Subject: Re: FFS reliability problems
To: der Mouse <mouse@Rodents.Montreal.QC.CA>
From: Manuel Bouyer <bouyer@antioche.eu.org>
List: tech-kern
Date: 05/17/2002 21:58:14
On Fri, May 17, 2002 at 02:42:55PM -0400, der Mouse wrote:
> >>> A file with refcount==0 has to be cleared, not reconnected.
> >> Why does it "[have] to be" cleared rather than being reconnected?
> > Because it mean it was unlinked.
> 
> Yes, but why does that mean it "has to be" cleared?  I can't see any
> harm that would come from relinking it

It's already hard to recover data from /lost+found when something got
wrong; adding deleted files will make more files to sort and won't help -
especially if you have to determine the last version of the file.

> at least optionally (ie, as an
> option to fsck_ffs).

optionally is't OK with me

> 
> > If it was deleted, why would someone want to recover it after a crash
> > ?
> 
> The obvious answer is "because it has valuable data in it", and I can't
> see anything wrong with that answer.  Certainly it seems it did in
> greywolf's case.
> 
> > If the machine didn't crash, this file would never have been seen
> > again.
> 
> But if the program that's holding it open doesn't get a chance to save
> in more permanent storage the data in it, it can still be valuable.
> (Yes, this is poor application design.  But users and even admins do
> not always have the option of redesigning the applications they run.)

OK, I didn't think of this case.

-- 
Manuel Bouyer <bouyer@antioche.eu.org>
--