Re: Random chmod errors

[Bob Proulx]

Peter Kratzer wrote:
> some of my users are encountering random errors with the chmod command. The
> errors are not exactly reproducible, so I cannot even narrow down the
> circumstances of their occurrance.

Problems like that can be disheartening and hard to track down.  I saw
James' question and your response that you were working on an NFS
server.  I am not familiar with "NAS filer".  But if there is any
ability to look at logs for errors there that would be good.  This may
very well be a problem in the NFS between the systems.

> What else do I know:
> - Following message occurs: "chmod: changing permissions of
> `filename' (requested: 0644, actual: 0444): Operation not permitted"

Seeing "operation not permitted" indicates to me that the filesystem
(NFS server in this case?) refused the operation.  That is also the
message you would get when trying to chown without the privilege.

> - Trying the same command with the same option again does the job as
> expected
> - In recursive calls (chmod -R) some of the files in the specified
> directory do not have the requested permissions when the error occurs
> - In some cases on single files the file's permission are actually changed
> (the were quite surely different before, when the file was created) to the
> requested value, even if the error message occurs

I have had enough problems with NFS in the past that now it is one of
my "usual suspects" whether it is really the problem now or not.

> - The message differs from attempting to chmod files owned by different
> users in the following part: "(requested: 0644, actual: 0444)"
> - I have tried to find this message part in brackets ("requested: 0644,
> actual: 0444") in the chmod.c source, but without success

This may have been introduced in a downstream patch to the GNU
sources.  You could look at the RH distro source and see.

> At this point I have completely run out of ideas. Does anybody have any
> idea where this problem comes from, what the reason might be?

Not I.

> I would appreciate any advice on this problem.

I would really suspect NFS is involved in this problem.  But I have
been burned too many times in the past and so I always suspect NFS.

I would build a pristine copy of latest coreutils.  I would run all of
the make checks because maybe in the tests on your system it would
turn up something.  I would place a copy of the pristine newly
compiled chmod in /usr/local/bin so that it would shadow the distro
supplied version for your users.  If the problem goes away and that is
the only change then you know it is due to something in the code.  You
should even be able to double-blind test it if you desired.
Unfortunately if the problem remains you don't know anything more.

STABLE
  ftp://ftp.gnu.org/gnu/coreutils/coreutils-5.2.1.tar.gz
  ftp://ftp.gnu.org/gnu/coreutils/coreutils-5.2.1.tar.bz2

BETA
  ftp://alpha.gnu.org/gnu/coreutils/coreutils-5.3.0.tar.gz
  ftp://alpha.gnu.org/gnu/coreutils/coreutils-5.3.0.tar.bz2

Bob