xref: /linux/kernel/exit.c (revision e0bf6c5ca2d3281f231c5f0c9bf145e9513644de)
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 
120 	/*
121 	 * Accumulate here the counters for all threads as they die. We could
122 	 * skip the group leader because it is the last user of signal_struct,
123 	 * but we want to avoid the race with thread_group_cputime() which can
124 	 * see the empty ->thread_head list.
125 	 */
126 	task_cputime(tsk, &utime, &stime);
127 	write_seqlock(&sig->stats_lock);
128 	sig->utime += utime;
129 	sig->stime += stime;
130 	sig->gtime += task_gtime(tsk);
131 	sig->min_flt += tsk->min_flt;
132 	sig->maj_flt += tsk->maj_flt;
133 	sig->nvcsw += tsk->nvcsw;
134 	sig->nivcsw += tsk->nivcsw;
135 	sig->inblock += task_io_get_inblock(tsk);
136 	sig->oublock += task_io_get_oublock(tsk);
137 	task_io_accounting_add(&sig->ioac, &tsk->ioac);
138 	sig->sum_sched_runtime += tsk->se.sum_exec_runtime;
139 	sig->nr_threads--;
140 	__unhash_process(tsk, group_dead);
141 	write_sequnlock(&sig->stats_lock);
142 
143 	/*
144 	 * Do this under ->siglock, we can race with another thread
145 	 * doing sigqueue_free() if we have SIGQUEUE_PREALLOC signals.
146 	 */
147 	flush_sigqueue(&tsk->pending);
148 	tsk->sighand = NULL;
149 	spin_unlock(&sighand->siglock);
150 
151 	__cleanup_sighand(sighand);
152 	clear_tsk_thread_flag(tsk, TIF_SIGPENDING);
153 	if (group_dead) {
154 		flush_sigqueue(&sig->shared_pending);
155 		tty_kref_put(tty);
156 	}
157 }
158 
159 static void delayed_put_task_struct(struct rcu_head *rhp)
160 {
161 	struct task_struct *tsk = container_of(rhp, struct task_struct, rcu);
162 
163 	perf_event_delayed_put(tsk);
164 	trace_sched_process_free(tsk);
165 	put_task_struct(tsk);
166 }
167 
168 
169 void release_task(struct task_struct *p)
170 {
171 	struct task_struct *leader;
172 	int zap_leader;
173 repeat:
174 	/* don't need to get the RCU readlock here - the process is dead and
175 	 * can't be modifying its own credentials. But shut RCU-lockdep up */
176 	rcu_read_lock();
177 	atomic_dec(&__task_cred(p)->user->processes);
178 	rcu_read_unlock();
179 
180 	proc_flush_task(p);
181 
182 	write_lock_irq(&tasklist_lock);
183 	ptrace_release_task(p);
184 	__exit_signal(p);
185 
186 	/*
187 	 * If we are the last non-leader member of the thread
188 	 * group, and the leader is zombie, then notify the
189 	 * group leader's parent process. (if it wants notification.)
190 	 */
191 	zap_leader = 0;
192 	leader = p->group_leader;
193 	if (leader != p && thread_group_empty(leader)
194 			&& leader->exit_state == EXIT_ZOMBIE) {
195 		/*
196 		 * If we were the last child thread and the leader has
197 		 * exited already, and the leader's parent ignores SIGCHLD,
198 		 * then we are the one who should release the leader.
199 		 */
200 		zap_leader = do_notify_parent(leader, leader->exit_signal);
201 		if (zap_leader)
202 			leader->exit_state = EXIT_DEAD;
203 	}
204 
205 	write_unlock_irq(&tasklist_lock);
206 	release_thread(p);
207 	call_rcu(&p->rcu, delayed_put_task_struct);
208 
209 	p = leader;
210 	if (unlikely(zap_leader))
211 		goto repeat;
212 }
213 
214 /*
215  * Determine if a process group is "orphaned", according to the POSIX
216  * definition in 2.2.2.52.  Orphaned process groups are not to be affected
217  * by terminal-generated stop signals.  Newly orphaned process groups are
218  * to receive a SIGHUP and a SIGCONT.
219  *
220  * "I ask you, have you ever known what it is to be an orphan?"
221  */
222 static int will_become_orphaned_pgrp(struct pid *pgrp,
223 					struct task_struct *ignored_task)
224 {
225 	struct task_struct *p;
226 
227 	do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
228 		if ((p == ignored_task) ||
229 		    (p->exit_state && thread_group_empty(p)) ||
230 		    is_global_init(p->real_parent))
231 			continue;
232 
233 		if (task_pgrp(p->real_parent) != pgrp &&
234 		    task_session(p->real_parent) == task_session(p))
235 			return 0;
236 	} while_each_pid_task(pgrp, PIDTYPE_PGID, p);
237 
238 	return 1;
239 }
240 
241 int is_current_pgrp_orphaned(void)
242 {
243 	int retval;
244 
245 	read_lock(&tasklist_lock);
246 	retval = will_become_orphaned_pgrp(task_pgrp(current), NULL);
247 	read_unlock(&tasklist_lock);
248 
249 	return retval;
250 }
251 
252 static bool has_stopped_jobs(struct pid *pgrp)
253 {
254 	struct task_struct *p;
255 
256 	do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
257 		if (p->signal->flags & SIGNAL_STOP_STOPPED)
258 			return true;
259 	} while_each_pid_task(pgrp, PIDTYPE_PGID, p);
260 
261 	return false;
262 }
263 
264 /*
265  * Check to see if any process groups have become orphaned as
266  * a result of our exiting, and if they have any stopped jobs,
267  * send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
268  */
269 static void
270 kill_orphaned_pgrp(struct task_struct *tsk, struct task_struct *parent)
271 {
272 	struct pid *pgrp = task_pgrp(tsk);
273 	struct task_struct *ignored_task = tsk;
274 
275 	if (!parent)
276 		/* exit: our father is in a different pgrp than
277 		 * we are and we were the only connection outside.
278 		 */
279 		parent = tsk->real_parent;
280 	else
281 		/* reparent: our child is in a different pgrp than
282 		 * we are, and it was the only connection outside.
283 		 */
284 		ignored_task = NULL;
285 
286 	if (task_pgrp(parent) != pgrp &&
287 	    task_session(parent) == task_session(tsk) &&
288 	    will_become_orphaned_pgrp(pgrp, ignored_task) &&
289 	    has_stopped_jobs(pgrp)) {
290 		__kill_pgrp_info(SIGHUP, SEND_SIG_PRIV, pgrp);
291 		__kill_pgrp_info(SIGCONT, SEND_SIG_PRIV, pgrp);
292 	}
293 }
294 
295 #ifdef CONFIG_MEMCG
296 /*
297  * A task is exiting.   If it owned this mm, find a new owner for the mm.
298  */
299 void mm_update_next_owner(struct mm_struct *mm)
300 {
301 	struct task_struct *c, *g, *p = current;
302 
303 retry:
304 	/*
305 	 * If the exiting or execing task is not the owner, it's
306 	 * someone else's problem.
307 	 */
308 	if (mm->owner != p)
309 		return;
310 	/*
311 	 * The current owner is exiting/execing and there are no other
312 	 * candidates.  Do not leave the mm pointing to a possibly
313 	 * freed task structure.
314 	 */
315 	if (atomic_read(&mm->mm_users) <= 1) {
316 		mm->owner = NULL;
317 		return;
318 	}
319 
320 	read_lock(&tasklist_lock);
321 	/*
322 	 * Search in the children
323 	 */
324 	list_for_each_entry(c, &p->children, sibling) {
325 		if (c->mm == mm)
326 			goto assign_new_owner;
327 	}
328 
329 	/*
330 	 * Search in the siblings
331 	 */
332 	list_for_each_entry(c, &p->real_parent->children, sibling) {
333 		if (c->mm == mm)
334 			goto assign_new_owner;
335 	}
336 
337 	/*
338 	 * Search through everything else, we should not get here often.
339 	 */
340 	for_each_process(g) {
341 		if (g->flags & PF_KTHREAD)
342 			continue;
343 		for_each_thread(g, c) {
344 			if (c->mm == mm)
345 				goto assign_new_owner;
346 			if (c->mm)
347 				break;
348 		}
349 	}
350 	read_unlock(&tasklist_lock);
351 	/*
352 	 * We found no owner yet mm_users > 1: this implies that we are
353 	 * most likely racing with swapoff (try_to_unuse()) or /proc or
354 	 * ptrace or page migration (get_task_mm()).  Mark owner as NULL.
355 	 */
356 	mm->owner = NULL;
357 	return;
358 
359 assign_new_owner:
360 	BUG_ON(c == p);
361 	get_task_struct(c);
362 	/*
363 	 * The task_lock protects c->mm from changing.
364 	 * We always want mm->owner->mm == mm
365 	 */
366 	task_lock(c);
367 	/*
368 	 * Delay read_unlock() till we have the task_lock()
369 	 * to ensure that c does not slip away underneath us
370 	 */
371 	read_unlock(&tasklist_lock);
372 	if (c->mm != mm) {
373 		task_unlock(c);
374 		put_task_struct(c);
375 		goto retry;
376 	}
377 	mm->owner = c;
378 	task_unlock(c);
379 	put_task_struct(c);
380 }
381 #endif /* CONFIG_MEMCG */
382 
383 /*
384  * Turn us into a lazy TLB process if we
385  * aren't already..
386  */
387 static void exit_mm(struct task_struct *tsk)
388 {
389 	struct mm_struct *mm = tsk->mm;
390 	struct core_state *core_state;
391 
392 	mm_release(tsk, mm);
393 	if (!mm)
394 		return;
395 	sync_mm_rss(mm);
396 	/*
397 	 * Serialize with any possible pending coredump.
398 	 * We must hold mmap_sem around checking core_state
399 	 * and clearing tsk->mm.  The core-inducing thread
400 	 * will increment ->nr_threads for each thread in the
401 	 * group with ->mm != NULL.
402 	 */
403 	down_read(&mm->mmap_sem);
404 	core_state = mm->core_state;
405 	if (core_state) {
406 		struct core_thread self;
407 
408 		up_read(&mm->mmap_sem);
409 
410 		self.task = tsk;
411 		self.next = xchg(&core_state->dumper.next, &self);
412 		/*
413 		 * Implies mb(), the result of xchg() must be visible
414 		 * to core_state->dumper.
415 		 */
416 		if (atomic_dec_and_test(&core_state->nr_threads))
417 			complete(&core_state->startup);
418 
419 		for (;;) {
420 			set_task_state(tsk, TASK_UNINTERRUPTIBLE);
421 			if (!self.task) /* see coredump_finish() */
422 				break;
423 			freezable_schedule();
424 		}
425 		__set_task_state(tsk, TASK_RUNNING);
426 		down_read(&mm->mmap_sem);
427 	}
428 	atomic_inc(&mm->mm_count);
429 	BUG_ON(mm != tsk->active_mm);
430 	/* more a memory barrier than a real lock */
431 	task_lock(tsk);
432 	tsk->mm = NULL;
433 	up_read(&mm->mmap_sem);
434 	enter_lazy_tlb(mm, current);
435 	task_unlock(tsk);
436 	mm_update_next_owner(mm);
437 	mmput(mm);
438 	if (test_thread_flag(TIF_MEMDIE))
439 		unmark_oom_victim();
440 }
441 
442 static struct task_struct *find_alive_thread(struct task_struct *p)
443 {
444 	struct task_struct *t;
445 
446 	for_each_thread(p, t) {
447 		if (!(t->flags & PF_EXITING))
448 			return t;
449 	}
450 	return NULL;
451 }
452 
453 static struct task_struct *find_child_reaper(struct task_struct *father)
454 	__releases(&tasklist_lock)
455 	__acquires(&tasklist_lock)
456 {
457 	struct pid_namespace *pid_ns = task_active_pid_ns(father);
458 	struct task_struct *reaper = pid_ns->child_reaper;
459 
460 	if (likely(reaper != father))
461 		return reaper;
462 
463 	reaper = find_alive_thread(father);
464 	if (reaper) {
465 		pid_ns->child_reaper = reaper;
466 		return reaper;
467 	}
468 
469 	write_unlock_irq(&tasklist_lock);
470 	if (unlikely(pid_ns == &init_pid_ns)) {
471 		panic("Attempted to kill init! exitcode=0x%08x\n",
472 			father->signal->group_exit_code ?: father->exit_code);
473 	}
474 	zap_pid_ns_processes(pid_ns);
475 	write_lock_irq(&tasklist_lock);
476 
477 	return father;
478 }
479 
480 /*
481  * When we die, we re-parent all our children, and try to:
482  * 1. give them to another thread in our thread group, if such a member exists
483  * 2. give it to the first ancestor process which prctl'd itself as a
484  *    child_subreaper for its children (like a service manager)
485  * 3. give it to the init process (PID 1) in our pid namespace
486  */
487 static struct task_struct *find_new_reaper(struct task_struct *father,
488 					   struct task_struct *child_reaper)
489 {
490 	struct task_struct *thread, *reaper;
491 
492 	thread = find_alive_thread(father);
493 	if (thread)
494 		return thread;
495 
496 	if (father->signal->has_child_subreaper) {
497 		/*
498 		 * Find the first ->is_child_subreaper ancestor in our pid_ns.
499 		 * We start from father to ensure we can not look into another
500 		 * namespace, this is safe because all its threads are dead.
501 		 */
502 		for (reaper = father;
503 		     !same_thread_group(reaper, child_reaper);
504 		     reaper = reaper->real_parent) {
505 			/* call_usermodehelper() descendants need this check */
506 			if (reaper == &init_task)
507 				break;
508 			if (!reaper->signal->is_child_subreaper)
509 				continue;
510 			thread = find_alive_thread(reaper);
511 			if (thread)
512 				return thread;
513 		}
514 	}
515 
516 	return child_reaper;
517 }
518 
519 /*
520 * Any that need to be release_task'd are put on the @dead list.
521  */
522 static void reparent_leader(struct task_struct *father, struct task_struct *p,
523 				struct list_head *dead)
524 {
525 	if (unlikely(p->exit_state == EXIT_DEAD))
526 		return;
527 
528 	/* We don't want people slaying init. */
529 	p->exit_signal = SIGCHLD;
530 
531 	/* If it has exited notify the new parent about this child's death. */
532 	if (!p->ptrace &&
533 	    p->exit_state == EXIT_ZOMBIE && thread_group_empty(p)) {
534 		if (do_notify_parent(p, p->exit_signal)) {
535 			p->exit_state = EXIT_DEAD;
536 			list_add(&p->ptrace_entry, dead);
537 		}
538 	}
539 
540 	kill_orphaned_pgrp(p, father);
541 }
542 
543 /*
544  * This does two things:
545  *
546  * A.  Make init inherit all the child processes
547  * B.  Check to see if any process groups have become orphaned
548  *	as a result of our exiting, and if they have any stopped
549  *	jobs, send them a SIGHUP and then a SIGCONT.  (POSIX 3.2.2.2)
550  */
551 static void forget_original_parent(struct task_struct *father,
552 					struct list_head *dead)
553 {
554 	struct task_struct *p, *t, *reaper;
555 
556 	if (unlikely(!list_empty(&father->ptraced)))
557 		exit_ptrace(father, dead);
558 
559 	/* Can drop and reacquire tasklist_lock */
560 	reaper = find_child_reaper(father);
561 	if (list_empty(&father->children))
562 		return;
563 
564 	reaper = find_new_reaper(father, reaper);
565 	list_for_each_entry(p, &father->children, sibling) {
566 		for_each_thread(p, t) {
567 			t->real_parent = reaper;
568 			BUG_ON((!t->ptrace) != (t->parent == father));
569 			if (likely(!t->ptrace))
570 				t->parent = t->real_parent;
571 			if (t->pdeath_signal)
572 				group_send_sig_info(t->pdeath_signal,
573 						    SEND_SIG_NOINFO, t);
574 		}
575 		/*
576 		 * If this is a threaded reparent there is no need to
577 		 * notify anyone anything has happened.
578 		 */
579 		if (!same_thread_group(reaper, father))
580 			reparent_leader(father, p, dead);
581 	}
582 	list_splice_tail_init(&father->children, &reaper->children);
583 }
584 
585 /*
586  * Send signals to all our closest relatives so that they know
587  * to properly mourn us..
588  */
589 static void exit_notify(struct task_struct *tsk, int group_dead)
590 {
591 	bool autoreap;
592 	struct task_struct *p, *n;
593 	LIST_HEAD(dead);
594 
595 	write_lock_irq(&tasklist_lock);
596 	forget_original_parent(tsk, &dead);
597 
598 	if (group_dead)
599 		kill_orphaned_pgrp(tsk->group_leader, NULL);
600 
601 	if (unlikely(tsk->ptrace)) {
602 		int sig = thread_group_leader(tsk) &&
603 				thread_group_empty(tsk) &&
604 				!ptrace_reparented(tsk) ?
605 			tsk->exit_signal : SIGCHLD;
606 		autoreap = do_notify_parent(tsk, sig);
607 	} else if (thread_group_leader(tsk)) {
608 		autoreap = thread_group_empty(tsk) &&
609 			do_notify_parent(tsk, tsk->exit_signal);
610 	} else {
611 		autoreap = true;
612 	}
613 
614 	tsk->exit_state = autoreap ? EXIT_DEAD : EXIT_ZOMBIE;
615 	if (tsk->exit_state == EXIT_DEAD)
616 		list_add(&tsk->ptrace_entry, &dead);
617 
618 	/* mt-exec, de_thread() is waiting for group leader */
619 	if (unlikely(tsk->signal->notify_count < 0))
620 		wake_up_process(tsk->signal->group_exit_task);
621 	write_unlock_irq(&tasklist_lock);
622 
623 	list_for_each_entry_safe(p, n, &dead, ptrace_entry) {
624 		list_del_init(&p->ptrace_entry);
625 		release_task(p);
626 	}
627 }
628 
629 #ifdef CONFIG_DEBUG_STACK_USAGE
630 static void check_stack_usage(void)
631 {
632 	static DEFINE_SPINLOCK(low_water_lock);
633 	static int lowest_to_date = THREAD_SIZE;
634 	unsigned long free;
635 
636 	free = stack_not_used(current);
637 
638 	if (free >= lowest_to_date)
639 		return;
640 
641 	spin_lock(&low_water_lock);
642 	if (free < lowest_to_date) {
643 		pr_warn("%s (%d) used greatest stack depth: %lu bytes left\n",
644 			current->comm, task_pid_nr(current), free);
645 		lowest_to_date = free;
646 	}
647 	spin_unlock(&low_water_lock);
648 }
649 #else
650 static inline void check_stack_usage(void) {}
651 #endif
652 
653 void do_exit(long code)
654 {
655 	struct task_struct *tsk = current;
656 	int group_dead;
657 	TASKS_RCU(int tasks_rcu_i);
658 
659 	profile_task_exit(tsk);
660 
661 	WARN_ON(blk_needs_flush_plug(tsk));
662 
663 	if (unlikely(in_interrupt()))
664 		panic("Aiee, killing interrupt handler!");
665 	if (unlikely(!tsk->pid))
666 		panic("Attempted to kill the idle task!");
667 
668 	/*
669 	 * If do_exit is called because this processes oopsed, it's possible
670 	 * that get_fs() was left as KERNEL_DS, so reset it to USER_DS before
671 	 * continuing. Amongst other possible reasons, this is to prevent
672 	 * mm_release()->clear_child_tid() from writing to a user-controlled
673 	 * kernel address.
674 	 */
675 	set_fs(USER_DS);
676 
677 	ptrace_event(PTRACE_EVENT_EXIT, code);
678 
679 	validate_creds_for_do_exit(tsk);
680 
681 	/*
682 	 * We're taking recursive faults here in do_exit. Safest is to just
683 	 * leave this task alone and wait for reboot.
684 	 */
685 	if (unlikely(tsk->flags & PF_EXITING)) {
686 		pr_alert("Fixing recursive fault but reboot is needed!\n");
687 		/*
688 		 * We can do this unlocked here. The futex code uses
689 		 * this flag just to verify whether the pi state
690 		 * cleanup has been done or not. In the worst case it
691 		 * loops once more. We pretend that the cleanup was
692 		 * done as there is no way to return. Either the
693 		 * OWNER_DIED bit is set by now or we push the blocked
694 		 * task into the wait for ever nirwana as well.
695 		 */
696 		tsk->flags |= PF_EXITPIDONE;
697 		set_current_state(TASK_UNINTERRUPTIBLE);
698 		schedule();
699 	}
700 
701 	exit_signals(tsk);  /* sets PF_EXITING */
702 	/*
703 	 * tsk->flags are checked in the futex code to protect against
704 	 * an exiting task cleaning up the robust pi futexes.
705 	 */
706 	smp_mb();
707 	raw_spin_unlock_wait(&tsk->pi_lock);
708 
709 	if (unlikely(in_atomic()))
710 		pr_info("note: %s[%d] exited with preempt_count %d\n",
711 			current->comm, task_pid_nr(current),
712 			preempt_count());
713 
714 	acct_update_integrals(tsk);
715 	/* sync mm's RSS info before statistics gathering */
716 	if (tsk->mm)
717 		sync_mm_rss(tsk->mm);
718 	group_dead = atomic_dec_and_test(&tsk->signal->live);
719 	if (group_dead) {
720 		hrtimer_cancel(&tsk->signal->real_timer);
721 		exit_itimers(tsk->signal);
722 		if (tsk->mm)
723 			setmax_mm_hiwater_rss(&tsk->signal->maxrss, tsk->mm);
724 	}
725 	acct_collect(code, group_dead);
726 	if (group_dead)
727 		tty_audit_exit();
728 	audit_free(tsk);
729 
730 	tsk->exit_code = code;
731 	taskstats_exit(tsk, group_dead);
732 
733 	exit_mm(tsk);
734 
735 	if (group_dead)
736 		acct_process();
737 	trace_sched_process_exit(tsk);
738 
739 	exit_sem(tsk);
740 	exit_shm(tsk);
741 	exit_files(tsk);
742 	exit_fs(tsk);
743 	if (group_dead)
744 		disassociate_ctty(1);
745 	exit_task_namespaces(tsk);
746 	exit_task_work(tsk);
747 	exit_thread();
748 
749 	/*
750 	 * Flush inherited counters to the parent - before the parent
751 	 * gets woken up by child-exit notifications.
752 	 *
753 	 * because of cgroup mode, must be called before cgroup_exit()
754 	 */
755 	perf_event_exit_task(tsk);
756 
757 	cgroup_exit(tsk);
758 
759 	module_put(task_thread_info(tsk)->exec_domain->module);
760 
761 	/*
762 	 * FIXME: do that only when needed, using sched_exit tracepoint
763 	 */
764 	flush_ptrace_hw_breakpoint(tsk);
765 
766 	TASKS_RCU(tasks_rcu_i = __srcu_read_lock(&tasks_rcu_exit_srcu));
767 	exit_notify(tsk, group_dead);
768 	proc_exit_connector(tsk);
769 #ifdef CONFIG_NUMA
770 	task_lock(tsk);
771 	mpol_put(tsk->mempolicy);
772 	tsk->mempolicy = NULL;
773 	task_unlock(tsk);
774 #endif
775 #ifdef CONFIG_FUTEX
776 	if (unlikely(current->pi_state_cache))
777 		kfree(current->pi_state_cache);
778 #endif
779 	/*
780 	 * Make sure we are holding no locks:
781 	 */
782 	debug_check_no_locks_held();
783 	/*
784 	 * We can do this unlocked here. The futex code uses this flag
785 	 * just to verify whether the pi state cleanup has been done
786 	 * or not. In the worst case it loops once more.
787 	 */
788 	tsk->flags |= PF_EXITPIDONE;
789 
790 	if (tsk->io_context)
791 		exit_io_context(tsk);
792 
793 	if (tsk->splice_pipe)
794 		free_pipe_info(tsk->splice_pipe);
795 
796 	if (tsk->task_frag.page)
797 		put_page(tsk->task_frag.page);
798 
799 	validate_creds_for_do_exit(tsk);
800 
801 	check_stack_usage();
802 	preempt_disable();
803 	if (tsk->nr_dirtied)
804 		__this_cpu_add(dirty_throttle_leaks, tsk->nr_dirtied);
805 	exit_rcu();
806 	TASKS_RCU(__srcu_read_unlock(&tasks_rcu_exit_srcu, tasks_rcu_i));
807 
808 	/*
809 	 * The setting of TASK_RUNNING by try_to_wake_up() may be delayed
810 	 * when the following two conditions become true.
811 	 *   - There is race condition of mmap_sem (It is acquired by
812 	 *     exit_mm()), and
813 	 *   - SMI occurs before setting TASK_RUNINNG.
814 	 *     (or hypervisor of virtual machine switches to other guest)
815 	 *  As a result, we may become TASK_RUNNING after becoming TASK_DEAD
816 	 *
817 	 * To avoid it, we have to wait for releasing tsk->pi_lock which
818 	 * is held by try_to_wake_up()
819 	 */
820 	smp_mb();
821 	raw_spin_unlock_wait(&tsk->pi_lock);
822 
823 	/* causes final put_task_struct in finish_task_switch(). */
824 	tsk->state = TASK_DEAD;
825 	tsk->flags |= PF_NOFREEZE;	/* tell freezer to ignore us */
826 	schedule();
827 	BUG();
828 	/* Avoid "noreturn function does return".  */
829 	for (;;)
830 		cpu_relax();	/* For when BUG is null */
831 }
832 EXPORT_SYMBOL_GPL(do_exit);
833 
834 void complete_and_exit(struct completion *comp, long code)
835 {
836 	if (comp)
837 		complete(comp);
838 
839 	do_exit(code);
840 }
841 EXPORT_SYMBOL(complete_and_exit);
842 
843 SYSCALL_DEFINE1(exit, int, error_code)
844 {
845 	do_exit((error_code&0xff)<<8);
846 }
847 
848 /*
849  * Take down every thread in the group.  This is called by fatal signals
850  * as well as by sys_exit_group (below).
851  */
852 void
853 do_group_exit(int exit_code)
854 {
855 	struct signal_struct *sig = current->signal;
856 
857 	BUG_ON(exit_code & 0x80); /* core dumps don't get here */
858 
859 	if (signal_group_exit(sig))
860 		exit_code = sig->group_exit_code;
861 	else if (!thread_group_empty(current)) {
862 		struct sighand_struct *const sighand = current->sighand;
863 
864 		spin_lock_irq(&sighand->siglock);
865 		if (signal_group_exit(sig))
866 			/* Another thread got here before we took the lock.  */
867 			exit_code = sig->group_exit_code;
868 		else {
869 			sig->group_exit_code = exit_code;
870 			sig->flags = SIGNAL_GROUP_EXIT;
871 			zap_other_threads(current);
872 		}
873 		spin_unlock_irq(&sighand->siglock);
874 	}
875 
876 	do_exit(exit_code);
877 	/* NOTREACHED */
878 }
879 
880 /*
881  * this kills every thread in the thread group. Note that any externally
882  * wait4()-ing process will get the correct exit code - even if this
883  * thread is not the thread group leader.
884  */
885 SYSCALL_DEFINE1(exit_group, int, error_code)
886 {
887 	do_group_exit((error_code & 0xff) << 8);
888 	/* NOTREACHED */
889 	return 0;
890 }
891 
892 struct wait_opts {
893 	enum pid_type		wo_type;
894 	int			wo_flags;
895 	struct pid		*wo_pid;
896 
897 	struct siginfo __user	*wo_info;
898 	int __user		*wo_stat;
899 	struct rusage __user	*wo_rusage;
900 
901 	wait_queue_t		child_wait;
902 	int			notask_error;
903 };
904 
905 static inline
906 struct pid *task_pid_type(struct task_struct *task, enum pid_type type)
907 {
908 	if (type != PIDTYPE_PID)
909 		task = task->group_leader;
910 	return task->pids[type].pid;
911 }
912 
913 static int eligible_pid(struct wait_opts *wo, struct task_struct *p)
914 {
915 	return	wo->wo_type == PIDTYPE_MAX ||
916 		task_pid_type(p, wo->wo_type) == wo->wo_pid;
917 }
918 
919 static int eligible_child(struct wait_opts *wo, struct task_struct *p)
920 {
921 	if (!eligible_pid(wo, p))
922 		return 0;
923 	/* Wait for all children (clone and not) if __WALL is set;
924 	 * otherwise, wait for clone children *only* if __WCLONE is
925 	 * set; otherwise, wait for non-clone children *only*.  (Note:
926 	 * A "clone" child here is one that reports to its parent
927 	 * using a signal other than SIGCHLD.) */
928 	if (((p->exit_signal != SIGCHLD) ^ !!(wo->wo_flags & __WCLONE))
929 	    && !(wo->wo_flags & __WALL))
930 		return 0;
931 
932 	return 1;
933 }
934 
935 static int wait_noreap_copyout(struct wait_opts *wo, struct task_struct *p,
936 				pid_t pid, uid_t uid, int why, int status)
937 {
938 	struct siginfo __user *infop;
939 	int retval = wo->wo_rusage
940 		? getrusage(p, RUSAGE_BOTH, wo->wo_rusage) : 0;
941 
942 	put_task_struct(p);
943 	infop = wo->wo_info;
944 	if (infop) {
945 		if (!retval)
946 			retval = put_user(SIGCHLD, &infop->si_signo);
947 		if (!retval)
948 			retval = put_user(0, &infop->si_errno);
949 		if (!retval)
950 			retval = put_user((short)why, &infop->si_code);
951 		if (!retval)
952 			retval = put_user(pid, &infop->si_pid);
953 		if (!retval)
954 			retval = put_user(uid, &infop->si_uid);
955 		if (!retval)
956 			retval = put_user(status, &infop->si_status);
957 	}
958 	if (!retval)
959 		retval = pid;
960 	return retval;
961 }
962 
963 /*
964  * Handle sys_wait4 work for one task in state EXIT_ZOMBIE.  We hold
965  * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
966  * the lock and this task is uninteresting.  If we return nonzero, we have
967  * released the lock and the system call should return.
968  */
969 static int wait_task_zombie(struct wait_opts *wo, struct task_struct *p)
970 {
971 	int state, retval, status;
972 	pid_t pid = task_pid_vnr(p);
973 	uid_t uid = from_kuid_munged(current_user_ns(), task_uid(p));
974 	struct siginfo __user *infop;
975 
976 	if (!likely(wo->wo_flags & WEXITED))
977 		return 0;
978 
979 	if (unlikely(wo->wo_flags & WNOWAIT)) {
980 		int exit_code = p->exit_code;
981 		int why;
982 
983 		get_task_struct(p);
984 		read_unlock(&tasklist_lock);
985 		sched_annotate_sleep();
986 
987 		if ((exit_code & 0x7f) == 0) {
988 			why = CLD_EXITED;
989 			status = exit_code >> 8;
990 		} else {
991 			why = (exit_code & 0x80) ? CLD_DUMPED : CLD_KILLED;
992 			status = exit_code & 0x7f;
993 		}
994 		return wait_noreap_copyout(wo, p, pid, uid, why, status);
995 	}
996 	/*
997 	 * Move the task's state to DEAD/TRACE, only one thread can do this.
998 	 */
999 	state = (ptrace_reparented(p) && thread_group_leader(p)) ?
1000 		EXIT_TRACE : EXIT_DEAD;
1001 	if (cmpxchg(&p->exit_state, EXIT_ZOMBIE, state) != EXIT_ZOMBIE)
1002 		return 0;
1003 	/*
1004 	 * We own this thread, nobody else can reap it.
1005 	 */
1006 	read_unlock(&tasklist_lock);
1007 	sched_annotate_sleep();
1008 
1009 	/*
1010 	 * Check thread_group_leader() to exclude the traced sub-threads.
1011 	 */
1012 	if (state == EXIT_DEAD && thread_group_leader(p)) {
1013 		struct signal_struct *sig = p->signal;
1014 		struct signal_struct *psig = current->signal;
1015 		unsigned long maxrss;
1016 		cputime_t tgutime, tgstime;
1017 
1018 		/*
1019 		 * The resource counters for the group leader are in its
1020 		 * own task_struct.  Those for dead threads in the group
1021 		 * are in its signal_struct, as are those for the child
1022 		 * processes it has previously reaped.  All these
1023 		 * accumulate in the parent's signal_struct c* fields.
1024 		 *
1025 		 * We don't bother to take a lock here to protect these
1026 		 * p->signal fields because the whole thread group is dead
1027 		 * and nobody can change them.
1028 		 *
1029 		 * psig->stats_lock also protects us from our sub-theads
1030 		 * which can reap other children at the same time. Until
1031 		 * we change k_getrusage()-like users to rely on this lock
1032 		 * we have to take ->siglock as well.
1033 		 *
1034 		 * We use thread_group_cputime_adjusted() to get times for
1035 		 * the thread group, which consolidates times for all threads
1036 		 * in the group including the group leader.
1037 		 */
1038 		thread_group_cputime_adjusted(p, &tgutime, &tgstime);
1039 		spin_lock_irq(&current->sighand->siglock);
1040 		write_seqlock(&psig->stats_lock);
1041 		psig->cutime += tgutime + sig->cutime;
1042 		psig->cstime += tgstime + sig->cstime;
1043 		psig->cgtime += task_gtime(p) + sig->gtime + sig->cgtime;
1044 		psig->cmin_flt +=
1045 			p->min_flt + sig->min_flt + sig->cmin_flt;
1046 		psig->cmaj_flt +=
1047 			p->maj_flt + sig->maj_flt + sig->cmaj_flt;
1048 		psig->cnvcsw +=
1049 			p->nvcsw + sig->nvcsw + sig->cnvcsw;
1050 		psig->cnivcsw +=
1051 			p->nivcsw + sig->nivcsw + sig->cnivcsw;
1052 		psig->cinblock +=
1053 			task_io_get_inblock(p) +
1054 			sig->inblock + sig->cinblock;
1055 		psig->coublock +=
1056 			task_io_get_oublock(p) +
1057 			sig->oublock + sig->coublock;
1058 		maxrss = max(sig->maxrss, sig->cmaxrss);
1059 		if (psig->cmaxrss < maxrss)
1060 			psig->cmaxrss = maxrss;
1061 		task_io_accounting_add(&psig->ioac, &p->ioac);
1062 		task_io_accounting_add(&psig->ioac, &sig->ioac);
1063 		write_sequnlock(&psig->stats_lock);
1064 		spin_unlock_irq(&current->sighand->siglock);
1065 	}
1066 
1067 	retval = wo->wo_rusage
1068 		? getrusage(p, RUSAGE_BOTH, wo->wo_rusage) : 0;
1069 	status = (p->signal->flags & SIGNAL_GROUP_EXIT)
1070 		? p->signal->group_exit_code : p->exit_code;
1071 	if (!retval && wo->wo_stat)
1072 		retval = put_user(status, wo->wo_stat);
1073 
1074 	infop = wo->wo_info;
1075 	if (!retval && infop)
1076 		retval = put_user(SIGCHLD, &infop->si_signo);
1077 	if (!retval && infop)
1078 		retval = put_user(0, &infop->si_errno);
1079 	if (!retval && infop) {
1080 		int why;
1081 
1082 		if ((status & 0x7f) == 0) {
1083 			why = CLD_EXITED;
1084 			status >>= 8;
1085 		} else {
1086 			why = (status & 0x80) ? CLD_DUMPED : CLD_KILLED;
1087 			status &= 0x7f;
1088 		}
1089 		retval = put_user((short)why, &infop->si_code);
1090 		if (!retval)
1091 			retval = put_user(status, &infop->si_status);
1092 	}
1093 	if (!retval && infop)
1094 		retval = put_user(pid, &infop->si_pid);
1095 	if (!retval && infop)
1096 		retval = put_user(uid, &infop->si_uid);
1097 	if (!retval)
1098 		retval = pid;
1099 
1100 	if (state == EXIT_TRACE) {
1101 		write_lock_irq(&tasklist_lock);
1102 		/* We dropped tasklist, ptracer could die and untrace */
1103 		ptrace_unlink(p);
1104 
1105 		/* If parent wants a zombie, don't release it now */
1106 		state = EXIT_ZOMBIE;
1107 		if (do_notify_parent(p, p->exit_signal))
1108 			state = EXIT_DEAD;
1109 		p->exit_state = state;
1110 		write_unlock_irq(&tasklist_lock);
1111 	}
1112 	if (state == EXIT_DEAD)
1113 		release_task(p);
1114 
1115 	return retval;
1116 }
1117 
1118 static int *task_stopped_code(struct task_struct *p, bool ptrace)
1119 {
1120 	if (ptrace) {
1121 		if (task_is_stopped_or_traced(p) &&
1122 		    !(p->jobctl & JOBCTL_LISTENING))
1123 			return &p->exit_code;
1124 	} else {
1125 		if (p->signal->flags & SIGNAL_STOP_STOPPED)
1126 			return &p->signal->group_exit_code;
1127 	}
1128 	return NULL;
1129 }
1130 
1131 /**
1132  * wait_task_stopped - Wait for %TASK_STOPPED or %TASK_TRACED
1133  * @wo: wait options
1134  * @ptrace: is the wait for ptrace
1135  * @p: task to wait for
1136  *
1137  * Handle sys_wait4() work for %p in state %TASK_STOPPED or %TASK_TRACED.
1138  *
1139  * CONTEXT:
1140  * read_lock(&tasklist_lock), which is released if return value is
1141  * non-zero.  Also, grabs and releases @p->sighand->siglock.
1142  *
1143  * RETURNS:
1144  * 0 if wait condition didn't exist and search for other wait conditions
1145  * should continue.  Non-zero return, -errno on failure and @p's pid on
1146  * success, implies that tasklist_lock is released and wait condition
1147  * search should terminate.
1148  */
1149 static int wait_task_stopped(struct wait_opts *wo,
1150 				int ptrace, struct task_struct *p)
1151 {
1152 	struct siginfo __user *infop;
1153 	int retval, exit_code, *p_code, why;
1154 	uid_t uid = 0; /* unneeded, required by compiler */
1155 	pid_t pid;
1156 
1157 	/*
1158 	 * Traditionally we see ptrace'd stopped tasks regardless of options.
1159 	 */
1160 	if (!ptrace && !(wo->wo_flags & WUNTRACED))
1161 		return 0;
1162 
1163 	if (!task_stopped_code(p, ptrace))
1164 		return 0;
1165 
1166 	exit_code = 0;
1167 	spin_lock_irq(&p->sighand->siglock);
1168 
1169 	p_code = task_stopped_code(p, ptrace);
1170 	if (unlikely(!p_code))
1171 		goto unlock_sig;
1172 
1173 	exit_code = *p_code;
1174 	if (!exit_code)
1175 		goto unlock_sig;
1176 
1177 	if (!unlikely(wo->wo_flags & WNOWAIT))
1178 		*p_code = 0;
1179 
1180 	uid = from_kuid_munged(current_user_ns(), task_uid(p));
1181 unlock_sig:
1182 	spin_unlock_irq(&p->sighand->siglock);
1183 	if (!exit_code)
1184 		return 0;
1185 
1186 	/*
1187 	 * Now we are pretty sure this task is interesting.
1188 	 * Make sure it doesn't get reaped out from under us while we
1189 	 * give up the lock and then examine it below.  We don't want to
1190 	 * keep holding onto the tasklist_lock while we call getrusage and
1191 	 * possibly take page faults for user memory.
1192 	 */
1193 	get_task_struct(p);
1194 	pid = task_pid_vnr(p);
1195 	why = ptrace ? CLD_TRAPPED : CLD_STOPPED;
1196 	read_unlock(&tasklist_lock);
1197 	sched_annotate_sleep();
1198 
1199 	if (unlikely(wo->wo_flags & WNOWAIT))
1200 		return wait_noreap_copyout(wo, p, pid, uid, why, exit_code);
1201 
1202 	retval = wo->wo_rusage
1203 		? getrusage(p, RUSAGE_BOTH, wo->wo_rusage) : 0;
1204 	if (!retval && wo->wo_stat)
1205 		retval = put_user((exit_code << 8) | 0x7f, wo->wo_stat);
1206 
1207 	infop = wo->wo_info;
1208 	if (!retval && infop)
1209 		retval = put_user(SIGCHLD, &infop->si_signo);
1210 	if (!retval && infop)
1211 		retval = put_user(0, &infop->si_errno);
1212 	if (!retval && infop)
1213 		retval = put_user((short)why, &infop->si_code);
1214 	if (!retval && infop)
1215 		retval = put_user(exit_code, &infop->si_status);
1216 	if (!retval && infop)
1217 		retval = put_user(pid, &infop->si_pid);
1218 	if (!retval && infop)
1219 		retval = put_user(uid, &infop->si_uid);
1220 	if (!retval)
1221 		retval = pid;
1222 	put_task_struct(p);
1223 
1224 	BUG_ON(!retval);
1225 	return retval;
1226 }
1227 
1228 /*
1229  * Handle do_wait work for one task in a live, non-stopped state.
1230  * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
1231  * the lock and this task is uninteresting.  If we return nonzero, we have
1232  * released the lock and the system call should return.
1233  */
1234 static int wait_task_continued(struct wait_opts *wo, struct task_struct *p)
1235 {
1236 	int retval;
1237 	pid_t pid;
1238 	uid_t uid;
1239 
1240 	if (!unlikely(wo->wo_flags & WCONTINUED))
1241 		return 0;
1242 
1243 	if (!(p->signal->flags & SIGNAL_STOP_CONTINUED))
1244 		return 0;
1245 
1246 	spin_lock_irq(&p->sighand->siglock);
1247 	/* Re-check with the lock held.  */
1248 	if (!(p->signal->flags & SIGNAL_STOP_CONTINUED)) {
1249 		spin_unlock_irq(&p->sighand->siglock);
1250 		return 0;
1251 	}
1252 	if (!unlikely(wo->wo_flags & WNOWAIT))
1253 		p->signal->flags &= ~SIGNAL_STOP_CONTINUED;
1254 	uid = from_kuid_munged(current_user_ns(), task_uid(p));
1255 	spin_unlock_irq(&p->sighand->siglock);
1256 
1257 	pid = task_pid_vnr(p);
1258 	get_task_struct(p);
1259 	read_unlock(&tasklist_lock);
1260 	sched_annotate_sleep();
1261 
1262 	if (!wo->wo_info) {
1263 		retval = wo->wo_rusage
1264 			? getrusage(p, RUSAGE_BOTH, wo->wo_rusage) : 0;
1265 		put_task_struct(p);
1266 		if (!retval && wo->wo_stat)
1267 			retval = put_user(0xffff, wo->wo_stat);
1268 		if (!retval)
1269 			retval = pid;
1270 	} else {
1271 		retval = wait_noreap_copyout(wo, p, pid, uid,
1272 					     CLD_CONTINUED, SIGCONT);
1273 		BUG_ON(retval == 0);
1274 	}
1275 
1276 	return retval;
1277 }
1278 
1279 /*
1280  * Consider @p for a wait by @parent.
1281  *
1282  * -ECHILD should be in ->notask_error before the first call.
1283  * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1284  * Returns zero if the search for a child should continue;
1285  * then ->notask_error is 0 if @p is an eligible child,
1286  * or another error from security_task_wait(), or still -ECHILD.
1287  */
1288 static int wait_consider_task(struct wait_opts *wo, int ptrace,
1289 				struct task_struct *p)
1290 {
1291 	/*
1292 	 * We can race with wait_task_zombie() from another thread.
1293 	 * Ensure that EXIT_ZOMBIE -> EXIT_DEAD/EXIT_TRACE transition
1294 	 * can't confuse the checks below.
1295 	 */
1296 	int exit_state = ACCESS_ONCE(p->exit_state);
1297 	int ret;
1298 
1299 	if (unlikely(exit_state == EXIT_DEAD))
1300 		return 0;
1301 
1302 	ret = eligible_child(wo, p);
1303 	if (!ret)
1304 		return ret;
1305 
1306 	ret = security_task_wait(p);
1307 	if (unlikely(ret < 0)) {
1308 		/*
1309 		 * If we have not yet seen any eligible child,
1310 		 * then let this error code replace -ECHILD.
1311 		 * A permission error will give the user a clue
1312 		 * to look for security policy problems, rather
1313 		 * than for mysterious wait bugs.
1314 		 */
1315 		if (wo->notask_error)
1316 			wo->notask_error = ret;
1317 		return 0;
1318 	}
1319 
1320 	if (unlikely(exit_state == EXIT_TRACE)) {
1321 		/*
1322 		 * ptrace == 0 means we are the natural parent. In this case
1323 		 * we should clear notask_error, debugger will notify us.
1324 		 */
1325 		if (likely(!ptrace))
1326 			wo->notask_error = 0;
1327 		return 0;
1328 	}
1329 
1330 	if (likely(!ptrace) && unlikely(p->ptrace)) {
1331 		/*
1332 		 * If it is traced by its real parent's group, just pretend
1333 		 * the caller is ptrace_do_wait() and reap this child if it
1334 		 * is zombie.
1335 		 *
1336 		 * This also hides group stop state from real parent; otherwise
1337 		 * a single stop can be reported twice as group and ptrace stop.
1338 		 * If a ptracer wants to distinguish these two events for its
1339 		 * own children it should create a separate process which takes
1340 		 * the role of real parent.
1341 		 */
1342 		if (!ptrace_reparented(p))
1343 			ptrace = 1;
1344 	}
1345 
1346 	/* slay zombie? */
1347 	if (exit_state == EXIT_ZOMBIE) {
1348 		/* we don't reap group leaders with subthreads */
1349 		if (!delay_group_leader(p)) {
1350 			/*
1351 			 * A zombie ptracee is only visible to its ptracer.
1352 			 * Notification and reaping will be cascaded to the
1353 			 * real parent when the ptracer detaches.
1354 			 */
1355 			if (unlikely(ptrace) || likely(!p->ptrace))
1356 				return wait_task_zombie(wo, p);
1357 		}
1358 
1359 		/*
1360 		 * Allow access to stopped/continued state via zombie by
1361 		 * falling through.  Clearing of notask_error is complex.
1362 		 *
1363 		 * When !@ptrace:
1364 		 *
1365 		 * If WEXITED is set, notask_error should naturally be
1366 		 * cleared.  If not, subset of WSTOPPED|WCONTINUED is set,
1367 		 * so, if there are live subthreads, there are events to
1368 		 * wait for.  If all subthreads are dead, it's still safe
1369 		 * to clear - this function will be called again in finite
1370 		 * amount time once all the subthreads are released and
1371 		 * will then return without clearing.
1372 		 *
1373 		 * When @ptrace:
1374 		 *
1375 		 * Stopped state is per-task and thus can't change once the
1376 		 * target task dies.  Only continued and exited can happen.
1377 		 * Clear notask_error if WCONTINUED | WEXITED.
1378 		 */
1379 		if (likely(!ptrace) || (wo->wo_flags & (WCONTINUED | WEXITED)))
1380 			wo->notask_error = 0;
1381 	} else {
1382 		/*
1383 		 * @p is alive and it's gonna stop, continue or exit, so
1384 		 * there always is something to wait for.
1385 		 */
1386 		wo->notask_error = 0;
1387 	}
1388 
1389 	/*
1390 	 * Wait for stopped.  Depending on @ptrace, different stopped state
1391 	 * is used and the two don't interact with each other.
1392 	 */
1393 	ret = wait_task_stopped(wo, ptrace, p);
1394 	if (ret)
1395 		return ret;
1396 
1397 	/*
1398 	 * Wait for continued.  There's only one continued state and the
1399 	 * ptracer can consume it which can confuse the real parent.  Don't
1400 	 * use WCONTINUED from ptracer.  You don't need or want it.
1401 	 */
1402 	return wait_task_continued(wo, p);
1403 }
1404 
1405 /*
1406  * Do the work of do_wait() for one thread in the group, @tsk.
1407  *
1408  * -ECHILD should be in ->notask_error before the first call.
1409  * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1410  * Returns zero if the search for a child should continue; then
1411  * ->notask_error is 0 if there were any eligible children,
1412  * or another error from security_task_wait(), or still -ECHILD.
1413  */
1414 static int do_wait_thread(struct wait_opts *wo, struct task_struct *tsk)
1415 {
1416 	struct task_struct *p;
1417 
1418 	list_for_each_entry(p, &tsk->children, sibling) {
1419 		int ret = wait_consider_task(wo, 0, p);
1420 
1421 		if (ret)
1422 			return ret;
1423 	}
1424 
1425 	return 0;
1426 }
1427 
1428 static int ptrace_do_wait(struct wait_opts *wo, struct task_struct *tsk)
1429 {
1430 	struct task_struct *p;
1431 
1432 	list_for_each_entry(p, &tsk->ptraced, ptrace_entry) {
1433 		int ret = wait_consider_task(wo, 1, p);
1434 
1435 		if (ret)
1436 			return ret;
1437 	}
1438 
1439 	return 0;
1440 }
1441 
1442 static int child_wait_callback(wait_queue_t *wait, unsigned mode,
1443 				int sync, void *key)
1444 {
1445 	struct wait_opts *wo = container_of(wait, struct wait_opts,
1446 						child_wait);
1447 	struct task_struct *p = key;
1448 
1449 	if (!eligible_pid(wo, p))
1450 		return 0;
1451 
1452 	if ((wo->wo_flags & __WNOTHREAD) && wait->private != p->parent)
1453 		return 0;
1454 
1455 	return default_wake_function(wait, mode, sync, key);
1456 }
1457 
1458 void __wake_up_parent(struct task_struct *p, struct task_struct *parent)
1459 {
1460 	__wake_up_sync_key(&parent->signal->wait_chldexit,
1461 				TASK_INTERRUPTIBLE, 1, p);
1462 }
1463 
1464 static long do_wait(struct wait_opts *wo)
1465 {
1466 	struct task_struct *tsk;
1467 	int retval;
1468 
1469 	trace_sched_process_wait(wo->wo_pid);
1470 
1471 	init_waitqueue_func_entry(&wo->child_wait, child_wait_callback);
1472 	wo->child_wait.private = current;
1473 	add_wait_queue(&current->signal->wait_chldexit, &wo->child_wait);
1474 repeat:
1475 	/*
1476 	 * If there is nothing that can match our critiera just get out.
1477 	 * We will clear ->notask_error to zero if we see any child that
1478 	 * might later match our criteria, even if we are not able to reap
1479 	 * it yet.
1480 	 */
1481 	wo->notask_error = -ECHILD;
1482 	if ((wo->wo_type < PIDTYPE_MAX) &&
1483 	   (!wo->wo_pid || hlist_empty(&wo->wo_pid->tasks[wo->wo_type])))
1484 		goto notask;
1485 
1486 	set_current_state(TASK_INTERRUPTIBLE);
1487 	read_lock(&tasklist_lock);
1488 	tsk = current;
1489 	do {
1490 		retval = do_wait_thread(wo, tsk);
1491 		if (retval)
1492 			goto end;
1493 
1494 		retval = ptrace_do_wait(wo, tsk);
1495 		if (retval)
1496 			goto end;
1497 
1498 		if (wo->wo_flags & __WNOTHREAD)
1499 			break;
1500 	} while_each_thread(current, tsk);
1501 	read_unlock(&tasklist_lock);
1502 
1503 notask:
1504 	retval = wo->notask_error;
1505 	if (!retval && !(wo->wo_flags & WNOHANG)) {
1506 		retval = -ERESTARTSYS;
1507 		if (!signal_pending(current)) {
1508 			schedule();
1509 			goto repeat;
1510 		}
1511 	}
1512 end:
1513 	__set_current_state(TASK_RUNNING);
1514 	remove_wait_queue(&current->signal->wait_chldexit, &wo->child_wait);
1515 	return retval;
1516 }
1517 
1518 SYSCALL_DEFINE5(waitid, int, which, pid_t, upid, struct siginfo __user *,
1519 		infop, int, options, struct rusage __user *, ru)
1520 {
1521 	struct wait_opts wo;
1522 	struct pid *pid = NULL;
1523 	enum pid_type type;
1524 	long ret;
1525 
1526 	if (options & ~(WNOHANG|WNOWAIT|WEXITED|WSTOPPED|WCONTINUED))
1527 		return -EINVAL;
1528 	if (!(options & (WEXITED|WSTOPPED|WCONTINUED)))
1529 		return -EINVAL;
1530 
1531 	switch (which) {
1532 	case P_ALL:
1533 		type = PIDTYPE_MAX;
1534 		break;
1535 	case P_PID:
1536 		type = PIDTYPE_PID;
1537 		if (upid <= 0)
1538 			return -EINVAL;
1539 		break;
1540 	case P_PGID:
1541 		type = PIDTYPE_PGID;
1542 		if (upid <= 0)
1543 			return -EINVAL;
1544 		break;
1545 	default:
1546 		return -EINVAL;
1547 	}
1548 
1549 	if (type < PIDTYPE_MAX)
1550 		pid = find_get_pid(upid);
1551 
1552 	wo.wo_type	= type;
1553 	wo.wo_pid	= pid;
1554 	wo.wo_flags	= options;
1555 	wo.wo_info	= infop;
1556 	wo.wo_stat	= NULL;
1557 	wo.wo_rusage	= ru;
1558 	ret = do_wait(&wo);
1559 
1560 	if (ret > 0) {
1561 		ret = 0;
1562 	} else if (infop) {
1563 		/*
1564 		 * For a WNOHANG return, clear out all the fields
1565 		 * we would set so the user can easily tell the
1566 		 * difference.
1567 		 */
1568 		if (!ret)
1569 			ret = put_user(0, &infop->si_signo);
1570 		if (!ret)
1571 			ret = put_user(0, &infop->si_errno);
1572 		if (!ret)
1573 			ret = put_user(0, &infop->si_code);
1574 		if (!ret)
1575 			ret = put_user(0, &infop->si_pid);
1576 		if (!ret)
1577 			ret = put_user(0, &infop->si_uid);
1578 		if (!ret)
1579 			ret = put_user(0, &infop->si_status);
1580 	}
1581 
1582 	put_pid(pid);
1583 	return ret;
1584 }
1585 
1586 SYSCALL_DEFINE4(wait4, pid_t, upid, int __user *, stat_addr,
1587 		int, options, struct rusage __user *, ru)
1588 {
1589 	struct wait_opts wo;
1590 	struct pid *pid = NULL;
1591 	enum pid_type type;
1592 	long ret;
1593 
1594 	if (options & ~(WNOHANG|WUNTRACED|WCONTINUED|
1595 			__WNOTHREAD|__WCLONE|__WALL))
1596 		return -EINVAL;
1597 
1598 	if (upid == -1)
1599 		type = PIDTYPE_MAX;
1600 	else if (upid < 0) {
1601 		type = PIDTYPE_PGID;
1602 		pid = find_get_pid(-upid);
1603 	} else if (upid == 0) {
1604 		type = PIDTYPE_PGID;
1605 		pid = get_task_pid(current, PIDTYPE_PGID);
1606 	} else /* upid > 0 */ {
1607 		type = PIDTYPE_PID;
1608 		pid = find_get_pid(upid);
1609 	}
1610 
1611 	wo.wo_type	= type;
1612 	wo.wo_pid	= pid;
1613 	wo.wo_flags	= options | WEXITED;
1614 	wo.wo_info	= NULL;
1615 	wo.wo_stat	= stat_addr;
1616 	wo.wo_rusage	= ru;
1617 	ret = do_wait(&wo);
1618 	put_pid(pid);
1619 
1620 	return ret;
1621 }
1622 
1623 #ifdef __ARCH_WANT_SYS_WAITPID
1624 
1625 /*
1626  * sys_waitpid() remains for compatibility. waitpid() should be
1627  * implemented by calling sys_wait4() from libc.a.
1628  */
1629 SYSCALL_DEFINE3(waitpid, pid_t, pid, int __user *, stat_addr, int, options)
1630 {
1631 	return sys_wait4(pid, stat_addr, options, NULL);
1632 }
1633 
1634 #endif
1635