kern_exit.c revision 1.234 1 /* $NetBSD: kern_exit.c,v 1.234 2011/06/06 22:04:34 matt Exp $ */
2
3 /*-
4 * Copyright (c) 1998, 1999, 2006, 2007, 2008 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 *
20 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
21 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
22 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
23 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
24 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
25 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
26 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
27 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
28 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
29 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
30 * POSSIBILITY OF SUCH DAMAGE.
31 */
32
33 /*
34 * Copyright (c) 1982, 1986, 1989, 1991, 1993
35 * The Regents of the University of California. All rights reserved.
36 * (c) UNIX System Laboratories, Inc.
37 * All or some portions of this file are derived from material licensed
38 * to the University of California by American Telephone and Telegraph
39 * Co. or Unix System Laboratories, Inc. and are reproduced herein with
40 * the permission of UNIX System Laboratories, Inc.
41 *
42 * Redistribution and use in source and binary forms, with or without
43 * modification, are permitted provided that the following conditions
44 * are met:
45 * 1. Redistributions of source code must retain the above copyright
46 * notice, this list of conditions and the following disclaimer.
47 * 2. Redistributions in binary form must reproduce the above copyright
48 * notice, this list of conditions and the following disclaimer in the
49 * documentation and/or other materials provided with the distribution.
50 * 3. Neither the name of the University nor the names of its contributors
51 * may be used to endorse or promote products derived from this software
52 * without specific prior written permission.
53 *
54 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
55 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
56 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
57 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
58 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
59 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
60 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
61 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
62 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
63 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
64 * SUCH DAMAGE.
65 *
66 * @(#)kern_exit.c 8.10 (Berkeley) 2/23/95
67 */
68
69 #include <sys/cdefs.h>
70 __KERNEL_RCSID(0, "$NetBSD: kern_exit.c,v 1.234 2011/06/06 22:04:34 matt Exp $");
71
72 #include "opt_ktrace.h"
73 #include "opt_perfctrs.h"
74 #include "opt_sa.h"
75 #include "opt_sysv.h"
76
77 #include <sys/param.h>
78 #include <sys/systm.h>
79 #include <sys/ioctl.h>
80 #include <sys/tty.h>
81 #include <sys/time.h>
82 #include <sys/resource.h>
83 #include <sys/kernel.h>
84 #include <sys/proc.h>
85 #include <sys/buf.h>
86 #include <sys/wait.h>
87 #include <sys/file.h>
88 #include <sys/vnode.h>
89 #include <sys/syslog.h>
90 #include <sys/pool.h>
91 #include <sys/uidinfo.h>
92 #if defined(PERFCTRS)
93 #include <sys/pmc.h>
94 #endif
95 #include <sys/ptrace.h>
96 #include <sys/acct.h>
97 #include <sys/filedesc.h>
98 #include <sys/ras.h>
99 #include <sys/signalvar.h>
100 #include <sys/sched.h>
101 #include <sys/sa.h>
102 #include <sys/savar.h>
103 #include <sys/mount.h>
104 #include <sys/syscallargs.h>
105 #include <sys/kauth.h>
106 #include <sys/sleepq.h>
107 #include <sys/lockdebug.h>
108 #include <sys/ktrace.h>
109 #include <sys/cpu.h>
110 #include <sys/lwpctl.h>
111 #include <sys/atomic.h>
112 #include <sys/sdt.h>
113
114 #include <uvm/uvm_extern.h>
115
116 #ifdef DEBUG_EXIT
117 int debug_exit = 0;
118 #define DPRINTF(x) if (debug_exit) printf x
119 #else
120 #define DPRINTF(x)
121 #endif
122
123 static int find_stopped_child(struct proc *, pid_t, int, struct proc **, int *);
124 static void proc_free(struct proc *, struct rusage *);
125
126 /*
127 * DTrace SDT provider definitions
128 */
129 SDT_PROBE_DEFINE(proc,,,exit,
130 "int", NULL, /* reason */
131 NULL, NULL, NULL, NULL,
132 NULL, NULL, NULL, NULL);
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 mutex_enter(p->p_lock);
177 if (p->p_sflag & PS_WEXIT) {
178 mutex_exit(p->p_lock);
179 lwp_exit(l);
180 }
181
182 /* exit1() will release the mutex. */
183 exit1(l, W_EXITCODE(SCARG(uap, rval), 0));
184 /* NOTREACHED */
185 return (0);
186 }
187
188 /*
189 * Exit: deallocate address space and other resources, change proc state
190 * to zombie, and unlink proc from allproc and parent's lists. Save exit
191 * status and rusage for wait(). Check for child processes and orphan them.
192 *
193 * Must be called with p->p_lock held. Does not return.
194 */
195 void
196 exit1(struct lwp *l, int rv)
197 {
198 struct proc *p, *q, *nq;
199 struct pgrp *pgrp;
200 ksiginfo_t ksi;
201 ksiginfoq_t kq;
202 int wakeinit, sa;
203
204 p = l->l_proc;
205
206 KASSERT(mutex_owned(p->p_lock));
207
208 if (__predict_false(p == initproc))
209 panic("init died (signal %d, exit %d)",
210 WTERMSIG(rv), WEXITSTATUS(rv));
211
212 /*
213 * Disable scheduler activation upcalls. We're trying to get out of
214 * here.
215 */
216 sa = 0;
217 #ifdef KERN_SA
218 if ((p->p_sa != NULL)) {
219 l->l_pflag |= LP_SA_NOBLOCK;
220 sa = 1;
221 }
222 #endif
223
224 p->p_sflag |= PS_WEXIT;
225
226 /*
227 * Force all other LWPs to exit before we do. Only then can we
228 * begin to tear down the rest of the process state.
229 */
230 if (sa || p->p_nlwps > 1)
231 exit_lwps(l);
232
233 ksiginfo_queue_init(&kq);
234
235 /*
236 * If we have been asked to stop on exit, do so now.
237 */
238 if (__predict_false(p->p_sflag & PS_STOPEXIT)) {
239 KERNEL_UNLOCK_ALL(l, &l->l_biglocks);
240 sigclearall(p, &contsigmask, &kq);
241 p->p_waited = 0;
242 membar_producer();
243 p->p_stat = SSTOP;
244 lwp_lock(l);
245 p->p_nrlwps--;
246 l->l_stat = LSSTOP;
247 lwp_unlock(l);
248 mutex_exit(p->p_lock);
249 lwp_lock(l);
250 mi_switch(l);
251 KERNEL_LOCK(l->l_biglocks, l);
252 mutex_enter(p->p_lock);
253 }
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 sigfillset(&p->p_sigctx.ps_sigignore);
260 sigclearall(p, NULL, &kq);
261 p->p_stat = SDYING;
262 mutex_exit(p->p_lock);
263 ksiginfo_queue_drain(&kq);
264
265 /* Destroy any lwpctl info. */
266 if (p->p_lwpctl != NULL)
267 lwp_ctl_exit();
268
269 /*
270 * Drain all remaining references that procfs, ptrace and others may
271 * have on the process.
272 */
273 rw_enter(&p->p_reflock, RW_WRITER);
274
275 DPRINTF(("exit1: %d.%d exiting.\n", p->p_pid, l->l_lid));
276
277 timers_free(p, TIMERS_ALL);
278 #if defined(__HAVE_RAS)
279 ras_purgeall();
280 #endif
281
282 /*
283 * Close open files, release open-file table and free signal
284 * actions. This may block!
285 */
286 fd_free();
287 cwdfree(p->p_cwdi);
288 p->p_cwdi = NULL;
289 doexithooks(p);
290 sigactsfree(p->p_sigacts);
291
292 /*
293 * Write out accounting data.
294 */
295 (void)acct_process(l);
296
297 #ifdef KTRACE
298 /*
299 * Release trace file.
300 */
301 if (p->p_tracep != NULL) {
302 mutex_enter(&ktrace_lock);
303 ktrderef(p);
304 mutex_exit(&ktrace_lock);
305 }
306 #endif
307
308 /*
309 * If emulation has process exit hook, call it now.
310 * Set the exit status now so that the exit hook has
311 * an opportunity to tweak it (COMPAT_LINUX requires
312 * this for thread group emulation)
313 */
314 p->p_xstat = rv;
315 if (p->p_emul->e_proc_exit)
316 (*p->p_emul->e_proc_exit)(p);
317
318 /*
319 * Free the VM resources we're still holding on to.
320 * We must do this from a valid thread because doing
321 * so may block. This frees vmspace, which we don't
322 * need anymore. The only remaining lwp is the one
323 * we run at this moment, nothing runs in userland
324 * anymore.
325 */
326 uvm_proc_exit(p);
327
328 /*
329 * Stop profiling.
330 */
331 if (__predict_false((p->p_stflag & PST_PROFIL) != 0)) {
332 mutex_spin_enter(&p->p_stmutex);
333 stopprofclock(p);
334 mutex_spin_exit(&p->p_stmutex);
335 }
336
337 /*
338 * If parent is waiting for us to exit or exec, PL_PPWAIT is set; we
339 * wake up the parent early to avoid deadlock. We can do this once
340 * the VM resources are released.
341 */
342 mutex_enter(proc_lock);
343 if (p->p_lflag & PL_PPWAIT) {
344 l->l_lwpctl = NULL; /* was on loan from blocked parent */
345 p->p_lflag &= ~PL_PPWAIT;
346 cv_broadcast(&p->p_pptr->p_waitcv);
347 }
348
349 if (SESS_LEADER(p)) {
350 struct vnode *vprele = NULL, *vprevoke = NULL;
351 struct session *sp = p->p_session;
352 struct tty *tp;
353
354 if (sp->s_ttyvp) {
355 /*
356 * Controlling process.
357 * Signal foreground pgrp,
358 * drain controlling terminal
359 * and revoke access to controlling terminal.
360 */
361 tp = sp->s_ttyp;
362 mutex_spin_enter(&tty_lock);
363 if (tp->t_session == sp) {
364 /* we can't guarantee the revoke will do this */
365 pgrp = tp->t_pgrp;
366 tp->t_pgrp = NULL;
367 tp->t_session = NULL;
368 mutex_spin_exit(&tty_lock);
369 if (pgrp != NULL) {
370 pgsignal(pgrp, SIGHUP, 1);
371 }
372 mutex_exit(proc_lock);
373 (void) ttywait(tp);
374 mutex_enter(proc_lock);
375
376 /* The tty could have been revoked. */
377 vprevoke = sp->s_ttyvp;
378 } else
379 mutex_spin_exit(&tty_lock);
380 vprele = sp->s_ttyvp;
381 sp->s_ttyvp = NULL;
382 /*
383 * s_ttyp is not zero'd; we use this to indicate
384 * that the session once had a controlling terminal.
385 * (for logging and informational purposes)
386 */
387 }
388 sp->s_leader = NULL;
389
390 if (vprevoke != NULL || vprele != NULL) {
391 if (vprevoke != NULL) {
392 /* Releases proc_lock. */
393 proc_sessrele(sp);
394 VOP_REVOKE(vprevoke, REVOKEALL);
395 } else
396 mutex_exit(proc_lock);
397 if (vprele != NULL)
398 vrele(vprele);
399 mutex_enter(proc_lock);
400 }
401 }
402 fixjobc(p, p->p_pgrp, 0);
403
404 /*
405 * Finalize the last LWP's specificdata, as well as the
406 * specificdata for the proc itself.
407 */
408 lwp_finispecific(l);
409 proc_finispecific(p);
410
411 /*
412 * Notify interested parties of our demise.
413 */
414 KNOTE(&p->p_klist, NOTE_EXIT);
415
416 SDT_PROBE(proc,,,exit,
417 (WCOREDUMP(rv) ? CLD_DUMPED :
418 (WIFSIGNALED(rv) ? CLD_KILLED : CLD_EXITED)),
419 0,0,0,0);
420
421 #if PERFCTRS
422 /*
423 * Save final PMC information in parent process & clean up.
424 */
425 if (PMC_ENABLED(p)) {
426 pmc_save_context(p);
427 pmc_accumulate(p->p_pptr, p);
428 pmc_process_exit(p);
429 }
430 #endif
431
432 /*
433 * Reset p_opptr pointer of all former children which got
434 * traced by another process and were reparented. We reset
435 * it to NULL here; the trace detach code then reparents
436 * the child to initproc. We only check allproc list, since
437 * eventual former children on zombproc list won't reference
438 * p_opptr anymore.
439 */
440 if (__predict_false(p->p_slflag & PSL_CHTRACED)) {
441 PROCLIST_FOREACH(q, &allproc) {
442 if (q->p_opptr == p)
443 q->p_opptr = NULL;
444 }
445 }
446
447 /*
448 * Give orphaned children to init(8).
449 */
450 q = LIST_FIRST(&p->p_children);
451 wakeinit = (q != NULL);
452 for (; q != NULL; q = nq) {
453 nq = LIST_NEXT(q, p_sibling);
454
455 /*
456 * Traced processes are killed since their existence
457 * means someone is screwing up. Since we reset the
458 * trace flags, the logic in sys_wait4() would not be
459 * triggered to reparent the process to its
460 * original parent, so we must do this here.
461 */
462 if (__predict_false(q->p_slflag & PSL_TRACED)) {
463 mutex_enter(p->p_lock);
464 q->p_slflag &= ~(PSL_TRACED|PSL_FSTRACE|PSL_SYSCALL);
465 mutex_exit(p->p_lock);
466 if (q->p_opptr != q->p_pptr) {
467 struct proc *t = q->p_opptr;
468 proc_reparent(q, t ? t : initproc);
469 q->p_opptr = NULL;
470 } else
471 proc_reparent(q, initproc);
472 killproc(q, "orphaned traced process");
473 } else
474 proc_reparent(q, initproc);
475 }
476
477 /*
478 * Move proc from allproc to zombproc, it's now nearly ready to be
479 * collected by parent.
480 */
481 LIST_REMOVE(l, l_list);
482 LIST_REMOVE(p, p_list);
483 LIST_INSERT_HEAD(&zombproc, p, p_list);
484
485 /*
486 * Mark the process as dead. We must do this before we signal
487 * the parent.
488 */
489 p->p_stat = SDEAD;
490
491 /* Put in front of parent's sibling list for parent to collect it */
492 q = p->p_pptr;
493 q->p_nstopchild++;
494 if (LIST_FIRST(&q->p_children) != p) {
495 /* Put child where it can be found quickly */
496 LIST_REMOVE(p, p_sibling);
497 LIST_INSERT_HEAD(&q->p_children, p, p_sibling);
498 }
499
500 /*
501 * Notify parent that we're gone. If parent has the P_NOCLDWAIT
502 * flag set, notify init instead (and hope it will handle
503 * this situation).
504 */
505 if (q->p_flag & (PK_NOCLDWAIT|PK_CLDSIGIGN)) {
506 proc_reparent(p, initproc);
507 wakeinit = 1;
508
509 /*
510 * If this was the last child of our parent, notify
511 * parent, so in case he was wait(2)ing, he will
512 * continue.
513 */
514 if (LIST_FIRST(&q->p_children) == NULL)
515 cv_broadcast(&q->p_waitcv);
516 }
517
518 /* Reload parent pointer, since p may have been reparented above */
519 q = p->p_pptr;
520
521 if (__predict_false((p->p_slflag & PSL_FSTRACE) == 0 &&
522 p->p_exitsig != 0)) {
523 exit_psignal(p, q, &ksi);
524 kpsignal(q, &ksi, NULL);
525 }
526
527 /* Calculate the final rusage info. */
528 calcru(p, &p->p_stats->p_ru.ru_utime, &p->p_stats->p_ru.ru_stime,
529 NULL, NULL);
530
531 if (wakeinit)
532 cv_broadcast(&initproc->p_waitcv);
533
534 callout_destroy(&l->l_timeout_ch);
535
536 /*
537 * Release any PCU resources before becoming a zombie.
538 */
539 pcu_discard_all(l);
540
541 /*
542 * Remaining lwp resources will be freed in lwp_exit2() once we've
543 * switch to idle context; at that point, we will be marked as a
544 * full blown zombie.
545 */
546 mutex_enter(p->p_lock);
547 lwp_drainrefs(l);
548 lwp_lock(l);
549 l->l_prflag &= ~LPR_DETACHED;
550 l->l_stat = LSZOMB;
551 lwp_unlock(l);
552 KASSERT(curlwp == l);
553 KASSERT(p->p_nrlwps == 1);
554 KASSERT(p->p_nlwps == 1);
555 p->p_stat = SZOMB;
556 p->p_nrlwps--;
557 p->p_nzlwps++;
558 p->p_ndlwps = 0;
559 mutex_exit(p->p_lock);
560
561 /*
562 * Signal the parent to collect us, and drop the proclist lock.
563 * Drop debugger/procfs lock; no new references can be gained.
564 */
565 cv_broadcast(&p->p_pptr->p_waitcv);
566 rw_exit(&p->p_reflock);
567 mutex_exit(proc_lock);
568
569 /* Verify that we hold no locks other than the kernel lock. */
570 LOCKDEBUG_BARRIER(&kernel_lock, 0);
571
572 /*
573 * NOTE: WE ARE NO LONGER ALLOWED TO SLEEP!
574 */
575
576 /*
577 * Give machine-dependent code a chance to free any MD LWP
578 * resources. This must be done before uvm_lwp_exit(), in
579 * case these resources are in the PCB.
580 */
581 cpu_lwp_free(l, 1);
582 pmap_deactivate(l);
583
584 /* This process no longer needs to hold the kernel lock. */
585 #ifdef notyet
586 /* XXXSMP hold in lwp_userret() */
587 KERNEL_UNLOCK_LAST(l);
588 #else
589 KERNEL_UNLOCK_ALL(l, NULL);
590 #endif
591
592 lwp_exit_switchaway(l);
593 }
594
595 void
596 exit_lwps(struct lwp *l)
597 {
598 struct proc *p;
599 struct lwp *l2;
600 int error;
601 lwpid_t waited;
602 int nlocks;
603
604 KERNEL_UNLOCK_ALL(l, &nlocks);
605
606 p = l->l_proc;
607 KASSERT(mutex_owned(p->p_lock));
608
609 #ifdef KERN_SA
610 if (p->p_sa != NULL) {
611 struct sadata_vp *vp;
612 SLIST_FOREACH(vp, &p->p_sa->sa_vps, savp_next) {
613 /*
614 * Make SA-cached LWPs normal process interruptable
615 * so that the exit code can wake them. Locking
616 * savp_mutex locks all the lwps on this vp that
617 * we need to adjust.
618 */
619 mutex_enter(&vp->savp_mutex);
620 DPRINTF(("exit_lwps: Making cached LWPs of %d on "
621 "VP %d interruptable: ", p->p_pid, vp->savp_id));
622 TAILQ_FOREACH(l2, &vp->savp_lwpcache, l_sleepchain) {
623 l2->l_flag |= LW_SINTR;
624 DPRINTF(("%d ", l2->l_lid));
625 }
626 DPRINTF(("\n"));
627
628 DPRINTF(("exit_lwps: Making unblocking LWPs of %d on "
629 "VP %d interruptable: ", p->p_pid, vp->savp_id));
630 TAILQ_FOREACH(l2, &vp->savp_woken, l_sleepchain) {
631 vp->savp_woken_count--;
632 l2->l_flag |= LW_SINTR;
633 DPRINTF(("%d ", l2->l_lid));
634 }
635 DPRINTF(("\n"));
636 mutex_exit(&vp->savp_mutex);
637 }
638 }
639 #endif
640
641 retry:
642 /*
643 * Interrupt LWPs in interruptable sleep, unsuspend suspended
644 * LWPs and then wait for everyone else to finish.
645 */
646 LIST_FOREACH(l2, &p->p_lwps, l_sibling) {
647 if (l2 == l)
648 continue;
649 lwp_lock(l2);
650 l2->l_flag &= ~LW_SA;
651 l2->l_flag |= LW_WEXIT;
652 if ((l2->l_stat == LSSLEEP && (l2->l_flag & LW_SINTR)) ||
653 l2->l_stat == LSSUSPENDED || l2->l_stat == LSSTOP) {
654 /* setrunnable() will release the lock. */
655 setrunnable(l2);
656 DPRINTF(("exit_lwps: Made %d.%d runnable\n",
657 p->p_pid, l2->l_lid));
658 continue;
659 }
660 lwp_unlock(l2);
661 }
662 while (p->p_nlwps > 1) {
663 DPRINTF(("exit_lwps: waiting for %d LWPs (%d zombies)\n",
664 p->p_nlwps, p->p_nzlwps));
665 error = lwp_wait1(l, 0, &waited, LWPWAIT_EXITCONTROL);
666 if (p->p_nlwps == 1)
667 break;
668 if (error == EDEADLK) {
669 /*
670 * LWPs can get suspended/slept behind us.
671 * (eg. sa_setwoken)
672 * kick them again and retry.
673 */
674 goto retry;
675 }
676 if (error)
677 panic("exit_lwps: lwp_wait1 failed with error %d",
678 error);
679 DPRINTF(("exit_lwps: Got LWP %d from lwp_wait1()\n", waited));
680 }
681
682 KERNEL_LOCK(nlocks, l);
683 KASSERT(p->p_nlwps == 1);
684 }
685
686 int
687 do_sys_wait(int *pid, int *status, int options, struct rusage *ru)
688 {
689 proc_t *child;
690 int error;
691
692 if (ru != NULL) {
693 memset(ru, 0, sizeof(*ru));
694 }
695 mutex_enter(proc_lock);
696 error = find_stopped_child(curproc, *pid, options, &child, status);
697 if (child == NULL) {
698 mutex_exit(proc_lock);
699 *pid = 0;
700 return error;
701 }
702 *pid = child->p_pid;
703
704 if (child->p_stat == SZOMB) {
705 /* proc_free() will release the proc_lock. */
706 if (options & WNOWAIT) {
707 mutex_exit(proc_lock);
708 } else {
709 proc_free(child, ru);
710 }
711 } else {
712 /* Child state must have been SSTOP. */
713 mutex_exit(proc_lock);
714 *status = W_STOPCODE(*status);
715 }
716 return 0;
717 }
718
719 int
720 sys___wait450(struct lwp *l, const struct sys___wait450_args *uap,
721 register_t *retval)
722 {
723 /* {
724 syscallarg(int) pid;
725 syscallarg(int *) status;
726 syscallarg(int) options;
727 syscallarg(struct rusage *) rusage;
728 } */
729 int error, status, pid = SCARG(uap, pid);
730 struct rusage ru;
731
732 error = do_sys_wait(&pid, &status, SCARG(uap, options),
733 SCARG(uap, rusage) != NULL ? &ru : NULL);
734
735 retval[0] = pid;
736 if (pid == 0) {
737 return error;
738 }
739 if (SCARG(uap, status)) {
740 error = copyout(&status, SCARG(uap, status), sizeof(status));
741 }
742 if (SCARG(uap, rusage) && error == 0) {
743 error = copyout(&ru, SCARG(uap, rusage), sizeof(ru));
744 }
745 return error;
746 }
747
748 /*
749 * Scan list of child processes for a child process that has stopped or
750 * exited. Used by sys_wait4 and 'compat' equivalents.
751 *
752 * Must be called with the proc_lock held, and may release while waiting.
753 */
754 static int
755 find_stopped_child(struct proc *parent, pid_t pid, int options,
756 struct proc **child_p, int *status_p)
757 {
758 struct proc *child, *dead;
759 int error;
760
761 KASSERT(mutex_owned(proc_lock));
762
763 if (options & ~(WUNTRACED|WNOHANG|WALTSIG|WALLSIG)
764 && !(options & WOPTSCHECKED)) {
765 *child_p = NULL;
766 return EINVAL;
767 }
768
769 if (pid == 0 && !(options & WOPTSCHECKED))
770 pid = -parent->p_pgid;
771
772 for (;;) {
773 error = ECHILD;
774 dead = NULL;
775
776 LIST_FOREACH(child, &parent->p_children, p_sibling) {
777 if (pid >= 0) {
778 if (child->p_pid != pid) {
779 child = proc_find_raw(pid);
780 if (child == NULL ||
781 child->p_stat == SIDL ||
782 child->p_pptr != parent) {
783 child = NULL;
784 break;
785 }
786 }
787 } else if (pid != WAIT_ANY && child->p_pgid != -pid) {
788 /* Child not in correct pgrp */
789 continue;
790 }
791
792 /*
793 * Wait for processes with p_exitsig != SIGCHLD
794 * processes only if WALTSIG is set; wait for
795 * processes with p_exitsig == SIGCHLD only
796 * if WALTSIG is clear.
797 */
798 if (((options & WALLSIG) == 0) &&
799 (options & WALTSIG ? child->p_exitsig == SIGCHLD
800 : P_EXITSIG(child) != SIGCHLD)){
801 if (child->p_pid == pid) {
802 child = NULL;
803 break;
804 }
805 continue;
806 }
807
808 error = 0;
809 if ((options & WNOZOMBIE) == 0) {
810 if (child->p_stat == SZOMB)
811 break;
812 if (child->p_stat == SDEAD) {
813 /*
814 * We may occasionally arrive here
815 * after receiving a signal, but
816 * immediately before the child
817 * process is zombified. The wait
818 * will be short, so avoid returning
819 * to userspace.
820 */
821 dead = child;
822 }
823 }
824
825 if (child->p_stat == SSTOP &&
826 child->p_waited == 0 &&
827 (child->p_slflag & PSL_TRACED ||
828 options & WUNTRACED)) {
829 if ((options & WNOWAIT) == 0) {
830 child->p_waited = 1;
831 parent->p_nstopchild--;
832 }
833 break;
834 }
835 if (parent->p_nstopchild == 0 || child->p_pid == pid) {
836 child = NULL;
837 break;
838 }
839 }
840
841 if (child != NULL || error != 0 ||
842 ((options & WNOHANG) != 0 && dead == NULL)) {
843 if (child != NULL) {
844 *status_p = child->p_xstat;
845 }
846 *child_p = child;
847 return error;
848 }
849
850 /*
851 * Wait for another child process to stop.
852 */
853 error = cv_wait_sig(&parent->p_waitcv, proc_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 = p->p_pptr;
872 struct lwp *l;
873 ksiginfo_t ksi;
874 kauth_cred_t cred1, cred2;
875 uid_t uid;
876
877 KASSERT(mutex_owned(proc_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 && p->p_opptr != parent) {
892 mutex_enter(p->p_lock);
893 p->p_slflag &= ~(PSL_TRACED|PSL_FSTRACE|PSL_SYSCALL);
894 mutex_exit(p->p_lock);
895 parent = (p->p_opptr == NULL) ? initproc : p->p_opptr;
896 proc_reparent(p, parent);
897 p->p_opptr = NULL;
898 if (p->p_exitsig != 0) {
899 exit_psignal(p, parent, &ksi);
900 kpsignal(parent, &ksi, NULL);
901 }
902 cv_broadcast(&parent->p_waitcv);
903 mutex_exit(proc_lock);
904 return;
905 }
906
907 sched_proc_exit(parent, p);
908
909 /*
910 * Add child times of exiting process onto its own times.
911 * This cannot be done any earlier else it might get done twice.
912 */
913 l = LIST_FIRST(&p->p_lwps);
914 p->p_stats->p_ru.ru_nvcsw += (l->l_ncsw - l->l_nivcsw);
915 p->p_stats->p_ru.ru_nivcsw += l->l_nivcsw;
916 ruadd(&p->p_stats->p_ru, &l->l_ru);
917 ruadd(&p->p_stats->p_ru, &p->p_stats->p_cru);
918 ruadd(&parent->p_stats->p_cru, &p->p_stats->p_ru);
919 if (ru != NULL)
920 *ru = p->p_stats->p_ru;
921 p->p_xstat = 0;
922
923 /* Release any SA state. */
924 #ifdef KERN_SA
925 if (p->p_sa)
926 sa_release(p);
927 #endif
928
929 /*
930 * At this point we are going to start freeing the final resources.
931 * If anyone tries to access the proc structure after here they will
932 * get a shock - bits are missing. Attempt to make it hard! We
933 * don't bother with any further locking past this point.
934 */
935 p->p_stat = SIDL; /* not even a zombie any more */
936 LIST_REMOVE(p, p_list); /* off zombproc */
937 parent->p_nstopchild--;
938 LIST_REMOVE(p, p_sibling);
939
940 /*
941 * Let pid be reallocated.
942 */
943 proc_free_pid(p->p_pid);
944
945 /*
946 * Unlink process from its process group.
947 * Releases the proc_lock.
948 */
949 proc_leavepgrp(p);
950
951 /*
952 * Delay release until after lwp_free.
953 */
954 cred2 = l->l_cred;
955
956 /*
957 * Free the last LWP's resources.
958 *
959 * lwp_free ensures the LWP is no longer running on another CPU.
960 */
961 lwp_free(l, false, true);
962
963 /*
964 * Now no one except us can reach the process p.
965 */
966
967 /*
968 * Decrement the count of procs running with this uid.
969 */
970 cred1 = p->p_cred;
971 uid = kauth_cred_getuid(cred1);
972 (void)chgproccnt(uid, -1);
973
974 /*
975 * Release substructures.
976 */
977
978 lim_free(p->p_limit);
979 pstatsfree(p->p_stats);
980 kauth_cred_free(cred1);
981 kauth_cred_free(cred2);
982
983 /*
984 * Release reference to text vnode
985 */
986 if (p->p_textvp)
987 vrele(p->p_textvp);
988
989 mutex_destroy(&p->p_auxlock);
990 mutex_obj_free(p->p_lock);
991 mutex_destroy(&p->p_stmutex);
992 cv_destroy(&p->p_waitcv);
993 cv_destroy(&p->p_lwpcv);
994 rw_destroy(&p->p_reflock);
995
996 proc_free_mem(p);
997 }
998
999 /*
1000 * make process 'parent' the new parent of process 'child'.
1001 *
1002 * Must be called with proc_lock held.
1003 */
1004 void
1005 proc_reparent(struct proc *child, struct proc *parent)
1006 {
1007
1008 KASSERT(mutex_owned(proc_lock));
1009
1010 if (child->p_pptr == parent)
1011 return;
1012
1013 if (child->p_stat == SZOMB ||
1014 (child->p_stat == SSTOP && !child->p_waited)) {
1015 child->p_pptr->p_nstopchild--;
1016 parent->p_nstopchild++;
1017 }
1018 if (parent == initproc)
1019 child->p_exitsig = SIGCHLD;
1020
1021 LIST_REMOVE(child, p_sibling);
1022 LIST_INSERT_HEAD(&parent->p_children, child, p_sibling);
1023 child->p_pptr = parent;
1024 child->p_ppid = parent->p_pid;
1025 }
1026