Re: swp_pager_meta_build DoS printf

On 15 August 2012 19:21, John Baldwin <jhb_at_freebsd.org> wrote:
> On Monday, August 13, 2012 1:49:38 am Sergey Kandaurov wrote:
>> On 2 July 2012 20:31, Bjoern A. Zeeb <bzeeb-lists_at_lists.zabbadoz.net> wrote:
>> >
>> > On 2. Jul 2012, at 14:36 , John Baldwin wrote:
>> >
>> >> On Sunday, July 01, 2012 8:23:31 am Bjoern A. Zeeb wrote:
>> >>> Hey,
>> >>>
>> >>> hitting this printf in swp_pager_meta_build()
>> >>>
>> >>>                        if (uma_zone_exhausted(swap_zone)) {
>> >>>                                printf("swap zone exhausted, increase
> kern.maxswzone\n");
>> >>>                                vm_pageout_oom(VM_OOM_SWAPZ);
>> >>>                                pause("swzonex", 10);
>> >>>                        } else
>> >>>
>> >>> seems to be an effective way to put the machine into a state of no
> recovery
>> >>> unless the memory situation would be able to clear itself.  Not that it
> wouldn't
>> >>> otherwise be any better but in addition having a couple of tenthousands
> of these
>> >>> going to console as well is really not helpful to try to do anything
> either.  Can
>> >>> we make it a log() call or something?
>> >>>
>> >>> /bz
>> >>>
>> >>> PS: I am not sure as I have seen it on someone else's machines and it's
>> >>> probably been ZFS that caused it.  I unfortunately neither had a way to
>> >>> get back in or break to a kernel debugger, so information is sparse.
>> >>
>> >> This used to be a silent deadlock before I added the printf() and the
> call to
>> >> OOM. :-P  Do you just want to ratelimit the printf?  We have an API to
> ratelimit
>> >> printf's already.
>> >
>> > Ratelimit would be fine;  I was writing that on the wrong time of the
> wrong day to
>> > just get it out;  could you do that?
>>
>> Hi, looks like the discussion was abandoned.
>> What about this patch? It enables to ratelimit the printf.
>> Also, are the new variables put in the right places from the style pov?
>
> I think DES has a newer variant of this now?

Yes, and for my taste it's better as it is closer to fix the cause
of the problem, not the consequences (unrated printf).  Not sure if
both should be combined, I believe the DES change is enough here.

-- 
wbr,
pluknet