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);
42 #define XFRM_TRANS_SKB_CB(__skb) ((struct xfrm_trans_cb *)&((__skb)->cb[0]))
44 static DEFINE_SPINLOCK(xfrm_input_afinfo_lock);
45 static struct xfrm_input_afinfo const __rcu *xfrm_input_afinfo[2][AF_INET6 + 1];
47 static struct gro_cells gro_cells;
48 static struct net_device xfrm_napi_dev;
50 static DEFINE_PER_CPU(struct xfrm_trans_tasklet, xfrm_trans_tasklet);
52 int xfrm_input_register_afinfo(const struct xfrm_input_afinfo *afinfo)
56 if (WARN_ON(afinfo->family > AF_INET6))
59 spin_lock_bh(&xfrm_input_afinfo_lock);
60 if (unlikely(xfrm_input_afinfo[afinfo->is_ipip][afinfo->family]))
63 rcu_assign_pointer(xfrm_input_afinfo[afinfo->is_ipip][afinfo->family], afinfo);
64 spin_unlock_bh(&xfrm_input_afinfo_lock);
67 EXPORT_SYMBOL(xfrm_input_register_afinfo);
69 int xfrm_input_unregister_afinfo(const struct xfrm_input_afinfo *afinfo)
73 spin_lock_bh(&xfrm_input_afinfo_lock);
74 if (likely(xfrm_input_afinfo[afinfo->is_ipip][afinfo->family])) {
75 if (unlikely(xfrm_input_afinfo[afinfo->is_ipip][afinfo->family] != afinfo))
78 RCU_INIT_POINTER(xfrm_input_afinfo[afinfo->is_ipip][afinfo->family], NULL);
80 spin_unlock_bh(&xfrm_input_afinfo_lock);
84 EXPORT_SYMBOL(xfrm_input_unregister_afinfo);
86 static const struct xfrm_input_afinfo *xfrm_input_get_afinfo(u8 family, bool is_ipip)
88 const struct xfrm_input_afinfo *afinfo;
90 if (WARN_ON_ONCE(family > AF_INET6))
94 afinfo = rcu_dereference(xfrm_input_afinfo[is_ipip][family]);
95 if (unlikely(!afinfo))
100 static int xfrm_rcv_cb(struct sk_buff *skb, unsigned int family, u8 protocol,
103 bool is_ipip = (protocol == IPPROTO_IPIP || protocol == IPPROTO_IPV6);
104 const struct xfrm_input_afinfo *afinfo;
107 afinfo = xfrm_input_get_afinfo(family, is_ipip);
109 return -EAFNOSUPPORT;
111 ret = afinfo->callback(skb, protocol, err);
117 struct sec_path *secpath_set(struct sk_buff *skb)
119 struct sec_path *sp, *tmp = skb_ext_find(skb, SKB_EXT_SEC_PATH);
121 sp = skb_ext_add(skb, SKB_EXT_SEC_PATH);
125 if (tmp) /* reused existing one (was COW'd if needed) */
128 /* allocated new secpath */
129 memset(sp->ovec, 0, sizeof(sp->ovec));
135 EXPORT_SYMBOL(secpath_set);
137 /* Fetch spi and seq from ipsec header */
139 int xfrm_parse_spi(struct sk_buff *skb, u8 nexthdr, __be32 *spi, __be32 *seq)
141 int offset, offset_seq;
146 hlen = sizeof(struct ip_auth_hdr);
147 offset = offsetof(struct ip_auth_hdr, spi);
148 offset_seq = offsetof(struct ip_auth_hdr, seq_no);
151 hlen = sizeof(struct ip_esp_hdr);
152 offset = offsetof(struct ip_esp_hdr, spi);
153 offset_seq = offsetof(struct ip_esp_hdr, seq_no);
156 if (!pskb_may_pull(skb, sizeof(struct ip_comp_hdr)))
158 *spi = htonl(ntohs(*(__be16 *)(skb_transport_header(skb) + 2)));
165 if (!pskb_may_pull(skb, hlen))
168 *spi = *(__be32 *)(skb_transport_header(skb) + offset);
169 *seq = *(__be32 *)(skb_transport_header(skb) + offset_seq);
172 EXPORT_SYMBOL(xfrm_parse_spi);
174 static int xfrm4_remove_beet_encap(struct xfrm_state *x, struct sk_buff *skb)
180 if (unlikely(XFRM_MODE_SKB_CB(skb)->protocol == IPPROTO_BEETPH)) {
181 struct ip_beet_phdr *ph;
184 if (!pskb_may_pull(skb, sizeof(*ph)))
187 ph = (struct ip_beet_phdr *)skb->data;
189 phlen = sizeof(*ph) + ph->padlen;
190 optlen = ph->hdrlen * 8 + (IPV4_BEET_PHMAXLEN - phlen);
191 if (optlen < 0 || optlen & 3 || optlen > 250)
194 XFRM_MODE_SKB_CB(skb)->protocol = ph->nexthdr;
196 if (!pskb_may_pull(skb, phlen))
198 __skb_pull(skb, phlen);
201 skb_push(skb, sizeof(*iph));
202 skb_reset_network_header(skb);
203 skb_mac_header_rebuild(skb);
205 xfrm4_beet_make_header(skb);
209 iph->ihl += optlen / 4;
210 iph->tot_len = htons(skb->len);
211 iph->daddr = x->sel.daddr.a4;
212 iph->saddr = x->sel.saddr.a4;
214 iph->check = ip_fast_csum(skb_network_header(skb), iph->ihl);
220 static void ipip_ecn_decapsulate(struct sk_buff *skb)
222 struct iphdr *inner_iph = ipip_hdr(skb);
224 if (INET_ECN_is_ce(XFRM_MODE_SKB_CB(skb)->tos))
225 IP_ECN_set_ce(inner_iph);
228 static int xfrm4_remove_tunnel_encap(struct xfrm_state *x, struct sk_buff *skb)
232 if (XFRM_MODE_SKB_CB(skb)->protocol != IPPROTO_IPIP)
235 if (!pskb_may_pull(skb, sizeof(struct iphdr)))
238 err = skb_unclone(skb, GFP_ATOMIC);
242 if (x->props.flags & XFRM_STATE_DECAP_DSCP)
243 ipv4_copy_dscp(XFRM_MODE_SKB_CB(skb)->tos, ipip_hdr(skb));
244 if (!(x->props.flags & XFRM_STATE_NOECN))
245 ipip_ecn_decapsulate(skb);
247 skb_reset_network_header(skb);
248 skb_mac_header_rebuild(skb);
250 eth_hdr(skb)->h_proto = skb->protocol;
258 static void ipip6_ecn_decapsulate(struct sk_buff *skb)
260 struct ipv6hdr *inner_iph = ipipv6_hdr(skb);
262 if (INET_ECN_is_ce(XFRM_MODE_SKB_CB(skb)->tos))
263 IP6_ECN_set_ce(skb, inner_iph);
266 static int xfrm6_remove_tunnel_encap(struct xfrm_state *x, struct sk_buff *skb)
270 if (XFRM_MODE_SKB_CB(skb)->protocol != IPPROTO_IPV6)
272 if (!pskb_may_pull(skb, sizeof(struct ipv6hdr)))
275 err = skb_unclone(skb, GFP_ATOMIC);
279 if (x->props.flags & XFRM_STATE_DECAP_DSCP)
280 ipv6_copy_dscp(ipv6_get_dsfield(ipv6_hdr(skb)),
282 if (!(x->props.flags & XFRM_STATE_NOECN))
283 ipip6_ecn_decapsulate(skb);
285 skb_reset_network_header(skb);
286 skb_mac_header_rebuild(skb);
288 eth_hdr(skb)->h_proto = skb->protocol;
296 static int xfrm6_remove_beet_encap(struct xfrm_state *x, struct sk_buff *skb)
298 struct ipv6hdr *ip6h;
299 int size = sizeof(struct ipv6hdr);
302 err = skb_cow_head(skb, size + skb->mac_len);
306 __skb_push(skb, size);
307 skb_reset_network_header(skb);
308 skb_mac_header_rebuild(skb);
310 xfrm6_beet_make_header(skb);
312 ip6h = ipv6_hdr(skb);
313 ip6h->payload_len = htons(skb->len - size);
314 ip6h->daddr = x->sel.daddr.in6;
315 ip6h->saddr = x->sel.saddr.in6;
321 /* Remove encapsulation header.
323 * The IP header will be moved over the top of the encapsulation
326 * On entry, the transport header shall point to where the IP header
327 * should be and the network header shall be set to where the IP
328 * header currently is. skb->data shall point to the start of the
332 xfrm_inner_mode_encap_remove(struct xfrm_state *x,
333 const struct xfrm_mode *inner_mode,
336 switch (inner_mode->encap) {
338 if (inner_mode->family == AF_INET)
339 return xfrm4_remove_beet_encap(x, skb);
340 if (inner_mode->family == AF_INET6)
341 return xfrm6_remove_beet_encap(x, skb);
343 case XFRM_MODE_TUNNEL:
344 if (inner_mode->family == AF_INET)
345 return xfrm4_remove_tunnel_encap(x, skb);
346 if (inner_mode->family == AF_INET6)
347 return xfrm6_remove_tunnel_encap(x, skb);
355 static int xfrm_prepare_input(struct xfrm_state *x, struct sk_buff *skb)
357 const struct xfrm_mode *inner_mode = &x->inner_mode;
359 switch (x->outer_mode.family) {
361 xfrm4_extract_header(skb);
364 xfrm6_extract_header(skb);
368 return -EAFNOSUPPORT;
371 if (x->sel.family == AF_UNSPEC) {
372 inner_mode = xfrm_ip2inner_mode(x, XFRM_MODE_SKB_CB(skb)->protocol);
374 return -EAFNOSUPPORT;
377 switch (inner_mode->family) {
379 skb->protocol = htons(ETH_P_IP);
382 skb->protocol = htons(ETH_P_IPV6);
389 return xfrm_inner_mode_encap_remove(x, inner_mode, skb);
392 /* Remove encapsulation header.
394 * The IP header will be moved over the top of the encapsulation header.
396 * On entry, skb_transport_header() shall point to where the IP header
397 * should be and skb_network_header() shall be set to where the IP header
398 * currently is. skb->data shall point to the start of the payload.
400 static int xfrm4_transport_input(struct xfrm_state *x, struct sk_buff *skb)
402 int ihl = skb->data - skb_transport_header(skb);
404 if (skb->transport_header != skb->network_header) {
405 memmove(skb_transport_header(skb),
406 skb_network_header(skb), ihl);
407 skb->network_header = skb->transport_header;
409 ip_hdr(skb)->tot_len = htons(skb->len + ihl);
410 skb_reset_transport_header(skb);
414 static int xfrm6_transport_input(struct xfrm_state *x, struct sk_buff *skb)
416 #if IS_ENABLED(CONFIG_IPV6)
417 int ihl = skb->data - skb_transport_header(skb);
419 if (skb->transport_header != skb->network_header) {
420 memmove(skb_transport_header(skb),
421 skb_network_header(skb), ihl);
422 skb->network_header = skb->transport_header;
424 ipv6_hdr(skb)->payload_len = htons(skb->len + ihl -
425 sizeof(struct ipv6hdr));
426 skb_reset_transport_header(skb);
430 return -EAFNOSUPPORT;
434 static int xfrm_inner_mode_input(struct xfrm_state *x,
435 const struct xfrm_mode *inner_mode,
438 switch (inner_mode->encap) {
440 case XFRM_MODE_TUNNEL:
441 return xfrm_prepare_input(x, skb);
442 case XFRM_MODE_TRANSPORT:
443 if (inner_mode->family == AF_INET)
444 return xfrm4_transport_input(x, skb);
445 if (inner_mode->family == AF_INET6)
446 return xfrm6_transport_input(x, skb);
448 case XFRM_MODE_ROUTEOPTIMIZATION:
459 int xfrm_input(struct sk_buff *skb, int nexthdr, __be32 spi, int encap_type)
461 const struct xfrm_state_afinfo *afinfo;
462 struct net *net = dev_net(skb->dev);
463 const struct xfrm_mode *inner_mode;
467 struct xfrm_state *x = NULL;
468 xfrm_address_t *daddr;
469 u32 mark = skb->mark;
470 unsigned int family = AF_UNSPEC;
473 bool xfrm_gro = false;
474 bool crypto_done = false;
475 struct xfrm_offload *xo = xfrm_offload(skb);
478 if (encap_type < 0) {
479 x = xfrm_input_state(skb);
481 if (unlikely(x->km.state != XFRM_STATE_VALID)) {
482 if (x->km.state == XFRM_STATE_ACQ)
483 XFRM_INC_STATS(net, LINUX_MIB_XFRMACQUIREERROR);
486 LINUX_MIB_XFRMINSTATEINVALID);
488 if (encap_type == -1)
493 family = x->outer_mode.family;
495 /* An encap_type of -1 indicates async resumption. */
496 if (encap_type == -1) {
498 seq = XFRM_SKB_CB(skb)->seq.input.low;
502 /* encap_type < -1 indicates a GRO call. */
504 seq = XFRM_SPI_SKB_CB(skb)->seq;
506 if (xo && (xo->flags & CRYPTO_DONE)) {
508 family = XFRM_SPI_SKB_CB(skb)->family;
510 if (!(xo->status & CRYPTO_SUCCESS)) {
512 (CRYPTO_TRANSPORT_AH_AUTH_FAILED |
513 CRYPTO_TRANSPORT_ESP_AUTH_FAILED |
514 CRYPTO_TUNNEL_AH_AUTH_FAILED |
515 CRYPTO_TUNNEL_ESP_AUTH_FAILED)) {
517 xfrm_audit_state_icvfail(x, skb,
519 x->stats.integrity_failed++;
520 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEPROTOERROR);
524 if (xo->status & CRYPTO_INVALID_PROTOCOL) {
525 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEPROTOERROR);
529 XFRM_INC_STATS(net, LINUX_MIB_XFRMINBUFFERERROR);
533 if ((err = xfrm_parse_spi(skb, nexthdr, &spi, &seq)) != 0) {
534 XFRM_INC_STATS(net, LINUX_MIB_XFRMINHDRERROR);
542 family = XFRM_SPI_SKB_CB(skb)->family;
544 /* if tunnel is present override skb->mark value with tunnel i_key */
547 if (XFRM_TUNNEL_SKB_CB(skb)->tunnel.ip4)
548 mark = be32_to_cpu(XFRM_TUNNEL_SKB_CB(skb)->tunnel.ip4->parms.i_key);
551 if (XFRM_TUNNEL_SKB_CB(skb)->tunnel.ip6)
552 mark = be32_to_cpu(XFRM_TUNNEL_SKB_CB(skb)->tunnel.ip6->parms.i_key);
556 sp = secpath_set(skb);
558 XFRM_INC_STATS(net, LINUX_MIB_XFRMINERROR);
563 if (!spi && (err = xfrm_parse_spi(skb, nexthdr, &spi, &seq)) != 0) {
565 XFRM_INC_STATS(net, LINUX_MIB_XFRMINHDRERROR);
569 daddr = (xfrm_address_t *)(skb_network_header(skb) +
570 XFRM_SPI_SKB_CB(skb)->daddroff);
572 sp = skb_sec_path(skb);
574 if (sp->len == XFRM_MAX_DEPTH) {
576 XFRM_INC_STATS(net, LINUX_MIB_XFRMINBUFFERERROR);
580 x = xfrm_state_lookup(net, mark, daddr, spi, nexthdr, family);
583 XFRM_INC_STATS(net, LINUX_MIB_XFRMINNOSTATES);
584 xfrm_audit_state_notfound(skb, family, spi, seq);
588 skb->mark = xfrm_smark_get(skb->mark, x);
590 sp->xvec[sp->len++] = x;
594 XFRM_INC_STATS(net, LINUX_MIB_XFRMINERROR);
601 if (unlikely(x->km.state != XFRM_STATE_VALID)) {
602 if (x->km.state == XFRM_STATE_ACQ)
603 XFRM_INC_STATS(net, LINUX_MIB_XFRMACQUIREERROR);
606 LINUX_MIB_XFRMINSTATEINVALID);
610 if ((x->encap ? x->encap->encap_type : 0) != encap_type) {
611 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMISMATCH);
615 if (x->repl->check(x, skb, seq)) {
616 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATESEQERROR);
620 if (xfrm_state_check_expire(x)) {
621 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEEXPIRED);
625 spin_unlock(&x->lock);
627 if (xfrm_tunnel_check(skb, x, family)) {
628 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMODEERROR);
632 seq_hi = htonl(xfrm_replay_seqhi(x, seq));
634 XFRM_SKB_CB(skb)->seq.input.low = seq;
635 XFRM_SKB_CB(skb)->seq.input.hi = seq_hi;
640 nexthdr = x->type_offload->input_tail(x, skb);
642 nexthdr = x->type->input(x, skb);
644 if (nexthdr == -EINPROGRESS)
651 if (nexthdr == -EBADMSG) {
652 xfrm_audit_state_icvfail(x, skb,
654 x->stats.integrity_failed++;
656 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEPROTOERROR);
660 /* only the first xfrm gets the encap type */
663 if (x->repl->recheck(x, skb, seq)) {
664 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATESEQERROR);
668 x->repl->advance(x, seq);
670 x->curlft.bytes += skb->len;
673 spin_unlock(&x->lock);
675 XFRM_MODE_SKB_CB(skb)->protocol = nexthdr;
677 inner_mode = &x->inner_mode;
679 if (x->sel.family == AF_UNSPEC) {
680 inner_mode = xfrm_ip2inner_mode(x, XFRM_MODE_SKB_CB(skb)->protocol);
681 if (inner_mode == NULL) {
682 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMODEERROR);
687 if (xfrm_inner_mode_input(x, inner_mode, skb)) {
688 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMODEERROR);
692 if (x->outer_mode.flags & XFRM_MODE_FLAG_TUNNEL) {
698 * We need the inner address. However, we only get here for
699 * transport mode so the outer address is identical.
701 daddr = &x->id.daddr;
702 family = x->outer_mode.family;
704 err = xfrm_parse_spi(skb, nexthdr, &spi, &seq);
706 XFRM_INC_STATS(net, LINUX_MIB_XFRMINHDRERROR);
712 err = xfrm_rcv_cb(skb, family, x->type->proto, 0);
719 sp = skb_sec_path(skb);
723 gro_cells_receive(&gro_cells, skb);
726 xo = xfrm_offload(skb);
728 xfrm_gro = xo->flags & XFRM_GRO;
732 afinfo = xfrm_state_afinfo_get_rcu(x->inner_mode.family);
734 err = afinfo->transport_finish(skb, xfrm_gro || async);
737 sp = skb_sec_path(skb);
741 gro_cells_receive(&gro_cells, skb);
749 spin_unlock(&x->lock);
751 xfrm_rcv_cb(skb, family, x && x->type ? x->type->proto : nexthdr, -1);
755 EXPORT_SYMBOL(xfrm_input);
757 int xfrm_input_resume(struct sk_buff *skb, int nexthdr)
759 return xfrm_input(skb, nexthdr, 0, -1);
761 EXPORT_SYMBOL(xfrm_input_resume);
763 static void xfrm_trans_reinject(struct tasklet_struct *t)
765 struct xfrm_trans_tasklet *trans = from_tasklet(trans, t, tasklet);
766 struct sk_buff_head queue;
769 __skb_queue_head_init(&queue);
770 skb_queue_splice_init(&trans->queue, &queue);
772 while ((skb = __skb_dequeue(&queue)))
773 XFRM_TRANS_SKB_CB(skb)->finish(XFRM_TRANS_SKB_CB(skb)->net,
777 int xfrm_trans_queue_net(struct net *net, struct sk_buff *skb,
778 int (*finish)(struct net *, struct sock *,
781 struct xfrm_trans_tasklet *trans;
783 trans = this_cpu_ptr(&xfrm_trans_tasklet);
785 if (skb_queue_len(&trans->queue) >= netdev_max_backlog)
788 BUILD_BUG_ON(sizeof(struct xfrm_trans_cb) > sizeof(skb->cb));
790 XFRM_TRANS_SKB_CB(skb)->finish = finish;
791 XFRM_TRANS_SKB_CB(skb)->net = net;
792 __skb_queue_tail(&trans->queue, skb);
793 tasklet_schedule(&trans->tasklet);
796 EXPORT_SYMBOL(xfrm_trans_queue_net);
798 int xfrm_trans_queue(struct sk_buff *skb,
799 int (*finish)(struct net *, struct sock *,
802 return xfrm_trans_queue_net(dev_net(skb->dev), skb, finish);
804 EXPORT_SYMBOL(xfrm_trans_queue);
806 void __init xfrm_input_init(void)
811 init_dummy_netdev(&xfrm_napi_dev);
812 err = gro_cells_init(&gro_cells, &xfrm_napi_dev);
814 gro_cells.cells = NULL;
816 for_each_possible_cpu(i) {
817 struct xfrm_trans_tasklet *trans;
819 trans = &per_cpu(xfrm_trans_tasklet, i);
820 __skb_queue_head_init(&trans->queue);
821 tasklet_setup(&trans->tasklet, xfrm_trans_reinject);