[parted-devel] [PATCH] Ignore devices where we can't create a new PedDisk

[David Cantrell]

On Wed, 01 Aug 2007 09:45:44 +0200
Jim Meyering <jim at meyering.net> wrote:

> Hi David,
> 
> What's the motivation for this patch?
> IMHO, such a change should be accompanied by
> a test case, or at the very least enough explanation
> to construct a test case.
> 
> This change (on the trunk) causes "make check" to fail
> all but two tests, so I've reverted it there.

First, know that I'm pretty much only working with the stable-1.8.x branch.  I don't even have time to work with trunk, ever.  Otavio suggested the patches be made to trunk, then cherry-picked to stable.  Well, I'm the stable branch maintainer and that's not a very effective way for me to work.  If a stable patch is applicable to trunk, we can cherry-pick the other way.

On to the patch.  While working on a RHEL bug, I discovered that the linux_probe_all() function in libparted/arch/linux.c walks three possible sources to find block devices:

	- A list of "standard" devices hardwired in linux.c (just hdX and sdX devices)
	- The contents of /sys/block with ., .., dm-*, and ram* filtered out
	- The contents of /proc/partitions

The /proc/partitions file is looked at only if /sys/block isn't available.  On my system, I had /sys/block available and there was one node that showed up that caused parted to crash.  That node was md0.  There were no software RAID devices on the system at all.  /sys/block/md0 was bogus.  But since libparted found it during the device probe, it tried to read a label from it.  Crash.

So I placed the ped_disk_new() test there because if it fails, the device wouldn't open, so I return NULL there.  Maybe that's the wrong place to do that, but the situation is valid.  We should test to make sure we can open the device during a probe all rather than assuming it can be read.

Should I have written a test case...yes, but after hacking on it for a few days and PM yelling at me, I really just didn't even think about it.  Apologies.

-- 
David Cantrell <dcantrell at redhat.com>
Red Hat / Westford, MA
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 189 bytes
Desc: not available
Url : http://lists.alioth.debian.org/pipermail/parted-devel/attachments/20070801/461e8446/attachment.pgp