Commit a985343ba906 ("vxlan: refactor verification and
application of configuration") introduced a change in the
behaviour of initial MTU setting: earlier, the MTU for a link
created on top of a given lower device, without an initial MTU
specification, was set to the MTU of the lower device minus
headroom as a result of this path in vxlan_dev_configure():
if (!conf->mtu)
dev->mtu = lowerdev->mtu -
(use_ipv6 ? VXLAN6_HEADROOM : VXLAN_HEADROOM);
which is now gone. Now, the initial MTU, in absence of a
configured value, is simply set by ether_setup() to ETH_DATA_LEN
(1500 bytes).
This breaks userspace expectations in case the MTU of
the lower device is higher than 1500 bytes minus headroom.
Restore the previous behaviour by calculating, for a new link,
the MTU from the lower device, if present, and if no value is
explicitly configured.
Reported-by: Junhan Yan <[email protected]>
Fixes: a985343ba906 ("vxlan: refactor verification and application of
configuration")
Signed-off-by: Stefano Brivio <[email protected]>
---
I guess this should be queued up for -stable, back to 4.13.
I'm actually introducing the third occurrence of this calculation (there's
another one in vxlan_config_apply(), and one in vxlan_change_mtu()). I would
anyway fix the userspace breakage first, and then plan on getting rid of several
bits of MTU logic duplication, which spans further than this.
drivers/net/vxlan.c | 3 +++
1 file changed, 3 insertions(+)
diff --git a/drivers/net/vxlan.c b/drivers/net/vxlan.c
index 19b9cc51079e..3a7e36cdf2c7 100644
--- a/drivers/net/vxlan.c
+++ b/drivers/net/vxlan.c
@@ -3085,6 +3085,9 @@ static void vxlan_config_apply(struct net_device *dev,
if (conf->mtu)
dev->mtu = conf->mtu;
+ else if (lowerdev)
+ dev->mtu = lowerdev->mtu - (use_ipv6 ? VXLAN6_HEADROOM :
+ VXLAN_HEADROOM);
vxlan->net = src_net;
}
--
2.9.4
