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#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
135 return sp;
136}
137EXPORT_SYMBOL(secpath_set);
138
139/* Fetch spi and seq from ipsec header */
140
141int xfrm_parse_spi(struct sk_buff *skb, u8 nexthdr, __be32 *spi, __be32 *seq)
142{
143 int offset, offset_seq;
144 int hlen;
145
146 switch (nexthdr) {
147 case IPPROTO_AH:
148 hlen = sizeof(struct ip_auth_hdr);
149 offset = offsetof(struct ip_auth_hdr, spi);
150 offset_seq = offsetof(struct ip_auth_hdr, seq_no);
151 break;
152 case IPPROTO_ESP:
153 hlen = sizeof(struct ip_esp_hdr);
154 offset = offsetof(struct ip_esp_hdr, spi);
155 offset_seq = offsetof(struct ip_esp_hdr, seq_no);
156 break;
157 case IPPROTO_COMP:
158 if (!pskb_may_pull(skb, sizeof(struct ip_comp_hdr)))
159 return -EINVAL;
160 *spi = htonl(ntohs(*(__be16 *)(skb_transport_header(skb) + 2)));
161 *seq = 0;
162 return 0;
163 default:
164 return 1;
165 }
166
167 if (!pskb_may_pull(skb, hlen))
168 return -EINVAL;
169
170 *spi = *(__be32 *)(skb_transport_header(skb) + offset);
171 *seq = *(__be32 *)(skb_transport_header(skb) + offset_seq);
172 return 0;
173}
174EXPORT_SYMBOL(xfrm_parse_spi);
175
176static int xfrm4_remove_beet_encap(struct xfrm_state *x, struct sk_buff *skb)
177{
178 struct iphdr *iph;
179 int optlen = 0;
180 int err = -EINVAL;
181
182 if (unlikely(XFRM_MODE_SKB_CB(skb)->protocol == IPPROTO_BEETPH)) {
183 struct ip_beet_phdr *ph;
184 int phlen;
185
186 if (!pskb_may_pull(skb, sizeof(*ph)))
187 goto out;
188
189 ph = (struct ip_beet_phdr *)skb->data;
190
191 phlen = sizeof(*ph) + ph->padlen;
192 optlen = ph->hdrlen * 8 + (IPV4_BEET_PHMAXLEN - phlen);
193 if (optlen < 0 || optlen & 3 || optlen > 250)
194 goto out;
195
196 XFRM_MODE_SKB_CB(skb)->protocol = ph->nexthdr;
197
198 if (!pskb_may_pull(skb, phlen))
199 goto out;
200 __skb_pull(skb, phlen);
201 }
202
203 skb_push(skb, sizeof(*iph));
204 skb_reset_network_header(skb);
205 skb_mac_header_rebuild(skb);
206
207 xfrm4_beet_make_header(skb);
208
209 iph = ip_hdr(skb);
210
211 iph->ihl += optlen / 4;
212 iph->tot_len = htons(skb->len);
213 iph->daddr = x->sel.daddr.a4;
214 iph->saddr = x->sel.saddr.a4;
215 iph->check = 0;
216 iph->check = ip_fast_csum(skb_network_header(skb), iph->ihl);
217 err = 0;
218out:
219 return err;
220}
221
222static void ipip_ecn_decapsulate(struct sk_buff *skb)
223{
224 struct iphdr *inner_iph = ipip_hdr(skb);
225
226 if (INET_ECN_is_ce(XFRM_MODE_SKB_CB(skb)->tos))
227 IP_ECN_set_ce(inner_iph);
228}
229
230static int xfrm4_remove_tunnel_encap(struct xfrm_state *x, struct sk_buff *skb)
231{
232 int err = -EINVAL;
233
234 if (XFRM_MODE_SKB_CB(skb)->protocol != IPPROTO_IPIP)
235 goto out;
236
237 if (!pskb_may_pull(skb, sizeof(struct iphdr)))
238 goto out;
239
240 err = skb_unclone(skb, GFP_ATOMIC);
241 if (err)
242 goto out;
243
244 if (x->props.flags & XFRM_STATE_DECAP_DSCP)
245 ipv4_copy_dscp(XFRM_MODE_SKB_CB(skb)->tos, ipip_hdr(skb));
246 if (!(x->props.flags & XFRM_STATE_NOECN))
247 ipip_ecn_decapsulate(skb);
248
249 skb_reset_network_header(skb);
250 skb_mac_header_rebuild(skb);
251 if (skb->mac_len)
252 eth_hdr(skb)->h_proto = skb->protocol;
253
254 err = 0;
255
256out:
257 return err;
258}
259
260static void ipip6_ecn_decapsulate(struct sk_buff *skb)
261{
262 struct ipv6hdr *inner_iph = ipipv6_hdr(skb);
263
264 if (INET_ECN_is_ce(XFRM_MODE_SKB_CB(skb)->tos))
265 IP6_ECN_set_ce(skb, inner_iph);
266}
267
268static int xfrm6_remove_tunnel_encap(struct xfrm_state *x, struct sk_buff *skb)
269{
270 int err = -EINVAL;
271
272 if (XFRM_MODE_SKB_CB(skb)->protocol != IPPROTO_IPV6)
273 goto out;
274 if (!pskb_may_pull(skb, sizeof(struct ipv6hdr)))
275 goto out;
276
277 err = skb_unclone(skb, GFP_ATOMIC);
278 if (err)
279 goto out;
280
281 if (x->props.flags & XFRM_STATE_DECAP_DSCP)
282 ipv6_copy_dscp(XFRM_MODE_SKB_CB(skb)->tos, ipipv6_hdr(skb));
283 if (!(x->props.flags & XFRM_STATE_NOECN))
284 ipip6_ecn_decapsulate(skb);
285
286 skb_reset_network_header(skb);
287 skb_mac_header_rebuild(skb);
288 if (skb->mac_len)
289 eth_hdr(skb)->h_proto = skb->protocol;
290
291 err = 0;
292
293out:
294 return err;
295}
296
297static int xfrm6_remove_beet_encap(struct xfrm_state *x, struct sk_buff *skb)
298{
299 struct ipv6hdr *ip6h;
300 int size = sizeof(struct ipv6hdr);
301 int err;
302
303 err = skb_cow_head(skb, size + skb->mac_len);
304 if (err)
305 goto out;
306
307 __skb_push(skb, size);
308 skb_reset_network_header(skb);
309 skb_mac_header_rebuild(skb);
310
311 xfrm6_beet_make_header(skb);
312
313 ip6h = ipv6_hdr(skb);
314 ip6h->payload_len = htons(skb->len - size);
315 ip6h->daddr = x->sel.daddr.in6;
316 ip6h->saddr = x->sel.saddr.in6;
317 err = 0;
318out:
319 return err;
320}
321
322/* Remove encapsulation header.
323 *
324 * The IP header will be moved over the top of the encapsulation
325 * header.
326 *
327 * On entry, the transport header shall point to where the IP header
328 * should be and the network header shall be set to where the IP
329 * header currently is. skb->data shall point to the start of the
330 * payload.
331 */
332static int
333xfrm_inner_mode_encap_remove(struct xfrm_state *x,
334 const struct xfrm_mode *inner_mode,
335 struct sk_buff *skb)
336{
337 switch (inner_mode->encap) {
338 case XFRM_MODE_BEET:
339 if (inner_mode->family == AF_INET)
340 return xfrm4_remove_beet_encap(x, skb);
341 if (inner_mode->family == AF_INET6)
342 return xfrm6_remove_beet_encap(x, skb);
343 break;
344 case XFRM_MODE_TUNNEL:
345 if (inner_mode->family == AF_INET)
346 return xfrm4_remove_tunnel_encap(x, skb);
347 if (inner_mode->family == AF_INET6)
348 return xfrm6_remove_tunnel_encap(x, skb);
349 break;
350 }
351
352 WARN_ON_ONCE(1);
353 return -EOPNOTSUPP;
354}
355
356static int xfrm_prepare_input(struct xfrm_state *x, struct sk_buff *skb)
357{
358 const struct xfrm_mode *inner_mode = &x->inner_mode;
359
360 switch (x->outer_mode.family) {
361 case AF_INET:
362 xfrm4_extract_header(skb);
363 break;
364 case AF_INET6:
365 xfrm6_extract_header(skb);
366 break;
367 default:
368 WARN_ON_ONCE(1);
369 return -EAFNOSUPPORT;
370 }
371
372 if (x->sel.family == AF_UNSPEC) {
373 inner_mode = xfrm_ip2inner_mode(x, XFRM_MODE_SKB_CB(skb)->protocol);
374 if (!inner_mode)
375 return -EAFNOSUPPORT;
376 }
377
378 switch (inner_mode->family) {
379 case AF_INET:
380 skb->protocol = htons(ETH_P_IP);
381 break;
382 case AF_INET6:
383 skb->protocol = htons(ETH_P_IPV6);
384 break;
385 default:
386 WARN_ON_ONCE(1);
387 break;
388 }
389
390 return xfrm_inner_mode_encap_remove(x, inner_mode, skb);
391}
392
393/* Remove encapsulation header.
394 *
395 * The IP header will be moved over the top of the encapsulation header.
396 *
397 * On entry, skb_transport_header() shall point to where the IP header
398 * should be and skb_network_header() shall be set to where the IP header
399 * currently is. skb->data shall point to the start of the payload.
400 */
401static int xfrm4_transport_input(struct xfrm_state *x, struct sk_buff *skb)
402{
403 int ihl = skb->data - skb_transport_header(skb);
404
405 if (skb->transport_header != skb->network_header) {
406 memmove(skb_transport_header(skb),
407 skb_network_header(skb), ihl);
408 skb->network_header = skb->transport_header;
409 }
410 ip_hdr(skb)->tot_len = htons(skb->len + ihl);
411 skb_reset_transport_header(skb);
412 return 0;
413}
414
415static int xfrm6_transport_input(struct xfrm_state *x, struct sk_buff *skb)
416{
417#if IS_ENABLED(CONFIG_IPV6)
418 int ihl = skb->data - skb_transport_header(skb);
419
420 if (skb->transport_header != skb->network_header) {
421 memmove(skb_transport_header(skb),
422 skb_network_header(skb), ihl);
423 skb->network_header = skb->transport_header;
424 }
425 ipv6_hdr(skb)->payload_len = htons(skb->len + ihl -
426 sizeof(struct ipv6hdr));
427 skb_reset_transport_header(skb);
428 return 0;
429#else
430 WARN_ON_ONCE(1);
431 return -EAFNOSUPPORT;
432#endif
433}
434
435static int xfrm_inner_mode_input(struct xfrm_state *x,
436 const struct xfrm_mode *inner_mode,
437 struct sk_buff *skb)
438{
439 switch (inner_mode->encap) {
440 case XFRM_MODE_BEET:
441 case XFRM_MODE_TUNNEL:
442 return xfrm_prepare_input(x, skb);
443 case XFRM_MODE_TRANSPORT:
444 if (inner_mode->family == AF_INET)
445 return xfrm4_transport_input(x, skb);
446 if (inner_mode->family == AF_INET6)
447 return xfrm6_transport_input(x, skb);
448 break;
449 case XFRM_MODE_ROUTEOPTIMIZATION:
450 WARN_ON_ONCE(1);
451 break;
452 default:
453 WARN_ON_ONCE(1);
454 break;
455 }
456
457 return -EOPNOTSUPP;
458}
459
460int xfrm_input(struct sk_buff *skb, int nexthdr, __be32 spi, int encap_type)
461{
462 const struct xfrm_state_afinfo *afinfo;
463 struct net *net = dev_net(skb->dev);
464 const struct xfrm_mode *inner_mode;
465 int err;
466 __be32 seq;
467 __be32 seq_hi;
468 struct xfrm_state *x = NULL;
469 xfrm_address_t *daddr;
470 u32 mark = skb->mark;
471 unsigned int family = AF_UNSPEC;
472 int decaps = 0;
473 int async = 0;
474 bool xfrm_gro = false;
475 bool crypto_done = false;
476 struct xfrm_offload *xo = xfrm_offload(skb);
477 struct sec_path *sp;
478
479 if (encap_type < 0) {
480 x = xfrm_input_state(skb);
481
482 if (unlikely(x->km.state != XFRM_STATE_VALID)) {
483 if (x->km.state == XFRM_STATE_ACQ)
484 XFRM_INC_STATS(net, LINUX_MIB_XFRMACQUIREERROR);
485 else
486 XFRM_INC_STATS(net,
487 LINUX_MIB_XFRMINSTATEINVALID);
488
489 if (encap_type == -1)
490 dev_put(skb->dev);
491 goto drop;
492 }
493
494 family = x->outer_mode.family;
495
496 /* An encap_type of -1 indicates async resumption. */
497 if (encap_type == -1) {
498 async = 1;
499 seq = XFRM_SKB_CB(skb)->seq.input.low;
500 goto resume;
501 }
502
503 /* encap_type < -1 indicates a GRO call. */
504 encap_type = 0;
505 seq = XFRM_SPI_SKB_CB(skb)->seq;
506
507 if (xo && (xo->flags & CRYPTO_DONE)) {
508 crypto_done = true;
509 family = XFRM_SPI_SKB_CB(skb)->family;
510
511 if (!(xo->status & CRYPTO_SUCCESS)) {
512 if (xo->status &
513 (CRYPTO_TRANSPORT_AH_AUTH_FAILED |
514 CRYPTO_TRANSPORT_ESP_AUTH_FAILED |
515 CRYPTO_TUNNEL_AH_AUTH_FAILED |
516 CRYPTO_TUNNEL_ESP_AUTH_FAILED)) {
517
518 xfrm_audit_state_icvfail(x, skb,
519 x->type->proto);
520 x->stats.integrity_failed++;
521 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEPROTOERROR);
522 goto drop;
523 }
524
525 if (xo->status & CRYPTO_INVALID_PROTOCOL) {
526 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEPROTOERROR);
527 goto drop;
528 }
529
530 XFRM_INC_STATS(net, LINUX_MIB_XFRMINBUFFERERROR);
531 goto drop;
532 }
533
534 if (xfrm_parse_spi(skb, nexthdr, &spi, &seq)) {
535 XFRM_INC_STATS(net, LINUX_MIB_XFRMINHDRERROR);
536 goto drop;
537 }
538 }
539
540 goto lock;
541 }
542
543 family = XFRM_SPI_SKB_CB(skb)->family;
544
545 /* if tunnel is present override skb->mark value with tunnel i_key */
546 switch (family) {
547 case AF_INET:
548 if (XFRM_TUNNEL_SKB_CB(skb)->tunnel.ip4)
549 mark = be32_to_cpu(XFRM_TUNNEL_SKB_CB(skb)->tunnel.ip4->parms.i_key);
550 break;
551 case AF_INET6:
552 if (XFRM_TUNNEL_SKB_CB(skb)->tunnel.ip6)
553 mark = be32_to_cpu(XFRM_TUNNEL_SKB_CB(skb)->tunnel.ip6->parms.i_key);
554 break;
555 }
556
557 sp = secpath_set(skb);
558 if (!sp) {
559 XFRM_INC_STATS(net, LINUX_MIB_XFRMINERROR);
560 goto drop;
561 }
562
563 seq = 0;
564 if (!spi && xfrm_parse_spi(skb, nexthdr, &spi, &seq)) {
565 secpath_reset(skb);
566 XFRM_INC_STATS(net, LINUX_MIB_XFRMINHDRERROR);
567 goto drop;
568 }
569
570 daddr = (xfrm_address_t *)(skb_network_header(skb) +
571 XFRM_SPI_SKB_CB(skb)->daddroff);
572 do {
573 sp = skb_sec_path(skb);
574
575 if (sp->len == XFRM_MAX_DEPTH) {
576 secpath_reset(skb);
577 XFRM_INC_STATS(net, LINUX_MIB_XFRMINBUFFERERROR);
578 goto drop;
579 }
580
581 x = xfrm_state_lookup(net, mark, daddr, spi, nexthdr, family);
582 if (x == NULL) {
583 secpath_reset(skb);
584 XFRM_INC_STATS(net, LINUX_MIB_XFRMINNOSTATES);
585 xfrm_audit_state_notfound(skb, family, spi, seq);
586 goto drop;
587 }
588
589 skb->mark = xfrm_smark_get(skb->mark, x);
590
591 sp->xvec[sp->len++] = x;
592
593 skb_dst_force(skb);
594 if (!skb_dst(skb)) {
595 XFRM_INC_STATS(net, LINUX_MIB_XFRMINERROR);
596 goto drop;
597 }
598
599lock:
600 spin_lock(&x->lock);
601
602 if (unlikely(x->km.state != XFRM_STATE_VALID)) {
603 if (x->km.state == XFRM_STATE_ACQ)
604 XFRM_INC_STATS(net, LINUX_MIB_XFRMACQUIREERROR);
605 else
606 XFRM_INC_STATS(net,
607 LINUX_MIB_XFRMINSTATEINVALID);
608 goto drop_unlock;
609 }
610
611 if ((x->encap ? x->encap->encap_type : 0) != encap_type) {
612 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMISMATCH);
613 goto drop_unlock;
614 }
615
616 if (xfrm_replay_check(x, skb, seq)) {
617 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATESEQERROR);
618 goto drop_unlock;
619 }
620
621 if (xfrm_state_check_expire(x)) {
622 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEEXPIRED);
623 goto drop_unlock;
624 }
625
626 spin_unlock(&x->lock);
627
628 if (xfrm_tunnel_check(skb, x, family)) {
629 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMODEERROR);
630 goto drop;
631 }
632
633 seq_hi = htonl(xfrm_replay_seqhi(x, seq));
634
635 XFRM_SKB_CB(skb)->seq.input.low = seq;
636 XFRM_SKB_CB(skb)->seq.input.hi = seq_hi;
637
638 dev_hold(skb->dev);
639
640 if (crypto_done)
641 nexthdr = x->type_offload->input_tail(x, skb);
642 else
643 nexthdr = x->type->input(x, skb);
644
645 if (nexthdr == -EINPROGRESS)
646 return 0;
647resume:
648 dev_put(skb->dev);
649
650 spin_lock(&x->lock);
651 if (nexthdr < 0) {
652 if (nexthdr == -EBADMSG) {
653 xfrm_audit_state_icvfail(x, skb,
654 x->type->proto);
655 x->stats.integrity_failed++;
656 }
657 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEPROTOERROR);
658 goto drop_unlock;
659 }
660
661 /* only the first xfrm gets the encap type */
662 encap_type = 0;
663
664 if (xfrm_replay_recheck(x, skb, seq)) {
665 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATESEQERROR);
666 goto drop_unlock;
667 }
668
669 xfrm_replay_advance(x, seq);
670
671 x->curlft.bytes += skb->len;
672 x->curlft.packets++;
673 x->lastused = ktime_get_real_seconds();
674
675 spin_unlock(&x->lock);
676
677 XFRM_MODE_SKB_CB(skb)->protocol = nexthdr;
678
679 inner_mode = &x->inner_mode;
680
681 if (x->sel.family == AF_UNSPEC) {
682 inner_mode = xfrm_ip2inner_mode(x, XFRM_MODE_SKB_CB(skb)->protocol);
683 if (inner_mode == NULL) {
684 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMODEERROR);
685 goto drop;
686 }
687 }
688
689 if (xfrm_inner_mode_input(x, inner_mode, skb)) {
690 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMODEERROR);
691 goto drop;
692 }
693
694 if (x->outer_mode.flags & XFRM_MODE_FLAG_TUNNEL) {
695 decaps = 1;
696 break;
697 }
698
699 /*
700 * We need the inner address. However, we only get here for
701 * transport mode so the outer address is identical.
702 */
703 daddr = &x->id.daddr;
704 family = x->outer_mode.family;
705
706 err = xfrm_parse_spi(skb, nexthdr, &spi, &seq);
707 if (err < 0) {
708 XFRM_INC_STATS(net, LINUX_MIB_XFRMINHDRERROR);
709 goto drop;
710 }
711 crypto_done = false;
712 } while (!err);
713
714 err = xfrm_rcv_cb(skb, family, x->type->proto, 0);
715 if (err)
716 goto drop;
717
718 nf_reset_ct(skb);
719
720 if (decaps) {
721 sp = skb_sec_path(skb);
722 if (sp)
723 sp->olen = 0;
724 if (skb_valid_dst(skb))
725 skb_dst_drop(skb);
726 gro_cells_receive(&gro_cells, skb);
727 return 0;
728 } else {
729 xo = xfrm_offload(skb);
730 if (xo)
731 xfrm_gro = xo->flags & XFRM_GRO;
732
733 err = -EAFNOSUPPORT;
734 rcu_read_lock();
735 afinfo = xfrm_state_afinfo_get_rcu(x->inner_mode.family);
736 if (likely(afinfo))
737 err = afinfo->transport_finish(skb, xfrm_gro || async);
738 rcu_read_unlock();
739 if (xfrm_gro) {
740 sp = skb_sec_path(skb);
741 if (sp)
742 sp->olen = 0;
743 if (skb_valid_dst(skb))
744 skb_dst_drop(skb);
745 gro_cells_receive(&gro_cells, skb);
746 return err;
747 }
748
749 return err;
750 }
751
752drop_unlock:
753 spin_unlock(&x->lock);
754drop:
755 xfrm_rcv_cb(skb, family, x && x->type ? x->type->proto : nexthdr, -1);
756 kfree_skb(skb);
757 return 0;
758}
759EXPORT_SYMBOL(xfrm_input);
760
761int xfrm_input_resume(struct sk_buff *skb, int nexthdr)
762{
763 return xfrm_input(skb, nexthdr, 0, -1);
764}
765EXPORT_SYMBOL(xfrm_input_resume);
766
767static void xfrm_trans_reinject(struct work_struct *work)
768{
769 struct xfrm_trans_tasklet *trans = container_of(work, struct xfrm_trans_tasklet, work);
770 struct sk_buff_head queue;
771 struct sk_buff *skb;
772
773 __skb_queue_head_init(&queue);
774 spin_lock_bh(&trans->queue_lock);
775 skb_queue_splice_init(&trans->queue, &queue);
776 spin_unlock_bh(&trans->queue_lock);
777
778 local_bh_disable();
779 while ((skb = __skb_dequeue(&queue)))
780 XFRM_TRANS_SKB_CB(skb)->finish(XFRM_TRANS_SKB_CB(skb)->net,
781 NULL, skb);
782 local_bh_enable();
783}
784
785int xfrm_trans_queue_net(struct net *net, struct sk_buff *skb,
786 int (*finish)(struct net *, struct sock *,
787 struct sk_buff *))
788{
789 struct xfrm_trans_tasklet *trans;
790
791 trans = this_cpu_ptr(&xfrm_trans_tasklet);
792
793 if (skb_queue_len(&trans->queue) >= READ_ONCE(netdev_max_backlog))
794 return -ENOBUFS;
795
796 BUILD_BUG_ON(sizeof(struct xfrm_trans_cb) > sizeof(skb->cb));
797
798 XFRM_TRANS_SKB_CB(skb)->finish = finish;
799 XFRM_TRANS_SKB_CB(skb)->net = net;
800 spin_lock_bh(&trans->queue_lock);
801 __skb_queue_tail(&trans->queue, skb);
802 spin_unlock_bh(&trans->queue_lock);
803 schedule_work(&trans->work);
804 return 0;
805}
806EXPORT_SYMBOL(xfrm_trans_queue_net);
807
808int xfrm_trans_queue(struct sk_buff *skb,
809 int (*finish)(struct net *, struct sock *,
810 struct sk_buff *))
811{
812 return xfrm_trans_queue_net(dev_net(skb->dev), skb, finish);
813}
814EXPORT_SYMBOL(xfrm_trans_queue);
815
816void __init xfrm_input_init(void)
817{
818 int err;
819 int i;
820
821 init_dummy_netdev(&xfrm_napi_dev);
822 err = gro_cells_init(&gro_cells, &xfrm_napi_dev);
823 if (err)
824 gro_cells.cells = NULL;
825
826 for_each_possible_cpu(i) {
827 struct xfrm_trans_tasklet *trans;
828
829 trans = &per_cpu(xfrm_trans_tasklet, i);
830 spin_lock_init(&trans->queue_lock);
831 __skb_queue_head_init(&trans->queue);
832 INIT_WORK(&trans->work, xfrm_trans_reinject);
833 }
834}
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 WARN_ON_ONCE(1);
450 break;
451 }
452
453 return -EOPNOTSUPP;
454}
455
456int xfrm_input(struct sk_buff *skb, int nexthdr, __be32 spi, int encap_type)
457{
458 const struct xfrm_state_afinfo *afinfo;
459 struct net *net = dev_net(skb->dev);
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 || (xo && xo->flags & XFRM_GRO)) {
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->props.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 /* GRO call */
498 seq = XFRM_SPI_SKB_CB(skb)->seq;
499
500 if (xo && (xo->flags & CRYPTO_DONE)) {
501 crypto_done = true;
502 family = XFRM_SPI_SKB_CB(skb)->family;
503
504 if (!(xo->status & CRYPTO_SUCCESS)) {
505 if (xo->status &
506 (CRYPTO_TRANSPORT_AH_AUTH_FAILED |
507 CRYPTO_TRANSPORT_ESP_AUTH_FAILED |
508 CRYPTO_TUNNEL_AH_AUTH_FAILED |
509 CRYPTO_TUNNEL_ESP_AUTH_FAILED)) {
510
511 xfrm_audit_state_icvfail(x, skb,
512 x->type->proto);
513 x->stats.integrity_failed++;
514 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEPROTOERROR);
515 goto drop;
516 }
517
518 if (xo->status & CRYPTO_INVALID_PROTOCOL) {
519 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEPROTOERROR);
520 goto drop;
521 }
522
523 XFRM_INC_STATS(net, LINUX_MIB_XFRMINBUFFERERROR);
524 goto drop;
525 }
526
527 if (xfrm_parse_spi(skb, nexthdr, &spi, &seq)) {
528 XFRM_INC_STATS(net, LINUX_MIB_XFRMINHDRERROR);
529 goto drop;
530 }
531 }
532
533 goto lock;
534 }
535
536 family = XFRM_SPI_SKB_CB(skb)->family;
537
538 /* if tunnel is present override skb->mark value with tunnel i_key */
539 switch (family) {
540 case AF_INET:
541 if (XFRM_TUNNEL_SKB_CB(skb)->tunnel.ip4)
542 mark = be32_to_cpu(XFRM_TUNNEL_SKB_CB(skb)->tunnel.ip4->parms.i_key);
543 break;
544 case AF_INET6:
545 if (XFRM_TUNNEL_SKB_CB(skb)->tunnel.ip6)
546 mark = be32_to_cpu(XFRM_TUNNEL_SKB_CB(skb)->tunnel.ip6->parms.i_key);
547 break;
548 }
549
550 sp = secpath_set(skb);
551 if (!sp) {
552 XFRM_INC_STATS(net, LINUX_MIB_XFRMINERROR);
553 goto drop;
554 }
555
556 seq = 0;
557 if (!spi && xfrm_parse_spi(skb, nexthdr, &spi, &seq)) {
558 secpath_reset(skb);
559 XFRM_INC_STATS(net, LINUX_MIB_XFRMINHDRERROR);
560 goto drop;
561 }
562
563 daddr = (xfrm_address_t *)(skb_network_header(skb) +
564 XFRM_SPI_SKB_CB(skb)->daddroff);
565 do {
566 sp = skb_sec_path(skb);
567
568 if (sp->len == XFRM_MAX_DEPTH) {
569 secpath_reset(skb);
570 XFRM_INC_STATS(net, LINUX_MIB_XFRMINBUFFERERROR);
571 goto drop;
572 }
573
574 x = xfrm_state_lookup(net, mark, daddr, spi, nexthdr, family);
575 if (x == NULL) {
576 secpath_reset(skb);
577 XFRM_INC_STATS(net, LINUX_MIB_XFRMINNOSTATES);
578 xfrm_audit_state_notfound(skb, family, spi, seq);
579 goto drop;
580 }
581
582 skb->mark = xfrm_smark_get(skb->mark, x);
583
584 sp->xvec[sp->len++] = x;
585
586 skb_dst_force(skb);
587 if (!skb_dst(skb)) {
588 XFRM_INC_STATS(net, LINUX_MIB_XFRMINERROR);
589 goto drop;
590 }
591
592lock:
593 spin_lock(&x->lock);
594
595 if (unlikely(x->km.state != XFRM_STATE_VALID)) {
596 if (x->km.state == XFRM_STATE_ACQ)
597 XFRM_INC_STATS(net, LINUX_MIB_XFRMACQUIREERROR);
598 else
599 XFRM_INC_STATS(net,
600 LINUX_MIB_XFRMINSTATEINVALID);
601 goto drop_unlock;
602 }
603
604 if ((x->encap ? x->encap->encap_type : 0) != encap_type) {
605 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMISMATCH);
606 goto drop_unlock;
607 }
608
609 if (xfrm_replay_check(x, skb, seq)) {
610 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATESEQERROR);
611 goto drop_unlock;
612 }
613
614 if (xfrm_state_check_expire(x)) {
615 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEEXPIRED);
616 goto drop_unlock;
617 }
618
619 spin_unlock(&x->lock);
620
621 if (xfrm_tunnel_check(skb, x, family)) {
622 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMODEERROR);
623 goto drop;
624 }
625
626 seq_hi = htonl(xfrm_replay_seqhi(x, seq));
627
628 XFRM_SKB_CB(skb)->seq.input.low = seq;
629 XFRM_SKB_CB(skb)->seq.input.hi = seq_hi;
630
631 dev_hold(skb->dev);
632
633 if (crypto_done)
634 nexthdr = x->type_offload->input_tail(x, skb);
635 else
636 nexthdr = x->type->input(x, skb);
637
638 if (nexthdr == -EINPROGRESS)
639 return 0;
640resume:
641 dev_put(skb->dev);
642
643 spin_lock(&x->lock);
644 if (nexthdr < 0) {
645 if (nexthdr == -EBADMSG) {
646 xfrm_audit_state_icvfail(x, skb,
647 x->type->proto);
648 x->stats.integrity_failed++;
649 }
650 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEPROTOERROR);
651 goto drop_unlock;
652 }
653
654 /* only the first xfrm gets the encap type */
655 encap_type = 0;
656
657 if (xfrm_replay_recheck(x, skb, seq)) {
658 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATESEQERROR);
659 goto drop_unlock;
660 }
661
662 xfrm_replay_advance(x, seq);
663
664 x->curlft.bytes += skb->len;
665 x->curlft.packets++;
666 x->lastused = ktime_get_real_seconds();
667
668 spin_unlock(&x->lock);
669
670 XFRM_MODE_SKB_CB(skb)->protocol = nexthdr;
671
672 if (xfrm_inner_mode_input(x, skb)) {
673 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMODEERROR);
674 goto drop;
675 }
676
677 if (x->outer_mode.flags & XFRM_MODE_FLAG_TUNNEL) {
678 decaps = 1;
679 break;
680 }
681
682 /*
683 * We need the inner address. However, we only get here for
684 * transport mode so the outer address is identical.
685 */
686 daddr = &x->id.daddr;
687 family = x->props.family;
688
689 err = xfrm_parse_spi(skb, nexthdr, &spi, &seq);
690 if (err < 0) {
691 XFRM_INC_STATS(net, LINUX_MIB_XFRMINHDRERROR);
692 goto drop;
693 }
694 crypto_done = false;
695 } while (!err);
696
697 err = xfrm_rcv_cb(skb, family, x->type->proto, 0);
698 if (err)
699 goto drop;
700
701 nf_reset_ct(skb);
702
703 if (decaps) {
704 sp = skb_sec_path(skb);
705 if (sp)
706 sp->olen = 0;
707 if (skb_valid_dst(skb))
708 skb_dst_drop(skb);
709 gro_cells_receive(&gro_cells, skb);
710 return 0;
711 } else {
712 xo = xfrm_offload(skb);
713 if (xo)
714 xfrm_gro = xo->flags & XFRM_GRO;
715
716 err = -EAFNOSUPPORT;
717 rcu_read_lock();
718 afinfo = xfrm_state_afinfo_get_rcu(x->props.family);
719 if (likely(afinfo))
720 err = afinfo->transport_finish(skb, xfrm_gro || async);
721 rcu_read_unlock();
722 if (xfrm_gro) {
723 sp = skb_sec_path(skb);
724 if (sp)
725 sp->olen = 0;
726 if (skb_valid_dst(skb))
727 skb_dst_drop(skb);
728 gro_cells_receive(&gro_cells, skb);
729 return err;
730 }
731
732 return err;
733 }
734
735drop_unlock:
736 spin_unlock(&x->lock);
737drop:
738 xfrm_rcv_cb(skb, family, x && x->type ? x->type->proto : nexthdr, -1);
739 kfree_skb(skb);
740 return 0;
741}
742EXPORT_SYMBOL(xfrm_input);
743
744int xfrm_input_resume(struct sk_buff *skb, int nexthdr)
745{
746 return xfrm_input(skb, nexthdr, 0, -1);
747}
748EXPORT_SYMBOL(xfrm_input_resume);
749
750static void xfrm_trans_reinject(struct work_struct *work)
751{
752 struct xfrm_trans_tasklet *trans = container_of(work, struct xfrm_trans_tasklet, work);
753 struct sk_buff_head queue;
754 struct sk_buff *skb;
755
756 __skb_queue_head_init(&queue);
757 spin_lock_bh(&trans->queue_lock);
758 skb_queue_splice_init(&trans->queue, &queue);
759 spin_unlock_bh(&trans->queue_lock);
760
761 local_bh_disable();
762 while ((skb = __skb_dequeue(&queue)))
763 XFRM_TRANS_SKB_CB(skb)->finish(XFRM_TRANS_SKB_CB(skb)->net,
764 NULL, skb);
765 local_bh_enable();
766}
767
768int xfrm_trans_queue_net(struct net *net, struct sk_buff *skb,
769 int (*finish)(struct net *, struct sock *,
770 struct sk_buff *))
771{
772 struct xfrm_trans_tasklet *trans;
773
774 trans = this_cpu_ptr(&xfrm_trans_tasklet);
775
776 if (skb_queue_len(&trans->queue) >= READ_ONCE(net_hotdata.max_backlog))
777 return -ENOBUFS;
778
779 BUILD_BUG_ON(sizeof(struct xfrm_trans_cb) > sizeof(skb->cb));
780
781 XFRM_TRANS_SKB_CB(skb)->finish = finish;
782 XFRM_TRANS_SKB_CB(skb)->net = net;
783 spin_lock_bh(&trans->queue_lock);
784 __skb_queue_tail(&trans->queue, skb);
785 spin_unlock_bh(&trans->queue_lock);
786 schedule_work(&trans->work);
787 return 0;
788}
789EXPORT_SYMBOL(xfrm_trans_queue_net);
790
791int xfrm_trans_queue(struct sk_buff *skb,
792 int (*finish)(struct net *, struct sock *,
793 struct sk_buff *))
794{
795 return xfrm_trans_queue_net(dev_net(skb->dev), skb, finish);
796}
797EXPORT_SYMBOL(xfrm_trans_queue);
798
799void __init xfrm_input_init(void)
800{
801 int err;
802 int i;
803
804 init_dummy_netdev(&xfrm_napi_dev);
805 err = gro_cells_init(&gro_cells, &xfrm_napi_dev);
806 if (err)
807 gro_cells.cells = NULL;
808
809 for_each_possible_cpu(i) {
810 struct xfrm_trans_tasklet *trans;
811
812 trans = &per_cpu(xfrm_trans_tasklet, i);
813 spin_lock_init(&trans->queue_lock);
814 __skb_queue_head_init(&trans->queue);
815 INIT_WORK(&trans->work, xfrm_trans_reinject);
816 }
817}