Re: 2.6.21-rc2-git3 Unstable for me

[Dave Kleikamp <shaggy@linux.vnet.ibm.com>]

On Mon, 2007-03-05 at 09:36 -0600, Dave Kleikamp wrote:
> On Mon, 2007-03-05 at 09:59 -0500, Chuck Ebbert wrote:
> > Will Trives wrote:
> > 
> > [adding JFS maintainer to cc:]
> 
> Thanks for adding me to the thread.
> 
> > Some kind of JFS problem?
> 
> I don' t know yet.  I was notified of the problem through bugzilla:
> http://bugzilla.kernel.org/show_bug.cgi?id=8128
> 
> The problem appears that a non-uptodate page has been marked dirty.  I
> haven't found much in the way of changes between 2.6.20 and the current
> git tree to explain this.  There's this one:
> http://git.kernel.org/gitweb.cgi?p=linux/kernel/git/torvalds/linux-2.6.git;a=commitdiff;h=22c8ca78f20724676b6006232bf06cc3e9299539
> 
> I'm looking for for a caller to ->prepare_write() which marks the page
> dirty without calling ->commit_write().

I think I just found it:

nobh_truncate_page():

	ret = a_ops->prepare_write(NULL, page, offset, to);
	if (ret == 0) {
		kaddr = kmap_atomic(page, KM_USER0);
		memset(kaddr + offset, 0, PAGE_CACHE_SIZE - offset);
		flush_dcache_page(page);
		kunmap_atomic(kaddr, KM_USER0);
		set_page_dirty(page);
	}
	unlock_page(page);
	page_cache_release(page);

> 
> Will, could you try reverting that patch?  If that fails, narrowing the
> problem down with git-bisect would be helpful, if you have the time.

Better yet, try this patch:

Signed-off-by: Dave Kleikamp <shaggy@linux.vnet.ibm.com>

diff -Nurp linux-orig/fs/buffer.c linux/fs/buffer.c
--- linux-orig/fs/buffer.c	2007-02-22 07:59:01.000000000 -0600
+++ linux/fs/buffer.c	2007-03-05 09:47:07.000000000 -0600
@@ -2466,6 +2466,7 @@ int nobh_truncate_page(struct address_sp
 		memset(kaddr + offset, 0, PAGE_CACHE_SIZE - offset);
 		flush_dcache_page(page);
 		kunmap_atomic(kaddr, KM_USER0);
+		SetPageUptodate(page);
 		set_page_dirty(page);
 	}
 	unlock_page(page);

-- 
David Kleikamp
IBM Linux Technology Center