linux-kernel - Re: sysfs permissions on dynamic attributes (led delay_on and delay

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: <CAMbhsRTNmfd7EgGGv3TU=qzBO0k81TLcKtgT57CAhvwPKCUWfw@mail.gmail.com>
Date:	Sat, 21 Jul 2012 09:22:20 -0700
From:	Colin Cross <ccross@...roid.com>
To:	Greg KH <gregkh@...uxfoundation.org>
Cc:	Richard Purdie <richard.purdie@...uxfoundation.org>,
	lkml <linux-kernel@...r.kernel.org>,
	Bryan Wu <bryan.wu@...onical.com>
Subject: Re: sysfs permissions on dynamic attributes (led delay_on and delay_off)

On Sat, Jul 21, 2012 at 9:13 AM, Greg KH <gregkh@...uxfoundation.org> wrote:
> On Sat, Jul 21, 2012 at 08:42:12AM -0700, Colin Cross wrote:
>> On Sat, Jul 21, 2012 at 4:21 AM, Richard Purdie
>> <richard.purdie@...uxfoundation.org> wrote:
>> > On Sat, 2012-07-21 at 01:26 -0700, Colin Cross wrote:
>> >> On Sat, Jul 21, 2012 at 12:33 AM, Richard Purdie
>> >> <richard.purdie@...uxfoundation.org> wrote:
>> >> > On Fri, 2012-07-20 at 21:08 -0700, Greg KH wrote:
>> >> >> On Fri, Jul 20, 2012 at 05:46:14PM -0700, Colin Cross wrote:
>> >> >> > I'm trying to use the standard ledtrig-timer.c code to handle led
>> >> >> > blinking for notifications on an Android device, and I'm hitting some
>> >> >> > issues with setting permissions on the dynamically created delay_on
>> >> >> > and delay_off attributes.  For most sysfs files, we have userspace
>> >> >> > uevent parser that watches for device add notifications and
>> >> >> > chowns/chmods attributes.  This doesn't work for delay_on and
>> >> >> > delay_off, because they are created later, when "timer" is written to
>> >> >> > the trigger attribute.  There is no uevent when the new files are
>> >> >> > created, and sysfs doesn't support inotify, so I don't see any way to
>> >> >> > receive an event to set the permissions.  This issue exists any time
>> >> >> > that device_create_file is called after device_add.
>> >> >> >
>> >> >> > What is the appropriate way to get an event to set the permissions?
>> >> >> > Add inotify support for sysfs file creation?  Send a KOBJ_CHANGE
>> >> >> > uevent in device_create_file?
>> >> >>
>> >> >> No.
>> >> >>
>> >> >> > Send a KOBJ_CHANGE uevent from the driver after calling
>> >> >> > device_create_file?
>> >> >>
>> >> >> Yes.
>> >> >>
>> >> >> > Dynamically create a timer device under /sys/class/leds/<led> so a new
>> >> >> > add uevent gets sent?
>> >> >>
>> >> >> Ick.
>> >> >>
>> >> >> > Promote blinking to be a core led feature instead of a trigger, so the
>> >> >> > files are always present?
>> >> >>
>> >> >> That's the best thing, why not just do that?
>> >> >
>> >> > This implies we should make every trigger a core led feature and
>> >> > effectively do away with triggers. I'm not sure that makes sense.
>> >>
>> >> Blinking is already effectively a core feature.  It is implemented in
>> >> led-core.c so it can be used by other triggers besides timer, it's
>> >> state is stored in the led_classdev structure, not in the trigger
>> >> data, and the only thing left in ledtrig-timer.c is the sysfs files.
>> >
>> > Having the attributes present all the time leads to some nasty questions
>> > like how the on/off delays interact with things like say a network
>> > activity trigger. Not all triggers are going to respect these delay
>> > values and I can imagine a whole new set of nasty bug reports with no
>> > easy solutions if this change is made...
>>
>> The delay_on and delay_off files could easily override the values from
>> the trigger.
>>
>> Sending a KOBJ_CHANGE uevent is not a great solution, it's still
>> horribly racy in userspace.  This script would never work reliably:
>> echo timer > trigger
>
> When this returned, the sysfs files would then be there, right?

Yes, but they would owned by root and not writable.  udev would be
triggered by the KOBJ_CHANGE event and eventually chown/chmod them,
but possibly too late.

>> echo 1000 > delay_on
>> echo 1000 > delay_off
>> echo 255 > brightness
>
> So this would work.
>
> What is racy here?

It's racy if the script is run as non-root, assuming udev has already
chowned/chmoded the trigger and brightness files.
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/