Home | History | Annotate | Line # | Download | only in kern
kern_exit.c revision 1.109
      1 /*	$NetBSD: kern_exit.c,v 1.109 2003/02/14 10:11:57 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 1.109 2003/02/14 10:11:57 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 	DPRINTF(("exit1: %d.%d exiting.\n", p->p_pid, l->l_lid));
    174 	/*
    175 	 * Disable scheduler activation upcalls.
    176 	 * We're trying to get out of here.
    177 	 */
    178 	sa = 0;
    179 	if (l->l_flag & L_SA) {
    180 		l->l_flag &= ~L_SA;
    181 		p->p_flag &= ~P_SA;
    182 		sa = 1;
    183 	}
    184 
    185 #ifdef PGINPROF
    186 	vmsizmon();
    187 #endif
    188 	if (p->p_flag & P_PROFIL)
    189 		stopprofclock(p);
    190 	p->p_ru = pool_get(&rusage_pool, PR_WAITOK);
    191 	/*
    192 	 * If parent is waiting for us to exit or exec, P_PPWAIT is set; we
    193 	 * wake up the parent early to avoid deadlock.
    194 	 */
    195 	p->p_flag |= P_WEXIT;
    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 	 * NOTE: WE ARE NO LONGER ALLOWED TO SLEEP!
    272 	 */
    273 	p->p_stat = SDEAD;
    274 	p->p_nrlwps--;
    275 	l->l_stat = SDEAD;
    276 
    277 	/*
    278 	 * Remove proc from pidhash chain so looking it up won't
    279 	 * work.  Move it from allproc to zombproc, but do not yet
    280 	 * wake up the reaper.  We will put the proc on the
    281 	 * deadproc list later (using the p_hash member), and
    282 	 * wake up the reaper when we do.
    283 	 */
    284 	s = proclist_lock_write();
    285 	LIST_REMOVE(p, p_hash);
    286 	LIST_REMOVE(p, p_list);
    287 	LIST_INSERT_HEAD(&zombproc, p, p_list);
    288 	LIST_REMOVE(l, l_list);
    289 	l->l_flag |= L_DETACHED;
    290 	proclist_unlock_write(s);
    291 
    292 	/*
    293 	 * Give orphaned children to init(8).
    294 	 */
    295 	q = LIST_FIRST(&p->p_children);
    296 	if (q)		/* only need this if any child is S_ZOMB */
    297 		wakeup((caddr_t)initproc);
    298 	for (; q != 0; q = nq) {
    299 		nq = LIST_NEXT(q, p_sibling);
    300 
    301 		/*
    302 		 * Traced processes are killed since their existence
    303 		 * means someone is screwing up. Since we reset the
    304 		 * trace flags, the logic in sys_wait4() would not be
    305 		 * triggered to reparent the process to its
    306 		 * original parent, so we must do this here.
    307 		 */
    308 		if (q->p_flag & P_TRACED) {
    309 			if (q->p_opptr != q->p_pptr) {
    310 				struct proc *t = q->p_opptr;
    311 				proc_reparent(q, t ? t : initproc);
    312 				q->p_opptr = NULL;
    313 			} else
    314 				proc_reparent(q, initproc);
    315 			q->p_flag &= ~(P_TRACED|P_WAITED|P_FSTRACE);
    316 			psignal(q, SIGKILL);
    317 		} else {
    318 			proc_reparent(q, initproc);
    319 		}
    320 	}
    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 	if (p->p_flag & P_CHTRACED) {
    331 		struct proc *t;
    332 
    333 		proclist_lock_read();
    334 
    335 		LIST_FOREACH(t, &allproc, p_list) {
    336 			if (t->p_opptr == p)
    337 				t->p_opptr = NULL;
    338 		}
    339 
    340 		proclist_unlock_read();
    341 	}
    342 
    343 	/*
    344 	 * Save exit status and final rusage info, adding in child rusage
    345 	 * info and self times.
    346 	 */
    347 	p->p_xstat = rv;
    348 	*p->p_ru = p->p_stats->p_ru;
    349 	calcru(p, &p->p_ru->ru_utime, &p->p_ru->ru_stime, NULL);
    350 	ruadd(p->p_ru, &p->p_stats->p_cru);
    351 
    352 	/*
    353 	 * Notify interested parties of our demise.
    354 	 */
    355 	KNOTE(&p->p_klist, NOTE_EXIT);
    356 
    357 #if PERFCTRS
    358 	/*
    359 	 * Save final PMC information in parent process & clean up.
    360 	 */
    361 	if (PMC_ENABLED(p)) {
    362 		pmc_save_context(p);
    363 		pmc_accumulate(p->p_pptr, p);
    364 		pmc_process_exit(p);
    365 	}
    366 #endif
    367 
    368 	/*
    369 	 * Notify parent that we're gone.  If parent has the P_NOCLDWAIT
    370 	 * flag set, notify init instead (and hope it will handle
    371 	 * this situation).
    372 	 */
    373 	if (p->p_pptr->p_flag & P_NOCLDWAIT) {
    374 		struct proc *pp = p->p_pptr;
    375 		proc_reparent(p, initproc);
    376 		/*
    377 		 * If this was the last child of our parent, notify
    378 		 * parent, so in case he was wait(2)ing, he will
    379 		 * continue.
    380 		 */
    381 		if (LIST_FIRST(&pp->p_children) == NULL)
    382 			wakeup((caddr_t)pp);
    383 	}
    384 
    385 	/*
    386 	 * Release the process's signal state.
    387 	 */
    388 	sigactsfree(p);
    389 
    390 	/*
    391 	 * Clear curlwp after we've done all operations
    392 	 * that could block, and before tearing down the rest
    393 	 * of the process state that might be used from clock, etc.
    394 	 * Also, can't clear curlwp while we're still runnable,
    395 	 * as we're not on a run queue (we are current, just not
    396 	 * a proper proc any longer!).
    397 	 *
    398 	 * Other substructures are freed from wait().
    399 	 */
    400 	curlwp = NULL;
    401 	limfree(p->p_limit);
    402 	pstatsfree(p->p_stats);
    403 	p->p_limit = NULL;
    404 
    405 	/* This process no longer needs to hold the kernel lock. */
    406 	KERNEL_PROC_UNLOCK(l);
    407 
    408 	/*
    409 	 * Finally, call machine-dependent code to switch to a new
    410 	 * context (possibly the idle context).  Once we are no longer
    411 	 * using the dead process's vmspace and stack, exit2() will be
    412 	 * called to schedule those resources to be released by the
    413 	 * reaper thread.
    414 	 *
    415 	 * Note that cpu_exit() will end with a call equivalent to
    416 	 * cpu_switch(), finishing our execution (pun intended).
    417 	 */
    418 
    419 	cpu_exit(l, 1);
    420 }
    421 
    422 void
    423 exit_lwps(struct lwp *l)
    424 {
    425 	struct proc *p;
    426 	struct lwp *l2;
    427 	int s, error;
    428 	lwpid_t		waited;
    429 
    430 	p = l->l_proc;
    431 
    432 	/* XXX SMP
    433 	 * This would be the right place to IPI any LWPs running on
    434 	 * other processors so that they can notice the userret exit hook.
    435 	 */
    436 	p->p_userret = lwp_exit_hook;
    437 	p->p_userret_arg = NULL;
    438 
    439 	/*
    440 	 * Make SA-cached LWPs normal process runnable LWPs so that
    441 	 * they'll also self-destruct.
    442 	 */
    443 	if (p->p_sa && p->p_sa->sa_ncached > 0) {
    444 		DPRINTF(("exit_lwps: Making cached LWPs of %d runnable: ",
    445 		    p->p_pid));
    446 		SCHED_LOCK(s);
    447 		while ((l2 = sa_getcachelwp(p)) != 0) {
    448 			l2->l_priority = l2->l_usrpri;
    449 			setrunnable(l2);
    450 			DPRINTF(("%d ", l2->l_lid));
    451 		}
    452 		DPRINTF(("\n"));
    453 		SCHED_UNLOCK(s);
    454 	}
    455 
    456 	/*
    457 	 * Interrupt LWPs in interruptable sleep, unsuspend suspended
    458 	 * LWPs, make detached LWPs undeached (so we can wait for
    459 	 * them) and then wait for everyone else to finish.
    460 	 */
    461 	LIST_FOREACH(l2, &p->p_lwps, l_sibling) {
    462 		l2->l_flag &= ~(L_DETACHED|L_SA);
    463 		if ((l2->l_stat == LSSLEEP && (l2->l_flag & L_SINTR)) ||
    464 		    l2->l_stat == LSSUSPENDED) {
    465 			SCHED_LOCK(s);
    466 			setrunnable(l2);
    467 			SCHED_UNLOCK(s);
    468 			DPRINTF(("exit_lwps: Made %d.%d runnable\n",
    469 			    p->p_pid, l2->l_lid));
    470 		}
    471 	}
    472 
    473 
    474 	while (p->p_nlwps > 1) {
    475 		DPRINTF(("exit_lwps: waiting for %d LWPs (%d runnable, %d zombies)\n",
    476 		    p->p_nlwps, p->p_nrlwps, p->p_nzlwps));
    477 		error = lwp_wait1(l, 0, &waited, LWPWAIT_EXITCONTROL);
    478 		if (error)
    479 			panic("exit_lwps: lwp_wait1 failed with error %d\n",
    480 			    error);
    481 		DPRINTF(("exit_lwps: Got LWP %d from lwp_wait1()\n", waited));
    482 	}
    483 
    484 	p->p_userret = NULL;
    485 }
    486 
    487 /* Wrapper function for use in p_userret */
    488 static void
    489 lwp_exit_hook(struct lwp *l, void *arg)
    490 {
    491 	KERNEL_PROC_LOCK(l);
    492 	lwp_exit(l);
    493 }
    494 
    495 /*
    496  * We are called from cpu_exit() once it is safe to schedule the
    497  * dead process's resources to be freed (i.e., once we've switched to
    498  * the idle PCB for the current CPU).
    499  *
    500  * NOTE: One must be careful with locking in this routine.  It's
    501  * called from a critical section in machine-dependent code, so
    502  * we should refrain from changing any interrupt state.
    503  *
    504  * We lock the deadproc list (a spin lock), place the proc on that
    505  * list (using the p_hash member), and wake up the reaper.
    506  */
    507 void
    508 exit2(struct lwp *l)
    509 {
    510 	struct proc *p = l->l_proc;
    511 
    512 	simple_lock(&deadproc_slock);
    513 	LIST_INSERT_HEAD(&deadproc, p, p_hash);
    514 	simple_unlock(&deadproc_slock);
    515 
    516 	/* lwp_exit2() will wake up deadproc for us. */
    517 	lwp_exit2(l);
    518 }
    519 
    520 /*
    521  * Process reaper.  This is run by a kernel thread to free the resources
    522  * of a dead process.  Once the resources are free, the process becomes
    523  * a zombie, and the parent is allowed to read the undead's status.
    524  */
    525 void
    526 reaper(void *arg)
    527 {
    528 	struct proc *p;
    529 	struct lwp *l;
    530 
    531 	KERNEL_PROC_UNLOCK(curlwp);
    532 
    533 	for (;;) {
    534 		simple_lock(&deadproc_slock);
    535 		p = LIST_FIRST(&deadproc);
    536 		l = LIST_FIRST(&deadlwp);
    537 		if (p == NULL && l == NULL) {
    538 			/* No work for us; go to sleep until someone exits. */
    539 			(void) ltsleep(&deadproc, PVM|PNORELOCK,
    540 			    "reaper", 0, &deadproc_slock);
    541 			continue;
    542 		}
    543 
    544 		if (l != NULL ) {
    545 			p = l->l_proc;
    546 
    547 			/* Remove us from the deadlwp list. */
    548 			LIST_REMOVE(l, l_list);
    549 			simple_unlock(&deadproc_slock);
    550 			KERNEL_PROC_LOCK(curlwp);
    551 
    552 			/*
    553 			 * Give machine-dependent code a chance to free any
    554 			 * resources it couldn't free while still running on
    555 			 * that process's context.  This must be done before
    556 			 * uvm_lwp_exit(), in case these resources are in the
    557 			 * PCB.
    558 			 */
    559 			cpu_wait(l);
    560 
    561 			/*
    562 			 * Free the VM resources we're still holding on to.
    563 			 */
    564 			uvm_lwp_exit(l);
    565 
    566 			l->l_stat = LSZOMB;
    567 			if (l->l_flag & L_DETACHED) {
    568 				/* Nobody waits for detached LWPs. */
    569 				LIST_REMOVE(l, l_sibling);
    570 				p->p_nlwps--;
    571 				pool_put(&lwp_pool, l);
    572 			} else {
    573 				p->p_nzlwps++;
    574 				wakeup((caddr_t)&p->p_nlwps);
    575 			}
    576 			/* XXXNJW where should this be with respect to
    577 			 * the wakeup() above? */
    578 			KERNEL_PROC_UNLOCK(curlwp);
    579 		} else {
    580 			/* Remove us from the deadproc list. */
    581 			LIST_REMOVE(p, p_hash);
    582 			simple_unlock(&deadproc_slock);
    583 			KERNEL_PROC_LOCK(curlwp);
    584 
    585 			/*
    586 			 * Free the VM resources we're still holding on to.
    587 			 * We must do this from a valid thread because doing
    588 			 * so may block.
    589 			 */
    590 			uvm_proc_exit(p);
    591 
    592 			/* Process is now a true zombie. */
    593 			p->p_stat = SZOMB;
    594 
    595 			/* Wake up the parent so it can get exit status. */
    596 			if ((p->p_flag & P_FSTRACE) == 0 && p->p_exitsig != 0)
    597 				psignal(p->p_pptr, P_EXITSIG(p));
    598 			KERNEL_PROC_UNLOCK(curlwp);
    599 			wakeup((caddr_t)p->p_pptr);
    600 		}
    601 	}
    602 }
    603 
    604 int
    605 sys_wait4(struct lwp *l, void *v, register_t *retval)
    606 {
    607 	struct sys_wait4_args /* {
    608 		syscallarg(int)			pid;
    609 		syscallarg(int *)		status;
    610 		syscallarg(int)			options;
    611 		syscallarg(struct rusage *)	rusage;
    612 	} */ *uap = v;
    613 	struct proc	*child, *parent;
    614 	int		status, error;
    615 
    616 	parent = l->l_proc;
    617 
    618 	if (SCARG(uap, pid) == 0)
    619 		SCARG(uap, pid) = -parent->p_pgid;
    620 	if (SCARG(uap, options) & ~(WUNTRACED|WNOHANG|WALTSIG|WALLSIG))
    621 		return (EINVAL);
    622 
    623 	error = find_stopped_child(parent, SCARG(uap,pid), SCARG(uap,options),
    624 				&child);
    625 	if (error != 0)
    626 		return error;
    627 	if (child == NULL) {
    628 		*retval = 0;
    629 		return 0;
    630 	}
    631 
    632 	retval[0] = child->p_pid;
    633 
    634 	if (child->p_stat == SZOMB) {
    635 		if (SCARG(uap, status)) {
    636 			status = child->p_xstat;	/* convert to int */
    637 			error = copyout((caddr_t)&status,
    638 					(caddr_t)SCARG(uap, status),
    639 					sizeof(status));
    640 			if (error)
    641 				return (error);
    642 		}
    643 		if (SCARG(uap, rusage)) {
    644 			error = copyout((caddr_t)child->p_ru,
    645 				    (caddr_t)SCARG(uap, rusage),
    646 				    sizeof(struct rusage));
    647 			if (error)
    648 				return (error);
    649 		}
    650 
    651 		proc_free(child);
    652 		return 0;
    653 	}
    654 
    655 	/* child state must be SSTOP */
    656 	if (SCARG(uap, status)) {
    657 		status = W_STOPCODE(child->p_xstat);
    658 		return copyout((caddr_t)&status,
    659 				(caddr_t)SCARG(uap, status),
    660 				sizeof(status));
    661 	}
    662 	return 0;
    663 }
    664 
    665 /*
    666  * Scan list of child processes for a child process that has stopped or
    667  * exited.  Used by sys_wait4 and 'compat' equivalents.
    668  */
    669 int
    670 find_stopped_child(struct proc *parent, pid_t pid, int options,
    671 	struct proc **child_p)
    672 {
    673 	struct proc *child;
    674 	int c_found, error;
    675 
    676 	 for (;;) {
    677 		c_found = 0;
    678 		LIST_FOREACH(child, &parent->p_children, p_sibling) {
    679 			if (pid != WAIT_ANY &&
    680 			    child->p_pid != pid &&
    681 			    child->p_pgid != -pid)
    682 				continue;
    683 			/*
    684 			 * Wait for processes with p_exitsig != SIGCHLD
    685 			 * processes only if WALTSIG is set; wait for
    686 			 * processes with p_exitsig == SIGCHLD only
    687 			 * if WALTSIG is clear.
    688 			 */
    689 			if (((options & WALLSIG) == 0) &&
    690 			    (options & WALTSIG ? child->p_exitsig == SIGCHLD
    691 						: P_EXITSIG(child) != SIGCHLD))
    692 				continue;
    693 
    694 			c_found = 1;
    695 			if (child->p_stat == SZOMB &&
    696 			    (options & WNOZOMBIE) == 0) {
    697 				*child_p = child;
    698 				return 0;
    699 			}
    700 
    701 			if (child->p_stat == SSTOP &&
    702 			    (child->p_flag & P_WAITED) == 0 &&
    703 			    (child->p_flag & P_TRACED || options & WUNTRACED)) {
    704 				if ((options & WNOWAIT) == 0)
    705 					child->p_flag |= P_WAITED;
    706 				*child_p = child;
    707 				return 0;
    708 			}
    709 		}
    710 		if (c_found == 0)
    711 			return ECHILD;
    712 		if (options & WNOHANG) {
    713 			*child_p = NULL;
    714 			return 0;
    715 		}
    716 		error = tsleep((caddr_t)parent, PWAIT | PCATCH, "wait", 0);
    717 		if (error != 0)
    718 			return error;
    719 	}
    720 }
    721 
    722 /*
    723  * Free a process after parent has taken all the state info.
    724  */
    725 void
    726 proc_free(struct proc *p)
    727 {
    728 	struct proc *parent = p->p_pptr;
    729 	int s;
    730 
    731 	/*
    732 	 * If we got the child via ptrace(2) or procfs, and
    733 	 * the parent is different (meaning the process was
    734 	 * attached, rather than run as a child), then we need
    735 	 * to give it back to the old parent, and send the
    736 	 * parent the exit signal.  The rest of the cleanup
    737 	 * will be done when the old parent waits on the child.
    738 	 */
    739 	if ((p->p_flag & P_TRACED) && p->p_opptr != parent){
    740 		parent = p->p_opptr;
    741 		if (parent == NULL)
    742 			parent = initproc;
    743 		proc_reparent(p, parent);
    744 		p->p_opptr = NULL;
    745 		p->p_flag &= ~(P_TRACED|P_WAITED|P_FSTRACE);
    746 		if (p->p_exitsig != 0)
    747 			psignal(parent, P_EXITSIG(p));
    748 		wakeup((caddr_t)parent);
    749 		return;
    750 	}
    751 
    752 	scheduler_wait_hook(parent, p);
    753 	p->p_xstat = 0;
    754 	ruadd(&parent->p_stats->p_cru, p->p_ru);
    755 	pool_put(&rusage_pool, p->p_ru);
    756 
    757 	/*
    758 	 * Finally finished with old proc entry.
    759 	 * Unlink it from its process group and free it.
    760 	 */
    761 	leavepgrp(p);
    762 
    763 	s = proclist_lock_write();
    764 	LIST_REMOVE(p, p_list);	/* off zombproc */
    765 	proclist_unlock_write(s);
    766 
    767 	LIST_REMOVE(p, p_sibling);
    768 
    769 	/*
    770 	 * Decrement the count of procs running with this uid.
    771 	 */
    772 	(void)chgproccnt(p->p_cred->p_ruid, -1);
    773 
    774 	/*
    775 	 * Free up credentials.
    776 	 */
    777 	if (--p->p_cred->p_refcnt == 0) {
    778 		crfree(p->p_cred->pc_ucred);
    779 		pool_put(&pcred_pool, p->p_cred);
    780 	}
    781 
    782 	/*
    783 	 * Release reference to text vnode
    784 	 */
    785 	if (p->p_textvp)
    786 		vrele(p->p_textvp);
    787 
    788 	/*
    789 	 * Release any SA state
    790 	 */
    791 	if (p->p_sa) {
    792 		free(p->p_sa->sa_stacks, M_SA);
    793 		pool_put(&sadata_pool, p->p_sa);
    794 	}
    795 
    796 	pool_put(&proc_pool, p);
    797 	nprocs--;
    798 	return;
    799 }
    800 
    801 /*
    802  * make process 'parent' the new parent of process 'child'.
    803  */
    804 void
    805 proc_reparent(struct proc *child, struct proc *parent)
    806 {
    807 
    808 	if (child->p_pptr == parent)
    809 		return;
    810 
    811 	if (parent == initproc)
    812 		child->p_exitsig = SIGCHLD;
    813 
    814 	LIST_REMOVE(child, p_sibling);
    815 	LIST_INSERT_HEAD(&parent->p_children, child, p_sibling);
    816 	child->p_pptr = parent;
    817 }
    818