--- /dev/null
+/* SPDX-License-Identifier: GPL-2.0 */
+#include <linux/module.h>
+#include <linux/netfilter/nf_tables.h>
+#include <net/netfilter/nf_tables.h>
+#include <net/netfilter/nf_tables_core.h>
+#include <net/netfilter/nf_tproxy.h>
+#include <net/inet_sock.h>
+#include <net/tcp.h>
+#include <linux/if_ether.h>
+#include <net/netfilter/ipv4/nf_defrag_ipv4.h>
+#if IS_ENABLED(CONFIG_NF_TABLES_IPV6)
+#include <net/netfilter/ipv6/nf_defrag_ipv6.h>
+#endif
+
+struct nft_tproxy {
+       enum nft_registers      sreg_addr:8;
+       enum nft_registers      sreg_port:8;
+       u8                      family;
+};
+
+static void nft_tproxy_eval_v4(const struct nft_expr *expr,
+                              struct nft_regs *regs,
+                              const struct nft_pktinfo *pkt)
+{
+       const struct nft_tproxy *priv = nft_expr_priv(expr);
+       struct sk_buff *skb = pkt->skb;
+       const struct iphdr *iph = ip_hdr(skb);
+       struct udphdr _hdr, *hp;
+       __be32 taddr = 0;
+       __be16 tport = 0;
+       struct sock *sk;
+
+       hp = skb_header_pointer(skb, ip_hdrlen(skb), sizeof(_hdr), &_hdr);
+       if (!hp) {
+               regs->verdict.code = NFT_BREAK;
+               return;
+       }
+
+       /* check if there's an ongoing connection on the packet addresses, this
+        * happens if the redirect already happened and the current packet
+        * belongs to an already established connection
+        */
+       sk = nf_tproxy_get_sock_v4(nft_net(pkt), skb, iph->protocol,
+                                  iph->saddr, iph->daddr,
+                                  hp->source, hp->dest,
+                                  skb->dev, NF_TPROXY_LOOKUP_ESTABLISHED);
+
+       if (priv->sreg_addr)
+               taddr = regs->data[priv->sreg_addr];
+       taddr = nf_tproxy_laddr4(skb, taddr, iph->daddr);
+
+       if (priv->sreg_port)
+               tport = regs->data[priv->sreg_port];
+       if (!tport)
+               tport = hp->dest;
+
+       /* UDP has no TCP_TIME_WAIT state, so we never enter here */
+       if (sk && sk->sk_state == TCP_TIME_WAIT) {
+               /* reopening a TIME_WAIT connection needs special handling */
+               sk = nf_tproxy_handle_time_wait4(nft_net(pkt), skb, taddr, tport, sk);
+       } else if (!sk) {
+               /* no, there's no established connection, check if
+                * there's a listener on the redirected addr/port
+                */
+               sk = nf_tproxy_get_sock_v4(nft_net(pkt), skb, iph->protocol,
+                                          iph->saddr, taddr,
+                                          hp->source, tport,
+                                          skb->dev, NF_TPROXY_LOOKUP_LISTENER);
+       }
+
+       if (sk && nf_tproxy_sk_is_transparent(sk))
+               nf_tproxy_assign_sock(skb, sk);
+       else
+               regs->verdict.code = NFT_BREAK;
+}
+
+#if IS_ENABLED(CONFIG_NF_TABLES_IPV6)
+static void nft_tproxy_eval_v6(const struct nft_expr *expr,
+                              struct nft_regs *regs,
+                              const struct nft_pktinfo *pkt)
+{
+       const struct nft_tproxy *priv = nft_expr_priv(expr);
+       struct sk_buff *skb = pkt->skb;
+       const struct ipv6hdr *iph = ipv6_hdr(skb);
+       struct in6_addr taddr = {0};
+       int thoff = pkt->xt.thoff;
+       struct udphdr _hdr, *hp;
+       __be16 tport = 0;
+       struct sock *sk;
+       int l4proto;
+
+       if (!pkt->tprot_set) {
+               regs->verdict.code = NFT_BREAK;
+               return;
+       }
+       l4proto = pkt->tprot;
+
+       hp = skb_header_pointer(skb, thoff, sizeof(_hdr), &_hdr);
+       if (hp == NULL) {
+               regs->verdict.code = NFT_BREAK;
+               return;
+       }
+
+       /* check if there's an ongoing connection on the packet addresses, this
+        * happens if the redirect already happened and the current packet
+        * belongs to an already established connection
+        */
+       sk = nf_tproxy_get_sock_v6(nft_net(pkt), skb, thoff, l4proto,
+                                  &iph->saddr, &iph->daddr,
+                                  hp->source, hp->dest,
+                                  nft_in(pkt), NF_TPROXY_LOOKUP_ESTABLISHED);
+
+       if (priv->sreg_addr)
+               memcpy(&taddr, ®s->data[priv->sreg_addr], sizeof(taddr));
+       taddr = *nf_tproxy_laddr6(skb, &taddr, &iph->daddr);
+
+       if (priv->sreg_port)
+               tport = regs->data[priv->sreg_port];
+       if (!tport)
+               tport = hp->dest;
+
+       /* UDP has no TCP_TIME_WAIT state, so we never enter here */
+       if (sk && sk->sk_state == TCP_TIME_WAIT) {
+               /* reopening a TIME_WAIT connection needs special handling */
+               sk = nf_tproxy_handle_time_wait6(skb, l4proto, thoff,
+                                                nft_net(pkt),
+                                                &taddr,
+                                                tport,
+                                                sk);
+       } else if (!sk) {
+               /* no there's no established connection, check if
+                * there's a listener on the redirected addr/port
+                */
+               sk = nf_tproxy_get_sock_v6(nft_net(pkt), skb, thoff,
+                                          l4proto, &iph->saddr, &taddr,
+                                          hp->source, tport,
+                                          nft_in(pkt), NF_TPROXY_LOOKUP_LISTENER);
+       }
+
+       /* NOTE: assign_sock consumes our sk reference */
+       if (sk && nf_tproxy_sk_is_transparent(sk))
+               nf_tproxy_assign_sock(skb, sk);
+       else
+               regs->verdict.code = NFT_BREAK;
+}
+#endif
+
+static void nft_tproxy_eval(const struct nft_expr *expr,
+                           struct nft_regs *regs,
+                           const struct nft_pktinfo *pkt)
+{
+       const struct nft_tproxy *priv = nft_expr_priv(expr);
+
+       switch (nft_pf(pkt)) {
+       case NFPROTO_IPV4:
+               switch (priv->family) {
+               case NFPROTO_IPV4:
+               case NFPROTO_UNSPEC:
+                       nft_tproxy_eval_v4(expr, regs, pkt);
+                       return;
+               }
+               break;
+#if IS_ENABLED(CONFIG_NF_TABLES_IPV6)
+       case NFPROTO_IPV6:
+               switch (priv->family) {
+               case NFPROTO_IPV6:
+               case NFPROTO_UNSPEC:
+                       nft_tproxy_eval_v6(expr, regs, pkt);
+                       return;
+               }
+#endif
+       }
+       regs->verdict.code = NFT_BREAK;
+}
+
+static const struct nla_policy nft_tproxy_policy[NFTA_TPROXY_MAX + 1] = {
+       [NFTA_TPROXY_FAMILY]   = { .type = NLA_U32 },
+       [NFTA_TPROXY_REG_ADDR] = { .type = NLA_U32 },
+       [NFTA_TPROXY_REG_PORT] = { .type = NLA_U32 },
+};
+
+static int nft_tproxy_init(const struct nft_ctx *ctx,
+                          const struct nft_expr *expr,
+                          const struct nlattr * const tb[])
+{
+       struct nft_tproxy *priv = nft_expr_priv(expr);
+       unsigned int alen = 0;
+       int err;
+
+       if (!tb[NFTA_TPROXY_FAMILY] ||
+           (!tb[NFTA_TPROXY_REG_ADDR] && !tb[NFTA_TPROXY_REG_PORT]))
+               return -EINVAL;
+
+       priv->family = ntohl(nla_get_be32(tb[NFTA_TPROXY_FAMILY]));
+
+       switch (ctx->family) {
+       case NFPROTO_IPV4:
+               if (priv->family != NFPROTO_IPV4)
+                       return -EINVAL;
+               break;
+#if IS_ENABLED(CONFIG_NF_TABLES_IPV6)
+       case NFPROTO_IPV6:
+               if (priv->family != NFPROTO_IPV6)
+                       return -EINVAL;
+               break;
+#endif
+       case NFPROTO_INET:
+               break;
+       default:
+               return -EOPNOTSUPP;
+       }
+
+       /* Address is specified but the rule family is not set accordingly */
+       if (priv->family == NFPROTO_UNSPEC && tb[NFTA_TPROXY_REG_ADDR])
+               return -EINVAL;
+
+       switch (priv->family) {
+       case NFPROTO_IPV4:
+               alen = FIELD_SIZEOF(union nf_inet_addr, in);
+               err = nf_defrag_ipv4_enable(ctx->net);
+               if (err)
+                       return err;
+               break;
+#if IS_ENABLED(CONFIG_NF_TABLES_IPV6)
+       case NFPROTO_IPV6:
+               alen = FIELD_SIZEOF(union nf_inet_addr, in6);
+               err = nf_defrag_ipv6_enable(ctx->net);
+               if (err)
+                       return err;
+               break;
+#endif
+       case NFPROTO_UNSPEC:
+               /* No address is specified here */
+               err = nf_defrag_ipv4_enable(ctx->net);
+               if (err)
+                       return err;
+               err = nf_defrag_ipv6_enable(ctx->net);
+               if (err)
+                       return err;
+               break;
+       default:
+               return -EOPNOTSUPP;
+       }
+
+       if (tb[NFTA_TPROXY_REG_ADDR]) {
+               priv->sreg_addr = nft_parse_register(tb[NFTA_TPROXY_REG_ADDR]);
+               err = nft_validate_register_load(priv->sreg_addr, alen);
+               if (err < 0)
+                       return err;
+       }
+
+       if (tb[NFTA_TPROXY_REG_PORT]) {
+               priv->sreg_port = nft_parse_register(tb[NFTA_TPROXY_REG_PORT]);
+               err = nft_validate_register_load(priv->sreg_port, sizeof(u16));
+               if (err < 0)
+                       return err;
+       }
+
+       return 0;
+}
+
+static int nft_tproxy_dump(struct sk_buff *skb,
+                          const struct nft_expr *expr)
+{
+       const struct nft_tproxy *priv = nft_expr_priv(expr);
+
+       if (nla_put_be32(skb, NFTA_TPROXY_FAMILY, htonl(priv->family)))
+               return -1;
+
+       if (priv->sreg_addr &&
+           nft_dump_register(skb, NFTA_TPROXY_REG_ADDR, priv->sreg_addr))
+               return -1;
+
+       if (priv->sreg_port &&
+           nft_dump_register(skb, NFTA_TPROXY_REG_PORT, priv->sreg_port))
+                       return -1;
+
+       return 0;
+}
+
+static struct nft_expr_type nft_tproxy_type;
+static const struct nft_expr_ops nft_tproxy_ops = {
+       .type           = &nft_tproxy_type,
+       .size           = NFT_EXPR_SIZE(sizeof(struct nft_tproxy)),
+       .eval           = nft_tproxy_eval,
+       .init           = nft_tproxy_init,
+       .dump           = nft_tproxy_dump,
+};
+
+static struct nft_expr_type nft_tproxy_type __read_mostly = {
+       .name           = "tproxy",
+       .ops            = &nft_tproxy_ops,
+       .policy         = nft_tproxy_policy,
+       .maxattr        = NFTA_TPROXY_MAX,
+       .owner          = THIS_MODULE,
+};
+
+static int __init nft_tproxy_module_init(void)
+{
+       return nft_register_expr(&nft_tproxy_type);
+}
+
+static void __exit nft_tproxy_module_exit(void)
+{
+       nft_unregister_expr(&nft_tproxy_type);
+}
+
+module_init(nft_tproxy_module_init);
+module_exit(nft_tproxy_module_exit);
+
+MODULE_LICENSE("GPL");
+MODULE_AUTHOR("Máté Eckl");
+MODULE_DESCRIPTION("nf_tables tproxy support module");
+MODULE_ALIAS_NFT_EXPR("tproxy");