linux-kernel - Re: clean up and streamline probe_kernel

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Date:   Wed, 13 May 2020 16:20:17 -0700
From:   Linus Torvalds <torvalds@...ux-foundation.org>
To:     Daniel Borkmann <daniel@...earbox.net>
Cc:     Christoph Hellwig <hch@....de>,
        "the arch/x86 maintainers" <x86@...nel.org>,
        Alexei Starovoitov <ast@...nel.org>,
        Masami Hiramatsu <mhiramat@...nel.org>,
        Andrew Morton <akpm@...ux-foundation.org>,
        linux-parisc@...r.kernel.org,
        linux-um <linux-um@...ts.infradead.org>,
        Netdev <netdev@...r.kernel.org>, bpf@...r.kernel.org,
        Linux-MM <linux-mm@...ck.org>,
        Linux Kernel Mailing List <linux-kernel@...r.kernel.org>
Subject: Re: clean up and streamline probe_kernel_* and friends v2

On Wed, May 13, 2020 at 4:04 PM Daniel Borkmann <daniel@...earbox.net> wrote:
>
> Aside from comments on list, the series looks reasonable to me. For BPF
> the bpf_probe_read() helper would be slightly penalized for probing user
> memory given we now test on copy_from_kernel_nofault() first and if that
> fails only then fall back to copy_from_user_nofault(),

Again, no.

If you can't tell that one or the other is always the right thing,
then that function is simply buggy and wrong.

On sparc and on s390, address X can be _both_ a kernel address and a
user address. You need to specify which it is (by using the proper
function). The whole "try one first, then the other" doesn't work.
They may both "work", and by virtue of that, unless you can state
"yes, we always want user space" or "yes, we always want kernel", that
"try one or the other" isn't valid.

And it can be a real security issue. If a user program can be made to
read kernel memory when BPF validated things as a user pointer, it's
an obvious security issue.

But it can be a security issue the other way around too: if the BPF
code expects to get a kernel string, but user space can fool it into
reading a user string instead by mapping something of its own into the
user space address that aliases the kernel space address, then you can
presumably fool the BPF program to do bad things too (eg mess up any
BPF packet switching routines?).

So BPF really really really needs to specify which one it is. Not
specifying it and saying "whichever" is a bug, and a security issue.

             Linus