ADSM-L

Re: ADSM/Tivoli client modifies time of last access

2001-05-29 08:08:28
Subject: Re: ADSM/Tivoli client modifies time of last access
From: Richard Sims <rbs AT BU DOT EDU>
Date: Tue, 29 May 2001 08:08:21 -0400
>Using the ADSM/Tivoli client Version 4, Release 1, Level 1.0 on _SGI IRIX_,
>I saw that backuping a file modifies the last access time of that
>file. One can reproduce this behaviour by:
>
>        ls -lu file.xyz # shows the last access
>        dsmc incr file.xyz # now make a backup
>        ls -lu file.xyz # shows the last access which is now the time of backup
>
>I think, this is a buggy behaviour, because backing up a file should
>not touch - in any case - the file's status bits.

Joachim - This is a known issue with the product, undocumented in the B/A
          client manual.  From http://people.bu.edu/rbs/ADSM.QuickFacts :

Access time                             When a file was last read: its "atime"
                                        value (stat struct st_atime).
                                        The Backup function changes the file's
                                        access as it backs up each file.
                                        Beginning with the Version 2 Release 1
                                        Level 0.1 PTF, UNIX backup and archive
                                        processes changed the ctime instead of
                                        user access time (atime). This was done
                                        because the HSM feature on AIX uses
                                        atime in assessing a file's eligibility
                                        and priority for migration.  However,
                                        since the change of ctime conflicts with
                                        other existing software, with this Level
                                        0.2 PTF, UNIX backup and archive
                                        functions now perform as they did with
                                        Version 1:  atime is updated, but not
                                        ctime.

We get around that by the rather painful step of making copies of our AIX
logical volumes, and then mounting those to back up the copy.  We were
expecting FlashCopy on the IBM ESS to be our salvation, but marketing promises
of its efficacy have apparently been exaggerated beyond the reality.
Network Appliance devices can make an instant snapshot image of a file system
for convenient backup, a la AFS design.
One thinks of maybe getting around the problem by remounting a mounted file
system read-only; but in AIX that doesn't work, as lower level mechanisms know
that the singular file has been touched.  (See topic "MOUNTING FILE SYSTEMS
READ-ONLY FOR BACKUP" in QuickFacts.)

>The LINUX client Version 4, Release 1, Level 2.0 does _not_ show this 
>behaviour.

I wonder if that inconsistency is an oversight in the client code.
In this case, then, I would expect the ctime value to have changed.
This is not too cool, either, in that security scans examine the ctime value
to detect if hackers have been "tailoring" system files to their purposes.

   Richard Sims, BU
<Prev in Thread] Current Thread [Next in Thread>