On Wed, 9 Feb 2000, Matthew Wilcox wrote:
> 2.4:
> Collapsed indirect blocks [readonly] (willy)
>
> 2.5:
> Journalling (sct)
> Access to lownumbered inodes
> Dynamic inode tables (phil)
> Btree directories (phil)
> ext2 allocation page size greater than cpu page size (bcrl)
> 256k blocks with 1k fragments (viro)
> [also 8k blocks with 256-byte fragments!]
^^^
HUH??? You want to deal with sharing sectors between the different files?
> fine-grained locking
> [remove test_and_set_bit()]
> e2compr
requires VM modifications (and basically nothing else).
> nounlink attribute flag
may go into 2.4.early, FWIC.
> Some interesting mm/fs projects:
>
> 2.4:
> Move directories to page cache (bcrl)
> Address spaces (viro)
Variant submitted to Linus.
> Move fhandle <-> dentry conversion functions to VFS (viro)
> move silly_rename to VFS (viro)
> Investigate buddy allocator algorithms with more interesting properties (bcrl)
> bdflush may need tuning (bcrl)
>
> 2.5:
> Removal of buffer heads (bcrl)
> fix posix fcntl locks (willy)
> sort out interface to block devices (viro)
I still hope to get at least the interface parts into 2.4...
Other 2.4.early stuff:
caching ext2_find_entry() results in dentry (patch exists,
obviously correct and well-tested).
caching the position of last lookup and doing cyclic lookups from
that place (literal copying from VFAT, where it worked since the early
Summer; it's an old BSD optimization).
pre-alloc for directories (yup, right now it's _off_).
Mandatory 2.3.late stuff:
serialization between truncate and write. There are races...
