xfrm_input.c

  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
 28struct xfrm_trans_tasklet {
 29	struct work_struct work;
 30	spinlock_t queue_lock;
 31	struct sk_buff_head queue;
 32};
 33
 34struct 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
 47static DEFINE_SPINLOCK(xfrm_input_afinfo_lock);
 48static struct xfrm_input_afinfo const __rcu *xfrm_input_afinfo[2][AF_INET6 + 1];
 49
 50static struct gro_cells gro_cells;
 51static struct net_device *xfrm_napi_dev;
 52
 53static DEFINE_PER_CPU(struct xfrm_trans_tasklet, xfrm_trans_tasklet);
 54
 55int 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}
 70EXPORT_SYMBOL(xfrm_input_register_afinfo);
 71
 72int 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}
 87EXPORT_SYMBOL(xfrm_input_unregister_afinfo);
 88
 89static 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
103static 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
120struct 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}
139EXPORT_SYMBOL(secpath_set);
140
141/* Fetch spi and seq from ipsec header */
142
143int 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}
176EXPORT_SYMBOL(xfrm_parse_spi);
177
178static 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;
222out:
223	return err;
224}
225
226static 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
234static 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
259out:
260	return err;
261}
262
263static 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
271static 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
296out:
297	return err;
298}
299
300static 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;
323out:
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 */
337static int
338xfrm_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
365static 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 */
390static int xfrm4_transport_input(struct xfrm_state *x, struct sk_buff *skb)
391{
392	struct xfrm_offload *xo = xfrm_offload(skb);
393	int ihl = skb->data - skb_transport_header(skb);
394
395	if (skb->transport_header != skb->network_header) {
396		memmove(skb_transport_header(skb),
397			skb_network_header(skb), ihl);
398		if (xo)
399			xo->orig_mac_len =
400				skb_mac_header_was_set(skb) ? skb_mac_header_len(skb) : 0;
401		skb->network_header = skb->transport_header;
402	}
403	ip_hdr(skb)->tot_len = htons(skb->len + ihl);
404	skb_reset_transport_header(skb);
405	return 0;
406}
407
408static int xfrm6_transport_input(struct xfrm_state *x, struct sk_buff *skb)
409{
410#if IS_ENABLED(CONFIG_IPV6)
411	struct xfrm_offload *xo = xfrm_offload(skb);
412	int ihl = skb->data - skb_transport_header(skb);
413
414	if (skb->transport_header != skb->network_header) {
415		memmove(skb_transport_header(skb),
416			skb_network_header(skb), ihl);
417		if (xo)
418			xo->orig_mac_len =
419				skb_mac_header_was_set(skb) ? skb_mac_header_len(skb) : 0;
420		skb->network_header = skb->transport_header;
421	}
422	ipv6_hdr(skb)->payload_len = htons(skb->len + ihl -
423					   sizeof(struct ipv6hdr));
424	skb_reset_transport_header(skb);
425	return 0;
426#else
427	WARN_ON_ONCE(1);
428	return -EAFNOSUPPORT;
429#endif
430}
431
432static int xfrm_inner_mode_input(struct xfrm_state *x,
433				 struct sk_buff *skb)
434{
435	switch (x->props.mode) {
436	case XFRM_MODE_BEET:
437	case XFRM_MODE_TUNNEL:
438		return xfrm_prepare_input(x, skb);
439	case XFRM_MODE_TRANSPORT:
440		if (x->props.family == AF_INET)
441			return xfrm4_transport_input(x, skb);
442		if (x->props.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		if (x->mode_cbs && x->mode_cbs->input)
450			return x->mode_cbs->input(x, skb);
451
452		WARN_ON_ONCE(1);
453		break;
454	}
455
456	return -EOPNOTSUPP;
457}
458
459/* NOTE: encap_type - In addition to the normal (non-negative) values for
460 * encap_type, a negative value of -1 or -2 can be used to resume/restart this
461 * function after a previous invocation early terminated for async operation.
462 */
463int xfrm_input(struct sk_buff *skb, int nexthdr, __be32 spi, int encap_type)
464{
465	const struct xfrm_state_afinfo *afinfo;
466	struct net *net = dev_net(skb->dev);
467	int err;
468	__be32 seq;
469	__be32 seq_hi;
470	struct xfrm_state *x = NULL;
471	xfrm_address_t *daddr;
472	u32 mark = skb->mark;
473	unsigned int family = AF_UNSPEC;
474	int decaps = 0;
475	int async = 0;
476	bool xfrm_gro = false;
477	bool crypto_done = false;
478	struct xfrm_offload *xo = xfrm_offload(skb);
479	struct sec_path *sp;
480
481	if (encap_type < 0 || (xo && (xo->flags & XFRM_GRO || encap_type == 0 ||
482				      encap_type == UDP_ENCAP_ESPINUDP))) {
483		x = xfrm_input_state(skb);
484
485		if (unlikely(x->km.state != XFRM_STATE_VALID)) {
486			if (x->km.state == XFRM_STATE_ACQ)
487				XFRM_INC_STATS(net, LINUX_MIB_XFRMACQUIREERROR);
488			else
489				XFRM_INC_STATS(net,
490					       LINUX_MIB_XFRMINSTATEINVALID);
491
492			if (encap_type == -1)
493				dev_put(skb->dev);
494			goto drop;
495		}
496
497		family = x->props.family;
498
499		/* An encap_type of -2 indicates reconstructed inner packet */
500		if (encap_type == -2)
501			goto resume_decapped;
502
503		/* An encap_type of -1 indicates async resumption. */
504		if (encap_type == -1) {
505			async = 1;
506			seq = XFRM_SKB_CB(skb)->seq.input.low;
507			spin_lock(&x->lock);
508			goto resume;
509		}
510		/* GRO call */
511		seq = XFRM_SPI_SKB_CB(skb)->seq;
512
513		if (xo && (xo->flags & CRYPTO_DONE)) {
514			crypto_done = true;
515			family = XFRM_SPI_SKB_CB(skb)->family;
516
517			if (!(xo->status & CRYPTO_SUCCESS)) {
518				if (xo->status &
519				    (CRYPTO_TRANSPORT_AH_AUTH_FAILED |
520				     CRYPTO_TRANSPORT_ESP_AUTH_FAILED |
521				     CRYPTO_TUNNEL_AH_AUTH_FAILED |
522				     CRYPTO_TUNNEL_ESP_AUTH_FAILED)) {
523
524					xfrm_audit_state_icvfail(x, skb,
525								 x->type->proto);
526					x->stats.integrity_failed++;
527					XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEPROTOERROR);
528					goto drop;
529				}
530
531				if (xo->status & CRYPTO_INVALID_PROTOCOL) {
532					XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEPROTOERROR);
533					goto drop;
534				}
535
536				XFRM_INC_STATS(net, LINUX_MIB_XFRMINBUFFERERROR);
537				goto drop;
538			}
539
540			if (xfrm_parse_spi(skb, nexthdr, &spi, &seq)) {
541				XFRM_INC_STATS(net, LINUX_MIB_XFRMINHDRERROR);
542				goto drop;
543			}
544
545			nexthdr = x->type_offload->input_tail(x, skb);
546		}
547
548		goto process;
549	}
550
551	family = XFRM_SPI_SKB_CB(skb)->family;
552
553	/* if tunnel is present override skb->mark value with tunnel i_key */
554	switch (family) {
555	case AF_INET:
556		if (XFRM_TUNNEL_SKB_CB(skb)->tunnel.ip4)
557			mark = be32_to_cpu(XFRM_TUNNEL_SKB_CB(skb)->tunnel.ip4->parms.i_key);
558		break;
559	case AF_INET6:
560		if (XFRM_TUNNEL_SKB_CB(skb)->tunnel.ip6)
561			mark = be32_to_cpu(XFRM_TUNNEL_SKB_CB(skb)->tunnel.ip6->parms.i_key);
562		break;
563	}
564
565	sp = secpath_set(skb);
566	if (!sp) {
567		XFRM_INC_STATS(net, LINUX_MIB_XFRMINERROR);
568		goto drop;
569	}
570
571	seq = 0;
572	if (!spi && xfrm_parse_spi(skb, nexthdr, &spi, &seq)) {
573		secpath_reset(skb);
574		XFRM_INC_STATS(net, LINUX_MIB_XFRMINHDRERROR);
575		goto drop;
576	}
577
578	daddr = (xfrm_address_t *)(skb_network_header(skb) +
579				   XFRM_SPI_SKB_CB(skb)->daddroff);
580	do {
581		sp = skb_sec_path(skb);
582
583		if (sp->len == XFRM_MAX_DEPTH) {
584			secpath_reset(skb);
585			XFRM_INC_STATS(net, LINUX_MIB_XFRMINBUFFERERROR);
586			goto drop;
587		}
588
589		x = xfrm_input_state_lookup(net, mark, daddr, spi, nexthdr, family);
590		if (x == NULL) {
591			secpath_reset(skb);
592			XFRM_INC_STATS(net, LINUX_MIB_XFRMINNOSTATES);
593			xfrm_audit_state_notfound(skb, family, spi, seq);
594			goto drop;
595		}
596
597		if (unlikely(x->dir && x->dir != XFRM_SA_DIR_IN)) {
598			secpath_reset(skb);
599			XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEDIRERROR);
600			xfrm_audit_state_notfound(skb, family, spi, seq);
601			xfrm_state_put(x);
602			x = NULL;
603			goto drop;
604		}
605
606		skb->mark = xfrm_smark_get(skb->mark, x);
607
608		sp->xvec[sp->len++] = x;
609
610		skb_dst_force(skb);
611		if (!skb_dst(skb)) {
612			XFRM_INC_STATS(net, LINUX_MIB_XFRMINERROR);
613			goto drop;
614		}
615
616process:
617		seq_hi = htonl(xfrm_replay_seqhi(x, seq));
618
619		XFRM_SKB_CB(skb)->seq.input.low = seq;
620		XFRM_SKB_CB(skb)->seq.input.hi = seq_hi;
621
622		spin_lock(&x->lock);
623
624		if (unlikely(x->km.state != XFRM_STATE_VALID)) {
625			if (x->km.state == XFRM_STATE_ACQ)
626				XFRM_INC_STATS(net, LINUX_MIB_XFRMACQUIREERROR);
627			else
628				XFRM_INC_STATS(net,
629					       LINUX_MIB_XFRMINSTATEINVALID);
630			goto drop_unlock;
631		}
632
633		if ((x->encap ? x->encap->encap_type : 0) != encap_type) {
634			XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMISMATCH);
635			goto drop_unlock;
636		}
637
638		if (xfrm_replay_check(x, skb, seq)) {
639			XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATESEQERROR);
640			goto drop_unlock;
641		}
642
643		if (xfrm_state_check_expire(x)) {
644			XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEEXPIRED);
645			goto drop_unlock;
646		}
647
648		if (xfrm_tunnel_check(skb, x, family)) {
649			XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMODEERROR);
650			goto drop_unlock;
651		}
652
653		if (!crypto_done) {
654			spin_unlock(&x->lock);
655			dev_hold(skb->dev);
656
657			nexthdr = x->type->input(x, skb);
658			if (nexthdr == -EINPROGRESS) {
659				if (async)
660					dev_put(skb->dev);
661				return 0;
662			}
663
664			dev_put(skb->dev);
665			spin_lock(&x->lock);
666		}
667resume:
668		if (nexthdr < 0) {
669			if (nexthdr == -EBADMSG) {
670				xfrm_audit_state_icvfail(x, skb,
671							 x->type->proto);
672				x->stats.integrity_failed++;
673			}
674			XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEPROTOERROR);
675			goto drop_unlock;
676		}
677
678		/* only the first xfrm gets the encap type */
679		encap_type = 0;
680
681		if (!crypto_done && xfrm_replay_recheck(x, skb, seq)) {
682			XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATESEQERROR);
683			goto drop_unlock;
684		}
685
686		xfrm_replay_advance(x, seq);
687
688		x->curlft.bytes += skb->len;
689		x->curlft.packets++;
690		x->lastused = ktime_get_real_seconds();
691
692		spin_unlock(&x->lock);
693
694		XFRM_MODE_SKB_CB(skb)->protocol = nexthdr;
695
696		err = xfrm_inner_mode_input(x, skb);
697		if (err == -EINPROGRESS) {
698			if (async)
699				dev_put(skb->dev);
700			return 0;
701		} else if (err) {
702			XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMODEERROR);
703			goto drop;
704		}
705resume_decapped:
706		if (x->outer_mode.flags & XFRM_MODE_FLAG_TUNNEL) {
707			decaps = 1;
708			break;
709		}
710
711		/*
712		 * We need the inner address.  However, we only get here for
713		 * transport mode so the outer address is identical.
714		 */
715		daddr = &x->id.daddr;
716		family = x->props.family;
717
718		err = xfrm_parse_spi(skb, nexthdr, &spi, &seq);
719		if (err < 0) {
720			XFRM_INC_STATS(net, LINUX_MIB_XFRMINHDRERROR);
721			goto drop;
722		}
723		crypto_done = false;
724	} while (!err);
725
726	err = xfrm_rcv_cb(skb, family, x->type->proto, 0);
727	if (err)
728		goto drop;
729
730	nf_reset_ct(skb);
731
732	if (decaps) {
733		sp = skb_sec_path(skb);
734		if (sp)
735			sp->olen = 0;
736		if (skb_valid_dst(skb))
737			skb_dst_drop(skb);
738		if (async)
739			dev_put(skb->dev);
740		gro_cells_receive(&gro_cells, skb);
741		return 0;
742	} else {
743		xo = xfrm_offload(skb);
744		if (xo)
745			xfrm_gro = xo->flags & XFRM_GRO;
746
747		err = -EAFNOSUPPORT;
748		rcu_read_lock();
749		afinfo = xfrm_state_afinfo_get_rcu(x->props.family);
750		if (likely(afinfo))
751			err = afinfo->transport_finish(skb, xfrm_gro || async);
752		rcu_read_unlock();
753		if (xfrm_gro) {
754			sp = skb_sec_path(skb);
755			if (sp)
756				sp->olen = 0;
757			if (skb_valid_dst(skb))
758				skb_dst_drop(skb);
759			if (async)
760				dev_put(skb->dev);
761			gro_cells_receive(&gro_cells, skb);
762			return err;
763		}
764
765		return err;
766	}
767
768drop_unlock:
769	spin_unlock(&x->lock);
770drop:
771	if (async)
772		dev_put(skb->dev);
773	xfrm_rcv_cb(skb, family, x && x->type ? x->type->proto : nexthdr, -1);
774	kfree_skb(skb);
775	return 0;
776}
777EXPORT_SYMBOL(xfrm_input);
778
779int xfrm_input_resume(struct sk_buff *skb, int nexthdr)
780{
781	return xfrm_input(skb, nexthdr, 0, -1);
782}
783EXPORT_SYMBOL(xfrm_input_resume);
784
785static void xfrm_trans_reinject(struct work_struct *work)
786{
787	struct xfrm_trans_tasklet *trans = container_of(work, struct xfrm_trans_tasklet, work);
788	struct sk_buff_head queue;
789	struct sk_buff *skb;
790
791	__skb_queue_head_init(&queue);
792	spin_lock_bh(&trans->queue_lock);
793	skb_queue_splice_init(&trans->queue, &queue);
794	spin_unlock_bh(&trans->queue_lock);
795
796	local_bh_disable();
797	while ((skb = __skb_dequeue(&queue)))
798		XFRM_TRANS_SKB_CB(skb)->finish(XFRM_TRANS_SKB_CB(skb)->net,
799					       NULL, skb);
800	local_bh_enable();
801}
802
803int xfrm_trans_queue_net(struct net *net, struct sk_buff *skb,
804			 int (*finish)(struct net *, struct sock *,
805				       struct sk_buff *))
806{
807	struct xfrm_trans_tasklet *trans;
808
809	trans = this_cpu_ptr(&xfrm_trans_tasklet);
810
811	if (skb_queue_len(&trans->queue) >= READ_ONCE(net_hotdata.max_backlog))
812		return -ENOBUFS;
813
814	BUILD_BUG_ON(sizeof(struct xfrm_trans_cb) > sizeof(skb->cb));
815
816	XFRM_TRANS_SKB_CB(skb)->finish = finish;
817	XFRM_TRANS_SKB_CB(skb)->net = net;
818	spin_lock_bh(&trans->queue_lock);
819	__skb_queue_tail(&trans->queue, skb);
820	spin_unlock_bh(&trans->queue_lock);
821	schedule_work(&trans->work);
822	return 0;
823}
824EXPORT_SYMBOL(xfrm_trans_queue_net);
825
826int xfrm_trans_queue(struct sk_buff *skb,
827		     int (*finish)(struct net *, struct sock *,
828				   struct sk_buff *))
829{
830	return xfrm_trans_queue_net(dev_net(skb->dev), skb, finish);
831}
832EXPORT_SYMBOL(xfrm_trans_queue);
833
834void __init xfrm_input_init(void)
835{
836	int err;
837	int i;
838
839	xfrm_napi_dev = alloc_netdev_dummy(0);
840	if (!xfrm_napi_dev)
841		panic("Failed to allocate XFRM dummy netdev\n");
842
843	err = gro_cells_init(&gro_cells, xfrm_napi_dev);
844	if (err)
845		gro_cells.cells = NULL;
846
847	for_each_possible_cpu(i) {
848		struct xfrm_trans_tasklet *trans;
849
850		trans = &per_cpu(xfrm_trans_tasklet, i);
851		spin_lock_init(&trans->queue_lock);
852		__skb_queue_head_init(&trans->queue);
853		INIT_WORK(&trans->work, xfrm_trans_reinject);
854	}
855}