From mboxrd@z Thu Jan 1 00:00:00 1970 From: Jeff Layton Subject: Re: [PATCH 4/4] cifs: remove /proc/fs/cifs/Experimental Date: Tue, 7 Dec 2010 19:09:29 -0500 Message-ID: <20101207190929.59caaea4@corrin.poochiereds.net> References: <1291731835-1120-1-git-send-email-jlayton@redhat.com> <1291731835-1120-5-git-send-email-jlayton@redhat.com> <20101207111328.6647c36c@tlielax.poochiereds.net> <20101207141059.0ddc6c7d@tlielax.poochiereds.net> Mime-Version: 1.0 Content-Type: text/plain; charset=ISO-8859-1 Content-Transfer-Encoding: QUOTED-PRINTABLE Cc: linux-cifs-u79uwXL29TY76Z2rM5mHXA@public.gmane.org To: Steve French Return-path: In-Reply-To: Sender: linux-cifs-owner-u79uwXL29TY76Z2rM5mHXA@public.gmane.org List-ID: On Tue, 7 Dec 2010 14:31:25 -0600 Steve French wrote: > On Tue, Dec 7, 2010 at 1:10 PM, Jeff Layton wrote= : > > On Tue, 7 Dec 2010 10:21:11 -0600 > > Steve French wrote: > > > >> The issue of how to prevent a page from being modified as it is > >> written out (due to signing in our case) has been discussed on lkm= l a > >> few times (e.g. T10 block devices). =A0 We should hold off on chan= ging > >> this until we have a way of handling the case of: =A0 we calculate= the > >> signature, but just before the page in the cache is remodified, we > >> send it with the wrong signature ... obviously if we reissue the w= rite > >> we are fine (the data is fine) but there may be better ways to loc= k > >> the page (and some suggestions have been made on lkml for similar > >> sounding problems). > >> > > > > This behavior has been "Experimental" for years. At what point do w= e > > remove this kludge? Who in their right mind is going to turn on a > > switch called "Experimental" to enable this? As a user, I'd certain= ly > > be reticent to do so. It's not clear what turning on "Experimental" > > would give me. >=20 > 20%+ better write performance >=20 > > If you think this behavior deserves to be switchable then let's put= a > > real usable switch on it. I don't think it ought to be hidden > > as /proc/fs/cifs/Experimental. Perhaps a module parameter for > > this would be more appropriate? >=20 > Probably best to work through how to fix the general issue on fsevel > (ie how to write a presumably unstable page from the cache out) >=20 >=20 >=20 Obviously there needs to be some sort of endgame for this code. It's not healthy to keep this under "Experimental" for years on end. That said, I don't really want to tackle that project at the moment.=20 I am however interested in moving this code out from under this "Experimental" procfile. Would you be amenable to a module option for this? If so, how would you describe this knob? What name should it have= ? The main reason I mention module option is that it's now very helpful to have this sort of knob based on a file under /proc/fs/cifs. It can't be set at boot time without a special script. --=20 Jeff Layton