bug-coreutils
[Top][All Lists]
Advanced

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: chmod set-gid/set-uid behavior change issues


From: Bob Proulx
Subject: Re: chmod set-gid/set-uid behavior change issues
Date: Wed, 26 Jul 2006 23:40:06 -0600
User-agent: Mutt/1.5.9i

Paul Eggert wrote:
> Bob Proulx writes:
> > Contrary to the statement above my testing shows that FreeBSD chmod
> > does not behave this way.
> 
> OK, thanks, clearly NEWS (at least :-) needs to get fixed.

:-)  Agreed.

Also I applaud your efforts to make working with setgid directories
nicer.  It is a useful paradigm and GNU should be able to make it a
nice way to operate.  But I am not sure how far things can go because
tar, cpio, etc. all set file permissions too and have their own ideas.
But at least the coreutils can be good there.  I just worry about
drifting too far afield.

> The behavior I observe on OpenBSD 3.9 is that sometimes the setgid bit
> is preserved, and sometimes not.

Hmm...  I actually think the mkdir -m case is wrong.  But HP-UX works
that way too.  I imagine it is a SysV'ism.  And even though I think it
is not behaving literally as I expect it is actually a nicer behavior
in many ways.

> Clearly there is no consistency here across systems, which gives us a
> bit of freedom to decide what is the best thing to do for GNU.

Also looking at what the behavior of 'cp -pR' is on systems shows that
things are different.  On HP-UX 'cp -pR' clears the sgid bit.

On HP-UX:
  mkdir /tmp/t && cd /tmp/t
  chmod g+s .
  mkdir foo
  ls -ld foo
  drwxrwsr-x   2 rwp        esl             96 Jul 25 15:29 foo

  /bin/cp -pR foo bar
  ls -ld bar
  drwxrwxr-x   2 rwp        esl             96 Jul 25 15:29 bar

That has burned me on occasion.  I imagine it is the final chown that
is clearing the setgid bit in the kernel but I have not traced through
it.  Because SysV allows you to give away ownership but only the owner
can chmod the file so the chmod must happen first there followed by
the chown which clears the setuid bits.  But I consider that
undesirable behavior.

On HP-UX:
  chmod g+s .
  /bin/mkdir -m 775 foo
  ls -ld foo
  drwxrwsr-x    2 rwp      esl            96 Jul 26 22:55 foo
  
  /bin/mkdir -m a=r foo
  ls -ld foo
  dr--r-Sr--    2 rwp      esl            96 Jul 26 23:11 foo
  
  /bin/mkdir -m a+rwx foo
  ls -ld foo
  d-----S-w-    2 rwp      esl            96 Jul 26 23:37 foo

So mkdir seems to work in your desired fashion.  The a+rwx case is
wacked though.  But FreeBSD is exactly the reverse in these two cases.

On FreeBSD:
  chmod g+s .
  mkdir -m 775 foo
  ls -ld foo
  drwxrwxr-x  2 rwp  user  4096 Jul 26 22:58 foo

  /bin/mkdir -m a=r foo
  ls -ld foo
  dr--r--r--  2 rwp  user  4096 Jul 26 23:11 foo

  mkdir foo
  ls -ld foo
  drwxr-sr-x  2 rwp  user  4096 Jul 26 23:05 foo

  /bin/cp -pR foo bar
  ls -ld bar
  drwxrwsr-x  2 rwp  user  4096 Jul 26 23:05 foo

In my opinion FreeBSD seems to have two cases right and one case
wrong.  Of course I am a literal minded person about programs and
think they should do exactly what I tell them.  Although the HP-UX
mkdir -m behavior is probably what you *want* more often.  But when
heuristics are involved to try to guess what I want instead of what I
say it usually works out badly for me in the long run.

But FreeBSD gets this case wrong and drops the 's' bit here.

  /bin/mkdir -m a+rwx foo
  ls -ld foo
  drwxrwxrwx  2 rwp  user  4096 Jul 26 23:36 foo

> However, in normal use it is more convenient if "chmod 755 DIR"
> preserves the setgid bit.

I don't know.  I could be convinced by group concensus otherwise.
Using 755 means to me a complete setting which would turn off set-id
bits.  Using symbolic modes 'chmod a+rwx DIR' I would expect to
preserve setgid bits.  But 'chmod a=rwx DIR' I would expect to clear
set-id.  There seems to be a trend to move away from numeric controls.
Maybe we preserve legacy behavior for the octal controls but introduce
new behavior when using the symbolic controls?  But a+rwx already
handles the problem.  I admit I have not completely thought this
problem out.  Sorry for rambling here.

> Also, it is more consistent with what Solaris does, since Solaris
> "chmod 755 DIR" preserves setgid.

I am really surprised that Solaris behaves that way.  It is opposite
of FreeBSD.  So the SysV vs BSD differences continue. :-)

> If we remove this feature, I'd like to change things to be 100%
> consistent with Solaris, and to preserve the setgid bit even if the
> user says "chmod 0755 DIR".

Personally I prefer the FreeBSD behavior, being the literal minded
person that I am.  :-)

Bob




reply via email to

[Prev in Thread] Current Thread [Next in Thread]