net/sunrpc/clnt.c:720:36: sparse: dereference of noderef expression

Hi Mel,

There are new sparse warnings show up in

tree:   git://git.kernel.org/pub/scm/linux/kernel/git/next/linux-next.git akpm
head:   247109fa112ac912f31efc28e130ed2a0cc1764c
commit: 5bb176c86a868289d5d93d89919c3b5a339d6d20 [147/308] nfs: enable swap on NFS

All sparse warnings:

net/sunrpc/clnt.c:1362:39: sparse: incorrect type in initializer (different address spaces)
net/sunrpc/clnt.c:1362:39: expected struct rpc_xprt *xprt
net/sunrpc/clnt.c:1362:39: got struct rpc_xprt [noderef] <asn:4>*cl_xprt
net/sunrpc/clnt.c:1954:59: sparse: incorrect type in argument 1 (different address spaces)
net/sunrpc/clnt.c:1954:59: expected struct rpc_xprt *xprt
net/sunrpc/clnt.c:1954:59: got struct rpc_xprt [noderef] <asn:4>*cl_xprt
net/sunrpc/clnt.c:1470:39: sparse: incorrect type in initializer (different address spaces)
net/sunrpc/clnt.c:1470:39: expected struct rpc_xprt *xprt
net/sunrpc/clnt.c:1470:39: got struct rpc_xprt [noderef] <asn:4>*cl_xprt
net/sunrpc/clnt.c:1648:40: sparse: incorrect type in argument 1 (different address spaces)
net/sunrpc/clnt.c:1648:40: expected struct rpc_wait_queue *<noident>
net/sunrpc/clnt.c:1648:40: got struct rpc_wait_queue [noderef] <asn:4>*<noident>
net/sunrpc/clnt.c:1748:51: sparse: incorrect type in argument 1 (different address spaces)
net/sunrpc/clnt.c:1748:51: expected struct rpc_xprt *xprt
net/sunrpc/clnt.c:1748:51: got struct rpc_xprt [noderef] <asn:4>*cl_xprt
net/sunrpc/clnt.c:1800:59: sparse: incorrect type in argument 1 (different address spaces)
net/sunrpc/clnt.c:1800:59: expected struct rpc_xprt *xprt
net/sunrpc/clnt.c:1800:59: got struct rpc_xprt [noderef] <asn:4>*cl_xprt
net/sunrpc/clnt.c:1564:39: sparse: incorrect type in initializer (different address spaces)
net/sunrpc/clnt.c:1564:39: expected struct rpc_xprt *xprt
net/sunrpc/clnt.c:1564:39: got struct rpc_xprt [noderef] <asn:4>*cl_xprt
net/sunrpc/clnt.c:1968:46: sparse: incorrect type in argument 1 (different address spaces)
net/sunrpc/clnt.c:1968:46: expected struct rpc_xprt *xprt
net/sunrpc/clnt.c:1968:46: got struct rpc_xprt [noderef] <asn:4>*cl_xprt
+ net/sunrpc/clnt.c:720:36: sparse: dereference of noderef expression net/sunrpc/clnt.c:1851:25: sparse: context imbalance in 'call_timeout' - different lock contexts for basic block

vim +720 net/sunrpc/clnt.c
717 atomic_inc(&clnt->cl_count);
718 if (clnt->cl_softrtry)
719 task->tk_flags |= RPC_TASK_SOFT;

721 task->tk_flags |= RPC_TASK_SWAPPER;
722 /* Add to the client's list of all tasks */
723 spin_lock(&clnt->cl_lock);

---
0-DAY kernel build testing backend Open Source Technology Centre
Fengguang Wu <[email protected]> Intel Corporation

Thanks (adding Andrew to cc for picking up in mm)

How about the following?

---8<---

sparsefix: nfs: enable swap on NFS

Fengguang Wu reports the following sparse problem

net/sunrpc/clnt.c:720:36: sparse: dereference of noderef expression

task->tk_client->cl_xprt is RCU protected and must be dereferenced using rcu_dereference(). This patch checks if a socket is being used for swap and if so it takes the RCU read lock and dereferences properly. This is a fix for "nfs: enable swap on NFS"

---
 net/sunrpc/clnt.c |   11 +++++++++--
 1 file changed, 9 insertions(+), 2 deletions(-)

index 2848e5b..df27890 100644
@@ -717,8 +717,15 @@ void rpc_task_set_client(struct rpc_task *task, struct rpc_clnt *clnt)
 		atomic_inc(&clnt->cl_count);
 		if (clnt->cl_softrtry)
 			task->tk_flags |= RPC_TASK_SOFT;
-		if (task->tk_client->cl_xprt->swapper)
-			task->tk_flags |= RPC_TASK_SWAPPER;
+		if (sk_memalloc_socks()) {
+			struct rpc_xprt *xprt;
+
+			rcu_read_lock();
+			xprt = rcu_dereference(clnt->cl_xprt);
+			if (xprt->swapper)
+				task->tk_flags |= RPC_TASK_SWAPPER;
+			rcu_read_unlock();
+		}
 		/* Add to the client's list of all tasks */
 		spin_lock(&clnt->cl_lock);
 		list_add_tail(&task->tk_task, &clnt->cl_tasks);

Looks good to me. And it does fix the warning.

I assume the fix will be folded into the original patch, otherwise the subject will look a bit strange ;)

Thanks,
Fengguang

That's what I'm expecting. I've sent a few fixes marked "buildfix:
patch subject" so Andrew knows exactly which patches to fold together. Ordinarily when dealing with just mmotm I would specify the filename in the quilt series but that does not make as much sense when talking about linux-next.

Mel Gorman
SUSE Labs

Ah yes! That title is actually pretty readable and precise for Andrew.

Thanks,
Fengguang