Received-SPF: pass (google.com: domain of linux-nfs-owner@vger.kernel.org designates 23.128.96.18 as permitted sender) client-ip=23.128.96.18;
MIME-Version: 1.0
References: <20210519160635.333057-1-Anna.Schumaker@Netapp.com> <9b54027028a9a7322b6f06748c4499174a238866.camel@hammerspace.com>
In-Reply-To: <9b54027028a9a7322b6f06748c4499174a238866.camel@hammerspace.com>
From:   Anna Schumaker <schumaker.anna@gmail.com>
Date:   Wed, 19 May 2021 12:47:23 -0400
Message-ID: <CAFX2Jf=m6nFNs4tfHXp4dyLpSK3k9Fpn11YDx1qJE5U1d-7CNw@mail.gmail.com>
Subject: Re: [PATCH] NFSv4: Fix a NULL pointer dereference in pnfs_mark_matching_lsegs_return()
To:     Trond Myklebust <trondmy@hammerspace.com>
Cc:     "linux-nfs@vger.kernel.org" <linux-nfs@vger.kernel.org>
Content-Type: text/plain; charset="UTF-8"
Precedence: bulk

On Wed, May 19, 2021 at 12:43 PM Trond Myklebust
<trondmy@hammerspace.com> wrote:
>
> On Wed, 2021-05-19 at 12:06 -0400, schumaker.anna@gmail.com wrote:
> > From: Anna Schumaker <Anna.Schumaker@Netapp.com>
> >
> > Commit de144ff4234f changes _pnfs_return_layout() to call
> > pnfs_mark_matching_lsegs_return() passing NULL as the struct
> > pnfs_layout_range argument. Unfortunately,
> > pnfs_mark_matching_lsegs_return() doesn't check if we have a value
> > here
> > before dereferencing it, causing an oops.
> >
> > I'm able to hit this crash consistently when running connectathon
> > basic
> > tests on NFS v4.1/v4.2 against Ontap.
> >
> > Fixes: de144ff4234f ("NFSv4: Don't discard segments marked for return
> > in _pnfs_return_layout()")
> > Cc: stable@vger.kernel.org
> > Signed-off-by: Anna Schumaker <Anna.Schumaker@Netapp.com>
> > ---
> >  fs/nfs/pnfs.c | 4 ++--
> >  1 file changed, 2 insertions(+), 2 deletions(-)
> >
> > diff --git a/fs/nfs/pnfs.c b/fs/nfs/pnfs.c
> > index 03e0b34c4a64..6d720afb7b70 100644
> > --- a/fs/nfs/pnfs.c
> > +++ b/fs/nfs/pnfs.c
> > @@ -2484,12 +2484,12 @@ pnfs_mark_matching_lsegs_return(struct
> > pnfs_layout_hdr *lo,
> >                         set_bit(NFS_LSEG_LAYOUTRETURN, &lseg-
> > >pls_flags);
> >                 }
> >
> > -       if (remaining) {
> > +       if (remaining && return_range) {
> >                 pnfs_set_plh_return_info(lo, return_range->iomode,
> > seq);
> >                 return -EBUSY;
> >         }
> >
> > -       if (!list_empty(&lo->plh_return_segs)) {
> > +       if (return_range && !list_empty(&lo->plh_return_segs)) {
> >                 pnfs_set_plh_return_info(lo, return_range->iomode,
> > seq);
> >                 return 0;
>
> This patch would mean we fail to mark the layout for return in
> situations where we clearly should be doing so. The lack of a
> return_range doesn't indicate that we don't want to return any layouts,
> but rather that we want to return all layouts.
>
> I suggest rather changing the caller, _pnfs_return_layout() to move
> that declaration of 'struct pnfs_layout_range range' out of the if
> statement so that it can be passed as an argument instead of NULL.

Sure, I'll try that now.

Thanks!
Anna

>
> >         }
>
> --
> Trond Myklebust
> Linux NFS client maintainer, Hammerspace
> trond.myklebust@hammerspace.com
>
>