RE: [PATCH] net/smc: mark as BROKEN due to remote memory exposure

[Parav Pandit <parav-VPRAkNaXOzVWk0Htik3J/w@public.gmane.org>]

Hi Dave,

> -----Original Message-----
> From: linux-rdma-owner-u79uwXL29TY76Z2rM5mHXA@public.gmane.org [mailto:linux-rdma-
> owner-u79uwXL29TY76Z2rM5mHXA@public.gmane.org] On Behalf Of David Miller
> Sent: Sunday, May 14, 2017 7:44 PM
> To: Bart.VanAssche-XdAiOPVOjttBDgjK7y7TUQ@public.gmane.org
> Cc: hch-jcswGhMUV9g@public.gmane.org; netdev-u79uwXL29TY76Z2rM5mHXA@public.gmane.org; linux-rdma-u79uwXL29TY76Z2rM5mHXA@public.gmane.org;
> stable-u79uwXL29TY76Z2rM5mHXA@public.gmane.org; ubraun-23VcF4HTsmIX0ybBhKVfKdBPR1lH4CV8@public.gmane.org
> Subject: Re: [PATCH] net/smc: mark as BROKEN due to remote memory
> exposure
> 
> From: Bart Van Assche <Bart.VanAssche-XdAiOPVOjttBDgjK7y7TUQ@public.gmane.org>
> Date: Sun, 14 May 2017 19:08:50 +0000
> 
> > What is your plan to avoid that applications start using and depending
> > on AF_SMC?
> 

status = socket(AF_SMC, field, IPPROT_TCP);
Here,
- AF_SMC actually means AF_INET IPv4 addresses!
- IPPROTO_TCP means TCP and RDMA both when socket is AF_SMC.
- When creating socket addresses, use AF_INET based addresses.
-  When invoking bind(), listen(), connect() APIs, use AF_INET addresses instead.
- Supporting IPv6 is TBD with AF_SMC sockets.
- At user level get_addrinfo will continue to return AF_INET addresses.

Such explanation for socket APIs doesn't sound correct.

The primary motivation for SMC protocol was to simplify the applications and library to make use of RDMA.
This kind of API is against such simplicity and creates more confusion.
RFC only gives example and doesn't asks to create new socket family.
I can provide more data, but a simple grep in get_addrinfo() and friend functions in user space has heavy dependence on AF_INET and AF_INET6.

> The API is out there already so we are out of luck, and neither you nor I nor
> anyone else can "stop" this from happening.

I think it is still not too late to fix this API. SMC is released in v4.11 very recently.
v4.12 is still not out.
Given the limitation of protocol being RoCEv1 only, we might not have many users whose applications will stop functioning.
(Which will anyway won't work for RoCEv2, and IPv6 addresses).

I propose,
(a) AF_SMC socket 43 can be marked reserved in future kernel versions to avoid use.
(b) New protocol family that represents TCP and RDMA protocol, may be named IPPROTO_SMC even though it is not a protocol in IP header.

We can possibly target to have this fix in 4.13 kernel timeframe.

> --
> To unsubscribe from this list: send the line "unsubscribe linux-rdma" in the
> body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org More majordomo info
> at  http://vger.kernel.org/majordomo-info.html
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html