Boot Windows? about btrfs HOT 59 CLOSED

Hello from ReactOS team!

We are looking for ways of using BTRFS as main FS in ReactOS. At the moment we already use WinBTRFS driver 1.0. We want to make possible booting of ReactOS from BTRFS partition.

from btrfs.

I've got it working on Windows 7 now as well. I'll submit the patches once I've cleaned them up.

from btrfs.

If anybody's interested, I've got this working with Vista now as well. I'll be submitting some patches to Freeloader shortly.

It's my understanding that Vista and Windows 7 aren't all that dissimilar internally, so hopefully getting this working with modern versions of Windows isn't that far away...

from btrfs.

If anybody's interested: https://github.com/maharmstone/quibble

from btrfs.

An update: I got fed up of trying to hack freeldr into shape, so I've been working on a new EFI bootloader, which I'll be releasing once it's done. Here's it running Windows 8 off a Btrfs partition:

from btrfs.

I've created a ReactOS pull request, which fixes the bug. Here's a binary with the fix applied: freeldr.zip.

With the modified version of freeldr and the latest btrfs patches, booting Windows Server 2003 from btrfs now seems to be stable.

from btrfs.

I'm happy to tell you that I successfully managed to boot Windows using FreeLoader (ReactOS boot loader) and the WinBtrfs' ReactOS port.

The video: https://youtu.be/zZ5JOwVbyWE

It's on drive D: due to drive lettering issues of some sort (happens on Windows 2003 only). Also, a random crash might occur ocassionaly while using the system (not sure wether it is FreeLDR or Btrfs driver fault).

from btrfs.

New update (29 July 2018): https://reactos.org/blogs/gsoc-2018-booting-btrfs-works

from btrfs.

Awesome! Thank you for sharing this.

from btrfs.

Here is the final GSoC blog update from our GSoC student (with a demo video linked): https://reactos.org/blogs/gsoc-2018-final-report
The demo build also includes the two PRs #102 and #103.

from btrfs.

In fact, Microsoft introduced 32-bit UEFI boot support as early as the beta version of Windows Vista. Of course, versions earlier than Windows 8 still require CSM support by default when UEFI boot.
The first Windows system build to support UEFI-IA32 boot was 6.0.5219, but this feature was removed by Microsoft after building 6.0.5477.

Until the 32-bit UEFI boot appeared again at build 6.1.7850 and has been retained to this day, the build is also the first Windows system to support exFAT boot.

from btrfs.

Latest update (14 July 2018): we have a VBR to boot directly into a btrfs volume, that is able to load our freeldr.sys : https://reactos.org/blogs/gsoc-2018-boot-sector-finished
The next step is to implement small btrfs filesystem read-only support in freeldr for it to be able to load arbitrary files from arbitrary directories. Currently work-in-progress.

from btrfs.

Thanks for the link, looks like it might be useful. There's symbols available for bootmgfw.efi, which say that there's functions called NtfsMount etc. It might be possible to hook the FS functions or somesuch, I don't know. As I said, it's a long-term aim, and I'm going to look at it properly once the driver itself is feature-complete.

from btrfs.

Thanks for letting me know, that sounds interesting. I'm intrigued to see how you get on!

from btrfs.

@ale5000-git - 32 so far - I'm hoping 64-bit support will be a relatively trivial addition...

@HBelusca - yes, I know: it's a mess and I'm not touching it. It looks very much like it comes from the run-it-through-IDA-Pro-and-clean-it-up school of reverse engineering. It copies, or appears to aim to copy, all of Microsoft's bad design decisions: the bootmgr/winload split, the inclusion of an XML parser, the inclusion of a FAT driver despite this being guaranteed by the UEFI specs...

from btrfs.

An update: I got fed up of trying to hack freeldr into shape, so I've been working on a new EFI bootloader, which I'll be releasing once it's done. Here's it running Windows 8 off a Btrfs partition:

Great, because there is one waiting for it to happen as well: https://git.reactos.org/?p=reactos.git;a=tree;f=boot/environ/app;h=504c60baef7c77f09ec9ae42f710da3496f95020;hb=HEAD xD

from btrfs.

@ale5000-git - I think that just means that Microsoft haven't written a 32-bit UEFI bootloader for Windows 7. As far as the kernel is concerned, Windows 8 doesn't require CSM, everything before it does.

from btrfs.

I was wondering if the "Have disk" button on the installer would let you load the driver - though it'd still break when it came to the first reboot.

Yeah, and I don't think this option would work for FS drivers in setup.

I'm surprised it works at all on Windows 2003, actually - I thought the minimum supported version was Windows 7. I don't know if that's behind your crashes.

I'm using a ReactOS version, which has lowered requirements to (at least) Windows 2003. Although ReactOS still really targets Windows 2003, and doesn't expose any NT 6+ functionality AFAIK. The driver itself works fine in ReactOS as well.

from btrfs.

I've spent some time looking into this, and I'm now fairly sure it's a bug in Freeloader...

When CmpInitializeSystemHive finishes, only the first sector is marked as dirty, rather than all the cells for which the bootloader has removed the volatile data. This means it ends up discarding dirty registry without it having been flushed, and later on HvpGetCellMapped ends up reading unsanitized data from the disk, and crashing when it tries to read an invalid pointer.

from btrfs.

It looks very much like it comes from the run-it-through-IDA-Pro-and-clean-it-up school of reverse engineering.

Blame @ionescu007 😅

32 so far - I'm hoping 64-bit support will be a relatively trivial addition...

Would be interesting to see it for Windows 10 ARM64 as well.

from btrfs.

@maharmstone: You can't trust the specs, the first PCs are EFI, not UEFI and may not comply completely with the current specs.

I suggest to NOT expect anything being guaranteed.

from btrfs.

It's a long-term aim. To get the whole thing on Btrfs, I'd have to do a new version of bootmgfw.efi, and I'm not even sure if you can do this without it being signed by Microsoft. Another possibility might be having a tiny NTFS partition with just a few files, and everything else on Btrfs.

from btrfs.

Thank you for clarifying.
AFAIK EFI has the concept of filesystem drivers (as separate .efi binaries loaded from another bootloader or EFI shell), and rEFInd already has btrfs driver, so maybe bootmgrfw can be patched to read btrfs if something loads the driver. Everything with disabled SecureBoot, of course.

Your idea of having tiny NTFS partition is interesting, but i am not sure how to achieve this. Symlinking everything from NTFS partition to btrfs one? I already did that trick with two NTFS drives and moved /ProgramData, /Program Files and some non-critical /Windows subfolders from SSD to HDD and it worked. Can you tell at what moment of booting windows loads filesystem drivers? Surely it loads some other drivers before, so i doubt if symlinking whole /Windows will work.

Anyway, i'll try to dissect and symlink as much folders as i can this weekend..

from btrfs.

To be honest, for all I know booting via rEFInd might be all that's required. If you load an EFI filesystem driver before the Windows bootloader, will Windows use it or ignore it? It's something I'll have to research.

Presumably the boot process goes something like this:

• UEFI runs bootmgr.efi
• bootmgr.efi uses functions in bootmgfw.efi to load ntoskrnl.exe
• bootmgfw.efi then gets used to load system.reg, which amongst other things lists the services needed to boot
• Either bootmgr.efi or ntoskrnl.exe then walks the tree looking for services where the start type is 0 ("Boot"), and loads the images into memory
• Windows then replaces bootmgfw.efi's NTFS commands with the proper ones contained in ntfs.sys, which it has just loaded
• We then proceed with the normal boot

from btrfs.

Everything with disabled SecureBoot, of course.

It's not like there isn't the handy Linux Foundation PreLoader that can sign whatever you want.

I mean, you could even modify re-sign bootmgfw after you have modified it (but is it even legal?).

To be honest, for all I know booting via rEFInd might be all that's required. If you load an EFI filesystem driver before the Windows bootloader, will Windows use it or ignore it? It's something I'll have to research.

I think it depends more on whether windows bootloader is more like GRUB (which loads its own set of drivers) or more like systemd-boot (which handle everything as simply as possible).
Given its functioning and how it can load winload.exe which is the system ntfs partition, I'd go with the former.

So either somebody manages to adds into it btrfs efi driver, or I guess the tiny ntfs partition idea is better.

from btrfs.

it'd be better to focus on making dism work, currently dies at a call to SetFileSize

from btrfs.

https://jira.reactos.org/browse/CORE-13769

from btrfs.

This discussion was mainly about UEFI boot, but has anyone tried to boot Windows 7 in legacy mode (not UEFI) from a btrfs partition?

from btrfs.

Any updates on this?

from btrfs.

from btrfs.

Awesome! Did you install it directly to Btrfs, or copy an installation from
NTFS?

I have installed Windows 2003 on FAT32 volume, then the WinBtrfs driver from ReactOS and finally moved the Windows installation to the Btrfs volume (which already had FreeLoader installed). Installing directly onto Btrfs might not be easy because Windows 1st stage installer doesn't support it.

from btrfs.

I was wondering if the "Have disk" button on the installer would let you load the driver - though it'd still break when it came to the first reboot.

I'm surprised it works at all on Windows 2003, actually - I thought the minimum supported version was Windows 7. I don't know if that's behind your crashes.

from btrfs.

Quote:

I've spent some time looking into this, and I'm now fairly sure it's a bug in Freeloader...

When CmpInitializeSystemHive finishes, only the first sector is marked as dirty, rather than all the cells for which the bootloader has removed the volatile data. This means it ends up discarding dirty registry without it having been flushed, and later on HvpGetCellMapped ends up reading unsanitized data from the disk, and crashing when it tries to read an invalid pointer.

@SergeGautherie
@Extravert-ir
It looks like a serious problem.

from btrfs.

Very nice! It truly is great to see FreeLDR being capable of booting Vista+... And Vista being booted from Btrfs as well. By the way, if you're not aware already, ReactOS also has an (incomplete) UEFI bootloader, it might be interesting when it comes to UEFI booting Windows from Btrfs.

from btrfs.

If anybody's interested, I've got this working with Vista now as well. I'll be submitting some patches to Freeloader shortly.

It's my understanding that Vista and Windows 7 aren't all that dissimilar internally, so hopefully getting this working with modern versions of Windows isn't that far away...

Is that a 32-bit or 64-bit version of Windows Vista?

from btrfs.

from btrfs.

That would explain why I couldn't boot into it. I was trying to boot a 64-bit version of Vista (which Freeldr doesn't support yet). By the way, can you please provide a compiled driver that works in that architecture of Vista? I got one to compile but it wouldn't load (it gave me a 31 error code).

from btrfs.

I was able to get Windows Vista x86 to load a driver compiled from the latest source. When I try to boot it via Freeldr off of a BTRFS drive, it gives me a STOP code of 0x000000A5. I am running Windows Vista x86 via a VirtualBox VM.

EDIT: Here's some debug output that I got by using Windbg to show what was being outputted via Vista's debug mode:

windbg output of ACPI crash - Vista via Freeldr on BTRFS drive.txt

from btrfs.

Is that with no changes to the registry? If so, I wonder why booting Windows Vista via Freeldr and BTRFS doesn't work on VMWare/VirtualBox but works on QEMU.

from btrfs.

I had to change the start type of tcpip, as you mentioned in the other thread - that's another issue I need to look into.

from btrfs.

Now I am getting a different crash when I try running Windows Vista via FreeLDR + BTRFS via WMWare Workstation Player 15 with only the start type of TCPIP changed. I am attaching the debug log to see if it can help in fixing the crash.

btrfs-vista-bsod.txt

from btrfs.

I have another crash after compiling it via ROSBE instead of Visual Studio 2019. The debug log is attached.

btrfs-vista-bsod-2.txt

from btrfs.

I managed to get it to work somewhat by using Fastcopy to preserve the ACLs of the files and directories when copying from a VHD that has a working copy of Vista on a NTFS FS to a VHD that has a BTRFS FS with the Freeldr files and bootsector applied. When I try to boot from the BTRFS VHD, it gets stuck on a blue background (meaning it can't load explorer.exe) until I change the group entry of BTRFS's key in the registry to include Boot and change the start dword from 1 to 0. Then it loads the taskbar and two Notepad windows with a desktop file showing its contents in both of them.

from btrfs.

@maharmstone How did you copy the files from the Vista image to the BTRFS image that has freeldr.sys and freeldr.ini on it?

from btrfs.

Can anyone do a tutorial to do it?

from btrfs.

I agree with the wanting of a tutorial on how to get Windows 2003 and Vista (and 7 when the patches are ready to apply) to boot when on a BTRFS drive.

from btrfs.

@maharmstone: Good work :)

Does it works on both 32 and 64-bit or only 32-bit?

from btrfs.

Also all of this need to be tested, before or later, on macOS (since it can install Windows), and very weird things can be expected from macOS firmware.

from btrfs.

@maharmstone Would the EFI bootloader allow the booting of earlier Windows versions that are on a BTRFS partition?

from btrfs.

You can't trust the specs, the first PCs are EFI, not UEFI and may not comply completely with the current specs.

To be fair, FAT32 is there supported from the get go.
On the other hand bugs happen. And I guess it has all its sense for microsoft not to trust firmware.

from btrfs.

@maharmstone Would the EFI bootloader allow the booting of earlier Windows versions that are on a BTRFS partition?

I expect older versions of Windows (pre-Vista) would require a CSM to function

from btrfs.

Actually, everything up to Windows 7 requires CSM (well, at least without hacks)
Contrarily to what most people think the compatibility support module doesn't automatically exclude UEFI.

A nice feature I'd like to see is instead, a bios bootloader that didn't require external disks to boot from GPT (but I digress) EDIT: cool

from btrfs.

Actually, everything up to Windows 7 requires CSM (well, at least without hacks)
Contrarily to what most people think the compatibility support module doesn't automatically exclude UEFI.

A nice feature I'd like to see is instead, is a bios bootloader that didn't require external disks to boot from GPT (but I digress)

Ah, I wasn't aware of that - thanks for some fascinating reads, too :)

from btrfs.

@mirh: If I'm not wrong 32-bit Windows 7 support only legacy mode but 64-bit Windows 7 support both legacy/UEFI mode.

from btrfs.

Booting via FreeLDR is easier in my opinion. All a user needs to do is get it it installed on a BTRFS partition somehow and then copy over the Windows installation.

from btrfs.

Did anybody try to create a Btrfs partition from GNU/Linux, wimapply the image from the install disc, install the bootloader and try to boot that?

Also, I think that Quibble is very nice, but computers with full and correct UEFI support appeared only several years ago and there are still many BIOS only machines with good specs running Windows 10. It would also be very nice to use Btrfs on the system partition on those machines.

from btrfs.

As you know, there is a completely different bootloader for BIOS systems
maharmstone/quibble#17
reactos/reactos#1905

from btrfs.

Hey, I want to boot windows from Btrfs subvolumes. I am tring to boot from quibble efi. But got error EFI_NOT_FOUND. I tried every config which I can think about:
`[FREELOADER]
TimeOut=10

[Operating Systems]
Windows_7="Windows 7"
Windows_8="Windows 8.1"
Windows_10="Windows 10"
Windows_11="Windows 11"
Windows_7_PATH="Windows 7"
Windows_8_PATH="Windows 8.1"
Windows_10_PATH="Windows 10"
Windows_11_PATH="Windows 11"
Windows_7_PATH2="Windows 7"
Windows_8_PATH2="Windows 8.1"
Windows_10_PATH2="Windows 10"
Windows_11_PATH2="Windows 11"

[Windows_7]
SystemPath=btrfs(756dadf8-2d04-4f43-bbc3-eb38a2fffb13)\Windows
Options=/SUBVOL=win7

[Windows_8]
SystemPath=btrfs(756dadf8-2d04-4f43-bbc3-eb38a2fffb13)\Windows
Options=/SUBVOL=win8

[Windows_10]
SystemPath=btrfs(756dadf8-2d04-4f43-bbc3-eb38a2fffb13)\Windows
Options=/SUBVOL=win10

[Windows_11]
SystemPath=btrfs(756dadf8-2d04-4f43-bbc3-eb38a2fffb13)\Windows
Options=/SUBVOL=win11

[Windows_7_PATH]
SystemPath=btrfs(756dadf8-2d04-4f43-bbc3-eb38a2fffb13)\win7\Windows
Options=/SUBVOL=win7

[Windows_8_PATH]
SystemPath=btrfs(756dadf8-2d04-4f43-bbc3-eb38a2fffb13)\win8\Windows
Options=/SUBVOL=win8

[Windows_10_PATH]
SystemPath=btrfs(756dadf8-2d04-4f43-bbc3-eb38a2fffb13)\win10\Windows
Options=/SUBVOL=win10

[Windows_11_PATH]
SystemPath=btrfs(756dadf8-2d04-4f43-bbc3-eb38a2fffb13)\win11\Windows
Options=/SUBVOL=win11

[Windows_7_PATH2]
SystemPath=btrfs(756dadf8-2d04-4f43-bbc3-eb38a2fffb13)\win7\Windows

[Windows_8_PATH2]
SystemPath=btrfs(756dadf8-2d04-4f43-bbc3-eb38a2fffb13)\win8\Windows

[Windows_10_PATH2]
SystemPath=btrfs(756dadf8-2d04-4f43-bbc3-eb38a2fffb13)\win10\Windows

[Windows_11_PATH2]
SystemPath=btrfs(756dadf8-2d04-4f43-bbc3-eb38a2fffb13)\win11\Windows`

Please tell me How I can approach booting from btrfs subvolume.

from btrfs.

Closing old issues

from btrfs.