Re: [PATCH net v2] net: Avoid calling WARN_ON() on -ENOMEM in netif_change_net_namespace()

[Ivan Abramov]

On Fri, 4 Apr 2025 10:53:35 +0200, Eric Dumazet wrote:
> On Fri, Apr 4, 2025 at 9:29 AM Ivan Abramov <i.abramov@xxxxxxxxxxxxxxxxx> wrote:
>>
>> On Thu, 3 Apr 2025 11:05:31 -0700, Stanislav Fomichev wrote:
>> > On 04/03, Ivan Abramov wrote:
>> >> It's pointless to call WARN_ON() in case of an allocation failure in
>> >> device_rename(), since it only leads to useless splats caused by deliberate
>> >> fault injections, so avoid it.
>>
>> > What if this happens in a non-fault injection environment? Suppose
>> > the user shows up and says that he's having an issue with device
>> > changing its name after netns change. There will be no way to diagnose
>> > it, right?
>>
>> Failure to allocate a few hundred bytes in kstrdup doesn't seem
>> practically possible and happens only in fault injection scenarios. Other
>> types of failures in device_rename will still trigger WARN_ON.
>
> If you want to fix this, fix it properly.
>
> Do not paper around the issue by silencing a warning.

As far as I know, WARN_ON call on -ENOMEM is the issue itself, since
it only fires in testing/deliberate scenarios. And this fixes just that,
without touching anything else.

How should proper fix of this look like? I would be glad to work up
some solution that satisfies maintainers' vision, but I don't see other
ways around it without some grand refactoring, which may bring more
problems than solutions.