Ext3 users - Feb 2002 - ext3 and chattr +S on postfix spools

Postfix <http://www.postfix.org> does a chattr +S on its spool directory
when it is installed on Linux.

This is what is written in the postfix startup script
--
#
# LINUX by default does not synchronously update directories -
# that's dangerous for mail.
#
if [ -f /usr/bin/chattr ]
then
    CHATTR="/usr/bin/chattr +S"
else
	CHATTR=echo
fi
--

Is this still required on ext3 ? If not, maybe some ext3 hacker can
inform Wietse about it

I recollect once that Andrew Morton had shown that setting data=journal
was quite beneficial for Postfix, I am not sure if that was with chattr
+S or without

Regards, Yusuf
-- 
Yusuf Goolamabbas
yusufg@outblaze.com

Hi,

On Mon, Feb 04, 2002 at 05:34:52PM +0800, Yusuf Goolamabbas
wrote:
> Postfix <http://www.postfix.org> does a chattr +S on its spool
directory
> when it is installed on Linux.
> 
> Is this still required on ext3 ?
I'm not sure --- it depends on what Postfix's assumptions are.  If it
expects all directory operations to be synchronous then the flag is
still needed.  If it is satisfied with fsync flushing dir information,
then no it is not needed.

Cheers,
 Stephen

On Tue, 2002-02-05 at 15:26, Wietse Venema wrote:
> Erik Smit:
> > On Tue, 2002-02-05 at 14:08, Wietse Venema wrote:
> > > Erik Smit:
> > > > On Tue, 2002-02-05 at 04:20, Wietse Venema wrote:
> > > > > ext3 is unsafe for maildir, and with softupdates, so is
ffs.
> > > > > One step forwards, two steps back.
> > > > > 
> > > > > Fortunately, some people still have a choice.
> > > > 
> > > > For the newbies among us (me :) would you mind explaining
what exactly
> > > > causes it to be unsafe? As far I know (but like I said,
I'm just a noob)
> > > > ext3 is journalling metadata and data so I cannot understand
how data
> > > > could be lost.
> > > 
> > > maildir delivery requires that certain file system operations are
> > > atomic and synchronous.
> > > 
> > > [...]
> > >
> > > Translated into ext3 speak that means: the link and unlink
operations
> > > are written to the log before the respective system call returns.
> > 
> > Excuse my ignorance, but since it has been written to log and the
> > machine would crash before the systemcall returns wouldn't this
mean it
> > would be recovered (by journal replaying) during boot?
> 
> IFF the information is written to the ON DISK log, then the
> recovery is possible.
Wasn't that the whole point to journaling filesystems (and chattr +S)?

Regards,

Erik Smit