Fix btrfs-subvolume install aborting at 99% after root GPT retag

[joshmccall221]

Problem

On btrfs installs with subvolumes, a fresh install aborts at step 7/8 (99%) — right after bootc install completes — with:

fisherman: fatal: writing hostname: finding composefs deploy etc:
reading composefs deploy base /mnt/fisherman-target/state/deploy:
no such file or directory

The OS is fully deployed on disk (composefs deploy, systemd-boot, UKI, loader entry all present), but the system is left unconfigured (no hostname written) and the installer reports failure.

Root cause

SetupBtrfsSubvolumes mounts the target with subvol=@, and the composefs deployment (state/deploy/<hash>) lives inside the @ subvolume.

After bootc install, for systemd-boot + composefs installs the root partition is retagged to the Linux-root GPT type for GPT auto-discovery. This unmounts and remounts the root — but the remount passed empty mount options:

disk.Mount(rootPart, activeTargetMount, "")   // drops subvol=@

So the root is remounted at the btrfs top-level instead of the @ subvolume. Post-install steps that write through state/deploy (WriteHostname, system Flatpak copy) then look at the top-level state/, which has no deploy/, and fail. The inline comment on that block ("…the empty /boot/efi directory in the XFS root") shows the path was written/validated for XFS and regressed on btrfs-subvolume layouts.

Fix

Remount the root with the same options SetupBtrfsSubvolumes used (subvol=@,compress=zstd:1) when the install uses btrfs subvolumes. To keep the two mount sites from drifting:

• Hoist the options into a shared disk.BtrfsRootMountOpts constant (used by both SetupBtrfsSubvolumes and the retag remount).
• Extract the retag remount into a testable disk.RemountRoot(diskDev, partNum, target, btrfsSubvols) helper.

Non-btrfs-subvolume installs (xfs/ext4) remount bare exactly as before.

Test

Adds TestRemountRoot, asserting the btrfs-subvolume remount carries subvol=@ (and compress=zstd:1) while non-subvolume installs remount with no options.

go test ./internal/disk/ -run TestRemountRoot   # PASS
GOOS=linux GOARCH=amd64 go build ./...          # OK
go vet ./internal/disk/ ./cmd/fisherman/         # clean

[joshmccall221]

Validated live end-to-end on a real machine (Snow Linux live ISO → install to internal NVMe).

Before the fix: install died at step 7/8 (99%) with:

fatal: writing hostname: finding composefs deploy etc: reading composefs deploy base /mnt/fisherman-target/state/deploy: no such file or directory

Root cause confirmed on-disk: after the GPT retag, root was remounted with empty options, exposing the btrfs top-level, so state/deploy/<hash> (which only exists inside the @ subvolume) was unreachable.

With this patch: a full install (btrfs + subvolumes, systemd-boot, composefs backend, no encryption) completed all 8 steps:

Writing hostname: snow
+ mkdir -p .../state/deploy/75afcb…/etc
wrote hostname "snow" to .../state/deploy/75afcb…/etc/hostname
...
Installation complete!

Post-install verification of the NVMe: composefs deploy present under @/state/deploy/<hash>, /etc/hostname written, user account created, and systemd-boot loader entry + systemd-bootx64.efi in place — i.e. a bootable system.

The remount now preserving subvol=@ is exactly what fixes the failure.

[bketelsen]

bootcrew vm tests expected to fail

[Copilot]

Pull request overview

Fixes a btrfs subvolume install regression where the root filesystem was remounted without subvol=@ after retagging the root GPT type for systemd-boot auto-discovery, causing post-install steps (e.g., hostname/flatpak writes through state/deploy) to fail and abort the install at 99%.

Changes:

• Introduces a shared disk.RemountRoot(...) helper to remount the root partition with the correct btrfs subvolume options when applicable.
• Hoists the btrfs root mount options into a shared constant to avoid drift between mount sites.
• Adds a regression unit test to assert the remount uses subvol=@,compress=zstd:1 for btrfs-subvolume installs and no opts otherwise.

Reviewed changes

Copilot reviewed 4 out of 4 changed files in this pull request and generated 2 comments.

File	Description
fisherman/internal/disk/format.go	Adds shared btrfs root mount options constant and RemountRoot helper; reuses the constant in SetupBtrfsSubvolumes.
fisherman/internal/disk/format_test.go	Adds TestRemountRoot regression coverage for btrfs-subvolume vs non-subvolume remount behavior.
fisherman/cmd/fisherman/main.go	Switches retag post-step to use disk.RemountRoot and preserve btrfs subvolume mount options.
CHANGELOG.md	Documents the bug fix and its user-visible impact in the Unreleased changelog.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[bketelsen]

GPT-5.6-sol review:

No blocking defects found. The fix correctly preserves `subvol=@,compress=zstd:1` after GPT retagging, while non-subvolume installs retain the previous bare mount behavior. I would approve the code change.
**Residual Risks**
- `fisherman/internal/disk/format_test.go:215` tests `RemountRoot` directly but not the production wiring at `fisherman/cmd/fisherman/main.go:714`. A future call-site regression could evade this test. Non-blocking given the monolithic `main()`.
- Pre-existing adjacent issue: a separately mounted `/var` is not remounted after the root retag sequence (`main.go:589-607`, `main.go:694-721`). Subsequent writes may land in the root filesystem instead. This is not introduced by PR #9, but should be tracked separately.
- GitHub CI is red because whole-tree lint failures already exist on base commit `44ae0de`; the PR’s changed lines are clean.
**Verification**
- `go test -race ./...`: passed
- `go vet ./...`: passed
- `go build ./...`: passed
- `git diff --check 44ae0de..b50eb02`: passed
- The author also documented a successful physical-machine end-to-end install.```
- 

[bketelsen]

Looks great - fix the conflicts please.

[bketelsen]

For the record this fix is "valid" but not currently useful because the subvol support isn't wired in fisherman. Will create tracking issue and update here.