Home | History | Annotate | Line # | Download | only in kern
kern_exit.c revision 1.201
      1 /*	$NetBSD: kern_exit.c,v 1.201 2008/03/23 16:53:45 ad Exp $	*/
      2 
      3 /*-
      4  * Copyright (c) 1998, 1999, 2006, 2007 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, and by Andrew Doran.
     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 1.201 2008/03/23 16:53:45 ad Exp $");
     78 
     79 #include "opt_ktrace.h"
     80 #include "opt_perfctrs.h"
     81 #include "opt_sysv.h"
     82 
     83 #include <sys/param.h>
     84 #include <sys/aio.h>
     85 #include <sys/systm.h>
     86 #include <sys/ioctl.h>
     87 #include <sys/tty.h>
     88 #include <sys/time.h>
     89 #include <sys/resource.h>
     90 #include <sys/kernel.h>
     91 #include <sys/proc.h>
     92 #include <sys/buf.h>
     93 #include <sys/wait.h>
     94 #include <sys/file.h>
     95 #include <sys/vnode.h>
     96 #include <sys/syslog.h>
     97 #include <sys/malloc.h>
     98 #include <sys/pool.h>
     99 #include <sys/resourcevar.h>
    100 #if defined(PERFCTRS)
    101 #include <sys/pmc.h>
    102 #endif
    103 #include <sys/ptrace.h>
    104 #include <sys/acct.h>
    105 #include <sys/filedesc.h>
    106 #include <sys/ras.h>
    107 #include <sys/signalvar.h>
    108 #include <sys/sched.h>
    109 #include <sys/mount.h>
    110 #include <sys/syscallargs.h>
    111 #include <sys/kauth.h>
    112 #include <sys/sleepq.h>
    113 #include <sys/lockdebug.h>
    114 #include <sys/ktrace.h>
    115 #include <sys/cpu.h>
    116 #include <sys/lwpctl.h>
    117 #include <sys/atomic.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 int find_stopped_child(struct proc *, pid_t, int, struct proc **, int *);
    131 static void proc_free(struct proc *, struct rusage *);
    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 	KSI_INIT(ksi);
    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 	 * It's safe to access p->p_cred unlocked here.
    154 	 */
    155 	ksi->ksi_pid = p->p_pid;
    156 	ksi->ksi_uid = kauth_cred_geteuid(p->p_cred);
    157 	ksi->ksi_status = p->p_xstat;
    158 	/* XXX: is this still valid? */
    159 	ksi->ksi_utime = p->p_stats->p_ru.ru_utime.tv_sec;
    160 	ksi->ksi_stime = p->p_stats->p_ru.ru_stime.tv_sec;
    161 }
    162 
    163 /*
    164  * exit --
    165  *	Death of process.
    166  */
    167 int
    168 sys_exit(struct lwp *l, const struct sys_exit_args *uap, register_t *retval)
    169 {
    170 	/* {
    171 		syscallarg(int)	rval;
    172 	} */
    173 	struct proc *p = l->l_proc;
    174 
    175 	/* Don't call exit1() multiple times in the same process. */
    176 	KERNEL_LOCK(1, NULL);
    177 	mutex_enter(&p->p_smutex);
    178 	if (p->p_sflag & PS_WEXIT) {
    179 		mutex_exit(&p->p_smutex);
    180 		lwp_exit(l);
    181 	}
    182 
    183 	/* exit1() will release the mutex. */
    184 	exit1(l, W_EXITCODE(SCARG(uap, rval), 0));
    185 	/* NOTREACHED */
    186 	return (0);
    187 }
    188 
    189 /*
    190  * Exit: deallocate address space and other resources, change proc state
    191  * to zombie, and unlink proc from allproc and parent's lists.  Save exit
    192  * status and rusage for wait().  Check for child processes and orphan them.
    193  *
    194  * Must be called with p->p_smutex held.  Does not return.
    195  */
    196 void
    197 exit1(struct lwp *l, int rv)
    198 {
    199 	struct proc	*p, *q, *nq;
    200 	struct pgrp	*pgrp;
    201 	ksiginfo_t	ksi;
    202 	ksiginfoq_t	kq;
    203 	int		wakeinit;
    204 
    205 	p = l->l_proc;
    206 
    207 	KASSERT(mutex_owned(&p->p_smutex));
    208 
    209 	if (__predict_false(p == initproc))
    210 		panic("init died (signal %d, exit %d)",
    211 		    WTERMSIG(rv), WEXITSTATUS(rv));
    212 
    213 	p->p_sflag |= PS_WEXIT;
    214 
    215 	/*
    216 	 * Force all other LWPs to exit before we do.  Only then can we
    217 	 * begin to tear down the rest of the process state.
    218 	 */
    219 	if (p->p_nlwps > 1)
    220 		exit_lwps(l);
    221 
    222 	ksiginfo_queue_init(&kq);
    223 
    224 	/*
    225 	 * If we have been asked to stop on exit, do so now.
    226 	 */
    227 	if (p->p_sflag & PS_STOPEXIT) {
    228 		KERNEL_UNLOCK_ALL(l, &l->l_biglocks);
    229 		sigclearall(p, &contsigmask, &kq);
    230 		p->p_waited = 0;
    231 		membar_producer();
    232 		p->p_stat = SSTOP;
    233 		lwp_lock(l);
    234 		p->p_nrlwps--;
    235 		l->l_stat = LSSTOP;
    236 		mutex_exit(&p->p_smutex);
    237 		mi_switch(l);
    238 		KERNEL_LOCK(l->l_biglocks, l);
    239 	} else
    240 		mutex_exit(&p->p_smutex);
    241 
    242 	/* Destroy any lwpctl info. */
    243 	if (p->p_lwpctl != NULL)
    244 		lwp_ctl_exit();
    245 
    246 	/* Destroy all AIO works */
    247 	aio_exit(p, p->p_aio);
    248 
    249 	/*
    250 	 * Drain all remaining references that procfs, ptrace and others may
    251 	 * have on the process.
    252 	 */
    253 	rw_enter(&p->p_reflock, RW_WRITER);
    254 
    255 	/*
    256 	 * Bin any remaining signals and mark the process as dying so it will
    257 	 * not be found for, e.g. signals.
    258 	 */
    259 	mutex_enter(&p->p_smutex);
    260 	sigfillset(&p->p_sigctx.ps_sigignore);
    261 	sigclearall(p, NULL, &kq);
    262 	p->p_stat = SDYING;
    263 	mutex_exit(&p->p_smutex);
    264 	ksiginfo_queue_drain(&kq);
    265 
    266 	DPRINTF(("exit1: %d.%d exiting.\n", p->p_pid, l->l_lid));
    267 
    268 #ifdef PGINPROF
    269 	vmsizmon();
    270 #endif
    271 	timers_free(p, TIMERS_ALL);
    272 #if defined(__HAVE_RAS)
    273 	ras_purgeall();
    274 #endif
    275 
    276 	/*
    277 	 * Close open files, release open-file table and free signal
    278 	 * actions.  This may block!
    279 	 */
    280 	fd_free();
    281 	cwdfree(p->p_cwdi);
    282 	p->p_cwdi = NULL;
    283 	doexithooks(p);
    284 	sigactsfree(p->p_sigacts);
    285 
    286 	/*
    287 	 * Write out accounting data.
    288 	 */
    289 	(void)acct_process(l);
    290 
    291 #ifdef KTRACE
    292 	/*
    293 	 * Release trace file.
    294 	 */
    295 	if (p->p_tracep != NULL) {
    296 		mutex_enter(&ktrace_lock);
    297 		ktrderef(p);
    298 		mutex_exit(&ktrace_lock);
    299 	}
    300 #endif
    301 
    302 	/*
    303 	 * If emulation has process exit hook, call it now.
    304 	 * Set the exit status now so that the exit hook has
    305 	 * an opportunity to tweak it (COMPAT_LINUX requires
    306 	 * this for thread group emulation)
    307 	 */
    308 	p->p_xstat = rv;
    309 	if (p->p_emul->e_proc_exit)
    310 		(*p->p_emul->e_proc_exit)(p);
    311 
    312 	/*
    313 	 * Free the VM resources we're still holding on to.
    314 	 * We must do this from a valid thread because doing
    315 	 * so may block. This frees vmspace, which we don't
    316 	 * need anymore. The only remaining lwp is the one
    317 	 * we run at this moment, nothing runs in userland
    318 	 * anymore.
    319 	 */
    320 	uvm_proc_exit(p);
    321 
    322 	/*
    323 	 * While we can still block, and mark the LWP as unswappable to
    324 	 * prevent conflicts with the with the swapper.  We also shouldn't
    325 	 * be swapped out, because we are about to exit and will release
    326 	 * memory.
    327 	 */
    328 	uvm_lwp_hold(l);
    329 
    330 	/*
    331 	 * Stop profiling.
    332 	 */
    333 	if ((p->p_stflag & PST_PROFIL) != 0) {
    334 		mutex_spin_enter(&p->p_stmutex);
    335 		stopprofclock(p);
    336 		mutex_spin_exit(&p->p_stmutex);
    337 	}
    338 
    339 	/*
    340 	 * If parent is waiting for us to exit or exec, P_PPWAIT is set; we
    341 	 * wake up the parent early to avoid deadlock.  We can do this once
    342 	 * the VM resources are released.
    343 	 */
    344 	mutex_enter(&proclist_lock);
    345 
    346 	mutex_enter(&p->p_smutex);
    347 	if (p->p_sflag & PS_PPWAIT) {
    348 		p->p_sflag &= ~PS_PPWAIT;
    349 		cv_broadcast(&p->p_pptr->p_waitcv);
    350 	}
    351 	mutex_exit(&p->p_smutex);
    352 
    353 	if (SESS_LEADER(p)) {
    354 		struct vnode *vprele = NULL, *vprevoke = NULL;
    355 		struct session *sp = p->p_session;
    356 		struct tty *tp;
    357 
    358 		if (sp->s_ttyvp) {
    359 			/*
    360 			 * Controlling process.
    361 			 * Signal foreground pgrp,
    362 			 * drain controlling terminal
    363 			 * and revoke access to controlling terminal.
    364 			 */
    365 			tp = sp->s_ttyp;
    366 			mutex_spin_enter(&tty_lock);
    367 			if (tp->t_session == sp) {
    368 				/* we can't guarantee the revoke will do this */
    369 				pgrp = tp->t_pgrp;
    370 				tp->t_pgrp = NULL;
    371 				tp->t_session = NULL;
    372 				mutex_spin_exit(&tty_lock);
    373 				if (pgrp != NULL) {
    374 					mutex_enter(&proclist_mutex);
    375 					pgsignal(pgrp, SIGHUP, 1);
    376 					mutex_exit(&proclist_mutex);
    377 				}
    378 				mutex_exit(&proclist_lock);
    379 				(void) ttywait(tp);
    380 				mutex_enter(&proclist_lock);
    381 
    382 				/* The tty could have been revoked. */
    383 				vprevoke = sp->s_ttyvp;
    384 			} else
    385 				mutex_spin_exit(&tty_lock);
    386 			vprele = sp->s_ttyvp;
    387 			sp->s_ttyvp = NULL;
    388 			/*
    389 			 * s_ttyp is not zero'd; we use this to indicate
    390 			 * that the session once had a controlling terminal.
    391 			 * (for logging and informational purposes)
    392 			 */
    393 		}
    394 		sp->s_leader = NULL;
    395 
    396 		if (vprevoke != NULL || vprele != NULL) {
    397 			if (vprevoke != NULL) {
    398 				SESSRELE(sp);
    399 				mutex_exit(&proclist_lock);
    400 				VOP_REVOKE(vprevoke, REVOKEALL);
    401 			} else
    402 				mutex_exit(&proclist_lock);
    403 			if (vprele != NULL)
    404 				vrele(vprele);
    405 			mutex_enter(&proclist_lock);
    406 		}
    407 	}
    408 	mutex_enter(&proclist_mutex);
    409 	fixjobc(p, p->p_pgrp, 0);
    410 	mutex_exit(&proclist_mutex);
    411 
    412 	/*
    413 	 * Finalize the last LWP's specificdata, as well as the
    414 	 * specificdata for the proc itself.
    415 	 */
    416 	lwp_finispecific(l);
    417 	proc_finispecific(p);
    418 
    419 	/*
    420 	 * Notify interested parties of our demise.
    421 	 */
    422 	KNOTE(&p->p_klist, NOTE_EXIT);
    423 
    424 
    425 
    426 #if PERFCTRS
    427 	/*
    428 	 * Save final PMC information in parent process & clean up.
    429 	 */
    430 	if (PMC_ENABLED(p)) {
    431 		pmc_save_context(p);
    432 		pmc_accumulate(p->p_pptr, p);
    433 		pmc_process_exit(p);
    434 	}
    435 #endif
    436 
    437 	/*
    438 	 * Reset p_opptr pointer of all former children which got
    439 	 * traced by another process and were reparented. We reset
    440 	 * it to NULL here; the trace detach code then reparents
    441 	 * the child to initproc. We only check allproc list, since
    442 	 * eventual former children on zombproc list won't reference
    443 	 * p_opptr anymore.
    444 	 */
    445 	if (p->p_slflag & PSL_CHTRACED) {
    446 		PROCLIST_FOREACH(q, &allproc) {
    447 			if (q->p_opptr == p)
    448 				q->p_opptr = NULL;
    449 		}
    450 	}
    451 
    452 	/*
    453 	 * Give orphaned children to init(8).
    454 	 */
    455 	q = LIST_FIRST(&p->p_children);
    456 	wakeinit = (q != NULL);
    457 	for (; q != NULL; q = nq) {
    458 		nq = LIST_NEXT(q, p_sibling);
    459 
    460 		/*
    461 		 * Traced processes are killed since their existence
    462 		 * means someone is screwing up. Since we reset the
    463 		 * trace flags, the logic in sys_wait4() would not be
    464 		 * triggered to reparent the process to its
    465 		 * original parent, so we must do this here.
    466 		 */
    467 		if (q->p_slflag & PSL_TRACED) {
    468 			mutex_enter(&p->p_smutex);
    469 			q->p_slflag &= ~(PSL_TRACED|PSL_FSTRACE|PSL_SYSCALL);
    470 			mutex_exit(&p->p_smutex);
    471 			if (q->p_opptr != q->p_pptr) {
    472 				struct proc *t = q->p_opptr;
    473 				proc_reparent(q, t ? t : initproc);
    474 				q->p_opptr = NULL;
    475 			} else
    476 				proc_reparent(q, initproc);
    477 			killproc(q, "orphaned traced process");
    478 		} else
    479 			proc_reparent(q, initproc);
    480 	}
    481 
    482 	/*
    483 	 * Move proc from allproc to zombproc, it's now nearly ready to be
    484 	 * collected by parent.
    485 	 */
    486 	mutex_enter(&proclist_mutex);
    487 	LIST_REMOVE(l, l_list);
    488 	LIST_REMOVE(p, p_list);
    489 	LIST_INSERT_HEAD(&zombproc, p, p_list);
    490 
    491 	/*
    492 	 * Mark the process as dead.  We must do this before we signal
    493 	 * the parent.
    494 	 */
    495 	p->p_stat = SDEAD;
    496 
    497 	/* Put in front of parent's sibling list for parent to collect it */
    498 	q = p->p_pptr;
    499 	q->p_nstopchild++;
    500 	if (LIST_FIRST(&q->p_children) != p) {
    501 		/* Put child where it can be found quickly */
    502 		LIST_REMOVE(p, p_sibling);
    503 		LIST_INSERT_HEAD(&q->p_children, p, p_sibling);
    504 	}
    505 	mutex_exit(&proclist_mutex);
    506 
    507 	/*
    508 	 * Notify parent that we're gone.  If parent has the P_NOCLDWAIT
    509 	 * flag set, notify init instead (and hope it will handle
    510 	 * this situation).
    511 	 */
    512 	mutex_enter(&q->p_mutex);
    513 	if (q->p_flag & (PK_NOCLDWAIT|PK_CLDSIGIGN)) {
    514 		proc_reparent(p, initproc);
    515 		wakeinit = 1;
    516 
    517 		/*
    518 		 * If this was the last child of our parent, notify
    519 		 * parent, so in case he was wait(2)ing, he will
    520 		 * continue.
    521 		 */
    522 		if (LIST_FIRST(&q->p_children) == NULL)
    523 			cv_broadcast(&q->p_waitcv);
    524 	}
    525 	mutex_exit(&q->p_mutex);
    526 
    527 	/* Reload parent pointer, since p may have been reparented above */
    528 	q = p->p_pptr;
    529 
    530 	if ((p->p_slflag & PSL_FSTRACE) == 0 && p->p_exitsig != 0) {
    531 		exit_psignal(p, q, &ksi);
    532 		mutex_enter(&proclist_mutex);
    533 		kpsignal(q, &ksi, NULL);
    534 		mutex_exit(&proclist_mutex);
    535 	}
    536 
    537 	/* Calculate the final rusage info.  */
    538 	calcru(p, &p->p_stats->p_ru.ru_utime, &p->p_stats->p_ru.ru_stime,
    539 	    NULL, NULL);
    540 
    541 	if (wakeinit)
    542 		cv_broadcast(&initproc->p_waitcv);
    543 
    544 	callout_destroy(&l->l_timeout_ch);
    545 
    546 	/*
    547 	 * Remaining lwp resources will be freed in lwp_exit2() once we've
    548 	 * switch to idle context; at that point, we will be marked as a
    549 	 * full blown zombie.
    550 	 *
    551 	 * XXXSMP disable preemption.
    552 	 */
    553 	mutex_enter(&p->p_smutex);
    554 	lwp_drainrefs(l);
    555 	lwp_lock(l);
    556 	l->l_prflag &= ~LPR_DETACHED;
    557 	l->l_stat = LSZOMB;
    558 	lwp_unlock(l);
    559 	KASSERT(curlwp == l);
    560 	KASSERT(p->p_nrlwps == 1);
    561 	KASSERT(p->p_nlwps == 1);
    562 	p->p_stat = SZOMB;
    563 	p->p_nrlwps--;
    564 	p->p_nzlwps++;
    565 	p->p_ndlwps = 0;
    566 	mutex_exit(&p->p_smutex);
    567 
    568 	/*
    569 	 * Signal the parent to collect us, and drop the proclist lock.
    570 	 * Drop debugger/procfs lock; no new references can be gained.
    571 	 */
    572 	cv_broadcast(&p->p_pptr->p_waitcv);
    573 	mutex_exit(&proclist_lock);
    574 	rw_exit(&p->p_reflock);
    575 
    576 	/* Verify that we hold no locks other than the kernel lock. */
    577 #ifdef MULTIPROCESSOR
    578 	LOCKDEBUG_BARRIER(&kernel_lock, 0);
    579 #else
    580 	LOCKDEBUG_BARRIER(NULL, 0);
    581 #endif
    582 
    583 	/*
    584 	 * NOTE: WE ARE NO LONGER ALLOWED TO SLEEP!
    585 	 */
    586 
    587 	/*
    588 	 * Give machine-dependent code a chance to free any MD LWP
    589 	 * resources.  This must be done before uvm_lwp_exit(), in
    590 	 * case these resources are in the PCB.
    591 	 */
    592 #ifndef __NO_CPU_LWP_FREE
    593 	cpu_lwp_free(l, 1);
    594 #endif
    595 	pmap_deactivate(l);
    596 
    597 	/* This process no longer needs to hold the kernel lock. */
    598 #ifdef notyet
    599 	/* XXXSMP hold in lwp_userret() */
    600 	KERNEL_UNLOCK_LAST(l);
    601 #else
    602 	KERNEL_UNLOCK_ALL(l, NULL);
    603 #endif
    604 
    605 	lwp_exit_switchaway(l);
    606 }
    607 
    608 void
    609 exit_lwps(struct lwp *l)
    610 {
    611 	struct proc *p;
    612 	struct lwp *l2;
    613 	int error;
    614 	lwpid_t waited;
    615 #if defined(MULTIPROCESSOR)
    616 	int nlocks;
    617 #endif
    618 
    619 	KERNEL_UNLOCK_ALL(l, &nlocks);
    620 
    621 	p = l->l_proc;
    622 	KASSERT(mutex_owned(&p->p_smutex));
    623 
    624  retry:
    625 	/*
    626 	 * Interrupt LWPs in interruptable sleep, unsuspend suspended
    627 	 * LWPs and then wait for everyone else to finish.
    628 	 */
    629 	LIST_FOREACH(l2, &p->p_lwps, l_sibling) {
    630 		if (l2 == l)
    631 			continue;
    632 		lwp_lock(l2);
    633 		l2->l_flag |= LW_WEXIT;
    634 		if ((l2->l_stat == LSSLEEP && (l2->l_flag & LW_SINTR)) ||
    635 		    l2->l_stat == LSSUSPENDED || l2->l_stat == LSSTOP) {
    636 		    	/* setrunnable() will release the lock. */
    637 			setrunnable(l2);
    638 			DPRINTF(("exit_lwps: Made %d.%d runnable\n",
    639 			    p->p_pid, l2->l_lid));
    640 			continue;
    641 		}
    642 		lwp_unlock(l2);
    643 	}
    644 	while (p->p_nlwps > 1) {
    645 		DPRINTF(("exit_lwps: waiting for %d LWPs (%d zombies)\n",
    646 		    p->p_nlwps, p->p_nzlwps));
    647 		error = lwp_wait1(l, 0, &waited, LWPWAIT_EXITCONTROL);
    648 		if (p->p_nlwps == 1)
    649 			break;
    650 		if (error == EDEADLK) {
    651 			/*
    652 			 * LWPs can get suspended/slept behind us.
    653 			 * (eg. sa_setwoken)
    654 			 * kick them again and retry.
    655 			 */
    656 			goto retry;
    657 		}
    658 		if (error)
    659 			panic("exit_lwps: lwp_wait1 failed with error %d",
    660 			    error);
    661 		DPRINTF(("exit_lwps: Got LWP %d from lwp_wait1()\n", waited));
    662 	}
    663 
    664 #if defined(MULTIPROCESSOR)
    665 	if (nlocks > 0) {
    666 		mutex_exit(&p->p_smutex);
    667 		KERNEL_LOCK(nlocks, l);
    668 		mutex_enter(&p->p_smutex);
    669 	}
    670 #endif /* defined(MULTIPROCESSOR) */
    671 	KASSERT(p->p_nlwps == 1);
    672 }
    673 
    674 int
    675 do_sys_wait(struct lwp *l, int *pid, int *status, int options,
    676     struct rusage *ru, int *was_zombie)
    677 {
    678 	struct proc	*child;
    679 	int		error;
    680 
    681 	KERNEL_LOCK(1, NULL);		/* XXXSMP */
    682 	mutex_enter(&proclist_lock);
    683 	error = find_stopped_child(l->l_proc, *pid, options, &child, status);
    684 	KERNEL_UNLOCK_ONE(NULL);	/* XXXSMP */
    685 
    686 	if (child == NULL) {
    687 		mutex_exit(&proclist_lock);
    688 		*pid = 0;
    689 		return error;
    690 	}
    691 
    692 	*pid = child->p_pid;
    693 
    694 	if (child->p_stat == SZOMB) {
    695 		/* proc_free() will release the proclist_lock. */
    696 		*was_zombie = 1;
    697 		if (options & WNOWAIT)
    698 			mutex_exit(&proclist_lock);
    699 		else {
    700 			proc_free(child, ru);
    701 		}
    702 	} else {
    703 		/* Child state must have been SSTOP. */
    704 		*was_zombie = 0;
    705 		mutex_exit(&proclist_lock);
    706 		*status = W_STOPCODE(*status);
    707 	}
    708 
    709 	return 0;
    710 }
    711 
    712 int
    713 sys_wait4(struct lwp *l, const struct sys_wait4_args *uap, register_t *retval)
    714 {
    715 	/* {
    716 		syscallarg(int)			pid;
    717 		syscallarg(int *)		status;
    718 		syscallarg(int)			options;
    719 		syscallarg(struct rusage *)	rusage;
    720 	} */
    721 	int		status, error;
    722 	int		was_zombie;
    723 	struct rusage	ru;
    724 	int pid = SCARG(uap, pid);
    725 
    726 	error = do_sys_wait(l, &pid, &status, SCARG(uap, options),
    727 	    SCARG(uap, rusage) != NULL ? &ru : NULL, &was_zombie);
    728 
    729 	retval[0] = pid;
    730 	if (pid == 0)
    731 		return error;
    732 
    733 	if (SCARG(uap, rusage))
    734 		error = copyout(&ru, SCARG(uap, rusage), sizeof(ru));
    735 
    736 	if (error == 0 && SCARG(uap, status))
    737 		error = copyout(&status, SCARG(uap, status), sizeof(status));
    738 
    739 	return error;
    740 }
    741 
    742 /*
    743  * Scan list of child processes for a child process that has stopped or
    744  * exited.  Used by sys_wait4 and 'compat' equivalents.
    745  *
    746  * Must be called with the proclist_lock held, and may release
    747  * while waiting.
    748  */
    749 static int
    750 find_stopped_child(struct proc *parent, pid_t pid, int options,
    751 		   struct proc **child_p, int *status_p)
    752 {
    753 	struct proc *child, *dead;
    754 	int error;
    755 
    756 	KASSERT(mutex_owned(&proclist_lock));
    757 
    758 	if (options & ~(WUNTRACED|WNOHANG|WALTSIG|WALLSIG)
    759 	    && !(options & WOPTSCHECKED)) {
    760 		*child_p = NULL;
    761 		return EINVAL;
    762 	}
    763 
    764 	if (pid == 0 && !(options & WOPTSCHECKED))
    765 		pid = -parent->p_pgid;
    766 
    767 	for (;;) {
    768 		error = ECHILD;
    769 		dead = NULL;
    770 
    771 		mutex_enter(&proclist_mutex);
    772 		LIST_FOREACH(child, &parent->p_children, p_sibling) {
    773 			if (pid >= 0) {
    774 				if (child->p_pid != pid) {
    775 					child = p_find(pid, PFIND_ZOMBIE |
    776 					    PFIND_LOCKED);
    777 					if (child == NULL ||
    778 					    child->p_pptr != parent) {
    779 						child = NULL;
    780 						break;
    781 					}
    782 				}
    783 			} else if (pid != WAIT_ANY && child->p_pgid != -pid) {
    784 				/* Child not in correct pgrp */
    785 				continue;
    786 			}
    787 
    788 			/*
    789 			 * Wait for processes with p_exitsig != SIGCHLD
    790 			 * processes only if WALTSIG is set; wait for
    791 			 * processes with p_exitsig == SIGCHLD only
    792 			 * if WALTSIG is clear.
    793 			 */
    794 			if (((options & WALLSIG) == 0) &&
    795 			    (options & WALTSIG ? child->p_exitsig == SIGCHLD
    796 						: P_EXITSIG(child) != SIGCHLD)){
    797 				if (child->p_pid == pid) {
    798 					child = NULL;
    799 					break;
    800 				}
    801 				continue;
    802 			}
    803 
    804 			error = 0;
    805 			if ((options & WNOZOMBIE) == 0) {
    806 				if (child->p_stat == SZOMB)
    807 					break;
    808 				if (child->p_stat == SDEAD) {
    809 					/*
    810 					 * We may occasionally arrive here
    811 					 * after receiving a signal, but
    812 					 * immediatley before the child
    813 					 * process is zombified.  The wait
    814 					 * will be short, so avoid returning
    815 					 * to userspace.
    816 					 */
    817 					dead = child;
    818 				}
    819 			}
    820 
    821 			if (child->p_stat == SSTOP &&
    822 			    child->p_waited == 0 &&
    823 			    (child->p_slflag & PSL_TRACED ||
    824 			    options & WUNTRACED)) {
    825 				if ((options & WNOWAIT) == 0) {
    826 					child->p_waited = 1;
    827 					parent->p_nstopchild--;
    828 				}
    829 				break;
    830 			}
    831 			if (parent->p_nstopchild == 0 || child->p_pid == pid) {
    832 				child = NULL;
    833 				break;
    834 			}
    835 		}
    836 
    837 		if (child != NULL || error != 0 ||
    838 		    ((options & WNOHANG) != 0 && dead == NULL)) {
    839 		    	if (child != NULL) {
    840 			    	*status_p = child->p_xstat;
    841 			}
    842 			mutex_exit(&proclist_mutex);
    843 			*child_p = child;
    844 			return error;
    845 		}
    846 
    847 		/*
    848 		 * Wait for another child process to stop.
    849 		 */
    850 		mutex_exit(&proclist_lock);
    851 		error = cv_wait_sig(&parent->p_waitcv, &proclist_mutex);
    852 		mutex_exit(&proclist_mutex);
    853 		mutex_enter(&proclist_lock);
    854 
    855 		if (error != 0) {
    856 			*child_p = NULL;
    857 			return error;
    858 		}
    859 	}
    860 }
    861 
    862 /*
    863  * Free a process after parent has taken all the state info.  Must be called
    864  * with the proclist lock held, and will release before returning.
    865  *
    866  * *ru is returned to the caller, and must be freed by the caller.
    867  */
    868 static void
    869 proc_free(struct proc *p, struct rusage *ru)
    870 {
    871 	struct proc *parent;
    872 	struct lwp *l;
    873 	ksiginfo_t ksi;
    874 	kauth_cred_t cred1, cred2;
    875 	uid_t uid;
    876 
    877 	KASSERT(mutex_owned(&proclist_lock));
    878 	KASSERT(p->p_nlwps == 1);
    879 	KASSERT(p->p_nzlwps == 1);
    880 	KASSERT(p->p_nrlwps == 0);
    881 	KASSERT(p->p_stat == SZOMB);
    882 
    883 	/*
    884 	 * If we got the child via ptrace(2) or procfs, and
    885 	 * the parent is different (meaning the process was
    886 	 * attached, rather than run as a child), then we need
    887 	 * to give it back to the old parent, and send the
    888 	 * parent the exit signal.  The rest of the cleanup
    889 	 * will be done when the old parent waits on the child.
    890 	 */
    891 	if ((p->p_slflag & PSL_TRACED) != 0) {
    892 		parent = p->p_pptr;
    893 		if (p->p_opptr != parent){
    894 			mutex_enter(&p->p_smutex);
    895 			p->p_slflag &= ~(PSL_TRACED|PSL_FSTRACE|PSL_SYSCALL);
    896 			mutex_exit(&p->p_smutex);
    897 			parent = p->p_opptr;
    898 			if (parent == NULL)
    899 				parent = initproc;
    900 			proc_reparent(p, parent);
    901 			p->p_opptr = NULL;
    902 			if (p->p_exitsig != 0) {
    903 				exit_psignal(p, parent, &ksi);
    904 				mutex_enter(&proclist_mutex);
    905 				kpsignal(parent, &ksi, NULL);
    906 				mutex_exit(&proclist_mutex);
    907 			}
    908 			KERNEL_LOCK(1, NULL);		/* XXXSMP */
    909 			cv_broadcast(&parent->p_waitcv);
    910 			KERNEL_UNLOCK_ONE(NULL);	/* XXXSMP */
    911 			mutex_exit(&proclist_lock);
    912 			return;
    913 		}
    914 	}
    915 
    916 	/*
    917 	 * Finally finished with old proc entry.  Unlink it from its process
    918 	 * group.
    919 	 */
    920 	leavepgrp(p);
    921 
    922 	parent = p->p_pptr;
    923 	sched_proc_exit(parent, p);
    924 	/*
    925 	 * Add child times of exiting process onto its own times.
    926 	 * This cannot be done any earlier else it might get done twice.
    927 	 */
    928 	ruadd(&p->p_stats->p_ru, &p->p_stats->p_cru);
    929 	ruadd(&parent->p_stats->p_cru, &p->p_stats->p_ru);
    930 	if (ru != NULL)
    931 		*ru = p->p_stats->p_ru;
    932 	p->p_xstat = 0;
    933 
    934 	/*
    935 	 * At this point we are going to start freeing the final resources.
    936 	 * If anyone tries to access the proc structure after here they will
    937 	 * get a shock - bits are missing.  Attempt to make it hard!  We
    938 	 * don't bother with any further locking past this point.
    939 	 */
    940 	mutex_enter(&proclist_mutex);
    941 	p->p_stat = SIDL;		/* not even a zombie any more */
    942 	LIST_REMOVE(p, p_list);	/* off zombproc */
    943 	parent = p->p_pptr;
    944 	p->p_pptr->p_nstopchild--;
    945 	mutex_exit(&proclist_mutex);
    946 	LIST_REMOVE(p, p_sibling);
    947 
    948 	/*
    949 	 * Let pid be reallocated.
    950 	 */
    951 	proc_free_pid(p);
    952 	mutex_exit(&proclist_lock);
    953 
    954 	l = LIST_FIRST(&p->p_lwps);
    955 
    956 	/*
    957 	 * Delay release until after lwp_free.
    958 	 */
    959 	cred2 = l->l_cred;
    960 
    961 	/*
    962 	 * Free the last LWP's resources.
    963 	 *
    964 	 * lwp_free ensures the LWP is no longer running on another CPU.
    965 	 */
    966 	lwp_free(l, false, true);
    967 
    968 	/*
    969 	 * Now no one except us can reach the process p.
    970 	 */
    971 
    972 	/*
    973 	 * Decrement the count of procs running with this uid.
    974 	 */
    975 	cred1 = p->p_cred;
    976 	uid = kauth_cred_getuid(cred1);
    977 	(void)chgproccnt(uid, -1);
    978 
    979 	/*
    980 	 * Release substructures.
    981 	 */
    982 
    983 	limfree(p->p_limit);
    984 	pstatsfree(p->p_stats);
    985 	kauth_cred_free(cred1);
    986 	kauth_cred_free(cred2);
    987 
    988 	/*
    989 	 * Release reference to text vnode
    990 	 */
    991 	if (p->p_textvp)
    992 		vrele(p->p_textvp);
    993 
    994 	mutex_destroy(&p->p_auxlock);
    995 	mutex_destroy(&p->p_mutex);
    996 	mutex_destroy(&p->p_stmutex);
    997 	mutex_destroy(&p->p_smutex);
    998 	cv_destroy(&p->p_waitcv);
    999 	cv_destroy(&p->p_lwpcv);
   1000 	rw_destroy(&p->p_reflock);
   1001 
   1002 	proc_free_mem(p);
   1003 }
   1004 
   1005 /*
   1006  * make process 'parent' the new parent of process 'child'.
   1007  *
   1008  * Must be called with proclist_lock lock held.
   1009  */
   1010 void
   1011 proc_reparent(struct proc *child, struct proc *parent)
   1012 {
   1013 
   1014 	KASSERT(mutex_owned(&proclist_lock));
   1015 
   1016 	if (child->p_pptr == parent)
   1017 		return;
   1018 
   1019 	mutex_enter(&proclist_mutex);
   1020 	if (child->p_stat == SZOMB ||
   1021 	    (child->p_stat == SSTOP && !child->p_waited)) {
   1022 		child->p_pptr->p_nstopchild--;
   1023 		parent->p_nstopchild++;
   1024 	}
   1025 	mutex_exit(&proclist_mutex);
   1026 	if (parent == initproc)
   1027 		child->p_exitsig = SIGCHLD;
   1028 
   1029 	LIST_REMOVE(child, p_sibling);
   1030 	LIST_INSERT_HEAD(&parent->p_children, child, p_sibling);
   1031 	child->p_pptr = parent;
   1032 }
   1033