Home | History | Annotate | Line # | Download | only in kern
kern_exit.c revision 1.132
      1 /*	$NetBSD: kern_exit.c,v 1.132 2004/01/03 19:43:55 jdolecek 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 1.132 2004/01/03 19:43:55 jdolecek 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 #if defined(__HAVE_RAS)
    252 	ras_purgeall(p);
    253 #endif
    254 
    255 	/*
    256 	 * Close open files and release open-file table.
    257 	 * This may block!
    258 	 */
    259 	fdfree(p);
    260 	cwdfree(p);
    261 
    262 	doexithooks(p);
    263 
    264 	if (SESS_LEADER(p)) {
    265 		struct session *sp = p->p_session;
    266 		struct tty *tp;
    267 
    268 		if (sp->s_ttyvp) {
    269 			/*
    270 			 * Controlling process.
    271 			 * Signal foreground pgrp,
    272 			 * drain controlling terminal
    273 			 * and revoke access to controlling terminal.
    274 			 */
    275 			tp = sp->s_ttyp;
    276 			TTY_LOCK(tp);
    277 			if (tp->t_session == sp) {
    278 				if (tp->t_pgrp)
    279 					pgsignal(tp->t_pgrp, SIGHUP, 1);
    280 				/* we can't guarantee the revoke will do this */
    281 				tp->t_pgrp = NULL;
    282 				tp->t_session = NULL;
    283 				TTY_UNLOCK(tp);
    284 				SESSRELE(sp);
    285 				(void) ttywait(tp);
    286 				/*
    287 				 * The tty could have been revoked
    288 				 * if we blocked.
    289 				 */
    290 				if (sp->s_ttyvp)
    291 					VOP_REVOKE(sp->s_ttyvp, REVOKEALL);
    292 			} else
    293 				TTY_UNLOCK(tp);
    294 			if (sp->s_ttyvp)
    295 				vrele(sp->s_ttyvp);
    296 			sp->s_ttyvp = NULL;
    297 			/*
    298 			 * s_ttyp is not zero'd; we use this to indicate
    299 			 * that the session once had a controlling terminal.
    300 			 * (for logging and informational purposes)
    301 			 */
    302 		}
    303 		sp->s_leader = NULL;
    304 	}
    305 	fixjobc(p, p->p_pgrp, 0);
    306 	(void)acct_process(p);
    307 #ifdef KTRACE
    308 	/*
    309 	 * release trace file
    310 	 */
    311 	ktrderef(p);
    312 #endif
    313 #ifdef SYSTRACE
    314 	systrace_sys_exit(p);
    315 #endif
    316 	/*
    317 	 * If emulation has process exit hook, call it now.
    318 	 */
    319 	if (p->p_emul->e_proc_exit)
    320 		(*p->p_emul->e_proc_exit)(p);
    321 
    322 	/*
    323 	 * Reset p_opptr pointer of all former children which got
    324 	 * traced by another process and were reparented. We reset
    325 	 * it to NULL here; the trace detach code then reparents
    326 	 * the child to initproc. We only check allproc list, since
    327 	 * eventual former children on zombproc list won't reference
    328 	 * p_opptr anymore.
    329 	 */
    330 	s = proclist_lock_write();
    331 	if (p->p_flag & P_CHTRACED) {
    332 		LIST_FOREACH(q, &allproc, p_list) {
    333 			if (q->p_opptr == p)
    334 				q->p_opptr = NULL;
    335 		}
    336 	}
    337 
    338 	/*
    339 	 * Give orphaned children to init(8).
    340 	 */
    341 	q = LIST_FIRST(&p->p_children);
    342 	if (q)		/* only need this if any child is SZOMB */
    343 		wakeup(initproc);
    344 	for (; q != NULL; q = nq) {
    345 		nq = LIST_NEXT(q, p_sibling);
    346 
    347 		/*
    348 		 * Traced processes are killed since their existence
    349 		 * means someone is screwing up. Since we reset the
    350 		 * trace flags, the logic in sys_wait4() would not be
    351 		 * triggered to reparent the process to its
    352 		 * original parent, so we must do this here.
    353 		 */
    354 		if (q->p_flag & P_TRACED) {
    355 			if (q->p_opptr != q->p_pptr) {
    356 				struct proc *t = q->p_opptr;
    357 				proc_reparent(q, t ? t : initproc);
    358 				q->p_opptr = NULL;
    359 			} else
    360 				proc_reparent(q, initproc);
    361 			q->p_flag &= ~(P_TRACED|P_WAITED|P_FSTRACE);
    362 			psignal(q, SIGKILL);
    363 		} else {
    364 			proc_reparent(q, initproc);
    365 		}
    366 	}
    367 	proclist_unlock_write(s);
    368 
    369 	/*
    370 	 * Save exit status and final rusage info, adding in child rusage
    371 	 * info and self times.
    372 	 * In order to pick up the time for the current execution, we must
    373 	 * do this before unlinking the lwp from l_list.
    374 	 */
    375 	p->p_xstat = rv;
    376 	*p->p_ru = p->p_stats->p_ru;
    377 	calcru(p, &p->p_ru->ru_utime, &p->p_ru->ru_stime, NULL);
    378 	ruadd(p->p_ru, &p->p_stats->p_cru);
    379 
    380 	/*
    381 	 * NOTE: WE ARE NO LONGER ALLOWED TO SLEEP!
    382 	 */
    383 
    384 	/*
    385 	 * Move proc from allproc to zombproc, but do not yet
    386 	 * wake up the reaper.  We will put the proc on the
    387 	 * deadproc list later (using the p_dead member), and
    388 	 * wake up the reaper when we do.
    389 	 * Changing the state to SDEAD stops it being found by pfind().
    390 	 */
    391 	s = proclist_lock_write();
    392 	p->p_stat = SDEAD;
    393 	p->p_nrlwps--;
    394 	l->l_stat = LSDEAD;
    395 	LIST_REMOVE(p, p_list);
    396 	LIST_INSERT_HEAD(&zombproc, p, p_list);
    397 	LIST_REMOVE(l, l_list);
    398 	l->l_flag |= L_DETACHED;
    399 	proclist_unlock_write(s);
    400 
    401 	/*
    402 	 * Notify interested parties of our demise.
    403 	 */
    404 	KNOTE(&p->p_klist, NOTE_EXIT);
    405 
    406 #if PERFCTRS
    407 	/*
    408 	 * Save final PMC information in parent process & clean up.
    409 	 */
    410 	if (PMC_ENABLED(p)) {
    411 		pmc_save_context(p);
    412 		pmc_accumulate(p->p_pptr, p);
    413 		pmc_process_exit(p);
    414 	}
    415 #endif
    416 
    417 	/*
    418 	 * Notify parent that we're gone.  If parent has the P_NOCLDWAIT
    419 	 * flag set, notify init instead (and hope it will handle
    420 	 * this situation).
    421 	 */
    422 	if (p->p_pptr->p_flag & P_NOCLDWAIT) {
    423 		struct proc *pp = p->p_pptr;
    424 		proc_reparent(p, initproc);
    425 		/*
    426 		 * If this was the last child of our parent, notify
    427 		 * parent, so in case he was wait(2)ing, he will
    428 		 * continue.
    429 		 */
    430 		if (LIST_FIRST(&pp->p_children) == NULL)
    431 			wakeup(pp);
    432 	}
    433 
    434 	/*
    435 	 * Release the process's signal state.
    436 	 */
    437 	sigactsfree(p);
    438 
    439 	/*
    440 	 * Clear curlwp after we've done all operations
    441 	 * that could block, and before tearing down the rest
    442 	 * of the process state that might be used from clock, etc.
    443 	 * Also, can't clear curlwp while we're still runnable,
    444 	 * as we're not on a run queue (we are current, just not
    445 	 * a proper proc any longer!).
    446 	 *
    447 	 * Other substructures are freed from wait().
    448 	 */
    449 	curlwp = NULL;
    450 	limfree(p->p_limit);
    451 	pstatsfree(p->p_stats);
    452 	p->p_limit = NULL;
    453 
    454 	/* This process no longer needs to hold the kernel lock. */
    455 	KERNEL_PROC_UNLOCK(l);
    456 
    457 	/*
    458 	 * Finally, call machine-dependent code to switch to a new
    459 	 * context (possibly the idle context).  Once we are no longer
    460 	 * using the dead process's vmspace and stack, exit2() will be
    461 	 * called to schedule those resources to be released by the
    462 	 * reaper thread.
    463 	 *
    464 	 * Note that cpu_exit() will end with a call equivalent to
    465 	 * cpu_switch(), finishing our execution (pun intended).
    466 	 */
    467 
    468 	cpu_exit(l, 1);
    469 }
    470 
    471 void
    472 exit_lwps(struct lwp *l)
    473 {
    474 	struct proc *p;
    475 	struct lwp *l2;
    476 	int s, error;
    477 	lwpid_t		waited;
    478 
    479 	p = l->l_proc;
    480 
    481 	/* XXX SMP
    482 	 * This would be the right place to IPI any LWPs running on
    483 	 * other processors so that they can notice the userret exit hook.
    484 	 */
    485 	p->p_userret = lwp_exit_hook;
    486 	p->p_userret_arg = NULL;
    487 
    488 	if (p->p_sa) {
    489 		/*
    490 		 * Make SA-cached LWPs normal process runnable LWPs so
    491 		 * that they'll also self-destruct.
    492 		 */
    493 		DPRINTF(("exit_lwps: Making cached LWPs of %d runnable: ",
    494 		    p->p_pid));
    495 		SCHED_LOCK(s);
    496 		while ((l2 = sa_getcachelwp(p)) != 0) {
    497 			l2->l_priority = l2->l_usrpri;
    498 			setrunnable(l2);
    499 			DPRINTF(("%d ", l2->l_lid));
    500 		}
    501 		DPRINTF(("\n"));
    502 		SCHED_UNLOCK(s);
    503 
    504 		/* Clear wokenq, the LWPs on the queue will run below */
    505 		p->p_sa->sa_wokenq_head = NULL;
    506 	}
    507 
    508 	/*
    509 	 * Interrupt LWPs in interruptable sleep, unsuspend suspended
    510 	 * LWPs, make detached LWPs undetached (so we can wait for
    511 	 * them) and then wait for everyone else to finish.
    512 	 */
    513 	LIST_FOREACH(l2, &p->p_lwps, l_sibling) {
    514 		l2->l_flag &= ~(L_DETACHED|L_SA);
    515 
    516 		if (p->p_sa && l2->l_wchan == p->p_sa)
    517 			l2->l_flag |= L_SINTR;
    518 
    519 		if ((l2->l_stat == LSSLEEP && (l2->l_flag & L_SINTR)) ||
    520 		    l2->l_stat == LSSUSPENDED || l2->l_stat == LSSTOP) {
    521 			SCHED_LOCK(s);
    522 			setrunnable(l2);
    523 			SCHED_UNLOCK(s);
    524 			DPRINTF(("exit_lwps: Made %d.%d runnable\n",
    525 			    p->p_pid, l2->l_lid));
    526 		}
    527 	}
    528 
    529 
    530 	while (p->p_nlwps > 1) {
    531 		DPRINTF(("exit_lwps: waiting for %d LWPs (%d runnable, %d zombies)\n",
    532 		    p->p_nlwps, p->p_nrlwps, p->p_nzlwps));
    533 		error = lwp_wait1(l, 0, &waited, LWPWAIT_EXITCONTROL);
    534 		if (error)
    535 			panic("exit_lwps: lwp_wait1 failed with error %d\n",
    536 			    error);
    537 		DPRINTF(("exit_lwps: Got LWP %d from lwp_wait1()\n", waited));
    538 	}
    539 
    540 	p->p_userret = NULL;
    541 }
    542 
    543 /* Wrapper function for use in p_userret */
    544 static void
    545 lwp_exit_hook(struct lwp *l, void *arg)
    546 {
    547 	KERNEL_PROC_LOCK(l);
    548 	lwp_exit(l);
    549 }
    550 
    551 /*
    552  * We are called from cpu_exit() once it is safe to schedule the
    553  * dead process's resources to be freed (i.e., once we've switched to
    554  * the idle PCB for the current CPU).
    555  *
    556  * NOTE: One must be careful with locking in this routine.  It's
    557  * called from a critical section in machine-dependent code, so
    558  * we should refrain from changing any interrupt state.
    559  *
    560  * We lock the deadproc list (a spin lock), place the proc on that
    561  * list (using the p_dead member), and wake up the reaper.
    562  */
    563 void
    564 exit2(struct lwp *l)
    565 {
    566 	struct proc *p = l->l_proc;
    567 
    568 	simple_lock(&deadproc_slock);
    569 	SLIST_INSERT_HEAD(&deadprocs, p, p_dead);
    570 	simple_unlock(&deadproc_slock);
    571 
    572 	/* lwp_exit2() will wake up deadproc for us. */
    573 	lwp_exit2(l);
    574 }
    575 
    576 /*
    577  * Process reaper.  This is run by a kernel thread to free the resources
    578  * of a dead process.  Once the resources are free, the process becomes
    579  * a zombie, and the parent is allowed to read the undead's status.
    580  */
    581 void
    582 reaper(void *arg)
    583 {
    584 	struct proc *p, *parent;
    585 	struct lwp *l;
    586 
    587 	KERNEL_PROC_UNLOCK(curlwp);
    588 
    589 	for (;;) {
    590 		simple_lock(&deadproc_slock);
    591 		p = SLIST_FIRST(&deadprocs);
    592 		l = LIST_FIRST(&deadlwp);
    593 		if (p == NULL && l == NULL) {
    594 			/* No work for us; go to sleep until someone exits. */
    595 			(void) ltsleep(&deadprocs, PVM|PNORELOCK,
    596 			    "reaper", 0, &deadproc_slock);
    597 			continue;
    598 		}
    599 
    600 		if (l != NULL ) {
    601 			p = l->l_proc;
    602 
    603 			/* Remove lwp from the deadlwp list. */
    604 			LIST_REMOVE(l, l_list);
    605 			simple_unlock(&deadproc_slock);
    606 			KERNEL_PROC_LOCK(curlwp);
    607 
    608 			/*
    609 			 * Give machine-dependent code a chance to free any
    610 			 * resources it couldn't free while still running on
    611 			 * that process's context.  This must be done before
    612 			 * uvm_lwp_exit(), in case these resources are in the
    613 			 * PCB.
    614 			 */
    615 			cpu_wait(l);
    616 
    617 			/*
    618 			 * Free the VM resources we're still holding on to.
    619 			 */
    620 			uvm_lwp_exit(l);
    621 
    622 			l->l_stat = LSZOMB;
    623 			if (l->l_flag & L_DETACHED) {
    624 				/* Nobody waits for detached LWPs. */
    625 				LIST_REMOVE(l, l_sibling);
    626 				p->p_nlwps--;
    627 				pool_put(&lwp_pool, l);
    628 			} else {
    629 				p->p_nzlwps++;
    630 				wakeup(&p->p_nlwps);
    631 			}
    632 			/* XXXNJW where should this be with respect to
    633 			 * the wakeup() above? */
    634 			KERNEL_PROC_UNLOCK(curlwp);
    635 		} else {
    636 			/* Remove proc from the deadproc list. */
    637 			SLIST_REMOVE_HEAD(&deadprocs, p_dead);
    638 			simple_unlock(&deadproc_slock);
    639 			KERNEL_PROC_LOCK(curlwp);
    640 
    641 			/*
    642 			 * Free the VM resources we're still holding on to.
    643 			 * We must do this from a valid thread because doing
    644 			 * so may block.
    645 			 */
    646 			uvm_proc_exit(p);
    647 
    648 			/* Process is now a true zombie. */
    649 			p->p_stat = SZOMB;
    650 			parent = p->p_pptr;
    651 			parent->p_nstopchild++;
    652 			if (LIST_FIRST(&parent->p_children) != p) {
    653 				/* Put child where it can be found quickly */
    654 				LIST_REMOVE(p, p_sibling);
    655 				LIST_INSERT_HEAD(&parent->p_children,
    656 						p, p_sibling);
    657 			}
    658 
    659 			/* Wake up the parent so it can get exit status. */
    660 			if ((p->p_flag & P_FSTRACE) == 0 && p->p_exitsig != 0)
    661 				exit_psignal(p, p->p_pptr);
    662 			KERNEL_PROC_UNLOCK(curlwp);
    663 			wakeup(p->p_pptr);
    664 		}
    665 	}
    666 }
    667 
    668 int
    669 sys_wait4(struct lwp *l, void *v, register_t *retval)
    670 {
    671 	struct sys_wait4_args /* {
    672 		syscallarg(int)			pid;
    673 		syscallarg(int *)		status;
    674 		syscallarg(int)			options;
    675 		syscallarg(struct rusage *)	rusage;
    676 	} */ *uap = v;
    677 	struct proc	*child, *parent;
    678 	int		status, error;
    679 
    680 	parent = l->l_proc;
    681 
    682 	if (SCARG(uap, pid) == 0)
    683 		SCARG(uap, pid) = -parent->p_pgid;
    684 	if (SCARG(uap, options) & ~(WUNTRACED|WNOHANG|WALTSIG|WALLSIG))
    685 		return (EINVAL);
    686 
    687 	error = find_stopped_child(parent, SCARG(uap,pid), SCARG(uap,options),
    688 				&child);
    689 	if (error != 0)
    690 		return error;
    691 	if (child == NULL) {
    692 		*retval = 0;
    693 		return 0;
    694 	}
    695 
    696 	retval[0] = child->p_pid;
    697 
    698 	if (child->p_stat == SZOMB) {
    699 		if (SCARG(uap, status)) {
    700 			status = child->p_xstat;	/* convert to int */
    701 			error = copyout(&status, SCARG(uap, status),
    702 					sizeof(status));
    703 			if (error)
    704 				return (error);
    705 		}
    706 		if (SCARG(uap, rusage)) {
    707 			error = copyout(child->p_ru, SCARG(uap, rusage),
    708 				    sizeof(struct rusage));
    709 			if (error)
    710 				return (error);
    711 		}
    712 
    713 		proc_free(child);
    714 		return 0;
    715 	}
    716 
    717 	/* child state must be SSTOP */
    718 	if (SCARG(uap, status)) {
    719 		status = W_STOPCODE(child->p_xstat);
    720 		return copyout(&status, SCARG(uap, status), sizeof(status));
    721 	}
    722 	return 0;
    723 }
    724 
    725 /*
    726  * Scan list of child processes for a child process that has stopped or
    727  * exited.  Used by sys_wait4 and 'compat' equivalents.
    728  */
    729 int
    730 find_stopped_child(struct proc *parent, pid_t pid, int options,
    731 	struct proc **child_p)
    732 {
    733 	struct proc *child;
    734 	int error;
    735 
    736 	for (;;) {
    737 		proclist_lock_read();
    738 		error = ECHILD;
    739 		LIST_FOREACH(child, &parent->p_children, p_sibling) {
    740 			if (pid >= 0) {
    741 				if (child->p_pid != pid) {
    742 					child = p_find(pid, PFIND_ZOMBIE |
    743 								PFIND_LOCKED);
    744 					if (child == NULL
    745 					    || child->p_pptr != parent) {
    746 						child = NULL;
    747 						break;
    748 					}
    749 				}
    750 			} else
    751 				if (pid != WAIT_ANY && child->p_pgid != -pid)
    752 					/* child not in correct pgrp */
    753 					continue;
    754 			/*
    755 			 * Wait for processes with p_exitsig != SIGCHLD
    756 			 * processes only if WALTSIG is set; wait for
    757 			 * processes with p_exitsig == SIGCHLD only
    758 			 * if WALTSIG is clear.
    759 			 */
    760 			if (((options & WALLSIG) == 0) &&
    761 			    (options & WALTSIG ? child->p_exitsig == SIGCHLD
    762 						: P_EXITSIG(child) != SIGCHLD)){
    763 				if (child->p_pid == pid) {
    764 					child = NULL;
    765 					break;
    766 				}
    767 				continue;
    768 			}
    769 
    770 			error = 0;
    771 			if (child->p_stat == SZOMB && !(options & WNOZOMBIE))
    772 				break;
    773 
    774 			if (child->p_stat == SSTOP &&
    775 			    (child->p_flag & P_WAITED) == 0 &&
    776 			    (child->p_flag & P_TRACED || options & WUNTRACED)) {
    777 				if ((options & WNOWAIT) == 0) {
    778 					child->p_flag |= P_WAITED;
    779 					parent->p_nstopchild--;
    780 				}
    781 				break;
    782 			}
    783 			if (parent->p_nstopchild == 0 || child->p_pid == pid) {
    784 				child = NULL;
    785 				break;
    786 			}
    787 		}
    788 		proclist_unlock_read();
    789 		if (child != NULL || error != 0 || options & WNOHANG) {
    790 			*child_p = child;
    791 			return error;
    792 		}
    793 		error = tsleep(parent, PWAIT | PCATCH, "wait", 0);
    794 		if (error != 0)
    795 			return error;
    796 	}
    797 }
    798 
    799 /*
    800  * Free a process after parent has taken all the state info.
    801  */
    802 void
    803 proc_free(struct proc *p)
    804 {
    805 	struct proc *parent = p->p_pptr;
    806 	int s;
    807 
    808 	/*
    809 	 * If we got the child via ptrace(2) or procfs, and
    810 	 * the parent is different (meaning the process was
    811 	 * attached, rather than run as a child), then we need
    812 	 * to give it back to the old parent, and send the
    813 	 * parent the exit signal.  The rest of the cleanup
    814 	 * will be done when the old parent waits on the child.
    815 	 */
    816 	if ((p->p_flag & P_TRACED) && p->p_opptr != parent){
    817 		parent = p->p_opptr;
    818 		if (parent == NULL)
    819 			parent = initproc;
    820 		proc_reparent(p, parent);
    821 		p->p_opptr = NULL;
    822 		p->p_flag &= ~(P_TRACED|P_WAITED|P_FSTRACE);
    823 		if (p->p_exitsig != 0)
    824 			exit_psignal(p, parent);
    825 		wakeup(parent);
    826 		return;
    827 	}
    828 
    829 	scheduler_wait_hook(parent, p);
    830 	p->p_xstat = 0;
    831 
    832 	ruadd(&parent->p_stats->p_cru, p->p_ru);
    833 
    834 	/*
    835 	 * At this point we are going to start freeing the final resources.
    836 	 * If anyone tries to access the proc structure after here they
    837 	 * will get a shock - bits are missing.
    838 	 * Attempt to make it hard!
    839 	 */
    840 
    841 	p->p_stat = SIDL;		/* not even a zombie any more */
    842 
    843 	pool_put(&rusage_pool, p->p_ru);
    844 
    845 	/*
    846 	 * Finally finished with old proc entry.
    847 	 * Unlink it from its process group and free it.
    848 	 */
    849 	leavepgrp(p);
    850 
    851 	s = proclist_lock_write();
    852 	LIST_REMOVE(p, p_list);	/* off zombproc */
    853 	p->p_pptr->p_nstopchild--;
    854 	LIST_REMOVE(p, p_sibling);
    855 	proclist_unlock_write(s);
    856 
    857 	/*
    858 	 * Decrement the count of procs running with this uid.
    859 	 */
    860 	(void)chgproccnt(p->p_cred->p_ruid, -1);
    861 
    862 	/*
    863 	 * Free up credentials.
    864 	 */
    865 	if (--p->p_cred->p_refcnt == 0) {
    866 		crfree(p->p_cred->pc_ucred);
    867 		pool_put(&pcred_pool, p->p_cred);
    868 	}
    869 
    870 	/*
    871 	 * Release reference to text vnode
    872 	 */
    873 	if (p->p_textvp)
    874 		vrele(p->p_textvp);
    875 
    876 	/* Release any SA state. */
    877 	if (p->p_sa)
    878 		sa_release(p);
    879 
    880 	/* Free proc structure and let pid be reallocated */
    881 	proc_free_mem(p);
    882 }
    883 
    884 /*
    885  * make process 'parent' the new parent of process 'child'.
    886  *
    887  * Must be called with proclist_lock_write() held.
    888  */
    889 void
    890 proc_reparent(struct proc *child, struct proc *parent)
    891 {
    892 
    893 	if (child->p_pptr == parent)
    894 		return;
    895 
    896 	if (child->p_stat == SZOMB
    897 	    || (child->p_stat == SSTOP && !(child->p_flag & P_WAITED))) {
    898 		child->p_pptr->p_nstopchild--;
    899 		parent->p_nstopchild++;
    900 	}
    901 	if (parent == initproc)
    902 		child->p_exitsig = SIGCHLD;
    903 
    904 	LIST_REMOVE(child, p_sibling);
    905 	LIST_INSERT_HEAD(&parent->p_children, child, p_sibling);
    906 	child->p_pptr = parent;
    907 }
    908