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));
134 sp->verified_cnt = 0;
138 EXPORT_SYMBOL(secpath_set);
140 /* Fetch spi and seq from ipsec header */
142 int xfrm_parse_spi(struct sk_buff *skb, u8 nexthdr, __be32 *spi, __be32 *seq)
144 int offset, offset_seq;
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);
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);
159 if (!pskb_may_pull(skb, sizeof(struct ip_comp_hdr)))
161 *spi = htonl(ntohs(*(__be16 *)(skb_transport_header(skb) + 2)));
168 if (!pskb_may_pull(skb, hlen))
171 *spi = *(__be32 *)(skb_transport_header(skb) + offset);
172 *seq = *(__be32 *)(skb_transport_header(skb) + offset_seq);
175 EXPORT_SYMBOL(xfrm_parse_spi);
177 static int xfrm4_remove_beet_encap(struct xfrm_state *x, struct sk_buff *skb)
183 skb->protocol = htons(ETH_P_IP);
185 if (unlikely(XFRM_MODE_SKB_CB(skb)->protocol == IPPROTO_BEETPH)) {
186 struct ip_beet_phdr *ph;
189 if (!pskb_may_pull(skb, sizeof(*ph)))
192 ph = (struct ip_beet_phdr *)skb->data;
194 phlen = sizeof(*ph) + ph->padlen;
195 optlen = ph->hdrlen * 8 + (IPV4_BEET_PHMAXLEN - phlen);
196 if (optlen < 0 || optlen & 3 || optlen > 250)
199 XFRM_MODE_SKB_CB(skb)->protocol = ph->nexthdr;
201 if (!pskb_may_pull(skb, phlen))
203 __skb_pull(skb, phlen);
206 skb_push(skb, sizeof(*iph));
207 skb_reset_network_header(skb);
208 skb_mac_header_rebuild(skb);
210 xfrm4_beet_make_header(skb);
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;
219 iph->check = ip_fast_csum(skb_network_header(skb), iph->ihl);
225 static void ipip_ecn_decapsulate(struct sk_buff *skb)
227 struct iphdr *inner_iph = ipip_hdr(skb);
229 if (INET_ECN_is_ce(XFRM_MODE_SKB_CB(skb)->tos))
230 IP_ECN_set_ce(inner_iph);
233 static int xfrm4_remove_tunnel_encap(struct xfrm_state *x, struct sk_buff *skb)
237 skb->protocol = htons(ETH_P_IP);
239 if (!pskb_may_pull(skb, sizeof(struct iphdr)))
242 err = skb_unclone(skb, GFP_ATOMIC);
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);
251 skb_reset_network_header(skb);
252 skb_mac_header_rebuild(skb);
254 eth_hdr(skb)->h_proto = skb->protocol;
262 static void ipip6_ecn_decapsulate(struct sk_buff *skb)
264 struct ipv6hdr *inner_iph = ipipv6_hdr(skb);
266 if (INET_ECN_is_ce(XFRM_MODE_SKB_CB(skb)->tos))
267 IP6_ECN_set_ce(skb, inner_iph);
270 static int xfrm6_remove_tunnel_encap(struct xfrm_state *x, struct sk_buff *skb)
274 skb->protocol = htons(ETH_P_IPV6);
276 if (!pskb_may_pull(skb, sizeof(struct ipv6hdr)))
279 err = skb_unclone(skb, GFP_ATOMIC);
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);
288 skb_reset_network_header(skb);
289 skb_mac_header_rebuild(skb);
291 eth_hdr(skb)->h_proto = skb->protocol;
299 static int xfrm6_remove_beet_encap(struct xfrm_state *x, struct sk_buff *skb)
301 struct ipv6hdr *ip6h;
302 int size = sizeof(struct ipv6hdr);
305 skb->protocol = htons(ETH_P_IPV6);
307 err = skb_cow_head(skb, size + skb->mac_len);
311 __skb_push(skb, size);
312 skb_reset_network_header(skb);
313 skb_mac_header_rebuild(skb);
315 xfrm6_beet_make_header(skb);
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;
326 /* Remove encapsulation header.
328 * The IP header will be moved over the top of the encapsulation
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
337 xfrm_inner_mode_encap_remove(struct xfrm_state *x,
340 switch (x->props.mode) {
342 switch (x->sel.family) {
344 return xfrm4_remove_beet_encap(x, skb);
346 return xfrm6_remove_beet_encap(x, skb);
349 case XFRM_MODE_TUNNEL:
350 switch (XFRM_MODE_SKB_CB(skb)->protocol) {
352 return xfrm4_remove_tunnel_encap(x, skb);
354 return xfrm6_remove_tunnel_encap(x, skb);
364 static int xfrm_prepare_input(struct xfrm_state *x, struct sk_buff *skb)
366 switch (x->props.family) {
368 xfrm4_extract_header(skb);
371 xfrm6_extract_header(skb);
375 return -EAFNOSUPPORT;
378 return xfrm_inner_mode_encap_remove(x, skb);
381 /* Remove encapsulation header.
383 * The IP header will be moved over the top of the encapsulation header.
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.
389 static int xfrm4_transport_input(struct xfrm_state *x, struct sk_buff *skb)
391 int ihl = skb->data - skb_transport_header(skb);
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;
398 ip_hdr(skb)->tot_len = htons(skb->len + ihl);
399 skb_reset_transport_header(skb);
403 static int xfrm6_transport_input(struct xfrm_state *x, struct sk_buff *skb)
405 #if IS_ENABLED(CONFIG_IPV6)
406 int ihl = skb->data - skb_transport_header(skb);
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;
413 ipv6_hdr(skb)->payload_len = htons(skb->len + ihl -
414 sizeof(struct ipv6hdr));
415 skb_reset_transport_header(skb);
419 return -EAFNOSUPPORT;
423 static int xfrm_inner_mode_input(struct xfrm_state *x,
426 switch (x->props.mode) {
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);
436 case XFRM_MODE_ROUTEOPTIMIZATION:
447 int xfrm_input(struct sk_buff *skb, int nexthdr, __be32 spi, int encap_type)
449 const struct xfrm_state_afinfo *afinfo;
450 struct net *net = dev_net(skb->dev);
454 struct xfrm_state *x = NULL;
455 xfrm_address_t *daddr;
456 u32 mark = skb->mark;
457 unsigned int family = AF_UNSPEC;
460 bool xfrm_gro = false;
461 bool crypto_done = false;
462 struct xfrm_offload *xo = xfrm_offload(skb);
465 if (encap_type < 0) {
466 x = xfrm_input_state(skb);
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);
473 LINUX_MIB_XFRMINSTATEINVALID);
475 if (encap_type == -1)
480 family = x->props.family;
482 /* An encap_type of -1 indicates async resumption. */
483 if (encap_type == -1) {
485 seq = XFRM_SKB_CB(skb)->seq.input.low;
489 /* encap_type < -1 indicates a GRO call. */
491 seq = XFRM_SPI_SKB_CB(skb)->seq;
493 if (xo && (xo->flags & CRYPTO_DONE)) {
495 family = XFRM_SPI_SKB_CB(skb)->family;
497 if (!(xo->status & CRYPTO_SUCCESS)) {
499 (CRYPTO_TRANSPORT_AH_AUTH_FAILED |
500 CRYPTO_TRANSPORT_ESP_AUTH_FAILED |
501 CRYPTO_TUNNEL_AH_AUTH_FAILED |
502 CRYPTO_TUNNEL_ESP_AUTH_FAILED)) {
504 xfrm_audit_state_icvfail(x, skb,
506 x->stats.integrity_failed++;
507 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEPROTOERROR);
511 if (xo->status & CRYPTO_INVALID_PROTOCOL) {
512 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEPROTOERROR);
516 XFRM_INC_STATS(net, LINUX_MIB_XFRMINBUFFERERROR);
520 if (xfrm_parse_spi(skb, nexthdr, &spi, &seq)) {
521 XFRM_INC_STATS(net, LINUX_MIB_XFRMINHDRERROR);
529 family = XFRM_SPI_SKB_CB(skb)->family;
531 /* if tunnel is present override skb->mark value with tunnel i_key */
534 if (XFRM_TUNNEL_SKB_CB(skb)->tunnel.ip4)
535 mark = be32_to_cpu(XFRM_TUNNEL_SKB_CB(skb)->tunnel.ip4->parms.i_key);
538 if (XFRM_TUNNEL_SKB_CB(skb)->tunnel.ip6)
539 mark = be32_to_cpu(XFRM_TUNNEL_SKB_CB(skb)->tunnel.ip6->parms.i_key);
543 sp = secpath_set(skb);
545 XFRM_INC_STATS(net, LINUX_MIB_XFRMINERROR);
550 if (!spi && xfrm_parse_spi(skb, nexthdr, &spi, &seq)) {
552 XFRM_INC_STATS(net, LINUX_MIB_XFRMINHDRERROR);
556 daddr = (xfrm_address_t *)(skb_network_header(skb) +
557 XFRM_SPI_SKB_CB(skb)->daddroff);
559 sp = skb_sec_path(skb);
561 if (sp->len == XFRM_MAX_DEPTH) {
563 XFRM_INC_STATS(net, LINUX_MIB_XFRMINBUFFERERROR);
567 x = xfrm_state_lookup(net, mark, daddr, spi, nexthdr, family);
570 XFRM_INC_STATS(net, LINUX_MIB_XFRMINNOSTATES);
571 xfrm_audit_state_notfound(skb, family, spi, seq);
575 skb->mark = xfrm_smark_get(skb->mark, x);
577 sp->xvec[sp->len++] = x;
581 XFRM_INC_STATS(net, LINUX_MIB_XFRMINERROR);
588 if (unlikely(x->km.state != XFRM_STATE_VALID)) {
589 if (x->km.state == XFRM_STATE_ACQ)
590 XFRM_INC_STATS(net, LINUX_MIB_XFRMACQUIREERROR);
593 LINUX_MIB_XFRMINSTATEINVALID);
597 if ((x->encap ? x->encap->encap_type : 0) != encap_type) {
598 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMISMATCH);
602 if (xfrm_replay_check(x, skb, seq)) {
603 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATESEQERROR);
607 if (xfrm_state_check_expire(x)) {
608 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEEXPIRED);
612 spin_unlock(&x->lock);
614 if (xfrm_tunnel_check(skb, x, family)) {
615 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMODEERROR);
619 seq_hi = htonl(xfrm_replay_seqhi(x, seq));
621 XFRM_SKB_CB(skb)->seq.input.low = seq;
622 XFRM_SKB_CB(skb)->seq.input.hi = seq_hi;
627 nexthdr = x->type_offload->input_tail(x, skb);
629 nexthdr = x->type->input(x, skb);
631 if (nexthdr == -EINPROGRESS)
638 if (nexthdr == -EBADMSG) {
639 xfrm_audit_state_icvfail(x, skb,
641 x->stats.integrity_failed++;
643 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEPROTOERROR);
647 /* only the first xfrm gets the encap type */
650 if (xfrm_replay_recheck(x, skb, seq)) {
651 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATESEQERROR);
655 xfrm_replay_advance(x, seq);
657 x->curlft.bytes += skb->len;
659 x->lastused = ktime_get_real_seconds();
661 spin_unlock(&x->lock);
663 XFRM_MODE_SKB_CB(skb)->protocol = nexthdr;
665 if (xfrm_inner_mode_input(x, skb)) {
666 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMODEERROR);
670 if (x->outer_mode.flags & XFRM_MODE_FLAG_TUNNEL) {
676 * We need the inner address. However, we only get here for
677 * transport mode so the outer address is identical.
679 daddr = &x->id.daddr;
680 family = x->props.family;
682 err = xfrm_parse_spi(skb, nexthdr, &spi, &seq);
684 XFRM_INC_STATS(net, LINUX_MIB_XFRMINHDRERROR);
690 err = xfrm_rcv_cb(skb, family, x->type->proto, 0);
697 sp = skb_sec_path(skb);
700 if (skb_valid_dst(skb))
702 gro_cells_receive(&gro_cells, skb);
705 xo = xfrm_offload(skb);
707 xfrm_gro = xo->flags & XFRM_GRO;
711 afinfo = xfrm_state_afinfo_get_rcu(x->props.family);
713 err = afinfo->transport_finish(skb, xfrm_gro || async);
716 sp = skb_sec_path(skb);
719 if (skb_valid_dst(skb))
721 gro_cells_receive(&gro_cells, skb);
729 spin_unlock(&x->lock);
731 xfrm_rcv_cb(skb, family, x && x->type ? x->type->proto : nexthdr, -1);
735 EXPORT_SYMBOL(xfrm_input);
737 int xfrm_input_resume(struct sk_buff *skb, int nexthdr)
739 return xfrm_input(skb, nexthdr, 0, -1);
741 EXPORT_SYMBOL(xfrm_input_resume);
743 static void xfrm_trans_reinject(struct work_struct *work)
745 struct xfrm_trans_tasklet *trans = container_of(work, struct xfrm_trans_tasklet, work);
746 struct sk_buff_head queue;
749 __skb_queue_head_init(&queue);
750 spin_lock_bh(&trans->queue_lock);
751 skb_queue_splice_init(&trans->queue, &queue);
752 spin_unlock_bh(&trans->queue_lock);
755 while ((skb = __skb_dequeue(&queue)))
756 XFRM_TRANS_SKB_CB(skb)->finish(XFRM_TRANS_SKB_CB(skb)->net,
761 int xfrm_trans_queue_net(struct net *net, struct sk_buff *skb,
762 int (*finish)(struct net *, struct sock *,
765 struct xfrm_trans_tasklet *trans;
767 trans = this_cpu_ptr(&xfrm_trans_tasklet);
769 if (skb_queue_len(&trans->queue) >= READ_ONCE(netdev_max_backlog))
772 BUILD_BUG_ON(sizeof(struct xfrm_trans_cb) > sizeof(skb->cb));
774 XFRM_TRANS_SKB_CB(skb)->finish = finish;
775 XFRM_TRANS_SKB_CB(skb)->net = net;
776 spin_lock_bh(&trans->queue_lock);
777 __skb_queue_tail(&trans->queue, skb);
778 spin_unlock_bh(&trans->queue_lock);
779 schedule_work(&trans->work);
782 EXPORT_SYMBOL(xfrm_trans_queue_net);
784 int xfrm_trans_queue(struct sk_buff *skb,
785 int (*finish)(struct net *, struct sock *,
788 return xfrm_trans_queue_net(dev_net(skb->dev), skb, finish);
790 EXPORT_SYMBOL(xfrm_trans_queue);
792 void __init xfrm_input_init(void)
797 init_dummy_netdev(&xfrm_napi_dev);
798 err = gro_cells_init(&gro_cells, &xfrm_napi_dev);
800 gro_cells.cells = NULL;
802 for_each_possible_cpu(i) {
803 struct xfrm_trans_tasklet *trans;
805 trans = &per_cpu(xfrm_trans_tasklet, i);
806 spin_lock_init(&trans->queue_lock);
807 __skb_queue_head_init(&trans->queue);
808 INIT_WORK(&trans->work, xfrm_trans_reinject);