Google Groups no longer supports new Usenet posts or subscriptions. Historical content remains viewable.
Dismiss

Current @ r335314 not bootable with Geli and ZFS

3 views
Skip to first unread message

Thomas Laus

unread,
Jun 18, 2018, 1:39:37 PM6/18/18
to
Something changed in /boot/gptzfsboot between r334610 and r335314. I
built current this morning and my system is un-bootable. I am using
redundant ZFS disks and only copied the updated /boot/gptzfsboot file to
my ada0 drive. I was able to boot the ada1 drive that still had the
gptzfsboot file from r334610.

I had a similar issue a few months ago with the upgrades to the Geli +
ZFS booting process. These were resolved and operation has been fine
since the last 'hick-up' in the testing process. I might not be the
only person running the combination of Geli encryption and using a ZFS
filesystem, but it should not be that much uncommon setup that I am the
first to report the problem.

Let me know far back I need to revert my sources to identify the commit
that broke gptzfsboot. My system goes into a continuous reboot loop
before presenting the password prompt. It is very early in the startup
process.

Tom

--
Public Keys:
PGP KeyID = 0x5F22FDC1
GnuPG KeyID = 0x620836CF
_______________________________________________
freebsd...@freebsd.org mailing list
https://lists.freebsd.org/mailman/listinfo/freebsd-current
To unsubscribe, send any mail to "freebsd-curre...@freebsd.org"

Allan Jude

unread,
Jun 19, 2018, 12:14:31 PM6/19/18
to
On 2018-06-18 12:42, Thomas Laus wrote:
> Something changed in /boot/gptzfsboot between r334610 and r335314. I
> built current this morning and my system is un-bootable. I am using
> redundant ZFS disks and only copied the updated /boot/gptzfsboot file to
> my ada0 drive. I was able to boot the ada1 drive that still had the
> gptzfsboot file from r334610.
>
> I had a similar issue a few months ago with the upgrades to the Geli +
> ZFS booting process. These were resolved and operation has been fine
> since the last 'hick-up' in the testing process. I might not be the
> only person running the combination of Geli encryption and using a ZFS
> filesystem, but it should not be that much uncommon setup that I am the
> first to report the problem.
>
> Let me know far back I need to revert my sources to identify the commit
> that broke gptzfsboot. My system goes into a continuous reboot loop
> before presenting the password prompt. It is very early in the startup
> process.
>
> Tom
>

We tested all of the changes with the setup in tools/boot/rootgen.sh, it
will be interesting to figure out what went wrong with your setup, and
add it as a test case to prevent this in the future.

The recent changes are:

r335245 (reading the size of the disk)
r335254 (reading past the end of the disk)
r335276 (enable the serial console sooner so the password prompt can be
used over serial)

There is also one outstanding fix: https://reviews.freebsd.org/D15847

--
Allan Jude

Thomas Laus

unread,
Jun 19, 2018, 2:27:14 PM6/19/18
to
On 06/19/18 12:09, Allan Jude wrote:
>
> We tested all of the changes with the setup in tools/boot/rootgen.sh, it
> will be interesting to figure out what went wrong with your setup, and
> add it as a test case to prevent this in the future.
>
> The recent changes are:
>
> r335245 (reading the size of the disk)
> r335254 (reading past the end of the disk)
> r335276 (enable the serial console sooner so the password prompt can be
> used over serial)
>
> There is also one outstanding fix: https://reviews.freebsd.org/D15847
>
I don't think that my issue is related to the fix described in the
review. I am using ~256G SSD's. They are slightly different sizes but
are both >200M. I will try backing out the commits starting with
r335276 and work backward from there. I will review the changes made to
gptzfsboot to get the date that it was last touched. When I replace
only gptzfsboot with one made 2 weeks ago (r334610) everything boots OK
and I get a password prompt for the geli password. If the gptzfs
bootcode after r335314 is copied to the boot record, the computer goes
into a continuous reboot loop and only a part of the password prompt is
shown before the reboot.

The filesize on the bad gptzfsboot file is 121922 bytes and the good one
is 121634 bytes. The filesize on that file has not changed in a few
months. I keep old versions of this file since imp fixed things for me
a few months ago when he made changes for loader code migration from
Forth that caused a similar issue.

Tom


--
Public Keys:
PGP KeyID = 0x5F22FDC1
GnuPG KeyID = 0x620836CF

Thomas Laus

unread,
Jun 19, 2018, 2:55:26 PM6/19/18
to
On 06/19/18 14:22, Thomas Laus wrote:
> On 06/19/18 12:09, Allan Jude wrote:
>>
>> We tested all of the changes with the setup in tools/boot/rootgen.sh, it
>> will be interesting to figure out what went wrong with your setup, and
>> add it as a test case to prevent this in the future.
>>
>> The recent changes are:
>>
>> r335245 (reading the size of the disk)
>> r335254 (reading past the end of the disk)
>> r335276 (enable the serial console sooner so the password prompt can be
>> used over serial)
>>
Allan:

It looks like the problem is with r335276. I reverted back to that
revision and built stand, my boot problem was still there. I reverted
further back to r335254 and everything worked fine.

Toomas Soome

unread,
Jun 19, 2018, 5:18:32 PM6/19/18
to

> On 19 Jun 2018, at 21:51, Thomas Laus <lau...@acm.org> wrote:
>
> On 06/19/18 14:22, Thomas Laus wrote:
>> On 06/19/18 12:09, Allan Jude wrote:
>>>
>>> We tested all of the changes with the setup in tools/boot/rootgen.sh, it
>>> will be interesting to figure out what went wrong with your setup, and
>>> add it as a test case to prevent this in the future.
>>>
>>> The recent changes are:
>>>
>>> r335245 (reading the size of the disk)
>>> r335254 (reading past the end of the disk)
>>> r335276 (enable the serial console sooner so the password prompt can be
>>> used over serial)
>>>
> Allan:
>
> It looks like the problem is with r335276. I reverted back to that
> revision and built stand, my boot problem was still there. I reverted
> further back to r335254 and everything worked fine.
>

ou, in illumos side there was an idea of having early boot printouts mirrored to serial, but it got turned down because there is no way to tell if any system has some bad side effects from it… (either directly or indirectly by having some weird device connected to the serial).

rgds,
toomas

0 new messages