Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 23.128.96.18 as permitted sender) client-ip=23.128.96.18;
Date:   Sun, 23 Jan 2022 20:18:36 -0800 (PST)
From:   Hugh Dickins <hughd@google.com>
To:     Aleksa Sarai <cyphar@cyphar.com>
cc:     Kir Kolyshkin <kolyshkin@gmail.com>, linux-man@vger.kernel.org,
        Michael Kerrisk <mtk.manpages@gmail.com>,
        David Herrmann <dh.herrmann@gmail.com>,
        Andrew Morton <akpm@linux-foundation.org>,
        Hugh Dickins <hughd@google.com>,
        Mike Kravetz <mike.kravetz@oracle.com>, linux-mm@kvack.org,
        linux-kernel@vger.kernel.org
Subject: Re: [PATCH] fcntl.2: document F_GET_SEALS on tmpfs peculiarity
In-Reply-To: <20220122090441.ktxh43lpgsd2dxj4@senku>
Message-ID: <d7ca2658-b63b-7437-9bd0-82bc59c7c981@google.com>
References: <20220122005251.1441343-1-kolyshkin@gmail.com> <20220122090441.ktxh43lpgsd2dxj4@senku>
MIME-Version: 1.0
Content-Type: text/plain; charset=US-ASCII
Precedence: bulk

On Sat, 22 Jan 2022, Aleksa Sarai wrote:

> Adding the maintainers of mm/{shmem,memfd}.c and fs/hugetlbfs/ just in
> case this was not intended behaviour.

Kir is correct - thanks - and it is intended behaviour.  Not consciously
intended to make for a difficult manpage, but the implementation was
intended to be simple, so tmpfs and hugetlbfs do not internally
distinguish memfd objects from filesystem files - their filesystem
files simply start off with F_SEAL_SEAL to rule out any sealing.

> 
> On 2022-01-21, Kir Kolyshkin <kolyshkin@gmail.com> wrote:
> > Currently, from the description of file sealing it can be deduced that
> > unless the fd is a memfd, all sealing operations fail with EINVAL.
> > 
> > Apparently, it's not true for tmpfs or hugetlbfs -- F_GET_SEALS returns
> > 1 (F_SEAL_SEAL) for an fd opened on these filesystems (probably because
> > those are used to back memfd files).
> > 
> > Fix the description to mention that peculiarity. Not knowing this can
> > result in incorrect code logic (see [1], where the code mistook a
> > descriptor of a file opened on on tmpfs for a memfd).
> > 
> > While at it, clarify that fcntl does not actually return EINVAL, but
> > sets errno to it (as it is usually said elsewhere).
> > 
> > [1] https://github.com/opencontainers/runc/pull/3342
> > 
> > Cc: Aleksa Sarai <cyphar@cyphar.com>
> > Cc: David Herrmann <dh.herrmann@gmail.com>
> > Signed-off-by: Kir Kolyshkin <kolyshkin@gmail.com>

Acked-by: Hugh Dickins <hughd@google.com>

> > ---
> >  man2/fcntl.2 | 17 +++++++++++++++--
> >  1 file changed, 15 insertions(+), 2 deletions(-)
> > 
> > diff --git a/man2/fcntl.2 b/man2/fcntl.2
> > index 7b5604e3a..f951b05ff 100644
> > --- a/man2/fcntl.2
> > +++ b/man2/fcntl.2
> > @@ -1402,10 +1402,23 @@ file seals can be applied only to a file descriptor returned by
> >  (if the
> >  .B MFD_ALLOW_SEALING
> >  was employed).
> > -On other filesystems, all
> > +On all other filesystems, except
> > +.BR tmpfs (5)
> > +and
> > +.BR hugetlbfs ,
> > +all
> >  .BR fcntl ()
> > -operations that operate on seals will return
> > +operations that operate on seals will fail with
> > +.I errno
> > +set to
> >  .BR EINVAL .
> > +For a descriptor from a file on
> > +.BR tmpfs (5)
> > +or
> > +.BR hugetlbfs ,
> > +.B F_GET_SEALS
> > +returns
> > +.BR F_SEAL_SEAL .
> >  .PP
> >  Seals are a property of an inode.
> >  Thus, all open file descriptors referring to the same inode share
> > -- 
> > 2.33.1
> > 
> 
> -- 
> Aleksa Sarai
> Senior Software Engineer (Containers)
> SUSE Linux GmbH
> <https://www.cyphar.com/>