Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:18 as permitted sender) client-ip=2620:137:e000::1:18;
MIME-Version: 1.0
References: <20220422054224.19527-1-matthew.ruffell@canonical.com>
In-Reply-To: <20220422054224.19527-1-matthew.ruffell@canonical.com>
From:   Josef Bacik <josef@toxicpanda.com>
Date:   Fri, 22 Apr 2022 11:23:58 -0400
Message-ID: <CAEzrpqe=LD3DQcEeLXmmFuq7cX_dAQ6DOCuJYWBoZWKKTmoTzA@mail.gmail.com>
Subject: Re: [PROBLEM] nbd requests become stuck when devices watched by
 inotify emit udev uevent changes
To:     Matthew Ruffell <matthew.ruffell@canonical.com>
Cc:     Jens Axboe <axboe@kernel.dk>,
        linux-block <linux-block@vger.kernel.org>,
        nbd <nbd@other.debian.org>,
        Linux Kernel <linux-kernel@vger.kernel.org>
Content-Type: text/plain; charset="UTF-8"
Precedence: bulk

On Fri, Apr 22, 2022 at 1:42 AM Matthew Ruffell
<matthew.ruffell@canonical.com> wrote:
>
> Dear maintainers of the nbd subsystem,
>
> A user has come across an issue which causes the nbd module to hang after a
> disconnect where a write has been made to a qemu qcow image file, with qemu-nbd
> being the server.
>

Ok there's two problems here, but I want to make sure I have the right
fix for the hang first.  Can you apply this patch

https://paste.centos.org/view/b1a2d01a

and make sure the hang goes away?  Once that part is fixed I'll fix
the IO errors, this is just us racing with systemd while we teardown
the device and then we're triggering a partition read while the device
is going down and it's complaining loudly.  Before we would
set_capacity to 0 whenever we disconnected, but that causes problems
with file systems that may still have the device open.  However now we
only do this if the server does the CLEAR_SOCK ioctl, which clearly
can race with systemd poking the device, so I need to make it
set_capacity(0) when the last opener closes the device to prevent this
style of race.

Let me know if that patch fixes the hang, and then I'll work up
something for the capacity problem.  Thanks,

Josef