Home | History | Annotate | Line # | Download | only in kern
kern_exit.c revision 1.128
      1 /*	$NetBSD: kern_exit.c,v 1.128 2003/11/12 21:07:38 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. 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 1.128 2003/11/12 21:07:38 dsl 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 kill 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 
    203 	DPRINTF(("exit1: %d.%d exiting.\n", p->p_pid, l->l_lid));
    204 	/*
    205 	 * Disable scheduler activation upcalls.
    206 	 * We're trying to get out of here.
    207 	 */
    208 	sa = 0;
    209 	if (p->p_sa != NULL) {
    210 		l->l_flag &= ~L_SA;
    211 #if 0
    212 		p->p_flag &= ~P_SA;
    213 #endif
    214 		sa = 1;
    215 	}
    216 
    217 #ifdef PGINPROF
    218 	vmsizmon();
    219 #endif
    220 	if (p->p_flag & P_PROFIL)
    221 		stopprofclock(p);
    222 	p->p_ru = pool_get(&rusage_pool, PR_WAITOK);
    223 	/*
    224 	 * If parent is waiting for us to exit or exec, P_PPWAIT is set; we
    225 	 * wake up the parent early to avoid deadlock.
    226 	 */
    227 	if (p->p_flag & P_PPWAIT) {
    228 		p->p_flag &= ~P_PPWAIT;
    229 		wakeup(p->p_pptr);
    230 	}
    231 	sigfillset(&p->p_sigctx.ps_sigignore);
    232 	sigemptyset(&p->p_sigctx.ps_siglist);
    233 	p->p_sigctx.ps_sigcheck = 0;
    234 	timers_free(p, TIMERS_ALL);
    235 
    236 	if (sa || (p->p_nlwps > 1))
    237 		exit_lwps(l);
    238 
    239 #if defined(__HAVE_RAS)
    240 	ras_purgeall(p);
    241 #endif
    242 
    243 	/*
    244 	 * Close open files and release open-file table.
    245 	 * This may block!
    246 	 */
    247 	fdfree(p);
    248 	cwdfree(p);
    249 
    250 	doexithooks(p);
    251 
    252 	if (SESS_LEADER(p)) {
    253 		struct session *sp = p->p_session;
    254 		struct tty *tp;
    255 
    256 		if (sp->s_ttyvp) {
    257 			/*
    258 			 * Controlling process.
    259 			 * Signal foreground pgrp,
    260 			 * drain controlling terminal
    261 			 * and revoke access to controlling terminal.
    262 			 */
    263 			tp = sp->s_ttyp;
    264 			TTY_LOCK(tp);
    265 			if (tp->t_session == sp) {
    266 				if (tp->t_pgrp)
    267 					pgsignal(tp->t_pgrp, SIGHUP, 1);
    268 				/* we can't guarantee the revoke will do this */
    269 				tp->t_pgrp = NULL;
    270 				tp->t_session = NULL;
    271 				TTY_UNLOCK(tp);
    272 				SESSRELE(sp);
    273 				(void) ttywait(tp);
    274 				/*
    275 				 * The tty could have been revoked
    276 				 * if we blocked.
    277 				 */
    278 				if (sp->s_ttyvp)
    279 					VOP_REVOKE(sp->s_ttyvp, REVOKEALL);
    280 			} else
    281 				TTY_UNLOCK(tp);
    282 			if (sp->s_ttyvp)
    283 				vrele(sp->s_ttyvp);
    284 			sp->s_ttyvp = NULL;
    285 			/*
    286 			 * s_ttyp is not zero'd; we use this to indicate
    287 			 * that the session once had a controlling terminal.
    288 			 * (for logging and informational purposes)
    289 			 */
    290 		}
    291 		sp->s_leader = NULL;
    292 	}
    293 	fixjobc(p, p->p_pgrp, 0);
    294 	(void)acct_process(p);
    295 #ifdef KTRACE
    296 	/*
    297 	 * release trace file
    298 	 */
    299 	ktrderef(p);
    300 #endif
    301 #ifdef SYSTRACE
    302 	systrace_sys_exit(p);
    303 #endif
    304 	/*
    305 	 * If emulation has process exit hook, call it now.
    306 	 */
    307 	if (p->p_emul->e_proc_exit)
    308 		(*p->p_emul->e_proc_exit)(p);
    309 
    310 	/*
    311 	 * Reset p_opptr pointer of all former children which got
    312 	 * traced by another process and were reparented. We reset
    313 	 * it to NULL here; the trace detach code then reparents
    314 	 * the child to initproc. We only check allproc list, since
    315 	 * eventual former children on zombproc list won't reference
    316 	 * p_opptr anymore.
    317 	 */
    318 	s = proclist_lock_write();
    319 	if (p->p_flag & P_CHTRACED) {
    320 		LIST_FOREACH(q, &allproc, p_list) {
    321 			if (q->p_opptr == p)
    322 				q->p_opptr = NULL;
    323 		}
    324 	}
    325 
    326 	/*
    327 	 * Give orphaned children to init(8).
    328 	 */
    329 	q = LIST_FIRST(&p->p_children);
    330 	if (q)		/* only need this if any child is S_ZOMB */
    331 		wakeup(initproc);
    332 	for (; q != 0; q = nq) {
    333 		nq = LIST_NEXT(q, p_sibling);
    334 
    335 		/*
    336 		 * Traced processes are killed since their existence
    337 		 * means someone is screwing up. Since we reset the
    338 		 * trace flags, the logic in sys_wait4() would not be
    339 		 * triggered to reparent the process to its
    340 		 * original parent, so we must do this here.
    341 		 */
    342 		if (q->p_flag & P_TRACED) {
    343 			if (q->p_opptr != q->p_pptr) {
    344 				struct proc *t = q->p_opptr;
    345 				proc_reparent(q, t ? t : initproc);
    346 				q->p_opptr = NULL;
    347 			} else
    348 				proc_reparent(q, initproc);
    349 			q->p_flag &= ~(P_TRACED|P_WAITED|P_FSTRACE);
    350 			psignal(q, SIGKILL);
    351 		} else {
    352 			proc_reparent(q, initproc);
    353 		}
    354 	}
    355 	proclist_unlock_write(s);
    356 
    357 	/*
    358 	 * Save exit status and final rusage info, adding in child rusage
    359 	 * info and self times.
    360 	 * In order to pick up the time for the current execution, we must
    361 	 * do this before unlinking the lwp from l_list.
    362 	 */
    363 	p->p_xstat = rv;
    364 	*p->p_ru = p->p_stats->p_ru;
    365 	calcru(p, &p->p_ru->ru_utime, &p->p_ru->ru_stime, NULL);
    366 	ruadd(p->p_ru, &p->p_stats->p_cru);
    367 
    368 	/*
    369 	 * NOTE: WE ARE NO LONGER ALLOWED TO SLEEP!
    370 	 */
    371 
    372 	/*
    373 	 * Move proc from allproc to zombproc, but do not yet
    374 	 * wake up the reaper.  We will put the proc on the
    375 	 * deadproc list later (using the p_dead member), and
    376 	 * wake up the reaper when we do.
    377 	 * Changing the state to SDEAD stops it being found by pfind().
    378 	 */
    379 	s = proclist_lock_write();
    380 	p->p_stat = SDEAD;
    381 	p->p_nrlwps--;
    382 	l->l_stat = LSDEAD;
    383 	LIST_REMOVE(p, p_list);
    384 	LIST_INSERT_HEAD(&zombproc, p, p_list);
    385 	LIST_REMOVE(l, l_list);
    386 	l->l_flag |= L_DETACHED;
    387 	proclist_unlock_write(s);
    388 
    389 	/*
    390 	 * Notify interested parties of our demise.
    391 	 */
    392 	KNOTE(&p->p_klist, NOTE_EXIT);
    393 
    394 #if PERFCTRS
    395 	/*
    396 	 * Save final PMC information in parent process & clean up.
    397 	 */
    398 	if (PMC_ENABLED(p)) {
    399 		pmc_save_context(p);
    400 		pmc_accumulate(p->p_pptr, p);
    401 		pmc_process_exit(p);
    402 	}
    403 #endif
    404 
    405 	/*
    406 	 * Notify parent that we're gone.  If parent has the P_NOCLDWAIT
    407 	 * flag set, notify init instead (and hope it will handle
    408 	 * this situation).
    409 	 */
    410 	if (p->p_pptr->p_flag & P_NOCLDWAIT) {
    411 		struct proc *pp = p->p_pptr;
    412 		proc_reparent(p, initproc);
    413 		/*
    414 		 * If this was the last child of our parent, notify
    415 		 * parent, so in case he was wait(2)ing, he will
    416 		 * continue.
    417 		 */
    418 		if (LIST_FIRST(&pp->p_children) == NULL)
    419 			wakeup(pp);
    420 	}
    421 
    422 	/*
    423 	 * Release the process's signal state.
    424 	 */
    425 	sigactsfree(p);
    426 
    427 	/*
    428 	 * Clear curlwp after we've done all operations
    429 	 * that could block, and before tearing down the rest
    430 	 * of the process state that might be used from clock, etc.
    431 	 * Also, can't clear curlwp while we're still runnable,
    432 	 * as we're not on a run queue (we are current, just not
    433 	 * a proper proc any longer!).
    434 	 *
    435 	 * Other substructures are freed from wait().
    436 	 */
    437 	curlwp = NULL;
    438 	limfree(p->p_limit);
    439 	pstatsfree(p->p_stats);
    440 	p->p_limit = NULL;
    441 
    442 	/* This process no longer needs to hold the kernel lock. */
    443 	KERNEL_PROC_UNLOCK(l);
    444 
    445 	/*
    446 	 * Finally, call machine-dependent code to switch to a new
    447 	 * context (possibly the idle context).  Once we are no longer
    448 	 * using the dead process's vmspace and stack, exit2() will be
    449 	 * called to schedule those resources to be released by the
    450 	 * reaper thread.
    451 	 *
    452 	 * Note that cpu_exit() will end with a call equivalent to
    453 	 * cpu_switch(), finishing our execution (pun intended).
    454 	 */
    455 
    456 	cpu_exit(l, 1);
    457 }
    458 
    459 void
    460 exit_lwps(struct lwp *l)
    461 {
    462 	struct proc *p;
    463 	struct lwp *l2;
    464 	int s, error;
    465 	lwpid_t		waited;
    466 
    467 	p = l->l_proc;
    468 
    469 	/* XXX SMP
    470 	 * This would be the right place to IPI any LWPs running on
    471 	 * other processors so that they can notice the userret exit hook.
    472 	 */
    473 	p->p_userret = lwp_exit_hook;
    474 	p->p_userret_arg = NULL;
    475 
    476 	/*
    477 	 * Make SA-cached LWPs normal process runnable LWPs so that
    478 	 * they'll also self-destruct.
    479 	 */
    480 	if (p->p_sa && p->p_sa->sa_ncached > 0) {
    481 		DPRINTF(("exit_lwps: Making cached LWPs of %d runnable: ",
    482 		    p->p_pid));
    483 		SCHED_LOCK(s);
    484 		while ((l2 = sa_getcachelwp(p)) != 0) {
    485 			l2->l_priority = l2->l_usrpri;
    486 			setrunnable(l2);
    487 			DPRINTF(("%d ", l2->l_lid));
    488 		}
    489 		DPRINTF(("\n"));
    490 		SCHED_UNLOCK(s);
    491 	}
    492 
    493 	/*
    494 	 * Interrupt LWPs in interruptable sleep, unsuspend suspended
    495 	 * LWPs, make detached LWPs undetached (so we can wait for
    496 	 * them) and then wait for everyone else to finish.
    497 	 */
    498 	LIST_FOREACH(l2, &p->p_lwps, l_sibling) {
    499 		l2->l_flag &= ~(L_DETACHED|L_SA);
    500 
    501 		if (l2->l_wchan == &l2->l_upcallstack)
    502 			wakeup(&l2->l_upcallstack);
    503 
    504 		if ((l2->l_stat == LSSLEEP && (l2->l_flag & L_SINTR)) ||
    505 		    l2->l_stat == LSSUSPENDED || l2->l_stat == LSSTOP) {
    506 			SCHED_LOCK(s);
    507 			setrunnable(l2);
    508 			SCHED_UNLOCK(s);
    509 			DPRINTF(("exit_lwps: Made %d.%d runnable\n",
    510 			    p->p_pid, l2->l_lid));
    511 		}
    512 	}
    513 
    514 
    515 	while (p->p_nlwps > 1) {
    516 		DPRINTF(("exit_lwps: waiting for %d LWPs (%d runnable, %d zombies)\n",
    517 		    p->p_nlwps, p->p_nrlwps, p->p_nzlwps));
    518 		error = lwp_wait1(l, 0, &waited, LWPWAIT_EXITCONTROL);
    519 		if (error)
    520 			panic("exit_lwps: lwp_wait1 failed with error %d\n",
    521 			    error);
    522 		DPRINTF(("exit_lwps: Got LWP %d from lwp_wait1()\n", waited));
    523 	}
    524 
    525 	p->p_userret = NULL;
    526 }
    527 
    528 /* Wrapper function for use in p_userret */
    529 static void
    530 lwp_exit_hook(struct lwp *l, void *arg)
    531 {
    532 	KERNEL_PROC_LOCK(l);
    533 	lwp_exit(l);
    534 }
    535 
    536 /*
    537  * We are called from cpu_exit() once it is safe to schedule the
    538  * dead process's resources to be freed (i.e., once we've switched to
    539  * the idle PCB for the current CPU).
    540  *
    541  * NOTE: One must be careful with locking in this routine.  It's
    542  * called from a critical section in machine-dependent code, so
    543  * we should refrain from changing any interrupt state.
    544  *
    545  * We lock the deadproc list (a spin lock), place the proc on that
    546  * list (using the p_dead member), and wake up the reaper.
    547  */
    548 void
    549 exit2(struct lwp *l)
    550 {
    551 	struct proc *p = l->l_proc;
    552 
    553 	simple_lock(&deadproc_slock);
    554 	SLIST_INSERT_HEAD(&deadprocs, p, p_dead);
    555 	simple_unlock(&deadproc_slock);
    556 
    557 	/* lwp_exit2() will wake up deadproc for us. */
    558 	lwp_exit2(l);
    559 }
    560 
    561 /*
    562  * Process reaper.  This is run by a kernel thread to free the resources
    563  * of a dead process.  Once the resources are free, the process becomes
    564  * a zombie, and the parent is allowed to read the undead's status.
    565  */
    566 void
    567 reaper(void *arg)
    568 {
    569 	struct proc *p, *parent;
    570 	struct lwp *l;
    571 
    572 	KERNEL_PROC_UNLOCK(curlwp);
    573 
    574 	for (;;) {
    575 		simple_lock(&deadproc_slock);
    576 		p = SLIST_FIRST(&deadprocs);
    577 		l = LIST_FIRST(&deadlwp);
    578 		if (p == NULL && l == NULL) {
    579 			/* No work for us; go to sleep until someone exits. */
    580 			(void) ltsleep(&deadprocs, PVM|PNORELOCK,
    581 			    "reaper", 0, &deadproc_slock);
    582 			continue;
    583 		}
    584 
    585 		if (l != NULL ) {
    586 			p = l->l_proc;
    587 
    588 			/* Remove lwp from the deadlwp list. */
    589 			LIST_REMOVE(l, l_list);
    590 			simple_unlock(&deadproc_slock);
    591 			KERNEL_PROC_LOCK(curlwp);
    592 
    593 			/*
    594 			 * Give machine-dependent code a chance to free any
    595 			 * resources it couldn't free while still running on
    596 			 * that process's context.  This must be done before
    597 			 * uvm_lwp_exit(), in case these resources are in the
    598 			 * PCB.
    599 			 */
    600 			cpu_wait(l);
    601 
    602 			/*
    603 			 * Free the VM resources we're still holding on to.
    604 			 */
    605 			uvm_lwp_exit(l);
    606 
    607 			l->l_stat = LSZOMB;
    608 			if (l->l_flag & L_DETACHED) {
    609 				/* Nobody waits for detached LWPs. */
    610 				LIST_REMOVE(l, l_sibling);
    611 				p->p_nlwps--;
    612 				pool_put(&lwp_pool, l);
    613 			} else {
    614 				p->p_nzlwps++;
    615 				wakeup(&p->p_nlwps);
    616 			}
    617 			/* XXXNJW where should this be with respect to
    618 			 * the wakeup() above? */
    619 			KERNEL_PROC_UNLOCK(curlwp);
    620 		} else {
    621 			/* Remove proc from the deadproc list. */
    622 			SLIST_REMOVE_HEAD(&deadprocs, p_dead);
    623 			simple_unlock(&deadproc_slock);
    624 			KERNEL_PROC_LOCK(curlwp);
    625 
    626 			/*
    627 			 * Free the VM resources we're still holding on to.
    628 			 * We must do this from a valid thread because doing
    629 			 * so may block.
    630 			 */
    631 			uvm_proc_exit(p);
    632 
    633 			/* Process is now a true zombie. */
    634 			p->p_stat = SZOMB;
    635 			parent = p->p_pptr;
    636 			parent->p_nstopchild++;
    637 			if (LIST_FIRST(&parent->p_children) != p) {
    638 				/* Put child where it can be found quickly */
    639 				LIST_REMOVE(p, p_sibling);
    640 				LIST_INSERT_HEAD(&parent->p_children,
    641 						p, p_sibling);
    642 			}
    643 
    644 			/* Wake up the parent so it can get exit status. */
    645 			if ((p->p_flag & P_FSTRACE) == 0 && p->p_exitsig != 0)
    646 				exit_psignal(p, p->p_pptr);
    647 			KERNEL_PROC_UNLOCK(curlwp);
    648 			wakeup(p->p_pptr);
    649 		}
    650 	}
    651 }
    652 
    653 int
    654 sys_wait4(struct lwp *l, void *v, register_t *retval)
    655 {
    656 	struct sys_wait4_args /* {
    657 		syscallarg(int)			pid;
    658 		syscallarg(int *)		status;
    659 		syscallarg(int)			options;
    660 		syscallarg(struct rusage *)	rusage;
    661 	} */ *uap = v;
    662 	struct proc	*child, *parent;
    663 	int		status, error;
    664 
    665 	parent = l->l_proc;
    666 
    667 	if (SCARG(uap, pid) == 0)
    668 		SCARG(uap, pid) = -parent->p_pgid;
    669 	if (SCARG(uap, options) & ~(WUNTRACED|WNOHANG|WALTSIG|WALLSIG))
    670 		return (EINVAL);
    671 
    672 	error = find_stopped_child(parent, SCARG(uap,pid), SCARG(uap,options),
    673 				&child);
    674 	if (error != 0)
    675 		return error;
    676 	if (child == NULL) {
    677 		*retval = 0;
    678 		return 0;
    679 	}
    680 
    681 	retval[0] = child->p_pid;
    682 
    683 	if (child->p_stat == SZOMB) {
    684 		if (SCARG(uap, status)) {
    685 			status = child->p_xstat;	/* convert to int */
    686 			error = copyout(&status, SCARG(uap, status),
    687 					sizeof(status));
    688 			if (error)
    689 				return (error);
    690 		}
    691 		if (SCARG(uap, rusage)) {
    692 			error = copyout(child->p_ru, SCARG(uap, rusage),
    693 				    sizeof(struct rusage));
    694 			if (error)
    695 				return (error);
    696 		}
    697 
    698 		proc_free(child);
    699 		return 0;
    700 	}
    701 
    702 	/* child state must be SSTOP */
    703 	if (SCARG(uap, status)) {
    704 		status = W_STOPCODE(child->p_xstat);
    705 		return copyout(&status, SCARG(uap, status), sizeof(status));
    706 	}
    707 	return 0;
    708 }
    709 
    710 /*
    711  * Scan list of child processes for a child process that has stopped or
    712  * exited.  Used by sys_wait4 and 'compat' equivalents.
    713  */
    714 int
    715 find_stopped_child(struct proc *parent, pid_t pid, int options,
    716 	struct proc **child_p)
    717 {
    718 	struct proc *child;
    719 	int error;
    720 
    721 	for (;;) {
    722 		proclist_lock_read();
    723 		error = ECHILD;
    724 		LIST_FOREACH(child, &parent->p_children, p_sibling) {
    725 			if (pid >= 0) {
    726 				if (child->p_pid != pid) {
    727 					child = p_find(pid, PFIND_ZOMBIE |
    728 								PFIND_LOCKED);
    729 					if (child == NULL
    730 					    || child->p_pptr != parent) {
    731 						child = NULL;
    732 						break;
    733 					}
    734 				}
    735 			} else
    736 				if (pid != WAIT_ANY && child->p_pgid != -pid)
    737 					/* child not in correct pgrp */
    738 					continue;
    739 			/*
    740 			 * Wait for processes with p_exitsig != SIGCHLD
    741 			 * processes only if WALTSIG is set; wait for
    742 			 * processes with p_exitsig == SIGCHLD only
    743 			 * if WALTSIG is clear.
    744 			 */
    745 			if (((options & WALLSIG) == 0) &&
    746 			    (options & WALTSIG ? child->p_exitsig == SIGCHLD
    747 						: P_EXITSIG(child) != SIGCHLD)){
    748 				if (child->p_pid == pid) {
    749 					child = NULL;
    750 					break;
    751 				}
    752 				continue;
    753 			}
    754 
    755 			error = 0;
    756 			if (child->p_stat == SZOMB && !(options & WNOZOMBIE))
    757 				break;
    758 
    759 			if (child->p_stat == SSTOP &&
    760 			    (child->p_flag & P_WAITED) == 0 &&
    761 			    (child->p_flag & P_TRACED || options & WUNTRACED)) {
    762 				if ((options & WNOWAIT) == 0) {
    763 					child->p_flag |= P_WAITED;
    764 					parent->p_nstopchild--;
    765 				}
    766 				break;
    767 			}
    768 			if (parent->p_nstopchild == 0 || child->p_pid == pid) {
    769 				child = NULL;
    770 				break;
    771 			}
    772 		}
    773 		proclist_unlock_read();
    774 		if (child != NULL || error != 0 || options & WNOHANG) {
    775 			*child_p = child;
    776 			return error;
    777 		}
    778 		error = tsleep(parent, PWAIT | PCATCH, "wait", 0);
    779 		if (error != 0)
    780 			return error;
    781 	}
    782 }
    783 
    784 /*
    785  * Free a process after parent has taken all the state info.
    786  */
    787 void
    788 proc_free(struct proc *p)
    789 {
    790 	struct proc *parent = p->p_pptr;
    791 	int s;
    792 
    793 	/*
    794 	 * If we got the child via ptrace(2) or procfs, and
    795 	 * the parent is different (meaning the process was
    796 	 * attached, rather than run as a child), then we need
    797 	 * to give it back to the old parent, and send the
    798 	 * parent the exit signal.  The rest of the cleanup
    799 	 * will be done when the old parent waits on the child.
    800 	 */
    801 	if ((p->p_flag & P_TRACED) && p->p_opptr != parent){
    802 		parent = p->p_opptr;
    803 		if (parent == NULL)
    804 			parent = initproc;
    805 		proc_reparent(p, parent);
    806 		p->p_opptr = NULL;
    807 		p->p_flag &= ~(P_TRACED|P_WAITED|P_FSTRACE);
    808 		if (p->p_exitsig != 0)
    809 			exit_psignal(p, parent);
    810 		wakeup(parent);
    811 		return;
    812 	}
    813 
    814 	scheduler_wait_hook(parent, p);
    815 	p->p_xstat = 0;
    816 
    817 	ruadd(&parent->p_stats->p_cru, p->p_ru);
    818 
    819 	/*
    820 	 * At this point we are going to start freeing the final resources.
    821 	 * If anyone tries to access the proc structure after here they
    822 	 * will get a shock - bits are missing.
    823 	 * Attempt to make it hard!
    824 	 */
    825 
    826 	p->p_stat = SIDL;		/* not even a zombie any more */
    827 
    828 	pool_put(&rusage_pool, p->p_ru);
    829 
    830 	/*
    831 	 * Finally finished with old proc entry.
    832 	 * Unlink it from its process group and free it.
    833 	 */
    834 	leavepgrp(p);
    835 
    836 	s = proclist_lock_write();
    837 	LIST_REMOVE(p, p_list);	/* off zombproc */
    838 	p->p_pptr->p_nstopchild--;
    839 	LIST_REMOVE(p, p_sibling);
    840 	proclist_unlock_write(s);
    841 
    842 	/*
    843 	 * Decrement the count of procs running with this uid.
    844 	 */
    845 	(void)chgproccnt(p->p_cred->p_ruid, -1);
    846 
    847 	/*
    848 	 * Free up credentials.
    849 	 */
    850 	if (--p->p_cred->p_refcnt == 0) {
    851 		crfree(p->p_cred->pc_ucred);
    852 		pool_put(&pcred_pool, p->p_cred);
    853 	}
    854 
    855 	/*
    856 	 * Release reference to text vnode
    857 	 */
    858 	if (p->p_textvp)
    859 		vrele(p->p_textvp);
    860 
    861 	/*
    862 	 * Release any SA state
    863 	 */
    864 	if (p->p_sa) {
    865 		free(p->p_sa->sa_stacks, M_SA);
    866 		pool_put(&sadata_pool, p->p_sa);
    867 	}
    868 
    869 	/* Free proc structure and let pid be reallocated */
    870 	proc_free_mem(p);
    871 }
    872 
    873 /*
    874  * make process 'parent' the new parent of process 'child'.
    875  *
    876  * Must be called with proclist_lock_write() held.
    877  */
    878 void
    879 proc_reparent(struct proc *child, struct proc *parent)
    880 {
    881 
    882 	if (child->p_pptr == parent)
    883 		return;
    884 
    885 	if (child->p_stat == SZOMB
    886 	    || (child->p_stat == SSTOP && !(child->p_flag & P_WAITED))) {
    887 		child->p_pptr->p_nstopchild--;
    888 		parent->p_nstopchild++;
    889 	}
    890 	if (parent == initproc)
    891 		child->p_exitsig = SIGCHLD;
    892 
    893 	LIST_REMOVE(child, p_sibling);
    894 	LIST_INSERT_HEAD(&parent->p_children, child, p_sibling);
    895 	child->p_pptr = parent;
    896 }
    897