On Thu, 3 Sep 2015 16:32:00 -0400
"J. Bruce Fields" <bfields@xxxxxxxxxxxx> wrote:
> On Thu, Sep 03, 2015 at 04:00:11PM -0400, Jeff Layton wrote:
> > On Thu, 3 Sep 2015 15:42:55 -0400
> > "J. Bruce Fields" <bfields@xxxxxxxxxxxx> wrote:
> >
> > > On Thu, Sep 03, 2015 at 03:05:48PM -0400, J. Bruce Fields wrote:
> > > > On Thu, Sep 03, 2015 at 02:45:44PM -0400, Jeff Layton wrote:
> > > > > On Thu, 3 Sep 2015 14:31:03 -0400
> > > > > "J. Bruce Fields" <bfields@xxxxxxxxxxxx> wrote:
> > > > >
> > > > > > On Thu, Sep 03, 2015 at 01:36:25PM -0400, Jeff Layton wrote:
> > > > > > > This is the userland companion patch for the patch to add a "nowcc"
> > > > > > > option to knfsd. This just adds the necessary code to allow userspace
> > > > > > > to set that option.
> > > > > > >
> > > > > > > Signed-off-by: Jeff Layton <jeff.layton@xxxxxxxxxxxxxxx>
> > > > > > > ---
> > > > > > > support/include/nfs/export.h | 3 ++-
> > > > > > > support/nfs/exports.c | 5 +++++
> > > > > > > utils/exportfs/exportfs.c | 2 ++
> > > > > > > utils/exportfs/exports.man | 14 ++++++++++++++
> > > > > > > 4 files changed, 23 insertions(+), 1 deletion(-)
> > > > > > >
> > > > > > > diff --git a/support/include/nfs/export.h b/support/include/nfs/export.h
> > > > > > > index 1194255899bd..68bee5c2f305 100644
> > > > > > > --- a/support/include/nfs/export.h
> > > > > > > +++ b/support/include/nfs/export.h
> > > > > > > @@ -26,7 +26,8 @@
> > > > > > > #define NFSEXP_CROSSMOUNT 0x4000
> > > > > > > #define NFSEXP_NOACL 0x8000 /* reserved for possible ACL related use */
> > > > > > > #define NFSEXP_V4ROOT 0x10000
> > > > > > > -#define NFSEXP_PNFS 0x20000
> > > > > > > +#define NFSEXP_PNFS 0x20000
> > > > > > > +#define NFSEXP_NOWCC 0x40000
> > > > > > > /*
> > > > > > > * All flags supported by the kernel before addition of the
> > > > > > > * export_features interface:
> > > > > > > diff --git a/support/nfs/exports.c b/support/nfs/exports.c
> > > > > > > index 0aea6f154f09..791b5f756c9c 100644
> > > > > > > --- a/support/nfs/exports.c
> > > > > > > +++ b/support/nfs/exports.c
> > > > > > > @@ -276,6 +276,7 @@ putexportent(struct exportent *ep)
> > > > > > > if (ep->e_flags & NFSEXP_NOREADDIRPLUS)
> > > > > > > fprintf(fp, "nordirplus,");
> > > > > > > fprintf(fp, "%spnfs,", (ep->e_flags & NFSEXP_PNFS)? "" : "no_");
> > > > > > > + fprintf(fp, "%swcc,", (ep->e_flags & NFSEXP_NOWCC) ? "no" : "");
> > > > > > > if (ep->e_flags & NFSEXP_FSID) {
> > > > > > > fprintf(fp, "fsid=%d,", ep->e_fsid);
> > > > > > > }
> > > > > > > @@ -586,6 +587,10 @@ parseopts(char *cp, struct exportent *ep, int warn, int *had_subtree_opt_ptr)
> > > > > > > setflags(NFSEXP_PNFS, active, ep);
> > > > > > > else if (!strcmp(opt, "no_pnfs"))
> > > > > > > clearflags(NFSEXP_PNFS, active, ep);
> > > > > > > + else if (!strcmp(opt, "wcc"))
> > > > > > > + clearflags(NFSEXP_NOWCC, active, ep);
> > > > > > > + else if (!strcmp(opt, "nowcc"))
> > > > > > > + setflags(NFSEXP_NOWCC, active, ep);
> > > > > > > else if (strncmp(opt, "anonuid=", 8) == 0) {
> > > > > > > char *oe;
> > > > > > > ep->e_anonuid = strtol(opt+8, &oe, 10);
> > > > > > > diff --git a/utils/exportfs/exportfs.c b/utils/exportfs/exportfs.c
> > > > > > > index 87582316b086..a2afc80ee6ea 100644
> > > > > > > --- a/utils/exportfs/exportfs.c
> > > > > > > +++ b/utils/exportfs/exportfs.c
> > > > > > > @@ -823,6 +823,8 @@ dump(int verbose, int export_format)
> > > > > > > c = dumpopt(c, "no_acl");
> > > > > > > if (ep->e_flags & NFSEXP_PNFS)
> > > > > > > c = dumpopt(c, "pnfs");
> > > > > > > + if (ep->e_flags & NFSEXP_NOWCC)
> > > > > > > + c = dumpopt(c, "nowcc");
> > > > > > > if (ep->e_flags & NFSEXP_FSID)
> > > > > > > c = dumpopt(c, "fsid=%d", ep->e_fsid);
> > > > > > > if (ep->e_uuid)
> > > > > > > diff --git a/utils/exportfs/exports.man b/utils/exportfs/exports.man
> > > > > > > index 93092463153b..9213a228bb04 100644
> > > > > > > --- a/utils/exportfs/exports.man
> > > > > > > +++ b/utils/exportfs/exports.man
> > > > > > > @@ -417,6 +417,20 @@ devices. The default can be explicitly requested with the
> > > > > > > .I no_pnfs
> > > > > > > option.
> > > > > > >
> > > > > > > +.TP
> > > > > > > +.IR nowcc
> > > > > > > +This option disables the collection and reporting of WCC (weak cache
> > > > > > > +consistency) data in NFSv3 requests. RFC 1813 recommends that all
> > > > > > > +servers report this information to the clients as it allows the client
> > > > > > > +to avoid some extra round trips to the server. In some underlying
> > > > > > > +filesystems however, collecting this data can be cost prohibitive and
> > > > > > > +atomicity is difficult to guarantee.
> > >
> > > By the way, independently of this change, maybe we should by default
> > > disable the pre-op attribute on distributed filesystems? As I
> > > understand it we're just depending on the i_mutex for atomicity, which
> > > can't be right in that case.
> > >
> > > --b.
> >
> > Right. In the case of something like a clustered or network filesystem
> > you really can't guarantee atomicity, so why even try?
> >
> > So, that is another option...we could add a flags field to the
> > export_operations struct and use one flag to indicate whether to send
> > wcc attrs. That'd give us the ability to add other flags in the future
> > as well.
> >
> > The reason I went with the export option is that I didn't really want
> > to add the flags field, and I figured the flexibility of being able to
> > do that on any filesystem might be helpful.
> >
> > That said, I'm open to doing it that way instead of (or in addition to)
> > the export option if you think it would be better.
>
> Well, it's sounding so far like:
>
> - nobody will want this on a non-distributed filesystem (hard to
> see what the benefit would be; and it might cost).
> - people probably *do* want this on any distributed filesystem.
> - we could save users some trouble by not giving them another
> export option which they might just set wrong.
>
> Eh, but I think I'm wrong about #2. Dropping the pre-op attribute is
> arguably necessary for correctness, but dropping the post-op attribute
> as well isn't necessary and might hurt, right? OK.
>
Oh, hmmm.
The pre-op attrs are generally pretty lightweight. It just scrapes the
info out of the in-core inode. The post-op attrs require a vfs_getattr
call though, and that's where the expensive part comes in. I guess I
looked at them as a unit. You either want both pre and post attrs or
none. The patch disables both when you add the nowcc option.
So I guess this means you're leaning toward a flags field in the export
operations instead of an export option? If so, I'm fine with that and I
can respin the patch accordingly.
There is probably some micro-optimization in turning them off on local
filesystems when you are doing DIO or having it act as a DS for
flexfiles, but I doubt it's significant enough to bother with.
--
Jeff Layton <jlayton@xxxxxxxxxxxxxxx>
--
To unsubscribe from this list: send the line "unsubscribe linux-nfs" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
