From mboxrd@z Thu Jan 1 00:00:00 1970 Return-path: From: ebiederm@xmission.com (Eric W. Biederman) References: <20210517203343.3941777-1-arnd@kernel.org> <20210517203343.3941777-2-arnd@kernel.org> Date: Tue, 18 May 2021 11:01:57 -0500 In-Reply-To: (Arnd Bergmann's message of "Tue, 18 May 2021 16:17:53 +0200") Message-ID: MIME-Version: 1.0 Subject: Re: [PATCH v3 1/4] kexec: simplify compat_sys_kexec_load List-Id: List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , Content-Type: text/plain; charset="us-ascii" Content-Transfer-Encoding: 7bit Sender: "kexec" Errors-To: kexec-bounces+dwmw2=infradead.org@lists.infradead.org To: Arnd Bergmann Cc: linux-arch , Christoph Hellwig , Alexander Viro , Andrew Morton , Borislav Petkov , Brian Gerst , Ingo Molnar , "H. Peter Anvin" , Thomas Gleixner , Linux ARM , Linux Kernel Mailing List , Linux-MM , kexec@lists.infradead.org Arnd Bergmann writes: > On Tue, May 18, 2021 at 4:05 PM Arnd Bergmann wrote: >> >> On Tue, May 18, 2021 at 3:41 PM Eric W. Biederman wrote: >> > >> > Arnd Bergmann writes: >> > >> > > From: Arnd Bergmann KEXEC_ARCH_DEFAULT >> > > >> > > The compat version of sys_kexec_load() uses compat_alloc_user_space to >> > > convert the user-provided arguments into the native format. >> > > >> > > Move the conversion into the regular implementation with >> > > an in_compat_syscall() check to simplify it and avoid the >> > > compat_alloc_user_space() call. >> > > >> > > compat_sys_kexec_load() now behaves the same as sys_kexec_load(). >> > >> > Nacked-by: "Eric W. Biederman" >> >KEXEC_ARCH_DEFAULT >> > The patch is wrong. >> > >> > The logic between the compat entry point and the ordinary entry point >> > are by necessity different. This unifies the logic and breaks the compat >> > entry point. >> > >> > The fundamentally necessity is that the code being loaded needs to know >> > which mode the kernel is running in so it can safely transition to the >> > new kernel. >> > >> > Given that the two entry points fundamentally need different logic, >> > and that difference was not preserved and the goal of this patchset >> > was to unify that which fundamentally needs to be different. I don't >> > think this patch series makes any sense for kexec. >> >> Sorry, I'm not following that explanation. Can you clarify what different >> modes of the kernel you are referring to here, and how my patch >> changes this? > > I think I figured it out now myself after comparing the two functions: > > --- a/kernel/kexec.c > +++ b/kernel/kexec.c > @@ -269,7 +269,8 @@ SYSCALL_DEFINE4(kexec_load, unsigned long, entry, > unsigned long, nr_segments, > > /* Verify we are on the appropriate architecture */ > if (((flags & KEXEC_ARCH_MASK) != KEXEC_ARCH) && > - ((flags & KEXEC_ARCH_MASK) != KEXEC_ARCH_DEFAULT)) > + (in_compat_syscall() || > + ((flags & KEXEC_ARCH_MASK) != KEXEC_ARCH_DEFAULT))) > return -EINVAL; > > /* Because we write directly to the reserved memory > > Not sure if that's the best way of doing it, but it looks like folding this > in restores the current behavior. Yes. That is pretty much all there is. I personally can't stand the sight of in_compat_syscall() doubly so when you have to lie to the type system with casts. The cognitive dissonance I experience is extreme. I will be happy to help you find another way to get rid of compat_alloc_user, but not that way. There is a whole mess in there that was introduced when someone added do_kexec_load while I was napping in 2017 that makes the system calls an absolute mess. It all needs to be cleaned up. Eric _______________________________________________ kexec mailing list kexec@lists.infradead.org http://lists.infradead.org/mailman/listinfo/kexec