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 * Destroy resource accounting information associated with the process.
1004 */
1005 #ifdef RACCT
1006 if (racct_enable) {
1007 PROC_LOCK(p);
1008 racct_sub(p, RACCT_NPROC, 1);
1009 PROC_UNLOCK(p);
1010 }
1011 #endif
1012 racct_proc_exit(p);
1013
1014 /*
1015 * Free credentials, arguments, and sigacts, and decrement the count of
1016 * processes running with this uid.
1017 */
1018 proc_unset_cred(p, true);
1019 pargs_drop(p->p_args);
1020 p->p_args = NULL;
1021 sigacts_free(p->p_sigacts);
1022 p->p_sigacts = NULL;
1023
1024 /*
1025 * Do any thread-system specific cleanups.
1026 */
1027 thread_wait(p);
1028
1029 /*
1030 * Give vm and machine-dependent layer a chance to free anything that
1031 * cpu_exit couldn't release while still running in process context.
1032 */
1033 vm_waitproc(p);
1034 #ifdef MAC
1035 mac_proc_destroy(p);
1036 #endif
1037
1038 KASSERT(FIRST_THREAD_IN_PROC(p),
1039 ("proc_reap: no residual thread!"));
1040 uma_zfree(proc_zone, p);
1041 atomic_add_int(&nprocs, -1);
1042 }
1043
1044 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)1045 proc_to_reap(struct thread *td, struct proc *p, idtype_t idtype, id_t id,
1046 int *status, int options, struct __wrusage *wrusage, siginfo_t *siginfo,
1047 int check_only)
1048 {
1049 struct rusage *rup;
1050
1051 sx_assert(&proctree_lock, SA_XLOCKED);
1052
1053 PROC_LOCK(p);
1054
1055 switch (idtype) {
1056 case P_ALL:
1057 if (p->p_procdesc == NULL ||
1058 (p->p_pptr == td->td_proc &&
1059 (p->p_flag & P_TRACED) != 0)) {
1060 break;
1061 }
1062
1063 PROC_UNLOCK(p);
1064 return (0);
1065 case P_PID:
1066 if (p->p_pid != (pid_t)id) {
1067 PROC_UNLOCK(p);
1068 return (0);
1069 }
1070 break;
1071 case P_PGID:
1072 if (p->p_pgid != (pid_t)id) {
1073 PROC_UNLOCK(p);
1074 return (0);
1075 }
1076 break;
1077 case P_SID:
1078 if (p->p_session->s_sid != (pid_t)id) {
1079 PROC_UNLOCK(p);
1080 return (0);
1081 }
1082 break;
1083 case P_UID:
1084 if (p->p_ucred->cr_uid != (uid_t)id) {
1085 PROC_UNLOCK(p);
1086 return (0);
1087 }
1088 break;
1089 case P_GID:
1090 if (p->p_ucred->cr_gid != (gid_t)id) {
1091 PROC_UNLOCK(p);
1092 return (0);
1093 }
1094 break;
1095 case P_JAILID:
1096 if (p->p_ucred->cr_prison->pr_id != (int)id) {
1097 PROC_UNLOCK(p);
1098 return (0);
1099 }
1100 break;
1101 /*
1102 * It seems that the thread structures get zeroed out
1103 * at process exit. This makes it impossible to
1104 * support P_SETID, P_CID or P_CPUID.
1105 */
1106 default:
1107 PROC_UNLOCK(p);
1108 return (0);
1109 }
1110
1111 if (p_canwait(td, p)) {
1112 PROC_UNLOCK(p);
1113 return (0);
1114 }
1115
1116 if (((options & WEXITED) == 0) && (p->p_state == PRS_ZOMBIE)) {
1117 PROC_UNLOCK(p);
1118 return (0);
1119 }
1120
1121 /*
1122 * This special case handles a kthread spawned by linux_clone
1123 * (see linux_misc.c). The linux_wait4 and linux_waitpid
1124 * functions need to be able to distinguish between waiting
1125 * on a process and waiting on a thread. It is a thread if
1126 * p_sigparent is not SIGCHLD, and the WLINUXCLONE option
1127 * signifies we want to wait for threads and not processes.
1128 */
1129 if ((p->p_sigparent != SIGCHLD) ^
1130 ((options & WLINUXCLONE) != 0)) {
1131 PROC_UNLOCK(p);
1132 return (0);
1133 }
1134
1135 if (siginfo != NULL) {
1136 bzero(siginfo, sizeof(*siginfo));
1137 siginfo->si_errno = 0;
1138
1139 /*
1140 * SUSv4 requires that the si_signo value is always
1141 * SIGCHLD. Obey it despite the rfork(2) interface
1142 * allows to request other signal for child exit
1143 * notification.
1144 */
1145 siginfo->si_signo = SIGCHLD;
1146
1147 /*
1148 * This is still a rough estimate. We will fix the
1149 * cases TRAPPED, STOPPED, and CONTINUED later.
1150 */
1151 if (WCOREDUMP(p->p_xsig)) {
1152 siginfo->si_code = CLD_DUMPED;
1153 siginfo->si_status = WTERMSIG(p->p_xsig);
1154 } else if (WIFSIGNALED(p->p_xsig)) {
1155 siginfo->si_code = CLD_KILLED;
1156 siginfo->si_status = WTERMSIG(p->p_xsig);
1157 } else {
1158 siginfo->si_code = CLD_EXITED;
1159 siginfo->si_status = p->p_xexit;
1160 }
1161
1162 siginfo->si_pid = p->p_pid;
1163 siginfo->si_uid = p->p_ucred->cr_uid;
1164
1165 /*
1166 * The si_addr field would be useful additional
1167 * detail, but apparently the PC value may be lost
1168 * when we reach this point. bzero() above sets
1169 * siginfo->si_addr to NULL.
1170 */
1171 }
1172
1173 /*
1174 * There should be no reason to limit resources usage info to
1175 * exited processes only. A snapshot about any resources used
1176 * by a stopped process may be exactly what is needed.
1177 */
1178 if (wrusage != NULL) {
1179 rup = &wrusage->wru_self;
1180 *rup = p->p_ru;
1181 PROC_STATLOCK(p);
1182 calcru(p, &rup->ru_utime, &rup->ru_stime);
1183 PROC_STATUNLOCK(p);
1184
1185 rup = &wrusage->wru_children;
1186 *rup = p->p_stats->p_cru;
1187 calccru(p, &rup->ru_utime, &rup->ru_stime);
1188 }
1189
1190 if (p->p_state == PRS_ZOMBIE && !check_only) {
1191 proc_reap(td, p, status, options);
1192 return (-1);
1193 }
1194 return (1);
1195 }
1196
1197 int
kern_wait(struct thread * td,pid_t pid,int * status,int options,struct rusage * rusage)1198 kern_wait(struct thread *td, pid_t pid, int *status, int options,
1199 struct rusage *rusage)
1200 {
1201 struct __wrusage wru, *wrup;
1202 idtype_t idtype;
1203 id_t id;
1204 int ret;
1205
1206 /*
1207 * Translate the special pid values into the (idtype, pid)
1208 * pair for kern_wait6. The WAIT_MYPGRP case is handled by
1209 * kern_wait6() on its own.
1210 */
1211 if (pid == WAIT_ANY) {
1212 idtype = P_ALL;
1213 id = 0;
1214 } else if (pid < 0) {
1215 idtype = P_PGID;
1216 id = (id_t)-pid;
1217 } else {
1218 idtype = P_PID;
1219 id = (id_t)pid;
1220 }
1221
1222 if (rusage != NULL)
1223 wrup = &wru;
1224 else
1225 wrup = NULL;
1226
1227 /*
1228 * For backward compatibility we implicitly add flags WEXITED
1229 * and WTRAPPED here.
1230 */
1231 options |= WEXITED | WTRAPPED;
1232 ret = kern_wait6(td, idtype, id, status, options, wrup, NULL);
1233 if (rusage != NULL)
1234 *rusage = wru.wru_self;
1235 return (ret);
1236 }
1237
1238 static void
report_alive_proc(struct thread * td,struct proc * p,siginfo_t * siginfo,int * status,int options,int si_code)1239 report_alive_proc(struct thread *td, struct proc *p, siginfo_t *siginfo,
1240 int *status, int options, int si_code)
1241 {
1242 bool cont;
1243
1244 PROC_LOCK_ASSERT(p, MA_OWNED);
1245 sx_assert(&proctree_lock, SA_XLOCKED);
1246 MPASS(si_code == CLD_TRAPPED || si_code == CLD_STOPPED ||
1247 si_code == CLD_CONTINUED);
1248
1249 cont = si_code == CLD_CONTINUED;
1250 if ((options & WNOWAIT) == 0) {
1251 if (cont)
1252 p->p_flag &= ~P_CONTINUED;
1253 else
1254 p->p_flag |= P_WAITED;
1255 if (kern_wait_dequeue_sigchld &&
1256 (td->td_proc->p_sysent->sv_flags & SV_SIG_WAITNDQ) == 0) {
1257 PROC_LOCK(td->td_proc);
1258 sigqueue_take(p->p_ksi);
1259 PROC_UNLOCK(td->td_proc);
1260 }
1261 }
1262 sx_xunlock(&proctree_lock);
1263 if (siginfo != NULL) {
1264 siginfo->si_code = si_code;
1265 siginfo->si_status = cont ? SIGCONT : p->p_xsig;
1266 }
1267 if (status != NULL)
1268 *status = cont ? SIGCONT : W_STOPCODE(p->p_xsig);
1269 td->td_retval[0] = p->p_pid;
1270 PROC_UNLOCK(p);
1271 }
1272
1273 int
kern_wait6(struct thread * td,idtype_t idtype,id_t id,int * status,int options,struct __wrusage * wrusage,siginfo_t * siginfo)1274 kern_wait6(struct thread *td, idtype_t idtype, id_t id, int *status,
1275 int options, struct __wrusage *wrusage, siginfo_t *siginfo)
1276 {
1277 struct proc *p, *q;
1278 pid_t pid;
1279 int error, nfound, ret;
1280 bool report;
1281
1282 AUDIT_ARG_VALUE((int)idtype); /* XXX - This is likely wrong! */
1283 AUDIT_ARG_PID((pid_t)id); /* XXX - This may be wrong! */
1284 AUDIT_ARG_VALUE(options);
1285
1286 q = td->td_proc;
1287
1288 if ((pid_t)id == WAIT_MYPGRP && (idtype == P_PID || idtype == P_PGID)) {
1289 PROC_LOCK(q);
1290 id = (id_t)q->p_pgid;
1291 PROC_UNLOCK(q);
1292 idtype = P_PGID;
1293 }
1294
1295 /* If we don't know the option, just return. */
1296 if ((options & ~(WUNTRACED | WNOHANG | WCONTINUED | WNOWAIT |
1297 WEXITED | WTRAPPED | WLINUXCLONE)) != 0)
1298 return (EINVAL);
1299 if ((options & (WEXITED | WUNTRACED | WCONTINUED | WTRAPPED)) == 0) {
1300 /*
1301 * We will be unable to find any matching processes,
1302 * because there are no known events to look for.
1303 * Prefer to return error instead of blocking
1304 * indefinitely.
1305 */
1306 return (EINVAL);
1307 }
1308
1309 loop:
1310 if (q->p_flag & P_STATCHILD) {
1311 PROC_LOCK(q);
1312 q->p_flag &= ~P_STATCHILD;
1313 PROC_UNLOCK(q);
1314 }
1315 sx_xlock(&proctree_lock);
1316 loop_locked:
1317 nfound = 0;
1318 LIST_FOREACH(p, &q->p_children, p_sibling) {
1319 pid = p->p_pid;
1320 ret = proc_to_reap(td, p, idtype, id, status, options,
1321 wrusage, siginfo, 0);
1322 if (ret == 0)
1323 continue;
1324 else if (ret != 1) {
1325 td->td_retval[0] = pid;
1326 return (0);
1327 }
1328
1329 /*
1330 * When running in capsicum(4) mode, make wait(2) ignore
1331 * processes created with pdfork(2). This is because one can
1332 * disown them - by passing their process descriptor to another
1333 * process - which means it needs to be prevented from touching
1334 * them afterwards.
1335 */
1336 if (IN_CAPABILITY_MODE(td) && p->p_procdesc != NULL) {
1337 PROC_UNLOCK(p);
1338 continue;
1339 }
1340
1341 nfound++;
1342 PROC_LOCK_ASSERT(p, MA_OWNED);
1343
1344 if ((options & WTRAPPED) != 0 &&
1345 (p->p_flag & P_TRACED) != 0) {
1346 PROC_SLOCK(p);
1347 report =
1348 ((p->p_flag & (P_STOPPED_TRACE | P_STOPPED_SIG)) &&
1349 p->p_suspcount == p->p_numthreads &&
1350 (p->p_flag & P_WAITED) == 0);
1351 PROC_SUNLOCK(p);
1352 if (report) {
1353 CTR4(KTR_PTRACE,
1354 "wait: returning trapped pid %d status %#x "
1355 "(xstat %d) xthread %d",
1356 p->p_pid, W_STOPCODE(p->p_xsig), p->p_xsig,
1357 p->p_xthread != NULL ?
1358 p->p_xthread->td_tid : -1);
1359 report_alive_proc(td, p, siginfo, status,
1360 options, CLD_TRAPPED);
1361 return (0);
1362 }
1363 }
1364 if ((options & WUNTRACED) != 0 &&
1365 (p->p_flag & P_STOPPED_SIG) != 0) {
1366 PROC_SLOCK(p);
1367 report = (p->p_suspcount == p->p_numthreads &&
1368 ((p->p_flag & P_WAITED) == 0));
1369 PROC_SUNLOCK(p);
1370 if (report) {
1371 report_alive_proc(td, p, siginfo, status,
1372 options, CLD_STOPPED);
1373 return (0);
1374 }
1375 }
1376 if ((options & WCONTINUED) != 0 &&
1377 (p->p_flag & P_CONTINUED) != 0) {
1378 report_alive_proc(td, p, siginfo, status, options,
1379 CLD_CONTINUED);
1380 return (0);
1381 }
1382 PROC_UNLOCK(p);
1383 }
1384
1385 /*
1386 * Look in the orphans list too, to allow the parent to
1387 * collect it's child exit status even if child is being
1388 * debugged.
1389 *
1390 * Debugger detaches from the parent upon successful
1391 * switch-over from parent to child. At this point due to
1392 * re-parenting the parent loses the child to debugger and a
1393 * wait4(2) call would report that it has no children to wait
1394 * for. By maintaining a list of orphans we allow the parent
1395 * to successfully wait until the child becomes a zombie.
1396 */
1397 if (nfound == 0) {
1398 LIST_FOREACH(p, &q->p_orphans, p_orphan) {
1399 ret = proc_to_reap(td, p, idtype, id, NULL, options,
1400 NULL, NULL, 1);
1401 if (ret != 0) {
1402 KASSERT(ret != -1, ("reaped an orphan (pid %d)",
1403 (int)td->td_retval[0]));
1404 PROC_UNLOCK(p);
1405 nfound++;
1406 break;
1407 }
1408 }
1409 }
1410 if (nfound == 0) {
1411 sx_xunlock(&proctree_lock);
1412 return (ECHILD);
1413 }
1414 if (options & WNOHANG) {
1415 sx_xunlock(&proctree_lock);
1416 td->td_retval[0] = 0;
1417 return (0);
1418 }
1419 PROC_LOCK(q);
1420 if (q->p_flag & P_STATCHILD) {
1421 q->p_flag &= ~P_STATCHILD;
1422 PROC_UNLOCK(q);
1423 goto loop_locked;
1424 }
1425 sx_xunlock(&proctree_lock);
1426 error = msleep(q, &q->p_mtx, PWAIT | PCATCH | PDROP, "wait", 0);
1427 if (error)
1428 return (error);
1429 goto loop;
1430 }
1431
1432 void
proc_add_orphan(struct proc * child,struct proc * parent)1433 proc_add_orphan(struct proc *child, struct proc *parent)
1434 {
1435
1436 sx_assert(&proctree_lock, SX_XLOCKED);
1437 KASSERT((child->p_flag & P_TRACED) != 0,
1438 ("proc_add_orphan: not traced"));
1439
1440 if (LIST_EMPTY(&parent->p_orphans)) {
1441 child->p_treeflag |= P_TREE_FIRST_ORPHAN;
1442 LIST_INSERT_HEAD(&parent->p_orphans, child, p_orphan);
1443 } else {
1444 LIST_INSERT_AFTER(LIST_FIRST(&parent->p_orphans),
1445 child, p_orphan);
1446 }
1447 child->p_treeflag |= P_TREE_ORPHANED;
1448 }
1449
1450 /*
1451 * Make process 'parent' the new parent of process 'child'.
1452 * Must be called with an exclusive hold of proctree lock.
1453 */
1454 void
proc_reparent(struct proc * child,struct proc * parent,bool set_oppid)1455 proc_reparent(struct proc *child, struct proc *parent, bool set_oppid)
1456 {
1457
1458 sx_assert(&proctree_lock, SX_XLOCKED);
1459 PROC_LOCK_ASSERT(child, MA_OWNED);
1460 if (child->p_pptr == parent)
1461 return;
1462
1463 PROC_LOCK(child->p_pptr);
1464 sigqueue_take(child->p_ksi);
1465 PROC_UNLOCK(child->p_pptr);
1466 LIST_REMOVE(child, p_sibling);
1467 LIST_INSERT_HEAD(&parent->p_children, child, p_sibling);
1468
1469 proc_clear_orphan(child);
1470 if ((child->p_flag & P_TRACED) != 0) {
1471 proc_add_orphan(child, child->p_pptr);
1472 }
1473
1474 child->p_pptr = parent;
1475 if (set_oppid)
1476 child->p_oppid = parent->p_pid;
1477 }
1478