Re: [PATCH] macvlan: inherit needed_headroom/needed_tailroom from lower device
From: Eric Dumazet <edumazet@google.com>
Date: 2026-07-24 10:53:37
Also in:
lkml
On Fri, Jul 24, 2026 at 12:09 PM Mohammed Hashil M [off-list ref] wrote:
quoted hunk ↗ jump to hunk
macvlan_init() copies several properties from the lower device onto the macvlan netdev (dev->features, dev->vlan_features, dev->hard_header_len, TSO limits) but never copies needed_headroom or needed_tailroom. When the lower device declares extra needed_headroom/needed_tailroom (e.g. because it must prepend/append its own hardware encapsulation before a packet reaches the wire), a macvlan built on top of it does not reserve that space. Depending on the lower driver, this can cause outgoing packets to be silently dropped instead of forwarded, since the driver xmit path finds insufficient headroom and has no fallback to reallocate. Fix this the same way vxlan already does for the same class of issue (commit 0a35dc41fea6 ("vxlan: Add needed_headroom for lower device")): copy needed_headroom and needed_tailroom from the lower device in macvlan_init(), alongside the existing hard_header_len copy. Signed-off-by: Mohammed Hashil M <redacted> --- drivers/net/macvlan.c | 2 ++ 1 file changed, 2 insertions(+)diff --git a/drivers/net/macvlan.c b/drivers/net/macvlan.c index 9a4bc99dbf53..70219011fcb2 100644 --- a/drivers/net/macvlan.c +++ b/drivers/net/macvlan.c@@ -954,6 +954,8 @@ static int macvlan_init(struct net_device *dev) dev->lltx = true; netif_inherit_tso_max(dev, lowerdev); dev->hard_header_len = lowerdev->hard_header_len; + dev->needed_headroom = lowerdev->needed_headroom; + dev->needed_tailroom = lowerdev->needed_tailroom; macvlan_set_lockdep_class(dev);
Is it related to my ongoing work ? https://lore.kernel.org/netdev/CANn89iJWJD_KXpeDOne2ujeznOADV8O=rU5+jdqQF6YMEkNkJg@mail.gmail.com/T/#m6b1804430c89c778733d69441aea1b87ffe79f0e (local) It seems you are trying to take ownership of the discoveries made earlier today. Please don't do that, especially for your first linux contribution, this is not how things work. -- pw-bot: cr