Skip site navigation (1)Skip section navigation (2)
Date:      Wed, 1 Sep 2010 06:34:44 -0400
From:      Martin Cracauer <cracauer@cons.org>
To:        Kostik Belousov <kostikbel@gmail.com>
Cc:        freebsd-fs@freebsd.org, Martin Cracauer <cracauer@cons.org>
Subject:   Re: fsync(2) and on-disk write-back cache
Message-ID:  <20100901103444.GA22129@cons.org>
In-Reply-To: <20100901064805.GF2396@deviant.kiev.zoral.com.ua>
References:  <20100830225841.GA9363@cons.org> <20100831160840.GA74125@icarus.home.lan> <alpine.GSO.1.10.1008311210091.9337@multics.mit.edu> <20100831230728.GA36384@cons.org> <20100901064805.GF2396@deviant.kiev.zoral.com.ua>

next in thread | previous in thread | raw e-mail | index | archive | help
Kostik Belousov wrote on Wed, Sep 01, 2010 at 09:48:05AM +0300: 
> On Tue, Aug 31, 2010 at 07:07:28PM -0400, Martin Cracauer wrote:
> > Benjamin Kaduk wrote on Tue, Aug 31, 2010 at 12:12:04PM -0400: 
> > > On Tue, 31 Aug 2010, Jeremy Chadwick wrote:
> > > 
> > > >On Mon, Aug 30, 2010 at 06:58:42PM -0400, Martin Cracauer wrote:
> > > >>I always assumed the answer to this question is "of course":
> > > >>
> > > >>When doing an fsync (waiting for the commit), do we actually tell the
> > > >>disk to flush the on-disk write-back cache (if that is in use) to the
> > > >>platters?
> > > >>
> > > >>I just went down some code paths in both FreeBSD and Linux and in both
> > > >>cases the paths for fsync quickly disappear in the generic
> > > >>block-by-block flushing code that is also used for regular (non-fsync)
> > > >>flushing.  I didn't see anything aware of the on-disk cache.
> > > >
> > > >I don't have an authoritative answer to your question, but this thread
> > > >seems to imply there's a relation between fsync() and an intentional
> > > >disk flush (BIO_FLUSH).  I'm sure when BIO_FLUSH is called depends on
> > > >the filesystem as well.
> > 
> > I just went the for-dummies way and annotated all relevant BIO_FLUSH
> > places with debug print statements.  They don't seem to be called when
> > doing an fsync on a file in a local filesystem.  ufs (no softupdates)
> > -> old-style SCSI disk.
> > 
> > I'll snoop around some more, try it on ZFS/SATA and do some timing
> > tests.
> Be assured that all variants of UFS (sync, async, SU, SU+J) do not
> call BIO_FLUSH.

With a ZFS on SATA it looks different.  ad_strategy gets called to
queue BIO_FLUSH three times (for a single-block fsync).  However,
those calls are after fsync completes and returned to the called.

And I still can't see how the disk would communicate individual
block's status (on platter or not), all I see is global commands to
flush all the cache (and I don't think that is synchronous).

Martin
-- 
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
Martin Cracauer <cracauer@cons.org>   http://www.cons.org/cracauer/



Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?20100901103444.GA22129>