linux-kernel - Re: [PATCH 1/2] fs/efivarfs: restrict inode permissions

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]

Date:   Wed, 21 Feb 2018 12:40:48 -0800
From:   Linus Torvalds <torvalds@...ux-foundation.org>
To:     "Luck, Tony" <tony.luck@...el.com>
Cc:     Andi Kleen <ak@...ux.intel.com>,
        Ard Biesheuvel <ard.biesheuvel@...aro.org>,
        Joe Konno <joe.konno@...ux.intel.com>,
        "linux-efi@...r.kernel.org" <linux-efi@...r.kernel.org>,
        Linux Kernel Mailing List <linux-kernel@...r.kernel.org>,
        Jeremy Kerr <jk@...abs.org>,
        Matthew Garrett <mjg59@...gle.com>,
        Peter Jones <pjones@...hat.com>,
        Andy Lutomirski <luto@...nel.org>,
        James Bottomley <james.bottomley@...senpartnership.com>
Subject: Re: [PATCH 1/2] fs/efivarfs: restrict inode permissions

On Wed, Feb 21, 2018 at 11:58 AM, Luck, Tony <tony.luck@...el.com> wrote:
>
> How are you envisioning this rate-limiting to be implemented? Are
> you going to fail an EFI call if the rate is too high?  I'm thinking that
> we just add a delay to each call so that we can't exceed the limit.

Delaying sounds ok, I guess.

But the "obvious" implementation may be simple:

    static void efi_ratelimit(void)
    {
        static DEFINE_RATELIMIT_STATE(ratelimit, HZ, 100);

        if (!__ratelimit(&ratelimit))
                msleep(10);
        }
    }

but the above is actually completely broken.

Why? If you have multiple processes, they can each only do a hundred
per second, but globally they can do millions per second by just
having a few thousand threads. They all sleep, but..

So how do you restrict it *globally*?

If you put this all inside a lock like a mutex, you can generate
basically arbitrary delays, and you're back to the DoS schenario. A
fair lock will allow thousands of waiters to line up and make the
delay be

But maybe I'm missing some really obvious way. You *can* make the
msleep be a spinning wait instead, and rely on the scheduler, I guess.

Or maybe I'm just stupid and am overlooking the obvious case.

Again, making the ratelimiting be per-user makes all of these issues
go away. Then one user cannot delay another one.

             Linus