/
TODO
104 lines (99 loc) · 5.54 KB
/
TODO
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
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
distro/packaging:
- (issue) check why systemd doesn't let me set scheduler priority
fixed for systemd unit file, LSB init + systemd still fails
https://bugzilla.redhat.com/show_bug.cgi?id=893015
link/host level:
- (issue) must add throttling on socket writing (in progress)
- (issue) SCTP: in kernel fragmentation appears to be broken and needs
some investigation. workaround in place by disabling
in-kernel SCTP fragmentation
- (issue) transports: plumb TX error handling on per-transport hooks
- (issue) transports: review all error handling code failures,
specially on clear_config, to not leak and better clean
in case of errors
- (issue) improve test suite to cover for all transports (needs transport list API)
- (issue) UDP/SCTP support for dynamic links (aka no dst_address)
this one needs plumbing inside RX thread
- (issue) change knet_bench to allow protocol specification per link
- (issue) review memory locking across
- (issue) simplify handling of DATA and HOSTINFO code paths in send/recv code
- (issue) need bind to interface for dynamic ip local interfaces vs src ip
address or find a way to autodetect the new ip on that interface
(listen to kernel netlink?)
- (issue) must implement link auth via user/passwd. This is necessary
in case key is leaked.
- (issue) standardize exit labels (ex out_unlock / exit_unlock) and variable/function
names.
- (issue) review how TX onwire pckt info are filled in between inbuf/socket/frags
- (rfe) add fd_tracker error exit check and perhaps use it for local sockets too
to make it easier to identify leaks and fd abuse
- (rfe) add transport list query API
- (rfe) link status callback notification
- (rfe) compress: should only compress user data, we will add a bit in the data
header to indicate if the pckt is compressed or not (save time).
this approach allow runtime change of compress.
open questions are: methods? level? zlib? lzo? bz? lzma? xz?
how much do we save by compressin our header?
compress must happen before encrypt
we can express compress data in packet type without adding extra
flags to the headers. DATA -> BZ/GZDATA and we can change that
right before encrypting. Using a similar approach to PING_MASK
- (rfe) crypto: expand API to support dual key for rekey process
- (rfe) link id made optional? right now we need the link id to match
on both sides of the connection. this is somewhat annoying
from a user perspective. Evaluate if we can make it optional.
- (rfe) make hostid autogenerated in a consistent way?
- (rfe) Check IPV6_NEXTHOP for v6 sockets and find equivalent for v4 (Jesper?)
this would allow using one IP address as destination via multiple links
- (rfe) add statistics at different levels (pckt per host/link, bytes, crypto
overhead, frame overhead, pure data...)
- (rfe) link connection access-list
(chrissie has working generic code for this one, needs merging and API)
- (rfe) improve host-to-host communication. Right now I am not satisfied with
the current implementation, even if it works.
- (rfe) implement link switching via scoring system based on:
1) latency
2) priority (auto/manual)
3) usage (over XX% traffic start RR)
4) flapping of the links (time/sec)
this requires complex rules setting and a super efficent way to look
up destination links
5) if links are stable, reduce the number of links in a-a
min 2
- (rfe) benchmark tests for all critical paths in switching threads
- (rfe) network convergence protocol (host exchange)
- (rfe) reswitching of packets
- (rfe) look into UDP+ECN bit set to avoid overloading sockets?
- (rfe) add openssl support?
- (rfe) consider adding threadpools to process data packets in parallel
libknet:
- (issue) review logging policy/levels in public api call
example is scanning for active links in a host that would return
a half gazzillion useless log entries
- (issue) add .3 man pages
libtap:
- (issue) add .3 man pages
- (issue) improve tests to cover thread safety and better error codes
specially from the up/down handling.
- (rfe) consider adding dhcp support for tap device
it can be done now via up.d/ scripts, but it's not intuitive
kronostnetd:
- (issue) beside the code that is as bad as it can possibly be and
will make you wish to have a tea spoon handy to carve your
eyeballs out, the vty needs a good clean/rewrite
- (issue) fix config file format. current one will make you scream
- (issue) missing output from several command execution failures in vty mode
- (issue) fix check_param for ip/prefix/crypto
(this is part of the rewrite as it needs more clever arg parsing
code/method)
- (rfe) add logging config (per subsystem/global)
- (rfe) split vty_cmd_files to be smaller. it's just too big to handle
nicely.
- (rfe) add equivalent of "description: ...." to various levels
- (rfe) add optional options. right now it's necessary to specify everything
all the time.
- (rfe) implement tab completion on options
general:
- (issue) missing unit tests on many many bits
- (issue) missing docs of all kind, devel, users, admin guide.
- (rfe) check code with coverity