Background: 2-disk mirror pool (2x4TB WD Red Plus) on a ZimaBlade running TrueNAS SCALE 25.10.5 went offline after a brief upgrade-then-revert. zpool import has failed in every mode and flag combination for over a week, always with either “device unavailable” or “insufficient replicas — destroy and re-create.” Full diagnostic history (with a very generous assist from someone working from exact source references) is here:
https://github.com/openzfs/zfs/discussions/18954
Where things stand now, confirmed via source-level analysis:
-
SMART, labels, uberblocks, partition tables: all clean on both disks.
-
zdb -efully loads the pool’s trusted config (spa_load(..., config trusted): LOADED) — the pool structure and data are intact and readable. -
The reason
zpool importfails in every mode, including readonly and withspa_load_verify_metadata=0:spa_load_verify(spa.c:2831) unconditionally walks every dataset viadmu_objset_find_dpbefore import can succeed, and one specific object (a DSL props ZAP, object 293) fails checksum (ECKSUM/errno 52) during that walk. This is a hard block — no flag, module parameter, or rewind mode skips it.-F/-Xdon’t help either, since they re-enable full data verification. -
A full
zdb -bcsvLtraversal of the whole pool found only 8 checksum errors total, out of 29.2 million block pointers — and every one of them is confined to pool-internal bookkeeping (SPA space map, DSL props) or disposable.system/samba4and.system/netdata-*service datasets. None of my actual data datasets (Photos, Documents, Archived, etc.) show any corruption.
Current blocker: Since zpool import can’t succeed, the recommended path is zdb -B (backup stream extraction), which reads a dataset directly without needing a successful import. But every dataset I try — encrypted or not, parent or child, small or large — fails identically:
dump_backup: dmu_send_obj: Permission denied
running as root, 0-byte output, exit code 0. zdb -O/-r (path lookups) also fail, but with a different error (Input/output error), suggesting a different cause than the -B failure.
What I’m hoping this forum specifically can help with:
-
Has anyone seen
dmu_send_obj: Permission deniedfromzdb -Bon TrueNAS SCALE specifically? Wondering if this is a TrueNAS-specific build/patch difference from upstream OpenZFS, a SCALE permission/capability restriction, or something environment-specific to this appliance. -
Is there a TrueNAS-supported/documented path for exactly this scenario (pool won’t import, but data is confirmed readable via zdb) that I’m missing — official data-recovery tooling, a support channel, or a known workaround for
zdb -Bon SCALE?
Not time-pressured, happy to provide any further diagnostics. Two 4TB drives, ~3TB used, non-time-sensitive personal data — just trying to get it off before doing anything destructive to the pool.