tune2fs: Update overhead when toggling journal feature

When adding or removing journal from a filesystem, we also need to add /
remove journal blocks from overhead stored in the superblock.  Otherwise
total number of blocks in the filesystem as reported by statfs(2) need
not match reality and could lead to odd results like negative number of
used blocks reported by df(1).

Signed-off-by: Jan Kara <jack@suse.cz>
---
 misc/tune2fs.c | 10 ++++++++--
 1 file changed, 8 insertions(+), 2 deletions(-)

> On Jun 14, 2021, at 3:28 PM, Jan Kara <jack@suse.cz> wrote:
> 
> When adding or removing journal from a filesystem, we also need to add /
> remove journal blocks from overhead stored in the superblock.  Otherwise
> total number of blocks in the filesystem as reported by statfs(2) need
> not match reality and could lead to odd results like negative number of
> used blocks reported by df(1).
> 
> Signed-off-by: Jan Kara <jack@suse.cz>

You could add:

Fixes: 9046b4dfd0ce ("mke2fs: set overhead in super block")

and

Reviewed-by: Andreas Dilger <adilger@dilger.ca>

That also makes me wonder if resize2fs also needs to recalculate or
invalidate the s_overhead_clusters field when adding new block groups.
It *looks* like that is done correctly in adjust_fs_info() already?

Cheers, Andreas

> ---
> misc/tune2fs.c | 10 ++++++++--
> 1 file changed, 8 insertions(+), 2 deletions(-)
> 
> diff --git a/misc/tune2fs.c b/misc/tune2fs.c
> index 4d4cf5a13384..2f6858abda32 100644
> --- a/misc/tune2fs.c
> +++ b/misc/tune2fs.c
> @@ -396,6 +396,8 @@ static errcode_t remove_journal_inode(ext2_filsys fs)
> 				_("while clearing journal inode"));
> 			return retval;
> 		}
> +		fs->super->s_overhead_clusters -=
> +			EXT2FS_NUM_B2C(fs, EXT2_I_SIZE(&inode) / fs->blocksize);
> 		memset(&inode, 0, sizeof(inode));
> 		ext2fs_mark_bb_dirty(fs);
> 		fs->flags &= ~EXT2_FLAG_SUPER_ONLY;
> @@ -1663,8 +1665,12 @@ static int add_journal(ext2_filsys fs)
> 			com_err(program_name, retval, "%s",
> 				_("\n\twhile trying to create journal file"));
> 			return retval;
> -		} else
> -			fputs(_("done\n"), stdout);
> +		}
> +		fs->super->s_overhead_clusters += EXT2FS_NUM_B2C(fs,
> +			jparams.num_journal_blocks + jparams.num_fc_blocks);
> +		ext2fs_mark_super_dirty(fs);
> +		fputs(_("done\n"), stdout);
> +
> 		/*
> 		 * If the filesystem wasn't mounted, we need to force
> 		 * the block group descriptors out.
> --
> 2.26.2
> 


Cheers, Andreas

On Mon 14-06-21 15:38:57, Andreas Dilger wrote:
> 
> > On Jun 14, 2021, at 3:28 PM, Jan Kara <jack@suse.cz> wrote:
> > 
> > When adding or removing journal from a filesystem, we also need to add /
> > remove journal blocks from overhead stored in the superblock.  Otherwise
> > total number of blocks in the filesystem as reported by statfs(2) need
> > not match reality and could lead to odd results like negative number of
> > used blocks reported by df(1).
> > 
> > Signed-off-by: Jan Kara <jack@suse.cz>
> 
> You could add:
> 
> Fixes: 9046b4dfd0ce ("mke2fs: set overhead in super block")
> 
> and
> 
> Reviewed-by: Andreas Dilger <adilger@dilger.ca>

Thanks!

> That also makes me wonder if resize2fs also needs to recalculate or
> invalidate the s_overhead_clusters field when adding new block groups.
> It *looks* like that is done correctly in adjust_fs_info() already?

Yes. From a quick look I had when doing this tune2fs patch I've noticed
that adjust_fs_info() just zeros s_overhead_clusters which makes the kernel
compute the overhead instead...

								Honza

On Mon, Jun 14, 2021 at 11:28:30PM +0200, Jan Kara wrote:
> When adding or removing journal from a filesystem, we also need to add /
> remove journal blocks from overhead stored in the superblock.  Otherwise
> total number of blocks in the filesystem as reported by statfs(2) need
> not match reality and could lead to odd results like negative number of
> used blocks reported by df(1).
> 
> Signed-off-by: Jan Kara <jack@suse.cz>

Applied thanks.

					- Ted