From mboxrd@z Thu Jan 1 00:00:00 1970 From: Edward Shishkin Subject: Re: [PATCH 3/3][reiser4] dont get radix-tree dirty tagging out of sync Date: Tue, 12 Aug 2008 16:56:16 +0400 Message-ID: <48A18870.6070500@gmail.com> References: <48A0BFC9.40502@gmail.com> <200808121236.51599.nickpiggin@yahoo.com.au> Mime-Version: 1.0 Content-Transfer-Encoding: 7bit Return-path: In-Reply-To: <200808121236.51599.nickpiggin@yahoo.com.au> Sender: reiserfs-devel-owner@vger.kernel.org List-ID: Content-Type: text/plain; charset="us-ascii" To: Nick Piggin , Ryan Hope Cc: Andrew Morton , LKML , Reiserfs mailing list , Vladimir Saveliev Nick Piggin wrote: > On Tuesday 12 August 2008 08:40, Ryan Hope wrote: > >> This was item #14 on the todo list for reiser4 inclusion in mainline: >> No. This patch is a nonsense. Where did you see radix-tree dirty tagging here? >> diff --git a/fs/reiser4/page_cache.c b/fs/reiser4/page_cache.c >> index fe71368..a662c25 100644 >> --- a/fs/reiser4/page_cache.c >> +++ b/fs/reiser4/page_cache.c >> @@ -467,15 +467,14 @@ int reiser4_set_page_dirty_internal(struct page >> *page) BUG_ON(mapping == NULL); >> >> if (!TestSetPageDirty(page)) { >> + spin_lock_irq(&mapping->tree_lock); >> if (mapping_cap_account_dirty(mapping)) >> inc_zone_page_state(page, NR_FILE_DIRTY); >> - >> + radix_tree_tag_set(&mapping->page_tree, >> + page_index(page), PAGECACHE_TAG_DIRTY); >> __mark_inode_dirty(mapping->host, I_DIRTY_PAGES); >> + spin_unlock_irq(&mapping->tree_lock); >> } >> - >> - /* znode must be dirty ? */ >> - if (mapping->host == reiser4_get_super_fake(mapping->host->i_sb)) >> - assert("", JF_ISSET(jprivate(page), JNODE_DIRTY)); >> return 0; >> } >> > > Any reason why this can't use a generic function such as > __set_page_dirty_nobuffers? Currently reiser4 is working around "anonymous" pages dirtied outside of reiser4 context (e.g. via mmap), where some reiser4 specific work (jnode creation, capturing by an atom) can not be done. > There are accounting changes gone in > there now which I suspetc may be wrong now in reiser4 (eg. task > io accounting). > Yes, this is definitely wrong. Thanks for pointing this out. Such poking around vfs internals should be fixed, otherwise we'll have permanent problems. I have cc-ed Vladimir: maybe he has some hints. At least, I know that he looked at this problem.. Thanks, Edward. > Actually every site that does a radix_tree_operation there should > be reviewed to try to use core functoins. > >