Simon Horman <horms@xxxxxxxxxx> writes:
> On Fri, Oct 18, 2024 at 12:46:18AM +0000, Hangbin Liu wrote:
>> On Thu, Oct 17, 2024 at 04:47:19PM +0200, Toke Høiland-Jørgensen wrote:
>> > > diff --git a/drivers/net/bonding/bond_main.c b/drivers/net/bonding/bond_main.c
>> > > index f0f76b6ac8be..6887a867fe8b 100644
>> > > --- a/drivers/net/bonding/bond_main.c
>> > > +++ b/drivers/net/bonding/bond_main.c
>> > > @@ -5699,7 +5699,7 @@ static int bond_xdp_set(struct net_device *dev, struct bpf_prog *prog,
>> > > if (dev_xdp_prog_count(slave_dev) > 0) {
>> > > SLAVE_NL_ERR(dev, slave_dev, extack,
>> > > "Slave has XDP program loaded, please unload before enslaving");
>> > > - err = -EOPNOTSUPP;
>> > > + err = -EEXIST;
>> >
>> > Hmm, this has been UAPI since kernel 5.15, so can we really change it
>> > now? What's the purpose of changing it, anyway?
>>
>> I just think it should return EXIST when the error is "Slave has XDP program
>> loaded". No special reason. If all others think we should not change it, I
>> can drop this patch.
>
> Hi Toke,
>
> Could you add some colour to what extent user's might rely on this error code?
>
> Basically I think that if they do then we shouldn't change this.
Well, that's the trouble with UAPI, we don't really know. In libxdp and
xdp-tools we look at the return code to provide a nicer error message,
like:
https://github.com/xdp-project/xdp-tools/blob/master/lib/libxdp/libxdp.c#L615
and as a signal to fall back to loading the programme without a dispatcher:
https://github.com/xdp-project/xdp-tools/blob/master/lib/libxdp/libxdp.c#L1824
Both of these cases would be unaffected (or even improved) by this
patch, so in that sense I don't have a concrete objection, just a
general "userspace may react to this". In other words, my concern is
more of a general "we don't know, so this seems risky". If any of you
have more information about how bonding XDP is generally used, that may
help get a better idea of this?
-Toke
