SG-1100 update from 23.09.1 to 24.03 keeps failing

mrneutron

I used the touch command to create files in the / directory and /var/log directories.
Both files were gone after a normal reboot.
I guess this is more evidence that the emmc is read-only, now?

I read that the mount command is supposed to show if the mounted partitions are ro or rw. It doesn't show ro or rw for any of them.

[23.09.1-RELEASE][admin@redacted]/: mount
pfSense/ROOT/default on / (zfs, local, noatime, nfsv4acls)
devfs on /dev (devfs)
pfSense/tmp on /tmp (zfs, local, noatime, nosuid, nfsv4acls)
pfSense/home on /home (zfs, local, noatime, nfsv4acls)
pfSense/var on /var (zfs, local, noatime, nfsv4acls)
pfSense/var/cache on /var/cache (zfs, local, noatime, noexec, nosuid, nfsv4acls)
pfSense/var/tmp on /var/tmp (zfs, local, noatime, nosuid, nfsv4acls)
pfSense/var/log on /var/log (zfs, local, noatime, noexec, nosuid, nfsv4acls)
pfSense/var/db on /var/db (zfs, local, noatime, noexec, nosuid, nfsv4acls)
pfSense/ROOT/default/cf on /cf (zfs, local, noatime, noexec, nosuid, nfsv4acls)
pfSense/ROOT/default/var_cache_pkg on /var/cache/pkg (zfs, local, noatime, noexec, nosuid, nfsv4acls)
pfSense/ROOT/default/var_db_pkg on /var/db/pkg (zfs, local, noatime, noexec, nosuid, nfsv4acls)
tmpfs on /var/run (tmpfs, local)
devfs on /var/dhcpd/dev (devfs)

stephenw10

I don't thing mount would show it. They are not mounted read-only. mount -p should show it though if they were.

mrneutron

Huh...all the Linux examples I saw of the mount command had only mount (no -p).
Is BSD usage different?

Using mount -p they are all supposedly mounted rw, but they don't hold what I write to them.

[23.09.1-RELEASE][admin@redacted]/: mount -p
pfSense/ROOT/default    /                       zfs     rw,noatime,nfsv4acls    0 0
devfs                   /dev                    devfs   rw              0 0
pfSense/tmp             /tmp                    zfs     rw,nosuid,noatime,nfsv4acls     0 0
pfSense/home            /home                   zfs     rw,noatime,nfsv4acls    0 0
pfSense/var             /var                    zfs     rw,noatime,nfsv4acls    0 0
pfSense/var/cache       /var/cache              zfs     rw,noexec,nosuid,noatime,nfsv4acls      0 0
pfSense/var/tmp         /var/tmp                zfs     rw,nosuid,noatime,nfsv4acls     0 0
pfSense/var/log         /var/log                zfs     rw,noexec,nosuid,noatime,nfsv4acls      0 0
pfSense/var/db          /var/db                 zfs     rw,noexec,nosuid,noatime,nfsv4acls      0 0
pfSense/ROOT/default/cf /cf                     zfs     rw,noexec,nosuid,noatime,nfsv4acls      0 0
pfSense/ROOT/default/var_cache_pkg /var/cache/pkg               zfs     rw,noexec,nosuid,noatime,nfsv4acls      0 0
pfSense/ROOT/default/var_db_pkg /var/db/pkg             zfs     rw,noexec,nosuid,noatime,nfsv4acls      0 0
tmpfs                   /var/run                tmpfs   rw              0 0
devfs                   /var/dhcpd/dev          devfs   rw              0 0

stephenw10

The -p switch cause it to print the output in fstab format and includes verbose info.
https://man.freebsd.org/cgi/man.cgi?query=mount

Yes so ZFS think it's all mounted read-write but that doesn't help if the actual storage is read-only.

mrneutron

Yeah, it sure looks like the emmc in my SG-1100 is no longer writable (upgradable).
It still boots in the configuration I had it set in, but I wonder how long before it goes non-functional?

"Certain types of disks, such as SSD and eMMC disks, may fail into a read only state where disk writes fail or are discarded, but data can still be read."

from: https://docs.netgate.com/pfsense/en/latest/troubleshooting/disk-lifetime.html

stephenw10

Yes unfortunately I have to agree. You can try installing to USB. I tested that yesterday.

mrneutron

@stephenw10 said in SG-1100 update from 23.09.1 to 24.03 keeps failing:

Yes unfortunately I have to agree. You can try installing to USB. I tested that yesterday.

You can run the pfsense OS off of a USB flash drive, plugged into the SG-1100?

stephenw10

Yes, the new Net Installer allows you select another USB drive as the target.

mrneutron

@stephenw10 said in SG-1100 update from 23.09.1 to 24.03 keeps failing:

Yes, the new Net Installer allows you select another USB drive as the target.

Can you use either USB port on the SG-1100? (like the USB3 port?)
Does loading into a USB flash drive automatically set the boot config to point to the USB flash drive as the boot drive?
If not, how do you manually edit the boot config?

stephenw10

Technically you could use either port and the USB3 port would likely be much faster. However when you run either usbrecovery or usbboot it boots the first USB device and if you have two drives inserted that is the USB3 port. Which means that when you boot the installer the target drive must be in the USB2 slot. That shouldn't matter, I'd expect to be able to move the drive after installing. However I was unable to do so. The 1100 can be picky about booting USB drives though so it could just be the drive I'm using.

The default boot device is not automatically changed. You need to set the bootcmd uboot env from the uboot prompt like:

setenv bootcmd 'run usbboot; run emmcboot;'
saveenv

Now in your case you may have an additional issue that the emmc is still holding a ZFS filesystem and cannot be wiped. If that is that case you'd need to either make sure the zfs pool names are different or use UFS on the USB drive.

mrneutron

@stephenw10 said in SG-1100 update from 23.09.1 to 24.03 keeps failing:

Technically you could use either port and the USB3 port would likely be much faster. However when you run either usbrecovery or usbboot it boots the first USB device and if you have two drives inserted that is the USB3 port. Which means that when you boot the installer the target drive must be in the USB2 slot. That shouldn't matter, I'd expect to be able to move the drive after installing. However I was unable to do so. The 1100 can be picky about booting USB drives though so it could just be the drive I'm using.

The default boot device is not automatically changed. You need to set the bootcmd uboot env from the uboot prompt like:
setenv bootcmd 'run usbboot; run emmcboot;'
saveenv
Now in your case you may have an additional issue that the emmc is still holding a ZFS filesystem and cannot be wiped. If that is that case you'd need to either make sure the zfs pool names are different or use UFS on the USB drive.

Yeah, given the read-only status of the emmc, it definitely appears the file system can't be wiped.
Using UFS on a USB flash drive seems like a less-complicated workaround - less other stuff to change.
How do you change the ZFS pool names, if you want/need to?

stephenw10

Yes, good question! There isn't an option to do that in the Net Installer TUI. Yet.

The legacy x86 installer allows setting the pool name. The legacy aarch64 installer never did because it used a prebuilt image.

So for now using UFS is the only option if you cannot remove a ZFS pool named 'pfSense' from the eMMC.

However I would argue that when running from USB the drive writes are even more important and running UFS with RAMdisks minimises that. So I would run UFS anyway.

Steve

flatsphere

@stephenw10 said in SG-1100 update from 23.09.1 to 24.03 keeps failing:

Yes, good question! There isn't an option to do that in the Net Installer TUI. Yet.

The legacy x86 installer allows setting the pool name. The legacy aarch64 installer never did because it used a prebuilt image.

So for now using UFS is the only option if you cannot remove a ZFS pool named 'pfSense' from the eMMC.

However I would argue that when running from USB the drive writes are even more important and running UFS with RAMdisks minimises that. So I would run UFS anyway.

Steve

Hi gang,

Is there a resolution to this problem yet? My SG-1100 is having the same problem.

Cheers!
Rafael

stephenw10

So you see the 8 disks shown by uboot?

flatsphere

@stephenw10 Hi Stephen,

The latest attempt shows that I'm running out of memory.
mount -p shoes this:

Following notices also appeared:

Upgrade
pkg (3): out of space @ 2024-05-15 09:06:30
PHP errors
PHP ERROR: Type: 1, File: /usr/local/pkg/pfblockerng/pfblockerng.inc, Line: 8804, Message: Uncaught TypeError: Unsupported operand types: int + string in /usr/local/pkg/pfblockerng/pfblockerng.inc:8804
Stack trace:
#0 /usr/local/www/pfblockerng/pfblockerng.php(628): sync_package_pfblockerng('cron')
#1 /usr/local/www/pfblockerng/pfblockerng.php(155): pfblockerng_sync_cron()
#2 {main}
thrown @ 2024-05-15 10:00:35

And following disk usage:

Cheers!
Rafael

SteveITS

@flatsphere given the 2.1 GB you probably have old boot environments to delete:
https://docs.netgate.com/pfsense/en/latest/troubleshooting/filesystem-shrink.html

Also if your RAM disk is full consider disabling that.

flatsphere

@SteveITS said in SG-1100 update from 23.09.1 to 24.03 keeps failing:

https://docs.netgate.com/pfsense/en/latest/troubleshooting/filesystem-shrink.html

I deleted all but the current boot environment, turned off VRAM and still failed.

R

stephenw10

Yeah you will either need bigger RAM disks or to just disable them for the update. I would disable them.

However this looks like a completely different error. It's not indicative of a failed eMMC. At least not what you've shown us so far.

flatsphere

@stephenw10 I just tried installing from USB. Here's the message:

R

stephenw10

Hmm, that was installing to the eMMC from the Net Installer?

Check the logs in /var/log/messages and /tmp/bsdinstall_log