Re: [patch] mm: bug in set_page_dirty_buffers

[Andrew Morton <akpm@osdl.org>]

On Tue, 10 Oct 2006 06:47:45 +0200
Nick Piggin <npiggin@suse.de> wrote:

> On Mon, Oct 09, 2006 at 09:38:06PM -0700, Andrew Morton wrote:
> > On Tue, 10 Oct 2006 06:21:44 +0200
> > Nick Piggin <npiggin@suse.de> wrote:
> > 
> > > On Mon, Oct 09, 2006 at 09:14:04PM -0700, Andrew Morton wrote:
> > > > Can we convert set_page_dirty_balance() to call set_page_dirty_lock()?
> > > 
> > > I think so. You can't in zap_pte_range though because you're under
> > > spinlocks.
> > 
> > There we're screwed.
> > 
> > > Same with try_to_unmap_{one|cluster}, and page_remove_rmap.
> > 
> > There we can trylock all the pages and bale if any fail.
> 
> Hmm, try_to_unmap is OK because the page is already locked. page_remove_rmap
> isn't allowed to fail.

I was talking about try_to_unmap_cluster().

> > > > And make set_page_dirty_lock() return if the page is already dirty?
> > > > 
> > > > > I think there are
> > > > > a whole lot more problems than just the unmapping path, though. Direct
> > > > > IO comes to mind.
> > > > 
> > > > Why?  direct-io locks the pages while invalidating them, and while marking
> > > > them dirty.
> > > 
> > > Uh, mistaken about dio. Point still stands.
> > 
> > But where?  locking the page is the preferred way to solve this stuff. 
> > (Well, locking the buffers might work, but isn't needed, and locking the
> > page covers other stuff)
> 
> AFAIKS, it is just fs/buffer.c that is racy.

Need to review all ->set_page_dirty, ->writepage, ->invalidatepage, ->etc
implementations before we can say that.

> Why can't it use
> mapping->private_lock or the buffer bit spinlock?

block_invalidatepage() wants to do lock_buffer().

It can probably be made to work.  But a sane interface is "when dinking
with page internals, lock the page".

Which leaves us with zap_pte_range().  Perhaps one could reuse
mmu_gather.pages[]: when it's full, drop locks, dirty pages (if needed),
retake locks, resume.

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>