On Tue, 12 Jun 2001, Alexander Viro wrote:
> 
> On Tue, 12 Jun 2001, Steve Lord wrote:
> 
> > We now return to truncate_complete_page() and remove the page from the inode,
> > the only reference to the page is now the buffer head.
> 
> The real problem is in block_flushpage() - we blindly convert page into
> buffer cache one, polluting the latter with highmem pages.

Tell me the problem again..

Yes, we will convert a mapped page into a "buffer page", but we will _not_
add any of the buffers to any LRU lists or anything like that, so the
buffers won't be reachable from anything that they weren't reachable from
before already.

So either the buffers had been reachable before - in which case the
anonymization adds no new cases, or they had not been reachable before -
in which case the anonymization will basically "lose" the buffers until
the page can later be reclaimed.

> Do we really need that logics at all? Reuse of page cache pages for
> buffer cache, that is...

We _do_ need the logic. And don't think of it as a "reuse of page cache
pages" issue. Think of it as a pure memory management issue - we will
release the page cache part of the page count, but we'll leave the buffers
untouched because somebody is still using them. It's really nothing but a
"the page is still busy, we'll delay the free" thing.

                Linus

-
To unsubscribe from this list: send the line "unsubscribe linux-fsdevel" in
the body of a message to [EMAIL PROTECTED]

Reply via email to