linux-kernel - Re: [PATCH] drivers: usb: host: fix NULL pointer dereferences triggered by unhandled errors in xhci_create_rhub_port

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] [day] [month] [year] [list]
Message-ID: <YmGCO0Km/pqPVgGK@kroah.com>
Date:   Thu, 21 Apr 2022 18:11:39 +0200
From:   Greg KH <gregkh@...uxfoundation.org>
To:     Mathias Nyman <mathias.nyman@...ux.intel.com>
Cc:     Fu Zixuan <r33s3n6@...il.com>,
        Heiner Kallweit <hkallweit1@...il.com>,
        mathias.nyman@...el.com, linux-usb@...r.kernel.org,
        linux-kernel@...r.kernel.org, baijiaju1990@...il.com,
        TOTE Robot <oslab@...nghua.edu.cn>
Subject: Re: [PATCH] drivers: usb: host: fix NULL pointer dereferences
 triggered by unhandled errors in xhci_create_rhub_port_array()

On Thu, Apr 21, 2022 at 03:52:40PM +0300, Mathias Nyman wrote:
> On 21.4.2022 15.21, Fu Zixuan wrote:
> > On Thu, 21 Apr 2022 at 20:06, Greg KH <gregkh@...uxfoundation.org> wrote:
> >>
> >> On Thu, Apr 21, 2022 at 07:55:28PM +0800, Fu Zixuan wrote:
> >>> On Thu, 21 Apr 2022 at 18:07, Greg KH <gregkh@...uxfoundation.org> wrote:
> >>>>
> >>>> On Thu, Apr 21, 2022 at 05:42:36PM +0800, Zixuan Fu wrote:
> >>>>> In xhci_create_rhub_port_array(), when rhub->num_ports is zero,
> >>>>> rhub->ports would not be set; when kcalloc_node() fails, rhub->ports
> >>>>> would be set to NULL. In these two cases, xhci_create_rhub_port_array()
> >>>>> just returns void, and thus its callers are unaware of the error.
> >>>>>
> >>>>> Then rhub->ports is dereferenced in xhci_usb3_hub_descriptor() or
> >>>>> xhci_usb2_hub_descriptor().
> >>>>>
> >>>>> To fix the bug, xhci_setup_port_arrays() should return an integer to
> >>>>> indicate a possible error, and its callers should handle the error.
> >>>>>
> >>>>> Here is the log when this bug occurred in our fault-injection testing:
> >>>>>
> >>>>> [   24.001309] BUG: kernel NULL pointer dereference, address: 0000000000000000
> >>>>> ...
> >>>>> [   24.003992] RIP: 0010:xhci_hub_control+0x3f5/0x60d0 [xhci_hcd]
> >>>>> ...
> >>>>> [   24.009803] Call Trace:
> >>>>> [   24.010014]  <TASK>
> >>>>> [   24.011310]  usb_hcd_submit_urb+0x1233/0x1fd0
> >>>>> [   24.017071]  usb_start_wait_urb+0x115/0x310
> >>>>> [   24.017641]  usb_control_msg+0x28a/0x450
> >>>>> [   24.019046]  hub_probe+0xb16/0x2320
> >>>>> [   24.019757]  usb_probe_interface+0x4f1/0x930
> >>>>> [   24.019765]  really_probe+0x33d/0x970
> >>>>> [   24.019768]  __driver_probe_device+0x157/0x210
> >>>>> [   24.019772]  driver_probe_device+0x4f/0x340
> >>>>> [   24.019775]  __device_attach_driver+0x2ee/0x3a0
> >>>>> ...
> >>>>>
> >>>>> Reported-by: TOTE Robot <oslab@...nghua.edu.cn>
> >>>>> Signed-off-by: Zixuan Fu <r33s3n6@...il.com>
> >>>>> ---
> >>>>>  drivers/usb/host/xhci-mem.c | 17 ++++++++++++-----
> >>>>>  1 file changed, 12 insertions(+), 5 deletions(-)
> >>>>>
> >>>>> diff --git a/drivers/usb/host/xhci-mem.c b/drivers/usb/host/xhci-mem.c
> >>>>> index bbb27ee2c6a3..024515346c39 100644
> >>>>> --- a/drivers/usb/host/xhci-mem.c
> >>>>> +++ b/drivers/usb/host/xhci-mem.c
> >>>>> @@ -2235,7 +2235,7 @@ static void xhci_add_in_port(struct xhci_hcd *xhci, unsigned int num_ports,
> >>>>>       /* FIXME: Should we disable ports not in the Extended Capabilities? */
> >>>>>  }
> >>>>>
> >>>>> -static void xhci_create_rhub_port_array(struct xhci_hcd *xhci,
> >>>>> +static int xhci_create_rhub_port_array(struct xhci_hcd *xhci,
> >>>>>                                       struct xhci_hub *rhub, gfp_t flags)
> >>>>>  {
> >>>>>       int port_index = 0;
> >>>>> @@ -2243,11 +2243,11 @@ static void xhci_create_rhub_port_array(struct xhci_hcd *xhci,
> >>>>>       struct device *dev = xhci_to_hcd(xhci)->self.sysdev;
> >>>>>
> >>>>>       if (!rhub->num_ports)
> >>>>> -             return;
> >>>>> +             return -EINVAL;
> >>>>>       rhub->ports = kcalloc_node(rhub->num_ports, sizeof(*rhub->ports),
> >>>>>                       flags, dev_to_node(dev));
> >>>>>       if (!rhub->ports)
> >>>>> -             return;
> >>>>> +             return -ENOMEM;
> >>>>>
> >>>>>       for (i = 0; i < HCS_MAX_PORTS(xhci->hcs_params1); i++) {
> >>>>>               if (xhci->hw_ports[i].rhub != rhub ||
> >>>>> @@ -2259,6 +2259,7 @@ static void xhci_create_rhub_port_array(struct xhci_hcd *xhci,
> >>>>>               if (port_index == rhub->num_ports)
> >>>>>                       break;
> >>>>>       }
> >>>>> +     return 0;
> >>>>>  }
> >>>>>
> >>>>>  /*
> >>>>> @@ -2277,6 +2278,7 @@ static int xhci_setup_port_arrays(struct xhci_hcd *xhci, gfp_t flags)
> >>>>>       int cap_count = 0;
> >>>>>       u32 cap_start;
> >>>>>       struct device *dev = xhci_to_hcd(xhci)->self.sysdev;
> >>>>> +     int ret;
> >>>>>
> >>>>>       num_ports = HCS_MAX_PORTS(xhci->hcs_params1);
> >>>>>       xhci->hw_ports = kcalloc_node(num_ports, sizeof(*xhci->hw_ports),
> >>>>> @@ -2367,8 +2369,13 @@ static int xhci_setup_port_arrays(struct xhci_hcd *xhci, gfp_t flags)
> >>>>>        * Not sure how the USB core will handle a hub with no ports...
> >>>>>        */
> >>>>>
> >>>>> -     xhci_create_rhub_port_array(xhci, &xhci->usb2_rhub, flags);
> >>>>> -     xhci_create_rhub_port_array(xhci, &xhci->usb3_rhub, flags);
> >>>>> +     ret = xhci_create_rhub_port_array(xhci, &xhci->usb2_rhub, flags);
> >>>>> +     if (ret)
> >>>>> +             return ret;
> >>>>> +
> >>>>> +     ret = xhci_create_rhub_port_array(xhci, &xhci->usb3_rhub, flags);
> >>>>> +     if (ret)
> >>>>> +             return ret;
> >>>>
> >>>> What about the memory allocated by the first call to
> >>>> xhci_create_rhub_port_array()?  Is that now lost?  Same for everything
> >>>> else allocated before these calls, how is that cleaned up properly?
> >>>>
> >>>> thanks,
> >>>>
> >>>> greg k-h
> >>>
> >>> Thanks for your swift reply. We understand your concern. In fact, we have
> >>> checked the related code carefully and found that xhci_create_rhub_port_array()
> >>> is only used in xhci_setup_port_arrays(). Moreover, only xhci_mem_init() calls
> >>> xhci_setup_port_arrays() and does all cleanup work when it fails. Specifically,
> >>> xhci_mem_init() calls xhci_mem_cleanup(), which eventually called
> >>> kfree(xhci->usb2_rhub.ports) and kfree(xhci->usb3_rhub.ports).
> >>
> >> Great, can you mention this in the changelog text to show that you have
> >> thought this through and it can be documented as such?
> >>
> >> thanks,
> >>
> >> greg k-h
> > 
> > Thanks for your reply! We will do that and submit the patch v2 soon.
> > 
> 
> Good to get this fixed, but there's a series by Heiner Kallweit that adds support
> for xHC controllers with just one roothub [1].
> It will conflict with this.  
> 
> We might need to change this a bit so that this can go to stable alone, but still
> being being able to somewhat neatly apply that new series on top of this.

As this is not anything a normal user will ever hit, it should be redone
on top of your xhci changes as it's not needed in any stable tree.

thanks,

greg k-h