linux-kernel - Re: [PATCH v2][next] sysctl: Avoid open coded arithmetic in memory allocator functions

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:   Sun, 24 Oct 2021 11:13:28 +0200
From:   Len Baker <len.baker@....com>
To:     Matthew Wilcox <willy@...radead.org>,
        "Gustavo A. R. Silva" <gustavoars@...nel.org>,
        Kees Cook <keescook@...omium.org>
Cc:     Len Baker <len.baker@....com>,
        Luis Chamberlain <mcgrof@...nel.org>,
        Iurii Zaikin <yzaikin@...gle.com>,
        linux-hardening@...r.kernel.org, linux-fsdevel@...r.kernel.org,
        linux-kernel@...r.kernel.org
Subject: Re: [PATCH v2][next] sysctl: Avoid open coded arithmetic in memory
 allocator functions

Hi Matthew,

thanks for looking at this. More below.

On Sat, Oct 23, 2021 at 03:27:17PM +0100, Matthew Wilcox wrote:
> On Sat, Oct 23, 2021 at 12:54:14PM +0200, Len Baker wrote:
> > Changelog v1 -> v2
> > - Remove the new_dir_size function and its use (Matthew Wilcox).
>
> Why do you think the other functions are any different?  Please
> provide reasoning.

I think it is better to be defensive. IMHO I believe that if the
struct_size() helper could be used in this patch, it would be more
easy to ACK. But it is not possible due to the complex memory
layouts. However, there are a lot of code in the kernel that uses the
struct_size() helper for memory allocator arguments where we know
that it don't overflow. For example:

1.- Function imx8mm_tmu_probe()
    Uses: struct_size(tmu, sensors, data->num_sensors)
    Where: tmu has a sizeof(struct imx8mm_tmu) -> Not very big
           data->num_sensors -> A little number

    So, almost certainly it doesn't overflow.

2.- Function igb_alloc_q_vector()
    Uses: struct_size(q_vector, ring, ring_count)
    Where: q_vector has a sizeof(struct igb_q_vector) -> Not very big
           ring_count -> At most two.

    So, almost certainly it doesn't overflow.

3.- And so on...

So, I think that these new functions for the size calculation are
helpers like struct_size (but specific due to the memory layouts).
I don't see any difference here. Also, I think that to be defensive
in memory allocation arguments it is better than a possible heap
overflow ;)

Also, under the KSPP [1][2][3] there is an effort to keep out of
code all the open-coded arithmetic (To avoid unwanted overflows).

[1] https://github.com/KSPP/linux/issues/83
[2] https://github.com/KSPP/linux/issues/92
[3] https://github.com/KSPP/linux/issues/160

Moreover, after writing these reasons and thinking for a while, I
think that the v1 it is correct patch to apply. This is my opinion
but I'm open minded. Any other solution that makes the code more
secure is welcome.

As a last point I would like to know the opinion of Kees and
Gustavo since they are also working on this task.

Kees and Gustavo, what do you think?

Regards,
Len