Re: [ext2resize] ext2resize on big endian, corrupt file systems and crashes?

[Andreas D. <ad...@cl...>]

On Mar 17, 2006  08:24 +0100, Petter Reinholdtsen wrote:
> [Andreas Dilger]
> > As good a place as any.  I'm actually somewhat interested in getting
> > rid of ext2resize itself and just merging the functionality into
> > e2fsprogs as RH has done, just to avoid duplication of maintenance
> > effort.
> 
> As far as I could see, the e2fsprogs souce RPM from redhat include the
> source for both e2fsprogs and ext2resize, and isn't a merge of
> functionallity.  So they still include ext2resize, but it does not
> seem that way from the list of RPMs.
> 
> But I would love to only have one package to maintain, so if e2fsprogs
> is up for the job, I am more than willing to drop ext2resize. :)

Ted Ts'o just announced he had added ext2online functionality to
resize2fs.  The only remaining holdout is ext2prepare (which doesn't
work for ext3 in any case, AFAIR, as it hasn't been updated to use
the ext3-style resize inode).

I think (without having looked) that it should be possible to modify
resize2fs to do a "resize" of the filesystem to the desired maximum
size (moving files, itable, etc out of the way of the group descriptor
table), but leaving the end of the filesystem alone.  Then, all it
would need to do is change s_reserved_gdt_blocks, and allocate the
reserved group descriptor blocks into the resize inode in the right order.

> > I'm also considering if we should at least install the ext2online
> > binary as ext3online and maybe have an (sym)link for compatibility.
> > It actually will not work on ext2 filesystems any more.
> 
> It does not?  Perhaps that is why I could not get the byteorder patch
> to work?  Please tell me more.

Well, the ext2online code in RH will not try the old ext2 "write to
block device from userspace" method (c.f. "disable-fallback" patch,
or similar).  To be honest, having ext3 do this work with the help of
the journal is much safer.

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