From 63b21caba17ef2df5982c7b75eb989100212b27b Mon Sep 17 00:00:00 2001 From: Antony Antony Date: Fri, 19 Jan 2024 12:27:42 +0100 Subject: [PATCH 1/4] xfrm: introduce forwarding of ICMP Error messages 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 v5: - fix return values bool->int, feedback from Steffen Changes since v4: - split the series to only ICMP erorr forwarding 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 Signed-off-by: Steffen Klassert --- net/xfrm/xfrm_policy.c | 142 ++++++++++++++++++++++++++++++++++++++++- 1 file changed, 140 insertions(+), 2 deletions(-) diff --git a/net/xfrm/xfrm_policy.c b/net/xfrm/xfrm_policy.c index 1b7e75159727..b4850a8f14ad 100644 --- a/net/xfrm/xfrm_policy.c +++ b/net/xfrm/xfrm_policy.c @@ -29,6 +29,7 @@ #include #include #include +#include #include #include #include @@ -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 = 1; + if (x->props.flags & XFRM_STATE_ICMP && + xfrm_selector_inner_icmp_match(skb, family, &x->sel, &fl)) + ret = 0; + 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; } From ab1e1a38de240057ed108075abd1309c02e2a2a4 Mon Sep 17 00:00:00 2001 From: Kunwu Chan Date: Wed, 24 Jan 2024 14:31:50 +0800 Subject: [PATCH 2/4] xfrm6_tunnel: Use KMEM_CACHE instead of kmem_cache_create Use the new KMEM_CACHE() macro instead of direct kmem_cache_create to simplify the creation of SLAB caches. Signed-off-by: Kunwu Chan Signed-off-by: Steffen Klassert --- net/ipv6/xfrm6_tunnel.c | 5 +---- 1 file changed, 1 insertion(+), 4 deletions(-) diff --git a/net/ipv6/xfrm6_tunnel.c b/net/ipv6/xfrm6_tunnel.c index 1323f2f6928e..0f3df26878a3 100644 --- a/net/ipv6/xfrm6_tunnel.c +++ b/net/ipv6/xfrm6_tunnel.c @@ -355,10 +355,7 @@ static int __init xfrm6_tunnel_init(void) { int rv; - xfrm6_tunnel_spi_kmem = kmem_cache_create("xfrm6_tunnel_spi", - sizeof(struct xfrm6_tunnel_spi), - 0, SLAB_HWCACHE_ALIGN, - NULL); + xfrm6_tunnel_spi_kmem = KMEM_CACHE(xfrm6_tunnel_spi, SLAB_HWCACHE_ALIGN); if (!xfrm6_tunnel_spi_kmem) return -ENOMEM; rv = register_pernet_subsys(&xfrm6_tunnel_net_ops); From 1476de6d2b578673e20fb4cf654ff61cf2782873 Mon Sep 17 00:00:00 2001 From: Kunwu Chan Date: Tue, 30 Jan 2024 16:14:11 +0800 Subject: [PATCH 3/4] xfrm: Simplify the allocation of slab caches in xfrm_policy_init commit 0a31bd5f2bbb ("KMEM_CACHE(): simplify slab cache creation") introduces a new macro. Use the new KMEM_CACHE() macro instead of direct kmem_cache_create to simplify the creation of SLAB caches. Signed-off-by: Kunwu Chan Signed-off-by: Steffen Klassert --- net/xfrm/xfrm_policy.c | 5 +---- 1 file changed, 1 insertion(+), 4 deletions(-) diff --git a/net/xfrm/xfrm_policy.c b/net/xfrm/xfrm_policy.c index b4850a8f14ad..53b7ce4a4db0 100644 --- a/net/xfrm/xfrm_policy.c +++ b/net/xfrm/xfrm_policy.c @@ -4163,10 +4163,7 @@ static int __net_init xfrm_policy_init(struct net *net) int dir, err; if (net_eq(net, &init_net)) { - xfrm_dst_cache = kmem_cache_create("xfrm_dst_cache", - sizeof(struct xfrm_dst), - 0, SLAB_HWCACHE_ALIGN|SLAB_PANIC, - NULL); + xfrm_dst_cache = KMEM_CACHE(xfrm_dst, SLAB_HWCACHE_ALIGN | SLAB_PANIC); err = rhashtable_init(&xfrm_policy_inexact_table, &xfrm_pol_inexact_params); BUG_ON(err); From aceb147b20a2eda9a10bdd4b5650718f731ef3e1 Mon Sep 17 00:00:00 2001 From: Breno Leitao Date: Thu, 22 Feb 2024 06:41:17 -0800 Subject: [PATCH 4/4] xfrm: Do not allocate stats in the driver With commit 34d21de99cea9 ("net: Move {l,t,d}stats allocation to core and convert veth & vrf"), stats allocation could be done on net core instead of this driver. With this new approach, the driver doesn't have to bother with error handling (allocation failure checking, making sure free happens in the right spot, etc). This is core responsibility now. Remove the allocation in the xfrm driver and leverage the network core allocation. Signed-off-by: Breno Leitao Reviewed-by: Simon Horman Signed-off-by: Steffen Klassert --- net/xfrm/xfrm_interface_core.c | 10 ++-------- 1 file changed, 2 insertions(+), 8 deletions(-) diff --git a/net/xfrm/xfrm_interface_core.c b/net/xfrm/xfrm_interface_core.c index 21d50d75c260..1a1f6613fda2 100644 --- a/net/xfrm/xfrm_interface_core.c +++ b/net/xfrm/xfrm_interface_core.c @@ -240,7 +240,6 @@ static void xfrmi_dev_free(struct net_device *dev) struct xfrm_if *xi = netdev_priv(dev); gro_cells_destroy(&xi->gro_cells); - free_percpu(dev->tstats); } static int xfrmi_create(struct net_device *dev) @@ -749,6 +748,7 @@ static void xfrmi_dev_setup(struct net_device *dev) dev->flags = IFF_NOARP; dev->needs_free_netdev = true; dev->priv_destructor = xfrmi_dev_free; + dev->pcpu_stat_type = NETDEV_PCPU_STAT_TSTATS; netif_keep_dst(dev); eth_broadcast_addr(dev->broadcast); @@ -765,15 +765,9 @@ static int xfrmi_dev_init(struct net_device *dev) struct net_device *phydev = __dev_get_by_index(xi->net, xi->p.link); int err; - dev->tstats = netdev_alloc_pcpu_stats(struct pcpu_sw_netstats); - if (!dev->tstats) - return -ENOMEM; - err = gro_cells_init(&xi->gro_cells, dev); - if (err) { - free_percpu(dev->tstats); + if (err) return err; - } dev->features |= NETIF_F_LLTX; dev->features |= XFRMI_FEATURES;