xref: /illumos-gate/usr/src/uts/common/os/exit.c (revision 9113a79cf228b8f7bd509b1328adf88659dfe218)
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 /*
23  * Copyright 2006 Sun Microsystems, Inc.  All rights reserved.
24  * Use is subject to license terms.
25  */
26 
27 /*	Copyright (c) 1984, 1986, 1987, 1988, 1989 AT&T	*/
28 
29 #pragma ident	"%Z%%M%	%I%	%E% SMI"	/* from SVr4.0 1.74 */
30 
31 #include <sys/types.h>
32 #include <sys/param.h>
33 #include <sys/sysmacros.h>
34 #include <sys/systm.h>
35 #include <sys/cred.h>
36 #include <sys/user.h>
37 #include <sys/errno.h>
38 #include <sys/proc.h>
39 #include <sys/ucontext.h>
40 #include <sys/procfs.h>
41 #include <sys/vnode.h>
42 #include <sys/acct.h>
43 #include <sys/var.h>
44 #include <sys/cmn_err.h>
45 #include <sys/debug.h>
46 #include <sys/wait.h>
47 #include <sys/siginfo.h>
48 #include <sys/procset.h>
49 #include <sys/class.h>
50 #include <sys/file.h>
51 #include <sys/session.h>
52 #include <sys/kmem.h>
53 #include <sys/vtrace.h>
54 #include <sys/prsystm.h>
55 #include <sys/ipc.h>
56 #include <sys/sem_impl.h>
57 #include <c2/audit.h>
58 #include <sys/aio_impl.h>
59 #include <vm/as.h>
60 #include <sys/poll.h>
61 #include <sys/door.h>
62 #include <sys/lwpchan_impl.h>
63 #include <sys/utrap.h>
64 #include <sys/task.h>
65 #include <sys/exacct.h>
66 #include <sys/cyclic.h>
67 #include <sys/schedctl.h>
68 #include <sys/rctl.h>
69 #include <sys/contract_impl.h>
70 #include <sys/contract/process_impl.h>
71 #include <sys/list.h>
72 #include <sys/dtrace.h>
73 #include <sys/pool.h>
74 #include <sys/sdt.h>
75 #include <sys/corectl.h>
76 #include <sys/brand.h>
77 #include <sys/libc_kernel.h>
78 
79 /*
80  * convert code/data pair into old style wait status
81  */
82 int
83 wstat(int code, int data)
84 {
85 	int stat = (data & 0377);
86 
87 	switch (code) {
88 	case CLD_EXITED:
89 		stat <<= 8;
90 		break;
91 	case CLD_DUMPED:
92 		stat |= WCOREFLG;
93 		break;
94 	case CLD_KILLED:
95 		break;
96 	case CLD_TRAPPED:
97 	case CLD_STOPPED:
98 		stat <<= 8;
99 		stat |= WSTOPFLG;
100 		break;
101 	case CLD_CONTINUED:
102 		stat = WCONTFLG;
103 		break;
104 	default:
105 		cmn_err(CE_PANIC, "wstat: bad code");
106 		/* NOTREACHED */
107 	}
108 	return (stat);
109 }
110 
111 static char *
112 exit_reason(char *buf, size_t bufsz, int what, int why)
113 {
114 	switch (why) {
115 	case CLD_EXITED:
116 		(void) snprintf(buf, bufsz, "exited with status %d", what);
117 		break;
118 	case CLD_KILLED:
119 		(void) snprintf(buf, bufsz, "exited on fatal signal %d", what);
120 		break;
121 	case CLD_DUMPED:
122 		(void) snprintf(buf, bufsz, "core dumped on signal %d", what);
123 		break;
124 	default:
125 		(void) snprintf(buf, bufsz, "encountered unknown error "
126 		    "(%d, %d)", why, what);
127 		break;
128 	}
129 
130 	return (buf);
131 }
132 
133 /*
134  * exit system call: pass back caller's arg.
135  */
136 void
137 rexit(int rval)
138 {
139 	exit(CLD_EXITED, rval);
140 }
141 
142 /*
143  * Called by proc_exit() when a zone's init exits, presumably because
144  * it failed.  As long as the given zone is still in the "running"
145  * state, we will re-exec() init, but first we need to reset things
146  * which are usually inherited across exec() but will break init's
147  * assumption that it is being exec()'d from a virgin process.  Most
148  * importantly this includes closing all file descriptors (exec only
149  * closes those marked close-on-exec) and resetting signals (exec only
150  * resets handled signals, and we need to clear any signals which
151  * killed init).  Anything else that exec(2) says would be inherited,
152  * but would affect the execution of init, needs to be reset.
153  */
154 static int
155 restart_init(int what, int why)
156 {
157 	kthread_t *t = curthread;
158 	klwp_t *lwp = ttolwp(t);
159 	proc_t *p = ttoproc(t);
160 	user_t *up = PTOU(p);
161 
162 	vnode_t *oldcd, *oldrd;
163 	int i, err;
164 	char reason_buf[64];
165 
166 	/*
167 	 * Let zone admin (and global zone admin if this is for a non-global
168 	 * zone) know that init has failed and will be restarted.
169 	 */
170 	zcmn_err(p->p_zone->zone_id, CE_WARN,
171 	    "init(1M) %s: restarting automatically",
172 	    exit_reason(reason_buf, sizeof (reason_buf), what, why));
173 
174 	if (!INGLOBALZONE(p)) {
175 		cmn_err(CE_WARN, "init(1M) for zone %s (pid %d) %s: "
176 		    "restarting automatically",
177 		    p->p_zone->zone_name, p->p_pid, reason_buf);
178 	}
179 
180 	/*
181 	 * Remove any fpollinfo_t's for this (last) thread from our file
182 	 * descriptors so closeall() can ASSERT() that they're all gone.
183 	 * Then close all open file descriptors in the process.
184 	 */
185 	pollcleanup();
186 	closeall(P_FINFO(p));
187 
188 	/*
189 	 * Grab p_lock and begin clearing miscellaneous global process
190 	 * state that needs to be reset before we exec the new init(1M).
191 	 */
192 
193 	mutex_enter(&p->p_lock);
194 	prbarrier(p);
195 
196 	p->p_flag &= ~(SKILLED | SEXTKILLED | SEXITING | SDOCORE);
197 	up->u_cmask = CMASK;
198 
199 	sigemptyset(&t->t_hold);
200 	sigemptyset(&t->t_sig);
201 	sigemptyset(&t->t_extsig);
202 
203 	sigemptyset(&p->p_sig);
204 	sigemptyset(&p->p_extsig);
205 
206 	sigdelq(p, t, 0);
207 	sigdelq(p, NULL, 0);
208 
209 	if (p->p_killsqp) {
210 		siginfofree(p->p_killsqp);
211 		p->p_killsqp = NULL;
212 	}
213 
214 	/*
215 	 * Reset any signals that are ignored back to the default disposition.
216 	 * Other u_signal members will be cleared when exec calls sigdefault().
217 	 */
218 	for (i = 1; i < NSIG; i++) {
219 		if (up->u_signal[i - 1] == SIG_IGN) {
220 			up->u_signal[i - 1] = SIG_DFL;
221 			sigemptyset(&up->u_sigmask[i - 1]);
222 		}
223 	}
224 
225 	/*
226 	 * Clear the current signal, any signal info associated with it, and
227 	 * any signal information from contracts and/or contract templates.
228 	 */
229 	lwp->lwp_cursig = 0;
230 	lwp->lwp_extsig = 0;
231 	if (lwp->lwp_curinfo != NULL) {
232 		siginfofree(lwp->lwp_curinfo);
233 		lwp->lwp_curinfo = NULL;
234 	}
235 	lwp_ctmpl_clear(lwp);
236 
237 	/*
238 	 * Reset both the process root directory and the current working
239 	 * directory to the root of the zone just as we do during boot.
240 	 */
241 	VN_HOLD(p->p_zone->zone_rootvp);
242 	oldrd = up->u_rdir;
243 	up->u_rdir = p->p_zone->zone_rootvp;
244 
245 	VN_HOLD(p->p_zone->zone_rootvp);
246 	oldcd = up->u_cdir;
247 	up->u_cdir = p->p_zone->zone_rootvp;
248 
249 	if (up->u_cwd != NULL) {
250 		refstr_rele(up->u_cwd);
251 		up->u_cwd = NULL;
252 	}
253 
254 	mutex_exit(&p->p_lock);
255 
256 	if (oldrd != NULL)
257 		VN_RELE(oldrd);
258 	if (oldcd != NULL)
259 		VN_RELE(oldcd);
260 
261 	/* Free the controlling tty.  (freectty() always assumes curproc.) */
262 	ASSERT(p == curproc);
263 	(void) freectty(B_TRUE);
264 
265 	/*
266 	 * Now exec() the new init(1M) on top of the current process.  If we
267 	 * succeed, the caller will treat this like a successful system call.
268 	 * If we fail, we issue messages and the caller will proceed with exit.
269 	 */
270 	err = exec_init(p->p_zone->zone_initname, NULL);
271 
272 	if (err == 0)
273 		return (0);
274 
275 	zcmn_err(p->p_zone->zone_id, CE_WARN,
276 	    "failed to restart init(1M) (err=%d): system reboot required", err);
277 
278 	if (!INGLOBALZONE(p)) {
279 		cmn_err(CE_WARN, "failed to restart init(1M) for zone %s "
280 		    "(pid %d, err=%d): zoneadm(1M) boot required",
281 		    p->p_zone->zone_name, p->p_pid, err);
282 	}
283 
284 	return (-1);
285 }
286 
287 /*
288  * Release resources.
289  * Enter zombie state.
290  * Wake up parent and init processes,
291  * and dispose of children.
292  */
293 void
294 exit(int why, int what)
295 {
296 	/*
297 	 * If proc_exit() fails, then some other lwp in the process
298 	 * got there first.  We just have to call lwp_exit() to allow
299 	 * the other lwp to finish exiting the process.  Otherwise we're
300 	 * restarting init, and should return.
301 	 */
302 	if (proc_exit(why, what) != 0) {
303 		mutex_enter(&curproc->p_lock);
304 		ASSERT(curproc->p_flag & SEXITLWPS);
305 		lwp_exit();
306 		/* NOTREACHED */
307 	}
308 }
309 
310 /*
311  * Set the SEXITING flag on the process, after making sure /proc does
312  * not have it locked.  This is done in more places than proc_exit(),
313  * so it is a separate function.
314  */
315 void
316 proc_is_exiting(proc_t *p)
317 {
318 	mutex_enter(&p->p_lock);
319 	prbarrier(p);
320 	p->p_flag |= SEXITING;
321 	mutex_exit(&p->p_lock);
322 }
323 
324 /*
325  * Return value:
326  *   1 - exitlwps() failed, call (or continue) lwp_exit()
327  *   0 - restarting init.  Return through system call path
328  */
329 int
330 proc_exit(int why, int what)
331 {
332 	kthread_t *t = curthread;
333 	klwp_t *lwp = ttolwp(t);
334 	proc_t *p = ttoproc(t);
335 	zone_t *z = p->p_zone;
336 	timeout_id_t tmp_id;
337 	int rv;
338 	proc_t *q;
339 	task_t *tk;
340 	vnode_t *exec_vp, *execdir_vp, *cdir, *rdir;
341 	sigqueue_t *sqp;
342 	lwpdir_t *lwpdir;
343 	uint_t lwpdir_sz;
344 	lwpdir_t **tidhash;
345 	uint_t tidhash_sz;
346 	refstr_t *cwd;
347 	hrtime_t hrutime, hrstime;
348 	int evaporate;
349 
350 	/*
351 	 * Stop and discard the process's lwps except for the current one,
352 	 * unless some other lwp beat us to it.  If exitlwps() fails then
353 	 * return and the calling lwp will call (or continue in) lwp_exit().
354 	 */
355 	proc_is_exiting(p);
356 	if (exitlwps(0) != 0)
357 		return (1);
358 
359 	DTRACE_PROC(lwp__exit);
360 	DTRACE_PROC1(exit, int, why);
361 
362 	/*
363 	 * Will perform any brand specific proc exit processing, since this
364 	 * is always the last lwp, will also perform lwp_exit and free brand
365 	 * data
366 	 */
367 	if (PROC_IS_BRANDED(p))
368 		BROP(p)->b_proc_exit(p, lwp);
369 
370 	/*
371 	 * Don't let init exit unless zone_start_init() failed its exec, or
372 	 * we are shutting down the zone or the machine.
373 	 *
374 	 * Since we are single threaded, we don't need to lock the
375 	 * following accesses to zone_proc_initpid.
376 	 */
377 	if (p->p_pid == z->zone_proc_initpid) {
378 		if (z->zone_boot_err == 0 &&
379 		    zone_status_get(z) < ZONE_IS_SHUTTING_DOWN &&
380 		    zone_status_get(global_zone) < ZONE_IS_SHUTTING_DOWN &&
381 		    z->zone_restart_init == B_TRUE &&
382 		    restart_init(what, why) == 0)
383 			return (0);
384 		/*
385 		 * Since we didn't or couldn't restart init, we clear
386 		 * the zone's init state and proceed with exit
387 		 * processing.
388 		 */
389 		z->zone_proc_initpid = -1;
390 	}
391 
392 	/*
393 	 * Allocate a sigqueue now, before we grab locks.
394 	 * It will be given to sigcld(), below.
395 	 * Special case:  If we will be making the process disappear
396 	 * without a trace (for the benefit of posix_spawn() in libc)
397 	 * don't bother to allocate a useless sigqueue.
398 	 */
399 	evaporate = ((p->p_flag & SVFORK) &&
400 	    why == CLD_EXITED && what == _EVAPORATE);
401 	if (!evaporate)
402 		sqp = kmem_zalloc(sizeof (sigqueue_t), KM_SLEEP);
403 
404 	/*
405 	 * revoke any doors created by the process.
406 	 */
407 	if (p->p_door_list)
408 		door_exit();
409 
410 	/*
411 	 * Release schedctl data structures.
412 	 */
413 	if (p->p_pagep)
414 		schedctl_proc_cleanup();
415 
416 	/*
417 	 * make sure all pending kaio has completed.
418 	 */
419 	if (p->p_aio)
420 		aio_cleanup_exit();
421 
422 	/*
423 	 * discard the lwpchan cache.
424 	 */
425 	if (p->p_lcp != NULL)
426 		lwpchan_destroy_cache(0);
427 
428 	/*
429 	 * Clean up any DTrace helper actions or probes for the process.
430 	 */
431 	if (p->p_dtrace_helpers != NULL) {
432 		ASSERT(dtrace_helpers_cleanup != NULL);
433 		(*dtrace_helpers_cleanup)();
434 	}
435 
436 	/* untimeout the realtime timers */
437 	if (p->p_itimer != NULL)
438 		timer_exit();
439 
440 	if ((tmp_id = p->p_alarmid) != 0) {
441 		p->p_alarmid = 0;
442 		(void) untimeout(tmp_id);
443 	}
444 
445 	/*
446 	 * Remove any fpollinfo_t's for this (last) thread from our file
447 	 * descriptors so closeall() can ASSERT() that they're all gone.
448 	 */
449 	pollcleanup();
450 
451 	if (p->p_rprof_cyclic != CYCLIC_NONE) {
452 		mutex_enter(&cpu_lock);
453 		cyclic_remove(p->p_rprof_cyclic);
454 		mutex_exit(&cpu_lock);
455 	}
456 
457 	mutex_enter(&p->p_lock);
458 
459 	/*
460 	 * Clean up any DTrace probes associated with this process.
461 	 */
462 	if (p->p_dtrace_probes) {
463 		ASSERT(dtrace_fasttrap_exit_ptr != NULL);
464 		dtrace_fasttrap_exit_ptr(p);
465 	}
466 
467 	while ((tmp_id = p->p_itimerid) != 0) {
468 		p->p_itimerid = 0;
469 		mutex_exit(&p->p_lock);
470 		(void) untimeout(tmp_id);
471 		mutex_enter(&p->p_lock);
472 	}
473 
474 	lwp_cleanup();
475 
476 	/*
477 	 * We are about to exit; prevent our resource associations from
478 	 * being changed.
479 	 */
480 	pool_barrier_enter();
481 
482 	/*
483 	 * Block the process against /proc now that we have really
484 	 * acquired p->p_lock (to manipulate p_tlist at least).
485 	 */
486 	prbarrier(p);
487 
488 #ifdef	SUN_SRC_COMPAT
489 	if (code == CLD_KILLED)
490 		u.u_acflag |= AXSIG;
491 #endif
492 	sigfillset(&p->p_ignore);
493 	sigemptyset(&p->p_siginfo);
494 	sigemptyset(&p->p_sig);
495 	sigemptyset(&p->p_extsig);
496 	sigemptyset(&t->t_sig);
497 	sigemptyset(&t->t_extsig);
498 	sigemptyset(&p->p_sigmask);
499 	sigdelq(p, t, 0);
500 	lwp->lwp_cursig = 0;
501 	lwp->lwp_extsig = 0;
502 	p->p_flag &= ~(SKILLED | SEXTKILLED);
503 	if (lwp->lwp_curinfo) {
504 		siginfofree(lwp->lwp_curinfo);
505 		lwp->lwp_curinfo = NULL;
506 	}
507 
508 	t->t_proc_flag |= TP_LWPEXIT;
509 	ASSERT(p->p_lwpcnt == 1 && p->p_zombcnt == 0);
510 	prlwpexit(t);		/* notify /proc */
511 	lwp_hash_out(p, t->t_tid);
512 	prexit(p);
513 
514 	p->p_lwpcnt = 0;
515 	p->p_tlist = NULL;
516 	sigqfree(p);
517 	term_mstate(t);
518 	p->p_mterm = gethrtime();
519 
520 	exec_vp = p->p_exec;
521 	execdir_vp = p->p_execdir;
522 	p->p_exec = NULLVP;
523 	p->p_execdir = NULLVP;
524 	mutex_exit(&p->p_lock);
525 	if (exec_vp)
526 		VN_RELE(exec_vp);
527 	if (execdir_vp)
528 		VN_RELE(execdir_vp);
529 
530 	pr_free_watched_pages(p);
531 
532 	closeall(P_FINFO(p));
533 
534 	/* Free the controlling tty.  (freectty() always assumes curproc.) */
535 	ASSERT(p == curproc);
536 	(void) freectty(B_TRUE);
537 
538 #if defined(__sparc)
539 	if (p->p_utraps != NULL)
540 		utrap_free(p);
541 #endif
542 	if (p->p_semacct)			/* IPC semaphore exit */
543 		semexit(p);
544 	rv = wstat(why, what);
545 
546 	acct(rv & 0xff);
547 	exacct_commit_proc(p, rv);
548 
549 	/*
550 	 * Release any resources associated with C2 auditing
551 	 */
552 #ifdef C2_AUDIT
553 	if (audit_active) {
554 		/*
555 		 * audit exit system call
556 		 */
557 		audit_exit(why, what);
558 	}
559 #endif
560 
561 	/*
562 	 * Free address space.
563 	 */
564 	relvm();
565 
566 	/*
567 	 * Release held contracts.
568 	 */
569 	contract_exit(p);
570 
571 	/*
572 	 * Depart our encapsulating process contract.
573 	 */
574 	if ((p->p_flag & SSYS) == 0) {
575 		ASSERT(p->p_ct_process);
576 		contract_process_exit(p->p_ct_process, p, rv);
577 	}
578 
579 	/*
580 	 * Remove pool association, and block if requested by pool_do_bind.
581 	 */
582 	mutex_enter(&p->p_lock);
583 	ASSERT(p->p_pool->pool_ref > 0);
584 	atomic_add_32(&p->p_pool->pool_ref, -1);
585 	p->p_pool = pool_default;
586 	/*
587 	 * Now that our address space has been freed and all other threads
588 	 * in this process have exited, set the PEXITED pool flag.  This
589 	 * tells the pools subsystems to ignore this process if it was
590 	 * requested to rebind this process to a new pool.
591 	 */
592 	p->p_poolflag |= PEXITED;
593 	pool_barrier_exit();
594 	mutex_exit(&p->p_lock);
595 
596 	mutex_enter(&pidlock);
597 
598 	/*
599 	 * Delete this process from the newstate list of its parent. We
600 	 * will put it in the right place in the sigcld in the end.
601 	 */
602 	delete_ns(p->p_parent, p);
603 
604 	/*
605 	 * Reassign the orphans to the next of kin.
606 	 * Don't rearrange init's orphanage.
607 	 */
608 	if ((q = p->p_orphan) != NULL && p != proc_init) {
609 
610 		proc_t *nokp = p->p_nextofkin;
611 
612 		for (;;) {
613 			q->p_nextofkin = nokp;
614 			if (q->p_nextorph == NULL)
615 				break;
616 			q = q->p_nextorph;
617 		}
618 		q->p_nextorph = nokp->p_orphan;
619 		nokp->p_orphan = p->p_orphan;
620 		p->p_orphan = NULL;
621 	}
622 
623 	/*
624 	 * Reassign the children to init.
625 	 * Don't try to assign init's children to init.
626 	 */
627 	if ((q = p->p_child) != NULL && p != proc_init) {
628 		struct proc	*np;
629 		struct proc	*initp = proc_init;
630 		boolean_t	setzonetop = B_FALSE;
631 
632 		if (!INGLOBALZONE(curproc))
633 			setzonetop = B_TRUE;
634 
635 		pgdetach(p);
636 
637 		do {
638 			np = q->p_sibling;
639 			/*
640 			 * Delete it from its current parent new state
641 			 * list and add it to init new state list
642 			 */
643 			delete_ns(q->p_parent, q);
644 
645 			q->p_ppid = 1;
646 			q->p_pidflag &= ~(CLDNOSIGCHLD | CLDWAITPID);
647 			if (setzonetop) {
648 				mutex_enter(&q->p_lock);
649 				q->p_flag |= SZONETOP;
650 				mutex_exit(&q->p_lock);
651 			}
652 			q->p_parent = initp;
653 
654 			/*
655 			 * Since q will be the first child,
656 			 * it will not have a previous sibling.
657 			 */
658 			q->p_psibling = NULL;
659 			if (initp->p_child) {
660 				initp->p_child->p_psibling = q;
661 			}
662 			q->p_sibling = initp->p_child;
663 			initp->p_child = q;
664 			if (q->p_proc_flag & P_PR_PTRACE) {
665 				mutex_enter(&q->p_lock);
666 				sigtoproc(q, NULL, SIGKILL);
667 				mutex_exit(&q->p_lock);
668 			}
669 			/*
670 			 * sigcld() will add the child to parents
671 			 * newstate list.
672 			 */
673 			if (q->p_stat == SZOMB)
674 				sigcld(q, NULL);
675 		} while ((q = np) != NULL);
676 
677 		p->p_child = NULL;
678 		ASSERT(p->p_child_ns == NULL);
679 	}
680 
681 	TRACE_1(TR_FAC_PROC, TR_PROC_EXIT, "proc_exit: %p", p);
682 
683 	mutex_enter(&p->p_lock);
684 	CL_EXIT(curthread); /* tell the scheduler that curthread is exiting */
685 
686 	hrutime = mstate_aggr_state(p, LMS_USER);
687 	hrstime = mstate_aggr_state(p, LMS_SYSTEM);
688 	p->p_utime = (clock_t)NSEC_TO_TICK(hrutime) + p->p_cutime;
689 	p->p_stime = (clock_t)NSEC_TO_TICK(hrstime) + p->p_cstime;
690 
691 	p->p_acct[LMS_USER]	+= p->p_cacct[LMS_USER];
692 	p->p_acct[LMS_SYSTEM]	+= p->p_cacct[LMS_SYSTEM];
693 	p->p_acct[LMS_TRAP]	+= p->p_cacct[LMS_TRAP];
694 	p->p_acct[LMS_TFAULT]	+= p->p_cacct[LMS_TFAULT];
695 	p->p_acct[LMS_DFAULT]	+= p->p_cacct[LMS_DFAULT];
696 	p->p_acct[LMS_KFAULT]	+= p->p_cacct[LMS_KFAULT];
697 	p->p_acct[LMS_USER_LOCK] += p->p_cacct[LMS_USER_LOCK];
698 	p->p_acct[LMS_SLEEP]	+= p->p_cacct[LMS_SLEEP];
699 	p->p_acct[LMS_WAIT_CPU]	+= p->p_cacct[LMS_WAIT_CPU];
700 	p->p_acct[LMS_STOPPED]	+= p->p_cacct[LMS_STOPPED];
701 
702 	p->p_ru.minflt	+= p->p_cru.minflt;
703 	p->p_ru.majflt	+= p->p_cru.majflt;
704 	p->p_ru.nswap	+= p->p_cru.nswap;
705 	p->p_ru.inblock	+= p->p_cru.inblock;
706 	p->p_ru.oublock	+= p->p_cru.oublock;
707 	p->p_ru.msgsnd	+= p->p_cru.msgsnd;
708 	p->p_ru.msgrcv	+= p->p_cru.msgrcv;
709 	p->p_ru.nsignals += p->p_cru.nsignals;
710 	p->p_ru.nvcsw	+= p->p_cru.nvcsw;
711 	p->p_ru.nivcsw	+= p->p_cru.nivcsw;
712 	p->p_ru.sysc	+= p->p_cru.sysc;
713 	p->p_ru.ioch	+= p->p_cru.ioch;
714 
715 	p->p_stat = SZOMB;
716 	p->p_proc_flag &= ~P_PR_PTRACE;
717 	p->p_wdata = what;
718 	p->p_wcode = (char)why;
719 
720 	cdir = PTOU(p)->u_cdir;
721 	rdir = PTOU(p)->u_rdir;
722 	cwd = PTOU(p)->u_cwd;
723 
724 	/*
725 	 * Release resource controls, as they are no longer enforceable.
726 	 */
727 	rctl_set_free(p->p_rctls);
728 
729 	/*
730 	 * Give up task and project memberships.  Decrement tk_nlwps counter
731 	 * for our task.max-lwps resource control.  An extended accounting
732 	 * record, if that facility is active, is scheduled to be written.
733 	 * Zombie processes are false members of task0 for the remainder of
734 	 * their lifetime; no accounting information is recorded for them.
735 	 */
736 	tk = p->p_task;
737 
738 	mutex_enter(&p->p_zone->zone_nlwps_lock);
739 	tk->tk_nlwps--;
740 	tk->tk_proj->kpj_nlwps--;
741 	p->p_zone->zone_nlwps--;
742 	mutex_exit(&p->p_zone->zone_nlwps_lock);
743 	task_detach(p);
744 	p->p_task = task0p;
745 
746 	/*
747 	 * Clear the lwp directory and the lwpid hash table
748 	 * now that /proc can't bother us any more.
749 	 * We free the memory below, after dropping p->p_lock.
750 	 */
751 	lwpdir = p->p_lwpdir;
752 	lwpdir_sz = p->p_lwpdir_sz;
753 	tidhash = p->p_tidhash;
754 	tidhash_sz = p->p_tidhash_sz;
755 	p->p_lwpdir = NULL;
756 	p->p_lwpfree = NULL;
757 	p->p_lwpdir_sz = 0;
758 	p->p_tidhash = NULL;
759 	p->p_tidhash_sz = 0;
760 
761 	/*
762 	 * If the process has context ops installed, call the exit routine
763 	 * on behalf of this last remaining thread. Normally exitpctx() is
764 	 * called during thread_exit() or lwp_exit(), but because this is the
765 	 * last thread in the process, we must call it here. By the time
766 	 * thread_exit() is called (below), the association with the relevant
767 	 * process has been lost.
768 	 *
769 	 * We also free the context here.
770 	 */
771 	if (p->p_pctx) {
772 		kpreempt_disable();
773 		exitpctx(p);
774 		kpreempt_enable();
775 
776 		freepctx(p, 0);
777 	}
778 
779 	/*
780 	 * curthread's proc pointer is changed to point at p0 because
781 	 * curthread's original proc pointer can be freed as soon as
782 	 * the child sends a SIGCLD to its parent.
783 	 */
784 	t->t_procp = &p0;
785 
786 	mutex_exit(&p->p_lock);
787 	if (!evaporate) {
788 		p->p_pidflag &= ~CLDPEND;
789 		sigcld(p, sqp);
790 	} else {
791 		/*
792 		 * Do what sigcld() would do if the disposition
793 		 * of the SIGCHLD signal were set to be ignored.
794 		 */
795 		cv_broadcast(&p->p_srwchan_cv);
796 		freeproc(p);
797 	}
798 	mutex_exit(&pidlock);
799 
800 	task_rele(tk);
801 
802 	kmem_free(lwpdir, lwpdir_sz * sizeof (lwpdir_t));
803 	kmem_free(tidhash, tidhash_sz * sizeof (lwpdir_t *));
804 
805 	/*
806 	 * We don't release u_cdir and u_rdir until SZOMB is set.
807 	 * This protects us against dofusers().
808 	 */
809 	VN_RELE(cdir);
810 	if (rdir)
811 		VN_RELE(rdir);
812 	if (cwd)
813 		refstr_rele(cwd);
814 
815 	lwp_pcb_exit();
816 
817 	thread_exit();
818 	/* NOTREACHED */
819 }
820 
821 /*
822  * Format siginfo structure for wait system calls.
823  */
824 void
825 winfo(proc_t *pp, k_siginfo_t *ip, int waitflag)
826 {
827 	ASSERT(MUTEX_HELD(&pidlock));
828 
829 	bzero(ip, sizeof (k_siginfo_t));
830 	ip->si_signo = SIGCLD;
831 	ip->si_code = pp->p_wcode;
832 	ip->si_pid = pp->p_pid;
833 	ip->si_ctid = PRCTID(pp);
834 	ip->si_zoneid = pp->p_zone->zone_id;
835 	ip->si_status = pp->p_wdata;
836 	ip->si_stime = pp->p_stime;
837 	ip->si_utime = pp->p_utime;
838 
839 	if (waitflag) {
840 		pp->p_wcode = 0;
841 		pp->p_wdata = 0;
842 		pp->p_pidflag &= ~CLDPEND;
843 	}
844 }
845 
846 /*
847  * Wait system call.
848  * Search for a terminated (zombie) child,
849  * finally lay it to rest, and collect its status.
850  * Look also for stopped children,
851  * and pass back status from them.
852  */
853 int
854 waitid(idtype_t idtype, id_t id, k_siginfo_t *ip, int options)
855 {
856 	int found;
857 	proc_t *cp, *pp;
858 	int proc_gone;
859 	int waitflag = !(options & WNOWAIT);
860 
861 	/*
862 	 * Obsolete flag, defined here only for binary compatibility
863 	 * with old statically linked executables.  Delete this when
864 	 * we no longer care about these old and broken applications.
865 	 */
866 #define	_WNOCHLD	0400
867 	options &= ~_WNOCHLD;
868 
869 	if (options == 0 || (options & ~WOPTMASK))
870 		return (EINVAL);
871 
872 	switch (idtype) {
873 	case P_PID:
874 	case P_PGID:
875 		if (id < 0 || id >= maxpid)
876 			return (EINVAL);
877 		/* FALLTHROUGH */
878 	case P_ALL:
879 		break;
880 	default:
881 		return (EINVAL);
882 	}
883 
884 	pp = ttoproc(curthread);
885 
886 	/*
887 	 * lock parent mutex so that sibling chain can be searched.
888 	 */
889 	mutex_enter(&pidlock);
890 
891 	/*
892 	 * if we are only looking for exited processes and child_ns list
893 	 * is empty no reason to look at all children.
894 	 */
895 	if (idtype == P_ALL &&
896 	    (options & ~WNOWAIT) == (WNOHANG | WEXITED) &&
897 	    pp->p_child_ns == NULL) {
898 		if (pp->p_child) {
899 			mutex_exit(&pidlock);
900 			bzero(ip, sizeof (k_siginfo_t));
901 			return (0);
902 		}
903 		mutex_exit(&pidlock);
904 		return (ECHILD);
905 	}
906 
907 	while (pp->p_child != NULL) {
908 
909 		proc_gone = 0;
910 
911 		for (cp = pp->p_child_ns; cp != NULL; cp = cp->p_sibling_ns) {
912 			if (idtype != P_PID && (cp->p_pidflag & CLDWAITPID))
913 				continue;
914 			if (idtype == P_PID && id != cp->p_pid)
915 				continue;
916 			if (idtype == P_PGID && id != cp->p_pgrp)
917 				continue;
918 
919 			switch (cp->p_wcode) {
920 
921 			case CLD_TRAPPED:
922 			case CLD_STOPPED:
923 			case CLD_CONTINUED:
924 				cmn_err(CE_PANIC,
925 				    "waitid: wrong state %d on the p_newstate"
926 				    " list", cp->p_wcode);
927 				break;
928 
929 			case CLD_EXITED:
930 			case CLD_DUMPED:
931 			case CLD_KILLED:
932 				if (!(options & WEXITED)) {
933 					/*
934 					 * Count how many are already gone
935 					 * for good.
936 					 */
937 					proc_gone++;
938 					break;
939 				}
940 				if (!waitflag) {
941 					winfo(cp, ip, 0);
942 				} else {
943 					winfo(cp, ip, 1);
944 					freeproc(cp);
945 				}
946 				mutex_exit(&pidlock);
947 				if (waitflag) {		/* accept SIGCLD */
948 					sigcld_delete(ip);
949 					sigcld_repost();
950 				}
951 				return (0);
952 			}
953 
954 			if (idtype == P_PID)
955 				break;
956 		}
957 
958 		/*
959 		 * Wow! None of the threads on the p_sibling_ns list were
960 		 * interesting threads. Check all the kids!
961 		 */
962 		found = 0;
963 		for (cp = pp->p_child; cp != NULL; cp = cp->p_sibling) {
964 			if (idtype == P_PID && id != cp->p_pid)
965 				continue;
966 			if (idtype == P_PGID && id != cp->p_pgrp)
967 				continue;
968 
969 			switch (cp->p_wcode) {
970 			case CLD_TRAPPED:
971 				if (!(options & WTRAPPED))
972 					break;
973 				winfo(cp, ip, waitflag);
974 				mutex_exit(&pidlock);
975 				if (waitflag) {		/* accept SIGCLD */
976 					sigcld_delete(ip);
977 					sigcld_repost();
978 				}
979 				return (0);
980 
981 			case CLD_STOPPED:
982 				if (!(options & WSTOPPED))
983 					break;
984 				/* Is it still stopped? */
985 				mutex_enter(&cp->p_lock);
986 				if (!jobstopped(cp)) {
987 					mutex_exit(&cp->p_lock);
988 					break;
989 				}
990 				mutex_exit(&cp->p_lock);
991 				winfo(cp, ip, waitflag);
992 				mutex_exit(&pidlock);
993 				if (waitflag) {		/* accept SIGCLD */
994 					sigcld_delete(ip);
995 					sigcld_repost();
996 				}
997 				return (0);
998 
999 			case CLD_CONTINUED:
1000 				if (!(options & WCONTINUED))
1001 					break;
1002 				winfo(cp, ip, waitflag);
1003 				mutex_exit(&pidlock);
1004 				if (waitflag) {		/* accept SIGCLD */
1005 					sigcld_delete(ip);
1006 					sigcld_repost();
1007 				}
1008 				return (0);
1009 
1010 			case CLD_EXITED:
1011 			case CLD_DUMPED:
1012 			case CLD_KILLED:
1013 				if (idtype != P_PID &&
1014 				    (cp->p_pidflag & CLDWAITPID))
1015 					continue;
1016 				/*
1017 				 * Don't complain if a process was found in
1018 				 * the first loop but we broke out of the loop
1019 				 * because of the arguments passed to us.
1020 				 */
1021 				if (proc_gone == 0) {
1022 					cmn_err(CE_PANIC,
1023 					    "waitid: wrong state on the"
1024 					    " p_child list");
1025 				} else {
1026 					break;
1027 				}
1028 			}
1029 
1030 			found++;
1031 
1032 			if (idtype == P_PID)
1033 				break;
1034 		}
1035 
1036 		/*
1037 		 * If we found no interesting processes at all,
1038 		 * break out and return ECHILD.
1039 		 */
1040 		if (found + proc_gone == 0)
1041 			break;
1042 
1043 		if (options & WNOHANG) {
1044 			mutex_exit(&pidlock);
1045 			bzero(ip, sizeof (k_siginfo_t));
1046 			/*
1047 			 * We should set ip->si_signo = SIGCLD,
1048 			 * but there is an SVVS test that expects
1049 			 * ip->si_signo to be zero in this case.
1050 			 */
1051 			return (0);
1052 		}
1053 
1054 		/*
1055 		 * If we found no processes of interest that could
1056 		 * change state while we wait, we don't wait at all.
1057 		 * Get out with ECHILD according to SVID.
1058 		 */
1059 		if (found == proc_gone)
1060 			break;
1061 
1062 		if (!cv_wait_sig_swap(&pp->p_cv, &pidlock)) {
1063 			mutex_exit(&pidlock);
1064 			return (EINTR);
1065 		}
1066 	}
1067 	mutex_exit(&pidlock);
1068 	return (ECHILD);
1069 }
1070 
1071 /*
1072  * The wait() system call trap is no longer invoked by libc.
1073  * It is retained only for the benefit of statically linked applications.
1074  * Delete this when we no longer care about these old and broken applications.
1075  */
1076 int64_t
1077 wait(void)
1078 {
1079 	int error;
1080 	k_siginfo_t info;
1081 	rval_t	r;
1082 
1083 	if (error =  waitid(P_ALL, (id_t)0, &info, WEXITED|WTRAPPED))
1084 		return (set_errno(error));
1085 	r.r_val1 = info.si_pid;
1086 	r.r_val2 = wstat(info.si_code, info.si_status);
1087 	return (r.r_vals);
1088 }
1089 
1090 int
1091 waitsys(idtype_t idtype, id_t id, siginfo_t *infop, int options)
1092 {
1093 	int error;
1094 	k_siginfo_t info;
1095 
1096 	if (error = waitid(idtype, id, &info, options))
1097 		return (set_errno(error));
1098 	if (copyout(&info, infop, sizeof (k_siginfo_t)))
1099 		return (set_errno(EFAULT));
1100 	return (0);
1101 }
1102 
1103 #ifdef _SYSCALL32_IMPL
1104 
1105 int
1106 waitsys32(idtype_t idtype, id_t id, siginfo_t *infop, int options)
1107 {
1108 	int error;
1109 	k_siginfo_t info;
1110 	siginfo32_t info32;
1111 
1112 	if (error = waitid(idtype, id, &info, options))
1113 		return (set_errno(error));
1114 	siginfo_kto32(&info, &info32);
1115 	if (copyout(&info32, infop, sizeof (info32)))
1116 		return (set_errno(EFAULT));
1117 	return (0);
1118 }
1119 
1120 #endif	/* _SYSCALL32_IMPL */
1121 
1122 void
1123 proc_detach(proc_t *p)
1124 {
1125 	proc_t *q;
1126 
1127 	ASSERT(MUTEX_HELD(&pidlock));
1128 
1129 	q = p->p_parent;
1130 	ASSERT(q != NULL);
1131 
1132 	/*
1133 	 * Take it off the newstate list of its parent
1134 	 */
1135 	delete_ns(q, p);
1136 
1137 	if (q->p_child == p) {
1138 		q->p_child = p->p_sibling;
1139 		/*
1140 		 * If the parent has no children, it better not
1141 		 * have any with new states either!
1142 		 */
1143 		ASSERT(q->p_child ? 1 : q->p_child_ns == NULL);
1144 	}
1145 
1146 	if (p->p_sibling) {
1147 		p->p_sibling->p_psibling = p->p_psibling;
1148 	}
1149 
1150 	if (p->p_psibling) {
1151 		p->p_psibling->p_sibling = p->p_sibling;
1152 	}
1153 }
1154 
1155 /*
1156  * Remove zombie children from the process table.
1157  */
1158 void
1159 freeproc(proc_t *p)
1160 {
1161 	proc_t *q;
1162 
1163 	ASSERT(p->p_stat == SZOMB);
1164 	ASSERT(p->p_tlist == NULL);
1165 	ASSERT(MUTEX_HELD(&pidlock));
1166 
1167 	sigdelq(p, NULL, 0);
1168 	if (p->p_killsqp) {
1169 		siginfofree(p->p_killsqp);
1170 		p->p_killsqp = NULL;
1171 	}
1172 
1173 	prfree(p);	/* inform /proc */
1174 
1175 	/*
1176 	 * Don't free the init processes.
1177 	 * Other dying processes will access it.
1178 	 */
1179 	if (p == proc_init)
1180 		return;
1181 
1182 
1183 	/*
1184 	 * We wait until now to free the cred structure because a
1185 	 * zombie process's credentials may be examined by /proc.
1186 	 * No cred locking needed because there are no threads at this point.
1187 	 */
1188 	upcount_dec(crgetruid(p->p_cred), crgetzoneid(p->p_cred));
1189 	crfree(p->p_cred);
1190 	if (p->p_corefile != NULL) {
1191 		corectl_path_rele(p->p_corefile);
1192 		p->p_corefile = NULL;
1193 	}
1194 	if (p->p_content != NULL) {
1195 		corectl_content_rele(p->p_content);
1196 		p->p_content = NULL;
1197 	}
1198 
1199 	if (p->p_nextofkin && !((p->p_nextofkin->p_flag & SNOWAIT) ||
1200 	    (PTOU(p->p_nextofkin)->u_signal[SIGCLD - 1] == SIG_IGN))) {
1201 		/*
1202 		 * This should still do the right thing since p_utime/stime
1203 		 * get set to the correct value on process exit, so it
1204 		 * should get properly updated
1205 		 */
1206 		p->p_nextofkin->p_cutime += p->p_utime;
1207 		p->p_nextofkin->p_cstime += p->p_stime;
1208 
1209 		p->p_nextofkin->p_cacct[LMS_USER] += p->p_acct[LMS_USER];
1210 		p->p_nextofkin->p_cacct[LMS_SYSTEM] += p->p_acct[LMS_SYSTEM];
1211 		p->p_nextofkin->p_cacct[LMS_TRAP] += p->p_acct[LMS_TRAP];
1212 		p->p_nextofkin->p_cacct[LMS_TFAULT] += p->p_acct[LMS_TFAULT];
1213 		p->p_nextofkin->p_cacct[LMS_DFAULT] += p->p_acct[LMS_DFAULT];
1214 		p->p_nextofkin->p_cacct[LMS_KFAULT] += p->p_acct[LMS_KFAULT];
1215 		p->p_nextofkin->p_cacct[LMS_USER_LOCK]
1216 		    += p->p_acct[LMS_USER_LOCK];
1217 		p->p_nextofkin->p_cacct[LMS_SLEEP] += p->p_acct[LMS_SLEEP];
1218 		p->p_nextofkin->p_cacct[LMS_WAIT_CPU]
1219 		    += p->p_acct[LMS_WAIT_CPU];
1220 		p->p_nextofkin->p_cacct[LMS_STOPPED] += p->p_acct[LMS_STOPPED];
1221 
1222 		p->p_nextofkin->p_cru.minflt	+= p->p_ru.minflt;
1223 		p->p_nextofkin->p_cru.majflt	+= p->p_ru.majflt;
1224 		p->p_nextofkin->p_cru.nswap	+= p->p_ru.nswap;
1225 		p->p_nextofkin->p_cru.inblock	+= p->p_ru.inblock;
1226 		p->p_nextofkin->p_cru.oublock	+= p->p_ru.oublock;
1227 		p->p_nextofkin->p_cru.msgsnd	+= p->p_ru.msgsnd;
1228 		p->p_nextofkin->p_cru.msgrcv	+= p->p_ru.msgrcv;
1229 		p->p_nextofkin->p_cru.nsignals	+= p->p_ru.nsignals;
1230 		p->p_nextofkin->p_cru.nvcsw	+= p->p_ru.nvcsw;
1231 		p->p_nextofkin->p_cru.nivcsw	+= p->p_ru.nivcsw;
1232 		p->p_nextofkin->p_cru.sysc	+= p->p_ru.sysc;
1233 		p->p_nextofkin->p_cru.ioch	+= p->p_ru.ioch;
1234 
1235 	}
1236 
1237 	q = p->p_nextofkin;
1238 	if (q && q->p_orphan == p)
1239 		q->p_orphan = p->p_nextorph;
1240 	else if (q) {
1241 		for (q = q->p_orphan; q; q = q->p_nextorph)
1242 			if (q->p_nextorph == p)
1243 				break;
1244 		ASSERT(q && q->p_nextorph == p);
1245 		q->p_nextorph = p->p_nextorph;
1246 	}
1247 
1248 	proc_detach(p);
1249 	pid_exit(p);	/* frees pid and proc structure */
1250 }
1251 
1252 /*
1253  * Delete process "child" from the newstate list of process "parent"
1254  */
1255 void
1256 delete_ns(proc_t *parent, proc_t *child)
1257 {
1258 	proc_t **ns;
1259 
1260 	ASSERT(MUTEX_HELD(&pidlock));
1261 	ASSERT(child->p_parent == parent);
1262 	for (ns = &parent->p_child_ns; *ns != NULL; ns = &(*ns)->p_sibling_ns) {
1263 		if (*ns == child) {
1264 
1265 			ASSERT((*ns)->p_parent == parent);
1266 
1267 			*ns = child->p_sibling_ns;
1268 			child->p_sibling_ns = NULL;
1269 			return;
1270 		}
1271 	}
1272 }
1273 
1274 /*
1275  * Add process "child" to the new state list of process "parent"
1276  */
1277 void
1278 add_ns(proc_t *parent, proc_t *child)
1279 {
1280 	ASSERT(child->p_sibling_ns == NULL);
1281 	child->p_sibling_ns = parent->p_child_ns;
1282 	parent->p_child_ns = child;
1283 }
1284