Messages in this thread | | | Date | Tue, 17 Nov 2009 07:47:15 -0500 | From | Christoph Hellwig <> | Subject | Re: [PATCH 0/7] Kill PF_MEMALLOC abuse |
| |
On Tue, Nov 17, 2009 at 09:24:24PM +0900, KOSAKI Motohiro wrote: > > On Tue, Nov 17, 2009 at 07:24:42PM +0900, KOSAKI Motohiro wrote: > > > if xfsbufd doesn't only write out dirty data but also drop page, > > > I agree you. > > > > It then drops the reference to the buffer which drops references to the > > pages, which often are the last references, yes. > > I though it is not typical case. Am I wrong? > if so, I'm sorry. I'm not XFS expert.
I think in the typical case it's the last reference. The are two reasons why it might not be:
- we're on a filesystem with block size < page size in which case two buffers can share a page and we'd need to write out and release both buffers to free the page - someone else might have a reference on the buffer. Offhand I can't remember a place where we do this for delayed write buffers (which is what xfsbufd writes out) as it would be a bit against the purpose of those delayed write buffers.
| |