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>
23 #include <net/dst_metadata.h>
25 #include "xfrm_inout.h"
27 struct xfrm_trans_tasklet {
28 struct work_struct work;
29 spinlock_t queue_lock;
30 struct sk_buff_head queue;
33 struct xfrm_trans_cb {
35 struct inet_skb_parm h4;
36 #if IS_ENABLED(CONFIG_IPV6)
37 struct inet6_skb_parm h6;
40 int (*finish)(struct net *net, struct sock *sk, struct sk_buff *skb);
44 #define XFRM_TRANS_SKB_CB(__skb) ((struct xfrm_trans_cb *)&((__skb)->cb[0]))
46 static DEFINE_SPINLOCK(xfrm_input_afinfo_lock);
47 static struct xfrm_input_afinfo const __rcu *xfrm_input_afinfo[2][AF_INET6 + 1];
49 static struct gro_cells gro_cells;
50 static struct net_device xfrm_napi_dev;
52 static DEFINE_PER_CPU(struct xfrm_trans_tasklet, xfrm_trans_tasklet);
54 int xfrm_input_register_afinfo(const struct xfrm_input_afinfo *afinfo)
58 if (WARN_ON(afinfo->family > AF_INET6))
61 spin_lock_bh(&xfrm_input_afinfo_lock);
62 if (unlikely(xfrm_input_afinfo[afinfo->is_ipip][afinfo->family]))
65 rcu_assign_pointer(xfrm_input_afinfo[afinfo->is_ipip][afinfo->family], afinfo);
66 spin_unlock_bh(&xfrm_input_afinfo_lock);
69 EXPORT_SYMBOL(xfrm_input_register_afinfo);
71 int xfrm_input_unregister_afinfo(const struct xfrm_input_afinfo *afinfo)
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))
80 RCU_INIT_POINTER(xfrm_input_afinfo[afinfo->is_ipip][afinfo->family], NULL);
82 spin_unlock_bh(&xfrm_input_afinfo_lock);
86 EXPORT_SYMBOL(xfrm_input_unregister_afinfo);
88 static const struct xfrm_input_afinfo *xfrm_input_get_afinfo(u8 family, bool is_ipip)
90 const struct xfrm_input_afinfo *afinfo;
92 if (WARN_ON_ONCE(family > AF_INET6))
96 afinfo = rcu_dereference(xfrm_input_afinfo[is_ipip][family]);
97 if (unlikely(!afinfo))
102 static int xfrm_rcv_cb(struct sk_buff *skb, unsigned int family, u8 protocol,
105 bool is_ipip = (protocol == IPPROTO_IPIP || protocol == IPPROTO_IPV6);
106 const struct xfrm_input_afinfo *afinfo;
109 afinfo = xfrm_input_get_afinfo(family, is_ipip);
111 return -EAFNOSUPPORT;
113 ret = afinfo->callback(skb, protocol, err);
119 struct sec_path *secpath_set(struct sk_buff *skb)
121 struct sec_path *sp, *tmp = skb_ext_find(skb, SKB_EXT_SEC_PATH);
123 sp = skb_ext_add(skb, SKB_EXT_SEC_PATH);
127 if (tmp) /* reused existing one (was COW'd if needed) */
130 /* allocated new secpath */
131 memset(sp->ovec, 0, sizeof(sp->ovec));
137 EXPORT_SYMBOL(secpath_set);
139 /* Fetch spi and seq from ipsec header */
141 int xfrm_parse_spi(struct sk_buff *skb, u8 nexthdr, __be32 *spi, __be32 *seq)
143 int offset, offset_seq;
148 hlen = sizeof(struct ip_auth_hdr);
149 offset = offsetof(struct ip_auth_hdr, spi);
150 offset_seq = offsetof(struct ip_auth_hdr, seq_no);
153 hlen = sizeof(struct ip_esp_hdr);
154 offset = offsetof(struct ip_esp_hdr, spi);
155 offset_seq = offsetof(struct ip_esp_hdr, seq_no);
158 if (!pskb_may_pull(skb, sizeof(struct ip_comp_hdr)))
160 *spi = htonl(ntohs(*(__be16 *)(skb_transport_header(skb) + 2)));
167 if (!pskb_may_pull(skb, hlen))
170 *spi = *(__be32 *)(skb_transport_header(skb) + offset);
171 *seq = *(__be32 *)(skb_transport_header(skb) + offset_seq);
174 EXPORT_SYMBOL(xfrm_parse_spi);
176 static int xfrm4_remove_beet_encap(struct xfrm_state *x, struct sk_buff *skb)
182 if (unlikely(XFRM_MODE_SKB_CB(skb)->protocol == IPPROTO_BEETPH)) {
183 struct ip_beet_phdr *ph;
186 if (!pskb_may_pull(skb, sizeof(*ph)))
189 ph = (struct ip_beet_phdr *)skb->data;
191 phlen = sizeof(*ph) + ph->padlen;
192 optlen = ph->hdrlen * 8 + (IPV4_BEET_PHMAXLEN - phlen);
193 if (optlen < 0 || optlen & 3 || optlen > 250)
196 XFRM_MODE_SKB_CB(skb)->protocol = ph->nexthdr;
198 if (!pskb_may_pull(skb, phlen))
200 __skb_pull(skb, phlen);
203 skb_push(skb, sizeof(*iph));
204 skb_reset_network_header(skb);
205 skb_mac_header_rebuild(skb);
207 xfrm4_beet_make_header(skb);
211 iph->ihl += optlen / 4;
212 iph->tot_len = htons(skb->len);
213 iph->daddr = x->sel.daddr.a4;
214 iph->saddr = x->sel.saddr.a4;
216 iph->check = ip_fast_csum(skb_network_header(skb), iph->ihl);
222 static void ipip_ecn_decapsulate(struct sk_buff *skb)
224 struct iphdr *inner_iph = ipip_hdr(skb);
226 if (INET_ECN_is_ce(XFRM_MODE_SKB_CB(skb)->tos))
227 IP_ECN_set_ce(inner_iph);
230 static int xfrm4_remove_tunnel_encap(struct xfrm_state *x, struct sk_buff *skb)
234 if (XFRM_MODE_SKB_CB(skb)->protocol != IPPROTO_IPIP)
237 if (!pskb_may_pull(skb, sizeof(struct iphdr)))
240 err = skb_unclone(skb, GFP_ATOMIC);
244 if (x->props.flags & XFRM_STATE_DECAP_DSCP)
245 ipv4_copy_dscp(XFRM_MODE_SKB_CB(skb)->tos, ipip_hdr(skb));
246 if (!(x->props.flags & XFRM_STATE_NOECN))
247 ipip_ecn_decapsulate(skb);
249 skb_reset_network_header(skb);
250 skb_mac_header_rebuild(skb);
252 eth_hdr(skb)->h_proto = skb->protocol;
260 static void ipip6_ecn_decapsulate(struct sk_buff *skb)
262 struct ipv6hdr *inner_iph = ipipv6_hdr(skb);
264 if (INET_ECN_is_ce(XFRM_MODE_SKB_CB(skb)->tos))
265 IP6_ECN_set_ce(skb, inner_iph);
268 static int xfrm6_remove_tunnel_encap(struct xfrm_state *x, struct sk_buff *skb)
272 if (XFRM_MODE_SKB_CB(skb)->protocol != IPPROTO_IPV6)
274 if (!pskb_may_pull(skb, sizeof(struct ipv6hdr)))
277 err = skb_unclone(skb, GFP_ATOMIC);
281 if (x->props.flags & XFRM_STATE_DECAP_DSCP)
282 ipv6_copy_dscp(XFRM_MODE_SKB_CB(skb)->tos, ipipv6_hdr(skb));
283 if (!(x->props.flags & XFRM_STATE_NOECN))
284 ipip6_ecn_decapsulate(skb);
286 skb_reset_network_header(skb);
287 skb_mac_header_rebuild(skb);
289 eth_hdr(skb)->h_proto = skb->protocol;
297 static int xfrm6_remove_beet_encap(struct xfrm_state *x, struct sk_buff *skb)
299 struct ipv6hdr *ip6h;
300 int size = sizeof(struct ipv6hdr);
303 err = skb_cow_head(skb, size + skb->mac_len);
307 __skb_push(skb, size);
308 skb_reset_network_header(skb);
309 skb_mac_header_rebuild(skb);
311 xfrm6_beet_make_header(skb);
313 ip6h = ipv6_hdr(skb);
314 ip6h->payload_len = htons(skb->len - size);
315 ip6h->daddr = x->sel.daddr.in6;
316 ip6h->saddr = x->sel.saddr.in6;
322 /* Remove encapsulation header.
324 * The IP header will be moved over the top of the encapsulation
327 * On entry, the transport header shall point to where the IP header
328 * should be and the network header shall be set to where the IP
329 * header currently is. skb->data shall point to the start of the
333 xfrm_inner_mode_encap_remove(struct xfrm_state *x,
334 const struct xfrm_mode *inner_mode,
337 switch (inner_mode->encap) {
339 if (inner_mode->family == AF_INET)
340 return xfrm4_remove_beet_encap(x, skb);
341 if (inner_mode->family == AF_INET6)
342 return xfrm6_remove_beet_encap(x, skb);
344 case XFRM_MODE_TUNNEL:
345 if (inner_mode->family == AF_INET)
346 return xfrm4_remove_tunnel_encap(x, skb);
347 if (inner_mode->family == AF_INET6)
348 return xfrm6_remove_tunnel_encap(x, skb);
356 static int xfrm_prepare_input(struct xfrm_state *x, struct sk_buff *skb)
358 const struct xfrm_mode *inner_mode = &x->inner_mode;
360 switch (x->outer_mode.family) {
362 xfrm4_extract_header(skb);
365 xfrm6_extract_header(skb);
369 return -EAFNOSUPPORT;
372 if (x->sel.family == AF_UNSPEC) {
373 inner_mode = xfrm_ip2inner_mode(x, XFRM_MODE_SKB_CB(skb)->protocol);
375 return -EAFNOSUPPORT;
378 switch (inner_mode->family) {
380 skb->protocol = htons(ETH_P_IP);
383 skb->protocol = htons(ETH_P_IPV6);
390 return xfrm_inner_mode_encap_remove(x, inner_mode, skb);
393 /* Remove encapsulation header.
395 * The IP header will be moved over the top of the encapsulation header.
397 * On entry, skb_transport_header() shall point to where the IP header
398 * should be and skb_network_header() shall be set to where the IP header
399 * currently is. skb->data shall point to the start of the payload.
401 static int xfrm4_transport_input(struct xfrm_state *x, struct sk_buff *skb)
403 int ihl = skb->data - skb_transport_header(skb);
405 if (skb->transport_header != skb->network_header) {
406 memmove(skb_transport_header(skb),
407 skb_network_header(skb), ihl);
408 skb->network_header = skb->transport_header;
410 ip_hdr(skb)->tot_len = htons(skb->len + ihl);
411 skb_reset_transport_header(skb);
415 static int xfrm6_transport_input(struct xfrm_state *x, struct sk_buff *skb)
417 #if IS_ENABLED(CONFIG_IPV6)
418 int ihl = skb->data - skb_transport_header(skb);
420 if (skb->transport_header != skb->network_header) {
421 memmove(skb_transport_header(skb),
422 skb_network_header(skb), ihl);
423 skb->network_header = skb->transport_header;
425 ipv6_hdr(skb)->payload_len = htons(skb->len + ihl -
426 sizeof(struct ipv6hdr));
427 skb_reset_transport_header(skb);
431 return -EAFNOSUPPORT;
435 static int xfrm_inner_mode_input(struct xfrm_state *x,
436 const struct xfrm_mode *inner_mode,
439 switch (inner_mode->encap) {
441 case XFRM_MODE_TUNNEL:
442 return xfrm_prepare_input(x, skb);
443 case XFRM_MODE_TRANSPORT:
444 if (inner_mode->family == AF_INET)
445 return xfrm4_transport_input(x, skb);
446 if (inner_mode->family == AF_INET6)
447 return xfrm6_transport_input(x, skb);
449 case XFRM_MODE_ROUTEOPTIMIZATION:
460 int xfrm_input(struct sk_buff *skb, int nexthdr, __be32 spi, int encap_type)
462 const struct xfrm_state_afinfo *afinfo;
463 struct net *net = dev_net(skb->dev);
464 const struct xfrm_mode *inner_mode;
468 struct xfrm_state *x = NULL;
469 xfrm_address_t *daddr;
470 u32 mark = skb->mark;
471 unsigned int family = AF_UNSPEC;
474 bool xfrm_gro = false;
475 bool crypto_done = false;
476 struct xfrm_offload *xo = xfrm_offload(skb);
479 if (encap_type < 0) {
480 x = xfrm_input_state(skb);
482 if (unlikely(x->km.state != XFRM_STATE_VALID)) {
483 if (x->km.state == XFRM_STATE_ACQ)
484 XFRM_INC_STATS(net, LINUX_MIB_XFRMACQUIREERROR);
487 LINUX_MIB_XFRMINSTATEINVALID);
489 if (encap_type == -1)
494 family = x->outer_mode.family;
496 /* An encap_type of -1 indicates async resumption. */
497 if (encap_type == -1) {
499 seq = XFRM_SKB_CB(skb)->seq.input.low;
503 /* encap_type < -1 indicates a GRO call. */
505 seq = XFRM_SPI_SKB_CB(skb)->seq;
507 if (xo && (xo->flags & CRYPTO_DONE)) {
509 family = XFRM_SPI_SKB_CB(skb)->family;
511 if (!(xo->status & CRYPTO_SUCCESS)) {
513 (CRYPTO_TRANSPORT_AH_AUTH_FAILED |
514 CRYPTO_TRANSPORT_ESP_AUTH_FAILED |
515 CRYPTO_TUNNEL_AH_AUTH_FAILED |
516 CRYPTO_TUNNEL_ESP_AUTH_FAILED)) {
518 xfrm_audit_state_icvfail(x, skb,
520 x->stats.integrity_failed++;
521 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEPROTOERROR);
525 if (xo->status & CRYPTO_INVALID_PROTOCOL) {
526 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEPROTOERROR);
530 XFRM_INC_STATS(net, LINUX_MIB_XFRMINBUFFERERROR);
534 if (xfrm_parse_spi(skb, nexthdr, &spi, &seq)) {
535 XFRM_INC_STATS(net, LINUX_MIB_XFRMINHDRERROR);
543 family = XFRM_SPI_SKB_CB(skb)->family;
545 /* if tunnel is present override skb->mark value with tunnel i_key */
548 if (XFRM_TUNNEL_SKB_CB(skb)->tunnel.ip4)
549 mark = be32_to_cpu(XFRM_TUNNEL_SKB_CB(skb)->tunnel.ip4->parms.i_key);
552 if (XFRM_TUNNEL_SKB_CB(skb)->tunnel.ip6)
553 mark = be32_to_cpu(XFRM_TUNNEL_SKB_CB(skb)->tunnel.ip6->parms.i_key);
557 sp = secpath_set(skb);
559 XFRM_INC_STATS(net, LINUX_MIB_XFRMINERROR);
564 if (!spi && xfrm_parse_spi(skb, nexthdr, &spi, &seq)) {
566 XFRM_INC_STATS(net, LINUX_MIB_XFRMINHDRERROR);
570 daddr = (xfrm_address_t *)(skb_network_header(skb) +
571 XFRM_SPI_SKB_CB(skb)->daddroff);
573 sp = skb_sec_path(skb);
575 if (sp->len == XFRM_MAX_DEPTH) {
577 XFRM_INC_STATS(net, LINUX_MIB_XFRMINBUFFERERROR);
581 x = xfrm_state_lookup(net, mark, daddr, spi, nexthdr, family);
584 XFRM_INC_STATS(net, LINUX_MIB_XFRMINNOSTATES);
585 xfrm_audit_state_notfound(skb, family, spi, seq);
589 skb->mark = xfrm_smark_get(skb->mark, x);
591 sp->xvec[sp->len++] = x;
595 XFRM_INC_STATS(net, LINUX_MIB_XFRMINERROR);
602 if (unlikely(x->km.state != XFRM_STATE_VALID)) {
603 if (x->km.state == XFRM_STATE_ACQ)
604 XFRM_INC_STATS(net, LINUX_MIB_XFRMACQUIREERROR);
607 LINUX_MIB_XFRMINSTATEINVALID);
611 if ((x->encap ? x->encap->encap_type : 0) != encap_type) {
612 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMISMATCH);
616 if (xfrm_replay_check(x, skb, seq)) {
617 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATESEQERROR);
621 if (xfrm_state_check_expire(x)) {
622 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEEXPIRED);
626 spin_unlock(&x->lock);
628 if (xfrm_tunnel_check(skb, x, family)) {
629 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMODEERROR);
633 seq_hi = htonl(xfrm_replay_seqhi(x, seq));
635 XFRM_SKB_CB(skb)->seq.input.low = seq;
636 XFRM_SKB_CB(skb)->seq.input.hi = seq_hi;
641 nexthdr = x->type_offload->input_tail(x, skb);
643 nexthdr = x->type->input(x, skb);
645 if (nexthdr == -EINPROGRESS)
652 if (nexthdr == -EBADMSG) {
653 xfrm_audit_state_icvfail(x, skb,
655 x->stats.integrity_failed++;
657 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEPROTOERROR);
661 /* only the first xfrm gets the encap type */
664 if (xfrm_replay_recheck(x, skb, seq)) {
665 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATESEQERROR);
669 xfrm_replay_advance(x, seq);
671 x->curlft.bytes += skb->len;
673 x->lastused = ktime_get_real_seconds();
675 spin_unlock(&x->lock);
677 XFRM_MODE_SKB_CB(skb)->protocol = nexthdr;
679 inner_mode = &x->inner_mode;
681 if (x->sel.family == AF_UNSPEC) {
682 inner_mode = xfrm_ip2inner_mode(x, XFRM_MODE_SKB_CB(skb)->protocol);
683 if (inner_mode == NULL) {
684 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMODEERROR);
689 if (xfrm_inner_mode_input(x, inner_mode, skb)) {
690 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMODEERROR);
694 if (x->outer_mode.flags & XFRM_MODE_FLAG_TUNNEL) {
700 * We need the inner address. However, we only get here for
701 * transport mode so the outer address is identical.
703 daddr = &x->id.daddr;
704 family = x->outer_mode.family;
706 err = xfrm_parse_spi(skb, nexthdr, &spi, &seq);
708 XFRM_INC_STATS(net, LINUX_MIB_XFRMINHDRERROR);
714 err = xfrm_rcv_cb(skb, family, x->type->proto, 0);
721 sp = skb_sec_path(skb);
724 if (skb_valid_dst(skb))
726 gro_cells_receive(&gro_cells, skb);
729 xo = xfrm_offload(skb);
731 xfrm_gro = xo->flags & XFRM_GRO;
735 afinfo = xfrm_state_afinfo_get_rcu(x->inner_mode.family);
737 err = afinfo->transport_finish(skb, xfrm_gro || async);
740 sp = skb_sec_path(skb);
743 if (skb_valid_dst(skb))
745 gro_cells_receive(&gro_cells, skb);
753 spin_unlock(&x->lock);
755 xfrm_rcv_cb(skb, family, x && x->type ? x->type->proto : nexthdr, -1);
759 EXPORT_SYMBOL(xfrm_input);
761 int xfrm_input_resume(struct sk_buff *skb, int nexthdr)
763 return xfrm_input(skb, nexthdr, 0, -1);
765 EXPORT_SYMBOL(xfrm_input_resume);
767 static void xfrm_trans_reinject(struct work_struct *work)
769 struct xfrm_trans_tasklet *trans = container_of(work, struct xfrm_trans_tasklet, work);
770 struct sk_buff_head queue;
773 __skb_queue_head_init(&queue);
774 spin_lock_bh(&trans->queue_lock);
775 skb_queue_splice_init(&trans->queue, &queue);
776 spin_unlock_bh(&trans->queue_lock);
779 while ((skb = __skb_dequeue(&queue)))
780 XFRM_TRANS_SKB_CB(skb)->finish(XFRM_TRANS_SKB_CB(skb)->net,
785 int xfrm_trans_queue_net(struct net *net, struct sk_buff *skb,
786 int (*finish)(struct net *, struct sock *,
789 struct xfrm_trans_tasklet *trans;
791 trans = this_cpu_ptr(&xfrm_trans_tasklet);
793 if (skb_queue_len(&trans->queue) >= READ_ONCE(netdev_max_backlog))
796 BUILD_BUG_ON(sizeof(struct xfrm_trans_cb) > sizeof(skb->cb));
798 XFRM_TRANS_SKB_CB(skb)->finish = finish;
799 XFRM_TRANS_SKB_CB(skb)->net = net;
800 spin_lock_bh(&trans->queue_lock);
801 __skb_queue_tail(&trans->queue, skb);
802 spin_unlock_bh(&trans->queue_lock);
803 schedule_work(&trans->work);
806 EXPORT_SYMBOL(xfrm_trans_queue_net);
808 int xfrm_trans_queue(struct sk_buff *skb,
809 int (*finish)(struct net *, struct sock *,
812 return xfrm_trans_queue_net(dev_net(skb->dev), skb, finish);
814 EXPORT_SYMBOL(xfrm_trans_queue);
816 void __init xfrm_input_init(void)
821 init_dummy_netdev(&xfrm_napi_dev);
822 err = gro_cells_init(&gro_cells, &xfrm_napi_dev);
824 gro_cells.cells = NULL;
826 for_each_possible_cpu(i) {
827 struct xfrm_trans_tasklet *trans;
829 trans = &per_cpu(xfrm_trans_tasklet, i);
830 spin_lock_init(&trans->queue_lock);
831 __skb_queue_head_init(&trans->queue);
832 INIT_WORK(&trans->work, xfrm_trans_reinject);