xfs
[Top] [All Lists]

Re: [PATCH] [PRELIMINARY] Add support for file flags to xfsdump #1

To: linux-xfs@xxxxxxxxxxx
Subject: Re: [PATCH] [PRELIMINARY] Add support for file flags to xfsdump #1
From: Ivan Rayner <ivanr@xxxxxxx>
Date: Tue, 28 Oct 2003 10:41:53 +1100
In-reply-to: <20031024080129.GF3648@xxxxxxxxxxxxxxx>
Organization: SGI
References: <20031020073151.GD3648@xxxxxxxxxxxxxxx> <20031024105919.5e22a10e.ivanr@xxxxxxx> <20031024080129.GF3648@xxxxxxxxxxxxxxx>
Sender: linux-xfs-bounce@xxxxxxxxxxx
[I sent this mail yesterday but it didn't seem to go through. So, try again.]

On Fri, 24 Oct 2003 00:01:29 -0800, Ethan Benson wrote:

> On Fri, Oct 24, 2003 at 10:59:19AM +1000, Ivan Rayner wrote:
> > On Sun, 19 Oct 2003 23:31:51 -0800, Ethan Benson wrote:
> > > What I have done to fix the above problems in order:
> > Thanks for your work Ethan.  Although I've not looked your code, it sounds
> > like you're doing the right sort of thing.
> hopefully, let me know when you get a chance to look at the code.

Well, I don't think I will get the time.  Besides, it's been a while since
I was responsible for xfsdump -- it's someone else's job now.

> > > * dump/inomap.c: Add warning when SGI_XFSDUMP_SKIP_FILE xattr is
> > >   present, but still honor it, this attribute is deprecated in favor of
> > >   the new nodump file flag, support for the obsolete xattr should be
> > >   removed in a future xfsdump (its not that old so this shouldn't be
> > >   much of a problem).  Add check for new nodump file flag.
> > 
> > The SKIP_FILE attribute is still the supported mechanism in IRIX,
> > therefore it'd be a good idea to keep support for it in the Linux version.
> > It doesn't hurt performance, so I don't see any reason to remove it.
> 
> well its very young, it was just added last year i think, maybe more
> recently then that (by request from linux users wanting chattr +d
> functionality).  and the file attributes will probably end up getting
> merged into irix anyway.  the inode flags are really a much cleaner
> way for this to work, which is why id rather see the xattr kludge go
> away.

A couple of things:

- It's almost 2 years old (Nov 2001).

- Excluding files has been an issue for xfsdump for ages (well before  
  Linux). The original request that sparked this implementation was via an  
  IRIX newsgroup and referred to BSD rather than Linux.  It was implemented 
  on IRIX and Linux simultaneously.

- It doesn't matter how old it is, SGI has the IRIX compatibility  
  guarantee, which means that we cannot remove this functionality from 
  IRIX, even if it had only been in a single release.  This is a good
  thing.  And given that we'd want xfsdump to be compatible across IRIX
  and Linux, we wouldn't want to remove it from Linux either.

- File attributes aren't in IRIX yet, and even if they were it wouldn't 
  matter -- see previous point.

I don't dispute that this solution may be better, but that doesn't mean we
can or should get rid of the old solution.

It's not in SGI's best interest to _force_ customers to change the way
they do things simply because we have another solution that we think is
better.  Even worse is the idea that xfsdump will suddenly not do what
they expect it to do. "But we issued warnings," we'd argue and they'll say
they had xfsdump in a script and never saw the warnings and now they have
a backup that contains terrabytes of temporary files which they don't
want.

I don't see the problem with just advertising the new feature, and those
people who don't care about IRIX compatibility can simply remove the old
SKIP_FILE attributes and start using the new system if and when they like.
Unless there's a security or data corruption issue, it isn't up to us to
determine how other people should work.

If there is a definite performance or functional benefit for the user,
then I suppose you could issue a notice instead of a warning, letting the
user know that there is a better solution.  Otherwise, I'd prefer just
changing the man page.

Ivan


<Prev in Thread] Current Thread [Next in Thread>