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#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 || encap_type == 0 ||
475 encap_type == UDP_ENCAP_ESPINUDP))) {
476 x = xfrm_input_state(skb);
477
478 if (unlikely(x->km.state != XFRM_STATE_VALID)) {
479 if (x->km.state == XFRM_STATE_ACQ)
480 XFRM_INC_STATS(net, LINUX_MIB_XFRMACQUIREERROR);
481 else
482 XFRM_INC_STATS(net,
483 LINUX_MIB_XFRMINSTATEINVALID);
484
485 if (encap_type == -1)
486 dev_put(skb->dev);
487 goto drop;
488 }
489
490 family = x->props.family;
491
492 /* An encap_type of -1 indicates async resumption. */
493 if (encap_type == -1) {
494 async = 1;
495 seq = XFRM_SKB_CB(skb)->seq.input.low;
496 goto resume;
497 }
498 /* GRO call */
499 seq = XFRM_SPI_SKB_CB(skb)->seq;
500
501 if (xo && (xo->flags & CRYPTO_DONE)) {
502 crypto_done = true;
503 family = XFRM_SPI_SKB_CB(skb)->family;
504
505 if (!(xo->status & CRYPTO_SUCCESS)) {
506 if (xo->status &
507 (CRYPTO_TRANSPORT_AH_AUTH_FAILED |
508 CRYPTO_TRANSPORT_ESP_AUTH_FAILED |
509 CRYPTO_TUNNEL_AH_AUTH_FAILED |
510 CRYPTO_TUNNEL_ESP_AUTH_FAILED)) {
511
512 xfrm_audit_state_icvfail(x, skb,
513 x->type->proto);
514 x->stats.integrity_failed++;
515 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEPROTOERROR);
516 goto drop;
517 }
518
519 if (xo->status & CRYPTO_INVALID_PROTOCOL) {
520 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEPROTOERROR);
521 goto drop;
522 }
523
524 XFRM_INC_STATS(net, LINUX_MIB_XFRMINBUFFERERROR);
525 goto drop;
526 }
527
528 if (xfrm_parse_spi(skb, nexthdr, &spi, &seq)) {
529 XFRM_INC_STATS(net, LINUX_MIB_XFRMINHDRERROR);
530 goto drop;
531 }
532 }
533
534 goto lock;
535 }
536
537 family = XFRM_SPI_SKB_CB(skb)->family;
538
539 /* if tunnel is present override skb->mark value with tunnel i_key */
540 switch (family) {
541 case AF_INET:
542 if (XFRM_TUNNEL_SKB_CB(skb)->tunnel.ip4)
543 mark = be32_to_cpu(XFRM_TUNNEL_SKB_CB(skb)->tunnel.ip4->parms.i_key);
544 break;
545 case AF_INET6:
546 if (XFRM_TUNNEL_SKB_CB(skb)->tunnel.ip6)
547 mark = be32_to_cpu(XFRM_TUNNEL_SKB_CB(skb)->tunnel.ip6->parms.i_key);
548 break;
549 }
550
551 sp = secpath_set(skb);
552 if (!sp) {
553 XFRM_INC_STATS(net, LINUX_MIB_XFRMINERROR);
554 goto drop;
555 }
556
557 seq = 0;
558 if (!spi && xfrm_parse_spi(skb, nexthdr, &spi, &seq)) {
559 secpath_reset(skb);
560 XFRM_INC_STATS(net, LINUX_MIB_XFRMINHDRERROR);
561 goto drop;
562 }
563
564 daddr = (xfrm_address_t *)(skb_network_header(skb) +
565 XFRM_SPI_SKB_CB(skb)->daddroff);
566 do {
567 sp = skb_sec_path(skb);
568
569 if (sp->len == XFRM_MAX_DEPTH) {
570 secpath_reset(skb);
571 XFRM_INC_STATS(net, LINUX_MIB_XFRMINBUFFERERROR);
572 goto drop;
573 }
574
575 x = xfrm_input_state_lookup(net, mark, daddr, spi, nexthdr, family);
576 if (x == NULL) {
577 secpath_reset(skb);
578 XFRM_INC_STATS(net, LINUX_MIB_XFRMINNOSTATES);
579 xfrm_audit_state_notfound(skb, family, spi, seq);
580 goto drop;
581 }
582
583 if (unlikely(x->dir && x->dir != XFRM_SA_DIR_IN)) {
584 secpath_reset(skb);
585 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEDIRERROR);
586 xfrm_audit_state_notfound(skb, family, spi, seq);
587 xfrm_state_put(x);
588 x = NULL;
589 goto drop;
590 }
591
592 skb->mark = xfrm_smark_get(skb->mark, x);
593
594 sp->xvec[sp->len++] = x;
595
596 skb_dst_force(skb);
597 if (!skb_dst(skb)) {
598 XFRM_INC_STATS(net, LINUX_MIB_XFRMINERROR);
599 goto drop;
600 }
601
602lock:
603 spin_lock(&x->lock);
604
605 if (unlikely(x->km.state != XFRM_STATE_VALID)) {
606 if (x->km.state == XFRM_STATE_ACQ)
607 XFRM_INC_STATS(net, LINUX_MIB_XFRMACQUIREERROR);
608 else
609 XFRM_INC_STATS(net,
610 LINUX_MIB_XFRMINSTATEINVALID);
611 goto drop_unlock;
612 }
613
614 if ((x->encap ? x->encap->encap_type : 0) != encap_type) {
615 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMISMATCH);
616 goto drop_unlock;
617 }
618
619 if (xfrm_replay_check(x, skb, seq)) {
620 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATESEQERROR);
621 goto drop_unlock;
622 }
623
624 if (xfrm_state_check_expire(x)) {
625 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEEXPIRED);
626 goto drop_unlock;
627 }
628
629 spin_unlock(&x->lock);
630
631 if (xfrm_tunnel_check(skb, x, family)) {
632 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMODEERROR);
633 goto drop;
634 }
635
636 seq_hi = htonl(xfrm_replay_seqhi(x, seq));
637
638 XFRM_SKB_CB(skb)->seq.input.low = seq;
639 XFRM_SKB_CB(skb)->seq.input.hi = seq_hi;
640
641 dev_hold(skb->dev);
642
643 if (crypto_done)
644 nexthdr = x->type_offload->input_tail(x, skb);
645 else
646 nexthdr = x->type->input(x, skb);
647
648 if (nexthdr == -EINPROGRESS)
649 return 0;
650resume:
651 dev_put(skb->dev);
652
653 spin_lock(&x->lock);
654 if (nexthdr < 0) {
655 if (nexthdr == -EBADMSG) {
656 xfrm_audit_state_icvfail(x, skb,
657 x->type->proto);
658 x->stats.integrity_failed++;
659 }
660 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEPROTOERROR);
661 goto drop_unlock;
662 }
663
664 /* only the first xfrm gets the encap type */
665 encap_type = 0;
666
667 if (xfrm_replay_recheck(x, skb, seq)) {
668 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATESEQERROR);
669 goto drop_unlock;
670 }
671
672 xfrm_replay_advance(x, seq);
673
674 x->curlft.bytes += skb->len;
675 x->curlft.packets++;
676 x->lastused = ktime_get_real_seconds();
677
678 spin_unlock(&x->lock);
679
680 XFRM_MODE_SKB_CB(skb)->protocol = nexthdr;
681
682 if (xfrm_inner_mode_input(x, skb)) {
683 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMODEERROR);
684 goto drop;
685 }
686
687 if (x->outer_mode.flags & XFRM_MODE_FLAG_TUNNEL) {
688 decaps = 1;
689 break;
690 }
691
692 /*
693 * We need the inner address. However, we only get here for
694 * transport mode so the outer address is identical.
695 */
696 daddr = &x->id.daddr;
697 family = x->props.family;
698
699 err = xfrm_parse_spi(skb, nexthdr, &spi, &seq);
700 if (err < 0) {
701 XFRM_INC_STATS(net, LINUX_MIB_XFRMINHDRERROR);
702 goto drop;
703 }
704 crypto_done = false;
705 } while (!err);
706
707 err = xfrm_rcv_cb(skb, family, x->type->proto, 0);
708 if (err)
709 goto drop;
710
711 nf_reset_ct(skb);
712
713 if (decaps) {
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 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->props.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 if (skb_valid_dst(skb))
737 skb_dst_drop(skb);
738 gro_cells_receive(&gro_cells, skb);
739 return err;
740 }
741
742 return err;
743 }
744
745drop_unlock:
746 spin_unlock(&x->lock);
747drop:
748 xfrm_rcv_cb(skb, family, x && x->type ? x->type->proto : nexthdr, -1);
749 kfree_skb(skb);
750 return 0;
751}
752EXPORT_SYMBOL(xfrm_input);
753
754int xfrm_input_resume(struct sk_buff *skb, int nexthdr)
755{
756 return xfrm_input(skb, nexthdr, 0, -1);
757}
758EXPORT_SYMBOL(xfrm_input_resume);
759
760static void xfrm_trans_reinject(struct work_struct *work)
761{
762 struct xfrm_trans_tasklet *trans = container_of(work, struct xfrm_trans_tasklet, work);
763 struct sk_buff_head queue;
764 struct sk_buff *skb;
765
766 __skb_queue_head_init(&queue);
767 spin_lock_bh(&trans->queue_lock);
768 skb_queue_splice_init(&trans->queue, &queue);
769 spin_unlock_bh(&trans->queue_lock);
770
771 local_bh_disable();
772 while ((skb = __skb_dequeue(&queue)))
773 XFRM_TRANS_SKB_CB(skb)->finish(XFRM_TRANS_SKB_CB(skb)->net,
774 NULL, skb);
775 local_bh_enable();
776}
777
778int xfrm_trans_queue_net(struct net *net, struct sk_buff *skb,
779 int (*finish)(struct net *, struct sock *,
780 struct sk_buff *))
781{
782 struct xfrm_trans_tasklet *trans;
783
784 trans = this_cpu_ptr(&xfrm_trans_tasklet);
785
786 if (skb_queue_len(&trans->queue) >= READ_ONCE(net_hotdata.max_backlog))
787 return -ENOBUFS;
788
789 BUILD_BUG_ON(sizeof(struct xfrm_trans_cb) > sizeof(skb->cb));
790
791 XFRM_TRANS_SKB_CB(skb)->finish = finish;
792 XFRM_TRANS_SKB_CB(skb)->net = net;
793 spin_lock_bh(&trans->queue_lock);
794 __skb_queue_tail(&trans->queue, skb);
795 spin_unlock_bh(&trans->queue_lock);
796 schedule_work(&trans->work);
797 return 0;
798}
799EXPORT_SYMBOL(xfrm_trans_queue_net);
800
801int xfrm_trans_queue(struct sk_buff *skb,
802 int (*finish)(struct net *, struct sock *,
803 struct sk_buff *))
804{
805 return xfrm_trans_queue_net(dev_net(skb->dev), skb, finish);
806}
807EXPORT_SYMBOL(xfrm_trans_queue);
808
809void __init xfrm_input_init(void)
810{
811 int err;
812 int i;
813
814 init_dummy_netdev(&xfrm_napi_dev);
815 err = gro_cells_init(&gro_cells, &xfrm_napi_dev);
816 if (err)
817 gro_cells.cells = NULL;
818
819 for_each_possible_cpu(i) {
820 struct xfrm_trans_tasklet *trans;
821
822 trans = &per_cpu(xfrm_trans_tasklet, i);
823 spin_lock_init(&trans->queue_lock);
824 __skb_queue_head_init(&trans->queue);
825 INIT_WORK(&trans->work, xfrm_trans_reinject);
826 }
827}
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
17static struct kmem_cache *secpath_cachep __read_mostly;
18
19void __secpath_destroy(struct sec_path *sp)
20{
21 int i;
22 for (i = 0; i < sp->len; i++)
23 xfrm_state_put(sp->xvec[i]);
24 kmem_cache_free(secpath_cachep, sp);
25}
26EXPORT_SYMBOL(__secpath_destroy);
27
28struct sec_path *secpath_dup(struct sec_path *src)
29{
30 struct sec_path *sp;
31
32 sp = kmem_cache_alloc(secpath_cachep, GFP_ATOMIC);
33 if (!sp)
34 return NULL;
35
36 sp->len = 0;
37 if (src) {
38 int i;
39
40 memcpy(sp, src, sizeof(*sp));
41 for (i = 0; i < sp->len; i++)
42 xfrm_state_hold(sp->xvec[i]);
43 }
44 atomic_set(&sp->refcnt, 1);
45 return sp;
46}
47EXPORT_SYMBOL(secpath_dup);
48
49/* Fetch spi and seq from ipsec header */
50
51int xfrm_parse_spi(struct sk_buff *skb, u8 nexthdr, __be32 *spi, __be32 *seq)
52{
53 int offset, offset_seq;
54 int hlen;
55
56 switch (nexthdr) {
57 case IPPROTO_AH:
58 hlen = sizeof(struct ip_auth_hdr);
59 offset = offsetof(struct ip_auth_hdr, spi);
60 offset_seq = offsetof(struct ip_auth_hdr, seq_no);
61 break;
62 case IPPROTO_ESP:
63 hlen = sizeof(struct ip_esp_hdr);
64 offset = offsetof(struct ip_esp_hdr, spi);
65 offset_seq = offsetof(struct ip_esp_hdr, seq_no);
66 break;
67 case IPPROTO_COMP:
68 if (!pskb_may_pull(skb, sizeof(struct ip_comp_hdr)))
69 return -EINVAL;
70 *spi = htonl(ntohs(*(__be16*)(skb_transport_header(skb) + 2)));
71 *seq = 0;
72 return 0;
73 default:
74 return 1;
75 }
76
77 if (!pskb_may_pull(skb, hlen))
78 return -EINVAL;
79
80 *spi = *(__be32*)(skb_transport_header(skb) + offset);
81 *seq = *(__be32*)(skb_transport_header(skb) + offset_seq);
82 return 0;
83}
84
85int xfrm_prepare_input(struct xfrm_state *x, struct sk_buff *skb)
86{
87 struct xfrm_mode *inner_mode = x->inner_mode;
88 int err;
89
90 err = x->outer_mode->afinfo->extract_input(x, skb);
91 if (err)
92 return err;
93
94 if (x->sel.family == AF_UNSPEC) {
95 inner_mode = xfrm_ip2inner_mode(x, XFRM_MODE_SKB_CB(skb)->protocol);
96 if (inner_mode == NULL)
97 return -EAFNOSUPPORT;
98 }
99
100 skb->protocol = inner_mode->afinfo->eth_proto;
101 return inner_mode->input2(x, skb);
102}
103EXPORT_SYMBOL(xfrm_prepare_input);
104
105int xfrm_input(struct sk_buff *skb, int nexthdr, __be32 spi, int encap_type)
106{
107 struct net *net = dev_net(skb->dev);
108 int err;
109 __be32 seq;
110 __be32 seq_hi;
111 struct xfrm_state *x;
112 xfrm_address_t *daddr;
113 struct xfrm_mode *inner_mode;
114 unsigned int family;
115 int decaps = 0;
116 int async = 0;
117
118 /* A negative encap_type indicates async resumption. */
119 if (encap_type < 0) {
120 async = 1;
121 x = xfrm_input_state(skb);
122 seq = XFRM_SKB_CB(skb)->seq.input.low;
123 goto resume;
124 }
125
126 /* Allocate new secpath or COW existing one. */
127 if (!skb->sp || atomic_read(&skb->sp->refcnt) != 1) {
128 struct sec_path *sp;
129
130 sp = secpath_dup(skb->sp);
131 if (!sp) {
132 XFRM_INC_STATS(net, LINUX_MIB_XFRMINERROR);
133 goto drop;
134 }
135 if (skb->sp)
136 secpath_put(skb->sp);
137 skb->sp = sp;
138 }
139
140 daddr = (xfrm_address_t *)(skb_network_header(skb) +
141 XFRM_SPI_SKB_CB(skb)->daddroff);
142 family = XFRM_SPI_SKB_CB(skb)->family;
143
144 seq = 0;
145 if (!spi && (err = xfrm_parse_spi(skb, nexthdr, &spi, &seq)) != 0) {
146 XFRM_INC_STATS(net, LINUX_MIB_XFRMINHDRERROR);
147 goto drop;
148 }
149
150 do {
151 if (skb->sp->len == XFRM_MAX_DEPTH) {
152 XFRM_INC_STATS(net, LINUX_MIB_XFRMINBUFFERERROR);
153 goto drop;
154 }
155
156 x = xfrm_state_lookup(net, skb->mark, daddr, spi, nexthdr, family);
157 if (x == NULL) {
158 XFRM_INC_STATS(net, LINUX_MIB_XFRMINNOSTATES);
159 xfrm_audit_state_notfound(skb, family, spi, seq);
160 goto drop;
161 }
162
163 skb->sp->xvec[skb->sp->len++] = x;
164
165 spin_lock(&x->lock);
166 if (unlikely(x->km.state != XFRM_STATE_VALID)) {
167 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEINVALID);
168 goto drop_unlock;
169 }
170
171 if ((x->encap ? x->encap->encap_type : 0) != encap_type) {
172 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMISMATCH);
173 goto drop_unlock;
174 }
175
176 if (x->repl->check(x, skb, seq)) {
177 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATESEQERROR);
178 goto drop_unlock;
179 }
180
181 if (xfrm_state_check_expire(x)) {
182 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEEXPIRED);
183 goto drop_unlock;
184 }
185
186 spin_unlock(&x->lock);
187
188 seq_hi = htonl(xfrm_replay_seqhi(x, seq));
189
190 XFRM_SKB_CB(skb)->seq.input.low = seq;
191 XFRM_SKB_CB(skb)->seq.input.hi = seq_hi;
192
193 skb_dst_force(skb);
194
195 nexthdr = x->type->input(x, skb);
196
197 if (nexthdr == -EINPROGRESS)
198 return 0;
199
200resume:
201 spin_lock(&x->lock);
202 if (nexthdr <= 0) {
203 if (nexthdr == -EBADMSG) {
204 xfrm_audit_state_icvfail(x, skb,
205 x->type->proto);
206 x->stats.integrity_failed++;
207 }
208 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEPROTOERROR);
209 goto drop_unlock;
210 }
211
212 /* only the first xfrm gets the encap type */
213 encap_type = 0;
214
215 if (async && x->repl->check(x, skb, seq)) {
216 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATESEQERROR);
217 goto drop_unlock;
218 }
219
220 x->repl->advance(x, seq);
221
222 x->curlft.bytes += skb->len;
223 x->curlft.packets++;
224
225 spin_unlock(&x->lock);
226
227 XFRM_MODE_SKB_CB(skb)->protocol = nexthdr;
228
229 inner_mode = x->inner_mode;
230
231 if (x->sel.family == AF_UNSPEC) {
232 inner_mode = xfrm_ip2inner_mode(x, XFRM_MODE_SKB_CB(skb)->protocol);
233 if (inner_mode == NULL)
234 goto drop;
235 }
236
237 if (inner_mode->input(x, skb)) {
238 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMODEERROR);
239 goto drop;
240 }
241
242 if (x->outer_mode->flags & XFRM_MODE_FLAG_TUNNEL) {
243 decaps = 1;
244 break;
245 }
246
247 /*
248 * We need the inner address. However, we only get here for
249 * transport mode so the outer address is identical.
250 */
251 daddr = &x->id.daddr;
252 family = x->outer_mode->afinfo->family;
253
254 err = xfrm_parse_spi(skb, nexthdr, &spi, &seq);
255 if (err < 0) {
256 XFRM_INC_STATS(net, LINUX_MIB_XFRMINHDRERROR);
257 goto drop;
258 }
259 } while (!err);
260
261 nf_reset(skb);
262
263 if (decaps) {
264 skb_dst_drop(skb);
265 netif_rx(skb);
266 return 0;
267 } else {
268 return x->inner_mode->afinfo->transport_finish(skb, async);
269 }
270
271drop_unlock:
272 spin_unlock(&x->lock);
273drop:
274 kfree_skb(skb);
275 return 0;
276}
277EXPORT_SYMBOL(xfrm_input);
278
279int xfrm_input_resume(struct sk_buff *skb, int nexthdr)
280{
281 return xfrm_input(skb, nexthdr, 0, -1);
282}
283EXPORT_SYMBOL(xfrm_input_resume);
284
285void __init xfrm_input_init(void)
286{
287 secpath_cachep = kmem_cache_create("secpath_cache",
288 sizeof(struct sec_path),
289 0, SLAB_HWCACHE_ALIGN|SLAB_PANIC,
290 NULL);
291}