Re: More parallel atomic_open/d_splice_alias fun with NFS and possibly more FSes.

From: Oleg Drokin
Date: Wed Jul 06 2016 - 00:03:58 EST



On Jul 5, 2016, at 11:20 PM, Al Viro wrote:

> On Tue, Jul 05, 2016 at 08:29:37PM -0400, Oleg Drokin wrote:
>>> + /* Otherwise we just unhash it to be rehashed afresh via
>>> + * lookup if necessary
>>> + */
>>> + d_drop(dentry);
>>
>> So we can even drop this part and retain the top condition as it was.
>> d_add does not care if the dentry we are feeding it was hashed or not,
>> so do you see any downsides to doing that I wonder?
>
> d_add() on hashed dentry will end up reaching this:
> static void __d_rehash(struct dentry * entry, struct hlist_bl_head *b)
> {
> BUG_ON(!d_unhashed(entry));

Ah, ok. Yes, I remember about it now from the older issue with nfs.

It's still puzzling why I did not hit it yet, but oh well.

I wonder if handling of negative dentries broke… Time for more investigations.