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

[Al Viro <viro@ZenIV.linux.org.uk>]

On Mon, Sep 09, 2013 at 07:06:04PM +0100, Al Viro wrote:
> I like my proposal for the set of primitives better:
> 
> static inline bool seqretry_and_lock(seqlock_t *lock, unsigned *seq):
> {
>         if ((*seq & 1) || !read_seqretry(lock, *seq))
>                 return true;
>         *seq |= 1;
>         write_seqlock(lock);
>         return false;
> }
> 
> static inline void seqretry_done(seqlock_t *lock, unsigned seq)
> {
>         if (seq & 1)
>                 write_sequnlock(lock);
> }
> 
> with the prepend_path() and friends becoming
> 
> 	rcu_read_lock();
> 	seq = read_seqbegin(&rename_lock);
> again:
> 	....
> 	if (!seqretry_and_lock(&rename_lock, seq))
> 		goto again;	/* now as writer */
> 	seqretry_done(&rename_lock, seq);
> 	rcu_read_unlock();

Actually, it's better for prepend_path() as well, because it's actually
 
	rcu_read_lock();
	seq = read_seqbegin(&rename_lock);
again:
	....
	if (error)
		goto done;
	....
	if (!seqretry_and_lock(&rename_lock, seq))
		goto again;	/* now as writer */
done:
	seqretry_done(&rename_lock, seq);
	rcu_read_unlock();

Posted variant will sometimes hit the following path:
	* seq_readlock()
	* start generating the output
	* hit an error
[another process has taken and released rename_lock for some reason]
	* hit read_seqretry_and_unlock(), which returns 1.
	* retry everything with seq_writelock(), despite the error.

It's not too horrible (we won't be looping indefinitely, ignoring error
all along), but it's certainly subtle enough...