linux-kernel - Re: Still a pretty bad time on 5.4.6 with fuse_request

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-ID: <CAJfpegv4iL=bW3TXP3F9w1z6-LUox8KiBmw7UBcWE-0jiK0YsA@mail.gmail.com>
Date:   Tue, 11 Feb 2020 11:55:46 +0100
From:   Miklos Szeredi <miklos@...redi.hu>
To:     michael+lkml@...pelberg.ch
Cc:     fuse-devel <fuse-devel@...ts.sourceforge.net>,
        Greg Kroah-Hartman <gregkh@...uxfoundation.org>,
        Kyle Sanderson <kyle.leet@...il.com>,
        linux-fsdevel@...r.kernel.org, linux-kernel@...r.kernel.org
Subject: Re: Still a pretty bad time on 5.4.6 with fuse_request_end.

On Sun, Feb 9, 2020 at 9:09 AM <michael+lkml@...pelberg.ch> wrote:
>
> From: Michael Stapelberg <michael+lkml@...pelberg.ch>
>
> Hey,
>
> I recently ran into this, too. The symptom for me is that processes using the
> affected FUSE file system hang indefinitely, sync(2) system calls hang
> indefinitely, and even triggering an abort via echo 1 >
> /sys/fs/fuse/connections/*/abort does not get the file system unstuck (there is
> always 1 request still pending). Only removing power will get the machine
> unstuck.
>
> I’m triggering this when building packages for https://distr1.org/, which uses a
> FUSE daemon (written in Go using the jacobsa/fuse package) to provide package
> contents.
>
> I bisected the issue to commit
> https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/?id=2b319d1f6f92a4ced9897678113d176ee16ae85d
>
> With that commit, I run into a kernel oops within ≈1 minute after starting my
> batch build. With the commit before, I can batch build for many minutes without
> issues.

Pretty weird.   I'm not seeing how this could change behavior, as the
args->end value is not changed after being initialized, and so moving
the test later should not make a difference.

Could you print out the complete contents of req->args?

Thanks,
Miklos