb->network_header = skb->transport_header;
398 	}
399 	ip_hdr(skb)->tot_len = htons(skb->len + ihl);
400 	skb_reset_transport_header(skb);
401 	return 0;
402 }
403 
404 static int xfrm6_transport_input(struct xfrm_state *x, struct sk_buff *skb)
405 {
406 #if IS_ENABLED(CONFIG_IPV6)
407 	int ihl = skb->data - skb_transport_header(skb);
408 
409 	if (skb->transport_header != skb->network_header) {
410 		memmove(skb_transport_header(skb),
411 			skb_network_header(skb), ihl);
412 		skb->network_header = skb->transport_header;
413 	}
414 	ipv6_hdr(skb)->payload_len = htons(skb->len + ihl -
415 					   sizeof(struct ipv6hdr));
416 	skb_reset_transport_header(skb);
417 	return 0;
418 #else
419 	WARN_ON_ONCE(1);
420 	return -EAFNOSUPPORT;
421 #endif
422 }
423 
424 static int xfrm_inner_mode_input(struct xfrm_state *x,
425 				 struct sk_buff *skb)
426 {
427 	switch (x->props.mode) {
428 	case XFRM_MODE_BEET:
429 	case XFRM_MODE_TUNNEL:
430 		return xfrm_prepare_input(x, skb);
431 	case XFRM_MODE_TRANSPORT:
432 		if (x->props.family == AF_INET)
433 			return xfrm4_transport_input(x, skb);
434 		if (x->props.family == AF_INET6)
435 			return xfrm6_transport_input(x, skb);
436 		break;
437 	case XFRM_MODE_ROUTEOPTIMIZATION:
438 		WARN_ON_ONCE(1);
439 		break;
440 	default:
441 		WARN_ON_ONCE(1);
442 		break;
443 	}
444 
445 	return -EOPNOTSUPP;
446 }
447 
448 int xfrm_input(struct sk_buff *skb, int nexthdr, __be32 spi, int encap_type)
449 {
450 	const struct xfrm_state_afinfo *afinfo;
451 	struct net *net = dev_net(skb->dev);
452 	int err;
453 	__be32 seq;
454 	__be32 seq_hi;
455 	struct xfrm_state *x = NULL;
456 	xfrm_address_t *daddr;
457 	u32 mark = skb->mark;
458 	unsigned int family = AF_UNSPEC;
459 	int decaps = 0;
460 	int async = 0;
461 	bool xfrm_gro = false;
462 	bool crypto_done = false;
463 	struct xfrm_offload *xo = xfrm_offload(skb);
464 	struct sec_path *sp;
465 
466 	if (encap_type < 0 || (xo && xo->flags & XFRM_GRO)) {
467 		x = xfrm_input_state(skb);
468 
469 		if (unlikely(x->km.state != XFRM_STATE_VALID)) {
470 			if (x->km.state == XFRM_STATE_ACQ)
471 				XFRM_INC_STATS(net, LINUX_MIB_XFRMACQUIREERROR);
472 			else
473 				XFRM_INC_STATS(net,
474 					       LINUX_MIB_XFRMINSTATEINVALID);
475 
476 			if (encap_type == -1)
477 				dev_put(skb->dev);
478 			goto drop;
479 		}
480 
481 		family = x->props.family;
482 
483 		/* An encap_type of -1 indicates async resumption. */
484 		if (encap_type == -1) {
485 			async = 1;
486 			seq = XFRM_SKB_CB(skb)->seq.input.low;
487 			goto resume;
488 		}
489 		/* GRO call */
490 		seq = XFRM_SPI_SKB_CB(skb)->seq;
491 
492 		if (xo && (xo->flags & CRYPTO_DONE)) {
493 			crypto_done = true;
494 			family = XFRM_SPI_SKB_CB(skb)->family;
495 
496 			if (!(xo->status & CRYPTO_SUCCESS)) {
497 				if (xo->status &
498 				    (CRYPTO_TRANSPORT_AH_AUTH_FAILED |
499 				     CRYPTO_TRANSPORT_ESP_AUTH_FAILED |
500 				     CRYPTO_TUNNEL_AH_AUTH_FAILED |
501 				     CRYPTO_TUNNEL_ESP_AUTH_FAILED)) {
502 
503 					xfrm_audit_state_icvfail(x, skb,
504 								 x->type->proto);
505 					x->stats.integrity_failed++;
506 					XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEPROTOERROR);
507 					goto drop;
508 				}
509 
510 				if (xo->status & CRYPTO_INVALID_PROTOCOL) {
511 					XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEPROTOERROR);
512 					goto drop;
513 				}
514 
515 				XFRM_INC_STATS(net, LINUX_MIB_XFRMINBUFFERERROR);
516 				goto drop;
517 			}
518 
519 			if (xfrm_parse_spi(skb, nexthdr, &spi, &seq)) {
520 				XFRM_INC_STATS(net, LINUX_MIB_XFRMINHDRERROR);
521 				goto drop;
522 			}
523 		}
524 
525 		goto lock;
526 	}
527 
528 	family = XFRM_SPI_SKB_CB(skb)->family;
529 
530 	/* if tunnel is present override skb->mark value with tunnel i_key */
531 	switch (family) {
532 	case AF_INET:
533 		if (XFRM_TUNNEL_SKB_CB(skb)->tunnel.ip4)
534 			mark = be32_to_cpu(XFRM_TUNNEL_SKB_CB(skb)->tunnel.ip4->parms.i_key);
535 		break;
536 	case AF_INET6:
537 		if (XFRM_TUNNEL_SKB_CB(skb)->tunnel.ip6)
538 			mark = be32_to_cpu(XFRM_TUNNEL_SKB_CB(skb)->tunnel.ip6->parms.i_key);
539 		break;
540 	}
541 
542 	sp = secpath_set(skb);
543 	if (!sp) {
544 		XFRM_INC_STATS(net, LINUX_MIB_XFRMINERROR);
545 		goto drop;
546 	}
547 
548 	seq = 0;
549 	if (!spi && xfrm_parse_spi(skb, nexthdr, &spi, &seq)) {
550 		secpath_reset(skb);
551 		XFRM_INC_STATS(net, LINUX_MIB_XFRMINHDRERROR);
552 		goto drop;
553 	}
554 
555 	daddr = (xfrm_address_t *)(skb_network_header(skb) +
556 				   XFRM_SPI_SKB_CB(skb)->daddroff);
557 	do {
558 		sp = skb_sec_path(skb);
559 
560 		if (sp->len == XFRM_MAX_DEPTH) {
561 			secpath_reset(skb);
562 			XFRM_INC_STATS(net, LINUX_MIB_XFRMINBUFFERERROR);
563 			goto drop;
564 		}
565 
566 		x = xfrm_state_lookup(net, mark, daddr, spi, nexthdr, family);
567 		if (x == NULL) {
568 			secpath_reset(skb);
569 			XFRM_INC_STATS(net, LINUX_MIB_XFRMINNOSTATES);
570 			xfrm_audit_state_notfound(skb, family, spi, seq);
571 			goto drop;
572 		}
573 
574 		skb->mark = xfrm_smark_get(skb->mark, x);
575 
576 		sp->xvec[sp->len++] = x;
577 
578 		skb_dst_force(skb);
579 		if (!skb_dst(skb)) {
580 			XFRM_INC_STATS(net, LINUX_MIB_XFRMINERROR);
581 			goto drop;
582 		}
583 
584 lock:
585 		spin_lock(&x->lock);
586 
587 		if (unlikely(x->km.state != XFRM_STATE_VALID)) {
588 			if (x->km.state == XFRM_STATE_ACQ)
589 				XFRM_INC_STATS(net, LINUX_MIB_XFRMACQUIREERROR);
590 			else
591 				XFRM_INC_STATS(net,
592 					       LINUX_MIB_XFRMINSTATEINVALID);
593 			goto drop_unlock;
594 		}
595 
596 		if ((x->encap ? x->encap->encap_type : 0) != encap_type) {
597 			XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMISMATCH);
598 			goto drop_unlock;
599 		}
600 
601 		if (xfrm_replay_check(x, skb, seq)) {
602 			XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATESEQERROR);
603 			goto drop_unlock;
604 		}
605 
606 		if (xfrm_state_check_expire(x)) {
607 			XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEEXPIRED);
608 			goto drop_unlock;
609 		}
610 
611 		spin_unlock(&x->lock);
612 
613 		if (xfrm_tunnel_check(skb, x, family)) {
614 			XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMODEERROR);
615 			goto drop;
616 		}
617 
618 		seq_hi = htonl(xfrm_replay_seqhi(x, seq));
619 
620 		XFRM_SKB_CB(skb)->seq.input.low = seq;
621 		XFRM_SKB_CB(skb)->seq.input.hi = seq_hi;
622 
623 		dev_hold(skb->dev);
624 
625 		if (crypto_done)
626 			nexthdr = x->type_offload->input_tail(x, skb);
627 		else
628 			nexthdr = x->type->input(x, skb);
629 
630 		if (nexthdr == -EINPROGRESS)
631 			return 0;
632 resume:
633 		dev_put(skb->dev);
634 
635 		spin_lock(&x->lock);
636 		if (nexthdr < 0) {
637 			if (nexthdr == -EBADMSG) {
638 				xfrm_audit_state_icvfail(x, skb,
639 							 x->type->proto);
640 				x->stats.integrity_failed++;
641 			}
642 			XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEPROTOERROR);
643 			goto drop_unlock;
644 		}
645 
646 		/* only the first xfrm gets the encap type */
647 		encap_type = 0;
648 
649 		if (xfrm_replay_recheck(x, skb, seq)) {
650 			XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATESEQERROR);
651 			goto drop_unlock;
652 		}
653 
654 		xfrm_replay_advance(x, seq);
655 
656 		x->curlft.bytes += skb->len;
657 		x->curlft.packets++;
658 		x->lastused = ktime_get_real_seconds();
659 
660 		spin_unlock(&x->lock);
661 
662 		XFRM_MODE_SKB_CB(skb)->protocol = nexthdr;
663 
664 		if (xfrm_inner_mode_input(x, skb)) {
665 			XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMODEERROR);
666 			goto drop;
667 		}
668 
669 		if (x->outer_mode.flags & XFRM_MODE_FLAG_TUNNEL) {
670 			decaps = 1;
671 			break;
672 		}
673 
674 		/*
675 		 * We need the inner address.  However, we only get here for
676 		 * transport mode so the outer address is identical.
677 		 */
678 		daddr = &x->id.daddr;
679 		family = x->props.family;
680 
681 		err = xfrm_parse_spi(skb, nexthdr, &spi, &seq);
682 		if (err < 0) {
683 			XFRM_INC_STATS(net, LINUX_MIB_XFRMINHDRERROR);
684 			goto drop;
685 		}
686 		crypto_done = false;
687 	} while (!err);
688 
689 	err = xfrm_rcv_cb(skb, family, x->type->proto, 0);
690 	if (err)
691 		goto drop;
692 
693 	nf_reset_ct(skb);
694 
695 	if (decaps) {
696 		sp = skb_sec_path(skb);
697 		if (sp)
698 			sp->olen = 0;
699 		if (skb_valid_dst(skb))
700 			skb_dst_drop(skb);
701 		gro_cells_receive(&gro_cells, skb);
702 		return 0;
703 	} else {
704 		xo = xfrm_offload(skb);
705 		if (xo)
706 			xfrm_gro = xo->flags & XFRM_GRO;
707 
708 		err = -EAFNOSUPPORT;
709 		rcu_read_lock();
710 		afinfo = xfrm_state_afinfo_get_rcu(x->props.family);
711 		if (likely(afinfo))
712 			err = afinfo->transport_finish(skb, xfrm_gro || async);
713 		rcu_read_unlock();
714 		if (xfrm_gro) {
715 			sp = skb_sec_path(skb);
716 			if (sp)
717 				sp->olen = 0;
718 			if (skb_valid_dst(skb))
719 				skb_dst_drop(skb);
720 			gro_cells_receive(&gro_cells, skb);
721 			return err;
722 		}
723 
724 		return err;
725 	}
726 
727 drop_unlock:
728 	spin_unlock(&x->lock);
729 drop:
730 	xfrm_rcv_cb(skb, family, x && x->type ? x->type->proto : nexthdr, -1);
731 	kfree_skb(skb);
732 	return 0;
733 }
734 EXPORT_SYMBOL(xfrm_input);
735 
736 int xfrm_input_resume(struct sk_buff *skb, int nexthdr)
737 {
738 	return xfrm_input(skb, nexthdr, 0, -1);
739 }
740 EXPORT_SYMBOL(xfrm_input_resume);
741 
742 static void xfrm_trans_reinject(struct work_struct *work)
743 {
744 	struct xfrm_trans_tasklet *trans = container_of(work, struct xfrm_trans_tasklet, work);
745 	struct sk_buff_head queue;
746 	struct sk_buff *skb;
747 
748 	__skb_queue_head_init(&queue);
749 	spin_lock_bh(&trans->queue_lock);
750 	skb_queue_splice_init(&trans->queue, &queue);
751 	spin_unlock_bh(&trans->queue_lock);
752 
753 	local_bh_disable();
754 	while ((skb = __skb_dequeue(&queue)))
755 		XFRM_TRANS_SKB_CB(skb)->finish(XFRM_TRANS_SKB_CB(skb)->net,
756 					       NULL, skb);
757 	local_bh_enable();
758 }
759 
760 int xfrm_trans_queue_net(struct net *net, struct sk_buff *skb,
761 			 int (*finish)(struct net *, struct sock *,
762 				       struct sk_buff *))
763 {
764 	struct xfrm_trans_tasklet *trans;
765 
766 	trans = this_cpu_ptr(&xfrm_trans_tasklet);
767 
768 	if (skb_queue_len(&trans->queue) >= READ_ONCE(net_hotdata.max_backlog))
769 		return -ENOBUFS;
770 
771 	BUILD_BUG_ON(sizeof(struct xfrm_trans_cb) > sizeof(skb->cb));
772 
773 	XFRM_TRANS_SKB_CB(skb)->finish = finish;
774 	XFRM_TRANS_SKB_CB(skb)->net = net;
775 	spin_lock_bh(&trans->queue_lock);
776 	__skb_queue_tail(&trans->queue, skb);
777 	spin_unlock_bh(&trans->queue_lock);
778 	schedule_work(&trans->work);
779 	return 0;
780 }
781 EXPORT_SYMBOL(xfrm_trans_queue_net);
782 
783 int xfrm_trans_queue(struct sk_buff *skb,
784 		     int (*finish)(struct net *, struct sock *,
785 				   struct sk_buff *))
786 {
787 	return xfrm_trans_queue_net(dev_net(skb->dev), skb, finish);
788 }
789 EXPORT_SYMBOL(xfrm_trans_queue);
790 
791 void __init xfrm_input_init(void)
792 {
793 	int err;
794 	int i;
795 
796 	init_dummy_netdev(&xfrm_napi_dev);
797 	err = gro_cells_init(&gro_cells, &xfrm_napi_dev);
798 	if (err)
799 		gro_cells.cells = NULL;
800 
801 	for_each_possible_cpu(i) {
802 		struct xfrm_trans_tasklet *trans;
803 
804 		trans = &per_cpu(xfrm_trans_tasklet, i);
805 		spin_lock_init(&trans->queue_lock);
806 		__skb_queue_head_init(&trans->queue);
807 		INIT_WORK(&trans->work, xfrm_trans_reinject);
808 	}
809 }
810