On Fri, 22 Feb 2019 04:58:06 +0000
Moshe Shemesh <mo...@mellanox.com> wrote:
> On 2/21/2019 7:54 PM, Stephen Hemminger wrote:
> > The netlink API for IFLA_VF_VLAN_LIST allows multiple VLAN tags to
> > be passed (and the message was validated) but only the first VLAN
> > tag was being passed to the device. Change to iterate over each tag
> > received.
> >
> > Fixes: 79aab093a0b5 ("net: Update API for VF vlan protocol 802.1ad support")
> > Signed-off-by: Stephen Hemminger <step...@networkplumber.org>
> > ---
> > net/core/rtnetlink.c | 19 ++++++++++---------
> > 1 file changed, 10 insertions(+), 9 deletions(-)
> >
> > diff --git a/net/core/rtnetlink.c b/net/core/rtnetlink.c
> > index a51cab95ba64..3a9ec988ae21 100644
> > --- a/net/core/rtnetlink.c
> > +++ b/net/core/rtnetlink.c
> > @@ -2207,11 +2207,10 @@ static int do_setvfinfo(struct net_device *dev,
> > struct nlattr **tb)
> > if (tb[IFLA_VF_VLAN_LIST]) {
> > struct ifla_vf_vlan_info *ivvl[MAX_VLAN_LIST_LEN];
> > struct nlattr *attr;
> > - int rem, len = 0;
> > + int i, rem, len = 0;
> >
> > - err = -EOPNOTSUPP;
> > if (!ops->ndo_set_vf_vlan)
> > - return err;
> > + return -EOPNOTSUPP;
> >
> > nla_for_each_nested(attr, tb[IFLA_VF_VLAN_LIST], rem) {
> > if (nla_type(attr) != IFLA_VF_VLAN_INFO ||
> > @@ -2224,13 +2223,15 @@ static int do_setvfinfo(struct net_device *dev,
> > struct nlattr **tb)
> >
> > len++;
> > }
> > - if (len == 0)
> > - return -EINVAL;
> >
> > - err = ops->ndo_set_vf_vlan(dev, ivvl[0]->vf, ivvl[0]->vlan,
> > - ivvl[0]->qos, ivvl[0]->vlan_proto);
> > - if (err < 0)
> > - return err;
> > + err = -EINVAL; /* empty list error */
> > + for (i = 0; i < len; i++) {
> > + err = ops->ndo_set_vf_vlan(dev, ivvl[i]->vf,
> > + ivvl[i]->vlan, ivvl[i]->qos,
> > + ivvl[i]->vlan_proto);
>
> Doing that each vlan will just overwrite the vf vlan configuration set
> by its preceding one.
> Note #define MAX_VLAN_LIST_LEN 1
> The point here was that I had to add the rtnl interface to set vf vlan
> with option to set vf vlan protocol. While doing that I was asked to add
> option to get a list of vlans from user to support QinQ, so once it will
> be needed there won't be a need to add another rtnl interface.
>
> The driver which will support setting double vlan or more per vf will
> change MAX_VLAN_LIST_LEN and change the ndo function.
>
OK, then the original code kind of makes sense now. Maybe the whole loop
could be eliminated.
I expected that the vlan list was for allowing multiple vlan's to go to
a single VF but it is really a future (unlikely to ever be implemented)
support for QinQ.
