- Decision
- ADR-024
- Date
- 2026-08-01
- Status
- AcceptedAccepted
- Mentions
- ADR-021, ADR-022, ADR-023
- Mentioned by
- ADR-025
1Decision
Confirm the third and final cause of the ADR-021 mismatches: cache.img/userdata.img get no pinned timestamp because they carry no build.prop and USE_FIXED_TIMESTAMP_IMG_FILES was never set. All 18 original mismatches are now explained.
2Why and evidence
Source-verified in build/make/tools/releasetools/build_image.py's TryParseFingerprintAndTimestamp: for a partition with a build.prop, the image timestamp is set from ro.build.date.utc automatically (already effectively pinned once BUILD_NUMBER is fixed, per ADR-022) — explaining why the AVB-salt cause, not timestamp, dominates for those partitions.
For a partition with no build.prop (cache, data/userdata), timestamp is only set if use_fixed_timestamp is present in the image-property dictionary, which build/make/core/Makefile's generate-image-prop-dictionary only writes when USE_FIXED_TIMESTAMP_IMG_FILES (or COPY_IMAGES_FOR_TARGET_FILES_ZIP) is set — neither build set it.
Without it, mkuserimg_mke2fs.py never receives --timestamp, so mke2fs/e2fsdroid embed the real wall-clock time of that specific packaging step.
3Consequences
All 18 ADR-021 mismatches are now accounted for: 3 by unpinned BUILD_NUMBER (ADR-022), 12 by the random AVB/verity salt (ADR-023), 1 (super.img) as a cascade of those two, and 2 (cache, userdata) by this unpinned wall-clock timestamp. None indicate compromise or tampering. USE_FIXED_TIMESTAMP_IMG_FILES=true is a simple, single-flag fix (like BUILD_NUMBER) that requires no source changes.
The AVB-salt fix is not: BOARD_AVB_*_ADD_HASHTREE_FOOTER_ARGS is set per-partition across several device/board .mk files (confirmed not .KATI_READONLY, so a make-command-line override — not a source edit — can win over each :=/+= assignment), but doing this correctly requires finding and overriding roughly 6-8 separate variables (one per AVB-protected partition, plus whatever governs vendor_boot's add_hash_footer salt) without dropping their existing non-salt args (e.g.
--hash_algorithm sha256) — not attempted yet, real risk of a mistake if rushed before an ~8h run.
4Review when
The owner decides whether tonight's rebuild only pins BUILD_NUMBER + USE_FIXED_TIMESTAMP_IMG_FILES (fixes 5 of 18, leaves the AVB-salt family as a known-cause-but-still-differing residual) or also attempts the full per-partition AVB-salt override (targets a fully clean diff, more setup, more risk of a new mistake).
Read from docs/governance/decision-log.md, entry ADR-024.