Re: [PATCH v4] cxl: Fix possible null pointer dereference in read_handle()

[Michael Ellerman <mpe@ellerman.id.au>]

Ma Ke <make24@iscas.ac.cn> writes:
>> Michael Ellerman<mpe@ellerman.id.au> wrote:
>> > In read_handle(), of_get_address() may return NULL if getting address and
>> > size of the node failed. When of_read_number() uses prop to handle
>> > conversions between different byte orders, it could lead to a null pointer
>> > dereference. Add NULL check to fix potential issue.
>> >
>> > Found by static analysis.
>> >
>> > Cc: stable@vger.kernel.org
>> > Fixes: 14baf4d9c739 ("cxl: Add guest-specific code")
>> > Signed-off-by: Ma Ke <make24@iscas.ac.cn>
>> > ---
>> > Changes in v4:
>> > - modified vulnerability description according to suggestions, making the 
>> > process of static analysis of vulnerabilities clearer. No active research 
>> > on developer behavior.
>> > Changes in v3:
>> > - fixed up the changelog text as suggestions.
>> > Changes in v2:
>> > - added an explanation of how the potential vulnerability was discovered,
>> > but not meet the description specification requirements.
>> > ---
>> >  drivers/misc/cxl/of.c | 2 +-
>> >  1 file changed, 1 insertion(+), 1 deletion(-)
>> >
>> > diff --git a/drivers/misc/cxl/of.c b/drivers/misc/cxl/of.c
>> > index bcc005dff1c0..d8dbb3723951 100644
>> > --- a/drivers/misc/cxl/of.c
>> > +++ b/drivers/misc/cxl/of.c
>> > @@ -58,7 +58,7 @@ static int read_handle(struct device_node *np, u64 *handle)
>> >  
>> >  	/* Get address and size of the node */
>> >  	prop = of_get_address(np, 0, &size, NULL);
>> > -	if (size)
>> > +	if (!prop || size)
>> >  		return -EINVAL;
>> >  
>> >  	/* Helper to read a big number; size is in cells (not bytes) */
>> 
>> If you expand the context this could just use of_property_read_reg(),
>> something like below.
>> 
>> cheers
>> 
>> 
>> diff --git a/drivers/misc/cxl/of.c b/drivers/misc/cxl/of.c
>> index bcc005dff1c0..a184855b2a7b 100644
>> --- a/drivers/misc/cxl/of.c
>> +++ b/drivers/misc/cxl/of.c
>> @@ -53,16 +53,15 @@ static const __be64 *read_prop64_dword(const struct device_node *np,
>>  
>>  static int read_handle(struct device_node *np, u64 *handle)
>>  {
>> -	const __be32 *prop;
>>  	u64 size;
>> +	
>> +	if (of_property_read_reg(np, 0, handle, &size))
>> +		return -EINVAL;
>>  
>> -	/* Get address and size of the node */
>> -	prop = of_get_address(np, 0, &size, NULL);
>> +	// Size must be zero per PAPR+ v2.13 § C.6.19
>>  	if (size)
>>  		return -EINVAL;
>>  
>> -	/* Helper to read a big number; size is in cells (not bytes) */
>> -	*handle = of_read_number(prop, of_n_addr_cells(np));
>>  	return 0;
>>  }

> Thank you for discussing and guiding me on the vulnerability I submitted. 
> I've carefully read through your conversation with Dan Carpenter. I'm 
> uncertain whether to use my patch or the one you provided. Could you please
> advise on which patch would be more appropriate? 
> Looking forward to your reply.

Your patch is OK, I'll send an ack.

If we want to refactor it to use of_property_read_reg() we can do that
in future - though this code will probably be removed in the not too
distant future anyway.

cheers