Implement nagios check mode in ping903q

2 files changed, 253 insertions, 116 deletions

diff --git a/doc/ping903q.1 b/doc/ping903q.1
index 3a2a91a..6e024d7 100644
--- a/doc/ping903q.1
+++ b/doc/ping903q.1
@@ -13,14 +13,22 @@
 .\"
 .\" You should have received a copy of the GNU General Public License
 .\" along with Ping903.  If not, see <http://www.gnu.org/licenses/>.
-.TH PING903Q 8 "February 12, 2020" "PING903Q" "User Commands"
+.TH PING903Q 8 "February 13, 2020" "PING903Q" "User Commands"
 .SH NAME
 ping903q \- ping903 query tool
 .SH SYNOPSIS
 \fBping903q\fR\
  [\fB\-hVv\fR]\
- [\fB\-c \fIFILE\fR]\
+ [\fB\-f \fIFILE\fR]\
  [\fIIP\fR]
+
+or
+
+\fBping903q\fR\
+  [\fB\-f \fIFILE\fR]\
+  \fB\-H \fIHOST\fR\
+  \fB\-c \fIRTA\fB,\fIPCT\fB%\fR\
+  \fB\-w \fIRTA\fB,\fIPCT\fB%\fR  
 .SH DESCRIPTION
 Queries monitoring statistics from the \fBping903\fR daemon.  When
 used with a single argument (\fIIP\fR), displays information about
@@ -41,40 +49,34 @@ exists, it will look obtain the socket address from the \fBlisten\fR
 statement.  See
 .BR ping903.conf (5),
 for detailed description of the configuration file.
+.SS Nagios check mode
+When the \fB\-H\fR, \fB\-c\fR, and \fB\-w\fR options are used, the
+program enters \fINagios check mode\fR.  In this mode its output
+complies with the requirements for external \fBNagios\fR check
+programs.
 .SH EXIT CODE
 When called with one argument, the program exits with code 0 (success)
-if the IP is alive and 1 otherwise.
+if the IP is alive and 2 otherwise.
 .PP
 When called without arguments, the program exits with code 0 if all
-monitored IP addresses are alive, 1 if none of them is reachable and 2
+monitored IP addresses are alive, 2 if none of them is reachable and 1
 if some of them are.
 .PP
-Other possible exit codes are:
-.TP
-.B 64
-Command line usage error.
-.TP
-.B 65
-Data format error.  The returned JSON is malformed.
-.TP
-.B 69
-Service unavailable.  This means some error occurred, which will be
-diagnosed on the standard error.
-.TP
-.B 70
-Internal software error.
+Exit codes in nagios check mode:
 .TP
-.B 71
-System error (e.g. error reading from socket or the like).
+.B 0
+Success
 .TP
-.B 76
-Protocol error when talking to the daemon.
+.B 1
+Warning condition.
 .TP
-.B 78
-Error in configuration file.
+.B 2
+Critical condition.
+.PP
+If any error is encountered, \fBping903q\fR exits with status \fB3\fR.
 .SH OPTIONS
 .TP
-\fB\-c \fIFILE\fR
+\fB\-f \fIFILE\fR
 Read configuration from \fIFILE\fR instead of from the default
 .BR /etc/ping903.conf .
 .TP
@@ -86,8 +88,28 @@ Print program version, copyright information, and exit.
 .TP
 .B \-v
 Turn on verbose output.
+.SS Options valid in Nagios check mode
+The presense of any of these options switches \fBping903q\fR to Nagios
+check mode.  For this mode to succeed, all three options must be specified.
+.TP
+\fB\-H \fIHOST\fR
+Sets host name or IP address to query for.
+.TP
+\fB\-c \fIRTA\fB,\fIPCT\fB%\fR
+Sets the critical threshold value.  \fIRTA\fR is the round-trip
+average and \fIPCT\fR is the package loss percentage values.  The
+critical condition is entered if either the returned round-trip
+average becomes greater than or equal to \fIRTA\fR or the lost package
+percentage becomes greater than or equal to \fIPCT\fR.  Note that both
+parts must be present and must be valid floating-point numbers.  Note
+also that the use of the percent sign is mandatory.
+.TP
+\fB\-w \fIRTA\fB,\fIPCT\fB%\fR
+Sets the warning threshold value.  See above for the discussion of the
+arguments.
 .SH SEE ALSO
-.BR ping903 (8).
+.BR ping903 (8),
+.BR Nagios <https://www.nagios.org/>.
 .SH COPYRIGHT
 Copyright \(co 2020 Sergey Poznyakoff
 .br
diff --git a/src/ping903q.c b/src/ping903q.c
index e709f64..ba5caa1 100644
--- a/src/ping903q.c
+++ b/src/ping903q.c
@@ -1,8 +1,10 @@
 #include <config.h>
 #include <stdio.h>
 #include <stdlib.h>
-#include <sysexits.h>
-#include <sys/types.h>          /* See NOTES */
+#include <syslog.h>
+#include <limits.h>
+#include <math.h>
+#include <sys/types.h>
 #include <sys/socket.h>
 #include <netdb.h>
 #include <errno.h>
@@ -18,11 +20,31 @@ int verbose;
 char const http_version[] = "HTTP/1.1";
 
 enum {
-	EX_P903_ALIVE = 0,
-	EX_P903_NOTALIVE = 1,
-	EX_P903_MIXED = 2
+	EX_NAGIOS_OK = 0,
+	EX_NAGIOS_WARNING = 1,
+	EX_NAGIOS_CRITICAL = 2,
+	EX_NAGIOS_UNKNOWN = 3,
+	EX_NAGIOS_DEPENDENT = 4
 };
 
+static char const *status_str[] = {
+	"OK",
+	"WARNING",
+	"CRITICAL",
+	"UNKNOWN",
+	"DEPENDENT"
+};
+
+static void
+abend(char *fmt, ...)
+{
+	va_list ap;
+	va_start(ap, fmt);
+	vlogger(LOG_CRIT, fmt, ap);
+	va_end(ap);
+	exit(EX_NAGIOS_UNKNOWN);
+}
+
 static char *
 read_listen(char **ret_service)
 {
@@ -38,11 +60,9 @@ read_listen(char **ret_service)
 	if (!fp) {
 		if (errno == ENOENT)
 			return NULL;
-		else {
-			fatal("can't open %s: %s",
+		else
+			abend("can't open %s: %s",
 			      config_file, strerror(errno));
-			exit(EX_CONFIG);
-		}
 	}
 
 	while (fgets(buf, sizeof(buf), fp)) {
@@ -113,26 +133,18 @@ http_connect(char *node, char *service)
 	hints.ai_socktype = SOCK_STREAM;
 	hints.ai_protocol = IPPROTO_TCP;
 	rc = getaddrinfo(node, service, &hints, &res);
-	if (rc) {
-		fatal("%s: %s", node, gai_strerror(rc));
-		exit(EX_UNAVAILABLE);
-	}
+	if (rc)
+		abend("%s: %s", node, gai_strerror(rc));
 
 	fd = socket(res->ai_family, res->ai_socktype, 0);
-	if (fd == -1) {
-		fatal("socket: %s", strerror(errno));
-		exit(EX_UNAVAILABLE);
-	}
-	if (connect(fd, (struct sockaddr *)res->ai_addr, res->ai_addrlen)) {
-		fatal("failed to connect to %s:%s: %s",
+	if (fd == -1)
+		abend("socket: %s", strerror(errno));
+	if (connect(fd, (struct sockaddr *)res->ai_addr, res->ai_addrlen))
+		abend("failed to connect to %s:%s: %s",
 		      node, service, strerror(errno));
-		exit(EX_UNAVAILABLE);
-	}
 	http = fdopen(fd, "w+");
-	if (http == NULL) {
-		fatal("failed to open socket: %s", strerror(errno));
-		exit(EX_UNAVAILABLE);
-	}
+	if (http == NULL)
+		abend("failed to open socket: %s", strerror(errno));
 	freeaddrinfo(res);
 }
 
@@ -180,14 +192,11 @@ http_readline(struct http_buf *hbuf)
 			hbuf->base = e2nrealloc(hbuf->base, &hbuf->size, 1);
 		}
 		if (fgets(hbuf->base + hbuf->len, hbuf->size, http) == NULL) {
-			if (feof(http)) {
-				fatal("connection closed prematurely");
-				exit(EX_PROTOCOL);
-			} else {
-				fatal("error reading from socket: %s",
+			if (feof(http))
+				abend("connection closed prematurely");
+			else
+				abend("error reading from socket: %s",
 				      strerror(errno));
-				exit(EX_OSERR);
-			}
 		}
 		hbuf->len += strlen(hbuf->base + hbuf->len);
 		if (hbuf->base[hbuf->len-1] == '\n') {
@@ -294,19 +303,16 @@ http_recv(struct http_resp *resp)
 			char *res[3];
 
 			if (strsplit3(hbuf.base, res, 0)) {
-				fatal("malformed HTTP response");
-				exit(EX_PROTOCOL);
+				abend("malformed HTTP response");
 			}
 			if (strcmp(res[0], http_version)) {
-				fatal("unsupported HTTP version: %s", res[0]);
-				exit(EX_PROTOCOL);
+				abend("unsupported HTTP version: %s", res[0]);
 			}
 			free(res[0]);
 
 			resp->code = atoi(res[1]);
 			if (resp->code <= 0 || resp->code > 559) {
-				fatal("bad response code: %s", res[1]);
-				exit(EX_PROTOCOL);
+				abend("bad response code: %s", res[1]);
 			}
 			free(res[1]);
 			resp->reason = res[2];
@@ -331,8 +337,7 @@ http_recv(struct http_resp *resp)
 		char *p;
 		unsigned long len = strtoul(hval, &p, 10);
 		if (*p) {
-			fatal("malformed content-length");
-			exit(EX_PROTOCOL);
+			abend("malformed content-length");
 		}
 		resp->content = emalloc(len + 1);
 		resp->content_length = len;
@@ -341,13 +346,11 @@ http_recv(struct http_resp *resp)
 		while (len) {
 			ssize_t n = fread(p, 1, len, http);
 			if (n == -1) {
-				fatal("socket read error: %s",
+				abend("socket read error: %s",
 				      strerror(errno));
-				exit(EX_OSERR);
 			}
 			if (n == 0) {
-				error("warning: short read from socket");
-				exit(EX_OSERR);
+				abend("short read from socket");
 			}
 			len -= n;
 			p += n;
@@ -395,18 +398,16 @@ ejson_get(struct json_value *obj, char *name, int type)
 {
 	struct json_value *jv;
 	if (json_object_get(obj, name, &jv)) {
-		fatal("no \"%s\" member in the response", name);
-		exit(EX_DATAERR);
+		abend("no \"%s\" member in the response", name);
 	}
 	if (jv->type != type) {
-		fatal("\"%s\" member has wrong type", name);
-		exit(EX_DATAERR);
+		abend("\"%s\" member has wrong type", name);
 	}
 	return jv;
 }
 
 static int
-print_host_status(struct json_value *obj)
+print_host_status(struct json_value *obj, void *unused)
 {
 	struct json_value *jv;
 	int alive;
@@ -418,7 +419,7 @@ print_host_status(struct json_value *obj)
 	jv = ejson_get(obj, "status", json_bool);
 	if (!jv->v.b) {
 		printf("no response from %s\n", name);
-		return EX_P903_NOTALIVE;
+		return EX_NAGIOS_CRITICAL;
 	}
 
 	alive = ejson_get(obj, "alive", json_bool)->v.b;
@@ -442,11 +443,12 @@ print_host_status(struct json_value *obj)
 		       ejson_get(obj, "tmax", json_number)->v.n,
 		       ejson_get(obj, "stddev", json_number)->v.n);
 	}
-	return alive ? EX_P903_ALIVE : EX_P903_NOTALIVE;
+	return alive ? EX_NAGIOS_OK : EX_NAGIOS_CRITICAL;
 }
 
 static void
-query_host(char const *name)
+query_host(char const *name, int (*report)(struct json_value *, void *),
+	   void *report_data)
 {
 	int rc;
 	struct http_resp resp;
@@ -456,33 +458,28 @@ query_host(char const *name)
 	char *p;
 	ssize_t n = snprintf(url, sizeof(url), "/host/%s", name);
 	if (n < 0 || n == sizeof(url)) {
-		fatal("bad host name or IP");
-		exit(EX_USAGE);
+		abend("bad host name or IP");
 	}
 	http_query("GET", url, NULL);
 	http_resp_init(&resp);
 	http_recv(&resp);
 	if (resp.code != 200) {
-		fatal("%s", resp.reason);
-		exit(EX_UNAVAILABLE);
+		abend("%s", resp.reason);
 	}
 
 	hval = http_resp_get_header(&resp, "content-type");
 	if (!hval || strcmp(hval, "application/json")) {
-		fatal("missing or unsupported content type");
-		exit(EX_UNAVAILABLE);
+		abend("missing or unsupported content type");
 	}
 
 	rc = json_parse_string(resp.content, &obj, &p);
 	if (rc != JSON_E_NOERR) {
-		fatal("%s near %s", json_strerror(rc), p);
-		exit(EX_DATAERR);
+		abend("%s near %s", json_strerror(rc), p);
 	}
 	if (obj->type != json_object) {
-		fatal("returned entity has wrong type");
-		exit(EX_DATAERR);
+		abend("returned entity has wrong type");
 	}
-	exit(print_host_status(obj));
+	exit(report(obj, report_data));
 }
 
 static void
@@ -501,53 +498,134 @@ query_all(void)
 	http_resp_init(&resp);
 	http_recv(&resp);
 	if (resp.code != 200) {
-		fatal("%s", resp.reason);
-		exit(EX_UNAVAILABLE);
+		abend("%s", resp.reason);
 	}
 
 	hval = http_resp_get_header(&resp, "content-type");
 	if (!hval || strcmp(hval, "application/json")) {
-		fatal("missing or unsupported content type");
-		exit(EX_UNAVAILABLE);
+		abend("missing or unsupported content type");
 	}
 
 	rc = json_parse_string(resp.content, &obj, &p);
 	if (rc != JSON_E_NOERR) {
-		fatal("%s near %s", json_strerror(rc), p);
-		exit(EX_DATAERR);
+		abend("%s near %s", json_strerror(rc), p);
 	}
 	if (obj->type != json_array) {
-		fatal("returned entity has wrong type");
-		exit(EX_DATAERR);
+		abend("returned entity has wrong type");
 	}
 	len = json_array_length(obj);
 	for (i = 0; i < len; i++) {
 		struct json_value *jv;
 		if (json_array_get(obj, i, &jv)) {
-			fatal("can't get element %lu", (unsigned long) i);
-			exit(EX_SOFTWARE);
+			abend("can't get element %lu", (unsigned long) i);
 		}
 		if (jv->type == json_object)
-			count[print_host_status(jv)]++;
+			count[print_host_status(jv, NULL) != EX_NAGIOS_OK]++;
 	}
 	if (count[1] == 0)
-		exit(EX_P903_ALIVE);
+		exit(EX_NAGIOS_OK);
 	if (count[0] == 0)
-		exit(EX_P903_NOTALIVE);
-	exit(EX_P903_MIXED);
+		exit(EX_NAGIOS_CRITICAL);
+	exit(EX_NAGIOS_WARNING);
+}
+
+struct nagios_threshold {
+	double round_trip;
+	double loss_pct;
+};
+
+struct nagios_check_data {
+	struct nagios_threshold wth;
+	struct nagios_threshold cth;
+};
+
+static void
+parse_nagios_threshold(char const *val, struct nagios_threshold *tp)
+{
+	char *p;
+	double d;
+
+	d = strtod(val, &p);
+	if (d <= 0 || d == HUGE_VAL) {
+		abend("invalid threshold: %s", val);
+	}
+	tp->round_trip = d;
+	if (*p != ',') {
+		abend("threshold missing percentage: %s", val);
+	}
+	d = strtod(p + 1, &p);
+	if (d < 0 || (d == 0 && errno) || d > 100 || *p != '%') {
+		abend("invalid threshold: %s", val);
+	}
+	tp->loss_pct = d;
+}
+
+static void
+print_perfdata(double rta, double pl, struct nagios_check_data *cd)
+{
+	printf("|rta=%.6fms;%.6f;%.6f;%.6f pl=%.6f%%;%.6f;%.6f;%d\n",
+	       rta, cd->wth.round_trip, cd->cth.round_trip, 0.0,
+	       pl, cd->wth.loss_pct, cd->cth.loss_pct, 0);
+}
+
+static inline int newstatus(int old, int new) {
+	return old > new ? old : new;
+}
+
+static int
+nagios_check(struct json_value *obj, void *data)
+{
+	struct nagios_check_data *cd = data;
+	struct json_value *jv;
+ 	char const *name;
+	int status;
+	double rta, loss;
+	
+	jv = ejson_get(obj, "name", json_string);
+	name = jv->v.s;
+
+	jv = ejson_get(obj, "status", json_bool);
+	if (!jv->v.b) {
+		printf("PING CRITICAL - Packet loss = 100%");
+		print_perfdata(cd->cth.round_trip, 100.0, cd);
+		exit(EX_NAGIOS_CRITICAL);
+	}
+
+	status = EX_NAGIOS_OK;
+	rta = ejson_get(obj, "avg", json_number)->v.n;
+	if (rta >= cd->cth.round_trip)
+		status = newstatus(status, EX_NAGIOS_CRITICAL);
+	if (rta >= cd->wth.round_trip)
+		status = newstatus(status, EX_NAGIOS_WARNING);
+	
+	loss = ejson_get(obj, "loss", json_number)->v.n;
+	if (loss >= cd->cth.loss_pct)
+		status = newstatus(status, EX_NAGIOS_CRITICAL);
+	if (loss >= cd->wth.loss_pct)
+		status = newstatus(status, EX_NAGIOS_WARNING);
+	printf("PING %s Packet loss = %d%%, RTA = %.2f ms",
+	       status_str[status], (int)loss, rta);
+	print_perfdata(rta, loss, cd);
+	exit(status);
 }
 
 void
 usage(void)
 {
-	printf("Usage: %s [-hVv] [-c FILE]\n", progname);
+	printf("Usage: %s [-hVv] [-f FILE] [HOST]\n", progname);
+	printf("   or: %s -H HOST -c RTA,PCT%% -w RTA,PCT%%\n", progname);
 	printf("Query ping903 daemon.\n");
 	printf("\n");
 	printf("Options:\n\n");
-	printf("  -c FILE        read configuration from FILE\n");
+	printf("  -f FILE        read configuration from FILE\n");
 	printf("  -h             print this help test\n");
 	printf("  -V             print program version and exit\n");
 	printf("  -v             additional verbosity\n");
+	printf("\nNagios check mode:\n\n");
+	printf("  -H HOST        query statistics for this host\n");
+	printf("  -c RTA,PCT%%    set critical threshold\n");
+	printf("  -w RTA,PCT%%    set warning threshold\n");
+	printf("\n(all three must be given in this mode)\n");
 	printf("\n");
 	printf("Report bugs to <%s>.\n", PACKAGE_BUGREPORT);
 	printf("%s home page: <%s>\n", PACKAGE_NAME, PACKAGE_URL);
@@ -565,8 +643,14 @@ main(int argc, char **argv)
 {
 	int c;
 	char *p, *node, *service;
-
+	int nagios_check_mode = 0;
+	char const *host = NULL;
+	char *c_opt = NULL;
+	char *w_opt = NULL;
+	struct nagios_check_data chkdata;
+	
 	set_progname(argv[0]);
+	
 	if (argc == 2) {
 		if (strcmp(argv[1], "--help") == 0) {
 			usage();
@@ -577,11 +661,19 @@ main(int argc, char **argv)
 			exit(0);
 		}
 	}
-	while ((c = getopt(argc, argv, "c:hVv")) != EOF) {
+	while ((c = getopt(argc, argv, "c:f:H:hVvw:")) != EOF) {
 		switch (c) {
 		case 'c':
+			c_opt = optarg;
+			nagios_check_mode = 1;
+			break;
+		case 'f':
 			config_file = optarg;
 			break;
+		case 'H':
+			host = optarg;
+			nagios_check_mode = 1;
+			break;
 		case 'h':
 			usage();
 			exit(0);
@@ -591,13 +683,32 @@ main(int argc, char **argv)
 		case 'v':
 			verbose++;
 			break;
+		case 'w':
+			w_opt = optarg;
+			nagios_check_mode = 1;
+			break;
 		default:
-			exit(EX_USAGE);
+			exit(EX_NAGIOS_UNKNOWN);
 		}
 	}
 	argc -= optind;
 	argv += optind;
 
+	if (nagios_check_mode) {
+		if (argc != 0) {
+			abend("bad number of arguments");
+		}
+		if (!host)
+			abend("hostname missing; use -H option");
+		if (!c_opt)
+			abend("critical threshold missing; use -c option");
+		parse_nagios_threshold(c_opt, &chkdata.cth);
+
+		if (!w_opt)
+			abend("warning threshold missing; use -w option");
+		parse_nagios_threshold(w_opt, &chkdata.wth);
+	}
+	
 	p = node = read_listen(&service);
 	if (!node || !node[0])
 		node = DEFAULT_ADDRESS;
@@ -606,13 +717,17 @@ main(int argc, char **argv)
 	http_connect(node, service);
 	free(p);
 
-	switch (argc) {
-	case 0:
-		query_all();
-	case 1:
-		query_host(argv[0]);
-	default:
-		fatal("bad number of arguments");
+	if (nagios_check_mode)
+		query_host(host, nagios_check, &chkdata);
+	else {
+		switch (argc) {
+		case 0:
+			query_all();
+		case 1:
+			query_host(argv[0], print_host_status, NULL);
+		default:
+			abend("bad number of arguments");
+		}
 	}
-	exit(EX_USAGE);
+	exit(EX_NAGIOS_UNKNOWN);
 }