OpenZFS 2.4.4 widens Linux kernel support to 7.2, ships zhack mmp reclaim for stranded pools

OpenZFS 2.4.4 is a maintenance update that bundles platform-compatibility work with a long list of reliability fixes. The officially supported Linux kernel range now runs from 4.18 through 7.2, and the superblock path was ported to the kernel's sget_fc() interface to track that change. FreeBSD support remains at 13.3+ and 14.0+. A smaller but operationally meaningful change is that systemd-udev-settle is now an optional dependency for systemd pool import units, which removes an unconditional wait for udev to finish before an import unit can proceed.

The headline new tool is `zhack mmp reclaim`, aimed at pools stranded by Multi-Modifier Protection. MMP is OpenZFS's safeguard against the same storage pool being imported and modified on multiple hosts simultaneously, a situation that would otherwise risk corruption. Alongside the reclaim command, the MMP code now does a better job of distinguishing a failed uberblock claim from genuine activity on another host, skips non-writable vdevs during activity checks, and handles degraded mirrors with absent legs more cleanly. For an operator whose pool is stuck after a Multi-Modifier Protection failure, this is the first built-in recovery path.

Several changes affect pool-management scripts and automation. `zpool export` now returns EBUSY when zvol minor devices belonging to the pool are still in use, instead of silently proceeding with an unclean export. A race between device-removal completion and pool export was also fixed. The net effect for scripting is that export success or failure is now a more reliable signal, but any code that relied on the previous behaviour to detach a busy pool will need updating. On the delegation side, a new `send:encrypted` permission lets admins permit or deny sending encrypted datasets per delegation.

Storage-internals fixes are concentrated in the Deduplication Table: negative time overflows during pruning, multiple pruning bugs, insufficient locking during dedup verification, and incorrect reference-count tagging in the DDT log. L2ARC rebuilding during the first sweep is now bounded by the write hand, capping how much work that initial scan can perform. Direct I/O verification events are rate-limited to prevent excessive event generation under load. Reads through blocks freed after cloning are corrected, and the POSIX ACL cache no longer goes stale after a rollback.

Receive-path fixes include split large blocks with a short trailing chunk, `zfs receive -x` behavior, and a null-pointer dereference in the DMU receive code. Linux-specific fixes cover an infinite loop in `zfs_write()` on unfaultable buffers, `mmap()` reads beyond end-of-file, and timestamp updates when `fallocate()` grows a file. Operators on a supported kernel should treat 2.4.4 as a routine maintenance upgrade, but should audit any code that calls `zpool export` against the new EBUSY semantics before deploying.