[PATCH v3 0/2] usb: gadget: use after free in dev_config

[PATCH v3 0/2] usb: gadget: use after free in dev_config

[Hangyu Hua]

There are two bugs:
dev->buf does not need to be released if it already exists before
executing dev_config.
dev->config and dev->hs_config and dev->dev need to be cleaned if
dev_config fails to avoid UAF.

v2:
1. break one patch up into two separate patches.
2. use "fail:" to clear all members.

v3:
fix a mistake in [PATCH v3 2/2]

Hangyu Hua (2):
  usb: gadget: don't release an existing dev->buf
  usb: gadget: clear related members when goto fail

 drivers/usb/gadget/legacy/inode.c | 17 +++++++++++++----
 1 file changed, 13 insertions(+), 4 deletions(-)

-- 
2.25.1

[PATCH v3 1/2] usb: gadget: don't release an existing dev->buf

[Hangyu Hua]

dev->buf does not need to be released if it already exists before
executing dev_config.

Acked-by: Alan Stern <stern@rowland.harvard.edu>
Signed-off-by: Hangyu Hua <hbh25y@gmail.com>
---
 drivers/usb/gadget/legacy/inode.c | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)

diff --git a/drivers/usb/gadget/legacy/inode.c b/drivers/usb/gadget/legacy/inode.c
index 3b58f4fc0a80..eaad03c0252f 100644
--- a/drivers/usb/gadget/legacy/inode.c
+++ b/drivers/usb/gadget/legacy/inode.c
@@ -1826,8 +1826,9 @@ dev_config (struct file *fd, const char __user *buf, size_t len, loff_t *ptr)
 	spin_lock_irq (&dev->lock);
 	value = -EINVAL;
 	if (dev->buf) {
+		spin_unlock_irq(&dev->lock);
 		kfree(kbuf);
-		goto fail;
+		return value;
 	}
 	dev->buf = kbuf;
 
-- 
2.25.1

[PATCH v3 2/2] usb: gadget: clear related members when goto fail

[Hangyu Hua]

dev->config and dev->hs_config and dev->dev need to be cleaned if
dev_config fails to avoid UAF.

Acked-by: Alan Stern <stern@rowland.harvard.edu>
Signed-off-by: Hangyu Hua <hbh25y@gmail.com>
---
 drivers/usb/gadget/legacy/inode.c | 14 +++++++++++---
 1 file changed, 11 insertions(+), 3 deletions(-)

diff --git a/drivers/usb/gadget/legacy/inode.c b/drivers/usb/gadget/legacy/inode.c
index eaad03c0252f..d2e88f3b9131 100644
--- a/drivers/usb/gadget/legacy/inode.c
+++ b/drivers/usb/gadget/legacy/inode.c
@@ -1847,7 +1847,7 @@ dev_config (struct file *fd, const char __user *buf, size_t len, loff_t *ptr)
 		total = le16_to_cpu(dev->hs_config->wTotalLength);
 		if (!is_valid_config(dev->hs_config, total) ||
 				total > length - USB_DT_DEVICE_SIZE)
-			goto fail;
+			goto fail1;
 		kbuf += total;
 		length -= total;
 	} else {
@@ -1858,12 +1858,12 @@ dev_config (struct file *fd, const char __user *buf, size_t len, loff_t *ptr)
 
 	/* device descriptor (tweaked for paranoia) */
 	if (length != USB_DT_DEVICE_SIZE)
-		goto fail;
+		goto fail1;
 	dev->dev = (void *)kbuf;
 	if (dev->dev->bLength != USB_DT_DEVICE_SIZE
 			|| dev->dev->bDescriptorType != USB_DT_DEVICE
 			|| dev->dev->bNumConfigurations != 1)
-		goto fail;
+		goto fail2;
 	dev->dev->bcdUSB = cpu_to_le16 (0x0200);
 
 	/* triggers gadgetfs_bind(); then we can enumerate. */
@@ -1875,6 +1875,9 @@ dev_config (struct file *fd, const char __user *buf, size_t len, loff_t *ptr)
 
 	value = usb_gadget_probe_driver(&gadgetfs_driver);
 	if (value != 0) {
+		dev->dev = NULL;
+		dev->hs_config = NULL;
+		dev->config = NULL;
 		kfree (dev->buf);
 		dev->buf = NULL;
 	} else {
@@ -1892,7 +1895,12 @@ dev_config (struct file *fd, const char __user *buf, size_t len, loff_t *ptr)
 	}
 	return value;
 
+fail2:
+	dev->dev = NULL;
+fail1:
+	dev->hs_config = NULL;
 fail:
+	dev->config = NULL;
 	spin_unlock_irq (&dev->lock);
 	pr_debug ("%s: %s fail %zd, %p\n", shortname, __func__, value, dev);
 	kfree (dev->buf);
-- 
2.25.1

Re: [PATCH v3 2/2] usb: gadget: clear related members when goto fail

[Alan Stern]

On Thu, Dec 30, 2021 at 01:11:32PM +0800, Hangyu Hua wrote:
> dev->config and dev->hs_config and dev->dev need to be cleaned if
> dev_config fails to avoid UAF.
> 
> Acked-by: Alan Stern <stern@rowland.harvard.edu>

You must not do this.  I never sent you an Acked-by for this patch; you 
shouldn't claim that I did.

> Signed-off-by: Hangyu Hua <hbh25y@gmail.com>
> ---
>  drivers/usb/gadget/legacy/inode.c | 14 +++++++++++---
>  1 file changed, 11 insertions(+), 3 deletions(-)
> 
> diff --git a/drivers/usb/gadget/legacy/inode.c b/drivers/usb/gadget/legacy/inode.c
> index eaad03c0252f..d2e88f3b9131 100644
> --- a/drivers/usb/gadget/legacy/inode.c
> +++ b/drivers/usb/gadget/legacy/inode.c
> @@ -1847,7 +1847,7 @@ dev_config (struct file *fd, const char __user *buf, size_t len, loff_t *ptr)
>  		total = le16_to_cpu(dev->hs_config->wTotalLength);
>  		if (!is_valid_config(dev->hs_config, total) ||
>  				total > length - USB_DT_DEVICE_SIZE)
> -			goto fail;
> +			goto fail1;
>  		kbuf += total;
>  		length -= total;
>  	} else {
> @@ -1858,12 +1858,12 @@ dev_config (struct file *fd, const char __user *buf, size_t len, loff_t *ptr)
>  
>  	/* device descriptor (tweaked for paranoia) */
>  	if (length != USB_DT_DEVICE_SIZE)
> -		goto fail;
> +		goto fail1;
>  	dev->dev = (void *)kbuf;
>  	if (dev->dev->bLength != USB_DT_DEVICE_SIZE
>  			|| dev->dev->bDescriptorType != USB_DT_DEVICE
>  			|| dev->dev->bNumConfigurations != 1)
> -		goto fail;
> +		goto fail2;
>  	dev->dev->bcdUSB = cpu_to_le16 (0x0200);
>  
>  	/* triggers gadgetfs_bind(); then we can enumerate. */
> @@ -1875,6 +1875,9 @@ dev_config (struct file *fd, const char __user *buf, size_t len, loff_t *ptr)
>  
>  	value = usb_gadget_probe_driver(&gadgetfs_driver);
>  	if (value != 0) {
> +		dev->dev = NULL;
> +		dev->hs_config = NULL;
> +		dev->config = NULL;
>  		kfree (dev->buf);
>  		dev->buf = NULL;

Why not just grep the lock and goto fail?

>  	} else {
> @@ -1892,7 +1895,12 @@ dev_config (struct file *fd, const char __user *buf, size_t len, loff_t *ptr)
>  	}
>  	return value;
>  
> +fail2:
> +	dev->dev = NULL;
> +fail1:
> +	dev->hs_config = NULL;

It is not necessary to have all these different statement labels.  You 
can simply have "fail:" clear all three pointers.

>  fail:
> +	dev->config = NULL;
>  	spin_unlock_irq (&dev->lock);
>  	pr_debug ("%s: %s fail %zd, %p\n", shortname, __func__, value, dev);
>  	kfree (dev->buf);

Alan Stern

Re: [PATCH v3 2/2] usb: gadget: clear related members when goto fail

[Alan Stern]

On Thu, Dec 30, 2021 at 02:46:00PM -0500, Alan Stern wrote:
> On Thu, Dec 30, 2021 at 01:11:32PM +0800, Hangyu Hua wrote:
> > dev->config and dev->hs_config and dev->dev need to be cleaned if
> > dev_config fails to avoid UAF.

> > @@ -1875,6 +1875,9 @@ dev_config (struct file *fd, const char __user *buf, size_t len, loff_t *ptr)
> >  
> >  	value = usb_gadget_probe_driver(&gadgetfs_driver);
> >  	if (value != 0) {
> > +		dev->dev = NULL;
> > +		dev->hs_config = NULL;
> > +		dev->config = NULL;
> >  		kfree (dev->buf);
> >  		dev->buf = NULL;
> 
> Why not just grep the lock and goto fail?

Wrong word: I meant to write "grab", not "grep".

Alan Stern

Re: [PATCH v3 2/2] usb: gadget: clear related members when goto fail

[Hangyu Hua]

On 2021/12/31 上午3:46, Alan Stern wrote:
> You must not do this.  I never sent you an Acked-by for this patch; you
> shouldn't claim that I did.

I am sorry about this. I should read the linux kernel community rules
more carefully.

>>      value = usb_gadget_probe_driver(&gadgetfs_driver);
>>      if (value != 0) {
>> +            dev->dev = NULL;
>> +            dev->hs_config = NULL;
>> +            dev->config = NULL;
>>              kfree (dev->buf);
>>              dev->buf = NULL;
> Why not just grep the lock and goto fail?

You are right. I will modify my patch later.

>>      } else {
>> @@ -1892,7 +1895,12 @@ dev_config (struct file *fd, const char __user *buf, size_t len, loff_t *ptr)
>>      }
>>      return value;
>>
>> +fail2:
>> +    dev->dev = NULL;
>> +fail1:
>> +    dev->hs_config = NULL;
>
> It is not necessary to have all these different statement labels.  You
> can simply have "fail:" clear all three pointers.
>
>>   fail:
>> +    dev->config = NULL;
>>      spin_unlock_irq (&dev->lock);
>>      pr_debug ("%s: %s fail %zd, %p\n", shortname, __func__, value, dev);
>>      kfree (dev->buf);
>
> Alan Stern
>

I don't think so. It is not necessary to clean all three pointers if
some of them aren't kbuf. I think it may be better to keep their own
pointers.

Thanks. Happy new year.

Re: [PATCH v3 2/2] usb: gadget: clear related members when goto fail

[Alan Stern]

On Fri, Dec 31, 2021 at 10:31:51AM +0800, Hangyu Hua wrote:
> On 2021/12/31 上午3:46, Alan Stern wrote:

> >> @@ -1892,7 +1895,12 @@ dev_config (struct file *fd, const char __user *buf, size_t len, loff_t *ptr)
> >>      }
> >>      return value;
> >>
> >> +fail2:
> >> +    dev->dev = NULL;
> >> +fail1:
> >> +    dev->hs_config = NULL;
> >
> > It is not necessary to have all these different statement labels.  You
> > can simply have "fail:" clear all three pointers.

> I don't think so. It is not necessary to clean all three pointers if
> some of them aren't kbuf. I think it may be better to keep their own
> pointers.

If the pointers aren't set to a region inside kbuf then they are 
meaningless.  There is no reason to keep the old values.  It is better 
to avoid multiple unnecessary statement labels.

Alan Stern

Re: [PATCH v3 2/2] usb: gadget: clear related members when goto fail

[Hangyu Hua]

I get that. I will fix and resubmit my patch.

Thanks again.

On 2021/12/31 下午11:57, Alan Stern wrote:
> On Fri, Dec 31, 2021 at 10:31:51AM +0800, Hangyu Hua wrote:
>> On 2021/12/31 上午3:46, Alan Stern wrote:
> 
>>>> @@ -1892,7 +1895,12 @@ dev_config (struct file *fd, const char __user *buf, size_t len, loff_t *ptr)
>>>>       }
>>>>       return value;
>>>>
>>>> +fail2:
>>>> +    dev->dev = NULL;
>>>> +fail1:
>>>> +    dev->hs_config = NULL;
>>>
>>> It is not necessary to have all these different statement labels.  You
>>> can simply have "fail:" clear all three pointers.
> 
>> I don't think so. It is not necessary to clean all three pointers if
>> some of them aren't kbuf. I think it may be better to keep their own
>> pointers.
> 
> If the pointers aren't set to a region inside kbuf then they are
> meaningless.  There is no reason to keep the old values.  It is better
> to avoid multiple unnecessary statement labels.
> 
> Alan Stern
>