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 Received: from lists.ozlabs.org (lists.ozlabs.org [112.213.38.117]) (using TLSv1.2 with cipher ECDHE-RSA-AES256-GCM-SHA384 (256/256 bits)) (No client certificate requested) by smtp.lore.kernel.org (Postfix) with ESMTPS id 0E9B9C433EF for ; Tue, 15 Feb 2022 02:48:35 +0000 (UTC) Received: from boromir.ozlabs.org (localhost [IPv6:::1]) by lists.ozlabs.org (Postfix) with ESMTP id 4JyQWj0gqmz3cY1 for ; Tue, 15 Feb 2022 13:48:33 +1100 (AEDT) Authentication-Results: lists.ozlabs.org; spf=none (no SPF record) smtp.mailfrom=ftp.linux.org.uk (client-ip=2607:5300:60:148a::1; helo=zeniv-ca.linux.org.uk; envelope-from=viro@ftp.linux.org.uk; receiver=) Received: from zeniv-ca.linux.org.uk (zeniv-ca.linux.org.uk [IPv6:2607:5300:60:148a::1]) (using TLSv1.3 with cipher TLS_AES_256_GCM_SHA384 (256/256 bits) key-exchange X25519 server-signature RSA-PSS (2048 bits) server-digest SHA256) (No client certificate requested) by lists.ozlabs.org (Postfix) with ESMTPS id 4JyQWD4QZWz2x9Q for ; Tue, 15 Feb 2022 13:48:07 +1100 (AEDT) Received: from viro by zeniv-ca.linux.org.uk with local (Exim 4.94.2 #2 (Red Hat Linux)) id 1nJnrt-001qwH-NG; Tue, 15 Feb 2022 02:47:05 +0000 Date: Tue, 15 Feb 2022 02:47:05 +0000 From: Al Viro To: Linus Torvalds Subject: Re: [PATCH 04/14] x86: use more conventional access_ok() definition Message-ID: References: <20220214163452.1568807-1-arnd@kernel.org> <20220214163452.1568807-5-arnd@kernel.org>

MIME-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Disposition: inline In-Reply-To: X-BeenThere: linuxppc-dev@lists.ozlabs.org X-Mailman-Version: 2.1.29 Precedence: list List-Id: Linux on PowerPC Developers Mail List List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , Cc: Mark Rutland , Rich Felker , linux-ia64@vger.kernel.org, Linux-sh list , Peter Zijlstra , "open list:BROADCOM NVRAM DRIVER" , Max Filippov , Guo Ren , sparclinux , linux-riscv , Will Deacon , Ard Biesheuvel , linux-arch , linux-s390 , Brian Cain , "open list:QUALCOMM HEXAGON..." , Helge Deller , the arch/x86 maintainers , Russell King - ARM Linux , linux-csky@vger.kernel.org, Christoph Hellwig , Christoph Hellwig , Ingo Molnar , Geert Uytterhoeven , "open list:SYNOPSYS ARC ARCHITECTURE" , "open list:TENSILICA XTENSA PORT \(xtensa\)" , Arnd Bergmann , Heiko Carstens , linux-um , Richard Weinberger , linux-m68k , Openrisc , Greentime Hu , Stafford Horne , Linux ARM , Arnd Bergmann , Michal Simek , Thomas Bogendoerfer , Nick Hu , Parisc List , Linux-MM , Linux API , Linux Kernel Mailing List , Dinh Nguyen , "Eric W . Biederman" , alpha , Andrew Morton , linuxppc-dev , David Miller Errors-To: linuxppc-dev-bounces+linuxppc-dev=archiver.kernel.org@lists.ozlabs.org Sender: "Linuxppc-dev" On Mon, Feb 14, 2022 at 08:17:07PM +0000, Al Viro wrote: > On Mon, Feb 14, 2022 at 12:01:05PM -0800, Linus Torvalds wrote: > > On Mon, Feb 14, 2022 at 11:46 AM Arnd Bergmann wrote: > > > > > > As Al pointed out, they turned out to be necessary on sparc64, but the only > > > definitions are on sparc64 and x86, so it's possible that they serve a similar > > > purpose here, in which case changing the limit from TASK_SIZE to > > > TASK_SIZE_MAX is probably wrong as well. > > > > x86-64 has always(*) used TASK_SIZE_MAX for access_ok(), and the > > get_user() assembler implementation does the same. > > > > I think any __range_not_ok() users that use TASK_SIZE are entirely > > historical, and should be just fixed. > > IIRC, that was mostly userland stack trace collection in perf. > I'll try to dig in archives and see what shows up - it's been > a while ago... After some digging: access_ok() needs only to make sure that MMU won't go anywhere near the kernel page tables; address limit for 32bit threads is none of its concern, so TASK_SIZE_MAX is right for it. valid_user_frame() in arch/x86/events/core.c: used while walking the userland call chain. The reason it's not access_ok() is only that perf_callchain_user() might've been called from interrupt that came while we'd been under KERNEL_DS. That had been back in 2015 and it had been obsoleted since 2017, commit 88b0193d9418 (perf/callchain: Force USER_DS when invoking perf_callchain_user()). We had been guaranteed USER_DS ever since. IOW, it could've reverted to use of access_ok() at any point after that. TASK_SIZE vs TASK_SIZE_MAX is pretty much an accident there - might've been TASK_SIZE_MAX from the very beginning. copy_stack_frame() in arch/x86/kernel/stacktrace.c: similar story, except the commit that made sure callers will have USER_DS - cac9b9a4b083 (stacktrace: Force USER_DS for stack_trace_save_user()) in this case. Also could've been using access_ok() just fine. Amusingly, access_ok() used to be there, until it had been replaced with explicit check on Jul 22 2019 - 4 days after that had been made useless by fix in the caller... copy_from_user_nmi(). That one is a bit more interesting. We have a call chain from perf_output_sample_ustack() (covered by force_uaccess_begin() these days, not that it mattered for x86 now), there's something odd in dumpstack.c:copy_code() (with explicit check for TASK_SIZE_MAX in the caller) and there's a couple of callers in Intel PMU code. AFAICS, there's no reason whatsoever to use TASK_SIZE in that one - the point is to prevent copyin from the kernel memory, and in that respect TASK_SIZE_MAX isn't any worse. The check in copy_code() probably should go. So all of those guys should be simply switched to access_ok(). Might be worth making that a preliminary patch - it's independent from everything else and there's no point folding it into any of the patches in the series.