Home | History | Annotate | Line # | Download | only in kern
kern_exit.c revision 1.120
      1 /*	$NetBSD: kern_exit.c,v 1.120 2003/08/20 13:50:53 yamt 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 1.120 2003/08/20 13:50:53 yamt 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  * exit --
    134  *	Death of process.
    135  */
    136 int
    137 sys_exit(struct lwp *l, void *v, register_t *retval)
    138 {
    139 	struct sys_exit_args /* {
    140 		syscallarg(int)	rval;
    141 	} */ *uap = v;
    142 
    143 	/* Don't call exit1() multiple times in the same process.*/
    144 	if (l->l_proc->p_flag & P_WEXIT)
    145 		lwp_exit(l);
    146 
    147 	exit1(l, W_EXITCODE(SCARG(uap, rval), 0));
    148 	/* NOTREACHED */
    149 	return (0);
    150 }
    151 
    152 /*
    153  * Exit: deallocate address space and other resources, change proc state
    154  * to zombie, and unlink proc from allproc and parent's lists.  Save exit
    155  * status and rusage for wait().  Check for child processes and orphan them.
    156  */
    157 void
    158 exit1(struct lwp *l, int rv)
    159 {
    160 	struct proc	*p, *q, *nq;
    161 	int		s, sa;
    162 
    163 	p = l->l_proc;
    164 
    165 	if (__predict_false(p == initproc))
    166 		panic("init died (signal %d, exit %d)",
    167 		    WTERMSIG(rv), WEXITSTATUS(rv));
    168 
    169 	p->p_flag |= P_WEXIT;
    170 
    171 	DPRINTF(("exit1: %d.%d exiting.\n", p->p_pid, l->l_lid));
    172 	/*
    173 	 * Disable scheduler activation upcalls.
    174 	 * We're trying to get out of here.
    175 	 */
    176 	sa = 0;
    177 	if (p->p_sa != NULL) {
    178 
    179 		l->l_flag &= ~L_SA;
    180 #if 0
    181 		p->p_flag &= ~P_SA;
    182 #endif
    183 		sa = 1;
    184 	}
    185 
    186 #ifdef PGINPROF
    187 	vmsizmon();
    188 #endif
    189 	if (p->p_flag & P_PROFIL)
    190 		stopprofclock(p);
    191 	p->p_ru = pool_get(&rusage_pool, PR_WAITOK);
    192 	/*
    193 	 * If parent is waiting for us to exit or exec, P_PPWAIT is set; we
    194 	 * wake up the parent early to avoid deadlock.
    195 	 */
    196 	if (p->p_flag & P_PPWAIT) {
    197 		p->p_flag &= ~P_PPWAIT;
    198 		wakeup((caddr_t)p->p_pptr);
    199 	}
    200 	sigfillset(&p->p_sigctx.ps_sigignore);
    201 	sigemptyset(&p->p_sigctx.ps_siglist);
    202 	p->p_sigctx.ps_sigcheck = 0;
    203 	timers_free(p, TIMERS_ALL);
    204 
    205 	if (sa || (p->p_nlwps > 1))
    206 		exit_lwps(l);
    207 
    208 #if defined(__HAVE_RAS)
    209 	ras_purgeall(p);
    210 #endif
    211 
    212 	/*
    213 	 * Close open files and release open-file table.
    214 	 * This may block!
    215 	 */
    216 	fdfree(p);
    217 	cwdfree(p);
    218 
    219 	doexithooks(p);
    220 
    221 	if (SESS_LEADER(p)) {
    222 		struct session *sp = p->p_session;
    223 
    224 		if (sp->s_ttyvp) {
    225 			/*
    226 			 * Controlling process.
    227 			 * Signal foreground pgrp,
    228 			 * drain controlling terminal
    229 			 * and revoke access to controlling terminal.
    230 			 */
    231 			if (sp->s_ttyp->t_session == sp) {
    232 				if (sp->s_ttyp->t_pgrp)
    233 					pgsignal(sp->s_ttyp->t_pgrp, SIGHUP, 1);
    234 				(void) ttywait(sp->s_ttyp);
    235 				/*
    236 				 * The tty could have been revoked
    237 				 * if we blocked.
    238 				 */
    239 				if (sp->s_ttyvp)
    240 					VOP_REVOKE(sp->s_ttyvp, REVOKEALL);
    241 			}
    242 			if (sp->s_ttyvp)
    243 				vrele(sp->s_ttyvp);
    244 			sp->s_ttyvp = NULL;
    245 			/*
    246 			 * s_ttyp is not zero'd; we use this to indicate
    247 			 * that the session once had a controlling terminal.
    248 			 * (for logging and informational purposes)
    249 			 */
    250 		}
    251 		sp->s_leader = NULL;
    252 	}
    253 	fixjobc(p, p->p_pgrp, 0);
    254 	(void)acct_process(p);
    255 #ifdef KTRACE
    256 	/*
    257 	 * release trace file
    258 	 */
    259 	ktrderef(p);
    260 #endif
    261 #ifdef SYSTRACE
    262 	systrace_sys_exit(p);
    263 #endif
    264 	/*
    265 	 * If emulation has process exit hook, call it now.
    266 	 */
    267 	if (p->p_emul->e_proc_exit)
    268 		(*p->p_emul->e_proc_exit)(p);
    269 
    270 	/*
    271 	 * Give orphaned children to init(8).
    272 	 */
    273 	q = LIST_FIRST(&p->p_children);
    274 	if (q)		/* only need this if any child is S_ZOMB */
    275 		wakeup((caddr_t)initproc);
    276 	for (; q != 0; q = nq) {
    277 		nq = LIST_NEXT(q, p_sibling);
    278 
    279 		/*
    280 		 * Traced processes are killed since their existence
    281 		 * means someone is screwing up. Since we reset the
    282 		 * trace flags, the logic in sys_wait4() would not be
    283 		 * triggered to reparent the process to its
    284 		 * original parent, so we must do this here.
    285 		 */
    286 		if (q->p_flag & P_TRACED) {
    287 			if (q->p_opptr != q->p_pptr) {
    288 				struct proc *t = q->p_opptr;
    289 				proc_reparent(q, t ? t : initproc);
    290 				q->p_opptr = NULL;
    291 			} else
    292 				proc_reparent(q, initproc);
    293 			q->p_flag &= ~(P_TRACED|P_WAITED|P_FSTRACE);
    294 			psignal(q, SIGKILL);
    295 		} else {
    296 			proc_reparent(q, initproc);
    297 		}
    298 	}
    299 
    300 	/*
    301 	 * Reset p_opptr pointer of all former children which got
    302 	 * traced by another process and were reparented. We reset
    303 	 * it to NULL here; the trace detach code then reparents
    304 	 * the child to initproc. We only check allproc list, since
    305 	 * eventual former children on zombproc list won't reference
    306 	 * p_opptr anymore.
    307 	 */
    308 	if (p->p_flag & P_CHTRACED) {
    309 		struct proc *t;
    310 
    311 		proclist_lock_read();
    312 
    313 		LIST_FOREACH(t, &allproc, p_list) {
    314 			if (t->p_opptr == p)
    315 				t->p_opptr = NULL;
    316 		}
    317 
    318 		proclist_unlock_read();
    319 	}
    320 
    321 	/*
    322 	 * Save exit status and final rusage info, adding in child rusage
    323 	 * info and self times.
    324 	 * In order to pick up the time for the current execution, we must
    325 	 * do this before unlinking the lwp from l_list.
    326 	 */
    327 	p->p_xstat = rv;
    328 	*p->p_ru = p->p_stats->p_ru;
    329 	calcru(p, &p->p_ru->ru_utime, &p->p_ru->ru_stime, NULL);
    330 	ruadd(p->p_ru, &p->p_stats->p_cru);
    331 
    332 	/*
    333 	 * NOTE: WE ARE NO LONGER ALLOWED TO SLEEP!
    334 	 */
    335 
    336 	/*
    337 	 * Move proc from allproc to zombproc, but do not yet
    338 	 * wake up the reaper.  We will put the proc on the
    339 	 * deadproc list later (using the p_dead member), and
    340 	 * wake up the reaper when we do.
    341 	 * Changing the state to SDEAD stops it being found by pfind().
    342 	 */
    343 	s = proclist_lock_write();
    344 	p->p_stat = SDEAD;
    345 	p->p_nrlwps--;
    346 	l->l_stat = SDEAD;
    347 	LIST_REMOVE(p, p_list);
    348 	LIST_INSERT_HEAD(&zombproc, p, p_list);
    349 	LIST_REMOVE(l, l_list);
    350 	l->l_flag |= L_DETACHED;
    351 	proclist_unlock_write(s);
    352 
    353 	/*
    354 	 * Notify interested parties of our demise.
    355 	 */
    356 	KNOTE(&p->p_klist, NOTE_EXIT);
    357 
    358 #if PERFCTRS
    359 	/*
    360 	 * Save final PMC information in parent process & clean up.
    361 	 */
    362 	if (PMC_ENABLED(p)) {
    363 		pmc_save_context(p);
    364 		pmc_accumulate(p->p_pptr, p);
    365 		pmc_process_exit(p);
    366 	}
    367 #endif
    368 
    369 	/*
    370 	 * Notify parent that we're gone.  If parent has the P_NOCLDWAIT
    371 	 * flag set, notify init instead (and hope it will handle
    372 	 * this situation).
    373 	 */
    374 	if (p->p_pptr->p_flag & P_NOCLDWAIT) {
    375 		struct proc *pp = p->p_pptr;
    376 		proc_reparent(p, initproc);
    377 		/*
    378 		 * If this was the last child of our parent, notify
    379 		 * parent, so in case he was wait(2)ing, he will
    380 		 * continue.
    381 		 */
    382 		if (LIST_FIRST(&pp->p_children) == NULL)
    383 			wakeup((caddr_t)pp);
    384 	}
    385 
    386 	/*
    387 	 * Release the process's signal state.
    388 	 */
    389 	sigactsfree(p);
    390 
    391 	/*
    392 	 * Clear curlwp after we've done all operations
    393 	 * that could block, and before tearing down the rest
    394 	 * of the process state that might be used from clock, etc.
    395 	 * Also, can't clear curlwp while we're still runnable,
    396 	 * as we're not on a run queue (we are current, just not
    397 	 * a proper proc any longer!).
    398 	 *
    399 	 * Other substructures are freed from wait().
    400 	 */
    401 	curlwp = NULL;
    402 	limfree(p->p_limit);
    403 	pstatsfree(p->p_stats);
    404 	p->p_limit = NULL;
    405 
    406 	/* This process no longer needs to hold the kernel lock. */
    407 	KERNEL_PROC_UNLOCK(l);
    408 
    409 	/*
    410 	 * Finally, call machine-dependent code to switch to a new
    411 	 * context (possibly the idle context).  Once we are no longer
    412 	 * using the dead process's vmspace and stack, exit2() will be
    413 	 * called to schedule those resources to be released by the
    414 	 * reaper thread.
    415 	 *
    416 	 * Note that cpu_exit() will end with a call equivalent to
    417 	 * cpu_switch(), finishing our execution (pun intended).
    418 	 */
    419 
    420 	cpu_exit(l, 1);
    421 }
    422 
    423 void
    424 exit_lwps(struct lwp *l)
    425 {
    426 	struct proc *p;
    427 	struct lwp *l2;
    428 	int s, error;
    429 	lwpid_t		waited;
    430 
    431 	p = l->l_proc;
    432 
    433 	/* XXX SMP
    434 	 * This would be the right place to IPI any LWPs running on
    435 	 * other processors so that they can notice the userret exit hook.
    436 	 */
    437 	p->p_userret = lwp_exit_hook;
    438 	p->p_userret_arg = NULL;
    439 
    440 	/*
    441 	 * Make SA-cached LWPs normal process runnable LWPs so that
    442 	 * they'll also self-destruct.
    443 	 */
    444 	if (p->p_sa && p->p_sa->sa_ncached > 0) {
    445 		DPRINTF(("exit_lwps: Making cached LWPs of %d runnable: ",
    446 		    p->p_pid));
    447 		SCHED_LOCK(s);
    448 		while ((l2 = sa_getcachelwp(p)) != 0) {
    449 			l2->l_priority = l2->l_usrpri;
    450 			setrunnable(l2);
    451 			DPRINTF(("%d ", l2->l_lid));
    452 		}
    453 		DPRINTF(("\n"));
    454 		SCHED_UNLOCK(s);
    455 	}
    456 
    457 	/*
    458 	 * Interrupt LWPs in interruptable sleep, unsuspend suspended
    459 	 * LWPs, make detached LWPs undetached (so we can wait for
    460 	 * them) and then wait for everyone else to finish.
    461 	 */
    462 	LIST_FOREACH(l2, &p->p_lwps, l_sibling) {
    463 #if 0
    464 		l2->l_flag &= ~(L_DETACHED|L_SA);
    465 #endif
    466 		l2->l_flag &= ~(L_DETACHED);
    467 
    468 		if(l2->l_flag & L_SA_WANTS_VP)
    469 		{
    470 			wakeup(l2);
    471 		}
    472 
    473 		if ((l2->l_stat == LSSLEEP && (l2->l_flag & L_SINTR)) ||
    474 		    l2->l_stat == LSSUSPENDED || l2->l_stat == LSSTOP) {
    475 			SCHED_LOCK(s);
    476 			setrunnable(l2);
    477 			SCHED_UNLOCK(s);
    478 			DPRINTF(("exit_lwps: Made %d.%d runnable\n",
    479 			    p->p_pid, l2->l_lid));
    480 		}
    481 	}
    482 
    483 
    484 	while (p->p_nlwps > 1) {
    485 		DPRINTF(("exit_lwps: waiting for %d LWPs (%d runnable, %d zombies)\n",
    486 		    p->p_nlwps, p->p_nrlwps, p->p_nzlwps));
    487 		error = lwp_wait1(l, 0, &waited, LWPWAIT_EXITCONTROL);
    488 		if (error)
    489 			panic("exit_lwps: lwp_wait1 failed with error %d\n",
    490 			    error);
    491 		DPRINTF(("exit_lwps: Got LWP %d from lwp_wait1()\n", waited));
    492 	}
    493 
    494 	p->p_userret = NULL;
    495 }
    496 
    497 /* Wrapper function for use in p_userret */
    498 static void
    499 lwp_exit_hook(struct lwp *l, void *arg)
    500 {
    501 	KERNEL_PROC_LOCK(l);
    502 	lwp_exit(l);
    503 }
    504 
    505 /*
    506  * We are called from cpu_exit() once it is safe to schedule the
    507  * dead process's resources to be freed (i.e., once we've switched to
    508  * the idle PCB for the current CPU).
    509  *
    510  * NOTE: One must be careful with locking in this routine.  It's
    511  * called from a critical section in machine-dependent code, so
    512  * we should refrain from changing any interrupt state.
    513  *
    514  * We lock the deadproc list (a spin lock), place the proc on that
    515  * list (using the p_dead member), and wake up the reaper.
    516  */
    517 void
    518 exit2(struct lwp *l)
    519 {
    520 	struct proc *p = l->l_proc;
    521 
    522 	simple_lock(&deadproc_slock);
    523 	SLIST_INSERT_HEAD(&deadprocs, p, p_dead);
    524 	simple_unlock(&deadproc_slock);
    525 
    526 	/* lwp_exit2() will wake up deadproc for us. */
    527 	lwp_exit2(l);
    528 }
    529 
    530 /*
    531  * Process reaper.  This is run by a kernel thread to free the resources
    532  * of a dead process.  Once the resources are free, the process becomes
    533  * a zombie, and the parent is allowed to read the undead's status.
    534  */
    535 void
    536 reaper(void *arg)
    537 {
    538 	struct proc *p;
    539 	struct lwp *l;
    540 
    541 	KERNEL_PROC_UNLOCK(curlwp);
    542 
    543 	for (;;) {
    544 		simple_lock(&deadproc_slock);
    545 		p = SLIST_FIRST(&deadprocs);
    546 		l = LIST_FIRST(&deadlwp);
    547 		if (p == NULL && l == NULL) {
    548 			/* No work for us; go to sleep until someone exits. */
    549 			(void) ltsleep(&deadprocs, PVM|PNORELOCK,
    550 			    "reaper", 0, &deadproc_slock);
    551 			continue;
    552 		}
    553 
    554 		if (l != NULL ) {
    555 			p = l->l_proc;
    556 
    557 			/* Remove lwp from the deadlwp list. */
    558 			LIST_REMOVE(l, l_list);
    559 			simple_unlock(&deadproc_slock);
    560 			KERNEL_PROC_LOCK(curlwp);
    561 
    562 			/*
    563 			 * Give machine-dependent code a chance to free any
    564 			 * resources it couldn't free while still running on
    565 			 * that process's context.  This must be done before
    566 			 * uvm_lwp_exit(), in case these resources are in the
    567 			 * PCB.
    568 			 */
    569 			cpu_wait(l);
    570 
    571 			/*
    572 			 * Free the VM resources we're still holding on to.
    573 			 */
    574 			uvm_lwp_exit(l);
    575 
    576 			l->l_stat = LSZOMB;
    577 			if (l->l_flag & L_DETACHED) {
    578 				/* Nobody waits for detached LWPs. */
    579 				LIST_REMOVE(l, l_sibling);
    580 				p->p_nlwps--;
    581 				pool_put(&lwp_pool, l);
    582 			} else {
    583 				p->p_nzlwps++;
    584 				wakeup((caddr_t)&p->p_nlwps);
    585 			}
    586 			/* XXXNJW where should this be with respect to
    587 			 * the wakeup() above? */
    588 			KERNEL_PROC_UNLOCK(curlwp);
    589 		} else {
    590 			/* Remove proc from the deadproc list. */
    591 			SLIST_REMOVE_HEAD(&deadprocs, p_dead);
    592 			simple_unlock(&deadproc_slock);
    593 			KERNEL_PROC_LOCK(curlwp);
    594 
    595 			/*
    596 			 * Free the VM resources we're still holding on to.
    597 			 * We must do this from a valid thread because doing
    598 			 * so may block.
    599 			 */
    600 			uvm_proc_exit(p);
    601 
    602 			/* Process is now a true zombie. */
    603 			p->p_stat = SZOMB;
    604 
    605 			/* Wake up the parent so it can get exit status. */
    606 			if ((p->p_flag & P_FSTRACE) == 0 && p->p_exitsig != 0)
    607 				psignal(p->p_pptr, P_EXITSIG(p));
    608 			KERNEL_PROC_UNLOCK(curlwp);
    609 			wakeup((caddr_t)p->p_pptr);
    610 		}
    611 	}
    612 }
    613 
    614 int
    615 sys_wait4(struct lwp *l, void *v, register_t *retval)
    616 {
    617 	struct sys_wait4_args /* {
    618 		syscallarg(int)			pid;
    619 		syscallarg(int *)		status;
    620 		syscallarg(int)			options;
    621 		syscallarg(struct rusage *)	rusage;
    622 	} */ *uap = v;
    623 	struct proc	*child, *parent;
    624 	int		status, error;
    625 
    626 	parent = l->l_proc;
    627 
    628 	if (SCARG(uap, pid) == 0)
    629 		SCARG(uap, pid) = -parent->p_pgid;
    630 	if (SCARG(uap, options) & ~(WUNTRACED|WNOHANG|WALTSIG|WALLSIG))
    631 		return (EINVAL);
    632 
    633 	error = find_stopped_child(parent, SCARG(uap,pid), SCARG(uap,options),
    634 				&child);
    635 	if (error != 0)
    636 		return error;
    637 	if (child == NULL) {
    638 		*retval = 0;
    639 		return 0;
    640 	}
    641 
    642 	retval[0] = child->p_pid;
    643 
    644 	if (child->p_stat == SZOMB) {
    645 		if (SCARG(uap, status)) {
    646 			status = child->p_xstat;	/* convert to int */
    647 			error = copyout((caddr_t)&status,
    648 					(caddr_t)SCARG(uap, status),
    649 					sizeof(status));
    650 			if (error)
    651 				return (error);
    652 		}
    653 		if (SCARG(uap, rusage)) {
    654 			error = copyout((caddr_t)child->p_ru,
    655 				    (caddr_t)SCARG(uap, rusage),
    656 				    sizeof(struct rusage));
    657 			if (error)
    658 				return (error);
    659 		}
    660 
    661 		proc_free(child);
    662 		return 0;
    663 	}
    664 
    665 	/* child state must be SSTOP */
    666 	if (SCARG(uap, status)) {
    667 		status = W_STOPCODE(child->p_xstat);
    668 		return copyout((caddr_t)&status,
    669 				(caddr_t)SCARG(uap, status),
    670 				sizeof(status));
    671 	}
    672 	return 0;
    673 }
    674 
    675 /*
    676  * Scan list of child processes for a child process that has stopped or
    677  * exited.  Used by sys_wait4 and 'compat' equivalents.
    678  */
    679 int
    680 find_stopped_child(struct proc *parent, pid_t pid, int options,
    681 	struct proc **child_p)
    682 {
    683 	struct proc *child;
    684 	int c_found, error;
    685 
    686 	for (;;) {
    687 		c_found = 0;
    688 		LIST_FOREACH(child, &parent->p_children, p_sibling) {
    689 			if (pid != WAIT_ANY &&
    690 			    child->p_pid != pid &&
    691 			    child->p_pgid != -pid)
    692 				continue;
    693 			/*
    694 			 * Wait for processes with p_exitsig != SIGCHLD
    695 			 * processes only if WALTSIG is set; wait for
    696 			 * processes with p_exitsig == SIGCHLD only
    697 			 * if WALTSIG is clear.
    698 			 */
    699 			if (((options & WALLSIG) == 0) &&
    700 			    (options & WALTSIG ? child->p_exitsig == SIGCHLD
    701 						: P_EXITSIG(child) != SIGCHLD))
    702 				continue;
    703 
    704 			c_found = 1;
    705 			if (child->p_stat == SZOMB &&
    706 			    (options & WNOZOMBIE) == 0) {
    707 				*child_p = child;
    708 				return 0;
    709 			}
    710 
    711 			if (child->p_stat == SSTOP &&
    712 			    (child->p_flag & P_WAITED) == 0 &&
    713 			    (child->p_flag & P_TRACED || options & WUNTRACED)) {
    714 				if ((options & WNOWAIT) == 0)
    715 					child->p_flag |= P_WAITED;
    716 				*child_p = child;
    717 				return 0;
    718 			}
    719 		}
    720 		if (c_found == 0)
    721 			return ECHILD;
    722 		if (options & WNOHANG) {
    723 			*child_p = NULL;
    724 			return 0;
    725 		}
    726 		error = tsleep((caddr_t)parent, PWAIT | PCATCH, "wait", 0);
    727 		if (error != 0)
    728 			return error;
    729 	}
    730 }
    731 
    732 /*
    733  * Free a process after parent has taken all the state info.
    734  */
    735 void
    736 proc_free(struct proc *p)
    737 {
    738 	struct proc *parent = p->p_pptr;
    739 	int s;
    740 
    741 	/*
    742 	 * If we got the child via ptrace(2) or procfs, and
    743 	 * the parent is different (meaning the process was
    744 	 * attached, rather than run as a child), then we need
    745 	 * to give it back to the old parent, and send the
    746 	 * parent the exit signal.  The rest of the cleanup
    747 	 * will be done when the old parent waits on the child.
    748 	 */
    749 	if ((p->p_flag & P_TRACED) && p->p_opptr != parent){
    750 		parent = p->p_opptr;
    751 		if (parent == NULL)
    752 			parent = initproc;
    753 		proc_reparent(p, parent);
    754 		p->p_opptr = NULL;
    755 		p->p_flag &= ~(P_TRACED|P_WAITED|P_FSTRACE);
    756 		if (p->p_exitsig != 0)
    757 			psignal(parent, P_EXITSIG(p));
    758 		wakeup((caddr_t)parent);
    759 		return;
    760 	}
    761 
    762 	scheduler_wait_hook(parent, p);
    763 	p->p_xstat = 0;
    764 
    765 	ruadd(&parent->p_stats->p_cru, p->p_ru);
    766 
    767 	/*
    768 	 * At this point we are going to start freeing the final resources.
    769 	 * If anyone tries to access the proc structure after here they
    770 	 * will get a shock - bits are missing.
    771 	 * Attempt to make it hard!
    772 	 */
    773 
    774 	p->p_stat = SIDL;		/* not even a zombie any more */
    775 
    776 	pool_put(&rusage_pool, p->p_ru);
    777 
    778 	/*
    779 	 * Finally finished with old proc entry.
    780 	 * Unlink it from its process group and free it.
    781 	 */
    782 	leavepgrp(p);
    783 
    784 	s = proclist_lock_write();
    785 	LIST_REMOVE(p, p_list);	/* off zombproc */
    786 	LIST_REMOVE(p, p_sibling);
    787 	proclist_unlock_write(s);
    788 
    789 	/*
    790 	 * Decrement the count of procs running with this uid.
    791 	 */
    792 	(void)chgproccnt(p->p_cred->p_ruid, -1);
    793 
    794 	/*
    795 	 * Free up credentials.
    796 	 */
    797 	if (--p->p_cred->p_refcnt == 0) {
    798 		crfree(p->p_cred->pc_ucred);
    799 		pool_put(&pcred_pool, p->p_cred);
    800 	}
    801 
    802 	/*
    803 	 * Release reference to text vnode
    804 	 */
    805 	if (p->p_textvp)
    806 		vrele(p->p_textvp);
    807 
    808 	/*
    809 	 * Release any SA state
    810 	 */
    811 	if (p->p_sa) {
    812 		free(p->p_sa->sa_stacks, M_SA);
    813 		pool_put(&sadata_pool, p->p_sa);
    814 	}
    815 
    816 	/* Free proc structure and let pid be reallocated */
    817 	proc_free_mem(p);
    818 }
    819 
    820 /*
    821  * make process 'parent' the new parent of process 'child'.
    822  */
    823 void
    824 proc_reparent(struct proc *child, struct proc *parent)
    825 {
    826 
    827 	if (child->p_pptr == parent)
    828 		return;
    829 
    830 	if (parent == initproc)
    831 		child->p_exitsig = SIGCHLD;
    832 
    833 	LIST_REMOVE(child, p_sibling);
    834 	LIST_INSERT_HEAD(&parent->p_children, child, p_sibling);
    835 	child->p_pptr = parent;
    836 }
    837