blob: 3e7fc47199667b73cc4199531ee761d8f5dc8bfa
1 | /* |
2 | * NTP client/server, based on OpenNTPD 3.9p1 |
3 | * |
4 | * Author: Adam Tkac <vonsch@gmail.com> |
5 | * |
6 | * Licensed under GPLv2, see file LICENSE in this source tree. |
7 | */ |
8 | #include "libbb.h" |
9 | #include <netinet/ip.h> /* For IPTOS_LOWDELAY definition */ |
10 | #include <sys/resource.h> /* setpriority */ |
11 | #ifndef IPTOS_LOWDELAY |
12 | # define IPTOS_LOWDELAY 0x10 |
13 | #endif |
14 | #ifndef IP_PKTINFO |
15 | # error "Sorry, your kernel has to support IP_PKTINFO" |
16 | #endif |
17 | |
18 | |
19 | /* Sync to peers every N secs */ |
20 | #define INTERVAL_QUERY_NORMAL 30 |
21 | #define INTERVAL_QUERY_PATHETIC 60 |
22 | #define INTERVAL_QUERY_AGRESSIVE 5 |
23 | |
24 | /* Bad if *less than* TRUSTLEVEL_BADPEER */ |
25 | #define TRUSTLEVEL_BADPEER 6 |
26 | #define TRUSTLEVEL_PATHETIC 2 |
27 | #define TRUSTLEVEL_AGRESSIVE 8 |
28 | #define TRUSTLEVEL_MAX 10 |
29 | |
30 | #define QSCALE_OFF_MIN 0.05 |
31 | #define QSCALE_OFF_MAX 0.50 |
32 | |
33 | /* Single query might take N secs max */ |
34 | #define QUERYTIME_MAX 15 |
35 | /* Min offset for settime at start. "man ntpd" says it's 128 ms */ |
36 | #define STEPTIME_MIN_OFFSET 0.128 |
37 | |
38 | typedef struct { |
39 | uint32_t int_partl; |
40 | uint32_t fractionl; |
41 | } l_fixedpt_t; |
42 | |
43 | typedef struct { |
44 | uint16_t int_parts; |
45 | uint16_t fractions; |
46 | } s_fixedpt_t; |
47 | |
48 | enum { |
49 | NTP_DIGESTSIZE = 16, |
50 | NTP_MSGSIZE_NOAUTH = 48, |
51 | NTP_MSGSIZE = (NTP_MSGSIZE_NOAUTH + 4 + NTP_DIGESTSIZE), |
52 | }; |
53 | |
54 | typedef struct { |
55 | uint8_t m_status; /* status of local clock and leap info */ |
56 | uint8_t m_stratum; /* stratum level */ |
57 | uint8_t m_ppoll; /* poll value */ |
58 | int8_t m_precision_exp; |
59 | s_fixedpt_t m_rootdelay; |
60 | s_fixedpt_t m_dispersion; |
61 | uint32_t m_refid; |
62 | l_fixedpt_t m_reftime; |
63 | l_fixedpt_t m_orgtime; |
64 | l_fixedpt_t m_rectime; |
65 | l_fixedpt_t m_xmttime; |
66 | uint32_t m_keyid; |
67 | uint8_t m_digest[NTP_DIGESTSIZE]; |
68 | } msg_t; |
69 | |
70 | enum { |
71 | NTP_VERSION = 4, |
72 | NTP_MAXSTRATUM = 15, |
73 | |
74 | /* Status Masks */ |
75 | MODE_MASK = (7 << 0), |
76 | VERSION_MASK = (7 << 3), |
77 | VERSION_SHIFT = 3, |
78 | LI_MASK = (3 << 6), |
79 | |
80 | /* Leap Second Codes (high order two bits of m_status) */ |
81 | LI_NOWARNING = (0 << 6), /* no warning */ |
82 | LI_PLUSSEC = (1 << 6), /* add a second (61 seconds) */ |
83 | LI_MINUSSEC = (2 << 6), /* minus a second (59 seconds) */ |
84 | LI_ALARM = (3 << 6), /* alarm condition */ |
85 | |
86 | /* Mode values */ |
87 | MODE_RES0 = 0, /* reserved */ |
88 | MODE_SYM_ACT = 1, /* symmetric active */ |
89 | MODE_SYM_PAS = 2, /* symmetric passive */ |
90 | MODE_CLIENT = 3, /* client */ |
91 | MODE_SERVER = 4, /* server */ |
92 | MODE_BROADCAST = 5, /* broadcast */ |
93 | MODE_RES1 = 6, /* reserved for NTP control message */ |
94 | MODE_RES2 = 7, /* reserved for private use */ |
95 | }; |
96 | |
97 | #define OFFSET_1900_1970 2208988800UL /* 1970 - 1900 in seconds */ |
98 | |
99 | typedef struct { |
100 | double d_offset; |
101 | double d_delay; |
102 | //UNUSED: double d_error; |
103 | time_t d_rcv_time; |
104 | uint32_t d_refid4; |
105 | uint8_t d_leap; |
106 | uint8_t d_stratum; |
107 | uint8_t d_good; |
108 | } datapoint_t; |
109 | |
110 | #define NUM_DATAPOINTS 8 |
111 | typedef struct { |
112 | len_and_sockaddr *p_lsa; |
113 | char *p_dotted; |
114 | /* When to send new query (if p_fd == -1) |
115 | * or when receive times out (if p_fd >= 0): */ |
116 | time_t next_action_time; |
117 | int p_fd; |
118 | uint8_t p_datapoint_idx; |
119 | uint8_t p_trustlevel; |
120 | double p_xmttime; |
121 | datapoint_t update; |
122 | datapoint_t p_datapoint[NUM_DATAPOINTS]; |
123 | msg_t p_xmt_msg; |
124 | } peer_t; |
125 | |
126 | enum { |
127 | OPT_n = (1 << 0), |
128 | OPT_q = (1 << 1), |
129 | OPT_N = (1 << 2), |
130 | OPT_x = (1 << 3), |
131 | /* Insert new options above this line. */ |
132 | /* Non-compat options: */ |
133 | OPT_p = (1 << 4), |
134 | OPT_l = (1 << 5) * ENABLE_FEATURE_NTPD_SERVER, |
135 | }; |
136 | |
137 | |
138 | struct globals { |
139 | /* total round trip delay to currently selected reference clock */ |
140 | double rootdelay; |
141 | /* reference timestamp: time when the system clock was last set or corrected */ |
142 | double reftime; |
143 | llist_t *ntp_peers; |
144 | #if ENABLE_FEATURE_NTPD_SERVER |
145 | int listen_fd; |
146 | #endif |
147 | unsigned verbose; |
148 | unsigned peer_cnt; |
149 | unsigned scale; |
150 | uint32_t refid; |
151 | uint32_t refid4; |
152 | uint8_t synced; |
153 | uint8_t leap; |
154 | #define G_precision_exp -6 |
155 | // int8_t precision_exp; |
156 | uint8_t stratum; |
157 | uint8_t time_was_stepped; |
158 | uint8_t first_adj_done; |
159 | }; |
160 | #define G (*ptr_to_globals) |
161 | |
162 | static const int const_IPTOS_LOWDELAY = IPTOS_LOWDELAY; |
163 | |
164 | |
165 | static void |
166 | set_next(peer_t *p, unsigned t) |
167 | { |
168 | p->next_action_time = time(NULL) + t; |
169 | } |
170 | |
171 | static void |
172 | add_peers(char *s) |
173 | { |
174 | peer_t *p; |
175 | |
176 | p = xzalloc(sizeof(*p)); |
177 | p->p_lsa = xhost2sockaddr(s, 123); |
178 | p->p_dotted = xmalloc_sockaddr2dotted_noport(&p->p_lsa->u.sa); |
179 | p->p_fd = -1; |
180 | p->p_xmt_msg.m_status = MODE_CLIENT | (NTP_VERSION << 3); |
181 | p->p_trustlevel = TRUSTLEVEL_PATHETIC; |
182 | p->next_action_time = time(NULL); /* = set_next(p, 0); */ |
183 | |
184 | llist_add_to(&G.ntp_peers, p); |
185 | G.peer_cnt++; |
186 | } |
187 | |
188 | static double |
189 | gettime1900d(void) |
190 | { |
191 | struct timeval tv; |
192 | gettimeofday(&tv, NULL); /* never fails */ |
193 | return (tv.tv_sec + 1.0e-6 * tv.tv_usec + OFFSET_1900_1970); |
194 | } |
195 | |
196 | static void |
197 | d_to_tv(double d, struct timeval *tv) |
198 | { |
199 | tv->tv_sec = (long)d; |
200 | tv->tv_usec = (d - tv->tv_sec) * 1000000; |
201 | } |
202 | |
203 | static double |
204 | lfp_to_d(l_fixedpt_t lfp) |
205 | { |
206 | double ret; |
207 | lfp.int_partl = ntohl(lfp.int_partl); |
208 | lfp.fractionl = ntohl(lfp.fractionl); |
209 | ret = (double)lfp.int_partl + ((double)lfp.fractionl / UINT_MAX); |
210 | return ret; |
211 | } |
212 | |
213 | #if 0 //UNUSED |
214 | static double |
215 | sfp_to_d(s_fixedpt_t sfp) |
216 | { |
217 | double ret; |
218 | sfp.int_parts = ntohs(sfp.int_parts); |
219 | sfp.fractions = ntohs(sfp.fractions); |
220 | ret = (double)sfp.int_parts + ((double)sfp.fractions / USHRT_MAX); |
221 | return ret; |
222 | } |
223 | #endif |
224 | |
225 | #if ENABLE_FEATURE_NTPD_SERVER |
226 | static l_fixedpt_t |
227 | d_to_lfp(double d) |
228 | { |
229 | l_fixedpt_t lfp; |
230 | lfp.int_partl = (uint32_t)d; |
231 | lfp.fractionl = (uint32_t)((d - lfp.int_partl) * UINT_MAX); |
232 | lfp.int_partl = htonl(lfp.int_partl); |
233 | lfp.fractionl = htonl(lfp.fractionl); |
234 | return lfp; |
235 | } |
236 | |
237 | static s_fixedpt_t |
238 | d_to_sfp(double d) |
239 | { |
240 | s_fixedpt_t sfp; |
241 | sfp.int_parts = (uint16_t)d; |
242 | sfp.fractions = (uint16_t)((d - sfp.int_parts) * USHRT_MAX); |
243 | sfp.int_parts = htons(sfp.int_parts); |
244 | sfp.fractions = htons(sfp.fractions); |
245 | return sfp; |
246 | } |
247 | #endif |
248 | |
249 | static unsigned |
250 | error_interval(void) |
251 | { |
252 | unsigned interval, r; |
253 | interval = INTERVAL_QUERY_PATHETIC * QSCALE_OFF_MAX / QSCALE_OFF_MIN; |
254 | r = (unsigned)random() % (unsigned)(interval / 10); |
255 | return (interval + r); |
256 | } |
257 | |
258 | static int |
259 | do_sendto(int fd, |
260 | const struct sockaddr *from, const struct sockaddr *to, socklen_t addrlen, |
261 | msg_t *msg, ssize_t len) |
262 | { |
263 | ssize_t ret; |
264 | |
265 | errno = 0; |
266 | if (!from) { |
267 | ret = sendto(fd, msg, len, MSG_DONTWAIT, to, addrlen); |
268 | } else { |
269 | ret = send_to_from(fd, msg, len, MSG_DONTWAIT, to, from, addrlen); |
270 | } |
271 | if (ret != len) { |
272 | bb_perror_msg("send failed"); |
273 | return -1; |
274 | } |
275 | return 0; |
276 | } |
277 | |
278 | static int |
279 | send_query_to_peer(peer_t *p) |
280 | { |
281 | // Why do we need to bind()? |
282 | // See what happens when we don't bind: |
283 | // |
284 | // socket(PF_INET, SOCK_DGRAM, IPPROTO_IP) = 3 |
285 | // setsockopt(3, SOL_IP, IP_TOS, [16], 4) = 0 |
286 | // gettimeofday({1259071266, 327885}, NULL) = 0 |
287 | // sendto(3, "xxx", 48, MSG_DONTWAIT, {sa_family=AF_INET, sin_port=htons(123), sin_addr=inet_addr("10.34.32.125")}, 16) = 48 |
288 | // ^^^ we sent it from some source port picked by kernel. |
289 | // time(NULL) = 1259071266 |
290 | // write(2, "ntpd: entering poll 15 secs\n", 28) = 28 |
291 | // poll([{fd=3, events=POLLIN}], 1, 15000) = 1 ([{fd=3, revents=POLLIN}]) |
292 | // recv(3, "yyy", 68, MSG_DONTWAIT) = 48 |
293 | // ^^^ this recv will receive packets to any local port! |
294 | // |
295 | // Uncomment this and use strace to see it in action: |
296 | #define PROBE_LOCAL_ADDR // { len_and_sockaddr lsa; lsa.len = LSA_SIZEOF_SA; getsockname(p->query.fd, &lsa.u.sa, &lsa.len); } |
297 | |
298 | if (p->p_fd == -1) { |
299 | int fd, family; |
300 | len_and_sockaddr *local_lsa; |
301 | |
302 | family = p->p_lsa->u.sa.sa_family; |
303 | p->p_fd = fd = xsocket_type(&local_lsa, family, SOCK_DGRAM); |
304 | /* local_lsa has "null" address and port 0 now. |
305 | * bind() ensures we have a *particular port* selected by kernel |
306 | * and remembered in p->p_fd, thus later recv(p->p_fd) |
307 | * receives only packets sent to this port. |
308 | */ |
309 | PROBE_LOCAL_ADDR |
310 | xbind(fd, &local_lsa->u.sa, local_lsa->len); |
311 | PROBE_LOCAL_ADDR |
312 | #if ENABLE_FEATURE_IPV6 |
313 | if (family == AF_INET) |
314 | #endif |
315 | setsockopt(fd, IPPROTO_IP, IP_TOS, &const_IPTOS_LOWDELAY, sizeof(const_IPTOS_LOWDELAY)); |
316 | free(local_lsa); |
317 | } |
318 | |
319 | /* |
320 | * Send out a random 64-bit number as our transmit time. The NTP |
321 | * server will copy said number into the originate field on the |
322 | * response that it sends us. This is totally legal per the SNTP spec. |
323 | * |
324 | * The impact of this is two fold: we no longer send out the current |
325 | * system time for the world to see (which may aid an attacker), and |
326 | * it gives us a (not very secure) way of knowing that we're not |
327 | * getting spoofed by an attacker that can't capture our traffic |
328 | * but can spoof packets from the NTP server we're communicating with. |
329 | * |
330 | * Save the real transmit timestamp locally. |
331 | */ |
332 | p->p_xmt_msg.m_xmttime.int_partl = random(); |
333 | p->p_xmt_msg.m_xmttime.fractionl = random(); |
334 | p->p_xmttime = gettime1900d(); |
335 | |
336 | if (do_sendto(p->p_fd, /*from:*/ NULL, /*to:*/ &p->p_lsa->u.sa, /*addrlen:*/ p->p_lsa->len, |
337 | &p->p_xmt_msg, NTP_MSGSIZE_NOAUTH) == -1 |
338 | ) { |
339 | close(p->p_fd); |
340 | p->p_fd = -1; |
341 | set_next(p, INTERVAL_QUERY_PATHETIC); |
342 | return -1; |
343 | } |
344 | |
345 | if (G.verbose) |
346 | bb_error_msg("sent query to %s", p->p_dotted); |
347 | set_next(p, QUERYTIME_MAX); |
348 | |
349 | return 0; |
350 | } |
351 | |
352 | |
353 | /* Time is stepped only once, when the first packet from a peer is received. |
354 | */ |
355 | static void |
356 | step_time_once(double offset) |
357 | { |
358 | double dtime; |
359 | llist_t *item; |
360 | struct timeval tv; |
361 | char buf[80]; |
362 | time_t tval; |
363 | |
364 | if (G.time_was_stepped) |
365 | goto bail; |
366 | G.time_was_stepped = 1; |
367 | |
368 | /* if the offset is small, don't step, slew (later) */ |
369 | if (offset < STEPTIME_MIN_OFFSET && offset > -STEPTIME_MIN_OFFSET) |
370 | goto bail; |
371 | |
372 | gettimeofday(&tv, NULL); /* never fails */ |
373 | dtime = offset + tv.tv_sec; |
374 | dtime += 1.0e-6 * tv.tv_usec; |
375 | d_to_tv(dtime, &tv); |
376 | |
377 | if (settimeofday(&tv, NULL) == -1) |
378 | bb_perror_msg_and_die("settimeofday"); |
379 | |
380 | tval = tv.tv_sec; |
381 | strftime_YYYYMMDDHHMMSS(buf, sizeof(buf), &tval); |
382 | |
383 | bb_error_msg("setting clock to %s (offset %fs)", buf, offset); |
384 | |
385 | for (item = G.ntp_peers; item != NULL; item = item->link) { |
386 | peer_t *p = (peer_t *) item->data; |
387 | p->next_action_time -= (time_t)offset; |
388 | } |
389 | |
390 | bail: |
391 | if (option_mask32 & OPT_q) |
392 | exit(0); |
393 | } |
394 | |
395 | |
396 | /* Time is periodically slewed when we collect enough |
397 | * good data points. |
398 | */ |
399 | static int |
400 | compare_offsets(const void *aa, const void *bb) |
401 | { |
402 | const peer_t *const *a = aa; |
403 | const peer_t *const *b = bb; |
404 | if ((*a)->update.d_offset < (*b)->update.d_offset) |
405 | return -1; |
406 | return ((*a)->update.d_offset > (*b)->update.d_offset); |
407 | } |
408 | static unsigned |
409 | updated_scale(double offset) |
410 | { |
411 | if (offset < 0) |
412 | offset = -offset; |
413 | if (offset > QSCALE_OFF_MAX) |
414 | return 1; |
415 | if (offset < QSCALE_OFF_MIN) |
416 | return QSCALE_OFF_MAX / QSCALE_OFF_MIN; |
417 | return QSCALE_OFF_MAX / offset; |
418 | } |
419 | static void |
420 | slew_time(void) |
421 | { |
422 | llist_t *item; |
423 | double offset_median; |
424 | struct timeval tv; |
425 | |
426 | { |
427 | peer_t **peers = xzalloc(sizeof(peers[0]) * G.peer_cnt); |
428 | unsigned goodpeer_cnt = 0; |
429 | unsigned middle; |
430 | |
431 | for (item = G.ntp_peers; item != NULL; item = item->link) { |
432 | peer_t *p = (peer_t *) item->data; |
433 | if (p->p_trustlevel < TRUSTLEVEL_BADPEER) |
434 | continue; |
435 | if (!p->update.d_good) { |
436 | free(peers); |
437 | return; |
438 | } |
439 | peers[goodpeer_cnt++] = p; |
440 | } |
441 | |
442 | if (goodpeer_cnt == 0) { |
443 | free(peers); |
444 | goto clear_good; |
445 | } |
446 | |
447 | qsort(peers, goodpeer_cnt, sizeof(peers[0]), compare_offsets); |
448 | |
449 | middle = goodpeer_cnt / 2; |
450 | if (middle != 0 && (goodpeer_cnt & 1) == 0) { |
451 | offset_median = (peers[middle-1]->update.d_offset + peers[middle]->update.d_offset) / 2; |
452 | G.rootdelay = (peers[middle-1]->update.d_delay + peers[middle]->update.d_delay) / 2; |
453 | G.stratum = 1 + MAX(peers[middle-1]->update.d_stratum, peers[middle]->update.d_stratum); |
454 | } else { |
455 | offset_median = peers[middle]->update.d_offset; |
456 | G.rootdelay = peers[middle]->update.d_delay; |
457 | G.stratum = 1 + peers[middle]->update.d_stratum; |
458 | } |
459 | G.leap = peers[middle]->update.d_leap; |
460 | G.refid4 = peers[middle]->update.d_refid4; |
461 | G.refid = |
462 | #if ENABLE_FEATURE_IPV6 |
463 | peers[middle]->p_lsa->u.sa.sa_family != AF_INET ? |
464 | G.refid4 : |
465 | #endif |
466 | peers[middle]->p_lsa->u.sin.sin_addr.s_addr; |
467 | free(peers); |
468 | } |
469 | //TODO: if (offset_median > BIG) step_time(offset_median)? |
470 | |
471 | G.scale = updated_scale(offset_median); |
472 | |
473 | bb_error_msg("adjusting clock by %fs, our stratum is %u, time scale %u", |
474 | offset_median, G.stratum, G.scale); |
475 | |
476 | errno = 0; |
477 | d_to_tv(offset_median, &tv); |
478 | if (adjtime(&tv, &tv) == -1) |
479 | bb_perror_msg_and_die("adjtime failed"); |
480 | if (G.verbose >= 2) |
481 | bb_error_msg("old adjust: %d.%06u", (int)tv.tv_sec, (unsigned)tv.tv_usec); |
482 | |
483 | if (G.first_adj_done) { |
484 | uint8_t synced = (tv.tv_sec == 0 && tv.tv_usec == 0); |
485 | if (synced != G.synced) { |
486 | G.synced = synced; |
487 | bb_error_msg("clock is %ssynced", synced ? "" : "un"); |
488 | } |
489 | } |
490 | G.first_adj_done = 1; |
491 | |
492 | G.reftime = gettime1900d(); |
493 | |
494 | clear_good: |
495 | for (item = G.ntp_peers; item != NULL; item = item->link) { |
496 | peer_t *p = (peer_t *) item->data; |
497 | p->update.d_good = 0; |
498 | } |
499 | } |
500 | |
501 | static void |
502 | update_peer_data(peer_t *p) |
503 | { |
504 | /* Clock filter. |
505 | * Find the datapoint with the lowest delay. |
506 | * Use that as the peer update. |
507 | * Invalidate it and all older ones. |
508 | */ |
509 | int i; |
510 | int best = -1; |
511 | int good = 0; |
512 | |
513 | for (i = 0; i < NUM_DATAPOINTS; i++) { |
514 | if (p->p_datapoint[i].d_good) { |
515 | good++; |
516 | if (best < 0 || p->p_datapoint[i].d_delay < p->p_datapoint[best].d_delay) |
517 | best = i; |
518 | } |
519 | } |
520 | |
521 | if (good < 8) //FIXME: was it meant to be NUM_DATAPOINTS, not 8? |
522 | return; |
523 | |
524 | p->update = p->p_datapoint[best]; /* struct copy */ |
525 | slew_time(); |
526 | |
527 | for (i = 0; i < NUM_DATAPOINTS; i++) |
528 | if (p->p_datapoint[i].d_rcv_time <= p->p_datapoint[best].d_rcv_time) |
529 | p->p_datapoint[i].d_good = 0; |
530 | } |
531 | |
532 | static unsigned |
533 | scale_interval(unsigned requested) |
534 | { |
535 | unsigned interval, r; |
536 | interval = requested * G.scale; |
537 | r = (unsigned)random() % (unsigned)(MAX(5, interval / 10)); |
538 | return (interval + r); |
539 | } |
540 | static void |
541 | recv_and_process_peer_pkt(peer_t *p) |
542 | { |
543 | ssize_t size; |
544 | msg_t msg; |
545 | double T1, T2, T3, T4; |
546 | unsigned interval; |
547 | datapoint_t *datapoint; |
548 | |
549 | /* We can recvfrom here and check from.IP, but some multihomed |
550 | * ntp servers reply from their *other IP*. |
551 | * TODO: maybe we should check at least what we can: from.port == 123? |
552 | */ |
553 | size = recv(p->p_fd, &msg, sizeof(msg), MSG_DONTWAIT); |
554 | if (size == -1) { |
555 | bb_perror_msg("recv(%s) error", p->p_dotted); |
556 | if (errno == EHOSTUNREACH || errno == EHOSTDOWN |
557 | || errno == ENETUNREACH || errno == ENETDOWN |
558 | || errno == ECONNREFUSED || errno == EADDRNOTAVAIL |
559 | || errno == EAGAIN |
560 | ) { |
561 | //TODO: always do this? |
562 | set_next(p, error_interval()); |
563 | goto close_sock; |
564 | } |
565 | xfunc_die(); |
566 | } |
567 | |
568 | if (size != NTP_MSGSIZE_NOAUTH && size != NTP_MSGSIZE) { |
569 | bb_error_msg("malformed packet received from %s", p->p_dotted); |
570 | goto bail; |
571 | } |
572 | |
573 | if (msg.m_orgtime.int_partl != p->p_xmt_msg.m_xmttime.int_partl |
574 | || msg.m_orgtime.fractionl != p->p_xmt_msg.m_xmttime.fractionl |
575 | ) { |
576 | goto bail; |
577 | } |
578 | |
579 | if ((msg.m_status & LI_ALARM) == LI_ALARM |
580 | || msg.m_stratum == 0 |
581 | || msg.m_stratum > NTP_MAXSTRATUM |
582 | ) { |
583 | // TODO: stratum 0 responses may have commands in 32-bit m_refid field: |
584 | // "DENY", "RSTR" - peer does not like us at all |
585 | // "RATE" - peer is overloaded, reduce polling freq |
586 | interval = error_interval(); |
587 | bb_error_msg("reply from %s: not synced, next query in %us", p->p_dotted, interval); |
588 | goto close_sock; |
589 | } |
590 | |
591 | /* |
592 | * From RFC 2030 (with a correction to the delay math): |
593 | * |
594 | * Timestamp Name ID When Generated |
595 | * ------------------------------------------------------------ |
596 | * Originate Timestamp T1 time request sent by client |
597 | * Receive Timestamp T2 time request received by server |
598 | * Transmit Timestamp T3 time reply sent by server |
599 | * Destination Timestamp T4 time reply received by client |
600 | * |
601 | * The roundtrip delay and local clock offset are defined as |
602 | * |
603 | * delay = (T4 - T1) - (T3 - T2); offset = ((T2 - T1) + (T3 - T4)) / 2 |
604 | */ |
605 | T1 = p->p_xmttime; |
606 | T2 = lfp_to_d(msg.m_rectime); |
607 | T3 = lfp_to_d(msg.m_xmttime); |
608 | T4 = gettime1900d(); |
609 | |
610 | datapoint = &p->p_datapoint[p->p_datapoint_idx]; |
611 | |
612 | datapoint->d_offset = ((T2 - T1) + (T3 - T4)) / 2; |
613 | datapoint->d_delay = (T4 - T1) - (T3 - T2); |
614 | if (datapoint->d_delay < 0) { |
615 | bb_error_msg("reply from %s: negative delay %f", p->p_dotted, datapoint->d_delay); |
616 | interval = error_interval(); |
617 | set_next(p, interval); |
618 | goto close_sock; |
619 | } |
620 | //UNUSED: datapoint->d_error = (T2 - T1) - (T3 - T4); |
621 | datapoint->d_rcv_time = (time_t)(T4 - OFFSET_1900_1970); /* = time(NULL); */ |
622 | datapoint->d_good = 1; |
623 | |
624 | datapoint->d_leap = (msg.m_status & LI_MASK); |
625 | //UNUSED: datapoint->o_precision = msg.m_precision_exp; |
626 | //UNUSED: datapoint->o_rootdelay = sfp_to_d(msg.m_rootdelay); |
627 | //UNUSED: datapoint->o_rootdispersion = sfp_to_d(msg.m_dispersion); |
628 | //UNUSED: datapoint->d_refid = ntohl(msg.m_refid); |
629 | datapoint->d_refid4 = msg.m_xmttime.fractionl; |
630 | //UNUSED: datapoint->o_reftime = lfp_to_d(msg.m_reftime); |
631 | //UNUSED: datapoint->o_poll = msg.m_ppoll; |
632 | datapoint->d_stratum = msg.m_stratum; |
633 | |
634 | if (p->p_trustlevel < TRUSTLEVEL_PATHETIC) |
635 | interval = scale_interval(INTERVAL_QUERY_PATHETIC); |
636 | else if (p->p_trustlevel < TRUSTLEVEL_AGRESSIVE) |
637 | interval = scale_interval(INTERVAL_QUERY_AGRESSIVE); |
638 | else |
639 | interval = scale_interval(INTERVAL_QUERY_NORMAL); |
640 | |
641 | set_next(p, interval); |
642 | |
643 | /* Every received reply which we do not discard increases trust */ |
644 | if (p->p_trustlevel < TRUSTLEVEL_MAX) { |
645 | p->p_trustlevel++; |
646 | if (p->p_trustlevel == TRUSTLEVEL_BADPEER) |
647 | bb_error_msg("peer %s now valid", p->p_dotted); |
648 | } |
649 | |
650 | if (G.verbose) |
651 | bb_error_msg("reply from %s: offset %f delay %f, next query in %us", p->p_dotted, |
652 | datapoint->d_offset, datapoint->d_delay, interval); |
653 | |
654 | update_peer_data(p); |
655 | //TODO: do it after all peers had a chance to return at least one reply? |
656 | step_time_once(datapoint->d_offset); |
657 | |
658 | p->p_datapoint_idx++; |
659 | if (p->p_datapoint_idx >= NUM_DATAPOINTS) |
660 | p->p_datapoint_idx = 0; |
661 | |
662 | close_sock: |
663 | /* We do not expect any more packets from this peer for now. |
664 | * Closing the socket informs kernel about it. |
665 | * We open a new socket when we send a new query. |
666 | */ |
667 | close(p->p_fd); |
668 | p->p_fd = -1; |
669 | bail: |
670 | return; |
671 | } |
672 | |
673 | #if ENABLE_FEATURE_NTPD_SERVER |
674 | static void |
675 | recv_and_process_client_pkt(void /*int fd*/) |
676 | { |
677 | ssize_t size; |
678 | uint8_t version; |
679 | double rectime; |
680 | len_and_sockaddr *to; |
681 | struct sockaddr *from; |
682 | msg_t msg; |
683 | uint8_t query_status; |
684 | uint8_t query_ppoll; |
685 | l_fixedpt_t query_xmttime; |
686 | |
687 | to = get_sock_lsa(G.listen_fd); |
688 | from = xzalloc(to->len); |
689 | |
690 | size = recv_from_to(G.listen_fd, &msg, sizeof(msg), MSG_DONTWAIT, from, &to->u.sa, to->len); |
691 | if (size != NTP_MSGSIZE_NOAUTH && size != NTP_MSGSIZE) { |
692 | char *addr; |
693 | if (size < 0) { |
694 | if (errno == EAGAIN) |
695 | goto bail; |
696 | bb_perror_msg_and_die("recv"); |
697 | } |
698 | addr = xmalloc_sockaddr2dotted_noport(from); |
699 | bb_error_msg("malformed packet received from %s: size %u", addr, (int)size); |
700 | free(addr); |
701 | goto bail; |
702 | } |
703 | |
704 | query_status = msg.m_status; |
705 | query_ppoll = msg.m_ppoll; |
706 | query_xmttime = msg.m_xmttime; |
707 | |
708 | /* Build a reply packet */ |
709 | memset(&msg, 0, sizeof(msg)); |
710 | msg.m_status = G.synced ? G.leap : LI_ALARM; |
711 | msg.m_status |= (query_status & VERSION_MASK); |
712 | msg.m_status |= ((query_status & MODE_MASK) == MODE_CLIENT) ? |
713 | MODE_SERVER : MODE_SYM_PAS; |
714 | msg.m_stratum = G.stratum; |
715 | msg.m_ppoll = query_ppoll; |
716 | msg.m_precision_exp = G_precision_exp; |
717 | rectime = gettime1900d(); |
718 | msg.m_xmttime = msg.m_rectime = d_to_lfp(rectime); |
719 | msg.m_reftime = d_to_lfp(G.reftime); |
720 | //msg.m_xmttime = d_to_lfp(gettime1900d()); // = msg.m_rectime |
721 | msg.m_orgtime = query_xmttime; |
722 | msg.m_rootdelay = d_to_sfp(G.rootdelay); |
723 | version = (query_status & VERSION_MASK); /* ... >> VERSION_SHIFT - done below instead */ |
724 | msg.m_refid = (version > (3 << VERSION_SHIFT)) ? G.refid4 : G.refid; |
725 | |
726 | /* We reply from the local address packet was sent to, |
727 | * this makes to/from look swapped here: */ |
728 | do_sendto(G.listen_fd, |
729 | /*from:*/ &to->u.sa, /*to:*/ from, /*addrlen:*/ to->len, |
730 | &msg, size); |
731 | |
732 | bail: |
733 | free(to); |
734 | free(from); |
735 | } |
736 | #endif |
737 | |
738 | /* Upstream ntpd's options: |
739 | * |
740 | * -4 Force DNS resolution of host names to the IPv4 namespace. |
741 | * -6 Force DNS resolution of host names to the IPv6 namespace. |
742 | * -a Require cryptographic authentication for broadcast client, |
743 | * multicast client and symmetric passive associations. |
744 | * This is the default. |
745 | * -A Do not require cryptographic authentication for broadcast client, |
746 | * multicast client and symmetric passive associations. |
747 | * This is almost never a good idea. |
748 | * -b Enable the client to synchronize to broadcast servers. |
749 | * -c conffile |
750 | * Specify the name and path of the configuration file, |
751 | * default /etc/ntp.conf |
752 | * -d Specify debugging mode. This option may occur more than once, |
753 | * with each occurrence indicating greater detail of display. |
754 | * -D level |
755 | * Specify debugging level directly. |
756 | * -f driftfile |
757 | * Specify the name and path of the frequency file. |
758 | * This is the same operation as the "driftfile FILE" |
759 | * configuration command. |
760 | * -g Normally, ntpd exits with a message to the system log |
761 | * if the offset exceeds the panic threshold, which is 1000 s |
762 | * by default. This option allows the time to be set to any value |
763 | * without restriction; however, this can happen only once. |
764 | * If the threshold is exceeded after that, ntpd will exit |
765 | * with a message to the system log. This option can be used |
766 | * with the -q and -x options. See the tinker command for other options. |
767 | * -i jaildir |
768 | * Chroot the server to the directory jaildir. This option also implies |
769 | * that the server attempts to drop root privileges at startup |
770 | * (otherwise, chroot gives very little additional security). |
771 | * You may need to also specify a -u option. |
772 | * -k keyfile |
773 | * Specify the name and path of the symmetric key file, |
774 | * default /etc/ntp/keys. This is the same operation |
775 | * as the "keys FILE" configuration command. |
776 | * -l logfile |
777 | * Specify the name and path of the log file. The default |
778 | * is the system log file. This is the same operation as |
779 | * the "logfile FILE" configuration command. |
780 | * -L Do not listen to virtual IPs. The default is to listen. |
781 | * -n Don't fork. |
782 | * -N To the extent permitted by the operating system, |
783 | * run the ntpd at the highest priority. |
784 | * -p pidfile |
785 | * Specify the name and path of the file used to record the ntpd |
786 | * process ID. This is the same operation as the "pidfile FILE" |
787 | * configuration command. |
788 | * -P priority |
789 | * To the extent permitted by the operating system, |
790 | * run the ntpd at the specified priority. |
791 | * -q Exit the ntpd just after the first time the clock is set. |
792 | * This behavior mimics that of the ntpdate program, which is |
793 | * to be retired. The -g and -x options can be used with this option. |
794 | * Note: The kernel time discipline is disabled with this option. |
795 | * -r broadcastdelay |
796 | * Specify the default propagation delay from the broadcast/multicast |
797 | * server to this client. This is necessary only if the delay |
798 | * cannot be computed automatically by the protocol. |
799 | * -s statsdir |
800 | * Specify the directory path for files created by the statistics |
801 | * facility. This is the same operation as the "statsdir DIR" |
802 | * configuration command. |
803 | * -t key |
804 | * Add a key number to the trusted key list. This option can occur |
805 | * more than once. |
806 | * -u user[:group] |
807 | * Specify a user, and optionally a group, to switch to. |
808 | * -v variable |
809 | * -V variable |
810 | * Add a system variable listed by default. |
811 | * -x Normally, the time is slewed if the offset is less than the step |
812 | * threshold, which is 128 ms by default, and stepped if above |
813 | * the threshold. This option sets the threshold to 600 s, which is |
814 | * well within the accuracy window to set the clock manually. |
815 | * Note: since the slew rate of typical Unix kernels is limited |
816 | * to 0.5 ms/s, each second of adjustment requires an amortization |
817 | * interval of 2000 s. Thus, an adjustment as much as 600 s |
818 | * will take almost 14 days to complete. This option can be used |
819 | * with the -g and -q options. See the tinker command for other options. |
820 | * Note: The kernel time discipline is disabled with this option. |
821 | */ |
822 | |
823 | /* By doing init in a separate function we decrease stack usage |
824 | * in main loop. |
825 | */ |
826 | static NOINLINE void ntp_init(char **argv) |
827 | { |
828 | unsigned opts; |
829 | llist_t *peers; |
830 | |
831 | srandom(getpid()); |
832 | |
833 | if (getuid()) |
834 | bb_error_msg_and_die(bb_msg_you_must_be_root); |
835 | |
836 | peers = NULL; |
837 | opt_complementary = "dd:p::"; /* d: counter, p: list */ |
838 | opts = getopt32(argv, |
839 | "nqNx" /* compat */ |
840 | "p:"IF_FEATURE_NTPD_SERVER("l") /* NOT compat */ |
841 | "d" /* compat */ |
842 | "46aAbgL", /* compat, ignored */ |
843 | &peers, &G.verbose); |
844 | if (!(opts & (OPT_p|OPT_l))) |
845 | bb_show_usage(); |
846 | if (opts & OPT_x) /* disable stepping, only slew is allowed */ |
847 | G.time_was_stepped = 1; |
848 | while (peers) |
849 | add_peers(llist_pop(&peers)); |
850 | if (!(opts & OPT_n)) { |
851 | bb_daemonize_or_rexec(DAEMON_DEVNULL_STDIO, argv); |
852 | logmode = LOGMODE_NONE; |
853 | } |
854 | #if ENABLE_FEATURE_NTPD_SERVER |
855 | G.listen_fd = -1; |
856 | if (opts & OPT_l) { |
857 | G.listen_fd = create_and_bind_dgram_or_die(NULL, 123); |
858 | socket_want_pktinfo(G.listen_fd); |
859 | setsockopt(G.listen_fd, IPPROTO_IP, IP_TOS, &const_IPTOS_LOWDELAY, sizeof(const_IPTOS_LOWDELAY)); |
860 | } |
861 | #endif |
862 | /* I hesitate to set -20 prio. -15 should be high enough for timekeeping */ |
863 | if (opts & OPT_N) |
864 | setpriority(PRIO_PROCESS, 0, -15); |
865 | |
866 | /* Set some globals */ |
867 | #if 0 |
868 | /* With constant b = 100, G.precision_exp is also constant -6. |
869 | * Uncomment this and you'll see */ |
870 | { |
871 | int prec = 0; |
872 | int b; |
873 | # if 0 |
874 | struct timespec tp; |
875 | /* We can use sys_clock_getres but assuming 10ms tick should be fine */ |
876 | clock_getres(CLOCK_REALTIME, &tp); |
877 | tp.tv_sec = 0; |
878 | tp.tv_nsec = 10000000; |
879 | b = 1000000000 / tp.tv_nsec; /* convert to Hz */ |
880 | # else |
881 | b = 100; /* b = 1000000000/10000000 = 100 */ |
882 | # endif |
883 | while (b > 1) |
884 | prec--, b >>= 1; |
885 | //G.precision_exp = prec; |
886 | bb_error_msg("G.precision_exp:%d", prec); /* -6 */ |
887 | } |
888 | #endif |
889 | G.scale = 1; |
890 | |
891 | bb_signals((1 << SIGTERM) | (1 << SIGINT), record_signo); |
892 | bb_signals((1 << SIGPIPE) | (1 << SIGHUP), SIG_IGN); |
893 | } |
894 | |
895 | int ntpd_main(int argc UNUSED_PARAM, char **argv) MAIN_EXTERNALLY_VISIBLE; |
896 | int ntpd_main(int argc UNUSED_PARAM, char **argv) |
897 | { |
898 | struct globals g; |
899 | struct pollfd *pfd; |
900 | peer_t **idx2peer; |
901 | |
902 | memset(&g, 0, sizeof(g)); |
903 | SET_PTR_TO_GLOBALS(&g); |
904 | |
905 | ntp_init(argv); |
906 | |
907 | { |
908 | /* if ENABLE_FEATURE_NTPD_SERVER, + 1 for listen_fd: */ |
909 | unsigned cnt = g.peer_cnt + ENABLE_FEATURE_NTPD_SERVER; |
910 | idx2peer = xzalloc(sizeof(idx2peer[0]) * cnt); |
911 | pfd = xzalloc(sizeof(pfd[0]) * cnt); |
912 | } |
913 | |
914 | while (!bb_got_signal) { |
915 | llist_t *item; |
916 | unsigned i, j; |
917 | unsigned sent_cnt, trial_cnt; |
918 | int nfds, timeout; |
919 | time_t cur_time, nextaction; |
920 | |
921 | /* Nothing between here and poll() blocks for any significant time */ |
922 | |
923 | cur_time = time(NULL); |
924 | nextaction = cur_time + 3600; |
925 | |
926 | i = 0; |
927 | #if ENABLE_FEATURE_NTPD_SERVER |
928 | if (g.listen_fd != -1) { |
929 | pfd[0].fd = g.listen_fd; |
930 | pfd[0].events = POLLIN; |
931 | i++; |
932 | } |
933 | #endif |
934 | /* Pass over peer list, send requests, time out on receives */ |
935 | sent_cnt = trial_cnt = 0; |
936 | for (item = g.ntp_peers; item != NULL; item = item->link) { |
937 | peer_t *p = (peer_t *) item->data; |
938 | |
939 | /* Overflow-safe "if (p->next_action_time <= cur_time) ..." */ |
940 | if ((int)(cur_time - p->next_action_time) >= 0) { |
941 | if (p->p_fd == -1) { |
942 | /* Time to send new req */ |
943 | trial_cnt++; |
944 | if (send_query_to_peer(p) == 0) |
945 | sent_cnt++; |
946 | } else { |
947 | /* Timed out waiting for reply */ |
948 | close(p->p_fd); |
949 | p->p_fd = -1; |
950 | timeout = error_interval(); |
951 | bb_error_msg("timed out waiting for %s, " |
952 | "next query in %us", p->p_dotted, timeout); |
953 | if (p->p_trustlevel >= TRUSTLEVEL_BADPEER) { |
954 | p->p_trustlevel /= 2; |
955 | if (p->p_trustlevel < TRUSTLEVEL_BADPEER) |
956 | bb_error_msg("peer %s now invalid", p->p_dotted); |
957 | } |
958 | set_next(p, timeout); |
959 | } |
960 | } |
961 | |
962 | if (p->next_action_time < nextaction) |
963 | nextaction = p->next_action_time; |
964 | |
965 | if (p->p_fd >= 0) { |
966 | /* Wait for reply from this peer */ |
967 | pfd[i].fd = p->p_fd; |
968 | pfd[i].events = POLLIN; |
969 | idx2peer[i] = p; |
970 | i++; |
971 | } |
972 | } |
973 | |
974 | if ((trial_cnt > 0 && sent_cnt == 0) || g.peer_cnt == 0) |
975 | step_time_once(0); /* no good peers, don't wait */ |
976 | |
977 | timeout = nextaction - cur_time; |
978 | if (timeout < 1) |
979 | timeout = 1; |
980 | |
981 | /* Here we may block */ |
982 | if (g.verbose >= 2) |
983 | bb_error_msg("poll %us, sockets:%u", timeout, i); |
984 | nfds = poll(pfd, i, timeout * 1000); |
985 | if (nfds <= 0) |
986 | continue; |
987 | |
988 | /* Process any received packets */ |
989 | j = 0; |
990 | #if ENABLE_FEATURE_NTPD_SERVER |
991 | if (g.listen_fd != -1) { |
992 | if (pfd[0].revents /* & (POLLIN|POLLERR)*/) { |
993 | nfds--; |
994 | recv_and_process_client_pkt(/*g.listen_fd*/); |
995 | } |
996 | j = 1; |
997 | } |
998 | #endif |
999 | for (; nfds != 0 && j < i; j++) { |
1000 | if (pfd[j].revents /* & (POLLIN|POLLERR)*/) { |
1001 | nfds--; |
1002 | recv_and_process_peer_pkt(idx2peer[j]); |
1003 | } |
1004 | } |
1005 | } /* while (!bb_got_signal) */ |
1006 | |
1007 | kill_myself_with_sig(bb_got_signal); |
1008 | } |
1009 |