support filter by ip6 address #26
Conversation
|
fix #6 |
1. assume that there are no other ipv6 extension headers and the default is to handle layer 4 protocols directly. 2. rename filter_mark to filter_meta, other metadata should be supported later Signed-off-by: Duan Jiong <[email protected]>
| u8 proto; | ||
| u8 pad[7]; | ||
| u8 l4_proto; | ||
| u16 l3_proto; |
There was a problem hiding this comment.
The tuple sizes don't match in C and Go. In Go it's 42 bytes, in C - 40 bytes. To make it 40 bytes in both, you need to swap l4_proto with l3_proto.
Because of that the event_t doesn't match with Event.
You can use the following to validate offsets and size (until #28 has been fixed):
$ clang -O2 -g -I./headers -target bpf -c kprobe_pwru.c -o foo.o
$ pahole foo.o | less
# meanwhile in Go add the following function:
func init() {
fmt.Println("tuple", reflect.TypeOf(Tuple{}).Size())
fmt.Println("meta", reflect.TypeOf(Meta{}).Size())
fmt.Println("event", reflect.TypeOf(Event{}).Size())
}
There was a problem hiding this comment.
Understood, thank you.
bpf/kprobe_pwru.c
Outdated
| } | ||
| } else if (ip_vsn == 6) { | ||
| /* | ||
| * The transport layer protocol is represented in ipv6 by the next header type, but there are other |
There was a problem hiding this comment.
I think it's fine to assume that IPv6 packets with extension headers can be ignored if the filter {s,d}port is set. So you can expect that the next header type will be pointing to the L4.
internal/pwru/output.go
Outdated
| u32ToNetIPv4(event.Tuple.Saddr), byteorder.NetworkToHost16(event.Tuple.Sport), | ||
| u32ToNetIPv4(event.Tuple.Daddr), byteorder.NetworkToHost16(event.Tuple.Dport), | ||
| protoToStr(event.Tuple.Proto)) | ||
| fmt.Printf(" [%s]:%d->[%s]:%d(%s)", |
There was a problem hiding this comment.
Let's use the square brackets only for the IPv6 type addrs.
bpf/kprobe_pwru.c
Outdated
| return true; | ||
| } | ||
|
|
||
| //Should we introduce builtin memcmp etc., like bpf in cilium |
There was a problem hiding this comment.
Let's instead create a GH issue to track this.
duanjiong
force-pushed
the
support-ipv6
branch
from
da6ca33 to
2ed5b6b
Compare
|
It was reworked as suggested. |
|
@duanjiong Hi, could you ACK #190? |
Signed-off-by: Duan Jiong [email protected]