net: ipv4: Make "ip route get" match iif lo rules again.
authorLorenzo Colitti <lorenzo@google.com>
Thu, 11 Jan 2018 09:36:26 +0000 (18:36 +0900)
committerGreg Kroah-Hartman <gregkh@linuxfoundation.org>
Wed, 31 Jan 2018 13:03:49 +0000 (14:03 +0100)
commitc2fd0b217068d6ef0b71bc5fbfd3c8f9f48c42eb
treef03972f5ff73dcff59e129bd6efb890d33aefeb6
parented10b9affb3af663d2cdb8a1e9134fb0a6358ab0
net: ipv4: Make "ip route get" match iif lo rules again.

[ Upstream commit 6503a30440962f1e1ccb8868816b4e18201218d4 ]

Commit 3765d35ed8b9 ("net: ipv4: Convert inet_rtm_getroute to rcu
versions of route lookup") broke "ip route get" in the presence
of rules that specify iif lo.

Host-originated traffic always has iif lo, because
ip_route_output_key_hash and ip6_route_output_flags set the flow
iif to LOOPBACK_IFINDEX. Thus, putting "iif lo" in an ip rule is a
convenient way to select only originated traffic and not forwarded
traffic.

inet_rtm_getroute used to match these rules correctly because
even though it sets the flow iif to 0, it called
ip_route_output_key which overwrites iif with LOOPBACK_IFINDEX.
But now that it calls ip_route_output_key_hash_rcu, the ifindex
will remain 0 and not match the iif lo in the rule. As a result,
"ip route get" will return ENETUNREACH.

Fixes: 3765d35ed8b9 ("net: ipv4: Convert inet_rtm_getroute to rcu versions of route lookup")
Tested: https://android.googlesource.com/kernel/tests/+/master/net/test/multinetwork_test.py passes again
Signed-off-by: Lorenzo Colitti <lorenzo@google.com>
Acked-by: David Ahern <dsahern@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
net/ipv4/route.c