http://bugzilla.novell.com/show_bug.cgi?id=539521 User aj@novell.com added comment http://bugzilla.novell.com/show_bug.cgi?id=539521#c1 Andreas Jaeger changed: What |Removed |Added ---------------------------------------------------------------------------- Status|NEW |NEEDINFO Info Provider| |patz1983@googlemail.com --- Comment #1 from Andreas Jaeger 2009-10-02 05:36:15 MDT --- Please add the YaST log files for this as described at http://en.opensuse.org/Bugs/YaST -- Configure bugmail: http://bugzilla.novell.com/userprefs.cgi?tab=email ------- You are receiving this mail because: ------- You are on the CC list for the bug.

http://bugzilla.novell.com/show_bug.cgi?id=539521 User patz1983@googlemail.com added comment http://bugzilla.novell.com/show_bug.cgi?id=539521#c2 Matthias Patz changed: What |Removed |Added ---------------------------------------------------------------------------- Status|NEEDINFO |NEW Info Provider|patz1983@googlemail.com | --- Comment #2 from Matthias Patz 2009-10-07 08:34:39 MDT --- Created an attachment (id=321497) --> (http://bugzilla.novell.com/attachment.cgi?id=321497) tar of the yast2 logs -- Configure bugmail: http://bugzilla.novell.com/userprefs.cgi?tab=email ------- You are receiving this mail because: ------- You are on the CC list for the bug.

http://bugzilla.novell.com/show_bug.cgi?id=539521 User aschnell@novell.com added comment http://bugzilla.novell.com/show_bug.cgi?id=539521#c4 Arvin Schnell changed: What |Removed |Added ---------------------------------------------------------------------------- AssignedTo|aschnell@novell.com |kernel-maintainers@forge.pr | |ovo.novell.com --- Comment #4 from Arvin Schnell 2009-10-07 09:36:23 MDT --- The information YaST reads from /proc/partitions is outdated. After deleting sda1 /proc/partition still includes sda1. Even after creating sda1 the entry in /proc/partitions contains the old size. Only after creating sda2 the entires is /proc/partitions are correct. Looks like a kernel or communication problem between kernel and parted. -- Configure bugmail: http://bugzilla.novell.com/userprefs.cgi?tab=email ------- You are receiving this mail because: ------- You are on the CC list for the bug.

http://bugzilla.novell.com/show_bug.cgi?id=539521 http://bugzilla.novell.com/show_bug.cgi?id=539521#c6 Jeff Mahoney changed: What |Removed |Added ---------------------------------------------------------------------------- AssignedTo|kernel-maintainers@forge.pr |bnc-team-screening@forge.pr |ovo.novell.com |ovo.novell.com --- Comment #6 from Jeff Mahoney 2009-11-25 20:39:12 UTC --- This isn't a kernel problem. It's working as expected, as documented by Thomas's test in comment #5. parted is a normal program that writes to a block device directly. From the kernel's perspective, there is no difference between it, mkfs, or fsck. Until the kernel is notified that the partition table has changed, it will use the old partitions. blockdev --rereadpt is how the kernel is notified. -- Configure bugmail: http://bugzilla.novell.com/userprefs.cgi?tab=email ------- You are receiving this mail because: ------- You are on the CC list for the bug.

http://bugzilla.novell.com/show_bug.cgi?id=539521 http://bugzilla.novell.com/show_bug.cgi?id=539521#c8 Arvin Schnell changed: What |Removed |Added ---------------------------------------------------------------------------- Priority|P3 - Medium |P2 - High Severity|Minor |Critical --- Comment #8 from Arvin Schnell 2009-11-30 14:12:09 UTC --- Mismatches between partition layouts "on disk" and "in kernel data" can lead to data loss. E.g. during mkfs the device is still 55 GB big. But after the next reboot the device is suddenly only 2 GB big so the filesystem cannot access all of it's data anymore. Increasing Severity and Pririty. -- Configure bugmail: http://bugzilla.novell.com/userprefs.cgi?tab=email ------- You are receiving this mail because: ------- You are on the CC list for the bug.

http://bugzilla.novell.com/show_bug.cgi?id=539521 http://bugzilla.novell.com/show_bug.cgi?id=539521#c10 Jeff Mahoney changed: What |Removed |Added ---------------------------------------------------------------------------- CC| |kasievers@novell.com --- Comment #10 from Jeff Mahoney 2009-11-30 19:30:40 UTC --- I have a script that fails quickly on this. It creates and destroys a partition in a loop, failing when a partition exists (or doesn't) when it shouldn't (or should). It writes out the strace output for each operation and breaks out when there is a failure for analysis. ----->8----- parted /dev/sda mkpart logical 87.8GB 120GB while true; do strace -o strace.out parted /dev/sda rm 13 if grep sda13 /proc/partitions; then echo FAIL1 break fi strace -o strace.out parted /dev/sda mkpart logical 87.8GB 120GB if ! grep sda13 /proc/partitions; then echo FAIL2 break fi done -----8<----- Here's what I got: ioctl(3, BLKPG, {BLKPG_DEL_PARTITION, flags=0, datalen=152, {start=0, length=0, pno=1, devname="", volname=""}}) = -1 EBUSY (Device or resource busy) ioctl(3, BLKPG, {BLKPG_DEL_PARTITION, flags=0, datalen=152, {start=0, length=0, pno=2, devname="", volname=""}}) = 0 ioctl(3, BLKPG, {BLKPG_DEL_PARTITION, flags=0, datalen=152, {start=0, length=0, pno=3, devname="", volname=""}}) = -1 ENXIO (No such device or address) ioctl(3, BLKPG, {BLKPG_DEL_PARTITION, flags=0, datalen=152, {start=0, length=0, pno=4, devname="", volname=""}}) = -1 ENXIO (No such device or address) ioctl(3, BLKPG, {BLKPG_DEL_PARTITION, flags=0, datalen=152, {start=0, length=0, pno=5, devname="", volname=""}}) = -1 EBUSY (Device or resource busy) ioctl(3, BLKPG, {BLKPG_DEL_PARTITION, flags=0, datalen=152, {start=0, length=0, pno=6, devname="", volname=""}}) = -1 EBUSY (Device or resource busy) ioctl(3, BLKPG, {BLKPG_DEL_PARTITION, flags=0, datalen=152, {start=0, length=0, pno=7, devname="", volname=""}}) = -1 EBUSY (Device or resource busy) ioctl(3, BLKPG, {BLKPG_DEL_PARTITION, flags=0, datalen=152, {start=0, length=0, pno=8, devname="", volname=""}}) = -1 EBUSY (Device or resource busy) ioctl(3, BLKPG, {BLKPG_DEL_PARTITION, flags=0, datalen=152, {start=0, length=0, pno=9, devname="", volname=""}}) = 0 ioctl(3, BLKPG, {BLKPG_DEL_PARTITION, flags=0, datalen=152, {start=0, length=0, pno=10, devname="", volname=""}}) = 0 ioctl(3, BLKPG, {BLKPG_DEL_PARTITION, flags=0, datalen=152, {start=0, length=0, pno=11, devname="", volname=""}}) = -1 EBUSY (Device or resource busy) ioctl(3, BLKPG, {BLKPG_DEL_PARTITION, flags=0, datalen=152, {start=0, length=0, pno=12, devname="", volname=""}}) = -1 EBUSY (Device or resource busy) ioctl(3, BLKPG, {BLKPG_DEL_PARTITION, flags=0, datalen=152, {start=0, length=0, pno=13, devname="", volname=""}}) = -1 EBUSY (Device or resource busy) ioctl(3, BLKPG, {BLKPG_DEL_PARTITION, flags=0, datalen=152, {start=0, length=0, pno=14, devname="", volname=""}}) = -1 ENXIO (No such device or address) ioctl(3, BLKPG, {BLKPG_DEL_PARTITION, flags=0, datalen=152, {start=0, length=0, pno=15, devname="", volname=""}}) = -1 ENXIO (No such device or address) ioctl(3, BLKPG, {BLKPG_DEL_PARTITION, flags=0, datalen=152, {start=0, length=0, pno=16, devname="", volname=""}}) = -1 ENXIO (No such device or address) ioctl(3, BLKPG, {BLKPG_ADD_PARTITION, flags=0, datalen=152, {start=139829760, length=512, pno=2, devname="/dev/sda2", volname=""}}) = 0 ioctl(3, BLKPG, {BLKPG_ADD_PARTITION, flags=0, datalen=152, {start=62446358016, length=6004422144, pno=9, devname="/dev/sda9", volname=""}}) = 0 ioctl(3, BLKPG, {BLKPG_ADD_PARTITION, flags=0, datalen=152, {start=68450812416, length=6448587264, pno=10, devname="/dev/sda10", volname=""}}) = 0 -- Configure bugmail: http://bugzilla.novell.com/userprefs.cgi?tab=email ------- You are receiving this mail because: ------- You are on the CC list for the bug.

http://bugzilla.novell.com/show_bug.cgi?id=539521 http://bugzilla.novell.com/show_bug.cgi?id=539521#c12 --- Comment #12 from Jeff Mahoney 2009-11-30 20:05:46 UTC --- Ah, but adding a udevadm settle in there doesn't account for the fact that the race is internal to parted. In my case, every BLKPG_DEL_PARTITION but sda2 (the extended partition table) fails with either -EBUSY or -ENXIO. udev could still be processing the events. When I add a system("/sbin/udevadm settle"); after the ioctl calls in parted, it seems to work fine. -- Configure bugmail: http://bugzilla.novell.com/userprefs.cgi?tab=email ------- You are receiving this mail because: ------- You are on the CC list for the bug.

http://bugzilla.novell.com/show_bug.cgi?id=539521 http://bugzilla.novell.com/show_bug.cgi?id=539521#c14 --- Comment #14 from Petr Uzel 2009-12-01 08:18:51 UTC --- Parted upstream recently switched back to use of BLKRRPART (from BLKPG_*) ioctls, which should fix this kind of problems - I'll check and report. -- Configure bugmail: http://bugzilla.novell.com/userprefs.cgi?tab=email ------- You are receiving this mail because: ------- You are on the CC list for the bug.

http://bugzilla.novell.com/show_bug.cgi?id=539521 http://bugzilla.novell.com/show_bug.cgi?id=539521#c16 --- Comment #16 from Petr Uzel 2009-12-04 13:54:15 UTC --- Patches [1] and [2] from parted repository remove some unnecessary open/close calls and thus avoiding unnecessary udev activity => fix the EBUSY issue for me. [1] http://git.debian.org/?p=parted/parted.git;a=commit;h=2a6936fab4d4499a4b812d... [2] http://git.debian.org/?p=parted/parted.git;a=commit;h=ad25892bb995f61b0ddf80... -- Configure bugmail: http://bugzilla.novell.com/userprefs.cgi?tab=email ------- You are receiving this mail because: ------- You are on the CC list for the bug.

http://bugzilla.novell.com/show_bug.cgi?id=539521 http://bugzilla.novell.com/show_bug.cgi?id=539521#c22 Swamp Workflow Management changed: What |Removed |Added ---------------------------------------------------------------------------- Status Whiteboard|maint:planned:update |maint:running:29521 --- Comment #22 from Swamp Workflow Management 2009-12-10 16:34:29 UTC --- The SWAMPID for this issue is 29521. Please submit the patch and patchinfo file using this ID. (https://swamp.suse.de/webswamp/wf/29521) -- Configure bugmail: http://bugzilla.novell.com/userprefs.cgi?tab=email ------- You are receiving this mail because: ------- You are on the CC list for the bug.

http://bugzilla.novell.com/show_bug.cgi?id=539521 http://bugzilla.novell.com/show_bug.cgi?id=539521#c27 --- Comment #27 from Petr Uzel 2009-12-16 12:02:02 UTC --- (In reply to comment #15)

Unfortunately we will completely loose the ability to modify partitions on a disk where at least one partition is in use. This will make a lot of usage scenarios of YaST2 in installed systems impossible and it would also most probably break installation scenarios like having installation sources on the disk.

Yes, that's correct. As pointed out in [1], can't we use addpart(8), delpart(8) instead? [1] http://www.mail-archive.com/parted-devel@lists.alioth.debian.org/msg02826.ht... -- Configure bugmail: http://bugzilla.novell.com/userprefs.cgi?tab=email ------- You are receiving this mail because: ------- You are on the CC list for the bug.

http://bugzilla.novell.com/show_bug.cgi?id=539521 http://bugzilla.novell.com/show_bug.cgi?id=539521#c29 Swamp Workflow Management changed: What |Removed |Added ---------------------------------------------------------------------------- Status Whiteboard|maint:running:29521 |maint:running:29521 |maint:released:sle10-sp2:29 |maint:released:sle10-sp3:29 |638 |641 --- Comment #29 from Swamp Workflow Management 2010-01-07 23:10:11 UTC --- Update released for: parted, parted-32bit, parted-64bit, parted-debuginfo, parted-devel, parted-x86 Products: SLE-DEBUGINFO 10-SP3 (i386, ia64, ppc, s390x, x86_64) SLE-DESKTOP 10-SP3 (i386, x86_64) SLE-SDK 10-SP3 (i386, ia64, ppc, s390x, x86_64) SLE-SERVER 10-SP3 (i386, ia64, ppc, s390x, x86_64) -- Configure bugmail: http://bugzilla.novell.com/userprefs.cgi?tab=email ------- You are receiving this mail because: ------- You are on the CC list for the bug.

http://bugzilla.novell.com/show_bug.cgi?id=539521 http://bugzilla.novell.com/show_bug.cgi?id=539521#c31 --- Comment #31 from Petr Uzel 2010-01-20 13:37:01 UTC --- Simple script: #!/bin/bash parted -s $/dev/sdb mkpart primary ext2 0 10M lsof +c 15 /dev/sdb1 parted -s /dev/sdb rm 1 grep sdb1 /proc/partitions && echo "FAIL" reveals that the problem is that after the partition is created, udev triggers hal to probe the device with hald-probe-volume. Later, when parted tries to remove the partition, hal has the device still open and thus parted's ioctl results in EBUSY => /proc/partitions not updated. udevadm settle isn't enough to serialize access to the disk. I can't think out anything better then doing simple sleep(1) after each modification of partition table. Does anybody have better idea? -- Configure bugmail: http://bugzilla.novell.com/userprefs.cgi?tab=email ------- You are receiving this mail because: ------- You are on the CC list for the bug.

http://bugzilla.novell.com/show_bug.cgi?id=539521 http://bugzilla.novell.com/show_bug.cgi?id=539521#c33 --- Comment #33 from Petr Uzel 2010-01-21 11:50:30 UTC --- (In reply to comment #32)

In libstorage we sent a signal SIGSTOP to process /usr/sbin/hald before executing parted and sent SIGCONT after parted is finished. Maybe this would also be an option for parted. I added this to fix bug #224516.

This does not help in case of two subsequent calls to parted like in comment #31 (SIGSTOPping hald does not make it release the device => EBUSY). hald would have to be stopped before the first call and resumed again after the second call - not doable in parted itself (I assume libstorage stops/starts hald before/after _each_ call to parted, right?). Wrapping multiple subsequent calls to parted in libstorage with hald stop/restart should work. Is that an option?

What tests did you do that made you reopen the bug in comment #30?

The one from comment #31.

If you were using yast2 partitioning the problem should have been fixed.

IMHO no - quick subsequent calls to parted could still fail. -- Configure bugmail: http://bugzilla.novell.com/userprefs.cgi?tab=email ------- You are receiving this mail because: ------- You are on the CC list for the bug.

http://bugzilla.novell.com/show_bug.cgi?id=539521 http://bugzilla.novell.com/show_bug.cgi?id=539521#c35 --- Comment #35 from Petr Uzel 2010-01-25 14:46:21 UTC --- I've implemented a patch that retries every BLKPG_DEL_PARTITION after a 1s sleep if the ioctl returns EBUSY (which either means the partition is mounted or something, like hal, has opened the partition). The drawback is that syncing kernel view of partition table will take longer on disks where some of the partitions are mounted (1s for every mounted partition) [*]. If nobody objects, I will push this fix tomorrow. [*] Checking whether the partition is mounted (to avoid unnecessary sleeping) is not an option because it would require big code changes I don't want to do in maintenance update. -- Configure bugmail: http://bugzilla.novell.com/userprefs.cgi?tab=email ------- You are receiving this mail because: ------- You are on the CC list for the bug.

http://bugzilla.novell.com/show_bug.cgi?id=539521 http://bugzilla.novell.com/show_bug.cgi?id=539521#c37 Marcus Meissner changed: What |Removed |Added ---------------------------------------------------------------------------- Status|REOPENED |NEEDINFO CC| |meissner@novell.com Info Provider| |puzel@novell.com --- Comment #37 from Marcus Meissner 2010-01-29 14:28:08 UTC --- i saw checkins from 26th... are they ok? what is supposed to be the top cvhanges? 11.2 has /work//SRC/old-versions/11.2/UPDATES/all/parted/parted.changes ------------------------------------------------------------------- Mon Jan 25 15:54:17 UTC 2010 - puzel@novell.com - use-ext-range.patch (bnc#567652) -- Configure bugmail: http://bugzilla.novell.com/userprefs.cgi?tab=email ------- You are receiving this mail because: ------- You are on the CC list for the bug.

http://bugzilla.novell.com/show_bug.cgi?id=539521 http://bugzilla.novell.com/show_bug.cgi?id=539521#c40 --- Comment #40 from Marcus Meissner 2010-03-01 13:18:13 UTC --- its in QA. just waiting. -- Configure bugmail: http://bugzilla.novell.com/userprefs.cgi?tab=email ------- You are receiving this mail because: ------- You are on the CC list for the bug.

http://bugzilla.novell.com/show_bug.cgi?id=539521 http://bugzilla.novell.com/show_bug.cgi?id=539521#c47 --- Comment #47 from Petr Uzel 2010-04-12 14:34:13 UTC --- I've submitted new fix based on what's being discussed in [1]. Parted retries the BLKPG ioctl() after short sleep if it fails. If also second attempt fails, parted tries to verify whether it is actually necessary to update in-kernel view of given partition (this check may fail as well, however). If none of the above steps helps, parted issues a warning that it wasn't able to inform kernel about changes of given partition(s) (before, parted silently ignored that the kernel hasn't been informed). [1] http://lists.alioth.debian.org/pipermail/parted-devel/2010-March/003518.html -- Configure bugmail: http://bugzilla.novell.com/userprefs.cgi?tab=email ------- You are receiving this mail because: ------- You are on the CC list for the bug.

http://bugzilla.novell.com/show_bug.cgi?id=539521 http://bugzilla.novell.com/show_bug.cgi?id=539521#c49 Swamp Workflow Management changed: What |Removed |Added ---------------------------------------------------------------------------- Status Whiteboard|maint:running:29521 |maint:running:29521 |maint:released:sle10-sp3:29 |maint:released:sle10-sp3:29 |641 |641 |maint:released:11.0:32631 |maint:released:11.1:32632 --- Comment #49 from Swamp Workflow Management 2010-04-16 14:06:46 UTC --- Update released for: parted, parted-debuginfo, parted-debugsource, parted-devel, parted-lang Products: openSUSE 11.1 (debug, i586, ppc, ppc64, x86_64) -- Configure bugmail: http://bugzilla.novell.com/userprefs.cgi?tab=email ------- You are receiving this mail because: ------- You are on the CC list for the bug.

http://bugzilla.novell.com/show_bug.cgi?id=539521 http://bugzilla.novell.com/show_bug.cgi?id=539521#c51 Swamp Workflow Management changed: What |Removed |Added ---------------------------------------------------------------------------- Status Whiteboard|maint:running:29521 |maint:running:29521 |maint:released:sle10-sp3:29 |maint:released:sle10-sp3:29 |641 |641 |maint:released:11.2:32633 |maint:released:sle11:31797 --- Comment #51 from Swamp Workflow Management 2010-04-16 14:07:41 UTC --- Update released for: parted, parted-32bit, parted-debuginfo, parted-debuginfo-32bit, parted-debuginfo-x86, parted-debugsource, parted-devel, parted-x86 Products: SLE-DEBUGINFO 11 (i386, ia64, ppc64, s390x, x86_64) SLE-DESKTOP 11 (i386, x86_64) SLE-SDK 11 (i386, ia64, ppc64, s390x, x86_64) SLE-SERVER 11 (i386, ia64, ppc64, s390x, x86_64) -- Configure bugmail: http://bugzilla.novell.com/userprefs.cgi?tab=email ------- You are receiving this mail because: ------- You are on the CC list for the bug.

http://bugzilla.novell.com/show_bug.cgi?id=539521 http://bugzilla.novell.com/show_bug.cgi?id=539521#c52 Petr Uzel changed: What |Removed |Added ---------------------------------------------------------------------------- Status|ASSIGNED |RESOLVED Resolution| |FIXED --- Comment #52 from Petr Uzel 2010-04-19 11:36:12 UTC --- Fixed. -- Configure bugmail: http://bugzilla.novell.com/userprefs.cgi?tab=email ------- You are receiving this mail because: ------- You are on the CC list for the bug.