LinuxPPC-Dev Archive on lore.kernel.org
 help / color / Atom feed
From: Greg KH <gregkh@linuxfoundation.org>
To: rananta@codeaurora.org
Cc: andrew@daynix.com, linuxppc-dev@lists.ozlabs.org,
	linux-kernel@vger.kernel.org, jslaby@suse.com
Subject: Re: [PATCH] tty: hvc: Fix data abort due to race in hvc_open
Date: Mon, 11 May 2020 09:39:13 +0200
Message-ID: <20200511073913.GA1347819@kroah.com> (raw)
In-Reply-To: <77d889be4e0cb0e6e30f96199e2d843d@codeaurora.org>

On Mon, May 11, 2020 at 12:23:58AM -0700, rananta@codeaurora.org wrote:
> On 2020-05-09 23:48, Greg KH wrote:
> > On Sat, May 09, 2020 at 06:30:56PM -0700, rananta@codeaurora.org wrote:
> > > On 2020-05-06 02:48, Greg KH wrote:
> > > > On Mon, Apr 27, 2020 at 08:26:01PM -0700, Raghavendra Rao Ananta wrote:
> > > > > Potentially, hvc_open() can be called in parallel when two tasks calls
> > > > > open() on /dev/hvcX. In such a scenario, if the
> > > > > hp->ops->notifier_add()
> > > > > callback in the function fails, where it sets the tty->driver_data to
> > > > > NULL, the parallel hvc_open() can see this NULL and cause a memory
> > > > > abort.
> > > > > Hence, serialize hvc_open and check if tty->private_data is NULL
> > > > > before
> > > > > proceeding ahead.
> > > > >
> > > > > The issue can be easily reproduced by launching two tasks
> > > > > simultaneously
> > > > > that does nothing but open() and close() on /dev/hvcX.
> > > > > For example:
> > > > > $ ./simple_open_close /dev/hvc0 & ./simple_open_close /dev/hvc0 &
> > > > >
> > > > > Signed-off-by: Raghavendra Rao Ananta <rananta@codeaurora.org>
> > > > > ---
> > > > >  drivers/tty/hvc/hvc_console.c | 16 ++++++++++++++--
> > > > >  1 file changed, 14 insertions(+), 2 deletions(-)
> > > > >
> > > > > diff --git a/drivers/tty/hvc/hvc_console.c
> > > > > b/drivers/tty/hvc/hvc_console.c
> > > > > index 436cc51c92c3..ebe26fe5ac09 100644
> > > > > --- a/drivers/tty/hvc/hvc_console.c
> > > > > +++ b/drivers/tty/hvc/hvc_console.c
> > > > > @@ -75,6 +75,8 @@ static LIST_HEAD(hvc_structs);
> > > > >   */
> > > > >  static DEFINE_MUTEX(hvc_structs_mutex);
> > > > >
> > > > > +/* Mutex to serialize hvc_open */
> > > > > +static DEFINE_MUTEX(hvc_open_mutex);
> > > > >  /*
> > > > >   * This value is used to assign a tty->index value to a hvc_struct
> > > > > based
> > > > >   * upon order of exposure via hvc_probe(), when we can not match it
> > > > > to
> > > > > @@ -346,16 +348,24 @@ static int hvc_install(struct tty_driver
> > > > > *driver, struct tty_struct *tty)
> > > > >   */
> > > > >  static int hvc_open(struct tty_struct *tty, struct file * filp)
> > > > >  {
> > > > > -	struct hvc_struct *hp = tty->driver_data;
> > > > > +	struct hvc_struct *hp;
> > > > >  	unsigned long flags;
> > > > >  	int rc = 0;
> > > > >
> > > > > +	mutex_lock(&hvc_open_mutex);
> > > > > +
> > > > > +	hp = tty->driver_data;
> > > > > +	if (!hp) {
> > > > > +		rc = -EIO;
> > > > > +		goto out;
> > > > > +	}
> > > > > +
> > > > >  	spin_lock_irqsave(&hp->port.lock, flags);
> > > > >  	/* Check and then increment for fast path open. */
> > > > >  	if (hp->port.count++ > 0) {
> > > > >  		spin_unlock_irqrestore(&hp->port.lock, flags);
> > > > >  		hvc_kick();
> > > > > -		return 0;
> > > > > +		goto out;
> > > > >  	} /* else count == 0 */
> > > > >  	spin_unlock_irqrestore(&hp->port.lock, flags);
> > > >
> > > > Wait, why isn't this driver just calling tty_port_open() instead of
> > > > trying to open-code all of this?
> > > >
> > > > Keeping a single mutext for open will not protect it from close, it will
> > > > just slow things down a bit.  There should already be a tty lock held by
> > > > the tty core for open() to keep it from racing things, right?
> > > The tty lock should have been held, but not likely across
> > > ->install() and
> > > ->open() callbacks, thus resulting in a race between hvc_install() and
> > > hvc_open(),
> > 
> > How?  The tty lock is held in install, and should not conflict with
> > open(), otherwise, we would be seeing this happen in all tty drivers,
> > right?
> > 
> Well, I was expecting the same, but IIRC, I see that the open() was being
> called in parallel for the same device node.

So open and install are happening at the same time?  And the tty_lock()
does not protect the needed fields from being protected properly?  If
not, what fields are being touched without the lock?

> Is it expected that the tty core would allow only one thread to
> access the dev-node, while blocking the other, or is it the client
> driver's responsibility to handle the exclusiveness?

The tty core should handle this correctly, for things that can mess
stuff up (like install and open at the same time).  A driver should not
have to worry about that.

> > > where hvc_install() sets a data and the hvc_open() clears it.
> > > hvc_open()
> > > doesn't
> > > check if the data was set to NULL and proceeds.
> > 
> > What data is being set that hvc_open is checking?
> hvc_install sets tty->private_data to hp, while hvc_open sets it to NULL (in
> one of the paths).

I see no use of private_data in drivers/tty/hvc/ so what exactly are you
referring to?  The file private_data or the port private_data or
something else?

> > And you are not grabbing a lock in your install callback, you are only
> > serializing your open call here, I don't see how this is fixing anything
> > other than perhaps slowing down your codepaths.
> Basically, my intention was to add a NULL check before accessing *hp in
> open().
> The intention of the lock was to protect against this check.
> If the tty layer would have taken care of this, then perhaps there won't be
> a
> need to check for NULL.

Ah, driver_data is what you are referring to, not private_data.

Look at hvc_close(), no locking is done there to test for private_data,
right?  Why not?  The only thing setting driver_data is in install, and
your lock is not touching that.

And again, install and open should not race, if so, the tty core needs
to be fixed.

> > As an arument why this isn't correct, can you answer why this same type
> > of change wouldn't be required for all tty drivers in the tree?
> > 
> I agree, that if it's already taken care by the tty-core, we don't need it
> here.
> Correct me if I'm wrong, but looks like the tty layer is allowing parallel
> accesses
> to open(),

I do not think that happens, try counting the calls to open(), there
should only be one.  If not, that's a bug somewhere else.

thanks,

greg k-h

  parent reply index

Thread overview: 18+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2020-04-28  3:26 Raghavendra Rao Ananta
2020-05-06  9:48 ` Greg KH
2020-05-10  1:30   ` rananta
2020-05-10  6:48     ` Greg KH
2020-05-11  7:23       ` rananta
2020-05-11  7:34         ` rananta
2020-05-11  7:41           ` Greg KH
2020-05-11  7:39         ` Greg KH [this message]
2020-05-12  7:22           ` Jiri Slaby
2020-05-12  8:25             ` Greg KH
2020-05-12 21:39               ` rananta
2020-05-13  7:04                 ` Greg KH
2020-05-14 23:22                   ` rananta
2020-05-15  7:30                     ` Greg KH
2020-05-15 19:21                       ` rananta
2020-05-20  9:38                     ` Jiri Slaby
2020-05-20 13:49                       ` rananta
2020-04-28 12:48 Markus Elfring

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=20200511073913.GA1347819@kroah.com \
    --to=gregkh@linuxfoundation.org \
    --cc=andrew@daynix.com \
    --cc=jslaby@suse.com \
    --cc=linux-kernel@vger.kernel.org \
    --cc=linuxppc-dev@lists.ozlabs.org \
    --cc=rananta@codeaurora.org \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link

LinuxPPC-Dev Archive on lore.kernel.org

Archives are clonable:
	git clone --mirror https://lore.kernel.org/linuxppc-dev/0 linuxppc-dev/git/0.git

	# If you have public-inbox 1.1+ installed, you may
	# initialize and index your mirror using the following commands:
	public-inbox-init -V2 linuxppc-dev linuxppc-dev/ https://lore.kernel.org/linuxppc-dev \
		linuxppc-dev@lists.ozlabs.org linuxppc-dev@ozlabs.org
	public-inbox-index linuxppc-dev

Example config snippet for mirrors

Newsgroup available over NNTP:
	nntp://nntp.lore.kernel.org/org.ozlabs.lists.linuxppc-dev


AGPL code for this site: git clone https://public-inbox.org/public-inbox.git