Home | History | Annotate | Line # | Download | only in kern
kern_exit.c revision 1.115
      1 /*	$NetBSD: kern_exit.c,v 1.115 2003/03/19 11:36:33 dsl 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. All advertising materials mentioning features or use of this software
     58  *    must display the following acknowledgement:
     59  *	This product includes software developed by the University of
     60  *	California, Berkeley and its contributors.
     61  * 4. Neither the name of the University nor the names of its contributors
     62  *    may be used to endorse or promote products derived from this software
     63  *    without specific prior written permission.
     64  *
     65  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
     66  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
     67  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
     68  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
     69  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
     70  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
     71  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
     72  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
     73  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
     74  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
     75  * SUCH DAMAGE.
     76  *
     77  *	@(#)kern_exit.c	8.10 (Berkeley) 2/23/95
     78  */
     79 
     80 #include <sys/cdefs.h>
     81 __KERNEL_RCSID(0, "$NetBSD: kern_exit.c,v 1.115 2003/03/19 11:36:33 dsl Exp $");
     82 
     83 #include "opt_ktrace.h"
     84 #include "opt_perfctrs.h"
     85 #include "opt_systrace.h"
     86 #include "opt_sysv.h"
     87 
     88 #include <sys/param.h>
     89 #include <sys/systm.h>
     90 #include <sys/ioctl.h>
     91 #include <sys/proc.h>
     92 #include <sys/tty.h>
     93 #include <sys/time.h>
     94 #include <sys/resource.h>
     95 #include <sys/kernel.h>
     96 #include <sys/ktrace.h>
     97 #include <sys/proc.h>
     98 #include <sys/buf.h>
     99 #include <sys/wait.h>
    100 #include <sys/file.h>
    101 #include <sys/vnode.h>
    102 #include <sys/syslog.h>
    103 #include <sys/malloc.h>
    104 #include <sys/pool.h>
    105 #include <sys/resourcevar.h>
    106 #if defined(PERFCTRS)
    107 #include <sys/pmc.h>
    108 #endif
    109 #include <sys/ptrace.h>
    110 #include <sys/acct.h>
    111 #include <sys/filedesc.h>
    112 #include <sys/ras.h>
    113 #include <sys/signalvar.h>
    114 #include <sys/sched.h>
    115 #include <sys/sa.h>
    116 #include <sys/savar.h>
    117 #include <sys/mount.h>
    118 #include <sys/syscallargs.h>
    119 #include <sys/systrace.h>
    120 
    121 #include <machine/cpu.h>
    122 
    123 #include <uvm/uvm_extern.h>
    124 
    125 #define DEBUG_EXIT
    126 
    127 #ifdef DEBUG_EXIT
    128 int debug_exit = 0;
    129 #define DPRINTF(x) if (debug_exit) printf x
    130 #else
    131 #define DPRINTF(x)
    132 #endif
    133 
    134 static void lwp_exit_hook(struct lwp *, void *);
    135 
    136 /*
    137  * exit --
    138  *	Death of process.
    139  */
    140 int
    141 sys_exit(struct lwp *l, void *v, register_t *retval)
    142 {
    143 	struct sys_exit_args /* {
    144 		syscallarg(int)	rval;
    145 	} */ *uap = v;
    146 
    147 	/* Don't call exit1() multiple times in the same process.*/
    148 	if (l->l_proc->p_flag & P_WEXIT)
    149 		lwp_exit(l);
    150 
    151 	exit1(l, W_EXITCODE(SCARG(uap, rval), 0));
    152 	/* NOTREACHED */
    153 	return (0);
    154 }
    155 
    156 /*
    157  * Exit: deallocate address space and other resources, change proc state
    158  * to zombie, and unlink proc from allproc and parent's lists.  Save exit
    159  * status and rusage for wait().  Check for child processes and orphan them.
    160  */
    161 void
    162 exit1(struct lwp *l, int rv)
    163 {
    164 	struct proc	*p, *q, *nq;
    165 	int		s, sa;
    166 
    167 	p = l->l_proc;
    168 
    169 	if (__predict_false(p == initproc))
    170 		panic("init died (signal %d, exit %d)",
    171 		    WTERMSIG(rv), WEXITSTATUS(rv));
    172 
    173 	p->p_flag |= P_WEXIT;
    174 
    175 	DPRINTF(("exit1: %d.%d exiting.\n", p->p_pid, l->l_lid));
    176 	/*
    177 	 * Disable scheduler activation upcalls.
    178 	 * We're trying to get out of here.
    179 	 */
    180 	sa = 0;
    181 	if (p->p_sa != NULL) {
    182 		l->l_flag &= ~L_SA;
    183 		p->p_flag &= ~P_SA;
    184 		sa = 1;
    185 	}
    186 
    187 #ifdef PGINPROF
    188 	vmsizmon();
    189 #endif
    190 	if (p->p_flag & P_PROFIL)
    191 		stopprofclock(p);
    192 	p->p_ru = pool_get(&rusage_pool, PR_WAITOK);
    193 	/*
    194 	 * If parent is waiting for us to exit or exec, P_PPWAIT is set; we
    195 	 * wake up the parent early to avoid deadlock.
    196 	 */
    197 	if (p->p_flag & P_PPWAIT) {
    198 		p->p_flag &= ~P_PPWAIT;
    199 		wakeup((caddr_t)p->p_pptr);
    200 	}
    201 	sigfillset(&p->p_sigctx.ps_sigignore);
    202 	sigemptyset(&p->p_sigctx.ps_siglist);
    203 	p->p_sigctx.ps_sigcheck = 0;
    204 	timers_free(p, TIMERS_ALL);
    205 
    206 	if (sa || (p->p_nlwps > 1))
    207 		exit_lwps(l);
    208 
    209 #if defined(__HAVE_RAS)
    210 	ras_purgeall(p);
    211 #endif
    212 
    213 	/*
    214 	 * Close open files and release open-file table.
    215 	 * This may block!
    216 	 */
    217 	fdfree(p);
    218 	cwdfree(p);
    219 
    220 	doexithooks(p);
    221 
    222 	if (SESS_LEADER(p)) {
    223 		struct session *sp = p->p_session;
    224 
    225 		if (sp->s_ttyvp) {
    226 			/*
    227 			 * Controlling process.
    228 			 * Signal foreground pgrp,
    229 			 * drain controlling terminal
    230 			 * and revoke access to controlling terminal.
    231 			 */
    232 			if (sp->s_ttyp->t_session == sp) {
    233 				if (sp->s_ttyp->t_pgrp)
    234 					pgsignal(sp->s_ttyp->t_pgrp, SIGHUP, 1);
    235 				(void) ttywait(sp->s_ttyp);
    236 				/*
    237 				 * The tty could have been revoked
    238 				 * if we blocked.
    239 				 */
    240 				if (sp->s_ttyvp)
    241 					VOP_REVOKE(sp->s_ttyvp, REVOKEALL);
    242 			}
    243 			if (sp->s_ttyvp)
    244 				vrele(sp->s_ttyvp);
    245 			sp->s_ttyvp = NULL;
    246 			/*
    247 			 * s_ttyp is not zero'd; we use this to indicate
    248 			 * that the session once had a controlling terminal.
    249 			 * (for logging and informational purposes)
    250 			 */
    251 		}
    252 		sp->s_leader = NULL;
    253 	}
    254 	fixjobc(p, p->p_pgrp, 0);
    255 	(void)acct_process(p);
    256 #ifdef KTRACE
    257 	/*
    258 	 * release trace file
    259 	 */
    260 	ktrderef(p);
    261 #endif
    262 #ifdef SYSTRACE
    263 	systrace_sys_exit(p);
    264 #endif
    265 	/*
    266 	 * If emulation has process exit hook, call it now.
    267 	 */
    268 	if (p->p_emul->e_proc_exit)
    269 		(*p->p_emul->e_proc_exit)(p);
    270 
    271 	/*
    272 	 * Give orphaned children to init(8).
    273 	 */
    274 	q = LIST_FIRST(&p->p_children);
    275 	if (q)		/* only need this if any child is S_ZOMB */
    276 		wakeup((caddr_t)initproc);
    277 	for (; q != 0; q = nq) {
    278 		nq = LIST_NEXT(q, p_sibling);
    279 
    280 		/*
    281 		 * Traced processes are killed since their existence
    282 		 * means someone is screwing up. Since we reset the
    283 		 * trace flags, the logic in sys_wait4() would not be
    284 		 * triggered to reparent the process to its
    285 		 * original parent, so we must do this here.
    286 		 */
    287 		if (q->p_flag & P_TRACED) {
    288 			if (q->p_opptr != q->p_pptr) {
    289 				struct proc *t = q->p_opptr;
    290 				proc_reparent(q, t ? t : initproc);
    291 				q->p_opptr = NULL;
    292 			} else
    293 				proc_reparent(q, initproc);
    294 			q->p_flag &= ~(P_TRACED|P_WAITED|P_FSTRACE);
    295 			psignal(q, SIGKILL);
    296 		} else {
    297 			proc_reparent(q, initproc);
    298 		}
    299 	}
    300 
    301 	/*
    302 	 * Reset p_opptr pointer of all former children which got
    303 	 * traced by another process and were reparented. We reset
    304 	 * it to NULL here; the trace detach code then reparents
    305 	 * the child to initproc. We only check allproc list, since
    306 	 * eventual former children on zombproc list won't reference
    307 	 * p_opptr anymore.
    308 	 */
    309 	if (p->p_flag & P_CHTRACED) {
    310 		struct proc *t;
    311 
    312 		proclist_lock_read();
    313 
    314 		LIST_FOREACH(t, &allproc, p_list) {
    315 			if (t->p_opptr == p)
    316 				t->p_opptr = NULL;
    317 		}
    318 
    319 		proclist_unlock_read();
    320 	}
    321 
    322 	/*
    323 	 * Save exit status and final rusage info, adding in child rusage
    324 	 * info and self times.
    325 	 * In order to pick up the time for the current execution, we must
    326 	 * do this before unlinking the lwp from l_list.
    327 	 */
    328 	p->p_xstat = rv;
    329 	*p->p_ru = p->p_stats->p_ru;
    330 	calcru(p, &p->p_ru->ru_utime, &p->p_ru->ru_stime, NULL);
    331 	ruadd(p->p_ru, &p->p_stats->p_cru);
    332 
    333 	/*
    334 	 * NOTE: WE ARE NO LONGER ALLOWED TO SLEEP!
    335 	 */
    336 
    337 	/*
    338 	 * Move proc from allproc to zombproc, but do not yet
    339 	 * wake up the reaper.  We will put the proc on the
    340 	 * deadproc list later (using the p_dead member), and
    341 	 * wake up the reaper when we do.
    342 	 * Changing the state to SDEAD stops it being found by pfind().
    343 	 */
    344 	s = proclist_lock_write();
    345 	p->p_stat = SDEAD;
    346 	p->p_nrlwps--;
    347 	l->l_stat = SDEAD;
    348 	LIST_REMOVE(p, p_list);
    349 	LIST_INSERT_HEAD(&zombproc, p, p_list);
    350 	LIST_REMOVE(l, l_list);
    351 	l->l_flag |= L_DETACHED;
    352 	proclist_unlock_write(s);
    353 
    354 	/*
    355 	 * Notify interested parties of our demise.
    356 	 */
    357 	KNOTE(&p->p_klist, NOTE_EXIT);
    358 
    359 #if PERFCTRS
    360 	/*
    361 	 * Save final PMC information in parent process & clean up.
    362 	 */
    363 	if (PMC_ENABLED(p)) {
    364 		pmc_save_context(p);
    365 		pmc_accumulate(p->p_pptr, p);
    366 		pmc_process_exit(p);
    367 	}
    368 #endif
    369 
    370 	/*
    371 	 * Notify parent that we're gone.  If parent has the P_NOCLDWAIT
    372 	 * flag set, notify init instead (and hope it will handle
    373 	 * this situation).
    374 	 */
    375 	if (p->p_pptr->p_flag & P_NOCLDWAIT) {
    376 		struct proc *pp = p->p_pptr;
    377 		proc_reparent(p, initproc);
    378 		/*
    379 		 * If this was the last child of our parent, notify
    380 		 * parent, so in case he was wait(2)ing, he will
    381 		 * continue.
    382 		 */
    383 		if (LIST_FIRST(&pp->p_children) == NULL)
    384 			wakeup((caddr_t)pp);
    385 	}
    386 
    387 	/*
    388 	 * Release the process's signal state.
    389 	 */
    390 	sigactsfree(p);
    391 
    392 	/*
    393 	 * Clear curlwp after we've done all operations
    394 	 * that could block, and before tearing down the rest
    395 	 * of the process state that might be used from clock, etc.
    396 	 * Also, can't clear curlwp while we're still runnable,
    397 	 * as we're not on a run queue (we are current, just not
    398 	 * a proper proc any longer!).
    399 	 *
    400 	 * Other substructures are freed from wait().
    401 	 */
    402 	curlwp = NULL;
    403 	limfree(p->p_limit);
    404 	pstatsfree(p->p_stats);
    405 	p->p_limit = NULL;
    406 
    407 	/* This process no longer needs to hold the kernel lock. */
    408 	KERNEL_PROC_UNLOCK(l);
    409 
    410 	/*
    411 	 * Finally, call machine-dependent code to switch to a new
    412 	 * context (possibly the idle context).  Once we are no longer
    413 	 * using the dead process's vmspace and stack, exit2() will be
    414 	 * called to schedule those resources to be released by the
    415 	 * reaper thread.
    416 	 *
    417 	 * Note that cpu_exit() will end with a call equivalent to
    418 	 * cpu_switch(), finishing our execution (pun intended).
    419 	 */
    420 
    421 	cpu_exit(l, 1);
    422 }
    423 
    424 void
    425 exit_lwps(struct lwp *l)
    426 {
    427 	struct proc *p;
    428 	struct lwp *l2;
    429 	int s, error;
    430 	lwpid_t		waited;
    431 
    432 	p = l->l_proc;
    433 
    434 	/* XXX SMP
    435 	 * This would be the right place to IPI any LWPs running on
    436 	 * other processors so that they can notice the userret exit hook.
    437 	 */
    438 	p->p_userret = lwp_exit_hook;
    439 	p->p_userret_arg = NULL;
    440 
    441 	/*
    442 	 * Make SA-cached LWPs normal process runnable LWPs so that
    443 	 * they'll also self-destruct.
    444 	 */
    445 	if (p->p_sa && p->p_sa->sa_ncached > 0) {
    446 		DPRINTF(("exit_lwps: Making cached LWPs of %d runnable: ",
    447 		    p->p_pid));
    448 		SCHED_LOCK(s);
    449 		while ((l2 = sa_getcachelwp(p)) != 0) {
    450 			l2->l_priority = l2->l_usrpri;
    451 			setrunnable(l2);
    452 			DPRINTF(("%d ", l2->l_lid));
    453 		}
    454 		DPRINTF(("\n"));
    455 		SCHED_UNLOCK(s);
    456 	}
    457 
    458 	/*
    459 	 * Interrupt LWPs in interruptable sleep, unsuspend suspended
    460 	 * LWPs, make detached LWPs undetached (so we can wait for
    461 	 * them) and then wait for everyone else to finish.
    462 	 */
    463 	LIST_FOREACH(l2, &p->p_lwps, l_sibling) {
    464 		l2->l_flag &= ~(L_DETACHED|L_SA);
    465 		if ((l2->l_stat == LSSLEEP && (l2->l_flag & L_SINTR)) ||
    466 		    l2->l_stat == LSSUSPENDED) {
    467 			SCHED_LOCK(s);
    468 			setrunnable(l2);
    469 			SCHED_UNLOCK(s);
    470 			DPRINTF(("exit_lwps: Made %d.%d runnable\n",
    471 			    p->p_pid, l2->l_lid));
    472 		}
    473 	}
    474 
    475 
    476 	while (p->p_nlwps > 1) {
    477 		DPRINTF(("exit_lwps: waiting for %d LWPs (%d runnable, %d zombies)\n",
    478 		    p->p_nlwps, p->p_nrlwps, p->p_nzlwps));
    479 		error = lwp_wait1(l, 0, &waited, LWPWAIT_EXITCONTROL);
    480 		if (error)
    481 			panic("exit_lwps: lwp_wait1 failed with error %d\n",
    482 			    error);
    483 		DPRINTF(("exit_lwps: Got LWP %d from lwp_wait1()\n", waited));
    484 	}
    485 
    486 	p->p_userret = NULL;
    487 }
    488 
    489 /* Wrapper function for use in p_userret */
    490 static void
    491 lwp_exit_hook(struct lwp *l, void *arg)
    492 {
    493 	KERNEL_PROC_LOCK(l);
    494 	lwp_exit(l);
    495 }
    496 
    497 /*
    498  * We are called from cpu_exit() once it is safe to schedule the
    499  * dead process's resources to be freed (i.e., once we've switched to
    500  * the idle PCB for the current CPU).
    501  *
    502  * NOTE: One must be careful with locking in this routine.  It's
    503  * called from a critical section in machine-dependent code, so
    504  * we should refrain from changing any interrupt state.
    505  *
    506  * We lock the deadproc list (a spin lock), place the proc on that
    507  * list (using the p_dead member), and wake up the reaper.
    508  */
    509 void
    510 exit2(struct lwp *l)
    511 {
    512 	struct proc *p = l->l_proc;
    513 
    514 	simple_lock(&deadproc_slock);
    515 	SLIST_INSERT_HEAD(&deadprocs, p, p_dead);
    516 	simple_unlock(&deadproc_slock);
    517 
    518 	/* lwp_exit2() will wake up deadproc for us. */
    519 	lwp_exit2(l);
    520 }
    521 
    522 /*
    523  * Process reaper.  This is run by a kernel thread to free the resources
    524  * of a dead process.  Once the resources are free, the process becomes
    525  * a zombie, and the parent is allowed to read the undead's status.
    526  */
    527 void
    528 reaper(void *arg)
    529 {
    530 	struct proc *p;
    531 	struct lwp *l;
    532 
    533 	KERNEL_PROC_UNLOCK(curlwp);
    534 
    535 	for (;;) {
    536 		simple_lock(&deadproc_slock);
    537 		p = SLIST_FIRST(&deadprocs);
    538 		l = LIST_FIRST(&deadlwp);
    539 		if (p == NULL && l == NULL) {
    540 			/* No work for us; go to sleep until someone exits. */
    541 			(void) ltsleep(&deadprocs, PVM|PNORELOCK,
    542 			    "reaper", 0, &deadproc_slock);
    543 			continue;
    544 		}
    545 
    546 		if (l != NULL ) {
    547 			p = l->l_proc;
    548 
    549 			/* Remove lwp from the deadlwp list. */
    550 			LIST_REMOVE(l, l_list);
    551 			simple_unlock(&deadproc_slock);
    552 			KERNEL_PROC_LOCK(curlwp);
    553 
    554 			/*
    555 			 * Give machine-dependent code a chance to free any
    556 			 * resources it couldn't free while still running on
    557 			 * that process's context.  This must be done before
    558 			 * uvm_lwp_exit(), in case these resources are in the
    559 			 * PCB.
    560 			 */
    561 			cpu_wait(l);
    562 
    563 			/*
    564 			 * Free the VM resources we're still holding on to.
    565 			 */
    566 			uvm_lwp_exit(l);
    567 
    568 			l->l_stat = LSZOMB;
    569 			if (l->l_flag & L_DETACHED) {
    570 				/* Nobody waits for detached LWPs. */
    571 				LIST_REMOVE(l, l_sibling);
    572 				p->p_nlwps--;
    573 				pool_put(&lwp_pool, l);
    574 			} else {
    575 				p->p_nzlwps++;
    576 				wakeup((caddr_t)&p->p_nlwps);
    577 			}
    578 			/* XXXNJW where should this be with respect to
    579 			 * the wakeup() above? */
    580 			KERNEL_PROC_UNLOCK(curlwp);
    581 		} else {
    582 			/* Remove proc from the deadproc list. */
    583 			SLIST_REMOVE_HEAD(&deadprocs, p_dead);
    584 			simple_unlock(&deadproc_slock);
    585 			KERNEL_PROC_LOCK(curlwp);
    586 
    587 			/*
    588 			 * Free the VM resources we're still holding on to.
    589 			 * We must do this from a valid thread because doing
    590 			 * so may block.
    591 			 */
    592 			uvm_proc_exit(p);
    593 
    594 			/* Process is now a true zombie. */
    595 			p->p_stat = SZOMB;
    596 
    597 			/* Wake up the parent so it can get exit status. */
    598 			if ((p->p_flag & P_FSTRACE) == 0 && p->p_exitsig != 0)
    599 				psignal(p->p_pptr, P_EXITSIG(p));
    600 			KERNEL_PROC_UNLOCK(curlwp);
    601 			wakeup((caddr_t)p->p_pptr);
    602 		}
    603 	}
    604 }
    605 
    606 int
    607 sys_wait4(struct lwp *l, void *v, register_t *retval)
    608 {
    609 	struct sys_wait4_args /* {
    610 		syscallarg(int)			pid;
    611 		syscallarg(int *)		status;
    612 		syscallarg(int)			options;
    613 		syscallarg(struct rusage *)	rusage;
    614 	} */ *uap = v;
    615 	struct proc	*child, *parent;
    616 	int		status, error;
    617 
    618 	parent = l->l_proc;
    619 
    620 	if (SCARG(uap, pid) == 0)
    621 		SCARG(uap, pid) = -parent->p_pgid;
    622 	if (SCARG(uap, options) & ~(WUNTRACED|WNOHANG|WALTSIG|WALLSIG))
    623 		return (EINVAL);
    624 
    625 	error = find_stopped_child(parent, SCARG(uap,pid), SCARG(uap,options),
    626 				&child);
    627 	if (error != 0)
    628 		return error;
    629 	if (child == NULL) {
    630 		*retval = 0;
    631 		return 0;
    632 	}
    633 
    634 	retval[0] = child->p_pid;
    635 
    636 	if (child->p_stat == SZOMB) {
    637 		if (SCARG(uap, status)) {
    638 			status = child->p_xstat;	/* convert to int */
    639 			error = copyout((caddr_t)&status,
    640 					(caddr_t)SCARG(uap, status),
    641 					sizeof(status));
    642 			if (error)
    643 				return (error);
    644 		}
    645 		if (SCARG(uap, rusage)) {
    646 			error = copyout((caddr_t)child->p_ru,
    647 				    (caddr_t)SCARG(uap, rusage),
    648 				    sizeof(struct rusage));
    649 			if (error)
    650 				return (error);
    651 		}
    652 
    653 		proc_free(child);
    654 		return 0;
    655 	}
    656 
    657 	/* child state must be SSTOP */
    658 	if (SCARG(uap, status)) {
    659 		status = W_STOPCODE(child->p_xstat);
    660 		return copyout((caddr_t)&status,
    661 				(caddr_t)SCARG(uap, status),
    662 				sizeof(status));
    663 	}
    664 	return 0;
    665 }
    666 
    667 /*
    668  * Scan list of child processes for a child process that has stopped or
    669  * exited.  Used by sys_wait4 and 'compat' equivalents.
    670  */
    671 int
    672 find_stopped_child(struct proc *parent, pid_t pid, int options,
    673 	struct proc **child_p)
    674 {
    675 	struct proc *child;
    676 	int c_found, error;
    677 
    678 	 for (;;) {
    679 		c_found = 0;
    680 		LIST_FOREACH(child, &parent->p_children, p_sibling) {
    681 			if (pid != WAIT_ANY &&
    682 			    child->p_pid != pid &&
    683 			    child->p_pgid != -pid)
    684 				continue;
    685 			/*
    686 			 * Wait for processes with p_exitsig != SIGCHLD
    687 			 * processes only if WALTSIG is set; wait for
    688 			 * processes with p_exitsig == SIGCHLD only
    689 			 * if WALTSIG is clear.
    690 			 */
    691 			if (((options & WALLSIG) == 0) &&
    692 			    (options & WALTSIG ? child->p_exitsig == SIGCHLD
    693 						: P_EXITSIG(child) != SIGCHLD))
    694 				continue;
    695 
    696 			c_found = 1;
    697 			if (child->p_stat == SZOMB &&
    698 			    (options & WNOZOMBIE) == 0) {
    699 				*child_p = child;
    700 				return 0;
    701 			}
    702 
    703 			if (child->p_stat == SSTOP &&
    704 			    (child->p_flag & P_WAITED) == 0 &&
    705 			    (child->p_flag & P_TRACED || options & WUNTRACED)) {
    706 				if ((options & WNOWAIT) == 0)
    707 					child->p_flag |= P_WAITED;
    708 				*child_p = child;
    709 				return 0;
    710 			}
    711 		}
    712 		if (c_found == 0)
    713 			return ECHILD;
    714 		if (options & WNOHANG) {
    715 			*child_p = NULL;
    716 			return 0;
    717 		}
    718 		error = tsleep((caddr_t)parent, PWAIT | PCATCH, "wait", 0);
    719 		if (error != 0)
    720 			return error;
    721 	}
    722 }
    723 
    724 /*
    725  * Free a process after parent has taken all the state info.
    726  */
    727 void
    728 proc_free(struct proc *p)
    729 {
    730 	struct proc *parent = p->p_pptr;
    731 	int s;
    732 
    733 	/*
    734 	 * If we got the child via ptrace(2) or procfs, and
    735 	 * the parent is different (meaning the process was
    736 	 * attached, rather than run as a child), then we need
    737 	 * to give it back to the old parent, and send the
    738 	 * parent the exit signal.  The rest of the cleanup
    739 	 * will be done when the old parent waits on the child.
    740 	 */
    741 	if ((p->p_flag & P_TRACED) && p->p_opptr != parent){
    742 		parent = p->p_opptr;
    743 		if (parent == NULL)
    744 			parent = initproc;
    745 		proc_reparent(p, parent);
    746 		p->p_opptr = NULL;
    747 		p->p_flag &= ~(P_TRACED|P_WAITED|P_FSTRACE);
    748 		if (p->p_exitsig != 0)
    749 			psignal(parent, P_EXITSIG(p));
    750 		wakeup((caddr_t)parent);
    751 		return;
    752 	}
    753 
    754 	scheduler_wait_hook(parent, p);
    755 	p->p_xstat = 0;
    756 
    757 	ruadd(&parent->p_stats->p_cru, p->p_ru);
    758 
    759 	/*
    760 	 * At this point we are going to start freeing the final resources.
    761 	 * If anyone tries to access the proc structure after here they
    762 	 * will get a shock - bits are missing.
    763 	 * Attempt to make it hard!
    764 	 */
    765 
    766 	p->p_stat = SIDL;		/* not even a zombie any more */
    767 
    768 	pool_put(&rusage_pool, p->p_ru);
    769 
    770 	/*
    771 	 * Finally finished with old proc entry.
    772 	 * Unlink it from its process group and free it.
    773 	 */
    774 	leavepgrp(p);
    775 
    776 	s = proclist_lock_write();
    777 	LIST_REMOVE(p, p_list);	/* off zombproc */
    778 	LIST_REMOVE(p, p_sibling);
    779 	proclist_unlock_write(s);
    780 
    781 	/*
    782 	 * Decrement the count of procs running with this uid.
    783 	 */
    784 	(void)chgproccnt(p->p_cred->p_ruid, -1);
    785 
    786 	/*
    787 	 * Free up credentials.
    788 	 */
    789 	if (--p->p_cred->p_refcnt == 0) {
    790 		crfree(p->p_cred->pc_ucred);
    791 		pool_put(&pcred_pool, p->p_cred);
    792 	}
    793 
    794 	/*
    795 	 * Release reference to text vnode
    796 	 */
    797 	if (p->p_textvp)
    798 		vrele(p->p_textvp);
    799 
    800 	/*
    801 	 * Release any SA state
    802 	 */
    803 	if (p->p_sa) {
    804 		free(p->p_sa->sa_stacks, M_SA);
    805 		pool_put(&sadata_pool, p->p_sa);
    806 	}
    807 
    808 	/* Free proc structure and let pid be reallocated */
    809 	proc_free_mem(p);
    810 }
    811 
    812 /*
    813  * make process 'parent' the new parent of process 'child'.
    814  */
    815 void
    816 proc_reparent(struct proc *child, struct proc *parent)
    817 {
    818 
    819 	if (child->p_pptr == parent)
    820 		return;
    821 
    822 	if (parent == initproc)
    823 		child->p_exitsig = SIGCHLD;
    824 
    825 	LIST_REMOVE(child, p_sibling);
    826 	LIST_INSERT_HEAD(&parent->p_children, child, p_sibling);
    827 	child->p_pptr = parent;
    828 }
    829