Re: [PATCH net-next 6/6] inet: use xa_array iterator to implement inet_dump_ifaddr()

[Jakub Kicinski <kuba@kernel.org>]

On Thu, 29 Feb 2024 16:50:45 +0100 Eric Dumazet wrote:
> > > You basically only want to return skb->len when message has
> > > overflown, so the somewhat idiomatic way to do this is:
> > >
> > >         err = (err == -EMSGSIZE) ? skb->len : err;  
> 
> This would set err to zero if skb is empty at this point.
> 
> I guess a more correct action would be:
> 
> if (err == -EMSGSIZE && likely(skb->len))
>     err = skb->len;

Ugh, fair point.
We should probably move the EMSGSIZE handling to the core, this is
getting too complicated for average humans.. Like this?

diff --git a/net/netlink/af_netlink.c b/net/netlink/af_netlink.c
index 765921cf1194..ce27003b90a8 100644
--- a/net/netlink/af_netlink.c
+++ b/net/netlink/af_netlink.c
@@ -2264,6 +2264,8 @@ static int netlink_dump(struct sock *sk, bool lock_taken)
 		if (extra_mutex)
 			mutex_lock(extra_mutex);
 		nlk->dump_done_errno = cb->dump(skb, cb);
+		if (nlk->dump_done_errno == -EMSGSIZE && skb->len)
+			nlk->dump_done_errno = skb->len;
 		if (extra_mutex)
 			mutex_unlock(extra_mutex);

> > >
> > > Assuming err can't be set to some weird positive value.
> > >
> > > IDK if you want to do this in future patches or it's risky, but I have
> > > the itch to tell you every time I see a conversion which doesn't follow
> > > this pattern :)  
> >
> > This totally makes sense.
> >
> > I will send a followup patch to fix all these in one go, if this is ok
> > with you ?

Definitely not a blocker