Date: Thu, 10 Aug 2017 09:12:46 -0700 (PDT)
Message-Id: <20170810.091246.434591449474758645.davem@davemloft.net>
To: otubo@redhat.com
Cc: netdev@vger.kernel.org, linux-kernel@vger.kernel.org,
        devel@linuxdriverproject.org, sthemmin@microsoft.com,
        haiyangz@microsoft.com, kys@microsoft.com
Subject: Re: [PATCH] hv_set_ifconfig.sh double check before setting ip
From: David Miller <davem@davemloft.net>
In-Reply-To: <0b0a89e2-250e-69d9-9b24-1cf49e9c1d26@redhat.com>
References: <20170808.211150.2070475124549651350.davem@davemloft.net>
        <494db0c8-8c31-f2ce-17fc-7b96b74f0b28@redhat.com>
        <0b0a89e2-250e-69d9-9b24-1cf49e9c1d26@redhat.com>
Mime-Version: 1.0
Content-Type: Text/Plain; charset=us-ascii
Content-Transfer-Encoding: 7bit
Sender: linux-kernel-owner@vger.kernel.org
Content-Length: 1710
Lines: 43

From: Eduardo Otubo <otubo@redhat.com>
Date: Thu, 10 Aug 2017 09:40:27 +0200

> On 08/09/2017 11:02 AM, Eduardo Otubo wrote:
>> On 08/09/2017 06:11 AM, David Miller wrote:
>>> From: Eduardo Otubo <otubo@redhat.com>
>>> Date: Tue,  8 Aug 2017 15:53:45 +0200
>>>
>>>> This patch fixes the behavior of the hv_set_ifconfig script when
>>>> setting
>>>> the interface ip. Sometimes the interface has already been configured
>>>> by
>>>> network daemon, in this case hv_set_ifconfig causes "RTNETLINK: file
>>>> exists error"; in order to avoid this error this patch makes sure
>>>> double
>>>> checks the interface before trying anything.
>>>>
>>>> Signed-off-by: Eduardo Otubo <otubo@redhat.com>
>>>
>>> And if the daemon sets the address after you test it but before
>>> you try to set it in the script, what happens?
>>>
>>> This is why I hate changes like this.  They don't remove the problem,
>>> they make it smaller.  And smaller in a bad way.  Smaller makes the
>>> problem even more harder to diagnose when it happens.
>>>
>>> There is implicitly no synchonization between network configuration
>>> daemons and things people run by hand like this script.
>>>
>>> So, caveat emptor.
>>>
>>> I'm not applying this, sorry.
> 
> But also, looking from a different point of view, the current upstream
> solution does not avoid the problems you mentioned. My fix at least
> avoids double configuration and RTNETLINK errors. So perhaps you could
> consider this as "a better version walking towards an ideal fix"?

I didn't say upstream avoids the problem.

In fact, that's the good thing.  It doesn't try to do something it
cannot do without explicit pieces of synchronization infrastructure
between such tools.