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