mtr/packet/probe_unix.c
Roger Wolff 62bd718e83
Merge pull request #478 from esimest/master
Add error code ETIMEOUT(110) handle logic
2023-10-03 14:00:59 +02:00

1049 lines
31 KiB
C

/*
mtr -- a network diagnostic tool
Copyright (C) 2016 Matt Kimball
This program is free software; you can redistribute it and/or modify
it under the terms of the GNU General Public License version 2 as
published by the Free Software Foundation.
This program is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
GNU General Public License for more details.
You should have received a copy of the GNU General Public License along
with this program; if not, write to the Free Software Foundation, Inc.,
51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
*/
#include "probe.h"
#include <assert.h>
#include <errno.h>
#include <fcntl.h>
#ifdef HAVE_ERROR_H
#include <error.h>
#else
#include "portability/error.h"
#endif
#ifdef HAVE_LINUX_ERRQUEUE_H
#include <linux/errqueue.h>
#endif
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <sys/socket.h>
#include <unistd.h>
#include "platform.h"
#include "protocols.h"
#include "sockaddr.h"
#include "construct_unix.h"
#include "deconstruct_unix.h"
#include "timeval.h"
/* A wrapper around sendto for mixed IPv4 and IPv6 sending */
static
int send_packet(
const struct net_state_t *net_state,
const struct probe_param_t *param,
int sequence,
const char *packet,
int packet_size,
const struct sockaddr_storage *sockaddr)
{
struct sockaddr_storage dst;
int send_socket = 0;
int sockaddr_length;
memcpy(&dst, sockaddr, sizeof(struct sockaddr_storage));
if (sockaddr->ss_family == AF_INET6) {
sockaddr_length = sizeof(struct sockaddr_in6);
if (param->protocol == IPPROTO_ICMP) {
if (net_state->platform.ip6_socket_raw) {
send_socket = net_state->platform.icmp6_send_socket;
} else {
send_socket = net_state->platform.ip6_txrx_icmp_socket;
}
} else if (param->protocol == IPPROTO_UDP) {
if (net_state->platform.ip6_socket_raw) {
send_socket = net_state->platform.udp6_send_socket;
/* we got a ipv6 udp raw socket
* the remote port is in the payload
* we do not set in the sockaddr
*/
*sockaddr_port_offset(&dst) = 0;
} else {
send_socket = net_state->platform.ip6_txrx_udp_socket;
if (param->dest_port) {
*sockaddr_port_offset(&dst) = htons(param->dest_port);
} else {
*sockaddr_port_offset(&dst) = sequence;
}
}
}
} else if (sockaddr->ss_family == AF_INET) {
sockaddr_length = sizeof(struct sockaddr_in);
if (param->protocol == IPPROTO_ICMP) {
if (net_state->platform.ip4_socket_raw) {
send_socket = net_state->platform.icmp4_send_socket;
} else {
send_socket = net_state->platform.ip4_txrx_icmp_socket;
}
} else if (param->protocol == IPPROTO_UDP) {
if (net_state->platform.ip4_socket_raw) {
send_socket = net_state->platform.udp4_send_socket;
/* we got a ipv4 udp raw socket
* the remote port is in the payload
* we do not set in the sockaddr
*/
*sockaddr_port_offset(&dst) = 0;
} else {
send_socket = net_state->platform.ip4_txrx_udp_socket;
if (param->dest_port) {
*sockaddr_port_offset(&dst) = htons(param->dest_port);
} else {
*sockaddr_port_offset(&dst) = sequence;
}
}
}
}
if (send_socket == 0) {
errno = EINVAL;
return -1;
}
return sendto(send_socket, packet, packet_size, 0,
(struct sockaddr *) &dst, sockaddr_length);
}
/*
Nearly all fields in the IP header should be encoded in network byte
order prior to passing to send(). However, the required byte order of
the length field of the IP header is inconsistent between operating
systems and operating system versions. FreeBSD 11 requires the length
field in network byte order, but some older versions of FreeBSD
require host byte order. OS X requires the length field in host
byte order. Linux will accept either byte order.
Test for a byte order which works by sending a ping to localhost.
*/
static
void check_length_order(
struct net_state_t *net_state)
{
char packet[PACKET_BUFFER_SIZE];
struct probe_param_t param;
struct probe_t p0 = {.sequence = MIN_PORT };
ssize_t bytes_sent;
int packet_size;
#ifdef __linux__
/* Linux will accept either byte order and check below fails to work
* in some cases due to sendto() returning EPERM. */
return;
#endif
memset(&param, 0, sizeof(struct probe_param_t));
param.ip_version = 4;
param.protocol = IPPROTO_ICMP;
param.ttl = 255;
param.remote_address = "127.0.0.1";
param.is_probing_byte_order = true;
if (resolve_probe_addresses(net_state, &param, &p0.remote_addr,
&p0.local_addr)) {
fprintf(stderr, "Error decoding localhost address (%s/%s)\n",
probe_err, strerror (errno));
exit(EXIT_FAILURE);
}
/* First attempt to ping the localhost with network byte order */
net_state->platform.ip_length_host_order = false;
packet_size = construct_packet(net_state, NULL, &p0,
packet, PACKET_BUFFER_SIZE,
&param);
if (packet_size < 0) {
error(EXIT_FAILURE, errno, "Unable to send to localhost");
}
bytes_sent =
send_packet(net_state, &param, MIN_PORT, packet, packet_size,
&p0.remote_addr);
if (bytes_sent > 0) {
return;
}
/* Since network byte order failed, try host byte order */
net_state->platform.ip_length_host_order = true;
packet_size = construct_packet(net_state, NULL, &p0,
packet, PACKET_BUFFER_SIZE,
&param);
if (packet_size < 0) {
error(EXIT_FAILURE, errno, "Unable to send to localhost");
}
bytes_sent =
send_packet(net_state, &param, MIN_PORT, packet, packet_size,
&p0.remote_addr);
if (bytes_sent < 0) {
error(EXIT_FAILURE, errno, "Unable to send with swapped length");
}
}
/*
Check to see if SCTP is support. We can't just rely on checking
if IPPROTO_SCTP is defined, because while that is necessary,
MacOS as of "Sierra" defines IPPROTO_SCTP, but creating an SCTP
socket results in an error.
*/
static
void check_sctp_support(
struct net_state_t *net_state)
{
#ifdef IPPROTO_SCTP
int sctp_socket;
sctp_socket = socket(AF_INET, SOCK_STREAM, IPPROTO_SCTP);
if (sctp_socket != -1) {
close(sctp_socket);
net_state->platform.sctp_support = true;
}
#endif
}
/* Set a socket to non-blocking mode */
void set_socket_nonblocking(
int socket)
{
int flags;
flags = fcntl(socket, F_GETFL, 0);
if (flags == -1) {
error(EXIT_FAILURE, errno, "Unexpected socket F_GETFL error");
}
if (fcntl(socket, F_SETFL, flags | O_NONBLOCK)) {
error(EXIT_FAILURE, errno, "Unexpected socket F_SETFL O_NONBLOCK error");
}
}
/* Open the raw sockets for sending/receiving IPv4 packets */
static
int open_ip4_sockets_raw(
struct net_state_t *net_state)
{
int send_socket_icmp;
int send_socket_udp;
int recv_socket;
send_socket_icmp = socket(AF_INET, SOCK_RAW, IPPROTO_ICMP);
if (send_socket_icmp == -1) {
return -1;
}
send_socket_udp = socket(AF_INET, SOCK_RAW, IPPROTO_UDP);
if (send_socket_udp == -1) {
close(send_socket_icmp);
return -1;
}
/*
Open a second socket with IPPROTO_ICMP because we are only
interested in receiving ICMP packets, not all packets.
*/
recv_socket = socket(AF_INET, SOCK_RAW, IPPROTO_ICMP);
if (recv_socket == -1) {
close(send_socket_icmp);
close(send_socket_udp);
return -1;
}
net_state->platform.ip4_present = true;
net_state->platform.ip4_socket_raw = true;
net_state->platform.icmp4_send_socket = send_socket_icmp;
net_state->platform.udp4_send_socket = send_socket_udp;
net_state->platform.ip4_recv_socket = recv_socket;
return 0;
}
#ifdef HAVE_LINUX_ERRQUEUE_H
/* Open DGRAM sockets for sending/receiving IPv4 packets */
static
int open_ip4_sockets_dgram(
struct net_state_t *net_state)
{
int udp_socket;
int icmp_socket, icmp_tmp_socket;
#ifdef HAVE_LINUX_ERRQUEUE_H
int val = 1;
#endif
icmp_socket = socket(AF_INET, SOCK_DGRAM, IPPROTO_ICMP);
if (icmp_socket == -1) {
return -1;
}
#ifdef HAVE_LINUX_ERRQUEUE_H
if (setsockopt(icmp_socket, SOL_IP, IP_RECVERR, &val, sizeof(val)) < 0) {
return -1;
}
#endif
udp_socket = socket(AF_INET, SOCK_DGRAM, IPPROTO_UDP);
if (udp_socket == -1) {
close(icmp_socket);
return -1;
}
#ifdef HAVE_LINUX_ERRQUEUE_H
if (setsockopt(udp_socket, SOL_IP, IP_RECVERR, &val, sizeof(val)) < 0) {
close(icmp_socket);
close(udp_socket);
return -1;
}
#endif
icmp_tmp_socket = socket(AF_INET, SOCK_DGRAM, IPPROTO_ICMP);
if (icmp_tmp_socket == -1) {
close(icmp_socket);
close(udp_socket);
return -1;
}
net_state->platform.ip4_present = true;
net_state->platform.ip4_socket_raw = false;
net_state->platform.ip4_txrx_icmp_socket = icmp_socket;
net_state->platform.ip4_tmp_icmp_socket = icmp_tmp_socket;
net_state->platform.ip4_txrx_udp_socket = udp_socket;
return 0;
}
#endif
/* Open the raw sockets for sending/receiving IPv6 packets */
static
int open_ip6_sockets_raw(
struct net_state_t *net_state)
{
int send_socket_icmp;
int send_socket_udp;
int recv_socket;
send_socket_icmp = socket(AF_INET6, SOCK_RAW, IPPROTO_ICMPV6);
if (send_socket_icmp == -1) {
return -1;
}
send_socket_udp = socket(AF_INET6, SOCK_RAW, IPPROTO_UDP);
if (send_socket_udp == -1) {
close(send_socket_icmp);
return -1;
}
recv_socket = socket(AF_INET6, SOCK_RAW, IPPROTO_ICMPV6);
if (recv_socket == -1) {
close(send_socket_icmp);
close(send_socket_udp);
return -1;
}
net_state->platform.ip6_present = true;
net_state->platform.ip6_socket_raw = true;
net_state->platform.icmp6_send_socket = send_socket_icmp;
net_state->platform.udp6_send_socket = send_socket_udp;
net_state->platform.ip6_recv_socket = recv_socket;
return 0;
}
#ifdef HAVE_LINUX_ERRQUEUE_H
/* Open DGRAM sockets for sending/receiving IPv6 packets */
static
int open_ip6_sockets_dgram(
struct net_state_t *net_state)
{
int icmp_socket;
int udp_socket;
#ifdef HAVE_LINUX_ERRQUEUE_H
int val = 1;
#endif
icmp_socket = socket(AF_INET6, SOCK_DGRAM, IPPROTO_ICMPV6);
if (icmp_socket == -1) {
return -1;
}
#ifdef HAVE_LINUX_ERRQUEUE_H
if (setsockopt(icmp_socket, SOL_IPV6, IPV6_RECVERR, &val, sizeof(val)) < 0) {
return -1;
}
#endif
udp_socket = socket(AF_INET6, SOCK_DGRAM, IPPROTO_UDP);
if (udp_socket == -1) {
close(icmp_socket);
return -1;
}
#ifdef HAVE_LINUX_ERRQUEUE_H
if (setsockopt(udp_socket, SOL_IPV6, IPV6_RECVERR, &val, sizeof(val)) < 0) {
close(icmp_socket);
close(udp_socket);
return -1;
}
#endif
net_state->platform.ip6_present = true;
net_state->platform.ip6_socket_raw = false;
net_state->platform.ip6_txrx_icmp_socket = icmp_socket;
net_state->platform.ip6_txrx_udp_socket = udp_socket;
return 0;
}
#endif
/*
The first half of the net state initialization. Since this
happens with elevated privileges, this is kept as minimal
as possible to minimize security risk.
*/
void init_net_state_privileged(
struct net_state_t *net_state)
{
int ip4_err = 0;
int ip6_err = 0;
memset(net_state, 0, sizeof(struct net_state_t));
net_state->platform.next_sequence = MIN_PORT;
if (open_ip4_sockets_raw(net_state)) {
#ifdef HAVE_LINUX_ERRQUEUE_H
/* fall back to using unprivileged sockets */
if (open_ip4_sockets_dgram(net_state)) {
ip4_err = errno;
}
#endif
}
if (open_ip6_sockets_raw(net_state)) {
#ifdef HAVE_LINUX_ERRQUEUE_H
/* fall back to using unprivileged sockets */
if (open_ip6_sockets_dgram(net_state)) {
ip6_err = errno;
}
#endif
}
/*
If we couldn't open either IPv4 or IPv6 sockets, we can't do
much, so print errors and exit.
*/
if (!net_state->platform.ip4_present
&& !net_state->platform.ip6_present) {
error(0, ip4_err, "Failure to open IPv4 sockets");
error(0, ip6_err, "Failure to open IPv6 sockets");
exit(EXIT_FAILURE);
}
}
/*
The second half of net state initialization, which is run
at normal privilege levels.
*/
void init_net_state(
struct net_state_t *net_state)
{
if (net_state->platform.ip4_socket_raw) {
set_socket_nonblocking(net_state->platform.ip4_recv_socket);
} else {
set_socket_nonblocking(net_state->platform.ip4_txrx_icmp_socket);
set_socket_nonblocking(net_state->platform.ip4_txrx_udp_socket);
}
if (net_state->platform.ip6_socket_raw) {
set_socket_nonblocking(net_state->platform.ip6_recv_socket);
} else {
set_socket_nonblocking(net_state->platform.ip6_txrx_icmp_socket);
set_socket_nonblocking(net_state->platform.ip6_txrx_udp_socket);
}
if (net_state->platform.ip4_present) {
check_length_order(net_state);
}
check_sctp_support(net_state);
}
/*
Returns true if we were able to open sockets for a particular
IP protocol version.
*/
bool is_ip_version_supported(
struct net_state_t *net_state,
int ip_version)
{
if (ip_version == 4) {
return net_state->platform.ip4_present;
} else if (ip_version == 6) {
return net_state->platform.ip6_present;
} else {
return false;
}
}
/* Returns true if we can transmit probes using the specified protocol */
bool is_protocol_supported(
struct net_state_t * net_state,
int protocol)
{
if (protocol == IPPROTO_ICMP) {
return true;
}
if (protocol == IPPROTO_UDP) {
return true;
}
if (protocol == IPPROTO_TCP) {
return true;
}
#ifdef IPPROTO_SCTP
if (protocol == IPPROTO_SCTP) {
return net_state->platform.sctp_support;
}
#endif
return false;
}
/* Report an error during send_probe based on the errno value */
static
void report_packet_error(
int command_token)
{
if (errno == EINVAL) {
printf("%d invalid-argument\n", command_token);
} else if (errno == ENETDOWN) {
printf("%d network-down\n", command_token);
} else if (errno == ENETUNREACH) {
printf("%d no-route\n", command_token);
} else if (errno == EHOSTUNREACH) {
printf("%d no-route\n", command_token);
} else if (errno == EPERM) {
printf("%d permission-denied\n", command_token);
} else if (errno == EADDRINUSE) {
printf("%d address-in-use\n", command_token);
} else if (errno == EADDRNOTAVAIL) {
printf("%d address-not-available\n", command_token);
} else if (errno == ETIMEDOUT) {
printf("%d wait-tcp-respone-timeout\n", command_token);
} else {
printf("%d unexpected-error errno %d\n", command_token, errno);
}
}
/* Craft a custom ICMP packet for a network probe. */
void send_probe(
struct net_state_t *net_state,
const struct probe_param_t *param)
{
char packet[PACKET_BUFFER_SIZE];
struct probe_t *probe;
int trytimes;
int packet_size;
probe = alloc_probe(net_state, param->command_token);
if (probe == NULL) {
printf("%d probes-exhausted\n", param->command_token);
return;
}
if (resolve_probe_addresses(net_state, param, &probe->remote_addr,
&probe->local_addr)) {
printf("%d invalid-argument\n", param->command_token);
free_probe(net_state, probe);
return;
}
if (gettimeofday(&probe->platform.departure_time, NULL)) {
error(EXIT_FAILURE, errno, "gettimeofday failure");
}
// there might be an off-by-one in the number of tries here.
// this is intentional. It is no use exhausting the very last
// open port. Max 10 retries would've been acceptable too I think.
for (trytimes=MIN_PORT; trytimes < MAX_PORT; trytimes++) {
packet_size = construct_packet(net_state, &probe->platform.socket,
probe, packet, PACKET_BUFFER_SIZE,
param);
if (packet_size > 0) break; // no retry if we succeed.
if ((param->protocol != IPPROTO_TCP) &&
(param->protocol != IPPROTO_SCTP)) break; // no retry if not TCP/SCTP
if ((errno != EADDRINUSE) && (errno != EADDRNOTAVAIL)) {
break; // no retry
}
probe->sequence = net_state->platform.next_sequence++;
if (net_state->platform.next_sequence > MAX_PORT) {
net_state->platform.next_sequence = MIN_PORT;
}
}
if (packet_size < 0) {
/*
When using a stream protocol, FreeBSD will return ECONNREFUSED
when connecting to localhost if the port doesn't exist,
even if the socket is non-blocking, so we should be
prepared for that.
*/
if (errno == ECONNREFUSED) {
receive_probe(net_state, probe, ICMP_ECHOREPLY,
&probe->remote_addr, NULL, 0, NULL);
} else {
report_packet_error(param->command_token);
free_probe(net_state, probe);
}
return;
}
if (packet_size > 0) {
if (send_packet(net_state, param, probe->sequence,
packet, packet_size, &probe->remote_addr) == -1) {
report_packet_error(param->command_token);
free_probe(net_state, probe);
return;
}
}
probe->platform.timeout_time = probe->platform.departure_time;
probe->platform.timeout_time.tv_sec += param->timeout;
}
/* When allocating a probe, assign it a unique port number */
void platform_alloc_probe(
struct net_state_t *net_state,
struct probe_t *probe)
{
probe->sequence = net_state->platform.next_sequence++;
if (net_state->platform.next_sequence > MAX_PORT) {
net_state->platform.next_sequence = MIN_PORT;
}
}
/*
When freeing the probe, close the socket for the probe,
if one has been opened
*/
void platform_free_probe(
struct probe_t *probe)
{
if (probe->platform.socket) {
close(probe->platform.socket);
probe->platform.socket = 0;
}
}
/*
Compute the round trip time of a just-received probe and pass it
to the platform agnostic response handling.
*/
void receive_probe(
struct net_state_t *net_state,
struct probe_t *probe,
int icmp_type,
const struct sockaddr_storage *remote_addr,
struct timeval *timestamp,
int mpls_count,
struct mpls_label_t *mpls)
{
unsigned int round_trip_us;
struct timeval *departure_time = &probe->platform.departure_time;
struct timeval now;
if (timestamp == NULL) {
if (gettimeofday(&now, NULL)) {
error(EXIT_FAILURE, errno, "gettimeofday failure");
}
timestamp = &now;
}
round_trip_us =
(timestamp->tv_sec - departure_time->tv_sec) * 1000000 +
timestamp->tv_usec - departure_time->tv_usec;
respond_to_probe(net_state, probe, icmp_type,
remote_addr, round_trip_us, mpls_count, mpls);
}
/*
Read all available packets through our receiving raw socket, and
handle any responses to probes we have previously sent.
*/
static
void receive_replies_from_recv_socket(
struct net_state_t *net_state,
int socket,
received_packet_func_t handle_received_packet)
{
char packet[PACKET_BUFFER_SIZE];
int packet_length;
struct sockaddr_storage remote_addr;
struct timeval timestamp;
int flag = 0;
#ifdef HAVE_LINUX_ERRQUEUE_H
struct cmsghdr *cm;
struct sock_extended_err *ee = NULL;
bool icmp_connrefused_received = false;
bool icmp_hostunreach_received = false;
#endif
/* Read until no more packets are available */
while (true) {
struct iovec iov;
struct msghdr msg;
char control[1024];
memset(&msg, 0, sizeof(msg));
memset(&iov, 0, sizeof(iov));
iov.iov_base = packet;
iov.iov_len = sizeof(packet);
msg.msg_iov = &iov;
msg.msg_iovlen = 1;
msg.msg_name = (struct sockaddr*) &remote_addr;
msg.msg_namelen = sizeof(remote_addr);
msg.msg_control = control;
msg.msg_controllen = sizeof(control);
packet_length = recvmsg(socket, &msg, flag);
/*
Get the time immediately after reading the packet to
keep the timing as precise as we can.
*/
if (gettimeofday(&timestamp, NULL)) {
error(EXIT_FAILURE, errno, "gettimeofday failure");
}
if (packet_length == -1) {
/*
EAGAIN will be returned if there is no current packet
available.
*/
if (errno == EAGAIN) {
return;
}
/*
EINTER will be returned if we received a signal during
receive.
*/
if (errno == EINTR) {
/* clear error */
int so_err;
socklen_t so_err_size = sizeof(so_err);
int err;
do {
err = getsockopt(socket, SOL_SOCKET, SO_ERROR, &so_err, &so_err_size);
} while (err < 0 && errno == EINTR);
continue;
}
/* handle error received in error queue */
if (errno == EHOSTUNREACH) {
/* potential error caused by ttl, read inner icmp hdr from err queue */
#ifdef HAVE_LINUX_ERRQUEUE_H
icmp_hostunreach_received = true;
flag |= MSG_ERRQUEUE;
#endif
continue;
}
if (errno == ECONNREFUSED) {
/* udp packet reached dst, read inner udp hdr from err queue */
#ifdef HAVE_LINUX_ERRQUEUE_H
icmp_connrefused_received = true;
flag |= MSG_ERRQUEUE;
#endif
continue;
}
error(EXIT_FAILURE, errno, "Failure receiving replies");
}
#ifdef HAVE_LINUX_ERRQUEUE_H
/* get src ip for packets read from err queue */
if (flag & MSG_ERRQUEUE) {
for (cm = CMSG_FIRSTHDR(&msg); cm; cm = CMSG_NXTHDR(&msg, cm)) {
if (cm->cmsg_level == SOL_IP) {
if (cm->cmsg_type == IP_RECVERR) {
ee = (struct sock_extended_err *) CMSG_DATA(cm);
}
}
else if (cm->cmsg_level == SOL_IPV6) {
if (cm->cmsg_type == IPV6_RECVERR) {
ee = (struct sock_extended_err *) CMSG_DATA(cm);
}
}
}
if (ee) {
memcpy(&remote_addr, SO_EE_OFFENDER(ee), sizeof(remote_addr));
}
}
#ifdef SO_PROTOCOL
if (icmp_connrefused_received) {
/* using ICMP type ICMP_ECHOREPLY is not a bug, it is an
indication of successfully reaching dst host.
*/
handle_error_queue_packet(net_state, &remote_addr, ICMP_ECHOREPLY, IPPROTO_UDP,
packet, packet_length, &timestamp);
} else if (icmp_hostunreach_received) {
/* handle packet based on send socket protocol */
int proto, length = sizeof(int);
if (getsockopt(socket, SOL_SOCKET, SO_PROTOCOL, &proto, &length) < 0) {
error(EXIT_FAILURE, errno, "getsockopt SO_PROTOCOL error");
}
handle_error_queue_packet(net_state, &remote_addr, ICMP_TIME_EXCEEDED, proto,
packet, packet_length, &timestamp);
} else {
#endif
#endif
/* ICMP packets received from raw socket */
handle_received_packet(net_state, &remote_addr, packet,
packet_length, &timestamp);
#ifdef HAVE_LINUX_ERRQUEUE_H
#ifdef SO_PROTOCOL
}
#endif
#endif
}
}
/*
Attempt to send using the probe's socket, in order to check whether
the connection has completed, for stream oriented protocols such as
TCP.
*/
static
void receive_replies_from_probe_socket(
struct net_state_t *net_state,
struct probe_t *probe)
{
int probe_socket;
struct timeval zero_time;
int err;
int err_length = sizeof(int);
fd_set write_set;
probe_socket = probe->platform.socket;
if (!probe_socket) {
return;
}
FD_ZERO(&write_set);
FD_SET(probe_socket, &write_set);
zero_time.tv_sec = 0;
zero_time.tv_usec = 0;
if (select(probe_socket + 1, NULL, &write_set, NULL, &zero_time) == -1) {
if (errno == EAGAIN) {
return;
} else {
error(EXIT_FAILURE, errno, "probe socket select error");
}
}
/*
If the socket is writable, the connection attempt has completed.
*/
if (!FD_ISSET(probe_socket, &write_set)) {
return;
}
if (getsockopt(probe_socket, SOL_SOCKET, SO_ERROR, &err, &err_length)) {
error(EXIT_FAILURE, errno, "probe socket SO_ERROR");
}
/*
If the connection complete successfully, or was refused, we can
assume our probe arrived at the destination.
*/
if (!err || err == ECONNREFUSED) {
receive_probe(net_state, probe, ICMP_ECHOREPLY,
&probe->remote_addr, NULL, 0, NULL);
} else {
errno = err;
report_packet_error(probe->token);
free_probe(net_state, probe);
}
}
/* Check both the IPv4 and IPv6 sockets for incoming packets */
void receive_replies(
struct net_state_t *net_state)
{
struct probe_t *probe;
struct probe_t *probe_safe_iter;
if (net_state->platform.ip4_present) {
if (net_state->platform.ip4_socket_raw) {
receive_replies_from_recv_socket(net_state,
net_state->platform.
ip4_recv_socket,
handle_received_ip4_packet);
} else {
receive_replies_from_recv_socket(net_state,
net_state->platform.
ip4_txrx_icmp_socket,
handle_received_ip4_packet);
receive_replies_from_recv_socket(net_state,
net_state->platform.
ip4_txrx_udp_socket,
handle_received_ip4_packet);
}
}
if (net_state->platform.ip6_present) {
if (net_state->platform.ip6_socket_raw) {
receive_replies_from_recv_socket(net_state,
net_state->platform.
ip6_recv_socket,
handle_received_ip6_packet);
} else {
receive_replies_from_recv_socket(net_state,
net_state->platform.
ip6_txrx_icmp_socket,
handle_received_ip6_packet);
receive_replies_from_recv_socket(net_state,
net_state->platform.
ip6_txrx_udp_socket,
handle_received_ip6_packet);
}
}
LIST_FOREACH_SAFE(probe, &net_state->outstanding_probes,
probe_list_entry, probe_safe_iter) {
receive_replies_from_probe_socket(net_state, probe);
}
}
/*
Put all of our probe sockets in the read set used for an upcoming
select so we can wake when any of them become readable.
*/
int gather_probe_sockets(
const struct net_state_t *net_state,
fd_set * write_set)
{
int probe_socket;
int nfds;
const struct probe_t *probe;
nfds = 0;
LIST_FOREACH(probe, &net_state->outstanding_probes, probe_list_entry) {
probe_socket = probe->platform.socket;
if (probe_socket) {
FD_SET(probe_socket, write_set);
if (probe_socket >= nfds) {
nfds = probe_socket + 1;
}
}
}
return nfds;
}
/*
Check for any probes for which we have not received a response
for some time, and report a time-out, assuming that we won't
receive a future reply.
*/
void check_probe_timeouts(
struct net_state_t *net_state)
{
struct timeval now;
struct probe_t *probe;
struct probe_t *probe_safe_iter;
if (gettimeofday(&now, NULL)) {
error(EXIT_FAILURE, errno, "gettimeofday failure");
}
LIST_FOREACH_SAFE(probe, &net_state->outstanding_probes,
probe_list_entry, probe_safe_iter) {
if (compare_timeval(probe->platform.timeout_time, now) < 0) {
/* Report timeout to the command stream */
printf("%d no-reply\n", probe->token);
free_probe(net_state, probe);
}
}
}
/*
Find the remaining time until the next probe times out.
This may be a negative value if the next probe timeout has
already elapsed.
Returns false if no probes are currently outstanding, and true
if a timeout value for the next probe exists.
*/
bool get_next_probe_timeout(
const struct net_state_t *net_state,
struct timeval *timeout)
{
bool have_timeout;
const struct probe_t *probe;
struct timeval now;
struct timeval probe_timeout;
if (gettimeofday(&now, NULL)) {
error(EXIT_FAILURE, errno, "gettimeofday failure");
}
have_timeout = false;
LIST_FOREACH(probe, &net_state->outstanding_probes, probe_list_entry) {
probe_timeout.tv_sec =
probe->platform.timeout_time.tv_sec - now.tv_sec;
probe_timeout.tv_usec =
probe->platform.timeout_time.tv_usec - now.tv_usec;
normalize_timeval(&probe_timeout);
if (have_timeout) {
if (compare_timeval(probe_timeout, *timeout) < 0) {
/* If this probe has a sooner timeout, store it instead */
*timeout = probe_timeout;
}
} else {
*timeout = probe_timeout;
have_timeout = true;
}
}
return have_timeout;
}