Home | History | Annotate | Line # | Download | only in kern
kern_exit.c revision 1.136
      1 /*	$NetBSD: kern_exit.c,v 1.136 2004/02/18 14:42:20 hannken 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 1.136 2004/02/18 14:42:20 hannken 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 static void exit_psignal(struct proc *, struct proc *);
    132 
    133 /*
    134  * Fill in the appropriate signal information, and signal the parent.
    135  */
    136 static void
    137 exit_psignal(struct proc *p, struct proc *pp)
    138 {
    139 	ksiginfo_t ksi;
    140 
    141 	(void)memset(&ksi, 0, sizeof(ksi));
    142 	if ((ksi.ksi_signo = P_EXITSIG(p)) == SIGCHLD) {
    143 		if (WIFSIGNALED(p->p_xstat)) {
    144 			if (WCOREDUMP(p->p_xstat))
    145 				ksi.ksi_code = CLD_DUMPED;
    146 			else
    147 				ksi.ksi_code = CLD_KILLED;
    148 		} else {
    149 			ksi.ksi_code = CLD_EXITED;
    150 		}
    151 	}
    152 	/*
    153 	 * we fill those in, even for non-SIGCHLD.
    154 	 */
    155 	ksi.ksi_pid = p->p_pid;
    156 	ksi.ksi_uid = p->p_ucred->cr_uid;
    157 	ksi.ksi_status = p->p_xstat;
    158 	/* XXX: is this still valid? */
    159 	ksi.ksi_utime = p->p_ru->ru_utime.tv_sec;
    160 	ksi.ksi_stime = p->p_ru->ru_stime.tv_sec;
    161 	kpsignal(pp, &ksi, NULL);
    162 }
    163 
    164 /*
    165  * exit --
    166  *	Death of process.
    167  */
    168 int
    169 sys_exit(struct lwp *l, void *v, register_t *retval)
    170 {
    171 	struct sys_exit_args /* {
    172 		syscallarg(int)	rval;
    173 	} */ *uap = v;
    174 
    175 	/* Don't call exit1() multiple times in the same process.*/
    176 	if (l->l_proc->p_flag & P_WEXIT)
    177 		lwp_exit(l);
    178 
    179 	exit1(l, W_EXITCODE(SCARG(uap, rval), 0));
    180 	/* NOTREACHED */
    181 	return (0);
    182 }
    183 
    184 /*
    185  * Exit: deallocate address space and other resources, change proc state
    186  * to zombie, and unlink proc from allproc and parent's lists.  Save exit
    187  * status and rusage for wait().  Check for child processes and orphan them.
    188  */
    189 void
    190 exit1(struct lwp *l, int rv)
    191 {
    192 	struct proc	*p, *q, *nq;
    193 	int		s, sa;
    194 
    195 	p = l->l_proc;
    196 
    197 	if (__predict_false(p == initproc))
    198 		panic("init died (signal %d, exit %d)",
    199 		    WTERMSIG(rv), WEXITSTATUS(rv));
    200 
    201 	p->p_flag |= P_WEXIT;
    202 	if (p->p_flag & P_STOPEXIT) {
    203 		int s;
    204 
    205 		sigminusset(&contsigmask, &p->p_sigctx.ps_siglist);
    206 		SCHED_LOCK(s);
    207 		p->p_stat = SSTOP;
    208 		l->l_stat = LSSTOP;
    209 		p->p_nrlwps--;
    210 		mi_switch(l, NULL);
    211 		SCHED_ASSERT_UNLOCKED();
    212 		splx(s);
    213 	}
    214 
    215 	DPRINTF(("exit1: %d.%d exiting.\n", p->p_pid, l->l_lid));
    216 	/*
    217 	 * Disable scheduler activation upcalls.
    218 	 * We're trying to get out of here.
    219 	 */
    220 	sa = 0;
    221 	if (p->p_sa != NULL) {
    222 		l->l_flag &= ~L_SA;
    223 #if 0
    224 		p->p_flag &= ~P_SA;
    225 #endif
    226 		sa = 1;
    227 	}
    228 
    229 #ifdef PGINPROF
    230 	vmsizmon();
    231 #endif
    232 	if (p->p_flag & P_PROFIL)
    233 		stopprofclock(p);
    234 	p->p_ru = pool_get(&rusage_pool, PR_WAITOK);
    235 	/*
    236 	 * If parent is waiting for us to exit or exec, P_PPWAIT is set; we
    237 	 * wake up the parent early to avoid deadlock.
    238 	 */
    239 	if (p->p_flag & P_PPWAIT) {
    240 		p->p_flag &= ~P_PPWAIT;
    241 		wakeup(p->p_pptr);
    242 	}
    243 	sigfillset(&p->p_sigctx.ps_sigignore);
    244 	sigemptyset(&p->p_sigctx.ps_siglist);
    245 	p->p_sigctx.ps_sigcheck = 0;
    246 	timers_free(p, TIMERS_ALL);
    247 
    248 	if (sa || (p->p_nlwps > 1)) {
    249 		exit_lwps(l);
    250 
    251 		/*
    252 		 * Collect thread u-areas.
    253 		 */
    254 		uvm_uarea_drain(FALSE);
    255 	}
    256 
    257 #if defined(__HAVE_RAS)
    258 	ras_purgeall(p);
    259 #endif
    260 
    261 	/*
    262 	 * Close open files and release open-file table.
    263 	 * This may block!
    264 	 */
    265 	fdfree(p);
    266 	cwdfree(p);
    267 
    268 	doexithooks(p);
    269 
    270 	if (SESS_LEADER(p)) {
    271 		struct session *sp = p->p_session;
    272 		struct tty *tp;
    273 
    274 		if (sp->s_ttyvp) {
    275 			/*
    276 			 * Controlling process.
    277 			 * Signal foreground pgrp,
    278 			 * drain controlling terminal
    279 			 * and revoke access to controlling terminal.
    280 			 */
    281 			tp = sp->s_ttyp;
    282 			TTY_LOCK(tp);
    283 			if (tp->t_session == sp) {
    284 				if (tp->t_pgrp)
    285 					pgsignal(tp->t_pgrp, SIGHUP, 1);
    286 				/* we can't guarantee the revoke will do this */
    287 				tp->t_pgrp = NULL;
    288 				tp->t_session = NULL;
    289 				TTY_UNLOCK(tp);
    290 				SESSRELE(sp);
    291 				(void) ttywait(tp);
    292 				/*
    293 				 * The tty could have been revoked
    294 				 * if we blocked.
    295 				 */
    296 				if (sp->s_ttyvp)
    297 					VOP_REVOKE(sp->s_ttyvp, REVOKEALL);
    298 			} else
    299 				TTY_UNLOCK(tp);
    300 			if (sp->s_ttyvp)
    301 				vrele(sp->s_ttyvp);
    302 			sp->s_ttyvp = NULL;
    303 			/*
    304 			 * s_ttyp is not zero'd; we use this to indicate
    305 			 * that the session once had a controlling terminal.
    306 			 * (for logging and informational purposes)
    307 			 */
    308 		}
    309 		sp->s_leader = NULL;
    310 	}
    311 	fixjobc(p, p->p_pgrp, 0);
    312 	(void)acct_process(p);
    313 #ifdef KTRACE
    314 	/*
    315 	 * release trace file
    316 	 */
    317 	ktrderef(p);
    318 #endif
    319 #ifdef SYSTRACE
    320 	systrace_sys_exit(p);
    321 #endif
    322 	/*
    323 	 * If emulation has process exit hook, call it now.
    324 	 */
    325 	if (p->p_emul->e_proc_exit)
    326 		(*p->p_emul->e_proc_exit)(p);
    327 
    328 	/*
    329 	 * Reset p_opptr pointer of all former children which got
    330 	 * traced by another process and were reparented. We reset
    331 	 * it to NULL here; the trace detach code then reparents
    332 	 * the child to initproc. We only check allproc list, since
    333 	 * eventual former children on zombproc list won't reference
    334 	 * p_opptr anymore.
    335 	 */
    336 	s = proclist_lock_write();
    337 	if (p->p_flag & P_CHTRACED) {
    338 		LIST_FOREACH(q, &allproc, p_list) {
    339 			if (q->p_opptr == p)
    340 				q->p_opptr = NULL;
    341 		}
    342 	}
    343 
    344 	/*
    345 	 * Give orphaned children to init(8).
    346 	 */
    347 	q = LIST_FIRST(&p->p_children);
    348 	if (q)		/* only need this if any child is SZOMB */
    349 		wakeup(initproc);
    350 	for (; q != NULL; q = nq) {
    351 		nq = LIST_NEXT(q, p_sibling);
    352 
    353 		/*
    354 		 * Traced processes are killed since their existence
    355 		 * means someone is screwing up. Since we reset the
    356 		 * trace flags, the logic in sys_wait4() would not be
    357 		 * triggered to reparent the process to its
    358 		 * original parent, so we must do this here.
    359 		 */
    360 		if (q->p_flag & P_TRACED) {
    361 			if (q->p_opptr != q->p_pptr) {
    362 				struct proc *t = q->p_opptr;
    363 				proc_reparent(q, t ? t : initproc);
    364 				q->p_opptr = NULL;
    365 			} else
    366 				proc_reparent(q, initproc);
    367 			q->p_flag &= ~(P_TRACED|P_WAITED|P_FSTRACE);
    368 			psignal(q, SIGKILL);
    369 		} else {
    370 			proc_reparent(q, initproc);
    371 		}
    372 	}
    373 	proclist_unlock_write(s);
    374 
    375 	/*
    376 	 * Free the VM resources we're still holding on to.
    377 	 * We must do this from a valid thread because doing
    378 	 * so may block. This frees vmspace, which we don't
    379 	 * need anymore. The only remaining lwp is the one
    380 	 * we run at this moment, nothing runs in userland
    381 	 * anymore.
    382 	 */
    383 	uvm_proc_exit(p);
    384 
    385 	/*
    386 	 * Give machine-dependent code a chance to free any
    387 	 * MD LWP resources while we can still block. This must be done
    388 	 * before uvm_lwp_exit(), in case these resources are in the
    389 	 * PCB.
    390 	 * THIS IS LAST BLOCKING OPERATION.
    391 	 */
    392 #ifndef __NO_CPU_LWP_FREE
    393 	cpu_lwp_free(l, 1);
    394 #endif
    395 
    396 	pmap_deactivate(l);
    397 
    398 	/*
    399 	 * NOTE: WE ARE NO LONGER ALLOWED TO SLEEP!
    400 	 */
    401 
    402 	/*
    403 	 * Save exit status and final rusage info, adding in child rusage
    404 	 * info and self times.
    405 	 * In order to pick up the time for the current execution, we must
    406 	 * do this before unlinking the lwp from l_list.
    407 	 */
    408 	p->p_xstat = rv;
    409 	*p->p_ru = p->p_stats->p_ru;
    410 	calcru(p, &p->p_ru->ru_utime, &p->p_ru->ru_stime, NULL);
    411 	ruadd(p->p_ru, &p->p_stats->p_cru);
    412 
    413 	/*
    414 	 * Move proc from allproc to zombproc, it's now ready
    415 	 * to be collected by parent. Remaining lwp resources
    416 	 * will be freed in lwp_exit2() once we'd switch to idle
    417 	 * context.
    418 	 * Changing the state to SZOMB stops it being found by pfind().
    419 	 */
    420 	s = proclist_lock_write();
    421 	LIST_REMOVE(p, p_list);
    422 	LIST_INSERT_HEAD(&zombproc, p, p_list);
    423 	p->p_stat = SZOMB;
    424 
    425 	LIST_REMOVE(l, l_list);
    426 	l->l_flag |= L_DETACHED|L_PROCEXIT;	/* detached from proc too */
    427 	l->l_stat = LSDEAD;
    428 
    429 	p->p_nrlwps--;
    430 	p->p_nlwps--;
    431 
    432 	/* Put in front of parent's sibling list for parent to collect it */
    433 	q = p->p_pptr;
    434 	q->p_nstopchild++;
    435 	if (LIST_FIRST(&q->p_children) != p) {
    436 		/* Put child where it can be found quickly */
    437 		LIST_REMOVE(p, p_sibling);
    438 		LIST_INSERT_HEAD(&q->p_children, p, p_sibling);
    439 	}
    440 
    441 	proclist_unlock_write(s);
    442 
    443 	/*
    444 	 * Notify interested parties of our demise.
    445 	 */
    446 	KNOTE(&p->p_klist, NOTE_EXIT);
    447 
    448 #if PERFCTRS
    449 	/*
    450 	 * Save final PMC information in parent process & clean up.
    451 	 */
    452 	if (PMC_ENABLED(p)) {
    453 		pmc_save_context(p);
    454 		pmc_accumulate(p->p_pptr, p);
    455 		pmc_process_exit(p);
    456 	}
    457 #endif
    458 
    459 	/*
    460 	 * Notify parent that we're gone.  If parent has the P_NOCLDWAIT
    461 	 * flag set, notify init instead (and hope it will handle
    462 	 * this situation).
    463 	 */
    464 	if (p->p_pptr->p_flag & P_NOCLDWAIT) {
    465 		struct proc *pp = p->p_pptr;
    466 		proc_reparent(p, initproc);
    467 
    468 		/*
    469 		 * If this was the last child of our parent, notify
    470 		 * parent, so in case he was wait(2)ing, he will
    471 		 * continue.
    472 		 */
    473 		if (LIST_FIRST(&pp->p_children) == NULL)
    474 			wakeup(pp);
    475 	}
    476 
    477 	/* Wake up the parent so it can get exit status. */
    478 	if ((p->p_flag & P_FSTRACE) == 0 && p->p_exitsig != 0)
    479 		exit_psignal(p, p->p_pptr);
    480 	wakeup(p->p_pptr);
    481 
    482 	/*
    483 	 * Release the process's signal state.
    484 	 */
    485 	sigactsfree(p);
    486 
    487 	/*
    488 	 * Clear curlwp after we've done all operations
    489 	 * that could block, and before tearing down the rest
    490 	 * of the process state that might be used from clock, etc.
    491 	 * Also, can't clear curlwp while we're still runnable,
    492 	 * as we're not on a run queue (we are current, just not
    493 	 * a proper proc any longer!).
    494 	 *
    495 	 * Other substructures are freed from wait().
    496 	 */
    497 	curlwp = NULL;
    498 	limfree(p->p_limit);
    499 	pstatsfree(p->p_stats);
    500 	p->p_limit = NULL;
    501 
    502 	/* This process no longer needs to hold the kernel lock. */
    503 	KERNEL_PROC_UNLOCK(l);
    504 
    505 #ifdef DEBUG
    506 	/* Nothing should use the process link anymore */
    507 	l->l_proc = NULL;
    508 #endif
    509 
    510 	/*
    511 	 * Finally, call machine-dependent code to switch to a new
    512 	 * context (possibly the idle context).  Once we are no longer
    513 	 * using the dead lwp's stack, lwp_exit2() will be called
    514 	 * to arrange for the resources to be released.
    515 	 *
    516 	 * Note that cpu_exit() will end with a call equivalent to
    517 	 * cpu_switch(), finishing our execution (pun intended).
    518 	 */
    519 
    520 	uvmexp.swtch++;
    521 	cpu_exit(l);
    522 }
    523 
    524 void
    525 exit_lwps(struct lwp *l)
    526 {
    527 	struct proc *p;
    528 	struct lwp *l2;
    529 	int s, error;
    530 	lwpid_t		waited;
    531 
    532 	p = l->l_proc;
    533 
    534 	/* XXX SMP
    535 	 * This would be the right place to IPI any LWPs running on
    536 	 * other processors so that they can notice the userret exit hook.
    537 	 */
    538 	p->p_userret = lwp_exit_hook;
    539 	p->p_userret_arg = NULL;
    540 
    541 	if (p->p_sa) {
    542 		/*
    543 		 * Make SA-cached LWPs normal process runnable LWPs so
    544 		 * that they'll also self-destruct.
    545 		 */
    546 		DPRINTF(("exit_lwps: Making cached LWPs of %d runnable: ",
    547 		    p->p_pid));
    548 		SCHED_LOCK(s);
    549 		while ((l2 = sa_getcachelwp(p)) != 0) {
    550 			l2->l_priority = l2->l_usrpri;
    551 			setrunnable(l2);
    552 			DPRINTF(("%d ", l2->l_lid));
    553 		}
    554 		DPRINTF(("\n"));
    555 		SCHED_UNLOCK(s);
    556 
    557 		/* Clear wokenq, the LWPs on the queue will run below */
    558 		p->p_sa->sa_wokenq_head = NULL;
    559 	}
    560 
    561 	/*
    562 	 * Interrupt LWPs in interruptable sleep, unsuspend suspended
    563 	 * LWPs, make detached LWPs undetached (so we can wait for
    564 	 * them) and then wait for everyone else to finish.
    565 	 */
    566 	LIST_FOREACH(l2, &p->p_lwps, l_sibling) {
    567 		l2->l_flag &= ~(L_DETACHED|L_SA);
    568 
    569 		if (p->p_sa && l2->l_wchan == p->p_sa)
    570 			l2->l_flag |= L_SINTR;
    571 
    572 		if ((l2->l_stat == LSSLEEP && (l2->l_flag & L_SINTR)) ||
    573 		    l2->l_stat == LSSUSPENDED || l2->l_stat == LSSTOP) {
    574 			SCHED_LOCK(s);
    575 			setrunnable(l2);
    576 			SCHED_UNLOCK(s);
    577 			DPRINTF(("exit_lwps: Made %d.%d runnable\n",
    578 			    p->p_pid, l2->l_lid));
    579 		}
    580 	}
    581 
    582 
    583 	while (p->p_nlwps > 1) {
    584 		DPRINTF(("exit_lwps: waiting for %d LWPs (%d runnable, %d zombies)\n",
    585 		    p->p_nlwps, p->p_nrlwps, p->p_nzlwps));
    586 		error = lwp_wait1(l, 0, &waited, LWPWAIT_EXITCONTROL);
    587 		if (error)
    588 			panic("exit_lwps: lwp_wait1 failed with error %d\n",
    589 			    error);
    590 		DPRINTF(("exit_lwps: Got LWP %d from lwp_wait1()\n", waited));
    591 	}
    592 
    593 	p->p_userret = NULL;
    594 }
    595 
    596 /* Wrapper function for use in p_userret */
    597 static void
    598 lwp_exit_hook(struct lwp *l, void *arg)
    599 {
    600 	KERNEL_PROC_LOCK(l);
    601 	lwp_exit(l);
    602 }
    603 
    604 int
    605 sys_wait4(struct lwp *l, void *v, register_t *retval)
    606 {
    607 	struct sys_wait4_args /* {
    608 		syscallarg(int)			pid;
    609 		syscallarg(int *)		status;
    610 		syscallarg(int)			options;
    611 		syscallarg(struct rusage *)	rusage;
    612 	} */ *uap = v;
    613 	struct proc	*child, *parent;
    614 	int		status, error;
    615 
    616 	parent = l->l_proc;
    617 
    618 	if (SCARG(uap, pid) == 0)
    619 		SCARG(uap, pid) = -parent->p_pgid;
    620 	if (SCARG(uap, options) & ~(WUNTRACED|WNOHANG|WALTSIG|WALLSIG))
    621 		return (EINVAL);
    622 
    623 	error = find_stopped_child(parent, SCARG(uap,pid), SCARG(uap,options),
    624 				&child);
    625 	if (error != 0)
    626 		return error;
    627 	if (child == NULL) {
    628 		*retval = 0;
    629 		return 0;
    630 	}
    631 
    632 	/*
    633 	 * Collect child u-areas.
    634 	 */
    635 	uvm_uarea_drain(FALSE);
    636 
    637 	retval[0] = child->p_pid;
    638 
    639 	if (child->p_stat == SZOMB) {
    640 		if (SCARG(uap, status)) {
    641 			status = child->p_xstat;	/* convert to int */
    642 			error = copyout(&status, SCARG(uap, status),
    643 					sizeof(status));
    644 			if (error)
    645 				return (error);
    646 		}
    647 		if (SCARG(uap, rusage)) {
    648 			error = copyout(child->p_ru, SCARG(uap, rusage),
    649 				    sizeof(struct rusage));
    650 			if (error)
    651 				return (error);
    652 		}
    653 
    654 		proc_free(child);
    655 		return 0;
    656 	}
    657 
    658 	/* child state must be SSTOP */
    659 	if (SCARG(uap, status)) {
    660 		status = W_STOPCODE(child->p_xstat);
    661 		return copyout(&status, SCARG(uap, status), sizeof(status));
    662 	}
    663 	return 0;
    664 }
    665 
    666 /*
    667  * Scan list of child processes for a child process that has stopped or
    668  * exited.  Used by sys_wait4 and 'compat' equivalents.
    669  */
    670 int
    671 find_stopped_child(struct proc *parent, pid_t pid, int options,
    672 	struct proc **child_p)
    673 {
    674 	struct proc *child;
    675 	int error;
    676 
    677 	for (;;) {
    678 		proclist_lock_read();
    679 		error = ECHILD;
    680 		LIST_FOREACH(child, &parent->p_children, p_sibling) {
    681 			if (pid >= 0) {
    682 				if (child->p_pid != pid) {
    683 					child = p_find(pid, PFIND_ZOMBIE |
    684 								PFIND_LOCKED);
    685 					if (child == NULL
    686 					    || child->p_pptr != parent) {
    687 						child = NULL;
    688 						break;
    689 					}
    690 				}
    691 			} else
    692 				if (pid != WAIT_ANY && child->p_pgid != -pid)
    693 					/* child not in correct pgrp */
    694 					continue;
    695 			/*
    696 			 * Wait for processes with p_exitsig != SIGCHLD
    697 			 * processes only if WALTSIG is set; wait for
    698 			 * processes with p_exitsig == SIGCHLD only
    699 			 * if WALTSIG is clear.
    700 			 */
    701 			if (((options & WALLSIG) == 0) &&
    702 			    (options & WALTSIG ? child->p_exitsig == SIGCHLD
    703 						: P_EXITSIG(child) != SIGCHLD)){
    704 				if (child->p_pid == pid) {
    705 					child = NULL;
    706 					break;
    707 				}
    708 				continue;
    709 			}
    710 
    711 			error = 0;
    712 			if (child->p_stat == SZOMB && !(options & WNOZOMBIE))
    713 				break;
    714 
    715 			if (child->p_stat == SSTOP &&
    716 			    (child->p_flag & P_WAITED) == 0 &&
    717 			    (child->p_flag & P_TRACED || options & WUNTRACED)) {
    718 				if ((options & WNOWAIT) == 0) {
    719 					child->p_flag |= P_WAITED;
    720 					parent->p_nstopchild--;
    721 				}
    722 				break;
    723 			}
    724 			if (parent->p_nstopchild == 0 || child->p_pid == pid) {
    725 				child = NULL;
    726 				break;
    727 			}
    728 		}
    729 		proclist_unlock_read();
    730 		if (child != NULL || error != 0 || options & WNOHANG) {
    731 			*child_p = child;
    732 			return error;
    733 		}
    734 		error = tsleep(parent, PWAIT | PCATCH, "wait", 0);
    735 		if (error != 0)
    736 			return error;
    737 	}
    738 }
    739 
    740 /*
    741  * Free a process after parent has taken all the state info.
    742  */
    743 void
    744 proc_free(struct proc *p)
    745 {
    746 	struct proc *parent = p->p_pptr;
    747 	int s;
    748 
    749 	/*
    750 	 * If we got the child via ptrace(2) or procfs, and
    751 	 * the parent is different (meaning the process was
    752 	 * attached, rather than run as a child), then we need
    753 	 * to give it back to the old parent, and send the
    754 	 * parent the exit signal.  The rest of the cleanup
    755 	 * will be done when the old parent waits on the child.
    756 	 */
    757 	if ((p->p_flag & P_TRACED) && p->p_opptr != parent){
    758 		parent = p->p_opptr;
    759 		if (parent == NULL)
    760 			parent = initproc;
    761 		proc_reparent(p, parent);
    762 		p->p_opptr = NULL;
    763 		p->p_flag &= ~(P_TRACED|P_WAITED|P_FSTRACE);
    764 		if (p->p_exitsig != 0)
    765 			exit_psignal(p, parent);
    766 		wakeup(parent);
    767 		return;
    768 	}
    769 
    770 	scheduler_wait_hook(parent, p);
    771 	p->p_xstat = 0;
    772 
    773 	ruadd(&parent->p_stats->p_cru, p->p_ru);
    774 
    775 	/*
    776 	 * At this point we are going to start freeing the final resources.
    777 	 * If anyone tries to access the proc structure after here they
    778 	 * will get a shock - bits are missing.
    779 	 * Attempt to make it hard!
    780 	 */
    781 
    782 	p->p_stat = SIDL;		/* not even a zombie any more */
    783 
    784 	pool_put(&rusage_pool, p->p_ru);
    785 
    786 	/*
    787 	 * Finally finished with old proc entry.
    788 	 * Unlink it from its process group and free it.
    789 	 */
    790 	leavepgrp(p);
    791 
    792 	s = proclist_lock_write();
    793 	LIST_REMOVE(p, p_list);	/* off zombproc */
    794 	p->p_pptr->p_nstopchild--;
    795 	LIST_REMOVE(p, p_sibling);
    796 	proclist_unlock_write(s);
    797 
    798 	/*
    799 	 * Decrement the count of procs running with this uid.
    800 	 */
    801 	(void)chgproccnt(p->p_cred->p_ruid, -1);
    802 
    803 	/*
    804 	 * Free up credentials.
    805 	 */
    806 	if (--p->p_cred->p_refcnt == 0) {
    807 		crfree(p->p_cred->pc_ucred);
    808 		pool_put(&pcred_pool, p->p_cred);
    809 	}
    810 
    811 	/*
    812 	 * Release reference to text vnode
    813 	 */
    814 	if (p->p_textvp)
    815 		vrele(p->p_textvp);
    816 
    817 	/* Release any SA state. */
    818 	if (p->p_sa)
    819 		sa_release(p);
    820 
    821 	/* Free proc structure and let pid be reallocated */
    822 	proc_free_mem(p);
    823 }
    824 
    825 /*
    826  * make process 'parent' the new parent of process 'child'.
    827  *
    828  * Must be called with proclist_lock_write() held.
    829  */
    830 void
    831 proc_reparent(struct proc *child, struct proc *parent)
    832 {
    833 
    834 	if (child->p_pptr == parent)
    835 		return;
    836 
    837 	if (child->p_stat == SZOMB
    838 	    || (child->p_stat == SSTOP && !(child->p_flag & P_WAITED))) {
    839 		child->p_pptr->p_nstopchild--;
    840 		parent->p_nstopchild++;
    841 	}
    842 	if (parent == initproc)
    843 		child->p_exitsig = SIGCHLD;
    844 
    845 	LIST_REMOVE(child, p_sibling);
    846 	LIST_INSERT_HEAD(&parent->p_children, child, p_sibling);
    847 	child->p_pptr = parent;
    848 }
    849