Re: [PATCH 6/6] kmod: make __request_module() killable

[Oleg Nesterov <oleg@redhat.com>]

On 02/15, Andrew Morton wrote:
>
> On Tue, 14 Feb 2012 17:49:14 +0100
> Oleg Nesterov <oleg@redhat.com> wrote:
>
> > As Tetsuo Handa pointed out, request_module() can stress the
> > system while the oom-killed caller sleeps in TASK_UNINTERRUPTIBLE.
>
> Whine.
>
> Solving this problem is the entire point of the entire patchset and you
> told us almost nothing about it.  Please, provide a complete
> description of the problem which is being solved, so we can understand
> the value of the patchset?

I did ;) from the message I sent to security list:

	Tetsuo has the test-cases, but the problem (well, one of the problems) is
	simple.

	The task T uses "almost all" memory, then it does something which triggers
	request_module(). Say, it can simply call sys_socket(). This in turn needs
	more memory and leads to OOM. oom-killer correctly chooses T and kills it,
	but this can't help because it sleeps in TASK_UNINTERRUPTIBLE and after
	that oom-killer becomes "disabled" by the TIF_MEMDIE task T.

Credits to Tetsuo.

But in fact I think this change is "obviously good" anyway. Assuming
it is correct of course. request_module() is heavy, it can take the
unpredictable amount of time/resources to finish. It is not good we
can't interrupt the task which waits for completion.

Yes, this adds some complications and initially I wasn't agree with
Tetsuo, I thought this doesn't worth the trouble. But I hope that
this code is simple/clean enough.



Btw, there is another example of "unbounded" sleep in UNINTERRUPTIBLE,
vfork. I already have the patches, will send today.

Oleg.