From: Kalpak Shah <Kalpak.Shah@Sun.COM>
Subject: [e2fsprogs] Failing extents tests
Date: Mon, 29 Sep 2008 19:17:35 +0530
Message-ID: <1222696055.3971.42.camel@localhost>
Mime-Version: 1.0
Content-Type: text/plain
Content-Transfer-Encoding: 7BIT
Cc: linux-ext4 <linux-ext4@vger.kernel.org>
To: TheodoreTso <tytso@mit.edu>
Sender: linux-ext4-owner@vger.kernel.org

Hi,

The extents support in e2fsprogs-1.41.1 doesn't pass a few of the
regression tests that we had created. 4 tests out of 20 are failing, I
have included some details about the failures.

e2fsprogs-tests-f_extents_ee_len.patch: Corrupt ee_len is not detected. 
e2fsprogs-tests-f_extents_ei_block.patch: Corrupt ei_block is not
detected. In the Sun extents code, we passed the previous extent to
ext2fs_extent_verify(), so additional sanity checks could be performed.
Here is a snippet of that:

/* Verify that a single extent @ex is valid.  If @ex_prev is passed in,
 * then this was the previous logical extent in this block and we can
 * do additional sanity checking (though in case of error we don't know
 * which of the two extents is bad).  Similarly, if @ix is passed in
 * we can check that this extent is logically part of the index that
 * refers to it (though again we can't know which of the two is bad). */
errcode_t ext2fs_extent_verify(ext2_filsys fs, struct ext3_extent *ex,
                               struct ext3_extent *ex_prev,
                               struct ext3_extent_idx *ix, int ix_len)


Similarly we also passed previous ext3_extent_idx to
ext2fs_extent_index_verify().

Maybe we can do something similar by adding ex_prev and ix_prev to
ext2_extent_handle_t?

e2fsprogs-tests-f_extents_imbalanced_tree.patch,
e2fsprogs-tests-f_extents_eh_depth.patch: In both these cases, e2fsck
aborts with a "Corrupt extent header on inode %i" error. Instead of
aborting we can clear the extent header and set i_blocks to 0?

I have uploaded these tests at
http://downloads.lustre.org/public/tools/e2fsprogs/extent_tests/

One problem with these tests is that, the high 16 bits of extent/index
block are set, so these tests cannot be used as-is, but the image should
be helpful to check if the problem is solved or not.

Thanks,
Kalpak