This version of lessfs now comes with a ‘per hash’ lock which protects a hash / data chunk while it is still being written to disk. The way this is done does not effect performance because it is highly granular. This said, I should still check if using a spinlock instead of mutex locking improves performance. Other then that this code looks pretty stable.
For those who are interested in performance. The graph below compares XFS with ZFS-fuse and Lessfs-1.1.0-betaX On read performance Lessfs is outperformed by ZFS. But when it comes to writing, Lessfs is a clear winner. Lessfs will be tuned for metadata operations in the upcoming 2.x release. LFS now uses the high level fuse API but will switch to the lowlevel API for better metadata performance.
