Home | History | Annotate | Line # | Download | only in kern
kern_exit.c revision 1.151
      1 /*	$NetBSD: kern_exit.c,v 1.151 2005/08/28 14:57:18 yamt Exp $	*/
      2 
      3 /*-
      4  * Copyright (c) 1998, 1999 The NetBSD Foundation, Inc.
      5  * All rights reserved.
      6  *
      7  * This code is derived from software contributed to The NetBSD Foundation
      8  * by Jason R. Thorpe of the Numerical Aerospace Simulation Facility,
      9  * NASA Ames Research Center.
     10  *
     11  * Redistribution and use in source and binary forms, with or without
     12  * modification, are permitted provided that the following conditions
     13  * are met:
     14  * 1. Redistributions of source code must retain the above copyright
     15  *    notice, this list of conditions and the following disclaimer.
     16  * 2. Redistributions in binary form must reproduce the above copyright
     17  *    notice, this list of conditions and the following disclaimer in the
     18  *    documentation and/or other materials provided with the distribution.
     19  * 3. All advertising materials mentioning features or use of this software
     20  *    must display the following acknowledgement:
     21  *	This product includes software developed by the NetBSD
     22  *	Foundation, Inc. and its contributors.
     23  * 4. Neither the name of The NetBSD Foundation nor the names of its
     24  *    contributors may be used to endorse or promote products derived
     25  *    from this software without specific prior written permission.
     26  *
     27  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
     28  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
     29  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
     30  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
     31  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
     32  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
     33  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
     34  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
     35  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
     36  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
     37  * POSSIBILITY OF SUCH DAMAGE.
     38  */
     39 
     40 /*
     41  * Copyright (c) 1982, 1986, 1989, 1991, 1993
     42  *	The Regents of the University of California.  All rights reserved.
     43  * (c) UNIX System Laboratories, Inc.
     44  * All or some portions of this file are derived from material licensed
     45  * to the University of California by American Telephone and Telegraph
     46  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
     47  * the permission of UNIX System Laboratories, Inc.
     48  *
     49  * Redistribution and use in source and binary forms, with or without
     50  * modification, are permitted provided that the following conditions
     51  * are met:
     52  * 1. Redistributions of source code must retain the above copyright
     53  *    notice, this list of conditions and the following disclaimer.
     54  * 2. Redistributions in binary form must reproduce the above copyright
     55  *    notice, this list of conditions and the following disclaimer in the
     56  *    documentation and/or other materials provided with the distribution.
     57  * 3. Neither the name of the University nor the names of its contributors
     58  *    may be used to endorse or promote products derived from this software
     59  *    without specific prior written permission.
     60  *
     61  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
     62  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
     63  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
     64  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
     65  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
     66  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
     67  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
     68  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
     69  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
     70  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
     71  * SUCH DAMAGE.
     72  *
     73  *	@(#)kern_exit.c	8.10 (Berkeley) 2/23/95
     74  */
     75 
     76 #include <sys/cdefs.h>
     77 __KERNEL_RCSID(0, "$NetBSD: kern_exit.c,v 1.151 2005/08/28 14:57:18 yamt Exp $");
     78 
     79 #include "opt_ktrace.h"
     80 #include "opt_perfctrs.h"
     81 #include "opt_systrace.h"
     82 #include "opt_sysv.h"
     83 
     84 #include <sys/param.h>
     85 #include <sys/systm.h>
     86 #include <sys/ioctl.h>
     87 #include <sys/proc.h>
     88 #include <sys/tty.h>
     89 #include <sys/time.h>
     90 #include <sys/resource.h>
     91 #include <sys/kernel.h>
     92 #include <sys/ktrace.h>
     93 #include <sys/proc.h>
     94 #include <sys/buf.h>
     95 #include <sys/wait.h>
     96 #include <sys/file.h>
     97 #include <sys/vnode.h>
     98 #include <sys/syslog.h>
     99 #include <sys/malloc.h>
    100 #include <sys/pool.h>
    101 #include <sys/resourcevar.h>
    102 #if defined(PERFCTRS)
    103 #include <sys/pmc.h>
    104 #endif
    105 #include <sys/ptrace.h>
    106 #include <sys/acct.h>
    107 #include <sys/filedesc.h>
    108 #include <sys/ras.h>
    109 #include <sys/signalvar.h>
    110 #include <sys/sched.h>
    111 #include <sys/sa.h>
    112 #include <sys/savar.h>
    113 #include <sys/mount.h>
    114 #include <sys/syscallargs.h>
    115 #include <sys/systrace.h>
    116 
    117 #include <machine/cpu.h>
    118 
    119 #include <uvm/uvm_extern.h>
    120 
    121 #define DEBUG_EXIT
    122 
    123 #ifdef DEBUG_EXIT
    124 int debug_exit = 0;
    125 #define DPRINTF(x) if (debug_exit) printf x
    126 #else
    127 #define DPRINTF(x)
    128 #endif
    129 
    130 static void lwp_exit_hook(struct lwp *, void *);
    131 
    132 /*
    133  * Fill in the appropriate signal information, and signal the parent.
    134  */
    135 static void
    136 exit_psignal(struct proc *p, struct proc *pp, ksiginfo_t *ksi)
    137 {
    138 
    139 	(void)memset(ksi, 0, sizeof(ksiginfo_t));
    140 	if ((ksi->ksi_signo = P_EXITSIG(p)) == SIGCHLD) {
    141 		if (WIFSIGNALED(p->p_xstat)) {
    142 			if (WCOREDUMP(p->p_xstat))
    143 				ksi->ksi_code = CLD_DUMPED;
    144 			else
    145 				ksi->ksi_code = CLD_KILLED;
    146 		} else {
    147 			ksi->ksi_code = CLD_EXITED;
    148 		}
    149 	}
    150 	/*
    151 	 * we fill those in, even for non-SIGCHLD.
    152 	 */
    153 	ksi->ksi_pid = p->p_pid;
    154 	ksi->ksi_uid = p->p_ucred->cr_uid;
    155 	ksi->ksi_status = p->p_xstat;
    156 	/* XXX: is this still valid? */
    157 	ksi->ksi_utime = p->p_ru->ru_utime.tv_sec;
    158 	ksi->ksi_stime = p->p_ru->ru_stime.tv_sec;
    159 }
    160 
    161 /*
    162  * exit --
    163  *	Death of process.
    164  */
    165 int
    166 sys_exit(struct lwp *l, void *v, register_t *retval)
    167 {
    168 	struct sys_exit_args /* {
    169 		syscallarg(int)	rval;
    170 	} */ *uap = v;
    171 
    172 	/* Don't call exit1() multiple times in the same process.*/
    173 	if (l->l_proc->p_flag & P_WEXIT)
    174 		lwp_exit(l);
    175 
    176 	exit1(l, W_EXITCODE(SCARG(uap, rval), 0));
    177 	/* NOTREACHED */
    178 	return (0);
    179 }
    180 
    181 /*
    182  * Exit: deallocate address space and other resources, change proc state
    183  * to zombie, and unlink proc from allproc and parent's lists.  Save exit
    184  * status and rusage for wait().  Check for child processes and orphan them.
    185  */
    186 void
    187 exit1(struct lwp *l, int rv)
    188 {
    189 	struct proc	*p, *q, *nq;
    190 	int		s, sa;
    191 	struct plimit	*plim;
    192 	struct pstats	*pstats;
    193 	struct sigacts	*ps;
    194 	ksiginfo_t	ksi;
    195 	int		do_psignal = 0;
    196 
    197 	p = l->l_proc;
    198 
    199 	if (__predict_false(p == initproc))
    200 		panic("init died (signal %d, exit %d)",
    201 		    WTERMSIG(rv), WEXITSTATUS(rv));
    202 
    203 	p->p_flag |= P_WEXIT;
    204 	if (p->p_flag & P_STOPEXIT) {
    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->p_cwdi);
    267 	p->p_cwdi = 0;
    268 
    269 	doexithooks(p);
    270 
    271 	if (SESS_LEADER(p)) {
    272 		struct session *sp = p->p_session;
    273 		struct tty *tp;
    274 
    275 		if (sp->s_ttyvp) {
    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 	 * Free the VM resources we're still holding on to.
    335 	 * We must do this from a valid thread because doing
    336 	 * so may block. This frees vmspace, which we don't
    337 	 * need anymore. The only remaining lwp is the one
    338 	 * we run at this moment, nothing runs in userland
    339 	 * anymore.
    340 	 */
    341 	uvm_proc_exit(p);
    342 
    343 	/*
    344 	 * Give machine-dependent code a chance to free any
    345 	 * MD LWP resources while we can still block. This must be done
    346 	 * before uvm_lwp_exit(), in case these resources are in the
    347 	 * PCB.
    348 	 * THIS IS LAST BLOCKING OPERATION.
    349 	 */
    350 #ifndef __NO_CPU_LWP_FREE
    351 	cpu_lwp_free(l, 1);
    352 #endif
    353 
    354 	pmap_deactivate(l);
    355 
    356 	/*
    357 	 * NOTE: WE ARE NO LONGER ALLOWED TO SLEEP!
    358 	 */
    359 
    360 	/*
    361 	 * Save exit status and final rusage info, adding in child rusage
    362 	 * info and self times.
    363 	 * In order to pick up the time for the current execution, we must
    364 	 * do this before unlinking the lwp from l_list.
    365 	 */
    366 	p->p_xstat = rv;
    367 	*p->p_ru = p->p_stats->p_ru;
    368 	calcru(p, &p->p_ru->ru_utime, &p->p_ru->ru_stime, NULL);
    369 	ruadd(p->p_ru, &p->p_stats->p_cru);
    370 
    371 	/*
    372 	 * Notify interested parties of our demise.
    373 	 */
    374 	KNOTE(&p->p_klist, NOTE_EXIT);
    375 
    376 #if PERFCTRS
    377 	/*
    378 	 * Save final PMC information in parent process & clean up.
    379 	 */
    380 	if (PMC_ENABLED(p)) {
    381 		pmc_save_context(p);
    382 		pmc_accumulate(p->p_pptr, p);
    383 		pmc_process_exit(p);
    384 	}
    385 #endif
    386 
    387 	s = proclist_lock_write();
    388 	/*
    389 	 * Reset p_opptr pointer of all former children which got
    390 	 * traced by another process and were reparented. We reset
    391 	 * it to NULL here; the trace detach code then reparents
    392 	 * the child to initproc. We only check allproc list, since
    393 	 * eventual former children on zombproc list won't reference
    394 	 * p_opptr anymore.
    395 	 */
    396 	if (p->p_flag & P_CHTRACED) {
    397 		PROCLIST_FOREACH(q, &allproc) {
    398 			if (q->p_opptr == p)
    399 				q->p_opptr = NULL;
    400 		}
    401 	}
    402 
    403 	/*
    404 	 * Give orphaned children to init(8).
    405 	 */
    406 	q = LIST_FIRST(&p->p_children);
    407 	if (q)		/* only need this if any child is SZOMB */
    408 		wakeup(initproc);
    409 	for (; q != NULL; q = nq) {
    410 		nq = LIST_NEXT(q, p_sibling);
    411 
    412 		/*
    413 		 * Traced processes are killed since their existence
    414 		 * means someone is screwing up. Since we reset the
    415 		 * trace flags, the logic in sys_wait4() would not be
    416 		 * triggered to reparent the process to its
    417 		 * original parent, so we must do this here.
    418 		 */
    419 		if (q->p_flag & P_TRACED) {
    420 			if (q->p_opptr != q->p_pptr) {
    421 				struct proc *t = q->p_opptr;
    422 				proc_reparent(q, t ? t : initproc);
    423 				q->p_opptr = NULL;
    424 			} else
    425 				proc_reparent(q, initproc);
    426 			q->p_flag &= ~(P_TRACED|P_WAITED|P_FSTRACE);
    427 			psignal(q, SIGKILL);
    428 		} else {
    429 			proc_reparent(q, initproc);
    430 		}
    431 	}
    432 
    433 	/*
    434 	 * Move proc from allproc to zombproc, it's now ready
    435 	 * to be collected by parent. Remaining lwp resources
    436 	 * will be freed in lwp_exit2() once we've switch to idle
    437 	 * context.
    438 	 * Changing the state to SZOMB stops it being found by pfind().
    439 	 */
    440 	LIST_REMOVE(p, p_list);
    441 	LIST_INSERT_HEAD(&zombproc, p, p_list);
    442 	p->p_stat = SZOMB;
    443 
    444 	LIST_REMOVE(l, l_list);
    445 	LIST_REMOVE(l, l_sibling);
    446 	l->l_flag |= L_DETACHED|L_PROCEXIT;	/* detached from proc too */
    447 	l->l_stat = LSDEAD;
    448 
    449 	KASSERT(p->p_nrlwps == 1);
    450 	KASSERT(p->p_nlwps == 1);
    451 	p->p_nrlwps--;
    452 	p->p_nlwps--;
    453 
    454 	/* Put in front of parent's sibling list for parent to collect it */
    455 	q = p->p_pptr;
    456 	q->p_nstopchild++;
    457 	if (LIST_FIRST(&q->p_children) != p) {
    458 		/* Put child where it can be found quickly */
    459 		LIST_REMOVE(p, p_sibling);
    460 		LIST_INSERT_HEAD(&q->p_children, p, p_sibling);
    461 	}
    462 
    463 	/*
    464 	 * Notify parent that we're gone.  If parent has the P_NOCLDWAIT
    465 	 * flag set, notify init instead (and hope it will handle
    466 	 * this situation).
    467 	 */
    468 	if (q->p_flag & (P_NOCLDWAIT|P_CLDSIGIGN)) {
    469 		proc_reparent(p, initproc);
    470 
    471 		/*
    472 		 * If this was the last child of our parent, notify
    473 		 * parent, so in case he was wait(2)ing, he will
    474 		 * continue.
    475 		 */
    476 		if (LIST_FIRST(&q->p_children) == NULL)
    477 			wakeup(q);
    478 	}
    479 
    480 	/*
    481 	 * Clear curlwp after we've done all operations
    482 	 * that could block, and before tearing down the rest
    483 	 * of the process state that might be used from clock, etc.
    484 	 * Also, can't clear curlwp while we're still runnable,
    485 	 * as we're not on a run queue (we are current, just not
    486 	 * a proper proc any longer!).
    487 	 *
    488 	 * Other substructures are freed from wait().
    489 	 */
    490 	curlwp = NULL;
    491 
    492 	/* Delay release until after dropping the proclist lock */
    493 	plim = p->p_limit;
    494 	pstats = p->p_stats;
    495 	ps = p->p_sigacts;
    496 
    497 	p->p_limit = NULL;
    498 	p->p_stats = NULL;
    499 	p->p_sigacts = NULL;
    500 
    501 	/* Reload parent pointer, since p may have been reparented above */
    502 	q = p->p_pptr;
    503 
    504 	if ((p->p_flag & P_FSTRACE) == 0 && p->p_exitsig != 0) {
    505 		exit_psignal(p, q, &ksi);
    506 		do_psignal = 1;
    507 	}
    508 
    509 	/*
    510 	 * Once we release the proclist lock, we shouldn't touch the
    511 	 * process structure anymore, since it's now on the zombie
    512 	 * list and available for collection by the parent.
    513 	 */
    514 	proclist_unlock_write(s);
    515 
    516 	if (do_psignal)
    517 		kpsignal(q, &ksi, NULL);
    518 
    519 	/* Wake up the parent so it can get exit status. */
    520 	wakeup(q);
    521 
    522 	/* Release substructures */
    523 	sigactsfree(ps);
    524 	limfree(plim);
    525 	pstatsfree(pstats);
    526 
    527 #ifdef DEBUG
    528 	/* Nothing should use the process link anymore */
    529 	l->l_proc = NULL;
    530 #endif
    531 
    532 	/* This process no longer needs to hold the kernel lock. */
    533 	KERNEL_PROC_UNLOCK(l);
    534 
    535 	/*
    536 	 * Finally, call machine-dependent code to switch to a new
    537 	 * context (possibly the idle context).  Once we are no longer
    538 	 * using the dead lwp's stack, lwp_exit2() will be called
    539 	 * to arrange for the resources to be released.
    540 	 *
    541 	 * Note that cpu_exit() will end with a call equivalent to
    542 	 * cpu_switch(), finishing our execution (pun intended).
    543 	 */
    544 
    545 	uvmexp.swtch++;
    546 	cpu_exit(l);
    547 }
    548 
    549 void
    550 exit_lwps(struct lwp *l)
    551 {
    552 	struct proc *p;
    553 	struct lwp *l2;
    554 	struct sadata_vp *vp;
    555 	int s, error;
    556 	lwpid_t		waited;
    557 
    558 	p = l->l_proc;
    559 
    560 	/* XXX SMP
    561 	 * This would be the right place to IPI any LWPs running on
    562 	 * other processors so that they can notice the userret exit hook.
    563 	 */
    564 	p->p_userret = lwp_exit_hook;
    565 	p->p_userret_arg = NULL;
    566 
    567 	if (p->p_sa) {
    568 		SLIST_FOREACH(vp, &p->p_sa->sa_vps, savp_next) {
    569 			/*
    570 			 * Make SA-cached LWPs normal process runnable
    571 			 * LWPs so that they'll also self-destruct.
    572 			 */
    573 			DPRINTF(("exit_lwps: Making cached LWPs of %d on VP %d runnable: ",
    574 				    p->p_pid, vp->savp_id));
    575 			SCHED_LOCK(s);
    576 			while ((l2 = sa_getcachelwp(vp)) != 0) {
    577 				l2->l_priority = l2->l_usrpri;
    578 				setrunnable(l2);
    579 				DPRINTF(("%d ", l2->l_lid));
    580 			}
    581 			SCHED_UNLOCK(s);
    582 			DPRINTF(("\n"));
    583 
    584 			/*
    585 			 * Clear wokenq, the LWPs on the queue will
    586 			 * run below.
    587 			 */
    588 			vp->savp_wokenq_head = NULL;
    589 		}
    590 	}
    591 
    592 retry:
    593 	/*
    594 	 * Interrupt LWPs in interruptable sleep, unsuspend suspended
    595 	 * LWPs, make detached LWPs undetached (so we can wait for
    596 	 * them) and then wait for everyone else to finish.
    597 	 */
    598 	LIST_FOREACH(l2, &p->p_lwps, l_sibling) {
    599 		l2->l_flag &= ~(L_DETACHED|L_SA);
    600 
    601 		SCHED_LOCK(s);
    602 		if ((l2->l_stat == LSSLEEP && (l2->l_flag & L_SINTR)) ||
    603 		    l2->l_stat == LSSUSPENDED || l2->l_stat == LSSTOP) {
    604 			setrunnable(l2);
    605 			DPRINTF(("exit_lwps: Made %d.%d runnable\n",
    606 			    p->p_pid, l2->l_lid));
    607 		}
    608 		SCHED_UNLOCK(s);
    609 	}
    610 
    611 
    612 	while (p->p_nlwps > 1) {
    613 		DPRINTF(("exit_lwps: waiting for %d LWPs (%d runnable, %d zombies)\n",
    614 		    p->p_nlwps, p->p_nrlwps, p->p_nzlwps));
    615 		error = lwp_wait1(l, 0, &waited, LWPWAIT_EXITCONTROL);
    616 		if (error == EDEADLK) {
    617 			/*
    618 			 * LWPs can get suspended/slept behind us.
    619 			 * (eg. sa_setwoken)
    620 			 * kick them again and retry.
    621 			 */
    622 			goto retry;
    623 		}
    624 		if (error)
    625 			panic("exit_lwps: lwp_wait1 failed with error %d",
    626 			    error);
    627 		DPRINTF(("exit_lwps: Got LWP %d from lwp_wait1()\n", waited));
    628 	}
    629 
    630 	p->p_userret = NULL;
    631 }
    632 
    633 /* Wrapper function for use in p_userret */
    634 static void
    635 lwp_exit_hook(struct lwp *l, void *arg)
    636 {
    637 	KERNEL_PROC_LOCK(l);
    638 	lwp_exit(l);
    639 }
    640 
    641 int
    642 sys_wait4(struct lwp *l, void *v, register_t *retval)
    643 {
    644 	struct sys_wait4_args /* {
    645 		syscallarg(int)			pid;
    646 		syscallarg(int *)		status;
    647 		syscallarg(int)			options;
    648 		syscallarg(struct rusage *)	rusage;
    649 	} */ *uap = v;
    650 	struct proc	*child, *parent;
    651 	int		status, error;
    652 
    653 	parent = l->l_proc;
    654 
    655 	if (SCARG(uap, pid) == 0)
    656 		SCARG(uap, pid) = -parent->p_pgid;
    657 	if (SCARG(uap, options) & ~(WUNTRACED|WNOHANG|WALTSIG|WALLSIG))
    658 		return (EINVAL);
    659 
    660 	error = find_stopped_child(parent, SCARG(uap,pid), SCARG(uap,options),
    661 				&child);
    662 	if (error != 0)
    663 		return error;
    664 	if (child == NULL) {
    665 		*retval = 0;
    666 		return 0;
    667 	}
    668 
    669 	/*
    670 	 * Collect child u-areas.
    671 	 */
    672 	uvm_uarea_drain(FALSE);
    673 
    674 	retval[0] = child->p_pid;
    675 
    676 	if (P_ZOMBIE(child)) {
    677 		if (SCARG(uap, status)) {
    678 			status = child->p_xstat;	/* convert to int */
    679 			error = copyout(&status, SCARG(uap, status),
    680 					sizeof(status));
    681 			if (error)
    682 				return (error);
    683 		}
    684 		if (SCARG(uap, rusage)) {
    685 			error = copyout(child->p_ru, SCARG(uap, rusage),
    686 				    sizeof(struct rusage));
    687 			if (error)
    688 				return (error);
    689 		}
    690 
    691 		proc_free(child);
    692 		return 0;
    693 	}
    694 
    695 	/* child state must be SSTOP */
    696 	if (SCARG(uap, status)) {
    697 		status = W_STOPCODE(child->p_xstat);
    698 		return copyout(&status, SCARG(uap, status), sizeof(status));
    699 	}
    700 	return 0;
    701 }
    702 
    703 /*
    704  * Scan list of child processes for a child process that has stopped or
    705  * exited.  Used by sys_wait4 and 'compat' equivalents.
    706  */
    707 int
    708 find_stopped_child(struct proc *parent, pid_t pid, int options,
    709 	struct proc **child_p)
    710 {
    711 	struct proc *child;
    712 	int error;
    713 
    714 	for (;;) {
    715 		proclist_lock_read();
    716 		error = ECHILD;
    717 		LIST_FOREACH(child, &parent->p_children, p_sibling) {
    718 			if (pid >= 0) {
    719 				if (child->p_pid != pid) {
    720 					child = p_find(pid, PFIND_ZOMBIE |
    721 								PFIND_LOCKED);
    722 					if (child == NULL
    723 					    || child->p_pptr != parent) {
    724 						child = NULL;
    725 						break;
    726 					}
    727 				}
    728 			} else
    729 				if (pid != WAIT_ANY && child->p_pgid != -pid)
    730 					/* child not in correct pgrp */
    731 					continue;
    732 			/*
    733 			 * Wait for processes with p_exitsig != SIGCHLD
    734 			 * processes only if WALTSIG is set; wait for
    735 			 * processes with p_exitsig == SIGCHLD only
    736 			 * if WALTSIG is clear.
    737 			 */
    738 			if (((options & WALLSIG) == 0) &&
    739 			    (options & WALTSIG ? child->p_exitsig == SIGCHLD
    740 						: P_EXITSIG(child) != SIGCHLD)){
    741 				if (child->p_pid == pid) {
    742 					child = NULL;
    743 					break;
    744 				}
    745 				continue;
    746 			}
    747 
    748 			error = 0;
    749 			if (child->p_stat == SZOMB && !(options & WNOZOMBIE))
    750 				break;
    751 
    752 			if (child->p_stat == SSTOP &&
    753 			    (child->p_flag & P_WAITED) == 0 &&
    754 			    (child->p_flag & P_TRACED || options & WUNTRACED)) {
    755 				if ((options & WNOWAIT) == 0) {
    756 					child->p_flag |= P_WAITED;
    757 					parent->p_nstopchild--;
    758 				}
    759 				break;
    760 			}
    761 			if (parent->p_nstopchild == 0 || child->p_pid == pid) {
    762 				child = NULL;
    763 				break;
    764 			}
    765 		}
    766 		proclist_unlock_read();
    767 		if (child != NULL || error != 0 || options & WNOHANG) {
    768 			*child_p = child;
    769 			return error;
    770 		}
    771 		error = tsleep(parent, PWAIT | PCATCH, "wait", 0);
    772 		if (error != 0)
    773 			return error;
    774 	}
    775 }
    776 
    777 /*
    778  * Free a process after parent has taken all the state info.
    779  */
    780 void
    781 proc_free(struct proc *p)
    782 {
    783 	struct proc *parent = p->p_pptr;
    784 	ksiginfo_t ksi;
    785 	int s;
    786 
    787 	KASSERT(p->p_nlwps == 0);
    788 	KASSERT(p->p_nzlwps == 0);
    789 	KASSERT(p->p_nrlwps == 0);
    790 	KASSERT(LIST_EMPTY(&p->p_lwps));
    791 
    792 	/*
    793 	 * If we got the child via ptrace(2) or procfs, and
    794 	 * the parent is different (meaning the process was
    795 	 * attached, rather than run as a child), then we need
    796 	 * to give it back to the old parent, and send the
    797 	 * parent the exit signal.  The rest of the cleanup
    798 	 * will be done when the old parent waits on the child.
    799 	 */
    800 	if ((p->p_flag & P_TRACED) && p->p_opptr != parent){
    801 		parent = p->p_opptr;
    802 		if (parent == NULL)
    803 			parent = initproc;
    804 		proc_reparent(p, parent);
    805 		p->p_opptr = NULL;
    806 		p->p_flag &= ~(P_TRACED|P_WAITED|P_FSTRACE);
    807 		if (p->p_exitsig != 0) {
    808 			exit_psignal(p, parent, &ksi);
    809 			kpsignal(parent, &ksi, NULL);
    810 		}
    811 		wakeup(parent);
    812 		return;
    813 	}
    814 
    815 	scheduler_wait_hook(parent, p);
    816 	p->p_xstat = 0;
    817 
    818 	ruadd(&parent->p_stats->p_cru, p->p_ru);
    819 
    820 	/*
    821 	 * At this point we are going to start freeing the final resources.
    822 	 * If anyone tries to access the proc structure after here they
    823 	 * will get a shock - bits are missing.
    824 	 * Attempt to make it hard!
    825 	 */
    826 
    827 	p->p_stat = SIDL;		/* not even a zombie any more */
    828 
    829 	pool_put(&rusage_pool, p->p_ru);
    830 
    831 	/*
    832 	 * Finally finished with old proc entry.
    833 	 * Unlink it from its process group and free it.
    834 	 */
    835 	leavepgrp(p);
    836 
    837 	s = proclist_lock_write();
    838 	LIST_REMOVE(p, p_list);	/* off zombproc */
    839 	p->p_pptr->p_nstopchild--;
    840 	LIST_REMOVE(p, p_sibling);
    841 	proclist_unlock_write(s);
    842 
    843 	/*
    844 	 * Decrement the count of procs running with this uid.
    845 	 */
    846 	(void)chgproccnt(p->p_cred->p_ruid, -1);
    847 
    848 	/*
    849 	 * Free up credentials.
    850 	 */
    851 	if (--p->p_cred->p_refcnt == 0) {
    852 		crfree(p->p_cred->pc_ucred);
    853 		pool_put(&pcred_pool, p->p_cred);
    854 	}
    855 
    856 	/*
    857 	 * Release reference to text vnode
    858 	 */
    859 	if (p->p_textvp)
    860 		vrele(p->p_textvp);
    861 
    862 	/* Release any SA state. */
    863 	if (p->p_sa)
    864 		sa_release(p);
    865 
    866 	/* Free proc structure and let pid be reallocated */
    867 	proc_free_mem(p);
    868 }
    869 
    870 /*
    871  * make process 'parent' the new parent of process 'child'.
    872  *
    873  * Must be called with proclist_lock_write() held.
    874  */
    875 void
    876 proc_reparent(struct proc *child, struct proc *parent)
    877 {
    878 
    879 	if (child->p_pptr == parent)
    880 		return;
    881 
    882 	if (child->p_stat == SZOMB
    883 	    || (child->p_stat == SSTOP && !(child->p_flag & P_WAITED))) {
    884 		child->p_pptr->p_nstopchild--;
    885 		parent->p_nstopchild++;
    886 	}
    887 	if (parent == initproc)
    888 		child->p_exitsig = SIGCHLD;
    889 
    890 	LIST_REMOVE(child, p_sibling);
    891 	LIST_INSERT_HEAD(&parent->p_children, child, p_sibling);
    892 	child->p_pptr = parent;
    893 }
    894