Re: [Q] missing unused dentry in prune_dcache()?

[Vasily Averin <vvs@sw.ru>]

David Howells wrote:
> Vasily Averin <vvs@sw.ru> wrote:
> 
>> Therefore I've removed break of cycle and insert this dentry to head of the
>> list. Theoretically it can lead to the second using of the same dentry,
>> however I do not think that it is a big problem.
> 
> Hmmm...  Or maybe it could be a problem.  If whenever we find the dentry we
> stick it back on the head of the list, this could be a problem as we're
> traversing the list from tail to head.

I still do not think that it could be a problem:
count argument of prune_dcache is 128 if prune_dcache is called from
shrink_dcache_memory() function and even lesser if prune_dcache is called from
shrink_dcache_parent() function. I think usually the size of unused_dentry list
(nr_usnused) is much greater and may be compared with these values only in case
of very hard memory shortage. From my point of view it is very rare situation
and I even not sure that it can really happen at all.

However even if this situation will happen I do not see here any seriously
troubles: Yes, we will try to free the same dentries, much probably without
success again, but why it is bad in case of hard memory shortage?

If my arguments are not convincing, I can protect second use of the same dentry:
we can compare counter of the skipped dentries with nr_unused value.

And what the alternatives we have?
1) We can move this dentry to end of list? But it is bad because it will prevent
shrink_dcache_memory().
2) we can move these dentries into some temporal list, and insert it back to
unused_list later? But it is bad because of we can be rescheduled and drop
dcache_lock and may confuse someone who will assume that these dentries are in
unused_list.

Therefore I believe that my patch is optimal solution.

Thank you,
	Vasily Averin
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/