Merge to Fedora kernel-2.6.18-1.2224_FC5 patched with stable patch-2.6.18.1-vs2.0...
[linux-2.6.git] / net / sunrpc / clnt.c
index aa8965e..3e19d32 100644 (file)
@@ -125,10 +125,9 @@ rpc_new_client(struct rpc_xprt *xprt, char *servname,
                goto out_err;
 
        err = -ENOMEM;
-       clnt = kmalloc(sizeof(*clnt), GFP_KERNEL);
+       clnt = kzalloc(sizeof(*clnt), GFP_KERNEL);
        if (!clnt)
                goto out_err;
-       memset(clnt, 0, sizeof(*clnt));
        atomic_set(&clnt->cl_users, 0);
        atomic_set(&clnt->cl_count, 1);
        clnt->cl_parent = clnt;
@@ -184,8 +183,7 @@ rpc_new_client(struct rpc_xprt *xprt, char *servname,
 
 out_no_auth:
        if (!IS_ERR(clnt->cl_dentry)) {
-               rpc_rmdir(clnt->cl_pathname);
-               dput(clnt->cl_dentry);
+               rpc_rmdir(clnt->cl_dentry);
                rpc_put_mount();
        }
 out_no_path:
@@ -252,10 +250,8 @@ rpc_clone_client(struct rpc_clnt *clnt)
        new->cl_autobind = 0;
        new->cl_oneshot = 0;
        new->cl_dead = 0;
-       if (!IS_ERR(new->cl_dentry)) {
+       if (!IS_ERR(new->cl_dentry))
                dget(new->cl_dentry);
-               rpc_get_mount();
-       }
        rpc_init_rtt(&new->cl_rtt_default, clnt->cl_xprt->timeout.to_initval);
        if (new->cl_auth)
                atomic_inc(&new->cl_auth->au_count);
@@ -318,11 +314,15 @@ rpc_destroy_client(struct rpc_clnt *clnt)
                clnt->cl_auth = NULL;
        }
        if (clnt->cl_parent != clnt) {
+               if (!IS_ERR(clnt->cl_dentry))
+                       dput(clnt->cl_dentry);
                rpc_destroy_client(clnt->cl_parent);
                goto out_free;
        }
-       if (clnt->cl_pathname[0])
-               rpc_rmdir(clnt->cl_pathname);
+       if (!IS_ERR(clnt->cl_dentry)) {
+               rpc_rmdir(clnt->cl_dentry);
+               rpc_put_mount();
+       }
        if (clnt->cl_xprt) {
                xprt_destroy(clnt->cl_xprt);
                clnt->cl_xprt = NULL;
@@ -332,10 +332,6 @@ rpc_destroy_client(struct rpc_clnt *clnt)
 out_free:
        rpc_free_iostats(clnt->cl_metrics);
        clnt->cl_metrics = NULL;
-       if (!IS_ERR(clnt->cl_dentry)) {
-               dput(clnt->cl_dentry);
-               rpc_put_mount();
-       }
        kfree(clnt);
        return 0;
 }
@@ -922,26 +918,43 @@ call_transmit(struct rpc_task *task)
        task->tk_status = xprt_prepare_transmit(task);
        if (task->tk_status != 0)
                return;
+       task->tk_action = call_transmit_status;
        /* Encode here so that rpcsec_gss can use correct sequence number. */
        if (rpc_task_need_encode(task)) {
-               task->tk_rqstp->rq_bytes_sent = 0;
+               BUG_ON(task->tk_rqstp->rq_bytes_sent != 0);
                call_encode(task);
                /* Did the encode result in an error condition? */
                if (task->tk_status != 0)
-                       goto out_nosend;
+                       return;
        }
-       task->tk_action = call_transmit_status;
        xprt_transmit(task);
        if (task->tk_status < 0)
                return;
-       if (!task->tk_msg.rpc_proc->p_decode) {
-               task->tk_action = rpc_exit_task;
-               rpc_wake_up_task(task);
-       }
-       return;
-out_nosend:
-       /* release socket write lock before attempting to handle error */
-       xprt_abort_transmit(task);
+       /*
+        * On success, ensure that we call xprt_end_transmit() before sleeping
+        * in order to allow access to the socket to other RPC requests.
+        */
+       call_transmit_status(task);
+       if (task->tk_msg.rpc_proc->p_decode != NULL)
+               return;
+       task->tk_action = rpc_exit_task;
+       rpc_wake_up_task(task);
+}
+
+/*
+ * 5a. Handle cleanup after a transmission
+ */
+static void
+call_transmit_status(struct rpc_task *task)
+{
+       task->tk_action = call_status;
+       /*
+        * Special case: if we've been waiting on the socket's write_space()
+        * callback, then don't call xprt_end_transmit().
+        */
+       if (task->tk_status == -EAGAIN)
+               return;
+       xprt_end_transmit(task);
        rpc_task_force_reencode(task);
 }
 
@@ -993,18 +1006,7 @@ call_status(struct rpc_task *task)
 }
 
 /*
- * 6a. Handle transmission errors.
- */
-static void
-call_transmit_status(struct rpc_task *task)
-{
-       if (task->tk_status != -EAGAIN)
-               rpc_task_force_reencode(task);
-       call_status(task);
-}
-
-/*
- * 6b. Handle RPC timeout
+ * 6a. Handle RPC timeout
  *     We do not release the request slot, so we keep using the
  *     same XID for all retransmits.
  */
@@ -1179,6 +1181,17 @@ call_verify(struct rpc_task *task)
        u32     *p = iov->iov_base, n;
        int error = -EACCES;
 
+       if ((task->tk_rqstp->rq_rcv_buf.len & 3) != 0) {
+               /* RFC-1014 says that the representation of XDR data must be a
+                * multiple of four bytes
+                * - if it isn't pointer subtraction in the NFS client may give
+                *   undefined results
+                */
+               printk(KERN_WARNING
+                      "call_verify: XDR representation not a multiple of"
+                      " 4 bytes: 0x%x\n", task->tk_rqstp->rq_rcv_buf.len);
+               goto out_eio;
+       }
        if ((len -= 3) < 0)
                goto out_overflow;
        p += 1; /* skip XID */