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  PHC 
Open Source and information security mailing list archives
 
Hash Suite for Android: free password hash cracker in your pocket
[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Date:   Sun, 3 Jan 2021 10:20:11 -0800
From:   Stephen Hemminger <stephen@...workplumber.org>
To:     Masahiro Yamada <masahiroy@...nel.org>
Cc:     Valdis Klētnieks <valdis.kletnieks@...edu>,
        Michal Marek <michal.lkml@...kovi.net>,
        "David S. Miller" <davem@...emloft.net>,
        Networking <netdev@...r.kernel.org>,
        Linux Kbuild mailing list <linux-kbuild@...r.kernel.org>,
        Linux Kernel Mailing List <linux-kernel@...r.kernel.org>
Subject: Re: Kconfig, DEFAULT_NETSCH, and shooting yourself in the foot..

On Sun, 3 Jan 2021 15:30:30 +0900
Masahiro Yamada <masahiroy@...nel.org> wrote:

> On Sun, Jan 3, 2021 at 5:14 AM Valdis Klētnieks <valdis.kletnieks@...edu> wrote:
> >
> > Consider the following own goal I just discovered I scored:
> >
> > [~] zgrep -i fq_codel /proc/config.gz
> > CONFIG_NET_SCH_FQ_CODEL=m
> > CONFIG_DEFAULT_FQ_CODEL=y
> > CONFIG_DEFAULT_NET_SCH="fq_codel"
> >
> > Obviously, fq_codel didn't get set as the default, because that happens
> > before the module gets loaded (which may never happen if the sysadmin
> > thinks the DEFAULT_NET_SCH already made it happen)
> >
> > Whoops. My bad, probably - but....
> >
> > The deeper question, part 1:
> >
> > There's this chunk in net/sched/Kconfig:
> >
> > config DEFAULT_NET_SCH
> >         string
> >         default "pfifo_fast" if DEFAULT_PFIFO_FAST
> >         default "fq" if DEFAULT_FQ
> >         default "fq_codel" if DEFAULT_FQ_CODEL
> >         default "fq_pie" if DEFAULT_FQ_PIE
> >         default "sfq" if DEFAULT_SFQ
> >         default "pfifo_fast"
> > endif
> >
> > (And a similar chunk right above it with a similar issue)
> >
> > Should those be "if (foo=y)" so =m can't be chosen? (I'll be
> > happy to write the patch if that's what we want)
> >
> > Deeper question, part 2:
> >
> > Should there be a way in the Kconfig language to ensure that
> > these two chunks can't accidentally get out of sync?  There's other
> > places in the kernel where similar issues arise - a few days ago I was
> > chasing a CPU governor issue where it looked like it was possible
> > to set a default that was a module and thus possibly not actually loaded.
> >  
> 
> 
> If there is a restriction where a modular discipline cannot be the default,
> I think you can add 'depends on FOO = y'.
> 
> 
> 
> For example,
> 
> 
> choice
>            prompt "Default"
> 
>            config DEFAULT_FOO
>                   bool "Use foo for default"
>                   depends on FOO = y
> 
>            config DEFAULT_BAR
>                   bool "Use bar for default"
>                   depends on BAR = y
> 
>            config DEFAULT_FALLBACK
>                   bool "fallback when nothing else is builtin"
> 
> endchoice
> 
> 
> 
> 
> 

You can use a qdisc that is a module, it just has to be available when device
is loaded. Typically that means putting it in initramfs.

Powered by blists - more mailing lists