1/* $NetBSD: kern_exit.c,v 1.267 2016/11/13 15:25:01 christos 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 1.267 2016/11/13 15:25:01 christos Exp $");
71
72#include "opt_ktrace.h"
73#include "opt_dtrace.h"
74#include "opt_perfctrs.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/mount.h>
102#include <sys/syscallargs.h>
103#include <sys/kauth.h>
104#include <sys/sleepq.h>
105#include <sys/lockdebug.h>
106#include <sys/ktrace.h>
107#include <sys/cpu.h>
108#include <sys/lwpctl.h>
109#include <sys/atomic.h>
110#include <sys/sdt.h>
111
112#include <uvm/uvm_extern.h>
113
114#ifdef DEBUG_EXIT
115int debug_exit = 0;
116#define DPRINTF(x) if (debug_exit) printf x
117#else
118#define DPRINTF(x)
119#endif
120
121static int find_stopped_child(struct proc *, idtype_t, id_t, int,
122 struct proc **, struct wrusage *, siginfo_t *);
123static void proc_free(struct proc *, struct wrusage *);
124
125/*
126 * DTrace SDT provider definitions
127 */
128SDT_PROVIDER_DECLARE(proc);
129SDT_PROBE_DEFINE1(proc, kernel, , exit, "int");
130
131/*
132 * Fill in the appropriate signal information, and signal the parent.
133 */
134/* XXX noclone works around a gcc 4.5 bug on arm */
135static void __noclone
136exit_psignal(struct proc *p, struct proc *pp, ksiginfo_t *ksi)
137{
138
139 KSI_INIT(ksi);
140 if ((ksi->ksi_signo = P_EXITSIG(p)) == SIGCHLD) {
141 if (p->p_xsig) {
142 if (p->p_sflag & PS_COREDUMP)
143 ksi->ksi_code = CLD_DUMPED;
144 else
145 ksi->ksi_code = CLD_KILLED;
146 ksi->ksi_status = p->p_xsig;
147 } else {
148 ksi->ksi_code = CLD_EXITED;
149 ksi->ksi_status = p->p_xexit;
150 }
151 } else {
152 ksi->ksi_code = SI_USER;
153 ksi->ksi_status = p->p_xsig;
154 }
155 /*
156 * We fill those in, even for non-SIGCHLD.
157 * It's safe to access p->p_cred unlocked here.
158 */
159 ksi->ksi_pid = p->p_pid;
160 ksi->ksi_uid = kauth_cred_geteuid(p->p_cred);
161 /* XXX: is this still valid? */
162 ksi->ksi_utime = p->p_stats->p_ru.ru_utime.tv_sec;
163 ksi->ksi_stime = p->p_stats->p_ru.ru_stime.tv_sec;
164}
165
166/*
167 * exit --
168 * Death of process.
169 */
170int
171sys_exit(struct lwp *l, const struct sys_exit_args *uap, register_t *retval)
172{
173 /* {
174 syscallarg(int) rval;
175 } */
176 struct proc *p = l->l_proc;
177
178 /* Don't call exit1() multiple times in the same process. */
179 mutex_enter(p->p_lock);
180 if (p->p_sflag & PS_WEXIT) {
181 mutex_exit(p->p_lock);
182 lwp_exit(l);
183 }
184
185 /* exit1() will release the mutex. */
186 exit1(l, SCARG(uap, rval), 0);
187 /* NOTREACHED */
188 return (0);
189}
190
191/*
192 * Exit: deallocate address space and other resources, change proc state
193 * to zombie, and unlink proc from allproc and parent's lists. Save exit
194 * status and rusage for wait(). Check for child processes and orphan them.
195 *
196 * Must be called with p->p_lock held. Does not return.
197 */
198void
199exit1(struct lwp *l, int exitcode, int signo)
200{
201 struct proc *p, *child, *next_child, *old_parent, *new_parent;
202 struct pgrp *pgrp;
203 ksiginfo_t ksi;
204 ksiginfoq_t kq;
205 int wakeinit;
206
207 p = l->l_proc;
208
209 KASSERT(mutex_owned(p->p_lock));
210 KASSERT(p->p_vmspace != NULL);
211
212 if (__predict_false(p == initproc)) {
213 panic("init died (signal %d, exit %d)", signo, exitcode);
214 }
215
216 p->p_sflag |= PS_WEXIT;
217
218 /*
219 * Force all other LWPs to exit before we do. Only then can we
220 * begin to tear down the rest of the process state.
221 */
222 if (p->p_nlwps > 1) {
223 exit_lwps(l);
224 }
225
226 ksiginfo_queue_init(&kq);
227
228 /*
229 * If we have been asked to stop on exit, do so now.
230 */
231 if (__predict_false(p->p_sflag & PS_STOPEXIT)) {
232 KERNEL_UNLOCK_ALL(l, &l->l_biglocks);
233 sigclearall(p, &contsigmask, &kq);
234
235 if (!mutex_tryenter(proc_lock)) {
236 mutex_exit(p->p_lock);
237 mutex_enter(proc_lock);
238 mutex_enter(p->p_lock);
239 }
240 p->p_waited = 0;
241 p->p_pptr->p_nstopchild++;
242 p->p_stat = SSTOP;
243 mutex_exit(proc_lock);
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(("%s: %d.%d exiting.\n", __func__, 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 p->p_xexit = exitcode;
309 p->p_xsig = signo;
310
311 /*
312 * If emulation has process exit hook, call it now.
313 * Set the exit status now so that the exit hook has
314 * an opportunity to tweak it (COMPAT_LINUX requires
315 * this for thread group emulation)
316 */
317 if (p->p_emul->e_proc_exit)
318 (*p->p_emul->e_proc_exit)(p);
319
320 /*
321 * Free the VM resources we're still holding on to.
322 * We must do this from a valid thread because doing
323 * so may block. This frees vmspace, which we don't
324 * need anymore. The only remaining lwp is the one
325 * we run at this moment, nothing runs in userland
326 * anymore.
327 */
328 uvm_proc_exit(p);
329
330 /*
331 * Stop profiling.
332 */
333 if (__predict_false((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, PL_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(proc_lock);
345 if (p->p_lflag & PL_PPWAIT) {
346#if 0
347 lwp_t *lp;
348
349 l->l_lwpctl = NULL; /* was on loan from blocked parent */
350 p->p_lflag &= ~PL_PPWAIT;
351
352 lp = p->p_vforklwp;
353 p->p_vforklwp = NULL;
354 lp->l_pflag &= ~LP_VFORKWAIT; /* XXX */
355 cv_broadcast(&lp->l_waitcv);
356#else
357 l->l_lwpctl = NULL; /* was on loan from blocked parent */
358 p->p_lflag &= ~PL_PPWAIT;
359 cv_broadcast(&p->p_pptr->p_waitcv);
360#endif
361 }
362
363 if (SESS_LEADER(p)) {
364 struct vnode *vprele = NULL, *vprevoke = NULL;
365 struct session *sp = p->p_session;
366 struct tty *tp;
367
368 if (sp->s_ttyvp) {
369 /*
370 * Controlling process.
371 * Signal foreground pgrp,
372 * drain controlling terminal
373 * and revoke access to controlling terminal.
374 */
375 tp = sp->s_ttyp;
376 mutex_spin_enter(&tty_lock);
377 if (tp->t_session == sp) {
378 /* we can't guarantee the revoke will do this */
379 pgrp = tp->t_pgrp;
380 tp->t_pgrp = NULL;
381 tp->t_session = NULL;
382 mutex_spin_exit(&tty_lock);
383 if (pgrp != NULL) {
384 pgsignal(pgrp, SIGHUP, 1);
385 }
386 mutex_exit(proc_lock);
387 (void) ttywait(tp);
388 mutex_enter(proc_lock);
389
390 /* The tty could have been revoked. */
391 vprevoke = sp->s_ttyvp;
392 } else
393 mutex_spin_exit(&tty_lock);
394 vprele = sp->s_ttyvp;
395 sp->s_ttyvp = NULL;
396 /*
397 * s_ttyp is not zero'd; we use this to indicate
398 * that the session once had a controlling terminal.
399 * (for logging and informational purposes)
400 */
401 }
402 sp->s_leader = NULL;
403
404 if (vprevoke != NULL || vprele != NULL) {
405 if (vprevoke != NULL) {
406 /* Releases proc_lock. */
407 proc_sessrele(sp);
408 VOP_REVOKE(vprevoke, REVOKEALL);
409 } else
410 mutex_exit(proc_lock);
411 if (vprele != NULL)
412 vrele(vprele);
413 mutex_enter(proc_lock);
414 }
415 }
416 fixjobc(p, p->p_pgrp, 0);
417
418 /*
419 * Finalize the last LWP's specificdata, as well as the
420 * specificdata for the proc itself.
421 */
422 lwp_finispecific(l);
423 proc_finispecific(p);
424
425 /*
426 * Notify interested parties of our demise.
427 */
428 KNOTE(&p->p_klist, NOTE_EXIT);
429
430 SDT_PROBE(proc, kernel, , exit,
431 ((p->p_sflag & PS_COREDUMP) ? CLD_DUMPED :
432 (p->p_xsig ? CLD_KILLED : CLD_EXITED)),
433 0,0,0,0);
434
435#if PERFCTRS
436 /*
437 * Save final PMC information in parent process & clean up.
438 */
439 if (PMC_ENABLED(p)) {
440 pmc_save_context(p);
441 pmc_accumulate(p->p_pptr, p);
442 pmc_process_exit(p);
443 }
444#endif
445
446 /*
447 * Reset p_opptr pointer of all former children which got
448 * traced by another process and were reparented. We reset
449 * it to NULL here; the trace detach code then reparents
450 * the child to initproc. We only check allproc list, since
451 * eventual former children on zombproc list won't reference
452 * p_opptr anymore.
453 */
454 if (__predict_false(p->p_slflag & PSL_CHTRACED)) {
455 struct proc *q;
456 PROCLIST_FOREACH(q, &allproc) {
457 if (q->p_opptr == p)
458 q->p_opptr = NULL;
459 }
460 PROCLIST_FOREACH(q, &zombproc) {
461 if (q->p_opptr == p)
462 q->p_opptr = NULL;
463 }
464 }
465
466 /*
467 * Give orphaned children to init(8).
468 */
469 child = LIST_FIRST(&p->p_children);
470 wakeinit = (child != NULL);
471 for (; child != NULL; child = next_child) {
472 next_child = LIST_NEXT(child, p_sibling);
473
474 /*
475 * Traced processes are killed since their existence
476 * means someone is screwing up. Since we reset the
477 * trace flags, the logic in sys_wait4() would not be
478 * triggered to reparent the process to its
479 * original parent, so we must do this here.
480 */
481 if (__predict_false(child->p_slflag & PSL_TRACED)) {
482 mutex_enter(p->p_lock);
483 child->p_slflag &=
484 ~(PSL_TRACED|PSL_FSTRACE|PSL_SYSCALL);
485 mutex_exit(p->p_lock);
486 if (child->p_opptr != child->p_pptr) {
487 struct proc *t = child->p_opptr;
488 proc_reparent(child, t ? t : initproc);
489 child->p_opptr = NULL;
490 } else
491 proc_reparent(child, initproc);
492 killproc(child, "orphaned traced process");
493 } else
494 proc_reparent(child, initproc);
495 }
496
497 /*
498 * Move proc from allproc to zombproc, it's now nearly ready to be
499 * collected by parent.
500 */
501 LIST_REMOVE(l, l_list);
502 LIST_REMOVE(p, p_list);
503 LIST_INSERT_HEAD(&zombproc, p, p_list);
504
505 /*
506 * Mark the process as dead. We must do this before we signal
507 * the parent.
508 */
509 p->p_stat = SDEAD;
510
511 /* Put in front of parent's sibling list for parent to collect it */
512 old_parent = p->p_pptr;
513 old_parent->p_nstopchild++;
514 if (LIST_FIRST(&old_parent->p_children) != p) {
515 /* Put child where it can be found quickly */
516 LIST_REMOVE(p, p_sibling);
517 LIST_INSERT_HEAD(&old_parent->p_children, p, p_sibling);
518 }
519
520 /*
521 * Notify parent that we're gone. If parent has the P_NOCLDWAIT
522 * flag set, notify init instead (and hope it will handle
523 * this situation).
524 */
525 if (old_parent->p_flag & (PK_NOCLDWAIT|PK_CLDSIGIGN)) {
526 proc_reparent(p, initproc);
527 wakeinit = 1;
528
529 /*
530 * If this was the last child of our parent, notify
531 * parent, so in case he was wait(2)ing, he will
532 * continue.
533 */
534 if (LIST_FIRST(&old_parent->p_children) == NULL)
535 cv_broadcast(&old_parent->p_waitcv);
536 }
537
538 /* Reload parent pointer, since p may have been reparented above */
539 new_parent = p->p_pptr;
540
541 if (__predict_false((p->p_slflag & PSL_FSTRACE) == 0 &&
542 p->p_exitsig != 0)) {
543 exit_psignal(p, new_parent, &ksi);
544 kpsignal(new_parent, &ksi, NULL);
545 }
546
547 /* Calculate the final rusage info. */
548 calcru(p, &p->p_stats->p_ru.ru_utime, &p->p_stats->p_ru.ru_stime,
549 NULL, NULL);
550
551 if (wakeinit)
552 cv_broadcast(&initproc->p_waitcv);
553
554 callout_destroy(&l->l_timeout_ch);
555
556 /*
557 * Release any PCU resources before becoming a zombie.
558 */
559 pcu_discard_all(l);
560
561 mutex_enter(p->p_lock);
562 /* Free the linux lwp id */
563 if ((l->l_pflag & LP_PIDLID) != 0 && l->l_lid != p->p_pid)
564 proc_free_pid(l->l_lid);
565 lwp_drainrefs(l);
566 lwp_lock(l);
567 l->l_prflag &= ~LPR_DETACHED;
568 l->l_stat = LSZOMB;
569 lwp_unlock(l);
570 KASSERT(curlwp == l);
571 KASSERT(p->p_nrlwps == 1);
572 KASSERT(p->p_nlwps == 1);
573 p->p_stat = SZOMB;
574 p->p_nrlwps--;
575 p->p_nzlwps++;
576 p->p_ndlwps = 0;
577 mutex_exit(p->p_lock);
578
579 /*
580 * Signal the parent to collect us, and drop the proclist lock.
581 * Drop debugger/procfs lock; no new references can be gained.
582 */
583 cv_broadcast(&p->p_pptr->p_waitcv);
584 rw_exit(&p->p_reflock);
585 mutex_exit(proc_lock);
586
587 /* Verify that we hold no locks other than the kernel lock. */
588 LOCKDEBUG_BARRIER(&kernel_lock, 0);
589
590 /*
591 * NOTE: WE ARE NO LONGER ALLOWED TO SLEEP!
592 */
593
594 /*
595 * Give machine-dependent code a chance to free any MD LWP
596 * resources. This must be done before uvm_lwp_exit(), in
597 * case these resources are in the PCB.
598 */
599 cpu_lwp_free(l, 1);
600
601 pmap_deactivate(l);
602
603 /* This process no longer needs to hold the kernel lock. */
604#ifdef notyet
605 /* XXXSMP hold in lwp_userret() */
606 KERNEL_UNLOCK_LAST(l);
607#else
608 KERNEL_UNLOCK_ALL(l, NULL);
609#endif
610
611 lwp_exit_switchaway(l);
612}
613
614void
615exit_lwps(struct lwp *l)
616{
617 proc_t *p = l->l_proc;
618 lwp_t *l2;
619 int nlocks;
620
621 KERNEL_UNLOCK_ALL(l, &nlocks);
622retry:
623 KASSERT(mutex_owned(p->p_lock));
624
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 continue;
639 }
640 lwp_unlock(l2);
641 }
642
643 /*
644 * Wait for every LWP to exit. Note: LWPs can get suspended/slept
645 * behind us or there may even be new LWPs created. Therefore, a
646 * full retry is required on error.
647 */
648 while (p->p_nlwps > 1) {
649 if (lwp_wait(l, 0, NULL, true)) {
650 goto retry;
651 }
652 }
653
654 KERNEL_LOCK(nlocks, l);
655 KASSERT(p->p_nlwps == 1);
656}
657
658int
659do_sys_waitid(idtype_t idtype, id_t id, int *pid, int *status, int options,
660 struct wrusage *wru, siginfo_t *si)
661{
662 proc_t *child;
663 int error;
664
665
666 if (wru != NULL)
667 memset(wru, 0, sizeof(*wru));
668 if (si != NULL)
669 memset(si, 0, sizeof(*si));
670
671 mutex_enter(proc_lock);
672 error = find_stopped_child(curproc, idtype, id, options, &child,
673 wru, si);
674 if (child == NULL) {
675 mutex_exit(proc_lock);
676 *pid = 0;
677 return error;
678 }
679 *pid = child->p_pid;
680
681 if (child->p_stat == SZOMB) {
682 /* Child is exiting */
683 *status = P_WAITSTATUS(child);
684 /* proc_free() will release the proc_lock. */
685 if (options & WNOWAIT) {
686 mutex_exit(proc_lock);
687 } else {
688 proc_free(child, wru);
689 }
690 } else {
691 /* Don't mark SIGCONT if we are being stopped */
692 *status = (child->p_xsig == SIGCONT && child->p_stat != SSTOP) ?
693 W_CONTCODE() : W_STOPCODE(child->p_xsig);
694 mutex_exit(proc_lock);
695 }
696 return 0;
697}
698
699int
700do_sys_wait(int *pid, int *status, int options, struct rusage *ru)
701{
702 idtype_t idtype;
703 id_t id;
704 int ret;
705 struct wrusage wru;
706
707 /*
708 * Translate the special pid values into the (idtype, pid)
709 * pair for wait6. The WAIT_MYPGRP case is handled by
710 * find_stopped_child() on its own.
711 */
712 if (*pid == WAIT_ANY) {
713 idtype = P_ALL;
714 id = 0;
715 } else if (*pid < 0) {
716 idtype = P_PGID;
717 id = (id_t)-*pid;
718 } else {
719 idtype = P_PID;
720 id = (id_t)*pid;
721 }
722 options |= WEXITED | WTRAPPED;
723 ret = do_sys_waitid(idtype, id, pid, status, options, ru ? &wru : NULL,
724 NULL);
725 if (ru)
726 *ru = wru.wru_self;
727 return ret;
728}
729
730int
731sys___wait450(struct lwp *l, const struct sys___wait450_args *uap,
732 register_t *retval)
733{
734 /* {
735 syscallarg(int) pid;
736 syscallarg(int *) status;
737 syscallarg(int) options;
738 syscallarg(struct rusage *) rusage;
739 } */
740 int error, status, pid = SCARG(uap, pid);
741 struct rusage ru;
742
743 error = do_sys_wait(&pid, &status, SCARG(uap, options),
744 SCARG(uap, rusage) != NULL ? &ru : NULL);
745
746 retval[0] = pid;
747 if (pid == 0) {
748 return error;
749 }
750 if (SCARG(uap, status)) {
751 error = copyout(&status, SCARG(uap, status), sizeof(status));
752 }
753 if (SCARG(uap, rusage) && error == 0) {
754 error = copyout(&ru, SCARG(uap, rusage), sizeof(ru));
755 }
756 return error;
757}
758
759int
760sys_wait6(struct lwp *l, const struct sys_wait6_args *uap, register_t *retval)
761{
762 /* {
763 syscallarg(idtype_t) idtype;
764 syscallarg(id_t) id;
765 syscallarg(int *) status;
766 syscallarg(int) options;
767 syscallarg(struct wrusage *) wru;
768 syscallarg(siginfo_t *) si;
769 } */
770 struct wrusage wru, *wrup;
771 siginfo_t si, *sip;
772 idtype_t idtype;
773 int pid;
774 id_t id;
775 int error, status;
776
777 idtype = SCARG(uap, idtype);
778 id = SCARG(uap, id);
779
780 if (SCARG(uap, wru) != NULL)
781 wrup = &wru;
782 else
783 wrup = NULL;
784
785 if (SCARG(uap, info) != NULL)
786 sip = &si;
787 else
788 sip = NULL;
789
790 /*
791 * We expect all callers of wait6() to know about WEXITED and
792 * WTRAPPED.
793 */
794 error = do_sys_waitid(idtype, id, &pid, &status, SCARG(uap, options),
795 wrup, sip);
796
797 retval[0] = pid; /* tell userland who it was */
798
799#if 0
800 /*
801 * should we copyout if there was no process, hence no useful data?
802 * We don't for an old sytle wait4() (etc) but I believe
803 * FreeBSD does for wait6(), so a tossup... Go with FreeBSD for now.
804 */
805 if (pid == 0)
806 return error;
807#endif
808
809 if (SCARG(uap, status) != NULL && error == 0)
810 error = copyout(&status, SCARG(uap, status), sizeof(status));
811 if (SCARG(uap, wru) != NULL && error == 0)
812 error = copyout(&wru, SCARG(uap, wru), sizeof(wru));
813 if (SCARG(uap, info) != NULL && error == 0)
814 error = copyout(&si, SCARG(uap, info), sizeof(si));
815 return error;
816}
817
818
819/*
820 * Find a process that matches the provided criteria, and fill siginfo
821 * and resources if found.
822 * Returns:
823 * -1: Not found, abort early
824 * 0: Not matched
825 * 1: Matched, there might be more matches
826 * 2: This is the only match
827 */
828static int
829match_process(const struct proc *pp, struct proc **q, idtype_t idtype, id_t id,
830 int options, struct wrusage *wrusage, siginfo_t *siginfo)
831{
832 struct rusage *rup;
833 struct proc *p = *q;
834 int rv = 1;
835
836 mutex_enter(p->p_lock);
837 switch (idtype) {
838 case P_ALL:
839 break;
840 case P_PID:
841 if (p->p_pid != (pid_t)id) {
842 mutex_exit(p->p_lock);
843 p = *q = proc_find_raw((pid_t)id);
844 if (p == NULL || p->p_stat == SIDL || p->p_pptr != pp) {
845 *q = NULL;
846 return -1;
847 }
848 mutex_enter(p->p_lock);
849 }
850 rv++;
851 break;
852 case P_PGID:
853 if (p->p_pgid != (pid_t)id)
854 goto out;
855 break;
856 case P_SID:
857 if (p->p_session->s_sid != (pid_t)id)
858 goto out;
859 break;
860 case P_UID:
861 if (kauth_cred_geteuid(p->p_cred) != (uid_t)id)
862 goto out;
863 break;
864 case P_GID:
865 if (kauth_cred_getegid(p->p_cred) != (gid_t)id)
866 goto out;
867 break;
868 case P_CID:
869 case P_PSETID:
870 case P_CPUID:
871 /* XXX: Implement me */
872 default:
873 out:
874 mutex_exit(p->p_lock);
875 return 0;
876 }
877
878 if ((options & WEXITED) == 0 && p->p_stat == SZOMB)
879 goto out;
880
881 if (siginfo != NULL) {
882 siginfo->si_errno = 0;
883
884 /*
885 * SUSv4 requires that the si_signo value is always
886 * SIGCHLD. Obey it despite the rfork(2) interface
887 * allows to request other signal for child exit
888 * notification.
889 */
890 siginfo->si_signo = SIGCHLD;
891
892 /*
893 * This is still a rough estimate. We will fix the
894 * cases TRAPPED, STOPPED, and CONTINUED later.
895 */
896 if (p->p_sflag & PS_COREDUMP) {
897 siginfo->si_code = CLD_DUMPED;
898 siginfo->si_status = p->p_xsig;
899 } else if (p->p_xsig) {
900 siginfo->si_code = CLD_KILLED;
901 siginfo->si_status = p->p_xsig;
902 } else {
903 siginfo->si_code = CLD_EXITED;
904 siginfo->si_status = p->p_xexit;
905 }
906
907 siginfo->si_pid = p->p_pid;
908 siginfo->si_uid = kauth_cred_geteuid(p->p_cred);
909 siginfo->si_utime = p->p_stats->p_ru.ru_utime.tv_sec;
910 siginfo->si_stime = p->p_stats->p_ru.ru_stime.tv_sec;
911 }
912
913 /*
914 * There should be no reason to limit resources usage info to
915 * exited processes only. A snapshot about any resources used
916 * by a stopped process may be exactly what is needed.
917 */
918 if (wrusage != NULL) {
919 rup = &wrusage->wru_self;
920 *rup = p->p_stats->p_ru;
921 calcru(p, &rup->ru_utime, &rup->ru_stime, NULL, NULL);
922
923 rup = &wrusage->wru_children;
924 *rup = p->p_stats->p_cru;
925 calcru(p, &rup->ru_utime, &rup->ru_stime, NULL, NULL);
926 }
927
928 mutex_exit(p->p_lock);
929 return rv;
930}
931
932/*
933 * Determine if there are existing processes being debugged
934 * that used to be (and sometime later will be again) children
935 * of a specific parent (while matching wait criteria)
936 */
937static bool
938debugged_child_exists(idtype_t idtype, id_t id, int options, siginfo_t *si,
939 const struct proc *parent)
940{
941 struct proc *pp;
942
943 /*
944 * If we are searching for a specific pid, we can optimise a little
945 */
946 if (idtype == P_PID) {
947 /*
948 * Check the specific process to see if its real parent is us
949 */
950 pp = proc_find_raw((pid_t)id);
951 if (pp != NULL && pp->p_stat != SIDL && pp->p_opptr == parent) {
952 /*
953 * using P_ALL here avoids match_process() doing the
954 * same work that we just did, but incorrectly for
955 * this scenario.
956 */
957 if (match_process(parent, &pp, P_ALL, id, options,
958 NULL, si))
959 return true;
960 }
961 return false;
962 }
963
964 /*
965 * For the hard cases, just look everywhere to see if some
966 * stolen (reparented) process is really our lost child.
967 * Then check if that process could satisfy the wait conditions.
968 */
969
970 /*
971 * XXX inefficient, but hopefully fairly rare.
972 * XXX should really use a list of reparented processes.
973 */
974 PROCLIST_FOREACH(pp, &allproc) {
975 if (pp->p_stat == SIDL) /* XXX impossible ?? */
976 continue;
977 if (pp->p_opptr == parent &&
978 match_process(parent, &pp, idtype, id, options, NULL, si))
979 return true;
980 }
981 PROCLIST_FOREACH(pp, &zombproc) {
982 if (pp->p_stat == SIDL) /* XXX impossible ?? */
983 continue;
984 if (pp->p_opptr == parent &&
985 match_process(parent, &pp, idtype, id, options, NULL, si))
986 return true;
987 }
988
989 return false;
990}
991
992/*
993 * Scan list of child processes for a child process that has stopped or
994 * exited. Used by sys_wait4 and 'compat' equivalents.
995 *
996 * Must be called with the proc_lock held, and may release while waiting.
997 */
998static int
999find_stopped_child(struct proc *parent, idtype_t idtype, id_t id, int options,
1000 struct proc **child_p, struct wrusage *wru, siginfo_t *si)
1001{
1002 struct proc *child, *dead;
1003 int error;
1004
1005 KASSERT(mutex_owned(proc_lock));
1006
1007 if (options & ~WALLOPTS) {
1008 *child_p = NULL;
1009 return EINVAL;
1010 }
1011
1012 if ((options & WSELECTOPTS) == 0) {
1013 /*
1014 * We will be unable to find any matching processes,
1015 * because there are no known events to look for.
1016 * Prefer to return error instead of blocking
1017 * indefinitely.
1018 */
1019 *child_p = NULL;
1020 return EINVAL;
1021 }
1022
1023 if ((pid_t)id == WAIT_MYPGRP && (idtype == P_PID || idtype == P_PGID)) {
1024 mutex_enter(parent->p_lock);
1025 id = (id_t)parent->p_pgid;
1026 mutex_exit(parent->p_lock);
1027 idtype = P_PGID;
1028 }
1029
1030 for (;;) {
1031 error = ECHILD;
1032 dead = NULL;
1033
1034 LIST_FOREACH(child, &parent->p_children, p_sibling) {
1035 int rv = match_process(parent, &child, idtype, id,
1036 options, wru, si);
1037 if (rv == -1)
1038 break;
1039 if (rv == 0)
1040 continue;
1041
1042 /*
1043 * Wait for processes with p_exitsig != SIGCHLD
1044 * processes only if WALTSIG is set; wait for
1045 * processes with p_exitsig == SIGCHLD only
1046 * if WALTSIG is clear.
1047 */
1048 if (((options & WALLSIG) == 0) &&
1049 (options & WALTSIG ? child->p_exitsig == SIGCHLD
1050 : P_EXITSIG(child) != SIGCHLD)){
1051 if (rv == 2) {
1052 child = NULL;
1053 break;
1054 }
1055 continue;
1056 }
1057
1058 error = 0;
1059 if ((options & WNOZOMBIE) == 0) {
1060 if (child->p_stat == SZOMB)
1061 break;
1062 if (child->p_stat == SDEAD) {
1063 /*
1064 * We may occasionally arrive here
1065 * after receiving a signal, but
1066 * immediately before the child
1067 * process is zombified. The wait
1068 * will be short, so avoid returning
1069 * to userspace.
1070 */
1071 dead = child;
1072 }
1073 }
1074
1075 if ((options & WCONTINUED) != 0 &&
1076 child->p_xsig == SIGCONT &&
1077 (child->p_sflag & PS_CONTINUED)) {
1078 if ((options & WNOWAIT) == 0) {
1079 child->p_sflag &= ~PS_CONTINUED;
1080 child->p_waited = 1;
1081 parent->p_nstopchild--;
1082 }
1083 if (si) {
1084 si->si_status = child->p_xsig;
1085 si->si_code = CLD_CONTINUED;
1086 }
1087 break;
1088 }
1089
1090 if ((options & (WTRAPPED|WSTOPPED)) != 0 &&
1091 child->p_stat == SSTOP &&
1092 child->p_waited == 0 &&
1093 ((child->p_slflag & PSL_TRACED) ||
1094 options & (WUNTRACED|WSTOPPED))) {
1095 if ((options & WNOWAIT) == 0) {
1096 child->p_waited = 1;
1097 parent->p_nstopchild--;
1098 }
1099 if (si) {
1100 si->si_status = child->p_xsig;
1101 si->si_code =
1102 (child->p_slflag & PSL_TRACED) ?
1103 CLD_TRAPPED : CLD_STOPPED;
1104 }
1105 break;
1106 }
1107 if (parent->p_nstopchild == 0 || rv == 2) {
1108 child = NULL;
1109 break;
1110 }
1111 }
1112
1113 /*
1114 * If we found nothing, but we are the bereaved parent
1115 * of a stolen child, look and see if that child (or
1116 * one of them) meets our search criteria. If so, then
1117 * we cannot succeed, but we can hang (wait...),
1118 * or if WNOHANG, return 0 instead of ECHILD
1119 */
1120 if (child == NULL && error == ECHILD &&
1121 (parent->p_slflag & PSL_CHTRACED) &&
1122 debugged_child_exists(idtype, id, options, si, parent))
1123 error = 0;
1124
1125 if (child != NULL || error != 0 ||
1126 ((options & WNOHANG) != 0 && dead == NULL)) {
1127 *child_p = child;
1128 return error;
1129 }
1130
1131 /*
1132 * Wait for another child process to stop.
1133 */
1134 error = cv_wait_sig(&parent->p_waitcv, proc_lock);
1135
1136 if (error != 0) {
1137 *child_p = NULL;
1138 return error;
1139 }
1140 }
1141}
1142
1143/*
1144 * Free a process after parent has taken all the state info. Must be called
1145 * with the proclist lock held, and will release before returning.
1146 *
1147 * *ru is returned to the caller, and must be freed by the caller.
1148 */
1149static void
1150proc_free(struct proc *p, struct wrusage *wru)
1151{
1152 struct proc *parent = p->p_pptr;
1153 struct lwp *l;
1154 ksiginfo_t ksi;
1155 kauth_cred_t cred1, cred2;
1156 uid_t uid;
1157
1158 KASSERT(mutex_owned(proc_lock));
1159 KASSERT(p->p_nlwps == 1);
1160 KASSERT(p->p_nzlwps == 1);
1161 KASSERT(p->p_nrlwps == 0);
1162 KASSERT(p->p_stat == SZOMB);
1163
1164 /*
1165 * If we got the child via ptrace(2) or procfs, and
1166 * the parent is different (meaning the process was
1167 * attached, rather than run as a child), then we need
1168 * to give it back to the old parent, and send the
1169 * parent the exit signal. The rest of the cleanup
1170 * will be done when the old parent waits on the child.
1171 */
1172 if ((p->p_slflag & PSL_TRACED) != 0 && p->p_opptr != parent) {
1173 mutex_enter(p->p_lock);
1174 p->p_slflag &= ~(PSL_TRACED|PSL_FSTRACE|PSL_SYSCALL);
1175 mutex_exit(p->p_lock);
1176 parent = (p->p_opptr == NULL) ? initproc : p->p_opptr;
1177 proc_reparent(p, parent);
1178 p->p_opptr = NULL;
1179 if (p->p_exitsig != 0) {
1180 exit_psignal(p, parent, &ksi);
1181 kpsignal(parent, &ksi, NULL);
1182 }
1183 cv_broadcast(&parent->p_waitcv);
1184 mutex_exit(proc_lock);
1185 return;
1186 }
1187
1188 sched_proc_exit(parent, p);
1189
1190 /*
1191 * Add child times of exiting process onto its own times.
1192 * This cannot be done any earlier else it might get done twice.
1193 */
1194 l = LIST_FIRST(&p->p_lwps);
1195 p->p_stats->p_ru.ru_nvcsw += (l->l_ncsw - l->l_nivcsw);
1196 p->p_stats->p_ru.ru_nivcsw += l->l_nivcsw;
1197 ruadd(&p->p_stats->p_ru, &l->l_ru);
1198 ruadd(&p->p_stats->p_ru, &p->p_stats->p_cru);
1199 ruadd(&parent->p_stats->p_cru, &p->p_stats->p_ru);
1200 if (wru != NULL) {
1201 wru->wru_self = p->p_stats->p_ru;
1202 wru->wru_children = p->p_stats->p_cru;
1203 }
1204 p->p_xsig = 0;
1205 p->p_xexit = 0;
1206
1207 /*
1208 * At this point we are going to start freeing the final resources.
1209 * If anyone tries to access the proc structure after here they will
1210 * get a shock - bits are missing. Attempt to make it hard! We
1211 * don't bother with any further locking past this point.
1212 */
1213 p->p_stat = SIDL; /* not even a zombie any more */
1214 LIST_REMOVE(p, p_list); /* off zombproc */
1215 parent->p_nstopchild--;
1216 LIST_REMOVE(p, p_sibling);
1217
1218 /*
1219 * Let pid be reallocated.
1220 */
1221 proc_free_pid(p->p_pid);
1222
1223 /*
1224 * Unlink process from its process group.
1225 * Releases the proc_lock.
1226 */
1227 proc_leavepgrp(p);
1228
1229 /*
1230 * Delay release until after lwp_free.
1231 */
1232 cred2 = l->l_cred;
1233
1234 /*
1235 * Free the last LWP's resources.
1236 *
1237 * lwp_free ensures the LWP is no longer running on another CPU.
1238 */
1239 lwp_free(l, false, true);
1240
1241 /*
1242 * Now no one except us can reach the process p.
1243 */
1244
1245 /*
1246 * Decrement the count of procs running with this uid.
1247 */
1248 cred1 = p->p_cred;
1249 uid = kauth_cred_getuid(cred1);
1250 (void)chgproccnt(uid, -1);
1251
1252 /*
1253 * Release substructures.
1254 */
1255
1256 lim_free(p->p_limit);
1257 pstatsfree(p->p_stats);
1258 kauth_cred_free(cred1);
1259 kauth_cred_free(cred2);
1260
1261 /*
1262 * Release reference to text vnode
1263 */
1264 if (p->p_textvp)
1265 vrele(p->p_textvp);
1266
1267 mutex_destroy(&p->p_auxlock);
1268 mutex_obj_free(p->p_lock);
1269 mutex_destroy(&p->p_stmutex);
1270 cv_destroy(&p->p_waitcv);
1271 cv_destroy(&p->p_lwpcv);
1272 rw_destroy(&p->p_reflock);
1273
1274 proc_free_mem(p);
1275}
1276
1277/*
1278 * Change the parent of a process for tracing purposes.
1279 */
1280void
1281proc_changeparent(struct proc *t, struct proc *p)
1282{
1283 SET(t->p_slflag, PSL_TRACED);
1284 t->p_opptr = t->p_pptr;
1285 if (t->p_pptr == p)
1286 return;
1287 struct proc *parent = t->p_pptr;
1288
1289 if (parent->p_lock < t->p_lock) {
1290 if (!mutex_tryenter(parent->p_lock)) {
1291 mutex_exit(t->p_lock);
1292 mutex_enter(parent->p_lock);
1293 mutex_enter(t->p_lock);
1294 }
1295 } else if (parent->p_lock > t->p_lock) {
1296 mutex_enter(parent->p_lock);
1297 }
1298 parent->p_slflag |= PSL_CHTRACED;
1299 proc_reparent(t, p);
1300 if (parent->p_lock != t->p_lock)
1301 mutex_exit(parent->p_lock);
1302}
1303
1304/*
1305 * make process 'parent' the new parent of process 'child'.
1306 *
1307 * Must be called with proc_lock held.
1308 */
1309void
1310proc_reparent(struct proc *child, struct proc *parent)
1311{
1312
1313 KASSERT(mutex_owned(proc_lock));
1314
1315 if (child->p_pptr == parent)
1316 return;
1317
1318 if (child->p_stat == SZOMB || child->p_stat == SDEAD ||
1319 (child->p_stat == SSTOP && !child->p_waited)) {
1320 child->p_pptr->p_nstopchild--;
1321 parent->p_nstopchild++;
1322 }
1323 if (parent == initproc) {
1324 child->p_exitsig = SIGCHLD;
1325 child->p_ppid = parent->p_pid;
1326 }
1327
1328 LIST_REMOVE(child, p_sibling);
1329 LIST_INSERT_HEAD(&parent->p_children, child, p_sibling);
1330 child->p_pptr = parent;
1331}
1332