From c1d25fe1ee355f3394cab0bd340896df84e9d549 Mon Sep 17 00:00:00 2001 From: Stanley Hopcroft Date: Thu, 2 Dec 2004 08:44:36 +0000 Subject: check_icmp plugin from A Ericsson git-svn-id: https://nagiosplug.svn.sourceforge.net/svnroot/nagiosplug/nagiosplug/trunk@975 f882894a-f735-0410-b71e-b25c423dba1c diff --git a/plugins/check_icmp.c b/plugins/check_icmp.c new file mode 100644 index 0000000..45bfcc5 --- /dev/null +++ b/plugins/check_icmp.c @@ -0,0 +1,1381 @@ +/* + * check_icmp - A hack of fping2 to work with nagios. + * This way we don't have to use the output parser. + * + * VIEWING NOTES: + * This file was formatted with tab indents at a tab stop of 4. + * + * It is highly recommended that your editor is set to this + * tab stop setting for viewing and editing. + * + * COPYLEFT; + * This programs copyright status is currently undetermined. Much of + * the code in it comes from the fping2 program which used to be licensed + * under the Stanford General Software License (available at + * http://graphics.stanford.edu/software/license.html). It is presently + * unclear what license (if any) applies to the original code at the + * moment. + * + * The fping website can be found at http://www.fping.com + */ + +#include +#include +#include +#include + +#include + +#include + +#include +#include + +#include +#include +#include + +#include + +#include +#include + +/* Linux has bizarre ip.h and ip_icmp.h */ +/* Taken from the fping distro. Thank you. */ +#if defined( __linux__ ) +#include "linux.h" +#else +#include +#include +#endif /* defined(__linux__) */ + +#include +#include + +/* RS6000 has sys/select.h */ +#ifdef HAVE_SYS_SELECT_H +#include +#endif /* HAVE_SYS_SELECT_H */ + +/* rta threshold values can't be larger than MAXTTL seconds */ +#ifndef MAXTTL +# define MAXTTL 255 +#endif +#ifndef IPDEFTTL +# define IPDEFTTL 64 +#endif + +/*** externals ***/ +extern char *optarg; +extern int optind, opterr; + +/*** Constants ***/ +#define REV_DATE "2004-09-06" +#define EMAIL "ae@op5.se" +#define VERSION "0.8" + +#ifndef INADDR_NONE +# define INADDR_NONE 0xffffffU +#endif + +/*** Ping packet defines ***/ +/* data added after ICMP header for our nefarious purposes */ +typedef struct ping_data { + unsigned int ping_count; /* counts up to -[n|p] count or 1 */ + struct timeval ping_ts; /* time sent */ +} PING_DATA; + +#define MIN_PING_DATA sizeof(PING_DATA) +#define MAX_IP_PACKET 65536 /* (theoretical) max IP packet size */ +#define SIZE_IP_HDR 20 +#define SIZE_ICMP_HDR ICMP_MINLEN /* from ip_icmp.h */ +#define MAX_PING_DATA (MAX_IP_PACKET - SIZE_IP_HDR - SIZE_ICMP_HDR) + +/* + * Interval is the minimum amount of time between sending a ping packet to + * any host. + * + * Perhost_interval is the minimum amount of time between sending a ping + * packet to a particular responding host + * + * Timeout is the initial amount of time between sending a ping packet to + * a particular non-responding host. + * + * Retry is the number of ping packets to send to a non-responding host + * before giving up (in is-it-alive mode). + * + * Backoff factor is how much longer to wait on successive retries. + */ +#ifndef DEFAULT_INTERVAL +#define DEFAULT_INTERVAL 25 /* default time between packets (msec) */ +#endif + +#ifndef DEFAULT_RETRY +#define DEFAULT_RETRY 1 /* number of times to retry a host */ +#endif + +#ifndef DEFAULT_TIMEOUT +# define DEFAULT_TIMEOUT 1000 +#endif + +#ifndef DEFAULT_BACKOFF_FACTOR +#define DEFAULT_BACKOFF_FACTOR 1.5 /* exponential timeout factor */ +#endif +#define MIN_BACKOFF_FACTOR 1.0 /* exponential timeout factor */ +#define MAX_BACKOFF_FACTOR 5.0 /* exponential timeout factor */ + +#ifndef DNS_TIMEOUT +#define DNS_TIMEOUT 1000 /* time in usec for dns retry */ +#endif + +#ifndef MAX_RTA_THRESHOLD_VALUE +# define MAX_RTA_THRESHOLD_VALUE 120*1000000 /* 2 minutes should be enough */ +#endif +#ifndef MIN_RTA_THRESHOLD_VALUE +# define MIN_RTA_THRESHOLD_VALUE 10000 /* minimum RTA threshold value */ +#endif + +/* sized so as to be like traditional ping */ +#define DEFAULT_PING_DATA_SIZE (MIN_PING_DATA + 44) + +/* maxima and minima */ +#define MAX_COUNT 50 /* max count even if we're root */ +#define MAX_RETRY 5 +#define MIN_INTERVAL 25 /* msecs */ +#define MIN_TIMEOUT 50 /* msecs */ + +/* response time array flags */ +#define RESP_WAITING -1 +#define RESP_UNUSED -2 + +#define ICMP_UNREACH_MAXTYPE 15 + +/* entry used to keep track of each host we are pinging */ +struct host_entry { + int i; /* index into array */ + char *name; /* name as given by user */ + char *host; /* text description of host */ + struct sockaddr_in saddr; /* internet address */ + unsigned short **pr; /* TCP port range to check for connectivity */ + struct timeval last_send_time; /* time of last packet sent */ + unsigned int num_sent; /* number of ping packets sent */ + unsigned int num_recv; /* number of pings received */ + unsigned int total_time; /* sum of response times */ + unsigned int status; /* this hosts status */ + unsigned int running; /* unset when through sending */ + unsigned int waiting; /* waiting for response */ + int *resp_times; /* individual response times */ + struct host_entry *prev, *next; /* doubly linked list */ +}; + +typedef struct host_entry HOST_ENTRY; + +struct host_name_list { + char *entry; + struct host_name_list *next; +}; + +/* threshold structure */ +struct threshold { + unsigned int pl; /* packet loss */ + unsigned int rta; /* roundtrip time average */ +}; +typedef struct threshold threshold; + +/***************************************************************************** + * Global Variables * + *****************************************************************************/ + +HOST_ENTRY *rrlist = NULL; /* linked list of hosts be pinged */ +HOST_ENTRY **table = NULL; /* array of pointers to items in the list */ +HOST_ENTRY *cursor; + +char *prog; /* our name */ +int ident; /* our pid, for marking icmp packets */ +int sock; /* socket */ +u_int debug = 0; + +/* threshold value defaults; + * WARNING; 60% packetloss or 200 msecs round trip average + * CRITICAL; 80% packetloss or 500 msecs round trip average */ +threshold warn = {60, 200 * 1000}; +threshold crit = {80, 500 * 1000}; + +/* times get *100 because all times are calculated in 10 usec units, not ms */ +unsigned int retry = DEFAULT_RETRY; +u_int timeout = DEFAULT_TIMEOUT * 100; +u_int interval = DEFAULT_INTERVAL * 100; +float backoff = DEFAULT_BACKOFF_FACTOR; +u_int select_time; /* calculated using maximum threshold rta value */ +u_int ping_data_size = DEFAULT_PING_DATA_SIZE; +u_int ping_pkt_size; +unsigned int count = 5; +unsigned int trials = 1; + +/* global stats */ +int total_replies = 0; +int num_jobs = 0; /* number of hosts still to do */ +int num_hosts = 0; /* total number of hosts */ +int num_alive = 0; /* total number alive */ +int num_unreachable = 0; /* total number unreachable */ +int num_noaddress = 0; /* total number of addresses not found */ +int num_timeout = 0; /* number of timed out packets */ +int num_pingsent = 0; /* total pings sent */ +int num_pingreceived = 0; /* total pings received */ +int num_othericmprcvd = 0; /* total non-echo-reply ICMP received */ + +struct timeval current_time; /* current time (pseudo) */ +struct timeval start_time; +struct timeval end_time; +struct timeval last_send_time; /* time last ping was sent */ +struct timezone tz; + +/* switches */ +int generate_flag = 0; /* flag for IP list generation */ +int stats_flag, unreachable_flag, alive_flag; +int elapsed_flag, version_flag, count_flag; +int name_flag, addr_flag, backoff_flag; +int multif_flag; + +/*** prototypes ***/ +void add_name(char *); +void add_addr(char *, char *, struct in_addr); +char *na_cat(char *, struct in_addr); +char *cpystr(char *); +void crash(char *); +char *get_host_by_address(struct in_addr); +int in_cksum(u_short *, int); +void u_sleep(int); +int recvfrom_wto(int, char *, int, struct sockaddr *, int); +void remove_job(HOST_ENTRY *); +void send_ping(int, HOST_ENTRY *); +long timeval_diff(struct timeval *, struct timeval *); +void usage(void); +int wait_for_reply(int); +void finish(void); +int handle_random_icmp(struct icmp *, struct sockaddr_in *); +char *sprint_tm(int); +int get_threshold(char *, threshold *); + +/*** the various exit-states */ +enum { + STATE_OK = 0, + STATE_WARNING, + STATE_CRITICAL, + STATE_UNKNOWN, + STATE_DEPENDANT, + STATE_OOB +}; +/* the strings that correspond to them */ +char *status_string[STATE_OOB] = { + "OK", + "WARNING", + "CRITICAL", + "UNKNOWN", + "DEPENDANT" +}; + +int status = STATE_OK; +int fin_stat = STATE_OK; + +/***************************************************************************** + * Code block start * + *****************************************************************************/ +int main(int argc, char **argv) +{ + int c; + u_int lt, ht; + int advance; + struct protoent *proto; + uid_t uid; + struct host_name_list *host_ptr, *host_base_ptr; + + if(strchr(argv[0], '/')) prog = strrchr(argv[0], '/') + 1; + else prog = argv[0]; + + /* check if we are root */ + if(geteuid()) { + printf("Root access needed (for raw sockets)\n"); + exit(STATE_UNKNOWN); + } + + /* confirm that ICMP is available on this machine */ + if((proto = getprotobyname("icmp")) == NULL) + crash("icmp: unknown protocol"); + + /* create raw socket for ICMP calls (ping) */ + sock = socket(AF_INET, SOCK_RAW, proto->p_proto); + + if(sock < 0) + crash("can't create raw socket"); + + /* drop privileges now that we have the socket */ + if((uid = getuid())) { + seteuid(uid); + } + + if(argc < 2) usage(); + + ident = getpid() & 0xFFFF; + + if(!(host_base_ptr = malloc(sizeof(struct host_name_list)))) { + crash("Unable to allocate memory for host name list\n"); + } + host_ptr = host_base_ptr; + + backoff_flag = 0; + opterr = 1; + + /* get command line options + * -H denotes a host (actually ignored and picked up later) + * -h for help + * -V or -v for version + * -d to display hostnames rather than addresses + * -t sets timeout for packets and tcp connects + * -r defines retries (persistence) + * -p or -n sets packet count (5) + * -b sets packet size (56) + * -w sets warning threshhold (200,40%) + * -c sets critical threshhold (500,80%) + * -i sets interval for both packet transmissions and connect attempts + */ +#define OPT_STR "amH:hvVDdAp:n:b:r:t:i:w:c:" + while((c = getopt(argc, argv, OPT_STR)) != EOF) { + switch (c) { + case 'H': + if(!(host_ptr->entry = malloc(strlen(optarg) + 1))) { + crash("Failed to allocate memory for hostname"); + } + memset(host_ptr->entry, 0, strlen(optarg) + 1); + host_ptr->entry = memcpy(host_ptr->entry, optarg, strlen(optarg)); + if(!(host_ptr->next = malloc(sizeof(struct host_name_list)))) + crash("Failed to allocate memory for hostname"); + host_ptr = host_ptr->next; + host_ptr->next = NULL; +// add_name(optarg); + break; + /* this is recognized, but silently ignored. + * host(s) are added later on */ + + break; + case 'w': + if(get_threshold(optarg, &warn)) { + printf("Illegal threshold pair specified for -%c", c); + usage(); + } + break; + + case 'c': + if(get_threshold(optarg, &crit)) { + printf("Illegal threshold pair specified for -%c", c); + usage(); + } + break; + + case 't': + if(!(timeout = (u_int) strtoul(optarg, NULL, 0) * 100)) { + printf("option -%c requires integer argument\n", c); + usage(); + } + break; + + case 'r': + if(!(retry = (u_int) strtoul(optarg, NULL, 0))) { + printf("option -%c requires integer argument\n", c); + usage(); + } + break; + + case 'i': + if(!(interval = (u_int) strtoul(optarg, NULL, 0) * 100)) { + printf("option -%c requires positive non-zero integer argument\n", c); + usage(); + } + break; + + case 'p': + case 'n': + if(!(count = (u_int) strtoul(optarg, NULL, 0))) { + printf("option -%c requires positive non-zero integer argument\n", c); + usage(); + } + break; + + case 'b': + if(!(ping_data_size = (u_int) strtoul(optarg, NULL, 0))) { + printf("option -%c requires integer argument\n", c); + usage(); + } + break; + + case 'h': + usage(); + break; + + case 'e': + elapsed_flag = 1; + break; + + case 'm': + multif_flag = 1; + break; + + case 'd': + name_flag = 1; + break; + + case 'A': + addr_flag = 1; + break; + + case 's': + stats_flag = 1; + break; + + case 'u': + unreachable_flag = 1; + break; + + case 'a': + alive_flag = 1; + break; + + case 'v': + printf("%s: Version %s $Date$\n", prog, VERSION, REV_DATE); + printf("%s: comments to %s\n", prog, EMAIL); + exit(STATE_OK); + + case 'g': + /* use IP list generation */ + /* mutex with file input or command line targets */ + generate_flag = 1; + break; + + default: + printf("option flag -%c specified, but not recognized\n", c); + usage(); + break; + } + } + + /* arguments are parsed, so now we validate them */ + + if(count > 1) count_flag = 1; + + /* set threshold values to 10usec units (inherited from fping.c) */ + crit.rta = crit.rta / 10; + warn.rta = warn.rta / 10; + select_time = crit.rta; + /* this isn't critical, but will most likely not be what the user expects + * so we tell him/her about it, but keep running anyways */ + if(warn.pl > crit.pl || warn.rta > crit.rta) { + select_time = warn.rta; + printf("(WARNING threshold > CRITICAL threshold) :: "); + fflush(stdout); + } + + /* A timeout smaller than maximum rta threshold makes no sense */ + if(timeout < crit.rta) timeout = crit.rta; + else if(timeout < warn.rta) timeout = warn.rta; + + if((interval < MIN_INTERVAL * 100 || retry > MAX_RETRY) && getuid()) { + printf("%s: these options are too risky for mere mortals.\n", prog); + printf("%s: You need i >= %u and r < %u\n", + prog, MIN_INTERVAL, MAX_RETRY); + printf("Current settings; i = %d, r = %d\n", + interval / 100, retry); + usage(); + } + + if((ping_data_size > MAX_PING_DATA) || (ping_data_size < MIN_PING_DATA)) { + printf("%s: data size %u not valid, must be between %u and %u\n", + prog, ping_data_size, MIN_PING_DATA, MAX_PING_DATA); + usage(); + + } + + if((backoff > MAX_BACKOFF_FACTOR) || (backoff < MIN_BACKOFF_FACTOR)) { + printf("%s: backoff factor %.1f not valid, must be between %.1f and %.1f\n", + prog, backoff, MIN_BACKOFF_FACTOR, MAX_BACKOFF_FACTOR); + usage(); + + } + + if(count > MAX_COUNT) { + printf("%s: count %u not valid, must be less than %u\n", + prog, count, MAX_COUNT); + usage(); + } + + if(count_flag) { + alive_flag = unreachable_flag = 0; + } + + trials = (count > retry + 1) ? count : retry + 1; + + /* handle host names supplied on command line or in a file */ + /* if the generate_flag is on, then generate the IP list */ + argv = &argv[optind]; + + /* cover allowable conditions */ + + /* generate requires command line parameters beyond the switches */ + if(generate_flag && !*argv) { + printf("generate flag requires command line parameters beyond switches\n"); + usage(); + } + + if(*argv && !generate_flag) { + while(*argv) { + if(!(host_ptr->entry = malloc(strlen(*argv) + 1))) { + crash("Failed to allocate memory for hostname"); + } + memset(host_ptr->entry, 0, strlen(*argv) + 1); + host_ptr->entry = memcpy(host_ptr->entry, *argv, strlen(*argv)); + if(!(host_ptr->next = malloc(sizeof(struct host_name_list)))) + crash("Failed to allocate memory for hostname"); + host_ptr = host_ptr->next; + host_ptr->next = NULL; + +// add_name(*argv); + argv++; + } + } + + // now add all the hosts + host_ptr = host_base_ptr; + while(host_ptr->next) { + add_name(host_ptr->entry); + host_ptr = host_ptr->next; + } + + if(!num_hosts) { + printf("No hosts to work with!\n\n"); + usage(); + } + + /* allocate array to hold outstanding ping requests */ + table = (HOST_ENTRY **) malloc(sizeof(HOST_ENTRY *) * num_hosts); + if(!table) crash("Can't malloc array of hosts"); + + cursor = rrlist; + + for(num_jobs = 0; num_jobs < num_hosts; num_jobs++) { + table[num_jobs] = cursor; + cursor->i = num_jobs; + + cursor = cursor->next; + } /* FOR */ + + ping_pkt_size = ping_data_size + SIZE_ICMP_HDR; + + signal(SIGINT, (void *)finish); + + gettimeofday(&start_time, &tz); + current_time = start_time; + + last_send_time.tv_sec = current_time.tv_sec - 10000; + + cursor = rrlist; + advance = 0; + + /* main loop */ + while(num_jobs) { + /* fetch all packets that receive within time boundaries */ + while(num_pingsent && + cursor && + cursor->num_sent > cursor->num_recv && + wait_for_reply(sock)) ; + + if(cursor && advance) { + cursor = cursor->next; + } + + gettimeofday(¤t_time, &tz); + lt = timeval_diff(¤t_time, &last_send_time); + ht = timeval_diff(¤t_time, &cursor->last_send_time); + + advance = 1; + + /* if it's OK to send while counting or looping or starting */ + if(lt > interval) { + /* send if starting or looping */ + if((cursor->num_sent == 0)) { + send_ping(sock, cursor); + continue; + } /* IF */ + + /* send if counting and count not exceeded */ + if(count_flag) { + if(cursor->num_sent < count) { + send_ping(sock, cursor); + continue; + } /* IF */ + } /* IF */ + } /* IF */ + + /* is-it-alive mode, and timeout exceeded while waiting for a reply */ + /* and we haven't exceeded our retries */ + if((lt > interval) && !count_flag && !cursor->num_recv && + (ht > timeout) && (cursor->waiting < retry + 1)) { + num_timeout++; + + /* try again */ + send_ping(sock, cursor); + continue; + } /* IF */ + + /* didn't send, can we remove? */ + + /* remove if counting and count exceeded */ + if(count_flag) { + if((cursor->num_sent >= count)) { + remove_job(cursor); + continue; + } /* IF */ + } /* IF */ + else { + /* normal mode, and we got one */ + if(cursor->num_recv) { + remove_job(cursor); + continue; + } /* IF */ + + /* normal mode, and timeout exceeded while waiting for a reply */ + /* and we've run out of retries, so node is unreachable */ + if((ht > timeout) && (cursor->waiting >= retry + 1)) { + num_timeout++; + remove_job(cursor); + continue; + + } /* IF */ + } /* ELSE */ + + /* could send to this host, so keep considering it */ + if(ht > interval) { + advance = 0; + } + } /* WHILE */ + + finish(); + return 0; +} /* main() */ + +/************************************************************ + * Description: + * + * Main program clean up and exit point + ************************************************************/ +void finish() +{ + int i; + HOST_ENTRY *h; + + gettimeofday(&end_time, &tz); + + /* tot up unreachables */ + for(i=0; inum_recv) { + num_unreachable++; + status = fin_stat = STATE_CRITICAL; + if(num_hosts == 1) { + printf("CRITICAL - %s is down (lost 100%%)|" + "rta=;%d;%d;; pl=100%%;%d;%d;;\n", + h->host, + warn.rta / 100, crit.rta / 100, + warn.pl, crit.pl); + } + else { + printf("%s is down (lost 100%%)", h->host); + } + } + else { + /* reset the status */ + status = STATE_OK; + + /* check for warning before critical, for debugging purposes */ + if(warn.rta <= h->total_time / h->num_recv) { +/* printf("warn.rta exceeded\n"); +*/ status = STATE_WARNING; + } + if(warn.pl <= ((h->num_sent - h->num_recv) * 100) / h->num_sent) { +/* printf("warn.pl exceeded (pl=%d)\n", + ((h->num_sent - h->num_recv) * 100) / h->num_sent); +*/ status = STATE_WARNING; + } + if(crit.rta <= h->total_time / h->num_recv) { +/* printf("crit.rta exceeded\n"); +*/ status = STATE_CRITICAL; + } + if(crit.pl <= ((h->num_sent - h->num_recv) * 100) / h->num_sent) { +/* printf("crit.pl exceeded (pl=%d)\n", + ((h->num_sent - h->num_recv) * 100) / h->num_sent); +*/ status = STATE_CRITICAL; + } + + if(num_hosts == 1 || status != STATE_OK) { + printf("%s - %s: rta %s ms, lost %d%%", + status_string[status], h->host, + sprint_tm(h->total_time / h->num_recv), + h->num_sent > 0 ? ((h->num_sent - h->num_recv) * 100) / h->num_sent : 0 + ); + /* perfdata only available for single-host stuff */ + if(num_hosts == 1) { + printf("|rta=%sms;%d;%d;; pl=%d%%;%d;%d;;\n", + sprint_tm(h->total_time / h->num_recv), warn.rta / 100, crit.rta / 100, + h->num_sent > 0 ? ((h->num_sent - h->num_recv) * 100) / h->num_sent : 0, warn.pl, crit.pl + ); + } + else printf(" :: "); + } + + /* fin_stat should always hold the WORST state */ + if(fin_stat != STATE_CRITICAL && status != STATE_OK) { + fin_stat = status; + } + } + } + + if(num_noaddress) { + printf("No hostaddress specified.\n"); + usage(); + } + else if(num_alive != num_hosts) { + /* for future multi-check support */ + /*printf("num_alive != num_hosts (%d : %d)\n", num_alive, num_hosts);*/ + fin_stat = STATE_CRITICAL; + } + + if(num_hosts > 1) { + if(num_alive == num_hosts) { + printf("OK - All %d hosts are alive\n", num_hosts); + } + else { + printf("CRITICAL - %d of %d hosts are alive\n", num_alive, num_hosts); + } + } + exit(fin_stat); +} + + +void send_ping(int lsock, HOST_ENTRY *h) +{ + char *buffer; + struct icmp *icp; + PING_DATA *pdp; + int n; + + buffer = (char *)malloc((size_t) ping_pkt_size); + if(!buffer) + crash("can't malloc ping packet"); + + memset(buffer, 0, ping_pkt_size * sizeof(char)); + icp = (struct icmp *)buffer; + + gettimeofday(&h->last_send_time, &tz); + + icp->icmp_type = ICMP_ECHO; + icp->icmp_code = 0; + icp->icmp_cksum = 0; + icp->icmp_seq = h->i; + icp->icmp_id = ident; + + pdp = (PING_DATA *) (buffer + SIZE_ICMP_HDR); + pdp->ping_ts = h->last_send_time; + pdp->ping_count = h->num_sent; + + icp->icmp_cksum = in_cksum((u_short *) icp, ping_pkt_size); + + n = sendto(lsock, buffer, ping_pkt_size, 0, + (struct sockaddr *)&h->saddr, sizeof(struct sockaddr_in)); + + if(n < 0 || (unsigned int)n != ping_pkt_size) { + if(unreachable_flag) { + printf("%s error while sending ping: %s\n", + h->host, strerror(errno)); + } /* IF */ + + num_unreachable++; + remove_job(h); + } /* IF */ + else { + /* mark this trial as outstanding */ + h->resp_times[h->num_sent] = RESP_WAITING; + + h->num_sent++; + h->waiting++; + num_pingsent++; + last_send_time = h->last_send_time; + } /* ELSE */ + + free(buffer); +} /* send_ping() */ + +int wait_for_reply(int lsock) +{ + int result; + static char buffer[4096]; + struct sockaddr_in response_addr; + struct ip *ip; + int hlen; + struct icmp *icp; + int n; + HOST_ENTRY *h = NULL; + long this_reply; + int this_count; + struct timeval sent_time; + + result = recvfrom_wto(lsock, buffer, sizeof(buffer), + (struct sockaddr *)&response_addr, select_time); + + if(result < 0) return 0; /* timeout */ + + ip = (struct ip *)buffer; + +#if defined( __alpha__ ) && __STDC__ && !defined( __GLIBC__ ) + /* The alpha headers are decidedly broken. + * Using an ANSI compiler, it provides ip_vhl instead of ip_hl and + * ip_v. So, to get ip_hl, we mask off the bottom four bits. + */ + hlen = (ip->ip_vhl & 0x0F) << 2; +#else + hlen = ip->ip_hl << 2; +#endif /* defined(__alpha__) && __STDC__ */ + + if(result < hlen + ICMP_MINLEN) { + printf("received packet too short for ICMP (%d bytes from %s)\n", result, + inet_ntoa(response_addr.sin_addr)); + + return (1); /* too short */ + } /* IF */ + + icp = (struct icmp *)(buffer + hlen); + if(icp->icmp_type != ICMP_ECHOREPLY) { + /* handle some problem */ + if(handle_random_icmp(icp, &response_addr)) + num_othericmprcvd++; + + return 1; + } /* IF */ + + if(icp->icmp_id != ident) + return 1; /* packet received, but not the one we are looking for! */ + + num_pingreceived++; + + if(icp->icmp_seq >= (n_short) num_hosts) + return(1); /* packet received, don't worry about it anymore */ + + n = icp->icmp_seq; + h = table[n]; + + /* received ping is cool, so process it */ + + gettimeofday(¤t_time, &tz); + h->waiting = 0; + h->num_recv++; + + memcpy(&sent_time, icp->icmp_data + offsetof(PING_DATA, ping_ts), + sizeof(sent_time)); + memcpy(&this_count, icp->icmp_data, sizeof(this_count)); + + this_reply = timeval_diff(¤t_time, &sent_time); + h->total_time += this_reply; + total_replies++; + + /* note reply time in array, probably */ + if((this_count >= 0) && ((unsigned int)this_count < trials)) { + if(h->resp_times[this_count] != RESP_WAITING) { + printf("%s : duplicate for [%d], %d bytes, %s ms", + h->host, this_count, result, sprint_tm(this_reply)); + + if(response_addr.sin_addr.s_addr != h->saddr.sin_addr.s_addr) + printf(" [<- %s]\n", inet_ntoa(response_addr.sin_addr)); + } /* IF */ + else h->resp_times[this_count] = this_reply; + } /* IF */ + else { + /* count is out of bounds?? */ + printf("%s : duplicate for [%d], %d bytes, %s ms\n", + h->host, this_count, result, sprint_tm(this_reply)); + } /* ELSE */ + + if(h->num_recv == 1) { + num_alive++; + } /* IF */ + + return num_jobs; +} /* wait_for_reply() */ + +int handle_random_icmp(struct icmp *p, struct sockaddr_in *addr) +{ + struct icmp *sent_icmp; + u_char *c; + HOST_ENTRY *h; + + c = (u_char *) p; + switch (p->icmp_type) { + case ICMP_UNREACH: + sent_icmp = (struct icmp *)(c + 28); + + if((sent_icmp->icmp_type == ICMP_ECHO) && + (sent_icmp->icmp_id == ident) && + (sent_icmp->icmp_seq < (n_short) num_hosts)) { + /* this is a response to a ping we sent */ + h = table[sent_icmp->icmp_seq]; + + if(p->icmp_code > ICMP_UNREACH_MAXTYPE) { + printf("ICMP Unreachable (Invalid Code) from %s for ICMP Echo sent to %s", + inet_ntoa(addr->sin_addr), h->host); + + } /* IF */ + else { + printf("ICMP Unreachable from %s for ICMP Echo sent to %s", + inet_ntoa(addr->sin_addr), h->host); + + } /* ELSE */ + + if(inet_addr(h->host) == INADDR_NONE) + printf(" (%s)", inet_ntoa(h->saddr.sin_addr)); + + printf("\n"); + + } /* IF */ + + return 1; + + case ICMP_SOURCEQUENCH: + case ICMP_REDIRECT: + case ICMP_TIMXCEED: + case ICMP_PARAMPROB: + sent_icmp = (struct icmp *)(c + 28); + if((sent_icmp->icmp_type = ICMP_ECHO) && + (sent_icmp->icmp_id = ident) && + (sent_icmp->icmp_seq < (n_short) num_hosts)) { + /* this is a response to a ping we sent */ + h = table[sent_icmp->icmp_seq]; + printf("ICMP Unreachable from %s for ICMP Echo sent to %s", + inet_ntoa(addr->sin_addr), h->host); + + if(inet_addr(h->host) == INADDR_NONE) + printf(" (%s)", inet_ntoa(h->saddr.sin_addr)); + + printf("\n"); + } /* IF */ + + return 2; + + /* no way to tell whether any of these are sent due to our ping */ + /* or not (shouldn't be, of course), so just discard */ + case ICMP_TSTAMP: + case ICMP_TSTAMPREPLY: + case ICMP_IREQ: + case ICMP_IREQREPLY: + case ICMP_MASKREQ: + case ICMP_MASKREPLY: + default: + return 0; + + } /* SWITCH */ + +} /* handle_random_icmp() */ + +int in_cksum(u_short * p, int n) +{ + register u_short answer; + register long sum = 0; + u_short odd_byte = 0; + + while(n > 1) { + sum += *p++; + n -= 2; + } /* WHILE */ + + /* mop up an odd byte, if necessary */ + if(n == 1) { + *(u_char *) (&odd_byte) = *(u_char *) p; + sum += odd_byte; + } /* IF */ + + sum = (sum >> 16) + (sum & 0xffff); /* add hi 16 to low 16 */ + sum += (sum >> 16); /* add carry */ + answer = ~sum; /* ones-complement, truncate */ + + return (answer); + +} /* in_cksum() */ + +void add_name(char *name) +{ + struct hostent *host_ent; + int ipaddress; + struct in_addr *ipa = (struct in_addr *)&ipaddress; + struct in_addr *host_add; + char *nm; + int i = 0; + + if((ipaddress = inet_addr(name)) != -1) { + /* input name is an IP addr, go with it */ + if(name_flag) { + if(addr_flag) + add_addr(name, na_cat(get_host_by_address(*ipa), *ipa), *ipa); + else { + nm = cpystr(get_host_by_address(*ipa)); + add_addr(name, nm, *ipa); + + } /* ELSE */ + } /* IF */ + else add_addr(name, name, *ipa); + + return; + } /* IF */ + + /* input name is not an IP addr, maybe it's a host name */ + host_ent = gethostbyname(name); + if(host_ent == NULL) { + if(h_errno == TRY_AGAIN) { + u_sleep(DNS_TIMEOUT); + host_ent = gethostbyname(name); + } /* IF */ + + if(host_ent == NULL) { + printf("%s address not found\n", name); + num_noaddress++; + return; + } /* IF */ + } /* IF */ + + host_add = (struct in_addr *)*(host_ent->h_addr_list); + if(host_add == NULL) { + printf("%s has no address data\n", name); + num_noaddress++; + return; + } /* IF */ + else { + /* it is indeed a hostname with a real address */ + while(host_add) { + if(name_flag && addr_flag) + add_addr(name, na_cat(name, *host_add), *host_add); + else if(addr_flag) { + nm = cpystr(inet_ntoa(*host_add)); + add_addr(name, nm, *host_add); + } /* ELSE IF */ + else { + add_addr(name, name, *host_add); + } + + if(!multif_flag) break; + + host_add = (struct in_addr *)(host_ent->h_addr_list[++i]); + } /* WHILE */ + } /* ELSE */ +} /* add_name() */ + + +char *na_cat(char *name, struct in_addr ipaddr) +{ + char *nm, *as; + + as = inet_ntoa(ipaddr); + nm = (char *)malloc(strlen(name) + strlen(as) + 4); + + if(!nm) + crash("can't allocate some space for a string"); + + strcpy(nm, name); + strcat(nm, " ("); + strcat(nm, as); + strcat(nm, ")"); + + return (nm); + +} /* na_cat() */ + + +void add_addr(char *name, char *host, struct in_addr ipaddr) +{ + HOST_ENTRY *p; + unsigned int n; + int *i; + + if(!(p = (HOST_ENTRY *) malloc(sizeof(HOST_ENTRY)))) { + crash("can't allocate HOST_ENTRY"); + } + + memset((char *)p, 0, sizeof(HOST_ENTRY)); + + p->name = name; + p->host = host; + p->saddr.sin_family = AF_INET; + p->saddr.sin_addr = ipaddr; + p->running = 1; + + /* array for response time results */ + if(!(i = (int *)malloc(trials * sizeof(int)))) { + crash("can't allocate resp_times array"); + } + + for(n = 1; n < trials; n++) + i[n] = RESP_UNUSED; + + p->resp_times = i; + + if(!rrlist) { + rrlist = p; + p->next = p; + p->prev = p; + } /* IF */ + else { + p->next = rrlist; + p->prev = rrlist->prev; + p->prev->next = p; + p->next->prev = p; + } /* ELSE */ + + num_hosts++; +} /* add_addr() */ + + +void remove_job(HOST_ENTRY * h) +{ + h->running = 0; + h->waiting = 0; + num_jobs--; + + + if(num_jobs) { + /* remove us from list of active jobs */ + h->prev->next = h->next; + h->next->prev = h->prev; + if(h == cursor) cursor = h->next; + } /* IF */ + else { + cursor = NULL; + rrlist = NULL; + } /* ELSE */ + +} /* remove_job() */ + + +char *get_host_by_address(struct in_addr in) +{ + struct hostent *h; + h = gethostbyaddr((char *)&in, sizeof(struct in_addr), AF_INET); + + if(h == NULL || h->h_name == NULL) + return inet_ntoa(in); + else + return (char *)h->h_name; + +} /* get_host_by_address() */ + + +char *cpystr(char *string) +{ + char *dst; + + if(string) { + dst = (char *)malloc(1 + strlen(string)); + if(!dst) crash("malloc() failed!"); + + strcpy(dst, string); + return dst; + + } /* IF */ + else return NULL; + +} /* cpystr() */ + + +void crash(char *msg) +{ + if(errno || h_errno) { + if(errno) + printf("%s: %s : %s\n", prog, msg, strerror(errno)); + if(h_errno) + printf("%s: %s : A network error occurred\n", prog, msg); + } + else printf("%s: %s\n", prog, msg); + + exit(STATE_UNKNOWN); +} /* crash() */ + + +long timeval_diff(struct timeval *a, struct timeval *b) +{ + double temp; + + temp = (((a->tv_sec * 1000000) + a->tv_usec) - + ((b->tv_sec * 1000000) + b->tv_usec)) / 10; + + return (long)temp; + +} /* timeval_diff() */ + + +char *sprint_tm(int t) +{ + static char buf[10]; + + /* <= 0.99 ms */ + if(t < 100) { + sprintf(buf, "0.%02d", t); + return (buf); + } /* IF */ + + /* 1.00 - 9.99 ms */ + if(t < 1000) { + sprintf(buf, "%d.%02d", t / 100, t % 100); + return (buf); + } /* IF */ + + /* 10.0 - 99.9 ms */ + if(t < 10000) { + sprintf(buf, "%d.%d", t / 100, (t % 100) / 10); + return (buf); + } /* IF */ + + /* >= 100 ms */ + sprintf(buf, "%d", t / 100); + return (buf); +} /* sprint_tm() */ + + +/* + * select() is posix, so we expect it to be around + */ +void u_sleep(int u_sec) +{ + int nfound; + struct timeval to; + fd_set readset, writeset; + + to.tv_sec = u_sec / 1000000; + to.tv_usec = u_sec - (to.tv_sec * 1000000); +/* printf("u_sleep :: to.tv_sec: %d, to_tv_usec: %d\n", + (int)to.tv_sec, (int)to.tv_usec); +*/ + FD_ZERO(&writeset); + FD_ZERO(&readset); + nfound = select(0, &readset, &writeset, NULL, &to); + if(nfound < 0) + crash("select() in u_sleep:"); + + return; +} /* u_sleep() */ + + +/************************************************************ + * Description: + * + * receive with timeout + * returns length of data read or -1 if timeout + * crash on any other errrors + ************************************************************/ +/* TODO: add MSG_DONTWAIT to recvfrom flags (currently 0) */ +int recvfrom_wto(int sock, char *buf, int len, struct sockaddr *saddr, int timo) +{ + int nfound = 0, slen, n; + struct timeval to; + fd_set readset, writeset; + + to.tv_sec = timo / 1000000; + to.tv_usec = (timo - (to.tv_sec * 1000000)) * 10; + +/* printf("to.tv_sec: %d, to.tv_usec: %d\n", (int)to.tv_sec, (int)to.tv_usec); +*/ + + FD_ZERO(&readset); + FD_ZERO(&writeset); + FD_SET(sock, &readset); + nfound = select(sock + 1, &readset, &writeset, NULL, &to); + if(nfound < 0) crash("select() in recvfrom_wto"); + + if(nfound == 0) return -1; /* timeout */ + + if(nfound) { + slen = sizeof(struct sockaddr); + n = recvfrom(sock, buf, len, 0, saddr, &slen); + if(n < 0) crash("recvfrom"); + return(n); + } + + return(0); /* 0 bytes read, so return it */ +} /* recvfrom_wto() */ + + +/* + * u = micro + * m = milli + * s = seconds + */ +int get_threshold(char *str, threshold *th) +{ + unsigned int i, factor = 0; + char *p = NULL; + + if(!str || !strlen(str) || !th) return -1; + + for(i=0; irta = (unsigned int)strtoul(str, NULL, 0) * factor; + th->pl = (unsigned int)strtoul(p, NULL, 0); + return 0; +} + +/* make a blahblah */ +void usage(void) +{ + printf("\nUsage: %s [options] [targets]\n", prog); + printf(" -H host target host\n"); + printf(" -b n ping packet size in bytes (default %d)\n", ping_data_size); + printf(" -n|p n number of pings to send to each target (default %d)\n", count); + printf(" -r n number of retries (default %d)\n", retry); + printf(" -t n timeout value (in msec) (default %d)\n", timeout / 100); + printf(" -i n packet interval (in msec) (default %d)\n", DEFAULT_INTERVAL); +/* XXX - possibly on todo-list + printf(" -m ping multiple interfaces on target host\n"); + printf(" -a show targets that are alive\n"); + printf(" -d show dead targets\n"); +*/ printf(" -v show version\n"); + printf(" -D increase debug output level\n"); + printf(" -w warning threshold pair, given as RTA[ums],PL[%%]\n"); + printf(" -c critical threshold pair, given as RTA[ums],PL[%%]\n"); + printf("\n"); + printf("Note:\n"); + printf("* This program requires root privileges to run properly.\n"); + printf(" If it is run as setuid root it will halt with an error if;\n"); + printf(" interval < 25 || retries > 5\n\n"); + printf("* Threshold pairs are given as such;\n"); + printf(" 100,40%%\n"); + printf(" to set a threshold value pair of 100 milliseconds and 40%% packetloss\n"); + printf(" The '%%' sign is optional, and if rta value is suffixed by;\n"); + printf(" s, rta time is set in seconds\n"); + printf(" m, rta time will be set in milliseconds (this is default)\n"); + printf(" u, rta time will be set in microseconds\n"); + printf(" Decimal points are silently ignored for sideways compatibility.\n"); + printf("\n"); + exit(3); +} /* usage() */ -- cgit v0.10-9-g596f