(cc's reinstated) On Wed, 07 Mar 2007 09:09:50 +0100 Miklos Szeredi <[EMAIL PROTECTED]> wrote:
> There's a race in clear_page_dirty_for_io() that allows a page to have > cleared PG_dirty, while being mapped read-write into the page table(s). I assume you refer to this: * FIXME! We still have a race here: if somebody * adds the page back to the page tables in * between the "page_mkclean()" and the "TestClearPageDirty()", * we might have it mapped without the dirty bit set. */ if (page_mkclean(page)) set_page_dirty(page); if (TestClearPageDirty(page)) { dec_zone_page_state(page, NR_FILE_DIRTY); return 1; } I guess the comment actually refers to a writefault after the set_page_dirty() and before the TestClearPageDirty(). The fault handler will run set_page_dirty() and will return to userspace to rerun the write. The page then gets set pte-dirty but this thread of control will now make the page !PageDirty() and will write it out. With Nick's proposed lock-the-page-in-pagefaults patches, we have lock_page() synchronisation between pagefaults and clear_page_dirty_for_io() which I think will fix this. - To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to [EMAIL PROTECTED] More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/