Re: [PATCH] updated radix-tree pagecache

[Hugh Dickins]

On Tue, 26 Mar 2002, Andrew Morton wrote:
> 
> Aside: This is not related to ratcache: shmem_getpage_locked() is
> setting PG_dirty but not adding the page to mapping->dirty_pages.  Is
> this intended?

Yes.  It used to be the case that if a tmpfs file page got on to
mapping->dirty_pages, fsync on that file would never escape from
filemap_fdatasync if there was no swap.  Hence also "SetPageDirty"
in several places which originally said "set_page_dirty".

Nowadays the "if (!PageLaunder(page)) return fail_writepage(page);"
at the start of shmem_writepage would prevent that hang, and
prevents a subtler tmpfs file corruption we realized later on.

But the dirty_pages list is still a waste of time for tmpfs:
its data does not need to be committed to stable storage.

Hugh