Re: net/smc and the RDMA core

3 messages, 3 authors, 2017-05-04 · open the first message on its own page

Re: net/smc and the RDMA core

From: hch-jcswGhMUV9g@public.gmane.org <hidden>
Date: 2017-05-04 08:48:25

On Thu, May 04, 2017 at 11:43:50AM +0300, Sagi Grimberg wrote:
I would also suggest that you stop exposing the DMA MR for remote
access (at least by default) and use a proper reg_mr operations with a
limited lifetime on a properly sized buffer.
Yes, exposing the default DMA MR is a _major_ security risk.  As soon
as SMC is enabled this will mean a remote system has full read/write
access to the local systems memory.

There іs a reason why I removed the ib_get_dma_mr function and replaced
it with the IB_PD_UNSAFE_GLOBAL_RKEY key that has _UNSAFE_ in the name
and a very long comment explaining why, and I'm really disappointed that
we got a driver merged that instead of asking on the relevant list on
why a change unexpertong a function it needed happened and instead
tried the hard way to keep a security vulnerarbility alive.
--
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

Re: net/smc and the RDMA core

From: Ursula Braun <hidden>
Date: 2017-05-04 13:08:49


On 05/04/2017 10:48 AM, hch@lst.de wrote:
On Thu, May 04, 2017 at 11:43:50AM +0300, Sagi Grimberg wrote:
quoted
I would also suggest that you stop exposing the DMA MR for remote
access (at least by default) and use a proper reg_mr operations with a
limited lifetime on a properly sized buffer.
Yes, exposing the default DMA MR is a _major_ security risk.  As soon
as SMC is enabled this will mean a remote system has full read/write
access to the local systems memory.

There іs a reason why I removed the ib_get_dma_mr function and replaced
it with the IB_PD_UNSAFE_GLOBAL_RKEY key that has _UNSAFE_ in the name
and a very long comment explaining why, and I'm really disappointed that
we got a driver merged that instead of asking on the relevant list on
why a change unexpertong a function it needed happened and instead
tried the hard way to keep a security vulnerarbility alive.
Thanks for pointing out these problems. We will address them.

Re: net/smc and the RDMA core

From: Jason Gunthorpe <hidden>
Date: 2017-05-04 15:32:08

On Thu, May 04, 2017 at 03:08:39PM +0200, Ursula Braun wrote:

On 05/04/2017 10:48 AM, hch@lst.de wrote:
quoted
On Thu, May 04, 2017 at 11:43:50AM +0300, Sagi Grimberg wrote:
quoted
I would also suggest that you stop exposing the DMA MR for remote
access (at least by default) and use a proper reg_mr operations with a
limited lifetime on a properly sized buffer.
Yes, exposing the default DMA MR is a _major_ security risk.  As soon
as SMC is enabled this will mean a remote system has full read/write
access to the local systems memory.

There ??s a reason why I removed the ib_get_dma_mr function and replaced
it with the IB_PD_UNSAFE_GLOBAL_RKEY key that has _UNSAFE_ in the name
and a very long comment explaining why, and I'm really disappointed that
we got a driver merged that instead of asking on the relevant list on
why a change unexpertong a function it needed happened and instead
tried the hard way to keep a security vulnerarbility alive.
Thanks for pointing out these problems. We will address them.
So, you've created a huge security hole in the kernel, anyone who
loads your smc module is vunerable.

What are you going to do *RIGHT NOW* to mitigate this?

Jason
Keyboard shortcuts
hback out one level
jnext message in thread
kprevious message in thread
ldrill in
Escclose help / fold thread tree
?toggle this help