xref: /freebsd/sys/kern/kern_exit.c (revision d0675399d09f02d347912e23d004329710338450)
1 /*-
2  * SPDX-License-Identifier: BSD-3-Clause
3  *
4  * Copyright (c) 1982, 1986, 1989, 1991, 1993
5  *	The Regents of the University of California.  All rights reserved.
6  * (c) UNIX System Laboratories, Inc.
7  * All or some portions of this file are derived from material licensed
8  * to the University of California by American Telephone and Telegraph
9  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
10  * the permission of UNIX System Laboratories, Inc.
11  *
12  * Redistribution and use in source and binary forms, with or without
13  * modification, are permitted provided that the following conditions
14  * are met:
15  * 1. Redistributions of source code must retain the above copyright
16  *    notice, this list of conditions and the following disclaimer.
17  * 2. Redistributions in binary form must reproduce the above copyright
18  *    notice, this list of conditions and the following disclaimer in the
19  *    documentation and/or other materials provided with the distribution.
20  * 3. Neither the name of the University nor the names of its contributors
21  *    may be used to endorse or promote products derived from this software
22  *    without specific prior written permission.
23  *
24  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
25  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
26  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
27  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
28  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
29  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
30  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
31  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
32  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
33  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34  * SUCH DAMAGE.
35  */
36 
37 #include <sys/cdefs.h>
38 #include "opt_ddb.h"
39 #include "opt_ktrace.h"
40 
41 #include <sys/param.h>
42 #include <sys/systm.h>
43 #include <sys/sysproto.h>
44 #include <sys/capsicum.h>
45 #include <sys/eventhandler.h>
46 #include <sys/kernel.h>
47 #include <sys/ktr.h>
48 #include <sys/malloc.h>
49 #include <sys/lock.h>
50 #include <sys/mutex.h>
51 #include <sys/proc.h>
52 #include <sys/procdesc.h>
53 #include <sys/jail.h>
54 #include <sys/tty.h>
55 #include <sys/wait.h>
56 #include <sys/vmmeter.h>
57 #include <sys/vnode.h>
58 #include <sys/racct.h>
59 #include <sys/resourcevar.h>
60 #include <sys/sbuf.h>
61 #include <sys/signalvar.h>
62 #include <sys/sched.h>
63 #include <sys/sx.h>
64 #include <sys/syscallsubr.h>
65 #include <sys/sysctl.h>
66 #include <sys/syslog.h>
67 #include <sys/ptrace.h>
68 #include <sys/acct.h>		/* for acct_process() function prototype */
69 #include <sys/filedesc.h>
70 #include <sys/sdt.h>
71 #include <sys/shm.h>
72 #include <sys/sem.h>
73 #include <sys/sysent.h>
74 #include <sys/timers.h>
75 #include <sys/umtxvar.h>
76 #ifdef KTRACE
77 #include <sys/ktrace.h>
78 #endif
79 
80 #include <security/audit/audit.h>
81 #include <security/mac/mac_framework.h>
82 
83 #include <vm/vm.h>
84 #include <vm/vm_extern.h>
85 #include <vm/vm_param.h>
86 #include <vm/pmap.h>
87 #include <vm/vm_map.h>
88 #include <vm/vm_page.h>
89 #include <vm/uma.h>
90 
91 #ifdef KDTRACE_HOOKS
92 #include <sys/dtrace_bsd.h>
93 dtrace_execexit_func_t	dtrace_fasttrap_exit;
94 #endif
95 
96 SDT_PROVIDER_DECLARE(proc);
97 SDT_PROBE_DEFINE1(proc, , , exit, "int");
98 
99 static int kern_kill_on_dbg_exit = 1;
100 SYSCTL_INT(_kern, OID_AUTO, kill_on_debugger_exit, CTLFLAG_RWTUN,
101     &kern_kill_on_dbg_exit, 0,
102     "Kill ptraced processes when debugger exits");
103 
104 static bool kern_wait_dequeue_sigchld = 1;
105 SYSCTL_BOOL(_kern, OID_AUTO, wait_dequeue_sigchld, CTLFLAG_RWTUN,
106     &kern_wait_dequeue_sigchld, 0,
107     "Dequeue SIGCHLD on wait(2) for live process");
108 
109 struct proc *
proc_realparent(struct proc * child)110 proc_realparent(struct proc *child)
111 {
112 	struct proc *p, *parent;
113 
114 	sx_assert(&proctree_lock, SX_LOCKED);
115 	if ((child->p_treeflag & P_TREE_ORPHANED) == 0)
116 		return (child->p_pptr->p_pid == child->p_oppid ?
117 		    child->p_pptr : child->p_reaper);
118 	for (p = child; (p->p_treeflag & P_TREE_FIRST_ORPHAN) == 0;) {
119 		/* Cannot use LIST_PREV(), since the list head is not known. */
120 		p = __containerof(p->p_orphan.le_prev, struct proc,
121 		    p_orphan.le_next);
122 		KASSERT((p->p_treeflag & P_TREE_ORPHANED) != 0,
123 		    ("missing P_ORPHAN %p", p));
124 	}
125 	parent = __containerof(p->p_orphan.le_prev, struct proc,
126 	    p_orphans.lh_first);
127 	return (parent);
128 }
129 
130 void
reaper_abandon_children(struct proc * p,bool exiting)131 reaper_abandon_children(struct proc *p, bool exiting)
132 {
133 	struct proc *p1, *p2, *ptmp;
134 
135 	sx_assert(&proctree_lock, SX_XLOCKED);
136 	KASSERT(p != initproc, ("reaper_abandon_children for initproc"));
137 	if ((p->p_treeflag & P_TREE_REAPER) == 0)
138 		return;
139 	p1 = p->p_reaper;
140 	LIST_FOREACH_SAFE(p2, &p->p_reaplist, p_reapsibling, ptmp) {
141 		LIST_REMOVE(p2, p_reapsibling);
142 		p2->p_reaper = p1;
143 		p2->p_reapsubtree = p->p_reapsubtree;
144 		LIST_INSERT_HEAD(&p1->p_reaplist, p2, p_reapsibling);
145 		if (exiting && p2->p_pptr == p) {
146 			PROC_LOCK(p2);
147 			proc_reparent(p2, p1, true);
148 			PROC_UNLOCK(p2);
149 		}
150 	}
151 	KASSERT(LIST_EMPTY(&p->p_reaplist), ("p_reaplist not empty"));
152 	p->p_treeflag &= ~P_TREE_REAPER;
153 }
154 
155 static void
reaper_clear(struct proc * p)156 reaper_clear(struct proc *p)
157 {
158 	struct proc *p1;
159 	bool clear;
160 
161 	sx_assert(&proctree_lock, SX_LOCKED);
162 	LIST_REMOVE(p, p_reapsibling);
163 	if (p->p_reapsubtree == 1)
164 		return;
165 	clear = true;
166 	LIST_FOREACH(p1, &p->p_reaper->p_reaplist, p_reapsibling) {
167 		if (p1->p_reapsubtree == p->p_reapsubtree) {
168 			clear = false;
169 			break;
170 		}
171 	}
172 	if (clear)
173 		proc_id_clear(PROC_ID_REAP, p->p_reapsubtree);
174 }
175 
176 void
proc_clear_orphan(struct proc * p)177 proc_clear_orphan(struct proc *p)
178 {
179 	struct proc *p1;
180 
181 	sx_assert(&proctree_lock, SA_XLOCKED);
182 	if ((p->p_treeflag & P_TREE_ORPHANED) == 0)
183 		return;
184 	if ((p->p_treeflag & P_TREE_FIRST_ORPHAN) != 0) {
185 		p1 = LIST_NEXT(p, p_orphan);
186 		if (p1 != NULL)
187 			p1->p_treeflag |= P_TREE_FIRST_ORPHAN;
188 		p->p_treeflag &= ~P_TREE_FIRST_ORPHAN;
189 	}
190 	LIST_REMOVE(p, p_orphan);
191 	p->p_treeflag &= ~P_TREE_ORPHANED;
192 }
193 
194 void
exit_onexit(struct proc * p)195 exit_onexit(struct proc *p)
196 {
197 	MPASS(p->p_numthreads == 1);
198 	umtx_thread_exit(FIRST_THREAD_IN_PROC(p));
199 }
200 
201 /*
202  * exit -- death of process.
203  */
204 int
sys_exit(struct thread * td,struct exit_args * uap)205 sys_exit(struct thread *td, struct exit_args *uap)
206 {
207 
208 	exit1(td, uap->rval, 0);
209 	__unreachable();
210 }
211 
212 void
proc_set_p2_wexit(struct proc * p)213 proc_set_p2_wexit(struct proc *p)
214 {
215 	PROC_LOCK_ASSERT(p, MA_OWNED);
216 	p->p_flag2 |= P2_WEXIT;
217 }
218 
219 /*
220  * Exit: deallocate address space and other resources, change proc state to
221  * zombie, and unlink proc from allproc and parent's lists.  Save exit status
222  * and rusage for wait().  Check for child processes and orphan them.
223  */
224 void
exit1(struct thread * td,int rval,int signo)225 exit1(struct thread *td, int rval, int signo)
226 {
227 	struct proc *p, *nq, *q, *t;
228 	struct thread *tdt;
229 	ksiginfo_t *ksi, *ksi1;
230 	int signal_parent;
231 
232 	mtx_assert(&Giant, MA_NOTOWNED);
233 	KASSERT(rval == 0 || signo == 0, ("exit1 rv %d sig %d", rval, signo));
234 	TSPROCEXIT(td->td_proc->p_pid);
235 
236 	p = td->td_proc;
237 	/*
238 	 * In case we're rebooting we just let init die in order to
239 	 * work around an issues where pid 1 might get a fatal signal.
240 	 * For instance, if network interface serving NFS root is
241 	 * going down due to reboot, page-in requests for text are
242 	 * failing.
243 	 */
244 	if (p == initproc && rebooting == 0) {
245 		printf("init died (signal %d, exit %d)\n", signo, rval);
246 		panic("Going nowhere without my init!");
247 	}
248 
249 	/*
250 	 * Process deferred operations, designated with ASTF_KCLEAR.
251 	 * For instance, we need to deref SU mp, since the thread does
252 	 * not return to userspace, and wait for geom to stabilize.
253 	 */
254 	ast_kclear(td);
255 
256 	/*
257 	 * MUST abort all other threads before proceeding past here.
258 	 */
259 	PROC_LOCK(p);
260 	proc_set_p2_wexit(p);
261 
262 	/*
263 	 * First check if some other thread or external request got
264 	 * here before us.  If so, act appropriately: exit or suspend.
265 	 * We must ensure that stop requests are handled before we set
266 	 * P_WEXIT.
267 	 */
268 	thread_suspend_check(0);
269 	while (p->p_flag & P_HADTHREADS) {
270 		/*
271 		 * Kill off the other threads. This requires
272 		 * some co-operation from other parts of the kernel
273 		 * so it may not be instantaneous.  With this state set
274 		 * any thread attempting to interruptibly
275 		 * sleep will return immediately with EINTR or EWOULDBLOCK
276 		 * which will hopefully force them to back out to userland
277 		 * freeing resources as they go.  Any thread attempting
278 		 * to return to userland will thread_exit() from ast().
279 		 * thread_exit() will unsuspend us when the last of the
280 		 * other threads exits.
281 		 * If there is already a thread singler after resumption,
282 		 * calling thread_single() will fail; in that case, we just
283 		 * re-check all suspension request, the thread should
284 		 * either be suspended there or exit.
285 		 */
286 		if (!thread_single(p, SINGLE_EXIT))
287 			/*
288 			 * All other activity in this process is now
289 			 * stopped.  Threading support has been turned
290 			 * off.
291 			 */
292 			break;
293 		/*
294 		 * Recheck for new stop or suspend requests which
295 		 * might appear while process lock was dropped in
296 		 * thread_single().
297 		 */
298 		thread_suspend_check(0);
299 	}
300 	KASSERT(p->p_numthreads == 1,
301 	    ("exit1: proc %p exiting with %d threads", p, p->p_numthreads));
302 	racct_sub(p, RACCT_NTHR, 1);
303 
304 	/* Let event handler change exit status */
305 	p->p_xexit = rval;
306 	p->p_xsig = signo;
307 
308 	/*
309 	 * Ignore any pending request to stop due to a stop signal.
310 	 * Once P_WEXIT is set, future requests will be ignored as
311 	 * well.
312 	 */
313 	p->p_flag &= ~P_STOPPED_SIG;
314 	KASSERT(!P_SHOULDSTOP(p), ("exiting process is stopped"));
315 
316 	/* Note that we are exiting. */
317 	p->p_flag |= P_WEXIT;
318 
319 	/*
320 	 * Wait for any processes that have a hold on our vmspace to
321 	 * release their reference.
322 	 */
323 	while (p->p_lock > 0)
324 		msleep(&p->p_lock, &p->p_mtx, PWAIT, "exithold", 0);
325 
326 	PROC_UNLOCK(p);
327 	/* Drain the limit callout while we don't have the proc locked */
328 	callout_drain(&p->p_limco);
329 
330 #ifdef AUDIT
331 	/*
332 	 * The Sun BSM exit token contains two components: an exit status as
333 	 * passed to exit(), and a return value to indicate what sort of exit
334 	 * it was.  The exit status is WEXITSTATUS(rv), but it's not clear
335 	 * what the return value is.
336 	 */
337 	AUDIT_ARG_EXIT(rval, 0);
338 	AUDIT_SYSCALL_EXIT(0, td);
339 #endif
340 
341 	/* Are we a task leader with peers? */
342 	if (p->p_peers != NULL && p == p->p_leader) {
343 		mtx_lock(&ppeers_lock);
344 		q = p->p_peers;
345 		while (q != NULL) {
346 			PROC_LOCK(q);
347 			kern_psignal(q, SIGKILL);
348 			PROC_UNLOCK(q);
349 			q = q->p_peers;
350 		}
351 		while (p->p_peers != NULL)
352 			msleep(p, &ppeers_lock, PWAIT, "exit1", 0);
353 		mtx_unlock(&ppeers_lock);
354 	}
355 
356 	itimers_exit(p);
357 
358 	/*
359 	 * Check if any loadable modules need anything done at process exit.
360 	 * E.g. SYSV IPC stuff.
361 	 * Event handler could change exit status.
362 	 * XXX what if one of these generates an error?
363 	 */
364 	EVENTHANDLER_DIRECT_INVOKE(process_exit, p);
365 
366 	/*
367 	 * If parent is waiting for us to exit or exec,
368 	 * P_PPWAIT is set; we will wakeup the parent below.
369 	 */
370 	PROC_LOCK(p);
371 	stopprofclock(p);
372 	p->p_ptevents = 0;
373 
374 	/*
375 	 * Stop the real interval timer.  If the handler is currently
376 	 * executing, prevent it from rearming itself and let it finish.
377 	 */
378 	if (timevalisset(&p->p_realtimer.it_value) &&
379 	    callout_stop(&p->p_itcallout) == 0) {
380 		timevalclear(&p->p_realtimer.it_interval);
381 		PROC_UNLOCK(p);
382 		callout_drain(&p->p_itcallout);
383 	} else {
384 		PROC_UNLOCK(p);
385 	}
386 
387 	if (p->p_sysent->sv_onexit != NULL)
388 		p->p_sysent->sv_onexit(p);
389 	seltdfini(td);
390 
391 	/*
392 	 * Reset any sigio structures pointing to us as a result of
393 	 * F_SETOWN with our pid.  The P_WEXIT flag interlocks with fsetown().
394 	 */
395 	funsetownlst(&p->p_sigiolst);
396 
397 	/*
398 	 * Close open files and release open-file table.
399 	 * This may block!
400 	 */
401 	pdescfree(td);
402 	fdescfree(td);
403 
404 	/*
405 	 * Remove ourself from our leader's peer list and wake our leader.
406 	 */
407 	if (p->p_leader->p_peers != NULL) {
408 		mtx_lock(&ppeers_lock);
409 		if (p->p_leader->p_peers != NULL) {
410 			q = p->p_leader;
411 			while (q->p_peers != p)
412 				q = q->p_peers;
413 			q->p_peers = p->p_peers;
414 			wakeup(p->p_leader);
415 		}
416 		mtx_unlock(&ppeers_lock);
417 	}
418 
419 	exec_free_abi_mappings(p);
420 	vmspace_exit(td);
421 	(void)acct_process(td);
422 
423 #ifdef KTRACE
424 	ktrprocexit(td);
425 #endif
426 	/*
427 	 * Release reference to text vnode etc
428 	 */
429 	if (p->p_textvp != NULL) {
430 		vrele(p->p_textvp);
431 		p->p_textvp = NULL;
432 	}
433 	if (p->p_textdvp != NULL) {
434 		vrele(p->p_textdvp);
435 		p->p_textdvp = NULL;
436 	}
437 	if (p->p_binname != NULL) {
438 		free(p->p_binname, M_PARGS);
439 		p->p_binname = NULL;
440 	}
441 
442 	/*
443 	 * Release our limits structure.
444 	 */
445 	lim_free(p->p_limit);
446 	p->p_limit = NULL;
447 
448 	tidhash_remove(td);
449 
450 	/*
451 	 * Call machine-dependent code to release any
452 	 * machine-dependent resources other than the address space.
453 	 * The address space is released by "vmspace_exitfree(p)" in
454 	 * vm_waitproc().
455 	 */
456 	cpu_exit(td);
457 
458 	WITNESS_WARN(WARN_PANIC, NULL, "process (pid %d) exiting", p->p_pid);
459 
460 	/*
461 	 * Remove from allproc. It still sits in the hash.
462 	 */
463 	sx_xlock(&allproc_lock);
464 	LIST_REMOVE(p, p_list);
465 
466 #ifdef DDB
467 	/*
468 	 * Used by ddb's 'ps' command to find this process via the
469 	 * pidhash.
470 	 */
471 	p->p_list.le_prev = NULL;
472 #endif
473 	prison_proc_unlink(p->p_ucred->cr_prison, p);
474 	sx_xunlock(&allproc_lock);
475 
476 	sx_xlock(&proctree_lock);
477 	if ((p->p_flag & (P_TRACED | P_PPWAIT | P_PPTRACE)) != 0) {
478 		PROC_LOCK(p);
479 		p->p_flag &= ~(P_TRACED | P_PPWAIT | P_PPTRACE);
480 		PROC_UNLOCK(p);
481 	}
482 
483 	/*
484 	 * killjobc() might drop and re-acquire proctree_lock to
485 	 * revoke control tty if exiting process was a session leader.
486 	 */
487 	killjobc();
488 
489 	/*
490 	 * Reparent all children processes:
491 	 * - traced ones to the original parent (or init if we are that parent)
492 	 * - the rest to init
493 	 */
494 	q = LIST_FIRST(&p->p_children);
495 	if (q != NULL)		/* only need this if any child is S_ZOMB */
496 		wakeup(q->p_reaper);
497 	for (; q != NULL; q = nq) {
498 		nq = LIST_NEXT(q, p_sibling);
499 		ksi = ksiginfo_alloc(M_WAITOK);
500 		PROC_LOCK(q);
501 		q->p_sigparent = SIGCHLD;
502 
503 		if ((q->p_flag & P_TRACED) == 0) {
504 			proc_reparent(q, q->p_reaper, true);
505 			if (q->p_state == PRS_ZOMBIE) {
506 				/*
507 				 * Inform reaper about the reparented
508 				 * zombie, since wait(2) has something
509 				 * new to report.  Guarantee queueing
510 				 * of the SIGCHLD signal, similar to
511 				 * the _exit() behaviour, by providing
512 				 * our ksiginfo.  Ksi is freed by the
513 				 * signal delivery.
514 				 */
515 				if (q->p_ksi == NULL) {
516 					ksi1 = NULL;
517 				} else {
518 					ksiginfo_copy(q->p_ksi, ksi);
519 					ksi->ksi_flags |= KSI_INS;
520 					ksi1 = ksi;
521 					ksi = NULL;
522 				}
523 				PROC_LOCK(q->p_reaper);
524 				pksignal(q->p_reaper, SIGCHLD, ksi1);
525 				PROC_UNLOCK(q->p_reaper);
526 			} else if (q->p_pdeathsig > 0) {
527 				/*
528 				 * The child asked to received a signal
529 				 * when we exit.
530 				 */
531 				kern_psignal(q, q->p_pdeathsig);
532 			}
533 		} else {
534 			/*
535 			 * Traced processes are killed by default
536 			 * since their existence means someone is
537 			 * screwing up.
538 			 */
539 			t = proc_realparent(q);
540 			if (t == p) {
541 				proc_reparent(q, q->p_reaper, true);
542 			} else {
543 				PROC_LOCK(t);
544 				proc_reparent(q, t, true);
545 				PROC_UNLOCK(t);
546 			}
547 			/*
548 			 * Since q was found on our children list, the
549 			 * proc_reparent() call moved q to the orphan
550 			 * list due to present P_TRACED flag. Clear
551 			 * orphan link for q now while q is locked.
552 			 */
553 			proc_clear_orphan(q);
554 			q->p_flag &= ~P_TRACED;
555 			q->p_flag2 &= ~P2_PTRACE_FSTP;
556 			q->p_ptevents = 0;
557 			p->p_xthread = NULL;
558 			FOREACH_THREAD_IN_PROC(q, tdt) {
559 				tdt->td_dbgflags &= ~(TDB_SUSPEND | TDB_XSIG |
560 				    TDB_FSTP);
561 				tdt->td_xsig = 0;
562 			}
563 			if (kern_kill_on_dbg_exit) {
564 				q->p_flag &= ~P_STOPPED_TRACE;
565 				kern_psignal(q, SIGKILL);
566 			} else if ((q->p_flag & (P_STOPPED_TRACE |
567 			    P_STOPPED_SIG)) != 0) {
568 				sigqueue_delete_proc(q, SIGTRAP);
569 				ptrace_unsuspend(q);
570 			}
571 		}
572 		PROC_UNLOCK(q);
573 		if (ksi != NULL)
574 			ksiginfo_free(ksi);
575 	}
576 
577 	/*
578 	 * Also get rid of our orphans.
579 	 */
580 	while ((q = LIST_FIRST(&p->p_orphans)) != NULL) {
581 		PROC_LOCK(q);
582 		KASSERT(q->p_oppid == p->p_pid,
583 		    ("orphan %p of %p has unexpected oppid %d", q, p,
584 		    q->p_oppid));
585 		q->p_oppid = q->p_reaper->p_pid;
586 
587 		/*
588 		 * If we are the real parent of this process
589 		 * but it has been reparented to a debugger, then
590 		 * check if it asked for a signal when we exit.
591 		 */
592 		if (q->p_pdeathsig > 0)
593 			kern_psignal(q, q->p_pdeathsig);
594 		CTR2(KTR_PTRACE, "exit: pid %d, clearing orphan %d", p->p_pid,
595 		    q->p_pid);
596 		proc_clear_orphan(q);
597 		PROC_UNLOCK(q);
598 	}
599 
600 #ifdef KDTRACE_HOOKS
601 	if (SDT_PROBES_ENABLED()) {
602 		int reason = CLD_EXITED;
603 		if (WCOREDUMP(signo))
604 			reason = CLD_DUMPED;
605 		else if (WIFSIGNALED(signo))
606 			reason = CLD_KILLED;
607 		SDT_PROBE1(proc, , , exit, reason);
608 	}
609 #endif
610 
611 	/* Save exit status. */
612 	PROC_LOCK(p);
613 	p->p_xthread = td;
614 
615 	if (p->p_sysent->sv_ontdexit != NULL)
616 		p->p_sysent->sv_ontdexit(td);
617 
618 #ifdef KDTRACE_HOOKS
619 	/*
620 	 * Tell the DTrace fasttrap provider about the exit if it
621 	 * has declared an interest.
622 	 */
623 	if (dtrace_fasttrap_exit)
624 		dtrace_fasttrap_exit(p);
625 #endif
626 
627 	/*
628 	 * Notify interested parties of our demise.
629 	 */
630 	KNOTE_LOCKED(p->p_klist, NOTE_EXIT);
631 
632 	/*
633 	 * If this is a process with a descriptor, we may not need to deliver
634 	 * a signal to the parent.  proctree_lock is held over
635 	 * procdesc_exit() to serialize concurrent calls to close() and
636 	 * exit().
637 	 */
638 	signal_parent = 0;
639 	if (p->p_procdesc == NULL || procdesc_exit(p)) {
640 		/*
641 		 * Notify parent that we're gone.  If parent has the
642 		 * PS_NOCLDWAIT flag set, or if the handler is set to SIG_IGN,
643 		 * notify process 1 instead (and hope it will handle this
644 		 * situation).
645 		 */
646 		PROC_LOCK(p->p_pptr);
647 		mtx_lock(&p->p_pptr->p_sigacts->ps_mtx);
648 		if (p->p_pptr->p_sigacts->ps_flag &
649 		    (PS_NOCLDWAIT | PS_CLDSIGIGN)) {
650 			struct proc *pp;
651 
652 			mtx_unlock(&p->p_pptr->p_sigacts->ps_mtx);
653 			pp = p->p_pptr;
654 			PROC_UNLOCK(pp);
655 			proc_reparent(p, p->p_reaper, true);
656 			p->p_sigparent = SIGCHLD;
657 			PROC_LOCK(p->p_pptr);
658 
659 			/*
660 			 * Notify parent, so in case he was wait(2)ing or
661 			 * executing waitpid(2) with our pid, he will
662 			 * continue.
663 			 */
664 			wakeup(pp);
665 		} else
666 			mtx_unlock(&p->p_pptr->p_sigacts->ps_mtx);
667 
668 		if (p->p_pptr == p->p_reaper || p->p_pptr == initproc) {
669 			signal_parent = 1;
670 		} else if (p->p_sigparent != 0) {
671 			if (p->p_sigparent == SIGCHLD) {
672 				signal_parent = 1;
673 			} else { /* LINUX thread */
674 				signal_parent = 2;
675 			}
676 		}
677 	} else
678 		PROC_LOCK(p->p_pptr);
679 	sx_xunlock(&proctree_lock);
680 
681 	if (signal_parent == 1) {
682 		childproc_exited(p);
683 	} else if (signal_parent == 2) {
684 		kern_psignal(p->p_pptr, p->p_sigparent);
685 	}
686 
687 	/* Tell the prison that we are gone. */
688 	prison_proc_free(p->p_ucred->cr_prison);
689 
690 	/*
691 	 * The state PRS_ZOMBIE prevents other processes from sending
692 	 * signal to the process, to avoid memory leak, we free memory
693 	 * for signal queue at the time when the state is set.
694 	 */
695 	sigqueue_flush(&p->p_sigqueue);
696 	sigqueue_flush(&td->td_sigqueue);
697 
698 	/*
699 	 * We have to wait until after acquiring all locks before
700 	 * changing p_state.  We need to avoid all possible context
701 	 * switches (including ones from blocking on a mutex) while
702 	 * marked as a zombie.  We also have to set the zombie state
703 	 * before we release the parent process' proc lock to avoid
704 	 * a lost wakeup.  So, we first call wakeup, then we grab the
705 	 * sched lock, update the state, and release the parent process'
706 	 * proc lock.
707 	 */
708 	wakeup(p->p_pptr);
709 	cv_broadcast(&p->p_pwait);
710 	sched_exit(p->p_pptr, td);
711 	PROC_SLOCK(p);
712 	p->p_state = PRS_ZOMBIE;
713 	PROC_UNLOCK(p->p_pptr);
714 
715 	/*
716 	 * Save our children's rusage information in our exit rusage.
717 	 */
718 	PROC_STATLOCK(p);
719 	ruadd(&p->p_ru, &p->p_rux, &p->p_stats->p_cru, &p->p_crux);
720 	PROC_STATUNLOCK(p);
721 
722 	/*
723 	 * Make sure the scheduler takes this thread out of its tables etc.
724 	 * This will also release this thread's reference to the ucred.
725 	 * Other thread parts to release include pcb bits and such.
726 	 */
727 	thread_exit();
728 }
729 
730 #ifndef _SYS_SYSPROTO_H_
731 struct abort2_args {
732 	char *why;
733 	int nargs;
734 	void **args;
735 };
736 #endif
737 
738 int
sys_abort2(struct thread * td,struct abort2_args * uap)739 sys_abort2(struct thread *td, struct abort2_args *uap)
740 {
741 	void *uargs[16];
742 	void **uargsp;
743 	int error, nargs;
744 
745 	nargs = uap->nargs;
746 	if (nargs < 0 || nargs > nitems(uargs))
747 		nargs = -1;
748 	uargsp = NULL;
749 	if (nargs > 0) {
750 		if (uap->args != NULL) {
751 			error = copyin(uap->args, uargs,
752 			    nargs * sizeof(void *));
753 			if (error != 0)
754 				nargs = -1;
755 			else
756 				uargsp = uargs;
757 		} else
758 			nargs = -1;
759 	}
760 	return (kern_abort2(td, uap->why, nargs, uargsp));
761 }
762 
763 /*
764  * kern_abort2()
765  * Arguments:
766  *  why - user pointer to why
767  *  nargs - number of arguments copied or -1 if an error occurred in copying
768  *  args - pointer to an array of pointers in kernel format
769  */
770 int
kern_abort2(struct thread * td,const char * why,int nargs,void ** uargs)771 kern_abort2(struct thread *td, const char *why, int nargs, void **uargs)
772 {
773 	struct proc *p = td->td_proc;
774 	struct sbuf *sb;
775 	int error, i, sig;
776 
777 	/*
778 	 * Do it right now so we can log either proper call of abort2(), or
779 	 * note, that invalid argument was passed. 512 is big enough to
780 	 * handle 16 arguments' descriptions with additional comments.
781 	 */
782 	sb = sbuf_new(NULL, NULL, 512, SBUF_FIXEDLEN);
783 	sbuf_clear(sb);
784 	sbuf_printf(sb, "%s(pid %d uid %d) aborted: ",
785 	    p->p_comm, p->p_pid, td->td_ucred->cr_uid);
786 	/*
787 	 * Since we can't return from abort2(), send SIGKILL in cases, where
788 	 * abort2() was called improperly
789 	 */
790 	sig = SIGKILL;
791 	/* Prevent from DoSes from user-space. */
792 	if (nargs == -1)
793 		goto out;
794 	KASSERT(nargs >= 0 && nargs <= 16, ("called with too many args (%d)",
795 	    nargs));
796 	/*
797 	 * Limit size of 'reason' string to 128. Will fit even when
798 	 * maximal number of arguments was chosen to be logged.
799 	 */
800 	if (why != NULL) {
801 		error = sbuf_copyin(sb, why, 128);
802 		if (error < 0)
803 			goto out;
804 	} else {
805 		sbuf_cat(sb, "(null)");
806 	}
807 	if (nargs > 0) {
808 		sbuf_putc(sb, '(');
809 		for (i = 0;i < nargs; i++)
810 			sbuf_printf(sb, "%s%p", i == 0 ? "" : ", ", uargs[i]);
811 		sbuf_putc(sb, ')');
812 	}
813 	/*
814 	 * Final stage: arguments were proper, string has been
815 	 * successfully copied from userspace, and copying pointers
816 	 * from user-space succeed.
817 	 */
818 	sig = SIGABRT;
819 out:
820 	if (sig == SIGKILL) {
821 		sbuf_trim(sb);
822 		sbuf_cat(sb, " (Reason text inaccessible)");
823 	}
824 	sbuf_cat(sb, "\n");
825 	sbuf_finish(sb);
826 	log(LOG_INFO, "%s", sbuf_data(sb));
827 	sbuf_delete(sb);
828 	PROC_LOCK(p);
829 	sigexit(td, sig);
830 	/* NOTREACHED */
831 }
832 
833 #ifdef COMPAT_43
834 /*
835  * The dirty work is handled by kern_wait().
836  */
837 int
owait(struct thread * td,struct owait_args * uap __unused)838 owait(struct thread *td, struct owait_args *uap __unused)
839 {
840 	int error, status;
841 
842 	error = kern_wait(td, WAIT_ANY, &status, 0, NULL);
843 	if (error == 0)
844 		td->td_retval[1] = status;
845 	return (error);
846 }
847 #endif /* COMPAT_43 */
848 
849 /*
850  * The dirty work is handled by kern_wait().
851  */
852 int
sys_wait4(struct thread * td,struct wait4_args * uap)853 sys_wait4(struct thread *td, struct wait4_args *uap)
854 {
855 	struct rusage ru, *rup;
856 	int error, status;
857 
858 	if (uap->rusage != NULL)
859 		rup = &ru;
860 	else
861 		rup = NULL;
862 	error = kern_wait(td, uap->pid, &status, uap->options, rup);
863 	if (uap->status != NULL && error == 0 && td->td_retval[0] != 0)
864 		error = copyout(&status, uap->status, sizeof(status));
865 	if (uap->rusage != NULL && error == 0 && td->td_retval[0] != 0)
866 		error = copyout(&ru, uap->rusage, sizeof(struct rusage));
867 	return (error);
868 }
869 
870 int
sys_wait6(struct thread * td,struct wait6_args * uap)871 sys_wait6(struct thread *td, struct wait6_args *uap)
872 {
873 	struct __wrusage wru, *wrup;
874 	siginfo_t si, *sip;
875 	idtype_t idtype;
876 	id_t id;
877 	int error, status;
878 
879 	idtype = uap->idtype;
880 	id = uap->id;
881 
882 	if (uap->wrusage != NULL)
883 		wrup = &wru;
884 	else
885 		wrup = NULL;
886 
887 	if (uap->info != NULL) {
888 		sip = &si;
889 		bzero(sip, sizeof(*sip));
890 	} else
891 		sip = NULL;
892 
893 	/*
894 	 *  We expect all callers of wait6() to know about WEXITED and
895 	 *  WTRAPPED.
896 	 */
897 	error = kern_wait6(td, idtype, id, &status, uap->options, wrup, sip);
898 
899 	if (uap->status != NULL && error == 0 && td->td_retval[0] != 0)
900 		error = copyout(&status, uap->status, sizeof(status));
901 	if (uap->wrusage != NULL && error == 0 && td->td_retval[0] != 0)
902 		error = copyout(&wru, uap->wrusage, sizeof(wru));
903 	if (uap->info != NULL && error == 0)
904 		error = copyout(&si, uap->info, sizeof(si));
905 	return (error);
906 }
907 
908 /*
909  * Reap the remains of a zombie process and optionally return status and
910  * rusage.  Asserts and will release both the proctree_lock and the process
911  * lock as part of its work.
912  */
913 void
proc_reap(struct thread * td,struct proc * p,int * status,int options)914 proc_reap(struct thread *td, struct proc *p, int *status, int options)
915 {
916 	struct proc *q, *t;
917 
918 	sx_assert(&proctree_lock, SA_XLOCKED);
919 	PROC_LOCK_ASSERT(p, MA_OWNED);
920 	KASSERT(p->p_state == PRS_ZOMBIE, ("proc_reap: !PRS_ZOMBIE"));
921 
922 	mtx_spin_wait_unlocked(&p->p_slock);
923 
924 	q = td->td_proc;
925 
926 	if (status)
927 		*status = KW_EXITCODE(p->p_xexit, p->p_xsig);
928 	if (options & WNOWAIT) {
929 		/*
930 		 *  Only poll, returning the status.  Caller does not wish to
931 		 * release the proc struct just yet.
932 		 */
933 		PROC_UNLOCK(p);
934 		sx_xunlock(&proctree_lock);
935 		return;
936 	}
937 
938 	PROC_LOCK(q);
939 	sigqueue_take(p->p_ksi);
940 	PROC_UNLOCK(q);
941 
942 	/*
943 	 * If we got the child via a ptrace 'attach', we need to give it back
944 	 * to the old parent.
945 	 */
946 	if (p->p_oppid != p->p_pptr->p_pid) {
947 		PROC_UNLOCK(p);
948 		t = proc_realparent(p);
949 		PROC_LOCK(t);
950 		PROC_LOCK(p);
951 		CTR2(KTR_PTRACE,
952 		    "wait: traced child %d moved back to parent %d", p->p_pid,
953 		    t->p_pid);
954 		proc_reparent(p, t, false);
955 		PROC_UNLOCK(p);
956 		pksignal(t, SIGCHLD, p->p_ksi);
957 		wakeup(t);
958 		cv_broadcast(&p->p_pwait);
959 		PROC_UNLOCK(t);
960 		sx_xunlock(&proctree_lock);
961 		return;
962 	}
963 	PROC_UNLOCK(p);
964 
965 	/*
966 	 * Remove other references to this process to ensure we have an
967 	 * exclusive reference.
968 	 */
969 	sx_xlock(PIDHASHLOCK(p->p_pid));
970 	LIST_REMOVE(p, p_hash);
971 	sx_xunlock(PIDHASHLOCK(p->p_pid));
972 	LIST_REMOVE(p, p_sibling);
973 	reaper_abandon_children(p, true);
974 	reaper_clear(p);
975 	PROC_LOCK(p);
976 	proc_clear_orphan(p);
977 	PROC_UNLOCK(p);
978 	leavepgrp(p);
979 	if (p->p_procdesc != NULL)
980 		procdesc_reap(p);
981 	sx_xunlock(&proctree_lock);
982 
983 	proc_id_clear(PROC_ID_PID, p->p_pid);
984 
985 	PROC_LOCK(p);
986 	knlist_detach(p->p_klist);
987 	p->p_klist = NULL;
988 	PROC_UNLOCK(p);
989 
990 	/*
991 	 * Removal from allproc list and process group list paired with
992 	 * PROC_LOCK which was executed during that time should guarantee
993 	 * nothing can reach this process anymore. As such further locking
994 	 * is unnecessary.
995 	 */
996 	p->p_xexit = p->p_xsig = 0;		/* XXX: why? */
997 
998 	PROC_LOCK(q);
999 	ruadd(&q->p_stats->p_cru, &q->p_crux, &p->p_ru, &p->p_rux);
1000 	PROC_UNLOCK(q);
1001 
1002 	/*
1003 	 * Decrement the count of procs running with this uid.
1004 	 */
1005 	(void)chgproccnt(p->p_ucred->cr_ruidinfo, -1, 0);
1006 
1007 	/*
1008 	 * Destroy resource accounting information associated with the process.
1009 	 */
1010 #ifdef RACCT
1011 	if (racct_enable) {
1012 		PROC_LOCK(p);
1013 		racct_sub(p, RACCT_NPROC, 1);
1014 		PROC_UNLOCK(p);
1015 	}
1016 #endif
1017 	racct_proc_exit(p);
1018 
1019 	/*
1020 	 * Free credentials, arguments, and sigacts.
1021 	 */
1022 	proc_unset_cred(p);
1023 	pargs_drop(p->p_args);
1024 	p->p_args = NULL;
1025 	sigacts_free(p->p_sigacts);
1026 	p->p_sigacts = NULL;
1027 
1028 	/*
1029 	 * Do any thread-system specific cleanups.
1030 	 */
1031 	thread_wait(p);
1032 
1033 	/*
1034 	 * Give vm and machine-dependent layer a chance to free anything that
1035 	 * cpu_exit couldn't release while still running in process context.
1036 	 */
1037 	vm_waitproc(p);
1038 #ifdef MAC
1039 	mac_proc_destroy(p);
1040 #endif
1041 
1042 	KASSERT(FIRST_THREAD_IN_PROC(p),
1043 	    ("proc_reap: no residual thread!"));
1044 	uma_zfree(proc_zone, p);
1045 	atomic_add_int(&nprocs, -1);
1046 }
1047 
1048 static int
proc_to_reap(struct thread * td,struct proc * p,idtype_t idtype,id_t id,int * status,int options,struct __wrusage * wrusage,siginfo_t * siginfo,int check_only)1049 proc_to_reap(struct thread *td, struct proc *p, idtype_t idtype, id_t id,
1050     int *status, int options, struct __wrusage *wrusage, siginfo_t *siginfo,
1051     int check_only)
1052 {
1053 	struct rusage *rup;
1054 
1055 	sx_assert(&proctree_lock, SA_XLOCKED);
1056 
1057 	PROC_LOCK(p);
1058 
1059 	switch (idtype) {
1060 	case P_ALL:
1061 		if (p->p_procdesc == NULL ||
1062 		   (p->p_pptr == td->td_proc &&
1063 		   (p->p_flag & P_TRACED) != 0)) {
1064 			break;
1065 		}
1066 
1067 		PROC_UNLOCK(p);
1068 		return (0);
1069 	case P_PID:
1070 		if (p->p_pid != (pid_t)id) {
1071 			PROC_UNLOCK(p);
1072 			return (0);
1073 		}
1074 		break;
1075 	case P_PGID:
1076 		if (p->p_pgid != (pid_t)id) {
1077 			PROC_UNLOCK(p);
1078 			return (0);
1079 		}
1080 		break;
1081 	case P_SID:
1082 		if (p->p_session->s_sid != (pid_t)id) {
1083 			PROC_UNLOCK(p);
1084 			return (0);
1085 		}
1086 		break;
1087 	case P_UID:
1088 		if (p->p_ucred->cr_uid != (uid_t)id) {
1089 			PROC_UNLOCK(p);
1090 			return (0);
1091 		}
1092 		break;
1093 	case P_GID:
1094 		if (p->p_ucred->cr_gid != (gid_t)id) {
1095 			PROC_UNLOCK(p);
1096 			return (0);
1097 		}
1098 		break;
1099 	case P_JAILID:
1100 		if (p->p_ucred->cr_prison->pr_id != (int)id) {
1101 			PROC_UNLOCK(p);
1102 			return (0);
1103 		}
1104 		break;
1105 	/*
1106 	 * It seems that the thread structures get zeroed out
1107 	 * at process exit.  This makes it impossible to
1108 	 * support P_SETID, P_CID or P_CPUID.
1109 	 */
1110 	default:
1111 		PROC_UNLOCK(p);
1112 		return (0);
1113 	}
1114 
1115 	if (p_canwait(td, p)) {
1116 		PROC_UNLOCK(p);
1117 		return (0);
1118 	}
1119 
1120 	if (((options & WEXITED) == 0) && (p->p_state == PRS_ZOMBIE)) {
1121 		PROC_UNLOCK(p);
1122 		return (0);
1123 	}
1124 
1125 	/*
1126 	 * This special case handles a kthread spawned by linux_clone
1127 	 * (see linux_misc.c).  The linux_wait4 and linux_waitpid
1128 	 * functions need to be able to distinguish between waiting
1129 	 * on a process and waiting on a thread.  It is a thread if
1130 	 * p_sigparent is not SIGCHLD, and the WLINUXCLONE option
1131 	 * signifies we want to wait for threads and not processes.
1132 	 */
1133 	if ((p->p_sigparent != SIGCHLD) ^
1134 	    ((options & WLINUXCLONE) != 0)) {
1135 		PROC_UNLOCK(p);
1136 		return (0);
1137 	}
1138 
1139 	if (siginfo != NULL) {
1140 		bzero(siginfo, sizeof(*siginfo));
1141 		siginfo->si_errno = 0;
1142 
1143 		/*
1144 		 * SUSv4 requires that the si_signo value is always
1145 		 * SIGCHLD. Obey it despite the rfork(2) interface
1146 		 * allows to request other signal for child exit
1147 		 * notification.
1148 		 */
1149 		siginfo->si_signo = SIGCHLD;
1150 
1151 		/*
1152 		 *  This is still a rough estimate.  We will fix the
1153 		 *  cases TRAPPED, STOPPED, and CONTINUED later.
1154 		 */
1155 		if (WCOREDUMP(p->p_xsig)) {
1156 			siginfo->si_code = CLD_DUMPED;
1157 			siginfo->si_status = WTERMSIG(p->p_xsig);
1158 		} else if (WIFSIGNALED(p->p_xsig)) {
1159 			siginfo->si_code = CLD_KILLED;
1160 			siginfo->si_status = WTERMSIG(p->p_xsig);
1161 		} else {
1162 			siginfo->si_code = CLD_EXITED;
1163 			siginfo->si_status = p->p_xexit;
1164 		}
1165 
1166 		siginfo->si_pid = p->p_pid;
1167 		siginfo->si_uid = p->p_ucred->cr_uid;
1168 
1169 		/*
1170 		 * The si_addr field would be useful additional
1171 		 * detail, but apparently the PC value may be lost
1172 		 * when we reach this point.  bzero() above sets
1173 		 * siginfo->si_addr to NULL.
1174 		 */
1175 	}
1176 
1177 	/*
1178 	 * There should be no reason to limit resources usage info to
1179 	 * exited processes only.  A snapshot about any resources used
1180 	 * by a stopped process may be exactly what is needed.
1181 	 */
1182 	if (wrusage != NULL) {
1183 		rup = &wrusage->wru_self;
1184 		*rup = p->p_ru;
1185 		PROC_STATLOCK(p);
1186 		calcru(p, &rup->ru_utime, &rup->ru_stime);
1187 		PROC_STATUNLOCK(p);
1188 
1189 		rup = &wrusage->wru_children;
1190 		*rup = p->p_stats->p_cru;
1191 		calccru(p, &rup->ru_utime, &rup->ru_stime);
1192 	}
1193 
1194 	if (p->p_state == PRS_ZOMBIE && !check_only) {
1195 		proc_reap(td, p, status, options);
1196 		return (-1);
1197 	}
1198 	return (1);
1199 }
1200 
1201 int
kern_wait(struct thread * td,pid_t pid,int * status,int options,struct rusage * rusage)1202 kern_wait(struct thread *td, pid_t pid, int *status, int options,
1203     struct rusage *rusage)
1204 {
1205 	struct __wrusage wru, *wrup;
1206 	idtype_t idtype;
1207 	id_t id;
1208 	int ret;
1209 
1210 	/*
1211 	 * Translate the special pid values into the (idtype, pid)
1212 	 * pair for kern_wait6.  The WAIT_MYPGRP case is handled by
1213 	 * kern_wait6() on its own.
1214 	 */
1215 	if (pid == WAIT_ANY) {
1216 		idtype = P_ALL;
1217 		id = 0;
1218 	} else if (pid < 0) {
1219 		idtype = P_PGID;
1220 		id = (id_t)-pid;
1221 	} else {
1222 		idtype = P_PID;
1223 		id = (id_t)pid;
1224 	}
1225 
1226 	if (rusage != NULL)
1227 		wrup = &wru;
1228 	else
1229 		wrup = NULL;
1230 
1231 	/*
1232 	 * For backward compatibility we implicitly add flags WEXITED
1233 	 * and WTRAPPED here.
1234 	 */
1235 	options |= WEXITED | WTRAPPED;
1236 	ret = kern_wait6(td, idtype, id, status, options, wrup, NULL);
1237 	if (rusage != NULL)
1238 		*rusage = wru.wru_self;
1239 	return (ret);
1240 }
1241 
1242 static void
report_alive_proc(struct thread * td,struct proc * p,siginfo_t * siginfo,int * status,int options,int si_code)1243 report_alive_proc(struct thread *td, struct proc *p, siginfo_t *siginfo,
1244     int *status, int options, int si_code)
1245 {
1246 	bool cont;
1247 
1248 	PROC_LOCK_ASSERT(p, MA_OWNED);
1249 	sx_assert(&proctree_lock, SA_XLOCKED);
1250 	MPASS(si_code == CLD_TRAPPED || si_code == CLD_STOPPED ||
1251 	    si_code == CLD_CONTINUED);
1252 
1253 	cont = si_code == CLD_CONTINUED;
1254 	if ((options & WNOWAIT) == 0) {
1255 		if (cont)
1256 			p->p_flag &= ~P_CONTINUED;
1257 		else
1258 			p->p_flag |= P_WAITED;
1259 		if (kern_wait_dequeue_sigchld &&
1260 		    (td->td_proc->p_sysent->sv_flags & SV_SIG_WAITNDQ) == 0) {
1261 			PROC_LOCK(td->td_proc);
1262 			sigqueue_take(p->p_ksi);
1263 			PROC_UNLOCK(td->td_proc);
1264 		}
1265 	}
1266 	sx_xunlock(&proctree_lock);
1267 	if (siginfo != NULL) {
1268 		siginfo->si_code = si_code;
1269 		siginfo->si_status = cont ? SIGCONT : p->p_xsig;
1270 	}
1271 	if (status != NULL)
1272 		*status = cont ? SIGCONT : W_STOPCODE(p->p_xsig);
1273 	td->td_retval[0] = p->p_pid;
1274 	PROC_UNLOCK(p);
1275 }
1276 
1277 int
kern_wait6(struct thread * td,idtype_t idtype,id_t id,int * status,int options,struct __wrusage * wrusage,siginfo_t * siginfo)1278 kern_wait6(struct thread *td, idtype_t idtype, id_t id, int *status,
1279     int options, struct __wrusage *wrusage, siginfo_t *siginfo)
1280 {
1281 	struct proc *p, *q;
1282 	pid_t pid;
1283 	int error, nfound, ret;
1284 	bool report;
1285 
1286 	AUDIT_ARG_VALUE((int)idtype);	/* XXX - This is likely wrong! */
1287 	AUDIT_ARG_PID((pid_t)id);	/* XXX - This may be wrong! */
1288 	AUDIT_ARG_VALUE(options);
1289 
1290 	q = td->td_proc;
1291 
1292 	if ((pid_t)id == WAIT_MYPGRP && (idtype == P_PID || idtype == P_PGID)) {
1293 		PROC_LOCK(q);
1294 		id = (id_t)q->p_pgid;
1295 		PROC_UNLOCK(q);
1296 		idtype = P_PGID;
1297 	}
1298 
1299 	/* If we don't know the option, just return. */
1300 	if ((options & ~(WUNTRACED | WNOHANG | WCONTINUED | WNOWAIT |
1301 	    WEXITED | WTRAPPED | WLINUXCLONE)) != 0)
1302 		return (EINVAL);
1303 	if ((options & (WEXITED | WUNTRACED | WCONTINUED | WTRAPPED)) == 0) {
1304 		/*
1305 		 * We will be unable to find any matching processes,
1306 		 * because there are no known events to look for.
1307 		 * Prefer to return error instead of blocking
1308 		 * indefinitely.
1309 		 */
1310 		return (EINVAL);
1311 	}
1312 
1313 loop:
1314 	if (q->p_flag & P_STATCHILD) {
1315 		PROC_LOCK(q);
1316 		q->p_flag &= ~P_STATCHILD;
1317 		PROC_UNLOCK(q);
1318 	}
1319 	sx_xlock(&proctree_lock);
1320 loop_locked:
1321 	nfound = 0;
1322 	LIST_FOREACH(p, &q->p_children, p_sibling) {
1323 		pid = p->p_pid;
1324 		ret = proc_to_reap(td, p, idtype, id, status, options,
1325 		    wrusage, siginfo, 0);
1326 		if (ret == 0)
1327 			continue;
1328 		else if (ret != 1) {
1329 			td->td_retval[0] = pid;
1330 			return (0);
1331 		}
1332 
1333 		/*
1334 		 * When running in capsicum(4) mode, make wait(2) ignore
1335 		 * processes created with pdfork(2).  This is because one can
1336 		 * disown them - by passing their process descriptor to another
1337 		 * process - which means it needs to be prevented from touching
1338 		 * them afterwards.
1339 		 */
1340 		if (IN_CAPABILITY_MODE(td) && p->p_procdesc != NULL) {
1341 			PROC_UNLOCK(p);
1342 			continue;
1343 		}
1344 
1345 		nfound++;
1346 		PROC_LOCK_ASSERT(p, MA_OWNED);
1347 
1348 		if ((options & WTRAPPED) != 0 &&
1349 		    (p->p_flag & P_TRACED) != 0) {
1350 			PROC_SLOCK(p);
1351 			report =
1352 			    ((p->p_flag & (P_STOPPED_TRACE | P_STOPPED_SIG)) &&
1353 			    p->p_suspcount == p->p_numthreads &&
1354 			    (p->p_flag & P_WAITED) == 0);
1355 			PROC_SUNLOCK(p);
1356 			if (report) {
1357 			CTR4(KTR_PTRACE,
1358 			    "wait: returning trapped pid %d status %#x "
1359 			    "(xstat %d) xthread %d",
1360 			    p->p_pid, W_STOPCODE(p->p_xsig), p->p_xsig,
1361 			    p->p_xthread != NULL ?
1362 			    p->p_xthread->td_tid : -1);
1363 				report_alive_proc(td, p, siginfo, status,
1364 				    options, CLD_TRAPPED);
1365 				return (0);
1366 			}
1367 		}
1368 		if ((options & WUNTRACED) != 0 &&
1369 		    (p->p_flag & P_STOPPED_SIG) != 0) {
1370 			PROC_SLOCK(p);
1371 			report = (p->p_suspcount == p->p_numthreads &&
1372 			    ((p->p_flag & P_WAITED) == 0));
1373 			PROC_SUNLOCK(p);
1374 			if (report) {
1375 				report_alive_proc(td, p, siginfo, status,
1376 				    options, CLD_STOPPED);
1377 				return (0);
1378 			}
1379 		}
1380 		if ((options & WCONTINUED) != 0 &&
1381 		    (p->p_flag & P_CONTINUED) != 0) {
1382 			report_alive_proc(td, p, siginfo, status, options,
1383 			    CLD_CONTINUED);
1384 			return (0);
1385 		}
1386 		PROC_UNLOCK(p);
1387 	}
1388 
1389 	/*
1390 	 * Look in the orphans list too, to allow the parent to
1391 	 * collect it's child exit status even if child is being
1392 	 * debugged.
1393 	 *
1394 	 * Debugger detaches from the parent upon successful
1395 	 * switch-over from parent to child.  At this point due to
1396 	 * re-parenting the parent loses the child to debugger and a
1397 	 * wait4(2) call would report that it has no children to wait
1398 	 * for.  By maintaining a list of orphans we allow the parent
1399 	 * to successfully wait until the child becomes a zombie.
1400 	 */
1401 	if (nfound == 0) {
1402 		LIST_FOREACH(p, &q->p_orphans, p_orphan) {
1403 			ret = proc_to_reap(td, p, idtype, id, NULL, options,
1404 			    NULL, NULL, 1);
1405 			if (ret != 0) {
1406 				KASSERT(ret != -1, ("reaped an orphan (pid %d)",
1407 				    (int)td->td_retval[0]));
1408 				PROC_UNLOCK(p);
1409 				nfound++;
1410 				break;
1411 			}
1412 		}
1413 	}
1414 	if (nfound == 0) {
1415 		sx_xunlock(&proctree_lock);
1416 		return (ECHILD);
1417 	}
1418 	if (options & WNOHANG) {
1419 		sx_xunlock(&proctree_lock);
1420 		td->td_retval[0] = 0;
1421 		return (0);
1422 	}
1423 	PROC_LOCK(q);
1424 	if (q->p_flag & P_STATCHILD) {
1425 		q->p_flag &= ~P_STATCHILD;
1426 		PROC_UNLOCK(q);
1427 		goto loop_locked;
1428 	}
1429 	sx_xunlock(&proctree_lock);
1430 	error = msleep(q, &q->p_mtx, PWAIT | PCATCH | PDROP, "wait", 0);
1431 	if (error)
1432 		return (error);
1433 	goto loop;
1434 }
1435 
1436 void
proc_add_orphan(struct proc * child,struct proc * parent)1437 proc_add_orphan(struct proc *child, struct proc *parent)
1438 {
1439 
1440 	sx_assert(&proctree_lock, SX_XLOCKED);
1441 	KASSERT((child->p_flag & P_TRACED) != 0,
1442 	    ("proc_add_orphan: not traced"));
1443 
1444 	if (LIST_EMPTY(&parent->p_orphans)) {
1445 		child->p_treeflag |= P_TREE_FIRST_ORPHAN;
1446 		LIST_INSERT_HEAD(&parent->p_orphans, child, p_orphan);
1447 	} else {
1448 		LIST_INSERT_AFTER(LIST_FIRST(&parent->p_orphans),
1449 		    child, p_orphan);
1450 	}
1451 	child->p_treeflag |= P_TREE_ORPHANED;
1452 }
1453 
1454 /*
1455  * Make process 'parent' the new parent of process 'child'.
1456  * Must be called with an exclusive hold of proctree lock.
1457  */
1458 void
proc_reparent(struct proc * child,struct proc * parent,bool set_oppid)1459 proc_reparent(struct proc *child, struct proc *parent, bool set_oppid)
1460 {
1461 
1462 	sx_assert(&proctree_lock, SX_XLOCKED);
1463 	PROC_LOCK_ASSERT(child, MA_OWNED);
1464 	if (child->p_pptr == parent)
1465 		return;
1466 
1467 	PROC_LOCK(child->p_pptr);
1468 	sigqueue_take(child->p_ksi);
1469 	PROC_UNLOCK(child->p_pptr);
1470 	LIST_REMOVE(child, p_sibling);
1471 	LIST_INSERT_HEAD(&parent->p_children, child, p_sibling);
1472 
1473 	proc_clear_orphan(child);
1474 	if ((child->p_flag & P_TRACED) != 0) {
1475 		proc_add_orphan(child, child->p_pptr);
1476 	}
1477 
1478 	child->p_pptr = parent;
1479 	if (set_oppid)
1480 		child->p_oppid = parent->p_pid;
1481 }
1482