| Message ID | 20131113182133.GA8760@amd64.fatal.se |
|---|---|
| State | RFC, archived |
| Delegated to: | David Miller |
| Headers | show |
diff --git a/net/ipv4/ip_tunnel.c b/net/ipv4/ip_tunnel.c index 63a6d6d..1dc4e41 100644 --- a/net/ipv4/ip_tunnel.c +++ b/net/ipv4/ip_tunnel.c @@ -732,8 +732,14 @@ int ip_tunnel_ioctl(struct net_device *dev, struct ip_tunnel_parm *p, int cmd) t = ip_tunnel_find(itn, p, itn->fb_tunnel_dev->type); - if (!t && (cmd == SIOCADDTUNNEL)) - t = ip_tunnel_create(net, itn, p); + if (cmd == SIOCADDTUNNEL) { + if (!t) { + t = ip_tunnel_create(net, itn, p); + } else { + err = -EEXIST; + break; + } + } if (dev != itn->fb_tunnel_dev && cmd == SIOCCHGTUNNEL) { if (t != NULL) {
After patch: $ sudo ip tunnel add test1 mode ipip remote 1.2.3.4 $ sudo ip tunnel add test2 mode ipip remote 1.2.3.4 add tunnel "tunl0" failed: File exists Before the patch, there would be no error and "test2" (silently) not added. Originally reported at http://bugs.debian.org/508450 The originally reported problem with sit tunnel seems to have been resolved since then: $ sudo ip tun add test3 mode sit add tunnel "sit0" failed: No buffer space available The problem still exists for (atleast) ipip tunnels though. Reported-by: martin f krafft <madduck@debian.org> Signed-off-by: Andreas Henriksson <andreas@fatal.se> --- net/ipv4/ip_tunnel.c | 10 ++++++++-- 1 file changed, 8 insertions(+), 2 deletions(-) Maybe also sit tunnels should explicitly use EEXISTS instead of ENOBUFS? Maybe ipip tunnel code should be refactored to be similar to sit tunnel addition code? Which other tunnel types should be checked for how they behave in similar situations? Other comments?