ICMP sockets (linux) - linux

Question

Is it possible to use ICMP sockets under the IP protocol? Maybe something like:
socket(PF_INET, <type>, IPPROTO_ICMP)?
What should I put in the <type> field? I saw some examples using SOCK_RAW, but won't that prevent the OS from doing his job handling the IP protocol?
And another thing. How can the OS know to which process he should send the ICMP datagrams, since there are no ports involved with the protocol?

Answer 1

Linux have a special ICMP socket type you can use with:
socket(PF_INET, SOCK_DGRAM, IPPROTO_ICMP);
This allows you to only send ICMP echo requests The kernel will handle it specially (match request/responses, fill in the checksum).
This only works if a special sysctl is set. By default not even root can use this kind of socket. You specify the user groups that can access it. To allow root (group 0) to use ICMP sockets, do:
sysctl -w net.ipv4.ping_group_range="0 0"
Here is an example program to demonstrate the very basic usage of sending an ICMP echo request:
#include <stdio.h>
#include <errno.h>
#include <string.h>
#include <stdlib.h>
#include <sys/socket.h>
#include <sys/socket.h>
#include <netinet/in.h>
#include <netinet/ip_icmp.h>
#include <arpa/inet.h>
#include <sys/select.h>
//note, to allow root to use icmp sockets, run:
//sysctl -w net.ipv4.ping_group_range="0 0"
void ping_it(struct in_addr *dst)
{
struct icmphdr icmp_hdr;
struct sockaddr_in addr;
int sequence = 0;
int sock = socket(AF_INET,SOCK_DGRAM,IPPROTO_ICMP);
if (sock < 0) {
perror("socket");
return ;
}
memset(&addr, 0, sizeof addr);
addr.sin_family = AF_INET;
addr.sin_addr = *dst;
memset(&icmp_hdr, 0, sizeof icmp_hdr);
icmp_hdr.type = ICMP_ECHO;
icmp_hdr.un.echo.id = 1234;//arbitrary id
for (;;) {
unsigned char data[2048];
int rc;
struct timeval timeout = {3, 0}; //wait max 3 seconds for a reply
fd_set read_set;
socklen_t slen;
struct icmphdr rcv_hdr;
icmp_hdr.un.echo.sequence = sequence++;
memcpy(data, &icmp_hdr, sizeof icmp_hdr);
memcpy(data + sizeof icmp_hdr, "hello", 5); //icmp payload
rc = sendto(sock, data, sizeof icmp_hdr + 5,
0, (struct sockaddr*)&addr, sizeof addr);
if (rc <= 0) {
perror("Sendto");
break;
}
puts("Sent ICMP\n");
memset(&read_set, 0, sizeof read_set);
FD_SET(sock, &read_set);
//wait for a reply with a timeout
rc = select(sock + 1, &read_set, NULL, NULL, &timeout);
if (rc == 0) {
puts("Got no reply\n");
continue;
} else if (rc < 0) {
perror("Select");
break;
}
//we don't care about the sender address in this example..
slen = 0;
rc = recvfrom(sock, data, sizeof data, 0, NULL, &slen);
if (rc <= 0) {
perror("recvfrom");
break;
} else if (rc < sizeof rcv_hdr) {
printf("Error, got short ICMP packet, %d bytes\n", rc);
break;
}
memcpy(&rcv_hdr, data, sizeof rcv_hdr);
if (rcv_hdr.type == ICMP_ECHOREPLY) {
printf("ICMP Reply, id=0x%x, sequence = 0x%x\n",
icmp_hdr.un.echo.id, icmp_hdr.un.echo.sequence);
} else {
printf("Got ICMP packet with type 0x%x ?!?\n", rcv_hdr.type);
}
}
}
int main(int argc, char *argv[])
{
if (argc != 2) {
printf("usage: %s destination_ip\n", argv[0]);
return 1;
}
struct in_addr dst;
if (inet_aton(argv[1], &dst) == 0) {
perror("inet_aton");
printf("%s isn't a valid IP address\n", argv[1]);
return 1;
}
ping_it(&dst);
return 0;
}
Note that the kernel will reject and fail the sendto() call if the data sent does not have room for a proper ICMP header, and the ICMP type must be 8 (ICMP_ECHO) and the ICMP code must be 0.

Answer 2

Yes it is possible, since the ping command does ICMP.
To find out the syscalls involved, you can strace that command (under root).
You could also glance into that command's source code, e.g. Debian's ping
And there is the liboping library to help you...