Re: [Fwd: ext2fs patch for large stores, RC1+20040304]

[Ognyan Kulev]

Michael Banck wrote:
> I tested this a bit by compiling glibc on  2.4GB partition. It seems to
> work quite well at first, but after a while, I got a couple of errors.
> Most of them were plain freezes or Resource lost. I also had quite some
> filesystem corruption when (forcibly) checking the file system. After
> reverting the patch and to a smaller partition, I was able to build
> glibc without problems :-/

Thank you for testing the patch :-)

> ext2fs.static: ../../ext2fs/pager.c:1045: disk_cache_block_ref:
>         Assertion '0 <= block && block <= store->size >>
>         log2_block_size' failed.

It seems that somewhere in code invalid block number (e.g., with highest 
bit set) is used.  Backtrace is very valuable for finding where is the 
problem.  I'll try to reproduce this bug.

> Also, I noticed that I am no longer able to mount my (rather small) root
> GNU/Linux partition, as it has less than 4kb blocks. I don't remember
> specifying anything special when I install GNU/Linux, so I wonder
> whether this is expected to be fixed/resolved?

The patched ext2fs doesn't work with block sizes != 4K.  If this is 
requirement for committing the patch, I'll fix it, if it's not, I prefer 
to address it later.  It's known fact that current ext2fs has problems 
with block sizes != 4K, but I don't know what are they.

> > And don't forget that if you test the patch, you'll probably want
> > patched e2fsprogs too.  You can get the patch from
> > http://debian.fmi.uni-sofia.bg/~ogi/hurd/ext3fs/ .  It applies cleanly
> > to e2fsprogs_1.35-2 (with some offsets, but without *.rej files).
>
> Did you try to get it applied upstream?

No, Marcus Brinkmann says we have to fix the 2G limit for files 
ourselves.  Original e2fsprogs uses Unix API (read/write syscalls on 
files) for working on file systems.  The Hurd API is 64-bit, but 
implementations (e.g. ext2fs and storeio) use libpager for all read and 
write operations, and libpager can't handle mappings (e.g. of files and 
devices) larger than 2G.  So either libpager's usage must be replaced 
with something else, or libpager should become 64-bit.  I have thought 
only about the latter, and it's not easy at all.  IMHO We should think 
again about applying the e2fsprogs patch.

Regards,
ogi