|
|
@@ -1,6 +1,6 @@
|
|
|
/******************************************************************************
|
|
|
*
|
|
|
-* Nagios check_ntp plugin
|
|
|
+* Nagios check_ntp_peer plugin
|
|
|
*
|
|
|
* License: GPL
|
|
|
* Copyright (c) 2006 sean finney <seanius@seanius.net>
|
|
|
@@ -10,11 +10,16 @@
|
|
|
*
|
|
|
* Description:
|
|
|
*
|
|
|
-* This file contains the check_ntp plugin
|
|
|
+* This file contains the check_ntp_peer plugin
|
|
|
*
|
|
|
-* This plugin to check ntp servers independant of any commandline
|
|
|
+* This plugin checks an NTP server independent of any commandline
|
|
|
* programs or external libraries.
|
|
|
*
|
|
|
+* Use this plugin to check the health of an NTP server. It supports
|
|
|
+* checking the offset with the sync peer, the jitter and stratum. This
|
|
|
+* plugin will not check the clock offset between the local host and NTP
|
|
|
+* server; please use check_ntp_time for that purpose.
|
|
|
+*
|
|
|
*
|
|
|
* License Information:
|
|
|
*
|
|
|
@@ -36,7 +41,7 @@
|
|
|
|
|
|
*****************************************************************************/
|
|
|
|
|
|
-const char *progname = "check_ntp";
|
|
|
+const char *progname = "check_ntp_peer";
|
|
|
const char *revision = "$Revision$";
|
|
|
const char *copyright = "2007";
|
|
|
const char *email = "nagiosplug-devel@lists.sourceforge.net";
|
|
|
@@ -47,51 +52,28 @@ const char *email = "nagiosplug-devel@lists.sourceforge.net";
|
|
|
|
|
|
static char *server_address=NULL;
|
|
|
static int verbose=0;
|
|
|
+static int quiet=0;
|
|
|
static short do_offset=0;
|
|
|
static char *owarn="60";
|
|
|
static char *ocrit="120";
|
|
|
+static short do_stratum=0;
|
|
|
+static char *swarn="-1:16";
|
|
|
+static char *scrit="-1:16";
|
|
|
static short do_jitter=0;
|
|
|
-static char *jwarn="5000";
|
|
|
-static char *jcrit="10000";
|
|
|
+static char *jwarn="-1:5000";
|
|
|
+static char *jcrit="-1:10000";
|
|
|
+static int syncsource_found=0;
|
|
|
|
|
|
int process_arguments (int, char **);
|
|
|
thresholds *offset_thresholds = NULL;
|
|
|
thresholds *jitter_thresholds = NULL;
|
|
|
+thresholds *stratum_thresholds = NULL;
|
|
|
void print_help (void);
|
|
|
void print_usage (void);
|
|
|
|
|
|
-/* number of times to perform each request to get a good average. */
|
|
|
-#define AVG_NUM 4
|
|
|
-
|
|
|
/* max size of control message data */
|
|
|
#define MAX_CM_SIZE 468
|
|
|
|
|
|
-/* this structure holds everything in an ntp request/response as per rfc1305 */
|
|
|
-typedef struct {
|
|
|
- uint8_t flags; /* byte with leapindicator,vers,mode. see macros */
|
|
|
- uint8_t stratum; /* clock stratum */
|
|
|
- int8_t poll; /* polling interval */
|
|
|
- int8_t precision; /* precision of the local clock */
|
|
|
- int32_t rtdelay; /* total rt delay, as a fixed point num. see macros */
|
|
|
- uint32_t rtdisp; /* like above, but for max err to primary src */
|
|
|
- uint32_t refid; /* ref clock identifier */
|
|
|
- uint64_t refts; /* reference timestamp. local time local clock */
|
|
|
- uint64_t origts; /* time at which request departed client */
|
|
|
- uint64_t rxts; /* time at which request arrived at server */
|
|
|
- uint64_t txts; /* time at which request departed server */
|
|
|
-} ntp_message;
|
|
|
-
|
|
|
-/* this structure holds data about results from querying offset from a peer */
|
|
|
-typedef struct {
|
|
|
- time_t waiting; /* ts set when we started waiting for a response */
|
|
|
- int num_responses; /* number of successfully recieved responses */
|
|
|
- uint8_t stratum; /* copied verbatim from the ntp_message */
|
|
|
- double rtdelay; /* converted from the ntp_message */
|
|
|
- double rtdisp; /* converted from the ntp_message */
|
|
|
- double offset[AVG_NUM]; /* offsets from each response */
|
|
|
- uint8_t flags; /* byte with leapindicator,vers,mode. see macros */
|
|
|
-} ntp_server_results;
|
|
|
-
|
|
|
/* this structure holds everything in an ntp control message as per rfc1305 */
|
|
|
typedef struct {
|
|
|
uint8_t flags; /* byte with leapindicator,vers,mode. see macros */
|
|
|
@@ -147,57 +129,6 @@ typedef struct {
|
|
|
#define PEER_INCLUDED 0x04
|
|
|
#define PEER_SYNCSOURCE 0x06
|
|
|
|
|
|
-/**
|
|
|
- ** a note about the 32-bit "fixed point" numbers:
|
|
|
- **
|
|
|
- they are divided into halves, each being a 16-bit int in network byte order:
|
|
|
- - the first 16 bits are an int on the left side of a decimal point.
|
|
|
- - the second 16 bits represent a fraction n/(2^16)
|
|
|
- likewise for the 64-bit "fixed point" numbers with everything doubled :)
|
|
|
- **/
|
|
|
-
|
|
|
-/* macros to access the left/right 16 bits of a 32-bit ntp "fixed point"
|
|
|
- number. note that these can be used as lvalues too */
|
|
|
-#define L16(x) (((uint16_t*)&x)[0])
|
|
|
-#define R16(x) (((uint16_t*)&x)[1])
|
|
|
-/* macros to access the left/right 32 bits of a 64-bit ntp "fixed point"
|
|
|
- number. these too can be used as lvalues */
|
|
|
-#define L32(x) (((uint32_t*)&x)[0])
|
|
|
-#define R32(x) (((uint32_t*)&x)[1])
|
|
|
-
|
|
|
-/* ntp wants seconds since 1/1/00, epoch is 1/1/70. this is the difference */
|
|
|
-#define EPOCHDIFF 0x83aa7e80UL
|
|
|
-
|
|
|
-/* extract a 32-bit ntp fixed point number into a double */
|
|
|
-#define NTP32asDOUBLE(x) (ntohs(L16(x)) + (double)ntohs(R16(x))/65536.0)
|
|
|
-
|
|
|
-/* likewise for a 64-bit ntp fp number */
|
|
|
-#define NTP64asDOUBLE(n) (double)(((uint64_t)n)?\
|
|
|
- (ntohl(L32(n))-EPOCHDIFF) + \
|
|
|
- (.00000001*(0.5+(double)(ntohl(R32(n))/42.94967296))):\
|
|
|
- 0)
|
|
|
-
|
|
|
-/* convert a struct timeval to a double */
|
|
|
-#define TVasDOUBLE(x) (double)(x.tv_sec+(0.000001*x.tv_usec))
|
|
|
-
|
|
|
-/* convert an ntp 64-bit fp number to a struct timeval */
|
|
|
-#define NTP64toTV(n,t) \
|
|
|
- do{ if(!n) t.tv_sec = t.tv_usec = 0; \
|
|
|
- else { \
|
|
|
- t.tv_sec=ntohl(L32(n))-EPOCHDIFF; \
|
|
|
- t.tv_usec=(int)(0.5+(double)(ntohl(R32(n))/4294.967296)); \
|
|
|
- } \
|
|
|
- }while(0)
|
|
|
-
|
|
|
-/* convert a struct timeval to an ntp 64-bit fp number */
|
|
|
-#define TVtoNTP64(t,n) \
|
|
|
- do{ if(!t.tv_usec && !t.tv_sec) n=0x0UL; \
|
|
|
- else { \
|
|
|
- L32(n)=htonl(t.tv_sec + EPOCHDIFF); \
|
|
|
- R32(n)=htonl((uint64_t)((4294.967296*t.tv_usec)+.5)); \
|
|
|
- } \
|
|
|
- } while(0)
|
|
|
-
|
|
|
/* NTP control message header is 12 bytes, plus any data in the data
|
|
|
* field, plus null padding to the nearest 32-bit boundary per rfc.
|
|
|
*/
|
|
|
@@ -210,42 +141,6 @@ typedef struct {
|
|
|
printf("%u.%u.%u.%u", (x>>24)&0xff, (x>>16)&0xff, (x>>8)&0xff, x&0xff);\
|
|
|
}while(0);
|
|
|
|
|
|
-/* calculate the offset of the local clock */
|
|
|
-static inline double calc_offset(const ntp_message *m, const struct timeval *t){
|
|
|
- double client_tx, peer_rx, peer_tx, client_rx;
|
|
|
- client_tx = NTP64asDOUBLE(m->origts);
|
|
|
- peer_rx = NTP64asDOUBLE(m->rxts);
|
|
|
- peer_tx = NTP64asDOUBLE(m->txts);
|
|
|
- client_rx=TVasDOUBLE((*t));
|
|
|
- return (.5*((peer_tx-client_rx)+(peer_rx-client_tx)));
|
|
|
-}
|
|
|
-
|
|
|
-/* print out a ntp packet in human readable/debuggable format */
|
|
|
-void print_ntp_message(const ntp_message *p){
|
|
|
- struct timeval ref, orig, rx, tx;
|
|
|
-
|
|
|
- NTP64toTV(p->refts,ref);
|
|
|
- NTP64toTV(p->origts,orig);
|
|
|
- NTP64toTV(p->rxts,rx);
|
|
|
- NTP64toTV(p->txts,tx);
|
|
|
-
|
|
|
- printf("packet contents:\n");
|
|
|
- printf("\tflags: 0x%.2x\n", p->flags);
|
|
|
- printf("\t li=%d (0x%.2x)\n", LI(p->flags), p->flags&LI_MASK);
|
|
|
- printf("\t vn=%d (0x%.2x)\n", VN(p->flags), p->flags&VN_MASK);
|
|
|
- printf("\t mode=%d (0x%.2x)\n", MODE(p->flags), p->flags&MODE_MASK);
|
|
|
- printf("\tstratum = %d\n", p->stratum);
|
|
|
- printf("\tpoll = %g\n", pow(2, p->poll));
|
|
|
- printf("\tprecision = %g\n", pow(2, p->precision));
|
|
|
- printf("\trtdelay = %-.16g\n", NTP32asDOUBLE(p->rtdelay));
|
|
|
- printf("\trtdisp = %-.16g\n", NTP32asDOUBLE(p->rtdisp));
|
|
|
- printf("\trefid = %x\n", p->refid);
|
|
|
- printf("\trefts = %-.16g\n", NTP64asDOUBLE(p->refts));
|
|
|
- printf("\torigts = %-.16g\n", NTP64asDOUBLE(p->origts));
|
|
|
- printf("\trxts = %-.16g\n", NTP64asDOUBLE(p->rxts));
|
|
|
- printf("\ttxts = %-.16g\n", NTP64asDOUBLE(p->txts));
|
|
|
-}
|
|
|
-
|
|
|
void print_ntp_control_message(const ntp_control_message *p){
|
|
|
int i=0, numpeers=0;
|
|
|
const ntp_assoc_status_pair *peer=NULL;
|
|
|
@@ -282,222 +177,25 @@ void print_ntp_control_message(const ntp_control_message *p){
|
|
|
}
|
|
|
}
|
|
|
|
|
|
-void setup_request(ntp_message *p){
|
|
|
- struct timeval t;
|
|
|
-
|
|
|
- memset(p, 0, sizeof(ntp_message));
|
|
|
- LI_SET(p->flags, LI_ALARM);
|
|
|
- VN_SET(p->flags, 4);
|
|
|
- MODE_SET(p->flags, MODE_CLIENT);
|
|
|
- p->poll=4;
|
|
|
- p->precision=(int8_t)0xfa;
|
|
|
- L16(p->rtdelay)=htons(1);
|
|
|
- L16(p->rtdisp)=htons(1);
|
|
|
-
|
|
|
- gettimeofday(&t, NULL);
|
|
|
- TVtoNTP64(t,p->txts);
|
|
|
-}
|
|
|
-
|
|
|
-/* select the "best" server from a list of servers, and return its index.
|
|
|
- * this is done by filtering servers based on stratum, dispersion, and
|
|
|
- * finally round-trip delay. */
|
|
|
-int best_offset_server(const ntp_server_results *slist, int nservers){
|
|
|
- int i=0, j=0, cserver=0, candidates[5], csize=0;
|
|
|
-
|
|
|
- /* for each server */
|
|
|
- for(cserver=0; cserver<nservers; cserver++){
|
|
|
- /* sort out servers with error flags */
|
|
|
- if ( LI(slist[cserver].flags) != LI_NOWARNING ){
|
|
|
- if (verbose) printf("discarding peer id %d: flags=%d\n", cserver, LI(slist[cserver].flags));
|
|
|
- break;
|
|
|
- }
|
|
|
-
|
|
|
- /* compare it to each of the servers already in the candidate list */
|
|
|
- for(i=0; i<csize; i++){
|
|
|
- /* does it have an equal or better stratum? */
|
|
|
- if(slist[cserver].stratum <= slist[i].stratum){
|
|
|
- /* does it have an equal or better dispersion? */
|
|
|
- if(slist[cserver].rtdisp <= slist[i].rtdisp){
|
|
|
- /* does it have a better rtdelay? */
|
|
|
- if(slist[cserver].rtdelay < slist[i].rtdelay){
|
|
|
- break;
|
|
|
- }
|
|
|
- }
|
|
|
- }
|
|
|
- }
|
|
|
-
|
|
|
- /* if we haven't reached the current list's end, move everyone
|
|
|
- * over one to the right, and insert the new candidate */
|
|
|
- if(i<csize){
|
|
|
- for(j=5; j>i; j--){
|
|
|
- candidates[j]=candidates[j-1];
|
|
|
- }
|
|
|
- }
|
|
|
- /* regardless, if they should be on the list... */
|
|
|
- if(i<5) {
|
|
|
- candidates[i]=cserver;
|
|
|
- if(csize<5) csize++;
|
|
|
- /* otherwise discard the server */
|
|
|
- } else {
|
|
|
- DBG(printf("discarding peer id %d\n", cserver));
|
|
|
- }
|
|
|
- }
|
|
|
-
|
|
|
- if(csize>0) {
|
|
|
- DBG(printf("best server selected: peer %d\n", candidates[0]));
|
|
|
- return candidates[0];
|
|
|
- } else {
|
|
|
- DBG(printf("no peers meeting synchronization criteria :(\n"));
|
|
|
- return -1;
|
|
|
- }
|
|
|
-}
|
|
|
-
|
|
|
-/* do everything we need to get the total average offset
|
|
|
- * - we use a certain amount of parallelization with poll() to ensure
|
|
|
- * we don't waste time sitting around waiting for single packets.
|
|
|
- * - we also "manually" handle resolving host names and connecting, because
|
|
|
- * we have to do it in a way that our lazy macros don't handle currently :( */
|
|
|
-double offset_request(const char *host, int *status){
|
|
|
- int i=0, j=0, ga_result=0, num_hosts=0, *socklist=NULL, respnum=0;
|
|
|
- int servers_completed=0, one_written=0, one_read=0, servers_readable=0, best_index=-1;
|
|
|
- time_t now_time=0, start_ts=0;
|
|
|
- ntp_message *req=NULL;
|
|
|
- double avg_offset=0.;
|
|
|
- struct timeval recv_time;
|
|
|
- struct addrinfo *ai=NULL, *ai_tmp=NULL, hints;
|
|
|
- struct pollfd *ufds=NULL;
|
|
|
- ntp_server_results *servers=NULL;
|
|
|
-
|
|
|
- /* setup hints to only return results from getaddrinfo that we'd like */
|
|
|
- memset(&hints, 0, sizeof(struct addrinfo));
|
|
|
- hints.ai_family = address_family;
|
|
|
- hints.ai_protocol = IPPROTO_UDP;
|
|
|
- hints.ai_socktype = SOCK_DGRAM;
|
|
|
-
|
|
|
- /* fill in ai with the list of hosts resolved by the host name */
|
|
|
- ga_result = getaddrinfo(host, "123", &hints, &ai);
|
|
|
- if(ga_result!=0){
|
|
|
- die(STATE_UNKNOWN, "error getting address for %s: %s\n",
|
|
|
- host, gai_strerror(ga_result));
|
|
|
- }
|
|
|
-
|
|
|
- /* count the number of returned hosts, and allocate stuff accordingly */
|
|
|
- for(ai_tmp=ai; ai_tmp!=NULL; ai_tmp=ai_tmp->ai_next){ num_hosts++; }
|
|
|
- req=(ntp_message*)malloc(sizeof(ntp_message)*num_hosts);
|
|
|
- if(req==NULL) die(STATE_UNKNOWN, "can not allocate ntp message array");
|
|
|
- socklist=(int*)malloc(sizeof(int)*num_hosts);
|
|
|
- if(socklist==NULL) die(STATE_UNKNOWN, "can not allocate socket array");
|
|
|
- ufds=(struct pollfd*)malloc(sizeof(struct pollfd)*num_hosts);
|
|
|
- if(ufds==NULL) die(STATE_UNKNOWN, "can not allocate socket array");
|
|
|
- servers=(ntp_server_results*)malloc(sizeof(ntp_server_results)*num_hosts);
|
|
|
- if(servers==NULL) die(STATE_UNKNOWN, "can not allocate server array");
|
|
|
- memset(servers, 0, sizeof(ntp_server_results)*num_hosts);
|
|
|
-
|
|
|
- /* setup each socket for writing, and the corresponding struct pollfd */
|
|
|
- ai_tmp=ai;
|
|
|
- for(i=0;ai_tmp;i++){
|
|
|
- socklist[i]=socket(ai_tmp->ai_family, SOCK_DGRAM, IPPROTO_UDP);
|
|
|
- if(socklist[i] == -1) {
|
|
|
- perror(NULL);
|
|
|
- die(STATE_UNKNOWN, "can not create new socket");
|
|
|
- }
|
|
|
- if(connect(socklist[i], ai_tmp->ai_addr, ai_tmp->ai_addrlen)){
|
|
|
- die(STATE_UNKNOWN, "can't create socket connection");
|
|
|
- } else {
|
|
|
- ufds[i].fd=socklist[i];
|
|
|
- ufds[i].events=POLLIN;
|
|
|
- ufds[i].revents=0;
|
|
|
- }
|
|
|
- ai_tmp = ai_tmp->ai_next;
|
|
|
- }
|
|
|
-
|
|
|
- /* now do AVG_NUM checks to each host. we stop before timeout/2 seconds
|
|
|
- * have passed in order to ensure post-processing and jitter time. */
|
|
|
- now_time=start_ts=time(NULL);
|
|
|
- while(servers_completed<num_hosts && now_time-start_ts <= socket_timeout/2){
|
|
|
- /* loop through each server and find each one which hasn't
|
|
|
- * been touched in the past second or so and is still lacking
|
|
|
- * some responses. for each of these servers, send a new request,
|
|
|
- * and update the "waiting" timestamp with the current time. */
|
|
|
- one_written=0;
|
|
|
- now_time=time(NULL);
|
|
|
-
|
|
|
- for(i=0; i<num_hosts; i++){
|
|
|
- if(servers[i].waiting<now_time && servers[i].num_responses<AVG_NUM){
|
|
|
- if(verbose && servers[i].waiting != 0) printf("re-");
|
|
|
- if(verbose) printf("sending request to peer %d\n", i);
|
|
|
- setup_request(&req[i]);
|
|
|
- write(socklist[i], &req[i], sizeof(ntp_message));
|
|
|
- servers[i].waiting=now_time;
|
|
|
- one_written=1;
|
|
|
- break;
|
|
|
- }
|
|
|
- }
|
|
|
-
|
|
|
- /* quickly poll for any sockets with pending data */
|
|
|
- servers_readable=poll(ufds, num_hosts, 100);
|
|
|
- if(servers_readable==-1){
|
|
|
- perror("polling ntp sockets");
|
|
|
- die(STATE_UNKNOWN, "communication errors");
|
|
|
- }
|
|
|
-
|
|
|
- /* read from any sockets with pending data */
|
|
|
- for(i=0; servers_readable && i<num_hosts; i++){
|
|
|
- if(ufds[i].revents&POLLIN && servers[i].num_responses < AVG_NUM){
|
|
|
- if(verbose) {
|
|
|
- printf("response from peer %d: ", i);
|
|
|
- }
|
|
|
+char *extract_value(const char *varlist, const char *name){
|
|
|
+ char *tmpvarlist=NULL, *tmpkey=NULL, *value=NULL;
|
|
|
+ int last=0;
|
|
|
|
|
|
- read(ufds[i].fd, &req[i], sizeof(ntp_message));
|
|
|
- gettimeofday(&recv_time, NULL);
|
|
|
- DBG(print_ntp_message(&req[i]));
|
|
|
- respnum=servers[i].num_responses++;
|
|
|
- servers[i].offset[respnum]=calc_offset(&req[i], &recv_time);
|
|
|
- if(verbose) {
|
|
|
- printf("offset %.10g\n", servers[i].offset[respnum]);
|
|
|
- }
|
|
|
- servers[i].stratum=req[i].stratum;
|
|
|
- servers[i].rtdisp=NTP32asDOUBLE(req[i].rtdisp);
|
|
|
- servers[i].rtdelay=NTP32asDOUBLE(req[i].rtdelay);
|
|
|
- servers[i].waiting=0;
|
|
|
- servers[i].flags=req[i].flags;
|
|
|
- servers_readable--;
|
|
|
- one_read = 1;
|
|
|
- if(servers[i].num_responses==AVG_NUM) servers_completed++;
|
|
|
- }
|
|
|
- }
|
|
|
- /* lather, rinse, repeat. */
|
|
|
- }
|
|
|
+ /* The following code require a non-empty varlist */
|
|
|
+ if(strlen(varlist) == 0)
|
|
|
+ return NULL;
|
|
|
|
|
|
- if (one_read == 0) {
|
|
|
- die(STATE_CRITICAL, "NTP CRITICAL: No response from NTP server\n");
|
|
|
- }
|
|
|
+ tmpvarlist = strdup(varlist);
|
|
|
+ tmpkey = strtok(tmpvarlist, "=");
|
|
|
|
|
|
- /* now, pick the best server from the list */
|
|
|
- best_index=best_offset_server(servers, num_hosts);
|
|
|
- if(best_index < 0){
|
|
|
- *status=STATE_UNKNOWN;
|
|
|
- } else {
|
|
|
- /* finally, calculate the average offset */
|
|
|
- for(i=0; i<servers[best_index].num_responses;i++){
|
|
|
- avg_offset+=servers[best_index].offset[j];
|
|
|
+ do {
|
|
|
+ if(strstr(tmpkey, name) != NULL) {
|
|
|
+ value = strtok(NULL, ",");
|
|
|
+ last = 1;
|
|
|
}
|
|
|
- avg_offset/=servers[best_index].num_responses;
|
|
|
- }
|
|
|
+ } while (last == 0 && (tmpkey = strtok(NULL, "=")));
|
|
|
|
|
|
- /* cleanup */
|
|
|
- /* FIXME: Not closing the socket to avoid re-use of the local port
|
|
|
- * which can cause old NTP packets to be read instead of NTP control
|
|
|
- * pactets in jitter_request(). THERE MUST BE ANOTHER WAY...
|
|
|
- * for(j=0; j<num_hosts; j++){ close(socklist[j]); } */
|
|
|
- free(socklist);
|
|
|
- free(ufds);
|
|
|
- free(servers);
|
|
|
- free(req);
|
|
|
- freeaddrinfo(ai);
|
|
|
-
|
|
|
- if(verbose) printf("overall average offset: %.10g\n", avg_offset);
|
|
|
- return avg_offset;
|
|
|
+ return value;
|
|
|
}
|
|
|
|
|
|
void
|
|
|
@@ -511,20 +209,36 @@ setup_control_request(ntp_control_message *p, uint8_t opcode, uint16_t seq){
|
|
|
/* Remaining fields are zero for requests */
|
|
|
}
|
|
|
|
|
|
-/* XXX handle responses with the error bit set */
|
|
|
-double jitter_request(const char *host, int *status){
|
|
|
- int conn=-1, i, npeers=0, num_candidates=0, syncsource_found=0;
|
|
|
- int run=0, min_peer_sel=PEER_INCLUDED, num_selected=0, num_valid=0;
|
|
|
+/* This function does all the actual work; roughly here's what it does
|
|
|
+ * beside setting the offest, jitter and stratum passed as argument:
|
|
|
+ * - offset can be negative, so if it cannot get the offset, offset_result
|
|
|
+ * is set to UNKNOWN, otherwise OK.
|
|
|
+ * - jitter and stratum are set to -1 if they cannot be retrieved so any
|
|
|
+ * positive value means a success retrieving the value.
|
|
|
+ * - status is set to WARNING if there's no sync.peer (otherwise OK) and is
|
|
|
+ * the return value of the function.
|
|
|
+ * status is pretty much useless as syncsource_found is a global variable
|
|
|
+ * used later in main to check is the server was synchronized. It works
|
|
|
+ * so I left it alone */
|
|
|
+int ntp_request(const char *host, double *offset, int *offset_result, double *jitter, int *stratum){
|
|
|
+ int conn=-1, i, npeers=0, num_candidates=0;
|
|
|
+ double tmp_offset = 0;
|
|
|
+ int min_peer_sel=PEER_INCLUDED;
|
|
|
int peers_size=0, peer_offset=0;
|
|
|
+ int status;
|
|
|
ntp_assoc_status_pair *peers=NULL;
|
|
|
ntp_control_message req;
|
|
|
- const char *getvar = "jitter";
|
|
|
- double rval = 0.0, jitter = -1.0;
|
|
|
- char *startofvalue=NULL, *nptr=NULL;
|
|
|
+ const char *getvar = "stratum,offset,jitter";
|
|
|
+ char *data, *value, *nptr;
|
|
|
void *tmp;
|
|
|
|
|
|
+ status = STATE_OK;
|
|
|
+ *offset_result = STATE_UNKNOWN;
|
|
|
+ *jitter = *stratum = -1;
|
|
|
+
|
|
|
/* Long-winded explanation:
|
|
|
- * Getting the jitter requires a number of steps:
|
|
|
+ * Getting the sync peer offset, jitter and stratum requires a number of
|
|
|
+ * steps:
|
|
|
* 1) Send a READSTAT request.
|
|
|
* 2) Interpret the READSTAT reply
|
|
|
* a) The data section contains a list of peer identifiers (16 bits)
|
|
|
@@ -535,7 +249,8 @@ double jitter_request(const char *host, int *status){
|
|
|
* set a minimum of warning.
|
|
|
* 3) Send a READVAR request for information on each peer identified
|
|
|
* in 2b greater than the minimum selection value.
|
|
|
- * 4) Extract the jitter value from the data[] (it's ASCII)
|
|
|
+ * 4) Extract the offset, jitter and stratum value from the data[]
|
|
|
+ * (it's ASCII)
|
|
|
*/
|
|
|
my_udp_connect(server_address, 123, &conn);
|
|
|
|
|
|
@@ -564,7 +279,7 @@ double jitter_request(const char *host, int *status){
|
|
|
} while(req.op&REM_MORE);
|
|
|
|
|
|
/* first, let's find out if we have a sync source, or if there are
|
|
|
- * at least some candidates. in the case of the latter we'll issue
|
|
|
+ * at least some candidates. In the latter case we'll issue
|
|
|
* a warning but go ahead with the check on them. */
|
|
|
for (i = 0; i < npeers; i++){
|
|
|
if (PEER_SEL(peers[i].status) >= PEER_INCLUDED){
|
|
|
@@ -578,21 +293,22 @@ double jitter_request(const char *host, int *status){
|
|
|
if(verbose) printf("%d candiate peers available\n", num_candidates);
|
|
|
if(verbose && syncsource_found) printf("synchronization source found\n");
|
|
|
if(! syncsource_found){
|
|
|
- *status = STATE_UNKNOWN;
|
|
|
+ status = STATE_WARNING;
|
|
|
if(verbose) printf("warning: no synchronization source found\n");
|
|
|
}
|
|
|
|
|
|
|
|
|
- for (run=0; run<AVG_NUM; run++){
|
|
|
- if(verbose) printf("jitter run %d of %d\n", run+1, AVG_NUM);
|
|
|
- for (i = 0; i < npeers; i++){
|
|
|
- /* Only query this server if it is the current sync source */
|
|
|
- if (PEER_SEL(peers[i].status) >= min_peer_sel){
|
|
|
- num_selected++;
|
|
|
+ for (i = 0; i < npeers; i++){
|
|
|
+ /* Only query this server if it is the current sync source */
|
|
|
+ /* If there's no sync.peer, query all candidates and use the best one */
|
|
|
+ if (PEER_SEL(peers[i].status) >= min_peer_sel){
|
|
|
+ if(verbose) printf("Getting offset, jitter and stratum for peer %.2x\n", ntohs(peers[i].assoc));
|
|
|
+ asprintf(&data, "");
|
|
|
+ do{
|
|
|
setup_control_request(&req, OP_READVAR, 2);
|
|
|
req.assoc = peers[i].assoc;
|
|
|
- /* By spec, putting the variable name "jitter" in the request
|
|
|
- * should cause the server to provide _only_ the jitter value.
|
|
|
+ /* Putting the wanted variable names in the request
|
|
|
+ * cause the server to provide _only_ the requested values.
|
|
|
* thus reducing net traffic, guaranteeing us only a single
|
|
|
* datagram in reply, and making intepretation much simpler
|
|
|
*/
|
|
|
@@ -605,48 +321,97 @@ double jitter_request(const char *host, int *status){
|
|
|
DBG(print_ntp_control_message(&req));
|
|
|
|
|
|
req.count = htons(MAX_CM_SIZE);
|
|
|
- DBG(printf("recieving READVAR response...\n"));
|
|
|
+ DBG(printf("receiving READVAR response...\n"));
|
|
|
read(conn, &req, SIZEOF_NTPCM(req));
|
|
|
DBG(print_ntp_control_message(&req));
|
|
|
|
|
|
- if(req.op&REM_ERROR && strstr(getvar, "jitter")) {
|
|
|
- if(verbose) printf("The 'jitter' command failed (old ntp server?)\nRestarting with 'dispersion'...\n");
|
|
|
- getvar = "dispersion";
|
|
|
- num_selected--;
|
|
|
+ if(!(req.op&REM_ERROR))
|
|
|
+ asprintf(&data, "%s%s", data, req.data);
|
|
|
+ } while(req.op&REM_MORE);
|
|
|
+
|
|
|
+ if(req.op&REM_ERROR) {
|
|
|
+ if(strstr(getvar, "jitter")) {
|
|
|
+ if(verbose) printf("The command failed. This is usually caused by servers refusing the 'jitter'\nvariable. Restarting with 'dispersion'...\n");
|
|
|
+ getvar = "stratum,offset,dispersion";
|
|
|
+ i--;
|
|
|
+ continue;
|
|
|
+ } else if(strlen(getvar)) {
|
|
|
+ if(verbose) printf("Server didn't like dispersion either; will retrieve everything\n");
|
|
|
+ getvar = "";
|
|
|
i--;
|
|
|
continue;
|
|
|
}
|
|
|
+ }
|
|
|
+
|
|
|
+ if(verbose > 1)
|
|
|
+ printf("Server responded: >>>%s<<<\n", data);
|
|
|
+
|
|
|
+ /* get the offset */
|
|
|
+ if(verbose)
|
|
|
+ printf("parsing offset from peer %.2x: ", ntohs(peers[i].assoc));
|
|
|
+
|
|
|
+ value = extract_value(data, "offset");
|
|
|
+ nptr=NULL;
|
|
|
+ /* Convert the value if we have one */
|
|
|
+ if(value != NULL)
|
|
|
+ tmp_offset = strtod(value, &nptr) / 1000;
|
|
|
+ /* If value is null or no conversion was performed */
|
|
|
+ if(value == NULL || value==nptr) {
|
|
|
+ if(verbose) printf("error: unable to read server offset response.\n");
|
|
|
+ } else {
|
|
|
+ if(verbose) printf("%.10g\n", tmp_offset);
|
|
|
+ if(*offset_result == STATE_UNKNOWN || fabs(tmp_offset) < fabs(*offset)) {
|
|
|
+ *offset = tmp_offset;
|
|
|
+ *offset_result = STATE_OK;
|
|
|
+ } else {
|
|
|
+ /* Skip this one; move to the next */
|
|
|
+ continue;
|
|
|
+ }
|
|
|
+ }
|
|
|
|
|
|
- /* get to the float value */
|
|
|
+ if(do_jitter) {
|
|
|
+ /* get the jitter */
|
|
|
if(verbose) {
|
|
|
- printf("parsing jitter from peer %.2x: ", ntohs(peers[i].assoc));
|
|
|
+ printf("parsing %s from peer %.2x: ", strstr(getvar, "dispersion") != NULL ? "dispersion" : "jitter", ntohs(peers[i].assoc));
|
|
|
}
|
|
|
- startofvalue = strchr(req.data, '=');
|
|
|
- if(startofvalue != NULL) {
|
|
|
- startofvalue++;
|
|
|
- jitter = strtod(startofvalue, &nptr);
|
|
|
+ value = extract_value(data, strstr(getvar, "dispersion") != NULL ? "dispersion" : "jitter");
|
|
|
+ nptr=NULL;
|
|
|
+ /* Convert the value if we have one */
|
|
|
+ if(value != NULL)
|
|
|
+ *jitter = strtod(value, &nptr);
|
|
|
+ /* If value is null or no conversion was performed */
|
|
|
+ if(value == NULL || value==nptr) {
|
|
|
+ if(verbose) printf("error: unable to read server jitter/dispersion response.\n");
|
|
|
+ *jitter = -1;
|
|
|
+ } else if(verbose) {
|
|
|
+ printf("%.10g\n", *jitter);
|
|
|
+ }
|
|
|
+ }
|
|
|
+
|
|
|
+ if(do_stratum) {
|
|
|
+ /* get the stratum */
|
|
|
+ if(verbose) {
|
|
|
+ printf("parsing stratum from peer %.2x: ", ntohs(peers[i].assoc));
|
|
|
}
|
|
|
- if(startofvalue == NULL || startofvalue==nptr){
|
|
|
- printf("warning: unable to read server jitter response.\n");
|
|
|
- *status = STATE_UNKNOWN;
|
|
|
+ value = extract_value(data, "stratum");
|
|
|
+ nptr=NULL;
|
|
|
+ /* Convert the value if we have one */
|
|
|
+ if(value != NULL)
|
|
|
+ *stratum = strtol(value, &nptr, 10);
|
|
|
+ if(value == NULL || value==nptr) {
|
|
|
+ if(verbose) printf("error: unable to read server stratum response.\n");
|
|
|
+ *stratum = -1;
|
|
|
} else {
|
|
|
- if(verbose) printf("%g\n", jitter);
|
|
|
- num_valid++;
|
|
|
- rval += jitter;
|
|
|
+ if(verbose) printf("%i\n", *stratum);
|
|
|
}
|
|
|
}
|
|
|
- }
|
|
|
- if(verbose){
|
|
|
- printf("jitter parsed from %d/%d peers\n", num_valid, num_selected);
|
|
|
- }
|
|
|
- }
|
|
|
-
|
|
|
- rval = num_valid ? rval / num_valid : -1.0;
|
|
|
+ } /* if (PEER_SEL(peers[i].status) >= min_peer_sel) */
|
|
|
+ } /* for (i = 0; i < npeers; i++) */
|
|
|
|
|
|
close(conn);
|
|
|
if(peers!=NULL) free(peers);
|
|
|
- /* If we return -1.0, it means no synchronization source was found */
|
|
|
- return rval;
|
|
|
+
|
|
|
+ return status;
|
|
|
}
|
|
|
|
|
|
int process_arguments(int argc, char **argv){
|
|
|
@@ -658,8 +423,11 @@ int process_arguments(int argc, char **argv){
|
|
|
{"verbose", no_argument, 0, 'v'},
|
|
|
{"use-ipv4", no_argument, 0, '4'},
|
|
|
{"use-ipv6", no_argument, 0, '6'},
|
|
|
+ {"quiet", no_argument, 0, 'q'},
|
|
|
{"warning", required_argument, 0, 'w'},
|
|
|
{"critical", required_argument, 0, 'c'},
|
|
|
+ {"swarn", required_argument, 0, 'W'},
|
|
|
+ {"scrit", required_argument, 0, 'C'},
|
|
|
{"jwarn", required_argument, 0, 'j'},
|
|
|
{"jcrit", required_argument, 0, 'k'},
|
|
|
{"timeout", required_argument, 0, 't'},
|
|
|
@@ -672,7 +440,7 @@ int process_arguments(int argc, char **argv){
|
|
|
usage ("\n");
|
|
|
|
|
|
while (1) {
|
|
|
- c = getopt_long (argc, argv, "Vhv46w:c:j:k:t:H:", longopts, &option);
|
|
|
+ c = getopt_long (argc, argv, "Vhv46qw:c:W:C:j:k:t:H:", longopts, &option);
|
|
|
if (c == -1 || c == EOF || c == 1)
|
|
|
break;
|
|
|
|
|
|
@@ -688,6 +456,9 @@ int process_arguments(int argc, char **argv){
|
|
|
case 'v':
|
|
|
verbose++;
|
|
|
break;
|
|
|
+ case 'q':
|
|
|
+ quiet = 1;
|
|
|
+ break;
|
|
|
case 'w':
|
|
|
do_offset=1;
|
|
|
owarn = optarg;
|
|
|
@@ -696,6 +467,14 @@ int process_arguments(int argc, char **argv){
|
|
|
do_offset=1;
|
|
|
ocrit = optarg;
|
|
|
break;
|
|
|
+ case 'W':
|
|
|
+ do_stratum=1;
|
|
|
+ swarn = optarg;
|
|
|
+ break;
|
|
|
+ case 'C':
|
|
|
+ do_stratum=1;
|
|
|
+ scrit = optarg;
|
|
|
+ break;
|
|
|
case 'j':
|
|
|
do_jitter=1;
|
|
|
jwarn = optarg;
|
|
|
@@ -746,24 +525,31 @@ char *perfd_offset (double offset)
|
|
|
|
|
|
char *perfd_jitter (double jitter)
|
|
|
{
|
|
|
- return fperfdata ("jitter", jitter, "s",
|
|
|
+ return fperfdata ("jitter", jitter, "",
|
|
|
do_jitter, jitter_thresholds->warning->end,
|
|
|
do_jitter, jitter_thresholds->critical->end,
|
|
|
TRUE, 0, FALSE, 0);
|
|
|
}
|
|
|
|
|
|
+char *perfd_stratum (int stratum)
|
|
|
+{
|
|
|
+ return perfdata ("stratum", stratum, "",
|
|
|
+ do_stratum, (int)stratum_thresholds->warning->end,
|
|
|
+ do_stratum, (int)stratum_thresholds->critical->end,
|
|
|
+ TRUE, 0, TRUE, 16);
|
|
|
+}
|
|
|
+
|
|
|
int main(int argc, char *argv[]){
|
|
|
- int result, offset_result, jitter_result;
|
|
|
+ int result, offset_result, stratum;
|
|
|
double offset=0, jitter=0;
|
|
|
char *result_line, *perfdata_line;
|
|
|
|
|
|
- result = offset_result = jitter_result = STATE_OK;
|
|
|
-
|
|
|
if (process_arguments (argc, argv) == ERROR)
|
|
|
usage4 (_("Could not parse arguments"));
|
|
|
|
|
|
set_thresholds(&offset_thresholds, owarn, ocrit);
|
|
|
set_thresholds(&jitter_thresholds, jwarn, jcrit);
|
|
|
+ set_thresholds(&stratum_thresholds, swarn, scrit);
|
|
|
|
|
|
/* initialize alarm signal handling */
|
|
|
signal (SIGALRM, socket_timeout_alarm_handler);
|
|
|
@@ -771,29 +557,24 @@ int main(int argc, char *argv[]){
|
|
|
/* set socket timeout */
|
|
|
alarm (socket_timeout);
|
|
|
|
|
|
- offset = offset_request(server_address, &offset_result);
|
|
|
- /* check_ntp used to always return CRITICAL if offset_result == STATE_UNKNOWN.
|
|
|
- * Now we'll only do that is the offset thresholds were set */
|
|
|
- if (do_offset && offset_result == STATE_UNKNOWN) {
|
|
|
- result = STATE_CRITICAL;
|
|
|
+ /* This returns either OK or WARNING (See comment preceeding ntp_request) */
|
|
|
+ result = ntp_request(server_address, &offset, &offset_result, &jitter, &stratum);
|
|
|
+
|
|
|
+ if(offset_result == STATE_UNKNOWN) {
|
|
|
+ /* if there's no sync peer (this overrides ntp_request output): */
|
|
|
+ result = (quiet == 1 ? STATE_UNKNOWN : STATE_CRITICAL);
|
|
|
} else {
|
|
|
- result = get_status(fabs(offset), offset_thresholds);
|
|
|
+ /* Be quiet if there's no candidates either */
|
|
|
+ if (quiet == 1 && result == STATE_WARNING)
|
|
|
+ result = STATE_UNKNOWN;
|
|
|
+ result = max_state_alt(result, get_status(fabs(offset), offset_thresholds));
|
|
|
}
|
|
|
|
|
|
- /* If not told to check the jitter, we don't even send packets.
|
|
|
- * jitter is checked using NTP control packets, which not all
|
|
|
- * servers recognize. Trying to check the jitter on OpenNTPD
|
|
|
- * (for example) will result in an error
|
|
|
- */
|
|
|
- if(do_jitter){
|
|
|
- jitter=jitter_request(server_address, &jitter_result);
|
|
|
+ if(do_stratum)
|
|
|
+ result = max_state_alt(result, get_status(stratum, stratum_thresholds));
|
|
|
+
|
|
|
+ if(do_jitter)
|
|
|
result = max_state_alt(result, get_status(jitter, jitter_thresholds));
|
|
|
- /* -1 indicates that we couldn't calculate the jitter
|
|
|
- * Only overrides STATE_OK from the offset */
|
|
|
- if(jitter == -1.0 && result == STATE_OK)
|
|
|
- result = STATE_UNKNOWN;
|
|
|
- }
|
|
|
- result = max_state_alt(result, jitter_result);
|
|
|
|
|
|
switch (result) {
|
|
|
case STATE_CRITICAL :
|
|
|
@@ -809,6 +590,9 @@ int main(int argc, char *argv[]){
|
|
|
asprintf(&result_line, "NTP UNKNOWN:");
|
|
|
break;
|
|
|
}
|
|
|
+ if(!syncsource_found)
|
|
|
+ asprintf(&result_line, "%s %s,", result_line, _("Server not synchronized"));
|
|
|
+
|
|
|
if(offset_result == STATE_UNKNOWN){
|
|
|
asprintf(&result_line, "%s %s", result_line, _("Offset unknown"));
|
|
|
asprintf(&perfdata_line, "");
|
|
|
@@ -818,7 +602,11 @@ int main(int argc, char *argv[]){
|
|
|
}
|
|
|
if (do_jitter) {
|
|
|
asprintf(&result_line, "%s, jitter=%f", result_line, jitter);
|
|
|
- asprintf(&perfdata_line, "%s %s", perfdata_line, perfd_jitter(jitter));
|
|
|
+ asprintf(&perfdata_line, "%s %s", perfdata_line, perfd_jitter(jitter));
|
|
|
+ }
|
|
|
+ if (do_stratum) {
|
|
|
+ asprintf(&result_line, "%s, stratum=%i", result_line, stratum);
|
|
|
+ asprintf(&perfdata_line, "%s %s", perfdata_line, perfd_stratum(stratum));
|
|
|
}
|
|
|
printf("%s|%s\n", result_line, perfdata_line);
|
|
|
|
|
|
@@ -833,18 +621,24 @@ void print_help(void){
|
|
|
|
|
|
printf ("Copyright (c) 2006 Sean Finney\n");
|
|
|
printf (COPYRIGHT, copyright, email);
|
|
|
-
|
|
|
- printf ("%s\n", _("This plugin checks the selected ntp server"));
|
|
|
|
|
|
- printf ("\n\n");
|
|
|
-
|
|
|
+ printf ("%s\n", _("This plugin checks the selected ntp server"));
|
|
|
+
|
|
|
+ printf ("\n\n");
|
|
|
+
|
|
|
print_usage();
|
|
|
printf (_(UT_HELP_VRSN));
|
|
|
printf (_(UT_HOST_PORT), 'p', "123");
|
|
|
+ printf (" %s\n", "-q, --quiet");
|
|
|
+ printf (" %s\n", _("Returns UNKNOWN instead of CRITICAL or WARNING if server isn't synchronized"));
|
|
|
printf (" %s\n", "-w, --warning=THRESHOLD");
|
|
|
printf (" %s\n", _("Offset to result in warning status (seconds)"));
|
|
|
printf (" %s\n", "-c, --critical=THRESHOLD");
|
|
|
printf (" %s\n", _("Offset to result in critical status (seconds)"));
|
|
|
+ printf (" %s\n", "-W, --warning=THRESHOLD");
|
|
|
+ printf (" %s\n", _("Warning threshold for stratum"));
|
|
|
+ printf (" %s\n", "-W, --critical=THRESHOLD");
|
|
|
+ printf (" %s\n", _("Critical threshold for stratum"));
|
|
|
printf (" %s\n", "-j, --warning=THRESHOLD");
|
|
|
printf (" %s\n", _("Warning threshold for jitter"));
|
|
|
printf (" %s\n", "-k, --critical=THRESHOLD");
|
|
|
@@ -854,17 +648,26 @@ void print_help(void){
|
|
|
|
|
|
printf("\n");
|
|
|
printf("%s\n", _("Notes:"));
|
|
|
+ printf(" %s\n", _("This plugin checks an NTP server independent of any commandline"));
|
|
|
+ printf(" %s\n\n", _("programs or external libraries."));
|
|
|
+ printf(" %s\n", _("Use this plugin to check the health of an NTP server. It supports"));
|
|
|
+ printf(" %s\n", _("checking the offset with the sync peer, the jitter and stratum. This"));
|
|
|
+ printf(" %s\n", _("plugin will not check the clock offset between the local host and NTP"));
|
|
|
+ printf(" %s\n\n", _("server; please use check_ntp_time for that purpose."));
|
|
|
+
|
|
|
printf(" %s\n", _("See:"));
|
|
|
printf(" %s\n", ("http://nagiosplug.sourceforge.net/developer-guidelines.html#THRESHOLDFORMAT"));
|
|
|
printf(" %s\n", _("for THRESHOLD format and examples."));
|
|
|
|
|
|
printf("\n");
|
|
|
printf("%s\n", _("Examples:"));
|
|
|
- printf(" %s\n", _("Normal offset check:"));
|
|
|
- printf(" %s\n", ("./check_ntp -H ntpserv -w 0.5 -c 1"));
|
|
|
+ printf(" %s\n", _("Normal NTP server check:"));
|
|
|
+ printf(" %s\n", ("./check_ntp_peer -H ntpserv -w 0.5 -c 1"));
|
|
|
printf(" %s\n", _("Check jitter too, avoiding critical notifications if jitter isn't available"));
|
|
|
printf(" %s\n", _("(See Notes above for more details on thresholds formats):"));
|
|
|
- printf(" %s\n", ("./check_ntp -H ntpserv -w 0.5 -c 1 -j -1:100 -k -1:200"));
|
|
|
+ printf(" %s\n", ("./check_ntp_peer -H ntpserv -w 0.5 -c 1 -j -1:100 -k -1:200"));
|
|
|
+ printf(" %s\n", _("Check only stratum:"));
|
|
|
+ printf(" %s\n", ("./check_ntp_peer -H ntpserv -W 4 -C 6"));
|
|
|
|
|
|
printf (_(UT_SUPPORT));
|
|
|
}
|
|
|
@@ -872,6 +675,7 @@ void print_help(void){
|
|
|
void
|
|
|
print_usage(void)
|
|
|
{
|
|
|
- printf (_("Usage:"));
|
|
|
- printf(" %s -H <host> [-w <warn>] [-c <crit>] [-j <warn>] [-k <crit>] [-v verbose]\n", progname);
|
|
|
+ printf (_("Usage:"));
|
|
|
+ printf(" %s -H <host> [-w <warn>] [-c <crit>] [-W <warn>] [-C <crit>]\n", progname);
|
|
|
+ printf(" [-j <warn>] [-k <crit>] [-v verbose]\n");
|
|
|
}
|
Thomas Guyot-Sionnest