I wrote a basic implementation of traceroute. I decided to keep it minimal and low-effort. I understand how traceroute works and just wanted a small project to refresh my C network programming.
I learned about “connected” UDP sockets in Beej’s Guide to Network Programming and thought I could cheat a little bit. Rather than fussing with a raw ICMP socket I could just use a connected UDP socket and let the kernel figure out what ICMP error messages belong to it.
The rest of this post is basically the dialogue I had with myself while trying to understand why it wasn’t working as expected.
I found various resources indicating that ICMP error messages are propagated on connected UDP sockets.
I wrote a small bit of code (below) to test out getting the ICMP error message for an unreachable destination port and used netcat to run a process listening for UDP packets nc -lu 9000.
I created a connected UDP socket to send() a packet to a different port and then called recv().
tty000 $ ./send_udp_recv_icmp localhost 33434
recv error: Connection refused
It works! I could also see the ICMP error message in Wireshark.
I confirmed that there were no errors when I sent packets to port 9000 and tested against a remote host as well.
tty999 $ nc -lu 9000
tty000 $ ./send_udp_recv_icmp localhost 9000
tty999 receives "Hello, World!" and hits enter.
tty000 $ <exited successfully after response>
Next up: ICMP time-exceeded messages
The code that follows is a pared down example with no error-checking, but should demonstrate what I was attempting.
/* send_udp_recv_icmp.c */
#include ...
#define MAXDATASIZE 1500
int main(int argc, char *argv[]) {
int sockfd;
int ttl;
char buf[MAXDATASIZE];
char message[] = "Hello, world!";
struct addrinfo *ai;
if (argc != 3) {
fprintf(stderr, "usage: hostname port\n");
exit(1);
}
getaddrinfo(argv[1], argv[2], NULL, &ai);
sockfd = socket(AF_INET, SOCK_DGRAM, IPPROTO_UDP);
connect(sockfd, ai->ai_addr, ai->ai_addrlen);
ttl = 1;
setsockopt(sockfd, IPPROTO_IP, IP_TTL, &ttl, sizeof(ttl));
if ((send(sockfd, message, sizeof(message), 0)) == -1) {
perror("send error");
exit(1);
}
// I expect this to fail if send() resulted in an ICMP
// error message but instead it hangs.
if ((recv(sockfd, buf, MAXDATASIZE - 1, 0)) == -1) {
perror("recv error");
exit(1);
}
return 0;
}
It just seemed to hang at recv() (i.e. it didn’t receive a response). Shouldn’t the error propagate? I tried sending more packets as well. Nothing.
I checked Wireshark for my outgoing UDP packet and for incoming ICMP time-exceeded messages.
19 3.121938 192.168.2.18 172.217.1.14 UDP 56 57501 → 33434 Len=14
20 3.124019 192.168.2.1 192.168.2.18 ICMP 84 Time-to-live exceeded (Time to live exceeded in transit)
There they are. I can tell the ICMP error message is for my socket by matching the source port of the UDP packet, 57501, to the source port found in the UDP header inside the ICMP message*. Same for the destination port 33434. (*assuming the source port is not being used by another socket for the same purposes).
Frame 20: 84 bytes on wire (672 bits), 84 bytes captured (672 bits) on interface 0
Ethernet II, Src: ...
Internet Protocol Version 4, Src: 192.168.2.1, Dst: 192.168.2.18
Internet Control Message Protocol
Type: 11 (Time-to-live exceeded)
Code: 0 (Time to live exceeded in transit)
Checksum: 0x65c9 [correct]
[Checksum Status: Good]
Internet Protocol Version 4, Src: 192.168.2.18, Dst: 172.217.1.14
User Datagram Protocol, Src Port: 57501, Dst Port: 33434
Source Port: 57501 // this matches!
Destination Port: 33434 // this matches!
Length: 22
Checksum: 0xea9b [unverified]
[Checksum Status: Unverified]
[Stream index: 1]
Data (14 bytes)
With TTL = 1 it looks like my router is sending the ICMP error messages.
While poking around various manpages and Google/StackOverflow, I came across SO_ERROR in the FreeBSD manpages:
SO_ERROR returns any pending error on the socket and clears the error status. It may be used to check for asynchronous errors on connected datagram sockets or for other asynchronous errors.
Sounds exactly like what I wanted! But, unfortunately, this didn’t help. I tried adding a sleep() before calling getsockopt() with SO_ERROR in case it was called too fast, but still no luck.
Linux has an option IP_RECVERR but I wanted my implementation to be portable.
The Linux manpages for UDP indicate:
All fatal errors will be passed to the user as an error return even when the socket is not connected.
Maybe it’s not considered “fatal”?
That manpage has a reference to RFC 1122 Requirements for Internet Hosts – Communication Layers.
on page 40:
An incoming Time Exceeded message MUST be passed to the transport layer.
on page 77:
UDP MUST pass to the application layer all ICMP error messages that it receives from the IP layer.
BCP 145 UDP Usage Guidelines also mentions on page 32:
On some stacks, a bound socket also allows an application to be notified when ICMP error messages are received for its transmissions [RFC1122].
Well, I decided to stop spending time on this particular issue and ended up using a raw ICMP socket to receive error messages.
There is a logical explanation to be heard. If you know what it is, please get in touch with me. I’d be happy to hear it.