On Fri, Dec 05, 2025 at 01:51:53AM +0100, Stefano Brivio wrote:
> IEEE 802.3 requires a minimum frame payload of 46 bytes, without a
> 802.1Q tag. Add padding for the simple tap_send_single() case using
> a zero-filled 60-byte buffer and copying data to it if needed.
> 
> In theory, we could add a further element in the iovec array, say:
> 
> 	uint8_t padding[ETH_ZLEN] = { 0 };
> 	struct iovec iov[3];
> 
> 	...
> 
> 		if (l2len < ETH_ZLEN) {
> 			iov[iovcnt].iov_base = (void *)padding;
> 			iov[iovcnt].iov_len = ETH_ZLEN - l2len;
> 			iovcnt++;
> 		}
> 
> and avoid a copy, but that would substantially complicate the
> vhost-user path, and it's questionable whether passing a reference
> to a further buffer actually causes lower overhead than the simple
> copy.
> 
> Link: https://bugs.passt.top/show_bug.cgi?id=166
> Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
> Reviewed-by: Laurent Vivier <lvivier@redhat.com>

Reviewed-by: David Gibson <david@gibson.dropbear.id.au>

> ---
>  tap.c | 11 ++++++++++-
>  1 file changed, 10 insertions(+), 1 deletion(-)
> 
> diff --git a/tap.c b/tap.c
> index 44b0644..e3ea61c 100644
> --- a/tap.c
> +++ b/tap.c
> @@ -130,9 +130,18 @@ unsigned long tap_l2_max_len(const struct ctx *c)
>   */
>  void tap_send_single(const struct ctx *c, const void *data, size_t l2len)
>  {
> -	uint32_t vnet_len = htonl(l2len);
> +	uint8_t padded[ETH_ZLEN] = { 0 };
>  	struct iovec iov[2];
>  	size_t iovcnt = 0;
> +	uint32_t vnet_len;
> +
> +	if (l2len < ETH_ZLEN) {
> +		memcpy(padded, data, l2len);
> +		data = padded;
> +		l2len = ETH_ZLEN;
> +	}
> +
> +	vnet_len = htonl(l2len);
>  
>  	switch (c->mode) {
>  	case MODE_PASST:
> -- 
> 2.43.0
> 

-- 
David Gibson (he or they)	| I'll have my music baroque, and my code
david AT gibson.dropbear.id.au	| minimalist, thank you, not the other way
				| around.
http://www.ozlabs.org/~dgibson