BUG with delayed allocation

BUG with delayed allocation

[Aneesh Kumar K.V]

Re: BUG with delayed allocation

[Eric Sandeen]

Aneesh Kumar K.V wrote:
> Hi,
> 
> Eric actually observed it yesterday. I am able to reproduce it locally.
> With delayed allocation we are observing wrong value of i_size.
> 
> kenrel: 2.6.25-rc5 +
> With all the patches in the patch queue

...

I also see this with only:

patches/delalloc-vfs.patch
patches/delalloc-ext4.patch

applied.

-Eric

Re: BUG with delayed allocation

[Aneesh Kumar K.V]

Re: BUG with delayed allocation

[Mingming Cao]

On Wed, 2008-03-19 at 14:22 +0530, Aneesh Kumar K.V wrote:
> Hi,
> 
> Eric actually observed it yesterday. I am able to reproduce it locally.
> With delayed allocation we are observing wrong value of i_size.
> 

The problem is current delalloc does not update on-disk i_size until
writeout time, the in-core i_size is updated though. 

Could you try the following patch? It updates the i_disksize at the
write_end time.


Signed-off-by: Mingming Cao <cmm@us.ibm.com>
---
 fs/ext4/inode.c |   39 ++++++++++++++++++++++++++++++++++++++-
 1 file changed, 38 insertions(+), 1 deletion(-)

Index: linux-2.6.25-rc5/fs/ext4/inode.c
===================================================================
--- linux-2.6.25-rc5.orig/fs/ext4/inode.c	2008-03-19 17:32:44.000000000 -0700
+++ linux-2.6.25-rc5/fs/ext4/inode.c	2008-03-19 17:43:19.000000000 -0700
@@ -1355,6 +1355,43 @@ static int ext4_writeback_write_end(stru
 	return ret ? ret : copied;
 }
 
+static int ext4_da_write_end(struct file *file,
+				struct address_space *mapping,
+				loff_t pos, unsigned len, unsigned copied,
+				struct page *page, void *fsdata)
+{
+	handle_t *handle;
+	struct inode *inode = file->f_mapping->host;
+	int needed_blocks = ext4_writepage_trans_blocks(inode);
+	int ret = 0, ret2;
+	loff_t new_i_size;
+
+	handle = ext4_journal_start(inode, needed_blocks);
+	if (IS_ERR(handle)) {
+		unlock_page(page);
+		page_cache_release(page);
+		ret = PTR_ERR(handle);
+		return ret;
+	}
+
+	new_i_size = pos + copied;
+	if (new_i_size > EXT4_I(inode)->i_disksize)
+		EXT4_I(inode)->i_disksize = new_i_size;
+
+	copied = ext4_generic_write_end(file, mapping, pos, len, copied,
+							page, fsdata);
+	if (copied < 0)
+		ret = copied;
+
+	ret2 = ext4_journal_stop(handle);
+	if (!ret)
+		ret = ret2;
+	unlock_page(page);
+	page_cache_release(page);
+
+	return ret ? ret : copied;
+}
+
 static int ext4_journalled_write_end(struct file *file,
 				struct address_space *mapping,
 				loff_t pos, unsigned len, unsigned copied,
@@ -2020,7 +2057,7 @@ static const struct address_space_operat
 	.writepages	= ext4_da_writepages,
 	.sync_page	= block_sync_page,
 	.write_begin	= ext4_da_write_begin,
-	.write_end	= generic_write_end,
+	.write_end	= ext4_da_write_end,
 	.bmap		= ext4_bmap,
 	.invalidatepage	= ext4_da_invalidatepage,
 	.releasepage	= ext4_releasepage,

Re: BUG with delayed allocation

[Aneesh Kumar K.V]

This mail didn't seem to reach the list.

-aneesh

On Thu, Mar 20, 2008 at 12:02:55AM +0530, Aneesh Kumar K.V wrote:
> On Wed, Mar 19, 2008 at 09:42:13AM -0500, Eric Sandeen wrote:
> > Aneesh Kumar K.V wrote:
> > > Hi,
> > > 
> > > Eric actually observed it yesterday. I am able to reproduce it locally.
> > > With delayed allocation we are observing wrong value of i_size.
> > > 
> > > kenrel: 2.6.25-rc5 +
> > > With all the patches in the patch queue
> > 
> > ...
> > 
> > I also see this with only:
> > 
> > patches/delalloc-vfs.patch
> > patches/delalloc-ext4.patch
> > 
> > applied.
> 
> This one seems to work for me. Need to think a bit more about parallel
> truncate.
> 
> -aneesh
> 
> diff --git a/fs/ext4/inode.c b/fs/ext4/inode.c
> index 79930df..b74426d 100644
> --- a/fs/ext4/inode.c
> +++ b/fs/ext4/inode.c
> @@ -1488,6 +1488,43 @@ out:
>  
>  	return ret;
>  }
> +static int ext4_da_writepage(struct page *page,
> +				struct writeback_control *wbc)
> +{
> +	struct inode *inode = page->mapping->host;
> +	handle_t *handle = NULL;
> +	int ret = 0;
> +	int err;
> +
> +	if (ext4_journal_current_handle())
> +		goto out_fail;
> +
> +	handle = ext4_journal_start(inode, ext4_writepage_trans_blocks(inode));
> +	if (IS_ERR(handle)) {
> +		ret = PTR_ERR(handle);
> +		goto out_fail;
> +	}
> +
> +	if (test_opt(inode->i_sb, NOBH) && ext4_should_writeback_data(inode))
> +		ret = nobh_writepage(page, ext4_get_block, wbc);
> +	else
> +		ret = block_write_full_page(page, ext4_get_block, wbc);
> +
> +	if (!ret && inode->i_size > EXT4_I(inode)->i_disksize) {
> +		EXT4_I(inode)->i_disksize = inode->i_size;
> +		ext4_mark_inode_dirty(handle, inode);
> +	}
> +
> +	err = ext4_journal_stop(handle);
> +	if (!ret)
> +		ret = err;
> +	return ret;
> +
> +out_fail:
> +	redirty_page_for_writepage(wbc, page);
> +	unlock_page(page);
> +	return ret;
> +}
>  
>  static int ext4_da_writepages(struct address_space *mapping,
>  				struct writeback_control *wbc)
> @@ -2015,7 +2052,7 @@ static const struct address_space_operations ext4_journalled_aops = {
>  static const struct address_space_operations ext4_da_aops = {
>  	.readpage	= ext4_readpage,
>  	.readpages	= ext4_readpages,
> -	.writepage	= ext4_writeback_writepage,
> +	.writepage	= ext4_da_writepage,
>  	.writepages	= ext4_da_writepages,
>  	.sync_page	= block_sync_page,
>  	.write_begin	= ext4_da_write_begin,

Re: BUG with delayed allocation

[Aneesh Kumar K.V]

On Wed, Mar 19, 2008 at 05:46:58PM -0700, Mingming Cao wrote:
> On Wed, 2008-03-19 at 14:22 +0530, Aneesh Kumar K.V wrote:
> > Hi,
> > 
> > Eric actually observed it yesterday. I am able to reproduce it locally.
> > With delayed allocation we are observing wrong value of i_size.
> > 
> 
> The problem is current delalloc does not update on-disk i_size until
> writeout time, the in-core i_size is updated though. 
> 

ext4_da_writepages actually update the i_disksize during writeout time.
The writepage callback for delalloc was using ext4_writeback_writepage,
which didn't update the i_disksize. That is why some of the files have
correct size while some doesn't. I tested the below change and sent this
to list. But i appears vger dropped the mail to the list.


diff --git a/fs/ext4/inode.c b/fs/ext4/inode.c
index 79930df..b74426d 100644
--- a/fs/ext4/inode.c
+++ b/fs/ext4/inode.c
@@ -1488,6 +1488,43 @@ out:
 
 	return ret;
 }
+static int ext4_da_writepage(struct page *page,
+				struct writeback_control *wbc)
+{
+	struct inode *inode = page->mapping->host;
+	handle_t *handle = NULL;
+	int ret = 0;
+	int err;
+
+	if (ext4_journal_current_handle())
+		goto out_fail;
+
+	handle = ext4_journal_start(inode, ext4_writepage_trans_blocks(inode));
+	if (IS_ERR(handle)) {
+		ret = PTR_ERR(handle);
+		goto out_fail;
+	}
+
+	if (test_opt(inode->i_sb, NOBH) && ext4_should_writeback_data(inode))
+		ret = nobh_writepage(page, ext4_get_block, wbc);
+	else
+		ret = block_write_full_page(page, ext4_get_block, wbc);
+
+	if (!ret && inode->i_size > EXT4_I(inode)->i_disksize) {
+		EXT4_I(inode)->i_disksize = inode->i_size;
+		ext4_mark_inode_dirty(handle, inode);
+	}
+
+	err = ext4_journal_stop(handle);
+	if (!ret)
+		ret = err;
+	return ret;
+
+out_fail:
+	redirty_page_for_writepage(wbc, page);
+	unlock_page(page);
+	return ret;
+}
 
 static int ext4_da_writepages(struct address_space *mapping,
 				struct writeback_control *wbc)
@@ -2015,7 +2052,7 @@ static const struct address_space_operations ext4_journalled_aops = {
 static const struct address_space_operations ext4_da_aops = {
 	.readpage	= ext4_readpage,
 	.readpages	= ext4_readpages,
-	.writepage	= ext4_writeback_writepage,
+	.writepage	= ext4_da_writepage,
 	.writepages	= ext4_da_writepages,
 	.sync_page	= block_sync_page,
 	.write_begin	= ext4_da_write_begin,

> Could you try the following patch? It updates the i_disksize at the
> write_end time.
> 

I will test the patch and update you. BTW shouldn't we update
i_disksize only after actual block got allocated ?


> 
> Signed-off-by: Mingming Cao <cmm@us.ibm.com>
> ---
>  fs/ext4/inode.c |   39 ++++++++++++++++++++++++++++++++++++++-
>  1 file changed, 38 insertions(+), 1 deletion(-)
> 
> Index: linux-2.6.25-rc5/fs/ext4/inode.c
> ===================================================================
> --- linux-2.6.25-rc5.orig/fs/ext4/inode.c	2008-03-19 17:32:44.000000000 -0700
> +++ linux-2.6.25-rc5/fs/ext4/inode.c	2008-03-19 17:43:19.000000000 -0700
> @@ -1355,6 +1355,43 @@ static int ext4_writeback_write_end(stru
>  	return ret ? ret : copied;
>  }
> 
> +static int ext4_da_write_end(struct file *file,
> +				struct address_space *mapping,
> +				loff_t pos, unsigned len, unsigned copied,
> +				struct page *page, void *fsdata)
> +{
> +	handle_t *handle;
> +	struct inode *inode = file->f_mapping->host;
> +	int needed_blocks = ext4_writepage_trans_blocks(inode);
> +	int ret = 0, ret2;
> +	loff_t new_i_size;
> +
> +	handle = ext4_journal_start(inode, needed_blocks);
> +	if (IS_ERR(handle)) {
> +		unlock_page(page);
> +		page_cache_release(page);
> +		ret = PTR_ERR(handle);
> +		return ret;
> +	}
> +
> +	new_i_size = pos + copied;
> +	if (new_i_size > EXT4_I(inode)->i_disksize)
> +		EXT4_I(inode)->i_disksize = new_i_size;
> +
> +	copied = ext4_generic_write_end(file, mapping, pos, len, copied,
> +							page, fsdata);
> +	if (copied < 0)
> +		ret = copied;
> +
> +	ret2 = ext4_journal_stop(handle);
> +	if (!ret)
> +		ret = ret2;
> +	unlock_page(page);
> +	page_cache_release(page);
> +
> +	return ret ? ret : copied;
> +}
> +
>  static int ext4_journalled_write_end(struct file *file,
>  				struct address_space *mapping,
>  				loff_t pos, unsigned len, unsigned copied,
> @@ -2020,7 +2057,7 @@ static const struct address_space_operat
>  	.writepages	= ext4_da_writepages,
>  	.sync_page	= block_sync_page,
>  	.write_begin	= ext4_da_write_begin,
> -	.write_end	= generic_write_end,
> +	.write_end	= ext4_da_write_end,
>  	.bmap		= ext4_bmap,
>  	.invalidatepage	= ext4_da_invalidatepage,
>  	.releasepage	= ext4_releasepage,
> 
> 

-aneesh

Re: BUG with delayed allocation

[Mingming Cao]

On Thu, 2008-03-20 at 11:09 +0530, Aneesh Kumar K.V wrote:

> > Could you try the following patch? It updates the i_disksize at the
> > write_end time.
> > 
> 
> I will test the patch and update you. BTW shouldn't we update
> i_disksize only after actual block got allocated ?
> 
> 
Hmm...I am not  100% sure but I think we should not to change the
behavior that the on-disk inode size should be updated when write()
returns to user. Right now the in-memory inode size is updated, user
would expecting the same when they run e2fsck, but e2fsck reads inode
size from disk. Pushing the inode i_disksize update at the writeout
(allocation) time will cause the window that i_size is different than
the i_disksize being enlarged quite big.

Mingming
> > 
> > Signed-off-by: Mingming Cao <cmm@us.ibm.com>
> > ---
> >  fs/ext4/inode.c |   39 ++++++++++++++++++++++++++++++++++++++-
> >  1 file changed, 38 insertions(+), 1 deletion(-)
> > 
> > Index: linux-2.6.25-rc5/fs/ext4/inode.c
> > ===================================================================
> > --- linux-2.6.25-rc5.orig/fs/ext4/inode.c	2008-03-19 17:32:44.000000000 -0700
> > +++ linux-2.6.25-rc5/fs/ext4/inode.c	2008-03-19 17:43:19.000000000 -0700
> > @@ -1355,6 +1355,43 @@ static int ext4_writeback_write_end(stru
> >  	return ret ? ret : copied;
> >  }
> > 
> > +static int ext4_da_write_end(struct file *file,
> > +				struct address_space *mapping,
> > +				loff_t pos, unsigned len, unsigned copied,
> > +				struct page *page, void *fsdata)
> > +{
> > +	handle_t *handle;
> > +	struct inode *inode = file->f_mapping->host;
> > +	int needed_blocks = ext4_writepage_trans_blocks(inode);
> > +	int ret = 0, ret2;
> > +	loff_t new_i_size;
> > +
> > +	handle = ext4_journal_start(inode, needed_blocks);
> > +	if (IS_ERR(handle)) {
> > +		unlock_page(page);
> > +		page_cache_release(page);
> > +		ret = PTR_ERR(handle);
> > +		return ret;
> > +	}
> > +
> > +	new_i_size = pos + copied;
> > +	if (new_i_size > EXT4_I(inode)->i_disksize)
> > +		EXT4_I(inode)->i_disksize = new_i_size;
> > +
> > +	copied = ext4_generic_write_end(file, mapping, pos, len, copied,
> > +							page, fsdata);
> > +	if (copied < 0)
> > +		ret = copied;
> > +
> > +	ret2 = ext4_journal_stop(handle);
> > +	if (!ret)
> > +		ret = ret2;
> > +	unlock_page(page);
> > +	page_cache_release(page);
> > +
> > +	return ret ? ret : copied;
> > +}
> > +
> >  static int ext4_journalled_write_end(struct file *file,
> >  				struct address_space *mapping,
> >  				loff_t pos, unsigned len, unsigned copied,
> > @@ -2020,7 +2057,7 @@ static const struct address_space_operat
> >  	.writepages	= ext4_da_writepages,
> >  	.sync_page	= block_sync_page,
> >  	.write_begin	= ext4_da_write_begin,
> > -	.write_end	= generic_write_end,
> > +	.write_end	= ext4_da_write_end,
> >  	.bmap		= ext4_bmap,
> >  	.invalidatepage	= ext4_da_invalidatepage,
> >  	.releasepage	= ext4_releasepage,
> > 
> > 
> 
> -aneesh
> --
> 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

Re: BUG with delayed allocation

[Aneesh Kumar K.V]

On Thu, Mar 20, 2008 at 10:29:50AM -0700, Mingming Cao wrote:
> On Thu, 2008-03-20 at 11:09 +0530, Aneesh Kumar K.V wrote:
> 
> > > Could you try the following patch? It updates the i_disksize at the
> > > write_end time.
> > > 
> > 
> > I will test the patch and update you. BTW shouldn't we update
> > i_disksize only after actual block got allocated ?
> > 
> > 
> Hmm...I am not  100% sure but I think we should not to change the
> behavior that the on-disk inode size should be updated when write()
> returns to user. Right now the in-memory inode size is updated, user
> would expecting the same when they run e2fsck, but e2fsck reads inode
> size from disk. Pushing the inode i_disksize update at the writeout
> (allocation) time will cause the window that i_size is different than
> the i_disksize being enlarged quite big.
> 

If we are updating i_disksize during write_end and if we crash before actually
allocating the blocks e2fsck will find errors because the inode doesn't
really have that many blocks right ?


-aneesh

Re: BUG with delayed allocation

[Andreas Dilger]

On Mar 20, 2008  23:26 +0530, Aneesh Kumar K.V wrote:
> On Thu, Mar 20, 2008 at 10:29:50AM -0700, Mingming Cao wrote:
> > On Thu, 2008-03-20 at 11:09 +0530, Aneesh Kumar K.V wrote:
> > > > Could you try the following patch? It updates the i_disksize at the
> > > > write_end time.
> > > 
> > > I will test the patch and update you. BTW shouldn't we update
> > > i_disksize only after actual block got allocated ?
> >
> > Hmm...I am not  100% sure but I think we should not to change the
> > behavior that the on-disk inode size should be updated when write()
> > returns to user. Right now the in-memory inode size is updated, user
> > would expecting the same when they run e2fsck, but e2fsck reads inode
> > size from disk. Pushing the inode i_disksize update at the writeout
> > (allocation) time will cause the window that i_size is different than
> > the i_disksize being enlarged quite big.
> 
> If we are updating i_disksize during write_end and if we crash before actually
> allocating the blocks e2fsck will find errors because the inode doesn't
> really have that many blocks right ?

No, it would just think the file is sparse and return \0 for the reads.

That said, I don't agree with Mingming - the i_disksize should only be
increased at the time the blocks are allocated on disk and not when the
file is extended in memory.  Even if the window where i_size is different
than i_disksize is large, this is only important after a crash, and at
that time ordered mode users want the file to have a shorter i_disksize
and the file contains only valid data, instead of the extended i_size
but the file contains \0 bytes at the end.

Cheers, Andreas
--
Andreas Dilger
Sr. Staff Engineer, Lustre Group
Sun Microsystems of Canada, Inc.