(Fwd) Re: DOSFS bug fixes, IDE drivers and sample released

Joel Sherrill joel.sherrill at OARcorp.com
Wed Apr 16 12:44:30 UTC 2003



"Eugeny S. Mints" wrote:
> 
> On Wed, 16 Apr 2003, Joel Sherrill wrote:
> 
> > Chris Caudle wrote:
> >
> > >>I have found that listing the directory after closing the file
> > >>seems to do the trick. It must purge the buffers to the disk.
> > >>
> > >>
> > >
> > >I haven't looked at the DOSFS code yet, does it have a sync() or fsync() function?
> > >If the file system keeps the allocation table cached in memory, you may need to flush the allocation table to the storage device, especially if you are shutting down power on the RTEMS system and moving the storage device to a different machine (or just re-booting into Windows, Linux, MacOS etc. on the same machine).
> > >
> > >
> > Or at least a flush() on the file itself?
> 
> Sorry, that I was silent for a long time but I was away from
> my  mail box.
> 
> As to sync/fsync/flush - sync call is implemented in DOSFS.
> I suppose that Angelo needs UNmount the volume to achieve
> his  goals. During 'unmount' a volume is automatically
> syncronized.

I understand that it has to sync when unmounting but shouldn't 
sync work even with it staying mounted?  I think the issue here
was simply that he was surprised it wasn't written until he closed.
[NOTE: This is standard behavior for many systems so is not unexpected.]

So the question is.. if I have an open file and want to ensure its
contents are written to disk, is it sufficient to sync() or should
it flush() and sync() and what?  I think unmounting is a bit too
heavy handed because it potentially impacts other tasks.

> >
> > Not to get me started on a tangent but my disertation included some
> > thought on the idea
> > that different embedded systems needed different disk caching and
> > scheduling algorithms.
> > Ideally those should be selectable on a per-file basis but at a
> > partition level would be
> > acceptable.  For example, a write-only log file should have data kept
> > only long enough to
> > get it to disk but a sequentially read file could use read-ahead and
> > discard as soon as
> > the application read the data.  With some (intelligent) hints from the
> > application,
> > the algorithms could be better selected.
> >
> > >-- Chris Caudle
> > >
> > >
> > >
> >
> >
> >
> 
> --
> Eugeny S. Mints
> OKTET Ltd.
> 1 Ulianovskaya st., Petergof, St.Petersburg, 198904 Russia
> Phone: +7(812)428-4384 Fax: +7(812)327-2246
> mailto:Eugeny.Mints at oktet.ru



More information about the users mailing list