Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
Feedback-ID: i787e41f1:Fastmail
Date:   Wed, 5 Apr 2023 16:44:37 +0200
From:   Greg KH <greg@kroah.com>
To:     Alan Stern <stern@rowland.harvard.edu>
Cc:     Hans de Goede <hdegoede@redhat.com>,
        syzbot <syzbot+4b3f8190f6e13b3efd74@syzkaller.appspotmail.com>,
        syzbot <syzbot+1cb937c125adb93fad2d@syzkaller.appspotmail.com>,
        linux-kernel@vger.kernel.org, linux-usb@vger.kernel.org,
        syzkaller-bugs@googlegroups.com
Subject: Re: [syzbot] WARNING in shark_write_reg/usb_submit_urb, WARNING in
 shark_write_val/usb_submit_urb
Message-ID: <2023040544-cuddly-glancing-f577@gregkh>
References: <00000000000096e4f905f81b2702@google.com>
 <e382763c-cf33-4871-a761-1ac85ae36f27@rowland.harvard.edu>
 <8896f261-9602-4663-aa87-1feb9bf3ec0f@redhat.com>
 <2023040148-aground-cornbread-84e2@gregkh>
 <f764a19d-858e-408c-a5f5-d6fe7306c4cb@rowland.harvard.edu>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <f764a19d-858e-408c-a5f5-d6fe7306c4cb@rowland.harvard.edu>
Precedence: bulk

On Sat, Apr 01, 2023 at 02:38:39PM -0400, Alan Stern wrote:
> On Sat, Apr 01, 2023 at 04:53:17PM +0200, Greg KH wrote:
> > On Sat, Apr 01, 2023 at 12:48:07PM +0200, Hans de Goede wrote:
> > > Hi Alan,
> > > 
> > > On 3/30/23 22:10, Alan Stern wrote:
> > > > Reference: https://syzkaller.appspot.com/bug?extid=4b3f8190f6e13b3efd74
> > > > Reference: https://syzkaller.appspot.com/bug?extid=1cb937c125adb93fad2d
> > > > 
> > > > The radio-shark driver just assumes that the endpoints it uses will be
> > > > present, without checking.  This adds an appropriate check.
> > > > 
> > > > Alan Stern
> > > > 
> > > > #syz test: https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/ v6.2
> > > 
> > > Thank you for working on this!
> > > 
> > > Both the core changes and the 2 radio-shark driver changes look good to me.
> > > 
> > > Please add my:
> > > 
> > > Reviewed-by: Hans de Goede <hdegoede@redhat.com>
> > > 
> > > When submitting these upstream.
> > > 
> > > Regards,
> > > 
> > > Hans
> > > 
> > > 
> > > 
> > > 
> > > 
> > > > 
> > > >  drivers/usb/core/usb.c |   70 +++++++++++++++++++++++++++++++++++++++++++++++++
> > > >  include/linux/usb.h    |    7 ++++
> > > >  2 files changed, 77 insertions(+)
> > > > 
> > > > Index: usb-devel/drivers/usb/core/usb.c
> > > > ===================================================================
> > > > --- usb-devel.orig/drivers/usb/core/usb.c
> > > > +++ usb-devel/drivers/usb/core/usb.c
> > > > @@ -207,6 +207,76 @@ int usb_find_common_endpoints_reverse(st
> > > >  EXPORT_SYMBOL_GPL(usb_find_common_endpoints_reverse);
> > > >  
> > > >  /**
> > > > + * usb_find_endpoint() - Given an endpoint address, search for the endpoint's
> > > > + * usb_host_endpoint structure in an interface's current altsetting.
> > > > + * @intf: the interface whose current altsetting should be searched
> > > > + * @ep_addr: the endpoint address (number and direction) to find
> > > > + *
> > > > + * Search the altsetting's list of endpoints for one with the specified address.
> > > > + *
> > > > + * Return: Pointer to the usb_host_endpoint if found, %NULL otherwise.
> > > > + */
> > > > +struct usb_host_endpoint __must_check *usb_find_endpoint(
> > > > +		const struct usb_interface *intf, unsigned int ep_addr)
> > > > +{
> > > > +	int n;
> > > > +	struct usb_host_endpoint *ep;
> > > > +
> > > > +	n = intf->cur_altsetting->desc.bNumEndpoints;
> > > > +	ep = intf->cur_altsetting->endpoint;
> > > > +	for (; n > 0; (--n, ++ep)) {
> > > > +		if (ep->desc.bEndpointAddress == ep_addr)
> > > > +			return ep;
> > > > +	}
> > > > +	return NULL;
> > > > +}
> > > > +EXPORT_SYMBOL_GPL(usb_find_endpoint);
> > > > +
> > > > +/**
> > > > + * usb_check_bulk_endpoint - Check whether an interface's current altsetting
> > > > + * contains a bulk endpoint with the given address.
> > > > + * @intf: the interface whose current altsetting should be searched
> > > > + * @ep_addr: the endpoint address (number and direction) to look for
> > > > + *
> > > > + * Search for an endpoint with the specified address and check its type.
> > > > + *
> > > > + * Return: %true if the endpoint is found and is bulk, %false otherwise.
> > > > + */
> > > > +bool usb_check_bulk_endpoint(
> > > > +		const struct usb_interface *intf, unsigned int ep_addr)
> > > > +{
> > > > +	const struct usb_host_endpoint *ep;
> > > > +
> > > > +	ep = usb_find_endpoint(intf, ep_addr);
> > > > +	if (!ep)
> > > > +		return false;
> > > > +	return usb_endpoint_xfer_bulk(&ep->desc);
> > > > +}
> > > > +EXPORT_SYMBOL_GPL(usb_check_bulk_endpoint);
> > > > +
> > > > +/**
> > > > + * usb_check_int_endpoint - Check whether an interface's current altsetting
> > > > + * contains an interrupt endpoint with the given address.
> > > > + * @intf: the interface whose current altsetting should be searched
> > > > + * @ep_addr: the endpoint address (number and direction) to look for
> > > > + *
> > > > + * Search for an endpoint with the specified address and check its type.
> > > > + *
> > > > + * Return: %true if the endpoint is found and is interrupt, %false otherwise.
> > > > + */
> > > > +bool usb_check_int_endpoint(
> > > > +		const struct usb_interface *intf, unsigned int ep_addr)
> > > > +{
> > > > +	const struct usb_host_endpoint *ep;
> > > > +
> > > > +	ep = usb_find_endpoint(intf, ep_addr);
> > > > +	if (!ep)
> > > > +		return false;
> > > > +	return usb_endpoint_xfer_int(&ep->desc);
> > > > +}
> > > > +EXPORT_SYMBOL_GPL(usb_check_int_endpoint);
> > 
> > Shouldn't you use the usb_find_bulk_in_endpoint() and friends functions
> > instead of these?  Many drivers hard-coded their "I know this endpoint
> > is this type" which breaks in fuzzing as you know (and see here), which
> > is why those functions were created to be used.
> 
> It's true, we could use the existing functions in this case.  I'm not sure 
> which approach would be better; it's probably mostly a matter of taste.
> 
> Using the existing functions, the probe function would get additional code 
> something like this:
> 
> 	struct usb_endpoint_descriptor *ep;
> 
> 	if (!(usb_find_int_in_endpoint(intf->cur_altsetting, &ep) == 0 &&
> 			usb_endpoint_num(&ep->desc) == SHARK_IN_EP) ||
> 	    !(usb_find_int_out_endpoint(intf->cur_altsetting, &ep) == 0 &&
> 			usb_endpoint_num(&ep->desc) == SHARK_OUT_EP)) {
> 		dev_err(...
> 
> Using the new functions (with a revised API, see the patch below) we would 
> instead add this:
> 
> 	static const u8 ep_addresses[] = {
> 		SHARK_IN_EP | USB_DIR_IN,
> 		SHARK_OUT_EP | USB_DIR_OUT,
> 		0};
> 
> 	if (!usb_check_int_endpoints(intf, ep_addresses)) {
> 		dev_err(...
> 
> In this case the difference is not particularly meaningful.  With the new 
> functions, the amount of code added to the driver is smaller, but of 
> course that's offset by adding the new functions themselves to the core.  
> (On the other hand there's probably a bunch of drivers that could stand 
> to be fixed up this way, which would amortize the cost to the core.)
> 
> The difference becomes a lot more noticeable with the sisusbvga driver 
> [1].  It has six endpoints that need to be checked: four bulk-OUT and two 
> bulk-IN.  While usb_find_common_endpoints() would be able to find them, 
> it's not well suited for checking that they have the expected addresses.  
> With the new functions, all the work can be done with a single call.
> 
> > I think just using them in the probe function would fix this issue 
> > instead of these functions which would only be used by that one driver.
> 
> It wouldn't be used just by these two drivers.  The new routines are 
> ideally suited for doing the checking in old drivers that have their 
> endpoint numbers and directions built-in, like these do.  While I haven't 
> looked to see, there must be quite a few of them.

Ok, fair enough, let's take this and see what other drivers can be fixed
up this way.

thanks,

greg k-h