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);
486 family = x->outer_mode.family;
488 /* An encap_type of -1 indicates async resumption. */
489 if (encap_type == -1) {
491 seq = XFRM_SKB_CB(skb)->seq.input.low;
495 /* encap_type < -1 indicates a GRO call. */
497 seq = XFRM_SPI_SKB_CB(skb)->seq;
499 if (xo && (xo->flags & CRYPTO_DONE)) {
501 family = XFRM_SPI_SKB_CB(skb)->family;
503 if (!(xo->status & CRYPTO_SUCCESS)) {
505 (CRYPTO_TRANSPORT_AH_AUTH_FAILED |
506 CRYPTO_TRANSPORT_ESP_AUTH_FAILED |
507 CRYPTO_TUNNEL_AH_AUTH_FAILED |
508 CRYPTO_TUNNEL_ESP_AUTH_FAILED)) {
510 xfrm_audit_state_icvfail(x, skb,
512 x->stats.integrity_failed++;
513 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEPROTOERROR);
517 if (xo->status & CRYPTO_INVALID_PROTOCOL) {
518 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEPROTOERROR);
522 XFRM_INC_STATS(net, LINUX_MIB_XFRMINBUFFERERROR);
526 if ((err = xfrm_parse_spi(skb, nexthdr, &spi, &seq)) != 0) {
527 XFRM_INC_STATS(net, LINUX_MIB_XFRMINHDRERROR);
535 family = XFRM_SPI_SKB_CB(skb)->family;
537 /* if tunnel is present override skb->mark value with tunnel i_key */
540 if (XFRM_TUNNEL_SKB_CB(skb)->tunnel.ip4)
541 mark = be32_to_cpu(XFRM_TUNNEL_SKB_CB(skb)->tunnel.ip4->parms.i_key);
544 if (XFRM_TUNNEL_SKB_CB(skb)->tunnel.ip6)
545 mark = be32_to_cpu(XFRM_TUNNEL_SKB_CB(skb)->tunnel.ip6->parms.i_key);
549 sp = secpath_set(skb);
551 XFRM_INC_STATS(net, LINUX_MIB_XFRMINERROR);
556 if (!spi && (err = xfrm_parse_spi(skb, nexthdr, &spi, &seq)) != 0) {
558 XFRM_INC_STATS(net, LINUX_MIB_XFRMINHDRERROR);
562 daddr = (xfrm_address_t *)(skb_network_header(skb) +
563 XFRM_SPI_SKB_CB(skb)->daddroff);
565 sp = skb_sec_path(skb);
567 if (sp->len == XFRM_MAX_DEPTH) {
569 XFRM_INC_STATS(net, LINUX_MIB_XFRMINBUFFERERROR);
573 x = xfrm_state_lookup(net, mark, daddr, spi, nexthdr, family);
576 XFRM_INC_STATS(net, LINUX_MIB_XFRMINNOSTATES);
577 xfrm_audit_state_notfound(skb, family, spi, seq);
581 skb->mark = xfrm_smark_get(skb->mark, x);
583 sp->xvec[sp->len++] = x;
587 XFRM_INC_STATS(net, LINUX_MIB_XFRMINERROR);
594 if (unlikely(x->km.state != XFRM_STATE_VALID)) {
595 if (x->km.state == XFRM_STATE_ACQ)
596 XFRM_INC_STATS(net, LINUX_MIB_XFRMACQUIREERROR);
599 LINUX_MIB_XFRMINSTATEINVALID);
603 if ((x->encap ? x->encap->encap_type : 0) != encap_type) {
604 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMISMATCH);
608 if (x->repl->check(x, skb, seq)) {
609 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATESEQERROR);
613 if (xfrm_state_check_expire(x)) {
614 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEEXPIRED);
618 spin_unlock(&x->lock);
620 if (xfrm_tunnel_check(skb, x, family)) {
621 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMODEERROR);
625 seq_hi = htonl(xfrm_replay_seqhi(x, seq));
627 XFRM_SKB_CB(skb)->seq.input.low = seq;
628 XFRM_SKB_CB(skb)->seq.input.hi = seq_hi;
633 nexthdr = x->type_offload->input_tail(x, skb);
635 nexthdr = x->type->input(x, skb);
637 if (nexthdr == -EINPROGRESS)
644 if (nexthdr == -EBADMSG) {
645 xfrm_audit_state_icvfail(x, skb,
647 x->stats.integrity_failed++;
649 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEPROTOERROR);
653 /* only the first xfrm gets the encap type */
656 if (async && x->repl->recheck(x, skb, seq)) {
657 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATESEQERROR);
661 x->repl->advance(x, seq);
663 x->curlft.bytes += skb->len;
666 spin_unlock(&x->lock);
668 XFRM_MODE_SKB_CB(skb)->protocol = nexthdr;
670 inner_mode = &x->inner_mode;
672 if (x->sel.family == AF_UNSPEC) {
673 inner_mode = xfrm_ip2inner_mode(x, XFRM_MODE_SKB_CB(skb)->protocol);
674 if (inner_mode == NULL) {
675 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMODEERROR);
680 if (xfrm_inner_mode_input(x, inner_mode, skb)) {
681 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMODEERROR);
685 if (x->outer_mode.flags & XFRM_MODE_FLAG_TUNNEL) {
691 * We need the inner address. However, we only get here for
692 * transport mode so the outer address is identical.
694 daddr = &x->id.daddr;
695 family = x->outer_mode.family;
697 err = xfrm_parse_spi(skb, nexthdr, &spi, &seq);
699 XFRM_INC_STATS(net, LINUX_MIB_XFRMINHDRERROR);
705 err = xfrm_rcv_cb(skb, family, x->type->proto, 0);
712 sp = skb_sec_path(skb);
716 gro_cells_receive(&gro_cells, skb);
719 xo = xfrm_offload(skb);
721 xfrm_gro = xo->flags & XFRM_GRO;
725 afinfo = xfrm_state_afinfo_get_rcu(x->inner_mode.family);
727 err = afinfo->transport_finish(skb, xfrm_gro || async);
730 sp = skb_sec_path(skb);
734 gro_cells_receive(&gro_cells, skb);
742 spin_unlock(&x->lock);
744 xfrm_rcv_cb(skb, family, x && x->type ? x->type->proto : nexthdr, -1);
748 EXPORT_SYMBOL(xfrm_input);
750 int xfrm_input_resume(struct sk_buff *skb, int nexthdr)
752 return xfrm_input(skb, nexthdr, 0, -1);
754 EXPORT_SYMBOL(xfrm_input_resume);
756 static void xfrm_trans_reinject(unsigned long data)
758 struct xfrm_trans_tasklet *trans = (void *)data;
759 struct sk_buff_head queue;
762 __skb_queue_head_init(&queue);
763 skb_queue_splice_init(&trans->queue, &queue);
765 while ((skb = __skb_dequeue(&queue)))
766 XFRM_TRANS_SKB_CB(skb)->finish(dev_net(skb->dev), NULL, skb);
769 int xfrm_trans_queue(struct sk_buff *skb,
770 int (*finish)(struct net *, struct sock *,
773 struct xfrm_trans_tasklet *trans;
775 trans = this_cpu_ptr(&xfrm_trans_tasklet);
777 if (skb_queue_len(&trans->queue) >= netdev_max_backlog)
780 XFRM_TRANS_SKB_CB(skb)->finish = finish;
781 __skb_queue_tail(&trans->queue, skb);
782 tasklet_schedule(&trans->tasklet);
785 EXPORT_SYMBOL(xfrm_trans_queue);
787 void __init xfrm_input_init(void)
792 init_dummy_netdev(&xfrm_napi_dev);
793 err = gro_cells_init(&gro_cells, &xfrm_napi_dev);
795 gro_cells.cells = NULL;
797 for_each_possible_cpu(i) {
798 struct xfrm_trans_tasklet *trans;
800 trans = &per_cpu(xfrm_trans_tasklet, i);
801 __skb_queue_head_init(&trans->queue);
802 tasklet_init(&trans->tasklet, xfrm_trans_reinject,
803 (unsigned long)trans);