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