pblk
pblk is a historical host-side flash translation layer that exposes open-channel SSD storage through a conventional block interface.
Mapping: pblk maps logical to physical addresses at 4 KiB granularity and handles garbage collection, I/O errors and recovery. It is a LightNVM target, not another name for the entire subsystem. [1]
Completion: In Linux v5.14, ordinary writes can complete after entering the host RAM buffer. Reads can return buffered data, so successful read-back alone does not establish persistence. A flush with outstanding data waits at a buffer flush point for media-write completion. [2] [3] [4] [5]
Capabilities: The v5.14 queue advertises write caching but no native FUA support. Do not infer native FUA handling from the existence of a flush path. [6]
Batching: In the FAST 2017 design, a flush before a full flash page is assembled adds unmapped padding. The amount depends on workload and media geometry, not just logical bytes written. [10]
Recovery: The v5.14 code rebuilds mappings in line-sequence order using end metadata or out-of-band metadata. Recovering that map cannot recover data that existed only in lost RAM. [7]
Lifecycle: pblk was removed with upstream LightNVM in Linux 5.15. The v5.14 configuration labels its disk format experimental; historical instructions are not current upstream support guidance. [8] [9]
Relations
Citations
- Linux v5.14: pblk Physical Block Device Target responsibilities.
- Linux v5.14 pblk-cache.c: pblk_write_to_cache and buffered completion.
- Linux v5.14 pblk-read.c: pblk_read_from_cache and address validation.
- Linux v5.14 pblk-rb.c: pblk_rb_may_write_flush and flush-point tracking.
- Linux v5.14 pblk-write.c: pblk_complete_write and pblk_end_w_bio.
- Linux v5.14 pblk-init.c: write-cache capability advertisement.
- Linux v5.14 pblk-recovery.c: line-sequence L2P recovery and metadata fallback.
- Linux v5.14 LightNVM Kconfig: experimental format and Linux 5.15 deprecation notice.
- Linux commit 9ea9b9c48387: remove the LightNVM subsystem, including pblk.
- Bjorling, Gonzalez and Bonnet. LightNVM: The Linux Open-Channel SSD Subsystem. FAST 2017, sections 4.2.1, 5 and 5.4.
Further reading
Sources & caveats
- Interpretation of version-pinned Linux v5.14 source, not a hardware power-loss experiment or a proof of every recovery path.
- EOL refers to the upstream Linux implementation, not all downstream forks or userspace open-channel software.