From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1755362Ab1GJQSS (ORCPT ); Sun, 10 Jul 2011 12:18:18 -0400 Received: from mail-pz0-f46.google.com ([209.85.210.46]:42332 "EHLO mail-pz0-f46.google.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1753477Ab1GJQSO convert rfc822-to-8bit (ORCPT ); Sun, 10 Jul 2011 12:18:14 -0400 MIME-Version: 1.0 In-Reply-To: References: <1310282825-14509-1-git-send-email-lacombar@gmail.com> <1310282825-14509-2-git-send-email-lacombar@gmail.com> Date: Sun, 10 Jul 2011 12:18:13 -0400 Message-ID: Subject: Re: [PATCH 2/2] kconfig/nconf: prevent segfault on empty menu From: Arnaud Lacombe To: Nir Tzachar Cc: linux-kbuild@vger.kernel.org, Michal Marek , linux-kernel@vger.kernel.org, Andrej Gelenberg Content-Type: text/plain; charset=ISO-8859-1 Content-Transfer-Encoding: 8BIT Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org Hi, On Sun, Jul 10, 2011 at 6:32 AM, Nir Tzachar wrote: > Hello. > > On Sun, Jul 10, 2011 at 10:27 AM, Arnaud Lacombe wrote: >> nconf does not check the validity of the current menu when help is requested >> (with either , '?' or 'h'). This leads to a NULL pointer dereference when an >> empty menu is encountered. >> >> The following reduced testcase exposes the problem: >> >> config DEP >>        bool >> >> menu "FOO" >> >> config BAR >>        bool "BAR" >>        depends on DEP >> >> endmenu >> >> Issue will happen when entering menu "FOO" and requesting help. >> >> nconf is the only front-end which do not filter the validity of the current >> menu. Such filter can not really happen beforehand as other key which does not >> deals with the current menu might be entered by the user, so just bails out >> earlier if we encounter an invalid menu. > > I do not believe the correct solution is silently ignoring the user's > key presses. I think the correct behavior would be to display a > nohelp_text window. > > something like this (untested): > show_scroll_win(main_window, nohelp_text); > This would be: show_scroll_win(main_window, _(menu_get_prompt(menu)), nohelp_text); but we would fall back on the original issue as requesting the prompt of an invalid menu does not make sense, so the menu_get_prompt(menu) would still have to be replaced with something valid. It would be nice to come up with a solution valid at least for nconf and mconf. Currently mconf just silently discard the key pressed. To some extend I do not think any frontend should display a menu which has no visible child, but that may require more changes and affect performance... - Arnaud > The other patches in the series seem gr8. > > Cheers. > >> Cc: Nir Tzachar >> Cc: Andrej Gelenberg >> Reported-by: Andrej Gelenberg >> Signed-off-by: Arnaud Lacombe >> --- >>  scripts/kconfig/nconf.c |    7 ++++++- >>  1 files changed, 6 insertions(+), 1 deletions(-) >> >> diff --git a/scripts/kconfig/nconf.c b/scripts/kconfig/nconf.c >> index 24fc79a..eb9e49d 100644 >> --- a/scripts/kconfig/nconf.c >> +++ b/scripts/kconfig/nconf.c >> @@ -1222,7 +1222,12 @@ static void conf_message_callback(const char *fmt, va_list ap) >> >>  static void show_help(struct menu *menu) >>  { >> -       struct gstr help = str_new(); >> +       struct gstr help; >> + >> +       if (!menu) >> +               return; >> + >> +       help = str_new(); >>        menu_get_ext_help(menu, &help); >>        show_scroll_win(main_window, _(menu_get_prompt(menu)), str_get(&help)); >>        str_free(&help); >> -- >> 1.7.3.4.574.g608b.dirty >> >> >