From: Jeff Layton <jlayton@redhat.com>
Date: Thu, 12 Jun 2008 09:08:11 -0400
Subject: [fs] cifs: wait on kthread_stop before thread exits
Message-id: 20080612090811.339c89f9@barsoom.rdu.redhat.com
O-Subject: Re: [RHEL5.3 PATCH] BZ#444865: [CIFS] don't allow demultiplex thread to exit until kthread_stop is called
Bugzilla: 444865
RH-Acked-by: Peter Staubach <staubach@redhat.com>

cifs_demultiplex_thread can exit under several conditions:

1) if it's signaled
2) if there's a problem with session setup
3) if kthread_stop is called on it

The first two are problems. If kthread_stop is called on the thread,
there is no guarantee that it will still be up. We need to have the
thread stay up until kthread_stop is called on it.

One option would be to not even try to tear things down until after
kthread_stop is called. However, in the case where there is a problem
setting up the session, there's no real reason to try continuing the
loop.

This patch allows the thread to continue to exit its main loop under all
three conditions, but it has the thread go to sleep until kthread_stop
is called. This allows us to simplify the shutdown code somewhat since
we can be reasonably sure that the thread won't exit after being
signaled but before kthread_stop is called.

It also removes the places where the thread itself set the tsk variable
since it appeared that it could have a potential race where the thread
might never be shut down.

This patch differs slightly from the one originally proposed. If the
server simply closes the connection in the middle of session setup,
then the mount process and cifsd can deadlock, both of them sleeping
and waiting for the other to wake them. To fix this, the wake_up_all()
call was added to wake up the mount process before cifsd goes to sleep.

diff --git a/fs/cifs/connect.c b/fs/cifs/connect.c
index 1396f37..b0b4ce0 100644
--- a/fs/cifs/connect.c
+++ b/fs/cifs/connect.c
@@ -375,7 +375,6 @@ cifs_demultiplex_thread(struct TCP_Server_Info *server)
 #endif
 
 	current->flags |= PF_MEMALLOC;
-	server->tsk = current;	/* save process info to wake at shutdown */
 	cFYI(1, ("Demultiplex PID: %d", current->pid));
 	write_lock(&GlobalSMBSeslock);
 	atomic_inc(&tcpSesAllocCount);
@@ -701,14 +700,25 @@ multi_t2_fnd:
 #endif /* CIFS_DEBUG2 */
 
 		}
-	} /* end while !EXITING */
+	} /* end while !kthread_should_stop() */
 
 	spin_lock(&GlobalMid_Lock);
 	server->tcpStatus = CifsExiting;
-	server->tsk = NULL;
+	spin_unlock(&GlobalMid_Lock);
+	wake_up_all(&server->response_q);
+
+	/* don't exit until kthread_stop is called */
+	set_current_state(TASK_UNINTERRUPTIBLE);
+	while (!kthread_should_stop()) {
+		schedule();
+		set_current_state(TASK_UNINTERRUPTIBLE);
+	}
+	set_current_state(TASK_RUNNING);
+
 	/* check if we have blocked requests that need to free */
 	/* Note that cifs_max_pending is normally 50, but
 	can be set at module install time to as little as two */
+	spin_lock(&GlobalMid_Lock);
 	if (atomic_read(&server->inFlight) >= cifs_max_pending)
 		atomic_set(&server->inFlight, cifs_max_pending - 1);
 	/* We do not want to set the max_pending too low or we
@@ -2237,15 +2247,12 @@ cifs_mount(struct super_block *sb, struct cifs_sb_info *cifs_sb,
 			srvTcp->tcpStatus = CifsExiting;
 			spin_unlock(&GlobalMid_Lock);
 			if (srvTcp->tsk) {
-				struct task_struct *tsk;
 				/* If we could verify that kthread_stop would
 				   always wake up processes blocked in
 				   tcp in recv_mesg then we could remove the
 				   send_sig call */
 				force_sig(SIGKILL, srvTcp->tsk);
-				tsk = srvTcp->tsk;
-				if (tsk)
-					kthread_stop(tsk);
+				kthread_stop(srvTcp->tsk);
 			}
 		}
 		 /* If find_unc succeeded then rc == 0 so we can not end */
@@ -2261,23 +2268,17 @@ cifs_mount(struct super_block *sb, struct cifs_sb_info *cifs_sb,
 					if ((temp_rc == -ESHUTDOWN) &&
 					    (pSesInfo->server) &&
 					    (pSesInfo->server->tsk)) {
-						struct task_struct *tsk;
 						force_sig(SIGKILL,
 							pSesInfo->server->tsk);
-						tsk = pSesInfo->server->tsk;
-						if (tsk)
-							kthread_stop(tsk);
+						kthread_stop(pSesInfo->server->tsk);
 					}
 				} else {
 					cFYI(1, ("No session or bad tcon"));
 					if ((pSesInfo->server) &&
 					    (pSesInfo->server->tsk)) {
-						struct task_struct *tsk;
 						force_sig(SIGKILL,
 							pSesInfo->server->tsk);
-						tsk = pSesInfo->server->tsk;
-						if (tsk)
-							kthread_stop(tsk);
+						kthread_stop(pSesInfo->server->tsk);
 					}
 				}
 				sesInfoFree(pSesInfo);