kern_exit.c revision 1.284 1 /* $NetBSD: kern_exit.c,v 1.284 2020/02/22 21:07:46 ad Exp $ */
2
3 /*-
4 * Copyright (c) 1998, 1999, 2006, 2007, 2008, 2020 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 1.284 2020/02/22 21:07:46 ad Exp $");
71
72 #include "opt_ktrace.h"
73 #include "opt_dtrace.h"
74 #include "opt_sysv.h"
75
76 #include <sys/param.h>
77 #include <sys/systm.h>
78 #include <sys/ioctl.h>
79 #include <sys/tty.h>
80 #include <sys/time.h>
81 #include <sys/resource.h>
82 #include <sys/kernel.h>
83 #include <sys/proc.h>
84 #include <sys/buf.h>
85 #include <sys/wait.h>
86 #include <sys/file.h>
87 #include <sys/fstrans.h>
88 #include <sys/vnode.h>
89 #include <sys/syslog.h>
90 #include <sys/pool.h>
91 #include <sys/uidinfo.h>
92 #include <sys/ptrace.h>
93 #include <sys/acct.h>
94 #include <sys/filedesc.h>
95 #include <sys/ras.h>
96 #include <sys/signalvar.h>
97 #include <sys/sched.h>
98 #include <sys/mount.h>
99 #include <sys/syscallargs.h>
100 #include <sys/kauth.h>
101 #include <sys/sleepq.h>
102 #include <sys/lock.h>
103 #include <sys/lockdebug.h>
104 #include <sys/ktrace.h>
105 #include <sys/cpu.h>
106 #include <sys/lwpctl.h>
107 #include <sys/atomic.h>
108 #include <sys/sdt.h>
109 #include <sys/psref.h>
110
111 #include <uvm/uvm_extern.h>
112
113 #ifdef DEBUG_EXIT
114 int debug_exit = 0;
115 #define DPRINTF(x) if (debug_exit) printf x
116 #else
117 #define DPRINTF(x)
118 #endif
119
120 static int find_stopped_child(struct proc *, idtype_t, id_t, int,
121 struct proc **, struct wrusage *, siginfo_t *);
122 static void proc_free(struct proc *, struct wrusage *);
123
124 /*
125 * DTrace SDT provider definitions
126 */
127 SDT_PROVIDER_DECLARE(proc);
128 SDT_PROBE_DEFINE1(proc, kernel, , exit, "int");
129
130 /*
131 * Fill in the appropriate signal information, and signal the parent.
132 */
133 /* XXX noclone works around a gcc 4.5 bug on arm */
134 static void __noclone
135 exit_psignal(struct proc *p, struct proc *pp, ksiginfo_t *ksi)
136 {
137
138 KSI_INIT(ksi);
139 if ((ksi->ksi_signo = P_EXITSIG(p)) == SIGCHLD) {
140 if (p->p_xsig) {
141 if (p->p_sflag & PS_COREDUMP)
142 ksi->ksi_code = CLD_DUMPED;
143 else
144 ksi->ksi_code = CLD_KILLED;
145 ksi->ksi_status = p->p_xsig;
146 } else {
147 ksi->ksi_code = CLD_EXITED;
148 ksi->ksi_status = p->p_xexit;
149 }
150 } else {
151 ksi->ksi_code = SI_USER;
152 ksi->ksi_status = p->p_xsig;
153 }
154 /*
155 * We fill those in, even for non-SIGCHLD.
156 * It's safe to access p->p_cred unlocked here.
157 */
158 ksi->ksi_pid = p->p_pid;
159 ksi->ksi_uid = kauth_cred_geteuid(p->p_cred);
160 /* XXX: is this still valid? */
161 ksi->ksi_utime = p->p_stats->p_ru.ru_utime.tv_sec;
162 ksi->ksi_stime = p->p_stats->p_ru.ru_stime.tv_sec;
163 }
164
165 /*
166 * exit --
167 * Death of process.
168 */
169 int
170 sys_exit(struct lwp *l, const struct sys_exit_args *uap, register_t *retval)
171 {
172 /* {
173 syscallarg(int) rval;
174 } */
175 struct proc *p = l->l_proc;
176
177 /* Don't call exit1() multiple times in the same process. */
178 mutex_enter(p->p_lock);
179 if (p->p_sflag & PS_WEXIT) {
180 mutex_exit(p->p_lock);
181 lwp_exit(l);
182 }
183
184 /* exit1() will release the mutex. */
185 exit1(l, SCARG(uap, rval), 0);
186 /* NOTREACHED */
187 return (0);
188 }
189
190 /*
191 * Exit: deallocate address space and other resources, change proc state
192 * to zombie, and unlink proc from allproc and parent's lists. Save exit
193 * status and rusage for wait(). Check for child processes and orphan them.
194 *
195 * Must be called with p->p_lock held. Does not return.
196 */
197 void
198 exit1(struct lwp *l, int exitcode, int signo)
199 {
200 struct proc *p, *child, *next_child, *old_parent, *new_parent;
201 struct pgrp *pgrp;
202 ksiginfo_t ksi;
203 ksiginfoq_t kq;
204 int wakeinit;
205 struct lwp *l2 __diagused;
206
207 p = l->l_proc;
208
209 /* XXX Temporary. */
210 kernel_lock_plug_leak();
211
212 /* Verify that we hold no locks other than p->p_lock. */
213 LOCKDEBUG_BARRIER(p->p_lock, 0);
214 KASSERTMSG(curcpu()->ci_biglock_count == 0, "kernel_lock leaked");
215 KASSERT(mutex_owned(p->p_lock));
216 KASSERT(p->p_vmspace != NULL);
217
218 if (__predict_false(p == initproc)) {
219 panic("init died (signal %d, exit %d)", signo, exitcode);
220 }
221
222 p->p_sflag |= PS_WEXIT;
223
224 /*
225 * Force all other LWPs to exit before we do. Only then can we
226 * begin to tear down the rest of the process state.
227 */
228 if (p->p_nlwps > 1) {
229 exit_lwps(l);
230 }
231
232 ksiginfo_queue_init(&kq);
233
234 /*
235 * If we have been asked to stop on exit, do so now.
236 */
237 if (__predict_false(p->p_sflag & PS_STOPEXIT)) {
238 KERNEL_UNLOCK_ALL(l, &l->l_biglocks);
239 sigclearall(p, &contsigmask, &kq);
240
241 if (!mutex_tryenter(proc_lock)) {
242 mutex_exit(p->p_lock);
243 mutex_enter(proc_lock);
244 mutex_enter(p->p_lock);
245 }
246 p->p_waited = 0;
247 p->p_pptr->p_nstopchild++;
248 p->p_stat = SSTOP;
249 mutex_exit(proc_lock);
250 lwp_lock(l);
251 p->p_nrlwps--;
252 l->l_stat = LSSTOP;
253 lwp_unlock(l);
254 mutex_exit(p->p_lock);
255 lwp_lock(l);
256 spc_lock(l->l_cpu);
257 mi_switch(l);
258 mutex_enter(p->p_lock);
259 }
260
261 /*
262 * Bin any remaining signals and mark the process as dying so it will
263 * not be found for, e.g. signals.
264 */
265 sigfillset(&p->p_sigctx.ps_sigignore);
266 sigclearall(p, NULL, &kq);
267 p->p_stat = SDYING;
268 mutex_exit(p->p_lock);
269 ksiginfo_queue_drain(&kq);
270
271 /* Destroy any lwpctl info. */
272 if (p->p_lwpctl != NULL)
273 lwp_ctl_exit();
274
275 /*
276 * Drain all remaining references that procfs, ptrace and others may
277 * have on the process.
278 */
279 rw_enter(&p->p_reflock, RW_WRITER);
280
281 DPRINTF(("%s: %d.%d exiting.\n", __func__, p->p_pid, l->l_lid));
282
283 timers_free(p, TIMERS_ALL);
284 #if defined(__HAVE_RAS)
285 ras_purgeall();
286 #endif
287
288 /*
289 * Close open files, release open-file table and free signal
290 * actions. This may block!
291 */
292 fd_free();
293 cwdfree(p->p_cwdi);
294 p->p_cwdi = NULL;
295 doexithooks(p);
296 sigactsfree(p->p_sigacts);
297
298 /*
299 * Write out accounting data.
300 */
301 (void)acct_process(l);
302
303 #ifdef KTRACE
304 /*
305 * Release trace file.
306 */
307 if (p->p_tracep != NULL) {
308 mutex_enter(&ktrace_lock);
309 ktrderef(p);
310 mutex_exit(&ktrace_lock);
311 }
312 #endif
313
314 p->p_xexit = exitcode;
315 p->p_xsig = signo;
316
317 /*
318 * If emulation has process exit hook, call it now.
319 * Set the exit status now so that the exit hook has
320 * an opportunity to tweak it (COMPAT_LINUX requires
321 * this for thread group emulation)
322 */
323 if (p->p_emul->e_proc_exit)
324 (*p->p_emul->e_proc_exit)(p);
325
326 /*
327 * Free the VM resources we're still holding on to.
328 * We must do this from a valid thread because doing
329 * so may block. This frees vmspace, which we don't
330 * need anymore. The only remaining lwp is the one
331 * we run at this moment, nothing runs in userland
332 * anymore.
333 */
334 ruspace(p); /* Update our vm resource use */
335 uvm_proc_exit(p);
336
337 /*
338 * Stop profiling.
339 */
340 if (__predict_false((p->p_stflag & PST_PROFIL) != 0)) {
341 mutex_spin_enter(&p->p_stmutex);
342 stopprofclock(p);
343 mutex_spin_exit(&p->p_stmutex);
344 }
345
346 /*
347 * If parent is waiting for us to exit or exec, PL_PPWAIT is set; we
348 * wake up the parent early to avoid deadlock. We can do this once
349 * the VM resources are released.
350 */
351 mutex_enter(proc_lock);
352 if (p->p_lflag & PL_PPWAIT) {
353 lwp_t *lp;
354
355 l->l_lwpctl = NULL; /* was on loan from blocked parent */
356 p->p_lflag &= ~PL_PPWAIT;
357
358 lp = p->p_vforklwp;
359 p->p_vforklwp = NULL;
360 lp->l_vforkwaiting = false;
361 cv_broadcast(&lp->l_waitcv);
362 }
363
364 if (SESS_LEADER(p)) {
365 struct vnode *vprele = NULL, *vprevoke = NULL;
366 struct session *sp = p->p_session;
367 struct tty *tp;
368
369 if (sp->s_ttyvp) {
370 /*
371 * Controlling process.
372 * Signal foreground pgrp,
373 * drain controlling terminal
374 * and revoke access to controlling terminal.
375 */
376 tp = sp->s_ttyp;
377 mutex_spin_enter(&tty_lock);
378 if (tp->t_session == sp) {
379 /* we can't guarantee the revoke will do this */
380 pgrp = tp->t_pgrp;
381 tp->t_pgrp = NULL;
382 tp->t_session = NULL;
383 mutex_spin_exit(&tty_lock);
384 if (pgrp != NULL) {
385 pgsignal(pgrp, SIGHUP, 1);
386 }
387 mutex_exit(proc_lock);
388 (void) ttywait(tp);
389 mutex_enter(proc_lock);
390
391 /* The tty could have been revoked. */
392 vprevoke = sp->s_ttyvp;
393 } else
394 mutex_spin_exit(&tty_lock);
395 vprele = sp->s_ttyvp;
396 sp->s_ttyvp = NULL;
397 /*
398 * s_ttyp is not zero'd; we use this to indicate
399 * that the session once had a controlling terminal.
400 * (for logging and informational purposes)
401 */
402 }
403 sp->s_leader = NULL;
404
405 if (vprevoke != NULL || vprele != NULL) {
406 if (vprevoke != NULL) {
407 /* Releases proc_lock. */
408 proc_sessrele(sp);
409 VOP_REVOKE(vprevoke, REVOKEALL);
410 } else
411 mutex_exit(proc_lock);
412 if (vprele != NULL)
413 vrele(vprele);
414 mutex_enter(proc_lock);
415 }
416 }
417 fixjobc(p, p->p_pgrp, 0);
418
419 /* Release fstrans private data. */
420 fstrans_lwp_dtor(l);
421
422 /*
423 * Finalize the last LWP's specificdata, as well as the
424 * specificdata for the proc itself.
425 */
426 lwp_finispecific(l);
427 proc_finispecific(p);
428
429 /*
430 * Notify interested parties of our demise.
431 */
432 KNOTE(&p->p_klist, NOTE_EXIT);
433
434 SDT_PROBE(proc, kernel, , exit,
435 ((p->p_sflag & PS_COREDUMP) ? CLD_DUMPED :
436 (p->p_xsig ? CLD_KILLED : CLD_EXITED)),
437 0,0,0,0);
438
439 /*
440 * Reset p_opptr pointer of all former children which got
441 * traced by another process and were reparented. We reset
442 * it to NULL here; the trace detach code then reparents
443 * the child to initproc. We only check allproc list, since
444 * eventual former children on zombproc list won't reference
445 * p_opptr anymore.
446 */
447 if (__predict_false(p->p_slflag & PSL_CHTRACED)) {
448 struct proc *q;
449 PROCLIST_FOREACH(q, &allproc) {
450 if (q->p_opptr == p)
451 q->p_opptr = NULL;
452 }
453 PROCLIST_FOREACH(q, &zombproc) {
454 if (q->p_opptr == p)
455 q->p_opptr = NULL;
456 }
457 }
458
459 /*
460 * Give orphaned children to init(8).
461 */
462 child = LIST_FIRST(&p->p_children);
463 wakeinit = (child != NULL);
464 for (; child != NULL; child = next_child) {
465 next_child = LIST_NEXT(child, p_sibling);
466
467 /*
468 * Traced processes are killed since their existence
469 * means someone is screwing up. Since we reset the
470 * trace flags, the logic in sys_wait4() would not be
471 * triggered to reparent the process to its
472 * original parent, so we must do this here.
473 */
474 if (__predict_false(child->p_slflag & PSL_TRACED)) {
475 mutex_enter(p->p_lock);
476 child->p_slflag &=
477 ~(PSL_TRACED|PSL_SYSCALL);
478 mutex_exit(p->p_lock);
479 if (child->p_opptr != child->p_pptr) {
480 struct proc *t = child->p_opptr;
481 proc_reparent(child, t ? t : initproc);
482 child->p_opptr = NULL;
483 } else
484 proc_reparent(child, initproc);
485 killproc(child, "orphaned traced process");
486 } else
487 proc_reparent(child, initproc);
488 }
489
490 /*
491 * Move proc from allproc to zombproc, it's now nearly ready to be
492 * collected by parent.
493 */
494 LIST_REMOVE(l, l_list);
495 LIST_REMOVE(p, p_list);
496 LIST_INSERT_HEAD(&zombproc, p, p_list);
497
498 /*
499 * Mark the process as dead. We must do this before we signal
500 * the parent.
501 */
502 p->p_stat = SDEAD;
503
504 /* Put in front of parent's sibling list for parent to collect it */
505 old_parent = p->p_pptr;
506 old_parent->p_nstopchild++;
507 if (LIST_FIRST(&old_parent->p_children) != p) {
508 /* Put child where it can be found quickly */
509 LIST_REMOVE(p, p_sibling);
510 LIST_INSERT_HEAD(&old_parent->p_children, p, p_sibling);
511 }
512
513 /*
514 * Notify parent that we're gone. If parent has the P_NOCLDWAIT
515 * flag set, notify init instead (and hope it will handle
516 * this situation).
517 */
518 if (old_parent->p_flag & (PK_NOCLDWAIT|PK_CLDSIGIGN)) {
519 proc_reparent(p, initproc);
520 wakeinit = 1;
521
522 /*
523 * If this was the last child of our parent, notify
524 * parent, so in case he was wait(2)ing, he will
525 * continue.
526 */
527 if (LIST_FIRST(&old_parent->p_children) == NULL)
528 cv_broadcast(&old_parent->p_waitcv);
529 }
530
531 /* Reload parent pointer, since p may have been reparented above */
532 new_parent = p->p_pptr;
533
534 if (__predict_false(p->p_exitsig != 0)) {
535 exit_psignal(p, new_parent, &ksi);
536 kpsignal(new_parent, &ksi, NULL);
537 }
538
539 /* Calculate the final rusage info. */
540 calcru(p, &p->p_stats->p_ru.ru_utime, &p->p_stats->p_ru.ru_stime,
541 NULL, NULL);
542
543 if (wakeinit)
544 cv_broadcast(&initproc->p_waitcv);
545
546 callout_destroy(&l->l_timeout_ch);
547
548 /*
549 * Release any PCU resources before becoming a zombie.
550 */
551 pcu_discard_all(l);
552
553 mutex_enter(p->p_lock);
554 /* Don't bother with p_treelock as no other LWPs remain. */
555 l2 = radix_tree_remove_node(&p->p_lwptree, (uint64_t)(l->l_lid - 1));
556 KASSERT(l2 == l);
557 KASSERT(radix_tree_empty_tree_p(&p->p_lwptree));
558 radix_tree_fini_tree(&p->p_lwptree);
559 /* Free the linux lwp id */
560 if ((l->l_pflag & LP_PIDLID) != 0 && l->l_lid != p->p_pid)
561 proc_free_pid(l->l_lid);
562 lwp_drainrefs(l);
563 lwp_lock(l);
564 l->l_prflag &= ~LPR_DETACHED;
565 l->l_stat = LSZOMB;
566 lwp_unlock(l);
567 KASSERT(curlwp == l);
568 KASSERT(p->p_nrlwps == 1);
569 KASSERT(p->p_nlwps == 1);
570 p->p_stat = SZOMB;
571 p->p_nrlwps--;
572 p->p_nzlwps++;
573 p->p_ndlwps = 0;
574 mutex_exit(p->p_lock);
575
576 /*
577 * Signal the parent to collect us, and drop the proclist lock.
578 * Drop debugger/procfs lock; no new references can be gained.
579 */
580 cv_broadcast(&p->p_pptr->p_waitcv);
581 rw_exit(&p->p_reflock);
582 mutex_exit(proc_lock);
583
584 /*
585 * NOTE: WE ARE NO LONGER ALLOWED TO SLEEP!
586 */
587
588 /*
589 * Give machine-dependent code a chance to free any MD LWP
590 * resources. This must be done before uvm_lwp_exit(), in
591 * case these resources are in the PCB.
592 */
593 cpu_lwp_free(l, 1);
594
595 /* Switch away into oblivion. */
596 lwp_lock(l);
597 spc_lock(l->l_cpu);
598 mi_switch(l);
599 panic("exit1");
600 }
601
602 void
603 exit_lwps(struct lwp *l)
604 {
605 proc_t *p = l->l_proc;
606 lwp_t *l2;
607
608 retry:
609 KASSERT(mutex_owned(p->p_lock));
610
611 /*
612 * Interrupt LWPs in interruptable sleep, unsuspend suspended
613 * LWPs and then wait for everyone else to finish.
614 */
615 LIST_FOREACH(l2, &p->p_lwps, l_sibling) {
616 if (l2 == l)
617 continue;
618 lwp_lock(l2);
619 l2->l_flag |= LW_WEXIT;
620 if ((l2->l_stat == LSSLEEP && (l2->l_flag & LW_SINTR)) ||
621 l2->l_stat == LSSUSPENDED || l2->l_stat == LSSTOP) {
622 l2->l_flag &= ~LW_DBGSUSPEND;
623 /* setrunnable() will release the lock. */
624 setrunnable(l2);
625 continue;
626 }
627 lwp_need_userret(l2);
628 lwp_unlock(l2);
629 }
630
631 /*
632 * Wait for every LWP to exit. Note: LWPs can get suspended/slept
633 * behind us or there may even be new LWPs created. Therefore, a
634 * full retry is required on error.
635 */
636 while (p->p_nlwps > 1) {
637 if (lwp_wait(l, 0, NULL, true)) {
638 goto retry;
639 }
640 }
641
642 KASSERT(p->p_nlwps == 1);
643 }
644
645 int
646 do_sys_waitid(idtype_t idtype, id_t id, int *pid, int *status, int options,
647 struct wrusage *wru, siginfo_t *si)
648 {
649 proc_t *child;
650 int error;
651
652
653 if (wru != NULL)
654 memset(wru, 0, sizeof(*wru));
655 if (si != NULL)
656 memset(si, 0, sizeof(*si));
657
658 mutex_enter(proc_lock);
659 error = find_stopped_child(curproc, idtype, id, options, &child,
660 wru, si);
661 if (child == NULL) {
662 mutex_exit(proc_lock);
663 *pid = 0;
664 *status = 0;
665 return error;
666 }
667 *pid = child->p_pid;
668
669 if (child->p_stat == SZOMB) {
670 /* Child is exiting */
671 *status = P_WAITSTATUS(child);
672 /* proc_free() will release the proc_lock. */
673 if (options & WNOWAIT) {
674 mutex_exit(proc_lock);
675 } else {
676 proc_free(child, wru);
677 }
678 } else {
679 /* Don't mark SIGCONT if we are being stopped */
680 *status = (child->p_xsig == SIGCONT && child->p_stat != SSTOP) ?
681 W_CONTCODE() : W_STOPCODE(child->p_xsig);
682 mutex_exit(proc_lock);
683 }
684 return 0;
685 }
686
687 int
688 do_sys_wait(int *pid, int *status, int options, struct rusage *ru)
689 {
690 idtype_t idtype;
691 id_t id;
692 int ret;
693 struct wrusage wru;
694
695 /*
696 * Translate the special pid values into the (idtype, pid)
697 * pair for wait6. The WAIT_MYPGRP case is handled by
698 * find_stopped_child() on its own.
699 */
700 if (*pid == WAIT_ANY) {
701 idtype = P_ALL;
702 id = 0;
703 } else if (*pid < 0) {
704 idtype = P_PGID;
705 id = (id_t)-*pid;
706 } else {
707 idtype = P_PID;
708 id = (id_t)*pid;
709 }
710 options |= WEXITED | WTRAPPED;
711 ret = do_sys_waitid(idtype, id, pid, status, options, ru ? &wru : NULL,
712 NULL);
713 if (ru)
714 *ru = wru.wru_self;
715 return ret;
716 }
717
718 int
719 sys___wait450(struct lwp *l, const struct sys___wait450_args *uap,
720 register_t *retval)
721 {
722 /* {
723 syscallarg(int) pid;
724 syscallarg(int *) status;
725 syscallarg(int) options;
726 syscallarg(struct rusage *) rusage;
727 } */
728 int error, status, pid = SCARG(uap, pid);
729 struct rusage ru;
730
731 error = do_sys_wait(&pid, &status, SCARG(uap, options),
732 SCARG(uap, rusage) != NULL ? &ru : NULL);
733
734 retval[0] = pid;
735 if (pid == 0) {
736 return error;
737 }
738 if (SCARG(uap, status)) {
739 error = copyout(&status, SCARG(uap, status), sizeof(status));
740 }
741 if (SCARG(uap, rusage) && error == 0) {
742 error = copyout(&ru, SCARG(uap, rusage), sizeof(ru));
743 }
744 return error;
745 }
746
747 int
748 sys_wait6(struct lwp *l, const struct sys_wait6_args *uap, register_t *retval)
749 {
750 /* {
751 syscallarg(idtype_t) idtype;
752 syscallarg(id_t) id;
753 syscallarg(int *) status;
754 syscallarg(int) options;
755 syscallarg(struct wrusage *) wru;
756 syscallarg(siginfo_t *) si;
757 } */
758 struct wrusage wru, *wrup;
759 siginfo_t si, *sip;
760 idtype_t idtype;
761 int pid;
762 id_t id;
763 int error, status;
764
765 idtype = SCARG(uap, idtype);
766 id = SCARG(uap, id);
767
768 if (SCARG(uap, wru) != NULL)
769 wrup = &wru;
770 else
771 wrup = NULL;
772
773 if (SCARG(uap, info) != NULL)
774 sip = &si;
775 else
776 sip = NULL;
777
778 /*
779 * We expect all callers of wait6() to know about WEXITED and
780 * WTRAPPED.
781 */
782 error = do_sys_waitid(idtype, id, &pid, &status, SCARG(uap, options),
783 wrup, sip);
784
785 retval[0] = pid; /* tell userland who it was */
786
787 #if 0
788 /*
789 * should we copyout if there was no process, hence no useful data?
790 * We don't for an old sytle wait4() (etc) but I believe
791 * FreeBSD does for wait6(), so a tossup... Go with FreeBSD for now.
792 */
793 if (pid == 0)
794 return error;
795 #endif
796
797 if (SCARG(uap, status) != NULL && error == 0)
798 error = copyout(&status, SCARG(uap, status), sizeof(status));
799 if (SCARG(uap, wru) != NULL && error == 0)
800 error = copyout(&wru, SCARG(uap, wru), sizeof(wru));
801 if (SCARG(uap, info) != NULL && error == 0)
802 error = copyout(&si, SCARG(uap, info), sizeof(si));
803 return error;
804 }
805
806
807 /*
808 * Find a process that matches the provided criteria, and fill siginfo
809 * and resources if found.
810 * Returns:
811 * -1: Not found, abort early
812 * 0: Not matched
813 * 1: Matched, there might be more matches
814 * 2: This is the only match
815 */
816 static int
817 match_process(const struct proc *pp, struct proc **q, idtype_t idtype, id_t id,
818 int options, struct wrusage *wrusage, siginfo_t *siginfo)
819 {
820 struct rusage *rup;
821 struct proc *p = *q;
822 int rv = 1;
823
824 mutex_enter(p->p_lock);
825 switch (idtype) {
826 case P_ALL:
827 break;
828 case P_PID:
829 if (p->p_pid != (pid_t)id) {
830 mutex_exit(p->p_lock);
831 p = *q = proc_find_raw((pid_t)id);
832 if (p == NULL || p->p_stat == SIDL || p->p_pptr != pp) {
833 *q = NULL;
834 return -1;
835 }
836 mutex_enter(p->p_lock);
837 }
838 rv++;
839 break;
840 case P_PGID:
841 if (p->p_pgid != (pid_t)id)
842 goto out;
843 break;
844 case P_SID:
845 if (p->p_session->s_sid != (pid_t)id)
846 goto out;
847 break;
848 case P_UID:
849 if (kauth_cred_geteuid(p->p_cred) != (uid_t)id)
850 goto out;
851 break;
852 case P_GID:
853 if (kauth_cred_getegid(p->p_cred) != (gid_t)id)
854 goto out;
855 break;
856 case P_CID:
857 case P_PSETID:
858 case P_CPUID:
859 /* XXX: Implement me */
860 default:
861 out:
862 mutex_exit(p->p_lock);
863 return 0;
864 }
865
866 if ((options & WEXITED) == 0 && p->p_stat == SZOMB)
867 goto out;
868
869 if (siginfo != NULL) {
870 siginfo->si_errno = 0;
871
872 /*
873 * SUSv4 requires that the si_signo value is always
874 * SIGCHLD. Obey it despite the rfork(2) interface
875 * allows to request other signal for child exit
876 * notification.
877 */
878 siginfo->si_signo = SIGCHLD;
879
880 /*
881 * This is still a rough estimate. We will fix the
882 * cases TRAPPED, STOPPED, and CONTINUED later.
883 */
884 if (p->p_sflag & PS_COREDUMP) {
885 siginfo->si_code = CLD_DUMPED;
886 siginfo->si_status = p->p_xsig;
887 } else if (p->p_xsig) {
888 siginfo->si_code = CLD_KILLED;
889 siginfo->si_status = p->p_xsig;
890 } else {
891 siginfo->si_code = CLD_EXITED;
892 siginfo->si_status = p->p_xexit;
893 }
894
895 siginfo->si_pid = p->p_pid;
896 siginfo->si_uid = kauth_cred_geteuid(p->p_cred);
897 siginfo->si_utime = p->p_stats->p_ru.ru_utime.tv_sec;
898 siginfo->si_stime = p->p_stats->p_ru.ru_stime.tv_sec;
899 }
900
901 /*
902 * There should be no reason to limit resources usage info to
903 * exited processes only. A snapshot about any resources used
904 * by a stopped process may be exactly what is needed.
905 */
906 if (wrusage != NULL) {
907 rup = &wrusage->wru_self;
908 *rup = p->p_stats->p_ru;
909 calcru(p, &rup->ru_utime, &rup->ru_stime, NULL, NULL);
910
911 rup = &wrusage->wru_children;
912 *rup = p->p_stats->p_cru;
913 calcru(p, &rup->ru_utime, &rup->ru_stime, NULL, NULL);
914 }
915
916 mutex_exit(p->p_lock);
917 return rv;
918 }
919
920 /*
921 * Determine if there are existing processes being debugged
922 * that used to be (and sometime later will be again) children
923 * of a specific parent (while matching wait criteria)
924 */
925 static bool
926 debugged_child_exists(idtype_t idtype, id_t id, int options, siginfo_t *si,
927 const struct proc *parent)
928 {
929 struct proc *pp;
930
931 /*
932 * If we are searching for a specific pid, we can optimise a little
933 */
934 if (idtype == P_PID) {
935 /*
936 * Check the specific process to see if its real parent is us
937 */
938 pp = proc_find_raw((pid_t)id);
939 if (pp != NULL && pp->p_stat != SIDL && pp->p_opptr == parent) {
940 /*
941 * using P_ALL here avoids match_process() doing the
942 * same work that we just did, but incorrectly for
943 * this scenario.
944 */
945 if (match_process(parent, &pp, P_ALL, id, options,
946 NULL, si))
947 return true;
948 }
949 return false;
950 }
951
952 /*
953 * For the hard cases, just look everywhere to see if some
954 * stolen (reparented) process is really our lost child.
955 * Then check if that process could satisfy the wait conditions.
956 */
957
958 /*
959 * XXX inefficient, but hopefully fairly rare.
960 * XXX should really use a list of reparented processes.
961 */
962 PROCLIST_FOREACH(pp, &allproc) {
963 if (pp->p_stat == SIDL) /* XXX impossible ?? */
964 continue;
965 if (pp->p_opptr == parent &&
966 match_process(parent, &pp, idtype, id, options, NULL, si))
967 return true;
968 }
969 PROCLIST_FOREACH(pp, &zombproc) {
970 if (pp->p_stat == SIDL) /* XXX impossible ?? */
971 continue;
972 if (pp->p_opptr == parent &&
973 match_process(parent, &pp, idtype, id, options, NULL, si))
974 return true;
975 }
976
977 return false;
978 }
979
980 /*
981 * Scan list of child processes for a child process that has stopped or
982 * exited. Used by sys_wait4 and 'compat' equivalents.
983 *
984 * Must be called with the proc_lock held, and may release while waiting.
985 */
986 static int
987 find_stopped_child(struct proc *parent, idtype_t idtype, id_t id, int options,
988 struct proc **child_p, struct wrusage *wru, siginfo_t *si)
989 {
990 struct proc *child, *dead;
991 int error;
992
993 KASSERT(mutex_owned(proc_lock));
994
995 if (options & ~WALLOPTS) {
996 *child_p = NULL;
997 return EINVAL;
998 }
999
1000 if ((options & WSELECTOPTS) == 0) {
1001 /*
1002 * We will be unable to find any matching processes,
1003 * because there are no known events to look for.
1004 * Prefer to return error instead of blocking
1005 * indefinitely.
1006 */
1007 *child_p = NULL;
1008 return EINVAL;
1009 }
1010
1011 if ((pid_t)id == WAIT_MYPGRP && (idtype == P_PID || idtype == P_PGID)) {
1012 mutex_enter(parent->p_lock);
1013 id = (id_t)parent->p_pgid;
1014 mutex_exit(parent->p_lock);
1015 idtype = P_PGID;
1016 }
1017
1018 for (;;) {
1019 error = ECHILD;
1020 dead = NULL;
1021
1022 LIST_FOREACH(child, &parent->p_children, p_sibling) {
1023 int rv = match_process(parent, &child, idtype, id,
1024 options, wru, si);
1025 if (rv == -1)
1026 break;
1027 if (rv == 0)
1028 continue;
1029
1030 /*
1031 * Wait for processes with p_exitsig != SIGCHLD
1032 * processes only if WALTSIG is set; wait for
1033 * processes with p_exitsig == SIGCHLD only
1034 * if WALTSIG is clear.
1035 */
1036 if (((options & WALLSIG) == 0) &&
1037 (options & WALTSIG ? child->p_exitsig == SIGCHLD
1038 : P_EXITSIG(child) != SIGCHLD)){
1039 if (rv == 2) {
1040 child = NULL;
1041 break;
1042 }
1043 continue;
1044 }
1045
1046 error = 0;
1047 if ((options & WNOZOMBIE) == 0) {
1048 if (child->p_stat == SZOMB)
1049 break;
1050 if (child->p_stat == SDEAD) {
1051 /*
1052 * We may occasionally arrive here
1053 * after receiving a signal, but
1054 * immediately before the child
1055 * process is zombified. The wait
1056 * will be short, so avoid returning
1057 * to userspace.
1058 */
1059 dead = child;
1060 }
1061 }
1062
1063 if ((options & WCONTINUED) != 0 &&
1064 child->p_xsig == SIGCONT &&
1065 (child->p_sflag & PS_CONTINUED)) {
1066 if ((options & WNOWAIT) == 0) {
1067 child->p_sflag &= ~PS_CONTINUED;
1068 child->p_waited = 1;
1069 parent->p_nstopchild--;
1070 }
1071 if (si) {
1072 si->si_status = child->p_xsig;
1073 si->si_code = CLD_CONTINUED;
1074 }
1075 break;
1076 }
1077
1078 if ((options & (WTRAPPED|WSTOPPED)) != 0 &&
1079 child->p_stat == SSTOP &&
1080 child->p_waited == 0 &&
1081 ((child->p_slflag & PSL_TRACED) ||
1082 options & (WUNTRACED|WSTOPPED))) {
1083 if ((options & WNOWAIT) == 0) {
1084 child->p_waited = 1;
1085 parent->p_nstopchild--;
1086 }
1087 if (si) {
1088 si->si_status = child->p_xsig;
1089 si->si_code =
1090 (child->p_slflag & PSL_TRACED) ?
1091 CLD_TRAPPED : CLD_STOPPED;
1092 }
1093 break;
1094 }
1095 if (parent->p_nstopchild == 0 || rv == 2) {
1096 child = NULL;
1097 break;
1098 }
1099 }
1100
1101 /*
1102 * If we found nothing, but we are the bereaved parent
1103 * of a stolen child, look and see if that child (or
1104 * one of them) meets our search criteria. If so, then
1105 * we cannot succeed, but we can hang (wait...),
1106 * or if WNOHANG, return 0 instead of ECHILD
1107 */
1108 if (child == NULL && error == ECHILD &&
1109 (parent->p_slflag & PSL_CHTRACED) &&
1110 debugged_child_exists(idtype, id, options, si, parent))
1111 error = 0;
1112
1113 if (child != NULL || error != 0 ||
1114 ((options & WNOHANG) != 0 && dead == NULL)) {
1115 *child_p = child;
1116 return error;
1117 }
1118
1119 /*
1120 * Wait for another child process to stop.
1121 */
1122 error = cv_wait_sig(&parent->p_waitcv, proc_lock);
1123
1124 if (error != 0) {
1125 *child_p = NULL;
1126 return error;
1127 }
1128 }
1129 }
1130
1131 /*
1132 * Free a process after parent has taken all the state info. Must be called
1133 * with the proclist lock held, and will release before returning.
1134 *
1135 * *ru is returned to the caller, and must be freed by the caller.
1136 */
1137 static void
1138 proc_free(struct proc *p, struct wrusage *wru)
1139 {
1140 struct proc *parent = p->p_pptr;
1141 struct lwp *l;
1142 ksiginfo_t ksi;
1143 kauth_cred_t cred1, cred2;
1144 uid_t uid;
1145
1146 KASSERT(mutex_owned(proc_lock));
1147 KASSERT(p->p_nlwps == 1);
1148 KASSERT(p->p_nzlwps == 1);
1149 KASSERT(p->p_nrlwps == 0);
1150 KASSERT(p->p_stat == SZOMB);
1151
1152 /*
1153 * If we got the child via ptrace(2) or procfs, and
1154 * the parent is different (meaning the process was
1155 * attached, rather than run as a child), then we need
1156 * to give it back to the old parent, and send the
1157 * parent the exit signal. The rest of the cleanup
1158 * will be done when the old parent waits on the child.
1159 */
1160 if ((p->p_slflag & PSL_TRACED) != 0 && p->p_opptr != parent) {
1161 mutex_enter(p->p_lock);
1162 p->p_slflag &= ~(PSL_TRACED|PSL_SYSCALL);
1163 mutex_exit(p->p_lock);
1164 parent = (p->p_opptr == NULL) ? initproc : p->p_opptr;
1165 proc_reparent(p, parent);
1166 p->p_opptr = NULL;
1167 if (p->p_exitsig != 0) {
1168 exit_psignal(p, parent, &ksi);
1169 kpsignal(parent, &ksi, NULL);
1170 }
1171 cv_broadcast(&parent->p_waitcv);
1172 mutex_exit(proc_lock);
1173 return;
1174 }
1175
1176 sched_proc_exit(parent, p);
1177
1178 /*
1179 * Add child times of exiting process onto its own times.
1180 * This cannot be done any earlier else it might get done twice.
1181 */
1182 l = LIST_FIRST(&p->p_lwps);
1183 p->p_stats->p_ru.ru_nvcsw += (l->l_ncsw - l->l_nivcsw);
1184 p->p_stats->p_ru.ru_nivcsw += l->l_nivcsw;
1185 ruadd(&p->p_stats->p_ru, &l->l_ru);
1186 ruadd(&p->p_stats->p_ru, &p->p_stats->p_cru);
1187 ruadd(&parent->p_stats->p_cru, &p->p_stats->p_ru);
1188 if (wru != NULL) {
1189 wru->wru_self = p->p_stats->p_ru;
1190 wru->wru_children = p->p_stats->p_cru;
1191 }
1192 p->p_xsig = 0;
1193 p->p_xexit = 0;
1194
1195 /*
1196 * At this point we are going to start freeing the final resources.
1197 * If anyone tries to access the proc structure after here they will
1198 * get a shock - bits are missing. Attempt to make it hard! We
1199 * don't bother with any further locking past this point.
1200 */
1201 p->p_stat = SIDL; /* not even a zombie any more */
1202 LIST_REMOVE(p, p_list); /* off zombproc */
1203 parent->p_nstopchild--;
1204 LIST_REMOVE(p, p_sibling);
1205
1206 /*
1207 * Let pid be reallocated.
1208 */
1209 proc_free_pid(p->p_pid);
1210
1211 /*
1212 * Unlink process from its process group.
1213 * Releases the proc_lock.
1214 */
1215 proc_leavepgrp(p);
1216
1217 /*
1218 * Delay release until after lwp_free.
1219 */
1220 cred2 = l->l_cred;
1221
1222 /*
1223 * Free the last LWP's resources.
1224 *
1225 * lwp_free ensures the LWP is no longer running on another CPU.
1226 */
1227 lwp_free(l, false, true);
1228
1229 /*
1230 * Now no one except us can reach the process p.
1231 */
1232
1233 /*
1234 * Decrement the count of procs running with this uid.
1235 */
1236 cred1 = p->p_cred;
1237 uid = kauth_cred_getuid(cred1);
1238 (void)chgproccnt(uid, -1);
1239
1240 /*
1241 * Release substructures.
1242 */
1243
1244 lim_free(p->p_limit);
1245 pstatsfree(p->p_stats);
1246 kauth_cred_free(cred1);
1247 kauth_cred_free(cred2);
1248
1249 /*
1250 * Release reference to text vnode
1251 */
1252 if (p->p_textvp)
1253 vrele(p->p_textvp);
1254 kmem_strfree(p->p_path);
1255
1256 mutex_destroy(&p->p_auxlock);
1257 mutex_obj_free(p->p_lock);
1258 mutex_destroy(&p->p_stmutex);
1259 cv_destroy(&p->p_waitcv);
1260 cv_destroy(&p->p_lwpcv);
1261 rw_destroy(&p->p_reflock);
1262 rw_destroy(&p->p_treelock);
1263
1264 proc_free_mem(p);
1265 }
1266
1267 /*
1268 * Change the parent of a process for tracing purposes.
1269 */
1270 void
1271 proc_changeparent(struct proc *t, struct proc *p)
1272 {
1273 SET(t->p_slflag, PSL_TRACED);
1274 t->p_opptr = t->p_pptr;
1275 if (t->p_pptr == p)
1276 return;
1277 struct proc *parent = t->p_pptr;
1278
1279 if (parent->p_lock < t->p_lock) {
1280 if (!mutex_tryenter(parent->p_lock)) {
1281 mutex_exit(t->p_lock);
1282 mutex_enter(parent->p_lock);
1283 mutex_enter(t->p_lock);
1284 }
1285 } else if (parent->p_lock > t->p_lock) {
1286 mutex_enter(parent->p_lock);
1287 }
1288 parent->p_slflag |= PSL_CHTRACED;
1289 proc_reparent(t, p);
1290 if (parent->p_lock != t->p_lock)
1291 mutex_exit(parent->p_lock);
1292 }
1293
1294 /*
1295 * make process 'parent' the new parent of process 'child'.
1296 *
1297 * Must be called with proc_lock held.
1298 */
1299 void
1300 proc_reparent(struct proc *child, struct proc *parent)
1301 {
1302
1303 KASSERT(mutex_owned(proc_lock));
1304
1305 if (child->p_pptr == parent)
1306 return;
1307
1308 if (child->p_stat == SZOMB || child->p_stat == SDEAD ||
1309 (child->p_stat == SSTOP && !child->p_waited)) {
1310 child->p_pptr->p_nstopchild--;
1311 parent->p_nstopchild++;
1312 }
1313 if (parent == initproc) {
1314 child->p_exitsig = SIGCHLD;
1315 child->p_ppid = parent->p_pid;
1316 }
1317
1318 LIST_REMOVE(child, p_sibling);
1319 LIST_INSERT_HEAD(&parent->p_children, child, p_sibling);
1320 child->p_pptr = parent;
1321 }
1322