[3.13.y-ckt stable] Patch "gadgetfs: use-after-free in ->aio_read()" has been added to staging queue

[Luis Henriques]

On Mon, Apr 06, 2015 at 02:58:30PM -0700, Kamal Mostafa wrote:
> This is a note to let you know that I have just added a patch titled
> 
>     gadgetfs: use-after-free in ->aio_read()
> 
> to the linux-3.13.y-queue branch of the 3.13.y-ckt extended stable tree 
> which can be found at:
> 
>  http://kernel.ubuntu.com/git?p=ubuntu/linux.git;a=shortlog;h=refs/heads/linux-3.13.y-queue
> 
> This patch is scheduled to be released in version 3.13.11-ckt19.
> 
> If you, or anyone else, feels it should not be added to this tree, please 
> reply to this email.
>

This backport will add a compilation warning because the 'value'
variable may be used uninitialized in function ep_aio_rwtail -- see
below.

> For more information about the 3.13.y-ckt tree, see
> https://wiki.ubuntu.com/Kernel/Dev/ExtendedStable
> 
> Thanks.
> -Kamal
> 
> ------
> 
> From 29e97f7298ed490b12f6c27e60b0607b4f06f081 Mon Sep 17 00:00:00 2001
> From: Al Viro <viro at zeniv.linux.org.uk>
> Date: Fri, 6 Feb 2015 02:07:45 -0500
> Subject: gadgetfs: use-after-free in ->aio_read()
> 
> commit f01d35a15fa04162a58b95970fc01fa70ec9dacd upstream.
> 
> AIO_PREAD requests call ->aio_read() with iovec on caller's stack, so if
> we are going to access it asynchronously, we'd better get ourselves
> a copy - the one on kernel stack of aio_run_iocb() won't be there
> anymore.  function/f_fs.c take care of doing that, legacy/inode.c
> doesn't...
> 
> Signed-off-by: Al Viro <viro at zeniv.linux.org.uk>
> Signed-off-by: Kamal Mostafa <kamal at canonical.com>
> ---
>  drivers/usb/gadget/inode.c | 15 ++++++++++++---
>  1 file changed, 12 insertions(+), 3 deletions(-)
> 
> diff --git a/drivers/usb/gadget/inode.c b/drivers/usb/gadget/inode.c
> index b94c049..ff3e6e6 100644
> --- a/drivers/usb/gadget/inode.c
> +++ b/drivers/usb/gadget/inode.c
> @@ -568,7 +568,6 @@ static ssize_t ep_copy_to_user(struct kiocb_priv *priv)
>  		if (total == 0)
>  			break;
>  	}
> -
>  	return len;
>  }
> 
> @@ -587,6 +586,7 @@ static void ep_user_copy_worker(struct work_struct *work)
>  	aio_complete(iocb, ret, ret);
> 
>  	kfree(priv->buf);
> +	kfree(priv->iv);
>  	kfree(priv);
>  }
> 
> @@ -607,6 +607,7 @@ static void ep_aio_complete(struct usb_ep *ep, struct usb_request *req)
>  	 */
>  	if (priv->iv == NULL || unlikely(req->actual == 0)) {
>  		kfree(req->buf);
> +		kfree(priv->iv);
>  		kfree(priv);
>  		iocb->private = NULL;
>  		/* aio_complete() reports bytes-transferred _and_ faults */
> @@ -642,7 +643,7 @@ ep_aio_rwtail(
>  	struct usb_request	*req;
>  	ssize_t			value;
> 
> -	priv = kmalloc(sizeof *priv, GFP_KERNEL);
> +	priv = kzalloc(sizeof *priv, GFP_KERNEL);
>  	if (!priv) {
>  		value = -ENOMEM;
>  fail:
> @@ -651,7 +652,14 @@ fail:
>  	}
>  	iocb->private = priv;
>  	priv->iocb = iocb;
> -	priv->iv = iv;
> +	if (iv) {
> +		priv->iv = kmemdup(iv, nr_segs * sizeof(struct iovec),
> +				   GFP_KERNEL);
> +		if (!priv->iv) {
> +			kfree(priv);
> +			goto fail;
> +		}
> +	}


This is where 'value' may be used without being initialized, if
kmemdup() fails and the goto statement is executed: 'value' is
returned in the 'fail' label.

This issue seems to have been present upstream but this code was
completely refactored by commit 7fe3976e0f3a ("gadget: switch
ep_io_operations to ->read_iter/->write_iter") and the issue is gone.

Probably, the best thing to do is to add a 'value = -ENOMEM;'
immediately before the goto statment.

Cheers,
--
Luís

>  	priv->nr_segs = nr_segs;
>  	INIT_WORK(&priv->work, ep_user_copy_worker);
> 
> @@ -691,6 +699,7 @@ fail:
>  	mutex_unlock(&epdata->lock);
> 
>  	if (unlikely(value)) {
> +		kfree(priv->iv);
>  		kfree(priv);
>  		put_ep(epdata);
>  	} else
> --
> 1.9.1
> 
> 
> -- 
> kernel-team mailing list
> kernel-team at lists.ubuntu.com
> https://lists.ubuntu.com/mailman/listinfo/kernel-team