Re: [PATCH v4 1/1] dcache: Translating dentry into pathname without taking rename_lock

[Waiman Long <waiman.long@hp.com>]

On 09/09/2013 03:28 PM, Al Viro wrote:
> On Mon, Sep 09, 2013 at 08:10:29PM +0100, Al Viro wrote:
>> On Mon, Sep 09, 2013 at 02:46:57PM -0400, Waiman Long wrote:
>>
>>> I am fine with your proposed change as long as it gets the job done.
>> I suspect that the real problem is the unlock part of read_seqretry_or_unlock();
>> for d_walk() we want to be able to check if we need retry and continue walking
>> if we do not.  Let's do it that way: I've applied your patch as is, with the
>> next step being
>> 	* split read_seqretry_or_unlock():
>> need_seqretry() (return (!(seq&  1)&&  read_seqretry(lock, seq))
>> done_seqretry() (if (seq&  1) write_sequnlock(lock, seq)),
>> your if (read_seqretry_or_unlock(&rename_lock,&seq))
>> 		goto restart;
>> becoming
>> 	if (need_seqretry(&rename_lock, seq)) {
>> 		seq = 1;
>> 		goto restart;
>> 	}
>> 	done_seqretry(&rename_lock, seq);
>>
>> Then d_walk() is trivially massaged to use of read_seqbegin_or_lock(),
>> need_seqretry() and done_seqretry().  Give me a few, I'll post it...
> OK, how about this?  It splits read_seqretry_or_unlock(), takes
> rcu_read_{lock,unlock} in the callers and converts d_walk() to those
> primitives.  I've pushed that and your commit into vfs.git#experimental
> (head at 48f5ec2, should propagate in a few); guys, please give it a look
> and comment.

The changes look good to me. I was planning to take rcu_read_lock() out 
and doing something similar, but your change is good. BTW, I think Linus 
want to add some comments on why RCU lock is needed without the 
rename_lock, but I can put that in with a follow-up patch once the 
current change is merged.

Thank for your help and inspiration on this patch.

-Longman