Home | History | Annotate | Line # | Download | only in kern
kern_exit.c revision 1.156
      1 /*	$NetBSD: kern_exit.c,v 1.156 2006/05/14 21:15:11 elad 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 1.156 2006/05/14 21:15:11 elad 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 #include <sys/kauth.h>
    117 
    118 #include <machine/cpu.h>
    119 
    120 #include <uvm/uvm_extern.h>
    121 
    122 #define DEBUG_EXIT
    123 
    124 #ifdef DEBUG_EXIT
    125 int debug_exit = 0;
    126 #define DPRINTF(x) if (debug_exit) printf x
    127 #else
    128 #define DPRINTF(x)
    129 #endif
    130 
    131 static void lwp_exit_hook(struct lwp *, void *);
    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, ksiginfo_t *ksi)
    138 {
    139 
    140 	(void)memset(ksi, 0, sizeof(ksiginfo_t));
    141 	if ((ksi->ksi_signo = P_EXITSIG(p)) == SIGCHLD) {
    142 		if (WIFSIGNALED(p->p_xstat)) {
    143 			if (WCOREDUMP(p->p_xstat))
    144 				ksi->ksi_code = CLD_DUMPED;
    145 			else
    146 				ksi->ksi_code = CLD_KILLED;
    147 		} else {
    148 			ksi->ksi_code = CLD_EXITED;
    149 		}
    150 	}
    151 	/*
    152 	 * we fill those in, even for non-SIGCHLD.
    153 	 */
    154 	ksi->ksi_pid = p->p_pid;
    155 	ksi->ksi_uid = kauth_cred_geteuid(p->p_cred);
    156 	ksi->ksi_status = p->p_xstat;
    157 	/* XXX: is this still valid? */
    158 	ksi->ksi_utime = p->p_ru->ru_utime.tv_sec;
    159 	ksi->ksi_stime = p->p_ru->ru_stime.tv_sec;
    160 }
    161 
    162 /*
    163  * exit --
    164  *	Death of process.
    165  */
    166 int
    167 sys_exit(struct lwp *l, void *v, register_t *retval)
    168 {
    169 	struct sys_exit_args /* {
    170 		syscallarg(int)	rval;
    171 	} */ *uap = v;
    172 
    173 	/* Don't call exit1() multiple times in the same process.*/
    174 	if (l->l_proc->p_flag & P_WEXIT)
    175 		lwp_exit(l);
    176 
    177 	exit1(l, W_EXITCODE(SCARG(uap, rval), 0));
    178 	/* NOTREACHED */
    179 	return (0);
    180 }
    181 
    182 /*
    183  * Exit: deallocate address space and other resources, change proc state
    184  * to zombie, and unlink proc from allproc and parent's lists.  Save exit
    185  * status and rusage for wait().  Check for child processes and orphan them.
    186  */
    187 void
    188 exit1(struct lwp *l, int rv)
    189 {
    190 	struct proc	*p, *q, *nq;
    191 	int		s, sa;
    192 	struct plimit	*plim;
    193 	struct pstats	*pstats;
    194 	struct sigacts	*ps;
    195 	ksiginfo_t	ksi;
    196 	int		do_psignal = 0;
    197 
    198 	p = l->l_proc;
    199 
    200 	if (__predict_false(p == initproc))
    201 		panic("init died (signal %d, exit %d)",
    202 		    WTERMSIG(rv), WEXITSTATUS(rv));
    203 
    204 	p->p_flag |= P_WEXIT;
    205 	if (p->p_flag & P_STOPEXIT) {
    206 		sigminusset(&contsigmask, &p->p_sigctx.ps_siglist);
    207 		SCHED_LOCK(s);
    208 		p->p_stat = SSTOP;
    209 		l->l_stat = LSSTOP;
    210 		p->p_nrlwps--;
    211 		mi_switch(l, NULL);
    212 		SCHED_ASSERT_UNLOCKED();
    213 		splx(s);
    214 	}
    215 
    216 	DPRINTF(("exit1: %d.%d exiting.\n", p->p_pid, l->l_lid));
    217 	/*
    218 	 * Disable scheduler activation upcalls.
    219 	 * We're trying to get out of here.
    220 	 */
    221 	sa = 0;
    222 	if (p->p_sa != NULL) {
    223 		l->l_flag &= ~L_SA;
    224 #if 0
    225 		p->p_flag &= ~P_SA;
    226 #endif
    227 		sa = 1;
    228 	}
    229 
    230 #ifdef PGINPROF
    231 	vmsizmon();
    232 #endif
    233 	if (p->p_flag & P_PROFIL)
    234 		stopprofclock(p);
    235 	p->p_ru = pool_get(&rusage_pool, PR_WAITOK);
    236 	/*
    237 	 * If parent is waiting for us to exit or exec, P_PPWAIT is set; we
    238 	 * wake up the parent early to avoid deadlock.
    239 	 */
    240 	if (p->p_flag & P_PPWAIT) {
    241 		p->p_flag &= ~P_PPWAIT;
    242 		wakeup(p->p_pptr);
    243 	}
    244 	sigfillset(&p->p_sigctx.ps_sigignore);
    245 	sigemptyset(&p->p_sigctx.ps_siglist);
    246 	p->p_sigctx.ps_sigcheck = 0;
    247 	timers_free(p, TIMERS_ALL);
    248 
    249 	if (sa || (p->p_nlwps > 1)) {
    250 		exit_lwps(l);
    251 
    252 		/*
    253 		 * Collect thread u-areas.
    254 		 */
    255 		uvm_uarea_drain(FALSE);
    256 	}
    257 
    258 #if defined(__HAVE_RAS)
    259 	ras_purgeall(p);
    260 #endif
    261 
    262 	/*
    263 	 * Close open files and release open-file table.
    264 	 * This may block!
    265 	 */
    266 	fdfree(l);
    267 	cwdfree(p->p_cwdi);
    268 	p->p_cwdi = 0;
    269 
    270 	doexithooks(p);
    271 
    272 	if (SESS_LEADER(p)) {
    273 		struct session *sp = p->p_session;
    274 		struct tty *tp;
    275 
    276 		if (sp->s_ttyvp) {
    277 			/*
    278 			 * Controlling process.
    279 			 * Signal foreground pgrp,
    280 			 * drain controlling terminal
    281 			 * and revoke access to controlling terminal.
    282 			 */
    283 			tp = sp->s_ttyp;
    284 			s = spltty();
    285 			TTY_LOCK(tp);
    286 			if (tp->t_session == sp) {
    287 				if (tp->t_pgrp)
    288 					pgsignal(tp->t_pgrp, SIGHUP, 1);
    289 				/* we can't guarantee the revoke will do this */
    290 				tp->t_pgrp = NULL;
    291 				tp->t_session = NULL;
    292 				TTY_UNLOCK(tp);
    293 				splx(s);
    294 				SESSRELE(sp);
    295 				(void) ttywait(tp);
    296 				/*
    297 				 * The tty could have been revoked
    298 				 * if we blocked.
    299 				 */
    300 				if (sp->s_ttyvp)
    301 					VOP_REVOKE(sp->s_ttyvp, REVOKEALL);
    302 			} else {
    303 				TTY_UNLOCK(tp);
    304 				splx(s);
    305 			}
    306 			if (sp->s_ttyvp)
    307 				vrele(sp->s_ttyvp);
    308 			sp->s_ttyvp = NULL;
    309 			/*
    310 			 * s_ttyp is not zero'd; we use this to indicate
    311 			 * that the session once had a controlling terminal.
    312 			 * (for logging and informational purposes)
    313 			 */
    314 		}
    315 		sp->s_leader = NULL;
    316 	}
    317 	fixjobc(p, p->p_pgrp, 0);
    318 	(void)acct_process(l);
    319 #ifdef KTRACE
    320 	/*
    321 	 * release trace file
    322 	 */
    323 	ktrderef(p);
    324 #endif
    325 #ifdef SYSTRACE
    326 	systrace_sys_exit(p);
    327 #endif
    328 	/*
    329 	 * If emulation has process exit hook, call it now.
    330 	 */
    331 	if (p->p_emul->e_proc_exit)
    332 		(*p->p_emul->e_proc_exit)(p);
    333 
    334 	/*
    335 	 * Free the VM resources we're still holding on to.
    336 	 * We must do this from a valid thread because doing
    337 	 * so may block. This frees vmspace, which we don't
    338 	 * need anymore. The only remaining lwp is the one
    339 	 * we run at this moment, nothing runs in userland
    340 	 * anymore.
    341 	 */
    342 	uvm_proc_exit(p);
    343 
    344 	/*
    345 	 * Give machine-dependent code a chance to free any
    346 	 * MD LWP resources while we can still block. This must be done
    347 	 * before uvm_lwp_exit(), in case these resources are in the
    348 	 * PCB.
    349 	 * THIS IS LAST BLOCKING OPERATION.
    350 	 */
    351 #ifndef __NO_CPU_LWP_FREE
    352 	cpu_lwp_free(l, 1);
    353 #endif
    354 
    355 	pmap_deactivate(l);
    356 
    357 	/*
    358 	 * NOTE: WE ARE NO LONGER ALLOWED TO SLEEP!
    359 	 */
    360 
    361 	/*
    362 	 * Save exit status and final rusage info, adding in child rusage
    363 	 * info and self times.
    364 	 * In order to pick up the time for the current execution, we must
    365 	 * do this before unlinking the lwp from l_list.
    366 	 */
    367 	p->p_xstat = rv;
    368 	*p->p_ru = p->p_stats->p_ru;
    369 	calcru(p, &p->p_ru->ru_utime, &p->p_ru->ru_stime, NULL);
    370 	ruadd(p->p_ru, &p->p_stats->p_cru);
    371 
    372 	/*
    373 	 * Notify interested parties of our demise.
    374 	 */
    375 	KNOTE(&p->p_klist, NOTE_EXIT);
    376 
    377 #if PERFCTRS
    378 	/*
    379 	 * Save final PMC information in parent process & clean up.
    380 	 */
    381 	if (PMC_ENABLED(p)) {
    382 		pmc_save_context(p);
    383 		pmc_accumulate(p->p_pptr, p);
    384 		pmc_process_exit(p);
    385 	}
    386 #endif
    387 
    388 	s = proclist_lock_write();
    389 	/*
    390 	 * Reset p_opptr pointer of all former children which got
    391 	 * traced by another process and were reparented. We reset
    392 	 * it to NULL here; the trace detach code then reparents
    393 	 * the child to initproc. We only check allproc list, since
    394 	 * eventual former children on zombproc list won't reference
    395 	 * p_opptr anymore.
    396 	 */
    397 	if (p->p_flag & P_CHTRACED) {
    398 		PROCLIST_FOREACH(q, &allproc) {
    399 			if (q->p_opptr == p)
    400 				q->p_opptr = NULL;
    401 		}
    402 	}
    403 
    404 	/*
    405 	 * Give orphaned children to init(8).
    406 	 */
    407 	q = LIST_FIRST(&p->p_children);
    408 	if (q)		/* only need this if any child is SZOMB */
    409 		wakeup(initproc);
    410 	for (; q != NULL; q = nq) {
    411 		nq = LIST_NEXT(q, p_sibling);
    412 
    413 		/*
    414 		 * Traced processes are killed since their existence
    415 		 * means someone is screwing up. Since we reset the
    416 		 * trace flags, the logic in sys_wait4() would not be
    417 		 * triggered to reparent the process to its
    418 		 * original parent, so we must do this here.
    419 		 */
    420 		if (q->p_flag & P_TRACED) {
    421 			if (q->p_opptr != q->p_pptr) {
    422 				struct proc *t = q->p_opptr;
    423 				proc_reparent(q, t ? t : initproc);
    424 				q->p_opptr = NULL;
    425 			} else
    426 				proc_reparent(q, initproc);
    427 			q->p_flag &= ~(P_TRACED|P_WAITED|P_FSTRACE|P_SYSCALL);
    428 			killproc(q, "orphaned traced process");
    429 		} else {
    430 			proc_reparent(q, initproc);
    431 		}
    432 	}
    433 
    434 	/*
    435 	 * Move proc from allproc to zombproc, it's now ready
    436 	 * to be collected by parent. Remaining lwp resources
    437 	 * will be freed in lwp_exit2() once we've switch to idle
    438 	 * context.
    439 	 * Changing the state to SZOMB stops it being found by pfind().
    440 	 */
    441 	LIST_REMOVE(p, p_list);
    442 	LIST_INSERT_HEAD(&zombproc, p, p_list);
    443 	p->p_stat = SZOMB;
    444 
    445 	LIST_REMOVE(l, l_list);
    446 	LIST_REMOVE(l, l_sibling);
    447 	l->l_flag |= L_DETACHED;	/* detached from proc too */
    448 	l->l_stat = LSDEAD;
    449 
    450 	KASSERT(p->p_nrlwps == 1);
    451 	KASSERT(p->p_nlwps == 1);
    452 	p->p_nrlwps--;
    453 	p->p_nlwps--;
    454 
    455 	/* Put in front of parent's sibling list for parent to collect it */
    456 	q = p->p_pptr;
    457 	q->p_nstopchild++;
    458 	if (LIST_FIRST(&q->p_children) != p) {
    459 		/* Put child where it can be found quickly */
    460 		LIST_REMOVE(p, p_sibling);
    461 		LIST_INSERT_HEAD(&q->p_children, p, p_sibling);
    462 	}
    463 
    464 	/*
    465 	 * Notify parent that we're gone.  If parent has the P_NOCLDWAIT
    466 	 * flag set, notify init instead (and hope it will handle
    467 	 * this situation).
    468 	 */
    469 	if (q->p_flag & (P_NOCLDWAIT|P_CLDSIGIGN)) {
    470 		proc_reparent(p, initproc);
    471 
    472 		/*
    473 		 * If this was the last child of our parent, notify
    474 		 * parent, so in case he was wait(2)ing, he will
    475 		 * continue.
    476 		 */
    477 		if (LIST_FIRST(&q->p_children) == NULL)
    478 			wakeup(q);
    479 	}
    480 
    481 	/*
    482 	 * Clear curlwp after we've done all operations
    483 	 * that could block, and before tearing down the rest
    484 	 * of the process state that might be used from clock, etc.
    485 	 * Also, can't clear curlwp while we're still runnable,
    486 	 * as we're not on a run queue (we are current, just not
    487 	 * a proper proc any longer!).
    488 	 *
    489 	 * Other substructures are freed from wait().
    490 	 */
    491 	curlwp = NULL;
    492 
    493 	/* Delay release until after dropping the proclist lock */
    494 	plim = p->p_limit;
    495 	pstats = p->p_stats;
    496 	ps = p->p_sigacts;
    497 
    498 	p->p_limit = NULL;
    499 	p->p_stats = NULL;
    500 	p->p_sigacts = NULL;
    501 
    502 	/* Reload parent pointer, since p may have been reparented above */
    503 	q = p->p_pptr;
    504 
    505 	if ((p->p_flag & P_FSTRACE) == 0 && p->p_exitsig != 0) {
    506 		exit_psignal(p, q, &ksi);
    507 		do_psignal = 1;
    508 	}
    509 
    510 	/*
    511 	 * Once we release the proclist lock, we shouldn't touch the
    512 	 * process structure anymore, since it's now on the zombie
    513 	 * list and available for collection by the parent.
    514 	 */
    515 	proclist_unlock_write(s);
    516 
    517 	if (do_psignal)
    518 		kpsignal(q, &ksi, NULL);
    519 
    520 	/* Wake up the parent so it can get exit status. */
    521 	wakeup(q);
    522 
    523 	/* Release substructures */
    524 	sigactsfree(ps);
    525 	limfree(plim);
    526 	pstatsfree(pstats);
    527 
    528 #ifdef DEBUG
    529 	/* Nothing should use the process link anymore */
    530 	l->l_proc = NULL;
    531 #endif
    532 
    533 	/* This process no longer needs to hold the kernel lock. */
    534 	KERNEL_PROC_UNLOCK(l);
    535 
    536 	/*
    537 	 * Finally, call machine-dependent code to switch to a new
    538 	 * context (possibly the idle context).  Once we are no longer
    539 	 * using the dead lwp's stack, lwp_exit2() will be called
    540 	 * to arrange for the resources to be released.
    541 	 *
    542 	 * Note that cpu_exit() will end with a call equivalent to
    543 	 * cpu_switch(), finishing our execution (pun intended).
    544 	 */
    545 
    546 	uvmexp.swtch++;
    547 	cpu_exit(l);
    548 }
    549 
    550 void
    551 exit_lwps(struct lwp *l)
    552 {
    553 	struct proc *p;
    554 	struct lwp *l2;
    555 	struct sadata_vp *vp;
    556 	int s, error;
    557 	lwpid_t		waited;
    558 
    559 	p = l->l_proc;
    560 
    561 	/* XXX SMP
    562 	 * This would be the right place to IPI any LWPs running on
    563 	 * other processors so that they can notice the userret exit hook.
    564 	 */
    565 	p->p_userret = lwp_exit_hook;
    566 	p->p_userret_arg = NULL;
    567 
    568 	if (p->p_sa) {
    569 		SLIST_FOREACH(vp, &p->p_sa->sa_vps, savp_next) {
    570 			/*
    571 			 * Make SA-cached LWPs normal process runnable
    572 			 * LWPs so that they'll also self-destruct.
    573 			 */
    574 			DPRINTF(("exit_lwps: Making cached LWPs of %d on VP %d runnable: ",
    575 				    p->p_pid, vp->savp_id));
    576 			SCHED_LOCK(s);
    577 			while ((l2 = sa_getcachelwp(vp)) != 0) {
    578 				l2->l_priority = l2->l_usrpri;
    579 				setrunnable(l2);
    580 				DPRINTF(("%d ", l2->l_lid));
    581 			}
    582 			SCHED_UNLOCK(s);
    583 			DPRINTF(("\n"));
    584 
    585 			/*
    586 			 * Clear wokenq, the LWPs on the queue will
    587 			 * run below.
    588 			 */
    589 			vp->savp_wokenq_head = NULL;
    590 		}
    591 	}
    592 
    593 retry:
    594 	/*
    595 	 * Interrupt LWPs in interruptable sleep, unsuspend suspended
    596 	 * LWPs, make detached LWPs undetached (so we can wait for
    597 	 * them) and then wait for everyone else to finish.
    598 	 */
    599 	LIST_FOREACH(l2, &p->p_lwps, l_sibling) {
    600 		l2->l_flag &= ~(L_DETACHED|L_SA);
    601 
    602 		SCHED_LOCK(s);
    603 		if ((l2->l_stat == LSSLEEP && (l2->l_flag & L_SINTR)) ||
    604 		    l2->l_stat == LSSUSPENDED || l2->l_stat == LSSTOP) {
    605 			setrunnable(l2);
    606 			DPRINTF(("exit_lwps: Made %d.%d runnable\n",
    607 			    p->p_pid, l2->l_lid));
    608 		}
    609 		SCHED_UNLOCK(s);
    610 	}
    611 
    612 
    613 	while (p->p_nlwps > 1) {
    614 		DPRINTF(("exit_lwps: waiting for %d LWPs (%d runnable, %d zombies)\n",
    615 		    p->p_nlwps, p->p_nrlwps, p->p_nzlwps));
    616 		error = lwp_wait1(l, 0, &waited, LWPWAIT_EXITCONTROL);
    617 		if (error == EDEADLK) {
    618 			/*
    619 			 * LWPs can get suspended/slept behind us.
    620 			 * (eg. sa_setwoken)
    621 			 * kick them again and retry.
    622 			 */
    623 			goto retry;
    624 		}
    625 		if (error)
    626 			panic("exit_lwps: lwp_wait1 failed with error %d",
    627 			    error);
    628 		DPRINTF(("exit_lwps: Got LWP %d from lwp_wait1()\n", waited));
    629 	}
    630 
    631 	p->p_userret = NULL;
    632 }
    633 
    634 /* Wrapper function for use in p_userret */
    635 static void
    636 lwp_exit_hook(struct lwp *l, void *arg)
    637 {
    638 	KERNEL_PROC_LOCK(l);
    639 	lwp_exit(l);
    640 }
    641 
    642 int
    643 sys_wait4(struct lwp *l, void *v, register_t *retval)
    644 {
    645 	struct sys_wait4_args /* {
    646 		syscallarg(int)			pid;
    647 		syscallarg(int *)		status;
    648 		syscallarg(int)			options;
    649 		syscallarg(struct rusage *)	rusage;
    650 	} */ *uap = v;
    651 	struct proc	*child, *parent;
    652 	int		status, error;
    653 
    654 	parent = l->l_proc;
    655 
    656 	if (SCARG(uap, pid) == 0)
    657 		SCARG(uap, pid) = -parent->p_pgid;
    658 	if (SCARG(uap, options) & ~(WUNTRACED|WNOHANG|WALTSIG|WALLSIG))
    659 		return (EINVAL);
    660 
    661 	error = find_stopped_child(parent, SCARG(uap,pid), SCARG(uap,options),
    662 				&child);
    663 	if (error != 0)
    664 		return error;
    665 	if (child == NULL) {
    666 		*retval = 0;
    667 		return 0;
    668 	}
    669 
    670 	/*
    671 	 * Collect child u-areas.
    672 	 */
    673 	uvm_uarea_drain(FALSE);
    674 
    675 	retval[0] = child->p_pid;
    676 
    677 	if (P_ZOMBIE(child)) {
    678 		if (SCARG(uap, status)) {
    679 			status = child->p_xstat;	/* convert to int */
    680 			error = copyout(&status, SCARG(uap, status),
    681 					sizeof(status));
    682 			if (error)
    683 				return (error);
    684 		}
    685 		if (SCARG(uap, rusage)) {
    686 			error = copyout(child->p_ru, SCARG(uap, rusage),
    687 				    sizeof(struct rusage));
    688 			if (error)
    689 				return (error);
    690 		}
    691 
    692 		proc_free(child);
    693 		return 0;
    694 	}
    695 
    696 	/* child state must be SSTOP */
    697 	if (SCARG(uap, status)) {
    698 		status = W_STOPCODE(child->p_xstat);
    699 		return copyout(&status, SCARG(uap, status), sizeof(status));
    700 	}
    701 	return 0;
    702 }
    703 
    704 /*
    705  * Scan list of child processes for a child process that has stopped or
    706  * exited.  Used by sys_wait4 and 'compat' equivalents.
    707  */
    708 int
    709 find_stopped_child(struct proc *parent, pid_t pid, int options,
    710 	struct proc **child_p)
    711 {
    712 	struct proc *child;
    713 	int error;
    714 
    715 	for (;;) {
    716 		proclist_lock_read();
    717 		error = ECHILD;
    718 		LIST_FOREACH(child, &parent->p_children, p_sibling) {
    719 			if (pid >= 0) {
    720 				if (child->p_pid != pid) {
    721 					child = p_find(pid, PFIND_ZOMBIE |
    722 								PFIND_LOCKED);
    723 					if (child == NULL
    724 					    || child->p_pptr != parent) {
    725 						child = NULL;
    726 						break;
    727 					}
    728 				}
    729 			} else
    730 				if (pid != WAIT_ANY && child->p_pgid != -pid)
    731 					/* child not in correct pgrp */
    732 					continue;
    733 			/*
    734 			 * Wait for processes with p_exitsig != SIGCHLD
    735 			 * processes only if WALTSIG is set; wait for
    736 			 * processes with p_exitsig == SIGCHLD only
    737 			 * if WALTSIG is clear.
    738 			 */
    739 			if (((options & WALLSIG) == 0) &&
    740 			    (options & WALTSIG ? child->p_exitsig == SIGCHLD
    741 						: P_EXITSIG(child) != SIGCHLD)){
    742 				if (child->p_pid == pid) {
    743 					child = NULL;
    744 					break;
    745 				}
    746 				continue;
    747 			}
    748 
    749 			error = 0;
    750 			if (child->p_stat == SZOMB && !(options & WNOZOMBIE))
    751 				break;
    752 
    753 			if (child->p_stat == SSTOP &&
    754 			    (child->p_flag & P_WAITED) == 0 &&
    755 			    (child->p_flag & P_TRACED || options & WUNTRACED)) {
    756 				if ((options & WNOWAIT) == 0) {
    757 					child->p_flag |= P_WAITED;
    758 					parent->p_nstopchild--;
    759 				}
    760 				break;
    761 			}
    762 			if (parent->p_nstopchild == 0 || child->p_pid == pid) {
    763 				child = NULL;
    764 				break;
    765 			}
    766 		}
    767 		proclist_unlock_read();
    768 		if (child != NULL || error != 0 || options & WNOHANG) {
    769 			*child_p = child;
    770 			return error;
    771 		}
    772 		error = tsleep(parent, PWAIT | PCATCH, "wait", 0);
    773 		if (error != 0)
    774 			return error;
    775 	}
    776 }
    777 
    778 /*
    779  * Free a process after parent has taken all the state info.
    780  */
    781 void
    782 proc_free(struct proc *p)
    783 {
    784 	struct proc *parent = p->p_pptr;
    785 	ksiginfo_t ksi;
    786 	int s;
    787 
    788 	KASSERT(p->p_nlwps == 0);
    789 	KASSERT(p->p_nzlwps == 0);
    790 	KASSERT(p->p_nrlwps == 0);
    791 	KASSERT(LIST_EMPTY(&p->p_lwps));
    792 
    793 	/*
    794 	 * If we got the child via ptrace(2) or procfs, and
    795 	 * the parent is different (meaning the process was
    796 	 * attached, rather than run as a child), then we need
    797 	 * to give it back to the old parent, and send the
    798 	 * parent the exit signal.  The rest of the cleanup
    799 	 * will be done when the old parent waits on the child.
    800 	 */
    801 	if ((p->p_flag & P_TRACED) && p->p_opptr != parent){
    802 		parent = p->p_opptr;
    803 		if (parent == NULL)
    804 			parent = initproc;
    805 		proc_reparent(p, parent);
    806 		p->p_opptr = NULL;
    807 		p->p_flag &= ~(P_TRACED|P_WAITED|P_FSTRACE|P_SYSCALL);
    808 		if (p->p_exitsig != 0) {
    809 			exit_psignal(p, parent, &ksi);
    810 			kpsignal(parent, &ksi, NULL);
    811 		}
    812 		wakeup(parent);
    813 		return;
    814 	}
    815 
    816 	scheduler_wait_hook(parent, p);
    817 	p->p_xstat = 0;
    818 
    819 	ruadd(&parent->p_stats->p_cru, p->p_ru);
    820 
    821 	/*
    822 	 * At this point we are going to start freeing the final resources.
    823 	 * If anyone tries to access the proc structure after here they
    824 	 * will get a shock - bits are missing.
    825 	 * Attempt to make it hard!
    826 	 */
    827 
    828 	p->p_stat = SIDL;		/* not even a zombie any more */
    829 
    830 	pool_put(&rusage_pool, p->p_ru);
    831 
    832 	/*
    833 	 * Finally finished with old proc entry.
    834 	 * Unlink it from its process group and free it.
    835 	 */
    836 	leavepgrp(p);
    837 
    838 	s = proclist_lock_write();
    839 	LIST_REMOVE(p, p_list);	/* off zombproc */
    840 	p->p_pptr->p_nstopchild--;
    841 	LIST_REMOVE(p, p_sibling);
    842 	proclist_unlock_write(s);
    843 
    844 	/*
    845 	 * Decrement the count of procs running with this uid.
    846 	 */
    847 	(void)chgproccnt(kauth_cred_getuid(p->p_cred), -1);
    848 
    849 	/*
    850 	 * Free up credentials.
    851 	 */
    852 	kauth_cred_free(p->p_cred);
    853 
    854 	/*
    855 	 * Release reference to text vnode
    856 	 */
    857 	if (p->p_textvp)
    858 		vrele(p->p_textvp);
    859 
    860 	/* Release any SA state. */
    861 	if (p->p_sa)
    862 		sa_release(p);
    863 
    864 	/* Free proc structure and let pid be reallocated */
    865 	proc_free_mem(p);
    866 }
    867 
    868 /*
    869  * make process 'parent' the new parent of process 'child'.
    870  *
    871  * Must be called with proclist_lock_write() held.
    872  */
    873 void
    874 proc_reparent(struct proc *child, struct proc *parent)
    875 {
    876 
    877 	if (child->p_pptr == parent)
    878 		return;
    879 
    880 	if (child->p_stat == SZOMB
    881 	    || (child->p_stat == SSTOP && !(child->p_flag & P_WAITED))) {
    882 		child->p_pptr->p_nstopchild--;
    883 		parent->p_nstopchild++;
    884 	}
    885 	if (parent == initproc)
    886 		child->p_exitsig = SIGCHLD;
    887 
    888 	LIST_REMOVE(child, p_sibling);
    889 	LIST_INSERT_HEAD(&parent->p_children, child, p_sibling);
    890 	child->p_pptr = parent;
    891 }
    892