net: ip: make ip_route_input_slow() return drop reasons

JIRA: https://issues.redhat.com/browse/RHEL-88891
Upstream Status: linux.git
Conflicts:\
- Minor code differences as the tos to dscp conversion wasn't done in
  c10s yet.

commit 5b92112
Author: Menglong Dong <menglong8.dong@gmail.com>
Date:   Thu Nov 7 20:55:56 2024 +0800

    net: ip: make ip_route_input_slow() return drop reasons

    In this commit, we make ip_route_input_slow() return skb drop reasons,
    and following new skb drop reasons are added:

      SKB_DROP_REASON_IP_INVALID_DEST

    The only caller of ip_route_input_slow() is ip_route_input_rcu(), and we
    adjust it by making it return -EINVAL on error.

    Signed-off-by: Menglong Dong <dongml2@chinatelecom.cn>
    Signed-off-by: Paolo Abeni <pabeni@redhat.com>

Signed-off-by: Antoine Tenart <atenart@redhat.com>

Author: atenart

include/net/dropreason-core.h

@@ -79,6 +79,7 @@
 	FN(IP_LOCAL_SOURCE)		\
 	FN(IP_INVALID_SOURCE)		\
 	FN(IP_LOCALNET)			\
+	FN(IP_INVALID_DEST)		\
 	FN(PKT_TOO_BIG)			\
 	FN(DUP_FRAG)			\
 	FN(FRAG_REASM_TIMEOUT)		\
@@ -386,6 +387,11 @@ enum skb_drop_reason {
 	SKB_DROP_REASON_IP_INVALID_SOURCE,
 	/** @SKB_DROP_REASON_IP_LOCALNET: source or dest ip is local net */
 	SKB_DROP_REASON_IP_LOCALNET,
+	/**
+	 * @SKB_DROP_REASON_IP_INVALID_DEST: the dest ip is invalid:
+	 * 1) dest ip is 0
+	 */
+	SKB_DROP_REASON_IP_INVALID_DEST,
 	/**
 	 * @SKB_DROP_REASON_PKT_TOO_BIG: packet size is too big (maybe exceed the
 	 * MTU)

net/ipv4/route.c

@@ -2219,9 +2219,10 @@ static struct net_device *ip_rt_get_dev(struct net *net,
  *	called with rcu_read_lock()
  */
 
-static int ip_route_input_slow(struct sk_buff *skb, __be32 daddr, __be32 saddr,
-			       u8 tos, struct net_device *dev,
-			       struct fib_result *res)
+static enum skb_drop_reason
+ip_route_input_slow(struct sk_buff *skb, __be32 daddr, __be32 saddr,
+		    u8 tos, struct net_device *dev,
+		    struct fib_result *res)
 {
 	enum skb_drop_reason reason = SKB_DROP_REASON_NOT_SPECIFIED;
 	struct in_device *in_dev = __in_dev_get_rcu(dev);
@@ -2251,8 +2252,10 @@ static int ip_route_input_slow(struct sk_buff *skb, __be32 daddr, __be32 saddr,
 		fl4.flowi4_tun_key.tun_id = 0;
 	skb_dst_drop(skb);
 
-	if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr))
+	if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr)) {
+		reason = SKB_DROP_REASON_IP_INVALID_SOURCE;
 		goto martian_source;
+	}
 
 	res->fi = NULL;
 	res->table = NULL;
@@ -2262,21 +2265,29 @@ static int ip_route_input_slow(struct sk_buff *skb, __be32 daddr, __be32 saddr,
 	/* Accept zero addresses only to limited broadcast;
 	 * I even do not know to fix it or not. Waiting for complains :-)
 	 */
-	if (ipv4_is_zeronet(saddr))
+	if (ipv4_is_zeronet(saddr)) {
+		reason = SKB_DROP_REASON_IP_INVALID_SOURCE;
 		goto martian_source;
+	}
 
-	if (ipv4_is_zeronet(daddr))
+	if (ipv4_is_zeronet(daddr)) {
+		reason = SKB_DROP_REASON_IP_INVALID_DEST;
 		goto martian_destination;
+	}
 
 	/* Following code try to avoid calling IN_DEV_NET_ROUTE_LOCALNET(),
 	 * and call it once if daddr or/and saddr are loopback addresses
 	 */
 	if (ipv4_is_loopback(daddr)) {
-		if (!IN_DEV_NET_ROUTE_LOCALNET(in_dev, net))
+		if (!IN_DEV_NET_ROUTE_LOCALNET(in_dev, net)) {
+			reason = SKB_DROP_REASON_IP_LOCALNET;
 			goto martian_destination;
+		}
 	} else if (ipv4_is_loopback(saddr)) {
-		if (!IN_DEV_NET_ROUTE_LOCALNET(in_dev, net))
+		if (!IN_DEV_NET_ROUTE_LOCALNET(in_dev, net)) {
+			reason = SKB_DROP_REASON_IP_LOCALNET;
 			goto martian_source;
+		}
 	}
 
 	/*
@@ -2331,19 +2342,26 @@ static int ip_route_input_slow(struct sk_buff *skb, __be32 daddr, __be32 saddr,
 		err = -EHOSTUNREACH;
 		goto no_route;
 	}
-	if (res->type != RTN_UNICAST)
+	if (res->type != RTN_UNICAST) {
+		reason = SKB_DROP_REASON_IP_INVALID_DEST;
 		goto martian_destination;
+	}
 
 make_route:
 	err = ip_mkroute_input(skb, res, in_dev, daddr, saddr, tos, flkeys);
-out:	return err;
+	if (!err)
+		reason = SKB_NOT_DROPPED_YET;
+
+out:
+	return reason;
 
 brd_input:
-	if (skb->protocol != htons(ETH_P_IP))
-		goto e_inval;
+	if (skb->protocol != htons(ETH_P_IP)) {
+		reason = SKB_DROP_REASON_INVALID_PROTO;
+		goto out;
+	}
 
 	if (!ipv4_is_zeronet(saddr)) {
-		err = -EINVAL;
 		reason = fib_validate_source_reason(skb, saddr, 0, tos, 0,
 						    dev, in_dev, &itag);
 		if (reason)
@@ -2364,7 +2382,7 @@ out:	return err;
 		rth = rcu_dereference(nhc->nhc_rth_input);
 		if (rt_cache_valid(rth)) {
 			skb_dst_set_noref(skb, &rth->dst);
-			err = 0;
+			reason = SKB_NOT_DROPPED_YET;
 			goto out;
 		}
 	}
@@ -2401,7 +2419,7 @@ out:	return err;
 			rt_add_uncached_list(rth);
 	}
 	skb_dst_set(skb, &rth->dst);
-	err = 0;
+	reason = SKB_NOT_DROPPED_YET;
 	goto out;
 
 no_route:
@@ -2422,12 +2440,8 @@ out:	return err;
 				     &daddr, &saddr, dev->name);
 #endif
 
-e_inval:
-	err = -EINVAL;
-	goto out;
-
 e_nobufs:
-	err = -ENOBUFS;
+	reason = SKB_DROP_REASON_NOMEM;
 	goto out;
 
 martian_source:
@@ -2483,7 +2497,7 @@ static int ip_route_input_rcu(struct sk_buff *skb, __be32 daddr, __be32 saddr,
 		return reason ? -EINVAL : 0;
 	}
 
-	return ip_route_input_slow(skb, daddr, saddr, tos, dev, res);
+	return ip_route_input_slow(skb, daddr, saddr, tos, dev, res) ? -EINVAL : 0;
 }
 
 int ip_route_input_noref(struct sk_buff *skb, __be32 daddr, __be32 saddr,