Re: [RFC 1/2] ext4 resize: Mark the added group with EXT4_BG_INODE_ZEROED flag

[Andreas Dilger <adilger@sun.com>]

On Nov 21, 2008  11:23 +0100, Solofo.Ramangalahy@bull.net wrote:
> The inode table has been zeroed in setup_new_group_blocks().
> Mark it as such in ext4_group_add().
> 
> As a side note, online resize and inode zeroing are "dual".
> 
> In order to obtain a filesystem with faster formating times one can
> do:
> . either format a smaller fs and then resize it,
> . or format the fs with lazy_itable_init

As discussed on the concall, it probably makes more sense to have the
resize code work by marking the inode tables UNINIT (if GDT_CSUM feature
is enabled) and then start the "itable zeroing" thread, if it isn't
already running, to do the zeroing of the itable.

If GDT_CSUM isn't set then the below fix is the right solution.

> Index: linux-2.6.28-rc4-itable_init/fs/ext4/resize.c
> ===================================================================
> --- linux-2.6.28-rc4-itable_init.orig/fs/ext4/resize.c
> +++ linux-2.6.28-rc4-itable_init/fs/ext4/resize.c
> @@ -865,7 +865,7 @@ int ext4_group_add(struct super_block *s
>  	gdp->bg_free_blocks_count = cpu_to_le16(input->free_blocks_count);
>  	gdp->bg_free_inodes_count = cpu_to_le16(EXT4_INODES_PER_GROUP(sb));
>  	gdp->bg_checksum = ext4_group_desc_csum(sbi, input->group, gdp);
> -
> +	gdp->bg_flags |= cpu_to_le16(EXT4_BG_INODE_ZEROED);
>  	/*
>  	 * We can allocate memory for mb_alloc based on the new group
>  	 * descriptor

Cheers, Andreas
--
Andreas Dilger
Sr. Staff Engineer, Lustre Group
Sun Microsystems of Canada, Inc.