linux-kernel - Re: [PATCH] usb: cdns3: fix possible buffer overflow caused by bad DMA value

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] [thread-next>] [day] [month] [year] [list]

Message-ID: <87366b916s.fsf@kernel.org>
Date:   Wed, 01 Jul 2020 09:52:43 +0300
From:   Felipe Balbi <balbi@...nel.org>
To:     Peter Chen <peter.chen@....com>,
        Jia-Ju Bai <baijiaju1990@...il.com>
Cc:     "gregkh\@linuxfoundation.org" <gregkh@...uxfoundation.org>,
        "pawell\@cadence.com" <pawell@...ence.com>,
        "rogerq\@ti.com" <rogerq@...com>,
        "colin.king\@canonical.com" <colin.king@...onical.com>,
        "yuehaibing\@huawei.com" <yuehaibing@...wei.com>,
        "linux-usb\@vger.kernel.org" <linux-usb@...r.kernel.org>,
        "linux-kernel\@vger.kernel.org" <linux-kernel@...r.kernel.org>
Subject: Re: [PATCH] usb: cdns3: fix possible buffer overflow caused by bad DMA value

Peter Chen <peter.chen@....com> writes:

> On 20-05-30 11:24:00, Jia-Ju Bai wrote:
>> In cdns3_ep0_setup_phase():
>>   struct usb_ctrlrequest *ctrl = priv_dev->setup_buf;
>> 
>> Because priv_dev->setup_buf (allocated in cdns3_gadget_start) is stored 
>> in DMA memory, and thus ctrl is a DMA value.
>> 
>> cdns3_ep0_setup_phase()
>>   cdns3_ep0_standard_request(priv_dev, ctrl)
>>     cdns3_req_ep0_get_status(priv_dev, ctrl)
>>       index = cdns3_ep_addr_to_index(ctrl->wIndex);
>>       priv_ep = priv_dev->eps[index];
>> 
>> cdns3_ep0_setup_phase()
>>   cdns3_ep0_standard_request(priv_dev, ctrl)
>>     cdns3_req_ep0_handle_feature(priv_dev, ctrl_req, 0)
>>       cdns3_ep0_feature_handle_endpoint(priv_dev, ctrl, set)
>>         index = cdns3_ep_addr_to_index(ctrl->wIndex);
>>         priv_ep = priv_dev->eps[index];
>> 
>> In these cases, ctrl->wIndex can be be modified at anytime by malicious
>> hardware, and thus a buffer overflow can occur when the code
>> "priv_dev->eps[index]" is executed.
>> 
>
> Did you see the setup buffer is overwritten before the setup handling is
> finished?
>
>> To fix these possible bugs, index is checked before being used.
>
> I think the better fix is to use one additional buffer for struct
> usb_ctrlrequest, and copy the dma_buf to it after setup packet
> has received, then use the value stored in this buffer for later
> operation, it could avoid quitting the code which is useful in fact.

Why is this a better fix? If you don't have that endpoint index, you
shouldn't try to access it. However, I think the problem here is
slightly easier to solve :-)

>> diff --git a/drivers/usb/cdns3/ep0.c b/drivers/usb/cdns3/ep0.c
>> index e71240b386b4..0a80c7ade613 100644
>> --- a/drivers/usb/cdns3/ep0.c
>> +++ b/drivers/usb/cdns3/ep0.c
>> @@ -265,6 +265,8 @@ static int cdns3_req_ep0_get_status(struct cdns3_device *priv_dev,
>>  		return cdns3_ep0_delegate_req(priv_dev, ctrl);
>>  	case USB_RECIP_ENDPOINT:
>>  		index = cdns3_ep_addr_to_index(ctrl->wIndex);

diff --git a/drivers/usb/cdns3/gadget.c b/drivers/usb/cdns3/gadget.c
index 5e24c2e57c0d..96ba3eec805c 100644
--- a/drivers/usb/cdns3/gadget.c
+++ b/drivers/usb/cdns3/gadget.c
@@ -107,7 +107,10 @@ void cdns3_set_register_bit(void __iomem *ptr, u32 mask)
  */
 u8 cdns3_ep_addr_to_index(u8 ep_addr)
 {
-       return (((ep_addr & 0x7F)) + ((ep_addr & USB_DIR_IN) ? 16 : 0));
+       u8 num = ep_addr & USB_ENDPOINT_NUMBER_MASK;
+       u8 dir = ep_addr & USB_ENDPOINT_DIR_MASK;
+
+       return num + dir ? 16 : 0;
 }
 
 static int cdns3_get_dma_pos(struct cdns3_device *priv_dev,

This will guarantee that the number is never over the limit.

-- 
balbi

Download attachment "signature.asc" of type "application/pgp-signature" (833 bytes)