On Tue, Sep 02, 2003 at 12:37:21PM -0700, Ed Alley wrote:
> 
> > On Tue, 2003-09-02 at 14:32, Ed Alley wrote:
> >> I'm running FreeBSD-4.8. Sometimes the file permissions for /dev/null get
> >> mysteriously changed by some unknown process to:
> >> 
> >>    crw------- 1 root wheel 2, 2 Sep 2 11:20 /dev/null
> 
> > On Tue, 2003-09-02 Adam McLaurin wrote:
> > That's very strange indeed. Have you tried using chflags to prevent the
> > permissions from being changed? This should do the trick, albeit a dirty
> > hack.
> 
> Sorry, I didn't mention that I tried setting flags on /dev/null:
> 
>       chflags schg /dev/null
> 
> What happens is that sendmail complains that it can't open /dev/null.
> 
> Hey! I just realized that this may be a clue! Does sendmail fiddle with
> /dev/null? What happens if sendmail tries to lock /dev/null after it
> opens it? Does schg prevent fcntl from locking /dev/null, if that is
> what sendmail uses?

Lock it why? There's no point locking the null device -- it's not like
it has contents that can be changed out from underneath a process...

Besides, a large number of processes tend to have open descriptors on
/dev/null -- any well behaved daemon process will close its stdin,
stdout and stderr and re-open them on /dev/null as part of the
standard setup for becoming a daemon.  See daemon(3).  Getting a
mandatory exclusive lock on /dev/null early in the boot process would
be a very effective way to cripple a system...

If you want to see what processes have an open file descriptor on
/dev/null, try:

    % fstat -f /dev | grep ' null '

There will be more than you expect.  As for tracking down what process
has mucked up the permissions on the device: that's going to be quite
laborious.  You'll probably have to do something horribly tedious like
not running each process (that uses /dev/null) in turn, and see if you
can identify when the chmod(2) doesn't happen.  It would have to be a
root-owned process to change the permissions on the device, which will
cut the list down a bit. Remember though that many daemon processes
will start as UID root in order to bind low-numbered network ports,
and then change their UID to something less privileged as a security
measure.

        Cheers,

        Matthew 

-- 
Dr Matthew J Seaman MA, D.Phil.                       26 The Paddocks
                                                      Savill Way
PGP: http://www.infracaninophile.co.uk/pgpkey         Marlow
Tel: +44 1628 476614                                  Bucks., SL7 1TH UK

Attachment: pgp00000.pgp
Description: PGP signature

Reply via email to