kread() and NULL pointers

kread() and NULL pointers

[Mike Mason]

Does kread() generate an error if you pass it a NULL pointer?  I'm trying to determine if NULL pointer checks are still necessary if I use kread().  Some earlier email implies that they aren't, but I want to be certain.

Thanks,
Mike

Re: kread() and NULL pointers

[Roland McGrath]

> Does kread() generate an error if you pass it a NULL pointer?  

Yes.  It's safely caught like all errors, but it only returns successfully
when given a valid pointer.

Re: kread() and NULL pointers

[Mike Mason]

Roland McGrath wrote:
>> Does kread() generate an error if you pass it a NULL pointer?  
> 
> Yes.  It's safely caught like all errors, but it only returns successfully
> when given a valid pointer.

So should the following generate an error?  It doesn't.  It just prints "ptr = 0" and exits.

function test_addr:long () %{
        void *nullptr = NULL;
        THIS->__retvalue = (long) kread(&(nullptr));
        CATCH_DEREF_FAULT();
%}


probe begin {
        ptr = test_addr()
        printf("ptr = %d\n", ptr)
        exit()
}

Re: kread() and NULL pointers

[Stone, Joshua I]

Mike Mason wrote:
> So should the following generate an error?  It doesn't.  It just prints 
> "ptr = 0" and exits.
> 
> function test_addr:long () %{
>        void *nullptr = NULL;
>        THIS->__retvalue = (long) kread(&(nullptr));
-----------------------------------------^

You're reading the address *of* your pointer, not the address in the 
pointer.  Thus kread() is dereferencing an address on the stack, and the 
value there is NULL.  Drop the '&', and change nullptr to a type that's 
meaningful to dereference (e.g., long*), and you'll get your error.

The "&(xxx)" pattern you see everywhere is because usually the "xxx" is 
a struct value, and so you pass kread() a pointer to that value.


Josh

Re: kread() and NULL pointers

[Roland McGrath]

> Roland McGrath wrote:
> >> Does kread() generate an error if you pass it a NULL pointer?  
> > 
> > Yes.  It's safely caught like all errors, but it only returns successfully
> > when given a valid pointer.
> 
> So should the following generate an error?  It doesn't.  It just prints "ptr = 0" and exits.
> 
> function test_addr:long () %{
>         void *nullptr = NULL;
>         THIS->__retvalue = (long) kread(&(nullptr));
>         CATCH_DEREF_FAULT();
> %}

As noted, this is not kread(NULL).  As to the question, CATCH_DEREF_FAULT
is the code that constitutes what "generate an error" means.  kread is a
macro using deref, which does "goto deref_fault" for errors.  And then:

	#define CATCH_DEREF_FAULT()				\
	  if (0) {						\
	deref_fault:						\
	    CONTEXT->last_error = "pointer dereference fault";	\
	  }

Re: kread() and NULL pointers

[Mike Mason]

Stone, Joshua I wrote:
> Mike Mason wrote:
>> So should the following generate an error?  It doesn't.  It just 
>> prints "ptr = 0" and exits.
>>
>> function test_addr:long () %{
>>        void *nullptr = NULL;
>>        THIS->__retvalue = (long) kread(&(nullptr));
> -----------------------------------------^
> 
> You're reading the address *of* your pointer, not the address in the 
> pointer.  Thus kread() is dereferencing an address on the stack, and the 
> value there is NULL.  Drop the '&', and change nullptr to a type that's 
> meaningful to dereference (e.g., long*), and you'll get your error.

You're right.  I fell into the trap of assuming the kread parameter should always have a '&' in front.  Bad assumption on my part.

> 
> The "&(xxx)" pattern you see everywhere is because usually the "xxx" is 
> a struct value, and so you pass kread() a pointer to that value.

Yep, this is what fooled me.

Never mind :-)

Thanks,
Mike

> 
> 
> Josh