samba - Jan 2002 - File Modify Times Are Unexpectedly Changed

Please repsond to me via email if possible to davepwhite@hotmail.com. No access
to the list at this time. Thanks.

I have 2 machines, one runs Windows NT 4.0 sp5, the other RH Linux 7.0.

The Linux box mounts an NT shared drive upon startup. The version number
reported by both smbclient and smbmount is 2.0.7.

Here is my problem: On the NT box, I create a file in the shared area. I then
check its various file date values using the
Explorer's properties dialog and write them down. Then I go to the Linux box
and simply cat the file's contents. I then return to
the NT box and again check the file's date properties. At this point, I find
that both the accessed time and the modified time have
changed from what was originally recorded. This seems strange to me as all that
has been done is that the file has been opened and
read. There has been no modification. The same thing occurs if I less the file
or do a touch -a to the file.

The result of all this is that my make process and my interaction with my
version control system (on the Windows box) are now hosed
up because the file is noted as having been changed when, in fact, it has not.

I see that in the docs for Samba, there is a config parameter that appears to
speak to this sort of problem. It is a share-level
parameter called "dos filetime resolution". However, since this is not
a Samba share, I cannot twiddle this parameter. I do not see
any similar parameter in the docs for smbclient or smbmount. Even if I did, I
still am not sure why the file's modify time has been
changed when no modification has been attempted.

Can anyone please tell me (1) what is going on here and (2) how to resolve the
problem so that my make and VCS are not affected?

Thanks in advance for your help.

David
Seattle, WA

On Tue, 1 Jan 2002, David White wrote:
> Here is my problem: On the NT box, I create a file in the shared area.
> I then check its various file date values using the Explorer's
> properties dialog and write them down. Then I go to the Linux box and
> simply cat the file's contents. I then return to the NT box and again
> check the file's date properties. At this point, I find that both the
> accessed time and the modified time have changed from what was
> originally recorded. This seems strange to me as all that has been
This is a known problem. I believe the same thing has been reported as tar
complaints about modified files and the VC++ IDE warning about modified
files.

It is related to smbfs doing open+close on the file.

smbfs opens files for writing if it can. Always. It also tries to write
back its idea of mtime and atime when a file is closed. Neither should be
a problem, although the timestamp writeback looks strange ... (why not do
it only when something has changed?)


For some reason I am unable to reproduce the problem you describe (and the
"tar-test" also passes). win2k instead of NT and possibly a newer
kernel,
but I don't know why it would work for me.

What kernel version are you running?

/Urban