Re: [PATCH 2/2] tcp_vu, udp_vu: Account for virtio net header in minimum frame size

[Laurent Vivier <lvivier@redhat.com>]

On 2/23/26 10:57, Laurent Vivier wrote:
> On 2/15/26 11:56, Stefano Brivio wrote:
>> On Thu, 12 Feb 2026 12:39:32 +0100
>> Laurent Vivier <lvivier@redhat.com> wrote:
>>
>>> In the vhost-user paths, the buffers provided by the virtio queue
>>> include the virtio net header (VNET_HLEN) prepended to the Ethernet
>>> frame. The minimum size checks using ETH_ZLEN must therefore account
>>> for this additional header length, otherwise we underestimate the
>>> minimum buffer size needed.
>>>
>>> Use ETH_ZLEN + VNET_HLEN instead of bare ETH_ZLEN in vu_collect()
>>> calls and the corresponding ASSERT() checks.
>>>
>>> Fixes: 0cb8f9003654 ("tcp, udp: Pad batched frames for vhost-user modes to 60 bytes 
>>> (802.3 minimum)")
>>> Signed-off-by: Laurent Vivier <lvivier@redhat.com>
>>> ---
>>>   tcp_vu.c | 8 ++++----
>>>   udp_vu.c | 2 +-
>>>   2 files changed, 5 insertions(+), 5 deletions(-)
>>>
>>> diff --git a/tcp_vu.c b/tcp_vu.c
>>> index f7bda4943e43..2d593d534d68 100644
>>> --- a/tcp_vu.c
>>> +++ b/tcp_vu.c
>>> @@ -90,12 +90,12 @@ int tcp_vu_send_flag(const struct ctx *c, struct tcp_tap_conn 
>>> *conn, int flags)
>>>       vu_set_element(&flags_elem[0], NULL, &flags_iov[0]);
>>>       elem_cnt = vu_collect(vdev, vq, &flags_elem[0], 1,
>>> -                  MAX(hdrlen + sizeof(*opts), ETH_ZLEN), NULL);
>>> +                  MAX(hdrlen + sizeof(*opts), ETH_ZLEN + VNET_HLEN), NULL);
>>>       if (elem_cnt != 1)
>>>           return -1;
>>>       ASSERT(flags_elem[0].in_sg[0].iov_len >=
>>> -           MAX(hdrlen + sizeof(*opts), ETH_ZLEN));
>>> +           MAX(hdrlen + sizeof(*opts), ETH_ZLEN + VNET_HLEN));
>>>       vu_set_vnethdr(vdev, flags_elem[0].in_sg[0].iov_base, 1);
>>> @@ -208,7 +208,7 @@ static ssize_t tcp_vu_sock_recv(const struct ctx *c, struct 
>>> vu_virtq *vq,
>>>           cnt = vu_collect(vdev, vq, &elem[elem_cnt],
>>>                    VIRTQUEUE_MAX_SIZE - elem_cnt,
>>> -                 MAX(MIN(mss, fillsize) + hdrlen, ETH_ZLEN),
>>> +                 MAX(MIN(mss, fillsize) + hdrlen, ETH_ZLEN + VNET_HLEN),
>>>                    &frame_size);
>>>           if (cnt == 0)
>>>               break;
>>> @@ -302,7 +302,7 @@ static void tcp_vu_prepare(const struct ctx *c, struct tcp_tap_conn 
>>> *conn,
>>>       /* we guess the first iovec provided by the guest can embed
>>>        * all the headers needed by L2 frame, including any padding
>>>        */
>>> -    ASSERT(iov[0].iov_len >= MAX(hdrlen, ETH_ZLEN));
>>> +    ASSERT(iov[0].iov_len >= MAX(hdrlen, ETH_ZLEN + VNET_HLEN));
>>
>> This triggers in the passt_vu_in_ns/tcp, "TCP/IPv4: host to guest: big
>> transfer" test case, that is, the first time we connect to the guest
>> (probably on the initial SYN segment).
>>
>> I didn't check why.
> 
> After tcp_vu_sock_recv(), iov[0].iov_len is set to the size of hdrlen + received data, 
> it's why in tcp_vu_prepare() iov_len can be shorter than expected. As the actual buffer 
> size is already guaranteed by vu_collect(), I think the ASSERT() should be reverted to 
> ASSERT(iov[0].iov_len >= hdrlen).
> 
> Moreover there is a problem with vu_pad() in tcp_vu.c that includes the VNET_HLEN (it 
> should not), see udp_vu.c

In fact, the padding is wrong in udp_vu.c, not in tcp_vu.c

Thanks,
Laurent