Add option to recompress/rechecksum existing data on property change

[unixdruid]

Describe the feature you would like to see added to OpenZFS

Add a command or flag to rewrite existing data in-place when compression or checksum properties are changed, allowing datasets to be fully migrated to new algorithms without manual workarounds.

Proposed syntax:

zfs rewrite pool/dataset

Or as a flag:

zfs set -o rewrite checksum=sha256 pool/dataset

This would traverse the dataset and rewrite blocks with the new compression/checksum algorithms while preserving all metadata, timestamps, permissions, and attributes.

How will this feature improve OpenZFS?

Problem: Changing checksum from fletcher4 to sha256 (for example NOOPwrite) or changing compression algorithms only affects new writes. Existing data remains in the old format indefinitely unless manually rewritten.

Benefits:

• Enable NOOPwrite on existing datasets (fletcher4→sha256 migration)
• Migrate to better compression algorithms (lz4→zstd) without space penalties
• Improve data integrity by upgrading to stronger checksums
• Single command instead of complex workarounds

Similar features:

• btrfs: btrfs balance for compression migration
• bcachefs: automatic recompression on algorithm change

Additional context

Real-world scenario: I have a dataset with fletcher4 checksums and want to enable NOOPwrite (requires sha256). With ~168GB of data, current options are:

1. Wait indefinitely for natural rewrites
2. Double my storage for send/recv
3. Write custom scripts with find | xargs | dd

A zfs rewrite command would make this trivial and safe.

Implementation considerations:

• Should be interruptible/resumable
• Progress reporting (similar to zpool scrub)
• Option to rewrite specific properties only (checksum vs compression vs both)
• Preserve refcounts for snapshots/clones
• Could be background operation like scrub/resilver

Thanks,
UD.

[tonyhutter]

zfs rewrite is included in zfs-2.3.4 and zfs-2.4.0.