xref: /linux/net/xfrm/xfrm_input.c (revision 8be4d31cb8aaeea27bde4b7ddb26e28a89062ebf)
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 
xfrm_input_register_afinfo(const struct xfrm_input_afinfo * afinfo)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 
xfrm_input_unregister_afinfo(const struct xfrm_input_afinfo * afinfo)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 
xfrm_input_get_afinfo(u8 family,bool is_ipip)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 
xfrm_rcv_cb(struct sk_buff * skb,unsigned int family,u8 protocol,int err)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 
secpath_set(struct sk_buff * skb)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 
xfrm_parse_spi(struct sk_buff * skb,u8 nexthdr,__be32 * spi,__be32 * seq)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 
xfrm4_remove_beet_encap(struct xfrm_state * x,struct sk_buff * skb)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 
ipip_ecn_decapsulate(struct sk_buff * skb)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 
xfrm4_remove_tunnel_encap(struct xfrm_state * x,struct sk_buff * skb)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 
ipip6_ecn_decapsulate(struct sk_buff * skb)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 
xfrm6_remove_tunnel_encap(struct xfrm_state * x,struct sk_buff * skb)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 
xfrm6_remove_beet_encap(struct xfrm_state * x,struct sk_buff * skb)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
xfrm_inner_mode_encap_remove(struct xfrm_state * x,struct sk_buff * skb)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 
xfrm_prepare_input(struct xfrm_state * x,struct sk_buff * skb)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  */
xfrm4_transport_input(struct xfrm_state * x,struct sk_buff * skb)390 static 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 
xfrm6_transport_input(struct xfrm_state * x,struct sk_buff * skb)408 static 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 
xfrm_inner_mode_input(struct xfrm_state * x,struct sk_buff * skb)432 static 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  */
xfrm_input(struct sk_buff * skb,int nexthdr,__be32 spi,int encap_type)463 int 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 			dev_put(skb->dev);
507 			seq = XFRM_SKB_CB(skb)->seq.input.low;
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 
546 		goto lock;
547 	}
548 
549 	family = XFRM_SPI_SKB_CB(skb)->family;
550 
551 	/* if tunnel is present override skb->mark value with tunnel i_key */
552 	switch (family) {
553 	case AF_INET:
554 		if (XFRM_TUNNEL_SKB_CB(skb)->tunnel.ip4)
555 			mark = be32_to_cpu(XFRM_TUNNEL_SKB_CB(skb)->tunnel.ip4->parms.i_key);
556 		break;
557 	case AF_INET6:
558 		if (XFRM_TUNNEL_SKB_CB(skb)->tunnel.ip6)
559 			mark = be32_to_cpu(XFRM_TUNNEL_SKB_CB(skb)->tunnel.ip6->parms.i_key);
560 		break;
561 	}
562 
563 	sp = secpath_set(skb);
564 	if (!sp) {
565 		XFRM_INC_STATS(net, LINUX_MIB_XFRMINERROR);
566 		goto drop;
567 	}
568 
569 	seq = 0;
570 	if (!spi && xfrm_parse_spi(skb, nexthdr, &spi, &seq)) {
571 		secpath_reset(skb);
572 		XFRM_INC_STATS(net, LINUX_MIB_XFRMINHDRERROR);
573 		goto drop;
574 	}
575 
576 	daddr = (xfrm_address_t *)(skb_network_header(skb) +
577 				   XFRM_SPI_SKB_CB(skb)->daddroff);
578 	do {
579 		sp = skb_sec_path(skb);
580 
581 		if (sp->len == XFRM_MAX_DEPTH) {
582 			secpath_reset(skb);
583 			XFRM_INC_STATS(net, LINUX_MIB_XFRMINBUFFERERROR);
584 			goto drop;
585 		}
586 
587 		x = xfrm_input_state_lookup(net, mark, daddr, spi, nexthdr, family);
588 		if (x == NULL) {
589 			secpath_reset(skb);
590 			XFRM_INC_STATS(net, LINUX_MIB_XFRMINNOSTATES);
591 			xfrm_audit_state_notfound(skb, family, spi, seq);
592 			goto drop;
593 		}
594 
595 		if (unlikely(x->dir && x->dir != XFRM_SA_DIR_IN)) {
596 			secpath_reset(skb);
597 			XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEDIRERROR);
598 			xfrm_audit_state_notfound(skb, family, spi, seq);
599 			xfrm_state_put(x);
600 			x = NULL;
601 			goto drop;
602 		}
603 
604 		skb->mark = xfrm_smark_get(skb->mark, x);
605 
606 		sp->xvec[sp->len++] = x;
607 
608 		skb_dst_force(skb);
609 		if (!skb_dst(skb)) {
610 			XFRM_INC_STATS(net, LINUX_MIB_XFRMINERROR);
611 			goto drop;
612 		}
613 
614 lock:
615 		spin_lock(&x->lock);
616 
617 		if (unlikely(x->km.state != XFRM_STATE_VALID)) {
618 			if (x->km.state == XFRM_STATE_ACQ)
619 				XFRM_INC_STATS(net, LINUX_MIB_XFRMACQUIREERROR);
620 			else
621 				XFRM_INC_STATS(net,
622 					       LINUX_MIB_XFRMINSTATEINVALID);
623 			goto drop_unlock;
624 		}
625 
626 		if ((x->encap ? x->encap->encap_type : 0) != encap_type) {
627 			XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMISMATCH);
628 			goto drop_unlock;
629 		}
630 
631 		if (xfrm_replay_check(x, skb, seq)) {
632 			XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATESEQERROR);
633 			goto drop_unlock;
634 		}
635 
636 		if (xfrm_state_check_expire(x)) {
637 			XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEEXPIRED);
638 			goto drop_unlock;
639 		}
640 
641 		spin_unlock(&x->lock);
642 
643 		if (xfrm_tunnel_check(skb, x, family)) {
644 			XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMODEERROR);
645 			goto drop;
646 		}
647 
648 		seq_hi = htonl(xfrm_replay_seqhi(x, seq));
649 
650 		XFRM_SKB_CB(skb)->seq.input.low = seq;
651 		XFRM_SKB_CB(skb)->seq.input.hi = seq_hi;
652 
653 		if (crypto_done) {
654 			nexthdr = x->type_offload->input_tail(x, skb);
655 		} else {
656 			dev_hold(skb->dev);
657 
658 			nexthdr = x->type->input(x, skb);
659 			if (nexthdr == -EINPROGRESS)
660 				return 0;
661 
662 			dev_put(skb->dev);
663 		}
664 resume:
665 		spin_lock(&x->lock);
666 		if (nexthdr < 0) {
667 			if (nexthdr == -EBADMSG) {
668 				xfrm_audit_state_icvfail(x, skb,
669 							 x->type->proto);
670 				x->stats.integrity_failed++;
671 			}
672 			XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEPROTOERROR);
673 			goto drop_unlock;
674 		}
675 
676 		/* only the first xfrm gets the encap type */
677 		encap_type = 0;
678 
679 		if (xfrm_replay_recheck(x, skb, seq)) {
680 			XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATESEQERROR);
681 			goto drop_unlock;
682 		}
683 
684 		xfrm_replay_advance(x, seq);
685 
686 		x->curlft.bytes += skb->len;
687 		x->curlft.packets++;
688 		x->lastused = ktime_get_real_seconds();
689 
690 		spin_unlock(&x->lock);
691 
692 		XFRM_MODE_SKB_CB(skb)->protocol = nexthdr;
693 
694 		err = xfrm_inner_mode_input(x, skb);
695 		if (err == -EINPROGRESS)
696 			return 0;
697 		else if (err) {
698 			XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMODEERROR);
699 			goto drop;
700 		}
701 resume_decapped:
702 		if (x->outer_mode.flags & XFRM_MODE_FLAG_TUNNEL) {
703 			decaps = 1;
704 			break;
705 		}
706 
707 		/*
708 		 * We need the inner address.  However, we only get here for
709 		 * transport mode so the outer address is identical.
710 		 */
711 		daddr = &x->id.daddr;
712 		family = x->props.family;
713 
714 		err = xfrm_parse_spi(skb, nexthdr, &spi, &seq);
715 		if (err < 0) {
716 			XFRM_INC_STATS(net, LINUX_MIB_XFRMINHDRERROR);
717 			goto drop;
718 		}
719 		crypto_done = false;
720 	} while (!err);
721 
722 	err = xfrm_rcv_cb(skb, family, x->type->proto, 0);
723 	if (err)
724 		goto drop;
725 
726 	nf_reset_ct(skb);
727 
728 	if (decaps) {
729 		sp = skb_sec_path(skb);
730 		if (sp)
731 			sp->olen = 0;
732 		if (skb_valid_dst(skb))
733 			skb_dst_drop(skb);
734 		gro_cells_receive(&gro_cells, skb);
735 		return 0;
736 	} else {
737 		xo = xfrm_offload(skb);
738 		if (xo)
739 			xfrm_gro = xo->flags & XFRM_GRO;
740 
741 		err = -EAFNOSUPPORT;
742 		rcu_read_lock();
743 		afinfo = xfrm_state_afinfo_get_rcu(x->props.family);
744 		if (likely(afinfo))
745 			err = afinfo->transport_finish(skb, xfrm_gro || async);
746 		rcu_read_unlock();
747 		if (xfrm_gro) {
748 			sp = skb_sec_path(skb);
749 			if (sp)
750 				sp->olen = 0;
751 			if (skb_valid_dst(skb))
752 				skb_dst_drop(skb);
753 			gro_cells_receive(&gro_cells, skb);
754 			return err;
755 		}
756 
757 		return err;
758 	}
759 
760 drop_unlock:
761 	spin_unlock(&x->lock);
762 drop:
763 	xfrm_rcv_cb(skb, family, x && x->type ? x->type->proto : nexthdr, -1);
764 	kfree_skb(skb);
765 	return 0;
766 }
767 EXPORT_SYMBOL(xfrm_input);
768 
xfrm_input_resume(struct sk_buff * skb,int nexthdr)769 int xfrm_input_resume(struct sk_buff *skb, int nexthdr)
770 {
771 	return xfrm_input(skb, nexthdr, 0, -1);
772 }
773 EXPORT_SYMBOL(xfrm_input_resume);
774 
xfrm_trans_reinject(struct work_struct * work)775 static void xfrm_trans_reinject(struct work_struct *work)
776 {
777 	struct xfrm_trans_tasklet *trans = container_of(work, struct xfrm_trans_tasklet, work);
778 	struct sk_buff_head queue;
779 	struct sk_buff *skb;
780 
781 	__skb_queue_head_init(&queue);
782 	spin_lock_bh(&trans->queue_lock);
783 	skb_queue_splice_init(&trans->queue, &queue);
784 	spin_unlock_bh(&trans->queue_lock);
785 
786 	local_bh_disable();
787 	while ((skb = __skb_dequeue(&queue)))
788 		XFRM_TRANS_SKB_CB(skb)->finish(XFRM_TRANS_SKB_CB(skb)->net,
789 					       NULL, skb);
790 	local_bh_enable();
791 }
792 
xfrm_trans_queue_net(struct net * net,struct sk_buff * skb,int (* finish)(struct net *,struct sock *,struct sk_buff *))793 int xfrm_trans_queue_net(struct net *net, struct sk_buff *skb,
794 			 int (*finish)(struct net *, struct sock *,
795 				       struct sk_buff *))
796 {
797 	struct xfrm_trans_tasklet *trans;
798 
799 	trans = this_cpu_ptr(&xfrm_trans_tasklet);
800 
801 	if (skb_queue_len(&trans->queue) >= READ_ONCE(net_hotdata.max_backlog))
802 		return -ENOBUFS;
803 
804 	BUILD_BUG_ON(sizeof(struct xfrm_trans_cb) > sizeof(skb->cb));
805 
806 	XFRM_TRANS_SKB_CB(skb)->finish = finish;
807 	XFRM_TRANS_SKB_CB(skb)->net = net;
808 	spin_lock_bh(&trans->queue_lock);
809 	__skb_queue_tail(&trans->queue, skb);
810 	spin_unlock_bh(&trans->queue_lock);
811 	schedule_work(&trans->work);
812 	return 0;
813 }
814 EXPORT_SYMBOL(xfrm_trans_queue_net);
815 
xfrm_trans_queue(struct sk_buff * skb,int (* finish)(struct net *,struct sock *,struct sk_buff *))816 int xfrm_trans_queue(struct sk_buff *skb,
817 		     int (*finish)(struct net *, struct sock *,
818 				   struct sk_buff *))
819 {
820 	return xfrm_trans_queue_net(dev_net(skb->dev), skb, finish);
821 }
822 EXPORT_SYMBOL(xfrm_trans_queue);
823 
xfrm_input_init(void)824 void __init xfrm_input_init(void)
825 {
826 	int err;
827 	int i;
828 
829 	xfrm_napi_dev = alloc_netdev_dummy(0);
830 	if (!xfrm_napi_dev)
831 		panic("Failed to allocate XFRM dummy netdev\n");
832 
833 	err = gro_cells_init(&gro_cells, xfrm_napi_dev);
834 	if (err)
835 		gro_cells.cells = NULL;
836 
837 	for_each_possible_cpu(i) {
838 		struct xfrm_trans_tasklet *trans;
839 
840 		trans = &per_cpu(xfrm_trans_tasklet, i);
841 		spin_lock_init(&trans->queue_lock);
842 		__skb_queue_head_init(&trans->queue);
843 		INIT_WORK(&trans->work, xfrm_trans_reinject);
844 	}
845 }
846