xref: /openbmc/linux/net/xfrm/xfrm_input.c (revision 4c004ed9)
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 	sp->verified_cnt = 0;
135 
136 	return sp;
137 }
138 EXPORT_SYMBOL(secpath_set);
139 
140 /* Fetch spi and seq from ipsec header */
141 
142 int xfrm_parse_spi(struct sk_buff *skb, u8 nexthdr, __be32 *spi, __be32 *seq)
143 {
144 	int offset, offset_seq;
145 	int hlen;
146 
147 	switch (nexthdr) {
148 	case IPPROTO_AH:
149 		hlen = sizeof(struct ip_auth_hdr);
150 		offset = offsetof(struct ip_auth_hdr, spi);
151 		offset_seq = offsetof(struct ip_auth_hdr, seq_no);
152 		break;
153 	case IPPROTO_ESP:
154 		hlen = sizeof(struct ip_esp_hdr);
155 		offset = offsetof(struct ip_esp_hdr, spi);
156 		offset_seq = offsetof(struct ip_esp_hdr, seq_no);
157 		break;
158 	case IPPROTO_COMP:
159 		if (!pskb_may_pull(skb, sizeof(struct ip_comp_hdr)))
160 			return -EINVAL;
161 		*spi = htonl(ntohs(*(__be16 *)(skb_transport_header(skb) + 2)));
162 		*seq = 0;
163 		return 0;
164 	default:
165 		return 1;
166 	}
167 
168 	if (!pskb_may_pull(skb, hlen))
169 		return -EINVAL;
170 
171 	*spi = *(__be32 *)(skb_transport_header(skb) + offset);
172 	*seq = *(__be32 *)(skb_transport_header(skb) + offset_seq);
173 	return 0;
174 }
175 EXPORT_SYMBOL(xfrm_parse_spi);
176 
177 static int xfrm4_remove_beet_encap(struct xfrm_state *x, struct sk_buff *skb)
178 {
179 	struct iphdr *iph;
180 	int optlen = 0;
181 	int err = -EINVAL;
182 
183 	skb->protocol = htons(ETH_P_IP);
184 
185 	if (unlikely(XFRM_MODE_SKB_CB(skb)->protocol == IPPROTO_BEETPH)) {
186 		struct ip_beet_phdr *ph;
187 		int phlen;
188 
189 		if (!pskb_may_pull(skb, sizeof(*ph)))
190 			goto out;
191 
192 		ph = (struct ip_beet_phdr *)skb->data;
193 
194 		phlen = sizeof(*ph) + ph->padlen;
195 		optlen = ph->hdrlen * 8 + (IPV4_BEET_PHMAXLEN - phlen);
196 		if (optlen < 0 || optlen & 3 || optlen > 250)
197 			goto out;
198 
199 		XFRM_MODE_SKB_CB(skb)->protocol = ph->nexthdr;
200 
201 		if (!pskb_may_pull(skb, phlen))
202 			goto out;
203 		__skb_pull(skb, phlen);
204 	}
205 
206 	skb_push(skb, sizeof(*iph));
207 	skb_reset_network_header(skb);
208 	skb_mac_header_rebuild(skb);
209 
210 	xfrm4_beet_make_header(skb);
211 
212 	iph = ip_hdr(skb);
213 
214 	iph->ihl += optlen / 4;
215 	iph->tot_len = htons(skb->len);
216 	iph->daddr = x->sel.daddr.a4;
217 	iph->saddr = x->sel.saddr.a4;
218 	iph->check = 0;
219 	iph->check = ip_fast_csum(skb_network_header(skb), iph->ihl);
220 	err = 0;
221 out:
222 	return err;
223 }
224 
225 static void ipip_ecn_decapsulate(struct sk_buff *skb)
226 {
227 	struct iphdr *inner_iph = ipip_hdr(skb);
228 
229 	if (INET_ECN_is_ce(XFRM_MODE_SKB_CB(skb)->tos))
230 		IP_ECN_set_ce(inner_iph);
231 }
232 
233 static int xfrm4_remove_tunnel_encap(struct xfrm_state *x, struct sk_buff *skb)
234 {
235 	int err = -EINVAL;
236 
237 	skb->protocol = htons(ETH_P_IP);
238 
239 	if (!pskb_may_pull(skb, sizeof(struct iphdr)))
240 		goto out;
241 
242 	err = skb_unclone(skb, GFP_ATOMIC);
243 	if (err)
244 		goto out;
245 
246 	if (x->props.flags & XFRM_STATE_DECAP_DSCP)
247 		ipv4_copy_dscp(XFRM_MODE_SKB_CB(skb)->tos, ipip_hdr(skb));
248 	if (!(x->props.flags & XFRM_STATE_NOECN))
249 		ipip_ecn_decapsulate(skb);
250 
251 	skb_reset_network_header(skb);
252 	skb_mac_header_rebuild(skb);
253 	if (skb->mac_len)
254 		eth_hdr(skb)->h_proto = skb->protocol;
255 
256 	err = 0;
257 
258 out:
259 	return err;
260 }
261 
262 static void ipip6_ecn_decapsulate(struct sk_buff *skb)
263 {
264 	struct ipv6hdr *inner_iph = ipipv6_hdr(skb);
265 
266 	if (INET_ECN_is_ce(XFRM_MODE_SKB_CB(skb)->tos))
267 		IP6_ECN_set_ce(skb, inner_iph);
268 }
269 
270 static int xfrm6_remove_tunnel_encap(struct xfrm_state *x, struct sk_buff *skb)
271 {
272 	int err = -EINVAL;
273 
274 	skb->protocol = htons(ETH_P_IPV6);
275 
276 	if (!pskb_may_pull(skb, sizeof(struct ipv6hdr)))
277 		goto out;
278 
279 	err = skb_unclone(skb, GFP_ATOMIC);
280 	if (err)
281 		goto out;
282 
283 	if (x->props.flags & XFRM_STATE_DECAP_DSCP)
284 		ipv6_copy_dscp(XFRM_MODE_SKB_CB(skb)->tos, ipipv6_hdr(skb));
285 	if (!(x->props.flags & XFRM_STATE_NOECN))
286 		ipip6_ecn_decapsulate(skb);
287 
288 	skb_reset_network_header(skb);
289 	skb_mac_header_rebuild(skb);
290 	if (skb->mac_len)
291 		eth_hdr(skb)->h_proto = skb->protocol;
292 
293 	err = 0;
294 
295 out:
296 	return err;
297 }
298 
299 static int xfrm6_remove_beet_encap(struct xfrm_state *x, struct sk_buff *skb)
300 {
301 	struct ipv6hdr *ip6h;
302 	int size = sizeof(struct ipv6hdr);
303 	int err;
304 
305 	skb->protocol = htons(ETH_P_IPV6);
306 
307 	err = skb_cow_head(skb, size + skb->mac_len);
308 	if (err)
309 		goto out;
310 
311 	__skb_push(skb, size);
312 	skb_reset_network_header(skb);
313 	skb_mac_header_rebuild(skb);
314 
315 	xfrm6_beet_make_header(skb);
316 
317 	ip6h = ipv6_hdr(skb);
318 	ip6h->payload_len = htons(skb->len - size);
319 	ip6h->daddr = x->sel.daddr.in6;
320 	ip6h->saddr = x->sel.saddr.in6;
321 	err = 0;
322 out:
323 	return err;
324 }
325 
326 /* Remove encapsulation header.
327  *
328  * The IP header will be moved over the top of the encapsulation
329  * header.
330  *
331  * On entry, the transport header shall point to where the IP header
332  * should be and the network header shall be set to where the IP
333  * header currently is.  skb->data shall point to the start of the
334  * payload.
335  */
336 static int
337 xfrm_inner_mode_encap_remove(struct xfrm_state *x,
338 			     struct sk_buff *skb)
339 {
340 	switch (x->props.mode) {
341 	case XFRM_MODE_BEET:
342 		switch (x->sel.family) {
343 		case AF_INET:
344 			return xfrm4_remove_beet_encap(x, skb);
345 		case AF_INET6:
346 			return xfrm6_remove_beet_encap(x, skb);
347 		}
348 		break;
349 	case XFRM_MODE_TUNNEL:
350 		switch (XFRM_MODE_SKB_CB(skb)->protocol) {
351 		case IPPROTO_IPIP:
352 			return xfrm4_remove_tunnel_encap(x, skb);
353 		case IPPROTO_IPV6:
354 			return xfrm6_remove_tunnel_encap(x, skb);
355 		break;
356 		}
357 		return -EINVAL;
358 	}
359 
360 	WARN_ON_ONCE(1);
361 	return -EOPNOTSUPP;
362 }
363 
364 static int xfrm_prepare_input(struct xfrm_state *x, struct sk_buff *skb)
365 {
366 	switch (x->props.family) {
367 	case AF_INET:
368 		xfrm4_extract_header(skb);
369 		break;
370 	case AF_INET6:
371 		xfrm6_extract_header(skb);
372 		break;
373 	default:
374 		WARN_ON_ONCE(1);
375 		return -EAFNOSUPPORT;
376 	}
377 
378 	return xfrm_inner_mode_encap_remove(x, skb);
379 }
380 
381 /* Remove encapsulation header.
382  *
383  * The IP header will be moved over the top of the encapsulation header.
384  *
385  * On entry, skb_transport_header() shall point to where the IP header
386  * should be and skb_network_header() shall be set to where the IP header
387  * currently is.  skb->data shall point to the start of the payload.
388  */
389 static int xfrm4_transport_input(struct xfrm_state *x, struct sk_buff *skb)
390 {
391 	int ihl = skb->data - skb_transport_header(skb);
392 
393 	if (skb->transport_header != skb->network_header) {
394 		memmove(skb_transport_header(skb),
395 			skb_network_header(skb), ihl);
396 		skb->network_header = skb->transport_header;
397 	}
398 	ip_hdr(skb)->tot_len = htons(skb->len + ihl);
399 	skb_reset_transport_header(skb);
400 	return 0;
401 }
402 
403 static int xfrm6_transport_input(struct xfrm_state *x, struct sk_buff *skb)
404 {
405 #if IS_ENABLED(CONFIG_IPV6)
406 	int ihl = skb->data - skb_transport_header(skb);
407 
408 	if (skb->transport_header != skb->network_header) {
409 		memmove(skb_transport_header(skb),
410 			skb_network_header(skb), ihl);
411 		skb->network_header = skb->transport_header;
412 	}
413 	ipv6_hdr(skb)->payload_len = htons(skb->len + ihl -
414 					   sizeof(struct ipv6hdr));
415 	skb_reset_transport_header(skb);
416 	return 0;
417 #else
418 	WARN_ON_ONCE(1);
419 	return -EAFNOSUPPORT;
420 #endif
421 }
422 
423 static int xfrm_inner_mode_input(struct xfrm_state *x,
424 				 struct sk_buff *skb)
425 {
426 	switch (x->props.mode) {
427 	case XFRM_MODE_BEET:
428 	case XFRM_MODE_TUNNEL:
429 		return xfrm_prepare_input(x, skb);
430 	case XFRM_MODE_TRANSPORT:
431 		if (x->props.family == AF_INET)
432 			return xfrm4_transport_input(x, skb);
433 		if (x->props.family == AF_INET6)
434 			return xfrm6_transport_input(x, skb);
435 		break;
436 	case XFRM_MODE_ROUTEOPTIMIZATION:
437 		WARN_ON_ONCE(1);
438 		break;
439 	default:
440 		WARN_ON_ONCE(1);
441 		break;
442 	}
443 
444 	return -EOPNOTSUPP;
445 }
446 
447 int xfrm_input(struct sk_buff *skb, int nexthdr, __be32 spi, int encap_type)
448 {
449 	const struct xfrm_state_afinfo *afinfo;
450 	struct net *net = dev_net(skb->dev);
451 	int err;
452 	__be32 seq;
453 	__be32 seq_hi;
454 	struct xfrm_state *x = NULL;
455 	xfrm_address_t *daddr;
456 	u32 mark = skb->mark;
457 	unsigned int family = AF_UNSPEC;
458 	int decaps = 0;
459 	int async = 0;
460 	bool xfrm_gro = false;
461 	bool crypto_done = false;
462 	struct xfrm_offload *xo = xfrm_offload(skb);
463 	struct sec_path *sp;
464 
465 	if (encap_type < 0) {
466 		x = xfrm_input_state(skb);
467 
468 		if (unlikely(x->km.state != XFRM_STATE_VALID)) {
469 			if (x->km.state == XFRM_STATE_ACQ)
470 				XFRM_INC_STATS(net, LINUX_MIB_XFRMACQUIREERROR);
471 			else
472 				XFRM_INC_STATS(net,
473 					       LINUX_MIB_XFRMINSTATEINVALID);
474 
475 			if (encap_type == -1)
476 				dev_put(skb->dev);
477 			goto drop;
478 		}
479 
480 		family = x->props.family;
481 
482 		/* An encap_type of -1 indicates async resumption. */
483 		if (encap_type == -1) {
484 			async = 1;
485 			seq = XFRM_SKB_CB(skb)->seq.input.low;
486 			goto resume;
487 		}
488 
489 		/* encap_type < -1 indicates a GRO call. */
490 		encap_type = 0;
491 		seq = XFRM_SPI_SKB_CB(skb)->seq;
492 
493 		if (xo && (xo->flags & CRYPTO_DONE)) {
494 			crypto_done = true;
495 			family = XFRM_SPI_SKB_CB(skb)->family;
496 
497 			if (!(xo->status & CRYPTO_SUCCESS)) {
498 				if (xo->status &
499 				    (CRYPTO_TRANSPORT_AH_AUTH_FAILED |
500 				     CRYPTO_TRANSPORT_ESP_AUTH_FAILED |
501 				     CRYPTO_TUNNEL_AH_AUTH_FAILED |
502 				     CRYPTO_TUNNEL_ESP_AUTH_FAILED)) {
503 
504 					xfrm_audit_state_icvfail(x, skb,
505 								 x->type->proto);
506 					x->stats.integrity_failed++;
507 					XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEPROTOERROR);
508 					goto drop;
509 				}
510 
511 				if (xo->status & CRYPTO_INVALID_PROTOCOL) {
512 					XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEPROTOERROR);
513 					goto drop;
514 				}
515 
516 				XFRM_INC_STATS(net, LINUX_MIB_XFRMINBUFFERERROR);
517 				goto drop;
518 			}
519 
520 			if (xfrm_parse_spi(skb, nexthdr, &spi, &seq)) {
521 				XFRM_INC_STATS(net, LINUX_MIB_XFRMINHDRERROR);
522 				goto drop;
523 			}
524 		}
525 
526 		goto lock;
527 	}
528 
529 	family = XFRM_SPI_SKB_CB(skb)->family;
530 
531 	/* if tunnel is present override skb->mark value with tunnel i_key */
532 	switch (family) {
533 	case AF_INET:
534 		if (XFRM_TUNNEL_SKB_CB(skb)->tunnel.ip4)
535 			mark = be32_to_cpu(XFRM_TUNNEL_SKB_CB(skb)->tunnel.ip4->parms.i_key);
536 		break;
537 	case AF_INET6:
538 		if (XFRM_TUNNEL_SKB_CB(skb)->tunnel.ip6)
539 			mark = be32_to_cpu(XFRM_TUNNEL_SKB_CB(skb)->tunnel.ip6->parms.i_key);
540 		break;
541 	}
542 
543 	sp = secpath_set(skb);
544 	if (!sp) {
545 		XFRM_INC_STATS(net, LINUX_MIB_XFRMINERROR);
546 		goto drop;
547 	}
548 
549 	seq = 0;
550 	if (!spi && xfrm_parse_spi(skb, nexthdr, &spi, &seq)) {
551 		secpath_reset(skb);
552 		XFRM_INC_STATS(net, LINUX_MIB_XFRMINHDRERROR);
553 		goto drop;
554 	}
555 
556 	daddr = (xfrm_address_t *)(skb_network_header(skb) +
557 				   XFRM_SPI_SKB_CB(skb)->daddroff);
558 	do {
559 		sp = skb_sec_path(skb);
560 
561 		if (sp->len == XFRM_MAX_DEPTH) {
562 			secpath_reset(skb);
563 			XFRM_INC_STATS(net, LINUX_MIB_XFRMINBUFFERERROR);
564 			goto drop;
565 		}
566 
567 		x = xfrm_state_lookup(net, mark, daddr, spi, nexthdr, family);
568 		if (x == NULL) {
569 			secpath_reset(skb);
570 			XFRM_INC_STATS(net, LINUX_MIB_XFRMINNOSTATES);
571 			xfrm_audit_state_notfound(skb, family, spi, seq);
572 			goto drop;
573 		}
574 
575 		skb->mark = xfrm_smark_get(skb->mark, x);
576 
577 		sp->xvec[sp->len++] = x;
578 
579 		skb_dst_force(skb);
580 		if (!skb_dst(skb)) {
581 			XFRM_INC_STATS(net, LINUX_MIB_XFRMINERROR);
582 			goto drop;
583 		}
584 
585 lock:
586 		spin_lock(&x->lock);
587 
588 		if (unlikely(x->km.state != XFRM_STATE_VALID)) {
589 			if (x->km.state == XFRM_STATE_ACQ)
590 				XFRM_INC_STATS(net, LINUX_MIB_XFRMACQUIREERROR);
591 			else
592 				XFRM_INC_STATS(net,
593 					       LINUX_MIB_XFRMINSTATEINVALID);
594 			goto drop_unlock;
595 		}
596 
597 		if ((x->encap ? x->encap->encap_type : 0) != encap_type) {
598 			XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMISMATCH);
599 			goto drop_unlock;
600 		}
601 
602 		if (xfrm_replay_check(x, skb, seq)) {
603 			XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATESEQERROR);
604 			goto drop_unlock;
605 		}
606 
607 		if (xfrm_state_check_expire(x)) {
608 			XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEEXPIRED);
609 			goto drop_unlock;
610 		}
611 
612 		spin_unlock(&x->lock);
613 
614 		if (xfrm_tunnel_check(skb, x, family)) {
615 			XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMODEERROR);
616 			goto drop;
617 		}
618 
619 		seq_hi = htonl(xfrm_replay_seqhi(x, seq));
620 
621 		XFRM_SKB_CB(skb)->seq.input.low = seq;
622 		XFRM_SKB_CB(skb)->seq.input.hi = seq_hi;
623 
624 		dev_hold(skb->dev);
625 
626 		if (crypto_done)
627 			nexthdr = x->type_offload->input_tail(x, skb);
628 		else
629 			nexthdr = x->type->input(x, skb);
630 
631 		if (nexthdr == -EINPROGRESS)
632 			return 0;
633 resume:
634 		dev_put(skb->dev);
635 
636 		spin_lock(&x->lock);
637 		if (nexthdr < 0) {
638 			if (nexthdr == -EBADMSG) {
639 				xfrm_audit_state_icvfail(x, skb,
640 							 x->type->proto);
641 				x->stats.integrity_failed++;
642 			}
643 			XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEPROTOERROR);
644 			goto drop_unlock;
645 		}
646 
647 		/* only the first xfrm gets the encap type */
648 		encap_type = 0;
649 
650 		if (xfrm_replay_recheck(x, skb, seq)) {
651 			XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATESEQERROR);
652 			goto drop_unlock;
653 		}
654 
655 		xfrm_replay_advance(x, seq);
656 
657 		x->curlft.bytes += skb->len;
658 		x->curlft.packets++;
659 		x->lastused = ktime_get_real_seconds();
660 
661 		spin_unlock(&x->lock);
662 
663 		XFRM_MODE_SKB_CB(skb)->protocol = nexthdr;
664 
665 		if (xfrm_inner_mode_input(x, skb)) {
666 			XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMODEERROR);
667 			goto drop;
668 		}
669 
670 		if (x->outer_mode.flags & XFRM_MODE_FLAG_TUNNEL) {
671 			decaps = 1;
672 			break;
673 		}
674 
675 		/*
676 		 * We need the inner address.  However, we only get here for
677 		 * transport mode so the outer address is identical.
678 		 */
679 		daddr = &x->id.daddr;
680 		family = x->props.family;
681 
682 		err = xfrm_parse_spi(skb, nexthdr, &spi, &seq);
683 		if (err < 0) {
684 			XFRM_INC_STATS(net, LINUX_MIB_XFRMINHDRERROR);
685 			goto drop;
686 		}
687 		crypto_done = false;
688 	} while (!err);
689 
690 	err = xfrm_rcv_cb(skb, family, x->type->proto, 0);
691 	if (err)
692 		goto drop;
693 
694 	nf_reset_ct(skb);
695 
696 	if (decaps) {
697 		sp = skb_sec_path(skb);
698 		if (sp)
699 			sp->olen = 0;
700 		if (skb_valid_dst(skb))
701 			skb_dst_drop(skb);
702 		gro_cells_receive(&gro_cells, skb);
703 		return 0;
704 	} else {
705 		xo = xfrm_offload(skb);
706 		if (xo)
707 			xfrm_gro = xo->flags & XFRM_GRO;
708 
709 		err = -EAFNOSUPPORT;
710 		rcu_read_lock();
711 		afinfo = xfrm_state_afinfo_get_rcu(x->props.family);
712 		if (likely(afinfo))
713 			err = afinfo->transport_finish(skb, xfrm_gro || async);
714 		rcu_read_unlock();
715 		if (xfrm_gro) {
716 			sp = skb_sec_path(skb);
717 			if (sp)
718 				sp->olen = 0;
719 			if (skb_valid_dst(skb))
720 				skb_dst_drop(skb);
721 			gro_cells_receive(&gro_cells, skb);
722 			return err;
723 		}
724 
725 		return err;
726 	}
727 
728 drop_unlock:
729 	spin_unlock(&x->lock);
730 drop:
731 	xfrm_rcv_cb(skb, family, x && x->type ? x->type->proto : nexthdr, -1);
732 	kfree_skb(skb);
733 	return 0;
734 }
735 EXPORT_SYMBOL(xfrm_input);
736 
737 int xfrm_input_resume(struct sk_buff *skb, int nexthdr)
738 {
739 	return xfrm_input(skb, nexthdr, 0, -1);
740 }
741 EXPORT_SYMBOL(xfrm_input_resume);
742 
743 static void xfrm_trans_reinject(struct work_struct *work)
744 {
745 	struct xfrm_trans_tasklet *trans = container_of(work, struct xfrm_trans_tasklet, work);
746 	struct sk_buff_head queue;
747 	struct sk_buff *skb;
748 
749 	__skb_queue_head_init(&queue);
750 	spin_lock_bh(&trans->queue_lock);
751 	skb_queue_splice_init(&trans->queue, &queue);
752 	spin_unlock_bh(&trans->queue_lock);
753 
754 	local_bh_disable();
755 	while ((skb = __skb_dequeue(&queue)))
756 		XFRM_TRANS_SKB_CB(skb)->finish(XFRM_TRANS_SKB_CB(skb)->net,
757 					       NULL, skb);
758 	local_bh_enable();
759 }
760 
761 int xfrm_trans_queue_net(struct net *net, struct sk_buff *skb,
762 			 int (*finish)(struct net *, struct sock *,
763 				       struct sk_buff *))
764 {
765 	struct xfrm_trans_tasklet *trans;
766 
767 	trans = this_cpu_ptr(&xfrm_trans_tasklet);
768 
769 	if (skb_queue_len(&trans->queue) >= READ_ONCE(netdev_max_backlog))
770 		return -ENOBUFS;
771 
772 	BUILD_BUG_ON(sizeof(struct xfrm_trans_cb) > sizeof(skb->cb));
773 
774 	XFRM_TRANS_SKB_CB(skb)->finish = finish;
775 	XFRM_TRANS_SKB_CB(skb)->net = net;
776 	spin_lock_bh(&trans->queue_lock);
777 	__skb_queue_tail(&trans->queue, skb);
778 	spin_unlock_bh(&trans->queue_lock);
779 	schedule_work(&trans->work);
780 	return 0;
781 }
782 EXPORT_SYMBOL(xfrm_trans_queue_net);
783 
784 int xfrm_trans_queue(struct sk_buff *skb,
785 		     int (*finish)(struct net *, struct sock *,
786 				   struct sk_buff *))
787 {
788 	return xfrm_trans_queue_net(dev_net(skb->dev), skb, finish);
789 }
790 EXPORT_SYMBOL(xfrm_trans_queue);
791 
792 void __init xfrm_input_init(void)
793 {
794 	int err;
795 	int i;
796 
797 	init_dummy_netdev(&xfrm_napi_dev);
798 	err = gro_cells_init(&gro_cells, &xfrm_napi_dev);
799 	if (err)
800 		gro_cells.cells = NULL;
801 
802 	for_each_possible_cpu(i) {
803 		struct xfrm_trans_tasklet *trans;
804 
805 		trans = &per_cpu(xfrm_trans_tasklet, i);
806 		spin_lock_init(&trans->queue_lock);
807 		__skb_queue_head_init(&trans->queue);
808 		INIT_WORK(&trans->work, xfrm_trans_reinject);
809 	}
810 }
811