Home | History | Annotate | Line # | Download | only in kern
kern_exit.c revision 1.140
      1 /*	$NetBSD: kern_exit.c,v 1.140 2004/05/04 21:23:39 pk Exp $	*/
      2 
      3 /*-
      4  * Copyright (c) 1998, 1999 The NetBSD Foundation, Inc.
      5  * All rights reserved.
      6  *
      7  * This code is derived from software contributed to The NetBSD Foundation
      8  * by Jason R. Thorpe of the Numerical Aerospace Simulation Facility,
      9  * NASA Ames Research Center.
     10  *
     11  * Redistribution and use in source and binary forms, with or without
     12  * modification, are permitted provided that the following conditions
     13  * are met:
     14  * 1. Redistributions of source code must retain the above copyright
     15  *    notice, this list of conditions and the following disclaimer.
     16  * 2. Redistributions in binary form must reproduce the above copyright
     17  *    notice, this list of conditions and the following disclaimer in the
     18  *    documentation and/or other materials provided with the distribution.
     19  * 3. All advertising materials mentioning features or use of this software
     20  *    must display the following acknowledgement:
     21  *	This product includes software developed by the NetBSD
     22  *	Foundation, Inc. and its contributors.
     23  * 4. Neither the name of The NetBSD Foundation nor the names of its
     24  *    contributors may be used to endorse or promote products derived
     25  *    from this software without specific prior written permission.
     26  *
     27  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
     28  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
     29  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
     30  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
     31  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
     32  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
     33  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
     34  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
     35  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
     36  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
     37  * POSSIBILITY OF SUCH DAMAGE.
     38  */
     39 
     40 /*
     41  * Copyright (c) 1982, 1986, 1989, 1991, 1993
     42  *	The Regents of the University of California.  All rights reserved.
     43  * (c) UNIX System Laboratories, Inc.
     44  * All or some portions of this file are derived from material licensed
     45  * to the University of California by American Telephone and Telegraph
     46  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
     47  * the permission of UNIX System Laboratories, Inc.
     48  *
     49  * Redistribution and use in source and binary forms, with or without
     50  * modification, are permitted provided that the following conditions
     51  * are met:
     52  * 1. Redistributions of source code must retain the above copyright
     53  *    notice, this list of conditions and the following disclaimer.
     54  * 2. Redistributions in binary form must reproduce the above copyright
     55  *    notice, this list of conditions and the following disclaimer in the
     56  *    documentation and/or other materials provided with the distribution.
     57  * 3. Neither the name of the University nor the names of its contributors
     58  *    may be used to endorse or promote products derived from this software
     59  *    without specific prior written permission.
     60  *
     61  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
     62  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
     63  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
     64  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
     65  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
     66  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
     67  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
     68  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
     69  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
     70  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
     71  * SUCH DAMAGE.
     72  *
     73  *	@(#)kern_exit.c	8.10 (Berkeley) 2/23/95
     74  */
     75 
     76 #include <sys/cdefs.h>
     77 __KERNEL_RCSID(0, "$NetBSD: kern_exit.c,v 1.140 2004/05/04 21:23:39 pk Exp $");
     78 
     79 #include "opt_ktrace.h"
     80 #include "opt_perfctrs.h"
     81 #include "opt_systrace.h"
     82 #include "opt_sysv.h"
     83 
     84 #include <sys/param.h>
     85 #include <sys/systm.h>
     86 #include <sys/ioctl.h>
     87 #include <sys/proc.h>
     88 #include <sys/tty.h>
     89 #include <sys/time.h>
     90 #include <sys/resource.h>
     91 #include <sys/kernel.h>
     92 #include <sys/ktrace.h>
     93 #include <sys/proc.h>
     94 #include <sys/buf.h>
     95 #include <sys/wait.h>
     96 #include <sys/file.h>
     97 #include <sys/vnode.h>
     98 #include <sys/syslog.h>
     99 #include <sys/malloc.h>
    100 #include <sys/pool.h>
    101 #include <sys/resourcevar.h>
    102 #if defined(PERFCTRS)
    103 #include <sys/pmc.h>
    104 #endif
    105 #include <sys/ptrace.h>
    106 #include <sys/acct.h>
    107 #include <sys/filedesc.h>
    108 #include <sys/ras.h>
    109 #include <sys/signalvar.h>
    110 #include <sys/sched.h>
    111 #include <sys/sa.h>
    112 #include <sys/savar.h>
    113 #include <sys/mount.h>
    114 #include <sys/syscallargs.h>
    115 #include <sys/systrace.h>
    116 
    117 #include <machine/cpu.h>
    118 
    119 #include <uvm/uvm_extern.h>
    120 
    121 #define DEBUG_EXIT
    122 
    123 #ifdef DEBUG_EXIT
    124 int debug_exit = 0;
    125 #define DPRINTF(x) if (debug_exit) printf x
    126 #else
    127 #define DPRINTF(x)
    128 #endif
    129 
    130 static void lwp_exit_hook(struct lwp *, void *);
    131 
    132 /*
    133  * Fill in the appropriate signal information, and signal the parent.
    134  */
    135 static void
    136 exit_psignal(struct proc *p, struct proc *pp, ksiginfo_t *ksi)
    137 {
    138 
    139 	(void)memset(ksi, 0, sizeof(ksiginfo_t));
    140 	if ((ksi->ksi_signo = P_EXITSIG(p)) == SIGCHLD) {
    141 		if (WIFSIGNALED(p->p_xstat)) {
    142 			if (WCOREDUMP(p->p_xstat))
    143 				ksi->ksi_code = CLD_DUMPED;
    144 			else
    145 				ksi->ksi_code = CLD_KILLED;
    146 		} else {
    147 			ksi->ksi_code = CLD_EXITED;
    148 		}
    149 	}
    150 	/*
    151 	 * we fill those in, even for non-SIGCHLD.
    152 	 */
    153 	ksi->ksi_pid = p->p_pid;
    154 	ksi->ksi_uid = p->p_ucred->cr_uid;
    155 	ksi->ksi_status = p->p_xstat;
    156 	/* XXX: is this still valid? */
    157 	ksi->ksi_utime = p->p_ru->ru_utime.tv_sec;
    158 	ksi->ksi_stime = p->p_ru->ru_stime.tv_sec;
    159 }
    160 
    161 /*
    162  * exit --
    163  *	Death of process.
    164  */
    165 int
    166 sys_exit(struct lwp *l, void *v, register_t *retval)
    167 {
    168 	struct sys_exit_args /* {
    169 		syscallarg(int)	rval;
    170 	} */ *uap = v;
    171 
    172 	/* Don't call exit1() multiple times in the same process.*/
    173 	if (l->l_proc->p_flag & P_WEXIT)
    174 		lwp_exit(l);
    175 
    176 	exit1(l, W_EXITCODE(SCARG(uap, rval), 0));
    177 	/* NOTREACHED */
    178 	return (0);
    179 }
    180 
    181 /*
    182  * Exit: deallocate address space and other resources, change proc state
    183  * to zombie, and unlink proc from allproc and parent's lists.  Save exit
    184  * status and rusage for wait().  Check for child processes and orphan them.
    185  */
    186 void
    187 exit1(struct lwp *l, int rv)
    188 {
    189 	struct proc	*p, *q, *nq;
    190 	int		s, sa;
    191 	struct plimit	*plim;
    192 	struct pstats	*pstats;
    193 	struct sigacts	*ps;
    194 	ksiginfo_t	ksi;
    195 	int		do_psignal = 0;
    196 
    197 	p = l->l_proc;
    198 
    199 	if (__predict_false(p == initproc))
    200 		panic("init died (signal %d, exit %d)",
    201 		    WTERMSIG(rv), WEXITSTATUS(rv));
    202 
    203 	p->p_flag |= P_WEXIT;
    204 	if (p->p_flag & P_STOPEXIT) {
    205 		int s;
    206 
    207 		sigminusset(&contsigmask, &p->p_sigctx.ps_siglist);
    208 		SCHED_LOCK(s);
    209 		p->p_stat = SSTOP;
    210 		l->l_stat = LSSTOP;
    211 		p->p_nrlwps--;
    212 		mi_switch(l, NULL);
    213 		SCHED_ASSERT_UNLOCKED();
    214 		splx(s);
    215 	}
    216 
    217 	DPRINTF(("exit1: %d.%d exiting.\n", p->p_pid, l->l_lid));
    218 	/*
    219 	 * Disable scheduler activation upcalls.
    220 	 * We're trying to get out of here.
    221 	 */
    222 	sa = 0;
    223 	if (p->p_sa != NULL) {
    224 		l->l_flag &= ~L_SA;
    225 #if 0
    226 		p->p_flag &= ~P_SA;
    227 #endif
    228 		sa = 1;
    229 	}
    230 
    231 #ifdef PGINPROF
    232 	vmsizmon();
    233 #endif
    234 	if (p->p_flag & P_PROFIL)
    235 		stopprofclock(p);
    236 	p->p_ru = pool_get(&rusage_pool, PR_WAITOK);
    237 	/*
    238 	 * If parent is waiting for us to exit or exec, P_PPWAIT is set; we
    239 	 * wake up the parent early to avoid deadlock.
    240 	 */
    241 	if (p->p_flag & P_PPWAIT) {
    242 		p->p_flag &= ~P_PPWAIT;
    243 		wakeup(p->p_pptr);
    244 	}
    245 	sigfillset(&p->p_sigctx.ps_sigignore);
    246 	sigemptyset(&p->p_sigctx.ps_siglist);
    247 	p->p_sigctx.ps_sigcheck = 0;
    248 	timers_free(p, TIMERS_ALL);
    249 
    250 	if (sa || (p->p_nlwps > 1)) {
    251 		exit_lwps(l);
    252 
    253 		/*
    254 		 * Collect thread u-areas.
    255 		 */
    256 		uvm_uarea_drain(FALSE);
    257 	}
    258 
    259 #if defined(__HAVE_RAS)
    260 	ras_purgeall(p);
    261 #endif
    262 
    263 	/*
    264 	 * Close open files and release open-file table.
    265 	 * This may block!
    266 	 */
    267 	fdfree(p);
    268 	cwdfree(p);
    269 
    270 	doexithooks(p);
    271 
    272 	if (SESS_LEADER(p)) {
    273 		struct session *sp = p->p_session;
    274 		struct tty *tp;
    275 
    276 		if (sp->s_ttyvp) {
    277 			int s;
    278 			/*
    279 			 * Controlling process.
    280 			 * Signal foreground pgrp,
    281 			 * drain controlling terminal
    282 			 * and revoke access to controlling terminal.
    283 			 */
    284 			tp = sp->s_ttyp;
    285 			s = spltty();
    286 			TTY_LOCK(tp);
    287 			if (tp->t_session == sp) {
    288 				if (tp->t_pgrp)
    289 					pgsignal(tp->t_pgrp, SIGHUP, 1);
    290 				/* we can't guarantee the revoke will do this */
    291 				tp->t_pgrp = NULL;
    292 				tp->t_session = NULL;
    293 				TTY_UNLOCK(tp);
    294 				splx(s);
    295 				SESSRELE(sp);
    296 				(void) ttywait(tp);
    297 				/*
    298 				 * The tty could have been revoked
    299 				 * if we blocked.
    300 				 */
    301 				if (sp->s_ttyvp)
    302 					VOP_REVOKE(sp->s_ttyvp, REVOKEALL);
    303 			} else {
    304 				TTY_UNLOCK(tp);
    305 				splx(s);
    306 			}
    307 			if (sp->s_ttyvp)
    308 				vrele(sp->s_ttyvp);
    309 			sp->s_ttyvp = NULL;
    310 			/*
    311 			 * s_ttyp is not zero'd; we use this to indicate
    312 			 * that the session once had a controlling terminal.
    313 			 * (for logging and informational purposes)
    314 			 */
    315 		}
    316 		sp->s_leader = NULL;
    317 	}
    318 	fixjobc(p, p->p_pgrp, 0);
    319 	(void)acct_process(p);
    320 #ifdef KTRACE
    321 	/*
    322 	 * release trace file
    323 	 */
    324 	ktrderef(p);
    325 #endif
    326 #ifdef SYSTRACE
    327 	systrace_sys_exit(p);
    328 #endif
    329 	/*
    330 	 * If emulation has process exit hook, call it now.
    331 	 */
    332 	if (p->p_emul->e_proc_exit)
    333 		(*p->p_emul->e_proc_exit)(p);
    334 
    335 	/*
    336 	 * Free the VM resources we're still holding on to.
    337 	 * We must do this from a valid thread because doing
    338 	 * so may block. This frees vmspace, which we don't
    339 	 * need anymore. The only remaining lwp is the one
    340 	 * we run at this moment, nothing runs in userland
    341 	 * anymore.
    342 	 */
    343 	uvm_proc_exit(p);
    344 
    345 	/*
    346 	 * Give machine-dependent code a chance to free any
    347 	 * MD LWP resources while we can still block. This must be done
    348 	 * before uvm_lwp_exit(), in case these resources are in the
    349 	 * PCB.
    350 	 * THIS IS LAST BLOCKING OPERATION.
    351 	 */
    352 #ifndef __NO_CPU_LWP_FREE
    353 	cpu_lwp_free(l, 1);
    354 #endif
    355 
    356 	pmap_deactivate(l);
    357 
    358 	/*
    359 	 * NOTE: WE ARE NO LONGER ALLOWED TO SLEEP!
    360 	 */
    361 
    362 	/*
    363 	 * Save exit status and final rusage info, adding in child rusage
    364 	 * info and self times.
    365 	 * In order to pick up the time for the current execution, we must
    366 	 * do this before unlinking the lwp from l_list.
    367 	 */
    368 	p->p_xstat = rv;
    369 	*p->p_ru = p->p_stats->p_ru;
    370 	calcru(p, &p->p_ru->ru_utime, &p->p_ru->ru_stime, NULL);
    371 	ruadd(p->p_ru, &p->p_stats->p_cru);
    372 
    373 	/*
    374 	 * Notify interested parties of our demise.
    375 	 */
    376 	KNOTE(&p->p_klist, NOTE_EXIT);
    377 
    378 #if PERFCTRS
    379 	/*
    380 	 * Save final PMC information in parent process & clean up.
    381 	 */
    382 	if (PMC_ENABLED(p)) {
    383 		pmc_save_context(p);
    384 		pmc_accumulate(p->p_pptr, p);
    385 		pmc_process_exit(p);
    386 	}
    387 #endif
    388 
    389 	s = proclist_lock_write();
    390 	/*
    391 	 * Reset p_opptr pointer of all former children which got
    392 	 * traced by another process and were reparented. We reset
    393 	 * it to NULL here; the trace detach code then reparents
    394 	 * the child to initproc. We only check allproc list, since
    395 	 * eventual former children on zombproc list won't reference
    396 	 * p_opptr anymore.
    397 	 */
    398 	if (p->p_flag & P_CHTRACED) {
    399 		LIST_FOREACH(q, &allproc, p_list) {
    400 			if (q->p_opptr == p)
    401 				q->p_opptr = NULL;
    402 		}
    403 	}
    404 
    405 	/*
    406 	 * Give orphaned children to init(8).
    407 	 */
    408 	q = LIST_FIRST(&p->p_children);
    409 	if (q)		/* only need this if any child is SZOMB */
    410 		wakeup(initproc);
    411 	for (; q != NULL; q = nq) {
    412 		nq = LIST_NEXT(q, p_sibling);
    413 
    414 		/*
    415 		 * Traced processes are killed since their existence
    416 		 * means someone is screwing up. Since we reset the
    417 		 * trace flags, the logic in sys_wait4() would not be
    418 		 * triggered to reparent the process to its
    419 		 * original parent, so we must do this here.
    420 		 */
    421 		if (q->p_flag & P_TRACED) {
    422 			if (q->p_opptr != q->p_pptr) {
    423 				struct proc *t = q->p_opptr;
    424 				proc_reparent(q, t ? t : initproc);
    425 				q->p_opptr = NULL;
    426 			} else
    427 				proc_reparent(q, initproc);
    428 			q->p_flag &= ~(P_TRACED|P_WAITED|P_FSTRACE);
    429 			psignal(q, SIGKILL);
    430 		} else {
    431 			proc_reparent(q, initproc);
    432 		}
    433 	}
    434 
    435 	/*
    436 	 * Move proc from allproc to zombproc, it's now ready
    437 	 * to be collected by parent. Remaining lwp resources
    438 	 * will be freed in lwp_exit2() once we've switch to idle
    439 	 * context.
    440 	 * Changing the state to SZOMB stops it being found by pfind().
    441 	 */
    442 	LIST_REMOVE(p, p_list);
    443 	LIST_INSERT_HEAD(&zombproc, p, p_list);
    444 	p->p_stat = SZOMB;
    445 
    446 	LIST_REMOVE(l, l_list);
    447 	LIST_REMOVE(l, l_sibling);
    448 	l->l_flag |= L_DETACHED|L_PROCEXIT;	/* detached from proc too */
    449 	l->l_stat = LSDEAD;
    450 
    451 	p->p_nrlwps--;
    452 	p->p_nlwps--;
    453 
    454 	/* Put in front of parent's sibling list for parent to collect it */
    455 	q = p->p_pptr;
    456 	q->p_nstopchild++;
    457 	if (LIST_FIRST(&q->p_children) != p) {
    458 		/* Put child where it can be found quickly */
    459 		LIST_REMOVE(p, p_sibling);
    460 		LIST_INSERT_HEAD(&q->p_children, p, p_sibling);
    461 	}
    462 
    463 	/*
    464 	 * Notify parent that we're gone.  If parent has the P_NOCLDWAIT
    465 	 * flag set, notify init instead (and hope it will handle
    466 	 * this situation).
    467 	 */
    468 	if (q->p_flag & P_NOCLDWAIT) {
    469 		proc_reparent(p, initproc);
    470 
    471 		/*
    472 		 * If this was the last child of our parent, notify
    473 		 * parent, so in case he was wait(2)ing, he will
    474 		 * continue.
    475 		 */
    476 		if (LIST_FIRST(&q->p_children) == NULL)
    477 			wakeup(q);
    478 	}
    479 
    480 	/*
    481 	 * Clear curlwp after we've done all operations
    482 	 * that could block, and before tearing down the rest
    483 	 * of the process state that might be used from clock, etc.
    484 	 * Also, can't clear curlwp while we're still runnable,
    485 	 * as we're not on a run queue (we are current, just not
    486 	 * a proper proc any longer!).
    487 	 *
    488 	 * Other substructures are freed from wait().
    489 	 */
    490 	curlwp = NULL;
    491 
    492 	/* Delay release until after dropping the proclist lock */
    493 	plim = p->p_limit;
    494 	pstats = p->p_stats;
    495 	ps = p->p_sigacts;
    496 
    497 	p->p_limit = NULL;
    498 	p->p_stats = NULL;
    499 	p->p_sigacts = NULL;
    500 
    501 	/* Reload parent pointer, since p may have been reparented above */
    502 	q = p->p_pptr;
    503 
    504 	if ((p->p_flag & P_FSTRACE) == 0 && p->p_exitsig != 0) {
    505 		exit_psignal(p, q, &ksi);
    506 		do_psignal = 1;
    507 	}
    508 
    509 	/*
    510 	 * Once we release the proclist lock, we shouldn't touch the
    511 	 * process structure anymore, since it's now on the zombie
    512 	 * list and available for collection by the parent.
    513 	 */
    514 	proclist_unlock_write(s);
    515 
    516 	if (do_psignal)
    517 		kpsignal(q, &ksi, NULL);
    518 
    519 	/* Wake up the parent so it can get exit status. */
    520 	wakeup(q);
    521 
    522 	/* Release substructures */
    523 	sigactsfree(ps);
    524 	limfree(plim);
    525 	pstatsfree(pstats);
    526 
    527 #ifdef DEBUG
    528 	/* Nothing should use the process link anymore */
    529 	l->l_proc = NULL;
    530 #endif
    531 
    532 	/* This process no longer needs to hold the kernel lock. */
    533 	KERNEL_PROC_UNLOCK(l);
    534 
    535 	/*
    536 	 * Finally, call machine-dependent code to switch to a new
    537 	 * context (possibly the idle context).  Once we are no longer
    538 	 * using the dead lwp's stack, lwp_exit2() will be called
    539 	 * to arrange for the resources to be released.
    540 	 *
    541 	 * Note that cpu_exit() will end with a call equivalent to
    542 	 * cpu_switch(), finishing our execution (pun intended).
    543 	 */
    544 
    545 	uvmexp.swtch++;
    546 	cpu_exit(l);
    547 }
    548 
    549 void
    550 exit_lwps(struct lwp *l)
    551 {
    552 	struct proc *p;
    553 	struct lwp *l2;
    554 	struct sadata_vp *vp;
    555 	int s, error;
    556 	lwpid_t		waited;
    557 
    558 	p = l->l_proc;
    559 
    560 	/* XXX SMP
    561 	 * This would be the right place to IPI any LWPs running on
    562 	 * other processors so that they can notice the userret exit hook.
    563 	 */
    564 	p->p_userret = lwp_exit_hook;
    565 	p->p_userret_arg = NULL;
    566 
    567 	if (p->p_sa) {
    568 		SLIST_FOREACH(vp, &p->p_sa->sa_vps, savp_next) {
    569 			/*
    570 			 * Make SA-cached LWPs normal process runnable
    571 			 * LWPs so that they'll also self-destruct.
    572 			 */
    573 			DPRINTF(("exit_lwps: Making cached LWPs of %d on VP %d runnable: ",
    574 				    p->p_pid, vp->savp_id));
    575 			SCHED_LOCK(s);
    576 			while ((l2 = sa_getcachelwp(vp)) != 0) {
    577 				l2->l_priority = l2->l_usrpri;
    578 				setrunnable(l2);
    579 				DPRINTF(("%d ", l2->l_lid));
    580 			}
    581 			SCHED_UNLOCK(s);
    582 			DPRINTF(("\n"));
    583 
    584 			/*
    585 			 * Clear wokenq, the LWPs on the queue will
    586 			 * run below.
    587 			 */
    588 			vp->savp_wokenq_head = NULL;
    589 		}
    590 	}
    591 
    592 	/*
    593 	 * Interrupt LWPs in interruptable sleep, unsuspend suspended
    594 	 * LWPs, make detached LWPs undetached (so we can wait for
    595 	 * them) and then wait for everyone else to finish.
    596 	 */
    597 	LIST_FOREACH(l2, &p->p_lwps, l_sibling) {
    598 		l2->l_flag &= ~(L_DETACHED|L_SA);
    599 
    600 		if ((l2->l_stat == LSSLEEP && (l2->l_flag & L_SINTR)) ||
    601 		    l2->l_stat == LSSUSPENDED || l2->l_stat == LSSTOP) {
    602 			SCHED_LOCK(s);
    603 			setrunnable(l2);
    604 			SCHED_UNLOCK(s);
    605 			DPRINTF(("exit_lwps: Made %d.%d runnable\n",
    606 			    p->p_pid, l2->l_lid));
    607 		}
    608 	}
    609 
    610 
    611 	while (p->p_nlwps > 1) {
    612 		DPRINTF(("exit_lwps: waiting for %d LWPs (%d runnable, %d zombies)\n",
    613 		    p->p_nlwps, p->p_nrlwps, p->p_nzlwps));
    614 		error = lwp_wait1(l, 0, &waited, LWPWAIT_EXITCONTROL);
    615 		if (error)
    616 			panic("exit_lwps: lwp_wait1 failed with error %d\n",
    617 			    error);
    618 		DPRINTF(("exit_lwps: Got LWP %d from lwp_wait1()\n", waited));
    619 	}
    620 
    621 	p->p_userret = NULL;
    622 }
    623 
    624 /* Wrapper function for use in p_userret */
    625 static void
    626 lwp_exit_hook(struct lwp *l, void *arg)
    627 {
    628 	KERNEL_PROC_LOCK(l);
    629 	lwp_exit(l);
    630 }
    631 
    632 int
    633 sys_wait4(struct lwp *l, void *v, register_t *retval)
    634 {
    635 	struct sys_wait4_args /* {
    636 		syscallarg(int)			pid;
    637 		syscallarg(int *)		status;
    638 		syscallarg(int)			options;
    639 		syscallarg(struct rusage *)	rusage;
    640 	} */ *uap = v;
    641 	struct proc	*child, *parent;
    642 	int		status, error;
    643 
    644 	parent = l->l_proc;
    645 
    646 	if (SCARG(uap, pid) == 0)
    647 		SCARG(uap, pid) = -parent->p_pgid;
    648 	if (SCARG(uap, options) & ~(WUNTRACED|WNOHANG|WALTSIG|WALLSIG))
    649 		return (EINVAL);
    650 
    651 	error = find_stopped_child(parent, SCARG(uap,pid), SCARG(uap,options),
    652 				&child);
    653 	if (error != 0)
    654 		return error;
    655 	if (child == NULL) {
    656 		*retval = 0;
    657 		return 0;
    658 	}
    659 
    660 	/*
    661 	 * Collect child u-areas.
    662 	 */
    663 	uvm_uarea_drain(FALSE);
    664 
    665 	retval[0] = child->p_pid;
    666 
    667 	if (child->p_stat == SZOMB) {
    668 		if (SCARG(uap, status)) {
    669 			status = child->p_xstat;	/* convert to int */
    670 			error = copyout(&status, SCARG(uap, status),
    671 					sizeof(status));
    672 			if (error)
    673 				return (error);
    674 		}
    675 		if (SCARG(uap, rusage)) {
    676 			error = copyout(child->p_ru, SCARG(uap, rusage),
    677 				    sizeof(struct rusage));
    678 			if (error)
    679 				return (error);
    680 		}
    681 
    682 		proc_free(child);
    683 		return 0;
    684 	}
    685 
    686 	/* child state must be SSTOP */
    687 	if (SCARG(uap, status)) {
    688 		status = W_STOPCODE(child->p_xstat);
    689 		return copyout(&status, SCARG(uap, status), sizeof(status));
    690 	}
    691 	return 0;
    692 }
    693 
    694 /*
    695  * Scan list of child processes for a child process that has stopped or
    696  * exited.  Used by sys_wait4 and 'compat' equivalents.
    697  */
    698 int
    699 find_stopped_child(struct proc *parent, pid_t pid, int options,
    700 	struct proc **child_p)
    701 {
    702 	struct proc *child;
    703 	int error;
    704 
    705 	for (;;) {
    706 		proclist_lock_read();
    707 		error = ECHILD;
    708 		LIST_FOREACH(child, &parent->p_children, p_sibling) {
    709 			if (pid >= 0) {
    710 				if (child->p_pid != pid) {
    711 					child = p_find(pid, PFIND_ZOMBIE |
    712 								PFIND_LOCKED);
    713 					if (child == NULL
    714 					    || child->p_pptr != parent) {
    715 						child = NULL;
    716 						break;
    717 					}
    718 				}
    719 			} else
    720 				if (pid != WAIT_ANY && child->p_pgid != -pid)
    721 					/* child not in correct pgrp */
    722 					continue;
    723 			/*
    724 			 * Wait for processes with p_exitsig != SIGCHLD
    725 			 * processes only if WALTSIG is set; wait for
    726 			 * processes with p_exitsig == SIGCHLD only
    727 			 * if WALTSIG is clear.
    728 			 */
    729 			if (((options & WALLSIG) == 0) &&
    730 			    (options & WALTSIG ? child->p_exitsig == SIGCHLD
    731 						: P_EXITSIG(child) != SIGCHLD)){
    732 				if (child->p_pid == pid) {
    733 					child = NULL;
    734 					break;
    735 				}
    736 				continue;
    737 			}
    738 
    739 			error = 0;
    740 			if (child->p_stat == SZOMB && !(options & WNOZOMBIE))
    741 				break;
    742 
    743 			if (child->p_stat == SSTOP &&
    744 			    (child->p_flag & P_WAITED) == 0 &&
    745 			    (child->p_flag & P_TRACED || options & WUNTRACED)) {
    746 				if ((options & WNOWAIT) == 0) {
    747 					child->p_flag |= P_WAITED;
    748 					parent->p_nstopchild--;
    749 				}
    750 				break;
    751 			}
    752 			if (parent->p_nstopchild == 0 || child->p_pid == pid) {
    753 				child = NULL;
    754 				break;
    755 			}
    756 		}
    757 		proclist_unlock_read();
    758 		if (child != NULL || error != 0 || options & WNOHANG) {
    759 			*child_p = child;
    760 			return error;
    761 		}
    762 		error = tsleep(parent, PWAIT | PCATCH, "wait", 0);
    763 		if (error != 0)
    764 			return error;
    765 	}
    766 }
    767 
    768 /*
    769  * Free a process after parent has taken all the state info.
    770  */
    771 void
    772 proc_free(struct proc *p)
    773 {
    774 	struct proc *parent = p->p_pptr;
    775 	ksiginfo_t ksi;
    776 	int s;
    777 
    778 	KASSERT(p->p_nlwps == 0);
    779 	KASSERT(p->p_nzlwps == 0);
    780 	KASSERT(p->p_nrlwps == 0);
    781 	KASSERT(LIST_EMPTY(&p->p_lwps));
    782 
    783 	/*
    784 	 * If we got the child via ptrace(2) or procfs, and
    785 	 * the parent is different (meaning the process was
    786 	 * attached, rather than run as a child), then we need
    787 	 * to give it back to the old parent, and send the
    788 	 * parent the exit signal.  The rest of the cleanup
    789 	 * will be done when the old parent waits on the child.
    790 	 */
    791 	if ((p->p_flag & P_TRACED) && p->p_opptr != parent){
    792 		parent = p->p_opptr;
    793 		if (parent == NULL)
    794 			parent = initproc;
    795 		proc_reparent(p, parent);
    796 		p->p_opptr = NULL;
    797 		p->p_flag &= ~(P_TRACED|P_WAITED|P_FSTRACE);
    798 		if (p->p_exitsig != 0) {
    799 			exit_psignal(p, parent, &ksi);
    800 			kpsignal(parent, &ksi, NULL);
    801 		}
    802 		wakeup(parent);
    803 		return;
    804 	}
    805 
    806 	scheduler_wait_hook(parent, p);
    807 	p->p_xstat = 0;
    808 
    809 	ruadd(&parent->p_stats->p_cru, p->p_ru);
    810 
    811 	/*
    812 	 * At this point we are going to start freeing the final resources.
    813 	 * If anyone tries to access the proc structure after here they
    814 	 * will get a shock - bits are missing.
    815 	 * Attempt to make it hard!
    816 	 */
    817 
    818 	p->p_stat = SIDL;		/* not even a zombie any more */
    819 
    820 	pool_put(&rusage_pool, p->p_ru);
    821 
    822 	/*
    823 	 * Finally finished with old proc entry.
    824 	 * Unlink it from its process group and free it.
    825 	 */
    826 	leavepgrp(p);
    827 
    828 	s = proclist_lock_write();
    829 	LIST_REMOVE(p, p_list);	/* off zombproc */
    830 	p->p_pptr->p_nstopchild--;
    831 	LIST_REMOVE(p, p_sibling);
    832 	proclist_unlock_write(s);
    833 
    834 	/*
    835 	 * Decrement the count of procs running with this uid.
    836 	 */
    837 	(void)chgproccnt(p->p_cred->p_ruid, -1);
    838 
    839 	/*
    840 	 * Free up credentials.
    841 	 */
    842 	if (--p->p_cred->p_refcnt == 0) {
    843 		crfree(p->p_cred->pc_ucred);
    844 		pool_put(&pcred_pool, p->p_cred);
    845 	}
    846 
    847 	/*
    848 	 * Release reference to text vnode
    849 	 */
    850 	if (p->p_textvp)
    851 		vrele(p->p_textvp);
    852 
    853 	/* Release any SA state. */
    854 	if (p->p_sa)
    855 		sa_release(p);
    856 
    857 	/* Free proc structure and let pid be reallocated */
    858 	proc_free_mem(p);
    859 }
    860 
    861 /*
    862  * make process 'parent' the new parent of process 'child'.
    863  *
    864  * Must be called with proclist_lock_write() held.
    865  */
    866 void
    867 proc_reparent(struct proc *child, struct proc *parent)
    868 {
    869 
    870 	if (child->p_pptr == parent)
    871 		return;
    872 
    873 	if (child->p_stat == SZOMB
    874 	    || (child->p_stat == SSTOP && !(child->p_flag & P_WAITED))) {
    875 		child->p_pptr->p_nstopchild--;
    876 		parent->p_nstopchild++;
    877 	}
    878 	if (parent == initproc)
    879 		child->p_exitsig = SIGCHLD;
    880 
    881 	LIST_REMOVE(child, p_sibling);
    882 	LIST_INSERT_HEAD(&parent->p_children, child, p_sibling);
    883 	child->p_pptr = parent;
    884 }
    885