linux-kernel - Re: [PATCH] acct: Prevent NULL pointer dereference when writing to sysfs

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: <20250211-anschaffen-leiht-9e803cc9b087@brauner>
Date: Tue, 11 Feb 2025 11:17:07 +0100
From: Christian Brauner <brauner@...nel.org>
To: Al Viro <viro@...iv.linux.org.uk>
Cc: Zicheng Qu <quzicheng@...wei.com>, 
	Linus Torvalds <torvalds@...ux-foundation.org>, jlayton@...nel.org, axboe@...nel.dk, joel.granados@...nel.org, 
	tglx@...utronix.de, linux-kernel@...r.kernel.org, hch@....de, len.brown@...el.com, 
	pavel@....cz, pengfei.xu@...el.com, rafael@...nel.org, 
	syzkaller-bugs@...glegroups.com, linux-pm@...r.kernel.org, tanghui20@...wei.com, 
	zhangqiao22@...wei.com, judy.chenhui@...wei.com
Subject: Re: [PATCH] acct: Prevent NULL pointer dereference when writing to
 sysfs

On Tue, Feb 11, 2025 at 12:23:08AM +0000, Al Viro wrote:
> On Mon, Feb 10, 2025 at 06:19:02PM +0000, Al Viro wrote:
> > On Mon, Feb 10, 2025 at 05:02:35PM +0100, Christian Brauner wrote:
> > > On Mon, Feb 10, 2025 at 03:21:46PM +0000, Al Viro wrote:
> > > > On Mon, Feb 10, 2025 at 04:12:54PM +0100, Christian Brauner wrote:
> > > > 
> > > > > One fix would be to move exit_fs() past exit_task_work(). It looks like
> > > > > that this should be doable without much of a problem and it would fix
> > > > > the path_init() problem.
> > > > > 
> > > > > There should hopefully be nothing relying on task->fs == NULL in
> > > > > exit_task_work().
> > > > 
> > > > There's a question of the task_work_add() issued by exit_task_fs(),
> > > > though.
> > > 
> > > Can't we simply remove the pins on the mounts of fs->root and fs->pwd in
> > > exit_fs() explicitly? If that works I think that's a fair enough
> > > compromise for this shite.
> > 
> > I'd rather go for a simpler approach...  Why do we need those writes
> > to be done in context of exiting process in the first place?  It's
> > not as if they needed to go out before it terminates, so what's to
> > stop us from having a kernel thread in background and queue the data
> > to be written for it to pick up?
> > 
> > Does anybody see problems with that approach?
> 
> Note, BTW, that games with rlimit and creds switching disappear if done
> that way.  
> 
> FWIW, I wonder if we should simply allocate a page worth of buffer,
> occupied by acct_t array, with count + pointer to buffer kept in acct,
> with acct->mutex used to protect the entire thing, so that do_acct_process()
> would add a record to that sucker and wake the kthread up, with kthread
> handling actual writes and emptying the buffer.  No need for exit(2)
> to wait unless the buffer is full...

I had thought about it but both LTP and the selftests want the buffer to
be filled after the process exits.

So let's not overly complicate this. I want this to be as simple as
possible and then start deprecating this api asap.

I have a patch that just moves the final write into the workqueue. But
let's keep the cred override because who knows what security hole we
open up if we skip the override cred.