[PATCH] fs: don't flush pagecache when expanding block device

[PATCH] fs: don't flush pagecache when expanding block device

[shunki-fujita]

When changing the size of a block device, its all caches are freed.
It's necessary on shrinking to prevent spurious I/Os to the disappeared region.
However, on expanding, such kind of I/Os doesn't happen.

Similar things can be considered for btrfs filesystem resize and resize2fs,
but they are designed not to cache drops when expanding.
Therefore this patch removes unnecessary cache drop.

Signed-off-by: Shunki Fujita <shunki-fujita@cybozu.co.jp>
---
 fs/block_dev.c | 9 ++++++++-
 1 file changed, 8 insertions(+), 1 deletion(-)

diff --git a/fs/block_dev.c b/fs/block_dev.c
index fe09ef9..243f4002 100644
--- a/fs/block_dev.c
+++ b/fs/block_dev.c
@@ -1337,7 +1337,14 @@ void check_disk_size_change(struct gendisk *disk, struct block_device *bdev)
 		       "%s: detected capacity change from %lld to %lld\n",
 		       disk->disk_name, bdev_size, disk_size);
 		i_size_write(bdev->bd_inode, disk_size);
-		flush_disk(bdev, false);
+		if (bdev_size > disk_size) {
+			flush_disk(bdev, false);
+		} else {
+			if (!bdev->bd_disk)
+				return;
+			if (disk_part_scan_enabled(bdev->bd_disk))
+				bdev->bd_invalidated = 1;
+		}
 	}
 }
 EXPORT_SYMBOL(check_disk_size_change);
-- 
2.7.4

Re: [PATCH] fs: don't flush pagecache when expanding block device

[Andrew Morton]

On Fri, 16 Mar 2018 15:55:53 +0900 shunki-fujita <shunki-fujita@cybozu.co.jp> wrote:

> When changing the size of a block device, its all caches are freed.
> It's necessary on shrinking to prevent spurious I/Os to the disappeared region.
> However, on expanding, such kind of I/Os doesn't happen.
> 
> Similar things can be considered for btrfs filesystem resize and resize2fs,
> but they are designed not to cache drops when expanding.
> Therefore this patch removes unnecessary cache drop.

Yes, the patch removes the flush_disk() call when shrinking.  But it
adds additional code which is unchangelogged and uncommented.  What's
happening here?

> --- a/fs/block_dev.c
> +++ b/fs/block_dev.c
> @@ -1337,7 +1337,14 @@ void check_disk_size_change(struct gendisk *disk, struct block_device *bdev)
>  		       "%s: detected capacity change from %lld to %lld\n",
>  		       disk->disk_name, bdev_size, disk_size);
>  		i_size_write(bdev->bd_inode, disk_size);
> -		flush_disk(bdev, false);
> +		if (bdev_size > disk_size) {
> +			flush_disk(bdev, false);
> +		} else {
> +			if (!bdev->bd_disk)
> +				return;
> +			if (disk_part_scan_enabled(bdev->bd_disk))
> +				bdev->bd_invalidated = 1;
> +		}
>  	}
>  }

Oh, I see.  It's a copy-n-paste of some mystery uncommented,
lost-in-the-mists-of-time code from flush_disk().  Argh.

The world would be a better place if we could move that bit of code
into a separate function, figure out what it does, document it and call
it from both sites.

Re: [PATCH] fs: don't flush pagecache when expanding block device

[Shunki Fujita]

Hi Andrew,

> On Fri, 16 Mar 2018 15:55:53 +0900 shunki-fujita <shunki-fujita@cybozu.co.jp> wrote:
> 
> > When changing the size of a block device, its all caches are freed.
> > It's necessary on shrinking to prevent spurious I/Os to the disappeared region.
> > However, on expanding, such kind of I/Os doesn't happen.
> > 
> > Similar things can be considered for btrfs filesystem resize and resize2fs,
> > but they are designed not to cache drops when expanding.
> > Therefore this patch removes unnecessary cache drop.
> 
> Yes, the patch removes the flush_disk() call when shrinking.  But it
> adds additional code which is unchangelogged and uncommented.  What's
> happening here?
> 
> > --- a/fs/block_dev.c
> > +++ b/fs/block_dev.c
> > @@ -1337,7 +1337,14 @@ void check_disk_size_change(struct gendisk *disk, struct block_device *bdev)
> >  		       "%s: detected capacity change from %lld to %lld\n",
> >  		       disk->disk_name, bdev_size, disk_size);
> >  		i_size_write(bdev->bd_inode, disk_size);
> > -		flush_disk(bdev, false);
> > +		if (bdev_size > disk_size) {
> > +			flush_disk(bdev, false);
> > +		} else {
> > +			if (!bdev->bd_disk)
> > +				return;
> > +			if (disk_part_scan_enabled(bdev->bd_disk))
> > +				bdev->bd_invalidated = 1;
> > +		}
> >  	}
> >  }
> 
> Oh, I see.  It's a copy-n-paste of some mystery uncommented,
> lost-in-the-mists-of-time code from flush_disk().  Argh.
> 
> The world would be a better place if we could move that bit of code
> into a separate function, figure out what it does, document it and call
> it from both sites.
> 

This was an unnecessary additional code.
Since bdev-> bd_invalidated always 0 after check_disk_size_change (), no additional code was required.
https://github.com/torvalds/linux/blob/master/fs/block_dev.c#L1365-L1366

I will repost the fix patch.

Thanks,
Shunki

[PATCH v2] fs: don't flush pagecache when expanding block device

[shunki-fujita]

When changing the size of a block device, its all caches are freed.
It's necessary on shrinking to prevent spurious I/Os to the disappeared region.
However, on expanding, such kind of I/Os doesn't happen.

Similar things can be considered for btrfs filesystem resize and resize2fs,
but they are designed not to cache drops when expanding.
Therefore this patch removes unnecessary cache drop.

Signed-off-by: Shunki Fujita <shunki-fujita@cybozu.co.jp>
---
Changes in v2:
   - Remove unnecessary changes (Since bdev-> bd_invalidated always 0 after check_disk_size_change ()[1])
   - Add document of check_disk_size_change ()

   [1] https://github.com/torvalds/linux/blob/master/fs/block_dev.c#L1365-L1366

 fs/block_dev.c | 6 ++++--
 1 file changed, 4 insertions(+), 2 deletions(-)

diff --git a/fs/block_dev.c b/fs/block_dev.c
index fe09ef9..63f1269 100644
--- a/fs/block_dev.c
+++ b/fs/block_dev.c
@@ -1324,7 +1324,8 @@ static void flush_disk(struct block_device *bdev, bool kill_dirty)
  * @bdev: struct bdev to adjust.
  *
  * This routine checks to see if the bdev size does not match the disk size
- * and adjusts it if it differs.
+ * and adjusts it if it differs. When shrinking the bdev size, its all caches 
+ * are freed.
  */
 void check_disk_size_change(struct gendisk *disk, struct block_device *bdev)
 {
@@ -1337,7 +1338,8 @@ void check_disk_size_change(struct gendisk *disk, struct block_device *bdev)
 		       "%s: detected capacity change from %lld to %lld\n",
 		       disk->disk_name, bdev_size, disk_size);
 		i_size_write(bdev->bd_inode, disk_size);
-		flush_disk(bdev, false);
+		if (bdev_size > disk_size)
+			flush_disk(bdev, false);
 	}
 }
 EXPORT_SYMBOL(check_disk_size_change);
-- 
2.7.4