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;
MIME-Version: 1.0
References: <20210226183357.28467-1-lhenriques@suse.de> <20210301163324.GC186178@redhat.com>
 <YD0wbmulcBVZ7VZy@suse.de> <20210302160033.GD220334@redhat.com> <20210302162554.GE220334@redhat.com>
In-Reply-To: <20210302162554.GE220334@redhat.com>
From:   Miklos Szeredi <miklos@szeredi.hu>
Date:   Wed, 3 Mar 2021 16:36:24 +0100
Message-ID: <CAJfpegt7U_-gv0QKFF8D9P6DmsHbQ=fNUOBrC17X6Z+O7WzrzQ@mail.gmail.com>
Subject: Re: [RFC PATCH] fuse: Clear SGID bit when setting mode in setacl
To:     Vivek Goyal <vgoyal@redhat.com>
Cc:     Luis Henriques <lhenriques@suse.de>, linux-fsdevel@vger.kernel.org,
        virtio-fs-list <virtio-fs@redhat.com>,
        linux-kernel@vger.kernel.org,
        "Dr. David Alan Gilbert" <dgilbert@redhat.com>
Content-Type: text/plain; charset="UTF-8"
Precedence: bulk

On Tue, Mar 2, 2021 at 5:26 PM Vivek Goyal <vgoyal@redhat.com> wrote:

> > I still feel that it should probably be fixed in virtiofsd, given fuse client
> > is expecting file server to take care of any change of mode (file
> > permission bits).
>
> Havid said that, there is one disadvantage of relying on server to
> do this. Now idmapped mount patches have been merged. If virtiofs
> were to ever support idmapped mounts, this will become an issue.
> Server does not know about idmapped mounts, and it does not have
> information on how to shift inode gid to determine if SGID should
> be cleared or not.
>
> So if we were to keep possible future support of idmapped mounts in mind,
> then solving it in client makes more sense.  (/me is afraid that there
> might be other dependencies like this elsewhere).
>
> Miklos, WDYT.

Hmm sounds like two different modes of operation.

1) shared, non-idmapped: need to take care of races, so do the sgid
clearing in the server
2) non-shared, idmapped: can only do it in client

The same applies to all the other FUSE_*_KILL* stuff, so I guess the
decision about the mode just needs to be tied to a flag in some way.
Not sure if an existing one could be used.

Thanks,
Miklos