2 * The Data Control Manager for MOIRA.
4 * Copyright 1987, 1988 by the Massachusetts Institute of Technology.
5 * For copying and distribution information, see the file
14 static char rcsid_dcm_c[] = "$Header$";
29 #include <moira_site.h>
31 #include "mit-copyright.h"
35 extern int log_flags, errno;
39 /* declared global so that we can get the current time from different places. */
55 dbg = s ? atoi(s) : 0;
59 setvbuf(stderr, NULL, _IOLBF, BUFSIZ);
60 setvbuf(stdout, NULL, _IOLBF, BUFSIZ);
62 while(++arg - argv < argc) {
66 dbg = atoi((*arg)[2]? *arg+2: *++arg);
70 set_com_err_hook(dcm_com_err_hook);
72 /* if /etc/nodcm exists, punt quietly. */
73 if (!access(NODCMFILE, F_OK)) {
77 if (status = mr_connect("")) {
78 com_err(whoami, status, " on mr_connect");
79 leave("connect failed");
82 if (status = mr_auth("dcm")) {
83 com_err(whoami, status, " on \"authenticate\"");
87 /* if DCM is not enabled, exit after logging */
88 qargv[0] = "dcm_enable";
89 if (status = mr_query("get_value", 1, qargv, gqval, &i)) {
90 com_err(whoami, status, " check dcm_enable");
91 leave("query failed");
95 leave("dcm_enable not set");
105 /* Used by the get_value query when checking for dcm_enable. */
107 gqval(argc, argv, hint)
112 *hint = atoi(argv[0]);
117 /* Used by qualified_get_server to make a list of servers to check */
119 qgetsv(argc, argv, sq)
122 struct save_queue *sq;
124 sq_save_data(sq, strsave(argv[0]));
129 /* Used by get_server_info to record all of the returned information */
131 getsvinfo(argc, argv, sserv)
134 struct service *sserv;
136 sserv->service = strsave(argv[0]);
137 sserv->interval = atoi(argv[1]);
138 sserv->target = strsave(argv[2]);
139 sserv->script = strsave(argv[3]);
140 sserv->dfgen = atoi(argv[4]);
141 sserv->dfcheck = atoi(argv[5]);
142 sserv->type = strsave(argv[6]);
143 sserv->enable = atoi(argv[7]);
144 sserv->inprogress = atoi(argv[8]);
145 sserv->harderror = atoi(argv[9]);
146 sserv->errmsg = strsave(argv[10]);
151 /* Scan the services and process any that need it. */
156 struct save_queue *sq, *sq_create();
157 char *service, dfgen_prog[64], dfgen_cmd[128];
159 int status, lock_fd, ex, (*cstat)();
163 struct sigaction action, prevaction;
165 if (dbg & DBG_VERBOSE)
166 com_err(whoami, 0, "starting pass over services");
169 qargv[1] = "dontcare";
172 if (status = mr_query("qualified_get_server", 3, qargv, qgetsv, sq)) {
173 com_err(whoami, status, " getting services");
174 leave("query failed");
176 while (sq_get_data(sq, &service)) {
177 for (p = service; *p; p++)
180 com_err(whoami, 0, "checking %s...", service);
182 sprintf(dfgen_prog, "%s/%s.gen", BIN_DIR, service);
183 if (!file_exists(dfgen_prog)) {
184 com_err(whoami, 0, "prog %s doesn't exist\n", dfgen_prog);
188 sprintf(dfgen_cmd, "exec %s %s/%s.out",
189 dfgen_prog, DCM_DIR, service);
190 gettimeofday(&tv, &tz);
191 if (status = mr_query("get_server_info", 1, qargv, getsvinfo, &svc)) {
192 com_err(whoami, status, " getting service %s info, skipping to next service", service);
195 svc.service = strsave(service);
196 qargv[0] = strsave(service);
197 qargv[1] = itoa(svc.dfgen);
198 qargv[2] = itoa(svc.dfcheck);
199 qargv[3] = strsave("0");
200 qargv[4] = itoa(svc.harderror);
201 qargv[5] = strsave(svc.errmsg);
202 if (svc.interval != 0) {
203 if (svc.interval * 60 + svc.dfcheck < tv.tv_sec) {
204 lock_fd = maybe_lock_update("@db@", service, 1);
210 qargv[3] = strsave("1");
211 qargv[4] = strsave("0");
212 qargv[5] = strsave("");
213 status = mr_query("set_server_internal_flags", 6,
214 qargv, scream, NULL);
215 if (status != MR_SUCCESS) {
216 com_err(whoami, status, " setting server state");
220 com_err(whoami, status, " running %s", dfgen_prog);
223 sigemptyset(&action.sa_mask);
224 action.sa_handler = SIG_DFL;
225 sigaction(SIGCHLD, &action, &prevaction);
226 waits = system(dfgen_cmd);
227 sigaction(SIGCHLD, &prevaction, NULL);
228 if (WIFSIGNALED(waits)) {
229 status = MR_COREDUMP;
230 com_err(whoami, status, " %s exited on signal %d",
231 dfgen_prog, WTERMSIG(waits));
232 } else if (WEXITSTATUS(waits)) {
233 /* extract the process's exit value */
234 status = WEXITSTATUS(waits) + ERROR_TABLE_BASE_sms;
235 com_err(whoami, status, " %s exited", dfgen_prog);
238 if (SOFT_FAIL(status)) {
240 qargv[5] = strsave(error_message(status));
241 } else if (status == MR_NO_CHANGE) {
243 qargv[2] = itoa(tv.tv_sec);
244 svc.dfcheck = tv.tv_sec;
245 } else if (status == MR_SUCCESS) {
248 qargv[1] = itoa(tv.tv_sec);
249 qargv[2] = strsave(qargv[1]);
250 svc.dfcheck = svc.dfgen = tv.tv_sec;
251 } else { /* HARD_FAIL(status) */
255 qargv[2] = itoa(tv.tv_sec);
256 svc.dfcheck = tv.tv_sec;
257 qargv[4] = itoa(status);
258 qargv[5] = strsave(error_message(status));
259 critical_alert("DCM","DCM building config files for %s: %s",
264 qargv[3] = strsave("0");
265 status = mr_query("set_server_internal_flags", 6,
266 qargv, scream, NULL);
268 com_err(whoami, status, " setting service state");
277 if (!strcmp(svc.type, "REPLICAT"))
281 lock_fd = maybe_lock_update("@db@", service, ex);
298 /* Used by qualified_get_server_host to make a list of hosts to check */
300 qgethost(argc, argv, sq)
303 struct save_queue *sq;
305 sq_save_data(sq, strsave(argv[1]));
310 /* Used by get_server_host_info to store all of the info about a host */
312 gethostinfo(argc, argv, shost)
315 struct svrhost *shost;
317 shost->service = strsave(argv[0]);
318 shost->machine = strsave(argv[1]);
319 shost->enable = atoi(argv[2]);
320 shost->override = atoi(argv[3]);
321 shost->success = atoi(argv[4]);
322 shost->inprogress = atoi(argv[5]);
323 shost->hosterror = atoi(argv[6]);
324 shost->errmsg = strsave(argv[7]);
325 shost->lasttry = atoi(argv[8]);
326 shost->lastsuccess = atoi(argv[9]);
327 shost->value1 = atoi(argv[10]);
328 shost->value2 = atoi(argv[11]);
329 shost->value3 = strsave(argv[12]);
334 /* Scans all of the hosts for a particular service, and processes them. */
339 char *argv[9], *machine;
341 struct save_queue *sq;
342 struct svrhost shost;
345 argv[0] = svc->service;
347 argv[2] = argv[3] = argv[4] = "DONTCARE";
349 status = mr_query("qualified_get_server_host", 6, argv, qgethost, sq);
350 if (status == MR_NO_MATCH) {
353 com_err(whoami, status, " getting server_hosts for %s", svc->service);
356 while (sq_get_data(sq, &machine)) {
358 com_err(whoami, 0, "checking %s...", machine);
360 status = mr_query("get_server_host_info", 2, argv,
361 gethostinfo, &shost);
363 com_err(whoami,status, " getting server_host_info for %s", machine);
366 if (!shost.enable || shost.hosterror ||
367 (shost.success && !shost.override &&
368 shost.lastsuccess >= svc->dfgen)) {
370 com_err(whoami, 0, "not updating %s:%s", svc->service, machine);
374 lock_fd = maybe_lock_update(machine, svc->service, 1);
377 argv[0] = svc->service;
379 argv[2] = argv[3] = argv[5] = "0";
381 argv[6] = strsave("");
382 argv[7] = itoa(tv.tv_sec);
383 argv[8] = itoa(shost.lastsuccess);
384 status = mr_query("set_server_host_internal", 9, argv, scream, NULL);
385 if (status != MR_SUCCESS) {
386 com_err(whoami,status," while setting internal state for %s:%s",
387 svc->service, machine);
390 status = mr_update_server(svc->service, machine, svc->target,
392 if (status == MR_SUCCESS) {
393 argv[2] = argv[4] = "0";
396 argv[8] = itoa(tv.tv_sec);
397 } else if (SOFT_FAIL(status)) {
400 argv[6] = strsave(error_message(status));
401 } else { /* HARD_FAIL */
402 argv[2] = itoa(shost.override);
404 argv[5] = itoa(status);
406 argv[6] = strsave(error_message(status));
407 critical_alert("DCM", "DCM updating %s:%s: %s",
408 machine, svc->service, argv[6]);
409 if (!strcmp(svc->type, "REPLICAT")) {
412 svc->harderror = status;
413 svc->errmsg = strsave(argv[6]);
414 qargv[0] = strsave(svc->service);
415 qargv[1] = itoa(svc->dfgen);
416 qargv[2] = itoa(svc->dfcheck);
417 qargv[3] = strsave("0");
418 qargv[4] = itoa(svc->harderror);
419 qargv[5] = strsave(svc->errmsg);
420 status = mr_query("set_server_internal_flags",
421 6, qargv, scream, NULL);
423 com_err(whoami, status, " setting service state again");
431 status = mr_query("set_server_host_internal",
432 9, argv,scream,NULL);
436 com_err(whoami, status, " setting host state again");
441 status = mr_query("set_server_host_internal", 9, argv, scream, NULL);
443 com_err(whoami, status, " setting host state again");