Home | History | Annotate | Line # | Download | only in kern
kern_exit.c revision 1.139
      1 /*	$NetBSD: kern_exit.c,v 1.139 2004/03/14 01:08:47 cl 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 1.139 2004/03/14 01:08:47 cl 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 		/*
    252 		 * Collect thread u-areas.
    253 		 */
    254 		uvm_uarea_drain(FALSE);
    255 	}
    256 
    257 #if defined(__HAVE_RAS)
    258 	ras_purgeall(p);
    259 #endif
    260 
    261 	/*
    262 	 * Close open files and release open-file table.
    263 	 * This may block!
    264 	 */
    265 	fdfree(p);
    266 	cwdfree(p);
    267 
    268 	doexithooks(p);
    269 
    270 	if (SESS_LEADER(p)) {
    271 		struct session *sp = p->p_session;
    272 		struct tty *tp;
    273 
    274 		if (sp->s_ttyvp) {
    275 			int s;
    276 			/*
    277 			 * Controlling process.
    278 			 * Signal foreground pgrp,
    279 			 * drain controlling terminal
    280 			 * and revoke access to controlling terminal.
    281 			 */
    282 			tp = sp->s_ttyp;
    283 			s = spltty();
    284 			TTY_LOCK(tp);
    285 			if (tp->t_session == sp) {
    286 				if (tp->t_pgrp)
    287 					pgsignal(tp->t_pgrp, SIGHUP, 1);
    288 				/* we can't guarantee the revoke will do this */
    289 				tp->t_pgrp = NULL;
    290 				tp->t_session = NULL;
    291 				TTY_UNLOCK(tp);
    292 				splx(s);
    293 				SESSRELE(sp);
    294 				(void) ttywait(tp);
    295 				/*
    296 				 * The tty could have been revoked
    297 				 * if we blocked.
    298 				 */
    299 				if (sp->s_ttyvp)
    300 					VOP_REVOKE(sp->s_ttyvp, REVOKEALL);
    301 			} else {
    302 				TTY_UNLOCK(tp);
    303 				splx(s);
    304 			}
    305 			if (sp->s_ttyvp)
    306 				vrele(sp->s_ttyvp);
    307 			sp->s_ttyvp = NULL;
    308 			/*
    309 			 * s_ttyp is not zero'd; we use this to indicate
    310 			 * that the session once had a controlling terminal.
    311 			 * (for logging and informational purposes)
    312 			 */
    313 		}
    314 		sp->s_leader = NULL;
    315 	}
    316 	fixjobc(p, p->p_pgrp, 0);
    317 	(void)acct_process(p);
    318 #ifdef KTRACE
    319 	/*
    320 	 * release trace file
    321 	 */
    322 	ktrderef(p);
    323 #endif
    324 #ifdef SYSTRACE
    325 	systrace_sys_exit(p);
    326 #endif
    327 	/*
    328 	 * If emulation has process exit hook, call it now.
    329 	 */
    330 	if (p->p_emul->e_proc_exit)
    331 		(*p->p_emul->e_proc_exit)(p);
    332 
    333 	/*
    334 	 * Reset p_opptr pointer of all former children which got
    335 	 * traced by another process and were reparented. We reset
    336 	 * it to NULL here; the trace detach code then reparents
    337 	 * the child to initproc. We only check allproc list, since
    338 	 * eventual former children on zombproc list won't reference
    339 	 * p_opptr anymore.
    340 	 */
    341 	s = proclist_lock_write();
    342 	if (p->p_flag & P_CHTRACED) {
    343 		LIST_FOREACH(q, &allproc, p_list) {
    344 			if (q->p_opptr == p)
    345 				q->p_opptr = NULL;
    346 		}
    347 	}
    348 
    349 	/*
    350 	 * Give orphaned children to init(8).
    351 	 */
    352 	q = LIST_FIRST(&p->p_children);
    353 	if (q)		/* only need this if any child is SZOMB */
    354 		wakeup(initproc);
    355 	for (; q != NULL; q = nq) {
    356 		nq = LIST_NEXT(q, p_sibling);
    357 
    358 		/*
    359 		 * Traced processes are killed since their existence
    360 		 * means someone is screwing up. Since we reset the
    361 		 * trace flags, the logic in sys_wait4() would not be
    362 		 * triggered to reparent the process to its
    363 		 * original parent, so we must do this here.
    364 		 */
    365 		if (q->p_flag & P_TRACED) {
    366 			if (q->p_opptr != q->p_pptr) {
    367 				struct proc *t = q->p_opptr;
    368 				proc_reparent(q, t ? t : initproc);
    369 				q->p_opptr = NULL;
    370 			} else
    371 				proc_reparent(q, initproc);
    372 			q->p_flag &= ~(P_TRACED|P_WAITED|P_FSTRACE);
    373 			psignal(q, SIGKILL);
    374 		} else {
    375 			proc_reparent(q, initproc);
    376 		}
    377 	}
    378 	proclist_unlock_write(s);
    379 
    380 	/*
    381 	 * Free the VM resources we're still holding on to.
    382 	 * We must do this from a valid thread because doing
    383 	 * so may block. This frees vmspace, which we don't
    384 	 * need anymore. The only remaining lwp is the one
    385 	 * we run at this moment, nothing runs in userland
    386 	 * anymore.
    387 	 */
    388 	uvm_proc_exit(p);
    389 
    390 	/*
    391 	 * Give machine-dependent code a chance to free any
    392 	 * MD LWP resources while we can still block. This must be done
    393 	 * before uvm_lwp_exit(), in case these resources are in the
    394 	 * PCB.
    395 	 * THIS IS LAST BLOCKING OPERATION.
    396 	 */
    397 #ifndef __NO_CPU_LWP_FREE
    398 	cpu_lwp_free(l, 1);
    399 #endif
    400 
    401 	pmap_deactivate(l);
    402 
    403 	/*
    404 	 * NOTE: WE ARE NO LONGER ALLOWED TO SLEEP!
    405 	 */
    406 
    407 	/*
    408 	 * Save exit status and final rusage info, adding in child rusage
    409 	 * info and self times.
    410 	 * In order to pick up the time for the current execution, we must
    411 	 * do this before unlinking the lwp from l_list.
    412 	 */
    413 	p->p_xstat = rv;
    414 	*p->p_ru = p->p_stats->p_ru;
    415 	calcru(p, &p->p_ru->ru_utime, &p->p_ru->ru_stime, NULL);
    416 	ruadd(p->p_ru, &p->p_stats->p_cru);
    417 
    418 	/*
    419 	 * Move proc from allproc to zombproc, it's now ready
    420 	 * to be collected by parent. Remaining lwp resources
    421 	 * will be freed in lwp_exit2() once we'd switch to idle
    422 	 * context.
    423 	 * Changing the state to SZOMB stops it being found by pfind().
    424 	 */
    425 	s = proclist_lock_write();
    426 	LIST_REMOVE(p, p_list);
    427 	LIST_INSERT_HEAD(&zombproc, p, p_list);
    428 	p->p_stat = SZOMB;
    429 
    430 	LIST_REMOVE(l, l_list);
    431 	LIST_REMOVE(l, l_sibling);
    432 	l->l_flag |= L_DETACHED|L_PROCEXIT;	/* detached from proc too */
    433 	l->l_stat = LSDEAD;
    434 
    435 	p->p_nrlwps--;
    436 	p->p_nlwps--;
    437 
    438 	/* Put in front of parent's sibling list for parent to collect it */
    439 	q = p->p_pptr;
    440 	q->p_nstopchild++;
    441 	if (LIST_FIRST(&q->p_children) != p) {
    442 		/* Put child where it can be found quickly */
    443 		LIST_REMOVE(p, p_sibling);
    444 		LIST_INSERT_HEAD(&q->p_children, p, p_sibling);
    445 	}
    446 
    447 	proclist_unlock_write(s);
    448 
    449 	/*
    450 	 * Notify interested parties of our demise.
    451 	 */
    452 	KNOTE(&p->p_klist, NOTE_EXIT);
    453 
    454 #if PERFCTRS
    455 	/*
    456 	 * Save final PMC information in parent process & clean up.
    457 	 */
    458 	if (PMC_ENABLED(p)) {
    459 		pmc_save_context(p);
    460 		pmc_accumulate(p->p_pptr, p);
    461 		pmc_process_exit(p);
    462 	}
    463 #endif
    464 
    465 	/*
    466 	 * Notify parent that we're gone.  If parent has the P_NOCLDWAIT
    467 	 * flag set, notify init instead (and hope it will handle
    468 	 * this situation).
    469 	 */
    470 	if (p->p_pptr->p_flag & P_NOCLDWAIT) {
    471 		struct proc *pp = p->p_pptr;
    472 		proc_reparent(p, initproc);
    473 
    474 		/*
    475 		 * If this was the last child of our parent, notify
    476 		 * parent, so in case he was wait(2)ing, he will
    477 		 * continue.
    478 		 */
    479 		if (LIST_FIRST(&pp->p_children) == NULL)
    480 			wakeup(pp);
    481 	}
    482 
    483 	/* Wake up the parent so it can get exit status. */
    484 	if ((p->p_flag & P_FSTRACE) == 0 && p->p_exitsig != 0)
    485 		exit_psignal(p, p->p_pptr);
    486 	wakeup(p->p_pptr);
    487 
    488 	/*
    489 	 * Release the process's signal state.
    490 	 */
    491 	sigactsfree(p);
    492 
    493 	/*
    494 	 * Clear curlwp after we've done all operations
    495 	 * that could block, and before tearing down the rest
    496 	 * of the process state that might be used from clock, etc.
    497 	 * Also, can't clear curlwp while we're still runnable,
    498 	 * as we're not on a run queue (we are current, just not
    499 	 * a proper proc any longer!).
    500 	 *
    501 	 * Other substructures are freed from wait().
    502 	 */
    503 	curlwp = NULL;
    504 	limfree(p->p_limit);
    505 	pstatsfree(p->p_stats);
    506 	p->p_limit = NULL;
    507 
    508 	/* This process no longer needs to hold the kernel lock. */
    509 	KERNEL_PROC_UNLOCK(l);
    510 
    511 #ifdef DEBUG
    512 	/* Nothing should use the process link anymore */
    513 	l->l_proc = NULL;
    514 #endif
    515 
    516 	/*
    517 	 * Finally, call machine-dependent code to switch to a new
    518 	 * context (possibly the idle context).  Once we are no longer
    519 	 * using the dead lwp's stack, lwp_exit2() will be called
    520 	 * to arrange for the resources to be released.
    521 	 *
    522 	 * Note that cpu_exit() will end with a call equivalent to
    523 	 * cpu_switch(), finishing our execution (pun intended).
    524 	 */
    525 
    526 	uvmexp.swtch++;
    527 	cpu_exit(l);
    528 }
    529 
    530 void
    531 exit_lwps(struct lwp *l)
    532 {
    533 	struct proc *p;
    534 	struct lwp *l2;
    535 	struct sadata_vp *vp;
    536 	int s, error;
    537 	lwpid_t		waited;
    538 
    539 	p = l->l_proc;
    540 
    541 	/* XXX SMP
    542 	 * This would be the right place to IPI any LWPs running on
    543 	 * other processors so that they can notice the userret exit hook.
    544 	 */
    545 	p->p_userret = lwp_exit_hook;
    546 	p->p_userret_arg = NULL;
    547 
    548 	if (p->p_sa) {
    549 		SLIST_FOREACH(vp, &p->p_sa->sa_vps, savp_next) {
    550 			/*
    551 			 * Make SA-cached LWPs normal process runnable
    552 			 * LWPs so that they'll also self-destruct.
    553 			 */
    554 			DPRINTF(("exit_lwps: Making cached LWPs of %d on VP %d runnable: ",
    555 				    p->p_pid, vp->savp_id));
    556 			SCHED_LOCK(s);
    557 			while ((l2 = sa_getcachelwp(vp)) != 0) {
    558 				l2->l_priority = l2->l_usrpri;
    559 				setrunnable(l2);
    560 				DPRINTF(("%d ", l2->l_lid));
    561 			}
    562 			SCHED_UNLOCK(s);
    563 			DPRINTF(("\n"));
    564 
    565 			/*
    566 			 * Clear wokenq, the LWPs on the queue will
    567 			 * run below.
    568 			 */
    569 			vp->savp_wokenq_head = NULL;
    570 		}
    571 	}
    572 
    573 	/*
    574 	 * Interrupt LWPs in interruptable sleep, unsuspend suspended
    575 	 * LWPs, make detached LWPs undetached (so we can wait for
    576 	 * them) and then wait for everyone else to finish.
    577 	 */
    578 	LIST_FOREACH(l2, &p->p_lwps, l_sibling) {
    579 		l2->l_flag &= ~(L_DETACHED|L_SA);
    580 
    581 		if ((l2->l_stat == LSSLEEP && (l2->l_flag & L_SINTR)) ||
    582 		    l2->l_stat == LSSUSPENDED || l2->l_stat == LSSTOP) {
    583 			SCHED_LOCK(s);
    584 			setrunnable(l2);
    585 			SCHED_UNLOCK(s);
    586 			DPRINTF(("exit_lwps: Made %d.%d runnable\n",
    587 			    p->p_pid, l2->l_lid));
    588 		}
    589 	}
    590 
    591 
    592 	while (p->p_nlwps > 1) {
    593 		DPRINTF(("exit_lwps: waiting for %d LWPs (%d runnable, %d zombies)\n",
    594 		    p->p_nlwps, p->p_nrlwps, p->p_nzlwps));
    595 		error = lwp_wait1(l, 0, &waited, LWPWAIT_EXITCONTROL);
    596 		if (error)
    597 			panic("exit_lwps: lwp_wait1 failed with error %d\n",
    598 			    error);
    599 		DPRINTF(("exit_lwps: Got LWP %d from lwp_wait1()\n", waited));
    600 	}
    601 
    602 	p->p_userret = NULL;
    603 }
    604 
    605 /* Wrapper function for use in p_userret */
    606 static void
    607 lwp_exit_hook(struct lwp *l, void *arg)
    608 {
    609 	KERNEL_PROC_LOCK(l);
    610 	lwp_exit(l);
    611 }
    612 
    613 int
    614 sys_wait4(struct lwp *l, void *v, register_t *retval)
    615 {
    616 	struct sys_wait4_args /* {
    617 		syscallarg(int)			pid;
    618 		syscallarg(int *)		status;
    619 		syscallarg(int)			options;
    620 		syscallarg(struct rusage *)	rusage;
    621 	} */ *uap = v;
    622 	struct proc	*child, *parent;
    623 	int		status, error;
    624 
    625 	parent = l->l_proc;
    626 
    627 	if (SCARG(uap, pid) == 0)
    628 		SCARG(uap, pid) = -parent->p_pgid;
    629 	if (SCARG(uap, options) & ~(WUNTRACED|WNOHANG|WALTSIG|WALLSIG))
    630 		return (EINVAL);
    631 
    632 	error = find_stopped_child(parent, SCARG(uap,pid), SCARG(uap,options),
    633 				&child);
    634 	if (error != 0)
    635 		return error;
    636 	if (child == NULL) {
    637 		*retval = 0;
    638 		return 0;
    639 	}
    640 
    641 	/*
    642 	 * Collect child u-areas.
    643 	 */
    644 	uvm_uarea_drain(FALSE);
    645 
    646 	retval[0] = child->p_pid;
    647 
    648 	if (child->p_stat == SZOMB) {
    649 		if (SCARG(uap, status)) {
    650 			status = child->p_xstat;	/* convert to int */
    651 			error = copyout(&status, SCARG(uap, status),
    652 					sizeof(status));
    653 			if (error)
    654 				return (error);
    655 		}
    656 		if (SCARG(uap, rusage)) {
    657 			error = copyout(child->p_ru, SCARG(uap, rusage),
    658 				    sizeof(struct rusage));
    659 			if (error)
    660 				return (error);
    661 		}
    662 
    663 		proc_free(child);
    664 		return 0;
    665 	}
    666 
    667 	/* child state must be SSTOP */
    668 	if (SCARG(uap, status)) {
    669 		status = W_STOPCODE(child->p_xstat);
    670 		return copyout(&status, SCARG(uap, status), 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 error;
    685 
    686 	for (;;) {
    687 		proclist_lock_read();
    688 		error = ECHILD;
    689 		LIST_FOREACH(child, &parent->p_children, p_sibling) {
    690 			if (pid >= 0) {
    691 				if (child->p_pid != pid) {
    692 					child = p_find(pid, PFIND_ZOMBIE |
    693 								PFIND_LOCKED);
    694 					if (child == NULL
    695 					    || child->p_pptr != parent) {
    696 						child = NULL;
    697 						break;
    698 					}
    699 				}
    700 			} else
    701 				if (pid != WAIT_ANY && child->p_pgid != -pid)
    702 					/* child not in correct pgrp */
    703 					continue;
    704 			/*
    705 			 * Wait for processes with p_exitsig != SIGCHLD
    706 			 * processes only if WALTSIG is set; wait for
    707 			 * processes with p_exitsig == SIGCHLD only
    708 			 * if WALTSIG is clear.
    709 			 */
    710 			if (((options & WALLSIG) == 0) &&
    711 			    (options & WALTSIG ? child->p_exitsig == SIGCHLD
    712 						: P_EXITSIG(child) != SIGCHLD)){
    713 				if (child->p_pid == pid) {
    714 					child = NULL;
    715 					break;
    716 				}
    717 				continue;
    718 			}
    719 
    720 			error = 0;
    721 			if (child->p_stat == SZOMB && !(options & WNOZOMBIE))
    722 				break;
    723 
    724 			if (child->p_stat == SSTOP &&
    725 			    (child->p_flag & P_WAITED) == 0 &&
    726 			    (child->p_flag & P_TRACED || options & WUNTRACED)) {
    727 				if ((options & WNOWAIT) == 0) {
    728 					child->p_flag |= P_WAITED;
    729 					parent->p_nstopchild--;
    730 				}
    731 				break;
    732 			}
    733 			if (parent->p_nstopchild == 0 || child->p_pid == pid) {
    734 				child = NULL;
    735 				break;
    736 			}
    737 		}
    738 		proclist_unlock_read();
    739 		if (child != NULL || error != 0 || options & WNOHANG) {
    740 			*child_p = child;
    741 			return error;
    742 		}
    743 		error = tsleep(parent, PWAIT | PCATCH, "wait", 0);
    744 		if (error != 0)
    745 			return error;
    746 	}
    747 }
    748 
    749 /*
    750  * Free a process after parent has taken all the state info.
    751  */
    752 void
    753 proc_free(struct proc *p)
    754 {
    755 	struct proc *parent = p->p_pptr;
    756 	int s;
    757 
    758 	KASSERT(p->p_nlwps == 0);
    759 	KASSERT(p->p_nzlwps == 0);
    760 	KASSERT(p->p_nrlwps == 0);
    761 	KASSERT(LIST_EMPTY(&p->p_lwps));
    762 
    763 	/*
    764 	 * If we got the child via ptrace(2) or procfs, and
    765 	 * the parent is different (meaning the process was
    766 	 * attached, rather than run as a child), then we need
    767 	 * to give it back to the old parent, and send the
    768 	 * parent the exit signal.  The rest of the cleanup
    769 	 * will be done when the old parent waits on the child.
    770 	 */
    771 	if ((p->p_flag & P_TRACED) && p->p_opptr != parent){
    772 		parent = p->p_opptr;
    773 		if (parent == NULL)
    774 			parent = initproc;
    775 		proc_reparent(p, parent);
    776 		p->p_opptr = NULL;
    777 		p->p_flag &= ~(P_TRACED|P_WAITED|P_FSTRACE);
    778 		if (p->p_exitsig != 0)
    779 			exit_psignal(p, parent);
    780 		wakeup(parent);
    781 		return;
    782 	}
    783 
    784 	scheduler_wait_hook(parent, p);
    785 	p->p_xstat = 0;
    786 
    787 	ruadd(&parent->p_stats->p_cru, p->p_ru);
    788 
    789 	/*
    790 	 * At this point we are going to start freeing the final resources.
    791 	 * If anyone tries to access the proc structure after here they
    792 	 * will get a shock - bits are missing.
    793 	 * Attempt to make it hard!
    794 	 */
    795 
    796 	p->p_stat = SIDL;		/* not even a zombie any more */
    797 
    798 	pool_put(&rusage_pool, p->p_ru);
    799 
    800 	/*
    801 	 * Finally finished with old proc entry.
    802 	 * Unlink it from its process group and free it.
    803 	 */
    804 	leavepgrp(p);
    805 
    806 	s = proclist_lock_write();
    807 	LIST_REMOVE(p, p_list);	/* off zombproc */
    808 	p->p_pptr->p_nstopchild--;
    809 	LIST_REMOVE(p, p_sibling);
    810 	proclist_unlock_write(s);
    811 
    812 	/*
    813 	 * Decrement the count of procs running with this uid.
    814 	 */
    815 	(void)chgproccnt(p->p_cred->p_ruid, -1);
    816 
    817 	/*
    818 	 * Free up credentials.
    819 	 */
    820 	if (--p->p_cred->p_refcnt == 0) {
    821 		crfree(p->p_cred->pc_ucred);
    822 		pool_put(&pcred_pool, p->p_cred);
    823 	}
    824 
    825 	/*
    826 	 * Release reference to text vnode
    827 	 */
    828 	if (p->p_textvp)
    829 		vrele(p->p_textvp);
    830 
    831 	/* Release any SA state. */
    832 	if (p->p_sa)
    833 		sa_release(p);
    834 
    835 	/* Free proc structure and let pid be reallocated */
    836 	proc_free_mem(p);
    837 }
    838 
    839 /*
    840  * make process 'parent' the new parent of process 'child'.
    841  *
    842  * Must be called with proclist_lock_write() held.
    843  */
    844 void
    845 proc_reparent(struct proc *child, struct proc *parent)
    846 {
    847 
    848 	if (child->p_pptr == parent)
    849 		return;
    850 
    851 	if (child->p_stat == SZOMB
    852 	    || (child->p_stat == SSTOP && !(child->p_flag & P_WAITED))) {
    853 		child->p_pptr->p_nstopchild--;
    854 		parent->p_nstopchild++;
    855 	}
    856 	if (parent == initproc)
    857 		child->p_exitsig = SIGCHLD;
    858 
    859 	LIST_REMOVE(child, p_sibling);
    860 	LIST_INSERT_HEAD(&parent->p_children, child, p_sibling);
    861 	child->p_pptr = parent;
    862 }
    863