linux-kernel - Re: [syzbot] [kernfs?] possible deadlock in kernfs_fop

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]

Date: Fri, 5 Apr 2024 16:52:40 +0200
From: Christoph Hellwig <hch@....de>
To: Christian Brauner <brauner@...nel.org>
Cc: Christoph Hellwig <hch@....de>, Amir Goldstein <amir73il@...il.com>,
	Al Viro <viro@...iv.linux.org.uk>,
	syzbot <syzbot+9a5b0ced8b1bfb238b56@...kaller.appspotmail.com>,
	gregkh@...uxfoundation.org, linux-fsdevel@...r.kernel.org,
	linux-kernel@...r.kernel.org, syzkaller-bugs@...glegroups.com,
	tj@...nel.org, valesini@...dex-team.ru, Jan Kara <jack@...e.cz>,
	Miklos Szeredi <miklos@...redi.hu>
Subject: Re: [syzbot] [kernfs?] possible deadlock in kernfs_fop_llseek

On Fri, Apr 05, 2024 at 03:48:20PM +0200, Christian Brauner wrote:
> * There's early_lookup_bdev() which deals with PARTUUID,
>   PARTLABEL, raw device number, and lookup based on /dev. No actual path
>   lookup involved in that.
> 
> * So the only interesting case is lookup_bdev() for /sys/power/suspend.
>   That one takes arbitrary paths. But being realistic for a moment...
>   How many people will specify a device path that's _not_ some variant
>   of /dev/...? IOW, how many people will specify a device path that's
>   not on devtmpfs or a symlink on devtmpfs? Probably almost no one.

That's not the point.  The poins is that trying to do the dumb name
to bdev translation in early_lookup_bdev is wrong.  Distro had and have
their own numbering schemes, and not using them bypasses access
control.  We should never use that at runtime.

> <brauner> So /sys/power/resume does systemd ever write anything other than a /dev/* path in to there?
> <maintainer> Hmm? You never do that? It only accepts devno.
> 
> So that takes away one of the main users of this api. So I really
> suspect that arbitrary device path is unused in practice. Maybe I'm all
> wrong though.

I'm all fine with just accepting a devno and no name.  But I fear it
will break something as someone added it for whatever use case they had
(and we should not have allowed that back then, but that ship has sailed
unfortunately)