Loading...
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
24#include "xfrm_inout.h"
25
26struct xfrm_trans_tasklet {
27 struct tasklet_struct tasklet;
28 struct sk_buff_head queue;
29};
30
31struct xfrm_trans_cb {
32 union {
33 struct inet_skb_parm h4;
34#if IS_ENABLED(CONFIG_IPV6)
35 struct inet6_skb_parm h6;
36#endif
37 } header;
38 int (*finish)(struct net *net, struct sock *sk, struct sk_buff *skb);
39};
40
41#define XFRM_TRANS_SKB_CB(__skb) ((struct xfrm_trans_cb *)&((__skb)->cb[0]))
42
43static DEFINE_SPINLOCK(xfrm_input_afinfo_lock);
44static struct xfrm_input_afinfo const __rcu *xfrm_input_afinfo[AF_INET6 + 1];
45
46static struct gro_cells gro_cells;
47static struct net_device xfrm_napi_dev;
48
49static DEFINE_PER_CPU(struct xfrm_trans_tasklet, xfrm_trans_tasklet);
50
51int xfrm_input_register_afinfo(const struct xfrm_input_afinfo *afinfo)
52{
53 int err = 0;
54
55 if (WARN_ON(afinfo->family >= ARRAY_SIZE(xfrm_input_afinfo)))
56 return -EAFNOSUPPORT;
57
58 spin_lock_bh(&xfrm_input_afinfo_lock);
59 if (unlikely(xfrm_input_afinfo[afinfo->family] != NULL))
60 err = -EEXIST;
61 else
62 rcu_assign_pointer(xfrm_input_afinfo[afinfo->family], afinfo);
63 spin_unlock_bh(&xfrm_input_afinfo_lock);
64 return err;
65}
66EXPORT_SYMBOL(xfrm_input_register_afinfo);
67
68int xfrm_input_unregister_afinfo(const struct xfrm_input_afinfo *afinfo)
69{
70 int err = 0;
71
72 spin_lock_bh(&xfrm_input_afinfo_lock);
73 if (likely(xfrm_input_afinfo[afinfo->family] != NULL)) {
74 if (unlikely(xfrm_input_afinfo[afinfo->family] != afinfo))
75 err = -EINVAL;
76 else
77 RCU_INIT_POINTER(xfrm_input_afinfo[afinfo->family], NULL);
78 }
79 spin_unlock_bh(&xfrm_input_afinfo_lock);
80 synchronize_rcu();
81 return err;
82}
83EXPORT_SYMBOL(xfrm_input_unregister_afinfo);
84
85static const struct xfrm_input_afinfo *xfrm_input_get_afinfo(unsigned int family)
86{
87 const struct xfrm_input_afinfo *afinfo;
88
89 if (WARN_ON_ONCE(family >= ARRAY_SIZE(xfrm_input_afinfo)))
90 return NULL;
91
92 rcu_read_lock();
93 afinfo = rcu_dereference(xfrm_input_afinfo[family]);
94 if (unlikely(!afinfo))
95 rcu_read_unlock();
96 return afinfo;
97}
98
99static int xfrm_rcv_cb(struct sk_buff *skb, unsigned int family, u8 protocol,
100 int err)
101{
102 int ret;
103 const struct xfrm_input_afinfo *afinfo = xfrm_input_get_afinfo(family);
104
105 if (!afinfo)
106 return -EAFNOSUPPORT;
107
108 ret = afinfo->callback(skb, protocol, err);
109 rcu_read_unlock();
110
111 return ret;
112}
113
114struct sec_path *secpath_set(struct sk_buff *skb)
115{
116 struct sec_path *sp, *tmp = skb_ext_find(skb, SKB_EXT_SEC_PATH);
117
118 sp = skb_ext_add(skb, SKB_EXT_SEC_PATH);
119 if (!sp)
120 return NULL;
121
122 if (tmp) /* reused existing one (was COW'd if needed) */
123 return sp;
124
125 /* allocated new secpath */
126 memset(sp->ovec, 0, sizeof(sp->ovec));
127 sp->olen = 0;
128 sp->len = 0;
129
130 return sp;
131}
132EXPORT_SYMBOL(secpath_set);
133
134/* Fetch spi and seq from ipsec header */
135
136int xfrm_parse_spi(struct sk_buff *skb, u8 nexthdr, __be32 *spi, __be32 *seq)
137{
138 int offset, offset_seq;
139 int hlen;
140
141 switch (nexthdr) {
142 case IPPROTO_AH:
143 hlen = sizeof(struct ip_auth_hdr);
144 offset = offsetof(struct ip_auth_hdr, spi);
145 offset_seq = offsetof(struct ip_auth_hdr, seq_no);
146 break;
147 case IPPROTO_ESP:
148 hlen = sizeof(struct ip_esp_hdr);
149 offset = offsetof(struct ip_esp_hdr, spi);
150 offset_seq = offsetof(struct ip_esp_hdr, seq_no);
151 break;
152 case IPPROTO_COMP:
153 if (!pskb_may_pull(skb, sizeof(struct ip_comp_hdr)))
154 return -EINVAL;
155 *spi = htonl(ntohs(*(__be16 *)(skb_transport_header(skb) + 2)));
156 *seq = 0;
157 return 0;
158 default:
159 return 1;
160 }
161
162 if (!pskb_may_pull(skb, hlen))
163 return -EINVAL;
164
165 *spi = *(__be32 *)(skb_transport_header(skb) + offset);
166 *seq = *(__be32 *)(skb_transport_header(skb) + offset_seq);
167 return 0;
168}
169EXPORT_SYMBOL(xfrm_parse_spi);
170
171static int xfrm4_remove_beet_encap(struct xfrm_state *x, struct sk_buff *skb)
172{
173 struct iphdr *iph;
174 int optlen = 0;
175 int err = -EINVAL;
176
177 if (unlikely(XFRM_MODE_SKB_CB(skb)->protocol == IPPROTO_BEETPH)) {
178 struct ip_beet_phdr *ph;
179 int phlen;
180
181 if (!pskb_may_pull(skb, sizeof(*ph)))
182 goto out;
183
184 ph = (struct ip_beet_phdr *)skb->data;
185
186 phlen = sizeof(*ph) + ph->padlen;
187 optlen = ph->hdrlen * 8 + (IPV4_BEET_PHMAXLEN - phlen);
188 if (optlen < 0 || optlen & 3 || optlen > 250)
189 goto out;
190
191 XFRM_MODE_SKB_CB(skb)->protocol = ph->nexthdr;
192
193 if (!pskb_may_pull(skb, phlen))
194 goto out;
195 __skb_pull(skb, phlen);
196 }
197
198 skb_push(skb, sizeof(*iph));
199 skb_reset_network_header(skb);
200 skb_mac_header_rebuild(skb);
201
202 xfrm4_beet_make_header(skb);
203
204 iph = ip_hdr(skb);
205
206 iph->ihl += optlen / 4;
207 iph->tot_len = htons(skb->len);
208 iph->daddr = x->sel.daddr.a4;
209 iph->saddr = x->sel.saddr.a4;
210 iph->check = 0;
211 iph->check = ip_fast_csum(skb_network_header(skb), iph->ihl);
212 err = 0;
213out:
214 return err;
215}
216
217static void ipip_ecn_decapsulate(struct sk_buff *skb)
218{
219 struct iphdr *inner_iph = ipip_hdr(skb);
220
221 if (INET_ECN_is_ce(XFRM_MODE_SKB_CB(skb)->tos))
222 IP_ECN_set_ce(inner_iph);
223}
224
225static int xfrm4_remove_tunnel_encap(struct xfrm_state *x, struct sk_buff *skb)
226{
227 int err = -EINVAL;
228
229 if (XFRM_MODE_SKB_CB(skb)->protocol != IPPROTO_IPIP)
230 goto out;
231
232 if (!pskb_may_pull(skb, sizeof(struct iphdr)))
233 goto out;
234
235 err = skb_unclone(skb, GFP_ATOMIC);
236 if (err)
237 goto out;
238
239 if (x->props.flags & XFRM_STATE_DECAP_DSCP)
240 ipv4_copy_dscp(XFRM_MODE_SKB_CB(skb)->tos, ipip_hdr(skb));
241 if (!(x->props.flags & XFRM_STATE_NOECN))
242 ipip_ecn_decapsulate(skb);
243
244 skb_reset_network_header(skb);
245 skb_mac_header_rebuild(skb);
246 if (skb->mac_len)
247 eth_hdr(skb)->h_proto = skb->protocol;
248
249 err = 0;
250
251out:
252 return err;
253}
254
255static void ipip6_ecn_decapsulate(struct sk_buff *skb)
256{
257 struct ipv6hdr *inner_iph = ipipv6_hdr(skb);
258
259 if (INET_ECN_is_ce(XFRM_MODE_SKB_CB(skb)->tos))
260 IP6_ECN_set_ce(skb, inner_iph);
261}
262
263static int xfrm6_remove_tunnel_encap(struct xfrm_state *x, struct sk_buff *skb)
264{
265 int err = -EINVAL;
266
267 if (XFRM_MODE_SKB_CB(skb)->protocol != IPPROTO_IPV6)
268 goto out;
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(ipv6_get_dsfield(ipv6_hdr(skb)),
278 ipipv6_hdr(skb));
279 if (!(x->props.flags & XFRM_STATE_NOECN))
280 ipip6_ecn_decapsulate(skb);
281
282 skb_reset_network_header(skb);
283 skb_mac_header_rebuild(skb);
284 if (skb->mac_len)
285 eth_hdr(skb)->h_proto = skb->protocol;
286
287 err = 0;
288
289out:
290 return err;
291}
292
293static int xfrm6_remove_beet_encap(struct xfrm_state *x, struct sk_buff *skb)
294{
295 struct ipv6hdr *ip6h;
296 int size = sizeof(struct ipv6hdr);
297 int err;
298
299 err = skb_cow_head(skb, size + skb->mac_len);
300 if (err)
301 goto out;
302
303 __skb_push(skb, size);
304 skb_reset_network_header(skb);
305 skb_mac_header_rebuild(skb);
306
307 xfrm6_beet_make_header(skb);
308
309 ip6h = ipv6_hdr(skb);
310 ip6h->payload_len = htons(skb->len - size);
311 ip6h->daddr = x->sel.daddr.in6;
312 ip6h->saddr = x->sel.saddr.in6;
313 err = 0;
314out:
315 return err;
316}
317
318/* Remove encapsulation header.
319 *
320 * The IP header will be moved over the top of the encapsulation
321 * header.
322 *
323 * On entry, the transport header shall point to where the IP header
324 * should be and the network header shall be set to where the IP
325 * header currently is. skb->data shall point to the start of the
326 * payload.
327 */
328static int
329xfrm_inner_mode_encap_remove(struct xfrm_state *x,
330 const struct xfrm_mode *inner_mode,
331 struct sk_buff *skb)
332{
333 switch (inner_mode->encap) {
334 case XFRM_MODE_BEET:
335 if (inner_mode->family == AF_INET)
336 return xfrm4_remove_beet_encap(x, skb);
337 if (inner_mode->family == AF_INET6)
338 return xfrm6_remove_beet_encap(x, skb);
339 break;
340 case XFRM_MODE_TUNNEL:
341 if (inner_mode->family == AF_INET)
342 return xfrm4_remove_tunnel_encap(x, skb);
343 if (inner_mode->family == AF_INET6)
344 return xfrm6_remove_tunnel_encap(x, skb);
345 break;
346 }
347
348 WARN_ON_ONCE(1);
349 return -EOPNOTSUPP;
350}
351
352static int xfrm_prepare_input(struct xfrm_state *x, struct sk_buff *skb)
353{
354 const struct xfrm_mode *inner_mode = &x->inner_mode;
355 const struct xfrm_state_afinfo *afinfo;
356 int err = -EAFNOSUPPORT;
357
358 rcu_read_lock();
359 afinfo = xfrm_state_afinfo_get_rcu(x->outer_mode.family);
360 if (likely(afinfo))
361 err = afinfo->extract_input(x, skb);
362 rcu_read_unlock();
363
364 if (err)
365 return err;
366
367 if (x->sel.family == AF_UNSPEC) {
368 inner_mode = xfrm_ip2inner_mode(x, XFRM_MODE_SKB_CB(skb)->protocol);
369 if (!inner_mode)
370 return -EAFNOSUPPORT;
371 }
372
373 switch (inner_mode->family) {
374 case AF_INET:
375 skb->protocol = htons(ETH_P_IP);
376 break;
377 case AF_INET6:
378 skb->protocol = htons(ETH_P_IPV6);
379 break;
380 default:
381 WARN_ON_ONCE(1);
382 break;
383 }
384
385 return xfrm_inner_mode_encap_remove(x, inner_mode, skb);
386}
387
388/* Remove encapsulation header.
389 *
390 * The IP header will be moved over the top of the encapsulation header.
391 *
392 * On entry, skb_transport_header() shall point to where the IP header
393 * should be and skb_network_header() shall be set to where the IP header
394 * currently is. skb->data shall point to the start of the payload.
395 */
396static int xfrm4_transport_input(struct xfrm_state *x, struct sk_buff *skb)
397{
398 int ihl = skb->data - skb_transport_header(skb);
399
400 if (skb->transport_header != skb->network_header) {
401 memmove(skb_transport_header(skb),
402 skb_network_header(skb), ihl);
403 skb->network_header = skb->transport_header;
404 }
405 ip_hdr(skb)->tot_len = htons(skb->len + ihl);
406 skb_reset_transport_header(skb);
407 return 0;
408}
409
410static int xfrm6_transport_input(struct xfrm_state *x, struct sk_buff *skb)
411{
412#if IS_ENABLED(CONFIG_IPV6)
413 int ihl = skb->data - skb_transport_header(skb);
414
415 if (skb->transport_header != skb->network_header) {
416 memmove(skb_transport_header(skb),
417 skb_network_header(skb), ihl);
418 skb->network_header = skb->transport_header;
419 }
420 ipv6_hdr(skb)->payload_len = htons(skb->len + ihl -
421 sizeof(struct ipv6hdr));
422 skb_reset_transport_header(skb);
423 return 0;
424#else
425 WARN_ON_ONCE(1);
426 return -EAFNOSUPPORT;
427#endif
428}
429
430static int xfrm_inner_mode_input(struct xfrm_state *x,
431 const struct xfrm_mode *inner_mode,
432 struct sk_buff *skb)
433{
434 switch (inner_mode->encap) {
435 case XFRM_MODE_BEET:
436 case XFRM_MODE_TUNNEL:
437 return xfrm_prepare_input(x, skb);
438 case XFRM_MODE_TRANSPORT:
439 if (inner_mode->family == AF_INET)
440 return xfrm4_transport_input(x, skb);
441 if (inner_mode->family == AF_INET6)
442 return xfrm6_transport_input(x, skb);
443 break;
444 case XFRM_MODE_ROUTEOPTIMIZATION:
445 WARN_ON_ONCE(1);
446 break;
447 default:
448 WARN_ON_ONCE(1);
449 break;
450 }
451
452 return -EOPNOTSUPP;
453}
454
455int xfrm_input(struct sk_buff *skb, int nexthdr, __be32 spi, int encap_type)
456{
457 const struct xfrm_state_afinfo *afinfo;
458 struct net *net = dev_net(skb->dev);
459 const struct xfrm_mode *inner_mode;
460 int err;
461 __be32 seq;
462 __be32 seq_hi;
463 struct xfrm_state *x = NULL;
464 xfrm_address_t *daddr;
465 u32 mark = skb->mark;
466 unsigned int family = AF_UNSPEC;
467 int decaps = 0;
468 int async = 0;
469 bool xfrm_gro = false;
470 bool crypto_done = false;
471 struct xfrm_offload *xo = xfrm_offload(skb);
472 struct sec_path *sp;
473
474 if (encap_type < 0) {
475 x = xfrm_input_state(skb);
476
477 if (unlikely(x->km.state != XFRM_STATE_VALID)) {
478 if (x->km.state == XFRM_STATE_ACQ)
479 XFRM_INC_STATS(net, LINUX_MIB_XFRMACQUIREERROR);
480 else
481 XFRM_INC_STATS(net,
482 LINUX_MIB_XFRMINSTATEINVALID);
483
484 if (encap_type == -1)
485 dev_put(skb->dev);
486 goto drop;
487 }
488
489 family = x->outer_mode.family;
490
491 /* An encap_type of -1 indicates async resumption. */
492 if (encap_type == -1) {
493 async = 1;
494 seq = XFRM_SKB_CB(skb)->seq.input.low;
495 goto resume;
496 }
497
498 /* encap_type < -1 indicates a GRO call. */
499 encap_type = 0;
500 seq = XFRM_SPI_SKB_CB(skb)->seq;
501
502 if (xo && (xo->flags & CRYPTO_DONE)) {
503 crypto_done = true;
504 family = XFRM_SPI_SKB_CB(skb)->family;
505
506 if (!(xo->status & CRYPTO_SUCCESS)) {
507 if (xo->status &
508 (CRYPTO_TRANSPORT_AH_AUTH_FAILED |
509 CRYPTO_TRANSPORT_ESP_AUTH_FAILED |
510 CRYPTO_TUNNEL_AH_AUTH_FAILED |
511 CRYPTO_TUNNEL_ESP_AUTH_FAILED)) {
512
513 xfrm_audit_state_icvfail(x, skb,
514 x->type->proto);
515 x->stats.integrity_failed++;
516 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEPROTOERROR);
517 goto drop;
518 }
519
520 if (xo->status & CRYPTO_INVALID_PROTOCOL) {
521 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEPROTOERROR);
522 goto drop;
523 }
524
525 XFRM_INC_STATS(net, LINUX_MIB_XFRMINBUFFERERROR);
526 goto drop;
527 }
528
529 if ((err = xfrm_parse_spi(skb, nexthdr, &spi, &seq)) != 0) {
530 XFRM_INC_STATS(net, LINUX_MIB_XFRMINHDRERROR);
531 goto drop;
532 }
533 }
534
535 goto lock;
536 }
537
538 family = XFRM_SPI_SKB_CB(skb)->family;
539
540 /* if tunnel is present override skb->mark value with tunnel i_key */
541 switch (family) {
542 case AF_INET:
543 if (XFRM_TUNNEL_SKB_CB(skb)->tunnel.ip4)
544 mark = be32_to_cpu(XFRM_TUNNEL_SKB_CB(skb)->tunnel.ip4->parms.i_key);
545 break;
546 case AF_INET6:
547 if (XFRM_TUNNEL_SKB_CB(skb)->tunnel.ip6)
548 mark = be32_to_cpu(XFRM_TUNNEL_SKB_CB(skb)->tunnel.ip6->parms.i_key);
549 break;
550 }
551
552 sp = secpath_set(skb);
553 if (!sp) {
554 XFRM_INC_STATS(net, LINUX_MIB_XFRMINERROR);
555 goto drop;
556 }
557
558 seq = 0;
559 if (!spi && (err = xfrm_parse_spi(skb, nexthdr, &spi, &seq)) != 0) {
560 secpath_reset(skb);
561 XFRM_INC_STATS(net, LINUX_MIB_XFRMINHDRERROR);
562 goto drop;
563 }
564
565 daddr = (xfrm_address_t *)(skb_network_header(skb) +
566 XFRM_SPI_SKB_CB(skb)->daddroff);
567 do {
568 sp = skb_sec_path(skb);
569
570 if (sp->len == XFRM_MAX_DEPTH) {
571 secpath_reset(skb);
572 XFRM_INC_STATS(net, LINUX_MIB_XFRMINBUFFERERROR);
573 goto drop;
574 }
575
576 x = xfrm_state_lookup(net, mark, daddr, spi, nexthdr, family);
577 if (x == NULL) {
578 secpath_reset(skb);
579 XFRM_INC_STATS(net, LINUX_MIB_XFRMINNOSTATES);
580 xfrm_audit_state_notfound(skb, family, spi, seq);
581 goto drop;
582 }
583
584 skb->mark = xfrm_smark_get(skb->mark, x);
585
586 sp->xvec[sp->len++] = x;
587
588 skb_dst_force(skb);
589 if (!skb_dst(skb)) {
590 XFRM_INC_STATS(net, LINUX_MIB_XFRMINERROR);
591 goto drop;
592 }
593
594lock:
595 spin_lock(&x->lock);
596
597 if (unlikely(x->km.state != XFRM_STATE_VALID)) {
598 if (x->km.state == XFRM_STATE_ACQ)
599 XFRM_INC_STATS(net, LINUX_MIB_XFRMACQUIREERROR);
600 else
601 XFRM_INC_STATS(net,
602 LINUX_MIB_XFRMINSTATEINVALID);
603 goto drop_unlock;
604 }
605
606 if ((x->encap ? x->encap->encap_type : 0) != encap_type) {
607 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMISMATCH);
608 goto drop_unlock;
609 }
610
611 if (x->repl->check(x, skb, seq)) {
612 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATESEQERROR);
613 goto drop_unlock;
614 }
615
616 if (xfrm_state_check_expire(x)) {
617 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEEXPIRED);
618 goto drop_unlock;
619 }
620
621 spin_unlock(&x->lock);
622
623 if (xfrm_tunnel_check(skb, x, family)) {
624 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMODEERROR);
625 goto drop;
626 }
627
628 seq_hi = htonl(xfrm_replay_seqhi(x, seq));
629
630 XFRM_SKB_CB(skb)->seq.input.low = seq;
631 XFRM_SKB_CB(skb)->seq.input.hi = seq_hi;
632
633 dev_hold(skb->dev);
634
635 if (crypto_done)
636 nexthdr = x->type_offload->input_tail(x, skb);
637 else
638 nexthdr = x->type->input(x, skb);
639
640 if (nexthdr == -EINPROGRESS)
641 return 0;
642resume:
643 dev_put(skb->dev);
644
645 spin_lock(&x->lock);
646 if (nexthdr <= 0) {
647 if (nexthdr == -EBADMSG) {
648 xfrm_audit_state_icvfail(x, skb,
649 x->type->proto);
650 x->stats.integrity_failed++;
651 }
652 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEPROTOERROR);
653 goto drop_unlock;
654 }
655
656 /* only the first xfrm gets the encap type */
657 encap_type = 0;
658
659 if (async && x->repl->recheck(x, skb, seq)) {
660 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATESEQERROR);
661 goto drop_unlock;
662 }
663
664 x->repl->advance(x, seq);
665
666 x->curlft.bytes += skb->len;
667 x->curlft.packets++;
668
669 spin_unlock(&x->lock);
670
671 XFRM_MODE_SKB_CB(skb)->protocol = nexthdr;
672
673 inner_mode = &x->inner_mode;
674
675 if (x->sel.family == AF_UNSPEC) {
676 inner_mode = xfrm_ip2inner_mode(x, XFRM_MODE_SKB_CB(skb)->protocol);
677 if (inner_mode == NULL) {
678 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMODEERROR);
679 goto drop;
680 }
681 }
682
683 if (xfrm_inner_mode_input(x, inner_mode, skb)) {
684 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMODEERROR);
685 goto drop;
686 }
687
688 if (x->outer_mode.flags & XFRM_MODE_FLAG_TUNNEL) {
689 decaps = 1;
690 break;
691 }
692
693 /*
694 * We need the inner address. However, we only get here for
695 * transport mode so the outer address is identical.
696 */
697 daddr = &x->id.daddr;
698 family = x->outer_mode.family;
699
700 err = xfrm_parse_spi(skb, nexthdr, &spi, &seq);
701 if (err < 0) {
702 XFRM_INC_STATS(net, LINUX_MIB_XFRMINHDRERROR);
703 goto drop;
704 }
705 crypto_done = false;
706 } while (!err);
707
708 err = xfrm_rcv_cb(skb, family, x->type->proto, 0);
709 if (err)
710 goto drop;
711
712 nf_reset_ct(skb);
713
714 if (decaps) {
715 sp = skb_sec_path(skb);
716 if (sp)
717 sp->olen = 0;
718 skb_dst_drop(skb);
719 gro_cells_receive(&gro_cells, skb);
720 return 0;
721 } else {
722 xo = xfrm_offload(skb);
723 if (xo)
724 xfrm_gro = xo->flags & XFRM_GRO;
725
726 err = -EAFNOSUPPORT;
727 rcu_read_lock();
728 afinfo = xfrm_state_afinfo_get_rcu(x->inner_mode.family);
729 if (likely(afinfo))
730 err = afinfo->transport_finish(skb, xfrm_gro || async);
731 rcu_read_unlock();
732 if (xfrm_gro) {
733 sp = skb_sec_path(skb);
734 if (sp)
735 sp->olen = 0;
736 skb_dst_drop(skb);
737 gro_cells_receive(&gro_cells, skb);
738 return err;
739 }
740
741 return err;
742 }
743
744drop_unlock:
745 spin_unlock(&x->lock);
746drop:
747 xfrm_rcv_cb(skb, family, x && x->type ? x->type->proto : nexthdr, -1);
748 kfree_skb(skb);
749 return 0;
750}
751EXPORT_SYMBOL(xfrm_input);
752
753int xfrm_input_resume(struct sk_buff *skb, int nexthdr)
754{
755 return xfrm_input(skb, nexthdr, 0, -1);
756}
757EXPORT_SYMBOL(xfrm_input_resume);
758
759static void xfrm_trans_reinject(unsigned long data)
760{
761 struct xfrm_trans_tasklet *trans = (void *)data;
762 struct sk_buff_head queue;
763 struct sk_buff *skb;
764
765 __skb_queue_head_init(&queue);
766 skb_queue_splice_init(&trans->queue, &queue);
767
768 while ((skb = __skb_dequeue(&queue)))
769 XFRM_TRANS_SKB_CB(skb)->finish(dev_net(skb->dev), NULL, skb);
770}
771
772int xfrm_trans_queue(struct sk_buff *skb,
773 int (*finish)(struct net *, struct sock *,
774 struct sk_buff *))
775{
776 struct xfrm_trans_tasklet *trans;
777
778 trans = this_cpu_ptr(&xfrm_trans_tasklet);
779
780 if (skb_queue_len(&trans->queue) >= netdev_max_backlog)
781 return -ENOBUFS;
782
783 XFRM_TRANS_SKB_CB(skb)->finish = finish;
784 __skb_queue_tail(&trans->queue, skb);
785 tasklet_schedule(&trans->tasklet);
786 return 0;
787}
788EXPORT_SYMBOL(xfrm_trans_queue);
789
790void __init xfrm_input_init(void)
791{
792 int err;
793 int i;
794
795 init_dummy_netdev(&xfrm_napi_dev);
796 err = gro_cells_init(&gro_cells, &xfrm_napi_dev);
797 if (err)
798 gro_cells.cells = NULL;
799
800 for_each_possible_cpu(i) {
801 struct xfrm_trans_tasklet *trans;
802
803 trans = &per_cpu(xfrm_trans_tasklet, i);
804 __skb_queue_head_init(&trans->queue);
805 tasklet_init(&trans->tasklet, xfrm_trans_reinject,
806 (unsigned long)trans);
807 }
808}
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
27struct xfrm_trans_tasklet {
28 struct work_struct work;
29 spinlock_t queue_lock;
30 struct sk_buff_head queue;
31};
32
33struct 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
46static DEFINE_SPINLOCK(xfrm_input_afinfo_lock);
47static struct xfrm_input_afinfo const __rcu *xfrm_input_afinfo[2][AF_INET6 + 1];
48
49static struct gro_cells gro_cells;
50static struct net_device xfrm_napi_dev;
51
52static DEFINE_PER_CPU(struct xfrm_trans_tasklet, xfrm_trans_tasklet);
53
54int 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}
69EXPORT_SYMBOL(xfrm_input_register_afinfo);
70
71int 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}
86EXPORT_SYMBOL(xfrm_input_unregister_afinfo);
87
88static 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
102static 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
119struct 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}
138EXPORT_SYMBOL(secpath_set);
139
140/* Fetch spi and seq from ipsec header */
141
142int 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}
175EXPORT_SYMBOL(xfrm_parse_spi);
176
177static 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;
221out:
222 return err;
223}
224
225static 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
233static 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
258out:
259 return err;
260}
261
262static 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
270static 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
295out:
296 return err;
297}
298
299static 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;
322out:
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 */
336static int
337xfrm_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
364static 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 */
389static 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
403static 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
423static 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
447int 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 || (xo && xo->flags & XFRM_GRO)) {
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 /* GRO call */
489 seq = XFRM_SPI_SKB_CB(skb)->seq;
490
491 if (xo && (xo->flags & CRYPTO_DONE)) {
492 crypto_done = true;
493 family = XFRM_SPI_SKB_CB(skb)->family;
494
495 if (!(xo->status & CRYPTO_SUCCESS)) {
496 if (xo->status &
497 (CRYPTO_TRANSPORT_AH_AUTH_FAILED |
498 CRYPTO_TRANSPORT_ESP_AUTH_FAILED |
499 CRYPTO_TUNNEL_AH_AUTH_FAILED |
500 CRYPTO_TUNNEL_ESP_AUTH_FAILED)) {
501
502 xfrm_audit_state_icvfail(x, skb,
503 x->type->proto);
504 x->stats.integrity_failed++;
505 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEPROTOERROR);
506 goto drop;
507 }
508
509 if (xo->status & CRYPTO_INVALID_PROTOCOL) {
510 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEPROTOERROR);
511 goto drop;
512 }
513
514 XFRM_INC_STATS(net, LINUX_MIB_XFRMINBUFFERERROR);
515 goto drop;
516 }
517
518 if (xfrm_parse_spi(skb, nexthdr, &spi, &seq)) {
519 XFRM_INC_STATS(net, LINUX_MIB_XFRMINHDRERROR);
520 goto drop;
521 }
522 }
523
524 goto lock;
525 }
526
527 family = XFRM_SPI_SKB_CB(skb)->family;
528
529 /* if tunnel is present override skb->mark value with tunnel i_key */
530 switch (family) {
531 case AF_INET:
532 if (XFRM_TUNNEL_SKB_CB(skb)->tunnel.ip4)
533 mark = be32_to_cpu(XFRM_TUNNEL_SKB_CB(skb)->tunnel.ip4->parms.i_key);
534 break;
535 case AF_INET6:
536 if (XFRM_TUNNEL_SKB_CB(skb)->tunnel.ip6)
537 mark = be32_to_cpu(XFRM_TUNNEL_SKB_CB(skb)->tunnel.ip6->parms.i_key);
538 break;
539 }
540
541 sp = secpath_set(skb);
542 if (!sp) {
543 XFRM_INC_STATS(net, LINUX_MIB_XFRMINERROR);
544 goto drop;
545 }
546
547 seq = 0;
548 if (!spi && xfrm_parse_spi(skb, nexthdr, &spi, &seq)) {
549 secpath_reset(skb);
550 XFRM_INC_STATS(net, LINUX_MIB_XFRMINHDRERROR);
551 goto drop;
552 }
553
554 daddr = (xfrm_address_t *)(skb_network_header(skb) +
555 XFRM_SPI_SKB_CB(skb)->daddroff);
556 do {
557 sp = skb_sec_path(skb);
558
559 if (sp->len == XFRM_MAX_DEPTH) {
560 secpath_reset(skb);
561 XFRM_INC_STATS(net, LINUX_MIB_XFRMINBUFFERERROR);
562 goto drop;
563 }
564
565 x = xfrm_state_lookup(net, mark, daddr, spi, nexthdr, family);
566 if (x == NULL) {
567 secpath_reset(skb);
568 XFRM_INC_STATS(net, LINUX_MIB_XFRMINNOSTATES);
569 xfrm_audit_state_notfound(skb, family, spi, seq);
570 goto drop;
571 }
572
573 skb->mark = xfrm_smark_get(skb->mark, x);
574
575 sp->xvec[sp->len++] = x;
576
577 skb_dst_force(skb);
578 if (!skb_dst(skb)) {
579 XFRM_INC_STATS(net, LINUX_MIB_XFRMINERROR);
580 goto drop;
581 }
582
583lock:
584 spin_lock(&x->lock);
585
586 if (unlikely(x->km.state != XFRM_STATE_VALID)) {
587 if (x->km.state == XFRM_STATE_ACQ)
588 XFRM_INC_STATS(net, LINUX_MIB_XFRMACQUIREERROR);
589 else
590 XFRM_INC_STATS(net,
591 LINUX_MIB_XFRMINSTATEINVALID);
592 goto drop_unlock;
593 }
594
595 if ((x->encap ? x->encap->encap_type : 0) != encap_type) {
596 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMISMATCH);
597 goto drop_unlock;
598 }
599
600 if (xfrm_replay_check(x, skb, seq)) {
601 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATESEQERROR);
602 goto drop_unlock;
603 }
604
605 if (xfrm_state_check_expire(x)) {
606 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEEXPIRED);
607 goto drop_unlock;
608 }
609
610 spin_unlock(&x->lock);
611
612 if (xfrm_tunnel_check(skb, x, family)) {
613 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMODEERROR);
614 goto drop;
615 }
616
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 dev_hold(skb->dev);
623
624 if (crypto_done)
625 nexthdr = x->type_offload->input_tail(x, skb);
626 else
627 nexthdr = x->type->input(x, skb);
628
629 if (nexthdr == -EINPROGRESS)
630 return 0;
631resume:
632 dev_put(skb->dev);
633
634 spin_lock(&x->lock);
635 if (nexthdr < 0) {
636 if (nexthdr == -EBADMSG) {
637 xfrm_audit_state_icvfail(x, skb,
638 x->type->proto);
639 x->stats.integrity_failed++;
640 }
641 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEPROTOERROR);
642 goto drop_unlock;
643 }
644
645 /* only the first xfrm gets the encap type */
646 encap_type = 0;
647
648 if (xfrm_replay_recheck(x, skb, seq)) {
649 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATESEQERROR);
650 goto drop_unlock;
651 }
652
653 xfrm_replay_advance(x, seq);
654
655 x->curlft.bytes += skb->len;
656 x->curlft.packets++;
657 x->lastused = ktime_get_real_seconds();
658
659 spin_unlock(&x->lock);
660
661 XFRM_MODE_SKB_CB(skb)->protocol = nexthdr;
662
663 if (xfrm_inner_mode_input(x, skb)) {
664 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMODEERROR);
665 goto drop;
666 }
667
668 if (x->outer_mode.flags & XFRM_MODE_FLAG_TUNNEL) {
669 decaps = 1;
670 break;
671 }
672
673 /*
674 * We need the inner address. However, we only get here for
675 * transport mode so the outer address is identical.
676 */
677 daddr = &x->id.daddr;
678 family = x->props.family;
679
680 err = xfrm_parse_spi(skb, nexthdr, &spi, &seq);
681 if (err < 0) {
682 XFRM_INC_STATS(net, LINUX_MIB_XFRMINHDRERROR);
683 goto drop;
684 }
685 crypto_done = false;
686 } while (!err);
687
688 err = xfrm_rcv_cb(skb, family, x->type->proto, 0);
689 if (err)
690 goto drop;
691
692 nf_reset_ct(skb);
693
694 if (decaps) {
695 sp = skb_sec_path(skb);
696 if (sp)
697 sp->olen = 0;
698 if (skb_valid_dst(skb))
699 skb_dst_drop(skb);
700 gro_cells_receive(&gro_cells, skb);
701 return 0;
702 } else {
703 xo = xfrm_offload(skb);
704 if (xo)
705 xfrm_gro = xo->flags & XFRM_GRO;
706
707 err = -EAFNOSUPPORT;
708 rcu_read_lock();
709 afinfo = xfrm_state_afinfo_get_rcu(x->props.family);
710 if (likely(afinfo))
711 err = afinfo->transport_finish(skb, xfrm_gro || async);
712 rcu_read_unlock();
713 if (xfrm_gro) {
714 sp = skb_sec_path(skb);
715 if (sp)
716 sp->olen = 0;
717 if (skb_valid_dst(skb))
718 skb_dst_drop(skb);
719 gro_cells_receive(&gro_cells, skb);
720 return err;
721 }
722
723 return err;
724 }
725
726drop_unlock:
727 spin_unlock(&x->lock);
728drop:
729 xfrm_rcv_cb(skb, family, x && x->type ? x->type->proto : nexthdr, -1);
730 kfree_skb(skb);
731 return 0;
732}
733EXPORT_SYMBOL(xfrm_input);
734
735int xfrm_input_resume(struct sk_buff *skb, int nexthdr)
736{
737 return xfrm_input(skb, nexthdr, 0, -1);
738}
739EXPORT_SYMBOL(xfrm_input_resume);
740
741static void xfrm_trans_reinject(struct work_struct *work)
742{
743 struct xfrm_trans_tasklet *trans = container_of(work, struct xfrm_trans_tasklet, work);
744 struct sk_buff_head queue;
745 struct sk_buff *skb;
746
747 __skb_queue_head_init(&queue);
748 spin_lock_bh(&trans->queue_lock);
749 skb_queue_splice_init(&trans->queue, &queue);
750 spin_unlock_bh(&trans->queue_lock);
751
752 local_bh_disable();
753 while ((skb = __skb_dequeue(&queue)))
754 XFRM_TRANS_SKB_CB(skb)->finish(XFRM_TRANS_SKB_CB(skb)->net,
755 NULL, skb);
756 local_bh_enable();
757}
758
759int xfrm_trans_queue_net(struct net *net, struct sk_buff *skb,
760 int (*finish)(struct net *, struct sock *,
761 struct sk_buff *))
762{
763 struct xfrm_trans_tasklet *trans;
764
765 trans = this_cpu_ptr(&xfrm_trans_tasklet);
766
767 if (skb_queue_len(&trans->queue) >= READ_ONCE(netdev_max_backlog))
768 return -ENOBUFS;
769
770 BUILD_BUG_ON(sizeof(struct xfrm_trans_cb) > sizeof(skb->cb));
771
772 XFRM_TRANS_SKB_CB(skb)->finish = finish;
773 XFRM_TRANS_SKB_CB(skb)->net = net;
774 spin_lock_bh(&trans->queue_lock);
775 __skb_queue_tail(&trans->queue, skb);
776 spin_unlock_bh(&trans->queue_lock);
777 schedule_work(&trans->work);
778 return 0;
779}
780EXPORT_SYMBOL(xfrm_trans_queue_net);
781
782int xfrm_trans_queue(struct sk_buff *skb,
783 int (*finish)(struct net *, struct sock *,
784 struct sk_buff *))
785{
786 return xfrm_trans_queue_net(dev_net(skb->dev), skb, finish);
787}
788EXPORT_SYMBOL(xfrm_trans_queue);
789
790void __init xfrm_input_init(void)
791{
792 int err;
793 int i;
794
795 init_dummy_netdev(&xfrm_napi_dev);
796 err = gro_cells_init(&gro_cells, &xfrm_napi_dev);
797 if (err)
798 gro_cells.cells = NULL;
799
800 for_each_possible_cpu(i) {
801 struct xfrm_trans_tasklet *trans;
802
803 trans = &per_cpu(xfrm_trans_tasklet, i);
804 spin_lock_init(&trans->queue_lock);
805 __skb_queue_head_init(&trans->queue);
806 INIT_WORK(&trans->work, xfrm_trans_reinject);
807 }
808}