Re: [PATCH] sysctl: add support for poll()

[Kay Sievers]

On Thu, Jun 2, 2011 at 15:12, Alan Cox <alan@xxxxxxxxxxxxxxxxxxx> wrote:
>> > Or to manage it properly.
>>
>> What if the user decides do invoke sethostname syscall "by hand"?
>> Hostname would change beneath any other process that is trying to
>> manage it properly. What this patch does is to notify that process
>> that something happened.
>
> That is a stupid argument. Shall we extend it to its logical idiotic end
> and ask
>
> "What if the user decides to recompile their kernel without sysfs poll
> support ?"

Alan please! This is not something we haven't thought through.

> You have to be root to run sethostname, at which point you are
> realistically at the command line, a superuser and you know what you are
> doing (eg using sethostname for non IP network naming, or cluster id, or
> other stuff).

Please stay to the actual problem this patch tries to resolve.

>> With this patch in, if anyone wants to manage a file under /proc/sys
>> there's really a small amount of code to write. He only has to define
>> the new poll struct for that file.
>
> Sure - and there is an 8 byte cost per sysctl node (of which we have
> rather a lot), and we really need to tackle sysfs not sysctl anyway.

It is. And we will very likely need poll() for other things in
/proc/sys too. It's the cost of providing functionality we just need
today.

I could understand arguing about things like: void *extra1; void
*extra2; in that very same structure, but not about something that
can't really be solved otherwise.

> I'm not averse to pollable sysfs/sysctl nodes at all although the memory
> hit on sysfs is going to be tricky to manage and need clever code.

Yeah, but not related to the problem this patch tries to solve.

> I just think the utsname is a completely misguided example and whoever is
> trying to do it doesn't actually understand the limits of utsname.

We are not talking about limits of a certain infrastructure. It is
used, it will not go away, we need to support it.

This is about propagating in-kernel state changes to userspace. Please
open a different conversation for everything else.

Thanks,
Kay
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/