On Thu, 2017-10-12 at 14:47 +0100, Lorenzo Pieralisi wrote:
> Hi Trond,
>
> On Wed, Oct 11, 2017 at 02:01:34PM -0400, Trond Myklebust wrote:
> > We know that the socket autoclose cannot be queued after we've set
> > the XPRT_LOCKED bit, so the call to cancel_work_sync() is
> > redundant.
> > In addition, it is causing lockdep to complain about a false ABA
> > lock dependency.
> >
> > Signed-off-by: Trond Myklebust <trond.myklebust@xxxxxxxxxxxxxxx>
> > ---
> > net/sunrpc/xprt.c | 1 -
> > 1 file changed, 1 deletion(-)
> >
> > diff --git a/net/sunrpc/xprt.c b/net/sunrpc/xprt.c
> > index e741ec2b4d8e..5f12fe145f02 100644
> > --- a/net/sunrpc/xprt.c
> > +++ b/net/sunrpc/xprt.c
> > @@ -1464,7 +1464,6 @@ static void xprt_destroy(struct rpc_xprt
> > *xprt)
> > rpc_destroy_wait_queue(&xprt->pending);
> > rpc_destroy_wait_queue(&xprt->sending);
> > rpc_destroy_wait_queue(&xprt->backlog);
> > - cancel_work_sync(&xprt->task_cleanup);
>
> This does not make the lockdep warning go away, actually the lockdep
> is triggered by the xs_destroy() cancel_work_sync() call but I do not
> know this code path so I can't really comment on it, let me know if
> there is any specific test I can carry out.
>
> Thanks for looking into this,
> Lorenzo
Sorry for the delay. Does this one help?
8<----------------------------------
>From 528fd3547bad0bdd31c8f987e5bd00c83df8af39 Mon Sep 17 00:00:00 2001
From: Trond Myklebust <trond.myklebust@xxxxxxxxxxxxxxx>
Date: Thu, 19 Oct 2017 12:13:10 -0400
Subject: [PATCH] SUNRPC: Destroy transport from the system workqueue
The transport may need to flush transport connect and receive tasks
that are running on rpciod. In order to do so safely, we need to
ensure that the caller of cancel_work_sync() etc is not itself
running on rpciod.
Do so by running the destroy task from the system workqueue.
Signed-off-by: Trond Myklebust <trond.myklebust@xxxxxxxxxxxxxxx>
---
net/sunrpc/xprt.c | 34 ++++++++++++++++++++++++----------
1 file changed, 24 insertions(+), 10 deletions(-)
diff --git a/net/sunrpc/xprt.c b/net/sunrpc/xprt.c
index 1a39ad14c42f..898485e3ece4 100644
--- a/net/sunrpc/xprt.c
+++ b/net/sunrpc/xprt.c
@@ -1445,6 +1445,23 @@ struct rpc_xprt *xprt_create_transport(struct xprt_create *args)
return xprt;
}
+static void xprt_destroy_cb(struct work_struct *work)
+{
+ struct rpc_xprt *xprt =
+ container_of(work, struct rpc_xprt, task_cleanup);
+
+ rpc_xprt_debugfs_unregister(xprt);
+ rpc_destroy_wait_queue(&xprt->binding);
+ rpc_destroy_wait_queue(&xprt->pending);
+ rpc_destroy_wait_queue(&xprt->sending);
+ rpc_destroy_wait_queue(&xprt->backlog);
+ kfree(xprt->servername);
+ /*
+ * Tear down transport state and free the rpc_xprt
+ */
+ xprt->ops->destroy(xprt);
+}
+
/**
* xprt_destroy - destroy an RPC transport, killing off all requests.
* @xprt: transport to destroy
@@ -1454,22 +1471,19 @@ static void xprt_destroy(struct rpc_xprt *xprt)
{
dprintk("RPC: destroying transport %p\n", xprt);
- /* Exclude transport connect/disconnect handlers */
+ /*
+ * Exclude transport connect/disconnect handlers and autoclose
+ */
wait_on_bit_lock(&xprt->state, XPRT_LOCKED, TASK_UNINTERRUPTIBLE);
del_timer_sync(&xprt->timer);
- rpc_xprt_debugfs_unregister(xprt);
- rpc_destroy_wait_queue(&xprt->binding);
- rpc_destroy_wait_queue(&xprt->pending);
- rpc_destroy_wait_queue(&xprt->sending);
- rpc_destroy_wait_queue(&xprt->backlog);
- cancel_work_sync(&xprt->task_cleanup);
- kfree(xprt->servername);
/*
- * Tear down transport state and free the rpc_xprt
+ * Destroy sockets etc from the system workqueue so they can
+ * safely flush receive work running on rpciod.
*/
- xprt->ops->destroy(xprt);
+ INIT_WORK(&xprt->task_cleanup, xprt_destroy_cb);
+ schedule_work(&xprt->task_cleanup);
}
static void xprt_destroy_kref(struct kref *kref)
--
2.13.6
--
Trond Myklebust
Linux NFS client maintainer, PrimaryData
trond.myklebust@xxxxxxxxxxxxxxx
��.n��������+%������w��{.n�����{��w��� jg��� �����ݢj����G���� ���j:+v���w�m������w�������h�����٥
