pevik
commented
8 months ago Thanks for working on this. no answer yet for first reply would need to be fixed.
$ ./builddir/ping/ping -OD ::1
PING ::1 (::1) 56 data bytes
[1708670330.871771] no answer yet for icmp_seq=1
[1708670330.871817] 64 bytes from ::1: icmp_seq=1 ttl=64 time=0.041 ms
[1708670330.871832] 64 bytes from ::1: icmp_seq=2 ttl=64 time=0.013 ms
[1708670331.889429] 64 bytes from ::1: icmp_seq=3 ttl=64 time=0.058 ms
[1708670332.873644] 64 bytes from ::1: icmp_seq=4 ttl=64 time=0.068 msHave you tried a different interval than just the default 1s? It's slightly increasing.
$ ./builddir/ping/ping 8.8.8.8 -OD -i2
PING 8.8.8.8 (8.8.8.8) 56(84) bytes of data.
[1708670068.819166] no answer yet for icmp_seq=1
[1708670068.831540] 64 bytes from 8.8.8.8: icmp_seq=1 ttl=116 time=12.3 ms
[1708670068.832339] 64 bytes from 8.8.8.8: icmp_seq=2 ttl=116 time=13.1 ms
[1708670070.837011] 64 bytes from 8.8.8.8: icmp_seq=3 ttl=116 time=15.4 ms
[1708670072.838528] 64 bytes from 8.8.8.8: icmp_seq=4 ttl=116 time=16.3 ms
[1708670074.835015] 64 bytes from 8.8.8.8: icmp_seq=5 ttl=116 time=11.2 ms
...
[1708670174.842987] 64 bytes from 8.8.8.8: icmp_seq=55 ttl=116 time=11.1 ms
[1708670176.846084] 64 bytes from 8.8.8.8: icmp_seq=56 ttl=116 time=13.9 ms
[1708670178.843404] 64 bytes from 8.8.8.8: icmp_seq=57 ttl=116 time=11.9 ms
[1708670180.844339] 64 bytes from 8.8.8.8: icmp_seq=58 ttl=116 time=10.7 msnit (early return adds readability):
+++ ping/ping.h
@@ -311,18 +311,19 @@ static inline void tvsub(struct timeval *out, struct timeval *in)
static inline void tsincr(struct timespec *out, int msec)
{
- if (msec) {
- long incr_msecs = msec % 1000;
- long incr_secs = msec / 1000;
- long incr_nsecs = incr_msecs * 1000000;
-
- if (out->tv_nsec + incr_nsecs > 1000000000) {
- incr_secs++;
- incr_nsecs -= 1000000000;
- }
- out->tv_sec += incr_secs;
- out->tv_nsec += incr_nsecs;
+ if (!msec)
+ return;
+
+ long incr_msecs = msec % 1000;
+ long incr_secs = msec / 1000;
+ long incr_nsecs = incr_msecs * 1000000;
+
+ if (out->tv_nsec + incr_nsecs > 1000000000) {
+ incr_secs++;
+ incr_nsecs -= 1000000000;
}
+ out->tv_sec += incr_secs;
+ out->tv_nsec += incr_nsecs;
}
/*
greearb
yvs2014
The existing code calculated 'tokens' from current time. As system calls used up some small amount of code, the ping transmit interval tends to be about 2ms slower than expected (on my test system).
Instead of trying to keep track of a tokens counter, use a timestamp that is incremented each ping send request by 'interval'. This keeps the interval the same even with the system call latency.
Patch attached fixes it, could probably use some more testing. 0001-ping-Ensure-ping-is-sent-on-correct-interval.txt