xref: /illumos-gate/usr/src/uts/common/os/exit.c (revision c0dd49bdd68c0d758a67d56f07826f3b45cfc664)
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 2010 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 #include <sys/types.h>
30 #include <sys/param.h>
31 #include <sys/sysmacros.h>
32 #include <sys/systm.h>
33 #include <sys/cred.h>
34 #include <sys/user.h>
35 #include <sys/errno.h>
36 #include <sys/proc.h>
37 #include <sys/ucontext.h>
38 #include <sys/procfs.h>
39 #include <sys/vnode.h>
40 #include <sys/acct.h>
41 #include <sys/var.h>
42 #include <sys/cmn_err.h>
43 #include <sys/debug.h>
44 #include <sys/wait.h>
45 #include <sys/siginfo.h>
46 #include <sys/procset.h>
47 #include <sys/class.h>
48 #include <sys/file.h>
49 #include <sys/session.h>
50 #include <sys/kmem.h>
51 #include <sys/vtrace.h>
52 #include <sys/prsystm.h>
53 #include <sys/ipc.h>
54 #include <sys/sem_impl.h>
55 #include <c2/audit.h>
56 #include <sys/aio_impl.h>
57 #include <vm/as.h>
58 #include <sys/poll.h>
59 #include <sys/door.h>
60 #include <sys/lwpchan_impl.h>
61 #include <sys/utrap.h>
62 #include <sys/task.h>
63 #include <sys/exacct.h>
64 #include <sys/cyclic.h>
65 #include <sys/schedctl.h>
66 #include <sys/rctl.h>
67 #include <sys/contract_impl.h>
68 #include <sys/contract/process_impl.h>
69 #include <sys/list.h>
70 #include <sys/dtrace.h>
71 #include <sys/pool.h>
72 #include <sys/sdt.h>
73 #include <sys/corectl.h>
74 #include <sys/brand.h>
75 #include <sys/libc_kernel.h>
76 
77 /*
78  * convert code/data pair into old style wait status
79  */
80 int
81 wstat(int code, int data)
82 {
83 	int stat = (data & 0377);
84 
85 	switch (code) {
86 	case CLD_EXITED:
87 		stat <<= 8;
88 		break;
89 	case CLD_DUMPED:
90 		stat |= WCOREFLG;
91 		break;
92 	case CLD_KILLED:
93 		break;
94 	case CLD_TRAPPED:
95 	case CLD_STOPPED:
96 		stat <<= 8;
97 		stat |= WSTOPFLG;
98 		break;
99 	case CLD_CONTINUED:
100 		stat = WCONTFLG;
101 		break;
102 	default:
103 		cmn_err(CE_PANIC, "wstat: bad code");
104 		/* NOTREACHED */
105 	}
106 	return (stat);
107 }
108 
109 static char *
110 exit_reason(char *buf, size_t bufsz, int what, int why)
111 {
112 	switch (why) {
113 	case CLD_EXITED:
114 		(void) snprintf(buf, bufsz, "exited with status %d", what);
115 		break;
116 	case CLD_KILLED:
117 		(void) snprintf(buf, bufsz, "exited on fatal signal %d", what);
118 		break;
119 	case CLD_DUMPED:
120 		(void) snprintf(buf, bufsz, "core dumped on signal %d", what);
121 		break;
122 	default:
123 		(void) snprintf(buf, bufsz, "encountered unknown error "
124 		    "(%d, %d)", why, what);
125 		break;
126 	}
127 
128 	return (buf);
129 }
130 
131 /*
132  * exit system call: pass back caller's arg.
133  */
134 void
135 rexit(int rval)
136 {
137 	exit(CLD_EXITED, rval);
138 }
139 
140 /*
141  * Called by proc_exit() when a zone's init exits, presumably because
142  * it failed.  As long as the given zone is still in the "running"
143  * state, we will re-exec() init, but first we need to reset things
144  * which are usually inherited across exec() but will break init's
145  * assumption that it is being exec()'d from a virgin process.  Most
146  * importantly this includes closing all file descriptors (exec only
147  * closes those marked close-on-exec) and resetting signals (exec only
148  * resets handled signals, and we need to clear any signals which
149  * killed init).  Anything else that exec(2) says would be inherited,
150  * but would affect the execution of init, needs to be reset.
151  */
152 static int
153 restart_init(int what, int why)
154 {
155 	kthread_t *t = curthread;
156 	klwp_t *lwp = ttolwp(t);
157 	proc_t *p = ttoproc(t);
158 	user_t *up = PTOU(p);
159 
160 	vnode_t *oldcd, *oldrd;
161 	int i, err;
162 	char reason_buf[64];
163 
164 	/*
165 	 * Let zone admin (and global zone admin if this is for a non-global
166 	 * zone) know that init has failed and will be restarted.
167 	 */
168 	zcmn_err(p->p_zone->zone_id, CE_WARN,
169 	    "init(1M) %s: restarting automatically",
170 	    exit_reason(reason_buf, sizeof (reason_buf), what, why));
171 
172 	if (!INGLOBALZONE(p)) {
173 		cmn_err(CE_WARN, "init(1M) for zone %s (pid %d) %s: "
174 		    "restarting automatically",
175 		    p->p_zone->zone_name, p->p_pid, reason_buf);
176 	}
177 
178 	/*
179 	 * Remove any fpollinfo_t's for this (last) thread from our file
180 	 * descriptors so closeall() can ASSERT() that they're all gone.
181 	 * Then close all open file descriptors in the process.
182 	 */
183 	pollcleanup();
184 	closeall(P_FINFO(p));
185 
186 	/*
187 	 * Grab p_lock and begin clearing miscellaneous global process
188 	 * state that needs to be reset before we exec the new init(1M).
189 	 */
190 
191 	mutex_enter(&p->p_lock);
192 	prbarrier(p);
193 
194 	p->p_flag &= ~(SKILLED | SEXTKILLED | SEXITING | SDOCORE);
195 	up->u_cmask = CMASK;
196 
197 	sigemptyset(&t->t_hold);
198 	sigemptyset(&t->t_sig);
199 	sigemptyset(&t->t_extsig);
200 
201 	sigemptyset(&p->p_sig);
202 	sigemptyset(&p->p_extsig);
203 
204 	sigdelq(p, t, 0);
205 	sigdelq(p, NULL, 0);
206 
207 	if (p->p_killsqp) {
208 		siginfofree(p->p_killsqp);
209 		p->p_killsqp = NULL;
210 	}
211 
212 	/*
213 	 * Reset any signals that are ignored back to the default disposition.
214 	 * Other u_signal members will be cleared when exec calls sigdefault().
215 	 */
216 	for (i = 1; i < NSIG; i++) {
217 		if (up->u_signal[i - 1] == SIG_IGN) {
218 			up->u_signal[i - 1] = SIG_DFL;
219 			sigemptyset(&up->u_sigmask[i - 1]);
220 		}
221 	}
222 
223 	/*
224 	 * Clear the current signal, any signal info associated with it, and
225 	 * any signal information from contracts and/or contract templates.
226 	 */
227 	lwp->lwp_cursig = 0;
228 	lwp->lwp_extsig = 0;
229 	if (lwp->lwp_curinfo != NULL) {
230 		siginfofree(lwp->lwp_curinfo);
231 		lwp->lwp_curinfo = NULL;
232 	}
233 	lwp_ctmpl_clear(lwp);
234 
235 	/*
236 	 * Reset both the process root directory and the current working
237 	 * directory to the root of the zone just as we do during boot.
238 	 */
239 	VN_HOLD(p->p_zone->zone_rootvp);
240 	oldrd = up->u_rdir;
241 	up->u_rdir = p->p_zone->zone_rootvp;
242 
243 	VN_HOLD(p->p_zone->zone_rootvp);
244 	oldcd = up->u_cdir;
245 	up->u_cdir = p->p_zone->zone_rootvp;
246 
247 	if (up->u_cwd != NULL) {
248 		refstr_rele(up->u_cwd);
249 		up->u_cwd = NULL;
250 	}
251 
252 	mutex_exit(&p->p_lock);
253 
254 	if (oldrd != NULL)
255 		VN_RELE(oldrd);
256 	if (oldcd != NULL)
257 		VN_RELE(oldcd);
258 
259 	/* Free the controlling tty.  (freectty() always assumes curproc.) */
260 	ASSERT(p == curproc);
261 	(void) freectty(B_TRUE);
262 
263 	/*
264 	 * Now exec() the new init(1M) on top of the current process.  If we
265 	 * succeed, the caller will treat this like a successful system call.
266 	 * If we fail, we issue messages and the caller will proceed with exit.
267 	 */
268 	err = exec_init(p->p_zone->zone_initname, NULL);
269 
270 	if (err == 0)
271 		return (0);
272 
273 	zcmn_err(p->p_zone->zone_id, CE_WARN,
274 	    "failed to restart init(1M) (err=%d): system reboot required", err);
275 
276 	if (!INGLOBALZONE(p)) {
277 		cmn_err(CE_WARN, "failed to restart init(1M) for zone %s "
278 		    "(pid %d, err=%d): zoneadm(1M) boot required",
279 		    p->p_zone->zone_name, p->p_pid, err);
280 	}
281 
282 	return (-1);
283 }
284 
285 /*
286  * Release resources.
287  * Enter zombie state.
288  * Wake up parent and init processes,
289  * and dispose of children.
290  */
291 void
292 exit(int why, int what)
293 {
294 	/*
295 	 * If proc_exit() fails, then some other lwp in the process
296 	 * got there first.  We just have to call lwp_exit() to allow
297 	 * the other lwp to finish exiting the process.  Otherwise we're
298 	 * restarting init, and should return.
299 	 */
300 	if (proc_exit(why, what) != 0) {
301 		mutex_enter(&curproc->p_lock);
302 		ASSERT(curproc->p_flag & SEXITLWPS);
303 		lwp_exit();
304 		/* NOTREACHED */
305 	}
306 }
307 
308 /*
309  * Set the SEXITING flag on the process, after making sure /proc does
310  * not have it locked.  This is done in more places than proc_exit(),
311  * so it is a separate function.
312  */
313 void
314 proc_is_exiting(proc_t *p)
315 {
316 	mutex_enter(&p->p_lock);
317 	prbarrier(p);
318 	p->p_flag |= SEXITING;
319 	mutex_exit(&p->p_lock);
320 }
321 
322 /*
323  * Return value:
324  *   1 - exitlwps() failed, call (or continue) lwp_exit()
325  *   0 - restarting init.  Return through system call path
326  */
327 int
328 proc_exit(int why, int what)
329 {
330 	kthread_t *t = curthread;
331 	klwp_t *lwp = ttolwp(t);
332 	proc_t *p = ttoproc(t);
333 	zone_t *z = p->p_zone;
334 	timeout_id_t tmp_id;
335 	int rv;
336 	proc_t *q;
337 	task_t *tk;
338 	vnode_t *exec_vp, *execdir_vp, *cdir, *rdir;
339 	sigqueue_t *sqp;
340 	lwpdir_t *lwpdir;
341 	uint_t lwpdir_sz;
342 	tidhash_t *tidhash;
343 	uint_t tidhash_sz;
344 	ret_tidhash_t *ret_tidhash;
345 	refstr_t *cwd;
346 	hrtime_t hrutime, hrstime;
347 	int evaporate;
348 
349 	/*
350 	 * Stop and discard the process's lwps except for the current one,
351 	 * unless some other lwp beat us to it.  If exitlwps() fails then
352 	 * return and the calling lwp will call (or continue in) lwp_exit().
353 	 */
354 	proc_is_exiting(p);
355 	if (exitlwps(0) != 0)
356 		return (1);
357 
358 	mutex_enter(&p->p_lock);
359 	if (p->p_ttime > 0) {
360 		/*
361 		 * Account any remaining ticks charged to this process
362 		 * on its way out.
363 		 */
364 		(void) task_cpu_time_incr(p->p_task, p->p_ttime);
365 		p->p_ttime = 0;
366 	}
367 	mutex_exit(&p->p_lock);
368 
369 	DTRACE_PROC(lwp__exit);
370 	DTRACE_PROC1(exit, int, why);
371 
372 	/*
373 	 * Will perform any brand specific proc exit processing, since this
374 	 * is always the last lwp, will also perform lwp_exit and free brand
375 	 * data
376 	 */
377 	if (PROC_IS_BRANDED(p)) {
378 		lwp_detach_brand_hdlrs(lwp);
379 		brand_clearbrand(p);
380 	}
381 
382 	/*
383 	 * Don't let init exit unless zone_start_init() failed its exec, or
384 	 * we are shutting down the zone or the machine.
385 	 *
386 	 * Since we are single threaded, we don't need to lock the
387 	 * following accesses to zone_proc_initpid.
388 	 */
389 	if (p->p_pid == z->zone_proc_initpid) {
390 		if (z->zone_boot_err == 0 &&
391 		    zone_status_get(z) < ZONE_IS_SHUTTING_DOWN &&
392 		    zone_status_get(global_zone) < ZONE_IS_SHUTTING_DOWN &&
393 		    z->zone_restart_init == B_TRUE &&
394 		    restart_init(what, why) == 0)
395 			return (0);
396 		/*
397 		 * Since we didn't or couldn't restart init, we clear
398 		 * the zone's init state and proceed with exit
399 		 * processing.
400 		 */
401 		z->zone_proc_initpid = -1;
402 	}
403 
404 	lwp_pcb_exit();
405 
406 	/*
407 	 * Allocate a sigqueue now, before we grab locks.
408 	 * It will be given to sigcld(), below.
409 	 * Special case:  If we will be making the process disappear
410 	 * without a trace because it is either:
411 	 *	* an exiting SSYS process, or
412 	 *	* a posix_spawn() vfork child who requests it,
413 	 * we don't bother to allocate a useless sigqueue.
414 	 */
415 	evaporate = (p->p_flag & SSYS) || ((p->p_flag & SVFORK) &&
416 	    why == CLD_EXITED && what == _EVAPORATE);
417 	if (!evaporate)
418 		sqp = kmem_zalloc(sizeof (sigqueue_t), KM_SLEEP);
419 
420 	/*
421 	 * revoke any doors created by the process.
422 	 */
423 	if (p->p_door_list)
424 		door_exit();
425 
426 	/*
427 	 * Release schedctl data structures.
428 	 */
429 	if (p->p_pagep)
430 		schedctl_proc_cleanup();
431 
432 	/*
433 	 * make sure all pending kaio has completed.
434 	 */
435 	if (p->p_aio)
436 		aio_cleanup_exit();
437 
438 	/*
439 	 * discard the lwpchan cache.
440 	 */
441 	if (p->p_lcp != NULL)
442 		lwpchan_destroy_cache(0);
443 
444 	/*
445 	 * Clean up any DTrace helper actions or probes for the process.
446 	 */
447 	if (p->p_dtrace_helpers != NULL) {
448 		ASSERT(dtrace_helpers_cleanup != NULL);
449 		(*dtrace_helpers_cleanup)();
450 	}
451 
452 	/* untimeout the realtime timers */
453 	if (p->p_itimer != NULL)
454 		timer_exit();
455 
456 	if ((tmp_id = p->p_alarmid) != 0) {
457 		p->p_alarmid = 0;
458 		(void) untimeout(tmp_id);
459 	}
460 
461 	/*
462 	 * Remove any fpollinfo_t's for this (last) thread from our file
463 	 * descriptors so closeall() can ASSERT() that they're all gone.
464 	 */
465 	pollcleanup();
466 
467 	if (p->p_rprof_cyclic != CYCLIC_NONE) {
468 		mutex_enter(&cpu_lock);
469 		cyclic_remove(p->p_rprof_cyclic);
470 		mutex_exit(&cpu_lock);
471 	}
472 
473 	mutex_enter(&p->p_lock);
474 
475 	/*
476 	 * Clean up any DTrace probes associated with this process.
477 	 */
478 	if (p->p_dtrace_probes) {
479 		ASSERT(dtrace_fasttrap_exit_ptr != NULL);
480 		dtrace_fasttrap_exit_ptr(p);
481 	}
482 
483 	while ((tmp_id = p->p_itimerid) != 0) {
484 		p->p_itimerid = 0;
485 		mutex_exit(&p->p_lock);
486 		(void) untimeout(tmp_id);
487 		mutex_enter(&p->p_lock);
488 	}
489 
490 	lwp_cleanup();
491 
492 	/*
493 	 * We are about to exit; prevent our resource associations from
494 	 * being changed.
495 	 */
496 	pool_barrier_enter();
497 
498 	/*
499 	 * Block the process against /proc now that we have really
500 	 * acquired p->p_lock (to manipulate p_tlist at least).
501 	 */
502 	prbarrier(p);
503 
504 #ifdef	SUN_SRC_COMPAT
505 	if (code == CLD_KILLED)
506 		u.u_acflag |= AXSIG;
507 #endif
508 	sigfillset(&p->p_ignore);
509 	sigemptyset(&p->p_siginfo);
510 	sigemptyset(&p->p_sig);
511 	sigemptyset(&p->p_extsig);
512 	sigemptyset(&t->t_sig);
513 	sigemptyset(&t->t_extsig);
514 	sigemptyset(&p->p_sigmask);
515 	sigdelq(p, t, 0);
516 	lwp->lwp_cursig = 0;
517 	lwp->lwp_extsig = 0;
518 	p->p_flag &= ~(SKILLED | SEXTKILLED);
519 	if (lwp->lwp_curinfo) {
520 		siginfofree(lwp->lwp_curinfo);
521 		lwp->lwp_curinfo = NULL;
522 	}
523 
524 	t->t_proc_flag |= TP_LWPEXIT;
525 	ASSERT(p->p_lwpcnt == 1 && p->p_zombcnt == 0);
526 	prlwpexit(t);		/* notify /proc */
527 	lwp_hash_out(p, t->t_tid);
528 	prexit(p);
529 
530 	p->p_lwpcnt = 0;
531 	p->p_tlist = NULL;
532 	sigqfree(p);
533 	term_mstate(t);
534 	p->p_mterm = gethrtime();
535 
536 	exec_vp = p->p_exec;
537 	execdir_vp = p->p_execdir;
538 	p->p_exec = NULLVP;
539 	p->p_execdir = NULLVP;
540 	mutex_exit(&p->p_lock);
541 
542 	pr_free_watched_pages(p);
543 
544 	closeall(P_FINFO(p));
545 
546 	/* Free the controlling tty.  (freectty() always assumes curproc.) */
547 	ASSERT(p == curproc);
548 	(void) freectty(B_TRUE);
549 
550 #if defined(__sparc)
551 	if (p->p_utraps != NULL)
552 		utrap_free(p);
553 #endif
554 	if (p->p_semacct)			/* IPC semaphore exit */
555 		semexit(p);
556 	rv = wstat(why, what);
557 
558 	acct(rv & 0xff);
559 	exacct_commit_proc(p, rv);
560 
561 	/*
562 	 * Release any resources associated with C2 auditing
563 	 */
564 	if (AU_AUDITING()) {
565 		/*
566 		 * audit exit system call
567 		 */
568 		audit_exit(why, what);
569 	}
570 
571 	/*
572 	 * Free address space.
573 	 */
574 	relvm();
575 
576 	if (exec_vp) {
577 		/*
578 		 * Close this executable which has been opened when the process
579 		 * was created by getproc().
580 		 */
581 		(void) VOP_CLOSE(exec_vp, FREAD, 1, (offset_t)0, CRED(), NULL);
582 		VN_RELE(exec_vp);
583 	}
584 	if (execdir_vp)
585 		VN_RELE(execdir_vp);
586 
587 	/*
588 	 * Release held contracts.
589 	 */
590 	contract_exit(p);
591 
592 	/*
593 	 * Depart our encapsulating process contract.
594 	 */
595 	if ((p->p_flag & SSYS) == 0) {
596 		ASSERT(p->p_ct_process);
597 		contract_process_exit(p->p_ct_process, p, rv);
598 	}
599 
600 	/*
601 	 * Remove pool association, and block if requested by pool_do_bind.
602 	 */
603 	mutex_enter(&p->p_lock);
604 	ASSERT(p->p_pool->pool_ref > 0);
605 	atomic_add_32(&p->p_pool->pool_ref, -1);
606 	p->p_pool = pool_default;
607 	/*
608 	 * Now that our address space has been freed and all other threads
609 	 * in this process have exited, set the PEXITED pool flag.  This
610 	 * tells the pools subsystems to ignore this process if it was
611 	 * requested to rebind this process to a new pool.
612 	 */
613 	p->p_poolflag |= PEXITED;
614 	pool_barrier_exit();
615 	mutex_exit(&p->p_lock);
616 
617 	mutex_enter(&pidlock);
618 
619 	/*
620 	 * Delete this process from the newstate list of its parent. We
621 	 * will put it in the right place in the sigcld in the end.
622 	 */
623 	delete_ns(p->p_parent, p);
624 
625 	/*
626 	 * Reassign the orphans to the next of kin.
627 	 * Don't rearrange init's orphanage.
628 	 */
629 	if ((q = p->p_orphan) != NULL && p != proc_init) {
630 
631 		proc_t *nokp = p->p_nextofkin;
632 
633 		for (;;) {
634 			q->p_nextofkin = nokp;
635 			if (q->p_nextorph == NULL)
636 				break;
637 			q = q->p_nextorph;
638 		}
639 		q->p_nextorph = nokp->p_orphan;
640 		nokp->p_orphan = p->p_orphan;
641 		p->p_orphan = NULL;
642 	}
643 
644 	/*
645 	 * Reassign the children to init.
646 	 * Don't try to assign init's children to init.
647 	 */
648 	if ((q = p->p_child) != NULL && p != proc_init) {
649 		struct proc	*np;
650 		struct proc	*initp = proc_init;
651 		boolean_t	setzonetop = B_FALSE;
652 
653 		if (!INGLOBALZONE(curproc))
654 			setzonetop = B_TRUE;
655 
656 		pgdetach(p);
657 
658 		do {
659 			np = q->p_sibling;
660 			/*
661 			 * Delete it from its current parent new state
662 			 * list and add it to init new state list
663 			 */
664 			delete_ns(q->p_parent, q);
665 
666 			q->p_ppid = 1;
667 			q->p_pidflag &= ~(CLDNOSIGCHLD | CLDWAITPID);
668 			if (setzonetop) {
669 				mutex_enter(&q->p_lock);
670 				q->p_flag |= SZONETOP;
671 				mutex_exit(&q->p_lock);
672 			}
673 			q->p_parent = initp;
674 
675 			/*
676 			 * Since q will be the first child,
677 			 * it will not have a previous sibling.
678 			 */
679 			q->p_psibling = NULL;
680 			if (initp->p_child) {
681 				initp->p_child->p_psibling = q;
682 			}
683 			q->p_sibling = initp->p_child;
684 			initp->p_child = q;
685 			if (q->p_proc_flag & P_PR_PTRACE) {
686 				mutex_enter(&q->p_lock);
687 				sigtoproc(q, NULL, SIGKILL);
688 				mutex_exit(&q->p_lock);
689 			}
690 			/*
691 			 * sigcld() will add the child to parents
692 			 * newstate list.
693 			 */
694 			if (q->p_stat == SZOMB)
695 				sigcld(q, NULL);
696 		} while ((q = np) != NULL);
697 
698 		p->p_child = NULL;
699 		ASSERT(p->p_child_ns == NULL);
700 	}
701 
702 	TRACE_1(TR_FAC_PROC, TR_PROC_EXIT, "proc_exit: %p", p);
703 
704 	mutex_enter(&p->p_lock);
705 	CL_EXIT(curthread); /* tell the scheduler that curthread is exiting */
706 
707 	/*
708 	 * Have our task accummulate our resource usage data before they
709 	 * become contaminated by p_cacct etc., and before we renounce
710 	 * membership of the task.
711 	 *
712 	 * We do this regardless of whether or not task accounting is active.
713 	 * This is to avoid having nonsense data reported for this task if
714 	 * task accounting is subsequently enabled. The overhead is minimal;
715 	 * by this point, this process has accounted for the usage of all its
716 	 * LWPs. We nonetheless do the work here, and under the protection of
717 	 * pidlock, so that the movement of the process's usage to the task
718 	 * happens at the same time as the removal of the process from the
719 	 * task, from the point of view of exacct_snapshot_task_usage().
720 	 */
721 	exacct_update_task_mstate(p);
722 
723 	hrutime = mstate_aggr_state(p, LMS_USER);
724 	hrstime = mstate_aggr_state(p, LMS_SYSTEM);
725 	p->p_utime = (clock_t)NSEC_TO_TICK(hrutime) + p->p_cutime;
726 	p->p_stime = (clock_t)NSEC_TO_TICK(hrstime) + p->p_cstime;
727 
728 	p->p_acct[LMS_USER]	+= p->p_cacct[LMS_USER];
729 	p->p_acct[LMS_SYSTEM]	+= p->p_cacct[LMS_SYSTEM];
730 	p->p_acct[LMS_TRAP]	+= p->p_cacct[LMS_TRAP];
731 	p->p_acct[LMS_TFAULT]	+= p->p_cacct[LMS_TFAULT];
732 	p->p_acct[LMS_DFAULT]	+= p->p_cacct[LMS_DFAULT];
733 	p->p_acct[LMS_KFAULT]	+= p->p_cacct[LMS_KFAULT];
734 	p->p_acct[LMS_USER_LOCK] += p->p_cacct[LMS_USER_LOCK];
735 	p->p_acct[LMS_SLEEP]	+= p->p_cacct[LMS_SLEEP];
736 	p->p_acct[LMS_WAIT_CPU]	+= p->p_cacct[LMS_WAIT_CPU];
737 	p->p_acct[LMS_STOPPED]	+= p->p_cacct[LMS_STOPPED];
738 
739 	p->p_ru.minflt	+= p->p_cru.minflt;
740 	p->p_ru.majflt	+= p->p_cru.majflt;
741 	p->p_ru.nswap	+= p->p_cru.nswap;
742 	p->p_ru.inblock	+= p->p_cru.inblock;
743 	p->p_ru.oublock	+= p->p_cru.oublock;
744 	p->p_ru.msgsnd	+= p->p_cru.msgsnd;
745 	p->p_ru.msgrcv	+= p->p_cru.msgrcv;
746 	p->p_ru.nsignals += p->p_cru.nsignals;
747 	p->p_ru.nvcsw	+= p->p_cru.nvcsw;
748 	p->p_ru.nivcsw	+= p->p_cru.nivcsw;
749 	p->p_ru.sysc	+= p->p_cru.sysc;
750 	p->p_ru.ioch	+= p->p_cru.ioch;
751 
752 	p->p_stat = SZOMB;
753 	p->p_proc_flag &= ~P_PR_PTRACE;
754 	p->p_wdata = what;
755 	p->p_wcode = (char)why;
756 
757 	cdir = PTOU(p)->u_cdir;
758 	rdir = PTOU(p)->u_rdir;
759 	cwd = PTOU(p)->u_cwd;
760 
761 	ASSERT(cdir != NULL || p->p_parent == &p0);
762 
763 	/*
764 	 * Release resource controls, as they are no longer enforceable.
765 	 */
766 	rctl_set_free(p->p_rctls);
767 
768 	/*
769 	 * Give up task and project memberships.  Decrement tk_nlwps counter
770 	 * for our task.max-lwps resource control.  An extended accounting
771 	 * record, if that facility is active, is scheduled to be written.
772 	 * Zombie processes are false members of task0 for the remainder of
773 	 * their lifetime; no accounting information is recorded for them.
774 	 */
775 	tk = p->p_task;
776 
777 	mutex_enter(&p->p_zone->zone_nlwps_lock);
778 	tk->tk_nlwps--;
779 	tk->tk_proj->kpj_nlwps--;
780 	p->p_zone->zone_nlwps--;
781 	mutex_exit(&p->p_zone->zone_nlwps_lock);
782 	task_detach(p);
783 	p->p_task = task0p;
784 
785 	/*
786 	 * Clear the lwp directory and the lwpid hash table
787 	 * now that /proc can't bother us any more.
788 	 * We free the memory below, after dropping p->p_lock.
789 	 */
790 	lwpdir = p->p_lwpdir;
791 	lwpdir_sz = p->p_lwpdir_sz;
792 	tidhash = p->p_tidhash;
793 	tidhash_sz = p->p_tidhash_sz;
794 	ret_tidhash = p->p_ret_tidhash;
795 	p->p_lwpdir = NULL;
796 	p->p_lwpfree = NULL;
797 	p->p_lwpdir_sz = 0;
798 	p->p_tidhash = NULL;
799 	p->p_tidhash_sz = 0;
800 	p->p_ret_tidhash = NULL;
801 
802 	/*
803 	 * If the process has context ops installed, call the exit routine
804 	 * on behalf of this last remaining thread. Normally exitpctx() is
805 	 * called during thread_exit() or lwp_exit(), but because this is the
806 	 * last thread in the process, we must call it here. By the time
807 	 * thread_exit() is called (below), the association with the relevant
808 	 * process has been lost.
809 	 *
810 	 * We also free the context here.
811 	 */
812 	if (p->p_pctx) {
813 		kpreempt_disable();
814 		exitpctx(p);
815 		kpreempt_enable();
816 
817 		freepctx(p, 0);
818 	}
819 
820 	/*
821 	 * curthread's proc pointer is changed to point to the 'sched'
822 	 * process for the corresponding zone, except in the case when
823 	 * the exiting process is in fact a zsched instance, in which
824 	 * case the proc pointer is set to p0.  We do so, so that the
825 	 * process still points at the right zone when we call the VN_RELE()
826 	 * below.
827 	 *
828 	 * This is because curthread's original proc pointer can be freed as
829 	 * soon as the child sends a SIGCLD to its parent.  We use zsched so
830 	 * that for user processes, even in the final moments of death, the
831 	 * process is still associated with its zone.
832 	 */
833 	if (p != t->t_procp->p_zone->zone_zsched)
834 		t->t_procp = t->t_procp->p_zone->zone_zsched;
835 	else
836 		t->t_procp = &p0;
837 
838 	mutex_exit(&p->p_lock);
839 	if (!evaporate) {
840 		p->p_pidflag &= ~CLDPEND;
841 		sigcld(p, sqp);
842 	} else {
843 		/*
844 		 * Do what sigcld() would do if the disposition
845 		 * of the SIGCHLD signal were set to be ignored.
846 		 */
847 		cv_broadcast(&p->p_srwchan_cv);
848 		freeproc(p);
849 	}
850 	mutex_exit(&pidlock);
851 
852 	/*
853 	 * We don't release u_cdir and u_rdir until SZOMB is set.
854 	 * This protects us against dofusers().
855 	 */
856 	if (cdir)
857 		VN_RELE(cdir);
858 	if (rdir)
859 		VN_RELE(rdir);
860 	if (cwd)
861 		refstr_rele(cwd);
862 
863 	/*
864 	 * task_rele() may ultimately cause the zone to go away (or
865 	 * may cause the last user process in a zone to go away, which
866 	 * signals zsched to go away).  So prior to this call, we must
867 	 * no longer point at zsched.
868 	 */
869 	t->t_procp = &p0;
870 	task_rele(tk);
871 
872 	kmem_free(lwpdir, lwpdir_sz * sizeof (lwpdir_t));
873 	kmem_free(tidhash, tidhash_sz * sizeof (tidhash_t));
874 	while (ret_tidhash != NULL) {
875 		ret_tidhash_t *next = ret_tidhash->rth_next;
876 		kmem_free(ret_tidhash->rth_tidhash,
877 		    ret_tidhash->rth_tidhash_sz * sizeof (tidhash_t));
878 		kmem_free(ret_tidhash, sizeof (*ret_tidhash));
879 		ret_tidhash = next;
880 	}
881 
882 	thread_exit();
883 	/* NOTREACHED */
884 }
885 
886 /*
887  * Format siginfo structure for wait system calls.
888  */
889 void
890 winfo(proc_t *pp, k_siginfo_t *ip, int waitflag)
891 {
892 	ASSERT(MUTEX_HELD(&pidlock));
893 
894 	bzero(ip, sizeof (k_siginfo_t));
895 	ip->si_signo = SIGCLD;
896 	ip->si_code = pp->p_wcode;
897 	ip->si_pid = pp->p_pid;
898 	ip->si_ctid = PRCTID(pp);
899 	ip->si_zoneid = pp->p_zone->zone_id;
900 	ip->si_status = pp->p_wdata;
901 	ip->si_stime = pp->p_stime;
902 	ip->si_utime = pp->p_utime;
903 
904 	if (waitflag) {
905 		pp->p_wcode = 0;
906 		pp->p_wdata = 0;
907 		pp->p_pidflag &= ~CLDPEND;
908 	}
909 }
910 
911 /*
912  * Wait system call.
913  * Search for a terminated (zombie) child,
914  * finally lay it to rest, and collect its status.
915  * Look also for stopped children,
916  * and pass back status from them.
917  */
918 int
919 waitid(idtype_t idtype, id_t id, k_siginfo_t *ip, int options)
920 {
921 	int found;
922 	proc_t *cp, *pp;
923 	int proc_gone;
924 	int waitflag = !(options & WNOWAIT);
925 
926 	/*
927 	 * Obsolete flag, defined here only for binary compatibility
928 	 * with old statically linked executables.  Delete this when
929 	 * we no longer care about these old and broken applications.
930 	 */
931 #define	_WNOCHLD	0400
932 	options &= ~_WNOCHLD;
933 
934 	if (options == 0 || (options & ~WOPTMASK))
935 		return (EINVAL);
936 
937 	switch (idtype) {
938 	case P_PID:
939 	case P_PGID:
940 		if (id < 0 || id >= maxpid)
941 			return (EINVAL);
942 		/* FALLTHROUGH */
943 	case P_ALL:
944 		break;
945 	default:
946 		return (EINVAL);
947 	}
948 
949 	pp = ttoproc(curthread);
950 
951 	/*
952 	 * lock parent mutex so that sibling chain can be searched.
953 	 */
954 	mutex_enter(&pidlock);
955 
956 	/*
957 	 * if we are only looking for exited processes and child_ns list
958 	 * is empty no reason to look at all children.
959 	 */
960 	if (idtype == P_ALL &&
961 	    (options & ~WNOWAIT) == (WNOHANG | WEXITED) &&
962 	    pp->p_child_ns == NULL) {
963 		if (pp->p_child) {
964 			mutex_exit(&pidlock);
965 			bzero(ip, sizeof (k_siginfo_t));
966 			return (0);
967 		}
968 		mutex_exit(&pidlock);
969 		return (ECHILD);
970 	}
971 
972 	while (pp->p_child != NULL) {
973 
974 		proc_gone = 0;
975 
976 		for (cp = pp->p_child_ns; cp != NULL; cp = cp->p_sibling_ns) {
977 			if (idtype != P_PID && (cp->p_pidflag & CLDWAITPID))
978 				continue;
979 			if (idtype == P_PID && id != cp->p_pid)
980 				continue;
981 			if (idtype == P_PGID && id != cp->p_pgrp)
982 				continue;
983 
984 			switch (cp->p_wcode) {
985 
986 			case CLD_TRAPPED:
987 			case CLD_STOPPED:
988 			case CLD_CONTINUED:
989 				cmn_err(CE_PANIC,
990 				    "waitid: wrong state %d on the p_newstate"
991 				    " list", cp->p_wcode);
992 				break;
993 
994 			case CLD_EXITED:
995 			case CLD_DUMPED:
996 			case CLD_KILLED:
997 				if (!(options & WEXITED)) {
998 					/*
999 					 * Count how many are already gone
1000 					 * for good.
1001 					 */
1002 					proc_gone++;
1003 					break;
1004 				}
1005 				if (!waitflag) {
1006 					winfo(cp, ip, 0);
1007 				} else {
1008 					winfo(cp, ip, 1);
1009 					freeproc(cp);
1010 				}
1011 				mutex_exit(&pidlock);
1012 				if (waitflag) {		/* accept SIGCLD */
1013 					sigcld_delete(ip);
1014 					sigcld_repost();
1015 				}
1016 				return (0);
1017 			}
1018 
1019 			if (idtype == P_PID)
1020 				break;
1021 		}
1022 
1023 		/*
1024 		 * Wow! None of the threads on the p_sibling_ns list were
1025 		 * interesting threads. Check all the kids!
1026 		 */
1027 		found = 0;
1028 		for (cp = pp->p_child; cp != NULL; cp = cp->p_sibling) {
1029 			if (idtype == P_PID && id != cp->p_pid)
1030 				continue;
1031 			if (idtype == P_PGID && id != cp->p_pgrp)
1032 				continue;
1033 
1034 			switch (cp->p_wcode) {
1035 			case CLD_TRAPPED:
1036 				if (!(options & WTRAPPED))
1037 					break;
1038 				winfo(cp, ip, waitflag);
1039 				mutex_exit(&pidlock);
1040 				if (waitflag) {		/* accept SIGCLD */
1041 					sigcld_delete(ip);
1042 					sigcld_repost();
1043 				}
1044 				return (0);
1045 
1046 			case CLD_STOPPED:
1047 				if (!(options & WSTOPPED))
1048 					break;
1049 				/* Is it still stopped? */
1050 				mutex_enter(&cp->p_lock);
1051 				if (!jobstopped(cp)) {
1052 					mutex_exit(&cp->p_lock);
1053 					break;
1054 				}
1055 				mutex_exit(&cp->p_lock);
1056 				winfo(cp, ip, waitflag);
1057 				mutex_exit(&pidlock);
1058 				if (waitflag) {		/* accept SIGCLD */
1059 					sigcld_delete(ip);
1060 					sigcld_repost();
1061 				}
1062 				return (0);
1063 
1064 			case CLD_CONTINUED:
1065 				if (!(options & WCONTINUED))
1066 					break;
1067 				winfo(cp, ip, waitflag);
1068 				mutex_exit(&pidlock);
1069 				if (waitflag) {		/* accept SIGCLD */
1070 					sigcld_delete(ip);
1071 					sigcld_repost();
1072 				}
1073 				return (0);
1074 
1075 			case CLD_EXITED:
1076 			case CLD_DUMPED:
1077 			case CLD_KILLED:
1078 				if (idtype != P_PID &&
1079 				    (cp->p_pidflag & CLDWAITPID))
1080 					continue;
1081 				/*
1082 				 * Don't complain if a process was found in
1083 				 * the first loop but we broke out of the loop
1084 				 * because of the arguments passed to us.
1085 				 */
1086 				if (proc_gone == 0) {
1087 					cmn_err(CE_PANIC,
1088 					    "waitid: wrong state on the"
1089 					    " p_child list");
1090 				} else {
1091 					break;
1092 				}
1093 			}
1094 
1095 			found++;
1096 
1097 			if (idtype == P_PID)
1098 				break;
1099 		}
1100 
1101 		/*
1102 		 * If we found no interesting processes at all,
1103 		 * break out and return ECHILD.
1104 		 */
1105 		if (found + proc_gone == 0)
1106 			break;
1107 
1108 		if (options & WNOHANG) {
1109 			mutex_exit(&pidlock);
1110 			bzero(ip, sizeof (k_siginfo_t));
1111 			/*
1112 			 * We should set ip->si_signo = SIGCLD,
1113 			 * but there is an SVVS test that expects
1114 			 * ip->si_signo to be zero in this case.
1115 			 */
1116 			return (0);
1117 		}
1118 
1119 		/*
1120 		 * If we found no processes of interest that could
1121 		 * change state while we wait, we don't wait at all.
1122 		 * Get out with ECHILD according to SVID.
1123 		 */
1124 		if (found == proc_gone)
1125 			break;
1126 
1127 		if (!cv_wait_sig_swap(&pp->p_cv, &pidlock)) {
1128 			mutex_exit(&pidlock);
1129 			return (EINTR);
1130 		}
1131 	}
1132 	mutex_exit(&pidlock);
1133 	return (ECHILD);
1134 }
1135 
1136 int
1137 waitsys(idtype_t idtype, id_t id, siginfo_t *infop, int options)
1138 {
1139 	int error;
1140 	k_siginfo_t info;
1141 
1142 	if (error = waitid(idtype, id, &info, options))
1143 		return (set_errno(error));
1144 	if (copyout(&info, infop, sizeof (k_siginfo_t)))
1145 		return (set_errno(EFAULT));
1146 	return (0);
1147 }
1148 
1149 #ifdef _SYSCALL32_IMPL
1150 
1151 int
1152 waitsys32(idtype_t idtype, id_t id, siginfo_t *infop, int options)
1153 {
1154 	int error;
1155 	k_siginfo_t info;
1156 	siginfo32_t info32;
1157 
1158 	if (error = waitid(idtype, id, &info, options))
1159 		return (set_errno(error));
1160 	siginfo_kto32(&info, &info32);
1161 	if (copyout(&info32, infop, sizeof (info32)))
1162 		return (set_errno(EFAULT));
1163 	return (0);
1164 }
1165 
1166 #endif	/* _SYSCALL32_IMPL */
1167 
1168 void
1169 proc_detach(proc_t *p)
1170 {
1171 	proc_t *q;
1172 
1173 	ASSERT(MUTEX_HELD(&pidlock));
1174 
1175 	q = p->p_parent;
1176 	ASSERT(q != NULL);
1177 
1178 	/*
1179 	 * Take it off the newstate list of its parent
1180 	 */
1181 	delete_ns(q, p);
1182 
1183 	if (q->p_child == p) {
1184 		q->p_child = p->p_sibling;
1185 		/*
1186 		 * If the parent has no children, it better not
1187 		 * have any with new states either!
1188 		 */
1189 		ASSERT(q->p_child ? 1 : q->p_child_ns == NULL);
1190 	}
1191 
1192 	if (p->p_sibling) {
1193 		p->p_sibling->p_psibling = p->p_psibling;
1194 	}
1195 
1196 	if (p->p_psibling) {
1197 		p->p_psibling->p_sibling = p->p_sibling;
1198 	}
1199 }
1200 
1201 /*
1202  * Remove zombie children from the process table.
1203  */
1204 void
1205 freeproc(proc_t *p)
1206 {
1207 	proc_t *q;
1208 
1209 	ASSERT(p->p_stat == SZOMB);
1210 	ASSERT(p->p_tlist == NULL);
1211 	ASSERT(MUTEX_HELD(&pidlock));
1212 
1213 	sigdelq(p, NULL, 0);
1214 	if (p->p_killsqp) {
1215 		siginfofree(p->p_killsqp);
1216 		p->p_killsqp = NULL;
1217 	}
1218 
1219 	prfree(p);	/* inform /proc */
1220 
1221 	/*
1222 	 * Don't free the init processes.
1223 	 * Other dying processes will access it.
1224 	 */
1225 	if (p == proc_init)
1226 		return;
1227 
1228 
1229 	/*
1230 	 * We wait until now to free the cred structure because a
1231 	 * zombie process's credentials may be examined by /proc.
1232 	 * No cred locking needed because there are no threads at this point.
1233 	 */
1234 	upcount_dec(crgetruid(p->p_cred), crgetzoneid(p->p_cred));
1235 	crfree(p->p_cred);
1236 	if (p->p_corefile != NULL) {
1237 		corectl_path_rele(p->p_corefile);
1238 		p->p_corefile = NULL;
1239 	}
1240 	if (p->p_content != NULL) {
1241 		corectl_content_rele(p->p_content);
1242 		p->p_content = NULL;
1243 	}
1244 
1245 	if (p->p_nextofkin && !((p->p_nextofkin->p_flag & SNOWAIT) ||
1246 	    (PTOU(p->p_nextofkin)->u_signal[SIGCLD - 1] == SIG_IGN))) {
1247 		/*
1248 		 * This should still do the right thing since p_utime/stime
1249 		 * get set to the correct value on process exit, so it
1250 		 * should get properly updated
1251 		 */
1252 		p->p_nextofkin->p_cutime += p->p_utime;
1253 		p->p_nextofkin->p_cstime += p->p_stime;
1254 
1255 		p->p_nextofkin->p_cacct[LMS_USER] += p->p_acct[LMS_USER];
1256 		p->p_nextofkin->p_cacct[LMS_SYSTEM] += p->p_acct[LMS_SYSTEM];
1257 		p->p_nextofkin->p_cacct[LMS_TRAP] += p->p_acct[LMS_TRAP];
1258 		p->p_nextofkin->p_cacct[LMS_TFAULT] += p->p_acct[LMS_TFAULT];
1259 		p->p_nextofkin->p_cacct[LMS_DFAULT] += p->p_acct[LMS_DFAULT];
1260 		p->p_nextofkin->p_cacct[LMS_KFAULT] += p->p_acct[LMS_KFAULT];
1261 		p->p_nextofkin->p_cacct[LMS_USER_LOCK]
1262 		    += p->p_acct[LMS_USER_LOCK];
1263 		p->p_nextofkin->p_cacct[LMS_SLEEP] += p->p_acct[LMS_SLEEP];
1264 		p->p_nextofkin->p_cacct[LMS_WAIT_CPU]
1265 		    += p->p_acct[LMS_WAIT_CPU];
1266 		p->p_nextofkin->p_cacct[LMS_STOPPED] += p->p_acct[LMS_STOPPED];
1267 
1268 		p->p_nextofkin->p_cru.minflt	+= p->p_ru.minflt;
1269 		p->p_nextofkin->p_cru.majflt	+= p->p_ru.majflt;
1270 		p->p_nextofkin->p_cru.nswap	+= p->p_ru.nswap;
1271 		p->p_nextofkin->p_cru.inblock	+= p->p_ru.inblock;
1272 		p->p_nextofkin->p_cru.oublock	+= p->p_ru.oublock;
1273 		p->p_nextofkin->p_cru.msgsnd	+= p->p_ru.msgsnd;
1274 		p->p_nextofkin->p_cru.msgrcv	+= p->p_ru.msgrcv;
1275 		p->p_nextofkin->p_cru.nsignals	+= p->p_ru.nsignals;
1276 		p->p_nextofkin->p_cru.nvcsw	+= p->p_ru.nvcsw;
1277 		p->p_nextofkin->p_cru.nivcsw	+= p->p_ru.nivcsw;
1278 		p->p_nextofkin->p_cru.sysc	+= p->p_ru.sysc;
1279 		p->p_nextofkin->p_cru.ioch	+= p->p_ru.ioch;
1280 
1281 	}
1282 
1283 	q = p->p_nextofkin;
1284 	if (q && q->p_orphan == p)
1285 		q->p_orphan = p->p_nextorph;
1286 	else if (q) {
1287 		for (q = q->p_orphan; q; q = q->p_nextorph)
1288 			if (q->p_nextorph == p)
1289 				break;
1290 		ASSERT(q && q->p_nextorph == p);
1291 		q->p_nextorph = p->p_nextorph;
1292 	}
1293 
1294 	proc_detach(p);
1295 	pid_exit(p);	/* frees pid and proc structure */
1296 }
1297 
1298 /*
1299  * Delete process "child" from the newstate list of process "parent"
1300  */
1301 void
1302 delete_ns(proc_t *parent, proc_t *child)
1303 {
1304 	proc_t **ns;
1305 
1306 	ASSERT(MUTEX_HELD(&pidlock));
1307 	ASSERT(child->p_parent == parent);
1308 	for (ns = &parent->p_child_ns; *ns != NULL; ns = &(*ns)->p_sibling_ns) {
1309 		if (*ns == child) {
1310 
1311 			ASSERT((*ns)->p_parent == parent);
1312 
1313 			*ns = child->p_sibling_ns;
1314 			child->p_sibling_ns = NULL;
1315 			return;
1316 		}
1317 	}
1318 }
1319 
1320 /*
1321  * Add process "child" to the new state list of process "parent"
1322  */
1323 void
1324 add_ns(proc_t *parent, proc_t *child)
1325 {
1326 	ASSERT(child->p_sibling_ns == NULL);
1327 	child->p_sibling_ns = parent->p_child_ns;
1328 	parent->p_child_ns = child;
1329 }
1330