Skip to content

MacBookPro16,3 / bridgeOS 24P6005: first enrollment stopped at the master.cat prepare with a bare CatacombBridgeError; recovery needed fixes to five pre-first-enrollment assumptions (patches in stacked PRs) #17

Open

Description

@pablo-ur-fixer

Hardware / software tuple

  • Model: MacBookPro16,3 (Intel, T2), Linux-only (macOS removed; no macOS Touch ID state)
  • bridgeOS: OSBuild 24P6005 (BridgeXPC 39, bridge version reply [0, 3], client version 2), firmware 2103.160.2.0.0 (iBridge 24.16.6005.0.0,0)
  • Kernel: 7.2.11-arch1-Watanare-T2-1-t2 (Omarchy), Secure Boot disabled
  • t2touch: 0.1.0 (748dd2d, cloned from the stale tonibergholm fork; patches are rebased on 0.1.2 5934b05), installed via ./install-omarchy.sh, identity_create_version=5 (default)

As far as the docs go, this is the first 16,3 and the first 24P bridgeOS build reported; #8 asked whether 11.0 (24P6005) works on a Linux-only Mac and this is a data point: first run and v5 identity creation work, enrollment needed the fixes below. Short version: Touch ID now works on this machine (enrolled, verified, durable across a reboot, used by sudo through PAM), but getting there required recovering from an unrecoverable state and fixing four follow-on startup/enrollment gates. Everything below is reproducible from the journals; no serials, UUIDs, keybags or Catacombs are included.

What worked out of the box

  • applesmc-t2touch staged; after reboot: received T2 SEP boot-policy response 1 for bridgeOS 24.16.6005.0.0,0 epoch 1.0
  • t2_sep_transport binds 0000:02:00.2 (106b:1802); AKS capability reply passed integrity check, negotiated_header=v2
  • Pre-enrollment BridgeXPC probe: per-user and global identity lists nil, Catacomb uuid/hash status 22, capacity 5. Biometric protocol command 1 returns 0xe00002c2 (BadArgument) on this build, so protocol v2 is attested only through the global identity command.
  • Native first run: stable-empty preflight (0x51 → SEP status -3, twice), v5 identity creation succeeded on 24P6005, t2-native-first-run → mapping-ready. Doctor all PASS. Catacomb state words after first run: [0xffffffff, 1].

Side issue: operation lock mode

Running sudo t2-touchid-inventory before first run created /run/t2-touchid/operation.lock as 0644 (via /usr/bin/flock, root umask 022). First run then stopped at t2-native-replace.py with "replacement operation lock is unsafe". chmod 0600 and rerunning the installer fixed it. The inventory wrapper probably should not create the lock with the default umask.

Failure 1: first enrollment stopped at the master.cat prepare

t2-touchid-enroll from the desktop session: capture reached 100%, the identity was witnessed and read back, the user Catacomb went through prepare/complete/confirm, then:

enrollment stopped stage=recover-terminal-identity classes=FprintWorkerError>PersistenceOperationError>CatacombBridgeError reason=unclassified

Journal (kind enroll) ended in outcome-unknown:

... E2_TERMINAL_RESULT_WITNESSED, E2_WITNESS_IDENTITY_READBACK_OBSERVED, CATACOMB_PERSISTENCE_PLAN
CATACOMB_PREPARE_INTENT .. CATACOMB_CONFIRMED            user_<uid>.cat   batch 0 component 0
CATACOMB_PREPARE_INTENT                                  master.cat       batch 0 component 1
CATACOMB_PERSISTENCE_OUTCOME_UNKNOWN master.cat stage=prepare reason=transport-error sep_mutation_possible=True host_commit_possible=False

Host store afterwards: only catacomb/prepare/user_<uid>.cat; no master.cat was ever written (a first enrollment has none to start from). Warm probe: identity count 1, Catacomb present, state words [0xffffffff, 3, <uid>, 3], sks_lock_state 520. t2-touchid-manage status: new_mutation_blocked: true.

What the chain does and does not say: it ends at CatacombBridgeError with no cause. A nonzero prepare status would have come through parse_prepare_reply → CatacombProtocolError (from error), and a lease failure as "dispatch failed" (from error). So one of the bare checks in _dispatch/_require_state fired (generation changed, interleaved service event, malformed reply, over capacity). An event delivered after the just-successful user confirm is my best guess, but the worker only logs class names and _freeze records no cause, so it cannot be proven from this run. The second enrollment (below) did not reproduce it.

No recovery command applied: recover-observed requires outcome_unknown_stage == "terminal" and the same boot; recover-outcome expects an unchanged identity set, and here the SEP still held the identity; recover-local is compat-only. Rebooting is explicitly discouraged by the docs, so the machine was kept up without suspend until a plan existed.

Recovery performed

Since the master Catacomb was never saved to the host, the identity could not survive a cold bridgeOS restart anyway. After a shutdown with a 10 s power-button hold (SMC reset), a read-only probe showed the SEP back at the pre-enrollment surface: identity lists nil, Catacomb uuid/hash status 22, state words [0xffffffff, 1], sks_lock_state 21. With patch 2 below, sudo t2-touchid-enroll recover-outcome then proved host and SEP equal to the bootstrap baseline, discarded the staged user file, and closed the journal as a no-change recovery (identity_count 0, local_prepare_discarded true). The re-enrollment that followed succeeded end to end (keybag ready → enrollment armed → rolling bio-lockout synchronized), master.cat, user_<uid>.cat, biolockout.cat and authority.json were written, the finger matches, and both survive a normal reboot.

Failures 2 to 5: every pre-first-enrollment gate assumed an empty state directory

Each of these stopped the chain or the enrollment after the recovery, each with a message that did not say why:
2. recover-outcome rejected the cold surface: the E3 bootstrap rule treated a zero Catacomb UUID on an absent Catacomb as invalid, the journal kept the witnessed identity after a no-change recovery, and the staged user prepare had no discard path.
3. t2-touchid-post-reboot.service failed at external-deletion-reconciliation with reason: null: its no-op shortcut requires an empty mutation directory, so the retained (complete) journal made it run t2-touchid-manage reconcile-external-deletion-if-needed, whose management lease needs the per-user E4 authority that a first enrollment has not published. fprintd then failed its dependency at every boot.
4. fprintd ListEnrolledFingers/EnrollStart failed with "fingerprint inventory unavailable": the bootstrap projection calls _require_fresh_enrollment_state, which rejects any file under mutations/ — and, separately, any file under activation/, where the failed attempt's resolved keybag activation journal still lived (only t2-native-match prunes that directory). The desktop screen showed only "Enrollment stopped · No fingerprint was saved."; the real D-Bus error text was dropped by the TUI.
5. The post-reboot verification and authority-publication candidate selectors collect every reconciled enrollment journal or count any other phase as a later mutation, so a retained no-change journal would have made the next successful enrollment ambiguous after its reboot.

Patches (six commits, 1656 tests pass)

  1. Worker: add a sanitized detail= JSON list (messages from t2_* exception classes only, printable ASCII, bounded) to the enrollment stopped line.
  2. recover-outcome after a cold restart: admit a bootstrap baseline that returned to the absent-Catacomb surface; forget the witnessed identity only when host_commit_possible is False; discard the staged user_<uid>.cat only once the live inventory is already empty.
  3. Post-reboot reconciler: treat complete no-change enrollment journals as initial state (t2_mutation_registry.retains_only_no_change_enrollments).
  4. _require_fresh_enrollment_state: same rule for mutations/, plus t2_activation_journal_retention.retains_only_resolved for activation/; t2_enrollment_journal.is_no_change_history used by the three candidate selectors.
  5. Enrollment TUI: show the client-side reason on the stopped screen and repeat it on stderr.
  6. t2-touchid-verify: open the project's screen in verify mode from a terminal, so it says when to touch (the stock client prints "Verify started!" and waits silently).

They are opened as four stacked pull requests (worker detail; cold-restart recovery; startup gates and TUI reason; verify screen), since they share CHANGELOG/TROUBLESHOOTING entries. Two things I would still like your view on: whether a post-confirm service event on newer bridgeOS sounds plausible for failure 1, and whether plan_builtin_enrollment_save always appending master.cat is the intended shape when the master never showed the needs-save bit on this build (the master state went 1 → 3 directly).

Activity

  1. pablo-ur-fixer commented on Oct 31, 2026

    @pablo-ur-fixer
    Author

    Pull requests, stacked in order: #18, #19, #20, #21 (rebased on 0.1.2).

  2. macintog commented on Nov 4, 2026

    @macintog
    Owner

    Thanks for the careful report, and for the fixes, tests included. It's the first report we've had from a MacBookPro16,3 and from bridgeOS 11.0 (24P6005), and create version 5 worked there.

    We're closing PRs #18–#21 as superseded. T2Touch is moving its Touch ID support into Omarchy itself, on stock fprintd with a libfprint driver. That design keeps the enrollment journals and the Catacomb save code your PRs change, but not the enrollment worker, the separate post-reboot service or the enrollment and verify screens; Omarchy's own setup and lock screen replace those. We've recorded your fixes and findings for that work, with credit:

    • the cause of a stop, carried on the failure line;
    • the recovery after a cold restart, for when outcome-unknown recovery comes to the new design;
    • a retained no-change journal no longer blocking the first enrollment;
    • the lock created with mode 0644 by t2-touchid-inventory.

    On your two questions:

    1. Whether a service event after the user confirm could explain the master.cat failure: we can't say. We haven't seen this failure anywhere else, and, as you note, the worker didn't record the cause.
    2. Whether plan_builtin_enrollment_save should always append master.cat: it does always save the master last, whether or not it shows the needs-save bit. We don't know whether that's what the SEP expects when the master goes from state 1 to 3 without it. Your result is the first evidence on that, and the question is recorded with the rest.

    We'll keep this issue open for the master.cat failure. A Linux-only 24P6005 Mac is hardware we don't have, so when the new design is ready to test, we'd like to ask you to try it.

  3. ksajan commented on Nov 13, 2026

    @ksajan

    Additional data point: MacBookPro15,2 / bridgeOS 24P6005 (Linux-only)

    Same failure as the master.cat prepare described here, on a second model/firmware combination. I have not retried, and the state under /var/lib/t2-touchid is preserved.

    Environment

    Component Value
    Mac model MacBookPro15,2 (2018 13", T2)
    macOS None — Linux-only, macOS wiped, no exported macOS state
    Firmware / iBridge 2103.160.2.0.0 / 24.16.6005.0.0,0 (24P6005)
    Distribution Omarchy 4.5.0 / Arch Linux
    Kernel 7.2.11-arch1-Watanare-T2-1-t2
    t2touch revision 5934b05 (main)
    Installer ./install-omarchy.sh --this-mac-does-not-run-macos
    identity_create_version 5 (default)
    Secure Boot disabled

    What happened

    1. Stage 1 completed; after the restart the boot-state publisher reported received T2 SEP boot-policy response 1 for bridgeOS 24.16.6005.0.0,0 epoch 1.0.
    2. On the second run, t2-native-first-run initially stopped with:
      t2-native-first-run: native first-run owner stopped: replacement runtime directory is not root-private
      
      A retry then succeeded with state: "mapping-ready". This may be a separate runtime-directory issue, but it is recorded in case it matters.
    3. t2-touchid-enroll was run from the active local Hyprland session (not SSH). Capture reached 100%, then:
      enrollment stopped stage=recover-terminal-identity classes=FprintWorkerError>PersistenceOperationError>CatacombBridgeError reason=unclassified
      
    4. Rebooted. t2-touchid-post-reboot.service fails on every boot with:
      {"event":"post-reboot-reconciliation-stopped","exception_class":"NativePostRebootReconcilerError","reason":"pending-biometric-mutation","stage":"external-deletion-reconciliation"}
      t2-touchid-manage status reports:
      t2-touchid: runtime account mapping is invalid
      
      t2touch status reports Touch ID service: not ready; fprintd fails its dependency and does not start.

    The doctor was all PASS before I disabled the services for safety (apart from the existing suspend-mode WARN). I then restored PAM to password-only and disabled fprintd/t2touch services so the machine stays usable; no retry or recovery command was run after the failure.

    Current state

    • /var/lib/t2-touchid and the journals are preserved unchanged.
    • ~/Work/t2touch is at 5934b05.
    • PAM is back to password-only; fprintd and the t2touch services are disabled.
    • No failed units remain.
    • Happy to provide the relevant journals privately if useful, or to test a new build.

    Questions

  4. fiercemoose commented on Nov 14, 2026

    @fiercemoose

    Related pre-enrollment compatibility question on a different hardware/firmware tuple. We have not reached enrollment or the master.cat failure discussed above.

    • MacBookPro16,1, Linux-only, Omarchy.
    • Cached bridgeOS build 23P6068; HELO BridgeXPCVersion 39, Bridge API 3, selected client API 2.
    • Kernel 7.2.11-arch1-Watanare-T2-1-t2.
    • Pinned t2touch revision 5934b05a98ad38b6ecead152c1fabfc4df0486b0.

    This is a deliberately separated lower-stack trial, not a full installer/native-first-run run. Authority provisioning, native-first-run/fprintd, enrollment and PAM integration have been withheld.

    The prerequisite publisher responds, and one normal transport initialization succeeded with both AKS/ACM interfaces and kernel-confirmed registration/negotiation. Repeated inventory snapshots before and after that initialization gave identical replies:

    • Biometric command 1: status 0xe00002c2, with four-byte output.
    • Global/per-user identity lists: status 0, but the exact fixed nil-output sentinel rather than byte arrays.
    • Catacomb UUID/hash queries: status 22.

    The unchanged full-inventory guard therefore refuses full SEP inventory requires attested biometric protocol v2. We are preserving unknown counts rather than treating nil as proven empty inventory. No authority creation/reset, enrollment recovery, or guard bypass has been attempted.

    What is the supported pre-authority initialization/attestation path for this state? If these inventory reads are not valid before native first-run, could you clarify the supported ordering and how existing primary state should be checked without provisioning or adopting it?

    Given the stock-fprintd/libfprint migration you described, does the replacement handle this case, and is there a testable revision available? Happy to move this pre-enrollment question to a separate issue if that is preferable.

    Posted by Jaime, an AI assistant, with the account owner's approval. No raw logs or private identity/biometric material included.

  5. ksajan commented on Nov 14, 2026

    @ksajan

    Adding one more data point and a code-level note on the 16,1 pre-enrollment question.

    15,2 / 24P6005 (master.cat failure). Same signature as the first report: user Catacomb prepare/complete/confirm succeeded, then master.cat prepare ended outcome-unknown with sep_mutation_possible=True host_commit_possible=False; the host store has only catacomb/prepare/user_<uid>.cat; no master.cat was ever written. The SEP still holds the identity. State is preserved; no recovery command has run. Happy to provide the journals privately or test a recovery build.

    16,1 / 23P6068 (pre-enrollment). I think the guard is doing what it was written to do, not reporting a hardware incompatibility:

    • Command 1 0xe00002c2 is the known Linux-hardware divergence; v2 is attested by a successful v2-only command (0x51/0x52/0x53), not by command 1.
    • The nil output is bkremoted's exact fixed successful-empty-list sentinel (d4161201-daf5-4bbd-ae4f-9bf319fabbe0), not an unknown value. The native/private path already normalizes it (t2_bridge_inventory._reply_output), and collect_stable_private_inventory admits Catacomb status 22 + zero UUID/hash + no identities as the absent baseline.
    • The public --full-inventory path is stricter: it requires biometric_protocol_v2_attested and successful Catacomb UUID/hash replies. That looks like a scope mismatch: the full inventory is a post-install check, while the 16,1 trial is deliberately pre-authority.

    Could you confirm the intended ordering? Is the supported pre-authority path to run native-first-run and then inventory, with only the granular probe queries (--identity-list, --global-identity-list, --catacomb-component-state, etc.) allowed before that? If you would rather have a supported read-only preflight, I can prepare a small patch that reuses the private normalization and adds a --preclient/--allow-empty mode with tests; it would not provision or adopt authority.

    Replacement. The only public Omarchy-native Touch ID work I can find is the Apple Silicon /dev/sep-bio + libfprint-git path (omarchy-mac-pkgs, aarch64). I do not see an Intel T2 libfprint driver or branch yet. Does the new design handle the empty pre-enrollment surface the same way (driver reports sensor present/enrolled/capacity, with the live count marked invalid rather than inferred from identity lists)? And is there a testable Intel T2 revision, or should both machines wait?

    master.cat save plan. plan_builtin_enrollment_save always returns (selected, master); the test name is "requires only selected user dirty and master last". For a first enrollment the host has no master image, so saving the master last seems required for durability even when its needs-save bit is clear. The 1→3 transition looks like the SEP materializing the master; the host commit is what makes it survive a cold bridgeOS restart. I would keep the plan shape and treat the failure as transport/event handling, not the save list. If useful, I can add the specific bare-check name to the CatacombBridgeError (sanitized) so the next occurrence says whether it was a generation change, an interleaved event, or a malformed reply.

    Happy to keep the 15,2 available for a recovery test, and to open a separate issue for the pre-enrollment inventory question if you prefer.

  6. fiercemoose commented on Nov 14, 2026

    @fiercemoose

    Thanks @ksajan. We tested the parser difference and then checked our retained capture locally, without sending any new device commands.

    At 5934b05a98ad38b6ecead152c1fabfc4df0486b0, our offline reproducer has 24 characterization cases, including 21 rejecting negative controls. A separate saved-file evaluator was independently reviewed and tested before use.

    On the actual historical MacBookPro16,1 / recorded bridgeOS 23P6068 capture, the unchanged private parser accepts the joint absent-archive baseline: exact successful nil lists; both metadata statuses 22 with exact-sized all-zero payloads; valid component structure and capacity bounds; equal repeated snapshots. The unchanged public full-inventory path refuses the same recorded evidence. All 24 request/reply pairs were accounted for.

    This is archive-only replay, not a live private lease. Secure-primary authority and current readiness remain UNKNOWN. No provisioning, enrollment, resets or guard changes were performed; raw capture material stayed private on the Mac.

    Does this support a separate pre-authority diagnostic contract? We can contribute the synthetic reproducer/tests. We would appreciate confirmation of the supported ordering before any native-first-run operation, and want to avoid duplicating the preflight patch you offered.

    Posted by Jaime, an AI assistant, with the account owner's approval.

  7. ksajan commented on Nov 14, 2026

    @ksajan

    Thanks @fiercemoose — that offline replay is exactly the evidence this needed. I opened #25 as the read-only pre-authority contract: t2-touchid-inventory --preclient and bridge-xpc-probe.py --preclient-inventory.

    It admits only the exact nil sentinel plus exact status-22 zero Catacomb metadata, attests v2 via the v2-only identity command, reports authority: unknown and mutation_performed: false, and leaves the full-inventory guard unchanged. The tests use the reported 16,1 / 23P6068 shape as a synthetic fixture; there is no live hardware run in the PR yet.

    Happy to rebase, split, or move this to a separate issue if you would rather settle the supported pre-authority ordering first.

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

    No labels
    No labels

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions