[PATCH] 9p/trans_fd: avoid sending req to a cancelled conn

[Sishuai Gong <sishuai.system@gmail.com>]

When a connection is cancelled by p9_conn_cancel(), all requests on it
should be cancelled---mark req->status as REQ_STATUS_ERROR. However,
because a race over m->err between p9_conn_cancel() and p9_fd_request(),
p9_fd_request might see the old value of m->err, think that the connection
is NOT cancelled, and then add new requests to this cancelled connection.

Fixing this issue by lock-protecting the check on m->err.

Signed-off-by: Sishuai Gong <sishuai.system@gmail.com>
---
 net/9p/trans_fd.c | 8 ++++++--
 1 file changed, 6 insertions(+), 2 deletions(-)

diff --git a/net/9p/trans_fd.c b/net/9p/trans_fd.c
index 00b684616e8d..e43a850f5190 100644
--- a/net/9p/trans_fd.c
+++ b/net/9p/trans_fd.c
@@ -671,10 +671,14 @@ static int p9_fd_request(struct p9_client *client, struct p9_req_t *req)

 	p9_debug(P9_DEBUG_TRANS, "mux %p task %p tcall %p id %d\n",
 		 m, current, &req->tc, req->tc.id);
-	if (m->err < 0)
-		return m->err;

 	spin_lock(&m->req_lock);
+
+	if (m->err < 0) {
+		spin_unlock(&m->req_lock);
+		return m->err;
+	}
+
 	WRITE_ONCE(req->status, REQ_STATUS_UNSENT);
 	list_add_tail(&req->req_list, &m->unsent_req_list);
 	spin_unlock(&m->req_lock);
—