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>
24 #include <net/hotdata.h>
26 #include "xfrm_inout.h"
28 struct xfrm_trans_tasklet {
29 struct work_struct work;
30 spinlock_t queue_lock;
31 struct sk_buff_head queue;
34 struct xfrm_trans_cb {
36 struct inet_skb_parm h4;
37 #if IS_ENABLED(CONFIG_IPV6)
38 struct inet6_skb_parm h6;
41 int (*finish)(struct net *net, struct sock *sk, struct sk_buff *skb);
45 #define XFRM_TRANS_SKB_CB(__skb) ((struct xfrm_trans_cb *)&((__skb)->cb[0]))
47 static DEFINE_SPINLOCK(xfrm_input_afinfo_lock);
48 static struct xfrm_input_afinfo const __rcu *xfrm_input_afinfo[2][AF_INET6 + 1];
50 static struct gro_cells gro_cells;
51 static struct net_device xfrm_napi_dev;
53 static DEFINE_PER_CPU(struct xfrm_trans_tasklet, xfrm_trans_tasklet);
55 int xfrm_input_register_afinfo(const struct xfrm_input_afinfo *afinfo)
59 if (WARN_ON(afinfo->family > AF_INET6))
62 spin_lock_bh(&xfrm_input_afinfo_lock);
63 if (unlikely(xfrm_input_afinfo[afinfo->is_ipip][afinfo->family]))
66 rcu_assign_pointer(xfrm_input_afinfo[afinfo->is_ipip][afinfo->family], afinfo);
67 spin_unlock_bh(&xfrm_input_afinfo_lock);
70 EXPORT_SYMBOL(xfrm_input_register_afinfo);
72 int xfrm_input_unregister_afinfo(const struct xfrm_input_afinfo *afinfo)
76 spin_lock_bh(&xfrm_input_afinfo_lock);
77 if (likely(xfrm_input_afinfo[afinfo->is_ipip][afinfo->family])) {
78 if (unlikely(xfrm_input_afinfo[afinfo->is_ipip][afinfo->family] != afinfo))
81 RCU_INIT_POINTER(xfrm_input_afinfo[afinfo->is_ipip][afinfo->family], NULL);
83 spin_unlock_bh(&xfrm_input_afinfo_lock);
87 EXPORT_SYMBOL(xfrm_input_unregister_afinfo);
89 static const struct xfrm_input_afinfo *xfrm_input_get_afinfo(u8 family, bool is_ipip)
91 const struct xfrm_input_afinfo *afinfo;
93 if (WARN_ON_ONCE(family > AF_INET6))
97 afinfo = rcu_dereference(xfrm_input_afinfo[is_ipip][family]);
98 if (unlikely(!afinfo))
103 static int xfrm_rcv_cb(struct sk_buff *skb, unsigned int family, u8 protocol,
106 bool is_ipip = (protocol == IPPROTO_IPIP || protocol == IPPROTO_IPV6);
107 const struct xfrm_input_afinfo *afinfo;
110 afinfo = xfrm_input_get_afinfo(family, is_ipip);
112 return -EAFNOSUPPORT;
114 ret = afinfo->callback(skb, protocol, err);
120 struct sec_path *secpath_set(struct sk_buff *skb)
122 struct sec_path *sp, *tmp = skb_ext_find(skb, SKB_EXT_SEC_PATH);
124 sp = skb_ext_add(skb, SKB_EXT_SEC_PATH);
128 if (tmp) /* reused existing one (was COW'd if needed) */
131 /* allocated new secpath */
132 memset(sp->ovec, 0, sizeof(sp->ovec));
135 sp->verified_cnt = 0;
139 EXPORT_SYMBOL(secpath_set);
141 /* Fetch spi and seq from ipsec header */
143 int xfrm_parse_spi(struct sk_buff *skb, u8 nexthdr, __be32 *spi, __be32 *seq)
145 int offset, offset_seq;
150 hlen = sizeof(struct ip_auth_hdr);
151 offset = offsetof(struct ip_auth_hdr, spi);
152 offset_seq = offsetof(struct ip_auth_hdr, seq_no);
155 hlen = sizeof(struct ip_esp_hdr);
156 offset = offsetof(struct ip_esp_hdr, spi);
157 offset_seq = offsetof(struct ip_esp_hdr, seq_no);
160 if (!pskb_may_pull(skb, sizeof(struct ip_comp_hdr)))
162 *spi = htonl(ntohs(*(__be16 *)(skb_transport_header(skb) + 2)));
169 if (!pskb_may_pull(skb, hlen))
172 *spi = *(__be32 *)(skb_transport_header(skb) + offset);
173 *seq = *(__be32 *)(skb_transport_header(skb) + offset_seq);
176 EXPORT_SYMBOL(xfrm_parse_spi);
178 static int xfrm4_remove_beet_encap(struct xfrm_state *x, struct sk_buff *skb)
184 skb->protocol = htons(ETH_P_IP);
186 if (unlikely(XFRM_MODE_SKB_CB(skb)->protocol == IPPROTO_BEETPH)) {
187 struct ip_beet_phdr *ph;
190 if (!pskb_may_pull(skb, sizeof(*ph)))
193 ph = (struct ip_beet_phdr *)skb->data;
195 phlen = sizeof(*ph) + ph->padlen;
196 optlen = ph->hdrlen * 8 + (IPV4_BEET_PHMAXLEN - phlen);
197 if (optlen < 0 || optlen & 3 || optlen > 250)
200 XFRM_MODE_SKB_CB(skb)->protocol = ph->nexthdr;
202 if (!pskb_may_pull(skb, phlen))
204 __skb_pull(skb, phlen);
207 skb_push(skb, sizeof(*iph));
208 skb_reset_network_header(skb);
209 skb_mac_header_rebuild(skb);
211 xfrm4_beet_make_header(skb);
215 iph->ihl += optlen / 4;
216 iph->tot_len = htons(skb->len);
217 iph->daddr = x->sel.daddr.a4;
218 iph->saddr = x->sel.saddr.a4;
220 iph->check = ip_fast_csum(skb_network_header(skb), iph->ihl);
226 static void ipip_ecn_decapsulate(struct sk_buff *skb)
228 struct iphdr *inner_iph = ipip_hdr(skb);
230 if (INET_ECN_is_ce(XFRM_MODE_SKB_CB(skb)->tos))
231 IP_ECN_set_ce(inner_iph);
234 static int xfrm4_remove_tunnel_encap(struct xfrm_state *x, struct sk_buff *skb)
238 skb->protocol = htons(ETH_P_IP);
240 if (!pskb_may_pull(skb, sizeof(struct iphdr)))
243 err = skb_unclone(skb, GFP_ATOMIC);
247 if (x->props.flags & XFRM_STATE_DECAP_DSCP)
248 ipv4_copy_dscp(XFRM_MODE_SKB_CB(skb)->tos, ipip_hdr(skb));
249 if (!(x->props.flags & XFRM_STATE_NOECN))
250 ipip_ecn_decapsulate(skb);
252 skb_reset_network_header(skb);
253 skb_mac_header_rebuild(skb);
255 eth_hdr(skb)->h_proto = skb->protocol;
263 static void ipip6_ecn_decapsulate(struct sk_buff *skb)
265 struct ipv6hdr *inner_iph = ipipv6_hdr(skb);
267 if (INET_ECN_is_ce(XFRM_MODE_SKB_CB(skb)->tos))
268 IP6_ECN_set_ce(skb, inner_iph);
271 static int xfrm6_remove_tunnel_encap(struct xfrm_state *x, struct sk_buff *skb)
275 skb->protocol = htons(ETH_P_IPV6);
277 if (!pskb_may_pull(skb, sizeof(struct ipv6hdr)))
280 err = skb_unclone(skb, GFP_ATOMIC);
284 if (x->props.flags & XFRM_STATE_DECAP_DSCP)
285 ipv6_copy_dscp(XFRM_MODE_SKB_CB(skb)->tos, ipipv6_hdr(skb));
286 if (!(x->props.flags & XFRM_STATE_NOECN))
287 ipip6_ecn_decapsulate(skb);
289 skb_reset_network_header(skb);
290 skb_mac_header_rebuild(skb);
292 eth_hdr(skb)->h_proto = skb->protocol;
300 static int xfrm6_remove_beet_encap(struct xfrm_state *x, struct sk_buff *skb)
302 struct ipv6hdr *ip6h;
303 int size = sizeof(struct ipv6hdr);
306 skb->protocol = htons(ETH_P_IPV6);
308 err = skb_cow_head(skb, size + skb->mac_len);
312 __skb_push(skb, size);
313 skb_reset_network_header(skb);
314 skb_mac_header_rebuild(skb);
316 xfrm6_beet_make_header(skb);
318 ip6h = ipv6_hdr(skb);
319 ip6h->payload_len = htons(skb->len - size);
320 ip6h->daddr = x->sel.daddr.in6;
321 ip6h->saddr = x->sel.saddr.in6;
327 /* Remove encapsulation header.
329 * The IP header will be moved over the top of the encapsulation
332 * On entry, the transport header shall point to where the IP header
333 * should be and the network header shall be set to where the IP
334 * header currently is. skb->data shall point to the start of the
338 xfrm_inner_mode_encap_remove(struct xfrm_state *x,
341 switch (x->props.mode) {
343 switch (x->sel.family) {
345 return xfrm4_remove_beet_encap(x, skb);
347 return xfrm6_remove_beet_encap(x, skb);
350 case XFRM_MODE_TUNNEL:
351 switch (XFRM_MODE_SKB_CB(skb)->protocol) {
353 return xfrm4_remove_tunnel_encap(x, skb);
355 return xfrm6_remove_tunnel_encap(x, skb);
365 static int xfrm_prepare_input(struct xfrm_state *x, struct sk_buff *skb)
367 switch (x->props.family) {
369 xfrm4_extract_header(skb);
372 xfrm6_extract_header(skb);
376 return -EAFNOSUPPORT;
379 return xfrm_inner_mode_encap_remove(x, skb);
382 /* Remove encapsulation header.
384 * The IP header will be moved over the top of the encapsulation header.
386 * On entry, skb_transport_header() shall point to where the IP header
387 * should be and skb_network_header() shall be set to where the IP header
388 * currently is. skb->data shall point to the start of the payload.
390 static int xfrm4_transport_input(struct xfrm_state *x, struct sk_buff *skb)
392 struct xfrm_offload *xo = xfrm_offload(skb);
393 int ihl = skb->data - skb_transport_header(skb);
395 if (skb->transport_header != skb->network_header) {
396 memmove(skb_transport_header(skb),
397 skb_network_header(skb), ihl);
400 skb_mac_header_was_set(skb) ? skb_mac_header_len(skb) : 0;
401 skb->network_header = skb->transport_header;
403 ip_hdr(skb)->tot_len = htons(skb->len + ihl);
404 skb_reset_transport_header(skb);
408 static int xfrm6_transport_input(struct xfrm_state *x, struct sk_buff *skb)
410 #if IS_ENABLED(CONFIG_IPV6)
411 struct xfrm_offload *xo = xfrm_offload(skb);
412 int ihl = skb->data - skb_transport_header(skb);
414 if (skb->transport_header != skb->network_header) {
415 memmove(skb_transport_header(skb),
416 skb_network_header(skb), ihl);
419 skb_mac_header_was_set(skb) ? skb_mac_header_len(skb) : 0;
420 skb->network_header = skb->transport_header;
422 ipv6_hdr(skb)->payload_len = htons(skb->len + ihl -
423 sizeof(struct ipv6hdr));
424 skb_reset_transport_header(skb);
428 return -EAFNOSUPPORT;
432 static int xfrm_inner_mode_input(struct xfrm_state *x,
435 switch (x->props.mode) {
437 case XFRM_MODE_TUNNEL:
438 return xfrm_prepare_input(x, skb);
439 case XFRM_MODE_TRANSPORT:
440 if (x->props.family == AF_INET)
441 return xfrm4_transport_input(x, skb);
442 if (x->props.family == AF_INET6)
443 return xfrm6_transport_input(x, skb);
445 case XFRM_MODE_ROUTEOPTIMIZATION:
456 int xfrm_input(struct sk_buff *skb, int nexthdr, __be32 spi, int encap_type)
458 const struct xfrm_state_afinfo *afinfo;
459 struct net *net = dev_net(skb->dev);
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 || (xo && xo->flags & XFRM_GRO)) {
475 x = xfrm_input_state(skb);
477 if (unlikely(x->dir && x->dir != XFRM_SA_DIR_IN)) {
478 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEDIRERROR);
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->props.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 seq = XFRM_SPI_SKB_CB(skb)->seq;
505 if (xo && (xo->flags & CRYPTO_DONE)) {
507 family = XFRM_SPI_SKB_CB(skb)->family;
509 if (!(xo->status & CRYPTO_SUCCESS)) {
511 (CRYPTO_TRANSPORT_AH_AUTH_FAILED |
512 CRYPTO_TRANSPORT_ESP_AUTH_FAILED |
513 CRYPTO_TUNNEL_AH_AUTH_FAILED |
514 CRYPTO_TUNNEL_ESP_AUTH_FAILED)) {
516 xfrm_audit_state_icvfail(x, skb,
518 x->stats.integrity_failed++;
519 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEPROTOERROR);
523 if (xo->status & CRYPTO_INVALID_PROTOCOL) {
524 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEPROTOERROR);
528 XFRM_INC_STATS(net, LINUX_MIB_XFRMINBUFFERERROR);
532 if (xfrm_parse_spi(skb, nexthdr, &spi, &seq)) {
533 XFRM_INC_STATS(net, LINUX_MIB_XFRMINHDRERROR);
541 family = XFRM_SPI_SKB_CB(skb)->family;
543 /* if tunnel is present override skb->mark value with tunnel i_key */
546 if (XFRM_TUNNEL_SKB_CB(skb)->tunnel.ip4)
547 mark = be32_to_cpu(XFRM_TUNNEL_SKB_CB(skb)->tunnel.ip4->parms.i_key);
550 if (XFRM_TUNNEL_SKB_CB(skb)->tunnel.ip6)
551 mark = be32_to_cpu(XFRM_TUNNEL_SKB_CB(skb)->tunnel.ip6->parms.i_key);
555 sp = secpath_set(skb);
557 XFRM_INC_STATS(net, LINUX_MIB_XFRMINERROR);
562 if (!spi && xfrm_parse_spi(skb, nexthdr, &spi, &seq)) {
564 XFRM_INC_STATS(net, LINUX_MIB_XFRMINHDRERROR);
568 daddr = (xfrm_address_t *)(skb_network_header(skb) +
569 XFRM_SPI_SKB_CB(skb)->daddroff);
571 sp = skb_sec_path(skb);
573 if (sp->len == XFRM_MAX_DEPTH) {
575 XFRM_INC_STATS(net, LINUX_MIB_XFRMINBUFFERERROR);
579 x = xfrm_state_lookup(net, mark, daddr, spi, nexthdr, family);
582 XFRM_INC_STATS(net, LINUX_MIB_XFRMINNOSTATES);
583 xfrm_audit_state_notfound(skb, family, spi, seq);
587 if (unlikely(x->dir && x->dir != XFRM_SA_DIR_IN)) {
588 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEDIRERROR);
593 skb->mark = xfrm_smark_get(skb->mark, x);
595 sp->xvec[sp->len++] = x;
599 XFRM_INC_STATS(net, LINUX_MIB_XFRMINERROR);
606 if (unlikely(x->km.state != XFRM_STATE_VALID)) {
607 if (x->km.state == XFRM_STATE_ACQ)
608 XFRM_INC_STATS(net, LINUX_MIB_XFRMACQUIREERROR);
611 LINUX_MIB_XFRMINSTATEINVALID);
615 if ((x->encap ? x->encap->encap_type : 0) != encap_type) {
616 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMISMATCH);
620 if (xfrm_replay_check(x, skb, seq)) {
621 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATESEQERROR);
625 if (xfrm_state_check_expire(x)) {
626 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEEXPIRED);
630 spin_unlock(&x->lock);
632 if (xfrm_tunnel_check(skb, x, family)) {
633 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMODEERROR);
637 seq_hi = htonl(xfrm_replay_seqhi(x, seq));
639 XFRM_SKB_CB(skb)->seq.input.low = seq;
640 XFRM_SKB_CB(skb)->seq.input.hi = seq_hi;
645 nexthdr = x->type_offload->input_tail(x, skb);
647 nexthdr = x->type->input(x, skb);
649 if (nexthdr == -EINPROGRESS)
656 if (nexthdr == -EBADMSG) {
657 xfrm_audit_state_icvfail(x, skb,
659 x->stats.integrity_failed++;
661 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEPROTOERROR);
665 /* only the first xfrm gets the encap type */
668 if (xfrm_replay_recheck(x, skb, seq)) {
669 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATESEQERROR);
673 xfrm_replay_advance(x, seq);
675 x->curlft.bytes += skb->len;
677 x->lastused = ktime_get_real_seconds();
679 spin_unlock(&x->lock);
681 XFRM_MODE_SKB_CB(skb)->protocol = nexthdr;
683 if (xfrm_inner_mode_input(x, 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->props.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);
718 if (skb_valid_dst(skb))
720 gro_cells_receive(&gro_cells, skb);
723 xo = xfrm_offload(skb);
725 xfrm_gro = xo->flags & XFRM_GRO;
729 afinfo = xfrm_state_afinfo_get_rcu(x->props.family);
731 err = afinfo->transport_finish(skb, xfrm_gro || async);
734 sp = skb_sec_path(skb);
737 if (skb_valid_dst(skb))
739 gro_cells_receive(&gro_cells, skb);
747 spin_unlock(&x->lock);
749 xfrm_rcv_cb(skb, family, x && x->type ? x->type->proto : nexthdr, -1);
753 EXPORT_SYMBOL(xfrm_input);
755 int xfrm_input_resume(struct sk_buff *skb, int nexthdr)
757 return xfrm_input(skb, nexthdr, 0, -1);
759 EXPORT_SYMBOL(xfrm_input_resume);
761 static void xfrm_trans_reinject(struct work_struct *work)
763 struct xfrm_trans_tasklet *trans = container_of(work, struct xfrm_trans_tasklet, work);
764 struct sk_buff_head queue;
767 __skb_queue_head_init(&queue);
768 spin_lock_bh(&trans->queue_lock);
769 skb_queue_splice_init(&trans->queue, &queue);
770 spin_unlock_bh(&trans->queue_lock);
773 while ((skb = __skb_dequeue(&queue)))
774 XFRM_TRANS_SKB_CB(skb)->finish(XFRM_TRANS_SKB_CB(skb)->net,
779 int xfrm_trans_queue_net(struct net *net, struct sk_buff *skb,
780 int (*finish)(struct net *, struct sock *,
783 struct xfrm_trans_tasklet *trans;
785 trans = this_cpu_ptr(&xfrm_trans_tasklet);
787 if (skb_queue_len(&trans->queue) >= READ_ONCE(net_hotdata.max_backlog))
790 BUILD_BUG_ON(sizeof(struct xfrm_trans_cb) > sizeof(skb->cb));
792 XFRM_TRANS_SKB_CB(skb)->finish = finish;
793 XFRM_TRANS_SKB_CB(skb)->net = net;
794 spin_lock_bh(&trans->queue_lock);
795 __skb_queue_tail(&trans->queue, skb);
796 spin_unlock_bh(&trans->queue_lock);
797 schedule_work(&trans->work);
800 EXPORT_SYMBOL(xfrm_trans_queue_net);
802 int xfrm_trans_queue(struct sk_buff *skb,
803 int (*finish)(struct net *, struct sock *,
806 return xfrm_trans_queue_net(dev_net(skb->dev), skb, finish);
808 EXPORT_SYMBOL(xfrm_trans_queue);
810 void __init xfrm_input_init(void)
815 init_dummy_netdev(&xfrm_napi_dev);
816 err = gro_cells_init(&gro_cells, &xfrm_napi_dev);
818 gro_cells.cells = NULL;
820 for_each_possible_cpu(i) {
821 struct xfrm_trans_tasklet *trans;
823 trans = &per_cpu(xfrm_trans_tasklet, i);
824 spin_lock_init(&trans->queue_lock);
825 __skb_queue_head_init(&trans->queue);
826 INIT_WORK(&trans->work, xfrm_trans_reinject);