@@ -29,6 +29,7 @@
#include <linux/audit.h>
#include <linux/rhashtable.h>
#include <linux/if_tunnel.h>
+#include <linux/icmp.h>
#include <net/dst.h>
#include <net/flow.h>
#include <net/inet_ecn.h>
@@ -3503,6 +3504,128 @@ static inline int secpath_has_nontransport(const struct sec_path *sp, int k, int
return 0;
}
+static bool icmp_err_packet(const struct flowi *fl, unsigned short family)
+{
+ const struct flowi4 *fl4 = &fl->u.ip4;
+
+ if (family == AF_INET &&
+ fl4->flowi4_proto == IPPROTO_ICMP &&
+ (fl4->fl4_icmp_type == ICMP_DEST_UNREACH ||
+ fl4->fl4_icmp_type == ICMP_TIME_EXCEEDED))
+ return true;
+
+#if IS_ENABLED(CONFIG_IPV6)
+ if (family == AF_INET6) {
+ const struct flowi6 *fl6 = &fl->u.ip6;
+
+ if (fl6->flowi6_proto == IPPROTO_ICMPV6 &&
+ (fl6->fl6_icmp_type == ICMPV6_DEST_UNREACH ||
+ fl6->fl6_icmp_type == ICMPV6_PKT_TOOBIG ||
+ fl6->fl6_icmp_type == ICMPV6_TIME_EXCEED))
+ return true;
+ }
+#endif
+ return false;
+}
+
+static bool xfrm_icmp_flow_decode(struct sk_buff *skb, unsigned short family,
+ const struct flowi *fl, struct flowi *fl1)
+{
+ bool ret = true;
+ struct sk_buff *newskb = skb_clone(skb, GFP_ATOMIC);
+ int hl = family == AF_INET ? (sizeof(struct iphdr) + sizeof(struct icmphdr)) :
+ (sizeof(struct ipv6hdr) + sizeof(struct icmp6hdr));
+
+ if (!newskb)
+ return true;
+
+ if (!pskb_pull(newskb, hl))
+ goto out;
+
+ skb_reset_network_header(newskb);
+
+ if (xfrm_decode_session_reverse(dev_net(skb->dev), newskb, fl1, family) < 0)
+ goto out;
+
+ fl1->flowi_oif = fl->flowi_oif;
+ fl1->flowi_mark = fl->flowi_mark;
+ fl1->flowi_tos = fl->flowi_tos;
+ nf_nat_decode_session(newskb, fl1, family);
+ ret = false;
+
+out:
+ consume_skb(newskb);
+ return ret;
+}
+
+static bool xfrm_selector_inner_icmp_match(struct sk_buff *skb, unsigned short family,
+ const struct xfrm_selector *sel,
+ const struct flowi *fl)
+{
+ bool ret = false;
+
+ if (icmp_err_packet(fl, family)) {
+ struct flowi fl1;
+
+ if (xfrm_icmp_flow_decode(skb, family, fl, &fl1))
+ return ret;
+
+ ret = xfrm_selector_match(sel, &fl1, family);
+ }
+
+ return ret;
+}
+
+static inline struct
+xfrm_policy *xfrm_in_fwd_icmp(struct sk_buff *skb,
+ const struct flowi *fl, unsigned short family,
+ u32 if_id)
+{
+ struct xfrm_policy *pol = NULL;
+
+ if (icmp_err_packet(fl, family)) {
+ struct flowi fl1;
+ struct net *net = dev_net(skb->dev);
+
+ if (xfrm_icmp_flow_decode(skb, family, fl, &fl1))
+ return pol;
+
+ pol = xfrm_policy_lookup(net, &fl1, family, XFRM_POLICY_FWD, if_id);
+ }
+
+ return pol;
+}
+
+static inline struct
+dst_entry *xfrm_out_fwd_icmp(struct sk_buff *skb, struct flowi *fl,
+ unsigned short family, struct dst_entry *dst)
+{
+ if (icmp_err_packet(fl, family)) {
+ struct net *net = dev_net(skb->dev);
+ struct dst_entry *dst2;
+ struct flowi fl1;
+
+ if (xfrm_icmp_flow_decode(skb, family, fl, &fl1))
+ return dst;
+
+ dst_hold(dst);
+
+ dst2 = xfrm_lookup(net, dst, &fl1, NULL, (XFRM_LOOKUP_QUEUE | XFRM_LOOKUP_ICMP));
+
+ if (IS_ERR(dst2))
+ return dst;
+
+ if (dst2->xfrm) {
+ dst_release(dst);
+ dst = dst2;
+ } else {
+ dst_release(dst2);
+ }
+ }
+
+ return dst;
+}
+
int __xfrm_policy_check(struct sock *sk, int dir, struct sk_buff *skb,
unsigned short family)
{
@@ -3549,9 +3672,17 @@ int __xfrm_policy_check(struct sock *sk, int dir, struct sk_buff *skb,
for (i = sp->len - 1; i >= 0; i--) {
struct xfrm_state *x = sp->xvec[i];
+ int ret = 0;
+
if (!xfrm_selector_match(&x->sel, &fl, family)) {
- XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMISMATCH);
- return 0;
+ ret = true;
+ if (x->props.flags & XFRM_STATE_ICMP &&
+ xfrm_selector_inner_icmp_match(skb, family, &x->sel, &fl))
+ ret = false;
+ if (ret) {
+ XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMISMATCH);
+ return 0;
+ }
}
}
}
@@ -3574,6 +3705,9 @@ int __xfrm_policy_check(struct sock *sk, int dir, struct sk_buff *skb,
return 0;
}
+ if (!pol && dir == XFRM_POLICY_FWD)
+ pol = xfrm_in_fwd_icmp(skb, &fl, family, if_id);
+
if (!pol) {
if (net->xfrm.policy_default[dir] == XFRM_USERPOLICY_BLOCK) {
XFRM_INC_STATS(net, LINUX_MIB_XFRMINNOPOLS);
@@ -3707,6 +3841,10 @@ int __xfrm_route_forward(struct sk_buff *skb, unsigned short family)
res = 0;
dst = NULL;
}
+
+ if (dst && !dst->xfrm)
+ dst = xfrm_out_fwd_icmp(skb, &fl, family, dst);
+
skb_dst_set(skb, dst);
return res;
}
This commit aligns with RFC 4301, Section 6, and addresses the requirement to forward unauthenticated ICMP error messages that do not match any xfrm policies. It utilizes the ICMP payload as an skb and performs a reverse lookup. If a policy match is found, forward the packet. The ICMP payload typically contains a partial IP packet that is likely responsible for the error message. The following error types will be forwarded: - IPv4 ICMP error types: ICMP_DEST_UNREACH & ICMP_TIME_EXCEEDED - IPv6 ICMPv6 error types: ICMPV6_DEST_UNREACH, ICMPV6_PKT_TOOBIG, ICMPV6_TIME_EXCEED To implement this feature, a reverse lookup has been added to the xfrm forward path, making use of the ICMP payload as the skb. To enable this functionality from user space, the XFRM_POLICY_ICMP flag should be added to the outgoing and forward policies, and the XFRM_STATE_ICMP flag should be set on incoming states. e.g. ip xfrm policy add flag icmp tmpl ip xfrm policy src 192.0.2.0/24 dst 192.0.1.0/25 dir out priority 2084302 ptype main flag icmp ip xfrm state add ...flag icmp ip xfrm state root@west:~#ip x s src 192.1.2.23 dst 192.1.2.45 proto esp spi 0xa7b76872 reqid 16389 mode tunnel replay-window 32 flag icmp af-unspec Changes since v3: no code chage - add missing white spaces detected by checkpatch.pl Changes since v2: reviewed by Steffen Klassert - user consume_skb instead of kfree_skb for the inner skb - fixed newskb leaks in error paths - free the newskb once inner flow is decoded with change due to commit 7a0207094f1b ("xfrm: policy: replace session decode with flow dissector") - if xfrm_decode_session_reverse() on inner payload fails ignore. do not increment error counter Changes since v1: - Move IPv6 variable declaration inside IS_ENABLED(CONFIG_IPV6) Changes since RFC: - Fix calculation of ICMPv6 header length Signed-off-by: Antony Antony <antony.antony@secunet.com> --- net/xfrm/xfrm_policy.c | 142 ++++++++++++++++++++++++++++++++++++++++- 1 file changed, 140 insertions(+), 2 deletions(-) -- 2.30.2