Re: Ext2fs 2GB limit status?

[Ognyan Kulev]

Marco Gerards wrote:
> Ogi (I hope he is subscribed to this list) wrote some patches for the Hurd so
> ext2fs can support large partitions.  It seems to me that not many
> people have looked at these patches :(.  Having a >2GB capable ext2fs
> is very important to me.

The people that commented on this patch are less that the fingers of one 
of my hands ;-)

> The thing that bothered me when reading these patches was that byte
> offsets were used quite often.  This will make it not possible to use
> partitions bigger than 4GB in some circumstances.

All used byte offsets should be 64-bit (off_t).  My main target was to 
_quickly_ release patch that works.  That's the reason that I made as 
few changes to the original design as possible.  Applying ext2fs patch 
that works fine is higher priority than improving the code.

> Most likely I forgot some.  Especially the pokel stuff might be bad.
> Another use for these functions was to read/write the superblock;
> although it isn't that bad, it would be better if you didn't use byte
> offsets for this.
>
> Perhaps this problem can be fixed by removing all these functions so
> they can't be used accidently.  I see no reason to have these
> functions anyway.

I will need some days to go deeper about these.

> Another thing I've noticed in the patch were changes in the amount of
> whitespaces.  This isn't a big problem (for me), it is just something
> to lookout for when sending in the definite patch, here is an example:
>
> -     * We have succeeded in finding a free byte in the block
> -     * bitmap.  Now search backwards up to 7 bits to find the
> -     * start of this group of free blocks.
> +   * We have succeeded in finding a free byte in the block
> +   * bitmap.  Now search backwards up to 7 bits to find the
> +   * start of this group of free blocks.

Yes, there is changed whitespace in some places.  In all of them, it's 
done the way Emacs do it.  In this particular example, I think that * 
are aligned to the leading /* or something like that.  AFAIR I've made 
such whitespace changes in the legacy ext2 code from Linux.

> The last thing I mentioned was this:
>
> +#if 1
>        err = find_block (node, offset, &block, &lock);
>        if (err)
>         break;
>        assert (block);
> +#else
> +      /* XXX: I still haven't investigated why this is needed...  */
> +      err = ext2_getblk (node, offset >> log2_block_size, 1, &block);
> +      if (err)
> +       break;
> +#endif
>
> I'm not really an ext2fs expert, but I think this is used for sparse
> files.  Ext2fs doesn't allocate blocks when seeking, it just makes the
> file sparse.  So the blocks get allocated just before they are written
> to disk.

As you see, the original code is compiled by default.  The #else is just 
garbage, it _was_ required because of some bugs that are gone now.  I 
believe it can be removed safely, as other #if 0 in the code.

> I also have a question about the code.  The assembly code "int $3" is
> used.  AFAIK this is used to tell the debugger a breakpoint is
> reached, right?  When is this function (abort) called?

Well, this is why it is called alpha ;-)  There was a time when 
assertions failed very often.  When they fail, they call abort(), which 
I replaced with the function you saw and which contains this "int $3". 
This is solely for debugging purposes.

Now, about the future of ext2fs.  I won't be able to work on it much 
till the end of November.  So if you, or anyone else, want to work on 
it, it will be best if we put it on hurd-extras or some other public 
place where changes can be made by many people.  (We can forget about 
changelogs till release candidate status.)

Regards,
ogi