Re: [PATCH v5 1/5] mm: add vm_insert_mixed_mkwrite()

[Kirill A. Shutemov]

On Tue, Jul 25, 2017 at 02:50:37PM +0200, Jan Kara wrote:
> On Tue 25-07-17 14:15:22, Christoph Hellwig wrote:
> > On Tue, Jul 25, 2017 at 11:35:08AM +0200, Jan Kara wrote:
> > > On Tue 25-07-17 10:01:58, Christoph Hellwig wrote:
> > > > On Tue, Jul 25, 2017 at 01:14:00AM +0300, Kirill A. Shutemov wrote:
> > > > > I guess it's up to filesystem if it wants to reuse the same spot to 
> > > > > write
> > > > > data or not. I think your assumptions works for ext4 and xfs. I 
> > > > > wouldn't
> > > > > be that sure for btrfs or other filesystems with CoW support.
> > > > 
> > > > Or XFS with reflinks for that matter.  Which currently can't be
> > > > combined with DAX, but I had a somewhat working version a few month
> > > > ago.
> > > 
> > > But in cases like COW when the block mapping changes, the process
> > > must run unmap_mapping_range() before installing the new PTE so that all
> > > processes mapping this file offset actually refault and see the new
> > > mapping. So this would go through pte_none() case. Am I missing something?
> > 
> > Yes, for DAX COW mappings we'd probably need something like this, unlike
> > the pagecache COW handling for which only the underlying block change,
> > but not the page.
> 
> Right. So again nothing where the WARN_ON should trigger.

Yes. I was confused on how COW is handled.

Acked-by: Kirill A. Shutemov 

-- 
 Kirill A. Shutemov
___
Linux-nvdimm mailing list
Linux-nvdimm@lists.01.org
https://lists.01.org/mailman/listinfo/linux-nvdimm

Re: [PATCH v5 1/5] mm: add vm_insert_mixed_mkwrite()

[Jan Kara]

On Tue 25-07-17 14:15:22, Christoph Hellwig wrote:
> On Tue, Jul 25, 2017 at 11:35:08AM +0200, Jan Kara wrote:
> > On Tue 25-07-17 10:01:58, Christoph Hellwig wrote:
> > > On Tue, Jul 25, 2017 at 01:14:00AM +0300, Kirill A. Shutemov wrote:
> > > > I guess it's up to filesystem if it wants to reuse the same spot to 
> > > > write
> > > > data or not. I think your assumptions works for ext4 and xfs. I wouldn't
> > > > be that sure for btrfs or other filesystems with CoW support.
> > > 
> > > Or XFS with reflinks for that matter.  Which currently can't be
> > > combined with DAX, but I had a somewhat working version a few month
> > > ago.
> > 
> > But in cases like COW when the block mapping changes, the process
> > must run unmap_mapping_range() before installing the new PTE so that all
> > processes mapping this file offset actually refault and see the new
> > mapping. So this would go through pte_none() case. Am I missing something?
> 
> Yes, for DAX COW mappings we'd probably need something like this, unlike
> the pagecache COW handling for which only the underlying block change,
> but not the page.

Right. So again nothing where the WARN_ON should trigger. That being said I
don't care about the WARN_ON too deeply but it can help to catch DAX bugs
so if we can keep it I'd prefer to do so...

Honza
-- 
Jan Kara 
SUSE Labs, CR
___
Linux-nvdimm mailing list
Linux-nvdimm@lists.01.org
https://lists.01.org/mailman/listinfo/linux-nvdimm

Re: [PATCH v5 1/5] mm: add vm_insert_mixed_mkwrite()

[Christoph Hellwig]

On Tue, Jul 25, 2017 at 11:35:08AM +0200, Jan Kara wrote:
> On Tue 25-07-17 10:01:58, Christoph Hellwig wrote:
> > On Tue, Jul 25, 2017 at 01:14:00AM +0300, Kirill A. Shutemov wrote:
> > > I guess it's up to filesystem if it wants to reuse the same spot to write
> > > data or not. I think your assumptions works for ext4 and xfs. I wouldn't
> > > be that sure for btrfs or other filesystems with CoW support.
> > 
> > Or XFS with reflinks for that matter.  Which currently can't be
> > combined with DAX, but I had a somewhat working version a few month
> > ago.
> 
> But in cases like COW when the block mapping changes, the process
> must run unmap_mapping_range() before installing the new PTE so that all
> processes mapping this file offset actually refault and see the new
> mapping. So this would go through pte_none() case. Am I missing something?

Yes, for DAX COW mappings we'd probably need something like this, unlike
the pagecache COW handling for which only the underlying block change,
but not the page.
___
Linux-nvdimm mailing list
Linux-nvdimm@lists.01.org
https://lists.01.org/mailman/listinfo/linux-nvdimm

Re: [PATCH v5 1/5] mm: add vm_insert_mixed_mkwrite()

[Jan Kara]

On Tue 25-07-17 10:01:58, Christoph Hellwig wrote:
> On Tue, Jul 25, 2017 at 01:14:00AM +0300, Kirill A. Shutemov wrote:
> > I guess it's up to filesystem if it wants to reuse the same spot to write
> > data or not. I think your assumptions works for ext4 and xfs. I wouldn't
> > be that sure for btrfs or other filesystems with CoW support.
> 
> Or XFS with reflinks for that matter.  Which currently can't be
> combined with DAX, but I had a somewhat working version a few month
> ago.

But in cases like COW when the block mapping changes, the process
must run unmap_mapping_range() before installing the new PTE so that all
processes mapping this file offset actually refault and see the new
mapping. So this would go through pte_none() case. Am I missing something?

Honza
-- 
Jan Kara 
SUSE Labs, CR
___
Linux-nvdimm mailing list
Linux-nvdimm@lists.01.org
https://lists.01.org/mailman/listinfo/linux-nvdimm

Re: [PATCH v5 1/5] mm: add vm_insert_mixed_mkwrite()

[Christoph Hellwig]

On Tue, Jul 25, 2017 at 01:14:00AM +0300, Kirill A. Shutemov wrote:
> I guess it's up to filesystem if it wants to reuse the same spot to write
> data or not. I think your assumptions works for ext4 and xfs. I wouldn't
> be that sure for btrfs or other filesystems with CoW support.

Or XFS with reflinks for that matter.  Which currently can't be
combined with DAX, but I had a somewhat working version a few month
ago.
___
Linux-nvdimm mailing list
Linux-nvdimm@lists.01.org
https://lists.01.org/mailman/listinfo/linux-nvdimm

Re: [PATCH v5 1/5] mm: add vm_insert_mixed_mkwrite()

[Kirill A. Shutemov]

On Mon, Jul 24, 2017 at 11:06:12AM -0600, Ross Zwisler wrote:
> @@ -1658,14 +1658,35 @@ static int insert_pfn(struct vm_area_struct *vma, 
> unsigned long addr,
>   if (!pte)
>   goto out;
>   retval = -EBUSY;
> - if (!pte_none(*pte))
> - goto out_unlock;
> + if (!pte_none(*pte)) {
> + if (mkwrite) {
> + /*
> +  * For read faults on private mappings the PFN passed
> +  * in may not match the PFN we have mapped if the
> +  * mapped PFN is a writeable COW page.  In the mkwrite
> +  * case we are creating a writable PTE for a shared
> +  * mapping and we expect the PFNs to match.
> +  */

Can we?

I guess it's up to filesystem if it wants to reuse the same spot to write
data or not. I think your assumptions works for ext4 and xfs. I wouldn't
be that sure for btrfs or other filesystems with CoW support.

-- 
 Kirill A. Shutemov
___
Linux-nvdimm mailing list
Linux-nvdimm@lists.01.org
https://lists.01.org/mailman/listinfo/linux-nvdimm