int zcip_main(int argc UNUSED_PARAM, char **argv)
{
int state;
char *r_opt;
const char *l_opt = "169.254.0.0";
unsigned opts;
// ugly trick, but I want these zeroed in one go
struct {
const struct in_addr null_ip;
const struct ether_addr null_addr;
struct in_addr ip;
struct ifreq ifr;
int timeout_ms; /* must be signed */
unsigned conflicts;
unsigned nprobes;
unsigned nclaims;
int ready;
int verbose;
} L;
#define null_ip (L.null_ip )
#define null_addr (L.null_addr )
#define ip (L.ip )
#define ifr (L.ifr )
#define timeout_ms (L.timeout_ms)
#define conflicts (L.conflicts )
#define nprobes (L.nprobes )
#define nclaims (L.nclaims )
#define ready (L.ready )
#define verbose (L.verbose )
memset(&L, 0, sizeof(L));
INIT_G();
#define FOREGROUND (opts & 1)
#define QUIT (opts & 2)
// parse commandline: prog [options] ifname script
// exactly 2 args; -v accumulates and implies -f
opt_complementary = "=2:vv:vf";
opts = getopt32(argv, "fqr:l:v", &r_opt, &l_opt, &verbose);
#if !BB_MMU
// on NOMMU reexec early (or else we will rerun things twice)
if (!FOREGROUND)
bb_daemonize_or_rexec(0 /*was: DAEMON_CHDIR_ROOT*/, argv);
#endif
// open an ARP socket
// (need to do it before openlog to prevent openlog from taking
// fd 3 (sock_fd==3))
xmove_fd(xsocket(AF_PACKET, SOCK_PACKET, htons(ETH_P_ARP)), sock_fd);
if (!FOREGROUND) {
// do it before all bb_xx_msg calls
openlog(applet_name, 0, LOG_DAEMON);
logmode |= LOGMODE_SYSLOG;
}
bb_logenv_override();
{ // -l n.n.n.n
struct in_addr net;
if (inet_aton(l_opt, &net) == 0
|| (net.s_addr & htonl(IN_CLASSB_NET)) != net.s_addr
) {
bb_error_msg_and_die("invalid network address");
}
G.localnet_ip = ntohl(net.s_addr);
}
if (opts & 4) { // -r n.n.n.n
if (inet_aton(r_opt, &ip) == 0
|| (ntohl(ip.s_addr) & IN_CLASSB_NET) != G.localnet_ip
) {
bb_error_msg_and_die("invalid link address");
}
}
argv += optind - 1;
/* Now: argv[0]:junk argv[1]:intf argv[2]:script argv[3]:NULL */
/* We need to make space for script argument: */
argv[0] = argv[1];
argv[1] = argv[2];
/* Now: argv[0]:intf argv[1]:script argv[2]:junk argv[3]:NULL */
#define argv_intf (argv[0])
xsetenv("interface", argv_intf);
// initialize the interface (modprobe, ifup, etc)
if (run(argv, "init", NULL))
return EXIT_FAILURE;
// initialize saddr
// saddr is: { u16 sa_family; u8 sa_data[14]; }
//memset(&saddr, 0, sizeof(saddr));
//TODO: are we leaving sa_family == 0 (AF_UNSPEC)?!
safe_strncpy(saddr.sa_data, argv_intf, sizeof(saddr.sa_data));
// bind to the interface's ARP socket
xbind(sock_fd, &saddr, sizeof(saddr));
// get the interface's ethernet address
//memset(&ifr, 0, sizeof(ifr));
strncpy_IFNAMSIZ(ifr.ifr_name, argv_intf);
xioctl(sock_fd, SIOCGIFHWADDR, &ifr);
memcpy(ð_addr, &ifr.ifr_hwaddr.sa_data, ETH_ALEN);
// start with some stable ip address, either a function of
// the hardware address or else the last address we used.
// we are taking low-order four bytes, as top-order ones
// aren't random enough.
// NOTE: the sequence of addresses we try changes only
// depending on when we detect conflicts.
{
uint32_t t;
move_from_unaligned32(t, ((char *)ð_addr + 2));
t += getpid();
srand(t);
}
if (ip.s_addr == 0)
ip.s_addr = pick_nip();
// FIXME cases to handle:
// - zcip already running!
// - link already has local address... just defend/update
// daemonize now; don't delay system startup
if (!FOREGROUND) {
#if BB_MMU
bb_daemonize(0 /*was: DAEMON_CHDIR_ROOT*/);
#endif
bb_info_msg("start, interface %s", argv_intf);
}
// run the dynamic address negotiation protocol,
// restarting after address conflicts:
// - start with some address we want to try
// - short random delay
// - arp probes to see if another host uses it
// - arp announcements that we're claiming it
// - use it
// - defend it, within limits
// exit if:
// - address is successfully obtained and -q was given:
// run "<script> config", then exit with exitcode 0
// - poll error (when does this happen?)
// - read error (when does this happen?)
// - sendto error (in arp()) (when does this happen?)
// - revents & POLLERR (link down). run "<script> deconfig" first
state = PROBE;
while (1) {
struct pollfd fds[1];
unsigned deadline_us;
struct arp_packet p;
int source_ip_conflict;
int target_ip_conflict;
fds[0].fd = sock_fd;
fds[0].events = POLLIN;
fds[0].revents = 0;
// poll, being ready to adjust current timeout
if (!timeout_ms) {
timeout_ms = random_delay_ms(PROBE_WAIT);
// FIXME setsockopt(sock_fd, SO_ATTACH_FILTER, ...) to
// make the kernel filter out all packets except
// ones we'd care about.
}
// set deadline_us to the point in time when we timeout
deadline_us = MONOTONIC_US() + timeout_ms * 1000;
VDBG("...wait %d %s nprobes=%u, nclaims=%u\n",
timeout_ms, argv_intf, nprobes, nclaims);
switch (safe_poll(fds, 1, timeout_ms)) {
default:
//bb_perror_msg("poll"); - done in safe_poll
return EXIT_FAILURE;
// timeout
case 0:
VDBG("state = %d\n", state);
switch (state) {
case PROBE:
// timeouts in the PROBE state mean no conflicting ARP packets
// have been received, so we can progress through the states
if (nprobes < PROBE_NUM) {
nprobes++;
VDBG("probe/%u %s@%s\n",
nprobes, argv_intf, inet_ntoa(ip));
timeout_ms = PROBE_MIN * 1000;
timeout_ms += random_delay_ms(PROBE_MAX - PROBE_MIN);
arp(/* ARPOP_REQUEST, */
/* ð_addr, */ null_ip,
&null_addr, ip);
}
else {
// Switch to announce state.
state = ANNOUNCE;
nclaims = 0;
VDBG("announce/%u %s@%s\n",
nclaims, argv_intf, inet_ntoa(ip));
timeout_ms = ANNOUNCE_INTERVAL * 1000;
arp(/* ARPOP_REQUEST, */
/* ð_addr, */ ip,
ð_addr, ip);
}
break;
case RATE_LIMIT_PROBE:
// timeouts in the RATE_LIMIT_PROBE state mean no conflicting ARP packets
// have been received, so we can move immediately to the announce state
state = ANNOUNCE;
nclaims = 0;
VDBG("announce/%u %s@%s\n",
nclaims, argv_intf, inet_ntoa(ip));
timeout_ms = ANNOUNCE_INTERVAL * 1000;
arp(/* ARPOP_REQUEST, */
/* ð_addr, */ ip,
ð_addr, ip);
break;
case ANNOUNCE:
// timeouts in the ANNOUNCE state mean no conflicting ARP packets
// have been received, so we can progress through the states
if (nclaims < ANNOUNCE_NUM) {
nclaims++;
VDBG("announce/%u %s@%s\n",
nclaims, argv_intf, inet_ntoa(ip));
timeout_ms = ANNOUNCE_INTERVAL * 1000;
arp(/* ARPOP_REQUEST, */
/* ð_addr, */ ip,
ð_addr, ip);
}
else {
// Switch to monitor state.
state = MONITOR;
// link is ok to use earlier
// FIXME update filters
run(argv, "config", &ip);
ready = 1;
conflicts = 0;
timeout_ms = -1; // Never timeout in the monitor state.
// NOTE: all other exit paths
// should deconfig ...
if (QUIT)
return EXIT_SUCCESS;
}
break;
case DEFEND:
// We won! No ARP replies, so just go back to monitor.
state = MONITOR;
timeout_ms = -1;
conflicts = 0;
break;
default:
// Invalid, should never happen. Restart the whole protocol.
state = PROBE;
ip.s_addr = pick_nip();
timeout_ms = 0;
nprobes = 0;
nclaims = 0;
break;
} // switch (state)
break; // case 0 (timeout)
// packets arriving, or link went down
case 1:
// We need to adjust the timeout in case we didn't receive
// a conflicting packet.
if (timeout_ms > 0) {
unsigned diff = deadline_us - MONOTONIC_US();
if ((int)(diff) < 0) {
// Current time is greater than the expected timeout time.
// Should never happen.
VDBG("missed an expected timeout\n");
timeout_ms = 0;
} else {
VDBG("adjusting timeout\n");
timeout_ms = (diff / 1000) | 1; /* never 0 */
}
}
if ((fds[0].revents & POLLIN) == 0) {
if (fds[0].revents & POLLERR) {
// FIXME: links routinely go down;
// this shouldn't necessarily exit.
bb_error_msg("iface %s is down", argv_intf);
if (ready) {
run(argv, "deconfig", &ip);
}
return EXIT_FAILURE;
}
continue;
}
// read ARP packet
if (safe_read(sock_fd, &p, sizeof(p)) < 0) {
bb_perror_msg_and_die(bb_msg_read_error);
}
if (p.eth.ether_type != htons(ETHERTYPE_ARP))
continue;
#ifdef DEBUG
{
struct ether_addr *sha = (struct ether_addr *) p.arp.arp_sha;
struct ether_addr *tha = (struct ether_addr *) p.arp.arp_tha;
struct in_addr *spa = (struct in_addr *) p.arp.arp_spa;
struct in_addr *tpa = (struct in_addr *) p.arp.arp_tpa;
VDBG("%s recv arp type=%d, op=%d,\n",
argv_intf, ntohs(p.eth.ether_type),
ntohs(p.arp.arp_op));
VDBG("\tsource=%s %s\n",
ether_ntoa(sha),
inet_ntoa(*spa));
VDBG("\ttarget=%s %s\n",
ether_ntoa(tha),
inet_ntoa(*tpa));
}
#endif
if (p.arp.arp_op != htons(ARPOP_REQUEST)
&& p.arp.arp_op != htons(ARPOP_REPLY)
) {
continue;
}
source_ip_conflict = 0;
target_ip_conflict = 0;
if (memcmp(&p.arp.arp_sha, ð_addr, ETH_ALEN) != 0) {
if (memcmp(p.arp.arp_spa, &ip.s_addr, sizeof(struct in_addr)) == 0) {
/* A probe or reply with source_ip == chosen ip */
source_ip_conflict = 1;
}
if (p.arp.arp_op == htons(ARPOP_REQUEST)
&& memcmp(p.arp.arp_spa, &null_ip, sizeof(struct in_addr)) == 0
&& memcmp(p.arp.arp_tpa, &ip.s_addr, sizeof(struct in_addr)) == 0
) {
/* A probe with source_ip == 0.0.0.0, target_ip == chosen ip:
* another host trying to claim this ip!
*/
target_ip_conflict = 1;
}
}
VDBG("state = %d, source ip conflict = %d, target ip conflict = %d\n",
state, source_ip_conflict, target_ip_conflict);
switch (state) {
case PROBE:
case ANNOUNCE:
// When probing or announcing, check for source IP conflicts
// and other hosts doing ARP probes (target IP conflicts).
if (source_ip_conflict || target_ip_conflict) {
conflicts++;
if (conflicts >= MAX_CONFLICTS) {
VDBG("%s ratelimit\n", argv_intf);
timeout_ms = RATE_LIMIT_INTERVAL * 1000;
state = RATE_LIMIT_PROBE;
}
// restart the whole protocol
ip.s_addr = pick_nip();
timeout_ms = 0;
nprobes = 0;
nclaims = 0;
}
break;
case MONITOR:
// If a conflict, we try to defend with a single ARP probe.
if (source_ip_conflict) {
VDBG("monitor conflict -- defending\n");
state = DEFEND;
timeout_ms = DEFEND_INTERVAL * 1000;
arp(/* ARPOP_REQUEST, */
/* ð_addr, */ ip,
ð_addr, ip);
}
break;
case DEFEND:
// Well, we tried. Start over (on conflict).
if (source_ip_conflict) {
state = PROBE;
VDBG("defend conflict -- starting over\n");
ready = 0;
run(argv, "deconfig", &ip);
// restart the whole protocol
ip.s_addr = pick_nip();
timeout_ms = 0;
nprobes = 0;
nclaims = 0;
}
break;
default:
// Invalid, should never happen. Restart the whole protocol.
VDBG("invalid state -- starting over\n");
state = PROBE;
ip.s_addr = pick_nip();
timeout_ms = 0;
nprobes = 0;
nclaims = 0;
break;
} // switch state
break; // case 1 (packets arriving)
} // switch poll
} // while (1)
#undef argv_intf
}
int zcip_main(int argc UNUSED_PARAM, char **argv)
{
char *r_opt;
const char *l_opt = "169.254.0.0";
int state;
int nsent;
unsigned opts;
// Ugly trick, but I want these zeroed in one go
struct {
const struct ether_addr null_ethaddr;
struct ifreq ifr;
uint32_t chosen_nip;
int conflicts;
int timeout_ms; // must be signed
int verbose;
} L;
#define null_ethaddr (L.null_ethaddr)
#define ifr (L.ifr )
#define chosen_nip (L.chosen_nip )
#define conflicts (L.conflicts )
#define timeout_ms (L.timeout_ms )
#define verbose (L.verbose )
memset(&L, 0, sizeof(L));
INIT_G();
#define FOREGROUND (opts & 1)
#define QUIT (opts & 2)
// Parse commandline: prog [options] ifname script
// exactly 2 args; -v accumulates and implies -f
opt_complementary = "=2:vv:vf";
opts = getopt32(argv, "fqr:l:v", &r_opt, &l_opt, &verbose);
#if !BB_MMU
// on NOMMU reexec early (or else we will rerun things twice)
if (!FOREGROUND)
bb_daemonize_or_rexec(0 /*was: DAEMON_CHDIR_ROOT*/, argv);
#endif
// Open an ARP socket
// (need to do it before openlog to prevent openlog from taking
// fd 3 (sock_fd==3))
xmove_fd(xsocket(AF_PACKET, SOCK_PACKET, htons(ETH_P_ARP)), sock_fd);
if (!FOREGROUND) {
// do it before all bb_xx_msg calls
openlog(applet_name, 0, LOG_DAEMON);
logmode |= LOGMODE_SYSLOG;
}
bb_logenv_override();
{ // -l n.n.n.n
struct in_addr net;
if (inet_aton(l_opt, &net) == 0
|| (net.s_addr & htonl(IN_CLASSB_NET)) != net.s_addr
) {
bb_error_msg_and_die("invalid network address");
}
G.localnet_ip = ntohl(net.s_addr);
}
if (opts & 4) { // -r n.n.n.n
struct in_addr ip;
if (inet_aton(r_opt, &ip) == 0
|| (ntohl(ip.s_addr) & IN_CLASSB_NET) != G.localnet_ip
) {
bb_error_msg_and_die("invalid link address");
}
chosen_nip = ip.s_addr;
}
argv += optind - 1;
/* Now: argv[0]:junk argv[1]:intf argv[2]:script argv[3]:NULL */
/* We need to make space for script argument: */
argv[0] = argv[1];
argv[1] = argv[2];
/* Now: argv[0]:intf argv[1]:script argv[2]:junk argv[3]:NULL */
#define argv_intf (argv[0])
xsetenv("interface", argv_intf);
// Initialize the interface (modprobe, ifup, etc)
if (run(argv, "init", 0))
return EXIT_FAILURE;
// Initialize G.iface_sockaddr
// G.iface_sockaddr is: { u16 sa_family; u8 sa_data[14]; }
//memset(&G.iface_sockaddr, 0, sizeof(G.iface_sockaddr));
//TODO: are we leaving sa_family == 0 (AF_UNSPEC)?!
safe_strncpy(G.iface_sockaddr.sa_data, argv_intf, sizeof(G.iface_sockaddr.sa_data));
// Bind to the interface's ARP socket
xbind(sock_fd, &G.iface_sockaddr, sizeof(G.iface_sockaddr));
// Get the interface's ethernet address
//memset(&ifr, 0, sizeof(ifr));
strncpy_IFNAMSIZ(ifr.ifr_name, argv_intf);
xioctl(sock_fd, SIOCGIFHWADDR, &ifr);
memcpy(&G.our_ethaddr, &ifr.ifr_hwaddr.sa_data, ETH_ALEN);
// Start with some stable ip address, either a function of
// the hardware address or else the last address we used.
// we are taking low-order four bytes, as top-order ones
// aren't random enough.
// NOTE: the sequence of addresses we try changes only
// depending on when we detect conflicts.
{
uint32_t t;
move_from_unaligned32(t, ((char *)&G.our_ethaddr + 2));
t += getpid();
srand(t);
}
// FIXME cases to handle:
// - zcip already running!
// - link already has local address... just defend/update
// Daemonize now; don't delay system startup
if (!FOREGROUND) {
#if BB_MMU
bb_daemonize(0 /*was: DAEMON_CHDIR_ROOT*/);
#endif
bb_info_msg("start, interface %s", argv_intf);
}
// Run the dynamic address negotiation protocol,
// restarting after address conflicts:
// - start with some address we want to try
// - short random delay
// - arp probes to see if another host uses it
// 00:04:e2:64:23:c2 > ff:ff:ff:ff:ff:ff arp who-has 169.254.194.171 tell 0.0.0.0
// - arp announcements that we're claiming it
// 00:04:e2:64:23:c2 > ff:ff:ff:ff:ff:ff arp who-has 169.254.194.171 (00:04:e2:64:23:c2) tell 169.254.194.171
// - use it
// - defend it, within limits
// exit if:
// - address is successfully obtained and -q was given:
// run "<script> config", then exit with exitcode 0
// - poll error (when does this happen?)
// - read error (when does this happen?)
// - sendto error (in send_arp_request()) (when does this happen?)
// - revents & POLLERR (link down). run "<script> deconfig" first
if (chosen_nip == 0) {
new_nip_and_PROBE:
chosen_nip = pick_nip();
}
nsent = 0;
state = PROBE;
while (1) {
struct pollfd fds[1];
unsigned deadline_us;
struct arp_packet p;
int ip_conflict;
int n;
fds[0].fd = sock_fd;
fds[0].events = POLLIN;
fds[0].revents = 0;
// Poll, being ready to adjust current timeout
if (!timeout_ms) {
timeout_ms = random_delay_ms(PROBE_WAIT);
// FIXME setsockopt(sock_fd, SO_ATTACH_FILTER, ...) to
// make the kernel filter out all packets except
// ones we'd care about.
}
// Set deadline_us to the point in time when we timeout
deadline_us = MONOTONIC_US() + timeout_ms * 1000;
VDBG("...wait %d %s nsent=%u\n",
timeout_ms, argv_intf, nsent);
n = safe_poll(fds, 1, timeout_ms);
if (n < 0) {
//bb_perror_msg("poll"); - done in safe_poll
return EXIT_FAILURE;
}
if (n == 0) { // timed out?
VDBG("state:%d\n", state);
switch (state) {
case PROBE:
// No conflicting ARP packets were seen:
// we can progress through the states
if (nsent < PROBE_NUM) {
nsent++;
VDBG("probe/%u %s@%s\n",
nsent, argv_intf, nip_to_a(chosen_nip));
timeout_ms = PROBE_MIN * 1000;
timeout_ms += random_delay_ms(PROBE_MAX - PROBE_MIN);
send_arp_request(0, &null_ethaddr, chosen_nip);
continue;
}
// Switch to announce state
nsent = 0;
state = ANNOUNCE;
goto send_announce;
case ANNOUNCE:
// No conflicting ARP packets were seen:
// we can progress through the states
if (nsent < ANNOUNCE_NUM) {
send_announce:
nsent++;
VDBG("announce/%u %s@%s\n",
nsent, argv_intf, nip_to_a(chosen_nip));
timeout_ms = ANNOUNCE_INTERVAL * 1000;
send_arp_request(chosen_nip, &G.our_ethaddr, chosen_nip);
continue;
}
// Switch to monitor state
// FIXME update filters
run(argv, "config", chosen_nip);
// NOTE: all other exit paths should deconfig...
if (QUIT)
return EXIT_SUCCESS;
// fall through: switch to MONITOR
default:
// case DEFEND:
// case MONITOR: (shouldn't happen, MONITOR timeout is infinite)
// Defend period ended with no ARP replies - we won
timeout_ms = -1; // never timeout in monitor state
state = MONITOR;
continue;
}
}
// Packet arrived, or link went down.
// We need to adjust the timeout in case we didn't receive
// a conflicting packet.
if (timeout_ms > 0) {
unsigned diff = deadline_us - MONOTONIC_US();
if ((int)(diff) < 0) {
// Current time is greater than the expected timeout time.
diff = 0;
}
VDBG("adjusting timeout\n");
timeout_ms = (diff / 1000) | 1; // never 0
}
if ((fds[0].revents & POLLIN) == 0) {
if (fds[0].revents & POLLERR) {
// FIXME: links routinely go down;
// this shouldn't necessarily exit.
bb_error_msg("iface %s is down", argv_intf);
if (state >= MONITOR) {
// Only if we are in MONITOR or DEFEND
run(argv, "deconfig", chosen_nip);
}
return EXIT_FAILURE;
}
continue;
}
// Read ARP packet
if (safe_read(sock_fd, &p, sizeof(p)) < 0) {
bb_perror_msg_and_die(bb_msg_read_error);
}
if (p.eth.ether_type != htons(ETHERTYPE_ARP))
continue;
if (p.arp.arp_op != htons(ARPOP_REQUEST)
&& p.arp.arp_op != htons(ARPOP_REPLY)
) {
continue;
}
#ifdef DEBUG
{
struct ether_addr *sha = (struct ether_addr *) p.arp.arp_sha;
struct ether_addr *tha = (struct ether_addr *) p.arp.arp_tha;
struct in_addr *spa = (struct in_addr *) p.arp.arp_spa;
struct in_addr *tpa = (struct in_addr *) p.arp.arp_tpa;
VDBG("source=%s %s\n", ether_ntoa(sha), inet_ntoa(*spa));
VDBG("target=%s %s\n", ether_ntoa(tha), inet_ntoa(*tpa));
}
#endif
ip_conflict = 0;
if (memcmp(&p.arp.arp_sha, &G.our_ethaddr, ETH_ALEN) != 0) {
if (memcmp(p.arp.arp_spa, &chosen_nip, 4) == 0) {
// A probe or reply with source_ip == chosen ip
ip_conflict = 1;
}
if (p.arp.arp_op == htons(ARPOP_REQUEST)
&& memcmp(p.arp.arp_spa, &const_int_0, 4) == 0
&& memcmp(p.arp.arp_tpa, &chosen_nip, 4) == 0
) {
// A probe with source_ip == 0.0.0.0, target_ip == chosen ip:
// another host trying to claim this ip!
ip_conflict |= 2;
}
}
VDBG("state:%d ip_conflict:%d\n", state, ip_conflict);
if (!ip_conflict)
continue;
// Either src or target IP conflict exists
if (state <= ANNOUNCE) {
// PROBE or ANNOUNCE
conflicts++;
timeout_ms = PROBE_MIN * 1000
+ CONFLICT_MULTIPLIER * random_delay_ms(conflicts);
goto new_nip_and_PROBE;
}
// MONITOR or DEFEND: only src IP conflict is a problem
if (ip_conflict & 1) {
if (state == MONITOR) {
// Src IP conflict, defend with a single ARP probe
VDBG("monitor conflict - defending\n");
timeout_ms = DEFEND_INTERVAL * 1000;
state = DEFEND;
send_arp_request(chosen_nip, &G.our_ethaddr, chosen_nip);
continue;
}
// state == DEFEND
// Another src IP conflict, start over
VDBG("defend conflict - starting over\n");
run(argv, "deconfig", chosen_nip);
conflicts = 0;
timeout_ms = 0;
goto new_nip_and_PROBE;
}
// Note: if we only have a target IP conflict here (ip_conflict & 2),
// IOW: if we just saw this sort of ARP packet:
// aa:bb:cc:dd:ee:ff > xx:xx:xx:xx:xx:xx arp who-has <chosen_nip> tell 0.0.0.0
// we expect _kernel_ to respond to that, because <chosen_nip>
// is (expected to be) configured on this iface.
} // while (1)
#undef argv_intf
}