X-Git-Url: https://git.librecmc.org/?a=blobdiff_plain;f=networking%2Fntpd.c;h=bfd5705fcd30a8ab2293eec6baa2c7a02ec6c119;hb=2b1559056cf32c42675ecd937796e1455bcb5c2c;hp=03fe448ca92eb6cea650bb7d61ba3024ee92fba1;hpb=c009d35f0060905842c779fdf57535e0127cb238;p=oweals%2Fbusybox.git diff --git a/networking/ntpd.c b/networking/ntpd.c index 03fe448ca..bfd5705fc 100644 --- a/networking/ntpd.c +++ b/networking/ntpd.c @@ -1,35 +1,74 @@ /* * NTP client/server, based on OpenNTPD 3.9p1 * - * Author: Adam Tkac + * Busybox port author: Adam Tkac (C) 2009 * - * Licensed under GPLv2, see file LICENSE in this source tree. + * OpenNTPd 3.9p1 copyright holders: + * Copyright (c) 2003, 2004 Henning Brauer + * Copyright (c) 2004 Alexander Guy + * + * OpenNTPd code is licensed under ISC-style licence: + * + * Permission to use, copy, modify, and distribute this software for any + * purpose with or without fee is hereby granted, provided that the above + * copyright notice and this permission notice appear in all copies. + * + * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES + * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF + * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR + * ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES + * WHATSOEVER RESULTING FROM LOSS OF MIND, USE, DATA OR PROFITS, WHETHER + * IN AN ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING + * OUT OF OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE. + *********************************************************************** * * Parts of OpenNTPD clock syncronization code is replaced by - * code which is based on ntp-4.2.6, whuch carries the following + * code which is based on ntp-4.2.6, which carries the following * copyright notice: * - *********************************************************************** - * * - * Copyright (c) University of Delaware 1992-2009 * - * * - * Permission to use, copy, modify, and distribute this software and * - * its documentation for any purpose with or without fee is hereby * - * granted, provided that the above copyright notice appears in all * - * copies and that both the copyright notice and this permission * - * notice appear in supporting documentation, and that the name * - * University of Delaware not be used in advertising or publicity * - * pertaining to distribution of the software without specific, * - * written prior permission. The University of Delaware makes no * - * representations about the suitability this software for any * - * purpose. It is provided "as is" without express or implied * - * warranty. * - * * + * Copyright (c) University of Delaware 1992-2009 + * + * Permission to use, copy, modify, and distribute this software and + * its documentation for any purpose with or without fee is hereby + * granted, provided that the above copyright notice appears in all + * copies and that both the copyright notice and this permission + * notice appear in supporting documentation, and that the name + * University of Delaware not be used in advertising or publicity + * pertaining to distribution of the software without specific, + * written prior permission. The University of Delaware makes no + * representations about the suitability this software for any + * purpose. It is provided "as is" without express or implied warranty. *********************************************************************** */ +//config:config NTPD +//config: bool "ntpd" +//config: default y +//config: select PLATFORM_LINUX +//config: help +//config: The NTP client/server daemon. +//config: +//config:config FEATURE_NTPD_SERVER +//config: bool "Make ntpd usable as a NTP server" +//config: default y +//config: depends on NTPD +//config: help +//config: Make ntpd usable as a NTP server. If you disable this option +//config: ntpd will be usable only as a NTP client. +//config: +//config:config FEATURE_NTPD_CONF +//config: bool "Make ntpd understand /etc/ntp.conf" +//config: default y +//config: depends on NTPD +//config: help +//config: Make ntpd look in /etc/ntp.conf for peers. Only "server address" +//config: is supported. + +//applet:IF_NTPD(APPLET(ntpd, BB_DIR_USR_SBIN, BB_SUID_DROP)) + +//kbuild:lib-$(CONFIG_NTPD) += ntpd.o //usage:#define ntpd_trivial_usage -//usage: "[-dnqNw"IF_FEATURE_NTPD_SERVER("l")"] [-S PROG] [-p PEER]..." +//usage: "[-dnqNw"IF_FEATURE_NTPD_SERVER("l -I IFACE")"] [-S PROG] [-p PEER]..." //usage:#define ntpd_full_usage "\n\n" //usage: "NTP client/server\n" //usage: "\n -d Verbose" @@ -37,11 +76,20 @@ //usage: "\n -q Quit after clock is set" //usage: "\n -N Run at high priority" //usage: "\n -w Do not set time (only query peers), implies -n" -//usage: IF_FEATURE_NTPD_SERVER( -//usage: "\n -l Run as server on port 123" -//usage: ) //usage: "\n -S PROG Run PROG after stepping time, stratum change, and every 11 mins" //usage: "\n -p PEER Obtain time from PEER (may be repeated)" +//usage: IF_FEATURE_NTPD_CONF( +//usage: "\n If -p is not given, 'server HOST' lines" +//usage: "\n from /etc/ntp.conf are used" +//usage: ) +//usage: IF_FEATURE_NTPD_SERVER( +//usage: "\n -l Also run as server on port 123" +//usage: "\n -I IFACE Bind server to IFACE, implies -l" +//usage: ) + +// -l and -p options are not compatible with "standard" ntpd: +// it has them as "-l logfile" and "-p pidfile". +// -S and -w are not compat either, "standard" ntpd has no such opts. #include "libbb.h" #include @@ -90,7 +138,7 @@ * * Made some changes to speed up re-syncing after our clock goes bad * (tested with suspending my laptop): - * - if largish offset (>= STEP_THRESHOLD * 8 == 1 sec) is seen + * - if largish offset (>= STEP_THRESHOLD == 1 sec) is seen * from a peer, schedule next query for this peer soon * without drastically lowering poll interval for everybody. * This makes us collect enough data for step much faster: @@ -101,36 +149,48 @@ * datapoints after the step. */ -#define RETRY_INTERVAL 5 /* on error, retry in N secs */ -#define RESPONSE_INTERVAL 15 /* wait for reply up to N secs */ #define INITIAL_SAMPLES 4 /* how many samples do we want for init */ #define BAD_DELAY_GROWTH 4 /* drop packet if its delay grew by more than this */ -/* Clock discipline parameters and constants */ +#define RETRY_INTERVAL 32 /* on send/recv error, retry in N secs (need to be power of 2) */ +#define NOREPLY_INTERVAL 512 /* sent, but got no reply: cap next query by this many seconds */ +#define RESPONSE_INTERVAL 16 /* wait for reply up to N secs */ +#define HOSTNAME_INTERVAL 5 /* hostname lookup failed. Wait N secs for next try */ /* Step threshold (sec). std ntpd uses 0.128. - * Using exact power of 2 (1/8) results in smaller code */ -#define STEP_THRESHOLD 0.125 -#define WATCH_THRESHOLD 128 /* stepout threshold (sec). std ntpd uses 900 (11 mins (!)) */ + */ +#define STEP_THRESHOLD 1 +/* Slew threshold (sec): adjtimex() won't accept offsets larger than this. + * Using exact power of 2 (1/8) results in smaller code + */ +#define SLEW_THRESHOLD 0.125 +/* Stepout threshold (sec). std ntpd uses 900 (11 mins (!)) */ +#define WATCH_THRESHOLD 128 /* NB: set WATCH_THRESHOLD to ~60 when debugging to save time) */ //UNUSED: #define PANIC_THRESHOLD 1000 /* panic threshold (sec) */ +/* + * If we got |offset| > BIGOFF from a peer, cap next query interval + * for this peer by this many seconds: + */ +#define BIGOFF STEP_THRESHOLD +#define BIGOFF_INTERVAL (1 << 7) /* 128 s */ + #define FREQ_TOLERANCE 0.000015 /* frequency tolerance (15 PPM) */ #define BURSTPOLL 0 /* initial poll */ #define MINPOLL 5 /* minimum poll interval. std ntpd uses 6 (6: 64 sec) */ -/* If we got largish offset from a peer, cap next query interval - * for this peer by this many seconds: - */ -#define BIGOFF_INTERVAL (1 << 6) -/* If offset > discipline_jitter * POLLADJ_GATE, and poll interval is >= 2^BIGPOLL, - * then it is decreased _at once_. (If < 2^BIGPOLL, it will be decreased _eventually_). +/* + * If offset > discipline_jitter * POLLADJ_GATE, and poll interval is > 2^BIGPOLL, + * then it is decreased _at once_. (If <= 2^BIGPOLL, it will be decreased _eventually_). */ -#define BIGPOLL 10 /* 2^10 sec ~= 17 min */ +#define BIGPOLL 9 /* 2^9 sec ~= 8.5 min */ #define MAXPOLL 12 /* maximum poll interval (12: 1.1h, 17: 36.4h). std ntpd uses 17 */ -/* Actively lower poll when we see such big offsets. - * With STEP_THRESHOLD = 0.125, it means we try to sync more aggressively - * if offset increases over ~0.04 sec */ -#define POLLDOWN_OFFSET (STEP_THRESHOLD / 3) +/* + * Actively lower poll when we see such big offsets. + * With SLEW_THRESHOLD = 0.125, it means we try to sync more aggressively + * if offset increases over ~0.04 sec + */ +//#define POLLDOWN_OFFSET (SLEW_THRESHOLD / 3) #define MINDISP 0.01 /* minimum dispersion (sec) */ #define MAXDISP 16 /* maximum dispersion (sec) */ #define MAXSTRAT 16 /* maximum stratum (infinity metric) */ @@ -245,6 +305,9 @@ typedef struct { * or when receive times out (if p_fd >= 0): */ double next_action_time; double p_xmttime; + double p_raw_delay; + /* p_raw_delay is set even by "high delay" packets */ + /* lastpkt_delay isn't */ double lastpkt_recv_time; double lastpkt_delay; double lastpkt_rootdelay; @@ -256,6 +319,7 @@ typedef struct { datapoint_t filter_datapoint[NUM_DATAPOINTS]; /* last sent packet: */ msg_t p_xmt_msg; + char p_hostname[1]; } peer_t; @@ -273,6 +337,7 @@ enum { OPT_p = (1 << 5), OPT_S = (1 << 6), OPT_l = (1 << 7) * ENABLE_FEATURE_NTPD_SERVER, + OPT_I = (1 << 8) * ENABLE_FEATURE_NTPD_SERVER, /* We hijack some bits for other purposes */ OPT_qq = (1 << 31), }; @@ -291,6 +356,7 @@ struct globals { llist_t *ntp_peers; #if ENABLE_FEATURE_NTPD_SERVER int listen_fd; + char *if_name; # define G_listen_fd (G.listen_fd) #else # define G_listen_fd (-1) @@ -342,8 +408,6 @@ struct globals { */ #define G_precision_sec 0.002 uint8_t stratum; - /* Bool. After set to 1, never goes back to 0: */ - smallint initial_poll_complete; #define STATE_NSET 0 /* initial state, "nothing is set" */ //#define STATE_FSET 1 /* frequency set from file */ @@ -372,8 +436,6 @@ struct globals { }; #define G (*ptr_to_globals) -static const int const_IPTOS_LOWDELAY = IPTOS_LOWDELAY; - #define VERB1 if (MAX_VERBOSE && G.verbose) #define VERB2 if (MAX_VERBOSE >= 2 && G.verbose >= 2) @@ -689,11 +751,11 @@ static void reset_peer_stats(peer_t *p, double offset) { int i; - bool small_ofs = fabs(offset) < 16 * STEP_THRESHOLD; + bool small_ofs = fabs(offset) < STEP_THRESHOLD; /* Used to set p->filter_datapoint[i].d_dispersion = MAXDISP * and clear reachable bits, but this proved to be too agressive: - * after step (tested with suspinding laptop for ~30 secs), + * after step (tested with suspending laptop for ~30 secs), * this caused all previous data to be considered invalid, * making us needing to collect full ~8 datapoins per peer * after step in order to start trusting them. @@ -729,18 +791,51 @@ reset_peer_stats(peer_t *p, double offset) VERB6 bb_error_msg("%s->lastpkt_recv_time=%f", p->p_dotted, p->lastpkt_recv_time); } +static len_and_sockaddr* +resolve_peer_hostname(peer_t *p) +{ + len_and_sockaddr *lsa = host2sockaddr(p->p_hostname, 123); + if (lsa) { + free(p->p_lsa); + free(p->p_dotted); + p->p_lsa = lsa; + p->p_dotted = xmalloc_sockaddr2dotted_noport(&lsa->u.sa); + } else { + /* error message is emitted by host2sockaddr() */ + set_next(p, HOSTNAME_INTERVAL); + } + return lsa; +} + static void -add_peers(char *s) +add_peers(const char *s) { + llist_t *item; peer_t *p; - p = xzalloc(sizeof(*p)); - p->p_lsa = xhost2sockaddr(s, 123); - p->p_dotted = xmalloc_sockaddr2dotted_noport(&p->p_lsa->u.sa); + p = xzalloc(sizeof(*p) + strlen(s)); + strcpy(p->p_hostname, s); p->p_fd = -1; p->p_xmt_msg.m_status = MODE_CLIENT | (NTP_VERSION << 3); p->next_action_time = G.cur_time; /* = set_next(p, 0); */ - reset_peer_stats(p, 16 * STEP_THRESHOLD); + reset_peer_stats(p, STEP_THRESHOLD); + + /* Names like N..pool.ntp.org are randomly resolved + * to a pool of machines. Sometimes different N's resolve to the same IP. + * It is not useful to have two peers with same IP. We skip duplicates. + */ + if (resolve_peer_hostname(p)) { + for (item = G.ntp_peers; item != NULL; item = item->link) { + peer_t *pp = (peer_t *) item->data; + if (pp->p_dotted && strcmp(p->p_dotted, pp->p_dotted) == 0) { + bb_error_msg("duplicate peer %s (%s)", s, p->p_dotted); + free(p->p_lsa); + free(p->p_dotted); + free(p); + return; + } + } + } llist_add_to(&G.ntp_peers, p); G.peer_cnt++; @@ -769,6 +864,11 @@ do_sendto(int fd, static void send_query_to_peer(peer_t *p) { + if (!p->p_lsa) { + if (!resolve_peer_hostname(p)) + return; + } + /* Why do we need to bind()? * See what happens when we don't bind: * @@ -804,7 +904,7 @@ send_query_to_peer(peer_t *p) #if ENABLE_FEATURE_IPV6 if (family == AF_INET) #endif - setsockopt(fd, IPPROTO_IP, IP_TOS, &const_IPTOS_LOWDELAY, sizeof(const_IPTOS_LOWDELAY)); + setsockopt_int(fd, IPPROTO_IP, IP_TOS, IPTOS_LOWDELAY); free(local_lsa); } @@ -827,8 +927,8 @@ send_query_to_peer(peer_t *p) * * Save the real transmit timestamp locally. */ - p->p_xmt_msg.m_xmttime.int_partl = random(); - p->p_xmt_msg.m_xmttime.fractionl = random(); + p->p_xmt_msg.m_xmttime.int_partl = rand(); + p->p_xmt_msg.m_xmttime.fractionl = rand(); p->p_xmttime = gettime1900d(); /* Were doing it only if sendto worked, but @@ -959,6 +1059,16 @@ step_time(double offset) } } +static void clamp_pollexp_and_set_MAXSTRAT(void) +{ + if (G.poll_exp < MINPOLL) + G.poll_exp = MINPOLL; + if (G.poll_exp > BIGPOLL) + G.poll_exp = BIGPOLL; + G.polladj_count = 0; + G.stratum = MAXSTRAT; +} + /* * Selection and clustering, and their helpers @@ -1040,7 +1150,7 @@ select_and_cluster(void) num_points = 0; item = G.ntp_peers; - if (G.initial_poll_complete) while (item != NULL) { + while (item != NULL) { double rd, offset; p = (peer_t *) item->data; @@ -1440,9 +1550,7 @@ update_local_clock(peer_t *p) exit(0); } - G.polladj_count = 0; - G.poll_exp = MINPOLL; - G.stratum = MAXSTRAT; + clamp_pollexp_and_set_MAXSTRAT(); run_script("step", offset); @@ -1456,14 +1564,12 @@ update_local_clock(peer_t *p) #endif abs_offset = offset = 0; set_new_values(STATE_SYNC, offset, recv_time); - } else { /* abs_offset <= STEP_THRESHOLD */ - if (G.poll_exp < MINPOLL && G.initial_poll_complete) { - VERB4 bb_error_msg("small offset:%+f, disabling burst mode", offset); - G.polladj_count = 0; - G.poll_exp = MINPOLL; - } + /* The ratio is calculated before jitter is updated to make + * poll adjust code more sensitive to large offsets. + */ + G.offset_to_jitter_ratio = abs_offset / G.discipline_jitter; /* Compute the clock jitter as the RMS of exponentially * weighted offset differences. Used by the poll adjust code. @@ -1471,6 +1577,8 @@ update_local_clock(peer_t *p) etemp = SQUARE(G.discipline_jitter); dtemp = SQUARE(offset - G.last_update_offset); G.discipline_jitter = SQRT(etemp + (dtemp - etemp) / AVG); + if (G.discipline_jitter < G_precision_sec) + G.discipline_jitter = G_precision_sec; switch (G.discipline_state) { case STATE_NSET: @@ -1547,10 +1655,6 @@ update_local_clock(peer_t *p) } } - if (G.discipline_jitter < G_precision_sec) - G.discipline_jitter = G_precision_sec; - G.offset_to_jitter_ratio = abs_offset / G.discipline_jitter; - G.reftime = G.cur_time; G.ntp_status = p->lastpkt_status; G.refid = p->lastpkt_refid; @@ -1603,14 +1707,7 @@ update_local_clock(peer_t *p) tmx.freq = G.discipline_freq_drift * 65536e6; #endif tmx.modes = ADJ_OFFSET | ADJ_STATUS | ADJ_TIMECONST;// | ADJ_MAXERROR | ADJ_ESTERROR; - tmx.offset = (offset * 1000000); /* usec */ - tmx.status = STA_PLL; - if (G.ntp_status & LI_PLUSSEC) - tmx.status |= STA_INS; - if (G.ntp_status & LI_MINUSSEC) - tmx.status |= STA_DEL; - - tmx.constant = G.poll_exp - 4; + tmx.constant = (int)G.poll_exp - 4; /* EXPERIMENTAL. * The below if statement should be unnecessary, but... * It looks like Linux kernel's PLL is far too gentle in changing @@ -1621,8 +1718,27 @@ update_local_clock(peer_t *p) * To be on a safe side, let's do it only if offset is significantly * larger than jitter. */ - if (tmx.constant > 0 && G.offset_to_jitter_ratio >= TIMECONST_HACK_GATE) + if (G.offset_to_jitter_ratio >= TIMECONST_HACK_GATE) tmx.constant--; + tmx.offset = (long)(offset * 1000000); /* usec */ + if (SLEW_THRESHOLD < STEP_THRESHOLD) { + if (tmx.offset > (long)(SLEW_THRESHOLD * 1000000)) { + tmx.offset = (long)(SLEW_THRESHOLD * 1000000); + tmx.constant--; + } + if (tmx.offset < -(long)(SLEW_THRESHOLD * 1000000)) { + tmx.offset = -(long)(SLEW_THRESHOLD * 1000000); + tmx.constant--; + } + } + if (tmx.constant < 0) + tmx.constant = 0; + + tmx.status = STA_PLL; + if (G.ntp_status & LI_PLUSSEC) + tmx.status |= STA_INS; + if (G.ntp_status & LI_MINUSSEC) + tmx.status |= STA_DEL; //tmx.esterror = (uint32_t)(clock_jitter * 1e6); //tmx.maxerror = (uint32_t)((sys_rootdelay / 2 + sys_rootdisp) * 1e6); @@ -1635,8 +1751,14 @@ update_local_clock(peer_t *p) VERB4 bb_error_msg("adjtimex:%d freq:%ld offset:%+ld status:0x%x", rc, tmx.freq, tmx.offset, tmx.status); G.kernel_freq_drift = tmx.freq / 65536; - VERB2 bb_error_msg("update from:%s offset:%+f jitter:%f clock drift:%+.3fppm tc:%d", - p->p_dotted, offset, G.discipline_jitter, (double)tmx.freq / 65536, (int)tmx.constant); + VERB2 bb_error_msg("update from:%s offset:%+f delay:%f jitter:%f clock drift:%+.3fppm tc:%d", + p->p_dotted, + offset, + p->lastpkt_delay, + G.discipline_jitter, + (double)tmx.freq / 65536, + (int)tmx.constant + ); return 1; /* "ok to increase poll interval" */ } @@ -1647,28 +1769,52 @@ update_local_clock(peer_t *p) * (helpers first) */ static unsigned -retry_interval(void) +poll_interval(int upper_bound) { - /* Local problem, want to retry soon */ - unsigned interval, r; - interval = RETRY_INTERVAL; - r = random(); - interval += r % (unsigned)(RETRY_INTERVAL / 4); - VERB4 bb_error_msg("chose retry interval:%u", interval); + unsigned interval, r, mask; + interval = 1 << G.poll_exp; + if (interval > upper_bound) + interval = upper_bound; + mask = ((interval-1) >> 4) | 1; + r = rand(); + interval += r & mask; /* ~ random(0..1) * interval/16 */ + VERB4 bb_error_msg("chose poll interval:%u (poll_exp:%d)", interval, G.poll_exp); return interval; } -static unsigned -poll_interval(int exponent) +static void +adjust_poll(int count) { - unsigned interval, r; - exponent = G.poll_exp + exponent; - if (exponent < 0) - exponent = 0; - interval = 1 << exponent; - r = random(); - interval += ((r & (interval-1)) >> 4) + ((r >> 8) & 1); /* + 1/16 of interval, max */ - VERB4 bb_error_msg("chose poll interval:%u (poll_exp:%d exp:%d)", interval, G.poll_exp, exponent); - return interval; + G.polladj_count += count; + if (G.polladj_count > POLLADJ_LIMIT) { + G.polladj_count = 0; + if (G.poll_exp < MAXPOLL) { + G.poll_exp++; + VERB4 bb_error_msg("polladj: discipline_jitter:%f ++poll_exp=%d", + G.discipline_jitter, G.poll_exp); + } + } else if (G.polladj_count < -POLLADJ_LIMIT || (count < 0 && G.poll_exp > BIGPOLL)) { + G.polladj_count = 0; + if (G.poll_exp > MINPOLL) { + llist_t *item; + + G.poll_exp--; + /* Correct p->next_action_time in each peer + * which waits for sending, so that they send earlier. + * Old pp->next_action_time are on the order + * of t + (1 << old_poll_exp) + small_random, + * we simply need to subtract ~half of that. + */ + for (item = G.ntp_peers; item != NULL; item = item->link) { + peer_t *pp = (peer_t *) item->data; + if (pp->p_fd < 0) + pp->next_action_time -= (1 << G.poll_exp); + } + VERB4 bb_error_msg("polladj: discipline_jitter:%f --poll_exp=%d", + G.discipline_jitter, G.poll_exp); + } + } else { + VERB4 bb_error_msg("polladj: count:%d", G.polladj_count); + } } static NOINLINE void recv_and_process_peer_pkt(peer_t *p) @@ -1677,7 +1823,8 @@ recv_and_process_peer_pkt(peer_t *p) ssize_t size; msg_t msg; double T1, T2, T3, T4; - double dv, offset; + double offset; + double prev_delay, delay; unsigned interval; datapoint_t *datapoint; peer_t *q; @@ -1688,19 +1835,23 @@ recv_and_process_peer_pkt(peer_t *p) * ntp servers reply from their *other IP*. * TODO: maybe we should check at least what we can: from.port == 123? */ + recv_again: size = recv(p->p_fd, &msg, sizeof(msg), MSG_DONTWAIT); - if (size == -1) { - bb_perror_msg("recv(%s) error", p->p_dotted); - if (errno == EHOSTUNREACH || errno == EHOSTDOWN - || errno == ENETUNREACH || errno == ENETDOWN - || errno == ECONNREFUSED || errno == EADDRNOTAVAIL - || errno == EAGAIN - ) { -//TODO: always do this? - interval = retry_interval(); - goto set_next_and_ret; - } - xfunc_die(); + if (size < 0) { + if (errno == EINTR) + /* Signal caught */ + goto recv_again; + if (errno == EAGAIN) + /* There was no packet after all + * (poll() returning POLLIN for a fd + * is not a ironclad guarantee that data is there) + */ + return; + /* + * If you need a different handling for a specific + * errno, always explain it in comment. + */ + bb_perror_msg_and_die("recv(%s) error", p->p_dotted); } if (size != NTP_MSGSIZE_NOAUTH && size != NTP_MSGSIZE) { @@ -1726,10 +1877,15 @@ recv_and_process_peer_pkt(peer_t *p) || msg.m_stratum == 0 || msg.m_stratum > NTP_MAXSTRATUM ) { -// TODO: stratum 0 responses may have commands in 32-bit m_refid field: -// "DENY", "RSTR" - peer does not like us at all -// "RATE" - peer is overloaded, reduce polling freq bb_error_msg("reply from %s: peer is unsynced", p->p_dotted); + /* + * Stratum 0 responses may have commands in 32-bit m_refid field: + * "DENY", "RSTR" - peer does not like us at all, + * "RATE" - peer is overloaded, reduce polling freq. + * If poll interval is small, increase it. + */ + if (G.poll_exp < BIGPOLL) + goto increase_interval; goto pick_normal_interval; } @@ -1737,12 +1893,6 @@ recv_and_process_peer_pkt(peer_t *p) // if (msg.m_rootdelay / 2 + msg.m_rootdisp >= MAXDISP || p->lastpkt_reftime > msg.m_xmt) // return; /* invalid header values */ - p->lastpkt_status = msg.m_status; - p->lastpkt_stratum = msg.m_stratum; - p->lastpkt_rootdelay = sfp_to_d(msg.m_rootdelay); - p->lastpkt_rootdisp = sfp_to_d(msg.m_rootdisp); - p->lastpkt_refid = msg.m_refid; - /* * From RFC 2030 (with a correction to the delay math): * @@ -1762,28 +1912,35 @@ recv_and_process_peer_pkt(peer_t *p) T3 = lfp_to_d(msg.m_xmttime); T4 = G.cur_time; - p->lastpkt_recv_time = T4; - VERB6 bb_error_msg("%s->lastpkt_recv_time=%f", p->p_dotted, p->lastpkt_recv_time); - /* The delay calculation is a special case. In cases where the * server and client clocks are running at different rates and * with very fast networks, the delay can appear negative. In * order to avoid violating the Principle of Least Astonishment, * the delay is clamped not less than the system precision. */ - dv = p->lastpkt_delay; - p->lastpkt_delay = (T4 - T1) - (T3 - T2); - if (p->lastpkt_delay < G_precision_sec) - p->lastpkt_delay = G_precision_sec; + delay = (T4 - T1) - (T3 - T2); + if (delay < G_precision_sec) + delay = G_precision_sec; /* * If this packet's delay is much bigger than the last one, * it's better to just ignore it than use its much less precise value. */ - if (p->reachable_bits && p->lastpkt_delay > dv * BAD_DELAY_GROWTH) { - bb_error_msg("reply from %s: delay %f is too high, ignoring", p->p_dotted, p->lastpkt_delay); + prev_delay = p->p_raw_delay; + p->p_raw_delay = delay; + if (p->reachable_bits && delay > prev_delay * BAD_DELAY_GROWTH) { + bb_error_msg("reply from %s: delay %f is too high, ignoring", p->p_dotted, delay); goto pick_normal_interval; } + p->lastpkt_delay = delay; + p->lastpkt_recv_time = T4; + VERB6 bb_error_msg("%s->lastpkt_recv_time=%f", p->p_dotted, p->lastpkt_recv_time); + p->lastpkt_status = msg.m_status; + p->lastpkt_stratum = msg.m_stratum; + p->lastpkt_rootdelay = sfp_to_d(msg.m_rootdelay); + p->lastpkt_rootdisp = sfp_to_d(msg.m_rootdisp); + p->lastpkt_refid = msg.m_refid; + p->datapoint_idx = p->reachable_bits ? (p->datapoint_idx + 1) % NUM_DATAPOINTS : 0; datapoint = &p->filter_datapoint[p->datapoint_idx]; datapoint->d_recv_time = T4; @@ -1817,21 +1974,36 @@ recv_and_process_peer_pkt(peer_t *p) /* Muck with statictics and update the clock */ filter_datapoints(p); q = select_and_cluster(); - rc = -1; + rc = 0; if (q) { - rc = 0; if (!(option_mask32 & OPT_w)) { rc = update_local_clock(q); +#if 0 +//Disabled this because there is a case where largish offsets +//are unavoidable: if network round-trip delay is, say, ~0.6s, +//error in offset estimation would be ~delay/2 ~= 0.3s. +//Thus, offsets will be usually in -0.3...0.3s range. +//In this case, this code would keep poll interval small, +//but it won't be helping. +//BIGOFF check below deals with a case of seeing multi-second offsets. + /* If drift is dangerously large, immediately * drop poll interval one step down. */ if (fabs(q->filter_offset) >= POLLDOWN_OFFSET) { VERB4 bb_error_msg("offset:%+f > POLLDOWN_OFFSET", q->filter_offset); - goto poll_down; + adjust_poll(-POLLADJ_LIMIT * 3); + rc = 0; } +#endif } + } else { + /* No peer selected. + * If poll interval is small, increase it. + */ + if (G.poll_exp < BIGPOLL) + goto increase_interval; } - /* else: no peer selected, rc = -1: we want to poll more often */ if (rc != 0) { /* Adjust the poll interval by comparing the current offset @@ -1843,51 +2015,21 @@ recv_and_process_peer_pkt(peer_t *p) if (rc > 0 && G.offset_to_jitter_ratio <= POLLADJ_GATE) { /* was += G.poll_exp but it is a bit * too optimistic for my taste at high poll_exp's */ - G.polladj_count += MINPOLL; - if (G.polladj_count > POLLADJ_LIMIT) { - G.polladj_count = 0; - if (G.poll_exp < MAXPOLL) { - G.poll_exp++; - VERB4 bb_error_msg("polladj: discipline_jitter:%f ++poll_exp=%d", - G.discipline_jitter, G.poll_exp); - } - } else { - VERB4 bb_error_msg("polladj: incr:%d", G.polladj_count); - } + increase_interval: + adjust_poll(MINPOLL); } else { - G.polladj_count -= G.poll_exp * 2; - if (G.polladj_count < -POLLADJ_LIMIT || G.poll_exp >= BIGPOLL) { - poll_down: - G.polladj_count = 0; - if (G.poll_exp > MINPOLL) { - llist_t *item; - - G.poll_exp--; - /* Correct p->next_action_time in each peer - * which waits for sending, so that they send earlier. - * Old pp->next_action_time are on the order - * of t + (1 << old_poll_exp) + small_random, - * we simply need to subtract ~half of that. - */ - for (item = G.ntp_peers; item != NULL; item = item->link) { - peer_t *pp = (peer_t *) item->data; - if (pp->p_fd < 0) - pp->next_action_time -= (1 << G.poll_exp); - } - VERB4 bb_error_msg("polladj: discipline_jitter:%f --poll_exp=%d", - G.discipline_jitter, G.poll_exp); - } - } else { - VERB4 bb_error_msg("polladj: decr:%d", G.polladj_count); - } + VERB3 if (rc > 0) + bb_error_msg("want smaller interval: offset/jitter = %u", + G.offset_to_jitter_ratio); + adjust_poll(-G.poll_exp * 2); } } /* Decide when to send new query for this peer */ pick_normal_interval: - interval = poll_interval(0); - if (fabs(offset) >= STEP_THRESHOLD * 8 && interval > BIGOFF_INTERVAL) { - /* If we are synced, offsets are less than STEP_THRESHOLD, + interval = poll_interval(INT_MAX); + if (fabs(offset) >= BIGOFF && interval > BIGOFF_INTERVAL) { + /* If we are synced, offsets are less than SLEW_THRESHOLD, * or at the very least not much larger than it. * Now we see a largish one. * Either this peer is feeling bad, or packet got corrupted, @@ -1901,7 +2043,6 @@ recv_and_process_peer_pkt(peer_t *p) interval = BIGOFF_INTERVAL; } - set_next_and_ret: set_next(p, interval); } @@ -1934,6 +2075,13 @@ recv_and_process_client_pkt(void /*int fd*/) goto bail; } + /* Respond only to client and symmetric active packets */ + if ((msg.m_status & MODE_MASK) != MODE_CLIENT + && (msg.m_status & MODE_MASK) != MODE_SYM_ACT + ) { + goto bail; + } + query_status = msg.m_status; query_xmttime = msg.m_xmttime; @@ -2066,12 +2214,13 @@ static NOINLINE void ntp_init(char **argv) unsigned opts; llist_t *peers; - srandom(getpid()); + srand(getpid()); if (getuid()) bb_error_msg_and_die(bb_msg_you_must_be_root); /* Set some globals */ + G.discipline_jitter = G_precision_sec; G.stratum = MAXSTRAT; if (BURSTPOLL != 0) G.poll_exp = BURSTPOLL; /* speeds up initial sync */ @@ -2079,40 +2228,72 @@ static NOINLINE void ntp_init(char **argv) /* Parse options */ peers = NULL; - opt_complementary = "dd:p::wn"; /* d: counter; p: list; -w implies -n */ + opt_complementary = "dd:wn" /* -d: counter; -p: list; -w implies -n */ + IF_FEATURE_NTPD_SERVER(":Il"); /* -I implies -l */ opts = getopt32(argv, "nqNx" /* compat */ - "wp:S:"IF_FEATURE_NTPD_SERVER("l") /* NOT compat */ + "wp:*S:"IF_FEATURE_NTPD_SERVER("l") /* NOT compat */ + IF_FEATURE_NTPD_SERVER("I:") /* compat */ "d" /* compat */ "46aAbgL", /* compat, ignored */ - &peers, &G.script_name, &G.verbose); - if (!(opts & (OPT_p|OPT_l))) - bb_show_usage(); + &peers, &G.script_name, +#if ENABLE_FEATURE_NTPD_SERVER + &G.if_name, +#endif + &G.verbose); + // if (opts & OPT_x) /* disable stepping, only slew is allowed */ // G.time_was_stepped = 1; - if (peers) { - while (peers) - add_peers(llist_pop(&peers)); - } else { - /* -l but no peers: "stratum 1 server" mode */ - G.stratum = 1; - } - if (!(opts & OPT_n)) { - bb_daemonize_or_rexec(DAEMON_DEVNULL_STDIO, argv); - logmode = LOGMODE_NONE; - } + #if ENABLE_FEATURE_NTPD_SERVER G_listen_fd = -1; if (opts & OPT_l) { G_listen_fd = create_and_bind_dgram_or_die(NULL, 123); + if (G.if_name) { + if (setsockopt_bindtodevice(G_listen_fd, G.if_name)) + xfunc_die(); + } socket_want_pktinfo(G_listen_fd); - setsockopt(G_listen_fd, IPPROTO_IP, IP_TOS, &const_IPTOS_LOWDELAY, sizeof(const_IPTOS_LOWDELAY)); + setsockopt_int(G_listen_fd, IPPROTO_IP, IP_TOS, IPTOS_LOWDELAY); } #endif /* I hesitate to set -20 prio. -15 should be high enough for timekeeping */ if (opts & OPT_N) setpriority(PRIO_PROCESS, 0, -15); + if (!(opts & OPT_n)) { + bb_daemonize_or_rexec(DAEMON_DEVNULL_STDIO, argv); + logmode = LOGMODE_NONE; + } + + if (peers) { + while (peers) + add_peers(llist_pop(&peers)); + } +#if ENABLE_FEATURE_NTPD_CONF + else { + parser_t *parser; + char *token[3]; + + parser = config_open("/etc/ntp.conf"); + while (config_read(parser, token, 3, 1, "# \t", PARSE_NORMAL)) { + if (strcmp(token[0], "server") == 0 && token[1]) { + add_peers(token[1]); + continue; + } + bb_error_msg("skipping %s:%u: unimplemented command '%s'", + "/etc/ntp.conf", parser->lineno, token[0] + ); + } + config_close(parser); + } +#endif + if (G.peer_cnt == 0) { + if (!(opts & OPT_l)) + bb_show_usage(); + /* -l but no peers: "stratum 1 server" mode */ + G.stratum = 1; + } /* If network is up, syncronization occurs in ~10 seconds. * We give "ntpd -q" 10 seconds to get first reply, * then another 50 seconds to finish syncing. @@ -2196,16 +2377,26 @@ int ntpd_main(int argc UNUSED_PARAM, char **argv) if (p->p_fd == -1) { /* Time to send new req */ if (--cnt == 0) { - G.initial_poll_complete = 1; + VERB4 bb_error_msg("disabling burst mode"); + G.polladj_count = 0; + G.poll_exp = MINPOLL; } send_query_to_peer(p); } else { /* Timed out waiting for reply */ close(p->p_fd); p->p_fd = -1; - timeout = poll_interval(-2); /* -2: try a bit sooner */ + /* If poll interval is small, increase it */ + if (G.poll_exp < BIGPOLL) + adjust_poll(MINPOLL); + timeout = poll_interval(NOREPLY_INTERVAL); bb_error_msg("timed out waiting for %s, reach 0x%02x, next query in %us", p->p_dotted, p->reachable_bits, timeout); + + /* What if don't see it because it changed its IP? */ + if (p->reachable_bits == 0) + resolve_peer_hostname(p); + set_next(p, timeout); } } @@ -2294,9 +2485,7 @@ int ntpd_main(int argc UNUSED_PARAM, char **argv) goto have_reachable_peer; } /* No peer responded for last 8 packets, panic */ - G.polladj_count = 0; - G.poll_exp = MINPOLL; - G.stratum = MAXSTRAT; + clamp_pollexp_and_set_MAXSTRAT(); run_script("unsync", 0.0); have_reachable_peer: ; }