linux-kernel - [PATCH 4.19 36/44] [STABLE PATCH] SUNRPC: Always drop the XPRT_LOCK on XPRT_CLOSE

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

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-Id: <20190213183654.822481720@linuxfoundation.org>
Date:   Wed, 13 Feb 2019 19:38:37 +0100
From:   Greg Kroah-Hartman <gregkh@...uxfoundation.org>
To:     linux-kernel@...r.kernel.org, stable@...r.kernel.org
Cc:     Greg Kroah-Hartman <gregkh@...uxfoundation.org>,
        Dave Wysochanski <dwysocha@...hat.com>,
        Trond Myklebust <trondmy@...merspace.com>,
        Benjamin Coddington <bcodding@...hat.com>
Subject: [PATCH 4.19 36/44] [STABLE PATCH] SUNRPC: Always drop the XPRT_LOCK on XPRT_CLOSE_WAIT

4.19-stable review patch.  If anyone has any objections, please let me know.

------------------

From: Benjamin Coddington <bcodding@...hat.com>

This patch is only appropriate for stable kernels v4.16 - v4.19

Since commit 9b30889c548a ("SUNRPC: Ensure we always close the socket after
a connection shuts down"), and until commit c544577daddb ("SUNRPC: Clean up
transport write space handling"), it is possible for the NFS client to spin
in the following tight loop:

269.964083: rpc_task_run_action: task:43@0 flags=5a81 state=0005 status=0 action=call_bind [sunrpc]
269.964083: rpc_task_run_action: task:43@0 flags=5a81 state=0005 status=0 action=call_connect [sunrpc]
269.964083: rpc_task_run_action: task:43@0 flags=5a81 state=0005 status=0 action=call_transmit [sunrpc]
269.964085: xprt_transmit: peer=[10.0.1.82]:2049 xid=0x761d3f77 status=-32
269.964085: rpc_task_run_action: task:43@0 flags=5a81 state=0005 status=-32 action=call_transmit_status [sunrpc]
269.964085: rpc_task_run_action: task:43@0 flags=5a81 state=0005 status=-32 action=call_status [sunrpc]
269.964085: rpc_call_status: task:43@0 status=-32

The issue is that the path through call_transmit_status does not release
the XPRT_LOCK when the transmit result is -EPIPE, so the socket cannot be
properly shut down.

The below commit fixed things up in mainline by unconditionally calling
xprt_end_transmit() and releasing the XPRT_LOCK after every pass through
call_transmit.  However, the entirety of this commit is not appropriate for
stable kernels because its original inclusion was part of a series that
modifies the sunrpc code to use a different queueing model.  As a result,
there are machinations within this patch that are not needed for a stable
fix and will not make sense without a larger backport of the mainline
series.

In this patch, we take the slightly modified bit of the mainline patch
below, which is to release the XPRT_LOCK on transmission error should we
detect that the transport is waiting to close.

commit c544577daddb618c7dd5fa7fb98d6a41782f020e upstream
Author: Trond Myklebust <trond.myklebust@...merspace.com>
Date:   Mon Sep 3 23:39:27 2018 -0400

    SUNRPC: Clean up transport write space handling

    Treat socket write space handling in the same way we now treat transport
    congestion: by denying the XPRT_LOCK until the transport signals that it
    has free buffer space.

    Signed-off-by: Trond Myklebust <trond.myklebust@...merspace.com>

The original discussion of the problem is here:

    https://lore.kernel.org/linux-nfs/20181212135157.4489-1-dwysocha@redhat.com/T/#t

This passes my usual cthon and xfstests on NFS as applied on v4.19 mainline.

Reported-by: Dave Wysochanski <dwysocha@...hat.com>
Suggested-by: Trond Myklebust <trondmy@...merspace.com>
Signed-off-by: Benjamin Coddington <bcodding@...hat.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@...uxfoundation.org>
---
 include/linux/sunrpc/xprt.h |    5 +++++
 net/sunrpc/clnt.c           |    6 ++++--
 2 files changed, 9 insertions(+), 2 deletions(-)

--- a/include/linux/sunrpc/xprt.h
+++ b/include/linux/sunrpc/xprt.h
@@ -443,6 +443,11 @@ static inline int xprt_test_and_set_conn
 	return test_and_set_bit(XPRT_CONNECTING, &xprt->state);
 }
 
+static inline int xprt_close_wait(struct rpc_xprt *xprt)
+{
+	return test_bit(XPRT_CLOSE_WAIT, &xprt->state);
+}
+
 static inline void xprt_set_bound(struct rpc_xprt *xprt)
 {
 	test_and_set_bit(XPRT_BOUND, &xprt->state);
--- a/net/sunrpc/clnt.c
+++ b/net/sunrpc/clnt.c
@@ -1992,13 +1992,15 @@ call_transmit(struct rpc_task *task)
 static void
 call_transmit_status(struct rpc_task *task)
 {
+	struct rpc_xprt *xprt = task->tk_rqstp->rq_xprt;
 	task->tk_action = call_status;
 
 	/*
 	 * Common case: success.  Force the compiler to put this
-	 * test first.
+	 * test first.  Or, if any error and xprt_close_wait,
+	 * release the xprt lock so the socket can close.
 	 */
-	if (task->tk_status == 0) {
+	if (task->tk_status == 0 || xprt_close_wait(xprt)) {
 		xprt_end_transmit(task);
 		rpc_task_force_reencode(task);
 		return;