Re: [RFC][PATCH 0/2]Extend ext3 filesystem limit from 8TB to 16TB

[Andreas Dilger <adilger@clusterfs.com>]

On Mar 29, 2006  17:38 -0800, Mingming Cao wrote:
> There are places in ext3 code to use "int" to represent block numbers in
> kernel(not on-disk). This seems the "only" reason that why we can only
> have 8TB ext3 rather than 16TB.  Most times it just a bug with no
> particular reason why not use unsigned 32 bit value, so the fix is easy.
> 
> However, it is not so straightforward fix for the ext3 block allocation
> code, as ext3_new_block() returns a block number, and "-1" to indicating
> block allocation failure. Ext3 block reservation code, called by
> ext3_new_block(), thus also use "int" for block numbers in some places.

What might make the code a lot clearer, easier to audit, and easier to
fix in the future is to declare new types for fs block offsets and group
block offsets.  Something like "ext3_fsblk" and "ext3_grblk".  That way,
we can declare ext3_fsblk as "unsigned long" and "ext3_grblk" as "unsigned
int", and we could optionally change ext3_fsblk to be "unsigned long long"
later to support 64-bit filesystems without having to re-patch all of the
code.

It would be more clear what type of block offset a function is handling
(fs-wide or group-relative).  If we wanted to be able to overload the
block number with an error code we could use ERR_PTR and PTR_ERR like
macros, and just restrict the filesystem to 2^32 - 1024 blocks until we
extend it to 64 bits.

Cheers, Andreas
--
Andreas Dilger
Principal Software Engineer
Cluster File Systems, Inc.