Add support for installing to devices with multiple parents by ckyrouac · Pull Request #1068 · coreos/bootupd

ckyrouac · 2026-03-05T18:51:59Z

See individual commits for details. Generally, this migrates to using the new blockdev crate funtions and standardize on the Device struct operating on block devices. On top of this, handle the --filesystem and multiple --device flags to install the bootloader to multiple devices, e.g. a /dev/md0 RAID1 device backed by 3 physical disks, each with an ESP partition.

Keeping this as a draft because it will break older bootc versions prior to bootc-dev/bootc#2048.

Leaving the TEMP code in to validate CI passes (previously validated on my fork).

openshift-ci · 2026-03-05T18:52:11Z

Hi @ckyrouac. Thanks for your PR.

I'm waiting for a coreos member to verify that this patch is reasonable to test. If it is, they should reply with /ok-to-test on its own line. Until that is done, I will not automatically test new commits in this PR, but the usual testing commands by org members will still work.

Tip

We noticed you've done this a few times! Consider joining the org to skip this step and gain /lgtm and other bot rights. We recommend asking approvers on your previous PRs to sponsor you.

Once the patch is verified, the new status will be reflected by the ok-to-test label.

I understand the commands that are listed here.

Details

Instructions for interacting with me using PR comments are available here. If you have questions or suggestions related to my behavior, please file an issue against the kubernetes-sigs/prow repository.

gemini-code-assist

Code Review

This pull request is a significant and well-executed refactoring to support installing bootloaders to multiple devices, such as in a RAID setup. The migration to the bootc-internal-blockdev crate and the Device struct is consistently applied across the codebase. The new CLI options --device and --filesystem are sensible additions. The logic for handling multiple devices, especially for EFI where it filters for ESPs, is robust. I've also noticed some nice correctness improvements, like verifying that a VFAT partition is a true mount point. I have a couple of suggestions to fix a potential panic when dealing with relative paths, and one for code simplification.

src/efi.rs

src/bootupd.rs

cgwalters

I didn't do a deep review, only superficial. There's some nontrivial regression risk here - CI covers some basics, but far from everything.

src/efi.rs

.github/workflows/workflow-test.yml

ckyrouac · 2026-03-11T13:34:17Z

marking as ready for review but waiting on bootc 1.14.0 to release to remove the temp commit and for the arm/s390/ppc builds for the tests

cgwalters

Offhand this is looking plausible. This may touch on e.g. bootc-dev/bootc#1648 and similar issues, I hope we can try to see if we've fixed them with this.

src/bootupd.rs

cgwalters · 2026-03-11T19:27:39Z

src/bootupd.rs

    }
 }

+/// Get the block device backing the current root by trying `/boot` first,


I think it'd be cleaner to use /sysroot if it exists and is a mountpoint, otherwise use /.

In bootc/ostree we call this "physical root" - there's similar code there.

This pattern was copied from the removed blockdev.rs code. I'm not entirely sure why it was done this way originally, but I think your suggestion makes sense.

Consider that if boot and root partitions are on separate disks, see bootc-dev/bootc#1916, maybe /boot makes sense?

That's a very good point! Indeed...we should support that case. So I think things will work correctly if we list /boot first.

That said, there's a related thing here which is: we should honor XBOOTLDR from the DPS spec if it exists.

At the current time we aren't setting that up in the Fedora-derivative ecosystem much that I know of, but we should support it.

src/efi.rs

Cargo.toml

src/bootupd.rs

Align cap-std-ext with the version used by bootc-internal-blockdev, eliminating duplicate cap-std/cap-primitives dependency chains. Assisted-by: Claude Code (Opus 4) Signed-off-by: ckyrouac <ckyrouac@redhat.com>

cgwalters · 2026-03-17T15:17:14Z

FYI I put up #1072 which may textually (but not semantically) conflict with this

Signed-off-by: ckyrouac <ckyrouac@redhat.com>

Instead of referencing devices as a &str, use Device from the blockdev crate and its related functions. Signed-off-by: ckyrouac <ckyrouac@redhat.com>

This adds a new --filesystem flag that will use the blockdev crate to determine the parent devices of the root filesystem and install the bootloader to each device. The devices can also be explicitly specified via multiple --device flags. Signed-off-by: ckyrouac <ckyrouac@redhat.com>

When running install-to-filesystem, the host may already have the ESP mounted (e.g. /dev/sda15 at /boot/efi). Previously mount_esp_device would unconditionally call mount, which fails with "already mounted". Check if the mount point already contains a vfat filesystem on a different device than its parent (i.e. is a real mount point) before attempting to mount, and reuse the existing mount if so. Assisted-by: Claude Code (Opus 4) Signed-off-by: ckyrouac <ckyrouac@redhat.com>

Signed-off-by: ckyrouac <ckyrouac@redhat.com>

This uses the latest bootc RPM built from main in the CI tests. Assisted-by: Claude Code (Opus 4) Signed-off-by: ckyrouac <ckyrouac@redhat.com>

ckyrouac · 2026-03-18T12:30:49Z

updated to use the bootc rpm from copr and the 0.2.0 release of the blockdev crate, should be ready for another test run and review

cgwalters · 2026-03-19T13:15:26Z

/ok-to-test

ckyrouac · 2026-03-19T15:01:30Z

@cgwalters is there a way to trigger the tests now that /ok-to-test was applied?

cgwalters · 2026-03-19T15:21:54Z

The prow job ran and passed, so that's good, although it's not testing too much more than Jenkins is I believe.

cgwalters

So...I think this looks good but we should talk about tests.

install the bootloader to multiple devices, e.g. a /dev/md0 RAID1 device backed by 3 physical disks, each with an ESP partition.

Does Anaconda support setting this up? I think we could automate that here. Other than that, another way would be a custom bit that spawns a bcvk ephemeral instance w/two virtio block devices attached, partitions both disks, then uses bootc install to-filesystem.

cgwalters · 2026-03-19T15:30:37Z

src/efi.rs

+            // Unmount any previously mounted ESP to ensure we install to the
+            // correct device. This is important for multi-device installs where
+            // each device has its own ESP.
+            self.unmount()?;


I think this is OK as is, but it'd probably be cleaner longer term to have the mount clearly scoped to the device.

cgwalters · 2026-03-19T15:42:26Z

Dockerfile

+dnf -y copr enable rhcontainerbot/bootc centos-stream-9-x86_64
+dnf -y install bootc
+dnf clean all
+rm -rf /var/log


Yeah...at some point we'll need to do rpm-software-management/dnf5#2011 (comment)

ckyrouac · 2026-03-19T17:31:31Z

So...I think this looks good but we should talk about tests.

install the bootloader to multiple devices, e.g. a /dev/md0 RAID1 device backed by 3 physical disks, each with an ESP partition.

Does Anaconda support setting this up? I think we could automate that here. Other than that, another way would be a custom bit that spawns a bcvk ephemeral instance w/two virtio block devices attached, partitions both disks, then uses bootc install to-filesystem.

Currently the tests for multi device are in the bootc PR, see https://github.com/bootc-dev/bootc/pull/1911/changes#diff-c8f79904b9f0bdcea90dd5a8e8cfd0c84e7f6da4a36868cf0b86421c767f452b

ckyrouac · 2026-03-19T17:33:52Z

The prow job ran and passed, so that's good, although it's not testing too much more than Jenkins is I believe.

The cross arch builds didn't re-run after the /ok-to-test. I'm not sure if that's blocking this from merging. The builds should pass now with the copr rpm in the test container.

openshift-ci bot added do-not-merge/work-in-progress needs-ok-to-test labels Mar 5, 2026

gemini-code-assist bot reviewed Mar 5, 2026

View reviewed changes

src/efi.rs Outdated Show resolved Hide resolved

src/efi.rs Outdated Show resolved Hide resolved

src/bootupd.rs Outdated Show resolved Hide resolved

cgwalters reviewed Mar 6, 2026

View reviewed changes

src/efi.rs Show resolved Hide resolved

src/efi.rs Show resolved Hide resolved

ckyrouac force-pushed the multi-device-commit-cleanup branch 8 times, most recently from 10341ba to d868b99 Compare March 10, 2026 14:22

cgwalters reviewed Mar 10, 2026

View reviewed changes

.github/workflows/workflow-test.yml Outdated Show resolved Hide resolved

ckyrouac force-pushed the multi-device-commit-cleanup branch from d868b99 to 1365054 Compare March 11, 2026 12:38

ckyrouac marked this pull request as ready for review March 11, 2026 13:33

openshift-ci bot removed the do-not-merge/work-in-progress label Mar 11, 2026

ckyrouac force-pushed the multi-device-commit-cleanup branch 3 times, most recently from 612732a to b9b9176 Compare March 11, 2026 19:17

cgwalters approved these changes Mar 11, 2026

View reviewed changes

ckyrouac force-pushed the multi-device-commit-cleanup branch from b9b9176 to 42500c3 Compare March 12, 2026 14:30

deps: Upgrade cap-std-ext from 4.x to 5.0

4a63f9c

Align cap-std-ext with the version used by bootc-internal-blockdev, eliminating duplicate cap-std/cap-primitives dependency chains. Assisted-by: Claude Code (Opus 4) Signed-off-by: ckyrouac <ckyrouac@redhat.com>

ckyrouac force-pushed the multi-device-commit-cleanup branch from 42500c3 to 313c930 Compare March 17, 2026 15:09

ckyrouac added 4 commits March 17, 2026 15:31

deps: Update blockdev crate

17c5ade

Signed-off-by: ckyrouac <ckyrouac@redhat.com>

component: Clearer docs for Component trait

6380a4f

Signed-off-by: ckyrouac <ckyrouac@redhat.com>

blockdev: Remove blockdev code, it is now in the blockdev crate

4d61579

Signed-off-by: ckyrouac <ckyrouac@redhat.com>

install: Migrate to using blockdev functions and Device

bacc54b

Instead of referencing devices as a &str, use Device from the blockdev crate and its related functions. Signed-off-by: ckyrouac <ckyrouac@redhat.com>

ckyrouac added 4 commits March 17, 2026 15:31

spec: Add conflict with bootc < 1.14.0

e13e60c

Signed-off-by: ckyrouac <ckyrouac@redhat.com>

ci: Install bootc from rhcontainerbot/bootc copr repo

7d0b735

This uses the latest bootc RPM built from main in the CI tests. Assisted-by: Claude Code (Opus 4) Signed-off-by: ckyrouac <ckyrouac@redhat.com>

ckyrouac force-pushed the multi-device-commit-cleanup branch from 313c930 to 7d0b735 Compare March 17, 2026 19:32

ckyrouac requested a review from cgwalters March 18, 2026 16:23

openshift-ci bot added ok-to-test and removed needs-ok-to-test labels Mar 19, 2026

cgwalters approved these changes Mar 19, 2026

View reviewed changes

cgwalters merged commit b788553 into coreos:main Mar 19, 2026
10 of 12 checks passed

Conversation

ckyrouac commented Mar 5, 2026

Uh oh!

openshift-ci bot commented Mar 5, 2026

Uh oh!

gemini-code-assist bot left a comment

Choose a reason for hiding this comment

Code Review

Uh oh!

Uh oh!

Uh oh!

Uh oh!

cgwalters left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

ckyrouac commented Mar 11, 2026

Uh oh!

cgwalters left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

cgwalters Mar 11, 2026

Choose a reason for hiding this comment

Uh oh!

ckyrouac Mar 12, 2026

Choose a reason for hiding this comment

Uh oh!

HuijingHei Mar 12, 2026

Choose a reason for hiding this comment

Uh oh!

cgwalters Mar 19, 2026

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

cgwalters commented Mar 17, 2026

Uh oh!

ckyrouac commented Mar 18, 2026

Uh oh!

cgwalters commented Mar 19, 2026

Uh oh!

ckyrouac commented Mar 19, 2026

Uh oh!

cgwalters commented Mar 19, 2026

Uh oh!

cgwalters left a comment

Choose a reason for hiding this comment

Uh oh!

cgwalters Mar 19, 2026

Choose a reason for hiding this comment

Uh oh!

cgwalters Mar 19, 2026

Choose a reason for hiding this comment

Uh oh!

ckyrouac commented Mar 19, 2026

Uh oh!

ckyrouac commented Mar 19, 2026

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants