Red Hat Bugzilla – Bug 269541
Processes hang and system is unrebootable after snapshot invalidation
Last modified: 2007-11-30 17:12:14 EST
Description of problem:
When a snapshot is taken of a logical volume and the snapshot fills up and
has to be invalidated, the process writing to the original volume hangs in an
unkillable wait state. Also, any subsequent processes that query the filesystem
tree hang as well. This includes any attempt to run "reboot" or "shutdown",
meaning the system has to be hard-power-cycled to restore it.
This is fixed upstream and seems to be fixed in the current development kernel
(2.6.23-0.42.rc3.git10.fc8) but that kernel is unusable on a Fedora Core 6
machine (requires a newer mkinitrd that segfaults if run on FC6, and has
numerous other dependency problems). So it would be nice to have an FC6 errata
issued, given that being hit by this bug requires a hard reset to restore normal
Version-Release number of selected component (if applicable):
Steps to Reproduce:
1. lvcreate -L 512M -n testlv vg1 [where vg1=name of a volume group with
2. mkfs /dev/vg1/testlv; mkdir /testmnt; mount /dev/vg1/testlv /testmnt
3. lvcreate -s -n testsnap -L 128M /dev/vg1/testlv
4. head -c 200m /dev/zero > /testmnt/largefile
"Invalidating snapshot" is logged to syslog (as expected).
The "head" process hangs in an unkillable wait state.
Any subsequent process that tries to access low-level filesystem tables also
hangs unkillably. This includes "reboot" and "shutdown".
"Invalidating snapshot" is logged.
The "head" process continues and finishes as normal.
Other processes all behave as normal as well, except that any attempt to
mount the snapshot or read data from it should give an I/O error.
Fixed in 2.6.23-0.42.rc3.git10.fc8 but that kernel as packaged can't be
installed on an FC6 system.
Patch found, will be in upcoming kernel (but not the next one that's already
submitted for updates-testing.)
In kernel-126.96.36.199-55.fc6, submitted for testing.