Corrupting a ZFS File on Purpose

9 hours ago

ZFS normally prevents corruption, but developers sometimes need controlled corruption for testing self-healing and observing system responses.
Use `zinject` for quick corruption without understanding the underlying disk layout; it injects checksum errors into live pools.
For educational purposes, manually corrupt a file on a throwaway pool built from file-based 'disks' using `dd` and hex editors.
The process involves mapping a file to its DVA (Data Virtual Address), adjusting for ZFS's reserved 4 MiB area, and calculating sector offsets.
Compression must be turned off (`zfs set compression=off`) to ensure predictable block sizes and avoid confusion with compressed data.
In RAIDZ2, the block mapping involves splitting data across child devices with parity; manually calculating column offsets is essential for targeted corruption.
After exporting the pool and corrupting the backing file with `dd conv=notrunc`, importing and scrubbing reveals errors: single vdevs show permanent errors, RAIDZ2 heals via parity.

Hasty Briefsbeta