xref: /openbmc/linux/kernel/exit.c (revision e23feb16)
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 }
78 
79 /*
80  * This function expects the tasklist_lock write-locked.
81  */
82 static void __exit_signal(struct task_struct *tsk)
83 {
84 	struct signal_struct *sig = tsk->signal;
85 	bool group_dead = thread_group_leader(tsk);
86 	struct sighand_struct *sighand;
87 	struct tty_struct *uninitialized_var(tty);
88 	cputime_t utime, stime;
89 
90 	sighand = rcu_dereference_check(tsk->sighand,
91 					lockdep_tasklist_lock_is_held());
92 	spin_lock(&sighand->siglock);
93 
94 	posix_cpu_timers_exit(tsk);
95 	if (group_dead) {
96 		posix_cpu_timers_exit_group(tsk);
97 		tty = sig->tty;
98 		sig->tty = NULL;
99 	} else {
100 		/*
101 		 * This can only happen if the caller is de_thread().
102 		 * FIXME: this is the temporary hack, we should teach
103 		 * posix-cpu-timers to handle this case correctly.
104 		 */
105 		if (unlikely(has_group_leader_pid(tsk)))
106 			posix_cpu_timers_exit_group(tsk);
107 
108 		/*
109 		 * If there is any task waiting for the group exit
110 		 * then notify it:
111 		 */
112 		if (sig->notify_count > 0 && !--sig->notify_count)
113 			wake_up_process(sig->group_exit_task);
114 
115 		if (tsk == sig->curr_target)
116 			sig->curr_target = next_thread(tsk);
117 		/*
118 		 * Accumulate here the counters for all threads but the
119 		 * group leader as they die, so they can be added into
120 		 * the process-wide totals when those are taken.
121 		 * The group leader stays around as a zombie as long
122 		 * as there are other threads.  When it gets reaped,
123 		 * the exit.c code will add its counts into these totals.
124 		 * We won't ever get here for the group leader, since it
125 		 * will have been the last reference on the signal_struct.
126 		 */
127 		task_cputime(tsk, &utime, &stime);
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 	}
140 
141 	sig->nr_threads--;
142 	__unhash_process(tsk, group_dead);
143 
144 	/*
145 	 * Do this under ->siglock, we can race with another thread
146 	 * doing sigqueue_free() if we have SIGQUEUE_PREALLOC signals.
147 	 */
148 	flush_sigqueue(&tsk->pending);
149 	tsk->sighand = NULL;
150 	spin_unlock(&sighand->siglock);
151 
152 	__cleanup_sighand(sighand);
153 	clear_tsk_thread_flag(tsk,TIF_SIGPENDING);
154 	if (group_dead) {
155 		flush_sigqueue(&sig->shared_pending);
156 		tty_kref_put(tty);
157 	}
158 }
159 
160 static void delayed_put_task_struct(struct rcu_head *rhp)
161 {
162 	struct task_struct *tsk = container_of(rhp, struct task_struct, rcu);
163 
164 	perf_event_delayed_put(tsk);
165 	trace_sched_process_free(tsk);
166 	put_task_struct(tsk);
167 }
168 
169 
170 void release_task(struct task_struct * p)
171 {
172 	struct task_struct *leader;
173 	int zap_leader;
174 repeat:
175 	/* don't need to get the RCU readlock here - the process is dead and
176 	 * can't be modifying its own credentials. But shut RCU-lockdep up */
177 	rcu_read_lock();
178 	atomic_dec(&__task_cred(p)->user->processes);
179 	rcu_read_unlock();
180 
181 	proc_flush_task(p);
182 
183 	write_lock_irq(&tasklist_lock);
184 	ptrace_release_task(p);
185 	__exit_signal(p);
186 
187 	/*
188 	 * If we are the last non-leader member of the thread
189 	 * group, and the leader is zombie, then notify the
190 	 * group leader's parent process. (if it wants notification.)
191 	 */
192 	zap_leader = 0;
193 	leader = p->group_leader;
194 	if (leader != p && thread_group_empty(leader) && 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  * This checks not only the pgrp, but falls back on the pid if no
216  * satisfactory pgrp is found. I dunno - gdb doesn't work correctly
217  * without this...
218  *
219  * The caller must hold rcu lock or the tasklist lock.
220  */
221 struct pid *session_of_pgrp(struct pid *pgrp)
222 {
223 	struct task_struct *p;
224 	struct pid *sid = NULL;
225 
226 	p = pid_task(pgrp, PIDTYPE_PGID);
227 	if (p == NULL)
228 		p = pid_task(pgrp, PIDTYPE_PID);
229 	if (p != NULL)
230 		sid = task_session(p);
231 
232 	return sid;
233 }
234 
235 /*
236  * Determine if a process group is "orphaned", according to the POSIX
237  * definition in 2.2.2.52.  Orphaned process groups are not to be affected
238  * by terminal-generated stop signals.  Newly orphaned process groups are
239  * to receive a SIGHUP and a SIGCONT.
240  *
241  * "I ask you, have you ever known what it is to be an orphan?"
242  */
243 static int will_become_orphaned_pgrp(struct pid *pgrp, struct task_struct *ignored_task)
244 {
245 	struct task_struct *p;
246 
247 	do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
248 		if ((p == ignored_task) ||
249 		    (p->exit_state && thread_group_empty(p)) ||
250 		    is_global_init(p->real_parent))
251 			continue;
252 
253 		if (task_pgrp(p->real_parent) != pgrp &&
254 		    task_session(p->real_parent) == task_session(p))
255 			return 0;
256 	} while_each_pid_task(pgrp, PIDTYPE_PGID, p);
257 
258 	return 1;
259 }
260 
261 int is_current_pgrp_orphaned(void)
262 {
263 	int retval;
264 
265 	read_lock(&tasklist_lock);
266 	retval = will_become_orphaned_pgrp(task_pgrp(current), NULL);
267 	read_unlock(&tasklist_lock);
268 
269 	return retval;
270 }
271 
272 static bool has_stopped_jobs(struct pid *pgrp)
273 {
274 	struct task_struct *p;
275 
276 	do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
277 		if (p->signal->flags & SIGNAL_STOP_STOPPED)
278 			return true;
279 	} while_each_pid_task(pgrp, PIDTYPE_PGID, p);
280 
281 	return false;
282 }
283 
284 /*
285  * Check to see if any process groups have become orphaned as
286  * a result of our exiting, and if they have any stopped jobs,
287  * send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
288  */
289 static void
290 kill_orphaned_pgrp(struct task_struct *tsk, struct task_struct *parent)
291 {
292 	struct pid *pgrp = task_pgrp(tsk);
293 	struct task_struct *ignored_task = tsk;
294 
295 	if (!parent)
296 		 /* exit: our father is in a different pgrp than
297 		  * we are and we were the only connection outside.
298 		  */
299 		parent = tsk->real_parent;
300 	else
301 		/* reparent: our child is in a different pgrp than
302 		 * we are, and it was the only connection outside.
303 		 */
304 		ignored_task = NULL;
305 
306 	if (task_pgrp(parent) != pgrp &&
307 	    task_session(parent) == task_session(tsk) &&
308 	    will_become_orphaned_pgrp(pgrp, ignored_task) &&
309 	    has_stopped_jobs(pgrp)) {
310 		__kill_pgrp_info(SIGHUP, SEND_SIG_PRIV, pgrp);
311 		__kill_pgrp_info(SIGCONT, SEND_SIG_PRIV, pgrp);
312 	}
313 }
314 
315 /*
316  * Let kernel threads use this to say that they allow a certain signal.
317  * Must not be used if kthread was cloned with CLONE_SIGHAND.
318  */
319 int allow_signal(int sig)
320 {
321 	if (!valid_signal(sig) || sig < 1)
322 		return -EINVAL;
323 
324 	spin_lock_irq(&current->sighand->siglock);
325 	/* This is only needed for daemonize()'ed kthreads */
326 	sigdelset(&current->blocked, sig);
327 	/*
328 	 * Kernel threads handle their own signals. Let the signal code
329 	 * know it'll be handled, so that they don't get converted to
330 	 * SIGKILL or just silently dropped.
331 	 */
332 	current->sighand->action[(sig)-1].sa.sa_handler = (void __user *)2;
333 	recalc_sigpending();
334 	spin_unlock_irq(&current->sighand->siglock);
335 	return 0;
336 }
337 
338 EXPORT_SYMBOL(allow_signal);
339 
340 int disallow_signal(int sig)
341 {
342 	if (!valid_signal(sig) || sig < 1)
343 		return -EINVAL;
344 
345 	spin_lock_irq(&current->sighand->siglock);
346 	current->sighand->action[(sig)-1].sa.sa_handler = SIG_IGN;
347 	recalc_sigpending();
348 	spin_unlock_irq(&current->sighand->siglock);
349 	return 0;
350 }
351 
352 EXPORT_SYMBOL(disallow_signal);
353 
354 #ifdef CONFIG_MM_OWNER
355 /*
356  * A task is exiting.   If it owned this mm, find a new owner for the mm.
357  */
358 void mm_update_next_owner(struct mm_struct *mm)
359 {
360 	struct task_struct *c, *g, *p = current;
361 
362 retry:
363 	/*
364 	 * If the exiting or execing task is not the owner, it's
365 	 * someone else's problem.
366 	 */
367 	if (mm->owner != p)
368 		return;
369 	/*
370 	 * The current owner is exiting/execing and there are no other
371 	 * candidates.  Do not leave the mm pointing to a possibly
372 	 * freed task structure.
373 	 */
374 	if (atomic_read(&mm->mm_users) <= 1) {
375 		mm->owner = NULL;
376 		return;
377 	}
378 
379 	read_lock(&tasklist_lock);
380 	/*
381 	 * Search in the children
382 	 */
383 	list_for_each_entry(c, &p->children, sibling) {
384 		if (c->mm == mm)
385 			goto assign_new_owner;
386 	}
387 
388 	/*
389 	 * Search in the siblings
390 	 */
391 	list_for_each_entry(c, &p->real_parent->children, sibling) {
392 		if (c->mm == mm)
393 			goto assign_new_owner;
394 	}
395 
396 	/*
397 	 * Search through everything else. We should not get
398 	 * here often
399 	 */
400 	do_each_thread(g, c) {
401 		if (c->mm == mm)
402 			goto assign_new_owner;
403 	} while_each_thread(g, c);
404 
405 	read_unlock(&tasklist_lock);
406 	/*
407 	 * We found no owner yet mm_users > 1: this implies that we are
408 	 * most likely racing with swapoff (try_to_unuse()) or /proc or
409 	 * ptrace or page migration (get_task_mm()).  Mark owner as NULL.
410 	 */
411 	mm->owner = NULL;
412 	return;
413 
414 assign_new_owner:
415 	BUG_ON(c == p);
416 	get_task_struct(c);
417 	/*
418 	 * The task_lock protects c->mm from changing.
419 	 * We always want mm->owner->mm == mm
420 	 */
421 	task_lock(c);
422 	/*
423 	 * Delay read_unlock() till we have the task_lock()
424 	 * to ensure that c does not slip away underneath us
425 	 */
426 	read_unlock(&tasklist_lock);
427 	if (c->mm != mm) {
428 		task_unlock(c);
429 		put_task_struct(c);
430 		goto retry;
431 	}
432 	mm->owner = c;
433 	task_unlock(c);
434 	put_task_struct(c);
435 }
436 #endif /* CONFIG_MM_OWNER */
437 
438 /*
439  * Turn us into a lazy TLB process if we
440  * aren't already..
441  */
442 static void exit_mm(struct task_struct * tsk)
443 {
444 	struct mm_struct *mm = tsk->mm;
445 	struct core_state *core_state;
446 
447 	mm_release(tsk, mm);
448 	if (!mm)
449 		return;
450 	sync_mm_rss(mm);
451 	/*
452 	 * Serialize with any possible pending coredump.
453 	 * We must hold mmap_sem around checking core_state
454 	 * and clearing tsk->mm.  The core-inducing thread
455 	 * will increment ->nr_threads for each thread in the
456 	 * group with ->mm != NULL.
457 	 */
458 	down_read(&mm->mmap_sem);
459 	core_state = mm->core_state;
460 	if (core_state) {
461 		struct core_thread self;
462 		up_read(&mm->mmap_sem);
463 
464 		self.task = tsk;
465 		self.next = xchg(&core_state->dumper.next, &self);
466 		/*
467 		 * Implies mb(), the result of xchg() must be visible
468 		 * to core_state->dumper.
469 		 */
470 		if (atomic_dec_and_test(&core_state->nr_threads))
471 			complete(&core_state->startup);
472 
473 		for (;;) {
474 			set_task_state(tsk, TASK_UNINTERRUPTIBLE);
475 			if (!self.task) /* see coredump_finish() */
476 				break;
477 			freezable_schedule();
478 		}
479 		__set_task_state(tsk, TASK_RUNNING);
480 		down_read(&mm->mmap_sem);
481 	}
482 	atomic_inc(&mm->mm_count);
483 	BUG_ON(mm != tsk->active_mm);
484 	/* more a memory barrier than a real lock */
485 	task_lock(tsk);
486 	tsk->mm = NULL;
487 	up_read(&mm->mmap_sem);
488 	enter_lazy_tlb(mm, current);
489 	task_unlock(tsk);
490 	mm_update_next_owner(mm);
491 	mmput(mm);
492 }
493 
494 /*
495  * When we die, we re-parent all our children, and try to:
496  * 1. give them to another thread in our thread group, if such a member exists
497  * 2. give it to the first ancestor process which prctl'd itself as a
498  *    child_subreaper for its children (like a service manager)
499  * 3. give it to the init process (PID 1) in our pid namespace
500  */
501 static struct task_struct *find_new_reaper(struct task_struct *father)
502 	__releases(&tasklist_lock)
503 	__acquires(&tasklist_lock)
504 {
505 	struct pid_namespace *pid_ns = task_active_pid_ns(father);
506 	struct task_struct *thread;
507 
508 	thread = father;
509 	while_each_thread(father, thread) {
510 		if (thread->flags & PF_EXITING)
511 			continue;
512 		if (unlikely(pid_ns->child_reaper == father))
513 			pid_ns->child_reaper = thread;
514 		return thread;
515 	}
516 
517 	if (unlikely(pid_ns->child_reaper == father)) {
518 		write_unlock_irq(&tasklist_lock);
519 		if (unlikely(pid_ns == &init_pid_ns)) {
520 			panic("Attempted to kill init! exitcode=0x%08x\n",
521 				father->signal->group_exit_code ?:
522 					father->exit_code);
523 		}
524 
525 		zap_pid_ns_processes(pid_ns);
526 		write_lock_irq(&tasklist_lock);
527 	} else if (father->signal->has_child_subreaper) {
528 		struct task_struct *reaper;
529 
530 		/*
531 		 * Find the first ancestor marked as child_subreaper.
532 		 * Note that the code below checks same_thread_group(reaper,
533 		 * pid_ns->child_reaper).  This is what we need to DTRT in a
534 		 * PID namespace. However we still need the check above, see
535 		 * http://marc.info/?l=linux-kernel&m=131385460420380
536 		 */
537 		for (reaper = father->real_parent;
538 		     reaper != &init_task;
539 		     reaper = reaper->real_parent) {
540 			if (same_thread_group(reaper, pid_ns->child_reaper))
541 				break;
542 			if (!reaper->signal->is_child_subreaper)
543 				continue;
544 			thread = reaper;
545 			do {
546 				if (!(thread->flags & PF_EXITING))
547 					return reaper;
548 			} while_each_thread(reaper, thread);
549 		}
550 	}
551 
552 	return pid_ns->child_reaper;
553 }
554 
555 /*
556 * Any that need to be release_task'd are put on the @dead list.
557  */
558 static void reparent_leader(struct task_struct *father, struct task_struct *p,
559 				struct list_head *dead)
560 {
561 	list_move_tail(&p->sibling, &p->real_parent->children);
562 
563 	if (p->exit_state == EXIT_DEAD)
564 		return;
565 	/*
566 	 * If this is a threaded reparent there is no need to
567 	 * notify anyone anything has happened.
568 	 */
569 	if (same_thread_group(p->real_parent, father))
570 		return;
571 
572 	/* We don't want people slaying init.  */
573 	p->exit_signal = SIGCHLD;
574 
575 	/* If it has exited notify the new parent about this child's death. */
576 	if (!p->ptrace &&
577 	    p->exit_state == EXIT_ZOMBIE && thread_group_empty(p)) {
578 		if (do_notify_parent(p, p->exit_signal)) {
579 			p->exit_state = EXIT_DEAD;
580 			list_move_tail(&p->sibling, dead);
581 		}
582 	}
583 
584 	kill_orphaned_pgrp(p, father);
585 }
586 
587 static void forget_original_parent(struct task_struct *father)
588 {
589 	struct task_struct *p, *n, *reaper;
590 	LIST_HEAD(dead_children);
591 
592 	write_lock_irq(&tasklist_lock);
593 	/*
594 	 * Note that exit_ptrace() and find_new_reaper() might
595 	 * drop tasklist_lock and reacquire it.
596 	 */
597 	exit_ptrace(father);
598 	reaper = find_new_reaper(father);
599 
600 	list_for_each_entry_safe(p, n, &father->children, sibling) {
601 		struct task_struct *t = p;
602 		do {
603 			t->real_parent = reaper;
604 			if (t->parent == father) {
605 				BUG_ON(t->ptrace);
606 				t->parent = t->real_parent;
607 			}
608 			if (t->pdeath_signal)
609 				group_send_sig_info(t->pdeath_signal,
610 						    SEND_SIG_NOINFO, t);
611 		} while_each_thread(p, t);
612 		reparent_leader(father, p, &dead_children);
613 	}
614 	write_unlock_irq(&tasklist_lock);
615 
616 	BUG_ON(!list_empty(&father->children));
617 
618 	list_for_each_entry_safe(p, n, &dead_children, sibling) {
619 		list_del_init(&p->sibling);
620 		release_task(p);
621 	}
622 }
623 
624 /*
625  * Send signals to all our closest relatives so that they know
626  * to properly mourn us..
627  */
628 static void exit_notify(struct task_struct *tsk, int group_dead)
629 {
630 	bool autoreap;
631 
632 	/*
633 	 * This does two things:
634 	 *
635   	 * A.  Make init inherit all the child processes
636 	 * B.  Check to see if any process groups have become orphaned
637 	 *	as a result of our exiting, and if they have any stopped
638 	 *	jobs, send them a SIGHUP and then a SIGCONT.  (POSIX 3.2.2.2)
639 	 */
640 	forget_original_parent(tsk);
641 
642 	write_lock_irq(&tasklist_lock);
643 	if (group_dead)
644 		kill_orphaned_pgrp(tsk->group_leader, NULL);
645 
646 	if (unlikely(tsk->ptrace)) {
647 		int sig = thread_group_leader(tsk) &&
648 				thread_group_empty(tsk) &&
649 				!ptrace_reparented(tsk) ?
650 			tsk->exit_signal : SIGCHLD;
651 		autoreap = do_notify_parent(tsk, sig);
652 	} else if (thread_group_leader(tsk)) {
653 		autoreap = thread_group_empty(tsk) &&
654 			do_notify_parent(tsk, tsk->exit_signal);
655 	} else {
656 		autoreap = true;
657 	}
658 
659 	tsk->exit_state = autoreap ? EXIT_DEAD : EXIT_ZOMBIE;
660 
661 	/* mt-exec, de_thread() is waiting for group leader */
662 	if (unlikely(tsk->signal->notify_count < 0))
663 		wake_up_process(tsk->signal->group_exit_task);
664 	write_unlock_irq(&tasklist_lock);
665 
666 	/* If the process is dead, release it - nobody will wait for it */
667 	if (autoreap)
668 		release_task(tsk);
669 }
670 
671 #ifdef CONFIG_DEBUG_STACK_USAGE
672 static void check_stack_usage(void)
673 {
674 	static DEFINE_SPINLOCK(low_water_lock);
675 	static int lowest_to_date = THREAD_SIZE;
676 	unsigned long free;
677 
678 	free = stack_not_used(current);
679 
680 	if (free >= lowest_to_date)
681 		return;
682 
683 	spin_lock(&low_water_lock);
684 	if (free < lowest_to_date) {
685 		printk(KERN_WARNING "%s (%d) used greatest stack depth: "
686 				"%lu bytes left\n",
687 				current->comm, task_pid_nr(current), free);
688 		lowest_to_date = free;
689 	}
690 	spin_unlock(&low_water_lock);
691 }
692 #else
693 static inline void check_stack_usage(void) {}
694 #endif
695 
696 void do_exit(long code)
697 {
698 	struct task_struct *tsk = current;
699 	int group_dead;
700 
701 	profile_task_exit(tsk);
702 
703 	WARN_ON(blk_needs_flush_plug(tsk));
704 
705 	if (unlikely(in_interrupt()))
706 		panic("Aiee, killing interrupt handler!");
707 	if (unlikely(!tsk->pid))
708 		panic("Attempted to kill the idle task!");
709 
710 	/*
711 	 * If do_exit is called because this processes oopsed, it's possible
712 	 * that get_fs() was left as KERNEL_DS, so reset it to USER_DS before
713 	 * continuing. Amongst other possible reasons, this is to prevent
714 	 * mm_release()->clear_child_tid() from writing to a user-controlled
715 	 * kernel address.
716 	 */
717 	set_fs(USER_DS);
718 
719 	ptrace_event(PTRACE_EVENT_EXIT, code);
720 
721 	validate_creds_for_do_exit(tsk);
722 
723 	/*
724 	 * We're taking recursive faults here in do_exit. Safest is to just
725 	 * leave this task alone and wait for reboot.
726 	 */
727 	if (unlikely(tsk->flags & PF_EXITING)) {
728 		printk(KERN_ALERT
729 			"Fixing recursive fault but reboot is needed!\n");
730 		/*
731 		 * We can do this unlocked here. The futex code uses
732 		 * this flag just to verify whether the pi state
733 		 * cleanup has been done or not. In the worst case it
734 		 * loops once more. We pretend that the cleanup was
735 		 * done as there is no way to return. Either the
736 		 * OWNER_DIED bit is set by now or we push the blocked
737 		 * task into the wait for ever nirwana as well.
738 		 */
739 		tsk->flags |= PF_EXITPIDONE;
740 		set_current_state(TASK_UNINTERRUPTIBLE);
741 		schedule();
742 	}
743 
744 	exit_signals(tsk);  /* sets PF_EXITING */
745 	/*
746 	 * tsk->flags are checked in the futex code to protect against
747 	 * an exiting task cleaning up the robust pi futexes.
748 	 */
749 	smp_mb();
750 	raw_spin_unlock_wait(&tsk->pi_lock);
751 
752 	if (unlikely(in_atomic()))
753 		printk(KERN_INFO "note: %s[%d] exited with preempt_count %d\n",
754 				current->comm, task_pid_nr(current),
755 				preempt_count());
756 
757 	acct_update_integrals(tsk);
758 	/* sync mm's RSS info before statistics gathering */
759 	if (tsk->mm)
760 		sync_mm_rss(tsk->mm);
761 	group_dead = atomic_dec_and_test(&tsk->signal->live);
762 	if (group_dead) {
763 		hrtimer_cancel(&tsk->signal->real_timer);
764 		exit_itimers(tsk->signal);
765 		if (tsk->mm)
766 			setmax_mm_hiwater_rss(&tsk->signal->maxrss, tsk->mm);
767 	}
768 	acct_collect(code, group_dead);
769 	if (group_dead)
770 		tty_audit_exit();
771 	audit_free(tsk);
772 
773 	tsk->exit_code = code;
774 	taskstats_exit(tsk, group_dead);
775 
776 	exit_mm(tsk);
777 
778 	if (group_dead)
779 		acct_process();
780 	trace_sched_process_exit(tsk);
781 
782 	exit_sem(tsk);
783 	exit_shm(tsk);
784 	exit_files(tsk);
785 	exit_fs(tsk);
786 	exit_task_namespaces(tsk);
787 	exit_task_work(tsk);
788 	check_stack_usage();
789 	exit_thread();
790 
791 	/*
792 	 * Flush inherited counters to the parent - before the parent
793 	 * gets woken up by child-exit notifications.
794 	 *
795 	 * because of cgroup mode, must be called before cgroup_exit()
796 	 */
797 	perf_event_exit_task(tsk);
798 
799 	cgroup_exit(tsk, 1);
800 
801 	if (group_dead)
802 		disassociate_ctty(1);
803 
804 	module_put(task_thread_info(tsk)->exec_domain->module);
805 
806 	proc_exit_connector(tsk);
807 
808 	/*
809 	 * FIXME: do that only when needed, using sched_exit tracepoint
810 	 */
811 	flush_ptrace_hw_breakpoint(tsk);
812 
813 	exit_notify(tsk, group_dead);
814 #ifdef CONFIG_NUMA
815 	task_lock(tsk);
816 	mpol_put(tsk->mempolicy);
817 	tsk->mempolicy = NULL;
818 	task_unlock(tsk);
819 #endif
820 #ifdef CONFIG_FUTEX
821 	if (unlikely(current->pi_state_cache))
822 		kfree(current->pi_state_cache);
823 #endif
824 	/*
825 	 * Make sure we are holding no locks:
826 	 */
827 	debug_check_no_locks_held();
828 	/*
829 	 * We can do this unlocked here. The futex code uses this flag
830 	 * just to verify whether the pi state cleanup has been done
831 	 * or not. In the worst case it loops once more.
832 	 */
833 	tsk->flags |= PF_EXITPIDONE;
834 
835 	if (tsk->io_context)
836 		exit_io_context(tsk);
837 
838 	if (tsk->splice_pipe)
839 		free_pipe_info(tsk->splice_pipe);
840 
841 	if (tsk->task_frag.page)
842 		put_page(tsk->task_frag.page);
843 
844 	validate_creds_for_do_exit(tsk);
845 
846 	preempt_disable();
847 	if (tsk->nr_dirtied)
848 		__this_cpu_add(dirty_throttle_leaks, tsk->nr_dirtied);
849 	exit_rcu();
850 
851 	/*
852 	 * The setting of TASK_RUNNING by try_to_wake_up() may be delayed
853 	 * when the following two conditions become true.
854 	 *   - There is race condition of mmap_sem (It is acquired by
855 	 *     exit_mm()), and
856 	 *   - SMI occurs before setting TASK_RUNINNG.
857 	 *     (or hypervisor of virtual machine switches to other guest)
858 	 *  As a result, we may become TASK_RUNNING after becoming TASK_DEAD
859 	 *
860 	 * To avoid it, we have to wait for releasing tsk->pi_lock which
861 	 * is held by try_to_wake_up()
862 	 */
863 	smp_mb();
864 	raw_spin_unlock_wait(&tsk->pi_lock);
865 
866 	/* causes final put_task_struct in finish_task_switch(). */
867 	tsk->state = TASK_DEAD;
868 	tsk->flags |= PF_NOFREEZE;	/* tell freezer to ignore us */
869 	schedule();
870 	BUG();
871 	/* Avoid "noreturn function does return".  */
872 	for (;;)
873 		cpu_relax();	/* For when BUG is null */
874 }
875 
876 EXPORT_SYMBOL_GPL(do_exit);
877 
878 void complete_and_exit(struct completion *comp, long code)
879 {
880 	if (comp)
881 		complete(comp);
882 
883 	do_exit(code);
884 }
885 
886 EXPORT_SYMBOL(complete_and_exit);
887 
888 SYSCALL_DEFINE1(exit, int, error_code)
889 {
890 	do_exit((error_code&0xff)<<8);
891 }
892 
893 /*
894  * Take down every thread in the group.  This is called by fatal signals
895  * as well as by sys_exit_group (below).
896  */
897 void
898 do_group_exit(int exit_code)
899 {
900 	struct signal_struct *sig = current->signal;
901 
902 	BUG_ON(exit_code & 0x80); /* core dumps don't get here */
903 
904 	if (signal_group_exit(sig))
905 		exit_code = sig->group_exit_code;
906 	else if (!thread_group_empty(current)) {
907 		struct sighand_struct *const sighand = current->sighand;
908 		spin_lock_irq(&sighand->siglock);
909 		if (signal_group_exit(sig))
910 			/* Another thread got here before we took the lock.  */
911 			exit_code = sig->group_exit_code;
912 		else {
913 			sig->group_exit_code = exit_code;
914 			sig->flags = SIGNAL_GROUP_EXIT;
915 			zap_other_threads(current);
916 		}
917 		spin_unlock_irq(&sighand->siglock);
918 	}
919 
920 	do_exit(exit_code);
921 	/* NOTREACHED */
922 }
923 
924 /*
925  * this kills every thread in the thread group. Note that any externally
926  * wait4()-ing process will get the correct exit code - even if this
927  * thread is not the thread group leader.
928  */
929 SYSCALL_DEFINE1(exit_group, int, error_code)
930 {
931 	do_group_exit((error_code & 0xff) << 8);
932 	/* NOTREACHED */
933 	return 0;
934 }
935 
936 struct wait_opts {
937 	enum pid_type		wo_type;
938 	int			wo_flags;
939 	struct pid		*wo_pid;
940 
941 	struct siginfo __user	*wo_info;
942 	int __user		*wo_stat;
943 	struct rusage __user	*wo_rusage;
944 
945 	wait_queue_t		child_wait;
946 	int			notask_error;
947 };
948 
949 static inline
950 struct pid *task_pid_type(struct task_struct *task, enum pid_type type)
951 {
952 	if (type != PIDTYPE_PID)
953 		task = task->group_leader;
954 	return task->pids[type].pid;
955 }
956 
957 static int eligible_pid(struct wait_opts *wo, struct task_struct *p)
958 {
959 	return	wo->wo_type == PIDTYPE_MAX ||
960 		task_pid_type(p, wo->wo_type) == wo->wo_pid;
961 }
962 
963 static int eligible_child(struct wait_opts *wo, struct task_struct *p)
964 {
965 	if (!eligible_pid(wo, p))
966 		return 0;
967 	/* Wait for all children (clone and not) if __WALL is set;
968 	 * otherwise, wait for clone children *only* if __WCLONE is
969 	 * set; otherwise, wait for non-clone children *only*.  (Note:
970 	 * A "clone" child here is one that reports to its parent
971 	 * using a signal other than SIGCHLD.) */
972 	if (((p->exit_signal != SIGCHLD) ^ !!(wo->wo_flags & __WCLONE))
973 	    && !(wo->wo_flags & __WALL))
974 		return 0;
975 
976 	return 1;
977 }
978 
979 static int wait_noreap_copyout(struct wait_opts *wo, struct task_struct *p,
980 				pid_t pid, uid_t uid, int why, int status)
981 {
982 	struct siginfo __user *infop;
983 	int retval = wo->wo_rusage
984 		? getrusage(p, RUSAGE_BOTH, wo->wo_rusage) : 0;
985 
986 	put_task_struct(p);
987 	infop = wo->wo_info;
988 	if (infop) {
989 		if (!retval)
990 			retval = put_user(SIGCHLD, &infop->si_signo);
991 		if (!retval)
992 			retval = put_user(0, &infop->si_errno);
993 		if (!retval)
994 			retval = put_user((short)why, &infop->si_code);
995 		if (!retval)
996 			retval = put_user(pid, &infop->si_pid);
997 		if (!retval)
998 			retval = put_user(uid, &infop->si_uid);
999 		if (!retval)
1000 			retval = put_user(status, &infop->si_status);
1001 	}
1002 	if (!retval)
1003 		retval = pid;
1004 	return retval;
1005 }
1006 
1007 /*
1008  * Handle sys_wait4 work for one task in state EXIT_ZOMBIE.  We hold
1009  * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
1010  * the lock and this task is uninteresting.  If we return nonzero, we have
1011  * released the lock and the system call should return.
1012  */
1013 static int wait_task_zombie(struct wait_opts *wo, struct task_struct *p)
1014 {
1015 	unsigned long state;
1016 	int retval, status, traced;
1017 	pid_t pid = task_pid_vnr(p);
1018 	uid_t uid = from_kuid_munged(current_user_ns(), task_uid(p));
1019 	struct siginfo __user *infop;
1020 
1021 	if (!likely(wo->wo_flags & WEXITED))
1022 		return 0;
1023 
1024 	if (unlikely(wo->wo_flags & WNOWAIT)) {
1025 		int exit_code = p->exit_code;
1026 		int why;
1027 
1028 		get_task_struct(p);
1029 		read_unlock(&tasklist_lock);
1030 		if ((exit_code & 0x7f) == 0) {
1031 			why = CLD_EXITED;
1032 			status = exit_code >> 8;
1033 		} else {
1034 			why = (exit_code & 0x80) ? CLD_DUMPED : CLD_KILLED;
1035 			status = exit_code & 0x7f;
1036 		}
1037 		return wait_noreap_copyout(wo, p, pid, uid, why, status);
1038 	}
1039 
1040 	/*
1041 	 * Try to move the task's state to DEAD
1042 	 * only one thread is allowed to do this:
1043 	 */
1044 	state = xchg(&p->exit_state, EXIT_DEAD);
1045 	if (state != EXIT_ZOMBIE) {
1046 		BUG_ON(state != EXIT_DEAD);
1047 		return 0;
1048 	}
1049 
1050 	traced = ptrace_reparented(p);
1051 	/*
1052 	 * It can be ptraced but not reparented, check
1053 	 * thread_group_leader() to filter out sub-threads.
1054 	 */
1055 	if (likely(!traced) && thread_group_leader(p)) {
1056 		struct signal_struct *psig;
1057 		struct signal_struct *sig;
1058 		unsigned long maxrss;
1059 		cputime_t tgutime, tgstime;
1060 
1061 		/*
1062 		 * The resource counters for the group leader are in its
1063 		 * own task_struct.  Those for dead threads in the group
1064 		 * are in its signal_struct, as are those for the child
1065 		 * processes it has previously reaped.  All these
1066 		 * accumulate in the parent's signal_struct c* fields.
1067 		 *
1068 		 * We don't bother to take a lock here to protect these
1069 		 * p->signal fields, because they are only touched by
1070 		 * __exit_signal, which runs with tasklist_lock
1071 		 * write-locked anyway, and so is excluded here.  We do
1072 		 * need to protect the access to parent->signal fields,
1073 		 * as other threads in the parent group can be right
1074 		 * here reaping other children at the same time.
1075 		 *
1076 		 * We use thread_group_cputime_adjusted() to get times for the thread
1077 		 * group, which consolidates times for all threads in the
1078 		 * group including the group leader.
1079 		 */
1080 		thread_group_cputime_adjusted(p, &tgutime, &tgstime);
1081 		spin_lock_irq(&p->real_parent->sighand->siglock);
1082 		psig = p->real_parent->signal;
1083 		sig = p->signal;
1084 		psig->cutime += tgutime + sig->cutime;
1085 		psig->cstime += tgstime + sig->cstime;
1086 		psig->cgtime += task_gtime(p) + sig->gtime + sig->cgtime;
1087 		psig->cmin_flt +=
1088 			p->min_flt + sig->min_flt + sig->cmin_flt;
1089 		psig->cmaj_flt +=
1090 			p->maj_flt + sig->maj_flt + sig->cmaj_flt;
1091 		psig->cnvcsw +=
1092 			p->nvcsw + sig->nvcsw + sig->cnvcsw;
1093 		psig->cnivcsw +=
1094 			p->nivcsw + sig->nivcsw + sig->cnivcsw;
1095 		psig->cinblock +=
1096 			task_io_get_inblock(p) +
1097 			sig->inblock + sig->cinblock;
1098 		psig->coublock +=
1099 			task_io_get_oublock(p) +
1100 			sig->oublock + sig->coublock;
1101 		maxrss = max(sig->maxrss, sig->cmaxrss);
1102 		if (psig->cmaxrss < maxrss)
1103 			psig->cmaxrss = maxrss;
1104 		task_io_accounting_add(&psig->ioac, &p->ioac);
1105 		task_io_accounting_add(&psig->ioac, &sig->ioac);
1106 		spin_unlock_irq(&p->real_parent->sighand->siglock);
1107 	}
1108 
1109 	/*
1110 	 * Now we are sure this task is interesting, and no other
1111 	 * thread can reap it because we set its state to EXIT_DEAD.
1112 	 */
1113 	read_unlock(&tasklist_lock);
1114 
1115 	retval = wo->wo_rusage
1116 		? getrusage(p, RUSAGE_BOTH, wo->wo_rusage) : 0;
1117 	status = (p->signal->flags & SIGNAL_GROUP_EXIT)
1118 		? p->signal->group_exit_code : p->exit_code;
1119 	if (!retval && wo->wo_stat)
1120 		retval = put_user(status, wo->wo_stat);
1121 
1122 	infop = wo->wo_info;
1123 	if (!retval && infop)
1124 		retval = put_user(SIGCHLD, &infop->si_signo);
1125 	if (!retval && infop)
1126 		retval = put_user(0, &infop->si_errno);
1127 	if (!retval && infop) {
1128 		int why;
1129 
1130 		if ((status & 0x7f) == 0) {
1131 			why = CLD_EXITED;
1132 			status >>= 8;
1133 		} else {
1134 			why = (status & 0x80) ? CLD_DUMPED : CLD_KILLED;
1135 			status &= 0x7f;
1136 		}
1137 		retval = put_user((short)why, &infop->si_code);
1138 		if (!retval)
1139 			retval = put_user(status, &infop->si_status);
1140 	}
1141 	if (!retval && infop)
1142 		retval = put_user(pid, &infop->si_pid);
1143 	if (!retval && infop)
1144 		retval = put_user(uid, &infop->si_uid);
1145 	if (!retval)
1146 		retval = pid;
1147 
1148 	if (traced) {
1149 		write_lock_irq(&tasklist_lock);
1150 		/* We dropped tasklist, ptracer could die and untrace */
1151 		ptrace_unlink(p);
1152 		/*
1153 		 * If this is not a sub-thread, notify the parent.
1154 		 * If parent wants a zombie, don't release it now.
1155 		 */
1156 		if (thread_group_leader(p) &&
1157 		    !do_notify_parent(p, p->exit_signal)) {
1158 			p->exit_state = EXIT_ZOMBIE;
1159 			p = NULL;
1160 		}
1161 		write_unlock_irq(&tasklist_lock);
1162 	}
1163 	if (p != NULL)
1164 		release_task(p);
1165 
1166 	return retval;
1167 }
1168 
1169 static int *task_stopped_code(struct task_struct *p, bool ptrace)
1170 {
1171 	if (ptrace) {
1172 		if (task_is_stopped_or_traced(p) &&
1173 		    !(p->jobctl & JOBCTL_LISTENING))
1174 			return &p->exit_code;
1175 	} else {
1176 		if (p->signal->flags & SIGNAL_STOP_STOPPED)
1177 			return &p->signal->group_exit_code;
1178 	}
1179 	return NULL;
1180 }
1181 
1182 /**
1183  * wait_task_stopped - Wait for %TASK_STOPPED or %TASK_TRACED
1184  * @wo: wait options
1185  * @ptrace: is the wait for ptrace
1186  * @p: task to wait for
1187  *
1188  * Handle sys_wait4() work for %p in state %TASK_STOPPED or %TASK_TRACED.
1189  *
1190  * CONTEXT:
1191  * read_lock(&tasklist_lock), which is released if return value is
1192  * non-zero.  Also, grabs and releases @p->sighand->siglock.
1193  *
1194  * RETURNS:
1195  * 0 if wait condition didn't exist and search for other wait conditions
1196  * should continue.  Non-zero return, -errno on failure and @p's pid on
1197  * success, implies that tasklist_lock is released and wait condition
1198  * search should terminate.
1199  */
1200 static int wait_task_stopped(struct wait_opts *wo,
1201 				int ptrace, struct task_struct *p)
1202 {
1203 	struct siginfo __user *infop;
1204 	int retval, exit_code, *p_code, why;
1205 	uid_t uid = 0; /* unneeded, required by compiler */
1206 	pid_t pid;
1207 
1208 	/*
1209 	 * Traditionally we see ptrace'd stopped tasks regardless of options.
1210 	 */
1211 	if (!ptrace && !(wo->wo_flags & WUNTRACED))
1212 		return 0;
1213 
1214 	if (!task_stopped_code(p, ptrace))
1215 		return 0;
1216 
1217 	exit_code = 0;
1218 	spin_lock_irq(&p->sighand->siglock);
1219 
1220 	p_code = task_stopped_code(p, ptrace);
1221 	if (unlikely(!p_code))
1222 		goto unlock_sig;
1223 
1224 	exit_code = *p_code;
1225 	if (!exit_code)
1226 		goto unlock_sig;
1227 
1228 	if (!unlikely(wo->wo_flags & WNOWAIT))
1229 		*p_code = 0;
1230 
1231 	uid = from_kuid_munged(current_user_ns(), task_uid(p));
1232 unlock_sig:
1233 	spin_unlock_irq(&p->sighand->siglock);
1234 	if (!exit_code)
1235 		return 0;
1236 
1237 	/*
1238 	 * Now we are pretty sure this task is interesting.
1239 	 * Make sure it doesn't get reaped out from under us while we
1240 	 * give up the lock and then examine it below.  We don't want to
1241 	 * keep holding onto the tasklist_lock while we call getrusage and
1242 	 * possibly take page faults for user memory.
1243 	 */
1244 	get_task_struct(p);
1245 	pid = task_pid_vnr(p);
1246 	why = ptrace ? CLD_TRAPPED : CLD_STOPPED;
1247 	read_unlock(&tasklist_lock);
1248 
1249 	if (unlikely(wo->wo_flags & WNOWAIT))
1250 		return wait_noreap_copyout(wo, p, pid, uid, why, exit_code);
1251 
1252 	retval = wo->wo_rusage
1253 		? getrusage(p, RUSAGE_BOTH, wo->wo_rusage) : 0;
1254 	if (!retval && wo->wo_stat)
1255 		retval = put_user((exit_code << 8) | 0x7f, wo->wo_stat);
1256 
1257 	infop = wo->wo_info;
1258 	if (!retval && infop)
1259 		retval = put_user(SIGCHLD, &infop->si_signo);
1260 	if (!retval && infop)
1261 		retval = put_user(0, &infop->si_errno);
1262 	if (!retval && infop)
1263 		retval = put_user((short)why, &infop->si_code);
1264 	if (!retval && infop)
1265 		retval = put_user(exit_code, &infop->si_status);
1266 	if (!retval && infop)
1267 		retval = put_user(pid, &infop->si_pid);
1268 	if (!retval && infop)
1269 		retval = put_user(uid, &infop->si_uid);
1270 	if (!retval)
1271 		retval = pid;
1272 	put_task_struct(p);
1273 
1274 	BUG_ON(!retval);
1275 	return retval;
1276 }
1277 
1278 /*
1279  * Handle do_wait work for one task in a live, non-stopped state.
1280  * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
1281  * the lock and this task is uninteresting.  If we return nonzero, we have
1282  * released the lock and the system call should return.
1283  */
1284 static int wait_task_continued(struct wait_opts *wo, struct task_struct *p)
1285 {
1286 	int retval;
1287 	pid_t pid;
1288 	uid_t uid;
1289 
1290 	if (!unlikely(wo->wo_flags & WCONTINUED))
1291 		return 0;
1292 
1293 	if (!(p->signal->flags & SIGNAL_STOP_CONTINUED))
1294 		return 0;
1295 
1296 	spin_lock_irq(&p->sighand->siglock);
1297 	/* Re-check with the lock held.  */
1298 	if (!(p->signal->flags & SIGNAL_STOP_CONTINUED)) {
1299 		spin_unlock_irq(&p->sighand->siglock);
1300 		return 0;
1301 	}
1302 	if (!unlikely(wo->wo_flags & WNOWAIT))
1303 		p->signal->flags &= ~SIGNAL_STOP_CONTINUED;
1304 	uid = from_kuid_munged(current_user_ns(), task_uid(p));
1305 	spin_unlock_irq(&p->sighand->siglock);
1306 
1307 	pid = task_pid_vnr(p);
1308 	get_task_struct(p);
1309 	read_unlock(&tasklist_lock);
1310 
1311 	if (!wo->wo_info) {
1312 		retval = wo->wo_rusage
1313 			? getrusage(p, RUSAGE_BOTH, wo->wo_rusage) : 0;
1314 		put_task_struct(p);
1315 		if (!retval && wo->wo_stat)
1316 			retval = put_user(0xffff, wo->wo_stat);
1317 		if (!retval)
1318 			retval = pid;
1319 	} else {
1320 		retval = wait_noreap_copyout(wo, p, pid, uid,
1321 					     CLD_CONTINUED, SIGCONT);
1322 		BUG_ON(retval == 0);
1323 	}
1324 
1325 	return retval;
1326 }
1327 
1328 /*
1329  * Consider @p for a wait by @parent.
1330  *
1331  * -ECHILD should be in ->notask_error before the first call.
1332  * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1333  * Returns zero if the search for a child should continue;
1334  * then ->notask_error is 0 if @p is an eligible child,
1335  * or another error from security_task_wait(), or still -ECHILD.
1336  */
1337 static int wait_consider_task(struct wait_opts *wo, int ptrace,
1338 				struct task_struct *p)
1339 {
1340 	int ret = eligible_child(wo, p);
1341 	if (!ret)
1342 		return ret;
1343 
1344 	ret = security_task_wait(p);
1345 	if (unlikely(ret < 0)) {
1346 		/*
1347 		 * If we have not yet seen any eligible child,
1348 		 * then let this error code replace -ECHILD.
1349 		 * A permission error will give the user a clue
1350 		 * to look for security policy problems, rather
1351 		 * than for mysterious wait bugs.
1352 		 */
1353 		if (wo->notask_error)
1354 			wo->notask_error = ret;
1355 		return 0;
1356 	}
1357 
1358 	/* dead body doesn't have much to contribute */
1359 	if (unlikely(p->exit_state == EXIT_DEAD)) {
1360 		/*
1361 		 * But do not ignore this task until the tracer does
1362 		 * wait_task_zombie()->do_notify_parent().
1363 		 */
1364 		if (likely(!ptrace) && unlikely(ptrace_reparented(p)))
1365 			wo->notask_error = 0;
1366 		return 0;
1367 	}
1368 
1369 	/* slay zombie? */
1370 	if (p->exit_state == EXIT_ZOMBIE) {
1371 		/*
1372 		 * A zombie ptracee is only visible to its ptracer.
1373 		 * Notification and reaping will be cascaded to the real
1374 		 * parent when the ptracer detaches.
1375 		 */
1376 		if (likely(!ptrace) && unlikely(p->ptrace)) {
1377 			/* it will become visible, clear notask_error */
1378 			wo->notask_error = 0;
1379 			return 0;
1380 		}
1381 
1382 		/* we don't reap group leaders with subthreads */
1383 		if (!delay_group_leader(p))
1384 			return wait_task_zombie(wo, p);
1385 
1386 		/*
1387 		 * Allow access to stopped/continued state via zombie by
1388 		 * falling through.  Clearing of notask_error is complex.
1389 		 *
1390 		 * When !@ptrace:
1391 		 *
1392 		 * If WEXITED is set, notask_error should naturally be
1393 		 * cleared.  If not, subset of WSTOPPED|WCONTINUED is set,
1394 		 * so, if there are live subthreads, there are events to
1395 		 * wait for.  If all subthreads are dead, it's still safe
1396 		 * to clear - this function will be called again in finite
1397 		 * amount time once all the subthreads are released and
1398 		 * will then return without clearing.
1399 		 *
1400 		 * When @ptrace:
1401 		 *
1402 		 * Stopped state is per-task and thus can't change once the
1403 		 * target task dies.  Only continued and exited can happen.
1404 		 * Clear notask_error if WCONTINUED | WEXITED.
1405 		 */
1406 		if (likely(!ptrace) || (wo->wo_flags & (WCONTINUED | WEXITED)))
1407 			wo->notask_error = 0;
1408 	} else {
1409 		/*
1410 		 * If @p is ptraced by a task in its real parent's group,
1411 		 * hide group stop/continued state when looking at @p as
1412 		 * the real parent; otherwise, a single stop can be
1413 		 * reported twice as group and ptrace stops.
1414 		 *
1415 		 * If a ptracer wants to distinguish the two events for its
1416 		 * own children, it should create a separate process which
1417 		 * takes the role of real parent.
1418 		 */
1419 		if (likely(!ptrace) && p->ptrace && !ptrace_reparented(p))
1420 			return 0;
1421 
1422 		/*
1423 		 * @p is alive and it's gonna stop, continue or exit, so
1424 		 * there always is something to wait for.
1425 		 */
1426 		wo->notask_error = 0;
1427 	}
1428 
1429 	/*
1430 	 * Wait for stopped.  Depending on @ptrace, different stopped state
1431 	 * is used and the two don't interact with each other.
1432 	 */
1433 	ret = wait_task_stopped(wo, ptrace, p);
1434 	if (ret)
1435 		return ret;
1436 
1437 	/*
1438 	 * Wait for continued.  There's only one continued state and the
1439 	 * ptracer can consume it which can confuse the real parent.  Don't
1440 	 * use WCONTINUED from ptracer.  You don't need or want it.
1441 	 */
1442 	return wait_task_continued(wo, p);
1443 }
1444 
1445 /*
1446  * Do the work of do_wait() for one thread in the group, @tsk.
1447  *
1448  * -ECHILD should be in ->notask_error before the first call.
1449  * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1450  * Returns zero if the search for a child should continue; then
1451  * ->notask_error is 0 if there were any eligible children,
1452  * or another error from security_task_wait(), or still -ECHILD.
1453  */
1454 static int do_wait_thread(struct wait_opts *wo, struct task_struct *tsk)
1455 {
1456 	struct task_struct *p;
1457 
1458 	list_for_each_entry(p, &tsk->children, sibling) {
1459 		int ret = wait_consider_task(wo, 0, p);
1460 		if (ret)
1461 			return ret;
1462 	}
1463 
1464 	return 0;
1465 }
1466 
1467 static int ptrace_do_wait(struct wait_opts *wo, struct task_struct *tsk)
1468 {
1469 	struct task_struct *p;
1470 
1471 	list_for_each_entry(p, &tsk->ptraced, ptrace_entry) {
1472 		int ret = wait_consider_task(wo, 1, p);
1473 		if (ret)
1474 			return ret;
1475 	}
1476 
1477 	return 0;
1478 }
1479 
1480 static int child_wait_callback(wait_queue_t *wait, unsigned mode,
1481 				int sync, void *key)
1482 {
1483 	struct wait_opts *wo = container_of(wait, struct wait_opts,
1484 						child_wait);
1485 	struct task_struct *p = key;
1486 
1487 	if (!eligible_pid(wo, p))
1488 		return 0;
1489 
1490 	if ((wo->wo_flags & __WNOTHREAD) && wait->private != p->parent)
1491 		return 0;
1492 
1493 	return default_wake_function(wait, mode, sync, key);
1494 }
1495 
1496 void __wake_up_parent(struct task_struct *p, struct task_struct *parent)
1497 {
1498 	__wake_up_sync_key(&parent->signal->wait_chldexit,
1499 				TASK_INTERRUPTIBLE, 1, p);
1500 }
1501 
1502 static long do_wait(struct wait_opts *wo)
1503 {
1504 	struct task_struct *tsk;
1505 	int retval;
1506 
1507 	trace_sched_process_wait(wo->wo_pid);
1508 
1509 	init_waitqueue_func_entry(&wo->child_wait, child_wait_callback);
1510 	wo->child_wait.private = current;
1511 	add_wait_queue(&current->signal->wait_chldexit, &wo->child_wait);
1512 repeat:
1513 	/*
1514 	 * If there is nothing that can match our critiera just get out.
1515 	 * We will clear ->notask_error to zero if we see any child that
1516 	 * might later match our criteria, even if we are not able to reap
1517 	 * it yet.
1518 	 */
1519 	wo->notask_error = -ECHILD;
1520 	if ((wo->wo_type < PIDTYPE_MAX) &&
1521 	   (!wo->wo_pid || hlist_empty(&wo->wo_pid->tasks[wo->wo_type])))
1522 		goto notask;
1523 
1524 	set_current_state(TASK_INTERRUPTIBLE);
1525 	read_lock(&tasklist_lock);
1526 	tsk = current;
1527 	do {
1528 		retval = do_wait_thread(wo, tsk);
1529 		if (retval)
1530 			goto end;
1531 
1532 		retval = ptrace_do_wait(wo, tsk);
1533 		if (retval)
1534 			goto end;
1535 
1536 		if (wo->wo_flags & __WNOTHREAD)
1537 			break;
1538 	} while_each_thread(current, tsk);
1539 	read_unlock(&tasklist_lock);
1540 
1541 notask:
1542 	retval = wo->notask_error;
1543 	if (!retval && !(wo->wo_flags & WNOHANG)) {
1544 		retval = -ERESTARTSYS;
1545 		if (!signal_pending(current)) {
1546 			schedule();
1547 			goto repeat;
1548 		}
1549 	}
1550 end:
1551 	__set_current_state(TASK_RUNNING);
1552 	remove_wait_queue(&current->signal->wait_chldexit, &wo->child_wait);
1553 	return retval;
1554 }
1555 
1556 SYSCALL_DEFINE5(waitid, int, which, pid_t, upid, struct siginfo __user *,
1557 		infop, int, options, struct rusage __user *, ru)
1558 {
1559 	struct wait_opts wo;
1560 	struct pid *pid = NULL;
1561 	enum pid_type type;
1562 	long ret;
1563 
1564 	if (options & ~(WNOHANG|WNOWAIT|WEXITED|WSTOPPED|WCONTINUED))
1565 		return -EINVAL;
1566 	if (!(options & (WEXITED|WSTOPPED|WCONTINUED)))
1567 		return -EINVAL;
1568 
1569 	switch (which) {
1570 	case P_ALL:
1571 		type = PIDTYPE_MAX;
1572 		break;
1573 	case P_PID:
1574 		type = PIDTYPE_PID;
1575 		if (upid <= 0)
1576 			return -EINVAL;
1577 		break;
1578 	case P_PGID:
1579 		type = PIDTYPE_PGID;
1580 		if (upid <= 0)
1581 			return -EINVAL;
1582 		break;
1583 	default:
1584 		return -EINVAL;
1585 	}
1586 
1587 	if (type < PIDTYPE_MAX)
1588 		pid = find_get_pid(upid);
1589 
1590 	wo.wo_type	= type;
1591 	wo.wo_pid	= pid;
1592 	wo.wo_flags	= options;
1593 	wo.wo_info	= infop;
1594 	wo.wo_stat	= NULL;
1595 	wo.wo_rusage	= ru;
1596 	ret = do_wait(&wo);
1597 
1598 	if (ret > 0) {
1599 		ret = 0;
1600 	} else if (infop) {
1601 		/*
1602 		 * For a WNOHANG return, clear out all the fields
1603 		 * we would set so the user can easily tell the
1604 		 * difference.
1605 		 */
1606 		if (!ret)
1607 			ret = put_user(0, &infop->si_signo);
1608 		if (!ret)
1609 			ret = put_user(0, &infop->si_errno);
1610 		if (!ret)
1611 			ret = put_user(0, &infop->si_code);
1612 		if (!ret)
1613 			ret = put_user(0, &infop->si_pid);
1614 		if (!ret)
1615 			ret = put_user(0, &infop->si_uid);
1616 		if (!ret)
1617 			ret = put_user(0, &infop->si_status);
1618 	}
1619 
1620 	put_pid(pid);
1621 	return ret;
1622 }
1623 
1624 SYSCALL_DEFINE4(wait4, pid_t, upid, int __user *, stat_addr,
1625 		int, options, struct rusage __user *, ru)
1626 {
1627 	struct wait_opts wo;
1628 	struct pid *pid = NULL;
1629 	enum pid_type type;
1630 	long ret;
1631 
1632 	if (options & ~(WNOHANG|WUNTRACED|WCONTINUED|
1633 			__WNOTHREAD|__WCLONE|__WALL))
1634 		return -EINVAL;
1635 
1636 	if (upid == -1)
1637 		type = PIDTYPE_MAX;
1638 	else if (upid < 0) {
1639 		type = PIDTYPE_PGID;
1640 		pid = find_get_pid(-upid);
1641 	} else if (upid == 0) {
1642 		type = PIDTYPE_PGID;
1643 		pid = get_task_pid(current, PIDTYPE_PGID);
1644 	} else /* upid > 0 */ {
1645 		type = PIDTYPE_PID;
1646 		pid = find_get_pid(upid);
1647 	}
1648 
1649 	wo.wo_type	= type;
1650 	wo.wo_pid	= pid;
1651 	wo.wo_flags	= options | WEXITED;
1652 	wo.wo_info	= NULL;
1653 	wo.wo_stat	= stat_addr;
1654 	wo.wo_rusage	= ru;
1655 	ret = do_wait(&wo);
1656 	put_pid(pid);
1657 
1658 	return ret;
1659 }
1660 
1661 #ifdef __ARCH_WANT_SYS_WAITPID
1662 
1663 /*
1664  * sys_waitpid() remains for compatibility. waitpid() should be
1665  * implemented by calling sys_wait4() from libc.a.
1666  */
1667 SYSCALL_DEFINE3(waitpid, pid_t, pid, int __user *, stat_addr, int, options)
1668 {
1669 	return sys_wait4(pid, stat_addr, options, NULL);
1670 }
1671 
1672 #endif
1673