ntpd: new applet by Adam Tkac. +5k.
authorAdam Tkac <vonsch@gmail.com>
Sun, 22 Nov 2009 02:43:55 +0000 (03:43 +0100)
committerDenys Vlasenko <vda.linux@googlemail.com>
Sun, 22 Nov 2009 02:43:55 +0000 (03:43 +0100)
Signed-off-by: Adam Tkac <vonsch@gmail.com>
Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>
include/applets.h
include/usage.h
networking/Config.in
networking/Kbuild
networking/ntpd.c [new file with mode: 0644]

index 15ccf9cea6a9a19647e7c47b84248a8cb87e31e3..a2d1e95a598833c9e44283e67578e404c0bbafa6 100644 (file)
@@ -285,6 +285,7 @@ IF_NICE(APPLET(nice, _BB_DIR_BIN, _BB_SUID_DROP))
 IF_NMETER(APPLET(nmeter, _BB_DIR_USR_BIN, _BB_SUID_DROP))
 IF_NOHUP(APPLET(nohup, _BB_DIR_USR_BIN, _BB_SUID_DROP))
 IF_NSLOOKUP(APPLET(nslookup, _BB_DIR_USR_BIN, _BB_SUID_DROP))
+IF_NTPD(APPLET(ntpd, _BB_DIR_USR_SBIN, _BB_SUID_DROP))
 IF_OD(APPLET(od, _BB_DIR_USR_BIN, _BB_SUID_DROP))
 IF_OPENVT(APPLET(openvt, _BB_DIR_USR_BIN, _BB_SUID_DROP))
 //IF_PARSE(APPLET(parse, _BB_DIR_USR_BIN, _BB_SUID_DROP))
index eae9650ca8fc33520e444461b7436f6bd6fe8817..3ccf160bd9b22e0fe125c9e4f2a898f25292e5db 100644 (file)
        "Name:       debian\n" \
        "Address:    127.0.0.1\n"
 
+#define ntpd_trivial_usage \
+       "[-dngl] [-p PEER]..."
+#define ntpd_full_usage "\n\n" \
+       "NTP client/server\n" \
+     "\nOptions:" \
+     "\n       -d      Verbose" \
+     "\n       -n      Do not daemonize" \
+     "\n       -g      Set system time even if offset is > 1000 sec" \
+     "\n       -l      Run as server on port 123" \
+     "\n       -p PEER Obtain time from PEER (may be repeated)" \
+
 #define od_trivial_usage \
        "[-aBbcDdeFfHhIiLlOovXx] " IF_DESKTOP("[-t TYPE] ") "[FILE]"
 #define od_full_usage "\n\n" \
index 5f1b6f63680297b2b640a5f1fdce677afc9dd5e1..3d29622da874747eef49799d2c896762b6373769 100644 (file)
@@ -661,6 +661,20 @@ config NSLOOKUP
        help
          nslookup is a tool to query Internet name servers.
 
+config NTPD
+       bool "ntpd"
+       default n
+       help
+         The NTP client/server daemon.
+
+config FEATURE_NTPD_SERVER
+        bool "Make ntpd usable as a NTP server"
+        default y
+        depends on NTPD
+        help
+         Make ntpd usable as a NTP server. If you disable this option 
+         ntpd will be usable only as a NTP client.
+
 config PING
        bool "ping"
        default n
index 866d42f6ba24753eacc2b154f261db4da3d9fa6e..b0765bcf61aca75c5ec5d9a0a2b70826f699ccca 100644 (file)
@@ -27,6 +27,7 @@ lib-$(CONFIG_NAMEIF)       += nameif.o
 lib-$(CONFIG_NC)           += nc.o
 lib-$(CONFIG_NETSTAT)      += netstat.o
 lib-$(CONFIG_NSLOOKUP)     += nslookup.o
+lib-$(CONFIG_NTPD)         += ntpd.o
 lib-$(CONFIG_PING)         += ping.o
 lib-$(CONFIG_PING6)        += ping.o
 lib-$(CONFIG_PSCAN)        += pscan.o
diff --git a/networking/ntpd.c b/networking/ntpd.c
new file mode 100644 (file)
index 0000000..62a00a5
--- /dev/null
@@ -0,0 +1,999 @@
+/*
+ * NTP client/server, based on OpenNTPD 3.9p1
+ *
+ * Author: Adam Tkac <vonsch@gmail.com>
+ *
+ * Licensed under GPLv2, see file LICENSE in this tarball for details.
+ */
+
+#include "libbb.h"
+#include <netinet/ip.h> /* For IPTOS_LOWDELAY definition */
+
+#ifndef IP_PKTINFO
+# error "Sorry, your kernel has to support IP_PKTINFO"
+#endif
+
+#define        INTERVAL_QUERY_NORMAL           30      /* sync to peers every n secs */
+#define        INTERVAL_QUERY_PATHETIC         60
+#define        INTERVAL_QUERY_AGRESSIVE        5
+
+#define        TRUSTLEVEL_BADPEER              6
+#define        TRUSTLEVEL_PATHETIC             2
+#define        TRUSTLEVEL_AGRESSIVE            8
+#define        TRUSTLEVEL_MAX                  10
+
+#define        QSCALE_OFF_MIN                  0.05
+#define        QSCALE_OFF_MAX                  0.50
+
+#define        QUERYTIME_MAX           15      /* single query might take n secs max */
+#define        OFFSET_ARRAY_SIZE       8
+#define        SETTIME_MIN_OFFSET      180     /* min offset for settime at start */
+#define        SETTIME_TIMEOUT         15      /* max seconds to wait with -s */
+
+/* Style borrowed from NTP ref/tcpdump and updated for SNTPv4 (RFC2030). */
+
+/*
+ * RFC Section 3
+ *
+ *    0                   1                   2                   3
+ *    0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1
+ *   +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
+ *   |                         Integer Part                          |
+ *   +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
+ *   |                         Fraction Part                         |
+ *   +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
+ *
+ *    0                   1                   2                   3
+ *    0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1
+ *   +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
+ *   |            Integer Part       |     Fraction Part             |
+ *   +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
+*/
+typedef struct {
+       uint32_t int_partl;
+       uint32_t fractionl;
+} l_fixedpt_t;
+
+typedef struct {
+       uint16_t int_parts;
+       uint16_t fractions;
+} s_fixedpt_t;
+
+#define        NTP_DIGESTSIZE          16
+#define        NTP_MSGSIZE_NOAUTH      48
+#define        NTP_MSGSIZE             (NTP_MSGSIZE_NOAUTH + 4 + NTP_DIGESTSIZE)
+
+typedef struct {
+       uint8_t status; /* status of local clock and leap info */
+       uint8_t stratum;        /* Stratum level */
+       uint8_t ppoll;          /* poll value */
+       int8_t precision;
+       s_fixedpt_t rootdelay;
+       s_fixedpt_t dispersion;
+       uint32_t refid;
+       l_fixedpt_t reftime;
+       l_fixedpt_t orgtime;
+       l_fixedpt_t rectime;
+       l_fixedpt_t xmttime;
+       uint32_t keyid;
+       uint8_t digest[NTP_DIGESTSIZE];
+} ntp_msg_t;
+
+typedef struct {
+       int                     fd;
+       ntp_msg_t               msg;
+       double                  xmttime;
+} ntp_query_t;
+
+enum {
+       NTP_VERSION     = 4,
+       NTP_MAXSTRATUM  = 15,
+       /* Leap Second Codes (high order two bits) */
+       LI_NOWARNING    = (0 << 6),     /* no warning */
+       LI_PLUSSEC      = (1 << 6),     /* add a second (61 seconds) */
+       LI_MINUSSEC     = (2 << 6),     /* minus a second (59 seconds) */
+       LI_ALARM        = (3 << 6),     /* alarm condition */
+
+       /* Status Masks */
+       MODE_MASK       = (7 << 0),
+       VERSION_MASK    = (7 << 3),
+       LI_MASK         = (3 << 6),
+
+       /* Mode values */
+       MODE_RES0       = 0,    /* reserved */
+       MODE_SYM_ACT    = 1,    /* symmetric active */
+       MODE_SYM_PAS    = 2,    /* symmetric passive */
+       MODE_CLIENT     = 3,    /* client */
+       MODE_SERVER     = 4,    /* server */
+       MODE_BROADCAST  = 5,    /* broadcast */
+       MODE_RES1       = 6,    /* reserved for NTP control message */
+       MODE_RES2       = 7     /* reserved for private use */
+};
+
+#define        JAN_1970        2208988800UL    /* 1970 - 1900 in seconds */
+
+enum client_state {
+       STATE_NONE,
+       STATE_QUERY_SENT,
+       STATE_REPLY_RECEIVED
+};
+
+typedef struct {
+       double          rootdelay;
+       double          rootdispersion;
+       double          reftime;
+       uint32_t        refid;
+       uint32_t        refid4;
+       uint8_t         synced;
+       uint8_t         leap;
+       int8_t          precision;
+       uint8_t         poll;
+       uint8_t         stratum;
+} ntp_status_t;
+
+typedef struct {
+       ntp_status_t            status;
+       double                  offset;
+       double                  delay;
+       double                  error;
+       time_t                  rcvd;
+       uint8_t                 good;
+} ntp_offset_t;
+
+typedef struct {
+       len_and_sockaddr        *lsa;
+       ntp_query_t              query;
+       ntp_offset_t             reply[OFFSET_ARRAY_SIZE];
+       ntp_offset_t             update;
+       enum client_state        state;
+       time_t                   next;
+       time_t                   deadline;
+       uint8_t                  shift;
+       uint8_t                  trustlevel;
+} ntp_peer_t;
+
+
+struct globals {
+       unsigned        verbose;
+#if ENABLE_FEATURE_NTPD_SERVER
+       int             listen_fd;
+#endif
+       llist_t         *ntp_peers;
+       ntp_status_t    status;
+       uint32_t        scale;
+       uint8_t         settime;
+       uint8_t         firstadj;
+       smallint        peer_cnt;
+
+};
+#define G (*ptr_to_globals)
+
+
+static const int const_IPTOS_LOWDELAY = IPTOS_LOWDELAY;
+
+
+static void
+set_next(ntp_peer_t *p, time_t t)
+{
+       p->next = time(NULL) + t;
+       p->deadline = 0;
+}
+
+static void
+add_peers(const char *s)
+{
+       ntp_peer_t *p;
+
+       p = xzalloc(sizeof(*p));
+//TODO: big ntpd uses all IPs, not just 1st, do we need to mimic that?
+       p->lsa = xhost2sockaddr(s, 123);
+       p->query.fd = -1;
+       p->query.msg.status = MODE_CLIENT | (NTP_VERSION << 3);
+       if (STATE_NONE != 0)
+               p->state = STATE_NONE;
+       p->trustlevel = TRUSTLEVEL_PATHETIC;
+       p->query.fd = -1;
+       set_next(p, 0);
+
+       llist_add_to(&G.ntp_peers, p);
+       G.peer_cnt++;
+}
+
+static double
+gettime(void)
+{
+       struct timeval  tv;
+
+       gettimeofday(&tv, NULL); /* never fails */
+       return (tv.tv_sec + JAN_1970 + 1.0e-6 * tv.tv_usec);
+}
+
+
+static void
+d_to_tv(double d, struct timeval *tv)
+{
+       tv->tv_sec = (long)d;
+       tv->tv_usec = (d - tv->tv_sec) * 1000000;
+}
+
+static double
+lfp_to_d(l_fixedpt_t lfp)
+{
+       double  ret;
+
+       lfp.int_partl = ntohl(lfp.int_partl);
+       lfp.fractionl = ntohl(lfp.fractionl);
+       ret = (double)(lfp.int_partl) + ((double)lfp.fractionl / UINT_MAX);
+       return ret;
+}
+
+#if ENABLE_FEATURE_NTPD_SERVER
+static l_fixedpt_t
+d_to_lfp(double d)
+{
+       l_fixedpt_t     lfp;
+
+       lfp.int_partl = htonl((uint32_t)d);
+       lfp.fractionl = htonl((uint32_t)((d - (u_int32_t)d) * UINT_MAX));
+       return lfp;
+}
+#endif
+
+static double
+sfp_to_d(s_fixedpt_t sfp)
+{
+       double  ret;
+
+       sfp.int_parts = ntohs(sfp.int_parts);
+       sfp.fractions = ntohs(sfp.fractions);
+       ret = (double)(sfp.int_parts) + ((double)sfp.fractions / USHRT_MAX);
+       return ret;
+}
+
+#if ENABLE_FEATURE_NTPD_SERVER
+static s_fixedpt_t
+d_to_sfp(double d)
+{
+       s_fixedpt_t     sfp;
+
+       sfp.int_parts = htons((uint16_t)d);
+       sfp.fractions = htons((uint16_t)((d - (u_int16_t)d) * USHRT_MAX));
+       return sfp;
+}
+#endif
+
+static void
+set_deadline(ntp_peer_t *p, time_t t)
+{
+       p->deadline = time(NULL) + t;
+       p->next = 0;
+}
+
+static time_t
+error_interval(void)
+{
+       time_t interval, r;
+
+       interval = INTERVAL_QUERY_PATHETIC * QSCALE_OFF_MAX / QSCALE_OFF_MIN;
+       r = random() % (interval / 10);
+       return (interval + r);
+}
+
+static int
+sendmsg_wrap(int fd,
+               const struct sockaddr *from, const struct sockaddr *to, socklen_t addrlen,
+               ntp_msg_t *msg, ssize_t len)
+{
+       ssize_t ret;
+
+       errno = 0;
+       if (!from) {
+               ret = sendto(fd, msg, len, 0, to, addrlen);
+       } else {
+               ret = send_to_from(fd, msg, len, 0, to, from, addrlen);
+       }
+       if (ret != len) {
+               bb_perror_msg("send failed");
+               return -1;
+       }
+       return 0;
+}
+
+static int
+client_query(ntp_peer_t *p)
+{
+       if (p->query.fd == -1) {
+               p->query.fd = xsocket(p->lsa->u.sa.sa_family, SOCK_DGRAM, 0);
+#if ENABLE_FEATURE_IPV6
+               if (p->lsa->u.sa.sa_family == AF_INET)
+#endif
+                       setsockopt(p->query.fd, IPPROTO_IP, IP_TOS, &const_IPTOS_LOWDELAY, sizeof(const_IPTOS_LOWDELAY));
+       }
+
+       /*
+        * Send out a random 64-bit number as our transmit time.  The NTP
+        * server will copy said number into the originate field on the
+        * response that it sends us.  This is totally legal per the SNTP spec.
+        *
+        * The impact of this is two fold: we no longer send out the current
+        * system time for the world to see (which may aid an attacker), and
+        * it gives us a (not very secure) way of knowing that we're not
+        * getting spoofed by an attacker that can't capture our traffic
+        * but can spoof packets from the NTP server we're communicating with.
+        *
+        * Save the real transmit timestamp locally.
+        */
+
+       p->query.msg.xmttime.int_partl = random();
+       p->query.msg.xmttime.fractionl = random();
+       p->query.xmttime = gettime();
+
+       if (sendmsg_wrap(p->query.fd, /*from:*/ NULL, /*to:*/ &p->lsa->u.sa, /*addrlen:*/ p->lsa->len,
+                       &p->query.msg, NTP_MSGSIZE_NOAUTH) == -1) {
+               set_next(p, INTERVAL_QUERY_PATHETIC);
+               return -1;
+       }
+
+       p->state = STATE_QUERY_SENT;
+       set_deadline(p, QUERYTIME_MAX);
+
+       return 0;
+}
+
+static int
+offset_compare(const void *aa, const void *bb)
+{
+       const ntp_peer_t * const *a;
+       const ntp_peer_t * const *b;
+
+       a = aa;
+       b = bb;
+
+       if ((*a)->update.offset < (*b)->update.offset)
+               return -1;
+       return ((*a)->update.offset > (*b)->update.offset);
+}
+
+static uint32_t
+updated_scale(double offset)
+{
+       if (offset < 0)
+               offset = -offset;
+
+       if (offset > QSCALE_OFF_MAX)
+               return 1;
+       if (offset < QSCALE_OFF_MIN)
+               return QSCALE_OFF_MAX / QSCALE_OFF_MIN;
+       return QSCALE_OFF_MAX / offset;
+}
+
+static void
+adjtime_wrap(void)
+{
+       ntp_peer_t       *p;
+       unsigned          offset_cnt;
+       int               i = 0;
+       ntp_peer_t      **peers;
+       double            offset_median;
+       llist_t          *item;
+       len_and_sockaddr *lsa;
+       struct timeval    tv, olddelta;
+
+       offset_cnt = 0;
+       for (item = G.ntp_peers; item != NULL; item = item->link) {
+               p = (ntp_peer_t *) item->data;
+               if (p->trustlevel < TRUSTLEVEL_BADPEER)
+                       continue;
+               if (!p->update.good)
+                       return;
+               offset_cnt++;
+       }
+
+       peers = xzalloc(sizeof(ntp_peer_t *) * offset_cnt);
+       for (item = G.ntp_peers; item != NULL; item = item->link) {
+               p = (ntp_peer_t *) item->data;
+               if (p->trustlevel < TRUSTLEVEL_BADPEER)
+                       continue;
+               peers[i++] = p;
+       }
+
+       qsort(peers, offset_cnt, sizeof(ntp_peer_t *), offset_compare);
+
+       if (offset_cnt != 0) {
+               if ((offset_cnt & 1) == 0) {
+//TODO: try offset_cnt /= 2...
+                       offset_median =
+                           (peers[offset_cnt / 2 - 1]->update.offset +
+                           peers[offset_cnt / 2]->update.offset) / 2;
+                       G.status.rootdelay =
+                           (peers[offset_cnt / 2 - 1]->update.delay +
+                           peers[offset_cnt / 2]->update.delay) / 2;
+                       G.status.stratum = MAX(
+                           peers[offset_cnt / 2 - 1]->update.status.stratum,
+                           peers[offset_cnt / 2]->update.status.stratum);
+               } else {
+                       offset_median = peers[offset_cnt / 2]->update.offset;
+                       G.status.rootdelay = peers[offset_cnt / 2]->update.delay;
+                       G.status.stratum = peers[offset_cnt / 2]->update.status.stratum;
+               }
+               G.status.leap = peers[offset_cnt / 2]->update.status.leap;
+
+               bb_info_msg("adjusting local clock by %fs", offset_median);
+
+               d_to_tv(offset_median, &tv);
+               if (adjtime(&tv, &olddelta) == -1)
+                       bb_error_msg("adjtime failed");
+               else if (!G.firstadj
+                && olddelta.tv_sec == 0
+                && olddelta.tv_usec == 0
+                && !G.status.synced
+               ) {
+                       bb_info_msg("clock synced");
+                       G.status.synced = 1;
+               } else if (G.status.synced) {
+                       bb_info_msg("clock unsynced");
+                       G.status.synced = 0;
+               }
+
+               G.firstadj = 0;
+               G.status.reftime = gettime();
+               G.status.stratum++;     /* one more than selected peer */
+               G.scale = updated_scale(offset_median);
+
+               G.status.refid4 = peers[offset_cnt / 2]->update.status.refid4;
+
+               lsa = peers[offset_cnt / 2]->lsa;
+               G.status.refid =
+#if ENABLE_FEATURE_IPV6
+                       lsa->u.sa.sa_family != AF_INET ?
+                               G.status.refid4 :
+#endif
+                               lsa->u.sin.sin_addr.s_addr;
+       }
+
+       free(peers);
+
+       for (item = G.ntp_peers; item != NULL; item = item->link) {
+               p = (ntp_peer_t *) item->data;
+               p->update.good = 0;
+       }
+}
+
+static void
+settime(double offset)
+{
+       ntp_peer_t *p;
+       llist_t         *item;
+       struct timeval  tv, curtime;
+       char            buf[80];
+       time_t          tval;
+
+#if 0
+       if (!G.settime)
+               return;
+#endif
+
+       /* if the offset is small, don't call settimeofday */
+       if (offset < SETTIME_MIN_OFFSET && offset > -SETTIME_MIN_OFFSET)
+               return;
+
+       gettimeofday(&curtime, NULL); /* never fails */
+
+       d_to_tv(offset, &tv);
+       curtime.tv_usec += tv.tv_usec + 1000000;
+       curtime.tv_sec += tv.tv_sec - 1 + (curtime.tv_usec / 1000000);
+       curtime.tv_usec %= 1000000;
+
+       if (settimeofday(&curtime, NULL) == -1) {
+               bb_error_msg("settimeofday");
+               return;
+       }
+
+       G.settime = 0;
+
+       tval = curtime.tv_sec;
+       strftime(buf, sizeof(buf), "%a %b %e %H:%M:%S %Z %Y", localtime(&tval));
+
+       /* Do we want to print message below to system log when daemonized? */
+       bb_info_msg("set local clock to %s (offset %fs)", buf, offset);
+
+       for (item = G.ntp_peers; item != NULL; item = item->link) {
+               p = (ntp_peer_t *) item->data;
+               if (p->next)
+                       p->next -= offset;
+               if (p->deadline)
+                       p->deadline -= offset;
+       }
+}
+
+static void
+client_update(ntp_peer_t *p)
+{
+       int i, best = 0, good = 0;
+
+       /*
+        * clock filter
+        * find the offset which arrived with the lowest delay
+        * use that as the peer update
+        * invalidate it and all older ones
+        */
+
+       for (i = 0; good == 0 && i < OFFSET_ARRAY_SIZE; i++) {
+               if (p->reply[i].good) {
+                       good++;
+                       best = i;
+               }
+       }
+
+       for (; i < OFFSET_ARRAY_SIZE; i++) {
+               if (p->reply[i].good) {
+                       good++;
+                       if (p->reply[i].delay < p->reply[best].delay)
+                               best = i;
+               }
+       }
+
+       if (good < 8)
+               return;
+
+       memcpy(&p->update, &p->reply[best], sizeof(p->update));
+       adjtime_wrap();
+
+       for (i = 0; i < OFFSET_ARRAY_SIZE; i++)
+               if (p->reply[i].rcvd <= p->reply[best].rcvd)
+                       p->reply[i].good = 0;
+}
+
+static time_t
+scale_interval(time_t requested)
+{
+       time_t interval, r;
+
+       interval = requested * G.scale;
+       r = random() % MAX(5, interval / 10);
+       return (interval + r);
+}
+
+static void
+client_dispatch(ntp_peer_t *p)
+{
+       char                     *addr;
+       ssize_t                  size;
+       ntp_msg_t                msg;
+       double                   T1, T2, T3, T4;
+       time_t                   interval;
+       ntp_offset_t            *offset;
+
+       addr = xmalloc_sockaddr2dotted_noport(&p->lsa->u.sa);
+
+       size = recvfrom(p->query.fd, &msg, sizeof(msg), 0, NULL, NULL);
+       if (size == -1) {
+               bb_perror_msg("recvfrom(%s) error", addr);
+               if (errno == EHOSTUNREACH || errno == EHOSTDOWN
+                || errno == ENETUNREACH || errno == ENETDOWN
+                || errno == ECONNREFUSED || errno == EADDRNOTAVAIL
+               ) {
+//TODO: always do this?
+                       set_next(p, error_interval());
+                       goto bail;
+               }
+               xfunc_die();
+       }
+
+       T4 = gettime();
+
+       if (size != NTP_MSGSIZE_NOAUTH && size != NTP_MSGSIZE) {
+               bb_error_msg("malformed packet received from %s", addr);
+               goto bail;
+       }
+
+       if (msg.orgtime.int_partl != p->query.msg.xmttime.int_partl
+        || msg.orgtime.fractionl != p->query.msg.xmttime.fractionl
+       ) {
+               goto bail;
+       }
+
+       if ((msg.status & LI_ALARM) == LI_ALARM
+        || msg.stratum == 0
+        || msg.stratum > NTP_MAXSTRATUM
+       ) {
+               interval = error_interval();
+               bb_info_msg("reply from %s: not synced, next query %ds", addr, (int) interval);
+               goto bail;
+       }
+
+       /*
+        * From RFC 2030 (with a correction to the delay math):
+        *
+        *     Timestamp Name          ID   When Generated
+        *     ------------------------------------------------------------
+        *     Originate Timestamp     T1   time request sent by client
+        *     Receive Timestamp       T2   time request received by server
+        *     Transmit Timestamp      T3   time reply sent by server
+        *     Destination Timestamp   T4   time reply received by client
+        *
+        *  The roundtrip delay d and local clock offset t are defined as
+        *
+        *    d = (T4 - T1) - (T3 - T2)     t = ((T2 - T1) + (T3 - T4)) / 2.
+        */
+
+       T1 = p->query.xmttime;
+       T2 = lfp_to_d(msg.rectime);
+       T3 = lfp_to_d(msg.xmttime);
+
+       offset = &p->reply[p->shift];
+
+       offset->offset = ((T2 - T1) + (T3 - T4)) / 2;
+       offset->delay = (T4 - T1) - (T3 - T2);
+       if (offset->delay < 0) {
+               interval = error_interval();
+               set_next(p, interval);
+               bb_info_msg("reply from %s: negative delay %f", addr, p->reply[p->shift].delay);
+               goto bail;
+       }
+       offset->error = (T2 - T1) - (T3 - T4);
+       offset->rcvd = time(NULL);
+       offset->good = 1;
+
+       offset->status.leap = (msg.status & LI_MASK);
+       offset->status.precision = msg.precision;
+       offset->status.rootdelay = sfp_to_d(msg.rootdelay);
+       offset->status.rootdispersion = sfp_to_d(msg.dispersion);
+       offset->status.refid = ntohl(msg.refid);
+       offset->status.refid4 = msg.xmttime.fractionl;
+       offset->status.reftime = lfp_to_d(msg.reftime);
+       offset->status.poll = msg.ppoll;
+       offset->status.stratum = msg.stratum;
+
+       if (p->trustlevel < TRUSTLEVEL_PATHETIC)
+               interval = scale_interval(INTERVAL_QUERY_PATHETIC);
+       else if (p->trustlevel < TRUSTLEVEL_AGRESSIVE)
+               interval = scale_interval(INTERVAL_QUERY_AGRESSIVE);
+       else
+               interval = scale_interval(INTERVAL_QUERY_NORMAL);
+
+       set_next(p, interval);
+       p->state = STATE_REPLY_RECEIVED;
+
+       /* every received reply which we do not discard increases trust */
+       if (p->trustlevel < TRUSTLEVEL_MAX) {
+               if (p->trustlevel < TRUSTLEVEL_BADPEER
+                && p->trustlevel + 1 >= TRUSTLEVEL_BADPEER
+               ) {
+                       bb_info_msg("peer %s now valid", addr);
+               }
+               p->trustlevel++;
+       }
+
+       bb_info_msg("reply from %s: offset %f delay %f, next query %ds", addr,
+                       offset->offset, offset->delay, (int) interval);
+
+       client_update(p);
+       if (!G.settime)
+               settime(offset->offset);
+
+       if (++p->shift >= OFFSET_ARRAY_SIZE)
+               p->shift = 0;
+
+ bail:
+       free(addr);
+}
+
+#if ENABLE_FEATURE_NTPD_SERVER
+static void
+server_dispatch(int fd)
+{
+       ssize_t                 size;
+       uint8_t                 version;
+       double                  rectime;
+       len_and_sockaddr        *to;
+       struct sockaddr         *from;
+       ntp_msg_t               query, reply;
+
+       to = get_sock_lsa(G.listen_fd);
+       from = xzalloc(to->len);
+
+       size = recv_from_to(fd, &query, sizeof(query), 0, from, &to->u.sa, to->len);
+       if (size == -1)
+               bb_error_msg_and_die("recv_from_to");
+       if (size != NTP_MSGSIZE_NOAUTH && size != NTP_MSGSIZE) {
+               char *addr = xmalloc_sockaddr2dotted_noport(from);
+               bb_error_msg("malformed packet received from %s", addr);
+               free(addr);
+               goto bail;
+       }
+
+       rectime = gettime();
+       version = (query.status & VERSION_MASK) >> 3;
+
+       memset(&reply, 0, sizeof(reply));
+       reply.status = G.status.synced ? G.status.leap : LI_ALARM;
+       reply.status |= (query.status & VERSION_MASK);
+       reply.status |= ((query.status & MODE_MASK) == MODE_CLIENT) ?
+                        MODE_SERVER : MODE_SYM_PAS;
+       reply.stratum = G.status.stratum;
+       reply.ppoll = query.ppoll;
+       reply.precision = G.status.precision;
+       reply.rectime = d_to_lfp(rectime);
+       reply.reftime = d_to_lfp(G.status.reftime);
+       reply.xmttime = d_to_lfp(gettime());
+       reply.orgtime = query.xmttime;
+       reply.rootdelay = d_to_sfp(G.status.rootdelay);
+       reply.refid = (version > 3) ? G.status.refid4 : G.status.refid;
+
+       /* We reply from the address packet was sent to,
+        * this makes to/from look swapped here: */
+       sendmsg_wrap(fd, /*from:*/ &to->u.sa, /*to:*/ from, /*addrlen:*/ to->len,
+                       &reply, size);
+
+ bail:
+       free(to);
+       free(from);
+}
+#endif
+
+/* Upstream ntpd's options:
+ *
+ * -4   Force DNS resolution of host names to the IPv4 namespace.
+ * -6   Force DNS resolution of host names to the IPv6 namespace.
+ * -a   Require cryptographic authentication for broadcast client,
+ *      multicast client and symmetric passive associations.
+ *      This is the default.
+ * -A   Do not require cryptographic authentication for broadcast client,
+ *      multicast client and symmetric passive associations.
+ *      This is almost never a good idea.
+ * -b   Enable the client to synchronize to broadcast servers.
+ * -c conffile
+ *      Specify the name and path of the configuration file,
+ *      default /etc/ntp.conf
+ * -d   Specify debugging mode. This option may occur more than once,
+ *      with each occurrence indicating greater detail of display.
+ * -D level
+ *      Specify debugging level directly.
+ * -f driftfile
+ *      Specify the name and path of the frequency file.
+ *      This is the same operation as the "driftfile FILE"
+ *      configuration command.
+ * -g   Normally, ntpd exits with a message to the system log
+ *      if the offset exceeds the panic threshold, which is 1000 s
+ *      by default. This option allows the time to be set to any value
+ *      without restriction; however, this can happen only once.
+ *      If the threshold is exceeded after that, ntpd will exit
+ *      with a message to the system log. This option can be used
+ *      with the -q and -x options. See the tinker command for other options.
+ * -i jaildir
+ *      Chroot the server to the directory jaildir. This option also implies
+ *      that the server attempts to drop root privileges at startup
+ *      (otherwise, chroot gives very little additional security).
+ *      You may need to also specify a -u option.
+ * -k keyfile
+ *      Specify the name and path of the symmetric key file,
+ *      default /etc/ntp/keys. This is the same operation
+ *      as the "keys FILE" configuration command.
+ * -l logfile
+ *      Specify the name and path of the log file. The default
+ *      is the system log file. This is the same operation as
+ *      the "logfile FILE" configuration command.
+ * -L   Do not listen to virtual IPs. The default is to listen.
+ * -n   Don't fork.
+ * -N   To the extent permitted by the operating system,
+ *      run the ntpd at the highest priority.
+ * -p pidfile
+ *      Specify the name and path of the file used to record the ntpd
+ *      process ID. This is the same operation as the "pidfile FILE"
+ *      configuration command.
+ * -P priority
+ *      To the extent permitted by the operating system,
+ *      run the ntpd at the specified priority.
+ * -q   Exit the ntpd just after the first time the clock is set.
+ *      This behavior mimics that of the ntpdate program, which is
+ *      to be retired. The -g and -x options can be used with this option.
+ *      Note: The kernel time discipline is disabled with this option.
+ * -r broadcastdelay
+ *      Specify the default propagation delay from the broadcast/multicast
+ *      server to this client. This is necessary only if the delay
+ *      cannot be computed automatically by the protocol.
+ * -s statsdir
+ *      Specify the directory path for files created by the statistics
+ *      facility. This is the same operation as the "statsdir DIR"
+ *      configuration command.
+ * -t key
+ *      Add a key number to the trusted key list. This option can occur
+ *      more than once.
+ * -u user[:group]
+ *      Specify a user, and optionally a group, to switch to.
+ * -v variable
+ * -V variable
+ *      Add a system variable listed by default.
+ * -x   Normally, the time is slewed if the offset is less than the step
+ *      threshold, which is 128 ms by default, and stepped if above
+ *      the threshold. This option sets the threshold to 600 s, which is
+ *      well within the accuracy window to set the clock manually.
+ *      Note: since the slew rate of typical Unix kernels is limited
+ *      to 0.5 ms/s, each second of adjustment requires an amortization
+ *      interval of 2000 s. Thus, an adjustment as much as 600 s
+ *      will take almost 14 days to complete. This option can be used
+ *      with the -g and -q options. See the tinker command for other options.
+ *      Note: The kernel time discipline is disabled with this option.
+ */
+
+enum {
+       OPT_n = (1 << 0),
+       OPT_g = (1 << 1),
+       OPT_p = (1 << 2),
+       OPT_l = (1 << 3),
+};
+
+/* By doing init in a separate function we decrease stack usage
+ * in main loop.
+ */
+static NOINLINE void ntp_init(char **argv)
+{
+       unsigned opts;
+       llist_t *peers;
+
+       tzset();
+
+       if (getuid())
+               bb_error_msg_and_die("need root privileges");
+
+       peers = NULL;
+       opt_complementary = "dd:p::"; /* d: counter, p: list */
+       opts = getopt32(argv,
+                       "ng" /* compat */
+                       "p:"IF_FEATURE_NTPD_SERVER("l") /* NOT compat */
+                       "d" /* compat */
+                       "46aAbLNx", /* compat, ignored */
+                       &peers, &G.verbose);
+#if ENABLE_FEATURE_NTPD_SERVER
+       G.listen_fd = -1;
+       if (opts & OPT_l) {
+               G.listen_fd = create_and_bind_dgram_or_die(NULL, 123);
+               socket_want_pktinfo(G.listen_fd);
+               setsockopt(G.listen_fd, IPPROTO_IP, IP_TOS, &const_IPTOS_LOWDELAY, sizeof(const_IPTOS_LOWDELAY));
+       }
+#endif
+       if (opts & OPT_g)
+               G.settime = 1;
+       while (peers)
+               add_peers(llist_pop(&peers));
+       if (!(opts & OPT_n)) {
+               logmode = LOGMODE_NONE;
+               bb_daemonize(DAEMON_DEVNULL_STDIO);
+       }
+
+       /* Set some globals */
+       {
+               int prec = 0;
+               int b;
+#if 0
+               struct timespec tp;
+               /* We can use sys_clock_getres but assuming 10ms tick should be fine */
+               clock_getres(CLOCK_REALTIME, &tp);
+               tp.tv_sec = 0;
+               tp.tv_nsec = 10000000;
+               b = 1000000000 / tp.tv_nsec;    /* convert to Hz */
+#else
+               b = 100; /* b = 1000000000/10000000 = 100 */
+#endif
+               while (b > 1)
+                       prec--, b >>= 1;
+               G.status.precision = prec;
+       }
+       G.scale = 1;
+       G.firstadj = 1;
+
+       bb_signals((1 << SIGTERM) | (1 << SIGINT), record_signo);
+       bb_signals((1 << SIGPIPE) | (1 << SIGHUP), SIG_IGN);
+}
+
+int ntpd_main(int argc UNUSED_PARAM, char **argv) MAIN_EXTERNALLY_VISIBLE;
+int ntpd_main(int argc UNUSED_PARAM, char **argv)
+{
+       struct globals g;
+       unsigned new_cnt;
+       struct pollfd *pfd;
+       ntp_peer_t **idx2peer;
+
+       memset(&g, 0, sizeof(g));
+       SET_PTR_TO_GLOBALS(&g);
+
+       ntp_init(argv);
+
+       new_cnt = g.peer_cnt;
+       idx2peer = xzalloc(sizeof(void *) * new_cnt);
+#if ENABLE_FEATURE_NTPD_SERVER
+       if (g.listen_fd != -1)
+               new_cnt++;
+#endif
+       pfd = xzalloc(sizeof(pfd[0]) * new_cnt);
+
+       while (!bb_got_signal) {
+               llist_t *item;
+               unsigned i, j, idx_peers;
+               unsigned sent_cnt, trial_cnt;
+               int nfds, timeout;
+               time_t nextaction;
+
+               nextaction = time(NULL) + 3600;
+
+               i = 0;
+#if ENABLE_FEATURE_NTPD_SERVER
+               if (g.listen_fd != -1) {
+                       pfd[0].fd = g.listen_fd;
+                       pfd[0].events = POLLIN;
+                       i++;
+               }
+#endif
+               idx_peers = i;
+               sent_cnt = trial_cnt = 0;
+               for (item = g.ntp_peers; item != NULL; item = item->link) {
+                       ntp_peer_t *p = (ntp_peer_t *) item->data;
+
+                       if (p->next > 0 && p->next <= time(NULL)) {
+                               trial_cnt++;
+                               if (client_query(p) == 0)
+                                       sent_cnt++;
+                       }
+                       if (p->next > 0 && p->next < nextaction)
+                               nextaction = p->next;
+                       if (p->deadline > 0 && p->deadline < nextaction)
+                               nextaction = p->deadline;
+
+                       if (p->deadline > 0 && p->deadline <= time(NULL)) {
+                               char *addr = xmalloc_sockaddr2dotted_noport(&p->lsa->u.sa);
+
+                               timeout = error_interval();
+                               bb_info_msg("no reply from %s received in time, "
+                                               "next query %ds", addr, timeout);
+                               if (p->trustlevel >= TRUSTLEVEL_BADPEER) {
+                                       p->trustlevel /= 2;
+                                       if (p->trustlevel < TRUSTLEVEL_BADPEER)
+                                               bb_info_msg("peer %s now invalid", addr);
+                               }
+                               free(addr);
+
+                               set_next(p, timeout);
+                       }
+
+                       if (p->state == STATE_QUERY_SENT) {
+                               pfd[i].fd = p->query.fd;
+                               pfd[i].events = POLLIN;
+                               idx2peer[i - idx_peers] = p;
+                               i++;
+                       }
+               }
+
+               if (g.settime
+                && ((trial_cnt > 0 && sent_cnt == 0) || g.peer_cnt == 0)
+               ) {
+                       settime(0);     /* no good peers, don't wait */
+               }
+
+               timeout = nextaction - time(NULL);
+               if (timeout < 0)
+                       timeout = 0;
+
+               if (g.verbose)
+                       bb_error_msg("entering poll %u secs", timeout);
+               nfds = poll(pfd, i, timeout * 1000);
+
+               j = 0;
+#if ENABLE_FEATURE_NTPD_SERVER
+               for (; nfds > 0 && j < idx_peers; j++) {
+                       if (pfd[j].revents & (POLLIN|POLLERR)) {
+                               nfds--;
+                               server_dispatch(pfd[j].fd);
+                       }
+               }
+#endif
+               for (; nfds > 0 && j < i; j++) {
+                       if (pfd[j].revents & (POLLIN|POLLERR)) {
+                               nfds--;
+                               client_dispatch(idx2peer[j - idx_peers]);
+                       }
+               }
+       } /* while (!bb_got_signal) */
+
+       kill_myself_with_sig(bb_got_signal);
+}