smb3: fix unmount hang in open_shroot

An earlier patch "CIFS: fix deadlock in cached root handling"
did not completely address the deadlock in open_shroot. This
patch addresses the deadlock.

In testing the recent patch:
  smb3: improve handling of share deleted (and share recreated)
we were able to reproduce the open_shroot deadlock to one
of the target servers in unmount in a delete share scenario.

Fixes: 7e5a70ad ("CIFS: fix deadlock in cached root handling")

This is version 2 of this patch. An earlier version of this
patch "smb3: fix unmount hang in open_shroot" had a problem
found by Dan.
Reported-by: kbuild test robot <lkp@intel.com>
Reported-by: Dan Carpenter <dan.carpenter@oracle.com>
Suggested-by: Pavel Shilovsky <pshilov@microsoft.com>
Reviewed-by: Pavel Shilovsky <pshilov@microsoft.com>
Signed-off-by: Steve French <stfrench@microsoft.com>
CC: Aurelien Aptel <aaptel@suse.com>
CC: Stable <stable@vger.kernel.org>

fs/cifs/smb2ops.c

@@ -658,6 +658,15 @@ int open_shroot(unsigned int xid, struct cifs_tcon *tcon, struct cifs_fid *pfid)
 		return 0;
 	}
 
+	/*
+	 * We do not hold the lock for the open because in case
+	 * SMB2_open needs to reconnect, it will end up calling
+	 * cifs_mark_open_files_invalid() which takes the lock again
+	 * thus causing a deadlock
+	 */
+
+	mutex_unlock(&tcon->crfid.fid_mutex);
+
 	if (smb3_encryption_required(tcon))
 		flags |= CIFS_TRANSFORM_REQ;
 
@@ -679,7 +688,7 @@ int open_shroot(unsigned int xid, struct cifs_tcon *tcon, struct cifs_fid *pfid)
 
 	rc = SMB2_open_init(tcon, &rqst[0], &oplock, &oparms, &utf16_path);
 	if (rc)
-		goto oshr_exit;
+		goto oshr_free;
 	smb2_set_next_command(tcon, &rqst[0]);
 
 	memset(&qi_iov, 0, sizeof(qi_iov));
@@ -692,18 +701,10 @@ int open_shroot(unsigned int xid, struct cifs_tcon *tcon, struct cifs_fid *pfid)
 				  sizeof(struct smb2_file_all_info) +
 				  PATH_MAX * 2, 0, NULL);
 	if (rc)
-		goto oshr_exit;
+		goto oshr_free;
 
 	smb2_set_related(&rqst[1]);
 
-	/*
-	 * We do not hold the lock for the open because in case
-	 * SMB2_open needs to reconnect, it will end up calling
-	 * cifs_mark_open_files_invalid() which takes the lock again
-	 * thus causing a deadlock
-	 */
-
-	mutex_unlock(&tcon->crfid.fid_mutex);
 	rc = compound_send_recv(xid, ses, flags, 2, rqst,
 				resp_buftype, rsp_iov);
 	mutex_lock(&tcon->crfid.fid_mutex);