Re: [PATCH net-next v2] bonding: add a vlan+mac tx hashing option

[Jarod Wilson]

On Thu, Jan 14, 2021 at 01:54:31PM -0800, Jay Vosburgh wrote:
> Jarod Wilson <jarod@xxxxxxxxxx> wrote:
> 
> >On Wed, Jan 13, 2021 at 05:58:18PM -0800, Jakub Kicinski wrote:
> >> On Wed, 13 Jan 2021 17:35:48 -0500 Jarod Wilson wrote:
> >> > This comes from an end-user request, where they're running multiple VMs on
> >> > hosts with bonded interfaces connected to some interest switch topologies,
> >> > where 802.3ad isn't an option. They're currently running a proprietary
> >> > solution that effectively achieves load-balancing of VMs and bandwidth
> >> > utilization improvements with a similar form of transmission algorithm.
> >> > 
> >> > Basically, each VM has it's own vlan, so it always sends its traffic out
> >> > the same interface, unless that interface fails. Traffic gets split
> >> > between the interfaces, maintaining a consistent path, with failover still
> >> > available if an interface goes down.
> >> > 
> >> > This has been rudimetarily tested to provide similar results, suitable for
> >> > them to use to move off their current proprietary solution. A patch for
> >> > iproute2 is forthcoming as well, to properly support the new mode there as
> >> > well.
> >> 
> >> > Signed-off-by: Jarod Wilson <jarod@xxxxxxxxxx>
> >> > ---
> >> > v2: verified netlink interfaces working, added Documentation, changed
> >> > tx hash mode name to vlan+mac for consistency and clarity.
> >> > 
> >> >  Documentation/networking/bonding.rst | 13 +++++++++++++
> >> >  drivers/net/bonding/bond_main.c      | 27 +++++++++++++++++++++++++--
> >> >  drivers/net/bonding/bond_options.c   |  1 +
> >> >  include/linux/netdevice.h            |  1 +
> >> >  include/uapi/linux/if_bonding.h      |  1 +
> >> >  5 files changed, 41 insertions(+), 2 deletions(-)
> >> > 
> >> > diff --git a/Documentation/networking/bonding.rst b/Documentation/networking/bonding.rst
> >> > index adc314639085..c78ceb7630a0 100644
> >> > --- a/Documentation/networking/bonding.rst
> >> > +++ b/Documentation/networking/bonding.rst
> >> > @@ -951,6 +951,19 @@ xmit_hash_policy
> >> >  		packets will be distributed according to the encapsulated
> >> >  		flows.
> >> >  
> >> > +	vlan+mac
> 
> 	I notice that the code calls it "VLAN_SRCMAC" but the
> user-facing nomenclature is "vlan+mac"; I tend to lean towards having
> the user visible name also be "vlan+srcmac".  Both for consistency, and
> just in case someone someday wants "vlan+dstmac".  And you did ask for
> preference on this in a separate email.

That's valid. I was trying to keep it short, but it does muddy the waters
a bit by not including src. I'll adjust accordingly and resend the
userspace bit too.

...
> 	Yah, the existing L2 hash is pretty weak.  It might be possible
> to squeeze this into the existing bond_xmit_hash a bit better, if the
> hash is two u32s.  The first being the first 32 bits of the MAC, and the
> second being the last 16 bits of the MAC combined with the 16 bit VLAN
> tag.
> 
> 	There's already logic at the end of bond_xmit_hash to reduce a
> u32 into the final hash that perhaps could be leveraged.  
> 
> 	Thinking about it, though, all the ways to combine that data
> together end up being pretty vile ("*(u32 *)&ethhdr->h_source[0]" sorts
> of things).

Yeah, I'd worry that bond_xmit_hash() is already getting a bit complicated
to follow and understand, and that would make it even more so.

> >Something like this instead maybe:
> >
> >static u32 bond_vlan_srcmac_hash(struct sk_buff *skb)
> >{
> >        struct ethhdr *mac_hdr = (struct ethhdr *)skb_mac_header(skb);
> >        u32 srcmac = 0;
> >        u16 vlan;
> >        int i;
> >
> >        for (i = 0; i < ETH_ALEN; i++)
> >                srcmac = (srcmac << 8) | mac_hdr->h_source[i];
> 
> 	I think this will shift h_source[0] and [1] into oblivion.

Argh, yep, 48 bits don't fit into a u32. Okay, so I'll replace that with a
u32 srcmac_vendor and u32 srcmac_dev, but they'll only have 24 bits of data
in them, then return vlan ^ srcmac_vendor ^ srcmac_dev, I think.

-- 
Jarod Wilson
jarod@xxxxxxxxxx