blob: 3c00e0c260d9ee6b15f330a489160cf08957ee04 [file] [log] [blame]
/* ping.c - check network connectivity
*
* Copyright 2014 Rob Landley <rob@landley.net>
*
* Not in SUSv4.
*
* Note: ping_group_range should never have existed. To disable it, do:
* echo 0 999999999 > /proc/sys/net/ipv4/ping_group_range
* (Android does this by default in its init script.)
*
* Yes, I wimped out and capped -s at sizeof(toybuf), waiting for a complaint...
// -s > 4064 = sizeof(toybuf)-sizeof(struct icmphdr)-CMSG_SPACE(sizeof(uint8_t)), then kernel adds 20 bytes
USE_PING(NEWTOY(ping, "<1>1m#t#<0>255=64c#<0=3s#<0>4064=56i%W#<0=3w#<0qf46I:[-46]", TOYFLAG_USR|TOYFLAG_BIN|TOYFLAG_LINEBUF))
USE_PING(OLDTOY(ping6, ping, TOYFLAG_USR|TOYFLAG_BIN|TOYFLAG_LINEBUF))
config PING
bool "ping"
default y
help
usage: ping [OPTIONS] HOST
Check network connectivity by sending packets to a host and reporting
its response.
Send ICMP ECHO_REQUEST packets to ipv4 or ipv6 addresses and prints each
echo it receives back, with round trip time. Returns true if host alive.
Options:
-4, -6 Force IPv4 or IPv6
-c CNT Send CNT many packets (default 3, 0 = infinite)
-f Flood (print . and \b to show drops, default -c 15 -i 0.2)
-i TIME Interval between packets (default 1, need root for < .2)
-I IFACE/IP Source interface or address
-m MARK Tag outgoing packets using SO_MARK
-q Quiet (stops after one returns true if host is alive)
-s SIZE Data SIZE in bytes (default 56)
-t TTL Set Time To Live (number of hops)
-W SEC Seconds to wait for response after last -c packet (default 3)
-w SEC Exit after this many seconds
*/
#define FOR_ping
#include "toys.h"
#include <ifaddrs.h>
#include <netinet/ip_icmp.h>
GLOBALS(
char *I;
long w, W, i, s, c, t, m;
struct sockaddr *sa;
int sock;
unsigned long sent, recv, fugit, min, max;
)
// Print a summary. Called as a single handler or at exit.
static void summary(int sig)
{
if (!FLAG(q) && TT.sent && TT.sa) {
printf("\n--- %s ping statistics ---\n", ntop(TT.sa));
printf("%lu packets transmitted, %lu received, %ld%% packet loss\n",
TT.sent, TT.recv, ((TT.sent-TT.recv)*100)/(TT.sent?TT.sent:1));
if (TT.recv)
printf("round-trip min/avg/max = %lu/%lu/%lu ms\n",
TT.min, TT.fugit/TT.recv, TT.max);
}
TT.sa = 0;
}
// assumes aligned and can read even number of bytes
static unsigned short pingchksum(unsigned short *data, int len)
{
unsigned short u = 0, d;
// circular carry is endian independent: bits from high byte go to low byte
while (len>0) {
d = *data++;
if (len == 1) d &= 255<<IS_BIG_ENDIAN;
if (d >= (u += d)) u++;
len -= 2;
}
return u;
}
static int xrecvmsgwait(int fd, struct msghdr *msg, int flag,
union socksaddr *sa, int timeout)
{
socklen_t sl = sizeof(*sa);
int len;
if (timeout >= 0) {
struct pollfd pfd;
pfd.fd = fd;
pfd.events = POLLIN;
if (!xpoll(&pfd, 1, timeout)) return 0;
}
msg->msg_name = (void *)sa;
msg->msg_namelen = sl;
len = recvmsg(fd, msg, flag);
if (len<0) perror_exit("recvmsg");
return len;
}
void ping_main(void)
{
struct addrinfo *ai, *ai2;
struct ifaddrs *ifa, *ifa2 = 0;
struct icmphdr *ih = (void *)toybuf;
struct msghdr msg;
struct cmsghdr *cmsg;
struct iovec iov;
union socksaddr srcaddr, srcaddr2;
struct sockaddr *sa = (void *)&srcaddr;
int family = 0, ttl = 0, len;
long long tnext, tW, tnow, tw;
unsigned short seq = 0, pkttime;
// Set nonstatic default values
if (!FLAG(i)) TT.i = FLAG(f) ? 200 : 1000;
else if (TT.i<200 && geteuid()) error_exit("need root for -i <200");
if (!FLAG(s)) TT.s = 56; // 64-PHDR_LEN
if (FLAG(f) && !FLAG(c)) TT.c = 15;
// ipv4 or ipv6? (0 = autodetect if -I or arg have only one address type.)
if (FLAG(6) || strchr(toys.which->name, '6')) family = AF_INET6;
else if (FLAG(4)) family = AF_INET;
else family = 0;
// If -I srcaddr look it up. Allow numeric address of correct type.
memset(&srcaddr, 0, sizeof(srcaddr));
if (TT.I) {
if (!FLAG(6) && inet_pton(AF_INET, TT.I, (void *)&srcaddr.in.sin_addr))
family = AF_INET;
else if (!FLAG(4) && inet_pton(AF_INET6, TT.I, (void *)&srcaddr.in6.sin6_addr))
family = AF_INET6;
else if (getifaddrs(&ifa2)) perror_exit("getifaddrs");
}
// Look up HOST address, filtering for correct type and interface.
// If -I but no -46 then find compatible type between -I and HOST
ai2 = xgetaddrinfo(*toys.optargs, 0, family, 0, 0, 0);
for (ai = ai2; ai; ai = ai->ai_next) {
// correct type?
if (family && family!=ai->ai_family) continue;
if (ai->ai_family!=AF_INET && ai->ai_family!=AF_INET6) continue;
// correct interface?
if (!TT.I || !ifa2) break;
for (ifa = ifa2; ifa; ifa = ifa->ifa_next) {
if (!ifa->ifa_addr || ifa->ifa_addr->sa_family!=ai->ai_family
|| strcmp(ifa->ifa_name, TT.I)) continue;
sa = (void *)ifa->ifa_addr;
break;
}
if (ifa) break;
}
if (!ai)
error_exit("no v%d addr for -I %s", 4+2*(family==AF_INET6), TT.I);
TT.sa = ai->ai_addr;
// Open DGRAM socket
sa->sa_family = ai->ai_family;
TT.sock = socket(ai->ai_family, SOCK_DGRAM,
len = (ai->ai_family == AF_INET) ? IPPROTO_ICMP : IPPROTO_ICMPV6);
if (TT.sock == -1) {
perror_msg("socket SOCK_DGRAM %x", len);
if (errno == EACCES) {
fprintf(stderr, "Kernel bug workaround:\n"
"echo 0 99999999 | sudo tee /proc/sys/net/ipv4/ping_group_range\n");
}
xexit();
}
if (TT.I) xbind(TT.sock, sa, sizeof(srcaddr));
len = 1;
xsetsockopt(TT.sock, SOL_IP, IP_RECVTTL, &len, sizeof(len));
if (FLAG(m)) {
len = TT.m;
xsetsockopt(TT.sock, SOL_SOCKET, SO_MARK, &len, sizeof(len));
}
if (TT.t) {
len = TT.t;
if (ai->ai_family == AF_INET)
xsetsockopt(TT.sock, IPPROTO_IP, IP_TTL, &len, 4);
else xsetsockopt(TT.sock, IPPROTO_IPV6, IPV6_UNICAST_HOPS, &len, sizeof(len));
}
if (!FLAG(q)) {
printf("Ping %s (%s)", *toys.optargs, ntop(TT.sa));
if (TT.I) {
*toybuf = 0;
printf(" from %s (%s)", TT.I, ntop(sa));
}
// 20 byte TCP header, 8 byte ICMP header, plus data payload
printf(": %ld(%ld) bytes.\n", TT.s, TT.s+28);
}
TT.min = ULONG_MAX;
toys.exitval = 1;
tW = tw = 0;
tnext = millitime();
if (TT.w) tw = TT.w*1000+tnext;
memset(&msg, 0, sizeof(msg));
// left enought space to store ttl value
len = CMSG_SPACE(sizeof(uint8_t));
iov.iov_base = (void *)toybuf;
iov.iov_len = sizeof(toybuf) - len;
msg.msg_iov = &iov;
msg.msg_iovlen = 1;
msg.msg_control = &toybuf[iov.iov_len];
msg.msg_controllen = len;
sigatexit(summary);
// Send/receive packets
for (;;) {
int waitms = INT_MAX;
// Exit due to timeout? (TODO: timeout is after last packet, waiting if
// any packets ever dropped. Not timeout since packet was dropped.)
tnow = millitime();
if (tW) {
if (0>=(waitms = tW-tnow) || !(TT.sent-TT.recv)) break;
waitms = tW-tnow;
}
if (tw) {
if (tnow>tw) break;
else if (waitms>tw-tnow) waitms = tw-tnow;
}
// Time to send the next packet?
if (!tW && tnext-tnow <= 0) {
tnext += TT.i;
memset(ih, 0, sizeof(*ih));
ih->type = (ai->ai_family == AF_INET) ? 8 : 128;
ih->un.echo.id = getpid();
ih->un.echo.sequence = ++seq;
if (TT.s >= 4) *(unsigned *)(ih+1) = tnow;
ih->checksum = pingchksum((void *)toybuf, TT.s+sizeof(*ih));
xsendto(TT.sock, toybuf, TT.s+sizeof(*ih), TT.sa);
TT.sent++;
if (FLAG(f) && !FLAG(q)) xputc('.');
// last packet?
if (TT.c) if (!--TT.c) {
tW = tnow + TT.W*1000;
waitms = 1; // check for immediate return even when W=0
}
}
// This is down here so it's against new period if we just sent a packet
if (!tW && waitms>tnext-tnow) waitms = tnext-tnow;
// wait for next packet or timeout
if (waitms<0) waitms = 0;
if (!(len = xrecvmsgwait(TT.sock, &msg, 0, &srcaddr2, waitms)))
continue;
TT.recv++;
TT.fugit += (pkttime = millitime()-*(unsigned *)(ih+1));
if (pkttime < TT.min) TT.min = pkttime;
if (pkttime > TT.max) TT.max = pkttime;
// reply id == 0 for ipv4, 129 for ipv6
cmsg = CMSG_FIRSTHDR(&msg);
for (cmsg = CMSG_FIRSTHDR(&msg); cmsg; cmsg = CMSG_NXTHDR(&msg, cmsg)) {
if (cmsg->cmsg_level == IPPROTO_IP
&& cmsg->cmsg_type == IP_TTL) {
ttl = *(uint8_t *)CMSG_DATA(cmsg);
break;
}
};
if (!FLAG(q)) {
if (FLAG(f)) xputc('\b');
else {
printf("%d bytes from %s: icmp_seq=%d ttl=%d", len, ntop(&srcaddr2.s),
ih->un.echo.sequence, ttl);
if (len >= sizeof(*ih)+4) printf(" time=%u ms", pkttime);
xputc('\n');
}
}
toys.exitval = 0;
}
// summary(0) gets called for us atexit.
if (CFG_TOYBOX_FREE) {
freeaddrinfo(ai2);
if (ifa2) freeifaddrs(ifa2);
}
}