SUNRPC: Move fault injection call sites

I've hit some crashes that occur in the xprt_rdma_inject_disconnect
path. It appears that, for some provides, rdma_disconnect() can
take so long that the transport can disconnect and release its
hardware resources while rdma_disconnect() is still running,
resulting in a UAF in the provider.

The transport's fault injection method may depend on the stability
of transport data structures. That means it needs to be invoked
only from contexts that hold the transport write lock.

Fixes: 4a0682583988 ("SUNRPC: Transport fault injection")
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Trond Myklebust <trond.myklebust@hammerspace.com>

diff --git a/net/sunrpc/clnt.c b/net/sunrpc/clnt.c
index 612f0a6..c2a0112 100644 (file)
--- a/net/sunrpc/clnt.c
+++ b/net/sunrpc/clnt.c
@@ -1799,7 +1799,6 @@ call_allocate(struct rpc_task *task)
 
        status = xprt->ops->buf_alloc(task);
        trace_rpc_buf_alloc(task, status);
-       xprt_inject_disconnect(xprt);
        if (status == 0)
                return;
        if (status != -ENOMEM) {

diff --git a/net/sunrpc/xprt.c b/net/sunrpc/xprt.c
index a853f75..80c94ea 100644 (file)
--- a/net/sunrpc/xprt.c
+++ b/net/sunrpc/xprt.c
@@ -1485,7 +1485,10 @@ bool xprt_prepare_transmit(struct rpc_task *task)
 
 void xprt_end_transmit(struct rpc_task *task)
 {
-       xprt_release_write(task->tk_rqstp->rq_xprt, task);
+       struct rpc_xprt *xprt = task->tk_rqstp->rq_xprt;
+
+       xprt_inject_disconnect(xprt);
+       xprt_release_write(xprt, task);
 }
 
 /**
@@ -1887,7 +1890,6 @@ void xprt_release(struct rpc_task *task)
        spin_unlock(&xprt->transport_lock);
        if (req->rq_buffer)
                xprt->ops->buf_free(task);
-       xprt_inject_disconnect(xprt);
        xdr_free_bvec(&req->rq_rcv_buf);
        xdr_free_bvec(&req->rq_snd_buf);
        if (req->rq_cred != NULL)

diff --git a/net/sunrpc/xprtrdma/transport.c b/net/sunrpc/xprtrdma/transport.c
index 78d29d1..0995359 100644 (file)
--- a/net/sunrpc/xprtrdma/transport.c
+++ b/net/sunrpc/xprtrdma/transport.c
@@ -262,8 +262,10 @@ xprt_rdma_connect_worker(struct work_struct *work)
  * xprt_rdma_inject_disconnect - inject a connection fault
  * @xprt: transport context
  *
- * If @xprt is connected, disconnect it to simulate spurious connection
- * loss.
+ * If @xprt is connected, disconnect it to simulate spurious
+ * connection loss. Caller must hold @xprt's send lock to
+ * ensure that data structures and hardware resources are
+ * stable during the rdma_disconnect() call.
  */
 static void
 xprt_rdma_inject_disconnect(struct rpc_xprt *xprt)