Repository navigation
Description
TL;DR: Works for me. Fingerprint enrollment and verification work on this MacBookAir9,1 with Ubuntu 26.04.1. It needed an Apple Revive to bridgeOS 24P6005, the 6.18.63 kernel, and a few local fixes (one of them is #28).
This is a redacted hardware report covering a Linux-only MacBookAir9,1 that reached successful first enrollment and positive/negative fingerprint verification after several local fixes. A later second enrollment reached the sensor but stalled during persistence; it was recovered and rolled back, preserving the first fingerprint. It is not a claim that an unmodified release works end to end on this configuration.
Related: #17 (24P6005 persistence/recovery), #23 and #24 (22P creation failures), #8 (firmware choice), and #26 (Linux 7.3/macsmc prerequisite proposal). The boundaries below differ, so similar error classes should not be assumed to have the same cause.
Environment
- MacBookAir9,1 / iBridge2,15, Ubuntu 26.04.1, Linux-only disk, no imported macOS Touch ID authority.
- Source base:
5934b05a98ad38b6ecead152c1fabfc4df0486b0, with local patches; not an unmodified release. - Initial kernel:
7.3.4-3-t2-resolute; working kernel:6.18.63-1-t2-resolute. - Initial firmware: bridgeOS 9.0 /
22P353; after Apple Revive:24P6005.
Confirmed Ubuntu compatibility defect: inhibitor registration
Native enrollment and identity management both query:
systemd-inhibit --list --json=short
The installed Ubuntu command rejects the JSON option. The enrollment worker stops before capture, while the desktop UI reports only “Enrollment stopped / No fingerprint was saved.” The same check later prevented the ordinary rollback operation with “sleep inhibitor could not be established.”
The local fix calls logind's ListInhibitors through busctl --system --json=short, validates the a(ssssuu) response, and requires the same live child PID, exact owner label, sleep scope, and block mode. Acquisition/lifetime protection remains in place. Enrollment and management then pass the inhibitor gate. #28 proposes this compatibility fix on its own, separate from the recovery and protocol work.
Additional enrollment: ambiguous user-component early confirm
After successful first enrollment and matching, a second enrollment captured one new identity. The journal then reached:
terminal identity observed
user component: prepare -> complete -> host staged -> confirm intent
CATACOMB_PERSISTENCE_OUTCOME_UNKNOWN
stage=early-confirm reason=transport-error
host_commit_possible=false
The worker logged:
stage=recover-terminal-identity
classes=FprintWorkerError>PersistenceOperationError>CatacombBridgeError
reason=unclassified
The bare exception chain does not establish the specific transport cause. In particular, an interleaved service event is only a hypothesis, not an observed root cause.
Fresh, stable read-only inventory showed:
- Both global and per-user inventories preserved the original identity and contained exactly the journaled new addition.
- The selected user component was clean; the master still needed saving.
- The committed host inventory still contained only the original fingerprint, with the new user component staged under its journaled hash.
Native recover-observed only admitted terminal-stage ambiguity, so it could not handle this persistence stage. recover-outcome cannot prove no change when the new identity is present. Additional observed recovery also attempted to obtain slot names from baseline records that contain only user/identity/entity fields.
A local recovery extension used the already-existing CATACOMB_EARLY_CONFIRM_RECOVERED milestone. It required the exact additional-enrollment early-confirm shape, current same-boot mapping bindings, unchanged host baseline, protected staged-file hashes and identity metadata, one stable SEP addition, a fresh connection, and clean-user/dirty-master save state. It continued the remaining persistence without repeating capture or the ambiguous confirmation. Slot names came from the protected committed archive. Recovery completed.
The normal rollback-unverified-addition operation then removed only the second fingerprint, because I wanted to enroll a different finger. Final independent inventory contained one fingerprint. No journals were deleted or manually rewritten to clear the block.
This differs from #17's first-enrollment master-prepare boundary, although the recovery and diagnostic limitations are related. The UI's “No fingerprint was saved” should not imply that retrying is safe. A lock-screen nonresponse was observed while this second enrollment remained pending; graphical authentication was not retested after recovery.
Account authority invalidated by an unrelated passwd addition
During setup, installing OpenSSH added a system passwd row. Because T2Touch includes the complete passwd database in the account generation, activation stopped even though the target account record, password digest, and home identity were unchanged. A protected prior passwd snapshot plus the current target credential/home reconstructed the exact saved generation.
A local explicit recovery tool proves that all old rows remain byte-for-byte present and only rows were added. Under the operation lock it rechecks current inputs, then publishes a protected compatibility record; subsequent assertions revalidate the proof. Target-password, home, mapping, or incompatible account changes still fail closed. No passwd database was restored and no mapping or provisioning journal was rewritten.
This suggests a separate account-authority lifecycle review, not an automatic relaxation of generation checks.
Older firmware investigation and revive result
- Creation versions 5 and 4 returned SEP status
-13on22P353. Original failures were preserved, and fresh stable absence evidence across boot generations preceded the explicitly authorized experimental attempts. Status -13 alone was not treated as proof of no mutation. - Static analysis of the matching bridgeOS host AppleSEPKeyStore code identified a version-3 creation layout. The SEP image remained encrypted; this was not a decoded SEP implementation.
- The experimental v3 codec completed creation, export, mapping, and activation-bundle replacement. Activation operation
0x21still returned-5on two boots. Saved authority passed its integrity checks. - Apple Revive changed
22P353to24P6005, preserving Ubuntu and the saved authority. After selecting the compatible kernel and resolving the separate passwd-generation mismatch, the existing activation-resume path succeeded on attempt 3, including independent activation proof. The saved identity was not recreated for that retry.
This establishes activation of one retained v3-created identity after this firmware transition. It does not establish the cause of -5, qualify v3 generally, or prove fresh v3 creation on 24P6005. No automatic protocol fallback is proposed.
Earlier integration findings and scope disposition
- Ubuntu puts
modprobein/usr/sbin; several T2Touch helpers assumed/usr/bin. Local helper-path fixes allowed transport startup. - Ubuntu's dedicated
root:shadowread permission on/etc/shadowwas rejected. A narrow account-collector compatibility change retained the other ownership/permission checks. - Linux 7.3 removed the ACPI interfaces expected by the applesmc prerequisite and uses the macsmc stack. Matching headers were installed; reinstalling headers was not a solution. The local workaround was a side-by-side 6.18.63 kernel and a machine-local GRUB default, with protected backup and rollback instructions. No kernel policy or GRUB change is proposed for T2Touch; Publish the T2 SEP boot state through macsmc on Linux 7.3+ #26 is the related product adaptation.
- Boot-policy response 3 became response 1 after a physical SMC reset; ordinary Linux reboots had not cleared it on this machine.
- KaiT2en was present during early failures and was removed with authorization. Its contribution to the creation failure was not established.
Validation and remaining limits
- First enrollment completed; I confirmed an enrolled-finger match and an unenrolled-finger rejection in separate visible transactions.
- Final recovery state: service ready, one fingerprint, no pending mutations and no T2Touch sleep inhibitor.
- Last local focused runs: 60 recovery/persistence/worker tests and 22 inhibitor/management tests passed. Earlier v3 work passed 100 AKS tests; account collection had 19 passing tests.
- The broad local suite had four installer subtest failures against the pre-existing Ubuntu adapter's differences from the Limine expectations. This is not an all-green full-suite claim for the machine's dirty checkout.
- Reboot persistence after enrollment, graphical unlock after repair, a separate PAM transaction, and suspend/resume remain unverified. Password fallback remains configured.
Private authority, raw journals/captures, firmware binaries, device/identity identifiers, and session logs are retained locally and are not attached. Maintainer guidance is requested on separate recovery/account-authority proposals; the inhibitor fix is proposed separately in #28.
Activity
[-]MacBookAir9,1 / Ubuntu: 22P353 to 24P6005 bring-up, inhibitor compatibility and additional-enrollment recovery gaps[/-][+]MacBookAir9,1 / Ubuntu 26.04: working after 22P353 → 24P6005 revive; inhibitor fix and recovery gaps[/+][-]MacBookAir9,1 / Ubuntu 26.04: working after 22P353 → 24P6005 revive; inhibitor fix and recovery gaps[/-][+]MacBookAir9,1 / Ubuntu: 22P353 to 24P6005 bring-up, inhibitor compatibility and additional-enrollment recovery gaps[/+]