blob: 28c49e218866b1221fe0da2aeac9bb0e77ab3721 [file] [log] [blame]
/* vi: set sw=4 ts=4: */
/*
* ifplugd for busybox, based on ifplugd 0.28 (written by Lennart Poettering).
*
* Copyright (C) 2009 Maksym Kryzhanovskyy <xmaks@email.cz>
*
* Licensed under GPLv2 or later, see file LICENSE in this source tree.
*/
//usage:#define ifplugd_trivial_usage
//usage: "[OPTIONS]"
//usage:#define ifplugd_full_usage "\n\n"
//usage: "Network interface plug detection daemon\n"
//usage: "\n -n Don't daemonize"
//usage: "\n -s Don't log to syslog"
//usage: "\n -i IFACE Interface"
//usage: "\n -f/-F Treat link detection error as link down/link up"
//usage: "\n (otherwise exit on error)"
//usage: "\n -a Don't up interface at each link probe"
//usage: "\n -M Monitor creation/destruction of interface"
//usage: "\n (otherwise it must exist)"
//usage: "\n -r PROG Script to run"
//usage: "\n -x ARG Extra argument for script"
//usage: "\n -I Don't exit on nonzero exit code from script"
//usage: "\n -p Don't run \"up\" script on startup"
//usage: "\n -q Don't run \"down\" script on exit"
//usage: "\n -l Always run script on startup"
//usage: "\n -t SECS Poll time in seconds"
//usage: "\n -u SECS Delay before running script after link up"
//usage: "\n -d SECS Delay after link down"
//usage: "\n -m MODE API mode (mii, priv, ethtool, wlan, iff, auto)"
//usage: "\n -k Kill running daemon"
#include "libbb.h"
#include "fix_u32.h"
#include <linux/if.h>
#include <linux/mii.h>
#include <linux/ethtool.h>
#ifdef HAVE_NET_ETHERNET_H
/* musl breakage:
* In file included from /usr/include/net/ethernet.h:10,
* from networking/ifplugd.c:41:
* /usr/include/netinet/if_ether.h:96: error: redefinition of 'struct ethhdr'
*
* Build succeeds without it on musl. Commented it out.
* If on your system you need it, consider removing <linux/ethtool.h>
* and copy-pasting its definitions here (<linux/ethtool.h> is what pulls in
* conflicting definition of struct ethhdr on musl).
*/
/* # include <net/ethernet.h> */
#endif
#include <linux/netlink.h>
#include <linux/rtnetlink.h>
#include <linux/sockios.h>
#include <syslog.h>
#define __user
#include <linux/wireless.h>
#ifndef ETH_ALEN
# define ETH_ALEN 6
#endif
/*
From initial port to busybox, removed most of the redundancy by
converting implementation of a polymorphic interface to the strict
functional style. The main role is run a script when link state
changed, other activities like audio signal or detailed reports
are on the script itself.
One questionable point of the design is netlink usage:
We have 1 second timeout by default to poll the link status,
it is short enough so that there are no real benefits in
using netlink to get "instantaneous" interface creation/deletion
notifications. We can check for interface existence by just
doing some fast ioctl using its name.
Netlink code then can be just dropped (1k or more?)
*/
#define IFPLUGD_ENV_PREVIOUS "IFPLUGD_PREVIOUS"
#define IFPLUGD_ENV_CURRENT "IFPLUGD_CURRENT"
enum {
FLAG_NO_AUTO = 1 << 0, // -a, Do not enable interface automatically
FLAG_NO_DAEMON = 1 << 1, // -n, Do not daemonize
FLAG_NO_SYSLOG = 1 << 2, // -s, Do not use syslog, use stderr instead
FLAG_IGNORE_FAIL = 1 << 3, // -f, Ignore detection failure, retry instead (failure is treated as DOWN)
FLAG_IGNORE_FAIL_POSITIVE = 1 << 4, // -F, Ignore detection failure, retry instead (failure is treated as UP)
FLAG_IFACE = 1 << 5, // -i, Specify ethernet interface
FLAG_RUN = 1 << 6, // -r, Specify program to execute
FLAG_IGNORE_RETVAL = 1 << 7, // -I, Don't exit on nonzero return value of program executed
FLAG_POLL_TIME = 1 << 8, // -t, Specify poll time in seconds
FLAG_DELAY_UP = 1 << 9, // -u, Specify delay for configuring interface
FLAG_DELAY_DOWN = 1 << 10, // -d, Specify delay for deconfiguring interface
FLAG_API_MODE = 1 << 11, // -m, Force API mode (mii, priv, ethtool, wlan, auto)
FLAG_NO_STARTUP = 1 << 12, // -p, Don't run script on daemon startup
FLAG_NO_SHUTDOWN = 1 << 13, // -q, Don't run script on daemon quit
FLAG_INITIAL_DOWN = 1 << 14, // -l, Run "down" script on startup if no cable is detected
FLAG_EXTRA_ARG = 1 << 15, // -x, Specify an extra argument for action script
FLAG_MONITOR = 1 << 16, // -M, Use interface monitoring
#if ENABLE_FEATURE_PIDFILE
FLAG_KILL = 1 << 17, // -k, Kill a running daemon
#endif
};
#if ENABLE_FEATURE_PIDFILE
# define OPTION_STR "+ansfFi:r:It:+u:+d:+m:pqlx:Mk"
#else
# define OPTION_STR "+ansfFi:r:It:+u:+d:+m:pqlx:M"
#endif
enum { // interface status
IFSTATUS_ERR = -1,
IFSTATUS_DOWN = 0,
IFSTATUS_UP = 1,
};
enum { // constant fds
ioctl_fd = 3,
netlink_fd = 4,
};
struct globals {
smallint iface_last_status;
smallint iface_prev_status;
smallint iface_exists;
smallint api_method_num;
/* Used in getopt32, must have sizeof == sizeof(int) */
unsigned poll_time;
unsigned delay_up;
unsigned delay_down;
const char *iface;
const char *api_mode;
const char *script_name;
const char *extra_arg;
};
#define G (*ptr_to_globals)
#define INIT_G() do { \
SET_PTR_TO_GLOBALS(xzalloc(sizeof(G))); \
G.iface_last_status = -1; \
G.iface_exists = 1; \
G.poll_time = 1; \
G.delay_down = 5; \
G.iface = "eth0"; \
G.api_mode = "a"; \
G.script_name = "/etc/ifplugd/ifplugd.action"; \
} while (0)
/* Utility routines */
static void set_ifreq_to_ifname(struct ifreq *ifreq)
{
memset(ifreq, 0, sizeof(struct ifreq));
strncpy_IFNAMSIZ(ifreq->ifr_name, G.iface);
}
static int network_ioctl(int request, void* data, const char *errmsg)
{
int r = ioctl(ioctl_fd, request, data);
if (r < 0 && errmsg)
bb_perror_msg("%s failed", errmsg);
return r;
}
/* Link detection routines and table */
static smallint detect_link_mii(void)
{
/* char buffer instead of bona-fide struct avoids aliasing warning */
char buf[sizeof(struct ifreq)];
struct ifreq *const ifreq = (void *)buf;
struct mii_ioctl_data *mii = (void *)&ifreq->ifr_data;
set_ifreq_to_ifname(ifreq);
if (network_ioctl(SIOCGMIIPHY, ifreq, "SIOCGMIIPHY") < 0) {
return IFSTATUS_ERR;
}
mii->reg_num = 1;
if (network_ioctl(SIOCGMIIREG, ifreq, "SIOCGMIIREG") < 0) {
return IFSTATUS_ERR;
}
return (mii->val_out & 0x0004) ? IFSTATUS_UP : IFSTATUS_DOWN;
}
static smallint detect_link_priv(void)
{
/* char buffer instead of bona-fide struct avoids aliasing warning */
char buf[sizeof(struct ifreq)];
struct ifreq *const ifreq = (void *)buf;
struct mii_ioctl_data *mii = (void *)&ifreq->ifr_data;
set_ifreq_to_ifname(ifreq);
if (network_ioctl(SIOCDEVPRIVATE, ifreq, "SIOCDEVPRIVATE") < 0) {
return IFSTATUS_ERR;
}
mii->reg_num = 1;
if (network_ioctl(SIOCDEVPRIVATE+1, ifreq, "SIOCDEVPRIVATE+1") < 0) {
return IFSTATUS_ERR;
}
return (mii->val_out & 0x0004) ? IFSTATUS_UP : IFSTATUS_DOWN;
}
static smallint detect_link_ethtool(void)
{
struct ifreq ifreq;
struct ethtool_value edata;
set_ifreq_to_ifname(&ifreq);
edata.cmd = ETHTOOL_GLINK;
ifreq.ifr_data = (void*) &edata;
if (network_ioctl(SIOCETHTOOL, &ifreq, "ETHTOOL_GLINK") < 0) {
return IFSTATUS_ERR;
}
return edata.data ? IFSTATUS_UP : IFSTATUS_DOWN;
}
static smallint detect_link_iff(void)
{
struct ifreq ifreq;
set_ifreq_to_ifname(&ifreq);
if (network_ioctl(SIOCGIFFLAGS, &ifreq, "SIOCGIFFLAGS") < 0) {
return IFSTATUS_ERR;
}
/* If IFF_UP is not set (interface is down), IFF_RUNNING is never set
* regardless of link status. Simply continue to report last status -
* no point in reporting spurious link downs if interface is disabled
* by admin. When/if it will be brought up,
* we'll report real link status.
*/
if (!(ifreq.ifr_flags & IFF_UP) && G.iface_last_status != IFSTATUS_ERR)
return G.iface_last_status;
return (ifreq.ifr_flags & IFF_RUNNING) ? IFSTATUS_UP : IFSTATUS_DOWN;
}
static smallint detect_link_wlan(void)
{
int i;
struct iwreq iwrequest;
uint8_t mac[ETH_ALEN];
memset(&iwrequest, 0, sizeof(iwrequest));
strncpy_IFNAMSIZ(iwrequest.ifr_ifrn.ifrn_name, G.iface);
if (network_ioctl(SIOCGIWAP, &iwrequest, "SIOCGIWAP") < 0) {
return IFSTATUS_ERR;
}
memcpy(mac, &iwrequest.u.ap_addr.sa_data, ETH_ALEN);
if (mac[0] == 0xFF || mac[0] == 0x44 || mac[0] == 0x00) {
for (i = 1; i < ETH_ALEN; ++i) {
if (mac[i] != mac[0])
return IFSTATUS_UP;
}
return IFSTATUS_DOWN;
}
return IFSTATUS_UP;
}
enum { // api mode
API_ETHTOOL, // 'e'
API_MII, // 'm'
API_PRIVATE, // 'p'
API_WLAN, // 'w'
API_IFF, // 'i'
API_AUTO, // 'a'
};
static const char api_modes[] ALIGN1 = "empwia";
static const struct {
const char *name;
smallint (*func)(void);
} method_table[] = {
{ "SIOCETHTOOL" , &detect_link_ethtool },
{ "SIOCGMIIPHY" , &detect_link_mii },
{ "SIOCDEVPRIVATE" , &detect_link_priv },
{ "wireless extension", &detect_link_wlan },
{ "IFF_RUNNING" , &detect_link_iff },
};
static const char *strstatus(int status)
{
if (status == IFSTATUS_ERR)
return "error";
return "down\0up" + (status * 5);
}
static int run_script(const char *action)
{
char *env_PREVIOUS, *env_CURRENT;
char *argv[5];
int r;
bb_error_msg("executing '%s %s %s'", G.script_name, G.iface, action);
argv[0] = (char*) G.script_name;
argv[1] = (char*) G.iface;
argv[2] = (char*) action;
argv[3] = (char*) G.extra_arg;
argv[4] = NULL;
env_PREVIOUS = xasprintf("%s=%s", IFPLUGD_ENV_PREVIOUS, strstatus(G.iface_prev_status));
putenv(env_PREVIOUS);
env_CURRENT = xasprintf("%s=%s", IFPLUGD_ENV_CURRENT, strstatus(G.iface_last_status));
putenv(env_CURRENT);
/* r < 0 - can't exec, 0 <= r < 0x180 - exited, >=0x180 - killed by sig (r-0x180) */
r = spawn_and_wait(argv);
unsetenv(IFPLUGD_ENV_PREVIOUS);
unsetenv(IFPLUGD_ENV_CURRENT);
free(env_PREVIOUS);
free(env_CURRENT);
bb_error_msg("exit code: %d", r & 0xff);
return (option_mask32 & FLAG_IGNORE_RETVAL) ? 0 : r;
}
static void up_iface(void)
{
struct ifreq ifrequest;
if (!G.iface_exists)
return;
set_ifreq_to_ifname(&ifrequest);
if (network_ioctl(SIOCGIFFLAGS, &ifrequest, "getting interface flags") < 0) {
G.iface_exists = 0;
return;
}
if (!(ifrequest.ifr_flags & IFF_UP)) {
ifrequest.ifr_flags |= IFF_UP;
/* Let user know we mess up with interface */
bb_error_msg("upping interface");
if (network_ioctl(SIOCSIFFLAGS, &ifrequest, "setting interface flags") < 0)
xfunc_die();
}
#if 0 /* why do we mess with IP addr? It's not our business */
if (network_ioctl(SIOCGIFADDR, &ifrequest, "can't get interface address") < 0) {
} else if (ifrequest.ifr_addr.sa_family != AF_INET) {
bb_perror_msg("the interface is not IP-based");
} else {
((struct sockaddr_in*)(&ifrequest.ifr_addr))->sin_addr.s_addr = INADDR_ANY;
network_ioctl(SIOCSIFADDR, &ifrequest, "can't set interface address");
}
network_ioctl(SIOCGIFFLAGS, &ifrequest, "can't get interface flags");
#endif
}
static void maybe_up_new_iface(void)
{
if (!(option_mask32 & FLAG_NO_AUTO))
up_iface();
#if 0 /* bloat */
struct ifreq ifrequest;
struct ethtool_drvinfo driver_info;
set_ifreq_to_ifname(&ifrequest);
driver_info.cmd = ETHTOOL_GDRVINFO;
ifrequest.ifr_data = &driver_info;
if (network_ioctl(SIOCETHTOOL, &ifrequest, NULL) == 0) {
char buf[sizeof("/xx:xx:xx:xx:xx:xx")];
/* Get MAC */
buf[0] = '\0';
set_ifreq_to_ifname(&ifrequest);
if (network_ioctl(SIOCGIFHWADDR, &ifrequest, NULL) == 0) {
sprintf(buf, "/%02X:%02X:%02X:%02X:%02X:%02X",
(uint8_t)(ifrequest.ifr_hwaddr.sa_data[0]),
(uint8_t)(ifrequest.ifr_hwaddr.sa_data[1]),
(uint8_t)(ifrequest.ifr_hwaddr.sa_data[2]),
(uint8_t)(ifrequest.ifr_hwaddr.sa_data[3]),
(uint8_t)(ifrequest.ifr_hwaddr.sa_data[4]),
(uint8_t)(ifrequest.ifr_hwaddr.sa_data[5]));
}
bb_error_msg("using interface %s%s with driver<%s> (version: %s)",
G.iface, buf, driver_info.driver, driver_info.version);
}
#endif
if (G.api_mode[0] == 'a')
G.api_method_num = API_AUTO;
}
static smallint detect_link(void)
{
smallint status;
if (!G.iface_exists)
return (option_mask32 & FLAG_MONITOR) ? IFSTATUS_DOWN : IFSTATUS_ERR;
/* Some drivers can't detect link status when the interface is down.
* I imagine detect_link_iff() is the most vulnerable.
* That's why -a "noauto" in an option, not a hardwired behavior.
*/
if (!(option_mask32 & FLAG_NO_AUTO))
up_iface();
if (G.api_method_num == API_AUTO) {
int i;
smallint sv_logmode;
sv_logmode = logmode;
for (i = 0; i < ARRAY_SIZE(method_table); i++) {
logmode = LOGMODE_NONE;
status = method_table[i].func();
logmode = sv_logmode;
if (status != IFSTATUS_ERR) {
G.api_method_num = i;
bb_error_msg("using %s detection mode", method_table[i].name);
break;
}
}
} else {
status = method_table[G.api_method_num].func();
}
if (status == IFSTATUS_ERR) {
if (option_mask32 & FLAG_IGNORE_FAIL)
status = IFSTATUS_DOWN;
else if (option_mask32 & FLAG_IGNORE_FAIL_POSITIVE)
status = IFSTATUS_UP;
else if (G.api_mode[0] == 'a')
bb_error_msg("can't detect link status");
}
if (status != G.iface_last_status) {
G.iface_prev_status = G.iface_last_status;
G.iface_last_status = status;
}
return status;
}
static NOINLINE int check_existence_through_netlink(void)
{
int iface_len;
/* Buffer was 1K, but on linux-3.9.9 it was reported to be too small.
* netlink.h: "limit to 8K to avoid MSG_TRUNC when PAGE_SIZE is very large".
* Note: on error returns (-1) we exit, no need to free replybuf.
*/
enum { BUF_SIZE = 8 * 1024 };
char *replybuf = xmalloc(BUF_SIZE);
iface_len = strlen(G.iface);
while (1) {
struct nlmsghdr *mhdr;
ssize_t bytes;
bytes = recv(netlink_fd, replybuf, BUF_SIZE, MSG_DONTWAIT);
if (bytes < 0) {
if (errno == EAGAIN)
goto ret;
if (errno == EINTR)
continue;
bb_perror_msg("netlink: recv");
return -1;
}
mhdr = (struct nlmsghdr*)replybuf;
while (bytes > 0) {
if (!NLMSG_OK(mhdr, bytes)) {
bb_error_msg("netlink packet too small or truncated");
return -1;
}
if (mhdr->nlmsg_type == RTM_NEWLINK || mhdr->nlmsg_type == RTM_DELLINK) {
struct rtattr *attr;
int attr_len;
if (mhdr->nlmsg_len < NLMSG_LENGTH(sizeof(struct ifinfomsg))) {
bb_error_msg("netlink packet too small or truncated");
return -1;
}
attr = IFLA_RTA(NLMSG_DATA(mhdr));
attr_len = IFLA_PAYLOAD(mhdr);
while (RTA_OK(attr, attr_len)) {
if (attr->rta_type == IFLA_IFNAME) {
int len = RTA_PAYLOAD(attr);
if (len > IFNAMSIZ)
len = IFNAMSIZ;
if (iface_len <= len
&& strncmp(G.iface, RTA_DATA(attr), len) == 0
) {
G.iface_exists = (mhdr->nlmsg_type == RTM_NEWLINK);
}
}
attr = RTA_NEXT(attr, attr_len);
}
}
mhdr = NLMSG_NEXT(mhdr, bytes);
}
}
ret:
free(replybuf);
return G.iface_exists;
}
#if ENABLE_FEATURE_PIDFILE
static NOINLINE pid_t read_pid(const char *filename)
{
int len;
char buf[128];
len = open_read_close(filename, buf, 127);
if (len > 0) {
buf[len] = '\0';
/* returns ULONG_MAX on error => -1 */
return bb_strtoul(buf, NULL, 10);
}
return 0;
}
#endif
int ifplugd_main(int argc, char **argv) MAIN_EXTERNALLY_VISIBLE;
int ifplugd_main(int argc UNUSED_PARAM, char **argv)
{
int iface_status;
int delay_time;
const char *iface_status_str;
struct pollfd netlink_pollfd[1];
unsigned opts;
const char *api_mode_found;
#if ENABLE_FEATURE_PIDFILE
char *pidfile_name;
pid_t pid_from_pidfile;
#endif
INIT_G();
opts = getopt32(argv, OPTION_STR,
&G.iface, &G.script_name, &G.poll_time, &G.delay_up,
&G.delay_down, &G.api_mode, &G.extra_arg);
G.poll_time *= 1000;
applet_name = xasprintf("ifplugd(%s)", G.iface);
#if ENABLE_FEATURE_PIDFILE
pidfile_name = xasprintf(CONFIG_PID_FILE_PATH "/ifplugd.%s.pid", G.iface);
pid_from_pidfile = read_pid(pidfile_name);
if (opts & FLAG_KILL) {
if (pid_from_pidfile > 0)
/* Upstream tool use SIGINT for -k */
kill(pid_from_pidfile, SIGINT);
return EXIT_SUCCESS;
}
if (pid_from_pidfile > 0 && kill(pid_from_pidfile, 0) == 0)
bb_error_msg_and_die("daemon already running");
#endif
api_mode_found = strchr(api_modes, G.api_mode[0]);
if (!api_mode_found)
bb_error_msg_and_die("unknown API mode '%s'", G.api_mode);
G.api_method_num = api_mode_found - api_modes;
if (!(opts & FLAG_NO_DAEMON))
bb_daemonize_or_rexec(DAEMON_CHDIR_ROOT, argv);
xmove_fd(xsocket(AF_INET, SOCK_DGRAM, 0), ioctl_fd);
if (opts & FLAG_MONITOR) {
struct sockaddr_nl addr;
int fd = xsocket(PF_NETLINK, SOCK_DGRAM, NETLINK_ROUTE);
memset(&addr, 0, sizeof(addr));
addr.nl_family = AF_NETLINK;
addr.nl_groups = RTMGRP_LINK;
addr.nl_pid = getpid();
xbind(fd, (struct sockaddr*)&addr, sizeof(addr));
xmove_fd(fd, netlink_fd);
}
write_pidfile(pidfile_name);
/* this can't be moved before socket creation */
if (!(opts & FLAG_NO_SYSLOG)) {
openlog(applet_name, 0, LOG_DAEMON);
logmode |= LOGMODE_SYSLOG;
}
bb_signals(0
| (1 << SIGINT )
| (1 << SIGTERM)
| (1 << SIGQUIT)
| (1 << SIGHUP ) /* why we ignore it? */
/* | (1 << SIGCHLD) - run_script does not use it anymore */
, record_signo);
bb_error_msg("started: %s", bb_banner);
if (opts & FLAG_MONITOR) {
struct ifreq ifrequest;
set_ifreq_to_ifname(&ifrequest);
G.iface_exists = (network_ioctl(SIOCGIFINDEX, &ifrequest, NULL) == 0);
}
if (G.iface_exists)
maybe_up_new_iface();
iface_status = detect_link();
if (iface_status == IFSTATUS_ERR)
goto exiting;
iface_status_str = strstatus(iface_status);
if (opts & FLAG_MONITOR) {
bb_error_msg("interface %s",
G.iface_exists ? "exists"
: "doesn't exist, waiting");
}
/* else we assume it always exists, but don't mislead user
* by potentially lying that it really exists */
if (G.iface_exists) {
bb_error_msg("link is %s", iface_status_str);
}
if ((!(opts & FLAG_NO_STARTUP)
&& iface_status == IFSTATUS_UP
)
|| (opts & FLAG_INITIAL_DOWN)
) {
if (run_script(iface_status_str) != 0)
goto exiting;
}
/* Main loop */
netlink_pollfd[0].fd = netlink_fd;
netlink_pollfd[0].events = POLLIN;
delay_time = 0;
while (1) {
int iface_status_old;
switch (bb_got_signal) {
case SIGINT:
case SIGTERM:
bb_got_signal = 0;
goto cleanup;
case SIGQUIT:
bb_got_signal = 0;
goto exiting;
default:
bb_got_signal = 0;
break;
}
if (poll(netlink_pollfd,
(opts & FLAG_MONITOR) ? 1 : 0,
G.poll_time
) < 0
) {
if (errno == EINTR)
continue;
bb_perror_msg("poll");
goto exiting;
}
if ((opts & FLAG_MONITOR)
&& (netlink_pollfd[0].revents & POLLIN)
) {
int iface_exists_old;
iface_exists_old = G.iface_exists;
G.iface_exists = check_existence_through_netlink();
if (G.iface_exists < 0) /* error */
goto exiting;
if (iface_exists_old != G.iface_exists) {
bb_error_msg("interface %sappeared",
G.iface_exists ? "" : "dis");
if (G.iface_exists)
maybe_up_new_iface();
}
}
/* note: if !G.iface_exists, returns DOWN */
iface_status_old = iface_status;
iface_status = detect_link();
if (iface_status == IFSTATUS_ERR) {
if (!(opts & FLAG_MONITOR))
goto exiting;
iface_status = IFSTATUS_DOWN;
}
iface_status_str = strstatus(iface_status);
if (iface_status_old != iface_status) {
bb_error_msg("link is %s", iface_status_str);
if (delay_time) {
/* link restored its old status before
* we ran script. don't run the script: */
delay_time = 0;
} else {
delay_time = monotonic_sec();
if (iface_status == IFSTATUS_UP)
delay_time += G.delay_up;
if (iface_status == IFSTATUS_DOWN)
delay_time += G.delay_down;
#if 0 /* if you are back in 1970... */
if (delay_time == 0) {
sleep(1);
delay_time = 1;
}
#endif
}
}
if (delay_time && (int)(monotonic_sec() - delay_time) >= 0) {
if (run_script(iface_status_str) != 0)
goto exiting;
delay_time = 0;
}
} /* while (1) */
cleanup:
if (!(opts & FLAG_NO_SHUTDOWN)
&& (iface_status == IFSTATUS_UP
|| (iface_status == IFSTATUS_DOWN && delay_time)
)
) {
setenv(IFPLUGD_ENV_PREVIOUS, strstatus(iface_status), 1);
setenv(IFPLUGD_ENV_CURRENT, strstatus(-1), 1);
run_script("down\0up"); /* reusing string */
}
exiting:
remove_pidfile(pidfile_name);
bb_error_msg_and_die("exiting");
}