[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <1c1a17caf13911270d02ae2fb0147136580f254c.camel@hammerspace.com>
Date: Mon, 17 Mar 2025 21:37:51 +0000
From: Trond Myklebust <trondmy@...merspace.com>
To: "davem@...emloft.net" <davem@...emloft.net>, "chuck.lever@...cle.com"
<chuck.lever@...cle.com>, "pabeni@...hat.com" <pabeni@...hat.com>,
"okorniev@...hat.com" <okorniev@...hat.com>, "tom@...pey.com"
<tom@...pey.com>, "anna@...nel.org" <anna@...nel.org>, "horms@...nel.org"
<horms@...nel.org>, "jlayton@...nel.org" <jlayton@...nel.org>,
"kuba@...nel.org" <kuba@...nel.org>, "Dai.Ngo@...cle.com"
<Dai.Ngo@...cle.com>, "edumazet@...gle.com" <edumazet@...gle.com>,
"neilb@...e.de" <neilb@...e.de>
CC: "josef@...icpanda.com" <josef@...icpanda.com>, "linux-nfs@...r.kernel.org"
<linux-nfs@...r.kernel.org>, "bcodding@...hat.com" <bcodding@...hat.com>,
"linux-kernel@...r.kernel.org" <linux-kernel@...r.kernel.org>,
"netdev@...r.kernel.org" <netdev@...r.kernel.org>
Subject: Re: [PATCH RFC 9/9] sunrpc: don't upgrade passive net reference in
xs_create_sock
On Mon, 2025-03-17 at 17:36 -0400, Jeff Layton wrote:
> On Mon, 2025-03-17 at 21:28 +0000, Trond Myklebust wrote:
> > On Mon, 2025-03-17 at 17:00 -0400, Jeff Layton wrote:
> > > With the move to having sunrpc client xprts not hold active
> > > references
> > > to the net namespace, there is no need to upgrade the socket's
> > > reference
> > > in xs_create_sock. Just keep the passive reference instead.
> > >
> > > Signed-off-by: Jeff Layton <jlayton@...nel.org>
> > > ---
> > > net/sunrpc/xprtsock.c | 3 ---
> > > 1 file changed, 3 deletions(-)
> > >
> > > diff --git a/net/sunrpc/xprtsock.c b/net/sunrpc/xprtsock.c
> > > index
> > > 83cc095846d356f24aed26e2f98525662a6cff1f..0c3d7552f772d6f8477a3ae
> > > d8f0
> > > c513b62cdf589 100644
> > > --- a/net/sunrpc/xprtsock.c
> > > +++ b/net/sunrpc/xprtsock.c
> > > @@ -1941,9 +1941,6 @@ static struct socket *xs_create_sock(struct
> > > rpc_xprt *xprt,
> > > goto out;
> > > }
> > >
> > > - if (protocol == IPPROTO_TCP)
> > > - sk_net_refcnt_upgrade(sock->sk);
> > > -
> > > filp = sock_alloc_file(sock, O_NONBLOCK, NULL);
> > > if (IS_ERR(filp))
> > > return ERR_CAST(filp);
> > >
> >
> > Is this not going to reintroduce the bug described by
> > https://lore.kernel.org/netdev/67b72aeb.050a0220.14d86d.0283.GAE@google.com/T/#u
> > ?
> >
> > As I understand it, the problem has nothing to do with whether or
> > not
> > NFS or the RPC layer holds a reference to the net namespace, but
> > rather
> > whether there are still packets in the socket queues at the time
> > when
> > that net namespace is being freed.
> >
> >
>
> I don't think so. That syzkaller report was closed by this patch:
>
> 5c70eb5c593d net: better track kernel sockets lifetime
>
> That says:
>
> "To fix this, make sure that kernel sockets own a reference on
> net
> passive."
>
> With this, we still do keep a passive reference on the net in the
> socket, which I think is enough.
No. You just removed that by reverting the patch that assigns the
passive reference.
>
> That said, I'd appreciate a look at this from the netdev folks.
--
Trond Myklebust
Linux NFS client maintainer, Hammerspace
trond.myklebust@...merspace.com
Powered by blists - more mailing lists