]>
Commit | Line | Data |
---|---|---|
7ac48069 | 1 | /* $Id$ |
2a2a3914 | 2 | * |
7ac48069 | 3 | * The Data Control Manager for Moira. |
2a2a3914 | 4 | * |
7ac48069 | 5 | * Copyright (C) 1987-1998 by the Massachusetts Institute of Technology. |
6 | * For copying and distribution information, see the file | |
7 | * <mit-copyright.h>. | |
2a2a3914 | 8 | */ |
9 | ||
7ac48069 | 10 | #include <mit-copyright.h> |
11 | #include <moira.h> | |
12 | #include <moira_site.h> | |
dfaf9b68 | 13 | #include <moira_schema.h> |
7ac48069 | 14 | #include "update.h" |
15 | ||
16 | #include <sys/param.h> | |
17 | #include <sys/stat.h> | |
18 | #include <sys/wait.h> | |
2a2a3914 | 19 | |
85330553 | 20 | #include <errno.h> |
2a2a3914 | 21 | #include <signal.h> |
22 | #include <stdio.h> | |
7ac48069 | 23 | #include <stdlib.h> |
2a2a3914 | 24 | #include <string.h> |
2a2a3914 | 25 | #include <unistd.h> |
2a2a3914 | 26 | |
2a2a3914 | 27 | EXEC SQL INCLUDE sqlca; |
7ac48069 | 28 | void sqlglm(char *, unsigned int *, unsigned int *); |
29 | ||
30 | RCSID("$Header$"); | |
31 | ||
32 | int generate_service(char *name, int force); | |
33 | void do_hosts(char *service); | |
34 | int dcm_send_file(char *service, char *host, char *target); | |
35 | int dcm_execute(char *service, char *host, char *script); | |
36 | void dbmserr(void); | |
37 | ||
2a2a3914 | 38 | #define SQL_NO_MATCH 1403 |
5eaef520 | 39 | #define SOFT_FAIL(x) (((x) == MR_NO_MEM) || ((x) == MR_CANT_CONNECT) || ((x) == MR_CCONFIG) || ((x) == MR_DEADLOCK) || ((x) == MR_BUSY) || ((x) == MR_ABORT)) |
2a2a3914 | 40 | |
5eaef520 | 41 | char whobuf[256], *whoami = whobuf, *db = "moira"; |
2a2a3914 | 42 | |
5eaef520 | 43 | int main(int argc, char **argv) |
2a2a3914 | 44 | { |
5eaef520 | 45 | int i; |
46 | EXEC SQL BEGIN DECLARE SECTION; | |
dfaf9b68 | 47 | char buf[SERVERS_NAME_SIZE], *name; |
5eaef520 | 48 | int enable; |
49 | EXEC SQL END DECLARE SECTION; | |
50 | struct save_queue *sq; | |
51 | int status; | |
52 | ||
53 | if (strchr(argv[0], '/')) | |
54 | strcpy(whoami, strrchr(argv[0], '/') + 1); | |
55 | else strcpy(whoami, argv[0]); | |
56 | umask(7); | |
57 | ||
58 | setvbuf(stderr, NULL, _IOLBF, BUFSIZ); | |
59 | setvbuf(stdout, NULL, _IOLBF, BUFSIZ); | |
60 | ||
61 | initialize_sms_error_table(); | |
62 | initialize_krb_error_table(); | |
63 | ||
64 | /* if services were specified on the command line, do just those ones */ | |
65 | if (argc > 1) | |
66 | { | |
67 | for (i = 1; i < argc; i++) | |
68 | { | |
69 | if (generate_service(argv[i], 1)) | |
70 | do_hosts(argv[i]); | |
2a2a3914 | 71 | } |
5eaef520 | 72 | exit(0); |
2a2a3914 | 73 | } |
74 | ||
5eaef520 | 75 | /* if DCM is not enabled, exit after logging */ |
76 | if (!access(NODCMFILE, F_OK)) | |
77 | { | |
78 | printf("/etc/nodcm exists -- exiting\n"); | |
79 | exit(1); | |
2a2a3914 | 80 | } |
5eaef520 | 81 | |
ff20de45 | 82 | EXEC SQL WHENEVER SQLERROR DO dbmserr(); |
83 | ||
5eaef520 | 84 | EXEC SQL CONNECT :db IDENTIFIED BY :db; |
85 | ||
86 | EXEC SQL SELECT value INTO :enable FROM numvalues WHERE name = 'dcm_enable'; | |
87 | if (enable == 0) | |
88 | { | |
89 | printf("dcm_enable not set -- exiting\n"); | |
90 | exit(1); | |
2a2a3914 | 91 | } |
5eaef520 | 92 | |
93 | /* fetch list of services */ | |
94 | EXEC SQL DECLARE csr_svc CURSOR FOR SELECT LOWER(name) FROM servers | |
95 | WHERE enable = 1 AND harderror = 0 AND update_int > 0; | |
96 | EXEC SQL OPEN csr_svc; | |
97 | sq = sq_create(); | |
98 | while (1) | |
99 | { | |
100 | EXEC SQL FETCH csr_svc INTO :buf; | |
101 | if (sqlca.sqlcode) | |
102 | break; | |
103 | ||
104 | sq_save_data(sq, strdup(strtrim(buf))); | |
2a2a3914 | 105 | } |
5eaef520 | 106 | EXEC SQL CLOSE csr_svc; |
107 | /* we will repeatedly open and close the db since it seems to get | |
108 | upset if you keep it open across a fork */ | |
109 | EXEC SQL COMMIT RELEASE; | |
110 | ||
111 | /* Now run through list */ | |
112 | while (sq_get_data(sq, &name)) | |
113 | { | |
114 | if (generate_service(name, 0)) | |
115 | { | |
116 | switch (fork()) | |
117 | { | |
118 | case -1: | |
ff20de45 | 119 | com_err(whoami, errno, "forking for service %s -- exiting", |
5eaef520 | 120 | name); |
121 | exit(1); | |
122 | case 0: | |
ff20de45 | 123 | sprintf(strchr(whoami, '\0'), " (%s:%ld)", name, (long)getpid()); |
5eaef520 | 124 | do_hosts(name); |
125 | com_err(whoami, 0, "exiting"); | |
126 | exit(0); | |
127 | default: | |
128 | break; | |
2a2a3914 | 129 | } |
130 | } | |
131 | } | |
5eaef520 | 132 | |
ff20de45 | 133 | com_err(whoami, 0, "All files generated. Waiting for children to exit"); |
5eaef520 | 134 | while (waitpid(0, &status, 0) > 0) |
135 | ; | |
136 | com_err(whoami, 0, "exiting"); | |
7ac48069 | 137 | exit(0); |
2a2a3914 | 138 | } |
139 | ||
140 | int generate_service(char *name, int force) | |
141 | { | |
5eaef520 | 142 | EXEC SQL BEGIN DECLARE SECTION; |
143 | int interval, dfcheck, status; | |
144 | time_t now; | |
7ac48069 | 145 | const char *errmsg; |
5eaef520 | 146 | EXEC SQL END DECLARE SECTION; |
dfaf9b68 | 147 | char dfgen_prog[MAXPATHLEN], dfgen_cmd[2 * MAXPATHLEN]; |
5eaef520 | 148 | struct sigaction action, prevaction; |
149 | int waits; | |
150 | ||
151 | EXEC SQL CONNECT :db IDENTIFIED BY :db; | |
152 | ||
dfaf9b68 | 153 | EXEC SQL SELECT update_int, dfcheck INTO :interval, :dfcheck |
5eaef520 | 154 | FROM servers WHERE name = UPPER(:name); |
155 | if (sqlca.sqlcode == SQL_NO_MATCH) | |
156 | { | |
157 | com_err(whoami, 0, "No such service `%s'", name); | |
158 | EXEC SQL COMMIT RELEASE; | |
159 | return 0; | |
2a2a3914 | 160 | } |
5eaef520 | 161 | |
162 | time(&now); | |
163 | ||
164 | if ((interval * 60 + dfcheck < now) || force) | |
165 | { | |
166 | sprintf(dfgen_prog, "%s/%s.gen", BIN_DIR, name); | |
167 | if (access(dfgen_prog, F_OK) != 0) | |
168 | { | |
169 | com_err(whoami, 0, "prog %s doesn't exist", dfgen_prog); | |
170 | EXEC SQL COMMIT RELEASE; | |
171 | return 0; | |
2a2a3914 | 172 | } |
5eaef520 | 173 | sprintf(dfgen_cmd, "exec %s %s/%s.out", dfgen_prog, DCM_DIR, name); |
174 | com_err(whoami, 0, "running %s", dfgen_prog); | |
175 | ||
ff20de45 | 176 | EXEC SQL WHENEVER SQLERROR GOTO gen_cleanup; |
177 | ||
5eaef520 | 178 | EXEC SQL UPDATE servers SET inprogress = 1 |
179 | WHERE name = UPPER(:name); | |
ff20de45 | 180 | EXEC SQL COMMIT; |
5eaef520 | 181 | |
182 | action.sa_flags = 0; | |
183 | sigemptyset(&action.sa_mask); | |
184 | action.sa_handler = SIG_DFL; | |
185 | sigaction(SIGCHLD, &action, &prevaction); | |
186 | waits = system(dfgen_cmd); | |
187 | sigaction(SIGCHLD, &prevaction, NULL); | |
188 | if (WIFSIGNALED(waits)) | |
189 | { | |
190 | status = MR_COREDUMP; | |
191 | com_err(whoami, status, " %s exited on signal %d", | |
192 | dfgen_prog, WTERMSIG(waits)); | |
193 | } | |
194 | else if (WEXITSTATUS(waits)) | |
195 | { | |
196 | /* extract the process's exit value */ | |
197 | status = WEXITSTATUS(waits) + ERROR_TABLE_BASE_sms; | |
198 | if (status != MR_NO_CHANGE) | |
199 | com_err(whoami, status, "in %s", dfgen_prog); | |
200 | } | |
201 | else | |
202 | status = MR_SUCCESS; | |
203 | ||
204 | if (status == MR_SUCCESS) | |
205 | { | |
206 | EXEC SQL UPDATE servers SET dfgen = :now, dfcheck = :now, | |
207 | inprogress = 0 WHERE name = UPPER(:name); | |
208 | EXEC SQL COMMIT RELEASE; | |
209 | return 1; | |
210 | } | |
211 | else if (status == MR_NO_CHANGE) | |
212 | { | |
213 | EXEC SQL UPDATE servers SET dfcheck = :now, inprogress = 0 | |
214 | WHERE name = UPPER(:name); | |
215 | } | |
216 | else if (SOFT_FAIL(status)) | |
217 | { | |
218 | errmsg = error_message(status); | |
219 | EXEC SQL UPDATE servers SET errmsg = :errmsg, inprogress = 0 | |
220 | WHERE name = UPPER(:name); | |
221 | } | |
222 | else /* HARD_FAIL(status) */ | |
223 | { | |
224 | errmsg = error_message(status); | |
ff20de45 | 225 | EXEC SQL UPDATE servers SET harderror = :status, errmsg = :errmsg, |
226 | inprogress = 0 WHERE name = UPPER(:name); | |
5eaef520 | 227 | critical_alert("DCM", "DCM building config files for %s: %s", |
228 | name, errmsg); | |
2a2a3914 | 229 | } |
230 | } | |
5eaef520 | 231 | EXEC SQL COMMIT RELEASE; |
232 | return 0; | |
ff20de45 | 233 | |
234 | gen_cleanup: | |
235 | EXEC SQL WHENEVER SQLERROR DO dbmserr(); | |
236 | EXEC SQL UPDATE servers SET inprogress = 0, harderror = MR_INTERNAL, | |
237 | errmsg = 'DBMS Internal Error' WHERE name = UPPER(:name); | |
238 | dbmserr(); | |
2a2a3914 | 239 | } |
240 | ||
241 | void do_hosts(char *service) | |
242 | { | |
5eaef520 | 243 | EXEC SQL BEGIN DECLARE SECTION; |
dae7c89e | 244 | char type[SERVERS_TYPE_SIZE], host[MACHINE_NAME_SIZE], *name; |
dfaf9b68 | 245 | char target[SERVERS_TARGET_FILE_SIZE], script[SERVERS_SCRIPT_SIZE]; |
7ac48069 | 246 | const char *errmsg; |
ff20de45 | 247 | int status = 0, dfgen, replicated, mid; |
5eaef520 | 248 | time_t now; |
249 | EXEC SQL END DECLARE SECTION; | |
dae7c89e | 250 | struct save_queue *sq; |
5eaef520 | 251 | |
252 | time(&now); | |
85330553 | 253 | mr_init(); |
5eaef520 | 254 | |
255 | EXEC SQL CONNECT :db IDENTIFIED BY :db; | |
256 | ||
257 | EXEC SQL SELECT dfgen, type, target_file, script | |
258 | INTO :dfgen, :type, :target, :script | |
259 | FROM servers WHERE name = UPPER(:service); | |
260 | replicated = !strncmp(type, "REPLICAT", 8); | |
ff20de45 | 261 | strtrim(target); |
262 | strtrim(script); | |
5eaef520 | 263 | |
264 | EXEC SQL DECLARE csr_hst1 CURSOR FOR | |
ff20de45 | 265 | SELECT m.name, m.mach_id FROM machine m, serverhosts sh |
5eaef520 | 266 | WHERE sh.service = UPPER(:service) |
267 | AND sh.enable = 1 AND sh.hosterror = 0 | |
ff20de45 | 268 | AND sh.lts < :dfgen AND sh.mach_id = m.mach_id; |
5eaef520 | 269 | EXEC SQL OPEN csr_hst1; |
ff20de45 | 270 | sq = sq_create(); |
5eaef520 | 271 | while (1) |
272 | { | |
ff20de45 | 273 | EXEC SQL FETCH csr_hst1 INTO :host, mid; |
5eaef520 | 274 | if (sqlca.sqlcode == SQL_NO_MATCH) |
275 | break; | |
276 | ||
ff20de45 | 277 | sq_save_data(sq, strdup(strtrim(host))); |
278 | sq_save_data(sq, (void *)mid); | |
279 | } | |
280 | EXEC SQL CLOSE csr_hst1; | |
281 | ||
282 | EXEC SQL WHENEVER SQLERROR GOTO host_cleanup; | |
283 | while (sq_get_data(sq, &name)) | |
284 | { | |
285 | sq_get_data(sq, &mid); | |
286 | com_err(whoami, 0, "sending %s data to %s", service, name); | |
5eaef520 | 287 | EXEC SQL UPDATE serverhosts SET inprogress = 1 |
ff20de45 | 288 | WHERE service = UPPER(:service) AND mach_id = :mid; |
289 | EXEC SQL COMMIT; | |
290 | status = dcm_send_file(service, name, target); | |
5eaef520 | 291 | if (status) |
292 | { | |
293 | errmsg = error_message(status); | |
294 | EXEC SQL UPDATE serverhosts SET hosterrmsg = :errmsg, | |
ff20de45 | 295 | inprogress = 0, success = 0, ltt = :now |
dae7c89e | 296 | WHERE service = UPPER(:service) AND mach_id = :mid; |
5eaef520 | 297 | if (!SOFT_FAIL(status)) |
298 | { | |
299 | EXEC SQL UPDATE serverhosts SET hosterror = :status | |
dae7c89e | 300 | WHERE service = UPPER(:service) AND mach_id = :mid; |
5eaef520 | 301 | critical_alert("DCM", "DCM updating %s:%s: %s", |
ff20de45 | 302 | service, name, errmsg); |
2a2a3914 | 303 | } |
ff20de45 | 304 | EXEC SQL COMMIT; |
5eaef520 | 305 | |
306 | if (replicated) | |
307 | break; | |
2a2a3914 | 308 | } |
309 | } | |
ff20de45 | 310 | sq_destroy(sq); |
5eaef520 | 311 | |
312 | if (status == MR_SUCCESS || !replicated) | |
313 | { | |
314 | EXEC SQL DECLARE csr_hst2 CURSOR FOR | |
ff20de45 | 315 | SELECT m.name, m.mach_id FROM machine m, serverhosts sh |
5eaef520 | 316 | WHERE sh.service = UPPER(:service) AND sh.inprogress = 1 |
ff20de45 | 317 | AND sh.mach_id = m.mach_id; |
5eaef520 | 318 | EXEC SQL OPEN csr_hst2; |
ff20de45 | 319 | sq = sq_create(); |
5eaef520 | 320 | |
321 | while (1) | |
322 | { | |
ff20de45 | 323 | EXEC SQL FETCH csr_hst2 INTO :host, :mid; |
5eaef520 | 324 | if (sqlca.sqlcode == SQL_NO_MATCH) |
325 | break; | |
326 | ||
ff20de45 | 327 | sq_save_data(sq, strdup(strtrim(host))); |
328 | sq_save_data(sq, (void *)mid); | |
329 | } | |
330 | EXEC SQL CLOSE csr_hst2; | |
331 | ||
332 | while (sq_get_data(sq, &name)) | |
333 | { | |
334 | sq_get_data(sq, &mid); | |
335 | ||
336 | com_err(whoami, 0, "executing instructions on %s", name); | |
337 | status = dcm_execute(service, name, script); | |
5eaef520 | 338 | if (status) |
339 | { | |
340 | errmsg = error_message(status); | |
341 | EXEC SQL UPDATE serverhosts SET hosterrmsg = :errmsg, | |
ff20de45 | 342 | inprogress = 0, success = 0, ltt = :now |
343 | WHERE service = UPPER(:service) AND mach_id = :mid; | |
5eaef520 | 344 | if (!SOFT_FAIL(status)) |
345 | { | |
346 | EXEC SQL UPDATE serverhosts SET hosterror = :status | |
ff20de45 | 347 | WHERE service = UPPER(:service) AND mach_id = :mid; |
5eaef520 | 348 | critical_alert("DCM", "DCM updating %s:%s: %s", |
f272b0e1 | 349 | service, name, errmsg); |
2a2a3914 | 350 | } |
5eaef520 | 351 | |
352 | if (replicated) | |
353 | { | |
354 | /* We're giving up, so clear the inprogress flag on | |
355 | any hosts in this service we haven't gotten to yet */ | |
356 | EXEC SQL UPDATE serverhosts SET inprogress = 0 | |
357 | WHERE service = UPPER(:service); | |
358 | break; | |
2a2a3914 | 359 | } |
5eaef520 | 360 | } |
361 | else | |
362 | { | |
ff20de45 | 363 | EXEC SQL UPDATE serverhosts SET inprogress = 0, ltt = :now, |
364 | lts = :now, success = 1 WHERE service = UPPER(:service) | |
365 | AND mach_id = :mid; | |
2a2a3914 | 366 | } |
ff20de45 | 367 | EXEC SQL COMMIT; |
2a2a3914 | 368 | } |
5eaef520 | 369 | EXEC SQL CLOSE csr_hst2; |
2a2a3914 | 370 | } |
371 | ||
ff20de45 | 372 | EXEC SQL WHENEVER SQLERROR DO dbmserr(); |
ed43b350 | 373 | if (status && !SOFT_FAIL(status) && replicated) |
5eaef520 | 374 | { |
375 | EXEC SQL UPDATE servers SET harderror = :status, errmsg = :errmsg | |
376 | WHERE name = UPPER(:service); | |
2a2a3914 | 377 | } |
378 | ||
5eaef520 | 379 | EXEC SQL COMMIT RELEASE; |
ff20de45 | 380 | return; |
381 | ||
382 | host_cleanup: | |
383 | EXEC SQL UPDATE serverhosts SET inprogress = 0, success = 0, ltt = :now, | |
384 | hosterror = MR_INTERNAL, hosterrmsg = 'DBMS Internal Error' | |
385 | WHERE service = UPPER(:service) AND mach_id = :mid; | |
386 | if (replicated) | |
387 | { | |
388 | EXEC SQL UPDATE servers SET harderror = MR_INTERNAL, | |
389 | errmsg = 'DBMS Internal Error' WHERE name = UPPER(:service); | |
390 | } | |
2a2a3914 | 391 | } |
392 | ||
393 | int dcm_send_file(char *service, char *host, char *target) | |
394 | { | |
85330553 | 395 | char data[MAXPATHLEN]; |
396 | int code, conn; | |
5eaef520 | 397 | |
85330553 | 398 | conn = mr_connect_internal(host, "moira_update"); |
399 | if (!conn) | |
5eaef520 | 400 | { |
85330553 | 401 | com_err(whoami, errno, "can't connect to %s", host); |
5eaef520 | 402 | return MR_CANT_CONNECT; |
2a2a3914 | 403 | } |
404 | ||
85330553 | 405 | code = send_auth(conn, host); |
5eaef520 | 406 | if (code) |
407 | { | |
408 | com_err(whoami, code, "authenticating to %s", host); | |
409 | goto done; | |
2a2a3914 | 410 | } |
411 | ||
5eaef520 | 412 | sprintf(data, "%s/%s.out", DCM_DIR, service); |
85330553 | 413 | code = send_file(conn, data, target, 0); |
5eaef520 | 414 | if (code) |
415 | com_err(whoami, code, "sending data to %s", host); | |
2a2a3914 | 416 | |
417 | done: | |
85330553 | 418 | send_quit(conn); |
419 | close(conn); | |
5eaef520 | 420 | return code; |
2a2a3914 | 421 | } |
422 | ||
423 | int dcm_execute(char *service, char *host, char *script) | |
424 | { | |
85330553 | 425 | char inst[MAXPATHLEN]; |
426 | int code, conn; | |
5eaef520 | 427 | |
85330553 | 428 | conn = mr_connect_internal(host, "moira_update"); |
429 | if (!conn) | |
5eaef520 | 430 | { |
85330553 | 431 | com_err(whoami, errno, "can't connect to %s", host); |
5eaef520 | 432 | return MR_CANT_CONNECT; |
2a2a3914 | 433 | } |
434 | ||
85330553 | 435 | code = send_auth(conn, host); |
5eaef520 | 436 | if (code) |
437 | { | |
438 | com_err(whoami, code, "authenticating to %s", host); | |
439 | goto done; | |
2a2a3914 | 440 | } |
441 | ||
5eaef520 | 442 | sprintf(inst, "/tmp/moira-update.XXXXXX"); |
443 | mktemp(inst); | |
85330553 | 444 | code = send_file(conn, script, inst, 0); |
5eaef520 | 445 | if (code) |
446 | { | |
447 | com_err(whoami, code, "sending instructions to %s", host); | |
448 | goto done; | |
2a2a3914 | 449 | } |
450 | ||
85330553 | 451 | code = execute(conn, inst); |
5eaef520 | 452 | if (code) |
453 | com_err(whoami, code, "executing instructions on %s", host); | |
2a2a3914 | 454 | |
455 | done: | |
85330553 | 456 | send_quit(conn); |
457 | close(conn); | |
5eaef520 | 458 | return code; |
459 | } | |
2a2a3914 | 460 | |
461 | void dbmserr(void) | |
462 | { | |
5eaef520 | 463 | EXEC SQL BEGIN DECLARE SECTION; |
464 | char err_msg[256]; | |
465 | EXEC SQL END DECLARE SECTION; | |
466 | int bufsize = 256, msglength = 0; | |
467 | ||
468 | sqlglm(err_msg, &bufsize, &msglength); | |
469 | err_msg[msglength] = '\0'; | |
470 | com_err(whoami, 0, "Encountered SQL error:\n%s", err_msg); | |
471 | com_err(whoami, 0, "exiting"); | |
472 | exit(1); | |
2a2a3914 | 473 | } |