linux-kernel - Re: [PATCH bpf] bpf: invalidate unused part of bpf_prog

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]

Message-ID: <8F788446-899C-4BA3-8236-612A94D98582@fb.com>
Date:   Thu, 21 Apr 2022 22:51:52 +0000
From:   Song Liu <songliubraving@...com>
To:     Linus Torvalds <torvalds@...ux-foundation.org>,
        Peter Zijlstra <peterz@...radead.org>
CC:     Song Liu <song@...nel.org>,
        Alexei Starovoitov <alexei.starovoitov@...il.com>,
        bpf <bpf@...r.kernel.org>, Linux-MM <linux-mm@...ck.org>,
        Linux Kernel Mailing List <linux-kernel@...r.kernel.org>,
        Alexei Starovoitov <ast@...nel.org>,
        Daniel Borkmann <daniel@...earbox.net>,
        Kernel Team <Kernel-team@...com>,
        Andrew Morton <akpm@...ux-foundation.org>,
        "Edgecombe, Rick P" <rick.p.edgecombe@...el.com>,
        Christoph Hellwig <hch@...radead.org>,
        Andrii Nakryiko <andrii@...nel.org>
Subject: Re: [PATCH bpf] bpf: invalidate unused part of bpf_prog_pack

Hi Linus,

> On Apr 21, 2022, at 3:30 PM, Linus Torvalds <torvalds@...ux-foundation.org> wrote:
> 
> On Thu, Apr 21, 2022 at 2:53 PM Song Liu <songliubraving@...com> wrote:
>> 
>> However, we cannot really use the same function at free time. The
>> huge page is RO+X at free time, but we are only zeroing out a chunk
>> of it. So regular memset/memcpy won’t work. Instead, we will need
>> something like bpf_arch_text_copy().
> 
> I actually think bpf_arch_text_copy() is another horribly badly done thing.
> 
> It seems only implemented on x86 (I'm not sure how anything else is
> supposed to work, I didn't go look), and there it is horribly badly
> done, using __text_poke() that does all these magical things just to
> make it atomic wrt concurrent code execution.
> 
> None of which is *AT*ALL* relevant for this case, since concurrent
> code execution simply isn't a thing (and if it were, you would already
> have lost).
> 
> And if that wasn't pointless enough, it does all that magic "map the
> page writably at a different virtual address using poking_addr in
> poking_mm" and a different address space entirely.
> 
> All of that is required for REAL KERNEL CODE.
> 
> But the thing is, for bpf_prog_pack, all of that is just completely
> pointless and stupid complexity.
> 
> We already *have* the other non-executable address that is writable:
> it's the actual pages that got vmalloc'ed. Just use vmalloc_to_page()
> and it's RIGHT THERE.

I think this won’t work, as set_memory_ro makes all the aliases of 
these pages read only. We can probably add set_memory_ro_noalias(), 
which will be similar to set_memory_np_noalias(). This approach was 
NACKed by Peter[1]. So we went with the bpf_arch_text_copy approach.

If we can loosen the W^X requirement for BPF programs, other parts 
of bpf_prog_pack could also be simplified. 

Thanks,
Song

[1] https://lore.kernel.org/netdev/20211116080051.GU174703@worktop.programming.kicks-ass.net/

> 
> At which point you just use the same bpf_jit_fill_hole() function, and
> you're done.
> 
> In other words, all of this seems excessively stupidly done, for no
> good reason.  It's only making it much too complicated, and just not
> doing the right thing at all.
> 
>                Linus