Re: Support for large filesystems?
On Mon, Apr 30, 2001 at 04:56:19PM +0200, Oystein Viggen wrote:
> I just discovered the following from the gnumach and hurd changelogs:
> gnumach (1:1.2-9) unstable; urgency=low
>
> * Update to CVS 20010111, which contains a new interface for
> block devices with a large number of blocks.
>
> hurd (20010111) unstable; urgency=low
>
> * New snapshot from CVS, containing lots of small bug fixes, and:
> + Together with gnumach 1.2-9, you can access large stores with
> storeinfo, storeread (because libstore uses the new interface in
> gnumach, and off64_t internally).
>
> Does this mean support for filesystems larger than 1GB, or at least a
> beginning for such support, or is it something else entirely?
It is the beginning. The changes above are meant to implement
the support for large and very large stores (for example, large
partitions and hard drives). There is something wrong with it, though.
My 10GB disk is corrected allright, but the 40 GB disk gets wrong values.
Check your own drives with "storeinfo /dev/hd0" etc, and if you see a
discrepancy, try to debug it in GNU Mach. I suspect there is some
bit-lossing conversion from blocks to bytes and back to blocks or so.
The GNU Mach change above was needed because the old calls returned bytes,
but using blocks we can get up to a terabyte (or couple thereof). The other
changes are needed because off_t is 32 bit.
Now, support for large filesystems requires special care in the filesystem
servers (ext2fs, ...). But the above requirements are a precondition to
make that work.
Thanks,
Marcus
Reply to: