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);
368 if (x->sel.family == AF_UNSPEC) {
369 inner_mode = xfrm_ip2inner_mode(x, XFRM_MODE_SKB_CB(skb)->protocol);
372 return -EAFNOSUPPORT;
376 afinfo = xfrm_state_afinfo_get_rcu(inner_mode->family);
377 if (unlikely(!afinfo)) {
379 return -EAFNOSUPPORT;
382 skb->protocol = afinfo->eth_proto;
384 return xfrm_inner_mode_encap_remove(x, inner_mode, skb);
387 /* Remove encapsulation header.
389 * The IP header will be moved over the top of the encapsulation header.
391 * On entry, skb_transport_header() shall point to where the IP header
392 * should be and skb_network_header() shall be set to where the IP header
393 * currently is. skb->data shall point to the start of the payload.
395 static int xfrm4_transport_input(struct xfrm_state *x, struct sk_buff *skb)
397 int ihl = skb->data - skb_transport_header(skb);
399 if (skb->transport_header != skb->network_header) {
400 memmove(skb_transport_header(skb),
401 skb_network_header(skb), ihl);
402 skb->network_header = skb->transport_header;
404 ip_hdr(skb)->tot_len = htons(skb->len + ihl);
405 skb_reset_transport_header(skb);
409 static int xfrm6_transport_input(struct xfrm_state *x, struct sk_buff *skb)
411 #if IS_ENABLED(CONFIG_IPV6)
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);
417 skb->network_header = skb->transport_header;
419 ipv6_hdr(skb)->payload_len = htons(skb->len + ihl -
420 sizeof(struct ipv6hdr));
421 skb_reset_transport_header(skb);
425 return -EAFNOSUPPORT;
429 static int xfrm_inner_mode_input(struct xfrm_state *x,
430 const struct xfrm_mode *inner_mode,
433 switch (inner_mode->encap) {
435 case XFRM_MODE_TUNNEL:
436 return xfrm_prepare_input(x, skb);
437 case XFRM_MODE_TRANSPORT:
438 if (inner_mode->family == AF_INET)
439 return xfrm4_transport_input(x, skb);
440 if (inner_mode->family == AF_INET6)
441 return xfrm6_transport_input(x, skb);
443 case XFRM_MODE_ROUTEOPTIMIZATION:
454 int xfrm_input(struct sk_buff *skb, int nexthdr, __be32 spi, int encap_type)
456 const struct xfrm_state_afinfo *afinfo;
457 struct net *net = dev_net(skb->dev);
458 const struct xfrm_mode *inner_mode;
462 struct xfrm_state *x = NULL;
463 xfrm_address_t *daddr;
464 u32 mark = skb->mark;
465 unsigned int family = AF_UNSPEC;
468 bool xfrm_gro = false;
469 bool crypto_done = false;
470 struct xfrm_offload *xo = xfrm_offload(skb);
473 if (encap_type < 0) {
474 x = xfrm_input_state(skb);
476 if (unlikely(x->km.state != XFRM_STATE_VALID)) {
477 if (x->km.state == XFRM_STATE_ACQ)
478 XFRM_INC_STATS(net, LINUX_MIB_XFRMACQUIREERROR);
481 LINUX_MIB_XFRMINSTATEINVALID);
485 family = x->outer_mode.family;
487 /* An encap_type of -1 indicates async resumption. */
488 if (encap_type == -1) {
490 seq = XFRM_SKB_CB(skb)->seq.input.low;
494 /* encap_type < -1 indicates a GRO call. */
496 seq = XFRM_SPI_SKB_CB(skb)->seq;
498 if (xo && (xo->flags & CRYPTO_DONE)) {
500 family = XFRM_SPI_SKB_CB(skb)->family;
502 if (!(xo->status & CRYPTO_SUCCESS)) {
504 (CRYPTO_TRANSPORT_AH_AUTH_FAILED |
505 CRYPTO_TRANSPORT_ESP_AUTH_FAILED |
506 CRYPTO_TUNNEL_AH_AUTH_FAILED |
507 CRYPTO_TUNNEL_ESP_AUTH_FAILED)) {
509 xfrm_audit_state_icvfail(x, skb,
511 x->stats.integrity_failed++;
512 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEPROTOERROR);
516 if (xo->status & CRYPTO_INVALID_PROTOCOL) {
517 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEPROTOERROR);
521 XFRM_INC_STATS(net, LINUX_MIB_XFRMINBUFFERERROR);
525 if ((err = xfrm_parse_spi(skb, nexthdr, &spi, &seq)) != 0) {
526 XFRM_INC_STATS(net, LINUX_MIB_XFRMINHDRERROR);
534 family = XFRM_SPI_SKB_CB(skb)->family;
536 /* if tunnel is present override skb->mark value with tunnel i_key */
539 if (XFRM_TUNNEL_SKB_CB(skb)->tunnel.ip4)
540 mark = be32_to_cpu(XFRM_TUNNEL_SKB_CB(skb)->tunnel.ip4->parms.i_key);
543 if (XFRM_TUNNEL_SKB_CB(skb)->tunnel.ip6)
544 mark = be32_to_cpu(XFRM_TUNNEL_SKB_CB(skb)->tunnel.ip6->parms.i_key);
548 sp = secpath_set(skb);
550 XFRM_INC_STATS(net, LINUX_MIB_XFRMINERROR);
555 if (!spi && (err = xfrm_parse_spi(skb, nexthdr, &spi, &seq)) != 0) {
557 XFRM_INC_STATS(net, LINUX_MIB_XFRMINHDRERROR);
561 daddr = (xfrm_address_t *)(skb_network_header(skb) +
562 XFRM_SPI_SKB_CB(skb)->daddroff);
564 sp = skb_sec_path(skb);
566 if (sp->len == XFRM_MAX_DEPTH) {
568 XFRM_INC_STATS(net, LINUX_MIB_XFRMINBUFFERERROR);
572 x = xfrm_state_lookup(net, mark, daddr, spi, nexthdr, family);
575 XFRM_INC_STATS(net, LINUX_MIB_XFRMINNOSTATES);
576 xfrm_audit_state_notfound(skb, family, spi, seq);
580 skb->mark = xfrm_smark_get(skb->mark, x);
582 sp->xvec[sp->len++] = x;
586 XFRM_INC_STATS(net, LINUX_MIB_XFRMINERROR);
593 if (unlikely(x->km.state != XFRM_STATE_VALID)) {
594 if (x->km.state == XFRM_STATE_ACQ)
595 XFRM_INC_STATS(net, LINUX_MIB_XFRMACQUIREERROR);
598 LINUX_MIB_XFRMINSTATEINVALID);
602 if ((x->encap ? x->encap->encap_type : 0) != encap_type) {
603 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMISMATCH);
607 if (x->repl->check(x, skb, seq)) {
608 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATESEQERROR);
612 if (xfrm_state_check_expire(x)) {
613 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEEXPIRED);
617 spin_unlock(&x->lock);
619 if (xfrm_tunnel_check(skb, x, family)) {
620 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMODEERROR);
624 seq_hi = htonl(xfrm_replay_seqhi(x, seq));
626 XFRM_SKB_CB(skb)->seq.input.low = seq;
627 XFRM_SKB_CB(skb)->seq.input.hi = seq_hi;
632 nexthdr = x->type_offload->input_tail(x, skb);
634 nexthdr = x->type->input(x, skb);
636 if (nexthdr == -EINPROGRESS)
643 if (nexthdr == -EBADMSG) {
644 xfrm_audit_state_icvfail(x, skb,
646 x->stats.integrity_failed++;
648 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEPROTOERROR);
652 /* only the first xfrm gets the encap type */
655 if (async && x->repl->recheck(x, skb, seq)) {
656 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATESEQERROR);
660 x->repl->advance(x, seq);
662 x->curlft.bytes += skb->len;
665 spin_unlock(&x->lock);
667 XFRM_MODE_SKB_CB(skb)->protocol = nexthdr;
669 inner_mode = &x->inner_mode;
671 if (x->sel.family == AF_UNSPEC) {
672 inner_mode = xfrm_ip2inner_mode(x, XFRM_MODE_SKB_CB(skb)->protocol);
673 if (inner_mode == NULL) {
674 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMODEERROR);
679 if (xfrm_inner_mode_input(x, inner_mode, skb)) {
680 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMODEERROR);
684 if (x->outer_mode.flags & XFRM_MODE_FLAG_TUNNEL) {
690 * We need the inner address. However, we only get here for
691 * transport mode so the outer address is identical.
693 daddr = &x->id.daddr;
694 family = x->outer_mode.family;
696 err = xfrm_parse_spi(skb, nexthdr, &spi, &seq);
698 XFRM_INC_STATS(net, LINUX_MIB_XFRMINHDRERROR);
704 err = xfrm_rcv_cb(skb, family, x->type->proto, 0);
711 sp = skb_sec_path(skb);
715 gro_cells_receive(&gro_cells, skb);
718 xo = xfrm_offload(skb);
720 xfrm_gro = xo->flags & XFRM_GRO;
724 afinfo = xfrm_state_afinfo_get_rcu(x->inner_mode.family);
726 err = afinfo->transport_finish(skb, xfrm_gro || async);
729 sp = skb_sec_path(skb);
733 gro_cells_receive(&gro_cells, skb);
741 spin_unlock(&x->lock);
743 xfrm_rcv_cb(skb, family, x && x->type ? x->type->proto : nexthdr, -1);
747 EXPORT_SYMBOL(xfrm_input);
749 int xfrm_input_resume(struct sk_buff *skb, int nexthdr)
751 return xfrm_input(skb, nexthdr, 0, -1);
753 EXPORT_SYMBOL(xfrm_input_resume);
755 static void xfrm_trans_reinject(unsigned long data)
757 struct xfrm_trans_tasklet *trans = (void *)data;
758 struct sk_buff_head queue;
761 __skb_queue_head_init(&queue);
762 skb_queue_splice_init(&trans->queue, &queue);
764 while ((skb = __skb_dequeue(&queue)))
765 XFRM_TRANS_SKB_CB(skb)->finish(dev_net(skb->dev), NULL, skb);
768 int xfrm_trans_queue(struct sk_buff *skb,
769 int (*finish)(struct net *, struct sock *,
772 struct xfrm_trans_tasklet *trans;
774 trans = this_cpu_ptr(&xfrm_trans_tasklet);
776 if (skb_queue_len(&trans->queue) >= netdev_max_backlog)
779 XFRM_TRANS_SKB_CB(skb)->finish = finish;
780 __skb_queue_tail(&trans->queue, skb);
781 tasklet_schedule(&trans->tasklet);
784 EXPORT_SYMBOL(xfrm_trans_queue);
786 void __init xfrm_input_init(void)
791 init_dummy_netdev(&xfrm_napi_dev);
792 err = gro_cells_init(&gro_cells, &xfrm_napi_dev);
794 gro_cells.cells = NULL;
796 for_each_possible_cpu(i) {
797 struct xfrm_trans_tasklet *trans;
799 trans = &per_cpu(xfrm_trans_tasklet, i);
800 __skb_queue_head_init(&trans->queue);
801 tasklet_init(&trans->tasklet, xfrm_trans_reinject,
802 (unsigned long)trans);