When using rule-based routing to send traffic via VXLAN, a routing
loop may occur. Say you have the following routing setup:
ip rule add from all fwmark 0x2/0x2 lookup 2
ip route add table 2 default via 10.244.2.0 dev vxlan1 onlink
The intention is to route packets with mark 2 through VXLAN, and
this works fine. However, the current vxlan code copies the mark
to the encapsulated packet. Immediately after egress on the VXLAN
interface, the encapsulated packet is routed, with no opportunity
to mangle the encapsulated packet. The mark is copied from the
inner packet to the outer packet, and the same routing rule and
table shown above will apply, resulting in ELOOP.
This patch simply doesn't copy the mark from the encapsulated packet.
I don't intend this code to land as is, but I want to start a
discussion on how to make separate routing of VXLAN inner and
encapsulated traffic easier.
Signed-off-by: Jethro Beekman <[email protected]>
---
drivers/net/vxlan.c | 2 --
1 file changed, 2 deletions(-)
diff --git a/drivers/net/vxlan.c b/drivers/net/vxlan.c
index 3d9bcc9..f9ed1b7 100644
--- a/drivers/net/vxlan.c
+++ b/drivers/net/vxlan.c
@@ -2236,7 +2236,6 @@ static struct rtable *vxlan_get_route(struct vxlan_dev *vxlan, struct net_device
memset(&fl4, 0, sizeof(fl4));
fl4.flowi4_oif = oif;
fl4.flowi4_tos = RT_TOS(tos);
- fl4.flowi4_mark = skb->mark;
fl4.flowi4_proto = IPPROTO_UDP;
fl4.daddr = daddr;
fl4.saddr = *saddr;
@@ -2294,7 +2293,6 @@ static struct dst_entry *vxlan6_get_route(struct vxlan_dev *vxlan,
fl6.daddr = *daddr;
fl6.saddr = *saddr;
fl6.flowlabel = ip6_make_flowinfo(RT_TOS(tos), label);
- fl6.flowi6_mark = skb->mark;
fl6.flowi6_proto = IPPROTO_UDP;
fl6.fl6_dport = dport;
fl6.fl6_sport = sport;
--
2.7.4
On Sun, Sep 29, 2019 at 11:27 PM Jethro Beekman <[email protected]> wrote:
>
> When using rule-based routing to send traffic via VXLAN, a routing
> loop may occur. Say you have the following routing setup:
>
> ip rule add from all fwmark 0x2/0x2 lookup 2
> ip route add table 2 default via 10.244.2.0 dev vxlan1 onlink
>
> The intention is to route packets with mark 2 through VXLAN, and
> this works fine. However, the current vxlan code copies the mark
> to the encapsulated packet. Immediately after egress on the VXLAN
> interface, the encapsulated packet is routed, with no opportunity
> to mangle the encapsulated packet. The mark is copied from the
> inner packet to the outer packet, and the same routing rule and
> table shown above will apply, resulting in ELOOP.
>
> This patch simply doesn't copy the mark from the encapsulated packet.
> I don't intend this code to land as is, but I want to start a
> discussion on how to make separate routing of VXLAN inner and
> encapsulated traffic easier.
yeah, i think the patch as is will break users who use mark to
influence the underlay route lookup.
When you say the mark is copied into the packet, what exactly are you
seeing and where is the copy happening ?
>
> Signed-off-by: Jethro Beekman <[email protected]>
> ---
> drivers/net/vxlan.c | 2 --
> 1 file changed, 2 deletions(-)
>
> diff --git a/drivers/net/vxlan.c b/drivers/net/vxlan.c
> index 3d9bcc9..f9ed1b7 100644
> --- a/drivers/net/vxlan.c
> +++ b/drivers/net/vxlan.c
> @@ -2236,7 +2236,6 @@ static struct rtable *vxlan_get_route(struct vxlan_dev *vxlan, struct net_device
> memset(&fl4, 0, sizeof(fl4));
> fl4.flowi4_oif = oif;
> fl4.flowi4_tos = RT_TOS(tos);
> - fl4.flowi4_mark = skb->mark;
> fl4.flowi4_proto = IPPROTO_UDP;
> fl4.daddr = daddr;
> fl4.saddr = *saddr;
> @@ -2294,7 +2293,6 @@ static struct dst_entry *vxlan6_get_route(struct vxlan_dev *vxlan,
> fl6.daddr = *daddr;
> fl6.saddr = *saddr;
> fl6.flowlabel = ip6_make_flowinfo(RT_TOS(tos), label);
> - fl6.flowi6_mark = skb->mark;
> fl6.flowi6_proto = IPPROTO_UDP;
> fl6.fl6_dport = dport;
> fl6.fl6_sport = sport;
> --
> 2.7.4
>
>
On 2019-10-01 17:30, Roopa Prabhu wrote:
> On Sun, Sep 29, 2019 at 11:27 PM Jethro Beekman <[email protected]> wrote:
>>
>> When using rule-based routing to send traffic via VXLAN, a routing
>> loop may occur. Say you have the following routing setup:
>>
>> ip rule add from all fwmark 0x2/0x2 lookup 2
>> ip route add table 2 default via 10.244.2.0 dev vxlan1 onlink
>>
>> The intention is to route packets with mark 2 through VXLAN, and
>> this works fine. However, the current vxlan code copies the mark
>> to the encapsulated packet. Immediately after egress on the VXLAN
>> interface, the encapsulated packet is routed, with no opportunity
>> to mangle the encapsulated packet. The mark is copied from the
>> inner packet to the outer packet, and the same routing rule and
>> table shown above will apply, resulting in ELOOP.
>>
>> This patch simply doesn't copy the mark from the encapsulated packet.
>> I don't intend this code to land as is, but I want to start a
>> discussion on how to make separate routing of VXLAN inner and
>> encapsulated traffic easier.
>
> yeah, i think the patch as is will break users who use mark to
> influence the underlay route lookup.
> When you say the mark is copied into the packet, what exactly are you
> seeing and where is the copy happening ?
>
Maybe the mark isn't actually copied? At least it's used in the route lookup as shown in the patch.
--
Jethro Beekman | Fortanix
>
>
>>
>> Signed-off-by: Jethro Beekman <[email protected]>
>> ---
>> drivers/net/vxlan.c | 2 --
>> 1 file changed, 2 deletions(-)
>>
>> diff --git a/drivers/net/vxlan.c b/drivers/net/vxlan.c
>> index 3d9bcc9..f9ed1b7 100644
>> --- a/drivers/net/vxlan.c
>> +++ b/drivers/net/vxlan.c
>> @@ -2236,7 +2236,6 @@ static struct rtable *vxlan_get_route(struct vxlan_dev *vxlan, struct net_device
>> memset(&fl4, 0, sizeof(fl4));
>> fl4.flowi4_oif = oif;
>> fl4.flowi4_tos = RT_TOS(tos);
>> - fl4.flowi4_mark = skb->mark;
>> fl4.flowi4_proto = IPPROTO_UDP;
>> fl4.daddr = daddr;
>> fl4.saddr = *saddr;
>> @@ -2294,7 +2293,6 @@ static struct dst_entry *vxlan6_get_route(struct vxlan_dev *vxlan,
>> fl6.daddr = *daddr;
>> fl6.saddr = *saddr;
>> fl6.flowlabel = ip6_make_flowinfo(RT_TOS(tos), label);
>> - fl6.flowi6_mark = skb->mark;
>> fl6.flowi6_proto = IPPROTO_UDP;
>> fl6.fl6_dport = dport;
>> fl6.fl6_sport = sport;
>> --
>> 2.7.4
>>
>>