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/*
2 * xfrm_input.c
3 *
4 * Changes:
5 * YOSHIFUJI Hideaki @USAGI
6 * Split up af-specific portion
7 *
8 */
9
10#include <linux/slab.h>
11#include <linux/module.h>
12#include <linux/netdevice.h>
13#include <net/dst.h>
14#include <net/ip.h>
15#include <net/xfrm.h>
16#include <net/ip_tunnels.h>
17#include <net/ip6_tunnel.h>
18
19static struct kmem_cache *secpath_cachep __read_mostly;
20
21static DEFINE_SPINLOCK(xfrm_input_afinfo_lock);
22static struct xfrm_input_afinfo __rcu *xfrm_input_afinfo[NPROTO];
23
24int xfrm_input_register_afinfo(struct xfrm_input_afinfo *afinfo)
25{
26 int err = 0;
27
28 if (unlikely(afinfo == NULL))
29 return -EINVAL;
30 if (unlikely(afinfo->family >= NPROTO))
31 return -EAFNOSUPPORT;
32 spin_lock_bh(&xfrm_input_afinfo_lock);
33 if (unlikely(xfrm_input_afinfo[afinfo->family] != NULL))
34 err = -EEXIST;
35 else
36 rcu_assign_pointer(xfrm_input_afinfo[afinfo->family], afinfo);
37 spin_unlock_bh(&xfrm_input_afinfo_lock);
38 return err;
39}
40EXPORT_SYMBOL(xfrm_input_register_afinfo);
41
42int xfrm_input_unregister_afinfo(struct xfrm_input_afinfo *afinfo)
43{
44 int err = 0;
45
46 if (unlikely(afinfo == NULL))
47 return -EINVAL;
48 if (unlikely(afinfo->family >= NPROTO))
49 return -EAFNOSUPPORT;
50 spin_lock_bh(&xfrm_input_afinfo_lock);
51 if (likely(xfrm_input_afinfo[afinfo->family] != NULL)) {
52 if (unlikely(xfrm_input_afinfo[afinfo->family] != afinfo))
53 err = -EINVAL;
54 else
55 RCU_INIT_POINTER(xfrm_input_afinfo[afinfo->family], NULL);
56 }
57 spin_unlock_bh(&xfrm_input_afinfo_lock);
58 synchronize_rcu();
59 return err;
60}
61EXPORT_SYMBOL(xfrm_input_unregister_afinfo);
62
63static struct xfrm_input_afinfo *xfrm_input_get_afinfo(unsigned int family)
64{
65 struct xfrm_input_afinfo *afinfo;
66
67 if (unlikely(family >= NPROTO))
68 return NULL;
69 rcu_read_lock();
70 afinfo = rcu_dereference(xfrm_input_afinfo[family]);
71 if (unlikely(!afinfo))
72 rcu_read_unlock();
73 return afinfo;
74}
75
76static void xfrm_input_put_afinfo(struct xfrm_input_afinfo *afinfo)
77{
78 rcu_read_unlock();
79}
80
81static int xfrm_rcv_cb(struct sk_buff *skb, unsigned int family, u8 protocol,
82 int err)
83{
84 int ret;
85 struct xfrm_input_afinfo *afinfo = xfrm_input_get_afinfo(family);
86
87 if (!afinfo)
88 return -EAFNOSUPPORT;
89
90 ret = afinfo->callback(skb, protocol, err);
91 xfrm_input_put_afinfo(afinfo);
92
93 return ret;
94}
95
96void __secpath_destroy(struct sec_path *sp)
97{
98 int i;
99 for (i = 0; i < sp->len; i++)
100 xfrm_state_put(sp->xvec[i]);
101 kmem_cache_free(secpath_cachep, sp);
102}
103EXPORT_SYMBOL(__secpath_destroy);
104
105struct sec_path *secpath_dup(struct sec_path *src)
106{
107 struct sec_path *sp;
108
109 sp = kmem_cache_alloc(secpath_cachep, GFP_ATOMIC);
110 if (!sp)
111 return NULL;
112
113 sp->len = 0;
114 if (src) {
115 int i;
116
117 memcpy(sp, src, sizeof(*sp));
118 for (i = 0; i < sp->len; i++)
119 xfrm_state_hold(sp->xvec[i]);
120 }
121 atomic_set(&sp->refcnt, 1);
122 return sp;
123}
124EXPORT_SYMBOL(secpath_dup);
125
126/* Fetch spi and seq from ipsec header */
127
128int xfrm_parse_spi(struct sk_buff *skb, u8 nexthdr, __be32 *spi, __be32 *seq)
129{
130 int offset, offset_seq;
131 int hlen;
132
133 switch (nexthdr) {
134 case IPPROTO_AH:
135 hlen = sizeof(struct ip_auth_hdr);
136 offset = offsetof(struct ip_auth_hdr, spi);
137 offset_seq = offsetof(struct ip_auth_hdr, seq_no);
138 break;
139 case IPPROTO_ESP:
140 hlen = sizeof(struct ip_esp_hdr);
141 offset = offsetof(struct ip_esp_hdr, spi);
142 offset_seq = offsetof(struct ip_esp_hdr, seq_no);
143 break;
144 case IPPROTO_COMP:
145 if (!pskb_may_pull(skb, sizeof(struct ip_comp_hdr)))
146 return -EINVAL;
147 *spi = htonl(ntohs(*(__be16 *)(skb_transport_header(skb) + 2)));
148 *seq = 0;
149 return 0;
150 default:
151 return 1;
152 }
153
154 if (!pskb_may_pull(skb, hlen))
155 return -EINVAL;
156
157 *spi = *(__be32 *)(skb_transport_header(skb) + offset);
158 *seq = *(__be32 *)(skb_transport_header(skb) + offset_seq);
159 return 0;
160}
161
162int xfrm_prepare_input(struct xfrm_state *x, struct sk_buff *skb)
163{
164 struct xfrm_mode *inner_mode = x->inner_mode;
165 int err;
166
167 err = x->outer_mode->afinfo->extract_input(x, skb);
168 if (err)
169 return err;
170
171 if (x->sel.family == AF_UNSPEC) {
172 inner_mode = xfrm_ip2inner_mode(x, XFRM_MODE_SKB_CB(skb)->protocol);
173 if (inner_mode == NULL)
174 return -EAFNOSUPPORT;
175 }
176
177 skb->protocol = inner_mode->afinfo->eth_proto;
178 return inner_mode->input2(x, skb);
179}
180EXPORT_SYMBOL(xfrm_prepare_input);
181
182int xfrm_input(struct sk_buff *skb, int nexthdr, __be32 spi, int encap_type)
183{
184 struct net *net = dev_net(skb->dev);
185 int err;
186 __be32 seq;
187 __be32 seq_hi;
188 struct xfrm_state *x = NULL;
189 xfrm_address_t *daddr;
190 struct xfrm_mode *inner_mode;
191 u32 mark = skb->mark;
192 unsigned int family;
193 int decaps = 0;
194 int async = 0;
195
196 /* A negative encap_type indicates async resumption. */
197 if (encap_type < 0) {
198 async = 1;
199 x = xfrm_input_state(skb);
200 seq = XFRM_SKB_CB(skb)->seq.input.low;
201 family = x->outer_mode->afinfo->family;
202 goto resume;
203 }
204
205 daddr = (xfrm_address_t *)(skb_network_header(skb) +
206 XFRM_SPI_SKB_CB(skb)->daddroff);
207 family = XFRM_SPI_SKB_CB(skb)->family;
208
209 /* if tunnel is present override skb->mark value with tunnel i_key */
210 switch (family) {
211 case AF_INET:
212 if (XFRM_TUNNEL_SKB_CB(skb)->tunnel.ip4)
213 mark = be32_to_cpu(XFRM_TUNNEL_SKB_CB(skb)->tunnel.ip4->parms.i_key);
214 break;
215 case AF_INET6:
216 if (XFRM_TUNNEL_SKB_CB(skb)->tunnel.ip6)
217 mark = be32_to_cpu(XFRM_TUNNEL_SKB_CB(skb)->tunnel.ip6->parms.i_key);
218 break;
219 }
220
221 /* Allocate new secpath or COW existing one. */
222 if (!skb->sp || atomic_read(&skb->sp->refcnt) != 1) {
223 struct sec_path *sp;
224
225 sp = secpath_dup(skb->sp);
226 if (!sp) {
227 XFRM_INC_STATS(net, LINUX_MIB_XFRMINERROR);
228 goto drop;
229 }
230 if (skb->sp)
231 secpath_put(skb->sp);
232 skb->sp = sp;
233 }
234
235 seq = 0;
236 if (!spi && (err = xfrm_parse_spi(skb, nexthdr, &spi, &seq)) != 0) {
237 XFRM_INC_STATS(net, LINUX_MIB_XFRMINHDRERROR);
238 goto drop;
239 }
240
241 do {
242 if (skb->sp->len == XFRM_MAX_DEPTH) {
243 XFRM_INC_STATS(net, LINUX_MIB_XFRMINBUFFERERROR);
244 goto drop;
245 }
246
247 x = xfrm_state_lookup(net, mark, daddr, spi, nexthdr, family);
248 if (x == NULL) {
249 XFRM_INC_STATS(net, LINUX_MIB_XFRMINNOSTATES);
250 xfrm_audit_state_notfound(skb, family, spi, seq);
251 goto drop;
252 }
253
254 skb->sp->xvec[skb->sp->len++] = x;
255
256 spin_lock(&x->lock);
257
258 if (unlikely(x->km.state != XFRM_STATE_VALID)) {
259 if (x->km.state == XFRM_STATE_ACQ)
260 XFRM_INC_STATS(net, LINUX_MIB_XFRMACQUIREERROR);
261 else
262 XFRM_INC_STATS(net,
263 LINUX_MIB_XFRMINSTATEINVALID);
264 goto drop_unlock;
265 }
266
267 if ((x->encap ? x->encap->encap_type : 0) != encap_type) {
268 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMISMATCH);
269 goto drop_unlock;
270 }
271
272 if (x->repl->check(x, skb, seq)) {
273 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATESEQERROR);
274 goto drop_unlock;
275 }
276
277 if (xfrm_state_check_expire(x)) {
278 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEEXPIRED);
279 goto drop_unlock;
280 }
281
282 spin_unlock(&x->lock);
283
284 if (xfrm_tunnel_check(skb, x, family)) {
285 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMODEERROR);
286 goto drop;
287 }
288
289 seq_hi = htonl(xfrm_replay_seqhi(x, seq));
290
291 XFRM_SKB_CB(skb)->seq.input.low = seq;
292 XFRM_SKB_CB(skb)->seq.input.hi = seq_hi;
293
294 skb_dst_force(skb);
295 dev_hold(skb->dev);
296
297 nexthdr = x->type->input(x, skb);
298
299 if (nexthdr == -EINPROGRESS)
300 return 0;
301resume:
302 dev_put(skb->dev);
303
304 spin_lock(&x->lock);
305 if (nexthdr <= 0) {
306 if (nexthdr == -EBADMSG) {
307 xfrm_audit_state_icvfail(x, skb,
308 x->type->proto);
309 x->stats.integrity_failed++;
310 }
311 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEPROTOERROR);
312 goto drop_unlock;
313 }
314
315 /* only the first xfrm gets the encap type */
316 encap_type = 0;
317
318 if (async && x->repl->recheck(x, skb, seq)) {
319 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATESEQERROR);
320 goto drop_unlock;
321 }
322
323 x->repl->advance(x, seq);
324
325 x->curlft.bytes += skb->len;
326 x->curlft.packets++;
327
328 spin_unlock(&x->lock);
329
330 XFRM_MODE_SKB_CB(skb)->protocol = nexthdr;
331
332 inner_mode = x->inner_mode;
333
334 if (x->sel.family == AF_UNSPEC) {
335 inner_mode = xfrm_ip2inner_mode(x, XFRM_MODE_SKB_CB(skb)->protocol);
336 if (inner_mode == NULL) {
337 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMODEERROR);
338 goto drop;
339 }
340 }
341
342 if (inner_mode->input(x, skb)) {
343 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMODEERROR);
344 goto drop;
345 }
346
347 if (x->outer_mode->flags & XFRM_MODE_FLAG_TUNNEL) {
348 decaps = 1;
349 break;
350 }
351
352 /*
353 * We need the inner address. However, we only get here for
354 * transport mode so the outer address is identical.
355 */
356 daddr = &x->id.daddr;
357 family = x->outer_mode->afinfo->family;
358
359 err = xfrm_parse_spi(skb, nexthdr, &spi, &seq);
360 if (err < 0) {
361 XFRM_INC_STATS(net, LINUX_MIB_XFRMINHDRERROR);
362 goto drop;
363 }
364 } while (!err);
365
366 err = xfrm_rcv_cb(skb, family, x->type->proto, 0);
367 if (err)
368 goto drop;
369
370 nf_reset(skb);
371
372 if (decaps) {
373 skb_dst_drop(skb);
374 netif_rx(skb);
375 return 0;
376 } else {
377 return x->inner_mode->afinfo->transport_finish(skb, async);
378 }
379
380drop_unlock:
381 spin_unlock(&x->lock);
382drop:
383 xfrm_rcv_cb(skb, family, x && x->type ? x->type->proto : nexthdr, -1);
384 kfree_skb(skb);
385 return 0;
386}
387EXPORT_SYMBOL(xfrm_input);
388
389int xfrm_input_resume(struct sk_buff *skb, int nexthdr)
390{
391 return xfrm_input(skb, nexthdr, 0, -1);
392}
393EXPORT_SYMBOL(xfrm_input_resume);
394
395void __init xfrm_input_init(void)
396{
397 secpath_cachep = kmem_cache_create("secpath_cache",
398 sizeof(struct sec_path),
399 0, SLAB_HWCACHE_ALIGN|SLAB_PANIC,
400 NULL);
401}