[<prev] [next>] [<thread-prev] [day] [month] [year] [list]
Message-ID: <20161122171911.bvzlhli24lscbqa3@grep.be>
Date: Tue, 22 Nov 2016 18:19:11 +0100
From: Wouter Verhelst <w@...r.be>
To: Josef Bacik <jbacik@...com>
Cc: axboe@...com, linux-block@...r.kernel.org,
linux-kernel@...r.kernel.org, kernel-team@...com, alex@...x.org.uk,
nbd-general@...ts.sourceforge.net
Subject: Re: [PATCH][V4] nbd: add multi-connection support
Hi Josef,
[cc to nbd-general added]
On Thu, Nov 17, 2016 at 03:27:30PM -0500, Josef Bacik wrote:
> NBD can become contended on its single connection. We have to serialize all
> writes and we can only process one read response at a time. Fix this by
> allowing userspace to provide multiple connections to a single nbd device. This
> coupled with block-mq drastically increases performance in multi-process cases.
> Thanks,
>
> Signed-off-by: Josef Bacik <jbacik@...com>
> ---
> V3->V4:
> -Fixed a problem where fast completes (or early completes) would crash because
> we were still accessing the bio's on the submit side.
> -Added a flag to disallow multi-connection support if the server doesn't
> explicitly allow for them.
[...]
> + if (num_connections > 1 &&
> + !(nbd->flags & NBD_FLAG_CAN_MULTI_CONN)) {
> + dev_err(disk_to_dev(nbd->disk), "server does not support multiple connections per device.\n");
> + goto out_err;
> + }
I'm not sure whether the kernel needs to check this. I agree that the
flag can be useful, but it's probably something for userspace to check,
rather than for the kernel; I could imagine a --force parameter to be
useful in some corner cases.
Having said that, implementing such a parameter could also be done by
artificially adding particular flags, so I'm certainly not opposed to
this.
[...]
> +#define NBD_FLAG_CAN_MULTI_CONN (1 << 6) /* Server supports multiple connections per export. */
NAK, that is already specified in the protocol spec at
https://github.com/NetworkBlockDevice/nbd/blob/master/doc/proto.md to be
NBD_FLAG_SEND_WRITE_ZEROES. Bit 7 is also taken already, so please use
bit 8 instead.
(I'll reserve that bit in that document to be "export is multi-conn safe" in a
minute)
--
< ron> I mean, the main *practical* problem with C++, is there's like a dozen
people in the world who think they really understand all of its rules,
and pretty much all of them are just lying to themselves too.
-- #debian-devel, OFTC, 2016-02-12
Powered by blists - more mailing lists