"Joseph I. Davida" wrote:
> If SGI does not push the Linux community
> into modifying the block IO layer to use
> 64 bit ints for representing the number of
> blocks in a device, we will be stuck with
> the 2 TB limit for a long time. One way to
> make this push is for SGI to go ahead and
> make the mods and let the rest catch up.
> The LARGEFILE support is already in place
> and it uses 64 bit ints for inodes, file
> sizes...etc. The only bit remaining is
> the LVM and block layer which still use
> 32 bit ints for device number of blocks.
That's not entirely true... inode number are still 32 bit.
the LARGEFILE support fixes the size fields for things
like stat and lseek but not inode numbers.
As far as the block layer stuff goes, the latest thinking from the 2.5
plans is to eliminate buffer_heads and implement a kiobuf based
io sub system. The 32 bit limitations should go away at that time.
I suspect that we will end up back porting much of this to 2.4.whatever
we will need large FS support long before 2.5 becomes 2.6.
For the moment we are a ways from having 2TB file systems available for
so we have waiting on rest of the linux developers to solve this
while we concentrate on stabilizing the file system itself.