lists.openwall.net   lists  /  announce  owl-users  owl-dev  john-users  john-dev  passwdqc-users  yescrypt  popa3d-users  /  oss-security  kernel-hardening  musl  sabotage  tlsify  passwords  /  crypt-dev  xvendor  /  Bugtraq  Full-Disclosure  linux-kernel  linux-netdev  linux-ext4  linux-hardening  linux-cve-announce  PHC 
Open Source and information security mailing list archives
 
Hash Suite: Windows password security audit tool. GUI, reports in PDF.
[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <20181018042739.GA650@jagdpanzerIV>
Date:   Thu, 18 Oct 2018 13:27:39 +0900
From:   Sergey Senozhatsky <sergey.senozhatsky.work@...il.com>
To:     Tetsuo Handa <penguin-kernel@...ove.sakura.ne.jp>
Cc:     Sergey Senozhatsky <sergey.senozhatsky.work@...il.com>,
        Michal Hocko <mhocko@...nel.org>,
        Johannes Weiner <hannes@...xchg.org>, linux-mm@...ck.org,
        syzkaller-bugs@...glegroups.com, guro@...com,
        kirill.shutemov@...ux.intel.com, linux-kernel@...r.kernel.org,
        rientjes@...gle.com, yang.s@...baba-inc.com,
        Andrew Morton <akpm@...ux-foundation.org>,
        Petr Mladek <pmladek@...e.com>,
        Sergey Senozhatsky <sergey.senozhatsky@...il.com>,
        Steven Rostedt <rostedt@...dmis.org>,
        syzbot <syzbot+77e6b28a7a7106ad0def@...kaller.appspotmail.com>
Subject: Re: [PATCH v3] mm: memcontrol: Don't flood OOM messages with no
 eligible task.

On (10/18/18 11:46), Tetsuo Handa wrote:
> Sergey Senozhatsky wrote:
> > 
> > int printk_ratelimit_interval(void)
> > {
> >        int ret = DEFAULT_RATELIMIT_INTERVAL;
> >        struct tty_driver *driver = NULL;
> >        speed_t min_baud = MAX_INT;
> > 
> >        console_lock();
> >        for_each_console(c) {
> >                speed_t br;
> > 
> >                if (!c->device)
> >                        continue;
> >                if (!(c->flags & CON_ENABLED))
> >                        continue;
> >                if (!c->write)
> >                        continue;
> >                driver = c->device(c, index);
> >                if (!driver)
> >                        continue;
> > 
> >                br = tty_get_baud_rate(tty_driver to tty_struct [???]);
> >                min_baud = min(min_baud, br);
> >        }
> >        console_unlock();
> > 
> >        switch (min_baud) {
> >        case 115200:
> >                return ret;
> > 
> >        case ...blah blah...:
> >                return ret * 2;
> > 
> >        case 9600:
> >                return ret * 4;
> >        }
> >        return ret;
> > }
> 
> I don't think that baud rate is relevant. Writing to console messes up
> operations by console users. What matters is that we don't mess up consoles
> to the level (or frequency) where console users cannot do their operations.
> That is, interval between the last moment we wrote to a console and the
> first moment we will write to a console for the next time matters. Roughly
> speaking, remember the time stamp when we called call_console_drivers() for
> the last time, and compare with that stamp before trying to call a sort of
> ratelimited printk(). My patch is doing it using per call-site stamp recording.

To my personal taste, "baud rate of registered and enabled consoles"
approach is drastically more relevant than hard coded 10 * HZ or
60 * HZ magic numbers... But not in the form of that "min baud rate"
brain fart, which I have posted.

What I'd do:

-- Iterate over all registered and enabled serial consoles
-- Sum up all the baud rates
-- Calculate (*roughly*) how many bytes per second/minute/etc my
   call_console_driver() can push

        -- we actually don't even have to iterate all consoles. Just
	   add a baud rate in register_console() and sub baud rate
	   in unregister_console() of each console individually
	-- and have a static unsigned long in printk.c, which OOM
	   can use in rate-limit interval check

-- Leave all the noise behind: e.g. console_sem can be locked by
   a preempted fbcon, etc. It's out of my control; bad luck, there
   is nothing I can do about it.
-- Then I would, probably, take the most recent, say, 100 OOM
   reports, and calculate the *average* strlen() value
   (including \r and \n at the end of each line)

	(strlen(oom_report1) + ... + strlen(omm_report100)) / 100

   Then I'd try to reach an agreement with MM people that we will
   use this "average" oom_report_strlen() in ratelimit interval
   calculation. Yes, some reports will be longer, some shorter.

	Say, suppose...
	
	I have 2 consoles, and I can write 250 bytes per second.
	And average oom_report is 5000 bytes.
	Then I can print one oom_report every (5000 / 250) seconds
	in the *best* case. That's the optimistic baseline. There
	can be printk()-s from other CPUs, etc. etc. No one can
	predict those things.

	Note, how things change when I have just 1 console enabled.

	I have 1 console, and I can write 500 bytes per second.
	And average oom_report is 5000 bytes.
	Then I can print one oom_report every (5000 / 500) seconds
	in the *best* case.

Just my $0.02. Who knows, may be it's dumb and ugly.
I don't have a dog in this fight.

	-ss

Powered by blists - more mailing lists

Powered by Openwall GNU/*/Linux Powered by OpenVZ