> On Fri, 24 Aug 2001 at 12:28, Hans Kratz wrote:
> > The Reiserfs slides state that link() on Reiserfs is not synchronous.
> > This is correct but is the same not true for XFS as well?
>
> This I do not know, either. Thank you for bringing this up. There is a
> clear FAQ entry about ReiserFS not treating link() as synchronous because
> of issues with QMail (and most other mail transfer agents, I think, but am
> not sure).
>
> I was under the impression that XFS treated the link() as a synchronous
> operation. Maybe the experts can shed some light on this?
>
> > Whatever is the case you should state if XFS link()/unlink() is
> > synchronous on the XFS slides for symmetry reasons.
>
> I agree. Thank you very much for the feedback. I will put information on
> this in my slides as soon as I get a response from the XFS development
> team.
>
> --> Jijo
>
link is not synchronous in xfs. The only way to acheive this is to make
all transactions synchronous using the wsync mount option. For performance
reasons this is not something you want to do unless you absolutely
have to. For a back door trick to make sure a transaction is on the
disk, you can either do a small O_SYNC write to a file, or some other
operation which will flush the log. I do not think there is a hard and
fast way to make sure that the link and just the link gets pushed to
disk immediately without doing some other operation.
Steve
|