Re: [PATCH] USB: gadget: Fix use-after-free Read in usb_udc_uevent()

[Greg KH <gregkh@linuxfoundation.org>]

On Thu, Jul 21, 2022 at 11:07:10AM -0400, Alan Stern wrote:
> The syzbot fuzzer found a race between uevent callbacks and gadget
> driver unregistration that can cause a use-after-free bug:
> 
> ---------------------------------------------------------------
> BUG: KASAN: use-after-free in usb_udc_uevent+0x11f/0x130
> drivers/usb/gadget/udc/core.c:1732
> Read of size 8 at addr ffff888078ce2050 by task udevd/2968
> 
> CPU: 1 PID: 2968 Comm: udevd Not tainted 5.19.0-rc4-next-20220628-syzkaller #0
> Hardware name: Google Google Compute Engine/Google Compute Engine, BIOS Google
> 06/29/2022
> Call Trace:
>  <TASK>
>  __dump_stack lib/dump_stack.c:88 [inline]
>  dump_stack_lvl+0xcd/0x134 lib/dump_stack.c:106
>  print_address_description mm/kasan/report.c:317 [inline]
>  print_report.cold+0x2ba/0x719 mm/kasan/report.c:433
>  kasan_report+0xbe/0x1f0 mm/kasan/report.c:495
>  usb_udc_uevent+0x11f/0x130 drivers/usb/gadget/udc/core.c:1732
>  dev_uevent+0x290/0x770 drivers/base/core.c:2424
> ---------------------------------------------------------------
> 
> The bug occurs because usb_udc_uevent() dereferences udc->driver but
> does so without acquiring the udc_lock mutex, which protects this
> field.  If the gadget driver is unbound from the udc concurrently with
> uevent processing, the driver structure may be accessed after it has
> been deallocated.
> 
> To prevent the race, we make sure that the routine holds the mutex
> around the racing accesses.
> 
> Signed-off-by: Alan Stern <stern@rowland.harvard.edu>
> Reported-and-tested-by: syzbot+b0de012ceb1e2a97891b@syzkaller.appspotmail.com
> CC: stable@vger.kernel.org
> Link: <https://lore.kernel.org/all/0000000000004de90405a719c951@google.com>
> 
> ---
> 
> As far as I can tell, this bug has always been present.  However, the
> udc_lock mutex used by the patch was added in commit fc274c1e9973
> ("USB: gadget: Add a new bus for gadgets"), so this patch won't apply
> to trees which don't include that commit or a backport of it.  I don't
> know what tag, if any, can express this requirement.

As per the stable_kernel_rules document, you can say:
	cc: stable@vger.kernel.org	# fc274c1e9973

and I should hopefully figure it out :)

I'll add that here and see how well it works...

thanks,

greg k-h