linux-kernel - Re: [GIT PULL] selinux/selinux-pr-20231030

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] [day] [month] [year] [list]

Message-ID: <CAHC9VhQs9H-f8qZbwMgcF=LHEwspxvdJSNhhjwGu5K-ExYjPZw@mail.gmail.com>
Date:   Tue, 31 Oct 2023 21:31:48 -0400
From:   Paul Moore <paul@...l-moore.com>
To:     Linus Torvalds <torvalds@...ux-foundation.org>
Cc:     Christian Göttsche <cgzones@...glemail.com>,
        selinux@...r.kernel.org, linux-security-module@...r.kernel.org,
        linux-kernel@...r.kernel.org
Subject: Re: [GIT PULL] selinux/selinux-pr-20231030

On Tue, Oct 31, 2023 at 2:13 AM Linus Torvalds
<torvalds@...ux-foundation.org> wrote:
> On Mon, 30 Oct 2023 at 16:16, Paul Moore <paul@...l-moore.com> wrote:
> >
> > * Use a better hashing function for the SELinux role tansition hash
> >   table.
>
> Bah.
>
> While the old hash function was garbage, the new one is quite expensive.
>
> Maybe it's worth it.
>
> But generally, if you find that "oh, just doing a modulus with a power
> of two drops all high bits", the first thing to try is probably to
> just do "hash_long(x, N)" to get N bits instead.
>
> Assuming the input is somewhat ok in one word, it does a fairly good
> job of mixing the bits with a simple multiply-and-shift.
>
> Yes, yes, jhash is a fine hash, but it does a quite *lot* of (simple)
> ALU ops. While "hash_long()" is often small enough to be inlined.

We probably should do some performance measurements of the various
hash tables in the SELinux code and use that to drive some decisions
on what functions we use.  There have been some in the past for
specific tables, but I don't think we've done anything comprehensive,
or recent.  This latest change obviously focused more on ensuring a
better distribution, which can help, but if the digest calculation is
too slow it probably doesn't matter.

> I also note that filenametr_hash() does the old "one byte at a time"
> hash and partial_name_hash(). Is there any reason that code doesn't
> use the "full_name_hash()" which does things a word at a time?

Likely just a matter of no one looking at it and realizing it can be
improved.  I'll toss this on the todo list, it should take all of five
minutes.

> Probably doesn't matter, but since I looked at this to see what the
> new hashing was, I noticed...

No harm in mentioning it, feedback is always welcome, but you know
what else is even more welcome?  Patches ;)

-- 
paul-moore.com