Bug 614357

Summary:	[RHEL6 Beta2] Unable to access a partition created by the parted command.
Product:	Red Hat Enterprise Linux 6	Reporter:	Mark Wu <dwu>
Component:	parted	Assignee:	Brian Lane <bcl>
Status:	CLOSED NOTABUG	QA Contact:	Release Test Team <release-test-team-automation>
Severity:	high	Docs Contact:
Priority:	high
Version:	6.0	CC:	adingman, atesin, bmr, david.duffey, david, gspurgeon, hdegoede, jlau, kmaiti, piyushgaur123, ricardo.arguello, samukawa-oxa, tao, vorpal, wboessen
Target Milestone:	rc	Keywords:	Reopened
Target Release:	---
Hardware:	All
OS:	Linux
Whiteboard:
Fixed In Version:		Doc Type:	Bug Fix
Doc Text:		Story Points:	---
Clone Of:		Environment:
Last Closed:	2011-06-16 22:05:37 UTC	Type:	---
Regression:	---	Mount Type:	---
Documentation:	---	CRM:
Verified Versions:		Category:	---
oVirt Team:	---	RHEL 7.3 requirements from Atomic Host:
Cloudforms Team:	---	Target Upstream Version:
Embargoed:
Bug Depends On:
Bug Blocks:	717126

Description Mark Wu 2010-07-14 09:00:38 UTC

Description of problem:
The new partition created by parted can't be accessed without reboot. The ioctl call with parameter BLKRRPART returns EBUSY, so kernel can't rescan the partition table after changing the partition table. No partition is mounted, used as swap, raid, or pv, so it shouldn't be busy.


Version-Release number of selected component (if applicable):
kernel-2.6.32-37.el6.x86_64
parted-2.1-8.el6.x86_64

How reproducible:
100%

Steps to Reproduce:
1.# parted -s /dev/sda unit MiB print
 Model: LSI MegaRAID SAS RMB (scsi)
 Disk /dev/sda: 100000MiB
 Sector size (logical/physical): 512B/512B
 Partition Table: msdos

 Number  Start     End       Size      Type     File system     Flags
  1      0.03MiB   500MiB    500MiB    primary  ext3            boot
  2      500MiB    40500MiB  40000MiB  primary  ext3
  3      40500MiB  44500MiB  4000MiB   primary  linux-swap(v1)
 <There is no 4th partition yet.>
2.parted -s /dev/sda mkpart primary 44502MiB 100000MiB
 Warning: WARNING: the kernel failed to re-read the partition table
 on /dev/sda (Device or resource busy).  As a result, it may not reflect
 all of your changes until after reboot.
3.[root@localhost ~]# partprobe 
Warning: WARNING: the kernel failed to re-read the partition table on /dev/sda (Device or resource busy).  As a result, it may not reflect all of your changes until after reboot.

  
Actual results:
can't see the partition new created in /proc/partitions 


Expected results:
The new created partition can be accessed without reboot


Additional info:
Even without change of partition table,  partprobe reports the same warning message. 
This problem has been reported on Fedora 12:
https://bugzilla.redhat.com/show_bug.cgi?id=533851

Comment 2 Mark Wu 2010-07-14 09:02:30 UTC

Unfortunately,  when I try to use systemtap to debug it, kernel panicked.

Comment 3 RHEL Program Management 2010-07-15 15:19:06 UTC

This issue has been proposed when we are only considering blocker
issues in the current Red Hat Enterprise Linux release. It has
been denied for the current Red Hat Enterprise Linux release.

** If you would still like this issue considered for the current
release, ask your support representative to file as a blocker on
your behalf. Otherwise ask that it be considered for the next
Red Hat Enterprise Linux release. **

Comment 4 Hans de Goede 2010-07-20 14:29:37 UTC

If you get an ebusy error something is using the disk in some way. You can try the following to find the culprit:

cat /proc/mounts
cat /proc/swaps
dmsetup table
lsof | grep sda

Comment 5 Andrew Dingman 2010-07-20 15:30:26 UTC

NOTABUG? Seriously?

Of *course* sda is in use. It will typically contain /boot, swap, and at least one LVM PV. So what? At least in RHEL5, I could add a new partition and expect it to be recognized without a reboot. (Heck, I'm pretty sure this could be done live on RHEL 3 and 4, but I don't have them handy to test.) Now if I add a new partition I have to re-boot. This is a loss of functionality that makes it more likely a customer has to re-boot a box. Why would this be OK?

Sure, everyone should be using LVM, which reduces the impact of the problem. But not everyone is.

Comment 6 Hans de Goede 2010-07-21 07:35:07 UTC

(In reply to comment #5)
> NOTABUG? Seriously?
> 

Yes I'm afraid so.

> Of *course* sda is in use. It will typically contain /boot, swap, and at least
> one LVM PV. So what? At least in RHEL5, I could add a new partition and expect
> it to be recognized without a reboot.

Correct, in RHEL-5 you can also change an in use partition, and parted will happily report that it had told the kernel about the changes, even though it did not. The problem with the RHEL-5 way of doing things is there was no error checking, so sometimes it worked sometimes it did not work, but did not report this.

So in Parted 2.x (2.1 I think) the code was changed from trying to be clever
to doing things the same way fdisk does them. Which is a single ioctl asking
the kernel to re-read the partition-table, which fails with -EBUSY when any partition is in use.

> Now if I add a new
> partition I have to re-boot. This is a loss of functionality that makes it more
> likely a customer has to re-boot a box. Why would this be OK?

Given the choice between silent failure and then creating for example a FS on a changed partition with the wrong size, or making it more likely a customer has to re-boot a box, the reboot option is indeed OK I'm afraid.

Comment 20 David Duffey 2010-12-03 17:47:36 UTC

It appears that various RHEL6 partitioning tools have different behaviors. For example, using the graphical palimpsest tool you are able to make and use more primary partitions when RHEL6 fdisk and partprobe fail.

Why the difference between the tools?

Second, the palimpsest tool is unable make a usable extended partition without a reboot, but after the reboot is able to make logical partitions within the extended partition.

David

Comment 21 Wander Boessenkool 2010-12-08 10:42:54 UTC

I don't see why forcing a reboot on people is a safer option. If I make an overlapping partition and run partprobe I get corruption NOW, if I create an overlapping partition and reboot I get corruption after the reboot. If we want to stoop people from shooting themselves in the foot fdisk/parted etc. should be fixed to warn about overlapping partitions.
Having to reboot a machine to add a partition should be seen as a major regression IMHO, not a NOTABUG.

Comment 22 John Lau 2011-05-24 08:50:23 UTC

*** Bug 707149 has been marked as a duplicate of this bug. ***

Comment 24 Brian Lane 2011-05-27 13:18:28 UTC

See comment 16.

Comment 25 Brian Lane 2011-05-27 18:43:05 UTC

I have re-evaluated this problem after discussing it with Jim on IRC:

1. parted 2.3 fixes this. It allows changes to unmounted partitions and blocks changes to both the disk label and mounted partitions.

2. The changes involved in getting from 2.1 to 2.3 are not simple. Back porting those changes could result in different bugs, cause problems with programs using libparted, or some other unknown.

In the interest of not breaking things that are currently working I will not be fixing this on 2.1

Comment 30 Brian Lane 2011-12-19 22:59:43 UTC

*** Bug 717126 has been marked as a duplicate of this bug. ***

Comment 31 Brian Lane 2012-01-23 17:41:57 UTC

*** Bug 783781 has been marked as a duplicate of this bug. ***

Comment 34 atesin 2013-06-11 15:09:39 UTC

(In reply to Wander Boessenkool from comment #21)
> I don't see why forcing a reboot on people is a safer option. If I make an
> overlapping partition and run partprobe I get corruption NOW, if I create an
> overlapping partition and reboot I get corruption after the reboot. If we
> want to stoop people from shooting themselves in the foot fdisk/parted etc.
> should be fixed to warn about overlapping partitions.
> Having to reboot a machine to add a partition should be seen as a major
> regression IMHO, not a NOTABUG.

i agree with you ... i think force reboot "as a safety measure" is pointless because you in fact CAN DELETE all your partitions even while system running, the system will fail for sure at partition update or at reboot, doesnt matter .. just removes functionality

you can even totally mess up your entire disk/partition/filesystem without any warning with dd, so you can bypass this -security measure- if you want, but if you are a good admin you know what are you doing

i think fdisk/partprobe/parted/etc. would check if changing a *partition* in use not whole disk, and if so display a confirmation (think you want live repartioning your disk for system reinstallation at next reboot).. sometimes one intentionally left unpartitioned space for future use on demand .. cannot reboot because a lot of services running, but need disk space NOW

i work in a big TV station and rebooting is not an option, even get the homepage offline a couple seconds would be a catastrophe