Re: [PATCH 4/7] Use vfs __set_page_dirty interface instead of doing it inside filesystem

On Thu, Jul 5, 2012 at 11:20 PM, Sage Weil [off-list ref] wrote:

On Wed, 4 Jul 2012, Sha Zhengju wrote:

quoted

On 07/02/2012 10:49 PM, Sage Weil wrote:

quoted

On Mon, 2 Jul 2012, Sha Zhengju wrote:

quoted

On 06/29/2012 01:21 PM, Sage Weil wrote:

quoted

On Thu, 28 Jun 2012, Sha Zhengju wrote:

quoted

From: Sha Zhengju<redacted>

Following we will treat SetPageDirty and dirty page accounting as an
integrated
operation. Filesystems had better use vfs interface directly to avoid
those details.

Signed-off-by: Sha Zhengju<redacted>
---
   fs/buffer.c                 |    2 +-
   fs/ceph/addr.c              |   20 ++------------------
   include/linux/buffer_head.h |    2 ++
   3 files changed, 5 insertions(+), 19 deletions(-)

diff --git a/fs/buffer.c b/fs/buffer.c
index e8d96b8..55522dd 100644
--- a/fs/buffer.c
+++ b/fs/buffer.c

@@ -610,7 +610,7 @@ EXPORT_SYMBOL(mark_buffer_dirty_inode);
    * If warn is true, then emit a warning if the page is not uptodate

and
has
    * not been truncated.
    */
-static int __set_page_dirty(struct page *page,
+int __set_page_dirty(struct page *page,
              struct address_space *mapping, int warn)
   {
      if (unlikely(!mapping))

This also needs an EXPORT_SYMBOL(__set_page_dirty) to allow ceph to
continue to build as a module.

With that fixed, the ceph bits are a welcome cleanup!

Acked-by: Sage Weil<redacted>

Further, I check the path again and may it be reworked as follows to avoid
undo?

__set_page_dirty();
__set_page_dirty();
ceph operations;                ==>                     if (page->mapping)
if (page->mapping)                                            ceph
operations;
     ;
else
     undo = 1;
if (undo)
     xxx;

Yep.  Taking another look at the original code, though, I'm worried that
one reason the __set_page_dirty() actions were spread out the way they are
is because we wanted to ensure that the ceph operations were always
performed when PagePrivate was set.

Sorry, I've lost something:

__set_page_dirty();                        __set_page_dirty();
ceph operations;
if(page->mapping)         ==>      if(page->mapping) {
       SetPagePrivate;                            SetPagePrivate;
else                                                      ceph operations;
    undo = 1;                                  }

if (undo)
    XXX;

I think this can ensure that ceph operations are performed together with
SetPagePrivate.

Yeah, that looks right, as long as the ceph accounting operations happen
before SetPagePrivate.  I think it's no more or less racy than before, at
least.

The patch doesn't apply without the previous ones in the series, it looks
like.  Do you want to prepare a new version or should I?

Good. I'm doing some test then I'll send out a new version patchset, please
wait a bit. : )


Thanks,
Sha

--
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>