Loading...
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * xfrm_input.c
4 *
5 * Changes:
6 * YOSHIFUJI Hideaki @USAGI
7 * Split up af-specific portion
8 *
9 */
10
11#include <linux/bottom_half.h>
12#include <linux/cache.h>
13#include <linux/interrupt.h>
14#include <linux/slab.h>
15#include <linux/module.h>
16#include <linux/netdevice.h>
17#include <linux/percpu.h>
18#include <net/dst.h>
19#include <net/ip.h>
20#include <net/xfrm.h>
21#include <net/ip_tunnels.h>
22#include <net/ip6_tunnel.h>
23
24#include "xfrm_inout.h"
25
26struct xfrm_trans_tasklet {
27 struct tasklet_struct tasklet;
28 struct sk_buff_head queue;
29};
30
31struct xfrm_trans_cb {
32 union {
33 struct inet_skb_parm h4;
34#if IS_ENABLED(CONFIG_IPV6)
35 struct inet6_skb_parm h6;
36#endif
37 } header;
38 int (*finish)(struct net *net, struct sock *sk, struct sk_buff *skb);
39 struct net *net;
40};
41
42#define XFRM_TRANS_SKB_CB(__skb) ((struct xfrm_trans_cb *)&((__skb)->cb[0]))
43
44static DEFINE_SPINLOCK(xfrm_input_afinfo_lock);
45static struct xfrm_input_afinfo const __rcu *xfrm_input_afinfo[2][AF_INET6 + 1];
46
47static struct gro_cells gro_cells;
48static struct net_device xfrm_napi_dev;
49
50static DEFINE_PER_CPU(struct xfrm_trans_tasklet, xfrm_trans_tasklet);
51
52int xfrm_input_register_afinfo(const struct xfrm_input_afinfo *afinfo)
53{
54 int err = 0;
55
56 if (WARN_ON(afinfo->family > AF_INET6))
57 return -EAFNOSUPPORT;
58
59 spin_lock_bh(&xfrm_input_afinfo_lock);
60 if (unlikely(xfrm_input_afinfo[afinfo->is_ipip][afinfo->family]))
61 err = -EEXIST;
62 else
63 rcu_assign_pointer(xfrm_input_afinfo[afinfo->is_ipip][afinfo->family], afinfo);
64 spin_unlock_bh(&xfrm_input_afinfo_lock);
65 return err;
66}
67EXPORT_SYMBOL(xfrm_input_register_afinfo);
68
69int xfrm_input_unregister_afinfo(const struct xfrm_input_afinfo *afinfo)
70{
71 int err = 0;
72
73 spin_lock_bh(&xfrm_input_afinfo_lock);
74 if (likely(xfrm_input_afinfo[afinfo->is_ipip][afinfo->family])) {
75 if (unlikely(xfrm_input_afinfo[afinfo->is_ipip][afinfo->family] != afinfo))
76 err = -EINVAL;
77 else
78 RCU_INIT_POINTER(xfrm_input_afinfo[afinfo->is_ipip][afinfo->family], NULL);
79 }
80 spin_unlock_bh(&xfrm_input_afinfo_lock);
81 synchronize_rcu();
82 return err;
83}
84EXPORT_SYMBOL(xfrm_input_unregister_afinfo);
85
86static const struct xfrm_input_afinfo *xfrm_input_get_afinfo(u8 family, bool is_ipip)
87{
88 const struct xfrm_input_afinfo *afinfo;
89
90 if (WARN_ON_ONCE(family > AF_INET6))
91 return NULL;
92
93 rcu_read_lock();
94 afinfo = rcu_dereference(xfrm_input_afinfo[is_ipip][family]);
95 if (unlikely(!afinfo))
96 rcu_read_unlock();
97 return afinfo;
98}
99
100static int xfrm_rcv_cb(struct sk_buff *skb, unsigned int family, u8 protocol,
101 int err)
102{
103 bool is_ipip = (protocol == IPPROTO_IPIP || protocol == IPPROTO_IPV6);
104 const struct xfrm_input_afinfo *afinfo;
105 int ret;
106
107 afinfo = xfrm_input_get_afinfo(family, is_ipip);
108 if (!afinfo)
109 return -EAFNOSUPPORT;
110
111 ret = afinfo->callback(skb, protocol, err);
112 rcu_read_unlock();
113
114 return ret;
115}
116
117struct sec_path *secpath_set(struct sk_buff *skb)
118{
119 struct sec_path *sp, *tmp = skb_ext_find(skb, SKB_EXT_SEC_PATH);
120
121 sp = skb_ext_add(skb, SKB_EXT_SEC_PATH);
122 if (!sp)
123 return NULL;
124
125 if (tmp) /* reused existing one (was COW'd if needed) */
126 return sp;
127
128 /* allocated new secpath */
129 memset(sp->ovec, 0, sizeof(sp->ovec));
130 sp->olen = 0;
131 sp->len = 0;
132
133 return sp;
134}
135EXPORT_SYMBOL(secpath_set);
136
137/* Fetch spi and seq from ipsec header */
138
139int xfrm_parse_spi(struct sk_buff *skb, u8 nexthdr, __be32 *spi, __be32 *seq)
140{
141 int offset, offset_seq;
142 int hlen;
143
144 switch (nexthdr) {
145 case IPPROTO_AH:
146 hlen = sizeof(struct ip_auth_hdr);
147 offset = offsetof(struct ip_auth_hdr, spi);
148 offset_seq = offsetof(struct ip_auth_hdr, seq_no);
149 break;
150 case IPPROTO_ESP:
151 hlen = sizeof(struct ip_esp_hdr);
152 offset = offsetof(struct ip_esp_hdr, spi);
153 offset_seq = offsetof(struct ip_esp_hdr, seq_no);
154 break;
155 case IPPROTO_COMP:
156 if (!pskb_may_pull(skb, sizeof(struct ip_comp_hdr)))
157 return -EINVAL;
158 *spi = htonl(ntohs(*(__be16 *)(skb_transport_header(skb) + 2)));
159 *seq = 0;
160 return 0;
161 default:
162 return 1;
163 }
164
165 if (!pskb_may_pull(skb, hlen))
166 return -EINVAL;
167
168 *spi = *(__be32 *)(skb_transport_header(skb) + offset);
169 *seq = *(__be32 *)(skb_transport_header(skb) + offset_seq);
170 return 0;
171}
172EXPORT_SYMBOL(xfrm_parse_spi);
173
174static int xfrm4_remove_beet_encap(struct xfrm_state *x, struct sk_buff *skb)
175{
176 struct iphdr *iph;
177 int optlen = 0;
178 int err = -EINVAL;
179
180 if (unlikely(XFRM_MODE_SKB_CB(skb)->protocol == IPPROTO_BEETPH)) {
181 struct ip_beet_phdr *ph;
182 int phlen;
183
184 if (!pskb_may_pull(skb, sizeof(*ph)))
185 goto out;
186
187 ph = (struct ip_beet_phdr *)skb->data;
188
189 phlen = sizeof(*ph) + ph->padlen;
190 optlen = ph->hdrlen * 8 + (IPV4_BEET_PHMAXLEN - phlen);
191 if (optlen < 0 || optlen & 3 || optlen > 250)
192 goto out;
193
194 XFRM_MODE_SKB_CB(skb)->protocol = ph->nexthdr;
195
196 if (!pskb_may_pull(skb, phlen))
197 goto out;
198 __skb_pull(skb, phlen);
199 }
200
201 skb_push(skb, sizeof(*iph));
202 skb_reset_network_header(skb);
203 skb_mac_header_rebuild(skb);
204
205 xfrm4_beet_make_header(skb);
206
207 iph = ip_hdr(skb);
208
209 iph->ihl += optlen / 4;
210 iph->tot_len = htons(skb->len);
211 iph->daddr = x->sel.daddr.a4;
212 iph->saddr = x->sel.saddr.a4;
213 iph->check = 0;
214 iph->check = ip_fast_csum(skb_network_header(skb), iph->ihl);
215 err = 0;
216out:
217 return err;
218}
219
220static void ipip_ecn_decapsulate(struct sk_buff *skb)
221{
222 struct iphdr *inner_iph = ipip_hdr(skb);
223
224 if (INET_ECN_is_ce(XFRM_MODE_SKB_CB(skb)->tos))
225 IP_ECN_set_ce(inner_iph);
226}
227
228static int xfrm4_remove_tunnel_encap(struct xfrm_state *x, struct sk_buff *skb)
229{
230 int err = -EINVAL;
231
232 if (XFRM_MODE_SKB_CB(skb)->protocol != IPPROTO_IPIP)
233 goto out;
234
235 if (!pskb_may_pull(skb, sizeof(struct iphdr)))
236 goto out;
237
238 err = skb_unclone(skb, GFP_ATOMIC);
239 if (err)
240 goto out;
241
242 if (x->props.flags & XFRM_STATE_DECAP_DSCP)
243 ipv4_copy_dscp(XFRM_MODE_SKB_CB(skb)->tos, ipip_hdr(skb));
244 if (!(x->props.flags & XFRM_STATE_NOECN))
245 ipip_ecn_decapsulate(skb);
246
247 skb_reset_network_header(skb);
248 skb_mac_header_rebuild(skb);
249 if (skb->mac_len)
250 eth_hdr(skb)->h_proto = skb->protocol;
251
252 err = 0;
253
254out:
255 return err;
256}
257
258static void ipip6_ecn_decapsulate(struct sk_buff *skb)
259{
260 struct ipv6hdr *inner_iph = ipipv6_hdr(skb);
261
262 if (INET_ECN_is_ce(XFRM_MODE_SKB_CB(skb)->tos))
263 IP6_ECN_set_ce(skb, inner_iph);
264}
265
266static int xfrm6_remove_tunnel_encap(struct xfrm_state *x, struct sk_buff *skb)
267{
268 int err = -EINVAL;
269
270 if (XFRM_MODE_SKB_CB(skb)->protocol != IPPROTO_IPV6)
271 goto out;
272 if (!pskb_may_pull(skb, sizeof(struct ipv6hdr)))
273 goto out;
274
275 err = skb_unclone(skb, GFP_ATOMIC);
276 if (err)
277 goto out;
278
279 if (x->props.flags & XFRM_STATE_DECAP_DSCP)
280 ipv6_copy_dscp(ipv6_get_dsfield(ipv6_hdr(skb)),
281 ipipv6_hdr(skb));
282 if (!(x->props.flags & XFRM_STATE_NOECN))
283 ipip6_ecn_decapsulate(skb);
284
285 skb_reset_network_header(skb);
286 skb_mac_header_rebuild(skb);
287 if (skb->mac_len)
288 eth_hdr(skb)->h_proto = skb->protocol;
289
290 err = 0;
291
292out:
293 return err;
294}
295
296static int xfrm6_remove_beet_encap(struct xfrm_state *x, struct sk_buff *skb)
297{
298 struct ipv6hdr *ip6h;
299 int size = sizeof(struct ipv6hdr);
300 int err;
301
302 err = skb_cow_head(skb, size + skb->mac_len);
303 if (err)
304 goto out;
305
306 __skb_push(skb, size);
307 skb_reset_network_header(skb);
308 skb_mac_header_rebuild(skb);
309
310 xfrm6_beet_make_header(skb);
311
312 ip6h = ipv6_hdr(skb);
313 ip6h->payload_len = htons(skb->len - size);
314 ip6h->daddr = x->sel.daddr.in6;
315 ip6h->saddr = x->sel.saddr.in6;
316 err = 0;
317out:
318 return err;
319}
320
321/* Remove encapsulation header.
322 *
323 * The IP header will be moved over the top of the encapsulation
324 * header.
325 *
326 * On entry, the transport header shall point to where the IP header
327 * should be and the network header shall be set to where the IP
328 * header currently is. skb->data shall point to the start of the
329 * payload.
330 */
331static int
332xfrm_inner_mode_encap_remove(struct xfrm_state *x,
333 const struct xfrm_mode *inner_mode,
334 struct sk_buff *skb)
335{
336 switch (inner_mode->encap) {
337 case XFRM_MODE_BEET:
338 if (inner_mode->family == AF_INET)
339 return xfrm4_remove_beet_encap(x, skb);
340 if (inner_mode->family == AF_INET6)
341 return xfrm6_remove_beet_encap(x, skb);
342 break;
343 case XFRM_MODE_TUNNEL:
344 if (inner_mode->family == AF_INET)
345 return xfrm4_remove_tunnel_encap(x, skb);
346 if (inner_mode->family == AF_INET6)
347 return xfrm6_remove_tunnel_encap(x, skb);
348 break;
349 }
350
351 WARN_ON_ONCE(1);
352 return -EOPNOTSUPP;
353}
354
355static int xfrm_prepare_input(struct xfrm_state *x, struct sk_buff *skb)
356{
357 const struct xfrm_mode *inner_mode = &x->inner_mode;
358
359 switch (x->outer_mode.family) {
360 case AF_INET:
361 xfrm4_extract_header(skb);
362 break;
363 case AF_INET6:
364 xfrm6_extract_header(skb);
365 break;
366 default:
367 WARN_ON_ONCE(1);
368 return -EAFNOSUPPORT;
369 }
370
371 if (x->sel.family == AF_UNSPEC) {
372 inner_mode = xfrm_ip2inner_mode(x, XFRM_MODE_SKB_CB(skb)->protocol);
373 if (!inner_mode)
374 return -EAFNOSUPPORT;
375 }
376
377 switch (inner_mode->family) {
378 case AF_INET:
379 skb->protocol = htons(ETH_P_IP);
380 break;
381 case AF_INET6:
382 skb->protocol = htons(ETH_P_IPV6);
383 break;
384 default:
385 WARN_ON_ONCE(1);
386 break;
387 }
388
389 return xfrm_inner_mode_encap_remove(x, inner_mode, skb);
390}
391
392/* Remove encapsulation header.
393 *
394 * The IP header will be moved over the top of the encapsulation header.
395 *
396 * On entry, skb_transport_header() shall point to where the IP header
397 * should be and skb_network_header() shall be set to where the IP header
398 * currently is. skb->data shall point to the start of the payload.
399 */
400static int xfrm4_transport_input(struct xfrm_state *x, struct sk_buff *skb)
401{
402 int ihl = skb->data - skb_transport_header(skb);
403
404 if (skb->transport_header != skb->network_header) {
405 memmove(skb_transport_header(skb),
406 skb_network_header(skb), ihl);
407 skb->network_header = skb->transport_header;
408 }
409 ip_hdr(skb)->tot_len = htons(skb->len + ihl);
410 skb_reset_transport_header(skb);
411 return 0;
412}
413
414static int xfrm6_transport_input(struct xfrm_state *x, struct sk_buff *skb)
415{
416#if IS_ENABLED(CONFIG_IPV6)
417 int ihl = skb->data - skb_transport_header(skb);
418
419 if (skb->transport_header != skb->network_header) {
420 memmove(skb_transport_header(skb),
421 skb_network_header(skb), ihl);
422 skb->network_header = skb->transport_header;
423 }
424 ipv6_hdr(skb)->payload_len = htons(skb->len + ihl -
425 sizeof(struct ipv6hdr));
426 skb_reset_transport_header(skb);
427 return 0;
428#else
429 WARN_ON_ONCE(1);
430 return -EAFNOSUPPORT;
431#endif
432}
433
434static int xfrm_inner_mode_input(struct xfrm_state *x,
435 const struct xfrm_mode *inner_mode,
436 struct sk_buff *skb)
437{
438 switch (inner_mode->encap) {
439 case XFRM_MODE_BEET:
440 case XFRM_MODE_TUNNEL:
441 return xfrm_prepare_input(x, skb);
442 case XFRM_MODE_TRANSPORT:
443 if (inner_mode->family == AF_INET)
444 return xfrm4_transport_input(x, skb);
445 if (inner_mode->family == AF_INET6)
446 return xfrm6_transport_input(x, skb);
447 break;
448 case XFRM_MODE_ROUTEOPTIMIZATION:
449 WARN_ON_ONCE(1);
450 break;
451 default:
452 WARN_ON_ONCE(1);
453 break;
454 }
455
456 return -EOPNOTSUPP;
457}
458
459int xfrm_input(struct sk_buff *skb, int nexthdr, __be32 spi, int encap_type)
460{
461 const struct xfrm_state_afinfo *afinfo;
462 struct net *net = dev_net(skb->dev);
463 const struct xfrm_mode *inner_mode;
464 int err;
465 __be32 seq;
466 __be32 seq_hi;
467 struct xfrm_state *x = NULL;
468 xfrm_address_t *daddr;
469 u32 mark = skb->mark;
470 unsigned int family = AF_UNSPEC;
471 int decaps = 0;
472 int async = 0;
473 bool xfrm_gro = false;
474 bool crypto_done = false;
475 struct xfrm_offload *xo = xfrm_offload(skb);
476 struct sec_path *sp;
477
478 if (encap_type < 0) {
479 x = xfrm_input_state(skb);
480
481 if (unlikely(x->km.state != XFRM_STATE_VALID)) {
482 if (x->km.state == XFRM_STATE_ACQ)
483 XFRM_INC_STATS(net, LINUX_MIB_XFRMACQUIREERROR);
484 else
485 XFRM_INC_STATS(net,
486 LINUX_MIB_XFRMINSTATEINVALID);
487
488 if (encap_type == -1)
489 dev_put(skb->dev);
490 goto drop;
491 }
492
493 family = x->outer_mode.family;
494
495 /* An encap_type of -1 indicates async resumption. */
496 if (encap_type == -1) {
497 async = 1;
498 seq = XFRM_SKB_CB(skb)->seq.input.low;
499 goto resume;
500 }
501
502 /* encap_type < -1 indicates a GRO call. */
503 encap_type = 0;
504 seq = XFRM_SPI_SKB_CB(skb)->seq;
505
506 if (xo && (xo->flags & CRYPTO_DONE)) {
507 crypto_done = true;
508 family = XFRM_SPI_SKB_CB(skb)->family;
509
510 if (!(xo->status & CRYPTO_SUCCESS)) {
511 if (xo->status &
512 (CRYPTO_TRANSPORT_AH_AUTH_FAILED |
513 CRYPTO_TRANSPORT_ESP_AUTH_FAILED |
514 CRYPTO_TUNNEL_AH_AUTH_FAILED |
515 CRYPTO_TUNNEL_ESP_AUTH_FAILED)) {
516
517 xfrm_audit_state_icvfail(x, skb,
518 x->type->proto);
519 x->stats.integrity_failed++;
520 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEPROTOERROR);
521 goto drop;
522 }
523
524 if (xo->status & CRYPTO_INVALID_PROTOCOL) {
525 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEPROTOERROR);
526 goto drop;
527 }
528
529 XFRM_INC_STATS(net, LINUX_MIB_XFRMINBUFFERERROR);
530 goto drop;
531 }
532
533 if ((err = xfrm_parse_spi(skb, nexthdr, &spi, &seq)) != 0) {
534 XFRM_INC_STATS(net, LINUX_MIB_XFRMINHDRERROR);
535 goto drop;
536 }
537 }
538
539 goto lock;
540 }
541
542 family = XFRM_SPI_SKB_CB(skb)->family;
543
544 /* if tunnel is present override skb->mark value with tunnel i_key */
545 switch (family) {
546 case AF_INET:
547 if (XFRM_TUNNEL_SKB_CB(skb)->tunnel.ip4)
548 mark = be32_to_cpu(XFRM_TUNNEL_SKB_CB(skb)->tunnel.ip4->parms.i_key);
549 break;
550 case AF_INET6:
551 if (XFRM_TUNNEL_SKB_CB(skb)->tunnel.ip6)
552 mark = be32_to_cpu(XFRM_TUNNEL_SKB_CB(skb)->tunnel.ip6->parms.i_key);
553 break;
554 }
555
556 sp = secpath_set(skb);
557 if (!sp) {
558 XFRM_INC_STATS(net, LINUX_MIB_XFRMINERROR);
559 goto drop;
560 }
561
562 seq = 0;
563 if (!spi && (err = xfrm_parse_spi(skb, nexthdr, &spi, &seq)) != 0) {
564 secpath_reset(skb);
565 XFRM_INC_STATS(net, LINUX_MIB_XFRMINHDRERROR);
566 goto drop;
567 }
568
569 daddr = (xfrm_address_t *)(skb_network_header(skb) +
570 XFRM_SPI_SKB_CB(skb)->daddroff);
571 do {
572 sp = skb_sec_path(skb);
573
574 if (sp->len == XFRM_MAX_DEPTH) {
575 secpath_reset(skb);
576 XFRM_INC_STATS(net, LINUX_MIB_XFRMINBUFFERERROR);
577 goto drop;
578 }
579
580 x = xfrm_state_lookup(net, mark, daddr, spi, nexthdr, family);
581 if (x == NULL) {
582 secpath_reset(skb);
583 XFRM_INC_STATS(net, LINUX_MIB_XFRMINNOSTATES);
584 xfrm_audit_state_notfound(skb, family, spi, seq);
585 goto drop;
586 }
587
588 skb->mark = xfrm_smark_get(skb->mark, x);
589
590 sp->xvec[sp->len++] = x;
591
592 skb_dst_force(skb);
593 if (!skb_dst(skb)) {
594 XFRM_INC_STATS(net, LINUX_MIB_XFRMINERROR);
595 goto drop;
596 }
597
598lock:
599 spin_lock(&x->lock);
600
601 if (unlikely(x->km.state != XFRM_STATE_VALID)) {
602 if (x->km.state == XFRM_STATE_ACQ)
603 XFRM_INC_STATS(net, LINUX_MIB_XFRMACQUIREERROR);
604 else
605 XFRM_INC_STATS(net,
606 LINUX_MIB_XFRMINSTATEINVALID);
607 goto drop_unlock;
608 }
609
610 if ((x->encap ? x->encap->encap_type : 0) != encap_type) {
611 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMISMATCH);
612 goto drop_unlock;
613 }
614
615 if (x->repl->check(x, skb, seq)) {
616 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATESEQERROR);
617 goto drop_unlock;
618 }
619
620 if (xfrm_state_check_expire(x)) {
621 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEEXPIRED);
622 goto drop_unlock;
623 }
624
625 spin_unlock(&x->lock);
626
627 if (xfrm_tunnel_check(skb, x, family)) {
628 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMODEERROR);
629 goto drop;
630 }
631
632 seq_hi = htonl(xfrm_replay_seqhi(x, seq));
633
634 XFRM_SKB_CB(skb)->seq.input.low = seq;
635 XFRM_SKB_CB(skb)->seq.input.hi = seq_hi;
636
637 dev_hold(skb->dev);
638
639 if (crypto_done)
640 nexthdr = x->type_offload->input_tail(x, skb);
641 else
642 nexthdr = x->type->input(x, skb);
643
644 if (nexthdr == -EINPROGRESS)
645 return 0;
646resume:
647 dev_put(skb->dev);
648
649 spin_lock(&x->lock);
650 if (nexthdr < 0) {
651 if (nexthdr == -EBADMSG) {
652 xfrm_audit_state_icvfail(x, skb,
653 x->type->proto);
654 x->stats.integrity_failed++;
655 }
656 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEPROTOERROR);
657 goto drop_unlock;
658 }
659
660 /* only the first xfrm gets the encap type */
661 encap_type = 0;
662
663 if (async && x->repl->recheck(x, skb, seq)) {
664 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATESEQERROR);
665 goto drop_unlock;
666 }
667
668 x->repl->advance(x, seq);
669
670 x->curlft.bytes += skb->len;
671 x->curlft.packets++;
672
673 spin_unlock(&x->lock);
674
675 XFRM_MODE_SKB_CB(skb)->protocol = nexthdr;
676
677 inner_mode = &x->inner_mode;
678
679 if (x->sel.family == AF_UNSPEC) {
680 inner_mode = xfrm_ip2inner_mode(x, XFRM_MODE_SKB_CB(skb)->protocol);
681 if (inner_mode == NULL) {
682 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMODEERROR);
683 goto drop;
684 }
685 }
686
687 if (xfrm_inner_mode_input(x, inner_mode, skb)) {
688 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMODEERROR);
689 goto drop;
690 }
691
692 if (x->outer_mode.flags & XFRM_MODE_FLAG_TUNNEL) {
693 decaps = 1;
694 break;
695 }
696
697 /*
698 * We need the inner address. However, we only get here for
699 * transport mode so the outer address is identical.
700 */
701 daddr = &x->id.daddr;
702 family = x->outer_mode.family;
703
704 err = xfrm_parse_spi(skb, nexthdr, &spi, &seq);
705 if (err < 0) {
706 XFRM_INC_STATS(net, LINUX_MIB_XFRMINHDRERROR);
707 goto drop;
708 }
709 crypto_done = false;
710 } while (!err);
711
712 err = xfrm_rcv_cb(skb, family, x->type->proto, 0);
713 if (err)
714 goto drop;
715
716 nf_reset_ct(skb);
717
718 if (decaps) {
719 sp = skb_sec_path(skb);
720 if (sp)
721 sp->olen = 0;
722 skb_dst_drop(skb);
723 gro_cells_receive(&gro_cells, skb);
724 return 0;
725 } else {
726 xo = xfrm_offload(skb);
727 if (xo)
728 xfrm_gro = xo->flags & XFRM_GRO;
729
730 err = -EAFNOSUPPORT;
731 rcu_read_lock();
732 afinfo = xfrm_state_afinfo_get_rcu(x->inner_mode.family);
733 if (likely(afinfo))
734 err = afinfo->transport_finish(skb, xfrm_gro || async);
735 rcu_read_unlock();
736 if (xfrm_gro) {
737 sp = skb_sec_path(skb);
738 if (sp)
739 sp->olen = 0;
740 skb_dst_drop(skb);
741 gro_cells_receive(&gro_cells, skb);
742 return err;
743 }
744
745 return err;
746 }
747
748drop_unlock:
749 spin_unlock(&x->lock);
750drop:
751 xfrm_rcv_cb(skb, family, x && x->type ? x->type->proto : nexthdr, -1);
752 kfree_skb(skb);
753 return 0;
754}
755EXPORT_SYMBOL(xfrm_input);
756
757int xfrm_input_resume(struct sk_buff *skb, int nexthdr)
758{
759 return xfrm_input(skb, nexthdr, 0, -1);
760}
761EXPORT_SYMBOL(xfrm_input_resume);
762
763static void xfrm_trans_reinject(unsigned long data)
764{
765 struct xfrm_trans_tasklet *trans = (void *)data;
766 struct sk_buff_head queue;
767 struct sk_buff *skb;
768
769 __skb_queue_head_init(&queue);
770 skb_queue_splice_init(&trans->queue, &queue);
771
772 while ((skb = __skb_dequeue(&queue)))
773 XFRM_TRANS_SKB_CB(skb)->finish(XFRM_TRANS_SKB_CB(skb)->net,
774 NULL, skb);
775}
776
777int xfrm_trans_queue_net(struct net *net, struct sk_buff *skb,
778 int (*finish)(struct net *, struct sock *,
779 struct sk_buff *))
780{
781 struct xfrm_trans_tasklet *trans;
782
783 trans = this_cpu_ptr(&xfrm_trans_tasklet);
784
785 if (skb_queue_len(&trans->queue) >= netdev_max_backlog)
786 return -ENOBUFS;
787
788 BUILD_BUG_ON(sizeof(struct xfrm_trans_cb) > sizeof(skb->cb));
789
790 XFRM_TRANS_SKB_CB(skb)->finish = finish;
791 XFRM_TRANS_SKB_CB(skb)->net = net;
792 __skb_queue_tail(&trans->queue, skb);
793 tasklet_schedule(&trans->tasklet);
794 return 0;
795}
796EXPORT_SYMBOL(xfrm_trans_queue_net);
797
798int xfrm_trans_queue(struct sk_buff *skb,
799 int (*finish)(struct net *, struct sock *,
800 struct sk_buff *))
801{
802 return xfrm_trans_queue_net(dev_net(skb->dev), skb, finish);
803}
804EXPORT_SYMBOL(xfrm_trans_queue);
805
806void __init xfrm_input_init(void)
807{
808 int err;
809 int i;
810
811 init_dummy_netdev(&xfrm_napi_dev);
812 err = gro_cells_init(&gro_cells, &xfrm_napi_dev);
813 if (err)
814 gro_cells.cells = NULL;
815
816 for_each_possible_cpu(i) {
817 struct xfrm_trans_tasklet *trans;
818
819 trans = &per_cpu(xfrm_trans_tasklet, i);
820 __skb_queue_head_init(&trans->queue);
821 tasklet_init(&trans->tasklet, xfrm_trans_reinject,
822 (unsigned long)trans);
823 }
824}
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
24struct xfrm_trans_tasklet {
25 struct tasklet_struct tasklet;
26 struct sk_buff_head queue;
27};
28
29struct xfrm_trans_cb {
30 union {
31 struct inet_skb_parm h4;
32#if IS_ENABLED(CONFIG_IPV6)
33 struct inet6_skb_parm h6;
34#endif
35 } header;
36 int (*finish)(struct net *net, struct sock *sk, struct sk_buff *skb);
37};
38
39#define XFRM_TRANS_SKB_CB(__skb) ((struct xfrm_trans_cb *)&((__skb)->cb[0]))
40
41static struct kmem_cache *secpath_cachep __ro_after_init;
42
43static DEFINE_SPINLOCK(xfrm_input_afinfo_lock);
44static struct xfrm_input_afinfo const __rcu *xfrm_input_afinfo[AF_INET6 + 1];
45
46static struct gro_cells gro_cells;
47static struct net_device xfrm_napi_dev;
48
49static DEFINE_PER_CPU(struct xfrm_trans_tasklet, xfrm_trans_tasklet);
50
51int xfrm_input_register_afinfo(const struct xfrm_input_afinfo *afinfo)
52{
53 int err = 0;
54
55 if (WARN_ON(afinfo->family >= ARRAY_SIZE(xfrm_input_afinfo)))
56 return -EAFNOSUPPORT;
57
58 spin_lock_bh(&xfrm_input_afinfo_lock);
59 if (unlikely(xfrm_input_afinfo[afinfo->family] != NULL))
60 err = -EEXIST;
61 else
62 rcu_assign_pointer(xfrm_input_afinfo[afinfo->family], afinfo);
63 spin_unlock_bh(&xfrm_input_afinfo_lock);
64 return err;
65}
66EXPORT_SYMBOL(xfrm_input_register_afinfo);
67
68int xfrm_input_unregister_afinfo(const struct xfrm_input_afinfo *afinfo)
69{
70 int err = 0;
71
72 spin_lock_bh(&xfrm_input_afinfo_lock);
73 if (likely(xfrm_input_afinfo[afinfo->family] != NULL)) {
74 if (unlikely(xfrm_input_afinfo[afinfo->family] != afinfo))
75 err = -EINVAL;
76 else
77 RCU_INIT_POINTER(xfrm_input_afinfo[afinfo->family], NULL);
78 }
79 spin_unlock_bh(&xfrm_input_afinfo_lock);
80 synchronize_rcu();
81 return err;
82}
83EXPORT_SYMBOL(xfrm_input_unregister_afinfo);
84
85static const struct xfrm_input_afinfo *xfrm_input_get_afinfo(unsigned int family)
86{
87 const struct xfrm_input_afinfo *afinfo;
88
89 if (WARN_ON_ONCE(family >= ARRAY_SIZE(xfrm_input_afinfo)))
90 return NULL;
91
92 rcu_read_lock();
93 afinfo = rcu_dereference(xfrm_input_afinfo[family]);
94 if (unlikely(!afinfo))
95 rcu_read_unlock();
96 return afinfo;
97}
98
99static int xfrm_rcv_cb(struct sk_buff *skb, unsigned int family, u8 protocol,
100 int err)
101{
102 int ret;
103 const struct xfrm_input_afinfo *afinfo = xfrm_input_get_afinfo(family);
104
105 if (!afinfo)
106 return -EAFNOSUPPORT;
107
108 ret = afinfo->callback(skb, protocol, err);
109 rcu_read_unlock();
110
111 return ret;
112}
113
114void __secpath_destroy(struct sec_path *sp)
115{
116 int i;
117 for (i = 0; i < sp->len; i++)
118 xfrm_state_put(sp->xvec[i]);
119 kmem_cache_free(secpath_cachep, sp);
120}
121EXPORT_SYMBOL(__secpath_destroy);
122
123struct sec_path *secpath_dup(struct sec_path *src)
124{
125 struct sec_path *sp;
126
127 sp = kmem_cache_alloc(secpath_cachep, GFP_ATOMIC);
128 if (!sp)
129 return NULL;
130
131 sp->len = 0;
132 sp->olen = 0;
133
134 memset(sp->ovec, 0, sizeof(sp->ovec[XFRM_MAX_OFFLOAD_DEPTH]));
135
136 if (src) {
137 int i;
138
139 memcpy(sp, src, sizeof(*sp));
140 for (i = 0; i < sp->len; i++)
141 xfrm_state_hold(sp->xvec[i]);
142 }
143 refcount_set(&sp->refcnt, 1);
144 return sp;
145}
146EXPORT_SYMBOL(secpath_dup);
147
148int secpath_set(struct sk_buff *skb)
149{
150 struct sec_path *sp;
151
152 /* Allocate new secpath or COW existing one. */
153 if (!skb->sp || refcount_read(&skb->sp->refcnt) != 1) {
154 sp = secpath_dup(skb->sp);
155 if (!sp)
156 return -ENOMEM;
157
158 if (skb->sp)
159 secpath_put(skb->sp);
160 skb->sp = sp;
161 }
162 return 0;
163}
164EXPORT_SYMBOL(secpath_set);
165
166/* Fetch spi and seq from ipsec header */
167
168int xfrm_parse_spi(struct sk_buff *skb, u8 nexthdr, __be32 *spi, __be32 *seq)
169{
170 int offset, offset_seq;
171 int hlen;
172
173 switch (nexthdr) {
174 case IPPROTO_AH:
175 hlen = sizeof(struct ip_auth_hdr);
176 offset = offsetof(struct ip_auth_hdr, spi);
177 offset_seq = offsetof(struct ip_auth_hdr, seq_no);
178 break;
179 case IPPROTO_ESP:
180 hlen = sizeof(struct ip_esp_hdr);
181 offset = offsetof(struct ip_esp_hdr, spi);
182 offset_seq = offsetof(struct ip_esp_hdr, seq_no);
183 break;
184 case IPPROTO_COMP:
185 if (!pskb_may_pull(skb, sizeof(struct ip_comp_hdr)))
186 return -EINVAL;
187 *spi = htonl(ntohs(*(__be16 *)(skb_transport_header(skb) + 2)));
188 *seq = 0;
189 return 0;
190 default:
191 return 1;
192 }
193
194 if (!pskb_may_pull(skb, hlen))
195 return -EINVAL;
196
197 *spi = *(__be32 *)(skb_transport_header(skb) + offset);
198 *seq = *(__be32 *)(skb_transport_header(skb) + offset_seq);
199 return 0;
200}
201EXPORT_SYMBOL(xfrm_parse_spi);
202
203int xfrm_prepare_input(struct xfrm_state *x, struct sk_buff *skb)
204{
205 struct xfrm_mode *inner_mode = x->inner_mode;
206 int err;
207
208 err = x->outer_mode->afinfo->extract_input(x, skb);
209 if (err)
210 return err;
211
212 if (x->sel.family == AF_UNSPEC) {
213 inner_mode = xfrm_ip2inner_mode(x, XFRM_MODE_SKB_CB(skb)->protocol);
214 if (inner_mode == NULL)
215 return -EAFNOSUPPORT;
216 }
217
218 skb->protocol = inner_mode->afinfo->eth_proto;
219 return inner_mode->input2(x, skb);
220}
221EXPORT_SYMBOL(xfrm_prepare_input);
222
223int xfrm_input(struct sk_buff *skb, int nexthdr, __be32 spi, int encap_type)
224{
225 struct net *net = dev_net(skb->dev);
226 int err;
227 __be32 seq;
228 __be32 seq_hi;
229 struct xfrm_state *x = NULL;
230 xfrm_address_t *daddr;
231 struct xfrm_mode *inner_mode;
232 u32 mark = skb->mark;
233 unsigned int family = AF_UNSPEC;
234 int decaps = 0;
235 int async = 0;
236 bool xfrm_gro = false;
237 bool crypto_done = false;
238 struct xfrm_offload *xo = xfrm_offload(skb);
239
240 if (encap_type < 0) {
241 x = xfrm_input_state(skb);
242
243 if (unlikely(x->km.state != XFRM_STATE_VALID)) {
244 if (x->km.state == XFRM_STATE_ACQ)
245 XFRM_INC_STATS(net, LINUX_MIB_XFRMACQUIREERROR);
246 else
247 XFRM_INC_STATS(net,
248 LINUX_MIB_XFRMINSTATEINVALID);
249 goto drop;
250 }
251
252 family = x->outer_mode->afinfo->family;
253
254 /* An encap_type of -1 indicates async resumption. */
255 if (encap_type == -1) {
256 async = 1;
257 seq = XFRM_SKB_CB(skb)->seq.input.low;
258 goto resume;
259 }
260
261 /* encap_type < -1 indicates a GRO call. */
262 encap_type = 0;
263 seq = XFRM_SPI_SKB_CB(skb)->seq;
264
265 if (xo && (xo->flags & CRYPTO_DONE)) {
266 crypto_done = true;
267 family = XFRM_SPI_SKB_CB(skb)->family;
268
269 if (!(xo->status & CRYPTO_SUCCESS)) {
270 if (xo->status &
271 (CRYPTO_TRANSPORT_AH_AUTH_FAILED |
272 CRYPTO_TRANSPORT_ESP_AUTH_FAILED |
273 CRYPTO_TUNNEL_AH_AUTH_FAILED |
274 CRYPTO_TUNNEL_ESP_AUTH_FAILED)) {
275
276 xfrm_audit_state_icvfail(x, skb,
277 x->type->proto);
278 x->stats.integrity_failed++;
279 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEPROTOERROR);
280 goto drop;
281 }
282
283 if (xo->status & CRYPTO_INVALID_PROTOCOL) {
284 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEPROTOERROR);
285 goto drop;
286 }
287
288 XFRM_INC_STATS(net, LINUX_MIB_XFRMINBUFFERERROR);
289 goto drop;
290 }
291
292 if ((err = xfrm_parse_spi(skb, nexthdr, &spi, &seq)) != 0) {
293 XFRM_INC_STATS(net, LINUX_MIB_XFRMINHDRERROR);
294 goto drop;
295 }
296 }
297
298 goto lock;
299 }
300
301 family = XFRM_SPI_SKB_CB(skb)->family;
302
303 /* if tunnel is present override skb->mark value with tunnel i_key */
304 switch (family) {
305 case AF_INET:
306 if (XFRM_TUNNEL_SKB_CB(skb)->tunnel.ip4)
307 mark = be32_to_cpu(XFRM_TUNNEL_SKB_CB(skb)->tunnel.ip4->parms.i_key);
308 break;
309 case AF_INET6:
310 if (XFRM_TUNNEL_SKB_CB(skb)->tunnel.ip6)
311 mark = be32_to_cpu(XFRM_TUNNEL_SKB_CB(skb)->tunnel.ip6->parms.i_key);
312 break;
313 }
314
315 err = secpath_set(skb);
316 if (err) {
317 XFRM_INC_STATS(net, LINUX_MIB_XFRMINERROR);
318 goto drop;
319 }
320
321 seq = 0;
322 if (!spi && (err = xfrm_parse_spi(skb, nexthdr, &spi, &seq)) != 0) {
323 XFRM_INC_STATS(net, LINUX_MIB_XFRMINHDRERROR);
324 goto drop;
325 }
326
327 daddr = (xfrm_address_t *)(skb_network_header(skb) +
328 XFRM_SPI_SKB_CB(skb)->daddroff);
329 do {
330 if (skb->sp->len == XFRM_MAX_DEPTH) {
331 XFRM_INC_STATS(net, LINUX_MIB_XFRMINBUFFERERROR);
332 goto drop;
333 }
334
335 x = xfrm_state_lookup(net, mark, daddr, spi, nexthdr, family);
336 if (x == NULL) {
337 XFRM_INC_STATS(net, LINUX_MIB_XFRMINNOSTATES);
338 xfrm_audit_state_notfound(skb, family, spi, seq);
339 goto drop;
340 }
341
342 skb->sp->xvec[skb->sp->len++] = x;
343
344lock:
345 spin_lock(&x->lock);
346
347 if (unlikely(x->km.state != XFRM_STATE_VALID)) {
348 if (x->km.state == XFRM_STATE_ACQ)
349 XFRM_INC_STATS(net, LINUX_MIB_XFRMACQUIREERROR);
350 else
351 XFRM_INC_STATS(net,
352 LINUX_MIB_XFRMINSTATEINVALID);
353 goto drop_unlock;
354 }
355
356 if ((x->encap ? x->encap->encap_type : 0) != encap_type) {
357 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMISMATCH);
358 goto drop_unlock;
359 }
360
361 if (x->repl->check(x, skb, seq)) {
362 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATESEQERROR);
363 goto drop_unlock;
364 }
365
366 if (xfrm_state_check_expire(x)) {
367 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEEXPIRED);
368 goto drop_unlock;
369 }
370
371 spin_unlock(&x->lock);
372
373 if (xfrm_tunnel_check(skb, x, family)) {
374 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMODEERROR);
375 goto drop;
376 }
377
378 seq_hi = htonl(xfrm_replay_seqhi(x, seq));
379
380 XFRM_SKB_CB(skb)->seq.input.low = seq;
381 XFRM_SKB_CB(skb)->seq.input.hi = seq_hi;
382
383 skb_dst_force(skb);
384 dev_hold(skb->dev);
385
386 if (crypto_done)
387 nexthdr = x->type_offload->input_tail(x, skb);
388 else
389 nexthdr = x->type->input(x, skb);
390
391 if (nexthdr == -EINPROGRESS)
392 return 0;
393resume:
394 dev_put(skb->dev);
395
396 spin_lock(&x->lock);
397 if (nexthdr <= 0) {
398 if (nexthdr == -EBADMSG) {
399 xfrm_audit_state_icvfail(x, skb,
400 x->type->proto);
401 x->stats.integrity_failed++;
402 }
403 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEPROTOERROR);
404 goto drop_unlock;
405 }
406
407 /* only the first xfrm gets the encap type */
408 encap_type = 0;
409
410 if (async && x->repl->recheck(x, skb, seq)) {
411 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATESEQERROR);
412 goto drop_unlock;
413 }
414
415 x->repl->advance(x, seq);
416
417 x->curlft.bytes += skb->len;
418 x->curlft.packets++;
419
420 spin_unlock(&x->lock);
421
422 XFRM_MODE_SKB_CB(skb)->protocol = nexthdr;
423
424 inner_mode = x->inner_mode;
425
426 if (x->sel.family == AF_UNSPEC) {
427 inner_mode = xfrm_ip2inner_mode(x, XFRM_MODE_SKB_CB(skb)->protocol);
428 if (inner_mode == NULL) {
429 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMODEERROR);
430 goto drop;
431 }
432 }
433
434 if (inner_mode->input(x, skb)) {
435 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMODEERROR);
436 goto drop;
437 }
438
439 if (x->outer_mode->flags & XFRM_MODE_FLAG_TUNNEL) {
440 decaps = 1;
441 break;
442 }
443
444 /*
445 * We need the inner address. However, we only get here for
446 * transport mode so the outer address is identical.
447 */
448 daddr = &x->id.daddr;
449 family = x->outer_mode->afinfo->family;
450
451 err = xfrm_parse_spi(skb, nexthdr, &spi, &seq);
452 if (err < 0) {
453 XFRM_INC_STATS(net, LINUX_MIB_XFRMINHDRERROR);
454 goto drop;
455 }
456 } while (!err);
457
458 err = xfrm_rcv_cb(skb, family, x->type->proto, 0);
459 if (err)
460 goto drop;
461
462 nf_reset(skb);
463
464 if (decaps) {
465 if (skb->sp)
466 skb->sp->olen = 0;
467 skb_dst_drop(skb);
468 gro_cells_receive(&gro_cells, skb);
469 return 0;
470 } else {
471 xo = xfrm_offload(skb);
472 if (xo)
473 xfrm_gro = xo->flags & XFRM_GRO;
474
475 err = x->inner_mode->afinfo->transport_finish(skb, xfrm_gro || async);
476 if (xfrm_gro) {
477 if (skb->sp)
478 skb->sp->olen = 0;
479 skb_dst_drop(skb);
480 gro_cells_receive(&gro_cells, skb);
481 return err;
482 }
483
484 return err;
485 }
486
487drop_unlock:
488 spin_unlock(&x->lock);
489drop:
490 xfrm_rcv_cb(skb, family, x && x->type ? x->type->proto : nexthdr, -1);
491 kfree_skb(skb);
492 return 0;
493}
494EXPORT_SYMBOL(xfrm_input);
495
496int xfrm_input_resume(struct sk_buff *skb, int nexthdr)
497{
498 return xfrm_input(skb, nexthdr, 0, -1);
499}
500EXPORT_SYMBOL(xfrm_input_resume);
501
502static void xfrm_trans_reinject(unsigned long data)
503{
504 struct xfrm_trans_tasklet *trans = (void *)data;
505 struct sk_buff_head queue;
506 struct sk_buff *skb;
507
508 __skb_queue_head_init(&queue);
509 skb_queue_splice_init(&trans->queue, &queue);
510
511 while ((skb = __skb_dequeue(&queue)))
512 XFRM_TRANS_SKB_CB(skb)->finish(dev_net(skb->dev), NULL, skb);
513}
514
515int xfrm_trans_queue(struct sk_buff *skb,
516 int (*finish)(struct net *, struct sock *,
517 struct sk_buff *))
518{
519 struct xfrm_trans_tasklet *trans;
520
521 trans = this_cpu_ptr(&xfrm_trans_tasklet);
522
523 if (skb_queue_len(&trans->queue) >= netdev_max_backlog)
524 return -ENOBUFS;
525
526 XFRM_TRANS_SKB_CB(skb)->finish = finish;
527 __skb_queue_tail(&trans->queue, skb);
528 tasklet_schedule(&trans->tasklet);
529 return 0;
530}
531EXPORT_SYMBOL(xfrm_trans_queue);
532
533void __init xfrm_input_init(void)
534{
535 int err;
536 int i;
537
538 init_dummy_netdev(&xfrm_napi_dev);
539 err = gro_cells_init(&gro_cells, &xfrm_napi_dev);
540 if (err)
541 gro_cells.cells = NULL;
542
543 secpath_cachep = kmem_cache_create("secpath_cache",
544 sizeof(struct sec_path),
545 0, SLAB_HWCACHE_ALIGN|SLAB_PANIC,
546 NULL);
547
548 for_each_possible_cpu(i) {
549 struct xfrm_trans_tasklet *trans;
550
551 trans = &per_cpu(xfrm_trans_tasklet, i);
552 __skb_queue_head_init(&trans->queue);
553 tasklet_init(&trans->tasklet, xfrm_trans_reinject,
554 (unsigned long)trans);
555 }
556}