2018-11-28, 14:27:57 -0800, Roopa Prabhu wrote:
> From: Roopa Prabhu <ro...@cumulusnetworks.com>
> 
> We started very conservative when supporting changelink
> especially because not all attribute changes could be
> tested. This patch opens up a few more attributes for
> changelink. The reason for choosing this set of attributes
> is based on code references for these attributes. I have
> tested TTL changes and did some changelink api testing
> to sanity test the others.
> 
> Signed-off-by: Roopa Prabhu <ro...@cumulusnetworks.com>
> ---
>  drivers/net/vxlan.c | 36 ++++--------------------------------
>  1 file changed, 4 insertions(+), 32 deletions(-)
> 
> diff --git a/drivers/net/vxlan.c b/drivers/net/vxlan.c
> index 9110662..73caa65 100644
> --- a/drivers/net/vxlan.c
> +++ b/drivers/net/vxlan.c
> @@ -3438,11 +3438,8 @@ static int vxlan_nl2conf(struct nlattr *tb[], struct 
> nlattr *data[],
>       if (data[IFLA_VXLAN_TTL])
>               conf->ttl = nla_get_u8(data[IFLA_VXLAN_TTL]);
>  
> -     if (data[IFLA_VXLAN_TTL_INHERIT]) {
> -             if (changelink)
> -                     return -EOPNOTSUPP;
> +     if (data[IFLA_VXLAN_TTL_INHERIT])
>               conf->flags |= VXLAN_F_TTL_INHERIT;
> -     }

This doesn't give us an option to disable TTL_INHERIT after it was
enabled once. Same thing with GBP, GPE, REMCSUM_NOPARTIAL.

> -     if (data[IFLA_VXLAN_GBP]) {
> -             if (changelink)
> -                     return -EOPNOTSUPP;
> +     if (data[IFLA_VXLAN_GBP])
>               conf->flags |= VXLAN_F_GBP;
> -     }
>  
> -     if (data[IFLA_VXLAN_GPE]) {
> -             if (changelink)
> -                     return -EOPNOTSUPP;
> +     if (data[IFLA_VXLAN_GPE])
>               conf->flags |= VXLAN_F_GPE;
> -     }

GPE implies running a different setup function (vxlan_raw_setup() vs
vxlan_ether_setup()), that vxlan_config_apply() only calls for
!changelink. I think this is incomplete.

I think we'd also end up with mixed tunnel types (GPE/!GPE) on the
same socket, I'm not sure how that would work. Normally, they each try
to create a separate socket, and pass the GPE flag on to the
associated vxlan_sock. I suspect that's also a problem with rx
offload.

> -     if (data[IFLA_VXLAN_REMCSUM_NOPARTIAL]) {
> -             if (changelink)
> -                     return -EOPNOTSUPP;
> +     if (data[IFLA_VXLAN_REMCSUM_NOPARTIAL])
>               conf->flags |= VXLAN_F_REMCSUM_NOPARTIAL;
> -     }
>  
>       if (tb[IFLA_MTU]) {
>               if (changelink)
> -- 
> 2.1.4
> 

-- 
Sabrina

Reply via email to