The below commit adds support for:
ip link set dummy0 down ip link set dummy0 master bond0 up
but breaks the opposite:
ip link set dummy0 up ip link set dummy0 master bond0 down
Let's add a workaround to have both commands working.
Cc: stable@vger.kernel.org Fixes: a4abfa627c38 ("net: rtnetlink: Enslave device before bringing it up") Signed-off-by: Nicolas Dichtel nicolas.dichtel@6wind.com Acked-by: Phil Sutter phil@nwl.cc Reviewed-by: David Ahern dsahern@kernel.org --- net/core/rtnetlink.c | 8 ++++++++ 1 file changed, 8 insertions(+)
diff --git a/net/core/rtnetlink.c b/net/core/rtnetlink.c index e8431c6c8490..dd79693c2d91 100644 --- a/net/core/rtnetlink.c +++ b/net/core/rtnetlink.c @@ -2905,6 +2905,14 @@ static int do_setlink(const struct sk_buff *skb, call_netdevice_notifiers(NETDEV_CHANGEADDR, dev); }
+ /* Backward compat: enable to set interface down before enslaving it */ + if (!(ifm->ifi_flags & IFF_UP) && ifm->ifi_change & IFF_UP) { + err = dev_change_flags(dev, rtnl_dev_combine_flags(dev, ifm), + extack); + if (err < 0) + goto errout; + } + if (tb[IFLA_MASTER]) { err = do_set_master(dev, nla_get_u32(tb[IFLA_MASTER]), extack); if (err)
Thu, Jan 04, 2024 at 05:42:59PM CET, nicolas.dichtel@6wind.com wrote:
The below commit adds support for:
ip link set dummy0 down ip link set dummy0 master bond0 up
but breaks the opposite:
ip link set dummy0 up ip link set dummy0 master bond0 down
It is a bit weird to see these 2 and assume some ordering. The first one assumes: dummy0 master bond 0, dummy0 up The second one assumes: dummy0 down, dummy0 master bond 0 But why?
What is the practival reason for a4abfa627c38 existence? I mean, bond/team bring up the device themselfs when needed. Phil? Wouldn't simple revert do better job here?
Let's add a workaround to have both commands working.
Cc: stable@vger.kernel.org Fixes: a4abfa627c38 ("net: rtnetlink: Enslave device before bringing it up") Signed-off-by: Nicolas Dichtel nicolas.dichtel@6wind.com Acked-by: Phil Sutter phil@nwl.cc Reviewed-by: David Ahern dsahern@kernel.org
net/core/rtnetlink.c | 8 ++++++++ 1 file changed, 8 insertions(+)
diff --git a/net/core/rtnetlink.c b/net/core/rtnetlink.c index e8431c6c8490..dd79693c2d91 100644 --- a/net/core/rtnetlink.c +++ b/net/core/rtnetlink.c @@ -2905,6 +2905,14 @@ static int do_setlink(const struct sk_buff *skb, call_netdevice_notifiers(NETDEV_CHANGEADDR, dev); }
- /* Backward compat: enable to set interface down before enslaving it */
- if (!(ifm->ifi_flags & IFF_UP) && ifm->ifi_change & IFF_UP) {
err = dev_change_flags(dev, rtnl_dev_combine_flags(dev, ifm),extack);if (err < 0)goto errout;- }
- if (tb[IFLA_MASTER]) { err = do_set_master(dev, nla_get_u32(tb[IFLA_MASTER]), extack); if (err)
-- 2.39.2
Le 05/01/2024 à 12:59, Jiri Pirko a écrit :
Thu, Jan 04, 2024 at 05:42:59PM CET, nicolas.dichtel@6wind.com wrote:
The below commit adds support for:
ip link set dummy0 down ip link set dummy0 master bond0 up
but breaks the opposite:
ip link set dummy0 up ip link set dummy0 master bond0 down
It is a bit weird to see these 2 and assume some ordering. The first one assumes: dummy0 master bond 0, dummy0 up The second one assumes: dummy0 down, dummy0 master bond 0 But why?
What is the practival reason for a4abfa627c38 existence? I mean, bond/team bring up the device themselfs when needed. Phil? Wouldn't simple revert do better job here?
Yeah, I assumed there was a good reason, but you're right.
Let's add a workaround to have both commands working.
Cc: stable@vger.kernel.org Fixes: a4abfa627c38 ("net: rtnetlink: Enslave device before bringing it up") Signed-off-by: Nicolas Dichtel nicolas.dichtel@6wind.com Acked-by: Phil Sutter phil@nwl.cc Reviewed-by: David Ahern dsahern@kernel.org
net/core/rtnetlink.c | 8 ++++++++ 1 file changed, 8 insertions(+)
diff --git a/net/core/rtnetlink.c b/net/core/rtnetlink.c index e8431c6c8490..dd79693c2d91 100644 --- a/net/core/rtnetlink.c +++ b/net/core/rtnetlink.c @@ -2905,6 +2905,14 @@ static int do_setlink(const struct sk_buff *skb, call_netdevice_notifiers(NETDEV_CHANGEADDR, dev); }
- /* Backward compat: enable to set interface down before enslaving it */
- if (!(ifm->ifi_flags & IFF_UP) && ifm->ifi_change & IFF_UP) {
err = dev_change_flags(dev, rtnl_dev_combine_flags(dev, ifm),extack);if (err < 0)goto errout;- }
- if (tb[IFLA_MASTER]) { err = do_set_master(dev, nla_get_u32(tb[IFLA_MASTER]), extack); if (err)
-- 2.39.2
Hi,
On Fri, Jan 05, 2024 at 12:59:24PM +0100, Jiri Pirko wrote:
Thu, Jan 04, 2024 at 05:42:59PM CET, nicolas.dichtel@6wind.com wrote:
The below commit adds support for:
ip link set dummy0 down ip link set dummy0 master bond0 up
but breaks the opposite:
ip link set dummy0 up ip link set dummy0 master bond0 down
It is a bit weird to see these 2 and assume some ordering. The first one assumes: dummy0 master bond 0, dummy0 up The second one assumes: dummy0 down, dummy0 master bond 0 But why?
What is the practival reason for a4abfa627c38 existence? I mean, bond/team bring up the device themselfs when needed. Phil? Wouldn't simple revert do better job here?
Ah, I wasn't aware bond master manipulates slaves' links itself and thus treated all types' link master setting the same by setting the slave up afterwards. This is basically what a4abfa627c38 is good for: Enabling 'ip link set X master Y up' regardless of Y's link type.
If setting a bond slave up manually is not recommended, the easiest solution is probbaly indeed to revert a4abfa627c38 and live with the quirk in bond driver.
Cheers, Phil
Sat, Jan 06, 2024 at 04:32:12AM CET, phil@nwl.cc wrote:
Hi,
On Fri, Jan 05, 2024 at 12:59:24PM +0100, Jiri Pirko wrote:
Thu, Jan 04, 2024 at 05:42:59PM CET, nicolas.dichtel@6wind.com wrote:
The below commit adds support for:
ip link set dummy0 down ip link set dummy0 master bond0 up
but breaks the opposite:
ip link set dummy0 up ip link set dummy0 master bond0 down
It is a bit weird to see these 2 and assume some ordering. The first one assumes: dummy0 master bond 0, dummy0 up The second one assumes: dummy0 down, dummy0 master bond 0 But why?
What is the practival reason for a4abfa627c38 existence? I mean, bond/team bring up the device themselfs when needed. Phil? Wouldn't simple revert do better job here?
Ah, I wasn't aware bond master manipulates slaves' links itself and thus treated all types' link master setting the same by setting the slave up afterwards. This is basically what a4abfa627c38 is good for: Enabling 'ip link set X master Y up' regardless of Y's link type.
If setting a bond slave up manually is not recommended, the easiest solution is probbaly indeed to revert a4abfa627c38 and live with the quirk in bond driver.
Okay, lets revert then.
Cheers, Phil
linux-stable-mirror@lists.linaro.org
-
Jiri Pirko -
Nicolas Dichtel -
Phil Sutter