Quibble – Custom Windows Bootloader(github.com)
One use case that would be amazing is actually be able to get 64bit Windows on those 64bit atoms with no 64bit bootloaders :(
Neat. ReactOS has a similar piece of code called "freeldr".
I wonder if any of this code might be of use to ReactOS. Even just to extend its bootloader to cover other Windows versions, if it can't already.
If so it goes both ways, because some files have referances to ReactOS. Like peloaddef.h
And the configuration file is freeldr.ini
One interesting use for this could be signing it with your own secure boot keys. This could let you lock a machine to only boot your system image and not others.
Standard Windows bootloader already does that, it's part of supported deployment flow in highest security setting - in fact part of why MS requires that computers with certain Windows-related branding (the part about stickers "designed for windows" etc., I just don't remember which one) have to enable end users to replace platform keys.
Except the same keys have been used to sign some EFI loaders from other vendors, namely Kaspersky Rescue Disk 18, so it isn't really as locked down as they would have us believe.
There are two keys distributed by MS - one for Windows, one for third party applications. The latter one is used by Linux shim bootloader and the like. MS also requires that all devices that are supposed to be "full featured" support replacing all keys and loading user keys, and the highest security deployment setup for Windows Enterprise requires that you sign windows bootloader (or, more likely, sign the hash of the specific binary you are using) yourself.
> the highest security deployment setup for Windows Enterprise requires that you sign windows bootloader (or, more likely, sign the hash of the specific binary you are using) yourself.
To my knowledge that's not possible with the Microsoft bootloader, you need to use Microsoft's keys, hence why I am suggesting that this open source bootloader could be useful. Can you provide some more information about such a setup?
This document is aimed at OEMs, but they specifically say that high-security environments can generate Secure Boot keys and use them on their devices in conjunction with their own Windows image. (I’d imagine that you need the Windows SDK and some know how/MS consulting to get it working, though)
This document is about generating the platform key, but changing the platform key alone doesn't allow you to do what I'm suggesting. Even after setting your own platform key you would still need to trust the key for Microsoft's bootloader. See section 1.4.1:
> The Microsoft Windows Production PCA 2011 with a SHA-1 Cert Hash of 58 0a 6f 4c c4 e4 b6 69 b9 eb dc 1b 2b 3e 08 7b 80 d0 67 8d must be included in db in order to allow the Windows OS Loader to load.
For OEM to allow reinstall done "normally", yes.
But DB entries can also contain SHA-256 hashes of the image to load (with the image stripped of the signature, which btw allows you to also resign it).
> which btw allows you to also resign it
Are you sure? I havent tried, but this disagrees: https://docs.microsoft.com/en-us/previous-versions/windows/i...
"Windows boot components: BootMgr, WinLoad, Windows Kernel Startup. Windows boot components verify the signature on each component. Any non-trusted components will not be loaded and instead will trigger Secure Boot remediation. "
> DB entries can also contain SHA-256 hashes
OK, fair enough, but it still doesn't really solve the problem because an attacker could just copy your modified bootmgr to be able to steal and use your workstation. In order for this to work you'd also have to add some kind of additional checks which we can now do with an open source version of the bootloader.
The signatures are, IIRC, checked in order, so a resigned BootMgr won't impact verification of WinLoad, etc.
As for abusing signed bootloader, the full process depends on verifying that the signed bootloader appropriately handles TPM, which is then also coupled with an encrypted disk, which in turn works to prevent loading unsanctioned code.
Of course, it's possible to defeat this, but the idea is to frustrate the efforts and raise the cost of an attack, as well as give you a longer timeframe to deal with an attack.
Those mechanisms only prevent an attacker from decrypting the disk. The scenario I'm envisioning is where the attacker steals your workstation, wipes the disk and reinstalls Windows.
That's not part of the threat model - the encrypted data is lost, the point of the defences are to prevent damage to things accessible from that specific machine, not to prevent stealing the machine.
The basic procedure is as follows:
- Remove all keys (switch to Setup mode) - Setup your own PKI and platform keys - Sign hash of specific EFI files and load those signatures into EFI
The last part doesn't require modifying files themselves, as you're locking specific files. The firmware will make a hash of the file, and verify that it's on permitted list (the list is signed)
If I'm not mistaken it's actually a hash of the file's signature that you have to register, not a hash of the file itself. And Bootmgr will not allow you to change its signature. Thus any computer that is configured to boot Windows with secure boot will be able to boot any copy of Windows.
Not if you replace the keys with your own.
Sure, and I do that, but most users will not, they will select Secure Boot and think that their boot is now secure.
That's super cool. I would be much more likely to keep windows around if it could boot from a btrfs volume.
I thought about that too, but looking through the README, there are some considerable complications (and funny bugs). The author has tested this on a lot of Windows versions!
I feel like this was a toy/thought experiment that turned into learning a lot about Windows, EFI, bootstraping, etc.
It looks like windows can't copy some in-use files, and Linux can't copy some attributes.
Why not copy as many files as possible with windows, and then go to Linux to copy the last few?
Most of funny bugs arise from not-really-exact filesystem copying during install. But making a perfect duplication tool seems to be out of scope of this project.
Nice experiment, anyway.
Issues would be solved as soon as you'd be able to install a WIM image directly, which should actually be possible.
I thought about that as well, can't we just load the WinBtrfs driver from the installer?
This looks like an awesome project; if the author is here: well done!
How do you prevent Windows Update from replacing it?