On Tue, Dec 01, 2020 at 11:43:31AM -0500, Jeff Layton wrote:
> On Mon, 2020-11-30 at 17:46 -0500, J. Bruce Fields wrote:
> > From: "J. Bruce Fields" <bfields@xxxxxxxxxx>
> >
> > inode_query_iversion() can modify i_version. Depending on the exported
> > filesystem, that may not be safe. For example, if you're re-exporting
> > NFS, NFS stores the server's change attribute in i_version and does not
> > expect it to be modified locally. This has been observed causing
> > unnecessary cache invalidations.
> >
> > The way a filesystem indicates that it's OK to call
> > inode_query_iverson() is by setting SB_I_VERSION.
> >
> > So, move the I_VERSION check out of encode_change(), where it's used
> > only in FATTR responses, to nfsd4_changeattr(), which is also called for
> > pre- and post- operation attributes.
> >
> > (Note we could also pull the NFSEXP_V4ROOT case into
> > nfsd4_change_attribute as well. That would actually be a no-op, since
> > pre/post attrs are only used for metadata-modifying operations, and
> > V4ROOT exports are read-only. But we might make the change in the
> > future just for simplicity.)
> >
> > Reported-by: Daire Byrne <daire@xxxxxxxx>
> > Signed-off-by: J. Bruce Fields <bfields@xxxxxxxxxx>
> > ---
> > fs/nfsd/nfs3xdr.c | 5 ++---
> > fs/nfsd/nfs4xdr.c | 6 +-----
> > fs/nfsd/nfsfh.h | 14 ++++++++++----
> > 3 files changed, 13 insertions(+), 12 deletions(-)
> >
> > diff --git a/fs/nfsd/nfs3xdr.c b/fs/nfsd/nfs3xdr.c
> > index 2277f83da250..dfbf390ff40c 100644
> > --- a/fs/nfsd/nfs3xdr.c
> > +++ b/fs/nfsd/nfs3xdr.c
> > @@ -291,14 +291,13 @@ void fill_post_wcc(struct svc_fh *fhp)
> > printk("nfsd: inode locked twice during operation.\n");
> >
> >
> >
> >
> > err = fh_getattr(fhp, &fhp->fh_post_attr);
> > - fhp->fh_post_change = nfsd4_change_attribute(&fhp->fh_post_attr,
> > - d_inode(fhp->fh_dentry));
> > if (err) {
> > fhp->fh_post_saved = false;
> > - /* Grab the ctime anyway - set_change_info might use it */
> > fhp->fh_post_attr.ctime = d_inode(fhp->fh_dentry)->i_ctime;
> > } else
> > fhp->fh_post_saved = true;
> > + fhp->fh_post_change = nfsd4_change_attribute(&fhp->fh_post_attr,
> > + d_inode(fhp->fh_dentry));
> > }
> >
> >
> >
> >
> > /*
> > diff --git a/fs/nfsd/nfs4xdr.c b/fs/nfsd/nfs4xdr.c
> > index 833a2c64dfe8..56fd5f6d5c44 100644
> > --- a/fs/nfsd/nfs4xdr.c
> > +++ b/fs/nfsd/nfs4xdr.c
> > @@ -2298,12 +2298,8 @@ static __be32 *encode_change(__be32 *p, struct kstat *stat, struct inode *inode,
> > if (exp->ex_flags & NFSEXP_V4ROOT) {
> > *p++ = cpu_to_be32(convert_to_wallclock(exp->cd->flush_time));
> > *p++ = 0;
> > - } else if (IS_I_VERSION(inode)) {
> > + } else
> > p = xdr_encode_hyper(p, nfsd4_change_attribute(stat, inode));
> > - } else {
> > - *p++ = cpu_to_be32(stat->ctime.tv_sec);
> > - *p++ = cpu_to_be32(stat->ctime.tv_nsec);
> > - }
> > return p;
> > }
> >
> >
> >
> >
> > diff --git a/fs/nfsd/nfsfh.h b/fs/nfsd/nfsfh.h
> > index 56cfbc361561..39d764b129fa 100644
> > --- a/fs/nfsd/nfsfh.h
> > +++ b/fs/nfsd/nfsfh.h
> > @@ -261,10 +261,16 @@ static inline u64 nfsd4_change_attribute(struct kstat *stat,
> > {
> > u64 chattr;
> >
> >
> >
> >
> > - chattr = stat->ctime.tv_sec;
> > - chattr <<= 30;
> > - chattr += stat->ctime.tv_nsec;
> > - chattr += inode_query_iversion(inode);
> > + if (IS_I_VERSION(inode)) {
> > + chattr = stat->ctime.tv_sec;
> > + chattr <<= 30;
> > + chattr += stat->ctime.tv_nsec;
> > + chattr += inode_query_iversion(inode);
> > + } else {
> > + chattr = stat->ctime.tv_sec;
> > + chattr <<= 32;
>
> Might be nice to annotate the shifts above and maybe make them named
> constants. I'm not sure where those values come from, tbh.
This shouldn't be changing the on-the-wire results at all; so
for example the latter is just doing exactly what:
> > - *p++ = cpu_to_be32(stat->ctime.tv_sec);
> > - *p++ = cpu_to_be32(stat->ctime.tv_nsec);
did (after xdr-encoding of the 64-bit chattr).
But yeah maybe it could use some explanation.
The 30-bit shift in the IS_I_VERSION case is weirder, I honestly don't
remember what I was thinking. Maybe just that nanoseconds really only
need 30 bits, and the shift would save a couple high bits in case we
wanted to change the format later without making the change attribute go
backwards. Probably overthinking it!
--b.
