[v2] dns_resolver: assure that dns_query() result is null-terminated

dns_query() credulously assumes that keys are null-terminated and
returns a copy of a memory block that is off by one.
---
 net/dns_resolver/dns_query.c |    4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

On Sat, 7 Jun 2014, Manuel Schölling wrote:

> dns_query() credulously assumes that keys are null-terminated and
> returns a copy of a memory block that is off by one.

No sign-off?  Please read Documentation/SubmittingPatches.

> ---
>  net/dns_resolver/dns_query.c |    4 ++--
>  1 file changed, 2 insertions(+), 2 deletions(-)
> 
> diff --git a/net/dns_resolver/dns_query.c b/net/dns_resolver/dns_query.c
> index e7b6d53..84871a2 100644
> --- a/net/dns_resolver/dns_query.c
> +++ b/net/dns_resolver/dns_query.c
> @@ -145,11 +145,11 @@ int dns_query(const char *type, const char *name, size_t namelen,
>  	len = upayload->datalen;
>  
>  	ret = -ENOMEM;
> -	*_result = kmalloc(len + 1, GFP_KERNEL);
> +	*_result = kzalloc(len + 1, GFP_KERNEL);
>  	if (!*_result)
>  		goto put;
>  
> -	memcpy(*_result, upayload->data, len + 1);
> +	memcpy(*_result, upayload->data, len);
>  	if (_expiry)
>  		*_expiry = rkey->expiry;
>  

kzalloc() would be unnecessary overhead (zeroing definitely comes with a 
cost) if you're going to copy to the memory immediately afterwards.  Just 
leave the kmalloc(), do the memcpy() and explicitly zero terminate it 
_result.

On Sa, 2014-06-07 at 14:42 -0700, David Rientjes wrote:
> On Sat, 7 Jun 2014, Manuel Schölling wrote:
> 
> > dns_query() credulously assumes that keys are null-terminated and
> > returns a copy of a memory block that is off by one.
> 
> No sign-off?  Please read Documentation/SubmittingPatches.
It's just not my day today.
Sorry, I forgot about the sign-off.

> > ---
> >  net/dns_resolver/dns_query.c |    4 ++--
> >  1 file changed, 2 insertions(+), 2 deletions(-)
> > 
> > diff --git a/net/dns_resolver/dns_query.c b/net/dns_resolver/dns_query.c
> > index e7b6d53..84871a2 100644
> > --- a/net/dns_resolver/dns_query.c
> > +++ b/net/dns_resolver/dns_query.c
> > @@ -145,11 +145,11 @@ int dns_query(const char *type, const char *name, size_t namelen,
> >  	len = upayload->datalen;
> >  
> >  	ret = -ENOMEM;
> > -	*_result = kmalloc(len + 1, GFP_KERNEL);
> > +	*_result = kzalloc(len + 1, GFP_KERNEL);
> >  	if (!*_result)
> >  		goto put;
> >  
> > -	memcpy(*_result, upayload->data, len + 1);
> > +	memcpy(*_result, upayload->data, len);
> >  	if (_expiry)
> >  		*_expiry = rkey->expiry;
> >  
> 
> kzalloc() would be unnecessary overhead (zeroing definitely comes with a 
> cost) if you're going to copy to the memory immediately afterwards.  Just 
> leave the kmalloc(), do the memcpy() and explicitly zero terminate it 
> _result.

Using kzalloc() was suggested of a developer on IRC (#kernelnewbies) but
if you prefer kmalloc, that's ok, too.
I'll send you a corrected patch in a second.


--
To unsubscribe from this list: send the line "unsubscribe netdev" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

On 06/08/2014 01:42 AM, David Rientjes wrote:

>> dns_query() credulously assumes that keys are null-terminated and
>> returns a copy of a memory block that is off by one.

> No sign-off?  Please read Documentation/SubmittingPatches.

>> ---
>>   net/dns_resolver/dns_query.c |    4 ++--
>>   1 file changed, 2 insertions(+), 2 deletions(-)
>>
>> diff --git a/net/dns_resolver/dns_query.c b/net/dns_resolver/dns_query.c
>> index e7b6d53..84871a2 100644
>> --- a/net/dns_resolver/dns_query.c
>> +++ b/net/dns_resolver/dns_query.c
>> @@ -145,11 +145,11 @@ int dns_query(const char *type, const char *name, size_t namelen,
>>   	len = upayload->datalen;
>>
>>   	ret = -ENOMEM;
>> -	*_result = kmalloc(len + 1, GFP_KERNEL);
>> +	*_result = kzalloc(len + 1, GFP_KERNEL);
>>   	if (!*_result)
>>   		goto put;
>>
>> -	memcpy(*_result, upayload->data, len + 1);
>> +	memcpy(*_result, upayload->data, len);
>>   	if (_expiry)
>>   		*_expiry = rkey->expiry;

> kzalloc() would be unnecessary overhead (zeroing definitely comes with a
> cost) if you're going to copy to the memory immediately afterwards.  Just
> leave the kmalloc(), do the memcpy() and explicitly zero terminate it
> _result.

    You can also replace kmalloc()/memcpy() with kmemdup().

WBR, Sergei

--
To unsubscribe from this list: send the line "unsubscribe netdev" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

On Sat, 7 Jun 2014, Manuel Schoelling wrote:

> > kzalloc() would be unnecessary overhead (zeroing definitely comes with a 
> > cost) if you're going to copy to the memory immediately afterwards.  Just 
> > leave the kmalloc(), do the memcpy() and explicitly zero terminate it 
> > _result.
> 
> Using kzalloc() was suggested of a developer on IRC (#kernelnewbies) but
> if you prefer kmalloc, that's ok, too.
> I'll send you a corrected patch in a second.
> 

Using kzalloc() here instead of kmalloc() is functionally equivalent to

	if (*_result) {
		memset(*_result, 0, len + 1);
		memcpy(*_result, upayload->data, len);
	}

so for anything with len > 1 there is an unnecessary overhead in doing 
this.  k?alloc() can return object sizes larger than len + 1 here as well 
(usually power-of-2 sizes are supported by the slab allocator) so 
depending on the value of len, you may be zeroing more memory than 
copying.

Your first patch had the right idea, it's just off by one.
--
To unsubscribe from this list: send the line "unsubscribe netdev" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html