Skip to content

security: independently establish LineageOS pdx235 source OTA authenticity and fingerprint provenance #114

Open

Description

@0cwa

Blocking evidence gap

Failed build 38003576753 used lineage-23.2-20261009-nightly-pdx235-signed.zip. Embedded care_map.pb reports Sony stock Android 15 fingerprints with sparse ranges, while avbroot 3.34.1 reconstructs LineageOS Android 16 fingerprints and contiguous ranges.

ModOS currently extracts avb_pkmd.bin and META-INF/com/android/otacert from the same untrusted OTA and verifies against those extracted identities. This checks internal consistency rather than a separately pinned publisher identity. The log also warns of public AOSP AVB test keys and root vbmeta insecure flags 0x3.

Required independent investigation

  • Obtain original OTA bytes and record SHA-256, officially published source/release identity, and verified publisher signature using independently trusted LineageOS material (e.g. documented build verification). Do not bootstrap trust solely from the ZIP's own cert/key.
  • Compare the exact signed archive's care_map.pb, target-files partition build props, META/misc_info.txt, and payload AVB property descriptors; document why fingerprints differ.
  • Escalate discrepant build metadata to LineageOS device/build maintainers and sparse-ranges/AVB equality behaviour to avbroot upstream.
  • Record the reviewable evidence before filling in any exception digest. A reviewed digest is a byte-identity check, not independent source authentication.
  • Re-run strict checks and nonpublishing OTA build; separate physical first-boot and OTA-persistence acceptance.

Do not treat a green mocked verifier or output re-signing as proof of source provenance. Do not publish/merge an unverified build.

Related: #97 and helper #7.

Activity

  1. 0cwa commented on Oct 10, 2026

    @0cwa
    OwnerAuthor

    Additional source-trust research (2026-10-10): ModOS src/ota_providers.sh already validates downloaded LineageOS ZIP bytes against a SHA-256 returned over HTTPS by the official download.lineageos.org/api/v2 builds API. This is useful delivery integrity, but does not independently authenticate the ZIP's signer or reconcile its fingerprints. The official LineageOS build-verification instructions reference LineageOS/update_verifier and lineageos_pubkey; the published pinned public key at 9ffcf56a has SHA-256 0848ca38cb604887cea216a7f39d7dd62bad20b513cfdeff357ece4e731d6372 when encoded as SPKI DER. Draft ModOS PR #116 now adds a second, independently pinned certificate public-key identity check that would be required before any source care-map exception. Actual Oct 9 original ZIP bytes have not been authenticated against that public key in this review, and their target-files fingerprint provenance is still unproven, so the byte-digest exception pin remains intentionally empty. Do not consider #114 resolved or allow publication on the basis of this preparatory pin alone.

  2. 0cwa commented on Oct 11, 2026

    @0cwa
    OwnerAuthor

    Independent forensic follow-up — 2026-10-11 (read-only; release remains blocked)

    New, independently located fingerprint provenance

    The exact Sony Android 15 fingerprint shown for all six care_map.pb partitions in failed job 38003576753 is an explicit upstream LineageOS pdx235 device build override, not merely a hypothetical stale stock artifact:

    • Official device tree, lineage_pdx235.mk at commit 0559de69466e1e9af69b121985f592c4f4609283, includes PRODUCT_BUILD_PROP_OVERRIDES += BuildFingerprint=Sony/XQ-DC72/XQ-DC72:15/68.2.A.3.107/068002A003010700521798587:user/release-keys, with BuildDesc of the matching Sony A15 stock build. This change is dated 2026-03-28, predating the October 9 nightly. The current lineage-23.2 branch resolves to that commit; the exact October 9 build manifest has not yet been inspected.
    • Official LineageOS Android build core/sysprop.mk applies PRODUCT_BUILD_PROP_OVERRIDES while generating partition build props and writes ro.<partition>.build.fingerprint from BuildFingerprint. Official releasetools add_img_to_target_files.py reads the partition's *.build.prop value into care_map.pb and uses sparse-image meaningful-block ranges where applicable.
    • In contrast, avbroot v3.34.1 care_map.rs reads the partition image's AVB com.android.build.<partition>.fingerprint property and produces contiguous 2,0,N ranges from hashtree image size. The failure log shows its LineageOS 16 fingerprint Sony/lineage_pdx235/pdx235:16/BP4A.251205.006/9d767828e9:userdebug/release-keys. odm, vendor, and vendor_dlkm already have matching contiguous ranges; product, system, and system_ext additionally differ on sparse ranges.

    Interpretation: This is a strong upstream source-code explanation for the Sony care-map fingerprint (intentional stock-fingerprint override) and the range differences (different construction methods). It is not proof that the Oct 9 archive's target-files and signed partition images were produced using precisely these inputs; nor does it establish that substituting a new care map is safe for update_verifier. The exact, signed-archive discrepancy must still be reconciled.

    Original ZIP and release evidence: located, NOT independently authenticated

    • Failed CI job 114067106505 downloaded https://mirrorbits.lineageos.org/full/pdx235/20261009/lineage-23.2-20261009-nightly-pdx235-signed.zip and passed the implementation's official builds API SHA-256 delivery check. The response/digest is not recorded in the log. avbroot verified its whole-file signature against the certificate extracted from that same ZIP, so this does not independently authenticate the publisher.
    • A public mirror's 20261009 directory currently lists the exact ZIP (approximately 1.1 GiB) and build-manifest.xml (279 KiB), both timestamped October 9. This is a recovery lead, not a trusted digest or signature. The current investigation environment cannot retrieve binary ZIP or XML manifest bytes. The original failed GitHub Actions run exposes no retained artifacts.
    • The independent official signing trust root is LineageOS/update_verifier/lineageos_pubkey at 9ffcf56a0fe152467da2971f0e6b2b79a42f7890, SPKI DER SHA-256 0848ca38cb604887cea216a7f39d7dd62bad20b513cfdeff357ece4e731d6372. The official build-verification instructions require running python3 update_verifier.py lineageos_pubkey <original.zip>. This has NOT been executed on the October 9 ZIP; no publisher-signature verification result or independently measured ZIP SHA-256 exists in this review.

    Required evidence and upstream questions

    1. Recover the exact original ZIP, preserve its bytes, record full SHA-256 and preferably the contemporaneous official builds API response; compare identities. Run the official LineageOS update_verifier.py with the separately pinned lineageos_pubkey, require success. Do not trust the ZIP's own certificate as a publisher trust anchor.
    2. Recover the accompanying dated build-manifest.xml and the actual target-files ZIP from the LineageOS build infrastructure; record their provenance. Compare six *.build.prop files' ro.<partition>.build.fingerprint, META/misc_info.txt, META/care_map.pb where present, signed OTA care_map.pb, payload metadata, and partition AVB com.android.build.<partition>.fingerprint property descriptors. In particular test whether the upstream stock BuildFingerprint override explains the original care map byte-for-byte; examine why AVB descriptors use LineageOS 16.
    3. Ask LineageOS pdx235 device/build maintainers whether the fingerprint override and care-map/AVB divergence are intentional and operationally safe for update_verifier; ask avbroot upstream (reference #639) whether the exact-equality check should account for AOSP sparse ranges and per-partition fingerprint overrides, while maintaining meaningful verification. Provide authenticated artifacts/minimal reproducible evidence, not assumptions.
    4. Only after authoritative source-authentication/provenance review: nonpublishing strict re-verification, then separate physical first-boot and OTA-persistence testing. Do not fill _SOURCE_CARE_MAP_PDX235_20261009_SHA256, activate --allow-source-care-map-mismatch, merge the unverified feature into main, or publish an OTA. PR security: fail closed for unreviewed pdx235 source care maps (#113) #116's production pin must stay empty; helper PR chore(deps): update dependency chenxiaolong/custota to v5.1 - autoclosed #8's output checks are orthogonal to source trust.

    Disposition: Source fingerprint origin now has strong independent source-code support; original ZIP authenticity and exact artifact-level provenance remain UNRESOLVED. Keep #114 open and strict verification enabled.

  3. 0cwa commented on Oct 11, 2026

    @0cwa
    OwnerAuthor

    October 11 forensic addendum — immutable build paths + first-boot safety (no release authorization)

    Source-level fingerprint discrepancy now explained more precisely: The dated official LineageOS Android build commit e5aaa62172df0f321e68133fa30f42316376bfe8 (2026-09-24) shows that:

    • core/sysprop.mk writes ro.<partition>.build.fingerprint using BuildFingerprint from PRODUCT_BUILD_PROP_OVERRIDES. The upstream pdx235 device tree at 0559de69 supplies the exact Sony Android 15 value in the original care map.
    • core/Makefile independently constructs com.android.build.<partition>.fingerprint AVB properties from the normal BUILD_FINGERPRINT_FROM_FILE (LineageOS Android 16 in the CI reconstruction), for all six partitions.
    • add_img_to_target_files.py reads care-map fingerprints from partition *.build.prop and uses AOSP sparse meaningful-block ranges; ota_from_target_files.py copies META/care_map.pb into the OTA. Thus two metadata-generation paths credibly explain both fingerprint classes without implying the source archive is corrupt. We still lack the exact October 9 target-files and manifest to tie the source revisions to this signed build.

    Additional security risk: rebuilt care maps must match runtime properties, not merely AVB metadata. In LineageOS update_verifier.cpp, ParseCareMap() excludes any partition whose runtime GetProperty(id) does not match the care-map fingerprint; if all entries are excluded it returns false (No partition to verify). The boot caller then logs Failed to parse the care map file, skipping verification, and can proceed without per-block verification. Actual device behavior depends on dm-verity and snapuserd status, and the October 9 runtime props are not yet recorded.

    Implication: A patched OTA that passes strict avbroot ota verify after AVB-derived care-map regeneration may still have all six first-boot care-map entries skipped if its partition runtime props retain Sony Android 15. Do not enable the exception based solely on source key/digest authentication or on output signer verification. Add an independent output/runtime-property compatibility check, and perform physical first-boot and OTA-persistence tests before any approval.

    Recovery attempt status: Princeton's public 20261009 mirror directory continues to list the ZIP and build-manifest.xml, but attempts to retrieve bytes/official builds API from this environment fail at DNS (curl exit 6) or unsupported ZIP/XML content types. Exact OTA SHA-256, original public-key-verified signature, release API response, target-files, and signed-archive property comparisons remain missing.

    Release gate unchanged: Keep #114 open; leave _SOURCE_CARE_MAP_PDX235_20261009_SHA256='', source care-map exception disabled and strict checks active. Do not merge unverified #97 into main, build/publish an exception OTA, or infer trust from previously successful CI. No upstream maintainers were contacted.

  4. 0cwa commented on Oct 11, 2026

    @0cwa
    OwnerAuthor

    2026-10-12 forensic verification milestone — exact October 9 source publisher-authenticated

    Independent whole-file signature and official release metadata: PASSED. A dedicated nonpublishing forensic run 38180526270 downloaded the original lineage-23.2-20261009-nightly-pdx235-signed.zip from the official LineageOS mirror and ran pinned upstream LineageOS/update_verifier at commit 9ffcf56a0fe152467da2971f0e6b2b79a42f7890 against its independently supplied lineageos_pubkey. Official verifier output: verified successfully (exit 0). PEM key SHA-256 07ba6d6614f6c118bb448a662d0777d7cf894a4096490b5bd062566055125290, SPKI DER SHA-256 0848ca38cb604887cea216a7f39d7dd62bad20b513cfdeff357ece4e731d6372.

    • Exact signed OTA SHA-256: b2b4269738df1d3092de49bde6c7d0ffcb62ae19b342213a819187c30b7ab1f8.
    • File size: 1,232,106,897 bytes.
    • Authoritative LineageOS builds API: returned the same exact ZIP SHA-256, byte size, source download URL, release type nightly, version 23.2, and datetime=1791520682. The separately downloaded ZIP's digest/size matched.
    • Original signed care-map SHA-256: e903a2c0e4a71aaadf757d7d479a6be2ca80939a084c1832796db61c75e6e147, decoded from the independently authenticated ZIP. Its six names, stock Sony Android 15 fingerprint IDs/values, and sparse ranges agree exactly with the Oct 9 failed CI log.
    • Signed OTA metadata: post-build=Sony/XQ-DC72/XQ-DC72:15/68.2.A.3.107/068002A003010700521798587:user/release-keys, post-build-incremental=9d767828e9, post-sdk-level=36, post-security-patch-level=2026-10-01, ota-type=AB.
    • Signed OTA payload manifest SHA-256: d580240e74fef8e0a7fe735276984fad516018834556c47a5676e0e7aee51f78, block size 4096, full update with 11 partitions. The manifest's new_partition_info.hash and size agree exactly with the official builds API entries for boot.img, dtbo.img, vbmeta.img, and vendor_boot.img.

    Exact release build-manifest recovered and matched to the official API. The follow-up successful run 38180766827 downloaded build-manifest.xml via the official mirrorbits endpoint and verified SHA-256 25ce9014292665e0d9887fce8dcafd67a25585e4f5a3f73f7c3865a1392e051e, size 285,724 bytes, against the official API. Its immutable project revisions include:

    • device/sony/pdx235 0559de69466e1e9af69b121985f592c4f4609283 (exact Sony Android 15 BuildFingerprint override).
    • build/make e5aaa62172df0f321e68133fa30f42316376bfe8 (care map from per-partition build props, separate AVB BUILD_FINGERPRINT_FROM_FILE path, sparse ranges, and OTA copy).
    • build/soong 9aa045a2aef10b8089e32e847fed26d9aa3d61be.
    • bootable/recovery acd99f7c48e0229eaedd40aa3e2adcbabe2dc283.
    • device/sony/sm6375-common af6d25a714d6868cb4ddd7d0f373aefc5c31c5b6.

    Output/runtime safety remains BLOCKING: The exact target-files ZIP (META/misc_info.txt, finalized *.build.prop), extracted six-partition payload AVB descriptors in the independent forensic environment, and actual boot-time ro.<partition>.build.fingerprint observations remain unreviewed. Source-code paths strongly support a legitimate Sony care-map fingerprint versus AVB Lineage fingerprint split. However, update_verifier skips care-map partitions when runtime fingerprints differ; replacing all six with AVB-only identities might cause first-boot care-map verification to be skipped even when the regenerated OTA passes strict avbroot verification. Maintain strict source verification and require output runtime-property/first-boot safety before compatibility exception approval.

    Dormant format bug found and fixed: The exact source's META-INF/com/android/otacert is PEM, while prior gate incorrectly parsed it as DER. PR #118 added PEM-based independently pinned signing-public-key matching and negative tests; CI 38180846239 passed. It was merged only into the existing feature branch as adbf48c647d20ef384a469c4ebf3b53a1bc5a977. Production _SOURCE_CARE_MAP_PDX235_20261009_SHA256='' remains empty.

    Disposition: Publisher authenticity and release source-revision provenance established. Care-map runtime compatibility is not established. Keep #114 open; do not fill the production digest pin, activate the source exception, merge #97 into main, or build/publish an exception OTA. No upstream messages were sent, and no patched OTA was built or published. Read-only evidence artifacts are retained in the forensic run; the 1.23 GB original ZIP was not uploaded/published.

  5. 0cwa commented on Oct 11, 2026

    @0cwa
    OwnerAuthor

    2026-10-12 — Signed source partition fingerprint provenance independently confirmed; runtime care-map compatibility still BLOCKED

    New original-signed-image evidence: In isolated read-only CI run 38181164127, the exact authenticated October 9 ZIP was downloaded again, its SHA-256 b2b4269738df1d3092de49bde6c7d0ffcb62ae19b342213a819187c30b7ab1f8 independently checked and publisher-authenticated again using upstream LineageOS/update_verifier with the pinned published LineageOS public key. The avbroot 3.34.1 executable bytes were verified against the ModOS existing executable lock before extracting the original signed payload partition images. No ZIP, system image, OTA or signed artifact was modified or published.

    For each of odm, product, system, system_ext, vendor, and vendor_dlkm, both independent sources were inspected directly inside the ORIGINAL payload image:

    Source within the very same signed partition image Observed value
    Actual filesystem ro.<partition>.build.fingerprint (read via debugfs) Sony/XQ-DC72/XQ-DC72:15/68.2.A.3.107/068002A003010700521798587:user/release-keys
    Its AVB com.android.build.<partition>.fingerprint (read via avbroot avb info) Sony/lineage_pdx235/pdx235:16/BP4A.251205.006/9d767828e9:userdebug/release-keys
    Actual filesystem ro.<partition>.build.version.release 16
    Actual filesystem ro.<partition>.build.type userdebug
    Authenticated source OTA's care_map.pb fingerprint matches the filesystem fingerprint in all six partitions
    avbroot 3.34.1 reconstructed fingerprint from original signed AVB descriptor matches the AVB fingerprint, disagrees with filesystem fingerprint in all six partitions

    The extracted image SHA-256 values logged for all six partitions also match the original publisher-authenticated OTA's signed payload manifest:

    • odm: 3364fade0f314f7598e6acd06dca65ae943bdfc5113ee42413753625d12b6497
    • product: 7d0d2327d3e0099fa19fc07f4d5a5ceb92c7807757d4b402a33989386226f2cd
    • system: e47686704a35acb3a38f262227437e4e3be9f21ebb22e9646aa45a174f497985
    • system_ext: 6435cd45daf483df3116ba899e257a44bb1b787dcf2936bd5ed31c1adc44901a
    • vendor: 5eac98c82259e7395418a569d87e3431eaa646563883257811e10deb27f7fb9a
    • vendor_dlkm: 95fc33431bd257adcfc80c39079a49e8c17569d5cde3eb4e81ea6f083b1e4ba1

    This resolves the causal fingerprint discrepancy at the original signed-image byte/metadata level, not just via plausible build-system source analysis. Both identities coexist intentionally in authentic Android 16 images. The original AOSP-generated care map's Sony fingerprint is consistent with actual filesystem build props, not stale Android 15 image content. Sparse vs contiguous range semantics still need safety review.

    Critical remaining boot semantics: The same release's exact manifest-pinned bootable/recovery/update_verifier.cpp at acd99f7c48e0229eaedd40aa3e2adcbabe2dc283 reads runtime ro.<partition>.build.fingerprint, excludes every partition with a mismatch, and can skip care-map block verification when the verified partition map is empty. A regenerated map using the LineageOS AVB property value is therefore at risk of excluding all six partitions, unless actual runtime props/verity/snapuserd conditions demonstrate otherwise. Merely making output pass avbroot strict care-map equality does not establish Android's runtime block-verification behavior.

    Proven elsewhere and unchanged: Original ZIP official API SHA-256/size and independent LineageOS publisher-signature verification passed run 38180526270; original build-manifest official API hash matched in run 38180766827, recording the exact 2026-10-09 release revisions. The original target-files ZIP/META/misc_info.txt, physical runtime getprop, and end-to-end nonpublishing/physical OTA-persistence acceptance remain unavailable/incomplete.

    Repository housekeeping: Isolated forensic draft PRs #117 and #119 were both reset to the feature branch head and closed without merge; temporary diagnostic workflow changes no longer reside on the feature branch. PR #118 corrected the PEM-versus-DER parsing of the source OTA signing certificate and passed CI; its feature-branch-only merge left the original production exception digest pin empty.

    Disposition: KEEP #114 OPEN. Do not fill _SOURCE_CARE_MAP_PDX235_20261009_SHA256, activate --allow-source-care-map-mismatch, merge unverified feature PR #97 into main, or publish any OTA. Independent source authenticity is resolved; output runtime equivalence remains unproven and potentially unsafe. No upstream maintainers were contacted.

  6. 0cwa commented on Oct 11, 2026

    @0cwa
    OwnerAuthor

    2026-10-12 — Fail-closed regenerated care-map fingerprint guard integrated into feature branches

    Why: The original October 9 pdx235 OTA was independently authenticated, and examination of the original publisher-signed partition images established that all six actual filesystem ro.<partition>.build.fingerprint values agree with the Sony Android 15 source care_map.pb, whereas AVB com.android.build.<partition>.fingerprint values are LineageOS Android 16. Strict avbroot ota verify of a regenerated output can therefore succeed on an AVB-derived map that Android update_verifier could ignore because its runtime fingerprint comparison fails.

    New helper protection (my-avbroot-setup PR #9, merged only into feature/locked-microg-20260919 as 70819893ff91488b4b40a8c6f45fde39854c6c68):

    • Only after the explicit source care-map exception actually fires, keep strict output signature/AVB/partition verification against the configured signing key and OTA certificate.
    • After each output's strict verification, parse the authenticated source and regenerated output care_map.pb with a bounded fail-closed protobuf parser.
    • Reject missing/extra/duplicate partitions, malformed entries/ranges, or changed ro.<partition>.build.fingerprint identity. Require identical source/output partition names, fingerprint property IDs and fingerprint values for both rootless and Magisk outputs. Range encoding may differ but avbroot's separate strict output verification still applies.
    • The check precedes generation of Custota sidecars and locked patch reports. Regression tests cover the known AVB-only substitution, malformed protobuf, wrong/missing output, and a secondary-output identity failure suppressing sidecars. Helper CI 38182023902 passed.

    ModOS pin updated (PR #120, merged only into feature/microg-pdx235-20260919 as b38377c6fd8803f8722e834e68076aa181e5d41b):

    • Exact helper commit 70819893ff91488b4b40a8c6f45fde39854c6c68.
    • Independently recalculated patch.py SHA-256: 3405142ea7d056a37ac5f7ff9301fff5d4fef8519ab63f32333ff2574fd387c5. The SHA-256 implementation was cross-checked against the previous known-good patch.py digest before advancing the compatibility manifest.
    • All existing pin assertions updated, no additional workflows. ModOS CI including strict helper compatibility 38182179606 passed.

    Limitation: necessary, not sufficient. This preserves the original authenticated source's known runtime fingerprint identities, but does not itself prove that output filesystem properties are unchanged by a patch, that AVB-generated care-map ranges are safe on the actual filesystem, or that dm-verity/snapuserd and first-boot update_verifier execute the intended checks on physical pdx235. With current avbroot 3.34.1, the known six changed fingerprints should cause fail-closed rejection, not an implicitly authorized OTA. No verification is disabled or bypassed.

    Disposition unchanged: Keep issue #114 open. _SOURCE_CARE_MAP_PDX235_20261009_SHA256='' remains empty, so production source compatibility cannot activate even though original publisher authenticity is proven. Feature PR #97 remains unmerged into main and helper PR #7 remains unmerged into master. No signed output OTA or publication was initiated; no upstream maintainers were messaged. Next prerequisite is a correct care-map generation/semantic verification design that respects actual output partition properties, followed by nonpublishing test builds and physical-device boot/OTA-persistence acceptance.

  7. 0cwa commented on Oct 11, 2026

    @0cwa
    OwnerAuthor

    Worklane 3 conformance interface available — test-only draft my-avbroot-setup #10

    The isolated test branch targets feature/locked-microg-20260919@70819893ff91488b4b40a8c6f45fde39854c6c68 and adds only test/fixture/evidence files. Its 1,086-byte reconstructed signed-source care_map.pb SHA-256 is e903a2c0e4a71aaadf757d7d479a6be2ca80939a084c1832796db61c75e6e147; six actual names, fingerprint IDs, stock Sony filesystem identities, and historical sparse ranges are included. The avbroot v3.34.1 AVB-derived contiguous/range mismatch is independently reconstructed from the historical CI evidence without implying corrupted signed bytes.

    Worklane 1 test contract: modos-care-map-conformance-v1 spec + coverage matrix. A test-only CLI adapter accepts one JSON request on stdin and returns one {name,ranges,id,fingerprint} JSON object on stdout. Input includes original_image_size, block_size, ordered RAW/FILL/DONT_CARE synthetic chunks (or null for nonsparse), independently supplied output-filesystem ro.<partition>.build.fingerprint property/id, and deliberately divergent AVB fingerprint. Invalid inputs exit nonzero. Run CARE_MAP_CONFORMANCE_REQUIRE_CANDIDATE=1 CARE_MAP_CANDIDATE_CMD='/path/to/adapter' uv run python -m unittest tests.test_care_map_candidate_contract -v once the candidate exists. Missing adapter then errors rather than marking integration green. No request to modify Worklane 1 production code for the adapter.

    Evidence: Existing offline CI 38185442489 is green: 256 tests, one explicit candidate-not-wired skip, with 30 new conformance tests passing. External output cryptographic verification is mocked in the small fixture tests; physical/runtime checks and a fresh full-OTA avbroot execution are NOT claimed. Current helper identity guard does not itself read rebuilt ext4 runtime props or enforce output-image bounds, so those remain mandatory Worklane 1/release-side gates. A separate nonpublishing full-OTA forensic harness is provided, but normal CI remains offline and unchanged.

    No production digest pin, source exception activation, merge, OTA generation or release authorization.

  8. 0cwa commented on Oct 11, 2026

    @0cwa
    OwnerAuthor

    Worklane 1 handoff — provenance-aware care-map proposal (2026-10-12; NO release authorization)

    Reviewable draft, documentation-only helper RFC: my-avbroot-setup PR #11. It is confined to research/worklane1-care-map-provenance-20261012 and must not be merged to production. A separate standalone research prototype passed 24 synthetic tests locally, but has not been run against real partition images or built/tested in Rust.

    Root cause (v3.34.1): avbroot selects care-map fingerprints from AVB com.android.build.<partition>.fingerprint and emits contiguous ranges. The original authenticated 2026-10-09 OTA's care-map properties match the signed filesystem ro.<partition>.build.fingerprint (Sony A15), not its AVB descriptors (LineageOS A16). AOSP sparse care ranges describe RAW/FILL chunks excluding DONT_CARE; they are not reconstructible in general from raw expanded images.

    Narrow proposed proof: Let S_p be the independently publisher-authenticated original care ranges and D_p all protected 4096-byte filesystem blocks changed in final vs source image. Reuse the original sparse ranges only if D_p ⊆ S_p, image layout and verified AVB hashtree/footer original size agree, SHA-256/size match independently verified signed manifests, source map runtime fingerprints match the original actual filesystem, and final map fingerprints match the final actual filesystem. Validate all range boundaries, IDs, partition lists and the entire candidate map. Fail closed for changes outside authenticated ranges, absent properties, unsupported filesystems, or resized extents. Never replace sparse ranges with 2,0,N merely to satisfy avbroot's old verifier.

    Worklane 3 interface: provide source signed map and dynamic manifest list; manifest-verified original raw partition images and digests/sizes; exact AVB hashtree/footer data sizes and canonical read-only build.prop paths; retain identity-preservation guard. Collect physical effective runtime getprop, update_verifier, dm-verity/snapuserd and slot-success evidence before approval.

    Worklane 4 interface: perform provisional proof on patch-time images, then independently verify the configured output OTA cert and AVB key, payload signature and hashes, AVB chain, recovery cert and flags; re-extract signed-output images and repeat image/diff/property proof against the independently authenticated source. Reject before any sidecar/report/publication on failure. Source digest pin remains disabled.

    Exact integration blocker: stock avbroot v3.34.1 ota verify still recomputes its wrong AVB/contiguous expectation, so it cannot strictly verify an output with a corrected fingerprint/sparse map. A new opt-in, provenance-aware expected-map verifier (in isolated avbroot development or equivalently independently verified boundary) is required. No blanket ignore flag, production pin edit, OTA build, upstream contact or release has occurred. Keep #114 open.

  9. 0cwa commented on Oct 12, 2026

    @0cwa
    OwnerAuthor

    Worklane 3 continued — signed-source property-path evidence + D⊆S conformance (2026-10-12)

    Independent test-only draft my-avbroot-setup PR #10 has been expanded on its own isolated branch, targeting the unchanged helper base 70819893ff91488b4b40a8c6f45fde39854c6c68. Exact diff: 12 new test/fixture/evidence files only, zero production or workflow modifications. Current head b4882de356a1eaf59d1764529ef35ce90374256e.

    CI: run 38187232104 succeeded: 287 tests, two explicitly skipped not-yet-wired Worklane 1 adapters, zero failures. Neither adapter's implementation nor a real signed OTA was tested; cryptographic strict-output verification in the small helper tests was mocked. Missing provenance candidate becomes a hard error under CARE_MAP_PROVENANCE_REQUIRED=1.

    New independent coverage: six-partition synthetic D_p ⊆ S_p range-preservation oracle aligned with Worklane 1 RFC #11, including changed protected sparse holes (reject), changed within authenticated ranges (conditionally accept), signed-manifest image SHA/size binding, AVB hashtree/footer data extent equality, no output property/partition/range drift, no AVB-only fingerprint substitution. Real original 1,086-byte signed care map and exact high sparse half-open endpoints retained and checked.

    New original-image evidence: the six actual signed image build.prop read-only debugfs paths are now pinned from authenticated forensic job 114593834255: odm/product/system_ext/vendor_dlkm /etc/build.prop, system /system/build.prop, vendor /build.prop. Their six signed payload image SHA-256 hashes remain pinned. A separate optional nonpublishing, read-only hash-first original-image inspection harness was added; normal CI tests it using tiny files and mocked debugfs. This does not claim output/ext4 or physical runtime attestation.

    Critical release blocks persist: avbroot 3.34.1's wrong AVB/contiguous expected-map computation still needs a proof-backed correction rather than mismatch waiver; actual final signed output images/properties must be re-extracted and verified for both rootless and Magisk. Worklane 4 draft #121 identifies cached outputs_ready reuse bypass risk: require fresh post-sign verification or no reuse. Effective pdx235 runtime getprops/update_verifier/verity/snapuserd and OTA persistence remain untested. Source digest pin stays disabled; no merge, OTA build, publisher contact or release authorization.

    Complete matrix and interface handoff.

  10. 0cwa commented on Oct 12, 2026

    @0cwa
    OwnerAuthor

    2026-10-12 — Worklane 1 executable prototype + independently required W3 conformance (RELEASE STILL BLOCKED)

    Research code now in isolated helper draft PR #11 at b5494594784bd8e4a72078929a2901913c6f0458. Scope: 5 new research/design/test files only, zero production helper modification. The research-only code computes output fingerprints from actual ext4 build.prop when provided verified image files, preserves authenticated source sparse ranges subject to all changed 4096-byte protected data blocks D⊆S, validates signed-manifest-supplied image digests/sizes and hashtree/footer data extents, requires exact output map ID/range/fingerprint equality, and offers two synthetic JSON candidate protocols to W3. Synthetic mode accepts mock digest/property values and must not be used as an authentication or output verifier.

    • 54/54 local W1 synthetic tests passed, including byte-exact 1,086-byte signed source map reconstruction (SHA-256 e903a2c0e4a71aaadf757d7d479a6be2ca80939a084c1832796db61c75e6e147). W1 CI #38187927075 success: 242 tests total.
    • Independent Worklane 3 required-mode candidate suites PASSED on ephemeral isolated draft integration PR #12 at aa1b4e4278aae0736eee2179f3db0e2fc6bfc4d1. Its base is unmodified W3 test branch at b4882de356a1eaf59d1764529ef35ce90374256e; diff only W1's two research modules and a required-mode test wrapper. CI #38188261521 success: 11 nested independent W3 candidate tests, both required flags active, zero nested skips. Outer CI had 288 tests and 2 expected default-mode candidate skips. This only establishes synthetic protocol conformance, not any real-device or signature-security guarantee.
    • Potential functionality blocker discovered: the pinned helper's microG module actually writes additional GmsCore/Companion APKs and XMLs into product ext4. The publisher-signed source product care map has wide DONT_CARE gaps; newly allocated product blocks may lie outside S_product, so the conservative proof must reject that output until an independently verified broader meaningful-range plus deterministic OTA block-write/verity argument is established. No genuine output images were generated/inspected; this remains a conditional risk.

    Absolute blockers unchanged: no corrected Rust avbroot v3.34.1 ota verify expected-map implementation, no signed-output payload/AVB whole-image postflight or real ext4 property validation, no physical effective runtime getprop/first-boot/update_verifier and OTA persistence, no approved handling of newly allocated blocks. The current avbroot AVB-fingerprint/contiguous verifier is incompatible with correct maps. A strict alternate provenance-aware expected-map verifier is needed; do NOT simply ignore mismatches. Keep _SOURCE_CARE_MAP_PDX235_20261009_SHA256='', output quarantine, independent publisher/signing validation, and issue #114 open. No production tool pin advanced, helper feature branch merge, OTA build/sign/release or upstream maintainer contact occurred.

  11. 0cwa commented on Oct 15, 2026

    @0cwa
    OwnerAuthor

    2026-10-15 pdx235 publishing-path audit — release quarantine is intentional (no publication authorized)

    The redundant integration draft #122 was closed as superseded by merged #97. Git ancestry confirms the #122 head b8b2a1e is already an ancestor of main. This cleanup does not change any release gate.

    Why scheduled ModOS pdx235 public releases are absent: On current main, .github/workflows/release-lineage.yml specifies 0 */6 * * *, loads .github/schedules/lineageos-pdx235.toml (nightly, both root modes, microG selected), and calls the shared builder with release-type: default and publish: true. .github/workflows/build-rom.yml therefore resolves OUTPUT_SCOPE=published. The early enforce_output_policy calls enforce_pdx235_care_map_quarantine, which explicitly refuses any LineageOS/pdx235 scope other than local-unpublished; both direct metadata publishers independently repeat the refusal. Thus a scheduled publication cannot succeed under the current checked-in code, regardless of successful private build-only artifacts or microG selection. The published-release listing and exact scheduled run history were not independently enumerated in this audit; this is a source-path conclusion.

    Additional non-release-ready boundaries: The permanent production source exception digest _SOURCE_CARE_MAP_PDX235_20261009_SHA256 remains empty. The reviewed repository-key private builds and the fixed October 9 rootless candidate establish real artifact proof and a physical first boot, not all release acceptance. The October 13 repair record confirms Android boot with locked/yellow verified boot, verity and SELinux enforcing, but successful-slot/snapshot/first-boot verification evidence, actual microG account/push/location and current pinned 0.3.17 firmware behavior, Magisk physical acceptance and a Custota A/B OTA-persistence cycle are still not fully established.

    New source-version boundary: The normal LineageOS provider selects the latest nightly rather than the explicitly frozen October 9 fixture selected by the care-map-pdx235 development profile. The documented October 12 provider source was LineageOS 24.0 / Android 17 / SDK 37, whereas verified October 9 acceptance concerns 23.2 / Android 16 / SDK 36. Do not assume the old fixture results authorize automatic publishing of a newer source. The approved 3.35 guarded source-helper qualification retains the vendor-signed 3.34.1 executable lock because the prebuilt 3.35 verifier has an unsigned-explicit-key acceptance regression.

    Disposition: Maintain the quarantine. For a future release, independently qualify the actual newly selected signed source, patched rootless and Magisk outputs, runtime versions and properties, and physical update/persistence; record an explicit security review and only then propose a separate narrowly scoped quarantine-lift PR. Consider making quarantined schedules explicitly skip publishing or perform controlled build-only diagnostics, rather than attempt a known-prohibited published scope. This audit neither dispatched a workflow nor installed/published an OTA.

  12. 0cwa commented on Oct 15, 2026

    @0cwa
    OwnerAuthor

    2026-10-15 correction to publication audit: avbroot 3.35 and rootless-first scope

    Correcting the prior audit: The unsigned-AVB-header/explicit expected-key acceptance is not a 3.35 regression. Compared upstream chenxiaolong/avbroot tags v3.34.1 and v3.35.0; avbroot/src/cli/avb.rs has identical blob SHA 10da0eedbaaceff1e9a89245ffbd283473b9e642 and avbroot/src/format/avb.rs identical blob SHA a48604adc24689d0d64b9fbf0da9b55ccf546807. Header::verify() returns None for AlgorithmType::None and verify_headers() logs the unsigned image without rejecting a supplied trust key. This behavior predates 3.35. The security invariant still matters: ModOS's postflight must explicitly reject unsigned root/chained AVB headers when a key is required, not interpret the vendor binary's zero exit code as proof of expected signatures. The guarded 3.35 source-engine helper was already merged via #130; independently authenticating and promoting the official 3.35 executable lock is distinct from that helper change, not inherently disqualified as a regression.

    Owner-reported functional acceptance: The device owner confirms microG works and Custota works on the installed device. Take these as positive runtime evidence; do not demand gratuitous re-testing of those features on the already running build. Specific evidence for an end-to-end future OS/OTA transition should be scoped to that transition rather than silently inferred from general functionality. The owner elects to defer Magisk testing and prioritizes keeping rootless pdx235 devices updated. Propose rootless-only initial scheduled publishing; do not make Magisk physical acceptance a prerequisite for rootless-only delivery. This is not permission to publish currently unreviewed artifacts.

    Precisely what “source verification” means here: “Source” means the original official LineageOS OTA ZIP chosen by src/ota_providers.sh, not the ModOS source tree. In the already documented baseline it was lineage-23.2-20261009-nightly-pdx235-signed.zip; the provider subsequently returned lineage-24.0-20261012-nightly-pdx235-signed.zip (Android 17/SDK37) in the October 13 build record. The API may offer a later build when the next run executes. Routine per-build checks should be automatic and actionable: (1) record the chosen ZIP filename/version/official API SHA-256 and verify downloaded bytes match; (2) cryptographically authenticate the ZIP signature against the independently pinned LineageOS public key; (3) verify that its partition metadata/care-map fingerprints agree with the actual images, and that the patched output's regenerated map and protected-block coverage are valid; (4) verify the finished rootless OTA against the configured OTA cert/AVB key with explicit fail-closed signed-header checks, and confirm non-disabled vbmeta flags. The main code already implements much of this, but the production care-map exception remains empty and the current source-builder/profile access is restricted to local-unpublished. Determine the minimal correct production path rather than treating a frozen October 9 development fixture as authorization for all subsequent nightlies.

    Since Android 17 is a major-version change, separately validate its actual build's compatibility, rather than requiring a new open-ended forensic investigation for every future nightly. Keep quarantine until a reviewed, explicit rootless-only publication policy change is backed by a qualifying artifact and update path. Do not alter keys, force-publish, enable Magisk, or remove quarantine as part of this informational comment.

  13. 0cwa commented on Oct 15, 2026

    @0cwa
    OwnerAuthor

    2026-10-15 LineageOS 24.0 development qualification milestone (no release authorized)

    Draft #131 now supports the lineage-24.0 nightly series for pdx235, preserves optimized ROOT_MODE=both, and independently pins the publisher-signed official avbroot 3.35 executable archive + exact member hashes. The prior unsigned-explicit-key acceptance is unchanged from 3.34.1; the guarded source-built expected-care-map verifier remains in use for the impacted source class.

    Real, nonpublishing paired OTA proof: Actions run 38743923150 succeeded on lineage-24.0-20261012-nightly-pdx235-signed.zip (official LineageOS API-selected nightly). It built rootless and Magisk v30.7 from one shared filesystem preparation with disposable signing keys. Both passed helper strict source/output signature, payload, signed AVB header and expected care-map verification, six partition filesystem fingerprint proof, Magisk/rootless boot separation, and output boot animation checks. The original OTA's vbmeta had flags 0x3; output verification showed signed root/chained headers. Release and publisher steps were skipped.

    Production path being qualified: A follow-on in the draft routes ordinary development-profile=none for authenticated pdx235 LineageOS 24.0 nightlies only through that same guarded source-built verifier. It requires an API-digest-shaped source record, actual ZIP digest matching the official API, the independently pinned original OTA signer, and authenticated extracted images; wrong ROM/device/channel or newer series do not match. The unconditional #114 publication quarantine remains intact. Current CI is also exercising that ordinary route with disposable keys; do not consider it passed until its actual live run completes.

    Private installable candidate preparation: The separately explicit care-map-pdx235-lineage24 profile can now be selected by the existing manual build-only workflow with configured repository signing keys and optional private artifact export. Both UI-side request gate and export script reject non-approved profiles, disposable signers, and publishing. No repository keys entered PR CI; no user device was flashed.

    The device owner confirms working microG and Custota on the prior installed build, and wants ROOT_MODE=both retained for build efficiency while deferring physical Magisk validation. Neither runtime claim substitutes for actual Android 17 first boot or an Android 16→17 rootless update test. Leave #114 open and the public release quarantine unchanged until a privately signed candidate is installed/tested and a separate publication-policy review explicitly authorizes release.

  14. 0cwa commented on Oct 15, 2026

    @0cwa
    OwnerAuthor

    2026-10-15 — Owner-authorized LineageOS 24.0 experimental public release dispatched

    The owner explicitly requested public test delivery for their own Custota installation and asked to keep optimized ROOT_MODE=both; physical Magisk testing remains deferred. This permits an experimental release attempt, not a claim of completed Android 17 on-device acceptance.

    • #131 merged into main: independent official avbroot 3.35.0 signed-executable lock and guarded runtime care-map path for LineageOS 24.0. Full nonpublishing paired disposable-signer build #38747818968 passed.
    • #132 merged as release(pdx235): 24.0 pilot, triggering release run #38761516870. It freezes the selected original publisher source to lineage-24.0-20261012-nightly-pdx235-signed.zip and official builds-API SHA-256 29cb72bbdf808807d27f834a889df7a96ad85cdeed42b45dfb6748cdf05c3820. The original ZIP is independently signer-verified by the maintained helper. Both signed output OTAs still require strict payload, AVB and care-map postflight before GitHub release or Custota metadata writes.
    • The publishing allowance is scoped to the explicit pilot workflow and that exact source identity; ordinary scheduled pdx235 publication remains blocked for later nightlies. No AVB/OTA signing key was changed or exposed to PR CI.
    • At last verification, GitHub Actions had queued the pilot run's schedule_config job and the new CI jobs were also queued. Publication has not been confirmed. Please use the actual release run and produced signed assets as evidence before treating the OTA as available in Custota.
    • Still open: owner first boot and OTA persistence on Android 17, any physical Magisk acceptance, and routine automatic later-nightly policy after successful testing. The owner reports microG and Custota working on their currently installed system.

    Keep #114 open until the explicit on-device upgrade result and subsequent publication policy review are recorded.

  15. 0cwa commented on Oct 15, 2026

    @0cwa
    OwnerAuthor

    2026-10-15 21:23 UTC — pdx235 Android 17 publication retry, exact archive source fix

    No October 12 pdx235 assets published yet. The first one-shot release workflow #38761516870 failed at the source-selection stage before signing, with Error: LineageOS returned invalid release metadata. A concrete source-parser defect was found: the exact source_filename path enforced the October 9 23.2 SHA-256 even when the pilot selected the October 12 24.0 source. The current builds API's historical retention is not guaranteed and should not be a hard dependency for this exact archived pilot.

    Merged #133 as release(pdx235): 24.0 pilot, triggering production-signed release run #38809442251. The selected input is now exactly lineage-24.0-20261012-nightly-pdx235-signed.zip retrieved over HTTPS from the FAU LineageOS archive, which lists the original ZIP. It must match the previously recorded historical official API SHA-256 29cb72bbdf808807d27f834a889df7a96ad85cdeed42b45dfb6748cdf05c3820. Independent publisher key authentication, strict source and both final output signatures, AVB, payload, care-map and file-system postflight remain required. This digest still must be checked against the downloaded bytes; the mirror listing alone is not trust. Later nightlies remain outside the publication allowance.

    Latest GitHub Actions check shows retry queued (no signed release/Custota pointer yet). The CI queue is also backed up; the retry was explicitly initiated under the owner's test-publishing approval. Do not report a successful public release until both GitHub OTA assets and gh-pages Custota metadata are independently observed. Physical Android 17 rootless boot is still not yet demonstrated; owner-reported microG and Custota on the prior installed system remain recorded.

  16. 0cwa commented on Oct 16, 2026

    @0cwa
    OwnerAuthor

    2026-10-16 — Published OTA confirmed; scheduled 24.0 releases restored

    Investigated latest owner report that releases were not appearing. The October 12 release is already publicly published: release 20261012 was created 2026-10-16 05:04 UTC by successful signed publishing run #38809442251, and contains rootless + Magisk OTA ZIPs, .csig signatures, and selection metadata. The checked-in gh-pages rootless/pdx235.json and magisk/pdx235.json each point to the respective October 12 release assets. The FAU pdx235 archive currently shows only October 9 and 12; there is no independently confirmed newer upstream pdx235 build yet.

    Actual bug for future releases: Schedule runs #38890820576 and earlier were skipped, because schedule_config and scheduled_build were guarded by a one-off push-commit-title predicate. Merged PR #134, commit 046154cc9df16f749e58507c0faacc77d63844f9. The normal six-hour LineageOS schedule now invokes the paired publish builder, selecting the newest official lineage-24.0-YYYYMMDD-nightly-pdx235-signed.zip from the LineageOS builds API and requiring its advertised SHA-256, independently pinned OTA publisher signature, guarded care-map AVB/payload verification, and full output signing/postflight. Future 24.1/25.x or unqualified sources fail closed. Existing signing keys, ROOT_MODE=both, and microG remain unchanged.

    PR CI #38892417239 passed. Merge triggered release run #38892562531: schedule_config passed and paired production builder began at last check; release output pending. The older October 12 signed assets remain available in GitHub Releases and the gh-pages pointer.

    If Custota on the device still does not show that release, it is a separate client-side/configuration problem not proven by repo publication state. Confirm that the app is configured to the repository's deployed rootless OTA server directory, not an old GitHub Pages path or the magisk flavor, then inspect Custota's check.log. Do not overwrite existing release metadata or claim a new build exists without official upstream source evidence. Physical Android 17 acceptance remains pending.

  17. 0cwa commented on Oct 17, 2026

    @0cwa
    OwnerAuthor

    2026-10-17 — Custota unable to see published OTA: physical device ID/feed mismatch fixed

    Owner provided Custota CHECK log from Oct 17 02:43 local time. Root cause now established from the actual runtime request:

    • Custota uses the Android OTA pre-device XQ-DC72, requesting https://0cwa.github.io/ModOS/rootless/XQ-DC72.json.
    • The stale gh-pages/rootless/XQ-DC72.json referenced 0cwa/PixeneOS 20260911 with security patch 2026-09-01. The installed device has security patch 2026-10-01, so Custota correctly refused to downgrade (ValidationException: Downgrading to older security patch is not allowed).
    • ModOS release workflows were updating only rootless/pdx235.json and magisk/pdx235.json, which the phone never requests. The October 16 ModOS rootless + Magisk release and canonical pdx235 feeds were already live and correct; the device-specific aliases had remained stale.

    Immediate repair: Pushed both gh-pages/rootless/XQ-DC72.json and gh-pages/magisk/XQ-DC72.json as byte-identical copies of their corresponding current pdx235.json canonical feed. All four now point to the already verified and published 20261016 ModOS release with appropriate paired OTA and .csig assets. Did not rebuild or resign OTA or change any signing keys, updater security policy or AVB checks.

    Permanent repair: Merged PR #135, commit 5a44b7489955da3bf0dcaeeffc0fee7171426d3e. Both single and dual publication helpers now stage the fixed Sony XQ-DC72.json alias from the exact verified canonical source, on each future published nightly. Strict scope to lineageos/pdx235 and rootless/magisk; reject unsafe paths/symlinks; archive, signed release asset, source and care-map verification unchanged. CI #38988392608 passed, including new alias tests, before merge. ROOT_MODE=both remains.

    User follow-up: No change to Custota server URL needed (https://0cwa.github.io/ModOS/rootless/). Retry CHECK after GitHub Pages deployment/cache refresh; it should request XQ-DC72.json and see the new ModOS release, not the old PixeneOS September 11 OTA. If not, inspect the new response and logs. We can confirm the gh-pages branch files have correct content but have not independently fetched the live github.io URL from this environment, so live CDN propagation remains to be checked on device. First Android 17 OTA installation still awaits owner acceptance.

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