1 // SPDX-License-Identifier: GPL-2.0
6 * YOSHIFUJI Hideaki @USAGI
7 * Split up af-specific portion
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>
21 #include <net/ip_tunnels.h>
22 #include <net/ip6_tunnel.h>
24 #include "xfrm_inout.h"
26 struct xfrm_trans_tasklet
{
27 struct tasklet_struct tasklet
;
28 struct sk_buff_head queue
;
31 struct xfrm_trans_cb
{
33 struct inet_skb_parm h4
;
34 #if IS_ENABLED(CONFIG_IPV6)
35 struct inet6_skb_parm h6
;
38 int (*finish
)(struct net
*net
, struct sock
*sk
, struct sk_buff
*skb
);
41 #define XFRM_TRANS_SKB_CB(__skb) ((struct xfrm_trans_cb *)&((__skb)->cb[0]))
43 static DEFINE_SPINLOCK(xfrm_input_afinfo_lock
);
44 static struct xfrm_input_afinfo
const __rcu
*xfrm_input_afinfo
[AF_INET6
+ 1];
46 static struct gro_cells gro_cells
;
47 static struct net_device xfrm_napi_dev
;
49 static DEFINE_PER_CPU(struct xfrm_trans_tasklet
, xfrm_trans_tasklet
);
51 int xfrm_input_register_afinfo(const struct xfrm_input_afinfo
*afinfo
)
55 if (WARN_ON(afinfo
->family
>= ARRAY_SIZE(xfrm_input_afinfo
)))
58 spin_lock_bh(&xfrm_input_afinfo_lock
);
59 if (unlikely(xfrm_input_afinfo
[afinfo
->family
] != NULL
))
62 rcu_assign_pointer(xfrm_input_afinfo
[afinfo
->family
], afinfo
);
63 spin_unlock_bh(&xfrm_input_afinfo_lock
);
66 EXPORT_SYMBOL(xfrm_input_register_afinfo
);
68 int xfrm_input_unregister_afinfo(const struct xfrm_input_afinfo
*afinfo
)
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
))
77 RCU_INIT_POINTER(xfrm_input_afinfo
[afinfo
->family
], NULL
);
79 spin_unlock_bh(&xfrm_input_afinfo_lock
);
83 EXPORT_SYMBOL(xfrm_input_unregister_afinfo
);
85 static const struct xfrm_input_afinfo
*xfrm_input_get_afinfo(unsigned int family
)
87 const struct xfrm_input_afinfo
*afinfo
;
89 if (WARN_ON_ONCE(family
>= ARRAY_SIZE(xfrm_input_afinfo
)))
93 afinfo
= rcu_dereference(xfrm_input_afinfo
[family
]);
94 if (unlikely(!afinfo
))
99 static int xfrm_rcv_cb(struct sk_buff
*skb
, unsigned int family
, u8 protocol
,
103 const struct xfrm_input_afinfo
*afinfo
= xfrm_input_get_afinfo(family
);
106 return -EAFNOSUPPORT
;
108 ret
= afinfo
->callback(skb
, protocol
, err
);
114 struct sec_path
*secpath_set(struct sk_buff
*skb
)
116 struct sec_path
*sp
, *tmp
= skb_ext_find(skb
, SKB_EXT_SEC_PATH
);
118 sp
= skb_ext_add(skb
, SKB_EXT_SEC_PATH
);
122 if (tmp
) /* reused existing one (was COW'd if needed) */
125 /* allocated new secpath */
126 memset(sp
->ovec
, 0, sizeof(sp
->ovec
));
132 EXPORT_SYMBOL(secpath_set
);
134 /* Fetch spi and seq from ipsec header */
136 int xfrm_parse_spi(struct sk_buff
*skb
, u8 nexthdr
, __be32
*spi
, __be32
*seq
)
138 int offset
, offset_seq
;
143 hlen
= sizeof(struct ip_auth_hdr
);
144 offset
= offsetof(struct ip_auth_hdr
, spi
);
145 offset_seq
= offsetof(struct ip_auth_hdr
, seq_no
);
148 hlen
= sizeof(struct ip_esp_hdr
);
149 offset
= offsetof(struct ip_esp_hdr
, spi
);
150 offset_seq
= offsetof(struct ip_esp_hdr
, seq_no
);
153 if (!pskb_may_pull(skb
, sizeof(struct ip_comp_hdr
)))
155 *spi
= htonl(ntohs(*(__be16
*)(skb_transport_header(skb
) + 2)));
162 if (!pskb_may_pull(skb
, hlen
))
165 *spi
= *(__be32
*)(skb_transport_header(skb
) + offset
);
166 *seq
= *(__be32
*)(skb_transport_header(skb
) + offset_seq
);
169 EXPORT_SYMBOL(xfrm_parse_spi
);
171 static int xfrm4_remove_beet_encap(struct xfrm_state
*x
, struct sk_buff
*skb
)
177 if (unlikely(XFRM_MODE_SKB_CB(skb
)->protocol
== IPPROTO_BEETPH
)) {
178 struct ip_beet_phdr
*ph
;
181 if (!pskb_may_pull(skb
, sizeof(*ph
)))
184 ph
= (struct ip_beet_phdr
*)skb
->data
;
186 phlen
= sizeof(*ph
) + ph
->padlen
;
187 optlen
= ph
->hdrlen
* 8 + (IPV4_BEET_PHMAXLEN
- phlen
);
188 if (optlen
< 0 || optlen
& 3 || optlen
> 250)
191 XFRM_MODE_SKB_CB(skb
)->protocol
= ph
->nexthdr
;
193 if (!pskb_may_pull(skb
, phlen
))
195 __skb_pull(skb
, phlen
);
198 skb_push(skb
, sizeof(*iph
));
199 skb_reset_network_header(skb
);
200 skb_mac_header_rebuild(skb
);
202 xfrm4_beet_make_header(skb
);
206 iph
->ihl
+= optlen
/ 4;
207 iph
->tot_len
= htons(skb
->len
);
208 iph
->daddr
= x
->sel
.daddr
.a4
;
209 iph
->saddr
= x
->sel
.saddr
.a4
;
211 iph
->check
= ip_fast_csum(skb_network_header(skb
), iph
->ihl
);
217 static void ipip_ecn_decapsulate(struct sk_buff
*skb
)
219 struct iphdr
*inner_iph
= ipip_hdr(skb
);
221 if (INET_ECN_is_ce(XFRM_MODE_SKB_CB(skb
)->tos
))
222 IP_ECN_set_ce(inner_iph
);
225 static int xfrm4_remove_tunnel_encap(struct xfrm_state
*x
, struct sk_buff
*skb
)
229 if (XFRM_MODE_SKB_CB(skb
)->protocol
!= IPPROTO_IPIP
)
232 if (!pskb_may_pull(skb
, sizeof(struct iphdr
)))
235 err
= skb_unclone(skb
, GFP_ATOMIC
);
239 if (x
->props
.flags
& XFRM_STATE_DECAP_DSCP
)
240 ipv4_copy_dscp(XFRM_MODE_SKB_CB(skb
)->tos
, ipip_hdr(skb
));
241 if (!(x
->props
.flags
& XFRM_STATE_NOECN
))
242 ipip_ecn_decapsulate(skb
);
244 skb_reset_network_header(skb
);
245 skb_mac_header_rebuild(skb
);
247 eth_hdr(skb
)->h_proto
= skb
->protocol
;
255 static void ipip6_ecn_decapsulate(struct sk_buff
*skb
)
257 struct ipv6hdr
*inner_iph
= ipipv6_hdr(skb
);
259 if (INET_ECN_is_ce(XFRM_MODE_SKB_CB(skb
)->tos
))
260 IP6_ECN_set_ce(skb
, inner_iph
);
263 static int xfrm6_remove_tunnel_encap(struct xfrm_state
*x
, struct sk_buff
*skb
)
267 if (XFRM_MODE_SKB_CB(skb
)->protocol
!= IPPROTO_IPV6
)
269 if (!pskb_may_pull(skb
, sizeof(struct ipv6hdr
)))
272 err
= skb_unclone(skb
, GFP_ATOMIC
);
276 if (x
->props
.flags
& XFRM_STATE_DECAP_DSCP
)
277 ipv6_copy_dscp(ipv6_get_dsfield(ipv6_hdr(skb
)),
279 if (!(x
->props
.flags
& XFRM_STATE_NOECN
))
280 ipip6_ecn_decapsulate(skb
);
282 skb_reset_network_header(skb
);
283 skb_mac_header_rebuild(skb
);
285 eth_hdr(skb
)->h_proto
= skb
->protocol
;
293 static int xfrm6_remove_beet_encap(struct xfrm_state
*x
, struct sk_buff
*skb
)
295 struct ipv6hdr
*ip6h
;
296 int size
= sizeof(struct ipv6hdr
);
299 err
= skb_cow_head(skb
, size
+ skb
->mac_len
);
303 __skb_push(skb
, size
);
304 skb_reset_network_header(skb
);
305 skb_mac_header_rebuild(skb
);
307 xfrm6_beet_make_header(skb
);
309 ip6h
= ipv6_hdr(skb
);
310 ip6h
->payload_len
= htons(skb
->len
- size
);
311 ip6h
->daddr
= x
->sel
.daddr
.in6
;
312 ip6h
->saddr
= x
->sel
.saddr
.in6
;
318 /* Remove encapsulation header.
320 * The IP header will be moved over the top of the encapsulation
323 * On entry, the transport header shall point to where the IP header
324 * should be and the network header shall be set to where the IP
325 * header currently is. skb->data shall point to the start of the
329 xfrm_inner_mode_encap_remove(struct xfrm_state
*x
,
330 const struct xfrm_mode
*inner_mode
,
333 switch (inner_mode
->encap
) {
335 if (inner_mode
->family
== AF_INET
)
336 return xfrm4_remove_beet_encap(x
, skb
);
337 if (inner_mode
->family
== AF_INET6
)
338 return xfrm6_remove_beet_encap(x
, skb
);
340 case XFRM_MODE_TUNNEL
:
341 if (inner_mode
->family
== AF_INET
)
342 return xfrm4_remove_tunnel_encap(x
, skb
);
343 if (inner_mode
->family
== AF_INET6
)
344 return xfrm6_remove_tunnel_encap(x
, skb
);
352 static int xfrm_prepare_input(struct xfrm_state
*x
, struct sk_buff
*skb
)
354 const struct xfrm_mode
*inner_mode
= &x
->inner_mode
;
355 const struct xfrm_state_afinfo
*afinfo
;
356 int err
= -EAFNOSUPPORT
;
359 afinfo
= xfrm_state_afinfo_get_rcu(x
->outer_mode
.family
);
361 err
= afinfo
->extract_input(x
, skb
);
367 if (x
->sel
.family
== AF_UNSPEC
) {
368 inner_mode
= xfrm_ip2inner_mode(x
, XFRM_MODE_SKB_CB(skb
)->protocol
);
370 return -EAFNOSUPPORT
;
373 switch (inner_mode
->family
) {
375 skb
->protocol
= htons(ETH_P_IP
);
378 skb
->protocol
= htons(ETH_P_IPV6
);
385 return xfrm_inner_mode_encap_remove(x
, inner_mode
, skb
);
388 /* Remove encapsulation header.
390 * The IP header will be moved over the top of the encapsulation header.
392 * On entry, skb_transport_header() shall point to where the IP header
393 * should be and skb_network_header() shall be set to where the IP header
394 * currently is. skb->data shall point to the start of the payload.
396 static int xfrm4_transport_input(struct xfrm_state
*x
, struct sk_buff
*skb
)
398 int ihl
= skb
->data
- skb_transport_header(skb
);
400 if (skb
->transport_header
!= skb
->network_header
) {
401 memmove(skb_transport_header(skb
),
402 skb_network_header(skb
), ihl
);
403 skb
->network_header
= skb
->transport_header
;
405 ip_hdr(skb
)->tot_len
= htons(skb
->len
+ ihl
);
406 skb_reset_transport_header(skb
);
410 static int xfrm6_transport_input(struct xfrm_state
*x
, struct sk_buff
*skb
)
412 #if IS_ENABLED(CONFIG_IPV6)
413 int ihl
= skb
->data
- skb_transport_header(skb
);
415 if (skb
->transport_header
!= skb
->network_header
) {
416 memmove(skb_transport_header(skb
),
417 skb_network_header(skb
), ihl
);
418 skb
->network_header
= skb
->transport_header
;
420 ipv6_hdr(skb
)->payload_len
= htons(skb
->len
+ ihl
-
421 sizeof(struct ipv6hdr
));
422 skb_reset_transport_header(skb
);
426 return -EAFNOSUPPORT
;
430 static int xfrm_inner_mode_input(struct xfrm_state
*x
,
431 const struct xfrm_mode
*inner_mode
,
434 switch (inner_mode
->encap
) {
436 case XFRM_MODE_TUNNEL
:
437 return xfrm_prepare_input(x
, skb
);
438 case XFRM_MODE_TRANSPORT
:
439 if (inner_mode
->family
== AF_INET
)
440 return xfrm4_transport_input(x
, skb
);
441 if (inner_mode
->family
== AF_INET6
)
442 return xfrm6_transport_input(x
, skb
);
444 case XFRM_MODE_ROUTEOPTIMIZATION
:
455 int xfrm_input(struct sk_buff
*skb
, int nexthdr
, __be32 spi
, int encap_type
)
457 const struct xfrm_state_afinfo
*afinfo
;
458 struct net
*net
= dev_net(skb
->dev
);
459 const struct xfrm_mode
*inner_mode
;
463 struct xfrm_state
*x
= NULL
;
464 xfrm_address_t
*daddr
;
465 u32 mark
= skb
->mark
;
466 unsigned int family
= AF_UNSPEC
;
469 bool xfrm_gro
= false;
470 bool crypto_done
= false;
471 struct xfrm_offload
*xo
= xfrm_offload(skb
);
474 if (encap_type
< 0) {
475 x
= xfrm_input_state(skb
);
477 if (unlikely(x
->km
.state
!= XFRM_STATE_VALID
)) {
478 if (x
->km
.state
== XFRM_STATE_ACQ
)
479 XFRM_INC_STATS(net
, LINUX_MIB_XFRMACQUIREERROR
);
482 LINUX_MIB_XFRMINSTATEINVALID
);
484 if (encap_type
== -1)
489 family
= x
->outer_mode
.family
;
491 /* An encap_type of -1 indicates async resumption. */
492 if (encap_type
== -1) {
494 seq
= XFRM_SKB_CB(skb
)->seq
.input
.low
;
498 /* encap_type < -1 indicates a GRO call. */
500 seq
= XFRM_SPI_SKB_CB(skb
)->seq
;
502 if (xo
&& (xo
->flags
& CRYPTO_DONE
)) {
504 family
= XFRM_SPI_SKB_CB(skb
)->family
;
506 if (!(xo
->status
& CRYPTO_SUCCESS
)) {
508 (CRYPTO_TRANSPORT_AH_AUTH_FAILED
|
509 CRYPTO_TRANSPORT_ESP_AUTH_FAILED
|
510 CRYPTO_TUNNEL_AH_AUTH_FAILED
|
511 CRYPTO_TUNNEL_ESP_AUTH_FAILED
)) {
513 xfrm_audit_state_icvfail(x
, skb
,
515 x
->stats
.integrity_failed
++;
516 XFRM_INC_STATS(net
, LINUX_MIB_XFRMINSTATEPROTOERROR
);
520 if (xo
->status
& CRYPTO_INVALID_PROTOCOL
) {
521 XFRM_INC_STATS(net
, LINUX_MIB_XFRMINSTATEPROTOERROR
);
525 XFRM_INC_STATS(net
, LINUX_MIB_XFRMINBUFFERERROR
);
529 if ((err
= xfrm_parse_spi(skb
, nexthdr
, &spi
, &seq
)) != 0) {
530 XFRM_INC_STATS(net
, LINUX_MIB_XFRMINHDRERROR
);
538 family
= XFRM_SPI_SKB_CB(skb
)->family
;
540 /* if tunnel is present override skb->mark value with tunnel i_key */
543 if (XFRM_TUNNEL_SKB_CB(skb
)->tunnel
.ip4
)
544 mark
= be32_to_cpu(XFRM_TUNNEL_SKB_CB(skb
)->tunnel
.ip4
->parms
.i_key
);
547 if (XFRM_TUNNEL_SKB_CB(skb
)->tunnel
.ip6
)
548 mark
= be32_to_cpu(XFRM_TUNNEL_SKB_CB(skb
)->tunnel
.ip6
->parms
.i_key
);
552 sp
= secpath_set(skb
);
554 XFRM_INC_STATS(net
, LINUX_MIB_XFRMINERROR
);
559 if (!spi
&& (err
= xfrm_parse_spi(skb
, nexthdr
, &spi
, &seq
)) != 0) {
561 XFRM_INC_STATS(net
, LINUX_MIB_XFRMINHDRERROR
);
565 daddr
= (xfrm_address_t
*)(skb_network_header(skb
) +
566 XFRM_SPI_SKB_CB(skb
)->daddroff
);
568 sp
= skb_sec_path(skb
);
570 if (sp
->len
== XFRM_MAX_DEPTH
) {
572 XFRM_INC_STATS(net
, LINUX_MIB_XFRMINBUFFERERROR
);
576 x
= xfrm_state_lookup(net
, mark
, daddr
, spi
, nexthdr
, family
);
579 XFRM_INC_STATS(net
, LINUX_MIB_XFRMINNOSTATES
);
580 xfrm_audit_state_notfound(skb
, family
, spi
, seq
);
584 skb
->mark
= xfrm_smark_get(skb
->mark
, x
);
586 sp
->xvec
[sp
->len
++] = x
;
590 XFRM_INC_STATS(net
, LINUX_MIB_XFRMINERROR
);
597 if (unlikely(x
->km
.state
!= XFRM_STATE_VALID
)) {
598 if (x
->km
.state
== XFRM_STATE_ACQ
)
599 XFRM_INC_STATS(net
, LINUX_MIB_XFRMACQUIREERROR
);
602 LINUX_MIB_XFRMINSTATEINVALID
);
606 if ((x
->encap
? x
->encap
->encap_type
: 0) != encap_type
) {
607 XFRM_INC_STATS(net
, LINUX_MIB_XFRMINSTATEMISMATCH
);
611 if (x
->repl
->check(x
, skb
, seq
)) {
612 XFRM_INC_STATS(net
, LINUX_MIB_XFRMINSTATESEQERROR
);
616 if (xfrm_state_check_expire(x
)) {
617 XFRM_INC_STATS(net
, LINUX_MIB_XFRMINSTATEEXPIRED
);
621 spin_unlock(&x
->lock
);
623 if (xfrm_tunnel_check(skb
, x
, family
)) {
624 XFRM_INC_STATS(net
, LINUX_MIB_XFRMINSTATEMODEERROR
);
628 seq_hi
= htonl(xfrm_replay_seqhi(x
, seq
));
630 XFRM_SKB_CB(skb
)->seq
.input
.low
= seq
;
631 XFRM_SKB_CB(skb
)->seq
.input
.hi
= seq_hi
;
636 nexthdr
= x
->type_offload
->input_tail(x
, skb
);
638 nexthdr
= x
->type
->input(x
, skb
);
640 if (nexthdr
== -EINPROGRESS
)
647 if (nexthdr
== -EBADMSG
) {
648 xfrm_audit_state_icvfail(x
, skb
,
650 x
->stats
.integrity_failed
++;
652 XFRM_INC_STATS(net
, LINUX_MIB_XFRMINSTATEPROTOERROR
);
656 /* only the first xfrm gets the encap type */
659 if (async
&& x
->repl
->recheck(x
, skb
, seq
)) {
660 XFRM_INC_STATS(net
, LINUX_MIB_XFRMINSTATESEQERROR
);
664 x
->repl
->advance(x
, seq
);
666 x
->curlft
.bytes
+= skb
->len
;
669 spin_unlock(&x
->lock
);
671 XFRM_MODE_SKB_CB(skb
)->protocol
= nexthdr
;
673 inner_mode
= &x
->inner_mode
;
675 if (x
->sel
.family
== AF_UNSPEC
) {
676 inner_mode
= xfrm_ip2inner_mode(x
, XFRM_MODE_SKB_CB(skb
)->protocol
);
677 if (inner_mode
== NULL
) {
678 XFRM_INC_STATS(net
, LINUX_MIB_XFRMINSTATEMODEERROR
);
683 if (xfrm_inner_mode_input(x
, inner_mode
, skb
)) {
684 XFRM_INC_STATS(net
, LINUX_MIB_XFRMINSTATEMODEERROR
);
688 if (x
->outer_mode
.flags
& XFRM_MODE_FLAG_TUNNEL
) {
694 * We need the inner address. However, we only get here for
695 * transport mode so the outer address is identical.
697 daddr
= &x
->id
.daddr
;
698 family
= x
->outer_mode
.family
;
700 err
= xfrm_parse_spi(skb
, nexthdr
, &spi
, &seq
);
702 XFRM_INC_STATS(net
, LINUX_MIB_XFRMINHDRERROR
);
708 err
= xfrm_rcv_cb(skb
, family
, x
->type
->proto
, 0);
715 sp
= skb_sec_path(skb
);
719 gro_cells_receive(&gro_cells
, skb
);
722 xo
= xfrm_offload(skb
);
724 xfrm_gro
= xo
->flags
& XFRM_GRO
;
728 afinfo
= xfrm_state_afinfo_get_rcu(x
->inner_mode
.family
);
730 err
= afinfo
->transport_finish(skb
, xfrm_gro
|| async
);
733 sp
= skb_sec_path(skb
);
737 gro_cells_receive(&gro_cells
, skb
);
745 spin_unlock(&x
->lock
);
747 xfrm_rcv_cb(skb
, family
, x
&& x
->type
? x
->type
->proto
: nexthdr
, -1);
751 EXPORT_SYMBOL(xfrm_input
);
753 int xfrm_input_resume(struct sk_buff
*skb
, int nexthdr
)
755 return xfrm_input(skb
, nexthdr
, 0, -1);
757 EXPORT_SYMBOL(xfrm_input_resume
);
759 static void xfrm_trans_reinject(unsigned long data
)
761 struct xfrm_trans_tasklet
*trans
= (void *)data
;
762 struct sk_buff_head queue
;
765 __skb_queue_head_init(&queue
);
766 skb_queue_splice_init(&trans
->queue
, &queue
);
768 while ((skb
= __skb_dequeue(&queue
)))
769 XFRM_TRANS_SKB_CB(skb
)->finish(dev_net(skb
->dev
), NULL
, skb
);
772 int xfrm_trans_queue(struct sk_buff
*skb
,
773 int (*finish
)(struct net
*, struct sock
*,
776 struct xfrm_trans_tasklet
*trans
;
778 trans
= this_cpu_ptr(&xfrm_trans_tasklet
);
780 if (skb_queue_len(&trans
->queue
) >= netdev_max_backlog
)
783 XFRM_TRANS_SKB_CB(skb
)->finish
= finish
;
784 __skb_queue_tail(&trans
->queue
, skb
);
785 tasklet_schedule(&trans
->tasklet
);
788 EXPORT_SYMBOL(xfrm_trans_queue
);
790 void __init
xfrm_input_init(void)
795 init_dummy_netdev(&xfrm_napi_dev
);
796 err
= gro_cells_init(&gro_cells
, &xfrm_napi_dev
);
798 gro_cells
.cells
= NULL
;
800 for_each_possible_cpu(i
) {
801 struct xfrm_trans_tasklet
*trans
;
803 trans
= &per_cpu(xfrm_trans_tasklet
, i
);
804 __skb_queue_head_init(&trans
->queue
);
805 tasklet_init(&trans
->tasklet
, xfrm_trans_reinject
,
806 (unsigned long)trans
);