xref: /illumos-gate/usr/src/uts/common/os/exit.c (revision 3323877de6db742e6657b6081ffe6acd0b007436)
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 2007 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 	if (audit_active) {
553 		/*
554 		 * audit exit system call
555 		 */
556 		audit_exit(why, what);
557 	}
558 
559 	/*
560 	 * Free address space.
561 	 */
562 	relvm();
563 
564 	/*
565 	 * Release held contracts.
566 	 */
567 	contract_exit(p);
568 
569 	/*
570 	 * Depart our encapsulating process contract.
571 	 */
572 	if ((p->p_flag & SSYS) == 0) {
573 		ASSERT(p->p_ct_process);
574 		contract_process_exit(p->p_ct_process, p, rv);
575 	}
576 
577 	/*
578 	 * Remove pool association, and block if requested by pool_do_bind.
579 	 */
580 	mutex_enter(&p->p_lock);
581 	ASSERT(p->p_pool->pool_ref > 0);
582 	atomic_add_32(&p->p_pool->pool_ref, -1);
583 	p->p_pool = pool_default;
584 	/*
585 	 * Now that our address space has been freed and all other threads
586 	 * in this process have exited, set the PEXITED pool flag.  This
587 	 * tells the pools subsystems to ignore this process if it was
588 	 * requested to rebind this process to a new pool.
589 	 */
590 	p->p_poolflag |= PEXITED;
591 	pool_barrier_exit();
592 	mutex_exit(&p->p_lock);
593 
594 	mutex_enter(&pidlock);
595 
596 	/*
597 	 * Delete this process from the newstate list of its parent. We
598 	 * will put it in the right place in the sigcld in the end.
599 	 */
600 	delete_ns(p->p_parent, p);
601 
602 	/*
603 	 * Reassign the orphans to the next of kin.
604 	 * Don't rearrange init's orphanage.
605 	 */
606 	if ((q = p->p_orphan) != NULL && p != proc_init) {
607 
608 		proc_t *nokp = p->p_nextofkin;
609 
610 		for (;;) {
611 			q->p_nextofkin = nokp;
612 			if (q->p_nextorph == NULL)
613 				break;
614 			q = q->p_nextorph;
615 		}
616 		q->p_nextorph = nokp->p_orphan;
617 		nokp->p_orphan = p->p_orphan;
618 		p->p_orphan = NULL;
619 	}
620 
621 	/*
622 	 * Reassign the children to init.
623 	 * Don't try to assign init's children to init.
624 	 */
625 	if ((q = p->p_child) != NULL && p != proc_init) {
626 		struct proc	*np;
627 		struct proc	*initp = proc_init;
628 		boolean_t	setzonetop = B_FALSE;
629 
630 		if (!INGLOBALZONE(curproc))
631 			setzonetop = B_TRUE;
632 
633 		pgdetach(p);
634 
635 		do {
636 			np = q->p_sibling;
637 			/*
638 			 * Delete it from its current parent new state
639 			 * list and add it to init new state list
640 			 */
641 			delete_ns(q->p_parent, q);
642 
643 			q->p_ppid = 1;
644 			q->p_pidflag &= ~(CLDNOSIGCHLD | CLDWAITPID);
645 			if (setzonetop) {
646 				mutex_enter(&q->p_lock);
647 				q->p_flag |= SZONETOP;
648 				mutex_exit(&q->p_lock);
649 			}
650 			q->p_parent = initp;
651 
652 			/*
653 			 * Since q will be the first child,
654 			 * it will not have a previous sibling.
655 			 */
656 			q->p_psibling = NULL;
657 			if (initp->p_child) {
658 				initp->p_child->p_psibling = q;
659 			}
660 			q->p_sibling = initp->p_child;
661 			initp->p_child = q;
662 			if (q->p_proc_flag & P_PR_PTRACE) {
663 				mutex_enter(&q->p_lock);
664 				sigtoproc(q, NULL, SIGKILL);
665 				mutex_exit(&q->p_lock);
666 			}
667 			/*
668 			 * sigcld() will add the child to parents
669 			 * newstate list.
670 			 */
671 			if (q->p_stat == SZOMB)
672 				sigcld(q, NULL);
673 		} while ((q = np) != NULL);
674 
675 		p->p_child = NULL;
676 		ASSERT(p->p_child_ns == NULL);
677 	}
678 
679 	TRACE_1(TR_FAC_PROC, TR_PROC_EXIT, "proc_exit: %p", p);
680 
681 	mutex_enter(&p->p_lock);
682 	CL_EXIT(curthread); /* tell the scheduler that curthread is exiting */
683 
684 	/*
685 	 * Have our task accummulate our resource usage data before they
686 	 * become contaminated by p_cacct etc., and before we renounce
687 	 * membership of the task.
688 	 *
689 	 * We do this regardless of whether or not task accounting is active.
690 	 * This is to avoid having nonsense data reported for this task if
691 	 * task accounting is subsequently enabled. The overhead is minimal;
692 	 * by this point, this process has accounted for the usage of all its
693 	 * LWPs. We nonetheless do the work here, and under the protection of
694 	 * pidlock, so that the movement of the process's usage to the task
695 	 * happens at the same time as the removal of the process from the
696 	 * task, from the point of view of exacct_snapshot_task_usage().
697 	 */
698 	exacct_update_task_mstate(p);
699 
700 	hrutime = mstate_aggr_state(p, LMS_USER);
701 	hrstime = mstate_aggr_state(p, LMS_SYSTEM);
702 	p->p_utime = (clock_t)NSEC_TO_TICK(hrutime) + p->p_cutime;
703 	p->p_stime = (clock_t)NSEC_TO_TICK(hrstime) + p->p_cstime;
704 
705 	p->p_acct[LMS_USER]	+= p->p_cacct[LMS_USER];
706 	p->p_acct[LMS_SYSTEM]	+= p->p_cacct[LMS_SYSTEM];
707 	p->p_acct[LMS_TRAP]	+= p->p_cacct[LMS_TRAP];
708 	p->p_acct[LMS_TFAULT]	+= p->p_cacct[LMS_TFAULT];
709 	p->p_acct[LMS_DFAULT]	+= p->p_cacct[LMS_DFAULT];
710 	p->p_acct[LMS_KFAULT]	+= p->p_cacct[LMS_KFAULT];
711 	p->p_acct[LMS_USER_LOCK] += p->p_cacct[LMS_USER_LOCK];
712 	p->p_acct[LMS_SLEEP]	+= p->p_cacct[LMS_SLEEP];
713 	p->p_acct[LMS_WAIT_CPU]	+= p->p_cacct[LMS_WAIT_CPU];
714 	p->p_acct[LMS_STOPPED]	+= p->p_cacct[LMS_STOPPED];
715 
716 	p->p_ru.minflt	+= p->p_cru.minflt;
717 	p->p_ru.majflt	+= p->p_cru.majflt;
718 	p->p_ru.nswap	+= p->p_cru.nswap;
719 	p->p_ru.inblock	+= p->p_cru.inblock;
720 	p->p_ru.oublock	+= p->p_cru.oublock;
721 	p->p_ru.msgsnd	+= p->p_cru.msgsnd;
722 	p->p_ru.msgrcv	+= p->p_cru.msgrcv;
723 	p->p_ru.nsignals += p->p_cru.nsignals;
724 	p->p_ru.nvcsw	+= p->p_cru.nvcsw;
725 	p->p_ru.nivcsw	+= p->p_cru.nivcsw;
726 	p->p_ru.sysc	+= p->p_cru.sysc;
727 	p->p_ru.ioch	+= p->p_cru.ioch;
728 
729 	p->p_stat = SZOMB;
730 	p->p_proc_flag &= ~P_PR_PTRACE;
731 	p->p_wdata = what;
732 	p->p_wcode = (char)why;
733 
734 	cdir = PTOU(p)->u_cdir;
735 	rdir = PTOU(p)->u_rdir;
736 	cwd = PTOU(p)->u_cwd;
737 
738 	/*
739 	 * Release resource controls, as they are no longer enforceable.
740 	 */
741 	rctl_set_free(p->p_rctls);
742 
743 	/*
744 	 * Give up task and project memberships.  Decrement tk_nlwps counter
745 	 * for our task.max-lwps resource control.  An extended accounting
746 	 * record, if that facility is active, is scheduled to be written.
747 	 * Zombie processes are false members of task0 for the remainder of
748 	 * their lifetime; no accounting information is recorded for them.
749 	 */
750 	tk = p->p_task;
751 
752 	mutex_enter(&p->p_zone->zone_nlwps_lock);
753 	tk->tk_nlwps--;
754 	tk->tk_proj->kpj_nlwps--;
755 	p->p_zone->zone_nlwps--;
756 	mutex_exit(&p->p_zone->zone_nlwps_lock);
757 	task_detach(p);
758 	p->p_task = task0p;
759 
760 	/*
761 	 * Clear the lwp directory and the lwpid hash table
762 	 * now that /proc can't bother us any more.
763 	 * We free the memory below, after dropping p->p_lock.
764 	 */
765 	lwpdir = p->p_lwpdir;
766 	lwpdir_sz = p->p_lwpdir_sz;
767 	tidhash = p->p_tidhash;
768 	tidhash_sz = p->p_tidhash_sz;
769 	p->p_lwpdir = NULL;
770 	p->p_lwpfree = NULL;
771 	p->p_lwpdir_sz = 0;
772 	p->p_tidhash = NULL;
773 	p->p_tidhash_sz = 0;
774 
775 	/*
776 	 * If the process has context ops installed, call the exit routine
777 	 * on behalf of this last remaining thread. Normally exitpctx() is
778 	 * called during thread_exit() or lwp_exit(), but because this is the
779 	 * last thread in the process, we must call it here. By the time
780 	 * thread_exit() is called (below), the association with the relevant
781 	 * process has been lost.
782 	 *
783 	 * We also free the context here.
784 	 */
785 	if (p->p_pctx) {
786 		kpreempt_disable();
787 		exitpctx(p);
788 		kpreempt_enable();
789 
790 		freepctx(p, 0);
791 	}
792 
793 	/*
794 	 * curthread's proc pointer is changed to point to the 'sched'
795 	 * process for the corresponding zone, except in the case when
796 	 * the exiting process is in fact a zsched instance, in which
797 	 * case the proc pointer is set to p0.  We do so, so that the
798 	 * process still points at the right zone when we call the VN_RELE()
799 	 * below.
800 	 *
801 	 * This is because curthread's original proc pointer can be freed as
802 	 * soon as the child sends a SIGCLD to its parent.  We use zsched so
803 	 * that for user processes, even in the final moments of death, the
804 	 * process is still associated with its zone.
805 	 */
806 	if (p != t->t_procp->p_zone->zone_zsched)
807 		t->t_procp = t->t_procp->p_zone->zone_zsched;
808 	else
809 		t->t_procp = &p0;
810 
811 	mutex_exit(&p->p_lock);
812 	if (!evaporate) {
813 		p->p_pidflag &= ~CLDPEND;
814 		sigcld(p, sqp);
815 	} else {
816 		/*
817 		 * Do what sigcld() would do if the disposition
818 		 * of the SIGCHLD signal were set to be ignored.
819 		 */
820 		cv_broadcast(&p->p_srwchan_cv);
821 		freeproc(p);
822 	}
823 	mutex_exit(&pidlock);
824 
825 	/*
826 	 * We don't release u_cdir and u_rdir until SZOMB is set.
827 	 * This protects us against dofusers().
828 	 */
829 	VN_RELE(cdir);
830 	if (rdir)
831 		VN_RELE(rdir);
832 	if (cwd)
833 		refstr_rele(cwd);
834 
835 	/*
836 	 * task_rele() may ultimately cause the zone to go away (or
837 	 * may cause the last user process in a zone to go away, which
838 	 * signals zsched to go away).  So prior to this call, we must
839 	 * no longer point at zsched.
840 	 */
841 	t->t_procp = &p0;
842 	task_rele(tk);
843 
844 	kmem_free(lwpdir, lwpdir_sz * sizeof (lwpdir_t));
845 	kmem_free(tidhash, tidhash_sz * sizeof (lwpdir_t *));
846 
847 	lwp_pcb_exit();
848 
849 	thread_exit();
850 	/* NOTREACHED */
851 }
852 
853 /*
854  * Format siginfo structure for wait system calls.
855  */
856 void
857 winfo(proc_t *pp, k_siginfo_t *ip, int waitflag)
858 {
859 	ASSERT(MUTEX_HELD(&pidlock));
860 
861 	bzero(ip, sizeof (k_siginfo_t));
862 	ip->si_signo = SIGCLD;
863 	ip->si_code = pp->p_wcode;
864 	ip->si_pid = pp->p_pid;
865 	ip->si_ctid = PRCTID(pp);
866 	ip->si_zoneid = pp->p_zone->zone_id;
867 	ip->si_status = pp->p_wdata;
868 	ip->si_stime = pp->p_stime;
869 	ip->si_utime = pp->p_utime;
870 
871 	if (waitflag) {
872 		pp->p_wcode = 0;
873 		pp->p_wdata = 0;
874 		pp->p_pidflag &= ~CLDPEND;
875 	}
876 }
877 
878 /*
879  * Wait system call.
880  * Search for a terminated (zombie) child,
881  * finally lay it to rest, and collect its status.
882  * Look also for stopped children,
883  * and pass back status from them.
884  */
885 int
886 waitid(idtype_t idtype, id_t id, k_siginfo_t *ip, int options)
887 {
888 	int found;
889 	proc_t *cp, *pp;
890 	int proc_gone;
891 	int waitflag = !(options & WNOWAIT);
892 
893 	/*
894 	 * Obsolete flag, defined here only for binary compatibility
895 	 * with old statically linked executables.  Delete this when
896 	 * we no longer care about these old and broken applications.
897 	 */
898 #define	_WNOCHLD	0400
899 	options &= ~_WNOCHLD;
900 
901 	if (options == 0 || (options & ~WOPTMASK))
902 		return (EINVAL);
903 
904 	switch (idtype) {
905 	case P_PID:
906 	case P_PGID:
907 		if (id < 0 || id >= maxpid)
908 			return (EINVAL);
909 		/* FALLTHROUGH */
910 	case P_ALL:
911 		break;
912 	default:
913 		return (EINVAL);
914 	}
915 
916 	pp = ttoproc(curthread);
917 
918 	/*
919 	 * lock parent mutex so that sibling chain can be searched.
920 	 */
921 	mutex_enter(&pidlock);
922 
923 	/*
924 	 * if we are only looking for exited processes and child_ns list
925 	 * is empty no reason to look at all children.
926 	 */
927 	if (idtype == P_ALL &&
928 	    (options & ~WNOWAIT) == (WNOHANG | WEXITED) &&
929 	    pp->p_child_ns == NULL) {
930 		if (pp->p_child) {
931 			mutex_exit(&pidlock);
932 			bzero(ip, sizeof (k_siginfo_t));
933 			return (0);
934 		}
935 		mutex_exit(&pidlock);
936 		return (ECHILD);
937 	}
938 
939 	while (pp->p_child != NULL) {
940 
941 		proc_gone = 0;
942 
943 		for (cp = pp->p_child_ns; cp != NULL; cp = cp->p_sibling_ns) {
944 			if (idtype != P_PID && (cp->p_pidflag & CLDWAITPID))
945 				continue;
946 			if (idtype == P_PID && id != cp->p_pid)
947 				continue;
948 			if (idtype == P_PGID && id != cp->p_pgrp)
949 				continue;
950 
951 			switch (cp->p_wcode) {
952 
953 			case CLD_TRAPPED:
954 			case CLD_STOPPED:
955 			case CLD_CONTINUED:
956 				cmn_err(CE_PANIC,
957 				    "waitid: wrong state %d on the p_newstate"
958 				    " list", cp->p_wcode);
959 				break;
960 
961 			case CLD_EXITED:
962 			case CLD_DUMPED:
963 			case CLD_KILLED:
964 				if (!(options & WEXITED)) {
965 					/*
966 					 * Count how many are already gone
967 					 * for good.
968 					 */
969 					proc_gone++;
970 					break;
971 				}
972 				if (!waitflag) {
973 					winfo(cp, ip, 0);
974 				} else {
975 					winfo(cp, ip, 1);
976 					freeproc(cp);
977 				}
978 				mutex_exit(&pidlock);
979 				if (waitflag) {		/* accept SIGCLD */
980 					sigcld_delete(ip);
981 					sigcld_repost();
982 				}
983 				return (0);
984 			}
985 
986 			if (idtype == P_PID)
987 				break;
988 		}
989 
990 		/*
991 		 * Wow! None of the threads on the p_sibling_ns list were
992 		 * interesting threads. Check all the kids!
993 		 */
994 		found = 0;
995 		for (cp = pp->p_child; cp != NULL; cp = cp->p_sibling) {
996 			if (idtype == P_PID && id != cp->p_pid)
997 				continue;
998 			if (idtype == P_PGID && id != cp->p_pgrp)
999 				continue;
1000 
1001 			switch (cp->p_wcode) {
1002 			case CLD_TRAPPED:
1003 				if (!(options & WTRAPPED))
1004 					break;
1005 				winfo(cp, ip, waitflag);
1006 				mutex_exit(&pidlock);
1007 				if (waitflag) {		/* accept SIGCLD */
1008 					sigcld_delete(ip);
1009 					sigcld_repost();
1010 				}
1011 				return (0);
1012 
1013 			case CLD_STOPPED:
1014 				if (!(options & WSTOPPED))
1015 					break;
1016 				/* Is it still stopped? */
1017 				mutex_enter(&cp->p_lock);
1018 				if (!jobstopped(cp)) {
1019 					mutex_exit(&cp->p_lock);
1020 					break;
1021 				}
1022 				mutex_exit(&cp->p_lock);
1023 				winfo(cp, ip, waitflag);
1024 				mutex_exit(&pidlock);
1025 				if (waitflag) {		/* accept SIGCLD */
1026 					sigcld_delete(ip);
1027 					sigcld_repost();
1028 				}
1029 				return (0);
1030 
1031 			case CLD_CONTINUED:
1032 				if (!(options & WCONTINUED))
1033 					break;
1034 				winfo(cp, ip, waitflag);
1035 				mutex_exit(&pidlock);
1036 				if (waitflag) {		/* accept SIGCLD */
1037 					sigcld_delete(ip);
1038 					sigcld_repost();
1039 				}
1040 				return (0);
1041 
1042 			case CLD_EXITED:
1043 			case CLD_DUMPED:
1044 			case CLD_KILLED:
1045 				if (idtype != P_PID &&
1046 				    (cp->p_pidflag & CLDWAITPID))
1047 					continue;
1048 				/*
1049 				 * Don't complain if a process was found in
1050 				 * the first loop but we broke out of the loop
1051 				 * because of the arguments passed to us.
1052 				 */
1053 				if (proc_gone == 0) {
1054 					cmn_err(CE_PANIC,
1055 					    "waitid: wrong state on the"
1056 					    " p_child list");
1057 				} else {
1058 					break;
1059 				}
1060 			}
1061 
1062 			found++;
1063 
1064 			if (idtype == P_PID)
1065 				break;
1066 		}
1067 
1068 		/*
1069 		 * If we found no interesting processes at all,
1070 		 * break out and return ECHILD.
1071 		 */
1072 		if (found + proc_gone == 0)
1073 			break;
1074 
1075 		if (options & WNOHANG) {
1076 			mutex_exit(&pidlock);
1077 			bzero(ip, sizeof (k_siginfo_t));
1078 			/*
1079 			 * We should set ip->si_signo = SIGCLD,
1080 			 * but there is an SVVS test that expects
1081 			 * ip->si_signo to be zero in this case.
1082 			 */
1083 			return (0);
1084 		}
1085 
1086 		/*
1087 		 * If we found no processes of interest that could
1088 		 * change state while we wait, we don't wait at all.
1089 		 * Get out with ECHILD according to SVID.
1090 		 */
1091 		if (found == proc_gone)
1092 			break;
1093 
1094 		if (!cv_wait_sig_swap(&pp->p_cv, &pidlock)) {
1095 			mutex_exit(&pidlock);
1096 			return (EINTR);
1097 		}
1098 	}
1099 	mutex_exit(&pidlock);
1100 	return (ECHILD);
1101 }
1102 
1103 /*
1104  * The wait() system call trap is no longer invoked by libc.
1105  * It is retained only for the benefit of statically linked applications.
1106  * Delete this when we no longer care about these old and broken applications.
1107  */
1108 int64_t
1109 wait(void)
1110 {
1111 	int error;
1112 	k_siginfo_t info;
1113 	rval_t	r;
1114 
1115 	if (error =  waitid(P_ALL, (id_t)0, &info, WEXITED|WTRAPPED))
1116 		return (set_errno(error));
1117 	r.r_val1 = info.si_pid;
1118 	r.r_val2 = wstat(info.si_code, info.si_status);
1119 	return (r.r_vals);
1120 }
1121 
1122 int
1123 waitsys(idtype_t idtype, id_t id, siginfo_t *infop, int options)
1124 {
1125 	int error;
1126 	k_siginfo_t info;
1127 
1128 	if (error = waitid(idtype, id, &info, options))
1129 		return (set_errno(error));
1130 	if (copyout(&info, infop, sizeof (k_siginfo_t)))
1131 		return (set_errno(EFAULT));
1132 	return (0);
1133 }
1134 
1135 #ifdef _SYSCALL32_IMPL
1136 
1137 int
1138 waitsys32(idtype_t idtype, id_t id, siginfo_t *infop, int options)
1139 {
1140 	int error;
1141 	k_siginfo_t info;
1142 	siginfo32_t info32;
1143 
1144 	if (error = waitid(idtype, id, &info, options))
1145 		return (set_errno(error));
1146 	siginfo_kto32(&info, &info32);
1147 	if (copyout(&info32, infop, sizeof (info32)))
1148 		return (set_errno(EFAULT));
1149 	return (0);
1150 }
1151 
1152 #endif	/* _SYSCALL32_IMPL */
1153 
1154 void
1155 proc_detach(proc_t *p)
1156 {
1157 	proc_t *q;
1158 
1159 	ASSERT(MUTEX_HELD(&pidlock));
1160 
1161 	q = p->p_parent;
1162 	ASSERT(q != NULL);
1163 
1164 	/*
1165 	 * Take it off the newstate list of its parent
1166 	 */
1167 	delete_ns(q, p);
1168 
1169 	if (q->p_child == p) {
1170 		q->p_child = p->p_sibling;
1171 		/*
1172 		 * If the parent has no children, it better not
1173 		 * have any with new states either!
1174 		 */
1175 		ASSERT(q->p_child ? 1 : q->p_child_ns == NULL);
1176 	}
1177 
1178 	if (p->p_sibling) {
1179 		p->p_sibling->p_psibling = p->p_psibling;
1180 	}
1181 
1182 	if (p->p_psibling) {
1183 		p->p_psibling->p_sibling = p->p_sibling;
1184 	}
1185 }
1186 
1187 /*
1188  * Remove zombie children from the process table.
1189  */
1190 void
1191 freeproc(proc_t *p)
1192 {
1193 	proc_t *q;
1194 
1195 	ASSERT(p->p_stat == SZOMB);
1196 	ASSERT(p->p_tlist == NULL);
1197 	ASSERT(MUTEX_HELD(&pidlock));
1198 
1199 	sigdelq(p, NULL, 0);
1200 	if (p->p_killsqp) {
1201 		siginfofree(p->p_killsqp);
1202 		p->p_killsqp = NULL;
1203 	}
1204 
1205 	prfree(p);	/* inform /proc */
1206 
1207 	/*
1208 	 * Don't free the init processes.
1209 	 * Other dying processes will access it.
1210 	 */
1211 	if (p == proc_init)
1212 		return;
1213 
1214 
1215 	/*
1216 	 * We wait until now to free the cred structure because a
1217 	 * zombie process's credentials may be examined by /proc.
1218 	 * No cred locking needed because there are no threads at this point.
1219 	 */
1220 	upcount_dec(crgetruid(p->p_cred), crgetzoneid(p->p_cred));
1221 	crfree(p->p_cred);
1222 	if (p->p_corefile != NULL) {
1223 		corectl_path_rele(p->p_corefile);
1224 		p->p_corefile = NULL;
1225 	}
1226 	if (p->p_content != NULL) {
1227 		corectl_content_rele(p->p_content);
1228 		p->p_content = NULL;
1229 	}
1230 
1231 	if (p->p_nextofkin && !((p->p_nextofkin->p_flag & SNOWAIT) ||
1232 	    (PTOU(p->p_nextofkin)->u_signal[SIGCLD - 1] == SIG_IGN))) {
1233 		/*
1234 		 * This should still do the right thing since p_utime/stime
1235 		 * get set to the correct value on process exit, so it
1236 		 * should get properly updated
1237 		 */
1238 		p->p_nextofkin->p_cutime += p->p_utime;
1239 		p->p_nextofkin->p_cstime += p->p_stime;
1240 
1241 		p->p_nextofkin->p_cacct[LMS_USER] += p->p_acct[LMS_USER];
1242 		p->p_nextofkin->p_cacct[LMS_SYSTEM] += p->p_acct[LMS_SYSTEM];
1243 		p->p_nextofkin->p_cacct[LMS_TRAP] += p->p_acct[LMS_TRAP];
1244 		p->p_nextofkin->p_cacct[LMS_TFAULT] += p->p_acct[LMS_TFAULT];
1245 		p->p_nextofkin->p_cacct[LMS_DFAULT] += p->p_acct[LMS_DFAULT];
1246 		p->p_nextofkin->p_cacct[LMS_KFAULT] += p->p_acct[LMS_KFAULT];
1247 		p->p_nextofkin->p_cacct[LMS_USER_LOCK]
1248 		    += p->p_acct[LMS_USER_LOCK];
1249 		p->p_nextofkin->p_cacct[LMS_SLEEP] += p->p_acct[LMS_SLEEP];
1250 		p->p_nextofkin->p_cacct[LMS_WAIT_CPU]
1251 		    += p->p_acct[LMS_WAIT_CPU];
1252 		p->p_nextofkin->p_cacct[LMS_STOPPED] += p->p_acct[LMS_STOPPED];
1253 
1254 		p->p_nextofkin->p_cru.minflt	+= p->p_ru.minflt;
1255 		p->p_nextofkin->p_cru.majflt	+= p->p_ru.majflt;
1256 		p->p_nextofkin->p_cru.nswap	+= p->p_ru.nswap;
1257 		p->p_nextofkin->p_cru.inblock	+= p->p_ru.inblock;
1258 		p->p_nextofkin->p_cru.oublock	+= p->p_ru.oublock;
1259 		p->p_nextofkin->p_cru.msgsnd	+= p->p_ru.msgsnd;
1260 		p->p_nextofkin->p_cru.msgrcv	+= p->p_ru.msgrcv;
1261 		p->p_nextofkin->p_cru.nsignals	+= p->p_ru.nsignals;
1262 		p->p_nextofkin->p_cru.nvcsw	+= p->p_ru.nvcsw;
1263 		p->p_nextofkin->p_cru.nivcsw	+= p->p_ru.nivcsw;
1264 		p->p_nextofkin->p_cru.sysc	+= p->p_ru.sysc;
1265 		p->p_nextofkin->p_cru.ioch	+= p->p_ru.ioch;
1266 
1267 	}
1268 
1269 	q = p->p_nextofkin;
1270 	if (q && q->p_orphan == p)
1271 		q->p_orphan = p->p_nextorph;
1272 	else if (q) {
1273 		for (q = q->p_orphan; q; q = q->p_nextorph)
1274 			if (q->p_nextorph == p)
1275 				break;
1276 		ASSERT(q && q->p_nextorph == p);
1277 		q->p_nextorph = p->p_nextorph;
1278 	}
1279 
1280 	proc_detach(p);
1281 	pid_exit(p);	/* frees pid and proc structure */
1282 }
1283 
1284 /*
1285  * Delete process "child" from the newstate list of process "parent"
1286  */
1287 void
1288 delete_ns(proc_t *parent, proc_t *child)
1289 {
1290 	proc_t **ns;
1291 
1292 	ASSERT(MUTEX_HELD(&pidlock));
1293 	ASSERT(child->p_parent == parent);
1294 	for (ns = &parent->p_child_ns; *ns != NULL; ns = &(*ns)->p_sibling_ns) {
1295 		if (*ns == child) {
1296 
1297 			ASSERT((*ns)->p_parent == parent);
1298 
1299 			*ns = child->p_sibling_ns;
1300 			child->p_sibling_ns = NULL;
1301 			return;
1302 		}
1303 	}
1304 }
1305 
1306 /*
1307  * Add process "child" to the new state list of process "parent"
1308  */
1309 void
1310 add_ns(proc_t *parent, proc_t *child)
1311 {
1312 	ASSERT(child->p_sibling_ns == NULL);
1313 	child->p_sibling_ns = parent->p_child_ns;
1314 	parent->p_child_ns = child;
1315 }
1316