From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: X-Spam-Checker-Version: SpamAssassin 3.4.0 (2014-02-07) on aws-us-west-2-korg-lkml-1.web.codeaurora.org X-Spam-Level: X-Spam-Status: No, score=-2.3 required=3.0 tests=DKIM_INVALID,DKIM_SIGNED, HEADER_FROM_DIFFERENT_DOMAINS,MAILING_LIST_MULTI,SPF_PASS,USER_AGENT_MUTT autolearn=ham autolearn_force=no version=3.4.0 Received: from mail.kernel.org (mail.kernel.org [198.145.29.99]) by smtp.lore.kernel.org (Postfix) with ESMTP id CD111C43381 for ; Thu, 28 Feb 2019 17:46:25 +0000 (UTC) Received: from vger.kernel.org (vger.kernel.org [209.132.180.67]) by mail.kernel.org (Postfix) with ESMTP id 9F453218AE for ; Thu, 28 Feb 2019 17:46:25 +0000 (UTC) Authentication-Results: mail.kernel.org; dkim=fail reason="signature verification failed" (2048-bit key) header.d=infradead.org header.i=@infradead.org header.b="Gf1b45VQ" Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S2387934AbfB1RqY (ORCPT ); Thu, 28 Feb 2019 12:46:24 -0500 Received: from merlin.infradead.org ([205.233.59.134]:48598 "EHLO merlin.infradead.org" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1732473AbfB1RqX (ORCPT ); Thu, 28 Feb 2019 12:46:23 -0500 DKIM-Signature: v=1; a=rsa-sha256; q=dns/txt; c=relaxed/relaxed; d=infradead.org; s=merlin.20170209; h=In-Reply-To:Content-Type:MIME-Version: References:Message-ID:Subject:Cc:To:From:Date:Sender:Reply-To: Content-Transfer-Encoding:Content-ID:Content-Description:Resent-Date: Resent-From:Resent-Sender:Resent-To:Resent-Cc:Resent-Message-ID:List-Id: List-Help:List-Unsubscribe:List-Subscribe:List-Post:List-Owner:List-Archive; bh=1tIRHaljqGjsTO1apj67lwRFzWZnSUB4T1+xsg6UD7c=; b=Gf1b45VQf+eom9qcjvjELnBgk RcZXJUq9jiBZ+fcUB73RGOHANCIX71hjazFHtWz2CAeE3Lb8PJmykQ53+rSRk7cwLVxsbNeTcfSTs 91Eh26lFT3SYY8hHf4q5zSzZetk1MSlj9K/O0+zpi4BALwXTJarclL3bKt4oK9OJoF+Edz0E9Plmw 1lbyiSERaPL6PyreFqddKdbm7+DlmVBjhNNYo4s9aIPSQB8k4ZSelmw4EsDHuai1U3H5pVlMKJZtd iIfozKsDAqkSvPxLwpOi32SjbOJ/TwPLSUoZruVymDMWOxh1n0RMSxuLX0NJY3Q5cGkmDtLLkmk4y yMRuBQjsw==; Received: from j217100.upc-j.chello.nl ([24.132.217.100] helo=hirez.programming.kicks-ass.net) by merlin.infradead.org with esmtpsa (Exim 4.90_1 #2 (Red Hat Linux)) id 1gzPl2-0007Mj-0p; Thu, 28 Feb 2019 17:46:08 +0000 Received: by hirez.programming.kicks-ass.net (Postfix, from userid 1000) id 834C8202C2DA7; Thu, 28 Feb 2019 18:46:05 +0100 (CET) Date: Thu, 28 Feb 2019 18:46:05 +0100 From: Peter Zijlstra To: Dmitry Vyukov Cc: Linus Torvalds , Thomas Gleixner , "H. Peter Anvin" , Julien Thierry , Will Deacon , Andy Lutomirski , Ingo Molnar , Catalin Marinas , James Morse , valentin.schneider@arm.com, Brian Gerst , Josh Poimboeuf , Andy Lutomirski , Borislav Petkov , Denys Vlasenko , LKML , Andrey Ryabinin Subject: Re: [PATCH 1/8] kasan,x86: Frob kasan_report() in an exception Message-ID: <20190228174605.GF32494@hirez.programming.kicks-ass.net> References: <20190228145450.289603901@infradead.org> <20190228150152.078767622@infradead.org> <20190228154551.GE32494@hirez.programming.kicks-ass.net> MIME-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Disposition: inline In-Reply-To: User-Agent: Mutt/1.10.1 (2018-07-13) Sender: linux-kernel-owner@vger.kernel.org Precedence: bulk List-ID: X-Mailing-List: linux-kernel@vger.kernel.org On Thu, Feb 28, 2019 at 05:03:09PM +0100, Dmitry Vyukov wrote: > I am missing some knowledge about SMAP to answer this. > In short, these tools insert lots of callbacks into runtime for memory > accesses, function entry/exit, atomicops and some other. These > callbacks can do things of different complexity. > Humm... perhaps we could just disable SMAP for KMSAN/KTSAN. It's > possible, right? If we have it enabled with KASAN, that should be > enough. SMAP detects access to _PAGE_USER pages; that is, such access is only allowed when EFLAGS.AC=1, otherwise they'll fault. I again don't know enough about KASAN to say if it does that; but I suspect it only tracks kernel memory state. > Also, what's the actual problem with KASAN+SMAP? Is it warnings from > static analysis tool? Or there are also some runtime effects? What > effects? Both; so because of the above semantics, things like copy_to_user() will have to do STAC (set EFLAGS.AC=1), then do the actual copies to the user addresses, and then CLAC (clear the AC flag again). The desire is to have AC=1 sections as small as possible, such that as much code as possible is ran with AC=0 and will trap on unintended accesses. Also; the scheduler doesn't (but I have a patch for that, but I'd prefer to not have to use it) context switch EFLAGS. This means that if we land in the scheduler while AC=1, the next task will resume with AC=1. Consequently, if that task returns to userspace before it gets scheduled again, we'll continue our previous task (that left with AC=1) with AC=0 and it'll then fault where no fault were expected. Anyway; the objtool annotation basically tracks the EFLAGS.AC state (through STAC/CLAC instructions -- no PUSHF/POPF) and disallows any CALL/RET while AC=1. This is where the __asan_{load,store}*() stuff went *splat*. GCC inserts those calls in the middle of STAC/CLAC (AC=1) and we then have to mark the functions as AC-safe. objtool validates those on the same rules, no further CALLs that are not also safe. Things like __fentry__ are inherently unsafe because they use preempt_disable/preempt_enable, where the latter has a CALL __preempt_schedule (and is thus very unsafe). Similarly with kasan_report(), it does all sorts of things that are not safe to do. > Is it possible to disable the SMAP runtime checks once we enter > kasan_report() past report_enabled() check? We could restrict it to > "just finish printing this bug report whatever it takes and then > whatever" if it makes things simpler. > It would be nice if we could restrict it to something like: > > @@ -291,6 +303,7 @@ void kasan_report(unsigned long addr, size_t size, > if (likely(!report_enabled())) > return; > + disable_smap(); > > And then enforce panic at the end of report if smap is enabled. That would be a CLAC, and the current rules disallow CLAC for AC-safe functions. Furthermore, kasan_report() isn't fatal, right? So it would have to restore the state on exit. That makes the validation state much more complicated. Let me try and frob some of the report_enabled() stuff before the #UD.