[Confirmed Bug] Parted does not notify kernel when modifying partition tables in partitionable md arrays

[Christopher White]

Greetings!

After some serious back-and-forth on the md mailing list, it finally 
turned out to be a bug in (g)parted, which has been reproduced by two 
people on separate machines using the latest versions of (g)parted. 
Here's how to reproduce the bug:

Step 1:
Create a PARTITIONABLE md array with mdadm (change parameters to 
whatever test parameters you use, but keep the --auto=partX flag to make 
it a partitionable md array):
mdadm --create --level=raid5 --auto=part2 /dev/md1 --metadata=1.2 
--raid-devices=4 /dev/sda2 /dev/sdb2 /dev/sdc2 /dev/sdd2

Step 2:
Verify that the device was created (ls /dev/md* will show /dev/md1 and 
nothing else since there are no partitions yet)

Step 3:
Run (g)parted on the MD array(this bug exists in both parted and 
gparted, I prefer gparted but phil verified it in parted as well):
parted /dev/md1

Step 4:
Create a partition table (type doesn't matter, I use GPT)

Step 5:
Create two partitions.

Step 6:
Apply the changes to write them to disk.

Step 7:
Check the device list (ls /dev/md*) and you will NOT see two partitions, 
you will still just see /dev/md1.

Step 8:
Manually force a device re-scan in the kernel (blockdev --rereadpt 
/dev/md1).

Step 9:
Check the device list (ls /dev/md*) and you WILL see two partitions 
(/dev/md1, /dev/md1p1 and /dev/md1p2).


This is a serious bug, because if you try to use (g)parted to not only 
make the partitions but also tell it to FORMAT the partitions in a 
particular way, it will actually FAIL when it tries to execute mkfs.XXX 
on each partition, since it cannot find /dev/md1p* due to the kernel not 
having re-scanned the partition table after the partitions were created.


Other tools such as fdisk work properly and will properly tell the 
kernel to re-scan the partition table after writing the partition table 
modifications.


(g)parted properly notifies the kernel about partition table changes 
when modifying PHYSICAL disks. The bug only applies to working with 
md-arrays.


Bug CAUSE:
Phil Turman managed to detect a probable cause for the bug, and I will 
quote him:

"I suspect (g)parted is in fact requesting a rescan, but is being ignored.
I just tried this on one of my servers, and parted (v2.3) choked on an 
assertion.  Hmm."


I hope this is enough information to get an idea of where the bug is, 
because I do not have the luxury of performing more testing, but all 
steps to fully reproduce the bug are above.

Useless side info but I'll mention it before anyone asks: Linux Mint 11 
RC, Kernel 2.6.38-8-generic.


------------------------
Christopher White