sulogin: use bb_error_msg instead of bb_info_msg; better message
[oweals/busybox.git] / networking / zcip.c
index 281f551d7e97e72028fdb9580cf56c5b61d49540..c930826194a3644fe0f383d994ee7c81be10f582 100644 (file)
@@ -6,7 +6,7 @@
  * Copyright (C) 2003 by Arthur van Hoff (avh@strangeberry.com)
  * Copyright (C) 2004 by David Brownell
  *
- * Licensed under the GPL v2 or later, see the file LICENSE in this tarball.
+ * Licensed under GPLv2 or later, see file LICENSE in this source tree.
  */
 
 /*
 // - avoid silent script failures, especially under load...
 // - link status monitoring (restart on link-up; stop on link-down)
 
-#include <syslog.h>
-#include <poll.h>
-#include <sys/wait.h>
+//usage:#define zcip_trivial_usage
+//usage:       "[OPTIONS] IFACE SCRIPT"
+//usage:#define zcip_full_usage "\n\n"
+//usage:       "Manage a ZeroConf IPv4 link-local address\n"
+//usage:     "\n       -f              Run in foreground"
+//usage:     "\n       -q              Quit after obtaining address"
+//usage:     "\n       -r 169.254.x.x  Request this address first"
+//usage:     "\n       -l x.x.0.0      Use this range instead of 169.254"
+//usage:     "\n       -v              Verbose"
+//usage:     "\n"
+//usage:     "\n$LOGGING=none          Suppress logging"
+//usage:     "\n$LOGGING=syslog        Log to syslog"
+//usage:     "\n"
+//usage:     "\nWith no -q, runs continuously monitoring for ARP conflicts,"
+//usage:     "\nexits only on I/O errors (link down etc)"
+
+#include "libbb.h"
 #include <netinet/ether.h>
-#include <net/ethernet.h>
 #include <net/if.h>
 #include <net/if_arp.h>
-#include <linux/if_packet.h>
 #include <linux/sockios.h>
 
-#include "libbb.h"
+#include <syslog.h>
 
 /* We don't need more than 32 bits of the counter */
 #define MONOTONIC_US() ((unsigned)monotonic_us())
 
 struct arp_packet {
-       struct ether_header hdr;
+       struct ether_header eth;
        struct ether_arp arp;
-} ATTRIBUTE_PACKED;
+} PACKED;
 
 enum {
-/* 169.254.0.0 */
-       LINKLOCAL_ADDR = 0xa9fe0000,
-
-/* protocol timeout parameters, specified in seconds */
+       /* 0-1 seconds before sending 1st probe */
        PROBE_WAIT = 1,
+       /* 1-2 seconds between probes */
        PROBE_MIN = 1,
        PROBE_MAX = 2,
-       PROBE_NUM = 3,
-       MAX_CONFLICTS = 10,
-       RATE_LIMIT_INTERVAL = 60,
-       ANNOUNCE_WAIT = 2,
-       ANNOUNCE_NUM = 2,
-       ANNOUNCE_INTERVAL = 2,
-       DEFEND_INTERVAL = 10
+       PROBE_NUM = 3,          /* total probes to send */
+       ANNOUNCE_INTERVAL = 2,  /* 2 seconds between announces */
+       ANNOUNCE_NUM = 3,       /* announces to send */
+       /* if probe/announce sees a conflict, multiply RANDOM(NUM_CONFLICT) by... */
+       CONFLICT_MULTIPLIER = 2,
+       /* if we monitor and see a conflict, how long is defend state? */
+       DEFEND_INTERVAL = 10,
 };
 
 /* States during the configuration process. */
 enum {
        PROBE = 0,
-       RATE_LIMIT_PROBE,
        ANNOUNCE,
        MONITOR,
        DEFEND
 };
 
-#define VDBG(fmt,args...) \
-       do { } while (0)
+#define VDBG(...) do { } while (0)
+
+
+enum {
+       sock_fd = 3
+};
+
+struct globals {
+       struct sockaddr iface_sockaddr;
+       struct ether_addr our_ethaddr;
+       uint32_t localnet_ip;
+} FIX_ALIASING;
+#define G (*(struct globals*)&bb_common_bufsiz1)
+#define INIT_G() do { } while (0)
+
 
 /**
  * Pick a random link local IP address on 169.254/16, except that
  * the first and last 256 addresses are reserved.
  */
-static void pick(struct in_addr *ip)
+static uint32_t pick_nip(void)
 {
        unsigned tmp;
 
        do {
                tmp = rand() & IN_CLASSB_HOST;
        } while (tmp > (IN_CLASSB_HOST - 0x0200));
-       ip->s_addr = htonl((LINKLOCAL_ADDR + 0x0100) + tmp);
+       return htonl((G.localnet_ip + 0x0100) + tmp);
+}
+
+static const char *nip_to_a(uint32_t nip)
+{
+       struct in_addr in;
+       in.s_addr = nip;
+       return inet_ntoa(in);
 }
 
 /**
  * Broadcast an ARP packet.
  */
-static void arp(int fd, struct sockaddr *saddr, int op,
-       const struct ether_addr *source_addr, struct in_addr source_ip,
-       const struct ether_addr *target_addr, struct in_addr target_ip)
+static void send_arp_request(
+       /* int op, - always ARPOP_REQUEST */
+       /* const struct ether_addr *source_eth, - always &G.our_ethaddr */
+                                       uint32_t source_nip,
+       const struct ether_addr *target_eth, uint32_t target_nip)
 {
+       enum { op = ARPOP_REQUEST };
+#define source_eth (&G.our_ethaddr)
+
        struct arp_packet p;
        memset(&p, 0, sizeof(p));
 
        // ether header
-       p.hdr.ether_type = htons(ETHERTYPE_ARP);
-       memcpy(p.hdr.ether_shost, source_addr, ETH_ALEN);
-       memset(p.hdr.ether_dhost, 0xff, ETH_ALEN);
+       p.eth.ether_type = htons(ETHERTYPE_ARP);
+       memcpy(p.eth.ether_shost, source_eth, ETH_ALEN);
+       memset(p.eth.ether_dhost, 0xff, ETH_ALEN);
 
        // arp request
        p.arp.arp_hrd = htons(ARPHRD_ETHER);
@@ -107,429 +141,387 @@ static void arp(int fd, struct sockaddr *saddr, int op,
        p.arp.arp_hln = ETH_ALEN;
        p.arp.arp_pln = 4;
        p.arp.arp_op = htons(op);
-       memcpy(&p.arp.arp_sha, source_addr, ETH_ALEN);
-       memcpy(&p.arp.arp_spa, &source_ip, sizeof(p.arp.arp_spa));
-       memcpy(&p.arp.arp_tha, target_addr, ETH_ALEN);
-       memcpy(&p.arp.arp_tpa, &target_ip, sizeof(p.arp.arp_tpa));
+       memcpy(&p.arp.arp_sha, source_eth, ETH_ALEN);
+       memcpy(&p.arp.arp_spa, &source_nip, 4);
+       memcpy(&p.arp.arp_tha, target_eth, ETH_ALEN);
+       memcpy(&p.arp.arp_tpa, &target_nip, 4);
 
        // send it
-       xsendto(fd, &p, sizeof(p), saddr, sizeof(*saddr));
-
-       // Currently all callers ignore errors, that's why returns are
-       // commented out...
-       //return 0;
+       // Even though sock_fd is already bound to G.iface_sockaddr, just send()
+       // won't work, because "socket is not connected"
+       // (and connect() won't fix that, "operation not supported").
+       // Thus we sendto() to G.iface_sockaddr. I wonder which sockaddr
+       // (from bind() or from sendto()?) kernel actually uses
+       // to determine iface to emit the packet from...
+       xsendto(sock_fd, &p, sizeof(p), &G.iface_sockaddr, sizeof(G.iface_sockaddr));
+#undef source_eth
 }
 
 /**
- * Run a script. argv[2] is already NULL.
+ * Run a script.
+ * argv[0]:intf argv[1]:script_name argv[2]:junk argv[3]:NULL
  */
-static int run(char *argv[3], const char *intf, struct in_addr *ip)
+static int run(char *argv[3], const char *param, uint32_t nip)
 {
        int status;
+       const char *addr = addr; /* for gcc */
+       const char *fmt = "%s %s %s" + 3;
 
-       VDBG("%s run %s %s\n", intf, argv[0], argv[1]);
+       argv[2] = (char*)param;
 
-       if (ip) {
-               char *addr = inet_ntoa(*ip);
-               setenv("ip", addr, 1);
-               bb_info_msg("%s %s %s", argv[1], intf, addr);
+       VDBG("%s run %s %s\n", argv[0], argv[1], argv[2]);
+
+       if (nip != 0) {
+               addr = nip_to_a(nip);
+               xsetenv("ip", addr);
+               fmt -= 3;
        }
+       bb_error_msg(fmt, argv[2], argv[0], addr);
 
-       status = wait4pid(spawn(argv));
+       status = spawn_and_wait(argv + 1);
        if (status < 0) {
-               bb_perror_msg("%s %s", argv[1], intf);
+               bb_perror_msg("%s %s %s" + 3, argv[2], argv[0]);
                return -errno;
        }
        if (status != 0)
-               bb_error_msg("script %s %s failed, exitcode=%d", argv[0], argv[1], status);
+               bb_error_msg("script %s %s failed, exitcode=%d", argv[1], argv[2], status & 0xff);
        return status;
 }
 
 /**
  * Return milliseconds of random delay, up to "secs" seconds.
  */
-static unsigned ALWAYS_INLINE ms_rdelay(unsigned secs)
+static ALWAYS_INLINE unsigned random_delay_ms(unsigned secs)
 {
-       return rand() % (secs * 1000);
+       return (unsigned)rand() % (secs * 1000);
 }
 
 /**
  * main program
  */
-int zcip_main(int argc, char **argv);
-int zcip_main(int argc, char **argv)
+int zcip_main(int argc, char **argv) MAIN_EXTERNALLY_VISIBLE;
+int zcip_main(int argc UNUSED_PARAM, char **argv)
 {
-       int state = PROBE;
-       struct ether_addr eth_addr;
-       const char *why;
-       int fd;
        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 */
+       // Ugly trick, but I want these zeroed in one go
        struct {
-               const struct in_addr null_ip;
-               const struct ether_addr null_addr;
-               struct sockaddr saddr;
-               struct in_addr ip;
+               const struct ether_addr null_ethaddr;
                struct ifreq ifr;
-               char *intf;
-               char *script_av[3];
-               int timeout_ms; /* must be signed */
-               unsigned conflicts;
-               unsigned nprobes;
-               unsigned nclaims;
-               int ready;
+               uint32_t chosen_nip;
+               int conflicts;
+               int timeout_ms; // must be signed
                int verbose;
        } L;
-#define null_ip    (L.null_ip   )
-#define null_addr  (L.null_addr )
-#define saddr      (L.saddr     )
-#define ip         (L.ip        )
-#define ifr        (L.ifr       )
-#define intf       (L.intf      )
-#define script_av  (L.script_av )
-#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   )
+#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));
-
-       srand(MONOTONIC_US());
+       INIT_G();
 
 #define FOREGROUND (opts & 1)
 #define QUIT       (opts & 2)
-       // parse commandline: prog [options] ifname script
+       // Parse commandline: prog [options] ifname script
        // exactly 2 args; -v accumulates and implies -f
        opt_complementary = "=2:vv:vf";
-       opts = getopt32(argc, argv, "fqr:v", &r_opt, &verbose);
+       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 early, before all bb_xx_msg calls */
+               // 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) != LINKLOCAL_ADDR
+                || (ntohl(ip.s_addr) & IN_CLASSB_NET) != G.localnet_ip
                ) {
                        bb_error_msg_and_die("invalid link address");
                }
+               chosen_nip = ip.s_addr;
        }
-       // On NOMMU reexec early (or else we will rerun things twice)
-#if !BB_MMU
-       if (!FOREGROUND)
-               bb_daemonize_or_rexec(DAEMON_CHDIR_ROOT, argv);
-#endif
-       argc -= optind;
-       argv += optind;
+       argv += optind - 1;
 
-       intf = argv[0];
-       script_av[0] = argv[1];
-       setenv("interface", intf, 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])
 
-       // initialize the interface (modprobe, ifup, etc)
-       script_av[1] = (char*)"init";
-       if (run(script_av, intf, NULL))
+       xsetenv("interface", argv_intf);
+
+       // Initialize the interface (modprobe, ifup, etc)
+       if (run(argv, "init", 0))
                return EXIT_FAILURE;
 
-       // initialize saddr
-       //memset(&saddr, 0, sizeof(saddr));
-       safe_strncpy(saddr.sa_data, intf, sizeof(saddr.sa_data));
+       // 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));
 
-       // open an ARP socket
-       fd = xsocket(PF_PACKET, SOCK_PACKET, htons(ETH_P_ARP));
-       // bind to the interface's ARP socket
-       xbind(fd, &saddr, sizeof(saddr));
+       // Bind to the interface's ARP socket
+       xbind(sock_fd, &G.iface_sockaddr, sizeof(G.iface_sockaddr));
 
-       // get the interface's ethernet address
+       // Get the interface's ethernet address
        //memset(&ifr, 0, sizeof(ifr));
-       strncpy(ifr.ifr_name, intf, sizeof(ifr.ifr_name));
-       xioctl(fd, SIOCGIFHWADDR, &ifr);
-       memcpy(&eth_addr, &ifr.ifr_hwaddr.sa_data, ETH_ALEN);
+       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
+       // 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.
-       // (SVID 3 bogon: who says that "short" is always 16 bits?)
-       seed48( (unsigned short*)&ifr.ifr_hwaddr.sa_data );
-       if (ip.s_addr == 0)
-               pick(&ip);
-
+       {
+               uint32_t t;
+               move_from_unaligned32(t, ((char *)&G.our_ethaddr + 2));
+               srand(t);
+       }
        // FIXME cases to handle:
        //  - zcip already running!
        //  - link already has local address... just defend/update
 
-       // daemonize now; don't delay system startup
+       // Daemonize now; don't delay system startup
        if (!FOREGROUND) {
 #if BB_MMU
-               bb_daemonize(DAEMON_CHDIR_ROOT);
+               bb_daemonize(0 /*was: DAEMON_CHDIR_ROOT*/);
 #endif
-               bb_info_msg("start, interface %s", intf);
+               bb_error_msg("start, interface %s", argv_intf);
        }
 
-       // run the dynamic address negotiation protocol,
+       // 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 else uses it
+       //  - 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;
+               unsigned deadline_us = deadline_us;
                struct arp_packet p;
+               int ip_conflict;
+               int n;
 
-               int source_ip_conflict = 0;
-               int target_ip_conflict = 0;
-
-               fds[0].fd = fd;
+               fds[0].fd = sock_fd;
                fds[0].events = POLLIN;
                fds[0].revents = 0;
 
-               // poll, being ready to adjust current timeout
+               // Poll, being ready to adjust current timeout
                if (!timeout_ms) {
-                       timeout_ms = ms_rdelay(PROBE_WAIT);
-                       // FIXME setsockopt(fd, SO_ATTACH_FILTER, ...) to
+                       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;
+               if (timeout_ms >= 0) {
+                       // 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, intf, nprobes, nclaims);
-               switch (poll(fds, 1, timeout_ms)) {
+               VDBG("...wait %d %s nsent=%u\n",
+                               timeout_ms, argv_intf, nsent);
 
-               // timeout
-               case 0:
-                       VDBG("state = %d\n", state);
+               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:
-                               // 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++;
+                               // No conflicting ARP packets were seen:
+                               // we can progress through the states
+                               if (nsent < PROBE_NUM) {
+                                       nsent++;
                                        VDBG("probe/%u %s@%s\n",
-                                                       nprobes, intf, inet_ntoa(ip));
-                                       arp(fd, &saddr, ARPOP_REQUEST,
-                                                       &eth_addr, null_ip,
-                                                       &null_addr, ip);
+                                                       nsent, argv_intf, nip_to_a(chosen_nip));
                                        timeout_ms = PROBE_MIN * 1000;
-                                       timeout_ms += ms_rdelay(PROBE_MAX - PROBE_MIN);
+                                       timeout_ms += random_delay_ms(PROBE_MAX - PROBE_MIN);
+                                       send_arp_request(0, &null_ethaddr, chosen_nip);
+                                       continue;
                                }
-                               else {
-                                       // Switch to announce state.
-                                       state = ANNOUNCE;
-                                       nclaims = 0;
-                                       VDBG("announce/%u %s@%s\n",
-                                                       nclaims, intf, inet_ntoa(ip));
-                                       arp(fd, &saddr, ARPOP_REQUEST,
-                                                       &eth_addr, ip,
-                                                       &eth_addr, ip);
-                                       timeout_ms = ANNOUNCE_INTERVAL * 1000;
-                               }
-                               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
+                               // Switch to announce state
+                               nsent = 0;
                                state = ANNOUNCE;
-                               nclaims = 0;
-                               VDBG("announce/%u %s@%s\n",
-                                               nclaims, intf, inet_ntoa(ip));
-                               arp(fd, &saddr, ARPOP_REQUEST,
-                                               &eth_addr, ip,
-                                               &eth_addr, ip);
-                               timeout_ms = ANNOUNCE_INTERVAL * 1000;
-                               break;
+                               goto send_announce;
                        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++;
+                               // 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",
-                                                       nclaims, intf, inet_ntoa(ip));
-                                       arp(fd, &saddr, ARPOP_REQUEST,
-                                                       &eth_addr, ip,
-                                                       &eth_addr, ip);
+                                                       nsent, argv_intf, nip_to_a(chosen_nip));
                                        timeout_ms = ANNOUNCE_INTERVAL * 1000;
+                                       send_arp_request(chosen_nip, &G.our_ethaddr, chosen_nip);
+                                       continue;
                                }
-                               else {
-                                       // Switch to monitor state.
-                                       state = MONITOR;
-                                       // link is ok to use earlier
-                                       // FIXME update filters
-                                       script_av[1] = (char*)"config";
-                                       run(script_av, intf, &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;
+                               // 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:
-                               // Invalid, should never happen.  Restart the whole protocol.
-                               state = PROBE;
-                               pick(&ip);
-                               timeout_ms = 0;
-                               nprobes = 0;
-                               nclaims = 0;
-                               break;
-                       } // switch (state)
-                       break; // case 0 (timeout)
-               // packets arriving
-               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;
-                                       if (!timeout_ms) timeout_ms = 1;
-                               }
+                       // 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;
                        }
+               }
 
-                       if ((fds[0].revents & POLLIN) == 0) {
-                               if (fds[0].revents & POLLERR) {
-                                       // FIXME: links routinely go down;
-                                       // this shouldn't necessarily exit.
-                                       bb_error_msg("%s: poll error", intf);
-                                       if (ready) {
-                                               script_av[1] = (char*)"deconfig";
-                                               run(script_av, intf, &ip);
-                                       }
-                                       return EXIT_FAILURE;
-                               }
-                               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
+               }
 
-                       // read ARP packet
-                       if (recv(fd, &p, sizeof(p), 0) < 0) {
-                               why = "recv";
-                               goto bad;
+               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;
                        }
-                       if (p.hdr.ether_type != htons(ETHERTYPE_ARP))
-                               continue;
+                       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("%s recv arp type=%d, op=%d,\n",
-                                       intf, ntohs(p.hdr.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));
-                       }
+               {
+                       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
-                       if (p.arp.arp_op != htons(ARPOP_REQUEST)
-                                       && p.arp.arp_op != htons(ARPOP_REPLY))
-                               continue;
-
-                       if (memcmp(p.arp.arp_spa, &ip.s_addr, sizeof(struct in_addr)) == 0 &&
-                               memcmp(&eth_addr, &p.arp.arp_sha, ETH_ALEN) != 0) {
-                               source_ip_conflict = 1;
+               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 (memcmp(p.arp.arp_tpa, &ip.s_addr, sizeof(struct in_addr)) == 0 &&
-                               p.arp.arp_op == htons(ARPOP_REQUEST) &&
-                               memcmp(&eth_addr, &p.arp.arp_tha, ETH_ALEN) != 0) {
-                               target_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;
+               }
 
-                       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", intf);
-                                               timeout_ms = RATE_LIMIT_INTERVAL * 1000;
-                                               state = RATE_LIMIT_PROBE;
-                                       }
-
-                                       // restart the whole protocol
-                                       pick(&ip);
-                                       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(fd, &saddr,
-                                                       ARPOP_REQUEST,
-                                                       &eth_addr, ip,
-                                                       &eth_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;
-                                       script_av[1] = (char*)"deconfig";
-                                       run(script_av, intf, &ip);
-
-                                       // restart the whole protocol
-                                       pick(&ip);
-                                       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;
-                               pick(&ip);
-                               timeout_ms = 0;
-                               nprobes = 0;
-                               nclaims = 0;
-                               break;
-                       } // switch state
-
-                       break; // case 1 (packets arriving)
-               default:
-                       why = "poll";
-                       goto bad;
-               } // switch poll
-       }
- bad:
-       bb_perror_msg("%s, %s", intf, why);
-       return EXIT_FAILURE;
+               // 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
 }