On Mon, 2018-06-04 at 18:41 -0400, Kent Overstreet wrote:
> On Mon, Jun 04, 2018 at 08:59:39AM +0000, Bart Van Assche wrote:
> > Please comment out the getuid_callout line from /etc/multipath.conf and try again.
> > Changing the "/..." part into the path of the srp-test software should also help.
> > That line is a left-over from the time when the NVMe initiator drivers supported
> > dm-multipath. I will update the srp-test README.md to make this more clear.
>
> Deleted that line and it still doesn't work (I also tried fixing that line so
> the path was correct, but got the same output as what I sent you earlier). With
> that line deleted, the output is different. test output still says SRP login
> failed, syslog says:
>
> Jun 04 18:35:31 moria-kvm-kvm-kvm-kvm-kvm multipathd[453]: --------start up--------
> Jun 04 18:35:31 moria-kvm-kvm-kvm-kvm-kvm multipathd[453]: read /etc/multipath.conf
> Jun 04 18:35:31 moria-kvm-kvm-kvm-kvm-kvm multipathd[453]: path checkers start up
> Jun 04 18:35:31 moria-kvm-kvm-kvm-kvm-kvm multipathd[453]: sda: fail to get serial
> Jun 04 18:35:31 moria-kvm-kvm-kvm-kvm-kvm multipathd[453]: sdb: fail to get serial
> Jun 04 18:35:31 moria-kvm-kvm-kvm-kvm-kvm multipathd[453]: nbd0: HDIO_GETGEO failed with 25
> Jun 04 18:35:31 moria-kvm-kvm-kvm-kvm-kvm multipathd[453]: nbd0: failed to get udev uid: Invalid argument
> Jun 04 18:35:31 moria-kvm-kvm-kvm-kvm-kvm multipathd[453]: nbd1: HDIO_GETGEO failed with 25
> Jun 04 18:35:31 moria-kvm-kvm-kvm-kvm-kvm multipathd[453]: nbd1: failed to get udev uid: Invalid argument
> Jun 04 18:35:31 moria-kvm-kvm-kvm-kvm-kvm multipathd[453]: nbd10: HDIO_GETGEO failed with 25
Thanks for having shared this output. In the multipathd source code I found that
the error message "fail to get serial" is produced by the following code:
if (get_vpd_sysfs(parent, 0x80, pp->serial, SERIAL_SIZE) <= 0) {
if (get_serial(pp->serial, SERIAL_SIZE, pp->fd)) {
condlog(2, "%s: fail to get serial", pp->dev);
return 0;
}
}
The message "fail to get serial" is only logged if reading
/sys/class/scsi_device/*/*/*/*/vpd_pg80 fails *and* submitting a SCSI INQUIRY
command through the SG_IO interface fails. This is something I haven't encountered
before. Is there something special about the VM? Was e.g. multipathd configured
to run as another user than root? Have some security settings been configured that
prevent multipathd from accessing sysfs attributes? Was SCSI generic support
disabled (CHR_DEV_SG)?
Thanks,
Bart.
