linux-kernel - Re: [PATCH 1/1] system call notification with self

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-ID: <20080910162008.GA401@tv-sign.ru>
Date:	Wed, 10 Sep 2008 20:20:08 +0400
From:	Oleg Nesterov <oleg@...sign.ru>
To:	Pierre Morel <pmorel@...ux.vnet.ibm.com>
Cc:	Andrew Morton <akpm@...ux-foundation.org>,
	linux-kernel@...r.kernel.org, Roland McGrath <roland@...hat.com>,
	Heiko Carstens <heicars2@...ux.vnet.ibm.com>,
	sameske@...ux.vnet.ibm.com,
	Martin Schwidefsky <schwidefsky@...ibm.com>,
	Ingo Molnar <mingo@...e.hu>, gregkh@...e.de,
	uml-devel <user-mode-linux-devel@...ts.sourceforge.net>,
	Dave Hansen <dave@...ux.vnet.ibm.com>,
	Cedric Le Goater <clg@...ibm.com>,
	Daniel Lezcano <dlezcano@...ibm.com>
Subject: Re: [PATCH 1/1] system call notification with self_ptrace

On 09/10, Pierre Morel wrote:
>
> Oleg Nesterov wrote:
> >
> >I still think this patch shouldn't change handle_signal().
> >
> >Once again. The signal handler for SIGSYS can first do
> >sys_ptrace(PTRACE_SELF_OFF) (which is filtered out), and then use any
> >other syscall, so this change is not needed, afaics.
> >
> Yes it can but what if the application forget to do it?
> It is a security so that the application do not bounce for ever.

The (buggy) task can be killed, this has nothing to do with security.

>From the security pov, this case doesn't differ from, say,

	void sigh(int sig)
	{
		kill(getpid(), sig);
	}

	void main(void)
	{
		signal(SIGSYS, sigh);
		kill(getpid(), SIGSYS);
	}

Or I missed something?

> >So, PTRACE_SELF_XXX disables the "normal" ptrace. Not sure this is good.
> >
> I think that having two tracing system one over the other may be
> quite difficult to handle.

Yes I see.

But... well, I think we need Roland's opinion. I must admit, I am a bit
sceptical about this patch ;) I mean, I don't really understand why it
is useful. We can do the same with fork() + ptrace(). Yes, in that
case we need an "extra" context switch for any traced syscall. But,
do you have any "real life" example to demonstrate that the user-space
solution sucks? We can even use CLONE_MM to speedup the context switch.

Pierre, don't get me wrong. I never used debuggers for myself, I will
be happy to know I am wrong. I just don't understand.

As for ->instrumentation. If you are going to remove PTS_INSTRUMENTED,
we need only one bit. We could use PF_PTS_SELF, but ->flags is already
"contended". Perhaps you can do something like

	--- include/linux/sched.h
	+++ include/linux/sched.h
	@@ -1088,6 +1088,7 @@ struct task_struct {
		/* ??? */
		unsigned int personality;
		unsigned did_exec:1;
	+	unsigned pts_self:1;
		pid_t pid;
		pid_t tgid;

Both did_exec and pts_self can only be changed by current, so it is
safe to share the same word. This way we don't enlarge task_struct.

Oleg.

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/