linux-kernel - Re: [PATCH qemu] x86: don't let decompressed kernel image clobber setup

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: <Y68Zf/MKmX3Rr18E@zn.tnic>
Date:   Fri, 30 Dec 2022 18:01:51 +0100
From:   Borislav Petkov <bp@...en8.de>
To:     "Jason A. Donenfeld" <Jason@...c4.com>
Cc:     "H. Peter Anvin" <hpa@...or.com>, pbonzini@...hat.com,
        ebiggers@...nel.org, x86@...nel.org, linux-kernel@...r.kernel.org,
        qemu-devel@...gnu.org, ardb@...nel.org, kraxel@...hat.com,
        philmd@...aro.org
Subject: Re: [PATCH qemu] x86: don't let decompressed kernel image clobber
 setup_data

On Fri, Dec 30, 2022 at 04:54:27PM +0100, Jason A. Donenfeld wrote:
> > Right, with CONFIG_X86_VERBOSE_BOOTUP=y in a guest here, it says:
> > 
> > early console in extract_kernel
> > input_data: 0x000000000be073a8
> > input_len: 0x00000000008cfc43
> > output: 0x0000000001000000
> > output_len: 0x000000000b600a98
> > kernel_total_size: 0x000000000ac26000
> > needed_size: 0x000000000b800000
> > trampoline_32bit: 0x000000000009d000
> > 
> > so that's a ~9M kernel which gets decompressed at 0x1000000 and the
> > output len is, what, ~180M which looks like plenty to me...
> 
> I think you might have misunderstood the thread.

Maybe...

I've been trying to make sense of all the decompression dancing we're doing and
the diagrams you've drawn and there's a comment over extract_kernel() which
basically says that the compressed image is at the back of the memory buffer

input_data: 0x000000000be073a8

and we decompress to a smaller address

output: 0x0000000001000000

And, it *looks* like setup_data is at an address somewhere >= 0x1000000 so when
we start decompressing, we overwrite it.

I guess extract_kernel() could also dump the setup_data addresses so that we can
verify that...

> First, to reproduce the bug that this patch fixes, you need a kernel with a
> compressed size of around 16 megs, not 9.

Not only that - you need setup_data to be placed somewhere just so that it gets
overwritten during decompression. Also, see below.

> Secondly, that crash is well understood and doesn't need to be reproduced;

Is it?

Where do you get the 0x100000 as the starting address of the kernel image?

Because input_data above says that the input address is somewhere higher...

Btw, here's an allyesconfig:

early console in setup code
No EFI environment detected.
early console in extract_kernel
input_data: 0x000000001bd003a8
input_len: 0x000000000cde2963
output: 0x0000000001000000
output_len: 0x0000000027849d74
kernel_total_size: 0x0000000025830000
needed_size: 0x0000000027a00000
trampoline_32bit: 0x000000000009d000
Physical KASLR using RDRAND RDTSC...
Virtual KASLR using RDRAND RDTSC...

That kernel has compressed size of ~205M and the output buffer is of size ~632M.

> this patch fixes it. Rather, the question now is how to improve this patch
> to remove the 62 meg limit.

Yeah, I'm wondering what that 62M limit is too, actually.

But maybe I'm misunderstanding...

-- 
Regards/Gruss,
    Boris.

https://people.kernel.org/tglx/notes-about-netiquette