[Openstack] bizarre segfaults with CentOS 7 and LVM/iscsi

Chris Friesen chris.friesen at windriver.com
Thu May 7 23:49:47 UTC 2015


Hi all,

I've got a bizarre situation.  I've got a couple of little openstack clouds, one 
where cinder uses LVM/iscsi and the other with cinder using rbd/ceph.

I'm trying to bring up the CentOS 7 qcow2 image from 
"http://cloud.centos.org/centos/7/images/".

In both clouds I can boot an instance from the image with no problems.

In the cloud using ceph for cinder I can boot from volume and it works as expected.

In the cloud using LVM/iscsi I get segfaults in the guest logs when it boots up:

[    3.672576] cloud-init[429]: segfault at 7fffe9bf4f50 ip 00007ffff7de69a9 sp 
00007fffffffc620 error 6 in ld-2.17.so[7ffff7ddb000+21000]
[   20.888077] postconf[711]: segfault at 8 ip 00007ffff7de66d8 sp 
00007fffffffea30 error 4 in ld-2.17.so[7ffff7ddb000+21000]
[   20.979021] cloud-init[708]: segfault at 7fffe9bf4f50 ip 00007ffff7de69a9 sp 
00007fffffffc630 error 6 in ld-2.17.so[7ffff7ddb000+21000]

If I log in anyways and check the md5sum of /bin/vi it's different than expected 
(and it's different every time I reboot the instance, but stays the same while 
running). If I shutdown the guest and mount the iscsi device directly on the 
compute host then the filesystem looks fine and the md5sum of /bin/vi on that 
mount is as expected.

Some final pieces of data--the stock Ubuntu 14.10 and Fedora 21 cloud images 
work fine booting from volume on both clouds.

Anyone got any ideas?  I've disabled selinux, I've turned off the 
"kernel.randomize_va_space" and "kernel.exec-shield" sysctl options, but so far 
nothing helps.

Thanks,
Chris




More information about the Openstack mailing list