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 sp->verified_cnt = 0;
135
136 return sp;
137}
138EXPORT_SYMBOL(secpath_set);
139
140/* Fetch spi and seq from ipsec header */
141
142int xfrm_parse_spi(struct sk_buff *skb, u8 nexthdr, __be32 *spi, __be32 *seq)
143{
144 int offset, offset_seq;
145 int hlen;
146
147 switch (nexthdr) {
148 case IPPROTO_AH:
149 hlen = sizeof(struct ip_auth_hdr);
150 offset = offsetof(struct ip_auth_hdr, spi);
151 offset_seq = offsetof(struct ip_auth_hdr, seq_no);
152 break;
153 case IPPROTO_ESP:
154 hlen = sizeof(struct ip_esp_hdr);
155 offset = offsetof(struct ip_esp_hdr, spi);
156 offset_seq = offsetof(struct ip_esp_hdr, seq_no);
157 break;
158 case IPPROTO_COMP:
159 if (!pskb_may_pull(skb, sizeof(struct ip_comp_hdr)))
160 return -EINVAL;
161 *spi = htonl(ntohs(*(__be16 *)(skb_transport_header(skb) + 2)));
162 *seq = 0;
163 return 0;
164 default:
165 return 1;
166 }
167
168 if (!pskb_may_pull(skb, hlen))
169 return -EINVAL;
170
171 *spi = *(__be32 *)(skb_transport_header(skb) + offset);
172 *seq = *(__be32 *)(skb_transport_header(skb) + offset_seq);
173 return 0;
174}
175EXPORT_SYMBOL(xfrm_parse_spi);
176
177static int xfrm4_remove_beet_encap(struct xfrm_state *x, struct sk_buff *skb)
178{
179 struct iphdr *iph;
180 int optlen = 0;
181 int err = -EINVAL;
182
183 skb->protocol = htons(ETH_P_IP);
184
185 if (unlikely(XFRM_MODE_SKB_CB(skb)->protocol == IPPROTO_BEETPH)) {
186 struct ip_beet_phdr *ph;
187 int phlen;
188
189 if (!pskb_may_pull(skb, sizeof(*ph)))
190 goto out;
191
192 ph = (struct ip_beet_phdr *)skb->data;
193
194 phlen = sizeof(*ph) + ph->padlen;
195 optlen = ph->hdrlen * 8 + (IPV4_BEET_PHMAXLEN - phlen);
196 if (optlen < 0 || optlen & 3 || optlen > 250)
197 goto out;
198
199 XFRM_MODE_SKB_CB(skb)->protocol = ph->nexthdr;
200
201 if (!pskb_may_pull(skb, phlen))
202 goto out;
203 __skb_pull(skb, phlen);
204 }
205
206 skb_push(skb, sizeof(*iph));
207 skb_reset_network_header(skb);
208 skb_mac_header_rebuild(skb);
209
210 xfrm4_beet_make_header(skb);
211
212 iph = ip_hdr(skb);
213
214 iph->ihl += optlen / 4;
215 iph->tot_len = htons(skb->len);
216 iph->daddr = x->sel.daddr.a4;
217 iph->saddr = x->sel.saddr.a4;
218 iph->check = 0;
219 iph->check = ip_fast_csum(skb_network_header(skb), iph->ihl);
220 err = 0;
221out:
222 return err;
223}
224
225static void ipip_ecn_decapsulate(struct sk_buff *skb)
226{
227 struct iphdr *inner_iph = ipip_hdr(skb);
228
229 if (INET_ECN_is_ce(XFRM_MODE_SKB_CB(skb)->tos))
230 IP_ECN_set_ce(inner_iph);
231}
232
233static int xfrm4_remove_tunnel_encap(struct xfrm_state *x, struct sk_buff *skb)
234{
235 int err = -EINVAL;
236
237 skb->protocol = htons(ETH_P_IP);
238
239 if (!pskb_may_pull(skb, sizeof(struct iphdr)))
240 goto out;
241
242 err = skb_unclone(skb, GFP_ATOMIC);
243 if (err)
244 goto out;
245
246 if (x->props.flags & XFRM_STATE_DECAP_DSCP)
247 ipv4_copy_dscp(XFRM_MODE_SKB_CB(skb)->tos, ipip_hdr(skb));
248 if (!(x->props.flags & XFRM_STATE_NOECN))
249 ipip_ecn_decapsulate(skb);
250
251 skb_reset_network_header(skb);
252 skb_mac_header_rebuild(skb);
253 if (skb->mac_len)
254 eth_hdr(skb)->h_proto = skb->protocol;
255
256 err = 0;
257
258out:
259 return err;
260}
261
262static void ipip6_ecn_decapsulate(struct sk_buff *skb)
263{
264 struct ipv6hdr *inner_iph = ipipv6_hdr(skb);
265
266 if (INET_ECN_is_ce(XFRM_MODE_SKB_CB(skb)->tos))
267 IP6_ECN_set_ce(skb, inner_iph);
268}
269
270static int xfrm6_remove_tunnel_encap(struct xfrm_state *x, struct sk_buff *skb)
271{
272 int err = -EINVAL;
273
274 skb->protocol = htons(ETH_P_IPV6);
275
276 if (!pskb_may_pull(skb, sizeof(struct ipv6hdr)))
277 goto out;
278
279 err = skb_unclone(skb, GFP_ATOMIC);
280 if (err)
281 goto out;
282
283 if (x->props.flags & XFRM_STATE_DECAP_DSCP)
284 ipv6_copy_dscp(XFRM_MODE_SKB_CB(skb)->tos, ipipv6_hdr(skb));
285 if (!(x->props.flags & XFRM_STATE_NOECN))
286 ipip6_ecn_decapsulate(skb);
287
288 skb_reset_network_header(skb);
289 skb_mac_header_rebuild(skb);
290 if (skb->mac_len)
291 eth_hdr(skb)->h_proto = skb->protocol;
292
293 err = 0;
294
295out:
296 return err;
297}
298
299static int xfrm6_remove_beet_encap(struct xfrm_state *x, struct sk_buff *skb)
300{
301 struct ipv6hdr *ip6h;
302 int size = sizeof(struct ipv6hdr);
303 int err;
304
305 skb->protocol = htons(ETH_P_IPV6);
306
307 err = skb_cow_head(skb, size + skb->mac_len);
308 if (err)
309 goto out;
310
311 __skb_push(skb, size);
312 skb_reset_network_header(skb);
313 skb_mac_header_rebuild(skb);
314
315 xfrm6_beet_make_header(skb);
316
317 ip6h = ipv6_hdr(skb);
318 ip6h->payload_len = htons(skb->len - size);
319 ip6h->daddr = x->sel.daddr.in6;
320 ip6h->saddr = x->sel.saddr.in6;
321 err = 0;
322out:
323 return err;
324}
325
326/* Remove encapsulation header.
327 *
328 * The IP header will be moved over the top of the encapsulation
329 * header.
330 *
331 * On entry, the transport header shall point to where the IP header
332 * should be and the network header shall be set to where the IP
333 * header currently is. skb->data shall point to the start of the
334 * payload.
335 */
336static int
337xfrm_inner_mode_encap_remove(struct xfrm_state *x,
338 struct sk_buff *skb)
339{
340 switch (x->props.mode) {
341 case XFRM_MODE_BEET:
342 switch (x->sel.family) {
343 case AF_INET:
344 return xfrm4_remove_beet_encap(x, skb);
345 case AF_INET6:
346 return xfrm6_remove_beet_encap(x, skb);
347 }
348 break;
349 case XFRM_MODE_TUNNEL:
350 switch (XFRM_MODE_SKB_CB(skb)->protocol) {
351 case IPPROTO_IPIP:
352 return xfrm4_remove_tunnel_encap(x, skb);
353 case IPPROTO_IPV6:
354 return xfrm6_remove_tunnel_encap(x, skb);
355 break;
356 }
357 return -EINVAL;
358 }
359
360 WARN_ON_ONCE(1);
361 return -EOPNOTSUPP;
362}
363
364static int xfrm_prepare_input(struct xfrm_state *x, struct sk_buff *skb)
365{
366 switch (x->props.family) {
367 case AF_INET:
368 xfrm4_extract_header(skb);
369 break;
370 case AF_INET6:
371 xfrm6_extract_header(skb);
372 break;
373 default:
374 WARN_ON_ONCE(1);
375 return -EAFNOSUPPORT;
376 }
377
378 return xfrm_inner_mode_encap_remove(x, skb);
379}
380
381/* Remove encapsulation header.
382 *
383 * The IP header will be moved over the top of the encapsulation header.
384 *
385 * On entry, skb_transport_header() shall point to where the IP header
386 * should be and skb_network_header() shall be set to where the IP header
387 * currently is. skb->data shall point to the start of the payload.
388 */
389static int xfrm4_transport_input(struct xfrm_state *x, struct sk_buff *skb)
390{
391 int ihl = skb->data - skb_transport_header(skb);
392
393 if (skb->transport_header != skb->network_header) {
394 memmove(skb_transport_header(skb),
395 skb_network_header(skb), ihl);
396 skb->network_header = skb->transport_header;
397 }
398 ip_hdr(skb)->tot_len = htons(skb->len + ihl);
399 skb_reset_transport_header(skb);
400 return 0;
401}
402
403static int xfrm6_transport_input(struct xfrm_state *x, struct sk_buff *skb)
404{
405#if IS_ENABLED(CONFIG_IPV6)
406 int ihl = skb->data - skb_transport_header(skb);
407
408 if (skb->transport_header != skb->network_header) {
409 memmove(skb_transport_header(skb),
410 skb_network_header(skb), ihl);
411 skb->network_header = skb->transport_header;
412 }
413 ipv6_hdr(skb)->payload_len = htons(skb->len + ihl -
414 sizeof(struct ipv6hdr));
415 skb_reset_transport_header(skb);
416 return 0;
417#else
418 WARN_ON_ONCE(1);
419 return -EAFNOSUPPORT;
420#endif
421}
422
423static int xfrm_inner_mode_input(struct xfrm_state *x,
424 struct sk_buff *skb)
425{
426 switch (x->props.mode) {
427 case XFRM_MODE_BEET:
428 case XFRM_MODE_TUNNEL:
429 return xfrm_prepare_input(x, skb);
430 case XFRM_MODE_TRANSPORT:
431 if (x->props.family == AF_INET)
432 return xfrm4_transport_input(x, skb);
433 if (x->props.family == AF_INET6)
434 return xfrm6_transport_input(x, skb);
435 break;
436 case XFRM_MODE_ROUTEOPTIMIZATION:
437 WARN_ON_ONCE(1);
438 break;
439 default:
440 WARN_ON_ONCE(1);
441 break;
442 }
443
444 return -EOPNOTSUPP;
445}
446
447int xfrm_input(struct sk_buff *skb, int nexthdr, __be32 spi, int encap_type)
448{
449 const struct xfrm_state_afinfo *afinfo;
450 struct net *net = dev_net(skb->dev);
451 int err;
452 __be32 seq;
453 __be32 seq_hi;
454 struct xfrm_state *x = NULL;
455 xfrm_address_t *daddr;
456 u32 mark = skb->mark;
457 unsigned int family = AF_UNSPEC;
458 int decaps = 0;
459 int async = 0;
460 bool xfrm_gro = false;
461 bool crypto_done = false;
462 struct xfrm_offload *xo = xfrm_offload(skb);
463 struct sec_path *sp;
464
465 if (encap_type < 0 || (xo && xo->flags & XFRM_GRO)) {
466 x = xfrm_input_state(skb);
467
468 if (unlikely(x->km.state != XFRM_STATE_VALID)) {
469 if (x->km.state == XFRM_STATE_ACQ)
470 XFRM_INC_STATS(net, LINUX_MIB_XFRMACQUIREERROR);
471 else
472 XFRM_INC_STATS(net,
473 LINUX_MIB_XFRMINSTATEINVALID);
474
475 if (encap_type == -1)
476 dev_put(skb->dev);
477 goto drop;
478 }
479
480 family = x->props.family;
481
482 /* An encap_type of -1 indicates async resumption. */
483 if (encap_type == -1) {
484 async = 1;
485 seq = XFRM_SKB_CB(skb)->seq.input.low;
486 goto resume;
487 }
488 /* GRO call */
489 seq = XFRM_SPI_SKB_CB(skb)->seq;
490
491 if (xo && (xo->flags & CRYPTO_DONE)) {
492 crypto_done = true;
493 family = XFRM_SPI_SKB_CB(skb)->family;
494
495 if (!(xo->status & CRYPTO_SUCCESS)) {
496 if (xo->status &
497 (CRYPTO_TRANSPORT_AH_AUTH_FAILED |
498 CRYPTO_TRANSPORT_ESP_AUTH_FAILED |
499 CRYPTO_TUNNEL_AH_AUTH_FAILED |
500 CRYPTO_TUNNEL_ESP_AUTH_FAILED)) {
501
502 xfrm_audit_state_icvfail(x, skb,
503 x->type->proto);
504 x->stats.integrity_failed++;
505 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEPROTOERROR);
506 goto drop;
507 }
508
509 if (xo->status & CRYPTO_INVALID_PROTOCOL) {
510 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEPROTOERROR);
511 goto drop;
512 }
513
514 XFRM_INC_STATS(net, LINUX_MIB_XFRMINBUFFERERROR);
515 goto drop;
516 }
517
518 if (xfrm_parse_spi(skb, nexthdr, &spi, &seq)) {
519 XFRM_INC_STATS(net, LINUX_MIB_XFRMINHDRERROR);
520 goto drop;
521 }
522 }
523
524 goto lock;
525 }
526
527 family = XFRM_SPI_SKB_CB(skb)->family;
528
529 /* if tunnel is present override skb->mark value with tunnel i_key */
530 switch (family) {
531 case AF_INET:
532 if (XFRM_TUNNEL_SKB_CB(skb)->tunnel.ip4)
533 mark = be32_to_cpu(XFRM_TUNNEL_SKB_CB(skb)->tunnel.ip4->parms.i_key);
534 break;
535 case AF_INET6:
536 if (XFRM_TUNNEL_SKB_CB(skb)->tunnel.ip6)
537 mark = be32_to_cpu(XFRM_TUNNEL_SKB_CB(skb)->tunnel.ip6->parms.i_key);
538 break;
539 }
540
541 sp = secpath_set(skb);
542 if (!sp) {
543 XFRM_INC_STATS(net, LINUX_MIB_XFRMINERROR);
544 goto drop;
545 }
546
547 seq = 0;
548 if (!spi && xfrm_parse_spi(skb, nexthdr, &spi, &seq)) {
549 secpath_reset(skb);
550 XFRM_INC_STATS(net, LINUX_MIB_XFRMINHDRERROR);
551 goto drop;
552 }
553
554 daddr = (xfrm_address_t *)(skb_network_header(skb) +
555 XFRM_SPI_SKB_CB(skb)->daddroff);
556 do {
557 sp = skb_sec_path(skb);
558
559 if (sp->len == XFRM_MAX_DEPTH) {
560 secpath_reset(skb);
561 XFRM_INC_STATS(net, LINUX_MIB_XFRMINBUFFERERROR);
562 goto drop;
563 }
564
565 x = xfrm_state_lookup(net, mark, daddr, spi, nexthdr, family);
566 if (x == NULL) {
567 secpath_reset(skb);
568 XFRM_INC_STATS(net, LINUX_MIB_XFRMINNOSTATES);
569 xfrm_audit_state_notfound(skb, family, spi, seq);
570 goto drop;
571 }
572
573 skb->mark = xfrm_smark_get(skb->mark, x);
574
575 sp->xvec[sp->len++] = x;
576
577 skb_dst_force(skb);
578 if (!skb_dst(skb)) {
579 XFRM_INC_STATS(net, LINUX_MIB_XFRMINERROR);
580 goto drop;
581 }
582
583lock:
584 spin_lock(&x->lock);
585
586 if (unlikely(x->km.state != XFRM_STATE_VALID)) {
587 if (x->km.state == XFRM_STATE_ACQ)
588 XFRM_INC_STATS(net, LINUX_MIB_XFRMACQUIREERROR);
589 else
590 XFRM_INC_STATS(net,
591 LINUX_MIB_XFRMINSTATEINVALID);
592 goto drop_unlock;
593 }
594
595 if ((x->encap ? x->encap->encap_type : 0) != encap_type) {
596 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMISMATCH);
597 goto drop_unlock;
598 }
599
600 if (xfrm_replay_check(x, skb, seq)) {
601 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATESEQERROR);
602 goto drop_unlock;
603 }
604
605 if (xfrm_state_check_expire(x)) {
606 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEEXPIRED);
607 goto drop_unlock;
608 }
609
610 spin_unlock(&x->lock);
611
612 if (xfrm_tunnel_check(skb, x, family)) {
613 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMODEERROR);
614 goto drop;
615 }
616
617 seq_hi = htonl(xfrm_replay_seqhi(x, seq));
618
619 XFRM_SKB_CB(skb)->seq.input.low = seq;
620 XFRM_SKB_CB(skb)->seq.input.hi = seq_hi;
621
622 dev_hold(skb->dev);
623
624 if (crypto_done)
625 nexthdr = x->type_offload->input_tail(x, skb);
626 else
627 nexthdr = x->type->input(x, skb);
628
629 if (nexthdr == -EINPROGRESS)
630 return 0;
631resume:
632 dev_put(skb->dev);
633
634 spin_lock(&x->lock);
635 if (nexthdr < 0) {
636 if (nexthdr == -EBADMSG) {
637 xfrm_audit_state_icvfail(x, skb,
638 x->type->proto);
639 x->stats.integrity_failed++;
640 }
641 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEPROTOERROR);
642 goto drop_unlock;
643 }
644
645 /* only the first xfrm gets the encap type */
646 encap_type = 0;
647
648 if (xfrm_replay_recheck(x, skb, seq)) {
649 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATESEQERROR);
650 goto drop_unlock;
651 }
652
653 xfrm_replay_advance(x, seq);
654
655 x->curlft.bytes += skb->len;
656 x->curlft.packets++;
657 x->lastused = ktime_get_real_seconds();
658
659 spin_unlock(&x->lock);
660
661 XFRM_MODE_SKB_CB(skb)->protocol = nexthdr;
662
663 if (xfrm_inner_mode_input(x, skb)) {
664 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMODEERROR);
665 goto drop;
666 }
667
668 if (x->outer_mode.flags & XFRM_MODE_FLAG_TUNNEL) {
669 decaps = 1;
670 break;
671 }
672
673 /*
674 * We need the inner address. However, we only get here for
675 * transport mode so the outer address is identical.
676 */
677 daddr = &x->id.daddr;
678 family = x->props.family;
679
680 err = xfrm_parse_spi(skb, nexthdr, &spi, &seq);
681 if (err < 0) {
682 XFRM_INC_STATS(net, LINUX_MIB_XFRMINHDRERROR);
683 goto drop;
684 }
685 crypto_done = false;
686 } while (!err);
687
688 err = xfrm_rcv_cb(skb, family, x->type->proto, 0);
689 if (err)
690 goto drop;
691
692 nf_reset_ct(skb);
693
694 if (decaps) {
695 sp = skb_sec_path(skb);
696 if (sp)
697 sp->olen = 0;
698 if (skb_valid_dst(skb))
699 skb_dst_drop(skb);
700 gro_cells_receive(&gro_cells, skb);
701 return 0;
702 } else {
703 xo = xfrm_offload(skb);
704 if (xo)
705 xfrm_gro = xo->flags & XFRM_GRO;
706
707 err = -EAFNOSUPPORT;
708 rcu_read_lock();
709 afinfo = xfrm_state_afinfo_get_rcu(x->props.family);
710 if (likely(afinfo))
711 err = afinfo->transport_finish(skb, xfrm_gro || async);
712 rcu_read_unlock();
713 if (xfrm_gro) {
714 sp = skb_sec_path(skb);
715 if (sp)
716 sp->olen = 0;
717 if (skb_valid_dst(skb))
718 skb_dst_drop(skb);
719 gro_cells_receive(&gro_cells, skb);
720 return err;
721 }
722
723 return err;
724 }
725
726drop_unlock:
727 spin_unlock(&x->lock);
728drop:
729 xfrm_rcv_cb(skb, family, x && x->type ? x->type->proto : nexthdr, -1);
730 kfree_skb(skb);
731 return 0;
732}
733EXPORT_SYMBOL(xfrm_input);
734
735int xfrm_input_resume(struct sk_buff *skb, int nexthdr)
736{
737 return xfrm_input(skb, nexthdr, 0, -1);
738}
739EXPORT_SYMBOL(xfrm_input_resume);
740
741static void xfrm_trans_reinject(struct work_struct *work)
742{
743 struct xfrm_trans_tasklet *trans = container_of(work, struct xfrm_trans_tasklet, work);
744 struct sk_buff_head queue;
745 struct sk_buff *skb;
746
747 __skb_queue_head_init(&queue);
748 spin_lock_bh(&trans->queue_lock);
749 skb_queue_splice_init(&trans->queue, &queue);
750 spin_unlock_bh(&trans->queue_lock);
751
752 local_bh_disable();
753 while ((skb = __skb_dequeue(&queue)))
754 XFRM_TRANS_SKB_CB(skb)->finish(XFRM_TRANS_SKB_CB(skb)->net,
755 NULL, skb);
756 local_bh_enable();
757}
758
759int xfrm_trans_queue_net(struct net *net, struct sk_buff *skb,
760 int (*finish)(struct net *, struct sock *,
761 struct sk_buff *))
762{
763 struct xfrm_trans_tasklet *trans;
764
765 trans = this_cpu_ptr(&xfrm_trans_tasklet);
766
767 if (skb_queue_len(&trans->queue) >= READ_ONCE(netdev_max_backlog))
768 return -ENOBUFS;
769
770 BUILD_BUG_ON(sizeof(struct xfrm_trans_cb) > sizeof(skb->cb));
771
772 XFRM_TRANS_SKB_CB(skb)->finish = finish;
773 XFRM_TRANS_SKB_CB(skb)->net = net;
774 spin_lock_bh(&trans->queue_lock);
775 __skb_queue_tail(&trans->queue, skb);
776 spin_unlock_bh(&trans->queue_lock);
777 schedule_work(&trans->work);
778 return 0;
779}
780EXPORT_SYMBOL(xfrm_trans_queue_net);
781
782int xfrm_trans_queue(struct sk_buff *skb,
783 int (*finish)(struct net *, struct sock *,
784 struct sk_buff *))
785{
786 return xfrm_trans_queue_net(dev_net(skb->dev), skb, finish);
787}
788EXPORT_SYMBOL(xfrm_trans_queue);
789
790void __init xfrm_input_init(void)
791{
792 int err;
793 int i;
794
795 init_dummy_netdev(&xfrm_napi_dev);
796 err = gro_cells_init(&gro_cells, &xfrm_napi_dev);
797 if (err)
798 gro_cells.cells = NULL;
799
800 for_each_possible_cpu(i) {
801 struct xfrm_trans_tasklet *trans;
802
803 trans = &per_cpu(xfrm_trans_tasklet, i);
804 spin_lock_init(&trans->queue_lock);
805 __skb_queue_head_init(&trans->queue);
806 INIT_WORK(&trans->work, xfrm_trans_reinject);
807 }
808}
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
19static DEFINE_SPINLOCK(xfrm_input_afinfo_lock);
20static struct xfrm_input_afinfo __rcu *xfrm_input_afinfo[NPROTO];
21
22int xfrm_input_register_afinfo(struct xfrm_input_afinfo *afinfo)
23{
24 int err = 0;
25
26 if (unlikely(afinfo == NULL))
27 return -EINVAL;
28 if (unlikely(afinfo->family >= NPROTO))
29 return -EAFNOSUPPORT;
30 spin_lock_bh(&xfrm_input_afinfo_lock);
31 if (unlikely(xfrm_input_afinfo[afinfo->family] != NULL))
32 err = -ENOBUFS;
33 else
34 rcu_assign_pointer(xfrm_input_afinfo[afinfo->family], afinfo);
35 spin_unlock_bh(&xfrm_input_afinfo_lock);
36 return err;
37}
38EXPORT_SYMBOL(xfrm_input_register_afinfo);
39
40int xfrm_input_unregister_afinfo(struct xfrm_input_afinfo *afinfo)
41{
42 int err = 0;
43
44 if (unlikely(afinfo == NULL))
45 return -EINVAL;
46 if (unlikely(afinfo->family >= NPROTO))
47 return -EAFNOSUPPORT;
48 spin_lock_bh(&xfrm_input_afinfo_lock);
49 if (likely(xfrm_input_afinfo[afinfo->family] != NULL)) {
50 if (unlikely(xfrm_input_afinfo[afinfo->family] != afinfo))
51 err = -EINVAL;
52 else
53 RCU_INIT_POINTER(xfrm_input_afinfo[afinfo->family], NULL);
54 }
55 spin_unlock_bh(&xfrm_input_afinfo_lock);
56 synchronize_rcu();
57 return err;
58}
59EXPORT_SYMBOL(xfrm_input_unregister_afinfo);
60
61static struct xfrm_input_afinfo *xfrm_input_get_afinfo(unsigned int family)
62{
63 struct xfrm_input_afinfo *afinfo;
64
65 if (unlikely(family >= NPROTO))
66 return NULL;
67 rcu_read_lock();
68 afinfo = rcu_dereference(xfrm_input_afinfo[family]);
69 if (unlikely(!afinfo))
70 rcu_read_unlock();
71 return afinfo;
72}
73
74static void xfrm_input_put_afinfo(struct xfrm_input_afinfo *afinfo)
75{
76 rcu_read_unlock();
77}
78
79static int xfrm_rcv_cb(struct sk_buff *skb, unsigned int family, u8 protocol,
80 int err)
81{
82 int ret;
83 struct xfrm_input_afinfo *afinfo = xfrm_input_get_afinfo(family);
84
85 if (!afinfo)
86 return -EAFNOSUPPORT;
87
88 ret = afinfo->callback(skb, protocol, err);
89 xfrm_input_put_afinfo(afinfo);
90
91 return ret;
92}
93
94void __secpath_destroy(struct sec_path *sp)
95{
96 int i;
97 for (i = 0; i < sp->len; i++)
98 xfrm_state_put(sp->xvec[i]);
99 kmem_cache_free(secpath_cachep, sp);
100}
101EXPORT_SYMBOL(__secpath_destroy);
102
103struct sec_path *secpath_dup(struct sec_path *src)
104{
105 struct sec_path *sp;
106
107 sp = kmem_cache_alloc(secpath_cachep, GFP_ATOMIC);
108 if (!sp)
109 return NULL;
110
111 sp->len = 0;
112 if (src) {
113 int i;
114
115 memcpy(sp, src, sizeof(*sp));
116 for (i = 0; i < sp->len; i++)
117 xfrm_state_hold(sp->xvec[i]);
118 }
119 atomic_set(&sp->refcnt, 1);
120 return sp;
121}
122EXPORT_SYMBOL(secpath_dup);
123
124/* Fetch spi and seq from ipsec header */
125
126int xfrm_parse_spi(struct sk_buff *skb, u8 nexthdr, __be32 *spi, __be32 *seq)
127{
128 int offset, offset_seq;
129 int hlen;
130
131 switch (nexthdr) {
132 case IPPROTO_AH:
133 hlen = sizeof(struct ip_auth_hdr);
134 offset = offsetof(struct ip_auth_hdr, spi);
135 offset_seq = offsetof(struct ip_auth_hdr, seq_no);
136 break;
137 case IPPROTO_ESP:
138 hlen = sizeof(struct ip_esp_hdr);
139 offset = offsetof(struct ip_esp_hdr, spi);
140 offset_seq = offsetof(struct ip_esp_hdr, seq_no);
141 break;
142 case IPPROTO_COMP:
143 if (!pskb_may_pull(skb, sizeof(struct ip_comp_hdr)))
144 return -EINVAL;
145 *spi = htonl(ntohs(*(__be16 *)(skb_transport_header(skb) + 2)));
146 *seq = 0;
147 return 0;
148 default:
149 return 1;
150 }
151
152 if (!pskb_may_pull(skb, hlen))
153 return -EINVAL;
154
155 *spi = *(__be32 *)(skb_transport_header(skb) + offset);
156 *seq = *(__be32 *)(skb_transport_header(skb) + offset_seq);
157 return 0;
158}
159
160int xfrm_prepare_input(struct xfrm_state *x, struct sk_buff *skb)
161{
162 struct xfrm_mode *inner_mode = x->inner_mode;
163 int err;
164
165 err = x->outer_mode->afinfo->extract_input(x, skb);
166 if (err)
167 return err;
168
169 if (x->sel.family == AF_UNSPEC) {
170 inner_mode = xfrm_ip2inner_mode(x, XFRM_MODE_SKB_CB(skb)->protocol);
171 if (inner_mode == NULL)
172 return -EAFNOSUPPORT;
173 }
174
175 skb->protocol = inner_mode->afinfo->eth_proto;
176 return inner_mode->input2(x, skb);
177}
178EXPORT_SYMBOL(xfrm_prepare_input);
179
180int xfrm_input(struct sk_buff *skb, int nexthdr, __be32 spi, int encap_type)
181{
182 struct net *net = dev_net(skb->dev);
183 int err;
184 __be32 seq;
185 __be32 seq_hi;
186 struct xfrm_state *x = NULL;
187 xfrm_address_t *daddr;
188 struct xfrm_mode *inner_mode;
189 unsigned int family;
190 int decaps = 0;
191 int async = 0;
192
193 /* A negative encap_type indicates async resumption. */
194 if (encap_type < 0) {
195 async = 1;
196 x = xfrm_input_state(skb);
197 seq = XFRM_SKB_CB(skb)->seq.input.low;
198 family = x->outer_mode->afinfo->family;
199 goto resume;
200 }
201
202 daddr = (xfrm_address_t *)(skb_network_header(skb) +
203 XFRM_SPI_SKB_CB(skb)->daddroff);
204 family = XFRM_SPI_SKB_CB(skb)->family;
205
206 /* Allocate new secpath or COW existing one. */
207 if (!skb->sp || atomic_read(&skb->sp->refcnt) != 1) {
208 struct sec_path *sp;
209
210 sp = secpath_dup(skb->sp);
211 if (!sp) {
212 XFRM_INC_STATS(net, LINUX_MIB_XFRMINERROR);
213 goto drop;
214 }
215 if (skb->sp)
216 secpath_put(skb->sp);
217 skb->sp = sp;
218 }
219
220 seq = 0;
221 if (!spi && (err = xfrm_parse_spi(skb, nexthdr, &spi, &seq)) != 0) {
222 XFRM_INC_STATS(net, LINUX_MIB_XFRMINHDRERROR);
223 goto drop;
224 }
225
226 do {
227 if (skb->sp->len == XFRM_MAX_DEPTH) {
228 XFRM_INC_STATS(net, LINUX_MIB_XFRMINBUFFERERROR);
229 goto drop;
230 }
231
232 x = xfrm_state_lookup(net, skb->mark, daddr, spi, nexthdr, family);
233 if (x == NULL) {
234 XFRM_INC_STATS(net, LINUX_MIB_XFRMINNOSTATES);
235 xfrm_audit_state_notfound(skb, family, spi, seq);
236 goto drop;
237 }
238
239 skb->sp->xvec[skb->sp->len++] = x;
240
241 if (xfrm_tunnel_check(skb, x, family)) {
242 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMODEERROR);
243 goto drop;
244 }
245
246 spin_lock(&x->lock);
247 if (unlikely(x->km.state == XFRM_STATE_ACQ)) {
248 XFRM_INC_STATS(net, LINUX_MIB_XFRMACQUIREERROR);
249 goto drop_unlock;
250 }
251
252 if (unlikely(x->km.state != XFRM_STATE_VALID)) {
253 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEINVALID);
254 goto drop_unlock;
255 }
256
257 if ((x->encap ? x->encap->encap_type : 0) != encap_type) {
258 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMISMATCH);
259 goto drop_unlock;
260 }
261
262 if (x->repl->check(x, skb, seq)) {
263 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATESEQERROR);
264 goto drop_unlock;
265 }
266
267 if (xfrm_state_check_expire(x)) {
268 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEEXPIRED);
269 goto drop_unlock;
270 }
271
272 spin_unlock(&x->lock);
273
274 seq_hi = htonl(xfrm_replay_seqhi(x, seq));
275
276 XFRM_SKB_CB(skb)->seq.input.low = seq;
277 XFRM_SKB_CB(skb)->seq.input.hi = seq_hi;
278
279 skb_dst_force(skb);
280
281 nexthdr = x->type->input(x, skb);
282
283 if (nexthdr == -EINPROGRESS)
284 return 0;
285resume:
286 spin_lock(&x->lock);
287 if (nexthdr <= 0) {
288 if (nexthdr == -EBADMSG) {
289 xfrm_audit_state_icvfail(x, skb,
290 x->type->proto);
291 x->stats.integrity_failed++;
292 }
293 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEPROTOERROR);
294 goto drop_unlock;
295 }
296
297 /* only the first xfrm gets the encap type */
298 encap_type = 0;
299
300 if (async && x->repl->recheck(x, skb, seq)) {
301 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATESEQERROR);
302 goto drop_unlock;
303 }
304
305 x->repl->advance(x, seq);
306
307 x->curlft.bytes += skb->len;
308 x->curlft.packets++;
309
310 spin_unlock(&x->lock);
311
312 XFRM_MODE_SKB_CB(skb)->protocol = nexthdr;
313
314 inner_mode = x->inner_mode;
315
316 if (x->sel.family == AF_UNSPEC) {
317 inner_mode = xfrm_ip2inner_mode(x, XFRM_MODE_SKB_CB(skb)->protocol);
318 if (inner_mode == NULL)
319 goto drop;
320 }
321
322 if (inner_mode->input(x, skb)) {
323 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMODEERROR);
324 goto drop;
325 }
326
327 if (x->outer_mode->flags & XFRM_MODE_FLAG_TUNNEL) {
328 decaps = 1;
329 break;
330 }
331
332 /*
333 * We need the inner address. However, we only get here for
334 * transport mode so the outer address is identical.
335 */
336 daddr = &x->id.daddr;
337 family = x->outer_mode->afinfo->family;
338
339 err = xfrm_parse_spi(skb, nexthdr, &spi, &seq);
340 if (err < 0) {
341 XFRM_INC_STATS(net, LINUX_MIB_XFRMINHDRERROR);
342 goto drop;
343 }
344 } while (!err);
345
346 err = xfrm_rcv_cb(skb, family, x->type->proto, 0);
347 if (err)
348 goto drop;
349
350 nf_reset(skb);
351
352 if (decaps) {
353 skb_dst_drop(skb);
354 netif_rx(skb);
355 return 0;
356 } else {
357 return x->inner_mode->afinfo->transport_finish(skb, async);
358 }
359
360drop_unlock:
361 spin_unlock(&x->lock);
362drop:
363 xfrm_rcv_cb(skb, family, x && x->type ? x->type->proto : nexthdr, -1);
364 kfree_skb(skb);
365 return 0;
366}
367EXPORT_SYMBOL(xfrm_input);
368
369int xfrm_input_resume(struct sk_buff *skb, int nexthdr)
370{
371 return xfrm_input(skb, nexthdr, 0, -1);
372}
373EXPORT_SYMBOL(xfrm_input_resume);
374
375void __init xfrm_input_init(void)
376{
377 secpath_cachep = kmem_cache_create("secpath_cache",
378 sizeof(struct sec_path),
379 0, SLAB_HWCACHE_ALIGN|SLAB_PANIC,
380 NULL);
381}