Re: [PATCH] Fix regression in O_DIRECT|O_SYNC writes to block devices

[Jan Kara <jack@suse.cz>]

On Thu 15-04-10 12:42:24, Christoph Hellwig wrote:
> On Thu, Apr 15, 2010 at 02:40:39PM +1000, Anton Blanchard wrote:
> >  int blkdev_fsync(struct file *filp, struct dentry *dentry, int datasync)
> >  {
> > -	struct block_device *bdev = I_BDEV(filp->f_mapping->host);
> > +	struct inode *bd_inode = filp->f_mapping->host;
> > +	struct block_device *bdev = I_BDEV(bd_inode);
> >  	int error;
> >  
> > +	mutex_unlock(&bd_inode->i_mutex);
> > +
> >  	error = sync_blockdev(bdev);
> 
> Actually you can just drop this call entirely.  sync_blockdev is an
> overcomplicated alias for filemap_write_and_wait on the  block device
> inode, which is exactl what we did just before calling into ->fsync
> 
> It might be worth to still drop i_mutex for the cache flush, though.
  Yeah, that's a good point... Anton, just remove sync_blockdev() from
blkdev_fsync completely please.

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