Re: [PATCH 3/3] VFS: close race between getcwd() and d_move()

[Nikolay Borisov]

On  9.11.2017 15:08, Matthew Wilcox wrote:
> On Thu, Nov 09, 2017 at 01:41:24PM +0200, Nikolay Borisov wrote:
>> On  9.11.2017 05:22, NeilBrown wrote:
>>> @@ -493,12 +496,18 @@ void __d_drop(struct dentry *dentry)
>>>  		} else
>>>  			hlist_bl_lock(b);
>>>  		__hlist_bl_del(&dentry->d_hash);
>>> -		dentry->d_hash.pprev = NULL;
>>> +		if (likely(!moving))
>>> +			dentry->d_hash.pprev = NULL;
>>
>> nit: isn't a bit more explicit if (unlikely(moving)). I suspect the end
>> result is the same, however it's easy to miss the !. It's not a big deal
>> but just wondering.
> 
> umm ... you just suggested the exact opposite of what the patch is
> intended to do.  likely()/unlikely() only hint to the compiler the
> probabilities of the branch; they don't change the meaning of the
> condition.

brainfart, disregard my comment doh....

> 
>