Received-SPF: pass (google.com: best guess record for domain of linux-kernel-owner@vger.kernel.org designates 209.132.180.67 as permitted sender) client-ip=209.132.180.67;
MIME-Version: 1.0
References: <155110348217.21156.3874419272673328527.stgit@devbox>
 <155110354092.21156.13871336589042178985.stgit@devbox> <20190225150603.GE32494@hirez.programming.kicks-ass.net>
 <CAHk-=wgNctC6PoCH9HE3hWFGq+5qBt0nsh+STDguD=jV9ovcag@mail.gmail.com>
In-Reply-To: <CAHk-=wgNctC6PoCH9HE3hWFGq+5qBt0nsh+STDguD=jV9ovcag@mail.gmail.com>
From:   Andy Lutomirski <luto@kernel.org>
Date:   Mon, 25 Feb 2019 10:16:20 -0800
Message-ID: <CALCETrU1dx+Om9k0Q7KbXSVRUpapFWe1bLr7ETm8Rdq=7c9AHw@mail.gmail.com>
Subject: Re: [RFC PATCH 2/4] uaccess: Add non-pagefault user-space read functions
To:     Linus Torvalds <torvalds@linux-foundation.org>
Cc:     Peter Zijlstra <peterz@infradead.org>,
        Masami Hiramatsu <mhiramat@kernel.org>,
        Steven Rostedt <rostedt@goodmis.org>,
        Linux List Kernel Mailing <linux-kernel@vger.kernel.org>,
        Ingo Molnar <mingo@kernel.org>,
        Andrew Morton <akpm@linux-foundation.org>,
        Changbin Du <changbin.du@gmail.com>,
        Jann Horn <jannh@google.com>,
        Kees Cook <keescook@chromium.org>,
        Andy Lutomirski <luto@kernel.org>,
        Alexei Starovoitov <alexei.starovoitov@gmail.com>,
        Nadav Amit <namit@vmware.com>
Content-Type: text/plain; charset="UTF-8"
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

On Mon, Feb 25, 2019 at 9:01 AM Linus Torvalds
<torvalds@linux-foundation.org> wrote:
>
> On Mon, Feb 25, 2019 at 7:06 AM Peter Zijlstra <peterz@infradead.org> wrote:
> >
> > Would something like so work for people?
>
> Looks reasonable to me.
>
> > Why not keep it simple:
> >
> >         mm_segment_t old_fs = get_fs();
> >
> >         set_fs(USER_DS);
> >         ret = __strncpy...();
> >         set_fs(old_fd);
> >
> >         return ret;
>
> So none of this code looks sane. First odd, there's no real reason to
> use __get_user(). The thing should never be used. It does the whole
> stac/clac for every byte.
>
> In the copy_from_user() case, I suggested re-doing it as one common
> routine without the set_fs() dance for the "already there" case to
> simplify error handling. Here it doesn't do that.
>
> But honestly, I think for the strncpy case, we could just do
>
>   long strncpy_from_unsafe_user(char *dst, const void __user *src, long count)
>   {
>       long ret;
>       mm_segment_t old_fs = get_fs();
>
>       set_fs(USER_DS);
>       pagefault_disable();
>       ret = strncpy_from_user(dst, src, count);
>       pagefault_enable();
>       set_fs(old_fs);
>       return ret;
>   }
>
> and be done with it. Efficient and simple.
>
> Note: the above will *only* work for actual user addresses, because
> strncpy_from_user() does that proper range check.
>

Can we also stick the nmi_uaccess_okay() thing in here and kill an
extra bird with the same stone?  Basically, this is "I have no idea
what context I'm in, but I want to try to read a string from current's
address space, so do your best."  In which case, we need to handle the
fact that we might be in KERNEL_DS *and* we need to handle the fact
that CR3 might be wrong.