linux-kernel - Re: possible deadlock in mon_bin_vma

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:   Wed, 20 Nov 2019 13:47:00 -0500 (EST)
From:   Alan Stern <stern@...land.harvard.edu>
To:     Pete Zaitcev <zaitcev@...hat.com>
cc:     syzbot <syzbot+56f9673bb4cdcbeb0e92@...kaller.appspotmail.com>,
        <arnd@...db.de>, <gregkh@...uxfoundation.org>,
        <jrdr.linux@...il.com>, <keescook@...omium.org>,
        <kstewart@...uxfoundation.org>,
        Kernel development list <linux-kernel@...r.kernel.org>,
        USB list <linux-usb@...r.kernel.org>,
        <syzkaller-bugs@...glegroups.com>, <tglx@...utronix.de>,
        <viro@...iv.linux.org.uk>
Subject: Re: possible deadlock in mon_bin_vma_fault

On Wed, 20 Nov 2019, Pete Zaitcev wrote:

> On Wed, 20 Nov 2019 11:14:05 -0500 (EST)
> Alan Stern <stern@...land.harvard.edu> wrote:
> 
> > As it happens, I spent a little time investigating this bug report just
> > yesterday.  It seems to me that the easiest fix would be to disallow
> > resizing the buffer while it is mapped by any users.  (Besides,
> > allowing that seems like a bad idea in any case.)
> > 
> > Pete, does that seem reasonable to you?
> 
> Yes, it does seem reasonable.
> 
> I think I understand it now. My fallacy was thinking that since everything
> is nailed down as long as fetch_lock is held, it was okay to grab whatever
> page from our pagemap. What happens later is an attempt to get pages of the
> new buffer while looking at them through the old VMA, in mon_bin_vma_fault.
> 
> It seems to me that the use counter, mmap_active, is correct and sufficient
> to check in the ioctl.
> 
> -- Pete
> 
> P.S. One thing that vaguely bothers me on this is that the bot
> bisected to the commit that clearly fixed worse issues.
> 
> P.P.S. Like this?
> 
> diff --git a/drivers/usb/mon/mon_bin.c b/drivers/usb/mon/mon_bin.c
> index ac2b4fcc265f..e27d99606adb 100644
> --- a/drivers/usb/mon/mon_bin.c
> +++ b/drivers/usb/mon/mon_bin.c
> @@ -1020,6 +1020,9 @@ static long mon_bin_ioctl(struct file *file, unsigned int cmd, unsigned long arg
>                 int size;
>                 struct mon_pgmap *vec;
>  
> +               if (rp->mmap_active)
> +                       return -EBUSY;
> +
>                 if (arg < BUFF_MIN || arg > BUFF_MAX)
>                         return -EINVAL;

Like that, yes, but the test has to be made while fetch_lock is held.  
Otherwise there's still a race: One thread could pass the test and then
do the resize, and in between another thread could map the buffer and
incur a fault.

Incidentally, the comment for fetch_lock says that it protects b_read 
and b_out, but mon_bin_vma_fault doesn't use either of those fields.

Alan Stern