Why is NFS using a_ops->freepage?

Why is NFS using a_ops->freepage?

[Jan Kara]

Hello,

I'm doing some work in page cache handling and I have noticed that NFS is
the only user of mapping->a_ops->freepage callback. From a quick look I
don't see why isn't NFS using ->releasepage / ->invalidatepage callback as
all other filesystems do? I agree you would have to set PagePrivate bit for
those to get called for the directory mapping however that would seem like
a cleaner thing to do anyway - in fact you do have private data in the
page.  Just they are not pointed to by page->private but instead are stored
as page data... Am I missing something?

								Honza
-- 
Jan Kara <jack@suse.com>
SUSE Labs, CR

Why is NFS using a_ops->freepage?

[Jan Kara]

Hello,

I'm doing some work in page cache handling and I have noticed that NFS is
the only user of mapping->a_ops->freepage callback. From a quick look I
don't see why isn't NFS using ->releasepage / ->invalidatepage callback as
all other filesystems do? I agree you would have to set PagePrivate bit for
those to get called for the directory mapping however that would seem like
a cleaner thing to do anyway - in fact you do have private data in the
page.  Just they are not pointed to by page->private but instead are stored
as page data... Am I missing something?

								Honza
-- 
Jan Kara <jack@suse.com>
SUSE Labs, CR

--
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>

Re: Why is NFS using a_ops->freepage?

[Trond Myklebust]
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Re: Why is NFS using a_ops->freepage?

[Trond Myklebust]

Hi Jan,

On Thu, 2017-10-05 at 10:36 +0200, Jan Kara wrote:
> Hello,
> 
> I'm doing some work in page cache handling and I have noticed that
> NFS is
> the only user of mapping->a_ops->freepage callback. From a quick look
> I
> don't see why isn't NFS using ->releasepage / ->invalidatepage
> callback as
> all other filesystems do? I agree you would have to set PagePrivate
> bit for
> those to get called for the directory mapping however that would seem
> like
> a cleaner thing to do anyway - in fact you do have private data in
> the
> page.  Just they are not pointed to by page->private but instead are
> stored
> as page data... Am I missing something?
> 
> 								Honza

I'm not understanding your point. delete_from_page_cache() doesn't call
releasepage AFAICS.

The point of freepage is that it is called after the page has been
removed from the page cache.

-- 
Trond Myklebust
Linux NFS client maintainer, PrimaryData
trond.myklebust@primarydata.com

Re: Why is NFS using a_ops->freepage?

[Jan Kara]

On Thu 05-10-17 13:39:23, Trond Myklebust wrote:
> Hi Jan,
> 
> On Thu, 2017-10-05 at 10:36 +0200, Jan Kara wrote:
> > Hello,
> > 
> > I'm doing some work in page cache handling and I have noticed that
> > NFS is
> > the only user of mapping->a_ops->freepage callback. From a quick look
> > I
> > don't see why isn't NFS using ->releasepage / ->invalidatepage
> > callback as
> > all other filesystems do? I agree you would have to set PagePrivate
> > bit for
> > those to get called for the directory mapping however that would seem
> > like
> > a cleaner thing to do anyway - in fact you do have private data in
> > the
> > page.  Just they are not pointed to by page->private but instead are
> > stored
> > as page data... Am I missing something?
> > 
> > 								Honza
> 
> I'm not understanding your point. delete_from_page_cache() doesn't call
> releasepage AFAICS.

No, but before getting to delete_from_page_cache() the filesystem is
guaranteed to get either ->invalidatepage or ->releasepage callback called
(if it defines them). And at that point the page is already locked and on
its way to be destroyed. So my point was you could use these callbacks
instead to achieve the same...

If you are afraid of races, I don't think those can happen for NFS. Page
can be destroyed either because of truncate - at that point there's no risk
of anyone else looking at that page for directories (i_rwsem) - or because
of page reclaim - at which point we are guaranteed nobody else holds a
reference to the page and new reference cannot be acquired.


								Honza
-- 
Jan Kara <jack@suse.com>
SUSE Labs, CR

Re: Why is NFS using a_ops->freepage?

[Jan Kara]

On Thu 05-10-17 13:39:23, Trond Myklebust wrote:
> Hi Jan,
> 
> On Thu, 2017-10-05 at 10:36 +0200, Jan Kara wrote:
> > Hello,
> > 
> > I'm doing some work in page cache handling and I have noticed that
> > NFS is
> > the only user of mapping->a_ops->freepage callback. From a quick look
> > I
> > don't see why isn't NFS using ->releasepage / ->invalidatepage
> > callback as
> > all other filesystems do? I agree you would have to set PagePrivate
> > bit for
> > those to get called for the directory mapping however that would seem
> > like
> > a cleaner thing to do anyway - in fact you do have private data in
> > the
> > page.  Just they are not pointed to by page->private but instead are
> > stored
> > as page data... Am I missing something?
> > 
> > 								Honza
> 
> I'm not understanding your point. delete_from_page_cache() doesn't call
> releasepage AFAICS.

No, but before getting to delete_from_page_cache() the filesystem is
guaranteed to get either ->invalidatepage or ->releasepage callback called
(if it defines them). And at that point the page is already locked and on
its way to be destroyed. So my point was you could use these callbacks
instead to achieve the same...

If you are afraid of races, I don't think those can happen for NFS. Page
can be destroyed either because of truncate - at that point there's no risk
of anyone else looking at that page for directories (i_rwsem) - or because
of page reclaim - at which point we are guaranteed nobody else holds a
reference to the page and new reference cannot be acquired.


								Honza
-- 
Jan Kara <jack@suse.com>
SUSE Labs, CR

--
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>