Re: Patch related with Fork Bombing Atack

[Jens Axboe <jens.axboe@oracle.com>]

On Fri, Jun 01 2007, Daniel Hazelton wrote:
> On Friday 01 June 2007 03:30:20 Jens Axboe wrote:
> > On Fri, Jun 01 2007, Daniel Hazelton wrote:
> > > On Friday 01 June 2007 02:48:59 Anand Jahagirdar wrote:
> > > > On 5/31/07, Jens Axboe <jens.axboe@oracle.com> wrote:
> > > > > On Thu, May 31 2007, Anand Jahagirdar wrote:
> > > > > > 2) Printk message in my patch will definitely help
> > > > > > Administrator/Root User to detect which particular user is trying
> > > > > > fork bombing attack on his machine by looking at /var/log/messages
> > > > > > or dmesg . he can take action against that particular user and kill
> > > > > > his processes.
> > > > >
> > > > > You just opened a DoS possibility for any user, they can now flood
> > > > > the syslog instead.
> > > >
> > > > Jens Axboe
> > > >
> > > >      when they try to flood the syslog using fork bombing attack,
> > > > their messge will be printed only once in syslog and it will show how
> > > > many times it has repeated. due to this he will not able to flood the
> > > > syslog.and i am using only one single variable in my printk messge so
> > > > it is quite not possible to flood the syslog.
> > > >
> > > >    am i missing something??
> > > >
> > > > anand
> > >
> > > Most definately. Each printk() call outputs to syslog - which means
> > > that every time your code outputs its message there is another line in
> > > the logs. It then becomes possible to use that to flood the syslog.
> >
> > I think Anand is assuming that because syslog may coalesce identical
> > messages into "repeated foo times" in the messages file, that it's not a
> > dos. That is of course wrong.
> 
> Well... The problem, IMHO, isn't the logfiles flood-filling the disc
> but the klogd -> syslogd communications sucking down processor
> bandwidth and memory.  A quick grep of the sources shows that the
> kernel doesn't do that nifty "repeated foo times" bit, but the
> userspace syslogd. That's what that closing tagline in my previous
> response to this thread was all about - the whole fact that the
> "repeated foo times" thing is from the userspace component of the
> syslog system. Any program fed the massive amount of input starting a
> recursive forkbomb would generate with the kernel doing this "I've
> detected a forkbomb" printk would start to lag as it struggled to keep
> up - so it starts using more and more processor until its stopped the
> machine from doing anything *but* process those messages. (Oh, wait -
> I just described the DoS that happens when the syslog gets flooded -
> silly me!)

Yep, otherwise there would be no need to have introduced
printk_ratelimit().

-- 
Jens Axboe