#44 big-endian mkudffs goes boom

[Pat LaVarre]

The released version of mkudffs lacks a
little/big-endian adjustment that the CVS version has.

Low priority until someone expert in sourceforge.net
volunteers to release what now works in CVS.

[Pat LaVarre]

Logged In: YES
user_id=964855

Kindly donated offline, from an old human memory, we have the claim:

---
with out the patch, ... mkudffs ... segmentation fault ...
disc->udf_lvd[0]->numPartitionMaps ... little endian ...
with the byte swap ...
the code runs past the end of the udf_lvd[] array.
---

[YoungdoLee]

Logged In: YES
user_id=399786

maybe this problem

diff -urN ../../udftools-1.0.0b3/mkudffs/mkudffs.c ./mkudffs.c
--- ../../udftools-1.0.0b3/mkudffs/mkudffs.c 2004-02-23
12:35:33.000000000 +0900
+++ ./mkudffs.c 2004-08-25 22:32:08.000000000 +0900
@@ -250,14 +250,18 @@
if (size % offsets[PSPACE_SIZE])
size -= (size % offsets[PSPACE_SIZE]);
set_extent(disc, PSPACE, start, size);
- for (i=0; i<disc->udf_lvd[0]->numPartitionMaps; i++)
+
+// tifler : bugs on big endian machine
+// for (i=0; i<disc->udf_lvd[0]->numPartitionMaps; i++)
+ for (i=0; i<le32_to_cpu(disc->udf_lvd[0]-
>numPartitionMaps); i++)
{
if (i == 1)
disc->udf_lvid->freeSpaceTable[i] =
cpu_to_le32(0xFFFFFFFF);
else
disc->udf_lvid->freeSpaceTable[i] =
cpu_to_le32(size);
}
- for (j=0; j<disc->udf_lvd[0]->numPartitionMaps; j++)
+// for (j=0; j<disc->udf_lvd[0]->numPartitionMaps; j++)
+ for (j=0; j<le32_to_cpu(disc->udf_lvd[0]-
>numPartitionMaps); j++)
{
if (j == 1)
disc->udf_lvid->sizeTable[i+j] = cpu_to_le32
(0xFFFFFFFF);

[Pat LaVarre]

Logged In: YES
user_id=964855

Hmmm. I'm not sure I made myself clear.

Me, I meant to say that we find two versions of mkudffs at linux-udf at
sourceforge.net. I see both versions work in x86, but I'm told one
version segfaults in PPC.

The mkudffs that purportedly segfaults appears in the released version of
1.0.0.b3 udftools at:

https://sourceforge.net/project/showfiles.php?
group_id=295&package_id=38129&release_id=218959

The mkudffs that works appears in CVS if your firewall lets you supply
an empty password in response to a command sequence like:

export CVSROOT=:pserver:anonymous@cvs.sourceforge.net:/cvsroot/
linux-udf
cvs login
cvs co udftools
cvs logout

The issue I relayed to here from the friend who discovered it is that ...

No one knowledgeable and privileged enough to fix this discrepancy has
fixed it, as yet.

That issue I don't see your reply addressing. However, yes indeed, the
diff between the source code of these two versions does include the code
you kindly highlighted. Specifically today, if I avoid counting absent
files, I see 179 lines of difference which include the 20 lines:

diff -urp udftools-1.0.0b3/mkudffs/mkudffs.c udftools-cvs/mkudffs/
mkudffs.c
--- udftools-1.0.0b3/mkudffs/mkudffs.c Sun Feb 22 20:35:33 2004
+++ udftools-cvs/mkudffs/mkudffs.c Mon Mar 1 19:09:39 2004
@@ -250,14 +250,14 @@ void split_space(struct udf_disc *disc)
if (size % offsets[PSPACE_SIZE])
size -= (size % offsets[PSPACE_SIZE]);
set_extent(disc, PSPACE, start, size);
- for (i=0; i<disc->udf_lvd[0]->numPartitionMaps; i++)
+ for (i=0; i<le32_to_cpu(disc->udf_lvd[0]->numPartitionMaps); i++)
{
if (i == 1)
disc->udf_lvid->freeSpaceTable[i] =
cpu_to_le32(0xFFFFFFFF);
else
disc->udf_lvid->freeSpaceTable[i] = cpu_to_le32(size);
}
- for (j=0; j<disc->udf_lvd[0]->numPartitionMaps; j++)
+ for (j=0; j<le32_to_cpu(disc->udf_lvd[0]->numPartitionMaps); j++)
{
if (j == 1)
disc->udf_lvid->sizeTable[i+j] = cpu_to_le32(0xFFFFFFFF);

[Pat LaVarre]

Logged In: YES
user_id=964855

With less than a couple of days of work, I have discovered how to run
Linux on a (big-endian) (PPC) Mac without sacrificing its hard drive, via
slightly reconfiguring the version of Gentoo Linux live CD that includes a
GUI desktop.

But I still can't build a mkudffs to try there myself because I haven't yet
figured out how to install gcc and make. My effort to date appears
detailed at:

Subject: bugs.gentoo.org
http://linux-pel.blog-city.com/read/782855.htm