Re: [PATCH v1 15/22] tune2fs: add inline_data feature in tune2fs

["Darrick J. Wong" <darrick.wong@oracle.com>]

On Sat, Oct 12, 2013 at 04:16:38PM +0800, Zheng Liu wrote:
> On Fri, Oct 11, 2013 at 05:39:45PM -0700, Darrick J. Wong wrote:
> > On Fri, Aug 02, 2013 at 05:49:42PM +0800, Zheng Liu wrote:
> > > From: Zheng Liu <wenqing.lz@taobao.com>
> > > 
> > > Inline_data feature can be set when ext_attr feature is enabled.  Now this
> > > feature only can be enabled because we may be out of space when disabling
> > > it.  If this feature is disabled, we need to allocate a block for every
> > > file and directory, and it might exhaust all space.
> > > 
> > > Signed-off-by: Theodore Ts'o <tytso@mit.edu>
> > > Signed-off-by: Zheng Liu <wenqing.lz@taobao.com>
> > > ---
> > >  misc/tune2fs.8.in |    5 +++++
> > >  misc/tune2fs.c    |   17 ++++++++++++++++-
> > >  2 files changed, 21 insertions(+), 1 deletion(-)
> > > 
> > > diff --git a/misc/tune2fs.8.in b/misc/tune2fs.8.in
> > > index 55c6dd9..b768fff 100644
> > > --- a/misc/tune2fs.8.in
> > > +++ b/misc/tune2fs.8.in
> > > @@ -531,6 +531,11 @@ Setting the filesystem feature is equivalent to using the
> > >  .B \-j
> > >  option.
> > >  .TP
> > > +.B inline_data
> > > +Allow data to be stored in inode.
> > > +.B Tune2fs
> > > +only supports setting this filesystem feature.
> > > +.TP
> > >  .B large_file
> > >  Filesystem can contain files that are greater than 2GB.  (Modern kernels
> > >  set this feature automatically when a file > 2GB is created.)
> > > diff --git a/misc/tune2fs.c b/misc/tune2fs.c
> > > index 7d6520e..f22a736 100644
> > > --- a/misc/tune2fs.c
> > > +++ b/misc/tune2fs.c
> > > @@ -137,7 +137,8 @@ static __u32 ok_features[3] = {
> > >  	EXT2_FEATURE_INCOMPAT_FILETYPE |
> > >  		EXT3_FEATURE_INCOMPAT_EXTENTS |
> > >  		EXT4_FEATURE_INCOMPAT_FLEX_BG |
> > > -		EXT4_FEATURE_INCOMPAT_MMP,
> > > +		EXT4_FEATURE_INCOMPAT_MMP |
> > > +		EXT4_FEATURE_INCOMPAT_INLINE_DATA,
> > >  	/* R/O compat */
> > >  	EXT2_FEATURE_RO_COMPAT_LARGE_FILE |
> > >  		EXT4_FEATURE_RO_COMPAT_HUGE_FILE|
> > > @@ -995,6 +996,20 @@ mmp_error:
> > >  		disable_uninit_bg(fs,
> > >  				EXT4_FEATURE_RO_COMPAT_GDT_CSUM);
> > >  
> > > +	if (FEATURE_ON(E2P_FEATURE_INCOMPAT,
> > > +		       EXT4_FEATURE_INCOMPAT_INLINE_DATA)) {
> > > +		/* Inline_data feature cannot be enabled if ext_attr is
> > > +		 * disabled.
> > > +		 */
> > 
> > Multiline comment issue.
> > 
> > > +		if (!(fs->super->s_feature_compat &
> > > +		      EXT2_FEATURE_COMPAT_EXT_ATTR)) {
> > 
> > EXT2_HAS_COMPAT_FEATURE()?
> > 
> > > +			fputs(_("The inline_data feature cannot "
> > > +				"be set if ext_attr feature is disabled.\n"),
> > > +				stderr);
> > > +			return 1;
> > > +		}
> > > +	}
> > > +
> > 
> > Maybe I'm missing something, but I don't see anything prohibiting the user from
> > disabling inline_data?  Or from doing the conversion?
> 
> Yes, now in tune2fs we can only enable inline_data feature.  If we want
> to disable inline_data, that means that we need to allocate a block for
> every inode with inline data, and we could exhaust all disk space.
> Another implementation might be just clear inline_data flag from
> super_block and doesn't allocate a block for these inodes.  The former
> obviously is not a good choice.  But the latter might be better.  I am
> really not sure for this.  What do you think?

I don't think it's a good idea to turn off inline_data once it's been enabled,
just like we don't let people turn off extents...

*Oh*, duh, I didn't notice that INLINE_DATA is *NOT* in the clear_ok_features
list.  Never mind this comment, I'm just making noise.

(The complaints about the comment issue and not using EXT2_HAS_COMPAT_FEATURE
remain.)

--D
> 
> Thanks,
>                                                 - Zheng