netfilter: add IPv6 SYNPROXY target
authorPatrick McHardy <kaber@trash.net>
Tue, 27 Aug 2013 06:50:16 +0000 (08:50 +0200)
committerPablo Neira Ayuso <pablo@netfilter.org>
Tue, 27 Aug 2013 22:28:13 +0000 (00:28 +0200)
Add an IPv6 version of the SYNPROXY target. The main differences to the
IPv4 version is routing and IP header construction.

Signed-off-by: Patrick McHardy <kaber@trash.net>
Tested-by: Martin Topholm <mph@one.com>
Signed-off-by: Jesper Dangaard Brouer <brouer@redhat.com>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
net/ipv6/netfilter/Kconfig
net/ipv6/netfilter/Makefile
net/ipv6/netfilter/ip6t_SYNPROXY.c [new file with mode: 0644]

index 4433ab4..a7f842b 100644 (file)
@@ -153,6 +153,19 @@ config IP6_NF_TARGET_REJECT
 
          To compile it as a module, choose M here.  If unsure, say N.
 
+config IP6_NF_TARGET_SYNPROXY
+       tristate "SYNPROXY target support"
+       depends on NF_CONNTRACK && NETFILTER_ADVANCED
+       select NETFILTER_SYNPROXY
+       select SYN_COOKIES
+       help
+         The SYNPROXY target allows you to intercept TCP connections and
+         establish them using syncookies before they are passed on to the
+         server. This allows to avoid conntrack and server resource usage
+         during SYN-flood attacks.
+
+         To compile it as a module, choose M here. If unsure, say N.
+
 config IP6_NF_MANGLE
        tristate "Packet mangling"
        default m if NETFILTER_ADVANCED=n
index ad13bf0..2b53738 100644 (file)
@@ -37,3 +37,4 @@ obj-$(CONFIG_IP6_NF_MATCH_RT) += ip6t_rt.o
 obj-$(CONFIG_IP6_NF_TARGET_MASQUERADE) += ip6t_MASQUERADE.o
 obj-$(CONFIG_IP6_NF_TARGET_NPT) += ip6t_NPT.o
 obj-$(CONFIG_IP6_NF_TARGET_REJECT) += ip6t_REJECT.o
+obj-$(CONFIG_IP6_NF_TARGET_SYNPROXY) += ip6t_SYNPROXY.o
diff --git a/net/ipv6/netfilter/ip6t_SYNPROXY.c b/net/ipv6/netfilter/ip6t_SYNPROXY.c
new file mode 100644 (file)
index 0000000..4270a9b
--- /dev/null
@@ -0,0 +1,495 @@
+/*
+ * Copyright (c) 2013 Patrick McHardy <kaber@trash.net>
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ */
+
+#include <linux/module.h>
+#include <linux/skbuff.h>
+#include <net/ip6_checksum.h>
+#include <net/ip6_route.h>
+#include <net/tcp.h>
+
+#include <linux/netfilter_ipv6/ip6_tables.h>
+#include <linux/netfilter/x_tables.h>
+#include <linux/netfilter/xt_SYNPROXY.h>
+#include <net/netfilter/nf_conntrack.h>
+#include <net/netfilter/nf_conntrack_seqadj.h>
+#include <net/netfilter/nf_conntrack_synproxy.h>
+
+static struct ipv6hdr *
+synproxy_build_ip(struct sk_buff *skb, const struct in6_addr *saddr,
+                                      const struct in6_addr *daddr)
+{
+       struct ipv6hdr *iph;
+
+       skb_reset_network_header(skb);
+       iph = (struct ipv6hdr *)skb_put(skb, sizeof(*iph));
+       ip6_flow_hdr(iph, 0, 0);
+       iph->hop_limit  = 64;   //XXX
+       iph->nexthdr    = IPPROTO_TCP;
+       iph->saddr      = *saddr;
+       iph->daddr      = *daddr;
+
+       return iph;
+}
+
+static void
+synproxy_send_tcp(const struct sk_buff *skb, struct sk_buff *nskb,
+                 struct nf_conntrack *nfct, enum ip_conntrack_info ctinfo,
+                 struct ipv6hdr *niph, struct tcphdr *nth,
+                 unsigned int tcp_hdr_size)
+{
+       struct net *net = nf_ct_net((struct nf_conn *)nfct);
+       struct dst_entry *dst;
+       struct flowi6 fl6;
+
+       nth->check = ~tcp_v6_check(tcp_hdr_size, &niph->saddr, &niph->daddr, 0);
+       nskb->ip_summed   = CHECKSUM_PARTIAL;
+       nskb->csum_start  = (unsigned char *)nth - nskb->head;
+       nskb->csum_offset = offsetof(struct tcphdr, check);
+
+       memset(&fl6, 0, sizeof(fl6));
+       fl6.flowi6_proto = IPPROTO_TCP;
+       fl6.saddr = niph->saddr;
+       fl6.daddr = niph->daddr;
+       fl6.fl6_sport = nth->source;
+       fl6.fl6_dport = nth->dest;
+       security_skb_classify_flow((struct sk_buff *)skb, flowi6_to_flowi(&fl6));
+       dst = ip6_route_output(net, NULL, &fl6);
+       if (dst == NULL || dst->error) {
+               dst_release(dst);
+               goto free_nskb;
+       }
+       dst = xfrm_lookup(net, dst, flowi6_to_flowi(&fl6), NULL, 0);
+       if (IS_ERR(dst))
+               goto free_nskb;
+
+       skb_dst_set(nskb, dst);
+
+       if (nfct) {
+               nskb->nfct = nfct;
+               nskb->nfctinfo = ctinfo;
+               nf_conntrack_get(nfct);
+       }
+
+       ip6_local_out(nskb);
+       return;
+
+free_nskb:
+       kfree_skb(nskb);
+}
+
+static void
+synproxy_send_client_synack(const struct sk_buff *skb, const struct tcphdr *th,
+                           const struct synproxy_options *opts)
+{
+       struct sk_buff *nskb;
+       struct ipv6hdr *iph, *niph;
+       struct tcphdr *nth;
+       unsigned int tcp_hdr_size;
+       u16 mss = opts->mss;
+
+       iph = ipv6_hdr(skb);
+
+       tcp_hdr_size = sizeof(*nth) + synproxy_options_size(opts);
+       nskb = alloc_skb(sizeof(*niph) + tcp_hdr_size + MAX_TCP_HEADER,
+                        GFP_ATOMIC);
+       if (nskb == NULL)
+               return;
+       skb_reserve(nskb, MAX_TCP_HEADER);
+
+       niph = synproxy_build_ip(nskb, &iph->daddr, &iph->saddr);
+
+       skb_reset_transport_header(nskb);
+       nth = (struct tcphdr *)skb_put(nskb, tcp_hdr_size);
+       nth->source     = th->dest;
+       nth->dest       = th->source;
+       nth->seq        = htonl(__cookie_v6_init_sequence(iph, th, &mss));
+       nth->ack_seq    = htonl(ntohl(th->seq) + 1);
+       tcp_flag_word(nth) = TCP_FLAG_SYN | TCP_FLAG_ACK;
+       if (opts->options & XT_SYNPROXY_OPT_ECN)
+               tcp_flag_word(nth) |= TCP_FLAG_ECE;
+       nth->doff       = tcp_hdr_size / 4;
+       nth->window     = 0;
+       nth->check      = 0;
+       nth->urg_ptr    = 0;
+
+       synproxy_build_options(nth, opts);
+
+       synproxy_send_tcp(skb, nskb, skb->nfct, IP_CT_ESTABLISHED_REPLY,
+                         niph, nth, tcp_hdr_size);
+}
+
+static void
+synproxy_send_server_syn(const struct synproxy_net *snet,
+                        const struct sk_buff *skb, const struct tcphdr *th,
+                        const struct synproxy_options *opts, u32 recv_seq)
+{
+       struct sk_buff *nskb;
+       struct ipv6hdr *iph, *niph;
+       struct tcphdr *nth;
+       unsigned int tcp_hdr_size;
+
+       iph = ipv6_hdr(skb);
+
+       tcp_hdr_size = sizeof(*nth) + synproxy_options_size(opts);
+       nskb = alloc_skb(sizeof(*niph) + tcp_hdr_size + MAX_TCP_HEADER,
+                        GFP_ATOMIC);
+       if (nskb == NULL)
+               return;
+       skb_reserve(nskb, MAX_TCP_HEADER);
+
+       niph = synproxy_build_ip(nskb, &iph->saddr, &iph->daddr);
+
+       skb_reset_transport_header(nskb);
+       nth = (struct tcphdr *)skb_put(nskb, tcp_hdr_size);
+       nth->source     = th->source;
+       nth->dest       = th->dest;
+       nth->seq        = htonl(recv_seq - 1);
+       /* ack_seq is used to relay our ISN to the synproxy hook to initialize
+        * sequence number translation once a connection tracking entry exists.
+        */
+       nth->ack_seq    = htonl(ntohl(th->ack_seq) - 1);
+       tcp_flag_word(nth) = TCP_FLAG_SYN;
+       if (opts->options & XT_SYNPROXY_OPT_ECN)
+               tcp_flag_word(nth) |= TCP_FLAG_ECE | TCP_FLAG_CWR;
+       nth->doff       = tcp_hdr_size / 4;
+       nth->window     = th->window;
+       nth->check      = 0;
+       nth->urg_ptr    = 0;
+
+       synproxy_build_options(nth, opts);
+
+       synproxy_send_tcp(skb, nskb, &snet->tmpl->ct_general, IP_CT_NEW,
+                         niph, nth, tcp_hdr_size);
+}
+
+static void
+synproxy_send_server_ack(const struct synproxy_net *snet,
+                        const struct ip_ct_tcp *state,
+                        const struct sk_buff *skb, const struct tcphdr *th,
+                        const struct synproxy_options *opts)
+{
+       struct sk_buff *nskb;
+       struct ipv6hdr *iph, *niph;
+       struct tcphdr *nth;
+       unsigned int tcp_hdr_size;
+
+       iph = ipv6_hdr(skb);
+
+       tcp_hdr_size = sizeof(*nth) + synproxy_options_size(opts);
+       nskb = alloc_skb(sizeof(*niph) + tcp_hdr_size + MAX_TCP_HEADER,
+                        GFP_ATOMIC);
+       if (nskb == NULL)
+               return;
+       skb_reserve(nskb, MAX_TCP_HEADER);
+
+       niph = synproxy_build_ip(nskb, &iph->daddr, &iph->saddr);
+
+       skb_reset_transport_header(nskb);
+       nth = (struct tcphdr *)skb_put(nskb, tcp_hdr_size);
+       nth->source     = th->dest;
+       nth->dest       = th->source;
+       nth->seq        = htonl(ntohl(th->ack_seq));
+       nth->ack_seq    = htonl(ntohl(th->seq) + 1);
+       tcp_flag_word(nth) = TCP_FLAG_ACK;
+       nth->doff       = tcp_hdr_size / 4;
+       nth->window     = htons(state->seen[IP_CT_DIR_ORIGINAL].td_maxwin);
+       nth->check      = 0;
+       nth->urg_ptr    = 0;
+
+       synproxy_build_options(nth, opts);
+
+       synproxy_send_tcp(skb, nskb, NULL, 0, niph, nth, tcp_hdr_size);
+}
+
+static void
+synproxy_send_client_ack(const struct synproxy_net *snet,
+                        const struct sk_buff *skb, const struct tcphdr *th,
+                        const struct synproxy_options *opts)
+{
+       struct sk_buff *nskb;
+       struct ipv6hdr *iph, *niph;
+       struct tcphdr *nth;
+       unsigned int tcp_hdr_size;
+
+       iph = ipv6_hdr(skb);
+
+       tcp_hdr_size = sizeof(*nth) + synproxy_options_size(opts);
+       nskb = alloc_skb(sizeof(*niph) + tcp_hdr_size + MAX_TCP_HEADER,
+                        GFP_ATOMIC);
+       if (nskb == NULL)
+               return;
+       skb_reserve(nskb, MAX_TCP_HEADER);
+
+       niph = synproxy_build_ip(nskb, &iph->saddr, &iph->daddr);
+
+       skb_reset_transport_header(nskb);
+       nth = (struct tcphdr *)skb_put(nskb, tcp_hdr_size);
+       nth->source     = th->source;
+       nth->dest       = th->dest;
+       nth->seq        = htonl(ntohl(th->seq) + 1);
+       nth->ack_seq    = th->ack_seq;
+       tcp_flag_word(nth) = TCP_FLAG_ACK;
+       nth->doff       = tcp_hdr_size / 4;
+       nth->window     = ntohs(htons(th->window) >> opts->wscale);
+       nth->check      = 0;
+       nth->urg_ptr    = 0;
+
+       synproxy_build_options(nth, opts);
+
+       synproxy_send_tcp(skb, nskb, NULL, 0, niph, nth, tcp_hdr_size);
+}
+
+static bool
+synproxy_recv_client_ack(const struct synproxy_net *snet,
+                        const struct sk_buff *skb, const struct tcphdr *th,
+                        struct synproxy_options *opts, u32 recv_seq)
+{
+       int mss;
+
+       mss = __cookie_v6_check(ipv6_hdr(skb), th, ntohl(th->ack_seq) - 1);
+       if (mss == 0) {
+               this_cpu_inc(snet->stats->cookie_invalid);
+               return false;
+       }
+
+       this_cpu_inc(snet->stats->cookie_valid);
+       opts->mss = mss;
+
+       if (opts->options & XT_SYNPROXY_OPT_TIMESTAMP)
+               synproxy_check_timestamp_cookie(opts);
+
+       synproxy_send_server_syn(snet, skb, th, opts, recv_seq);
+       return true;
+}
+
+static unsigned int
+synproxy_tg6(struct sk_buff *skb, const struct xt_action_param *par)
+{
+       const struct xt_synproxy_info *info = par->targinfo;
+       struct synproxy_net *snet = synproxy_pernet(dev_net(par->in));
+       struct synproxy_options opts = {};
+       struct tcphdr *th, _th;
+
+       if (nf_ip6_checksum(skb, par->hooknum, par->thoff, IPPROTO_TCP))
+               return NF_DROP;
+
+       th = skb_header_pointer(skb, par->thoff, sizeof(_th), &_th);
+       if (th == NULL)
+               return NF_DROP;
+
+       synproxy_parse_options(skb, par->thoff, th, &opts);
+
+       if (th->syn) {
+               /* Initial SYN from client */
+               this_cpu_inc(snet->stats->syn_received);
+
+               if (th->ece && th->cwr)
+                       opts.options |= XT_SYNPROXY_OPT_ECN;
+
+               opts.options &= info->options;
+               if (opts.options & XT_SYNPROXY_OPT_TIMESTAMP)
+                       synproxy_init_timestamp_cookie(info, &opts);
+               else
+                       opts.options &= ~(XT_SYNPROXY_OPT_WSCALE |
+                                         XT_SYNPROXY_OPT_SACK_PERM |
+                                         XT_SYNPROXY_OPT_ECN);
+
+               synproxy_send_client_synack(skb, th, &opts);
+       } else if (th->ack && !(th->fin || th->rst))
+               /* ACK from client */
+               synproxy_recv_client_ack(snet, skb, th, &opts, ntohl(th->seq));
+
+       return NF_DROP;
+}
+
+static unsigned int ipv6_synproxy_hook(unsigned int hooknum,
+                                      struct sk_buff *skb,
+                                      const struct net_device *in,
+                                      const struct net_device *out,
+                                      int (*okfn)(struct sk_buff *))
+{
+       struct synproxy_net *snet = synproxy_pernet(dev_net(in ? : out));
+       enum ip_conntrack_info ctinfo;
+       struct nf_conn *ct;
+       struct nf_conn_synproxy *synproxy;
+       struct synproxy_options opts = {};
+       const struct ip_ct_tcp *state;
+       struct tcphdr *th, _th;
+       __be16 frag_off;
+       u8 nexthdr;
+       int thoff;
+
+       ct = nf_ct_get(skb, &ctinfo);
+       if (ct == NULL)
+               return NF_ACCEPT;
+
+       synproxy = nfct_synproxy(ct);
+       if (synproxy == NULL)
+               return NF_ACCEPT;
+
+       if (nf_is_loopback_packet(skb))
+               return NF_ACCEPT;
+
+       nexthdr = ipv6_hdr(skb)->nexthdr;
+       thoff = ipv6_skip_exthdr(skb, sizeof(struct ipv6hdr), &nexthdr,
+                                &frag_off);
+       if (thoff < 0)
+               return NF_ACCEPT;
+
+       th = skb_header_pointer(skb, thoff, sizeof(_th), &_th);
+       if (th == NULL)
+               return NF_DROP;
+
+       state = &ct->proto.tcp;
+       switch (state->state) {
+       case TCP_CONNTRACK_CLOSE:
+               if (th->rst && !test_bit(IPS_SEEN_REPLY_BIT, &ct->status)) {
+                       nf_ct_seqadj_init(ct, ctinfo, synproxy->isn -
+                                                     ntohl(th->seq) + 1);
+                       break;
+               }
+
+               if (!th->syn || th->ack ||
+                   CTINFO2DIR(ctinfo) != IP_CT_DIR_ORIGINAL)
+                       break;
+
+               /* Reopened connection - reset the sequence number and timestamp
+                * adjustments, they will get initialized once the connection is
+                * reestablished.
+                */
+               nf_ct_seqadj_init(ct, ctinfo, 0);
+               synproxy->tsoff = 0;
+               this_cpu_inc(snet->stats->conn_reopened);
+
+               /* fall through */
+       case TCP_CONNTRACK_SYN_SENT:
+               synproxy_parse_options(skb, thoff, th, &opts);
+
+               if (!th->syn && th->ack &&
+                   CTINFO2DIR(ctinfo) == IP_CT_DIR_ORIGINAL) {
+                       /* Keep-Alives are sent with SEG.SEQ = SND.NXT-1,
+                        * therefore we need to add 1 to make the SYN sequence
+                        * number match the one of first SYN.
+                        */
+                       if (synproxy_recv_client_ack(snet, skb, th, &opts,
+                                                    ntohl(th->seq) + 1))
+                               this_cpu_inc(snet->stats->cookie_retrans);
+
+                       return NF_DROP;
+               }
+
+               synproxy->isn = ntohl(th->ack_seq);
+               if (opts.options & XT_SYNPROXY_OPT_TIMESTAMP)
+                       synproxy->its = opts.tsecr;
+               break;
+       case TCP_CONNTRACK_SYN_RECV:
+               if (!th->syn || !th->ack)
+                       break;
+
+               synproxy_parse_options(skb, thoff, th, &opts);
+               if (opts.options & XT_SYNPROXY_OPT_TIMESTAMP)
+                       synproxy->tsoff = opts.tsval - synproxy->its;
+
+               opts.options &= ~(XT_SYNPROXY_OPT_MSS |
+                                 XT_SYNPROXY_OPT_WSCALE |
+                                 XT_SYNPROXY_OPT_SACK_PERM);
+
+               swap(opts.tsval, opts.tsecr);
+               synproxy_send_server_ack(snet, state, skb, th, &opts);
+
+               nf_ct_seqadj_init(ct, ctinfo, synproxy->isn - ntohl(th->seq));
+
+               swap(opts.tsval, opts.tsecr);
+               synproxy_send_client_ack(snet, skb, th, &opts);
+
+               consume_skb(skb);
+               return NF_STOLEN;
+       default:
+               break;
+       }
+
+       synproxy_tstamp_adjust(skb, thoff, th, ct, ctinfo, synproxy);
+       return NF_ACCEPT;
+}
+
+static int synproxy_tg6_check(const struct xt_tgchk_param *par)
+{
+       const struct ip6t_entry *e = par->entryinfo;
+
+       if (!(e->ipv6.flags & IP6T_F_PROTO) ||
+           e->ipv6.proto != IPPROTO_TCP ||
+           e->ipv6.invflags & XT_INV_PROTO)
+               return -EINVAL;
+
+       return nf_ct_l3proto_try_module_get(par->family);
+}
+
+static void synproxy_tg6_destroy(const struct xt_tgdtor_param *par)
+{
+       nf_ct_l3proto_module_put(par->family);
+}
+
+static struct xt_target synproxy_tg6_reg __read_mostly = {
+       .name           = "SYNPROXY",
+       .family         = NFPROTO_IPV6,
+       .target         = synproxy_tg6,
+       .targetsize     = sizeof(struct xt_synproxy_info),
+       .checkentry     = synproxy_tg6_check,
+       .destroy        = synproxy_tg6_destroy,
+       .me             = THIS_MODULE,
+};
+
+static struct nf_hook_ops ipv6_synproxy_ops[] __read_mostly = {
+       {
+               .hook           = ipv6_synproxy_hook,
+               .owner          = THIS_MODULE,
+               .pf             = NFPROTO_IPV6,
+               .hooknum        = NF_INET_LOCAL_IN,
+               .priority       = NF_IP_PRI_CONNTRACK_CONFIRM - 1,
+       },
+       {
+               .hook           = ipv6_synproxy_hook,
+               .owner          = THIS_MODULE,
+               .pf             = NFPROTO_IPV6,
+               .hooknum        = NF_INET_POST_ROUTING,
+               .priority       = NF_IP_PRI_CONNTRACK_CONFIRM - 1,
+       },
+};
+
+static int __init synproxy_tg6_init(void)
+{
+       int err;
+
+       err = nf_register_hooks(ipv6_synproxy_ops,
+                               ARRAY_SIZE(ipv6_synproxy_ops));
+       if (err < 0)
+               goto err1;
+
+       err = xt_register_target(&synproxy_tg6_reg);
+       if (err < 0)
+               goto err2;
+
+       return 0;
+
+err2:
+       nf_unregister_hooks(ipv6_synproxy_ops, ARRAY_SIZE(ipv6_synproxy_ops));
+err1:
+       return err;
+}
+
+static void __exit synproxy_tg6_exit(void)
+{
+       xt_unregister_target(&synproxy_tg6_reg);
+       nf_unregister_hooks(ipv6_synproxy_ops, ARRAY_SIZE(ipv6_synproxy_ops));
+}
+
+module_init(synproxy_tg6_init);
+module_exit(synproxy_tg6_exit);
+
+MODULE_LICENSE("GPL");
+MODULE_AUTHOR("Patrick McHardy <kaber@trash.net>");