]>
Commit | Line | Data |
---|---|---|
846841f4 | 1 | /* |
2ce085d2 | 2 | * The Data Control Manager for MOIRA. |
846841f4 | 3 | * |
4 | * Copyright 1987, 1988 by the Massachusetts Institute of Technology. | |
5 | * For copying and distribution information, see the file | |
6 | * "mit-copyright.h". | |
7 | * | |
8 | * $Source$ | |
9 | * $Author$ | |
10 | * $Header$ | |
11 | */ | |
12 | ||
13 | #ifndef lint | |
14 | static char rcsid_dcm_c[] = "$Header$"; | |
15 | #endif lint | |
16 | ||
17 | #include <stdio.h> | |
18 | #include <update.h> | |
19 | #include <sys/file.h> | |
20 | #include <sys/time.h> | |
21 | #include <sys/wait.h> | |
a1a0dadf | 22 | #include <signal.h> |
846841f4 | 23 | #include <ctype.h> |
2ce085d2 | 24 | #include <moira.h> |
25 | #include <moira_site.h> | |
846841f4 | 26 | #include "dcm.h" |
27 | #include "mit-copyright.h" | |
28 | ||
29 | extern char *ctime(); | |
30 | extern char *getenv(); | |
31 | extern int log_flags; | |
32 | extern char *error_message(); | |
33 | char *itoa(); | |
34 | int gqval(); | |
35 | long time(); | |
36 | ||
37 | ||
38 | /* declared global so that we can get the current time from different places. */ | |
39 | struct timeval tv; | |
40 | ||
41 | ||
42 | main(argc, argv) | |
43 | int argc; | |
44 | char *argv[]; | |
45 | { | |
46 | int i; | |
47 | char **arg = argv; | |
48 | char *qargv[3]; | |
49 | int status; | |
50 | ||
51 | whoami = argv[0]; | |
52 | dbg = atoi(getenv("DEBUG")); | |
53 | umask(UMASK); | |
54 | log_flags = 0; | |
55 | setlinebuf(stderr); | |
56 | setlinebuf(stdout); | |
57 | ||
58 | while(++arg - argv < argc) { | |
59 | if (**arg == '-') | |
60 | switch((*arg)[1]) { | |
61 | case 'd': | |
62 | dbg = atoi((*arg)[2]? *arg+2: *++arg); | |
63 | break; | |
64 | } | |
65 | } | |
66 | set_com_err_hook(dcm_com_err_hook); | |
67 | ||
68 | /* if /etc/nodcm exists, punt quietly. */ | |
185f76ce | 69 | if (!access(NODCMFILE, F_OK)) { |
846841f4 | 70 | exit(1); |
71 | } | |
72 | ||
2ce085d2 | 73 | if (status = mr_connect("")) { |
74 | com_err(whoami, status, " on mr_connect"); | |
846841f4 | 75 | leave("connect failed"); |
76 | } | |
77 | ||
2ce085d2 | 78 | if (status = mr_auth("dcm")) { |
846841f4 | 79 | com_err(whoami, status, " on \"authenticate\""); |
80 | leave("auth failed"); | |
81 | } | |
82 | ||
83 | /* if DCM is not enabled, exit after logging */ | |
84 | qargv[0] = "dcm_enable"; | |
2ce085d2 | 85 | if (status = mr_query("get_value", 1, qargv, gqval, &i)) { |
846841f4 | 86 | com_err(whoami, status, " check dcm_enable"); |
87 | leave("query failed"); | |
88 | } | |
89 | if (i == 0) { | |
90 | errno = 0; | |
91 | leave("dcm_enable not set"); | |
92 | } | |
93 | ||
94 | /* do it! */ | |
846841f4 | 95 | do_services(); |
96 | errno = 0; | |
97 | leave(""); | |
98 | } | |
99 | ||
100 | ||
101 | /* Used by the get_value query when checking for dcm_enable. */ | |
102 | ||
103 | gqval(argc, argv, hint) | |
104 | int argc; | |
105 | char **argv; | |
106 | int *hint; | |
107 | { | |
108 | *hint = atoi(argv[0]); | |
109 | return(UPCALL_STOP); | |
110 | } | |
111 | ||
112 | ||
113 | /* Used by qualified_get_server to make a list of servers to check */ | |
114 | ||
115 | qgetsv(argc, argv, sq) | |
116 | int argc; | |
117 | char **argv; | |
118 | struct save_queue *sq; | |
119 | { | |
120 | sq_save_data(sq, strsave(argv[0])); | |
121 | return(UPCALL_CONT); | |
122 | } | |
123 | ||
124 | ||
125 | /* Used by get_server_info to record all of the returned information */ | |
126 | ||
127 | getsvinfo(argc, argv, sserv) | |
128 | int argc; | |
129 | char **argv; | |
130 | struct service *sserv; | |
131 | { | |
132 | sserv->service = strsave(argv[0]); | |
133 | sserv->interval = atoi(argv[1]); | |
134 | sserv->target = strsave(argv[2]); | |
135 | sserv->script = strsave(argv[3]); | |
136 | sserv->dfgen = atoi(argv[4]); | |
137 | sserv->dfcheck = atoi(argv[5]); | |
138 | sserv->type = strsave(argv[6]); | |
139 | sserv->enable = atoi(argv[7]); | |
140 | sserv->inprogress = atoi(argv[8]); | |
141 | sserv->harderror = atoi(argv[9]); | |
142 | sserv->errmsg = strsave(argv[10]); | |
143 | return(UPCALL_STOP); | |
144 | } | |
145 | ||
146 | ||
147 | /* Scan the services and process any that need it. */ | |
148 | ||
149 | do_services() | |
150 | { | |
151 | char *qargv[6]; | |
152 | struct save_queue *sq, *sq_create(); | |
153 | char *service, dfgen_prog[64], dfgen_cmd[128]; | |
154 | struct service svc; | |
a1a0dadf | 155 | int status, lock_fd, ex, (*cstat)(); |
846841f4 | 156 | struct timezone tz; |
157 | register char *p; | |
158 | union wait waits; | |
159 | ||
160 | if (dbg & DBG_VERBOSE) | |
161 | com_err(whoami, 0, "starting pass over services"); | |
162 | ||
163 | qargv[0] = "true"; | |
164 | qargv[1] = "dontcare"; | |
165 | qargv[2] = "false"; | |
166 | sq = sq_create(); | |
2ce085d2 | 167 | if (status = mr_query_with_retry("qualified_get_server", 3, qargv, |
0967ae03 | 168 | qgetsv, sq)) { |
846841f4 | 169 | com_err(whoami, status, " getting services"); |
170 | leave("query failed"); | |
171 | } | |
172 | while (sq_get_data(sq, &service)) { | |
173 | for (p = service; *p; p++) | |
174 | if (isupper(*p)) | |
175 | *p = tolower(*p); | |
176 | com_err(whoami, 0, "checking %s...", service); | |
177 | qargv[0] = service; | |
185f76ce | 178 | sprintf(dfgen_prog, "%s/%s.gen", BIN_DIR, service); |
846841f4 | 179 | if (!file_exists(dfgen_prog)) { |
180 | com_err(whoami, 0, "prog %s doesn't exist\n", dfgen_prog); | |
181 | free(service); | |
182 | continue; | |
183 | } | |
185f76ce | 184 | sprintf(dfgen_cmd, "exec %s %s/%s.out", |
185 | dfgen_prog, DCM_DIR, service); | |
846841f4 | 186 | gettimeofday(&tv, &tz); |
2ce085d2 | 187 | if (status = mr_query_with_retry("get_server_info", 1, qargv, |
0967ae03 | 188 | getsvinfo, &svc)) { |
4e5690ff | 189 | com_err(whoami, status, " getting service %s info, skipping to next service", service); |
190 | continue; | |
846841f4 | 191 | } |
192 | svc.service = strsave(service); | |
193 | qargv[0] = strsave(service); | |
194 | qargv[1] = itoa(svc.dfgen); | |
195 | qargv[2] = itoa(svc.dfcheck); | |
196 | qargv[3] = strsave("0"); | |
197 | qargv[4] = itoa(svc.harderror); | |
198 | qargv[5] = strsave(svc.errmsg); | |
199 | if (svc.interval != 0) { | |
200 | if (svc.interval * 60 + svc.dfcheck < tv.tv_sec) { | |
185f76ce | 201 | lock_fd = maybe_lock_update("@db@", service, 1); |
846841f4 | 202 | if (lock_fd < 0) |
203 | goto free_service; | |
204 | free(qargv[3]); | |
205 | free(qargv[4]); | |
206 | free(qargv[5]); | |
207 | qargv[3] = strsave("1"); | |
208 | qargv[4] = strsave("0"); | |
209 | qargv[5] = strsave(""); | |
2ce085d2 | 210 | status = mr_query_with_retry("set_server_internal_flags", 6, |
0967ae03 | 211 | qargv, scream, NULL); |
2ce085d2 | 212 | if (status != MR_SUCCESS) { |
846841f4 | 213 | com_err(whoami, status, " setting server state"); |
214 | goto free_service; | |
215 | } | |
216 | ||
217 | com_err(whoami, status, " running %s", dfgen_prog); | |
a1a0dadf | 218 | cstat = signal(SIGCHLD, SIG_DFL); |
846841f4 | 219 | waits.w_status = system(dfgen_cmd); |
a1a0dadf | 220 | signal(SIGCHLD, cstat); |
c9c95b8a | 221 | if (waits.w_termsig) { |
2ce085d2 | 222 | status = MR_TAR_FAIL; |
c9c95b8a | 223 | com_err(whoami, status, " %s exited on signal %d", |
224 | dfgen_prog, waits.w_termsig); | |
225 | } else if (waits.w_retcode) { | |
226 | /* extract the process's exit value */ | |
2ce085d2 | 227 | status = waits.w_retcode + ERROR_TABLE_BASE_sms; |
846841f4 | 228 | com_err(whoami, status, " %s exited", dfgen_prog); |
229 | } | |
230 | if (SOFT_FAIL(status)) { | |
231 | free(qargv[5]); | |
232 | qargv[5] = strsave(error_message(status)); | |
2ce085d2 | 233 | } else if (status == MR_NO_CHANGE) { |
846841f4 | 234 | free(qargv[2]); |
235 | qargv[2] = itoa(tv.tv_sec); | |
236 | svc.dfcheck = tv.tv_sec; | |
2ce085d2 | 237 | } else if (status == MR_SUCCESS) { |
846841f4 | 238 | free(qargv[1]); |
239 | free(qargv[2]); | |
240 | qargv[1] = itoa(tv.tv_sec); | |
241 | qargv[2] = strsave(qargv[1]); | |
242 | svc.dfcheck = svc.dfgen = tv.tv_sec; | |
243 | } else { /* HARD_FAIL(status) */ | |
244 | free(qargv[2]); | |
245 | free(qargv[4]); | |
246 | free(qargv[5]); | |
247 | qargv[2] = itoa(tv.tv_sec); | |
248 | svc.dfcheck = tv.tv_sec; | |
249 | qargv[4] = itoa(status); | |
250 | qargv[5] = strsave(error_message(status)); | |
251 | critical_alert("DCM","DCM building config files for %s: %s", | |
252 | service, qargv[5]); | |
253 | } | |
254 | free_service: | |
255 | free(qargv[3]); | |
256 | qargv[3] = strsave("0"); | |
2ce085d2 | 257 | status = mr_query_with_retry("set_server_internal_flags", 6, |
0967ae03 | 258 | qargv, scream, NULL); |
259 | if (status) | |
260 | com_err(whoami, status, " setting service state"); | |
846841f4 | 261 | close(lock_fd); |
262 | free(qargv[0]); | |
263 | free(qargv[1]); | |
264 | free(qargv[2]); | |
265 | free(qargv[3]); | |
266 | free(qargv[4]); | |
267 | free(qargv[5]); | |
268 | } | |
269 | if (!strcmp(svc.type, "REPLICAT")) | |
270 | ex = 1; | |
271 | else | |
272 | ex = 0; | |
185f76ce | 273 | lock_fd = maybe_lock_update("@db@", service, ex); |
846841f4 | 274 | if (lock_fd >= 0) { |
275 | do_hosts(&svc); | |
276 | close(lock_fd); | |
277 | } | |
278 | } | |
279 | free(svc.service); | |
280 | free(svc.target); | |
281 | free(svc.script); | |
282 | free(svc.type); | |
283 | free(svc.errmsg); | |
284 | free(service); | |
285 | } | |
286 | sq_destroy(sq); | |
287 | } | |
288 | ||
289 | ||
290 | /* Used by qualified_get_server_host to make a list of hosts to check */ | |
291 | ||
292 | qgethost(argc, argv, sq) | |
293 | int argc; | |
294 | char **argv; | |
295 | struct save_queue *sq; | |
296 | { | |
297 | sq_save_data(sq, strsave(argv[1])); | |
298 | return(UPCALL_CONT); | |
299 | } | |
300 | ||
301 | ||
302 | /* Used by get_server_host_info to store all of the info about a host */ | |
303 | ||
304 | gethostinfo(argc, argv, shost) | |
305 | int argc; | |
306 | char **argv; | |
307 | struct svrhost *shost; | |
308 | { | |
309 | shost->service = strsave(argv[0]); | |
310 | shost->machine = strsave(argv[1]); | |
311 | shost->enable = atoi(argv[2]); | |
312 | shost->override = atoi(argv[3]); | |
313 | shost->success = atoi(argv[4]); | |
314 | shost->inprogress = atoi(argv[5]); | |
315 | shost->hosterror = atoi(argv[6]); | |
316 | shost->errmsg = strsave(argv[7]); | |
317 | shost->lasttry = atoi(argv[8]); | |
318 | shost->lastsuccess = atoi(argv[9]); | |
319 | shost->value1 = atoi(argv[10]); | |
320 | shost->value2 = atoi(argv[11]); | |
321 | shost->value3 = strsave(argv[12]); | |
322 | return(UPCALL_STOP); | |
323 | } | |
324 | ||
325 | ||
326 | /* Scans all of the hosts for a particular service, and processes them. */ | |
327 | ||
328 | do_hosts(svc) | |
329 | struct service *svc; | |
330 | { | |
331 | char *argv[9], *machine; | |
332 | int status, lock_fd; | |
333 | struct save_queue *sq; | |
334 | struct svrhost shost; | |
335 | ||
336 | sq = sq_create(); | |
337 | argv[0] = svc->service; | |
338 | argv[1] = "TRUE"; | |
339 | argv[2] = argv[3] = argv[4] = "DONTCARE"; | |
340 | argv[5] = "FALSE"; | |
2ce085d2 | 341 | status = mr_query_with_retry("qualified_get_server_host", 6, argv, |
0967ae03 | 342 | qgethost, sq); |
2ce085d2 | 343 | if (status == MR_NO_MATCH) { |
846841f4 | 344 | return; |
345 | } else if (status) { | |
346 | com_err(whoami, status, " getting server_hosts for %s", svc->service); | |
347 | return; | |
348 | } | |
349 | while (sq_get_data(sq, &machine)) { | |
350 | if (dbg & DBG_TRACE) | |
351 | com_err(whoami, 0, "checking %s...", machine); | |
352 | argv[1] = machine; | |
2ce085d2 | 353 | status = mr_query_with_retry("get_server_host_info", 2, argv, |
0967ae03 | 354 | gethostinfo, &shost); |
846841f4 | 355 | if (status) { |
356 | com_err(whoami,status, " getting server_host_info for %s", machine); | |
357 | goto free_mach; | |
358 | } | |
359 | if (!shost.enable || shost.hosterror || | |
4112693b | 360 | (shost.success && !shost.override && |
6899dfbe | 361 | shost.lastsuccess >= svc->dfgen)) { |
846841f4 | 362 | if (dbg & DBG_TRACE) |
363 | com_err(whoami, 0, "not updating %s:%s", svc->service, machine); | |
364 | goto free_mach; | |
365 | } | |
062079b1 | 366 | |
185f76ce | 367 | lock_fd = maybe_lock_update(machine, svc->service, 1); |
062079b1 | 368 | if (lock_fd < 0) |
369 | goto free_mach; | |
370 | argv[0] = svc->service; | |
371 | argv[1] = machine; | |
372 | argv[2] = argv[3] = argv[5] = "0"; | |
373 | argv[4] = "1"; | |
374 | argv[6] = strsave(""); | |
375 | argv[7] = itoa(tv.tv_sec); | |
376 | argv[8] = itoa(shost.lastsuccess); | |
2ce085d2 | 377 | status = mr_query_with_retry("set_server_host_internal", 9, argv, |
0967ae03 | 378 | scream, NULL); |
2ce085d2 | 379 | if (status != MR_SUCCESS) { |
062079b1 | 380 | com_err(whoami,status," while setting internal state for %s:%s", |
381 | svc->service, machine); | |
382 | goto free_mach; | |
383 | } | |
2ce085d2 | 384 | status = mr_update_server(svc->service, machine, svc->target, |
062079b1 | 385 | svc->script); |
2ce085d2 | 386 | if (status == MR_SUCCESS) { |
062079b1 | 387 | argv[2] = "0"; |
388 | argv[3] = "1"; | |
389 | free(argv[8]); | |
390 | argv[8] = itoa(tv.tv_sec); | |
391 | } else if (SOFT_FAIL(status)) { | |
392 | free(argv[6]); | |
393 | argv[6] = strsave(error_message(status)); | |
394 | } else { /* HARD_FAIL */ | |
395 | argv[2] = itoa(shost.override); | |
396 | argv[5] = itoa(status); | |
397 | free(argv[6]); | |
398 | argv[6] = strsave(error_message(status)); | |
399 | critical_alert("DCM", "DCM updating %s:%s: %s", | |
400 | machine, svc->service, argv[6]); | |
401 | if (!strcmp(svc->type, "REPLICAT")) { | |
402 | char *qargv[6]; | |
403 | ||
404 | svc->harderror = status; | |
405 | svc->errmsg = strsave(argv[6]); | |
406 | qargv[0] = strsave(svc->service); | |
407 | qargv[1] = itoa(svc->dfgen); | |
408 | qargv[2] = itoa(svc->dfcheck); | |
409 | qargv[3] = strsave("0"); | |
410 | qargv[4] = itoa(svc->harderror); | |
411 | qargv[5] = strsave(svc->errmsg); | |
2ce085d2 | 412 | status = mr_query_with_retry("set_server_internal_flags", |
0967ae03 | 413 | 6, qargv, scream, NULL); |
414 | if (status) | |
415 | com_err(whoami, status, " setting service state again"); | |
062079b1 | 416 | free(qargv[0]); |
417 | free(qargv[1]); | |
418 | free(qargv[2]); | |
419 | free(qargv[3]); | |
420 | free(qargv[4]); | |
421 | free(qargv[5]); | |
422 | close(lock_fd); | |
846841f4 | 423 | free(argv[2]); |
062079b1 | 424 | argv[4] = "0"; |
846841f4 | 425 | free(argv[5]); |
2ce085d2 | 426 | status = mr_query_with_retry("set_server_host_internal", |
0967ae03 | 427 | 9, argv,scream,NULL); |
428 | if (status) | |
429 | com_err(whoami, status, " setting host state again"); | |
062079b1 | 430 | return(-1); |
846841f4 | 431 | } |
062079b1 | 432 | free(argv[2]); |
433 | free(argv[5]); | |
846841f4 | 434 | } |
062079b1 | 435 | argv[4] = "0"; |
436 | close(lock_fd); | |
2ce085d2 | 437 | status = mr_query_with_retry("set_server_host_internal", 9, argv, |
0967ae03 | 438 | scream, NULL); |
439 | if (status) | |
440 | com_err(whoami, status, " setting host state again"); | |
846841f4 | 441 | free_mach: |
442 | free(machine); | |
443 | close(lock_fd); | |
444 | } | |
445 | return(0); | |
446 | } |