[ovs-dev] [PATCH v4] gre: Rename fallback devices to avoid udev's interference

Gregory Rose gvrose8192 at gmail.com
Mon Sep 17 21:28:05 UTC 2018


On 9/17/2018 1:40 PM, Yifeng Sun wrote:
> Hi Greg,
>
> I checked that the 'else' part in the above patch is actually
> executed. So it should be necessary to keep them.
>
> Thanks,
> Yifeng

That's interesting but I expect that it is only executed once during 
creation of the FB device.  Does it continue
to get invoked if you keep adding new tunnels?  Can you try adding new 
ip gre tunnels and see if we would ever actually get to 100 of them? 
Enough to create a 3 digit trailing number?

The reason I ask is that the kernel code is checking IFNAMSIZ - 3 and I 
think that check is sufficient because
only the FB devices, of which there should only be one, will cause that 
block to be executed.  I'm not sure
though!  If you could try that out I'd appreciate it but I do believe 
that for all non-FB devices cases the name
of the tunnel will be in parms->name.  Confirming that would tell us 
whether you need to make a patch
for upstream or not.

Thanks Yifeng!

- Greg

>
> On Mon, Sep 17, 2018 at 9:51 AM Yifeng Sun <pkusunyifeng at gmail.com 
> <mailto:pkusunyifeng at gmail.com>> wrote:
>
>     Hi Greg,
>
>     Sure, thanks for the review. I will work on it.
>
>     Yifeng
>
>     On Mon, Sep 17, 2018 at 8:49 AM Gregory Rose <gvrose8192 at gmail.com
>     <mailto:gvrose8192 at gmail.com>> wrote:
>
>         On 9/15/2018 9:24 PM, Yifeng Sun wrote:
>         > On certain kernel versions, when openvswitch kernel module
>         creates
>         > a gre0 interface, the kernel’s gre module will jump out and
>         compete
>         > to control the gre0 interface. This will cause the failure of
>         > openvswitch kernel module loading.
>         >
>         > This fix renames fallback devices by adding a prefix "ovs-".
>         >
>         > Signed-off-by: Yifeng Sun <pkusunyifeng at gmail.com
>         <mailto:pkusunyifeng at gmail.com>>
>         > VMware Issue: #2162866
>         > ---
>         > Please backport this patch to upstream OVS down to 2.9, thanks.
>         >
>         > v1->v2: Added sanity check for device names, thanks Justin.
>         > v2->v3: Fix an indent error.
>         > v3->v4: Fix by code review, also fix a potenial bug.
>         >
>         >   datapath/linux/compat/ip6_gre.c    | 5 +++--
>         >   datapath/linux/compat/ip6_tunnel.c | 6 +++---
>         >   datapath/linux/compat/ip_gre.c     | 2 +-
>         >   datapath/linux/compat/ip_tunnel.c  | 7 +++----
>         >   4 files changed, 10 insertions(+), 10 deletions(-)
>         >
>         > diff --git a/datapath/linux/compat/ip6_gre.c
>         b/datapath/linux/compat/ip6_gre.c
>         > index 00dbefc9b099..182785273c6f 100644
>         > --- a/datapath/linux/compat/ip6_gre.c
>         > +++ b/datapath/linux/compat/ip6_gre.c
>         > @@ -377,7 +377,7 @@ static struct ip6_tnl
>         *ip6gre_tunnel_locate(struct net *net,
>         >       if (parms->name[0])
>         >               strlcpy(name, parms->name, IFNAMSIZ);
>         >       else
>         > -             strcpy(name, "ip6gre%d");
>         > +             strlcpy(name, "ovs-ip6gre%d", IFNAMSIZ);
>         >
>         >       dev = alloc_netdev(sizeof(*t), name, NET_NAME_UNKNOWN,
>         >                          ip6gre_tunnel_setup);
>         > @@ -1712,7 +1712,8 @@ static int __net_init
>         ip6gre_init_net(struct net *net)
>         >       struct ip6gre_net *ign = net_generic(net, ip6gre_net_id);
>         >       int err;
>         >
>         > -     ign->fb_tunnel_dev = alloc_netdev(sizeof(struct
>         ip6_tnl), "ip6gre0",
>         > +     ign->fb_tunnel_dev = alloc_netdev(sizeof(struct ip6_tnl),
>         > +  "ovs-ip6gre0",
>         >  NET_NAME_UNKNOWN,
>         >  ip6gre_tunnel_setup);
>         >       if (!ign->fb_tunnel_dev) {
>         > diff --git a/datapath/linux/compat/ip6_tunnel.c
>         b/datapath/linux/compat/ip6_tunnel.c
>         > index 56fd8b4dd342..9f4bae7dd3d1 100644
>         > --- a/datapath/linux/compat/ip6_tunnel.c
>         > +++ b/datapath/linux/compat/ip6_tunnel.c
>         > @@ -355,7 +355,7 @@ static struct ip6_tnl
>         *ip6_tnl_create(struct net *net, struct __ip6_tnl_parm *p)
>         >       if (p->name[0])
>         >               strlcpy(name, p->name, IFNAMSIZ);
>         >       else
>         > -             sprintf(name, "ip6tnl%%d");
>         > +             strlcpy(name, "ovs-ip6tnl%d", IFNAMSIZ);
>         >
>         >       dev = alloc_netdev(sizeof(*t), name, NET_NAME_UNKNOWN,
>         >                          ip6_tnl_dev_setup);
>         > @@ -1410,7 +1410,7 @@ ip6_tnl_parm_to_user(struct
>         ip6_tnl_parm *u, const struct __ip6_tnl_parm *p)
>         >    *     %SIOCCHGTUNNEL: change tunnel parameters to those given
>         >    *     %SIOCDELTUNNEL: delete tunnel
>         >    *
>         > - *   The fallback device "ip6tnl0", created during module
>         > + *   The fallback device "ovs-ip6tnl0", created during module
>         >    *   initialization, can be used for creating other tunnel
>         devices.
>         >    *
>         >    * Return:
>         > @@ -2093,7 +2093,7 @@ static int __net_init
>         ip6_tnl_init_net(struct net *net)
>         >       ip6n->tnls[1] = ip6n->tnls_r_l;
>         >
>         >       err = -ENOMEM;
>         > -     ip6n->fb_tnl_dev = alloc_netdev(sizeof(struct
>         ip6_tnl), "ip6tnl0",
>         > +     ip6n->fb_tnl_dev = alloc_netdev(sizeof(struct
>         ip6_tnl), "ovs-ip6tnl0",
>         >  NET_NAME_UNKNOWN, ip6_tnl_dev_setup);
>         >
>         >       if (!ip6n->fb_tnl_dev)
>         > diff --git a/datapath/linux/compat/ip_gre.c
>         b/datapath/linux/compat/ip_gre.c
>         > index 05132ba9494a..b7322c58e420 100644
>         > --- a/datapath/linux/compat/ip_gre.c
>         > +++ b/datapath/linux/compat/ip_gre.c
>         > @@ -1463,7 +1463,7 @@ static struct pernet_operations
>         erspan_net_ops = {
>         >
>         >   static int __net_init ipgre_tap_init_net(struct net *net)
>         >   {
>         > -     return ip_tunnel_init_net(net, gre_tap_net_id,
>         &ipgre_tap_ops, "gretap0");
>         > +     return ip_tunnel_init_net(net, gre_tap_net_id,
>         &ipgre_tap_ops, "ovs-gretap0");
>         >   }
>         >
>         >   static void __net_exit ipgre_tap_exit_net(struct net *net)
>         > diff --git a/datapath/linux/compat/ip_tunnel.c
>         b/datapath/linux/compat/ip_tunnel.c
>         > index d16e60fbfef0..ae5d5a058663 100644
>         > --- a/datapath/linux/compat/ip_tunnel.c
>         > +++ b/datapath/linux/compat/ip_tunnel.c
>         > @@ -130,12 +130,11 @@ static struct net_device
>         *__ip_tunnel_create(struct net *net,
>         >       if (parms->name[0])
>         >               strlcpy(name, parms->name, IFNAMSIZ);
>         >       else {
>         > -             if (strlen(ops->kind) > (IFNAMSIZ - 3)) {
>         > +             err = snprintf(name, IFNAMSIZ - 3,
>         "ovs-%s%%d", ops->kind);
>         > +             if (err >= IFNAMSIZ - 3)
>         >                       err = -E2BIG;
>         > +             if (err < 0)
>         >                       goto failed;
>         > -             }
>         > -             strlcpy(name, ops->kind, IFNAMSIZ);
>         > -             strncat(name, "%d", 2);
>         >       }
>         >
>         >       ASSERT_RTNL();
>
>         Thanks Yifeng for all your work on this.
>
>         This looks OK but I'm not sure the additional protections
>         against a
>         device name being too long are
>         really needed.  I think in all cases the parms->name will be
>         filled out
>         since OVS is the only user.
>         Have you tested whether the else code path will ever be entered?
>         Sometimes kernel code gets
>         into our backports even though it would never be executed
>         within the
>         context of OVS.  It would
>         be good to check this.
>
>         Thanks,
>
>         - Greg
>



More information about the dev mailing list