xref: /titanic_41/usr/src/cmd/svc/startd/fork.c (revision b7d3956b92a285d8dac2c7f5f7e28d2ef5347ef8)
1 /*
2  * CDDL HEADER START
3  *
4  * The contents of this file are subject to the terms of the
5  * Common Development and Distribution License (the "License").
6  * You may not use this file except in compliance with the License.
7  *
8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9  * or http://www.opensolaris.org/os/licensing.
10  * See the License for the specific language governing permissions
11  * and limitations under the License.
12  *
13  * When distributing Covered Code, include this CDDL HEADER in each
14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15  * If applicable, add the following below this CDDL HEADER, with the
16  * fields enclosed by brackets "[]" replaced with your own identifying
17  * information: Portions Copyright [yyyy] [name of copyright owner]
18  *
19  * CDDL HEADER END
20  */
21 /*
22  * Copyright 2008 Sun Microsystems, Inc.  All rights reserved.
23  * Use is subject to license terms.
24  */
25 
26 #pragma ident	"%Z%%M%	%I%	%E% SMI"
27 
28 /*
29  * fork.c - safe forking for svc.startd
30  *
31  * fork_configd() and fork_sulogin() are related, special cases that handle the
32  * spawning of specific client processes for svc.startd.
33  */
34 
35 #include <sys/contract/process.h>
36 #include <sys/corectl.h>
37 #include <sys/ctfs.h>
38 #include <sys/stat.h>
39 #include <sys/types.h>
40 #include <sys/uio.h>
41 #include <sys/wait.h>
42 #include <assert.h>
43 #include <errno.h>
44 #include <fcntl.h>
45 #include <libcontract.h>
46 #include <libcontract_priv.h>
47 #include <libscf_priv.h>
48 #include <limits.h>
49 #include <port.h>
50 #include <signal.h>
51 #include <stdarg.h>
52 #include <stdio.h>
53 #include <stdlib.h>
54 #include <string.h>
55 #include <unistd.h>
56 #include <utmpx.h>
57 
58 #include "configd_exit.h"
59 #include "protocol.h"
60 #include "startd.h"
61 
62 static	struct	utmpx	*utmpp;	/* pointer for getutxent() */
63 
64 pid_t
65 startd_fork1(int *forkerr)
66 {
67 	pid_t p;
68 
69 	/*
70 	 * prefork stack
71 	 */
72 	wait_prefork();
73 
74 	p = fork1();
75 
76 	if (p == -1 && forkerr != NULL)
77 		*forkerr = errno;
78 
79 	/*
80 	 * postfork stack
81 	 */
82 	wait_postfork(p);
83 
84 	return (p);
85 }
86 
87 /*
88  * void fork_mount(char *, char *)
89  *   Run mount(1M) with the given options and mount point.  (mount(1M) has much
90  *   hidden knowledge; it's much less correct to reimplement that logic here to
91  *   save a fork(2)/exec(2) invocation.)
92  */
93 int
94 fork_mount(char *path, char *opts)
95 {
96 	pid_t pid;
97 	uint_t tries = 0;
98 	int status;
99 
100 	for (pid = fork1(); pid == -1; pid = fork1()) {
101 		if (++tries > MAX_MOUNT_RETRIES)
102 			return (-1);
103 
104 		(void) sleep(tries);
105 	}
106 
107 	if (pid != 0) {
108 		(void) waitpid(pid, &status, 0);
109 
110 		/*
111 		 * If our mount(1M) invocation exited by peculiar means, or with
112 		 * a non-zero status, our mount likelihood is low.
113 		 */
114 		if (!WIFEXITED(status) ||
115 		    WEXITSTATUS(status) != 0)
116 			return (-1);
117 
118 		return (0);
119 	}
120 
121 	(void) execl("/sbin/mount", "mount", "-o", opts, path, NULL);
122 
123 	return (-1);
124 }
125 
126 /*
127  * pid_t fork_common(...)
128  *   Common routine used by fork_sulogin and fork_configd to fork a
129  *   process in a contract with the provided terms.  Invokes
130  *   fork_sulogin (with its no-fork argument set) on errors.
131  */
132 static pid_t
133 fork_common(const char *name, const char *svc_fmri, int retries, ctid_t *ctidp,
134     uint_t inf, uint_t crit, uint_t fatal, uint_t param, uint64_t cookie)
135 {
136 	uint_t tries = 0;
137 	int ctfd, err;
138 	pid_t pid;
139 
140 	/*
141 	 * Establish process contract terms.
142 	 */
143 	if ((ctfd = open64(CTFS_ROOT "/process/template", O_RDWR)) == -1) {
144 		fork_sulogin(B_TRUE, "Could not open process contract template "
145 		    "for %s: %s\n", name, strerror(errno));
146 		/* NOTREACHED */
147 	}
148 
149 	err = ct_tmpl_set_critical(ctfd, crit);
150 	err |= ct_pr_tmpl_set_fatal(ctfd, fatal);
151 	err |= ct_tmpl_set_informative(ctfd, inf);
152 	err |= ct_pr_tmpl_set_param(ctfd, param);
153 	err |= ct_tmpl_set_cookie(ctfd, cookie);
154 	err |= ct_pr_tmpl_set_svc_fmri(ctfd, svc_fmri);
155 	err |= ct_pr_tmpl_set_svc_aux(ctfd, name);
156 	if (err) {
157 		(void) close(ctfd);
158 		fork_sulogin(B_TRUE, "Could not set %s process contract "
159 		    "terms\n", name);
160 		/* NOTREACHED */
161 	}
162 
163 	if (err = ct_tmpl_activate(ctfd)) {
164 		(void) close(ctfd);
165 		fork_sulogin(B_TRUE, "Could not activate %s process contract "
166 		    "template: %s\n", name, strerror(err));
167 		/* NOTREACHED */
168 	}
169 
170 	/*
171 	 * Attempt to fork "retries" times.
172 	 */
173 	for (pid = fork1(); pid == -1; pid = fork1()) {
174 		if (++tries > retries) {
175 			/*
176 			 * When we exit the sulogin session, init(1M)
177 			 * will restart svc.startd(1M).
178 			 */
179 			err = errno;
180 			(void) ct_tmpl_clear(ctfd);
181 			(void) close(ctfd);
182 			fork_sulogin(B_TRUE, "Could not fork to start %s: %s\n",
183 			    name, strerror(err));
184 			/* NOTREACHED */
185 		}
186 		(void) sleep(tries);
187 	}
188 
189 	/*
190 	 * Clean up, return pid and ctid.
191 	 */
192 	if (pid != 0 && (errno = contract_latest(ctidp)) != 0)
193 		uu_die("Could not get new contract id for %s\n", name);
194 	(void) ct_tmpl_clear(ctfd);
195 	(void) close(ctfd);
196 
197 	return (pid);
198 }
199 
200 /*
201  * void fork_sulogin(boolean_t, const char *, ...)
202  *   When we are invoked with the -s flag from boot (or run into an unfixable
203  *   situation), we run a private copy of sulogin.  When the sulogin session
204  *   is ended, we continue.  This is the last fallback action for system
205  *   maintenance.
206  *
207  *   If immediate is true, fork_sulogin() executes sulogin(1M) directly, without
208  *   forking.
209  *
210  *   Because fork_sulogin() is needed potentially before we daemonize, we leave
211  *   it outside the wait_register() framework.
212  */
213 /*PRINTFLIKE2*/
214 void
215 fork_sulogin(boolean_t immediate, const char *format, ...)
216 {
217 	va_list args;
218 	int fd_console;
219 
220 	(void) printf("Requesting System Maintenance Mode\n");
221 
222 	if (!booting_to_single_user)
223 		(void) printf("(See /lib/svc/share/README for more "
224 		    "information.)\n");
225 
226 	va_start(args, format);
227 	(void) vprintf(format, args);
228 	va_end(args);
229 
230 	if (!immediate) {
231 		ctid_t	ctid;
232 		pid_t	pid;
233 
234 		pid = fork_common("sulogin", SVC_SULOGIN_FMRI,
235 		    MAX_SULOGIN_RETRIES, &ctid, CT_PR_EV_HWERR, 0,
236 		    CT_PR_EV_HWERR, CT_PR_PGRPONLY, SULOGIN_COOKIE);
237 
238 		if (pid != 0) {
239 			(void) waitpid(pid, NULL, 0);
240 			contract_abandon(ctid);
241 			return;
242 		}
243 		/* close all inherited fds */
244 		closefrom(0);
245 	} else {
246 		(void) printf("Directly executing sulogin.\n");
247 		/*
248 		 * Can't call closefrom() in this MT section
249 		 * so safely close a minimum set of fds.
250 		 */
251 		(void) close(STDIN_FILENO);
252 		(void) close(STDOUT_FILENO);
253 		(void) close(STDERR_FILENO);
254 	}
255 
256 	(void) setpgrp();
257 
258 	/* open the console for sulogin */
259 	if ((fd_console = open("/dev/console", O_RDWR)) >= 0) {
260 		if (fd_console != STDIN_FILENO)
261 			while (dup2(fd_console, STDIN_FILENO) < 0 &&
262 			    errno == EINTR)
263 				;
264 		if (fd_console != STDOUT_FILENO)
265 			while (dup2(fd_console, STDOUT_FILENO) < 0 &&
266 			    errno == EINTR)
267 				;
268 		if (fd_console != STDERR_FILENO)
269 			while (dup2(fd_console, STDERR_FILENO) < 0 &&
270 			    errno == EINTR)
271 				;
272 		if (fd_console > STDERR_FILENO)
273 			(void) close(fd_console);
274 	}
275 
276 	setutxent();
277 	while ((utmpp = getutxent()) != NULL) {
278 		if (strcmp(utmpp->ut_user, "LOGIN") != 0) {
279 			if (strcmp(utmpp->ut_line, "console") == 0) {
280 				(void) kill(utmpp->ut_pid, 9);
281 				break;
282 			}
283 		}
284 	}
285 
286 	(void) execl("/sbin/sulogin", "sulogin", NULL);
287 
288 	uu_warn("Could not exec() sulogin");
289 
290 	exit(1);
291 }
292 
293 #define	CONFIGD_PATH	"/lib/svc/bin/svc.configd"
294 
295 /*
296  * void fork_configd(int status)
297  *   We are interested in exit events (since the parent's exiting means configd
298  *   is ready to run and since the child's exiting indicates an error case) and
299  *   in empty events.  This means we have a unique template for initiating
300  *   configd.
301  */
302 /*ARGSUSED*/
303 void
304 fork_configd(int exitstatus)
305 {
306 	pid_t pid;
307 	ctid_t ctid = -1;
308 	int err;
309 	char path[PATH_MAX];
310 
311 retry:
312 	log_framework(LOG_DEBUG, "fork_configd trying to start svc.configd\n");
313 
314 	/*
315 	 * If we're retrying, we will have an old contract lying around
316 	 * from the failure.  Since we're going to be creating a new
317 	 * contract shortly, we abandon the old one now.
318 	 */
319 	if (ctid != -1)
320 		contract_abandon(ctid);
321 	ctid = -1;
322 
323 	pid = fork_common("svc.configd", SCF_SERVICE_CONFIGD,
324 	    MAX_CONFIGD_RETRIES, &ctid, 0, CT_PR_EV_EXIT, 0,
325 	    CT_PR_INHERIT | CT_PR_REGENT, CONFIGD_COOKIE);
326 
327 	if (pid != 0) {
328 		int exitstatus;
329 
330 		st->st_configd_pid = pid;
331 
332 		if (waitpid(pid, &exitstatus, 0) == -1) {
333 			fork_sulogin(B_FALSE, "waitpid on svc.configd "
334 			    "failed: %s\n", strerror(errno));
335 		} else if (WIFEXITED(exitstatus)) {
336 			char *errstr;
337 
338 			/*
339 			 * Examine exitstatus.  This will eventually get more
340 			 * complicated, as we will want to teach startd how to
341 			 * invoke configd with alternate repositories, etc.
342 			 *
343 			 * Note that exec(2) failure results in an exit status
344 			 * of 1, resulting in the default clause below.
345 			 */
346 
347 			/*
348 			 * Assign readable strings to cases we don't handle, or
349 			 * have error outcomes that cannot be eliminated.
350 			 */
351 			switch (WEXITSTATUS(exitstatus)) {
352 			case CONFIGD_EXIT_BAD_ARGS:
353 				errstr = "bad arguments";
354 				break;
355 
356 			case CONFIGD_EXIT_DATABASE_BAD:
357 				errstr = "database corrupt";
358 				break;
359 
360 			case CONFIGD_EXIT_DATABASE_LOCKED:
361 				errstr = "database locked";
362 				break;
363 			case CONFIGD_EXIT_INIT_FAILED:
364 				errstr = "initialization failure";
365 				break;
366 			case CONFIGD_EXIT_DOOR_INIT_FAILED:
367 				errstr = "door initialization failure";
368 				break;
369 			case CONFIGD_EXIT_DATABASE_INIT_FAILED:
370 				errstr = "database initialization failure";
371 				break;
372 			case CONFIGD_EXIT_NO_THREADS:
373 				errstr = "no threads available";
374 				break;
375 			case CONFIGD_EXIT_LOST_MAIN_DOOR:
376 				errstr = "lost door server attachment";
377 				break;
378 			case 1:
379 				errstr = "execution failure";
380 				break;
381 			default:
382 				errstr = "unknown error";
383 				break;
384 			}
385 
386 			/*
387 			 * Remedial actions for various configd failures.
388 			 */
389 			switch (WEXITSTATUS(exitstatus)) {
390 			case CONFIGD_EXIT_OKAY:
391 				break;
392 
393 			case CONFIGD_EXIT_DATABASE_LOCKED:
394 				/* attempt remount of / read-write */
395 				if (fs_is_read_only("/", NULL) == 1) {
396 					if (fs_remount("/") == -1)
397 						fork_sulogin(B_FALSE,
398 						    "remount of root "
399 						    "filesystem failed\n");
400 
401 					goto retry;
402 				}
403 				break;
404 
405 			default:
406 				fork_sulogin(B_FALSE, "svc.configd exited "
407 				    "with status %d (%s)\n",
408 				    WEXITSTATUS(exitstatus), errstr);
409 				goto retry;
410 			}
411 		} else if (WIFSIGNALED(exitstatus)) {
412 			char signame[SIG2STR_MAX];
413 
414 			if (sig2str(WTERMSIG(exitstatus), signame))
415 				(void) snprintf(signame, SIG2STR_MAX,
416 				    "signum %d", WTERMSIG(exitstatus));
417 
418 			fork_sulogin(B_FALSE, "svc.configd signalled:"
419 			    " %s\n", signame);
420 
421 			goto retry;
422 		} else {
423 			fork_sulogin(B_FALSE, "svc.configd non-exit "
424 			    "condition: 0x%x\n", exitstatus);
425 
426 			goto retry;
427 		}
428 
429 		/*
430 		 * Announce that we have a valid svc.configd status.
431 		 */
432 		MUTEX_LOCK(&st->st_configd_live_lock);
433 		st->st_configd_lives = 1;
434 		err = pthread_cond_broadcast(&st->st_configd_live_cv);
435 		assert(err == 0);
436 		MUTEX_UNLOCK(&st->st_configd_live_lock);
437 
438 		log_framework(LOG_DEBUG, "fork_configd broadcasts configd is "
439 		    "live\n");
440 		return;
441 	}
442 
443 	/*
444 	 * Set our per-process core file path to leave core files in
445 	 * /etc/svc/volatile directory, named after the PID to aid in debugging.
446 	 */
447 	(void) snprintf(path, sizeof (path),
448 	    "/etc/svc/volatile/core.configd.%%p");
449 
450 	(void) core_set_process_path(path, strlen(path) + 1, getpid());
451 
452 	log_framework(LOG_DEBUG, "executing svc.configd\n");
453 
454 	(void) execl(CONFIGD_PATH, CONFIGD_PATH, NULL);
455 
456 	/*
457 	 * Status code is used above to identify configd exec failure.
458 	 */
459 	exit(1);
460 }
461 
462 void *
463 fork_configd_thread(void *vctid)
464 {
465 	int fd, err;
466 	ctid_t configd_ctid = (ctid_t)vctid;
467 
468 	if (configd_ctid == -1) {
469 		log_framework(LOG_DEBUG,
470 		    "fork_configd_thread starting svc.configd\n");
471 		fork_configd(0);
472 	} else {
473 		/*
474 		 * configd_ctid is known:  we broadcast and continue.
475 		 * test contract for appropriate state by verifying that
476 		 * there is one or more processes within it?
477 		 */
478 		log_framework(LOG_DEBUG,
479 		    "fork_configd_thread accepting svc.configd with CTID %ld\n",
480 		    configd_ctid);
481 		MUTEX_LOCK(&st->st_configd_live_lock);
482 		st->st_configd_lives = 1;
483 		(void) pthread_cond_broadcast(&st->st_configd_live_cv);
484 		MUTEX_UNLOCK(&st->st_configd_live_lock);
485 	}
486 
487 	fd = open64(CTFS_ROOT "/process/pbundle", O_RDONLY);
488 	if (fd == -1)
489 		uu_die("process bundle open failed");
490 
491 	/*
492 	 * Make sure we get all events (including those generated by configd
493 	 * before this thread was started).
494 	 */
495 	err = ct_event_reset(fd);
496 	assert(err == 0);
497 
498 	for (;;) {
499 		int efd, sfd;
500 		ct_evthdl_t ev;
501 		uint32_t type;
502 		ctevid_t evid;
503 		ct_stathdl_t status;
504 		ctid_t ctid;
505 		uint64_t cookie;
506 		pid_t pid;
507 
508 		if (err = ct_event_read_critical(fd, &ev)) {
509 			assert(err != EINVAL && err != EAGAIN);
510 			log_error(LOG_WARNING,
511 			    "Error reading next contract event: %s",
512 			    strerror(err));
513 			continue;
514 		}
515 
516 		evid = ct_event_get_evid(ev);
517 		ctid = ct_event_get_ctid(ev);
518 		type = ct_event_get_type(ev);
519 
520 		/* Fetch cookie. */
521 		sfd = contract_open(ctid, "process", "status", O_RDONLY);
522 		if (sfd < 0) {
523 			ct_event_free(ev);
524 			continue;
525 		}
526 
527 		if (err = ct_status_read(sfd, CTD_COMMON, &status)) {
528 			log_framework(LOG_WARNING, "Could not get status for "
529 			    "contract %ld: %s\n", ctid, strerror(err));
530 
531 			ct_event_free(ev);
532 			startd_close(sfd);
533 			continue;
534 		}
535 
536 		cookie = ct_status_get_cookie(status);
537 
538 		ct_status_free(status);
539 
540 		startd_close(sfd);
541 
542 		/*
543 		 * Don't process events from contracts we aren't interested in.
544 		 */
545 		if (cookie != CONFIGD_COOKIE) {
546 			ct_event_free(ev);
547 			continue;
548 		}
549 
550 		if (type == CT_PR_EV_EXIT) {
551 			int exitstatus;
552 
553 			(void) ct_pr_event_get_pid(ev, &pid);
554 			(void) ct_pr_event_get_exitstatus(ev,
555 			    &exitstatus);
556 
557 			if (st->st_configd_pid != pid) {
558 				/*
559 				 * This is the child exiting, so we
560 				 * abandon the contract and restart
561 				 * configd.
562 				 */
563 				contract_abandon(ctid);
564 				fork_configd(exitstatus);
565 			}
566 		}
567 
568 		efd = contract_open(ctid, "process", "ctl", O_WRONLY);
569 		if (efd != -1) {
570 			(void) ct_ctl_ack(efd, evid);
571 			startd_close(efd);
572 		}
573 
574 		ct_event_free(ev);
575 
576 	}
577 
578 	/*NOTREACHED*/
579 	return (NULL);
580 }
581 
582 void
583 fork_rc_script(char rl, const char *arg, boolean_t wait)
584 {
585 	pid_t pid;
586 	int tmpl, err, stat;
587 	char path[20] = "/sbin/rc.", log[20] = "rc..log", timebuf[20];
588 	time_t now;
589 	struct tm ltime;
590 	size_t sz;
591 	char *pathenv;
592 	char **nenv;
593 
594 	path[8] = rl;
595 
596 	tmpl = open64(CTFS_ROOT "/process/template", O_RDWR);
597 	if (tmpl >= 0) {
598 		err = ct_tmpl_set_critical(tmpl, 0);
599 		assert(err == 0);
600 
601 		err = ct_tmpl_set_informative(tmpl, 0);
602 		assert(err == 0);
603 
604 		err = ct_pr_tmpl_set_fatal(tmpl, 0);
605 		assert(err == 0);
606 
607 		err = ct_tmpl_activate(tmpl);
608 		assert(err == 0);
609 
610 		err = close(tmpl);
611 		assert(err == 0);
612 	} else {
613 		uu_warn("Could not create contract template for %s.\n", path);
614 	}
615 
616 	pid = startd_fork1(NULL);
617 	if (pid < 0) {
618 		return;
619 	} else if (pid != 0) {
620 		/* parent */
621 		if (wait) {
622 			do
623 				err = waitpid(pid, &stat, 0);
624 			while (err != 0 && errno == EINTR)
625 				;
626 
627 			if (!WIFEXITED(stat)) {
628 				log_framework(LOG_INFO,
629 				    "%s terminated with waitpid() status %d.\n",
630 				    path, stat);
631 			} else if (WEXITSTATUS(stat) != 0) {
632 				log_framework(LOG_INFO,
633 				    "%s failed with status %d.\n", path,
634 				    WEXITSTATUS(stat));
635 			}
636 		}
637 
638 		return;
639 	}
640 
641 	/* child */
642 
643 	log[2] = rl;
644 
645 	setlog(log);
646 
647 	now = time(NULL);
648 	sz = strftime(timebuf, sizeof (timebuf), "%b %e %T",
649 	    localtime_r(&now, &ltime));
650 	assert(sz != 0);
651 
652 	(void) fprintf(stderr, "%s Executing %s %s\n", timebuf, path, arg);
653 
654 	if (rl == 'S')
655 		pathenv = "PATH=/sbin:/usr/sbin:/usr/bin";
656 	else
657 		pathenv = "PATH=/usr/sbin:/usr/bin";
658 
659 	nenv = set_smf_env(NULL, 0, pathenv, NULL, NULL);
660 
661 	(void) execle(path, path, arg, 0, nenv);
662 
663 	perror("exec");
664 	exit(0);
665 }
666