[PATCH net-next] netfilter: Update ip_route_me_harder to consider L3 domain

[PATCH net-next] netfilter: Update ip_route_me_harder to consider L3 domain

[David Ahern]

ip_route_me_harder is not considering the L3 domain and sending lookups
to the wrong table. For example consider the following output rule:

iptables -I OUTPUT -p tcp --dport 12345 -j REJECT --reject-with tcp-reset

using perf to analyze lookups via the fib_table_lookup tracepoint shows:

vrf-test  1187 [001] 46887.295927: fib:fib_table_lookup: table 255 oif 0 iif 0 src 0.0.0.0 dst 10.100.1.254 tos 0 scope 0 flags 0
        ffffffff8143922c perf_trace_fib_table_lookup ([kernel.kallsyms])
        ffffffff81493aac fib_table_lookup ([kernel.kallsyms])
        ffffffff8148dda3 __inet_dev_addr_type ([kernel.kallsyms])
        ffffffff8148ddf6 inet_addr_type ([kernel.kallsyms])
        ffffffff8149e344 ip_route_me_harder ([kernel.kallsyms])

and

vrf-test  1187 [001] 46887.295933: fib:fib_table_lookup: table 255 oif 0 iif 1 src 10.100.1.254 dst 10.100.1.2 tos 0 scope 0 flags
        ffffffff8143922c perf_trace_fib_table_lookup ([kernel.kallsyms])
        ffffffff81493aac fib_table_lookup ([kernel.kallsyms])
        ffffffff814998ff fib4_rule_action ([kernel.kallsyms])
        ffffffff81437f35 fib_rules_lookup ([kernel.kallsyms])
        ffffffff81499758 __fib_lookup ([kernel.kallsyms])
        ffffffff8144f010 fib_lookup.constprop.34 ([kernel.kallsyms])
        ffffffff8144f759 __ip_route_output_key_hash ([kernel.kallsyms])
        ffffffff8144fc6a ip_route_output_flow ([kernel.kallsyms])
        ffffffff8149e39b ip_route_me_harder ([kernel.kallsyms])

Updating both lookups to pull the L3 domain from the dst currently
attached to the skb directs both lookups to the correct table.

Signed-off-by: David Ahern <dsa@cumulusnetworks.com>
---
Pablo: from a code review it seems ip_route_me_harder is only called in
       the output path and after skb_dst is set.

 net/ipv4/netfilter.c | 5 ++++-
 1 file changed, 4 insertions(+), 1 deletion(-)

diff --git a/net/ipv4/netfilter.c b/net/ipv4/netfilter.c
index c3776ff6749f..b3cc1335adbc 100644
--- a/net/ipv4/netfilter.c
+++ b/net/ipv4/netfilter.c
@@ -24,10 +24,11 @@ int ip_route_me_harder(struct net *net, struct sk_buff *skb, unsigned int addr_t
 	struct flowi4 fl4 = {};
 	__be32 saddr = iph->saddr;
 	__u8 flags = skb->sk ? inet_sk_flowi_flags(skb->sk) : 0;
+	struct net_device *dev = skb_dst(skb)->dev;
 	unsigned int hh_len;
 
 	if (addr_type == RTN_UNSPEC)
-		addr_type = inet_addr_type(net, saddr);
+		addr_type = inet_addr_type_dev_table(net, dev, saddr);
 	if (addr_type == RTN_LOCAL || addr_type == RTN_UNICAST)
 		flags |= FLOWI_FLAG_ANYSRC;
 	else
@@ -40,6 +41,8 @@ int ip_route_me_harder(struct net *net, struct sk_buff *skb, unsigned int addr_t
 	fl4.saddr = saddr;
 	fl4.flowi4_tos = RT_TOS(iph->tos);
 	fl4.flowi4_oif = skb->sk ? skb->sk->sk_bound_dev_if : 0;
+	if (!fl4.flowi4_oif)
+		fl4.flowi4_oif = l3mdev_master_ifindex(dev);
 	fl4.flowi4_mark = skb->mark;
 	fl4.flowi4_flags = flags;
 	rt = ip_route_output_key(net, &fl4);
-- 
2.1.4

Re: [PATCH net-next] netfilter: Update ip_route_me_harder to consider L3 domain

[David Ahern]

On 11/3/16 11:43 AM, David Ahern wrote:
> ip_route_me_harder is not considering the L3 domain and sending lookups
> to the wrong table. For example consider the following output rule:
> 
> iptables -I OUTPUT -p tcp --dport 12345 -j REJECT --reject-with tcp-reset
> 
> using perf to analyze lookups via the fib_table_lookup tracepoint shows:
> 
> vrf-test  1187 [001] 46887.295927: fib:fib_table_lookup: table 255 oif 0 iif 0 src 0.0.0.0 dst 10.100.1.254 tos 0 scope 0 flags 0
>         ffffffff8143922c perf_trace_fib_table_lookup ([kernel.kallsyms])
>         ffffffff81493aac fib_table_lookup ([kernel.kallsyms])
>         ffffffff8148dda3 __inet_dev_addr_type ([kernel.kallsyms])
>         ffffffff8148ddf6 inet_addr_type ([kernel.kallsyms])
>         ffffffff8149e344 ip_route_me_harder ([kernel.kallsyms])
> 
> and
> 
> vrf-test  1187 [001] 46887.295933: fib:fib_table_lookup: table 255 oif 0 iif 1 src 10.100.1.254 dst 10.100.1.2 tos 0 scope 0 flags
>         ffffffff8143922c perf_trace_fib_table_lookup ([kernel.kallsyms])
>         ffffffff81493aac fib_table_lookup ([kernel.kallsyms])
>         ffffffff814998ff fib4_rule_action ([kernel.kallsyms])
>         ffffffff81437f35 fib_rules_lookup ([kernel.kallsyms])
>         ffffffff81499758 __fib_lookup ([kernel.kallsyms])
>         ffffffff8144f010 fib_lookup.constprop.34 ([kernel.kallsyms])
>         ffffffff8144f759 __ip_route_output_key_hash ([kernel.kallsyms])
>         ffffffff8144fc6a ip_route_output_flow ([kernel.kallsyms])
>         ffffffff8149e39b ip_route_me_harder ([kernel.kallsyms])
> 
> Updating both lookups to pull the L3 domain from the dst currently
> attached to the skb directs both lookups to the correct table.
> 
> Signed-off-by: David Ahern <dsa@cumulusnetworks.com>
> ---
> Pablo: from a code review it seems ip_route_me_harder is only called in
>        the output path and after skb_dst is set.

Correction: the 'output path' comment is wrong, but dst appears to always be set for both rx and tx paths.