watchdog.c

watchdog source watchdog source

/* $Header: /cvsroot/watchdog/watchdog/src/watchdog.c,v 1.4 2007/08/17 09:24:54 meskes Exp $ */

/*************************************************************/
/* Original version was an example in the kernel source tree */
/*                                                           */
/* Most of the rest was written by me, Michael Meskes        */
/* [e-mail]                                         */
/*                                                           */
/*************************************************************/
#ifdef HAVE_CONFIG_H
#include "config.h"
#endif

#include "extern.h"

#include <errno.h>
#include <fcntl.h>
#include <getopt.h>
#include <netdb.h>
#include <sched.h>
#include <signal.h>
#include <stdlib.h>
#include <arpa/inet.h>
#include <sys/mman.h>
#include <sys/wait.h>
#define __USE_GNU
#include <string.h>

#if !defined(__GLIBC__)		/* __GLIBC__ */
extern char *basename(const char *);
#endif				/* __GLIBC__ */

#include <unistd.h>

#include "watch_err.h"

#if USE_SYSLOG
#include <syslog.h>
#endif				/* USE_SYSLOG */

static int no_act = FALSE;

#if USE_SYSLOG
int verbose = FALSE;
#endif				/* USE_SYSLOG */

#define ADMIN		"admin"
#define CHANGE		"change"
#define DEVICE		"watchdog-device"
#define	FILENAME	"file"
#define INTERFACE	"interface"
#define INTERVAL	"interval"
#define LOGTICK		"logtick"
#define MAXLOAD1	"max-load-1"
#define MAXLOAD5	"max-load-5"
#define MAXLOAD15	"max-load-15"
#define MAXTEMP		"max-temperature"
#define MINMEM		"min-memory"
#define SERVERPIDFILE	"pidfile"
#define PING		"ping"
#define PINGCOUNT	"ping-count"
#define PRIORITY	"priority"
#define REALTIME	"realtime"
#define REPAIRBIN	"repair-binary"
#define TEMP		"temperature-device"
#define TESTBIN		"test-binary"
#define TESTTIMEOUT	"test-timeout"
#define HEARTBEAT	"heartbeat-file"
#define HBSTAMPS	"heartbeat-stamps"

pid_t pid;
int softboot = FALSE, watchdog = -1, load = -1, mem = -1, temp = -1;
int tint = 10, logtick = 1, ticker = 1, schedprio = 1;
int maxload1 = 0, maxload5 = 0, maxload15 = 0, minpages = 0;
int maxtemp = 120, hbstamps = 300, lastts, nrts;
int pingcount = 3;
char *tempname = NULL, *devname = NULL, *admin = "root", *progname;
char *timestamps, *heartbeat;
time_t timeout = 0;
FILE *hb;

#if defined(_POSIX_MEMLOCK)
int mlocked = FALSE, realtime = FALSE;
#endif

static void usage(void)
{
    fprintf(stderr, "%s version %d.%d, usage:\n", progname, MAJOR_VERSION, MINOR_VERSION);
#if USE_SYSLOG
    fprintf(stderr, "%s [-i <interval> [-f]] [-l <max load avg>] [-v] [-s] [-b] [-m <max temperature>]\n", progname);
#else				/* USE_SYSLOG */
    fprintf(stderr, "%s [-i <interval> [-f]] [-l <max load avg>] [-v] [-b] [-m <max temperature>]\n", progname);
#endif				/* USE_SYSLOG */
    exit(1);
}

/* Try to sync */
static int sync_system(int sync_it)
{
    if (sync_it) {
	sync();
	sync();
    }
    return (0);
}

/* execute repair binary */
static int repair(char *rbinary, int result)
{
    pid_t child_pid;
    char parm[5];
    int ret;

    /* no binary given, we have to reboot */
    if (rbinary == NULL)
	return (result);

    sprintf(parm, "%d", result);

    child_pid = fork();
    if (!child_pid) {
	/* Don't want the stdin and stdout of our test program
	 * to cause trouble
	 * So make stdout and stderr go to their respective files */
        if (!freopen("/var/log/watchdog/repair-bin.stdout", "a+", stdout))
        	exit (errno);
        if (!freopen("/var/log/watchdog/repair-bin.stderr", "a+", stderr))
        	exit (errno);

        /* else start binary */
	execl(rbinary, rbinary, parm, NULL);

	/* execl should only return in case of an error */
	/* so we return the reboot code */
	return (errno);
    } else if (child_pid < 0) {	/* fork failed */
	int err = errno;

	if (errno == EAGAIN) {	/* process table full */
#if USE_SYSLOG
	    syslog(LOG_ERR, "process table is full!");
#endif				/* USE_SYSLOG */
	    return (EREBOOT);
	} else if (softboot)
	    return (err);
	else
	    return (ENOERR);
    }
    if (waitpid(child_pid, &result, 0) != child_pid) {
	int err = errno;

#if USE_SYSLOG
	syslog(LOG_ERR, "child %d does not exist (errno = %d = '%m')", child_pid, err);
#else				/* USE_SYSLOG */
	perror(progname);
#endif				/* USE_SYSLOG */
	if (softboot)
	    return (err);
    }
    
    /* check result */
    ret = WEXITSTATUS(result);
    if (ret != 0) {
#if USE_SYSLOG
	syslog(LOG_ERR, "repair binary returned %d", ret);
#endif				/* USE_SYSLOG */

	if (ret == ERESET) /* repair script says force hard reset, we give it a try */
		sleep(TIMER_MARGIN * 4);
	
	/* for all other errors or if we still live, we let shutdown handle it */
	return (ret);
    }
    
    return (ENOERR);
}

static void wd_action(int result, char *rbinary)
{
    /* if no-action flag set, do nothing */
    /* no error, keep on working */
    if (result == ENOERR || no_act == TRUE)
	return;

    /* error that might be repairable */
    if (result != EREBOOT)
	result = repair(rbinary, result);

    /* if still error, reboot */
    if (result != ENOERR)
	do_shutdown(result);

}

static void do_check(int res, char *rbinary)
{
    wd_action(res, rbinary);
    wd_action(keep_alive(), rbinary);
}

struct list *file = NULL, *target = NULL, *pidfile = NULL, *iface = NULL;
char *tbinary, *rbinary, *admin;

static void add_list(struct list **list, char *name)
{
    struct list *new, *act;

    if ((new = (struct list *) calloc(1, sizeof(struct list))) == NULL) {
	fprintf(stderr, "%s: out of memory\n", progname);
	exit(1);
    }
    new->name = name;
    memset((char *) (&(new->parameter)), '\0', sizeof(union wdog_options));

    if (*list == NULL)
	*list = new;
    else {
	for (act = *list; act->next != NULL; act = act->next);
	act->next = new;
    }
}

static int spool(char *line, int *i, int offset)
{
    for ((*i) += offset; line[*i] == ' ' || line[*i] == '\t'; (*i)++);
    if (line[*i] == '=')
	(*i)++;
    for (; line[*i] == ' ' || line[*i] == '\t'; (*i)++);
    if (line[*i] == '\0')
    	return(1);
    else
    	return(0);
}

static void read_config(char *filename, char *progname)
{
    FILE *wc;
    int gotload5 = FALSE, gotload15 = FALSE;

    if ((wc = fopen(filename, "r")) == NULL) {
	perror(progname);
	exit(1);
    }

    while (!feof(wc)) {
	char line[CONFIG_LINE_LEN];

	if (fgets(line, CONFIG_LINE_LEN, wc) == NULL) {
	    if (!ferror(wc))
		break;
	    else {
		perror(progname);
		exit(1);
	    }
	} else {
	    int i, j;

	    /* scan the actual line for an option */
	    /* first remove the leading blanks */
	    for (i = 0; line[i] == ' ' || line[i] == '\t'; i++);

	    /* if the next sign is a '#' we have a comment */
	    if (line[i] == '#')
		continue;

	    /* also remove the trailing blanks and the \n */
	    for (j = strlen(line) - 1; line[j] == ' ' || line[j] == '\t' || line[j] == '\n'; j--);
	    line[j + 1] = '\0';

	    /* if the line is empty now, we don't have to parse it */
	    if (strlen(line + i) == 0)
		continue;

	    /* now check for an option */
	    if (strncmp(line + i, FILENAME, strlen(FILENAME)) == 0) {
		if (spool(line, &i, strlen(FILENAME)))
		    fprintf(stderr, "Ignoring invalid line in config file:\n%s\n", line);
		else
		    add_list(&file, strdup(line + i));
	    } else if (strncmp(line + i, CHANGE, strlen(CHANGE)) == 0) {
		struct list *ptr;

		if (spool(line, &i, strlen(CHANGE)))
		    continue;

		if (!file) {	/* no file entered yet */
		    fprintf(stderr, "Ignoring invalid line in config file:\n%s\n", line);
		    continue;
		}
		for (ptr = file; ptr->next != NULL; ptr = ptr->next);
		if (ptr->parameter.file.mtime != 0)
		    fprintf(stderr, "Duplicate change interval option in config file. Ignoring first entry.\n");

		ptr->parameter.file.mtime = atoi(line + i);
	    } else if (strncmp(line + i, SERVERPIDFILE, strlen(SERVERPIDFILE)) == 0) {
		if (spool(line, &i, strlen(SERVERPIDFILE)))
		    fprintf(stderr, "Ignoring invalid line in config file:\n%s\n", line);
		else
		    add_list(&pidfile, strdup(line + i));
	    } else if (strncmp(line + i, PINGCOUNT, strlen(PINGCOUNT)) == 0) {
		if (spool(line, &i, strlen(PINGCOUNT)))
		    fprintf(stderr, "Ignoring invalid line in config file:\n%s\n", line);
		else
		    pingcount = atol(line + i);
	    } else if (strncmp(line + i, PING, strlen(PING)) == 0) {
		if (spool(line, &i, strlen(PING)))
		    fprintf(stderr, "Ignoring invalid line in config file:\n%s\n", line);
		else
		    add_list(&target, strdup(line + i));
	    } else if (strncmp(line + i, INTERFACE, strlen(INTERFACE)) == 0) {
		if (spool(line, &i, strlen(INTERFACE)))
		    fprintf(stderr, "Ignoring invalid line in config file:\n%s\n", line);
		else
		    add_list(&iface, strdup(line + i));
	    } else if (strncmp(line + i, REALTIME, strlen(REALTIME)) == 0) {
		(void)spool(line, &i, strlen(REALTIME));
		realtime = (strncmp(line + i, "yes", 3) == 0) ? TRUE : FALSE;
	    } else if (strncmp(line + i, PRIORITY, strlen(PRIORITY)) == 0) {
		if (spool(line, &i, strlen(PRIORITY)))
		    fprintf(stderr, "Ignoring invalid line in config file:\n%s\n", line);
		else
		    schedprio = atol(line + i);
	    } else if (strncmp(line + i, REPAIRBIN, strlen(REPAIRBIN)) == 0) {
		if (spool(line, &i, strlen(REPAIRBIN)))
			rbinary = NULL;
		else
			rbinary = strdup(line + i);
	    } else if (strncmp(line + i, TESTBIN, strlen(TESTBIN)) == 0) {
		if (spool(line, &i, strlen(TESTBIN)))
			tbinary = NULL;
		else
			tbinary = strdup(line + i);
	    } else if (strncmp(line + i, TESTTIMEOUT, strlen(TESTTIMEOUT)) == 0) {
		if (spool(line, &i, strlen(TESTTIMEOUT)))
			timeout = 0;
		else
			timeout = atol(line + i);
	    } else if (strncmp(line + i, HEARTBEAT, strlen(HEARTBEAT)) == 0) {
		if (spool(line, &i, strlen(HEARTBEAT)))
			heartbeat = NULL;
		else
			heartbeat = strdup(line + i);
	    } else if (strncmp(line + i, HBSTAMPS, strlen(HBSTAMPS)) == 0) {
		if (spool(line, &i, strlen(HBSTAMPS)))
		        fprintf(stderr, "Ignoring invalid line in config file:\n%s\n", line);
		else
			hbstamps = atol(line + i);
	    } else if (strncmp(line + i, ADMIN, strlen(ADMIN)) == 0) {
		if (spool(line, &i, strlen(ADMIN)))
			admin = NULL;
		else
			admin = strdup(line + i);
	    } else if (strncmp(line + i, INTERVAL, strlen(INTERVAL)) == 0) {
		if (spool(line, &i, strlen(INTERVAL)))
		    fprintf(stderr, "Ignoring invalid line in config file:\n%s\n", line);
		else
		    tint = atol(line + i);
	    } else if (strncmp(line + i, LOGTICK, strlen(LOGTICK)) == 0) {
		if (spool(line, &i, strlen(LOGTICK)))
		    logtick = ticker = 1;
		else
		    logtick = ticker = atol(line + i);
	    } else if (strncmp(line + i, DEVICE, strlen(DEVICE)) == 0) {
		if (spool(line, &i, strlen(DEVICE)))
			devname = NULL;
		else
			devname = strdup(line + i);
	    } else if (strncmp(line + i, TEMP, strlen(TEMP)) == 0) {
		if (spool(line, &i, strlen(TEMP)))
			tempname = NULL;
		else
			tempname = strdup(line + i);
	    } else if (strncmp(line + i, MAXTEMP, strlen(MAXTEMP)) == 0) {
		if (spool(line, &i, strlen(MAXTEMP)))
			fprintf(stderr, "Ignoring invalid line in config file:\n%s\n", line);
		else
			maxtemp = atol(line + i);
	    } else if (strncmp(line + i, MAXLOAD15, strlen(MAXLOAD15)) == 0) {
		if (spool(line, &i, strlen(MAXLOAD15)))
			fprintf(stderr, "Ignoring invalid line in config file:\n%s\n", line);
		else {
			maxload15 = atol(line + i);
			gotload15 = TRUE;
		}
	    } else if (strncmp(line + i, MAXLOAD1, strlen(MAXLOAD1)) == 0) {
		if (spool(line, &i, strlen(MAXLOAD1)))
			fprintf(stderr, "Ignoring invalid line in config file:\n%s\n", line);
		else {
			maxload1 = atol(line + i);
			if (!gotload5)
			    maxload5 = maxload1 * 3 / 4;
			if (!gotload15)
			    maxload15 = maxload1 / 2;
		}
	    } else if (strncmp(line + i, MAXLOAD5, strlen(MAXLOAD5)) == 0) {
		if (spool(line, &i, strlen(MAXLOAD5)))
			fprintf(stderr, "Ignoring invalid line in config file:\n%s\n", line);
		else {
			maxload5 = atol(line + i);
			gotload5 = TRUE;
		}
	    } else if (strncmp(line + i, MINMEM, strlen(MINMEM)) == 0) {
		if (spool(line, &i, strlen(MINMEM)))
			fprintf(stderr, "Ignoring invalid line in config file:\n%s\n", line);
		else {
			minpages = atol(line + i);
		}
	    } else {
		fprintf(stderr, "Ignoring invalid line in config file:\n%s\n", line);
	    }
	}
    }

    if (fclose(wc) != 0) {
	perror(progname);
	exit(1);
    }
}

static void old_option(int c, char *filename)
{
    fprintf(stderr, "Option -%c is no longer valid, please specify it in %s.\n", c, filename);
    usage();
}