Re: [ext2resize] Patches for ext2resize?

[Takashi S. <sh...@tn...>]

Hi,

I had gone out for three days so the reply to you was delayed.

> Can you explain why the rest of the patch is needed?  This is the
> patch updated to the current CVS version.

OK, I described comments below.
 
> Index: src/ext2.h
> ===================================================================
> RCS file: /cvsroot/ext2resize/ext2resize/src/ext2.h,v
> retrieving revision 1.28
> diff -u -3 -p -u -r1.28 ext2.h
> --- src/ext2.h 10 Jun 2006 07:45:55 -0000 1.28
> +++ src/ext2.h 10 Jun 2006 14:27:56 -0000
> @@ -264,7 +264,7 @@ static __inline__ int ext2_is_data_block
>  group = blk / fs->sb.s_blocks_per_group;
>  blk %= fs->sb.s_blocks_per_group;
> 
> - if (ext2_bg_has_super(fs, group) && blk <= fs->gdblocks)
> + if (ext2_bg_has_super(fs, group) && blk <= fs->gdblocks + fs->resgdblocks)
>  return 0;

If blk belongs to group descriptor blocks(GDB) or reserved GDB,
ext2_is_data_block should return 0.
Without this fix, it would return 1 though blk belongs to reserved GDB.

>  if (block == fs->gd[group].bg_block_bitmap ||
> Index: src/ext2_block_relocator.c
> ===================================================================
> RCS file: /cvsroot/ext2resize/ext2resize/src/ext2_block_relocator.c,v
> retrieving revision 1.22
> diff -u -3 -p -u -r1.22 ext2_block_relocator.c
> --- src/ext2_block_relocator.c 10 Jun 2006 07:45:55 -0000 1.22
> +++ src/ext2_block_relocator.c 10 Jun 2006 14:27:56 -0000
> @@ -497,6 +497,10 @@ static int ext2_block_relocator_grab_blo
>  int raid_bb, raid_ib;
>  int itend = state->new_alloc_offset;
> 
> + if (!ext2_bg_has_super(fs, group))
> + itend = fs->gd[group].bg_inode_table
> + +fs->inodeblocks - start;
> +

state->new_alloc_offset  is set to the offset of the first data block
in the block group which has a super block.  So, if the target group
doesn't have a super block, it should be re-calculated.

>  bpg = fs->sb.s_blocks_per_group;
>  if (start + bpg > fs->newblocks)
>  bpg = fs->newblocks - start;
> Index: src/ext2_meta.c
> ===================================================================
> RCS file: /cvsroot/ext2resize/ext2resize/src/ext2_meta.c,v
> retrieving revision 1.14
> diff -u -3 -p -u -r1.14 ext2_meta.c
> --- src/ext2_meta.c 17 Feb 2006 12:04:09 -0000 1.14
> +++ src/ext2_meta.c 10 Jun 2006 14:27:56 -0000
> @@ -192,7 +192,8 @@ int ext2_metadata_push(struct ext2_fs *f
>         group + 1, fs->numgroups);
>  }
> 
> - fs->itoffset = new_itoffset;
> + fs->itoffset = fs->newgdblocks + 3;
> +

fs->itoffset should be set to the offset of an inode table in a group
which has a super block.  Without this fix, fs->itoffset is possible to
be set to the offset in a group which doesn't have a super block and
following procedure runs incorrectly.

>  if (fs->flags & FL_VERBOSE)
>  printf("\n");
> Index: src/ext2_resize.c
> ===================================================================
> RCS file: /cvsroot/ext2resize/ext2resize/src/ext2_resize.c,v
> retrieving revision 1.19
> diff -u -3 -p -u -r1.19 ext2_resize.c
> --- src/ext2_resize.c 10 Jun 2006 06:45:22 -0000 1.19
> +++ src/ext2_resize.c 10 Jun 2006 14:27:56 -0000
> @@ -157,7 +157,7 @@ static int ext2_add_group(struct ext2_fs
>  return 1;
> }
> 
> -static int ext2_del_group(struct ext2_fs *fs)
> +static int ext2_del_group(struct ext2_fs *fs, int old_gdblocks)
> {
>  blk_t admin;
>  int   group = fs->numgroups - 1;
> @@ -170,7 +170,7 @@ static int ext2_del_group(struct ext2_fs
> 
>  has_sb = ext2_bg_has_super(fs, group);
> 
> - admin = fs->inodeblocks + (has_sb ? fs->gdblocks + 3 : 2);
> + admin = fs->inodeblocks + (has_sb ? old_gdblocks + 3 : 2);
> 
>  bpg = fs->sb.s_blocks_count - fs->sb.s_first_data_block -
>  group * fs->sb.s_blocks_per_group;
> @@ -407,6 +407,7 @@ static int ext2_grow_fs(struct ext2_fs *
> 
> static int ext2_shrink_fs(struct ext2_fs *fs)
> {
> + int old_gdblocks;
>  if (fs->flags & FL_DEBUG)
>  printf("%s\n", __FUNCTION__);
> 
> @@ -432,8 +433,7 @@ static int ext2_shrink_fs(struct ext2_fs
>  if (!ext2_inode_relocate(fs))
>  return 0;
> 
> - if (!ext2_block_relocate(fs))
> - return 0;
> + old_gdblocks = fs->gdblocks;

Until the size of a filesystem becomes the expected size,
groups will be removed at the following loop.

1. ext2_del_group() decreases the number of GDB(fs->gdblocks)
    after deleting all groups on one GBD.

2. When ext2_del_group() is called at the next iteration, it calculates
    the number of meta-data blocks in a group by using fs->gdblocks,
    so the result of the calculation is less than the actual meta-data
    blocks and following procedure runs incorrectly.

So we fixed this to save the original number of GDB(fs->gdblocks)
before deleting groups and passed the saved number to
ext2_del_group() as an argument.

>  while (fs->sb.s_blocks_count > fs->newblocks) {
>  blk_t sizelast = (fs->sb.s_blocks_count -
> @@ -445,7 +445,7 @@ static int ext2_shrink_fs(struct ext2_fs
>         fs->sb.s_blocks_count))
>  return 0;
>  } else {
> - if (!ext2_del_group(fs))
> + if (!ext2_del_group(fs, old_gdblocks))
>  return 0;
>  }
>  }

Cheers, sho