]>
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 | |
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)) | |
35 | ||
36 | char whobuf[256], *whoami=whobuf, *db="moira"; | |
37 | extern CONNECTION conn; | |
38 | ||
39 | int main(argc, argv) | |
40 | int argc; | |
41 | char *argv[]; | |
42 | { | |
43 | int i; | |
44 | EXEC SQL BEGIN DECLARE SECTION; | |
45 | char buf[16], *name; | |
46 | int enable; | |
47 | EXEC SQL END DECLARE SECTION; | |
48 | struct save_queue *sq; | |
49 | int status; | |
50 | ||
51 | if (strchr(argv[0], '/')) strcpy(whoami, strrchr(argv[0], '/')+1); | |
52 | else strcpy(whoami, argv[0]); | |
53 | umask(7); | |
54 | ||
55 | setvbuf(stderr, NULL, _IOLBF, BUFSIZ); | |
56 | setvbuf(stdout, NULL, _IOLBF, BUFSIZ); | |
57 | ||
58 | initialize_sms_error_table(); | |
59 | initialize_krb_error_table(); | |
60 | ||
61 | /* if services were specified on the command line, do just those ones */ | |
62 | if (argc > 1) { | |
63 | for (i = 1; i < argc; i++) { | |
64 | if (generate_service(argv[i], 1)) | |
65 | do_hosts(argv[i]); | |
66 | } | |
67 | exit(0); | |
68 | } | |
69 | ||
70 | /* if DCM is not enabled, exit after logging */ | |
71 | if (!access(NODCMFILE, F_OK)) { | |
72 | printf("/etc/nodcm exists -- exiting\n"); | |
73 | exit(1); | |
74 | } | |
75 | ||
76 | EXEC SQL CONNECT :db IDENTIFIED BY :db; | |
77 | ||
78 | EXEC SQL SELECT value INTO :enable FROM numvalues | |
79 | WHERE name='dcm_enable'; | |
80 | if (enable == 0) { | |
81 | printf("dcm_enable not set -- exiting\n"); | |
82 | exit(1); | |
83 | } | |
84 | ||
85 | /* fetch list of services */ | |
86 | EXEC SQL DECLARE csr_svc CURSOR FOR SELECT LOWER(name) FROM servers | |
87 | WHERE enable=1 AND harderror=0 AND update_int>0; | |
88 | EXEC SQL OPEN csr_svc; | |
89 | sq = sq_create(); | |
90 | while(1) { | |
91 | EXEC SQL FETCH csr_svc INTO :buf; | |
92 | if (sqlca.sqlcode) break; | |
93 | ||
94 | sq_save_data(sq, strdup(strtrim(buf))); | |
95 | } | |
96 | EXEC SQL CLOSE csr_svc; | |
97 | /* we will repeatedly open and close the db since it seems to get | |
98 | upset if you keep it open across a fork */ | |
99 | EXEC SQL COMMIT RELEASE; | |
100 | ||
101 | /* Now run through list */ | |
102 | while (sq_get_data(sq, &name)) { | |
103 | if (generate_service(name, 0)) { | |
104 | switch (fork()) { | |
105 | case -1: | |
106 | fprintf(stderr, | |
9ce35ab9 | 107 | "dcm: could not fork for service %s -- exiting", |
2a2a3914 | 108 | name); |
109 | exit(1); | |
110 | case 0: | |
111 | sprintf(strchr(whoami, '\0'), " (%s)", name); | |
112 | do_hosts(name); | |
113 | com_err(whoami, 0, "exiting"); | |
114 | exit(0); | |
115 | default: | |
116 | break; | |
117 | } | |
118 | } | |
119 | } | |
120 | ||
121 | /* wait for children */ | |
122 | while (waitpid(0, &status, 0) > 0) ; | |
123 | com_err(whoami, 0, "exiting"); | |
124 | } | |
125 | ||
126 | int generate_service(char *name, int force) | |
127 | { | |
128 | EXEC SQL BEGIN DECLARE SECTION; | |
129 | int interval, dfcheck, status; | |
130 | time_t now; | |
131 | char *errmsg; | |
132 | EXEC SQL END DECLARE SECTION; | |
133 | char dfgen_prog[64], dfgen_cmd[128]; | |
134 | struct sigaction action, prevaction; | |
135 | int waits; | |
136 | ||
137 | EXEC SQL CONNECT :db IDENTIFIED BY :db; | |
138 | ||
139 | EXEC SQL SELECT update_int, dfcheck INTO :interval, :dfcheck | |
140 | FROM servers WHERE name=UPPER(:name); | |
141 | if (sqlca.sqlcode == SQL_NO_MATCH) { | |
142 | com_err(whoami, 0, "No such service `%s'", name); | |
143 | EXEC SQL COMMIT RELEASE; | |
144 | return 0; | |
145 | } | |
146 | ||
147 | time(&now); | |
148 | ||
149 | if ((interval * 60 + dfcheck < now) || force) { | |
150 | sprintf(dfgen_prog, "%s/%s.gen", BIN_DIR, name); | |
151 | if (access(dfgen_prog, F_OK) != 0) { | |
152 | com_err(whoami, 0, "prog %s doesn't exist", dfgen_prog); | |
153 | EXEC SQL COMMIT RELEASE; | |
154 | return 0; | |
155 | } | |
156 | sprintf(dfgen_cmd, "exec %s %s/%s.out", dfgen_prog, DCM_DIR, name); | |
157 | com_err(whoami, 0, "running %s", dfgen_prog); | |
158 | ||
159 | EXEC SQL UPDATE servers SET inprogress=1 | |
160 | WHERE name=UPPER(:name); | |
161 | ||
162 | action.sa_flags = 0; | |
163 | sigemptyset(&action.sa_mask); | |
164 | action.sa_handler = SIG_DFL; | |
165 | sigaction(SIGCHLD, &action, &prevaction); | |
166 | waits = system(dfgen_cmd); | |
167 | sigaction(SIGCHLD, &prevaction, NULL); | |
168 | if (WIFSIGNALED(waits)) { | |
169 | status = MR_COREDUMP; | |
170 | com_err(whoami, status, " %s exited on signal %d", | |
171 | dfgen_prog, WTERMSIG(waits)); | |
172 | } else if (WEXITSTATUS(waits)) { | |
173 | /* extract the process's exit value */ | |
174 | status = WEXITSTATUS(waits) + ERROR_TABLE_BASE_sms; | |
175 | if (status != MR_NO_CHANGE) | |
176 | com_err(whoami, status, "in %s", dfgen_prog); | |
177 | } else status = MR_SUCCESS; | |
178 | ||
179 | if (status == MR_SUCCESS) { | |
180 | EXEC SQL UPDATE servers SET dfgen = :now, dfcheck = :now, | |
181 | inprogress = 0 WHERE name=UPPER(:name); | |
182 | EXEC SQL COMMIT RELEASE; | |
183 | return 1; | |
184 | } else if (status == MR_NO_CHANGE) { | |
185 | EXEC SQL UPDATE servers SET dfcheck = :now, inprogress = 0 | |
186 | WHERE name=UPPER(:name); | |
187 | } else if (SOFT_FAIL(status)) { | |
188 | errmsg = error_message(status); | |
189 | EXEC SQL UPDATE servers SET errmsg = :errmsg, inprogress = 0 | |
190 | WHERE name=UPPER(:name); | |
191 | } else { /* HARD_FAIL(status) */ | |
192 | errmsg = error_message(status); | |
193 | EXEC SQL UPDATE servers SET dfcheck = :now, harderror = :status, | |
194 | errmsg = :errmsg, inprogress = 0 WHERE name=UPPER(:name); | |
2a2a3914 | 195 | critical_alert("DCM","DCM building config files for %s: %s", |
196 | name, errmsg); | |
2a2a3914 | 197 | } |
198 | } | |
199 | EXEC SQL COMMIT RELEASE; | |
200 | return 0; | |
201 | } | |
202 | ||
203 | void do_hosts(char *service) | |
204 | { | |
205 | EXEC SQL BEGIN DECLARE SECTION; | |
206 | char type[16], host[73], target[64], script[128], *errmsg; | |
207 | int status = 0, mid, dfgen, replicated; | |
208 | time_t now; | |
209 | EXEC SQL END DECLARE SECTION; | |
210 | ||
211 | time(&now); | |
212 | gdb_init(); | |
213 | ||
214 | EXEC SQL CONNECT :db IDENTIFIED BY :db; | |
215 | ||
216 | EXEC SQL SELECT dfgen, type, target_file, script | |
217 | INTO :dfgen, :type, :target, :script | |
218 | FROM servers WHERE name=UPPER(:service); | |
219 | replicated = !strncmp(type, "REPLICAT", 8); | |
220 | ||
221 | EXEC SQL DECLARE csr_hst1 CURSOR FOR | |
222 | SELECT m.name FROM machine m, serverhosts sh | |
223 | WHERE sh.service=UPPER(:service) AND sh.enable=1 AND sh.hosterror=0 | |
224 | AND sh.lts<:dfgen AND sh.mach_id=m.mach_id | |
225 | FOR UPDATE OF sh.inprogress, sh.hosterror, sh.hosterrmsg; | |
226 | EXEC SQL OPEN csr_hst1; | |
227 | ||
228 | while (1) { | |
229 | EXEC SQL FETCH csr_hst1 INTO :host; | |
230 | if (sqlca.sqlcode == SQL_NO_MATCH) break; | |
231 | ||
232 | com_err(whoami, 0, "sending %s data to %s", service, strtrim(host)); | |
233 | EXEC SQL UPDATE serverhosts SET inprogress = 1 | |
234 | WHERE CURRENT OF csr_hst1; | |
235 | status = dcm_send_file(service, host, strtrim(target)); | |
236 | if (status) { | |
237 | errmsg = error_message(status); | |
238 | EXEC SQL UPDATE serverhosts SET hosterrmsg = :errmsg, | |
239 | inprogress = 0 WHERE CURRENT OF csr_hst1; | |
240 | if (!SOFT_FAIL(status)) { | |
241 | EXEC SQL UPDATE serverhosts SET hosterror = :status | |
242 | WHERE CURRENT OF csr_hst1; | |
2a2a3914 | 243 | critical_alert("DCM", "DCM updating %s:%s: %s", |
244 | service, host, errmsg); | |
2a2a3914 | 245 | } |
246 | ||
247 | if (replicated) break; | |
248 | } | |
249 | } | |
250 | EXEC SQL CLOSE csr_hst1; | |
251 | ||
252 | if (status == MR_SUCCESS || !replicated) { | |
253 | EXEC SQL DECLARE csr_hst2 CURSOR FOR | |
254 | SELECT m.name FROM machine m, serverhosts sh | |
255 | WHERE sh.service=UPPER(:service) AND sh.inprogress=1 | |
256 | AND sh.mach_id=m.mach_id | |
257 | FOR UPDATE OF sh.hosterror, sh.hosterrmsg, sh.inprogress; | |
258 | EXEC SQL OPEN csr_hst2; | |
259 | ||
260 | while (1) { | |
261 | EXEC SQL FETCH csr_hst2 INTO :host; | |
262 | if (sqlca.sqlcode == SQL_NO_MATCH) break; | |
263 | ||
264 | com_err(whoami, 0, "executing instructions on %s", strtrim(host)); | |
265 | status = dcm_execute(service, host, strtrim(script)); | |
266 | if (status) { | |
267 | errmsg = error_message(status); | |
268 | EXEC SQL UPDATE serverhosts SET hosterrmsg = :errmsg, | |
269 | inprogress = 0 WHERE CURRENT OF csr_hst2; | |
270 | if (!SOFT_FAIL(status)) { | |
271 | EXEC SQL UPDATE serverhosts SET hosterror = :status | |
272 | WHERE CURRENT OF csr_hst2; | |
2a2a3914 | 273 | critical_alert("DCM", "DCM updating %s:%s: %s", |
274 | service, host, errmsg); | |
2a2a3914 | 275 | } |
276 | ||
277 | if (replicated) { | |
278 | /* We're giving up, so clear the inprogress flag on | |
279 | any hosts in this service we haven't gotten to yet */ | |
280 | EXEC SQL UPDATE serverhosts SET inprogress = 0 | |
281 | WHERE service=UPPER(:service); | |
282 | break; | |
283 | } | |
284 | } else { | |
285 | EXEC SQL UPDATE serverhosts SET inprogress=0, lts=:now | |
286 | WHERE CURRENT OF csr_hst2; | |
287 | } | |
288 | } | |
289 | EXEC SQL CLOSE csr_hst2; | |
290 | } | |
291 | ||
292 | if (status && replicated) { | |
293 | EXEC SQL UPDATE servers SET harderror = :status, errmsg = :errmsg | |
294 | WHERE name = UPPER(:service); | |
295 | } | |
296 | ||
297 | EXEC SQL COMMIT RELEASE; | |
298 | } | |
299 | ||
300 | int dcm_send_file(char *service, char *host, char *target) | |
301 | { | |
302 | char addr[256], data[MAXPATHLEN]; | |
303 | int code; | |
304 | ||
305 | sprintf(addr, "%s:moira_update", host); | |
306 | conn = start_server_connection(addr, ""); | |
307 | if (!conn || (connection_status(conn) == CON_STOPPED)) { | |
308 | com_err(whoami, connection_errno(conn), "can't connect to %s", addr); | |
309 | return MR_CANT_CONNECT; | |
310 | } | |
311 | ||
312 | code = send_auth(host); | |
313 | if (code) { | |
314 | com_err(whoami, code, "authenticating to %s", host); | |
315 | goto done; | |
316 | } | |
317 | ||
318 | sprintf(data, "%s/%s.out", DCM_DIR, service); | |
319 | code = send_file(data, target, 1); | |
320 | if (code == MR_UNKNOWN_PROC) code = send_file(data, target, 0); | |
321 | if (code) com_err(whoami, code, "sending data to %s", host); | |
322 | ||
323 | done: | |
324 | send_quit(); | |
325 | sever_connection(conn); | |
326 | return(code); | |
327 | } | |
328 | ||
329 | int dcm_execute(char *service, char *host, char *script) | |
330 | { | |
331 | char addr[256], inst[MAXPATHLEN]; | |
332 | int code; | |
333 | ||
334 | sprintf(addr, "%s:moira_update", host); | |
335 | conn = start_server_connection(addr, ""); | |
336 | if (!conn || (connection_status(conn) == CON_STOPPED)) { | |
337 | com_err(whoami, connection_errno(conn), "can't connect to %s", addr); | |
338 | return MR_CANT_CONNECT; | |
339 | } | |
340 | ||
341 | code = send_auth(host); | |
342 | if (code) { | |
343 | com_err(whoami, code, "authenticating to %s", host); | |
344 | goto done; | |
345 | } | |
346 | ||
347 | sprintf(inst, "/tmp/moira-update.XXXXXX"); | |
348 | mktemp(inst); | |
349 | code = send_file(script, inst, 0); | |
350 | if (code) { | |
351 | com_err(whoami, code, "sending instructions to %s", host); | |
352 | goto done; | |
353 | } | |
354 | ||
355 | code = execute(inst); | |
356 | if (code) com_err(whoami, code, "executing instructions on %s", host); | |
357 | ||
358 | done: | |
359 | send_quit(); | |
360 | sever_connection(conn); | |
361 | return(code); | |
362 | } | |
363 | ||
364 | void dbmserr(void) | |
365 | { | |
366 | EXEC SQL BEGIN DECLARE SECTION; | |
367 | char err_msg[256]; | |
368 | EXEC SQL END DECLARE SECTION; | |
369 | int bufsize=256, msglength=0; | |
370 | ||
371 | sqlglm(err_msg, &bufsize, &msglength); | |
372 | err_msg[msglength]=0; | |
373 | com_err(whoami, 0, "Encountered SQL error:\n%s", err_msg); | |
374 | com_err(whoami, 0, "exiting"); | |
375 | exit(1); | |
376 | } |