Question on set_page_dirty()

Question on set_page_dirty()

[Martin Maletinsky]

Hello,

Looking at the function set_page_dirty() (in linux 2.4.18-3 - see below) I noticed, that it not only sets the pages PG_dirty bit (as the SetPageDirty() macro does), but
additionnally may link the page onto a queue (more precisely the dirty queue of it's 'mapping').
What is the meaning of this dirty queue, what is the effect of linking a page onto that queue, and when should the set_page_dirty() function be used rather than the
SetPageDirty() macro?

Thanks in advance for any help
with best regards
Martin Maletinsky

P.S. Please put me on CC: in your reply, since I am not in the mailing list.

*
153  * Add a page to the dirty page list.
154  */
155 void set_page_dirty(struct page *page)
156 {
157         if (!test_and_set_bit(PG_dirty, &page->flags)) {
158                 struct address_space *mapping = page->mapping;
159 
160                 if (mapping) {
161                         spin_lock(&pagecache_lock);
162                         list_del(&page->list);
163                         list_add(&page->list, &mapping->dirty_pages);
164                         spin_unlock(&pagecache_lock);
165 
166                         if (mapping->host)
167                                 mark_inode_dirty_pages(mapping->host);
168                 }
169         }
170 }


--
Supercomputing System AG          email: maletinsky@scs.ch
Martin Maletinsky                 phone: +41 (0)1 445 16 05
Technoparkstrasse 1               fax:   +41 (0)1 445 16 10
CH-8005 Zurich
--
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/

Re: Question on set_page_dirty()

[Jan Hudec]

On Tue, Dec 10, 2002 at 10:59:34AM +0100, Martin Maletinsky wrote:
> Hello,
> 
> Looking at the function set_page_dirty() (in linux 2.4.18-3 - see
> below) I noticed, that it not only sets the pages PG_dirty bit (as the
> SetPageDirty() macro does), but additionnally may link the page onto
> a queue (more precisely the dirty queue of it's 'mapping').

That's the most important bit of it all. All dirty pages must at some
point be cleaned. The list keeps track of which pages need to be
cleaned, so kernel can do it quickly either when it needs to free the
mapping (close the file, terminate process, exec) or when it's just time
to flush some pages (in kflushd).

> What is the meaning of this dirty queue, what is the effect of linking
> a page onto that queue, and when should the set_page_dirty() function
> be used rather than the
> SetPageDirty() macro?

If you use the SetPageDirty macro, then the page is marked dirty, but
kernel can't find it when it should clean it. Thus it eventualy won't
flush the data (it won't call writepage on it).

-------------------------------------------------------------------------------
						 Jan 'Bulb' Hudec <bulb@ucw.cz>
--
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/

Re: Question on set_page_dirty()

[Ingo Oeser]

Hi,

On Wed, Dec 11, 2002 at 09:01:02AM +0100, Jan Hudec wrote:
> > What is the meaning of this dirty queue, what is the effect of linking
> > a page onto that queue, and when should the set_page_dirty() function
> > be used rather than the
> > SetPageDirty() macro?
> 
> If you use the SetPageDirty macro, then the page is marked dirty, but
> kernel can't find it when it should clean it. Thus it eventualy won't
> flush the data (it won't call writepage on it).

set_page_dirty() can be used in all cases, IMHO, since it:
   - will not sleep
   - will not call the set_page_dirty() method, if page has been dirty
     before (test_and_set_XXX is atomic an guarantees to trigger
     once only)
   - will not do anything besides settingt the PG_Dirty bit,
     if the page contains no mapping, or does not contain a
     set_page_dirty_method

So if set_page_dirty() exists on a certain kernel you want to
support, it should be used in all cases. Accounting code can also
be hooked into this, if it is used properly.

Regards

Ingo Oeser
-- 
Science is what we can tell a computer. Art is everything else. --- D.E.Knuth
--
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/

Re: Question on set_page_dirty()

[Stephen C. Tweedie]

Hi,

On Thu, 2002-12-12 at 09:29, Ingo Oeser wrote:
> set_page_dirty() can be used in all cases, IMHO, since it:
>    - will not sleep
...

Unfortunately, it can take both the inode_lock and pagecache_lock
spinlocks, so if you use it in the wrong place, with other locks already
held, you can cause a deadlock.  So you _do_ need to be a bit careful,
and you can't just use it with abandon.

Cheers,
 Stephen

--
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/

Re: Question on set_page_dirty()

[Andrew Morton]

"Stephen C. Tweedie" wrote:
> 
> Hi,
> 
> On Thu, 2002-12-12 at 09:29, Ingo Oeser wrote:
> > set_page_dirty() can be used in all cases, IMHO, since it:
> >    - will not sleep
> ...
> 
> Unfortunately, it can take both the inode_lock and pagecache_lock
> spinlocks, so if you use it in the wrong place, with other locks already
> held, you can cause a deadlock.  So you _do_ need to be a bit careful,
> and you can't just use it with abandon.
> 

And in 2.5 the pagecache_lock is per-inode.  This ends up meaning that
it is not legal to run set_page_dirty(page) unless the caller has
done something to prevent the thing at page->mapping from being freed.

If that has not been done, it is conceivable (but hugely unlikely) that
the page could be truncated from its mapping and that mapping could be
thrown away while set_page_dirty() is trying to claim its ->page_lock.

So in 2.5, set_page_dirty() is only legal if the caller has a ref against
page->mapping->host, or if the page is locked.

It's currently wrong in a couple of places.  I have local fixes for
the VM, and direct-IO still needs to be done.  It will just be:

	lock_page(page);		/* pins page->mapping */
	set_page_dirty(page);
	unlock_page(page);
--
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/