contrib/package/freifunk-watchdog/src/watchdog.c

   1 /*
   2  *   This program is free software; you can redistribute it and/or modify
   3  *   it under the terms of the GNU General Public License as published by
   4  *   the Free Software Foundation; either version 2 of the License, or
   5  *   (at your option) any later version.
   6  *
   7  *   This program is distributed in the hope that it will be useful,
   8  *   but WITHOUT ANY WARRANTY; without even the implied warranty of
   9  *   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  10  *   GNU General Public License for more details.
  11  *
  12  *   You should have received a copy of the GNU General Public License
  13  *   along with this program; if not, write to the Free Software
  14  *   Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307, USA.
  15  *
  16  *   Copyright (C) 2009 Jo-Philipp Wich <xm@subsignal.org>
  17  */
  18
  19 #include "watchdog.h"
  20
  21 /* Global watchdog fd, required by signal handler */
  22 int wdfd = -1;
  23
  24 /* Handle finished childs */
  25 static void sigchld_handler(int sig)
  26 {
  27         pid_t pid;
  28
  29         while( (pid = waitpid(-1, NULL, WNOHANG)) > 0 )
  30                 syslog(LOG_INFO, "Child returned (pid %d)", pid);
  31 }
  32
  33 /* Watchdog shutdown helper */
  34 static void shutdown_watchdog(int sig)
  35 {
  36         static const char wshutdown = WATCH_SHUTDOWN;
  37
  38         if( wdfd > -1 )
  39         {
  40                 syslog(LOG_INFO, "Stopping watchdog timer");
  41                 write(wdfd, &wshutdown, 1);
  42                 close(wdfd);
  43                 wdfd = -1;
  44         }
  45
  46         exit(0);
  47 }
  48
  49 /* Get BSSID of given interface */
  50 static int iw_get_bssid(int iwfd, const char *ifname, char *bssid)
  51 {
  52         struct iwreq iwrq;
  53
  54         if( iw_ioctl(iwfd, ifname, SIOCGIWAP, &iwrq) >= 0 )
  55         {
  56                 unsigned char *addr = (unsigned char *)iwrq.u.ap_addr.sa_data;
  57
  58                 sprintf(bssid, "%02X:%02X:%02X:%02X:%02X:%02X",
  59                         addr[0], addr[1], addr[2], addr[3], addr[4], addr[5]);
  60
  61                 return 0;
  62         }
  63
  64         return -1;
  65 }
  66
  67 /* Get channel of given interface */
  68 static int iw_get_channel(int iwfd, const char *ifname, int *channel)
  69 {
  70         int i;
  71         char buffer[sizeof(struct iw_range)];
  72         double cur_freq, cmp_freq;
  73         struct iwreq iwrq;
  74         struct iw_range *range;
  75
  76         memset(buffer, 0, sizeof(buffer));
  77
  78         iwrq.u.data.pointer = (char *)buffer;
  79         iwrq.u.data.length = sizeof(buffer);
  80         iwrq.u.data.flags = 0;
  81
  82         if( iw_ioctl(iwfd, ifname, SIOCGIWRANGE, &iwrq) < 0)
  83         {
  84                 *channel = -1;
  85                 return -1;
  86         }
  87
  88         range = (struct iw_range *)buffer;
  89
  90         if( iw_ioctl(iwfd, ifname, SIOCGIWFREQ, &iwrq) >= 0 )
  91         {
  92                 cur_freq = ((double)iwrq.u.freq.m) * pow(10, iwrq.u.freq.e);
  93                 if( cur_freq < 1000.00 )
  94                 {
  95                         *channel = (int)cur_freq;
  96                         return 0;
  97                 }
  98
  99                 for(i = 0; i < range->num_frequency; i++)
 100                 {
 101                         cmp_freq = ((double)range->freq[i].m) * pow(10, range->freq[i].e);
 102                         if( cmp_freq == cur_freq )
 103                         {
 104                                 *channel = (int)range->freq[i].i;
 105                                 return 0;
 106                         }
 107                 }
 108         }
 109
 110         *channel = -1;
 111         return -1;
 112 }
 113
 114 /* Get the (first) pid of given process name */
 115 static int find_process(const char *name)
 116 {
 117         int pid = -1;
 118         int file;
 119         char buffer[128];
 120         char cmpname[128];
 121         DIR *dir;
 122         struct dirent *entry;
 123
 124         if( (dir = opendir("/proc")) != NULL )
 125         {
 126                 snprintf(cmpname, sizeof(cmpname), "Name:\t%s\n", name);
 127
 128                 while( (entry = readdir(dir)) != NULL )
 129                 {
 130                         if( !strcmp(entry->d_name, "..") || !isdigit(*entry->d_name) )
 131                                 continue;
 132
 133                         sprintf(buffer, "/proc/%s/status", entry->d_name);
 134                         if( (file = open(buffer, O_RDONLY)) > -1 )
 135                         {
 136                                 read(file, buffer, sizeof(buffer));
 137                                 close(file);
 138
 139                                 if( strstr(buffer, cmpname) == buffer )
 140                                 {
 141                                         pid = atoi(entry->d_name);
 142
 143                                         /* Skip myself ... */
 144                                         if( pid == getpid() )
 145                                                 pid = -1;
 146                                         else
 147                                                 break;
 148                                 }
 149                         }
 150                 }
 151
 152                 closedir(dir);
 153                 return pid;
 154         }
 155
 156         syslog(LOG_CRIT, "Unable to open /proc: %s",
 157                 strerror(errno));
 158
 159         return -1;
 160 }
 161
 162 /* Get the 5 minute load average */
 163 static double find_loadavg(void)
 164 {
 165         int fd;
 166         char buffer[10];
 167         double load = 0.00;
 168
 169         if( (fd = open("/proc/loadavg", O_RDONLY)) > -1 )
 170         {
 171                 if( read(fd, buffer, sizeof(buffer)) == sizeof(buffer) )
 172                         load = atof(&buffer[5]);
 173
 174                 close(fd);
 175         }
 176
 177         return load;
 178 }
 179
 180 /* Check if given uci file was updated */
 181 static int check_uci_update(const char *config, time_t *mtime)
 182 {
 183         struct stat s;
 184         char path[128];
 185
 186         snprintf(path, sizeof(path), "/var/state/%s", config);
 187         if( stat(path, &s) > -1 )
 188         {
 189                 if( (*mtime == 0) || (s.st_mtime > *mtime) )
 190                 {
 191                         *mtime = s.st_mtime;
 192                         return 1;
 193                 }
 194         }
 195
 196         return 0;
 197 }
 198
 199 /* Add tuple */
 200 static void load_wifi_uci_add_iface(const char *section, struct uci_wifi_iface_itr_ctx *itr)
 201 {
 202         wifi_tuple_t *t;
 203         const char *ucitmp;
 204         int val = 0;
 205
 206         ucitmp = ucix_get_option(itr->ctx, "wireless", section, "mode");
 207         if( ucitmp && !strncmp(ucitmp, "adhoc", 5) )
 208         {
 209                 if( (t = (wifi_tuple_t *)malloc(sizeof(wifi_tuple_t))) != NULL )
 210                 {
 211                         ucitmp = ucix_get_option(itr->ctx, "wireless", section, "ifname");
 212                         if(ucitmp)
 213                         {
 214                                 strncpy(t->ifname, ucitmp, sizeof(t->ifname));
 215                                 val++;
 216                         }
 217
 218                         ucitmp = ucix_get_option(itr->ctx, "wireless", section, "bssid");
 219                         if(ucitmp)
 220                         {
 221                                 strncpy(t->bssid, ucitmp, sizeof(t->bssid));
 222                                 val++;
 223                         }
 224
 225                         ucitmp = ucix_get_option(itr->ctx, "wireless", section, "device");
 226                         if(ucitmp)
 227                         {
 228                                 ucitmp = ucix_get_option(itr->ctx, "wireless", ucitmp, "channel");
 229                                 if(ucitmp)
 230                                 {
 231                                         t->channel = atoi(ucitmp);
 232                                         val++;
 233                                 }
 234                         }
 235
 236                         if( val == 3 )
 237                         {
 238                                 syslog(LOG_INFO, "Monitoring %s: bssid=%s channel=%d",
 239                                         t->ifname, t->bssid, t->channel);
 240
 241                                 t->next = itr->list;
 242                                 itr->list = t;
 243                         }
 244                         else
 245                         {
 246                                 free(t);
 247                         }
 248                 }
 249         }
 250 }
 251
 252 /* Load config */
 253 static wifi_tuple_t * load_wifi_uci(wifi_tuple_t *ifs, time_t *modtime)
 254 {
 255         struct uci_context *ctx;
 256         struct uci_wifi_iface_itr_ctx itr;
 257         wifi_tuple_t *cur, *next;
 258
 259         if( check_uci_update("wireless", modtime) )
 260         {
 261                 syslog(LOG_INFO, "Wireless config changed, reloading");
 262
 263                 if( (ctx = ucix_init("wireless")) != NULL )
 264                 {
 265                         if( ifs != NULL )
 266                         {
 267                                 for(cur = ifs; cur; cur = next)
 268                                 {
 269                                         next = cur->next;
 270                                         free(cur);
 271                                 }
 272                         }
 273
 274                         itr.list = NULL;
 275                         itr.ctx = ctx;
 276
 277                         ucix_for_each_section_type(ctx, "wireless", "wifi-iface",
 278                                 (void *)load_wifi_uci_add_iface, &itr);
 279
 280                         return itr.list;
 281                 }
 282         }
 283
 284         return ifs;
 285 }
 286
 287 /* Add tuple */
 288 static void load_watchdog_uci_add_process(const char *section, struct uci_process_itr_ctx *itr)
 289 {
 290         process_tuple_t *t;
 291         const char *ucitmp;
 292         int val = 0;
 293
 294         if( (t = (process_tuple_t *)malloc(sizeof(process_tuple_t))) != NULL )
 295         {
 296                 t->restart = 0;
 297
 298                 ucitmp = ucix_get_option(itr->ctx, "freifunk-watchdog", section, "process");
 299                 if(ucitmp)
 300                 {
 301                         strncpy(t->process, ucitmp, sizeof(t->process));
 302                         val++;
 303                 }
 304
 305                 ucitmp = ucix_get_option(itr->ctx, "freifunk-watchdog", section, "initscript");
 306                 if(ucitmp)
 307                 {
 308                         strncpy(t->initscript, ucitmp, sizeof(t->initscript));
 309                         val++;
 310                 }
 311
 312                 if( val == 2 )
 313                 {
 314                         syslog(LOG_INFO, "Monitoring %s: initscript=%s",
 315                                 t->process, t->initscript);
 316
 317                                 t->next = itr->list;
 318                                 itr->list = t;
 319                 }
 320                 else
 321                 {
 322                         free(t);
 323                 }
 324         }
 325 }
 326
 327 /* Load config */
 328 static process_tuple_t * load_watchdog_uci(process_tuple_t *procs)
 329 {
 330         struct uci_context *ctx;
 331         struct uci_process_itr_ctx itr;
 332         process_tuple_t *cur, *next;
 333
 334         syslog(LOG_INFO, "Loading watchdog config");
 335
 336         if( (ctx = ucix_init("freifunk-watchdog")) != NULL )
 337         {
 338                 if( procs != NULL )
 339                 {
 340                         for(cur = procs; cur; cur = next)
 341                         {
 342                                 next = cur->next;
 343                                 free(cur);
 344                         }
 345                 }
 346
 347                 itr.list = NULL;
 348                 itr.ctx = ctx;
 349
 350                 ucix_for_each_section_type(ctx, "freifunk-watchdog", "process",
 351                         (void *)load_watchdog_uci_add_process, &itr);
 352
 353                 return itr.list;
 354         }
 355
 356         return procs;
 357 }
 358
 359 /* Daemon implementation */
 360 static int do_daemon(void)
 361 {
 362         static int wdtrigger = 1;
 363         static int wdtimeout = BASE_INTERVAL * 2;
 364         static const char wdkeepalive = WATCH_KEEPALIVE;
 365
 366         int iwfd;
 367         int channel;
 368         char bssid[18];
 369         struct sigaction sa;
 370
 371         wifi_tuple_t *ifs = NULL, *curr_if;
 372         process_tuple_t *procs = NULL, *curr_proc;
 373         time_t wireless_modtime = 0;
 374
 375         int action_intv = 0;
 376         int restart_wifi = 0;
 377         int loadavg_panic = 0;
 378
 379         openlog(SYSLOG_IDENT, 0, LOG_DAEMON);
 380         memset(&sa, 0, sizeof(sa));
 381
 382         if( (iwfd = socket(AF_INET, SOCK_DGRAM, 0)) == -1 )
 383         {
 384                 syslog(LOG_ERR, "Can not open wireless control socket: %s",
 385                         strerror(errno));
 386
 387                 return 1;
 388         }
 389
 390         if( (wdfd = open(WATCH_DEVICE, O_WRONLY)) > -1 )
 391         {
 392                 syslog(LOG_INFO, "Opened %s - polling every %i seconds",
 393                         WATCH_DEVICE, BASE_INTERVAL);
 394
 395                 /* Install signal handler to halt watchdog on shutdown */
 396                 sa.sa_handler = shutdown_watchdog;
 397                 sa.sa_flags = SA_NOCLDWAIT | SA_RESTART;
 398                 sigaction(SIGHUP,  &sa, NULL);
 399                 sigaction(SIGINT,  &sa, NULL);
 400                 sigaction(SIGPIPE, &sa, NULL);
 401                 sigaction(SIGTERM, &sa, NULL);
 402                 sigaction(SIGUSR1, &sa, NULL);
 403                 sigaction(SIGUSR2, &sa, NULL);
 404
 405                 /* Set watchdog timeout to twice the interval */
 406                 ioctl(wdfd, WDIOC_SETTIMEOUT, &wdtimeout);
 407         }
 408
 409         /* Install signal handler to reap childs */
 410         sa.sa_handler = sigchld_handler;
 411         sa.sa_flags = 0;
 412         sigaction(SIGCHLD, &sa, NULL);
 413
 414         /* Load watchdog configuration only once */
 415         procs = load_watchdog_uci(procs);
 416
 417         while( 1 )
 418         {
 419                 /* Check/increment action interval */
 420                 if( ++action_intv >= ACTION_INTERVAL )
 421                 {
 422                         /* Reset action interval */
 423                         action_intv = 0;
 424
 425                         /* Check average load */
 426                         if( find_loadavg() >= LOAD_TRESHOLD )
 427                                 loadavg_panic++;
 428                         else
 429                                 loadavg_panic = 0;
 430
 431                         /* Check wireless interfaces */
 432                         ifs = load_wifi_uci(ifs, &wireless_modtime);
 433                         for( curr_if = ifs; curr_if; curr_if = curr_if->next )
 434                         {
 435                                 /* Get current channel and bssid */
 436                                 if( (iw_get_bssid(iwfd, curr_if->ifname, bssid) == 0) &&
 437                             (iw_get_channel(iwfd, curr_if->ifname, &channel) == 0) )
 438                                 {
 439                                         /* Check BSSID */
 440                                         if( strcasecmp(bssid, curr_if->bssid) != 0 )
 441                                         {
 442                                                 syslog(LOG_WARNING, "BSSID mismatch on %s: current=%s wanted=%s",
 443                                                         curr_if->ifname, bssid, curr_if->bssid);
 444
 445                                                 restart_wifi++;
 446                                         }
 447
 448                                         /* Check channel */
 449                                         else if( channel != curr_if->channel )
 450                                         {
 451                                                 syslog(LOG_WARNING, "Channel mismatch on %s: current=%d wanted=%d",
 452                                                         curr_if->ifname, channel, curr_if->channel);
 453
 454                                                 restart_wifi++;
 455                                         }
 456                                 }
 457                                 else
 458                                 {
 459                                         syslog(LOG_WARNING, "Requested interface %s not present", curr_if->ifname);
 460                                 }
 461                         }
 462
 463                         /* Check processes */
 464                         for( curr_proc = procs; curr_proc; curr_proc = curr_proc->next )
 465                         {
 466                                 if( find_process(curr_proc->process) < 0 )
 467                                         curr_proc->restart++;
 468                                 else
 469                                         curr_proc->restart = 0;
 470
 471                                 /* Process restart required? */
 472                                 if( curr_proc->restart >= HYSTERESIS )
 473                                 {
 474                                         curr_proc->restart = 0;
 475                                         syslog(LOG_WARNING, "The %s process died, restarting", curr_proc->process);
 476                                         EXEC(PROC_ACTION);
 477                                 }
 478                         }
 479
 480
 481                         /* Wifi restart required? */
 482                         if( restart_wifi >= HYSTERESIS )
 483                         {
 484                                 restart_wifi = 0;
 485                                 syslog(LOG_WARNING, "Channel or BSSID mismatch on wireless interface, restarting");
 486                                 EXEC(WIFI_ACTION);
 487                         }
 488
 489                         /* Is there a load problem? */
 490                         if( loadavg_panic >= HYSTERESIS )
 491                         {
 492                                 syslog(LOG_EMERG, "Critical system load level, triggering reset!");
 493
 494                                 /* Try watchdog, fall back to reboot */
 495                                 if( wdfd > -1 )
 496                                         ioctl(wdfd, WDIOC_SETTIMEOUT, &wdtrigger);
 497                                 else
 498                                         EXEC(LOAD_ACTION);
 499                         }
 500                 }
 501
 502
 503                 /* Reset watchdog timer */
 504                 if( wdfd > -1 )
 505                         write(wdfd, &wdkeepalive, 1);
 506
 507                 sleep(BASE_INTERVAL);
 508         }
 509
 510         shutdown_watchdog(0);
 511         closelog();
 512
 513         return 0;
 514 }
 515
 516
 517 int main(int argc, char *argv[])
 518 {
 519         /* Check if watchdog is running ... */
 520         if( (argc > 1) && (strcmp(argv[1], "running") == 0) )
 521         {
 522                 return (find_process(BINARY) == -1);
 523         }
 524
 525         /* Start daemon */
 526         return do_daemon();
 527 }