Re: Question on huge_file

[Andreas Dilger <adilger@sun.com>]

On Jul 10, 2009  08:32 -0700, Curt Wohlgemuth wrote:
> I apologize if this is a dumb question, but I'm having trouble
> understanding the huge_file superblock flag.
> 
> I see how, if this flag is set, that the inode can have a size > 2**32
> bytes, using the i_size_lo/i_size_high fields.

Actually, it is RO_COMPAT_LARGE_FILE that indicates support for size larger
than 2^32 _bytes_.  The RO_COMPAT_HUGE_FILE indicates support for blocks
more than 2^32 512-byte _sectors_ (2TB).

> But since an ext4_extent only uses 32-bits for for its ee_block field
> to represent the logical block, how can an extent describe any block
> range of a file past the 4GiB boundary?

There are two different mechanisms used with HUGE_FILE.  It allows
storing a high word of data (2^48 sectors) and it ALSO changes the
units to be in terms of filesystem blocksize instead of 512-byte sectors.

While both of these mechanisms are not strictly necessary with the
current extent format, which only handles 2^32 filesystem blocks,
there were some good reasons to make both changes:
- having the inode i_blocks field be in 512-byte sectors was confusing
  to many coders and wasted 3 bits (for 4kB blocks) of dynamic range.
- if we ever implement a new extent format that handles more than 2^32
  filesystem blocks, or use larger filesystem blocks we don't need to
  rework this code again.

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