blob: f989c3a91818b1ba7dd8bb2fc3c26838c3ecf84d (
plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
|
The hfs_fs "to do" list.
------------------------
Items are broken down into groups and the groups are listed in order
from most important to least important. The items within each group
are not placed in any particular order. The order in which items are
listed probably doesn't correlate well with the order they will be
addressed.
Genuine bugs:
1. Header files have compiled-in limit (currently 10) on descriptors.
Missing features:
1. The partition code should be migrated into the kernel to allow
simultaneous access to multiple partitions on a single disk.
2. 1k block support is needed for some devices.
3. An ioctl()-based interface is needed to provide a consistent way
to do things under all of the representations of forked files.
Possible additional "fork" mount options:
1. AppleSingle.
2. The scheme MacOS uses on FAT disks (PC Exchange).
3. "Flat" (no resource forks or metadata).
Performance issues:
1. Use drAllocPtr to speed block allocations.
2. Keep a real cache of bnodes, rather than just a hash table of
the ones that are currently in use.
3. Keep a real cache of extent records, rather than just a linked
list of the ones that are currently in use and the one most
recently used. This is particularly needed to get acceptable
performance with multiple readers on a file. Perhaps simply
keep them in memory once they've been read until the file is
closed.
Implementation details:
1. Allocation scheme could/should be closer to that used by Apple.
2. B*-tree insertion could/should be closer to that used by Apple.
3. Magic-number checks on data structures are rarely done.
4. Error recovery is needed for failed binsert(), bdelete() and rename().
5. Deadlock detection is needed to make insert_empty_bnode() and
bdelete() less likely to hang on a corrupted B-tree.
6. Metadata for covered directories shouldn't appear in the filesystem.
Under CAP and AppleDouble it currently does. However, the obvious
solution is a real performance killer and is not worth implementing.
Fantasy features:
1. Access Desktop file/database for comment and icon.
2. Implement mmap() for AppleDouble header files and CAP info files.
3. Implement AppleShare client support.
Suggestions/comments/questions are welcome.
Code addressing any of the issues listed above is especially welcome.
Paul H. Hargrove
hargrove@sccm.Stanford.EDU
|
