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
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
|
Value range check:
- tcc currently lets problem constructs like
$b = trTCM(cir 10bps,cbs 0B,pir 10bps,pbs 0B);
pass (causes failure in tc)
Protocol-aware if_u32:
- algorithm isn't fully tested:
- does protocol selection always yield correct selection ?
- does protocol selection always yield formally acceptable data ?
- need to enable protocol awareness permanently
- needs to fix compatibility with policing
Small IPv6-related issues:
- tcsim should use typed values, so that "send ::" knows that "::" is an IPv6
address and doesn't need ipv6: qualifier (and likewise for IPv4 addresses)
- tcc issues "ipv4" classifiers for the "tc" target even if using IPv6
- "C" target does not support 128 bit access and operations
Big IPv6-related issues:
- tcng can't loop, so it can't usefully parse extension headers
- neither does tcng have a useful way for expressing alternative branches, so
adding TCP after IPv6 would either override the IPv4 TCP header offset, or
a new "tcp6" would have to be defined
Other:
- "egress" clobbers skb->tc_index by implicitly setting default_index, which
breaks configurations that set skb->tc_index at ingress and use it for
classification at egress
More:
- should make kernel patch to enable "prio" to drop, like lots of tcng
regression tests show
- tcsim should use better timer resolution than "jiffies"
- add limited macro capabilities to eliminate tcc_var2fix and to allow
addition of compound queuing constructs
- provide "qdisc primitives", like metering primitives
- port tcsim to recent 2.5 kernels
- tcsim: also generate IPv6 packet length
- tcsim: generate IPv4, TCP, UDP checksums
Stuff below the double line is *very* old ...
===============================================================================
* nicer message if aborting with signal ?
Short-term issues
-----------------
param.c:
- param_print gets confused if parameter exists at qdisc and class, but is not
defaulted (e.g. priomap and most other CBQ parameters)
- check: are there parameters at CBQ classes that aren't defaulted among child
classes (again, priomap comes to mind, but we hide this from the external
interface anyway) ?
external interface:
- fix "dump_all found no matching choice for value 0x0 in tcindex" problem
(tests/misfeature)
Long-term issues
----------------
classes, general:
- check to what extent forward-declared variables can be abused to
construct invalid references
documentation (tomorrow is always a good day for writing documentation):
- add complete flow description to doc/external.tex
- merge README.tccext into doc/external.tex
ext_all:
- FFS: consider either always suppressing or always generating optional
parameters with default values
- document that we print flag parameters with numeric argument
- dumping really has three modes: "only filter", "one device", "all devices".
Should also support "one device".
iflib_bit:
- support "if" via u32 using bit tree algorithms
- add reordering constraints and reorder where useful (need to define this
first)
- make bit tree functions generate more compact matches (i.e. find local
optimum for leaf balance)
- optimize bit tree functions
iflib_arith:
- make c:build_le generate more compact expressions by testing multiple bits
per comparison
policing:
- support "drop on" (like "drop if")
qdisc.c:
- for defaults, also take into account "reclassify" (or at least report it if
we can't support it)
target support:
- support more than one external target
tccext:
- include user-provided label in element lines
tunnels:
- integrate Milena's tunneling suggestions
object identifiers:
- check: qdisc:class:... may "class" be empty also for tunnel ?
dsmark:
- consider using fields instead of mask/value
"C" target:
- should implement u32 classifier using perfect hashes, see
http://burtleburtle.net/bob/hash/perfect.html
(link courtesy of Mario Luoni)
Strings:
- strings have a moderate memory leak problem
- general: consider switching data handling to pointer passing, along with
reference counts
Distribution:
- add RPM requirements and make RPMs portable across distributions (e.g.
RH's cpp vs. SuSE's c_preprocessor)
|
