netdev - Re: [RFC PATCH bpf-next v4 4/7] bpf: add bpf queue and stack maps

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

Message-ID: <20181005184139.lo3agre2tyn73csr@ast-mbp.dhcp.thefacebook.com>
Date:   Fri, 5 Oct 2018 11:41:40 -0700
From:   Alexei Starovoitov <alexei.starovoitov@...il.com>
To:     Mauricio Vasquez <mauricio.vasquez@...ito.it>
Cc:     Alexei Starovoitov <ast@...nel.org>,
        Daniel Borkmann <daniel@...earbox.net>, netdev@...r.kernel.org
Subject: Re: [RFC PATCH bpf-next v4 4/7] bpf: add bpf queue and stack maps

On Thu, Oct 04, 2018 at 10:40:55PM -0500, Mauricio Vasquez wrote:
> 
> > > +	/* Round up queue size to nearest power of 2 */
> > > +	max_entries = index_mask + 1;
> > what's the point of roundup ?
> 
> If the size of the buffer is power of two we can wrap the indexes with an
> AND operation instead of MOD.
> 
> > The memory waste becomes quite large when max_entries are high.
> Yes, you are right, we have the different choices described below.
> 
> > 
> > If queue/stack is sized to exact max_entries,
> > then 'count' can be removed too, right?
> 
> If we don't use 'count' and we want to use the AND operation for wrapping
> indexes, the max entries should be 2^ - 1  because a slot is lost to
> distinguish between full/empty queue/stack.
> 
> Just to summarize, we have these options:
> 1. Allow any size, round up, use the AND operation and 'count' (current).
> 2. Allow only power of 2 sizes, use the AND operation and 'count'.
> 3. Allow any size, no roundup, use the MOD operation and leaving an empty
> slot.
> 
> I prefer 1 or 2, but I don't have a strong opinion, maybe allowing only
> power of two max entries could be too limiting.
> Another consideration: is this really too bad to waste memory when user
> requires a size far away of the next power of 2?

I think there is 4th option. Neither AND nor MOD is necessary.
Pls take a look at ptr_ring implementation.