Re: [PATCH] fs/eventpoll: error handling micro-cleanup

[Jeff Garzik <jeff@garzik.org>]

Davide Libenzi wrote:
> On Sun, 1 Oct 2006, Jeff Garzik wrote:
> 
>> Davide Libenzi wrote:
>>
>>> I just tried a `find /usr/src/linux-2.6.16/ -type f -exec grep -H -C 2
>>> PTR_ERR {} \;`
>>> and looked at the cases where the error variable is assigned in any case
>>> before the test. Same code pattern as, like:
>>>
>>> error = -EFAULT;
>>> if (copy_from_user(...))
>>> 	goto kaboom;
>> No, that's quite different.  I'm talking about
>>
>> 	ptr = get_a_pointer_from_somewhere()
>> 	error = PTR_ERR(ptr)
>>
>> See the difference?  The error variable is directly assigned from a
>> potentially-valid pointer.
> 
> So? Is PTR_ERR() defined and documented in a way that, if called with a 
> valid pointer, has an unexpected/faulty behaviour?

When called with a valid pointer, the value assigned to the return-code 
integer is essentially a random number.


> Again, I don't care either ways, but don't tell me you're not sure about 
> the countless occurrences. Take a look at:
> 
> `find $LINUXSRC -type f -exec grep -H -C 2 PTR_ERR {} \;`

Perhaps 1 out of every 100 or so hits from this find(1) is unprotected 
by IS_ERR().  IOW, what I've been describing here is quite rare.

	Jeff