[<prev] [next>] [<thread-prev] [day] [month] [year] [list]
Message-ID: <598c983f-d28a-3b77-9d9c-3a8686836630@gmail.com>
Date: Mon, 7 Oct 2019 15:55:29 -0600
From: David Ahern <dsahern@...il.com>
To: Leon Romanovsky <leon@...nel.org>
Cc: Leon Romanovsky <leonro@...lanox.com>,
netdev <netdev@...r.kernel.org>,
RDMA mailing list <linux-rdma@...r.kernel.org>,
Stephen Hemminger <stephen@...workplumber.org>
Subject: Re: [PATCH iproute2-next] rdma: Relax requirement to have PID for HW
objects
On 10/2/19 7:49 AM, Leon Romanovsky wrote:
> From: Leon Romanovsky <leonro@...lanox.com>
>
> RDMA has weak connection between PIDs and HW objects, because
> the latter tied to file descriptors for their lifetime management.
>
> The outcome of such connection is that for the following scenario,
> the returned PID will be 0 (not-valid):
> 1. Create FD and context
> 2. Share it with ephemeral child
> 3. Create any object and exit that child
>
> This flow was revealed in testing environment and of course real users
> are not running such scenario, because it makes no sense at all in RDMA
> world.
>
> Let's do two changes in the code to support such workflow anyway:
> 1. Remove need to provide PID/kernel name. Code already supports it,
> just need to remove extra validation.
> 2. Ball-out in case PID is 0.
>
> Link: https://lore.kernel.org/linux-rdma/20191002123245.18153-2-leon@kernel.org
> Signed-off-by: Leon Romanovsky <leonro@...lanox.com>
> ---
> rdma/res-cmid.c | 5 +----
> rdma/res-cq.c | 5 +----
> rdma/res-mr.c | 5 +----
> rdma/res-pd.c | 5 +----
> rdma/res-qp.c | 5 +----
> rdma/res.c | 3 +++
> 6 files changed, 8 insertions(+), 20 deletions(-)
>
applied to iproute2-next
Powered by blists - more mailing lists