Posts made by Flyer

@pauleb Intriguing. So you didn’t need to create a EFI entry for the drive? I wonder if it was already there for that drive from the previous image or something.

I guess mine did not work because I first imaged my test machine with an MBR image, which made the EFI entry no longer show up.

Good luck, and thanks for the info.

@pauleb I recently went through this and used the script I posted in this thread to create the EFI entry for the windows boot.

The Image must be in GPT format, or at least I had no luck with MBR…

I never got to the Windows boot loader at all with CSM disabled until I used GPT and the script I posted at the link above. My machines would just hang at a black screen, and the drive was not seen in the EFI BIOS until I set the EFI entry as described using efibootmgr in that script. Though I have to say I am not sure I ever tried a GPT image without the script.

I made my image in VirtualBox set to EFI, though I had to turn off EFI to get Virtual Box to network boot and not fire up the golden image, ruining it due to it already being sysprepped, but that is a different story.

@george1421 Great stuff. Thanks!

@george1421 Thanks for the reply. I had seen a lot of your (and others’) scripts. It does look like there are methods to do it from the script, no doubt.

It just seems like the scripts are generally either host or image specific, which is why I was thinking it would be good to link those in the DB for easy management. Perhaps they could be applied to groups or some other collection of hosts/images such that they could applied to many hundreds or thousands of hosts at once. I have not looked at groups; so I have no clue what functionality they offer at this point. I do understand it may be more trouble than it’s worth, since you can do it right now using the methods in the script.

As a side note, is there a way to drop into a shell in the OS some time between when the image is written and reboot? I guess the postdownload script could just invoke a shell? Would be useful for debugging without have to re-image repeatedly.

@Sebastian-Roth OK on the monolithic kernel vs. module loading. That explains the modprobe error I received (I had tried adding a manual call to modprobe in my script).

It turns out the issue must have been with that particular motherboard/BIOS. It is a much older machine (for anyone searching it was an ASUS Sabertooth X79 with BIOS version 3501). But there may have been other issues as well. I have yet to go back and try my now-seemingly-working script on that machine.

Below is my postdownload script I am now using to update the UEFI boot records for some newer machines (7950X on Gigabyte x670 boards). So far this is working well. I need to test it with an existing boot record with the same name (which is the default Windows name). I know efibootmgr will throw a warning in this case. It took me a bit to realize that I had my Windows image in MBR format, and needed to convert it to GPT (web search for mbr2gpt).

#!/bin/bash

#
# Note that the Windows disk needs to be in GPT format, not MBR or this
# will appear to succeed, but still not boot.
#
# Public Domain - Modify and use at will. No Warranty.
#

. /usr/share/fog/lib/funcs.sh
[[ -z $postdownpath ]] && postdownpath="/images/postdownloadscripts/"

case $osid in
    5|6|7|9)
        clear
        echo "OS Type is Windows"

        getHardDisk

        if [[ -z $hd ]]; then
            handleError "Could not find hdd to use for EFI Boot Entry"
        fi
        echo "Found Install Disk: $hd"

        dots "Creating Win EFI Boot Entry for Disk: $hd Part: 1 using efibootmgr"

        #Windows UEFI partition is usually partition 1
        efibootmgr -c -L "Windows Boot Manager" -l "\\EFI\\Microsoft\\Boot\\bootmgfw.efi" -d $hd -p 1

        if [[ ! $? -eq 0 ]]; then
                echo "Failed. Error: $?"

                # We opt not to fail here, as some machines may not support EFI Boot entries.
                #debugPause
                #handleError "Failed to Create Windows EFI Boot Entry. Error: $?"

                echo " *** Failed to Create Windows EFI Boot Entry. Error: $? *** (pause 15sec)"
                sleep 15
        fi

        echo "done"
        echo ""
        # Output the resulting boot records for reference/debug
        efibootmgr -v
        echo ""

        debugPause
        ;;
    *)
        echo "Non-Windows Deployment"
        debugPause
        return
        ;;
esac

I do think that the postdownload scripts should be per host or per image. A global post download script means you then need to check the image or host in the script, which is not a clean way to do it. Better if this is moved into the UI so there is a way to set post download scripts for Images and for Hosts. This way you could run separate scripts for either or both.

@Sebastian-Roth When I try to run efibootmgr from the postdownload script I am getting modprobe telling me it can’t find the efivars kernel module. Any idea as to the best way to get that where it needs to be, and more importantly where does it need to be?

I am curious how @ClementK was able to run it.

Anyone?

Edit: Perhaps I don’t need to run modprobe, and the motherboard truly does not support efi vars, which is the error I am getting from efibootmgr without first calling modprobe.

Errno 2 “EFI variables are not supported on this system.”

@george1421 Thank you for the reply. I meant that I was using ipxe.efi (Not my mistyped ipxe.pxe above) for the newer machines. I selected that based on the info on this page, since I do not use the built in DHCP server:

https://docs.fogproject.org/en/latest/installation/network-setup/

Perhaps those docs are out of date?

So it sounds like I should configure the UEFI machines to use snp.efi/snponly.efi
(and fall back only if I have problems)

??

Thanks.

Is there some good documentation on which pxe file to use? I just realized from another post that there are some other vendor specific pxe files like intel.kpxe, and a realtek one, etc.

I have been trying to use ipxe.pxe, and when that fails (typically says out of memory on older mobos) the I use undionly.pxe . (The basic intro docs I found only listed those two.) However, honestly, I don’t fully understand the difference, or why I would want one over the other, or if I should be using an Intel pxe file with Intel NICs, etc.

My DHCP server let’s me set the pxe file for each MAC address individually if needed.

I am guessing this is really a case of if it works it’s fine, but I wondered if there was any reason to pick a particular pxe file, other than the fact that it boots OK and exits OK.

Also many motherboards seem to offer the UEFI option to net boot (Which usually works with the ipxe.pxe) as well as a older op ROM method, like with an Intel card. Any reason to use one over the other if they both work for a given setup? (I know some mobos only support the legacy ROM booting in CSM mode).

As an aside, but related note for anyone wishing to learn more I found this YouTube video very helpful in explaining UEFI for legacy ROM in sufficient detail to really know what was going on:

https://www.youtube.com/watch?v=G_qKrJPuAmg&t=503s

Thanks.

@Sebastian-Roth If I set that key correctly in the post install do you think the value will hold after a driver installation (via snapin)? I guess they only way to know is to test.

@Sebastian-Roth Thank you so much. I could not find that registry setting via searching. Maybe a DuckDuckGo thing, or perhaps user error. This is excellent! Thanks.

@Sebastian-Roth Thankfully the installer from Realtek is indeed silent. The only issue is that it of course kills network connectivity for a moment, and I see the Fog client looses its connection, but then retires about a minute and a half later and completes the updating of the server that the install completed.

The automatic driver install (like that which is built in to Windows… unless you are referring to some Sysprep specific option) installs a generic driver for both Intel and Realtek NICs. The Intel driver installed with automatic updates supports WOL, and it is enabled by default. The Realtek driver installed by Windows does not support WOL. Instead I have to install the Realtek driver (eoither from realtek or from the mobo manufacturer-- I go straight to the source form Realtek).

Luckily the Realtek driver does indeed support the completely silent option and it works via a snapin.

The only thing that I cannot figure how to get to work automatically (or remotely, like via a snapin) is the enabling of the “Only allow a magic packet to wake the computer” option. This default to OFF in Realtek’s driver, but ON (like it should) in Intel’s.

@Sebastian-Roth Oh, nice! I had not seen that. Still pretty new to the ecosystem. Trying to learn all I can.

Will check it out. Thanks.

@Sebastian-Roth Yes, after manually installing the Realtek NIC driver the WOL gets turned on. I can’t set the WOL with powercfg on the machines with the Realtek 2.5GbE NIcs until after I install the driver.

Of course I really need to find a way to set those to ONLY magic packet can wake because they default to any traffic, which results in a lot of waking.

It looks like on the Intel NICs I can set WOL to on via powercfg

So I am trying to find the “Foggy” way of handling the heterogeneous NIC setup. I can’t just put the drivers in the golden image because there are different NICs. (Or could I actually install both? Not sure the installers will even let me install in a VM.)

I was wondering about a Snapin. The Realtek drivers use an InstallShield EXE, which supposedly support command like install (haven’t tried it yet).

So installing the NIC driver manually seems to fix this. Any thoughts on the best way to install the proper driver for each machine? Would this be best handled with a SnapIn?

Thanks.

I am finding that none of my machine will respond to Wake On Lan after being imaged. I thought this was the default.

I can’t imaging I somehow turned this off in the BIOS of all the machines.

They are running Windows. Perhaps I missed something in the Golden Image. Any thoughts? The NICs are a bit heterogeneous. Some Intel, some of the newer Realtek 2.5Gig. So I am not installing any drivers, but rather letting Windows 10 find the drivers (which normally works fine with fresh manual installs).

Thanks.

I believe this lack of a EUFI entry is the same reason why I am finding that I need to enable CSM booting after I image Windows.

Is there a way to set up this script so that it always uses the correct device? For instance, some machines are NVME, some SSD, some rust, etc. So they would have different device names and partition numbers.

On one motherboard I just setup FOG on I couldn’t even see the nvme at all as a boot option in the BIOS until I enabled the CSM boot.

Thanks.

With the bad RAM removed the system imaged just fine with the same image. Thanks again.

Thanks for the feedback, everyone.

You are absolutely correct! The new nvme had no effect

Memtest revealed a bad ram stick. Going to RMA that.

Thank you. Now I will try the imaging again after memtest passes a few times on the remaining RAM.

On one machine we are having the image constantly fail.

However, I just sued that image on another machine (and did it again to test). I have another nvme coming overnight. But this one passed the extended self test, but not sure that means much.

Strange that it says “read error” but I am guessing that was a read of a verification. But I have no clue.

I will report back if the new nvme fixes it. In the mean time any thoughts?

Thanks.

Thanks for the reply.

Initially I was creating the user with the unattend.xml, but I think I changed because of the mapped drives. When creating form scratch they weren’t there at all. Also, there are some BG applications (like sys tray stuff) that need to run for the user to make the rendering process work. So I configured these for that user on the golden image.

The users are not set to autologin (I’d like at least to have to enter a password to login, even if the same pwd for all).

My latest experiments led be to making a batch file that I could run once as the user, but I would still have to manually enter the credentials to save the user and pass for the shares. (The option to save credentials seems to be incompatible with specifying the credentials on the command line, which is never a great plan anyway.)

Maybe this is silly, and I should just store the credentials in the batch file, as either way, the user has access to the shares anyway (and the credentials are not shared with anything else). The batch file could simply be setup as a traditional startup item script, rather than involving the deployment process.

I guess I need some further testing. And I would eventually like to try to image a dual boot win/linux system that can be remotely (via script on win or linux) reboot to the other OS.