xref: /linux/kernel/exit.c (revision d91517839e5d95adc0cf4b28caa7af62a71de526)
1 /*
2  *  linux/kernel/exit.c
3  *
4  *  Copyright (C) 1991, 1992  Linus Torvalds
5  */
6 
7 #include <linux/mm.h>
8 #include <linux/slab.h>
9 #include <linux/interrupt.h>
10 #include <linux/module.h>
11 #include <linux/capability.h>
12 #include <linux/completion.h>
13 #include <linux/personality.h>
14 #include <linux/tty.h>
15 #include <linux/iocontext.h>
16 #include <linux/key.h>
17 #include <linux/security.h>
18 #include <linux/cpu.h>
19 #include <linux/acct.h>
20 #include <linux/tsacct_kern.h>
21 #include <linux/file.h>
22 #include <linux/fdtable.h>
23 #include <linux/freezer.h>
24 #include <linux/binfmts.h>
25 #include <linux/nsproxy.h>
26 #include <linux/pid_namespace.h>
27 #include <linux/ptrace.h>
28 #include <linux/profile.h>
29 #include <linux/mount.h>
30 #include <linux/proc_fs.h>
31 #include <linux/kthread.h>
32 #include <linux/mempolicy.h>
33 #include <linux/taskstats_kern.h>
34 #include <linux/delayacct.h>
35 #include <linux/cgroup.h>
36 #include <linux/syscalls.h>
37 #include <linux/signal.h>
38 #include <linux/posix-timers.h>
39 #include <linux/cn_proc.h>
40 #include <linux/mutex.h>
41 #include <linux/futex.h>
42 #include <linux/pipe_fs_i.h>
43 #include <linux/audit.h> /* for audit_free() */
44 #include <linux/resource.h>
45 #include <linux/blkdev.h>
46 #include <linux/task_io_accounting_ops.h>
47 #include <linux/tracehook.h>
48 #include <linux/fs_struct.h>
49 #include <linux/init_task.h>
50 #include <linux/perf_event.h>
51 #include <trace/events/sched.h>
52 #include <linux/hw_breakpoint.h>
53 #include <linux/oom.h>
54 #include <linux/writeback.h>
55 #include <linux/shm.h>
56 
57 #include <asm/uaccess.h>
58 #include <asm/unistd.h>
59 #include <asm/pgtable.h>
60 #include <asm/mmu_context.h>
61 
62 static void exit_mm(struct task_struct * tsk);
63 
64 static void __unhash_process(struct task_struct *p, bool group_dead)
65 {
66 	nr_threads--;
67 	detach_pid(p, PIDTYPE_PID);
68 	if (group_dead) {
69 		detach_pid(p, PIDTYPE_PGID);
70 		detach_pid(p, PIDTYPE_SID);
71 
72 		list_del_rcu(&p->tasks);
73 		list_del_init(&p->sibling);
74 		__this_cpu_dec(process_counts);
75 	}
76 	list_del_rcu(&p->thread_group);
77 	list_del_rcu(&p->thread_node);
78 }
79 
80 /*
81  * This function expects the tasklist_lock write-locked.
82  */
83 static void __exit_signal(struct task_struct *tsk)
84 {
85 	struct signal_struct *sig = tsk->signal;
86 	bool group_dead = thread_group_leader(tsk);
87 	struct sighand_struct *sighand;
88 	struct tty_struct *uninitialized_var(tty);
89 	cputime_t utime, stime;
90 
91 	sighand = rcu_dereference_check(tsk->sighand,
92 					lockdep_tasklist_lock_is_held());
93 	spin_lock(&sighand->siglock);
94 
95 	posix_cpu_timers_exit(tsk);
96 	if (group_dead) {
97 		posix_cpu_timers_exit_group(tsk);
98 		tty = sig->tty;
99 		sig->tty = NULL;
100 	} else {
101 		/*
102 		 * This can only happen if the caller is de_thread().
103 		 * FIXME: this is the temporary hack, we should teach
104 		 * posix-cpu-timers to handle this case correctly.
105 		 */
106 		if (unlikely(has_group_leader_pid(tsk)))
107 			posix_cpu_timers_exit_group(tsk);
108 
109 		/*
110 		 * If there is any task waiting for the group exit
111 		 * then notify it:
112 		 */
113 		if (sig->notify_count > 0 && !--sig->notify_count)
114 			wake_up_process(sig->group_exit_task);
115 
116 		if (tsk == sig->curr_target)
117 			sig->curr_target = next_thread(tsk);
118 		/*
119 		 * Accumulate here the counters for all threads but the
120 		 * group leader as they die, so they can be added into
121 		 * the process-wide totals when those are taken.
122 		 * The group leader stays around as a zombie as long
123 		 * as there are other threads.  When it gets reaped,
124 		 * the exit.c code will add its counts into these totals.
125 		 * We won't ever get here for the group leader, since it
126 		 * will have been the last reference on the signal_struct.
127 		 */
128 		task_cputime(tsk, &utime, &stime);
129 		sig->utime += utime;
130 		sig->stime += stime;
131 		sig->gtime += task_gtime(tsk);
132 		sig->min_flt += tsk->min_flt;
133 		sig->maj_flt += tsk->maj_flt;
134 		sig->nvcsw += tsk->nvcsw;
135 		sig->nivcsw += tsk->nivcsw;
136 		sig->inblock += task_io_get_inblock(tsk);
137 		sig->oublock += task_io_get_oublock(tsk);
138 		task_io_accounting_add(&sig->ioac, &tsk->ioac);
139 		sig->sum_sched_runtime += tsk->se.sum_exec_runtime;
140 	}
141 
142 	sig->nr_threads--;
143 	__unhash_process(tsk, group_dead);
144 
145 	/*
146 	 * Do this under ->siglock, we can race with another thread
147 	 * doing sigqueue_free() if we have SIGQUEUE_PREALLOC signals.
148 	 */
149 	flush_sigqueue(&tsk->pending);
150 	tsk->sighand = NULL;
151 	spin_unlock(&sighand->siglock);
152 
153 	__cleanup_sighand(sighand);
154 	clear_tsk_thread_flag(tsk,TIF_SIGPENDING);
155 	if (group_dead) {
156 		flush_sigqueue(&sig->shared_pending);
157 		tty_kref_put(tty);
158 	}
159 }
160 
161 static void delayed_put_task_struct(struct rcu_head *rhp)
162 {
163 	struct task_struct *tsk = container_of(rhp, struct task_struct, rcu);
164 
165 	perf_event_delayed_put(tsk);
166 	trace_sched_process_free(tsk);
167 	put_task_struct(tsk);
168 }
169 
170 
171 void release_task(struct task_struct * p)
172 {
173 	struct task_struct *leader;
174 	int zap_leader;
175 repeat:
176 	/* don't need to get the RCU readlock here - the process is dead and
177 	 * can't be modifying its own credentials. But shut RCU-lockdep up */
178 	rcu_read_lock();
179 	atomic_dec(&__task_cred(p)->user->processes);
180 	rcu_read_unlock();
181 
182 	proc_flush_task(p);
183 
184 	write_lock_irq(&tasklist_lock);
185 	ptrace_release_task(p);
186 	__exit_signal(p);
187 
188 	/*
189 	 * If we are the last non-leader member of the thread
190 	 * group, and the leader is zombie, then notify the
191 	 * group leader's parent process. (if it wants notification.)
192 	 */
193 	zap_leader = 0;
194 	leader = p->group_leader;
195 	if (leader != p && thread_group_empty(leader) && leader->exit_state == EXIT_ZOMBIE) {
196 		/*
197 		 * If we were the last child thread and the leader has
198 		 * exited already, and the leader's parent ignores SIGCHLD,
199 		 * then we are the one who should release the leader.
200 		 */
201 		zap_leader = do_notify_parent(leader, leader->exit_signal);
202 		if (zap_leader)
203 			leader->exit_state = EXIT_DEAD;
204 	}
205 
206 	write_unlock_irq(&tasklist_lock);
207 	release_thread(p);
208 	call_rcu(&p->rcu, delayed_put_task_struct);
209 
210 	p = leader;
211 	if (unlikely(zap_leader))
212 		goto repeat;
213 }
214 
215 /*
216  * This checks not only the pgrp, but falls back on the pid if no
217  * satisfactory pgrp is found. I dunno - gdb doesn't work correctly
218  * without this...
219  *
220  * The caller must hold rcu lock or the tasklist lock.
221  */
222 struct pid *session_of_pgrp(struct pid *pgrp)
223 {
224 	struct task_struct *p;
225 	struct pid *sid = NULL;
226 
227 	p = pid_task(pgrp, PIDTYPE_PGID);
228 	if (p == NULL)
229 		p = pid_task(pgrp, PIDTYPE_PID);
230 	if (p != NULL)
231 		sid = task_session(p);
232 
233 	return sid;
234 }
235 
236 /*
237  * Determine if a process group is "orphaned", according to the POSIX
238  * definition in 2.2.2.52.  Orphaned process groups are not to be affected
239  * by terminal-generated stop signals.  Newly orphaned process groups are
240  * to receive a SIGHUP and a SIGCONT.
241  *
242  * "I ask you, have you ever known what it is to be an orphan?"
243  */
244 static int will_become_orphaned_pgrp(struct pid *pgrp, struct task_struct *ignored_task)
245 {
246 	struct task_struct *p;
247 
248 	do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
249 		if ((p == ignored_task) ||
250 		    (p->exit_state && thread_group_empty(p)) ||
251 		    is_global_init(p->real_parent))
252 			continue;
253 
254 		if (task_pgrp(p->real_parent) != pgrp &&
255 		    task_session(p->real_parent) == task_session(p))
256 			return 0;
257 	} while_each_pid_task(pgrp, PIDTYPE_PGID, p);
258 
259 	return 1;
260 }
261 
262 int is_current_pgrp_orphaned(void)
263 {
264 	int retval;
265 
266 	read_lock(&tasklist_lock);
267 	retval = will_become_orphaned_pgrp(task_pgrp(current), NULL);
268 	read_unlock(&tasklist_lock);
269 
270 	return retval;
271 }
272 
273 static bool has_stopped_jobs(struct pid *pgrp)
274 {
275 	struct task_struct *p;
276 
277 	do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
278 		if (p->signal->flags & SIGNAL_STOP_STOPPED)
279 			return true;
280 	} while_each_pid_task(pgrp, PIDTYPE_PGID, p);
281 
282 	return false;
283 }
284 
285 /*
286  * Check to see if any process groups have become orphaned as
287  * a result of our exiting, and if they have any stopped jobs,
288  * send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
289  */
290 static void
291 kill_orphaned_pgrp(struct task_struct *tsk, struct task_struct *parent)
292 {
293 	struct pid *pgrp = task_pgrp(tsk);
294 	struct task_struct *ignored_task = tsk;
295 
296 	if (!parent)
297 		 /* exit: our father is in a different pgrp than
298 		  * we are and we were the only connection outside.
299 		  */
300 		parent = tsk->real_parent;
301 	else
302 		/* reparent: our child is in a different pgrp than
303 		 * we are, and it was the only connection outside.
304 		 */
305 		ignored_task = NULL;
306 
307 	if (task_pgrp(parent) != pgrp &&
308 	    task_session(parent) == task_session(tsk) &&
309 	    will_become_orphaned_pgrp(pgrp, ignored_task) &&
310 	    has_stopped_jobs(pgrp)) {
311 		__kill_pgrp_info(SIGHUP, SEND_SIG_PRIV, pgrp);
312 		__kill_pgrp_info(SIGCONT, SEND_SIG_PRIV, pgrp);
313 	}
314 }
315 
316 /*
317  * Let kernel threads use this to say that they allow a certain signal.
318  * Must not be used if kthread was cloned with CLONE_SIGHAND.
319  */
320 int allow_signal(int sig)
321 {
322 	if (!valid_signal(sig) || sig < 1)
323 		return -EINVAL;
324 
325 	spin_lock_irq(&current->sighand->siglock);
326 	/* This is only needed for daemonize()'ed kthreads */
327 	sigdelset(&current->blocked, sig);
328 	/*
329 	 * Kernel threads handle their own signals. Let the signal code
330 	 * know it'll be handled, so that they don't get converted to
331 	 * SIGKILL or just silently dropped.
332 	 */
333 	current->sighand->action[(sig)-1].sa.sa_handler = (void __user *)2;
334 	recalc_sigpending();
335 	spin_unlock_irq(&current->sighand->siglock);
336 	return 0;
337 }
338 
339 EXPORT_SYMBOL(allow_signal);
340 
341 int disallow_signal(int sig)
342 {
343 	if (!valid_signal(sig) || sig < 1)
344 		return -EINVAL;
345 
346 	spin_lock_irq(&current->sighand->siglock);
347 	current->sighand->action[(sig)-1].sa.sa_handler = SIG_IGN;
348 	recalc_sigpending();
349 	spin_unlock_irq(&current->sighand->siglock);
350 	return 0;
351 }
352 
353 EXPORT_SYMBOL(disallow_signal);
354 
355 #ifdef CONFIG_MM_OWNER
356 /*
357  * A task is exiting.   If it owned this mm, find a new owner for the mm.
358  */
359 void mm_update_next_owner(struct mm_struct *mm)
360 {
361 	struct task_struct *c, *g, *p = current;
362 
363 retry:
364 	/*
365 	 * If the exiting or execing task is not the owner, it's
366 	 * someone else's problem.
367 	 */
368 	if (mm->owner != p)
369 		return;
370 	/*
371 	 * The current owner is exiting/execing and there are no other
372 	 * candidates.  Do not leave the mm pointing to a possibly
373 	 * freed task structure.
374 	 */
375 	if (atomic_read(&mm->mm_users) <= 1) {
376 		mm->owner = NULL;
377 		return;
378 	}
379 
380 	read_lock(&tasklist_lock);
381 	/*
382 	 * Search in the children
383 	 */
384 	list_for_each_entry(c, &p->children, sibling) {
385 		if (c->mm == mm)
386 			goto assign_new_owner;
387 	}
388 
389 	/*
390 	 * Search in the siblings
391 	 */
392 	list_for_each_entry(c, &p->real_parent->children, sibling) {
393 		if (c->mm == mm)
394 			goto assign_new_owner;
395 	}
396 
397 	/*
398 	 * Search through everything else. We should not get
399 	 * here often
400 	 */
401 	do_each_thread(g, c) {
402 		if (c->mm == mm)
403 			goto assign_new_owner;
404 	} while_each_thread(g, c);
405 
406 	read_unlock(&tasklist_lock);
407 	/*
408 	 * We found no owner yet mm_users > 1: this implies that we are
409 	 * most likely racing with swapoff (try_to_unuse()) or /proc or
410 	 * ptrace or page migration (get_task_mm()).  Mark owner as NULL.
411 	 */
412 	mm->owner = NULL;
413 	return;
414 
415 assign_new_owner:
416 	BUG_ON(c == p);
417 	get_task_struct(c);
418 	/*
419 	 * The task_lock protects c->mm from changing.
420 	 * We always want mm->owner->mm == mm
421 	 */
422 	task_lock(c);
423 	/*
424 	 * Delay read_unlock() till we have the task_lock()
425 	 * to ensure that c does not slip away underneath us
426 	 */
427 	read_unlock(&tasklist_lock);
428 	if (c->mm != mm) {
429 		task_unlock(c);
430 		put_task_struct(c);
431 		goto retry;
432 	}
433 	mm->owner = c;
434 	task_unlock(c);
435 	put_task_struct(c);
436 }
437 #endif /* CONFIG_MM_OWNER */
438 
439 /*
440  * Turn us into a lazy TLB process if we
441  * aren't already..
442  */
443 static void exit_mm(struct task_struct * tsk)
444 {
445 	struct mm_struct *mm = tsk->mm;
446 	struct core_state *core_state;
447 
448 	mm_release(tsk, mm);
449 	if (!mm)
450 		return;
451 	sync_mm_rss(mm);
452 	/*
453 	 * Serialize with any possible pending coredump.
454 	 * We must hold mmap_sem around checking core_state
455 	 * and clearing tsk->mm.  The core-inducing thread
456 	 * will increment ->nr_threads for each thread in the
457 	 * group with ->mm != NULL.
458 	 */
459 	down_read(&mm->mmap_sem);
460 	core_state = mm->core_state;
461 	if (core_state) {
462 		struct core_thread self;
463 		up_read(&mm->mmap_sem);
464 
465 		self.task = tsk;
466 		self.next = xchg(&core_state->dumper.next, &self);
467 		/*
468 		 * Implies mb(), the result of xchg() must be visible
469 		 * to core_state->dumper.
470 		 */
471 		if (atomic_dec_and_test(&core_state->nr_threads))
472 			complete(&core_state->startup);
473 
474 		for (;;) {
475 			set_task_state(tsk, TASK_UNINTERRUPTIBLE);
476 			if (!self.task) /* see coredump_finish() */
477 				break;
478 			freezable_schedule();
479 		}
480 		__set_task_state(tsk, TASK_RUNNING);
481 		down_read(&mm->mmap_sem);
482 	}
483 	atomic_inc(&mm->mm_count);
484 	BUG_ON(mm != tsk->active_mm);
485 	/* more a memory barrier than a real lock */
486 	task_lock(tsk);
487 	tsk->mm = NULL;
488 	up_read(&mm->mmap_sem);
489 	enter_lazy_tlb(mm, current);
490 	task_unlock(tsk);
491 	mm_update_next_owner(mm);
492 	mmput(mm);
493 }
494 
495 /*
496  * When we die, we re-parent all our children, and try to:
497  * 1. give them to another thread in our thread group, if such a member exists
498  * 2. give it to the first ancestor process which prctl'd itself as a
499  *    child_subreaper for its children (like a service manager)
500  * 3. give it to the init process (PID 1) in our pid namespace
501  */
502 static struct task_struct *find_new_reaper(struct task_struct *father)
503 	__releases(&tasklist_lock)
504 	__acquires(&tasklist_lock)
505 {
506 	struct pid_namespace *pid_ns = task_active_pid_ns(father);
507 	struct task_struct *thread;
508 
509 	thread = father;
510 	while_each_thread(father, thread) {
511 		if (thread->flags & PF_EXITING)
512 			continue;
513 		if (unlikely(pid_ns->child_reaper == father))
514 			pid_ns->child_reaper = thread;
515 		return thread;
516 	}
517 
518 	if (unlikely(pid_ns->child_reaper == father)) {
519 		write_unlock_irq(&tasklist_lock);
520 		if (unlikely(pid_ns == &init_pid_ns)) {
521 			panic("Attempted to kill init! exitcode=0x%08x\n",
522 				father->signal->group_exit_code ?:
523 					father->exit_code);
524 		}
525 
526 		zap_pid_ns_processes(pid_ns);
527 		write_lock_irq(&tasklist_lock);
528 	} else if (father->signal->has_child_subreaper) {
529 		struct task_struct *reaper;
530 
531 		/*
532 		 * Find the first ancestor marked as child_subreaper.
533 		 * Note that the code below checks same_thread_group(reaper,
534 		 * pid_ns->child_reaper).  This is what we need to DTRT in a
535 		 * PID namespace. However we still need the check above, see
536 		 * http://marc.info/?l=linux-kernel&m=131385460420380
537 		 */
538 		for (reaper = father->real_parent;
539 		     reaper != &init_task;
540 		     reaper = reaper->real_parent) {
541 			if (same_thread_group(reaper, pid_ns->child_reaper))
542 				break;
543 			if (!reaper->signal->is_child_subreaper)
544 				continue;
545 			thread = reaper;
546 			do {
547 				if (!(thread->flags & PF_EXITING))
548 					return reaper;
549 			} while_each_thread(reaper, thread);
550 		}
551 	}
552 
553 	return pid_ns->child_reaper;
554 }
555 
556 /*
557 * Any that need to be release_task'd are put on the @dead list.
558  */
559 static void reparent_leader(struct task_struct *father, struct task_struct *p,
560 				struct list_head *dead)
561 {
562 	list_move_tail(&p->sibling, &p->real_parent->children);
563 
564 	if (p->exit_state == EXIT_DEAD)
565 		return;
566 	/*
567 	 * If this is a threaded reparent there is no need to
568 	 * notify anyone anything has happened.
569 	 */
570 	if (same_thread_group(p->real_parent, father))
571 		return;
572 
573 	/* We don't want people slaying init.  */
574 	p->exit_signal = SIGCHLD;
575 
576 	/* If it has exited notify the new parent about this child's death. */
577 	if (!p->ptrace &&
578 	    p->exit_state == EXIT_ZOMBIE && thread_group_empty(p)) {
579 		if (do_notify_parent(p, p->exit_signal)) {
580 			p->exit_state = EXIT_DEAD;
581 			list_move_tail(&p->sibling, dead);
582 		}
583 	}
584 
585 	kill_orphaned_pgrp(p, father);
586 }
587 
588 static void forget_original_parent(struct task_struct *father)
589 {
590 	struct task_struct *p, *n, *reaper;
591 	LIST_HEAD(dead_children);
592 
593 	write_lock_irq(&tasklist_lock);
594 	/*
595 	 * Note that exit_ptrace() and find_new_reaper() might
596 	 * drop tasklist_lock and reacquire it.
597 	 */
598 	exit_ptrace(father);
599 	reaper = find_new_reaper(father);
600 
601 	list_for_each_entry_safe(p, n, &father->children, sibling) {
602 		struct task_struct *t = p;
603 		do {
604 			t->real_parent = reaper;
605 			if (t->parent == father) {
606 				BUG_ON(t->ptrace);
607 				t->parent = t->real_parent;
608 			}
609 			if (t->pdeath_signal)
610 				group_send_sig_info(t->pdeath_signal,
611 						    SEND_SIG_NOINFO, t);
612 		} while_each_thread(p, t);
613 		reparent_leader(father, p, &dead_children);
614 	}
615 	write_unlock_irq(&tasklist_lock);
616 
617 	BUG_ON(!list_empty(&father->children));
618 
619 	list_for_each_entry_safe(p, n, &dead_children, sibling) {
620 		list_del_init(&p->sibling);
621 		release_task(p);
622 	}
623 }
624 
625 /*
626  * Send signals to all our closest relatives so that they know
627  * to properly mourn us..
628  */
629 static void exit_notify(struct task_struct *tsk, int group_dead)
630 {
631 	bool autoreap;
632 
633 	/*
634 	 * This does two things:
635 	 *
636   	 * A.  Make init inherit all the child processes
637 	 * B.  Check to see if any process groups have become orphaned
638 	 *	as a result of our exiting, and if they have any stopped
639 	 *	jobs, send them a SIGHUP and then a SIGCONT.  (POSIX 3.2.2.2)
640 	 */
641 	forget_original_parent(tsk);
642 
643 	write_lock_irq(&tasklist_lock);
644 	if (group_dead)
645 		kill_orphaned_pgrp(tsk->group_leader, NULL);
646 
647 	if (unlikely(tsk->ptrace)) {
648 		int sig = thread_group_leader(tsk) &&
649 				thread_group_empty(tsk) &&
650 				!ptrace_reparented(tsk) ?
651 			tsk->exit_signal : SIGCHLD;
652 		autoreap = do_notify_parent(tsk, sig);
653 	} else if (thread_group_leader(tsk)) {
654 		autoreap = thread_group_empty(tsk) &&
655 			do_notify_parent(tsk, tsk->exit_signal);
656 	} else {
657 		autoreap = true;
658 	}
659 
660 	tsk->exit_state = autoreap ? EXIT_DEAD : EXIT_ZOMBIE;
661 
662 	/* mt-exec, de_thread() is waiting for group leader */
663 	if (unlikely(tsk->signal->notify_count < 0))
664 		wake_up_process(tsk->signal->group_exit_task);
665 	write_unlock_irq(&tasklist_lock);
666 
667 	/* If the process is dead, release it - nobody will wait for it */
668 	if (autoreap)
669 		release_task(tsk);
670 }
671 
672 #ifdef CONFIG_DEBUG_STACK_USAGE
673 static void check_stack_usage(void)
674 {
675 	static DEFINE_SPINLOCK(low_water_lock);
676 	static int lowest_to_date = THREAD_SIZE;
677 	unsigned long free;
678 
679 	free = stack_not_used(current);
680 
681 	if (free >= lowest_to_date)
682 		return;
683 
684 	spin_lock(&low_water_lock);
685 	if (free < lowest_to_date) {
686 		printk(KERN_WARNING "%s (%d) used greatest stack depth: "
687 				"%lu bytes left\n",
688 				current->comm, task_pid_nr(current), free);
689 		lowest_to_date = free;
690 	}
691 	spin_unlock(&low_water_lock);
692 }
693 #else
694 static inline void check_stack_usage(void) {}
695 #endif
696 
697 void do_exit(long code)
698 {
699 	struct task_struct *tsk = current;
700 	int group_dead;
701 
702 	profile_task_exit(tsk);
703 
704 	WARN_ON(blk_needs_flush_plug(tsk));
705 
706 	if (unlikely(in_interrupt()))
707 		panic("Aiee, killing interrupt handler!");
708 	if (unlikely(!tsk->pid))
709 		panic("Attempted to kill the idle task!");
710 
711 	/*
712 	 * If do_exit is called because this processes oopsed, it's possible
713 	 * that get_fs() was left as KERNEL_DS, so reset it to USER_DS before
714 	 * continuing. Amongst other possible reasons, this is to prevent
715 	 * mm_release()->clear_child_tid() from writing to a user-controlled
716 	 * kernel address.
717 	 */
718 	set_fs(USER_DS);
719 
720 	ptrace_event(PTRACE_EVENT_EXIT, code);
721 
722 	validate_creds_for_do_exit(tsk);
723 
724 	/*
725 	 * We're taking recursive faults here in do_exit. Safest is to just
726 	 * leave this task alone and wait for reboot.
727 	 */
728 	if (unlikely(tsk->flags & PF_EXITING)) {
729 		printk(KERN_ALERT
730 			"Fixing recursive fault but reboot is needed!\n");
731 		/*
732 		 * We can do this unlocked here. The futex code uses
733 		 * this flag just to verify whether the pi state
734 		 * cleanup has been done or not. In the worst case it
735 		 * loops once more. We pretend that the cleanup was
736 		 * done as there is no way to return. Either the
737 		 * OWNER_DIED bit is set by now or we push the blocked
738 		 * task into the wait for ever nirwana as well.
739 		 */
740 		tsk->flags |= PF_EXITPIDONE;
741 		set_current_state(TASK_UNINTERRUPTIBLE);
742 		schedule();
743 	}
744 
745 	exit_signals(tsk);  /* sets PF_EXITING */
746 	/*
747 	 * tsk->flags are checked in the futex code to protect against
748 	 * an exiting task cleaning up the robust pi futexes.
749 	 */
750 	smp_mb();
751 	raw_spin_unlock_wait(&tsk->pi_lock);
752 
753 	if (unlikely(in_atomic()))
754 		printk(KERN_INFO "note: %s[%d] exited with preempt_count %d\n",
755 				current->comm, task_pid_nr(current),
756 				preempt_count());
757 
758 	acct_update_integrals(tsk);
759 	/* sync mm's RSS info before statistics gathering */
760 	if (tsk->mm)
761 		sync_mm_rss(tsk->mm);
762 	group_dead = atomic_dec_and_test(&tsk->signal->live);
763 	if (group_dead) {
764 		hrtimer_cancel(&tsk->signal->real_timer);
765 		exit_itimers(tsk->signal);
766 		if (tsk->mm)
767 			setmax_mm_hiwater_rss(&tsk->signal->maxrss, tsk->mm);
768 	}
769 	acct_collect(code, group_dead);
770 	if (group_dead)
771 		tty_audit_exit();
772 	audit_free(tsk);
773 
774 	tsk->exit_code = code;
775 	taskstats_exit(tsk, group_dead);
776 
777 	exit_mm(tsk);
778 
779 	if (group_dead)
780 		acct_process();
781 	trace_sched_process_exit(tsk);
782 
783 	exit_sem(tsk);
784 	exit_shm(tsk);
785 	exit_files(tsk);
786 	exit_fs(tsk);
787 	exit_task_namespaces(tsk);
788 	exit_task_work(tsk);
789 	check_stack_usage();
790 	exit_thread();
791 
792 	/*
793 	 * Flush inherited counters to the parent - before the parent
794 	 * gets woken up by child-exit notifications.
795 	 *
796 	 * because of cgroup mode, must be called before cgroup_exit()
797 	 */
798 	perf_event_exit_task(tsk);
799 
800 	cgroup_exit(tsk, 1);
801 
802 	if (group_dead)
803 		disassociate_ctty(1);
804 
805 	module_put(task_thread_info(tsk)->exec_domain->module);
806 
807 	proc_exit_connector(tsk);
808 
809 	/*
810 	 * FIXME: do that only when needed, using sched_exit tracepoint
811 	 */
812 	flush_ptrace_hw_breakpoint(tsk);
813 
814 	exit_notify(tsk, group_dead);
815 #ifdef CONFIG_NUMA
816 	task_lock(tsk);
817 	mpol_put(tsk->mempolicy);
818 	tsk->mempolicy = NULL;
819 	task_unlock(tsk);
820 #endif
821 #ifdef CONFIG_FUTEX
822 	if (unlikely(current->pi_state_cache))
823 		kfree(current->pi_state_cache);
824 #endif
825 	/*
826 	 * Make sure we are holding no locks:
827 	 */
828 	debug_check_no_locks_held();
829 	/*
830 	 * We can do this unlocked here. The futex code uses this flag
831 	 * just to verify whether the pi state cleanup has been done
832 	 * or not. In the worst case it loops once more.
833 	 */
834 	tsk->flags |= PF_EXITPIDONE;
835 
836 	if (tsk->io_context)
837 		exit_io_context(tsk);
838 
839 	if (tsk->splice_pipe)
840 		free_pipe_info(tsk->splice_pipe);
841 
842 	if (tsk->task_frag.page)
843 		put_page(tsk->task_frag.page);
844 
845 	validate_creds_for_do_exit(tsk);
846 
847 	preempt_disable();
848 	if (tsk->nr_dirtied)
849 		__this_cpu_add(dirty_throttle_leaks, tsk->nr_dirtied);
850 	exit_rcu();
851 
852 	/*
853 	 * The setting of TASK_RUNNING by try_to_wake_up() may be delayed
854 	 * when the following two conditions become true.
855 	 *   - There is race condition of mmap_sem (It is acquired by
856 	 *     exit_mm()), and
857 	 *   - SMI occurs before setting TASK_RUNINNG.
858 	 *     (or hypervisor of virtual machine switches to other guest)
859 	 *  As a result, we may become TASK_RUNNING after becoming TASK_DEAD
860 	 *
861 	 * To avoid it, we have to wait for releasing tsk->pi_lock which
862 	 * is held by try_to_wake_up()
863 	 */
864 	smp_mb();
865 	raw_spin_unlock_wait(&tsk->pi_lock);
866 
867 	/* causes final put_task_struct in finish_task_switch(). */
868 	tsk->state = TASK_DEAD;
869 	tsk->flags |= PF_NOFREEZE;	/* tell freezer to ignore us */
870 	schedule();
871 	BUG();
872 	/* Avoid "noreturn function does return".  */
873 	for (;;)
874 		cpu_relax();	/* For when BUG is null */
875 }
876 
877 EXPORT_SYMBOL_GPL(do_exit);
878 
879 void complete_and_exit(struct completion *comp, long code)
880 {
881 	if (comp)
882 		complete(comp);
883 
884 	do_exit(code);
885 }
886 
887 EXPORT_SYMBOL(complete_and_exit);
888 
889 SYSCALL_DEFINE1(exit, int, error_code)
890 {
891 	do_exit((error_code&0xff)<<8);
892 }
893 
894 /*
895  * Take down every thread in the group.  This is called by fatal signals
896  * as well as by sys_exit_group (below).
897  */
898 void
899 do_group_exit(int exit_code)
900 {
901 	struct signal_struct *sig = current->signal;
902 
903 	BUG_ON(exit_code & 0x80); /* core dumps don't get here */
904 
905 	if (signal_group_exit(sig))
906 		exit_code = sig->group_exit_code;
907 	else if (!thread_group_empty(current)) {
908 		struct sighand_struct *const sighand = current->sighand;
909 		spin_lock_irq(&sighand->siglock);
910 		if (signal_group_exit(sig))
911 			/* Another thread got here before we took the lock.  */
912 			exit_code = sig->group_exit_code;
913 		else {
914 			sig->group_exit_code = exit_code;
915 			sig->flags = SIGNAL_GROUP_EXIT;
916 			zap_other_threads(current);
917 		}
918 		spin_unlock_irq(&sighand->siglock);
919 	}
920 
921 	do_exit(exit_code);
922 	/* NOTREACHED */
923 }
924 
925 /*
926  * this kills every thread in the thread group. Note that any externally
927  * wait4()-ing process will get the correct exit code - even if this
928  * thread is not the thread group leader.
929  */
930 SYSCALL_DEFINE1(exit_group, int, error_code)
931 {
932 	do_group_exit((error_code & 0xff) << 8);
933 	/* NOTREACHED */
934 	return 0;
935 }
936 
937 struct wait_opts {
938 	enum pid_type		wo_type;
939 	int			wo_flags;
940 	struct pid		*wo_pid;
941 
942 	struct siginfo __user	*wo_info;
943 	int __user		*wo_stat;
944 	struct rusage __user	*wo_rusage;
945 
946 	wait_queue_t		child_wait;
947 	int			notask_error;
948 };
949 
950 static inline
951 struct pid *task_pid_type(struct task_struct *task, enum pid_type type)
952 {
953 	if (type != PIDTYPE_PID)
954 		task = task->group_leader;
955 	return task->pids[type].pid;
956 }
957 
958 static int eligible_pid(struct wait_opts *wo, struct task_struct *p)
959 {
960 	return	wo->wo_type == PIDTYPE_MAX ||
961 		task_pid_type(p, wo->wo_type) == wo->wo_pid;
962 }
963 
964 static int eligible_child(struct wait_opts *wo, struct task_struct *p)
965 {
966 	if (!eligible_pid(wo, p))
967 		return 0;
968 	/* Wait for all children (clone and not) if __WALL is set;
969 	 * otherwise, wait for clone children *only* if __WCLONE is
970 	 * set; otherwise, wait for non-clone children *only*.  (Note:
971 	 * A "clone" child here is one that reports to its parent
972 	 * using a signal other than SIGCHLD.) */
973 	if (((p->exit_signal != SIGCHLD) ^ !!(wo->wo_flags & __WCLONE))
974 	    && !(wo->wo_flags & __WALL))
975 		return 0;
976 
977 	return 1;
978 }
979 
980 static int wait_noreap_copyout(struct wait_opts *wo, struct task_struct *p,
981 				pid_t pid, uid_t uid, int why, int status)
982 {
983 	struct siginfo __user *infop;
984 	int retval = wo->wo_rusage
985 		? getrusage(p, RUSAGE_BOTH, wo->wo_rusage) : 0;
986 
987 	put_task_struct(p);
988 	infop = wo->wo_info;
989 	if (infop) {
990 		if (!retval)
991 			retval = put_user(SIGCHLD, &infop->si_signo);
992 		if (!retval)
993 			retval = put_user(0, &infop->si_errno);
994 		if (!retval)
995 			retval = put_user((short)why, &infop->si_code);
996 		if (!retval)
997 			retval = put_user(pid, &infop->si_pid);
998 		if (!retval)
999 			retval = put_user(uid, &infop->si_uid);
1000 		if (!retval)
1001 			retval = put_user(status, &infop->si_status);
1002 	}
1003 	if (!retval)
1004 		retval = pid;
1005 	return retval;
1006 }
1007 
1008 /*
1009  * Handle sys_wait4 work for one task in state EXIT_ZOMBIE.  We hold
1010  * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
1011  * the lock and this task is uninteresting.  If we return nonzero, we have
1012  * released the lock and the system call should return.
1013  */
1014 static int wait_task_zombie(struct wait_opts *wo, struct task_struct *p)
1015 {
1016 	unsigned long state;
1017 	int retval, status, traced;
1018 	pid_t pid = task_pid_vnr(p);
1019 	uid_t uid = from_kuid_munged(current_user_ns(), task_uid(p));
1020 	struct siginfo __user *infop;
1021 
1022 	if (!likely(wo->wo_flags & WEXITED))
1023 		return 0;
1024 
1025 	if (unlikely(wo->wo_flags & WNOWAIT)) {
1026 		int exit_code = p->exit_code;
1027 		int why;
1028 
1029 		get_task_struct(p);
1030 		read_unlock(&tasklist_lock);
1031 		if ((exit_code & 0x7f) == 0) {
1032 			why = CLD_EXITED;
1033 			status = exit_code >> 8;
1034 		} else {
1035 			why = (exit_code & 0x80) ? CLD_DUMPED : CLD_KILLED;
1036 			status = exit_code & 0x7f;
1037 		}
1038 		return wait_noreap_copyout(wo, p, pid, uid, why, status);
1039 	}
1040 
1041 	/*
1042 	 * Try to move the task's state to DEAD
1043 	 * only one thread is allowed to do this:
1044 	 */
1045 	state = xchg(&p->exit_state, EXIT_DEAD);
1046 	if (state != EXIT_ZOMBIE) {
1047 		BUG_ON(state != EXIT_DEAD);
1048 		return 0;
1049 	}
1050 
1051 	traced = ptrace_reparented(p);
1052 	/*
1053 	 * It can be ptraced but not reparented, check
1054 	 * thread_group_leader() to filter out sub-threads.
1055 	 */
1056 	if (likely(!traced) && thread_group_leader(p)) {
1057 		struct signal_struct *psig;
1058 		struct signal_struct *sig;
1059 		unsigned long maxrss;
1060 		cputime_t tgutime, tgstime;
1061 
1062 		/*
1063 		 * The resource counters for the group leader are in its
1064 		 * own task_struct.  Those for dead threads in the group
1065 		 * are in its signal_struct, as are those for the child
1066 		 * processes it has previously reaped.  All these
1067 		 * accumulate in the parent's signal_struct c* fields.
1068 		 *
1069 		 * We don't bother to take a lock here to protect these
1070 		 * p->signal fields, because they are only touched by
1071 		 * __exit_signal, which runs with tasklist_lock
1072 		 * write-locked anyway, and so is excluded here.  We do
1073 		 * need to protect the access to parent->signal fields,
1074 		 * as other threads in the parent group can be right
1075 		 * here reaping other children at the same time.
1076 		 *
1077 		 * We use thread_group_cputime_adjusted() to get times for the thread
1078 		 * group, which consolidates times for all threads in the
1079 		 * group including the group leader.
1080 		 */
1081 		thread_group_cputime_adjusted(p, &tgutime, &tgstime);
1082 		spin_lock_irq(&p->real_parent->sighand->siglock);
1083 		psig = p->real_parent->signal;
1084 		sig = p->signal;
1085 		psig->cutime += tgutime + sig->cutime;
1086 		psig->cstime += tgstime + sig->cstime;
1087 		psig->cgtime += task_gtime(p) + sig->gtime + sig->cgtime;
1088 		psig->cmin_flt +=
1089 			p->min_flt + sig->min_flt + sig->cmin_flt;
1090 		psig->cmaj_flt +=
1091 			p->maj_flt + sig->maj_flt + sig->cmaj_flt;
1092 		psig->cnvcsw +=
1093 			p->nvcsw + sig->nvcsw + sig->cnvcsw;
1094 		psig->cnivcsw +=
1095 			p->nivcsw + sig->nivcsw + sig->cnivcsw;
1096 		psig->cinblock +=
1097 			task_io_get_inblock(p) +
1098 			sig->inblock + sig->cinblock;
1099 		psig->coublock +=
1100 			task_io_get_oublock(p) +
1101 			sig->oublock + sig->coublock;
1102 		maxrss = max(sig->maxrss, sig->cmaxrss);
1103 		if (psig->cmaxrss < maxrss)
1104 			psig->cmaxrss = maxrss;
1105 		task_io_accounting_add(&psig->ioac, &p->ioac);
1106 		task_io_accounting_add(&psig->ioac, &sig->ioac);
1107 		spin_unlock_irq(&p->real_parent->sighand->siglock);
1108 	}
1109 
1110 	/*
1111 	 * Now we are sure this task is interesting, and no other
1112 	 * thread can reap it because we set its state to EXIT_DEAD.
1113 	 */
1114 	read_unlock(&tasklist_lock);
1115 
1116 	retval = wo->wo_rusage
1117 		? getrusage(p, RUSAGE_BOTH, wo->wo_rusage) : 0;
1118 	status = (p->signal->flags & SIGNAL_GROUP_EXIT)
1119 		? p->signal->group_exit_code : p->exit_code;
1120 	if (!retval && wo->wo_stat)
1121 		retval = put_user(status, wo->wo_stat);
1122 
1123 	infop = wo->wo_info;
1124 	if (!retval && infop)
1125 		retval = put_user(SIGCHLD, &infop->si_signo);
1126 	if (!retval && infop)
1127 		retval = put_user(0, &infop->si_errno);
1128 	if (!retval && infop) {
1129 		int why;
1130 
1131 		if ((status & 0x7f) == 0) {
1132 			why = CLD_EXITED;
1133 			status >>= 8;
1134 		} else {
1135 			why = (status & 0x80) ? CLD_DUMPED : CLD_KILLED;
1136 			status &= 0x7f;
1137 		}
1138 		retval = put_user((short)why, &infop->si_code);
1139 		if (!retval)
1140 			retval = put_user(status, &infop->si_status);
1141 	}
1142 	if (!retval && infop)
1143 		retval = put_user(pid, &infop->si_pid);
1144 	if (!retval && infop)
1145 		retval = put_user(uid, &infop->si_uid);
1146 	if (!retval)
1147 		retval = pid;
1148 
1149 	if (traced) {
1150 		write_lock_irq(&tasklist_lock);
1151 		/* We dropped tasklist, ptracer could die and untrace */
1152 		ptrace_unlink(p);
1153 		/*
1154 		 * If this is not a sub-thread, notify the parent.
1155 		 * If parent wants a zombie, don't release it now.
1156 		 */
1157 		if (thread_group_leader(p) &&
1158 		    !do_notify_parent(p, p->exit_signal)) {
1159 			p->exit_state = EXIT_ZOMBIE;
1160 			p = NULL;
1161 		}
1162 		write_unlock_irq(&tasklist_lock);
1163 	}
1164 	if (p != NULL)
1165 		release_task(p);
1166 
1167 	return retval;
1168 }
1169 
1170 static int *task_stopped_code(struct task_struct *p, bool ptrace)
1171 {
1172 	if (ptrace) {
1173 		if (task_is_stopped_or_traced(p) &&
1174 		    !(p->jobctl & JOBCTL_LISTENING))
1175 			return &p->exit_code;
1176 	} else {
1177 		if (p->signal->flags & SIGNAL_STOP_STOPPED)
1178 			return &p->signal->group_exit_code;
1179 	}
1180 	return NULL;
1181 }
1182 
1183 /**
1184  * wait_task_stopped - Wait for %TASK_STOPPED or %TASK_TRACED
1185  * @wo: wait options
1186  * @ptrace: is the wait for ptrace
1187  * @p: task to wait for
1188  *
1189  * Handle sys_wait4() work for %p in state %TASK_STOPPED or %TASK_TRACED.
1190  *
1191  * CONTEXT:
1192  * read_lock(&tasklist_lock), which is released if return value is
1193  * non-zero.  Also, grabs and releases @p->sighand->siglock.
1194  *
1195  * RETURNS:
1196  * 0 if wait condition didn't exist and search for other wait conditions
1197  * should continue.  Non-zero return, -errno on failure and @p's pid on
1198  * success, implies that tasklist_lock is released and wait condition
1199  * search should terminate.
1200  */
1201 static int wait_task_stopped(struct wait_opts *wo,
1202 				int ptrace, struct task_struct *p)
1203 {
1204 	struct siginfo __user *infop;
1205 	int retval, exit_code, *p_code, why;
1206 	uid_t uid = 0; /* unneeded, required by compiler */
1207 	pid_t pid;
1208 
1209 	/*
1210 	 * Traditionally we see ptrace'd stopped tasks regardless of options.
1211 	 */
1212 	if (!ptrace && !(wo->wo_flags & WUNTRACED))
1213 		return 0;
1214 
1215 	if (!task_stopped_code(p, ptrace))
1216 		return 0;
1217 
1218 	exit_code = 0;
1219 	spin_lock_irq(&p->sighand->siglock);
1220 
1221 	p_code = task_stopped_code(p, ptrace);
1222 	if (unlikely(!p_code))
1223 		goto unlock_sig;
1224 
1225 	exit_code = *p_code;
1226 	if (!exit_code)
1227 		goto unlock_sig;
1228 
1229 	if (!unlikely(wo->wo_flags & WNOWAIT))
1230 		*p_code = 0;
1231 
1232 	uid = from_kuid_munged(current_user_ns(), task_uid(p));
1233 unlock_sig:
1234 	spin_unlock_irq(&p->sighand->siglock);
1235 	if (!exit_code)
1236 		return 0;
1237 
1238 	/*
1239 	 * Now we are pretty sure this task is interesting.
1240 	 * Make sure it doesn't get reaped out from under us while we
1241 	 * give up the lock and then examine it below.  We don't want to
1242 	 * keep holding onto the tasklist_lock while we call getrusage and
1243 	 * possibly take page faults for user memory.
1244 	 */
1245 	get_task_struct(p);
1246 	pid = task_pid_vnr(p);
1247 	why = ptrace ? CLD_TRAPPED : CLD_STOPPED;
1248 	read_unlock(&tasklist_lock);
1249 
1250 	if (unlikely(wo->wo_flags & WNOWAIT))
1251 		return wait_noreap_copyout(wo, p, pid, uid, why, exit_code);
1252 
1253 	retval = wo->wo_rusage
1254 		? getrusage(p, RUSAGE_BOTH, wo->wo_rusage) : 0;
1255 	if (!retval && wo->wo_stat)
1256 		retval = put_user((exit_code << 8) | 0x7f, wo->wo_stat);
1257 
1258 	infop = wo->wo_info;
1259 	if (!retval && infop)
1260 		retval = put_user(SIGCHLD, &infop->si_signo);
1261 	if (!retval && infop)
1262 		retval = put_user(0, &infop->si_errno);
1263 	if (!retval && infop)
1264 		retval = put_user((short)why, &infop->si_code);
1265 	if (!retval && infop)
1266 		retval = put_user(exit_code, &infop->si_status);
1267 	if (!retval && infop)
1268 		retval = put_user(pid, &infop->si_pid);
1269 	if (!retval && infop)
1270 		retval = put_user(uid, &infop->si_uid);
1271 	if (!retval)
1272 		retval = pid;
1273 	put_task_struct(p);
1274 
1275 	BUG_ON(!retval);
1276 	return retval;
1277 }
1278 
1279 /*
1280  * Handle do_wait work for one task in a live, non-stopped state.
1281  * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
1282  * the lock and this task is uninteresting.  If we return nonzero, we have
1283  * released the lock and the system call should return.
1284  */
1285 static int wait_task_continued(struct wait_opts *wo, struct task_struct *p)
1286 {
1287 	int retval;
1288 	pid_t pid;
1289 	uid_t uid;
1290 
1291 	if (!unlikely(wo->wo_flags & WCONTINUED))
1292 		return 0;
1293 
1294 	if (!(p->signal->flags & SIGNAL_STOP_CONTINUED))
1295 		return 0;
1296 
1297 	spin_lock_irq(&p->sighand->siglock);
1298 	/* Re-check with the lock held.  */
1299 	if (!(p->signal->flags & SIGNAL_STOP_CONTINUED)) {
1300 		spin_unlock_irq(&p->sighand->siglock);
1301 		return 0;
1302 	}
1303 	if (!unlikely(wo->wo_flags & WNOWAIT))
1304 		p->signal->flags &= ~SIGNAL_STOP_CONTINUED;
1305 	uid = from_kuid_munged(current_user_ns(), task_uid(p));
1306 	spin_unlock_irq(&p->sighand->siglock);
1307 
1308 	pid = task_pid_vnr(p);
1309 	get_task_struct(p);
1310 	read_unlock(&tasklist_lock);
1311 
1312 	if (!wo->wo_info) {
1313 		retval = wo->wo_rusage
1314 			? getrusage(p, RUSAGE_BOTH, wo->wo_rusage) : 0;
1315 		put_task_struct(p);
1316 		if (!retval && wo->wo_stat)
1317 			retval = put_user(0xffff, wo->wo_stat);
1318 		if (!retval)
1319 			retval = pid;
1320 	} else {
1321 		retval = wait_noreap_copyout(wo, p, pid, uid,
1322 					     CLD_CONTINUED, SIGCONT);
1323 		BUG_ON(retval == 0);
1324 	}
1325 
1326 	return retval;
1327 }
1328 
1329 /*
1330  * Consider @p for a wait by @parent.
1331  *
1332  * -ECHILD should be in ->notask_error before the first call.
1333  * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1334  * Returns zero if the search for a child should continue;
1335  * then ->notask_error is 0 if @p is an eligible child,
1336  * or another error from security_task_wait(), or still -ECHILD.
1337  */
1338 static int wait_consider_task(struct wait_opts *wo, int ptrace,
1339 				struct task_struct *p)
1340 {
1341 	int ret = eligible_child(wo, p);
1342 	if (!ret)
1343 		return ret;
1344 
1345 	ret = security_task_wait(p);
1346 	if (unlikely(ret < 0)) {
1347 		/*
1348 		 * If we have not yet seen any eligible child,
1349 		 * then let this error code replace -ECHILD.
1350 		 * A permission error will give the user a clue
1351 		 * to look for security policy problems, rather
1352 		 * than for mysterious wait bugs.
1353 		 */
1354 		if (wo->notask_error)
1355 			wo->notask_error = ret;
1356 		return 0;
1357 	}
1358 
1359 	/* dead body doesn't have much to contribute */
1360 	if (unlikely(p->exit_state == EXIT_DEAD)) {
1361 		/*
1362 		 * But do not ignore this task until the tracer does
1363 		 * wait_task_zombie()->do_notify_parent().
1364 		 */
1365 		if (likely(!ptrace) && unlikely(ptrace_reparented(p)))
1366 			wo->notask_error = 0;
1367 		return 0;
1368 	}
1369 
1370 	/* slay zombie? */
1371 	if (p->exit_state == EXIT_ZOMBIE) {
1372 		/*
1373 		 * A zombie ptracee is only visible to its ptracer.
1374 		 * Notification and reaping will be cascaded to the real
1375 		 * parent when the ptracer detaches.
1376 		 */
1377 		if (likely(!ptrace) && unlikely(p->ptrace)) {
1378 			/* it will become visible, clear notask_error */
1379 			wo->notask_error = 0;
1380 			return 0;
1381 		}
1382 
1383 		/* we don't reap group leaders with subthreads */
1384 		if (!delay_group_leader(p))
1385 			return wait_task_zombie(wo, p);
1386 
1387 		/*
1388 		 * Allow access to stopped/continued state via zombie by
1389 		 * falling through.  Clearing of notask_error is complex.
1390 		 *
1391 		 * When !@ptrace:
1392 		 *
1393 		 * If WEXITED is set, notask_error should naturally be
1394 		 * cleared.  If not, subset of WSTOPPED|WCONTINUED is set,
1395 		 * so, if there are live subthreads, there are events to
1396 		 * wait for.  If all subthreads are dead, it's still safe
1397 		 * to clear - this function will be called again in finite
1398 		 * amount time once all the subthreads are released and
1399 		 * will then return without clearing.
1400 		 *
1401 		 * When @ptrace:
1402 		 *
1403 		 * Stopped state is per-task and thus can't change once the
1404 		 * target task dies.  Only continued and exited can happen.
1405 		 * Clear notask_error if WCONTINUED | WEXITED.
1406 		 */
1407 		if (likely(!ptrace) || (wo->wo_flags & (WCONTINUED | WEXITED)))
1408 			wo->notask_error = 0;
1409 	} else {
1410 		/*
1411 		 * If @p is ptraced by a task in its real parent's group,
1412 		 * hide group stop/continued state when looking at @p as
1413 		 * the real parent; otherwise, a single stop can be
1414 		 * reported twice as group and ptrace stops.
1415 		 *
1416 		 * If a ptracer wants to distinguish the two events for its
1417 		 * own children, it should create a separate process which
1418 		 * takes the role of real parent.
1419 		 */
1420 		if (likely(!ptrace) && p->ptrace && !ptrace_reparented(p))
1421 			return 0;
1422 
1423 		/*
1424 		 * @p is alive and it's gonna stop, continue or exit, so
1425 		 * there always is something to wait for.
1426 		 */
1427 		wo->notask_error = 0;
1428 	}
1429 
1430 	/*
1431 	 * Wait for stopped.  Depending on @ptrace, different stopped state
1432 	 * is used and the two don't interact with each other.
1433 	 */
1434 	ret = wait_task_stopped(wo, ptrace, p);
1435 	if (ret)
1436 		return ret;
1437 
1438 	/*
1439 	 * Wait for continued.  There's only one continued state and the
1440 	 * ptracer can consume it which can confuse the real parent.  Don't
1441 	 * use WCONTINUED from ptracer.  You don't need or want it.
1442 	 */
1443 	return wait_task_continued(wo, p);
1444 }
1445 
1446 /*
1447  * Do the work of do_wait() for one thread in the group, @tsk.
1448  *
1449  * -ECHILD should be in ->notask_error before the first call.
1450  * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1451  * Returns zero if the search for a child should continue; then
1452  * ->notask_error is 0 if there were any eligible children,
1453  * or another error from security_task_wait(), or still -ECHILD.
1454  */
1455 static int do_wait_thread(struct wait_opts *wo, struct task_struct *tsk)
1456 {
1457 	struct task_struct *p;
1458 
1459 	list_for_each_entry(p, &tsk->children, sibling) {
1460 		int ret = wait_consider_task(wo, 0, p);
1461 		if (ret)
1462 			return ret;
1463 	}
1464 
1465 	return 0;
1466 }
1467 
1468 static int ptrace_do_wait(struct wait_opts *wo, struct task_struct *tsk)
1469 {
1470 	struct task_struct *p;
1471 
1472 	list_for_each_entry(p, &tsk->ptraced, ptrace_entry) {
1473 		int ret = wait_consider_task(wo, 1, p);
1474 		if (ret)
1475 			return ret;
1476 	}
1477 
1478 	return 0;
1479 }
1480 
1481 static int child_wait_callback(wait_queue_t *wait, unsigned mode,
1482 				int sync, void *key)
1483 {
1484 	struct wait_opts *wo = container_of(wait, struct wait_opts,
1485 						child_wait);
1486 	struct task_struct *p = key;
1487 
1488 	if (!eligible_pid(wo, p))
1489 		return 0;
1490 
1491 	if ((wo->wo_flags & __WNOTHREAD) && wait->private != p->parent)
1492 		return 0;
1493 
1494 	return default_wake_function(wait, mode, sync, key);
1495 }
1496 
1497 void __wake_up_parent(struct task_struct *p, struct task_struct *parent)
1498 {
1499 	__wake_up_sync_key(&parent->signal->wait_chldexit,
1500 				TASK_INTERRUPTIBLE, 1, p);
1501 }
1502 
1503 static long do_wait(struct wait_opts *wo)
1504 {
1505 	struct task_struct *tsk;
1506 	int retval;
1507 
1508 	trace_sched_process_wait(wo->wo_pid);
1509 
1510 	init_waitqueue_func_entry(&wo->child_wait, child_wait_callback);
1511 	wo->child_wait.private = current;
1512 	add_wait_queue(&current->signal->wait_chldexit, &wo->child_wait);
1513 repeat:
1514 	/*
1515 	 * If there is nothing that can match our critiera just get out.
1516 	 * We will clear ->notask_error to zero if we see any child that
1517 	 * might later match our criteria, even if we are not able to reap
1518 	 * it yet.
1519 	 */
1520 	wo->notask_error = -ECHILD;
1521 	if ((wo->wo_type < PIDTYPE_MAX) &&
1522 	   (!wo->wo_pid || hlist_empty(&wo->wo_pid->tasks[wo->wo_type])))
1523 		goto notask;
1524 
1525 	set_current_state(TASK_INTERRUPTIBLE);
1526 	read_lock(&tasklist_lock);
1527 	tsk = current;
1528 	do {
1529 		retval = do_wait_thread(wo, tsk);
1530 		if (retval)
1531 			goto end;
1532 
1533 		retval = ptrace_do_wait(wo, tsk);
1534 		if (retval)
1535 			goto end;
1536 
1537 		if (wo->wo_flags & __WNOTHREAD)
1538 			break;
1539 	} while_each_thread(current, tsk);
1540 	read_unlock(&tasklist_lock);
1541 
1542 notask:
1543 	retval = wo->notask_error;
1544 	if (!retval && !(wo->wo_flags & WNOHANG)) {
1545 		retval = -ERESTARTSYS;
1546 		if (!signal_pending(current)) {
1547 			schedule();
1548 			goto repeat;
1549 		}
1550 	}
1551 end:
1552 	__set_current_state(TASK_RUNNING);
1553 	remove_wait_queue(&current->signal->wait_chldexit, &wo->child_wait);
1554 	return retval;
1555 }
1556 
1557 SYSCALL_DEFINE5(waitid, int, which, pid_t, upid, struct siginfo __user *,
1558 		infop, int, options, struct rusage __user *, ru)
1559 {
1560 	struct wait_opts wo;
1561 	struct pid *pid = NULL;
1562 	enum pid_type type;
1563 	long ret;
1564 
1565 	if (options & ~(WNOHANG|WNOWAIT|WEXITED|WSTOPPED|WCONTINUED))
1566 		return -EINVAL;
1567 	if (!(options & (WEXITED|WSTOPPED|WCONTINUED)))
1568 		return -EINVAL;
1569 
1570 	switch (which) {
1571 	case P_ALL:
1572 		type = PIDTYPE_MAX;
1573 		break;
1574 	case P_PID:
1575 		type = PIDTYPE_PID;
1576 		if (upid <= 0)
1577 			return -EINVAL;
1578 		break;
1579 	case P_PGID:
1580 		type = PIDTYPE_PGID;
1581 		if (upid <= 0)
1582 			return -EINVAL;
1583 		break;
1584 	default:
1585 		return -EINVAL;
1586 	}
1587 
1588 	if (type < PIDTYPE_MAX)
1589 		pid = find_get_pid(upid);
1590 
1591 	wo.wo_type	= type;
1592 	wo.wo_pid	= pid;
1593 	wo.wo_flags	= options;
1594 	wo.wo_info	= infop;
1595 	wo.wo_stat	= NULL;
1596 	wo.wo_rusage	= ru;
1597 	ret = do_wait(&wo);
1598 
1599 	if (ret > 0) {
1600 		ret = 0;
1601 	} else if (infop) {
1602 		/*
1603 		 * For a WNOHANG return, clear out all the fields
1604 		 * we would set so the user can easily tell the
1605 		 * difference.
1606 		 */
1607 		if (!ret)
1608 			ret = put_user(0, &infop->si_signo);
1609 		if (!ret)
1610 			ret = put_user(0, &infop->si_errno);
1611 		if (!ret)
1612 			ret = put_user(0, &infop->si_code);
1613 		if (!ret)
1614 			ret = put_user(0, &infop->si_pid);
1615 		if (!ret)
1616 			ret = put_user(0, &infop->si_uid);
1617 		if (!ret)
1618 			ret = put_user(0, &infop->si_status);
1619 	}
1620 
1621 	put_pid(pid);
1622 	return ret;
1623 }
1624 
1625 SYSCALL_DEFINE4(wait4, pid_t, upid, int __user *, stat_addr,
1626 		int, options, struct rusage __user *, ru)
1627 {
1628 	struct wait_opts wo;
1629 	struct pid *pid = NULL;
1630 	enum pid_type type;
1631 	long ret;
1632 
1633 	if (options & ~(WNOHANG|WUNTRACED|WCONTINUED|
1634 			__WNOTHREAD|__WCLONE|__WALL))
1635 		return -EINVAL;
1636 
1637 	if (upid == -1)
1638 		type = PIDTYPE_MAX;
1639 	else if (upid < 0) {
1640 		type = PIDTYPE_PGID;
1641 		pid = find_get_pid(-upid);
1642 	} else if (upid == 0) {
1643 		type = PIDTYPE_PGID;
1644 		pid = get_task_pid(current, PIDTYPE_PGID);
1645 	} else /* upid > 0 */ {
1646 		type = PIDTYPE_PID;
1647 		pid = find_get_pid(upid);
1648 	}
1649 
1650 	wo.wo_type	= type;
1651 	wo.wo_pid	= pid;
1652 	wo.wo_flags	= options | WEXITED;
1653 	wo.wo_info	= NULL;
1654 	wo.wo_stat	= stat_addr;
1655 	wo.wo_rusage	= ru;
1656 	ret = do_wait(&wo);
1657 	put_pid(pid);
1658 
1659 	return ret;
1660 }
1661 
1662 #ifdef __ARCH_WANT_SYS_WAITPID
1663 
1664 /*
1665  * sys_waitpid() remains for compatibility. waitpid() should be
1666  * implemented by calling sys_wait4() from libc.a.
1667  */
1668 SYSCALL_DEFINE3(waitpid, pid_t, pid, int __user *, stat_addr, int, options)
1669 {
1670 	return sys_wait4(pid, stat_addr, options, NULL);
1671 }
1672 
1673 #endif
1674