- We currently do no block accounting when blocks are written. Since
- allocation is not performed until blocks in the buffer cache are
- written to disk, it is possible to return success on a write, only
- to discover later that there is insufficient space get the block
- on disk.
-
- We intend to support multiple block sizes rather than fragments.
- This is not implemented.
-
- Since blocks are laid out contiguously, we can miss rotations reading
- sequentially. We need to read in contiguous blocks to avoid that.
- See McVoy's Winter 1991 Usenix paper for details on how to do that.