Re: [PATCH net-next v2 3/9] tun: correct drop statistics in tun_put_user

[Willem de Bruijn]

Jon Kohler wrote:
> 
> 
> > On Nov 28, 2025, at 10:07 PM, Willem de Bruijn <willemdebruijn.kernel@xxxxxxxxx> wrote:
> > 
> > Jon Kohler wrote:
> >> Fold kfree_skb and consume_skb for tun_put_user into tun_put_user and
> >> rework kfree_skb to take a drop reason. Add drop reason to all drop
> >> sites and ensure that all failing paths properly increment drop
> >> counter.
> >> 
> >> Signed-off-by: Jon Kohler <jon@xxxxxxxxxxx>
> >> ---
> >> drivers/net/tun.c | 51 +++++++++++++++++++++++++++++++----------------
> >> 1 file changed, 34 insertions(+), 17 deletions(-)
> >> 
> >> diff --git a/drivers/net/tun.c b/drivers/net/tun.c
> >> index 68ad46ab04a4..e0f5e1fe4bd0 100644
> >> --- a/drivers/net/tun.c
> >> +++ b/drivers/net/tun.c
> >> @@ -2035,6 +2035,7 @@ static ssize_t tun_put_user(struct tun_struct *tun,
> >>     struct sk_buff *skb,
> >>     struct iov_iter *iter)
> >> {
> >> + enum skb_drop_reason drop_reason = SKB_DROP_REASON_NOT_SPECIFIED;
> >> struct tun_pi pi = { 0, skb->protocol };
> >> ssize_t total;
> >> int vlan_offset = 0;
> >> @@ -2051,8 +2052,11 @@ static ssize_t tun_put_user(struct tun_struct *tun,
> >> total = skb->len + vlan_hlen + vnet_hdr_sz;
> >> 
> >> if (!(tun->flags & IFF_NO_PI)) {
> >> - if (iov_iter_count(iter) < sizeof(pi))
> >> - return -EINVAL;
> >> + if (iov_iter_count(iter) < sizeof(pi)) {
> >> + ret = -EINVAL;
> >> + drop_reason = SKB_DROP_REASON_PKT_TOO_SMALL;
> > 
> > PI counts as SKB_DROP_REASON_DEV_HDR?
> 
> Are you saying I should change this use case to DEV_HDR?
> 
> This one seemed like a pretty straight forward “It’s too small” case,
> no? Or am I misreading into what you’re saying here?
> 
> Happy to take a suggestion if I’ve got the drop reason wired
> wrong (or if we need to cook up a brand new drop reason for any of
> these)

I agree that it's a clear case of the buffer being too small. But I
consider PI not part of the packet itself, but bad device headers.
It's borderline nitpicking. With that context, pick which you see fits
best.