ext3 0.9.6 for kernel 2.4.9

Andrew Morton (akpm@zip.com.au)
Sat, 18 Aug 2001 14:23:15 -0700


This is just a rediff and retest - no ext3 changes in this release.

http://www.uow.edu.au/~andrewm/linux/ext3/

The latest diff is against linux-2.4.9. The version of ext3 in
-ac kernels is current.

-ac's ext3 has the "buffer tracing" debug code removed from the
non-ext3 files, so the 2.4.9 diff is more useful for bug hunting.

I should generate a diff against -ac to enable buffer tracing, but
there doesn't seem to be a lot of call for it now - ext3 is in
Red Hat's "Rawhide" kernels, is in -ac and downloads from the
above site are still running at >100/day. The bug report rate from
all of this is exceedingly low.

The current things outstanding for a 1.0 release are:

- There are now two reports of an interaction between ext3 and [v]fat.
The symptoms are that fatfs decides to return ENOTDIR when its
root directory is read. This is rather bizarre - it may be due to
ext3 leaving some unexpected state in a buffer_head, and fatfs
is assuming that its buffers are coming in with that state zeroed.

If anyone can reproduce this, please send me your .config, details
of mountpoints, filesystem sizes, system config, shoe size, Mother's
maiden name, etc. Thanks.

- Two reports of an assertion failure in journal_revoke() - where the
block was already revoked. One report was from the intermezzo developers
who were using `chattr +J' to enable data journalling on just part of a
filesystem (I really haven't been testing this feature at all). The
other reporter had never used `chattr +J'.

We can make this go away by clearing the revoke bits when refiling
buffers for checkpoint writeback, but this should not be necessary.
Still hunting for this one.

- Stephen has been working a largish change to the ext3 error handling.
At present ext3 tends to go BUG() if any internal inconsistency is
detected (ext3 is *very* defensive of your data). But this is rather
harsh so Stephen's changes are aimed at handling these errors more
selectively - generally by turning the filesystem read-only when
an inconsistency is detected.

- Documentation/filesystems/ext3.txt :)

Also, Andreas is working on changes to ext3's external journal
management. This is post-1.0 material.

-
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/