netfilter: flowtable: split IPv6 datapath in helper functions
authorPablo Neira Ayuso <pablo@netfilter.org>
Thu, 11 May 2023 05:35:35 +0000 (07:35 +0200)
committerFlorian Westphal <fw@strlen.de>
Thu, 18 May 2023 06:48:55 +0000 (08:48 +0200)
Add context structure and helper functions to look up for a matching
IPv6 entry in the flowtable and to forward packets.

No functional changes are intended.

Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
Signed-off-by: Florian Westphal <fw@strlen.de>
net/netfilter/nf_flow_table_ip.c

index 3fb4761..d248763 100644 (file)
@@ -570,32 +570,31 @@ static void nf_flow_nat_ipv6(const struct flow_offload *flow,
        }
 }
 
-static int nf_flow_tuple_ipv6(struct sk_buff *skb, const struct net_device *dev,
-                             struct flow_offload_tuple *tuple, u32 *hdrsize,
-                             u32 offset)
+static int nf_flow_tuple_ipv6(struct nf_flowtable_ctx *ctx, struct sk_buff *skb,
+                             struct flow_offload_tuple *tuple)
 {
        struct flow_ports *ports;
        struct ipv6hdr *ip6h;
        unsigned int thoff;
        u8 nexthdr;
 
-       thoff = sizeof(*ip6h) + offset;
+       thoff = sizeof(*ip6h) + ctx->offset;
        if (!pskb_may_pull(skb, thoff))
                return -1;
 
-       ip6h = (struct ipv6hdr *)(skb_network_header(skb) + offset);
+       ip6h = (struct ipv6hdr *)(skb_network_header(skb) + ctx->offset);
 
        nexthdr = ip6h->nexthdr;
        switch (nexthdr) {
        case IPPROTO_TCP:
-               *hdrsize = sizeof(struct tcphdr);
+               ctx->hdrsize = sizeof(struct tcphdr);
                break;
        case IPPROTO_UDP:
-               *hdrsize = sizeof(struct udphdr);
+               ctx->hdrsize = sizeof(struct udphdr);
                break;
 #ifdef CONFIG_NF_CT_PROTO_GRE
        case IPPROTO_GRE:
-               *hdrsize = sizeof(struct gre_base_hdr);
+               ctx->hdrsize = sizeof(struct gre_base_hdr);
                break;
 #endif
        default:
@@ -605,7 +604,7 @@ static int nf_flow_tuple_ipv6(struct sk_buff *skb, const struct net_device *dev,
        if (ip6h->hop_limit <= 1)
                return -1;
 
-       if (!pskb_may_pull(skb, thoff + *hdrsize))
+       if (!pskb_may_pull(skb, thoff + ctx->hdrsize))
                return -1;
 
        switch (nexthdr) {
@@ -625,65 +624,47 @@ static int nf_flow_tuple_ipv6(struct sk_buff *skb, const struct net_device *dev,
        }
        }
 
-       ip6h = (struct ipv6hdr *)(skb_network_header(skb) + offset);
+       ip6h = (struct ipv6hdr *)(skb_network_header(skb) + ctx->offset);
 
        tuple->src_v6           = ip6h->saddr;
        tuple->dst_v6           = ip6h->daddr;
        tuple->l3proto          = AF_INET6;
        tuple->l4proto          = nexthdr;
-       tuple->iifidx           = dev->ifindex;
+       tuple->iifidx           = ctx->in->ifindex;
        nf_flow_tuple_encap(skb, tuple);
 
        return 0;
 }
 
-unsigned int
-nf_flow_offload_ipv6_hook(void *priv, struct sk_buff *skb,
-                         const struct nf_hook_state *state)
+static int nf_flow_offload_ipv6_forward(struct nf_flowtable_ctx *ctx,
+                                       struct nf_flowtable *flow_table,
+                                       struct flow_offload_tuple_rhash *tuplehash,
+                                       struct sk_buff *skb)
 {
-       struct flow_offload_tuple_rhash *tuplehash;
-       struct nf_flowtable *flow_table = priv;
-       struct flow_offload_tuple tuple = {};
        enum flow_offload_tuple_dir dir;
-       const struct in6_addr *nexthop;
        struct flow_offload *flow;
-       struct net_device *outdev;
        unsigned int thoff, mtu;
-       u32 hdrsize, offset = 0;
        struct ipv6hdr *ip6h;
-       struct rt6_info *rt;
-       int ret;
-
-       if (skb->protocol != htons(ETH_P_IPV6) &&
-           !nf_flow_skb_encap_protocol(skb, htons(ETH_P_IPV6), &offset))
-               return NF_ACCEPT;
-
-       if (nf_flow_tuple_ipv6(skb, state->in, &tuple, &hdrsize, offset) < 0)
-               return NF_ACCEPT;
-
-       tuplehash = flow_offload_lookup(flow_table, &tuple);
-       if (tuplehash == NULL)
-               return NF_ACCEPT;
 
        dir = tuplehash->tuple.dir;
        flow = container_of(tuplehash, struct flow_offload, tuplehash[dir]);
 
-       mtu = flow->tuplehash[dir].tuple.mtu + offset;
+       mtu = flow->tuplehash[dir].tuple.mtu + ctx->offset;
        if (unlikely(nf_flow_exceeds_mtu(skb, mtu)))
-               return NF_ACCEPT;
+               return 0;
 
-       ip6h = (struct ipv6hdr *)(skb_network_header(skb) + offset);
-       thoff = sizeof(*ip6h) + offset;
+       ip6h = (struct ipv6hdr *)(skb_network_header(skb) + ctx->offset);
+       thoff = sizeof(*ip6h) + ctx->offset;
        if (nf_flow_state_check(flow, ip6h->nexthdr, skb, thoff))
-               return NF_ACCEPT;
+               return 0;
 
        if (!nf_flow_dst_check(&tuplehash->tuple)) {
                flow_offload_teardown(flow);
-               return NF_ACCEPT;
+               return 0;
        }
 
-       if (skb_try_make_writable(skb, thoff + hdrsize))
-               return NF_DROP;
+       if (skb_try_make_writable(skb, thoff + ctx->hdrsize))
+               return -1;
 
        flow_offload_refresh(flow_table, flow);
 
@@ -698,6 +679,52 @@ nf_flow_offload_ipv6_hook(void *priv, struct sk_buff *skb,
        if (flow_table->flags & NF_FLOWTABLE_COUNTER)
                nf_ct_acct_update(flow->ct, tuplehash->tuple.dir, skb->len);
 
+       return 1;
+}
+
+static struct flow_offload_tuple_rhash *
+nf_flow_offload_ipv6_lookup(struct nf_flowtable_ctx *ctx,
+                           struct nf_flowtable *flow_table,
+                           struct sk_buff *skb)
+{
+       struct flow_offload_tuple tuple = {};
+
+       if (skb->protocol != htons(ETH_P_IPV6) &&
+           !nf_flow_skb_encap_protocol(skb, htons(ETH_P_IPV6), &ctx->offset))
+               return NULL;
+
+       if (nf_flow_tuple_ipv6(ctx, skb, &tuple) < 0)
+               return NULL;
+
+       return flow_offload_lookup(flow_table, &tuple);
+}
+
+unsigned int
+nf_flow_offload_ipv6_hook(void *priv, struct sk_buff *skb,
+                         const struct nf_hook_state *state)
+{
+       struct flow_offload_tuple_rhash *tuplehash;
+       struct nf_flowtable *flow_table = priv;
+       enum flow_offload_tuple_dir dir;
+       struct nf_flowtable_ctx ctx = {
+               .in     = state->in,
+       };
+       const struct in6_addr *nexthop;
+       struct flow_offload *flow;
+       struct net_device *outdev;
+       struct rt6_info *rt;
+       int ret;
+
+       tuplehash = nf_flow_offload_ipv6_lookup(&ctx, flow_table, skb);
+       if (tuplehash == NULL)
+               return NF_ACCEPT;
+
+       ret = nf_flow_offload_ipv6_forward(&ctx, flow_table, tuplehash, skb);
+       if (ret < 0)
+               return NF_DROP;
+       else if (ret == 0)
+               return NF_ACCEPT;
+
        if (unlikely(tuplehash->tuple.xmit_type == FLOW_OFFLOAD_XMIT_XFRM)) {
                rt = (struct rt6_info *)tuplehash->tuple.dst_cache;
                memset(skb->cb, 0, sizeof(struct inet6_skb_parm));
@@ -706,6 +733,9 @@ nf_flow_offload_ipv6_hook(void *priv, struct sk_buff *skb,
                return nf_flow_xmit_xfrm(skb, state, &rt->dst);
        }
 
+       dir = tuplehash->tuple.dir;
+       flow = container_of(tuplehash, struct flow_offload, tuplehash[dir]);
+
        switch (tuplehash->tuple.xmit_type) {
        case FLOW_OFFLOAD_XMIT_NEIGH:
                rt = (struct rt6_info *)tuplehash->tuple.dst_cache;