ndhc/ifchd/ifchd.c

965 lines
27 KiB
C
Raw Normal View History

2010-11-12 14:32:18 +05:30
/* ifchd.c - interface change daemon
*
* Copyright (c) 2004-2011 Nicholas J. Kain <njkain at gmail dot com>
* All rights reserved.
2010-11-12 14:32:18 +05:30
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions are met:
2010-11-12 14:32:18 +05:30
*
* - Redistributions of source code must retain the above copyright notice,
* this list of conditions and the following disclaimer.
2010-11-12 14:32:18 +05:30
*
* - Redistributions in binary form must reproduce the above copyright notice,
* this list of conditions and the following disclaimer in the documentation
* and/or other materials provided with the distribution.
*
* THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
* AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
* IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
* ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE
* LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
* CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
* SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
* INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
* CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
* ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
* POSSIBILITY OF SUCH DAMAGE.
2010-11-12 14:32:18 +05:30
*/
#include <unistd.h>
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <sys/socket.h>
#include <sys/un.h>
#include <sys/time.h>
#include <sys/types.h>
2010-12-01 22:52:08 +05:30
#include <sys/epoll.h>
#include <sys/signalfd.h>
2010-11-12 14:32:18 +05:30
#include <sys/stat.h>
#include <fcntl.h>
#include <time.h>
#include <pwd.h>
#include <grp.h>
#include <signal.h>
#include <errno.h>
#include <getopt.h>
#include "ifchd-defines.h"
2010-11-12 16:12:07 +05:30
#include "malloc.h"
2010-11-12 14:32:18 +05:30
#include "log.h"
#include "chroot.h"
#include "pidfile.h"
#include "signals.h"
#include "strlist.h"
#include "ifch_proto.h"
2010-11-12 16:12:07 +05:30
#include "strl.h"
2010-11-12 19:34:43 +05:30
#include "cap.h"
#include "io.h"
2010-11-12 14:32:18 +05:30
#include "linux.h"
enum states {
STATE_NOTHING,
STATE_INTERFACE,
STATE_IP,
STATE_SUBNET,
STATE_TIMEZONE,
STATE_ROUTER,
STATE_DNS,
STATE_LPRSVR,
STATE_HOSTNAME,
STATE_DOMAIN,
STATE_IPTTL,
STATE_MTU,
STATE_BROADCAST,
2011-07-25 12:41:47 +05:30
STATE_NTPSVR,
2010-11-12 14:32:18 +05:30
STATE_WINS
};
2010-12-01 22:52:08 +05:30
static int epollfd, signalFd;
/* Extra two event slots are for signalFd and the listen socket. */
static struct epoll_event events[SOCK_QUEUE+2];
2010-11-12 14:32:18 +05:30
/* Socket fd, current state, and idle time for connections. */
static int sks[SOCK_QUEUE], state[SOCK_QUEUE], idle_time[SOCK_QUEUE];
/* Per-connection buffers. */
static char ibuf[SOCK_QUEUE][MAX_BUF];
/*
* Per-connection pointers into the command lists. Respectively, the
* topmost item on the list, the current item, and the last item on the list.
*/
static strlist_t *head[SOCK_QUEUE], *curl[SOCK_QUEUE], *last[SOCK_QUEUE];
int resolv_conf_fd = -1;
/* int ntp_conf_fd = -1; */
/* If true, allow HOSTNAME changes from dhcp server. */
int allow_hostname = 0;
static uid_t peer_uid;
static gid_t peer_gid;
static pid_t peer_pid;
static int gflags_verbose = 0;
2010-11-12 14:32:18 +05:30
/* Lists of nameservers and search domains. Unfortunately they must be
* per-connection, since otherwise seperate clients could race against
* one another to write out unpredictable data.
*/
static strlist_t *namesvrs[SOCK_QUEUE];
static strlist_t *domains[SOCK_QUEUE];
static void die_nulstr(strlist_t *p)
{
if (!p)
2010-11-12 16:35:46 +05:30
suicide("FATAL - NULL passed to die_nulstr");
2010-11-12 14:32:18 +05:30
if (!p->str)
2010-11-12 16:35:46 +05:30
suicide("FATAL - NULL string in strlist");
2010-11-12 14:32:18 +05:30
}
static void writeordie(int fd, const char *buf, int len)
{
if (safe_write(fd, buf, len) == -1)
suicide("write returned error");
}
2010-12-01 22:52:08 +05:30
static void epoll_add(int fd)
{
struct epoll_event ev;
int r;
ev.events = EPOLLIN | EPOLLRDHUP | EPOLLERR | EPOLLHUP;
ev.data.fd = fd;
r = epoll_ctl(epollfd, EPOLL_CTL_ADD, fd, &ev);
if (r == -1)
suicide("epoll_add failed %s", strerror(errno));
}
static void epoll_del(int fd)
{
struct epoll_event ev;
int r;
ev.events = EPOLLIN | EPOLLRDHUP | EPOLLERR | EPOLLHUP;
ev.data.fd = fd;
r = epoll_ctl(epollfd, EPOLL_CTL_DEL, fd, &ev);
if (r == -1)
suicide("epoll_del failed %s", strerror(errno));
}
/* Abstracts away the details of accept()ing a socket connection. */
2010-11-12 14:32:18 +05:30
/* Writes out each element in a strlist as an argument to a keyword in
* a file. */
static void write_resolve_list(const char *keyword, strlist_t *list)
{
char *buf;
strlist_t *p = list;
unsigned int len;
if (!keyword || resolv_conf_fd == -1)
2010-11-12 16:35:46 +05:30
return;
2010-11-12 14:32:18 +05:30
while (p) {
2010-11-12 16:35:46 +05:30
buf = p->str;
len = strlen(buf);
if (len) {
writeordie(resolv_conf_fd, keyword, strlen(keyword));
writeordie(resolv_conf_fd, buf, strlen(buf));
writeordie(resolv_conf_fd, "\n", 1);
2010-11-12 16:35:46 +05:30
}
p = p->next;
2010-11-12 14:32:18 +05:30
}
}
/* Writes a new resolv.conf based on the information we have received. */
static void write_resolve_conf(int idx)
{
int r;
2010-11-12 14:32:18 +05:30
off_t off;
if (resolv_conf_fd == -1)
2010-11-12 16:35:46 +05:30
return;
2010-11-12 14:32:18 +05:30
if (lseek(resolv_conf_fd, 0, SEEK_SET) == -1)
2010-11-12 16:35:46 +05:30
return;
2010-11-12 14:32:18 +05:30
write_resolve_list("nameserver ", namesvrs[idx]);
write_resolve_list("search ", domains[idx]);
off = lseek(resolv_conf_fd, 0, SEEK_CUR);
if (off == -1) {
2010-11-12 16:35:46 +05:30
log_line("write_resolve_conf: lseek returned error: %s\n",
strerror(errno));
return;
2010-11-12 14:32:18 +05:30
}
retry:
r = ftruncate(resolv_conf_fd, off);
if (r == -1) {
2010-11-12 16:35:46 +05:30
if (errno == EINTR)
goto retry;
log_line("write_resolve_conf: ftruncate returned error: %s\n",
strerror(errno));
return;
}
r = fsync(resolv_conf_fd);
if (r == -1) {
2010-11-12 16:35:46 +05:30
log_line("write_resolve_conf: fsync returned error: %s\n",
strerror(errno));
return;
}
2010-11-12 14:32:18 +05:30
}
/* Decomposes a ' '-delimited flat character array onto a strlist, then
* calls the given function to perform work on the generated strlist. */
static void parse_list(int idx, char *str, strlist_t **toplist,
void (*fn)(int))
{
char *p, n[256];
unsigned int i;
strlist_t *newn = 0;
if (!str || !toplist || !fn)
2010-11-12 16:35:46 +05:30
return;
2010-11-12 14:32:18 +05:30
p = str;
while (p != '\0') {
2010-11-12 16:35:46 +05:30
memset(n, '\0', sizeof n);
for (i = 0; i < sizeof n - 1 && *p != '\0' && *p != ' '; ++p, ++i)
n[i] = *p;
if (*p == ' ')
++p;
add_to_strlist(&newn, n);
2010-11-12 14:32:18 +05:30
}
if (newn) {
2010-11-12 16:35:46 +05:30
free_strlist(*toplist);
*toplist = newn;
2010-11-12 14:32:18 +05:30
} else
2010-11-12 16:35:46 +05:30
return;
2010-11-12 14:32:18 +05:30
(*fn)(idx);
}
/* XXX: addme */
static void perform_timezone(int idx, char *str)
{}
/* Add a dns server to the /etc/resolv.conf -- we already have a fd. */
static void perform_dns(int idx, char *str)
{
if (!str || resolv_conf_fd == -1)
2010-11-12 16:35:46 +05:30
return;
2010-11-12 14:32:18 +05:30
parse_list(idx, str, &(namesvrs[idx]), &write_resolve_conf);
}
/* Updates for print daemons are too non-standard to be useful. */
static void perform_lprsvr(int idx, char *str)
{}
/* Sets machine hostname. */
static void perform_hostname(int idx, char *str)
{
if (!allow_hostname || !str)
2010-11-12 16:35:46 +05:30
return;
2010-11-12 14:32:18 +05:30
if (sethostname(str, strlen(str) + 1) == -1)
2010-11-12 16:35:46 +05:30
log_line("sethostname returned %s\n", strerror(errno));
2010-11-12 14:32:18 +05:30
}
/* update "search" in /etc/resolv.conf */
static void perform_domain(int idx, char *str)
{
if (!str || resolv_conf_fd == -1)
2010-11-12 16:35:46 +05:30
return;
2010-11-12 14:32:18 +05:30
parse_list(idx, str, &(domains[idx]), &write_resolve_conf);
}
/* I don't think this can be done without a netfilter extension
* that isn't in the mainline kernels. */
static void perform_ipttl(int idx, char *str)
{}
/* XXX: addme */
static void perform_ntpsrv(int idx, char *str)
{}
/* Maybe Samba cares about this feature? I don't know. */
static void perform_wins(int idx, char *str)
{}
/* Wipes all state associated with a given connection. */
2010-11-12 16:35:46 +05:30
static void new_sk(int idx, int val)
2010-11-12 14:32:18 +05:30
{
sks[idx] = val;
memset(ibuf[idx], '\0', sizeof(ibuf[idx]));
free_strlist(head[idx]);
free_strlist(namesvrs[idx]);
free_strlist(domains[idx]);
head[idx] = NULL;
curl[idx] = NULL;
last[idx] = NULL;
namesvrs[idx] = NULL;
domains[idx] = NULL;
idle_time[idx] = time(NULL);
state[idx] = STATE_NOTHING;
clear_if_data(idx);
}
/* Conditionally accepts a new connection and initializes data structures. */
static void add_sk(int sk)
{
int i;
if (authorized_peer(sk, peer_pid, peer_uid, peer_gid)) {
2010-12-01 22:52:08 +05:30
for (i = 0; i < SOCK_QUEUE; i++)
2010-11-12 16:35:46 +05:30
if (sks[i] == -1) {
new_sk(i, sk);
2010-12-01 22:52:08 +05:30
epoll_add(sk);
2010-11-12 16:35:46 +05:30
return;
}
2010-11-12 14:32:18 +05:30
}
close(sk);
}
/* Closes idle connections. */
static void close_idle_sk(void)
{
int i;
for (i=0; i<SOCK_QUEUE; i++) {
2010-11-12 16:35:46 +05:30
if (sks[i] == -1)
continue;
if (time(NULL) - idle_time[i] > CONN_TIMEOUT) {
2010-12-01 22:52:08 +05:30
epoll_del(sks[i]);
2010-11-12 16:35:46 +05:30
close(sks[i]);
new_sk(i, -1);
}
2010-11-12 14:32:18 +05:30
}
}
/* Decomposes a ':'-delimited flat character array onto a strlist. */
static int stream_onto_list(int i)
{
int e, s;
for (e = 0, s = 0; ibuf[i][e] != '\0'; e++) {
2010-11-12 16:35:46 +05:30
if (ibuf[i][e] == ':') {
/* Zero-length command: skip. */
if (s == e) {
s = e + 1;
continue;
}
curl[i] = xmalloc(sizeof(strlist_t));
if (head[i] == NULL) {
head[i] = curl[i];
last[i] = NULL;
}
curl[i]->next = NULL;
if (last[i] != NULL)
last[i]->next = curl[i];
curl[i]->str = xmalloc(e - s + 1);
strlcpy(curl[i]->str, ibuf[i] + s, e - s + 1);
2010-11-12 16:35:46 +05:30
last[i] = curl[i];
s = e + 1;
}
2010-11-12 14:32:18 +05:30
}
return s;
}
/* State machine that runs over the command and argument list,
* executing commands. */
static void execute_list(int i)
{
char *p;
for (;;) {
2010-11-12 16:35:46 +05:30
if (!curl[i])
break;
die_nulstr(curl[i]);
p = curl[i]->str;
if (gflags_verbose)
log_line("execute_list - p = '%s'", p);
2010-11-12 16:35:46 +05:30
switch (state[i]) {
case STATE_NOTHING:
if (strncmp(p, CMD_INTERFACE, sizeof(CMD_INTERFACE)) == 0)
state[i] = STATE_INTERFACE;
if (strncmp(p, CMD_IP, sizeof(CMD_IP)) == 0)
state[i] = STATE_IP;
if (strncmp(p, CMD_SUBNET, sizeof(CMD_SUBNET)) == 0)
state[i] = STATE_SUBNET;
if (strncmp(p, CMD_TIMEZONE, sizeof(CMD_TIMEZONE)) == 0)
state[i] = STATE_TIMEZONE;
if (strncmp(p, CMD_ROUTER, sizeof(CMD_ROUTER)) == 0)
state[i] = STATE_ROUTER;
if (strncmp(p, CMD_DNS, sizeof(CMD_DNS)) == 0)
state[i] = STATE_DNS;
if (strncmp(p, CMD_LPRSVR, sizeof(CMD_LPRSVR)) == 0)
state[i] = STATE_LPRSVR;
if (strncmp(p, CMD_HOSTNAME, sizeof(CMD_HOSTNAME)) == 0)
state[i] = STATE_HOSTNAME;
if (strncmp(p, CMD_DOMAIN, sizeof(CMD_DOMAIN)) == 0)
state[i] = STATE_DOMAIN;
if (strncmp(p, CMD_IPTTL, sizeof(CMD_IPTTL)) == 0)
state[i] = STATE_IPTTL;
if (strncmp(p, CMD_MTU, sizeof(CMD_MTU)) == 0)
state[i] = STATE_MTU;
if (strncmp(p, CMD_BROADCAST, sizeof(CMD_BROADCAST)) == 0)
state[i] = STATE_BROADCAST;
2011-07-25 12:41:47 +05:30
if (strncmp(p, CMD_NTPSVR, sizeof(CMD_NTPSVR)) == 0)
state[i] = STATE_NTPSVR;
2010-11-12 16:35:46 +05:30
if (strncmp(p, CMD_WINS, sizeof(CMD_WINS)) == 0)
state[i] = STATE_WINS;
free_stritem(&(curl[i]));
break;
case STATE_INTERFACE:
perform_interface(i, p);
free_stritem(&(curl[i]));
state[i] = STATE_NOTHING;
break;
case STATE_IP:
perform_ip(i, p);
free_stritem(&(curl[i]));
state[i] = STATE_NOTHING;
break;
case STATE_SUBNET:
perform_subnet(i, p);
free_stritem(&(curl[i]));
state[i] = STATE_NOTHING;
break;
case STATE_TIMEZONE:
perform_timezone(i, p);
free_stritem(&(curl[i]));
state[i] = STATE_NOTHING;
break;
case STATE_ROUTER:
perform_router(i, p);
free_stritem(&(curl[i]));
state[i] = STATE_NOTHING;
break;
case STATE_DNS:
perform_dns(i, p);
free_stritem(&(curl[i]));
state[i] = STATE_NOTHING;
break;
case STATE_LPRSVR:
perform_lprsvr(i, p);
free_stritem(&(curl[i]));
state[i] = STATE_NOTHING;
break;
case STATE_HOSTNAME:
perform_hostname(i, p);
free_stritem(&(curl[i]));
state[i] = STATE_NOTHING;
break;
case STATE_DOMAIN:
perform_domain(i, p);
free_stritem(&(curl[i]));
state[i] = STATE_NOTHING;
break;
case STATE_IPTTL:
perform_ipttl(i, p);
free_stritem(&(curl[i]));
state[i] = STATE_NOTHING;
break;
case STATE_MTU:
perform_mtu(i, p);
free_stritem(&(curl[i]));
state[i] = STATE_NOTHING;
break;
case STATE_BROADCAST:
perform_broadcast(i, p);
free_stritem(&(curl[i]));
state[i] = STATE_NOTHING;
break;
2011-07-25 12:41:47 +05:30
case STATE_NTPSVR:
2010-11-12 16:35:46 +05:30
perform_ntpsrv(i, p);
free_stritem(&(curl[i]));
state[i] = STATE_NOTHING;
break;
case STATE_WINS:
perform_wins(i, p);
free_stritem(&(curl[i]));
state[i] = STATE_NOTHING;
break;
default:
log_line("warning: invalid state in dispatch_work\n");
break;
}
2010-11-12 14:32:18 +05:30
}
head[i] = curl[i];
}
/* Opens a non-blocking listening socket with the appropriate properties. */
static int get_listen(void)
{
int lsock, ret;
struct sockaddr_un lsock_addr =
{
2010-11-12 16:35:46 +05:30
.sun_family = AF_UNIX,
.sun_path = "/var/state/ifchange"
2010-11-12 14:32:18 +05:30
};
lsock = socket(AF_UNIX, SOCK_STREAM, 0);
2010-11-12 14:32:18 +05:30
if (lsock == -1)
2010-11-12 16:35:46 +05:30
suicide("dispatch_work - failed to create socket");
2010-11-12 14:32:18 +05:30
fcntl(lsock, F_SETFL, O_NONBLOCK);
(void) unlink("/var/state/ifchange");
2010-11-12 14:32:18 +05:30
ret = bind(lsock, (struct sockaddr *) &lsock_addr, sizeof(lsock_addr));
if (ret)
2010-11-12 16:35:46 +05:30
suicide("dispatch_work - failed to bind socket");
ret = chmod("/var/state/ifchange", S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP);
2010-11-12 14:32:18 +05:30
if (ret)
2010-11-12 16:35:46 +05:30
suicide("dispatch_work - failed to chmod socket");
2010-11-12 14:32:18 +05:30
ret = listen(lsock, SOCK_QUEUE);
if (ret)
2010-11-12 16:35:46 +05:30
suicide("dispatch_work - failed to listen on socket");
2010-11-12 14:32:18 +05:30
return lsock;
}
static void accept_conns(int *lsock)
{
int ret;
struct sockaddr_un sock_addr;
socklen_t sock_len = sizeof(sock_addr);
for(;;)
{
2010-11-12 16:35:46 +05:30
ret = accept(*lsock, (struct sockaddr *) &sock_addr, &sock_len);
if (ret != -1) {
add_sk(ret);
return;
}
switch (errno) {
case EAGAIN:
2010-11-12 14:32:18 +05:30
#ifdef LINUX
2010-11-12 16:35:46 +05:30
case ENETDOWN:
case EPROTO:
case ENOPROTOOPT:
case EHOSTDOWN:
case ENONET:
case EHOSTUNREACH:
case EOPNOTSUPP:
case ENETUNREACH:
2010-11-12 14:32:18 +05:30
#endif
2010-11-12 16:35:46 +05:30
return;
case EINTR:
continue;
case EBADF:
case ENOTSOCK:
case EINVAL:
log_line("warning: accept returned %s!\n", strerror(errno));
2010-12-01 22:52:08 +05:30
epoll_del(*lsock);
2010-11-12 16:35:46 +05:30
close(*lsock);
2010-12-01 22:52:08 +05:30
2010-11-12 16:35:46 +05:30
*lsock = get_listen();
2010-12-01 22:52:08 +05:30
epoll_add(*lsock);
2010-11-12 16:35:46 +05:30
return;
case ECONNABORTED:
case EMFILE:
case ENFILE:
log_line("warning: accept returned %s!\n", strerror(errno));
return;
default:
log_line("warning: accept returned a mysterious error: %s\n",
strerror(errno));
return;
}
2010-11-12 14:32:18 +05:30
}
}
2010-12-01 22:52:08 +05:30
static void setup_signals()
2010-11-12 14:32:18 +05:30
{
2010-12-01 22:52:08 +05:30
sigset_t mask;
sigemptyset(&mask);
sigaddset(&mask, SIGPIPE);
sigaddset(&mask, SIGUSR1);
sigaddset(&mask, SIGUSR2);
sigaddset(&mask, SIGTSTP);
sigaddset(&mask, SIGTTIN);
sigaddset(&mask, SIGCHLD);
sigaddset(&mask, SIGHUP);
sigaddset(&mask, SIGINT);
sigaddset(&mask, SIGTERM);
if (sigprocmask(SIG_BLOCK, &mask, NULL) < 0)
suicide("sigprocmask failed");
signalFd = signalfd(-1, &mask, SFD_NONBLOCK);
if (signalFd < 0)
suicide("signalfd failed");
}
2010-11-12 14:32:18 +05:30
2010-12-01 22:52:08 +05:30
static void signal_dispatch()
{
int t, off = 0;
struct signalfd_siginfo si;
again:
t = read(signalFd, (char *)&si + off, sizeof si - off);
if (t < sizeof si - off) {
if (t < 0) {
if (t == EAGAIN || t == EWOULDBLOCK || t == EINTR)
goto again;
else
suicide("signalfd read error");
2010-11-12 16:35:46 +05:30
}
2010-12-01 22:52:08 +05:30
off += t;
}
switch (si.ssi_signo) {
case SIGINT:
case SIGTERM:
exit(EXIT_SUCCESS);
default:
break;
}
}
2010-11-12 16:35:46 +05:30
2010-12-01 22:52:08 +05:30
static void process_client_fd(int fd)
{
char buf[MAX_BUF];
int r, index, sqidx = -1;
for (int j = 0; j < SOCK_QUEUE; ++j) {
if (sks[j] == fd) {
sqidx = j;
break;
2010-11-12 16:35:46 +05:30
}
2010-12-01 22:52:08 +05:30
}
if (sqidx == -1)
suicide("epoll returned pending read for untracked fd");
2010-11-12 16:35:46 +05:30
2010-12-01 22:52:08 +05:30
idle_time[sqidx] = time(NULL);
memset(buf, '\0', sizeof buf);
2010-11-12 16:35:46 +05:30
2010-12-01 22:52:08 +05:30
r = safe_read(sks[sqidx], buf, sizeof buf / 2 - 1);
if (r == 0)
2010-12-01 22:52:08 +05:30
goto fail;
else if (r < 0) {
if (errno == EAGAIN || errno == EWOULDBLOCK)
return;
log_line("error reading from client fd: %s", strerror(errno));
2010-12-01 22:52:08 +05:30
}
2010-11-12 16:35:46 +05:30
2010-12-01 22:52:08 +05:30
/* Discard everything and close connection if we risk overflow.
* This approach is maximally conservative... worst case is that
* some client requests will get dropped. */
index = strlen(ibuf[sqidx]);
if (index + strlen(buf) > sizeof buf - 2)
goto fail;
2010-11-12 16:35:46 +05:30
2010-12-01 22:52:08 +05:30
/* Append new stream input avoiding overflow. */
strlcpy(ibuf[sqidx] + index, buf, sizeof ibuf[sqidx] - index);
2010-11-12 14:32:18 +05:30
2010-12-01 22:52:08 +05:30
/* Decompose ibuf contents onto strlist. */
index = stream_onto_list(sqidx);
2010-11-12 16:35:46 +05:30
2010-12-01 22:52:08 +05:30
/* Remove everything that we've parsed into the list. */
strlcpy(buf, ibuf[sqidx] + index, sizeof buf);
strlcpy(ibuf[sqidx], buf, sizeof ibuf[sqidx]);
2010-11-12 16:35:46 +05:30
2010-12-01 22:52:08 +05:30
/* Now we have a strlist of commands and arguments.
* Decompose and execute it. */
if (!head[sqidx])
return;
curl[sqidx] = head[sqidx];
execute_list(sqidx);
return;
fail:
epoll_del(sks[sqidx]);
close(sks[sqidx]);
new_sk(sqidx, -1);
}
2010-11-12 16:35:46 +05:30
2010-12-01 22:52:08 +05:30
/* Core function that handles connections, gathers input, and calls
* the state machine to do actual work. */
static void dispatch_work(void)
{
int lsock;
2010-11-12 16:35:46 +05:30
2010-12-01 22:52:08 +05:30
/* Initialize all structures to blank state. */
for (int i = 0; i < SOCK_QUEUE; i++)
sks[i] = -1;
initialize_if_data();
2010-11-12 16:35:46 +05:30
2010-12-01 22:52:08 +05:30
lsock = get_listen();
2010-11-12 16:35:46 +05:30
2010-12-01 22:52:08 +05:30
epollfd = epoll_create1(0);
if (epollfd == -1)
suicide("epoll_create1 failed");
epoll_add(lsock);
epoll_add(signalFd);
2010-11-12 16:35:46 +05:30
2010-12-01 22:52:08 +05:30
for (;;) {
int r = epoll_wait(epollfd, events, SOCK_QUEUE + 2, -1);
if (r == -1) {
if (errno == EINTR)
2010-11-12 16:35:46 +05:30
continue;
2010-12-01 22:52:08 +05:30
else
suicide("epoll_wait failed");
}
for (int i = 0; i < r; ++i) {
int fd = events[i].data.fd;
if (fd == lsock)
accept_conns(&lsock);
else if (fd == signalFd)
signal_dispatch();
else
process_client_fd(fd);
2010-11-12 16:35:46 +05:30
}
close_idle_sk();
2010-11-12 14:32:18 +05:30
}
}
int main(int argc, char** argv) {
2010-11-12 16:12:07 +05:30
int c, t;
uid_t uid = 0;
gid_t gid = 0;
2010-11-12 14:32:18 +05:30
char pidfile[MAX_PATH_LENGTH] = PID_FILE_DEFAULT;
char chrootd[MAX_PATH_LENGTH] = "";
char resolv_conf_d[MAX_PATH_LENGTH] = "";
char *p;
struct passwd *pws;
struct group *grp;
while (1) {
2010-11-12 16:35:46 +05:30
int option_index = 0;
static struct option long_options[] = {
{"detach", 0, 0, 'd'},
{"nodetach", 0, 0, 'n'},
{"pidfile", 1, 0, 'p'},
{"quiet", 0, 0, 'q'},
{"chroot", 1, 0, 'c'},
{"resolve", 1, 0, 'r'},
{"hostname", 0, 0, 'o'},
{"user", 1, 0, 'u'},
{"group", 1, 0, 'g'},
{"cuser", 1, 0, 'U'},
{"cgroup", 1, 0, 'G'},
{"cpid", 1, 0, 'P'},
{"interface", 1, 0, 'i'},
{"help", 0, 0, 'h'},
{"version", 0, 0, 'v'},
{"verbose", 0, 0, 'V'},
2010-11-12 16:35:46 +05:30
{0, 0, 0, 0}
};
c = getopt_long(argc, argv, "dnp:qc:r:ou:g:U:G:P:i:hvV", long_options,
2010-11-12 16:35:46 +05:30
&option_index);
if (c == -1)
break;
switch (c) {
case 'h':
printf(
2010-11-12 14:32:18 +05:30
"ifchd %s, if change daemon. Licensed under GNU GPL.\n", IFCHD_VERSION);
2010-11-12 16:35:46 +05:30
printf(
"Copyright (C) 2004-2011 Nicholas J. Kain\n"
2010-11-12 14:32:18 +05:30
"Usage: ifchd [OPTIONS]\n"
" -d, --detach detach from TTY and daemonize\n"
" -n, --nodetach stay attached to TTY\n"
" -q, --quiet don't print to std(out|err) or log\n"
" -c, --chroot path where ifchd should chroot\n"
" -r, --resolve path to resolv.conf or equiv\n"
" -o, --hostname allow dhcp to set machine hostname\n"
" -p, --pidfile pidfile path\n");
2010-11-12 16:35:46 +05:30
printf(
2010-11-12 14:32:18 +05:30
" -u, --user user name that ifchd should run as\n"
" -g, --group group name that ifchd should run as\n"
" -U, --cuser user name of clients\n"
" -G, --cgroup group name of clients\n"
" -P, --cpid process id of client\n"
" -i, --interface ifchd clients may modify this interface\n"
" -V, --verbose log detailed messages\n"
2010-11-12 14:32:18 +05:30
" -h, --help print this help and exit\n"
" -v, --version print version information and exit\n");
2010-11-12 16:35:46 +05:30
exit(EXIT_FAILURE);
break;
2010-11-12 14:32:18 +05:30
2010-11-12 16:35:46 +05:30
case 'v':
printf(
2010-11-12 14:32:18 +05:30
"ifchd %s, if change daemon. Licensed under GNU GPL.\n", IFCHD_VERSION);
2010-11-12 16:35:46 +05:30
printf(
"Copyright (C) 2004-2011 Nicholas J. Kain\n"
2010-11-12 14:32:18 +05:30
"This is free software; see the source for copying conditions. There is NO\n"
"WARRANTY; not even for MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.\n");
2010-11-12 16:35:46 +05:30
exit(EXIT_FAILURE);
break;
case 'd':
gflags_detach = 1;
break;
case 'n':
gflags_detach = 0;
break;
case 'q':
gflags_quiet = 1;
break;
case 'c':
strlcpy(chrootd, optarg, MAX_PATH_LENGTH);
break;
case 'p':
strlcpy(pidfile, optarg, MAX_PATH_LENGTH);
break;
case 'r':
strlcpy(resolv_conf_d, optarg, MAX_PATH_LENGTH);
break;
case 'o':
allow_hostname = 1;
break;
case 'u':
t = (unsigned int) strtol(optarg, &p, 10);
if (*p != '\0') {
pws = getpwnam(optarg);
if (pws) {
uid = (int)pws->pw_uid;
if (!gid)
gid = (int)pws->pw_gid;
} else suicide("FATAL - Invalid uid specified.");
} else
uid = t;
break;
case 'g':
t = (unsigned int) strtol(optarg, &p, 10);
if (*p != '\0') {
grp = getgrnam(optarg);
if (grp) {
gid = (int)grp->gr_gid;
} else
suicide("FATAL - Invalid gid specified.");
} else
gid = t;
break;
case 'U':
t = (unsigned int) strtol(optarg, &p, 10);
if (*p != '\0') {
pws = getpwnam(optarg);
if (pws) {
peer_uid = (int)pws->pw_uid;
if (!peer_gid)
peer_gid = (int)pws->pw_gid;
} else
suicide("FATAL - Invalid uid specified.");
} else
peer_uid = t;
break;
case 'G':
t = (unsigned int) strtol(optarg, &p, 10);
if (*p != '\0') {
grp = getgrnam(optarg);
if (grp) {
peer_gid = (int)grp->gr_gid;
} else
suicide("FATAL - Invalid gid specified.");
} else
peer_gid = t;
break;
case 'P':
t = (unsigned int) strtol(optarg, &p, 10);
if (*p == '\0')
peer_pid = t;
break;
case 'i':
add_permitted_if(optarg);
break;
case 'V':
gflags_verbose = 1;
break;
2010-11-12 16:35:46 +05:30
}
2010-11-12 14:32:18 +05:30
}
if (getuid())
2010-11-12 16:35:46 +05:30
suicide("FATAL - I need root for CAP_NET_ADMIN and chroot!");
2010-11-12 14:32:18 +05:30
if (gflags_detach)
2010-11-12 16:35:46 +05:30
if (daemon(0,0)) {
log_line("FATAL - detaching fork failed\n");
exit(EXIT_FAILURE);
}
2010-11-12 14:32:18 +05:30
if (file_exists(pidfile, "w") == -1) {
2010-11-12 16:35:46 +05:30
log_line("FATAL - cannot open pidfile for write!");
exit(EXIT_FAILURE);
2010-11-12 14:32:18 +05:30
}
write_pid(pidfile);
umask(077);
2010-12-01 22:52:08 +05:30
setup_signals();
2010-11-12 14:32:18 +05:30
/* If we are requested to update resolv.conf, preopen the fd before
* we drop root privileges, making sure that if we create
* resolv.conf, it will be world-readable.
*/
if (strncmp(resolv_conf_d, "", MAX_PATH_LENGTH)) {
2010-11-12 16:35:46 +05:30
umask(022);
resolv_conf_fd = open(resolv_conf_d, O_RDWR | O_CREAT, 644);
umask(077);
if (resolv_conf_fd == -1) {
suicide("FATAL - unable to open resolv.conf");
}
2010-11-12 14:32:18 +05:30
}
if (!strncmp(chrootd, "", MAX_PATH_LENGTH))
2010-11-12 16:35:46 +05:30
suicide("FATAL - No chroot path specified. Refusing to run.");
2010-11-12 14:32:18 +05:30
/* Note that failure cases are handled by called fns. */
imprison(chrootd);
2010-11-12 16:12:07 +05:30
set_cap(uid, gid, "cap_net_admin=ep");
drop_root(uid, gid);
2010-11-12 14:32:18 +05:30
/* Cover our tracks... */
memset(chrootd, '\0', sizeof(chrootd));
memset(resolv_conf_d, '\0', sizeof(resolv_conf_d));
memset(pidfile, '\0', sizeof(pidfile));
dispatch_work();
/* Explicitly freed so memory debugger output has less static. */
for (c=0; c<SOCK_QUEUE; ++c) {
2010-11-12 16:35:46 +05:30
free_strlist(head[c]);
free_strlist(namesvrs[c]);
free_strlist(domains[c]);
2010-11-12 14:32:18 +05:30
}
exit(EXIT_SUCCESS);
}