Skip to content

os/v3/puppy: an installed 1bit OS does not boot - init resolves pupsfs=1BIT-DATA but not pupsfs=UUID=... #865

Closed

Description

@bong-water-water-bong

A disk installed by 1bit-install does not boot on the cycle-13 image. Puppy's init resolves
pupsfs= as a label but not as UUID=…, and installer step 5 rewrites every entry to
pupsfs=UUID=<uuid>. So the installed system cannot find its own data partition, drops to the rescue
console, and the install gate's "boot the installed disk" row cannot pass.

Found while verifying the install docs (lane pi-rlk, ~/briefs/TDOCS-INSTALL.md), on the built artifact
~/wt/puppy-out-c13/1bit-os-puppy.img (sha256 dcc619ff74882f484e4c7dfb373980f294ce2f39b02291fb816ac7c62776beb7)
in QEMU, Secure Boot on with our MOK in db.

Measured

Install to a bare 8 GiB disk (1bit-install /dev/vda --yes): all seven steps green, exit 0, and the
installed grub.cfg carries what step 5 wrote:

linux /1bit/vmlinuz pmedia=usbhd pupsfs=UUID=7fa88580-b15a-469d-ab9b-f0215012eaf4 1bit.data=UUID=7fa88580-... psubdir=1bit pfix=copy,nox

Booting that disk (as a virtio disk, and again as the only USB device with the stick removed) fails in
init:

Waiting for partition [UUID=7fa88580-b15a-469d-ab9b-f0215012eaf4]...............failed
0: PMEDIA=usbhd PDRV=UUID=7fa88580-... PSUBDIR=/1bit pfix=copy,nox
1: PDRV= P_BP_ID=UUID=7fa88580-... P_BP_FN=
Partition UUID=7fa88580-b15a-469d-ab9b-f0215012eaf4 not found.
*** UUID=7fa88580-... /1bit/puppy_upup64r_26.04.sfs not found.
*** Error is too critical, dropping out to console...

The partition is there — from the rescue console on that same boot:

~ # blkid
/dev/sda2: LABEL="1BIT-DATA" UUID="7fa88580-b15a-469d-ab9b-f0215012eaf4" TYPE="ext4"
/dev/sda1: LABEL="1BIT-OS" UUID="3E36-7A05" TYPE="vfat"

So it is not enumeration, not pmedia=usbhd (the install booted from a USB device both times), and not the
disk: it is the UUID= form of the pinned partition that init does not resolve.

One-token A/B, same installed disk, nothing else changed

grub.cfg boot line result
pupsfs=UUID=7fa88580-… (what step 5 writes) init: "Partition UUID=… not found", rescue console
pupsfs=1BIT-DATA (label), 1bit.data=UUID=… kept boots: puppypc20143, /data/.installed present, lemond up, GET /api/v1/models 401 without the key and 200 with it, SSH on the stick's key

Suggested fix

The pinning has to survive two 1BIT-DATA partitions (engine#539/#711), so the fix is in init, not in
installer step 5: teach the pupsfs= handling to resolve UUID=<uuid> (the same blkid -t path
1bit.data= already uses through findfs). If init can only take a label this cycle, step 5 should keep
pupsfs=1BIT-DATA and rely on 1bit.data=UUID=… — but then a machine with the stick in and a disk
installed can load the wrong save layer, which is the bug the pinning exists for.

Also noticed, cosmetic: step 5's sed rewrites the string pupsfs=1BIT-DATA inside the comment above
the menu entries too, so the installed grub.cfg says "pupsfs=UUID=… sets Puppy's PDRV" in a comment and
grep -c reports 4 pinned entries for 3 menuentries.

Verification rig and full transcript: ~/wt/rlk-vm/ (verify.sh, results.txt) on strixhalo; the
install/clients docs that found it are in the same lane's branch.

Activity

  1. added
    bugSomething isn't working
    os1bit OS image, installer and appliance console (os/mini, os/v2, os/v3)
    on Nov 3, 2026
  2. bong-water-water-bong commented on Nov 3, 2026

    @bong-water-water-bong
    CollaboratorAuthor

    Triaged at origin/main 692cfa25c — and the fix is in a file that is not in main yet: os/v3/puppy/1bit-install.sh, patch-init.sh and the embed/etc/init.d/* scripts live only on PR #828's branch (os/v3-fragment-parents, head 5a310e89). Verified there, so the report stands on the branch tip and the row inherits #828.

    Confirmed, with the lines the fix has to touch:

    • 1bit-install.sh:158-161 — step 5 rewrites every entry with s|pupsfs=1BIT-DATA|pupsfs=UUID=$datauuid 1bit.data=UUID=$datauuid|g.
    • patch-init.sh never touches pupsfs= handling at all: its whole surface is the one_rescue trap, the three exec /bin/sh failure paths and the two cycle-14 progress lines (git show pr828:os/v3/puppy/patch-init.sh | grep -c pupsfs → 0). So nothing in our tree teaches init the UUID form; the A/B in the report is the whole story.

    The step that should have caught this is blind by construction

    1bit-install.sh:163-164 is the installer's own assertion:

    n=$(grep -c "pupsfs=UUID=$datauuid 1bit.data=UUID=$datauuid" "$gcfg")
    [ "$n" -ge 1 ] || fail "no grub entry pins this disk's data partition (pupsfs=UUID=...)"

    It greps for the exact string step 5 just wrote. A self-consistent check cannot fail, and it proves nothing about whether init resolves the value — which is why seven green steps, exit 0, and an unbootable disk coexist. That is the same class as #709's secureboot SKIP and #814's footer: a row that asserts its own write is not a gate. Whatever fixes the init side, the assertion should become a readback of the thing init needs (e.g. findfs on the value that ends up on the boot line, in the installer's own environment).

    Our own data layer already speaks this form — so the fix is not a new mechanism

    05onboard-data:39-42 takes spec from 1bit.data= and :46-53 resolves it with findfs "$spec"; findfs accepts both forms, one line on this box:

    $ findfs UUID=30801e38-5b73-40f3-9741-fb3dab9d9f12
    /dev/sda2
    

    So the init-side change is "resolve pupsfs=UUID=… with the same findfs our script already uses", not a new parser. If init cannot be patched this cycle, the fallback in the report (pupsfs=1BIT-DATA + 1bit.data=UUID=…) is not equivalent: it reopens the wrong-save-layer bug #539/#711 on any machine with the stick in, and our layer would load UUID= correctly while Puppy loads the SFS from the wrong partition — a split brain, worse than a rescue console.

    Two more, both cheap to fix in the same PR:

    1. The cosmetic one is a real assertion hazard: sed -i rewrites the pupsfs=1BIT-DATA string inside the comment above the menuentries too, so the installed grub.cfg documents a value it does not use and grep -c counts 4 pinned lines for 3 entries. Anchor the sed on ^linux (or -e '/^linux /s|…|…|').
    2. Nothing in the install path boots what it wrote. 1bit-install.sh ends with say "reboot and remove the stick"; the only thing that ever tests the installed disk is the QEMU rig in ~/wt/rlk-vm/. os/v3/puppy: 1bit-install step 7 reads the wrong Boot#### id from efibootmgr, so BootNext points at the BootManagerMenuApp #866 (same report series) is the other half of "an install that says green and does not boot", so the two should land together with one row that boots the installed disk and reads it back — that row is also os: 1bit-install registers no firmware boot entry and the live kernel has no efivarfs - the installed disk only boots by firmware fallback #709's outstanding acceptance (efibootmgr -v + BootOrder naming the entry after one install).

    Next action: patch pupsfs= resolution on the init side inside PR #828 (this file set only exists there), turn :163 into a findfs readback, and anchor the sed on linux lines. Owner pi-os3 (PR #828), with #709's readback row as the shared gate. Bucket A — with the caveat that it closes on #828 (or a PR stacked on it), since the file is not in main.

  3. bong-water-water-bong commented on Nov 3, 2026

    @bong-water-water-bong
    CollaboratorAuthor

    Closed by the merge of #890 (8251ae597511e038e3f28879ee76c2a16c60201b). Verified at 65460184 in os/v3/puppy/patch-init.sh: the resolver is there (:247-250 names this issue and the broken pattern, and prefixed UUID=/LABEL= tokens are matched against blkid's own quoted fields), and the build-time guard refuses an init that cannot resolve the form (:136-139). The acceptance in the issue — an installed disk reaching userspace from its own pupsfs=UUID=… boot line — was measured on QEMU with Secure Boot on in the PR (PASS installed … the installed disk reached userspace on its own boot line).

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    bugSomething isn't workingos1bit OS image, installer and appliance console (os/mini, os/v2, os/v3)

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions