[RFC,09/10,v1,RESEND] ext4: don't need to writeout all dirty pages in punch hole

From: Zheng Liu <wenqing.lz@taobao.com>

Now we don't need to writeout all dirty pages when punching a hole.  The i_mutex
locking is taken to avoid concurrent writes.  In truncate_pagecache_range, all
pages in this hole is released, and ext4_es_remove_space is called to update
extent status tree. 

Signed-off-by: Zheng Liu <wenqing.lz@taobao.com>
---
 fs/ext4/extents.c |   28 ++++++++++++----------------
 1 files changed, 12 insertions(+), 16 deletions(-)

On Sun, 22 Jul 2012, Zheng Liu wrote:

> Date: Sun, 22 Jul 2012 15:59:45 +0800
> From: Zheng Liu <gnehzuil.liu@gmail.com>
> To: linux-ext4@vger.kernel.org, linux-fsdevel@vger.kernel.org
> Cc: xiaoqiangnk@gmail.com, achender@linux.vnet.ibm.com, wenqing.lz@taobao.com
> Subject: [RFC][PATCH 09/10 v1][RESEND] ext4: don't need to writeout all dirty
>     pages in punch hole
> 
> From: Zheng Liu <wenqing.lz@taobao.com>
> 
> Now we don't need to writeout all dirty pages when punching a hole.  The i_mutex
> locking is taken to avoid concurrent writes.  In truncate_pagecache_range, all
> pages in this hole is released, and ext4_es_remove_space is called to update
> extent status tree. 

Hi Zheng,

I am currently in the middle of reworking punch hole for ext4 so
there will be some changes in this area. See the patch set

http://www.spinics.net/lists/linux-ext4/msg33014.html

Moreover I think that we should avoid taking i_mutex if we can and I
believe that we can in this case, because we only need to prevent
allocation. So I just want to let you know that this part is
probably going to change anyway.

Thanks!
-Lukas

> 
> Signed-off-by: Zheng Liu <wenqing.lz@taobao.com>
> ---
>  fs/ext4/extents.c |   28 ++++++++++++----------------
>  1 files changed, 12 insertions(+), 16 deletions(-)
> 
> diff --git a/fs/ext4/extents.c b/fs/ext4/extents.c
> index f2c5294..2a526b4 100644
> --- a/fs/ext4/extents.c
> +++ b/fs/ext4/extents.c
> @@ -4529,9 +4529,11 @@ int ext4_ext_punch_hole(struct file *file, loff_t offset, loff_t length)
>  	loff_t first_page_offset, last_page_offset;
>  	int credits, err = 0;
>  
> +	mutex_lock(&inode->i_mutex);
> +
>  	/* No need to punch hole beyond i_size */
>  	if (offset >= inode->i_size)
> -		return 0;
> +		goto error;
>  
>  	/*
>  	 * If the hole extends beyond i_size, set the hole
> @@ -4549,18 +4551,6 @@ int ext4_ext_punch_hole(struct file *file, loff_t offset, loff_t length)
>  	first_page_offset = first_page << PAGE_CACHE_SHIFT;
>  	last_page_offset = last_page << PAGE_CACHE_SHIFT;
>  
> -	/*
> -	 * Write out all dirty pages to avoid race conditions
> -	 * Then release them.
> -	 */
> -	if (mapping->nrpages && mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) {
> -		err = filemap_write_and_wait_range(mapping,
> -			offset, offset + length - 1);
> -
> -		if (err)
> -			return err;
> -	}
> -
>  	/* Now release the pages */
>  	if (last_page_offset > first_page_offset) {
>  		truncate_pagecache_range(inode, first_page_offset,
> @@ -4572,12 +4562,14 @@ int ext4_ext_punch_hole(struct file *file, loff_t offset, loff_t length)
>  
>  	credits = ext4_writepage_trans_blocks(inode);
>  	handle = ext4_journal_start(inode, credits);
> -	if (IS_ERR(handle))
> -		return PTR_ERR(handle);
> +	if (IS_ERR(handle)) {
> +		err = PTR_ERR(handle);
> +		goto error;
> +	}
>  
>  	err = ext4_orphan_add(handle, inode);
>  	if (err)
> -		goto out;
> +		goto error;
>  
>  	/*
>  	 * Now we need to zero out the non-page-aligned data in the
> @@ -4652,6 +4644,8 @@ int ext4_ext_punch_hole(struct file *file, loff_t offset, loff_t length)
>  	ext4_ext_invalidate_cache(inode);
>  	ext4_discard_preallocations(inode);
>  
> +	err = ext4_es_remove_space(inode, first_block,
> +				   stop_block - first_block);
>  	err = ext4_ext_remove_space(inode, first_block, stop_block - 1);
>  
>  	ext4_ext_invalidate_cache(inode);
> @@ -4667,6 +4661,8 @@ out:
>  	inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
>  	ext4_mark_inode_dirty(handle, inode);
>  	ext4_journal_stop(handle);
> +error:
> +	mutex_unlock(&inode->i_mutex);
>  	return err;
>  }
>  int ext4_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
> 
--
To unsubscribe from this list: send the line "unsubscribe linux-ext4" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Hi Lukas,

On Mon, Jul 23, 2012 at 01:01:53PM +0200, Lukáš Czerner wrote:
> On Sun, 22 Jul 2012, Zheng Liu wrote:
> 
> > Date: Sun, 22 Jul 2012 15:59:45 +0800
> > From: Zheng Liu <gnehzuil.liu@gmail.com>
> > To: linux-ext4@vger.kernel.org, linux-fsdevel@vger.kernel.org
> > Cc: xiaoqiangnk@gmail.com, achender@linux.vnet.ibm.com, wenqing.lz@taobao.com
> > Subject: [RFC][PATCH 09/10 v1][RESEND] ext4: don't need to writeout all dirty
> >     pages in punch hole
> > 
> > From: Zheng Liu <wenqing.lz@taobao.com>
> > 
> > Now we don't need to writeout all dirty pages when punching a hole.  The i_mutex
> > locking is taken to avoid concurrent writes.  In truncate_pagecache_range, all
> > pages in this hole is released, and ext4_es_remove_space is called to update
> > extent status tree. 
> 
> Hi Zheng,
> 
> I am currently in the middle of reworking punch hole for ext4 so
> there will be some changes in this area. See the patch set
> 
> http://www.spinics.net/lists/linux-ext4/msg33014.html

Thank you for your reminding.  I have seen your patch set and it is
pretty cool.  IMHO your patch set shall be merged into upstream kernel
before applying io tree because io tree still has a lot of things that
needs to be done.  So I will change it according to your patches. :-)

> Moreover I think that we should avoid taking i_mutex if we can and I
> believe that we can in this case, because we only need to prevent
> allocation. So I just want to let you know that this part is
> probably going to change anyway.

It seems that we need to take i_mutex locking to prevent from buffered
writes after page cache has been truncated by truncate_pagecache_range.
If a buffered write without delalloc occurs after truncating page cache
and before taking i_data_sem, that means that the allocated block for
this buffered write will be removed in ext4_ext_remove_space when the
offset is within the range of the hole.  Am I missing something?

Regards,
Zheng

> > 
> > Signed-off-by: Zheng Liu <wenqing.lz@taobao.com>
> > ---
> >  fs/ext4/extents.c |   28 ++++++++++++----------------
> >  1 files changed, 12 insertions(+), 16 deletions(-)
> > 
> > diff --git a/fs/ext4/extents.c b/fs/ext4/extents.c
> > index f2c5294..2a526b4 100644
> > --- a/fs/ext4/extents.c
> > +++ b/fs/ext4/extents.c
> > @@ -4529,9 +4529,11 @@ int ext4_ext_punch_hole(struct file *file, loff_t offset, loff_t length)
> >  	loff_t first_page_offset, last_page_offset;
> >  	int credits, err = 0;
> >  
> > +	mutex_lock(&inode->i_mutex);
> > +
> >  	/* No need to punch hole beyond i_size */
> >  	if (offset >= inode->i_size)
> > -		return 0;
> > +		goto error;
> >  
> >  	/*
> >  	 * If the hole extends beyond i_size, set the hole
> > @@ -4549,18 +4551,6 @@ int ext4_ext_punch_hole(struct file *file, loff_t offset, loff_t length)
> >  	first_page_offset = first_page << PAGE_CACHE_SHIFT;
> >  	last_page_offset = last_page << PAGE_CACHE_SHIFT;
> >  
> > -	/*
> > -	 * Write out all dirty pages to avoid race conditions
> > -	 * Then release them.
> > -	 */
> > -	if (mapping->nrpages && mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) {
> > -		err = filemap_write_and_wait_range(mapping,
> > -			offset, offset + length - 1);
> > -
> > -		if (err)
> > -			return err;
> > -	}
> > -
> >  	/* Now release the pages */
> >  	if (last_page_offset > first_page_offset) {
> >  		truncate_pagecache_range(inode, first_page_offset,
> > @@ -4572,12 +4562,14 @@ int ext4_ext_punch_hole(struct file *file, loff_t offset, loff_t length)
> >  
> >  	credits = ext4_writepage_trans_blocks(inode);
> >  	handle = ext4_journal_start(inode, credits);
> > -	if (IS_ERR(handle))
> > -		return PTR_ERR(handle);
> > +	if (IS_ERR(handle)) {
> > +		err = PTR_ERR(handle);
> > +		goto error;
> > +	}
> >  
> >  	err = ext4_orphan_add(handle, inode);
> >  	if (err)
> > -		goto out;
> > +		goto error;
> >  
> >  	/*
> >  	 * Now we need to zero out the non-page-aligned data in the
> > @@ -4652,6 +4644,8 @@ int ext4_ext_punch_hole(struct file *file, loff_t offset, loff_t length)
> >  	ext4_ext_invalidate_cache(inode);
> >  	ext4_discard_preallocations(inode);
> >  
> > +	err = ext4_es_remove_space(inode, first_block,
> > +				   stop_block - first_block);
> >  	err = ext4_ext_remove_space(inode, first_block, stop_block - 1);
> >  
> >  	ext4_ext_invalidate_cache(inode);
> > @@ -4667,6 +4661,8 @@ out:
> >  	inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
> >  	ext4_mark_inode_dirty(handle, inode);
> >  	ext4_journal_stop(handle);
> > +error:
> > +	mutex_unlock(&inode->i_mutex);
> >  	return err;
> >  }
> >  int ext4_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
> > 
--
To unsubscribe from this list: send the line "unsubscribe linux-ext4" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

On Mon, 23 Jul 2012, Zheng Liu wrote:

> Date: Mon, 23 Jul 2012 19:57:22 +0800
> From: Zheng Liu <gnehzuil.liu@gmail.com>
> To: Lukáš Czerner <lczerner@redhat.com>
> Cc: linux-ext4@vger.kernel.org, linux-fsdevel@vger.kernel.org,
>     xiaoqiangnk@gmail.com, achender@linux.vnet.ibm.com, wenqing.lz@taobao.com
> Subject: Re: [RFC][PATCH 09/10 v1][RESEND] ext4: don't need to writeout all
>     dirty pages in punch hole
> 
> Hi Lukas,
> 
> On Mon, Jul 23, 2012 at 01:01:53PM +0200, Lukáš Czerner wrote:
> > On Sun, 22 Jul 2012, Zheng Liu wrote:
> > 
> > > Date: Sun, 22 Jul 2012 15:59:45 +0800
> > > From: Zheng Liu <gnehzuil.liu@gmail.com>
> > > To: linux-ext4@vger.kernel.org, linux-fsdevel@vger.kernel.org
> > > Cc: xiaoqiangnk@gmail.com, achender@linux.vnet.ibm.com, wenqing.lz@taobao.com
> > > Subject: [RFC][PATCH 09/10 v1][RESEND] ext4: don't need to writeout all dirty
> > >     pages in punch hole
> > > 
> > > From: Zheng Liu <wenqing.lz@taobao.com>
> > > 
> > > Now we don't need to writeout all dirty pages when punching a hole.  The i_mutex
> > > locking is taken to avoid concurrent writes.  In truncate_pagecache_range, all
> > > pages in this hole is released, and ext4_es_remove_space is called to update
> > > extent status tree. 
> > 
> > Hi Zheng,
> > 
> > I am currently in the middle of reworking punch hole for ext4 so
> > there will be some changes in this area. See the patch set
> > 
> > http://www.spinics.net/lists/linux-ext4/msg33014.html
> 
> Thank you for your reminding.  I have seen your patch set and it is
> pretty cool.  IMHO your patch set shall be merged into upstream kernel
> before applying io tree because io tree still has a lot of things that
> needs to be done.  So I will change it according to your patches. :-)

Great, thanks!

> 
> > Moreover I think that we should avoid taking i_mutex if we can and I
> > believe that we can in this case, because we only need to prevent
> > allocation. So I just want to let you know that this part is
> > probably going to change anyway.
> 
> It seems that we need to take i_mutex locking to prevent from buffered
> writes after page cache has been truncated by truncate_pagecache_range.
> If a buffered write without delalloc occurs after truncating page cache
> and before taking i_data_sem, that means that the allocated block for
> this buffered write will be removed in ext4_ext_remove_space when the
> offset is within the range of the hole.  Am I missing something?

You're absolutely right, currently this is possible. But I think that we
can take i_data_sem before truncating the pagecache hence preventing anyone
from mapping new blocks. However this is not yet implemented in my
patch set.

...
hmm, looking at the ext4_write_begin() it seems like it might not be
such good idea after all. It seems to take page lock before
i_data_sem so we might get deadlock, moreover if the punch hole
happened in the middle of the ext4_write_begin() we might have only
part of the data written, moreover this does not have to be hole
aligned, which is bad. I need to revise that.

Thanks!
-Lukas


> 
> Regards,
> Zheng
> 
> > > 
> > > Signed-off-by: Zheng Liu <wenqing.lz@taobao.com>
> > > ---
> > >  fs/ext4/extents.c |   28 ++++++++++++----------------
> > >  1 files changed, 12 insertions(+), 16 deletions(-)
> > > 
> > > diff --git a/fs/ext4/extents.c b/fs/ext4/extents.c
> > > index f2c5294..2a526b4 100644
> > > --- a/fs/ext4/extents.c
> > > +++ b/fs/ext4/extents.c
> > > @@ -4529,9 +4529,11 @@ int ext4_ext_punch_hole(struct file *file, loff_t offset, loff_t length)
> > >  	loff_t first_page_offset, last_page_offset;
> > >  	int credits, err = 0;
> > >  
> > > +	mutex_lock(&inode->i_mutex);
> > > +
> > >  	/* No need to punch hole beyond i_size */
> > >  	if (offset >= inode->i_size)
> > > -		return 0;
> > > +		goto error;
> > >  
> > >  	/*
> > >  	 * If the hole extends beyond i_size, set the hole
> > > @@ -4549,18 +4551,6 @@ int ext4_ext_punch_hole(struct file *file, loff_t offset, loff_t length)
> > >  	first_page_offset = first_page << PAGE_CACHE_SHIFT;
> > >  	last_page_offset = last_page << PAGE_CACHE_SHIFT;
> > >  
> > > -	/*
> > > -	 * Write out all dirty pages to avoid race conditions
> > > -	 * Then release them.
> > > -	 */
> > > -	if (mapping->nrpages && mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) {
> > > -		err = filemap_write_and_wait_range(mapping,
> > > -			offset, offset + length - 1);
> > > -
> > > -		if (err)
> > > -			return err;
> > > -	}
> > > -
> > >  	/* Now release the pages */
> > >  	if (last_page_offset > first_page_offset) {
> > >  		truncate_pagecache_range(inode, first_page_offset,
> > > @@ -4572,12 +4562,14 @@ int ext4_ext_punch_hole(struct file *file, loff_t offset, loff_t length)
> > >  
> > >  	credits = ext4_writepage_trans_blocks(inode);
> > >  	handle = ext4_journal_start(inode, credits);
> > > -	if (IS_ERR(handle))
> > > -		return PTR_ERR(handle);
> > > +	if (IS_ERR(handle)) {
> > > +		err = PTR_ERR(handle);
> > > +		goto error;
> > > +	}
> > >  
> > >  	err = ext4_orphan_add(handle, inode);
> > >  	if (err)
> > > -		goto out;
> > > +		goto error;
> > >  
> > >  	/*
> > >  	 * Now we need to zero out the non-page-aligned data in the
> > > @@ -4652,6 +4644,8 @@ int ext4_ext_punch_hole(struct file *file, loff_t offset, loff_t length)
> > >  	ext4_ext_invalidate_cache(inode);
> > >  	ext4_discard_preallocations(inode);
> > >  
> > > +	err = ext4_es_remove_space(inode, first_block,
> > > +				   stop_block - first_block);
> > >  	err = ext4_ext_remove_space(inode, first_block, stop_block - 1);
> > >  
> > >  	ext4_ext_invalidate_cache(inode);
> > > @@ -4667,6 +4661,8 @@ out:
> > >  	inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
> > >  	ext4_mark_inode_dirty(handle, inode);
> > >  	ext4_journal_stop(handle);
> > > +error:
> > > +	mutex_unlock(&inode->i_mutex);
> > >  	return err;
> > >  }
> > >  int ext4_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
> > > 
> --
> To unsubscribe from this list: send the line "unsubscribe linux-ext4" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html
>

On Mon, Jul 23, 2012 at 02:20:57PM +0200, Lukáš Czerner wrote:
[cut...]
> > > Moreover I think that we should avoid taking i_mutex if we can and I
> > > believe that we can in this case, because we only need to prevent
> > > allocation. So I just want to let you know that this part is
> > > probably going to change anyway.
> > 
> > It seems that we need to take i_mutex locking to prevent from buffered
> > writes after page cache has been truncated by truncate_pagecache_range.
> > If a buffered write without delalloc occurs after truncating page cache
> > and before taking i_data_sem, that means that the allocated block for
> > this buffered write will be removed in ext4_ext_remove_space when the
> > offset is within the range of the hole.  Am I missing something?
> 
> You're absolutely right, currently this is possible. But I think that we
> can take i_data_sem before truncating the pagecache hence preventing anyone
> from mapping new blocks. However this is not yet implemented in my
> patch set.
> 
> ...
> hmm, looking at the ext4_write_begin() it seems like it might not be
> such good idea after all. It seems to take page lock before
> i_data_sem so we might get deadlock, moreover if the punch hole
> happened in the middle of the ext4_write_begin() we might have only
> part of the data written, moreover this does not have to be hole
> aligned, which is bad. I need to revise that.

Yes, this is why I think that i_mutex locking should be taken.  At least
we are safty when we take the i_mutex. :-)

Regards,
Zheng
--
To unsubscribe from this list: send the line "unsubscribe linux-ext4" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html