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
In-Reply-To: <b4841a84-cf65-f3fb-13b3-ea0a15f47bde@iogearbox.net>
References: <000000000000d48c8e056f5b6c67@google.com> <alpine.DEB.2.21.1806240849150.8650@nanos.tec.linutronix.de>
 <20180624.161411.1560796210597132716.davem@davemloft.net> <20180624100249.GA9493@gmail.com>
 <b4841a84-cf65-f3fb-13b3-ea0a15f47bde@iogearbox.net>
From:   Kees Cook <keescook@chromium.org>
Date:   Tue, 26 Jun 2018 17:26:01 -0700
Message-ID: <CAGXu5jKndtFL7D7T00GmziYaazs7v0t4DWPp_2fdGh2tRnuuEQ@mail.gmail.com>
Subject: Re: set_memory_* (was: Re: BUG: unable to handle kernel paging
 request in bpf_int_jit_compile)
To:     Daniel Borkmann <daniel@iogearbox.net>
Cc:     Ingo Molnar <mingo@kernel.org>, David Miller <davem@davemloft.net>,
        Thomas Gleixner <tglx@linutronix.de>,
        syzbot+a4eb8c7766952a1ca872@syzkaller.appspotmail.com,
        Alexei Starovoitov <ast@kernel.org>,
        "H. Peter Anvin" <hpa@zytor.com>,
        Alexey Kuznetsov <kuznet@ms2.inr.ac.ru>,
        LKML <linux-kernel@vger.kernel.org>,
        Ingo Molnar <mingo@redhat.com>,
        Network Development <netdev@vger.kernel.org>,
        syzkaller-bugs@googlegroups.com, X86 ML <x86@kernel.org>,
        Hideaki YOSHIFUJI <yoshfuji@linux-ipv6.org>,
        Peter Zijlstra <peterz@infradead.org>,
        Laura Abbott <labbott@redhat.com>,
        Linus Torvalds <torvalds@linux-foundation.org>,
        Eric Dumazet <edumazet@google.com>,
        Rik van Riel <riel@redhat.com>,
        Ard Biesheuvel <ard.biesheuvel@linaro.org>
Content-Type: text/plain; charset="UTF-8"
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

On Tue, Jun 26, 2018 at 3:53 PM, Daniel Borkmann <daniel@iogearbox.net> wrote:
> In any case, for pairs like set_memory_ro() + set_memory_rw() that are also used
> outside of bpf e.g. STRICT_MODULE_RWX and friends which are mostly default these
> days for some archs, is the choice to not check errors from there by design or from
> historical context that it originated from 'debugging code' in that sense (DEBUG_RODATA /
> DEBUG_SET_MODULE_RONX) earlier? Also if no-one checks for errors (and if that would
> infact be the recommendation it is agreed upon) should the API be changed to void,
> or generally should actual error checking occur on these + potential rollback; but
> then question is what about restoring part from prior set_memory_ro() via set_memory_rw()?
> Kees/others, do you happen to have some more context on recommended use around this
> by any chance? (Would probably also help if we add some doc around assumptions into
> include/linux/set_memory.h for future users.)

If set_memory_* can fail, I think it needs to be __must_check, and all
the callers need to deal with it gracefully. Those markings aren't
"advisory": they're expected to actually do what they say.

-Kees

-- 
Kees Cook
Pixel Security