plugins/check_ntp.c

   1 /*****************************************************************************
   2 *
   3 * Monitoring check_ntp plugin
   4 *
   5 * License: GPL
   6 * Copyright (c) 2006 Sean Finney <seanius@seanius.net>
   7 * Copyright (c) 2006-2024 Monitoring Plugins Development Team
   8 *
   9 * Description:
  10 *
  11 * This file contains the check_ntp plugin
  12 *
  13 * This plugin to check ntp servers independent of any commandline
  14 * programs or external libraries.
  15 *
  16 *
  17 * This program is free software: you can redistribute it and/or modify
  18 * it under the terms of the GNU General Public License as published by
  19 * the Free Software Foundation, either version 3 of the License, or
  20 * (at your option) any later version.
  21 *
  22 * This program is distributed in the hope that it will be useful,
  23 * but WITHOUT ANY WARRANTY; without even the implied warranty of
  24 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  25 * GNU General Public License for more details.
  26 *
  27 * You should have received a copy of the GNU General Public License
  28 * along with this program.  If not, see <http://www.gnu.org/licenses/>.
  29 *
  30 *
  31 *****************************************************************************/
  32
  33 const char *progname = "check_ntp";
  34 const char *copyright = "2006-2024";
  35 const char *email = "devel@monitoring-plugins.org";
  36
  37 #include "common.h"
  38 #include "netutils.h"
  39 #include "utils.h"
  40
  41 static char *server_address=NULL;
  42 static int verbose=0;
  43 static bool do_offset = false;
  44 static char *owarn="60";
  45 static char *ocrit="120";
  46 static bool do_jitter = false;
  47 static char *jwarn="5000";
  48 static char *jcrit="10000";
  49
  50 static int process_arguments (int /*argc*/, char ** /*argv*/);
  51 static thresholds *offset_thresholds = NULL;
  52 static thresholds *jitter_thresholds = NULL;
  53 static void print_help (void);
  54 void print_usage (void);
  55
  56 /* number of times to perform each request to get a good average. */
  57 #ifndef AVG_NUM
  58 #define AVG_NUM 4
  59 #endif
  60
  61 /* max size of control message data */
  62 #define MAX_CM_SIZE 468
  63
  64 /* this structure holds everything in an ntp request/response as per rfc1305 */
  65 typedef struct {
  66         uint8_t flags;       /* byte with leapindicator,vers,mode. see macros */
  67         uint8_t stratum;     /* clock stratum */
  68         int8_t poll;         /* polling interval */
  69         int8_t precision;    /* precision of the local clock */
  70         int32_t rtdelay;     /* total rt delay, as a fixed point num. see macros */
  71         uint32_t rtdisp;     /* like above, but for max err to primary src */
  72         uint32_t refid;      /* ref clock identifier */
  73         uint64_t refts;      /* reference timestamp.  local time local clock */
  74         uint64_t origts;     /* time at which request departed client */
  75         uint64_t rxts;       /* time at which request arrived at server */
  76         uint64_t txts;       /* time at which request departed server */
  77 } ntp_message;
  78
  79 /* this structure holds data about results from querying offset from a peer */
  80 typedef struct {
  81         time_t waiting;         /* ts set when we started waiting for a response */
  82         int num_responses;      /* number of successfully received responses */
  83         uint8_t stratum;        /* copied verbatim from the ntp_message */
  84         double rtdelay;         /* converted from the ntp_message */
  85         double rtdisp;          /* converted from the ntp_message */
  86         double offset[AVG_NUM]; /* offsets from each response */
  87         uint8_t flags;       /* byte with leapindicator,vers,mode. see macros */
  88 } ntp_server_results;
  89
  90 /* this structure holds everything in an ntp control message as per rfc1305 */
  91 typedef struct {
  92         uint8_t flags;       /* byte with leapindicator,vers,mode. see macros */
  93         uint8_t op;          /* R,E,M bits and Opcode */
  94         uint16_t seq;        /* Packet sequence */
  95         uint16_t status;     /* Clock status */
  96         uint16_t assoc;      /* Association */
  97         uint16_t offset;     /* Similar to TCP sequence # */
  98         uint16_t count;      /* # bytes of data */
  99         char data[MAX_CM_SIZE]; /* ASCII data of the request */
 100                                 /* NB: not necessarily NULL terminated! */
 101 } ntp_control_message;
 102
 103 /* this is an association/status-word pair found in control packet responses */
 104 typedef struct {
 105         uint16_t assoc;
 106         uint16_t status;
 107 } ntp_assoc_status_pair;
 108
 109 /* bits 1,2 are the leap indicator */
 110 #define LI_MASK 0xc0
 111 #define LI(x) ((x&LI_MASK)>>6)
 112 #define LI_SET(x,y) do{ x |= ((y<<6)&LI_MASK); }while(0)
 113 /* and these are the values of the leap indicator */
 114 #define LI_NOWARNING 0x00
 115 #define LI_EXTRASEC 0x01
 116 #define LI_MISSINGSEC 0x02
 117 #define LI_ALARM 0x03
 118 /* bits 3,4,5 are the ntp version */
 119 #define VN_MASK 0x38
 120 #define VN(x)   ((x&VN_MASK)>>3)
 121 #define VN_SET(x,y)     do{ x |= ((y<<3)&VN_MASK); }while(0)
 122 #define VN_RESERVED 0x02
 123 /* bits 6,7,8 are the ntp mode */
 124 #define MODE_MASK 0x07
 125 #define MODE(x) (x&MODE_MASK)
 126 #define MODE_SET(x,y)   do{ x |= (y&MODE_MASK); }while(0)
 127 /* here are some values */
 128 #define MODE_CLIENT 0x03
 129 #define MODE_CONTROLMSG 0x06
 130 /* In control message, bits 8-10 are R,E,M bits */
 131 #define REM_MASK 0xe0
 132 #define REM_RESP 0x80
 133 #define REM_ERROR 0x40
 134 #define REM_MORE 0x20
 135 /* In control message, bits 11 - 15 are opcode */
 136 #define OP_MASK 0x1f
 137 #define OP_SET(x,y)   do{ x |= (y&OP_MASK); }while(0)
 138 #define OP_READSTAT 0x01
 139 #define OP_READVAR  0x02
 140 /* In peer status bytes, bits 6,7,8 determine clock selection status */
 141 #define PEER_SEL(x) ((ntohs(x)>>8)&0x07)
 142 #define PEER_INCLUDED 0x04
 143 #define PEER_SYNCSOURCE 0x06
 144
 145 /**
 146  ** a note about the 32-bit "fixed point" numbers:
 147  **
 148  they are divided into halves, each being a 16-bit int in network byte order:
 149  - the first 16 bits are an int on the left side of a decimal point.
 150  - the second 16 bits represent a fraction n/(2^16)
 151  likewise for the 64-bit "fixed point" numbers with everything doubled :)
 152  **/
 153
 154 /* macros to access the left/right 16 bits of a 32-bit ntp "fixed point"
 155    number.  note that these can be used as lvalues too */
 156 #define L16(x) (((uint16_t*)&x)[0])
 157 #define R16(x) (((uint16_t*)&x)[1])
 158 /* macros to access the left/right 32 bits of a 64-bit ntp "fixed point"
 159    number.  these too can be used as lvalues */
 160 #define L32(x) (((uint32_t*)&x)[0])
 161 #define R32(x) (((uint32_t*)&x)[1])
 162
 163 /* ntp wants seconds since 1/1/00, epoch is 1/1/70.  this is the difference */
 164 #define EPOCHDIFF 0x83aa7e80UL
 165
 166 /* extract a 32-bit ntp fixed point number into a double */
 167 #define NTP32asDOUBLE(x) (ntohs(L16(x)) + (double)ntohs(R16(x))/65536.0)
 168
 169 /* likewise for a 64-bit ntp fp number */
 170 #define NTP64asDOUBLE(n) (double)(((uint64_t)n)?\
 171                          (ntohl(L32(n))-EPOCHDIFF) + \
 172                          (.00000001*(0.5+(double)(ntohl(R32(n))/42.94967296))):\
 173                          0)
 174
 175 /* convert a struct timeval to a double */
 176 #define TVasDOUBLE(x) (double)(x.tv_sec+(0.000001*x.tv_usec))
 177
 178 /* convert an ntp 64-bit fp number to a struct timeval */
 179 #define NTP64toTV(n,t) \
 180         do{ if(!n) t.tv_sec = t.tv_usec = 0; \
 181             else { \
 182                         t.tv_sec=ntohl(L32(n))-EPOCHDIFF; \
 183                         t.tv_usec=(int)(0.5+(double)(ntohl(R32(n))/4294.967296)); \
 184                 } \
 185         }while(0)
 186
 187 /* convert a struct timeval to an ntp 64-bit fp number */
 188 #define TVtoNTP64(t,n) \
 189         do{ if(!t.tv_usec && !t.tv_sec) n=0x0UL; \
 190                 else { \
 191                         L32(n)=htonl(t.tv_sec + EPOCHDIFF); \
 192                         R32(n)=htonl((uint64_t)((4294.967296*t.tv_usec)+.5)); \
 193                 } \
 194         } while(0)
 195
 196 /* NTP control message header is 12 bytes, plus any data in the data
 197  * field, plus null padding to the nearest 32-bit boundary per rfc.
 198  */
 199 #define SIZEOF_NTPCM(m) (12+ntohs(m.count)+((ntohs(m.count)%4)?4-(ntohs(m.count)%4):0))
 200
 201 /* finally, a little helper or two for debugging: */
 202 #define DBG(x) do{if(verbose>1){ x; }}while(0);
 203 #define PRINTSOCKADDR(x) \
 204         do{ \
 205                 printf("%u.%u.%u.%u", (x>>24)&0xff, (x>>16)&0xff, (x>>8)&0xff, x&0xff);\
 206         }while(0);
 207
 208 /* calculate the offset of the local clock */
 209 static inline double calc_offset(const ntp_message *m, const struct timeval *t){
 210         double client_tx, peer_rx, peer_tx, client_rx;
 211         client_tx = NTP64asDOUBLE(m->origts);
 212         peer_rx = NTP64asDOUBLE(m->rxts);
 213         peer_tx = NTP64asDOUBLE(m->txts);
 214         client_rx=TVasDOUBLE((*t));
 215         return (.5*((peer_tx-client_rx)+(peer_rx-client_tx)));
 216 }
 217
 218 /* print out a ntp packet in human readable/debuggable format */
 219 void print_ntp_message(const ntp_message *p){
 220         struct timeval ref, orig, rx, tx;
 221
 222         NTP64toTV(p->refts,ref);
 223         NTP64toTV(p->origts,orig);
 224         NTP64toTV(p->rxts,rx);
 225         NTP64toTV(p->txts,tx);
 226
 227         printf("packet contents:\n");
 228         printf("\tflags: 0x%.2x\n", p->flags);
 229         printf("\t  li=%d (0x%.2x)\n", LI(p->flags), p->flags&LI_MASK);
 230         printf("\t  vn=%d (0x%.2x)\n", VN(p->flags), p->flags&VN_MASK);
 231         printf("\t  mode=%d (0x%.2x)\n", MODE(p->flags), p->flags&MODE_MASK);
 232         printf("\tstratum = %d\n", p->stratum);
 233         printf("\tpoll = %g\n", pow(2, p->poll));
 234         printf("\tprecision = %g\n", pow(2, p->precision));
 235         printf("\trtdelay = %-.16g\n", NTP32asDOUBLE(p->rtdelay));
 236         printf("\trtdisp = %-.16g\n", NTP32asDOUBLE(p->rtdisp));
 237         printf("\trefid = %x\n", p->refid);
 238         printf("\trefts = %-.16g\n", NTP64asDOUBLE(p->refts));
 239         printf("\torigts = %-.16g\n", NTP64asDOUBLE(p->origts));
 240         printf("\trxts = %-.16g\n", NTP64asDOUBLE(p->rxts));
 241         printf("\ttxts = %-.16g\n", NTP64asDOUBLE(p->txts));
 242 }
 243
 244 void print_ntp_control_message(const ntp_control_message *p){
 245         int i=0, numpeers=0;
 246         const ntp_assoc_status_pair *peer=NULL;
 247
 248         printf("control packet contents:\n");
 249         printf("\tflags: 0x%.2x , 0x%.2x\n", p->flags, p->op);
 250         printf("\t  li=%d (0x%.2x)\n", LI(p->flags), p->flags&LI_MASK);
 251         printf("\t  vn=%d (0x%.2x)\n", VN(p->flags), p->flags&VN_MASK);
 252         printf("\t  mode=%d (0x%.2x)\n", MODE(p->flags), p->flags&MODE_MASK);
 253         printf("\t  response=%d (0x%.2x)\n", (p->op&REM_RESP)>0, p->op&REM_RESP);
 254         printf("\t  more=%d (0x%.2x)\n", (p->op&REM_MORE)>0, p->op&REM_MORE);
 255         printf("\t  error=%d (0x%.2x)\n", (p->op&REM_ERROR)>0, p->op&REM_ERROR);
 256         printf("\t  op=%d (0x%.2x)\n", p->op&OP_MASK, p->op&OP_MASK);
 257         printf("\tsequence: %d (0x%.2x)\n", ntohs(p->seq), ntohs(p->seq));
 258         printf("\tstatus: %d (0x%.2x)\n", ntohs(p->status), ntohs(p->status));
 259         printf("\tassoc: %d (0x%.2x)\n", ntohs(p->assoc), ntohs(p->assoc));
 260         printf("\toffset: %d (0x%.2x)\n", ntohs(p->offset), ntohs(p->offset));
 261         printf("\tcount: %d (0x%.2x)\n", ntohs(p->count), ntohs(p->count));
 262         numpeers=ntohs(p->count)/(sizeof(ntp_assoc_status_pair));
 263         if(p->op&REM_RESP && p->op&OP_READSTAT){
 264                 peer=(ntp_assoc_status_pair*)p->data;
 265                 for(i=0;i<numpeers;i++){
 266                         printf("\tpeer id %.2x status %.2x",
 267                                ntohs(peer[i].assoc), ntohs(peer[i].status));
 268                         if (PEER_SEL(peer[i].status) >= PEER_INCLUDED){
 269                                 if(PEER_SEL(peer[i].status) >= PEER_SYNCSOURCE){
 270                                         printf(" <-- current sync source");
 271                                 } else {
 272                                         printf(" <-- current sync candidate");
 273                                 }
 274                         }
 275                         printf("\n");
 276                 }
 277         }
 278 }
 279
 280 void setup_request(ntp_message *p){
 281         struct timeval t;
 282
 283         memset(p, 0, sizeof(ntp_message));
 284         LI_SET(p->flags, LI_ALARM);
 285         VN_SET(p->flags, 4);
 286         MODE_SET(p->flags, MODE_CLIENT);
 287         p->poll=4;
 288         p->precision=(int8_t)0xfa;
 289         L16(p->rtdelay)=htons(1);
 290         L16(p->rtdisp)=htons(1);
 291
 292         gettimeofday(&t, NULL);
 293         TVtoNTP64(t,p->txts);
 294 }
 295
 296 /* select the "best" server from a list of servers, and return its index.
 297  * this is done by filtering servers based on stratum, dispersion, and
 298  * finally round-trip delay. */
 299 int best_offset_server(const ntp_server_results *slist, int nservers){
 300         int cserver=0, best_server=-1;
 301
 302         /* for each server */
 303         for(cserver=0; cserver<nservers; cserver++){
 304                 /* We don't want any servers that fails these tests */
 305                 /* Sort out servers that didn't respond or responede with a 0 stratum;
 306                  * stratum 0 is for reference clocks so no NTP server should ever report
 307                  * a stratum 0 */
 308                 if ( slist[cserver].stratum == 0){
 309                         if (verbose) printf("discarding peer %d: stratum=%d\n", cserver, slist[cserver].stratum);
 310                         continue;
 311                 }
 312                 /* Sort out servers with error flags */
 313                 if ( LI(slist[cserver].flags) == LI_ALARM ){
 314                         if (verbose) printf("discarding peer %d: flags=%d\n", cserver, LI(slist[cserver].flags));
 315                         continue;
 316                 }
 317
 318                 /* If we don't have a server yet, use the first one */
 319                 if (best_server == -1) {
 320                         best_server = cserver;
 321                         DBG(printf("using peer %d as our first candidate\n", best_server));
 322                         continue;
 323                 }
 324
 325                 /* compare the server to the best one we've seen so far */
 326                 /* does it have an equal or better stratum? */
 327                 DBG(printf("comparing peer %d with peer %d\n", cserver, best_server));
 328                 if(slist[cserver].stratum <= slist[best_server].stratum){
 329                         DBG(printf("stratum for peer %d <= peer %d\n", cserver, best_server));
 330                         /* does it have an equal or better dispersion? */
 331                         if(slist[cserver].rtdisp <= slist[best_server].rtdisp){
 332                                 DBG(printf("dispersion for peer %d <= peer %d\n", cserver, best_server));
 333                                 /* does it have a better rtdelay? */
 334                                 if(slist[cserver].rtdelay < slist[best_server].rtdelay){
 335                                         DBG(printf("rtdelay for peer %d < peer %d\n", cserver, best_server));
 336                                         best_server = cserver;
 337                                         DBG(printf("peer %d is now our best candidate\n", best_server));
 338                                 }
 339                         }
 340                 }
 341         }
 342
 343         if(best_server >= 0) {
 344                 DBG(printf("best server selected: peer %d\n", best_server));
 345                 return best_server;
 346         } else {
 347                 DBG(printf("no peers meeting synchronization criteria :(\n"));
 348                 return -1;
 349         }
 350 }
 351
 352 /* do everything we need to get the total average offset
 353  * - we use a certain amount of parallelization with poll() to ensure
 354  *   we don't waste time sitting around waiting for single packets.
 355  * - we also "manually" handle resolving host names and connecting, because
 356  *   we have to do it in a way that our lazy macros don't handle currently :( */
 357 double offset_request(const char *host, int *status){
 358         int i=0, ga_result=0, num_hosts=0, *socklist=NULL, respnum=0;
 359         int servers_completed=0, one_read=0, servers_readable=0, best_index=-1;
 360         time_t now_time=0, start_ts=0;
 361         ntp_message *req=NULL;
 362         double avg_offset=0.;
 363         struct timeval recv_time;
 364         struct addrinfo *ai=NULL, *ai_tmp=NULL, hints;
 365         struct pollfd *ufds=NULL;
 366         ntp_server_results *servers=NULL;
 367
 368         /* setup hints to only return results from getaddrinfo that we'd like */
 369         memset(&hints, 0, sizeof(struct addrinfo));
 370         hints.ai_family = address_family;
 371         hints.ai_protocol = IPPROTO_UDP;
 372         hints.ai_socktype = SOCK_DGRAM;
 373
 374         /* fill in ai with the list of hosts resolved by the host name */
 375         ga_result = getaddrinfo(host, "123", &hints, &ai);
 376         if(ga_result!=0){
 377                 die(STATE_UNKNOWN, "error getting address for %s: %s\n",
 378                     host, gai_strerror(ga_result));
 379         }
 380
 381         /* count the number of returned hosts, and allocate stuff accordingly */
 382         for(ai_tmp=ai; ai_tmp!=NULL; ai_tmp=ai_tmp->ai_next){ num_hosts++; }
 383         req=(ntp_message*)malloc(sizeof(ntp_message)*num_hosts);
 384         if(req==NULL) die(STATE_UNKNOWN, "can not allocate ntp message array");
 385         socklist=(int*)malloc(sizeof(int)*num_hosts);
 386         if(socklist==NULL) die(STATE_UNKNOWN, "can not allocate socket array");
 387         ufds=(struct pollfd*)malloc(sizeof(struct pollfd)*num_hosts);
 388         if(ufds==NULL) die(STATE_UNKNOWN, "can not allocate socket array");
 389         servers=(ntp_server_results*)malloc(sizeof(ntp_server_results)*num_hosts);
 390         if(servers==NULL) die(STATE_UNKNOWN, "can not allocate server array");
 391         memset(servers, 0, sizeof(ntp_server_results)*num_hosts);
 392         DBG(printf("Found %d peers to check\n", num_hosts));
 393
 394         /* setup each socket for writing, and the corresponding struct pollfd */
 395         ai_tmp=ai;
 396         for(i=0;ai_tmp;i++){
 397                 socklist[i]=socket(ai_tmp->ai_family, SOCK_DGRAM, IPPROTO_UDP);
 398                 if(socklist[i] == -1) {
 399                         perror(NULL);
 400                         die(STATE_UNKNOWN, "can not create new socket");
 401                 }
 402                 if(connect(socklist[i], ai_tmp->ai_addr, ai_tmp->ai_addrlen)){
 403                         /* don't die here, because it is enough if there is one server
 404                            answering in time. This also would break for dual ipv4/6 stacked
 405                            ntp servers when the client only supports on of them.
 406                          */
 407                         DBG(printf("can't create socket connection on peer %i: %s\n", i, strerror(errno)));
 408                 } else {
 409                         ufds[i].fd=socklist[i];
 410                         ufds[i].events=POLLIN;
 411                         ufds[i].revents=0;
 412                 }
 413                 ai_tmp = ai_tmp->ai_next;
 414         }
 415
 416         /* now do AVG_NUM checks to each host.  we stop before timeout/2 seconds
 417          * have passed in order to ensure post-processing and jitter time. */
 418         now_time=start_ts=time(NULL);
 419         while(servers_completed<num_hosts && now_time-start_ts <= socket_timeout/2){
 420                 /* loop through each server and find each one which hasn't
 421                  * been touched in the past second or so and is still lacking
 422                  * some responses.  for each of these servers, send a new request,
 423                  * and update the "waiting" timestamp with the current time. */
 424                 now_time=time(NULL);
 425
 426                 for(i=0; i<num_hosts; i++){
 427                         if(servers[i].waiting<now_time && servers[i].num_responses<AVG_NUM){
 428                                 if(verbose && servers[i].waiting != 0) printf("re-");
 429                                 if(verbose) printf("sending request to peer %d\n", i);
 430                                 setup_request(&req[i]);
 431                                 write(socklist[i], &req[i], sizeof(ntp_message));
 432                                 servers[i].waiting=now_time;
 433                                 break;
 434                         }
 435                 }
 436
 437                 /* quickly poll for any sockets with pending data */
 438                 servers_readable=poll(ufds, num_hosts, 100);
 439                 if(servers_readable==-1){
 440                         perror("polling ntp sockets");
 441                         die(STATE_UNKNOWN, "communication errors");
 442                 }
 443
 444                 /* read from any sockets with pending data */
 445                 for(i=0; servers_readable && i<num_hosts; i++){
 446                         if(ufds[i].revents&POLLIN && servers[i].num_responses < AVG_NUM){
 447                                 if(verbose) {
 448                                         printf("response from peer %d: ", i);
 449                                 }
 450
 451                                 read(ufds[i].fd, &req[i], sizeof(ntp_message));
 452                                 gettimeofday(&recv_time, NULL);
 453                                 DBG(print_ntp_message(&req[i]));
 454                                 respnum=servers[i].num_responses++;
 455                                 servers[i].offset[respnum]=calc_offset(&req[i], &recv_time);
 456                                 if(verbose) {
 457                                         printf("offset %.10g\n", servers[i].offset[respnum]);
 458                                 }
 459                                 servers[i].stratum=req[i].stratum;
 460                                 servers[i].rtdisp=NTP32asDOUBLE(req[i].rtdisp);
 461                                 servers[i].rtdelay=NTP32asDOUBLE(req[i].rtdelay);
 462                                 servers[i].waiting=0;
 463                                 servers[i].flags=req[i].flags;
 464                                 servers_readable--;
 465                                 one_read = 1;
 466                                 if(servers[i].num_responses==AVG_NUM) servers_completed++;
 467                         }
 468                 }
 469                 /* lather, rinse, repeat. */
 470         }
 471
 472         if (one_read == 0) {
 473                 die(STATE_CRITICAL, "NTP CRITICAL: No response from NTP server\n");
 474         }
 475
 476         /* now, pick the best server from the list */
 477         best_index=best_offset_server(servers, num_hosts);
 478         if(best_index < 0){
 479                 *status=STATE_UNKNOWN;
 480         } else {
 481                 /* finally, calculate the average offset */
 482                 for(i=0; i<servers[best_index].num_responses;i++){
 483                         avg_offset+=servers[best_index].offset[i];
 484                 }
 485                 avg_offset/=servers[best_index].num_responses;
 486         }
 487
 488         /* cleanup */
 489         /* FIXME: Not closing the socket to avoid reuse of the local port
 490          * which can cause old NTP packets to be read instead of NTP control
 491          * packets in jitter_request(). THERE MUST BE ANOTHER WAY...
 492          * for(j=0; j<num_hosts; j++){ close(socklist[j]); } */
 493         free(socklist);
 494         free(ufds);
 495         free(servers);
 496         free(req);
 497         freeaddrinfo(ai);
 498
 499         if(verbose) printf("overall average offset: %.10g\n", avg_offset);
 500         return avg_offset;
 501 }
 502
 503 void
 504 setup_control_request(ntp_control_message *p, uint8_t opcode, uint16_t seq){
 505         memset(p, 0, sizeof(ntp_control_message));
 506         LI_SET(p->flags, LI_NOWARNING);
 507         VN_SET(p->flags, VN_RESERVED);
 508         MODE_SET(p->flags, MODE_CONTROLMSG);
 509         OP_SET(p->op, opcode);
 510         p->seq = htons(seq);
 511         /* Remaining fields are zero for requests */
 512 }
 513
 514 /* XXX handle responses with the error bit set */
 515 double jitter_request(int *status){
 516         int conn=-1, i, npeers=0, num_candidates=0;
 517         bool syncsource_found = false;
 518         int run=0, min_peer_sel=PEER_INCLUDED, num_selected=0, num_valid=0;
 519         int peers_size=0, peer_offset=0;
 520         ntp_assoc_status_pair *peers=NULL;
 521         ntp_control_message req;
 522         const char *getvar = "jitter";
 523         double rval = 0.0, jitter = -1.0;
 524         char *startofvalue=NULL, *nptr=NULL;
 525         void *tmp;
 526
 527         /* Long-winded explanation:
 528          * Getting the jitter requires a number of steps:
 529          * 1) Send a READSTAT request.
 530          * 2) Interpret the READSTAT reply
 531          *  a) The data section contains a list of peer identifiers (16 bits)
 532          *     and associated status words (16 bits)
 533          *  b) We want the value of 0x06 in the SEL (peer selection) value,
 534          *     which means "current synchronizatin source".  If that's missing,
 535          *     we take anything better than 0x04 (see the rfc for details) but
 536          *     set a minimum of warning.
 537          * 3) Send a READVAR request for information on each peer identified
 538          *    in 2b greater than the minimum selection value.
 539          * 4) Extract the jitter value from the data[] (it's ASCII)
 540          */
 541         my_udp_connect(server_address, 123, &conn);
 542
 543         /* keep sending requests until the server stops setting the
 544          * REM_MORE bit, though usually this is only 1 packet. */
 545         do{
 546                 setup_control_request(&req, OP_READSTAT, 1);
 547                 DBG(printf("sending READSTAT request"));
 548                 write(conn, &req, SIZEOF_NTPCM(req));
 549                 DBG(print_ntp_control_message(&req));
 550                 /* Attempt to read the largest size packet possible */
 551                 req.count=htons(MAX_CM_SIZE);
 552                 DBG(printf("receiving READSTAT response"))
 553                 read(conn, &req, SIZEOF_NTPCM(req));
 554                 DBG(print_ntp_control_message(&req));
 555                 /* Each peer identifier is 4 bytes in the data section, which
 556                  * we represent as a ntp_assoc_status_pair datatype.
 557                  */
 558                 peers_size+=ntohs(req.count);
 559                 if((tmp=realloc(peers, peers_size)) == NULL)
 560                         free(peers), die(STATE_UNKNOWN, "can not (re)allocate 'peers' buffer\n");
 561                 peers=tmp;
 562                 memcpy((void*)((ptrdiff_t)peers+peer_offset), (void*)req.data, ntohs(req.count));
 563                 npeers=peers_size/sizeof(ntp_assoc_status_pair);
 564                 peer_offset+=ntohs(req.count);
 565         } while(req.op&REM_MORE);
 566
 567         /* first, let's find out if we have a sync source, or if there are
 568          * at least some candidates.  in the case of the latter we'll issue
 569          * a warning but go ahead with the check on them. */
 570         for (i = 0; i < npeers; i++){
 571                 if (PEER_SEL(peers[i].status) >= PEER_INCLUDED){
 572                         num_candidates++;
 573                         if(PEER_SEL(peers[i].status) >= PEER_SYNCSOURCE){
 574                                 syncsource_found = true;
 575                                 min_peer_sel=PEER_SYNCSOURCE;
 576                         }
 577                 }
 578         }
 579         if(verbose) printf("%d candidate peers available\n", num_candidates);
 580         if(verbose && syncsource_found) printf("synchronization source found\n");
 581         if(! syncsource_found){
 582                 *status = STATE_UNKNOWN;
 583                 if(verbose) printf("warning: no synchronization source found\n");
 584         }
 585
 586
 587         for (run=0; run<AVG_NUM; run++){
 588                 if(verbose) printf("jitter run %d of %d\n", run+1, AVG_NUM);
 589                 for (i = 0; i < npeers; i++){
 590                         /* Only query this server if it is the current sync source */
 591                         if (PEER_SEL(peers[i].status) >= min_peer_sel){
 592                                 char jitter_data[MAX_CM_SIZE+1];
 593                                 size_t jitter_data_count;
 594
 595                                 num_selected++;
 596                                 setup_control_request(&req, OP_READVAR, 2);
 597                                 req.assoc = peers[i].assoc;
 598                                 /* By spec, putting the variable name "jitter"  in the request
 599                                  * should cause the server to provide _only_ the jitter value.
 600                                  * thus reducing net traffic, guaranteeing us only a single
 601                                  * datagram in reply, and making interpretation much simpler
 602                                  */
 603                                 /* Older servers doesn't know what jitter is, so if we get an
 604                                  * error on the first pass we redo it with "dispersion" */
 605                                 strncpy(req.data, getvar, MAX_CM_SIZE-1);
 606                                 req.count = htons(strlen(getvar));
 607                                 DBG(printf("sending READVAR request...\n"));
 608                                 write(conn, &req, SIZEOF_NTPCM(req));
 609                                 DBG(print_ntp_control_message(&req));
 610
 611                                 req.count = htons(MAX_CM_SIZE);
 612                                 DBG(printf("receiving READVAR response...\n"));
 613                                 read(conn, &req, SIZEOF_NTPCM(req));
 614                                 DBG(print_ntp_control_message(&req));
 615
 616                                 if(req.op&REM_ERROR && strstr(getvar, "jitter")) {
 617                                         if(verbose) printf("The 'jitter' command failed (old ntp server?)\nRestarting with 'dispersion'...\n");
 618                                         getvar = "dispersion";
 619                                         num_selected--;
 620                                         i--;
 621                                         continue;
 622                                 }
 623
 624                                 /* get to the float value */
 625                                 if(verbose) {
 626                                         printf("parsing jitter from peer %.2x: ", ntohs(peers[i].assoc));
 627                                 }
 628                                 if((jitter_data_count = ntohs(req.count)) >= sizeof(jitter_data)){
 629                                         die(STATE_UNKNOWN,
 630                                             _("jitter response too large (%lu bytes)\n"),
 631                                             (unsigned long)jitter_data_count);
 632                                 }
 633                                 memcpy(jitter_data, req.data, jitter_data_count);
 634                                 jitter_data[jitter_data_count] = '\0';
 635                                 startofvalue = strchr(jitter_data, '=');
 636                                 if(startofvalue != NULL) {
 637                                         startofvalue++;
 638                                         jitter = strtod(startofvalue, &nptr);
 639                                 }
 640                                 if(startofvalue == NULL || startofvalue==nptr){
 641                                         printf("warning: unable to read server jitter response.\n");
 642                                         *status = STATE_UNKNOWN;
 643                                 } else {
 644                                         if(verbose) printf("%g\n", jitter);
 645                                         num_valid++;
 646                                         rval += jitter;
 647                                 }
 648                         }
 649                 }
 650                 if(verbose){
 651                         printf("jitter parsed from %d/%d peers\n", num_valid, num_selected);
 652                 }
 653         }
 654
 655         rval = num_valid ? rval / num_valid : -1.0;
 656
 657         close(conn);
 658         if(peers!=NULL) free(peers);
 659         /* If we return -1.0, it means no synchronization source was found */
 660         return rval;
 661 }
 662
 663 int process_arguments(int argc, char **argv){
 664         int c;
 665         int option=0;
 666         static struct option longopts[] = {
 667                 {"version", no_argument, 0, 'V'},
 668                 {"help", no_argument, 0, 'h'},
 669                 {"verbose", no_argument, 0, 'v'},
 670                 {"use-ipv4", no_argument, 0, '4'},
 671                 {"use-ipv6", no_argument, 0, '6'},
 672                 {"warning", required_argument, 0, 'w'},
 673                 {"critical", required_argument, 0, 'c'},
 674                 {"jwarn", required_argument, 0, 'j'},
 675                 {"jcrit", required_argument, 0, 'k'},
 676                 {"timeout", required_argument, 0, 't'},
 677                 {"hostname", required_argument, 0, 'H'},
 678                 {0, 0, 0, 0}
 679         };
 680
 681
 682         if (argc < 2)
 683                 usage ("\n");
 684
 685         while (1) {
 686                 c = getopt_long (argc, argv, "Vhv46w:c:j:k:t:H:", longopts, &option);
 687                 if (c == -1 || c == EOF || c == 1)
 688                         break;
 689
 690                 switch (c) {
 691                 case 'h':
 692                         print_help();
 693                         exit(STATE_UNKNOWN);
 694                         break;
 695                 case 'V':
 696                         print_revision(progname, NP_VERSION);
 697                         exit(STATE_UNKNOWN);
 698                         break;
 699                 case 'v':
 700                         verbose++;
 701                         break;
 702                 case 'w':
 703                         do_offset = true;
 704                         owarn = optarg;
 705                         break;
 706                 case 'c':
 707                         do_offset = true;
 708                         ocrit = optarg;
 709                         break;
 710                 case 'j':
 711                         do_jitter = true;
 712                         jwarn = optarg;
 713                         break;
 714                 case 'k':
 715                         do_jitter = true;
 716                         jcrit = optarg;
 717                         break;
 718                 case 'H':
 719                         if(!is_host(optarg))
 720                                 usage2(_("Invalid hostname/address"), optarg);
 721                         server_address = strdup(optarg);
 722                         break;
 723                 case 't':
 724                         socket_timeout=atoi(optarg);
 725                         break;
 726                 case '4':
 727                         address_family = AF_INET;
 728                         break;
 729                 case '6':
 730 #ifdef USE_IPV6
 731                         address_family = AF_INET6;
 732 #else
 733                         usage4 (_("IPv6 support not available"));
 734 #endif
 735                         break;
 736                 case '?':
 737                         /* print short usage statement if args not parsable */
 738                         usage5 ();
 739                         break;
 740                 }
 741         }
 742
 743         if(server_address == NULL){
 744                 usage4(_("Hostname was not supplied"));
 745         }
 746
 747         return 0;
 748 }
 749
 750 char *perfd_offset (double offset)
 751 {
 752         return fperfdata ("offset", offset, "s",
 753                 true, offset_thresholds->warning->end,
 754                 true, offset_thresholds->critical->end,
 755                 false, 0, false, 0);
 756 }
 757
 758 char *perfd_jitter (double jitter)
 759 {
 760         return fperfdata ("jitter", jitter, "s",
 761                 do_jitter, jitter_thresholds->warning->end,
 762                 do_jitter, jitter_thresholds->critical->end,
 763                 true, 0, false, 0);
 764 }
 765
 766 int main(int argc, char *argv[]){
 767         int result, offset_result, jitter_result;
 768         double offset=0, jitter=0;
 769         char *result_line, *perfdata_line;
 770
 771         setlocale (LC_ALL, "");
 772         bindtextdomain (PACKAGE, LOCALEDIR);
 773         textdomain (PACKAGE);
 774
 775         result = offset_result = jitter_result = STATE_OK;
 776
 777         /* Parse extra opts if any */
 778         argv=np_extra_opts (&argc, argv, progname);
 779
 780         if (process_arguments (argc, argv) == ERROR)
 781                 usage4 (_("Could not parse arguments"));
 782
 783         set_thresholds(&offset_thresholds, owarn, ocrit);
 784         set_thresholds(&jitter_thresholds, jwarn, jcrit);
 785
 786         /* initialize alarm signal handling */
 787         signal (SIGALRM, socket_timeout_alarm_handler);
 788
 789         /* set socket timeout */
 790         alarm (socket_timeout);
 791
 792         offset = offset_request(server_address, &offset_result);
 793         /* check_ntp used to always return CRITICAL if offset_result == STATE_UNKNOWN.
 794          * Now we'll only do that is the offset thresholds were set */
 795         if (do_offset && offset_result == STATE_UNKNOWN) {
 796                 result = STATE_CRITICAL;
 797         } else {
 798                 result = get_status(fabs(offset), offset_thresholds);
 799         }
 800
 801         /* If not told to check the jitter, we don't even send packets.
 802          * jitter is checked using NTP control packets, which not all
 803          * servers recognize.  Trying to check the jitter on OpenNTPD
 804          * (for example) will result in an error
 805          */
 806         if(do_jitter){
 807                 jitter=jitter_request(&jitter_result);
 808                 result = max_state_alt(result, get_status(jitter, jitter_thresholds));
 809                 /* -1 indicates that we couldn't calculate the jitter
 810                  * Only overrides STATE_OK from the offset */
 811                 if(jitter == -1.0 && result == STATE_OK)
 812                         result = STATE_UNKNOWN;
 813         }
 814         result = max_state_alt(result, jitter_result);
 815
 816         switch (result) {
 817                 case STATE_CRITICAL :
 818                         xasprintf(&result_line, _("NTP CRITICAL:"));
 819                         break;
 820                 case STATE_WARNING :
 821                         xasprintf(&result_line, _("NTP WARNING:"));
 822                         break;
 823                 case STATE_OK :
 824                         xasprintf(&result_line, _("NTP OK:"));
 825                         break;
 826                 default :
 827                         xasprintf(&result_line, _("NTP UNKNOWN:"));
 828                         break;
 829         }
 830         if(offset_result == STATE_UNKNOWN){
 831                 xasprintf(&result_line, "%s %s", result_line, _("Offset unknown"));
 832                 xasprintf(&perfdata_line, "");
 833         } else {
 834                 xasprintf(&result_line, "%s %s %.10g secs", result_line, _("Offset"), offset);
 835                 xasprintf(&perfdata_line, "%s", perfd_offset(offset));
 836         }
 837         if (do_jitter) {
 838                 xasprintf(&result_line, "%s, jitter=%f", result_line, jitter);
 839                 xasprintf(&perfdata_line, "%s %s", perfdata_line,  perfd_jitter(jitter));
 840         }
 841         printf("%s|%s\n", result_line, perfdata_line);
 842
 843         if(server_address!=NULL) free(server_address);
 844         return result;
 845 }
 846
 847
 848
 849 void print_help(void){
 850         print_revision(progname, NP_VERSION);
 851
 852         printf ("Copyright (c) 2006 Sean Finney\n");
 853         printf (COPYRIGHT, copyright, email);
 854
 855         printf ("%s\n", _("This plugin checks the selected ntp server"));
 856
 857         printf ("\n\n");
 858
 859         print_usage();
 860         printf (UT_HELP_VRSN);
 861         printf (UT_EXTRA_OPTS);
 862         printf (UT_HOST_PORT, 'p', "123");
 863         printf (UT_IPv46);
 864         printf (" %s\n", "-w, --warning=THRESHOLD");
 865         printf ("    %s\n", _("Offset to result in warning status (seconds)"));
 866         printf (" %s\n", "-c, --critical=THRESHOLD");
 867         printf ("    %s\n", _("Offset to result in critical status (seconds)"));
 868         printf (" %s\n", "-j, --jwarn=THRESHOLD");
 869         printf ("    %s\n", _("Warning threshold for jitter"));
 870         printf (" %s\n", "-k, --jcrit=THRESHOLD");
 871         printf ("    %s\n", _("Critical threshold for jitter"));
 872         printf (UT_CONN_TIMEOUT, DEFAULT_SOCKET_TIMEOUT);
 873         printf (UT_VERBOSE);
 874
 875         printf("\n");
 876         printf("%s\n", _("Notes:"));
 877         printf(UT_THRESHOLDS_NOTES);
 878
 879         printf("\n");
 880         printf("%s\n", _("Examples:"));
 881         printf(" %s\n", _("Normal offset check:"));
 882         printf("  %s\n", ("./check_ntp -H ntpserv -w 0.5 -c 1"));
 883         printf("\n");
 884         printf(" %s\n", _("Check jitter too, avoiding critical notifications if jitter isn't available"));
 885         printf(" %s\n", _("(See Notes above for more details on thresholds formats):"));
 886         printf("  %s\n", ("./check_ntp -H ntpserv -w 0.5 -c 1 -j -1:100 -k -1:200"));
 887
 888         printf (UT_SUPPORT);
 889
 890         printf ("%s\n", _("WARNING: check_ntp is deprecated. Please use check_ntp_peer or"));
 891         printf ("%s\n\n", _("check_ntp_time instead."));
 892 }
 893
 894 void
 895 print_usage(void)
 896 {
 897         printf ("%s\n", _("WARNING: check_ntp is deprecated. Please use check_ntp_peer or"));
 898         printf ("%s\n\n", _("check_ntp_time instead."));
 899         printf ("%s\n", _("Usage:"));
 900         printf(" %s -H <host> [-w <warn>] [-c <crit>] [-j <warn>] [-k <crit>] [-4|-6] [-v verbose]\n", progname);
 901 }