| lists.openwall.net | lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC | |
|
Open Source and information security mailing list archives
| ||
|
Date: Fri, 18 Sep 2015 18:00:07 -0400
From: Trond Myklebust <trond.myklebust@...marydata.com>
To: "Suzuki K. Poulose" <Suzuki.Poulose@....com>,
Jeff Layton <jlayton@...chiereds.net>
Cc: Anna Schumaker <anna.schumaker@...app.com>,
"J. Bruce Fields" <bfields@...ldses.org>,
"David S. Miller" <davem@...emloft.net>,
"linux-nfs@...r.kernel.org" <linux-nfs@...r.kernel.org>,
"linux-kernel@...r.kernel.org" <linux-kernel@...r.kernel.org>,
Marc Zyngier <Marc.Zyngier@....com>
Subject: Re: [PATCHv2] SUNRPC: Fix a race in xs_reset_transport
On Fri, 2015-09-18 at 12:51 -0400, Trond Myklebust wrote:
> On Fri, 2015-09-18 at 12:19 +0100, Suzuki K. Poulose wrote:
> > On 16/09/15 12:17, Jeff Layton wrote:
> > > On Wed, 16 Sep 2015 10:35:49 +0100
> > > "Suzuki K. Poulose" <suzuki.poulose@....com> wrote:
> > >
> > > > From: "Suzuki K. Poulose" <suzuki.poulose@....com>
> > > >
> >
> > ...
> >
> > > > + write_unlock_bh(&sk->sk_callback_lock);
> > > > + return;
> > > > + }
> > > > + sock = transport->sock;
> > > > +
> > > > transport->inet = NULL;
> > > > transport->sock = NULL;
> > > >
> > > > @@ -833,6 +838,10 @@ static void xs_reset_transport(struct
> > > > sock_xprt *transport)
> > > > xs_restore_old_callbacks(transport, sk);
> > > > xprt_clear_connected(xprt);
> > > > write_unlock_bh(&sk->sk_callback_lock);
> > > > +
> > > > + if (sock)
> > > > + kernel_sock_shutdown(sock, SHUT_RDWR);
> > > > +
> > > > xs_sock_reset_connection_flags(xprt);
> > > >
> > > > trace_rpc_socket_close(xprt, sock);
> > >
> > > Better, but now I'm wondering...is it problematic to restore the
> > > old
> > > callbacks before calling kernel_sock_shutdown? I can't quite tell
> > > whether it matters in all cases.
> > >
> > > It might be best to just go ahead and take the spinlock twice
> > > here.
> > > Do
> > > it once to clear the transport->sock pointer, call
> > > kernel_sock_shutdown, and then take it again to restore the old
> > > callbacks, etc.
> > >
> > > I don't know though...I get the feeling there are races all over
> > > the
> > > place in this code. It seems like there's a similar one wrt to
> > > the
> > > transport->inet pointer. It seems a little silly that we clear it
> > > under
> > > the sk->sk_callback_lock. You have to dereference that pointer
> > > in order to get to the lock.
> > >
> > > Maybe the right solution is to use an xchg to swap the inet
> > > pointer
> > > with NULL so it can act as a gatekeeper. Whoever gets there first
> > > does
> > > the rest of the shutdown.
> > >
> > > Something like this maybe? Would this also fix the original
> > > problem?
> > > Note that this patch is untested...
> > >
> > > [PATCH] sunrpc: use xchg to fetch and clear the transport->inet
> > > pointer in xs_reset_transport
> > >
> > > Reported-by: "Suzuki K. Poulose" <Suzuki.Poulose@....com>
> > > Signed-off-by: Jeff Layton <jeff.layton@...marydata.com>
> > > ---
> > > net/sunrpc/xprtsock.c | 4 ++--
> > > 1 file changed, 2 insertions(+), 2 deletions(-)
> > >
> > > diff --git a/net/sunrpc/xprtsock.c b/net/sunrpc/xprtsock.c
> > > index 7be90bc1a7c2..57f79dcab493 100644
> > > --- a/net/sunrpc/xprtsock.c
> > > +++ b/net/sunrpc/xprtsock.c
> > > @@ -813,9 +813,10 @@ static void xs_error_report(struct sock *sk)
> > > static void xs_reset_transport(struct sock_xprt *transport)
> > > {
> > > struct socket *sock = transport->sock;
> > > - struct sock *sk = transport->inet;
> > > + struct sock *sk;
> > > struct rpc_xprt *xprt = &transport->xprt;
> > >
> > > + sk = xchg(&transport->inet, NULL);
> > > if (sk == NULL)
> > > return;
> > >
> > > @@ -825,7 +826,6 @@ static void xs_reset_transport(struct
> > > sock_xprt
> > > *transport)
> > > kernel_sock_shutdown(sock, SHUT_RDWR);
> > >
> > > write_lock_bh(&sk->sk_callback_lock);
> > > - transport->inet = NULL;
> > > transport->sock = NULL;
> > >
> > > sk->sk_user_data = NULL;
> > >
> >
> >
> > This one seemed to fix it, so if it matters :
> >
> > Tested-by: Suzuki K. Poulose <suzuki.poulose@....com>
>
>
> I don't think it does. It addresses a symptom, but the actual problem
> is that we're running 2 parallel close() calls on the same socket
> during a shutdown. That must not happen because it means we have
> something else trying to use the socket while it is being freed.
>
> I think what is happening is that we're triggering the socket
> autoclose
> mechanism from the state change callback. You're seeing the problem
> more frequently because we added the call to kernel_sock_shutdown()
> as
> part of the socket shutdown, but AFAICS, it could still be triggered
> from some external event such as a server-initiated shutdown that
> happened at the same time.
> In fact, looking at the code, it could even be triggered from the
> data
> receive side of things.
> Both of these things are bad, because autoclose puts the transport
> struct that is being freed onto a workqueue. That again can lead to a
> really bad use-after-free situation if the timing is just a little
> different.
>
> So how about the following patch? It should apply cleanly on top of
> the
> first one (which is still needed, btw).
Having thought some more about this, I think the safest thing in order
to avoid races is simply to have the shutdown set XPRT_LOCKED. That way
we can keep the current desirable behaviour of closing the socket
automatically any time the server initiates a close, while still
preventing it during shutdown.
8<-------------------------------------------------------------------
>From 3e1c9d8092e2fa4509d84a00fcf21e7e0c581fe2 Mon Sep 17 00:00:00 2001
From: Trond Myklebust <trond.myklebust@...marydata.com>
Date: Fri, 18 Sep 2015 15:53:24 -0400
Subject: [PATCH] SUNRPC: Lock the transport layer on shutdown
Avoid all races with the connect/disconnect handlers by taking the
transport lock.
Signed-off-by: Trond Myklebust <trond.myklebust@...marydata.com>
---
net/sunrpc/xprt.c | 6 ++++++
1 file changed, 6 insertions(+)
diff --git a/net/sunrpc/xprt.c b/net/sunrpc/xprt.c
index ab5dd621ae0c..2e98f4a243e5 100644
--- a/net/sunrpc/xprt.c
+++ b/net/sunrpc/xprt.c
@@ -614,6 +614,7 @@ static void xprt_autoclose(struct work_struct *work)
clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
xprt->ops->close(xprt);
xprt_release_write(xprt, NULL);
+ wake_up_bit(&xprt->state, XPRT_LOCKED);
}
/**
@@ -723,6 +724,7 @@ void xprt_unlock_connect(struct rpc_xprt *xprt, void *cookie)
xprt->ops->release_xprt(xprt, NULL);
out:
spin_unlock_bh(&xprt->transport_lock);
+ wake_up_bit(&xprt->state, XPRT_LOCKED);
}
/**
@@ -1394,6 +1396,10 @@ out:
static void xprt_destroy(struct rpc_xprt *xprt)
{
dprintk("RPC: destroying transport %p\n", xprt);
+
+ /* Exclude transport connect/disconnect handlers */
+ wait_on_bit_lock(&xprt->state, XPRT_LOCKED, TASK_UNINTERRUPTIBLE);
+
del_timer_sync(&xprt->timer);
rpc_xprt_debugfs_unregister(xprt);
--
2.4.3
--
Trond Myklebust
Linux NFS client maintainer, PrimaryData
trond.myklebust@...marydata.com
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/
Powered by blists - more mailing lists