tracer_t is a toy Rust traceroute(8), for Linux.
$ tracer_t 8.8.8.8
tracer_t to 8.8.8.8, 12 hops max
1 192.168.0.1 0.349 ms 0.460 ms 0.600 ms
2 63.231.10.66 4.251 ms 4.296 ms 4.353 ms
3 63.226.198.9 4.358 ms 4.396 ms 4.448 ms
4 67.14.41.158 4.391 ms 4.448 ms 4.574 ms
5 72.14.221.108 4.523 ms 4.670 ms 72.14.221.110 4.803 ms
6 74.125.243.193 5.016 ms 10.252.164.126 5.585 ms 74.125.243.177 6.391 ms
7 8.8.8.8 4.578 ms 4.722 ms 4.833 ms
tracer_t sends UDP datagrams to the target IP address, on sequential, high-numbered ports, with varying IP TTL (time to live) values.
tracer_t uses the Linux IPPROTO_IP IP_RECVERR socket option feature to
receive ICMP error results associated with the low-TTL probes from the Linux
kernel. In addition to delegating some of the work to the kernel, this
approach is nice because it does not require elevated privileges.
tracer_t blasts out all probes immediately, without rate-limiting, and then
uses the poll(2) system call to monitor responses without polling. Output
lines are emitted when all probes for the next TTL have received responses, or
timed out.
tracer_t employs the same max/here/near timeout mechanism as traceroute(8).
Probes are subject to the minimum of several possible timeouts. The first is the 'max', which always applies (default 5 seconds). Second, if any other responses have been received for the same hop, the 'here' timeout is the quickest response for the same hop multiplied by a small factor (default 3x). Finally, the 'near' timeout is the quickest response for any subsequent hop multiplied by a larger factor (default 10x).
This heuristic should work tolerably in consistently high-latency network conditions, while being responsive in low-latency networks with some packet loss, or with middleboxes that do not send ICMP Time Exceeded error responses.
tracer_t depends on the nix, libc, and anyhow crates. Of these, nix and libc are essential; anyhow could be dropped. Indirect dependencies are autocfg, bitflags, cfg-if, and memoffset.